Loading...
1/*
2 * Functions related to setting various queue properties from drivers
3 */
4#include <linux/kernel.h>
5#include <linux/module.h>
6#include <linux/init.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
10#include <linux/gcd.h>
11#include <linux/lcm.h>
12#include <linux/jiffies.h>
13#include <linux/gfp.h>
14
15#include "blk.h"
16
17unsigned long blk_max_low_pfn;
18EXPORT_SYMBOL(blk_max_low_pfn);
19
20unsigned long blk_max_pfn;
21
22/**
23 * blk_queue_prep_rq - set a prepare_request function for queue
24 * @q: queue
25 * @pfn: prepare_request function
26 *
27 * It's possible for a queue to register a prepare_request callback which
28 * is invoked before the request is handed to the request_fn. The goal of
29 * the function is to prepare a request for I/O, it can be used to build a
30 * cdb from the request data for instance.
31 *
32 */
33void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
34{
35 q->prep_rq_fn = pfn;
36}
37EXPORT_SYMBOL(blk_queue_prep_rq);
38
39/**
40 * blk_queue_unprep_rq - set an unprepare_request function for queue
41 * @q: queue
42 * @ufn: unprepare_request function
43 *
44 * It's possible for a queue to register an unprepare_request callback
45 * which is invoked before the request is finally completed. The goal
46 * of the function is to deallocate any data that was allocated in the
47 * prepare_request callback.
48 *
49 */
50void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn)
51{
52 q->unprep_rq_fn = ufn;
53}
54EXPORT_SYMBOL(blk_queue_unprep_rq);
55
56/**
57 * blk_queue_merge_bvec - set a merge_bvec function for queue
58 * @q: queue
59 * @mbfn: merge_bvec_fn
60 *
61 * Usually queues have static limitations on the max sectors or segments that
62 * we can put in a request. Stacking drivers may have some settings that
63 * are dynamic, and thus we have to query the queue whether it is ok to
64 * add a new bio_vec to a bio at a given offset or not. If the block device
65 * has such limitations, it needs to register a merge_bvec_fn to control
66 * the size of bio's sent to it. Note that a block device *must* allow a
67 * single page to be added to an empty bio. The block device driver may want
68 * to use the bio_split() function to deal with these bio's. By default
69 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
70 * honored.
71 */
72void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
73{
74 q->merge_bvec_fn = mbfn;
75}
76EXPORT_SYMBOL(blk_queue_merge_bvec);
77
78void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
79{
80 q->softirq_done_fn = fn;
81}
82EXPORT_SYMBOL(blk_queue_softirq_done);
83
84void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
85{
86 q->rq_timeout = timeout;
87}
88EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
89
90void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
91{
92 q->rq_timed_out_fn = fn;
93}
94EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
95
96void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
97{
98 q->lld_busy_fn = fn;
99}
100EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
101
102/**
103 * blk_set_default_limits - reset limits to default values
104 * @lim: the queue_limits structure to reset
105 *
106 * Description:
107 * Returns a queue_limit struct to its default state.
108 */
109void blk_set_default_limits(struct queue_limits *lim)
110{
111 lim->max_segments = BLK_MAX_SEGMENTS;
112 lim->max_integrity_segments = 0;
113 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
114 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
115 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
116 lim->max_discard_sectors = 0;
117 lim->discard_granularity = 0;
118 lim->discard_alignment = 0;
119 lim->discard_misaligned = 0;
120 lim->discard_zeroes_data = 0;
121 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
122 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
123 lim->alignment_offset = 0;
124 lim->io_opt = 0;
125 lim->misaligned = 0;
126 lim->cluster = 1;
127}
128EXPORT_SYMBOL(blk_set_default_limits);
129
130/**
131 * blk_set_stacking_limits - set default limits for stacking devices
132 * @lim: the queue_limits structure to reset
133 *
134 * Description:
135 * Returns a queue_limit struct to its default state. Should be used
136 * by stacking drivers like DM that have no internal limits.
137 */
138void blk_set_stacking_limits(struct queue_limits *lim)
139{
140 blk_set_default_limits(lim);
141
142 /* Inherit limits from component devices */
143 lim->discard_zeroes_data = 1;
144 lim->max_segments = USHRT_MAX;
145 lim->max_hw_sectors = UINT_MAX;
146
147 lim->max_sectors = BLK_DEF_MAX_SECTORS;
148}
149EXPORT_SYMBOL(blk_set_stacking_limits);
150
151/**
152 * blk_queue_make_request - define an alternate make_request function for a device
153 * @q: the request queue for the device to be affected
154 * @mfn: the alternate make_request function
155 *
156 * Description:
157 * The normal way for &struct bios to be passed to a device
158 * driver is for them to be collected into requests on a request
159 * queue, and then to allow the device driver to select requests
160 * off that queue when it is ready. This works well for many block
161 * devices. However some block devices (typically virtual devices
162 * such as md or lvm) do not benefit from the processing on the
163 * request queue, and are served best by having the requests passed
164 * directly to them. This can be achieved by providing a function
165 * to blk_queue_make_request().
166 *
167 * Caveat:
168 * The driver that does this *must* be able to deal appropriately
169 * with buffers in "highmemory". This can be accomplished by either calling
170 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
171 * blk_queue_bounce() to create a buffer in normal memory.
172 **/
173void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
174{
175 /*
176 * set defaults
177 */
178 q->nr_requests = BLKDEV_MAX_RQ;
179
180 q->make_request_fn = mfn;
181 blk_queue_dma_alignment(q, 511);
182 blk_queue_congestion_threshold(q);
183 q->nr_batching = BLK_BATCH_REQ;
184
185 blk_set_default_limits(&q->limits);
186
187 /*
188 * by default assume old behaviour and bounce for any highmem page
189 */
190 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
191}
192EXPORT_SYMBOL(blk_queue_make_request);
193
194/**
195 * blk_queue_bounce_limit - set bounce buffer limit for queue
196 * @q: the request queue for the device
197 * @dma_mask: the maximum address the device can handle
198 *
199 * Description:
200 * Different hardware can have different requirements as to what pages
201 * it can do I/O directly to. A low level driver can call
202 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
203 * buffers for doing I/O to pages residing above @dma_mask.
204 **/
205void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
206{
207 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
208 int dma = 0;
209
210 q->bounce_gfp = GFP_NOIO;
211#if BITS_PER_LONG == 64
212 /*
213 * Assume anything <= 4GB can be handled by IOMMU. Actually
214 * some IOMMUs can handle everything, but I don't know of a
215 * way to test this here.
