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 q->backing_dev_info.ra_pages =
133 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
134 q->backing_dev_info.state = 0;
135 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
136 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
137 blk_queue_logical_block_size(q, 512);
138 blk_queue_dma_alignment(q, 511);
139 blk_queue_congestion_threshold(q);
140 q->nr_batching = BLK_BATCH_REQ;
142 q->unplug_thresh = 4; /* hmm */
143 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
144 if (q->unplug_delay == 0)
147 q->unplug_timer.function = blk_unplug_timeout;
148 q->unplug_timer.data = (unsigned long)q;
151 * by default assume old behaviour and bounce for any highmem page
153 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
155 EXPORT_SYMBOL(blk_queue_make_request);
158 * blk_queue_bounce_limit - set bounce buffer limit for queue
159 * @q: the request queue for the device
160 * @dma_mask: the maximum address the device can handle
163 * Different hardware can have different requirements as to what pages
164 * it can do I/O directly to. A low level driver can call
165 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
166 * buffers for doing I/O to pages residing above @dma_mask.
168 void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
170 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
173 q->bounce_gfp = GFP_NOIO;
174 #if BITS_PER_LONG == 64
176 * Assume anything <= 4GB can be handled by IOMMU. Actually
177 * some IOMMUs can handle everything, but I don't know of a
178 * way to test this here.
180 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
182 q->limits.bounce_pfn = max_low_pfn;
184 if (b_pfn < blk_max_low_pfn)
186 q->limits.bounce_pfn = b_pfn;
189 init_emergency_isa_pool();
190 q->bounce_gfp = GFP_NOIO | GFP_DMA;
191 q->limits.bounce_pfn = b_pfn;
194 EXPORT_SYMBOL(blk_queue_bounce_limit);
197 * blk_queue_bounce_pfn - set the bounce buffer limit for queue
198 * @q: the request queue for the device
202 * This function is similar to blk_queue_bounce_limit except it
203 * neither changes allocation flags, nor does it set up the ISA DMA
204 * pool. This function should only be used by stacking drivers.
205 * Hardware drivers should use blk_queue_bounce_limit instead.
207 void blk_queue_bounce_pfn(struct request_queue *q, u64 pfn)
209 q->limits.bounce_pfn = pfn;
211 EXPORT_SYMBOL(blk_queue_bounce_pfn);
214 * blk_queue_max_sectors - set max sectors for a request for this queue
215 * @q: the request queue for the device
216 * @max_sectors: max sectors in the usual 512b unit
219 * Enables a low level driver to set an upper limit on the size of
222 void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
224 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
225 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
226 printk(KERN_INFO "%s: set to minimum %d\n",
227 __func__, max_sectors);
230 if (BLK_DEF_MAX_SECTORS > max_sectors)
231 q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors;
233 q->limits.max_sectors = BLK_DEF_MAX_SECTORS;
234 q->limits.max_hw_sectors = max_sectors;
237 EXPORT_SYMBOL(blk_queue_max_sectors);
239 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors)
241 if (BLK_DEF_MAX_SECTORS > max_sectors)
242 q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS;
244 q->limits.max_hw_sectors = max_sectors;
246 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
249 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
250 * @q: the request queue for the device
251 * @max_segments: max number of segments
254 * Enables a low level driver to set an upper limit on the number of
255 * physical data segments in a request. This would be the largest sized
256 * scatter list the driver could handle.
258 void blk_queue_max_phys_segments(struct request_queue *q,
259 unsigned short max_segments)
263 printk(KERN_INFO "%s: set to minimum %d\n",
264 __func__, max_segments);
267 q->limits.max_phys_segments = max_segments;
269 EXPORT_SYMBOL(blk_queue_max_phys_segments);
272 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
273 * @q: the request queue for the device
274 * @max_segments: max number of segments
277 * Enables a low level driver to set an upper limit on the number of
278 * hw data segments in a request. This would be the largest number of
279 * address/length pairs the host adapter can actually give at once
282 void blk_queue_max_hw_segments(struct request_queue *q,
283 unsigned short max_segments)
287 printk(KERN_INFO "%s: set to minimum %d\n",
288 __func__, max_segments);
291 q->limits.max_hw_segments = max_segments;
293 EXPORT_SYMBOL(blk_queue_max_hw_segments);
296 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
297 * @q: the request queue for the device
298 * @max_size: max size of segment in bytes
301 * Enables a low level driver to set an upper limit on the size of a
304 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
306 if (max_size < PAGE_CACHE_SIZE) {
307 max_size = PAGE_CACHE_SIZE;
308 printk(KERN_INFO "%s: set to minimum %d\n",
312 q->limits.max_segment_size = max_size;
314 EXPORT_SYMBOL(blk_queue_max_segment_size);
317 * blk_queue_logical_block_size - set logical block size for the queue
318 * @q: the request queue for the device
319 * @size: the logical block size, in bytes
322 * This should be set to the lowest possible block size that the
323 * storage device can address. The default of 512 covers most
326 void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
328 q->limits.logical_block_size = size;
330 if (q->limits.physical_block_size < size)
331 q->limits.physical_block_size = size;
333 if (q->limits.io_min < q->limits.physical_block_size)
334 q->limits.io_min = q->limits.physical_block_size;
336 EXPORT_SYMBOL(blk_queue_logical_block_size);
339 * blk_queue_physical_block_size - set physical block size for the queue
340 * @q: the request queue for the device
341 * @size: the physical block size, in bytes
344 * This should be set to the lowest possible sector size that the
345 * hardware can operate on without reverting to read-modify-write
348 void blk_queue_physical_block_size(struct request_queue *q, unsigned short size)
