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
;
289 rq
->end_io_data
= NULL
;
293 * blk_queue_ordered - does this queue support ordered writes
294 * @q: the request queue
298 * For journalled file systems, doing ordered writes on a commit
299 * block instead of explicitly doing wait_on_buffer (which is bad
300 * for performance) can be a big win. Block drivers supporting this
301 * feature should call this function and indicate so.
304 void blk_queue_ordered(request_queue_t
*q
, int flag
)
307 case QUEUE_ORDERED_NONE
:
309 kmem_cache_free(request_cachep
, q
->flush_rq
);
313 case QUEUE_ORDERED_TAG
:
316 case QUEUE_ORDERED_FLUSH
:
319 q
->flush_rq
= kmem_cache_alloc(request_cachep
,
323 printk("blk_queue_ordered: bad value %d\n", flag
);
328 EXPORT_SYMBOL(blk_queue_ordered
);
331 * blk_queue_issue_flush_fn - set function for issuing a flush
332 * @q: the request queue
333 * @iff: the function to be called issuing the flush
336 * If a driver supports issuing a flush command, the support is notified
337 * to the block layer by defining it through this call.
340 void blk_queue_issue_flush_fn(request_queue_t
*q
, issue_flush_fn
*iff
)
342 q
->issue_flush_fn
= iff
;
345 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
348 * Cache flushing for ordered writes handling
350 static void blk_pre_flush_end_io(struct request
*flush_rq
)
352 struct request
*rq
= flush_rq
->end_io_data
;
353 request_queue_t
*q
= rq
->q
;
355 rq
->flags
|= REQ_BAR_PREFLUSH
;
357 if (!flush_rq
->errors
)
358 elv_requeue_request(q
, rq
);
360 q
->end_flush_fn(q
, flush_rq
);
361 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
366 static void blk_post_flush_end_io(struct request
*flush_rq
)
368 struct request
*rq
= flush_rq
->end_io_data
;
369 request_queue_t
*q
= rq
->q
;
371 rq
->flags
|= REQ_BAR_POSTFLUSH
;
373 q
->end_flush_fn(q
, flush_rq
);
374 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
378 struct request
*blk_start_pre_flush(request_queue_t
*q
, struct request
*rq
)
380 struct request
*flush_rq
= q
->flush_rq
;
382 BUG_ON(!blk_barrier_rq(rq
));
384 if (test_and_set_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
))
387 rq_init(q
, flush_rq
);
388 flush_rq
->elevator_private
= NULL
;
389 flush_rq
->flags
= REQ_BAR_FLUSH
;
390 flush_rq
->rq_disk
= rq
->rq_disk
;
394 * prepare_flush returns 0 if no flush is needed, just mark both
395 * pre and post flush as done in that case
397 if (!q
->prepare_flush_fn(q
, flush_rq
)) {
398 rq
->flags
|= REQ_BAR_PREFLUSH
| REQ_BAR_POSTFLUSH
;
399 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
404 * some drivers dequeue requests right away, some only after io
405 * completion. make sure the request is dequeued.
407 if (!list_empty(&rq
->queuelist
))
408 blkdev_dequeue_request(rq
);
410 elv_deactivate_request(q
, rq
);
412 flush_rq
->end_io_data
= rq
;
413 flush_rq
->end_io
= blk_pre_flush_end_io
;
415 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
419 static void blk_start_post_flush(request_queue_t
*q
, struct request
*rq
)
421 struct request
*flush_rq
= q
->flush_rq
;
423 BUG_ON(!blk_barrier_rq(rq
));
425 rq_init(q
, flush_rq
);
426 flush_rq
->elevator_private
= NULL
;
427 flush_rq
->flags
= REQ_BAR_FLUSH
;
428 flush_rq
->rq_disk
= rq
->rq_disk
;
431 if (q
->prepare_flush_fn(q
, flush_rq
)) {
432 flush_rq
->end_io_data
= rq
;
433 flush_rq
->end_io
= blk_post_flush_end_io
;
435 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
440 static inline int blk_check_end_barrier(request_queue_t
*q
, struct request
*rq
,
443 if (sectors
> rq
->nr_sectors
)
444 sectors
= rq
->nr_sectors
;
446 rq
->nr_sectors
-= sectors
;
447 return rq
->nr_sectors
;
450 static int __blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
,
451 int sectors
, int queue_locked
)
453 if (q
->ordered
!= QUEUE_ORDERED_FLUSH
)
455 if (!blk_fs_request(rq
) || !blk_barrier_rq(rq
))
457 if (blk_barrier_postflush(rq
))
460 if (!blk_check_end_barrier(q
, rq
, sectors
)) {
461 unsigned long flags
= 0;
464 spin_lock_irqsave(q
->queue_lock
, flags
);
466 blk_start_post_flush(q
, rq
);
469 spin_unlock_irqrestore(q
->queue_lock
, flags
);
476 * blk_complete_barrier_rq - complete possible barrier request
477 * @q: the request queue for the device
479 * @sectors: number of sectors to complete
482 * Used in driver end_io handling to determine whether to postpone
483 * completion of a barrier request until a post flush has been done. This
484 * is the unlocked variant, used if the caller doesn't already hold the
487 int blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
, int sectors
)
489 return __blk_complete_barrier_rq(q
, rq
, sectors
, 0);
491 EXPORT_SYMBOL(blk_complete_barrier_rq
);
494 * blk_complete_barrier_rq_locked - complete possible barrier request
495 * @q: the request queue for the device
497 * @sectors: number of sectors to complete
500 * See blk_complete_barrier_rq(). This variant must be used if the caller
501 * holds the queue lock.
503 int blk_complete_barrier_rq_locked(request_queue_t
*q
, struct request
*rq
,
506 return __blk_complete_barrier_rq(q
, rq
, sectors
, 1);
508 EXPORT_SYMBOL(blk_complete_barrier_rq_locked
);
511 * blk_queue_bounce_limit - set bounce buffer limit for queue
512 * @q: the request queue for the device
513 * @dma_addr: bus address limit
516 * Different hardware can have different requirements as to what pages
517 * it can do I/O directly to. A low level driver can call
518 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
519 * buffers for doing I/O to pages residing above @page. By default
520 * the block layer sets this to the highest numbered "low" memory page.
522 void blk_queue_bounce_limit(request_queue_t
*q
, u64 dma_addr
)
524 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
527 * set appropriate bounce gfp mask -- unfortunately we don't have a
528 * full 4GB zone, so we have to resort to low memory for any bounces.
529 * ISA has its own < 16MB zone.
531 if (bounce_pfn
< blk_max_low_pfn
) {
532 BUG_ON(dma_addr
< BLK_BOUNCE_ISA
);
533 init_emergency_isa_pool();
534 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
536 q
->bounce_gfp
= GFP_NOIO
;
538 q
->bounce_pfn
= bounce_pfn
;
541 EXPORT_SYMBOL(blk_queue_bounce_limit
);
544 * blk_queue_max_sectors - set max sectors for a request for this queue
545 * @q: the request queue for the device
546 * @max_sectors: max sectors in the usual 512b unit
549 * Enables a low level driver to set an upper limit on the size of
552 void blk_queue_max_sectors(request_queue_t
*q
, unsigned short max_sectors
)
554 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
555 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
556 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
559 q
->max_sectors
= q
->max_hw_sectors
= max_sectors
;
562 EXPORT_SYMBOL(blk_queue_max_sectors
);
565 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
566 * @q: the request queue for the device
567 * @max_segments: max number of segments
570 * Enables a low level driver to set an upper limit on the number of
571 * physical data segments in a request. This would be the largest sized
572 * scatter list the driver could handle.
574 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
578 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
581 q
->max_phys_segments
= max_segments
;
584 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
587 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
588 * @q: the request queue for the device
589 * @max_segments: max number of segments
592 * Enables a low level driver to set an upper limit on the number of
593 * hw data segments in a request. This would be the largest number of
594 * address/length pairs the host adapter can actually give as once
597 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
601 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
604 q
->max_hw_segments
= max_segments
;
607 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
610 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
611 * @q: the request queue for the device
612 * @max_size: max size of segment in bytes
615 * Enables a low level driver to set an upper limit on the size of a
618 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
620 if (max_size
< PAGE_CACHE_SIZE
) {
621 max_size
= PAGE_CACHE_SIZE
;
622 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
625 q
->max_segment_size
= max_size
;
628 EXPORT_SYMBOL(blk_queue_max_segment_size
);
631 * blk_queue_hardsect_size - set hardware sector size for the queue
632 * @q: the request queue for the device
633 * @size: the hardware sector size, in bytes
636 * This should typically be set to the lowest possible sector size
637 * that the hardware can operate on (possible without reverting to
638 * even internal read-modify-write operations). Usually the default
639 * of 512 covers most hardware.
641 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
643 q
->hardsect_size
= size
;
646 EXPORT_SYMBOL(blk_queue_hardsect_size
);
649 * Returns the minimum that is _not_ zero, unless both are zero.