216 */
217 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
218 dma = 1;
219 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
220#else
221 if (b_pfn < blk_max_low_pfn)
222 dma = 1;
223 q->limits.bounce_pfn = b_pfn;
224#endif
225 if (dma) {
226 init_emergency_isa_pool();
227 q->bounce_gfp = GFP_NOIO | GFP_DMA;
228 q->limits.bounce_pfn = b_pfn;
229 }
230}
231EXPORT_SYMBOL(blk_queue_bounce_limit);
232
233/**
234 * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request
235 * @limits: the queue limits
236 * @max_hw_sectors: max hardware sectors in the usual 512b unit
237 *
238 * Description:
239 * Enables a low level driver to set a hard upper limit,
240 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
241 * the device driver based upon the combined capabilities of I/O
242 * controller and storage device.
243 *
244 * max_sectors is a soft limit imposed by the block layer for
245 * filesystem type requests. This value can be overridden on a
246 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
247 * The soft limit can not exceed max_hw_sectors.
248 **/
249void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors)
250{
251 if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
252 max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
253 printk(KERN_INFO "%s: set to minimum %d\n",
254 __func__, max_hw_sectors);
255 }
256
257 limits->max_hw_sectors = max_hw_sectors;
258 limits->max_sectors = min_t(unsigned int, max_hw_sectors,
259 BLK_DEF_MAX_SECTORS);
260}
261EXPORT_SYMBOL(blk_limits_max_hw_sectors);
262
263/**
264 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
265 * @q: the request queue for the device
266 * @max_hw_sectors: max hardware sectors in the usual 512b unit
267 *
268 * Description:
269 * See description for blk_limits_max_hw_sectors().
270 **/
271void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
272{
273 blk_limits_max_hw_sectors(&q->limits, max_hw_sectors);
274}
275EXPORT_SYMBOL(blk_queue_max_hw_sectors);
276
277/**
278 * blk_queue_max_discard_sectors - set max sectors for a single discard
279 * @q: the request queue for the device
280 * @max_discard_sectors: maximum number of sectors to discard
281 **/
282void blk_queue_max_discard_sectors(struct request_queue *q,
283 unsigned int max_discard_sectors)
284{
285 q->limits.max_discard_sectors = max_discard_sectors;
286}
287EXPORT_SYMBOL(blk_queue_max_discard_sectors);
288
289/**
290 * blk_queue_max_segments - set max hw segments for a request for this queue
291 * @q: the request queue for the device
292 * @max_segments: max number of segments
293 *
294 * Description:
295 * Enables a low level driver to set an upper limit on the number of
296 * hw data segments in a request.
297 **/
298void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
299{
300 if (!max_segments) {
301 max_segments = 1;
302 printk(KERN_INFO "%s: set to minimum %d\n",
303 __func__, max_segments);
304 }
305
306 q->limits.max_segments = max_segments;
307}
308EXPORT_SYMBOL(blk_queue_max_segments);
309
310/**
311 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
312 * @q: the request queue for the device
313 * @max_size: max size of segment in bytes
314 *
315 * Description:
316 * Enables a low level driver to set an upper limit on the size of a
317 * coalesced segment
318 **/
319void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
320{
321 if (max_size < PAGE_CACHE_SIZE) {
322 max_size = PAGE_CACHE_SIZE;
323 printk(KERN_INFO "%s: set to minimum %d\n",
324 __func__, max_size);
325 }
326
327 q->limits.max_segment_size = max_size;
328}
329EXPORT_SYMBOL(blk_queue_max_segment_size);
330
331/**
332 * blk_queue_logical_block_size - set logical block size for the queue
333 * @q: the request queue for the device
334 * @size: the logical block size, in bytes
335 *
336 * Description:
337 * This should be set to the lowest possible block size that the
338 * storage device can address. The default of 512 covers most
339 * hardware.
340 **/
341void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
342{
343 q->limits.logical_block_size = size;
344
345 if (q->limits.physical_block_size < size)
346 q->limits.physical_block_size = size;
347
348 if (q->limits.io_min < q->limits.physical_block_size)
349 q->limits.io_min = q->limits.physical_block_size;
350}
351EXPORT_SYMBOL(blk_queue_logical_block_size);
352
353/**
354 * blk_queue_physical_block_size - set physical block size for the queue
355 * @q: the request queue for the device
356 * @size: the physical block size, in bytes
357 *
358 * Description:
359 * This should be set to the lowest possible sector size that the
360 * hardware can operate on without reverting to read-modify-write
361 * operations.
362 */
363void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
364{
365 q->limits.physical_block_size = size;
366
367 if (q->limits.physical_block_size < q->limits.logical_block_size)
368 q->limits.physical_block_size = q->limits.logical_block_size;
369
370 if (q->limits.io_min < q->limits.physical_block_size)
371 q->limits.io_min = q->limits.physical_block_size;
372}
373EXPORT_SYMBOL(blk_queue_physical_block_size);
374
375/**
376 * blk_queue_alignment_offset - set physical block alignment offset
377 * @q: the request queue for the device
378 * @offset: alignment offset in bytes
379 *
380 * Description:
381 * Some devices are naturally misaligned to compensate for things like
382 * the legacy DOS partition table 63-sector offset. Low-level drivers
383 * should call this function for devices whose first sector is not
384 * naturally aligned.
385 */
386void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
387{
388 q->limits.alignment_offset =
389 offset & (q->limits.physical_block_size - 1);
390 q->limits.misaligned = 0;
391}
392EXPORT_SYMBOL(blk_queue_alignment_offset);
393
394/**
395 * blk_limits_io_min - set minimum request size for a device
396 * @limits: the queue limits
397 * @min: smallest I/O size in bytes
398 *
399 * Description:
400 * Some devices have an internal block size bigger than the reported
401 * hardware sector size. This function can be used to signal the
402 * smallest I/O the device can perform without incurring a performance
403 * penalty.
404 */
405void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
406{
407 limits->io_min = min;
408
409 if (limits->io_min < limits->logical_block_size)
410 limits->io_min = limits->logical_block_size;
411
412 if (limits->io_min < limits->physical_block_size)
413 limits->io_min = limits->physical_block_size;
414}
415EXPORT_SYMBOL(blk_limits_io_min);
416
417/**
418 * blk_queue_io_min - set minimum request size for the queue
419 * @q: the request queue for the device
420 * @min: smallest I/O size in bytes
421 *
422 * Description:
423 * Storage devices may report a granularity or preferred minimum I/O
424 * size which is the smallest request the device can perform without
425 * incurring a performance penalty. For disk drives this is often the
426 * physical block size. For RAID arrays it is often the stripe chunk
427 * size. A properly aligned multiple of minimum_io_size is the
428 * preferred request size for workloads where a high number of I/O
429 * operations is desired.
