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#include "blk-wbt.h"
17
18unsigned long blk_max_low_pfn;
19EXPORT_SYMBOL(blk_max_low_pfn);
20
21unsigned long blk_max_pfn;
22
23/**
24 * blk_queue_prep_rq - set a prepare_request function for queue
25 * @q: queue
26 * @pfn: prepare_request function
27 *
28 * It's possible for a queue to register a prepare_request callback which
29 * is invoked before the request is handed to the request_fn. The goal of
30 * the function is to prepare a request for I/O, it can be used to build a
31 * cdb from the request data for instance.
32 *
33 */
34void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
35{
36 q->prep_rq_fn = pfn;
37}
38EXPORT_SYMBOL(blk_queue_prep_rq);
39
40/**
41 * blk_queue_unprep_rq - set an unprepare_request function for queue
42 * @q: queue
43 * @ufn: unprepare_request function
44 *
45 * It's possible for a queue to register an unprepare_request callback
46 * which is invoked before the request is finally completed. The goal
47 * of the function is to deallocate any data that was allocated in the
48 * prepare_request callback.
49 *
50 */
51void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn)
52{
53 q->unprep_rq_fn = ufn;
54}
55EXPORT_SYMBOL(blk_queue_unprep_rq);
56
57void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
58{
59 q->softirq_done_fn = fn;
60}
61EXPORT_SYMBOL(blk_queue_softirq_done);
62
63void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
64{
65 q->rq_timeout = timeout;
66}
67EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
68
69void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
70{
71 WARN_ON_ONCE(q->mq_ops);
72 q->rq_timed_out_fn = fn;
73}
74EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
75
76void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
77{
78 q->lld_busy_fn = fn;
79}
80EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
81
82/**
83 * blk_set_default_limits - reset limits to default values
84 * @lim: the queue_limits structure to reset
85 *
86 * Description:
87 * Returns a queue_limit struct to its default state.
88 */
89void blk_set_default_limits(struct queue_limits *lim)
90{
91 lim->max_segments = BLK_MAX_SEGMENTS;
92 lim->max_discard_segments = 1;
93 lim->max_integrity_segments = 0;
94 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
95 lim->virt_boundary_mask = 0;
96 lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
97 lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
98 lim->max_dev_sectors = 0;
99 lim->chunk_sectors = 0;
100 lim->max_write_same_sectors = 0;
101 lim->max_write_zeroes_sectors = 0;
102 lim->max_discard_sectors = 0;
103 lim->max_hw_discard_sectors = 0;
104 lim->discard_granularity = 0;
105 lim->discard_alignment = 0;
106 lim->discard_misaligned = 0;
107 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
108 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
109 lim->alignment_offset = 0;
110 lim->io_opt = 0;
111 lim->misaligned = 0;
112 lim->cluster = 1;
113 lim->zoned = BLK_ZONED_NONE;
114}
115EXPORT_SYMBOL(blk_set_default_limits);
116
117/**
118 * blk_set_stacking_limits - set default limits for stacking devices
119 * @lim: the queue_limits structure to reset
120 *
121 * Description:
122 * Returns a queue_limit struct to its default state. Should be used
123 * by stacking drivers like DM that have no internal limits.
124 */
125void blk_set_stacking_limits(struct queue_limits *lim)
126{
127 blk_set_default_limits(lim);
128
129 /* Inherit limits from component devices */
130 lim->max_segments = USHRT_MAX;
131 lim->max_discard_segments = 1;
132 lim->max_hw_sectors = UINT_MAX;
133 lim->max_segment_size = UINT_MAX;
134 lim->max_sectors = UINT_MAX;
135 lim->max_dev_sectors = UINT_MAX;
136 lim->max_write_same_sectors = UINT_MAX;
137 lim->max_write_zeroes_sectors = UINT_MAX;
138}
139EXPORT_SYMBOL(blk_set_stacking_limits);
140
141/**
142 * blk_queue_make_request - define an alternate make_request function for a device
143 * @q: the request queue for the device to be affected
144 * @mfn: the alternate make_request function
145 *
146 * Description:
147 * The normal way for &struct bios to be passed to a device
148 * driver is for them to be collected into requests on a request
149 * queue, and then to allow the device driver to select requests
150 * off that queue when it is ready. This works well for many block
151 * devices. However some block devices (typically virtual devices
152 * such as md or lvm) do not benefit from the processing on the
153 * request queue, and are served best by having the requests passed
154 * directly to them. This can be achieved by providing a function
155 * to blk_queue_make_request().