651 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
654 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
655 * @t: the stacking driver (top)
656 * @b: the underlying device (bottom)
658 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
660 /* zero is "infinity" */
661 t
->max_sectors
= t
->max_hw_sectors
=
662 min_not_zero(t
->max_sectors
,b
->max_sectors
);
664 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
665 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
666 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
667 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
670 EXPORT_SYMBOL(blk_queue_stack_limits
);
673 * blk_queue_segment_boundary - set boundary rules for segment merging
674 * @q: the request queue for the device
675 * @mask: the memory boundary mask
677 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
679 if (mask
< PAGE_CACHE_SIZE
- 1) {
680 mask
= PAGE_CACHE_SIZE
- 1;
681 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
684 q
->seg_boundary_mask
= mask
;
687 EXPORT_SYMBOL(blk_queue_segment_boundary
);
690 * blk_queue_dma_alignment - set dma length and memory alignment
691 * @q: the request queue for the device
692 * @mask: alignment mask
695 * set required memory and length aligment for direct dma transactions.
696 * this is used when buiding direct io requests for the queue.
699 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
701 q
->dma_alignment
= mask
;
704 EXPORT_SYMBOL(blk_queue_dma_alignment
);
707 * blk_queue_find_tag - find a request by its tag and queue
709 * @q: The request queue for the device
710 * @tag: The tag of the request
713 * Should be used when a device returns a tag and you want to match
716 * no locks need be held.
718 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
720 struct blk_queue_tag
*bqt
= q
->queue_tags
;
722 if (unlikely(bqt
== NULL
|| tag
>= bqt
->max_depth
))
725 return bqt
->tag_index
[tag
];
728 EXPORT_SYMBOL(blk_queue_find_tag
);
731 * __blk_queue_free_tags - release tag maintenance info
732 * @q: the request queue for the device
735 * blk_cleanup_queue() will take care of calling this function, if tagging
736 * has been used. So there's no need to call this directly.
738 static void __blk_queue_free_tags(request_queue_t
*q
)
740 struct blk_queue_tag
*bqt
= q
->queue_tags
;
745 if (atomic_dec_and_test(&bqt
->refcnt
)) {
747 BUG_ON(!list_empty(&bqt
->busy_list
));
749 kfree(bqt
->tag_index
);
750 bqt
->tag_index
= NULL
;
758 q
->queue_tags
= NULL
;
759 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
763 * blk_queue_free_tags - release tag maintenance info
764 * @q: the request queue for the device
767 * This is used to disabled tagged queuing to a device, yet leave
770 void blk_queue_free_tags(request_queue_t
*q
)
772 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
775 EXPORT_SYMBOL(blk_queue_free_tags
);
778 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
780 struct request
**tag_index
;
781 unsigned long *tag_map
;
784 if (depth
> q
->nr_requests
* 2) {
785 depth
= q
->nr_requests
* 2;
786 printk(KERN_ERR
"%s: adjusted depth to %d\n",
787 __FUNCTION__
, depth
);
790 tag_index
= kmalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
794 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
795 tag_map
= kmalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
799 memset(tag_index
, 0, depth
* sizeof(struct request
*));
800 memset(tag_map
, 0, nr_ulongs
* sizeof(unsigned long));
801 tags
->max_depth
= depth
;
802 tags
->tag_index
= tag_index
;
803 tags
->tag_map
= tag_map
;
812 * blk_queue_init_tags - initialize the queue tag info
813 * @q: the request queue for the device
814 * @depth: the maximum queue depth supported
815 * @tags: the tag to use
817 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
818 struct blk_queue_tag
*tags
)
822 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
824 if (!tags
&& !q
->queue_tags
) {
825 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
829 if (init_tag_map(q
, tags
, depth
))
832 INIT_LIST_HEAD(&tags
->busy_list
);
834 atomic_set(&tags
->refcnt
, 1);
835 } else if (q
->queue_tags
) {
836 if ((rc
= blk_queue_resize_tags(q
, depth
)))
838 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
841 atomic_inc(&tags
->refcnt
);
844 * assign it, all done
846 q
->queue_tags
= tags
;
847 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
854 EXPORT_SYMBOL(blk_queue_init_tags
);
857 * blk_queue_resize_tags - change the queueing depth
858 * @q: the request queue for the device
859 * @new_depth: the new max command queueing depth
862 * Must be called with the queue lock held.
864 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
866 struct blk_queue_tag
*bqt
= q
->queue_tags
;
867 struct request
**tag_index
;
868 unsigned long *tag_map
;
869 int max_depth
, nr_ulongs
;
875 * save the old state info, so we can copy it back
877 tag_index
= bqt
->tag_index
;
878 tag_map
= bqt
->tag_map
;
879 max_depth
= bqt
->max_depth
;
881 if (init_tag_map(q
, bqt
, new_depth
))
884 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
885 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
886 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
893 EXPORT_SYMBOL(blk_queue_resize_tags
);
896 * blk_queue_end_tag - end tag operations for a request
897 * @q: the request queue for the device
898 * @rq: the request that has completed
901 * Typically called when end_that_request_first() returns 0, meaning
902 * all transfers have been done for a request. It's important to call
903 * this function before end_that_request_last(), as that will put the
904 * request back on the free list thus corrupting the internal tag list.
907 * queue lock must be held.
909 void blk_queue_end_tag(request_queue_t
*q
, struct request
*rq
)
911 struct blk_queue_tag
*bqt
= q
->queue_tags
;
916 if (unlikely(tag
>= bqt
->max_depth
))
918 * This can happen after tag depth has been reduced.
919 * FIXME: how about a warning or info message here?
923 if (unlikely(!__test_and_clear_bit(tag
, bqt
->tag_map
))) {
924 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
929 list_del_init(&rq
->queuelist
);
930 rq
->flags
&= ~REQ_QUEUED
;
933 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
934 printk(KERN_ERR
"%s: tag %d is missing\n",
937 bqt
->tag_index
[tag
] = NULL
;
941 EXPORT_SYMBOL(blk_queue_end_tag
);
944 * blk_queue_start_tag - find a free tag and assign it
945 * @q: the request queue for the device
946 * @rq: the block request that needs tagging
949 * This can either be used as a stand-alone helper, or possibly be
950 * assigned as the queue &prep_rq_fn (in which case &struct request
951 * automagically gets a tag assigned). Note that this function
952 * assumes that any type of request can be queued! if this is not
953 * true for your device, you must check the request type before
954 * calling this function. The request will also be removed from
955 * the request queue, so it's the drivers responsibility to readd
956 * it if it should need to be restarted for some reason.
959 * queue lock must be held.
961 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
963 struct blk_queue_tag
*bqt
= q
->queue_tags
;
966 if (unlikely((rq
->flags
& REQ_QUEUED
))) {
968 "%s: request %p for device [%s] already tagged %d",
970 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
974 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
975 if (tag
>= bqt
->max_depth
)
978 __set_bit(tag
, bqt
->tag_map
);
980 rq
->flags
|= REQ_QUEUED
;
982 bqt
->tag_index
[tag
] = rq
;
983 blkdev_dequeue_request(rq
);
984 list_add(&rq
->queuelist
, &bqt
->busy_list
);
989 EXPORT_SYMBOL(blk_queue_start_tag
);
992 * blk_queue_invalidate_tags - invalidate all pending tags
993 * @q: the request queue for the device
996 * Hardware conditions may dictate a need to stop all pending requests.
997 * In this case, we will safely clear the block side of the tag queue and
998 * readd all requests to the request queue in the right order.
1001 * queue lock must be held.
1003 void blk_queue_invalidate_tags(request_queue_t
*q
)
1005 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1006 struct list_head
*tmp
, *n
;
1009 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1010 rq
= list_entry_rq(tmp
);
1012 if (rq
->tag
== -1) {
1014 "%s: bad tag found on list\n", __FUNCTION__
);
1015 list_del_init(&rq
->queuelist
);
1016 rq
->flags
&= ~REQ_QUEUED
;
1018 blk_queue_end_tag(q
, rq
);
1020 rq
->flags
&= ~REQ_STARTED
;
1021 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1025 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1027 static char *rq_flags
[] = {
1045 "REQ_DRIVE_TASKFILE",
1052 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1056 printk("%s: dev %s: flags = ", msg
,
1057 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?");
1060 if (rq
->flags
& (1 << bit
))
1061 printk("%s ", rq_flags
[bit
]);
1063 } while (bit
< __REQ_NR_BITS
);
1065 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1067 rq
->current_nr_sectors
);
1068 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1070 if (rq
->flags
& (REQ_BLOCK_PC
| REQ_PC
)) {
1072 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1073 printk("%02x ", rq
->cmd
[bit
]);
1078 EXPORT_SYMBOL(blk_dump_rq_flags
);
1080 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1082 struct bio_vec
*bv
, *bvprv
= NULL
;
1083 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1084 int high
, highprv
= 1;
1086 if (unlikely(!bio
->bi_io_vec
))
1089 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1090 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1091 bio_for_each_segment(bv
, bio
, i
) {
1093 * the trick here is making sure that a high page is never
1094 * considered part of another segment, since that might
1095 * change with the bounce page.
