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 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_update_dma_alignment - update dma length and memory alignment
764 * @q: the request queue for the device
765 * @mask: alignment mask
768 * update required memory and length aligment for direct dma transactions.
769 * If the requested alignment is larger than the current alignment, then
770 * the current queue alignment is updated to the new value, otherwise it
771 * is left alone. The design of this is to allow multiple objects
772 * (driver, device, transport etc) to set their respective
773 * alignments without having them interfere.
776 void blk_queue_update_dma_alignment(struct request_queue
*q
, int mask
)
778 BUG_ON(mask
> PAGE_SIZE
);
780 if (mask
> q
->dma_alignment
)
781 q
->dma_alignment
= mask
;
784 EXPORT_SYMBOL(blk_queue_update_dma_alignment
);
787 * blk_queue_find_tag - find a request by its tag and queue
788 * @q: The request queue for the device
789 * @tag: The tag of the request
792 * Should be used when a device returns a tag and you want to match
795 * no locks need be held.
797 struct request
*blk_queue_find_tag(struct request_queue
*q
, int tag
)
799 return blk_map_queue_find_tag(q
->queue_tags
, tag
);
802 EXPORT_SYMBOL(blk_queue_find_tag
);
805 * __blk_free_tags - release a given set of tag maintenance info
806 * @bqt: the tag map to free
808 * Tries to free the specified @bqt@. Returns true if it was
809 * actually freed and false if there are still references using it
811 static int __blk_free_tags(struct blk_queue_tag
*bqt
)
815 retval
= atomic_dec_and_test(&bqt
->refcnt
);
819 kfree(bqt
->tag_index
);
820 bqt
->tag_index
= NULL
;
833 * __blk_queue_free_tags - release tag maintenance info
834 * @q: the request queue for the device
837 * blk_cleanup_queue() will take care of calling this function, if tagging
838 * has been used. So there's no need to call this directly.
840 static void __blk_queue_free_tags(struct request_queue
*q
)
842 struct blk_queue_tag
*bqt
= q
->queue_tags
;
847 __blk_free_tags(bqt
);
849 q
->queue_tags
= NULL
;
850 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
855 * blk_free_tags - release a given set of tag maintenance info
856 * @bqt: the tag map to free
858 * For externally managed @bqt@ frees the map. Callers of this
859 * function must guarantee to have released all the queues that
860 * might have been using this tag map.
862 void blk_free_tags(struct blk_queue_tag
*bqt
)
864 if (unlikely(!__blk_free_tags(bqt
)))
867 EXPORT_SYMBOL(blk_free_tags
);
870 * blk_queue_free_tags - release tag maintenance info
871 * @q: the request queue for the device
874 * This is used to disabled tagged queuing to a device, yet leave
877 void blk_queue_free_tags(struct request_queue
*q
)
879 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
882 EXPORT_SYMBOL(blk_queue_free_tags
);
885 init_tag_map(struct request_queue
*q
, struct blk_queue_tag
*tags
, int depth
)
887 struct request
**tag_index
;
888 unsigned long *tag_map
;
891 if (q
&& depth
> q
->nr_requests
* 2) {
892 depth
= q
->nr_requests
* 2;
893 printk(KERN_ERR
"%s: adjusted depth to %d\n",
894 __FUNCTION__
, depth
);
897 tag_index
= kzalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
901 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
902 tag_map
= kzalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
906 tags
->real_max_depth
= depth
;
907 tags
->max_depth
= depth
;
908 tags
->tag_index
= tag_index
;
909 tags
->tag_map
= tag_map
;
917 static struct blk_queue_tag
*__blk_queue_init_tags(struct request_queue
*q
,
920 struct blk_queue_tag
*tags
;
922 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
926 if (init_tag_map(q
, tags
, depth
))
930 atomic_set(&tags
->refcnt
, 1);
938 * blk_init_tags - initialize the tag info for an external tag map
939 * @depth: the maximum queue depth supported
940 * @tags: the tag to use
942 struct blk_queue_tag
*blk_init_tags(int depth
)
944 return __blk_queue_init_tags(NULL
, depth
);
946 EXPORT_SYMBOL(blk_init_tags
);
949 * blk_queue_init_tags - initialize the queue tag info
950 * @q: the request queue for the device
951 * @depth: the maximum queue depth supported
952 * @tags: the tag to use
954 int blk_queue_init_tags(struct request_queue
*q
, int depth
,
955 struct blk_queue_tag
*tags
)
959 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
961 if (!tags
&& !q
->queue_tags
) {
962 tags
= __blk_queue_init_tags(q
, depth
);
966 } else if (q
->queue_tags
) {
967 if ((rc
= blk_queue_resize_tags(q
, depth
)))
969 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
972 atomic_inc(&tags
->refcnt
);
975 * assign it, all done
977 q
->queue_tags
= tags
;
978 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
979 INIT_LIST_HEAD(&q
->tag_busy_list
);
986 EXPORT_SYMBOL(blk_queue_init_tags
);
989 * blk_queue_resize_tags - change the queueing depth
990 * @q: the request queue for the device
991 * @new_depth: the new max command queueing depth
994 * Must be called with the queue lock held.
996 int blk_queue_resize_tags(struct request_queue
*q
, int new_depth
)
998 struct blk_queue_tag
*bqt
= q
->queue_tags
;
999 struct request
**tag_index
;
1000 unsigned long *tag_map
;
1001 int max_depth
, nr_ulongs
;
1007 * if we already have large enough real_max_depth. just
1008 * adjust max_depth. *NOTE* as requests with tag value
1009 * between new_depth and real_max_depth can be in-flight, tag
1010 * map can not be shrunk blindly here.
1012 if (new_depth
<= bqt
->real_max_depth
) {
1013 bqt
->max_depth
= new_depth
;
1018 * Currently cannot replace a shared tag map with a new
1019 * one, so error out if this is the case
1021 if (atomic_read(&bqt
->refcnt
) != 1)
1025 * save the old state info, so we can copy it back
1027 tag_index
= bqt
->tag_index
;
1028 tag_map
= bqt
->tag_map
;
1029 max_depth
= bqt
->real_max_depth
;
1031 if (init_tag_map(q
, bqt
, new_depth
))
1034 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
1035 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
1036 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
1043 EXPORT_SYMBOL(blk_queue_resize_tags
);
1046 * blk_queue_end_tag - end tag operations for a request
1047 * @q: the request queue for the device
1048 * @rq: the request that has completed
1051 * Typically called when end_that_request_first() returns 0, meaning
1052 * all transfers have been done for a request. It's important to call
1053 * this function before end_that_request_last(), as that will put the
1054 * request back on the free list thus corrupting the internal tag list.
1057 * queue lock must be held.
1059 void blk_queue_end_tag(struct request_queue
*q
, struct request
*rq
)
1061 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1066 if (unlikely(tag
>= bqt
->real_max_depth
))
1068 * This can happen after tag depth has been reduced.
1069 * FIXME: how about a warning or info message here?
1073 list_del_init(&rq
->queuelist
);
1074 rq
->cmd_flags
&= ~REQ_QUEUED
;
1077 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
1078 printk(KERN_ERR
"%s: tag %d is missing\n",
1081 bqt
->tag_index
[tag
] = NULL
;
1083 if (unlikely(!test_bit(tag
, bqt
->tag_map
))) {
1084 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
1089 * The tag_map bit acts as a lock for tag_index[bit], so we need
1090 * unlock memory barrier semantics.
1092 clear_bit_unlock(tag
, bqt
->tag_map
);
1096 EXPORT_SYMBOL(blk_queue_end_tag
);
1099 * blk_queue_start_tag - find a free tag and assign it
1100 * @q: the request queue for the device
1101 * @rq: the block request that needs tagging
1104 * This can either be used as a stand-alone helper, or possibly be
1105 * assigned as the queue &prep_rq_fn (in which case &struct request
1106 * automagically gets a tag assigned). Note that this function
1107 * assumes that any type of request can be queued! if this is not
1108 * true for your device, you must check the request type before
1109 * calling this function. The request will also be removed from
1110 * the request queue, so it's the drivers responsibility to readd
1111 * it if it should need to be restarted for some reason.
1114 * queue lock must be held.
1116 int blk_queue_start_tag(struct request_queue
*q
, struct request
*rq
)
1118 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1121 if (unlikely((rq
->cmd_flags
& REQ_QUEUED
))) {
1123 "%s: request %p for device [%s] already tagged %d",
1125 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
1130 * Protect against shared tag maps, as we may not have exclusive
1131 * access to the tag map.
1134 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
1135 if (tag
>= bqt
->max_depth
)
1138 } while (test_and_set_bit_lock(tag
, bqt
->tag_map
));
1140 * We need lock ordering semantics given by test_and_set_bit_lock.
1141 * See blk_queue_end_tag for details.
1144 rq
->cmd_flags
|= REQ_QUEUED
;
1146 bqt
->tag_index
[tag
] = rq
;
1147 blkdev_dequeue_request(rq
);
1148 list_add(&rq
->queuelist
, &q
->tag_busy_list
);
1153 EXPORT_SYMBOL(blk_queue_start_tag
);
1156 * blk_queue_invalidate_tags - invalidate all pending tags
1157 * @q: the request queue for the device
1160 * Hardware conditions may dictate a need to stop all pending requests.
1161 * In this case, we will safely clear the block side of the tag queue and
1162 * readd all requests to the request queue in the right order.
1165 * queue lock must be held.
1167 void blk_queue_invalidate_tags(struct request_queue
*q
)
1169 struct list_head
*tmp
, *n
;
1171 list_for_each_safe(tmp
, n
, &q
->tag_busy_list
)
1172 blk_requeue_request(q
, list_entry_rq(tmp
));
1175 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1177 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1181 printk("%s: dev %s: type=%x, flags=%x\n", msg
,
1182 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
1185 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1187 rq
->current_nr_sectors
);
1188 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1190 if (blk_pc_request(rq
)) {
1192 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1193 printk("%02x ", rq
->cmd
[bit
]);
1198 EXPORT_SYMBOL(blk_dump_rq_flags
);
1200 void blk_recount_segments(struct request_queue
*q
, struct bio
*bio
)
1203 struct bio
*nxt
= bio
->bi_next
;
1205 rq
.bio
= rq
.biotail
= bio
;
1206 bio
->bi_next
= NULL
;
1207 blk_recalc_rq_segments(&rq
);
1209 bio
->bi_phys_segments
= rq
.nr_phys_segments
;
1210 bio
->bi_hw_segments
= rq
.nr_hw_segments
;
1211 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1213 EXPORT_SYMBOL(blk_recount_segments
);
1215 static void blk_recalc_rq_segments(struct request
*rq
)
1219 unsigned int phys_size
;
1220 unsigned int hw_size
;
1221 struct bio_vec
*bv
, *bvprv
= NULL
;
1225 struct req_iterator iter
;
1226 int high
, highprv
= 1;
1227 struct request_queue
*q
= rq
->q
;
1232 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1233 hw_seg_size
= seg_size
= 0;
1234 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
1235 rq_for_each_segment(bv
, rq
, iter
) {
1237 * the trick here is making sure that a high page is never
1238 * considered part of another segment, since that might
1239 * change with the bounce page.
