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
);
42 * For the allocated request tables
44 static kmem_cache_t
*request_cachep
;
47 * For queue allocation
49 static kmem_cache_t
*requestq_cachep
;
52 * For io context allocations
54 static kmem_cache_t
*iocontext_cachep
;
56 static wait_queue_head_t congestion_wqh
[2] = {
57 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
58 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
62 * Controlling structure to kblockd
64 static struct workqueue_struct
*kblockd_workqueue
;
66 unsigned long blk_max_low_pfn
, blk_max_pfn
;
68 EXPORT_SYMBOL(blk_max_low_pfn
);
69 EXPORT_SYMBOL(blk_max_pfn
);
71 /* Amount of time in which a process may batch requests */
72 #define BLK_BATCH_TIME (HZ/50UL)
74 /* Number of requests a "batching" process may submit */
75 #define BLK_BATCH_REQ 32
78 * Return the threshold (number of used requests) at which the queue is
79 * considered to be congested. It include a little hysteresis to keep the
80 * context switch rate down.
82 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
84 return q
->nr_congestion_on
;
88 * The threshold at which a queue is considered to be uncongested
90 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
92 return q
->nr_congestion_off
;
95 static void blk_queue_congestion_threshold(struct request_queue
*q
)
99 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
100 if (nr
> q
->nr_requests
)
102 q
->nr_congestion_on
= nr
;
104 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
107 q
->nr_congestion_off
= nr
;
111 * A queue has just exitted congestion. Note this in the global counter of
112 * congested queues, and wake up anyone who was waiting for requests to be
115 static void clear_queue_congested(request_queue_t
*q
, int rw
)
118 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
120 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
121 clear_bit(bit
, &q
->backing_dev_info
.state
);
122 smp_mb__after_clear_bit();
123 if (waitqueue_active(wqh
))
128 * A queue has just entered congestion. Flag that in the queue's VM-visible
129 * state flags and increment the global gounter of congested queues.
131 static void set_queue_congested(request_queue_t
*q
, int rw
)
135 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
136 set_bit(bit
, &q
->backing_dev_info
.state
);
140 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
143 * Locates the passed device's request queue and returns the address of its
146 * Will return NULL if the request queue cannot be located.
148 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
150 struct backing_dev_info
*ret
= NULL
;
151 request_queue_t
*q
= bdev_get_queue(bdev
);
154 ret
= &q
->backing_dev_info
;
158 EXPORT_SYMBOL(blk_get_backing_dev_info
);
160 void blk_queue_activity_fn(request_queue_t
*q
, activity_fn
*fn
, void *data
)
163 q
->activity_data
= data
;
166 EXPORT_SYMBOL(blk_queue_activity_fn
);
169 * blk_queue_prep_rq - set a prepare_request function for queue
171 * @pfn: prepare_request function
173 * It's possible for a queue to register a prepare_request callback which
174 * is invoked before the request is handed to the request_fn. The goal of
175 * the function is to prepare a request for I/O, it can be used to build a
176 * cdb from the request data for instance.
179 void blk_queue_prep_rq(request_queue_t
*q
, prep_rq_fn
*pfn
)
184 EXPORT_SYMBOL(blk_queue_prep_rq
);
187 * blk_queue_merge_bvec - set a merge_bvec function for queue
189 * @mbfn: merge_bvec_fn
191 * Usually queues have static limitations on the max sectors or segments that
192 * we can put in a request. Stacking drivers may have some settings that
193 * are dynamic, and thus we have to query the queue whether it is ok to
194 * add a new bio_vec to a bio at a given offset or not. If the block device
195 * has such limitations, it needs to register a merge_bvec_fn to control
196 * the size of bio's sent to it. Note that a block device *must* allow a
197 * single page to be added to an empty bio. The block device driver may want
198 * to use the bio_split() function to deal with these bio's. By default
199 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
202 void blk_queue_merge_bvec(request_queue_t
*q
, merge_bvec_fn
*mbfn
)
204 q
->merge_bvec_fn
= mbfn
;
207 EXPORT_SYMBOL(blk_queue_merge_bvec
);
210 * blk_queue_make_request - define an alternate make_request function for a device
211 * @q: the request queue for the device to be affected
212 * @mfn: the alternate make_request function
215 * The normal way for &struct bios to be passed to a device
216 * driver is for them to be collected into requests on a request
217 * queue, and then to allow the device driver to select requests
218 * off that queue when it is ready. This works well for many block
219 * devices. However some block devices (typically virtual devices
220 * such as md or lvm) do not benefit from the processing on the
221 * request queue, and are served best by having the requests passed
222 * directly to them. This can be achieved by providing a function
223 * to blk_queue_make_request().
226 * The driver that does this *must* be able to deal appropriately
227 * with buffers in "highmemory". This can be accomplished by either calling
228 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
229 * blk_queue_bounce() to create a buffer in normal memory.
231 void blk_queue_make_request(request_queue_t
* q
, make_request_fn
* mfn
)
236 q
->nr_requests
= BLKDEV_MAX_RQ
;
237 q
->max_phys_segments
= MAX_PHYS_SEGMENTS
;
238 q
->max_hw_segments
= MAX_HW_SEGMENTS
;
239 q
->make_request_fn
= mfn
;
240 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
241 q
->backing_dev_info
.state
= 0;
242 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
243 blk_queue_max_sectors(q
, MAX_SECTORS
);
244 blk_queue_hardsect_size(q
, 512);
245 blk_queue_dma_alignment(q
, 511);
246 blk_queue_congestion_threshold(q
);
247 q
->nr_batching
= BLK_BATCH_REQ
;
249 q
->unplug_thresh
= 4; /* hmm */
250 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
251 if (q
->unplug_delay
== 0)
254 INIT_WORK(&q
->unplug_work
, blk_unplug_work
, q
);
256 q
->unplug_timer
.function
= blk_unplug_timeout
;
257 q
->unplug_timer
.data
= (unsigned long)q
;
260 * by default assume old behaviour and bounce for any highmem page
262 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
264 blk_queue_activity_fn(q
, NULL
, NULL
);
266 INIT_LIST_HEAD(&q
->drain_list
);
269 EXPORT_SYMBOL(blk_queue_make_request
);
271 static inline void rq_init(request_queue_t
*q
, struct request
*rq
)
273 INIT_LIST_HEAD(&rq
->queuelist
);
276 rq
->rq_status
= RQ_ACTIVE
;
277 rq
->bio
= rq
->biotail
= NULL
;
287 rq
->end_io_data
= NULL
;
291 * blk_queue_ordered - does this queue support ordered writes
292 * @q: the request queue
296 * For journalled file systems, doing ordered writes on a commit
297 * block instead of explicitly doing wait_on_buffer (which is bad
298 * for performance) can be a big win. Block drivers supporting this
299 * feature should call this function and indicate so.
302 void blk_queue_ordered(request_queue_t
*q
, int flag
)
305 case QUEUE_ORDERED_NONE
:
307 kmem_cache_free(request_cachep
, q
->flush_rq
);
311 case QUEUE_ORDERED_TAG
:
314 case QUEUE_ORDERED_FLUSH
:
317 q
->flush_rq
= kmem_cache_alloc(request_cachep
,
321 printk("blk_queue_ordered: bad value %d\n", flag
);
326 EXPORT_SYMBOL(blk_queue_ordered
);
329 * blk_queue_issue_flush_fn - set function for issuing a flush
330 * @q: the request queue
331 * @iff: the function to be called issuing the flush
334 * If a driver supports issuing a flush command, the support is notified
335 * to the block layer by defining it through this call.
338 void blk_queue_issue_flush_fn(request_queue_t
*q
, issue_flush_fn
*iff
)
340 q
->issue_flush_fn
= iff
;
343 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
346 * Cache flushing for ordered writes handling
348 static void blk_pre_flush_end_io(struct request
*flush_rq
)
350 struct request
*rq
= flush_rq
->end_io_data
;
351 request_queue_t
*q
= rq
->q
;
353 rq
->flags
|= REQ_BAR_PREFLUSH
;
355 if (!flush_rq
->errors
)
356 elv_requeue_request(q
, rq
);
358 q
->end_flush_fn(q
, flush_rq
);
359 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
364 static void blk_post_flush_end_io(struct request
*flush_rq
)
366 struct request
*rq
= flush_rq
->end_io_data
;
367 request_queue_t
*q
= rq
->q
;
369 rq
->flags
|= REQ_BAR_POSTFLUSH
;
371 q
->end_flush_fn(q
, flush_rq
);
372 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
376 struct request
*blk_start_pre_flush(request_queue_t
*q
, struct request
*rq
)
378 struct request
*flush_rq
= q
->flush_rq
;
380 BUG_ON(!blk_barrier_rq(rq
));
382 if (test_and_set_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
))
385 rq_init(q
, flush_rq
);
386 flush_rq
->elevator_private
= NULL
;
387 flush_rq
->flags
= REQ_BAR_FLUSH
;
388 flush_rq
->rq_disk
= rq
->rq_disk
;
392 * prepare_flush returns 0 if no flush is needed, just mark both
393 * pre and post flush as done in that case
395 if (!q
->prepare_flush_fn(q
, flush_rq
)) {
396 rq
->flags
|= REQ_BAR_PREFLUSH
| REQ_BAR_POSTFLUSH
;
397 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
402 * some drivers dequeue requests right away, some only after io
403 * completion. make sure the request is dequeued.
