2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
7 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
11 * This handles all read/write requests to block devices
13 #include <linux/config.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/blkdev.h>
34 #include <scsi/scsi_cmnd.h>
36 static void blk_unplug_work(void *data
);
37 static void blk_unplug_timeout(unsigned long data
);
38 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
39 static void init_request_from_bio(struct request
*req
, struct bio
*bio
);
40 static int __make_request(request_queue_t
*q
, struct bio
*bio
);
43 * For the allocated request tables
45 static kmem_cache_t
*request_cachep
;
48 * For queue allocation
50 static kmem_cache_t
*requestq_cachep
;
53 * For io context allocations
55 static kmem_cache_t
*iocontext_cachep
;
57 static wait_queue_head_t congestion_wqh
[2] = {
58 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
59 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
63 * Controlling structure to kblockd
65 static struct workqueue_struct
*kblockd_workqueue
;
67 unsigned long blk_max_low_pfn
, blk_max_pfn
;
69 EXPORT_SYMBOL(blk_max_low_pfn
);
70 EXPORT_SYMBOL(blk_max_pfn
);
72 /* Amount of time in which a process may batch requests */
73 #define BLK_BATCH_TIME (HZ/50UL)
75 /* Number of requests a "batching" process may submit */
76 #define BLK_BATCH_REQ 32
79 * Return the threshold (number of used requests) at which the queue is
80 * considered to be congested. It include a little hysteresis to keep the
81 * context switch rate down.
83 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
85 return q
->nr_congestion_on
;
89 * The threshold at which a queue is considered to be uncongested
91 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
93 return q
->nr_congestion_off
;
96 static void blk_queue_congestion_threshold(struct request_queue
*q
)
100 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
101 if (nr
> q
->nr_requests
)
103 q
->nr_congestion_on
= nr
;
105 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
108 q
->nr_congestion_off
= nr
;
112 * A queue has just exitted congestion. Note this in the global counter of
113 * congested queues, and wake up anyone who was waiting for requests to be
116 static void clear_queue_congested(request_queue_t
*q
, int rw
)
119 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
121 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
122 clear_bit(bit
, &q
->backing_dev_info
.state
);
123 smp_mb__after_clear_bit();
124 if (waitqueue_active(wqh
))
129 * A queue has just entered congestion. Flag that in the queue's VM-visible
130 * state flags and increment the global gounter of congested queues.
132 static void set_queue_congested(request_queue_t
*q
, int rw
)
136 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
137 set_bit(bit
, &q
->backing_dev_info
.state
);
141 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
144 * Locates the passed device's request queue and returns the address of its
147 * Will return NULL if the request queue cannot be located.
149 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
151 struct backing_dev_info
*ret
= NULL
;
152 request_queue_t
*q
= bdev_get_queue(bdev
);
155 ret
= &q
->backing_dev_info
;
159 EXPORT_SYMBOL(blk_get_backing_dev_info
);
161 void blk_queue_activity_fn(request_queue_t
*q
, activity_fn
*fn
, void *data
)
164 q
->activity_data
= data
;
167 EXPORT_SYMBOL(blk_queue_activity_fn
);
170 * blk_queue_prep_rq - set a prepare_request function for queue
172 * @pfn: prepare_request function
174 * It's possible for a queue to register a prepare_request callback which
175 * is invoked before the request is handed to the request_fn. The goal of
176 * the function is to prepare a request for I/O, it can be used to build a
177 * cdb from the request data for instance.
180 void blk_queue_prep_rq(request_queue_t
*q
, prep_rq_fn
*pfn
)
185 EXPORT_SYMBOL(blk_queue_prep_rq
);
188 * blk_queue_merge_bvec - set a merge_bvec function for queue
190 * @mbfn: merge_bvec_fn
192 * Usually queues have static limitations on the max sectors or segments that
193 * we can put in a request. Stacking drivers may have some settings that
194 * are dynamic, and thus we have to query the queue whether it is ok to
195 * add a new bio_vec to a bio at a given offset or not. If the block device
196 * has such limitations, it needs to register a merge_bvec_fn to control
197 * the size of bio's sent to it. Note that a block device *must* allow a
198 * single page to be added to an empty bio. The block device driver may want
199 * to use the bio_split() function to deal with these bio's. By default
200 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
203 void blk_queue_merge_bvec(request_queue_t
*q
, merge_bvec_fn
*mbfn
)
205 q
->merge_bvec_fn
= mbfn
;
208 EXPORT_SYMBOL(blk_queue_merge_bvec
);
211 * blk_queue_make_request - define an alternate make_request function for a device
212 * @q: the request queue for the device to be affected
213 * @mfn: the alternate make_request function
216 * The normal way for &struct bios to be passed to a device
217 * driver is for them to be collected into requests on a request
218 * queue, and then to allow the device driver to select requests
219 * off that queue when it is ready. This works well for many block
220 * devices. However some block devices (typically virtual devices
221 * such as md or lvm) do not benefit from the processing on the
222 * request queue, and are served best by having the requests passed
223 * directly to them. This can be achieved by providing a function
224 * to blk_queue_make_request().
227 * The driver that does this *must* be able to deal appropriately
228 * with buffers in "highmemory". This can be accomplished by either calling
229 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
230 * blk_queue_bounce() to create a buffer in normal memory.
232 void blk_queue_make_request(request_queue_t
* q
, make_request_fn
* mfn
)
237 q
->nr_requests
= BLKDEV_MAX_RQ
;
238 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
239 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
240 q
->make_request_fn
= mfn
;
241 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
242 q
->backing_dev_info
.state
= 0;
243 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
244 blk_queue_max_sectors(q
, SAFE_MAX_SECTORS
);
245 blk_queue_hardsect_size(q
, 512);
246 blk_queue_dma_alignment(q
, 511);
247 blk_queue_congestion_threshold(q
);
248 q
->nr_batching
= BLK_BATCH_REQ
;
250 q
->unplug_thresh
= 4; /* hmm */
251 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
252 if (q
->unplug_delay
== 0)
255 INIT_WORK(&q
->unplug_work
, blk_unplug_work
, q
);
257 q
->unplug_timer
.function
= blk_unplug_timeout
;
258 q
->unplug_timer
.data
= (unsigned long)q
;
261 * by default assume old behaviour and bounce for any highmem page
263 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
265 blk_queue_activity_fn(q
, NULL
, NULL
);
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
;
285 rq
->nr_phys_segments
= 0;
288 rq
->end_io_data
= NULL
;
292 * blk_queue_ordered - does this queue support ordered writes
293 * @q: the request queue
297 * For journalled file systems, doing ordered writes on a commit
298 * block instead of explicitly doing wait_on_buffer (which is bad
299 * for performance) can be a big win. Block drivers supporting this
300 * feature should call this function and indicate so.
303 void blk_queue_ordered(request_queue_t
*q
, int flag
)
306 case QUEUE_ORDERED_NONE
:
308 kmem_cache_free(request_cachep
, q
->flush_rq
);
312 case QUEUE_ORDERED_TAG
:
315 case QUEUE_ORDERED_FLUSH
:
318 q
->flush_rq
= kmem_cache_alloc(request_cachep
,
322 printk("blk_queue_ordered: bad value %d\n", flag
);
327 EXPORT_SYMBOL(blk_queue_ordered
);
330 * blk_queue_issue_flush_fn - set function for issuing a flush
331 * @q: the request queue
332 * @iff: the function to be called issuing the flush
335 * If a driver supports issuing a flush command, the support is notified
336 * to the block layer by defining it through this call.
339 void blk_queue_issue_flush_fn(request_queue_t
*q
, issue_flush_fn
*iff
)
341 q
->issue_flush_fn
= iff
;
344 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
347 * Cache flushing for ordered writes handling
349 static void blk_pre_flush_end_io(struct request
*flush_rq
, int error
)
351 struct request
*rq
= flush_rq
->end_io_data
;
352 request_queue_t
*q
= rq
->q
;
354 elv_completed_request(q
, flush_rq
);
356 rq
->flags
|= REQ_BAR_PREFLUSH
;
358 if (!flush_rq
->errors
)
359 elv_requeue_request(q
, rq
);
361 q
->end_flush_fn(q
, flush_rq
);
362 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
367 static void blk_post_flush_end_io(struct request
*flush_rq
, int error
)
369 struct request
*rq
= flush_rq
->end_io_data
;
370 request_queue_t
*q
= rq
->q
;
372 elv_completed_request(q
, flush_rq
);
374 rq
->flags
|= REQ_BAR_POSTFLUSH
;
376 q
->end_flush_fn(q
, flush_rq
);
377 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
381 struct request
*blk_start_pre_flush(request_queue_t
*q
, struct request
*rq
)
383 struct request
*flush_rq
= q
->flush_rq
;
385 BUG_ON(!blk_barrier_rq(rq
));
387 if (test_and_set_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
))
390 rq_init(q
, flush_rq
);
391 flush_rq
->elevator_private
= NULL
;
392 flush_rq
->flags
= REQ_BAR_FLUSH
;
393 flush_rq
->rq_disk
= rq
->rq_disk
;
397 * prepare_flush returns 0 if no flush is needed, just mark both
398 * pre and post flush as done in that case
400 if (!q
->prepare_flush_fn(q
, flush_rq
)) {
401 rq
->flags
|= REQ_BAR_PREFLUSH
| REQ_BAR_POSTFLUSH
;
402 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
407 * some drivers dequeue requests right away, some only after io
408 * completion. make sure the request is dequeued.
