2 * linux/drivers/block/ll_rw_blk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
6 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
7 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
8 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
13 * This handles all read/write requests to block devices
15 #include <linux/config.h>
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/backing-dev.h>
19 #include <linux/bio.h>
20 #include <linux/blkdev.h>
21 #include <linux/highmem.h>
23 #include <linux/kernel_stat.h>
24 #include <linux/string.h>
25 #include <linux/init.h>
26 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/blkdev.h>
36 #include <scsi/scsi_cmnd.h>
38 static void blk_unplug_work(void *data
);
39 static void blk_unplug_timeout(unsigned long data
);
40 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
43 * For the allocated request tables
45 static kmem_cache_t
*request_cachep
;
48 * For queue allocation
50 static kmem_cache_t
*requestq_cachep
;
53 * For io context allocations
55 static kmem_cache_t
*iocontext_cachep
;
57 static wait_queue_head_t congestion_wqh
[2] = {
58 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
59 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
63 * Controlling structure to kblockd
65 static struct workqueue_struct
*kblockd_workqueue
;
67 unsigned long blk_max_low_pfn
, blk_max_pfn
;
69 EXPORT_SYMBOL(blk_max_low_pfn
);
70 EXPORT_SYMBOL(blk_max_pfn
);
72 /* Amount of time in which a process may batch requests */
73 #define BLK_BATCH_TIME (HZ/50UL)
75 /* Number of requests a "batching" process may submit */
76 #define BLK_BATCH_REQ 32
79 * Return the threshold (number of used requests) at which the queue is
80 * considered to be congested. It include a little hysteresis to keep the
81 * context switch rate down.
83 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
85 return q
->nr_congestion_on
;
89 * The threshold at which a queue is considered to be uncongested
91 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
93 return q
->nr_congestion_off
;
96 static void blk_queue_congestion_threshold(struct request_queue
*q
)
100 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
101 if (nr
> q
->nr_requests
)
103 q
->nr_congestion_on
= nr
;
105 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
108 q
->nr_congestion_off
= nr
;
112 * A queue has just exitted congestion. Note this in the global counter of
113 * congested queues, and wake up anyone who was waiting for requests to be
116 static void clear_queue_congested(request_queue_t
*q
, int rw
)
119 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
121 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
122 clear_bit(bit
, &q
->backing_dev_info
.state
);
123 smp_mb__after_clear_bit();
124 if (waitqueue_active(wqh
))
129 * A queue has just entered congestion. Flag that in the queue's VM-visible
130 * state flags and increment the global gounter of congested queues.
132 static void set_queue_congested(request_queue_t
*q
, int rw
)
136 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
137 set_bit(bit
, &q
->backing_dev_info
.state
);
141 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
144 * Locates the passed device's request queue and returns the address of its
147 * Will return NULL if the request queue cannot be located.
149 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
151 struct backing_dev_info
*ret
= NULL
;
152 request_queue_t
*q
= bdev_get_queue(bdev
);
155 ret
= &q
->backing_dev_info
;
159 EXPORT_SYMBOL(blk_get_backing_dev_info
);
161 void blk_queue_activity_fn(request_queue_t
*q
, activity_fn
*fn
, void *data
)
164 q
->activity_data
= data
;
167 EXPORT_SYMBOL(blk_queue_activity_fn
);
170 * blk_queue_prep_rq - set a prepare_request function for queue
172 * @pfn: prepare_request function
174 * It's possible for a queue to register a prepare_request callback which
175 * is invoked before the request is handed to the request_fn. The goal of
176 * the function is to prepare a request for I/O, it can be used to build a
177 * cdb from the request data for instance.
180 void blk_queue_prep_rq(request_queue_t
*q
, prep_rq_fn
*pfn
)
185 EXPORT_SYMBOL(blk_queue_prep_rq
);
188 * blk_queue_merge_bvec - set a merge_bvec function for queue
190 * @mbfn: merge_bvec_fn
192 * Usually queues have static limitations on the max sectors or segments that
193 * we can put in a request. Stacking drivers may have some settings that
194 * are dynamic, and thus we have to query the queue whether it is ok to
195 * add a new bio_vec to a bio at a given offset or not. If the block device
196 * has such limitations, it needs to register a merge_bvec_fn to control
197 * the size of bio's sent to it. Note that a block device *must* allow a
198 * single page to be added to an empty bio. The block device driver may want
199 * to use the bio_split() function to deal with these bio's. By default
200 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
203 void blk_queue_merge_bvec(request_queue_t
*q
, merge_bvec_fn
*mbfn
)
205 q
->merge_bvec_fn
= mbfn
;
208 EXPORT_SYMBOL(blk_queue_merge_bvec
);
211 * blk_queue_make_request - define an alternate make_request function for a device
212 * @q: the request queue for the device to be affected
213 * @mfn: the alternate make_request function
216 * The normal way for &struct bios to be passed to a device
217 * driver is for them to be collected into requests on a request
218 * queue, and then to allow the device driver to select requests
219 * off that queue when it is ready. This works well for many block
220 * devices. However some block devices (typically virtual devices
221 * such as md or lvm) do not benefit from the processing on the
222 * request queue, and are served best by having the requests passed
223 * directly to them. This can be achieved by providing a function
224 * to blk_queue_make_request().
227 * The driver that does this *must* be able to deal appropriately
228 * with buffers in "highmemory". This can be accomplished by either calling
229 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
230 * blk_queue_bounce() to create a buffer in normal memory.
232 void blk_queue_make_request(request_queue_t
* q
, make_request_fn
* mfn
)
237 q
->nr_requests
= BLKDEV_MAX_RQ
;
238 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
, 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
)
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
)
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 q
->max_sectors
= q
->max_hw_sectors
= max_sectors
;
563 EXPORT_SYMBOL(blk_queue_max_sectors
);
566 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
567 * @q: the request queue for the device
568 * @max_segments: max number of segments
571 * Enables a low level driver to set an upper limit on the number of
572 * physical data segments in a request. This would be the largest sized
573 * scatter list the driver could handle.
575 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
579 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
582 q
->max_phys_segments
= max_segments
;
585 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
588 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
589 * @q: the request queue for the device
590 * @max_segments: max number of segments
593 * Enables a low level driver to set an upper limit on the number of
594 * hw data segments in a request. This would be the largest number of
595 * address/length pairs the host adapter can actually give as once
598 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
602 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
605 q
->max_hw_segments
= max_segments
;
608 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
611 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
612 * @q: the request queue for the device
613 * @max_size: max size of segment in bytes
616 * Enables a low level driver to set an upper limit on the size of a
619 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
621 if (max_size
< PAGE_CACHE_SIZE
) {
622 max_size
= PAGE_CACHE_SIZE
;
623 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
626 q
->max_segment_size
= max_size
;
629 EXPORT_SYMBOL(blk_queue_max_segment_size
);
632 * blk_queue_hardsect_size - set hardware sector size for the queue
633 * @q: the request queue for the device
634 * @size: the hardware sector size, in bytes
637 * This should typically be set to the lowest possible sector size
638 * that the hardware can operate on (possible without reverting to
639 * even internal read-modify-write operations). Usually the default
640 * of 512 covers most hardware.
642 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
644 q
->hardsect_size
= size
;
647 EXPORT_SYMBOL(blk_queue_hardsect_size
);
650 * Returns the minimum that is _not_ zero, unless both are zero.
652 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
655 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
656 * @t: the stacking driver (top)
657 * @b: the underlying device (bottom)
659 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
661 /* zero is "infinity" */
662 t
->max_sectors
= t
->max_hw_sectors
=
663 min_not_zero(t
->max_sectors
,b
->max_sectors
);
665 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
666 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
667 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
668 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
671 EXPORT_SYMBOL(blk_queue_stack_limits
);
674 * blk_queue_segment_boundary - set boundary rules for segment merging
675 * @q: the request queue for the device
676 * @mask: the memory boundary mask
678 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
680 if (mask
< PAGE_CACHE_SIZE
- 1) {
681 mask
= PAGE_CACHE_SIZE
- 1;
682 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
685 q
->seg_boundary_mask
= mask
;
688 EXPORT_SYMBOL(blk_queue_segment_boundary
);
691 * blk_queue_dma_alignment - set dma length and memory alignment
692 * @q: the request queue for the device
693 * @mask: alignment mask
696 * set required memory and length aligment for direct dma transactions.
697 * this is used when buiding direct io requests for the queue.
700 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
702 q
->dma_alignment
= mask
;
705 EXPORT_SYMBOL(blk_queue_dma_alignment
);
708 * blk_queue_find_tag - find a request by its tag and queue
710 * @q: The request queue for the device
711 * @tag: The tag of the request
714 * Should be used when a device returns a tag and you want to match
717 * no locks need be held.
719 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
721 struct blk_queue_tag
*bqt
= q
->queue_tags
;
723 if (unlikely(bqt
== NULL
|| tag
>= bqt
->real_max_depth
))
726 return bqt
->tag_index
[tag
];
729 EXPORT_SYMBOL(blk_queue_find_tag
);
732 * __blk_queue_free_tags - release tag maintenance info
733 * @q: the request queue for the device
736 * blk_cleanup_queue() will take care of calling this function, if tagging
737 * has been used. So there's no need to call this directly.