405 if (!list_empty(&rq
->queuelist
))
406 blkdev_dequeue_request(rq
);
408 elv_deactivate_request(q
, rq
);
410 flush_rq
->end_io_data
= rq
;
411 flush_rq
->end_io
= blk_pre_flush_end_io
;
413 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
417 static void blk_start_post_flush(request_queue_t
*q
, struct request
*rq
)
419 struct request
*flush_rq
= q
->flush_rq
;
421 BUG_ON(!blk_barrier_rq(rq
));
423 rq_init(q
, flush_rq
);
424 flush_rq
->elevator_private
= NULL
;
425 flush_rq
->flags
= REQ_BAR_FLUSH
;
426 flush_rq
->rq_disk
= rq
->rq_disk
;
429 if (q
->prepare_flush_fn(q
, flush_rq
)) {
430 flush_rq
->end_io_data
= rq
;
431 flush_rq
->end_io
= blk_post_flush_end_io
;
433 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
438 static inline int blk_check_end_barrier(request_queue_t
*q
, struct request
*rq
,
441 if (sectors
> rq
->nr_sectors
)
442 sectors
= rq
->nr_sectors
;
444 rq
->nr_sectors
-= sectors
;
445 return rq
->nr_sectors
;
448 static int __blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
,
449 int sectors
, int queue_locked
)
451 if (q
->ordered
!= QUEUE_ORDERED_FLUSH
)
453 if (!blk_fs_request(rq
) || !blk_barrier_rq(rq
))
455 if (blk_barrier_postflush(rq
))
458 if (!blk_check_end_barrier(q
, rq
, sectors
)) {
459 unsigned long flags
= 0;
462 spin_lock_irqsave(q
->queue_lock
, flags
);
464 blk_start_post_flush(q
, rq
);
467 spin_unlock_irqrestore(q
->queue_lock
, flags
);
474 * blk_complete_barrier_rq - complete possible barrier request
475 * @q: the request queue for the device
477 * @sectors: number of sectors to complete
480 * Used in driver end_io handling to determine whether to postpone
481 * completion of a barrier request until a post flush has been done. This
482 * is the unlocked variant, used if the caller doesn't already hold the
485 int blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
, int sectors
)
487 return __blk_complete_barrier_rq(q
, rq
, sectors
, 0);
489 EXPORT_SYMBOL(blk_complete_barrier_rq
);
492 * blk_complete_barrier_rq_locked - complete possible barrier request
493 * @q: the request queue for the device
495 * @sectors: number of sectors to complete
498 * See blk_complete_barrier_rq(). This variant must be used if the caller
499 * holds the queue lock.
501 int blk_complete_barrier_rq_locked(request_queue_t
*q
, struct request
*rq
,
504 return __blk_complete_barrier_rq(q
, rq
, sectors
, 1);
506 EXPORT_SYMBOL(blk_complete_barrier_rq_locked
);
509 * blk_queue_bounce_limit - set bounce buffer limit for queue
510 * @q: the request queue for the device
511 * @dma_addr: bus address limit
514 * Different hardware can have different requirements as to what pages
515 * it can do I/O directly to. A low level driver can call
516 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
517 * buffers for doing I/O to pages residing above @page. By default
518 * the block layer sets this to the highest numbered "low" memory page.
520 void blk_queue_bounce_limit(request_queue_t
*q
, u64 dma_addr
)
522 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
525 * set appropriate bounce gfp mask -- unfortunately we don't have a
526 * full 4GB zone, so we have to resort to low memory for any bounces.
527 * ISA has its own < 16MB zone.
529 if (bounce_pfn
< blk_max_low_pfn
) {
530 BUG_ON(dma_addr
< BLK_BOUNCE_ISA
);
531 init_emergency_isa_pool();
532 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
534 q
->bounce_gfp
= GFP_NOIO
;
536 q
->bounce_pfn
= bounce_pfn
;
539 EXPORT_SYMBOL(blk_queue_bounce_limit
);
542 * blk_queue_max_sectors - set max sectors for a request for this queue
543 * @q: the request queue for the device
544 * @max_sectors: max sectors in the usual 512b unit
547 * Enables a low level driver to set an upper limit on the size of
550 void blk_queue_max_sectors(request_queue_t
*q
, unsigned short max_sectors
)
552 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
553 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
554 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
557 q
->max_sectors
= q
->max_hw_sectors
= max_sectors
;
560 EXPORT_SYMBOL(blk_queue_max_sectors
);
563 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
564 * @q: the request queue for the device
565 * @max_segments: max number of segments
568 * Enables a low level driver to set an upper limit on the number of
569 * physical data segments in a request. This would be the largest sized
570 * scatter list the driver could handle.
572 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
576 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
579 q
->max_phys_segments
= max_segments
;
582 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
585 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
586 * @q: the request queue for the device
587 * @max_segments: max number of segments
590 * Enables a low level driver to set an upper limit on the number of
591 * hw data segments in a request. This would be the largest number of
592 * address/length pairs the host adapter can actually give as once
595 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
599 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
602 q
->max_hw_segments
= max_segments
;
605 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
608 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
609 * @q: the request queue for the device
610 * @max_size: max size of segment in bytes
613 * Enables a low level driver to set an upper limit on the size of a
616 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
618 if (max_size
< PAGE_CACHE_SIZE
) {
619 max_size
= PAGE_CACHE_SIZE
;
620 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
623 q
->max_segment_size
= max_size
;
626 EXPORT_SYMBOL(blk_queue_max_segment_size
);
629 * blk_queue_hardsect_size - set hardware sector size for the queue
630 * @q: the request queue for the device
631 * @size: the hardware sector size, in bytes
634 * This should typically be set to the lowest possible sector size
635 * that the hardware can operate on (possible without reverting to
636 * even internal read-modify-write operations). Usually the default
637 * of 512 covers most hardware.
639 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
641 q
->hardsect_size
= size
;
644 EXPORT_SYMBOL(blk_queue_hardsect_size
);
647 * Returns the minimum that is _not_ zero, unless both are zero.
649 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
652 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
653 * @t: the stacking driver (top)
654 * @b: the underlying device (bottom)
656 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
658 /* zero is "infinity" */
659 t
->max_sectors
= t
->max_hw_sectors
=
660 min_not_zero(t
->max_sectors
,b
->max_sectors
);
662 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
663 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
664 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
665 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
668 EXPORT_SYMBOL(blk_queue_stack_limits
);
671 * blk_queue_segment_boundary - set boundary rules for segment merging
672 * @q: the request queue for the device
673 * @mask: the memory boundary mask
675 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
677 if (mask
< PAGE_CACHE_SIZE
- 1) {
678 mask
= PAGE_CACHE_SIZE
- 1;
679 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
682 q
->seg_boundary_mask
= mask
;
685 EXPORT_SYMBOL(blk_queue_segment_boundary
);
688 * blk_queue_dma_alignment - set dma length and memory alignment
689 * @q: the request queue for the device
690 * @mask: alignment mask
693 * set required memory and length aligment for direct dma transactions.
694 * this is used when buiding direct io requests for the queue.
697 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
699 q
->dma_alignment
= mask
;
702 EXPORT_SYMBOL(blk_queue_dma_alignment
);
705 * blk_queue_find_tag - find a request by its tag and queue
707 * @q: The request queue for the device
708 * @tag: The tag of the request
711 * Should be used when a device returns a tag and you want to match
714 * no locks need be held.
716 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
718 struct blk_queue_tag
*bqt
= q
->queue_tags
;
720 if (unlikely(bqt
== NULL
|| tag
>= bqt
->real_max_depth
))
723 return bqt
->tag_index
[tag
];
726 EXPORT_SYMBOL(blk_queue_find_tag
);
729 * __blk_queue_free_tags - release tag maintenance info
730 * @q: the request queue for the device
733 * blk_cleanup_queue() will take care of calling this function, if tagging
734 * has been used. So there's no need to call this directly.
736 static void __blk_queue_free_tags(request_queue_t
*q
)
738 struct blk_queue_tag
*bqt
= q
->queue_tags
;
743 if (atomic_dec_and_test(&bqt
->refcnt
)) {
745 BUG_ON(!list_empty(&bqt
->busy_list
));
747 kfree(bqt
->tag_index
);
748 bqt
->tag_index
= NULL
;
756 q
->queue_tags
= NULL
;
757 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
761 * blk_queue_free_tags - release tag maintenance info
762 * @q: the request queue for the device
765 * This is used to disabled tagged queuing to a device, yet leave
768 void blk_queue_free_tags(request_queue_t
*q
)
770 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
773 EXPORT_SYMBOL(blk_queue_free_tags
);
776 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
779 struct request
**tag_index
;
780 unsigned long *tag_map
;
782 if (depth
> q
->nr_requests
* 2) {
783 depth
= q
->nr_requests
* 2;
784 printk(KERN_ERR
"%s: adjusted depth to %d\n",
785 __FUNCTION__
, depth
);
788 tag_index
= kmalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
792 bits
= (depth
/ BLK_TAGS_PER_LONG
) + 1;
793 tag_map
= kmalloc(bits
* sizeof(unsigned long), GFP_ATOMIC
);
797 memset(tag_index
, 0, depth
* sizeof(struct request
*));
798 memset(tag_map
, 0, bits
* sizeof(unsigned long));
799 tags
->max_depth
= depth
;
800 tags
->real_max_depth
= bits
* BITS_PER_LONG
;
801 tags
->tag_index
= tag_index
;
802 tags
->tag_map
= tag_map
;
805 * set the upper bits if the depth isn't a multiple of the word size
807 for (i
= depth
; i
< bits
* BLK_TAGS_PER_LONG
; i
++)
808 __set_bit(i
, tag_map
);
817 * blk_queue_init_tags - initialize the queue tag info
818 * @q: the request queue for the device
819 * @depth: the maximum queue depth supported
820 * @tags: the tag to use
822 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
823 struct blk_queue_tag
*tags
)
827 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
829 if (!tags
&& !q
->queue_tags
) {
830 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
834 if (init_tag_map(q
, tags
, depth
))
837 INIT_LIST_HEAD(&tags
->busy_list
);
839 atomic_set(&tags
->refcnt
, 1);
840 } else if (q
->queue_tags
) {
841 if ((rc
= blk_queue_resize_tags(q
, depth
)))
843 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
846 atomic_inc(&tags
->refcnt
);
849 * assign it, all done
851 q
->queue_tags
= tags
;
852 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
859 EXPORT_SYMBOL(blk_queue_init_tags
);
862 * blk_queue_resize_tags - change the queueing depth
863 * @q: the request queue for the device
864 * @new_depth: the new max command queueing depth
867 * Must be called with the queue lock held.
869 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
871 struct blk_queue_tag
*bqt
= q
->queue_tags
;
872 struct request
**tag_index
;
873 unsigned long *tag_map
;
880 * don't bother sizing down
882 if (new_depth
<= bqt
->real_max_depth
) {
883 bqt
->max_depth
= new_depth
;
888 * save the old state info, so we can copy it back
890 tag_index
= bqt
->tag_index
;
891 tag_map
= bqt
->tag_map
;
892 max_depth
= bqt
->real_max_depth
;
894 if (init_tag_map(q
, bqt
, new_depth
))
897 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
898 bits
= max_depth
/ BLK_TAGS_PER_LONG
;
899 memcpy(bqt
->tag_map
, tag_map
, bits
* sizeof(unsigned long));
906 EXPORT_SYMBOL(blk_queue_resize_tags
);
909 * blk_queue_end_tag - end tag operations for a request
910 * @q: the request queue for the device
911 * @rq: the request that has completed
914 * Typically called when end_that_request_first() returns 0, meaning
915 * all transfers have been done for a request. It's important to call
916 * this function before end_that_request_last(), as that will put the
917 * request back on the free list thus corrupting the internal tag list.
920 * queue lock must be held.