410 if (!list_empty(&rq
->queuelist
))
411 blkdev_dequeue_request(rq
);
413 flush_rq
->end_io_data
= rq
;
414 flush_rq
->end_io
= blk_pre_flush_end_io
;
416 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
420 static void blk_start_post_flush(request_queue_t
*q
, struct request
*rq
)
422 struct request
*flush_rq
= q
->flush_rq
;
424 BUG_ON(!blk_barrier_rq(rq
));
426 rq_init(q
, flush_rq
);
427 flush_rq
->elevator_private
= NULL
;
428 flush_rq
->flags
= REQ_BAR_FLUSH
;
429 flush_rq
->rq_disk
= rq
->rq_disk
;
432 if (q
->prepare_flush_fn(q
, flush_rq
)) {
433 flush_rq
->end_io_data
= rq
;
434 flush_rq
->end_io
= blk_post_flush_end_io
;
436 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
441 static inline int blk_check_end_barrier(request_queue_t
*q
, struct request
*rq
,
444 if (sectors
> rq
->nr_sectors
)
445 sectors
= rq
->nr_sectors
;
447 rq
->nr_sectors
-= sectors
;
448 return rq
->nr_sectors
;
451 static int __blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
,
452 int sectors
, int queue_locked
)
454 if (q
->ordered
!= QUEUE_ORDERED_FLUSH
)
456 if (!blk_fs_request(rq
) || !blk_barrier_rq(rq
))
458 if (blk_barrier_postflush(rq
))
461 if (!blk_check_end_barrier(q
, rq
, sectors
)) {
462 unsigned long flags
= 0;
465 spin_lock_irqsave(q
->queue_lock
, flags
);
467 blk_start_post_flush(q
, rq
);
470 spin_unlock_irqrestore(q
->queue_lock
, flags
);
477 * blk_complete_barrier_rq - complete possible barrier request
478 * @q: the request queue for the device
480 * @sectors: number of sectors to complete
483 * Used in driver end_io handling to determine whether to postpone
484 * completion of a barrier request until a post flush has been done. This
485 * is the unlocked variant, used if the caller doesn't already hold the
488 int blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
, int sectors
)
490 return __blk_complete_barrier_rq(q
, rq
, sectors
, 0);
492 EXPORT_SYMBOL(blk_complete_barrier_rq
);
495 * blk_complete_barrier_rq_locked - complete possible barrier request
496 * @q: the request queue for the device
498 * @sectors: number of sectors to complete
501 * See blk_complete_barrier_rq(). This variant must be used if the caller
502 * holds the queue lock.
504 int blk_complete_barrier_rq_locked(request_queue_t
*q
, struct request
*rq
,
507 return __blk_complete_barrier_rq(q
, rq
, sectors
, 1);
509 EXPORT_SYMBOL(blk_complete_barrier_rq_locked
);
512 * blk_queue_bounce_limit - set bounce buffer limit for queue
513 * @q: the request queue for the device
514 * @dma_addr: bus address limit
517 * Different hardware can have different requirements as to what pages
518 * it can do I/O directly to. A low level driver can call
519 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
520 * buffers for doing I/O to pages residing above @page. By default
521 * the block layer sets this to the highest numbered "low" memory page.
523 void blk_queue_bounce_limit(request_queue_t
*q
, u64 dma_addr
)
525 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
528 * set appropriate bounce gfp mask -- unfortunately we don't have a
529 * full 4GB zone, so we have to resort to low memory for any bounces.
530 * ISA has its own < 16MB zone.
532 if (bounce_pfn
< blk_max_low_pfn
) {
533 BUG_ON(dma_addr
< BLK_BOUNCE_ISA
);
534 init_emergency_isa_pool();
535 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
537 q
->bounce_gfp
= GFP_NOIO
;
539 q
->bounce_pfn
= bounce_pfn
;
542 EXPORT_SYMBOL(blk_queue_bounce_limit
);
545 * blk_queue_max_sectors - set max sectors for a request for this queue
546 * @q: the request queue for the device
547 * @max_sectors: max sectors in the usual 512b unit
550 * Enables a low level driver to set an upper limit on the size of
553 void blk_queue_max_sectors(request_queue_t
*q
, unsigned short max_sectors
)
555 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
556 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
557 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
560 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
561 q
->max_hw_sectors
= q
->max_sectors
= max_sectors
;
563 q
->max_sectors
= BLK_DEF_MAX_SECTORS
;
564 q
->max_hw_sectors
= max_sectors
;
568 EXPORT_SYMBOL(blk_queue_max_sectors
);
571 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
572 * @q: the request queue for the device
573 * @max_segments: max number of segments
576 * Enables a low level driver to set an upper limit on the number of
577 * physical data segments in a request. This would be the largest sized
578 * scatter list the driver could handle.
580 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
584 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
587 q
->max_phys_segments
= max_segments
;
590 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
593 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
594 * @q: the request queue for the device
595 * @max_segments: max number of segments
598 * Enables a low level driver to set an upper limit on the number of
599 * hw data segments in a request. This would be the largest number of
600 * address/length pairs the host adapter can actually give as once
603 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
607 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
610 q
->max_hw_segments
= max_segments
;
613 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
616 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
617 * @q: the request queue for the device
618 * @max_size: max size of segment in bytes
621 * Enables a low level driver to set an upper limit on the size of a
624 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
626 if (max_size
< PAGE_CACHE_SIZE
) {
627 max_size
= PAGE_CACHE_SIZE
;
628 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
631 q
->max_segment_size
= max_size
;
634 EXPORT_SYMBOL(blk_queue_max_segment_size
);
637 * blk_queue_hardsect_size - set hardware sector size for the queue
638 * @q: the request queue for the device
639 * @size: the hardware sector size, in bytes
642 * This should typically be set to the lowest possible sector size
643 * that the hardware can operate on (possible without reverting to
644 * even internal read-modify-write operations). Usually the default
645 * of 512 covers most hardware.
647 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
649 q
->hardsect_size
= size
;
652 EXPORT_SYMBOL(blk_queue_hardsect_size
);
655 * Returns the minimum that is _not_ zero, unless both are zero.
657 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
660 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
661 * @t: the stacking driver (top)
662 * @b: the underlying device (bottom)
664 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
666 /* zero is "infinity" */
667 t
->max_sectors
= min_not_zero(t
->max_sectors
,b
->max_sectors
);
668 t
->max_hw_sectors
= min_not_zero(t
->max_hw_sectors
,b
->max_hw_sectors
);
670 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
671 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
672 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
673 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
676 EXPORT_SYMBOL(blk_queue_stack_limits
);
679 * blk_queue_segment_boundary - set boundary rules for segment merging
680 * @q: the request queue for the device
681 * @mask: the memory boundary mask
683 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
685 if (mask
< PAGE_CACHE_SIZE
- 1) {
686 mask
= PAGE_CACHE_SIZE
- 1;
687 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
690 q
->seg_boundary_mask
= mask
;
693 EXPORT_SYMBOL(blk_queue_segment_boundary
);
696 * blk_queue_dma_alignment - set dma length and memory alignment
697 * @q: the request queue for the device
698 * @mask: alignment mask
701 * set required memory and length aligment for direct dma transactions.
702 * this is used when buiding direct io requests for the queue.
705 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
707 q
->dma_alignment
= mask
;
710 EXPORT_SYMBOL(blk_queue_dma_alignment
);
713 * blk_queue_find_tag - find a request by its tag and queue
714 * @q: The request queue for the device
715 * @tag: The tag of the request
718 * Should be used when a device returns a tag and you want to match
721 * no locks need be held.
723 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
725 struct blk_queue_tag
*bqt
= q
->queue_tags
;
727 if (unlikely(bqt
== NULL
|| tag
>= bqt
->real_max_depth
))
730 return bqt
->tag_index
[tag
];
733 EXPORT_SYMBOL(blk_queue_find_tag
);
736 * __blk_queue_free_tags - release tag maintenance info
737 * @q: the request queue for the device
740 * blk_cleanup_queue() will take care of calling this function, if tagging
741 * has been used. So there's no need to call this directly.
743 static void __blk_queue_free_tags(request_queue_t
*q
)
745 struct blk_queue_tag
*bqt
= q
->queue_tags
;
750 if (atomic_dec_and_test(&bqt
->refcnt
)) {
752 BUG_ON(!list_empty(&bqt
->busy_list
));
754 kfree(bqt
->tag_index
);
755 bqt
->tag_index
= NULL
;
763 q
->queue_tags
= NULL
;
764 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
768 * blk_queue_free_tags - release tag maintenance info
769 * @q: the request queue for the device
772 * This is used to disabled tagged queuing to a device, yet leave
775 void blk_queue_free_tags(request_queue_t
*q
)
777 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
780 EXPORT_SYMBOL(blk_queue_free_tags
);
783 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
785 struct request
**tag_index
;
786 unsigned long *tag_map
;
789 if (depth
> q
->nr_requests
* 2) {
790 depth
= q
->nr_requests
* 2;
791 printk(KERN_ERR
"%s: adjusted depth to %d\n",
792 __FUNCTION__
, depth
);
795 tag_index
= kmalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
799 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
800 tag_map
= kmalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
804 memset(tag_index
, 0, depth
* sizeof(struct request
*));
805 memset(tag_map
, 0, nr_ulongs
* sizeof(unsigned long));
806 tags
->real_max_depth
= depth
;
807 tags
->max_depth
= depth
;
808 tags
->tag_index
= tag_index
;
809 tags
->tag_map
= tag_map
;
818 * blk_queue_init_tags - initialize the queue tag info
819 * @q: the request queue for the device
820 * @depth: the maximum queue depth supported
821 * @tags: the tag to use
823 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
824 struct blk_queue_tag
*tags
)
828 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
830 if (!tags
&& !q
->queue_tags
) {
831 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
835 if (init_tag_map(q
, tags
, depth
))
838 INIT_LIST_HEAD(&tags
->busy_list
);
840 atomic_set(&tags
->refcnt
, 1);
841 } else if (q
->queue_tags
) {
842 if ((rc
= blk_queue_resize_tags(q
, depth
)))
844 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
847 atomic_inc(&tags
->refcnt
);
850 * assign it, all done
852 q
->queue_tags
= tags
;
853 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
860 EXPORT_SYMBOL(blk_queue_init_tags
);
863 * blk_queue_resize_tags - change the queueing depth
864 * @q: the request queue for the device
865 * @new_depth: the new max command queueing depth
868 * Must be called with the queue lock held.
870 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
872 struct blk_queue_tag
*bqt
= q
->queue_tags
;
873 struct request
**tag_index
;
874 unsigned long *tag_map
;
875 int max_depth
, nr_ulongs
;
881 * if we already have large enough real_max_depth. just
882 * adjust max_depth. *NOTE* as requests with tag value
883 * between new_depth and real_max_depth can be in-flight, tag
884 * map can not be shrunk blindly here.
886 if (new_depth
<= bqt
->real_max_depth
) {
887 bqt
->max_depth
= new_depth
;
892 * save the old state info, so we can copy it back
894 tag_index
= bqt
->tag_index
;
895 tag_map
= bqt
->tag_map
;
896 max_depth
= bqt
->real_max_depth
;
898 if (init_tag_map(q
, bqt
, new_depth
))
901 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
902 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
903 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
910 EXPORT_SYMBOL(blk_queue_resize_tags
);
913 * blk_queue_end_tag - end tag operations for a request
914 * @q: the request queue for the device
915 * @rq: the request that has completed
918 * Typically called when end_that_request_first() returns 0, meaning
919 * all transfers have been done for a request. It's important to call
920 * this function before end_that_request_last(), as that will put the
921 * request back on the free list thus corrupting the internal tag list.
924 * queue lock must be held.
