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
)
351 if (error
&& !q
->orderr
)
354 BUG_ON(q
->ordseq
& seq
);
357 if (blk_ordered_cur_seq(q
) != QUEUE_ORDSEQ_DONE
)
361 * Okay, sequence complete.
366 if (__blk_end_request(rq
, q
->orderr
, blk_rq_bytes(rq
)))
370 static void pre_flush_end_io(struct request
*rq
, int error
)
372 elv_completed_request(rq
->q
, rq
);
373 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_PREFLUSH
, error
);
376 static void bar_end_io(struct request
*rq
, int error
)
378 elv_completed_request(rq
->q
, rq
);
379 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_BAR
, error
);
382 static void post_flush_end_io(struct request
*rq
, int error
)
384 elv_completed_request(rq
->q
, rq
);
385 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_POSTFLUSH
, error
);
388 static void queue_flush(struct request_queue
*q
, unsigned which
)
391 rq_end_io_fn
*end_io
;
393 if (which
== QUEUE_ORDERED_PREFLUSH
) {
394 rq
= &q
->pre_flush_rq
;
395 end_io
= pre_flush_end_io
;
397 rq
= &q
->post_flush_rq
;
398 end_io
= post_flush_end_io
;
401 rq
->cmd_flags
= REQ_HARDBARRIER
;
403 rq
->elevator_private
= NULL
;
404 rq
->elevator_private2
= NULL
;
405 rq
->rq_disk
= q
->bar_rq
.rq_disk
;
407 q
->prepare_flush_fn(q
, rq
);
409 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
412 static inline struct request
*start_ordered(struct request_queue
*q
,
416 q
->ordered
= q
->next_ordered
;
417 q
->ordseq
|= QUEUE_ORDSEQ_STARTED
;
420 * Prep proxy barrier request.
422 blkdev_dequeue_request(rq
);
427 if (bio_data_dir(q
->orig_bar_rq
->bio
) == WRITE
)
428 rq
->cmd_flags
|= REQ_RW
;
429 if (q
->ordered
& QUEUE_ORDERED_FUA
)
430 rq
->cmd_flags
|= REQ_FUA
;
431 rq
->elevator_private
= NULL
;
432 rq
->elevator_private2
= NULL
;
433 init_request_from_bio(rq
, q
->orig_bar_rq
->bio
);
434 rq
->end_io
= bar_end_io
;
437 * Queue ordered sequence. As we stack them at the head, we
438 * need to queue in reverse order. Note that we rely on that
439 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
440 * request gets inbetween ordered sequence. If this request is
441 * an empty barrier, we don't need to do a postflush ever since
442 * there will be no data written between the pre and post flush.
443 * Hence a single flush will suffice.
445 if ((q
->ordered
& QUEUE_ORDERED_POSTFLUSH
) && !blk_empty_barrier(rq
))
446 queue_flush(q
, QUEUE_ORDERED_POSTFLUSH
);
448 q
->ordseq
|= QUEUE_ORDSEQ_POSTFLUSH
;
450 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
452 if (q
->ordered
& QUEUE_ORDERED_PREFLUSH
) {
453 queue_flush(q
, QUEUE_ORDERED_PREFLUSH
);
454 rq
= &q
->pre_flush_rq
;
456 q
->ordseq
|= QUEUE_ORDSEQ_PREFLUSH
;
458 if ((q
->ordered
& QUEUE_ORDERED_TAG
) || q
->in_flight
== 0)
459 q
->ordseq
|= QUEUE_ORDSEQ_DRAIN
;
466 int blk_do_ordered(struct request_queue
*q
, struct request
**rqp
)
468 struct request
*rq
= *rqp
;
469 const int is_barrier
= blk_fs_request(rq
) && blk_barrier_rq(rq
);
475 if (q
->next_ordered
!= QUEUE_ORDERED_NONE
) {
476 *rqp
= start_ordered(q
, rq
);
480 * This can happen when the queue switches to
481 * ORDERED_NONE while this request is on it.
483 blkdev_dequeue_request(rq
);
484 if (__blk_end_request(rq
, -EOPNOTSUPP
,
493 * Ordered sequence in progress
496 /* Special requests are not subject to ordering rules. */
497 if (!blk_fs_request(rq
) &&
498 rq
!= &q
->pre_flush_rq
&& rq
!= &q
->post_flush_rq
)
501 if (q
->ordered
& QUEUE_ORDERED_TAG
) {
502 /* Ordered by tag. Blocking the next barrier is enough. */
503 if (is_barrier
&& rq
!= &q
->bar_rq
)
506 /* Ordered by draining. Wait for turn. */
507 WARN_ON(blk_ordered_req_seq(rq
) < blk_ordered_cur_seq(q
));
508 if (blk_ordered_req_seq(rq
) > blk_ordered_cur_seq(q
))
515 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
516 unsigned int nbytes
, int error
)
518 struct request_queue
*q
= rq
->q
;
520 if (&q
->bar_rq
!= rq
) {
522 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
523 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
526 if (unlikely(nbytes
> bio
->bi_size
)) {
527 printk("%s: want %u bytes done, only %u left\n",
528 __FUNCTION__
, nbytes
, bio
->bi_size
);
529 nbytes
= bio
->bi_size
;
532 bio
->bi_size
-= nbytes
;
533 bio
->bi_sector
+= (nbytes
>> 9);
534 if (bio
->bi_size
== 0)
535 bio_endio(bio
, error
);
539 * Okay, this is the barrier request in progress, just
542 if (error
&& !q
->orderr
)
548 * blk_queue_bounce_limit - set bounce buffer limit for queue
549 * @q: the request queue for the device
550 * @dma_addr: bus address limit
553 * Different hardware can have different requirements as to what pages
554 * it can do I/O directly to. A low level driver can call
555 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
556 * buffers for doing I/O to pages residing above @page.
558 void blk_queue_bounce_limit(struct request_queue
*q
, u64 dma_addr
)
560 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
563 q
->bounce_gfp
= GFP_NOIO
;
564 #if BITS_PER_LONG == 64
565 /* Assume anything <= 4GB can be handled by IOMMU.
566 Actually some IOMMUs can handle everything, but I don't
567 know of a way to test this here. */
568 if (bounce_pfn
< (min_t(u64
,0xffffffff,BLK_BOUNCE_HIGH
) >> PAGE_SHIFT
))
570 q
->bounce_pfn
= max_low_pfn
;
572 if (bounce_pfn
< blk_max_low_pfn
)
574 q
->bounce_pfn
= bounce_pfn
;
577 init_emergency_isa_pool();
578 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
579 q
->bounce_pfn
= bounce_pfn
;
583 EXPORT_SYMBOL(blk_queue_bounce_limit
);
586 * blk_queue_max_sectors - set max sectors for a request for this queue
587 * @q: the request queue for the device
588 * @max_sectors: max sectors in the usual 512b unit
591 * Enables a low level driver to set an upper limit on the size of
594 void blk_queue_max_sectors(struct request_queue
*q
, unsigned int max_sectors
)
596 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
597 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
598 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
601 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
602 q
->max_hw_sectors
= q
->max_sectors
= max_sectors
;
604 q
->max_sectors
= BLK_DEF_MAX_SECTORS
;
605 q
->max_hw_sectors
= max_sectors
;
609 EXPORT_SYMBOL(blk_queue_max_sectors
);
612 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
613 * @q: the request queue for the device
614 * @max_segments: max number of segments
617 * Enables a low level driver to set an upper limit on the number of
618 * physical data segments in a request. This would be the largest sized
619 * scatter list the driver could handle.
621 void blk_queue_max_phys_segments(struct request_queue
*q
,
622 unsigned short max_segments
)
626 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
629 q
->max_phys_segments
= max_segments
;
632 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
635 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
636 * @q: the request queue for the device
637 * @max_segments: max number of segments
640 * Enables a low level driver to set an upper limit on the number of
641 * hw data segments in a request. This would be the largest number of
642 * address/length pairs the host adapter can actually give as once
645 void blk_queue_max_hw_segments(struct request_queue
*q
,
646 unsigned short max_segments
)
650 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
653 q
->max_hw_segments
= max_segments
;
656 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
659 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
660 * @q: the request queue for the device
661 * @max_size: max size of segment in bytes
664 * Enables a low level driver to set an upper limit on the size of a
667 void blk_queue_max_segment_size(struct request_queue
*q
, unsigned int max_size
)
669 if (max_size
< PAGE_CACHE_SIZE
) {
670 max_size
= PAGE_CACHE_SIZE
;
671 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
674 q
->max_segment_size
= max_size
;
677 EXPORT_SYMBOL(blk_queue_max_segment_size
);
680 * blk_queue_hardsect_size - set hardware sector size for the queue
681 * @q: the request queue for the device
682 * @size: the hardware sector size, in bytes
685 * This should typically be set to the lowest possible sector size
686 * that the hardware can operate on (possible without reverting to
687 * even internal read-modify-write operations). Usually the default
688 * of 512 covers most hardware.
690 void blk_queue_hardsect_size(struct request_queue
*q
, unsigned short size
)
692 q
->hardsect_size
= size
;
695 EXPORT_SYMBOL(blk_queue_hardsect_size
);
698 * Returns the minimum that is _not_ zero, unless both are zero.
700 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
703 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
704 * @t: the stacking driver (top)
705 * @b: the underlying device (bottom)
707 void blk_queue_stack_limits(struct request_queue
*t
, struct request_queue
*b
)
709 /* zero is "infinity" */
710 t
->max_sectors
= min_not_zero(t
->max_sectors
,b
->max_sectors
);
711 t
->max_hw_sectors
= min_not_zero(t
->max_hw_sectors
,b
->max_hw_sectors
);
713 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
714 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
715 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
716 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
717 if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
))
718 clear_bit(QUEUE_FLAG_CLUSTER
, &t
->queue_flags
);
721 EXPORT_SYMBOL(blk_queue_stack_limits
);
724 * blk_queue_segment_boundary - set boundary rules for segment merging
725 * @q: the request queue for the device
726 * @mask: the memory boundary mask
728 void blk_queue_segment_boundary(struct request_queue
*q
, unsigned long mask
)
730 if (mask
< PAGE_CACHE_SIZE
- 1) {
731 mask
= PAGE_CACHE_SIZE
- 1;
732 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
735 q
->seg_boundary_mask
= mask
;
738 EXPORT_SYMBOL(blk_queue_segment_boundary
);
741 * blk_queue_dma_alignment - set dma length and memory alignment
742 * @q: the request queue for the device
743 * @mask: alignment mask
746 * set required memory and length aligment for direct dma transactions.
747 * this is used when buiding direct io requests for the queue.
750 void blk_queue_dma_alignment(struct request_queue
*q
, int mask
)
752 q
->dma_alignment
= mask
;
755 EXPORT_SYMBOL(blk_queue_dma_alignment
);
758 * blk_queue_update_dma_alignment - update dma length and memory alignment
759 * @q: the request queue for the device
760 * @mask: alignment mask
763 * update required memory and length aligment for direct dma transactions.
764 * If the requested alignment is larger than the current alignment, then
765 * the current queue alignment is updated to the new value, otherwise it
766 * is left alone. The design of this is to allow multiple objects
767 * (driver, device, transport etc) to set their respective
768 * alignments without having them interfere.
771 void blk_queue_update_dma_alignment(struct request_queue
*q
, int mask
)
773 BUG_ON(mask
> PAGE_SIZE
);
775 if (mask
> q
->dma_alignment
)
776 q
->dma_alignment
= mask
;
779 EXPORT_SYMBOL(blk_queue_update_dma_alignment
);
782 * blk_queue_find_tag - find a request by its tag and queue
783 * @q: The request queue for the device
784 * @tag: The tag of the request
787 * Should be used when a device returns a tag and you want to match
790 * no locks need be held.
792 struct request
*blk_queue_find_tag(struct request_queue
*q
, int tag
)
794 return blk_map_queue_find_tag(q
->queue_tags
, tag
);
797 EXPORT_SYMBOL(blk_queue_find_tag
);
800 * __blk_free_tags - release a given set of tag maintenance info
801 * @bqt: the tag map to free
803 * Tries to free the specified @bqt@. Returns true if it was
804 * actually freed and false if there are still references using it
806 static int __blk_free_tags(struct blk_queue_tag
*bqt
)
810 retval
= atomic_dec_and_test(&bqt
->refcnt
);
814 kfree(bqt
->tag_index
);
815 bqt
->tag_index
= NULL
;
828 * __blk_queue_free_tags - release tag maintenance info
829 * @q: the request queue for the device
832 * blk_cleanup_queue() will take care of calling this function, if tagging
833 * has been used. So there's no need to call this directly.