1241 high
= page_to_pfn(bv
->bv_page
) > q
->bounce_pfn
;
1242 if (high
|| highprv
)
1243 goto new_hw_segment
;
1245 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1247 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1249 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1251 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1252 goto new_hw_segment
;
1254 seg_size
+= bv
->bv_len
;
1255 hw_seg_size
+= bv
->bv_len
;
1260 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1261 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1262 hw_seg_size
+= bv
->bv_len
;
1265 if (nr_hw_segs
== 1 &&
1266 hw_seg_size
> rq
->bio
->bi_hw_front_size
)
1267 rq
->bio
->bi_hw_front_size
= hw_seg_size
;
1268 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1274 seg_size
= bv
->bv_len
;
1278 if (nr_hw_segs
== 1 &&
1279 hw_seg_size
> rq
->bio
->bi_hw_front_size
)
1280 rq
->bio
->bi_hw_front_size
= hw_seg_size
;
1281 if (hw_seg_size
> rq
->biotail
->bi_hw_back_size
)
1282 rq
->biotail
->bi_hw_back_size
= hw_seg_size
;
1283 rq
->nr_phys_segments
= nr_phys_segs
;
1284 rq
->nr_hw_segments
= nr_hw_segs
;
1287 static int blk_phys_contig_segment(struct request_queue
*q
, struct bio
*bio
,
1290 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1293 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1295 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1299 * bio and nxt are contigous in memory, check if the queue allows
1300 * these two to be merged into one
1302 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1308 static int blk_hw_contig_segment(struct request_queue
*q
, struct bio
*bio
,
1311 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1312 blk_recount_segments(q
, bio
);
1313 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1314 blk_recount_segments(q
, nxt
);
1315 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1316 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_back_size
+ nxt
->bi_hw_front_size
))
1318 if (bio
->bi_hw_back_size
+ nxt
->bi_hw_front_size
> q
->max_segment_size
)
1325 * map a request to scatterlist, return number of sg entries setup. Caller
1326 * must make sure sg can hold rq->nr_phys_segments entries
1328 int blk_rq_map_sg(struct request_queue
*q
, struct request
*rq
,
1329 struct scatterlist
*sglist
)
1331 struct bio_vec
*bvec
, *bvprv
;
1332 struct req_iterator iter
;
1333 struct scatterlist
*sg
;
1337 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1340 * for each bio in rq
1344 rq_for_each_segment(bvec
, rq
, iter
) {
1345 int nbytes
= bvec
->bv_len
;
1347 if (bvprv
&& cluster
) {
1348 if (sg
->length
+ nbytes
> q
->max_segment_size
)
1351 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1353 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1356 sg
->length
+= nbytes
;
1363 * If the driver previously mapped a shorter
1364 * list, we could see a termination bit
1365 * prematurely unless it fully inits the sg
1366 * table on each mapping. We KNOW that there
1367 * must be more entries here or the driver
1368 * would be buggy, so force clear the
1369 * termination bit to avoid doing a full
1370 * sg_init_table() in drivers for each command.
1372 sg
->page_link
&= ~0x02;
1376 sg_set_page(sg
, bvec
->bv_page
, nbytes
, bvec
->bv_offset
);
1380 } /* segments in rq */
1388 EXPORT_SYMBOL(blk_rq_map_sg
);
1391 * the standard queue merge functions, can be overridden with device
1392 * specific ones if so desired
1395 static inline int ll_new_mergeable(struct request_queue
*q
,
1396 struct request
*req
,
1399 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1401 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1402 req
->cmd_flags
|= REQ_NOMERGE
;
1403 if (req
== q
->last_merge
)
1404 q
->last_merge
= NULL
;
1409 * A hw segment is just getting larger, bump just the phys
1412 req
->nr_phys_segments
+= nr_phys_segs
;
1416 static inline int ll_new_hw_segment(struct request_queue
*q
,
1417 struct request
*req
,
1420 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1421 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1423 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1424 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1425 req
->cmd_flags
|= REQ_NOMERGE
;
1426 if (req
== q
->last_merge
)
1427 q
->last_merge
= NULL
;
1432 * This will form the start of a new hw segment. Bump both
1435 req
->nr_hw_segments
+= nr_hw_segs
;
1436 req
->nr_phys_segments
+= nr_phys_segs
;
1440 static int ll_back_merge_fn(struct request_queue
*q
, struct request
*req
,
1443 unsigned short max_sectors
;
1446 if (unlikely(blk_pc_request(req
)))
1447 max_sectors
= q
->max_hw_sectors
;
1449 max_sectors
= q
->max_sectors
;
1451 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1452 req
->cmd_flags
|= REQ_NOMERGE
;
1453 if (req
== q
->last_merge
)
1454 q
->last_merge
= NULL
;
1457 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1458 blk_recount_segments(q
, req
->biotail
);
1459 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1460 blk_recount_segments(q
, bio
);
1461 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1462 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1463 !BIOVEC_VIRT_OVERSIZE(len
)) {
1464 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1467 if (req
->nr_hw_segments
== 1)
1468 req
->bio
->bi_hw_front_size
= len
;
1469 if (bio
->bi_hw_segments
== 1)
1470 bio
->bi_hw_back_size
= len
;
1475 return ll_new_hw_segment(q
, req
, bio
);
1478 static int ll_front_merge_fn(struct request_queue
*q
, struct request
*req
,
1481 unsigned short max_sectors
;
1484 if (unlikely(blk_pc_request(req
)))
1485 max_sectors
= q
->max_hw_sectors
;
1487 max_sectors
= q
->max_sectors
;
1490 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1491 req
->cmd_flags
|= REQ_NOMERGE
;
1492 if (req
== q
->last_merge
)
1493 q
->last_merge
= NULL
;
1496 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1497 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1498 blk_recount_segments(q
, bio
);
1499 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1500 blk_recount_segments(q
, req
->bio
);
1501 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1502 !BIOVEC_VIRT_OVERSIZE(len
)) {
1503 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1506 if (bio
->bi_hw_segments
== 1)
1507 bio
->bi_hw_front_size
= len
;
1508 if (req
->nr_hw_segments
== 1)
1509 req
->biotail
->bi_hw_back_size
= len
;
1514 return ll_new_hw_segment(q
, req
, bio
);
1517 static int ll_merge_requests_fn(struct request_queue
*q
, struct request
*req
,
1518 struct request
*next
)
1520 int total_phys_segments
;
1521 int total_hw_segments
;
1524 * First check if the either of the requests are re-queued
1525 * requests. Can't merge them if they are.
1527 if (req
->special
|| next
->special
)
1531 * Will it become too large?
1533 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1536 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1537 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1538 total_phys_segments
--;
1540 if (total_phys_segments
> q
->max_phys_segments
)
1543 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1544 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1545 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1547 * propagate the combined length to the end of the requests
1549 if (req
->nr_hw_segments
== 1)
1550 req
->bio
->bi_hw_front_size
= len
;
1551 if (next
->nr_hw_segments
== 1)
1552 next
->biotail
->bi_hw_back_size
= len
;
1553 total_hw_segments
--;
1556 if (total_hw_segments
> q
->max_hw_segments
)
1559 /* Merge is OK... */
1560 req
->nr_phys_segments
= total_phys_segments
;
1561 req
->nr_hw_segments
= total_hw_segments
;
1566 * "plug" the device if there are no outstanding requests: this will
1567 * force the transfer to start only after we have put all the requests
1570 * This is called with interrupts off and no requests on the queue and
1571 * with the queue lock held.
1573 void blk_plug_device(struct request_queue
*q
)
1575 WARN_ON(!irqs_disabled());
1578 * don't plug a stopped queue, it must be paired with blk_start_queue()
1579 * which will restart the queueing
1581 if (blk_queue_stopped(q
))
1584 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
1585 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1586 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
1590 EXPORT_SYMBOL(blk_plug_device
);
1593 * remove the queue from the plugged list, if present. called with
1594 * queue lock held and interrupts disabled.
1596 int blk_remove_plug(struct request_queue
*q
)
1598 WARN_ON(!irqs_disabled());
1600 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1603 del_timer(&q
->unplug_timer
);
1607 EXPORT_SYMBOL(blk_remove_plug
);
1610 * remove the plug and let it rip..
1612 void __generic_unplug_device(struct request_queue
*q
)
1614 if (unlikely(blk_queue_stopped(q
)))
1617 if (!blk_remove_plug(q
))
1622 EXPORT_SYMBOL(__generic_unplug_device
);
1625 * generic_unplug_device - fire a request queue
1626 * @q: The &struct request_queue in question
1629 * Linux uses plugging to build bigger requests queues before letting
1630 * the device have at them. If a queue is plugged, the I/O scheduler
1631 * is still adding and merging requests on the queue. Once the queue
1632 * gets unplugged, the request_fn defined for the queue is invoked and
1633 * transfers started.
1635 void generic_unplug_device(struct request_queue
*q
)
1637 spin_lock_irq(q
->queue_lock
);
1638 __generic_unplug_device(q
);
1639 spin_unlock_irq(q
->queue_lock
);
1641 EXPORT_SYMBOL(generic_unplug_device
);
1643 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1646 struct request_queue
*q
= bdi
->unplug_io_data
;
1651 static void blk_unplug_work(struct work_struct
*work
)
1653 struct request_queue
*q
=
1654 container_of(work
, struct request_queue
, unplug_work
);
1656 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1657 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1662 static void blk_unplug_timeout(unsigned long data
)
1664 struct request_queue
*q
= (struct request_queue
*)data
;
1666 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
1667 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1669 kblockd_schedule_work(&q
->unplug_work
);
1672 void blk_unplug(struct request_queue
*q
)
1675 * devices don't necessarily have an ->unplug_fn defined
1678 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1679 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1684 EXPORT_SYMBOL(blk_unplug
);
1687 * blk_start_queue - restart a previously stopped queue
1688 * @q: The &struct request_queue in question
1691 * blk_start_queue() will clear the stop flag on the queue, and call
1692 * the request_fn for the queue if it was in a stopped state when
1693 * entered. Also see blk_stop_queue(). Queue lock must be held.
