2 * linux/drivers/block/ll_rw_blk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
6 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
7 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
8 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
13 * This handles all read/write requests to block devices
15 #include <linux/config.h>
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/backing-dev.h>
19 #include <linux/bio.h>
20 #include <linux/blkdev.h>
21 #include <linux/highmem.h>
23 #include <linux/kernel_stat.h>
24 #include <linux/string.h>
25 #include <linux/init.h>
26 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/blkdev.h>
36 #include <scsi/scsi_cmnd.h>
38 static void blk_unplug_work(void *data);
39 static void blk_unplug_timeout(unsigned long data);
42 * For the allocated request tables
44 static kmem_cache_t *request_cachep;
47 * For queue allocation
49 static kmem_cache_t *requestq_cachep;
52 * For io context allocations
54 static kmem_cache_t *iocontext_cachep;
56 static wait_queue_head_t congestion_wqh[2] = {
57 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
58 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
62 * Controlling structure to kblockd
64 static struct workqueue_struct *kblockd_workqueue;
66 unsigned long blk_max_low_pfn, blk_max_pfn;
68 EXPORT_SYMBOL(blk_max_low_pfn);
69 EXPORT_SYMBOL(blk_max_pfn);
71 /* Amount of time in which a process may batch requests */
72 #define BLK_BATCH_TIME (HZ/50UL)
74 /* Number of requests a "batching" process may submit */
75 #define BLK_BATCH_REQ 32
78 * Return the threshold (number of used requests) at which the queue is
79 * considered to be congested. It include a little hysteresis to keep the
80 * context switch rate down.
82 static inline int queue_congestion_on_threshold(struct request_queue *q)
84 return q->nr_congestion_on;
88 * The threshold at which a queue is considered to be uncongested
90 static inline int queue_congestion_off_threshold(struct request_queue *q)
92 return q->nr_congestion_off;
95 static void blk_queue_congestion_threshold(struct request_queue *q)
99 nr = q->nr_requests - (q->nr_requests / 8) + 1;
100 if (nr > q->nr_requests)
102 q->nr_congestion_on = nr;
104 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
107 q->nr_congestion_off = nr;
111 * A queue has just exitted congestion. Note this in the global counter of
112 * congested queues, and wake up anyone who was waiting for requests to be
115 static void clear_queue_congested(request_queue_t *q, int rw)
118 wait_queue_head_t *wqh = &congestion_wqh[rw];
120 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
121 clear_bit(bit, &q->backing_dev_info.state);
122 smp_mb__after_clear_bit();
123 if (waitqueue_active(wqh))
128 * A queue has just entered congestion. Flag that in the queue's VM-visible
129 * state flags and increment the global gounter of congested queues.
131 static void set_queue_congested(request_queue_t *q, int rw)
135 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
136 set_bit(bit, &q->backing_dev_info.state);
140 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
143 * Locates the passed device's request queue and returns the address of its
146 * Will return NULL if the request queue cannot be located.
148 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
150 struct backing_dev_info *ret = NULL;
151 request_queue_t *q = bdev_get_queue(bdev);
154 ret = &q->backing_dev_info;
158 EXPORT_SYMBOL(blk_get_backing_dev_info);
160 void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data)
163 q->activity_data = data;
166 EXPORT_SYMBOL(blk_queue_activity_fn);
169 * blk_queue_prep_rq - set a prepare_request function for queue
171 * @pfn: prepare_request function
173 * It's possible for a queue to register a prepare_request callback which
174 * is invoked before the request is handed to the request_fn. The goal of
175 * the function is to prepare a request for I/O, it can be used to build a
176 * cdb from the request data for instance.
179 void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
184 EXPORT_SYMBOL(blk_queue_prep_rq);
187 * blk_queue_merge_bvec - set a merge_bvec function for queue
189 * @mbfn: merge_bvec_fn
191 * Usually queues have static limitations on the max sectors or segments that
192 * we can put in a request. Stacking drivers may have some settings that
193 * are dynamic, and thus we have to query the queue whether it is ok to
194 * add a new bio_vec to a bio at a given offset or not. If the block device
195 * has such limitations, it needs to register a merge_bvec_fn to control
196 * the size of bio's sent to it. Note that a block device *must* allow a
197 * single page to be added to an empty bio. The block device driver may want
198 * to use the bio_split() function to deal with these bio's. By default
199 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
202 void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
204 q->merge_bvec_fn = mbfn;
207 EXPORT_SYMBOL(blk_queue_merge_bvec);
210 * blk_queue_make_request - define an alternate make_request function for a device
211 * @q: the request queue for the device to be affected
212 * @mfn: the alternate make_request function
215 * The normal way for &struct bios to be passed to a device
216 * driver is for them to be collected into requests on a request
217 * queue, and then to allow the device driver to select requests
218 * off that queue when it is ready. This works well for many block
219 * devices. However some block devices (typically virtual devices
220 * such as md or lvm) do not benefit from the processing on the
221 * request queue, and are served best by having the requests passed
222 * directly to them. This can be achieved by providing a function
223 * to blk_queue_make_request().
226 * The driver that does this *must* be able to deal appropriately
227 * with buffers in "highmemory". This can be accomplished by either calling
228 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
229 * blk_queue_bounce() to create a buffer in normal memory.
231 void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
236 q->nr_requests = BLKDEV_MAX_RQ;
237 q->max_phys_segments = MAX_PHYS_SEGMENTS;
238 q->max_hw_segments = MAX_HW_SEGMENTS;
239 q->make_request_fn = mfn;
240 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
241 q->backing_dev_info.state = 0;
242 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
243 blk_queue_max_sectors(q, MAX_SECTORS);
244 blk_queue_hardsect_size(q, 512);
245 blk_queue_dma_alignment(q, 511);
246 blk_queue_congestion_threshold(q);
247 q->nr_batching = BLK_BATCH_REQ;
249 q->unplug_thresh = 4; /* hmm */
250 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
251 if (q->unplug_delay == 0)
254 INIT_WORK(&q->unplug_work, blk_unplug_work, q);
256 q->unplug_timer.function = blk_unplug_timeout;
257 q->unplug_timer.data = (unsigned long)q;
260 * by default assume old behaviour and bounce for any highmem page
262 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
264 blk_queue_activity_fn(q, NULL, NULL);
266 INIT_LIST_HEAD(&q->drain_list);
269 EXPORT_SYMBOL(blk_queue_make_request);
271 static inline void rq_init(request_queue_t *q, struct request *rq)
273 INIT_LIST_HEAD(&rq->queuelist);
276 rq->rq_status = RQ_ACTIVE;
277 rq->bio = rq->biotail = NULL;
287 rq->end_io_data = NULL;
291 * blk_queue_ordered - does this queue support ordered writes
292 * @q: the request queue
296 * For journalled file systems, doing ordered writes on a commit
297 * block instead of explicitly doing wait_on_buffer (which is bad
298 * for performance) can be a big win. Block drivers supporting this
299 * feature should call this function and indicate so.
302 void blk_queue_ordered(request_queue_t *q, int flag)
305 case QUEUE_ORDERED_NONE:
307 kmem_cache_free(request_cachep, q->flush_rq);
311 case QUEUE_ORDERED_TAG:
314 case QUEUE_ORDERED_FLUSH:
317 q->flush_rq = kmem_cache_alloc(request_cachep,
321 printk("blk_queue_ordered: bad value %d\n", flag);
326 EXPORT_SYMBOL(blk_queue_ordered);
329 * blk_queue_issue_flush_fn - set function for issuing a flush
330 * @q: the request queue
331 * @iff: the function to be called issuing the flush
334 * If a driver supports issuing a flush command, the support is notified
335 * to the block layer by defining it through this call.
338 void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff)
340 q->issue_flush_fn = iff;
343 EXPORT_SYMBOL(blk_queue_issue_flush_fn);
346 * Cache flushing for ordered writes handling
348 static void blk_pre_flush_end_io(struct request *flush_rq)
350 struct request *rq = flush_rq->end_io_data;
351 request_queue_t *q = rq->q;
353 rq->flags |= REQ_BAR_PREFLUSH;
355 if (!flush_rq->errors)
356 elv_requeue_request(q, rq);
358 q->end_flush_fn(q, flush_rq);
359 clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags);
364 static void blk_post_flush_end_io(struct request *flush_rq)
366 struct request *rq = flush_rq->end_io_data;
367 request_queue_t *q = rq->q;
369 rq->flags |= REQ_BAR_POSTFLUSH;
371 q->end_flush_fn(q, flush_rq);
372 clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags);
376 struct request *blk_start_pre_flush(request_queue_t *q, struct request *rq)
378 struct request *flush_rq = q->flush_rq;
380 BUG_ON(!blk_barrier_rq(rq));
382 if (test_and_set_bit(QUEUE_FLAG_FLUSH, &q->queue_flags))
385 rq_init(q, flush_rq);
386 flush_rq->elevator_private = NULL;
387 flush_rq->flags = REQ_BAR_FLUSH;
388 flush_rq->rq_disk = rq->rq_disk;
392 * prepare_flush returns 0 if no flush is needed, just mark both
393 * pre and post flush as done in that case
395 if (!q->prepare_flush_fn(q, flush_rq)) {
396 rq->flags |= REQ_BAR_PREFLUSH | REQ_BAR_POSTFLUSH;
397 clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags);
402 * some drivers dequeue requests right away, some only after io
403 * completion. make sure the request is dequeued.