430 */
431void blk_queue_io_min(struct request_queue *q, unsigned int min)
432{
433 blk_limits_io_min(&q->limits, min);
434}
435EXPORT_SYMBOL(blk_queue_io_min);
436
437/**
438 * blk_limits_io_opt - set optimal request size for a device
439 * @limits: the queue limits
440 * @opt: smallest I/O size in bytes
441 *
442 * Description:
443 * Storage devices may report an optimal I/O size, which is the
444 * device's preferred unit for sustained I/O. This is rarely reported
445 * for disk drives. For RAID arrays it is usually the stripe width or
446 * the internal track size. A properly aligned multiple of
447 * optimal_io_size is the preferred request size for workloads where
448 * sustained throughput is desired.
449 */
450void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
451{
452 limits->io_opt = opt;
453}
454EXPORT_SYMBOL(blk_limits_io_opt);
455
456/**
457 * blk_queue_io_opt - set optimal request size for the queue
458 * @q: the request queue for the device
459 * @opt: optimal request size in bytes
460 *
461 * Description:
462 * Storage devices may report an optimal I/O size, which is the
463 * device's preferred unit for sustained I/O. This is rarely reported
464 * for disk drives. For RAID arrays it is usually the stripe width or
465 * the internal track size. A properly aligned multiple of
466 * optimal_io_size is the preferred request size for workloads where
467 * sustained throughput is desired.
468 */
469void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
470{
471 blk_limits_io_opt(&q->limits, opt);
472}
473EXPORT_SYMBOL(blk_queue_io_opt);
474
475/**
476 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
477 * @t: the stacking driver (top)
478 * @b: the underlying device (bottom)
479 **/
480void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
481{
482 blk_stack_limits(&t->limits, &b->limits, 0);
483}
484EXPORT_SYMBOL(blk_queue_stack_limits);
485
486/**
487 * blk_stack_limits - adjust queue_limits for stacked devices
488 * @t: the stacking driver limits (top device)
489 * @b: the underlying queue limits (bottom, component device)
490 * @start: first data sector within component device
491 *
492 * Description:
493 * This function is used by stacking drivers like MD and DM to ensure
494 * that all component devices have compatible block sizes and
495 * alignments. The stacking driver must provide a queue_limits
496 * struct (top) and then iteratively call the stacking function for
497 * all component (bottom) devices. The stacking function will
498 * attempt to combine the values and ensure proper alignment.
499 *
500 * Returns 0 if the top and bottom queue_limits are compatible. The
501 * top device's block sizes and alignment offsets may be adjusted to
502 * ensure alignment with the bottom device. If no compatible sizes
503 * and alignments exist, -1 is returned and the resulting top
504 * queue_limits will have the misaligned flag set to indicate that
505 * the alignment_offset is undefined.
506 */
507int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
508 sector_t start)
509{
510 unsigned int top, bottom, alignment, ret = 0;
511
512 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
513 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
514 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
515
516 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
517 b->seg_boundary_mask);
518
519 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
520 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
521 b->max_integrity_segments);
522
523 t->max_segment_size = min_not_zero(t->max_segment_size,
524 b->max_segment_size);
525
526 t->misaligned |= b->misaligned;
527
528 alignment = queue_limit_alignment_offset(b, start);
529
530 /* Bottom device has different alignment. Check that it is
531 * compatible with the current top alignment.
532 */
533 if (t->alignment_offset != alignment) {
534
535 top = max(t->physical_block_size, t->io_min)
536 + t->alignment_offset;
537 bottom = max(b->physical_block_size, b->io_min) + alignment;
538
539 /* Verify that top and bottom intervals line up */
540 if (max(top, bottom) & (min(top, bottom) - 1)) {
541 t->misaligned = 1;
542 ret = -1;
543 }
544 }
545
546 t->logical_block_size = max(t->logical_block_size,
547 b->logical_block_size);
548
549 t->physical_block_size = max(t->physical_block_size,
550 b->physical_block_size);
551
552 t->io_min = max(t->io_min, b->io_min);
553 t->io_opt = lcm(t->io_opt, b->io_opt);
554
555 t->cluster &= b->cluster;
556 t->discard_zeroes_data &= b->discard_zeroes_data;
557
558 /* Physical block size a multiple of the logical block size? */
559 if (t->physical_block_size & (t->logical_block_size - 1)) {
560 t->physical_block_size = t->logical_block_size;
561 t->misaligned = 1;
562 ret = -1;
563 }
564
565 /* Minimum I/O a multiple of the physical block size? */
566 if (t->io_min & (t->physical_block_size - 1)) {
567 t->io_min = t->physical_block_size;
568 t->misaligned = 1;
569 ret = -1;
570 }
571
572 /* Optimal I/O a multiple of the physical block size? */
573 if (t->io_opt & (t->physical_block_size - 1)) {
574 t->io_opt = 0;
575 t->misaligned = 1;
576 ret = -1;
577 }
578
579 /* Find lowest common alignment_offset */
580 t->alignment_offset = lcm(t->alignment_offset, alignment)
581 & (max(t->physical_block_size, t->io_min) - 1);
582
583 /* Verify that new alignment_offset is on a logical block boundary */
584 if (t->alignment_offset & (t->logical_block_size - 1)) {
585 t->misaligned = 1;
586 ret = -1;
587 }
588
589 /* Discard alignment and granularity */
590 if (b->discard_granularity) {
591 alignment = queue_limit_discard_alignment(b, start);
592
593 if (t->discard_granularity != 0 &&
594 t->discard_alignment != alignment) {
595 top = t->discard_granularity + t->discard_alignment;
596 bottom = b->discard_granularity + alignment;
597
598 /* Verify that top and bottom intervals line up */
599 if (max(top, bottom) & (min(top, bottom) - 1))
600 t->discard_misaligned = 1;
601 }
602
603 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
604 b->max_discard_sectors);
605 t->discard_granularity = max(t->discard_granularity,
606 b->discard_granularity);
607 t->discard_alignment = lcm(t->discard_alignment, alignment) &
608 (t->discard_granularity - 1);
609 }
610
611 return ret;
612}
613EXPORT_SYMBOL(blk_stack_limits);
614
615/**
616 * bdev_stack_limits - adjust queue limits for stacked drivers
617 * @t: the stacking driver limits (top device)
618 * @bdev: the component block_device (bottom)
619 * @start: first data sector within component device
620 *
621 * Description:
622 * Merges queue limits for a top device and a block_device. Returns
623 * 0 if alignment didn't change. Returns -1 if adding the bottom
624 * device caused misalignment.