922 void blk_queue_end_tag(request_queue_t
*q
, struct request
*rq
)
924 struct blk_queue_tag
*bqt
= q
->queue_tags
;
929 if (unlikely(tag
>= bqt
->real_max_depth
))
932 if (unlikely(!__test_and_clear_bit(tag
, bqt
->tag_map
))) {
933 printk("attempt to clear non-busy tag (%d)\n", tag
);
937 list_del_init(&rq
->queuelist
);
938 rq
->flags
&= ~REQ_QUEUED
;
941 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
942 printk("tag %d is missing\n", tag
);
944 bqt
->tag_index
[tag
] = NULL
;
948 EXPORT_SYMBOL(blk_queue_end_tag
);
951 * blk_queue_start_tag - find a free tag and assign it
952 * @q: the request queue for the device
953 * @rq: the block request that needs tagging
956 * This can either be used as a stand-alone helper, or possibly be
957 * assigned as the queue &prep_rq_fn (in which case &struct request
958 * automagically gets a tag assigned). Note that this function
959 * assumes that any type of request can be queued! if this is not
960 * true for your device, you must check the request type before
961 * calling this function. The request will also be removed from
962 * the request queue, so it's the drivers responsibility to readd
963 * it if it should need to be restarted for some reason.
966 * queue lock must be held.
968 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
970 struct blk_queue_tag
*bqt
= q
->queue_tags
;
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 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
981 if (tag
>= bqt
->max_depth
)
984 __set_bit(tag
, bqt
->tag_map
);
986 rq
->flags
|= REQ_QUEUED
;
988 bqt
->tag_index
[tag
] = rq
;
989 blkdev_dequeue_request(rq
);
990 list_add(&rq
->queuelist
, &bqt
->busy_list
);
995 EXPORT_SYMBOL(blk_queue_start_tag
);
998 * blk_queue_invalidate_tags - invalidate all pending tags
999 * @q: the request queue for the device
1002 * Hardware conditions may dictate a need to stop all pending requests.
1003 * In this case, we will safely clear the block side of the tag queue and
1004 * readd all requests to the request queue in the right order.
1007 * queue lock must be held.
1009 void blk_queue_invalidate_tags(request_queue_t
*q
)
1011 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1012 struct list_head
*tmp
, *n
;
1015 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1016 rq
= list_entry_rq(tmp
);
1018 if (rq
->tag
== -1) {
1019 printk("bad tag found on list\n");
1020 list_del_init(&rq
->queuelist
);
1021 rq
->flags
&= ~REQ_QUEUED
;
1023 blk_queue_end_tag(q
, rq
);
1025 rq
->flags
&= ~REQ_STARTED
;
1026 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1030 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1032 static char *rq_flags
[] = {
1050 "REQ_DRIVE_TASKFILE",
1057 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1061 printk("%s: dev %s: flags = ", msg
,
1062 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?");
1065 if (rq
->flags
& (1 << bit
))
1066 printk("%s ", rq_flags
[bit
]);
1068 } while (bit
< __REQ_NR_BITS
);
1070 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1072 rq
->current_nr_sectors
);
1073 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1075 if (rq
->flags
& (REQ_BLOCK_PC
| REQ_PC
)) {
1077 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1078 printk("%02x ", rq
->cmd
[bit
]);
1083 EXPORT_SYMBOL(blk_dump_rq_flags
);
1085 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1087 struct bio_vec
*bv
, *bvprv
= NULL
;
1088 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1089 int high
, highprv
= 1;
1091 if (unlikely(!bio
->bi_io_vec
))
1094 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1095 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1096 bio_for_each_segment(bv
, bio
, i
) {
1098 * the trick here is making sure that a high page is never
1099 * considered part of another segment, since that might
1100 * change with the bounce page.
1102 high
= page_to_pfn(bv
->bv_page
) >= q
->bounce_pfn
;
1103 if (high
|| highprv
)
1104 goto new_hw_segment
;
1106 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1108 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1110 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1112 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1113 goto new_hw_segment
;
1115 seg_size
+= bv
->bv_len
;
1116 hw_seg_size
+= bv
->bv_len
;
1121 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1122 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1123 hw_seg_size
+= bv
->bv_len
;
1126 if (hw_seg_size
> bio
->bi_hw_front_size
)
1127 bio
->bi_hw_front_size
= hw_seg_size
;
1128 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1134 seg_size
= bv
->bv_len
;
1137 if (hw_seg_size
> bio
->bi_hw_back_size
)
1138 bio
->bi_hw_back_size
= hw_seg_size
;
1139 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1140 bio
->bi_hw_front_size
= hw_seg_size
;
1141 bio
->bi_phys_segments
= nr_phys_segs
;
1142 bio
->bi_hw_segments
= nr_hw_segs
;
1143 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1147 int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1150 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1153 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1155 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1159 * bio and nxt are contigous in memory, check if the queue allows
1160 * these two to be merged into one
1162 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1168 EXPORT_SYMBOL(blk_phys_contig_segment
);
1170 int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1173 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1174 blk_recount_segments(q
, bio
);
1175 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1176 blk_recount_segments(q
, nxt
);
1177 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1178 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1180 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1186 EXPORT_SYMBOL(blk_hw_contig_segment
);
1189 * map a request to scatterlist, return number of sg entries setup. Caller
1190 * must make sure sg can hold rq->nr_phys_segments entries
1192 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1194 struct bio_vec
*bvec
, *bvprv
;
1196 int nsegs
, i
, cluster
;
1199 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1202 * for each bio in rq
1205 rq_for_each_bio(bio
, rq
) {
1207 * for each segment in bio
1209 bio_for_each_segment(bvec
, bio
, i
) {
1210 int nbytes
= bvec
->bv_len
;
1212 if (bvprv
&& cluster
) {
1213 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1216 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1218 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1221 sg
[nsegs
- 1].length
+= nbytes
;
1224 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1225 sg
[nsegs
].page
= bvec
->bv_page
;
1226 sg
[nsegs
].length
= nbytes
;
1227 sg
[nsegs
].offset
= bvec
->bv_offset
;
1232 } /* segments in bio */
1238 EXPORT_SYMBOL(blk_rq_map_sg
);
1241 * the standard queue merge functions, can be overridden with device
1242 * specific ones if so desired
1245 static inline int ll_new_mergeable(request_queue_t
*q
,
1246 struct request
*req
,
1249 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1251 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1252 req
->flags
|= REQ_NOMERGE
;
1253 if (req
== q
->last_merge
)
1254 q
->last_merge
= NULL
;
1259 * A hw segment is just getting larger, bump just the phys
1262 req
->nr_phys_segments
+= nr_phys_segs
;
1266 static inline int ll_new_hw_segment(request_queue_t
*q
,
1267 struct request
*req
,
1270 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1271 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1273 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1274 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1275 req
->flags
|= REQ_NOMERGE
;
1276 if (req
== q
->last_merge
)
1277 q
->last_merge
= NULL
;
1282 * This will form the start of a new hw segment. Bump both
1285 req
->nr_hw_segments
+= nr_hw_segs
;
1286 req
->nr_phys_segments
+= nr_phys_segs
;
1290 static int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
,
1295 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1296 req
->flags
|= REQ_NOMERGE
;
1297 if (req
== q
->last_merge
)
1298 q
->last_merge
= NULL
;
1301 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1302 blk_recount_segments(q
, req
->biotail
);
1303 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1304 blk_recount_segments(q
, bio
);
1305 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1306 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1307 !BIOVEC_VIRT_OVERSIZE(len
)) {
1308 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1311 if (req
->nr_hw_segments
== 1)
1312 req
->bio
->bi_hw_front_size
= len
;
1313 if (bio
->bi_hw_segments
== 1)
1314 bio
->bi_hw_back_size
= len
;
1319 return ll_new_hw_segment(q
, req
, bio
);
1322 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1327 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1328 req
->flags
|= REQ_NOMERGE
;
1329 if (req
== q
->last_merge
)
1330 q
->last_merge
= NULL
;
1333 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1334 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1335 blk_recount_segments(q
, bio
);
1336 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1337 blk_recount_segments(q
, req
->bio
);
1338 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1339 !BIOVEC_VIRT_OVERSIZE(len
)) {
1340 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1343 if (bio
->bi_hw_segments
== 1)
1344 bio
->bi_hw_front_size
= len
;
1345 if (req
->nr_hw_segments
== 1)
1346 req
->biotail
->bi_hw_back_size
= len
;
1351 return ll_new_hw_segment(q
, req
, bio
);
1354 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1355 struct request
*next
)
1357 int total_phys_segments
= req
->nr_phys_segments
+next
->nr_phys_segments
;
1358 int total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1361 * First check if the either of the requests are re-queued
1362 * requests. Can't merge them if they are.
1364 if (req
->special
|| next
->special
)
1368 * Will it become to large?
1370 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1373 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1374 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1375 total_phys_segments
--;
1377 if (total_phys_segments
> q
->max_phys_segments
)
1380 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1381 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1382 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1384 * propagate the combined length to the end of the requests
1386 if (req
->nr_hw_segments
== 1)
1387 req
->bio
->bi_hw_front_size
= len
;
1388 if (next
->nr_hw_segments
== 1)
1389 next
->biotail
->bi_hw_back_size
= len
;
1390 total_hw_segments
--;
1393 if (total_hw_segments
> q
->max_hw_segments
)
1396 /* Merge is OK... */
1397 req
->nr_phys_segments
= total_phys_segments
;
1398 req
->nr_hw_segments
= total_hw_segments
;
1403 * "plug" the device if there are no outstanding requests: this will
1404 * force the transfer to start only after we have put all the requests
1407 * This is called with interrupts off and no requests on the queue and
1408 * with the queue lock held.
1410 void blk_plug_device(request_queue_t
*q
)
1412 WARN_ON(!irqs_disabled());
1415 * don't plug a stopped queue, it must be paired with blk_start_queue()
1416 * which will restart the queueing
1418 if (test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
))
1421 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1422 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1425 EXPORT_SYMBOL(blk_plug_device
);
1428 * remove the queue from the plugged list, if present. called with
1429 * queue lock held and interrupts disabled.
1431 int blk_remove_plug(request_queue_t
*q
)
1433 WARN_ON(!irqs_disabled());
1435 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1438 del_timer(&q
->unplug_timer
);
1442 EXPORT_SYMBOL(blk_remove_plug
);
1445 * remove the plug and let it rip..