835 static void __blk_queue_free_tags(struct request_queue
*q
)
837 struct blk_queue_tag
*bqt
= q
->queue_tags
;
842 __blk_free_tags(bqt
);
844 q
->queue_tags
= NULL
;
845 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
850 * blk_free_tags - release a given set of tag maintenance info
851 * @bqt: the tag map to free
853 * For externally managed @bqt@ frees the map. Callers of this
854 * function must guarantee to have released all the queues that
855 * might have been using this tag map.
857 void blk_free_tags(struct blk_queue_tag
*bqt
)
859 if (unlikely(!__blk_free_tags(bqt
)))
862 EXPORT_SYMBOL(blk_free_tags
);
865 * blk_queue_free_tags - release tag maintenance info
866 * @q: the request queue for the device
869 * This is used to disabled tagged queuing to a device, yet leave
872 void blk_queue_free_tags(struct request_queue
*q
)
874 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
877 EXPORT_SYMBOL(blk_queue_free_tags
);
880 init_tag_map(struct request_queue
*q
, struct blk_queue_tag
*tags
, int depth
)
882 struct request
**tag_index
;
883 unsigned long *tag_map
;
886 if (q
&& depth
> q
->nr_requests
* 2) {
887 depth
= q
->nr_requests
* 2;
888 printk(KERN_ERR
"%s: adjusted depth to %d\n",
889 __FUNCTION__
, depth
);
892 tag_index
= kzalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
896 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
897 tag_map
= kzalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
901 tags
->real_max_depth
= depth
;
902 tags
->max_depth
= depth
;
903 tags
->tag_index
= tag_index
;
904 tags
->tag_map
= tag_map
;
912 static struct blk_queue_tag
*__blk_queue_init_tags(struct request_queue
*q
,
915 struct blk_queue_tag
*tags
;
917 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
921 if (init_tag_map(q
, tags
, depth
))
925 atomic_set(&tags
->refcnt
, 1);
933 * blk_init_tags - initialize the tag info for an external tag map
934 * @depth: the maximum queue depth supported
935 * @tags: the tag to use
937 struct blk_queue_tag
*blk_init_tags(int depth
)
939 return __blk_queue_init_tags(NULL
, depth
);
941 EXPORT_SYMBOL(blk_init_tags
);
944 * blk_queue_init_tags - initialize the queue tag info
945 * @q: the request queue for the device
946 * @depth: the maximum queue depth supported
947 * @tags: the tag to use
949 int blk_queue_init_tags(struct request_queue
*q
, int depth
,
950 struct blk_queue_tag
*tags
)
954 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
956 if (!tags
&& !q
->queue_tags
) {
957 tags
= __blk_queue_init_tags(q
, depth
);
961 } else if (q
->queue_tags
) {
962 if ((rc
= blk_queue_resize_tags(q
, depth
)))
964 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
967 atomic_inc(&tags
->refcnt
);
970 * assign it, all done
972 q
->queue_tags
= tags
;
973 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
974 INIT_LIST_HEAD(&q
->tag_busy_list
);
981 EXPORT_SYMBOL(blk_queue_init_tags
);
984 * blk_queue_resize_tags - change the queueing depth
985 * @q: the request queue for the device
986 * @new_depth: the new max command queueing depth
989 * Must be called with the queue lock held.
991 int blk_queue_resize_tags(struct request_queue
*q
, int new_depth
)
993 struct blk_queue_tag
*bqt
= q
->queue_tags
;
994 struct request
**tag_index
;
995 unsigned long *tag_map
;
996 int max_depth
, nr_ulongs
;
1002 * if we already have large enough real_max_depth. just
1003 * adjust max_depth. *NOTE* as requests with tag value
1004 * between new_depth and real_max_depth can be in-flight, tag
1005 * map can not be shrunk blindly here.
1007 if (new_depth
<= bqt
->real_max_depth
) {
1008 bqt
->max_depth
= new_depth
;
1013 * Currently cannot replace a shared tag map with a new
1014 * one, so error out if this is the case
1016 if (atomic_read(&bqt
->refcnt
) != 1)
1020 * save the old state info, so we can copy it back
1022 tag_index
= bqt
->tag_index
;
1023 tag_map
= bqt
->tag_map
;
1024 max_depth
= bqt
->real_max_depth
;
1026 if (init_tag_map(q
, bqt
, new_depth
))
1029 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
1030 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
1031 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
1038 EXPORT_SYMBOL(blk_queue_resize_tags
);
1041 * blk_queue_end_tag - end tag operations for a request
1042 * @q: the request queue for the device
1043 * @rq: the request that has completed
1046 * Typically called when end_that_request_first() returns 0, meaning
1047 * all transfers have been done for a request. It's important to call
1048 * this function before end_that_request_last(), as that will put the
1049 * request back on the free list thus corrupting the internal tag list.
1052 * queue lock must be held.
1054 void blk_queue_end_tag(struct request_queue
*q
, struct request
*rq
)
1056 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1061 if (unlikely(tag
>= bqt
->real_max_depth
))
1063 * This can happen after tag depth has been reduced.
1064 * FIXME: how about a warning or info message here?
1068 list_del_init(&rq
->queuelist
);
1069 rq
->cmd_flags
&= ~REQ_QUEUED
;
1072 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
1073 printk(KERN_ERR
"%s: tag %d is missing\n",
1076 bqt
->tag_index
[tag
] = NULL
;
1078 if (unlikely(!test_bit(tag
, bqt
->tag_map
))) {
1079 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
1084 * The tag_map bit acts as a lock for tag_index[bit], so we need
1085 * unlock memory barrier semantics.
1087 clear_bit_unlock(tag
, bqt
->tag_map
);
1091 EXPORT_SYMBOL(blk_queue_end_tag
);
1094 * blk_queue_start_tag - find a free tag and assign it
1095 * @q: the request queue for the device
1096 * @rq: the block request that needs tagging
1099 * This can either be used as a stand-alone helper, or possibly be
1100 * assigned as the queue &prep_rq_fn (in which case &struct request
1101 * automagically gets a tag assigned). Note that this function
1102 * assumes that any type of request can be queued! if this is not
1103 * true for your device, you must check the request type before
1104 * calling this function. The request will also be removed from
1105 * the request queue, so it's the drivers responsibility to readd
1106 * it if it should need to be restarted for some reason.
1109 * queue lock must be held.
1111 int blk_queue_start_tag(struct request_queue
*q
, struct request
*rq
)
1113 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1116 if (unlikely((rq
->cmd_flags
& REQ_QUEUED
))) {
1118 "%s: request %p for device [%s] already tagged %d",
1120 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
1125 * Protect against shared tag maps, as we may not have exclusive
1126 * access to the tag map.
1129 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
1130 if (tag
>= bqt
->max_depth
)
1133 } while (test_and_set_bit_lock(tag
, bqt
->tag_map
));
1135 * We need lock ordering semantics given by test_and_set_bit_lock.
1136 * See blk_queue_end_tag for details.
1139 rq
->cmd_flags
|= REQ_QUEUED
;
1141 bqt
->tag_index
[tag
] = rq
;
1142 blkdev_dequeue_request(rq
);
1143 list_add(&rq
->queuelist
, &q
->tag_busy_list
);
1148 EXPORT_SYMBOL(blk_queue_start_tag
);
1151 * blk_queue_invalidate_tags - invalidate all pending tags
1152 * @q: the request queue for the device
1155 * Hardware conditions may dictate a need to stop all pending requests.
1156 * In this case, we will safely clear the block side of the tag queue and
1157 * readd all requests to the request queue in the right order.
1160 * queue lock must be held.
1162 void blk_queue_invalidate_tags(struct request_queue
*q
)
1164 struct list_head
*tmp
, *n
;
1166 list_for_each_safe(tmp
, n
, &q
->tag_busy_list
)
1167 blk_requeue_request(q
, list_entry_rq(tmp
));
1170 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1172 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1176 printk("%s: dev %s: type=%x, flags=%x\n", msg
,
1177 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
1180 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1182 rq
->current_nr_sectors
);
1183 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1185 if (blk_pc_request(rq
)) {
1187 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1188 printk("%02x ", rq
->cmd
[bit
]);
1193 EXPORT_SYMBOL(blk_dump_rq_flags
);
1195 void blk_recount_segments(struct request_queue
*q
, struct bio
*bio
)
1198 struct bio
*nxt
= bio
->bi_next
;
1200 rq
.bio
= rq
.biotail
= bio
;
1201 bio
->bi_next
= NULL
;
1202 blk_recalc_rq_segments(&rq
);
1204 bio
->bi_phys_segments
= rq
.nr_phys_segments
;
1205 bio
->bi_hw_segments
= rq
.nr_hw_segments
;
1206 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1208 EXPORT_SYMBOL(blk_recount_segments
);
1210 static void blk_recalc_rq_segments(struct request
*rq
)
1214 unsigned int phys_size
;
1215 unsigned int hw_size
;
1216 struct bio_vec
*bv
, *bvprv
= NULL
;
1220 struct req_iterator iter
;
1221 int high
, highprv
= 1;
1222 struct request_queue
*q
= rq
->q
;
1227 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1228 hw_seg_size
= seg_size
= 0;
1229 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
1230 rq_for_each_segment(bv
, rq
, iter
) {
1232 * the trick here is making sure that a high page is never
1233 * considered part of another segment, since that might
1234 * change with the bounce page.
1236 high
= page_to_pfn(bv
->bv_page
) > q
->bounce_pfn
;
1237 if (high
|| highprv
)
1238 goto new_hw_segment
;
1240 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1242 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1244 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1246 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1247 goto new_hw_segment
;
1249 seg_size
+= bv
->bv_len
;
1250 hw_seg_size
+= bv
->bv_len
;
1255 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1256 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1257 hw_seg_size
+= bv
->bv_len
;
1260 if (nr_hw_segs
== 1 &&
1261 hw_seg_size
> rq
->bio
->bi_hw_front_size
)
1262 rq
->bio
->bi_hw_front_size
= hw_seg_size
;
1263 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1269 seg_size
= bv
->bv_len
;
1273 if (nr_hw_segs
== 1 &&
1274 hw_seg_size
> rq
->bio
->bi_hw_front_size
)
1275 rq
->bio
->bi_hw_front_size
= hw_seg_size
;
1276 if (hw_seg_size
> rq
->biotail
->bi_hw_back_size
)
1277 rq
->biotail
->bi_hw_back_size
= hw_seg_size
;
1278 rq
->nr_phys_segments
= nr_phys_segs
;
1279 rq
->nr_hw_segments
= nr_hw_segs
;
1282 static int blk_phys_contig_segment(struct request_queue
*q
, struct bio
*bio
,
1285 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1288 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1290 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1294 * bio and nxt are contigous in memory, check if the queue allows
1295 * these two to be merged into one
1297 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1303 static int blk_hw_contig_segment(struct request_queue
*q
, struct bio
*bio
,
1306 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1307 blk_recount_segments(q
, bio
);
1308 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1309 blk_recount_segments(q
, nxt
);
1310 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1311 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_back_size
+ nxt
->bi_hw_front_size
))
1313 if (bio
->bi_hw_back_size
+ nxt
->bi_hw_front_size
> q
->max_segment_size
)
1320 * map a request to scatterlist, return number of sg entries setup. Caller
1321 * must make sure sg can hold rq->nr_phys_segments entries
1323 int blk_rq_map_sg(struct request_queue
*q
, struct request
*rq
,
1324 struct scatterlist
*sglist
)
1326 struct bio_vec
*bvec
, *bvprv
;
1327 struct req_iterator iter
;
1328 struct scatterlist
*sg
;
1332 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1335 * for each bio in rq
1339 rq_for_each_segment(bvec
, rq
, iter
) {
1340 int nbytes
= bvec
->bv_len
;
1342 if (bvprv
&& cluster
) {
1343 if (sg
->length
+ nbytes
> q
->max_segment_size
)
1346 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1348 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1351 sg
->length
+= nbytes
;
1358 * If the driver previously mapped a shorter
1359 * list, we could see a termination bit
1360 * prematurely unless it fully inits the sg
1361 * table on each mapping. We KNOW that there
1362 * must be more entries here or the driver
1363 * would be buggy, so force clear the
1364 * termination bit to avoid doing a full
1365 * sg_init_table() in drivers for each command.
