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_set_default_limits - reset limits to default values
100 * @lim: the queue_limits structure to reset
103 * Returns a queue_limit struct to its default state. Can be used by
104 * stacking drivers like DM that stage table swaps and reuse an
105 * existing device queue.
107 void blk_set_default_limits(struct queue_limits
*lim
)
109 lim
->max_phys_segments
= MAX_PHYS_SEGMENTS
;
110 lim
->max_hw_segments
= MAX_HW_SEGMENTS
;
111 lim
->seg_boundary_mask
= BLK_SEG_BOUNDARY_MASK
;
112 lim
->max_segment_size
= MAX_SEGMENT_SIZE
;
113 lim
->max_sectors
= lim
->max_hw_sectors
= SAFE_MAX_SECTORS
;
114 lim
->logical_block_size
= lim
->physical_block_size
= lim
->io_min
= 512;
115 lim
->bounce_pfn
= (unsigned long)(BLK_BOUNCE_ANY
>> PAGE_SHIFT
);
116 lim
->alignment_offset
= 0;
121 EXPORT_SYMBOL(blk_set_default_limits
);
124 * blk_queue_make_request - define an alternate make_request function for a device
125 * @q: the request queue for the device to be affected
126 * @mfn: the alternate make_request function
129 * The normal way for &struct bios to be passed to a device
130 * driver is for them to be collected into requests on a request
131 * queue, and then to allow the device driver to select requests
132 * off that queue when it is ready. This works well for many block
133 * devices. However some block devices (typically virtual devices
134 * such as md or lvm) do not benefit from the processing on the
135 * request queue, and are served best by having the requests passed
136 * directly to them. This can be achieved by providing a function
137 * to blk_queue_make_request().
140 * The driver that does this *must* be able to deal appropriately
141 * with buffers in "highmemory". This can be accomplished by either calling
142 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
143 * blk_queue_bounce() to create a buffer in normal memory.
145 void blk_queue_make_request(struct request_queue
*q
, make_request_fn
*mfn
)
150 q
->nr_requests
= BLKDEV_MAX_RQ
;
152 q
->make_request_fn
= mfn
;
153 blk_queue_dma_alignment(q
, 511);
154 blk_queue_congestion_threshold(q
);
155 q
->nr_batching
= BLK_BATCH_REQ
;
157 q
->unplug_thresh
= 4; /* hmm */
158 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
159 if (q
->unplug_delay
== 0)
162 q
->unplug_timer
.function
= blk_unplug_timeout
;
163 q
->unplug_timer
.data
= (unsigned long)q
;
165 blk_set_default_limits(&q
->limits
);
168 * If the caller didn't supply a lock, fall back to our embedded
172 q
->queue_lock
= &q
->__queue_lock
;
175 * by default assume old behaviour and bounce for any highmem page
177 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
179 EXPORT_SYMBOL(blk_queue_make_request
);
182 * blk_queue_bounce_limit - set bounce buffer limit for queue
183 * @q: the request queue for the device
184 * @dma_mask: the maximum address the device can handle
187 * Different hardware can have different requirements as to what pages
188 * it can do I/O directly to. A low level driver can call
189 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
190 * buffers for doing I/O to pages residing above @dma_mask.
192 void blk_queue_bounce_limit(struct request_queue
*q
, u64 dma_mask
)
194 unsigned long b_pfn
= dma_mask
>> PAGE_SHIFT
;
197 q
->bounce_gfp
= GFP_NOIO
;
198 #if BITS_PER_LONG == 64
200 * Assume anything <= 4GB can be handled by IOMMU. Actually
201 * some IOMMUs can handle everything, but I don't know of a
202 * way to test this here.