405 if (!list_empty(&rq->queuelist))
406 blkdev_dequeue_request(rq);
408 elv_deactivate_request(q, rq);
410 flush_rq->end_io_data = rq;
411 flush_rq->end_io = blk_pre_flush_end_io;
413 __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0);
417 static void blk_start_post_flush(request_queue_t *q, struct request *rq)
419 struct request *flush_rq = q->flush_rq;
421 BUG_ON(!blk_barrier_rq(rq));
423 rq_init(q, flush_rq);
424 flush_rq->elevator_private = NULL;
425 flush_rq->flags = REQ_BAR_FLUSH;
426 flush_rq->rq_disk = rq->rq_disk;
429 if (q->prepare_flush_fn(q, flush_rq)) {
430 flush_rq->end_io_data = rq;
431 flush_rq->end_io = blk_post_flush_end_io;
433 __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0);
438 static inline int blk_check_end_barrier(request_queue_t *q, struct request *rq,
441 if (sectors > rq->nr_sectors)
442 sectors = rq->nr_sectors;
444 rq->nr_sectors -= sectors;
445 return rq->nr_sectors;
448 static int __blk_complete_barrier_rq(request_queue_t *q, struct request *rq,
449 int sectors, int queue_locked)
451 if (q->ordered != QUEUE_ORDERED_FLUSH)
453 if (!blk_fs_request(rq) || !blk_barrier_rq(rq))
455 if (blk_barrier_postflush(rq))
458 if (!blk_check_end_barrier(q, rq, sectors)) {
459 unsigned long flags = 0;
462 spin_lock_irqsave(q->queue_lock, flags);
464 blk_start_post_flush(q, rq);
467 spin_unlock_irqrestore(q->queue_lock, flags);
474 * blk_complete_barrier_rq - complete possible barrier request
475 * @q: the request queue for the device
477 * @sectors: number of sectors to complete
480 * Used in driver end_io handling to determine whether to postpone
481 * completion of a barrier request until a post flush has been done. This
482 * is the unlocked variant, used if the caller doesn't already hold the
485 int blk_complete_barrier_rq(request_queue_t *q, struct request *rq, int sectors)
487 return __blk_complete_barrier_rq(q, rq, sectors, 0);
489 EXPORT_SYMBOL(blk_complete_barrier_rq);
492 * blk_complete_barrier_rq_locked - complete possible barrier request
493 * @q: the request queue for the device
495 * @sectors: number of sectors to complete
498 * See blk_complete_barrier_rq(). This variant must be used if the caller
499 * holds the queue lock.
501 int blk_complete_barrier_rq_locked(request_queue_t *q, struct request *rq,
504 return __blk_complete_barrier_rq(q, rq, sectors, 1);
506 EXPORT_SYMBOL(blk_complete_barrier_rq_locked);
509 * blk_queue_bounce_limit - set bounce buffer limit for queue
510 * @q: the request queue for the device
511 * @dma_addr: bus address limit
514 * Different hardware can have different requirements as to what pages
515 * it can do I/O directly to. A low level driver can call
516 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
517 * buffers for doing I/O to pages residing above @page. By default
518 * the block layer sets this to the highest numbered "low" memory page.
520 void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
522 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
525 * set appropriate bounce gfp mask -- unfortunately we don't have a
526 * full 4GB zone, so we have to resort to low memory for any bounces.
527 * ISA has its own < 16MB zone.
529 if (bounce_pfn < blk_max_low_pfn) {
530 BUG_ON(dma_addr < BLK_BOUNCE_ISA);
531 init_emergency_isa_pool();
532 q->bounce_gfp = GFP_NOIO | GFP_DMA;
534 q->bounce_gfp = GFP_NOIO;
536 q->bounce_pfn = bounce_pfn;
539 EXPORT_SYMBOL(blk_queue_bounce_limit);
542 * blk_queue_max_sectors - set max sectors for a request for this queue
543 * @q: the request queue for the device
544 * @max_sectors: max sectors in the usual 512b unit
547 * Enables a low level driver to set an upper limit on the size of
550 void blk_queue_max_sectors(request_queue_t *q, unsigned short max_sectors)
552 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
553 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
554 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
557 q->max_sectors = q->max_hw_sectors = max_sectors;
560 EXPORT_SYMBOL(blk_queue_max_sectors);
563 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
564 * @q: the request queue for the device
565 * @max_segments: max number of segments
568 * Enables a low level driver to set an upper limit on the number of
569 * physical data segments in a request. This would be the largest sized
570 * scatter list the driver could handle.
572 void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
576 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
579 q->max_phys_segments = max_segments;
582 EXPORT_SYMBOL(blk_queue_max_phys_segments);
585 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
586 * @q: the request queue for the device
587 * @max_segments: max number of segments
590 * Enables a low level driver to set an upper limit on the number of
591 * hw data segments in a request. This would be the largest number of
592 * address/length pairs the host adapter can actually give as once
595 void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
599 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
602 q->max_hw_segments = max_segments;
605 EXPORT_SYMBOL(blk_queue_max_hw_segments);
608 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
609 * @q: the request queue for the device
610 * @max_size: max size of segment in bytes
613 * Enables a low level driver to set an upper limit on the size of a
616 void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
618 if (max_size < PAGE_CACHE_SIZE) {
619 max_size = PAGE_CACHE_SIZE;
620 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
623 q->max_segment_size = max_size;
626 EXPORT_SYMBOL(blk_queue_max_segment_size);
629 * blk_queue_hardsect_size - set hardware sector size for the queue
630 * @q: the request queue for the device
631 * @size: the hardware sector size, in bytes
634 * This should typically be set to the lowest possible sector size
635 * that the hardware can operate on (possible without reverting to
636 * even internal read-modify-write operations). Usually the default
637 * of 512 covers most hardware.
639 void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
641 q->hardsect_size = size;
644 EXPORT_SYMBOL(blk_queue_hardsect_size);
647 * Returns the minimum that is _not_ zero, unless both are zero.
649 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
652 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
653 * @t: the stacking driver (top)
654 * @b: the underlying device (bottom)
656 void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
658 /* zero is "infinity" */
659 t->max_sectors = t->max_hw_sectors =
660 min_not_zero(t->max_sectors,b->max_sectors);
662 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
663 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
664 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
665 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
668 EXPORT_SYMBOL(blk_queue_stack_limits);
671 * blk_queue_segment_boundary - set boundary rules for segment merging
672 * @q: the request queue for the device
673 * @mask: the memory boundary mask
675 void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
677 if (mask < PAGE_CACHE_SIZE - 1) {
678 mask = PAGE_CACHE_SIZE - 1;
679 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
682 q->seg_boundary_mask = mask;
685 EXPORT_SYMBOL(blk_queue_segment_boundary);
688 * blk_queue_dma_alignment - set dma length and memory alignment
689 * @q: the request queue for the device
690 * @mask: alignment mask
693 * set required memory and length aligment for direct dma transactions.
694 * this is used when buiding direct io requests for the queue.
697 void blk_queue_dma_alignment(request_queue_t *q, int mask)
699 q->dma_alignment = mask;
702 EXPORT_SYMBOL(blk_queue_dma_alignment);
705 * blk_queue_find_tag - find a request by its tag and queue
707 * @q: The request queue for the device
708 * @tag: The tag of the request
711 * Should be used when a device returns a tag and you want to match
714 * no locks need be held.
716 struct request *blk_queue_find_tag(request_queue_t *q, int tag)
718 struct blk_queue_tag *bqt = q->queue_tags;
720 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
723 return bqt->tag_index[tag];
726 EXPORT_SYMBOL(blk_queue_find_tag);
729 * __blk_queue_free_tags - release tag maintenance info
730 * @q: the request queue for the device
733 * blk_cleanup_queue() will take care of calling this function, if tagging
734 * has been used. So there's no need to call this directly.
736 static void __blk_queue_free_tags(request_queue_t *q)
738 struct blk_queue_tag *bqt = q->queue_tags;
743 if (atomic_dec_and_test(&bqt->refcnt)) {
745 BUG_ON(!list_empty(&bqt->busy_list));
747 kfree(bqt->tag_index);
748 bqt->tag_index = NULL;
756 q->queue_tags = NULL;
757 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
761 * blk_queue_free_tags - release tag maintenance info
762 * @q: the request queue for the device
765 * This is used to disabled tagged queuing to a device, yet leave
768 void blk_queue_free_tags(request_queue_t *q)
770 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
773 EXPORT_SYMBOL(blk_queue_free_tags);
776 init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
779 struct request **tag_index;
780 unsigned long *tag_map;
782 if (depth > q->nr_requests * 2) {
783 depth = q->nr_requests * 2;
784 printk(KERN_ERR "%s: adjusted depth to %d\n",
785 __FUNCTION__, depth);
788 tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC);
792 bits = (depth / BLK_TAGS_PER_LONG) + 1;
793 tag_map = kmalloc(bits * sizeof(unsigned long), GFP_ATOMIC);
797 memset(tag_index, 0, depth * sizeof(struct request *));
798 memset(tag_map, 0, bits * sizeof(unsigned long));
799 tags->max_depth = depth;
800 tags->real_max_depth = bits * BITS_PER_LONG;
801 tags->tag_index = tag_index;
802 tags->tag_map = tag_map;
805 * set the upper bits if the depth isn't a multiple of the word size
807 for (i = depth; i < bits * BLK_TAGS_PER_LONG; i++)
808 __set_bit(i, tag_map);
817 * blk_queue_init_tags - initialize the queue tag info
818 * @q: the request queue for the device
819 * @depth: the maximum queue depth supported
820 * @tags: the tag to use
822 int blk_queue_init_tags(request_queue_t *q, int depth,
823 struct blk_queue_tag *tags)
827 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
829 if (!tags && !q->queue_tags) {
830 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
834 if (init_tag_map(q, tags, depth))
837 INIT_LIST_HEAD(&tags->busy_list);
839 atomic_set(&tags->refcnt, 1);
840 } else if (q->queue_tags) {
841 if ((rc = blk_queue_resize_tags(q, depth)))
843 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
846 atomic_inc(&tags->refcnt);
849 * assign it, all done
851 q->queue_tags = tags;
852 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
859 EXPORT_SYMBOL(blk_queue_init_tags);
862 * blk_queue_resize_tags - change the queueing depth
863 * @q: the request queue for the device
864 * @new_depth: the new max command queueing depth
867 * Must be called with the queue lock held.
869 int blk_queue_resize_tags(request_queue_t *q, int new_depth)
871 struct blk_queue_tag *bqt = q->queue_tags;
872 struct request **tag_index;
873 unsigned long *tag_map;
880 * don't bother sizing down
882 if (new_depth <= bqt->real_max_depth) {
883 bqt->max_depth = new_depth;
888 * save the old state info, so we can copy it back
890 tag_index = bqt->tag_index;
891 tag_map = bqt->tag_map;
892 max_depth = bqt->real_max_depth;
894 if (init_tag_map(q, bqt, new_depth))
897 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
898 bits = max_depth / BLK_TAGS_PER_LONG;
899 memcpy(bqt->tag_map, tag_map, bits * sizeof(unsigned long));
906 EXPORT_SYMBOL(blk_queue_resize_tags);
909 * blk_queue_end_tag - end tag operations for a request
910 * @q: the request queue for the device
911 * @rq: the request that has completed
914 * Typically called when end_that_request_first() returns 0, meaning
915 * all transfers have been done for a request. It's important to call
916 * this function before end_that_request_last(), as that will put the
917 * request back on the free list thus corrupting the internal tag list.
920 * queue lock must be held.
