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// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to setting various queue properties from drivers
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/bio.h>
9#include <linux/blkdev.h>
10#include <linux/memblock.h> /* for max_pfn/max_low_pfn */
11#include <linux/gcd.h>
12#include <linux/lcm.h>
13#include <linux/jiffies.h>
14#include <linux/gfp.h>
15#include <linux/dma-mapping.h>
16
17#include "blk.h"
18#include "blk-wbt.h"
19
20unsigned long blk_max_low_pfn;
21EXPORT_SYMBOL(blk_max_low_pfn);
22
23unsigned long blk_max_pfn;
24
25void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
26{
27 q->rq_timeout = timeout;
28}
29EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
30
31/**
32 * blk_set_default_limits - reset limits to default values
33 * @lim: the queue_limits structure to reset
34 *
35 * Description:
36 * Returns a queue_limit struct to its default state.
37 */
38void blk_set_default_limits(struct queue_limits *lim)
39{
40 lim->max_segments = BLK_MAX_SEGMENTS;
41 lim->max_discard_segments = 1;
42 lim->max_integrity_segments = 0;
43 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
44 lim->virt_boundary_mask = 0;
45 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
46 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
47 lim->max_dev_sectors = 0;
48 lim->chunk_sectors = 0;
49 lim->max_write_same_sectors = 0;
50 lim->max_write_zeroes_sectors = 0;
51 lim->max_discard_sectors = 0;
52 lim->max_hw_discard_sectors = 0;
53 lim->discard_granularity = 0;
54 lim->discard_alignment = 0;
55 lim->discard_misaligned = 0;
56 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
57 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
58 lim->alignment_offset = 0;
59 lim->io_opt = 0;
60 lim->misaligned = 0;
61 lim->zoned = BLK_ZONED_NONE;
62}
63EXPORT_SYMBOL(blk_set_default_limits);
64
65/**
66 * blk_set_stacking_limits - set default limits for stacking devices
67 * @lim: the queue_limits structure to reset
68 *
69 * Description:
70 * Returns a queue_limit struct to its default state. Should be used
71 * by stacking drivers like DM that have no internal limits.
72 */
73void blk_set_stacking_limits(struct queue_limits *lim)
74{
75 blk_set_default_limits(lim);
76
77 /* Inherit limits from component devices */
78 lim->max_segments = USHRT_MAX;
79 lim->max_discard_segments = USHRT_MAX;
80 lim->max_hw_sectors = UINT_MAX;
81 lim->max_segment_size = UINT_MAX;
82 lim->max_sectors = UINT_MAX;
83 lim->max_dev_sectors = UINT_MAX;
84 lim->max_write_same_sectors = UINT_MAX;
85 lim->max_write_zeroes_sectors = UINT_MAX;
86}
87EXPORT_SYMBOL(blk_set_stacking_limits);
88
89/**
90 * blk_queue_make_request - define an alternate make_request function for a device
91 * @q: the request queue for the device to be affected
92 * @mfn: the alternate make_request function
93 *
94 * Description:
95 * The normal way for &struct bios to be passed to a device
96 * driver is for them to be collected into requests on a request
97 * queue, and then to allow the device driver to select requests
98 * off that queue when it is ready. This works well for many block
99 * devices. However some block devices (typically virtual devices
100 * such as md or lvm) do not benefit from the processing on the
101 * request queue, and are served best by having the requests passed
102 * directly to them. This can be achieved by providing a function
103 * to blk_queue_make_request().
104 *
105 * Caveat:
106 * The driver that does this *must* be able to deal appropriately
107 * with buffers in "highmemory". This can be accomplished by either calling
108 * kmap_atomic() to get a temporary kernel mapping, or by calling
109 * blk_queue_bounce() to create a buffer in normal memory.
