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>
33 #include <linux/scatterlist.h>
38 #include <scsi/scsi_cmnd.h>
40 static void blk_unplug_work(struct work_struct
*work
);
41 static void blk_unplug_timeout(unsigned long data
);
42 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
43 static void init_request_from_bio(struct request
*req
, struct bio
*bio
);
44 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
45 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
);
46 static void blk_recalc_rq_segments(struct request
*rq
);
47 static void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
51 * For the allocated request tables
53 static struct kmem_cache
*request_cachep
;
56 * For queue allocation
58 static struct kmem_cache
*requestq_cachep
;
61 * For io context allocations
63 static struct kmem_cache
*iocontext_cachep
;
66 * Controlling structure to kblockd
68 static struct workqueue_struct
*kblockd_workqueue
;
70 unsigned long blk_max_low_pfn
, blk_max_pfn
;
72 EXPORT_SYMBOL(blk_max_low_pfn
);
73 EXPORT_SYMBOL(blk_max_pfn
);
75 static DEFINE_PER_CPU(struct list_head
, blk_cpu_done
);
77 /* Amount of time in which a process may batch requests */
78 #define BLK_BATCH_TIME (HZ/50UL)
80 /* Number of requests a "batching" process may submit */
81 #define BLK_BATCH_REQ 32
84 * Return the threshold (number of used requests) at which the queue is
85 * considered to be congested. It include a little hysteresis to keep the
86 * context switch rate down.
88 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
90 return q
->nr_congestion_on
;
94 * The threshold at which a queue is considered to be uncongested
96 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
98 return q
->nr_congestion_off
;
101 static void blk_queue_congestion_threshold(struct request_queue
*q
)
105 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
106 if (nr
> q
->nr_requests
)
108 q
->nr_congestion_on
= nr
;
110 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
113 q
->nr_congestion_off
= nr
;
117 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
120 * Locates the passed device's request queue and returns the address of its
123 * Will return NULL if the request queue cannot be located.
125 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
127 struct backing_dev_info
*ret
= NULL
;
128 struct request_queue
*q
= bdev_get_queue(bdev
);
131 ret
= &q
->backing_dev_info
;
134 EXPORT_SYMBOL(blk_get_backing_dev_info
);
137 * blk_queue_prep_rq - set a prepare_request function for queue
139 * @pfn: prepare_request function
141 * It's possible for a queue to register a prepare_request callback which
142 * is invoked before the request is handed to the request_fn. The goal of
143 * the function is to prepare a request for I/O, it can be used to build a
144 * cdb from the request data for instance.
147 void blk_queue_prep_rq(struct request_queue
*q
, prep_rq_fn
*pfn
)
152 EXPORT_SYMBOL(blk_queue_prep_rq
);
155 * blk_queue_merge_bvec - set a merge_bvec function for queue
157 * @mbfn: merge_bvec_fn
159 * Usually queues have static limitations on the max sectors or segments that
160 * we can put in a request. Stacking drivers may have some settings that
161 * are dynamic, and thus we have to query the queue whether it is ok to
162 * add a new bio_vec to a bio at a given offset or not. If the block device
163 * has such limitations, it needs to register a merge_bvec_fn to control
164 * the size of bio's sent to it. Note that a block device *must* allow a
165 * single page to be added to an empty bio. The block device driver may want
166 * to use the bio_split() function to deal with these bio's. By default
167 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
170 void blk_queue_merge_bvec(struct request_queue
*q
, merge_bvec_fn
*mbfn
)
172 q
->merge_bvec_fn
= mbfn
;
175 EXPORT_SYMBOL(blk_queue_merge_bvec
);
177 void blk_queue_softirq_done(struct request_queue
*q
, softirq_done_fn
*fn
)
179 q
->softirq_done_fn
= fn
;
182 EXPORT_SYMBOL(blk_queue_softirq_done
);
185 * blk_queue_make_request - define an alternate make_request function for a device
186 * @q: the request queue for the device to be affected
187 * @mfn: the alternate make_request function
190 * The normal way for &struct bios to be passed to a device
191 * driver is for them to be collected into requests on a request
192 * queue, and then to allow the device driver to select requests
193 * off that queue when it is ready. This works well for many block
194 * devices. However some block devices (typically virtual devices
195 * such as md or lvm) do not benefit from the processing on the
196 * request queue, and are served best by having the requests passed
197 * directly to them. This can be achieved by providing a function
198 * to blk_queue_make_request().
201 * The driver that does this *must* be able to deal appropriately
202 * with buffers in "highmemory". This can be accomplished by either calling
203 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
204 * blk_queue_bounce() to create a buffer in normal memory.
206 void blk_queue_make_request(struct request_queue
* q
, make_request_fn
* mfn
)
211 q
->nr_requests
= BLKDEV_MAX_RQ
;
212 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
213 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
214 q
->make_request_fn
= mfn
;
215 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
216 q
->backing_dev_info
.state
= 0;
217 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
218 blk_queue_max_sectors(q
, SAFE_MAX_SECTORS
);
219 blk_queue_hardsect_size(q
, 512);
220 blk_queue_dma_alignment(q
, 511);
221 blk_queue_congestion_threshold(q
);
222 q
->nr_batching
= BLK_BATCH_REQ
;
224 q
->unplug_thresh
= 4; /* hmm */
225 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
226 if (q
->unplug_delay
== 0)
229 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
231 q
->unplug_timer
.function
= blk_unplug_timeout
;
232 q
->unplug_timer
.data
= (unsigned long)q
;
235 * by default assume old behaviour and bounce for any highmem page
237 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
240 EXPORT_SYMBOL(blk_queue_make_request
);
242 static void rq_init(struct request_queue
*q
, struct request
*rq
)
244 INIT_LIST_HEAD(&rq
->queuelist
);
245 INIT_LIST_HEAD(&rq
->donelist
);
248 rq
->bio
= rq
->biotail
= NULL
;
249 INIT_HLIST_NODE(&rq
->hash
);
250 RB_CLEAR_NODE(&rq
->rb_node
);
258 rq
->nr_phys_segments
= 0;
261 rq
->end_io_data
= NULL
;
262 rq
->completion_data
= NULL
;
267 * blk_queue_ordered - does this queue support ordered writes
268 * @q: the request queue
269 * @ordered: one of QUEUE_ORDERED_*
270 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
273 * For journalled file systems, doing ordered writes on a commit
274 * block instead of explicitly doing wait_on_buffer (which is bad
275 * for performance) can be a big win. Block drivers supporting this
276 * feature should call this function and indicate so.
279 int blk_queue_ordered(struct request_queue
*q
, unsigned ordered
,
280 prepare_flush_fn
*prepare_flush_fn
)
282 if (ordered
& (QUEUE_ORDERED_PREFLUSH
| QUEUE_ORDERED_POSTFLUSH
) &&
283 prepare_flush_fn
== NULL
) {
284 printk(KERN_ERR
"blk_queue_ordered: prepare_flush_fn required\n");
288 if (ordered
!= QUEUE_ORDERED_NONE
&&
289 ordered
!= QUEUE_ORDERED_DRAIN
&&
290 ordered
!= QUEUE_ORDERED_DRAIN_FLUSH
&&
291 ordered
!= QUEUE_ORDERED_DRAIN_FUA
&&
292 ordered
!= QUEUE_ORDERED_TAG
&&
293 ordered
!= QUEUE_ORDERED_TAG_FLUSH
&&
294 ordered
!= QUEUE_ORDERED_TAG_FUA
) {
295 printk(KERN_ERR
"blk_queue_ordered: bad value %d\n", ordered
);
299 q
->ordered
= ordered
;
300 q
->next_ordered
= ordered
;
301 q
->prepare_flush_fn
= prepare_flush_fn
;
306 EXPORT_SYMBOL(blk_queue_ordered
);
309 * Cache flushing for ordered writes handling
311 inline unsigned blk_ordered_cur_seq(struct request_queue
*q
)
315 return 1 << ffz(q
->ordseq
);
318 unsigned blk_ordered_req_seq(struct request
*rq
)
320 struct request_queue
*q
= rq
->q
;
322 BUG_ON(q
->ordseq
== 0);
324 if (rq
== &q
->pre_flush_rq
)
325 return QUEUE_ORDSEQ_PREFLUSH
;
326 if (rq
== &q
->bar_rq
)
327 return QUEUE_ORDSEQ_BAR
;
328 if (rq
== &q
->post_flush_rq
)
329 return QUEUE_ORDSEQ_POSTFLUSH
;
332 * !fs requests don't need to follow barrier ordering. Always
333 * put them at the front. This fixes the following deadlock.
335 * http://thread.gmane.org/gmane.linux.kernel/537473
337 if (!blk_fs_request(rq
))
338 return QUEUE_ORDSEQ_DRAIN
;
340 if ((rq
->cmd_flags
& REQ_ORDERED_COLOR
) ==
341 (q
->orig_bar_rq
->cmd_flags
& REQ_ORDERED_COLOR
))
342 return QUEUE_ORDSEQ_DRAIN
;
344 return QUEUE_ORDSEQ_DONE
;
347 void blk_ordered_complete_seq(struct request_queue
*q
, unsigned seq
, int error
)
352 if (error
&& !q
->orderr
)
355 BUG_ON(q
->ordseq
& seq
);
358 if (blk_ordered_cur_seq(q
) != QUEUE_ORDSEQ_DONE
)
362 * Okay, sequence complete.
366 uptodate
= q
->orderr
;
371 end_that_request_first(rq
, uptodate
, rq
->hard_nr_sectors
);
372 end_that_request_last(rq
, uptodate
);
375 static void pre_flush_end_io(struct request
*rq
, int error
)
377 elv_completed_request(rq
->q
, rq
);
378 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_PREFLUSH
, error
);
381 static void bar_end_io(struct request
*rq
, int error
)
383 elv_completed_request(rq
->q
, rq
);
384 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_BAR
, error
);
387 static void post_flush_end_io(struct request
*rq
, int error
)
389 elv_completed_request(rq
->q
, rq
);
390 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_POSTFLUSH
, error
);
393 static void queue_flush(struct request_queue
*q
, unsigned which
)
396 rq_end_io_fn
*end_io
;
398 if (which
== QUEUE_ORDERED_PREFLUSH
) {
399 rq
= &q
->pre_flush_rq
;
400 end_io
= pre_flush_end_io
;
402 rq
= &q
->post_flush_rq
;
403 end_io
= post_flush_end_io
;
406 rq
->cmd_flags
= REQ_HARDBARRIER
;
408 rq
->elevator_private
= NULL
;
409 rq
->elevator_private2
= NULL
;
410 rq
->rq_disk
= q
->bar_rq
.rq_disk
;
412 q
->prepare_flush_fn(q
, rq
);
414 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
417 static inline struct request
*start_ordered(struct request_queue
*q
,
421 q
->ordered
= q
->next_ordered
;
422 q
->ordseq
|= QUEUE_ORDSEQ_STARTED
;
425 * Prep proxy barrier request.
427 blkdev_dequeue_request(rq
);
432 if (bio_data_dir(q
->orig_bar_rq
->bio
) == WRITE
)
433 rq
->cmd_flags
|= REQ_RW
;
434 if (q
->ordered
& QUEUE_ORDERED_FUA
)
435 rq
->cmd_flags
|= REQ_FUA
;
436 rq
->elevator_private
= NULL
;
437 rq
->elevator_private2
= NULL
;
438 init_request_from_bio(rq
, q
->orig_bar_rq
->bio
);
439 rq
->end_io
= bar_end_io
;
442 * Queue ordered sequence. As we stack them at the head, we
443 * need to queue in reverse order. Note that we rely on that
444 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
445 * request gets inbetween ordered sequence. If this request is
446 * an empty barrier, we don't need to do a postflush ever since
447 * there will be no data written between the pre and post flush.
448 * Hence a single flush will suffice.
450 if ((q
->ordered
& QUEUE_ORDERED_POSTFLUSH
) && !blk_empty_barrier(rq
))
451 queue_flush(q
, QUEUE_ORDERED_POSTFLUSH
);
453 q
->ordseq
|= QUEUE_ORDSEQ_POSTFLUSH
;
455 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
457 if (q
->ordered
& QUEUE_ORDERED_PREFLUSH
) {
458 queue_flush(q
, QUEUE_ORDERED_PREFLUSH
);
459 rq
= &q
->pre_flush_rq
;
461 q
->ordseq
|= QUEUE_ORDSEQ_PREFLUSH
;
463 if ((q
->ordered
& QUEUE_ORDERED_TAG
) || q
->in_flight
== 0)
464 q
->ordseq
|= QUEUE_ORDSEQ_DRAIN
;
471 int blk_do_ordered(struct request_queue
*q
, struct request
**rqp
)
473 struct request
*rq
= *rqp
;
474 const int is_barrier
= blk_fs_request(rq
) && blk_barrier_rq(rq
);
480 if (q
->next_ordered
!= QUEUE_ORDERED_NONE
) {
481 *rqp
= start_ordered(q
, rq
);
485 * This can happen when the queue switches to
486 * ORDERED_NONE while this request is on it.
