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>
35 #include <scsi/scsi_cmnd.h>
37 static void blk_unplug_work(void *data
);
38 static void blk_unplug_timeout(unsigned long data
);
41 * For the allocated request tables
43 static kmem_cache_t
*request_cachep
;
46 * For queue allocation
48 static kmem_cache_t
*requestq_cachep
;
51 * For io context allocations
53 static kmem_cache_t
*iocontext_cachep
;
55 static wait_queue_head_t congestion_wqh
[2] = {
56 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
57 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
61 * Controlling structure to kblockd
63 static struct workqueue_struct
*kblockd_workqueue
;
65 unsigned long blk_max_low_pfn
, blk_max_pfn
;
67 EXPORT_SYMBOL(blk_max_low_pfn
);
68 EXPORT_SYMBOL(blk_max_pfn
);
70 /* Amount of time in which a process may batch requests */
71 #define BLK_BATCH_TIME (HZ/50UL)
73 /* Number of requests a "batching" process may submit */
74 #define BLK_BATCH_REQ 32
77 * Return the threshold (number of used requests) at which the queue is
78 * considered to be congested. It include a little hysteresis to keep the
79 * context switch rate down.
81 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
83 return q
->nr_congestion_on
;
87 * The threshold at which a queue is considered to be uncongested
89 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
91 return q
->nr_congestion_off
;
94 static void blk_queue_congestion_threshold(struct request_queue
*q
)
98 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
99 if (nr
> q
->nr_requests
)
101 q
->nr_congestion_on
= nr
;
103 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
106 q
->nr_congestion_off
= nr
;
110 * A queue has just exitted congestion. Note this in the global counter of
111 * congested queues, and wake up anyone who was waiting for requests to be
114 static void clear_queue_congested(request_queue_t
*q
, int rw
)
117 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
119 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
120 clear_bit(bit
, &q
->backing_dev_info
.state
);
121 smp_mb__after_clear_bit();
122 if (waitqueue_active(wqh
))
127 * A queue has just entered congestion. Flag that in the queue's VM-visible
128 * state flags and increment the global gounter of congested queues.
130 static void set_queue_congested(request_queue_t
*q
, int rw
)
134 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
135 set_bit(bit
, &q
->backing_dev_info
.state
);
139 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
142 * Locates the passed device's request queue and returns the address of its
145 * Will return NULL if the request queue cannot be located.
147 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
149 struct backing_dev_info
*ret
= NULL
;
150 request_queue_t
*q
= bdev_get_queue(bdev
);
153 ret
= &q
->backing_dev_info
;
157 EXPORT_SYMBOL(blk_get_backing_dev_info
);
159 void blk_queue_activity_fn(request_queue_t
*q
, activity_fn
*fn
, void *data
)
162 q
->activity_data
= data
;
165 EXPORT_SYMBOL(blk_queue_activity_fn
);
168 * blk_queue_prep_rq - set a prepare_request function for queue
170 * @pfn: prepare_request function
172 * It's possible for a queue to register a prepare_request callback which
173 * is invoked before the request is handed to the request_fn. The goal of
174 * the function is to prepare a request for I/O, it can be used to build a
175 * cdb from the request data for instance.
178 void blk_queue_prep_rq(request_queue_t
*q
, prep_rq_fn
*pfn
)
183 EXPORT_SYMBOL(blk_queue_prep_rq
);
186 * blk_queue_merge_bvec - set a merge_bvec function for queue
188 * @mbfn: merge_bvec_fn
190 * Usually queues have static limitations on the max sectors or segments that
191 * we can put in a request. Stacking drivers may have some settings that
192 * are dynamic, and thus we have to query the queue whether it is ok to
193 * add a new bio_vec to a bio at a given offset or not. If the block device
194 * has such limitations, it needs to register a merge_bvec_fn to control
195 * the size of bio's sent to it. Note that a block device *must* allow a
196 * single page to be added to an empty bio. The block device driver may want
197 * to use the bio_split() function to deal with these bio's. By default
198 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
201 void blk_queue_merge_bvec(request_queue_t
*q
, merge_bvec_fn
*mbfn
)
203 q
->merge_bvec_fn
= mbfn
;
206 EXPORT_SYMBOL(blk_queue_merge_bvec
);
209 * blk_queue_make_request - define an alternate make_request function for a device
210 * @q: the request queue for the device to be affected
211 * @mfn: the alternate make_request function
214 * The normal way for &struct bios to be passed to a device
215 * driver is for them to be collected into requests on a request
216 * queue, and then to allow the device driver to select requests
217 * off that queue when it is ready. This works well for many block
218 * devices. However some block devices (typically virtual devices
219 * such as md or lvm) do not benefit from the processing on the
220 * request queue, and are served best by having the requests passed
221 * directly to them. This can be achieved by providing a function
222 * to blk_queue_make_request().
225 * The driver that does this *must* be able to deal appropriately
226 * with buffers in "highmemory". This can be accomplished by either calling
227 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
228 * blk_queue_bounce() to create a buffer in normal memory.
230 void blk_queue_make_request(request_queue_t
* q
, make_request_fn
* mfn
)
235 q
->nr_requests
= BLKDEV_MAX_RQ
;
236 q
->max_phys_segments
= MAX_PHYS_SEGMENTS
;
237 q
->max_hw_segments
= MAX_HW_SEGMENTS
;
238 q
->make_request_fn
= mfn
;
239 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
240 q
->backing_dev_info
.state
= 0;
241 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
242 blk_queue_max_sectors(q
, MAX_SECTORS
);
243 blk_queue_hardsect_size(q
, 512);
244 blk_queue_dma_alignment(q
, 511);
245 blk_queue_congestion_threshold(q
);
246 q
->nr_batching
= BLK_BATCH_REQ
;
248 q
->unplug_thresh
= 4; /* hmm */
249 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
250 if (q
->unplug_delay
== 0)
253 INIT_WORK(&q
->unplug_work
, blk_unplug_work
, q
);
255 q
->unplug_timer
.function
= blk_unplug_timeout
;
256 q
->unplug_timer
.data
= (unsigned long)q
;
259 * by default assume old behaviour and bounce for any highmem page
261 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
263 blk_queue_activity_fn(q
, NULL
, NULL
);
265 INIT_LIST_HEAD(&q
->drain_list
);
268 EXPORT_SYMBOL(blk_queue_make_request
);
270 static inline void rq_init(request_queue_t
*q
, struct request
*rq
)
272 INIT_LIST_HEAD(&rq
->queuelist
);
275 rq
->rq_status
= RQ_ACTIVE
;
276 rq
->bio
= rq
->biotail
= NULL
;
286 rq
->end_io_data
= NULL
;
290 * blk_queue_ordered - does this queue support ordered writes
291 * @q: the request queue
295 * For journalled file systems, doing ordered writes on a commit
296 * block instead of explicitly doing wait_on_buffer (which is bad
297 * for performance) can be a big win. Block drivers supporting this
298 * feature should call this function and indicate so.
301 void blk_queue_ordered(request_queue_t
*q
, int flag
)
304 case QUEUE_ORDERED_NONE
:
306 kmem_cache_free(request_cachep
, q
->flush_rq
);
310 case QUEUE_ORDERED_TAG
:
313 case QUEUE_ORDERED_FLUSH
:
316 q
->flush_rq
= kmem_cache_alloc(request_cachep
,
320 printk("blk_queue_ordered: bad value %d\n", flag
);
325 EXPORT_SYMBOL(blk_queue_ordered
);
328 * blk_queue_issue_flush_fn - set function for issuing a flush
329 * @q: the request queue
330 * @iff: the function to be called issuing the flush
333 * If a driver supports issuing a flush command, the support is notified
334 * to the block layer by defining it through this call.
337 void blk_queue_issue_flush_fn(request_queue_t
*q
, issue_flush_fn
*iff
)
339 q
->issue_flush_fn
= iff
;
342 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
345 * Cache flushing for ordered writes handling
347 static void blk_pre_flush_end_io(struct request
*flush_rq
)
349 struct request
*rq
= flush_rq
->end_io_data
;
350 request_queue_t
*q
= rq
->q
;
352 rq
->flags
|= REQ_BAR_PREFLUSH
;
354 if (!flush_rq
->errors
)
355 elv_requeue_request(q
, rq
);
357 q
->end_flush_fn(q
, flush_rq
);
358 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
363 static void blk_post_flush_end_io(struct request
*flush_rq
)
365 struct request
*rq
= flush_rq
->end_io_data
;
366 request_queue_t
*q
= rq
->q
;
368 rq
->flags
|= REQ_BAR_POSTFLUSH
;
370 q
->end_flush_fn(q
, flush_rq
);
371 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
375 struct request
*blk_start_pre_flush(request_queue_t
*q
, struct request
*rq
)
377 struct request
*flush_rq
= q
->flush_rq
;
379 BUG_ON(!blk_barrier_rq(rq
));
381 if (test_and_set_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
))
384 rq_init(q
, flush_rq
);
385 flush_rq
->elevator_private
= NULL
;
386 flush_rq
->flags
= REQ_BAR_FLUSH
;
387 flush_rq
->rq_disk
= rq
->rq_disk
;
391 * prepare_flush returns 0 if no flush is needed, just mark both
392 * pre and post flush as done in that case
394 if (!q
->prepare_flush_fn(q
, flush_rq
)) {
395 rq
->flags
|= REQ_BAR_PREFLUSH
| REQ_BAR_POSTFLUSH
;
396 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
401 * some drivers dequeue requests right away, some only after io
402 * completion. make sure the request is dequeued.