926 void blk_queue_end_tag(request_queue_t
*q
, struct request
*rq
)
928 struct blk_queue_tag
*bqt
= q
->queue_tags
;
933 if (unlikely(tag
>= bqt
->real_max_depth
))
935 * This can happen after tag depth has been reduced.
936 * FIXME: how about a warning or info message here?
940 if (unlikely(!__test_and_clear_bit(tag
, bqt
->tag_map
))) {
941 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
946 list_del_init(&rq
->queuelist
);
947 rq
->flags
&= ~REQ_QUEUED
;
950 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
951 printk(KERN_ERR
"%s: tag %d is missing\n",
954 bqt
->tag_index
[tag
] = NULL
;
958 EXPORT_SYMBOL(blk_queue_end_tag
);
961 * blk_queue_start_tag - find a free tag and assign it
962 * @q: the request queue for the device
963 * @rq: the block request that needs tagging
966 * This can either be used as a stand-alone helper, or possibly be
967 * assigned as the queue &prep_rq_fn (in which case &struct request
968 * automagically gets a tag assigned). Note that this function
969 * assumes that any type of request can be queued! if this is not
970 * true for your device, you must check the request type before
971 * calling this function. The request will also be removed from
972 * the request queue, so it's the drivers responsibility to readd
973 * it if it should need to be restarted for some reason.
976 * queue lock must be held.
978 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
980 struct blk_queue_tag
*bqt
= q
->queue_tags
;
983 if (unlikely((rq
->flags
& REQ_QUEUED
))) {
985 "%s: request %p for device [%s] already tagged %d",
987 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
991 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
992 if (tag
>= bqt
->max_depth
)
995 __set_bit(tag
, bqt
->tag_map
);
997 rq
->flags
|= REQ_QUEUED
;
999 bqt
->tag_index
[tag
] = rq
;
1000 blkdev_dequeue_request(rq
);
1001 list_add(&rq
->queuelist
, &bqt
->busy_list
);
1006 EXPORT_SYMBOL(blk_queue_start_tag
);
1009 * blk_queue_invalidate_tags - invalidate all pending tags
1010 * @q: the request queue for the device
1013 * Hardware conditions may dictate a need to stop all pending requests.
1014 * In this case, we will safely clear the block side of the tag queue and
1015 * readd all requests to the request queue in the right order.
1018 * queue lock must be held.
1020 void blk_queue_invalidate_tags(request_queue_t
*q
)
1022 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1023 struct list_head
*tmp
, *n
;
1026 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1027 rq
= list_entry_rq(tmp
);
1029 if (rq
->tag
== -1) {
1031 "%s: bad tag found on list\n", __FUNCTION__
);
1032 list_del_init(&rq
->queuelist
);
1033 rq
->flags
&= ~REQ_QUEUED
;
1035 blk_queue_end_tag(q
, rq
);
1037 rq
->flags
&= ~REQ_STARTED
;
1038 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1042 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1044 static const char * const rq_flags
[] = {
1064 "REQ_DRIVE_TASKFILE",
1071 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1075 printk("%s: dev %s: flags = ", msg
,
1076 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?");
1079 if (rq
->flags
& (1 << bit
))
1080 printk("%s ", rq_flags
[bit
]);
1082 } while (bit
< __REQ_NR_BITS
);
1084 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1086 rq
->current_nr_sectors
);
1087 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1089 if (rq
->flags
& (REQ_BLOCK_PC
| REQ_PC
)) {
1091 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1092 printk("%02x ", rq
->cmd
[bit
]);
1097 EXPORT_SYMBOL(blk_dump_rq_flags
);
1099 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1101 struct bio_vec
*bv
, *bvprv
= NULL
;
1102 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1103 int high
, highprv
= 1;
1105 if (unlikely(!bio
->bi_io_vec
))
1108 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1109 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1110 bio_for_each_segment(bv
, bio
, i
) {
1112 * the trick here is making sure that a high page is never
1113 * considered part of another segment, since that might
1114 * change with the bounce page.
1116 high
= page_to_pfn(bv
->bv_page
) >= q
->bounce_pfn
;
1117 if (high
|| highprv
)
1118 goto new_hw_segment
;
1120 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1122 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1124 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1126 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1127 goto new_hw_segment
;
1129 seg_size
+= bv
->bv_len
;
1130 hw_seg_size
+= bv
->bv_len
;
1135 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1136 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1137 hw_seg_size
+= bv
->bv_len
;
1140 if (hw_seg_size
> bio
->bi_hw_front_size
)
1141 bio
->bi_hw_front_size
= hw_seg_size
;
1142 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1148 seg_size
= bv
->bv_len
;
1151 if (hw_seg_size
> bio
->bi_hw_back_size
)
1152 bio
->bi_hw_back_size
= hw_seg_size
;
1153 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1154 bio
->bi_hw_front_size
= hw_seg_size
;
1155 bio
->bi_phys_segments
= nr_phys_segs
;
1156 bio
->bi_hw_segments
= nr_hw_segs
;
1157 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1161 static int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1164 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1167 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1169 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1173 * bio and nxt are contigous in memory, check if the queue allows
1174 * these two to be merged into one
1176 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1182 static int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1185 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1186 blk_recount_segments(q
, bio
);
1187 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1188 blk_recount_segments(q
, nxt
);
1189 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1190 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1192 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1199 * map a request to scatterlist, return number of sg entries setup. Caller
1200 * must make sure sg can hold rq->nr_phys_segments entries
1202 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1204 struct bio_vec
*bvec
, *bvprv
;
1206 int nsegs
, i
, cluster
;
1209 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1212 * for each bio in rq
1215 rq_for_each_bio(bio
, rq
) {
1217 * for each segment in bio
1219 bio_for_each_segment(bvec
, bio
, i
) {
1220 int nbytes
= bvec
->bv_len
;
1222 if (bvprv
&& cluster
) {
1223 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1226 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1228 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1231 sg
[nsegs
- 1].length
+= nbytes
;
1234 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1235 sg
[nsegs
].page
= bvec
->bv_page
;
1236 sg
[nsegs
].length
= nbytes
;
1237 sg
[nsegs
].offset
= bvec
->bv_offset
;
1242 } /* segments in bio */
1248 EXPORT_SYMBOL(blk_rq_map_sg
);
1251 * the standard queue merge functions, can be overridden with device
1252 * specific ones if so desired
1255 static inline int ll_new_mergeable(request_queue_t
*q
,
1256 struct request
*req
,
1259 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1261 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1262 req
->flags
|= REQ_NOMERGE
;
1263 if (req
== q
->last_merge
)
1264 q
->last_merge
= NULL
;
1269 * A hw segment is just getting larger, bump just the phys
1272 req
->nr_phys_segments
+= nr_phys_segs
;
1276 static inline int ll_new_hw_segment(request_queue_t
*q
,
1277 struct request
*req
,
1280 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1281 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1283 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1284 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1285 req
->flags
|= REQ_NOMERGE
;
1286 if (req
== q
->last_merge
)
1287 q
->last_merge
= NULL
;
1292 * This will form the start of a new hw segment. Bump both
1295 req
->nr_hw_segments
+= nr_hw_segs
;
1296 req
->nr_phys_segments
+= nr_phys_segs
;
1300 static int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
,
1303 unsigned short max_sectors
;
1306 if (unlikely(blk_pc_request(req
)))
1307 max_sectors
= q
->max_hw_sectors
;
1309 max_sectors
= q
->max_sectors
;
1311 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1312 req
->flags
|= REQ_NOMERGE
;
1313 if (req
== q
->last_merge
)
1314 q
->last_merge
= NULL
;
1317 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1318 blk_recount_segments(q
, req
->biotail
);
1319 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1320 blk_recount_segments(q
, bio
);
1321 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1322 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1323 !BIOVEC_VIRT_OVERSIZE(len
)) {
1324 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1327 if (req
->nr_hw_segments
== 1)
1328 req
->bio
->bi_hw_front_size
= len
;
1329 if (bio
->bi_hw_segments
== 1)
1330 bio
->bi_hw_back_size
= len
;
1335 return ll_new_hw_segment(q
, req
, bio
);
1338 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1341 unsigned short max_sectors
;
1344 if (unlikely(blk_pc_request(req
)))
1345 max_sectors
= q
->max_hw_sectors
;
1347 max_sectors
= q
->max_sectors
;
1350 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1351 req
->flags
|= REQ_NOMERGE
;
1352 if (req
== q
->last_merge
)
1353 q
->last_merge
= NULL
;
1356 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1357 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1358 blk_recount_segments(q
, bio
);
1359 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1360 blk_recount_segments(q
, req
->bio
);
1361 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1362 !BIOVEC_VIRT_OVERSIZE(len
)) {
1363 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1366 if (bio
->bi_hw_segments
== 1)
1367 bio
->bi_hw_front_size
= len
;
1368 if (req
->nr_hw_segments
== 1)
1369 req
->biotail
->bi_hw_back_size
= len
;
1374 return ll_new_hw_segment(q
, req
, bio
);
1377 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1378 struct request
*next
)
1380 int total_phys_segments
;
1381 int total_hw_segments
;
1384 * First check if the either of the requests are re-queued
1385 * requests. Can't merge them if they are.
1387 if (req
->special
|| next
->special
)
1391 * Will it become too large?
1393 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1396 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1397 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1398 total_phys_segments
--;
1400 if (total_phys_segments
> q
->max_phys_segments
)
1403 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1404 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1405 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1407 * propagate the combined length to the end of the requests
1409 if (req
->nr_hw_segments
== 1)
1410 req
->bio
->bi_hw_front_size
= len
;
1411 if (next
->nr_hw_segments
== 1)
1412 next
->biotail
->bi_hw_back_size
= len
;
1413 total_hw_segments
--;
1416 if (total_hw_segments
> q
->max_hw_segments
)
1419 /* Merge is OK... */
1420 req
->nr_phys_segments
= total_phys_segments
;
1421 req
->nr_hw_segments
= total_hw_segments
;
1426 * "plug" the device if there are no outstanding requests: this will
1427 * force the transfer to start only after we have put all the requests
1430 * This is called with interrupts off and no requests on the queue and
1431 * with the queue lock held.
1433 void blk_plug_device(request_queue_t
*q
)
1435 WARN_ON(!irqs_disabled());
1438 * don't plug a stopped queue, it must be paired with blk_start_queue()
1439 * which will restart the queueing
1441 if (test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
))
1444 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1445 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1448 EXPORT_SYMBOL(blk_plug_device
);
1451 * remove the queue from the plugged list, if present. called with
1452 * queue lock held and interrupts disabled.