488 blkdev_dequeue_request(rq
);
489 end_that_request_first(rq
, -EOPNOTSUPP
,
490 rq
->hard_nr_sectors
);
491 end_that_request_last(rq
, -EOPNOTSUPP
);
498 * Ordered sequence in progress
501 /* Special requests are not subject to ordering rules. */
502 if (!blk_fs_request(rq
) &&
503 rq
!= &q
->pre_flush_rq
&& rq
!= &q
->post_flush_rq
)
506 if (q
->ordered
& QUEUE_ORDERED_TAG
) {
507 /* Ordered by tag. Blocking the next barrier is enough. */
508 if (is_barrier
&& rq
!= &q
->bar_rq
)
511 /* Ordered by draining. Wait for turn. */
512 WARN_ON(blk_ordered_req_seq(rq
) < blk_ordered_cur_seq(q
));
513 if (blk_ordered_req_seq(rq
) > blk_ordered_cur_seq(q
))
520 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
521 unsigned int nbytes
, int error
)
523 struct request_queue
*q
= rq
->q
;
525 if (&q
->bar_rq
!= rq
) {
527 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
528 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
531 if (unlikely(nbytes
> bio
->bi_size
)) {
532 printk("%s: want %u bytes done, only %u left\n",
533 __FUNCTION__
, nbytes
, bio
->bi_size
);
534 nbytes
= bio
->bi_size
;
537 bio
->bi_size
-= nbytes
;
538 bio
->bi_sector
+= (nbytes
>> 9);
539 if (bio
->bi_size
== 0)
540 bio_endio(bio
, error
);
544 * Okay, this is the barrier request in progress, just
547 if (error
&& !q
->orderr
)
553 * blk_queue_bounce_limit - set bounce buffer limit for queue
554 * @q: the request queue for the device
555 * @dma_addr: bus address limit
558 * Different hardware can have different requirements as to what pages
559 * it can do I/O directly to. A low level driver can call
560 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
561 * buffers for doing I/O to pages residing above @page.
563 void blk_queue_bounce_limit(struct request_queue
*q
, u64 dma_addr
)
565 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
568 q
->bounce_gfp
= GFP_NOIO
;
569 #if BITS_PER_LONG == 64
570 /* Assume anything <= 4GB can be handled by IOMMU.
571 Actually some IOMMUs can handle everything, but I don't
572 know of a way to test this here. */
573 if (bounce_pfn
< (min_t(u64
,0xffffffff,BLK_BOUNCE_HIGH
) >> PAGE_SHIFT
))
575 q
->bounce_pfn
= max_low_pfn
;
577 if (bounce_pfn
< blk_max_low_pfn
)
579 q
->bounce_pfn
= bounce_pfn
;
582 init_emergency_isa_pool();
583 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
584 q
->bounce_pfn
= bounce_pfn
;
588 EXPORT_SYMBOL(blk_queue_bounce_limit
);
591 * blk_queue_max_sectors - set max sectors for a request for this queue
592 * @q: the request queue for the device
593 * @max_sectors: max sectors in the usual 512b unit
596 * Enables a low level driver to set an upper limit on the size of
599 void blk_queue_max_sectors(struct request_queue
*q
, unsigned int max_sectors
)
601 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
602 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
603 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
606 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
607 q
->max_hw_sectors
= q
->max_sectors
= max_sectors
;
609 q
->max_sectors
= BLK_DEF_MAX_SECTORS
;
610 q
->max_hw_sectors
= max_sectors
;
614 EXPORT_SYMBOL(blk_queue_max_sectors
);
617 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
618 * @q: the request queue for the device
619 * @max_segments: max number of segments
622 * Enables a low level driver to set an upper limit on the number of
623 * physical data segments in a request. This would be the largest sized
624 * scatter list the driver could handle.
626 void blk_queue_max_phys_segments(struct request_queue
*q
,
627 unsigned short max_segments
)
631 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
634 q
->max_phys_segments
= max_segments
;
637 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
640 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
641 * @q: the request queue for the device
642 * @max_segments: max number of segments
645 * Enables a low level driver to set an upper limit on the number of
646 * hw data segments in a request. This would be the largest number of
647 * address/length pairs the host adapter can actually give as once
650 void blk_queue_max_hw_segments(struct request_queue
*q
,
651 unsigned short max_segments
)
655 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
658 q
->max_hw_segments
= max_segments
;
661 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
664 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
665 * @q: the request queue for the device
666 * @max_size: max size of segment in bytes
669 * Enables a low level driver to set an upper limit on the size of a
672 void blk_queue_max_segment_size(struct request_queue
*q
, unsigned int max_size
)
674 if (max_size
< PAGE_CACHE_SIZE
) {
675 max_size
= PAGE_CACHE_SIZE
;
676 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
679 q
->max_segment_size
= max_size
;
682 EXPORT_SYMBOL(blk_queue_max_segment_size
);
685 * blk_queue_hardsect_size - set hardware sector size for the queue
686 * @q: the request queue for the device
687 * @size: the hardware sector size, in bytes
690 * This should typically be set to the lowest possible sector size
691 * that the hardware can operate on (possible without reverting to
692 * even internal read-modify-write operations). Usually the default
693 * of 512 covers most hardware.
695 void blk_queue_hardsect_size(struct request_queue
*q
, unsigned short size
)
697 q
->hardsect_size
= size
;
700 EXPORT_SYMBOL(blk_queue_hardsect_size
);
703 * Returns the minimum that is _not_ zero, unless both are zero.
705 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
708 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
709 * @t: the stacking driver (top)
710 * @b: the underlying device (bottom)
712 void blk_queue_stack_limits(struct request_queue
*t
, struct request_queue
*b
)
714 /* zero is "infinity" */
715 t
->max_sectors
= min_not_zero(t
->max_sectors
,b
->max_sectors
);
716 t
->max_hw_sectors
= min_not_zero(t
->max_hw_sectors
,b
->max_hw_sectors
);
718 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
719 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
720 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
721 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
722 if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
))
723 clear_bit(QUEUE_FLAG_CLUSTER
, &t
->queue_flags
);
726 EXPORT_SYMBOL(blk_queue_stack_limits
);
729 * blk_queue_segment_boundary - set boundary rules for segment merging
730 * @q: the request queue for the device
731 * @mask: the memory boundary mask
733 void blk_queue_segment_boundary(struct request_queue
*q
, unsigned long mask
)
735 if (mask
< PAGE_CACHE_SIZE
- 1) {
736 mask
= PAGE_CACHE_SIZE
- 1;
737 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
740 q
->seg_boundary_mask
= mask
;
743 EXPORT_SYMBOL(blk_queue_segment_boundary
);
746 * blk_queue_dma_alignment - set dma length and memory alignment
747 * @q: the request queue for the device
748 * @mask: alignment mask
751 * set required memory and length aligment for direct dma transactions.
752 * this is used when buiding direct io requests for the queue.
755 void blk_queue_dma_alignment(struct request_queue
*q
, int mask
)
757 q
->dma_alignment
= mask
;
760 EXPORT_SYMBOL(blk_queue_dma_alignment
);
763 * blk_queue_find_tag - find a request by its tag and queue
764 * @q: The request queue for the device
765 * @tag: The tag of the request
768 * Should be used when a device returns a tag and you want to match
771 * no locks need be held.
773 struct request
*blk_queue_find_tag(struct request_queue
*q
, int tag
)
775 return blk_map_queue_find_tag(q
->queue_tags
, tag
);
778 EXPORT_SYMBOL(blk_queue_find_tag
);
781 * __blk_free_tags - release a given set of tag maintenance info
782 * @bqt: the tag map to free
784 * Tries to free the specified @bqt@. Returns true if it was
785 * actually freed and false if there are still references using it
787 static int __blk_free_tags(struct blk_queue_tag
*bqt
)
791 retval
= atomic_dec_and_test(&bqt
->refcnt
);
794 BUG_ON(!list_empty(&bqt
->busy_list
));
796 kfree(bqt
->tag_index
);
797 bqt
->tag_index
= NULL
;
810 * __blk_queue_free_tags - release tag maintenance info
811 * @q: the request queue for the device
814 * blk_cleanup_queue() will take care of calling this function, if tagging
815 * has been used. So there's no need to call this directly.
817 static void __blk_queue_free_tags(struct request_queue
*q
)
819 struct blk_queue_tag
*bqt
= q
->queue_tags
;
824 __blk_free_tags(bqt
);
826 q
->queue_tags
= NULL
;
827 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
832 * blk_free_tags - release a given set of tag maintenance info
833 * @bqt: the tag map to free
835 * For externally managed @bqt@ frees the map. Callers of this
836 * function must guarantee to have released all the queues that
837 * might have been using this tag map.
839 void blk_free_tags(struct blk_queue_tag
*bqt
)
841 if (unlikely(!__blk_free_tags(bqt
)))
844 EXPORT_SYMBOL(blk_free_tags
);
847 * blk_queue_free_tags - release tag maintenance info
848 * @q: the request queue for the device
851 * This is used to disabled tagged queuing to a device, yet leave
854 void blk_queue_free_tags(struct request_queue
*q
)
856 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
859 EXPORT_SYMBOL(blk_queue_free_tags
);
862 init_tag_map(struct request_queue
*q
, struct blk_queue_tag
*tags
, int depth
)
864 struct request
**tag_index
;
865 unsigned long *tag_map
;
868 if (q
&& depth
> q
->nr_requests
* 2) {
869 depth
= q
->nr_requests
* 2;
870 printk(KERN_ERR
"%s: adjusted depth to %d\n",
871 __FUNCTION__
, depth
);
874 tag_index
= kzalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
878 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
879 tag_map
= kzalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
883 tags
->real_max_depth
= depth
;
884 tags
->max_depth
= depth
;
885 tags
->tag_index
= tag_index
;
886 tags
->tag_map
= tag_map
;
894 static struct blk_queue_tag
*__blk_queue_init_tags(struct request_queue
*q
,
897 struct blk_queue_tag
*tags
;
899 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
903 if (init_tag_map(q
, tags
, depth
))
906 INIT_LIST_HEAD(&tags
->busy_list
);
908 atomic_set(&tags
->refcnt
, 1);
916 * blk_init_tags - initialize the tag info for an external tag map
917 * @depth: the maximum queue depth supported
918 * @tags: the tag to use
920 struct blk_queue_tag
*blk_init_tags(int depth
)
922 return __blk_queue_init_tags(NULL
, depth
);
924 EXPORT_SYMBOL(blk_init_tags
);
927 * blk_queue_init_tags - initialize the queue tag info
928 * @q: the request queue for the device
929 * @depth: the maximum queue depth supported
930 * @tags: the tag to use
932 int blk_queue_init_tags(struct request_queue
*q
, int depth
,
933 struct blk_queue_tag
*tags
)
937 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
939 if (!tags
&& !q
->queue_tags
) {
940 tags
= __blk_queue_init_tags(q
, depth
);
944 } else if (q
->queue_tags
) {
945 if ((rc
= blk_queue_resize_tags(q
, depth
)))
947 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
950 atomic_inc(&tags
->refcnt
);
953 * assign it, all done
955 q
->queue_tags
= tags
;
956 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
963 EXPORT_SYMBOL(blk_queue_init_tags
);
966 * blk_queue_resize_tags - change the queueing depth
967 * @q: the request queue for the device
968 * @new_depth: the new max command queueing depth
971 * Must be called with the queue lock held.
973 int blk_queue_resize_tags(struct request_queue
*q
, int new_depth
)
975 struct blk_queue_tag
*bqt
= q
->queue_tags
;
976 struct request
**tag_index
;
977 unsigned long *tag_map
;
978 int max_depth
, nr_ulongs
;
984 * if we already have large enough real_max_depth. just
985 * adjust max_depth. *NOTE* as requests with tag value
986 * between new_depth and real_max_depth can be in-flight, tag
987 * map can not be shrunk blindly here.
989 if (new_depth
<= bqt
->real_max_depth
) {
990 bqt
->max_depth
= new_depth
;
995 * Currently cannot replace a shared tag map with a new
996 * one, so error out if this is the case
998 if (atomic_read(&bqt
->refcnt
) != 1)
1002 * save the old state info, so we can copy it back
1004 tag_index
= bqt
->tag_index
;
1005 tag_map
= bqt
->tag_map
;
1006 max_depth
= bqt
->real_max_depth
;
1008 if (init_tag_map(q
, bqt
, new_depth
))
1011 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
1012 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
1013 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
1020 EXPORT_SYMBOL(blk_queue_resize_tags
);
1023 * blk_queue_end_tag - end tag operations for a request
1024 * @q: the request queue for the device
1025 * @rq: the request that has completed
1028 * Typically called when end_that_request_first() returns 0, meaning
1029 * all transfers have been done for a request. It's important to call
1030 * this function before end_that_request_last(), as that will put the
1031 * request back on the free list thus corrupting the internal tag list.
