2 * linux/drivers/block/ll_rw_blk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
6 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
7 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
8 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
13 * This handles all read/write requests to block devices
15 #include <linux/config.h>
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/backing-dev.h>
19 #include <linux/bio.h>
20 #include <linux/blkdev.h>
21 #include <linux/highmem.h>
23 #include <linux/kernel_stat.h>
24 #include <linux/string.h>
25 #include <linux/init.h>
26 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/blkdev.h>
36 #include <scsi/scsi_cmnd.h>
38 static void blk_unplug_work(void *data
);
39 static void blk_unplug_timeout(unsigned long data
);
40 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
43 * For the allocated request tables
45 static kmem_cache_t
*request_cachep
;
48 * For queue allocation
50 static kmem_cache_t
*requestq_cachep
;
53 * For io context allocations
55 static kmem_cache_t
*iocontext_cachep
;
57 static wait_queue_head_t congestion_wqh
[2] = {
58 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
59 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
63 * Controlling structure to kblockd
65 static struct workqueue_struct
*kblockd_workqueue
;
67 unsigned long blk_max_low_pfn
, blk_max_pfn
;
69 EXPORT_SYMBOL(blk_max_low_pfn
);
70 EXPORT_SYMBOL(blk_max_pfn
);
72 /* Amount of time in which a process may batch requests */
73 #define BLK_BATCH_TIME (HZ/50UL)
75 /* Number of requests a "batching" process may submit */
76 #define BLK_BATCH_REQ 32
79 * Return the threshold (number of used requests) at which the queue is
80 * considered to be congested. It include a little hysteresis to keep the
81 * context switch rate down.
83 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
85 return q
->nr_congestion_on
;
89 * The threshold at which a queue is considered to be uncongested
91 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
93 return q
->nr_congestion_off
;
96 static void blk_queue_congestion_threshold(struct request_queue
*q
)
100 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
101 if (nr
> q
->nr_requests
)
103 q
->nr_congestion_on
= nr
;
105 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
108 q
->nr_congestion_off
= nr
;
112 * A queue has just exitted congestion. Note this in the global counter of
113 * congested queues, and wake up anyone who was waiting for requests to be
116 static void clear_queue_congested(request_queue_t
*q
, int rw
)
119 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
121 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
122 clear_bit(bit
, &q
->backing_dev_info
.state
);
123 smp_mb__after_clear_bit();
124 if (waitqueue_active(wqh
))
129 * A queue has just entered congestion. Flag that in the queue's VM-visible
130 * state flags and increment the global gounter of congested queues.
132 static void set_queue_congested(request_queue_t
*q
, int rw
)
136 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
137 set_bit(bit
, &q
->backing_dev_info
.state
);
141 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
144 * Locates the passed device's request queue and returns the address of its
147 * Will return NULL if the request queue cannot be located.
149 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
151 struct backing_dev_info
*ret
= NULL
;
152 request_queue_t
*q
= bdev_get_queue(bdev
);
155 ret
= &q
->backing_dev_info
;
159 EXPORT_SYMBOL(blk_get_backing_dev_info
);
161 void blk_queue_activity_fn(request_queue_t
*q
, activity_fn
*fn
, void *data
)
164 q
->activity_data
= data
;
167 EXPORT_SYMBOL(blk_queue_activity_fn
);
170 * blk_queue_prep_rq - set a prepare_request function for queue
172 * @pfn: prepare_request function
174 * It's possible for a queue to register a prepare_request callback which
175 * is invoked before the request is handed to the request_fn. The goal of
176 * the function is to prepare a request for I/O, it can be used to build a
177 * cdb from the request data for instance.
180 void blk_queue_prep_rq(request_queue_t
*q
, prep_rq_fn
*pfn
)
185 EXPORT_SYMBOL(blk_queue_prep_rq
);
188 * blk_queue_merge_bvec - set a merge_bvec function for queue
190 * @mbfn: merge_bvec_fn
192 * Usually queues have static limitations on the max sectors or segments that
193 * we can put in a request. Stacking drivers may have some settings that
194 * are dynamic, and thus we have to query the queue whether it is ok to
195 * add a new bio_vec to a bio at a given offset or not. If the block device
196 * has such limitations, it needs to register a merge_bvec_fn to control
197 * the size of bio's sent to it. Note that a block device *must* allow a
198 * single page to be added to an empty bio. The block device driver may want
199 * to use the bio_split() function to deal with these bio's. By default
200 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
203 void blk_queue_merge_bvec(request_queue_t
*q
, merge_bvec_fn
*mbfn
)
205 q
->merge_bvec_fn
= mbfn
;
208 EXPORT_SYMBOL(blk_queue_merge_bvec
);
211 * blk_queue_make_request - define an alternate make_request function for a device
212 * @q: the request queue for the device to be affected
213 * @mfn: the alternate make_request function
216 * The normal way for &struct bios to be passed to a device
217 * driver is for them to be collected into requests on a request
218 * queue, and then to allow the device driver to select requests
219 * off that queue when it is ready. This works well for many block
220 * devices. However some block devices (typically virtual devices
221 * such as md or lvm) do not benefit from the processing on the
222 * request queue, and are served best by having the requests passed
223 * directly to them. This can be achieved by providing a function
224 * to blk_queue_make_request().
227 * The driver that does this *must* be able to deal appropriately
228 * with buffers in "highmemory". This can be accomplished by either calling
229 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
230 * blk_queue_bounce() to create a buffer in normal memory.
232 void blk_queue_make_request(request_queue_t
* q
, make_request_fn
* mfn
)
237 q
->nr_requests
= BLKDEV_MAX_RQ
;
238 q
->max_phys_segments
= MAX_PHYS_SEGMENTS
;
239 q
->max_hw_segments
= MAX_HW_SEGMENTS
;
240 q
->make_request_fn
= mfn
;
241 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
242 q
->backing_dev_info
.state
= 0;
243 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
244 blk_queue_max_sectors(q
, MAX_SECTORS
);
245 blk_queue_hardsect_size(q
, 512);
246 blk_queue_dma_alignment(q
, 511);
247 blk_queue_congestion_threshold(q
);
248 q
->nr_batching
= BLK_BATCH_REQ
;
250 q
->unplug_thresh
= 4; /* hmm */
251 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
252 if (q
->unplug_delay
== 0)
255 INIT_WORK(&q
->unplug_work
, blk_unplug_work
, q
);
257 q
->unplug_timer
.function
= blk_unplug_timeout
;
258 q
->unplug_timer
.data
= (unsigned long)q
;
261 * by default assume old behaviour and bounce for any highmem page
263 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
265 blk_queue_activity_fn(q
, NULL
, NULL
);
267 INIT_LIST_HEAD(&q
->drain_list
);
270 EXPORT_SYMBOL(blk_queue_make_request
);
272 static inline void rq_init(request_queue_t
*q
, struct request
*rq
)
274 INIT_LIST_HEAD(&rq
->queuelist
);
277 rq
->rq_status
= RQ_ACTIVE
;
278 rq
->bio
= rq
->biotail
= NULL
;
288 rq
->end_io_data
= NULL
;
292 * blk_queue_ordered - does this queue support ordered writes
293 * @q: the request queue
297 * For journalled file systems, doing ordered writes on a commit
298 * block instead of explicitly doing wait_on_buffer (which is bad
299 * for performance) can be a big win. Block drivers supporting this
300 * feature should call this function and indicate so.
303 void blk_queue_ordered(request_queue_t
*q
, int flag
)
306 case QUEUE_ORDERED_NONE
:
308 kmem_cache_free(request_cachep
, q
->flush_rq
);
312 case QUEUE_ORDERED_TAG
:
315 case QUEUE_ORDERED_FLUSH
:
318 q
->flush_rq
= kmem_cache_alloc(request_cachep
,
322 printk("blk_queue_ordered: bad value %d\n", flag
);
327 EXPORT_SYMBOL(blk_queue_ordered
);
330 * blk_queue_issue_flush_fn - set function for issuing a flush
331 * @q: the request queue
332 * @iff: the function to be called issuing the flush
335 * If a driver supports issuing a flush command, the support is notified
336 * to the block layer by defining it through this call.
339 void blk_queue_issue_flush_fn(request_queue_t
*q
, issue_flush_fn
*iff
)
341 q
->issue_flush_fn
= iff
;
344 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
347 * Cache flushing for ordered writes handling
349 static void blk_pre_flush_end_io(struct request
*flush_rq
)
351 struct request
*rq
= flush_rq
->end_io_data
;
352 request_queue_t
*q
= rq
->q
;
354 rq
->flags
|= REQ_BAR_PREFLUSH
;
356 if (!flush_rq
->errors
)
357 elv_requeue_request(q
, rq
);
359 q
->end_flush_fn(q
, flush_rq
);
360 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
365 static void blk_post_flush_end_io(struct request
*flush_rq
)
367 struct request
*rq
= flush_rq
->end_io_data
;
368 request_queue_t
*q
= rq
->q
;
370 rq
->flags
|= REQ_BAR_POSTFLUSH
;
372 q
->end_flush_fn(q
, flush_rq
);
373 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
377 struct request
*blk_start_pre_flush(request_queue_t
*q
, struct request
*rq
)
379 struct request
*flush_rq
= q
->flush_rq
;
381 BUG_ON(!blk_barrier_rq(rq
));
383 if (test_and_set_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
))
386 rq_init(q
, flush_rq
);
387 flush_rq
->elevator_private
= NULL
;
388 flush_rq
->flags
= REQ_BAR_FLUSH
;
389 flush_rq
->rq_disk
= rq
->rq_disk
;
393 * prepare_flush returns 0 if no flush is needed, just mark both
394 * pre and post flush as done in that case
396 if (!q
->prepare_flush_fn(q
, flush_rq
)) {
397 rq
->flags
|= REQ_BAR_PREFLUSH
| REQ_BAR_POSTFLUSH
;
398 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
403 * some drivers dequeue requests right away, some only after io
404 * completion. make sure the request is dequeued.