404 if (!list_empty(&rq
->queuelist
))
405 blkdev_dequeue_request(rq
);
407 elv_deactivate_request(q
, rq
);
409 flush_rq
->end_io_data
= rq
;
410 flush_rq
->end_io
= blk_pre_flush_end_io
;
412 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
416 static void blk_start_post_flush(request_queue_t
*q
, struct request
*rq
)
418 struct request
*flush_rq
= q
->flush_rq
;
420 BUG_ON(!blk_barrier_rq(rq
));
422 rq_init(q
, flush_rq
);
423 flush_rq
->elevator_private
= NULL
;
424 flush_rq
->flags
= REQ_BAR_FLUSH
;
425 flush_rq
->rq_disk
= rq
->rq_disk
;
428 if (q
->prepare_flush_fn(q
, flush_rq
)) {
429 flush_rq
->end_io_data
= rq
;
430 flush_rq
->end_io
= blk_post_flush_end_io
;
432 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
437 static inline int blk_check_end_barrier(request_queue_t
*q
, struct request
*rq
,
440 if (sectors
> rq
->nr_sectors
)
441 sectors
= rq
->nr_sectors
;
443 rq
->nr_sectors
-= sectors
;
444 return rq
->nr_sectors
;
447 static int __blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
,
448 int sectors
, int queue_locked
)
450 if (q
->ordered
!= QUEUE_ORDERED_FLUSH
)
452 if (!blk_fs_request(rq
) || !blk_barrier_rq(rq
))
454 if (blk_barrier_postflush(rq
))
457 if (!blk_check_end_barrier(q
, rq
, sectors
)) {
458 unsigned long flags
= 0;
461 spin_lock_irqsave(q
->queue_lock
, flags
);
463 blk_start_post_flush(q
, rq
);
466 spin_unlock_irqrestore(q
->queue_lock
, flags
);
473 * blk_complete_barrier_rq - complete possible barrier request
474 * @q: the request queue for the device
476 * @sectors: number of sectors to complete
479 * Used in driver end_io handling to determine whether to postpone
480 * completion of a barrier request until a post flush has been done. This
481 * is the unlocked variant, used if the caller doesn't already hold the
484 int blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
, int sectors
)
486 return __blk_complete_barrier_rq(q
, rq
, sectors
, 0);
488 EXPORT_SYMBOL(blk_complete_barrier_rq
);
491 * blk_complete_barrier_rq_locked - complete possible barrier request
492 * @q: the request queue for the device
494 * @sectors: number of sectors to complete
497 * See blk_complete_barrier_rq(). This variant must be used if the caller
498 * holds the queue lock.
500 int blk_complete_barrier_rq_locked(request_queue_t
*q
, struct request
*rq
,
503 return __blk_complete_barrier_rq(q
, rq
, sectors
, 1);
505 EXPORT_SYMBOL(blk_complete_barrier_rq_locked
);
508 * blk_queue_bounce_limit - set bounce buffer limit for queue
509 * @q: the request queue for the device
510 * @dma_addr: bus address limit
513 * Different hardware can have different requirements as to what pages
514 * it can do I/O directly to. A low level driver can call
515 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
516 * buffers for doing I/O to pages residing above @page. By default
517 * the block layer sets this to the highest numbered "low" memory page.
519 void blk_queue_bounce_limit(request_queue_t
*q
, u64 dma_addr
)
521 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
524 * set appropriate bounce gfp mask -- unfortunately we don't have a
525 * full 4GB zone, so we have to resort to low memory for any bounces.
526 * ISA has its own < 16MB zone.
528 if (bounce_pfn
< blk_max_low_pfn
) {
529 BUG_ON(dma_addr
< BLK_BOUNCE_ISA
);
530 init_emergency_isa_pool();
531 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
533 q
->bounce_gfp
= GFP_NOIO
;
535 q
->bounce_pfn
= bounce_pfn
;
538 EXPORT_SYMBOL(blk_queue_bounce_limit
);
541 * blk_queue_max_sectors - set max sectors for a request for this queue
542 * @q: the request queue for the device
543 * @max_sectors: max sectors in the usual 512b unit
546 * Enables a low level driver to set an upper limit on the size of
549 void blk_queue_max_sectors(request_queue_t
*q
, unsigned short max_sectors
)
551 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
552 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
553 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
556 q
->max_sectors
= q
->max_hw_sectors
= max_sectors
;
559 EXPORT_SYMBOL(blk_queue_max_sectors
);
562 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
563 * @q: the request queue for the device
564 * @max_segments: max number of segments
567 * Enables a low level driver to set an upper limit on the number of
568 * physical data segments in a request. This would be the largest sized
569 * scatter list the driver could handle.
571 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
575 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
578 q
->max_phys_segments
= max_segments
;
581 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
584 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
585 * @q: the request queue for the device
586 * @max_segments: max number of segments
589 * Enables a low level driver to set an upper limit on the number of
590 * hw data segments in a request. This would be the largest number of
591 * address/length pairs the host adapter can actually give as once
594 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
598 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
601 q
->max_hw_segments
= max_segments
;
604 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
607 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
608 * @q: the request queue for the device
609 * @max_size: max size of segment in bytes
612 * Enables a low level driver to set an upper limit on the size of a
615 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
617 if (max_size
< PAGE_CACHE_SIZE
) {
618 max_size
= PAGE_CACHE_SIZE
;
619 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
622 q
->max_segment_size
= max_size
;
625 EXPORT_SYMBOL(blk_queue_max_segment_size
);
628 * blk_queue_hardsect_size - set hardware sector size for the queue
629 * @q: the request queue for the device
630 * @size: the hardware sector size, in bytes
633 * This should typically be set to the lowest possible sector size
634 * that the hardware can operate on (possible without reverting to
635 * even internal read-modify-write operations). Usually the default
636 * of 512 covers most hardware.
638 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
640 q
->hardsect_size
= size
;
643 EXPORT_SYMBOL(blk_queue_hardsect_size
);
646 * Returns the minimum that is _not_ zero, unless both are zero.
648 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
651 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
652 * @t: the stacking driver (top)
653 * @b: the underlying device (bottom)
655 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
657 /* zero is "infinity" */
658 t
->max_sectors
= t
->max_hw_sectors
=
659 min_not_zero(t
->max_sectors
,b
->max_sectors
);
661 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
662 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
663 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
664 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
667 EXPORT_SYMBOL(blk_queue_stack_limits
);
670 * blk_queue_segment_boundary - set boundary rules for segment merging
671 * @q: the request queue for the device
672 * @mask: the memory boundary mask
674 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
676 if (mask
< PAGE_CACHE_SIZE
- 1) {
677 mask
= PAGE_CACHE_SIZE
- 1;
678 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
681 q
->seg_boundary_mask
= mask
;
684 EXPORT_SYMBOL(blk_queue_segment_boundary
);
687 * blk_queue_dma_alignment - set dma length and memory alignment
688 * @q: the request queue for the device
689 * @mask: alignment mask
692 * set required memory and length aligment for direct dma transactions.
693 * this is used when buiding direct io requests for the queue.
696 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
698 q
->dma_alignment
= mask
;
701 EXPORT_SYMBOL(blk_queue_dma_alignment
);
704 * blk_queue_find_tag - find a request by its tag and queue
706 * @q: The request queue for the device
707 * @tag: The tag of the request
710 * Should be used when a device returns a tag and you want to match
713 * no locks need be held.
715 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
717 struct blk_queue_tag
*bqt
= q
->queue_tags
;
719 if (unlikely(bqt
== NULL
|| tag
>= bqt
->real_max_depth
))
722 return bqt
->tag_index
[tag
];
725 EXPORT_SYMBOL(blk_queue_find_tag
);
728 * __blk_queue_free_tags - release tag maintenance info
729 * @q: the request queue for the device
732 * blk_cleanup_queue() will take care of calling this function, if tagging
733 * has been used. So there's no need to call this directly.
735 static void __blk_queue_free_tags(request_queue_t
*q
)
737 struct blk_queue_tag
*bqt
= q
->queue_tags
;
742 if (atomic_dec_and_test(&bqt
->refcnt
)) {
744 BUG_ON(!list_empty(&bqt
->busy_list
));
746 kfree(bqt
->tag_index
);
747 bqt
->tag_index
= NULL
;
755 q
->queue_tags
= NULL
;
756 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
760 * blk_queue_free_tags - release tag maintenance info
761 * @q: the request queue for the device
764 * This is used to disabled tagged queuing to a device, yet leave
767 void blk_queue_free_tags(request_queue_t
*q
)
769 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
772 EXPORT_SYMBOL(blk_queue_free_tags
);
775 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
778 struct request
**tag_index
;
779 unsigned long *tag_map
;
781 if (depth
> q
->nr_requests
* 2) {
782 depth
= q
->nr_requests
* 2;
783 printk(KERN_ERR
"%s: adjusted depth to %d\n",
784 __FUNCTION__
, depth
);
787 tag_index
= kmalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
791 bits
= (depth
/ BLK_TAGS_PER_LONG
) + 1;
792 tag_map
= kmalloc(bits
* sizeof(unsigned long), GFP_ATOMIC
);
796 memset(tag_index
, 0, depth
* sizeof(struct request
*));
797 memset(tag_map
, 0, bits
* sizeof(unsigned long));
798 tags
->max_depth
= depth
;
799 tags
->real_max_depth
= bits
* BITS_PER_LONG
;
800 tags
->tag_index
= tag_index
;
801 tags
->tag_map
= tag_map
;
804 * set the upper bits if the depth isn't a multiple of the word size
806 for (i
= depth
; i
< bits
* BLK_TAGS_PER_LONG
; i
++)
807 __set_bit(i
, tag_map
);
816 * blk_queue_init_tags - initialize the queue tag info
817 * @q: the request queue for the device
818 * @depth: the maximum queue depth supported
819 * @tags: the tag to use
821 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
822 struct blk_queue_tag
*tags
)
826 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
828 if (!tags
&& !q
->queue_tags
) {
829 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
833 if (init_tag_map(q
, tags
, depth
))
836 INIT_LIST_HEAD(&tags
->busy_list
);
838 atomic_set(&tags
->refcnt
, 1);
839 } else if (q
->queue_tags
) {
840 if ((rc
= blk_queue_resize_tags(q
, depth
)))
842 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
845 atomic_inc(&tags
->refcnt
);
848 * assign it, all done
850 q
->queue_tags
= tags
;
851 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
858 EXPORT_SYMBOL(blk_queue_init_tags
);
861 * blk_queue_resize_tags - change the queueing depth
862 * @q: the request queue for the device
863 * @new_depth: the new max command queueing depth
866 * Must be called with the queue lock held.
868 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
870 struct blk_queue_tag
*bqt
= q
->queue_tags
;
871 struct request
**tag_index
;
872 unsigned long *tag_map
;
879 * don't bother sizing down
881 if (new_depth
<= bqt
->real_max_depth
) {
882 bqt
->max_depth
= new_depth
;
887 * save the old state info, so we can copy it back
889 tag_index
= bqt
->tag_index
;
890 tag_map
= bqt
->tag_map
;
891 max_depth
= bqt
->real_max_depth
;
893 if (init_tag_map(q
, bqt
, new_depth
))
896 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
897 bits
= max_depth
/ BLK_TAGS_PER_LONG
;
898 memcpy(bqt
->tag_map
, tag_map
, bits
* sizeof(unsigned long));
905 EXPORT_SYMBOL(blk_queue_resize_tags
);
908 * blk_queue_end_tag - end tag operations for a request
909 * @q: the request queue for the device
910 * @rq: the request that has completed
913 * Typically called when end_that_request_first() returns 0, meaning
914 * all transfers have been done for a request. It's important to call
915 * this function before end_that_request_last(), as that will put the
916 * request back on the free list thus corrupting the internal tag list.