739 static void __blk_queue_free_tags(request_queue_t
*q
)
741 struct blk_queue_tag
*bqt
= q
->queue_tags
;
746 if (atomic_dec_and_test(&bqt
->refcnt
)) {
748 BUG_ON(!list_empty(&bqt
->busy_list
));
750 kfree(bqt
->tag_index
);
751 bqt
->tag_index
= NULL
;
759 q
->queue_tags
= NULL
;
760 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
764 * blk_queue_free_tags - release tag maintenance info
765 * @q: the request queue for the device
768 * This is used to disabled tagged queuing to a device, yet leave
771 void blk_queue_free_tags(request_queue_t
*q
)
773 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
776 EXPORT_SYMBOL(blk_queue_free_tags
);
779 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
781 struct request
**tag_index
;
782 unsigned long *tag_map
;
785 if (depth
> q
->nr_requests
* 2) {
786 depth
= q
->nr_requests
* 2;
787 printk(KERN_ERR
"%s: adjusted depth to %d\n",
788 __FUNCTION__
, depth
);
791 tag_index
= kmalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
795 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
796 tag_map
= kmalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
800 memset(tag_index
, 0, depth
* sizeof(struct request
*));
801 memset(tag_map
, 0, nr_ulongs
* sizeof(unsigned long));
802 tags
->real_max_depth
= depth
;
803 tags
->max_depth
= depth
;
804 tags
->tag_index
= tag_index
;
805 tags
->tag_map
= tag_map
;
814 * blk_queue_init_tags - initialize the queue tag info
815 * @q: the request queue for the device
816 * @depth: the maximum queue depth supported
817 * @tags: the tag to use
819 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
820 struct blk_queue_tag
*tags
)
824 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
826 if (!tags
&& !q
->queue_tags
) {
827 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
831 if (init_tag_map(q
, tags
, depth
))
834 INIT_LIST_HEAD(&tags
->busy_list
);
836 atomic_set(&tags
->refcnt
, 1);
837 } else if (q
->queue_tags
) {
838 if ((rc
= blk_queue_resize_tags(q
, depth
)))
840 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
843 atomic_inc(&tags
->refcnt
);
846 * assign it, all done
848 q
->queue_tags
= tags
;
849 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
856 EXPORT_SYMBOL(blk_queue_init_tags
);
859 * blk_queue_resize_tags - change the queueing depth
860 * @q: the request queue for the device
861 * @new_depth: the new max command queueing depth
864 * Must be called with the queue lock held.
866 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
868 struct blk_queue_tag
*bqt
= q
->queue_tags
;
869 struct request
**tag_index
;
870 unsigned long *tag_map
;
871 int max_depth
, nr_ulongs
;
877 * if we already have large enough real_max_depth. just
878 * adjust max_depth. *NOTE* as requests with tag value
879 * between new_depth and real_max_depth can be in-flight, tag
880 * map can not be shrunk blindly here.
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 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
899 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* 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
))
931 * This can happen after tag depth has been reduced.
932 * FIXME: how about a warning or info message here?
936 if (unlikely(!__test_and_clear_bit(tag
, bqt
->tag_map
))) {
937 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
942 list_del_init(&rq
->queuelist
);
943 rq
->flags
&= ~REQ_QUEUED
;
946 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
947 printk(KERN_ERR
"%s: tag %d is missing\n",
950 bqt
->tag_index
[tag
] = NULL
;
954 EXPORT_SYMBOL(blk_queue_end_tag
);
957 * blk_queue_start_tag - find a free tag and assign it
958 * @q: the request queue for the device
959 * @rq: the block request that needs tagging
962 * This can either be used as a stand-alone helper, or possibly be
963 * assigned as the queue &prep_rq_fn (in which case &struct request
964 * automagically gets a tag assigned). Note that this function
965 * assumes that any type of request can be queued! if this is not
966 * true for your device, you must check the request type before
967 * calling this function. The request will also be removed from
968 * the request queue, so it's the drivers responsibility to readd
969 * it if it should need to be restarted for some reason.
972 * queue lock must be held.
974 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
976 struct blk_queue_tag
*bqt
= q
->queue_tags
;
979 if (unlikely((rq
->flags
& REQ_QUEUED
))) {
981 "%s: request %p for device [%s] already tagged %d",
983 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
987 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
988 if (tag
>= bqt
->max_depth
)
991 __set_bit(tag
, bqt
->tag_map
);
993 rq
->flags
|= REQ_QUEUED
;
995 bqt
->tag_index
[tag
] = rq
;
996 blkdev_dequeue_request(rq
);
997 list_add(&rq
->queuelist
, &bqt
->busy_list
);
1002 EXPORT_SYMBOL(blk_queue_start_tag
);
1005 * blk_queue_invalidate_tags - invalidate all pending tags
1006 * @q: the request queue for the device
1009 * Hardware conditions may dictate a need to stop all pending requests.
1010 * In this case, we will safely clear the block side of the tag queue and
1011 * readd all requests to the request queue in the right order.
1014 * queue lock must be held.
1016 void blk_queue_invalidate_tags(request_queue_t
*q
)
1018 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1019 struct list_head
*tmp
, *n
;
1022 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1023 rq
= list_entry_rq(tmp
);
1025 if (rq
->tag
== -1) {
1027 "%s: bad tag found on list\n", __FUNCTION__
);
1028 list_del_init(&rq
->queuelist
);
1029 rq
->flags
&= ~REQ_QUEUED
;
1031 blk_queue_end_tag(q
, rq
);
1033 rq
->flags
&= ~REQ_STARTED
;
1034 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1038 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1040 static char *rq_flags
[] = {
1060 "REQ_DRIVE_TASKFILE",
1067 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1071 printk("%s: dev %s: flags = ", msg
,
1072 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?");
1075 if (rq
->flags
& (1 << bit
))
1076 printk("%s ", rq_flags
[bit
]);
1078 } while (bit
< __REQ_NR_BITS
);
1080 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1082 rq
->current_nr_sectors
);
1083 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1085 if (rq
->flags
& (REQ_BLOCK_PC
| REQ_PC
)) {
1087 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1088 printk("%02x ", rq
->cmd
[bit
]);
1093 EXPORT_SYMBOL(blk_dump_rq_flags
);
1095 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1097 struct bio_vec
*bv
, *bvprv
= NULL
;
1098 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1099 int high
, highprv
= 1;
1101 if (unlikely(!bio
->bi_io_vec
))
1104 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1105 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1106 bio_for_each_segment(bv
, bio
, i
) {
1108 * the trick here is making sure that a high page is never
1109 * considered part of another segment, since that might
1110 * change with the bounce page.
1112 high
= page_to_pfn(bv
->bv_page
) >= q
->bounce_pfn
;
1113 if (high
|| highprv
)
1114 goto new_hw_segment
;
1116 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1118 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1120 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1122 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1123 goto new_hw_segment
;
1125 seg_size
+= bv
->bv_len
;
1126 hw_seg_size
+= bv
->bv_len
;
1131 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1132 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1133 hw_seg_size
+= bv
->bv_len
;
1136 if (hw_seg_size
> bio
->bi_hw_front_size
)
1137 bio
->bi_hw_front_size
= hw_seg_size
;
1138 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1144 seg_size
= bv
->bv_len
;
1147 if (hw_seg_size
> bio
->bi_hw_back_size
)
1148 bio
->bi_hw_back_size
= hw_seg_size
;
1149 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1150 bio
->bi_hw_front_size
= hw_seg_size
;
1151 bio
->bi_phys_segments
= nr_phys_segs
;
1152 bio
->bi_hw_segments
= nr_hw_segs
;
1153 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1157 static int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1160 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1163 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1165 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1169 * bio and nxt are contigous in memory, check if the queue allows
1170 * these two to be merged into one
1172 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1178 static int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1181 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1182 blk_recount_segments(q
, bio
);
1183 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1184 blk_recount_segments(q
, nxt
);
1185 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1186 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1188 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1195 * map a request to scatterlist, return number of sg entries setup. Caller
1196 * must make sure sg can hold rq->nr_phys_segments entries
1198 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1200 struct bio_vec
*bvec
, *bvprv
;
1202 int nsegs
, i
, cluster
;
1205 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1208 * for each bio in rq
1211 rq_for_each_bio(bio
, rq
) {
1213 * for each segment in bio
1215 bio_for_each_segment(bvec
, bio
, i
) {
1216 int nbytes
= bvec
->bv_len
;
1218 if (bvprv
&& cluster
) {
1219 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1222 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1224 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1227 sg
[nsegs
- 1].length
+= nbytes
;
1230 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1231 sg
[nsegs
].page
= bvec
->bv_page
;
1232 sg
[nsegs
].length
= nbytes
;
1233 sg
[nsegs
].offset
= bvec
->bv_offset
;
1238 } /* segments in bio */
1244 EXPORT_SYMBOL(blk_rq_map_sg
);
1247 * the standard queue merge functions, can be overridden with device
1248 * specific ones if so desired
1251 static inline int ll_new_mergeable(request_queue_t
*q
,
1252 struct request
*req
,
1255 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1257 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1258 req
->flags
|= REQ_NOMERGE
;
1259 if (req
== q
->last_merge
)
1260 q
->last_merge
= NULL
;
1265 * A hw segment is just getting larger, bump just the phys
1268 req
->nr_phys_segments
+= nr_phys_segs
;
1272 static inline int ll_new_hw_segment(request_queue_t
*q
,
1273 struct request
*req
,
1276 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1277 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1279 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1280 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1281 req
->flags
|= REQ_NOMERGE
;
1282 if (req
== q
->last_merge
)
1283 q
->last_merge
= NULL
;
1288 * This will form the start of a new hw segment. Bump both
1291 req
->nr_hw_segments
+= nr_hw_segs
;
1292 req
->nr_phys_segments
+= nr_phys_segs
;
1296 static int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
,
1301 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1302 req
->flags
|= REQ_NOMERGE
;
1303 if (req
== q
->last_merge
)
1304 q
->last_merge
= NULL
;
1307 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1308 blk_recount_segments(q
, req
->biotail
);
1309 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1310 blk_recount_segments(q
, bio
);
1311 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1312 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1313 !BIOVEC_VIRT_OVERSIZE(len
)) {
1314 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1317 if (req
->nr_hw_segments
== 1)
1318 req
->bio
->bi_hw_front_size
= len
;
1319 if (bio
->bi_hw_segments
== 1)
1320 bio
->bi_hw_back_size
= len
;
1325 return ll_new_hw_segment(q
, req
, bio
);
1328 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1333 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1334 req
->flags
|= REQ_NOMERGE
;
1335 if (req
== q
->last_merge
)
1336 q
->last_merge
= NULL
;
1339 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1340 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1341 blk_recount_segments(q
, bio
);
1342 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1343 blk_recount_segments(q
, req
->bio
);
1344 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1345 !BIOVEC_VIRT_OVERSIZE(len
)) {
1346 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1349 if (bio
->bi_hw_segments
== 1)
1350 bio
->bi_hw_front_size
= len
;
1351 if (req
->nr_hw_segments
== 1)
1352 req
->biotail
->bi_hw_back_size
= len
;
1357 return ll_new_hw_segment(q
, req
, bio
);
1360 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1361 struct request
*next
)
1363 int total_phys_segments
;
1364 int total_hw_segments
;
1367 * First check if the either of the requests are re-queued
1368 * requests. Can't merge them if they are.