1447 void __generic_unplug_device(request_queue_t
*q
)
1449 if (test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
))
1452 if (!blk_remove_plug(q
))
1456 * was plugged, fire request_fn if queue has stuff to do
1458 if (elv_next_request(q
))
1461 EXPORT_SYMBOL(__generic_unplug_device
);
1464 * generic_unplug_device - fire a request queue
1465 * @q: The &request_queue_t in question
1468 * Linux uses plugging to build bigger requests queues before letting
1469 * the device have at them. If a queue is plugged, the I/O scheduler
1470 * is still adding and merging requests on the queue. Once the queue
1471 * gets unplugged, the request_fn defined for the queue is invoked and
1472 * transfers started.
1474 void generic_unplug_device(request_queue_t
*q
)
1476 spin_lock_irq(q
->queue_lock
);
1477 __generic_unplug_device(q
);
1478 spin_unlock_irq(q
->queue_lock
);
1480 EXPORT_SYMBOL(generic_unplug_device
);
1482 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1485 request_queue_t
*q
= bdi
->unplug_io_data
;
1488 * devices don't necessarily have an ->unplug_fn defined
1494 static void blk_unplug_work(void *data
)
1496 request_queue_t
*q
= data
;
1501 static void blk_unplug_timeout(unsigned long data
)
1503 request_queue_t
*q
= (request_queue_t
*)data
;
1505 kblockd_schedule_work(&q
->unplug_work
);
1509 * blk_start_queue - restart a previously stopped queue
1510 * @q: The &request_queue_t in question
1513 * blk_start_queue() will clear the stop flag on the queue, and call
1514 * the request_fn for the queue if it was in a stopped state when
1515 * entered. Also see blk_stop_queue(). Queue lock must be held.
1517 void blk_start_queue(request_queue_t
*q
)
1519 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1522 * one level of recursion is ok and is much faster than kicking
1523 * the unplug handling
1525 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1527 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1530 kblockd_schedule_work(&q
->unplug_work
);
1534 EXPORT_SYMBOL(blk_start_queue
);
1537 * blk_stop_queue - stop a queue
1538 * @q: The &request_queue_t in question
1541 * The Linux block layer assumes that a block driver will consume all
1542 * entries on the request queue when the request_fn strategy is called.
1543 * Often this will not happen, because of hardware limitations (queue
1544 * depth settings). If a device driver gets a 'queue full' response,
1545 * or if it simply chooses not to queue more I/O at one point, it can
1546 * call this function to prevent the request_fn from being called until
1547 * the driver has signalled it's ready to go again. This happens by calling
1548 * blk_start_queue() to restart queue operations. Queue lock must be held.
1550 void blk_stop_queue(request_queue_t
*q
)
1553 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1555 EXPORT_SYMBOL(blk_stop_queue
);
1558 * blk_sync_queue - cancel any pending callbacks on a queue
1562 * The block layer may perform asynchronous callback activity
1563 * on a queue, such as calling the unplug function after a timeout.
1564 * A block device may call blk_sync_queue to ensure that any
1565 * such activity is cancelled, thus allowing it to release resources
1566 * the the callbacks might use. The caller must already have made sure
1567 * that its ->make_request_fn will not re-add plugging prior to calling
1571 void blk_sync_queue(struct request_queue
*q
)
1573 del_timer_sync(&q
->unplug_timer
);
1576 EXPORT_SYMBOL(blk_sync_queue
);
1579 * blk_run_queue - run a single device queue
1580 * @q: The queue to run
1582 void blk_run_queue(struct request_queue
*q
)
1584 unsigned long flags
;
1586 spin_lock_irqsave(q
->queue_lock
, flags
);
1588 if (!elv_queue_empty(q
))
1590 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1592 EXPORT_SYMBOL(blk_run_queue
);
1595 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1596 * @q: the request queue to be released
1599 * blk_cleanup_queue is the pair to blk_init_queue() or
1600 * blk_queue_make_request(). It should be called when a request queue is
1601 * being released; typically when a block device is being de-registered.
1602 * Currently, its primary task it to free all the &struct request
1603 * structures that were allocated to the queue and the queue itself.
1606 * Hopefully the low level driver will have finished any
1607 * outstanding requests first...
1609 void blk_cleanup_queue(request_queue_t
* q
)
1611 struct request_list
*rl
= &q
->rq
;
1613 if (!atomic_dec_and_test(&q
->refcnt
))
1617 elevator_exit(q
->elevator
);
1622 mempool_destroy(rl
->rq_pool
);
1625 __blk_queue_free_tags(q
);
1627 blk_queue_ordered(q
, QUEUE_ORDERED_NONE
);
1629 kmem_cache_free(requestq_cachep
, q
);
1632 EXPORT_SYMBOL(blk_cleanup_queue
);
1634 static int blk_init_free_list(request_queue_t
*q
)
1636 struct request_list
*rl
= &q
->rq
;
1638 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1639 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1640 init_waitqueue_head(&rl
->wait
[READ
]);
1641 init_waitqueue_head(&rl
->wait
[WRITE
]);
1642 init_waitqueue_head(&rl
->drain
);
1644 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1645 mempool_free_slab
, request_cachep
, q
->node
);
1653 static int __make_request(request_queue_t
*, struct bio
*);
1655 request_queue_t
*blk_alloc_queue(int gfp_mask
)
1657 return blk_alloc_queue_node(gfp_mask
, -1);
1659 EXPORT_SYMBOL(blk_alloc_queue
);
1661 request_queue_t
*blk_alloc_queue_node(int gfp_mask
, int node_id
)
1665 q
= kmem_cache_alloc_node(requestq_cachep
, gfp_mask
, node_id
);
1669 memset(q
, 0, sizeof(*q
));
1670 init_timer(&q
->unplug_timer
);
1671 atomic_set(&q
->refcnt
, 1);
1673 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1674 q
->backing_dev_info
.unplug_io_data
= q
;
1678 EXPORT_SYMBOL(blk_alloc_queue_node
);
1681 * blk_init_queue - prepare a request queue for use with a block device
1682 * @rfn: The function to be called to process requests that have been
1683 * placed on the queue.
1684 * @lock: Request queue spin lock
1687 * If a block device wishes to use the standard request handling procedures,
1688 * which sorts requests and coalesces adjacent requests, then it must
1689 * call blk_init_queue(). The function @rfn will be called when there
1690 * are requests on the queue that need to be processed. If the device
1691 * supports plugging, then @rfn may not be called immediately when requests
1692 * are available on the queue, but may be called at some time later instead.
1693 * Plugged queues are generally unplugged when a buffer belonging to one
1694 * of the requests on the queue is needed, or due to memory pressure.
1696 * @rfn is not required, or even expected, to remove all requests off the
1697 * queue, but only as many as it can handle at a time. If it does leave
1698 * requests on the queue, it is responsible for arranging that the requests
1699 * get dealt with eventually.
1701 * The queue spin lock must be held while manipulating the requests on the
1704 * Function returns a pointer to the initialized request queue, or NULL if
1705 * it didn't succeed.
1708 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1709 * when the block device is deactivated (such as at module unload).
1712 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1714 return blk_init_queue_node(rfn
, lock
, -1);
1716 EXPORT_SYMBOL(blk_init_queue
);
1719 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1721 request_queue_t
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1727 if (blk_init_free_list(q
))
1731 * if caller didn't supply a lock, they get per-queue locking with
1735 spin_lock_init(&q
->__queue_lock
);
1736 lock
= &q
->__queue_lock
;
1739 q
->request_fn
= rfn
;
1740 q
->back_merge_fn
= ll_back_merge_fn
;
1741 q
->front_merge_fn
= ll_front_merge_fn
;
1742 q
->merge_requests_fn
= ll_merge_requests_fn
;
1743 q
->prep_rq_fn
= NULL
;
1744 q
->unplug_fn
= generic_unplug_device
;
1745 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1746 q
->queue_lock
= lock
;
1748 blk_queue_segment_boundary(q
, 0xffffffff);
1750 blk_queue_make_request(q
, __make_request
);
1751 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1753 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1754 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1759 if (!elevator_init(q
, NULL
)) {
1760 blk_queue_congestion_threshold(q
);
1764 blk_cleanup_queue(q
);
1766 kmem_cache_free(requestq_cachep
, q
);
1769 EXPORT_SYMBOL(blk_init_queue_node
);
1771 int blk_get_queue(request_queue_t
*q
)
1773 if (!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
1774 atomic_inc(&q
->refcnt
);
1781 EXPORT_SYMBOL(blk_get_queue
);
1783 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
1785 elv_put_request(q
, rq
);
1786 mempool_free(rq
, q
->rq
.rq_pool
);
1789 static inline struct request
*blk_alloc_request(request_queue_t
*q
, int rw
,
1792 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1798 * first three bits are identical in rq->flags and bio->bi_rw,
1799 * see bio.h and blkdev.h
1803 if (!elv_set_request(q
, rq
, gfp_mask
))
1806 mempool_free(rq
, q
->rq
.rq_pool
);
1811 * ioc_batching returns true if the ioc is a valid batching request and
1812 * should be given priority access to a request.
1814 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
1820 * Make sure the process is able to allocate at least 1 request
1821 * even if the batch times out, otherwise we could theoretically
1824 return ioc
->nr_batch_requests
== q
->nr_batching
||
1825 (ioc
->nr_batch_requests
> 0
1826 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1830 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1831 * will cause the process to be a "batcher" on all queues in the system. This
1832 * is the behaviour we want though - once it gets a wakeup it should be given
1835 void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
1837 if (!ioc
|| ioc_batching(q
, ioc
))
1840 ioc
->nr_batch_requests
= q
->nr_batching
;
1841 ioc
->last_waited
= jiffies
;
1844 static void __freed_request(request_queue_t
*q
, int rw
)
1846 struct request_list
*rl
= &q
->rq
;
1848 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
1849 clear_queue_congested(q
, rw
);
1851 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
1853 if (waitqueue_active(&rl
->wait
[rw
]))
1854 wake_up(&rl
->wait
[rw
]);
1856 blk_clear_queue_full(q
, rw
);
1861 * A request has just been released. Account for it, update the full and
1862 * congestion status, wake up any waiters. Called under q->queue_lock.