625 */
626int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
627 sector_t start)
628{
629 struct request_queue *bq = bdev_get_queue(bdev);
630
631 start += get_start_sect(bdev);
632
633 return blk_stack_limits(t, &bq->limits, start);
634}
635EXPORT_SYMBOL(bdev_stack_limits);
636
637/**
638 * disk_stack_limits - adjust queue limits for stacked drivers
639 * @disk: MD/DM gendisk (top)
640 * @bdev: the underlying block device (bottom)
641 * @offset: offset to beginning of data within component device
642 *
643 * Description:
644 * Merges the limits for a top level gendisk and a bottom level
645 * block_device.
646 */
647void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
648 sector_t offset)
649{
650 struct request_queue *t = disk->queue;
651
652 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
653 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
654
655 disk_name(disk, 0, top);
656 bdevname(bdev, bottom);
657
658 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
659 top, bottom);
660 }
661}
662EXPORT_SYMBOL(disk_stack_limits);
663
664/**
665 * blk_queue_dma_pad - set pad mask
666 * @q: the request queue for the device
667 * @mask: pad mask
668 *
669 * Set dma pad mask.
670 *
671 * Appending pad buffer to a request modifies the last entry of a
672 * scatter list such that it includes the pad buffer.
673 **/
674void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
675{
676 q->dma_pad_mask = mask;
677}
678EXPORT_SYMBOL(blk_queue_dma_pad);
679
680/**
681 * blk_queue_update_dma_pad - update pad mask
682 * @q: the request queue for the device
683 * @mask: pad mask
684 *
685 * Update dma pad mask.
686 *
687 * Appending pad buffer to a request modifies the last entry of a
688 * scatter list such that it includes the pad buffer.
689 **/
690void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
691{
692 if (mask > q->dma_pad_mask)
693 q->dma_pad_mask = mask;
694}
695EXPORT_SYMBOL(blk_queue_update_dma_pad);
696
697/**
698 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
699 * @q: the request queue for the device
700 * @dma_drain_needed: fn which returns non-zero if drain is necessary
701 * @buf: physically contiguous buffer
702 * @size: size of the buffer in bytes
703 *
704 * Some devices have excess DMA problems and can't simply discard (or
705 * zero fill) the unwanted piece of the transfer. They have to have a
706 * real area of memory to transfer it into. The use case for this is
707 * ATAPI devices in DMA mode. If the packet command causes a transfer
708 * bigger than the transfer size some HBAs will lock up if there
709 * aren't DMA elements to contain the excess transfer. What this API
710 * does is adjust the queue so that the buf is always appended
711 * silently to the scatterlist.
712 *
713 * Note: This routine adjusts max_hw_segments to make room for appending
714 * the drain buffer. If you call blk_queue_max_segments() after calling
715 * this routine, you must set the limit to one fewer than your device
716 * can support otherwise there won't be room for the drain buffer.
717 */
718int blk_queue_dma_drain(struct request_queue *q,
719 dma_drain_needed_fn *dma_drain_needed,
720 void *buf, unsigned int size)
721{
722 if (queue_max_segments(q) < 2)
723 return -EINVAL;
724 /* make room for appending the drain */
725 blk_queue_max_segments(q, queue_max_segments(q) - 1);
726 q->dma_drain_needed = dma_drain_needed;
727 q->dma_drain_buffer = buf;
728 q->dma_drain_size = size;
729
730 return 0;
731}
732EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
733
734/**
735 * blk_queue_segment_boundary - set boundary rules for segment merging
736 * @q: the request queue for the device
737 * @mask: the memory boundary mask
738 **/
739void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
740{
741 if (mask < PAGE_CACHE_SIZE - 1) {
742 mask = PAGE_CACHE_SIZE - 1;
743 printk(KERN_INFO "%s: set to minimum %lx\n",
744 __func__, mask);
745 }
746
747 q->limits.seg_boundary_mask = mask;
748}
749EXPORT_SYMBOL(blk_queue_segment_boundary);
750
751/**
752 * blk_queue_dma_alignment - set dma length and memory alignment
753 * @q: the request queue for the device
754 * @mask: alignment mask
755 *
756 * description:
757 * set required memory and length alignment for direct dma transactions.
758 * this is used when building direct io requests for the queue.
759 *
760 **/
761void blk_queue_dma_alignment(struct request_queue *q, int mask)
762{
763 q->dma_alignment = mask;
764}
765EXPORT_SYMBOL(blk_queue_dma_alignment);
766
767/**
768 * blk_queue_update_dma_alignment - update dma length and memory alignment
769 * @q: the request queue for the device
770 * @mask: alignment mask
771 *
772 * description:
773 * update required memory and length alignment for direct dma transactions.
774 * If the requested alignment is larger than the current alignment, then
775 * the current queue alignment is updated to the new value, otherwise it
776 * is left alone. The design of this is to allow multiple objects
777 * (driver, device, transport etc) to set their respective
778 * alignments without having them interfere.
779 *
780 **/
781void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
782{
783 BUG_ON(mask > PAGE_SIZE);
784
785 if (mask > q->dma_alignment)
786 q->dma_alignment = mask;
787}
788EXPORT_SYMBOL(blk_queue_update_dma_alignment);
789
790/**
791 * blk_queue_flush - configure queue's cache flush capability
792 * @q: the request queue for the device
793 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
794 *
795 * Tell block layer cache flush capability of @q. If it supports
796 * flushing, REQ_FLUSH should be set. If it supports bypassing
797 * write cache for individual writes, REQ_FUA should be set.
798 */
799void blk_queue_flush(struct request_queue *q, unsigned int flush)
800{
801 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
802
803 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
804 flush &= ~REQ_FUA;
805
806 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
807}
808EXPORT_SYMBOL_GPL(blk_queue_flush);
809
810void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
811{
812 q->flush_not_queueable = !queueable;
813}
814EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
815
816static int __init blk_settings_init(void)
817{
818 blk_max_low_pfn = max_low_pfn - 1;
819 blk_max_pfn = max_pfn - 1;
820 return 0;
821}
822subsys_initcall(blk_settings_init);
1/*
2 * Functions related to setting various queue properties from drivers
3 */
4#include <linux/kernel.h>
5#include <linux/module.h>
6#include <linux/init.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
10#include <linux/gcd.h>
11#include <linux/lcm.h>
12#include <linux/jiffies.h>
13#include <linux/gfp.h>
14
15#include "blk.h"
16
17unsigned long blk_max_low_pfn;
18EXPORT_SYMBOL(blk_max_low_pfn);
19
20unsigned long blk_max_pfn;
21
22/**
23 * blk_queue_prep_rq - set a prepare_request function for queue
24 * @q: queue
25 * @pfn: prepare_request function
26 *
27 * It's possible for a queue to register a prepare_request callback which
28 * is invoked before the request is handed to the request_fn. The goal of
29 * the function is to prepare a request for I/O, it can be used to build a
30 * cdb from the request data for instance.
