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
56void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
57{
58 q->softirq_done_fn = fn;
59}
60EXPORT_SYMBOL(blk_queue_softirq_done);
61
62void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
63{
64 q->rq_timeout = timeout;
65}
66EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
67
68void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
69{
70 q->rq_timed_out_fn = fn;
71}
72EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
73
74void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
75{
76 q->lld_busy_fn = fn;
77}
78EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
79
80/**
81 * blk_set_default_limits - reset limits to default values
82 * @lim: the queue_limits structure to reset
83 *
84 * Description:
85 * Returns a queue_limit struct to its default state.
86 */
87void blk_set_default_limits(struct queue_limits *lim)
88{
89 lim->max_segments = BLK_MAX_SEGMENTS;
90 lim->max_integrity_segments = 0;
91 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
92 lim->virt_boundary_mask = 0;
93 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
94 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
95 lim->max_dev_sectors = 0;
96 lim->chunk_sectors = 0;
97 lim->max_write_same_sectors = 0;
98 lim->max_discard_sectors = 0;
99 lim->max_hw_discard_sectors = 0;
100 lim->discard_granularity = 0;
101 lim->discard_alignment = 0;
102 lim->discard_misaligned = 0;
103 lim->discard_zeroes_data = 0;
104 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
105 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
106 lim->alignment_offset = 0;
107 lim->io_opt = 0;
108 lim->misaligned = 0;
109 lim->cluster = 1;
110}
111EXPORT_SYMBOL(blk_set_default_limits);
112
113/**
114 * blk_set_stacking_limits - set default limits for stacking devices
115 * @lim: the queue_limits structure to reset
116 *
117 * Description:
118 * Returns a queue_limit struct to its default state. Should be used
119 * by stacking drivers like DM that have no internal limits.
120 */
121void blk_set_stacking_limits(struct queue_limits *lim)
122{
123 blk_set_default_limits(lim);
124
125 /* Inherit limits from component devices */
126 lim->discard_zeroes_data = 1;
127 lim->max_segments = USHRT_MAX;
128 lim->max_hw_sectors = UINT_MAX;
129 lim->max_segment_size = UINT_MAX;
130 lim->max_sectors = UINT_MAX;
131 lim->max_dev_sectors = UINT_MAX;
132 lim->max_write_same_sectors = UINT_MAX;
133}
134EXPORT_SYMBOL(blk_set_stacking_limits);
135
136/**
137 * blk_queue_make_request - define an alternate make_request function for a device
138 * @q: the request queue for the device to be affected
139 * @mfn: the alternate make_request function
140 *
141 * Description:
142 * The normal way for &struct bios to be passed to a device
143 * driver is for them to be collected into requests on a request
144 * queue, and then to allow the device driver to select requests
145 * off that queue when it is ready. This works well for many block
146 * devices. However some block devices (typically virtual devices
147 * such as md or lvm) do not benefit from the processing on the
148 * request queue, and are served best by having the requests passed
149 * directly to them. This can be achieved by providing a function
150 * to blk_queue_make_request().
151 *
152 * Caveat:
153 * The driver that does this *must* be able to deal appropriately
154 * with buffers in "highmemory". This can be accomplished by either calling
155 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
156 * blk_queue_bounce() to create a buffer in normal memory.
157 **/
158void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
159{
160 /*
161 * set defaults
162 */
163 q->nr_requests = BLKDEV_MAX_RQ;
164
165 q->make_request_fn = mfn;
166 blk_queue_dma_alignment(q, 511);
167 blk_queue_congestion_threshold(q);
168 q->nr_batching = BLK_BATCH_REQ;
169
170 blk_set_default_limits(&q->limits);
171
172 /*
173 * by default assume old behaviour and bounce for any highmem page
174 */
175 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
176}
177EXPORT_SYMBOL(blk_queue_make_request);
178
179/**
180 * blk_queue_bounce_limit - set bounce buffer limit for queue
181 * @q: the request queue for the device
182 * @max_addr: the maximum address the device can handle
183 *
184 * Description:
185 * Different hardware can have different requirements as to what pages
186 * it can do I/O directly to. A low level driver can call
187 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
188 * buffers for doing I/O to pages residing above @max_addr.
189 **/
190void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
191{
192 unsigned long b_pfn = max_addr >> PAGE_SHIFT;
193 int dma = 0;
194
195 q->bounce_gfp = GFP_NOIO;
196#if BITS_PER_LONG == 64
197 /*
198 * Assume anything <= 4GB can be handled by IOMMU. Actually
199 * some IOMMUs can handle everything, but I don't know of a
200 * way to test this here.