204 if (b_pfn
< (min_t(u64
, 0xffffffffUL
, BLK_BOUNCE_HIGH
) >> PAGE_SHIFT
))
206 q
->limits
.bounce_pfn
= max_low_pfn
;
208 if (b_pfn
< blk_max_low_pfn
)
210 q
->limits
.bounce_pfn
= b_pfn
;
213 init_emergency_isa_pool();
214 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
215 q
->limits
.bounce_pfn
= b_pfn
;
218 EXPORT_SYMBOL(blk_queue_bounce_limit
);
221 * blk_queue_max_sectors - set max sectors for a request for this queue
222 * @q: the request queue for the device
223 * @max_sectors: max sectors in the usual 512b unit
226 * Enables a low level driver to set an upper limit on the size of
229 void blk_queue_max_sectors(struct request_queue
*q
, unsigned int max_sectors
)
231 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
232 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
233 printk(KERN_INFO
"%s: set to minimum %d\n",
234 __func__
, max_sectors
);
237 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
238 q
->limits
.max_hw_sectors
= q
->limits
.max_sectors
= max_sectors
;
240 q
->limits
.max_sectors
= BLK_DEF_MAX_SECTORS
;
241 q
->limits
.max_hw_sectors
= max_sectors
;
244 EXPORT_SYMBOL(blk_queue_max_sectors
);
246 void blk_queue_max_hw_sectors(struct request_queue
*q
, unsigned int max_sectors
)
248 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
249 q
->limits
.max_hw_sectors
= BLK_DEF_MAX_SECTORS
;
251 q
->limits
.max_hw_sectors
= max_sectors
;
253 EXPORT_SYMBOL(blk_queue_max_hw_sectors
);
256 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
257 * @q: the request queue for the device
258 * @max_segments: max number of segments
261 * Enables a low level driver to set an upper limit on the number of
262 * physical data segments in a request. This would be the largest sized
263 * scatter list the driver could handle.
265 void blk_queue_max_phys_segments(struct request_queue
*q
,
266 unsigned short max_segments
)
270 printk(KERN_INFO
"%s: set to minimum %d\n",
271 __func__
, max_segments
);
274 q
->limits
.max_phys_segments
= max_segments
;
276 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
279 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
280 * @q: the request queue for the device
281 * @max_segments: max number of segments
284 * Enables a low level driver to set an upper limit on the number of
285 * hw data segments in a request. This would be the largest number of
286 * address/length pairs the host adapter can actually give at once
289 void blk_queue_max_hw_segments(struct request_queue
*q
,
290 unsigned short max_segments
)
294 printk(KERN_INFO
"%s: set to minimum %d\n",
295 __func__
, max_segments
);
298 q
->limits
.max_hw_segments
= max_segments
;
300 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
303 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
304 * @q: the request queue for the device
305 * @max_size: max size of segment in bytes
308 * Enables a low level driver to set an upper limit on the size of a
311 void blk_queue_max_segment_size(struct request_queue
*q
, unsigned int max_size
)
313 if (max_size
< PAGE_CACHE_SIZE
) {
314 max_size
= PAGE_CACHE_SIZE
;
315 printk(KERN_INFO
"%s: set to minimum %d\n",
319 q
->limits
.max_segment_size
= max_size
;
321 EXPORT_SYMBOL(blk_queue_max_segment_size
);
324 * blk_queue_logical_block_size - set logical block size for the queue
325 * @q: the request queue for the device
326 * @size: the logical block size, in bytes
329 * This should be set to the lowest possible block size that the
330 * storage device can address. The default of 512 covers most