1695 void blk_start_queue(struct request_queue
*q
)
1697 WARN_ON(!irqs_disabled());
1699 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1702 * one level of recursion is ok and is much faster than kicking
1703 * the unplug handling
1705 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1707 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1710 kblockd_schedule_work(&q
->unplug_work
);
1714 EXPORT_SYMBOL(blk_start_queue
);
1717 * blk_stop_queue - stop a queue
1718 * @q: The &struct request_queue in question
1721 * The Linux block layer assumes that a block driver will consume all
1722 * entries on the request queue when the request_fn strategy is called.
1723 * Often this will not happen, because of hardware limitations (queue
1724 * depth settings). If a device driver gets a 'queue full' response,
1725 * or if it simply chooses not to queue more I/O at one point, it can
1726 * call this function to prevent the request_fn from being called until
1727 * the driver has signalled it's ready to go again. This happens by calling
1728 * blk_start_queue() to restart queue operations. Queue lock must be held.
1730 void blk_stop_queue(struct request_queue
*q
)
1733 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1735 EXPORT_SYMBOL(blk_stop_queue
);
1738 * blk_sync_queue - cancel any pending callbacks on a queue
1742 * The block layer may perform asynchronous callback activity
1743 * on a queue, such as calling the unplug function after a timeout.
1744 * A block device may call blk_sync_queue to ensure that any
1745 * such activity is cancelled, thus allowing it to release resources
1746 * that the callbacks might use. The caller must already have made sure
1747 * that its ->make_request_fn will not re-add plugging prior to calling
1751 void blk_sync_queue(struct request_queue
*q
)
1753 del_timer_sync(&q
->unplug_timer
);
1754 kblockd_flush_work(&q
->unplug_work
);
1756 EXPORT_SYMBOL(blk_sync_queue
);
1759 * blk_run_queue - run a single device queue
1760 * @q: The queue to run
1762 void blk_run_queue(struct request_queue
*q
)
1764 unsigned long flags
;
1766 spin_lock_irqsave(q
->queue_lock
, flags
);
1770 * Only recurse once to avoid overrunning the stack, let the unplug
1771 * handling reinvoke the handler shortly if we already got there.
1773 if (!elv_queue_empty(q
)) {
1774 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1776 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1779 kblockd_schedule_work(&q
->unplug_work
);
1783 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1785 EXPORT_SYMBOL(blk_run_queue
);
1788 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
1789 * @kobj: the kobj belonging of the request queue to be released
1792 * blk_cleanup_queue is the pair to blk_init_queue() or
1793 * blk_queue_make_request(). It should be called when a request queue is
1794 * being released; typically when a block device is being de-registered.
1795 * Currently, its primary task it to free all the &struct request
1796 * structures that were allocated to the queue and the queue itself.
1799 * Hopefully the low level driver will have finished any
1800 * outstanding requests first...
1802 static void blk_release_queue(struct kobject
*kobj
)
1804 struct request_queue
*q
=
1805 container_of(kobj
, struct request_queue
, kobj
);
1806 struct request_list
*rl
= &q
->rq
;
1811 mempool_destroy(rl
->rq_pool
);
1814 __blk_queue_free_tags(q
);
1816 blk_trace_shutdown(q
);
1818 bdi_destroy(&q
->backing_dev_info
);
1819 kmem_cache_free(requestq_cachep
, q
);
1822 void blk_put_queue(struct request_queue
*q
)
1824 kobject_put(&q
->kobj
);
1826 EXPORT_SYMBOL(blk_put_queue
);
1828 void blk_cleanup_queue(struct request_queue
* q
)
1830 mutex_lock(&q
->sysfs_lock
);
1831 set_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
);
1832 mutex_unlock(&q
->sysfs_lock
);
1835 elevator_exit(q
->elevator
);
1840 EXPORT_SYMBOL(blk_cleanup_queue
);
1842 static int blk_init_free_list(struct request_queue
*q
)
1844 struct request_list
*rl
= &q
->rq
;
1846 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1847 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1849 init_waitqueue_head(&rl
->wait
[READ
]);
1850 init_waitqueue_head(&rl
->wait
[WRITE
]);
1852 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1853 mempool_free_slab
, request_cachep
, q
->node
);
1861 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
1863 return blk_alloc_queue_node(gfp_mask
, -1);
1865 EXPORT_SYMBOL(blk_alloc_queue
);
1867 static struct kobj_type queue_ktype
;
1869 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1871 struct request_queue
*q
;
1874 q
= kmem_cache_alloc_node(requestq_cachep
,
1875 gfp_mask
| __GFP_ZERO
, node_id
);
1879 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1880 q
->backing_dev_info
.unplug_io_data
= q
;
1881 err
= bdi_init(&q
->backing_dev_info
);
1883 kmem_cache_free(requestq_cachep
, q
);
1887 init_timer(&q
->unplug_timer
);
1889 kobject_set_name(&q
->kobj
, "%s", "queue");
1890 q
->kobj
.ktype
= &queue_ktype
;
1891 kobject_init(&q
->kobj
);
1893 mutex_init(&q
->sysfs_lock
);
1897 EXPORT_SYMBOL(blk_alloc_queue_node
);
1900 * blk_init_queue - prepare a request queue for use with a block device
1901 * @rfn: The function to be called to process requests that have been
1902 * placed on the queue.
1903 * @lock: Request queue spin lock
1906 * If a block device wishes to use the standard request handling procedures,
1907 * which sorts requests and coalesces adjacent requests, then it must
1908 * call blk_init_queue(). The function @rfn will be called when there
1909 * are requests on the queue that need to be processed. If the device
1910 * supports plugging, then @rfn may not be called immediately when requests
1911 * are available on the queue, but may be called at some time later instead.
1912 * Plugged queues are generally unplugged when a buffer belonging to one
1913 * of the requests on the queue is needed, or due to memory pressure.
1915 * @rfn is not required, or even expected, to remove all requests off the
1916 * queue, but only as many as it can handle at a time. If it does leave
1917 * requests on the queue, it is responsible for arranging that the requests
1918 * get dealt with eventually.
1920 * The queue spin lock must be held while manipulating the requests on the
1921 * request queue; this lock will be taken also from interrupt context, so irq
1922 * disabling is needed for it.
1924 * Function returns a pointer to the initialized request queue, or NULL if
1925 * it didn't succeed.
1928 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1929 * when the block device is deactivated (such as at module unload).
1932 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1934 return blk_init_queue_node(rfn
, lock
, -1);
1936 EXPORT_SYMBOL(blk_init_queue
);
1938 struct request_queue
*
1939 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1941 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1947 if (blk_init_free_list(q
)) {
1948 kmem_cache_free(requestq_cachep
, q
);
1953 * if caller didn't supply a lock, they get per-queue locking with
1957 spin_lock_init(&q
->__queue_lock
);
1958 lock
= &q
->__queue_lock
;
1961 q
->request_fn
= rfn
;
1962 q
->prep_rq_fn
= NULL
;
1963 q
->unplug_fn
= generic_unplug_device
;
1964 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1965 q
->queue_lock
= lock
;
1967 blk_queue_segment_boundary(q
, 0xffffffff);
1969 blk_queue_make_request(q
, __make_request
);
1970 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1972 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1973 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1975 q
->sg_reserved_size
= INT_MAX
;
1980 if (!elevator_init(q
, NULL
)) {
1981 blk_queue_congestion_threshold(q
);
1988 EXPORT_SYMBOL(blk_init_queue_node
);
1990 int blk_get_queue(struct request_queue
*q
)
1992 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1993 kobject_get(&q
->kobj
);
2000 EXPORT_SYMBOL(blk_get_queue
);
2002 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
2004 if (rq
->cmd_flags
& REQ_ELVPRIV
)
2005 elv_put_request(q
, rq
);
2006 mempool_free(rq
, q
->rq
.rq_pool
);
2009 static struct request
*
2010 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
2012 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
2018 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
2019 * see bio.h and blkdev.h
2021 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
2024 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
2025 mempool_free(rq
, q
->rq
.rq_pool
);
2028 rq
->cmd_flags
|= REQ_ELVPRIV
;
2035 * ioc_batching returns true if the ioc is a valid batching request and
2036 * should be given priority access to a request.
2038 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
2044 * Make sure the process is able to allocate at least 1 request
2045 * even if the batch times out, otherwise we could theoretically
2048 return ioc
->nr_batch_requests
== q
->nr_batching
||
2049 (ioc
->nr_batch_requests
> 0
2050 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
2054 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2055 * will cause the process to be a "batcher" on all queues in the system. This
2056 * is the behaviour we want though - once it gets a wakeup it should be given
2059 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
2061 if (!ioc
|| ioc_batching(q
, ioc
))
2064 ioc
->nr_batch_requests
= q
->nr_batching
;
2065 ioc
->last_waited
= jiffies
;
2068 static void __freed_request(struct request_queue
*q
, int rw
)
2070 struct request_list
*rl
= &q
->rq
;
2072 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
2073 blk_clear_queue_congested(q
, rw
);
2075 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
2076 if (waitqueue_active(&rl
->wait
[rw
]))
2077 wake_up(&rl
->wait
[rw
]);
2079 blk_clear_queue_full(q
, rw
);
2084 * A request has just been released. Account for it, update the full and
2085 * congestion status, wake up any waiters. Called under q->queue_lock.
2087 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
2089 struct request_list
*rl
= &q
->rq
;
2095 __freed_request(q
, rw
);
2097 if (unlikely(rl
->starved
[rw
^ 1]))
2098 __freed_request(q
, rw
^ 1);
2101 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2103 * Get a free request, queue_lock must be held.
2104 * Returns NULL on failure, with queue_lock held.
2105 * Returns !NULL on success, with queue_lock *not held*.