922 void blk_queue_end_tag(request_queue_t *q, struct request *rq)
924 struct blk_queue_tag *bqt = q->queue_tags;
929 if (unlikely(tag >= bqt->real_max_depth))
932 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
933 printk("attempt to clear non-busy tag (%d)\n", tag);
937 list_del_init(&rq->queuelist);
938 rq->flags &= ~REQ_QUEUED;
941 if (unlikely(bqt->tag_index[tag] == NULL))
942 printk("tag %d is missing\n", tag);
944 bqt->tag_index[tag] = NULL;
948 EXPORT_SYMBOL(blk_queue_end_tag);
951 * blk_queue_start_tag - find a free tag and assign it
952 * @q: the request queue for the device
953 * @rq: the block request that needs tagging
956 * This can either be used as a stand-alone helper, or possibly be
957 * assigned as the queue &prep_rq_fn (in which case &struct request
958 * automagically gets a tag assigned). Note that this function
959 * assumes that any type of request can be queued! if this is not
960 * true for your device, you must check the request type before
961 * calling this function. The request will also be removed from
962 * the request queue, so it's the drivers responsibility to readd
963 * it if it should need to be restarted for some reason.
966 * queue lock must be held.
968 int blk_queue_start_tag(request_queue_t *q, struct request *rq)
970 struct blk_queue_tag *bqt = q->queue_tags;
971 unsigned long *map = bqt->tag_map;
974 if (unlikely((rq->flags & REQ_QUEUED))) {
976 "request %p for device [%s] already tagged %d",
977 rq, rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
981 for (map = bqt->tag_map; *map == -1UL; map++) {
982 tag += BLK_TAGS_PER_LONG;
984 if (tag >= bqt->max_depth)
989 __set_bit(tag, bqt->tag_map);
991 rq->flags |= REQ_QUEUED;
993 bqt->tag_index[tag] = rq;
994 blkdev_dequeue_request(rq);
995 list_add(&rq->queuelist, &bqt->busy_list);
1000 EXPORT_SYMBOL(blk_queue_start_tag);
1003 * blk_queue_invalidate_tags - invalidate all pending tags
1004 * @q: the request queue for the device
1007 * Hardware conditions may dictate a need to stop all pending requests.
1008 * In this case, we will safely clear the block side of the tag queue and
1009 * readd all requests to the request queue in the right order.
1012 * queue lock must be held.
1014 void blk_queue_invalidate_tags(request_queue_t *q)
1016 struct blk_queue_tag *bqt = q->queue_tags;
1017 struct list_head *tmp, *n;
1020 list_for_each_safe(tmp, n, &bqt->busy_list) {
1021 rq = list_entry_rq(tmp);
1023 if (rq->tag == -1) {
1024 printk("bad tag found on list\n");
1025 list_del_init(&rq->queuelist);
1026 rq->flags &= ~REQ_QUEUED;
1028 blk_queue_end_tag(q, rq);
1030 rq->flags &= ~REQ_STARTED;
1031 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1035 EXPORT_SYMBOL(blk_queue_invalidate_tags);
1037 static char *rq_flags[] = {
1055 "REQ_DRIVE_TASKFILE",
1062 void blk_dump_rq_flags(struct request *rq, char *msg)
1066 printk("%s: dev %s: flags = ", msg,
1067 rq->rq_disk ? rq->rq_disk->disk_name : "?");
1070 if (rq->flags & (1 << bit))
1071 printk("%s ", rq_flags[bit]);
1073 } while (bit < __REQ_NR_BITS);
1075 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1077 rq->current_nr_sectors);
1078 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1080 if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) {
1082 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1083 printk("%02x ", rq->cmd[bit]);
1088 EXPORT_SYMBOL(blk_dump_rq_flags);
1090 void blk_recount_segments(request_queue_t *q, struct bio *bio)
1092 struct bio_vec *bv, *bvprv = NULL;
1093 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1094 int high, highprv = 1;
1096 if (unlikely(!bio->bi_io_vec))
1099 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1100 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1101 bio_for_each_segment(bv, bio, i) {
1103 * the trick here is making sure that a high page is never
1104 * considered part of another segment, since that might
1105 * change with the bounce page.
1107 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
1108 if (high || highprv)
1109 goto new_hw_segment;
1111 if (seg_size + bv->bv_len > q->max_segment_size)
1113 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1115 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1117 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1118 goto new_hw_segment;
1120 seg_size += bv->bv_len;
1121 hw_seg_size += bv->bv_len;
1126 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1127 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1128 hw_seg_size += bv->bv_len;
1131 if (hw_seg_size > bio->bi_hw_front_size)
1132 bio->bi_hw_front_size = hw_seg_size;
1133 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1139 seg_size = bv->bv_len;
1142 if (hw_seg_size > bio->bi_hw_back_size)
1143 bio->bi_hw_back_size = hw_seg_size;
1144 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1145 bio->bi_hw_front_size = hw_seg_size;
1146 bio->bi_phys_segments = nr_phys_segs;
1147 bio->bi_hw_segments = nr_hw_segs;
1148 bio->bi_flags |= (1 << BIO_SEG_VALID);
1152 int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
1155 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1158 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1160 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1164 * bio and nxt are contigous in memory, check if the queue allows
1165 * these two to be merged into one
1167 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1173 EXPORT_SYMBOL(blk_phys_contig_segment);
1175 int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
1178 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1179 blk_recount_segments(q, bio);
1180 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1181 blk_recount_segments(q, nxt);
1182 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1183 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1185 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1191 EXPORT_SYMBOL(blk_hw_contig_segment);
1194 * map a request to scatterlist, return number of sg entries setup. Caller
1195 * must make sure sg can hold rq->nr_phys_segments entries
1197 int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1199 struct bio_vec *bvec, *bvprv;
1201 int nsegs, i, cluster;
1204 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1207 * for each bio in rq
1210 rq_for_each_bio(bio, rq) {
1212 * for each segment in bio
1214 bio_for_each_segment(bvec, bio, i) {
1215 int nbytes = bvec->bv_len;
1217 if (bvprv && cluster) {
1218 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1221 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1223 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1226 sg[nsegs - 1].length += nbytes;
1229 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1230 sg[nsegs].page = bvec->bv_page;
1231 sg[nsegs].length = nbytes;
1232 sg[nsegs].offset = bvec->bv_offset;
1237 } /* segments in bio */
1243 EXPORT_SYMBOL(blk_rq_map_sg);
1246 * the standard queue merge functions, can be overridden with device
1247 * specific ones if so desired
1250 static inline int ll_new_mergeable(request_queue_t *q,
1251 struct request *req,
1254 int nr_phys_segs = bio_phys_segments(q, bio);
1256 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1257 req->flags |= REQ_NOMERGE;
1258 if (req == q->last_merge)
1259 q->last_merge = NULL;
1264 * A hw segment is just getting larger, bump just the phys
1267 req->nr_phys_segments += nr_phys_segs;
1271 static inline int ll_new_hw_segment(request_queue_t *q,
1272 struct request *req,
1275 int nr_hw_segs = bio_hw_segments(q, bio);
1276 int nr_phys_segs = bio_phys_segments(q, bio);
1278 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1279 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1280 req->flags |= REQ_NOMERGE;
1281 if (req == q->last_merge)
1282 q->last_merge = NULL;
1287 * This will form the start of a new hw segment. Bump both
1290 req->nr_hw_segments += nr_hw_segs;
1291 req->nr_phys_segments += nr_phys_segs;
1295 static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1300 if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) {
1301 req->flags |= REQ_NOMERGE;
1302 if (req == q->last_merge)
1303 q->last_merge = NULL;
1306 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1307 blk_recount_segments(q, req->biotail);
1308 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1309 blk_recount_segments(q, bio);
1310 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1311 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1312 !BIOVEC_VIRT_OVERSIZE(len)) {
1313 int mergeable = ll_new_mergeable(q, req, bio);
1316 if (req->nr_hw_segments == 1)
1317 req->bio->bi_hw_front_size = len;
1318 if (bio->bi_hw_segments == 1)
1319 bio->bi_hw_back_size = len;
1324 return ll_new_hw_segment(q, req, bio);
1327 static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1332 if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) {
1333 req->flags |= REQ_NOMERGE;
1334 if (req == q->last_merge)
1335 q->last_merge = NULL;
1338 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1339 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1340 blk_recount_segments(q, bio);
1341 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1342 blk_recount_segments(q, req->bio);
1343 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1344 !BIOVEC_VIRT_OVERSIZE(len)) {
1345 int mergeable = ll_new_mergeable(q, req, bio);
1348 if (bio->bi_hw_segments == 1)
1349 bio->bi_hw_front_size = len;
1350 if (req->nr_hw_segments == 1)
1351 req->biotail->bi_hw_back_size = len;
1356 return ll_new_hw_segment(q, req, bio);
1359 static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1360 struct request *next)
1362 int total_phys_segments = req->nr_phys_segments +next->nr_phys_segments;
1363 int total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1366 * First check if the either of the requests are re-queued
1367 * requests. Can't merge them if they are.
1369 if (req->special || next->special)
1373 * Will it become to large?
1375 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1378 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1379 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1380 total_phys_segments--;
1382 if (total_phys_segments > q->max_phys_segments)
1385 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1386 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1387 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1389 * propagate the combined length to the end of the requests
1391 if (req->nr_hw_segments == 1)
1392 req->bio->bi_hw_front_size = len;
1393 if (next->nr_hw_segments == 1)
1394 next->biotail->bi_hw_back_size = len;
1395 total_hw_segments--;
1398 if (total_hw_segments > q->max_hw_segments)
1401 /* Merge is OK... */
1402 req->nr_phys_segments = total_phys_segments;
1403 req->nr_hw_segments = total_hw_segments;
1408 * "plug" the device if there are no outstanding requests: this will
1409 * force the transfer to start only after we have put all the requests
1412 * This is called with interrupts off and no requests on the queue and
1413 * with the queue lock held.
1415 void blk_plug_device(request_queue_t *q)
1417 WARN_ON(!irqs_disabled());
1420 * don't plug a stopped queue, it must be paired with blk_start_queue()
1421 * which will restart the queueing
1423 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1426 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1427 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
1430 EXPORT_SYMBOL(blk_plug_device);
1433 * remove the queue from the plugged list, if present. called with
1434 * queue lock held and interrupts disabled.
1436 int blk_remove_plug(request_queue_t *q)
1438 WARN_ON(!irqs_disabled());
1440 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1443 del_timer(&q->unplug_timer);
1447 EXPORT_SYMBOL(blk_remove_plug);
1450 * remove the plug and let it rip..