1367 sg
->page_link
&= ~0x02;
1371 sg_set_page(sg
, bvec
->bv_page
, nbytes
, bvec
->bv_offset
);
1375 } /* segments in rq */
1383 EXPORT_SYMBOL(blk_rq_map_sg
);
1386 * the standard queue merge functions, can be overridden with device
1387 * specific ones if so desired
1390 static inline int ll_new_mergeable(struct request_queue
*q
,
1391 struct request
*req
,
1394 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1396 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1397 req
->cmd_flags
|= REQ_NOMERGE
;
1398 if (req
== q
->last_merge
)
1399 q
->last_merge
= NULL
;
1404 * A hw segment is just getting larger, bump just the phys
1407 req
->nr_phys_segments
+= nr_phys_segs
;
1411 static inline int ll_new_hw_segment(struct request_queue
*q
,
1412 struct request
*req
,
1415 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1416 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1418 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1419 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1420 req
->cmd_flags
|= REQ_NOMERGE
;
1421 if (req
== q
->last_merge
)
1422 q
->last_merge
= NULL
;
1427 * This will form the start of a new hw segment. Bump both
1430 req
->nr_hw_segments
+= nr_hw_segs
;
1431 req
->nr_phys_segments
+= nr_phys_segs
;
1435 static int ll_back_merge_fn(struct request_queue
*q
, struct request
*req
,
1438 unsigned short max_sectors
;
1441 if (unlikely(blk_pc_request(req
)))
1442 max_sectors
= q
->max_hw_sectors
;
1444 max_sectors
= q
->max_sectors
;
1446 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1447 req
->cmd_flags
|= REQ_NOMERGE
;
1448 if (req
== q
->last_merge
)
1449 q
->last_merge
= NULL
;
1452 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1453 blk_recount_segments(q
, req
->biotail
);
1454 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1455 blk_recount_segments(q
, bio
);
1456 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1457 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1458 !BIOVEC_VIRT_OVERSIZE(len
)) {
1459 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1462 if (req
->nr_hw_segments
== 1)
1463 req
->bio
->bi_hw_front_size
= len
;
1464 if (bio
->bi_hw_segments
== 1)
1465 bio
->bi_hw_back_size
= len
;
1470 return ll_new_hw_segment(q
, req
, bio
);
1473 static int ll_front_merge_fn(struct request_queue
*q
, struct request
*req
,
1476 unsigned short max_sectors
;
1479 if (unlikely(blk_pc_request(req
)))
1480 max_sectors
= q
->max_hw_sectors
;
1482 max_sectors
= q
->max_sectors
;
1485 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1486 req
->cmd_flags
|= REQ_NOMERGE
;
1487 if (req
== q
->last_merge
)
1488 q
->last_merge
= NULL
;
1491 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1492 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1493 blk_recount_segments(q
, bio
);
1494 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1495 blk_recount_segments(q
, req
->bio
);
1496 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1497 !BIOVEC_VIRT_OVERSIZE(len
)) {
1498 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1501 if (bio
->bi_hw_segments
== 1)
1502 bio
->bi_hw_front_size
= len
;
1503 if (req
->nr_hw_segments
== 1)
1504 req
->biotail
->bi_hw_back_size
= len
;
1509 return ll_new_hw_segment(q
, req
, bio
);
1512 static int ll_merge_requests_fn(struct request_queue
*q
, struct request
*req
,
1513 struct request
*next
)
1515 int total_phys_segments
;
1516 int total_hw_segments
;
1519 * First check if the either of the requests are re-queued
1520 * requests. Can't merge them if they are.
1522 if (req
->special
|| next
->special
)
1526 * Will it become too large?
1528 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1531 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1532 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1533 total_phys_segments
--;
1535 if (total_phys_segments
> q
->max_phys_segments
)
1538 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1539 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1540 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1542 * propagate the combined length to the end of the requests
1544 if (req
->nr_hw_segments
== 1)
1545 req
->bio
->bi_hw_front_size
= len
;
1546 if (next
->nr_hw_segments
== 1)
1547 next
->biotail
->bi_hw_back_size
= len
;
1548 total_hw_segments
--;
1551 if (total_hw_segments
> q
->max_hw_segments
)
1554 /* Merge is OK... */
1555 req
->nr_phys_segments
= total_phys_segments
;
1556 req
->nr_hw_segments
= total_hw_segments
;
1561 * "plug" the device if there are no outstanding requests: this will
1562 * force the transfer to start only after we have put all the requests
1565 * This is called with interrupts off and no requests on the queue and
1566 * with the queue lock held.
1568 void blk_plug_device(struct request_queue
*q
)
1570 WARN_ON(!irqs_disabled());
1573 * don't plug a stopped queue, it must be paired with blk_start_queue()
1574 * which will restart the queueing
1576 if (blk_queue_stopped(q
))
1579 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
1580 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1581 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
1585 EXPORT_SYMBOL(blk_plug_device
);
1588 * remove the queue from the plugged list, if present. called with
1589 * queue lock held and interrupts disabled.
1591 int blk_remove_plug(struct request_queue
*q
)
1593 WARN_ON(!irqs_disabled());
1595 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1598 del_timer(&q
->unplug_timer
);
1602 EXPORT_SYMBOL(blk_remove_plug
);
1605 * remove the plug and let it rip..
1607 void __generic_unplug_device(struct request_queue
*q
)
1609 if (unlikely(blk_queue_stopped(q
)))
1612 if (!blk_remove_plug(q
))
1617 EXPORT_SYMBOL(__generic_unplug_device
);
1620 * generic_unplug_device - fire a request queue
1621 * @q: The &struct request_queue in question
1624 * Linux uses plugging to build bigger requests queues before letting
1625 * the device have at them. If a queue is plugged, the I/O scheduler
1626 * is still adding and merging requests on the queue. Once the queue
1627 * gets unplugged, the request_fn defined for the queue is invoked and
1628 * transfers started.
1630 void generic_unplug_device(struct request_queue
*q
)
1632 spin_lock_irq(q
->queue_lock
);
1633 __generic_unplug_device(q
);
1634 spin_unlock_irq(q
->queue_lock
);
1636 EXPORT_SYMBOL(generic_unplug_device
);
1638 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1641 struct request_queue
*q
= bdi
->unplug_io_data
;
1646 static void blk_unplug_work(struct work_struct
*work
)
1648 struct request_queue
*q
=
1649 container_of(work
, struct request_queue
, unplug_work
);
1651 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1652 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1657 static void blk_unplug_timeout(unsigned long data
)
1659 struct request_queue
*q
= (struct request_queue
*)data
;
1661 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
1662 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1664 kblockd_schedule_work(&q
->unplug_work
);
1667 void blk_unplug(struct request_queue
*q
)
1670 * devices don't necessarily have an ->unplug_fn defined
1673 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1674 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1679 EXPORT_SYMBOL(blk_unplug
);
1682 * blk_start_queue - restart a previously stopped queue
1683 * @q: The &struct request_queue in question
1686 * blk_start_queue() will clear the stop flag on the queue, and call
1687 * the request_fn for the queue if it was in a stopped state when
1688 * entered. Also see blk_stop_queue(). Queue lock must be held.
1690 void blk_start_queue(struct request_queue
*q
)
1692 WARN_ON(!irqs_disabled());
1694 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1697 * one level of recursion is ok and is much faster than kicking
1698 * the unplug handling
1700 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1702 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1705 kblockd_schedule_work(&q
->unplug_work
);
1709 EXPORT_SYMBOL(blk_start_queue
);
1712 * blk_stop_queue - stop a queue
1713 * @q: The &struct request_queue in question
1716 * The Linux block layer assumes that a block driver will consume all
1717 * entries on the request queue when the request_fn strategy is called.
1718 * Often this will not happen, because of hardware limitations (queue
1719 * depth settings). If a device driver gets a 'queue full' response,
1720 * or if it simply chooses not to queue more I/O at one point, it can
1721 * call this function to prevent the request_fn from being called until
1722 * the driver has signalled it's ready to go again. This happens by calling
1723 * blk_start_queue() to restart queue operations. Queue lock must be held.
1725 void blk_stop_queue(struct request_queue
*q
)
1728 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1730 EXPORT_SYMBOL(blk_stop_queue
);
1733 * blk_sync_queue - cancel any pending callbacks on a queue
1737 * The block layer may perform asynchronous callback activity
1738 * on a queue, such as calling the unplug function after a timeout.
1739 * A block device may call blk_sync_queue to ensure that any
1740 * such activity is cancelled, thus allowing it to release resources
1741 * that the callbacks might use. The caller must already have made sure
1742 * that its ->make_request_fn will not re-add plugging prior to calling
1746 void blk_sync_queue(struct request_queue
*q
)
1748 del_timer_sync(&q
->unplug_timer
);
1749 kblockd_flush_work(&q
->unplug_work
);
1751 EXPORT_SYMBOL(blk_sync_queue
);
1754 * blk_run_queue - run a single device queue
1755 * @q: The queue to run
1757 void blk_run_queue(struct request_queue
*q
)
1759 unsigned long flags
;
1761 spin_lock_irqsave(q
->queue_lock
, flags
);
1765 * Only recurse once to avoid overrunning the stack, let the unplug
1766 * handling reinvoke the handler shortly if we already got there.
1768 if (!elv_queue_empty(q
)) {
1769 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1771 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1774 kblockd_schedule_work(&q
->unplug_work
);
1778 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1780 EXPORT_SYMBOL(blk_run_queue
);
1783 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
1784 * @kobj: the kobj belonging of the request queue to be released
1787 * blk_cleanup_queue is the pair to blk_init_queue() or
1788 * blk_queue_make_request(). It should be called when a request queue is
1789 * being released; typically when a block device is being de-registered.
1790 * Currently, its primary task it to free all the &struct request
1791 * structures that were allocated to the queue and the queue itself.
1794 * Hopefully the low level driver will have finished any
1795 * outstanding requests first...
1797 static void blk_release_queue(struct kobject
*kobj
)
1799 struct request_queue
*q
=
1800 container_of(kobj
, struct request_queue
, kobj
);
1801 struct request_list
*rl
= &q
->rq
;
1806 mempool_destroy(rl
->rq_pool
);
1809 __blk_queue_free_tags(q
);
1811 blk_trace_shutdown(q
);
1813 bdi_destroy(&q
->backing_dev_info
);
1814 kmem_cache_free(requestq_cachep
, q
);
1817 void blk_put_queue(struct request_queue
*q
)
1819 kobject_put(&q
->kobj
);
1821 EXPORT_SYMBOL(blk_put_queue
);
1823 void blk_cleanup_queue(struct request_queue
* q
)
1825 mutex_lock(&q
->sysfs_lock
);
1826 set_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
);
1827 mutex_unlock(&q
->sysfs_lock
);
1830 elevator_exit(q
->elevator
);
1835 EXPORT_SYMBOL(blk_cleanup_queue
);
1837 static int blk_init_free_list(struct request_queue
*q
)
1839 struct request_list
*rl
= &q
->rq
;
1841 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1842 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1844 init_waitqueue_head(&rl
->wait
[READ
]);
1845 init_waitqueue_head(&rl
->wait
[WRITE
]);
1847 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1848 mempool_free_slab
, request_cachep
, q
->node
);
1856 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
1858 return blk_alloc_queue_node(gfp_mask
, -1);
1860 EXPORT_SYMBOL(blk_alloc_queue
);
1862 static struct kobj_type queue_ktype
;
1864 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1866 struct request_queue
*q
;
1869 q
= kmem_cache_alloc_node(requestq_cachep
,
1870 gfp_mask
| __GFP_ZERO
, node_id
);
1874 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1875 q
->backing_dev_info
.unplug_io_data
= q
;
1876 err
= bdi_init(&q
->backing_dev_info
);
1878 kmem_cache_free(requestq_cachep
, q
);
1882 init_timer(&q
->unplug_timer
);
1884 kobject_init(&q
->kobj
, &queue_ktype
);
1886 mutex_init(&q
->sysfs_lock
);
1890 EXPORT_SYMBOL(blk_alloc_queue_node
);
1893 * blk_init_queue - prepare a request queue for use with a block device
1894 * @rfn: The function to be called to process requests that have been
1895 * placed on the queue.