201 */
202 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
203 dma = 1;
204 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
205#else
206 if (b_pfn < blk_max_low_pfn)
207 dma = 1;
208 q->limits.bounce_pfn = b_pfn;
209#endif
210 if (dma) {
211 init_emergency_isa_pool();
212 q->bounce_gfp = GFP_NOIO | GFP_DMA;
213 q->limits.bounce_pfn = b_pfn;
214 }
215}
216EXPORT_SYMBOL(blk_queue_bounce_limit);
217
218/**
219 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
220 * @q: the request queue for the device
221 * @max_hw_sectors: max hardware sectors in the usual 512b unit
222 *
223 * Description:
224 * Enables a low level driver to set a hard upper limit,
225 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
226 * the device driver based upon the capabilities of the I/O
227 * controller.
228 *
229 * max_dev_sectors is a hard limit imposed by the storage device for
230 * READ/WRITE requests. It is set by the disk driver.
231 *
232 * max_sectors is a soft limit imposed by the block layer for
233 * filesystem type requests. This value can be overridden on a
234 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
235 * The soft limit can not exceed max_hw_sectors.
236 **/
237void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
238{
239 struct queue_limits *limits = &q->limits;
240 unsigned int max_sectors;
241
242 if ((max_hw_sectors << 9) < PAGE_SIZE) {
243 max_hw_sectors = 1 << (PAGE_SHIFT - 9);
244 printk(KERN_INFO "%s: set to minimum %d\n",
245 __func__, max_hw_sectors);
246 }
247
248 limits->max_hw_sectors = max_hw_sectors;
249 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
250 max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
251 limits->max_sectors = max_sectors;
252}
253EXPORT_SYMBOL(blk_queue_max_hw_sectors);
254
255/**
256 * blk_queue_chunk_sectors - set size of the chunk for this queue
257 * @q: the request queue for the device
258 * @chunk_sectors: chunk sectors in the usual 512b unit
259 *
260 * Description:
261 * If a driver doesn't want IOs to cross a given chunk size, it can set
262 * this limit and prevent merging across chunks. Note that the chunk size
263 * must currently be a power-of-2 in sectors. Also note that the block
264 * layer must accept a page worth of data at any offset. So if the
265 * crossing of chunks is a hard limitation in the driver, it must still be
266 * prepared to split single page bios.
267 **/
268void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
269{
270 BUG_ON(!is_power_of_2(chunk_sectors));
271 q->limits.chunk_sectors = chunk_sectors;
272}
273EXPORT_SYMBOL(blk_queue_chunk_sectors);
274
275/**
276 * blk_queue_max_discard_sectors - set max sectors for a single discard
277 * @q: the request queue for the device
278 * @max_discard_sectors: maximum number of sectors to discard
279 **/
280void blk_queue_max_discard_sectors(struct request_queue *q,
281 unsigned int max_discard_sectors)
282{
283 q->limits.max_hw_discard_sectors = max_discard_sectors;
284 q->limits.max_discard_sectors = max_discard_sectors;
285}
286EXPORT_SYMBOL(blk_queue_max_discard_sectors);
287
288/**
289 * blk_queue_max_write_same_sectors - set max sectors for a single write same
290 * @q: the request queue for the device
291 * @max_write_same_sectors: maximum number of sectors to write per command
292 **/
293void blk_queue_max_write_same_sectors(struct request_queue *q,
294 unsigned int max_write_same_sectors)
295{
296 q->limits.max_write_same_sectors = max_write_same_sectors;
297}
298EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
299
300/**
301 * blk_queue_max_segments - set max hw segments for a request for this queue
302 * @q: the request queue for the device
303 * @max_segments: max number of segments
304 *
305 * Description:
306 * Enables a low level driver to set an upper limit on the number of
307 * hw data segments in a request.
308 **/
309void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
310{
311 if (!max_segments) {
312 max_segments = 1;
313 printk(KERN_INFO "%s: set to minimum %d\n",
314 __func__, max_segments);
315 }
316
317 q->limits.max_segments = max_segments;
318}
319EXPORT_SYMBOL(blk_queue_max_segments);
320
321/**
322 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
323 * @q: the request queue for the device
324 * @max_size: max size of segment in bytes
325 *
326 * Description:
327 * Enables a low level driver to set an upper limit on the size of a
328 * coalesced segment
329 **/
330void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
331{
332 if (max_size < PAGE_SIZE) {
333 max_size = PAGE_SIZE;
334 printk(KERN_INFO "%s: set to minimum %d\n",
335 __func__, max_size);
336 }
337
338 q->limits.max_segment_size = max_size;
339}
340EXPORT_SYMBOL(blk_queue_max_segment_size);
341
342/**
343 * blk_queue_logical_block_size - set logical block size for the queue
344 * @q: the request queue for the device
345 * @size: the logical block size, in bytes
346 *
347 * Description:
348 * This should be set to the lowest possible block size that the
349 * storage device can address. The default of 512 covers most
350 * hardware.