1034 * queue lock must be held.
1036 void blk_queue_end_tag(struct request_queue
*q
, struct request
*rq
)
1038 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1043 if (unlikely(tag
>= bqt
->real_max_depth
))
1045 * This can happen after tag depth has been reduced.
1046 * FIXME: how about a warning or info message here?
1050 list_del_init(&rq
->queuelist
);
1051 rq
->cmd_flags
&= ~REQ_QUEUED
;
1054 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
1055 printk(KERN_ERR
"%s: tag %d is missing\n",
1058 bqt
->tag_index
[tag
] = NULL
;
1061 * We use test_and_clear_bit's memory ordering properties here.
1062 * The tag_map bit acts as a lock for tag_index[bit], so we need
1063 * a barrer before clearing the bit (precisely: release semantics).
1064 * Could use clear_bit_unlock when it is merged.
1066 if (unlikely(!test_and_clear_bit(tag
, bqt
->tag_map
))) {
1067 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
1075 EXPORT_SYMBOL(blk_queue_end_tag
);
1078 * blk_queue_start_tag - find a free tag and assign it
1079 * @q: the request queue for the device
1080 * @rq: the block request that needs tagging
1083 * This can either be used as a stand-alone helper, or possibly be
1084 * assigned as the queue &prep_rq_fn (in which case &struct request
1085 * automagically gets a tag assigned). Note that this function
1086 * assumes that any type of request can be queued! if this is not
1087 * true for your device, you must check the request type before
1088 * calling this function. The request will also be removed from
1089 * the request queue, so it's the drivers responsibility to readd
1090 * it if it should need to be restarted for some reason.
1093 * queue lock must be held.
1095 int blk_queue_start_tag(struct request_queue
*q
, struct request
*rq
)
1097 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1100 if (unlikely((rq
->cmd_flags
& REQ_QUEUED
))) {
1102 "%s: request %p for device [%s] already tagged %d",
1104 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
1109 * Protect against shared tag maps, as we may not have exclusive
1110 * access to the tag map.
1113 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
1114 if (tag
>= bqt
->max_depth
)
1117 } while (test_and_set_bit(tag
, bqt
->tag_map
));
1119 * We rely on test_and_set_bit providing lock memory ordering semantics
1120 * (could use test_and_set_bit_lock when it is merged).
1123 rq
->cmd_flags
|= REQ_QUEUED
;
1125 bqt
->tag_index
[tag
] = rq
;
1126 blkdev_dequeue_request(rq
);
1127 list_add(&rq
->queuelist
, &bqt
->busy_list
);
1132 EXPORT_SYMBOL(blk_queue_start_tag
);
1135 * blk_queue_invalidate_tags - invalidate all pending tags
1136 * @q: the request queue for the device
1139 * Hardware conditions may dictate a need to stop all pending requests.
1140 * In this case, we will safely clear the block side of the tag queue and
1141 * readd all requests to the request queue in the right order.
1144 * queue lock must be held.
1146 void blk_queue_invalidate_tags(struct request_queue
*q
)
1148 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1149 struct list_head
*tmp
, *n
;
1152 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1153 rq
= list_entry_rq(tmp
);
1155 if (rq
->tag
== -1) {
1157 "%s: bad tag found on list\n", __FUNCTION__
);
1158 list_del_init(&rq
->queuelist
);
1159 rq
->cmd_flags
&= ~REQ_QUEUED
;
1161 blk_queue_end_tag(q
, rq
);
1163 rq
->cmd_flags
&= ~REQ_STARTED
;
1164 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1168 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1170 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1174 printk("%s: dev %s: type=%x, flags=%x\n", msg
,
1175 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
1178 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1180 rq
->current_nr_sectors
);
1181 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1183 if (blk_pc_request(rq
)) {
1185 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1186 printk("%02x ", rq
->cmd
[bit
]);
1191 EXPORT_SYMBOL(blk_dump_rq_flags
);
1193 void blk_recount_segments(struct request_queue
*q
, struct bio
*bio
)
1196 struct bio
*nxt
= bio
->bi_next
;
1198 rq
.bio
= rq
.biotail
= bio
;
1199 bio
->bi_next
= NULL
;
1200 blk_recalc_rq_segments(&rq
);
1202 bio
->bi_phys_segments
= rq
.nr_phys_segments
;
1203 bio
->bi_hw_segments
= rq
.nr_hw_segments
;
1204 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1206 EXPORT_SYMBOL(blk_recount_segments
);
1208 static void blk_recalc_rq_segments(struct request
*rq
)
1212 unsigned int phys_size
;
1213 unsigned int hw_size
;
1214 struct bio_vec
*bv
, *bvprv
= NULL
;
1218 struct req_iterator iter
;
1219 int high
, highprv
= 1;
1220 struct request_queue
*q
= rq
->q
;
1225 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1226 hw_seg_size
= seg_size
= 0;
1227 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
1228 rq_for_each_segment(bv
, rq
, iter
) {
1230 * the trick here is making sure that a high page is never
1231 * considered part of another segment, since that might
1232 * change with the bounce page.
1234 high
= page_to_pfn(bv
->bv_page
) > q
->bounce_pfn
;
1235 if (high
|| highprv
)
1236 goto new_hw_segment
;
1238 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1240 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1242 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1244 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1245 goto new_hw_segment
;
1247 seg_size
+= bv
->bv_len
;
1248 hw_seg_size
+= bv
->bv_len
;
1253 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1254 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1255 hw_seg_size
+= bv
->bv_len
;
1258 if (nr_hw_segs
== 1 &&
1259 hw_seg_size
> rq
->bio
->bi_hw_front_size
)
1260 rq
->bio
->bi_hw_front_size
= hw_seg_size
;
1261 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1267 seg_size
= bv
->bv_len
;
1271 if (nr_hw_segs
== 1 &&
1272 hw_seg_size
> rq
->bio
->bi_hw_front_size
)
1273 rq
->bio
->bi_hw_front_size
= hw_seg_size
;
1274 if (hw_seg_size
> rq
->biotail
->bi_hw_back_size
)
1275 rq
->biotail
->bi_hw_back_size
= hw_seg_size
;
1276 rq
->nr_phys_segments
= nr_phys_segs
;
1277 rq
->nr_hw_segments
= nr_hw_segs
;
1280 static int blk_phys_contig_segment(struct request_queue
*q
, struct bio
*bio
,
1283 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1286 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1288 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1292 * bio and nxt are contigous in memory, check if the queue allows
1293 * these two to be merged into one
1295 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1301 static int blk_hw_contig_segment(struct request_queue
*q
, struct bio
*bio
,
1304 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1305 blk_recount_segments(q
, bio
);
1306 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1307 blk_recount_segments(q
, nxt
);
1308 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1309 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_back_size
+ nxt
->bi_hw_front_size
))
1311 if (bio
->bi_hw_back_size
+ nxt
->bi_hw_front_size
> q
->max_segment_size
)
1318 * map a request to scatterlist, return number of sg entries setup. Caller
1319 * must make sure sg can hold rq->nr_phys_segments entries
1321 int blk_rq_map_sg(struct request_queue
*q
, struct request
*rq
,
1322 struct scatterlist
*sglist
)
1324 struct bio_vec
*bvec
, *bvprv
;
1325 struct scatterlist
*next_sg
, *sg
;
1326 struct req_iterator iter
;
1330 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1333 * for each bio in rq
1336 sg
= next_sg
= &sglist
[0];
1337 rq_for_each_segment(bvec
, rq
, iter
) {
1338 int nbytes
= bvec
->bv_len
;
1340 if (bvprv
&& cluster
) {
1341 if (sg
->length
+ nbytes
> q
->max_segment_size
)
1344 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1346 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1349 sg
->length
+= nbytes
;
1353 next_sg
= sg_next(sg
);
1355 memset(sg
, 0, sizeof(*sg
));
1356 sg
->page
= bvec
->bv_page
;
1357 sg
->length
= nbytes
;
1358 sg
->offset
= bvec
->bv_offset
;
1362 } /* segments in rq */
1367 EXPORT_SYMBOL(blk_rq_map_sg
);
1370 * the standard queue merge functions, can be overridden with device
1371 * specific ones if so desired
1374 static inline int ll_new_mergeable(struct request_queue
*q
,
1375 struct request
*req
,
1378 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1380 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1381 req
->cmd_flags
|= REQ_NOMERGE
;
1382 if (req
== q
->last_merge
)
1383 q
->last_merge
= NULL
;
1388 * A hw segment is just getting larger, bump just the phys
1391 req
->nr_phys_segments
+= nr_phys_segs
;
1395 static inline int ll_new_hw_segment(struct request_queue
*q
,
1396 struct request
*req
,
1399 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1400 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1402 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1403 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1404 req
->cmd_flags
|= REQ_NOMERGE
;
1405 if (req
== q
->last_merge
)
1406 q
->last_merge
= NULL
;
1411 * This will form the start of a new hw segment. Bump both
1414 req
->nr_hw_segments
+= nr_hw_segs
;
1415 req
->nr_phys_segments
+= nr_phys_segs
;
1419 static int ll_back_merge_fn(struct request_queue
*q
, struct request
*req
,
1422 unsigned short max_sectors
;
1425 if (unlikely(blk_pc_request(req
)))
1426 max_sectors
= q
->max_hw_sectors
;
1428 max_sectors
= q
->max_sectors
;
1430 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1431 req
->cmd_flags
|= REQ_NOMERGE
;
1432 if (req
== q
->last_merge
)
1433 q
->last_merge
= NULL
;
1436 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1437 blk_recount_segments(q
, req
->biotail
);
1438 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1439 blk_recount_segments(q
, bio
);
1440 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1441 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1442 !BIOVEC_VIRT_OVERSIZE(len
)) {
1443 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1446 if (req
->nr_hw_segments
== 1)
1447 req
->bio
->bi_hw_front_size
= len
;
1448 if (bio
->bi_hw_segments
== 1)
1449 bio
->bi_hw_back_size
= len
;
1454 return ll_new_hw_segment(q
, req
, bio
);
1457 static int ll_front_merge_fn(struct request_queue
*q
, struct request
*req
,
1460 unsigned short max_sectors
;
1463 if (unlikely(blk_pc_request(req
)))
1464 max_sectors
= q
->max_hw_sectors
;
1466 max_sectors
= q
->max_sectors
;
1469 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1470 req
->cmd_flags
|= REQ_NOMERGE
;
1471 if (req
== q
->last_merge
)
1472 q
->last_merge
= NULL
;
1475 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1476 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1477 blk_recount_segments(q
, bio
);
1478 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1479 blk_recount_segments(q
, req
->bio
);
1480 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1481 !BIOVEC_VIRT_OVERSIZE(len
)) {
1482 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1485 if (bio
->bi_hw_segments
== 1)
1486 bio
->bi_hw_front_size
= len
;
1487 if (req
->nr_hw_segments
== 1)
1488 req
->biotail
->bi_hw_back_size
= len
;
1493 return ll_new_hw_segment(q
, req
, bio
);
1496 static int ll_merge_requests_fn(struct request_queue
*q
, struct request
*req
,
1497 struct request
*next
)
1499 int total_phys_segments
;
1500 int total_hw_segments
;
1503 * First check if the either of the requests are re-queued
1504 * requests. Can't merge them if they are.
1506 if (req
->special
|| next
->special
)
1510 * Will it become too large?
1512 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1515 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1516 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1517 total_phys_segments
--;
1519 if (total_phys_segments
> q
->max_phys_segments
)
1522 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1523 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1524 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1526 * propagate the combined length to the end of the requests
1528 if (req
->nr_hw_segments
== 1)
1529 req
->bio
->bi_hw_front_size
= len
;
1530 if (next
->nr_hw_segments
== 1)
1531 next
->biotail
->bi_hw_back_size
= len
;
1532 total_hw_segments
--;
1535 if (total_hw_segments
> q
->max_hw_segments
)
1538 /* Merge is OK... */
1539 req
->nr_phys_segments
= total_phys_segments
;
1540 req
->nr_hw_segments
= total_hw_segments
;
1545 * "plug" the device if there are no outstanding requests: this will
1546 * force the transfer to start only after we have put all the requests
1549 * This is called with interrupts off and no requests on the queue and
1550 * with the queue lock held.