1896 * @lock: Request queue spin lock
1899 * If a block device wishes to use the standard request handling procedures,
1900 * which sorts requests and coalesces adjacent requests, then it must
1901 * call blk_init_queue(). The function @rfn will be called when there
1902 * are requests on the queue that need to be processed. If the device
1903 * supports plugging, then @rfn may not be called immediately when requests
1904 * are available on the queue, but may be called at some time later instead.
1905 * Plugged queues are generally unplugged when a buffer belonging to one
1906 * of the requests on the queue is needed, or due to memory pressure.
1908 * @rfn is not required, or even expected, to remove all requests off the
1909 * queue, but only as many as it can handle at a time. If it does leave
1910 * requests on the queue, it is responsible for arranging that the requests
1911 * get dealt with eventually.
1913 * The queue spin lock must be held while manipulating the requests on the
1914 * request queue; this lock will be taken also from interrupt context, so irq
1915 * disabling is needed for it.
1917 * Function returns a pointer to the initialized request queue, or NULL if
1918 * it didn't succeed.
1921 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1922 * when the block device is deactivated (such as at module unload).
1925 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1927 return blk_init_queue_node(rfn
, lock
, -1);
1929 EXPORT_SYMBOL(blk_init_queue
);
1931 struct request_queue
*
1932 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1934 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1940 if (blk_init_free_list(q
)) {
1941 kmem_cache_free(requestq_cachep
, q
);
1946 * if caller didn't supply a lock, they get per-queue locking with
1950 spin_lock_init(&q
->__queue_lock
);
1951 lock
= &q
->__queue_lock
;
1954 q
->request_fn
= rfn
;
1955 q
->prep_rq_fn
= NULL
;
1956 q
->unplug_fn
= generic_unplug_device
;
1957 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1958 q
->queue_lock
= lock
;
1960 blk_queue_segment_boundary(q
, 0xffffffff);
1962 blk_queue_make_request(q
, __make_request
);
1963 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1965 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1966 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1968 q
->sg_reserved_size
= INT_MAX
;
1973 if (!elevator_init(q
, NULL
)) {
1974 blk_queue_congestion_threshold(q
);
1981 EXPORT_SYMBOL(blk_init_queue_node
);
1983 int blk_get_queue(struct request_queue
*q
)
1985 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1986 kobject_get(&q
->kobj
);
1993 EXPORT_SYMBOL(blk_get_queue
);
1995 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
1997 if (rq
->cmd_flags
& REQ_ELVPRIV
)
1998 elv_put_request(q
, rq
);
1999 mempool_free(rq
, q
->rq
.rq_pool
);
2002 static struct request
*
2003 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
2005 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
2011 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
2012 * see bio.h and blkdev.h
2014 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
2017 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
2018 mempool_free(rq
, q
->rq
.rq_pool
);
2021 rq
->cmd_flags
|= REQ_ELVPRIV
;
2028 * ioc_batching returns true if the ioc is a valid batching request and
2029 * should be given priority access to a request.
2031 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
2037 * Make sure the process is able to allocate at least 1 request
2038 * even if the batch times out, otherwise we could theoretically
2041 return ioc
->nr_batch_requests
== q
->nr_batching
||
2042 (ioc
->nr_batch_requests
> 0
2043 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
2047 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2048 * will cause the process to be a "batcher" on all queues in the system. This
2049 * is the behaviour we want though - once it gets a wakeup it should be given
2052 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
2054 if (!ioc
|| ioc_batching(q
, ioc
))
2057 ioc
->nr_batch_requests
= q
->nr_batching
;
2058 ioc
->last_waited
= jiffies
;
2061 static void __freed_request(struct request_queue
*q
, int rw
)
2063 struct request_list
*rl
= &q
->rq
;
2065 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
2066 blk_clear_queue_congested(q
, rw
);
2068 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
2069 if (waitqueue_active(&rl
->wait
[rw
]))
2070 wake_up(&rl
->wait
[rw
]);
2072 blk_clear_queue_full(q
, rw
);
2077 * A request has just been released. Account for it, update the full and
2078 * congestion status, wake up any waiters. Called under q->queue_lock.
2080 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
2082 struct request_list
*rl
= &q
->rq
;
2088 __freed_request(q
, rw
);
2090 if (unlikely(rl
->starved
[rw
^ 1]))
2091 __freed_request(q
, rw
^ 1);
2094 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2096 * Get a free request, queue_lock must be held.
2097 * Returns NULL on failure, with queue_lock held.
2098 * Returns !NULL on success, with queue_lock *not held*.
2100 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
2101 struct bio
*bio
, gfp_t gfp_mask
)
2103 struct request
*rq
= NULL
;
2104 struct request_list
*rl
= &q
->rq
;
2105 struct io_context
*ioc
= NULL
;
2106 const int rw
= rw_flags
& 0x01;
2107 int may_queue
, priv
;
2109 may_queue
= elv_may_queue(q
, rw_flags
);
2110 if (may_queue
== ELV_MQUEUE_NO
)
2113 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
2114 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
2115 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
2117 * The queue will fill after this allocation, so set
2118 * it as full, and mark this process as "batching".
2119 * This process will be allowed to complete a batch of
2120 * requests, others will be blocked.
2122 if (!blk_queue_full(q
, rw
)) {
2123 ioc_set_batching(q
, ioc
);
2124 blk_set_queue_full(q
, rw
);
2126 if (may_queue
!= ELV_MQUEUE_MUST
2127 && !ioc_batching(q
, ioc
)) {
2129 * The queue is full and the allocating
2130 * process is not a "batcher", and not
2131 * exempted by the IO scheduler
2137 blk_set_queue_congested(q
, rw
);
2141 * Only allow batching queuers to allocate up to 50% over the defined
2142 * limit of requests, otherwise we could have thousands of requests
2143 * allocated with any setting of ->nr_requests
2145 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
2149 rl
->starved
[rw
] = 0;
2151 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
2155 spin_unlock_irq(q
->queue_lock
);
2157 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
2158 if (unlikely(!rq
)) {
2160 * Allocation failed presumably due to memory. Undo anything
2161 * we might have messed up.
2163 * Allocating task should really be put onto the front of the
2164 * wait queue, but this is pretty rare.
2166 spin_lock_irq(q
->queue_lock
);
2167 freed_request(q
, rw
, priv
);
2170 * in the very unlikely event that allocation failed and no
2171 * requests for this direction was pending, mark us starved
2172 * so that freeing of a request in the other direction will
2173 * notice us. another possible fix would be to split the
2174 * rq mempool into READ and WRITE
2177 if (unlikely(rl
->count
[rw
] == 0))
2178 rl
->starved
[rw
] = 1;
2184 * ioc may be NULL here, and ioc_batching will be false. That's
2185 * OK, if the queue is under the request limit then requests need
2186 * not count toward the nr_batch_requests limit. There will always
2187 * be some limit enforced by BLK_BATCH_TIME.
2189 if (ioc_batching(q
, ioc
))
2190 ioc
->nr_batch_requests
--;
2194 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
2200 * No available requests for this queue, unplug the device and wait for some
2201 * requests to become available.
2203 * Called with q->queue_lock held, and returns with it unlocked.
2205 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
2208 const int rw
= rw_flags
& 0x01;
2211 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
2214 struct request_list
*rl
= &q
->rq
;
2216 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
2217 TASK_UNINTERRUPTIBLE
);
2219 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
2222 struct io_context
*ioc
;
2224 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
2226 __generic_unplug_device(q
);
2227 spin_unlock_irq(q
->queue_lock
);
2231 * After sleeping, we become a "batching" process and
2232 * will be able to allocate at least one request, and
2233 * up to a big batch of them for a small period time.
2234 * See ioc_batching, ioc_set_batching
2236 ioc
= current_io_context(GFP_NOIO
, q
->node
);
2237 ioc_set_batching(q
, ioc
);
2239 spin_lock_irq(q
->queue_lock
);
2241 finish_wait(&rl
->wait
[rw
], &wait
);
2247 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
2251 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2253 spin_lock_irq(q
->queue_lock
);
2254 if (gfp_mask
& __GFP_WAIT
) {
2255 rq
= get_request_wait(q
, rw
, NULL
);
2257 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2259 spin_unlock_irq(q
->queue_lock
);
2261 /* q->queue_lock is unlocked at this point */
2265 EXPORT_SYMBOL(blk_get_request
);
2268 * blk_start_queueing - initiate dispatch of requests to device
2269 * @q: request queue to kick into gear
2271 * This is basically a helper to remove the need to know whether a queue
2272 * is plugged or not if someone just wants to initiate dispatch of requests
2275 * The queue lock must be held with interrupts disabled.
2277 void blk_start_queueing(struct request_queue
*q
)
2279 if (!blk_queue_plugged(q
))
2282 __generic_unplug_device(q
);
2284 EXPORT_SYMBOL(blk_start_queueing
);
2287 * blk_requeue_request - put a request back on queue
2288 * @q: request queue where request should be inserted
2289 * @rq: request to be inserted
2292 * Drivers often keep queueing requests until the hardware cannot accept
2293 * more, when that condition happens we need to put the request back
2294 * on the queue. Must be called with queue lock held.
2296 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
2298 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
2300 if (blk_rq_tagged(rq
))
2301 blk_queue_end_tag(q
, rq
);
2303 elv_requeue_request(q
, rq
);
2306 EXPORT_SYMBOL(blk_requeue_request
);
2309 * blk_insert_request - insert a special request in to a request queue
2310 * @q: request queue where request should be inserted
2311 * @rq: request to be inserted
2312 * @at_head: insert request at head or tail of queue
2313 * @data: private data
2316 * Many block devices need to execute commands asynchronously, so they don't
2317 * block the whole kernel from preemption during request execution. This is
2318 * accomplished normally by inserting aritficial requests tagged as
2319 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2320 * scheduled for actual execution by the request queue.
2322 * We have the option of inserting the head or the tail of the queue.
2323 * Typically we use the tail for new ioctls and so forth. We use the head
2324 * of the queue for things like a QUEUE_FULL message from a device, or a
2325 * host that is unable to accept a particular command.
2327 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
2328 int at_head
, void *data
)
2330 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2331 unsigned long flags
;
2334 * tell I/O scheduler that this isn't a regular read/write (ie it
2335 * must not attempt merges on this) and that it acts as a soft
2338 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
2339 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
2343 spin_lock_irqsave(q
->queue_lock
, flags
);
2346 * If command is tagged, release the tag
2348 if (blk_rq_tagged(rq
))
2349 blk_queue_end_tag(q
, rq
);
2351 drive_stat_acct(rq
, 1);
2352 __elv_add_request(q
, rq
, where
, 0);
2353 blk_start_queueing(q
);
2354 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2357 EXPORT_SYMBOL(blk_insert_request
);
2359 static int __blk_rq_unmap_user(struct bio
*bio
)
2364 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2365 bio_unmap_user(bio
);
2367 ret
= bio_uncopy_user(bio
);
2373 int blk_rq_append_bio(struct request_queue
*q
, struct request
*rq
,
2377 blk_rq_bio_prep(q
, rq
, bio
);
2378 else if (!ll_back_merge_fn(q
, rq
, bio
))
2381 rq
->biotail
->bi_next
= bio
;
2384 rq
->data_len
+= bio
->bi_size
;
2388 EXPORT_SYMBOL(blk_rq_append_bio
);
2390 static int __blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
2391 void __user
*ubuf
, unsigned int len
)
2393 unsigned long uaddr
;
2394 struct bio
*bio
, *orig_bio
;
2397 reading
= rq_data_dir(rq
) == READ
;
2400 * if alignment requirement is satisfied, map in user pages for
2401 * direct dma. else, set up kernel bounce buffers
2403 uaddr
= (unsigned long) ubuf
;
2404 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2405 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2407 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2410 return PTR_ERR(bio
);
2413 blk_queue_bounce(q
, &bio
);
2416 * We link the bounce buffer in and could have to traverse it
2417 * later so we have to get a ref to prevent it from being freed
2421 ret
= blk_rq_append_bio(q
, rq
, bio
);
2423 return bio
->bi_size
;
2425 /* if it was boucned we must call the end io function */
2427 __blk_rq_unmap_user(orig_bio
);
2433 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2434 * @q: request queue where request should be inserted
2435 * @rq: request structure to fill
2436 * @ubuf: the user buffer
2437 * @len: length of user data
2440 * Data will be mapped directly for zero copy io, if possible. Otherwise
2441 * a kernel bounce buffer is used.