1370 if (req
->special
|| next
->special
)
1374 * Will it become too large?
1376 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1379 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1380 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1381 total_phys_segments
--;
1383 if (total_phys_segments
> q
->max_phys_segments
)
1386 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1387 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1388 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1390 * propagate the combined length to the end of the requests
1392 if (req
->nr_hw_segments
== 1)
1393 req
->bio
->bi_hw_front_size
= len
;
1394 if (next
->nr_hw_segments
== 1)
1395 next
->biotail
->bi_hw_back_size
= len
;
1396 total_hw_segments
--;
1399 if (total_hw_segments
> q
->max_hw_segments
)
1402 /* Merge is OK... */
1403 req
->nr_phys_segments
= total_phys_segments
;
1404 req
->nr_hw_segments
= total_hw_segments
;
1409 * "plug" the device if there are no outstanding requests: this will
1410 * force the transfer to start only after we have put all the requests
1413 * This is called with interrupts off and no requests on the queue and
1414 * with the queue lock held.
1416 void blk_plug_device(request_queue_t
*q
)
1418 WARN_ON(!irqs_disabled());
1421 * don't plug a stopped queue, it must be paired with blk_start_queue()
1422 * which will restart the queueing
1424 if (test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
))
1427 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1428 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1431 EXPORT_SYMBOL(blk_plug_device
);
1434 * remove the queue from the plugged list, if present. called with
1435 * queue lock held and interrupts disabled.
1437 int blk_remove_plug(request_queue_t
*q
)
1439 WARN_ON(!irqs_disabled());
1441 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1444 del_timer(&q
->unplug_timer
);
1448 EXPORT_SYMBOL(blk_remove_plug
);
1451 * remove the plug and let it rip..
1453 void __generic_unplug_device(request_queue_t
*q
)
1455 if (unlikely(test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
)))
1458 if (!blk_remove_plug(q
))
1463 EXPORT_SYMBOL(__generic_unplug_device
);
1466 * generic_unplug_device - fire a request queue
1467 * @q: The &request_queue_t in question
1470 * Linux uses plugging to build bigger requests queues before letting
1471 * the device have at them. If a queue is plugged, the I/O scheduler
1472 * is still adding and merging requests on the queue. Once the queue
1473 * gets unplugged, the request_fn defined for the queue is invoked and
1474 * transfers started.
1476 void generic_unplug_device(request_queue_t
*q
)
1478 spin_lock_irq(q
->queue_lock
);
1479 __generic_unplug_device(q
);
1480 spin_unlock_irq(q
->queue_lock
);
1482 EXPORT_SYMBOL(generic_unplug_device
);
1484 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1487 request_queue_t
*q
= bdi
->unplug_io_data
;
1490 * devices don't necessarily have an ->unplug_fn defined
1496 static void blk_unplug_work(void *data
)
1498 request_queue_t
*q
= data
;
1503 static void blk_unplug_timeout(unsigned long data
)
1505 request_queue_t
*q
= (request_queue_t
*)data
;
1507 kblockd_schedule_work(&q
->unplug_work
);
1511 * blk_start_queue - restart a previously stopped queue
1512 * @q: The &request_queue_t in question
1515 * blk_start_queue() will clear the stop flag on the queue, and call
1516 * the request_fn for the queue if it was in a stopped state when
1517 * entered. Also see blk_stop_queue(). Queue lock must be held.
1519 void blk_start_queue(request_queue_t
*q
)
1521 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1524 * one level of recursion is ok and is much faster than kicking
1525 * the unplug handling
1527 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1529 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1532 kblockd_schedule_work(&q
->unplug_work
);
1536 EXPORT_SYMBOL(blk_start_queue
);
1539 * blk_stop_queue - stop a queue
1540 * @q: The &request_queue_t in question
1543 * The Linux block layer assumes that a block driver will consume all
1544 * entries on the request queue when the request_fn strategy is called.
1545 * Often this will not happen, because of hardware limitations (queue
1546 * depth settings). If a device driver gets a 'queue full' response,
1547 * or if it simply chooses not to queue more I/O at one point, it can
1548 * call this function to prevent the request_fn from being called until
1549 * the driver has signalled it's ready to go again. This happens by calling
1550 * blk_start_queue() to restart queue operations. Queue lock must be held.
1552 void blk_stop_queue(request_queue_t
*q
)
1555 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1557 EXPORT_SYMBOL(blk_stop_queue
);
1560 * blk_sync_queue - cancel any pending callbacks on a queue
1564 * The block layer may perform asynchronous callback activity
1565 * on a queue, such as calling the unplug function after a timeout.
1566 * A block device may call blk_sync_queue to ensure that any
1567 * such activity is cancelled, thus allowing it to release resources
1568 * the the callbacks might use. The caller must already have made sure
1569 * that its ->make_request_fn will not re-add plugging prior to calling
1573 void blk_sync_queue(struct request_queue
*q
)
1575 del_timer_sync(&q
->unplug_timer
);
1578 EXPORT_SYMBOL(blk_sync_queue
);
1581 * blk_run_queue - run a single device queue
1582 * @q: The queue to run
1584 void blk_run_queue(struct request_queue
*q
)
1586 unsigned long flags
;
1588 spin_lock_irqsave(q
->queue_lock
, flags
);
1590 if (!elv_queue_empty(q
))
1592 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1594 EXPORT_SYMBOL(blk_run_queue
);
1597 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1598 * @q: the request queue to be released
1601 * blk_cleanup_queue is the pair to blk_init_queue() or
1602 * blk_queue_make_request(). It should be called when a request queue is
1603 * being released; typically when a block device is being de-registered.
1604 * Currently, its primary task it to free all the &struct request
1605 * structures that were allocated to the queue and the queue itself.
1608 * Hopefully the low level driver will have finished any
1609 * outstanding requests first...
1611 void blk_cleanup_queue(request_queue_t
* q
)
1613 struct request_list
*rl
= &q
->rq
;
1615 if (!atomic_dec_and_test(&q
->refcnt
))
1619 elevator_exit(q
->elevator
);
1624 mempool_destroy(rl
->rq_pool
);
1627 __blk_queue_free_tags(q
);
1629 blk_queue_ordered(q
, QUEUE_ORDERED_NONE
);
1631 kmem_cache_free(requestq_cachep
, q
);
1634 EXPORT_SYMBOL(blk_cleanup_queue
);
1636 static int blk_init_free_list(request_queue_t
*q
)
1638 struct request_list
*rl
= &q
->rq
;
1640 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1641 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1643 init_waitqueue_head(&rl
->wait
[READ
]);
1644 init_waitqueue_head(&rl
->wait
[WRITE
]);
1646 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1647 mempool_free_slab
, request_cachep
, q
->node
);
1655 static int __make_request(request_queue_t
*, struct bio
*);
1657 request_queue_t
*blk_alloc_queue(gfp_t gfp_mask
)
1659 return blk_alloc_queue_node(gfp_mask
, -1);
1661 EXPORT_SYMBOL(blk_alloc_queue
);
1663 request_queue_t
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1667 q
= kmem_cache_alloc_node(requestq_cachep
, gfp_mask
, node_id
);
1671 memset(q
, 0, sizeof(*q
));
1672 init_timer(&q
->unplug_timer
);
1673 atomic_set(&q
->refcnt
, 1);
1675 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1676 q
->backing_dev_info
.unplug_io_data
= q
;
1680 EXPORT_SYMBOL(blk_alloc_queue_node
);
1683 * blk_init_queue - prepare a request queue for use with a block device
1684 * @rfn: The function to be called to process requests that have been
1685 * placed on the queue.