333 void blk_queue_logical_block_size(struct request_queue
*q
, unsigned short size
)
335 q
->limits
.logical_block_size
= size
;
337 if (q
->limits
.physical_block_size
< size
)
338 q
->limits
.physical_block_size
= size
;
340 if (q
->limits
.io_min
< q
->limits
.physical_block_size
)
341 q
->limits
.io_min
= q
->limits
.physical_block_size
;
343 EXPORT_SYMBOL(blk_queue_logical_block_size
);
346 * blk_queue_physical_block_size - set physical block size for the queue
347 * @q: the request queue for the device
348 * @size: the physical block size, in bytes
351 * This should be set to the lowest possible sector size that the
352 * hardware can operate on without reverting to read-modify-write
355 void blk_queue_physical_block_size(struct request_queue
*q
, unsigned short size
)
357 q
->limits
.physical_block_size
= size
;
359 if (q
->limits
.physical_block_size
< q
->limits
.logical_block_size
)
360 q
->limits
.physical_block_size
= q
->limits
.logical_block_size
;
362 if (q
->limits
.io_min
< q
->limits
.physical_block_size
)
363 q
->limits
.io_min
= q
->limits
.physical_block_size
;
365 EXPORT_SYMBOL(blk_queue_physical_block_size
);
368 * blk_queue_alignment_offset - set physical block alignment offset
369 * @q: the request queue for the device
370 * @offset: alignment offset in bytes
373 * Some devices are naturally misaligned to compensate for things like
374 * the legacy DOS partition table 63-sector offset. Low-level drivers
375 * should call this function for devices whose first sector is not
378 void blk_queue_alignment_offset(struct request_queue
*q
, unsigned int offset
)
380 q
->limits
.alignment_offset
=
381 offset
& (q
->limits
.physical_block_size
- 1);
382 q
->limits
.misaligned
= 0;
384 EXPORT_SYMBOL(blk_queue_alignment_offset
);
387 * blk_limits_io_min - set minimum request size for a device
388 * @limits: the queue limits
389 * @min: smallest I/O size in bytes
392 * Some devices have an internal block size bigger than the reported
393 * hardware sector size. This function can be used to signal the
394 * smallest I/O the device can perform without incurring a performance
397 void blk_limits_io_min(struct queue_limits
*limits
, unsigned int min
)
399 limits
->io_min
= min
;
401 if (limits
->io_min
< limits
->logical_block_size
)
402 limits
->io_min
= limits
->logical_block_size
;
404 if (limits
->io_min
< limits
->physical_block_size
)
405 limits
->io_min
= limits
->physical_block_size
;
407 EXPORT_SYMBOL(blk_limits_io_min
);
410 * blk_queue_io_min - set minimum request size for the queue
411 * @q: the request queue for the device
412 * @min: smallest I/O size in bytes
415 * Some devices have an internal block size bigger than the reported
416 * hardware sector size. This function can be used to signal the
417 * smallest I/O the device can perform without incurring a performance
420 void blk_queue_io_min(struct request_queue
*q
, unsigned int min
)
422 blk_limits_io_min(&q
->limits
, min
);
424 EXPORT_SYMBOL(blk_queue_io_min
);
427 * blk_queue_io_opt - set optimal request size for the queue
428 * @q: the request queue for the device
429 * @opt: optimal request size in bytes
432 * Drivers can call this function to set the preferred I/O request
433 * size for devices that report such a value.
435 void blk_queue_io_opt(struct request_queue
*q
, unsigned int opt
)
437 q
->limits
.io_opt
= opt
;
439 EXPORT_SYMBOL(blk_queue_io_opt
);
442 * Returns the minimum that is _not_ zero, unless both are zero.