31 *
32 */
33void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
34{
35 q->prep_rq_fn = pfn;
36}
37EXPORT_SYMBOL(blk_queue_prep_rq);
38
39/**
40 * blk_queue_unprep_rq - set an unprepare_request function for queue
41 * @q: queue
42 * @ufn: unprepare_request function
43 *
44 * It's possible for a queue to register an unprepare_request callback
45 * which is invoked before the request is finally completed. The goal
46 * of the function is to deallocate any data that was allocated in the
47 * prepare_request callback.
48 *
49 */
50void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn)
51{
52 q->unprep_rq_fn = ufn;
53}
54EXPORT_SYMBOL(blk_queue_unprep_rq);
55
56/**
57 * blk_queue_merge_bvec - set a merge_bvec function for queue
58 * @q: queue
59 * @mbfn: merge_bvec_fn
60 *
61 * Usually queues have static limitations on the max sectors or segments that
62 * we can put in a request. Stacking drivers may have some settings that
63 * are dynamic, and thus we have to query the queue whether it is ok to
64 * add a new bio_vec to a bio at a given offset or not. If the block device
65 * has such limitations, it needs to register a merge_bvec_fn to control
66 * the size of bio's sent to it. Note that a block device *must* allow a
67 * single page to be added to an empty bio. The block device driver may want
68 * to use the bio_split() function to deal with these bio's. By default
69 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
70 * honored.
71 */
72void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
73{
74 q->merge_bvec_fn = mbfn;
75}
76EXPORT_SYMBOL(blk_queue_merge_bvec);
77
78void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
79{
80 q->softirq_done_fn = fn;
81}
82EXPORT_SYMBOL(blk_queue_softirq_done);
83
84void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
85{
86 q->rq_timeout = timeout;
87}
88EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
89
90void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
91{
92 q->rq_timed_out_fn = fn;
93}
94EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
95
96void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
97{
98 q->lld_busy_fn = fn;
99}
100EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
101
102/**
103 * blk_set_default_limits - reset limits to default values
104 * @lim: the queue_limits structure to reset
105 *
106 * Description:
107 * Returns a queue_limit struct to its default state.
108 */
109void blk_set_default_limits(struct queue_limits *lim)
110{
111 lim->max_segments = BLK_MAX_SEGMENTS;
112 lim->max_integrity_segments = 0;
113 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
114 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
115 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
116 lim->max_write_same_sectors = 0;
117 lim->max_discard_sectors = 0;
118 lim->discard_granularity = 0;
119 lim->discard_alignment = 0;
120 lim->discard_misaligned = 0;
121 lim->discard_zeroes_data = 0;
122 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
123 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
124 lim->alignment_offset = 0;
125 lim->io_opt = 0;
126 lim->misaligned = 0;
127 lim->cluster = 1;
128}
129EXPORT_SYMBOL(blk_set_default_limits);
130
131/**
132 * blk_set_stacking_limits - set default limits for stacking devices
133 * @lim: the queue_limits structure to reset
134 *
135 * Description:
136 * Returns a queue_limit struct to its default state. Should be used
137 * by stacking drivers like DM that have no internal limits.
138 */
139void blk_set_stacking_limits(struct queue_limits *lim)
140{
141 blk_set_default_limits(lim);
142
143 /* Inherit limits from component devices */
144 lim->discard_zeroes_data = 1;
145 lim->max_segments = USHRT_MAX;
146 lim->max_hw_sectors = UINT_MAX;
147 lim->max_segment_size = UINT_MAX;
148 lim->max_sectors = UINT_MAX;
149 lim->max_write_same_sectors = UINT_MAX;
150}
151EXPORT_SYMBOL(blk_set_stacking_limits);
152
153/**
154 * blk_queue_make_request - define an alternate make_request function for a device
155 * @q: the request queue for the device to be affected
156 * @mfn: the alternate make_request function
157 *
158 * Description:
159 * The normal way for &struct bios to be passed to a device
160 * driver is for them to be collected into requests on a request
161 * queue, and then to allow the device driver to select requests
162 * off that queue when it is ready. This works well for many block
163 * devices. However some block devices (typically virtual devices
164 * such as md or lvm) do not benefit from the processing on the
165 * request queue, and are served best by having the requests passed
166 * directly to them. This can be achieved by providing a function
167 * to blk_queue_make_request().
168 *
169 * Caveat:
170 * The driver that does this *must* be able to deal appropriately
171 * with buffers in "highmemory". This can be accomplished by either calling
172 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
173 * blk_queue_bounce() to create a buffer in normal memory.
174 **/
175void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
176{
177 /*
178 * set defaults
179 */
180 q->nr_requests = BLKDEV_MAX_RQ;
181
182 q->make_request_fn = mfn;
183 blk_queue_dma_alignment(q, 511);
184 blk_queue_congestion_threshold(q);
185 q->nr_batching = BLK_BATCH_REQ;
186
187 blk_set_default_limits(&q->limits);
188
189 /*
190 * by default assume old behaviour and bounce for any highmem page
191 */
192 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
193}
194EXPORT_SYMBOL(blk_queue_make_request);
195
196/**
197 * blk_queue_bounce_limit - set bounce buffer limit for queue
198 * @q: the request queue for the device
199 * @max_addr: the maximum address the device can handle
200 *
201 * Description:
202 * Different hardware can have different requirements as to what pages
203 * it can do I/O directly to. A low level driver can call
204 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
205 * buffers for doing I/O to pages residing above @max_addr.
206 **/
207void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
208{
209 unsigned long b_pfn = max_addr >> PAGE_SHIFT;
210 int dma = 0;
211
212 q->bounce_gfp = GFP_NOIO;
213#if BITS_PER_LONG == 64
214 /*
215 * Assume anything <= 4GB can be handled by IOMMU. Actually
216 * some IOMMUs can handle everything, but I don't know of a
217 * way to test this here.