919 * queue lock must be held.
921 void blk_queue_end_tag(request_queue_t
*q
, struct request
*rq
)
923 struct blk_queue_tag
*bqt
= q
->queue_tags
;
928 if (unlikely(tag
>= bqt
->real_max_depth
))
931 if (unlikely(!__test_and_clear_bit(tag
, bqt
->tag_map
))) {
932 printk("attempt to clear non-busy tag (%d)\n", tag
);
936 list_del_init(&rq
->queuelist
);
937 rq
->flags
&= ~REQ_QUEUED
;
940 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
941 printk("tag %d is missing\n", tag
);
943 bqt
->tag_index
[tag
] = NULL
;
947 EXPORT_SYMBOL(blk_queue_end_tag
);
950 * blk_queue_start_tag - find a free tag and assign it
951 * @q: the request queue for the device
952 * @rq: the block request that needs tagging
955 * This can either be used as a stand-alone helper, or possibly be
956 * assigned as the queue &prep_rq_fn (in which case &struct request
957 * automagically gets a tag assigned). Note that this function
958 * assumes that any type of request can be queued! if this is not
959 * true for your device, you must check the request type before
960 * calling this function. The request will also be removed from
961 * the request queue, so it's the drivers responsibility to readd
962 * it if it should need to be restarted for some reason.
965 * queue lock must be held.
967 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
969 struct blk_queue_tag
*bqt
= q
->queue_tags
;
970 unsigned long *map
= bqt
->tag_map
;
973 if (unlikely((rq
->flags
& REQ_QUEUED
))) {
975 "request %p for device [%s] already tagged %d",
976 rq
, rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
980 for (map
= bqt
->tag_map
; *map
== -1UL; map
++) {
981 tag
+= BLK_TAGS_PER_LONG
;
983 if (tag
>= bqt
->max_depth
)
988 __set_bit(tag
, bqt
->tag_map
);
990 rq
->flags
|= REQ_QUEUED
;
992 bqt
->tag_index
[tag
] = rq
;
993 blkdev_dequeue_request(rq
);
994 list_add(&rq
->queuelist
, &bqt
->busy_list
);
999 EXPORT_SYMBOL(blk_queue_start_tag
);
1002 * blk_queue_invalidate_tags - invalidate all pending tags
1003 * @q: the request queue for the device
1006 * Hardware conditions may dictate a need to stop all pending requests.
1007 * In this case, we will safely clear the block side of the tag queue and
1008 * readd all requests to the request queue in the right order.
1011 * queue lock must be held.
1013 void blk_queue_invalidate_tags(request_queue_t
*q
)
1015 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1016 struct list_head
*tmp
, *n
;
1019 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1020 rq
= list_entry_rq(tmp
);
1022 if (rq
->tag
== -1) {
1023 printk("bad tag found on list\n");
1024 list_del_init(&rq
->queuelist
);
1025 rq
->flags
&= ~REQ_QUEUED
;
1027 blk_queue_end_tag(q
, rq
);
1029 rq
->flags
&= ~REQ_STARTED
;
1030 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1034 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1036 static char *rq_flags
[] = {
1054 "REQ_DRIVE_TASKFILE",
1061 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1065 printk("%s: dev %s: flags = ", msg
,
1066 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?");
1069 if (rq
->flags
& (1 << bit
))
1070 printk("%s ", rq_flags
[bit
]);
1072 } while (bit
< __REQ_NR_BITS
);
1074 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1076 rq
->current_nr_sectors
);
1077 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1079 if (rq
->flags
& (REQ_BLOCK_PC
| REQ_PC
)) {
1081 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1082 printk("%02x ", rq
->cmd
[bit
]);
1087 EXPORT_SYMBOL(blk_dump_rq_flags
);
1089 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1091 struct bio_vec
*bv
, *bvprv
= NULL
;
1092 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1093 int high
, highprv
= 1;
1095 if (unlikely(!bio
->bi_io_vec
))
1098 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1099 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1100 bio_for_each_segment(bv
, bio
, i
) {
1102 * the trick here is making sure that a high page is never
1103 * considered part of another segment, since that might
1104 * change with the bounce page.
1106 high
= page_to_pfn(bv
->bv_page
) >= q
->bounce_pfn
;
1107 if (high
|| highprv
)
1108 goto new_hw_segment
;
1110 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1112 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1114 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1116 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1117 goto new_hw_segment
;
1119 seg_size
+= bv
->bv_len
;
1120 hw_seg_size
+= bv
->bv_len
;
1125 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1126 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1127 hw_seg_size
+= bv
->bv_len
;
1130 if (hw_seg_size
> bio
->bi_hw_front_size
)
1131 bio
->bi_hw_front_size
= hw_seg_size
;
1132 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1138 seg_size
= bv
->bv_len
;
1141 if (hw_seg_size
> bio
->bi_hw_back_size
)
1142 bio
->bi_hw_back_size
= hw_seg_size
;
1143 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1144 bio
->bi_hw_front_size
= hw_seg_size
;
1145 bio
->bi_phys_segments
= nr_phys_segs
;
1146 bio
->bi_hw_segments
= nr_hw_segs
;
1147 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1151 int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1154 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1157 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1159 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1163 * bio and nxt are contigous in memory, check if the queue allows
1164 * these two to be merged into one
1166 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1172 EXPORT_SYMBOL(blk_phys_contig_segment
);
1174 int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1177 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1178 blk_recount_segments(q
, bio
);
1179 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1180 blk_recount_segments(q
, nxt
);
1181 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1182 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1184 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1190 EXPORT_SYMBOL(blk_hw_contig_segment
);
1193 * map a request to scatterlist, return number of sg entries setup. Caller
1194 * must make sure sg can hold rq->nr_phys_segments entries
1196 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1198 struct bio_vec
*bvec
, *bvprv
;
1200 int nsegs
, i
, cluster
;
1203 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1206 * for each bio in rq
1209 rq_for_each_bio(bio
, rq
) {
1211 * for each segment in bio
1213 bio_for_each_segment(bvec
, bio
, i
) {
1214 int nbytes
= bvec
->bv_len
;
1216 if (bvprv
&& cluster
) {
1217 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1220 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1222 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1225 sg
[nsegs
- 1].length
+= nbytes
;
1228 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1229 sg
[nsegs
].page
= bvec
->bv_page
;
1230 sg
[nsegs
].length
= nbytes
;
1231 sg
[nsegs
].offset
= bvec
->bv_offset
;
1236 } /* segments in bio */
1242 EXPORT_SYMBOL(blk_rq_map_sg
);
1245 * the standard queue merge functions, can be overridden with device
1246 * specific ones if so desired
1249 static inline int ll_new_mergeable(request_queue_t
*q
,
1250 struct request
*req
,
1253 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1255 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1256 req
->flags
|= REQ_NOMERGE
;
1257 if (req
== q
->last_merge
)
1258 q
->last_merge
= NULL
;
1263 * A hw segment is just getting larger, bump just the phys
1266 req
->nr_phys_segments
+= nr_phys_segs
;
1270 static inline int ll_new_hw_segment(request_queue_t
*q
,
1271 struct request
*req
,
1274 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1275 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1277 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1278 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1279 req
->flags
|= REQ_NOMERGE
;
1280 if (req
== q
->last_merge
)
1281 q
->last_merge
= NULL
;
1286 * This will form the start of a new hw segment. Bump both
1289 req
->nr_hw_segments
+= nr_hw_segs
;
1290 req
->nr_phys_segments
+= nr_phys_segs
;
1294 static int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
,
1299 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1300 req
->flags
|= REQ_NOMERGE
;
1301 if (req
== q
->last_merge
)
1302 q
->last_merge
= NULL
;
1305 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1306 blk_recount_segments(q
, req
->biotail
);
1307 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1308 blk_recount_segments(q
, bio
);
1309 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1310 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1311 !BIOVEC_VIRT_OVERSIZE(len
)) {
1312 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1315 if (req
->nr_hw_segments
== 1)
1316 req
->bio
->bi_hw_front_size
= len
;
1317 if (bio
->bi_hw_segments
== 1)
1318 bio
->bi_hw_back_size
= len
;
1323 return ll_new_hw_segment(q
, req
, bio
);
1326 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1331 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1332 req
->flags
|= REQ_NOMERGE
;
1333 if (req
== q
->last_merge
)
1334 q
->last_merge
= NULL
;
1337 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1338 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1339 blk_recount_segments(q
, bio
);
1340 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1341 blk_recount_segments(q
, req
->bio
);
1342 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1343 !BIOVEC_VIRT_OVERSIZE(len
)) {
1344 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1347 if (bio
->bi_hw_segments
== 1)
1348 bio
->bi_hw_front_size
= len
;
1349 if (req
->nr_hw_segments
== 1)
1350 req
->biotail
->bi_hw_back_size
= len
;
1355 return ll_new_hw_segment(q
, req
, bio
);
1358 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1359 struct request
*next
)
1361 int total_phys_segments
= req
->nr_phys_segments
+next
->nr_phys_segments
;
1362 int total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1365 * First check if the either of the requests are re-queued
1366 * requests. Can't merge them if they are.
1368 if (req
->special
|| next
->special
)
1372 * Will it become to large?
1374 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1377 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1378 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1379 total_phys_segments
--;
1381 if (total_phys_segments
> q
->max_phys_segments
)
1384 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1385 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1386 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1388 * propagate the combined length to the end of the requests
1390 if (req
->nr_hw_segments
== 1)
1391 req
->bio
->bi_hw_front_size
= len
;
1392 if (next
->nr_hw_segments
== 1)
1393 next
->biotail
->bi_hw_back_size
= len
;
1394 total_hw_segments
--;
1397 if (total_hw_segments
> q
->max_hw_segments
)
1400 /* Merge is OK... */
1401 req
->nr_phys_segments
= total_phys_segments
;
1402 req
->nr_hw_segments
= total_hw_segments
;
1407 * "plug" the device if there are no outstanding requests: this will
1408 * force the transfer to start only after we have put all the requests
1411 * This is called with interrupts off and no requests on the queue and
1412 * with the queue lock held.
1414 void blk_plug_device(request_queue_t
*q
)
1416 WARN_ON(!irqs_disabled());
1419 * don't plug a stopped queue, it must be paired with blk_start_queue()
1420 * which will restart the queueing
1422 if (test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
))
1425 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1426 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1429 EXPORT_SYMBOL(blk_plug_device
);
1432 * remove the queue from the plugged list, if present. called with
1433 * queue lock held and interrupts disabled.
1435 int blk_remove_plug(request_queue_t
*q
)
1437 WARN_ON(!irqs_disabled());
1439 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1442 del_timer(&q
->unplug_timer
);
1446 EXPORT_SYMBOL(blk_remove_plug
);
1449 * remove the plug and let it rip..