1454 int blk_remove_plug(request_queue_t
*q
)
1456 WARN_ON(!irqs_disabled());
1458 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1461 del_timer(&q
->unplug_timer
);
1465 EXPORT_SYMBOL(blk_remove_plug
);
1468 * remove the plug and let it rip..
1470 void __generic_unplug_device(request_queue_t
*q
)
1472 if (unlikely(test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
)))
1475 if (!blk_remove_plug(q
))
1480 EXPORT_SYMBOL(__generic_unplug_device
);
1483 * generic_unplug_device - fire a request queue
1484 * @q: The &request_queue_t in question
1487 * Linux uses plugging to build bigger requests queues before letting
1488 * the device have at them. If a queue is plugged, the I/O scheduler
1489 * is still adding and merging requests on the queue. Once the queue
1490 * gets unplugged, the request_fn defined for the queue is invoked and
1491 * transfers started.
1493 void generic_unplug_device(request_queue_t
*q
)
1495 spin_lock_irq(q
->queue_lock
);
1496 __generic_unplug_device(q
);
1497 spin_unlock_irq(q
->queue_lock
);
1499 EXPORT_SYMBOL(generic_unplug_device
);
1501 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1504 request_queue_t
*q
= bdi
->unplug_io_data
;
1507 * devices don't necessarily have an ->unplug_fn defined
1513 static void blk_unplug_work(void *data
)
1515 request_queue_t
*q
= data
;
1520 static void blk_unplug_timeout(unsigned long data
)
1522 request_queue_t
*q
= (request_queue_t
*)data
;
1524 kblockd_schedule_work(&q
->unplug_work
);
1528 * blk_start_queue - restart a previously stopped queue
1529 * @q: The &request_queue_t in question
1532 * blk_start_queue() will clear the stop flag on the queue, and call
1533 * the request_fn for the queue if it was in a stopped state when
1534 * entered. Also see blk_stop_queue(). Queue lock must be held.
1536 void blk_start_queue(request_queue_t
*q
)
1538 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1541 * one level of recursion is ok and is much faster than kicking
1542 * the unplug handling
1544 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1546 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1549 kblockd_schedule_work(&q
->unplug_work
);
1553 EXPORT_SYMBOL(blk_start_queue
);
1556 * blk_stop_queue - stop a queue
1557 * @q: The &request_queue_t in question
1560 * The Linux block layer assumes that a block driver will consume all
1561 * entries on the request queue when the request_fn strategy is called.
1562 * Often this will not happen, because of hardware limitations (queue
1563 * depth settings). If a device driver gets a 'queue full' response,
1564 * or if it simply chooses not to queue more I/O at one point, it can
1565 * call this function to prevent the request_fn from being called until
1566 * the driver has signalled it's ready to go again. This happens by calling
1567 * blk_start_queue() to restart queue operations. Queue lock must be held.
1569 void blk_stop_queue(request_queue_t
*q
)
1572 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1574 EXPORT_SYMBOL(blk_stop_queue
);
1577 * blk_sync_queue - cancel any pending callbacks on a queue
1581 * The block layer may perform asynchronous callback activity
1582 * on a queue, such as calling the unplug function after a timeout.
1583 * A block device may call blk_sync_queue to ensure that any
1584 * such activity is cancelled, thus allowing it to release resources
1585 * the the callbacks might use. The caller must already have made sure
1586 * that its ->make_request_fn will not re-add plugging prior to calling
1590 void blk_sync_queue(struct request_queue
*q
)
1592 del_timer_sync(&q
->unplug_timer
);
1595 EXPORT_SYMBOL(blk_sync_queue
);
1598 * blk_run_queue - run a single device queue
1599 * @q: The queue to run
1601 void blk_run_queue(struct request_queue
*q
)
1603 unsigned long flags
;
1605 spin_lock_irqsave(q
->queue_lock
, flags
);
1607 if (!elv_queue_empty(q
))
1609 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1611 EXPORT_SYMBOL(blk_run_queue
);
1614 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1615 * @q: the request queue to be released
1618 * blk_cleanup_queue is the pair to blk_init_queue() or
1619 * blk_queue_make_request(). It should be called when a request queue is
1620 * being released; typically when a block device is being de-registered.
1621 * Currently, its primary task it to free all the &struct request
1622 * structures that were allocated to the queue and the queue itself.
1625 * Hopefully the low level driver will have finished any
1626 * outstanding requests first...
1628 void blk_cleanup_queue(request_queue_t
* q
)
1630 struct request_list
*rl
= &q
->rq
;
1632 if (!atomic_dec_and_test(&q
->refcnt
))
1636 elevator_exit(q
->elevator
);
1641 mempool_destroy(rl
->rq_pool
);
1644 __blk_queue_free_tags(q
);
1646 blk_queue_ordered(q
, QUEUE_ORDERED_NONE
);
1648 kmem_cache_free(requestq_cachep
, q
);
1651 EXPORT_SYMBOL(blk_cleanup_queue
);
1653 static int blk_init_free_list(request_queue_t
*q
)
1655 struct request_list
*rl
= &q
->rq
;
1657 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1658 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1660 init_waitqueue_head(&rl
->wait
[READ
]);
1661 init_waitqueue_head(&rl
->wait
[WRITE
]);
1663 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1664 mempool_free_slab
, request_cachep
, q
->node
);
1672 request_queue_t
*blk_alloc_queue(gfp_t gfp_mask
)
1674 return blk_alloc_queue_node(gfp_mask
, -1);
1676 EXPORT_SYMBOL(blk_alloc_queue
);
1678 request_queue_t
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1682 q
= kmem_cache_alloc_node(requestq_cachep
, gfp_mask
, node_id
);
1686 memset(q
, 0, sizeof(*q
));
1687 init_timer(&q
->unplug_timer
);
1688 atomic_set(&q
->refcnt
, 1);
1690 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1691 q
->backing_dev_info
.unplug_io_data
= q
;
1695 EXPORT_SYMBOL(blk_alloc_queue_node
);
1698 * blk_init_queue - prepare a request queue for use with a block device
1699 * @rfn: The function to be called to process requests that have been
1700 * placed on the queue.
1701 * @lock: Request queue spin lock
1704 * If a block device wishes to use the standard request handling procedures,
1705 * which sorts requests and coalesces adjacent requests, then it must
1706 * call blk_init_queue(). The function @rfn will be called when there
1707 * are requests on the queue that need to be processed. If the device
1708 * supports plugging, then @rfn may not be called immediately when requests
1709 * are available on the queue, but may be called at some time later instead.
1710 * Plugged queues are generally unplugged when a buffer belonging to one
1711 * of the requests on the queue is needed, or due to memory pressure.
1713 * @rfn is not required, or even expected, to remove all requests off the
1714 * queue, but only as many as it can handle at a time. If it does leave
1715 * requests on the queue, it is responsible for arranging that the requests
1716 * get dealt with eventually.
1718 * The queue spin lock must be held while manipulating the requests on the
1721 * Function returns a pointer to the initialized request queue, or NULL if
1722 * it didn't succeed.
1725 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1726 * when the block device is deactivated (such as at module unload).
1729 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1731 return blk_init_queue_node(rfn
, lock
, -1);
1733 EXPORT_SYMBOL(blk_init_queue
);
1736 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1738 request_queue_t
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1744 if (blk_init_free_list(q
))
1748 * if caller didn't supply a lock, they get per-queue locking with
1752 spin_lock_init(&q
->__queue_lock
);
1753 lock
= &q
->__queue_lock
;
1756 q
->request_fn
= rfn
;
1757 q
->back_merge_fn
= ll_back_merge_fn
;
1758 q
->front_merge_fn
= ll_front_merge_fn
;
1759 q
->merge_requests_fn
= ll_merge_requests_fn
;
1760 q
->prep_rq_fn
= NULL
;
1761 q
->unplug_fn
= generic_unplug_device
;
1762 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1763 q
->queue_lock
= lock
;
1765 blk_queue_segment_boundary(q
, 0xffffffff);
1767 blk_queue_make_request(q
, __make_request
);
1768 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1770 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1771 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1776 if (!elevator_init(q
, NULL
)) {
1777 blk_queue_congestion_threshold(q
);
1781 blk_cleanup_queue(q
);
1783 kmem_cache_free(requestq_cachep
, q
);
1786 EXPORT_SYMBOL(blk_init_queue_node
);
1788 int blk_get_queue(request_queue_t
*q
)
1790 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1791 atomic_inc(&q
->refcnt
);
1798 EXPORT_SYMBOL(blk_get_queue
);
1800 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
1802 if (rq
->flags
& REQ_ELVPRIV
)
1803 elv_put_request(q
, rq
);
1804 mempool_free(rq
, q
->rq
.rq_pool
);
1807 static inline struct request
*
1808 blk_alloc_request(request_queue_t
*q
, int rw
, struct bio
*bio
,
1809 int priv
, gfp_t gfp_mask
)
1811 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1817 * first three bits are identical in rq->flags and bio->bi_rw,
1818 * see bio.h and blkdev.h
1823 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
))) {
1824 mempool_free(rq
, q
->rq
.rq_pool
);
1827 rq
->flags
|= REQ_ELVPRIV
;
1834 * ioc_batching returns true if the ioc is a valid batching request and
1835 * should be given priority access to a request.
1837 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
1843 * Make sure the process is able to allocate at least 1 request
1844 * even if the batch times out, otherwise we could theoretically
1847 return ioc
->nr_batch_requests
== q
->nr_batching
||
1848 (ioc
->nr_batch_requests
> 0
1849 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1853 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1854 * will cause the process to be a "batcher" on all queues in the system. This
1855 * is the behaviour we want though - once it gets a wakeup it should be given
1858 static void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
1860 if (!ioc
|| ioc_batching(q
, ioc
))
1863 ioc
->nr_batch_requests
= q
->nr_batching
;
1864 ioc
->last_waited
= jiffies
;
1867 static void __freed_request(request_queue_t
*q
, int rw
)
1869 struct request_list
*rl
= &q
->rq
;
1871 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
1872 clear_queue_congested(q
, rw
);
1874 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
1875 if (waitqueue_active(&rl
->wait
[rw
]))
1876 wake_up(&rl
->wait
[rw
]);
1878 blk_clear_queue_full(q
, rw
);
1883 * A request has just been released. Account for it, update the full and
1884 * congestion status, wake up any waiters. Called under q->queue_lock.
1886 static void freed_request(request_queue_t
*q
, int rw
, int priv
)
1888 struct request_list
*rl
= &q
->rq
;
1894 __freed_request(q
, rw
);
1896 if (unlikely(rl
->starved
[rw
^ 1]))
1897 __freed_request(q
, rw
^ 1);
1900 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1902 * Get a free request, queue_lock must be held.