218 */
219 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
220 dma = 1;
221 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
222#else
223 if (b_pfn < blk_max_low_pfn)
224 dma = 1;
225 q->limits.bounce_pfn = b_pfn;
226#endif
227 if (dma) {
228 init_emergency_isa_pool();
229 q->bounce_gfp = GFP_NOIO | GFP_DMA;
230 q->limits.bounce_pfn = b_pfn;
231 }
232}
233EXPORT_SYMBOL(blk_queue_bounce_limit);
234
235/**
236 * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request
237 * @limits: the queue limits
238 * @max_hw_sectors: max hardware sectors in the usual 512b unit
239 *
240 * Description:
241 * Enables a low level driver to set a hard upper limit,
242 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
243 * the device driver based upon the combined capabilities of I/O
244 * controller and storage device.
245 *
246 * max_sectors is a soft limit imposed by the block layer for
247 * filesystem type requests. This value can be overridden on a
248 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
249 * The soft limit can not exceed max_hw_sectors.
250 **/
251void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors)
252{
253 if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
254 max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
255 printk(KERN_INFO "%s: set to minimum %d\n",
256 __func__, max_hw_sectors);
257 }
258
259 limits->max_hw_sectors = max_hw_sectors;
260 limits->max_sectors = min_t(unsigned int, max_hw_sectors,
261 BLK_DEF_MAX_SECTORS);
262}
263EXPORT_SYMBOL(blk_limits_max_hw_sectors);
264
265/**
266 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
267 * @q: the request queue for the device
268 * @max_hw_sectors: max hardware sectors in the usual 512b unit
269 *
270 * Description:
271 * See description for blk_limits_max_hw_sectors().
272 **/
273void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
274{
275 blk_limits_max_hw_sectors(&q->limits, max_hw_sectors);
276}
277EXPORT_SYMBOL(blk_queue_max_hw_sectors);
278
279/**
280 * blk_queue_max_discard_sectors - set max sectors for a single discard
281 * @q: the request queue for the device
282 * @max_discard_sectors: maximum number of sectors to discard
283 **/
284void blk_queue_max_discard_sectors(struct request_queue *q,
285 unsigned int max_discard_sectors)
286{
287 q->limits.max_discard_sectors = max_discard_sectors;
288}
289EXPORT_SYMBOL(blk_queue_max_discard_sectors);
290
291/**
292 * blk_queue_max_write_same_sectors - set max sectors for a single write same
293 * @q: the request queue for the device
294 * @max_write_same_sectors: maximum number of sectors to write per command
295 **/
296void blk_queue_max_write_same_sectors(struct request_queue *q,
297 unsigned int max_write_same_sectors)
298{
299 q->limits.max_write_same_sectors = max_write_same_sectors;
300}
301EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
302
303/**
304 * blk_queue_max_segments - set max hw segments for a request for this queue
305 * @q: the request queue for the device
306 * @max_segments: max number of segments
307 *
308 * Description:
309 * Enables a low level driver to set an upper limit on the number of
310 * hw data segments in a request.
311 **/
312void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
313{
314 if (!max_segments) {
315 max_segments = 1;
316 printk(KERN_INFO "%s: set to minimum %d\n",
317 __func__, max_segments);
318 }
319
320 q->limits.max_segments = max_segments;
321}
322EXPORT_SYMBOL(blk_queue_max_segments);
323
324/**
325 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
326 * @q: the request queue for the device
327 * @max_size: max size of segment in bytes
328 *
329 * Description:
330 * Enables a low level driver to set an upper limit on the size of a
331 * coalesced segment
332 **/
333void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
334{
335 if (max_size < PAGE_CACHE_SIZE) {
336 max_size = PAGE_CACHE_SIZE;
337 printk(KERN_INFO "%s: set to minimum %d\n",
338 __func__, max_size);
339 }
340
341 q->limits.max_segment_size = max_size;
342}
343EXPORT_SYMBOL(blk_queue_max_segment_size);
344
345/**
346 * blk_queue_logical_block_size - set logical block size for the queue
347 * @q: the request queue for the device
348 * @size: the logical block size, in bytes
349 *
350 * Description:
351 * This should be set to the lowest possible block size that the
352 * storage device can address. The default of 512 covers most
353 * hardware.
354 **/
355void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
356{
357 q->limits.logical_block_size = size;
358
359 if (q->limits.physical_block_size < size)
360 q->limits.physical_block_size = size;
361
362 if (q->limits.io_min < q->limits.physical_block_size)
363 q->limits.io_min = q->limits.physical_block_size;
364}
365EXPORT_SYMBOL(blk_queue_logical_block_size);
366
367/**
368 * blk_queue_physical_block_size - set physical block size for the queue
369 * @q: the request queue for the device
370 * @size: the physical block size, in bytes
371 *
372 * Description:
373 * This should be set to the lowest possible sector size that the
374 * hardware can operate on without reverting to read-modify-write
375 * operations.
376 */
377void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
378{
379 q->limits.physical_block_size = size;
380
381 if (q->limits.physical_block_size < q->limits.logical_block_size)
382 q->limits.physical_block_size = q->limits.logical_block_size;
383
384 if (q->limits.io_min < q->limits.physical_block_size)
385 q->limits.io_min = q->limits.physical_block_size;
386}
387EXPORT_SYMBOL(blk_queue_physical_block_size);
388
389/**
390 * blk_queue_alignment_offset - set physical block alignment offset
391 * @q: the request queue for the device
392 * @offset: alignment offset in bytes
393 *
394 * Description:
395 * Some devices are naturally misaligned to compensate for things like
396 * the legacy DOS partition table 63-sector offset. Low-level drivers
397 * should call this function for devices whose first sector is not
398 * naturally aligned.
399 */
400void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
401{
402 q->limits.alignment_offset =
403 offset & (q->limits.physical_block_size - 1);
404 q->limits.misaligned = 0;
405}
406EXPORT_SYMBOL(blk_queue_alignment_offset);
407
408/**
409 * blk_limits_io_min - set minimum request size for a device
410 * @limits: the queue limits
411 * @min: smallest I/O size in bytes
412 *
413 * Description:
414 * Some devices have an internal block size bigger than the reported
415 * hardware sector size. This function can be used to signal the
416 * smallest I/O the device can perform without incurring a performance
417 * penalty.
418 */
419void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
420{
421 limits->io_min = min;
422
423 if (limits->io_min < limits->logical_block_size)
424 limits->io_min = limits->logical_block_size;
425
426 if (limits->io_min < limits->physical_block_size)
427 limits->io_min = limits->physical_block_size;
428}
429EXPORT_SYMBOL(blk_limits_io_min);
430
431/**
432 * blk_queue_io_min - set minimum request size for the queue
433 * @q: the request queue for the device
434 * @min: smallest I/O size in bytes
435 *
436 * Description:
437 * Storage devices may report a granularity or preferred minimum I/O
438 * size which is the smallest request the device can perform without
439 * incurring a performance penalty. For disk drives this is often the
440 * physical block size. For RAID arrays it is often the stripe chunk
441 * size. A properly aligned multiple of minimum_io_size is the
442 * preferred request size for workloads where a high number of I/O
443 * operations is desired.