406 if (!list_empty(&rq
->queuelist
))
407 blkdev_dequeue_request(rq
);
409 elv_deactivate_request(q
, rq
);
411 flush_rq
->end_io_data
= rq
;
412 flush_rq
->end_io
= blk_pre_flush_end_io
;
414 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
418 static void blk_start_post_flush(request_queue_t
*q
, struct request
*rq
)
420 struct request
*flush_rq
= q
->flush_rq
;
422 BUG_ON(!blk_barrier_rq(rq
));
424 rq_init(q
, flush_rq
);
425 flush_rq
->elevator_private
= NULL
;
426 flush_rq
->flags
= REQ_BAR_FLUSH
;
427 flush_rq
->rq_disk
= rq
->rq_disk
;
430 if (q
->prepare_flush_fn(q
, flush_rq
)) {
431 flush_rq
->end_io_data
= rq
;
432 flush_rq
->end_io
= blk_post_flush_end_io
;
434 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
439 static inline int blk_check_end_barrier(request_queue_t
*q
, struct request
*rq
,
442 if (sectors
> rq
->nr_sectors
)
443 sectors
= rq
->nr_sectors
;
445 rq
->nr_sectors
-= sectors
;
446 return rq
->nr_sectors
;
449 static int __blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
,
450 int sectors
, int queue_locked
)
452 if (q
->ordered
!= QUEUE_ORDERED_FLUSH
)
454 if (!blk_fs_request(rq
) || !blk_barrier_rq(rq
))
456 if (blk_barrier_postflush(rq
))
459 if (!blk_check_end_barrier(q
, rq
, sectors
)) {
460 unsigned long flags
= 0;
463 spin_lock_irqsave(q
->queue_lock
, flags
);
465 blk_start_post_flush(q
, rq
);
468 spin_unlock_irqrestore(q
->queue_lock
, flags
);
475 * blk_complete_barrier_rq - complete possible barrier request
476 * @q: the request queue for the device
478 * @sectors: number of sectors to complete
481 * Used in driver end_io handling to determine whether to postpone
482 * completion of a barrier request until a post flush has been done. This
483 * is the unlocked variant, used if the caller doesn't already hold the
486 int blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
, int sectors
)
488 return __blk_complete_barrier_rq(q
, rq
, sectors
, 0);
490 EXPORT_SYMBOL(blk_complete_barrier_rq
);
493 * blk_complete_barrier_rq_locked - complete possible barrier request
494 * @q: the request queue for the device
496 * @sectors: number of sectors to complete
499 * See blk_complete_barrier_rq(). This variant must be used if the caller
500 * holds the queue lock.
502 int blk_complete_barrier_rq_locked(request_queue_t
*q
, struct request
*rq
,
505 return __blk_complete_barrier_rq(q
, rq
, sectors
, 1);
507 EXPORT_SYMBOL(blk_complete_barrier_rq_locked
);
510 * blk_queue_bounce_limit - set bounce buffer limit for queue
511 * @q: the request queue for the device
512 * @dma_addr: bus address limit
515 * Different hardware can have different requirements as to what pages
516 * it can do I/O directly to. A low level driver can call
517 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
518 * buffers for doing I/O to pages residing above @page. By default
519 * the block layer sets this to the highest numbered "low" memory page.
521 void blk_queue_bounce_limit(request_queue_t
*q
, u64 dma_addr
)
523 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
526 * set appropriate bounce gfp mask -- unfortunately we don't have a
527 * full 4GB zone, so we have to resort to low memory for any bounces.
528 * ISA has its own < 16MB zone.
530 if (bounce_pfn
< blk_max_low_pfn
) {
531 BUG_ON(dma_addr
< BLK_BOUNCE_ISA
);
532 init_emergency_isa_pool();
533 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
535 q
->bounce_gfp
= GFP_NOIO
;
537 q
->bounce_pfn
= bounce_pfn
;
540 EXPORT_SYMBOL(blk_queue_bounce_limit
);
543 * blk_queue_max_sectors - set max sectors for a request for this queue
544 * @q: the request queue for the device
545 * @max_sectors: max sectors in the usual 512b unit
548 * Enables a low level driver to set an upper limit on the size of
551 void blk_queue_max_sectors(request_queue_t
*q
, unsigned short max_sectors
)
553 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
554 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
555 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
558 q
->max_sectors
= q
->max_hw_sectors
= max_sectors
;
561 EXPORT_SYMBOL(blk_queue_max_sectors
);
564 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
565 * @q: the request queue for the device
566 * @max_segments: max number of segments
569 * Enables a low level driver to set an upper limit on the number of
570 * physical data segments in a request. This would be the largest sized
571 * scatter list the driver could handle.
573 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
577 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
580 q
->max_phys_segments
= max_segments
;
583 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
586 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
587 * @q: the request queue for the device
588 * @max_segments: max number of segments
591 * Enables a low level driver to set an upper limit on the number of
592 * hw data segments in a request. This would be the largest number of
593 * address/length pairs the host adapter can actually give as once
596 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
600 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
603 q
->max_hw_segments
= max_segments
;
606 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
609 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
610 * @q: the request queue for the device
611 * @max_size: max size of segment in bytes
614 * Enables a low level driver to set an upper limit on the size of a
617 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
619 if (max_size
< PAGE_CACHE_SIZE
) {
620 max_size
= PAGE_CACHE_SIZE
;
621 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
624 q
->max_segment_size
= max_size
;
627 EXPORT_SYMBOL(blk_queue_max_segment_size
);
630 * blk_queue_hardsect_size - set hardware sector size for the queue
631 * @q: the request queue for the device
632 * @size: the hardware sector size, in bytes
635 * This should typically be set to the lowest possible sector size
636 * that the hardware can operate on (possible without reverting to
637 * even internal read-modify-write operations). Usually the default
638 * of 512 covers most hardware.
640 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
642 q
->hardsect_size
= size
;
645 EXPORT_SYMBOL(blk_queue_hardsect_size
);
648 * Returns the minimum that is _not_ zero, unless both are zero.
650 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
653 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
654 * @t: the stacking driver (top)
655 * @b: the underlying device (bottom)
657 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
659 /* zero is "infinity" */
660 t
->max_sectors
= t
->max_hw_sectors
=
661 min_not_zero(t
->max_sectors
,b
->max_sectors
);
663 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
664 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
665 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
666 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
669 EXPORT_SYMBOL(blk_queue_stack_limits
);
672 * blk_queue_segment_boundary - set boundary rules for segment merging
673 * @q: the request queue for the device
674 * @mask: the memory boundary mask
676 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
678 if (mask
< PAGE_CACHE_SIZE
- 1) {
679 mask
= PAGE_CACHE_SIZE
- 1;
680 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
683 q
->seg_boundary_mask
= mask
;
686 EXPORT_SYMBOL(blk_queue_segment_boundary
);
689 * blk_queue_dma_alignment - set dma length and memory alignment
690 * @q: the request queue for the device
691 * @mask: alignment mask
694 * set required memory and length aligment for direct dma transactions.
695 * this is used when buiding direct io requests for the queue.
698 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
700 q
->dma_alignment
= mask
;
703 EXPORT_SYMBOL(blk_queue_dma_alignment
);
706 * blk_queue_find_tag - find a request by its tag and queue
708 * @q: The request queue for the device
709 * @tag: The tag of the request
712 * Should be used when a device returns a tag and you want to match
715 * no locks need be held.
717 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
719 struct blk_queue_tag
*bqt
= q
->queue_tags
;
721 if (unlikely(bqt
== NULL
|| tag
>= bqt
->max_depth
))
724 return bqt
->tag_index
[tag
];
727 EXPORT_SYMBOL(blk_queue_find_tag
);
730 * __blk_queue_free_tags - release tag maintenance info
731 * @q: the request queue for the device
734 * blk_cleanup_queue() will take care of calling this function, if tagging
735 * has been used. So there's no need to call this directly.
737 static void __blk_queue_free_tags(request_queue_t
*q
)
739 struct blk_queue_tag
*bqt
= q
->queue_tags
;
744 if (atomic_dec_and_test(&bqt
->refcnt
)) {
746 BUG_ON(!list_empty(&bqt
->busy_list
));
748 kfree(bqt
->tag_index
);
749 bqt
->tag_index
= NULL
;
757 q
->queue_tags
= NULL
;
758 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
762 * blk_queue_free_tags - release tag maintenance info
763 * @q: the request queue for the device
766 * This is used to disabled tagged queuing to a device, yet leave
769 void blk_queue_free_tags(request_queue_t
*q
)
771 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
774 EXPORT_SYMBOL(blk_queue_free_tags
);
777 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
779 struct request
**tag_index
;
780 unsigned long *tag_map
;
783 if (depth
> q
->nr_requests
* 2) {
784 depth
= q
->nr_requests
* 2;
785 printk(KERN_ERR
"%s: adjusted depth to %d\n",
786 __FUNCTION__
, depth
);
789 tag_index
= kmalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
793 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
794 tag_map
= kmalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
798 memset(tag_index
, 0, depth
* sizeof(struct request
*));
799 memset(tag_map
, 0, nr_ulongs
* sizeof(unsigned long));
800 tags
->max_depth
= depth
;
801 tags
->tag_index
= tag_index
;
802 tags
->tag_map
= tag_map
;
811 * blk_queue_init_tags - initialize the queue tag info
812 * @q: the request queue for the device
813 * @depth: the maximum queue depth supported
814 * @tags: the tag to use
816 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
817 struct blk_queue_tag
*tags
)
821 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
823 if (!tags
&& !q
->queue_tags
) {
824 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
828 if (init_tag_map(q
, tags
, depth
))
831 INIT_LIST_HEAD(&tags
->busy_list
);
833 atomic_set(&tags
->refcnt
, 1);
834 } else if (q
->queue_tags
) {
835 if ((rc
= blk_queue_resize_tags(q
, depth
)))
837 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
840 atomic_inc(&tags
->refcnt
);
843 * assign it, all done
845 q
->queue_tags
= tags
;
846 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
853 EXPORT_SYMBOL(blk_queue_init_tags
);
856 * blk_queue_resize_tags - change the queueing depth
857 * @q: the request queue for the device
858 * @new_depth: the new max command queueing depth
861 * Must be called with the queue lock held.
863 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
865 struct blk_queue_tag
*bqt
= q
->queue_tags
;
866 struct request
**tag_index
;
867 unsigned long *tag_map
;
868 int max_depth
, nr_ulongs
;
874 * save the old state info, so we can copy it back
876 tag_index
= bqt
->tag_index
;
877 tag_map
= bqt
->tag_map
;
878 max_depth
= bqt
->max_depth
;
880 if (init_tag_map(q
, bqt
, new_depth
))
883 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
884 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
885 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
892 EXPORT_SYMBOL(blk_queue_resize_tags
);
895 * blk_queue_end_tag - end tag operations for a request
896 * @q: the request queue for the device
897 * @rq: the request that has completed
900 * Typically called when end_that_request_first() returns 0, meaning
901 * all transfers have been done for a request. It's important to call
902 * this function before end_that_request_last(), as that will put the
903 * request back on the free list thus corrupting the internal tag list.
