xref: /linux/block/blk-settings.c (revision 63307d015b91e626c97bb82e88054af3d0b74643)
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 
16 #include "blk.h"
17 #include "blk-wbt.h"
18 
19 unsigned long blk_max_low_pfn;
20 EXPORT_SYMBOL(blk_max_low_pfn);
21 
22 unsigned long blk_max_pfn;
23 
24 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
25 {
26 	q->rq_timeout = timeout;
27 }
28 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
29 
30 /**
31  * blk_set_default_limits - reset limits to default values
32  * @lim:  the queue_limits structure to reset
33  *
34  * Description:
35  *   Returns a queue_limit struct to its default state.
36  */
37 void blk_set_default_limits(struct queue_limits *lim)
38 {
39 	lim->max_segments = BLK_MAX_SEGMENTS;
40 	lim->max_discard_segments = 1;
41 	lim->max_integrity_segments = 0;
42 	lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
43 	lim->virt_boundary_mask = 0;
44 	lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
45 	lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
46 	lim->max_dev_sectors = 0;
47 	lim->chunk_sectors = 0;
48 	lim->max_write_same_sectors = 0;
49 	lim->max_write_zeroes_sectors = 0;
50 	lim->max_discard_sectors = 0;
51 	lim->max_hw_discard_sectors = 0;
52 	lim->discard_granularity = 0;
53 	lim->discard_alignment = 0;
54 	lim->discard_misaligned = 0;
55 	lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
56 	lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
57 	lim->alignment_offset = 0;
58 	lim->io_opt = 0;
59 	lim->misaligned = 0;
60 	lim->zoned = BLK_ZONED_NONE;
61 }
62 EXPORT_SYMBOL(blk_set_default_limits);
63 
64 /**
65  * blk_set_stacking_limits - set default limits for stacking devices
66  * @lim:  the queue_limits structure to reset
67  *
68  * Description:
69  *   Returns a queue_limit struct to its default state. Should be used
70  *   by stacking drivers like DM that have no internal limits.
71  */
72 void blk_set_stacking_limits(struct queue_limits *lim)
73 {
74 	blk_set_default_limits(lim);
75 
76 	/* Inherit limits from component devices */
77 	lim->max_segments = USHRT_MAX;
78 	lim->max_discard_segments = USHRT_MAX;
79 	lim->max_hw_sectors = UINT_MAX;
80 	lim->max_segment_size = UINT_MAX;
81 	lim->max_sectors = UINT_MAX;
82 	lim->max_dev_sectors = UINT_MAX;
83 	lim->max_write_same_sectors = UINT_MAX;
84 	lim->max_write_zeroes_sectors = UINT_MAX;
85 }
86 EXPORT_SYMBOL(blk_set_stacking_limits);
87 
88 /**
89  * blk_queue_make_request - define an alternate make_request function for a device
90  * @q:  the request queue for the device to be affected
91  * @mfn: the alternate make_request function
92  *
93  * Description:
94  *    The normal way for &struct bios to be passed to a device
95  *    driver is for them to be collected into requests on a request
96  *    queue, and then to allow the device driver to select requests
97  *    off that queue when it is ready.  This works well for many block
98  *    devices. However some block devices (typically virtual devices
99  *    such as md or lvm) do not benefit from the processing on the
100  *    request queue, and are served best by having the requests passed
101  *    directly to them.  This can be achieved by providing a function
102  *    to blk_queue_make_request().
103  *
104  * Caveat:
105  *    The driver that does this *must* be able to deal appropriately
106  *    with buffers in "highmemory". This can be accomplished by either calling
107  *    kmap_atomic() to get a temporary kernel mapping, or by calling
108  *    blk_queue_bounce() to create a buffer in normal memory.
109  **/
110 void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
111 {
112 	/*
113 	 * set defaults
114 	 */
115 	q->nr_requests = BLKDEV_MAX_RQ;
116 
117 	q->make_request_fn = mfn;
118 	blk_queue_dma_alignment(q, 511);
119 
120 	blk_set_default_limits(&q->limits);
121 }
122 EXPORT_SYMBOL(blk_queue_make_request);
123 
124 /**
125  * blk_queue_bounce_limit - set bounce buffer limit for queue
126  * @q: the request queue for the device
127  * @max_addr: the maximum address the device can handle
128  *
129  * Description:
130  *    Different hardware can have different requirements as to what pages
131  *    it can do I/O directly to. A low level driver can call
132  *    blk_queue_bounce_limit to have lower memory pages allocated as bounce
133  *    buffers for doing I/O to pages residing above @max_addr.