1451 void __generic_unplug_device(request_queue_t
*q
)
1453 if (test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
))
1456 if (!blk_remove_plug(q
))
1460 * was plugged, fire request_fn if queue has stuff to do
1462 if (elv_next_request(q
))
1465 EXPORT_SYMBOL(__generic_unplug_device
);
1468 * generic_unplug_device - fire a request queue
1469 * @q: The &request_queue_t in question
1472 * Linux uses plugging to build bigger requests queues before letting
1473 * the device have at them. If a queue is plugged, the I/O scheduler
1474 * is still adding and merging requests on the queue. Once the queue
1475 * gets unplugged, the request_fn defined for the queue is invoked and
1476 * transfers started.
1478 void generic_unplug_device(request_queue_t
*q
)
1480 spin_lock_irq(q
->queue_lock
);
1481 __generic_unplug_device(q
);
1482 spin_unlock_irq(q
->queue_lock
);
1484 EXPORT_SYMBOL(generic_unplug_device
);
1486 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1489 request_queue_t
*q
= bdi
->unplug_io_data
;
1492 * devices don't necessarily have an ->unplug_fn defined
1498 static void blk_unplug_work(void *data
)
1500 request_queue_t
*q
= data
;
1505 static void blk_unplug_timeout(unsigned long data
)
1507 request_queue_t
*q
= (request_queue_t
*)data
;
1509 kblockd_schedule_work(&q
->unplug_work
);
1513 * blk_start_queue - restart a previously stopped queue
1514 * @q: The &request_queue_t in question
1517 * blk_start_queue() will clear the stop flag on the queue, and call
1518 * the request_fn for the queue if it was in a stopped state when
1519 * entered. Also see blk_stop_queue(). Queue lock must be held.
1521 void blk_start_queue(request_queue_t
*q
)
1523 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1526 * one level of recursion is ok and is much faster than kicking
1527 * the unplug handling
1529 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1531 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1534 kblockd_schedule_work(&q
->unplug_work
);
1538 EXPORT_SYMBOL(blk_start_queue
);
1541 * blk_stop_queue - stop a queue
1542 * @q: The &request_queue_t in question
1545 * The Linux block layer assumes that a block driver will consume all
1546 * entries on the request queue when the request_fn strategy is called.
1547 * Often this will not happen, because of hardware limitations (queue
1548 * depth settings). If a device driver gets a 'queue full' response,
1549 * or if it simply chooses not to queue more I/O at one point, it can
1550 * call this function to prevent the request_fn from being called until
1551 * the driver has signalled it's ready to go again. This happens by calling
1552 * blk_start_queue() to restart queue operations. Queue lock must be held.
1554 void blk_stop_queue(request_queue_t
*q
)
1557 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1559 EXPORT_SYMBOL(blk_stop_queue
);
1562 * blk_sync_queue - cancel any pending callbacks on a queue
1566 * The block layer may perform asynchronous callback activity
1567 * on a queue, such as calling the unplug function after a timeout.
1568 * A block device may call blk_sync_queue to ensure that any
1569 * such activity is cancelled, thus allowing it to release resources
1570 * the the callbacks might use. The caller must already have made sure
1571 * that its ->make_request_fn will not re-add plugging prior to calling
1575 void blk_sync_queue(struct request_queue
*q
)
1577 del_timer_sync(&q
->unplug_timer
);
1580 EXPORT_SYMBOL(blk_sync_queue
);
1583 * blk_run_queue - run a single device queue
1584 * @q: The queue to run
1586 void blk_run_queue(struct request_queue
*q
)
1588 unsigned long flags
;
1590 spin_lock_irqsave(q
->queue_lock
, flags
);
1593 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1595 EXPORT_SYMBOL(blk_run_queue
);
1598 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1599 * @q: the request queue to be released
1602 * blk_cleanup_queue is the pair to blk_init_queue() or
1603 * blk_queue_make_request(). It should be called when a request queue is
1604 * being released; typically when a block device is being de-registered.
1605 * Currently, its primary task it to free all the &struct request
1606 * structures that were allocated to the queue and the queue itself.
1609 * Hopefully the low level driver will have finished any
1610 * outstanding requests first...
1612 void blk_cleanup_queue(request_queue_t
* q
)
1614 struct request_list
*rl
= &q
->rq
;
1616 if (!atomic_dec_and_test(&q
->refcnt
))
1620 elevator_exit(q
->elevator
);
1625 mempool_destroy(rl
->rq_pool
);
1628 __blk_queue_free_tags(q
);
1630 blk_queue_ordered(q
, QUEUE_ORDERED_NONE
);
1632 kmem_cache_free(requestq_cachep
, q
);
1635 EXPORT_SYMBOL(blk_cleanup_queue
);
1637 static int blk_init_free_list(request_queue_t
*q
)
1639 struct request_list
*rl
= &q
->rq
;
1641 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1642 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1643 init_waitqueue_head(&rl
->wait
[READ
]);
1644 init_waitqueue_head(&rl
->wait
[WRITE
]);
1645 init_waitqueue_head(&rl
->drain
);
1647 rl
->rq_pool
= mempool_create(BLKDEV_MIN_RQ
, mempool_alloc_slab
, mempool_free_slab
, request_cachep
);
1655 static int __make_request(request_queue_t
*, struct bio
*);
1657 request_queue_t
*blk_alloc_queue(int gfp_mask
)
1659 request_queue_t
*q
= kmem_cache_alloc(requestq_cachep
, gfp_mask
);
1664 memset(q
, 0, sizeof(*q
));
1665 init_timer(&q
->unplug_timer
);
1666 atomic_set(&q
->refcnt
, 1);
1668 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1669 q
->backing_dev_info
.unplug_io_data
= q
;
1674 EXPORT_SYMBOL(blk_alloc_queue
);
1677 * blk_init_queue - prepare a request queue for use with a block device
1678 * @rfn: The function to be called to process requests that have been
1679 * placed on the queue.
1680 * @lock: Request queue spin lock
1683 * If a block device wishes to use the standard request handling procedures,
1684 * which sorts requests and coalesces adjacent requests, then it must
1685 * call blk_init_queue(). The function @rfn will be called when there
1686 * are requests on the queue that need to be processed. If the device
1687 * supports plugging, then @rfn may not be called immediately when requests
1688 * are available on the queue, but may be called at some time later instead.
1689 * Plugged queues are generally unplugged when a buffer belonging to one
1690 * of the requests on the queue is needed, or due to memory pressure.
1692 * @rfn is not required, or even expected, to remove all requests off the
1693 * queue, but only as many as it can handle at a time. If it does leave
1694 * requests on the queue, it is responsible for arranging that the requests
1695 * get dealt with eventually.
1697 * The queue spin lock must be held while manipulating the requests on the
1700 * Function returns a pointer to the initialized request queue, or NULL if
1701 * it didn't succeed.
1704 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1705 * when the block device is deactivated (such as at module unload).
1707 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1709 request_queue_t
*q
= blk_alloc_queue(GFP_KERNEL
);
1714 if (blk_init_free_list(q
))
1717 q
->request_fn
= rfn
;
1718 q
->back_merge_fn
= ll_back_merge_fn
;
1719 q
->front_merge_fn
= ll_front_merge_fn
;
1720 q
->merge_requests_fn
= ll_merge_requests_fn
;
1721 q
->prep_rq_fn
= NULL
;
1722 q
->unplug_fn
= generic_unplug_device
;
1723 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1724 q
->queue_lock
= lock
;
1726 blk_queue_segment_boundary(q
, 0xffffffff);
1728 blk_queue_make_request(q
, __make_request
);
1729 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1731 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1732 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1737 if (!elevator_init(q
, NULL
)) {
1738 blk_queue_congestion_threshold(q
);
1742 blk_cleanup_queue(q
);
1744 kmem_cache_free(requestq_cachep
, q
);
1748 EXPORT_SYMBOL(blk_init_queue
);
1750 int blk_get_queue(request_queue_t
*q
)
1752 if (!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
1753 atomic_inc(&q
->refcnt
);
1760 EXPORT_SYMBOL(blk_get_queue
);
1762 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
1764 elv_put_request(q
, rq
);
1765 mempool_free(rq
, q
->rq
.rq_pool
);
1768 static inline struct request
*blk_alloc_request(request_queue_t
*q
, int rw
,
1771 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1777 * first three bits are identical in rq->flags and bio->bi_rw,
1778 * see bio.h and blkdev.h
1782 if (!elv_set_request(q
, rq
, gfp_mask
))
1785 mempool_free(rq
, q
->rq
.rq_pool
);
1790 * ioc_batching returns true if the ioc is a valid batching request and
1791 * should be given priority access to a request.
1793 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
1799 * Make sure the process is able to allocate at least 1 request
1800 * even if the batch times out, otherwise we could theoretically
1803 return ioc
->nr_batch_requests
== q
->nr_batching
||
1804 (ioc
->nr_batch_requests
> 0
1805 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1809 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1810 * will cause the process to be a "batcher" on all queues in the system. This
1811 * is the behaviour we want though - once it gets a wakeup it should be given
1814 void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
1816 if (!ioc
|| ioc_batching(q
, ioc
))
1819 ioc
->nr_batch_requests
= q
->nr_batching
;
1820 ioc
->last_waited
= jiffies
;
1823 static void __freed_request(request_queue_t
*q
, int rw
)
1825 struct request_list
*rl
= &q
->rq
;
1827 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
1828 clear_queue_congested(q
, rw
);
1830 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
1832 if (waitqueue_active(&rl
->wait
[rw
]))
1833 wake_up(&rl
->wait
[rw
]);
1835 blk_clear_queue_full(q
, rw
);
1840 * A request has just been released. Account for it, update the full and
1841 * congestion status, wake up any waiters. Called under q->queue_lock.