1903 * Returns NULL on failure, with queue_lock held.
1904 * Returns !NULL on success, with queue_lock *not held*.
1906 static struct request
*get_request(request_queue_t
*q
, int rw
, struct bio
*bio
,
1909 struct request
*rq
= NULL
;
1910 struct request_list
*rl
= &q
->rq
;
1911 struct io_context
*ioc
= NULL
;
1912 int may_queue
, priv
;
1914 may_queue
= elv_may_queue(q
, rw
, bio
);
1915 if (may_queue
== ELV_MQUEUE_NO
)
1918 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
1919 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
1920 ioc
= current_io_context(GFP_ATOMIC
);
1922 * The queue will fill after this allocation, so set
1923 * it as full, and mark this process as "batching".
1924 * This process will be allowed to complete a batch of
1925 * requests, others will be blocked.
1927 if (!blk_queue_full(q
, rw
)) {
1928 ioc_set_batching(q
, ioc
);
1929 blk_set_queue_full(q
, rw
);
1931 if (may_queue
!= ELV_MQUEUE_MUST
1932 && !ioc_batching(q
, ioc
)) {
1934 * The queue is full and the allocating
1935 * process is not a "batcher", and not
1936 * exempted by the IO scheduler
1942 set_queue_congested(q
, rw
);
1946 * Only allow batching queuers to allocate up to 50% over the defined
1947 * limit of requests, otherwise we could have thousands of requests
1948 * allocated with any setting of ->nr_requests
1950 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
1954 rl
->starved
[rw
] = 0;
1956 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
1960 spin_unlock_irq(q
->queue_lock
);
1962 rq
= blk_alloc_request(q
, rw
, bio
, priv
, gfp_mask
);
1963 if (unlikely(!rq
)) {
1965 * Allocation failed presumably due to memory. Undo anything
1966 * we might have messed up.
1968 * Allocating task should really be put onto the front of the
1969 * wait queue, but this is pretty rare.
1971 spin_lock_irq(q
->queue_lock
);
1972 freed_request(q
, rw
, priv
);
1975 * in the very unlikely event that allocation failed and no
1976 * requests for this direction was pending, mark us starved
1977 * so that freeing of a request in the other direction will
1978 * notice us. another possible fix would be to split the
1979 * rq mempool into READ and WRITE
1982 if (unlikely(rl
->count
[rw
] == 0))
1983 rl
->starved
[rw
] = 1;
1989 * ioc may be NULL here, and ioc_batching will be false. That's
1990 * OK, if the queue is under the request limit then requests need
1991 * not count toward the nr_batch_requests limit. There will always
1992 * be some limit enforced by BLK_BATCH_TIME.
1994 if (ioc_batching(q
, ioc
))
1995 ioc
->nr_batch_requests
--;
2004 * No available requests for this queue, unplug the device and wait for some
2005 * requests to become available.
2007 * Called with q->queue_lock held, and returns with it unlocked.
2009 static struct request
*get_request_wait(request_queue_t
*q
, int rw
,
2014 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
2017 struct request_list
*rl
= &q
->rq
;
2019 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
2020 TASK_UNINTERRUPTIBLE
);
2022 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
2025 struct io_context
*ioc
;
2027 __generic_unplug_device(q
);
2028 spin_unlock_irq(q
->queue_lock
);
2032 * After sleeping, we become a "batching" process and
2033 * will be able to allocate at least one request, and
2034 * up to a big batch of them for a small period time.
2035 * See ioc_batching, ioc_set_batching
2037 ioc
= current_io_context(GFP_NOIO
);
2038 ioc_set_batching(q
, ioc
);
2040 spin_lock_irq(q
->queue_lock
);
2042 finish_wait(&rl
->wait
[rw
], &wait
);
2048 struct request
*blk_get_request(request_queue_t
*q
, int rw
, gfp_t gfp_mask
)
2052 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2054 spin_lock_irq(q
->queue_lock
);
2055 if (gfp_mask
& __GFP_WAIT
) {
2056 rq
= get_request_wait(q
, rw
, NULL
);
2058 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2060 spin_unlock_irq(q
->queue_lock
);
2062 /* q->queue_lock is unlocked at this point */
2066 EXPORT_SYMBOL(blk_get_request
);
2069 * blk_requeue_request - put a request back on queue
2070 * @q: request queue where request should be inserted
2071 * @rq: request to be inserted
2074 * Drivers often keep queueing requests until the hardware cannot accept
2075 * more, when that condition happens we need to put the request back
2076 * on the queue. Must be called with queue lock held.
2078 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2080 if (blk_rq_tagged(rq
))
2081 blk_queue_end_tag(q
, rq
);
2083 elv_requeue_request(q
, rq
);
2086 EXPORT_SYMBOL(blk_requeue_request
);
2089 * blk_insert_request - insert a special request in to a request queue
2090 * @q: request queue where request should be inserted
2091 * @rq: request to be inserted
2092 * @at_head: insert request at head or tail of queue
2093 * @data: private data
2096 * Many block devices need to execute commands asynchronously, so they don't
2097 * block the whole kernel from preemption during request execution. This is
2098 * accomplished normally by inserting aritficial requests tagged as
2099 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2100 * scheduled for actual execution by the request queue.
2102 * We have the option of inserting the head or the tail of the queue.
2103 * Typically we use the tail for new ioctls and so forth. We use the head
2104 * of the queue for things like a QUEUE_FULL message from a device, or a
2105 * host that is unable to accept a particular command.
2107 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2108 int at_head
, void *data
)
2110 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2111 unsigned long flags
;
2114 * tell I/O scheduler that this isn't a regular read/write (ie it
2115 * must not attempt merges on this) and that it acts as a soft
2118 rq
->flags
|= REQ_SPECIAL
| REQ_SOFTBARRIER
;
2122 spin_lock_irqsave(q
->queue_lock
, flags
);
2125 * If command is tagged, release the tag
2127 if (blk_rq_tagged(rq
))
2128 blk_queue_end_tag(q
, rq
);
2130 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2131 __elv_add_request(q
, rq
, where
, 0);
2133 if (blk_queue_plugged(q
))
2134 __generic_unplug_device(q
);
2137 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2140 EXPORT_SYMBOL(blk_insert_request
);
2143 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2144 * @q: request queue where request should be inserted
2145 * @rq: request structure to fill
2146 * @ubuf: the user buffer
2147 * @len: length of user data
2150 * Data will be mapped directly for zero copy io, if possible. Otherwise
2151 * a kernel bounce buffer is used.
2153 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2154 * still in process context.
2156 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2157 * before being submitted to the device, as pages mapped may be out of
2158 * reach. It's the callers responsibility to make sure this happens. The
2159 * original bio must be passed back in to blk_rq_unmap_user() for proper
2162 int blk_rq_map_user(request_queue_t
*q
, struct request
*rq
, void __user
*ubuf
,
2165 unsigned long uaddr
;
2169 if (len
> (q
->max_hw_sectors
<< 9))
2174 reading
= rq_data_dir(rq
) == READ
;
2177 * if alignment requirement is satisfied, map in user pages for
2178 * direct dma. else, set up kernel bounce buffers
2180 uaddr
= (unsigned long) ubuf
;
2181 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2182 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2184 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2187 rq
->bio
= rq
->biotail
= bio
;
2188 blk_rq_bio_prep(q
, rq
, bio
);
2190 rq
->buffer
= rq
->data
= NULL
;
2196 * bio is the err-ptr
2198 return PTR_ERR(bio
);
2201 EXPORT_SYMBOL(blk_rq_map_user
);
2204 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2205 * @q: request queue where request should be inserted
2206 * @rq: request to map data to
2207 * @iov: pointer to the iovec
2208 * @iov_count: number of elements in the iovec
2211 * Data will be mapped directly for zero copy io, if possible. Otherwise
2212 * a kernel bounce buffer is used.
2214 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2215 * still in process context.
2217 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2218 * before being submitted to the device, as pages mapped may be out of
2219 * reach. It's the callers responsibility to make sure this happens. The
2220 * original bio must be passed back in to blk_rq_unmap_user() for proper
2223 int blk_rq_map_user_iov(request_queue_t
*q
, struct request
*rq
,
2224 struct sg_iovec
*iov
, int iov_count
)
2228 if (!iov
|| iov_count
<= 0)
2231 /* we don't allow misaligned data like bio_map_user() does. If the
2232 * user is using sg, they're expected to know the alignment constraints
2233 * and respect them accordingly */
2234 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2236 return PTR_ERR(bio
);
2238 rq
->bio
= rq
->biotail
= bio
;
2239 blk_rq_bio_prep(q
, rq
, bio
);
2240 rq
->buffer
= rq
->data
= NULL
;
2241 rq
->data_len
= bio
->bi_size
;
2245 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2248 * blk_rq_unmap_user - unmap a request with user data
2249 * @bio: bio to be unmapped
2250 * @ulen: length of user buffer
2253 * Unmap a bio previously mapped by blk_rq_map_user().
2255 int blk_rq_unmap_user(struct bio
*bio
, unsigned int ulen
)
2260 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2261 bio_unmap_user(bio
);
2263 ret
= bio_uncopy_user(bio
);
2269 EXPORT_SYMBOL(blk_rq_unmap_user
);
2272 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2273 * @q: request queue where request should be inserted
2274 * @rq: request to fill
2275 * @kbuf: the kernel buffer
2276 * @len: length of user data
2277 * @gfp_mask: memory allocation flags
2279 int blk_rq_map_kern(request_queue_t
*q
, struct request
*rq
, void *kbuf
,
2280 unsigned int len
, gfp_t gfp_mask
)
2284 if (len
> (q
->max_hw_sectors
<< 9))
2289 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2291 return PTR_ERR(bio
);
2293 if (rq_data_dir(rq
) == WRITE
)
2294 bio
->bi_rw
|= (1 << BIO_RW
);
2296 rq
->bio
= rq
->biotail
= bio
;
2297 blk_rq_bio_prep(q
, rq
, bio
);
2299 rq
->buffer
= rq
->data
= NULL
;
2304 EXPORT_SYMBOL(blk_rq_map_kern
);
2307 * blk_execute_rq_nowait - insert a request into queue for execution
2308 * @q: queue to insert the request in
2309 * @bd_disk: matching gendisk
2310 * @rq: request to insert
2311 * @at_head: insert request at head or tail of queue
2312 * @done: I/O completion handler
2315 * Insert a fully prepared request at the back of the io scheduler queue
2316 * for execution. Don't wait for completion.