1097 high
= page_to_pfn(bv
->bv_page
) >= q
->bounce_pfn
;
1098 if (high
|| highprv
)
1099 goto new_hw_segment
;
1101 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1103 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1105 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1107 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1108 goto new_hw_segment
;
1110 seg_size
+= bv
->bv_len
;
1111 hw_seg_size
+= bv
->bv_len
;
1116 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1117 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1118 hw_seg_size
+= bv
->bv_len
;
1121 if (hw_seg_size
> bio
->bi_hw_front_size
)
1122 bio
->bi_hw_front_size
= hw_seg_size
;
1123 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1129 seg_size
= bv
->bv_len
;
1132 if (hw_seg_size
> bio
->bi_hw_back_size
)
1133 bio
->bi_hw_back_size
= hw_seg_size
;
1134 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1135 bio
->bi_hw_front_size
= hw_seg_size
;
1136 bio
->bi_phys_segments
= nr_phys_segs
;
1137 bio
->bi_hw_segments
= nr_hw_segs
;
1138 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1142 static int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1145 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1148 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1150 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1154 * bio and nxt are contigous in memory, check if the queue allows
1155 * these two to be merged into one
1157 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1163 static int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1166 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1167 blk_recount_segments(q
, bio
);
1168 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1169 blk_recount_segments(q
, nxt
);
1170 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1171 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1173 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1180 * map a request to scatterlist, return number of sg entries setup. Caller
1181 * must make sure sg can hold rq->nr_phys_segments entries
1183 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1185 struct bio_vec
*bvec
, *bvprv
;
1187 int nsegs
, i
, cluster
;
1190 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1193 * for each bio in rq
1196 rq_for_each_bio(bio
, rq
) {
1198 * for each segment in bio
1200 bio_for_each_segment(bvec
, bio
, i
) {
1201 int nbytes
= bvec
->bv_len
;
1203 if (bvprv
&& cluster
) {
1204 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1207 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1209 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1212 sg
[nsegs
- 1].length
+= nbytes
;
1215 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1216 sg
[nsegs
].page
= bvec
->bv_page
;
1217 sg
[nsegs
].length
= nbytes
;
1218 sg
[nsegs
].offset
= bvec
->bv_offset
;
1223 } /* segments in bio */
1229 EXPORT_SYMBOL(blk_rq_map_sg
);
1232 * the standard queue merge functions, can be overridden with device
1233 * specific ones if so desired
1236 static inline int ll_new_mergeable(request_queue_t
*q
,
1237 struct request
*req
,
1240 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1242 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1243 req
->flags
|= REQ_NOMERGE
;
1244 if (req
== q
->last_merge
)
1245 q
->last_merge
= NULL
;
1250 * A hw segment is just getting larger, bump just the phys
1253 req
->nr_phys_segments
+= nr_phys_segs
;
1257 static inline int ll_new_hw_segment(request_queue_t
*q
,
1258 struct request
*req
,
1261 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1262 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1264 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1265 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1266 req
->flags
|= REQ_NOMERGE
;
1267 if (req
== q
->last_merge
)
1268 q
->last_merge
= NULL
;
1273 * This will form the start of a new hw segment. Bump both
1276 req
->nr_hw_segments
+= nr_hw_segs
;
1277 req
->nr_phys_segments
+= nr_phys_segs
;
1281 static int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
,
1286 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1287 req
->flags
|= REQ_NOMERGE
;
1288 if (req
== q
->last_merge
)
1289 q
->last_merge
= NULL
;
1292 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1293 blk_recount_segments(q
, req
->biotail
);
1294 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1295 blk_recount_segments(q
, bio
);
1296 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1297 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1298 !BIOVEC_VIRT_OVERSIZE(len
)) {
1299 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1302 if (req
->nr_hw_segments
== 1)
1303 req
->bio
->bi_hw_front_size
= len
;
1304 if (bio
->bi_hw_segments
== 1)
1305 bio
->bi_hw_back_size
= len
;
1310 return ll_new_hw_segment(q
, req
, bio
);
1313 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1318 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1319 req
->flags
|= REQ_NOMERGE
;
1320 if (req
== q
->last_merge
)
1321 q
->last_merge
= NULL
;
1324 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1325 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1326 blk_recount_segments(q
, bio
);
1327 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1328 blk_recount_segments(q
, req
->bio
);
1329 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1330 !BIOVEC_VIRT_OVERSIZE(len
)) {
1331 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1334 if (bio
->bi_hw_segments
== 1)
1335 bio
->bi_hw_front_size
= len
;
1336 if (req
->nr_hw_segments
== 1)
1337 req
->biotail
->bi_hw_back_size
= len
;
1342 return ll_new_hw_segment(q
, req
, bio
);
1345 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1346 struct request
*next
)
1348 int total_phys_segments
;
1349 int total_hw_segments
;
1352 * First check if the either of the requests are re-queued
1353 * requests. Can't merge them if they are.
1355 if (req
->special
|| next
->special
)
1359 * Will it become too large?
1361 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1364 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1365 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1366 total_phys_segments
--;
1368 if (total_phys_segments
> q
->max_phys_segments
)
1371 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1372 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1373 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1375 * propagate the combined length to the end of the requests
1377 if (req
->nr_hw_segments
== 1)
1378 req
->bio
->bi_hw_front_size
= len
;
1379 if (next
->nr_hw_segments
== 1)
1380 next
->biotail
->bi_hw_back_size
= len
;
1381 total_hw_segments
--;
1384 if (total_hw_segments
> q
->max_hw_segments
)
1387 /* Merge is OK... */
1388 req
->nr_phys_segments
= total_phys_segments
;
1389 req
->nr_hw_segments
= total_hw_segments
;
1394 * "plug" the device if there are no outstanding requests: this will
1395 * force the transfer to start only after we have put all the requests
1398 * This is called with interrupts off and no requests on the queue and
1399 * with the queue lock held.
1401 void blk_plug_device(request_queue_t
*q
)
1403 WARN_ON(!irqs_disabled());
1406 * don't plug a stopped queue, it must be paired with blk_start_queue()
1407 * which will restart the queueing
1409 if (test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
))
1412 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1413 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1416 EXPORT_SYMBOL(blk_plug_device
);
1419 * remove the queue from the plugged list, if present. called with
1420 * queue lock held and interrupts disabled.
1422 int blk_remove_plug(request_queue_t
*q
)
1424 WARN_ON(!irqs_disabled());
1426 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1429 del_timer(&q
->unplug_timer
);
1433 EXPORT_SYMBOL(blk_remove_plug
);
1436 * remove the plug and let it rip..
1438 void __generic_unplug_device(request_queue_t
*q
)
1440 if (unlikely(test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
)))
1443 if (!blk_remove_plug(q
))
1448 EXPORT_SYMBOL(__generic_unplug_device
);
1451 * generic_unplug_device - fire a request queue
1452 * @q: The &request_queue_t in question
1455 * Linux uses plugging to build bigger requests queues before letting
1456 * the device have at them. If a queue is plugged, the I/O scheduler
1457 * is still adding and merging requests on the queue. Once the queue
1458 * gets unplugged, the request_fn defined for the queue is invoked and
1459 * transfers started.
1461 void generic_unplug_device(request_queue_t
*q
)
1463 spin_lock_irq(q
->queue_lock
);
1464 __generic_unplug_device(q
);
1465 spin_unlock_irq(q
->queue_lock
);
1467 EXPORT_SYMBOL(generic_unplug_device
);
1469 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1472 request_queue_t
*q
= bdi
->unplug_io_data
;
1475 * devices don't necessarily have an ->unplug_fn defined
1481 static void blk_unplug_work(void *data
)
1483 request_queue_t
*q
= data
;
1488 static void blk_unplug_timeout(unsigned long data
)
1490 request_queue_t
*q
= (request_queue_t
*)data
;
1492 kblockd_schedule_work(&q
->unplug_work
);
1496 * blk_start_queue - restart a previously stopped queue
1497 * @q: The &request_queue_t in question
1500 * blk_start_queue() will clear the stop flag on the queue, and call
1501 * the request_fn for the queue if it was in a stopped state when
1502 * entered. Also see blk_stop_queue(). Queue lock must be held.
1504 void blk_start_queue(request_queue_t
*q
)
1506 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1509 * one level of recursion is ok and is much faster than kicking
1510 * the unplug handling
1512 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1514 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1517 kblockd_schedule_work(&q
->unplug_work
);
1521 EXPORT_SYMBOL(blk_start_queue
);
1524 * blk_stop_queue - stop a queue
1525 * @q: The &request_queue_t in question
1528 * The Linux block layer assumes that a block driver will consume all
1529 * entries on the request queue when the request_fn strategy is called.
1530 * Often this will not happen, because of hardware limitations (queue
1531 * depth settings). If a device driver gets a 'queue full' response,
1532 * or if it simply chooses not to queue more I/O at one point, it can
1533 * call this function to prevent the request_fn from being called until
1534 * the driver has signalled it's ready to go again. This happens by calling
1535 * blk_start_queue() to restart queue operations. Queue lock must be held.
1537 void blk_stop_queue(request_queue_t
*q
)
1540 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1542 EXPORT_SYMBOL(blk_stop_queue
);
1545 * blk_sync_queue - cancel any pending callbacks on a queue
1549 * The block layer may perform asynchronous callback activity
1550 * on a queue, such as calling the unplug function after a timeout.
1551 * A block device may call blk_sync_queue to ensure that any
1552 * such activity is cancelled, thus allowing it to release resources
1553 * the the callbacks might use. The caller must already have made sure
1554 * that its ->make_request_fn will not re-add plugging prior to calling
1558 void blk_sync_queue(struct request_queue
*q
)
1560 del_timer_sync(&q
->unplug_timer
);
1563 EXPORT_SYMBOL(blk_sync_queue
);
1566 * blk_run_queue - run a single device queue
1567 * @q: The queue to run
1569 void blk_run_queue(struct request_queue
*q
)
1571 unsigned long flags
;
1573 spin_lock_irqsave(q
->queue_lock
, flags
);
1575 if (!elv_queue_empty(q
))
1577 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1579 EXPORT_SYMBOL(blk_run_queue
);
1582 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1583 * @q: the request queue to be released
1586 * blk_cleanup_queue is the pair to blk_init_queue() or
1587 * blk_queue_make_request(). It should be called when a request queue is
1588 * being released; typically when a block device is being de-registered.