156 *
157 * Caveat:
158 * The driver that does this *must* be able to deal appropriately
159 * with buffers in "highmemory". This can be accomplished by either calling
160 * kmap_atomic() to get a temporary kernel mapping, or by calling
161 * blk_queue_bounce() to create a buffer in normal memory.
162 **/
163void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
164{
165 /*
166 * set defaults
167 */
168 q->nr_requests = BLKDEV_MAX_RQ;
169
170 q->make_request_fn = mfn;
171 blk_queue_dma_alignment(q, 511);
172 blk_queue_congestion_threshold(q);
173 q->nr_batching = BLK_BATCH_REQ;
174
175 blk_set_default_limits(&q->limits);
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 q->backing_dev_info->io_pages = max_sectors >> (PAGE_SHIFT - 9);
253}
254EXPORT_SYMBOL(blk_queue_max_hw_sectors);
255
256/**
257 * blk_queue_chunk_sectors - set size of the chunk for this queue
258 * @q: the request queue for the device
259 * @chunk_sectors: chunk sectors in the usual 512b unit
260 *
261 * Description:
262 * If a driver doesn't want IOs to cross a given chunk size, it can set
263 * this limit and prevent merging across chunks. Note that the chunk size
264 * must currently be a power-of-2 in sectors. Also note that the block
265 * layer must accept a page worth of data at any offset. So if the
266 * crossing of chunks is a hard limitation in the driver, it must still be
267 * prepared to split single page bios.
268 **/
269void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
270{
271 BUG_ON(!is_power_of_2(chunk_sectors));
272 q->limits.chunk_sectors = chunk_sectors;
273}
274EXPORT_SYMBOL(blk_queue_chunk_sectors);
275
276/**
277 * blk_queue_max_discard_sectors - set max sectors for a single discard
278 * @q: the request queue for the device
279 * @max_discard_sectors: maximum number of sectors to discard
280 **/
281void blk_queue_max_discard_sectors(struct request_queue *q,
282 unsigned int max_discard_sectors)
283{
284 q->limits.max_hw_discard_sectors = max_discard_sectors;
285 q->limits.max_discard_sectors = max_discard_sectors;
286}
287EXPORT_SYMBOL(blk_queue_max_discard_sectors);
288
289/**
290 * blk_queue_max_write_same_sectors - set max sectors for a single write same
291 * @q: the request queue for the device
292 * @max_write_same_sectors: maximum number of sectors to write per command
293 **/
294void blk_queue_max_write_same_sectors(struct request_queue *q,
295 unsigned int max_write_same_sectors)
296{
297 q->limits.max_write_same_sectors = max_write_same_sectors;
298}
299EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
300
301/**
302 * blk_queue_max_write_zeroes_sectors - set max sectors for a single
303 * write zeroes
304 * @q: the request queue for the device
305 * @max_write_zeroes_sectors: maximum number of sectors to write per command
306 **/
307void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
308 unsigned int max_write_zeroes_sectors)
309{
310 q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
311}
312EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
313
314/**
315 * blk_queue_max_segments - set max hw segments for a request for this queue
316 * @q: the request queue for the device
317 * @max_segments: max number of segments
318 *
319 * Description:
320 * Enables a low level driver to set an upper limit on the number of
321 * hw data segments in a request.
322 **/
323void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
324{
325 if (!max_segments) {
326 max_segments = 1;
327 printk(KERN_INFO "%s: set to minimum %d\n",
328 __func__, max_segments);
329 }
330
331 q->limits.max_segments = max_segments;
332}
333EXPORT_SYMBOL(blk_queue_max_segments);
334
335/**
336 * blk_queue_max_discard_segments - set max segments for discard requests
337 * @q: the request queue for the device
338 * @max_segments: max number of segments
339 *
340 * Description:
341 * Enables a low level driver to set an upper limit on the number of
342 * segments in a discard request.
343 **/
344void blk_queue_max_discard_segments(struct request_queue *q,
345 unsigned short max_segments)
346{
347 q->limits.max_discard_segments = max_segments;
348}
349EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
350
351/**
352 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
353 * @q: the request queue for the device
354 * @max_size: max size of segment in bytes
355 *
356 * Description:
357 * Enables a low level driver to set an upper limit on the size of a
358 * coalesced segment
359 **/
360void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
361{
362 if (max_size < PAGE_SIZE) {
363 max_size = PAGE_SIZE;
364 printk(KERN_INFO "%s: set to minimum %d\n",
365 __func__, max_size);
366 }
367
368 q->limits.max_segment_size = max_size;
369}
370EXPORT_SYMBOL(blk_queue_max_segment_size);
371
372/**
373 * blk_queue_logical_block_size - set logical block size for the queue
374 * @q: the request queue for the device
375 * @size: the logical block size, in bytes
376 *
377 * Description:
378 * This should be set to the lowest possible block size that the
379 * storage device can address. The default of 512 covers most
380 * hardware.