1552 void blk_plug_device(struct request_queue
*q
)
1554 WARN_ON(!irqs_disabled());
1557 * don't plug a stopped queue, it must be paired with blk_start_queue()
1558 * which will restart the queueing
1560 if (blk_queue_stopped(q
))
1563 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
1564 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1565 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
1569 EXPORT_SYMBOL(blk_plug_device
);
1572 * remove the queue from the plugged list, if present. called with
1573 * queue lock held and interrupts disabled.
1575 int blk_remove_plug(struct request_queue
*q
)
1577 WARN_ON(!irqs_disabled());
1579 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1582 del_timer(&q
->unplug_timer
);
1586 EXPORT_SYMBOL(blk_remove_plug
);
1589 * remove the plug and let it rip..
1591 void __generic_unplug_device(struct request_queue
*q
)
1593 if (unlikely(blk_queue_stopped(q
)))
1596 if (!blk_remove_plug(q
))
1601 EXPORT_SYMBOL(__generic_unplug_device
);
1604 * generic_unplug_device - fire a request queue
1605 * @q: The &struct request_queue in question
1608 * Linux uses plugging to build bigger requests queues before letting
1609 * the device have at them. If a queue is plugged, the I/O scheduler
1610 * is still adding and merging requests on the queue. Once the queue
1611 * gets unplugged, the request_fn defined for the queue is invoked and
1612 * transfers started.
1614 void generic_unplug_device(struct request_queue
*q
)
1616 spin_lock_irq(q
->queue_lock
);
1617 __generic_unplug_device(q
);
1618 spin_unlock_irq(q
->queue_lock
);
1620 EXPORT_SYMBOL(generic_unplug_device
);
1622 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1625 struct request_queue
*q
= bdi
->unplug_io_data
;
1628 * devices don't necessarily have an ->unplug_fn defined
1631 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1632 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1638 static void blk_unplug_work(struct work_struct
*work
)
1640 struct request_queue
*q
=
1641 container_of(work
, struct request_queue
, unplug_work
);
1643 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1644 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1649 static void blk_unplug_timeout(unsigned long data
)
1651 struct request_queue
*q
= (struct request_queue
*)data
;
1653 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
1654 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1656 kblockd_schedule_work(&q
->unplug_work
);
1660 * blk_start_queue - restart a previously stopped queue
1661 * @q: The &struct request_queue in question
1664 * blk_start_queue() will clear the stop flag on the queue, and call
1665 * the request_fn for the queue if it was in a stopped state when
1666 * entered. Also see blk_stop_queue(). Queue lock must be held.
1668 void blk_start_queue(struct request_queue
*q
)
1670 WARN_ON(!irqs_disabled());
1672 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1675 * one level of recursion is ok and is much faster than kicking
1676 * the unplug handling
1678 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1680 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1683 kblockd_schedule_work(&q
->unplug_work
);
1687 EXPORT_SYMBOL(blk_start_queue
);
1690 * blk_stop_queue - stop a queue
1691 * @q: The &struct request_queue in question
1694 * The Linux block layer assumes that a block driver will consume all
1695 * entries on the request queue when the request_fn strategy is called.
1696 * Often this will not happen, because of hardware limitations (queue
1697 * depth settings). If a device driver gets a 'queue full' response,
1698 * or if it simply chooses not to queue more I/O at one point, it can
1699 * call this function to prevent the request_fn from being called until
1700 * the driver has signalled it's ready to go again. This happens by calling
1701 * blk_start_queue() to restart queue operations. Queue lock must be held.
1703 void blk_stop_queue(struct request_queue
*q
)
1706 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1708 EXPORT_SYMBOL(blk_stop_queue
);
1711 * blk_sync_queue - cancel any pending callbacks on a queue
1715 * The block layer may perform asynchronous callback activity
1716 * on a queue, such as calling the unplug function after a timeout.
1717 * A block device may call blk_sync_queue to ensure that any
1718 * such activity is cancelled, thus allowing it to release resources
1719 * that the callbacks might use. The caller must already have made sure
1720 * that its ->make_request_fn will not re-add plugging prior to calling
1724 void blk_sync_queue(struct request_queue
*q
)
1726 del_timer_sync(&q
->unplug_timer
);
1728 EXPORT_SYMBOL(blk_sync_queue
);
1731 * blk_run_queue - run a single device queue
1732 * @q: The queue to run
1734 void blk_run_queue(struct request_queue
*q
)
1736 unsigned long flags
;
1738 spin_lock_irqsave(q
->queue_lock
, flags
);
1742 * Only recurse once to avoid overrunning the stack, let the unplug
1743 * handling reinvoke the handler shortly if we already got there.
1745 if (!elv_queue_empty(q
)) {
1746 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1748 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1751 kblockd_schedule_work(&q
->unplug_work
);
1755 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1757 EXPORT_SYMBOL(blk_run_queue
);
1760 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
1761 * @kobj: the kobj belonging of the request queue to be released
1764 * blk_cleanup_queue is the pair to blk_init_queue() or
1765 * blk_queue_make_request(). It should be called when a request queue is
1766 * being released; typically when a block device is being de-registered.
1767 * Currently, its primary task it to free all the &struct request
1768 * structures that were allocated to the queue and the queue itself.
1771 * Hopefully the low level driver will have finished any
1772 * outstanding requests first...
1774 static void blk_release_queue(struct kobject
*kobj
)
1776 struct request_queue
*q
=
1777 container_of(kobj
, struct request_queue
, kobj
);
1778 struct request_list
*rl
= &q
->rq
;
1783 mempool_destroy(rl
->rq_pool
);
1786 __blk_queue_free_tags(q
);
1788 blk_trace_shutdown(q
);
1790 kmem_cache_free(requestq_cachep
, q
);
1793 void blk_put_queue(struct request_queue
*q
)
1795 kobject_put(&q
->kobj
);
1797 EXPORT_SYMBOL(blk_put_queue
);
1799 void blk_cleanup_queue(struct request_queue
* q
)
1801 mutex_lock(&q
->sysfs_lock
);
1802 set_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
);
1803 mutex_unlock(&q
->sysfs_lock
);
1806 elevator_exit(q
->elevator
);
1811 EXPORT_SYMBOL(blk_cleanup_queue
);
1813 static int blk_init_free_list(struct request_queue
*q
)
1815 struct request_list
*rl
= &q
->rq
;
1817 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1818 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1820 init_waitqueue_head(&rl
->wait
[READ
]);
1821 init_waitqueue_head(&rl
->wait
[WRITE
]);
1823 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1824 mempool_free_slab
, request_cachep
, q
->node
);
1832 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
1834 return blk_alloc_queue_node(gfp_mask
, -1);
1836 EXPORT_SYMBOL(blk_alloc_queue
);
1838 static struct kobj_type queue_ktype
;
1840 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1842 struct request_queue
*q
;
1844 q
= kmem_cache_alloc_node(requestq_cachep
,
1845 gfp_mask
| __GFP_ZERO
, node_id
);
1849 init_timer(&q
->unplug_timer
);
1851 kobject_set_name(&q
->kobj
, "%s", "queue");
1852 q
->kobj
.ktype
= &queue_ktype
;
1853 kobject_init(&q
->kobj
);
1855 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1856 q
->backing_dev_info
.unplug_io_data
= q
;
1858 mutex_init(&q
->sysfs_lock
);
1862 EXPORT_SYMBOL(blk_alloc_queue_node
);
1865 * blk_init_queue - prepare a request queue for use with a block device
1866 * @rfn: The function to be called to process requests that have been
1867 * placed on the queue.
1868 * @lock: Request queue spin lock
1871 * If a block device wishes to use the standard request handling procedures,
1872 * which sorts requests and coalesces adjacent requests, then it must
1873 * call blk_init_queue(). The function @rfn will be called when there
1874 * are requests on the queue that need to be processed. If the device
1875 * supports plugging, then @rfn may not be called immediately when requests
1876 * are available on the queue, but may be called at some time later instead.
1877 * Plugged queues are generally unplugged when a buffer belonging to one
1878 * of the requests on the queue is needed, or due to memory pressure.
1880 * @rfn is not required, or even expected, to remove all requests off the
1881 * queue, but only as many as it can handle at a time. If it does leave
1882 * requests on the queue, it is responsible for arranging that the requests
1883 * get dealt with eventually.
1885 * The queue spin lock must be held while manipulating the requests on the
1886 * request queue; this lock will be taken also from interrupt context, so irq
1887 * disabling is needed for it.
1889 * Function returns a pointer to the initialized request queue, or NULL if
1890 * it didn't succeed.
1893 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1894 * when the block device is deactivated (such as at module unload).
1897 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1899 return blk_init_queue_node(rfn
, lock
, -1);
1901 EXPORT_SYMBOL(blk_init_queue
);
1903 struct request_queue
*
1904 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1906 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1912 if (blk_init_free_list(q
)) {
1913 kmem_cache_free(requestq_cachep
, q
);
1918 * if caller didn't supply a lock, they get per-queue locking with
1922 spin_lock_init(&q
->__queue_lock
);
1923 lock
= &q
->__queue_lock
;
1926 q
->request_fn
= rfn
;
1927 q
->prep_rq_fn
= NULL
;
1928 q
->unplug_fn
= generic_unplug_device
;
1929 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1930 q
->queue_lock
= lock
;
1932 blk_queue_segment_boundary(q
, 0xffffffff);
1934 blk_queue_make_request(q
, __make_request
);
1935 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1937 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1938 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1940 q
->sg_reserved_size
= INT_MAX
;
1945 if (!elevator_init(q
, NULL
)) {
1946 blk_queue_congestion_threshold(q
);
1953 EXPORT_SYMBOL(blk_init_queue_node
);
1955 int blk_get_queue(struct request_queue
*q
)
1957 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1958 kobject_get(&q
->kobj
);
1965 EXPORT_SYMBOL(blk_get_queue
);
1967 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
1969 if (rq
->cmd_flags
& REQ_ELVPRIV
)
1970 elv_put_request(q
, rq
);
1971 mempool_free(rq
, q
->rq
.rq_pool
);
1974 static struct request
*
1975 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
1977 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1983 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
1984 * see bio.h and blkdev.h
1986 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
1989 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
1990 mempool_free(rq
, q
->rq
.rq_pool
);
1993 rq
->cmd_flags
|= REQ_ELVPRIV
;
2000 * ioc_batching returns true if the ioc is a valid batching request and
2001 * should be given priority access to a request.
2003 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
2009 * Make sure the process is able to allocate at least 1 request
2010 * even if the batch times out, otherwise we could theoretically
2013 return ioc
->nr_batch_requests
== q
->nr_batching
||
2014 (ioc
->nr_batch_requests
> 0
2015 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
2019 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2020 * will cause the process to be a "batcher" on all queues in the system. This
2021 * is the behaviour we want though - once it gets a wakeup it should be given
2024 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
2026 if (!ioc
|| ioc_batching(q
, ioc
))
2029 ioc
->nr_batch_requests
= q
->nr_batching
;
2030 ioc
->last_waited
= jiffies
;
2033 static void __freed_request(struct request_queue
*q
, int rw
)
2035 struct request_list
*rl
= &q
->rq
;
2037 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
2038 blk_clear_queue_congested(q
, rw
);
2040 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
2041 if (waitqueue_active(&rl
->wait
[rw
]))
2042 wake_up(&rl
->wait
[rw
]);
2044 blk_clear_queue_full(q
, rw
);
2049 * A request has just been released. Account for it, update the full and
2050 * congestion status, wake up any waiters. Called under q->queue_lock.
2052 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
2054 struct request_list
*rl
= &q
->rq
;
2060 __freed_request(q
, rw
);
2062 if (unlikely(rl
->starved
[rw
^ 1]))
2063 __freed_request(q
, rw
^ 1);
2066 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2068 * Get a free request, queue_lock must be held.
2069 * Returns NULL on failure, with queue_lock held.
2070 * Returns !NULL on success, with queue_lock *not held*.
2072 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
2073 struct bio
*bio
, gfp_t gfp_mask
)
2075 struct request
*rq
= NULL
;
2076 struct request_list
*rl
= &q
->rq
;
2077 struct io_context
*ioc
= NULL
;
2078 const int rw
= rw_flags
& 0x01;
2079 int may_queue
, priv
;
2081 may_queue
= elv_may_queue(q
, rw_flags
);
2082 if (may_queue
== ELV_MQUEUE_NO
)
2085 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
2086 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
2087 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
2089 * The queue will fill after this allocation, so set
2090 * it as full, and mark this process as "batching".
2091 * This process will be allowed to complete a batch of
2092 * requests, others will be blocked.