1589 * Currently, its primary task it to free all the &struct request
1590 * structures that were allocated to the queue and the queue itself.
1593 * Hopefully the low level driver will have finished any
1594 * outstanding requests first...
1596 void blk_cleanup_queue(request_queue_t
* q
)
1598 struct request_list
*rl
= &q
->rq
;
1600 if (!atomic_dec_and_test(&q
->refcnt
))
1604 elevator_exit(q
->elevator
);
1609 mempool_destroy(rl
->rq_pool
);
1612 __blk_queue_free_tags(q
);
1614 blk_queue_ordered(q
, QUEUE_ORDERED_NONE
);
1616 kmem_cache_free(requestq_cachep
, q
);
1619 EXPORT_SYMBOL(blk_cleanup_queue
);
1621 static int blk_init_free_list(request_queue_t
*q
)
1623 struct request_list
*rl
= &q
->rq
;
1625 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1626 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1627 init_waitqueue_head(&rl
->wait
[READ
]);
1628 init_waitqueue_head(&rl
->wait
[WRITE
]);
1629 init_waitqueue_head(&rl
->drain
);
1631 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1632 mempool_free_slab
, request_cachep
, q
->node
);
1640 static int __make_request(request_queue_t
*, struct bio
*);
1642 request_queue_t
*blk_alloc_queue(int gfp_mask
)
1644 return blk_alloc_queue_node(gfp_mask
, -1);
1646 EXPORT_SYMBOL(blk_alloc_queue
);
1648 request_queue_t
*blk_alloc_queue_node(int gfp_mask
, int node_id
)
1652 q
= kmem_cache_alloc_node(requestq_cachep
, gfp_mask
, node_id
);
1656 memset(q
, 0, sizeof(*q
));
1657 init_timer(&q
->unplug_timer
);
1658 atomic_set(&q
->refcnt
, 1);
1660 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1661 q
->backing_dev_info
.unplug_io_data
= q
;
1665 EXPORT_SYMBOL(blk_alloc_queue_node
);
1668 * blk_init_queue - prepare a request queue for use with a block device
1669 * @rfn: The function to be called to process requests that have been
1670 * placed on the queue.
1671 * @lock: Request queue spin lock
1674 * If a block device wishes to use the standard request handling procedures,
1675 * which sorts requests and coalesces adjacent requests, then it must
1676 * call blk_init_queue(). The function @rfn will be called when there
1677 * are requests on the queue that need to be processed. If the device
1678 * supports plugging, then @rfn may not be called immediately when requests
1679 * are available on the queue, but may be called at some time later instead.
1680 * Plugged queues are generally unplugged when a buffer belonging to one
1681 * of the requests on the queue is needed, or due to memory pressure.
1683 * @rfn is not required, or even expected, to remove all requests off the
1684 * queue, but only as many as it can handle at a time. If it does leave
1685 * requests on the queue, it is responsible for arranging that the requests
1686 * get dealt with eventually.
1688 * The queue spin lock must be held while manipulating the requests on the
1691 * Function returns a pointer to the initialized request queue, or NULL if
1692 * it didn't succeed.
1695 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1696 * when the block device is deactivated (such as at module unload).
1699 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1701 return blk_init_queue_node(rfn
, lock
, -1);
1703 EXPORT_SYMBOL(blk_init_queue
);
1706 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1708 request_queue_t
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1714 if (blk_init_free_list(q
))
1718 * if caller didn't supply a lock, they get per-queue locking with
1722 spin_lock_init(&q
->__queue_lock
);
1723 lock
= &q
->__queue_lock
;
1726 q
->request_fn
= rfn
;
1727 q
->back_merge_fn
= ll_back_merge_fn
;
1728 q
->front_merge_fn
= ll_front_merge_fn
;
1729 q
->merge_requests_fn
= ll_merge_requests_fn
;
1730 q
->prep_rq_fn
= NULL
;
1731 q
->unplug_fn
= generic_unplug_device
;
1732 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1733 q
->queue_lock
= lock
;
1735 blk_queue_segment_boundary(q
, 0xffffffff);
1737 blk_queue_make_request(q
, __make_request
);
1738 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1740 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1741 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1746 if (!elevator_init(q
, NULL
)) {
1747 blk_queue_congestion_threshold(q
);
1751 blk_cleanup_queue(q
);
1753 kmem_cache_free(requestq_cachep
, q
);
1756 EXPORT_SYMBOL(blk_init_queue_node
);
1758 int blk_get_queue(request_queue_t
*q
)
1760 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1761 atomic_inc(&q
->refcnt
);
1768 EXPORT_SYMBOL(blk_get_queue
);
1770 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
1772 elv_put_request(q
, rq
);
1773 mempool_free(rq
, q
->rq
.rq_pool
);
1776 static inline struct request
*
1777 blk_alloc_request(request_queue_t
*q
, int rw
, struct bio
*bio
, int gfp_mask
)
1779 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1785 * first three bits are identical in rq->flags and bio->bi_rw,
1786 * see bio.h and blkdev.h
1790 if (!elv_set_request(q
, rq
, bio
, gfp_mask
))
1793 mempool_free(rq
, q
->rq
.rq_pool
);
1798 * ioc_batching returns true if the ioc is a valid batching request and
1799 * should be given priority access to a request.
1801 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
1807 * Make sure the process is able to allocate at least 1 request
1808 * even if the batch times out, otherwise we could theoretically
1811 return ioc
->nr_batch_requests
== q
->nr_batching
||
1812 (ioc
->nr_batch_requests
> 0
1813 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1817 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1818 * will cause the process to be a "batcher" on all queues in the system. This
1819 * is the behaviour we want though - once it gets a wakeup it should be given
1822 static void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
1824 if (!ioc
|| ioc_batching(q
, ioc
))
1827 ioc
->nr_batch_requests
= q
->nr_batching
;
1828 ioc
->last_waited
= jiffies
;
1831 static void __freed_request(request_queue_t
*q
, int rw
)
1833 struct request_list
*rl
= &q
->rq
;
1835 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
1836 clear_queue_congested(q
, rw
);
1838 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
1839 if (waitqueue_active(&rl
->wait
[rw
]))
1840 wake_up(&rl
->wait
[rw
]);
1842 blk_clear_queue_full(q
, rw
);
1847 * A request has just been released. Account for it, update the full and
1848 * congestion status, wake up any waiters. Called under q->queue_lock.
1850 static void freed_request(request_queue_t
*q
, int rw
)
1852 struct request_list
*rl
= &q
->rq
;
1856 __freed_request(q
, rw
);
1858 if (unlikely(rl
->starved
[rw
^ 1]))
1859 __freed_request(q
, rw
^ 1);
1861 if (!rl
->count
[READ
] && !rl
->count
[WRITE
]) {
1863 if (unlikely(waitqueue_active(&rl
->drain
)))
1864 wake_up(&rl
->drain
);
1868 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1870 * Get a free request, queue_lock must not be held
1872 static struct request
*get_request(request_queue_t
*q
, int rw
, struct bio
*bio
,
1875 struct request
*rq
= NULL
;
1876 struct request_list
*rl
= &q
->rq
;
1877 struct io_context
*ioc
= get_io_context(gfp_mask
);
1879 if (unlikely(test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
)))
1882 spin_lock_irq(q
->queue_lock
);
1883 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
1885 * The queue will fill after this allocation, so set it as
1886 * full, and mark this process as "batching". This process
1887 * will be allowed to complete a batch of requests, others
1890 if (!blk_queue_full(q
, rw
)) {
1891 ioc_set_batching(q
, ioc
);
1892 blk_set_queue_full(q
, rw
);
1896 switch (elv_may_queue(q
, rw
, bio
)) {
1899 case ELV_MQUEUE_MAY
:
1901 case ELV_MQUEUE_MUST
:
1905 if (blk_queue_full(q
, rw
) && !ioc_batching(q
, ioc
)) {
1907 * The queue is full and the allocating process is not a
1908 * "batcher", and not exempted by the IO scheduler
1910 spin_unlock_irq(q
->queue_lock
);
1916 rl
->starved
[rw
] = 0;
1917 if (rl
->count
[rw
] >= queue_congestion_on_threshold(q
))
1918 set_queue_congested(q
, rw
);
1919 spin_unlock_irq(q
->queue_lock
);
1921 rq
= blk_alloc_request(q
, rw
, bio
, gfp_mask
);
1924 * Allocation failed presumably due to memory. Undo anything
1925 * we might have messed up.
1927 * Allocating task should really be put onto the front of the
1928 * wait queue, but this is pretty rare.
1930 spin_lock_irq(q
->queue_lock
);
1931 freed_request(q
, rw
);
1934 * in the very unlikely event that allocation failed and no
1935 * requests for this direction was pending, mark us starved
1936 * so that freeing of a request in the other direction will
1937 * notice us. another possible fix would be to split the
1938 * rq mempool into READ and WRITE
1941 if (unlikely(rl
->count
[rw
] == 0))
1942 rl
->starved
[rw
] = 1;
1944 spin_unlock_irq(q
->queue_lock
);
1948 if (ioc_batching(q
, ioc
))
1949 ioc
->nr_batch_requests
--;
1954 put_io_context(ioc
);
1959 * No available requests for this queue, unplug the device and wait for some
1960 * requests to become available.