906 * queue lock must be held.
908 void blk_queue_end_tag(request_queue_t
*q
, struct request
*rq
)
910 struct blk_queue_tag
*bqt
= q
->queue_tags
;
915 if (unlikely(tag
>= bqt
->max_depth
))
917 * This can happen after tag depth has been reduced.
918 * FIXME: how about a warning or info message here?
922 if (unlikely(!__test_and_clear_bit(tag
, bqt
->tag_map
))) {
923 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
928 list_del_init(&rq
->queuelist
);
929 rq
->flags
&= ~REQ_QUEUED
;
932 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
933 printk(KERN_ERR
"%s: tag %d is missing\n",
936 bqt
->tag_index
[tag
] = NULL
;
940 EXPORT_SYMBOL(blk_queue_end_tag
);
943 * blk_queue_start_tag - find a free tag and assign it
944 * @q: the request queue for the device
945 * @rq: the block request that needs tagging
948 * This can either be used as a stand-alone helper, or possibly be
949 * assigned as the queue &prep_rq_fn (in which case &struct request
950 * automagically gets a tag assigned). Note that this function
951 * assumes that any type of request can be queued! if this is not
952 * true for your device, you must check the request type before
953 * calling this function. The request will also be removed from
954 * the request queue, so it's the drivers responsibility to readd
955 * it if it should need to be restarted for some reason.
958 * queue lock must be held.
960 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
962 struct blk_queue_tag
*bqt
= q
->queue_tags
;
965 if (unlikely((rq
->flags
& REQ_QUEUED
))) {
967 "%s: request %p for device [%s] already tagged %d",
969 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
973 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
974 if (tag
>= bqt
->max_depth
)
977 __set_bit(tag
, bqt
->tag_map
);
979 rq
->flags
|= REQ_QUEUED
;
981 bqt
->tag_index
[tag
] = rq
;
982 blkdev_dequeue_request(rq
);
983 list_add(&rq
->queuelist
, &bqt
->busy_list
);
988 EXPORT_SYMBOL(blk_queue_start_tag
);
991 * blk_queue_invalidate_tags - invalidate all pending tags
992 * @q: the request queue for the device
995 * Hardware conditions may dictate a need to stop all pending requests.
996 * In this case, we will safely clear the block side of the tag queue and
997 * readd all requests to the request queue in the right order.
1000 * queue lock must be held.
1002 void blk_queue_invalidate_tags(request_queue_t
*q
)
1004 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1005 struct list_head
*tmp
, *n
;
1008 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1009 rq
= list_entry_rq(tmp
);
1011 if (rq
->tag
== -1) {
1013 "%s: bad tag found on list\n", __FUNCTION__
);
1014 list_del_init(&rq
->queuelist
);
1015 rq
->flags
&= ~REQ_QUEUED
;
1017 blk_queue_end_tag(q
, rq
);
1019 rq
->flags
&= ~REQ_STARTED
;
1020 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1024 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1026 static char *rq_flags
[] = {
1044 "REQ_DRIVE_TASKFILE",
1051 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1055 printk("%s: dev %s: flags = ", msg
,
1056 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?");
1059 if (rq
->flags
& (1 << bit
))
1060 printk("%s ", rq_flags
[bit
]);
1062 } while (bit
< __REQ_NR_BITS
);
1064 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1066 rq
->current_nr_sectors
);
1067 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1069 if (rq
->flags
& (REQ_BLOCK_PC
| REQ_PC
)) {
1071 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1072 printk("%02x ", rq
->cmd
[bit
]);
1077 EXPORT_SYMBOL(blk_dump_rq_flags
);
1079 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1081 struct bio_vec
*bv
, *bvprv
= NULL
;
1082 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1083 int high
, highprv
= 1;
1085 if (unlikely(!bio
->bi_io_vec
))
1088 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1089 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1090 bio_for_each_segment(bv
, bio
, i
) {
1092 * the trick here is making sure that a high page is never
1093 * considered part of another segment, since that might
1094 * change with the bounce page.
1096 high
= page_to_pfn(bv
->bv_page
) >= q
->bounce_pfn
;
1097 if (high
|| highprv
)
1098 goto new_hw_segment
;
1100 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1102 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1104 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1106 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1107 goto new_hw_segment
;
1109 seg_size
+= bv
->bv_len
;
1110 hw_seg_size
+= bv
->bv_len
;
1115 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1116 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1117 hw_seg_size
+= bv
->bv_len
;
1120 if (hw_seg_size
> bio
->bi_hw_front_size
)
1121 bio
->bi_hw_front_size
= hw_seg_size
;
1122 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1128 seg_size
= bv
->bv_len
;
1131 if (hw_seg_size
> bio
->bi_hw_back_size
)
1132 bio
->bi_hw_back_size
= hw_seg_size
;
1133 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1134 bio
->bi_hw_front_size
= hw_seg_size
;
1135 bio
->bi_phys_segments
= nr_phys_segs
;
1136 bio
->bi_hw_segments
= nr_hw_segs
;
1137 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1141 static int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1144 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1147 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1149 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1153 * bio and nxt are contigous in memory, check if the queue allows
1154 * these two to be merged into one
1156 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1162 static int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1165 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1166 blk_recount_segments(q
, bio
);
1167 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1168 blk_recount_segments(q
, nxt
);
1169 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1170 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1172 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1179 * map a request to scatterlist, return number of sg entries setup. Caller
1180 * must make sure sg can hold rq->nr_phys_segments entries
1182 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1184 struct bio_vec
*bvec
, *bvprv
;
1186 int nsegs
, i
, cluster
;
1189 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1192 * for each bio in rq
1195 rq_for_each_bio(bio
, rq
) {
1197 * for each segment in bio
1199 bio_for_each_segment(bvec
, bio
, i
) {
1200 int nbytes
= bvec
->bv_len
;
1202 if (bvprv
&& cluster
) {
1203 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1206 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1208 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1211 sg
[nsegs
- 1].length
+= nbytes
;
1214 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1215 sg
[nsegs
].page
= bvec
->bv_page
;
1216 sg
[nsegs
].length
= nbytes
;
1217 sg
[nsegs
].offset
= bvec
->bv_offset
;
1222 } /* segments in bio */
1228 EXPORT_SYMBOL(blk_rq_map_sg
);
1231 * the standard queue merge functions, can be overridden with device
1232 * specific ones if so desired
1235 static inline int ll_new_mergeable(request_queue_t
*q
,
1236 struct request
*req
,
1239 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1241 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1242 req
->flags
|= REQ_NOMERGE
;
1243 if (req
== q
->last_merge
)
1244 q
->last_merge
= NULL
;
1249 * A hw segment is just getting larger, bump just the phys
1252 req
->nr_phys_segments
+= nr_phys_segs
;
1256 static inline int ll_new_hw_segment(request_queue_t
*q
,
1257 struct request
*req
,
1260 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1261 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1263 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1264 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1265 req
->flags
|= REQ_NOMERGE
;
1266 if (req
== q
->last_merge
)
1267 q
->last_merge
= NULL
;
1272 * This will form the start of a new hw segment. Bump both
1275 req
->nr_hw_segments
+= nr_hw_segs
;
1276 req
->nr_phys_segments
+= nr_phys_segs
;
1280 static int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
,
1285 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1286 req
->flags
|= REQ_NOMERGE
;
1287 if (req
== q
->last_merge
)
1288 q
->last_merge
= NULL
;
1291 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1292 blk_recount_segments(q
, req
->biotail
);
1293 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1294 blk_recount_segments(q
, bio
);
1295 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1296 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1297 !BIOVEC_VIRT_OVERSIZE(len
)) {
1298 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1301 if (req
->nr_hw_segments
== 1)
1302 req
->bio
->bi_hw_front_size
= len
;
1303 if (bio
->bi_hw_segments
== 1)
1304 bio
->bi_hw_back_size
= len
;
1309 return ll_new_hw_segment(q
, req
, bio
);
1312 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1317 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1318 req
->flags
|= REQ_NOMERGE
;
1319 if (req
== q
->last_merge
)
1320 q
->last_merge
= NULL
;
1323 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1324 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1325 blk_recount_segments(q
, bio
);
1326 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1327 blk_recount_segments(q
, req
->bio
);
1328 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1329 !BIOVEC_VIRT_OVERSIZE(len
)) {
1330 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1333 if (bio
->bi_hw_segments
== 1)
1334 bio
->bi_hw_front_size
= len
;
1335 if (req
->nr_hw_segments
== 1)
1336 req
->biotail
->bi_hw_back_size
= len
;
1341 return ll_new_hw_segment(q
, req
, bio
);
1344 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1345 struct request
*next
)
1347 int total_phys_segments
;
1348 int total_hw_segments
;
1351 * First check if the either of the requests are re-queued
1352 * requests. Can't merge them if they are.
1354 if (req
->special
|| next
->special
)
1358 * Will it become too large?
1360 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1363 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1364 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1365 total_phys_segments
--;
1367 if (total_phys_segments
> q
->max_phys_segments
)
1370 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1371 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1372 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1374 * propagate the combined length to the end of the requests
1376 if (req
->nr_hw_segments
== 1)
1377 req
->bio
->bi_hw_front_size
= len
;
1378 if (next
->nr_hw_segments
== 1)
1379 next
->biotail
->bi_hw_back_size
= len
;
1380 total_hw_segments
--;
1383 if (total_hw_segments
> q
->max_hw_segments
)
1386 /* Merge is OK... */
1387 req
->nr_phys_segments
= total_phys_segments
;
1388 req
->nr_hw_segments
= total_hw_segments
;
1393 * "plug" the device if there are no outstanding requests: this will
1394 * force the transfer to start only after we have put all the requests
1397 * This is called with interrupts off and no requests on the queue and
1398 * with the queue lock held.
1400 void blk_plug_device(request_queue_t
*q
)
1402 WARN_ON(!irqs_disabled());
1405 * don't plug a stopped queue, it must be paired with blk_start_queue()
1406 * which will restart the queueing
1408 if (test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
))
1411 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1412 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1415 EXPORT_SYMBOL(blk_plug_device
);
1418 * remove the queue from the plugged list, if present. called with
1419 * queue lock held and interrupts disabled.
1421 int blk_remove_plug(request_queue_t
*q
)
1423 WARN_ON(!irqs_disabled());
1425 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1428 del_timer(&q
->unplug_timer
);
1432 EXPORT_SYMBOL(blk_remove_plug
);
1435 * remove the plug and let it rip..