1452 void __generic_unplug_device(request_queue_t *q)
1454 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1457 if (!blk_remove_plug(q))
1461 * was plugged, fire request_fn if queue has stuff to do
1463 if (elv_next_request(q))
1466 EXPORT_SYMBOL(__generic_unplug_device);
1469 * generic_unplug_device - fire a request queue
1470 * @q: The &request_queue_t in question
1473 * Linux uses plugging to build bigger requests queues before letting
1474 * the device have at them. If a queue is plugged, the I/O scheduler
1475 * is still adding and merging requests on the queue. Once the queue
1476 * gets unplugged, the request_fn defined for the queue is invoked and
1477 * transfers started.
1479 void generic_unplug_device(request_queue_t *q)
1481 spin_lock_irq(q->queue_lock);
1482 __generic_unplug_device(q);
1483 spin_unlock_irq(q->queue_lock);
1485 EXPORT_SYMBOL(generic_unplug_device);
1487 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1490 request_queue_t *q = bdi->unplug_io_data;
1493 * devices don't necessarily have an ->unplug_fn defined
1499 static void blk_unplug_work(void *data)
1501 request_queue_t *q = data;
1506 static void blk_unplug_timeout(unsigned long data)
1508 request_queue_t *q = (request_queue_t *)data;
1510 kblockd_schedule_work(&q->unplug_work);
1514 * blk_start_queue - restart a previously stopped queue
1515 * @q: The &request_queue_t in question
1518 * blk_start_queue() will clear the stop flag on the queue, and call
1519 * the request_fn for the queue if it was in a stopped state when
1520 * entered. Also see blk_stop_queue(). Queue lock must be held.
1522 void blk_start_queue(request_queue_t *q)
1524 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1527 * one level of recursion is ok and is much faster than kicking
1528 * the unplug handling
1530 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1532 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1535 kblockd_schedule_work(&q->unplug_work);
1539 EXPORT_SYMBOL(blk_start_queue);
1542 * blk_stop_queue - stop a queue
1543 * @q: The &request_queue_t in question
1546 * The Linux block layer assumes that a block driver will consume all
1547 * entries on the request queue when the request_fn strategy is called.
1548 * Often this will not happen, because of hardware limitations (queue
1549 * depth settings). If a device driver gets a 'queue full' response,
1550 * or if it simply chooses not to queue more I/O at one point, it can
1551 * call this function to prevent the request_fn from being called until
1552 * the driver has signalled it's ready to go again. This happens by calling
1553 * blk_start_queue() to restart queue operations. Queue lock must be held.
1555 void blk_stop_queue(request_queue_t *q)
1558 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1560 EXPORT_SYMBOL(blk_stop_queue);
1563 * blk_sync_queue - cancel any pending callbacks on a queue
1567 * The block layer may perform asynchronous callback activity
1568 * on a queue, such as calling the unplug function after a timeout.
1569 * A block device may call blk_sync_queue to ensure that any
1570 * such activity is cancelled, thus allowing it to release resources
1571 * the the callbacks might use. The caller must already have made sure
1572 * that its ->make_request_fn will not re-add plugging prior to calling
1576 void blk_sync_queue(struct request_queue *q)
1578 del_timer_sync(&q->unplug_timer);
1581 EXPORT_SYMBOL(blk_sync_queue);
1584 * blk_run_queue - run a single device queue
1585 * @q: The queue to run
1587 void blk_run_queue(struct request_queue *q)
1589 unsigned long flags;
1591 spin_lock_irqsave(q->queue_lock, flags);
1593 if (!elv_queue_empty(q))
1595 spin_unlock_irqrestore(q->queue_lock, flags);
1597 EXPORT_SYMBOL(blk_run_queue);
1600 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1601 * @q: the request queue to be released
1604 * blk_cleanup_queue is the pair to blk_init_queue() or
1605 * blk_queue_make_request(). It should be called when a request queue is
1606 * being released; typically when a block device is being de-registered.
1607 * Currently, its primary task it to free all the &struct request
1608 * structures that were allocated to the queue and the queue itself.
1611 * Hopefully the low level driver will have finished any
1612 * outstanding requests first...
1614 void blk_cleanup_queue(request_queue_t * q)
1616 struct request_list *rl = &q->rq;
1618 if (!atomic_dec_and_test(&q->refcnt))
1622 elevator_exit(q->elevator);
1627 mempool_destroy(rl->rq_pool);
1630 __blk_queue_free_tags(q);
1632 blk_queue_ordered(q, QUEUE_ORDERED_NONE);
1634 kmem_cache_free(requestq_cachep, q);
1637 EXPORT_SYMBOL(blk_cleanup_queue);
1639 static int blk_init_free_list(request_queue_t *q)
1641 struct request_list *rl = &q->rq;
1643 rl->count[READ] = rl->count[WRITE] = 0;
1644 rl->starved[READ] = rl->starved[WRITE] = 0;
1645 init_waitqueue_head(&rl->wait[READ]);
1646 init_waitqueue_head(&rl->wait[WRITE]);
1647 init_waitqueue_head(&rl->drain);
1649 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1650 mempool_free_slab, request_cachep, q->node);
1658 static int __make_request(request_queue_t *, struct bio *);
1660 request_queue_t *blk_alloc_queue(int gfp_mask)
1662 return blk_alloc_queue_node(gfp_mask, -1);
1664 EXPORT_SYMBOL(blk_alloc_queue);
1666 request_queue_t *blk_alloc_queue_node(int gfp_mask, int node_id)
1670 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
1674 memset(q, 0, sizeof(*q));
1675 init_timer(&q->unplug_timer);
1676 atomic_set(&q->refcnt, 1);
1678 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1679 q->backing_dev_info.unplug_io_data = q;
1683 EXPORT_SYMBOL(blk_alloc_queue_node);
1686 * blk_init_queue - prepare a request queue for use with a block device
1687 * @rfn: The function to be called to process requests that have been
1688 * placed on the queue.
1689 * @lock: Request queue spin lock
1692 * If a block device wishes to use the standard request handling procedures,
1693 * which sorts requests and coalesces adjacent requests, then it must
1694 * call blk_init_queue(). The function @rfn will be called when there
1695 * are requests on the queue that need to be processed. If the device
1696 * supports plugging, then @rfn may not be called immediately when requests
1697 * are available on the queue, but may be called at some time later instead.
1698 * Plugged queues are generally unplugged when a buffer belonging to one
1699 * of the requests on the queue is needed, or due to memory pressure.
1701 * @rfn is not required, or even expected, to remove all requests off the
1702 * queue, but only as many as it can handle at a time. If it does leave
1703 * requests on the queue, it is responsible for arranging that the requests
1704 * get dealt with eventually.
1706 * The queue spin lock must be held while manipulating the requests on the
1709 * Function returns a pointer to the initialized request queue, or NULL if
1710 * it didn't succeed.
1713 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1714 * when the block device is deactivated (such as at module unload).
1717 request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1719 return blk_init_queue_node(rfn, lock, -1);
1721 EXPORT_SYMBOL(blk_init_queue);
1724 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1726 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
1732 if (blk_init_free_list(q))
1736 * if caller didn't supply a lock, they get per-queue locking with
1740 spin_lock_init(&q->__queue_lock);
1741 lock = &q->__queue_lock;
1744 q->request_fn = rfn;
1745 q->back_merge_fn = ll_back_merge_fn;
1746 q->front_merge_fn = ll_front_merge_fn;
1747 q->merge_requests_fn = ll_merge_requests_fn;
1748 q->prep_rq_fn = NULL;
1749 q->unplug_fn = generic_unplug_device;
1750 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1751 q->queue_lock = lock;
1753 blk_queue_segment_boundary(q, 0xffffffff);
1755 blk_queue_make_request(q, __make_request);
1756 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1758 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1759 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1764 if (!elevator_init(q, NULL)) {
1765 blk_queue_congestion_threshold(q);
1769 blk_cleanup_queue(q);
1771 kmem_cache_free(requestq_cachep, q);
1774 EXPORT_SYMBOL(blk_init_queue_node);
1776 int blk_get_queue(request_queue_t *q)
1778 if (!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
1779 atomic_inc(&q->refcnt);
1786 EXPORT_SYMBOL(blk_get_queue);
1788 static inline void blk_free_request(request_queue_t *q, struct request *rq)
1790 elv_put_request(q, rq);
1791 mempool_free(rq, q->rq.rq_pool);
1794 static inline struct request *blk_alloc_request(request_queue_t *q, int rw,
1797 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1803 * first three bits are identical in rq->flags and bio->bi_rw,
1804 * see bio.h and blkdev.h
1808 if (!elv_set_request(q, rq, gfp_mask))
1811 mempool_free(rq, q->rq.rq_pool);
1816 * ioc_batching returns true if the ioc is a valid batching request and
1817 * should be given priority access to a request.
1819 static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
1825 * Make sure the process is able to allocate at least 1 request
1826 * even if the batch times out, otherwise we could theoretically
1829 return ioc->nr_batch_requests == q->nr_batching ||
1830 (ioc->nr_batch_requests > 0
1831 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1835 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1836 * will cause the process to be a "batcher" on all queues in the system. This
1837 * is the behaviour we want though - once it gets a wakeup it should be given
1840 void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
1842 if (!ioc || ioc_batching(q, ioc))
1845 ioc->nr_batch_requests = q->nr_batching;
1846 ioc->last_waited = jiffies;
1849 static void __freed_request(request_queue_t *q, int rw)
1851 struct request_list *rl = &q->rq;
1853 if (rl->count[rw] < queue_congestion_off_threshold(q))
1854 clear_queue_congested(q, rw);
1856 if (rl->count[rw] + 1 <= q->nr_requests) {
1858 if (waitqueue_active(&rl->wait[rw]))
1859 wake_up(&rl->wait[rw]);
1861 blk_clear_queue_full(q, rw);
1866 * A request has just been released. Account for it, update the full and
1867 * congestion status, wake up any waiters. Called under q->queue_lock.