2107 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
2108 struct bio
*bio
, gfp_t gfp_mask
)
2110 struct request
*rq
= NULL
;
2111 struct request_list
*rl
= &q
->rq
;
2112 struct io_context
*ioc
= NULL
;
2113 const int rw
= rw_flags
& 0x01;
2114 int may_queue
, priv
;
2116 may_queue
= elv_may_queue(q
, rw_flags
);
2117 if (may_queue
== ELV_MQUEUE_NO
)
2120 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
2121 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
2122 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
2124 * The queue will fill after this allocation, so set
2125 * it as full, and mark this process as "batching".
2126 * This process will be allowed to complete a batch of
2127 * requests, others will be blocked.
2129 if (!blk_queue_full(q
, rw
)) {
2130 ioc_set_batching(q
, ioc
);
2131 blk_set_queue_full(q
, rw
);
2133 if (may_queue
!= ELV_MQUEUE_MUST
2134 && !ioc_batching(q
, ioc
)) {
2136 * The queue is full and the allocating
2137 * process is not a "batcher", and not
2138 * exempted by the IO scheduler
2144 blk_set_queue_congested(q
, rw
);
2148 * Only allow batching queuers to allocate up to 50% over the defined
2149 * limit of requests, otherwise we could have thousands of requests
2150 * allocated with any setting of ->nr_requests
2152 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
2156 rl
->starved
[rw
] = 0;
2158 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
2162 spin_unlock_irq(q
->queue_lock
);
2164 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
2165 if (unlikely(!rq
)) {
2167 * Allocation failed presumably due to memory. Undo anything
2168 * we might have messed up.
2170 * Allocating task should really be put onto the front of the
2171 * wait queue, but this is pretty rare.
2173 spin_lock_irq(q
->queue_lock
);
2174 freed_request(q
, rw
, priv
);
2177 * in the very unlikely event that allocation failed and no
2178 * requests for this direction was pending, mark us starved
2179 * so that freeing of a request in the other direction will
2180 * notice us. another possible fix would be to split the
2181 * rq mempool into READ and WRITE
2184 if (unlikely(rl
->count
[rw
] == 0))
2185 rl
->starved
[rw
] = 1;
2191 * ioc may be NULL here, and ioc_batching will be false. That's
2192 * OK, if the queue is under the request limit then requests need
2193 * not count toward the nr_batch_requests limit. There will always
2194 * be some limit enforced by BLK_BATCH_TIME.
2196 if (ioc_batching(q
, ioc
))
2197 ioc
->nr_batch_requests
--;
2201 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
2207 * No available requests for this queue, unplug the device and wait for some
2208 * requests to become available.
2210 * Called with q->queue_lock held, and returns with it unlocked.
2212 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
2215 const int rw
= rw_flags
& 0x01;
2218 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
2221 struct request_list
*rl
= &q
->rq
;
2223 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
2224 TASK_UNINTERRUPTIBLE
);
2226 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
2229 struct io_context
*ioc
;
2231 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
2233 __generic_unplug_device(q
);
2234 spin_unlock_irq(q
->queue_lock
);
2238 * After sleeping, we become a "batching" process and
2239 * will be able to allocate at least one request, and
2240 * up to a big batch of them for a small period time.
2241 * See ioc_batching, ioc_set_batching
2243 ioc
= current_io_context(GFP_NOIO
, q
->node
);
2244 ioc_set_batching(q
, ioc
);
2246 spin_lock_irq(q
->queue_lock
);
2248 finish_wait(&rl
->wait
[rw
], &wait
);
2254 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
2258 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2260 spin_lock_irq(q
->queue_lock
);
2261 if (gfp_mask
& __GFP_WAIT
) {
2262 rq
= get_request_wait(q
, rw
, NULL
);
2264 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2266 spin_unlock_irq(q
->queue_lock
);
2268 /* q->queue_lock is unlocked at this point */
2272 EXPORT_SYMBOL(blk_get_request
);
2275 * blk_start_queueing - initiate dispatch of requests to device
2276 * @q: request queue to kick into gear
2278 * This is basically a helper to remove the need to know whether a queue
2279 * is plugged or not if someone just wants to initiate dispatch of requests
2282 * The queue lock must be held with interrupts disabled.
2284 void blk_start_queueing(struct request_queue
*q
)
2286 if (!blk_queue_plugged(q
))
2289 __generic_unplug_device(q
);
2291 EXPORT_SYMBOL(blk_start_queueing
);
2294 * blk_requeue_request - put a request back on queue
2295 * @q: request queue where request should be inserted
2296 * @rq: request to be inserted
2299 * Drivers often keep queueing requests until the hardware cannot accept
2300 * more, when that condition happens we need to put the request back
2301 * on the queue. Must be called with queue lock held.
2303 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
2305 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
2307 if (blk_rq_tagged(rq
))
2308 blk_queue_end_tag(q
, rq
);
2310 elv_requeue_request(q
, rq
);
2313 EXPORT_SYMBOL(blk_requeue_request
);
2316 * blk_insert_request - insert a special request in to a request queue
2317 * @q: request queue where request should be inserted
2318 * @rq: request to be inserted
2319 * @at_head: insert request at head or tail of queue
2320 * @data: private data
2323 * Many block devices need to execute commands asynchronously, so they don't
2324 * block the whole kernel from preemption during request execution. This is
2325 * accomplished normally by inserting aritficial requests tagged as
2326 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2327 * scheduled for actual execution by the request queue.
2329 * We have the option of inserting the head or the tail of the queue.
2330 * Typically we use the tail for new ioctls and so forth. We use the head
2331 * of the queue for things like a QUEUE_FULL message from a device, or a
2332 * host that is unable to accept a particular command.
2334 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
2335 int at_head
, void *data
)
2337 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2338 unsigned long flags
;
2341 * tell I/O scheduler that this isn't a regular read/write (ie it
2342 * must not attempt merges on this) and that it acts as a soft
2345 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
2346 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
2350 spin_lock_irqsave(q
->queue_lock
, flags
);
2353 * If command is tagged, release the tag
2355 if (blk_rq_tagged(rq
))
2356 blk_queue_end_tag(q
, rq
);
2358 drive_stat_acct(rq
, 1);
2359 __elv_add_request(q
, rq
, where
, 0);
2360 blk_start_queueing(q
);
2361 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2364 EXPORT_SYMBOL(blk_insert_request
);
2366 static int __blk_rq_unmap_user(struct bio
*bio
)
2371 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2372 bio_unmap_user(bio
);
2374 ret
= bio_uncopy_user(bio
);
2380 int blk_rq_append_bio(struct request_queue
*q
, struct request
*rq
,
2384 blk_rq_bio_prep(q
, rq
, bio
);
2385 else if (!ll_back_merge_fn(q
, rq
, bio
))
2388 rq
->biotail
->bi_next
= bio
;
2391 rq
->data_len
+= bio
->bi_size
;
2395 EXPORT_SYMBOL(blk_rq_append_bio
);
2397 static int __blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
2398 void __user
*ubuf
, unsigned int len
)
2400 unsigned long uaddr
;
2401 struct bio
*bio
, *orig_bio
;
2404 reading
= rq_data_dir(rq
) == READ
;
2407 * if alignment requirement is satisfied, map in user pages for
2408 * direct dma. else, set up kernel bounce buffers
2410 uaddr
= (unsigned long) ubuf
;
2411 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2412 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2414 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2417 return PTR_ERR(bio
);
2420 blk_queue_bounce(q
, &bio
);
2423 * We link the bounce buffer in and could have to traverse it
2424 * later so we have to get a ref to prevent it from being freed
2428 ret
= blk_rq_append_bio(q
, rq
, bio
);
2430 return bio
->bi_size
;
2432 /* if it was boucned we must call the end io function */
2434 __blk_rq_unmap_user(orig_bio
);
2440 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2441 * @q: request queue where request should be inserted
2442 * @rq: request structure to fill
2443 * @ubuf: the user buffer
2444 * @len: length of user data
2447 * Data will be mapped directly for zero copy io, if possible. Otherwise
2448 * a kernel bounce buffer is used.
2450 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2451 * still in process context.
2453 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2454 * before being submitted to the device, as pages mapped may be out of
2455 * reach. It's the callers responsibility to make sure this happens. The
2456 * original bio must be passed back in to blk_rq_unmap_user() for proper
2459 int blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
2460 void __user
*ubuf
, unsigned long len
)
2462 unsigned long bytes_read
= 0;
2463 struct bio
*bio
= NULL
;
2466 if (len
> (q
->max_hw_sectors
<< 9))
2471 while (bytes_read
!= len
) {
2472 unsigned long map_len
, end
, start
;
2474 map_len
= min_t(unsigned long, len
- bytes_read
, BIO_MAX_SIZE
);
2475 end
= ((unsigned long)ubuf
+ map_len
+ PAGE_SIZE
- 1)
2477 start
= (unsigned long)ubuf
>> PAGE_SHIFT
;
2480 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2481 * pages. If this happens we just lower the requested
2482 * mapping len by a page so that we can fit
2484 if (end
- start
> BIO_MAX_PAGES
)
2485 map_len
-= PAGE_SIZE
;
2487 ret
= __blk_rq_map_user(q
, rq
, ubuf
, map_len
);
2496 rq
->buffer
= rq
->data
= NULL
;
2499 blk_rq_unmap_user(bio
);
2503 EXPORT_SYMBOL(blk_rq_map_user
);
2506 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2507 * @q: request queue where request should be inserted
2508 * @rq: request to map data to
2509 * @iov: pointer to the iovec
2510 * @iov_count: number of elements in the iovec
2511 * @len: I/O byte count
2514 * Data will be mapped directly for zero copy io, if possible. Otherwise
2515 * a kernel bounce buffer is used.
2517 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2518 * still in process context.
2520 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2521 * before being submitted to the device, as pages mapped may be out of
2522 * reach. It's the callers responsibility to make sure this happens. The
2523 * original bio must be passed back in to blk_rq_unmap_user() for proper
2526 int blk_rq_map_user_iov(struct request_queue
*q
, struct request
*rq
,
2527 struct sg_iovec
*iov
, int iov_count
, unsigned int len
)
2531 if (!iov
|| iov_count
<= 0)
2534 /* we don't allow misaligned data like bio_map_user() does. If the
2535 * user is using sg, they're expected to know the alignment constraints
2536 * and respect them accordingly */
2537 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2539 return PTR_ERR(bio
);
2541 if (bio
->bi_size
!= len
) {
2543 bio_unmap_user(bio
);
2548 blk_rq_bio_prep(q
, rq
, bio
);
2549 rq
->buffer
= rq
->data
= NULL
;
2553 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2556 * blk_rq_unmap_user - unmap a request with user data
2557 * @bio: start of bio list
2560 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2561 * supply the original rq->bio from the blk_rq_map_user() return, since
2562 * the io completion may have changed rq->bio.