1864 static void freed_request(request_queue_t
*q
, int rw
)
1866 struct request_list
*rl
= &q
->rq
;
1870 __freed_request(q
, rw
);
1872 if (unlikely(rl
->starved
[rw
^ 1]))
1873 __freed_request(q
, rw
^ 1);
1875 if (!rl
->count
[READ
] && !rl
->count
[WRITE
]) {
1877 if (unlikely(waitqueue_active(&rl
->drain
)))
1878 wake_up(&rl
->drain
);
1882 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1884 * Get a free request, queue_lock must not be held
1886 static struct request
*get_request(request_queue_t
*q
, int rw
, int gfp_mask
)
1888 struct request
*rq
= NULL
;
1889 struct request_list
*rl
= &q
->rq
;
1890 struct io_context
*ioc
= get_io_context(gfp_mask
);
1892 if (unlikely(test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
)))
1895 spin_lock_irq(q
->queue_lock
);
1896 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
1898 * The queue will fill after this allocation, so set it as
1899 * full, and mark this process as "batching". This process
1900 * will be allowed to complete a batch of requests, others
1903 if (!blk_queue_full(q
, rw
)) {
1904 ioc_set_batching(q
, ioc
);
1905 blk_set_queue_full(q
, rw
);
1909 switch (elv_may_queue(q
, rw
)) {
1912 case ELV_MQUEUE_MAY
:
1914 case ELV_MQUEUE_MUST
:
1918 if (blk_queue_full(q
, rw
) && !ioc_batching(q
, ioc
)) {
1920 * The queue is full and the allocating process is not a
1921 * "batcher", and not exempted by the IO scheduler
1923 spin_unlock_irq(q
->queue_lock
);
1929 rl
->starved
[rw
] = 0;
1930 if (rl
->count
[rw
] >= queue_congestion_on_threshold(q
))
1931 set_queue_congested(q
, rw
);
1932 spin_unlock_irq(q
->queue_lock
);
1934 rq
= blk_alloc_request(q
, rw
, gfp_mask
);
1937 * Allocation failed presumably due to memory. Undo anything
1938 * we might have messed up.
1940 * Allocating task should really be put onto the front of the
1941 * wait queue, but this is pretty rare.
1943 spin_lock_irq(q
->queue_lock
);
1944 freed_request(q
, rw
);
1947 * in the very unlikely event that allocation failed and no
1948 * requests for this direction was pending, mark us starved
1949 * so that freeing of a request in the other direction will
1950 * notice us. another possible fix would be to split the
1951 * rq mempool into READ and WRITE
1954 if (unlikely(rl
->count
[rw
] == 0))
1955 rl
->starved
[rw
] = 1;
1957 spin_unlock_irq(q
->queue_lock
);
1961 if (ioc_batching(q
, ioc
))
1962 ioc
->nr_batch_requests
--;
1967 put_io_context(ioc
);
1972 * No available requests for this queue, unplug the device and wait for some
1973 * requests to become available.
1975 static struct request
*get_request_wait(request_queue_t
*q
, int rw
)
1980 generic_unplug_device(q
);
1982 struct request_list
*rl
= &q
->rq
;
1984 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
1985 TASK_UNINTERRUPTIBLE
);
1987 rq
= get_request(q
, rw
, GFP_NOIO
);
1990 struct io_context
*ioc
;
1995 * After sleeping, we become a "batching" process and
1996 * will be able to allocate at least one request, and
1997 * up to a big batch of them for a small period time.
1998 * See ioc_batching, ioc_set_batching
2000 ioc
= get_io_context(GFP_NOIO
);
2001 ioc_set_batching(q
, ioc
);
2002 put_io_context(ioc
);
2004 finish_wait(&rl
->wait
[rw
], &wait
);
2010 struct request
*blk_get_request(request_queue_t
*q
, int rw
, int gfp_mask
)
2014 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2016 if (gfp_mask
& __GFP_WAIT
)
2017 rq
= get_request_wait(q
, rw
);
2019 rq
= get_request(q
, rw
, gfp_mask
);
2024 EXPORT_SYMBOL(blk_get_request
);
2027 * blk_requeue_request - put a request back on queue
2028 * @q: request queue where request should be inserted
2029 * @rq: request to be inserted
2032 * Drivers often keep queueing requests until the hardware cannot accept
2033 * more, when that condition happens we need to put the request back
2034 * on the queue. Must be called with queue lock held.
2036 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2038 if (blk_rq_tagged(rq
))
2039 blk_queue_end_tag(q
, rq
);
2041 elv_requeue_request(q
, rq
);
2044 EXPORT_SYMBOL(blk_requeue_request
);
2047 * blk_insert_request - insert a special request in to a request queue
2048 * @q: request queue where request should be inserted
2049 * @rq: request to be inserted
2050 * @at_head: insert request at head or tail of queue
2051 * @data: private data
2054 * Many block devices need to execute commands asynchronously, so they don't
2055 * block the whole kernel from preemption during request execution. This is
2056 * accomplished normally by inserting aritficial requests tagged as
2057 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2058 * scheduled for actual execution by the request queue.
2060 * We have the option of inserting the head or the tail of the queue.
2061 * Typically we use the tail for new ioctls and so forth. We use the head
2062 * of the queue for things like a QUEUE_FULL message from a device, or a
2063 * host that is unable to accept a particular command.
2065 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2066 int at_head
, void *data
)
2068 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2069 unsigned long flags
;
2072 * tell I/O scheduler that this isn't a regular read/write (ie it
2073 * must not attempt merges on this) and that it acts as a soft
2076 rq
->flags
|= REQ_SPECIAL
| REQ_SOFTBARRIER
;
2080 spin_lock_irqsave(q
->queue_lock
, flags
);
2083 * If command is tagged, release the tag
2085 if (blk_rq_tagged(rq
))
2086 blk_queue_end_tag(q
, rq
);
2088 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2089 __elv_add_request(q
, rq
, where
, 0);
2091 if (blk_queue_plugged(q
))
2092 __generic_unplug_device(q
);
2095 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2098 EXPORT_SYMBOL(blk_insert_request
);
2101 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2102 * @q: request queue where request should be inserted
2103 * @rw: READ or WRITE data
2104 * @ubuf: the user buffer
2105 * @len: length of user data
2108 * Data will be mapped directly for zero copy io, if possible. Otherwise
2109 * a kernel bounce buffer is used.
2111 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2112 * still in process context.
2114 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2115 * before being submitted to the device, as pages mapped may be out of
2116 * reach. It's the callers responsibility to make sure this happens. The
2117 * original bio must be passed back in to blk_rq_unmap_user() for proper
2120 struct request
*blk_rq_map_user(request_queue_t
*q
, int rw
, void __user
*ubuf
,
2123 unsigned long uaddr
;
2127 if (len
> (q
->max_sectors
<< 9))
2128 return ERR_PTR(-EINVAL
);
2129 if ((!len
&& ubuf
) || (len
&& !ubuf
))
2130 return ERR_PTR(-EINVAL
);
2132 rq
= blk_get_request(q
, rw
, __GFP_WAIT
);
2134 return ERR_PTR(-ENOMEM
);
2137 * if alignment requirement is satisfied, map in user pages for
2138 * direct dma. else, set up kernel bounce buffers
2140 uaddr
= (unsigned long) ubuf
;
2141 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2142 bio
= bio_map_user(q
, NULL
, uaddr
, len
, rw
== READ
);
2144 bio
= bio_copy_user(q
, uaddr
, len
, rw
== READ
);
2147 rq
->bio
= rq
->biotail
= bio
;
2148 blk_rq_bio_prep(q
, rq
, bio
);
2150 rq
->buffer
= rq
->data
= NULL
;
2156 * bio is the err-ptr
2158 blk_put_request(rq
);
2159 return (struct request
*) bio
;
2162 EXPORT_SYMBOL(blk_rq_map_user
);
2165 * blk_rq_unmap_user - unmap a request with user data
2166 * @rq: request to be unmapped
2167 * @bio: bio for the request
2168 * @ulen: length of user buffer
2171 * Unmap a request previously mapped by blk_rq_map_user().
2173 int blk_rq_unmap_user(struct request
*rq
, struct bio
*bio
, unsigned int ulen
)
2178 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2179 bio_unmap_user(bio
);
2181 ret
= bio_uncopy_user(bio
);
2184 blk_put_request(rq
);
2188 EXPORT_SYMBOL(blk_rq_unmap_user
);
2191 * blk_execute_rq - insert a request into queue for execution
2192 * @q: queue to insert the request in
2193 * @bd_disk: matching gendisk
2194 * @rq: request to insert
2197 * Insert a fully prepared request at the back of the io scheduler queue
2200 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2203 DECLARE_COMPLETION(wait
);
2204 char sense
[SCSI_SENSE_BUFFERSIZE
];
2207 rq
->rq_disk
= bd_disk
;
2210 * we need an extra reference to the request, so we can look at
2211 * it after io completion
2216 memset(sense
, 0, sizeof(sense
));
2221 rq
->flags
|= REQ_NOMERGE
;
2222 rq
->waiting
= &wait
;
2223 rq
->end_io
= blk_end_sync_rq
;
2224 elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 1);
2225 generic_unplug_device(q
);
2226 wait_for_completion(&wait
);
2235 EXPORT_SYMBOL(blk_execute_rq
);
2238 * blkdev_issue_flush - queue a flush
2239 * @bdev: blockdev to issue flush for
2240 * @error_sector: error sector
2243 * Issue a flush for the block device in question. Caller can supply
2244 * room for storing the error offset in case of a flush error, if they
2245 * wish to. Caller must run wait_for_completion() on its own.
2247 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2251 if (bdev
->bd_disk
== NULL
)
2254 q
= bdev_get_queue(bdev
);
2257 if (!q
->issue_flush_fn
)
2260 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2263 EXPORT_SYMBOL(blkdev_issue_flush
);
2266 * blkdev_scsi_issue_flush_fn - issue flush for SCSI devices
2269 * @error_sector: error offset
2272 * Devices understanding the SCSI command set, can use this function as
2273 * a helper for issuing a cache flush. Note: driver is required to store
2274 * the error offset (in case of error flushing) in ->sector of struct
2277 int blkdev_scsi_issue_flush_fn(request_queue_t
*q
, struct gendisk
*disk
,
2278 sector_t
*error_sector
)
2280 struct request
*rq
= blk_get_request(q
, WRITE
, __GFP_WAIT
);
2283 rq
->flags
|= REQ_BLOCK_PC
| REQ_SOFTBARRIER
;
2285 memset(rq
->cmd
, 0, sizeof(rq
->cmd
));
2290 rq
->timeout
= 60 * HZ
;
2292 ret
= blk_execute_rq(q
, disk
, rq
);
2294 if (ret
&& error_sector
)
2295 *error_sector
= rq
->sector
;
2297 blk_put_request(rq
);
2301 EXPORT_SYMBOL(blkdev_scsi_issue_flush_fn
);
2303 void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2305 int rw
= rq_data_dir(rq
);
2307 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2311 __disk_stat_add(rq
->rq_disk
, read_sectors
, nr_sectors
);
2313 __disk_stat_inc(rq
->rq_disk
, read_merges
);
2314 } else if (rw
== WRITE
) {
2315 __disk_stat_add(rq
->rq_disk
, write_sectors
, nr_sectors
);
2317 __disk_stat_inc(rq
->rq_disk
, write_merges
);
2320 disk_round_stats(rq
->rq_disk
);
2321 rq
->rq_disk
->in_flight
++;
2326 * add-request adds a request to the linked list.