1869 static void freed_request(request_queue_t *q, int rw)
1871 struct request_list *rl = &q->rq;
1875 __freed_request(q, rw);
1877 if (unlikely(rl->starved[rw ^ 1]))
1878 __freed_request(q, rw ^ 1);
1880 if (!rl->count[READ] && !rl->count[WRITE]) {
1882 if (unlikely(waitqueue_active(&rl->drain)))
1883 wake_up(&rl->drain);
1887 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1889 * Get a free request, queue_lock must not be held
1891 static struct request *get_request(request_queue_t *q, int rw, int gfp_mask)
1893 struct request *rq = NULL;
1894 struct request_list *rl = &q->rq;
1895 struct io_context *ioc = get_io_context(gfp_mask);
1897 if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)))
1900 spin_lock_irq(q->queue_lock);
1901 if (rl->count[rw]+1 >= q->nr_requests) {
1903 * The queue will fill after this allocation, so set it as
1904 * full, and mark this process as "batching". This process
1905 * will be allowed to complete a batch of requests, others
1908 if (!blk_queue_full(q, rw)) {
1909 ioc_set_batching(q, ioc);
1910 blk_set_queue_full(q, rw);
1914 switch (elv_may_queue(q, rw)) {
1917 case ELV_MQUEUE_MAY:
1919 case ELV_MQUEUE_MUST:
1923 if (blk_queue_full(q, rw) && !ioc_batching(q, ioc)) {
1925 * The queue is full and the allocating process is not a
1926 * "batcher", and not exempted by the IO scheduler
1928 spin_unlock_irq(q->queue_lock);
1934 rl->starved[rw] = 0;
1935 if (rl->count[rw] >= queue_congestion_on_threshold(q))
1936 set_queue_congested(q, rw);
1937 spin_unlock_irq(q->queue_lock);
1939 rq = blk_alloc_request(q, rw, gfp_mask);
1942 * Allocation failed presumably due to memory. Undo anything
1943 * we might have messed up.
1945 * Allocating task should really be put onto the front of the
1946 * wait queue, but this is pretty rare.
1948 spin_lock_irq(q->queue_lock);
1949 freed_request(q, rw);
1952 * in the very unlikely event that allocation failed and no
1953 * requests for this direction was pending, mark us starved
1954 * so that freeing of a request in the other direction will
1955 * notice us. another possible fix would be to split the
1956 * rq mempool into READ and WRITE
1959 if (unlikely(rl->count[rw] == 0))
1960 rl->starved[rw] = 1;
1962 spin_unlock_irq(q->queue_lock);
1966 if (ioc_batching(q, ioc))
1967 ioc->nr_batch_requests--;
1972 put_io_context(ioc);
1977 * No available requests for this queue, unplug the device and wait for some
1978 * requests to become available.
1980 static struct request *get_request_wait(request_queue_t *q, int rw)
1985 generic_unplug_device(q);
1987 struct request_list *rl = &q->rq;
1989 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
1990 TASK_UNINTERRUPTIBLE);
1992 rq = get_request(q, rw, GFP_NOIO);
1995 struct io_context *ioc;
2000 * After sleeping, we become a "batching" process and
2001 * will be able to allocate at least one request, and
2002 * up to a big batch of them for a small period time.
2003 * See ioc_batching, ioc_set_batching
2005 ioc = get_io_context(GFP_NOIO);
2006 ioc_set_batching(q, ioc);
2007 put_io_context(ioc);
2009 finish_wait(&rl->wait[rw], &wait);
2015 struct request *blk_get_request(request_queue_t *q, int rw, int gfp_mask)
2019 BUG_ON(rw != READ && rw != WRITE);
2021 if (gfp_mask & __GFP_WAIT)
2022 rq = get_request_wait(q, rw);
2024 rq = get_request(q, rw, gfp_mask);
2029 EXPORT_SYMBOL(blk_get_request);
2032 * blk_requeue_request - put a request back on queue
2033 * @q: request queue where request should be inserted
2034 * @rq: request to be inserted
2037 * Drivers often keep queueing requests until the hardware cannot accept
2038 * more, when that condition happens we need to put the request back
2039 * on the queue. Must be called with queue lock held.
2041 void blk_requeue_request(request_queue_t *q, struct request *rq)
2043 if (blk_rq_tagged(rq))
2044 blk_queue_end_tag(q, rq);
2046 elv_requeue_request(q, rq);
2049 EXPORT_SYMBOL(blk_requeue_request);
2052 * blk_insert_request - insert a special request in to a request queue
2053 * @q: request queue where request should be inserted
2054 * @rq: request to be inserted
2055 * @at_head: insert request at head or tail of queue
2056 * @data: private data
2059 * Many block devices need to execute commands asynchronously, so they don't
2060 * block the whole kernel from preemption during request execution. This is
2061 * accomplished normally by inserting aritficial requests tagged as
2062 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2063 * scheduled for actual execution by the request queue.
2065 * We have the option of inserting the head or the tail of the queue.
2066 * Typically we use the tail for new ioctls and so forth. We use the head
2067 * of the queue for things like a QUEUE_FULL message from a device, or a
2068 * host that is unable to accept a particular command.
2070 void blk_insert_request(request_queue_t *q, struct request *rq,
2071 int at_head, void *data)
2073 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2074 unsigned long flags;
2077 * tell I/O scheduler that this isn't a regular read/write (ie it
2078 * must not attempt merges on this) and that it acts as a soft
2081 rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER;
2085 spin_lock_irqsave(q->queue_lock, flags);
2088 * If command is tagged, release the tag
2090 if (blk_rq_tagged(rq))
2091 blk_queue_end_tag(q, rq);
2093 drive_stat_acct(rq, rq->nr_sectors, 1);
2094 __elv_add_request(q, rq, where, 0);
2096 if (blk_queue_plugged(q))
2097 __generic_unplug_device(q);
2100 spin_unlock_irqrestore(q->queue_lock, flags);
2103 EXPORT_SYMBOL(blk_insert_request);
2106 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2107 * @q: request queue where request should be inserted
2108 * @rw: READ or WRITE data
2109 * @ubuf: the user buffer
2110 * @len: length of user data
2113 * Data will be mapped directly for zero copy io, if possible. Otherwise
2114 * a kernel bounce buffer is used.
2116 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2117 * still in process context.
2119 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2120 * before being submitted to the device, as pages mapped may be out of
2121 * reach. It's the callers responsibility to make sure this happens. The
2122 * original bio must be passed back in to blk_rq_unmap_user() for proper
2125 struct request *blk_rq_map_user(request_queue_t *q, int rw, void __user *ubuf,
2128 unsigned long uaddr;
2132 if (len > (q->max_sectors << 9))
2133 return ERR_PTR(-EINVAL);
2134 if ((!len && ubuf) || (len && !ubuf))
2135 return ERR_PTR(-EINVAL);
2137 rq = blk_get_request(q, rw, __GFP_WAIT);
2139 return ERR_PTR(-ENOMEM);
2142 * if alignment requirement is satisfied, map in user pages for
2143 * direct dma. else, set up kernel bounce buffers
2145 uaddr = (unsigned long) ubuf;
2146 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2147 bio = bio_map_user(q, NULL, uaddr, len, rw == READ);
2149 bio = bio_copy_user(q, uaddr, len, rw == READ);
2152 rq->bio = rq->biotail = bio;
2153 blk_rq_bio_prep(q, rq, bio);
2155 rq->buffer = rq->data = NULL;
2161 * bio is the err-ptr
2163 blk_put_request(rq);
2164 return (struct request *) bio;
2167 EXPORT_SYMBOL(blk_rq_map_user);
2170 * blk_rq_unmap_user - unmap a request with user data
2171 * @rq: request to be unmapped
2172 * @bio: bio for the request
2173 * @ulen: length of user buffer
2176 * Unmap a request previously mapped by blk_rq_map_user().
2178 int blk_rq_unmap_user(struct request *rq, struct bio *bio, unsigned int ulen)
2183 if (bio_flagged(bio, BIO_USER_MAPPED))
2184 bio_unmap_user(bio);
2186 ret = bio_uncopy_user(bio);
2189 blk_put_request(rq);
2193 EXPORT_SYMBOL(blk_rq_unmap_user);
2196 * blk_execute_rq - insert a request into queue for execution
2197 * @q: queue to insert the request in
2198 * @bd_disk: matching gendisk
2199 * @rq: request to insert
2202 * Insert a fully prepared request at the back of the io scheduler queue
2205 int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
2208 DECLARE_COMPLETION(wait);
2209 char sense[SCSI_SENSE_BUFFERSIZE];
2212 rq->rq_disk = bd_disk;
2215 * we need an extra reference to the request, so we can look at
2216 * it after io completion
2221 memset(sense, 0, sizeof(sense));
2226 rq->flags |= REQ_NOMERGE;
2227 rq->waiting = &wait;
2228 rq->end_io = blk_end_sync_rq;
2229 elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 1);
2230 generic_unplug_device(q);
2231 wait_for_completion(&wait);
2240 EXPORT_SYMBOL(blk_execute_rq);
2243 * blkdev_issue_flush - queue a flush
2244 * @bdev: blockdev to issue flush for
2245 * @error_sector: error sector
2248 * Issue a flush for the block device in question. Caller can supply
2249 * room for storing the error offset in case of a flush error, if they
2250 * wish to. Caller must run wait_for_completion() on its own.
2252 int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2256 if (bdev->bd_disk == NULL)
2259 q = bdev_get_queue(bdev);
2262 if (!q->issue_flush_fn)
2265 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2268 EXPORT_SYMBOL(blkdev_issue_flush);
2271 * blkdev_scsi_issue_flush_fn - issue flush for SCSI devices
2274 * @error_sector: error offset
2277 * Devices understanding the SCSI command set, can use this function as
2278 * a helper for issuing a cache flush. Note: driver is required to store
2279 * the error offset (in case of error flushing) in ->sector of struct
2282 int blkdev_scsi_issue_flush_fn(request_queue_t *q, struct gendisk *disk,
2283 sector_t *error_sector)
2285 struct request *rq = blk_get_request(q, WRITE, __GFP_WAIT);
2288 rq->flags |= REQ_BLOCK_PC | REQ_SOFTBARRIER;
2290 memset(rq->cmd, 0, sizeof(rq->cmd));
2295 rq->timeout = 60 * HZ;
2297 ret = blk_execute_rq(q, disk, rq);
2299 if (ret && error_sector)
2300 *error_sector = rq->sector;
2302 blk_put_request(rq);
2306 EXPORT_SYMBOL(blkdev_scsi_issue_flush_fn);
2308 void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
2310 int rw = rq_data_dir(rq);
2312 if (!blk_fs_request(rq) || !rq->rq_disk)
2316 __disk_stat_add(rq->rq_disk, read_sectors, nr_sectors);
2318 __disk_stat_inc(rq->rq_disk, read_merges);
2319 } else if (rw == WRITE) {
2320 __disk_stat_add(rq->rq_disk, write_sectors, nr_sectors);
2322 __disk_stat_inc(rq->rq_disk, write_merges);
2325 disk_round_stats(rq->rq_disk);
2326 rq->rq_disk->in_flight++;
2331 * add-request adds a request to the linked list.
2332 * queue lock is held and interrupts disabled, as we muck with the
2333 * request queue list.