134  **/
135 void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
136 {
137 	unsigned long b_pfn = max_addr >> PAGE_SHIFT;
138 	int dma = 0;
139 
140 	q->bounce_gfp = GFP_NOIO;
141 #if BITS_PER_LONG == 64
142 	/*
143 	 * Assume anything <= 4GB can be handled by IOMMU.  Actually
144 	 * some IOMMUs can handle everything, but I don't know of a
145 	 * way to test this here.
146 	 */
147 	if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
148 		dma = 1;
149 	q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
150 #else
151 	if (b_pfn < blk_max_low_pfn)
152 		dma = 1;
153 	q->limits.bounce_pfn = b_pfn;
154 #endif
155 	if (dma) {
156 		init_emergency_isa_pool();
157 		q->bounce_gfp = GFP_NOIO | GFP_DMA;
158 		q->limits.bounce_pfn = b_pfn;
159 	}
160 }
161 EXPORT_SYMBOL(blk_queue_bounce_limit);
162 
163 /**
164  * blk_queue_max_hw_sectors - set max sectors for a request for this queue
165  * @q:  the request queue for the device
166  * @max_hw_sectors:  max hardware sectors in the usual 512b unit
167  *
168  * Description:
169  *    Enables a low level driver to set a hard upper limit,
170  *    max_hw_sectors, on the size of requests.  max_hw_sectors is set by
171  *    the device driver based upon the capabilities of the I/O
172  *    controller.
173  *
174  *    max_dev_sectors is a hard limit imposed by the storage device for
175  *    READ/WRITE requests. It is set by the disk driver.
176  *
177  *    max_sectors is a soft limit imposed by the block layer for
178  *    filesystem type requests.  This value can be overridden on a
179  *    per-device basis in /sys/block/<device>/queue/max_sectors_kb.
180  *    The soft limit can not exceed max_hw_sectors.
181  **/
182 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
183 {
184 	struct queue_limits *limits = &q->limits;
185 	unsigned int max_sectors;
186 
187 	if ((max_hw_sectors << 9) < PAGE_SIZE) {
188 		max_hw_sectors = 1 << (PAGE_SHIFT - 9);
189 		printk(KERN_INFO "%s: set to minimum %d\n",
190 		       __func__, max_hw_sectors);
191 	}
192 
193 	limits->max_hw_sectors = max_hw_sectors;
194 	max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
195 	max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
196 	limits->max_sectors = max_sectors;
197 	q->backing_dev_info->io_pages = max_sectors >> (PAGE_SHIFT - 9);
198 }
199 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
200 
201 /**
202  * blk_queue_chunk_sectors - set size of the chunk for this queue
203  * @q:  the request queue for the device
204  * @chunk_sectors:  chunk sectors in the usual 512b unit
205  *
206  * Description:
207  *    If a driver doesn't want IOs to cross a given chunk size, it can set
208  *    this limit and prevent merging across chunks. Note that the chunk size
209  *    must currently be a power-of-2 in sectors. Also note that the block
210  *    layer must accept a page worth of data at any offset. So if the
211  *    crossing of chunks is a hard limitation in the driver, it must still be
212  *    prepared to split single page bios.
213  **/
214 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
215 {
216 	BUG_ON(!is_power_of_2(chunk_sectors));
217 	q->limits.chunk_sectors = chunk_sectors;
218 }
219 EXPORT_SYMBOL(blk_queue_chunk_sectors);
220 
221 /**
222  * blk_queue_max_discard_sectors - set max sectors for a single discard
223  * @q:  the request queue for the device
224  * @max_discard_sectors: maximum number of sectors to discard
225  **/
226 void blk_queue_max_discard_sectors(struct request_queue *q,
227 		unsigned int max_discard_sectors)
228 {
229 	q->limits.max_hw_discard_sectors = max_discard_sectors;
230 	q->limits.max_discard_sectors = max_discard_sectors;
231 }
232 EXPORT_SYMBOL(blk_queue_max_discard_sectors);
233 
234 /**
235  * blk_queue_max_write_same_sectors - set max sectors for a single write same
236  * @q:  the request queue for the device
237  * @max_write_same_sectors: maximum number of sectors to write per command
238  **/
239 void blk_queue_max_write_same_sectors(struct request_queue *q,
240 				      unsigned int max_write_same_sectors)
241 {
242 	q->limits.max_write_same_sectors = max_write_same_sectors;
243 }
244 EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
245 
246 /**
247  * blk_queue_max_write_zeroes_sectors - set max sectors for a single
248  *                                      write zeroes
249  * @q:  the request queue for the device
250  * @max_write_zeroes_sectors: maximum number of sectors to write per command
251  **/
252 void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
253 		unsigned int max_write_zeroes_sectors)
254 {
255 	q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
256 }
257 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
258 
259 /**
260  * blk_queue_max_segments - set max hw segments for a request for this queue
261  * @q:  the request queue for the device
262  * @max_segments:  max number of segments
263  *
264  * Description:
265  *    Enables a low level driver to set an upper limit on the number of
266  *    hw data segments in a request.
