2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
7 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
11 * This handles all read/write requests to block devices
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/bio.h>
17 #include <linux/blkdev.h>
18 #include <linux/highmem.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/string.h>
22 #include <linux/init.h>
23 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/interrupt.h>
30 #include <linux/cpu.h>
31 #include <linux/blktrace_api.h>
32 #include <linux/fault-inject.h>
37 #include <scsi/scsi_cmnd.h>
39 static void blk_unplug_work(struct work_struct
*work
);
40 static void blk_unplug_timeout(unsigned long data
);
41 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
42 static void init_request_from_bio(struct request
*req
, struct bio
*bio
);
43 static int __make_request(request_queue_t
*q
, struct bio
*bio
);
44 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
);
47 * For the allocated request tables
49 static struct kmem_cache
*request_cachep
;
52 * For queue allocation
54 static struct kmem_cache
*requestq_cachep
;
57 * For io context allocations
59 static struct kmem_cache
*iocontext_cachep
;
62 * Controlling structure to kblockd
64 static struct workqueue_struct
*kblockd_workqueue
;
66 unsigned long blk_max_low_pfn
, blk_max_pfn
;
68 EXPORT_SYMBOL(blk_max_low_pfn
);
69 EXPORT_SYMBOL(blk_max_pfn
);
71 static DEFINE_PER_CPU(struct list_head
, blk_cpu_done
);
73 /* Amount of time in which a process may batch requests */
74 #define BLK_BATCH_TIME (HZ/50UL)
76 /* Number of requests a "batching" process may submit */
77 #define BLK_BATCH_REQ 32
80 * Return the threshold (number of used requests) at which the queue is
81 * considered to be congested. It include a little hysteresis to keep the
82 * context switch rate down.
84 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
86 return q
->nr_congestion_on
;
90 * The threshold at which a queue is considered to be uncongested
92 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
94 return q
->nr_congestion_off
;
97 static void blk_queue_congestion_threshold(struct request_queue
*q
)
101 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
102 if (nr
> q
->nr_requests
)
104 q
->nr_congestion_on
= nr
;
106 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
109 q
->nr_congestion_off
= nr
;
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
116 * Locates the passed device's request queue and returns the address of its
119 * Will return NULL if the request queue cannot be located.
121 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
123 struct backing_dev_info
*ret
= NULL
;
124 request_queue_t
*q
= bdev_get_queue(bdev
);
127 ret
= &q
->backing_dev_info
;
130 EXPORT_SYMBOL(blk_get_backing_dev_info
);
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(request_queue_t
*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(request_queue_t
*q
, merge_bvec_fn
*mbfn
)
168 q
->merge_bvec_fn
= mbfn
;
171 EXPORT_SYMBOL(blk_queue_merge_bvec
);
173 void blk_queue_softirq_done(request_queue_t
*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(request_queue_t
* 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(request_queue_t
*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
;
262 * blk_queue_ordered - does this queue support ordered writes
263 * @q: the request queue
264 * @ordered: one of QUEUE_ORDERED_*
265 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
268 * For journalled file systems, doing ordered writes on a commit
269 * block instead of explicitly doing wait_on_buffer (which is bad
270 * for performance) can be a big win. Block drivers supporting this
271 * feature should call this function and indicate so.
274 int blk_queue_ordered(request_queue_t
*q
, unsigned ordered
,
275 prepare_flush_fn
*prepare_flush_fn
)
277 if (ordered
& (QUEUE_ORDERED_PREFLUSH
| QUEUE_ORDERED_POSTFLUSH
) &&
278 prepare_flush_fn
== NULL
) {
279 printk(KERN_ERR
"blk_queue_ordered: prepare_flush_fn required\n");
283 if (ordered
!= QUEUE_ORDERED_NONE
&&
284 ordered
!= QUEUE_ORDERED_DRAIN
&&
285 ordered
!= QUEUE_ORDERED_DRAIN_FLUSH
&&
286 ordered
!= QUEUE_ORDERED_DRAIN_FUA
&&
287 ordered
!= QUEUE_ORDERED_TAG
&&
288 ordered
!= QUEUE_ORDERED_TAG_FLUSH
&&
289 ordered
!= QUEUE_ORDERED_TAG_FUA
) {
290 printk(KERN_ERR
"blk_queue_ordered: bad value %d\n", ordered
);
294 q
->ordered
= ordered
;
295 q
->next_ordered
= ordered
;
296 q
->prepare_flush_fn
= prepare_flush_fn
;
301 EXPORT_SYMBOL(blk_queue_ordered
);
304 * blk_queue_issue_flush_fn - set function for issuing a flush
305 * @q: the request queue
306 * @iff: the function to be called issuing the flush
309 * If a driver supports issuing a flush command, the support is notified
310 * to the block layer by defining it through this call.
313 void blk_queue_issue_flush_fn(request_queue_t
*q
, issue_flush_fn
*iff
)
315 q
->issue_flush_fn
= iff
;
318 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
321 * Cache flushing for ordered writes handling
323 inline unsigned blk_ordered_cur_seq(request_queue_t
*q
)
327 return 1 << ffz(q
->ordseq
);
330 unsigned blk_ordered_req_seq(struct request
*rq
)
332 request_queue_t
*q
= rq
->q
;
334 BUG_ON(q
->ordseq
== 0);
336 if (rq
== &q
->pre_flush_rq
)
337 return QUEUE_ORDSEQ_PREFLUSH
;
338 if (rq
== &q
->bar_rq
)
339 return QUEUE_ORDSEQ_BAR
;
340 if (rq
== &q
->post_flush_rq
)
341 return QUEUE_ORDSEQ_POSTFLUSH
;
344 * !fs requests don't need to follow barrier ordering. Always
345 * put them at the front. This fixes the following deadlock.
347 * http://thread.gmane.org/gmane.linux.kernel/537473
349 if (!blk_fs_request(rq
))
350 return QUEUE_ORDSEQ_DRAIN
;
352 if ((rq
->cmd_flags
& REQ_ORDERED_COLOR
) ==
353 (q
->orig_bar_rq
->cmd_flags
& REQ_ORDERED_COLOR
))
354 return QUEUE_ORDSEQ_DRAIN
;
356 return QUEUE_ORDSEQ_DONE
;
359 void blk_ordered_complete_seq(request_queue_t
*q
, unsigned seq
, int error
)
363 if (error
&& !q
->orderr
)
366 BUG_ON(q
->ordseq
& seq
);
369 if (blk_ordered_cur_seq(q
) != QUEUE_ORDSEQ_DONE
)
373 * Okay, sequence complete.
378 if (__blk_end_request(rq
, q
->orderr
, blk_rq_bytes(rq
)))
382 static void pre_flush_end_io(struct request
*rq
, int error
)
384 elv_completed_request(rq
->q
, rq
);
385 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_PREFLUSH
, error
);
388 static void bar_end_io(struct request
*rq
, int error
)
390 elv_completed_request(rq
->q
, rq
);
391 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_BAR
, error
);
394 static void post_flush_end_io(struct request
*rq
, int error
)
396 elv_completed_request(rq
->q
, rq
);
397 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_POSTFLUSH
, error
);
400 static void queue_flush(request_queue_t
*q
, unsigned which
)
403 rq_end_io_fn
*end_io
;
405 if (which
== QUEUE_ORDERED_PREFLUSH
) {
406 rq
= &q
->pre_flush_rq
;
407 end_io
= pre_flush_end_io
;
409 rq
= &q
->post_flush_rq
;
410 end_io
= post_flush_end_io
;
413 rq
->cmd_flags
= REQ_HARDBARRIER
;
415 rq
->elevator_private
= NULL
;
416 rq
->elevator_private2
= NULL
;
417 rq
->rq_disk
= q
->bar_rq
.rq_disk
;
419 q
->prepare_flush_fn(q
, rq
);
421 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
424 static inline struct request
*start_ordered(request_queue_t
*q
,
429 q
->ordered
= q
->next_ordered
;
430 q
->ordseq
|= QUEUE_ORDSEQ_STARTED
;
433 * Prep proxy barrier request.
435 blkdev_dequeue_request(rq
);
440 if (bio_data_dir(q
->orig_bar_rq
->bio
) == WRITE
)
441 rq
->cmd_flags
|= REQ_RW
;
442 rq
->cmd_flags
|= q
->ordered
& QUEUE_ORDERED_FUA
? REQ_FUA
: 0;
443 rq
->elevator_private
= NULL
;
444 rq
->elevator_private2
= NULL
;
445 init_request_from_bio(rq
, q
->orig_bar_rq
->bio
);
446 rq
->end_io
= bar_end_io
;
449 * Queue ordered sequence. As we stack them at the head, we
450 * need to queue in reverse order. Note that we rely on that
451 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
452 * request gets inbetween ordered sequence.
454 if (q
->ordered
& QUEUE_ORDERED_POSTFLUSH
)
455 queue_flush(q
, QUEUE_ORDERED_POSTFLUSH
);
457 q
->ordseq
|= QUEUE_ORDSEQ_POSTFLUSH
;
459 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
461 if (q
->ordered
& QUEUE_ORDERED_PREFLUSH
) {
462 queue_flush(q
, QUEUE_ORDERED_PREFLUSH
);
463 rq
= &q
->pre_flush_rq
;
465 q
->ordseq
|= QUEUE_ORDSEQ_PREFLUSH
;
467 if ((q
->ordered
& QUEUE_ORDERED_TAG
) || q
->in_flight
== 0)
468 q
->ordseq
|= QUEUE_ORDSEQ_DRAIN
;
475 int blk_do_ordered(request_queue_t
*q
, struct request
**rqp
)
477 struct request
*rq
= *rqp
;
478 int is_barrier
= blk_fs_request(rq
) && blk_barrier_rq(rq
);
484 if (q
->next_ordered
!= QUEUE_ORDERED_NONE
) {
485 *rqp
= start_ordered(q
, rq
);
489 * This can happen when the queue switches to
490 * ORDERED_NONE while this request is on it.
492 blkdev_dequeue_request(rq
);
493 if (__blk_end_request(rq
, -EOPNOTSUPP
,
502 * Ordered sequence in progress
505 /* Special requests are not subject to ordering rules. */
506 if (!blk_fs_request(rq
) &&
507 rq
!= &q
->pre_flush_rq
&& rq
!= &q
->post_flush_rq
)
510 if (q
->ordered
& QUEUE_ORDERED_TAG
) {
511 /* Ordered by tag. Blocking the next barrier is enough. */
512 if (is_barrier
&& rq
!= &q
->bar_rq
)
515 /* Ordered by draining. Wait for turn. */
516 WARN_ON(blk_ordered_req_seq(rq
) < blk_ordered_cur_seq(q
));
517 if (blk_ordered_req_seq(rq
) > blk_ordered_cur_seq(q
))
524 static int flush_dry_bio_endio(struct bio
*bio
, unsigned int bytes
, int error
)
526 request_queue_t
*q
= bio
->bi_private
;
527 struct bio_vec
*bvec
;
531 * This is dry run, restore bio_sector and size. We'll finish
532 * this request again with the original bi_end_io after an
533 * error occurs or post flush is complete.
542 bio_for_each_segment(bvec
, bio
, i
) {
543 bvec
->bv_len
+= bvec
->bv_offset
;
548 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
549 bio
->bi_size
= q
->bi_size
;
550 bio
->bi_sector
-= (q
->bi_size
>> 9);
556 static int ordered_bio_endio(struct request
*rq
, struct bio
*bio
,
557 unsigned int nbytes
, int error
)
559 request_queue_t
*q
= rq
->q
;
563 if (&q
->bar_rq
!= rq
)
567 * Okay, this is the barrier request in progress, dry finish it.
569 if (error
&& !q
->orderr
)
572 endio
= bio
->bi_end_io
;
573 private = bio
->bi_private
;
574 bio
->bi_end_io
= flush_dry_bio_endio
;
577 bio_endio(bio
, nbytes
, error
);
579 bio
->bi_end_io
= endio
;
580 bio
->bi_private
= private;
586 * blk_queue_bounce_limit - set bounce buffer limit for queue
587 * @q: the request queue for the device
588 * @dma_addr: bus address limit
591 * Different hardware can have different requirements as to what pages
592 * it can do I/O directly to. A low level driver can call
593 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
594 * buffers for doing I/O to pages residing above @page.