110 **/
111void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
112{
113 /*
114 * set defaults
115 */
116 q->nr_requests = BLKDEV_MAX_RQ;
117
118 q->make_request_fn = mfn;
119 blk_queue_dma_alignment(q, 511);
120
121 blk_set_default_limits(&q->limits);
122}
123EXPORT_SYMBOL(blk_queue_make_request);
124
125/**
126 * blk_queue_bounce_limit - set bounce buffer limit for queue
127 * @q: the request queue for the device
128 * @max_addr: the maximum address the device can handle
129 *
130 * Description:
131 * Different hardware can have different requirements as to what pages
132 * it can do I/O directly to. A low level driver can call
133 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
134 * buffers for doing I/O to pages residing above @max_addr.
135 **/
136void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
137{
138 unsigned long b_pfn = max_addr >> PAGE_SHIFT;
139 int dma = 0;
140
141 q->bounce_gfp = GFP_NOIO;
142#if BITS_PER_LONG == 64
143 /*
144 * Assume anything <= 4GB can be handled by IOMMU. Actually
145 * some IOMMUs can handle everything, but I don't know of a
146 * way to test this here.
147 */
148 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
149 dma = 1;
150 q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
151#else
152 if (b_pfn < blk_max_low_pfn)
153 dma = 1;
154 q->limits.bounce_pfn = b_pfn;
155#endif
156 if (dma) {
157 init_emergency_isa_pool();
158 q->bounce_gfp = GFP_NOIO | GFP_DMA;
159 q->limits.bounce_pfn = b_pfn;
160 }
161}
162EXPORT_SYMBOL(blk_queue_bounce_limit);
163
164/**
165 * blk_queue_max_hw_sectors - set max sectors for a request for this queue
166 * @q: the request queue for the device
167 * @max_hw_sectors: max hardware sectors in the usual 512b unit
168 *
169 * Description:
170 * Enables a low level driver to set a hard upper limit,
171 * max_hw_sectors, on the size of requests. max_hw_sectors is set by
172 * the device driver based upon the capabilities of the I/O
173 * controller.
174 *
175 * max_dev_sectors is a hard limit imposed by the storage device for
176 * READ/WRITE requests. It is set by the disk driver.
177 *
178 * max_sectors is a soft limit imposed by the block layer for
179 * filesystem type requests. This value can be overridden on a
180 * per-device basis in /sys/block/<device>/queue/max_sectors_kb.
181 * The soft limit can not exceed max_hw_sectors.
182 **/
183void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
184{
185 struct queue_limits *limits = &q->limits;
186 unsigned int max_sectors;
187
188 if ((max_hw_sectors << 9) < PAGE_SIZE) {
189 max_hw_sectors = 1 << (PAGE_SHIFT - 9);
190 printk(KERN_INFO "%s: set to minimum %d\n",
191 __func__, max_hw_sectors);
192 }
193
194 limits->max_hw_sectors = max_hw_sectors;
195 max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
196 max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
197 limits->max_sectors = max_sectors;
198 q->backing_dev_info->io_pages = max_sectors >> (PAGE_SHIFT - 9);
199}
200EXPORT_SYMBOL(blk_queue_max_hw_sectors);
201
202/**
203 * blk_queue_chunk_sectors - set size of the chunk for this queue
204 * @q: the request queue for the device
205 * @chunk_sectors: chunk sectors in the usual 512b unit
206 *
207 * Description:
208 * If a driver doesn't want IOs to cross a given chunk size, it can set
209 * this limit and prevent merging across chunks. Note that the chunk size
210 * must currently be a power-of-2 in sectors. Also note that the block
211 * layer must accept a page worth of data at any offset. So if the
212 * crossing of chunks is a hard limitation in the driver, it must still be
213 * prepared to split single page bios.