2443 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2444 * still in process context.
2446 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2447 * before being submitted to the device, as pages mapped may be out of
2448 * reach. It's the callers responsibility to make sure this happens. The
2449 * original bio must be passed back in to blk_rq_unmap_user() for proper
2452 int blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
2453 void __user
*ubuf
, unsigned long len
)
2455 unsigned long bytes_read
= 0;
2456 struct bio
*bio
= NULL
;
2459 if (len
> (q
->max_hw_sectors
<< 9))
2464 while (bytes_read
!= len
) {
2465 unsigned long map_len
, end
, start
;
2467 map_len
= min_t(unsigned long, len
- bytes_read
, BIO_MAX_SIZE
);
2468 end
= ((unsigned long)ubuf
+ map_len
+ PAGE_SIZE
- 1)
2470 start
= (unsigned long)ubuf
>> PAGE_SHIFT
;
2473 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2474 * pages. If this happens we just lower the requested
2475 * mapping len by a page so that we can fit
2477 if (end
- start
> BIO_MAX_PAGES
)
2478 map_len
-= PAGE_SIZE
;
2480 ret
= __blk_rq_map_user(q
, rq
, ubuf
, map_len
);
2489 rq
->buffer
= rq
->data
= NULL
;
2492 blk_rq_unmap_user(bio
);
2496 EXPORT_SYMBOL(blk_rq_map_user
);
2499 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2500 * @q: request queue where request should be inserted
2501 * @rq: request to map data to
2502 * @iov: pointer to the iovec
2503 * @iov_count: number of elements in the iovec
2504 * @len: I/O byte count
2507 * Data will be mapped directly for zero copy io, if possible. Otherwise
2508 * a kernel bounce buffer is used.
2510 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2511 * still in process context.
2513 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2514 * before being submitted to the device, as pages mapped may be out of
2515 * reach. It's the callers responsibility to make sure this happens. The
2516 * original bio must be passed back in to blk_rq_unmap_user() for proper
2519 int blk_rq_map_user_iov(struct request_queue
*q
, struct request
*rq
,
2520 struct sg_iovec
*iov
, int iov_count
, unsigned int len
)
2524 if (!iov
|| iov_count
<= 0)
2527 /* we don't allow misaligned data like bio_map_user() does. If the
2528 * user is using sg, they're expected to know the alignment constraints
2529 * and respect them accordingly */
2530 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2532 return PTR_ERR(bio
);
2534 if (bio
->bi_size
!= len
) {
2536 bio_unmap_user(bio
);
2541 blk_rq_bio_prep(q
, rq
, bio
);
2542 rq
->buffer
= rq
->data
= NULL
;
2546 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2549 * blk_rq_unmap_user - unmap a request with user data
2550 * @bio: start of bio list
2553 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2554 * supply the original rq->bio from the blk_rq_map_user() return, since
2555 * the io completion may have changed rq->bio.
2557 int blk_rq_unmap_user(struct bio
*bio
)
2559 struct bio
*mapped_bio
;
2564 if (unlikely(bio_flagged(bio
, BIO_BOUNCED
)))
2565 mapped_bio
= bio
->bi_private
;
2567 ret2
= __blk_rq_unmap_user(mapped_bio
);
2573 bio_put(mapped_bio
);
2579 EXPORT_SYMBOL(blk_rq_unmap_user
);
2582 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2583 * @q: request queue where request should be inserted
2584 * @rq: request to fill
2585 * @kbuf: the kernel buffer
2586 * @len: length of user data
2587 * @gfp_mask: memory allocation flags
2589 int blk_rq_map_kern(struct request_queue
*q
, struct request
*rq
, void *kbuf
,
2590 unsigned int len
, gfp_t gfp_mask
)
2594 if (len
> (q
->max_hw_sectors
<< 9))
2599 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2601 return PTR_ERR(bio
);
2603 if (rq_data_dir(rq
) == WRITE
)
2604 bio
->bi_rw
|= (1 << BIO_RW
);
2606 blk_rq_bio_prep(q
, rq
, bio
);
2607 blk_queue_bounce(q
, &rq
->bio
);
2608 rq
->buffer
= rq
->data
= NULL
;
2612 EXPORT_SYMBOL(blk_rq_map_kern
);
2615 * blk_execute_rq_nowait - insert a request into queue for execution
2616 * @q: queue to insert the request in
2617 * @bd_disk: matching gendisk
2618 * @rq: request to insert
2619 * @at_head: insert request at head or tail of queue
2620 * @done: I/O completion handler
2623 * Insert a fully prepared request at the back of the io scheduler queue
2624 * for execution. Don't wait for completion.
2626 void blk_execute_rq_nowait(struct request_queue
*q
, struct gendisk
*bd_disk
,
2627 struct request
*rq
, int at_head
,
2630 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2632 rq
->rq_disk
= bd_disk
;
2633 rq
->cmd_flags
|= REQ_NOMERGE
;
2635 WARN_ON(irqs_disabled());
2636 spin_lock_irq(q
->queue_lock
);
2637 __elv_add_request(q
, rq
, where
, 1);
2638 __generic_unplug_device(q
);
2639 spin_unlock_irq(q
->queue_lock
);
2641 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait
);
2644 * blk_execute_rq - insert a request into queue for execution
2645 * @q: queue to insert the request in
2646 * @bd_disk: matching gendisk
2647 * @rq: request to insert
2648 * @at_head: insert request at head or tail of queue
2651 * Insert a fully prepared request at the back of the io scheduler queue
2652 * for execution and wait for completion.
2654 int blk_execute_rq(struct request_queue
*q
, struct gendisk
*bd_disk
,
2655 struct request
*rq
, int at_head
)
2657 DECLARE_COMPLETION_ONSTACK(wait
);
2658 char sense
[SCSI_SENSE_BUFFERSIZE
];
2662 * we need an extra reference to the request, so we can look at
2663 * it after io completion
2668 memset(sense
, 0, sizeof(sense
));
2673 rq
->end_io_data
= &wait
;
2674 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2675 wait_for_completion(&wait
);
2683 EXPORT_SYMBOL(blk_execute_rq
);
2685 static void bio_end_empty_barrier(struct bio
*bio
, int err
)
2688 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2690 complete(bio
->bi_private
);
2694 * blkdev_issue_flush - queue a flush
2695 * @bdev: blockdev to issue flush for
2696 * @error_sector: error sector
2699 * Issue a flush for the block device in question. Caller can supply
2700 * room for storing the error offset in case of a flush error, if they
2701 * wish to. Caller must run wait_for_completion() on its own.
2703 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2705 DECLARE_COMPLETION_ONSTACK(wait
);
2706 struct request_queue
*q
;
2710 if (bdev
->bd_disk
== NULL
)
2713 q
= bdev_get_queue(bdev
);
2717 bio
= bio_alloc(GFP_KERNEL
, 0);
2721 bio
->bi_end_io
= bio_end_empty_barrier
;
2722 bio
->bi_private
= &wait
;
2723 bio
->bi_bdev
= bdev
;
2724 submit_bio(1 << BIO_RW_BARRIER
, bio
);
2726 wait_for_completion(&wait
);
2729 * The driver must store the error location in ->bi_sector, if
2730 * it supports it. For non-stacked drivers, this should be copied
2734 *error_sector
= bio
->bi_sector
;
2737 if (!bio_flagged(bio
, BIO_UPTODATE
))
2744 EXPORT_SYMBOL(blkdev_issue_flush
);
2746 static void drive_stat_acct(struct request
*rq
, int new_io
)
2748 int rw
= rq_data_dir(rq
);
2750 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2754 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2756 disk_round_stats(rq
->rq_disk
);
2757 rq
->rq_disk
->in_flight
++;
2762 * add-request adds a request to the linked list.
2763 * queue lock is held and interrupts disabled, as we muck with the
2764 * request queue list.
2766 static inline void add_request(struct request_queue
* q
, struct request
* req
)
2768 drive_stat_acct(req
, 1);
2771 * elevator indicated where it wants this request to be
2772 * inserted at elevator_merge time
2774 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2778 * disk_round_stats() - Round off the performance stats on a struct
2781 * The average IO queue length and utilisation statistics are maintained
2782 * by observing the current state of the queue length and the amount of
2783 * time it has been in this state for.
2785 * Normally, that accounting is done on IO completion, but that can result
2786 * in more than a second's worth of IO being accounted for within any one
2787 * second, leading to >100% utilisation. To deal with that, we call this
2788 * function to do a round-off before returning the results when reading
2789 * /proc/diskstats. This accounts immediately for all queue usage up to
2790 * the current jiffies and restarts the counters again.
2792 void disk_round_stats(struct gendisk
*disk
)
2794 unsigned long now
= jiffies
;
2796 if (now
== disk
->stamp
)
2799 if (disk
->in_flight
) {
2800 __disk_stat_add(disk
, time_in_queue
,
2801 disk
->in_flight
* (now
- disk
->stamp
));
2802 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2807 EXPORT_SYMBOL_GPL(disk_round_stats
);
2810 * queue lock must be held
2812 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
2816 if (unlikely(--req
->ref_count
))
2819 elv_completed_request(q
, req
);
2822 * Request may not have originated from ll_rw_blk. if not,
2823 * it didn't come out of our reserved rq pools
2825 if (req
->cmd_flags
& REQ_ALLOCED
) {
2826 int rw
= rq_data_dir(req
);
2827 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
2829 BUG_ON(!list_empty(&req
->queuelist
));
2830 BUG_ON(!hlist_unhashed(&req
->hash
));
2832 blk_free_request(q
, req
);
2833 freed_request(q
, rw
, priv
);
2837 EXPORT_SYMBOL_GPL(__blk_put_request
);
2839 void blk_put_request(struct request
*req
)
2841 unsigned long flags
;
2842 struct request_queue
*q
= req
->q
;
2845 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2846 * following if (q) test.
2849 spin_lock_irqsave(q
->queue_lock
, flags
);
2850 __blk_put_request(q
, req
);
2851 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2855 EXPORT_SYMBOL(blk_put_request
);
2858 * blk_end_sync_rq - executes a completion event on a request
2859 * @rq: request to complete
2860 * @error: end io status of the request
2862 void blk_end_sync_rq(struct request
*rq
, int error
)
2864 struct completion
*waiting
= rq
->end_io_data
;
2866 rq
->end_io_data
= NULL
;
2867 __blk_put_request(rq
->q
, rq
);
2870 * complete last, if this is a stack request the process (and thus
2871 * the rq pointer) could be invalid right after this complete()
2875 EXPORT_SYMBOL(blk_end_sync_rq
);
2878 * Has to be called with the request spinlock acquired
2880 static int attempt_merge(struct request_queue
*q
, struct request
*req
,
2881 struct request
*next
)
2883 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2889 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2892 if (rq_data_dir(req
) != rq_data_dir(next
)
2893 || req
->rq_disk
!= next
->rq_disk
2898 * If we are allowed to merge, then append bio list
2899 * from next to rq and release next. merge_requests_fn
2900 * will have updated segment counts, update sector
2903 if (!ll_merge_requests_fn(q
, req
, next
))
2907 * At this point we have either done a back merge
2908 * or front merge. We need the smaller start_time of
2909 * the merged requests to be the current request
2910 * for accounting purposes.