1962 static struct request
*get_request_wait(request_queue_t
*q
, int rw
,
1969 struct request_list
*rl
= &q
->rq
;
1971 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
1972 TASK_UNINTERRUPTIBLE
);
1974 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
1977 struct io_context
*ioc
;
1979 generic_unplug_device(q
);
1983 * After sleeping, we become a "batching" process and
1984 * will be able to allocate at least one request, and
1985 * up to a big batch of them for a small period time.
1986 * See ioc_batching, ioc_set_batching
1988 ioc
= get_io_context(GFP_NOIO
);
1989 ioc_set_batching(q
, ioc
);
1990 put_io_context(ioc
);
1992 finish_wait(&rl
->wait
[rw
], &wait
);
1998 struct request
*blk_get_request(request_queue_t
*q
, int rw
, int gfp_mask
)
2002 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2004 if (gfp_mask
& __GFP_WAIT
)
2005 rq
= get_request_wait(q
, rw
, NULL
);
2007 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2012 EXPORT_SYMBOL(blk_get_request
);
2015 * blk_requeue_request - put a request back on queue
2016 * @q: request queue where request should be inserted
2017 * @rq: request to be inserted
2020 * Drivers often keep queueing requests until the hardware cannot accept
2021 * more, when that condition happens we need to put the request back
2022 * on the queue. Must be called with queue lock held.
2024 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2026 if (blk_rq_tagged(rq
))
2027 blk_queue_end_tag(q
, rq
);
2029 elv_requeue_request(q
, rq
);
2032 EXPORT_SYMBOL(blk_requeue_request
);
2035 * blk_insert_request - insert a special request in to a request queue
2036 * @q: request queue where request should be inserted
2037 * @rq: request to be inserted
2038 * @at_head: insert request at head or tail of queue
2039 * @data: private data
2042 * Many block devices need to execute commands asynchronously, so they don't
2043 * block the whole kernel from preemption during request execution. This is
2044 * accomplished normally by inserting aritficial requests tagged as
2045 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2046 * scheduled for actual execution by the request queue.
2048 * We have the option of inserting the head or the tail of the queue.
2049 * Typically we use the tail for new ioctls and so forth. We use the head
2050 * of the queue for things like a QUEUE_FULL message from a device, or a
2051 * host that is unable to accept a particular command.
2053 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2054 int at_head
, void *data
)
2056 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2057 unsigned long flags
;
2060 * tell I/O scheduler that this isn't a regular read/write (ie it
2061 * must not attempt merges on this) and that it acts as a soft
2064 rq
->flags
|= REQ_SPECIAL
| REQ_SOFTBARRIER
;
2068 spin_lock_irqsave(q
->queue_lock
, flags
);
2071 * If command is tagged, release the tag
2073 if (blk_rq_tagged(rq
))
2074 blk_queue_end_tag(q
, rq
);
2076 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2077 __elv_add_request(q
, rq
, where
, 0);
2079 if (blk_queue_plugged(q
))
2080 __generic_unplug_device(q
);
2083 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2086 EXPORT_SYMBOL(blk_insert_request
);
2089 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2090 * @q: request queue where request should be inserted
2091 * @rw: READ or WRITE data
2092 * @ubuf: the user buffer
2093 * @len: length of user data
2096 * Data will be mapped directly for zero copy io, if possible. Otherwise
2097 * a kernel bounce buffer is used.
2099 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2100 * still in process context.
2102 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2103 * before being submitted to the device, as pages mapped may be out of
2104 * reach. It's the callers responsibility to make sure this happens. The
2105 * original bio must be passed back in to blk_rq_unmap_user() for proper
2108 struct request
*blk_rq_map_user(request_queue_t
*q
, int rw
, void __user
*ubuf
,
2111 unsigned long uaddr
;
2115 if (len
> (q
->max_sectors
<< 9))
2116 return ERR_PTR(-EINVAL
);
2117 if ((!len
&& ubuf
) || (len
&& !ubuf
))
2118 return ERR_PTR(-EINVAL
);
2120 rq
= blk_get_request(q
, rw
, __GFP_WAIT
);
2122 return ERR_PTR(-ENOMEM
);
2125 * if alignment requirement is satisfied, map in user pages for
2126 * direct dma. else, set up kernel bounce buffers
2128 uaddr
= (unsigned long) ubuf
;
2129 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2130 bio
= bio_map_user(q
, NULL
, uaddr
, len
, rw
== READ
);
2132 bio
= bio_copy_user(q
, uaddr
, len
, rw
== READ
);
2135 rq
->bio
= rq
->biotail
= bio
;
2136 blk_rq_bio_prep(q
, rq
, bio
);
2138 rq
->buffer
= rq
->data
= NULL
;
2144 * bio is the err-ptr
2146 blk_put_request(rq
);
2147 return (struct request
*) bio
;
2150 EXPORT_SYMBOL(blk_rq_map_user
);
2153 * blk_rq_unmap_user - unmap a request with user data
2154 * @rq: request to be unmapped
2155 * @bio: bio for the request
2156 * @ulen: length of user buffer
2159 * Unmap a request previously mapped by blk_rq_map_user().
2161 int blk_rq_unmap_user(struct request
*rq
, struct bio
*bio
, unsigned int ulen
)
2166 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2167 bio_unmap_user(bio
);
2169 ret
= bio_uncopy_user(bio
);
2172 blk_put_request(rq
);
2176 EXPORT_SYMBOL(blk_rq_unmap_user
);
2179 * blk_execute_rq - insert a request into queue for execution
2180 * @q: queue to insert the request in
2181 * @bd_disk: matching gendisk
2182 * @rq: request to insert
2185 * Insert a fully prepared request at the back of the io scheduler queue
2188 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2191 DECLARE_COMPLETION(wait
);
2192 char sense
[SCSI_SENSE_BUFFERSIZE
];
2195 rq
->rq_disk
= bd_disk
;
2198 * we need an extra reference to the request, so we can look at
2199 * it after io completion
2204 memset(sense
, 0, sizeof(sense
));
2209 rq
->flags
|= REQ_NOMERGE
;
2210 rq
->waiting
= &wait
;
2211 rq
->end_io
= blk_end_sync_rq
;
2212 elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 1);
2213 generic_unplug_device(q
);
2214 wait_for_completion(&wait
);
2223 EXPORT_SYMBOL(blk_execute_rq
);
2226 * blkdev_issue_flush - queue a flush
2227 * @bdev: blockdev to issue flush for
2228 * @error_sector: error sector
2231 * Issue a flush for the block device in question. Caller can supply
2232 * room for storing the error offset in case of a flush error, if they
2233 * wish to. Caller must run wait_for_completion() on its own.
2235 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2239 if (bdev
->bd_disk
== NULL
)
2242 q
= bdev_get_queue(bdev
);
2245 if (!q
->issue_flush_fn
)
2248 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2251 EXPORT_SYMBOL(blkdev_issue_flush
);
2253 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2255 int rw
= rq_data_dir(rq
);
2257 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2261 __disk_stat_add(rq
->rq_disk
, read_sectors
, nr_sectors
);
2263 __disk_stat_inc(rq
->rq_disk
, read_merges
);
2264 } else if (rw
== WRITE
) {
2265 __disk_stat_add(rq
->rq_disk
, write_sectors
, nr_sectors
);
2267 __disk_stat_inc(rq
->rq_disk
, write_merges
);
2270 disk_round_stats(rq
->rq_disk
);
2271 rq
->rq_disk
->in_flight
++;
2276 * add-request adds a request to the linked list.
2277 * queue lock is held and interrupts disabled, as we muck with the
2278 * request queue list.
2280 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2282 drive_stat_acct(req
, req
->nr_sectors
, 1);
2285 q
->activity_fn(q
->activity_data
, rq_data_dir(req
));
2288 * elevator indicated where it wants this request to be
2289 * inserted at elevator_merge time
2291 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2295 * disk_round_stats() - Round off the performance stats on a struct
2298 * The average IO queue length and utilisation statistics are maintained
2299 * by observing the current state of the queue length and the amount of
2300 * time it has been in this state for.
2302 * Normally, that accounting is done on IO completion, but that can result
2303 * in more than a second's worth of IO being accounted for within any one
2304 * second, leading to >100% utilisation. To deal with that, we call this
2305 * function to do a round-off before returning the results when reading
2306 * /proc/diskstats. This accounts immediately for all queue usage up to
2307 * the current jiffies and restarts the counters again.