214 **/
215void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
216{
217 BUG_ON(!is_power_of_2(chunk_sectors));
218 q->limits.chunk_sectors = chunk_sectors;
219}
220EXPORT_SYMBOL(blk_queue_chunk_sectors);
221
222/**
223 * blk_queue_max_discard_sectors - set max sectors for a single discard
224 * @q: the request queue for the device
225 * @max_discard_sectors: maximum number of sectors to discard
226 **/
227void blk_queue_max_discard_sectors(struct request_queue *q,
228 unsigned int max_discard_sectors)
229{
230 q->limits.max_hw_discard_sectors = max_discard_sectors;
231 q->limits.max_discard_sectors = max_discard_sectors;
232}
233EXPORT_SYMBOL(blk_queue_max_discard_sectors);
234
235/**
236 * blk_queue_max_write_same_sectors - set max sectors for a single write same
237 * @q: the request queue for the device
238 * @max_write_same_sectors: maximum number of sectors to write per command
239 **/
240void blk_queue_max_write_same_sectors(struct request_queue *q,
241 unsigned int max_write_same_sectors)
242{
243 q->limits.max_write_same_sectors = max_write_same_sectors;
244}
245EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
246
247/**
248 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
249 * write zeroes
250 * @q: the request queue for the device
251 * @max_write_zeroes_sectors: maximum number of sectors to write per command
252 **/
253void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
254 unsigned int max_write_zeroes_sectors)
255{
256 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
257}
258EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
259
260/**
261 * blk_queue_max_segments - set max hw segments for a request for this queue
262 * @q: the request queue for the device
263 * @max_segments: max number of segments
264 *
265 * Description:
266 * Enables a low level driver to set an upper limit on the number of
267 * hw data segments in a request.
268 **/
269void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
270{
271 if (!max_segments) {
272 max_segments = 1;
273 printk(KERN_INFO "%s: set to minimum %d\n",
274 __func__, max_segments);
275 }
276
277 q->limits.max_segments = max_segments;
278}
279EXPORT_SYMBOL(blk_queue_max_segments);
280
281/**
282 * blk_queue_max_discard_segments - set max segments for discard requests
283 * @q: the request queue for the device
284 * @max_segments: max number of segments
285 *
286 * Description:
287 * Enables a low level driver to set an upper limit on the number of
288 * segments in a discard request.
289 **/
290void blk_queue_max_discard_segments(struct request_queue *q,
291 unsigned short max_segments)
292{
293 q->limits.max_discard_segments = max_segments;
294}
295EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
296
297/**
298 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
299 * @q: the request queue for the device
300 * @max_size: max size of segment in bytes
301 *
302 * Description:
303 * Enables a low level driver to set an upper limit on the size of a
304 * coalesced segment
305 **/
306void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
307{
308 if (max_size < PAGE_SIZE) {
309 max_size = PAGE_SIZE;
310 printk(KERN_INFO "%s: set to minimum %d\n",
311 __func__, max_size);
312 }
313
314 /* see blk_queue_virt_boundary() for the explanation */
315 WARN_ON_ONCE(q->limits.virt_boundary_mask);
316
317 q->limits.max_segment_size = max_size;
318}
319EXPORT_SYMBOL(blk_queue_max_segment_size);
320
321/**
322 * blk_queue_logical_block_size - set logical block size for the queue
323 * @q: the request queue for the device
324 * @size: the logical block size, in bytes
325 *
326 * Description:
327 * This should be set to the lowest possible block size that the
328 * storage device can address. The default of 512 covers most
329 * hardware.
330 **/
331void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
332{
333 q->limits.logical_block_size = size;
334
335 if (q->limits.physical_block_size < size)
336 q->limits.physical_block_size = size;
337
338 if (q->limits.io_min < q->limits.physical_block_size)
339 q->limits.io_min = q->limits.physical_block_size;
340}
341EXPORT_SYMBOL(blk_queue_logical_block_size);
342
343/**
344 * blk_queue_physical_block_size - set physical block size for the queue
345 * @q: the request queue for the device
346 * @size: the physical block size, in bytes
347 *
348 * Description:
349 * This should be set to the lowest possible sector size that the
350 * hardware can operate on without reverting to read-modify-write
351 * operations.