2335 static inline void add_request(request_queue_t * q, struct request * req)
2337 drive_stat_acct(req, req->nr_sectors, 1);
2340 q->activity_fn(q->activity_data, rq_data_dir(req));
2343 * elevator indicated where it wants this request to be
2344 * inserted at elevator_merge time
2346 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2350 * disk_round_stats() - Round off the performance stats on a struct
2353 * The average IO queue length and utilisation statistics are maintained
2354 * by observing the current state of the queue length and the amount of
2355 * time it has been in this state for.
2357 * Normally, that accounting is done on IO completion, but that can result
2358 * in more than a second's worth of IO being accounted for within any one
2359 * second, leading to >100% utilisation. To deal with that, we call this
2360 * function to do a round-off before returning the results when reading
2361 * /proc/diskstats. This accounts immediately for all queue usage up to
2362 * the current jiffies and restarts the counters again.
2364 void disk_round_stats(struct gendisk *disk)
2366 unsigned long now = jiffies;
2368 __disk_stat_add(disk, time_in_queue,
2369 disk->in_flight * (now - disk->stamp));
2372 if (disk->in_flight)
2373 __disk_stat_add(disk, io_ticks, (now - disk->stamp_idle));
2374 disk->stamp_idle = now;
2378 * queue lock must be held
2380 static void __blk_put_request(request_queue_t *q, struct request *req)
2382 struct request_list *rl = req->rl;
2386 if (unlikely(--req->ref_count))
2389 req->rq_status = RQ_INACTIVE;
2394 * Request may not have originated from ll_rw_blk. if not,
2395 * it didn't come out of our reserved rq pools
2398 int rw = rq_data_dir(req);
2400 elv_completed_request(q, req);
2402 BUG_ON(!list_empty(&req->queuelist));
2404 blk_free_request(q, req);
2405 freed_request(q, rw);
2409 void blk_put_request(struct request *req)
2412 * if req->rl isn't set, this request didnt originate from the
2413 * block layer, so it's safe to just disregard it
2416 unsigned long flags;
2417 request_queue_t *q = req->q;
2419 spin_lock_irqsave(q->queue_lock, flags);
2420 __blk_put_request(q, req);
2421 spin_unlock_irqrestore(q->queue_lock, flags);
2425 EXPORT_SYMBOL(blk_put_request);
2428 * blk_end_sync_rq - executes a completion event on a request
2429 * @rq: request to complete
2431 void blk_end_sync_rq(struct request *rq)
2433 struct completion *waiting = rq->waiting;
2436 __blk_put_request(rq->q, rq);
2439 * complete last, if this is a stack request the process (and thus
2440 * the rq pointer) could be invalid right after this complete()
2444 EXPORT_SYMBOL(blk_end_sync_rq);
2447 * blk_congestion_wait - wait for a queue to become uncongested
2448 * @rw: READ or WRITE
2449 * @timeout: timeout in jiffies
2451 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2452 * If no queues are congested then just wait for the next request to be
2455 long blk_congestion_wait(int rw, long timeout)
2459 wait_queue_head_t *wqh = &congestion_wqh[rw];
2461 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2462 ret = io_schedule_timeout(timeout);
2463 finish_wait(wqh, &wait);
2467 EXPORT_SYMBOL(blk_congestion_wait);
2470 * Has to be called with the request spinlock acquired
2472 static int attempt_merge(request_queue_t *q, struct request *req,
2473 struct request *next)
2475 if (!rq_mergeable(req) || !rq_mergeable(next))
2481 if (req->sector + req->nr_sectors != next->sector)
2484 if (rq_data_dir(req) != rq_data_dir(next)
2485 || req->rq_disk != next->rq_disk
2486 || next->waiting || next->special)
2490 * If we are allowed to merge, then append bio list
2491 * from next to rq and release next. merge_requests_fn
2492 * will have updated segment counts, update sector
2495 if (!q->merge_requests_fn(q, req, next))
2499 * At this point we have either done a back merge
2500 * or front merge. We need the smaller start_time of
2501 * the merged requests to be the current request
2502 * for accounting purposes.
2504 if (time_after(req->start_time, next->start_time))
2505 req->start_time = next->start_time;
2507 req->biotail->bi_next = next->bio;
2508 req->biotail = next->biotail;
2510 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2512 elv_merge_requests(q, req, next);
2515 disk_round_stats(req->rq_disk);
2516 req->rq_disk->in_flight--;
2519 __blk_put_request(q, next);
2523 static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2525 struct request *next = elv_latter_request(q, rq);
2528 return attempt_merge(q, rq, next);
2533 static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2535 struct request *prev = elv_former_request(q, rq);
2538 return attempt_merge(q, prev, rq);
2544 * blk_attempt_remerge - attempt to remerge active head with next request
2545 * @q: The &request_queue_t belonging to the device
2546 * @rq: The head request (usually)
2549 * For head-active devices, the queue can easily be unplugged so quickly
2550 * that proper merging is not done on the front request. This may hurt
2551 * performance greatly for some devices. The block layer cannot safely
2552 * do merging on that first request for these queues, but the driver can
2553 * call this function and make it happen any way. Only the driver knows
2554 * when it is safe to do so.
2556 void blk_attempt_remerge(request_queue_t *q, struct request *rq)
2558 unsigned long flags;
2560 spin_lock_irqsave(q->queue_lock, flags);
2561 attempt_back_merge(q, rq);
2562 spin_unlock_irqrestore(q->queue_lock, flags);
2565 EXPORT_SYMBOL(blk_attempt_remerge);
2568 * Non-locking blk_attempt_remerge variant.
2570 void __blk_attempt_remerge(request_queue_t *q, struct request *rq)
2572 attempt_back_merge(q, rq);
2575 EXPORT_SYMBOL(__blk_attempt_remerge);
2577 static int __make_request(request_queue_t *q, struct bio *bio)
2579 struct request *req, *freereq = NULL;
2580 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync;
2583 sector = bio->bi_sector;
2584 nr_sectors = bio_sectors(bio);
2585 cur_nr_sectors = bio_cur_sectors(bio);
2587 rw = bio_data_dir(bio);
2588 sync = bio_sync(bio);
2591 * low level driver can indicate that it wants pages above a
2592 * certain limit bounced to low memory (ie for highmem, or even
2593 * ISA dma in theory)
2595 blk_queue_bounce(q, &bio);
2597 spin_lock_prefetch(q->queue_lock);
2599 barrier = bio_barrier(bio);
2600 if (barrier && (q->ordered == QUEUE_ORDERED_NONE)) {
2606 spin_lock_irq(q->queue_lock);
2608 if (elv_queue_empty(q)) {
2615 el_ret = elv_merge(q, &req, bio);
2617 case ELEVATOR_BACK_MERGE:
2618 BUG_ON(!rq_mergeable(req));
2620 if (!q->back_merge_fn(q, req, bio))
2623 req->biotail->bi_next = bio;
2625 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2626 drive_stat_acct(req, nr_sectors, 0);
2627 if (!attempt_back_merge(q, req))
2628 elv_merged_request(q, req);
2631 case ELEVATOR_FRONT_MERGE:
2632 BUG_ON(!rq_mergeable(req));
2634 if (!q->front_merge_fn(q, req, bio))
2637 bio->bi_next = req->bio;
2641 * may not be valid. if the low level driver said
2642 * it didn't need a bounce buffer then it better
2643 * not touch req->buffer either...
2645 req->buffer = bio_data(bio);
2646 req->current_nr_sectors = cur_nr_sectors;
2647 req->hard_cur_sectors = cur_nr_sectors;
2648 req->sector = req->hard_sector = sector;
2649 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2650 drive_stat_acct(req, nr_sectors, 0);
2651 if (!attempt_front_merge(q, req))
2652 elv_merged_request(q, req);
2656 * elevator says don't/can't merge. get new request
2658 case ELEVATOR_NO_MERGE:
2662 printk("elevator returned crap (%d)\n", el_ret);
2667 * Grab a free request from the freelist - if that is empty, check
2668 * if we are doing read ahead and abort instead of blocking for
2676 spin_unlock_irq(q->queue_lock);
2677 if ((freereq = get_request(q, rw, GFP_ATOMIC)) == NULL) {
2682 if (bio_rw_ahead(bio))
2685 freereq = get_request_wait(q, rw);
2690 req->flags |= REQ_CMD;
2693 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2695 if (bio_rw_ahead(bio) || bio_failfast(bio))
2696 req->flags |= REQ_FAILFAST;
2699 * REQ_BARRIER implies no merging, but lets make it explicit
2702 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
2705 req->hard_sector = req->sector = sector;
2706 req->hard_nr_sectors = req->nr_sectors = nr_sectors;
2707 req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors;
2708 req->nr_phys_segments = bio_phys_segments(q, bio);
2709 req->nr_hw_segments = bio_hw_segments(q, bio);
2710 req->buffer = bio_data(bio); /* see ->buffer comment above */
2711 req->waiting = NULL;
2712 req->bio = req->biotail = bio;
2713 req->rq_disk = bio->bi_bdev->bd_disk;
2714 req->start_time = jiffies;
2716 add_request(q, req);
2719 __blk_put_request(q, freereq);
2721 __generic_unplug_device(q);
2723 spin_unlock_irq(q->queue_lock);
2727 bio_endio(bio, nr_sectors << 9, err);
2732 * If bio->bi_dev is a partition, remap the location
2734 static inline void blk_partition_remap(struct bio *bio)
2736 struct block_device *bdev = bio->bi_bdev;
2738 if (bdev != bdev->bd_contains) {
2739 struct hd_struct *p = bdev->bd_part;
2741 switch (bio->bi_rw) {
2743 p->read_sectors += bio_sectors(bio);
2747 p->write_sectors += bio_sectors(bio);
2751 bio->bi_sector += p->start_sect;
2752 bio->bi_bdev = bdev->bd_contains;
2756 void blk_finish_queue_drain(request_queue_t *q)
2758 struct request_list *rl = &q->rq;
2761 spin_lock_irq(q->queue_lock);
2762 clear_bit(QUEUE_FLAG_DRAIN, &q->queue_flags);
2764 while (!list_empty(&q->drain_list)) {
2765 rq = list_entry_rq(q->drain_list.next);
2767 list_del_init(&rq->queuelist);
2768 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 1);
2771 spin_unlock_irq(q->queue_lock);
2773 wake_up(&rl->wait[0]);
2774 wake_up(&rl->wait[1]);
2775 wake_up(&rl->drain);
2778 static int wait_drain(request_queue_t *q, struct request_list *rl, int dispatch)
2780 int wait = rl->count[READ] + rl->count[WRITE];
2783 wait += !list_empty(&q->queue_head);
2789 * We rely on the fact that only requests allocated through blk_alloc_request()
2790 * have io scheduler private data structures associated with them. Any other
2791 * type of request (allocated on stack or through kmalloc()) should not go
2792 * to the io scheduler core, but be attached to the queue head instead.