1686 * @lock: Request queue spin lock
1689 * If a block device wishes to use the standard request handling procedures,
1690 * which sorts requests and coalesces adjacent requests, then it must
1691 * call blk_init_queue(). The function @rfn will be called when there
1692 * are requests on the queue that need to be processed. If the device
1693 * supports plugging, then @rfn may not be called immediately when requests
1694 * are available on the queue, but may be called at some time later instead.
1695 * Plugged queues are generally unplugged when a buffer belonging to one
1696 * of the requests on the queue is needed, or due to memory pressure.
1698 * @rfn is not required, or even expected, to remove all requests off the
1699 * queue, but only as many as it can handle at a time. If it does leave
1700 * requests on the queue, it is responsible for arranging that the requests
1701 * get dealt with eventually.
1703 * The queue spin lock must be held while manipulating the requests on the
1706 * Function returns a pointer to the initialized request queue, or NULL if
1707 * it didn't succeed.
1710 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1711 * when the block device is deactivated (such as at module unload).
1714 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1716 return blk_init_queue_node(rfn
, lock
, -1);
1718 EXPORT_SYMBOL(blk_init_queue
);
1721 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1723 request_queue_t
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1729 if (blk_init_free_list(q
))
1733 * if caller didn't supply a lock, they get per-queue locking with
1737 spin_lock_init(&q
->__queue_lock
);
1738 lock
= &q
->__queue_lock
;
1741 q
->request_fn
= rfn
;
1742 q
->back_merge_fn
= ll_back_merge_fn
;
1743 q
->front_merge_fn
= ll_front_merge_fn
;
1744 q
->merge_requests_fn
= ll_merge_requests_fn
;
1745 q
->prep_rq_fn
= NULL
;
1746 q
->unplug_fn
= generic_unplug_device
;
1747 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1748 q
->queue_lock
= lock
;
1750 blk_queue_segment_boundary(q
, 0xffffffff);
1752 blk_queue_make_request(q
, __make_request
);
1753 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1755 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1756 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1761 if (!elevator_init(q
, NULL
)) {
1762 blk_queue_congestion_threshold(q
);
1766 blk_cleanup_queue(q
);
1768 kmem_cache_free(requestq_cachep
, q
);
1771 EXPORT_SYMBOL(blk_init_queue_node
);
1773 int blk_get_queue(request_queue_t
*q
)
1775 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1776 atomic_inc(&q
->refcnt
);
1783 EXPORT_SYMBOL(blk_get_queue
);
1785 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
1787 if (rq
->flags
& REQ_ELVPRIV
)
1788 elv_put_request(q
, rq
);
1789 mempool_free(rq
, q
->rq
.rq_pool
);
1792 static inline struct request
*
1793 blk_alloc_request(request_queue_t
*q
, int rw
, struct bio
*bio
,
1794 int priv
, gfp_t gfp_mask
)
1796 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1802 * first three bits are identical in rq->flags and bio->bi_rw,
1803 * see bio.h and blkdev.h
1808 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
))) {
1809 mempool_free(rq
, q
->rq
.rq_pool
);
1812 rq
->flags
|= REQ_ELVPRIV
;
1819 * ioc_batching returns true if the ioc is a valid batching request and
1820 * should be given priority access to a request.
1822 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
1828 * Make sure the process is able to allocate at least 1 request
1829 * even if the batch times out, otherwise we could theoretically
1832 return ioc
->nr_batch_requests
== q
->nr_batching
||
1833 (ioc
->nr_batch_requests
> 0
1834 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1838 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1839 * will cause the process to be a "batcher" on all queues in the system. This
1840 * is the behaviour we want though - once it gets a wakeup it should be given
1843 static void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
1845 if (!ioc
|| ioc_batching(q
, ioc
))
1848 ioc
->nr_batch_requests
= q
->nr_batching
;
1849 ioc
->last_waited
= jiffies
;
1852 static void __freed_request(request_queue_t
*q
, int rw
)
1854 struct request_list
*rl
= &q
->rq
;
1856 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
1857 clear_queue_congested(q
, rw
);
1859 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
1860 if (waitqueue_active(&rl
->wait
[rw
]))
1861 wake_up(&rl
->wait
[rw
]);
1863 blk_clear_queue_full(q
, rw
);
1868 * A request has just been released. Account for it, update the full and
1869 * congestion status, wake up any waiters. Called under q->queue_lock.
1871 static void freed_request(request_queue_t
*q
, int rw
, int priv
)
1873 struct request_list
*rl
= &q
->rq
;
1879 __freed_request(q
, rw
);
1881 if (unlikely(rl
->starved
[rw
^ 1]))
1882 __freed_request(q
, rw
^ 1);
1885 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1887 * Get a free request, queue_lock must be held.
1888 * Returns NULL on failure, with queue_lock held.
1889 * Returns !NULL on success, with queue_lock *not held*.
1891 static struct request
*get_request(request_queue_t
*q
, int rw
, struct bio
*bio
,
1894 struct request
*rq
= NULL
;
1895 struct request_list
*rl
= &q
->rq
;
1896 struct io_context
*ioc
= current_io_context(GFP_ATOMIC
);
1899 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
1901 * The queue will fill after this allocation, so set it as
1902 * full, and mark this process as "batching". This process
1903 * will be allowed to complete a batch of requests, others
1906 if (!blk_queue_full(q
, rw
)) {
1907 ioc_set_batching(q
, ioc
);
1908 blk_set_queue_full(q
, rw
);
1912 switch (elv_may_queue(q
, rw
, bio
)) {
1915 case ELV_MQUEUE_MAY
:
1917 case ELV_MQUEUE_MUST
:
1921 if (blk_queue_full(q
, rw
) && !ioc_batching(q
, ioc
)) {
1923 * The queue is full and the allocating process is not a
1924 * "batcher", and not exempted by the IO scheduler
1931 * Only allow batching queuers to allocate up to 50% over the defined
1932 * limit of requests, otherwise we could have thousands of requests
1933 * allocated with any setting of ->nr_requests
1935 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
1939 rl
->starved
[rw
] = 0;
1940 if (rl
->count
[rw
] >= queue_congestion_on_threshold(q
))
1941 set_queue_congested(q
, rw
);
1943 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
1947 spin_unlock_irq(q
->queue_lock
);
1949 rq
= blk_alloc_request(q
, rw
, bio
, priv
, gfp_mask
);
1952 * Allocation failed presumably due to memory. Undo anything
1953 * we might have messed up.
1955 * Allocating task should really be put onto the front of the
1956 * wait queue, but this is pretty rare.
1958 spin_lock_irq(q
->queue_lock
);
1959 freed_request(q
, rw
, priv
);
1962 * in the very unlikely event that allocation failed and no
1963 * requests for this direction was pending, mark us starved
1964 * so that freeing of a request in the other direction will
1965 * notice us. another possible fix would be to split the
1966 * rq mempool into READ and WRITE
1969 if (unlikely(rl
->count
[rw
] == 0))
1970 rl
->starved
[rw
] = 1;
1975 if (ioc_batching(q
, ioc
))
1976 ioc
->nr_batch_requests
--;
1985 * No available requests for this queue, unplug the device and wait for some
1986 * requests to become available.
1988 * Called with q->queue_lock held, and returns with it unlocked.
1990 static struct request
*get_request_wait(request_queue_t
*q
, int rw
,
1995 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
1998 struct request_list
*rl
= &q
->rq
;
2000 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
2001 TASK_UNINTERRUPTIBLE
);
2003 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
2006 struct io_context
*ioc
;
2008 __generic_unplug_device(q
);
2009 spin_unlock_irq(q
->queue_lock
);
2013 * After sleeping, we become a "batching" process and
2014 * will be able to allocate at least one request, and
2015 * up to a big batch of them for a small period time.
2016 * See ioc_batching, ioc_set_batching
2018 ioc
= current_io_context(GFP_NOIO
);
2019 ioc_set_batching(q
, ioc
);
2021 spin_lock_irq(q
->queue_lock
);
2023 finish_wait(&rl
->wait
[rw
], &wait
);
2029 struct request
*blk_get_request(request_queue_t
*q
, int rw
, gfp_t gfp_mask
)
2033 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2035 spin_lock_irq(q
->queue_lock
);
2036 if (gfp_mask
& __GFP_WAIT
) {
2037 rq
= get_request_wait(q
, rw
, NULL
);
2039 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2041 spin_unlock_irq(q
->queue_lock
);
2043 /* q->queue_lock is unlocked at this point */
2047 EXPORT_SYMBOL(blk_get_request
);
2050 * blk_requeue_request - put a request back on queue
2051 * @q: request queue where request should be inserted
2052 * @rq: request to be inserted
2055 * Drivers often keep queueing requests until the hardware cannot accept
2056 * more, when that condition happens we need to put the request back
2057 * on the queue. Must be called with queue lock held.