444 */
445void blk_queue_io_min(struct request_queue *q, unsigned int min)
446{
447 blk_limits_io_min(&q->limits, min);
448}
449EXPORT_SYMBOL(blk_queue_io_min);
450
451/**
452 * blk_limits_io_opt - set optimal request size for a device
453 * @limits: the queue limits
454 * @opt: smallest I/O size in bytes
455 *
456 * Description:
457 * Storage devices may report an optimal I/O size, which is the
458 * device's preferred unit for sustained I/O. This is rarely reported
459 * for disk drives. For RAID arrays it is usually the stripe width or
460 * the internal track size. A properly aligned multiple of
461 * optimal_io_size is the preferred request size for workloads where
462 * sustained throughput is desired.
463 */
464void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
465{
466 limits->io_opt = opt;
467}
468EXPORT_SYMBOL(blk_limits_io_opt);
469
470/**
471 * blk_queue_io_opt - set optimal request size for the queue
472 * @q: the request queue for the device
473 * @opt: optimal request size in bytes
474 *
475 * Description:
476 * Storage devices may report an optimal I/O size, which is the
477 * device's preferred unit for sustained I/O. This is rarely reported
478 * for disk drives. For RAID arrays it is usually the stripe width or
479 * the internal track size. A properly aligned multiple of
480 * optimal_io_size is the preferred request size for workloads where
481 * sustained throughput is desired.
482 */
483void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
484{
485 blk_limits_io_opt(&q->limits, opt);
486}
487EXPORT_SYMBOL(blk_queue_io_opt);
488
489/**
490 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
491 * @t: the stacking driver (top)
492 * @b: the underlying device (bottom)
493 **/
494void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
495{
496 blk_stack_limits(&t->limits, &b->limits, 0);
497}
498EXPORT_SYMBOL(blk_queue_stack_limits);
499
500/**
501 * blk_stack_limits - adjust queue_limits for stacked devices
502 * @t: the stacking driver limits (top device)
503 * @b: the underlying queue limits (bottom, component device)
504 * @start: first data sector within component device
505 *
506 * Description:
507 * This function is used by stacking drivers like MD and DM to ensure
508 * that all component devices have compatible block sizes and
509 * alignments. The stacking driver must provide a queue_limits
510 * struct (top) and then iteratively call the stacking function for
511 * all component (bottom) devices. The stacking function will
512 * attempt to combine the values and ensure proper alignment.
513 *
514 * Returns 0 if the top and bottom queue_limits are compatible. The
515 * top device's block sizes and alignment offsets may be adjusted to
516 * ensure alignment with the bottom device. If no compatible sizes
517 * and alignments exist, -1 is returned and the resulting top
518 * queue_limits will have the misaligned flag set to indicate that
519 * the alignment_offset is undefined.
520 */
521int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
522 sector_t start)
523{
524 unsigned int top, bottom, alignment, ret = 0;
525
526 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
527 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
528 t->max_write_same_sectors = min(t->max_write_same_sectors,
529 b->max_write_same_sectors);
530 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
531
532 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
533 b->seg_boundary_mask);
534
535 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
536 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
537 b->max_integrity_segments);
538
539 t->max_segment_size = min_not_zero(t->max_segment_size,
540 b->max_segment_size);
541
542 t->misaligned |= b->misaligned;
543
544 alignment = queue_limit_alignment_offset(b, start);
545
546 /* Bottom device has different alignment. Check that it is
547 * compatible with the current top alignment.
548 */
549 if (t->alignment_offset != alignment) {
550
551 top = max(t->physical_block_size, t->io_min)
552 + t->alignment_offset;
553 bottom = max(b->physical_block_size, b->io_min) + alignment;
554
555 /* Verify that top and bottom intervals line up */
556 if (max(top, bottom) & (min(top, bottom) - 1)) {
557 t->misaligned = 1;
558 ret = -1;
559 }
560 }
561
562 t->logical_block_size = max(t->logical_block_size,
563 b->logical_block_size);
564
565 t->physical_block_size = max(t->physical_block_size,
566 b->physical_block_size);
567
568 t->io_min = max(t->io_min, b->io_min);
569 t->io_opt = lcm(t->io_opt, b->io_opt);
570
571 t->cluster &= b->cluster;
572 t->discard_zeroes_data &= b->discard_zeroes_data;
573
574 /* Physical block size a multiple of the logical block size? */
575 if (t->physical_block_size & (t->logical_block_size - 1)) {
576 t->physical_block_size = t->logical_block_size;
577 t->misaligned = 1;
578 ret = -1;
579 }
580
581 /* Minimum I/O a multiple of the physical block size? */
582 if (t->io_min & (t->physical_block_size - 1)) {
583 t->io_min = t->physical_block_size;
584 t->misaligned = 1;
585 ret = -1;
586 }
587
588 /* Optimal I/O a multiple of the physical block size? */
589 if (t->io_opt & (t->physical_block_size - 1)) {
590 t->io_opt = 0;
591 t->misaligned = 1;
592 ret = -1;
593 }
594
595 t->raid_partial_stripes_expensive =
596 max(t->raid_partial_stripes_expensive,
597 b->raid_partial_stripes_expensive);
598
599 /* Find lowest common alignment_offset */
600 t->alignment_offset = lcm(t->alignment_offset, alignment)
601 & (max(t->physical_block_size, t->io_min) - 1);
602
603 /* Verify that new alignment_offset is on a logical block boundary */
604 if (t->alignment_offset & (t->logical_block_size - 1)) {
605 t->misaligned = 1;
606 ret = -1;
607 }
608
609 /* Discard alignment and granularity */
610 if (b->discard_granularity) {
611 alignment = queue_limit_discard_alignment(b, start);
612
613 if (t->discard_granularity != 0 &&
614 t->discard_alignment != alignment) {
615 top = t->discard_granularity + t->discard_alignment;
616 bottom = b->discard_granularity + alignment;
617
618 /* Verify that top and bottom intervals line up */
619 if ((max(top, bottom) % min(top, bottom)) != 0)
620 t->discard_misaligned = 1;
621 }
622
623 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
624 b->max_discard_sectors);
625 t->discard_granularity = max(t->discard_granularity,
626 b->discard_granularity);
627 t->discard_alignment = lcm(t->discard_alignment, alignment) %
628 t->discard_granularity;
629 }
630
631 return ret;
632}
633EXPORT_SYMBOL(blk_stack_limits);
634
635/**
636 * bdev_stack_limits - adjust queue limits for stacked drivers
637 * @t: the stacking driver limits (top device)
638 * @bdev: the component block_device (bottom)
639 * @start: first data sector within component device
640 *
641 * Description:
642 * Merges queue limits for a top device and a block_device. Returns
643 * 0 if alignment didn't change. Returns -1 if adding the bottom
644 * device caused misalignment.