596 void blk_queue_bounce_limit(request_queue_t
*q
, u64 dma_addr
)
598 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
601 q
->bounce_gfp
= GFP_NOIO
;
602 #if BITS_PER_LONG == 64
603 /* Assume anything <= 4GB can be handled by IOMMU.
604 Actually some IOMMUs can handle everything, but I don't
605 know of a way to test this here. */
606 if (bounce_pfn
< (min_t(u64
,0xffffffff,BLK_BOUNCE_HIGH
) >> PAGE_SHIFT
))
608 q
->bounce_pfn
= max_low_pfn
;
610 if (bounce_pfn
< blk_max_low_pfn
)
612 q
->bounce_pfn
= bounce_pfn
;
615 init_emergency_isa_pool();
616 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
617 q
->bounce_pfn
= bounce_pfn
;
621 EXPORT_SYMBOL(blk_queue_bounce_limit
);
624 * blk_queue_max_sectors - set max sectors for a request for this queue
625 * @q: the request queue for the device
626 * @max_sectors: max sectors in the usual 512b unit
629 * Enables a low level driver to set an upper limit on the size of
632 void blk_queue_max_sectors(request_queue_t
*q
, unsigned int max_sectors
)
634 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
635 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
636 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
639 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
640 q
->max_hw_sectors
= q
->max_sectors
= max_sectors
;
642 q
->max_sectors
= BLK_DEF_MAX_SECTORS
;
643 q
->max_hw_sectors
= max_sectors
;
647 EXPORT_SYMBOL(blk_queue_max_sectors
);
650 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
651 * @q: the request queue for the device
652 * @max_segments: max number of segments
655 * Enables a low level driver to set an upper limit on the number of
656 * physical data segments in a request. This would be the largest sized
657 * scatter list the driver could handle.
659 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
663 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
666 q
->max_phys_segments
= max_segments
;
669 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
672 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
673 * @q: the request queue for the device
674 * @max_segments: max number of segments
677 * Enables a low level driver to set an upper limit on the number of
678 * hw data segments in a request. This would be the largest number of
679 * address/length pairs the host adapter can actually give as once
682 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
686 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
689 q
->max_hw_segments
= max_segments
;
692 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
695 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
696 * @q: the request queue for the device
697 * @max_size: max size of segment in bytes
700 * Enables a low level driver to set an upper limit on the size of a
703 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
705 if (max_size
< PAGE_CACHE_SIZE
) {
706 max_size
= PAGE_CACHE_SIZE
;
707 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
710 q
->max_segment_size
= max_size
;
713 EXPORT_SYMBOL(blk_queue_max_segment_size
);
716 * blk_queue_hardsect_size - set hardware sector size for the queue
717 * @q: the request queue for the device
718 * @size: the hardware sector size, in bytes
721 * This should typically be set to the lowest possible sector size
722 * that the hardware can operate on (possible without reverting to
723 * even internal read-modify-write operations). Usually the default
724 * of 512 covers most hardware.
726 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
728 q
->hardsect_size
= size
;
731 EXPORT_SYMBOL(blk_queue_hardsect_size
);
734 * Returns the minimum that is _not_ zero, unless both are zero.
736 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
739 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
740 * @t: the stacking driver (top)
741 * @b: the underlying device (bottom)
743 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
745 /* zero is "infinity" */
746 t
->max_sectors
= min_not_zero(t
->max_sectors
,b
->max_sectors
);
747 t
->max_hw_sectors
= min_not_zero(t
->max_hw_sectors
,b
->max_hw_sectors
);
749 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
750 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
751 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
752 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
753 if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
))
754 clear_bit(QUEUE_FLAG_CLUSTER
, &t
->queue_flags
);
757 EXPORT_SYMBOL(blk_queue_stack_limits
);
760 * blk_queue_segment_boundary - set boundary rules for segment merging
761 * @q: the request queue for the device
762 * @mask: the memory boundary mask
764 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
766 if (mask
< PAGE_CACHE_SIZE
- 1) {
767 mask
= PAGE_CACHE_SIZE
- 1;
768 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
771 q
->seg_boundary_mask
= mask
;
774 EXPORT_SYMBOL(blk_queue_segment_boundary
);
777 * blk_queue_dma_alignment - set dma length and memory alignment
778 * @q: the request queue for the device
779 * @mask: alignment mask
782 * set required memory and length aligment for direct dma transactions.
783 * this is used when buiding direct io requests for the queue.
786 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
788 q
->dma_alignment
= mask
;
791 EXPORT_SYMBOL(blk_queue_dma_alignment
);
794 * blk_queue_update_dma_alignment - update dma length and memory alignment
795 * @q: the request queue for the device
796 * @mask: alignment mask
799 * update required memory and length aligment for direct dma transactions.
800 * If the requested alignment is larger than the current alignment, then
801 * the current queue alignment is updated to the new value, otherwise it
802 * is left alone. The design of this is to allow multiple objects
803 * (driver, device, transport etc) to set their respective
804 * alignments without having them interfere.
807 void blk_queue_update_dma_alignment(request_queue_t
*q
, int mask
)
809 BUG_ON(mask
> PAGE_SIZE
);
811 if (mask
> q
->dma_alignment
)
812 q
->dma_alignment
= mask
;
815 EXPORT_SYMBOL(blk_queue_update_dma_alignment
);
818 * blk_queue_find_tag - find a request by its tag and queue
819 * @q: The request queue for the device
820 * @tag: The tag of the request
823 * Should be used when a device returns a tag and you want to match
826 * no locks need be held.
828 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
830 return blk_map_queue_find_tag(q
->queue_tags
, tag
);
833 EXPORT_SYMBOL(blk_queue_find_tag
);
836 * __blk_free_tags - release a given set of tag maintenance info
837 * @bqt: the tag map to free
839 * Tries to free the specified @bqt@. Returns true if it was
840 * actually freed and false if there are still references using it
842 static int __blk_free_tags(struct blk_queue_tag
*bqt
)
846 retval
= atomic_dec_and_test(&bqt
->refcnt
);
850 kfree(bqt
->tag_index
);
851 bqt
->tag_index
= NULL
;
864 * __blk_queue_free_tags - release tag maintenance info
865 * @q: the request queue for the device
868 * blk_cleanup_queue() will take care of calling this function, if tagging
869 * has been used. So there's no need to call this directly.
871 static void __blk_queue_free_tags(request_queue_t
*q
)
873 struct blk_queue_tag
*bqt
= q
->queue_tags
;
878 __blk_free_tags(bqt
);
880 q
->queue_tags
= NULL
;
881 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
886 * blk_free_tags - release a given set of tag maintenance info
887 * @bqt: the tag map to free
889 * For externally managed @bqt@ frees the map. Callers of this
890 * function must guarantee to have released all the queues that
891 * might have been using this tag map.
893 void blk_free_tags(struct blk_queue_tag
*bqt
)
895 if (unlikely(!__blk_free_tags(bqt
)))
898 EXPORT_SYMBOL(blk_free_tags
);
901 * blk_queue_free_tags - release tag maintenance info
902 * @q: the request queue for the device
905 * This is used to disabled tagged queuing to a device, yet leave
908 void blk_queue_free_tags(request_queue_t
*q
)
910 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
913 EXPORT_SYMBOL(blk_queue_free_tags
);
916 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
918 struct request
**tag_index
;
919 unsigned long *tag_map
;
922 if (q
&& depth
> q
->nr_requests
* 2) {
923 depth
= q
->nr_requests
* 2;
924 printk(KERN_ERR
"%s: adjusted depth to %d\n",
925 __FUNCTION__
, depth
);
928 tag_index
= kzalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
932 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
933 tag_map
= kzalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
937 tags
->real_max_depth
= depth
;
938 tags
->max_depth
= depth
;
939 tags
->tag_index
= tag_index
;
940 tags
->tag_map
= tag_map
;
948 static struct blk_queue_tag
*__blk_queue_init_tags(struct request_queue
*q
,
951 struct blk_queue_tag
*tags
;
953 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
957 if (init_tag_map(q
, tags
, depth
))
961 atomic_set(&tags
->refcnt
, 1);
969 * blk_init_tags - initialize the tag info for an external tag map
970 * @depth: the maximum queue depth supported
971 * @tags: the tag to use
973 struct blk_queue_tag
*blk_init_tags(int depth
)
975 return __blk_queue_init_tags(NULL
, depth
);
977 EXPORT_SYMBOL(blk_init_tags
);
980 * blk_queue_init_tags - initialize the queue tag info
981 * @q: the request queue for the device
982 * @depth: the maximum queue depth supported
983 * @tags: the tag to use
985 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
986 struct blk_queue_tag
*tags
)
990 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
992 if (!tags
&& !q
->queue_tags
) {
993 tags
= __blk_queue_init_tags(q
, depth
);
997 } else if (q
->queue_tags
) {
998 if ((rc
= blk_queue_resize_tags(q
, depth
)))
1000 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
1003 atomic_inc(&tags
->refcnt
);
1006 * assign it, all done
1008 q
->queue_tags
= tags
;
1009 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
1010 INIT_LIST_HEAD(&q
->tag_busy_list
);
1017 EXPORT_SYMBOL(blk_queue_init_tags
);
1020 * blk_queue_resize_tags - change the queueing depth
1021 * @q: the request queue for the device
1022 * @new_depth: the new max command queueing depth
1025 * Must be called with the queue lock held.
1027 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
1029 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1030 struct request
**tag_index
;
1031 unsigned long *tag_map
;
1032 int max_depth
, nr_ulongs
;
1038 * if we already have large enough real_max_depth. just
1039 * adjust max_depth. *NOTE* as requests with tag value
1040 * between new_depth and real_max_depth can be in-flight, tag
1041 * map can not be shrunk blindly here.
1043 if (new_depth
<= bqt
->real_max_depth
) {
1044 bqt
->max_depth
= new_depth
;
1049 * Currently cannot replace a shared tag map with a new
1050 * one, so error out if this is the case
1052 if (atomic_read(&bqt
->refcnt
) != 1)
1056 * save the old state info, so we can copy it back
1058 tag_index
= bqt
->tag_index
;
1059 tag_map
= bqt
->tag_map
;
1060 max_depth
= bqt
->real_max_depth
;
1062 if (init_tag_map(q
, bqt
, new_depth
))
1065 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
1066 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
1067 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
1074 EXPORT_SYMBOL(blk_queue_resize_tags
);
1077 * blk_queue_end_tag - end tag operations for a request
1078 * @q: the request queue for the device
1079 * @rq: the request that has completed
1082 * Typically called when end_that_request_first() returns 0, meaning
1083 * all transfers have been done for a request. It's important to call
1084 * this function before end_that_request_last(), as that will put the
1085 * request back on the free list thus corrupting the internal tag list.
1088 * queue lock must be held.
1090 void blk_queue_end_tag(request_queue_t
*q
, struct request
*rq
)
1092 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1097 if (unlikely(tag
>= bqt
->real_max_depth
))
1099 * This can happen after tag depth has been reduced.
1100 * FIXME: how about a warning or info message here?
1104 list_del_init(&rq
->queuelist
);
1105 rq
->cmd_flags
&= ~REQ_QUEUED
;
1108 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
1109 printk(KERN_ERR
"%s: tag %d is missing\n",
1112 bqt
->tag_index
[tag
] = NULL
;
1114 if (unlikely(!test_and_clear_bit(tag
, bqt
->tag_map
))) {
1115 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
1123 EXPORT_SYMBOL(blk_queue_end_tag
);
1126 * blk_queue_start_tag - find a free tag and assign it
1127 * @q: the request queue for the device
1128 * @rq: the block request that needs tagging
1131 * This can either be used as a stand-alone helper, or possibly be
1132 * assigned as the queue &prep_rq_fn (in which case &struct request
1133 * automagically gets a tag assigned). Note that this function
1134 * assumes that any type of request can be queued! if this is not
1135 * true for your device, you must check the request type before
1136 * calling this function. The request will also be removed from
1137 * the request queue, so it's the drivers responsibility to readd
1138 * it if it should need to be restarted for some reason.
1141 * queue lock must be held.
1143 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
1145 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1148 if (unlikely((rq
->cmd_flags
& REQ_QUEUED
))) {
1150 "%s: request %p for device [%s] already tagged %d",
1152 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
1157 * Protect against shared tag maps, as we may not have exclusive
1158 * access to the tag map.