2059 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2061 if (blk_rq_tagged(rq
))
2062 blk_queue_end_tag(q
, rq
);
2064 elv_requeue_request(q
, rq
);
2067 EXPORT_SYMBOL(blk_requeue_request
);
2070 * blk_insert_request - insert a special request in to a request queue
2071 * @q: request queue where request should be inserted
2072 * @rq: request to be inserted
2073 * @at_head: insert request at head or tail of queue
2074 * @data: private data
2077 * Many block devices need to execute commands asynchronously, so they don't
2078 * block the whole kernel from preemption during request execution. This is
2079 * accomplished normally by inserting aritficial requests tagged as
2080 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2081 * scheduled for actual execution by the request queue.
2083 * We have the option of inserting the head or the tail of the queue.
2084 * Typically we use the tail for new ioctls and so forth. We use the head
2085 * of the queue for things like a QUEUE_FULL message from a device, or a
2086 * host that is unable to accept a particular command.
2088 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2089 int at_head
, void *data
)
2091 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2092 unsigned long flags
;
2095 * tell I/O scheduler that this isn't a regular read/write (ie it
2096 * must not attempt merges on this) and that it acts as a soft
2099 rq
->flags
|= REQ_SPECIAL
| REQ_SOFTBARRIER
;
2103 spin_lock_irqsave(q
->queue_lock
, flags
);
2106 * If command is tagged, release the tag
2108 if (blk_rq_tagged(rq
))
2109 blk_queue_end_tag(q
, rq
);
2111 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2112 __elv_add_request(q
, rq
, where
, 0);
2114 if (blk_queue_plugged(q
))
2115 __generic_unplug_device(q
);
2118 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2121 EXPORT_SYMBOL(blk_insert_request
);
2124 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2125 * @q: request queue where request should be inserted
2126 * @rq: request structure to fill
2127 * @ubuf: the user buffer
2128 * @len: length of user data
2131 * Data will be mapped directly for zero copy io, if possible. Otherwise
2132 * a kernel bounce buffer is used.
2134 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2135 * still in process context.
2137 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2138 * before being submitted to the device, as pages mapped may be out of
2139 * reach. It's the callers responsibility to make sure this happens. The
2140 * original bio must be passed back in to blk_rq_unmap_user() for proper
2143 int blk_rq_map_user(request_queue_t
*q
, struct request
*rq
, void __user
*ubuf
,
2146 unsigned long uaddr
;
2150 if (len
> (q
->max_sectors
<< 9))
2155 reading
= rq_data_dir(rq
) == READ
;
2158 * if alignment requirement is satisfied, map in user pages for
2159 * direct dma. else, set up kernel bounce buffers
2161 uaddr
= (unsigned long) ubuf
;
2162 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2163 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2165 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2168 rq
->bio
= rq
->biotail
= bio
;
2169 blk_rq_bio_prep(q
, rq
, bio
);
2171 rq
->buffer
= rq
->data
= NULL
;
2177 * bio is the err-ptr
2179 return PTR_ERR(bio
);
2182 EXPORT_SYMBOL(blk_rq_map_user
);
2185 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2186 * @q: request queue where request should be inserted
2187 * @rq: request to map data to
2188 * @iov: pointer to the iovec
2189 * @iov_count: number of elements in the iovec
2192 * Data will be mapped directly for zero copy io, if possible. Otherwise
2193 * a kernel bounce buffer is used.
2195 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2196 * still in process context.
2198 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2199 * before being submitted to the device, as pages mapped may be out of
2200 * reach. It's the callers responsibility to make sure this happens. The
2201 * original bio must be passed back in to blk_rq_unmap_user() for proper
2204 int blk_rq_map_user_iov(request_queue_t
*q
, struct request
*rq
,
2205 struct sg_iovec
*iov
, int iov_count
)
2209 if (!iov
|| iov_count
<= 0)
2212 /* we don't allow misaligned data like bio_map_user() does. If the
2213 * user is using sg, they're expected to know the alignment constraints
2214 * and respect them accordingly */
2215 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2217 return PTR_ERR(bio
);
2219 rq
->bio
= rq
->biotail
= bio
;
2220 blk_rq_bio_prep(q
, rq
, bio
);
2221 rq
->buffer
= rq
->data
= NULL
;
2222 rq
->data_len
= bio
->bi_size
;
2226 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2229 * blk_rq_unmap_user - unmap a request with user data
2230 * @bio: bio to be unmapped
2231 * @ulen: length of user buffer
2234 * Unmap a bio previously mapped by blk_rq_map_user().
2236 int blk_rq_unmap_user(struct bio
*bio
, unsigned int ulen
)
2241 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2242 bio_unmap_user(bio
);
2244 ret
= bio_uncopy_user(bio
);
2250 EXPORT_SYMBOL(blk_rq_unmap_user
);
2253 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2254 * @q: request queue where request should be inserted
2255 * @rq: request to fill
2256 * @kbuf: the kernel buffer
2257 * @len: length of user data
2258 * @gfp_mask: memory allocation flags
2260 int blk_rq_map_kern(request_queue_t
*q
, struct request
*rq
, void *kbuf
,
2261 unsigned int len
, gfp_t gfp_mask
)
2265 if (len
> (q
->max_sectors
<< 9))
2270 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2272 return PTR_ERR(bio
);
2274 if (rq_data_dir(rq
) == WRITE
)
2275 bio
->bi_rw
|= (1 << BIO_RW
);
2277 rq
->bio
= rq
->biotail
= bio
;
2278 blk_rq_bio_prep(q
, rq
, bio
);
2280 rq
->buffer
= rq
->data
= NULL
;
2285 EXPORT_SYMBOL(blk_rq_map_kern
);
2288 * blk_execute_rq_nowait - insert a request into queue for execution
2289 * @q: queue to insert the request in
2290 * @bd_disk: matching gendisk
2291 * @rq: request to insert
2292 * @at_head: insert request at head or tail of queue
2293 * @done: I/O completion handler
2296 * Insert a fully prepared request at the back of the io scheduler queue
2297 * for execution. Don't wait for completion.
2299 void blk_execute_rq_nowait(request_queue_t
*q
, struct gendisk
*bd_disk
,
2300 struct request
*rq
, int at_head
,
2301 void (*done
)(struct request
*))
2303 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2305 rq
->rq_disk
= bd_disk
;
2306 rq
->flags
|= REQ_NOMERGE
;
2308 elv_add_request(q
, rq
, where
, 1);
2309 generic_unplug_device(q
);
2313 * blk_execute_rq - insert a request into queue for execution
2314 * @q: queue to insert the request in
2315 * @bd_disk: matching gendisk
2316 * @rq: request to insert
2317 * @at_head: insert request at head or tail of queue
2320 * Insert a fully prepared request at the back of the io scheduler queue
2321 * for execution and wait for completion.
2323 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2324 struct request
*rq
, int at_head
)
2326 DECLARE_COMPLETION(wait
);
2327 char sense
[SCSI_SENSE_BUFFERSIZE
];
2331 * we need an extra reference to the request, so we can look at
2332 * it after io completion
2337 memset(sense
, 0, sizeof(sense
));
2342 rq
->waiting
= &wait
;
2343 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2344 wait_for_completion(&wait
);
2353 EXPORT_SYMBOL(blk_execute_rq
);
2356 * blkdev_issue_flush - queue a flush
2357 * @bdev: blockdev to issue flush for
2358 * @error_sector: error sector
2361 * Issue a flush for the block device in question. Caller can supply
2362 * room for storing the error offset in case of a flush error, if they
2363 * wish to. Caller must run wait_for_completion() on its own.
2365 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2369 if (bdev
->bd_disk
== NULL
)
2372 q
= bdev_get_queue(bdev
);
2375 if (!q
->issue_flush_fn
)
2378 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2381 EXPORT_SYMBOL(blkdev_issue_flush
);
2383 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2385 int rw
= rq_data_dir(rq
);
2387 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2391 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2393 disk_round_stats(rq
->rq_disk
);
2394 rq
->rq_disk
->in_flight
++;
2399 * add-request adds a request to the linked list.
2400 * queue lock is held and interrupts disabled, as we muck with the
2401 * request queue list.
2403 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2405 drive_stat_acct(req
, req
->nr_sectors
, 1);
2408 q
->activity_fn(q
->activity_data
, rq_data_dir(req
));
2411 * elevator indicated where it wants this request to be
2412 * inserted at elevator_merge time
2414 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2418 * disk_round_stats() - Round off the performance stats on a struct
2421 * The average IO queue length and utilisation statistics are maintained
2422 * by observing the current state of the queue length and the amount of
2423 * time it has been in this state for.
2425 * Normally, that accounting is done on IO completion, but that can result
2426 * in more than a second's worth of IO being accounted for within any one
2427 * second, leading to >100% utilisation. To deal with that, we call this
2428 * function to do a round-off before returning the results when reading
2429 * /proc/diskstats. This accounts immediately for all queue usage up to
2430 * the current jiffies and restarts the counters again.