1437 void __generic_unplug_device(request_queue_t
*q
)
1439 if (unlikely(test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
)))
1442 if (!blk_remove_plug(q
))
1446 * was plugged, fire request_fn if queue has stuff to do
1448 if (elv_next_request(q
))
1451 EXPORT_SYMBOL(__generic_unplug_device
);
1454 * generic_unplug_device - fire a request queue
1455 * @q: The &request_queue_t in question
1458 * Linux uses plugging to build bigger requests queues before letting
1459 * the device have at them. If a queue is plugged, the I/O scheduler
1460 * is still adding and merging requests on the queue. Once the queue
1461 * gets unplugged, the request_fn defined for the queue is invoked and
1462 * transfers started.
1464 void generic_unplug_device(request_queue_t
*q
)
1466 spin_lock_irq(q
->queue_lock
);
1467 __generic_unplug_device(q
);
1468 spin_unlock_irq(q
->queue_lock
);
1470 EXPORT_SYMBOL(generic_unplug_device
);
1472 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1475 request_queue_t
*q
= bdi
->unplug_io_data
;
1478 * devices don't necessarily have an ->unplug_fn defined
1484 static void blk_unplug_work(void *data
)
1486 request_queue_t
*q
= data
;
1491 static void blk_unplug_timeout(unsigned long data
)
1493 request_queue_t
*q
= (request_queue_t
*)data
;
1495 kblockd_schedule_work(&q
->unplug_work
);
1499 * blk_start_queue - restart a previously stopped queue
1500 * @q: The &request_queue_t in question
1503 * blk_start_queue() will clear the stop flag on the queue, and call
1504 * the request_fn for the queue if it was in a stopped state when
1505 * entered. Also see blk_stop_queue(). Queue lock must be held.
1507 void blk_start_queue(request_queue_t
*q
)
1509 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1512 * one level of recursion is ok and is much faster than kicking
1513 * the unplug handling
1515 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1517 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1520 kblockd_schedule_work(&q
->unplug_work
);
1524 EXPORT_SYMBOL(blk_start_queue
);
1527 * blk_stop_queue - stop a queue
1528 * @q: The &request_queue_t in question
1531 * The Linux block layer assumes that a block driver will consume all
1532 * entries on the request queue when the request_fn strategy is called.
1533 * Often this will not happen, because of hardware limitations (queue
1534 * depth settings). If a device driver gets a 'queue full' response,
1535 * or if it simply chooses not to queue more I/O at one point, it can
1536 * call this function to prevent the request_fn from being called until
1537 * the driver has signalled it's ready to go again. This happens by calling
1538 * blk_start_queue() to restart queue operations. Queue lock must be held.
1540 void blk_stop_queue(request_queue_t
*q
)
1543 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1545 EXPORT_SYMBOL(blk_stop_queue
);
1548 * blk_sync_queue - cancel any pending callbacks on a queue
1552 * The block layer may perform asynchronous callback activity
1553 * on a queue, such as calling the unplug function after a timeout.
1554 * A block device may call blk_sync_queue to ensure that any
1555 * such activity is cancelled, thus allowing it to release resources
1556 * the the callbacks might use. The caller must already have made sure
1557 * that its ->make_request_fn will not re-add plugging prior to calling
1561 void blk_sync_queue(struct request_queue
*q
)
1563 del_timer_sync(&q
->unplug_timer
);
1566 EXPORT_SYMBOL(blk_sync_queue
);
1569 * blk_run_queue - run a single device queue
1570 * @q: The queue to run
1572 void blk_run_queue(struct request_queue
*q
)
1574 unsigned long flags
;
1576 spin_lock_irqsave(q
->queue_lock
, flags
);
1578 if (!elv_queue_empty(q
))
1580 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1582 EXPORT_SYMBOL(blk_run_queue
);
1585 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1586 * @q: the request queue to be released
1589 * blk_cleanup_queue is the pair to blk_init_queue() or
1590 * blk_queue_make_request(). It should be called when a request queue is
1591 * being released; typically when a block device is being de-registered.
1592 * Currently, its primary task it to free all the &struct request
1593 * structures that were allocated to the queue and the queue itself.
1596 * Hopefully the low level driver will have finished any
1597 * outstanding requests first...
1599 void blk_cleanup_queue(request_queue_t
* q
)
1601 struct request_list
*rl
= &q
->rq
;
1603 if (!atomic_dec_and_test(&q
->refcnt
))
1607 elevator_exit(q
->elevator
);
1612 mempool_destroy(rl
->rq_pool
);
1615 __blk_queue_free_tags(q
);
1617 blk_queue_ordered(q
, QUEUE_ORDERED_NONE
);
1619 kmem_cache_free(requestq_cachep
, q
);
1622 EXPORT_SYMBOL(blk_cleanup_queue
);
1624 static int blk_init_free_list(request_queue_t
*q
)
1626 struct request_list
*rl
= &q
->rq
;
1628 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1629 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1630 init_waitqueue_head(&rl
->wait
[READ
]);
1631 init_waitqueue_head(&rl
->wait
[WRITE
]);
1632 init_waitqueue_head(&rl
->drain
);
1634 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1635 mempool_free_slab
, request_cachep
, q
->node
);
1643 static int __make_request(request_queue_t
*, struct bio
*);
1645 request_queue_t
*blk_alloc_queue(int gfp_mask
)
1647 return blk_alloc_queue_node(gfp_mask
, -1);
1649 EXPORT_SYMBOL(blk_alloc_queue
);
1651 request_queue_t
*blk_alloc_queue_node(int gfp_mask
, int node_id
)
1655 q
= kmem_cache_alloc_node(requestq_cachep
, gfp_mask
, node_id
);
1659 memset(q
, 0, sizeof(*q
));
1660 init_timer(&q
->unplug_timer
);
1661 atomic_set(&q
->refcnt
, 1);
1663 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1664 q
->backing_dev_info
.unplug_io_data
= q
;
1668 EXPORT_SYMBOL(blk_alloc_queue_node
);
1671 * blk_init_queue - prepare a request queue for use with a block device
1672 * @rfn: The function to be called to process requests that have been
1673 * placed on the queue.
1674 * @lock: Request queue spin lock
1677 * If a block device wishes to use the standard request handling procedures,
1678 * which sorts requests and coalesces adjacent requests, then it must
1679 * call blk_init_queue(). The function @rfn will be called when there
1680 * are requests on the queue that need to be processed. If the device
1681 * supports plugging, then @rfn may not be called immediately when requests
1682 * are available on the queue, but may be called at some time later instead.
1683 * Plugged queues are generally unplugged when a buffer belonging to one
1684 * of the requests on the queue is needed, or due to memory pressure.
1686 * @rfn is not required, or even expected, to remove all requests off the
1687 * queue, but only as many as it can handle at a time. If it does leave
1688 * requests on the queue, it is responsible for arranging that the requests
1689 * get dealt with eventually.
1691 * The queue spin lock must be held while manipulating the requests on the
1694 * Function returns a pointer to the initialized request queue, or NULL if
1695 * it didn't succeed.
1698 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1699 * when the block device is deactivated (such as at module unload).
1702 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1704 return blk_init_queue_node(rfn
, lock
, -1);
1706 EXPORT_SYMBOL(blk_init_queue
);
1709 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1711 request_queue_t
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1717 if (blk_init_free_list(q
))
1721 * if caller didn't supply a lock, they get per-queue locking with
1725 spin_lock_init(&q
->__queue_lock
);
1726 lock
= &q
->__queue_lock
;
1729 q
->request_fn
= rfn
;
1730 q
->back_merge_fn
= ll_back_merge_fn
;
1731 q
->front_merge_fn
= ll_front_merge_fn
;
1732 q
->merge_requests_fn
= ll_merge_requests_fn
;
1733 q
->prep_rq_fn
= NULL
;
1734 q
->unplug_fn
= generic_unplug_device
;
1735 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1736 q
->queue_lock
= lock
;
1738 blk_queue_segment_boundary(q
, 0xffffffff);
1740 blk_queue_make_request(q
, __make_request
);
1741 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1743 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1744 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1749 if (!elevator_init(q
, NULL
)) {
1750 blk_queue_congestion_threshold(q
);
1754 blk_cleanup_queue(q
);
1756 kmem_cache_free(requestq_cachep
, q
);
1759 EXPORT_SYMBOL(blk_init_queue_node
);
1761 int blk_get_queue(request_queue_t
*q
)
1763 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1764 atomic_inc(&q
->refcnt
);
1771 EXPORT_SYMBOL(blk_get_queue
);
1773 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
1775 elv_put_request(q
, rq
);
1776 mempool_free(rq
, q
->rq
.rq_pool
);
1779 static inline struct request
*blk_alloc_request(request_queue_t
*q
, int rw
,
1782 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1788 * first three bits are identical in rq->flags and bio->bi_rw,
1789 * see bio.h and blkdev.h
1793 if (!elv_set_request(q
, rq
, gfp_mask
))
1796 mempool_free(rq
, q
->rq
.rq_pool
);
1801 * ioc_batching returns true if the ioc is a valid batching request and
1802 * should be given priority access to a request.
1804 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
1810 * Make sure the process is able to allocate at least 1 request
1811 * even if the batch times out, otherwise we could theoretically
1814 return ioc
->nr_batch_requests
== q
->nr_batching
||
1815 (ioc
->nr_batch_requests
> 0
1816 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1820 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1821 * will cause the process to be a "batcher" on all queues in the system. This
1822 * is the behaviour we want though - once it gets a wakeup it should be given
1825 static void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
1827 if (!ioc
|| ioc_batching(q
, ioc
))
1830 ioc
->nr_batch_requests
= q
->nr_batching
;
1831 ioc
->last_waited
= jiffies
;
1834 static void __freed_request(request_queue_t
*q
, int rw
)
1836 struct request_list
*rl
= &q
->rq
;
1838 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
1839 clear_queue_congested(q
, rw
);
1841 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
1842 if (waitqueue_active(&rl
->wait
[rw
]))
1843 wake_up(&rl
->wait
[rw
]);
1845 blk_clear_queue_full(q
, rw
);
1850 * A request has just been released. Account for it, update the full and
1851 * congestion status, wake up any waiters. Called under q->queue_lock.