1161 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
1162 if (tag
>= bqt
->max_depth
)
1165 } while (test_and_set_bit(tag
, bqt
->tag_map
));
1167 rq
->cmd_flags
|= REQ_QUEUED
;
1169 bqt
->tag_index
[tag
] = rq
;
1170 blkdev_dequeue_request(rq
);
1171 list_add(&rq
->queuelist
, &q
->tag_busy_list
);
1176 EXPORT_SYMBOL(blk_queue_start_tag
);
1179 * blk_queue_invalidate_tags - invalidate all pending tags
1180 * @q: the request queue for the device
1183 * Hardware conditions may dictate a need to stop all pending requests.
1184 * In this case, we will safely clear the block side of the tag queue and
1185 * readd all requests to the request queue in the right order.
1188 * queue lock must be held.
1190 void blk_queue_invalidate_tags(request_queue_t
*q
)
1192 struct list_head
*tmp
, *n
;
1195 list_for_each_safe(tmp
, n
, &q
->tag_busy_list
) {
1196 rq
= list_entry_rq(tmp
);
1198 if (rq
->tag
== -1) {
1200 "%s: bad tag found on list\n", __FUNCTION__
);
1201 list_del_init(&rq
->queuelist
);
1202 rq
->cmd_flags
&= ~REQ_QUEUED
;
1204 blk_queue_end_tag(q
, rq
);
1206 rq
->cmd_flags
&= ~REQ_STARTED
;
1207 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1211 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1213 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1217 printk("%s: dev %s: type=%x, flags=%x\n", msg
,
1218 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
1221 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1223 rq
->current_nr_sectors
);
1224 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1226 if (blk_pc_request(rq
)) {
1228 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1229 printk("%02x ", rq
->cmd
[bit
]);
1234 EXPORT_SYMBOL(blk_dump_rq_flags
);
1236 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1238 struct bio_vec
*bv
, *bvprv
= NULL
;
1239 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1240 int high
, highprv
= 1;
1242 if (unlikely(!bio
->bi_io_vec
))
1245 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1246 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1247 bio_for_each_segment(bv
, bio
, i
) {
1249 * the trick here is making sure that a high page is never
1250 * considered part of another segment, since that might
1251 * change with the bounce page.
1253 high
= page_to_pfn(bv
->bv_page
) > q
->bounce_pfn
;
1254 if (high
|| highprv
)
1255 goto new_hw_segment
;
1257 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1259 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1261 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1263 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1264 goto new_hw_segment
;
1266 seg_size
+= bv
->bv_len
;
1267 hw_seg_size
+= bv
->bv_len
;
1272 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1273 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1274 hw_seg_size
+= bv
->bv_len
;
1277 if (hw_seg_size
> bio
->bi_hw_front_size
)
1278 bio
->bi_hw_front_size
= hw_seg_size
;
1279 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1285 seg_size
= bv
->bv_len
;
1288 if (hw_seg_size
> bio
->bi_hw_back_size
)
1289 bio
->bi_hw_back_size
= hw_seg_size
;
1290 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1291 bio
->bi_hw_front_size
= hw_seg_size
;
1292 bio
->bi_phys_segments
= nr_phys_segs
;
1293 bio
->bi_hw_segments
= nr_hw_segs
;
1294 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1296 EXPORT_SYMBOL(blk_recount_segments
);
1298 static int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1301 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1304 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1306 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1310 * bio and nxt are contigous in memory, check if the queue allows
1311 * these two to be merged into one
1313 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1319 static int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1322 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1323 blk_recount_segments(q
, bio
);
1324 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1325 blk_recount_segments(q
, nxt
);
1326 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1327 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1329 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1336 * map a request to scatterlist, return number of sg entries setup. Caller
1337 * must make sure sg can hold rq->nr_phys_segments entries
1339 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1341 struct bio_vec
*bvec
, *bvprv
;
1343 int nsegs
, i
, cluster
;
1346 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1349 * for each bio in rq
1352 rq_for_each_bio(bio
, rq
) {
1354 * for each segment in bio
1356 bio_for_each_segment(bvec
, bio
, i
) {
1357 int nbytes
= bvec
->bv_len
;
1359 if (bvprv
&& cluster
) {
1360 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1363 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1365 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1368 sg
[nsegs
- 1].length
+= nbytes
;
1371 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1372 sg
[nsegs
].page
= bvec
->bv_page
;
1373 sg
[nsegs
].length
= nbytes
;
1374 sg
[nsegs
].offset
= bvec
->bv_offset
;
1379 } /* segments in bio */
1385 EXPORT_SYMBOL(blk_rq_map_sg
);
1388 * the standard queue merge functions, can be overridden with device
1389 * specific ones if so desired
1392 static inline int ll_new_mergeable(request_queue_t
*q
,
1393 struct request
*req
,
1396 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1398 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1399 req
->cmd_flags
|= REQ_NOMERGE
;
1400 if (req
== q
->last_merge
)
1401 q
->last_merge
= NULL
;
1406 * A hw segment is just getting larger, bump just the phys
1409 req
->nr_phys_segments
+= nr_phys_segs
;
1413 static inline int ll_new_hw_segment(request_queue_t
*q
,
1414 struct request
*req
,
1417 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1418 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1420 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1421 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1422 req
->cmd_flags
|= REQ_NOMERGE
;
1423 if (req
== q
->last_merge
)
1424 q
->last_merge
= NULL
;
1429 * This will form the start of a new hw segment. Bump both
1432 req
->nr_hw_segments
+= nr_hw_segs
;
1433 req
->nr_phys_segments
+= nr_phys_segs
;
1437 int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
, struct bio
*bio
)
1439 unsigned short max_sectors
;
1442 if (unlikely(blk_pc_request(req
)))
1443 max_sectors
= q
->max_hw_sectors
;
1445 max_sectors
= q
->max_sectors
;
1447 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1448 req
->cmd_flags
|= REQ_NOMERGE
;
1449 if (req
== q
->last_merge
)
1450 q
->last_merge
= NULL
;
1453 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1454 blk_recount_segments(q
, req
->biotail
);
1455 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1456 blk_recount_segments(q
, bio
);
1457 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1458 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1459 !BIOVEC_VIRT_OVERSIZE(len
)) {
1460 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1463 if (req
->nr_hw_segments
== 1)
1464 req
->bio
->bi_hw_front_size
= len
;
1465 if (bio
->bi_hw_segments
== 1)
1466 bio
->bi_hw_back_size
= len
;
1471 return ll_new_hw_segment(q
, req
, bio
);
1473 EXPORT_SYMBOL(ll_back_merge_fn
);
1475 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1478 unsigned short max_sectors
;
1481 if (unlikely(blk_pc_request(req
)))
1482 max_sectors
= q
->max_hw_sectors
;
1484 max_sectors
= q
->max_sectors
;
1487 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1488 req
->cmd_flags
|= REQ_NOMERGE
;
1489 if (req
== q
->last_merge
)
1490 q
->last_merge
= NULL
;
1493 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1494 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1495 blk_recount_segments(q
, bio
);
1496 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1497 blk_recount_segments(q
, req
->bio
);
1498 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1499 !BIOVEC_VIRT_OVERSIZE(len
)) {
1500 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1503 if (bio
->bi_hw_segments
== 1)
1504 bio
->bi_hw_front_size
= len
;
1505 if (req
->nr_hw_segments
== 1)
1506 req
->biotail
->bi_hw_back_size
= len
;
1511 return ll_new_hw_segment(q
, req
, bio
);
1514 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1515 struct request
*next
)
1517 int total_phys_segments
;
1518 int total_hw_segments
;
1521 * First check if the either of the requests are re-queued
1522 * requests. Can't merge them if they are.
1524 if (req
->special
|| next
->special
)
1528 * Will it become too large?
1530 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1533 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1534 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1535 total_phys_segments
--;
1537 if (total_phys_segments
> q
->max_phys_segments
)
1540 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1541 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1542 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1544 * propagate the combined length to the end of the requests
1546 if (req
->nr_hw_segments
== 1)
1547 req
->bio
->bi_hw_front_size
= len
;
1548 if (next
->nr_hw_segments
== 1)
1549 next
->biotail
->bi_hw_back_size
= len
;
1550 total_hw_segments
--;
1553 if (total_hw_segments
> q
->max_hw_segments
)
1556 /* Merge is OK... */
1557 req
->nr_phys_segments
= total_phys_segments
;
1558 req
->nr_hw_segments
= total_hw_segments
;
1563 * "plug" the device if there are no outstanding requests: this will
1564 * force the transfer to start only after we have put all the requests
1567 * This is called with interrupts off and no requests on the queue and
1568 * with the queue lock held.
1570 void blk_plug_device(request_queue_t
*q
)
1572 WARN_ON(!irqs_disabled());
1575 * don't plug a stopped queue, it must be paired with blk_start_queue()
1576 * which will restart the queueing
1578 if (blk_queue_stopped(q
))
1581 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
1582 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1583 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
1587 EXPORT_SYMBOL(blk_plug_device
);
1590 * remove the queue from the plugged list, if present. called with
1591 * queue lock held and interrupts disabled.
1593 int blk_remove_plug(request_queue_t
*q
)
1595 WARN_ON(!irqs_disabled());
1597 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1600 del_timer(&q
->unplug_timer
);
1604 EXPORT_SYMBOL(blk_remove_plug
);
1607 * remove the plug and let it rip..
1609 void __generic_unplug_device(request_queue_t
*q
)
1611 if (unlikely(blk_queue_stopped(q
)))
1614 if (!blk_remove_plug(q
))
1619 EXPORT_SYMBOL(__generic_unplug_device
);
1622 * generic_unplug_device - fire a request queue
1623 * @q: The &request_queue_t in question
1626 * Linux uses plugging to build bigger requests queues before letting
1627 * the device have at them. If a queue is plugged, the I/O scheduler
1628 * is still adding and merging requests on the queue. Once the queue
1629 * gets unplugged, the request_fn defined for the queue is invoked and
1630 * transfers started.
1632 void generic_unplug_device(request_queue_t
*q
)
1634 spin_lock_irq(q
->queue_lock
);
1635 __generic_unplug_device(q
);
1636 spin_unlock_irq(q
->queue_lock
);
1638 EXPORT_SYMBOL(generic_unplug_device
);
1640 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1643 request_queue_t
*q
= bdi
->unplug_io_data
;
1646 * devices don't necessarily have an ->unplug_fn defined
1649 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1650 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1656 static void blk_unplug_work(struct work_struct
*work
)
1658 request_queue_t
*q
= container_of(work
, request_queue_t
, unplug_work
);
1660 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1661 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1666 static void blk_unplug_timeout(unsigned long data
)
1668 request_queue_t
*q
= (request_queue_t
*)data
;
1670 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
1671 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1673 kblockd_schedule_work(&q
->unplug_work
);
1677 * blk_start_queue - restart a previously stopped queue
1678 * @q: The &request_queue_t in question
1681 * blk_start_queue() will clear the stop flag on the queue, and call
1682 * the request_fn for the queue if it was in a stopped state when
1683 * entered. Also see blk_stop_queue(). Queue lock must be held.
1685 void blk_start_queue(request_queue_t
*q
)
1687 WARN_ON(!irqs_disabled());
1689 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1692 * one level of recursion is ok and is much faster than kicking
1693 * the unplug handling
1695 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1697 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1700 kblockd_schedule_work(&q
->unplug_work
);
1704 EXPORT_SYMBOL(blk_start_queue
);
1707 * blk_stop_queue - stop a queue
1708 * @q: The &request_queue_t in question
1711 * The Linux block layer assumes that a block driver will consume all
1712 * entries on the request queue when the request_fn strategy is called.
1713 * Often this will not happen, because of hardware limitations (queue
1714 * depth settings). If a device driver gets a 'queue full' response,
1715 * or if it simply chooses not to queue more I/O at one point, it can
1716 * call this function to prevent the request_fn from being called until
1717 * the driver has signalled it's ready to go again. This happens by calling
1718 * blk_start_queue() to restart queue operations. Queue lock must be held.
1720 void blk_stop_queue(request_queue_t
*q
)
1723 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1725 EXPORT_SYMBOL(blk_stop_queue
);
1728 * blk_sync_queue - cancel any pending callbacks on a queue
1732 * The block layer may perform asynchronous callback activity
1733 * on a queue, such as calling the unplug function after a timeout.