350 q->limits.physical_block_size = size;
352 if (q->limits.physical_block_size < q->limits.logical_block_size)
353 q->limits.physical_block_size = q->limits.logical_block_size;
355 if (q->limits.io_min < q->limits.physical_block_size)
356 q->limits.io_min = q->limits.physical_block_size;
358 EXPORT_SYMBOL(blk_queue_physical_block_size);
361 * blk_queue_alignment_offset - set physical block alignment offset
362 * @q: the request queue for the device
363 * @alignment: alignment offset in bytes
366 * Some devices are naturally misaligned to compensate for things like
367 * the legacy DOS partition table 63-sector offset. Low-level drivers
368 * should call this function for devices whose first sector is not
371 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
373 q->limits.alignment_offset =
374 offset & (q->limits.physical_block_size - 1);
375 q->limits.misaligned = 0;
377 EXPORT_SYMBOL(blk_queue_alignment_offset);
380 * blk_queue_io_min - set minimum request size for the queue
381 * @q: the request queue for the device
382 * @io_min: smallest I/O size in bytes
385 * Some devices have an internal block size bigger than the reported
386 * hardware sector size. This function can be used to signal the
387 * smallest I/O the device can perform without incurring a performance
390 void blk_queue_io_min(struct request_queue *q, unsigned int min)
392 q->limits.io_min = min;
394 if (q->limits.io_min < q->limits.logical_block_size)
395 q->limits.io_min = q->limits.logical_block_size;
397 if (q->limits.io_min < q->limits.physical_block_size)
398 q->limits.io_min = q->limits.physical_block_size;
400 EXPORT_SYMBOL(blk_queue_io_min);
403 * blk_queue_io_opt - set optimal request size for the queue
404 * @q: the request queue for the device
405 * @io_opt: optimal request size in bytes
408 * Drivers can call this function to set the preferred I/O request
409 * size for devices that report such a value.
411 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
413 q->limits.io_opt = opt;
415 EXPORT_SYMBOL(blk_queue_io_opt);
418 * Returns the minimum that is _not_ zero, unless both are zero.
420 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
423 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
424 * @t: the stacking driver (top)
425 * @b: the underlying device (bottom)
427 void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
429 /* zero is "infinity" */
430 t->limits.max_sectors = min_not_zero(queue_max_sectors(t),
431 queue_max_sectors(b));
433 t->limits.max_hw_sectors = min_not_zero(queue_max_hw_sectors(t),
434 queue_max_hw_sectors(b));
436 t->limits.seg_boundary_mask = min_not_zero(queue_segment_boundary(t),
437 queue_segment_boundary(b));
439 t->limits.max_phys_segments = min_not_zero(queue_max_phys_segments(t),
440 queue_max_phys_segments(b));
442 t->limits.max_hw_segments = min_not_zero(queue_max_hw_segments(t),
443 queue_max_hw_segments(b));
445 t->limits.max_segment_size = min_not_zero(queue_max_segment_size(t),
446 queue_max_segment_size(b));
448 t->limits.logical_block_size = max(queue_logical_block_size(t),
449 queue_logical_block_size(b));
453 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
455 spin_lock_irqsave(t->queue_lock, flags);
456 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
457 spin_unlock_irqrestore(t->queue_lock, flags);
460 EXPORT_SYMBOL(blk_queue_stack_limits);
463 * blk_stack_limits - adjust queue_limits for stacked devices
464 * @t: the stacking driver limits (top)
465 * @bdev: the underlying queue limits (bottom)
466 * @offset: offset to beginning of data within component device
469 * Merges two queue_limit structs. Returns 0 if alignment didn't
470 * change. Returns -1 if adding the bottom device caused
473 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
476 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
477 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
479 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
480 b->seg_boundary_mask);
482 t->max_phys_segments = min_not_zero(t->max_phys_segments,
483 b->max_phys_segments);
485 t->max_hw_segments = min_not_zero(t->max_hw_segments,
488 t->max_segment_size = min_not_zero(t->max_segment_size,
489 b->max_segment_size);
491 t->logical_block_size = max(t->logical_block_size,
492 b->logical_block_size);
494 t->physical_block_size = max(t->physical_block_size,
495 b->physical_block_size);
497 t->io_min = max(t->io_min, b->io_min);
498 t->no_cluster |= b->no_cluster;
500 /* Bottom device offset aligned? */
502 (offset & (b->physical_block_size - 1)) != b->alignment_offset) {
507 /* If top has no alignment offset, inherit from bottom */
508 if (!t->alignment_offset)
509 t->alignment_offset =
510 b->alignment_offset & (b->physical_block_size - 1);
512 /* Top device aligned on logical block boundary? */
513 if (t->alignment_offset & (t->logical_block_size - 1)) {
520 EXPORT_SYMBOL(blk_stack_limits);
523 * disk_stack_limits - adjust queue limits for stacked drivers
524 * @t: MD/DM gendisk (top)
525 * @bdev: the underlying block device (bottom)
526 * @offset: offset to beginning of data within component device
529 * Merges the limits for two queues. Returns 0 if alignment
530 * didn't change. Returns -1 if adding the bottom device caused
533 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
536 struct request_queue *t = disk->queue;
537 struct request_queue *b = bdev_get_queue(bdev);
539 offset += get_start_sect(bdev) << 9;
541 if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) {
542 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
544 disk_name(disk, 0, top);
545 bdevname(bdev, bottom);
547 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
553 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
556 spin_lock_irqsave(t->queue_lock, flags);
557 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
558 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
559 spin_unlock_irqrestore(t->queue_lock, flags);
562 EXPORT_SYMBOL(disk_stack_limits);
565 * blk_queue_dma_pad - set pad mask
566 * @q: the request queue for the device
571 * Appending pad buffer to a request modifies the last entry of a
572 * scatter list such that it includes the pad buffer.