1853 static void freed_request(request_queue_t
*q
, int rw
)
1855 struct request_list
*rl
= &q
->rq
;
1859 __freed_request(q
, rw
);
1861 if (unlikely(rl
->starved
[rw
^ 1]))
1862 __freed_request(q
, rw
^ 1);
1864 if (!rl
->count
[READ
] && !rl
->count
[WRITE
]) {
1866 if (unlikely(waitqueue_active(&rl
->drain
)))
1867 wake_up(&rl
->drain
);
1871 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1873 * Get a free request, queue_lock must not be held
1875 static struct request
*get_request(request_queue_t
*q
, int rw
, int gfp_mask
)
1877 struct request
*rq
= NULL
;
1878 struct request_list
*rl
= &q
->rq
;
1879 struct io_context
*ioc
= get_io_context(gfp_mask
);
1881 if (unlikely(test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
)))
1884 spin_lock_irq(q
->queue_lock
);
1885 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
1887 * The queue will fill after this allocation, so set it as
1888 * full, and mark this process as "batching". This process
1889 * will be allowed to complete a batch of requests, others
1892 if (!blk_queue_full(q
, rw
)) {
1893 ioc_set_batching(q
, ioc
);
1894 blk_set_queue_full(q
, rw
);
1898 switch (elv_may_queue(q
, rw
)) {
1901 case ELV_MQUEUE_MAY
:
1903 case ELV_MQUEUE_MUST
:
1907 if (blk_queue_full(q
, rw
) && !ioc_batching(q
, ioc
)) {
1909 * The queue is full and the allocating process is not a
1910 * "batcher", and not exempted by the IO scheduler
1912 spin_unlock_irq(q
->queue_lock
);
1918 rl
->starved
[rw
] = 0;
1919 if (rl
->count
[rw
] >= queue_congestion_on_threshold(q
))
1920 set_queue_congested(q
, rw
);
1921 spin_unlock_irq(q
->queue_lock
);
1923 rq
= blk_alloc_request(q
, rw
, gfp_mask
);
1926 * Allocation failed presumably due to memory. Undo anything
1927 * we might have messed up.
1929 * Allocating task should really be put onto the front of the
1930 * wait queue, but this is pretty rare.
1932 spin_lock_irq(q
->queue_lock
);
1933 freed_request(q
, rw
);
1936 * in the very unlikely event that allocation failed and no
1937 * requests for this direction was pending, mark us starved
1938 * so that freeing of a request in the other direction will
1939 * notice us. another possible fix would be to split the
1940 * rq mempool into READ and WRITE
1943 if (unlikely(rl
->count
[rw
] == 0))
1944 rl
->starved
[rw
] = 1;
1946 spin_unlock_irq(q
->queue_lock
);
1950 if (ioc_batching(q
, ioc
))
1951 ioc
->nr_batch_requests
--;
1956 put_io_context(ioc
);
1961 * No available requests for this queue, unplug the device and wait for some
1962 * requests to become available.
1964 static struct request
*get_request_wait(request_queue_t
*q
, int rw
)
1970 struct request_list
*rl
= &q
->rq
;
1972 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
1973 TASK_UNINTERRUPTIBLE
);
1975 rq
= get_request(q
, rw
, GFP_NOIO
);
1978 struct io_context
*ioc
;
1980 generic_unplug_device(q
);
1984 * After sleeping, we become a "batching" process and
1985 * will be able to allocate at least one request, and
1986 * up to a big batch of them for a small period time.
1987 * See ioc_batching, ioc_set_batching
1989 ioc
= get_io_context(GFP_NOIO
);
1990 ioc_set_batching(q
, ioc
);
1991 put_io_context(ioc
);
1993 finish_wait(&rl
->wait
[rw
], &wait
);
1999 struct request
*blk_get_request(request_queue_t
*q
, int rw
, int gfp_mask
)
2003 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2005 if (gfp_mask
& __GFP_WAIT
)
2006 rq
= get_request_wait(q
, rw
);
2008 rq
= get_request(q
, rw
, gfp_mask
);
2013 EXPORT_SYMBOL(blk_get_request
);
2016 * blk_requeue_request - put a request back on queue
2017 * @q: request queue where request should be inserted
2018 * @rq: request to be inserted
2021 * Drivers often keep queueing requests until the hardware cannot accept
2022 * more, when that condition happens we need to put the request back
2023 * on the queue. Must be called with queue lock held.
2025 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2027 if (blk_rq_tagged(rq
))
2028 blk_queue_end_tag(q
, rq
);
2030 elv_requeue_request(q
, rq
);
2033 EXPORT_SYMBOL(blk_requeue_request
);
2036 * blk_insert_request - insert a special request in to a request queue
2037 * @q: request queue where request should be inserted
2038 * @rq: request to be inserted
2039 * @at_head: insert request at head or tail of queue
2040 * @data: private data
2043 * Many block devices need to execute commands asynchronously, so they don't
2044 * block the whole kernel from preemption during request execution. This is
2045 * accomplished normally by inserting aritficial requests tagged as
2046 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2047 * scheduled for actual execution by the request queue.
2049 * We have the option of inserting the head or the tail of the queue.
2050 * Typically we use the tail for new ioctls and so forth. We use the head
2051 * of the queue for things like a QUEUE_FULL message from a device, or a
2052 * host that is unable to accept a particular command.
2054 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2055 int at_head
, void *data
)
2057 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2058 unsigned long flags
;
2061 * tell I/O scheduler that this isn't a regular read/write (ie it
2062 * must not attempt merges on this) and that it acts as a soft
2065 rq
->flags
|= REQ_SPECIAL
| REQ_SOFTBARRIER
;
2069 spin_lock_irqsave(q
->queue_lock
, flags
);
2072 * If command is tagged, release the tag
2074 if (blk_rq_tagged(rq
))
2075 blk_queue_end_tag(q
, rq
);
2077 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2078 __elv_add_request(q
, rq
, where
, 0);
2080 if (blk_queue_plugged(q
))
2081 __generic_unplug_device(q
);
2084 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2087 EXPORT_SYMBOL(blk_insert_request
);
2090 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2091 * @q: request queue where request should be inserted
2092 * @rw: READ or WRITE data
2093 * @ubuf: the user buffer
2094 * @len: length of user data
2097 * Data will be mapped directly for zero copy io, if possible. Otherwise
2098 * a kernel bounce buffer is used.
2100 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2101 * still in process context.
2103 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2104 * before being submitted to the device, as pages mapped may be out of
2105 * reach. It's the callers responsibility to make sure this happens. The
2106 * original bio must be passed back in to blk_rq_unmap_user() for proper
2109 struct request
*blk_rq_map_user(request_queue_t
*q
, int rw
, void __user
*ubuf
,
2112 unsigned long uaddr
;
2116 if (len
> (q
->max_sectors
<< 9))
2117 return ERR_PTR(-EINVAL
);
2118 if ((!len
&& ubuf
) || (len
&& !ubuf
))
2119 return ERR_PTR(-EINVAL
);
2121 rq
= blk_get_request(q
, rw
, __GFP_WAIT
);
2123 return ERR_PTR(-ENOMEM
);
2126 * if alignment requirement is satisfied, map in user pages for
2127 * direct dma. else, set up kernel bounce buffers
2129 uaddr
= (unsigned long) ubuf
;
2130 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2131 bio
= bio_map_user(q
, NULL
, uaddr
, len
, rw
== READ
);
2133 bio
= bio_copy_user(q
, uaddr
, len
, rw
== READ
);
2136 rq
->bio
= rq
->biotail
= bio
;
2137 blk_rq_bio_prep(q
, rq
, bio
);
2139 rq
->buffer
= rq
->data
= NULL
;
2145 * bio is the err-ptr
2147 blk_put_request(rq
);
2148 return (struct request
*) bio
;
2151 EXPORT_SYMBOL(blk_rq_map_user
);
2154 * blk_rq_unmap_user - unmap a request with user data
2155 * @rq: request to be unmapped
2156 * @bio: bio for the request
2157 * @ulen: length of user buffer
2160 * Unmap a request previously mapped by blk_rq_map_user().
2162 int blk_rq_unmap_user(struct request
*rq
, struct bio
*bio
, unsigned int ulen
)
2167 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2168 bio_unmap_user(bio
);
2170 ret
= bio_uncopy_user(bio
);
2173 blk_put_request(rq
);
2177 EXPORT_SYMBOL(blk_rq_unmap_user
);
2180 * blk_execute_rq - insert a request into queue for execution
2181 * @q: queue to insert the request in
2182 * @bd_disk: matching gendisk
2183 * @rq: request to insert
2186 * Insert a fully prepared request at the back of the io scheduler queue
2189 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2192 DECLARE_COMPLETION(wait
);
2193 char sense
[SCSI_SENSE_BUFFERSIZE
];
2196 rq
->rq_disk
= bd_disk
;
2199 * we need an extra reference to the request, so we can look at
2200 * it after io completion
2205 memset(sense
, 0, sizeof(sense
));
2210 rq
->flags
|= REQ_NOMERGE
;
2211 rq
->waiting
= &wait
;
2212 rq
->end_io
= blk_end_sync_rq
;
2213 elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 1);
2214 generic_unplug_device(q
);
2215 wait_for_completion(&wait
);
2224 EXPORT_SYMBOL(blk_execute_rq
);
2227 * blkdev_issue_flush - queue a flush
2228 * @bdev: blockdev to issue flush for
2229 * @error_sector: error sector
2232 * Issue a flush for the block device in question. Caller can supply
2233 * room for storing the error offset in case of a flush error, if they
2234 * wish to. Caller must run wait_for_completion() on its own.
2236 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2240 if (bdev
->bd_disk
== NULL
)
2243 q
= bdev_get_queue(bdev
);
2246 if (!q
->issue_flush_fn
)
2249 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2252 EXPORT_SYMBOL(blkdev_issue_flush
);
2254 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2256 int rw
= rq_data_dir(rq
);
2258 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2262 __disk_stat_add(rq
->rq_disk
, read_sectors
, nr_sectors
);
2264 __disk_stat_inc(rq
->rq_disk
, read_merges
);
2265 } else if (rw
== WRITE
) {
2266 __disk_stat_add(rq
->rq_disk
, write_sectors
, nr_sectors
);
2268 __disk_stat_inc(rq
->rq_disk
, write_merges
);
2271 disk_round_stats(rq
->rq_disk
);
2272 rq
->rq_disk
->in_flight
++;
2277 * add-request adds a request to the linked list.
2278 * queue lock is held and interrupts disabled, as we muck with the
2279 * request queue list.