1734 * A block device may call blk_sync_queue to ensure that any
1735 * such activity is cancelled, thus allowing it to release resources
1736 * that the callbacks might use. The caller must already have made sure
1737 * that its ->make_request_fn will not re-add plugging prior to calling
1741 void blk_sync_queue(struct request_queue
*q
)
1743 del_timer_sync(&q
->unplug_timer
);
1745 EXPORT_SYMBOL(blk_sync_queue
);
1748 * blk_run_queue - run a single device queue
1749 * @q: The queue to run
1751 void blk_run_queue(struct request_queue
*q
)
1753 unsigned long flags
;
1755 spin_lock_irqsave(q
->queue_lock
, flags
);
1759 * Only recurse once to avoid overrunning the stack, let the unplug
1760 * handling reinvoke the handler shortly if we already got there.
1762 if (!elv_queue_empty(q
)) {
1763 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1765 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1768 kblockd_schedule_work(&q
->unplug_work
);
1772 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1774 EXPORT_SYMBOL(blk_run_queue
);
1777 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1778 * @kobj: the kobj belonging of the request queue to be released
1781 * blk_cleanup_queue is the pair to blk_init_queue() or
1782 * blk_queue_make_request(). It should be called when a request queue is
1783 * being released; typically when a block device is being de-registered.
1784 * Currently, its primary task it to free all the &struct request
1785 * structures that were allocated to the queue and the queue itself.
1788 * Hopefully the low level driver will have finished any
1789 * outstanding requests first...
1791 static void blk_release_queue(struct kobject
*kobj
)
1793 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
1794 struct request_list
*rl
= &q
->rq
;
1799 mempool_destroy(rl
->rq_pool
);
1802 __blk_queue_free_tags(q
);
1804 blk_trace_shutdown(q
);
1806 kmem_cache_free(requestq_cachep
, q
);
1809 void blk_put_queue(request_queue_t
*q
)
1811 kobject_put(&q
->kobj
);
1813 EXPORT_SYMBOL(blk_put_queue
);
1815 void blk_cleanup_queue(request_queue_t
* q
)
1817 mutex_lock(&q
->sysfs_lock
);
1818 set_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
);
1819 mutex_unlock(&q
->sysfs_lock
);
1822 elevator_exit(q
->elevator
);
1827 EXPORT_SYMBOL(blk_cleanup_queue
);
1829 static int blk_init_free_list(request_queue_t
*q
)
1831 struct request_list
*rl
= &q
->rq
;
1833 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1834 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1836 init_waitqueue_head(&rl
->wait
[READ
]);
1837 init_waitqueue_head(&rl
->wait
[WRITE
]);
1839 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1840 mempool_free_slab
, request_cachep
, q
->node
);
1848 request_queue_t
*blk_alloc_queue(gfp_t gfp_mask
)
1850 return blk_alloc_queue_node(gfp_mask
, -1);
1852 EXPORT_SYMBOL(blk_alloc_queue
);
1854 static struct kobj_type queue_ktype
;
1856 request_queue_t
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1860 q
= kmem_cache_alloc_node(requestq_cachep
, gfp_mask
, node_id
);
1864 memset(q
, 0, sizeof(*q
));
1865 init_timer(&q
->unplug_timer
);
1867 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
1868 q
->kobj
.ktype
= &queue_ktype
;
1869 kobject_init(&q
->kobj
);
1871 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1872 q
->backing_dev_info
.unplug_io_data
= q
;
1874 mutex_init(&q
->sysfs_lock
);
1878 EXPORT_SYMBOL(blk_alloc_queue_node
);
1881 * blk_init_queue - prepare a request queue for use with a block device
1882 * @rfn: The function to be called to process requests that have been
1883 * placed on the queue.
1884 * @lock: Request queue spin lock
1887 * If a block device wishes to use the standard request handling procedures,
1888 * which sorts requests and coalesces adjacent requests, then it must
1889 * call blk_init_queue(). The function @rfn will be called when there
1890 * are requests on the queue that need to be processed. If the device
1891 * supports plugging, then @rfn may not be called immediately when requests
1892 * are available on the queue, but may be called at some time later instead.
1893 * Plugged queues are generally unplugged when a buffer belonging to one
1894 * of the requests on the queue is needed, or due to memory pressure.
1896 * @rfn is not required, or even expected, to remove all requests off the
1897 * queue, but only as many as it can handle at a time. If it does leave
1898 * requests on the queue, it is responsible for arranging that the requests
1899 * get dealt with eventually.
1901 * The queue spin lock must be held while manipulating the requests on the
1902 * request queue; this lock will be taken also from interrupt context, so irq
1903 * disabling is needed for it.
1905 * Function returns a pointer to the initialized request queue, or NULL if
1906 * it didn't succeed.
1909 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1910 * when the block device is deactivated (such as at module unload).
1913 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1915 return blk_init_queue_node(rfn
, lock
, -1);
1917 EXPORT_SYMBOL(blk_init_queue
);
1920 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1922 request_queue_t
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1928 if (blk_init_free_list(q
)) {
1929 kmem_cache_free(requestq_cachep
, q
);
1934 * if caller didn't supply a lock, they get per-queue locking with
1938 spin_lock_init(&q
->__queue_lock
);
1939 lock
= &q
->__queue_lock
;
1942 q
->request_fn
= rfn
;
1943 q
->prep_rq_fn
= NULL
;
1944 q
->unplug_fn
= generic_unplug_device
;
1945 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1946 q
->queue_lock
= lock
;
1948 blk_queue_segment_boundary(q
, 0xffffffff);
1950 blk_queue_make_request(q
, __make_request
);
1951 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1953 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1954 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1956 q
->sg_reserved_size
= INT_MAX
;
1961 if (!elevator_init(q
, NULL
)) {
1962 blk_queue_congestion_threshold(q
);
1969 EXPORT_SYMBOL(blk_init_queue_node
);
1971 int blk_get_queue(request_queue_t
*q
)
1973 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1974 kobject_get(&q
->kobj
);
1981 EXPORT_SYMBOL(blk_get_queue
);
1983 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
1985 if (rq
->cmd_flags
& REQ_ELVPRIV
)
1986 elv_put_request(q
, rq
);
1987 mempool_free(rq
, q
->rq
.rq_pool
);
1990 static struct request
*
1991 blk_alloc_request(request_queue_t
*q
, int rw
, int priv
, gfp_t gfp_mask
)
1993 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1999 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
2000 * see bio.h and blkdev.h
2002 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
2005 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
2006 mempool_free(rq
, q
->rq
.rq_pool
);
2009 rq
->cmd_flags
|= REQ_ELVPRIV
;
2016 * ioc_batching returns true if the ioc is a valid batching request and
2017 * should be given priority access to a request.
2019 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
2025 * Make sure the process is able to allocate at least 1 request
2026 * even if the batch times out, otherwise we could theoretically
2029 return ioc
->nr_batch_requests
== q
->nr_batching
||
2030 (ioc
->nr_batch_requests
> 0
2031 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
2035 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2036 * will cause the process to be a "batcher" on all queues in the system. This
2037 * is the behaviour we want though - once it gets a wakeup it should be given
2040 static void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
2042 if (!ioc
|| ioc_batching(q
, ioc
))
2045 ioc
->nr_batch_requests
= q
->nr_batching
;
2046 ioc
->last_waited
= jiffies
;
2049 static void __freed_request(request_queue_t
*q
, int rw
)
2051 struct request_list
*rl
= &q
->rq
;
2053 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
2054 blk_clear_queue_congested(q
, rw
);
2056 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
2057 if (waitqueue_active(&rl
->wait
[rw
]))
2058 wake_up(&rl
->wait
[rw
]);
2060 blk_clear_queue_full(q
, rw
);
2065 * A request has just been released. Account for it, update the full and
2066 * congestion status, wake up any waiters. Called under q->queue_lock.
2068 static void freed_request(request_queue_t
*q
, int rw
, int priv
)
2070 struct request_list
*rl
= &q
->rq
;
2076 __freed_request(q
, rw
);
2078 if (unlikely(rl
->starved
[rw
^ 1]))
2079 __freed_request(q
, rw
^ 1);
2082 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2084 * Get a free request, queue_lock must be held.
2085 * Returns NULL on failure, with queue_lock held.
2086 * Returns !NULL on success, with queue_lock *not held*.
2088 static struct request
*get_request(request_queue_t
*q
, int rw_flags
,
2089 struct bio
*bio
, gfp_t gfp_mask
)
2091 struct request
*rq
= NULL
;
2092 struct request_list
*rl
= &q
->rq
;
2093 struct io_context
*ioc
= NULL
;
2094 const int rw
= rw_flags
& 0x01;
2095 int may_queue
, priv
;
2097 may_queue
= elv_may_queue(q
, rw_flags
);
2098 if (may_queue
== ELV_MQUEUE_NO
)
2101 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
2102 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
2103 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
2105 * The queue will fill after this allocation, so set
2106 * it as full, and mark this process as "batching".
2107 * This process will be allowed to complete a batch of
2108 * requests, others will be blocked.
2110 if (!blk_queue_full(q
, rw
)) {
2111 ioc_set_batching(q
, ioc
);
2112 blk_set_queue_full(q
, rw
);
2114 if (may_queue
!= ELV_MQUEUE_MUST
2115 && !ioc_batching(q
, ioc
)) {
2117 * The queue is full and the allocating
2118 * process is not a "batcher", and not
2119 * exempted by the IO scheduler
2125 blk_set_queue_congested(q
, rw
);
2129 * Only allow batching queuers to allocate up to 50% over the defined
2130 * limit of requests, otherwise we could have thousands of requests
2131 * allocated with any setting of ->nr_requests
2133 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
2137 rl
->starved
[rw
] = 0;
2139 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
2143 spin_unlock_irq(q
->queue_lock
);
2145 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
2146 if (unlikely(!rq
)) {
2148 * Allocation failed presumably due to memory. Undo anything
2149 * we might have messed up.
2151 * Allocating task should really be put onto the front of the
2152 * wait queue, but this is pretty rare.
2154 spin_lock_irq(q
->queue_lock
);
2155 freed_request(q
, rw
, priv
);
2158 * in the very unlikely event that allocation failed and no
2159 * requests for this direction was pending, mark us starved
2160 * so that freeing of a request in the other direction will
2161 * notice us. another possible fix would be to split the
2162 * rq mempool into READ and WRITE
2165 if (unlikely(rl
->count
[rw
] == 0))
2166 rl
->starved
[rw
] = 1;
2172 * ioc may be NULL here, and ioc_batching will be false. That's
2173 * OK, if the queue is under the request limit then requests need
2174 * not count toward the nr_batch_requests limit. There will always
2175 * be some limit enforced by BLK_BATCH_TIME.
2177 if (ioc_batching(q
, ioc
))
2178 ioc
->nr_batch_requests
--;
2182 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
2188 * No available requests for this queue, unplug the device and wait for some
2189 * requests to become available.
2191 * Called with q->queue_lock held, and returns with it unlocked.
2193 static struct request
*get_request_wait(request_queue_t
*q
, int rw_flags
,
2196 const int rw
= rw_flags
& 0x01;
2199 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
2202 struct request_list
*rl
= &q
->rq
;
2204 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
2205 TASK_UNINTERRUPTIBLE
);
2207 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
2210 struct io_context
*ioc
;
2212 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
2214 __generic_unplug_device(q
);
2215 spin_unlock_irq(q
->queue_lock
);
2219 * After sleeping, we become a "batching" process and
2220 * will be able to allocate at least one request, and
2221 * up to a big batch of them for a small period time.
2222 * See ioc_batching, ioc_set_batching
2224 ioc
= current_io_context(GFP_NOIO
, q
->node
);
2225 ioc_set_batching(q
, ioc
);
2227 spin_lock_irq(q
->queue_lock
);
2229 finish_wait(&rl
->wait
[rw
], &wait
);
2235 struct request
*blk_get_request(request_queue_t
*q
, int rw
, gfp_t gfp_mask
)
2239 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2241 spin_lock_irq(q
->queue_lock
);
2242 if (gfp_mask
& __GFP_WAIT
) {
2243 rq
= get_request_wait(q
, rw
, NULL
);
2245 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2247 spin_unlock_irq(q
->queue_lock
);
2249 /* q->queue_lock is unlocked at this point */
2253 EXPORT_SYMBOL(blk_get_request
);
2256 * blk_start_queueing - initiate dispatch of requests to device
2257 * @q: request queue to kick into gear
2259 * This is basically a helper to remove the need to know whether a queue
2260 * is plugged or not if someone just wants to initiate dispatch of requests
2263 * The queue lock must be held with interrupts disabled.