351 **/
352void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
353{
354 q->limits.logical_block_size = size;
355
356 if (q->limits.physical_block_size < size)
357 q->limits.physical_block_size = size;
358
359 if (q->limits.io_min < q->limits.physical_block_size)
360 q->limits.io_min = q->limits.physical_block_size;
361}
362EXPORT_SYMBOL(blk_queue_logical_block_size);
363
364/**
365 * blk_queue_physical_block_size - set physical block size for the queue
366 * @q: the request queue for the device
367 * @size: the physical block size, in bytes
368 *
369 * Description:
370 * This should be set to the lowest possible sector size that the
371 * hardware can operate on without reverting to read-modify-write
372 * operations.
373 */
374void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
375{
376 q->limits.physical_block_size = size;
377
378 if (q->limits.physical_block_size < q->limits.logical_block_size)
379 q->limits.physical_block_size = q->limits.logical_block_size;
380
381 if (q->limits.io_min < q->limits.physical_block_size)
382 q->limits.io_min = q->limits.physical_block_size;
383}
384EXPORT_SYMBOL(blk_queue_physical_block_size);
385
386/**
387 * blk_queue_alignment_offset - set physical block alignment offset
388 * @q: the request queue for the device
389 * @offset: alignment offset in bytes
390 *
391 * Description:
392 * Some devices are naturally misaligned to compensate for things like
393 * the legacy DOS partition table 63-sector offset. Low-level drivers
394 * should call this function for devices whose first sector is not
395 * naturally aligned.
396 */
397void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
398{
399 q->limits.alignment_offset =
400 offset & (q->limits.physical_block_size - 1);
401 q->limits.misaligned = 0;
402}
403EXPORT_SYMBOL(blk_queue_alignment_offset);
404
405/**
406 * blk_limits_io_min - set minimum request size for a device
407 * @limits: the queue limits
408 * @min: smallest I/O size in bytes
409 *
410 * Description:
411 * Some devices have an internal block size bigger than the reported
412 * hardware sector size. This function can be used to signal the
413 * smallest I/O the device can perform without incurring a performance
414 * penalty.
415 */
416void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
417{
418 limits->io_min = min;
419
420 if (limits->io_min < limits->logical_block_size)
421 limits->io_min = limits->logical_block_size;
422
423 if (limits->io_min < limits->physical_block_size)
424 limits->io_min = limits->physical_block_size;
425}
426EXPORT_SYMBOL(blk_limits_io_min);
427
428/**
429 * blk_queue_io_min - set minimum request size for the queue
430 * @q: the request queue for the device
431 * @min: smallest I/O size in bytes
432 *
433 * Description:
434 * Storage devices may report a granularity or preferred minimum I/O
435 * size which is the smallest request the device can perform without
436 * incurring a performance penalty. For disk drives this is often the
437 * physical block size. For RAID arrays it is often the stripe chunk
438 * size. A properly aligned multiple of minimum_io_size is the
439 * preferred request size for workloads where a high number of I/O
440 * operations is desired.
441 */
442void blk_queue_io_min(struct request_queue *q, unsigned int min)
443{
444 blk_limits_io_min(&q->limits, min);
445}
446EXPORT_SYMBOL(blk_queue_io_min);
447
448/**
449 * blk_limits_io_opt - set optimal request size for a device
450 * @limits: the queue limits
451 * @opt: smallest I/O size in bytes
452 *
453 * Description:
454 * Storage devices may report an optimal I/O size, which is the
455 * device's preferred unit for sustained I/O. This is rarely reported
456 * for disk drives. For RAID arrays it is usually the stripe width or
457 * the internal track size. A properly aligned multiple of
458 * optimal_io_size is the preferred request size for workloads where
459 * sustained throughput is desired.
460 */
461void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
462{
463 limits->io_opt = opt;
464}
465EXPORT_SYMBOL(blk_limits_io_opt);
466
467/**
468 * blk_queue_io_opt - set optimal request size for the queue
469 * @q: the request queue for the device
470 * @opt: optimal request size in bytes
471 *
472 * Description:
473 * Storage devices may report an optimal I/O size, which is the
474 * device's preferred unit for sustained I/O. This is rarely reported
475 * for disk drives. For RAID arrays it is usually the stripe width or
476 * the internal track size. A properly aligned multiple of
477 * optimal_io_size is the preferred request size for workloads where
478 * sustained throughput is desired.