645 */
646int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
647 sector_t start)
648{
649 struct request_queue *bq = bdev_get_queue(bdev);
650
651 start += get_start_sect(bdev);
652
653 return blk_stack_limits(t, &bq->limits, start);
654}
655EXPORT_SYMBOL(bdev_stack_limits);
656
657/**
658 * disk_stack_limits - adjust queue limits for stacked drivers
659 * @disk: MD/DM gendisk (top)
660 * @bdev: the underlying block device (bottom)
661 * @offset: offset to beginning of data within component device
662 *
663 * Description:
664 * Merges the limits for a top level gendisk and a bottom level
665 * block_device.
666 */
667void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
668 sector_t offset)
669{
670 struct request_queue *t = disk->queue;
671
672 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
673 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
674
675 disk_name(disk, 0, top);
676 bdevname(bdev, bottom);
677
678 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
679 top, bottom);
680 }
681}
682EXPORT_SYMBOL(disk_stack_limits);
683
684/**
685 * blk_queue_dma_pad - set pad mask
686 * @q: the request queue for the device
687 * @mask: pad mask
688 *
689 * Set dma pad mask.
690 *
691 * Appending pad buffer to a request modifies the last entry of a
692 * scatter list such that it includes the pad buffer.
693 **/
694void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
695{
696 q->dma_pad_mask = mask;
697}
698EXPORT_SYMBOL(blk_queue_dma_pad);
699
700/**
701 * blk_queue_update_dma_pad - update pad mask
702 * @q: the request queue for the device
703 * @mask: pad mask
704 *
705 * Update dma pad mask.
706 *
707 * Appending pad buffer to a request modifies the last entry of a
708 * scatter list such that it includes the pad buffer.
709 **/
710void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
711{
712 if (mask > q->dma_pad_mask)
713 q->dma_pad_mask = mask;
714}
715EXPORT_SYMBOL(blk_queue_update_dma_pad);
716
717/**
718 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
719 * @q: the request queue for the device
720 * @dma_drain_needed: fn which returns non-zero if drain is necessary
721 * @buf: physically contiguous buffer
722 * @size: size of the buffer in bytes
723 *
724 * Some devices have excess DMA problems and can't simply discard (or
725 * zero fill) the unwanted piece of the transfer. They have to have a
726 * real area of memory to transfer it into. The use case for this is
727 * ATAPI devices in DMA mode. If the packet command causes a transfer
728 * bigger than the transfer size some HBAs will lock up if there
729 * aren't DMA elements to contain the excess transfer. What this API
730 * does is adjust the queue so that the buf is always appended
731 * silently to the scatterlist.
732 *
733 * Note: This routine adjusts max_hw_segments to make room for appending
734 * the drain buffer. If you call blk_queue_max_segments() after calling
735 * this routine, you must set the limit to one fewer than your device
736 * can support otherwise there won't be room for the drain buffer.
737 */
738int blk_queue_dma_drain(struct request_queue *q,
739 dma_drain_needed_fn *dma_drain_needed,
740 void *buf, unsigned int size)
741{
742 if (queue_max_segments(q) < 2)
743 return -EINVAL;
744 /* make room for appending the drain */
745 blk_queue_max_segments(q, queue_max_segments(q) - 1);
746 q->dma_drain_needed = dma_drain_needed;
747 q->dma_drain_buffer = buf;
748 q->dma_drain_size = size;
749
750 return 0;
751}
752EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
753
754/**
755 * blk_queue_segment_boundary - set boundary rules for segment merging
756 * @q: the request queue for the device
757 * @mask: the memory boundary mask
758 **/
759void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
760{
761 if (mask < PAGE_CACHE_SIZE - 1) {
762 mask = PAGE_CACHE_SIZE - 1;
763 printk(KERN_INFO "%s: set to minimum %lx\n",
764 __func__, mask);
765 }
766
767 q->limits.seg_boundary_mask = mask;
768}
769EXPORT_SYMBOL(blk_queue_segment_boundary);
770
771/**
772 * blk_queue_dma_alignment - set dma length and memory alignment
773 * @q: the request queue for the device
774 * @mask: alignment mask
775 *
776 * description:
777 * set required memory and length alignment for direct dma transactions.
778 * this is used when building direct io requests for the queue.
779 *
780 **/
781void blk_queue_dma_alignment(struct request_queue *q, int mask)
782{
783 q->dma_alignment = mask;
784}
785EXPORT_SYMBOL(blk_queue_dma_alignment);
786
787/**
788 * blk_queue_update_dma_alignment - update dma length and memory alignment
789 * @q: the request queue for the device
790 * @mask: alignment mask
791 *
792 * description:
793 * update required memory and length alignment for direct dma transactions.
794 * If the requested alignment is larger than the current alignment, then
795 * the current queue alignment is updated to the new value, otherwise it
796 * is left alone. The design of this is to allow multiple objects
797 * (driver, device, transport etc) to set their respective
798 * alignments without having them interfere.
799 *
800 **/
801void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
802{
803 BUG_ON(mask > PAGE_SIZE);
804
805 if (mask > q->dma_alignment)
806 q->dma_alignment = mask;
807}
808EXPORT_SYMBOL(blk_queue_update_dma_alignment);
809
810/**
811 * blk_queue_flush - configure queue's cache flush capability
812 * @q: the request queue for the device
813 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
814 *
815 * Tell block layer cache flush capability of @q. If it supports
816 * flushing, REQ_FLUSH should be set. If it supports bypassing
817 * write cache for individual writes, REQ_FUA should be set.
818 */
819void blk_queue_flush(struct request_queue *q, unsigned int flush)
820{
821 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
822
823 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
824 flush &= ~REQ_FUA;
825
826 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
827}
828EXPORT_SYMBOL_GPL(blk_queue_flush);
829
830void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
831{
832 q->flush_not_queueable = !queueable;
833}
834EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
835
836static int __init blk_settings_init(void)
837{
838 blk_max_low_pfn = max_low_pfn - 1;
839 blk_max_pfn = max_pfn - 1;
840 return 0;
841}
842subsys_initcall(blk_settings_init);