352 */
353void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
354{
355 q->limits.physical_block_size = size;
356
357 if (q->limits.physical_block_size < q->limits.logical_block_size)
358 q->limits.physical_block_size = q->limits.logical_block_size;
359
360 if (q->limits.io_min < q->limits.physical_block_size)
361 q->limits.io_min = q->limits.physical_block_size;
362}
363EXPORT_SYMBOL(blk_queue_physical_block_size);
364
365/**
366 * blk_queue_alignment_offset - set physical block alignment offset
367 * @q: the request queue for the device
368 * @offset: alignment offset in bytes
369 *
370 * Description:
371 * Some devices are naturally misaligned to compensate for things like
372 * the legacy DOS partition table 63-sector offset. Low-level drivers
373 * should call this function for devices whose first sector is not
374 * naturally aligned.
375 */
376void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
377{
378 q->limits.alignment_offset =
379 offset & (q->limits.physical_block_size - 1);
380 q->limits.misaligned = 0;
381}
382EXPORT_SYMBOL(blk_queue_alignment_offset);
383
384/**
385 * blk_limits_io_min - set minimum request size for a device
386 * @limits: the queue limits
387 * @min: smallest I/O size in bytes
388 *
389 * Description:
390 * Some devices have an internal block size bigger than the reported
391 * hardware sector size. This function can be used to signal the
392 * smallest I/O the device can perform without incurring a performance
393 * penalty.
394 */
395void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
396{
397 limits->io_min = min;
398
399 if (limits->io_min < limits->logical_block_size)
400 limits->io_min = limits->logical_block_size;
401
402 if (limits->io_min < limits->physical_block_size)
403 limits->io_min = limits->physical_block_size;
404}
405EXPORT_SYMBOL(blk_limits_io_min);
406
407/**
408 * blk_queue_io_min - set minimum request size for the queue
409 * @q: the request queue for the device
410 * @min: smallest I/O size in bytes
411 *
412 * Description:
413 * Storage devices may report a granularity or preferred minimum I/O
414 * size which is the smallest request the device can perform without
415 * incurring a performance penalty. For disk drives this is often the
416 * physical block size. For RAID arrays it is often the stripe chunk
417 * size. A properly aligned multiple of minimum_io_size is the
418 * preferred request size for workloads where a high number of I/O
419 * operations is desired.
420 */
421void blk_queue_io_min(struct request_queue *q, unsigned int min)
422{
423 blk_limits_io_min(&q->limits, min);
424}
425EXPORT_SYMBOL(blk_queue_io_min);
426
427/**
428 * blk_limits_io_opt - set optimal request size for a device
429 * @limits: the queue limits
430 * @opt: smallest I/O size in bytes
431 *
432 * Description:
433 * Storage devices may report an optimal I/O size, which is the
434 * device's preferred unit for sustained I/O. This is rarely reported
435 * for disk drives. For RAID arrays it is usually the stripe width or
436 * the internal track size. A properly aligned multiple of
437 * optimal_io_size is the preferred request size for workloads where
438 * sustained throughput is desired.
439 */
440void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
441{
442 limits->io_opt = opt;
443}
444EXPORT_SYMBOL(blk_limits_io_opt);
445
446/**
447 * blk_queue_io_opt - set optimal request size for the queue
448 * @q: the request queue for the device
449 * @opt: optimal request size in bytes
450 *
451 * Description:
452 * Storage devices may report an optimal I/O size, which is the
453 * device's preferred unit for sustained I/O. This is rarely reported
454 * for disk drives. For RAID arrays it is usually the stripe width or
455 * the internal track size. A properly aligned multiple of
456 * optimal_io_size is the preferred request size for workloads where
457 * sustained throughput is desired.