2912 if (time_after(req
->start_time
, next
->start_time
))
2913 req
->start_time
= next
->start_time
;
2915 req
->biotail
->bi_next
= next
->bio
;
2916 req
->biotail
= next
->biotail
;
2918 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2920 elv_merge_requests(q
, req
, next
);
2923 disk_round_stats(req
->rq_disk
);
2924 req
->rq_disk
->in_flight
--;
2927 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2929 __blk_put_request(q
, next
);
2933 static inline int attempt_back_merge(struct request_queue
*q
,
2936 struct request
*next
= elv_latter_request(q
, rq
);
2939 return attempt_merge(q
, rq
, next
);
2944 static inline int attempt_front_merge(struct request_queue
*q
,
2947 struct request
*prev
= elv_former_request(q
, rq
);
2950 return attempt_merge(q
, prev
, rq
);
2955 static void init_request_from_bio(struct request
*req
, struct bio
*bio
)
2957 req
->cmd_type
= REQ_TYPE_FS
;
2960 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2962 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2963 req
->cmd_flags
|= REQ_FAILFAST
;
2966 * REQ_BARRIER implies no merging, but lets make it explicit
2968 if (unlikely(bio_barrier(bio
)))
2969 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2972 req
->cmd_flags
|= REQ_RW_SYNC
;
2973 if (bio_rw_meta(bio
))
2974 req
->cmd_flags
|= REQ_RW_META
;
2977 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
2978 req
->ioprio
= bio_prio(bio
);
2979 req
->start_time
= jiffies
;
2980 blk_rq_bio_prep(req
->q
, req
, bio
);
2983 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
2985 struct request
*req
;
2986 int el_ret
, nr_sectors
, barrier
, err
;
2987 const unsigned short prio
= bio_prio(bio
);
2988 const int sync
= bio_sync(bio
);
2991 nr_sectors
= bio_sectors(bio
);
2994 * low level driver can indicate that it wants pages above a
2995 * certain limit bounced to low memory (ie for highmem, or even
2996 * ISA dma in theory)
2998 blk_queue_bounce(q
, &bio
);
3000 barrier
= bio_barrier(bio
);
3001 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
3006 spin_lock_irq(q
->queue_lock
);
3008 if (unlikely(barrier
) || elv_queue_empty(q
))
3011 el_ret
= elv_merge(q
, &req
, bio
);
3013 case ELEVATOR_BACK_MERGE
:
3014 BUG_ON(!rq_mergeable(req
));
3016 if (!ll_back_merge_fn(q
, req
, bio
))
3019 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
3021 req
->biotail
->bi_next
= bio
;
3023 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
3024 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
3025 drive_stat_acct(req
, 0);
3026 if (!attempt_back_merge(q
, req
))
3027 elv_merged_request(q
, req
, el_ret
);
3030 case ELEVATOR_FRONT_MERGE
:
3031 BUG_ON(!rq_mergeable(req
));
3033 if (!ll_front_merge_fn(q
, req
, bio
))
3036 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
3038 bio
->bi_next
= req
->bio
;
3042 * may not be valid. if the low level driver said
3043 * it didn't need a bounce buffer then it better
3044 * not touch req->buffer either...
3046 req
->buffer
= bio_data(bio
);
3047 req
->current_nr_sectors
= bio_cur_sectors(bio
);
3048 req
->hard_cur_sectors
= req
->current_nr_sectors
;
3049 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
3050 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
3051 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
3052 drive_stat_acct(req
, 0);
3053 if (!attempt_front_merge(q
, req
))
3054 elv_merged_request(q
, req
, el_ret
);
3057 /* ELV_NO_MERGE: elevator says don't/can't merge. */
3064 * This sync check and mask will be re-done in init_request_from_bio(),
3065 * but we need to set it earlier to expose the sync flag to the
3066 * rq allocator and io schedulers.
3068 rw_flags
= bio_data_dir(bio
);
3070 rw_flags
|= REQ_RW_SYNC
;
3073 * Grab a free request. This is might sleep but can not fail.
3074 * Returns with the queue unlocked.
3076 req
= get_request_wait(q
, rw_flags
, bio
);
3079 * After dropping the lock and possibly sleeping here, our request
3080 * may now be mergeable after it had proven unmergeable (above).
3081 * We don't worry about that case for efficiency. It won't happen
3082 * often, and the elevators are able to handle it.
3084 init_request_from_bio(req
, bio
);
3086 spin_lock_irq(q
->queue_lock
);
3087 if (elv_queue_empty(q
))
3089 add_request(q
, req
);
3092 __generic_unplug_device(q
);
3094 spin_unlock_irq(q
->queue_lock
);
3098 bio_endio(bio
, err
);
3103 * If bio->bi_dev is a partition, remap the location
3105 static inline void blk_partition_remap(struct bio
*bio
)
3107 struct block_device
*bdev
= bio
->bi_bdev
;
3109 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
3110 struct hd_struct
*p
= bdev
->bd_part
;
3111 const int rw
= bio_data_dir(bio
);
3113 p
->sectors
[rw
] += bio_sectors(bio
);
3116 bio
->bi_sector
+= p
->start_sect
;
3117 bio
->bi_bdev
= bdev
->bd_contains
;
3119 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
3120 bdev
->bd_dev
, bio
->bi_sector
,
3121 bio
->bi_sector
- p
->start_sect
);
3125 static void handle_bad_sector(struct bio
*bio
)
3127 char b
[BDEVNAME_SIZE
];
3129 printk(KERN_INFO
"attempt to access beyond end of device\n");
3130 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
3131 bdevname(bio
->bi_bdev
, b
),
3133 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
3134 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
3136 set_bit(BIO_EOF
, &bio
->bi_flags
);
3139 #ifdef CONFIG_FAIL_MAKE_REQUEST
3141 static DECLARE_FAULT_ATTR(fail_make_request
);
3143 static int __init
setup_fail_make_request(char *str
)
3145 return setup_fault_attr(&fail_make_request
, str
);
3147 __setup("fail_make_request=", setup_fail_make_request
);
3149 static int should_fail_request(struct bio
*bio
)
3151 if ((bio
->bi_bdev
->bd_disk
->flags
& GENHD_FL_FAIL
) ||
3152 (bio
->bi_bdev
->bd_part
&& bio
->bi_bdev
->bd_part
->make_it_fail
))
3153 return should_fail(&fail_make_request
, bio
->bi_size
);
3158 static int __init
fail_make_request_debugfs(void)
3160 return init_fault_attr_dentries(&fail_make_request
,
3161 "fail_make_request");
3164 late_initcall(fail_make_request_debugfs
);
3166 #else /* CONFIG_FAIL_MAKE_REQUEST */
3168 static inline int should_fail_request(struct bio
*bio
)
3173 #endif /* CONFIG_FAIL_MAKE_REQUEST */
3176 * Check whether this bio extends beyond the end of the device.
3178 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
3185 /* Test device or partition size, when known. */
3186 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3188 sector_t sector
= bio
->bi_sector
;
3190 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
3192 * This may well happen - the kernel calls bread()
3193 * without checking the size of the device, e.g., when
3194 * mounting a device.
3196 handle_bad_sector(bio
);
3205 * generic_make_request: hand a buffer to its device driver for I/O
3206 * @bio: The bio describing the location in memory and on the device.
3208 * generic_make_request() is used to make I/O requests of block
3209 * devices. It is passed a &struct bio, which describes the I/O that needs
3212 * generic_make_request() does not return any status. The
3213 * success/failure status of the request, along with notification of
3214 * completion, is delivered asynchronously through the bio->bi_end_io
3215 * function described (one day) else where.
3217 * The caller of generic_make_request must make sure that bi_io_vec
3218 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3219 * set to describe the device address, and the
3220 * bi_end_io and optionally bi_private are set to describe how
3221 * completion notification should be signaled.
3223 * generic_make_request and the drivers it calls may use bi_next if this
3224 * bio happens to be merged with someone else, and may change bi_dev and
3225 * bi_sector for remaps as it sees fit. So the values of these fields
3226 * should NOT be depended on after the call to generic_make_request.
3228 static inline void __generic_make_request(struct bio
*bio
)
3230 struct request_queue
*q
;
3231 sector_t old_sector
;
3232 int ret
, nr_sectors
= bio_sectors(bio
);
3238 if (bio_check_eod(bio
, nr_sectors
))
3242 * Resolve the mapping until finished. (drivers are
3243 * still free to implement/resolve their own stacking
3244 * by explicitly returning 0)
3246 * NOTE: we don't repeat the blk_size check for each new device.
3247 * Stacking drivers are expected to know what they are doing.
3252 char b
[BDEVNAME_SIZE
];
3254 q
= bdev_get_queue(bio
->bi_bdev
);
3257 "generic_make_request: Trying to access "
3258 "nonexistent block-device %s (%Lu)\n",
3259 bdevname(bio
->bi_bdev
, b
),
3260 (long long) bio
->bi_sector
);
3262 bio_endio(bio
, err
);
3266 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
3267 printk("bio too big device %s (%u > %u)\n",
3268 bdevname(bio
->bi_bdev
, b
),
3274 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
3277 if (should_fail_request(bio
))
3281 * If this device has partitions, remap block n
3282 * of partition p to block n+start(p) of the disk.
3284 blk_partition_remap(bio
);
3286 if (old_sector
!= -1)
3287 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
3290 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
3292 old_sector
= bio
->bi_sector
;
3293 old_dev
= bio
->bi_bdev
->bd_dev
;
3295 if (bio_check_eod(bio
, nr_sectors
))
3297 if (bio_empty_barrier(bio
) && !q
->prepare_flush_fn
) {
3302 ret
= q
->make_request_fn(q
, bio
);
3307 * We only want one ->make_request_fn to be active at a time,
3308 * else stack usage with stacked devices could be a problem.
3309 * So use current->bio_{list,tail} to keep a list of requests
3310 * submited by a make_request_fn function.
3311 * current->bio_tail is also used as a flag to say if
3312 * generic_make_request is currently active in this task or not.
3313 * If it is NULL, then no make_request is active. If it is non-NULL,
3314 * then a make_request is active, and new requests should be added
3317 void generic_make_request(struct bio
*bio
)
3319 if (current
->bio_tail
) {
3320 /* make_request is active */
3321 *(current
->bio_tail
) = bio
;
3322 bio
->bi_next
= NULL
;
3323 current
->bio_tail
= &bio
->bi_next
;
3326 /* following loop may be a bit non-obvious, and so deserves some
3328 * Before entering the loop, bio->bi_next is NULL (as all callers
3329 * ensure that) so we have a list with a single bio.
3330 * We pretend that we have just taken it off a longer list, so
3331 * we assign bio_list to the next (which is NULL) and bio_tail
3332 * to &bio_list, thus initialising the bio_list of new bios to be
3333 * added. __generic_make_request may indeed add some more bios
3334 * through a recursive call to generic_make_request. If it
3335 * did, we find a non-NULL value in bio_list and re-enter the loop
3336 * from the top. In this case we really did just take the bio
3337 * of the top of the list (no pretending) and so fixup bio_list and
3338 * bio_tail or bi_next, and call into __generic_make_request again.
3340 * The loop was structured like this to make only one call to
3341 * __generic_make_request (which is important as it is large and
3342 * inlined) and to keep the structure simple.
3344 BUG_ON(bio
->bi_next
);
3346 current
->bio_list
= bio
->bi_next
;
3347 if (bio
->bi_next
== NULL
)
3348 current
->bio_tail
= ¤t
->bio_list
;
3350 bio
->bi_next
= NULL
;
3351 __generic_make_request(bio
);
3352 bio
= current
->bio_list
;
3354 current
->bio_tail
= NULL
; /* deactivate */
3357 EXPORT_SYMBOL(generic_make_request
);
3360 * submit_bio: submit a bio to the block device layer for I/O
3361 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3362 * @bio: The &struct bio which describes the I/O
3364 * submit_bio() is very similar in purpose to generic_make_request(), and
3365 * uses that function to do most of the work. Both are fairly rough
3366 * interfaces, @bio must be presetup and ready for I/O.
3369 void submit_bio(int rw
, struct bio
*bio
)
3371 int count
= bio_sectors(bio
);
3376 * If it's a regular read/write or a barrier with data attached,
3377 * go through the normal accounting stuff before submission.