2432 void disk_round_stats(struct gendisk
*disk
)
2434 unsigned long now
= jiffies
;
2436 if (now
== disk
->stamp
)
2439 if (disk
->in_flight
) {
2440 __disk_stat_add(disk
, time_in_queue
,
2441 disk
->in_flight
* (now
- disk
->stamp
));
2442 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2448 * queue lock must be held
2450 static void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2452 struct request_list
*rl
= req
->rl
;
2456 if (unlikely(--req
->ref_count
))
2459 elv_completed_request(q
, req
);
2461 req
->rq_status
= RQ_INACTIVE
;
2465 * Request may not have originated from ll_rw_blk. if not,
2466 * it didn't come out of our reserved rq pools
2469 int rw
= rq_data_dir(req
);
2470 int priv
= req
->flags
& REQ_ELVPRIV
;
2472 BUG_ON(!list_empty(&req
->queuelist
));
2474 blk_free_request(q
, req
);
2475 freed_request(q
, rw
, priv
);
2479 void blk_put_request(struct request
*req
)
2481 unsigned long flags
;
2482 request_queue_t
*q
= req
->q
;
2485 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2486 * following if (q) test.
2489 spin_lock_irqsave(q
->queue_lock
, flags
);
2490 __blk_put_request(q
, req
);
2491 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2495 EXPORT_SYMBOL(blk_put_request
);
2498 * blk_end_sync_rq - executes a completion event on a request
2499 * @rq: request to complete
2501 void blk_end_sync_rq(struct request
*rq
)
2503 struct completion
*waiting
= rq
->waiting
;
2506 __blk_put_request(rq
->q
, rq
);
2509 * complete last, if this is a stack request the process (and thus
2510 * the rq pointer) could be invalid right after this complete()
2514 EXPORT_SYMBOL(blk_end_sync_rq
);
2517 * blk_congestion_wait - wait for a queue to become uncongested
2518 * @rw: READ or WRITE
2519 * @timeout: timeout in jiffies
2521 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2522 * If no queues are congested then just wait for the next request to be
2525 long blk_congestion_wait(int rw
, long timeout
)
2529 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
2531 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
2532 ret
= io_schedule_timeout(timeout
);
2533 finish_wait(wqh
, &wait
);
2537 EXPORT_SYMBOL(blk_congestion_wait
);
2540 * Has to be called with the request spinlock acquired
2542 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2543 struct request
*next
)
2545 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2551 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2554 if (rq_data_dir(req
) != rq_data_dir(next
)
2555 || req
->rq_disk
!= next
->rq_disk
2556 || next
->waiting
|| next
->special
)
2560 * If we are allowed to merge, then append bio list
2561 * from next to rq and release next. merge_requests_fn
2562 * will have updated segment counts, update sector
2565 if (!q
->merge_requests_fn(q
, req
, next
))
2569 * At this point we have either done a back merge
2570 * or front merge. We need the smaller start_time of
2571 * the merged requests to be the current request
2572 * for accounting purposes.
2574 if (time_after(req
->start_time
, next
->start_time
))
2575 req
->start_time
= next
->start_time
;
2577 req
->biotail
->bi_next
= next
->bio
;
2578 req
->biotail
= next
->biotail
;
2580 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2582 elv_merge_requests(q
, req
, next
);
2585 disk_round_stats(req
->rq_disk
);
2586 req
->rq_disk
->in_flight
--;
2589 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2591 __blk_put_request(q
, next
);
2595 static inline int attempt_back_merge(request_queue_t
*q
, struct request
*rq
)
2597 struct request
*next
= elv_latter_request(q
, rq
);
2600 return attempt_merge(q
, rq
, next
);
2605 static inline int attempt_front_merge(request_queue_t
*q
, struct request
*rq
)
2607 struct request
*prev
= elv_former_request(q
, rq
);
2610 return attempt_merge(q
, prev
, rq
);
2616 * blk_attempt_remerge - attempt to remerge active head with next request
2617 * @q: The &request_queue_t belonging to the device
2618 * @rq: The head request (usually)
2621 * For head-active devices, the queue can easily be unplugged so quickly
2622 * that proper merging is not done on the front request. This may hurt
2623 * performance greatly for some devices. The block layer cannot safely
2624 * do merging on that first request for these queues, but the driver can
2625 * call this function and make it happen any way. Only the driver knows
2626 * when it is safe to do so.
2628 void blk_attempt_remerge(request_queue_t
*q
, struct request
*rq
)
2630 unsigned long flags
;
2632 spin_lock_irqsave(q
->queue_lock
, flags
);
2633 attempt_back_merge(q
, rq
);
2634 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2637 EXPORT_SYMBOL(blk_attempt_remerge
);
2639 static int __make_request(request_queue_t
*q
, struct bio
*bio
)
2641 struct request
*req
;
2642 int el_ret
, rw
, nr_sectors
, cur_nr_sectors
, barrier
, err
, sync
;
2643 unsigned short prio
;
2646 sector
= bio
->bi_sector
;
2647 nr_sectors
= bio_sectors(bio
);
2648 cur_nr_sectors
= bio_cur_sectors(bio
);
2649 prio
= bio_prio(bio
);
2651 rw
= bio_data_dir(bio
);
2652 sync
= bio_sync(bio
);
2655 * low level driver can indicate that it wants pages above a
2656 * certain limit bounced to low memory (ie for highmem, or even
2657 * ISA dma in theory)
2659 blk_queue_bounce(q
, &bio
);
2661 spin_lock_prefetch(q
->queue_lock
);
2663 barrier
= bio_barrier(bio
);
2664 if (unlikely(barrier
) && (q
->ordered
== QUEUE_ORDERED_NONE
)) {
2669 spin_lock_irq(q
->queue_lock
);
2671 if (unlikely(barrier
) || elv_queue_empty(q
))
2674 el_ret
= elv_merge(q
, &req
, bio
);
2676 case ELEVATOR_BACK_MERGE
:
2677 BUG_ON(!rq_mergeable(req
));
2679 if (!q
->back_merge_fn(q
, req
, bio
))
2682 req
->biotail
->bi_next
= bio
;
2684 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2685 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2686 drive_stat_acct(req
, nr_sectors
, 0);
2687 if (!attempt_back_merge(q
, req
))
2688 elv_merged_request(q
, req
);
2691 case ELEVATOR_FRONT_MERGE
:
2692 BUG_ON(!rq_mergeable(req
));
2694 if (!q
->front_merge_fn(q
, req
, bio
))
2697 bio
->bi_next
= req
->bio
;
2701 * may not be valid. if the low level driver said
2702 * it didn't need a bounce buffer then it better
2703 * not touch req->buffer either...
2705 req
->buffer
= bio_data(bio
);
2706 req
->current_nr_sectors
= cur_nr_sectors
;
2707 req
->hard_cur_sectors
= cur_nr_sectors
;
2708 req
->sector
= req
->hard_sector
= sector
;
2709 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2710 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2711 drive_stat_acct(req
, nr_sectors
, 0);
2712 if (!attempt_front_merge(q
, req
))
2713 elv_merged_request(q
, req
);
2716 /* ELV_NO_MERGE: elevator says don't/can't merge. */
2723 * Grab a free request. This is might sleep but can not fail.
2724 * Returns with the queue unlocked.
2726 req
= get_request_wait(q
, rw
, bio
);
2729 * After dropping the lock and possibly sleeping here, our request
2730 * may now be mergeable after it had proven unmergeable (above).
2731 * We don't worry about that case for efficiency. It won't happen
2732 * often, and the elevators are able to handle it.
2735 req
->flags
|= REQ_CMD
;
2738 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2740 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2741 req
->flags
|= REQ_FAILFAST
;
2744 * REQ_BARRIER implies no merging, but lets make it explicit
2746 if (unlikely(barrier
))
2747 req
->flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2750 req
->hard_sector
= req
->sector
= sector
;
2751 req
->hard_nr_sectors
= req
->nr_sectors
= nr_sectors
;
2752 req
->current_nr_sectors
= req
->hard_cur_sectors
= cur_nr_sectors
;
2753 req
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2754 req
->nr_hw_segments
= bio_hw_segments(q
, bio
);
2755 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2756 req
->waiting
= NULL
;
2757 req
->bio
= req
->biotail
= bio
;
2759 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2760 req
->start_time
= jiffies
;
2762 spin_lock_irq(q
->queue_lock
);
2763 if (elv_queue_empty(q
))
2765 add_request(q
, req
);
2768 __generic_unplug_device(q
);
2770 spin_unlock_irq(q
->queue_lock
);
2774 bio_endio(bio
, nr_sectors
<< 9, err
);
2779 * If bio->bi_dev is a partition, remap the location
2781 static inline void blk_partition_remap(struct bio
*bio
)
2783 struct block_device
*bdev
= bio
->bi_bdev
;
2785 if (bdev
!= bdev
->bd_contains
) {
2786 struct hd_struct
*p
= bdev
->bd_part
;
2787 const int rw
= bio_data_dir(bio
);
2789 p
->sectors
[rw
] += bio_sectors(bio
);
2792 bio
->bi_sector
+= p
->start_sect
;
2793 bio
->bi_bdev
= bdev
->bd_contains
;
2797 static void handle_bad_sector(struct bio
*bio
)
2799 char b
[BDEVNAME_SIZE
];
2801 printk(KERN_INFO
"attempt to access beyond end of device\n");
2802 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
2803 bdevname(bio
->bi_bdev
, b
),
2805 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
2806 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
2808 set_bit(BIO_EOF
, &bio
->bi_flags
);
2812 * generic_make_request: hand a buffer to its device driver for I/O
2813 * @bio: The bio describing the location in memory and on the device.