267  **/
268 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
269 {
270 	if (!max_segments) {
271 		max_segments = 1;
272 		printk(KERN_INFO "%s: set to minimum %d\n",
273 		       __func__, max_segments);
274 	}
275 
276 	q->limits.max_segments = max_segments;
277 }
278 EXPORT_SYMBOL(blk_queue_max_segments);
279 
280 /**
281  * blk_queue_max_discard_segments - set max segments for discard requests
282  * @q:  the request queue for the device
283  * @max_segments:  max number of segments
284  *
285  * Description:
286  *    Enables a low level driver to set an upper limit on the number of
287  *    segments in a discard request.
288  **/
289 void blk_queue_max_discard_segments(struct request_queue *q,
290 		unsigned short max_segments)
291 {
292 	q->limits.max_discard_segments = max_segments;
293 }
294 EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
295 
296 /**
297  * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
298  * @q:  the request queue for the device
299  * @max_size:  max size of segment in bytes
300  *
301  * Description:
302  *    Enables a low level driver to set an upper limit on the size of a
303  *    coalesced segment
304  **/
305 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
306 {
307 	if (max_size < PAGE_SIZE) {
308 		max_size = PAGE_SIZE;
309 		printk(KERN_INFO "%s: set to minimum %d\n",
310 		       __func__, max_size);
311 	}
312 
313 	/* see blk_queue_virt_boundary() for the explanation */
314 	WARN_ON_ONCE(q->limits.virt_boundary_mask);
315 
316 	q->limits.max_segment_size = max_size;
317 }
318 EXPORT_SYMBOL(blk_queue_max_segment_size);
319 
320 /**
321  * blk_queue_logical_block_size - set logical block size for the queue
322  * @q:  the request queue for the device
323  * @size:  the logical block size, in bytes
324  *
325  * Description:
326  *   This should be set to the lowest possible block size that the
327  *   storage device can address.  The default of 512 covers most
328  *   hardware.
329  **/
330 void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
331 {
332 	q->limits.logical_block_size = size;
333 
334 	if (q->limits.physical_block_size < size)
335 		q->limits.physical_block_size = size;
336 
337 	if (q->limits.io_min < q->limits.physical_block_size)
338 		q->limits.io_min = q->limits.physical_block_size;
339 }
340 EXPORT_SYMBOL(blk_queue_logical_block_size);
341 
342 /**
343  * blk_queue_physical_block_size - set physical block size for the queue
344  * @q:  the request queue for the device
345  * @size:  the physical block size, in bytes
346  *
347  * Description:
348  *   This should be set to the lowest possible sector size that the
349  *   hardware can operate on without reverting to read-modify-write
350  *   operations.
351  */
352 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
353 {
354 	q->limits.physical_block_size = size;
355 
356 	if (q->limits.physical_block_size < q->limits.logical_block_size)
357 		q->limits.physical_block_size = q->limits.logical_block_size;
358 
359 	if (q->limits.io_min < q->limits.physical_block_size)
360 		q->limits.io_min = q->limits.physical_block_size;
361 }
362 EXPORT_SYMBOL(blk_queue_physical_block_size);
363 
364 /**
365  * blk_queue_alignment_offset - set physical block alignment offset
366  * @q:	the request queue for the device
367  * @offset: alignment offset in bytes
368  *
369  * Description:
370  *   Some devices are naturally misaligned to compensate for things like
371  *   the legacy DOS partition table 63-sector offset.  Low-level drivers
372  *   should call this function for devices whose first sector is not
373  *   naturally aligned.