2094 if (!blk_queue_full(q
, rw
)) {
2095 ioc_set_batching(q
, ioc
);
2096 blk_set_queue_full(q
, rw
);
2098 if (may_queue
!= ELV_MQUEUE_MUST
2099 && !ioc_batching(q
, ioc
)) {
2101 * The queue is full and the allocating
2102 * process is not a "batcher", and not
2103 * exempted by the IO scheduler
2109 blk_set_queue_congested(q
, rw
);
2113 * Only allow batching queuers to allocate up to 50% over the defined
2114 * limit of requests, otherwise we could have thousands of requests
2115 * allocated with any setting of ->nr_requests
2117 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
2121 rl
->starved
[rw
] = 0;
2123 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
2127 spin_unlock_irq(q
->queue_lock
);
2129 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
2130 if (unlikely(!rq
)) {
2132 * Allocation failed presumably due to memory. Undo anything
2133 * we might have messed up.
2135 * Allocating task should really be put onto the front of the
2136 * wait queue, but this is pretty rare.
2138 spin_lock_irq(q
->queue_lock
);
2139 freed_request(q
, rw
, priv
);
2142 * in the very unlikely event that allocation failed and no
2143 * requests for this direction was pending, mark us starved
2144 * so that freeing of a request in the other direction will
2145 * notice us. another possible fix would be to split the
2146 * rq mempool into READ and WRITE
2149 if (unlikely(rl
->count
[rw
] == 0))
2150 rl
->starved
[rw
] = 1;
2156 * ioc may be NULL here, and ioc_batching will be false. That's
2157 * OK, if the queue is under the request limit then requests need
2158 * not count toward the nr_batch_requests limit. There will always
2159 * be some limit enforced by BLK_BATCH_TIME.
2161 if (ioc_batching(q
, ioc
))
2162 ioc
->nr_batch_requests
--;
2166 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
2172 * No available requests for this queue, unplug the device and wait for some
2173 * requests to become available.
2175 * Called with q->queue_lock held, and returns with it unlocked.
2177 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
2180 const int rw
= rw_flags
& 0x01;
2183 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
2186 struct request_list
*rl
= &q
->rq
;
2188 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
2189 TASK_UNINTERRUPTIBLE
);
2191 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
2194 struct io_context
*ioc
;
2196 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
2198 __generic_unplug_device(q
);
2199 spin_unlock_irq(q
->queue_lock
);
2203 * After sleeping, we become a "batching" process and
2204 * will be able to allocate at least one request, and
2205 * up to a big batch of them for a small period time.
2206 * See ioc_batching, ioc_set_batching
2208 ioc
= current_io_context(GFP_NOIO
, q
->node
);
2209 ioc_set_batching(q
, ioc
);
2211 spin_lock_irq(q
->queue_lock
);
2213 finish_wait(&rl
->wait
[rw
], &wait
);
2219 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
2223 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2225 spin_lock_irq(q
->queue_lock
);
2226 if (gfp_mask
& __GFP_WAIT
) {
2227 rq
= get_request_wait(q
, rw
, NULL
);
2229 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2231 spin_unlock_irq(q
->queue_lock
);
2233 /* q->queue_lock is unlocked at this point */
2237 EXPORT_SYMBOL(blk_get_request
);
2240 * blk_start_queueing - initiate dispatch of requests to device
2241 * @q: request queue to kick into gear
2243 * This is basically a helper to remove the need to know whether a queue
2244 * is plugged or not if someone just wants to initiate dispatch of requests
2247 * The queue lock must be held with interrupts disabled.
2249 void blk_start_queueing(struct request_queue
*q
)
2251 if (!blk_queue_plugged(q
))
2254 __generic_unplug_device(q
);
2256 EXPORT_SYMBOL(blk_start_queueing
);
2259 * blk_requeue_request - put a request back on queue
2260 * @q: request queue where request should be inserted
2261 * @rq: request to be inserted
2264 * Drivers often keep queueing requests until the hardware cannot accept
2265 * more, when that condition happens we need to put the request back
2266 * on the queue. Must be called with queue lock held.
2268 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
2270 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
2272 if (blk_rq_tagged(rq
))
2273 blk_queue_end_tag(q
, rq
);
2275 elv_requeue_request(q
, rq
);
2278 EXPORT_SYMBOL(blk_requeue_request
);
2281 * blk_insert_request - insert a special request in to a request queue
2282 * @q: request queue where request should be inserted
2283 * @rq: request to be inserted
2284 * @at_head: insert request at head or tail of queue
2285 * @data: private data
2288 * Many block devices need to execute commands asynchronously, so they don't
2289 * block the whole kernel from preemption during request execution. This is
2290 * accomplished normally by inserting aritficial requests tagged as
2291 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2292 * scheduled for actual execution by the request queue.
2294 * We have the option of inserting the head or the tail of the queue.
2295 * Typically we use the tail for new ioctls and so forth. We use the head
2296 * of the queue for things like a QUEUE_FULL message from a device, or a
2297 * host that is unable to accept a particular command.
2299 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
2300 int at_head
, void *data
)
2302 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2303 unsigned long flags
;
2306 * tell I/O scheduler that this isn't a regular read/write (ie it
2307 * must not attempt merges on this) and that it acts as a soft
2310 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
2311 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
2315 spin_lock_irqsave(q
->queue_lock
, flags
);
2318 * If command is tagged, release the tag
2320 if (blk_rq_tagged(rq
))
2321 blk_queue_end_tag(q
, rq
);
2323 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2324 __elv_add_request(q
, rq
, where
, 0);
2325 blk_start_queueing(q
);
2326 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2329 EXPORT_SYMBOL(blk_insert_request
);
2331 static int __blk_rq_unmap_user(struct bio
*bio
)
2336 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2337 bio_unmap_user(bio
);
2339 ret
= bio_uncopy_user(bio
);
2345 int blk_rq_append_bio(struct request_queue
*q
, struct request
*rq
,
2349 blk_rq_bio_prep(q
, rq
, bio
);
2350 else if (!ll_back_merge_fn(q
, rq
, bio
))
2353 rq
->biotail
->bi_next
= bio
;
2356 rq
->data_len
+= bio
->bi_size
;
2360 EXPORT_SYMBOL(blk_rq_append_bio
);
2362 static int __blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
2363 void __user
*ubuf
, unsigned int len
)
2365 unsigned long uaddr
;
2366 struct bio
*bio
, *orig_bio
;
2369 reading
= rq_data_dir(rq
) == READ
;
2372 * if alignment requirement is satisfied, map in user pages for
2373 * direct dma. else, set up kernel bounce buffers
2375 uaddr
= (unsigned long) ubuf
;
2376 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2377 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2379 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2382 return PTR_ERR(bio
);
2385 blk_queue_bounce(q
, &bio
);
2388 * We link the bounce buffer in and could have to traverse it
2389 * later so we have to get a ref to prevent it from being freed
2393 ret
= blk_rq_append_bio(q
, rq
, bio
);
2395 return bio
->bi_size
;
2397 /* if it was boucned we must call the end io function */
2399 __blk_rq_unmap_user(orig_bio
);
2405 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2406 * @q: request queue where request should be inserted
2407 * @rq: request structure to fill
2408 * @ubuf: the user buffer
2409 * @len: length of user data
2412 * Data will be mapped directly for zero copy io, if possible. Otherwise
2413 * a kernel bounce buffer is used.
2415 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2416 * still in process context.
2418 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2419 * before being submitted to the device, as pages mapped may be out of
2420 * reach. It's the callers responsibility to make sure this happens. The
2421 * original bio must be passed back in to blk_rq_unmap_user() for proper
2424 int blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
2425 void __user
*ubuf
, unsigned long len
)
2427 unsigned long bytes_read
= 0;
2428 struct bio
*bio
= NULL
;
2431 if (len
> (q
->max_hw_sectors
<< 9))
2436 while (bytes_read
!= len
) {
2437 unsigned long map_len
, end
, start
;
2439 map_len
= min_t(unsigned long, len
- bytes_read
, BIO_MAX_SIZE
);
2440 end
= ((unsigned long)ubuf
+ map_len
+ PAGE_SIZE
- 1)
2442 start
= (unsigned long)ubuf
>> PAGE_SHIFT
;
2445 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2446 * pages. If this happens we just lower the requested
2447 * mapping len by a page so that we can fit
2449 if (end
- start
> BIO_MAX_PAGES
)
2450 map_len
-= PAGE_SIZE
;
2452 ret
= __blk_rq_map_user(q
, rq
, ubuf
, map_len
);
2461 rq
->buffer
= rq
->data
= NULL
;
2464 blk_rq_unmap_user(bio
);
2468 EXPORT_SYMBOL(blk_rq_map_user
);
2471 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2472 * @q: request queue where request should be inserted
2473 * @rq: request to map data to
2474 * @iov: pointer to the iovec
2475 * @iov_count: number of elements in the iovec
2476 * @len: I/O byte count
2479 * Data will be mapped directly for zero copy io, if possible. Otherwise
2480 * a kernel bounce buffer is used.
2482 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2483 * still in process context.
2485 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2486 * before being submitted to the device, as pages mapped may be out of
2487 * reach. It's the callers responsibility to make sure this happens. The
2488 * original bio must be passed back in to blk_rq_unmap_user() for proper
2491 int blk_rq_map_user_iov(struct request_queue
*q
, struct request
*rq
,
2492 struct sg_iovec
*iov
, int iov_count
, unsigned int len
)
2496 if (!iov
|| iov_count
<= 0)
2499 /* we don't allow misaligned data like bio_map_user() does. If the
2500 * user is using sg, they're expected to know the alignment constraints
2501 * and respect them accordingly */
2502 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2504 return PTR_ERR(bio
);
2506 if (bio
->bi_size
!= len
) {
2508 bio_unmap_user(bio
);
2513 blk_rq_bio_prep(q
, rq
, bio
);
2514 rq
->buffer
= rq
->data
= NULL
;
2518 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2521 * blk_rq_unmap_user - unmap a request with user data
2522 * @bio: start of bio list
2525 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2526 * supply the original rq->bio from the blk_rq_map_user() return, since
2527 * the io completion may have changed rq->bio.
2529 int blk_rq_unmap_user(struct bio
*bio
)
2531 struct bio
*mapped_bio
;
2536 if (unlikely(bio_flagged(bio
, BIO_BOUNCED
)))
2537 mapped_bio
= bio
->bi_private
;
2539 ret2
= __blk_rq_unmap_user(mapped_bio
);
2545 bio_put(mapped_bio
);
2551 EXPORT_SYMBOL(blk_rq_unmap_user
);
2554 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2555 * @q: request queue where request should be inserted
2556 * @rq: request to fill
2557 * @kbuf: the kernel buffer
2558 * @len: length of user data
2559 * @gfp_mask: memory allocation flags
2561 int blk_rq_map_kern(struct request_queue
*q
, struct request
*rq
, void *kbuf
,
2562 unsigned int len
, gfp_t gfp_mask
)
2566 if (len
> (q
->max_hw_sectors
<< 9))
2571 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2573 return PTR_ERR(bio
);
2575 if (rq_data_dir(rq
) == WRITE
)
2576 bio
->bi_rw
|= (1 << BIO_RW
);
2578 blk_rq_bio_prep(q
, rq
, bio
);
2579 blk_queue_bounce(q
, &rq
->bio
);
2580 rq
->buffer
= rq
->data
= NULL
;
2584 EXPORT_SYMBOL(blk_rq_map_kern
);
2587 * blk_execute_rq_nowait - insert a request into queue for execution
2588 * @q: queue to insert the request in
2589 * @bd_disk: matching gendisk
2590 * @rq: request to insert
2591 * @at_head: insert request at head or tail of queue
2592 * @done: I/O completion handler
2595 * Insert a fully prepared request at the back of the io scheduler queue
2596 * for execution. Don't wait for completion.
2598 void blk_execute_rq_nowait(struct request_queue
*q
, struct gendisk
*bd_disk
,
2599 struct request
*rq
, int at_head
,
2602 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2604 rq
->rq_disk
= bd_disk
;
2605 rq
->cmd_flags
|= REQ_NOMERGE
;
2607 WARN_ON(irqs_disabled());
2608 spin_lock_irq(q
->queue_lock
);
2609 __elv_add_request(q
, rq
, where
, 1);
2610 __generic_unplug_device(q
);
2611 spin_unlock_irq(q
->queue_lock
);
2613 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait
);
2616 * blk_execute_rq - insert a request into queue for execution
2617 * @q: queue to insert the request in
2618 * @bd_disk: matching gendisk
2619 * @rq: request to insert
2620 * @at_head: insert request at head or tail of queue
2623 * Insert a fully prepared request at the back of the io scheduler queue
2624 * for execution and wait for completion.