458 */
459void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
460{
461 blk_limits_io_opt(&q->limits, opt);
462}
463EXPORT_SYMBOL(blk_queue_io_opt);
464
465/**
466 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
467 * @t: the stacking driver (top)
468 * @b: the underlying device (bottom)
469 **/
470void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
471{
472 blk_stack_limits(&t->limits, &b->limits, 0);
473}
474EXPORT_SYMBOL(blk_queue_stack_limits);
475
476/**
477 * blk_stack_limits - adjust queue_limits for stacked devices
478 * @t: the stacking driver limits (top device)
479 * @b: the underlying queue limits (bottom, component device)
480 * @start: first data sector within component device
481 *
482 * Description:
483 * This function is used by stacking drivers like MD and DM to ensure
484 * that all component devices have compatible block sizes and
485 * alignments. The stacking driver must provide a queue_limits
486 * struct (top) and then iteratively call the stacking function for
487 * all component (bottom) devices. The stacking function will
488 * attempt to combine the values and ensure proper alignment.
489 *
490 * Returns 0 if the top and bottom queue_limits are compatible. The
491 * top device's block sizes and alignment offsets may be adjusted to
492 * ensure alignment with the bottom device. If no compatible sizes
493 * and alignments exist, -1 is returned and the resulting top
494 * queue_limits will have the misaligned flag set to indicate that
495 * the alignment_offset is undefined.
496 */
497int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
498 sector_t start)
499{
500 unsigned int top, bottom, alignment, ret = 0;
501
502 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
503 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
504 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
505 t->max_write_same_sectors = min(t->max_write_same_sectors,
506 b->max_write_same_sectors);
507 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
508 b->max_write_zeroes_sectors);
509 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
510
511 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
512 b->seg_boundary_mask);
513 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
514 b->virt_boundary_mask);
515
516 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
517 t->max_discard_segments = min_not_zero(t->max_discard_segments,
518 b->max_discard_segments);
519 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
520 b->max_integrity_segments);
521
522 t->max_segment_size = min_not_zero(t->max_segment_size,
523 b->max_segment_size);
524
525 t->misaligned |= b->misaligned;
526
527 alignment = queue_limit_alignment_offset(b, start);
528
529 /* Bottom device has different alignment. Check that it is
530 * compatible with the current top alignment.
531 */
532 if (t->alignment_offset != alignment) {
533
534 top = max(t->physical_block_size, t->io_min)
535 + t->alignment_offset;
536 bottom = max(b->physical_block_size, b->io_min) + alignment;
537
538 /* Verify that top and bottom intervals line up */
539 if (max(top, bottom) % min(top, bottom)) {
540 t->misaligned = 1;
541 ret = -1;
542 }
543 }
544
545 t->logical_block_size = max(t->logical_block_size,
546 b->logical_block_size);
547
548 t->physical_block_size = max(t->physical_block_size,
549 b->physical_block_size);
550
551 t->io_min = max(t->io_min, b->io_min);
552 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
553
554 /* Physical block size a multiple of the logical block size? */
555 if (t->physical_block_size & (t->logical_block_size - 1)) {
556 t->physical_block_size = t->logical_block_size;
557 t->misaligned = 1;
558 ret = -1;
559 }
560
561 /* Minimum I/O a multiple of the physical block size? */
562 if (t->io_min & (t->physical_block_size - 1)) {
563 t->io_min = t->physical_block_size;
564 t->misaligned = 1;
565 ret = -1;
566 }
567
568 /* Optimal I/O a multiple of the physical block size? */
569 if (t->io_opt & (t->physical_block_size - 1)) {
570 t->io_opt = 0;
571 t->misaligned = 1;
572 ret = -1;
573 }
574
575 t->raid_partial_stripes_expensive =
576 max(t->raid_partial_stripes_expensive,
577 b->raid_partial_stripes_expensive);
578
579 /* Find lowest common alignment_offset */
580 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
581 % max(t->physical_block_size, t->io_min);
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)) != 0)
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->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
606 b->max_hw_discard_sectors);
607 t->discard_granularity = max(t->discard_granularity,
608 b->discard_granularity);
609 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
610 t->discard_granularity;
611 }
612
613 if (b->chunk_sectors)
614 t->chunk_sectors = min_not_zero(t->chunk_sectors,
615 b->chunk_sectors);
616
617 return ret;
618}
619EXPORT_SYMBOL(blk_stack_limits);
620
621/**
622 * bdev_stack_limits - adjust queue limits for stacked drivers
623 * @t: the stacking driver limits (top device)
624 * @bdev: the component block_device (bottom)
625 * @start: first data sector within component device
626 *
627 * Description:
628 * Merges queue limits for a top device and a block_device. Returns
629 * 0 if alignment didn't change. Returns -1 if adding the bottom
630 * device caused misalignment.