479 */
480void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
481{
482 blk_limits_io_opt(&q->limits, opt);
483}
484EXPORT_SYMBOL(blk_queue_io_opt);
485
486/**
487 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
488 * @t: the stacking driver (top)
489 * @b: the underlying device (bottom)
490 **/
491void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
492{
493 blk_stack_limits(&t->limits, &b->limits, 0);
494}
495EXPORT_SYMBOL(blk_queue_stack_limits);
496
497/**
498 * blk_stack_limits - adjust queue_limits for stacked devices
499 * @t: the stacking driver limits (top device)
500 * @b: the underlying queue limits (bottom, component device)
501 * @start: first data sector within component device
502 *
503 * Description:
504 * This function is used by stacking drivers like MD and DM to ensure
505 * that all component devices have compatible block sizes and
506 * alignments. The stacking driver must provide a queue_limits
507 * struct (top) and then iteratively call the stacking function for
508 * all component (bottom) devices. The stacking function will
509 * attempt to combine the values and ensure proper alignment.
510 *
511 * Returns 0 if the top and bottom queue_limits are compatible. The
512 * top device's block sizes and alignment offsets may be adjusted to
513 * ensure alignment with the bottom device. If no compatible sizes
514 * and alignments exist, -1 is returned and the resulting top
515 * queue_limits will have the misaligned flag set to indicate that
516 * the alignment_offset is undefined.
517 */
518int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
519 sector_t start)
520{
521 unsigned int top, bottom, alignment, ret = 0;
522
523 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
524 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
525 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
526 t->max_write_same_sectors = min(t->max_write_same_sectors,
527 b->max_write_same_sectors);
528 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
529
530 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
531 b->seg_boundary_mask);
532 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
533 b->virt_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)) {
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_not_zero(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_not_zero(t->alignment_offset, alignment)
601 % max(t->physical_block_size, t->io_min);
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->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
626 b->max_hw_discard_sectors);
627 t->discard_granularity = max(t->discard_granularity,
628 b->discard_granularity);
629 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
630 t->discard_granularity;
631 }
632
633 return ret;
634}
635EXPORT_SYMBOL(blk_stack_limits);
636
637/**
638 * bdev_stack_limits - adjust queue limits for stacked drivers
639 * @t: the stacking driver limits (top device)
640 * @bdev: the component block_device (bottom)
641 * @start: first data sector within component device
642 *
643 * Description:
644 * Merges queue limits for a top device and a block_device. Returns
645 * 0 if alignment didn't change. Returns -1 if adding the bottom
646 * device caused misalignment.
647 */
648int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
649 sector_t start)
650{
651 struct request_queue *bq = bdev_get_queue(bdev);
652
653 start += get_start_sect(bdev);
654
655 return blk_stack_limits(t, &bq->limits, start);
656}
657EXPORT_SYMBOL(bdev_stack_limits);
658
659/**
660 * disk_stack_limits - adjust queue limits for stacked drivers
661 * @disk: MD/DM gendisk (top)
662 * @bdev: the underlying block device (bottom)
663 * @offset: offset to beginning of data within component device
664 *
665 * Description:
666 * Merges the limits for a top level gendisk and a bottom level
667 * block_device.
668 */
669void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
670 sector_t offset)
671{
672 struct request_queue *t = disk->queue;
673
674 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
675 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
676
677 disk_name(disk, 0, top);
678 bdevname(bdev, bottom);
679
680 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
681 top, bottom);
682 }
683}
684EXPORT_SYMBOL(disk_stack_limits);
685
686/**
687 * blk_queue_dma_pad - set pad mask
688 * @q: the request queue for the device
689 * @mask: pad mask
690 *
691 * Set dma pad mask.
692 *
693 * Appending pad buffer to a request modifies the last entry of a
694 * scatter list such that it includes the pad buffer.
695 **/
696void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
697{
698 q->dma_pad_mask = mask;
699}
700EXPORT_SYMBOL(blk_queue_dma_pad);
701
702/**
703 * blk_queue_update_dma_pad - update pad mask
704 * @q: the request queue for the device
705 * @mask: pad mask
706 *
707 * Update dma pad mask.
708 *
709 * Appending pad buffer to a request modifies the last entry of a
710 * scatter list such that it includes the pad buffer.