2626 int blk_execute_rq(struct request_queue
*q
, struct gendisk
*bd_disk
,
2627 struct request
*rq
, int at_head
)
2629 DECLARE_COMPLETION_ONSTACK(wait
);
2630 char sense
[SCSI_SENSE_BUFFERSIZE
];
2634 * we need an extra reference to the request, so we can look at
2635 * it after io completion
2640 memset(sense
, 0, sizeof(sense
));
2645 rq
->end_io_data
= &wait
;
2646 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2647 wait_for_completion(&wait
);
2655 EXPORT_SYMBOL(blk_execute_rq
);
2657 static void bio_end_empty_barrier(struct bio
*bio
, int err
)
2660 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2662 complete(bio
->bi_private
);
2666 * blkdev_issue_flush - queue a flush
2667 * @bdev: blockdev to issue flush for
2668 * @error_sector: error sector
2671 * Issue a flush for the block device in question. Caller can supply
2672 * room for storing the error offset in case of a flush error, if they
2673 * wish to. Caller must run wait_for_completion() on its own.
2675 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2677 DECLARE_COMPLETION_ONSTACK(wait
);
2678 struct request_queue
*q
;
2682 if (bdev
->bd_disk
== NULL
)
2685 q
= bdev_get_queue(bdev
);
2689 bio
= bio_alloc(GFP_KERNEL
, 0);
2693 bio
->bi_end_io
= bio_end_empty_barrier
;
2694 bio
->bi_private
= &wait
;
2695 bio
->bi_bdev
= bdev
;
2696 submit_bio(1 << BIO_RW_BARRIER
, bio
);
2698 wait_for_completion(&wait
);
2701 * The driver must store the error location in ->bi_sector, if
2702 * it supports it. For non-stacked drivers, this should be copied
2706 *error_sector
= bio
->bi_sector
;
2709 if (!bio_flagged(bio
, BIO_UPTODATE
))
2716 EXPORT_SYMBOL(blkdev_issue_flush
);
2718 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2720 int rw
= rq_data_dir(rq
);
2722 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2726 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2728 disk_round_stats(rq
->rq_disk
);
2729 rq
->rq_disk
->in_flight
++;
2734 * add-request adds a request to the linked list.
2735 * queue lock is held and interrupts disabled, as we muck with the
2736 * request queue list.
2738 static inline void add_request(struct request_queue
* q
, struct request
* req
)
2740 drive_stat_acct(req
, req
->nr_sectors
, 1);
2743 * elevator indicated where it wants this request to be
2744 * inserted at elevator_merge time
2746 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2750 * disk_round_stats() - Round off the performance stats on a struct
2753 * The average IO queue length and utilisation statistics are maintained
2754 * by observing the current state of the queue length and the amount of
2755 * time it has been in this state for.
2757 * Normally, that accounting is done on IO completion, but that can result
2758 * in more than a second's worth of IO being accounted for within any one
2759 * second, leading to >100% utilisation. To deal with that, we call this
2760 * function to do a round-off before returning the results when reading
2761 * /proc/diskstats. This accounts immediately for all queue usage up to
2762 * the current jiffies and restarts the counters again.
2764 void disk_round_stats(struct gendisk
*disk
)
2766 unsigned long now
= jiffies
;
2768 if (now
== disk
->stamp
)
2771 if (disk
->in_flight
) {
2772 __disk_stat_add(disk
, time_in_queue
,
2773 disk
->in_flight
* (now
- disk
->stamp
));
2774 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2779 EXPORT_SYMBOL_GPL(disk_round_stats
);
2782 * queue lock must be held
2784 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
2788 if (unlikely(--req
->ref_count
))
2791 elv_completed_request(q
, req
);
2794 * Request may not have originated from ll_rw_blk. if not,
2795 * it didn't come out of our reserved rq pools
2797 if (req
->cmd_flags
& REQ_ALLOCED
) {
2798 int rw
= rq_data_dir(req
);
2799 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
2801 BUG_ON(!list_empty(&req
->queuelist
));
2802 BUG_ON(!hlist_unhashed(&req
->hash
));
2804 blk_free_request(q
, req
);
2805 freed_request(q
, rw
, priv
);
2809 EXPORT_SYMBOL_GPL(__blk_put_request
);
2811 void blk_put_request(struct request
*req
)
2813 unsigned long flags
;
2814 struct request_queue
*q
= req
->q
;
2817 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2818 * following if (q) test.
2821 spin_lock_irqsave(q
->queue_lock
, flags
);
2822 __blk_put_request(q
, req
);
2823 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2827 EXPORT_SYMBOL(blk_put_request
);
2830 * blk_end_sync_rq - executes a completion event on a request
2831 * @rq: request to complete
2832 * @error: end io status of the request
2834 void blk_end_sync_rq(struct request
*rq
, int error
)
2836 struct completion
*waiting
= rq
->end_io_data
;
2838 rq
->end_io_data
= NULL
;
2839 __blk_put_request(rq
->q
, rq
);
2842 * complete last, if this is a stack request the process (and thus
2843 * the rq pointer) could be invalid right after this complete()
2847 EXPORT_SYMBOL(blk_end_sync_rq
);
2850 * Has to be called with the request spinlock acquired
2852 static int attempt_merge(struct request_queue
*q
, struct request
*req
,
2853 struct request
*next
)
2855 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2861 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2864 if (rq_data_dir(req
) != rq_data_dir(next
)
2865 || req
->rq_disk
!= next
->rq_disk
2870 * If we are allowed to merge, then append bio list
2871 * from next to rq and release next. merge_requests_fn
2872 * will have updated segment counts, update sector
2875 if (!ll_merge_requests_fn(q
, req
, next
))
2879 * At this point we have either done a back merge
2880 * or front merge. We need the smaller start_time of
2881 * the merged requests to be the current request
2882 * for accounting purposes.
2884 if (time_after(req
->start_time
, next
->start_time
))
2885 req
->start_time
= next
->start_time
;
2887 req
->biotail
->bi_next
= next
->bio
;
2888 req
->biotail
= next
->biotail
;
2890 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2892 elv_merge_requests(q
, req
, next
);
2895 disk_round_stats(req
->rq_disk
);
2896 req
->rq_disk
->in_flight
--;
2899 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2901 __blk_put_request(q
, next
);
2905 static inline int attempt_back_merge(struct request_queue
*q
,
2908 struct request
*next
= elv_latter_request(q
, rq
);
2911 return attempt_merge(q
, rq
, next
);
2916 static inline int attempt_front_merge(struct request_queue
*q
,
2919 struct request
*prev
= elv_former_request(q
, rq
);
2922 return attempt_merge(q
, prev
, rq
);
2927 static void init_request_from_bio(struct request
*req
, struct bio
*bio
)
2929 req
->cmd_type
= REQ_TYPE_FS
;
2932 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2934 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2935 req
->cmd_flags
|= REQ_FAILFAST
;
2938 * REQ_BARRIER implies no merging, but lets make it explicit
2940 if (unlikely(bio_barrier(bio
)))
2941 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2944 req
->cmd_flags
|= REQ_RW_SYNC
;
2945 if (bio_rw_meta(bio
))
2946 req
->cmd_flags
|= REQ_RW_META
;
2949 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
2950 req
->ioprio
= bio_prio(bio
);
2951 req
->start_time
= jiffies
;
2952 blk_rq_bio_prep(req
->q
, req
, bio
);
2955 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
2957 struct request
*req
;
2958 int el_ret
, nr_sectors
, barrier
, err
;
2959 const unsigned short prio
= bio_prio(bio
);
2960 const int sync
= bio_sync(bio
);
2963 nr_sectors
= bio_sectors(bio
);
2966 * low level driver can indicate that it wants pages above a
2967 * certain limit bounced to low memory (ie for highmem, or even
2968 * ISA dma in theory)
2970 blk_queue_bounce(q
, &bio
);
2972 barrier
= bio_barrier(bio
);
2973 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
2978 spin_lock_irq(q
->queue_lock
);
2980 if (unlikely(barrier
) || elv_queue_empty(q
))
2983 el_ret
= elv_merge(q
, &req
, bio
);
2985 case ELEVATOR_BACK_MERGE
:
2986 BUG_ON(!rq_mergeable(req
));
2988 if (!ll_back_merge_fn(q
, req
, bio
))
2991 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
2993 req
->biotail
->bi_next
= bio
;
2995 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2996 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2997 drive_stat_acct(req
, nr_sectors
, 0);
2998 if (!attempt_back_merge(q
, req
))
2999 elv_merged_request(q
, req
, el_ret
);
3002 case ELEVATOR_FRONT_MERGE
:
3003 BUG_ON(!rq_mergeable(req
));
3005 if (!ll_front_merge_fn(q
, req
, bio
))
3008 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
3010 bio
->bi_next
= req
->bio
;
3014 * may not be valid. if the low level driver said
3015 * it didn't need a bounce buffer then it better
3016 * not touch req->buffer either...
3018 req
->buffer
= bio_data(bio
);
3019 req
->current_nr_sectors
= bio_cur_sectors(bio
);
3020 req
->hard_cur_sectors
= req
->current_nr_sectors
;
3021 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
3022 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
3023 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
3024 drive_stat_acct(req
, nr_sectors
, 0);
3025 if (!attempt_front_merge(q
, req
))
3026 elv_merged_request(q
, req
, el_ret
);
3029 /* ELV_NO_MERGE: elevator says don't/can't merge. */
3036 * This sync check and mask will be re-done in init_request_from_bio(),
3037 * but we need to set it earlier to expose the sync flag to the
3038 * rq allocator and io schedulers.
3040 rw_flags
= bio_data_dir(bio
);
3042 rw_flags
|= REQ_RW_SYNC
;
3045 * Grab a free request. This is might sleep but can not fail.
3046 * Returns with the queue unlocked.
3048 req
= get_request_wait(q
, rw_flags
, bio
);
3051 * After dropping the lock and possibly sleeping here, our request
3052 * may now be mergeable after it had proven unmergeable (above).
3053 * We don't worry about that case for efficiency. It won't happen
3054 * often, and the elevators are able to handle it.
3056 init_request_from_bio(req
, bio
);
3058 spin_lock_irq(q
->queue_lock
);
3059 if (elv_queue_empty(q
))
3061 add_request(q
, req
);
3064 __generic_unplug_device(q
);
3066 spin_unlock_irq(q
->queue_lock
);
3070 bio_endio(bio
, err
);
3075 * If bio->bi_dev is a partition, remap the location
3077 static inline void blk_partition_remap(struct bio
*bio
)
3079 struct block_device
*bdev
= bio
->bi_bdev
;
3081 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
3082 struct hd_struct
*p
= bdev
->bd_part
;
3083 const int rw
= bio_data_dir(bio
);
3085 p
->sectors
[rw
] += bio_sectors(bio
);
3088 bio
->bi_sector
+= p
->start_sect
;
3089 bio
->bi_bdev
= bdev
->bd_contains
;
3091 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
3092 bdev
->bd_dev
, bio
->bi_sector
,
3093 bio
->bi_sector
- p
->start_sect
);
3097 static void handle_bad_sector(struct bio
*bio
)
3099 char b
[BDEVNAME_SIZE
];
3101 printk(KERN_INFO
"attempt to access beyond end of device\n");
3102 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
3103 bdevname(bio
->bi_bdev
, b
),
3105 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
3106 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
3108 set_bit(BIO_EOF
, &bio
->bi_flags
);
3111 #ifdef CONFIG_FAIL_MAKE_REQUEST
3113 static DECLARE_FAULT_ATTR(fail_make_request
);
3115 static int __init
setup_fail_make_request(char *str
)
3117 return setup_fault_attr(&fail_make_request
, str
);
3119 __setup("fail_make_request=", setup_fail_make_request
);
3121 static int should_fail_request(struct bio
*bio
)
3123 if ((bio
->bi_bdev
->bd_disk
->flags
& GENHD_FL_FAIL
) ||
3124 (bio
->bi_bdev
->bd_part
&& bio
->bi_bdev
->bd_part
->make_it_fail
))
3125 return should_fail(&fail_make_request
, bio
->bi_size
);
3130 static int __init
fail_make_request_debugfs(void)
3132 return init_fault_attr_dentries(&fail_make_request
,
3133 "fail_make_request");
3136 late_initcall(fail_make_request_debugfs
);
3138 #else /* CONFIG_FAIL_MAKE_REQUEST */
3140 static inline int should_fail_request(struct bio
*bio
)
3145 #endif /* CONFIG_FAIL_MAKE_REQUEST */
3148 * Check whether this bio extends beyond the end of the device.
3150 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
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
);
3177 * generic_make_request: hand a buffer to its device driver for I/O
3178 * @bio: The bio describing the location in memory and on the device.
3180 * generic_make_request() is used to make I/O requests of block
3181 * devices. It is passed a &struct bio, which describes the I/O that needs
3184 * generic_make_request() does not return any status. The
3185 * success/failure status of the request, along with notification of
3186 * completion, is delivered asynchronously through the bio->bi_end_io
3187 * function described (one day) else where.
3189 * The caller of generic_make_request must make sure that bi_io_vec
3190 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3191 * set to describe the device address, and the
3192 * bi_end_io and optionally bi_private are set to describe how
3193 * completion notification should be signaled.