631 */
632int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
633 sector_t start)
634{
635 struct request_queue *bq = bdev_get_queue(bdev);
636
637 start += get_start_sect(bdev);
638
639 return blk_stack_limits(t, &bq->limits, start);
640}
641EXPORT_SYMBOL(bdev_stack_limits);
642
643/**
644 * disk_stack_limits - adjust queue limits for stacked drivers
645 * @disk: MD/DM gendisk (top)
646 * @bdev: the underlying block device (bottom)
647 * @offset: offset to beginning of data within component device
648 *
649 * Description:
650 * Merges the limits for a top level gendisk and a bottom level
651 * block_device.
652 */
653void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
654 sector_t offset)
655{
656 struct request_queue *t = disk->queue;
657
658 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
659 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
660
661 disk_name(disk, 0, top);
662 bdevname(bdev, bottom);
663
664 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
665 top, bottom);
666 }
667}
668EXPORT_SYMBOL(disk_stack_limits);
669
670/**
671 * blk_queue_update_dma_pad - update pad mask
672 * @q: the request queue for the device
673 * @mask: pad mask
674 *
675 * Update dma pad mask.
676 *
677 * Appending pad buffer to a request modifies the last entry of a
678 * scatter list such that it includes the pad buffer.
679 **/
680void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
681{
682 if (mask > q->dma_pad_mask)
683 q->dma_pad_mask = mask;
684}
685EXPORT_SYMBOL(blk_queue_update_dma_pad);
686
687/**
688 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
689 * @q: the request queue for the device
690 * @dma_drain_needed: fn which returns non-zero if drain is necessary
691 * @buf: physically contiguous buffer
692 * @size: size of the buffer in bytes
693 *
694 * Some devices have excess DMA problems and can't simply discard (or
695 * zero fill) the unwanted piece of the transfer. They have to have a
696 * real area of memory to transfer it into. The use case for this is
697 * ATAPI devices in DMA mode. If the packet command causes a transfer
698 * bigger than the transfer size some HBAs will lock up if there
699 * aren't DMA elements to contain the excess transfer. What this API
700 * does is adjust the queue so that the buf is always appended
701 * silently to the scatterlist.
702 *
703 * Note: This routine adjusts max_hw_segments to make room for appending
704 * the drain buffer. If you call blk_queue_max_segments() after calling
705 * this routine, you must set the limit to one fewer than your device
706 * can support otherwise there won't be room for the drain buffer.
707 */
708int blk_queue_dma_drain(struct request_queue *q,
709 dma_drain_needed_fn *dma_drain_needed,
710 void *buf, unsigned int size)
711{
712 if (queue_max_segments(q) < 2)
713 return -EINVAL;
714 /* make room for appending the drain */
715 blk_queue_max_segments(q, queue_max_segments(q) - 1);
716 q->dma_drain_needed = dma_drain_needed;
717 q->dma_drain_buffer = buf;
718 q->dma_drain_size = size;
719
720 return 0;
721}
722EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
723
724/**
725 * blk_queue_segment_boundary - set boundary rules for segment merging
726 * @q: the request queue for the device
727 * @mask: the memory boundary mask
728 **/
729void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
730{
731 if (mask < PAGE_SIZE - 1) {
732 mask = PAGE_SIZE - 1;
733 printk(KERN_INFO "%s: set to minimum %lx\n",
734 __func__, mask);
735 }
736
737 q->limits.seg_boundary_mask = mask;
738}
739EXPORT_SYMBOL(blk_queue_segment_boundary);
740
741/**
742 * blk_queue_virt_boundary - set boundary rules for bio merging
743 * @q: the request queue for the device
744 * @mask: the memory boundary mask
745 **/
746void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
747{
748 q->limits.virt_boundary_mask = mask;
749
750 /*
751 * Devices that require a virtual boundary do not support scatter/gather
752 * I/O natively, but instead require a descriptor list entry for each
753 * page (which might not be idential to the Linux PAGE_SIZE). Because
754 * of that they are not limited by our notion of "segment size".