2265 void blk_start_queueing(request_queue_t
*q
)
2267 if (!blk_queue_plugged(q
))
2270 __generic_unplug_device(q
);
2272 EXPORT_SYMBOL(blk_start_queueing
);
2275 * blk_requeue_request - put a request back on queue
2276 * @q: request queue where request should be inserted
2277 * @rq: request to be inserted
2280 * Drivers often keep queueing requests until the hardware cannot accept
2281 * more, when that condition happens we need to put the request back
2282 * on the queue. Must be called with queue lock held.
2284 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2286 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
2288 if (blk_rq_tagged(rq
))
2289 blk_queue_end_tag(q
, rq
);
2291 elv_requeue_request(q
, rq
);
2294 EXPORT_SYMBOL(blk_requeue_request
);
2297 * blk_insert_request - insert a special request in to a request queue
2298 * @q: request queue where request should be inserted
2299 * @rq: request to be inserted
2300 * @at_head: insert request at head or tail of queue
2301 * @data: private data
2304 * Many block devices need to execute commands asynchronously, so they don't
2305 * block the whole kernel from preemption during request execution. This is
2306 * accomplished normally by inserting aritficial requests tagged as
2307 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2308 * scheduled for actual execution by the request queue.
2310 * We have the option of inserting the head or the tail of the queue.
2311 * Typically we use the tail for new ioctls and so forth. We use the head
2312 * of the queue for things like a QUEUE_FULL message from a device, or a
2313 * host that is unable to accept a particular command.
2315 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2316 int at_head
, void *data
)
2318 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2319 unsigned long flags
;
2322 * tell I/O scheduler that this isn't a regular read/write (ie it
2323 * must not attempt merges on this) and that it acts as a soft
2326 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
2327 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
2331 spin_lock_irqsave(q
->queue_lock
, flags
);
2334 * If command is tagged, release the tag
2336 if (blk_rq_tagged(rq
))
2337 blk_queue_end_tag(q
, rq
);
2339 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2340 __elv_add_request(q
, rq
, where
, 0);
2341 blk_start_queueing(q
);
2342 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2345 EXPORT_SYMBOL(blk_insert_request
);
2347 static int __blk_rq_unmap_user(struct bio
*bio
)
2352 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2353 bio_unmap_user(bio
);
2355 ret
= bio_uncopy_user(bio
);
2361 static int __blk_rq_map_user(request_queue_t
*q
, struct request
*rq
,
2362 void __user
*ubuf
, unsigned int len
)
2364 unsigned long uaddr
;
2365 struct bio
*bio
, *orig_bio
;
2368 reading
= rq_data_dir(rq
) == READ
;
2371 * if alignment requirement is satisfied, map in user pages for
2372 * direct dma. else, set up kernel bounce buffers
2374 uaddr
= (unsigned long) ubuf
;
2375 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2376 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2378 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2381 return PTR_ERR(bio
);
2384 blk_queue_bounce(q
, &bio
);
2387 * We link the bounce buffer in and could have to traverse it
2388 * later so we have to get a ref to prevent it from being freed
2393 blk_rq_bio_prep(q
, rq
, bio
);
2394 else if (!ll_back_merge_fn(q
, rq
, bio
)) {
2398 rq
->biotail
->bi_next
= bio
;
2401 rq
->data_len
+= bio
->bi_size
;
2404 return bio
->bi_size
;
2407 /* if it was boucned we must call the end io function */
2408 bio_endio(bio
, bio
->bi_size
, 0);
2409 __blk_rq_unmap_user(orig_bio
);
2415 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2416 * @q: request queue where request should be inserted
2417 * @rq: request structure to fill
2418 * @ubuf: the user buffer
2419 * @len: length of user data
2422 * Data will be mapped directly for zero copy io, if possible. Otherwise
2423 * a kernel bounce buffer is used.
2425 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2426 * still in process context.
2428 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2429 * before being submitted to the device, as pages mapped may be out of
2430 * reach. It's the callers responsibility to make sure this happens. The
2431 * original bio must be passed back in to blk_rq_unmap_user() for proper
2434 int blk_rq_map_user(request_queue_t
*q
, struct request
*rq
, void __user
*ubuf
,
2437 unsigned long bytes_read
= 0;
2438 struct bio
*bio
= NULL
;
2441 if (len
> (q
->max_hw_sectors
<< 9))
2446 while (bytes_read
!= len
) {
2447 unsigned long map_len
, end
, start
;
2449 map_len
= min_t(unsigned long, len
- bytes_read
, BIO_MAX_SIZE
);
2450 end
= ((unsigned long)ubuf
+ map_len
+ PAGE_SIZE
- 1)
2452 start
= (unsigned long)ubuf
>> PAGE_SHIFT
;
2455 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2456 * pages. If this happens we just lower the requested
2457 * mapping len by a page so that we can fit
2459 if (end
- start
> BIO_MAX_PAGES
)
2460 map_len
-= PAGE_SIZE
;
2462 ret
= __blk_rq_map_user(q
, rq
, ubuf
, map_len
);
2471 rq
->buffer
= rq
->data
= NULL
;
2474 blk_rq_unmap_user(bio
);
2478 EXPORT_SYMBOL(blk_rq_map_user
);
2481 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2482 * @q: request queue where request should be inserted
2483 * @rq: request to map data to
2484 * @iov: pointer to the iovec
2485 * @iov_count: number of elements in the iovec
2486 * @len: I/O byte count
2489 * Data will be mapped directly for zero copy io, if possible. Otherwise
2490 * a kernel bounce buffer is used.
2492 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2493 * still in process context.
2495 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2496 * before being submitted to the device, as pages mapped may be out of
2497 * reach. It's the callers responsibility to make sure this happens. The
2498 * original bio must be passed back in to blk_rq_unmap_user() for proper
2501 int blk_rq_map_user_iov(request_queue_t
*q
, struct request
*rq
,
2502 struct sg_iovec
*iov
, int iov_count
, unsigned int len
)
2506 if (!iov
|| iov_count
<= 0)
2509 /* we don't allow misaligned data like bio_map_user() does. If the
2510 * user is using sg, they're expected to know the alignment constraints
2511 * and respect them accordingly */
2512 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2514 return PTR_ERR(bio
);
2516 if (bio
->bi_size
!= len
) {
2517 bio_endio(bio
, bio
->bi_size
, 0);
2518 bio_unmap_user(bio
);
2523 blk_rq_bio_prep(q
, rq
, bio
);
2524 rq
->buffer
= rq
->data
= NULL
;
2528 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2531 * blk_rq_unmap_user - unmap a request with user data
2532 * @bio: start of bio list
2535 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2536 * supply the original rq->bio from the blk_rq_map_user() return, since
2537 * the io completion may have changed rq->bio.
2539 int blk_rq_unmap_user(struct bio
*bio
)
2541 struct bio
*mapped_bio
;
2546 if (unlikely(bio_flagged(bio
, BIO_BOUNCED
)))
2547 mapped_bio
= bio
->bi_private
;
2549 ret2
= __blk_rq_unmap_user(mapped_bio
);
2555 bio_put(mapped_bio
);
2561 EXPORT_SYMBOL(blk_rq_unmap_user
);
2564 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2565 * @q: request queue where request should be inserted
2566 * @rq: request to fill
2567 * @kbuf: the kernel buffer
2568 * @len: length of user data
2569 * @gfp_mask: memory allocation flags
2571 int blk_rq_map_kern(request_queue_t
*q
, struct request
*rq
, void *kbuf
,
2572 unsigned int len
, gfp_t gfp_mask
)
2576 if (len
> (q
->max_hw_sectors
<< 9))
2581 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2583 return PTR_ERR(bio
);
2585 if (rq_data_dir(rq
) == WRITE
)
2586 bio
->bi_rw
|= (1 << BIO_RW
);
2588 blk_rq_bio_prep(q
, rq
, bio
);
2589 blk_queue_bounce(q
, &rq
->bio
);
2590 rq
->buffer
= rq
->data
= NULL
;
2594 EXPORT_SYMBOL(blk_rq_map_kern
);
2597 * blk_execute_rq_nowait - insert a request into queue for execution
2598 * @q: queue to insert the request in
2599 * @bd_disk: matching gendisk
2600 * @rq: request to insert
2601 * @at_head: insert request at head or tail of queue
2602 * @done: I/O completion handler
2605 * Insert a fully prepared request at the back of the io scheduler queue
2606 * for execution. Don't wait for completion.
2608 void blk_execute_rq_nowait(request_queue_t
*q
, struct gendisk
*bd_disk
,
2609 struct request
*rq
, int at_head
,
2612 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2614 rq
->rq_disk
= bd_disk
;
2615 rq
->cmd_flags
|= REQ_NOMERGE
;
2617 WARN_ON(irqs_disabled());
2618 spin_lock_irq(q
->queue_lock
);
2619 __elv_add_request(q
, rq
, where
, 1);
2620 __generic_unplug_device(q
);
2621 spin_unlock_irq(q
->queue_lock
);
2623 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait
);
2626 * blk_execute_rq - insert a request into queue for execution
2627 * @q: queue to insert the request in
2628 * @bd_disk: matching gendisk
2629 * @rq: request to insert
2630 * @at_head: insert request at head or tail of queue
2633 * Insert a fully prepared request at the back of the io scheduler queue
2634 * for execution and wait for completion.
2636 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2637 struct request
*rq
, int at_head
)
2639 DECLARE_COMPLETION_ONSTACK(wait
);
2640 char sense
[SCSI_SENSE_BUFFERSIZE
];
2644 * we need an extra reference to the request, so we can look at
2645 * it after io completion
2650 memset(sense
, 0, sizeof(sense
));
2655 rq
->end_io_data
= &wait
;
2656 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2657 wait_for_completion(&wait
);
2665 EXPORT_SYMBOL(blk_execute_rq
);
2668 * blkdev_issue_flush - queue a flush
2669 * @bdev: blockdev to issue flush for
2670 * @error_sector: error sector
2673 * Issue a flush for the block device in question. Caller can supply
2674 * room for storing the error offset in case of a flush error, if they
2675 * wish to. Caller must run wait_for_completion() on its own.
2677 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2681 if (bdev
->bd_disk
== NULL
)
2684 q
= bdev_get_queue(bdev
);
2687 if (!q
->issue_flush_fn
)
2690 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2693 EXPORT_SYMBOL(blkdev_issue_flush
);
2695 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2697 int rw
= rq_data_dir(rq
);
2699 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2703 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2705 disk_round_stats(rq
->rq_disk
);
2706 rq
->rq_disk
->in_flight
++;
2711 * add-request adds a request to the linked list.
2712 * queue lock is held and interrupts disabled, as we muck with the
2713 * request queue list.
2715 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2717 drive_stat_acct(req
, req
->nr_sectors
, 1);
2720 * elevator indicated where it wants this request to be
2721 * inserted at elevator_merge time
2723 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2727 * disk_round_stats() - Round off the performance stats on a struct
2730 * The average IO queue length and utilisation statistics are maintained
2731 * by observing the current state of the queue length and the amount of
2732 * time it has been in this state for.
2734 * Normally, that accounting is done on IO completion, but that can result
2735 * in more than a second's worth of IO being accounted for within any one
2736 * second, leading to >100% utilisation. To deal with that, we call this
2737 * function to do a round-off before returning the results when reading
2738 * /proc/diskstats. This accounts immediately for all queue usage up to
2739 * the current jiffies and restarts the counters again.