374  */
375 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
376 {
377 	q->limits.alignment_offset =
378 		offset & (q->limits.physical_block_size - 1);
379 	q->limits.misaligned = 0;
380 }
381 EXPORT_SYMBOL(blk_queue_alignment_offset);
382 
383 /**
384  * blk_limits_io_min - set minimum request size for a device
385  * @limits: the queue limits
386  * @min:  smallest I/O size in bytes
387  *
388  * Description:
389  *   Some devices have an internal block size bigger than the reported
390  *   hardware sector size.  This function can be used to signal the
391  *   smallest I/O the device can perform without incurring a performance
392  *   penalty.
393  */
394 void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
395 {
396 	limits->io_min = min;
397 
398 	if (limits->io_min < limits->logical_block_size)
399 		limits->io_min = limits->logical_block_size;
400 
401 	if (limits->io_min < limits->physical_block_size)
402 		limits->io_min = limits->physical_block_size;
403 }
404 EXPORT_SYMBOL(blk_limits_io_min);
405 
406 /**
407  * blk_queue_io_min - set minimum request size for the queue
408  * @q:	the request queue for the device
409  * @min:  smallest I/O size in bytes
410  *
411  * Description:
412  *   Storage devices may report a granularity or preferred minimum I/O
413  *   size which is the smallest request the device can perform without
414  *   incurring a performance penalty.  For disk drives this is often the
415  *   physical block size.  For RAID arrays it is often the stripe chunk
416  *   size.  A properly aligned multiple of minimum_io_size is the
417  *   preferred request size for workloads where a high number of I/O
418  *   operations is desired.
419  */
420 void blk_queue_io_min(struct request_queue *q, unsigned int min)
421 {
422 	blk_limits_io_min(&q->limits, min);
423 }
424 EXPORT_SYMBOL(blk_queue_io_min);
425 
426 /**
427  * blk_limits_io_opt - set optimal request size for a device
428  * @limits: the queue limits
429  * @opt:  smallest I/O size in bytes
430  *
431  * Description:
432  *   Storage devices may report an optimal I/O size, which is the
433  *   device's preferred unit for sustained I/O.  This is rarely reported
434  *   for disk drives.  For RAID arrays it is usually the stripe width or
435  *   the internal track size.  A properly aligned multiple of
436  *   optimal_io_size is the preferred request size for workloads where
437  *   sustained throughput is desired.
438  */
439 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
440 {
441 	limits->io_opt = opt;
442 }
443 EXPORT_SYMBOL(blk_limits_io_opt);
444 
445 /**
446  * blk_queue_io_opt - set optimal request size for the queue
447  * @q:	the request queue for the device
448  * @opt:  optimal request size in bytes
449  *
450  * Description:
451  *   Storage devices may report an optimal I/O size, which is the
452  *   device's preferred unit for sustained I/O.  This is rarely reported
453  *   for disk drives.  For RAID arrays it is usually the stripe width or
454  *   the internal track size.  A properly aligned multiple of
455  *   optimal_io_size is the preferred request size for workloads where
456  *   sustained throughput is desired.
457  */
458 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
459 {
460 	blk_limits_io_opt(&q->limits, opt);
461 }
462 EXPORT_SYMBOL(blk_queue_io_opt);
463 
464 /**
465  * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
466  * @t:	the stacking driver (top)
467  * @b:  the underlying device (bottom)
468  **/
469 void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
470 {
471 	blk_stack_limits(&t->limits, &b->limits, 0);
472 }
473 EXPORT_SYMBOL(blk_queue_stack_limits);
474 
475 /**
476  * blk_stack_limits - adjust queue_limits for stacked devices
477  * @t:	the stacking driver limits (top device)
478  * @b:  the underlying queue limits (bottom, component device)
479  * @start:  first data sector within component device
480  *
481  * Description:
482  *    This function is used by stacking drivers like MD and DM to ensure
483  *    that all component devices have compatible block sizes and
484  *    alignments.  The stacking driver must provide a queue_limits
485  *    struct (top) and then iteratively call the stacking function for
486  *    all component (bottom) devices.  The stacking function will
487  *    attempt to combine the values and ensure proper alignment.