1843 static void freed_request(request_queue_t
*q
, int rw
)
1845 struct request_list
*rl
= &q
->rq
;
1849 __freed_request(q
, rw
);
1851 if (unlikely(rl
->starved
[rw
^ 1]))
1852 __freed_request(q
, rw
^ 1);
1854 if (!rl
->count
[READ
] && !rl
->count
[WRITE
]) {
1856 if (unlikely(waitqueue_active(&rl
->drain
)))
1857 wake_up(&rl
->drain
);
1861 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1863 * Get a free request, queue_lock must not be held
1865 static struct request
*get_request(request_queue_t
*q
, int rw
, int gfp_mask
)
1867 struct request
*rq
= NULL
;
1868 struct request_list
*rl
= &q
->rq
;
1869 struct io_context
*ioc
= get_io_context(gfp_mask
);
1871 if (unlikely(test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
)))
1874 spin_lock_irq(q
->queue_lock
);
1875 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
1877 * The queue will fill after this allocation, so set it as
1878 * full, and mark this process as "batching". This process
1879 * will be allowed to complete a batch of requests, others
1882 if (!blk_queue_full(q
, rw
)) {
1883 ioc_set_batching(q
, ioc
);
1884 blk_set_queue_full(q
, rw
);
1888 switch (elv_may_queue(q
, rw
)) {
1891 case ELV_MQUEUE_MAY
:
1893 case ELV_MQUEUE_MUST
:
1897 if (blk_queue_full(q
, rw
) && !ioc_batching(q
, ioc
)) {
1899 * The queue is full and the allocating process is not a
1900 * "batcher", and not exempted by the IO scheduler
1902 spin_unlock_irq(q
->queue_lock
);
1908 rl
->starved
[rw
] = 0;
1909 if (rl
->count
[rw
] >= queue_congestion_on_threshold(q
))
1910 set_queue_congested(q
, rw
);
1911 spin_unlock_irq(q
->queue_lock
);
1913 rq
= blk_alloc_request(q
, rw
, gfp_mask
);
1916 * Allocation failed presumably due to memory. Undo anything
1917 * we might have messed up.
1919 * Allocating task should really be put onto the front of the
1920 * wait queue, but this is pretty rare.
1922 spin_lock_irq(q
->queue_lock
);
1923 freed_request(q
, rw
);
1926 * in the very unlikely event that allocation failed and no
1927 * requests for this direction was pending, mark us starved
1928 * so that freeing of a request in the other direction will
1929 * notice us. another possible fix would be to split the
1930 * rq mempool into READ and WRITE
1933 if (unlikely(rl
->count
[rw
] == 0))
1934 rl
->starved
[rw
] = 1;
1936 spin_unlock_irq(q
->queue_lock
);
1940 if (ioc_batching(q
, ioc
))
1941 ioc
->nr_batch_requests
--;
1946 put_io_context(ioc
);
1951 * No available requests for this queue, unplug the device and wait for some
1952 * requests to become available.
1954 static struct request
*get_request_wait(request_queue_t
*q
, int rw
)
1959 generic_unplug_device(q
);
1961 struct request_list
*rl
= &q
->rq
;
1963 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
1964 TASK_UNINTERRUPTIBLE
);
1966 rq
= get_request(q
, rw
, GFP_NOIO
);
1969 struct io_context
*ioc
;
1974 * After sleeping, we become a "batching" process and
1975 * will be able to allocate at least one request, and
1976 * up to a big batch of them for a small period time.
1977 * See ioc_batching, ioc_set_batching
1979 ioc
= get_io_context(GFP_NOIO
);
1980 ioc_set_batching(q
, ioc
);
1981 put_io_context(ioc
);
1983 finish_wait(&rl
->wait
[rw
], &wait
);
1989 struct request
*blk_get_request(request_queue_t
*q
, int rw
, int gfp_mask
)
1993 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1995 if (gfp_mask
& __GFP_WAIT
)
1996 rq
= get_request_wait(q
, rw
);
1998 rq
= get_request(q
, rw
, gfp_mask
);
2003 EXPORT_SYMBOL(blk_get_request
);
2006 * blk_requeue_request - put a request back on queue
2007 * @q: request queue where request should be inserted
2008 * @rq: request to be inserted
2011 * Drivers often keep queueing requests until the hardware cannot accept
2012 * more, when that condition happens we need to put the request back
2013 * on the queue. Must be called with queue lock held.
2015 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2017 if (blk_rq_tagged(rq
))
2018 blk_queue_end_tag(q
, rq
);
2020 elv_requeue_request(q
, rq
);
2023 EXPORT_SYMBOL(blk_requeue_request
);
2026 * blk_insert_request - insert a special request in to a request queue
2027 * @q: request queue where request should be inserted
2028 * @rq: request to be inserted
2029 * @at_head: insert request at head or tail of queue
2030 * @data: private data
2031 * @reinsert: true if request it a reinsertion of previously processed one
2034 * Many block devices need to execute commands asynchronously, so they don't
2035 * block the whole kernel from preemption during request execution. This is
2036 * accomplished normally by inserting aritficial requests tagged as
2037 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2038 * scheduled for actual execution by the request queue.
2040 * We have the option of inserting the head or the tail of the queue.
2041 * Typically we use the tail for new ioctls and so forth. We use the head
2042 * of the queue for things like a QUEUE_FULL message from a device, or a
2043 * host that is unable to accept a particular command.
2045 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2046 int at_head
, void *data
, int reinsert
)
2048 unsigned long flags
;
2051 * tell I/O scheduler that this isn't a regular read/write (ie it
2052 * must not attempt merges on this) and that it acts as a soft
2055 rq
->flags
|= REQ_SPECIAL
| REQ_SOFTBARRIER
;
2059 spin_lock_irqsave(q
->queue_lock
, flags
);
2062 * If command is tagged, release the tag
2065 blk_requeue_request(q
, rq
);
2067 int where
= ELEVATOR_INSERT_BACK
;
2070 where
= ELEVATOR_INSERT_FRONT
;
2072 if (blk_rq_tagged(rq
))
2073 blk_queue_end_tag(q
, rq
);
2075 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2076 __elv_add_request(q
, rq
, where
, 0);
2078 if (blk_queue_plugged(q
))
2079 __generic_unplug_device(q
);
2082 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2085 EXPORT_SYMBOL(blk_insert_request
);
2088 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2089 * @q: request queue where request should be inserted
2090 * @rw: READ or WRITE data
2091 * @ubuf: the user buffer
2092 * @len: length of user data
2095 * Data will be mapped directly for zero copy io, if possible. Otherwise
2096 * a kernel bounce buffer is used.
2098 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2099 * still in process context.
2101 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2102 * before being submitted to the device, as pages mapped may be out of
2103 * reach. It's the callers responsibility to make sure this happens. The
2104 * original bio must be passed back in to blk_rq_unmap_user() for proper
2107 struct request
*blk_rq_map_user(request_queue_t
*q
, int rw
, void __user
*ubuf
,
2110 unsigned long uaddr
;
2114 if (len
> (q
->max_sectors
<< 9))
2115 return ERR_PTR(-EINVAL
);
2116 if ((!len
&& ubuf
) || (len
&& !ubuf
))
2117 return ERR_PTR(-EINVAL
);
2119 rq
= blk_get_request(q
, rw
, __GFP_WAIT
);
2121 return ERR_PTR(-ENOMEM
);
2124 * if alignment requirement is satisfied, map in user pages for
2125 * direct dma. else, set up kernel bounce buffers
2127 uaddr
= (unsigned long) ubuf
;
2128 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2129 bio
= bio_map_user(q
, NULL
, uaddr
, len
, rw
== READ
);
2131 bio
= bio_copy_user(q
, uaddr
, len
, rw
== READ
);
2134 rq
->bio
= rq
->biotail
= bio
;
2135 blk_rq_bio_prep(q
, rq
, bio
);
2137 rq
->buffer
= rq
->data
= NULL
;
2143 * bio is the err-ptr
2145 blk_put_request(rq
);
2146 return (struct request
*) bio
;
2149 EXPORT_SYMBOL(blk_rq_map_user
);
2152 * blk_rq_unmap_user - unmap a request with user data
2153 * @rq: request to be unmapped
2154 * @bio: bio for the request
2155 * @ulen: length of user buffer
2158 * Unmap a request previously mapped by blk_rq_map_user().
2160 int blk_rq_unmap_user(struct request
*rq
, struct bio
*bio
, unsigned int ulen
)
2165 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2166 bio_unmap_user(bio
);
2168 ret
= bio_uncopy_user(bio
);
2171 blk_put_request(rq
);
2175 EXPORT_SYMBOL(blk_rq_unmap_user
);
2178 * blk_execute_rq - insert a request into queue for execution
2179 * @q: queue to insert the request in
2180 * @bd_disk: matching gendisk
2181 * @rq: request to insert
2184 * Insert a fully prepared request at the back of the io scheduler queue
2187 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2190 DECLARE_COMPLETION(wait
);
2191 char sense
[SCSI_SENSE_BUFFERSIZE
];
2194 rq
->rq_disk
= bd_disk
;
2197 * we need an extra reference to the request, so we can look at
2198 * it after io completion
2203 memset(sense
, 0, sizeof(sense
));
2208 rq
->flags
|= REQ_NOMERGE
;
2209 rq
->waiting
= &wait
;
2210 rq
->end_io
= blk_end_sync_rq
;
2211 elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 1);
2212 generic_unplug_device(q
);
2213 wait_for_completion(&wait
);
2222 EXPORT_SYMBOL(blk_execute_rq
);
2225 * blkdev_issue_flush - queue a flush
2226 * @bdev: blockdev to issue flush for
2227 * @error_sector: error sector
2230 * Issue a flush for the block device in question. Caller can supply
2231 * room for storing the error offset in case of a flush error, if they
2232 * wish to. Caller must run wait_for_completion() on its own.
2234 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2238 if (bdev
->bd_disk
== NULL
)
2241 q
= bdev_get_queue(bdev
);
2244 if (!q
->issue_flush_fn
)
2247 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2250 EXPORT_SYMBOL(blkdev_issue_flush
);
2253 * blkdev_scsi_issue_flush_fn - issue flush for SCSI devices
2256 * @error_sector: error offset
2259 * Devices understanding the SCSI command set, can use this function as
2260 * a helper for issuing a cache flush. Note: driver is required to store
2261 * the error offset (in case of error flushing) in ->sector of struct
2264 int blkdev_scsi_issue_flush_fn(request_queue_t
*q
, struct gendisk
*disk
,
2265 sector_t
*error_sector
)
2267 struct request
*rq
= blk_get_request(q
, WRITE
, __GFP_WAIT
);
2270 rq
->flags
|= REQ_BLOCK_PC
| REQ_SOFTBARRIER
;
2272 memset(rq
->cmd
, 0, sizeof(rq
->cmd
));
2277 rq
->timeout
= 60 * HZ
;
2279 ret
= blk_execute_rq(q
, disk
, rq
);
2281 if (ret
&& error_sector
)
2282 *error_sector
= rq
->sector
;
2284 blk_put_request(rq
);
2288 EXPORT_SYMBOL(blkdev_scsi_issue_flush_fn
);
2290 void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2292 int rw
= rq_data_dir(rq
);
2294 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2298 __disk_stat_add(rq
->rq_disk
, read_sectors
, nr_sectors
);
2300 __disk_stat_inc(rq
->rq_disk
, read_merges
);
2301 } else if (rw
== WRITE
) {
2302 __disk_stat_add(rq
->rq_disk
, write_sectors
, nr_sectors
);
2304 __disk_stat_inc(rq
->rq_disk
, write_merges
);
2307 disk_round_stats(rq
->rq_disk
);
2308 rq
->rq_disk
->in_flight
++;
2313 * add-request adds a request to the linked list.