2815 * generic_make_request() is used to make I/O requests of block
2816 * devices. It is passed a &struct bio, which describes the I/O that needs
2819 * generic_make_request() does not return any status. The
2820 * success/failure status of the request, along with notification of
2821 * completion, is delivered asynchronously through the bio->bi_end_io
2822 * function described (one day) else where.
2824 * The caller of generic_make_request must make sure that bi_io_vec
2825 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2826 * set to describe the device address, and the
2827 * bi_end_io and optionally bi_private are set to describe how
2828 * completion notification should be signaled.
2830 * generic_make_request and the drivers it calls may use bi_next if this
2831 * bio happens to be merged with someone else, and may change bi_dev and
2832 * bi_sector for remaps as it sees fit. So the values of these fields
2833 * should NOT be depended on after the call to generic_make_request.
2835 void generic_make_request(struct bio
*bio
)
2839 int ret
, nr_sectors
= bio_sectors(bio
);
2842 /* Test device or partition size, when known. */
2843 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
2845 sector_t sector
= bio
->bi_sector
;
2847 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2849 * This may well happen - the kernel calls bread()
2850 * without checking the size of the device, e.g., when
2851 * mounting a device.
2853 handle_bad_sector(bio
);
2859 * Resolve the mapping until finished. (drivers are
2860 * still free to implement/resolve their own stacking
2861 * by explicitly returning 0)
2863 * NOTE: we don't repeat the blk_size check for each new device.
2864 * Stacking drivers are expected to know what they are doing.
2867 char b
[BDEVNAME_SIZE
];
2869 q
= bdev_get_queue(bio
->bi_bdev
);
2872 "generic_make_request: Trying to access "
2873 "nonexistent block-device %s (%Lu)\n",
2874 bdevname(bio
->bi_bdev
, b
),
2875 (long long) bio
->bi_sector
);
2877 bio_endio(bio
, bio
->bi_size
, -EIO
);
2881 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
2882 printk("bio too big device %s (%u > %u)\n",
2883 bdevname(bio
->bi_bdev
, b
),
2889 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
2893 * If this device has partitions, remap block n
2894 * of partition p to block n+start(p) of the disk.
2896 blk_partition_remap(bio
);
2898 ret
= q
->make_request_fn(q
, bio
);
2902 EXPORT_SYMBOL(generic_make_request
);
2905 * submit_bio: submit a bio to the block device layer for I/O
2906 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2907 * @bio: The &struct bio which describes the I/O
2909 * submit_bio() is very similar in purpose to generic_make_request(), and
2910 * uses that function to do most of the work. Both are fairly rough
2911 * interfaces, @bio must be presetup and ready for I/O.
2914 void submit_bio(int rw
, struct bio
*bio
)
2916 int count
= bio_sectors(bio
);
2918 BIO_BUG_ON(!bio
->bi_size
);
2919 BIO_BUG_ON(!bio
->bi_io_vec
);
2922 mod_page_state(pgpgout
, count
);
2924 mod_page_state(pgpgin
, count
);
2926 if (unlikely(block_dump
)) {
2927 char b
[BDEVNAME_SIZE
];
2928 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
2929 current
->comm
, current
->pid
,
2930 (rw
& WRITE
) ? "WRITE" : "READ",
2931 (unsigned long long)bio
->bi_sector
,
2932 bdevname(bio
->bi_bdev
,b
));
2935 generic_make_request(bio
);
2938 EXPORT_SYMBOL(submit_bio
);
2940 static void blk_recalc_rq_segments(struct request
*rq
)
2942 struct bio
*bio
, *prevbio
= NULL
;
2943 int nr_phys_segs
, nr_hw_segs
;
2944 unsigned int phys_size
, hw_size
;
2945 request_queue_t
*q
= rq
->q
;
2950 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
2951 rq_for_each_bio(bio
, rq
) {
2952 /* Force bio hw/phys segs to be recalculated. */
2953 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
2955 nr_phys_segs
+= bio_phys_segments(q
, bio
);
2956 nr_hw_segs
+= bio_hw_segments(q
, bio
);
2958 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
2959 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
2961 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
2962 pseg
<= q
->max_segment_size
) {
2964 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
2968 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
2969 hseg
<= q
->max_segment_size
) {
2971 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
2978 rq
->nr_phys_segments
= nr_phys_segs
;
2979 rq
->nr_hw_segments
= nr_hw_segs
;
2982 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
2984 if (blk_fs_request(rq
)) {
2985 rq
->hard_sector
+= nsect
;
2986 rq
->hard_nr_sectors
-= nsect
;
2989 * Move the I/O submission pointers ahead if required.
2991 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
2992 (rq
->sector
<= rq
->hard_sector
)) {
2993 rq
->sector
= rq
->hard_sector
;
2994 rq
->nr_sectors
= rq
->hard_nr_sectors
;
2995 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
2996 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
2997 rq
->buffer
= bio_data(rq
->bio
);
3001 * if total number of sectors is less than the first segment
3002 * size, something has gone terribly wrong
3004 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3005 printk("blk: request botched\n");
3006 rq
->nr_sectors
= rq
->current_nr_sectors
;
3011 static int __end_that_request_first(struct request
*req
, int uptodate
,
3014 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3018 * extend uptodate bool to allow < 0 value to be direct io error
3021 if (end_io_error(uptodate
))
3022 error
= !uptodate
? -EIO
: uptodate
;
3025 * for a REQ_BLOCK_PC request, we want to carry any eventual
3026 * sense key with us all the way through
3028 if (!blk_pc_request(req
))
3032 if (blk_fs_request(req
) && !(req
->flags
& REQ_QUIET
))
3033 printk("end_request: I/O error, dev %s, sector %llu\n",
3034 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3035 (unsigned long long)req
->sector
);
3038 if (blk_fs_request(req
) && req
->rq_disk
) {
3039 const int rw
= rq_data_dir(req
);
3041 __disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3044 total_bytes
= bio_nbytes
= 0;
3045 while ((bio
= req
->bio
) != NULL
) {
3048 if (nr_bytes
>= bio
->bi_size
) {
3049 req
->bio
= bio
->bi_next
;
3050 nbytes
= bio
->bi_size
;
3051 bio_endio(bio
, nbytes
, error
);
3055 int idx
= bio
->bi_idx
+ next_idx
;
3057 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3058 blk_dump_rq_flags(req
, "__end_that");
3059 printk("%s: bio idx %d >= vcnt %d\n",
3061 bio
->bi_idx
, bio
->bi_vcnt
);
3065 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3066 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3069 * not a complete bvec done
3071 if (unlikely(nbytes
> nr_bytes
)) {
3072 bio_nbytes
+= nr_bytes
;
3073 total_bytes
+= nr_bytes
;
3078 * advance to the next vector
3081 bio_nbytes
+= nbytes
;
3084 total_bytes
+= nbytes
;
3087 if ((bio
= req
->bio
)) {
3089 * end more in this run, or just return 'not-done'
3091 if (unlikely(nr_bytes
<= 0))
3103 * if the request wasn't completed, update state
3106 bio_endio(bio
, bio_nbytes
, error
);
3107 bio
->bi_idx
+= next_idx
;
3108 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3109 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3112 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3113 blk_recalc_rq_segments(req
);
3118 * end_that_request_first - end I/O on a request
3119 * @req: the request being processed
3120 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3121 * @nr_sectors: number of sectors to end I/O on
3124 * Ends I/O on a number of sectors attached to @req, and sets it up
3125 * for the next range of segments (if any) in the cluster.
3128 * 0 - we are done with this request, call end_that_request_last()
3129 * 1 - still buffers pending for this request
3131 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3133 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3136 EXPORT_SYMBOL(end_that_request_first
);
3139 * end_that_request_chunk - end I/O on a request
3140 * @req: the request being processed
3141 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3142 * @nr_bytes: number of bytes to complete
3145 * Ends I/O on a number of bytes attached to @req, and sets it up
3146 * for the next range of segments (if any). Like end_that_request_first(),
3147 * but deals with bytes instead of sectors.