2281 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2283 drive_stat_acct(req
, req
->nr_sectors
, 1);
2286 q
->activity_fn(q
->activity_data
, rq_data_dir(req
));
2289 * elevator indicated where it wants this request to be
2290 * inserted at elevator_merge time
2292 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2296 * disk_round_stats() - Round off the performance stats on a struct
2299 * The average IO queue length and utilisation statistics are maintained
2300 * by observing the current state of the queue length and the amount of
2301 * time it has been in this state for.
2303 * Normally, that accounting is done on IO completion, but that can result
2304 * in more than a second's worth of IO being accounted for within any one
2305 * second, leading to >100% utilisation. To deal with that, we call this
2306 * function to do a round-off before returning the results when reading
2307 * /proc/diskstats. This accounts immediately for all queue usage up to
2308 * the current jiffies and restarts the counters again.
2310 void disk_round_stats(struct gendisk
*disk
)
2312 unsigned long now
= jiffies
;
2314 __disk_stat_add(disk
, time_in_queue
,
2315 disk
->in_flight
* (now
- disk
->stamp
));
2318 if (disk
->in_flight
)
2319 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp_idle
));
2320 disk
->stamp_idle
= now
;
2324 * queue lock must be held
2326 static void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2328 struct request_list
*rl
= req
->rl
;
2332 if (unlikely(--req
->ref_count
))
2335 req
->rq_status
= RQ_INACTIVE
;
2340 * Request may not have originated from ll_rw_blk. if not,
2341 * it didn't come out of our reserved rq pools
2344 int rw
= rq_data_dir(req
);
2346 elv_completed_request(q
, req
);
2348 BUG_ON(!list_empty(&req
->queuelist
));
2350 blk_free_request(q
, req
);
2351 freed_request(q
, rw
);
2355 void blk_put_request(struct request
*req
)
2358 * if req->rl isn't set, this request didnt originate from the
2359 * block layer, so it's safe to just disregard it
2362 unsigned long flags
;
2363 request_queue_t
*q
= req
->q
;
2365 spin_lock_irqsave(q
->queue_lock
, flags
);
2366 __blk_put_request(q
, req
);
2367 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2371 EXPORT_SYMBOL(blk_put_request
);
2374 * blk_end_sync_rq - executes a completion event on a request
2375 * @rq: request to complete
2377 void blk_end_sync_rq(struct request
*rq
)
2379 struct completion
*waiting
= rq
->waiting
;
2382 __blk_put_request(rq
->q
, rq
);
2385 * complete last, if this is a stack request the process (and thus
2386 * the rq pointer) could be invalid right after this complete()
2390 EXPORT_SYMBOL(blk_end_sync_rq
);
2393 * blk_congestion_wait - wait for a queue to become uncongested
2394 * @rw: READ or WRITE
2395 * @timeout: timeout in jiffies
2397 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2398 * If no queues are congested then just wait for the next request to be
2401 long blk_congestion_wait(int rw
, long timeout
)
2405 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
2407 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
2408 ret
= io_schedule_timeout(timeout
);
2409 finish_wait(wqh
, &wait
);
2413 EXPORT_SYMBOL(blk_congestion_wait
);
2416 * Has to be called with the request spinlock acquired
2418 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2419 struct request
*next
)
2421 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2427 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2430 if (rq_data_dir(req
) != rq_data_dir(next
)
2431 || req
->rq_disk
!= next
->rq_disk
2432 || next
->waiting
|| next
->special
)
2436 * If we are allowed to merge, then append bio list
2437 * from next to rq and release next. merge_requests_fn
2438 * will have updated segment counts, update sector
2441 if (!q
->merge_requests_fn(q
, req
, next
))
2445 * At this point we have either done a back merge
2446 * or front merge. We need the smaller start_time of
2447 * the merged requests to be the current request
2448 * for accounting purposes.
2450 if (time_after(req
->start_time
, next
->start_time
))
2451 req
->start_time
= next
->start_time
;
2453 req
->biotail
->bi_next
= next
->bio
;
2454 req
->biotail
= next
->biotail
;
2456 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2458 elv_merge_requests(q
, req
, next
);
2461 disk_round_stats(req
->rq_disk
);
2462 req
->rq_disk
->in_flight
--;
2465 __blk_put_request(q
, next
);
2469 static inline int attempt_back_merge(request_queue_t
*q
, struct request
*rq
)
2471 struct request
*next
= elv_latter_request(q
, rq
);
2474 return attempt_merge(q
, rq
, next
);
2479 static inline int attempt_front_merge(request_queue_t
*q
, struct request
*rq
)
2481 struct request
*prev
= elv_former_request(q
, rq
);
2484 return attempt_merge(q
, prev
, rq
);
2490 * blk_attempt_remerge - attempt to remerge active head with next request
2491 * @q: The &request_queue_t belonging to the device
2492 * @rq: The head request (usually)
2495 * For head-active devices, the queue can easily be unplugged so quickly
2496 * that proper merging is not done on the front request. This may hurt
2497 * performance greatly for some devices. The block layer cannot safely
2498 * do merging on that first request for these queues, but the driver can
2499 * call this function and make it happen any way. Only the driver knows
2500 * when it is safe to do so.
2502 void blk_attempt_remerge(request_queue_t
*q
, struct request
*rq
)
2504 unsigned long flags
;
2506 spin_lock_irqsave(q
->queue_lock
, flags
);
2507 attempt_back_merge(q
, rq
);
2508 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2511 EXPORT_SYMBOL(blk_attempt_remerge
);
2513 static int __make_request(request_queue_t
*q
, struct bio
*bio
)
2515 struct request
*req
, *freereq
= NULL
;
2516 int el_ret
, rw
, nr_sectors
, cur_nr_sectors
, barrier
, err
, sync
;
2519 sector
= bio
->bi_sector
;
2520 nr_sectors
= bio_sectors(bio
);
2521 cur_nr_sectors
= bio_cur_sectors(bio
);
2523 rw
= bio_data_dir(bio
);
2524 sync
= bio_sync(bio
);
2527 * low level driver can indicate that it wants pages above a
2528 * certain limit bounced to low memory (ie for highmem, or even
2529 * ISA dma in theory)
2531 blk_queue_bounce(q
, &bio
);
2533 spin_lock_prefetch(q
->queue_lock
);
2535 barrier
= bio_barrier(bio
);
2536 if (unlikely(barrier
) && (q
->ordered
== QUEUE_ORDERED_NONE
)) {
2542 spin_lock_irq(q
->queue_lock
);
2544 if (elv_queue_empty(q
)) {
2551 el_ret
= elv_merge(q
, &req
, bio
);
2553 case ELEVATOR_BACK_MERGE
:
2554 BUG_ON(!rq_mergeable(req
));
2556 if (!q
->back_merge_fn(q
, req
, bio
))
2559 req
->biotail
->bi_next
= bio
;
2561 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2562 drive_stat_acct(req
, nr_sectors
, 0);
2563 if (!attempt_back_merge(q
, req
))
2564 elv_merged_request(q
, req
);
2567 case ELEVATOR_FRONT_MERGE
:
2568 BUG_ON(!rq_mergeable(req
));
2570 if (!q
->front_merge_fn(q
, req
, bio
))
2573 bio
->bi_next
= req
->bio
;
2577 * may not be valid. if the low level driver said
2578 * it didn't need a bounce buffer then it better
2579 * not touch req->buffer either...
2581 req
->buffer
= bio_data(bio
);
2582 req
->current_nr_sectors
= cur_nr_sectors
;
2583 req
->hard_cur_sectors
= cur_nr_sectors
;
2584 req
->sector
= req
->hard_sector
= sector
;
2585 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2586 drive_stat_acct(req
, nr_sectors
, 0);
2587 if (!attempt_front_merge(q
, req
))
2588 elv_merged_request(q
, req
);
2592 * elevator says don't/can't merge. get new request
2594 case ELEVATOR_NO_MERGE
:
2598 printk("elevator returned crap (%d)\n", el_ret
);
2603 * Grab a free request from the freelist - if that is empty, check
2604 * if we are doing read ahead and abort instead of blocking for
2612 spin_unlock_irq(q
->queue_lock
);
2613 if ((freereq
= get_request(q
, rw
, GFP_ATOMIC
)) == NULL
) {
2618 if (bio_rw_ahead(bio
))
2621 freereq
= get_request_wait(q
, rw
);
2626 req
->flags
|= REQ_CMD
;
2629 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2631 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2632 req
->flags
|= REQ_FAILFAST
;
2635 * REQ_BARRIER implies no merging, but lets make it explicit
2637 if (unlikely(barrier
))
2638 req
->flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2641 req
->hard_sector
= req
->sector
= sector
;
2642 req
->hard_nr_sectors
= req
->nr_sectors
= nr_sectors
;
2643 req
->current_nr_sectors
= req
->hard_cur_sectors
= cur_nr_sectors
;
2644 req
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2645 req
->nr_hw_segments
= bio_hw_segments(q
, bio
);
2646 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2647 req
->waiting
= NULL
;
2648 req
->bio
= req
->biotail
= bio
;
2649 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2650 req
->start_time
= jiffies
;
2652 add_request(q
, req
);
2655 __blk_put_request(q
, freereq
);
2657 __generic_unplug_device(q
);
2659 spin_unlock_irq(q
->queue_lock
);
2663 bio_endio(bio
, nr_sectors
<< 9, err
);
2668 * If bio->bi_dev is a partition, remap the location
2670 static inline void blk_partition_remap(struct bio
*bio
)
2672 struct block_device
*bdev
= bio
->bi_bdev
;
2674 if (bdev
!= bdev
->bd_contains
) {
2675 struct hd_struct
*p
= bdev
->bd_part
;
2677 switch (bio
->bi_rw
) {
2679 p
->read_sectors
+= bio_sectors(bio
);
2683 p
->write_sectors
+= bio_sectors(bio
);
2687 bio
->bi_sector
+= p
->start_sect
;
2688 bio
->bi_bdev
= bdev
->bd_contains
;
2692 void blk_finish_queue_drain(request_queue_t
*q
)
2694 struct request_list
*rl
= &q
->rq
;
2697 spin_lock_irq(q
->queue_lock
);
2698 clear_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
);
2700 while (!list_empty(&q
->drain_list
)) {
2701 rq
= list_entry_rq(q
->drain_list
.next
);
2703 list_del_init(&rq
->queuelist
);
2704 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 1);
2707 spin_unlock_irq(q
->queue_lock
);
2709 wake_up(&rl
->wait
[0]);
2710 wake_up(&rl
->wait
[1]);
2711 wake_up(&rl
->drain
);
2714 static int wait_drain(request_queue_t
*q
, struct request_list
*rl
, int dispatch
)
2716 int wait
= rl
->count
[READ
] + rl
->count
[WRITE
];
2719 wait
+= !list_empty(&q
->queue_head
);
2725 * We rely on the fact that only requests allocated through blk_alloc_request()
2726 * have io scheduler private data structures associated with them. Any other
2727 * type of request (allocated on stack or through kmalloc()) should not go
2728 * to the io scheduler core, but be attached to the queue head instead.