2309 void disk_round_stats(struct gendisk
*disk
)
2311 unsigned long now
= jiffies
;
2313 __disk_stat_add(disk
, time_in_queue
,
2314 disk
->in_flight
* (now
- disk
->stamp
));
2317 if (disk
->in_flight
)
2318 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp_idle
));
2319 disk
->stamp_idle
= now
;
2323 * queue lock must be held
2325 static void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2327 struct request_list
*rl
= req
->rl
;
2331 if (unlikely(--req
->ref_count
))
2334 req
->rq_status
= RQ_INACTIVE
;
2338 * Request may not have originated from ll_rw_blk. if not,
2339 * it didn't come out of our reserved rq pools
2342 int rw
= rq_data_dir(req
);
2344 elv_completed_request(q
, req
);
2346 BUG_ON(!list_empty(&req
->queuelist
));
2348 blk_free_request(q
, req
);
2349 freed_request(q
, rw
);
2353 void blk_put_request(struct request
*req
)
2356 * if req->rl isn't set, this request didnt originate from the
2357 * block layer, so it's safe to just disregard it
2360 unsigned long flags
;
2361 request_queue_t
*q
= req
->q
;
2363 spin_lock_irqsave(q
->queue_lock
, flags
);
2364 __blk_put_request(q
, req
);
2365 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2369 EXPORT_SYMBOL(blk_put_request
);
2372 * blk_end_sync_rq - executes a completion event on a request
2373 * @rq: request to complete
2375 void blk_end_sync_rq(struct request
*rq
)
2377 struct completion
*waiting
= rq
->waiting
;
2380 __blk_put_request(rq
->q
, rq
);
2383 * complete last, if this is a stack request the process (and thus
2384 * the rq pointer) could be invalid right after this complete()
2388 EXPORT_SYMBOL(blk_end_sync_rq
);
2391 * blk_congestion_wait - wait for a queue to become uncongested
2392 * @rw: READ or WRITE
2393 * @timeout: timeout in jiffies
2395 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2396 * If no queues are congested then just wait for the next request to be
2399 long blk_congestion_wait(int rw
, long timeout
)
2403 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
2405 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
2406 ret
= io_schedule_timeout(timeout
);
2407 finish_wait(wqh
, &wait
);
2411 EXPORT_SYMBOL(blk_congestion_wait
);
2414 * Has to be called with the request spinlock acquired
2416 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2417 struct request
*next
)
2419 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2425 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2428 if (rq_data_dir(req
) != rq_data_dir(next
)
2429 || req
->rq_disk
!= next
->rq_disk
2430 || next
->waiting
|| next
->special
)
2434 * If we are allowed to merge, then append bio list
2435 * from next to rq and release next. merge_requests_fn
2436 * will have updated segment counts, update sector
2439 if (!q
->merge_requests_fn(q
, req
, next
))
2443 * At this point we have either done a back merge
2444 * or front merge. We need the smaller start_time of
2445 * the merged requests to be the current request
2446 * for accounting purposes.
2448 if (time_after(req
->start_time
, next
->start_time
))
2449 req
->start_time
= next
->start_time
;
2451 req
->biotail
->bi_next
= next
->bio
;
2452 req
->biotail
= next
->biotail
;
2454 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2456 elv_merge_requests(q
, req
, next
);
2459 disk_round_stats(req
->rq_disk
);
2460 req
->rq_disk
->in_flight
--;
2463 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
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
;
2517 unsigned short prio
;
2520 sector
= bio
->bi_sector
;
2521 nr_sectors
= bio_sectors(bio
);
2522 cur_nr_sectors
= bio_cur_sectors(bio
);
2523 prio
= bio_prio(bio
);
2525 rw
= bio_data_dir(bio
);
2526 sync
= bio_sync(bio
);
2529 * low level driver can indicate that it wants pages above a
2530 * certain limit bounced to low memory (ie for highmem, or even
2531 * ISA dma in theory)
2533 blk_queue_bounce(q
, &bio
);
2535 spin_lock_prefetch(q
->queue_lock
);
2537 barrier
= bio_barrier(bio
);
2538 if (unlikely(barrier
) && (q
->ordered
== QUEUE_ORDERED_NONE
)) {
2544 spin_lock_irq(q
->queue_lock
);
2546 if (elv_queue_empty(q
)) {
2553 el_ret
= elv_merge(q
, &req
, bio
);
2555 case ELEVATOR_BACK_MERGE
:
2556 BUG_ON(!rq_mergeable(req
));
2558 if (!q
->back_merge_fn(q
, req
, bio
))
2561 req
->biotail
->bi_next
= bio
;
2563 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2564 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2565 drive_stat_acct(req
, nr_sectors
, 0);
2566 if (!attempt_back_merge(q
, req
))
2567 elv_merged_request(q
, req
);
2570 case ELEVATOR_FRONT_MERGE
:
2571 BUG_ON(!rq_mergeable(req
));
2573 if (!q
->front_merge_fn(q
, req
, bio
))
2576 bio
->bi_next
= req
->bio
;
2580 * may not be valid. if the low level driver said
2581 * it didn't need a bounce buffer then it better
2582 * not touch req->buffer either...
2584 req
->buffer
= bio_data(bio
);
2585 req
->current_nr_sectors
= cur_nr_sectors
;
2586 req
->hard_cur_sectors
= cur_nr_sectors
;
2587 req
->sector
= req
->hard_sector
= sector
;
2588 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2589 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2590 drive_stat_acct(req
, nr_sectors
, 0);
2591 if (!attempt_front_merge(q
, req
))
2592 elv_merged_request(q
, req
);
2596 * elevator says don't/can't merge. get new request
2598 case ELEVATOR_NO_MERGE
:
2602 printk("elevator returned crap (%d)\n", el_ret
);
2607 * Grab a free request from the freelist - if that is empty, check
2608 * if we are doing read ahead and abort instead of blocking for
2616 spin_unlock_irq(q
->queue_lock
);
2617 if ((freereq
= get_request(q
, rw
, bio
, GFP_ATOMIC
)) == NULL
) {
2622 if (bio_rw_ahead(bio
))
2625 freereq
= get_request_wait(q
, rw
, bio
);
2630 req
->flags
|= REQ_CMD
;
2633 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2635 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2636 req
->flags
|= REQ_FAILFAST
;
2639 * REQ_BARRIER implies no merging, but lets make it explicit
2641 if (unlikely(barrier
))
2642 req
->flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2645 req
->hard_sector
= req
->sector
= sector
;
2646 req
->hard_nr_sectors
= req
->nr_sectors
= nr_sectors
;
2647 req
->current_nr_sectors
= req
->hard_cur_sectors
= cur_nr_sectors
;
2648 req
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2649 req
->nr_hw_segments
= bio_hw_segments(q
, bio
);
2650 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2651 req
->waiting
= NULL
;
2652 req
->bio
= req
->biotail
= bio
;
2654 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2655 req
->start_time
= jiffies
;
2657 add_request(q
, req
);
2660 __blk_put_request(q
, freereq
);
2662 __generic_unplug_device(q
);
2664 spin_unlock_irq(q
->queue_lock
);
2668 bio_endio(bio
, nr_sectors
<< 9, err
);
2673 * If bio->bi_dev is a partition, remap the location
2675 static inline void blk_partition_remap(struct bio
*bio
)
2677 struct block_device
*bdev
= bio
->bi_bdev
;
2679 if (bdev
!= bdev
->bd_contains
) {
2680 struct hd_struct
*p
= bdev
->bd_part
;
2682 switch (bio_data_dir(bio
)) {
2684 p
->read_sectors
+= bio_sectors(bio
);
2688 p
->write_sectors
+= bio_sectors(bio
);
2692 bio
->bi_sector
+= p
->start_sect
;
2693 bio
->bi_bdev
= bdev
->bd_contains
;
2697 void blk_finish_queue_drain(request_queue_t
*q
)
2699 struct request_list
*rl
= &q
->rq
;
2703 spin_lock_irq(q
->queue_lock
);
2704 clear_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
);
2706 while (!list_empty(&q
->drain_list
)) {
2707 rq
= list_entry_rq(q
->drain_list
.next
);
2709 list_del_init(&rq
->queuelist
);
2710 elv_requeue_request(q
, rq
);
2717 spin_unlock_irq(q
->queue_lock
);
2719 wake_up(&rl
->wait
[0]);
2720 wake_up(&rl
->wait
[1]);
2721 wake_up(&rl
->drain
);
2724 static int wait_drain(request_queue_t
*q
, struct request_list
*rl
, int dispatch
)
2726 int wait
= rl
->count
[READ
] + rl
->count
[WRITE
];
2729 wait
+= !list_empty(&q
->queue_head
);
2735 * We rely on the fact that only requests allocated through blk_alloc_request()
2736 * have io scheduler private data structures associated with them. Any other
2737 * type of request (allocated on stack or through kmalloc()) should not go
2738 * to the io scheduler core, but be attached to the queue head instead.
2740 void blk_wait_queue_drained(request_queue_t
*q
, int wait_dispatch
)
2742 struct request_list
*rl
= &q
->rq
;
2745 spin_lock_irq(q
->queue_lock
);
2746 set_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
);
2748 while (wait_drain(q
, rl
, wait_dispatch
)) {
2749 prepare_to_wait(&rl
->drain
, &wait
, TASK_UNINTERRUPTIBLE
);
2751 if (wait_drain(q
, rl
, wait_dispatch
)) {
2752 __generic_unplug_device(q
);
2753 spin_unlock_irq(q
->queue_lock
);
2755 spin_lock_irq(q
->queue_lock
);
2758 finish_wait(&rl
->drain
, &wait
);
2761 spin_unlock_irq(q
->queue_lock
);
2765 * block waiting for the io scheduler being started again.
2767 static inline void block_wait_queue_running(request_queue_t
*q
)
2771 while (unlikely(test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
))) {
2772 struct request_list
*rl
= &q
->rq
;
2774 prepare_to_wait_exclusive(&rl
->drain
, &wait
,
2775 TASK_UNINTERRUPTIBLE
);
2778 * re-check the condition. avoids using prepare_to_wait()
2779 * in the fast path (queue is running)
2781 if (test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
))
2784 finish_wait(&rl
->drain
, &wait
);
2788 static void handle_bad_sector(struct bio
*bio
)
2790 char b
[BDEVNAME_SIZE
];
2792 printk(KERN_INFO
"attempt to access beyond end of device\n");
2793 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
2794 bdevname(bio
->bi_bdev
, b
),
2796 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
2797 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
2799 set_bit(BIO_EOF
, &bio
->bi_flags
);
2803 * generic_make_request: hand a buffer to its device driver for I/O
2804 * @bio: The bio describing the location in memory and on the device.