2314 * queue lock is held and interrupts disabled, as we muck with the
2315 * request queue list.
2317 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2319 drive_stat_acct(req
, req
->nr_sectors
, 1);
2322 q
->activity_fn(q
->activity_data
, rq_data_dir(req
));
2325 * elevator indicated where it wants this request to be
2326 * inserted at elevator_merge time
2328 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2332 * disk_round_stats() - Round off the performance stats on a struct
2335 * The average IO queue length and utilisation statistics are maintained
2336 * by observing the current state of the queue length and the amount of
2337 * time it has been in this state for.
2339 * Normally, that accounting is done on IO completion, but that can result
2340 * in more than a second's worth of IO being accounted for within any one
2341 * second, leading to >100% utilisation. To deal with that, we call this
2342 * function to do a round-off before returning the results when reading
2343 * /proc/diskstats. This accounts immediately for all queue usage up to
2344 * the current jiffies and restarts the counters again.
2346 void disk_round_stats(struct gendisk
*disk
)
2348 unsigned long now
= jiffies
;
2350 __disk_stat_add(disk
, time_in_queue
,
2351 disk
->in_flight
* (now
- disk
->stamp
));
2354 if (disk
->in_flight
)
2355 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp_idle
));
2356 disk
->stamp_idle
= now
;
2360 * queue lock must be held
2362 static void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2364 struct request_list
*rl
= req
->rl
;
2368 if (unlikely(--req
->ref_count
))
2371 req
->rq_status
= RQ_INACTIVE
;
2376 * Request may not have originated from ll_rw_blk. if not,
2377 * it didn't come out of our reserved rq pools
2380 int rw
= rq_data_dir(req
);
2382 elv_completed_request(q
, req
);
2384 BUG_ON(!list_empty(&req
->queuelist
));
2386 blk_free_request(q
, req
);
2387 freed_request(q
, rw
);
2391 void blk_put_request(struct request
*req
)
2394 * if req->rl isn't set, this request didnt originate from the
2395 * block layer, so it's safe to just disregard it
2398 unsigned long flags
;
2399 request_queue_t
*q
= req
->q
;
2401 spin_lock_irqsave(q
->queue_lock
, flags
);
2402 __blk_put_request(q
, req
);
2403 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2407 EXPORT_SYMBOL(blk_put_request
);
2410 * blk_end_sync_rq - executes a completion event on a request
2411 * @rq: request to complete
2413 void blk_end_sync_rq(struct request
*rq
)
2415 struct completion
*waiting
= rq
->waiting
;
2418 __blk_put_request(rq
->q
, rq
);
2421 * complete last, if this is a stack request the process (and thus
2422 * the rq pointer) could be invalid right after this complete()
2426 EXPORT_SYMBOL(blk_end_sync_rq
);
2429 * blk_congestion_wait - wait for a queue to become uncongested
2430 * @rw: READ or WRITE
2431 * @timeout: timeout in jiffies
2433 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2434 * If no queues are congested then just wait for the next request to be
2437 long blk_congestion_wait(int rw
, long timeout
)
2441 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
2443 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
2444 ret
= io_schedule_timeout(timeout
);
2445 finish_wait(wqh
, &wait
);
2449 EXPORT_SYMBOL(blk_congestion_wait
);
2452 * Has to be called with the request spinlock acquired
2454 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2455 struct request
*next
)
2457 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2463 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2466 if (rq_data_dir(req
) != rq_data_dir(next
)
2467 || req
->rq_disk
!= next
->rq_disk
2468 || next
->waiting
|| next
->special
)
2472 * If we are allowed to merge, then append bio list
2473 * from next to rq and release next. merge_requests_fn
2474 * will have updated segment counts, update sector
2477 if (!q
->merge_requests_fn(q
, req
, next
))
2481 * At this point we have either done a back merge
2482 * or front merge. We need the smaller start_time of
2483 * the merged requests to be the current request
2484 * for accounting purposes.
2486 if (time_after(req
->start_time
, next
->start_time
))
2487 req
->start_time
= next
->start_time
;
2489 req
->biotail
->bi_next
= next
->bio
;
2490 req
->biotail
= next
->biotail
;
2492 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2494 elv_merge_requests(q
, req
, next
);
2497 disk_round_stats(req
->rq_disk
);
2498 req
->rq_disk
->in_flight
--;
2501 __blk_put_request(q
, next
);
2505 static inline int attempt_back_merge(request_queue_t
*q
, struct request
*rq
)
2507 struct request
*next
= elv_latter_request(q
, rq
);
2510 return attempt_merge(q
, rq
, next
);
2515 static inline int attempt_front_merge(request_queue_t
*q
, struct request
*rq
)
2517 struct request
*prev
= elv_former_request(q
, rq
);
2520 return attempt_merge(q
, prev
, rq
);
2526 * blk_attempt_remerge - attempt to remerge active head with next request
2527 * @q: The &request_queue_t belonging to the device
2528 * @rq: The head request (usually)
2531 * For head-active devices, the queue can easily be unplugged so quickly
2532 * that proper merging is not done on the front request. This may hurt
2533 * performance greatly for some devices. The block layer cannot safely
2534 * do merging on that first request for these queues, but the driver can
2535 * call this function and make it happen any way. Only the driver knows
2536 * when it is safe to do so.
2538 void blk_attempt_remerge(request_queue_t
*q
, struct request
*rq
)
2540 unsigned long flags
;
2542 spin_lock_irqsave(q
->queue_lock
, flags
);
2543 attempt_back_merge(q
, rq
);
2544 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2547 EXPORT_SYMBOL(blk_attempt_remerge
);
2550 * Non-locking blk_attempt_remerge variant.
2552 void __blk_attempt_remerge(request_queue_t
*q
, struct request
*rq
)
2554 attempt_back_merge(q
, rq
);
2557 EXPORT_SYMBOL(__blk_attempt_remerge
);
2559 static int __make_request(request_queue_t
*q
, struct bio
*bio
)
2561 struct request
*req
, *freereq
= NULL
;
2562 int el_ret
, rw
, nr_sectors
, cur_nr_sectors
, barrier
, err
;
2565 sector
= bio
->bi_sector
;
2566 nr_sectors
= bio_sectors(bio
);
2567 cur_nr_sectors
= bio_cur_sectors(bio
);
2569 rw
= bio_data_dir(bio
);
2572 * low level driver can indicate that it wants pages above a
2573 * certain limit bounced to low memory (ie for highmem, or even
2574 * ISA dma in theory)
2576 blk_queue_bounce(q
, &bio
);
2578 spin_lock_prefetch(q
->queue_lock
);
2580 barrier
= bio_barrier(bio
);
2581 if (barrier
&& (q
->ordered
== QUEUE_ORDERED_NONE
)) {
2587 spin_lock_irq(q
->queue_lock
);
2589 if (elv_queue_empty(q
)) {
2596 el_ret
= elv_merge(q
, &req
, bio
);
2598 case ELEVATOR_BACK_MERGE
:
2599 BUG_ON(!rq_mergeable(req
));
2601 if (!q
->back_merge_fn(q
, req
, bio
))
2604 req
->biotail
->bi_next
= bio
;
2606 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2607 drive_stat_acct(req
, nr_sectors
, 0);
2608 if (!attempt_back_merge(q
, req
))
2609 elv_merged_request(q
, req
);
2612 case ELEVATOR_FRONT_MERGE
:
2613 BUG_ON(!rq_mergeable(req
));
2615 if (!q
->front_merge_fn(q
, req
, bio
))
2618 bio
->bi_next
= req
->bio
;
2622 * may not be valid. if the low level driver said
2623 * it didn't need a bounce buffer then it better
2624 * not touch req->buffer either...
2626 req
->buffer
= bio_data(bio
);
2627 req
->current_nr_sectors
= cur_nr_sectors
;
2628 req
->hard_cur_sectors
= cur_nr_sectors
;
2629 req
->sector
= req
->hard_sector
= sector
;
2630 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2631 drive_stat_acct(req
, nr_sectors
, 0);
2632 if (!attempt_front_merge(q
, req
))
2633 elv_merged_request(q
, req
);
2637 * elevator says don't/can't merge. get new request
2639 case ELEVATOR_NO_MERGE
:
2643 printk("elevator returned crap (%d)\n", el_ret
);
2648 * Grab a free request from the freelist - if that is empty, check
2649 * if we are doing read ahead and abort instead of blocking for
2657 spin_unlock_irq(q
->queue_lock
);
2658 if ((freereq
= get_request(q
, rw
, GFP_ATOMIC
)) == NULL
) {
2663 if (bio_rw_ahead(bio
))
2666 freereq
= get_request_wait(q
, rw
);
2671 req
->flags
|= REQ_CMD
;
2674 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2676 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2677 req
->flags
|= REQ_FAILFAST
;
2680 * REQ_BARRIER implies no merging, but lets make it explicit
2683 req
->flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2686 req
->hard_sector
= req
->sector
= sector
;
2687 req
->hard_nr_sectors
= req
->nr_sectors
= nr_sectors
;
2688 req
->current_nr_sectors
= req
->hard_cur_sectors
= cur_nr_sectors
;
2689 req
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2690 req
->nr_hw_segments
= bio_hw_segments(q
, bio
);
2691 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2692 req
->waiting
= NULL
;
2693 req
->bio
= req
->biotail
= bio
;
2694 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2695 req
->start_time
= jiffies
;
2697 add_request(q
, req
);
2700 __blk_put_request(q
, freereq
);
2702 __generic_unplug_device(q
);
2704 spin_unlock_irq(q
->queue_lock
);
2708 bio_endio(bio
, nr_sectors
<< 9, err
);
2713 * If bio->bi_dev is a partition, remap the location
2715 static inline void blk_partition_remap(struct bio
*bio
)
2717 struct block_device
*bdev
= bio
->bi_bdev
;
2719 if (bdev
!= bdev
->bd_contains
) {
2720 struct hd_struct
*p
= bdev
->bd_part
;
2722 switch (bio
->bi_rw
) {
2724 p
->read_sectors
+= bio_sectors(bio
);
2728 p
->write_sectors
+= bio_sectors(bio
);
2732 bio
->bi_sector
+= p
->start_sect
;
2733 bio
->bi_bdev
= bdev
->bd_contains
;
2737 void blk_finish_queue_drain(request_queue_t
*q
)
2739 struct request_list
*rl
= &q
->rq
;
2742 spin_lock_irq(q
->queue_lock
);
2743 clear_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
);
2745 while (!list_empty(&q
->drain_list
)) {
2746 rq
= list_entry_rq(q
->drain_list
.next
);
2748 list_del_init(&rq
->queuelist
);
2749 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 1);
2752 spin_unlock_irq(q
->queue_lock
);
2754 wake_up(&rl
->wait
[0]);
2755 wake_up(&rl
->wait
[1]);
2756 wake_up(&rl
->drain
);
2759 static int wait_drain(request_queue_t
*q
, struct request_list
*rl
, int dispatch
)
2761 int wait
= rl
->count
[READ
] + rl
->count
[WRITE
];
2764 wait
+= !list_empty(&q
->queue_head
);
2770 * We rely on the fact that only requests allocated through blk_alloc_request()
2771 * have io scheduler private data structures associated with them. Any other
2772 * type of request (allocated on stack or through kmalloc()) should not go
2773 * to the io scheduler core, but be attached to the queue head instead.
