2 * Functions related to setting various queue properties from drivers
4 #include <linux/kernel.h>
5 #include <linux/module.h>
6 #include <linux/init.h>
8 #include <linux/blkdev.h>
9 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
13 unsigned long blk_max_low_pfn;
14 EXPORT_SYMBOL(blk_max_low_pfn);
16 unsigned long blk_max_pfn;
19 * blk_queue_prep_rq - set a prepare_request function for queue
21 * @pfn: prepare_request function
23 * It's possible for a queue to register a prepare_request callback which
24 * is invoked before the request is handed to the request_fn. The goal of
25 * the function is to prepare a request for I/O, it can be used to build a
26 * cdb from the request data for instance.
29 void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
33 EXPORT_SYMBOL(blk_queue_prep_rq);
36 * blk_queue_set_discard - set a discard_sectors function for queue
38 * @dfn: prepare_discard function
40 * It's possible for a queue to register a discard callback which is used
41 * to transform a discard request into the appropriate type for the
42 * hardware. If none is registered, then discard requests are failed
46 void blk_queue_set_discard(struct request_queue *q, prepare_discard_fn *dfn)
48 q->prepare_discard_fn = dfn;
50 EXPORT_SYMBOL(blk_queue_set_discard);
53 * blk_queue_merge_bvec - set a merge_bvec function for queue
55 * @mbfn: merge_bvec_fn
57 * Usually queues have static limitations on the max sectors or segments that
58 * we can put in a request. Stacking drivers may have some settings that
59 * are dynamic, and thus we have to query the queue whether it is ok to
60 * add a new bio_vec to a bio at a given offset or not. If the block device
61 * has such limitations, it needs to register a merge_bvec_fn to control
62 * the size of bio's sent to it. Note that a block device *must* allow a
63 * single page to be added to an empty bio. The block device driver may want
64 * to use the bio_split() function to deal with these bio's. By default
65 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
68 void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
70 q->merge_bvec_fn = mbfn;
72 EXPORT_SYMBOL(blk_queue_merge_bvec);
74 void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
76 q->softirq_done_fn = fn;
78 EXPORT_SYMBOL(blk_queue_softirq_done);
80 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
82 q->rq_timeout = timeout;
84 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
86 void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
88 q->rq_timed_out_fn = fn;
90 EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
92 void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
96 EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
99 * blk_queue_make_request - define an alternate make_request function for a device
100 * @q: the request queue for the device to be affected
101 * @mfn: the alternate make_request function
104 * The normal way for &struct bios to be passed to a device
105 * driver is for them to be collected into requests on a request
106 * queue, and then to allow the device driver to select requests
107 * off that queue when it is ready. This works well for many block
108 * devices. However some block devices (typically virtual devices
109 * such as md or lvm) do not benefit from the processing on the
110 * request queue, and are served best by having the requests passed
111 * directly to them. This can be achieved by providing a function
112 * to blk_queue_make_request().
115 * The driver that does this *must* be able to deal appropriately
116 * with buffers in "highmemory". This can be accomplished by either calling
117 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
118 * blk_queue_bounce() to create a buffer in normal memory.
120 void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
125 q->nr_requests = BLKDEV_MAX_RQ;
126 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
127 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
128 blk_queue_segment_boundary(q, BLK_SEG_BOUNDARY_MASK);
129 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
131 q->make_request_fn = mfn;
132 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
133 blk_queue_logical_block_size(q, 512);
134 blk_queue_dma_alignment(q, 511);
135 blk_queue_congestion_threshold(q);
136 q->nr_batching = BLK_BATCH_REQ;
138 q->unplug_thresh = 4; /* hmm */
139 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
140 if (q->unplug_delay == 0)
143 q->unplug_timer.function = blk_unplug_timeout;
144 q->unplug_timer.data = (unsigned long)q;
147 * by default assume old behaviour and bounce for any highmem page
149 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
151 EXPORT_SYMBOL(blk_queue_make_request);
154 * blk_queue_bounce_limit - set bounce buffer limit for queue
155 * @q: the request queue for the device
156 * @dma_mask: the maximum address the device can handle
159 * Different hardware can have different requirements as to what pages
160 * it can do I/O directly to. A low level driver can call
161 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
162 * buffers for doing I/O to pages residing above @dma_mask.
164 void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
166 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
169 q->bounce_gfp = GFP_NOIO;
170 #if BITS_PER_LONG == 64
172 * Assume anything <= 4GB can be handled by IOMMU. Actually
173 * some IOMMUs can handle everything, but I don't know of a
174 * way to test this here.