2564 int blk_rq_unmap_user(struct bio
*bio
)
2566 struct bio
*mapped_bio
;
2571 if (unlikely(bio_flagged(bio
, BIO_BOUNCED
)))
2572 mapped_bio
= bio
->bi_private
;
2574 ret2
= __blk_rq_unmap_user(mapped_bio
);
2580 bio_put(mapped_bio
);
2586 EXPORT_SYMBOL(blk_rq_unmap_user
);
2589 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2590 * @q: request queue where request should be inserted
2591 * @rq: request to fill
2592 * @kbuf: the kernel buffer
2593 * @len: length of user data
2594 * @gfp_mask: memory allocation flags
2596 int blk_rq_map_kern(struct request_queue
*q
, struct request
*rq
, void *kbuf
,
2597 unsigned int len
, gfp_t gfp_mask
)
2601 if (len
> (q
->max_hw_sectors
<< 9))
2606 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2608 return PTR_ERR(bio
);
2610 if (rq_data_dir(rq
) == WRITE
)
2611 bio
->bi_rw
|= (1 << BIO_RW
);
2613 blk_rq_bio_prep(q
, rq
, bio
);
2614 blk_queue_bounce(q
, &rq
->bio
);
2615 rq
->buffer
= rq
->data
= NULL
;
2619 EXPORT_SYMBOL(blk_rq_map_kern
);
2622 * blk_execute_rq_nowait - insert a request into queue for execution
2623 * @q: queue to insert the request in
2624 * @bd_disk: matching gendisk
2625 * @rq: request to insert
2626 * @at_head: insert request at head or tail of queue
2627 * @done: I/O completion handler
2630 * Insert a fully prepared request at the back of the io scheduler queue
2631 * for execution. Don't wait for completion.
2633 void blk_execute_rq_nowait(struct request_queue
*q
, struct gendisk
*bd_disk
,
2634 struct request
*rq
, int at_head
,
2637 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2639 rq
->rq_disk
= bd_disk
;
2640 rq
->cmd_flags
|= REQ_NOMERGE
;
2642 WARN_ON(irqs_disabled());
2643 spin_lock_irq(q
->queue_lock
);
2644 __elv_add_request(q
, rq
, where
, 1);
2645 __generic_unplug_device(q
);
2646 spin_unlock_irq(q
->queue_lock
);
2648 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait
);
2651 * blk_execute_rq - insert a request into queue for execution
2652 * @q: queue to insert the request in
2653 * @bd_disk: matching gendisk
2654 * @rq: request to insert
2655 * @at_head: insert request at head or tail of queue
2658 * Insert a fully prepared request at the back of the io scheduler queue
2659 * for execution and wait for completion.
2661 int blk_execute_rq(struct request_queue
*q
, struct gendisk
*bd_disk
,
2662 struct request
*rq
, int at_head
)
2664 DECLARE_COMPLETION_ONSTACK(wait
);
2665 char sense
[SCSI_SENSE_BUFFERSIZE
];
2669 * we need an extra reference to the request, so we can look at
2670 * it after io completion
2675 memset(sense
, 0, sizeof(sense
));
2680 rq
->end_io_data
= &wait
;
2681 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2682 wait_for_completion(&wait
);
2690 EXPORT_SYMBOL(blk_execute_rq
);
2692 static void bio_end_empty_barrier(struct bio
*bio
, int err
)
2695 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2697 complete(bio
->bi_private
);
2701 * blkdev_issue_flush - queue a flush
2702 * @bdev: blockdev to issue flush for
2703 * @error_sector: error sector
2706 * Issue a flush for the block device in question. Caller can supply
2707 * room for storing the error offset in case of a flush error, if they
2708 * wish to. Caller must run wait_for_completion() on its own.
2710 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2712 DECLARE_COMPLETION_ONSTACK(wait
);
2713 struct request_queue
*q
;
2717 if (bdev
->bd_disk
== NULL
)
2720 q
= bdev_get_queue(bdev
);
2724 bio
= bio_alloc(GFP_KERNEL
, 0);
2728 bio
->bi_end_io
= bio_end_empty_barrier
;
2729 bio
->bi_private
= &wait
;
2730 bio
->bi_bdev
= bdev
;
2731 submit_bio(1 << BIO_RW_BARRIER
, bio
);
2733 wait_for_completion(&wait
);
2736 * The driver must store the error location in ->bi_sector, if
2737 * it supports it. For non-stacked drivers, this should be copied
2741 *error_sector
= bio
->bi_sector
;
2744 if (!bio_flagged(bio
, BIO_UPTODATE
))
2751 EXPORT_SYMBOL(blkdev_issue_flush
);
2753 static void drive_stat_acct(struct request
*rq
, int new_io
)
2755 int rw
= rq_data_dir(rq
);
2757 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2761 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2763 disk_round_stats(rq
->rq_disk
);
2764 rq
->rq_disk
->in_flight
++;
2769 * add-request adds a request to the linked list.
2770 * queue lock is held and interrupts disabled, as we muck with the
2771 * request queue list.
2773 static inline void add_request(struct request_queue
* q
, struct request
* req
)
2775 drive_stat_acct(req
, 1);
2778 * elevator indicated where it wants this request to be
2779 * inserted at elevator_merge time
2781 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2785 * disk_round_stats() - Round off the performance stats on a struct
2788 * The average IO queue length and utilisation statistics are maintained
2789 * by observing the current state of the queue length and the amount of
2790 * time it has been in this state for.
2792 * Normally, that accounting is done on IO completion, but that can result
2793 * in more than a second's worth of IO being accounted for within any one
2794 * second, leading to >100% utilisation. To deal with that, we call this
2795 * function to do a round-off before returning the results when reading
2796 * /proc/diskstats. This accounts immediately for all queue usage up to
2797 * the current jiffies and restarts the counters again.
2799 void disk_round_stats(struct gendisk
*disk
)
2801 unsigned long now
= jiffies
;
2803 if (now
== disk
->stamp
)
2806 if (disk
->in_flight
) {
2807 __disk_stat_add(disk
, time_in_queue
,
2808 disk
->in_flight
* (now
- disk
->stamp
));
2809 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2814 EXPORT_SYMBOL_GPL(disk_round_stats
);
2817 * queue lock must be held
2819 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
2823 if (unlikely(--req
->ref_count
))
2826 elv_completed_request(q
, req
);
2829 * Request may not have originated from ll_rw_blk. if not,
2830 * it didn't come out of our reserved rq pools
2832 if (req
->cmd_flags
& REQ_ALLOCED
) {
2833 int rw
= rq_data_dir(req
);
2834 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
2836 BUG_ON(!list_empty(&req
->queuelist
));
2837 BUG_ON(!hlist_unhashed(&req
->hash
));
2839 blk_free_request(q
, req
);
2840 freed_request(q
, rw
, priv
);
2844 EXPORT_SYMBOL_GPL(__blk_put_request
);
2846 void blk_put_request(struct request
*req
)
2848 unsigned long flags
;
2849 struct request_queue
*q
= req
->q
;
2852 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2853 * following if (q) test.
2856 spin_lock_irqsave(q
->queue_lock
, flags
);
2857 __blk_put_request(q
, req
);
2858 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2862 EXPORT_SYMBOL(blk_put_request
);
2865 * blk_end_sync_rq - executes a completion event on a request
2866 * @rq: request to complete
2867 * @error: end io status of the request
2869 void blk_end_sync_rq(struct request
*rq
, int error
)
2871 struct completion
*waiting
= rq
->end_io_data
;
2873 rq
->end_io_data
= NULL
;
2874 __blk_put_request(rq
->q
, rq
);
2877 * complete last, if this is a stack request the process (and thus
2878 * the rq pointer) could be invalid right after this complete()
2882 EXPORT_SYMBOL(blk_end_sync_rq
);
2885 * Has to be called with the request spinlock acquired
2887 static int attempt_merge(struct request_queue
*q
, struct request
*req
,
2888 struct request
*next
)
2890 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2896 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2899 if (rq_data_dir(req
) != rq_data_dir(next
)
2900 || req
->rq_disk
!= next
->rq_disk
2905 * If we are allowed to merge, then append bio list
2906 * from next to rq and release next. merge_requests_fn
2907 * will have updated segment counts, update sector
2910 if (!ll_merge_requests_fn(q
, req
, next
))
2914 * At this point we have either done a back merge
2915 * or front merge. We need the smaller start_time of
2916 * the merged requests to be the current request
2917 * for accounting purposes.