2775 void blk_wait_queue_drained(request_queue_t
*q
, int wait_dispatch
)
2777 struct request_list
*rl
= &q
->rq
;
2780 spin_lock_irq(q
->queue_lock
);
2781 set_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
);
2783 while (wait_drain(q
, rl
, wait_dispatch
)) {
2784 prepare_to_wait(&rl
->drain
, &wait
, TASK_UNINTERRUPTIBLE
);
2786 if (wait_drain(q
, rl
, wait_dispatch
)) {
2787 __generic_unplug_device(q
);
2788 spin_unlock_irq(q
->queue_lock
);
2790 spin_lock_irq(q
->queue_lock
);
2793 finish_wait(&rl
->drain
, &wait
);
2796 spin_unlock_irq(q
->queue_lock
);
2800 * block waiting for the io scheduler being started again.
2802 static inline void block_wait_queue_running(request_queue_t
*q
)
2806 while (test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
)) {
2807 struct request_list
*rl
= &q
->rq
;
2809 prepare_to_wait_exclusive(&rl
->drain
, &wait
,
2810 TASK_UNINTERRUPTIBLE
);
2813 * re-check the condition. avoids using prepare_to_wait()
2814 * in the fast path (queue is running)
2816 if (test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
))
2819 finish_wait(&rl
->drain
, &wait
);
2823 static void handle_bad_sector(struct bio
*bio
)
2825 char b
[BDEVNAME_SIZE
];
2827 printk(KERN_INFO
"attempt to access beyond end of device\n");
2828 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
2829 bdevname(bio
->bi_bdev
, b
),
2831 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
2832 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
2834 set_bit(BIO_EOF
, &bio
->bi_flags
);
2838 * generic_make_request: hand a buffer to its device driver for I/O
2839 * @bio: The bio describing the location in memory and on the device.
2841 * generic_make_request() is used to make I/O requests of block
2842 * devices. It is passed a &struct bio, which describes the I/O that needs
2845 * generic_make_request() does not return any status. The
2846 * success/failure status of the request, along with notification of
2847 * completion, is delivered asynchronously through the bio->bi_end_io
2848 * function described (one day) else where.
2850 * The caller of generic_make_request must make sure that bi_io_vec
2851 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2852 * set to describe the device address, and the
2853 * bi_end_io and optionally bi_private are set to describe how
2854 * completion notification should be signaled.
2856 * generic_make_request and the drivers it calls may use bi_next if this
2857 * bio happens to be merged with someone else, and may change bi_dev and
2858 * bi_sector for remaps as it sees fit. So the values of these fields
2859 * should NOT be depended on after the call to generic_make_request.
2861 void generic_make_request(struct bio
*bio
)
2865 int ret
, nr_sectors
= bio_sectors(bio
);
2868 /* Test device or partition size, when known. */
2869 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
2871 sector_t sector
= bio
->bi_sector
;
2873 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2875 * This may well happen - the kernel calls bread()
2876 * without checking the size of the device, e.g., when
2877 * mounting a device.
2879 handle_bad_sector(bio
);
2885 * Resolve the mapping until finished. (drivers are
2886 * still free to implement/resolve their own stacking
2887 * by explicitly returning 0)
2889 * NOTE: we don't repeat the blk_size check for each new device.
2890 * Stacking drivers are expected to know what they are doing.
2893 char b
[BDEVNAME_SIZE
];
2895 q
= bdev_get_queue(bio
->bi_bdev
);
2898 "generic_make_request: Trying to access "
2899 "nonexistent block-device %s (%Lu)\n",
2900 bdevname(bio
->bi_bdev
, b
),
2901 (long long) bio
->bi_sector
);
2903 bio_endio(bio
, bio
->bi_size
, -EIO
);
2907 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
2908 printk("bio too big device %s (%u > %u)\n",
2909 bdevname(bio
->bi_bdev
, b
),
2915 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))
2918 block_wait_queue_running(q
);
2921 * If this device has partitions, remap block n
2922 * of partition p to block n+start(p) of the disk.
2924 blk_partition_remap(bio
);
2926 ret
= q
->make_request_fn(q
, bio
);
2930 EXPORT_SYMBOL(generic_make_request
);
2933 * submit_bio: submit a bio to the block device layer for I/O
2934 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2935 * @bio: The &struct bio which describes the I/O
2937 * submit_bio() is very similar in purpose to generic_make_request(), and
2938 * uses that function to do most of the work. Both are fairly rough
2939 * interfaces, @bio must be presetup and ready for I/O.
2942 void submit_bio(int rw
, struct bio
*bio
)
2944 int count
= bio_sectors(bio
);
2946 BIO_BUG_ON(!bio
->bi_size
);
2947 BIO_BUG_ON(!bio
->bi_io_vec
);
2950 mod_page_state(pgpgout
, count
);
2952 mod_page_state(pgpgin
, count
);
2954 if (unlikely(block_dump
)) {
2955 char b
[BDEVNAME_SIZE
];
2956 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
2957 current
->comm
, current
->pid
,
2958 (rw
& WRITE
) ? "WRITE" : "READ",
2959 (unsigned long long)bio
->bi_sector
,
2960 bdevname(bio
->bi_bdev
,b
));
2963 generic_make_request(bio
);
2966 EXPORT_SYMBOL(submit_bio
);
2968 void blk_recalc_rq_segments(struct request
*rq
)
2970 struct bio
*bio
, *prevbio
= NULL
;
2971 int nr_phys_segs
, nr_hw_segs
;
2972 unsigned int phys_size
, hw_size
;
2973 request_queue_t
*q
= rq
->q
;
2978 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
2979 rq_for_each_bio(bio
, rq
) {
2980 /* Force bio hw/phys segs to be recalculated. */
2981 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
2983 nr_phys_segs
+= bio_phys_segments(q
, bio
);
2984 nr_hw_segs
+= bio_hw_segments(q
, bio
);
2986 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
2987 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
2989 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
2990 pseg
<= q
->max_segment_size
) {
2992 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
2996 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
2997 hseg
<= q
->max_segment_size
) {
2999 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3006 rq
->nr_phys_segments
= nr_phys_segs
;
3007 rq
->nr_hw_segments
= nr_hw_segs
;
3010 void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3012 if (blk_fs_request(rq
)) {
3013 rq
->hard_sector
+= nsect
;
3014 rq
->hard_nr_sectors
-= nsect
;
3017 * Move the I/O submission pointers ahead if required.
3019 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3020 (rq
->sector
<= rq
->hard_sector
)) {
3021 rq
->sector
= rq
->hard_sector
;
3022 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3023 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3024 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3025 rq
->buffer
= bio_data(rq
->bio
);
3029 * if total number of sectors is less than the first segment
3030 * size, something has gone terribly wrong
3032 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3033 printk("blk: request botched\n");
3034 rq
->nr_sectors
= rq
->current_nr_sectors
;
3039 static int __end_that_request_first(struct request
*req
, int uptodate
,
3042 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3046 * extend uptodate bool to allow < 0 value to be direct io error
3049 if (end_io_error(uptodate
))
3050 error
= !uptodate
? -EIO
: uptodate
;
3053 * for a REQ_BLOCK_PC request, we want to carry any eventual
3054 * sense key with us all the way through
3056 if (!blk_pc_request(req
))
3060 if (blk_fs_request(req
) && !(req
->flags
& REQ_QUIET
))
3061 printk("end_request: I/O error, dev %s, sector %llu\n",
3062 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3063 (unsigned long long)req
->sector
);
3066 total_bytes
= bio_nbytes
= 0;
3067 while ((bio
= req
->bio
) != NULL
) {
3070 if (nr_bytes
>= bio
->bi_size
) {
3071 req
->bio
= bio
->bi_next
;
3072 nbytes
= bio
->bi_size
;
3073 bio_endio(bio
, nbytes
, error
);
3077 int idx
= bio
->bi_idx
+ next_idx
;
3079 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3080 blk_dump_rq_flags(req
, "__end_that");
3081 printk("%s: bio idx %d >= vcnt %d\n",
3083 bio
->bi_idx
, bio
->bi_vcnt
);
3087 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3088 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3091 * not a complete bvec done
3093 if (unlikely(nbytes
> nr_bytes
)) {
3094 bio_nbytes
+= nr_bytes
;
3095 total_bytes
+= nr_bytes
;
3100 * advance to the next vector
3103 bio_nbytes
+= nbytes
;
3106 total_bytes
+= nbytes
;
3109 if ((bio
= req
->bio
)) {
3111 * end more in this run, or just return 'not-done'
3113 if (unlikely(nr_bytes
<= 0))
3125 * if the request wasn't completed, update state
3128 bio_endio(bio
, bio_nbytes
, error
);
3129 bio
->bi_idx
+= next_idx
;
3130 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3131 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3134 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3135 blk_recalc_rq_segments(req
);
3140 * end_that_request_first - end I/O on a request
3141 * @req: the request being processed
3142 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3143 * @nr_sectors: number of sectors to end I/O on
3146 * Ends I/O on a number of sectors attached to @req, and sets it up
3147 * for the next range of segments (if any) in the cluster.
3150 * 0 - we are done with this request, call end_that_request_last()
3151 * 1 - still buffers pending for this request
3153 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3155 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3158 EXPORT_SYMBOL(end_that_request_first
);
3161 * end_that_request_chunk - end I/O on a request
3162 * @req: the request being processed
3163 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3164 * @nr_bytes: number of bytes to complete
3167 * Ends I/O on a number of bytes attached to @req, and sets it up
3168 * for the next range of segments (if any). Like end_that_request_first(),
3169 * but deals with bytes instead of sectors.