176 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
178 q->limits.bounce_pfn = max_low_pfn;
180 if (b_pfn < blk_max_low_pfn)
182 q->limits.bounce_pfn = b_pfn;
185 init_emergency_isa_pool();
186 q->bounce_gfp = GFP_NOIO | GFP_DMA;
187 q->limits.bounce_pfn = b_pfn;
190 EXPORT_SYMBOL(blk_queue_bounce_limit);
193 * blk_queue_max_sectors - set max sectors for a request for this queue
194 * @q: the request queue for the device
195 * @max_sectors: max sectors in the usual 512b unit
198 * Enables a low level driver to set an upper limit on the size of
201 void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
203 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
204 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
205 printk(KERN_INFO "%s: set to minimum %d\n",
206 __func__, max_sectors);
209 if (BLK_DEF_MAX_SECTORS > max_sectors)
210 q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors;
212 q->limits.max_sectors = BLK_DEF_MAX_SECTORS;
213 q->limits.max_hw_sectors = max_sectors;
216 EXPORT_SYMBOL(blk_queue_max_sectors);
218 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors)
220 if (BLK_DEF_MAX_SECTORS > max_sectors)
221 q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS;
223 q->limits.max_hw_sectors = max_sectors;
225 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
228 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
229 * @q: the request queue for the device
230 * @max_segments: max number of segments
233 * Enables a low level driver to set an upper limit on the number of
234 * physical data segments in a request. This would be the largest sized
235 * scatter list the driver could handle.
237 void blk_queue_max_phys_segments(struct request_queue *q,
238 unsigned short max_segments)
242 printk(KERN_INFO "%s: set to minimum %d\n",
243 __func__, max_segments);
246 q->limits.max_phys_segments = max_segments;
248 EXPORT_SYMBOL(blk_queue_max_phys_segments);
251 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
252 * @q: the request queue for the device
253 * @max_segments: max number of segments
256 * Enables a low level driver to set an upper limit on the number of
257 * hw data segments in a request. This would be the largest number of
258 * address/length pairs the host adapter can actually give at once
261 void blk_queue_max_hw_segments(struct request_queue *q,
262 unsigned short max_segments)
266 printk(KERN_INFO "%s: set to minimum %d\n",
267 __func__, max_segments);
270 q->limits.max_hw_segments = max_segments;
272 EXPORT_SYMBOL(blk_queue_max_hw_segments);
275 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
276 * @q: the request queue for the device
277 * @max_size: max size of segment in bytes
280 * Enables a low level driver to set an upper limit on the size of a
283 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
285 if (max_size < PAGE_CACHE_SIZE) {
286 max_size = PAGE_CACHE_SIZE;
287 printk(KERN_INFO "%s: set to minimum %d\n",
291 q->limits.max_segment_size = max_size;
293 EXPORT_SYMBOL(blk_queue_max_segment_size);
296 * blk_queue_logical_block_size - set logical block size for the queue
297 * @q: the request queue for the device
298 * @size: the logical block size, in bytes
301 * This should be set to the lowest possible block size that the
302 * storage device can address. The default of 512 covers most
305 void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
307 q->limits.logical_block_size = size;
309 if (q->limits.physical_block_size < size)
310 q->limits.physical_block_size = size;
312 if (q->limits.io_min < q->limits.physical_block_size)
313 q->limits.io_min = q->limits.physical_block_size;
315 EXPORT_SYMBOL(blk_queue_logical_block_size);
318 * blk_queue_physical_block_size - set physical block size for the queue
319 * @q: the request queue for the device
320 * @size: the physical block size, in bytes
323 * This should be set to the lowest possible sector size that the
324 * hardware can operate on without reverting to read-modify-write
327 void blk_queue_physical_block_size(struct request_queue *q, unsigned short size)
329 q->limits.physical_block_size = size;
331 if (q->limits.physical_block_size < q->limits.logical_block_size)
332 q->limits.physical_block_size = q->limits.logical_block_size;
334 if (q->limits.io_min < q->limits.physical_block_size)
335 q->limits.io_min = q->limits.physical_block_size;
337 EXPORT_SYMBOL(blk_queue_physical_block_size);
340 * blk_queue_alignment_offset - set physical block alignment offset
341 * @q: the request queue for the device
342 * @offset: alignment offset in bytes
345 * Some devices are naturally misaligned to compensate for things like
346 * the legacy DOS partition table 63-sector offset. Low-level drivers
347 * should call this function for devices whose first sector is not
350 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
352 q->limits.alignment_offset =
353 offset & (q->limits.physical_block_size - 1);
354 q->limits.misaligned = 0;
356 EXPORT_SYMBOL(blk_queue_alignment_offset);
359 * blk_queue_io_min - set minimum request size for the queue
360 * @q: the request queue for the device
361 * @min: smallest I/O size in bytes
364 * Some devices have an internal block size bigger than the reported
365 * hardware sector size. This function can be used to signal the
366 * smallest I/O the device can perform without incurring a performance
369 void blk_queue_io_min(struct request_queue *q, unsigned int min)
371 q->limits.io_min = min;
373 if (q->limits.io_min < q->limits.logical_block_size)
374 q->limits.io_min = q->limits.logical_block_size;
376 if (q->limits.io_min < q->limits.physical_block_size)
377 q->limits.io_min = q->limits.physical_block_size;
379 EXPORT_SYMBOL(blk_queue_io_min);
382 * blk_queue_io_opt - set optimal request size for the queue
383 * @q: the request queue for the device
384 * @opt: optimal request size in bytes
387 * Drivers can call this function to set the preferred I/O request
388 * size for devices that report such a value.