2806 * generic_make_request() is used to make I/O requests of block
2807 * devices. It is passed a &struct bio, which describes the I/O that needs
2810 * generic_make_request() does not return any status. The
2811 * success/failure status of the request, along with notification of
2812 * completion, is delivered asynchronously through the bio->bi_end_io
2813 * function described (one day) else where.
2815 * The caller of generic_make_request must make sure that bi_io_vec
2816 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2817 * set to describe the device address, and the
2818 * bi_end_io and optionally bi_private are set to describe how
2819 * completion notification should be signaled.
2821 * generic_make_request and the drivers it calls may use bi_next if this
2822 * bio happens to be merged with someone else, and may change bi_dev and
2823 * bi_sector for remaps as it sees fit. So the values of these fields
2824 * should NOT be depended on after the call to generic_make_request.
2826 void generic_make_request(struct bio
*bio
)
2830 int ret
, nr_sectors
= bio_sectors(bio
);
2833 /* Test device or partition size, when known. */
2834 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
2836 sector_t sector
= bio
->bi_sector
;
2838 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2840 * This may well happen - the kernel calls bread()
2841 * without checking the size of the device, e.g., when
2842 * mounting a device.
2844 handle_bad_sector(bio
);
2850 * Resolve the mapping until finished. (drivers are
2851 * still free to implement/resolve their own stacking
2852 * by explicitly returning 0)
2854 * NOTE: we don't repeat the blk_size check for each new device.
2855 * Stacking drivers are expected to know what they are doing.
2858 char b
[BDEVNAME_SIZE
];
2860 q
= bdev_get_queue(bio
->bi_bdev
);
2863 "generic_make_request: Trying to access "
2864 "nonexistent block-device %s (%Lu)\n",
2865 bdevname(bio
->bi_bdev
, b
),
2866 (long long) bio
->bi_sector
);
2868 bio_endio(bio
, bio
->bi_size
, -EIO
);
2872 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
2873 printk("bio too big device %s (%u > %u)\n",
2874 bdevname(bio
->bi_bdev
, b
),
2880 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
2883 block_wait_queue_running(q
);
2886 * If this device has partitions, remap block n
2887 * of partition p to block n+start(p) of the disk.
2889 blk_partition_remap(bio
);
2891 ret
= q
->make_request_fn(q
, bio
);
2895 EXPORT_SYMBOL(generic_make_request
);
2898 * submit_bio: submit a bio to the block device layer for I/O
2899 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2900 * @bio: The &struct bio which describes the I/O
2902 * submit_bio() is very similar in purpose to generic_make_request(), and
2903 * uses that function to do most of the work. Both are fairly rough
2904 * interfaces, @bio must be presetup and ready for I/O.
2907 void submit_bio(int rw
, struct bio
*bio
)
2909 int count
= bio_sectors(bio
);
2911 BIO_BUG_ON(!bio
->bi_size
);
2912 BIO_BUG_ON(!bio
->bi_io_vec
);
2915 mod_page_state(pgpgout
, count
);
2917 mod_page_state(pgpgin
, count
);
2919 if (unlikely(block_dump
)) {
2920 char b
[BDEVNAME_SIZE
];
2921 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
2922 current
->comm
, current
->pid
,
2923 (rw
& WRITE
) ? "WRITE" : "READ",
2924 (unsigned long long)bio
->bi_sector
,
2925 bdevname(bio
->bi_bdev
,b
));
2928 generic_make_request(bio
);
2931 EXPORT_SYMBOL(submit_bio
);
2933 static void blk_recalc_rq_segments(struct request
*rq
)
2935 struct bio
*bio
, *prevbio
= NULL
;
2936 int nr_phys_segs
, nr_hw_segs
;
2937 unsigned int phys_size
, hw_size
;
2938 request_queue_t
*q
= rq
->q
;
2943 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
2944 rq_for_each_bio(bio
, rq
) {
2945 /* Force bio hw/phys segs to be recalculated. */
2946 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
2948 nr_phys_segs
+= bio_phys_segments(q
, bio
);
2949 nr_hw_segs
+= bio_hw_segments(q
, bio
);
2951 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
2952 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
2954 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
2955 pseg
<= q
->max_segment_size
) {
2957 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
2961 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
2962 hseg
<= q
->max_segment_size
) {
2964 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
2971 rq
->nr_phys_segments
= nr_phys_segs
;
2972 rq
->nr_hw_segments
= nr_hw_segs
;
2975 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
2977 if (blk_fs_request(rq
)) {
2978 rq
->hard_sector
+= nsect
;
2979 rq
->hard_nr_sectors
-= nsect
;
2982 * Move the I/O submission pointers ahead if required.
2984 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
2985 (rq
->sector
<= rq
->hard_sector
)) {
2986 rq
->sector
= rq
->hard_sector
;
2987 rq
->nr_sectors
= rq
->hard_nr_sectors
;
2988 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
2989 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
2990 rq
->buffer
= bio_data(rq
->bio
);
2994 * if total number of sectors is less than the first segment
2995 * size, something has gone terribly wrong
2997 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
2998 printk("blk: request botched\n");
2999 rq
->nr_sectors
= rq
->current_nr_sectors
;
3004 static int __end_that_request_first(struct request
*req
, int uptodate
,
3007 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3011 * extend uptodate bool to allow < 0 value to be direct io error
3014 if (end_io_error(uptodate
))
3015 error
= !uptodate
? -EIO
: uptodate
;
3018 * for a REQ_BLOCK_PC request, we want to carry any eventual
3019 * sense key with us all the way through
3021 if (!blk_pc_request(req
))
3025 if (blk_fs_request(req
) && !(req
->flags
& REQ_QUIET
))
3026 printk("end_request: I/O error, dev %s, sector %llu\n",
3027 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3028 (unsigned long long)req
->sector
);
3031 total_bytes
= bio_nbytes
= 0;
3032 while ((bio
= req
->bio
) != NULL
) {
3035 if (nr_bytes
>= bio
->bi_size
) {
3036 req
->bio
= bio
->bi_next
;
3037 nbytes
= bio
->bi_size
;
3038 bio_endio(bio
, nbytes
, error
);
3042 int idx
= bio
->bi_idx
+ next_idx
;
3044 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3045 blk_dump_rq_flags(req
, "__end_that");
3046 printk("%s: bio idx %d >= vcnt %d\n",
3048 bio
->bi_idx
, bio
->bi_vcnt
);
3052 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3053 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3056 * not a complete bvec done
3058 if (unlikely(nbytes
> nr_bytes
)) {
3059 bio_nbytes
+= nr_bytes
;
3060 total_bytes
+= nr_bytes
;
3065 * advance to the next vector
3068 bio_nbytes
+= nbytes
;
3071 total_bytes
+= nbytes
;
3074 if ((bio
= req
->bio
)) {
3076 * end more in this run, or just return 'not-done'
3078 if (unlikely(nr_bytes
<= 0))
3090 * if the request wasn't completed, update state
3093 bio_endio(bio
, bio_nbytes
, error
);
3094 bio
->bi_idx
+= next_idx
;
3095 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3096 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3099 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3100 blk_recalc_rq_segments(req
);
3105 * end_that_request_first - end I/O on a request
3106 * @req: the request being processed
3107 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3108 * @nr_sectors: number of sectors to end I/O on
3111 * Ends I/O on a number of sectors attached to @req, and sets it up
3112 * for the next range of segments (if any) in the cluster.
3115 * 0 - we are done with this request, call end_that_request_last()
3116 * 1 - still buffers pending for this request
3118 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3120 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3123 EXPORT_SYMBOL(end_that_request_first
);
3126 * end_that_request_chunk - end I/O on a request
3127 * @req: the request being processed
3128 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3129 * @nr_bytes: number of bytes to complete
3132 * Ends I/O on a number of bytes attached to @req, and sets it up
3133 * for the next range of segments (if any). Like end_that_request_first(),
3134 * but deals with bytes instead of sectors.