444 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
447 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
448 * @t: the stacking driver (top)
449 * @b: the underlying device (bottom)
451 void blk_queue_stack_limits(struct request_queue
*t
, struct request_queue
*b
)
453 blk_stack_limits(&t
->limits
, &b
->limits
, 0);
457 else if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
)) {
459 spin_lock_irqsave(t
->queue_lock
, flags
);
460 queue_flag_clear(QUEUE_FLAG_CLUSTER
, t
);
461 spin_unlock_irqrestore(t
->queue_lock
, flags
);
464 EXPORT_SYMBOL(blk_queue_stack_limits
);
467 * blk_stack_limits - adjust queue_limits for stacked devices
468 * @t: the stacking driver limits (top)
469 * @b: the underlying queue limits (bottom)
470 * @offset: offset to beginning of data within component device
473 * Merges two queue_limit structs. Returns 0 if alignment didn't
474 * change. Returns -1 if adding the bottom device caused
477 int blk_stack_limits(struct queue_limits
*t
, struct queue_limits
*b
,
480 t
->max_sectors
= min_not_zero(t
->max_sectors
, b
->max_sectors
);
481 t
->max_hw_sectors
= min_not_zero(t
->max_hw_sectors
, b
->max_hw_sectors
);
482 t
->bounce_pfn
= min_not_zero(t
->bounce_pfn
, b
->bounce_pfn
);
484 t
->seg_boundary_mask
= min_not_zero(t
->seg_boundary_mask
,
485 b
->seg_boundary_mask
);
487 t
->max_phys_segments
= min_not_zero(t
->max_phys_segments
,
488 b
->max_phys_segments
);
490 t
->max_hw_segments
= min_not_zero(t
->max_hw_segments
,
493 t
->max_segment_size
= min_not_zero(t
->max_segment_size
,
494 b
->max_segment_size
);
496 t
->logical_block_size
= max(t
->logical_block_size
,
497 b
->logical_block_size
);
499 t
->physical_block_size
= max(t
->physical_block_size
,
500 b
->physical_block_size
);
502 t
->io_min
= max(t
->io_min
, b
->io_min
);
503 t
->no_cluster
|= b
->no_cluster
;
505 /* Bottom device offset aligned? */
507 (offset
& (b
->physical_block_size
- 1)) != b
->alignment_offset
) {
512 /* If top has no alignment offset, inherit from bottom */
513 if (!t
->alignment_offset
)
514 t
->alignment_offset
=
515 b
->alignment_offset
& (b
->physical_block_size
- 1);
517 /* Top device aligned on logical block boundary? */
518 if (t
->alignment_offset
& (t
->logical_block_size
- 1)) {
525 EXPORT_SYMBOL(blk_stack_limits
);
528 * disk_stack_limits - adjust queue limits for stacked drivers
529 * @disk: MD/DM gendisk (top)
530 * @bdev: the underlying block device (bottom)
531 * @offset: offset to beginning of data within component device
534 * Merges the limits for two queues. Returns 0 if alignment
535 * didn't change. Returns -1 if adding the bottom device caused
538 void disk_stack_limits(struct gendisk
*disk
, struct block_device
*bdev
,
541 struct request_queue
*t
= disk
->queue
;
542 struct request_queue
*b
= bdev_get_queue(bdev
);
544 offset
+= get_start_sect(bdev
) << 9;
546 if (blk_stack_limits(&t
->limits
, &b
->limits
, offset
) < 0) {
547 char top
[BDEVNAME_SIZE
], bottom
[BDEVNAME_SIZE
];
549 disk_name(disk
, 0, top
);
550 bdevname(bdev
, bottom
);
552 printk(KERN_NOTICE
"%s: Warning: Device %s is misaligned\n",
558 else if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
)) {
561 spin_lock_irqsave(t
->queue_lock
, flags
);
562 if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
))
563 queue_flag_clear(QUEUE_FLAG_CLUSTER
, t
);
564 spin_unlock_irqrestore(t
->queue_lock
, flags
);
567 EXPORT_SYMBOL(disk_stack_limits
);
570 * blk_queue_dma_pad - set pad mask
571 * @q: the request queue for the device
576 * Appending pad buffer to a request modifies the last entry of a
577 * scatter list such that it includes the pad buffer.
579 void blk_queue_dma_pad(struct request_queue
*q
, unsigned int mask
)
581 q
->dma_pad_mask
= mask
;
583 EXPORT_SYMBOL(blk_queue_dma_pad
);
586 * blk_queue_update_dma_pad - update pad mask
587 * @q: the request queue for the device
590 * Update dma pad mask.
592 * Appending pad buffer to a request modifies the last entry of a
593 * scatter list such that it includes the pad buffer.