2730 void blk_wait_queue_drained(request_queue_t
*q
, int wait_dispatch
)
2732 struct request_list
*rl
= &q
->rq
;
2735 spin_lock_irq(q
->queue_lock
);
2736 set_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
);
2738 while (wait_drain(q
, rl
, wait_dispatch
)) {
2739 prepare_to_wait(&rl
->drain
, &wait
, TASK_UNINTERRUPTIBLE
);
2741 if (wait_drain(q
, rl
, wait_dispatch
)) {
2742 __generic_unplug_device(q
);
2743 spin_unlock_irq(q
->queue_lock
);
2745 spin_lock_irq(q
->queue_lock
);
2748 finish_wait(&rl
->drain
, &wait
);
2751 spin_unlock_irq(q
->queue_lock
);
2755 * block waiting for the io scheduler being started again.
2757 static inline void block_wait_queue_running(request_queue_t
*q
)
2761 while (unlikely(test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
))) {
2762 struct request_list
*rl
= &q
->rq
;
2764 prepare_to_wait_exclusive(&rl
->drain
, &wait
,
2765 TASK_UNINTERRUPTIBLE
);
2768 * re-check the condition. avoids using prepare_to_wait()
2769 * in the fast path (queue is running)
2771 if (test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
))
2774 finish_wait(&rl
->drain
, &wait
);
2778 static void handle_bad_sector(struct bio
*bio
)
2780 char b
[BDEVNAME_SIZE
];
2782 printk(KERN_INFO
"attempt to access beyond end of device\n");
2783 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
2784 bdevname(bio
->bi_bdev
, b
),
2786 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
2787 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
2789 set_bit(BIO_EOF
, &bio
->bi_flags
);
2793 * generic_make_request: hand a buffer to its device driver for I/O
2794 * @bio: The bio describing the location in memory and on the device.
2796 * generic_make_request() is used to make I/O requests of block
2797 * devices. It is passed a &struct bio, which describes the I/O that needs
2800 * generic_make_request() does not return any status. The
2801 * success/failure status of the request, along with notification of
2802 * completion, is delivered asynchronously through the bio->bi_end_io
2803 * function described (one day) else where.
2805 * The caller of generic_make_request must make sure that bi_io_vec
2806 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2807 * set to describe the device address, and the
2808 * bi_end_io and optionally bi_private are set to describe how
2809 * completion notification should be signaled.
2811 * generic_make_request and the drivers it calls may use bi_next if this
2812 * bio happens to be merged with someone else, and may change bi_dev and
2813 * bi_sector for remaps as it sees fit. So the values of these fields
2814 * should NOT be depended on after the call to generic_make_request.
2816 void generic_make_request(struct bio
*bio
)
2820 int ret
, nr_sectors
= bio_sectors(bio
);
2823 /* Test device or partition size, when known. */
2824 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
2826 sector_t sector
= bio
->bi_sector
;
2828 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2830 * This may well happen - the kernel calls bread()
2831 * without checking the size of the device, e.g., when
2832 * mounting a device.
2834 handle_bad_sector(bio
);
2840 * Resolve the mapping until finished. (drivers are
2841 * still free to implement/resolve their own stacking
2842 * by explicitly returning 0)
2844 * NOTE: we don't repeat the blk_size check for each new device.
2845 * Stacking drivers are expected to know what they are doing.
2848 char b
[BDEVNAME_SIZE
];
2850 q
= bdev_get_queue(bio
->bi_bdev
);
2853 "generic_make_request: Trying to access "
2854 "nonexistent block-device %s (%Lu)\n",
2855 bdevname(bio
->bi_bdev
, b
),
2856 (long long) bio
->bi_sector
);
2858 bio_endio(bio
, bio
->bi_size
, -EIO
);
2862 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
2863 printk("bio too big device %s (%u > %u)\n",
2864 bdevname(bio
->bi_bdev
, b
),
2870 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
2873 block_wait_queue_running(q
);
2876 * If this device has partitions, remap block n
2877 * of partition p to block n+start(p) of the disk.
2879 blk_partition_remap(bio
);
2881 ret
= q
->make_request_fn(q
, bio
);
2885 EXPORT_SYMBOL(generic_make_request
);
2888 * submit_bio: submit a bio to the block device layer for I/O
2889 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2890 * @bio: The &struct bio which describes the I/O
2892 * submit_bio() is very similar in purpose to generic_make_request(), and
2893 * uses that function to do most of the work. Both are fairly rough
2894 * interfaces, @bio must be presetup and ready for I/O.
2897 void submit_bio(int rw
, struct bio
*bio
)
2899 int count
= bio_sectors(bio
);
2901 BIO_BUG_ON(!bio
->bi_size
);
2902 BIO_BUG_ON(!bio
->bi_io_vec
);
2905 mod_page_state(pgpgout
, count
);
2907 mod_page_state(pgpgin
, count
);
2909 if (unlikely(block_dump
)) {
2910 char b
[BDEVNAME_SIZE
];
2911 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
2912 current
->comm
, current
->pid
,
2913 (rw
& WRITE
) ? "WRITE" : "READ",
2914 (unsigned long long)bio
->bi_sector
,
2915 bdevname(bio
->bi_bdev
,b
));
2918 generic_make_request(bio
);
2921 EXPORT_SYMBOL(submit_bio
);
2923 static void blk_recalc_rq_segments(struct request
*rq
)
2925 struct bio
*bio
, *prevbio
= NULL
;
2926 int nr_phys_segs
, nr_hw_segs
;
2927 unsigned int phys_size
, hw_size
;
2928 request_queue_t
*q
= rq
->q
;
2933 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
2934 rq_for_each_bio(bio
, rq
) {
2935 /* Force bio hw/phys segs to be recalculated. */
2936 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
2938 nr_phys_segs
+= bio_phys_segments(q
, bio
);
2939 nr_hw_segs
+= bio_hw_segments(q
, bio
);
2941 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
2942 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
2944 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
2945 pseg
<= q
->max_segment_size
) {
2947 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
2951 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
2952 hseg
<= q
->max_segment_size
) {
2954 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
2961 rq
->nr_phys_segments
= nr_phys_segs
;
2962 rq
->nr_hw_segments
= nr_hw_segs
;
2965 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
2967 if (blk_fs_request(rq
)) {
2968 rq
->hard_sector
+= nsect
;
2969 rq
->hard_nr_sectors
-= nsect
;
2972 * Move the I/O submission pointers ahead if required.
2974 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
2975 (rq
->sector
<= rq
->hard_sector
)) {
2976 rq
->sector
= rq
->hard_sector
;
2977 rq
->nr_sectors
= rq
->hard_nr_sectors
;
2978 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
2979 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
2980 rq
->buffer
= bio_data(rq
->bio
);
2984 * if total number of sectors is less than the first segment
2985 * size, something has gone terribly wrong
2987 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
2988 printk("blk: request botched\n");
2989 rq
->nr_sectors
= rq
->current_nr_sectors
;
2994 static int __end_that_request_first(struct request
*req
, int uptodate
,
2997 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3001 * extend uptodate bool to allow < 0 value to be direct io error
3004 if (end_io_error(uptodate
))
3005 error
= !uptodate
? -EIO
: uptodate
;
3008 * for a REQ_BLOCK_PC request, we want to carry any eventual
3009 * sense key with us all the way through
3011 if (!blk_pc_request(req
))
3015 if (blk_fs_request(req
) && !(req
->flags
& REQ_QUIET
))
3016 printk("end_request: I/O error, dev %s, sector %llu\n",
3017 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3018 (unsigned long long)req
->sector
);
3021 total_bytes
= bio_nbytes
= 0;
3022 while ((bio
= req
->bio
) != NULL
) {
3025 if (nr_bytes
>= bio
->bi_size
) {
3026 req
->bio
= bio
->bi_next
;
3027 nbytes
= bio
->bi_size
;
3028 bio_endio(bio
, nbytes
, error
);
3032 int idx
= bio
->bi_idx
+ next_idx
;
3034 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3035 blk_dump_rq_flags(req
, "__end_that");
3036 printk("%s: bio idx %d >= vcnt %d\n",
3038 bio
->bi_idx
, bio
->bi_vcnt
);
3042 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3043 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3046 * not a complete bvec done
3048 if (unlikely(nbytes
> nr_bytes
)) {
3049 bio_nbytes
+= nr_bytes
;
3050 total_bytes
+= nr_bytes
;
3055 * advance to the next vector
3058 bio_nbytes
+= nbytes
;
3061 total_bytes
+= nbytes
;
3064 if ((bio
= req
->bio
)) {
3066 * end more in this run, or just return 'not-done'
3068 if (unlikely(nr_bytes
<= 0))
3080 * if the request wasn't completed, update state
3083 bio_endio(bio
, bio_nbytes
, error
);
3084 bio
->bi_idx
+= next_idx
;
3085 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3086 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3089 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3090 blk_recalc_rq_segments(req
);
3095 * end_that_request_first - end I/O on a request
3096 * @req: the request being processed
3097 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3098 * @nr_sectors: number of sectors to end I/O on
3101 * Ends I/O on a number of sectors attached to @req, and sets it up
3102 * for the next range of segments (if any) in the cluster.
3105 * 0 - we are done with this request, call end_that_request_last()
3106 * 1 - still buffers pending for this request
3108 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3110 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3113 EXPORT_SYMBOL(end_that_request_first
);
3116 * end_that_request_chunk - end I/O on a request
3117 * @req: the request being processed
3118 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3119 * @nr_bytes: number of bytes to complete
3122 * Ends I/O on a number of bytes attached to @req, and sets it up
3123 * for the next range of segments (if any). Like end_that_request_first(),
3124 * but deals with bytes instead of sectors.