381 **/
382void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
383{
384 q->limits.logical_block_size = size;
385
386 if (q->limits.physical_block_size < size)
387 q->limits.physical_block_size = size;
388
389 if (q->limits.io_min < q->limits.physical_block_size)
390 q->limits.io_min = q->limits.physical_block_size;
391}
392EXPORT_SYMBOL(blk_queue_logical_block_size);
393
394/**
395 * blk_queue_physical_block_size - set physical block size for the queue
396 * @q: the request queue for the device
397 * @size: the physical block size, in bytes
398 *
399 * Description:
400 * This should be set to the lowest possible sector size that the
401 * hardware can operate on without reverting to read-modify-write
402 * operations.
403 */
404void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
405{
406 q->limits.physical_block_size = size;
407
408 if (q->limits.physical_block_size < q->limits.logical_block_size)
409 q->limits.physical_block_size = q->limits.logical_block_size;
410
411 if (q->limits.io_min < q->limits.physical_block_size)
412 q->limits.io_min = q->limits.physical_block_size;
413}
414EXPORT_SYMBOL(blk_queue_physical_block_size);
415
416/**
417 * blk_queue_alignment_offset - set physical block alignment offset
418 * @q: the request queue for the device
419 * @offset: alignment offset in bytes
420 *
421 * Description:
422 * Some devices are naturally misaligned to compensate for things like
423 * the legacy DOS partition table 63-sector offset. Low-level drivers
424 * should call this function for devices whose first sector is not
425 * naturally aligned.
426 */
427void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
428{
429 q->limits.alignment_offset =
430 offset & (q->limits.physical_block_size - 1);
431 q->limits.misaligned = 0;
432}
433EXPORT_SYMBOL(blk_queue_alignment_offset);
434
435/**
436 * blk_limits_io_min - set minimum request size for a device
437 * @limits: the queue limits
438 * @min: smallest I/O size in bytes
439 *
440 * Description:
441 * Some devices have an internal block size bigger than the reported
442 * hardware sector size. This function can be used to signal the
443 * smallest I/O the device can perform without incurring a performance
444 * penalty.
445 */
446void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
447{
448 limits->io_min = min;
449
450 if (limits->io_min < limits->logical_block_size)
451 limits->io_min = limits->logical_block_size;
452
453 if (limits->io_min < limits->physical_block_size)
454 limits->io_min = limits->physical_block_size;
455}
456EXPORT_SYMBOL(blk_limits_io_min);
457
458/**
459 * blk_queue_io_min - set minimum request size for the queue
460 * @q: the request queue for the device
461 * @min: smallest I/O size in bytes
462 *
463 * Description:
464 * Storage devices may report a granularity or preferred minimum I/O
465 * size which is the smallest request the device can perform without
466 * incurring a performance penalty. For disk drives this is often the
467 * physical block size. For RAID arrays it is often the stripe chunk
468 * size. A properly aligned multiple of minimum_io_size is the
469 * preferred request size for workloads where a high number of I/O
470 * operations is desired.
471 */
472void blk_queue_io_min(struct request_queue *q, unsigned int min)
473{
474 blk_limits_io_min(&q->limits, min);
475}
476EXPORT_SYMBOL(blk_queue_io_min);
477
478/**
479 * blk_limits_io_opt - set optimal request size for a device
480 * @limits: the queue limits
481 * @opt: smallest I/O size in bytes
482 *
483 * Description:
484 * Storage devices may report an optimal I/O size, which is the
485 * device's preferred unit for sustained I/O. This is rarely reported
486 * for disk drives. For RAID arrays it is usually the stripe width or
487 * the internal track size. A properly aligned multiple of
488 * optimal_io_size is the preferred request size for workloads where
489 * sustained throughput is desired.