2794 void blk_wait_queue_drained(request_queue_t *q, int wait_dispatch)
2796 struct request_list *rl = &q->rq;
2799 spin_lock_irq(q->queue_lock);
2800 set_bit(QUEUE_FLAG_DRAIN, &q->queue_flags);
2802 while (wait_drain(q, rl, wait_dispatch)) {
2803 prepare_to_wait(&rl->drain, &wait, TASK_UNINTERRUPTIBLE);
2805 if (wait_drain(q, rl, wait_dispatch)) {
2806 __generic_unplug_device(q);
2807 spin_unlock_irq(q->queue_lock);
2809 spin_lock_irq(q->queue_lock);
2812 finish_wait(&rl->drain, &wait);
2815 spin_unlock_irq(q->queue_lock);
2819 * block waiting for the io scheduler being started again.
2821 static inline void block_wait_queue_running(request_queue_t *q)
2825 while (test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)) {
2826 struct request_list *rl = &q->rq;
2828 prepare_to_wait_exclusive(&rl->drain, &wait,
2829 TASK_UNINTERRUPTIBLE);
2832 * re-check the condition. avoids using prepare_to_wait()
2833 * in the fast path (queue is running)
2835 if (test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))
2838 finish_wait(&rl->drain, &wait);
2842 static void handle_bad_sector(struct bio *bio)
2844 char b[BDEVNAME_SIZE];
2846 printk(KERN_INFO "attempt to access beyond end of device\n");
2847 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
2848 bdevname(bio->bi_bdev, b),
2850 (unsigned long long)bio->bi_sector + bio_sectors(bio),
2851 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
2853 set_bit(BIO_EOF, &bio->bi_flags);
2857 * generic_make_request: hand a buffer to its device driver for I/O
2858 * @bio: The bio describing the location in memory and on the device.
2860 * generic_make_request() is used to make I/O requests of block
2861 * devices. It is passed a &struct bio, which describes the I/O that needs
2864 * generic_make_request() does not return any status. The
2865 * success/failure status of the request, along with notification of
2866 * completion, is delivered asynchronously through the bio->bi_end_io
2867 * function described (one day) else where.
2869 * The caller of generic_make_request must make sure that bi_io_vec
2870 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2871 * set to describe the device address, and the
2872 * bi_end_io and optionally bi_private are set to describe how
2873 * completion notification should be signaled.
2875 * generic_make_request and the drivers it calls may use bi_next if this
2876 * bio happens to be merged with someone else, and may change bi_dev and
2877 * bi_sector for remaps as it sees fit. So the values of these fields
2878 * should NOT be depended on after the call to generic_make_request.
2880 void generic_make_request(struct bio *bio)
2884 int ret, nr_sectors = bio_sectors(bio);
2887 /* Test device or partition size, when known. */
2888 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
2890 sector_t sector = bio->bi_sector;
2892 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2894 * This may well happen - the kernel calls bread()
2895 * without checking the size of the device, e.g., when
2896 * mounting a device.
2898 handle_bad_sector(bio);
2904 * Resolve the mapping until finished. (drivers are
2905 * still free to implement/resolve their own stacking
2906 * by explicitly returning 0)
2908 * NOTE: we don't repeat the blk_size check for each new device.
2909 * Stacking drivers are expected to know what they are doing.
2912 char b[BDEVNAME_SIZE];
2914 q = bdev_get_queue(bio->bi_bdev);
2917 "generic_make_request: Trying to access "
2918 "nonexistent block-device %s (%Lu)\n",
2919 bdevname(bio->bi_bdev, b),
2920 (long long) bio->bi_sector);
2922 bio_endio(bio, bio->bi_size, -EIO);
2926 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
2927 printk("bio too big device %s (%u > %u)\n",
2928 bdevname(bio->bi_bdev, b),
2934 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))
2937 block_wait_queue_running(q);
2940 * If this device has partitions, remap block n
2941 * of partition p to block n+start(p) of the disk.
2943 blk_partition_remap(bio);
2945 ret = q->make_request_fn(q, bio);
2949 EXPORT_SYMBOL(generic_make_request);
2952 * submit_bio: submit a bio to the block device layer for I/O
2953 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2954 * @bio: The &struct bio which describes the I/O
2956 * submit_bio() is very similar in purpose to generic_make_request(), and
2957 * uses that function to do most of the work. Both are fairly rough
2958 * interfaces, @bio must be presetup and ready for I/O.
2961 void submit_bio(int rw, struct bio *bio)
2963 int count = bio_sectors(bio);
2965 BIO_BUG_ON(!bio->bi_size);
2966 BIO_BUG_ON(!bio->bi_io_vec);
2969 mod_page_state(pgpgout, count);
2971 mod_page_state(pgpgin, count);
2973 if (unlikely(block_dump)) {
2974 char b[BDEVNAME_SIZE];
2975 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
2976 current->comm, current->pid,
2977 (rw & WRITE) ? "WRITE" : "READ",
2978 (unsigned long long)bio->bi_sector,
2979 bdevname(bio->bi_bdev,b));
2982 generic_make_request(bio);
2985 EXPORT_SYMBOL(submit_bio);
2987 void blk_recalc_rq_segments(struct request *rq)
2989 struct bio *bio, *prevbio = NULL;
2990 int nr_phys_segs, nr_hw_segs;
2991 unsigned int phys_size, hw_size;
2992 request_queue_t *q = rq->q;
2997 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
2998 rq_for_each_bio(bio, rq) {
2999 /* Force bio hw/phys segs to be recalculated. */
3000 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3002 nr_phys_segs += bio_phys_segments(q, bio);
3003 nr_hw_segs += bio_hw_segments(q, bio);
3005 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3006 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3008 if (blk_phys_contig_segment(q, prevbio, bio) &&
3009 pseg <= q->max_segment_size) {
3011 phys_size += prevbio->bi_size + bio->bi_size;
3015 if (blk_hw_contig_segment(q, prevbio, bio) &&
3016 hseg <= q->max_segment_size) {
3018 hw_size += prevbio->bi_size + bio->bi_size;
3025 rq->nr_phys_segments = nr_phys_segs;
3026 rq->nr_hw_segments = nr_hw_segs;
3029 void blk_recalc_rq_sectors(struct request *rq, int nsect)
3031 if (blk_fs_request(rq)) {
3032 rq->hard_sector += nsect;
3033 rq->hard_nr_sectors -= nsect;
3036 * Move the I/O submission pointers ahead if required.
3038 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3039 (rq->sector <= rq->hard_sector)) {
3040 rq->sector = rq->hard_sector;
3041 rq->nr_sectors = rq->hard_nr_sectors;
3042 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3043 rq->current_nr_sectors = rq->hard_cur_sectors;
3044 rq->buffer = bio_data(rq->bio);
3048 * if total number of sectors is less than the first segment
3049 * size, something has gone terribly wrong
3051 if (rq->nr_sectors < rq->current_nr_sectors) {
3052 printk("blk: request botched\n");
3053 rq->nr_sectors = rq->current_nr_sectors;
3058 static int __end_that_request_first(struct request *req, int uptodate,
3061 int total_bytes, bio_nbytes, error, next_idx = 0;
3065 * extend uptodate bool to allow < 0 value to be direct io error
3068 if (end_io_error(uptodate))
3069 error = !uptodate ? -EIO : uptodate;
3072 * for a REQ_BLOCK_PC request, we want to carry any eventual
3073 * sense key with us all the way through
3075 if (!blk_pc_request(req))
3079 if (blk_fs_request(req) && !(req->flags & REQ_QUIET))
3080 printk("end_request: I/O error, dev %s, sector %llu\n",
3081 req->rq_disk ? req->rq_disk->disk_name : "?",
3082 (unsigned long long)req->sector);
3085 total_bytes = bio_nbytes = 0;
3086 while ((bio = req->bio) != NULL) {
3089 if (nr_bytes >= bio->bi_size) {
3090 req->bio = bio->bi_next;
3091 nbytes = bio->bi_size;
3092 bio_endio(bio, nbytes, error);
3096 int idx = bio->bi_idx + next_idx;
3098 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3099 blk_dump_rq_flags(req, "__end_that");
3100 printk("%s: bio idx %d >= vcnt %d\n",
3102 bio->bi_idx, bio->bi_vcnt);
3106 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3107 BIO_BUG_ON(nbytes > bio->bi_size);
3110 * not a complete bvec done
3112 if (unlikely(nbytes > nr_bytes)) {
3113 bio_nbytes += nr_bytes;
3114 total_bytes += nr_bytes;
3119 * advance to the next vector
3122 bio_nbytes += nbytes;
3125 total_bytes += nbytes;
3128 if ((bio = req->bio)) {
3130 * end more in this run, or just return 'not-done'
3132 if (unlikely(nr_bytes <= 0))
3144 * if the request wasn't completed, update state
3147 bio_endio(bio, bio_nbytes, error);
3148 bio->bi_idx += next_idx;
3149 bio_iovec(bio)->bv_offset += nr_bytes;
3150 bio_iovec(bio)->bv_len -= nr_bytes;
3153 blk_recalc_rq_sectors(req, total_bytes >> 9);
3154 blk_recalc_rq_segments(req);
3159 * end_that_request_first - end I/O on a request
3160 * @req: the request being processed
3161 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3162 * @nr_sectors: number of sectors to end I/O on
3165 * Ends I/O on a number of sectors attached to @req, and sets it up
3166 * for the next range of segments (if any) in the cluster.
3169 * 0 - we are done with this request, call end_that_request_last()
3170 * 1 - still buffers pending for this request
3172 int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3174 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3177 EXPORT_SYMBOL(end_that_request_first);
3180 * end_that_request_chunk - end I/O on a request
3181 * @req: the request being processed
3182 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3183 * @nr_bytes: number of bytes to complete
3186 * Ends I/O on a number of bytes attached to @req, and sets it up
3187 * for the next range of segments (if any). Like end_that_request_first(),
3188 * but deals with bytes instead of sectors.