2318 void blk_execute_rq_nowait(request_queue_t
*q
, struct gendisk
*bd_disk
,
2319 struct request
*rq
, int at_head
,
2322 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2324 rq
->rq_disk
= bd_disk
;
2325 rq
->flags
|= REQ_NOMERGE
;
2327 elv_add_request(q
, rq
, where
, 1);
2328 generic_unplug_device(q
);
2331 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait
);
2334 * blk_execute_rq - insert a request into queue for execution
2335 * @q: queue to insert the request in
2336 * @bd_disk: matching gendisk
2337 * @rq: request to insert
2338 * @at_head: insert request at head or tail of queue
2341 * Insert a fully prepared request at the back of the io scheduler queue
2342 * for execution and wait for completion.
2344 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2345 struct request
*rq
, int at_head
)
2347 DECLARE_COMPLETION(wait
);
2348 char sense
[SCSI_SENSE_BUFFERSIZE
];
2352 * we need an extra reference to the request, so we can look at
2353 * it after io completion
2358 memset(sense
, 0, sizeof(sense
));
2363 rq
->waiting
= &wait
;
2364 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2365 wait_for_completion(&wait
);
2374 EXPORT_SYMBOL(blk_execute_rq
);
2377 * blkdev_issue_flush - queue a flush
2378 * @bdev: blockdev to issue flush for
2379 * @error_sector: error sector
2382 * Issue a flush for the block device in question. Caller can supply
2383 * room for storing the error offset in case of a flush error, if they
2384 * wish to. Caller must run wait_for_completion() on its own.
2386 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2390 if (bdev
->bd_disk
== NULL
)
2393 q
= bdev_get_queue(bdev
);
2396 if (!q
->issue_flush_fn
)
2399 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2402 EXPORT_SYMBOL(blkdev_issue_flush
);
2404 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2406 int rw
= rq_data_dir(rq
);
2408 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2412 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2414 disk_round_stats(rq
->rq_disk
);
2415 rq
->rq_disk
->in_flight
++;
2420 * add-request adds a request to the linked list.
2421 * queue lock is held and interrupts disabled, as we muck with the
2422 * request queue list.
2424 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2426 drive_stat_acct(req
, req
->nr_sectors
, 1);
2429 q
->activity_fn(q
->activity_data
, rq_data_dir(req
));
2432 * elevator indicated where it wants this request to be
2433 * inserted at elevator_merge time
2435 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2439 * disk_round_stats() - Round off the performance stats on a struct
2442 * The average IO queue length and utilisation statistics are maintained
2443 * by observing the current state of the queue length and the amount of
2444 * time it has been in this state for.
2446 * Normally, that accounting is done on IO completion, but that can result
2447 * in more than a second's worth of IO being accounted for within any one
2448 * second, leading to >100% utilisation. To deal with that, we call this
2449 * function to do a round-off before returning the results when reading
2450 * /proc/diskstats. This accounts immediately for all queue usage up to
2451 * the current jiffies and restarts the counters again.
2453 void disk_round_stats(struct gendisk
*disk
)
2455 unsigned long now
= jiffies
;
2457 if (now
== disk
->stamp
)
2460 if (disk
->in_flight
) {
2461 __disk_stat_add(disk
, time_in_queue
,
2462 disk
->in_flight
* (now
- disk
->stamp
));
2463 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2469 * queue lock must be held
2471 void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2473 struct request_list
*rl
= req
->rl
;
2477 if (unlikely(--req
->ref_count
))
2480 elv_completed_request(q
, req
);
2482 req
->rq_status
= RQ_INACTIVE
;
2486 * Request may not have originated from ll_rw_blk. if not,
2487 * it didn't come out of our reserved rq pools
2490 int rw
= rq_data_dir(req
);
2491 int priv
= req
->flags
& REQ_ELVPRIV
;
2493 BUG_ON(!list_empty(&req
->queuelist
));
2495 blk_free_request(q
, req
);
2496 freed_request(q
, rw
, priv
);
2500 EXPORT_SYMBOL_GPL(__blk_put_request
);
2502 void blk_put_request(struct request
*req
)
2504 unsigned long flags
;
2505 request_queue_t
*q
= req
->q
;
2508 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2509 * following if (q) test.
2512 spin_lock_irqsave(q
->queue_lock
, flags
);
2513 __blk_put_request(q
, req
);
2514 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2518 EXPORT_SYMBOL(blk_put_request
);
2521 * blk_end_sync_rq - executes a completion event on a request
2522 * @rq: request to complete
2524 void blk_end_sync_rq(struct request
*rq
, int error
)
2526 struct completion
*waiting
= rq
->waiting
;
2529 __blk_put_request(rq
->q
, rq
);
2532 * complete last, if this is a stack request the process (and thus
2533 * the rq pointer) could be invalid right after this complete()
2537 EXPORT_SYMBOL(blk_end_sync_rq
);
2540 * blk_congestion_wait - wait for a queue to become uncongested
2541 * @rw: READ or WRITE
2542 * @timeout: timeout in jiffies
2544 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2545 * If no queues are congested then just wait for the next request to be
2548 long blk_congestion_wait(int rw
, long timeout
)
2552 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
2554 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
2555 ret
= io_schedule_timeout(timeout
);
2556 finish_wait(wqh
, &wait
);
2560 EXPORT_SYMBOL(blk_congestion_wait
);
2563 * Has to be called with the request spinlock acquired
2565 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2566 struct request
*next
)
2568 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2574 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2577 if (rq_data_dir(req
) != rq_data_dir(next
)
2578 || req
->rq_disk
!= next
->rq_disk
2579 || next
->waiting
|| next
->special
)
2583 * If we are allowed to merge, then append bio list
2584 * from next to rq and release next. merge_requests_fn
2585 * will have updated segment counts, update sector
2588 if (!q
->merge_requests_fn(q
, req
, next
))
2592 * At this point we have either done a back merge
2593 * or front merge. We need the smaller start_time of
2594 * the merged requests to be the current request
2595 * for accounting purposes.
2597 if (time_after(req
->start_time
, next
->start_time
))
2598 req
->start_time
= next
->start_time
;
2600 req
->biotail
->bi_next
= next
->bio
;
2601 req
->biotail
= next
->biotail
;
2603 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2605 elv_merge_requests(q
, req
, next
);
2608 disk_round_stats(req
->rq_disk
);
2609 req
->rq_disk
->in_flight
--;
2612 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2614 __blk_put_request(q
, next
);
2618 static inline int attempt_back_merge(request_queue_t
*q
, struct request
*rq
)
2620 struct request
*next
= elv_latter_request(q
, rq
);
2623 return attempt_merge(q
, rq
, next
);
2628 static inline int attempt_front_merge(request_queue_t
*q
, struct request
*rq
)
2630 struct request
*prev
= elv_former_request(q
, rq
);
2633 return attempt_merge(q
, prev
, rq
);
2639 * blk_attempt_remerge - attempt to remerge active head with next request
2640 * @q: The &request_queue_t belonging to the device
2641 * @rq: The head request (usually)
2644 * For head-active devices, the queue can easily be unplugged so quickly
2645 * that proper merging is not done on the front request. This may hurt
2646 * performance greatly for some devices. The block layer cannot safely
2647 * do merging on that first request for these queues, but the driver can
2648 * call this function and make it happen any way. Only the driver knows
2649 * when it is safe to do so.
2651 void blk_attempt_remerge(request_queue_t
*q
, struct request
*rq
)
2653 unsigned long flags
;
2655 spin_lock_irqsave(q
->queue_lock
, flags
);
2656 attempt_back_merge(q
, rq
);
2657 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2660 EXPORT_SYMBOL(blk_attempt_remerge
);
2662 static void init_request_from_bio(struct request
*req
, struct bio
*bio
)
2664 req
->flags
|= REQ_CMD
;
2667 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2669 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2670 req
->flags
|= REQ_FAILFAST
;
2673 * REQ_BARRIER implies no merging, but lets make it explicit
2675 if (unlikely(bio_barrier(bio
)))
2676 req
->flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2679 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
2680 req
->hard_nr_sectors
= req
->nr_sectors
= bio_sectors(bio
);
2681 req
->current_nr_sectors
= req
->hard_cur_sectors
= bio_cur_sectors(bio
);
2682 req
->nr_phys_segments
= bio_phys_segments(req
->q
, bio
);
2683 req
->nr_hw_segments
= bio_hw_segments(req
->q
, bio
);
2684 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2685 req
->waiting
= NULL
;
2686 req
->bio
= req
->biotail
= bio
;
2687 req
->ioprio
= bio_prio(bio
);
2688 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2689 req
->start_time
= jiffies
;
2692 static int __make_request(request_queue_t
*q
, struct bio
*bio
)
2694 struct request
*req
;
2695 int el_ret
, rw
, nr_sectors
, cur_nr_sectors
, barrier
, err
, sync
;
2696 unsigned short prio
;
2699 sector
= bio
->bi_sector
;
2700 nr_sectors
= bio_sectors(bio
);
2701 cur_nr_sectors
= bio_cur_sectors(bio
);
2702 prio
= bio_prio(bio
);
2704 rw
= bio_data_dir(bio
);
2705 sync
= bio_sync(bio
);
2708 * low level driver can indicate that it wants pages above a
2709 * certain limit bounced to low memory (ie for highmem, or even
2710 * ISA dma in theory)
2712 blk_queue_bounce(q
, &bio
);
2714 spin_lock_prefetch(q
->queue_lock
);
2716 barrier
= bio_barrier(bio
);
2717 if (unlikely(barrier
) && (q
->ordered
== QUEUE_ORDERED_NONE
)) {
2722 spin_lock_irq(q
->queue_lock
);
2724 if (unlikely(barrier
) || elv_queue_empty(q
))
2727 el_ret
= elv_merge(q
, &req
, bio
);
2729 case ELEVATOR_BACK_MERGE
:
2730 BUG_ON(!rq_mergeable(req
));
2732 if (!q
->back_merge_fn(q
, req
, bio
))
2735 req
->biotail
->bi_next
= bio
;
2737 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2738 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2739 drive_stat_acct(req
, nr_sectors
, 0);
2740 if (!attempt_back_merge(q
, req
))
2741 elv_merged_request(q
, req
);
2744 case ELEVATOR_FRONT_MERGE
:
2745 BUG_ON(!rq_mergeable(req
));
2747 if (!q
->front_merge_fn(q
, req
, bio
))
2750 bio
->bi_next
= req
->bio
;
2754 * may not be valid. if the low level driver said
2755 * it didn't need a bounce buffer then it better
2756 * not touch req->buffer either...