490 */
491void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
492{
493 limits->io_opt = opt;
494}
495EXPORT_SYMBOL(blk_limits_io_opt);
496
497/**
498 * blk_queue_io_opt - set optimal request size for the queue
499 * @q: the request queue for the device
500 * @opt: optimal request size in bytes
501 *
502 * Description:
503 * Storage devices may report an optimal I/O size, which is the
504 * device's preferred unit for sustained I/O. This is rarely reported
505 * for disk drives. For RAID arrays it is usually the stripe width or
506 * the internal track size. A properly aligned multiple of
507 * optimal_io_size is the preferred request size for workloads where
508 * sustained throughput is desired.
509 */
510void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
511{
512 blk_limits_io_opt(&q->limits, opt);
513}
514EXPORT_SYMBOL(blk_queue_io_opt);
515
516/**
517 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
518 * @t: the stacking driver (top)
519 * @b: the underlying device (bottom)
520 **/
521void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
522{
523 blk_stack_limits(&t->limits, &b->limits, 0);
524}
525EXPORT_SYMBOL(blk_queue_stack_limits);
526
527/**
528 * blk_stack_limits - adjust queue_limits for stacked devices
529 * @t: the stacking driver limits (top device)
530 * @b: the underlying queue limits (bottom, component device)
531 * @start: first data sector within component device
532 *
533 * Description:
534 * This function is used by stacking drivers like MD and DM to ensure
535 * that all component devices have compatible block sizes and
536 * alignments. The stacking driver must provide a queue_limits
537 * struct (top) and then iteratively call the stacking function for
538 * all component (bottom) devices. The stacking function will
539 * attempt to combine the values and ensure proper alignment.
540 *
541 * Returns 0 if the top and bottom queue_limits are compatible. The
542 * top device's block sizes and alignment offsets may be adjusted to
543 * ensure alignment with the bottom device. If no compatible sizes
544 * and alignments exist, -1 is returned and the resulting top
545 * queue_limits will have the misaligned flag set to indicate that
546 * the alignment_offset is undefined.
547 */
548int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
549 sector_t start)
550{
551 unsigned int top, bottom, alignment, ret = 0;
552
553 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
554 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
555 t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
556 t->max_write_same_sectors = min(t->max_write_same_sectors,
557 b->max_write_same_sectors);
558 t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
559 b->max_write_zeroes_sectors);
560 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
561
562 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
563 b->seg_boundary_mask);
564 t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
565 b->virt_boundary_mask);
566
567 t->max_segments = min_not_zero(t->max_segments, b->max_segments);
568 t->max_discard_segments = min_not_zero(t->max_discard_segments,
569 b->max_discard_segments);
570 t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
571 b->max_integrity_segments);
572
573 t->max_segment_size = min_not_zero(t->max_segment_size,
574 b->max_segment_size);
575
576 t->misaligned |= b->misaligned;
577
578 alignment = queue_limit_alignment_offset(b, start);
579
580 /* Bottom device has different alignment. Check that it is
581 * compatible with the current top alignment.
582 */
583 if (t->alignment_offset != alignment) {
584
585 top = max(t->physical_block_size, t->io_min)
586 + t->alignment_offset;
587 bottom = max(b->physical_block_size, b->io_min) + alignment;
588
589 /* Verify that top and bottom intervals line up */
590 if (max(top, bottom) % min(top, bottom)) {
591 t->misaligned = 1;
592 ret = -1;
593 }
594 }
595
596 t->logical_block_size = max(t->logical_block_size,
597 b->logical_block_size);
598
599 t->physical_block_size = max(t->physical_block_size,
600 b->physical_block_size);
601
602 t->io_min = max(t->io_min, b->io_min);
603 t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
604
605 t->cluster &= b->cluster;
606
607 /* Physical block size a multiple of the logical block size? */
608 if (t->physical_block_size & (t->logical_block_size - 1)) {
609 t->physical_block_size = t->logical_block_size;
610 t->misaligned = 1;
611 ret = -1;
612 }
613
614 /* Minimum I/O a multiple of the physical block size? */
615 if (t->io_min & (t->physical_block_size - 1)) {
616 t->io_min = t->physical_block_size;
617 t->misaligned = 1;
618 ret = -1;
619 }
620
621 /* Optimal I/O a multiple of the physical block size? */
622 if (t->io_opt & (t->physical_block_size - 1)) {
623 t->io_opt = 0;
624 t->misaligned = 1;
625 ret = -1;
626 }
627
628 t->raid_partial_stripes_expensive =