2741 void disk_round_stats(struct gendisk
*disk
)
2743 unsigned long now
= jiffies
;
2745 if (now
== disk
->stamp
)
2748 if (disk
->in_flight
) {
2749 __disk_stat_add(disk
, time_in_queue
,
2750 disk
->in_flight
* (now
- disk
->stamp
));
2751 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2756 EXPORT_SYMBOL_GPL(disk_round_stats
);
2759 * queue lock must be held
2761 void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2765 if (unlikely(--req
->ref_count
))
2768 elv_completed_request(q
, req
);
2771 * Request may not have originated from ll_rw_blk. if not,
2772 * it didn't come out of our reserved rq pools
2774 if (req
->cmd_flags
& REQ_ALLOCED
) {
2775 int rw
= rq_data_dir(req
);
2776 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
2778 BUG_ON(!list_empty(&req
->queuelist
));
2779 BUG_ON(!hlist_unhashed(&req
->hash
));
2781 blk_free_request(q
, req
);
2782 freed_request(q
, rw
, priv
);
2786 EXPORT_SYMBOL_GPL(__blk_put_request
);
2788 void blk_put_request(struct request
*req
)
2790 unsigned long flags
;
2791 request_queue_t
*q
= req
->q
;
2794 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2795 * following if (q) test.
2798 spin_lock_irqsave(q
->queue_lock
, flags
);
2799 __blk_put_request(q
, req
);
2800 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2804 EXPORT_SYMBOL(blk_put_request
);
2807 * blk_end_sync_rq - executes a completion event on a request
2808 * @rq: request to complete
2809 * @error: end io status of the request
2811 void blk_end_sync_rq(struct request
*rq
, int error
)
2813 struct completion
*waiting
= rq
->end_io_data
;
2815 rq
->end_io_data
= NULL
;
2816 __blk_put_request(rq
->q
, rq
);
2819 * complete last, if this is a stack request the process (and thus
2820 * the rq pointer) could be invalid right after this complete()
2824 EXPORT_SYMBOL(blk_end_sync_rq
);
2827 * Has to be called with the request spinlock acquired
2829 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2830 struct request
*next
)
2832 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2838 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2841 if (rq_data_dir(req
) != rq_data_dir(next
)
2842 || req
->rq_disk
!= next
->rq_disk
2847 * If we are allowed to merge, then append bio list
2848 * from next to rq and release next. merge_requests_fn
2849 * will have updated segment counts, update sector
2852 if (!ll_merge_requests_fn(q
, req
, next
))
2856 * At this point we have either done a back merge
2857 * or front merge. We need the smaller start_time of
2858 * the merged requests to be the current request
2859 * for accounting purposes.
2861 if (time_after(req
->start_time
, next
->start_time
))
2862 req
->start_time
= next
->start_time
;
2864 req
->biotail
->bi_next
= next
->bio
;
2865 req
->biotail
= next
->biotail
;
2867 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2869 elv_merge_requests(q
, req
, next
);
2872 disk_round_stats(req
->rq_disk
);
2873 req
->rq_disk
->in_flight
--;
2876 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2878 __blk_put_request(q
, next
);
2882 static inline int attempt_back_merge(request_queue_t
*q
, struct request
*rq
)
2884 struct request
*next
= elv_latter_request(q
, rq
);
2887 return attempt_merge(q
, rq
, next
);
2892 static inline int attempt_front_merge(request_queue_t
*q
, struct request
*rq
)
2894 struct request
*prev
= elv_former_request(q
, rq
);
2897 return attempt_merge(q
, prev
, rq
);
2902 static void init_request_from_bio(struct request
*req
, struct bio
*bio
)
2904 req
->cmd_type
= REQ_TYPE_FS
;
2907 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2909 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2910 req
->cmd_flags
|= REQ_FAILFAST
;
2913 * REQ_BARRIER implies no merging, but lets make it explicit
2915 if (unlikely(bio_barrier(bio
)))
2916 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2919 req
->cmd_flags
|= REQ_RW_SYNC
;
2920 if (bio_rw_meta(bio
))
2921 req
->cmd_flags
|= REQ_RW_META
;
2924 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
2925 req
->hard_nr_sectors
= req
->nr_sectors
= bio_sectors(bio
);
2926 req
->current_nr_sectors
= req
->hard_cur_sectors
= bio_cur_sectors(bio
);
2927 req
->nr_phys_segments
= bio_phys_segments(req
->q
, bio
);
2928 req
->nr_hw_segments
= bio_hw_segments(req
->q
, bio
);
2929 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2930 req
->bio
= req
->biotail
= bio
;
2931 req
->ioprio
= bio_prio(bio
);
2932 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2933 req
->start_time
= jiffies
;
2936 static int __make_request(request_queue_t
*q
, struct bio
*bio
)
2938 struct request
*req
;
2939 int el_ret
, nr_sectors
, barrier
, err
;
2940 const unsigned short prio
= bio_prio(bio
);
2941 const int sync
= bio_sync(bio
);
2944 nr_sectors
= bio_sectors(bio
);
2947 * low level driver can indicate that it wants pages above a
2948 * certain limit bounced to low memory (ie for highmem, or even
2949 * ISA dma in theory)
2951 blk_queue_bounce(q
, &bio
);
2953 barrier
= bio_barrier(bio
);
2954 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
2959 spin_lock_irq(q
->queue_lock
);
2961 if (unlikely(barrier
) || elv_queue_empty(q
))
2964 el_ret
= elv_merge(q
, &req
, bio
);
2966 case ELEVATOR_BACK_MERGE
:
2967 BUG_ON(!rq_mergeable(req
));
2969 if (!ll_back_merge_fn(q
, req
, bio
))
2972 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
2974 req
->biotail
->bi_next
= bio
;
2976 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2977 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2978 drive_stat_acct(req
, nr_sectors
, 0);
2979 if (!attempt_back_merge(q
, req
))
2980 elv_merged_request(q
, req
, el_ret
);
2983 case ELEVATOR_FRONT_MERGE
:
2984 BUG_ON(!rq_mergeable(req
));
2986 if (!ll_front_merge_fn(q
, req
, bio
))
2989 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
2991 bio
->bi_next
= req
->bio
;
2995 * may not be valid. if the low level driver said
2996 * it didn't need a bounce buffer then it better
2997 * not touch req->buffer either...
2999 req
->buffer
= bio_data(bio
);
3000 req
->current_nr_sectors
= bio_cur_sectors(bio
);
3001 req
->hard_cur_sectors
= req
->current_nr_sectors
;
3002 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
3003 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
3004 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
3005 drive_stat_acct(req
, nr_sectors
, 0);
3006 if (!attempt_front_merge(q
, req
))
3007 elv_merged_request(q
, req
, el_ret
);
3010 /* ELV_NO_MERGE: elevator says don't/can't merge. */
3017 * This sync check and mask will be re-done in init_request_from_bio(),
3018 * but we need to set it earlier to expose the sync flag to the
3019 * rq allocator and io schedulers.
3021 rw_flags
= bio_data_dir(bio
);
3023 rw_flags
|= REQ_RW_SYNC
;
3026 * Grab a free request. This is might sleep but can not fail.
3027 * Returns with the queue unlocked.
3029 req
= get_request_wait(q
, rw_flags
, bio
);
3032 * After dropping the lock and possibly sleeping here, our request
3033 * may now be mergeable after it had proven unmergeable (above).
3034 * We don't worry about that case for efficiency. It won't happen
3035 * often, and the elevators are able to handle it.
3037 init_request_from_bio(req
, bio
);
3039 spin_lock_irq(q
->queue_lock
);
3040 if (elv_queue_empty(q
))
3042 add_request(q
, req
);
3045 __generic_unplug_device(q
);
3047 spin_unlock_irq(q
->queue_lock
);
3051 bio_endio(bio
, nr_sectors
<< 9, err
);
3056 * If bio->bi_dev is a partition, remap the location
3058 static inline void blk_partition_remap(struct bio
*bio
)
3060 struct block_device
*bdev
= bio
->bi_bdev
;
3062 if (bdev
!= bdev
->bd_contains
) {
3063 struct hd_struct
*p
= bdev
->bd_part
;
3064 const int rw
= bio_data_dir(bio
);
3066 p
->sectors
[rw
] += bio_sectors(bio
);
3069 bio
->bi_sector
+= p
->start_sect
;
3070 bio
->bi_bdev
= bdev
->bd_contains
;
3074 static void handle_bad_sector(struct bio
*bio
)
3076 char b
[BDEVNAME_SIZE
];
3078 printk(KERN_INFO
"attempt to access beyond end of device\n");
3079 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
3080 bdevname(bio
->bi_bdev
, b
),
3082 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
3083 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
3085 set_bit(BIO_EOF
, &bio
->bi_flags
);
3088 #ifdef CONFIG_FAIL_MAKE_REQUEST
3090 static DECLARE_FAULT_ATTR(fail_make_request
);
3092 static int __init
setup_fail_make_request(char *str
)
3094 return setup_fault_attr(&fail_make_request
, str
);
3096 __setup("fail_make_request=", setup_fail_make_request
);
3098 static int should_fail_request(struct bio
*bio
)
3100 if ((bio
->bi_bdev
->bd_disk
->flags
& GENHD_FL_FAIL
) ||
3101 (bio
->bi_bdev
->bd_part
&& bio
->bi_bdev
->bd_part
->make_it_fail
))
3102 return should_fail(&fail_make_request
, bio
->bi_size
);
3107 static int __init
fail_make_request_debugfs(void)
3109 return init_fault_attr_dentries(&fail_make_request
,
3110 "fail_make_request");
3113 late_initcall(fail_make_request_debugfs
);
3115 #else /* CONFIG_FAIL_MAKE_REQUEST */
3117 static inline int should_fail_request(struct bio
*bio
)
3122 #endif /* CONFIG_FAIL_MAKE_REQUEST */
3125 * generic_make_request: hand a buffer to its device driver for I/O
3126 * @bio: The bio describing the location in memory and on the device.
3128 * generic_make_request() is used to make I/O requests of block
3129 * devices. It is passed a &struct bio, which describes the I/O that needs
3132 * generic_make_request() does not return any status. The
3133 * success/failure status of the request, along with notification of
3134 * completion, is delivered asynchronously through the bio->bi_end_io
3135 * function described (one day) else where.
3137 * The caller of generic_make_request must make sure that bi_io_vec
3138 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3139 * set to describe the device address, and the
3140 * bi_end_io and optionally bi_private are set to describe how
3141 * completion notification should be signaled.
3143 * generic_make_request and the drivers it calls may use bi_next if this
3144 * bio happens to be merged with someone else, and may change bi_dev and
3145 * bi_sector for remaps as it sees fit. So the values of these fields
3146 * should NOT be depended on after the call to generic_make_request.
3148 static inline void __generic_make_request(struct bio
*bio
)
3152 sector_t old_sector
;
3153 int ret
, nr_sectors
= bio_sectors(bio
);
3157 /* Test device or partition size, when known. */
3158 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3160 sector_t sector
= bio
->bi_sector
;
3162 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
3164 * This may well happen - the kernel calls bread()
3165 * without checking the size of the device, e.g., when
3166 * mounting a device.
3168 handle_bad_sector(bio
);
3174 * Resolve the mapping until finished. (drivers are
3175 * still free to implement/resolve their own stacking
3176 * by explicitly returning 0)
3178 * NOTE: we don't repeat the blk_size check for each new device.
3179 * Stacking drivers are expected to know what they are doing.
3184 char b
[BDEVNAME_SIZE
];
3186 q
= bdev_get_queue(bio
->bi_bdev
);
3189 "generic_make_request: Trying to access "
3190 "nonexistent block-device %s (%Lu)\n",
3191 bdevname(bio
->bi_bdev
, b
),
3192 (long long) bio
->bi_sector
);
3194 bio_endio(bio
, bio
->bi_size
, -EIO
);
3198 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
3199 printk("bio too big device %s (%u > %u)\n",
3200 bdevname(bio
->bi_bdev
, b
),
3206 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
3209 if (should_fail_request(bio
))
3213 * If this device has partitions, remap block n
3214 * of partition p to block n+start(p) of the disk.
3216 blk_partition_remap(bio
);
3218 if (old_sector
!= -1)
3219 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
3222 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
3224 old_sector
= bio
->bi_sector
;
3225 old_dev
= bio
->bi_bdev
->bd_dev
;
3227 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3229 sector_t sector
= bio
->bi_sector
;
3231 if (maxsector
< nr_sectors
||
3232 maxsector
- nr_sectors
< sector
) {
3234 * This may well happen - partitions are not
3235 * checked to make sure they are within the size
3236 * of the whole device.
3238 handle_bad_sector(bio
);
3243 ret
= q
->make_request_fn(q
, bio
);
3248 * We only want one ->make_request_fn to be active at a time,
3249 * else stack usage with stacked devices could be a problem.