2327 * queue lock is held and interrupts disabled, as we muck with the
2328 * request queue list.
2330 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2332 drive_stat_acct(req
, req
->nr_sectors
, 1);
2335 q
->activity_fn(q
->activity_data
, rq_data_dir(req
));
2338 * elevator indicated where it wants this request to be
2339 * inserted at elevator_merge time
2341 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2345 * disk_round_stats() - Round off the performance stats on a struct
2348 * The average IO queue length and utilisation statistics are maintained
2349 * by observing the current state of the queue length and the amount of
2350 * time it has been in this state for.
2352 * Normally, that accounting is done on IO completion, but that can result
2353 * in more than a second's worth of IO being accounted for within any one
2354 * second, leading to >100% utilisation. To deal with that, we call this
2355 * function to do a round-off before returning the results when reading
2356 * /proc/diskstats. This accounts immediately for all queue usage up to
2357 * the current jiffies and restarts the counters again.
2359 void disk_round_stats(struct gendisk
*disk
)
2361 unsigned long now
= jiffies
;
2363 __disk_stat_add(disk
, time_in_queue
,
2364 disk
->in_flight
* (now
- disk
->stamp
));
2367 if (disk
->in_flight
)
2368 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp_idle
));
2369 disk
->stamp_idle
= now
;
2373 * queue lock must be held
2375 static void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2377 struct request_list
*rl
= req
->rl
;
2381 if (unlikely(--req
->ref_count
))
2384 req
->rq_status
= RQ_INACTIVE
;
2389 * Request may not have originated from ll_rw_blk. if not,
2390 * it didn't come out of our reserved rq pools
2393 int rw
= rq_data_dir(req
);
2395 elv_completed_request(q
, req
);
2397 BUG_ON(!list_empty(&req
->queuelist
));
2399 blk_free_request(q
, req
);
2400 freed_request(q
, rw
);
2404 void blk_put_request(struct request
*req
)
2407 * if req->rl isn't set, this request didnt originate from the
2408 * block layer, so it's safe to just disregard it
2411 unsigned long flags
;
2412 request_queue_t
*q
= req
->q
;
2414 spin_lock_irqsave(q
->queue_lock
, flags
);
2415 __blk_put_request(q
, req
);
2416 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2420 EXPORT_SYMBOL(blk_put_request
);
2423 * blk_end_sync_rq - executes a completion event on a request
2424 * @rq: request to complete
2426 void blk_end_sync_rq(struct request
*rq
)
2428 struct completion
*waiting
= rq
->waiting
;
2431 __blk_put_request(rq
->q
, rq
);
2434 * complete last, if this is a stack request the process (and thus
2435 * the rq pointer) could be invalid right after this complete()
2439 EXPORT_SYMBOL(blk_end_sync_rq
);
2442 * blk_congestion_wait - wait for a queue to become uncongested
2443 * @rw: READ or WRITE
2444 * @timeout: timeout in jiffies
2446 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2447 * If no queues are congested then just wait for the next request to be
2450 long blk_congestion_wait(int rw
, long timeout
)
2454 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
2456 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
2457 ret
= io_schedule_timeout(timeout
);
2458 finish_wait(wqh
, &wait
);
2462 EXPORT_SYMBOL(blk_congestion_wait
);
2465 * Has to be called with the request spinlock acquired
2467 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2468 struct request
*next
)
2470 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2476 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2479 if (rq_data_dir(req
) != rq_data_dir(next
)
2480 || req
->rq_disk
!= next
->rq_disk
2481 || next
->waiting
|| next
->special
)
2485 * If we are allowed to merge, then append bio list
2486 * from next to rq and release next. merge_requests_fn
2487 * will have updated segment counts, update sector
2490 if (!q
->merge_requests_fn(q
, req
, next
))
2494 * At this point we have either done a back merge
2495 * or front merge. We need the smaller start_time of
2496 * the merged requests to be the current request
2497 * for accounting purposes.
2499 if (time_after(req
->start_time
, next
->start_time
))
2500 req
->start_time
= next
->start_time
;
2502 req
->biotail
->bi_next
= next
->bio
;
2503 req
->biotail
= next
->biotail
;
2505 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2507 elv_merge_requests(q
, req
, next
);
2510 disk_round_stats(req
->rq_disk
);
2511 req
->rq_disk
->in_flight
--;
2514 __blk_put_request(q
, next
);
2518 static inline int attempt_back_merge(request_queue_t
*q
, struct request
*rq
)
2520 struct request
*next
= elv_latter_request(q
, rq
);
2523 return attempt_merge(q
, rq
, next
);
2528 static inline int attempt_front_merge(request_queue_t
*q
, struct request
*rq
)
2530 struct request
*prev
= elv_former_request(q
, rq
);
2533 return attempt_merge(q
, prev
, rq
);
2539 * blk_attempt_remerge - attempt to remerge active head with next request
2540 * @q: The &request_queue_t belonging to the device
2541 * @rq: The head request (usually)
2544 * For head-active devices, the queue can easily be unplugged so quickly
2545 * that proper merging is not done on the front request. This may hurt
2546 * performance greatly for some devices. The block layer cannot safely
2547 * do merging on that first request for these queues, but the driver can
2548 * call this function and make it happen any way. Only the driver knows
2549 * when it is safe to do so.
2551 void blk_attempt_remerge(request_queue_t
*q
, struct request
*rq
)
2553 unsigned long flags
;
2555 spin_lock_irqsave(q
->queue_lock
, flags
);
2556 attempt_back_merge(q
, rq
);
2557 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2560 EXPORT_SYMBOL(blk_attempt_remerge
);
2563 * Non-locking blk_attempt_remerge variant.
2565 void __blk_attempt_remerge(request_queue_t
*q
, struct request
*rq
)
2567 attempt_back_merge(q
, rq
);
2570 EXPORT_SYMBOL(__blk_attempt_remerge
);
2572 static int __make_request(request_queue_t
*q
, struct bio
*bio
)
2574 struct request
*req
, *freereq
= NULL
;
2575 int el_ret
, rw
, nr_sectors
, cur_nr_sectors
, barrier
, err
, sync
;
2578 sector
= bio
->bi_sector
;
2579 nr_sectors
= bio_sectors(bio
);
2580 cur_nr_sectors
= bio_cur_sectors(bio
);
2582 rw
= bio_data_dir(bio
);
2583 sync
= bio_sync(bio
);
2586 * low level driver can indicate that it wants pages above a
2587 * certain limit bounced to low memory (ie for highmem, or even
2588 * ISA dma in theory)
2590 blk_queue_bounce(q
, &bio
);
2592 spin_lock_prefetch(q
->queue_lock
);
2594 barrier
= bio_barrier(bio
);
2595 if (barrier
&& (q
->ordered
== QUEUE_ORDERED_NONE
)) {
2601 spin_lock_irq(q
->queue_lock
);
2603 if (elv_queue_empty(q
)) {
2610 el_ret
= elv_merge(q
, &req
, bio
);
2612 case ELEVATOR_BACK_MERGE
:
2613 BUG_ON(!rq_mergeable(req
));
2615 if (!q
->back_merge_fn(q
, req
, bio
))
2618 req
->biotail
->bi_next
= bio
;
2620 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2621 drive_stat_acct(req
, nr_sectors
, 0);
2622 if (!attempt_back_merge(q
, req
))
2623 elv_merged_request(q
, req
);
2626 case ELEVATOR_FRONT_MERGE
:
2627 BUG_ON(!rq_mergeable(req
));
2629 if (!q
->front_merge_fn(q
, req
, bio
))
2632 bio
->bi_next
= req
->bio
;
2636 * may not be valid. if the low level driver said
2637 * it didn't need a bounce buffer then it better
2638 * not touch req->buffer either...
2640 req
->buffer
= bio_data(bio
);
2641 req
->current_nr_sectors
= cur_nr_sectors
;
2642 req
->hard_cur_sectors
= cur_nr_sectors
;
2643 req
->sector
= req
->hard_sector
= sector
;
2644 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2645 drive_stat_acct(req
, nr_sectors
, 0);
2646 if (!attempt_front_merge(q
, req
))
2647 elv_merged_request(q
, req
);
2651 * elevator says don't/can't merge. get new request
2653 case ELEVATOR_NO_MERGE
:
2657 printk("elevator returned crap (%d)\n", el_ret
);
2662 * Grab a free request from the freelist - if that is empty, check
2663 * if we are doing read ahead and abort instead of blocking for
2671 spin_unlock_irq(q
->queue_lock
);
2672 if ((freereq
= get_request(q
, rw
, GFP_ATOMIC
)) == NULL
) {
2677 if (bio_rw_ahead(bio
))
2680 freereq
= get_request_wait(q
, rw
);
2685 req
->flags
|= REQ_CMD
;
2688 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2690 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2691 req
->flags
|= REQ_FAILFAST
;
2694 * REQ_BARRIER implies no merging, but lets make it explicit
2697 req
->flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2700 req
->hard_sector
= req
->sector
= sector
;
2701 req
->hard_nr_sectors
= req
->nr_sectors
= nr_sectors
;
2702 req
->current_nr_sectors
= req
->hard_cur_sectors
= cur_nr_sectors
;
2703 req
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2704 req
->nr_hw_segments
= bio_hw_segments(q
, bio
);
2705 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2706 req
->waiting
= NULL
;
2707 req
->bio
= req
->biotail
= bio
;
2708 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2709 req
->start_time
= jiffies
;
2711 add_request(q
, req
);
2714 __blk_put_request(q
, freereq
);
2716 __generic_unplug_device(q
);
2718 spin_unlock_irq(q
->queue_lock
);
2722 bio_endio(bio
, nr_sectors
<< 9, err
);
2727 * If bio->bi_dev is a partition, remap the location
2729 static inline void blk_partition_remap(struct bio
*bio
)
2731 struct block_device
*bdev
= bio
->bi_bdev
;
2733 if (bdev
!= bdev
->bd_contains
) {
2734 struct hd_struct
*p
= bdev
->bd_part
;
2736 switch (bio
->bi_rw
) {
2738 p
->read_sectors
+= bio_sectors(bio
);
2742 p
->write_sectors
+= bio_sectors(bio
);
2746 bio
->bi_sector
+= p
->start_sect
;
2747 bio
->bi_bdev
= bdev
->bd_contains
;
2751 void blk_finish_queue_drain(request_queue_t
*q
)
2753 struct request_list
*rl
= &q
->rq
;
2756 spin_lock_irq(q
->queue_lock
);
2757 clear_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
);
2759 while (!list_empty(&q
->drain_list
)) {
2760 rq
= list_entry_rq(q
->drain_list
.next
);
2762 list_del_init(&rq
->queuelist
);
2763 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 1);
2766 spin_unlock_irq(q
->queue_lock
);
2768 wake_up(&rl
->wait
[0]);
2769 wake_up(&rl
->wait
[1]);
2770 wake_up(&rl
->drain
);
2773 static int wait_drain(request_queue_t
*q
, struct request_list
*rl
, int dispatch
)
2775 int wait
= rl
->count
[READ
] + rl
->count
[WRITE
];
2778 wait
+= !list_empty(&q
->queue_head
);
2784 * We rely on the fact that only requests allocated through blk_alloc_request()
2785 * have io scheduler private data structures associated with them. Any other
2786 * type of request (allocated on stack or through kmalloc()) should not go
2787 * to the io scheduler core, but be attached to the queue head instead.