629 max(t->raid_partial_stripes_expensive,
630 b->raid_partial_stripes_expensive);
631
632 /* Find lowest common alignment_offset */
633 t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
634 % max(t->physical_block_size, t->io_min);
635
636 /* Verify that new alignment_offset is on a logical block boundary */
637 if (t->alignment_offset & (t->logical_block_size - 1)) {
638 t->misaligned = 1;
639 ret = -1;
640 }
641
642 /* Discard alignment and granularity */
643 if (b->discard_granularity) {
644 alignment = queue_limit_discard_alignment(b, start);
645
646 if (t->discard_granularity != 0 &&
647 t->discard_alignment != alignment) {
648 top = t->discard_granularity + t->discard_alignment;
649 bottom = b->discard_granularity + alignment;
650
651 /* Verify that top and bottom intervals line up */
652 if ((max(top, bottom) % min(top, bottom)) != 0)
653 t->discard_misaligned = 1;
654 }
655
656 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
657 b->max_discard_sectors);
658 t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
659 b->max_hw_discard_sectors);
660 t->discard_granularity = max(t->discard_granularity,
661 b->discard_granularity);
662 t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
663 t->discard_granularity;
664 }
665
666 if (b->chunk_sectors)
667 t->chunk_sectors = min_not_zero(t->chunk_sectors,
668 b->chunk_sectors);
669
670 return ret;
671}
672EXPORT_SYMBOL(blk_stack_limits);
673
674/**
675 * bdev_stack_limits - adjust queue limits for stacked drivers
676 * @t: the stacking driver limits (top device)
677 * @bdev: the component block_device (bottom)
678 * @start: first data sector within component device
679 *
680 * Description:
681 * Merges queue limits for a top device and a block_device. Returns
682 * 0 if alignment didn't change. Returns -1 if adding the bottom
683 * device caused misalignment.
684 */
685int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
686 sector_t start)
687{
688 struct request_queue *bq = bdev_get_queue(bdev);
689
690 start += get_start_sect(bdev);
691
692 return blk_stack_limits(t, &bq->limits, start);
693}
694EXPORT_SYMBOL(bdev_stack_limits);
695
696/**
697 * disk_stack_limits - adjust queue limits for stacked drivers
698 * @disk: MD/DM gendisk (top)
699 * @bdev: the underlying block device (bottom)
700 * @offset: offset to beginning of data within component device
701 *
702 * Description:
703 * Merges the limits for a top level gendisk and a bottom level
704 * block_device.
705 */
706void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
707 sector_t offset)
708{
709 struct request_queue *t = disk->queue;
710
711 if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
712 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
713
714 disk_name(disk, 0, top);
715 bdevname(bdev, bottom);
716
717 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
718 top, bottom);
719 }
720}
721EXPORT_SYMBOL(disk_stack_limits);
722
723/**
724 * blk_queue_dma_pad - set pad mask
725 * @q: the request queue for the device
726 * @mask: pad mask
727 *
728 * Set dma pad mask.
729 *
730 * Appending pad buffer to a request modifies the last entry of a
731 * scatter list such that it includes the pad buffer.
732 **/
733void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
734{
735 q->dma_pad_mask = mask;
736}
737EXPORT_SYMBOL(blk_queue_dma_pad);
738
739/**
740 * blk_queue_update_dma_pad - update pad mask
741 * @q: the request queue for the device
742 * @mask: pad mask
743 *
744 * Update dma pad mask.
745 *
746 * Appending pad buffer to a request modifies the last entry of a
747 * scatter list such that it includes the pad buffer.
748 **/
749void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
750{
751 if (mask > q->dma_pad_mask)
752 q->dma_pad_mask = mask;
753}
754EXPORT_SYMBOL(blk_queue_update_dma_pad);
755
756/**
757 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
758 * @q: the request queue for the device
759 * @dma_drain_needed: fn which returns non-zero if drain is necessary
760 * @buf: physically contiguous buffer
761 * @size: size of the buffer in bytes
762 *
763 * Some devices have excess DMA problems and can't simply discard (or
764 * zero fill) the unwanted piece of the transfer. They have to have a
765 * real area of memory to transfer it into. The use case for this is
766 * ATAPI devices in DMA mode. If the packet command causes a transfer
767 * bigger than the transfer size some HBAs will lock up if there
768 * aren't DMA elements to contain the excess transfer. What this API
769 * does is adjust the queue so that the buf is always appended
770 * silently to the scatterlist.
771 *
772 * Note: This routine adjusts max_hw_segments to make room for appending
773 * the drain buffer. If you call blk_queue_max_segments() after calling
774 * this routine, you must set the limit to one fewer than your device
775 * can support otherwise there won't be room for the drain buffer.