755 */
756 if (mask)
757 q->limits.max_segment_size = UINT_MAX;
758}
759EXPORT_SYMBOL(blk_queue_virt_boundary);
760
761/**
762 * blk_queue_dma_alignment - set dma length and memory alignment
763 * @q: the request queue for the device
764 * @mask: alignment mask
765 *
766 * description:
767 * set required memory and length alignment for direct dma transactions.
768 * this is used when building direct io requests for the queue.
769 *
770 **/
771void blk_queue_dma_alignment(struct request_queue *q, int mask)
772{
773 q->dma_alignment = mask;
774}
775EXPORT_SYMBOL(blk_queue_dma_alignment);
776
777/**
778 * blk_queue_update_dma_alignment - update dma length and memory alignment
779 * @q: the request queue for the device
780 * @mask: alignment mask
781 *
782 * description:
783 * update required memory and length alignment for direct dma transactions.
784 * If the requested alignment is larger than the current alignment, then
785 * the current queue alignment is updated to the new value, otherwise it
786 * is left alone. The design of this is to allow multiple objects
787 * (driver, device, transport etc) to set their respective
788 * alignments without having them interfere.
789 *
790 **/
791void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
792{
793 BUG_ON(mask > PAGE_SIZE);
794
795 if (mask > q->dma_alignment)
796 q->dma_alignment = mask;
797}
798EXPORT_SYMBOL(blk_queue_update_dma_alignment);
799
800/**
801 * blk_set_queue_depth - tell the block layer about the device queue depth
802 * @q: the request queue for the device
803 * @depth: queue depth
804 *
805 */
806void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
807{
808 q->queue_depth = depth;
809 rq_qos_queue_depth_changed(q);
810}
811EXPORT_SYMBOL(blk_set_queue_depth);
812
813/**
814 * blk_queue_write_cache - configure queue's write cache
815 * @q: the request queue for the device
816 * @wc: write back cache on or off
817 * @fua: device supports FUA writes, if true
818 *
819 * Tell the block layer about the write cache of @q.
820 */
821void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
822{
823 if (wc)
824 blk_queue_flag_set(QUEUE_FLAG_WC, q);
825 else
826 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
827 if (fua)
828 blk_queue_flag_set(QUEUE_FLAG_FUA, q);
829 else
830 blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
831
832 wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
833}
834EXPORT_SYMBOL_GPL(blk_queue_write_cache);
835
836/**
837 * blk_queue_required_elevator_features - Set a queue required elevator features
838 * @q: the request queue for the target device
839 * @features: Required elevator features OR'ed together
840 *
841 * Tell the block layer that for the device controlled through @q, only the
842 * only elevators that can be used are those that implement at least the set of
843 * features specified by @features.
844 */
845void blk_queue_required_elevator_features(struct request_queue *q,
846 unsigned int features)
847{
848 q->required_elevator_features = features;
849}
850EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
851
852/**
853 * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
854 * @q: the request queue for the device
855 * @dev: the device pointer for dma
856 *
857 * Tell the block layer about merging the segments by dma map of @q.
858 */
859bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
860 struct device *dev)
861{
862 unsigned long boundary = dma_get_merge_boundary(dev);
863
864 if (!boundary)
865 return false;
866
867 /* No need to update max_segment_size. see blk_queue_virt_boundary() */
868 blk_queue_virt_boundary(q, boundary);
869
870 return true;
871}
872EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
873
874static int __init blk_settings_init(void)
875{
876 blk_max_low_pfn = max_low_pfn - 1;
877 blk_max_pfn = max_pfn - 1;
878 return 0;
879}
880subsys_initcall(blk_settings_init);