2758 req
->buffer
= bio_data(bio
);
2759 req
->current_nr_sectors
= cur_nr_sectors
;
2760 req
->hard_cur_sectors
= cur_nr_sectors
;
2761 req
->sector
= req
->hard_sector
= sector
;
2762 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2763 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2764 drive_stat_acct(req
, nr_sectors
, 0);
2765 if (!attempt_front_merge(q
, req
))
2766 elv_merged_request(q
, req
);
2769 /* ELV_NO_MERGE: elevator says don't/can't merge. */
2776 * Grab a free request. This is might sleep but can not fail.
2777 * Returns with the queue unlocked.
2779 req
= get_request_wait(q
, rw
, bio
);
2782 * After dropping the lock and possibly sleeping here, our request
2783 * may now be mergeable after it had proven unmergeable (above).
2784 * We don't worry about that case for efficiency. It won't happen
2785 * often, and the elevators are able to handle it.
2787 init_request_from_bio(req
, bio
);
2789 spin_lock_irq(q
->queue_lock
);
2790 if (elv_queue_empty(q
))
2792 add_request(q
, req
);
2795 __generic_unplug_device(q
);
2797 spin_unlock_irq(q
->queue_lock
);
2801 bio_endio(bio
, nr_sectors
<< 9, err
);
2806 * If bio->bi_dev is a partition, remap the location
2808 static inline void blk_partition_remap(struct bio
*bio
)
2810 struct block_device
*bdev
= bio
->bi_bdev
;
2812 if (bdev
!= bdev
->bd_contains
) {
2813 struct hd_struct
*p
= bdev
->bd_part
;
2814 const int rw
= bio_data_dir(bio
);
2816 p
->sectors
[rw
] += bio_sectors(bio
);
2819 bio
->bi_sector
+= p
->start_sect
;
2820 bio
->bi_bdev
= bdev
->bd_contains
;
2824 static void handle_bad_sector(struct bio
*bio
)
2826 char b
[BDEVNAME_SIZE
];
2828 printk(KERN_INFO
"attempt to access beyond end of device\n");
2829 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
2830 bdevname(bio
->bi_bdev
, b
),
2832 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
2833 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
2835 set_bit(BIO_EOF
, &bio
->bi_flags
);
2839 * generic_make_request: hand a buffer to its device driver for I/O
2840 * @bio: The bio describing the location in memory and on the device.
2842 * generic_make_request() is used to make I/O requests of block
2843 * devices. It is passed a &struct bio, which describes the I/O that needs
2846 * generic_make_request() does not return any status. The
2847 * success/failure status of the request, along with notification of
2848 * completion, is delivered asynchronously through the bio->bi_end_io
2849 * function described (one day) else where.
2851 * The caller of generic_make_request must make sure that bi_io_vec
2852 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2853 * set to describe the device address, and the
2854 * bi_end_io and optionally bi_private are set to describe how
2855 * completion notification should be signaled.
2857 * generic_make_request and the drivers it calls may use bi_next if this
2858 * bio happens to be merged with someone else, and may change bi_dev and
2859 * bi_sector for remaps as it sees fit. So the values of these fields
2860 * should NOT be depended on after the call to generic_make_request.
2862 void generic_make_request(struct bio
*bio
)
2866 int ret
, nr_sectors
= bio_sectors(bio
);
2869 /* Test device or partition size, when known. */
2870 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
2872 sector_t sector
= bio
->bi_sector
;
2874 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2876 * This may well happen - the kernel calls bread()
2877 * without checking the size of the device, e.g., when
2878 * mounting a device.
2880 handle_bad_sector(bio
);
2886 * Resolve the mapping until finished. (drivers are
2887 * still free to implement/resolve their own stacking
2888 * by explicitly returning 0)
2890 * NOTE: we don't repeat the blk_size check for each new device.
2891 * Stacking drivers are expected to know what they are doing.
2894 char b
[BDEVNAME_SIZE
];
2896 q
= bdev_get_queue(bio
->bi_bdev
);
2899 "generic_make_request: Trying to access "
2900 "nonexistent block-device %s (%Lu)\n",
2901 bdevname(bio
->bi_bdev
, b
),
2902 (long long) bio
->bi_sector
);
2904 bio_endio(bio
, bio
->bi_size
, -EIO
);
2908 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
2909 printk("bio too big device %s (%u > %u)\n",
2910 bdevname(bio
->bi_bdev
, b
),
2916 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
2920 * If this device has partitions, remap block n
2921 * of partition p to block n+start(p) of the disk.
2923 blk_partition_remap(bio
);
2925 ret
= q
->make_request_fn(q
, bio
);
2929 EXPORT_SYMBOL(generic_make_request
);
2932 * submit_bio: submit a bio to the block device layer for I/O
2933 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2934 * @bio: The &struct bio which describes the I/O
2936 * submit_bio() is very similar in purpose to generic_make_request(), and
2937 * uses that function to do most of the work. Both are fairly rough
2938 * interfaces, @bio must be presetup and ready for I/O.
2941 void submit_bio(int rw
, struct bio
*bio
)
2943 int count
= bio_sectors(bio
);
2945 BIO_BUG_ON(!bio
->bi_size
);
2946 BIO_BUG_ON(!bio
->bi_io_vec
);
2949 mod_page_state(pgpgout
, count
);
2951 mod_page_state(pgpgin
, count
);
2953 if (unlikely(block_dump
)) {
2954 char b
[BDEVNAME_SIZE
];
2955 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
2956 current
->comm
, current
->pid
,
2957 (rw
& WRITE
) ? "WRITE" : "READ",
2958 (unsigned long long)bio
->bi_sector
,
2959 bdevname(bio
->bi_bdev
,b
));
2962 generic_make_request(bio
);
2965 EXPORT_SYMBOL(submit_bio
);
2967 static void blk_recalc_rq_segments(struct request
*rq
)
2969 struct bio
*bio
, *prevbio
= NULL
;
2970 int nr_phys_segs
, nr_hw_segs
;
2971 unsigned int phys_size
, hw_size
;
2972 request_queue_t
*q
= rq
->q
;
2977 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
2978 rq_for_each_bio(bio
, rq
) {
2979 /* Force bio hw/phys segs to be recalculated. */
2980 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
2982 nr_phys_segs
+= bio_phys_segments(q
, bio
);
2983 nr_hw_segs
+= bio_hw_segments(q
, bio
);
2985 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
2986 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
2988 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
2989 pseg
<= q
->max_segment_size
) {
2991 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
2995 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
2996 hseg
<= q
->max_segment_size
) {
2998 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3005 rq
->nr_phys_segments
= nr_phys_segs
;
3006 rq
->nr_hw_segments
= nr_hw_segs
;
3009 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3011 if (blk_fs_request(rq
)) {
3012 rq
->hard_sector
+= nsect
;
3013 rq
->hard_nr_sectors
-= nsect
;
3016 * Move the I/O submission pointers ahead if required.
3018 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3019 (rq
->sector
<= rq
->hard_sector
)) {
3020 rq
->sector
= rq
->hard_sector
;
3021 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3022 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3023 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3024 rq
->buffer
= bio_data(rq
->bio
);
3028 * if total number of sectors is less than the first segment
3029 * size, something has gone terribly wrong
3031 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3032 printk("blk: request botched\n");
3033 rq
->nr_sectors
= rq
->current_nr_sectors
;
3038 static int __end_that_request_first(struct request
*req
, int uptodate
,
3041 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3045 * extend uptodate bool to allow < 0 value to be direct io error
3048 if (end_io_error(uptodate
))
3049 error
= !uptodate
? -EIO
: uptodate
;
3052 * for a REQ_BLOCK_PC request, we want to carry any eventual
3053 * sense key with us all the way through
3055 if (!blk_pc_request(req
))
3059 if (blk_fs_request(req
) && !(req
->flags
& REQ_QUIET
))
3060 printk("end_request: I/O error, dev %s, sector %llu\n",
3061 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3062 (unsigned long long)req
->sector
);
3065 if (blk_fs_request(req
) && req
->rq_disk
) {
3066 const int rw
= rq_data_dir(req
);
3068 __disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3071 total_bytes
= bio_nbytes
= 0;
3072 while ((bio
= req
->bio
) != NULL
) {
3075 if (nr_bytes
>= bio
->bi_size
) {
3076 req
->bio
= bio
->bi_next
;
3077 nbytes
= bio
->bi_size
;
3078 bio_endio(bio
, nbytes
, error
);
3082 int idx
= bio
->bi_idx
+ next_idx
;
3084 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3085 blk_dump_rq_flags(req
, "__end_that");
3086 printk("%s: bio idx %d >= vcnt %d\n",
3088 bio
->bi_idx
, bio
->bi_vcnt
);
3092 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3093 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3096 * not a complete bvec done
3098 if (unlikely(nbytes
> nr_bytes
)) {
3099 bio_nbytes
+= nr_bytes
;
3100 total_bytes
+= nr_bytes
;
3105 * advance to the next vector
3108 bio_nbytes
+= nbytes
;
3111 total_bytes
+= nbytes
;
3114 if ((bio
= req
->bio
)) {
3116 * end more in this run, or just return 'not-done'
3118 if (unlikely(nr_bytes
<= 0))
3130 * if the request wasn't completed, update state
3133 bio_endio(bio
, bio_nbytes
, error
);
3134 bio
->bi_idx
+= next_idx
;
3135 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3136 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3139 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3140 blk_recalc_rq_segments(req
);
3145 * end_that_request_first - end I/O on a request
3146 * @req: the request being processed
3147 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3148 * @nr_sectors: number of sectors to end I/O on
3151 * Ends I/O on a number of sectors attached to @req, and sets it up
3152 * for the next range of segments (if any) in the cluster.
3155 * 0 - we are done with this request, call end_that_request_last()
3156 * 1 - still buffers pending for this request
3158 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3160 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3163 EXPORT_SYMBOL(end_that_request_first
);
3166 * end_that_request_chunk - end I/O on a request
3167 * @req: the request being processed
3168 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3169 * @nr_bytes: number of bytes to complete
3172 * Ends I/O on a number of bytes attached to @req, and sets it up
3173 * for the next range of segments (if any). Like end_that_request_first(),
3174 * but deals with bytes instead of sectors.