3250 * So use current->bio_{list,tail} to keep a list of requests
3251 * submited by a make_request_fn function.
3252 * current->bio_tail is also used as a flag to say if
3253 * generic_make_request is currently active in this task or not.
3254 * If it is NULL, then no make_request is active. If it is non-NULL,
3255 * then a make_request is active, and new requests should be added
3258 void generic_make_request(struct bio
*bio
)
3260 if (current
->bio_tail
) {
3261 /* make_request is active */
3262 *(current
->bio_tail
) = bio
;
3263 bio
->bi_next
= NULL
;
3264 current
->bio_tail
= &bio
->bi_next
;
3267 /* following loop may be a bit non-obvious, and so deserves some
3269 * Before entering the loop, bio->bi_next is NULL (as all callers
3270 * ensure that) so we have a list with a single bio.
3271 * We pretend that we have just taken it off a longer list, so
3272 * we assign bio_list to the next (which is NULL) and bio_tail
3273 * to &bio_list, thus initialising the bio_list of new bios to be
3274 * added. __generic_make_request may indeed add some more bios
3275 * through a recursive call to generic_make_request. If it
3276 * did, we find a non-NULL value in bio_list and re-enter the loop
3277 * from the top. In this case we really did just take the bio
3278 * of the top of the list (no pretending) and so fixup bio_list and
3279 * bio_tail or bi_next, and call into __generic_make_request again.
3281 * The loop was structured like this to make only one call to
3282 * __generic_make_request (which is important as it is large and
3283 * inlined) and to keep the structure simple.
3285 BUG_ON(bio
->bi_next
);
3287 current
->bio_list
= bio
->bi_next
;
3288 if (bio
->bi_next
== NULL
)
3289 current
->bio_tail
= ¤t
->bio_list
;
3291 bio
->bi_next
= NULL
;
3292 __generic_make_request(bio
);
3293 bio
= current
->bio_list
;
3295 current
->bio_tail
= NULL
; /* deactivate */
3298 EXPORT_SYMBOL(generic_make_request
);
3301 * submit_bio: submit a bio to the block device layer for I/O
3302 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3303 * @bio: The &struct bio which describes the I/O
3305 * submit_bio() is very similar in purpose to generic_make_request(), and
3306 * uses that function to do most of the work. Both are fairly rough
3307 * interfaces, @bio must be presetup and ready for I/O.
3310 void submit_bio(int rw
, struct bio
*bio
)
3312 int count
= bio_sectors(bio
);
3314 BIO_BUG_ON(!bio
->bi_size
);
3315 BIO_BUG_ON(!bio
->bi_io_vec
);
3318 count_vm_events(PGPGOUT
, count
);
3320 task_io_account_read(bio
->bi_size
);
3321 count_vm_events(PGPGIN
, count
);
3324 if (unlikely(block_dump
)) {
3325 char b
[BDEVNAME_SIZE
];
3326 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
3327 current
->comm
, current
->pid
,
3328 (rw
& WRITE
) ? "WRITE" : "READ",
3329 (unsigned long long)bio
->bi_sector
,
3330 bdevname(bio
->bi_bdev
,b
));
3333 generic_make_request(bio
);
3336 EXPORT_SYMBOL(submit_bio
);
3338 static void blk_recalc_rq_segments(struct request
*rq
)
3340 struct bio
*bio
, *prevbio
= NULL
;
3341 int nr_phys_segs
, nr_hw_segs
;
3342 unsigned int phys_size
, hw_size
;
3343 request_queue_t
*q
= rq
->q
;
3348 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
3349 rq_for_each_bio(bio
, rq
) {
3350 /* Force bio hw/phys segs to be recalculated. */
3351 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3353 nr_phys_segs
+= bio_phys_segments(q
, bio
);
3354 nr_hw_segs
+= bio_hw_segments(q
, bio
);
3356 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3357 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3359 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
3360 pseg
<= q
->max_segment_size
) {
3362 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3366 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
3367 hseg
<= q
->max_segment_size
) {
3369 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3376 rq
->nr_phys_segments
= nr_phys_segs
;
3377 rq
->nr_hw_segments
= nr_hw_segs
;
3380 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3382 if (blk_fs_request(rq
)) {
3383 rq
->hard_sector
+= nsect
;
3384 rq
->hard_nr_sectors
-= nsect
;
3387 * Move the I/O submission pointers ahead if required.
3389 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3390 (rq
->sector
<= rq
->hard_sector
)) {
3391 rq
->sector
= rq
->hard_sector
;
3392 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3393 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3394 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3395 rq
->buffer
= bio_data(rq
->bio
);
3399 * if total number of sectors is less than the first segment
3400 * size, something has gone terribly wrong
3402 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3403 printk("blk: request botched\n");
3404 rq
->nr_sectors
= rq
->current_nr_sectors
;
3410 * __end_that_request_first - end I/O on a request
3411 * @req: the request being processed
3412 * @error: 0 for success, < 0 for error
3413 * @nr_bytes: number of bytes to complete
3416 * Ends I/O on a number of bytes attached to @req, and sets it up
3417 * for the next range of segments (if any) in the cluster.
3420 * 0 - we are done with this request, call end_that_request_last()
3421 * 1 - still buffers pending for this request
3423 static int __end_that_request_first(struct request
*req
, int error
,
3426 int total_bytes
, bio_nbytes
, next_idx
= 0;
3429 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
3432 * for a REQ_BLOCK_PC request, we want to carry any eventual
3433 * sense key with us all the way through
3435 if (!blk_pc_request(req
))
3439 if (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))
3440 printk("end_request: I/O error, dev %s, sector %llu\n",
3441 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3442 (unsigned long long)req
->sector
);
3445 if (blk_fs_request(req
) && req
->rq_disk
) {
3446 const int rw
= rq_data_dir(req
);
3448 disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3451 total_bytes
= bio_nbytes
= 0;
3452 while ((bio
= req
->bio
) != NULL
) {
3455 if (nr_bytes
>= bio
->bi_size
) {
3456 req
->bio
= bio
->bi_next
;
3457 nbytes
= bio
->bi_size
;
3458 if (!ordered_bio_endio(req
, bio
, nbytes
, error
))
3459 bio_endio(bio
, nbytes
, error
);
3463 int idx
= bio
->bi_idx
+ next_idx
;
3465 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3466 blk_dump_rq_flags(req
, "__end_that");
3467 printk("%s: bio idx %d >= vcnt %d\n",
3469 bio
->bi_idx
, bio
->bi_vcnt
);
3473 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3474 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3477 * not a complete bvec done
3479 if (unlikely(nbytes
> nr_bytes
)) {
3480 bio_nbytes
+= nr_bytes
;
3481 total_bytes
+= nr_bytes
;
3486 * advance to the next vector
3489 bio_nbytes
+= nbytes
;
3492 total_bytes
+= nbytes
;
3495 if ((bio
= req
->bio
)) {
3497 * end more in this run, or just return 'not-done'
3499 if (unlikely(nr_bytes
<= 0))
3511 * if the request wasn't completed, update state
3514 if (!ordered_bio_endio(req
, bio
, bio_nbytes
, error
))
3515 bio_endio(bio
, bio_nbytes
, error
);
3516 bio
->bi_idx
+= next_idx
;
3517 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3518 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3521 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3522 blk_recalc_rq_segments(req
);
3527 * splice the completion data to a local structure and hand off to
3528 * process_completion_queue() to complete the requests
3530 static void blk_done_softirq(struct softirq_action
*h
)
3532 struct list_head
*cpu_list
, local_list
;
3534 local_irq_disable();
3535 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3536 list_replace_init(cpu_list
, &local_list
);
3539 while (!list_empty(&local_list
)) {
3540 struct request
*rq
= list_entry(local_list
.next
, struct request
, donelist
);
3542 list_del_init(&rq
->donelist
);
3543 rq
->q
->softirq_done_fn(rq
);
3547 static int blk_cpu_notify(struct notifier_block
*self
, unsigned long action
,
3551 * If a CPU goes away, splice its entries to the current CPU
3552 * and trigger a run of the softirq
3554 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3555 int cpu
= (unsigned long) hcpu
;
3557 local_irq_disable();
3558 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
3559 &__get_cpu_var(blk_cpu_done
));
3560 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3568 static struct notifier_block __devinitdata blk_cpu_notifier
= {
3569 .notifier_call
= blk_cpu_notify
,
3573 * blk_complete_request - end I/O on a request
3574 * @req: the request being processed
3577 * Ends all I/O on a request. It does not handle partial completions,
3578 * unless the driver actually implements this in its completion callback
3579 * through requeueing. Theh actual completion happens out-of-order,
3580 * through a softirq handler. The user must have registered a completion
3581 * callback through blk_queue_softirq_done().
3584 void blk_complete_request(struct request
*req
)
3586 struct list_head
*cpu_list
;
3587 unsigned long flags
;
3589 BUG_ON(!req
->q
->softirq_done_fn
);
3591 local_irq_save(flags
);
3593 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3594 list_add_tail(&req
->donelist
, cpu_list
);
3595 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3597 local_irq_restore(flags
);
3600 EXPORT_SYMBOL(blk_complete_request
);
3603 * queue lock must be held
3605 static void end_that_request_last(struct request
*req
, int error
)
3607 struct gendisk
*disk
= req
->rq_disk
;
3609 if (blk_rq_tagged(req
))
3610 blk_queue_end_tag(req
->q
, req
);
3612 if (blk_queued_rq(req
))
3613 blkdev_dequeue_request(req
);
3615 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3616 laptop_io_completion();
3619 * Account IO completion. bar_rq isn't accounted as a normal
3620 * IO on queueing nor completion. Accounting the containing
3621 * request is enough.
3623 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
3624 unsigned long duration
= jiffies
- req
->start_time
;
3625 const int rw
= rq_data_dir(req
);
3627 __disk_stat_inc(disk
, ios
[rw
]);
3628 __disk_stat_add(disk
, ticks
[rw
], duration
);
3629 disk_round_stats(disk
);
3634 req
->end_io(req
, error
);
3636 __blk_put_request(req
->q
, req
);
3640 static inline void __end_request(struct request
*rq
, int uptodate
,
3641 unsigned int nr_bytes
)
3646 error
= uptodate
? uptodate
: -EIO
;
3648 __blk_end_request(rq
, error
, nr_bytes
);
3652 * blk_rq_bytes - Returns bytes left to complete in the entire request
3654 unsigned int blk_rq_bytes(struct request
*rq
)
3656 if (blk_fs_request(rq
))
3657 return rq
->hard_nr_sectors
<< 9;
3659 return rq
->data_len
;
3661 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
3664 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
3666 unsigned int blk_rq_cur_bytes(struct request
*rq
)
3668 if (blk_fs_request(rq
))
3669 return rq
->current_nr_sectors
<< 9;
3672 return rq
->bio
->bi_size
;
3674 return rq
->data_len
;
3676 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
3679 * end_queued_request - end all I/O on a queued request
3680 * @rq: the request being processed
3681 * @uptodate: error value or 0/1 uptodate flag
3684 * Ends all I/O on a request, and removes it from the block layer queues.
3685 * Not suitable for normal IO completion, unless the driver still has
3686 * the request attached to the block layer.
3689 void end_queued_request(struct request
*rq
, int uptodate
)
3691 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
3693 EXPORT_SYMBOL(end_queued_request
);
3696 * end_dequeued_request - end all I/O on a dequeued request
3697 * @rq: the request being processed
3698 * @uptodate: error value or 0/1 uptodate flag
3701 * Ends all I/O on a request. The request must already have been
3702 * dequeued using blkdev_dequeue_request(), as is normally the case
3706 void end_dequeued_request(struct request
*rq
, int uptodate
)
3708 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
3710 EXPORT_SYMBOL(end_dequeued_request
);
3714 * end_request - end I/O on the current segment of the request
3715 * @rq: the request being processed
3716 * @uptodate: error value or 0/1 uptodate flag
3719 * Ends I/O on the current segment of a request. If that is the only
3720 * remaining segment, the request is also completed and freed.
3722 * This is a remnant of how older block drivers handled IO completions.