2919 if (time_after(req
->start_time
, next
->start_time
))
2920 req
->start_time
= next
->start_time
;
2922 req
->biotail
->bi_next
= next
->bio
;
2923 req
->biotail
= next
->biotail
;
2925 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2927 elv_merge_requests(q
, req
, next
);
2930 disk_round_stats(req
->rq_disk
);
2931 req
->rq_disk
->in_flight
--;
2934 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2936 __blk_put_request(q
, next
);
2940 static inline int attempt_back_merge(struct request_queue
*q
,
2943 struct request
*next
= elv_latter_request(q
, rq
);
2946 return attempt_merge(q
, rq
, next
);
2951 static inline int attempt_front_merge(struct request_queue
*q
,
2954 struct request
*prev
= elv_former_request(q
, rq
);
2957 return attempt_merge(q
, prev
, rq
);
2962 static void init_request_from_bio(struct request
*req
, struct bio
*bio
)
2964 req
->cmd_type
= REQ_TYPE_FS
;
2967 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2969 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2970 req
->cmd_flags
|= REQ_FAILFAST
;
2973 * REQ_BARRIER implies no merging, but lets make it explicit
2975 if (unlikely(bio_barrier(bio
)))
2976 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2979 req
->cmd_flags
|= REQ_RW_SYNC
;
2980 if (bio_rw_meta(bio
))
2981 req
->cmd_flags
|= REQ_RW_META
;
2984 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
2985 req
->ioprio
= bio_prio(bio
);
2986 req
->start_time
= jiffies
;
2987 blk_rq_bio_prep(req
->q
, req
, bio
);
2990 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
2992 struct request
*req
;
2993 int el_ret
, nr_sectors
, barrier
, err
;
2994 const unsigned short prio
= bio_prio(bio
);
2995 const int sync
= bio_sync(bio
);
2998 nr_sectors
= bio_sectors(bio
);
3001 * low level driver can indicate that it wants pages above a
3002 * certain limit bounced to low memory (ie for highmem, or even
3003 * ISA dma in theory)
3005 blk_queue_bounce(q
, &bio
);
3007 barrier
= bio_barrier(bio
);
3008 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
3013 spin_lock_irq(q
->queue_lock
);
3015 if (unlikely(barrier
) || elv_queue_empty(q
))
3018 el_ret
= elv_merge(q
, &req
, bio
);
3020 case ELEVATOR_BACK_MERGE
:
3021 BUG_ON(!rq_mergeable(req
));
3023 if (!ll_back_merge_fn(q
, req
, bio
))
3026 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
3028 req
->biotail
->bi_next
= bio
;
3030 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
3031 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
3032 drive_stat_acct(req
, 0);
3033 if (!attempt_back_merge(q
, req
))
3034 elv_merged_request(q
, req
, el_ret
);
3037 case ELEVATOR_FRONT_MERGE
:
3038 BUG_ON(!rq_mergeable(req
));
3040 if (!ll_front_merge_fn(q
, req
, bio
))
3043 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
3045 bio
->bi_next
= req
->bio
;
3049 * may not be valid. if the low level driver said
3050 * it didn't need a bounce buffer then it better
3051 * not touch req->buffer either...
3053 req
->buffer
= bio_data(bio
);
3054 req
->current_nr_sectors
= bio_cur_sectors(bio
);
3055 req
->hard_cur_sectors
= req
->current_nr_sectors
;
3056 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
3057 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
3058 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
3059 drive_stat_acct(req
, 0);
3060 if (!attempt_front_merge(q
, req
))
3061 elv_merged_request(q
, req
, el_ret
);
3064 /* ELV_NO_MERGE: elevator says don't/can't merge. */
3071 * This sync check and mask will be re-done in init_request_from_bio(),
3072 * but we need to set it earlier to expose the sync flag to the
3073 * rq allocator and io schedulers.
3075 rw_flags
= bio_data_dir(bio
);
3077 rw_flags
|= REQ_RW_SYNC
;
3080 * Grab a free request. This is might sleep but can not fail.
3081 * Returns with the queue unlocked.
3083 req
= get_request_wait(q
, rw_flags
, bio
);
3086 * After dropping the lock and possibly sleeping here, our request
3087 * may now be mergeable after it had proven unmergeable (above).
3088 * We don't worry about that case for efficiency. It won't happen
3089 * often, and the elevators are able to handle it.
3091 init_request_from_bio(req
, bio
);
3093 spin_lock_irq(q
->queue_lock
);
3094 if (elv_queue_empty(q
))
3096 add_request(q
, req
);
3099 __generic_unplug_device(q
);
3101 spin_unlock_irq(q
->queue_lock
);
3105 bio_endio(bio
, err
);
3110 * If bio->bi_dev is a partition, remap the location
3112 static inline void blk_partition_remap(struct bio
*bio
)
3114 struct block_device
*bdev
= bio
->bi_bdev
;
3116 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
3117 struct hd_struct
*p
= bdev
->bd_part
;
3118 const int rw
= bio_data_dir(bio
);
3120 p
->sectors
[rw
] += bio_sectors(bio
);
3123 bio
->bi_sector
+= p
->start_sect
;
3124 bio
->bi_bdev
= bdev
->bd_contains
;
3126 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
3127 bdev
->bd_dev
, bio
->bi_sector
,
3128 bio
->bi_sector
- p
->start_sect
);
3132 static void handle_bad_sector(struct bio
*bio
)
3134 char b
[BDEVNAME_SIZE
];
3136 printk(KERN_INFO
"attempt to access beyond end of device\n");
3137 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
3138 bdevname(bio
->bi_bdev
, b
),
3140 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
3141 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
3143 set_bit(BIO_EOF
, &bio
->bi_flags
);
3146 #ifdef CONFIG_FAIL_MAKE_REQUEST
3148 static DECLARE_FAULT_ATTR(fail_make_request
);
3150 static int __init
setup_fail_make_request(char *str
)
3152 return setup_fault_attr(&fail_make_request
, str
);
3154 __setup("fail_make_request=", setup_fail_make_request
);
3156 static int should_fail_request(struct bio
*bio
)
3158 if ((bio
->bi_bdev
->bd_disk
->flags
& GENHD_FL_FAIL
) ||
3159 (bio
->bi_bdev
->bd_part
&& bio
->bi_bdev
->bd_part
->make_it_fail
))
3160 return should_fail(&fail_make_request
, bio
->bi_size
);
3165 static int __init
fail_make_request_debugfs(void)
3167 return init_fault_attr_dentries(&fail_make_request
,
3168 "fail_make_request");
3171 late_initcall(fail_make_request_debugfs
);
3173 #else /* CONFIG_FAIL_MAKE_REQUEST */
3175 static inline int should_fail_request(struct bio
*bio
)
3180 #endif /* CONFIG_FAIL_MAKE_REQUEST */
3183 * Check whether this bio extends beyond the end of the device.
3185 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
3192 /* Test device or partition size, when known. */
3193 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3195 sector_t sector
= bio
->bi_sector
;
3197 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
3199 * This may well happen - the kernel calls bread()
3200 * without checking the size of the device, e.g., when
3201 * mounting a device.
3203 handle_bad_sector(bio
);
3212 * generic_make_request: hand a buffer to its device driver for I/O
3213 * @bio: The bio describing the location in memory and on the device.
3215 * generic_make_request() is used to make I/O requests of block
3216 * devices. It is passed a &struct bio, which describes the I/O that needs
3219 * generic_make_request() does not return any status. The
3220 * success/failure status of the request, along with notification of
3221 * completion, is delivered asynchronously through the bio->bi_end_io
3222 * function described (one day) else where.
3224 * The caller of generic_make_request must make sure that bi_io_vec
3225 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3226 * set to describe the device address, and the
3227 * bi_end_io and optionally bi_private are set to describe how
3228 * completion notification should be signaled.
3230 * generic_make_request and the drivers it calls may use bi_next if this
3231 * bio happens to be merged with someone else, and may change bi_dev and
3232 * bi_sector for remaps as it sees fit. So the values of these fields
3233 * should NOT be depended on after the call to generic_make_request.
3235 static inline void __generic_make_request(struct bio
*bio
)
3237 struct request_queue
*q
;
3238 sector_t old_sector
;
3239 int ret
, nr_sectors
= bio_sectors(bio
);
3245 if (bio_check_eod(bio
, nr_sectors
))
3249 * Resolve the mapping until finished. (drivers are
3250 * still free to implement/resolve their own stacking
3251 * by explicitly returning 0)
3253 * NOTE: we don't repeat the blk_size check for each new device.
3254 * Stacking drivers are expected to know what they are doing.
3259 char b
[BDEVNAME_SIZE
];
3261 q
= bdev_get_queue(bio
->bi_bdev
);
3264 "generic_make_request: Trying to access "
3265 "nonexistent block-device %s (%Lu)\n",
3266 bdevname(bio
->bi_bdev
, b
),
3267 (long long) bio
->bi_sector
);
3269 bio_endio(bio
, err
);
3273 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
3274 printk("bio too big device %s (%u > %u)\n",
3275 bdevname(bio
->bi_bdev
, b
),
3281 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
3284 if (should_fail_request(bio
))
3288 * If this device has partitions, remap block n
3289 * of partition p to block n+start(p) of the disk.
3291 blk_partition_remap(bio
);
3293 if (old_sector
!= -1)
3294 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
3297 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
3299 old_sector
= bio
->bi_sector
;
3300 old_dev
= bio
->bi_bdev
->bd_dev
;
3302 if (bio_check_eod(bio
, nr_sectors
))
3304 if (bio_empty_barrier(bio
) && !q
->prepare_flush_fn
) {
3309 ret
= q
->make_request_fn(q
, bio
);
3314 * We only want one ->make_request_fn to be active at a time,
3315 * else stack usage with stacked devices could be a problem.
3316 * So use current->bio_{list,tail} to keep a list of requests
3317 * submited by a make_request_fn function.
3318 * current->bio_tail is also used as a flag to say if
3319 * generic_make_request is currently active in this task or not.
3320 * If it is NULL, then no make_request is active. If it is non-NULL,
3321 * then a make_request is active, and new requests should be added
3324 void generic_make_request(struct bio
*bio
)
3326 if (current
->bio_tail
) {
3327 /* make_request is active */
3328 *(current
->bio_tail
) = bio
;
3329 bio
->bi_next
= NULL
;
3330 current
->bio_tail
= &bio
->bi_next
;
3333 /* following loop may be a bit non-obvious, and so deserves some
3335 * Before entering the loop, bio->bi_next is NULL (as all callers
3336 * ensure that) so we have a list with a single bio.
3337 * We pretend that we have just taken it off a longer list, so
3338 * we assign bio_list to the next (which is NULL) and bio_tail
3339 * to &bio_list, thus initialising the bio_list of new bios to be
3340 * added. __generic_make_request may indeed add some more bios
3341 * through a recursive call to generic_make_request. If it
3342 * did, we find a non-NULL value in bio_list and re-enter the loop
3343 * from the top. In this case we really did just take the bio
3344 * of the top of the list (no pretending) and so fixup bio_list and
3345 * bio_tail or bi_next, and call into __generic_make_request again.
3347 * The loop was structured like this to make only one call to
3348 * __generic_make_request (which is important as it is large and
3349 * inlined) and to keep the structure simple.
3351 BUG_ON(bio
->bi_next
);
3353 current
->bio_list
= bio
->bi_next
;
3354 if (bio
->bi_next
== NULL
)
3355 current
->bio_tail
= ¤t
->bio_list
;
3357 bio
->bi_next
= NULL
;
3358 __generic_make_request(bio
);
3359 bio
= current
->bio_list
;
3361 current
->bio_tail
= NULL
; /* deactivate */
3364 EXPORT_SYMBOL(generic_make_request
);
3367 * submit_bio: submit a bio to the block device layer for I/O
3368 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3369 * @bio: The &struct bio which describes the I/O
3371 * submit_bio() is very similar in purpose to generic_make_request(), and
3372 * uses that function to do most of the work. Both are fairly rough
3373 * interfaces, @bio must be presetup and ready for I/O.