711 **/
712void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
713{
714 if (mask > q->dma_pad_mask)
715 q->dma_pad_mask = mask;
716}
717EXPORT_SYMBOL(blk_queue_update_dma_pad);
718
719/**
720 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
721 * @q: the request queue for the device
722 * @dma_drain_needed: fn which returns non-zero if drain is necessary
723 * @buf: physically contiguous buffer
724 * @size: size of the buffer in bytes
725 *
726 * Some devices have excess DMA problems and can't simply discard (or
727 * zero fill) the unwanted piece of the transfer. They have to have a
728 * real area of memory to transfer it into. The use case for this is
729 * ATAPI devices in DMA mode. If the packet command causes a transfer
730 * bigger than the transfer size some HBAs will lock up if there
731 * aren't DMA elements to contain the excess transfer. What this API
732 * does is adjust the queue so that the buf is always appended
733 * silently to the scatterlist.
734 *
735 * Note: This routine adjusts max_hw_segments to make room for appending
736 * the drain buffer. If you call blk_queue_max_segments() after calling
737 * this routine, you must set the limit to one fewer than your device
738 * can support otherwise there won't be room for the drain buffer.
739 */
740int blk_queue_dma_drain(struct request_queue *q,
741 dma_drain_needed_fn *dma_drain_needed,
742 void *buf, unsigned int size)
743{
744 if (queue_max_segments(q) < 2)
745 return -EINVAL;
746 /* make room for appending the drain */
747 blk_queue_max_segments(q, queue_max_segments(q) - 1);
748 q->dma_drain_needed = dma_drain_needed;
749 q->dma_drain_buffer = buf;
750 q->dma_drain_size = size;
751
752 return 0;
753}
754EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
755
756/**
757 * blk_queue_segment_boundary - set boundary rules for segment merging
758 * @q: the request queue for the device
759 * @mask: the memory boundary mask
760 **/
761void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
762{
763 if (mask < PAGE_SIZE - 1) {
764 mask = PAGE_SIZE - 1;
765 printk(KERN_INFO "%s: set to minimum %lx\n",
766 __func__, mask);
767 }
768
769 q->limits.seg_boundary_mask = mask;
770}
771EXPORT_SYMBOL(blk_queue_segment_boundary);
772
773/**
774 * blk_queue_virt_boundary - set boundary rules for bio merging
775 * @q: the request queue for the device
776 * @mask: the memory boundary mask
777 **/
778void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
779{
780 q->limits.virt_boundary_mask = mask;
781}
782EXPORT_SYMBOL(blk_queue_virt_boundary);
783
784/**
785 * blk_queue_dma_alignment - set dma length and memory alignment
786 * @q: the request queue for the device
787 * @mask: alignment mask
788 *
789 * description:
790 * set required memory and length alignment for direct dma transactions.
791 * this is used when building direct io requests for the queue.
792 *
793 **/
794void blk_queue_dma_alignment(struct request_queue *q, int mask)
795{
796 q->dma_alignment = mask;
797}
798EXPORT_SYMBOL(blk_queue_dma_alignment);
799
800/**
801 * blk_queue_update_dma_alignment - update dma length and memory alignment
802 * @q: the request queue for the device
803 * @mask: alignment mask
804 *
805 * description:
806 * update required memory and length alignment for direct dma transactions.
807 * If the requested alignment is larger than the current alignment, then
808 * the current queue alignment is updated to the new value, otherwise it
809 * is left alone. The design of this is to allow multiple objects
810 * (driver, device, transport etc) to set their respective
811 * alignments without having them interfere.
812 *
813 **/
814void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
815{
816 BUG_ON(mask > PAGE_SIZE);
817
818 if (mask > q->dma_alignment)
819 q->dma_alignment = mask;
820}
821EXPORT_SYMBOL(blk_queue_update_dma_alignment);
822
823/**
824 * blk_queue_flush - configure queue's cache flush capability
825 * @q: the request queue for the device
826 * @flush: 0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
827 *
828 * Tell block layer cache flush capability of @q. If it supports
829 * flushing, REQ_FLUSH should be set. If it supports bypassing
830 * write cache for individual writes, REQ_FUA should be set.
831 */
832void blk_queue_flush(struct request_queue *q, unsigned int flush)
833{
834 WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
835
836 if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
837 flush &= ~REQ_FUA;
838
839 q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
840}
841EXPORT_SYMBOL_GPL(blk_queue_flush);
842
843void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
844{
845 q->flush_not_queueable = !queueable;
846}
847EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
848
849static int __init blk_settings_init(void)
850{
851 blk_max_low_pfn = max_low_pfn - 1;
852 blk_max_pfn = max_pfn - 1;
853 return 0;
854}
855subsys_initcall(blk_settings_init);