3723 * Modern drivers typically end IO on the full request in one go, unless
3724 * they have a residual value to account for. For that case this function
3725 * isn't really useful, unless the residual just happens to be the
3726 * full current segment. In other words, don't use this function in new
3727 * code. Either use end_request_completely(), or the
3728 * end_that_request_chunk() (along with end_that_request_last()) for
3729 * partial completions.
3732 void end_request(struct request
*req
, int uptodate
)
3734 __end_request(req
, uptodate
, req
->hard_cur_sectors
<< 9);
3736 EXPORT_SYMBOL(end_request
);
3739 * blk_end_request - Helper function for drivers to complete the request.
3740 * @rq: the request being processed
3741 * @error: 0 for success, < 0 for error
3742 * @nr_bytes: number of bytes to complete
3745 * Ends I/O on a number of bytes attached to @rq.
3746 * If @rq has leftover, sets it up for the next range of segments.
3749 * 0 - we are done with this request
3750 * 1 - still buffers pending for this request
3752 int blk_end_request(struct request
*rq
, int error
, int nr_bytes
)
3754 struct request_queue
*q
= rq
->q
;
3755 unsigned long flags
= 0UL;
3757 if (blk_fs_request(rq
) || blk_pc_request(rq
)) {
3758 if (__end_that_request_first(rq
, error
, nr_bytes
))
3762 add_disk_randomness(rq
->rq_disk
);
3764 spin_lock_irqsave(q
->queue_lock
, flags
);
3765 end_that_request_last(rq
, error
);
3766 spin_unlock_irqrestore(q
->queue_lock
, flags
);
3770 EXPORT_SYMBOL_GPL(blk_end_request
);
3773 * __blk_end_request - Helper function for drivers to complete the request.
3774 * @rq: the request being processed
3775 * @error: 0 for success, < 0 for error
3776 * @nr_bytes: number of bytes to complete
3779 * Must be called with queue lock held unlike blk_end_request().
3782 * 0 - we are done with this request
3783 * 1 - still buffers pending for this request
3785 int __blk_end_request(struct request
*rq
, int error
, int nr_bytes
)
3787 if (blk_fs_request(rq
) || blk_pc_request(rq
)) {
3788 if (__end_that_request_first(rq
, error
, nr_bytes
))
3792 add_disk_randomness(rq
->rq_disk
);
3794 end_that_request_last(rq
, error
);
3798 EXPORT_SYMBOL_GPL(__blk_end_request
);
3800 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3802 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3803 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
3805 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3806 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3807 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3808 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3809 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3810 rq
->buffer
= bio_data(bio
);
3811 rq
->data_len
= bio
->bi_size
;
3813 rq
->bio
= rq
->biotail
= bio
;
3816 EXPORT_SYMBOL(blk_rq_bio_prep
);
3818 int kblockd_schedule_work(struct work_struct
*work
)
3820 return queue_work(kblockd_workqueue
, work
);
3823 EXPORT_SYMBOL(kblockd_schedule_work
);
3825 void kblockd_flush_work(struct work_struct
*work
)
3827 cancel_work_sync(work
);
3829 EXPORT_SYMBOL(kblockd_flush_work
);
3831 int __init
blk_dev_init(void)
3835 kblockd_workqueue
= create_workqueue("kblockd");
3836 if (!kblockd_workqueue
)
3837 panic("Failed to create kblockd\n");
3839 request_cachep
= kmem_cache_create("blkdev_requests",
3840 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3842 requestq_cachep
= kmem_cache_create("blkdev_queue",
3843 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3845 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3846 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3848 for_each_possible_cpu(i
)
3849 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
3851 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
3852 register_hotcpu_notifier(&blk_cpu_notifier
);
3854 blk_max_low_pfn
= max_low_pfn
- 1;
3855 blk_max_pfn
= max_pfn
- 1;
3861 * IO Context helper functions
3863 void put_io_context(struct io_context
*ioc
)
3868 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3870 if (atomic_dec_and_test(&ioc
->refcount
)) {
3871 struct cfq_io_context
*cic
;
3874 if (ioc
->aic
&& ioc
->aic
->dtor
)
3875 ioc
->aic
->dtor(ioc
->aic
);
3876 if (ioc
->cic_root
.rb_node
!= NULL
) {
3877 struct rb_node
*n
= rb_first(&ioc
->cic_root
);
3879 cic
= rb_entry(n
, struct cfq_io_context
, rb_node
);
3884 kmem_cache_free(iocontext_cachep
, ioc
);
3887 EXPORT_SYMBOL(put_io_context
);
3889 /* Called by the exitting task */
3890 void exit_io_context(void)
3892 struct io_context
*ioc
;
3893 struct cfq_io_context
*cic
;
3896 ioc
= current
->io_context
;
3897 current
->io_context
= NULL
;
3898 task_unlock(current
);
3901 if (ioc
->aic
&& ioc
->aic
->exit
)
3902 ioc
->aic
->exit(ioc
->aic
);
3903 if (ioc
->cic_root
.rb_node
!= NULL
) {
3904 cic
= rb_entry(rb_first(&ioc
->cic_root
), struct cfq_io_context
, rb_node
);
3908 put_io_context(ioc
);
3912 * If the current task has no IO context then create one and initialise it.
3913 * Otherwise, return its existing IO context.
3915 * This returned IO context doesn't have a specifically elevated refcount,
3916 * but since the current task itself holds a reference, the context can be
3917 * used in general code, so long as it stays within `current` context.
3919 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
)
3921 struct task_struct
*tsk
= current
;
3922 struct io_context
*ret
;
3924 ret
= tsk
->io_context
;
3928 ret
= kmem_cache_alloc_node(iocontext_cachep
, gfp_flags
, node
);
3930 atomic_set(&ret
->refcount
, 1);
3931 ret
->task
= current
;
3932 ret
->ioprio_changed
= 0;
3933 ret
->last_waited
= jiffies
; /* doesn't matter... */
3934 ret
->nr_batch_requests
= 0; /* because this is 0 */
3936 ret
->cic_root
.rb_node
= NULL
;
3937 ret
->ioc_data
= NULL
;
3938 /* make sure set_task_ioprio() sees the settings above */
3940 tsk
->io_context
= ret
;
3947 * If the current task has no IO context then create one and initialise it.
3948 * If it does have a context, take a ref on it.
3950 * This is always called in the context of the task which submitted the I/O.
3952 struct io_context
*get_io_context(gfp_t gfp_flags
, int node
)
3954 struct io_context
*ret
;
3955 ret
= current_io_context(gfp_flags
, node
);
3957 atomic_inc(&ret
->refcount
);
3960 EXPORT_SYMBOL(get_io_context
);
3962 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3964 struct io_context
*src
= *psrc
;
3965 struct io_context
*dst
= *pdst
;
3968 BUG_ON(atomic_read(&src
->refcount
) == 0);
3969 atomic_inc(&src
->refcount
);
3970 put_io_context(dst
);
3974 EXPORT_SYMBOL(copy_io_context
);
3976 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3978 struct io_context
*temp
;
3983 EXPORT_SYMBOL(swap_io_context
);
3988 struct queue_sysfs_entry
{
3989 struct attribute attr
;
3990 ssize_t (*show
)(struct request_queue
*, char *);
3991 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3995 queue_var_show(unsigned int var
, char *page
)
3997 return sprintf(page
, "%d\n", var
);
4001 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
4003 char *p
= (char *) page
;
4005 *var
= simple_strtoul(p
, &p
, 10);
4009 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
4011 return queue_var_show(q
->nr_requests
, (page
));
4015 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
4017 struct request_list
*rl
= &q
->rq
;
4019 int ret
= queue_var_store(&nr
, page
, count
);
4020 if (nr
< BLKDEV_MIN_RQ
)
4023 spin_lock_irq(q
->queue_lock
);
4024 q
->nr_requests
= nr
;
4025 blk_queue_congestion_threshold(q
);
4027 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
4028 blk_set_queue_congested(q
, READ
);
4029 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
4030 blk_clear_queue_congested(q
, READ
);
4032 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
4033 blk_set_queue_congested(q
, WRITE
);
4034 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
4035 blk_clear_queue_congested(q
, WRITE
);
4037 if (rl
->count
[READ
] >= q
->nr_requests
) {
4038 blk_set_queue_full(q
, READ
);
4039 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
4040 blk_clear_queue_full(q
, READ
);
4041 wake_up(&rl
->wait
[READ
]);
4044 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
4045 blk_set_queue_full(q
, WRITE
);
4046 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
4047 blk_clear_queue_full(q
, WRITE
);
4048 wake_up(&rl
->wait
[WRITE
]);
4050 spin_unlock_irq(q
->queue_lock
);
4054 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
4056 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
4058 return queue_var_show(ra_kb
, (page
));
4062 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
4064 unsigned long ra_kb
;
4065 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
4067 spin_lock_irq(q
->queue_lock
);
4068 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
4069 spin_unlock_irq(q
->queue_lock
);
4074 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
4076 int max_sectors_kb
= q
->max_sectors
>> 1;
4078 return queue_var_show(max_sectors_kb
, (page
));
4082 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
4084 unsigned long max_sectors_kb
,
4085 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
4086 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
4087 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
4090 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
4093 * Take the queue lock to update the readahead and max_sectors
4094 * values synchronously:
4096 spin_lock_irq(q
->queue_lock
);
4098 * Trim readahead window as well, if necessary:
4100 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
4101 if (ra_kb
> max_sectors_kb
)
4102 q
->backing_dev_info
.ra_pages
=
4103 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
4105 q
->max_sectors
= max_sectors_kb
<< 1;
4106 spin_unlock_irq(q
->queue_lock
);
4111 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
4113 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
4115 return queue_var_show(max_hw_sectors_kb
, (page
));
4119 static struct queue_sysfs_entry queue_requests_entry
= {
4120 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
4121 .show
= queue_requests_show
,
4122 .store
= queue_requests_store
,
4125 static struct queue_sysfs_entry queue_ra_entry
= {
4126 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
4127 .show
= queue_ra_show
,
4128 .store
= queue_ra_store
,
4131 static struct queue_sysfs_entry queue_max_sectors_entry
= {
4132 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
4133 .show
= queue_max_sectors_show
,
4134 .store
= queue_max_sectors_store
,
4137 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
4138 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
4139 .show
= queue_max_hw_sectors_show
,
4142 static struct queue_sysfs_entry queue_iosched_entry
= {
4143 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
4144 .show
= elv_iosched_show
,
4145 .store
= elv_iosched_store
,
4148 static struct attribute
*default_attrs
[] = {
4149 &queue_requests_entry
.attr
,
4150 &queue_ra_entry
.attr
,
4151 &queue_max_hw_sectors_entry
.attr
,
4152 &queue_max_sectors_entry
.attr
,
4153 &queue_iosched_entry
.attr
,
4157 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4160 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4162 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4163 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
4168 mutex_lock(&q
->sysfs_lock
);
4169 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4170 mutex_unlock(&q
->sysfs_lock
);
4173 res
= entry
->show(q
, page
);
4174 mutex_unlock(&q
->sysfs_lock
);
4179 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4180 const char *page
, size_t length
)
4182 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4183 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
4189 mutex_lock(&q
->sysfs_lock
);
4190 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4191 mutex_unlock(&q
->sysfs_lock
);
4194 res
= entry
->store(q
, page
, length
);
4195 mutex_unlock(&q
->sysfs_lock
);
4199 static struct sysfs_ops queue_sysfs_ops
= {
4200 .show
= queue_attr_show
,
4201 .store
= queue_attr_store
,
4204 static struct kobj_type queue_ktype
= {
4205 .sysfs_ops
= &queue_sysfs_ops
,
4206 .default_attrs
= default_attrs
,
4207 .release
= blk_release_queue
,
4210 int blk_register_queue(struct gendisk
*disk
)
4214 request_queue_t
*q
= disk
->queue
;
4216 if (!q
|| !q
->request_fn
)
4219 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
4221 ret
= kobject_add(&q
->kobj
);
4225 kobject_uevent(&q
->kobj
, KOBJ_ADD
);
4227 ret
= elv_register_queue(q
);
4229 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4230 kobject_del(&q
->kobj
);
4237 void blk_unregister_queue(struct gendisk
*disk
)
4239 request_queue_t
*q
= disk
->queue
;
4241 if (q
&& q
->request_fn
) {
4242 elv_unregister_queue(q
);
4244 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4245 kobject_del(&q
->kobj
);
4246 kobject_put(&disk
->kobj
);