3379 if (!bio_empty_barrier(bio
)) {
3381 BIO_BUG_ON(!bio
->bi_size
);
3382 BIO_BUG_ON(!bio
->bi_io_vec
);
3385 count_vm_events(PGPGOUT
, count
);
3387 task_io_account_read(bio
->bi_size
);
3388 count_vm_events(PGPGIN
, count
);
3391 if (unlikely(block_dump
)) {
3392 char b
[BDEVNAME_SIZE
];
3393 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
3394 current
->comm
, task_pid_nr(current
),
3395 (rw
& WRITE
) ? "WRITE" : "READ",
3396 (unsigned long long)bio
->bi_sector
,
3397 bdevname(bio
->bi_bdev
,b
));
3401 generic_make_request(bio
);
3404 EXPORT_SYMBOL(submit_bio
);
3406 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3408 if (blk_fs_request(rq
)) {
3409 rq
->hard_sector
+= nsect
;
3410 rq
->hard_nr_sectors
-= nsect
;
3413 * Move the I/O submission pointers ahead if required.
3415 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3416 (rq
->sector
<= rq
->hard_sector
)) {
3417 rq
->sector
= rq
->hard_sector
;
3418 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3419 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3420 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3421 rq
->buffer
= bio_data(rq
->bio
);
3425 * if total number of sectors is less than the first segment
3426 * size, something has gone terribly wrong
3428 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3429 printk("blk: request botched\n");
3430 rq
->nr_sectors
= rq
->current_nr_sectors
;
3435 static int __end_that_request_first(struct request
*req
, int uptodate
,
3438 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3441 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
3444 * extend uptodate bool to allow < 0 value to be direct io error
3447 if (end_io_error(uptodate
))
3448 error
= !uptodate
? -EIO
: uptodate
;
3451 * for a REQ_BLOCK_PC request, we want to carry any eventual
3452 * sense key with us all the way through
3454 if (!blk_pc_request(req
))
3458 if (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))
3459 printk("end_request: I/O error, dev %s, sector %llu\n",
3460 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3461 (unsigned long long)req
->sector
);
3464 if (blk_fs_request(req
) && req
->rq_disk
) {
3465 const int rw
= rq_data_dir(req
);
3467 disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3470 total_bytes
= bio_nbytes
= 0;
3471 while ((bio
= req
->bio
) != NULL
) {
3475 * For an empty barrier request, the low level driver must
3476 * store a potential error location in ->sector. We pass
3477 * that back up in ->bi_sector.
3479 if (blk_empty_barrier(req
))
3480 bio
->bi_sector
= req
->sector
;
3482 if (nr_bytes
>= bio
->bi_size
) {
3483 req
->bio
= bio
->bi_next
;
3484 nbytes
= bio
->bi_size
;
3485 req_bio_endio(req
, bio
, nbytes
, error
);
3489 int idx
= bio
->bi_idx
+ next_idx
;
3491 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3492 blk_dump_rq_flags(req
, "__end_that");
3493 printk("%s: bio idx %d >= vcnt %d\n",
3495 bio
->bi_idx
, bio
->bi_vcnt
);
3499 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3500 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3503 * not a complete bvec done
3505 if (unlikely(nbytes
> nr_bytes
)) {
3506 bio_nbytes
+= nr_bytes
;
3507 total_bytes
+= nr_bytes
;
3512 * advance to the next vector
3515 bio_nbytes
+= nbytes
;
3518 total_bytes
+= nbytes
;
3521 if ((bio
= req
->bio
)) {
3523 * end more in this run, or just return 'not-done'
3525 if (unlikely(nr_bytes
<= 0))
3537 * if the request wasn't completed, update state
3540 req_bio_endio(req
, bio
, bio_nbytes
, error
);
3541 bio
->bi_idx
+= next_idx
;
3542 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3543 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3546 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3547 blk_recalc_rq_segments(req
);
3552 * end_that_request_first - end I/O on a request
3553 * @req: the request being processed
3554 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3555 * @nr_sectors: number of sectors to end I/O on
3558 * Ends I/O on a number of sectors attached to @req, and sets it up
3559 * for the next range of segments (if any) in the cluster.
3562 * 0 - we are done with this request, call end_that_request_last()
3563 * 1 - still buffers pending for this request
3565 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3567 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3570 EXPORT_SYMBOL(end_that_request_first
);
3573 * end_that_request_chunk - end I/O on a request
3574 * @req: the request being processed
3575 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3576 * @nr_bytes: number of bytes to complete
3579 * Ends I/O on a number of bytes attached to @req, and sets it up
3580 * for the next range of segments (if any). Like end_that_request_first(),
3581 * but deals with bytes instead of sectors.
3584 * 0 - we are done with this request, call end_that_request_last()
3585 * 1 - still buffers pending for this request
3587 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3589 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3592 EXPORT_SYMBOL(end_that_request_chunk
);
3595 * splice the completion data to a local structure and hand off to
3596 * process_completion_queue() to complete the requests
3598 static void blk_done_softirq(struct softirq_action
*h
)
3600 struct list_head
*cpu_list
, local_list
;
3602 local_irq_disable();
3603 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3604 list_replace_init(cpu_list
, &local_list
);
3607 while (!list_empty(&local_list
)) {
3608 struct request
*rq
= list_entry(local_list
.next
, struct request
, donelist
);
3610 list_del_init(&rq
->donelist
);
3611 rq
->q
->softirq_done_fn(rq
);
3615 static int __cpuinit
blk_cpu_notify(struct notifier_block
*self
, unsigned long action
,
3619 * If a CPU goes away, splice its entries to the current CPU
3620 * and trigger a run of the softirq
3622 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3623 int cpu
= (unsigned long) hcpu
;
3625 local_irq_disable();
3626 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
3627 &__get_cpu_var(blk_cpu_done
));
3628 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3636 static struct notifier_block blk_cpu_notifier __cpuinitdata
= {
3637 .notifier_call
= blk_cpu_notify
,
3641 * blk_complete_request - end I/O on a request
3642 * @req: the request being processed
3645 * Ends all I/O on a request. It does not handle partial completions,
3646 * unless the driver actually implements this in its completion callback
3647 * through requeueing. The actual completion happens out-of-order,
3648 * through a softirq handler. The user must have registered a completion
3649 * callback through blk_queue_softirq_done().
3652 void blk_complete_request(struct request
*req
)
3654 struct list_head
*cpu_list
;
3655 unsigned long flags
;
3657 BUG_ON(!req
->q
->softirq_done_fn
);
3659 local_irq_save(flags
);
3661 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3662 list_add_tail(&req
->donelist
, cpu_list
);
3663 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3665 local_irq_restore(flags
);
3668 EXPORT_SYMBOL(blk_complete_request
);
3671 * queue lock must be held
3673 void end_that_request_last(struct request
*req
, int uptodate
)
3675 struct gendisk
*disk
= req
->rq_disk
;
3679 * extend uptodate bool to allow < 0 value to be direct io error
3682 if (end_io_error(uptodate
))
3683 error
= !uptodate
? -EIO
: uptodate
;
3685 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3686 laptop_io_completion();
3689 * Account IO completion. bar_rq isn't accounted as a normal
3690 * IO on queueing nor completion. Accounting the containing
3691 * request is enough.
3693 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
3694 unsigned long duration
= jiffies
- req
->start_time
;
3695 const int rw
= rq_data_dir(req
);
3697 __disk_stat_inc(disk
, ios
[rw
]);
3698 __disk_stat_add(disk
, ticks
[rw
], duration
);
3699 disk_round_stats(disk
);
3703 req
->end_io(req
, error
);
3705 __blk_put_request(req
->q
, req
);
3708 EXPORT_SYMBOL(end_that_request_last
);
3710 static inline void __end_request(struct request
*rq
, int uptodate
,
3711 unsigned int nr_bytes
)
3716 error
= uptodate
? uptodate
: -EIO
;
3718 __blk_end_request(rq
, error
, nr_bytes
);
3722 * blk_rq_bytes - Returns bytes left to complete in the entire request
3724 unsigned int blk_rq_bytes(struct request
*rq
)
3726 if (blk_fs_request(rq
))
3727 return rq
->hard_nr_sectors
<< 9;
3729 return rq
->data_len
;
3731 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
3734 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
3736 unsigned int blk_rq_cur_bytes(struct request
*rq
)
3738 if (blk_fs_request(rq
))
3739 return rq
->current_nr_sectors
<< 9;
3742 return rq
->bio
->bi_size
;
3744 return rq
->data_len
;
3746 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
3749 * end_queued_request - end all I/O on a queued request
3750 * @rq: the request being processed
3751 * @uptodate: error value or 0/1 uptodate flag
3754 * Ends all I/O on a request, and removes it from the block layer queues.
3755 * Not suitable for normal IO completion, unless the driver still has
3756 * the request attached to the block layer.
3759 void end_queued_request(struct request
*rq
, int uptodate
)
3761 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
3763 EXPORT_SYMBOL(end_queued_request
);
3766 * end_dequeued_request - end all I/O on a dequeued request
3767 * @rq: the request being processed
3768 * @uptodate: error value or 0/1 uptodate flag
3771 * Ends all I/O on a request. The request must already have been
3772 * dequeued using blkdev_dequeue_request(), as is normally the case
3776 void end_dequeued_request(struct request
*rq
, int uptodate
)
3778 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
3780 EXPORT_SYMBOL(end_dequeued_request
);
3784 * end_request - end I/O on the current segment of the request
3785 * @req: the request being processed
3786 * @uptodate: error value or 0/1 uptodate flag
3789 * Ends I/O on the current segment of a request. If that is the only
3790 * remaining segment, the request is also completed and freed.
3792 * This is a remnant of how older block drivers handled IO completions.
3793 * Modern drivers typically end IO on the full request in one go, unless
3794 * they have a residual value to account for. For that case this function
3795 * isn't really useful, unless the residual just happens to be the
3796 * full current segment. In other words, don't use this function in new
3797 * code. Either use end_request_completely(), or the
3798 * end_that_request_chunk() (along with end_that_request_last()) for
3799 * partial completions.
3802 void end_request(struct request
*req
, int uptodate
)
3804 __end_request(req
, uptodate
, req
->hard_cur_sectors
<< 9);
3806 EXPORT_SYMBOL(end_request
);
3808 static void complete_request(struct request
*rq
, int error
)
3811 * REMOVEME: This conversion is transitional and will be removed
3812 * when old end_that_request_* are unexported.
3816 uptodate
= (error
== -EIO
) ? 0 : error
;
3818 if (blk_rq_tagged(rq
))
3819 blk_queue_end_tag(rq
->q
, rq
);
3821 if (blk_queued_rq(rq
))
3822 blkdev_dequeue_request(rq
);
3824 end_that_request_last(rq
, uptodate
);
3828 * blk_end_io - Generic end_io function to complete a request.
3829 * @rq: the request being processed
3830 * @error: 0 for success, < 0 for error
3831 * @nr_bytes: number of bytes to complete
3832 * @drv_callback: function called between completion of bios in the request
3833 * and completion of the request.
3834 * If the callback returns non 0, this helper returns without
3835 * completion of the request.
3838 * Ends I/O on a number of bytes attached to @rq.
3839 * If @rq has leftover, sets it up for the next range of segments.
3842 * 0 - we are done with this request
3843 * 1 - this request is not freed yet, it still has pending buffers.
3845 static int blk_end_io(struct request
*rq
, int error
, int nr_bytes
,
3846 int (drv_callback
)(struct request
*))
3848 struct request_queue
*q
= rq
->q
;
3849 unsigned long flags
= 0UL;
3851 * REMOVEME: This conversion is transitional and will be removed
3852 * when old end_that_request_* are unexported.
3856 uptodate
= (error
== -EIO
) ? 0 : error
;
3858 if (blk_fs_request(rq
) || blk_pc_request(rq
)) {
3859 if (__end_that_request_first(rq
, uptodate
, nr_bytes
))
3863 /* Special feature for tricky drivers */
3864 if (drv_callback
&& drv_callback(rq
))
3867 add_disk_randomness(rq
->rq_disk
);
3869 spin_lock_irqsave(q
->queue_lock
, flags
);
3870 complete_request(rq
, error
);
3871 spin_unlock_irqrestore(q
->queue_lock
, flags
);
3877 * blk_end_request - Helper function for drivers to complete the request.
3878 * @rq: the request being processed
3879 * @error: 0 for success, < 0 for error
3880 * @nr_bytes: number of bytes to complete
3883 * Ends I/O on a number of bytes attached to @rq.
3884 * If @rq has leftover, sets it up for the next range of segments.
3887 * 0 - we are done with this request
3888 * 1 - still buffers pending for this request
3890 int blk_end_request(struct request
*rq
, int error
, int nr_bytes
)
3892 return blk_end_io(rq
, error
, nr_bytes
, NULL
);
3894 EXPORT_SYMBOL_GPL(blk_end_request
);
3897 * __blk_end_request - Helper function for drivers to complete the request.