3376 void submit_bio(int rw
, struct bio
*bio
)
3378 int count
= bio_sectors(bio
);
3383 * If it's a regular read/write or a barrier with data attached,
3384 * go through the normal accounting stuff before submission.
3386 if (!bio_empty_barrier(bio
)) {
3388 BIO_BUG_ON(!bio
->bi_size
);
3389 BIO_BUG_ON(!bio
->bi_io_vec
);
3392 count_vm_events(PGPGOUT
, count
);
3394 task_io_account_read(bio
->bi_size
);
3395 count_vm_events(PGPGIN
, count
);
3398 if (unlikely(block_dump
)) {
3399 char b
[BDEVNAME_SIZE
];
3400 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
3401 current
->comm
, task_pid_nr(current
),
3402 (rw
& WRITE
) ? "WRITE" : "READ",
3403 (unsigned long long)bio
->bi_sector
,
3404 bdevname(bio
->bi_bdev
,b
));
3408 generic_make_request(bio
);
3411 EXPORT_SYMBOL(submit_bio
);
3413 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3415 if (blk_fs_request(rq
)) {
3416 rq
->hard_sector
+= nsect
;
3417 rq
->hard_nr_sectors
-= nsect
;
3420 * Move the I/O submission pointers ahead if required.
3422 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3423 (rq
->sector
<= rq
->hard_sector
)) {
3424 rq
->sector
= rq
->hard_sector
;
3425 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3426 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3427 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3428 rq
->buffer
= bio_data(rq
->bio
);
3432 * if total number of sectors is less than the first segment
3433 * size, something has gone terribly wrong
3435 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3436 printk("blk: request botched\n");
3437 rq
->nr_sectors
= rq
->current_nr_sectors
;
3442 static int __end_that_request_first(struct request
*req
, int uptodate
,
3445 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3448 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
3451 * extend uptodate bool to allow < 0 value to be direct io error
3454 if (end_io_error(uptodate
))
3455 error
= !uptodate
? -EIO
: uptodate
;
3458 * for a REQ_BLOCK_PC request, we want to carry any eventual
3459 * sense key with us all the way through
3461 if (!blk_pc_request(req
))
3465 if (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))
3466 printk("end_request: I/O error, dev %s, sector %llu\n",
3467 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3468 (unsigned long long)req
->sector
);
3471 if (blk_fs_request(req
) && req
->rq_disk
) {
3472 const int rw
= rq_data_dir(req
);
3474 disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3477 total_bytes
= bio_nbytes
= 0;
3478 while ((bio
= req
->bio
) != NULL
) {
3482 * For an empty barrier request, the low level driver must
3483 * store a potential error location in ->sector. We pass
3484 * that back up in ->bi_sector.
3486 if (blk_empty_barrier(req
))
3487 bio
->bi_sector
= req
->sector
;
3489 if (nr_bytes
>= bio
->bi_size
) {
3490 req
->bio
= bio
->bi_next
;
3491 nbytes
= bio
->bi_size
;
3492 req_bio_endio(req
, bio
, nbytes
, error
);
3496 int idx
= bio
->bi_idx
+ next_idx
;
3498 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3499 blk_dump_rq_flags(req
, "__end_that");
3500 printk("%s: bio idx %d >= vcnt %d\n",
3502 bio
->bi_idx
, bio
->bi_vcnt
);
3506 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3507 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3510 * not a complete bvec done
3512 if (unlikely(nbytes
> nr_bytes
)) {
3513 bio_nbytes
+= nr_bytes
;
3514 total_bytes
+= nr_bytes
;
3519 * advance to the next vector
3522 bio_nbytes
+= nbytes
;
3525 total_bytes
+= nbytes
;
3528 if ((bio
= req
->bio
)) {
3530 * end more in this run, or just return 'not-done'
3532 if (unlikely(nr_bytes
<= 0))
3544 * if the request wasn't completed, update state
3547 req_bio_endio(req
, bio
, bio_nbytes
, error
);
3548 bio
->bi_idx
+= next_idx
;
3549 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3550 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3553 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3554 blk_recalc_rq_segments(req
);
3559 * end_that_request_first - end I/O on a request
3560 * @req: the request being processed
3561 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3562 * @nr_sectors: number of sectors to end I/O on
3565 * Ends I/O on a number of sectors attached to @req, and sets it up
3566 * for the next range of segments (if any) in the cluster.
3569 * 0 - we are done with this request, call end_that_request_last()
3570 * 1 - still buffers pending for this request
3572 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3574 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3577 EXPORT_SYMBOL(end_that_request_first
);
3580 * end_that_request_chunk - end I/O on a request
3581 * @req: the request being processed
3582 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3583 * @nr_bytes: number of bytes to complete
3586 * Ends I/O on a number of bytes attached to @req, and sets it up
3587 * for the next range of segments (if any). Like end_that_request_first(),
3588 * but deals with bytes instead of sectors.
3591 * 0 - we are done with this request, call end_that_request_last()
3592 * 1 - still buffers pending for this request
3594 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3596 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3599 EXPORT_SYMBOL(end_that_request_chunk
);
3602 * splice the completion data to a local structure and hand off to
3603 * process_completion_queue() to complete the requests
3605 static void blk_done_softirq(struct softirq_action
*h
)
3607 struct list_head
*cpu_list
, local_list
;
3609 local_irq_disable();
3610 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3611 list_replace_init(cpu_list
, &local_list
);
3614 while (!list_empty(&local_list
)) {
3615 struct request
*rq
= list_entry(local_list
.next
, struct request
, donelist
);
3617 list_del_init(&rq
->donelist
);
3618 rq
->q
->softirq_done_fn(rq
);
3622 static int __cpuinit
blk_cpu_notify(struct notifier_block
*self
, unsigned long action
,
3626 * If a CPU goes away, splice its entries to the current CPU
3627 * and trigger a run of the softirq
3629 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3630 int cpu
= (unsigned long) hcpu
;
3632 local_irq_disable();
3633 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
3634 &__get_cpu_var(blk_cpu_done
));
3635 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3643 static struct notifier_block blk_cpu_notifier __cpuinitdata
= {
3644 .notifier_call
= blk_cpu_notify
,
3648 * blk_complete_request - end I/O on a request
3649 * @req: the request being processed
3652 * Ends all I/O on a request. It does not handle partial completions,
3653 * unless the driver actually implements this in its completion callback
3654 * through requeueing. The actual completion happens out-of-order,
3655 * through a softirq handler. The user must have registered a completion
3656 * callback through blk_queue_softirq_done().
3659 void blk_complete_request(struct request
*req
)
3661 struct list_head
*cpu_list
;
3662 unsigned long flags
;
3664 BUG_ON(!req
->q
->softirq_done_fn
);
3666 local_irq_save(flags
);
3668 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3669 list_add_tail(&req
->donelist
, cpu_list
);
3670 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3672 local_irq_restore(flags
);
3675 EXPORT_SYMBOL(blk_complete_request
);
3678 * queue lock must be held
3680 void end_that_request_last(struct request
*req
, int uptodate
)
3682 struct gendisk
*disk
= req
->rq_disk
;
3686 * extend uptodate bool to allow < 0 value to be direct io error
3689 if (end_io_error(uptodate
))
3690 error
= !uptodate
? -EIO
: uptodate
;
3692 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3693 laptop_io_completion();
3696 * Account IO completion. bar_rq isn't accounted as a normal
3697 * IO on queueing nor completion. Accounting the containing
3698 * request is enough.
3700 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
3701 unsigned long duration
= jiffies
- req
->start_time
;
3702 const int rw
= rq_data_dir(req
);
3704 __disk_stat_inc(disk
, ios
[rw
]);
3705 __disk_stat_add(disk
, ticks
[rw
], duration
);
3706 disk_round_stats(disk
);
3710 req
->end_io(req
, error
);
3712 __blk_put_request(req
->q
, req
);
3715 EXPORT_SYMBOL(end_that_request_last
);
3717 static inline void __end_request(struct request
*rq
, int uptodate
,
3718 unsigned int nr_bytes
, int dequeue
)
3720 if (!end_that_request_chunk(rq
, uptodate
, nr_bytes
)) {
3722 blkdev_dequeue_request(rq
);
3723 add_disk_randomness(rq
->rq_disk
);
3724 end_that_request_last(rq
, uptodate
);
3728 static unsigned int rq_byte_size(struct request
*rq
)
3730 if (blk_fs_request(rq
))
3731 return rq
->hard_nr_sectors
<< 9;
3733 return rq
->data_len
;
3737 * end_queued_request - end all I/O on a queued request
3738 * @rq: the request being processed
3739 * @uptodate: error value or 0/1 uptodate flag
3742 * Ends all I/O on a request, and removes it from the block layer queues.
3743 * Not suitable for normal IO completion, unless the driver still has
3744 * the request attached to the block layer.
3747 void end_queued_request(struct request
*rq
, int uptodate
)
3749 __end_request(rq
, uptodate
, rq_byte_size(rq
), 1);
3751 EXPORT_SYMBOL(end_queued_request
);
3754 * end_dequeued_request - end all I/O on a dequeued request
3755 * @rq: the request being processed
3756 * @uptodate: error value or 0/1 uptodate flag
3759 * Ends all I/O on a request. The request must already have been
3760 * dequeued using blkdev_dequeue_request(), as is normally the case
3764 void end_dequeued_request(struct request
*rq
, int uptodate
)
3766 __end_request(rq
, uptodate
, rq_byte_size(rq
), 0);
3768 EXPORT_SYMBOL(end_dequeued_request
);
3772 * end_request - end I/O on the current segment of the request
3773 * @req: the request being processed
3774 * @uptodate: error value or 0/1 uptodate flag
3777 * Ends I/O on the current segment of a request. If that is the only
3778 * remaining segment, the request is also completed and freed.
3780 * This is a remnant of how older block drivers handled IO completions.
3781 * Modern drivers typically end IO on the full request in one go, unless
3782 * they have a residual value to account for. For that case this function
3783 * isn't really useful, unless the residual just happens to be the
3784 * full current segment. In other words, don't use this function in new
3785 * code. Either use end_request_completely(), or the
3786 * end_that_request_chunk() (along with end_that_request_last()) for
3787 * partial completions.