3172 * 0 - we are done with this request, call end_that_request_last()
3173 * 1 - still buffers pending for this request
3175 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3177 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3180 EXPORT_SYMBOL(end_that_request_chunk
);
3183 * queue lock must be held
3185 void end_that_request_last(struct request
*req
)
3187 struct gendisk
*disk
= req
->rq_disk
;
3189 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3190 laptop_io_completion();
3192 if (disk
&& blk_fs_request(req
)) {
3193 unsigned long duration
= jiffies
- req
->start_time
;
3194 switch (rq_data_dir(req
)) {
3196 __disk_stat_inc(disk
, writes
);
3197 __disk_stat_add(disk
, write_ticks
, duration
);
3200 __disk_stat_inc(disk
, reads
);
3201 __disk_stat_add(disk
, read_ticks
, duration
);
3204 disk_round_stats(disk
);
3210 __blk_put_request(req
->q
, req
);
3213 EXPORT_SYMBOL(end_that_request_last
);
3215 void end_request(struct request
*req
, int uptodate
)
3217 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3218 add_disk_randomness(req
->rq_disk
);
3219 blkdev_dequeue_request(req
);
3220 end_that_request_last(req
);
3224 EXPORT_SYMBOL(end_request
);
3226 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3228 /* first three bits are identical in rq->flags and bio->bi_rw */
3229 rq
->flags
|= (bio
->bi_rw
& 7);
3231 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3232 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3233 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3234 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3235 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3236 rq
->buffer
= bio_data(bio
);
3238 rq
->bio
= rq
->biotail
= bio
;
3241 EXPORT_SYMBOL(blk_rq_bio_prep
);
3243 int kblockd_schedule_work(struct work_struct
*work
)
3245 return queue_work(kblockd_workqueue
, work
);
3248 EXPORT_SYMBOL(kblockd_schedule_work
);
3250 void kblockd_flush(void)
3252 flush_workqueue(kblockd_workqueue
);
3254 EXPORT_SYMBOL(kblockd_flush
);
3256 int __init
blk_dev_init(void)
3258 kblockd_workqueue
= create_workqueue("kblockd");
3259 if (!kblockd_workqueue
)
3260 panic("Failed to create kblockd\n");
3262 request_cachep
= kmem_cache_create("blkdev_requests",
3263 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3265 requestq_cachep
= kmem_cache_create("blkdev_queue",
3266 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3268 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3269 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3271 blk_max_low_pfn
= max_low_pfn
;
3272 blk_max_pfn
= max_pfn
;
3278 * IO Context helper functions
3280 void put_io_context(struct io_context
*ioc
)
3285 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3287 if (atomic_dec_and_test(&ioc
->refcount
)) {
3288 if (ioc
->aic
&& ioc
->aic
->dtor
)
3289 ioc
->aic
->dtor(ioc
->aic
);
3290 if (ioc
->cic
&& ioc
->cic
->dtor
)
3291 ioc
->cic
->dtor(ioc
->cic
);
3293 kmem_cache_free(iocontext_cachep
, ioc
);
3296 EXPORT_SYMBOL(put_io_context
);
3298 /* Called by the exitting task */
3299 void exit_io_context(void)
3301 unsigned long flags
;
3302 struct io_context
*ioc
;
3304 local_irq_save(flags
);
3305 ioc
= current
->io_context
;
3306 current
->io_context
= NULL
;
3307 local_irq_restore(flags
);
3309 if (ioc
->aic
&& ioc
->aic
->exit
)
3310 ioc
->aic
->exit(ioc
->aic
);
3311 if (ioc
->cic
&& ioc
->cic
->exit
)
3312 ioc
->cic
->exit(ioc
->cic
);
3314 put_io_context(ioc
);
3318 * If the current task has no IO context then create one and initialise it.
3319 * If it does have a context, take a ref on it.
3321 * This is always called in the context of the task which submitted the I/O.
3322 * But weird things happen, so we disable local interrupts to ensure exclusive
3323 * access to *current.
3325 struct io_context
*get_io_context(int gfp_flags
)
3327 struct task_struct
*tsk
= current
;
3328 unsigned long flags
;
3329 struct io_context
*ret
;
3331 local_irq_save(flags
);
3332 ret
= tsk
->io_context
;
3336 local_irq_restore(flags
);
3338 ret
= kmem_cache_alloc(iocontext_cachep
, gfp_flags
);
3340 atomic_set(&ret
->refcount
, 1);
3341 ret
->pid
= tsk
->pid
;
3342 ret
->last_waited
= jiffies
; /* doesn't matter... */
3343 ret
->nr_batch_requests
= 0; /* because this is 0 */
3346 spin_lock_init(&ret
->lock
);
3348 local_irq_save(flags
);
3351 * very unlikely, someone raced with us in setting up the task
3352 * io context. free new context and just grab a reference.
3354 if (!tsk
->io_context
)
3355 tsk
->io_context
= ret
;
3357 kmem_cache_free(iocontext_cachep
, ret
);
3358 ret
= tsk
->io_context
;
3362 atomic_inc(&ret
->refcount
);
3363 local_irq_restore(flags
);
3368 EXPORT_SYMBOL(get_io_context
);
3370 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3372 struct io_context
*src
= *psrc
;
3373 struct io_context
*dst
= *pdst
;
3376 BUG_ON(atomic_read(&src
->refcount
) == 0);
3377 atomic_inc(&src
->refcount
);
3378 put_io_context(dst
);
3382 EXPORT_SYMBOL(copy_io_context
);
3384 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3386 struct io_context
*temp
;
3391 EXPORT_SYMBOL(swap_io_context
);
3396 struct queue_sysfs_entry
{
3397 struct attribute attr
;
3398 ssize_t (*show
)(struct request_queue
*, char *);
3399 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3403 queue_var_show(unsigned int var
, char *page
)
3405 return sprintf(page
, "%d\n", var
);
3409 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3411 char *p
= (char *) page
;
3413 *var
= simple_strtoul(p
, &p
, 10);
3417 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3419 return queue_var_show(q
->nr_requests
, (page
));
3423 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3425 struct request_list
*rl
= &q
->rq
;
3427 int ret
= queue_var_store(&q
->nr_requests
, page
, count
);
3428 if (q
->nr_requests
< BLKDEV_MIN_RQ
)
3429 q
->nr_requests
= BLKDEV_MIN_RQ
;
3430 blk_queue_congestion_threshold(q
);
3432 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3433 set_queue_congested(q
, READ
);
3434 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3435 clear_queue_congested(q
, READ
);
3437 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3438 set_queue_congested(q
, WRITE
);
3439 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3440 clear_queue_congested(q
, WRITE
);
3442 if (rl
->count
[READ
] >= q
->nr_requests
) {
3443 blk_set_queue_full(q
, READ
);
3444 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3445 blk_clear_queue_full(q
, READ
);
3446 wake_up(&rl
->wait
[READ
]);
3449 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3450 blk_set_queue_full(q
, WRITE
);
3451 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3452 blk_clear_queue_full(q
, WRITE
);
3453 wake_up(&rl
->wait
[WRITE
]);
3458 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3460 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3462 return queue_var_show(ra_kb
, (page
));
3466 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3468 unsigned long ra_kb
;
3469 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3471 spin_lock_irq(q
->queue_lock
);
3472 if (ra_kb
> (q
->max_sectors
>> 1))
3473 ra_kb
= (q
->max_sectors
>> 1);
3475 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3476 spin_unlock_irq(q
->queue_lock
);
3481 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3483 int max_sectors_kb
= q
->max_sectors
>> 1;
3485 return queue_var_show(max_sectors_kb
, (page
));
3489 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3491 unsigned long max_sectors_kb
,
3492 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3493 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3494 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3497 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3500 * Take the queue lock to update the readahead and max_sectors
3501 * values synchronously:
3503 spin_lock_irq(q
->queue_lock
);
3505 * Trim readahead window as well, if necessary:
3507 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3508 if (ra_kb
> max_sectors_kb
)
3509 q
->backing_dev_info
.ra_pages
=
3510 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3512 q
->max_sectors
= max_sectors_kb
<< 1;
3513 spin_unlock_irq(q
->queue_lock
);
3518 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3520 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3522 return queue_var_show(max_hw_sectors_kb
, (page
));
3526 static struct queue_sysfs_entry queue_requests_entry
= {
3527 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
3528 .show
= queue_requests_show
,
3529 .store
= queue_requests_store
,
3532 static struct queue_sysfs_entry queue_ra_entry
= {
3533 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
3534 .show
= queue_ra_show
,
3535 .store
= queue_ra_store
,
3538 static struct queue_sysfs_entry queue_max_sectors_entry
= {
3539 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
3540 .show
= queue_max_sectors_show
,
3541 .store
= queue_max_sectors_store
,
3544 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
3545 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
3546 .show
= queue_max_hw_sectors_show
,
3549 static struct queue_sysfs_entry queue_iosched_entry
= {
3550 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
3551 .show
= elv_iosched_show
,
3552 .store
= elv_iosched_store
,
3555 static struct attribute
*default_attrs
[] = {
3556 &queue_requests_entry
.attr
,
3557 &queue_ra_entry
.attr
,
3558 &queue_max_hw_sectors_entry
.attr
,
3559 &queue_max_sectors_entry
.attr
,
3560 &queue_iosched_entry
.attr
,
3564 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3567 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3569 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3570 struct request_queue
*q
;
3572 q
= container_of(kobj
, struct request_queue
, kobj
);
3576 return entry
->show(q
, page
);
3580 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3581 const char *page
, size_t length
)
3583 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3584 struct request_queue
*q
;
3586 q
= container_of(kobj
, struct request_queue
, kobj
);
3590 return entry
->store(q
, page
, length
);
3593 static struct sysfs_ops queue_sysfs_ops
= {
3594 .show
= queue_attr_show
,
3595 .store
= queue_attr_store
,
3598 struct kobj_type queue_ktype
= {
3599 .sysfs_ops
= &queue_sysfs_ops
,
3600 .default_attrs
= default_attrs
,
3603 int blk_register_queue(struct gendisk
*disk
)
3607 request_queue_t
*q
= disk
->queue
;
3609 if (!q
|| !q
->request_fn
)
3612 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
3613 if (!q
->kobj
.parent
)
3616 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
3617 q
->kobj
.ktype
= &queue_ktype
;
3619 ret
= kobject_register(&q
->kobj
);
3623 ret
= elv_register_queue(q
);
3625 kobject_unregister(&q
->kobj
);
3632 void blk_unregister_queue(struct gendisk
*disk
)
3634 request_queue_t
*q
= disk
->queue
;
3636 if (q
&& q
->request_fn
) {
3637 elv_unregister_queue(q
);
3639 kobject_unregister(&q
->kobj
);
3640 kobject_put(&disk
->kobj
);