390 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
392 q->limits.io_opt = opt;
394 EXPORT_SYMBOL(blk_queue_io_opt);
397 * Returns the minimum that is _not_ zero, unless both are zero.
399 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
402 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
403 * @t: the stacking driver (top)
404 * @b: the underlying device (bottom)
406 void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
408 /* zero is "infinity" */
409 t->limits.max_sectors = min_not_zero(queue_max_sectors(t),
410 queue_max_sectors(b));
412 t->limits.max_hw_sectors = min_not_zero(queue_max_hw_sectors(t),
413 queue_max_hw_sectors(b));
415 t->limits.seg_boundary_mask = min_not_zero(queue_segment_boundary(t),
416 queue_segment_boundary(b));
418 t->limits.max_phys_segments = min_not_zero(queue_max_phys_segments(t),
419 queue_max_phys_segments(b));
421 t->limits.max_hw_segments = min_not_zero(queue_max_hw_segments(t),
422 queue_max_hw_segments(b));
424 t->limits.max_segment_size = min_not_zero(queue_max_segment_size(t),
425 queue_max_segment_size(b));
427 t->limits.logical_block_size = max(queue_logical_block_size(t),
428 queue_logical_block_size(b));
432 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
434 spin_lock_irqsave(t->queue_lock, flags);
435 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
436 spin_unlock_irqrestore(t->queue_lock, flags);
439 EXPORT_SYMBOL(blk_queue_stack_limits);
442 * blk_stack_limits - adjust queue_limits for stacked devices
443 * @t: the stacking driver limits (top)
444 * @b: the underlying queue limits (bottom)
445 * @offset: offset to beginning of data within component device
448 * Merges two queue_limit structs. Returns 0 if alignment didn't
449 * change. Returns -1 if adding the bottom device caused
452 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
455 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
456 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
457 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
459 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
460 b->seg_boundary_mask);
462 t->max_phys_segments = min_not_zero(t->max_phys_segments,
463 b->max_phys_segments);
465 t->max_hw_segments = min_not_zero(t->max_hw_segments,
468 t->max_segment_size = min_not_zero(t->max_segment_size,
469 b->max_segment_size);
471 t->logical_block_size = max(t->logical_block_size,
472 b->logical_block_size);
474 t->physical_block_size = max(t->physical_block_size,
475 b->physical_block_size);
477 t->io_min = max(t->io_min, b->io_min);
478 t->no_cluster |= b->no_cluster;
480 /* Bottom device offset aligned? */
482 (offset & (b->physical_block_size - 1)) != b->alignment_offset) {
487 /* If top has no alignment offset, inherit from bottom */
488 if (!t->alignment_offset)
489 t->alignment_offset =
490 b->alignment_offset & (b->physical_block_size - 1);
492 /* Top device aligned on logical block boundary? */
493 if (t->alignment_offset & (t->logical_block_size - 1)) {
500 EXPORT_SYMBOL(blk_stack_limits);
503 * disk_stack_limits - adjust queue limits for stacked drivers
504 * @disk: MD/DM gendisk (top)
505 * @bdev: the underlying block device (bottom)
506 * @offset: offset to beginning of data within component device
509 * Merges the limits for two queues. Returns 0 if alignment
510 * didn't change. Returns -1 if adding the bottom device caused
513 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
516 struct request_queue *t = disk->queue;
517 struct request_queue *b = bdev_get_queue(bdev);
519 offset += get_start_sect(bdev) << 9;
521 if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) {
522 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
524 disk_name(disk, 0, top);
525 bdevname(bdev, bottom);
527 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
533 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
536 spin_lock_irqsave(t->queue_lock, flags);
537 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
538 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
539 spin_unlock_irqrestore(t->queue_lock, flags);
542 EXPORT_SYMBOL(disk_stack_limits);
545 * blk_queue_dma_pad - set pad mask
546 * @q: the request queue for the device
551 * Appending pad buffer to a request modifies the last entry of a
552 * scatter list such that it includes the pad buffer.