3790 void end_request(struct request
*req
, int uptodate
)
3792 __end_request(req
, uptodate
, req
->hard_cur_sectors
<< 9, 1);
3794 EXPORT_SYMBOL(end_request
);
3796 static void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3799 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3800 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
3802 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3803 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3804 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3805 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3806 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3807 rq
->buffer
= bio_data(bio
);
3808 rq
->data_len
= bio
->bi_size
;
3810 rq
->bio
= rq
->biotail
= bio
;
3813 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
3816 int kblockd_schedule_work(struct work_struct
*work
)
3818 return queue_work(kblockd_workqueue
, work
);
3821 EXPORT_SYMBOL(kblockd_schedule_work
);
3823 void kblockd_flush_work(struct work_struct
*work
)
3825 cancel_work_sync(work
);
3827 EXPORT_SYMBOL(kblockd_flush_work
);
3829 int __init
blk_dev_init(void)
3833 kblockd_workqueue
= create_workqueue("kblockd");
3834 if (!kblockd_workqueue
)
3835 panic("Failed to create kblockd\n");
3837 request_cachep
= kmem_cache_create("blkdev_requests",
3838 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3840 requestq_cachep
= kmem_cache_create("blkdev_queue",
3841 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3843 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3844 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
);
3846 for_each_possible_cpu(i
)
3847 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
3849 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
3850 register_hotcpu_notifier(&blk_cpu_notifier
);
3852 blk_max_low_pfn
= max_low_pfn
- 1;
3853 blk_max_pfn
= max_pfn
- 1;
3859 * IO Context helper functions
3861 void put_io_context(struct io_context
*ioc
)
3866 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3868 if (atomic_dec_and_test(&ioc
->refcount
)) {
3869 struct cfq_io_context
*cic
;
3872 if (ioc
->aic
&& ioc
->aic
->dtor
)
3873 ioc
->aic
->dtor(ioc
->aic
);
3874 if (ioc
->cic_root
.rb_node
!= NULL
) {
3875 struct rb_node
*n
= rb_first(&ioc
->cic_root
);
3877 cic
= rb_entry(n
, struct cfq_io_context
, rb_node
);
3882 kmem_cache_free(iocontext_cachep
, ioc
);
3885 EXPORT_SYMBOL(put_io_context
);
3887 /* Called by the exitting task */
3888 void exit_io_context(void)
3890 struct io_context
*ioc
;
3891 struct cfq_io_context
*cic
;
3894 ioc
= current
->io_context
;
3895 current
->io_context
= NULL
;
3896 task_unlock(current
);
3899 if (ioc
->aic
&& ioc
->aic
->exit
)
3900 ioc
->aic
->exit(ioc
->aic
);
3901 if (ioc
->cic_root
.rb_node
!= NULL
) {
3902 cic
= rb_entry(rb_first(&ioc
->cic_root
), struct cfq_io_context
, rb_node
);
3906 put_io_context(ioc
);
3910 * If the current task has no IO context then create one and initialise it.
3911 * Otherwise, return its existing IO context.
3913 * This returned IO context doesn't have a specifically elevated refcount,
3914 * but since the current task itself holds a reference, the context can be
3915 * used in general code, so long as it stays within `current` context.
3917 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
)
3919 struct task_struct
*tsk
= current
;
3920 struct io_context
*ret
;
3922 ret
= tsk
->io_context
;
3926 ret
= kmem_cache_alloc_node(iocontext_cachep
, gfp_flags
, node
);
3928 atomic_set(&ret
->refcount
, 1);
3929 ret
->task
= current
;
3930 ret
->ioprio_changed
= 0;
3931 ret
->last_waited
= jiffies
; /* doesn't matter... */
3932 ret
->nr_batch_requests
= 0; /* because this is 0 */
3934 ret
->cic_root
.rb_node
= NULL
;
3935 ret
->ioc_data
= NULL
;
3936 /* make sure set_task_ioprio() sees the settings above */
3938 tsk
->io_context
= ret
;
3945 * If the current task has no IO context then create one and initialise it.
3946 * If it does have a context, take a ref on it.
3948 * This is always called in the context of the task which submitted the I/O.
3950 struct io_context
*get_io_context(gfp_t gfp_flags
, int node
)
3952 struct io_context
*ret
;
3953 ret
= current_io_context(gfp_flags
, node
);
3955 atomic_inc(&ret
->refcount
);
3958 EXPORT_SYMBOL(get_io_context
);
3960 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3962 struct io_context
*src
= *psrc
;
3963 struct io_context
*dst
= *pdst
;
3966 BUG_ON(atomic_read(&src
->refcount
) == 0);
3967 atomic_inc(&src
->refcount
);
3968 put_io_context(dst
);
3972 EXPORT_SYMBOL(copy_io_context
);
3974 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3976 struct io_context
*temp
;
3981 EXPORT_SYMBOL(swap_io_context
);
3986 struct queue_sysfs_entry
{
3987 struct attribute attr
;
3988 ssize_t (*show
)(struct request_queue
*, char *);
3989 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3993 queue_var_show(unsigned int var
, char *page
)
3995 return sprintf(page
, "%d\n", var
);
3999 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
4001 char *p
= (char *) page
;
4003 *var
= simple_strtoul(p
, &p
, 10);
4007 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
4009 return queue_var_show(q
->nr_requests
, (page
));
4013 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
4015 struct request_list
*rl
= &q
->rq
;
4017 int ret
= queue_var_store(&nr
, page
, count
);
4018 if (nr
< BLKDEV_MIN_RQ
)
4021 spin_lock_irq(q
->queue_lock
);
4022 q
->nr_requests
= nr
;
4023 blk_queue_congestion_threshold(q
);
4025 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
4026 blk_set_queue_congested(q
, READ
);
4027 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
4028 blk_clear_queue_congested(q
, READ
);
4030 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
4031 blk_set_queue_congested(q
, WRITE
);
4032 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
4033 blk_clear_queue_congested(q
, WRITE
);
4035 if (rl
->count
[READ
] >= q
->nr_requests
) {
4036 blk_set_queue_full(q
, READ
);
4037 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
4038 blk_clear_queue_full(q
, READ
);
4039 wake_up(&rl
->wait
[READ
]);
4042 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
4043 blk_set_queue_full(q
, WRITE
);
4044 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
4045 blk_clear_queue_full(q
, WRITE
);
4046 wake_up(&rl
->wait
[WRITE
]);
4048 spin_unlock_irq(q
->queue_lock
);
4052 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
4054 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
4056 return queue_var_show(ra_kb
, (page
));
4060 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
4062 unsigned long ra_kb
;
4063 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
4065 spin_lock_irq(q
->queue_lock
);
4066 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
4067 spin_unlock_irq(q
->queue_lock
);
4072 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
4074 int max_sectors_kb
= q
->max_sectors
>> 1;
4076 return queue_var_show(max_sectors_kb
, (page
));
4080 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
4082 unsigned long max_sectors_kb
,
4083 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
4084 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
4085 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
4087 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
4090 * Take the queue lock to update the readahead and max_sectors
4091 * values synchronously:
4093 spin_lock_irq(q
->queue_lock
);
4094 q
->max_sectors
= max_sectors_kb
<< 1;
4095 spin_unlock_irq(q
->queue_lock
);
4100 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
4102 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
4104 return queue_var_show(max_hw_sectors_kb
, (page
));
4108 static struct queue_sysfs_entry queue_requests_entry
= {
4109 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
4110 .show
= queue_requests_show
,
4111 .store
= queue_requests_store
,
4114 static struct queue_sysfs_entry queue_ra_entry
= {
4115 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
4116 .show
= queue_ra_show
,
4117 .store
= queue_ra_store
,
4120 static struct queue_sysfs_entry queue_max_sectors_entry
= {
4121 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
4122 .show
= queue_max_sectors_show
,
4123 .store
= queue_max_sectors_store
,
4126 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
4127 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
4128 .show
= queue_max_hw_sectors_show
,
4131 static struct queue_sysfs_entry queue_iosched_entry
= {
4132 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
4133 .show
= elv_iosched_show
,
4134 .store
= elv_iosched_store
,
4137 static struct attribute
*default_attrs
[] = {
4138 &queue_requests_entry
.attr
,
4139 &queue_ra_entry
.attr
,
4140 &queue_max_hw_sectors_entry
.attr
,
4141 &queue_max_sectors_entry
.attr
,
4142 &queue_iosched_entry
.attr
,
4146 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4149 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4151 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4152 struct request_queue
*q
=
4153 container_of(kobj
, struct request_queue
, kobj
);
4158 mutex_lock(&q
->sysfs_lock
);
4159 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4160 mutex_unlock(&q
->sysfs_lock
);
4163 res
= entry
->show(q
, page
);
4164 mutex_unlock(&q
->sysfs_lock
);
4169 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4170 const char *page
, size_t length
)
4172 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4173 struct request_queue
*q
= container_of(kobj
, struct request_queue
, kobj
);
4179 mutex_lock(&q
->sysfs_lock
);
4180 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4181 mutex_unlock(&q
->sysfs_lock
);
4184 res
= entry
->store(q
, page
, length
);
4185 mutex_unlock(&q
->sysfs_lock
);
4189 static struct sysfs_ops queue_sysfs_ops
= {
4190 .show
= queue_attr_show
,
4191 .store
= queue_attr_store
,
4194 static struct kobj_type queue_ktype
= {
4195 .sysfs_ops
= &queue_sysfs_ops
,
4196 .default_attrs
= default_attrs
,
4197 .release
= blk_release_queue
,
4200 int blk_register_queue(struct gendisk
*disk
)
4204 struct request_queue
*q
= disk
->queue
;
4206 if (!q
|| !q
->request_fn
)
4209 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
4211 ret
= kobject_add(&q
->kobj
);
4215 kobject_uevent(&q
->kobj
, KOBJ_ADD
);
4217 ret
= elv_register_queue(q
);
4219 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4220 kobject_del(&q
->kobj
);
4227 void blk_unregister_queue(struct gendisk
*disk
)
4229 struct request_queue
*q
= disk
->queue
;
4231 if (q
&& q
->request_fn
) {
4232 elv_unregister_queue(q
);
4234 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4235 kobject_del(&q
->kobj
);
4236 kobject_put(&disk
->kobj
);