776 */
777int blk_queue_dma_drain(struct request_queue *q,
778 dma_drain_needed_fn *dma_drain_needed,
779 void *buf, unsigned int size)
780{
781 if (queue_max_segments(q) < 2)
782 return -EINVAL;
783 /* make room for appending the drain */
784 blk_queue_max_segments(q, queue_max_segments(q) - 1);
785 q->dma_drain_needed = dma_drain_needed;
786 q->dma_drain_buffer = buf;
787 q->dma_drain_size = size;
788
789 return 0;
790}
791EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
792
793/**
794 * blk_queue_segment_boundary - set boundary rules for segment merging
795 * @q: the request queue for the device
796 * @mask: the memory boundary mask
797 **/
798void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
799{
800 if (mask < PAGE_SIZE - 1) {
801 mask = PAGE_SIZE - 1;
802 printk(KERN_INFO "%s: set to minimum %lx\n",
803 __func__, mask);
804 }
805
806 q->limits.seg_boundary_mask = mask;
807}
808EXPORT_SYMBOL(blk_queue_segment_boundary);
809
810/**
811 * blk_queue_virt_boundary - set boundary rules for bio merging
812 * @q: the request queue for the device
813 * @mask: the memory boundary mask
814 **/
815void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
816{
817 q->limits.virt_boundary_mask = mask;
818}
819EXPORT_SYMBOL(blk_queue_virt_boundary);
820
821/**
822 * blk_queue_dma_alignment - set dma length and memory alignment
823 * @q: the request queue for the device
824 * @mask: alignment mask
825 *
826 * description:
827 * set required memory and length alignment for direct dma transactions.
828 * this is used when building direct io requests for the queue.
829 *
830 **/
831void blk_queue_dma_alignment(struct request_queue *q, int mask)
832{
833 q->dma_alignment = mask;
834}
835EXPORT_SYMBOL(blk_queue_dma_alignment);
836
837/**
838 * blk_queue_update_dma_alignment - update dma length and memory alignment
839 * @q: the request queue for the device
840 * @mask: alignment mask
841 *
842 * description:
843 * update required memory and length alignment for direct dma transactions.
844 * If the requested alignment is larger than the current alignment, then
845 * the current queue alignment is updated to the new value, otherwise it
846 * is left alone. The design of this is to allow multiple objects
847 * (driver, device, transport etc) to set their respective
848 * alignments without having them interfere.
849 *
850 **/
851void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
852{
853 BUG_ON(mask > PAGE_SIZE);
854
855 if (mask > q->dma_alignment)
856 q->dma_alignment = mask;
857}
858EXPORT_SYMBOL(blk_queue_update_dma_alignment);
859
860void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
861{
862 if (queueable)
863 blk_queue_flag_clear(QUEUE_FLAG_FLUSH_NQ, q);
864 else
865 blk_queue_flag_set(QUEUE_FLAG_FLUSH_NQ, q);
866}
867EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
868
869/**
870 * blk_set_queue_depth - tell the block layer about the device queue depth
871 * @q: the request queue for the device
872 * @depth: queue depth
873 *
874 */
875void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
876{
877 q->queue_depth = depth;
878 wbt_set_queue_depth(q->rq_wb, depth);
879}
880EXPORT_SYMBOL(blk_set_queue_depth);
881
882/**
883 * blk_queue_write_cache - configure queue's write cache
884 * @q: the request queue for the device
885 * @wc: write back cache on or off
886 * @fua: device supports FUA writes, if true
887 *
888 * Tell the block layer about the write cache of @q.
889 */
890void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
891{
892 spin_lock_irq(q->queue_lock);
893 if (wc)
894 queue_flag_set(QUEUE_FLAG_WC, q);
895 else
896 queue_flag_clear(QUEUE_FLAG_WC, q);
897 if (fua)
898 queue_flag_set(QUEUE_FLAG_FUA, q);
899 else
900 queue_flag_clear(QUEUE_FLAG_FUA, q);
901 spin_unlock_irq(q->queue_lock);
902
903 wbt_set_write_cache(q->rq_wb, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
904}
905EXPORT_SYMBOL_GPL(blk_queue_write_cache);
906
907static int __init blk_settings_init(void)
908{
909 blk_max_low_pfn = max_low_pfn - 1;
910 blk_max_pfn = max_pfn - 1;
911 return 0;
912}
913subsys_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);