3137 * 0 - we are done with this request, call end_that_request_last()
3138 * 1 - still buffers pending for this request
3140 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3142 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3145 EXPORT_SYMBOL(end_that_request_chunk
);
3148 * queue lock must be held
3150 void end_that_request_last(struct request
*req
)
3152 struct gendisk
*disk
= req
->rq_disk
;
3154 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3155 laptop_io_completion();
3157 if (disk
&& blk_fs_request(req
)) {
3158 unsigned long duration
= jiffies
- req
->start_time
;
3159 switch (rq_data_dir(req
)) {
3161 __disk_stat_inc(disk
, writes
);
3162 __disk_stat_add(disk
, write_ticks
, duration
);
3165 __disk_stat_inc(disk
, reads
);
3166 __disk_stat_add(disk
, read_ticks
, duration
);
3169 disk_round_stats(disk
);
3175 __blk_put_request(req
->q
, req
);
3178 EXPORT_SYMBOL(end_that_request_last
);
3180 void end_request(struct request
*req
, int uptodate
)
3182 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3183 add_disk_randomness(req
->rq_disk
);
3184 blkdev_dequeue_request(req
);
3185 end_that_request_last(req
);
3189 EXPORT_SYMBOL(end_request
);
3191 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3193 /* first three bits are identical in rq->flags and bio->bi_rw */
3194 rq
->flags
|= (bio
->bi_rw
& 7);
3196 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3197 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3198 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3199 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3200 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3201 rq
->buffer
= bio_data(bio
);
3203 rq
->bio
= rq
->biotail
= bio
;
3206 EXPORT_SYMBOL(blk_rq_bio_prep
);
3208 int kblockd_schedule_work(struct work_struct
*work
)
3210 return queue_work(kblockd_workqueue
, work
);
3213 EXPORT_SYMBOL(kblockd_schedule_work
);
3215 void kblockd_flush(void)
3217 flush_workqueue(kblockd_workqueue
);
3219 EXPORT_SYMBOL(kblockd_flush
);
3221 int __init
blk_dev_init(void)
3223 kblockd_workqueue
= create_workqueue("kblockd");
3224 if (!kblockd_workqueue
)
3225 panic("Failed to create kblockd\n");
3227 request_cachep
= kmem_cache_create("blkdev_requests",
3228 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3230 requestq_cachep
= kmem_cache_create("blkdev_queue",
3231 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3233 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3234 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3236 blk_max_low_pfn
= max_low_pfn
;
3237 blk_max_pfn
= max_pfn
;
3243 * IO Context helper functions
3245 void put_io_context(struct io_context
*ioc
)
3250 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3252 if (atomic_dec_and_test(&ioc
->refcount
)) {
3253 if (ioc
->aic
&& ioc
->aic
->dtor
)
3254 ioc
->aic
->dtor(ioc
->aic
);
3255 if (ioc
->cic
&& ioc
->cic
->dtor
)
3256 ioc
->cic
->dtor(ioc
->cic
);
3258 kmem_cache_free(iocontext_cachep
, ioc
);
3261 EXPORT_SYMBOL(put_io_context
);
3263 /* Called by the exitting task */
3264 void exit_io_context(void)
3266 unsigned long flags
;
3267 struct io_context
*ioc
;
3269 local_irq_save(flags
);
3271 ioc
= current
->io_context
;
3272 current
->io_context
= NULL
;
3274 task_unlock(current
);
3275 local_irq_restore(flags
);
3277 if (ioc
->aic
&& ioc
->aic
->exit
)
3278 ioc
->aic
->exit(ioc
->aic
);
3279 if (ioc
->cic
&& ioc
->cic
->exit
)
3280 ioc
->cic
->exit(ioc
->cic
);
3282 put_io_context(ioc
);
3286 * If the current task has no IO context then create one and initialise it.
3287 * If it does have a context, take a ref on it.
3289 * This is always called in the context of the task which submitted the I/O.
3290 * But weird things happen, so we disable local interrupts to ensure exclusive
3291 * access to *current.
3293 struct io_context
*get_io_context(int gfp_flags
)
3295 struct task_struct
*tsk
= current
;
3296 unsigned long flags
;
3297 struct io_context
*ret
;
3299 local_irq_save(flags
);
3300 ret
= tsk
->io_context
;
3304 local_irq_restore(flags
);
3306 ret
= kmem_cache_alloc(iocontext_cachep
, gfp_flags
);
3308 atomic_set(&ret
->refcount
, 1);
3309 ret
->task
= current
;
3310 ret
->set_ioprio
= NULL
;
3311 ret
->last_waited
= jiffies
; /* doesn't matter... */
3312 ret
->nr_batch_requests
= 0; /* because this is 0 */
3316 local_irq_save(flags
);
3319 * very unlikely, someone raced with us in setting up the task
3320 * io context. free new context and just grab a reference.
3322 if (!tsk
->io_context
)
3323 tsk
->io_context
= ret
;
3325 kmem_cache_free(iocontext_cachep
, ret
);
3326 ret
= tsk
->io_context
;
3330 atomic_inc(&ret
->refcount
);
3331 local_irq_restore(flags
);
3336 EXPORT_SYMBOL(get_io_context
);
3338 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3340 struct io_context
*src
= *psrc
;
3341 struct io_context
*dst
= *pdst
;
3344 BUG_ON(atomic_read(&src
->refcount
) == 0);
3345 atomic_inc(&src
->refcount
);
3346 put_io_context(dst
);
3350 EXPORT_SYMBOL(copy_io_context
);
3352 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3354 struct io_context
*temp
;
3359 EXPORT_SYMBOL(swap_io_context
);
3364 struct queue_sysfs_entry
{
3365 struct attribute attr
;
3366 ssize_t (*show
)(struct request_queue
*, char *);
3367 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3371 queue_var_show(unsigned int var
, char *page
)
3373 return sprintf(page
, "%d\n", var
);
3377 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3379 char *p
= (char *) page
;
3381 *var
= simple_strtoul(p
, &p
, 10);
3385 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3387 return queue_var_show(q
->nr_requests
, (page
));
3391 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3393 struct request_list
*rl
= &q
->rq
;
3395 int ret
= queue_var_store(&q
->nr_requests
, page
, count
);
3396 if (q
->nr_requests
< BLKDEV_MIN_RQ
)
3397 q
->nr_requests
= BLKDEV_MIN_RQ
;
3398 blk_queue_congestion_threshold(q
);
3400 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3401 set_queue_congested(q
, READ
);
3402 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3403 clear_queue_congested(q
, READ
);
3405 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3406 set_queue_congested(q
, WRITE
);
3407 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3408 clear_queue_congested(q
, WRITE
);
3410 if (rl
->count
[READ
] >= q
->nr_requests
) {
3411 blk_set_queue_full(q
, READ
);
3412 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3413 blk_clear_queue_full(q
, READ
);
3414 wake_up(&rl
->wait
[READ
]);
3417 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3418 blk_set_queue_full(q
, WRITE
);
3419 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3420 blk_clear_queue_full(q
, WRITE
);
3421 wake_up(&rl
->wait
[WRITE
]);
3426 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3428 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3430 return queue_var_show(ra_kb
, (page
));
3434 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3436 unsigned long ra_kb
;
3437 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3439 spin_lock_irq(q
->queue_lock
);
3440 if (ra_kb
> (q
->max_sectors
>> 1))
3441 ra_kb
= (q
->max_sectors
>> 1);
3443 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3444 spin_unlock_irq(q
->queue_lock
);
3449 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3451 int max_sectors_kb
= q
->max_sectors
>> 1;
3453 return queue_var_show(max_sectors_kb
, (page
));
3457 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3459 unsigned long max_sectors_kb
,
3460 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3461 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3462 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3465 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3468 * Take the queue lock to update the readahead and max_sectors
3469 * values synchronously:
3471 spin_lock_irq(q
->queue_lock
);
3473 * Trim readahead window as well, if necessary:
3475 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3476 if (ra_kb
> max_sectors_kb
)
3477 q
->backing_dev_info
.ra_pages
=
3478 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3480 q
->max_sectors
= max_sectors_kb
<< 1;
3481 spin_unlock_irq(q
->queue_lock
);
3486 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3488 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3490 return queue_var_show(max_hw_sectors_kb
, (page
));
3494 static struct queue_sysfs_entry queue_requests_entry
= {
3495 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
3496 .show
= queue_requests_show
,
3497 .store
= queue_requests_store
,
3500 static struct queue_sysfs_entry queue_ra_entry
= {
3501 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
3502 .show
= queue_ra_show
,
3503 .store
= queue_ra_store
,
3506 static struct queue_sysfs_entry queue_max_sectors_entry
= {
3507 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
3508 .show
= queue_max_sectors_show
,
3509 .store
= queue_max_sectors_store
,
3512 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
3513 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
3514 .show
= queue_max_hw_sectors_show
,
3517 static struct queue_sysfs_entry queue_iosched_entry
= {
3518 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
3519 .show
= elv_iosched_show
,
3520 .store
= elv_iosched_store
,
3523 static struct attribute
*default_attrs
[] = {
3524 &queue_requests_entry
.attr
,
3525 &queue_ra_entry
.attr
,
3526 &queue_max_hw_sectors_entry
.attr
,
3527 &queue_max_sectors_entry
.attr
,
3528 &queue_iosched_entry
.attr
,
3532 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3535 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3537 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3538 struct request_queue
*q
;
3540 q
= container_of(kobj
, struct request_queue
, kobj
);
3544 return entry
->show(q
, page
);
3548 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3549 const char *page
, size_t length
)
3551 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3552 struct request_queue
*q
;
3554 q
= container_of(kobj
, struct request_queue
, kobj
);
3558 return entry
->store(q
, page
, length
);
3561 static struct sysfs_ops queue_sysfs_ops
= {
3562 .show
= queue_attr_show
,
3563 .store
= queue_attr_store
,
3566 static struct kobj_type queue_ktype
= {
3567 .sysfs_ops
= &queue_sysfs_ops
,
3568 .default_attrs
= default_attrs
,
3571 int blk_register_queue(struct gendisk
*disk
)
3575 request_queue_t
*q
= disk
->queue
;
3577 if (!q
|| !q
->request_fn
)
3580 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
3581 if (!q
->kobj
.parent
)
3584 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
3585 q
->kobj
.ktype
= &queue_ktype
;
3587 ret
= kobject_register(&q
->kobj
);
3591 ret
= elv_register_queue(q
);
3593 kobject_unregister(&q
->kobj
);
3600 void blk_unregister_queue(struct gendisk
*disk
)
3602 request_queue_t
*q
= disk
->queue
;
3604 if (q
&& q
->request_fn
) {
3605 elv_unregister_queue(q
);
3607 kobject_unregister(&q
->kobj
);
3608 kobject_put(&disk
->kobj
);