3898 * @rq: the request being processed
3899 * @error: 0 for success, < 0 for error
3900 * @nr_bytes: number of bytes to complete
3903 * Must be called with queue lock held unlike blk_end_request().
3906 * 0 - we are done with this request
3907 * 1 - still buffers pending for this request
3909 int __blk_end_request(struct request
*rq
, int error
, int nr_bytes
)
3912 * REMOVEME: This conversion is transitional and will be removed
3913 * when old end_that_request_* are unexported.
3917 uptodate
= (error
== -EIO
) ? 0 : error
;
3919 if (blk_fs_request(rq
) || blk_pc_request(rq
)) {
3920 if (__end_that_request_first(rq
, uptodate
, nr_bytes
))
3924 add_disk_randomness(rq
->rq_disk
);
3926 complete_request(rq
, error
);
3930 EXPORT_SYMBOL_GPL(__blk_end_request
);
3933 * blk_end_request_callback - Special helper function for tricky drivers
3934 * @rq: the request being processed
3935 * @error: 0 for success, < 0 for error
3936 * @nr_bytes: number of bytes to complete
3937 * @drv_callback: function called between completion of bios in the request
3938 * and completion of the request.
3939 * If the callback returns non 0, this helper returns without
3940 * completion of the request.
3943 * Ends I/O on a number of bytes attached to @rq.
3944 * If @rq has leftover, sets it up for the next range of segments.
3946 * This special helper function is used only for existing tricky drivers.
3947 * (e.g. cdrom_newpc_intr() of ide-cd)
3948 * This interface will be removed when such drivers are rewritten.
3949 * Don't use this interface in other places anymore.
3952 * 0 - we are done with this request
3953 * 1 - this request is not freed yet.
3954 * this request still has pending buffers or
3955 * the driver doesn't want to finish this request yet.
3957 int blk_end_request_callback(struct request
*rq
, int error
, int nr_bytes
,
3958 int (drv_callback
)(struct request
*))
3960 return blk_end_io(rq
, error
, nr_bytes
, drv_callback
);
3962 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
3964 static void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3967 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3968 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
3970 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3971 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3972 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3973 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3974 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3975 rq
->buffer
= bio_data(bio
);
3976 rq
->data_len
= bio
->bi_size
;
3978 rq
->bio
= rq
->biotail
= bio
;
3981 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
3984 int kblockd_schedule_work(struct work_struct
*work
)
3986 return queue_work(kblockd_workqueue
, work
);
3989 EXPORT_SYMBOL(kblockd_schedule_work
);
3991 void kblockd_flush_work(struct work_struct
*work
)
3993 cancel_work_sync(work
);
3995 EXPORT_SYMBOL(kblockd_flush_work
);
3997 int __init
blk_dev_init(void)
4001 kblockd_workqueue
= create_workqueue("kblockd");
4002 if (!kblockd_workqueue
)
4003 panic("Failed to create kblockd\n");
4005 request_cachep
= kmem_cache_create("blkdev_requests",
4006 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
4008 requestq_cachep
= kmem_cache_create("blkdev_queue",
4009 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
4011 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
4012 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
);
4014 for_each_possible_cpu(i
)
4015 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
4017 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
4018 register_hotcpu_notifier(&blk_cpu_notifier
);
4020 blk_max_low_pfn
= max_low_pfn
- 1;
4021 blk_max_pfn
= max_pfn
- 1;
4027 * IO Context helper functions
4029 void put_io_context(struct io_context
*ioc
)
4034 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
4036 if (atomic_dec_and_test(&ioc
->refcount
)) {
4037 struct cfq_io_context
*cic
;
4040 if (ioc
->aic
&& ioc
->aic
->dtor
)
4041 ioc
->aic
->dtor(ioc
->aic
);
4042 if (ioc
->cic_root
.rb_node
!= NULL
) {
4043 struct rb_node
*n
= rb_first(&ioc
->cic_root
);
4045 cic
= rb_entry(n
, struct cfq_io_context
, rb_node
);
4050 kmem_cache_free(iocontext_cachep
, ioc
);
4053 EXPORT_SYMBOL(put_io_context
);
4055 /* Called by the exitting task */
4056 void exit_io_context(void)
4058 struct io_context
*ioc
;
4059 struct cfq_io_context
*cic
;
4062 ioc
= current
->io_context
;
4063 current
->io_context
= NULL
;
4064 task_unlock(current
);
4067 if (ioc
->aic
&& ioc
->aic
->exit
)
4068 ioc
->aic
->exit(ioc
->aic
);
4069 if (ioc
->cic_root
.rb_node
!= NULL
) {
4070 cic
= rb_entry(rb_first(&ioc
->cic_root
), struct cfq_io_context
, rb_node
);
4074 put_io_context(ioc
);
4078 * If the current task has no IO context then create one and initialise it.
4079 * Otherwise, return its existing IO context.
4081 * This returned IO context doesn't have a specifically elevated refcount,
4082 * but since the current task itself holds a reference, the context can be
4083 * used in general code, so long as it stays within `current` context.
4085 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
)
4087 struct task_struct
*tsk
= current
;
4088 struct io_context
*ret
;
4090 ret
= tsk
->io_context
;
4094 ret
= kmem_cache_alloc_node(iocontext_cachep
, gfp_flags
, node
);
4096 atomic_set(&ret
->refcount
, 1);
4097 ret
->task
= current
;
4098 ret
->ioprio_changed
= 0;
4099 ret
->last_waited
= jiffies
; /* doesn't matter... */
4100 ret
->nr_batch_requests
= 0; /* because this is 0 */
4102 ret
->cic_root
.rb_node
= NULL
;
4103 ret
->ioc_data
= NULL
;
4104 /* make sure set_task_ioprio() sees the settings above */
4106 tsk
->io_context
= ret
;
4113 * If the current task has no IO context then create one and initialise it.
4114 * If it does have a context, take a ref on it.
4116 * This is always called in the context of the task which submitted the I/O.
4118 struct io_context
*get_io_context(gfp_t gfp_flags
, int node
)
4120 struct io_context
*ret
;
4121 ret
= current_io_context(gfp_flags
, node
);
4123 atomic_inc(&ret
->refcount
);
4126 EXPORT_SYMBOL(get_io_context
);
4128 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
4130 struct io_context
*src
= *psrc
;
4131 struct io_context
*dst
= *pdst
;
4134 BUG_ON(atomic_read(&src
->refcount
) == 0);
4135 atomic_inc(&src
->refcount
);
4136 put_io_context(dst
);
4140 EXPORT_SYMBOL(copy_io_context
);
4142 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
4144 struct io_context
*temp
;
4149 EXPORT_SYMBOL(swap_io_context
);
4154 struct queue_sysfs_entry
{
4155 struct attribute attr
;
4156 ssize_t (*show
)(struct request_queue
*, char *);
4157 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
4161 queue_var_show(unsigned int var
, char *page
)
4163 return sprintf(page
, "%d\n", var
);
4167 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
4169 char *p
= (char *) page
;
4171 *var
= simple_strtoul(p
, &p
, 10);
4175 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
4177 return queue_var_show(q
->nr_requests
, (page
));
4181 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
4183 struct request_list
*rl
= &q
->rq
;
4185 int ret
= queue_var_store(&nr
, page
, count
);
4186 if (nr
< BLKDEV_MIN_RQ
)
4189 spin_lock_irq(q
->queue_lock
);
4190 q
->nr_requests
= nr
;
4191 blk_queue_congestion_threshold(q
);
4193 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
4194 blk_set_queue_congested(q
, READ
);
4195 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
4196 blk_clear_queue_congested(q
, READ
);
4198 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
4199 blk_set_queue_congested(q
, WRITE
);
4200 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
4201 blk_clear_queue_congested(q
, WRITE
);
4203 if (rl
->count
[READ
] >= q
->nr_requests
) {
4204 blk_set_queue_full(q
, READ
);
4205 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
4206 blk_clear_queue_full(q
, READ
);
4207 wake_up(&rl
->wait
[READ
]);
4210 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
4211 blk_set_queue_full(q
, WRITE
);
4212 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
4213 blk_clear_queue_full(q
, WRITE
);
4214 wake_up(&rl
->wait
[WRITE
]);
4216 spin_unlock_irq(q
->queue_lock
);
4220 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
4222 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
4224 return queue_var_show(ra_kb
, (page
));
4228 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
4230 unsigned long ra_kb
;
4231 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
4233 spin_lock_irq(q
->queue_lock
);
4234 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
4235 spin_unlock_irq(q
->queue_lock
);
4240 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
4242 int max_sectors_kb
= q
->max_sectors
>> 1;
4244 return queue_var_show(max_sectors_kb
, (page
));
4248 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
4250 unsigned long max_sectors_kb
,
4251 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
4252 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
4253 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
4255 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
4258 * Take the queue lock to update the readahead and max_sectors
4259 * values synchronously:
4261 spin_lock_irq(q
->queue_lock
);
4262 q
->max_sectors
= max_sectors_kb
<< 1;
4263 spin_unlock_irq(q
->queue_lock
);
4268 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
4270 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
4272 return queue_var_show(max_hw_sectors_kb
, (page
));
4276 static struct queue_sysfs_entry queue_requests_entry
= {
4277 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
4278 .show
= queue_requests_show
,
4279 .store
= queue_requests_store
,
4282 static struct queue_sysfs_entry queue_ra_entry
= {
4283 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
4284 .show
= queue_ra_show
,
4285 .store
= queue_ra_store
,
4288 static struct queue_sysfs_entry queue_max_sectors_entry
= {
4289 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
4290 .show
= queue_max_sectors_show
,
4291 .store
= queue_max_sectors_store
,
4294 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
4295 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
4296 .show
= queue_max_hw_sectors_show
,
4299 static struct queue_sysfs_entry queue_iosched_entry
= {
4300 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
4301 .show
= elv_iosched_show
,
4302 .store
= elv_iosched_store
,
4305 static struct attribute
*default_attrs
[] = {
4306 &queue_requests_entry
.attr
,
4307 &queue_ra_entry
.attr
,
4308 &queue_max_hw_sectors_entry
.attr
,
4309 &queue_max_sectors_entry
.attr
,
4310 &queue_iosched_entry
.attr
,
4314 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4317 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4319 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4320 struct request_queue
*q
=
4321 container_of(kobj
, struct request_queue
, kobj
);
4326 mutex_lock(&q
->sysfs_lock
);
4327 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4328 mutex_unlock(&q
->sysfs_lock
);
4331 res
= entry
->show(q
, page
);
4332 mutex_unlock(&q
->sysfs_lock
);
4337 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4338 const char *page
, size_t length
)
4340 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4341 struct request_queue
*q
= container_of(kobj
, struct request_queue
, kobj
);
4347 mutex_lock(&q
->sysfs_lock
);
4348 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4349 mutex_unlock(&q
->sysfs_lock
);
4352 res
= entry
->store(q
, page
, length
);
4353 mutex_unlock(&q
->sysfs_lock
);
4357 static struct sysfs_ops queue_sysfs_ops
= {
4358 .show
= queue_attr_show
,
4359 .store
= queue_attr_store
,
4362 static struct kobj_type queue_ktype
= {
4363 .sysfs_ops
= &queue_sysfs_ops
,
4364 .default_attrs
= default_attrs
,
4365 .release
= blk_release_queue
,
4368 int blk_register_queue(struct gendisk
*disk
)
4372 struct request_queue
*q
= disk
->queue
;
4374 if (!q
|| !q
->request_fn
)
4377 ret
= kobject_add(&q
->kobj
, kobject_get(&disk
->dev
.kobj
),
4382 kobject_uevent(&q
->kobj
, KOBJ_ADD
);
4384 ret
= elv_register_queue(q
);
4386 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4387 kobject_del(&q
->kobj
);
4394 void blk_unregister_queue(struct gendisk
*disk
)
4396 struct request_queue
*q
= disk
->queue
;
4398 if (q
&& q
->request_fn
) {
4399 elv_unregister_queue(q
);
4401 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4402 kobject_del(&q
->kobj
);
4403 kobject_put(&disk
->dev
.kobj
);