2789 void blk_wait_queue_drained(request_queue_t
*q
, int wait_dispatch
)
2791 struct request_list
*rl
= &q
->rq
;
2794 spin_lock_irq(q
->queue_lock
);
2795 set_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
);
2797 while (wait_drain(q
, rl
, wait_dispatch
)) {
2798 prepare_to_wait(&rl
->drain
, &wait
, TASK_UNINTERRUPTIBLE
);
2800 if (wait_drain(q
, rl
, wait_dispatch
)) {
2801 __generic_unplug_device(q
);
2802 spin_unlock_irq(q
->queue_lock
);
2804 spin_lock_irq(q
->queue_lock
);
2807 finish_wait(&rl
->drain
, &wait
);
2810 spin_unlock_irq(q
->queue_lock
);
2814 * block waiting for the io scheduler being started again.
2816 static inline void block_wait_queue_running(request_queue_t
*q
)
2820 while (test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
)) {
2821 struct request_list
*rl
= &q
->rq
;
2823 prepare_to_wait_exclusive(&rl
->drain
, &wait
,
2824 TASK_UNINTERRUPTIBLE
);
2827 * re-check the condition. avoids using prepare_to_wait()
2828 * in the fast path (queue is running)
2830 if (test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
))
2833 finish_wait(&rl
->drain
, &wait
);
2837 static void handle_bad_sector(struct bio
*bio
)
2839 char b
[BDEVNAME_SIZE
];
2841 printk(KERN_INFO
"attempt to access beyond end of device\n");
2842 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
2843 bdevname(bio
->bi_bdev
, b
),
2845 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
2846 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
2848 set_bit(BIO_EOF
, &bio
->bi_flags
);
2852 * generic_make_request: hand a buffer to its device driver for I/O
2853 * @bio: The bio describing the location in memory and on the device.
2855 * generic_make_request() is used to make I/O requests of block
2856 * devices. It is passed a &struct bio, which describes the I/O that needs
2859 * generic_make_request() does not return any status. The
2860 * success/failure status of the request, along with notification of
2861 * completion, is delivered asynchronously through the bio->bi_end_io
2862 * function described (one day) else where.
2864 * The caller of generic_make_request must make sure that bi_io_vec
2865 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2866 * set to describe the device address, and the
2867 * bi_end_io and optionally bi_private are set to describe how
2868 * completion notification should be signaled.
2870 * generic_make_request and the drivers it calls may use bi_next if this
2871 * bio happens to be merged with someone else, and may change bi_dev and
2872 * bi_sector for remaps as it sees fit. So the values of these fields
2873 * should NOT be depended on after the call to generic_make_request.
2875 void generic_make_request(struct bio
*bio
)
2879 int ret
, nr_sectors
= bio_sectors(bio
);
2882 /* Test device or partition size, when known. */
2883 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
2885 sector_t sector
= bio
->bi_sector
;
2887 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2889 * This may well happen - the kernel calls bread()
2890 * without checking the size of the device, e.g., when
2891 * mounting a device.
2893 handle_bad_sector(bio
);
2899 * Resolve the mapping until finished. (drivers are
2900 * still free to implement/resolve their own stacking
2901 * by explicitly returning 0)
2903 * NOTE: we don't repeat the blk_size check for each new device.
2904 * Stacking drivers are expected to know what they are doing.
2907 char b
[BDEVNAME_SIZE
];
2909 q
= bdev_get_queue(bio
->bi_bdev
);
2912 "generic_make_request: Trying to access "
2913 "nonexistent block-device %s (%Lu)\n",
2914 bdevname(bio
->bi_bdev
, b
),
2915 (long long) bio
->bi_sector
);
2917 bio_endio(bio
, bio
->bi_size
, -EIO
);
2921 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
2922 printk("bio too big device %s (%u > %u)\n",
2923 bdevname(bio
->bi_bdev
, b
),
2929 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))
2932 block_wait_queue_running(q
);
2935 * If this device has partitions, remap block n
2936 * of partition p to block n+start(p) of the disk.
2938 blk_partition_remap(bio
);
2940 ret
= q
->make_request_fn(q
, bio
);
2944 EXPORT_SYMBOL(generic_make_request
);
2947 * submit_bio: submit a bio to the block device layer for I/O
2948 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2949 * @bio: The &struct bio which describes the I/O
2951 * submit_bio() is very similar in purpose to generic_make_request(), and
2952 * uses that function to do most of the work. Both are fairly rough
2953 * interfaces, @bio must be presetup and ready for I/O.
2956 void submit_bio(int rw
, struct bio
*bio
)
2958 int count
= bio_sectors(bio
);
2960 BIO_BUG_ON(!bio
->bi_size
);
2961 BIO_BUG_ON(!bio
->bi_io_vec
);
2964 mod_page_state(pgpgout
, count
);
2966 mod_page_state(pgpgin
, count
);
2968 if (unlikely(block_dump
)) {
2969 char b
[BDEVNAME_SIZE
];
2970 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
2971 current
->comm
, current
->pid
,
2972 (rw
& WRITE
) ? "WRITE" : "READ",
2973 (unsigned long long)bio
->bi_sector
,
2974 bdevname(bio
->bi_bdev
,b
));
2977 generic_make_request(bio
);
2980 EXPORT_SYMBOL(submit_bio
);
2982 void blk_recalc_rq_segments(struct request
*rq
)
2984 struct bio
*bio
, *prevbio
= NULL
;
2985 int nr_phys_segs
, nr_hw_segs
;
2986 unsigned int phys_size
, hw_size
;
2987 request_queue_t
*q
= rq
->q
;
2992 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
2993 rq_for_each_bio(bio
, rq
) {
2994 /* Force bio hw/phys segs to be recalculated. */
2995 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
2997 nr_phys_segs
+= bio_phys_segments(q
, bio
);
2998 nr_hw_segs
+= bio_hw_segments(q
, bio
);
3000 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3001 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3003 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
3004 pseg
<= q
->max_segment_size
) {
3006 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3010 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
3011 hseg
<= q
->max_segment_size
) {
3013 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3020 rq
->nr_phys_segments
= nr_phys_segs
;
3021 rq
->nr_hw_segments
= nr_hw_segs
;
3024 void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3026 if (blk_fs_request(rq
)) {
3027 rq
->hard_sector
+= nsect
;
3028 rq
->hard_nr_sectors
-= nsect
;
3031 * Move the I/O submission pointers ahead if required.
3033 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3034 (rq
->sector
<= rq
->hard_sector
)) {
3035 rq
->sector
= rq
->hard_sector
;
3036 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3037 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3038 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3039 rq
->buffer
= bio_data(rq
->bio
);
3043 * if total number of sectors is less than the first segment
3044 * size, something has gone terribly wrong
3046 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3047 printk("blk: request botched\n");
3048 rq
->nr_sectors
= rq
->current_nr_sectors
;
3053 static int __end_that_request_first(struct request
*req
, int uptodate
,
3056 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3060 * extend uptodate bool to allow < 0 value to be direct io error
3063 if (end_io_error(uptodate
))
3064 error
= !uptodate
? -EIO
: uptodate
;
3067 * for a REQ_BLOCK_PC request, we want to carry any eventual
3068 * sense key with us all the way through
3070 if (!blk_pc_request(req
))
3074 if (blk_fs_request(req
) && !(req
->flags
& REQ_QUIET
))
3075 printk("end_request: I/O error, dev %s, sector %llu\n",
3076 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3077 (unsigned long long)req
->sector
);
3080 total_bytes
= bio_nbytes
= 0;
3081 while ((bio
= req
->bio
) != NULL
) {
3084 if (nr_bytes
>= bio
->bi_size
) {
3085 req
->bio
= bio
->bi_next
;
3086 nbytes
= bio
->bi_size
;
3087 bio_endio(bio
, nbytes
, error
);
3091 int idx
= bio
->bi_idx
+ next_idx
;
3093 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3094 blk_dump_rq_flags(req
, "__end_that");
3095 printk("%s: bio idx %d >= vcnt %d\n",
3097 bio
->bi_idx
, bio
->bi_vcnt
);
3101 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3102 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3105 * not a complete bvec done
3107 if (unlikely(nbytes
> nr_bytes
)) {
3108 bio_nbytes
+= nr_bytes
;
3109 total_bytes
+= nr_bytes
;
3114 * advance to the next vector
3117 bio_nbytes
+= nbytes
;
3120 total_bytes
+= nbytes
;
3123 if ((bio
= req
->bio
)) {
3125 * end more in this run, or just return 'not-done'
3127 if (unlikely(nr_bytes
<= 0))
3139 * if the request wasn't completed, update state
3142 bio_endio(bio
, bio_nbytes
, error
);
3143 bio
->bi_idx
+= next_idx
;
3144 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3145 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3148 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3149 blk_recalc_rq_segments(req
);
3154 * end_that_request_first - end I/O on a request
3155 * @req: the request being processed
3156 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3157 * @nr_sectors: number of sectors to end I/O on
3160 * Ends I/O on a number of sectors attached to @req, and sets it up
3161 * for the next range of segments (if any) in the cluster.
3164 * 0 - we are done with this request, call end_that_request_last()
3165 * 1 - still buffers pending for this request
3167 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3169 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3172 EXPORT_SYMBOL(end_that_request_first
);
3175 * end_that_request_chunk - end I/O on a request
3176 * @req: the request being processed
3177 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3178 * @nr_bytes: number of bytes to complete
3181 * Ends I/O on a number of bytes attached to @req, and sets it up
3182 * for the next range of segments (if any). Like end_that_request_first(),
3183 * but deals with bytes instead of sectors.