3150 * 0 - we are done with this request, call end_that_request_last()
3151 * 1 - still buffers pending for this request
3153 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3155 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3158 EXPORT_SYMBOL(end_that_request_chunk
);
3161 * queue lock must be held
3163 void end_that_request_last(struct request
*req
)
3165 struct gendisk
*disk
= req
->rq_disk
;
3167 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3168 laptop_io_completion();
3170 if (disk
&& blk_fs_request(req
)) {
3171 unsigned long duration
= jiffies
- req
->start_time
;
3172 const int rw
= rq_data_dir(req
);
3174 __disk_stat_inc(disk
, ios
[rw
]);
3175 __disk_stat_add(disk
, ticks
[rw
], duration
);
3176 disk_round_stats(disk
);
3182 __blk_put_request(req
->q
, req
);
3185 EXPORT_SYMBOL(end_that_request_last
);
3187 void end_request(struct request
*req
, int uptodate
)
3189 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3190 add_disk_randomness(req
->rq_disk
);
3191 blkdev_dequeue_request(req
);
3192 end_that_request_last(req
);
3196 EXPORT_SYMBOL(end_request
);
3198 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3200 /* first three bits are identical in rq->flags and bio->bi_rw */
3201 rq
->flags
|= (bio
->bi_rw
& 7);
3203 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3204 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3205 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3206 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3207 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3208 rq
->buffer
= bio_data(bio
);
3210 rq
->bio
= rq
->biotail
= bio
;
3213 EXPORT_SYMBOL(blk_rq_bio_prep
);
3215 int kblockd_schedule_work(struct work_struct
*work
)
3217 return queue_work(kblockd_workqueue
, work
);
3220 EXPORT_SYMBOL(kblockd_schedule_work
);
3222 void kblockd_flush(void)
3224 flush_workqueue(kblockd_workqueue
);
3226 EXPORT_SYMBOL(kblockd_flush
);
3228 int __init
blk_dev_init(void)
3230 kblockd_workqueue
= create_workqueue("kblockd");
3231 if (!kblockd_workqueue
)
3232 panic("Failed to create kblockd\n");
3234 request_cachep
= kmem_cache_create("blkdev_requests",
3235 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3237 requestq_cachep
= kmem_cache_create("blkdev_queue",
3238 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3240 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3241 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3243 blk_max_low_pfn
= max_low_pfn
;
3244 blk_max_pfn
= max_pfn
;
3250 * IO Context helper functions
3252 void put_io_context(struct io_context
*ioc
)
3257 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3259 if (atomic_dec_and_test(&ioc
->refcount
)) {
3260 if (ioc
->aic
&& ioc
->aic
->dtor
)
3261 ioc
->aic
->dtor(ioc
->aic
);
3262 if (ioc
->cic
&& ioc
->cic
->dtor
)
3263 ioc
->cic
->dtor(ioc
->cic
);
3265 kmem_cache_free(iocontext_cachep
, ioc
);
3268 EXPORT_SYMBOL(put_io_context
);
3270 /* Called by the exitting task */
3271 void exit_io_context(void)
3273 unsigned long flags
;
3274 struct io_context
*ioc
;
3276 local_irq_save(flags
);
3278 ioc
= current
->io_context
;
3279 current
->io_context
= NULL
;
3281 task_unlock(current
);
3282 local_irq_restore(flags
);
3284 if (ioc
->aic
&& ioc
->aic
->exit
)
3285 ioc
->aic
->exit(ioc
->aic
);
3286 if (ioc
->cic
&& ioc
->cic
->exit
)
3287 ioc
->cic
->exit(ioc
->cic
);
3289 put_io_context(ioc
);
3293 * If the current task has no IO context then create one and initialise it.
3294 * Otherwise, return its existing IO context.
3296 * This returned IO context doesn't have a specifically elevated refcount,
3297 * but since the current task itself holds a reference, the context can be
3298 * used in general code, so long as it stays within `current` context.
3300 struct io_context
*current_io_context(gfp_t gfp_flags
)
3302 struct task_struct
*tsk
= current
;
3303 struct io_context
*ret
;
3305 ret
= tsk
->io_context
;
3309 ret
= kmem_cache_alloc(iocontext_cachep
, gfp_flags
);
3311 atomic_set(&ret
->refcount
, 1);
3312 ret
->task
= current
;
3313 ret
->set_ioprio
= NULL
;
3314 ret
->last_waited
= jiffies
; /* doesn't matter... */
3315 ret
->nr_batch_requests
= 0; /* because this is 0 */
3318 tsk
->io_context
= ret
;
3323 EXPORT_SYMBOL(current_io_context
);
3326 * If the current task has no IO context then create one and initialise it.
3327 * If it does have a context, take a ref on it.
3329 * This is always called in the context of the task which submitted the I/O.
3331 struct io_context
*get_io_context(gfp_t gfp_flags
)
3333 struct io_context
*ret
;
3334 ret
= current_io_context(gfp_flags
);
3336 atomic_inc(&ret
->refcount
);
3339 EXPORT_SYMBOL(get_io_context
);
3341 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3343 struct io_context
*src
= *psrc
;
3344 struct io_context
*dst
= *pdst
;
3347 BUG_ON(atomic_read(&src
->refcount
) == 0);
3348 atomic_inc(&src
->refcount
);
3349 put_io_context(dst
);
3353 EXPORT_SYMBOL(copy_io_context
);
3355 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3357 struct io_context
*temp
;
3362 EXPORT_SYMBOL(swap_io_context
);
3367 struct queue_sysfs_entry
{
3368 struct attribute attr
;
3369 ssize_t (*show
)(struct request_queue
*, char *);
3370 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3374 queue_var_show(unsigned int var
, char *page
)
3376 return sprintf(page
, "%d\n", var
);
3380 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3382 char *p
= (char *) page
;
3384 *var
= simple_strtoul(p
, &p
, 10);
3388 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3390 return queue_var_show(q
->nr_requests
, (page
));
3394 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3396 struct request_list
*rl
= &q
->rq
;
3398 int ret
= queue_var_store(&q
->nr_requests
, page
, count
);
3399 if (q
->nr_requests
< BLKDEV_MIN_RQ
)
3400 q
->nr_requests
= BLKDEV_MIN_RQ
;
3401 blk_queue_congestion_threshold(q
);
3403 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3404 set_queue_congested(q
, READ
);
3405 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3406 clear_queue_congested(q
, READ
);
3408 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3409 set_queue_congested(q
, WRITE
);
3410 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3411 clear_queue_congested(q
, WRITE
);
3413 if (rl
->count
[READ
] >= q
->nr_requests
) {
3414 blk_set_queue_full(q
, READ
);
3415 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3416 blk_clear_queue_full(q
, READ
);
3417 wake_up(&rl
->wait
[READ
]);
3420 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3421 blk_set_queue_full(q
, WRITE
);
3422 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3423 blk_clear_queue_full(q
, WRITE
);
3424 wake_up(&rl
->wait
[WRITE
]);
3429 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3431 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3433 return queue_var_show(ra_kb
, (page
));
3437 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3439 unsigned long ra_kb
;
3440 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3442 spin_lock_irq(q
->queue_lock
);
3443 if (ra_kb
> (q
->max_sectors
>> 1))
3444 ra_kb
= (q
->max_sectors
>> 1);
3446 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3447 spin_unlock_irq(q
->queue_lock
);
3452 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3454 int max_sectors_kb
= q
->max_sectors
>> 1;
3456 return queue_var_show(max_sectors_kb
, (page
));
3460 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3462 unsigned long max_sectors_kb
,
3463 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3464 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3465 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3468 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3471 * Take the queue lock to update the readahead and max_sectors
3472 * values synchronously:
3474 spin_lock_irq(q
->queue_lock
);
3476 * Trim readahead window as well, if necessary:
3478 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3479 if (ra_kb
> max_sectors_kb
)
3480 q
->backing_dev_info
.ra_pages
=
3481 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3483 q
->max_sectors
= max_sectors_kb
<< 1;
3484 spin_unlock_irq(q
->queue_lock
);
3489 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3491 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3493 return queue_var_show(max_hw_sectors_kb
, (page
));
3497 static struct queue_sysfs_entry queue_requests_entry
= {
3498 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
3499 .show
= queue_requests_show
,
3500 .store
= queue_requests_store
,
3503 static struct queue_sysfs_entry queue_ra_entry
= {
3504 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
3505 .show
= queue_ra_show
,
3506 .store
= queue_ra_store
,
3509 static struct queue_sysfs_entry queue_max_sectors_entry
= {
3510 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
3511 .show
= queue_max_sectors_show
,
3512 .store
= queue_max_sectors_store
,
3515 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
3516 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
3517 .show
= queue_max_hw_sectors_show
,
3520 static struct queue_sysfs_entry queue_iosched_entry
= {
3521 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
3522 .show
= elv_iosched_show
,
3523 .store
= elv_iosched_store
,
3526 static struct attribute
*default_attrs
[] = {
3527 &queue_requests_entry
.attr
,
3528 &queue_ra_entry
.attr
,
3529 &queue_max_hw_sectors_entry
.attr
,
3530 &queue_max_sectors_entry
.attr
,
3531 &queue_iosched_entry
.attr
,
3535 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3538 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3540 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3541 struct request_queue
*q
;
3543 q
= container_of(kobj
, struct request_queue
, kobj
);
3547 return entry
->show(q
, page
);
3551 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3552 const char *page
, size_t length
)
3554 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3555 struct request_queue
*q
;
3557 q
= container_of(kobj
, struct request_queue
, kobj
);
3561 return entry
->store(q
, page
, length
);
3564 static struct sysfs_ops queue_sysfs_ops
= {
3565 .show
= queue_attr_show
,
3566 .store
= queue_attr_store
,
3569 static struct kobj_type queue_ktype
= {
3570 .sysfs_ops
= &queue_sysfs_ops
,
3571 .default_attrs
= default_attrs
,
3574 int blk_register_queue(struct gendisk
*disk
)
3578 request_queue_t
*q
= disk
->queue
;
3580 if (!q
|| !q
->request_fn
)
3583 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
3584 if (!q
->kobj
.parent
)
3587 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
3588 q
->kobj
.ktype
= &queue_ktype
;
3590 ret
= kobject_register(&q
->kobj
);
3594 ret
= elv_register_queue(q
);
3596 kobject_unregister(&q
->kobj
);
3603 void blk_unregister_queue(struct gendisk
*disk
)
3605 request_queue_t
*q
= disk
->queue
;
3607 if (q
&& q
->request_fn
) {
3608 elv_unregister_queue(q
);
3610 kobject_unregister(&q
->kobj
);
3611 kobject_put(&disk
->kobj
);