3177 * 0 - we are done with this request, call end_that_request_last()
3178 * 1 - still buffers pending for this request
3180 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3182 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3185 EXPORT_SYMBOL(end_that_request_chunk
);
3188 * queue lock must be held
3190 void end_that_request_last(struct request
*req
, int uptodate
)
3192 struct gendisk
*disk
= req
->rq_disk
;
3196 * extend uptodate bool to allow < 0 value to be direct io error
3199 if (end_io_error(uptodate
))
3200 error
= !uptodate
? -EIO
: uptodate
;
3202 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3203 laptop_io_completion();
3205 if (disk
&& blk_fs_request(req
)) {
3206 unsigned long duration
= jiffies
- req
->start_time
;
3207 const int rw
= rq_data_dir(req
);
3209 __disk_stat_inc(disk
, ios
[rw
]);
3210 __disk_stat_add(disk
, ticks
[rw
], duration
);
3211 disk_round_stats(disk
);
3215 req
->end_io(req
, error
);
3217 __blk_put_request(req
->q
, req
);
3220 EXPORT_SYMBOL(end_that_request_last
);
3222 void end_request(struct request
*req
, int uptodate
)
3224 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3225 add_disk_randomness(req
->rq_disk
);
3226 blkdev_dequeue_request(req
);
3227 end_that_request_last(req
, uptodate
);
3231 EXPORT_SYMBOL(end_request
);
3233 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3235 /* first three bits are identical in rq->flags and bio->bi_rw */
3236 rq
->flags
|= (bio
->bi_rw
& 7);
3238 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3239 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3240 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3241 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3242 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3243 rq
->buffer
= bio_data(bio
);
3245 rq
->bio
= rq
->biotail
= bio
;
3248 EXPORT_SYMBOL(blk_rq_bio_prep
);
3250 int kblockd_schedule_work(struct work_struct
*work
)
3252 return queue_work(kblockd_workqueue
, work
);
3255 EXPORT_SYMBOL(kblockd_schedule_work
);
3257 void kblockd_flush(void)
3259 flush_workqueue(kblockd_workqueue
);
3261 EXPORT_SYMBOL(kblockd_flush
);
3263 int __init
blk_dev_init(void)
3265 kblockd_workqueue
= create_workqueue("kblockd");
3266 if (!kblockd_workqueue
)
3267 panic("Failed to create kblockd\n");
3269 request_cachep
= kmem_cache_create("blkdev_requests",
3270 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3272 requestq_cachep
= kmem_cache_create("blkdev_queue",
3273 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3275 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3276 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3278 blk_max_low_pfn
= max_low_pfn
;
3279 blk_max_pfn
= max_pfn
;
3285 * IO Context helper functions
3287 void put_io_context(struct io_context
*ioc
)
3292 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3294 if (atomic_dec_and_test(&ioc
->refcount
)) {
3295 if (ioc
->aic
&& ioc
->aic
->dtor
)
3296 ioc
->aic
->dtor(ioc
->aic
);
3297 if (ioc
->cic
&& ioc
->cic
->dtor
)
3298 ioc
->cic
->dtor(ioc
->cic
);
3300 kmem_cache_free(iocontext_cachep
, ioc
);
3303 EXPORT_SYMBOL(put_io_context
);
3305 /* Called by the exitting task */
3306 void exit_io_context(void)
3308 unsigned long flags
;
3309 struct io_context
*ioc
;
3311 local_irq_save(flags
);
3313 ioc
= current
->io_context
;
3314 current
->io_context
= NULL
;
3316 task_unlock(current
);
3317 local_irq_restore(flags
);
3319 if (ioc
->aic
&& ioc
->aic
->exit
)
3320 ioc
->aic
->exit(ioc
->aic
);
3321 if (ioc
->cic
&& ioc
->cic
->exit
)
3322 ioc
->cic
->exit(ioc
->cic
);
3324 put_io_context(ioc
);
3328 * If the current task has no IO context then create one and initialise it.
3329 * Otherwise, return its existing IO context.
3331 * This returned IO context doesn't have a specifically elevated refcount,
3332 * but since the current task itself holds a reference, the context can be
3333 * used in general code, so long as it stays within `current` context.
3335 struct io_context
*current_io_context(gfp_t gfp_flags
)
3337 struct task_struct
*tsk
= current
;
3338 struct io_context
*ret
;
3340 ret
= tsk
->io_context
;
3344 ret
= kmem_cache_alloc(iocontext_cachep
, gfp_flags
);
3346 atomic_set(&ret
->refcount
, 1);
3347 ret
->task
= current
;
3348 ret
->set_ioprio
= NULL
;
3349 ret
->last_waited
= jiffies
; /* doesn't matter... */
3350 ret
->nr_batch_requests
= 0; /* because this is 0 */
3353 tsk
->io_context
= ret
;
3358 EXPORT_SYMBOL(current_io_context
);
3361 * If the current task has no IO context then create one and initialise it.
3362 * If it does have a context, take a ref on it.
3364 * This is always called in the context of the task which submitted the I/O.
3366 struct io_context
*get_io_context(gfp_t gfp_flags
)
3368 struct io_context
*ret
;
3369 ret
= current_io_context(gfp_flags
);
3371 atomic_inc(&ret
->refcount
);
3374 EXPORT_SYMBOL(get_io_context
);
3376 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3378 struct io_context
*src
= *psrc
;
3379 struct io_context
*dst
= *pdst
;
3382 BUG_ON(atomic_read(&src
->refcount
) == 0);
3383 atomic_inc(&src
->refcount
);
3384 put_io_context(dst
);
3388 EXPORT_SYMBOL(copy_io_context
);
3390 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3392 struct io_context
*temp
;
3397 EXPORT_SYMBOL(swap_io_context
);
3402 struct queue_sysfs_entry
{
3403 struct attribute attr
;
3404 ssize_t (*show
)(struct request_queue
*, char *);
3405 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3409 queue_var_show(unsigned int var
, char *page
)
3411 return sprintf(page
, "%d\n", var
);
3415 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3417 char *p
= (char *) page
;
3419 *var
= simple_strtoul(p
, &p
, 10);
3423 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3425 return queue_var_show(q
->nr_requests
, (page
));
3429 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3431 struct request_list
*rl
= &q
->rq
;
3433 int ret
= queue_var_store(&q
->nr_requests
, page
, count
);
3434 if (q
->nr_requests
< BLKDEV_MIN_RQ
)
3435 q
->nr_requests
= BLKDEV_MIN_RQ
;
3436 blk_queue_congestion_threshold(q
);
3438 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3439 set_queue_congested(q
, READ
);
3440 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3441 clear_queue_congested(q
, READ
);
3443 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3444 set_queue_congested(q
, WRITE
);
3445 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3446 clear_queue_congested(q
, WRITE
);
3448 if (rl
->count
[READ
] >= q
->nr_requests
) {
3449 blk_set_queue_full(q
, READ
);
3450 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3451 blk_clear_queue_full(q
, READ
);
3452 wake_up(&rl
->wait
[READ
]);
3455 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3456 blk_set_queue_full(q
, WRITE
);
3457 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3458 blk_clear_queue_full(q
, WRITE
);
3459 wake_up(&rl
->wait
[WRITE
]);
3464 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3466 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3468 return queue_var_show(ra_kb
, (page
));
3472 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3474 unsigned long ra_kb
;
3475 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3477 spin_lock_irq(q
->queue_lock
);
3478 if (ra_kb
> (q
->max_sectors
>> 1))
3479 ra_kb
= (q
->max_sectors
>> 1);
3481 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3482 spin_unlock_irq(q
->queue_lock
);
3487 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3489 int max_sectors_kb
= q
->max_sectors
>> 1;
3491 return queue_var_show(max_sectors_kb
, (page
));
3495 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3497 unsigned long max_sectors_kb
,
3498 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3499 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3500 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3503 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3506 * Take the queue lock to update the readahead and max_sectors
3507 * values synchronously:
3509 spin_lock_irq(q
->queue_lock
);
3511 * Trim readahead window as well, if necessary:
3513 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3514 if (ra_kb
> max_sectors_kb
)
3515 q
->backing_dev_info
.ra_pages
=
3516 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3518 q
->max_sectors
= max_sectors_kb
<< 1;
3519 spin_unlock_irq(q
->queue_lock
);
3524 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3526 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3528 return queue_var_show(max_hw_sectors_kb
, (page
));
3532 static struct queue_sysfs_entry queue_requests_entry
= {
3533 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
3534 .show
= queue_requests_show
,
3535 .store
= queue_requests_store
,
3538 static struct queue_sysfs_entry queue_ra_entry
= {
3539 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
3540 .show
= queue_ra_show
,
3541 .store
= queue_ra_store
,
3544 static struct queue_sysfs_entry queue_max_sectors_entry
= {
3545 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
3546 .show
= queue_max_sectors_show
,
3547 .store
= queue_max_sectors_store
,
3550 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
3551 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
3552 .show
= queue_max_hw_sectors_show
,
3555 static struct queue_sysfs_entry queue_iosched_entry
= {
3556 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
3557 .show
= elv_iosched_show
,
3558 .store
= elv_iosched_store
,
3561 static struct attribute
*default_attrs
[] = {
3562 &queue_requests_entry
.attr
,
3563 &queue_ra_entry
.attr
,
3564 &queue_max_hw_sectors_entry
.attr
,
3565 &queue_max_sectors_entry
.attr
,
3566 &queue_iosched_entry
.attr
,
3570 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3573 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3575 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3576 struct request_queue
*q
;
3578 q
= container_of(kobj
, struct request_queue
, kobj
);
3582 return entry
->show(q
, page
);
3586 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3587 const char *page
, size_t length
)
3589 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3590 struct request_queue
*q
;
3592 q
= container_of(kobj
, struct request_queue
, kobj
);
3596 return entry
->store(q
, page
, length
);
3599 static struct sysfs_ops queue_sysfs_ops
= {
3600 .show
= queue_attr_show
,
3601 .store
= queue_attr_store
,
3604 static struct kobj_type queue_ktype
= {
3605 .sysfs_ops
= &queue_sysfs_ops
,
3606 .default_attrs
= default_attrs
,
3609 int blk_register_queue(struct gendisk
*disk
)
3613 request_queue_t
*q
= disk
->queue
;
3615 if (!q
|| !q
->request_fn
)
3618 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
3619 if (!q
->kobj
.parent
)
3622 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
3623 q
->kobj
.ktype
= &queue_ktype
;
3625 ret
= kobject_register(&q
->kobj
);
3629 ret
= elv_register_queue(q
);
3631 kobject_unregister(&q
->kobj
);
3638 void blk_unregister_queue(struct gendisk
*disk
)
3640 request_queue_t
*q
= disk
->queue
;
3642 if (q
&& q
->request_fn
) {
3643 elv_unregister_queue(q
);
3645 kobject_unregister(&q
->kobj
);
3646 kobject_put(&disk
->kobj
);