3127 * 0 - we are done with this request, call end_that_request_last()
3128 * 1 - still buffers pending for this request
3130 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3132 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3135 EXPORT_SYMBOL(end_that_request_chunk
);
3138 * queue lock must be held
3140 void end_that_request_last(struct request
*req
)
3142 struct gendisk
*disk
= req
->rq_disk
;
3144 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3145 laptop_io_completion();
3147 if (disk
&& blk_fs_request(req
)) {
3148 unsigned long duration
= jiffies
- req
->start_time
;
3149 switch (rq_data_dir(req
)) {
3151 __disk_stat_inc(disk
, writes
);
3152 __disk_stat_add(disk
, write_ticks
, duration
);
3155 __disk_stat_inc(disk
, reads
);
3156 __disk_stat_add(disk
, read_ticks
, duration
);
3159 disk_round_stats(disk
);
3165 __blk_put_request(req
->q
, req
);
3168 EXPORT_SYMBOL(end_that_request_last
);
3170 void end_request(struct request
*req
, int uptodate
)
3172 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3173 add_disk_randomness(req
->rq_disk
);
3174 blkdev_dequeue_request(req
);
3175 end_that_request_last(req
);
3179 EXPORT_SYMBOL(end_request
);
3181 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3183 /* first three bits are identical in rq->flags and bio->bi_rw */
3184 rq
->flags
|= (bio
->bi_rw
& 7);
3186 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3187 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3188 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3189 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3190 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3191 rq
->buffer
= bio_data(bio
);
3193 rq
->bio
= rq
->biotail
= bio
;
3196 EXPORT_SYMBOL(blk_rq_bio_prep
);
3198 int kblockd_schedule_work(struct work_struct
*work
)
3200 return queue_work(kblockd_workqueue
, work
);
3203 EXPORT_SYMBOL(kblockd_schedule_work
);
3205 void kblockd_flush(void)
3207 flush_workqueue(kblockd_workqueue
);
3209 EXPORT_SYMBOL(kblockd_flush
);
3211 int __init
blk_dev_init(void)
3213 kblockd_workqueue
= create_workqueue("kblockd");
3214 if (!kblockd_workqueue
)
3215 panic("Failed to create kblockd\n");
3217 request_cachep
= kmem_cache_create("blkdev_requests",
3218 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3220 requestq_cachep
= kmem_cache_create("blkdev_queue",
3221 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3223 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3224 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3226 blk_max_low_pfn
= max_low_pfn
;
3227 blk_max_pfn
= max_pfn
;
3233 * IO Context helper functions
3235 void put_io_context(struct io_context
*ioc
)
3240 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3242 if (atomic_dec_and_test(&ioc
->refcount
)) {
3243 if (ioc
->aic
&& ioc
->aic
->dtor
)
3244 ioc
->aic
->dtor(ioc
->aic
);
3245 if (ioc
->cic
&& ioc
->cic
->dtor
)
3246 ioc
->cic
->dtor(ioc
->cic
);
3248 kmem_cache_free(iocontext_cachep
, ioc
);
3251 EXPORT_SYMBOL(put_io_context
);
3253 /* Called by the exitting task */
3254 void exit_io_context(void)
3256 unsigned long flags
;
3257 struct io_context
*ioc
;
3259 local_irq_save(flags
);
3260 ioc
= current
->io_context
;
3261 current
->io_context
= NULL
;
3262 local_irq_restore(flags
);
3264 if (ioc
->aic
&& ioc
->aic
->exit
)
3265 ioc
->aic
->exit(ioc
->aic
);
3266 if (ioc
->cic
&& ioc
->cic
->exit
)
3267 ioc
->cic
->exit(ioc
->cic
);
3269 put_io_context(ioc
);
3273 * If the current task has no IO context then create one and initialise it.
3274 * If it does have a context, take a ref on it.
3276 * This is always called in the context of the task which submitted the I/O.
3277 * But weird things happen, so we disable local interrupts to ensure exclusive
3278 * access to *current.
3280 struct io_context
*get_io_context(int gfp_flags
)
3282 struct task_struct
*tsk
= current
;
3283 unsigned long flags
;
3284 struct io_context
*ret
;
3286 local_irq_save(flags
);
3287 ret
= tsk
->io_context
;
3291 local_irq_restore(flags
);
3293 ret
= kmem_cache_alloc(iocontext_cachep
, gfp_flags
);
3295 atomic_set(&ret
->refcount
, 1);
3296 ret
->pid
= tsk
->pid
;
3297 ret
->last_waited
= jiffies
; /* doesn't matter... */
3298 ret
->nr_batch_requests
= 0; /* because this is 0 */
3301 spin_lock_init(&ret
->lock
);
3303 local_irq_save(flags
);
3306 * very unlikely, someone raced with us in setting up the task
3307 * io context. free new context and just grab a reference.
3309 if (!tsk
->io_context
)
3310 tsk
->io_context
= ret
;
3312 kmem_cache_free(iocontext_cachep
, ret
);
3313 ret
= tsk
->io_context
;
3317 atomic_inc(&ret
->refcount
);
3318 local_irq_restore(flags
);
3323 EXPORT_SYMBOL(get_io_context
);
3325 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3327 struct io_context
*src
= *psrc
;
3328 struct io_context
*dst
= *pdst
;
3331 BUG_ON(atomic_read(&src
->refcount
) == 0);
3332 atomic_inc(&src
->refcount
);
3333 put_io_context(dst
);
3337 EXPORT_SYMBOL(copy_io_context
);
3339 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3341 struct io_context
*temp
;
3346 EXPORT_SYMBOL(swap_io_context
);
3351 struct queue_sysfs_entry
{
3352 struct attribute attr
;
3353 ssize_t (*show
)(struct request_queue
*, char *);
3354 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3358 queue_var_show(unsigned int var
, char *page
)
3360 return sprintf(page
, "%d\n", var
);
3364 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3366 char *p
= (char *) page
;
3368 *var
= simple_strtoul(p
, &p
, 10);
3372 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3374 return queue_var_show(q
->nr_requests
, (page
));
3378 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3380 struct request_list
*rl
= &q
->rq
;
3382 int ret
= queue_var_store(&q
->nr_requests
, page
, count
);
3383 if (q
->nr_requests
< BLKDEV_MIN_RQ
)
3384 q
->nr_requests
= BLKDEV_MIN_RQ
;
3385 blk_queue_congestion_threshold(q
);
3387 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3388 set_queue_congested(q
, READ
);
3389 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3390 clear_queue_congested(q
, READ
);
3392 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3393 set_queue_congested(q
, WRITE
);
3394 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3395 clear_queue_congested(q
, WRITE
);
3397 if (rl
->count
[READ
] >= q
->nr_requests
) {
3398 blk_set_queue_full(q
, READ
);
3399 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3400 blk_clear_queue_full(q
, READ
);
3401 wake_up(&rl
->wait
[READ
]);
3404 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3405 blk_set_queue_full(q
, WRITE
);
3406 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3407 blk_clear_queue_full(q
, WRITE
);
3408 wake_up(&rl
->wait
[WRITE
]);
3413 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3415 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3417 return queue_var_show(ra_kb
, (page
));
3421 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3423 unsigned long ra_kb
;
3424 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3426 spin_lock_irq(q
->queue_lock
);
3427 if (ra_kb
> (q
->max_sectors
>> 1))
3428 ra_kb
= (q
->max_sectors
>> 1);
3430 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3431 spin_unlock_irq(q
->queue_lock
);
3436 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3438 int max_sectors_kb
= q
->max_sectors
>> 1;
3440 return queue_var_show(max_sectors_kb
, (page
));
3444 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3446 unsigned long max_sectors_kb
,
3447 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3448 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3449 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3452 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3455 * Take the queue lock to update the readahead and max_sectors
3456 * values synchronously:
3458 spin_lock_irq(q
->queue_lock
);
3460 * Trim readahead window as well, if necessary:
3462 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3463 if (ra_kb
> max_sectors_kb
)
3464 q
->backing_dev_info
.ra_pages
=
3465 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3467 q
->max_sectors
= max_sectors_kb
<< 1;
3468 spin_unlock_irq(q
->queue_lock
);
3473 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3475 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3477 return queue_var_show(max_hw_sectors_kb
, (page
));
3481 static struct queue_sysfs_entry queue_requests_entry
= {
3482 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
3483 .show
= queue_requests_show
,
3484 .store
= queue_requests_store
,
3487 static struct queue_sysfs_entry queue_ra_entry
= {
3488 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
3489 .show
= queue_ra_show
,
3490 .store
= queue_ra_store
,
3493 static struct queue_sysfs_entry queue_max_sectors_entry
= {
3494 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
3495 .show
= queue_max_sectors_show
,
3496 .store
= queue_max_sectors_store
,
3499 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
3500 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
3501 .show
= queue_max_hw_sectors_show
,
3504 static struct queue_sysfs_entry queue_iosched_entry
= {
3505 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
3506 .show
= elv_iosched_show
,
3507 .store
= elv_iosched_store
,
3510 static struct attribute
*default_attrs
[] = {
3511 &queue_requests_entry
.attr
,
3512 &queue_ra_entry
.attr
,
3513 &queue_max_hw_sectors_entry
.attr
,
3514 &queue_max_sectors_entry
.attr
,
3515 &queue_iosched_entry
.attr
,
3519 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3522 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3524 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3525 struct request_queue
*q
;
3527 q
= container_of(kobj
, struct request_queue
, kobj
);
3531 return entry
->show(q
, page
);
3535 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3536 const char *page
, size_t length
)
3538 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3539 struct request_queue
*q
;
3541 q
= container_of(kobj
, struct request_queue
, kobj
);
3545 return entry
->store(q
, page
, length
);
3548 static struct sysfs_ops queue_sysfs_ops
= {
3549 .show
= queue_attr_show
,
3550 .store
= queue_attr_store
,
3553 static struct kobj_type queue_ktype
= {
3554 .sysfs_ops
= &queue_sysfs_ops
,
3555 .default_attrs
= default_attrs
,
3558 int blk_register_queue(struct gendisk
*disk
)
3562 request_queue_t
*q
= disk
->queue
;
3564 if (!q
|| !q
->request_fn
)
3567 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
3568 if (!q
->kobj
.parent
)
3571 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
3572 q
->kobj
.ktype
= &queue_ktype
;
3574 ret
= kobject_register(&q
->kobj
);
3578 ret
= elv_register_queue(q
);
3580 kobject_unregister(&q
->kobj
);
3587 void blk_unregister_queue(struct gendisk
*disk
)
3589 request_queue_t
*q
= disk
->queue
;
3591 if (q
&& q
->request_fn
) {
3592 elv_unregister_queue(q
);
3594 kobject_unregister(&q
->kobj
);
3595 kobject_put(&disk
->kobj
);