554 void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
556 q->dma_pad_mask = mask;
558 EXPORT_SYMBOL(blk_queue_dma_pad);
561 * blk_queue_update_dma_pad - update pad mask
562 * @q: the request queue for the device
565 * Update dma pad mask.
567 * Appending pad buffer to a request modifies the last entry of a
568 * scatter list such that it includes the pad buffer.
570 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
572 if (mask > q->dma_pad_mask)
573 q->dma_pad_mask = mask;
575 EXPORT_SYMBOL(blk_queue_update_dma_pad);
578 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
579 * @q: the request queue for the device
580 * @dma_drain_needed: fn which returns non-zero if drain is necessary
581 * @buf: physically contiguous buffer
582 * @size: size of the buffer in bytes
584 * Some devices have excess DMA problems and can't simply discard (or
585 * zero fill) the unwanted piece of the transfer. They have to have a
586 * real area of memory to transfer it into. The use case for this is
587 * ATAPI devices in DMA mode. If the packet command causes a transfer
588 * bigger than the transfer size some HBAs will lock up if there
589 * aren't DMA elements to contain the excess transfer. What this API
590 * does is adjust the queue so that the buf is always appended
591 * silently to the scatterlist.
593 * Note: This routine adjusts max_hw_segments to make room for
594 * appending the drain buffer. If you call
595 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
596 * calling this routine, you must set the limit to one fewer than your
597 * device can support otherwise there won't be room for the drain
600 int blk_queue_dma_drain(struct request_queue *q,
601 dma_drain_needed_fn *dma_drain_needed,
602 void *buf, unsigned int size)
604 if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2)
606 /* make room for appending the drain */
607 blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1);
608 blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1);
609 q->dma_drain_needed = dma_drain_needed;
610 q->dma_drain_buffer = buf;
611 q->dma_drain_size = size;
615 EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
618 * blk_queue_segment_boundary - set boundary rules for segment merging
619 * @q: the request queue for the device
620 * @mask: the memory boundary mask
622 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
624 if (mask < PAGE_CACHE_SIZE - 1) {
625 mask = PAGE_CACHE_SIZE - 1;
626 printk(KERN_INFO "%s: set to minimum %lx\n",
630 q->limits.seg_boundary_mask = mask;
632 EXPORT_SYMBOL(blk_queue_segment_boundary);
635 * blk_queue_dma_alignment - set dma length and memory alignment
636 * @q: the request queue for the device
637 * @mask: alignment mask
640 * set required memory and length alignment for direct dma transactions.
641 * this is used when building direct io requests for the queue.
644 void blk_queue_dma_alignment(struct request_queue *q, int mask)
646 q->dma_alignment = mask;
648 EXPORT_SYMBOL(blk_queue_dma_alignment);
651 * blk_queue_update_dma_alignment - update dma length and memory alignment
652 * @q: the request queue for the device
653 * @mask: alignment mask
656 * update required memory and length alignment for direct dma transactions.
657 * If the requested alignment is larger than the current alignment, then
658 * the current queue alignment is updated to the new value, otherwise it
659 * is left alone. The design of this is to allow multiple objects
660 * (driver, device, transport etc) to set their respective
661 * alignments without having them interfere.
664 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
666 BUG_ON(mask > PAGE_SIZE);
668 if (mask > q->dma_alignment)
669 q->dma_alignment = mask;
671 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
673 static int __init blk_settings_init(void)
675 blk_max_low_pfn = max_low_pfn - 1;
676 blk_max_pfn = max_pfn - 1;
679 subsys_initcall(blk_settings_init);