3186 * 0 - we are done with this request, call end_that_request_last()
3187 * 1 - still buffers pending for this request
3189 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3191 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3194 EXPORT_SYMBOL(end_that_request_chunk
);
3197 * queue lock must be held
3199 void end_that_request_last(struct request
*req
)
3201 struct gendisk
*disk
= req
->rq_disk
;
3203 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3204 laptop_io_completion();
3206 if (disk
&& blk_fs_request(req
)) {
3207 unsigned long duration
= jiffies
- req
->start_time
;
3208 switch (rq_data_dir(req
)) {
3210 __disk_stat_inc(disk
, writes
);
3211 __disk_stat_add(disk
, write_ticks
, duration
);
3214 __disk_stat_inc(disk
, reads
);
3215 __disk_stat_add(disk
, read_ticks
, duration
);
3218 disk_round_stats(disk
);
3224 __blk_put_request(req
->q
, req
);
3227 EXPORT_SYMBOL(end_that_request_last
);
3229 void end_request(struct request
*req
, int uptodate
)
3231 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3232 add_disk_randomness(req
->rq_disk
);
3233 blkdev_dequeue_request(req
);
3234 end_that_request_last(req
);
3238 EXPORT_SYMBOL(end_request
);
3240 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3242 /* first three bits are identical in rq->flags and bio->bi_rw */
3243 rq
->flags
|= (bio
->bi_rw
& 7);
3245 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3246 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3247 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3248 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3249 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3250 rq
->buffer
= bio_data(bio
);
3252 rq
->bio
= rq
->biotail
= bio
;
3255 EXPORT_SYMBOL(blk_rq_bio_prep
);
3257 int kblockd_schedule_work(struct work_struct
*work
)
3259 return queue_work(kblockd_workqueue
, work
);
3262 EXPORT_SYMBOL(kblockd_schedule_work
);
3264 void kblockd_flush(void)
3266 flush_workqueue(kblockd_workqueue
);
3268 EXPORT_SYMBOL(kblockd_flush
);
3270 int __init
blk_dev_init(void)
3272 kblockd_workqueue
= create_workqueue("kblockd");
3273 if (!kblockd_workqueue
)
3274 panic("Failed to create kblockd\n");
3276 request_cachep
= kmem_cache_create("blkdev_requests",
3277 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3279 requestq_cachep
= kmem_cache_create("blkdev_queue",
3280 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3282 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3283 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3285 blk_max_low_pfn
= max_low_pfn
;
3286 blk_max_pfn
= max_pfn
;
3292 * IO Context helper functions
3294 void put_io_context(struct io_context
*ioc
)
3299 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3301 if (atomic_dec_and_test(&ioc
->refcount
)) {
3302 if (ioc
->aic
&& ioc
->aic
->dtor
)
3303 ioc
->aic
->dtor(ioc
->aic
);
3304 if (ioc
->cic
&& ioc
->cic
->dtor
)
3305 ioc
->cic
->dtor(ioc
->cic
);
3307 kmem_cache_free(iocontext_cachep
, ioc
);
3310 EXPORT_SYMBOL(put_io_context
);
3312 /* Called by the exitting task */
3313 void exit_io_context(void)
3315 unsigned long flags
;
3316 struct io_context
*ioc
;
3318 local_irq_save(flags
);
3319 ioc
= current
->io_context
;
3320 current
->io_context
= NULL
;
3321 local_irq_restore(flags
);
3323 if (ioc
->aic
&& ioc
->aic
->exit
)
3324 ioc
->aic
->exit(ioc
->aic
);
3325 if (ioc
->cic
&& ioc
->cic
->exit
)
3326 ioc
->cic
->exit(ioc
->cic
);
3328 put_io_context(ioc
);
3332 * If the current task has no IO context then create one and initialise it.
3333 * If it does have a context, take a ref on it.
3335 * This is always called in the context of the task which submitted the I/O.
3336 * But weird things happen, so we disable local interrupts to ensure exclusive
3337 * access to *current.
3339 struct io_context
*get_io_context(int gfp_flags
)
3341 struct task_struct
*tsk
= current
;
3342 unsigned long flags
;
3343 struct io_context
*ret
;
3345 local_irq_save(flags
);
3346 ret
= tsk
->io_context
;
3350 local_irq_restore(flags
);
3352 ret
= kmem_cache_alloc(iocontext_cachep
, gfp_flags
);
3354 atomic_set(&ret
->refcount
, 1);
3355 ret
->pid
= tsk
->pid
;
3356 ret
->last_waited
= jiffies
; /* doesn't matter... */
3357 ret
->nr_batch_requests
= 0; /* because this is 0 */
3360 spin_lock_init(&ret
->lock
);
3362 local_irq_save(flags
);
3365 * very unlikely, someone raced with us in setting up the task
3366 * io context. free new context and just grab a reference.
3368 if (!tsk
->io_context
)
3369 tsk
->io_context
= ret
;
3371 kmem_cache_free(iocontext_cachep
, ret
);
3372 ret
= tsk
->io_context
;
3376 atomic_inc(&ret
->refcount
);
3377 local_irq_restore(flags
);
3382 EXPORT_SYMBOL(get_io_context
);
3384 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3386 struct io_context
*src
= *psrc
;
3387 struct io_context
*dst
= *pdst
;
3390 BUG_ON(atomic_read(&src
->refcount
) == 0);
3391 atomic_inc(&src
->refcount
);
3392 put_io_context(dst
);
3396 EXPORT_SYMBOL(copy_io_context
);
3398 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3400 struct io_context
*temp
;
3405 EXPORT_SYMBOL(swap_io_context
);
3410 struct queue_sysfs_entry
{
3411 struct attribute attr
;
3412 ssize_t (*show
)(struct request_queue
*, char *);
3413 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3417 queue_var_show(unsigned int var
, char *page
)
3419 return sprintf(page
, "%d\n", var
);
3423 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3425 char *p
= (char *) page
;
3427 *var
= simple_strtoul(p
, &p
, 10);
3431 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3433 return queue_var_show(q
->nr_requests
, (page
));
3437 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3439 struct request_list
*rl
= &q
->rq
;
3441 int ret
= queue_var_store(&q
->nr_requests
, page
, count
);
3442 if (q
->nr_requests
< BLKDEV_MIN_RQ
)
3443 q
->nr_requests
= BLKDEV_MIN_RQ
;
3444 blk_queue_congestion_threshold(q
);
3446 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3447 set_queue_congested(q
, READ
);
3448 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3449 clear_queue_congested(q
, READ
);
3451 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3452 set_queue_congested(q
, WRITE
);
3453 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3454 clear_queue_congested(q
, WRITE
);
3456 if (rl
->count
[READ
] >= q
->nr_requests
) {
3457 blk_set_queue_full(q
, READ
);
3458 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3459 blk_clear_queue_full(q
, READ
);
3460 wake_up(&rl
->wait
[READ
]);
3463 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3464 blk_set_queue_full(q
, WRITE
);
3465 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3466 blk_clear_queue_full(q
, WRITE
);
3467 wake_up(&rl
->wait
[WRITE
]);
3472 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3474 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3476 return queue_var_show(ra_kb
, (page
));
3480 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3482 unsigned long ra_kb
;
3483 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3485 spin_lock_irq(q
->queue_lock
);
3486 if (ra_kb
> (q
->max_sectors
>> 1))
3487 ra_kb
= (q
->max_sectors
>> 1);
3489 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3490 spin_unlock_irq(q
->queue_lock
);
3495 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3497 int max_sectors_kb
= q
->max_sectors
>> 1;
3499 return queue_var_show(max_sectors_kb
, (page
));
3503 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3505 unsigned long max_sectors_kb
,
3506 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3507 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3508 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3511 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3514 * Take the queue lock to update the readahead and max_sectors
3515 * values synchronously:
3517 spin_lock_irq(q
->queue_lock
);
3519 * Trim readahead window as well, if necessary:
3521 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3522 if (ra_kb
> max_sectors_kb
)
3523 q
->backing_dev_info
.ra_pages
=
3524 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3526 q
->max_sectors
= max_sectors_kb
<< 1;
3527 spin_unlock_irq(q
->queue_lock
);
3532 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3534 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3536 return queue_var_show(max_hw_sectors_kb
, (page
));
3540 static struct queue_sysfs_entry queue_requests_entry
= {
3541 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
3542 .show
= queue_requests_show
,
3543 .store
= queue_requests_store
,
3546 static struct queue_sysfs_entry queue_ra_entry
= {
3547 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
3548 .show
= queue_ra_show
,
3549 .store
= queue_ra_store
,
3552 static struct queue_sysfs_entry queue_max_sectors_entry
= {
3553 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
3554 .show
= queue_max_sectors_show
,
3555 .store
= queue_max_sectors_store
,
3558 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
3559 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
3560 .show
= queue_max_hw_sectors_show
,
3563 static struct queue_sysfs_entry queue_iosched_entry
= {
3564 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
3565 .show
= elv_iosched_show
,
3566 .store
= elv_iosched_store
,
3569 static struct attribute
*default_attrs
[] = {
3570 &queue_requests_entry
.attr
,
3571 &queue_ra_entry
.attr
,
3572 &queue_max_hw_sectors_entry
.attr
,
3573 &queue_max_sectors_entry
.attr
,
3574 &queue_iosched_entry
.attr
,
3578 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3581 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
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
->show(q
, page
);
3594 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3595 const char *page
, size_t length
)
3597 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3598 struct request_queue
*q
;
3600 q
= container_of(kobj
, struct request_queue
, kobj
);
3604 return entry
->store(q
, page
, length
);
3607 static struct sysfs_ops queue_sysfs_ops
= {
3608 .show
= queue_attr_show
,
3609 .store
= queue_attr_store
,
3612 struct kobj_type queue_ktype
= {
3613 .sysfs_ops
= &queue_sysfs_ops
,
3614 .default_attrs
= default_attrs
,
3617 int blk_register_queue(struct gendisk
*disk
)
3621 request_queue_t
*q
= disk
->queue
;
3623 if (!q
|| !q
->request_fn
)
3626 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
3627 if (!q
->kobj
.parent
)
3630 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
3631 q
->kobj
.ktype
= &queue_ktype
;
3633 ret
= kobject_register(&q
->kobj
);
3637 ret
= elv_register_queue(q
);
3639 kobject_unregister(&q
->kobj
);
3646 void blk_unregister_queue(struct gendisk
*disk
)
3648 request_queue_t
*q
= disk
->queue
;
3650 if (q
&& q
->request_fn
) {
3651 elv_unregister_queue(q
);
3653 kobject_unregister(&q
->kobj
);
3654 kobject_put(&disk
->kobj
);