595 void blk_queue_update_dma_pad(struct request_queue
*q
, unsigned int mask
)
597 if (mask
> q
->dma_pad_mask
)
598 q
->dma_pad_mask
= mask
;
600 EXPORT_SYMBOL(blk_queue_update_dma_pad
);
603 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
604 * @q: the request queue for the device
605 * @dma_drain_needed: fn which returns non-zero if drain is necessary
606 * @buf: physically contiguous buffer
607 * @size: size of the buffer in bytes
609 * Some devices have excess DMA problems and can't simply discard (or
610 * zero fill) the unwanted piece of the transfer. They have to have a
611 * real area of memory to transfer it into. The use case for this is
612 * ATAPI devices in DMA mode. If the packet command causes a transfer
613 * bigger than the transfer size some HBAs will lock up if there
614 * aren't DMA elements to contain the excess transfer. What this API
615 * does is adjust the queue so that the buf is always appended
616 * silently to the scatterlist.
618 * Note: This routine adjusts max_hw_segments to make room for
619 * appending the drain buffer. If you call
620 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
621 * calling this routine, you must set the limit to one fewer than your
622 * device can support otherwise there won't be room for the drain
625 int blk_queue_dma_drain(struct request_queue
*q
,
626 dma_drain_needed_fn
*dma_drain_needed
,
627 void *buf
, unsigned int size
)
629 if (queue_max_hw_segments(q
) < 2 || queue_max_phys_segments(q
) < 2)
631 /* make room for appending the drain */
632 blk_queue_max_hw_segments(q
, queue_max_hw_segments(q
) - 1);
633 blk_queue_max_phys_segments(q
, queue_max_phys_segments(q
) - 1);
634 q
->dma_drain_needed
= dma_drain_needed
;
635 q
->dma_drain_buffer
= buf
;
636 q
->dma_drain_size
= size
;
640 EXPORT_SYMBOL_GPL(blk_queue_dma_drain
);
643 * blk_queue_segment_boundary - set boundary rules for segment merging
644 * @q: the request queue for the device
645 * @mask: the memory boundary mask
647 void blk_queue_segment_boundary(struct request_queue
*q
, unsigned long mask
)
649 if (mask
< PAGE_CACHE_SIZE
- 1) {
650 mask
= PAGE_CACHE_SIZE
- 1;
651 printk(KERN_INFO
"%s: set to minimum %lx\n",
655 q
->limits
.seg_boundary_mask
= mask
;
657 EXPORT_SYMBOL(blk_queue_segment_boundary
);
660 * blk_queue_dma_alignment - set dma length and memory alignment
661 * @q: the request queue for the device
662 * @mask: alignment mask
665 * set required memory and length alignment for direct dma transactions.
666 * this is used when building direct io requests for the queue.
669 void blk_queue_dma_alignment(struct request_queue
*q
, int mask
)
671 q
->dma_alignment
= mask
;
673 EXPORT_SYMBOL(blk_queue_dma_alignment
);
676 * blk_queue_update_dma_alignment - update dma length and memory alignment
677 * @q: the request queue for the device
678 * @mask: alignment mask
681 * update required memory and length alignment for direct dma transactions.
682 * If the requested alignment is larger than the current alignment, then
683 * the current queue alignment is updated to the new value, otherwise it
684 * is left alone. The design of this is to allow multiple objects
685 * (driver, device, transport etc) to set their respective
686 * alignments without having them interfere.
689 void blk_queue_update_dma_alignment(struct request_queue
*q
, int mask
)
691 BUG_ON(mask
> PAGE_SIZE
);
693 if (mask
> q
->dma_alignment
)
694 q
->dma_alignment
= mask
;
696 EXPORT_SYMBOL(blk_queue_update_dma_alignment
);
698 static int __init
blk_settings_init(void)
700 blk_max_low_pfn
= max_low_pfn
- 1;
701 blk_max_pfn
= max_pfn
- 1;
704 subsys_initcall(blk_settings_init
);