574 void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
576 q->dma_pad_mask = mask;
578 EXPORT_SYMBOL(blk_queue_dma_pad);
581 * blk_queue_update_dma_pad - update pad mask
582 * @q: the request queue for the device
585 * Update dma pad mask.
587 * Appending pad buffer to a request modifies the last entry of a
588 * scatter list such that it includes the pad buffer.
590 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
592 if (mask > q->dma_pad_mask)
593 q->dma_pad_mask = mask;
595 EXPORT_SYMBOL(blk_queue_update_dma_pad);
598 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
599 * @q: the request queue for the device
600 * @dma_drain_needed: fn which returns non-zero if drain is necessary
601 * @buf: physically contiguous buffer
602 * @size: size of the buffer in bytes
604 * Some devices have excess DMA problems and can't simply discard (or
605 * zero fill) the unwanted piece of the transfer. They have to have a
606 * real area of memory to transfer it into. The use case for this is
607 * ATAPI devices in DMA mode. If the packet command causes a transfer
608 * bigger than the transfer size some HBAs will lock up if there
609 * aren't DMA elements to contain the excess transfer. What this API
610 * does is adjust the queue so that the buf is always appended
611 * silently to the scatterlist.
613 * Note: This routine adjusts max_hw_segments to make room for
614 * appending the drain buffer. If you call
615 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
616 * calling this routine, you must set the limit to one fewer than your
617 * device can support otherwise there won't be room for the drain
620 int blk_queue_dma_drain(struct request_queue *q,
621 dma_drain_needed_fn *dma_drain_needed,
622 void *buf, unsigned int size)
624 if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2)
626 /* make room for appending the drain */
627 blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1);
628 blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1);
629 q->dma_drain_needed = dma_drain_needed;
630 q->dma_drain_buffer = buf;
631 q->dma_drain_size = size;
635 EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
638 * blk_queue_segment_boundary - set boundary rules for segment merging
639 * @q: the request queue for the device
640 * @mask: the memory boundary mask
642 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
644 if (mask < PAGE_CACHE_SIZE - 1) {
645 mask = PAGE_CACHE_SIZE - 1;
646 printk(KERN_INFO "%s: set to minimum %lx\n",
650 q->limits.seg_boundary_mask = mask;
652 EXPORT_SYMBOL(blk_queue_segment_boundary);
655 * blk_queue_dma_alignment - set dma length and memory alignment
656 * @q: the request queue for the device
657 * @mask: alignment mask
660 * set required memory and length alignment for direct dma transactions.
661 * this is used when building direct io requests for the queue.
664 void blk_queue_dma_alignment(struct request_queue *q, int mask)
666 q->dma_alignment = mask;
668 EXPORT_SYMBOL(blk_queue_dma_alignment);
671 * blk_queue_update_dma_alignment - update dma length and memory alignment
672 * @q: the request queue for the device
673 * @mask: alignment mask
676 * update required memory and length alignment for direct dma transactions.
677 * If the requested alignment is larger than the current alignment, then
678 * the current queue alignment is updated to the new value, otherwise it
679 * is left alone. The design of this is to allow multiple objects
680 * (driver, device, transport etc) to set their respective
681 * alignments without having them interfere.
684 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
686 BUG_ON(mask > PAGE_SIZE);
688 if (mask > q->dma_alignment)
689 q->dma_alignment = mask;
691 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
693 static int __init blk_settings_init(void)
695 blk_max_low_pfn = max_low_pfn - 1;
696 blk_max_pfn = max_pfn - 1;
699 subsys_initcall(blk_settings_init);