3195 * generic_make_request and the drivers it calls may use bi_next if this
3196 * bio happens to be merged with someone else, and may change bi_dev and
3197 * bi_sector for remaps as it sees fit. So the values of these fields
3198 * should NOT be depended on after the call to generic_make_request.
3200 static inline void __generic_make_request(struct bio
*bio
)
3202 struct request_queue
*q
;
3203 sector_t old_sector
;
3204 int ret
, nr_sectors
= bio_sectors(bio
);
3209 if (bio_check_eod(bio
, nr_sectors
))
3213 * Resolve the mapping until finished. (drivers are
3214 * still free to implement/resolve their own stacking
3215 * by explicitly returning 0)
3217 * NOTE: we don't repeat the blk_size check for each new device.
3218 * Stacking drivers are expected to know what they are doing.
3223 char b
[BDEVNAME_SIZE
];
3225 q
= bdev_get_queue(bio
->bi_bdev
);
3228 "generic_make_request: Trying to access "
3229 "nonexistent block-device %s (%Lu)\n",
3230 bdevname(bio
->bi_bdev
, b
),
3231 (long long) bio
->bi_sector
);
3233 bio_endio(bio
, -EIO
);
3237 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
3238 printk("bio too big device %s (%u > %u)\n",
3239 bdevname(bio
->bi_bdev
, b
),
3245 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
3248 if (should_fail_request(bio
))
3252 * If this device has partitions, remap block n
3253 * of partition p to block n+start(p) of the disk.
3255 blk_partition_remap(bio
);
3257 if (old_sector
!= -1)
3258 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
3261 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
3263 old_sector
= bio
->bi_sector
;
3264 old_dev
= bio
->bi_bdev
->bd_dev
;
3266 if (bio_check_eod(bio
, nr_sectors
))
3269 ret
= q
->make_request_fn(q
, bio
);
3274 * We only want one ->make_request_fn to be active at a time,
3275 * else stack usage with stacked devices could be a problem.
3276 * So use current->bio_{list,tail} to keep a list of requests
3277 * submited by a make_request_fn function.
3278 * current->bio_tail is also used as a flag to say if
3279 * generic_make_request is currently active in this task or not.
3280 * If it is NULL, then no make_request is active. If it is non-NULL,
3281 * then a make_request is active, and new requests should be added
3284 void generic_make_request(struct bio
*bio
)
3286 if (current
->bio_tail
) {
3287 /* make_request is active */
3288 *(current
->bio_tail
) = bio
;
3289 bio
->bi_next
= NULL
;
3290 current
->bio_tail
= &bio
->bi_next
;
3293 /* following loop may be a bit non-obvious, and so deserves some
3295 * Before entering the loop, bio->bi_next is NULL (as all callers
3296 * ensure that) so we have a list with a single bio.
3297 * We pretend that we have just taken it off a longer list, so
3298 * we assign bio_list to the next (which is NULL) and bio_tail
3299 * to &bio_list, thus initialising the bio_list of new bios to be
3300 * added. __generic_make_request may indeed add some more bios
3301 * through a recursive call to generic_make_request. If it
3302 * did, we find a non-NULL value in bio_list and re-enter the loop
3303 * from the top. In this case we really did just take the bio
3304 * of the top of the list (no pretending) and so fixup bio_list and
3305 * bio_tail or bi_next, and call into __generic_make_request again.
3307 * The loop was structured like this to make only one call to
3308 * __generic_make_request (which is important as it is large and
3309 * inlined) and to keep the structure simple.
3311 BUG_ON(bio
->bi_next
);
3313 current
->bio_list
= bio
->bi_next
;
3314 if (bio
->bi_next
== NULL
)
3315 current
->bio_tail
= ¤t
->bio_list
;
3317 bio
->bi_next
= NULL
;
3318 __generic_make_request(bio
);
3319 bio
= current
->bio_list
;
3321 current
->bio_tail
= NULL
; /* deactivate */
3324 EXPORT_SYMBOL(generic_make_request
);
3327 * submit_bio: submit a bio to the block device layer for I/O
3328 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3329 * @bio: The &struct bio which describes the I/O
3331 * submit_bio() is very similar in purpose to generic_make_request(), and
3332 * uses that function to do most of the work. Both are fairly rough
3333 * interfaces, @bio must be presetup and ready for I/O.
3336 void submit_bio(int rw
, struct bio
*bio
)
3338 int count
= bio_sectors(bio
);
3343 * If it's a regular read/write or a barrier with data attached,
3344 * go through the normal accounting stuff before submission.
3346 if (!bio_empty_barrier(bio
)) {
3348 BIO_BUG_ON(!bio
->bi_size
);
3349 BIO_BUG_ON(!bio
->bi_io_vec
);
3352 count_vm_events(PGPGOUT
, count
);
3354 task_io_account_read(bio
->bi_size
);
3355 count_vm_events(PGPGIN
, count
);
3358 if (unlikely(block_dump
)) {
3359 char b
[BDEVNAME_SIZE
];
3360 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
3361 current
->comm
, current
->pid
,
3362 (rw
& WRITE
) ? "WRITE" : "READ",
3363 (unsigned long long)bio
->bi_sector
,
3364 bdevname(bio
->bi_bdev
,b
));
3368 generic_make_request(bio
);
3371 EXPORT_SYMBOL(submit_bio
);
3373 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3375 if (blk_fs_request(rq
)) {
3376 rq
->hard_sector
+= nsect
;
3377 rq
->hard_nr_sectors
-= nsect
;
3380 * Move the I/O submission pointers ahead if required.
3382 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3383 (rq
->sector
<= rq
->hard_sector
)) {
3384 rq
->sector
= rq
->hard_sector
;
3385 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3386 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3387 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3388 rq
->buffer
= bio_data(rq
->bio
);
3392 * if total number of sectors is less than the first segment
3393 * size, something has gone terribly wrong
3395 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3396 printk("blk: request botched\n");
3397 rq
->nr_sectors
= rq
->current_nr_sectors
;
3402 static int __end_that_request_first(struct request
*req
, int uptodate
,
3405 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3408 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
3411 * extend uptodate bool to allow < 0 value to be direct io error
3414 if (end_io_error(uptodate
))
3415 error
= !uptodate
? -EIO
: uptodate
;
3418 * for a REQ_BLOCK_PC request, we want to carry any eventual
3419 * sense key with us all the way through
3421 if (!blk_pc_request(req
))
3425 if (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))
3426 printk("end_request: I/O error, dev %s, sector %llu\n",
3427 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3428 (unsigned long long)req
->sector
);
3431 if (blk_fs_request(req
) && req
->rq_disk
) {
3432 const int rw
= rq_data_dir(req
);
3434 disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3437 total_bytes
= bio_nbytes
= 0;
3438 while ((bio
= req
->bio
) != NULL
) {
3442 * For an empty barrier request, the low level driver must
3443 * store a potential error location in ->sector. We pass
3444 * that back up in ->bi_sector.
3446 if (blk_empty_barrier(req
))
3447 bio
->bi_sector
= req
->sector
;
3449 if (nr_bytes
>= bio
->bi_size
) {
3450 req
->bio
= bio
->bi_next
;
3451 nbytes
= bio
->bi_size
;
3452 req_bio_endio(req
, bio
, nbytes
, error
);
3456 int idx
= bio
->bi_idx
+ next_idx
;
3458 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3459 blk_dump_rq_flags(req
, "__end_that");
3460 printk("%s: bio idx %d >= vcnt %d\n",
3462 bio
->bi_idx
, bio
->bi_vcnt
);
3466 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3467 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3470 * not a complete bvec done
3472 if (unlikely(nbytes
> nr_bytes
)) {
3473 bio_nbytes
+= nr_bytes
;
3474 total_bytes
+= nr_bytes
;
3479 * advance to the next vector
3482 bio_nbytes
+= nbytes
;
3485 total_bytes
+= nbytes
;
3488 if ((bio
= req
->bio
)) {
3490 * end more in this run, or just return 'not-done'
3492 if (unlikely(nr_bytes
<= 0))
3504 * if the request wasn't completed, update state
3507 req_bio_endio(req
, bio
, bio_nbytes
, error
);
3508 bio
->bi_idx
+= next_idx
;
3509 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3510 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3513 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3514 blk_recalc_rq_segments(req
);
3519 * end_that_request_first - end I/O on a request
3520 * @req: the request being processed
3521 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3522 * @nr_sectors: number of sectors to end I/O on
3525 * Ends I/O on a number of sectors attached to @req, and sets it up
3526 * for the next range of segments (if any) in the cluster.
3529 * 0 - we are done with this request, call end_that_request_last()
3530 * 1 - still buffers pending for this request
3532 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3534 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3537 EXPORT_SYMBOL(end_that_request_first
);
3540 * end_that_request_chunk - end I/O on a request
3541 * @req: the request being processed
3542 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3543 * @nr_bytes: number of bytes to complete
3546 * Ends I/O on a number of bytes attached to @req, and sets it up
3547 * for the next range of segments (if any). Like end_that_request_first(),
3548 * but deals with bytes instead of sectors.
3551 * 0 - we are done with this request, call end_that_request_last()
3552 * 1 - still buffers pending for this request
3554 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3556 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3559 EXPORT_SYMBOL(end_that_request_chunk
);
3562 * splice the completion data to a local structure and hand off to
3563 * process_completion_queue() to complete the requests
3565 static void blk_done_softirq(struct softirq_action
*h
)
3567 struct list_head
*cpu_list
, local_list
;
3569 local_irq_disable();
3570 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3571 list_replace_init(cpu_list
, &local_list
);
3574 while (!list_empty(&local_list
)) {
3575 struct request
*rq
= list_entry(local_list
.next
, struct request
, donelist
);
3577 list_del_init(&rq
->donelist
);
3578 rq
->q
->softirq_done_fn(rq
);
3582 static int __cpuinit
blk_cpu_notify(struct notifier_block
*self
, unsigned long action
,
3586 * If a CPU goes away, splice its entries to the current CPU
3587 * and trigger a run of the softirq
3589 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3590 int cpu
= (unsigned long) hcpu
;
3592 local_irq_disable();
3593 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
3594 &__get_cpu_var(blk_cpu_done
));
3595 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3603 static struct notifier_block blk_cpu_notifier __cpuinitdata
= {
3604 .notifier_call
= blk_cpu_notify
,
3608 * blk_complete_request - end I/O on a request
3609 * @req: the request being processed
3612 * Ends all I/O on a request. It does not handle partial completions,
3613 * unless the driver actually implements this in its completion callback
3614 * through requeueing. The actual completion happens out-of-order,
3615 * through a softirq handler. The user must have registered a completion
3616 * callback through blk_queue_softirq_done().
3619 void blk_complete_request(struct request
*req
)
3621 struct list_head
*cpu_list
;
3622 unsigned long flags
;
3624 BUG_ON(!req
->q
->softirq_done_fn
);
3626 local_irq_save(flags
);
3628 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3629 list_add_tail(&req
->donelist
, cpu_list
);
3630 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3632 local_irq_restore(flags
);
3635 EXPORT_SYMBOL(blk_complete_request
);
3638 * queue lock must be held
3640 void end_that_request_last(struct request
*req
, int uptodate
)
3642 struct gendisk
*disk
= req
->rq_disk
;
3646 * extend uptodate bool to allow < 0 value to be direct io error
3649 if (end_io_error(uptodate
))
3650 error
= !uptodate
? -EIO
: uptodate
;
3652 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3653 laptop_io_completion();
3656 * Account IO completion. bar_rq isn't accounted as a normal
3657 * IO on queueing nor completion. Accounting the containing
3658 * request is enough.
3660 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
3661 unsigned long duration
= jiffies
- req
->start_time
;
3662 const int rw
= rq_data_dir(req
);
3664 __disk_stat_inc(disk
, ios
[rw
]);
3665 __disk_stat_add(disk
, ticks
[rw
], duration
);
3666 disk_round_stats(disk
);
3670 req
->end_io(req
, error
);
3672 __blk_put_request(req
->q
, req
);
3675 EXPORT_SYMBOL(end_that_request_last
);
3677 static inline void __end_request(struct request
*rq
, int uptodate
,
3678 unsigned int nr_bytes
, int dequeue
)
3680 if (!end_that_request_chunk(rq
, uptodate
, nr_bytes
)) {
3682 blkdev_dequeue_request(rq
);
3683 add_disk_randomness(rq
->rq_disk
);
3684 end_that_request_last(rq
, uptodate
);
3688 static unsigned int rq_byte_size(struct request
*rq
)
3690 if (blk_fs_request(rq
))
3691 return rq
->hard_nr_sectors
<< 9;
3693 return rq
->data_len
;
3697 * end_queued_request - end all I/O on a queued request
3698 * @rq: the request being processed
3699 * @uptodate: error value or 0/1 uptodate flag
3702 * Ends all I/O on a request, and removes it from the block layer queues.