3191 * 0 - we are done with this request, call end_that_request_last()
3192 * 1 - still buffers pending for this request
3194 int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3196 return __end_that_request_first(req, uptodate, nr_bytes);
3199 EXPORT_SYMBOL(end_that_request_chunk);
3202 * queue lock must be held
3204 void end_that_request_last(struct request *req)
3206 struct gendisk *disk = req->rq_disk;
3208 if (unlikely(laptop_mode) && blk_fs_request(req))
3209 laptop_io_completion();
3211 if (disk && blk_fs_request(req)) {
3212 unsigned long duration = jiffies - req->start_time;
3213 switch (rq_data_dir(req)) {
3215 __disk_stat_inc(disk, writes);
3216 __disk_stat_add(disk, write_ticks, duration);
3219 __disk_stat_inc(disk, reads);
3220 __disk_stat_add(disk, read_ticks, duration);
3223 disk_round_stats(disk);
3229 __blk_put_request(req->q, req);
3232 EXPORT_SYMBOL(end_that_request_last);
3234 void end_request(struct request *req, int uptodate)
3236 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3237 add_disk_randomness(req->rq_disk);
3238 blkdev_dequeue_request(req);
3239 end_that_request_last(req);
3243 EXPORT_SYMBOL(end_request);
3245 void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3247 /* first three bits are identical in rq->flags and bio->bi_rw */
3248 rq->flags |= (bio->bi_rw & 7);
3250 rq->nr_phys_segments = bio_phys_segments(q, bio);
3251 rq->nr_hw_segments = bio_hw_segments(q, bio);
3252 rq->current_nr_sectors = bio_cur_sectors(bio);
3253 rq->hard_cur_sectors = rq->current_nr_sectors;
3254 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3255 rq->buffer = bio_data(bio);
3257 rq->bio = rq->biotail = bio;
3260 EXPORT_SYMBOL(blk_rq_bio_prep);
3262 int kblockd_schedule_work(struct work_struct *work)
3264 return queue_work(kblockd_workqueue, work);
3267 EXPORT_SYMBOL(kblockd_schedule_work);
3269 void kblockd_flush(void)
3271 flush_workqueue(kblockd_workqueue);
3273 EXPORT_SYMBOL(kblockd_flush);
3275 int __init blk_dev_init(void)
3277 kblockd_workqueue = create_workqueue("kblockd");
3278 if (!kblockd_workqueue)
3279 panic("Failed to create kblockd\n");
3281 request_cachep = kmem_cache_create("blkdev_requests",
3282 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3284 requestq_cachep = kmem_cache_create("blkdev_queue",
3285 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3287 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3288 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3290 blk_max_low_pfn = max_low_pfn;
3291 blk_max_pfn = max_pfn;
3297 * IO Context helper functions
3299 void put_io_context(struct io_context *ioc)
3304 BUG_ON(atomic_read(&ioc->refcount) == 0);
3306 if (atomic_dec_and_test(&ioc->refcount)) {
3307 if (ioc->aic && ioc->aic->dtor)
3308 ioc->aic->dtor(ioc->aic);
3309 if (ioc->cic && ioc->cic->dtor)
3310 ioc->cic->dtor(ioc->cic);
3312 kmem_cache_free(iocontext_cachep, ioc);
3315 EXPORT_SYMBOL(put_io_context);
3317 /* Called by the exitting task */
3318 void exit_io_context(void)
3320 unsigned long flags;
3321 struct io_context *ioc;
3323 local_irq_save(flags);
3324 ioc = current->io_context;
3325 current->io_context = NULL;
3326 local_irq_restore(flags);
3328 if (ioc->aic && ioc->aic->exit)
3329 ioc->aic->exit(ioc->aic);
3330 if (ioc->cic && ioc->cic->exit)
3331 ioc->cic->exit(ioc->cic);
3333 put_io_context(ioc);
3337 * If the current task has no IO context then create one and initialise it.
3338 * If it does have a context, take a ref on it.
3340 * This is always called in the context of the task which submitted the I/O.
3341 * But weird things happen, so we disable local interrupts to ensure exclusive
3342 * access to *current.
3344 struct io_context *get_io_context(int gfp_flags)
3346 struct task_struct *tsk = current;
3347 unsigned long flags;
3348 struct io_context *ret;
3350 local_irq_save(flags);
3351 ret = tsk->io_context;
3355 local_irq_restore(flags);
3357 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags);
3359 atomic_set(&ret->refcount, 1);
3360 ret->pid = tsk->pid;
3361 ret->last_waited = jiffies; /* doesn't matter... */
3362 ret->nr_batch_requests = 0; /* because this is 0 */
3365 spin_lock_init(&ret->lock);
3367 local_irq_save(flags);
3370 * very unlikely, someone raced with us in setting up the task
3371 * io context. free new context and just grab a reference.
3373 if (!tsk->io_context)
3374 tsk->io_context = ret;
3376 kmem_cache_free(iocontext_cachep, ret);
3377 ret = tsk->io_context;
3381 atomic_inc(&ret->refcount);
3382 local_irq_restore(flags);
3387 EXPORT_SYMBOL(get_io_context);
3389 void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3391 struct io_context *src = *psrc;
3392 struct io_context *dst = *pdst;
3395 BUG_ON(atomic_read(&src->refcount) == 0);
3396 atomic_inc(&src->refcount);
3397 put_io_context(dst);
3401 EXPORT_SYMBOL(copy_io_context);
3403 void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3405 struct io_context *temp;
3410 EXPORT_SYMBOL(swap_io_context);
3415 struct queue_sysfs_entry {
3416 struct attribute attr;
3417 ssize_t (*show)(struct request_queue *, char *);
3418 ssize_t (*store)(struct request_queue *, const char *, size_t);
3422 queue_var_show(unsigned int var, char *page)
3424 return sprintf(page, "%d\n", var);
3428 queue_var_store(unsigned long *var, const char *page, size_t count)
3430 char *p = (char *) page;
3432 *var = simple_strtoul(p, &p, 10);
3436 static ssize_t queue_requests_show(struct request_queue *q, char *page)
3438 return queue_var_show(q->nr_requests, (page));
3442 queue_requests_store(struct request_queue *q, const char *page, size_t count)
3444 struct request_list *rl = &q->rq;
3446 int ret = queue_var_store(&q->nr_requests, page, count);
3447 if (q->nr_requests < BLKDEV_MIN_RQ)
3448 q->nr_requests = BLKDEV_MIN_RQ;
3449 blk_queue_congestion_threshold(q);
3451 if (rl->count[READ] >= queue_congestion_on_threshold(q))
3452 set_queue_congested(q, READ);
3453 else if (rl->count[READ] < queue_congestion_off_threshold(q))
3454 clear_queue_congested(q, READ);
3456 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3457 set_queue_congested(q, WRITE);
3458 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3459 clear_queue_congested(q, WRITE);
3461 if (rl->count[READ] >= q->nr_requests) {
3462 blk_set_queue_full(q, READ);
3463 } else if (rl->count[READ]+1 <= q->nr_requests) {
3464 blk_clear_queue_full(q, READ);
3465 wake_up(&rl->wait[READ]);
3468 if (rl->count[WRITE] >= q->nr_requests) {
3469 blk_set_queue_full(q, WRITE);
3470 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3471 blk_clear_queue_full(q, WRITE);
3472 wake_up(&rl->wait[WRITE]);
3477 static ssize_t queue_ra_show(struct request_queue *q, char *page)
3479 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3481 return queue_var_show(ra_kb, (page));
3485 queue_ra_store(struct request_queue *q, const char *page, size_t count)
3487 unsigned long ra_kb;
3488 ssize_t ret = queue_var_store(&ra_kb, page, count);
3490 spin_lock_irq(q->queue_lock);
3491 if (ra_kb > (q->max_sectors >> 1))
3492 ra_kb = (q->max_sectors >> 1);
3494 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3495 spin_unlock_irq(q->queue_lock);
3500 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3502 int max_sectors_kb = q->max_sectors >> 1;
3504 return queue_var_show(max_sectors_kb, (page));
3508 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3510 unsigned long max_sectors_kb,
3511 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3512 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3513 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3516 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3519 * Take the queue lock to update the readahead and max_sectors
3520 * values synchronously:
3522 spin_lock_irq(q->queue_lock);
3524 * Trim readahead window as well, if necessary:
3526 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3527 if (ra_kb > max_sectors_kb)
3528 q->backing_dev_info.ra_pages =
3529 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3531 q->max_sectors = max_sectors_kb << 1;
3532 spin_unlock_irq(q->queue_lock);
3537 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3539 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3541 return queue_var_show(max_hw_sectors_kb, (page));
3545 static struct queue_sysfs_entry queue_requests_entry = {
3546 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3547 .show = queue_requests_show,
3548 .store = queue_requests_store,
3551 static struct queue_sysfs_entry queue_ra_entry = {
3552 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3553 .show = queue_ra_show,
3554 .store = queue_ra_store,
3557 static struct queue_sysfs_entry queue_max_sectors_entry = {
3558 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3559 .show = queue_max_sectors_show,
3560 .store = queue_max_sectors_store,
3563 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3564 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3565 .show = queue_max_hw_sectors_show,
3568 static struct queue_sysfs_entry queue_iosched_entry = {
3569 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3570 .show = elv_iosched_show,
3571 .store = elv_iosched_store,
3574 static struct attribute *default_attrs[] = {
3575 &queue_requests_entry.attr,
3576 &queue_ra_entry.attr,
3577 &queue_max_hw_sectors_entry.attr,
3578 &queue_max_sectors_entry.attr,
3579 &queue_iosched_entry.attr,
3583 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3586 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3588 struct queue_sysfs_entry *entry = to_queue(attr);
3589 struct request_queue *q;
3591 q = container_of(kobj, struct request_queue, kobj);
3595 return entry->show(q, page);
3599 queue_attr_store(struct kobject *kobj, struct attribute *attr,
3600 const char *page, size_t length)
3602 struct queue_sysfs_entry *entry = to_queue(attr);
3603 struct request_queue *q;
3605 q = container_of(kobj, struct request_queue, kobj);
3609 return entry->store(q, page, length);
3612 static struct sysfs_ops queue_sysfs_ops = {
3613 .show = queue_attr_show,
3614 .store = queue_attr_store,
3617 struct kobj_type queue_ktype = {
3618 .sysfs_ops = &queue_sysfs_ops,
3619 .default_attrs = default_attrs,
3622 int blk_register_queue(struct gendisk *disk)
3626 request_queue_t *q = disk->queue;
3628 if (!q || !q->request_fn)
3631 q->kobj.parent = kobject_get(&disk->kobj);
3632 if (!q->kobj.parent)
3635 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
3636 q->kobj.ktype = &queue_ktype;
3638 ret = kobject_register(&q->kobj);
3642 ret = elv_register_queue(q);
3644 kobject_unregister(&q->kobj);
3651 void blk_unregister_queue(struct gendisk *disk)
3653 request_queue_t *q = disk->queue;
3655 if (q && q->request_fn) {
3656 elv_unregister_queue(q);
3658 kobject_unregister(&q->kobj);
3659 kobject_put(&disk->kobj);