3703 * Not suitable for normal IO completion, unless the driver still has
3704 * the request attached to the block layer.
3707 void end_queued_request(struct request
*rq
, int uptodate
)
3709 __end_request(rq
, uptodate
, rq_byte_size(rq
), 1);
3711 EXPORT_SYMBOL(end_queued_request
);
3714 * end_dequeued_request - end all I/O on a dequeued request
3715 * @rq: the request being processed
3716 * @uptodate: error value or 0/1 uptodate flag
3719 * Ends all I/O on a request. The request must already have been
3720 * dequeued using blkdev_dequeue_request(), as is normally the case
3724 void end_dequeued_request(struct request
*rq
, int uptodate
)
3726 __end_request(rq
, uptodate
, rq_byte_size(rq
), 0);
3728 EXPORT_SYMBOL(end_dequeued_request
);
3732 * end_request - end I/O on the current segment of the request
3733 * @rq: the request being processed
3734 * @uptodate: error value or 0/1 uptodate flag
3737 * Ends I/O on the current segment of a request. If that is the only
3738 * remaining segment, the request is also completed and freed.
3740 * This is a remnant of how older block drivers handled IO completions.
3741 * Modern drivers typically end IO on the full request in one go, unless
3742 * they have a residual value to account for. For that case this function
3743 * isn't really useful, unless the residual just happens to be the
3744 * full current segment. In other words, don't use this function in new
3745 * code. Either use end_request_completely(), or the
3746 * end_that_request_chunk() (along with end_that_request_last()) for
3747 * partial completions.
3750 void end_request(struct request
*req
, int uptodate
)
3752 __end_request(req
, uptodate
, req
->hard_cur_sectors
<< 9, 1);
3754 EXPORT_SYMBOL(end_request
);
3756 static void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3759 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3760 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
3762 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3763 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3764 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3765 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3766 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3767 rq
->buffer
= bio_data(bio
);
3768 rq
->data_len
= bio
->bi_size
;
3770 rq
->bio
= rq
->biotail
= bio
;
3773 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
3776 int kblockd_schedule_work(struct work_struct
*work
)
3778 return queue_work(kblockd_workqueue
, work
);
3781 EXPORT_SYMBOL(kblockd_schedule_work
);
3783 void kblockd_flush_work(struct work_struct
*work
)
3785 cancel_work_sync(work
);
3787 EXPORT_SYMBOL(kblockd_flush_work
);
3789 int __init
blk_dev_init(void)
3793 kblockd_workqueue
= create_workqueue("kblockd");
3794 if (!kblockd_workqueue
)
3795 panic("Failed to create kblockd\n");
3797 request_cachep
= kmem_cache_create("blkdev_requests",
3798 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3800 requestq_cachep
= kmem_cache_create("blkdev_queue",
3801 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3803 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3804 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
);
3806 for_each_possible_cpu(i
)
3807 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
3809 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
3810 register_hotcpu_notifier(&blk_cpu_notifier
);
3812 blk_max_low_pfn
= max_low_pfn
- 1;
3813 blk_max_pfn
= max_pfn
- 1;
3819 * IO Context helper functions
3821 void put_io_context(struct io_context
*ioc
)
3826 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3828 if (atomic_dec_and_test(&ioc
->refcount
)) {
3829 struct cfq_io_context
*cic
;
3832 if (ioc
->aic
&& ioc
->aic
->dtor
)
3833 ioc
->aic
->dtor(ioc
->aic
);
3834 if (ioc
->cic_root
.rb_node
!= NULL
) {
3835 struct rb_node
*n
= rb_first(&ioc
->cic_root
);
3837 cic
= rb_entry(n
, struct cfq_io_context
, rb_node
);
3842 kmem_cache_free(iocontext_cachep
, ioc
);
3845 EXPORT_SYMBOL(put_io_context
);
3847 /* Called by the exitting task */
3848 void exit_io_context(void)
3850 struct io_context
*ioc
;
3851 struct cfq_io_context
*cic
;
3854 ioc
= current
->io_context
;
3855 current
->io_context
= NULL
;
3856 task_unlock(current
);
3859 if (ioc
->aic
&& ioc
->aic
->exit
)
3860 ioc
->aic
->exit(ioc
->aic
);
3861 if (ioc
->cic_root
.rb_node
!= NULL
) {
3862 cic
= rb_entry(rb_first(&ioc
->cic_root
), struct cfq_io_context
, rb_node
);
3866 put_io_context(ioc
);
3870 * If the current task has no IO context then create one and initialise it.
3871 * Otherwise, return its existing IO context.
3873 * This returned IO context doesn't have a specifically elevated refcount,
3874 * but since the current task itself holds a reference, the context can be
3875 * used in general code, so long as it stays within `current` context.
3877 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
)
3879 struct task_struct
*tsk
= current
;
3880 struct io_context
*ret
;
3882 ret
= tsk
->io_context
;
3886 ret
= kmem_cache_alloc_node(iocontext_cachep
, gfp_flags
, node
);
3888 atomic_set(&ret
->refcount
, 1);
3889 ret
->task
= current
;
3890 ret
->ioprio_changed
= 0;
3891 ret
->last_waited
= jiffies
; /* doesn't matter... */
3892 ret
->nr_batch_requests
= 0; /* because this is 0 */
3894 ret
->cic_root
.rb_node
= NULL
;
3895 ret
->ioc_data
= NULL
;
3896 /* make sure set_task_ioprio() sees the settings above */
3898 tsk
->io_context
= ret
;
3905 * If the current task has no IO context then create one and initialise it.
3906 * If it does have a context, take a ref on it.
3908 * This is always called in the context of the task which submitted the I/O.
3910 struct io_context
*get_io_context(gfp_t gfp_flags
, int node
)
3912 struct io_context
*ret
;
3913 ret
= current_io_context(gfp_flags
, node
);
3915 atomic_inc(&ret
->refcount
);
3918 EXPORT_SYMBOL(get_io_context
);
3920 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3922 struct io_context
*src
= *psrc
;
3923 struct io_context
*dst
= *pdst
;
3926 BUG_ON(atomic_read(&src
->refcount
) == 0);
3927 atomic_inc(&src
->refcount
);
3928 put_io_context(dst
);
3932 EXPORT_SYMBOL(copy_io_context
);
3934 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3936 struct io_context
*temp
;
3941 EXPORT_SYMBOL(swap_io_context
);
3946 struct queue_sysfs_entry
{
3947 struct attribute attr
;
3948 ssize_t (*show
)(struct request_queue
*, char *);
3949 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3953 queue_var_show(unsigned int var
, char *page
)
3955 return sprintf(page
, "%d\n", var
);
3959 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3961 char *p
= (char *) page
;
3963 *var
= simple_strtoul(p
, &p
, 10);
3967 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3969 return queue_var_show(q
->nr_requests
, (page
));
3973 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3975 struct request_list
*rl
= &q
->rq
;
3977 int ret
= queue_var_store(&nr
, page
, count
);
3978 if (nr
< BLKDEV_MIN_RQ
)
3981 spin_lock_irq(q
->queue_lock
);
3982 q
->nr_requests
= nr
;
3983 blk_queue_congestion_threshold(q
);
3985 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3986 blk_set_queue_congested(q
, READ
);
3987 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3988 blk_clear_queue_congested(q
, READ
);
3990 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3991 blk_set_queue_congested(q
, WRITE
);
3992 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3993 blk_clear_queue_congested(q
, WRITE
);
3995 if (rl
->count
[READ
] >= q
->nr_requests
) {
3996 blk_set_queue_full(q
, READ
);
3997 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3998 blk_clear_queue_full(q
, READ
);
3999 wake_up(&rl
->wait
[READ
]);
4002 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
4003 blk_set_queue_full(q
, WRITE
);
4004 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
4005 blk_clear_queue_full(q
, WRITE
);
4006 wake_up(&rl
->wait
[WRITE
]);
4008 spin_unlock_irq(q
->queue_lock
);
4012 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
4014 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
4016 return queue_var_show(ra_kb
, (page
));
4020 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
4022 unsigned long ra_kb
;
4023 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
4025 spin_lock_irq(q
->queue_lock
);
4026 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
4027 spin_unlock_irq(q
->queue_lock
);
4032 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
4034 int max_sectors_kb
= q
->max_sectors
>> 1;
4036 return queue_var_show(max_sectors_kb
, (page
));
4040 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
4042 unsigned long max_sectors_kb
,
4043 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
4044 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
4045 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
4047 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
4050 * Take the queue lock to update the readahead and max_sectors
4051 * values synchronously:
4053 spin_lock_irq(q
->queue_lock
);
4054 q
->max_sectors
= max_sectors_kb
<< 1;
4055 spin_unlock_irq(q
->queue_lock
);
4060 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
4062 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
4064 return queue_var_show(max_hw_sectors_kb
, (page
));
4067 static ssize_t
queue_max_segments_show(struct request_queue
*q
, char *page
)
4069 return queue_var_show(q
->max_phys_segments
, page
);
4072 static ssize_t
queue_max_segments_store(struct request_queue
*q
,
4073 const char *page
, size_t count
)
4075 unsigned long segments
;
4076 ssize_t ret
= queue_var_store(&segments
, page
, count
);
4078 spin_lock_irq(q
->queue_lock
);
4079 q
->max_phys_segments
= segments
;
4080 spin_unlock_irq(q
->queue_lock
);
4084 static struct queue_sysfs_entry queue_requests_entry
= {
4085 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
4086 .show
= queue_requests_show
,
4087 .store
= queue_requests_store
,
4090 static struct queue_sysfs_entry queue_ra_entry
= {
4091 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
4092 .show
= queue_ra_show
,
4093 .store
= queue_ra_store
,
4096 static struct queue_sysfs_entry queue_max_sectors_entry
= {
4097 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
4098 .show
= queue_max_sectors_show
,
4099 .store
= queue_max_sectors_store
,
4102 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
4103 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
4104 .show
= queue_max_hw_sectors_show
,
4107 static struct queue_sysfs_entry queue_max_segments_entry
= {
4108 .attr
= {.name
= "max_segments", .mode
= S_IRUGO
| S_IWUSR
},
4109 .show
= queue_max_segments_show
,
4110 .store
= queue_max_segments_store
,
4113 static struct queue_sysfs_entry queue_iosched_entry
= {
4114 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
4115 .show
= elv_iosched_show
,
4116 .store
= elv_iosched_store
,
4119 static struct attribute
*default_attrs
[] = {
4120 &queue_requests_entry
.attr
,
4121 &queue_ra_entry
.attr
,
4122 &queue_max_hw_sectors_entry
.attr
,
4123 &queue_max_sectors_entry
.attr
,
4124 &queue_max_segments_entry
.attr
,
4125 &queue_iosched_entry
.attr
,
4129 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4132 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4134 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4135 struct request_queue
*q
=
4136 container_of(kobj
, struct request_queue
, kobj
);
4141 mutex_lock(&q
->sysfs_lock
);
4142 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4143 mutex_unlock(&q
->sysfs_lock
);
4146 res
= entry
->show(q
, page
);
4147 mutex_unlock(&q
->sysfs_lock
);
4152 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4153 const char *page
, size_t length
)
4155 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4156 struct request_queue
*q
= container_of(kobj
, struct request_queue
, kobj
);
4162 mutex_lock(&q
->sysfs_lock
);
4163 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4164 mutex_unlock(&q
->sysfs_lock
);
4167 res
= entry
->store(q
, page
, length
);
4168 mutex_unlock(&q
->sysfs_lock
);
4172 static struct sysfs_ops queue_sysfs_ops
= {
4173 .show
= queue_attr_show
,
4174 .store
= queue_attr_store
,
4177 static struct kobj_type queue_ktype
= {
4178 .sysfs_ops
= &queue_sysfs_ops
,
4179 .default_attrs
= default_attrs
,
4180 .release
= blk_release_queue
,
4183 int blk_register_queue(struct gendisk
*disk
)
4187 struct request_queue
*q
= disk
->queue
;
4189 if (!q
|| !q
->request_fn
)
4192 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
4194 ret
= kobject_add(&q
->kobj
);
4198 kobject_uevent(&q
->kobj
, KOBJ_ADD
);
4200 ret
= elv_register_queue(q
);
4202 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4203 kobject_del(&q
->kobj
);
4210 void blk_unregister_queue(struct gendisk
*disk
)
4212 struct request_queue
*q
= disk
->queue
;
4214 if (q
&& q
->request_fn
) {
4215 elv_unregister_queue(q
);
4217 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4218 kobject_del(&q
->kobj
);
4219 kobject_put(&disk
->kobj
);