488  *
489  *    Returns 0 if the top and bottom queue_limits are compatible.  The
490  *    top device's block sizes and alignment offsets may be adjusted to
491  *    ensure alignment with the bottom device. If no compatible sizes
492  *    and alignments exist, -1 is returned and the resulting top
493  *    queue_limits will have the misaligned flag set to indicate that
494  *    the alignment_offset is undefined.
495  */
496 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
497 		     sector_t start)
498 {
499 	unsigned int top, bottom, alignment, ret = 0;
500 
501 	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
502 	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
503 	t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
504 	t->max_write_same_sectors = min(t->max_write_same_sectors,
505 					b->max_write_same_sectors);
506 	t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
507 					b->max_write_zeroes_sectors);
508 	t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
509 
510 	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
511 					    b->seg_boundary_mask);
512 	t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
513 					    b->virt_boundary_mask);
514 
515 	t->max_segments = min_not_zero(t->max_segments, b->max_segments);
516 	t->max_discard_segments = min_not_zero(t->max_discard_segments,
517 					       b->max_discard_segments);
518 	t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
519 						 b->max_integrity_segments);
520 
521 	t->max_segment_size = min_not_zero(t->max_segment_size,
522 					   b->max_segment_size);
523 
524 	t->misaligned |= b->misaligned;
525 
526 	alignment = queue_limit_alignment_offset(b, start);
527 
528 	/* Bottom device has different alignment.  Check that it is
529 	 * compatible with the current top alignment.
530 	 */
531 	if (t->alignment_offset != alignment) {
532 
533 		top = max(t->physical_block_size, t->io_min)
534 			+ t->alignment_offset;
535 		bottom = max(b->physical_block_size, b->io_min) + alignment;
536 
537 		/* Verify that top and bottom intervals line up */
538 		if (max(top, bottom) % min(top, bottom)) {
539 			t->misaligned = 1;
540 			ret = -1;
541 		}
542 	}
543 
544 	t->logical_block_size = max(t->logical_block_size,
545 				    b->logical_block_size);
546 
547 	t->physical_block_size = max(t->physical_block_size,
548 				     b->physical_block_size);
549 
550 	t->io_min = max(t->io_min, b->io_min);
551 	t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
552 
553 	/* Physical block size a multiple of the logical block size? */
554 	if (t->physical_block_size & (t->logical_block_size - 1)) {
555 		t->physical_block_size = t->logical_block_size;
556 		t->misaligned = 1;
557 		ret = -1;
558 	}
559 
560 	/* Minimum I/O a multiple of the physical block size? */
561 	if (t->io_min & (t->physical_block_size - 1)) {
562 		t->io_min = t->physical_block_size;
563 		t->misaligned = 1;
564 		ret = -1;
565 	}
566 
567 	/* Optimal I/O a multiple of the physical block size? */
568 	if (t->io_opt & (t->physical_block_size - 1)) {
569 		t->io_opt = 0;
570 		t->misaligned = 1;
571 		ret = -1;
572 	}
573 
574 	t->raid_partial_stripes_expensive =
575 		max(t->raid_partial_stripes_expensive,
576 		    b->raid_partial_stripes_expensive);
577 
578 	/* Find lowest common alignment_offset */
579 	t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
580 		% max(t->physical_block_size, t->io_min);
581 
582 	/* Verify that new alignment_offset is on a logical block boundary */
583 	if (t->alignment_offset & (t->logical_block_size - 1)) {
584 		t->misaligned = 1;
585 		ret = -1;
586 	}
587 
588 	/* Discard alignment and granularity */
589 	if (b->discard_granularity) {
590 		alignment = queue_limit_discard_alignment(b, start);
591 
592 		if (t->discard_granularity != 0 &&
593 		    t->discard_alignment != alignment) {
594 			top = t->discard_granularity + t->discard_alignment;
595 			bottom = b->discard_granularity + alignment;
596 
597 			/* Verify that top and bottom intervals line up */
598 			if ((max(top, bottom) % min(top, bottom)) != 0)
599 				t->discard_misaligned = 1;
600 		}
601 
602 		t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
603 						      b->max_discard_sectors);
604 		t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
605 							 b->max_hw_discard_sectors);
606 		t->discard_granularity = max(t->discard_granularity,
607 					     b->discard_granularity);
608 		t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
609 			t->discard_granularity;
610 	}
611 
612 	if (b->chunk_sectors)
613 		t->chunk_sectors = min_not_zero(t->chunk_sectors,
614 						b->chunk_sectors);
615 
616 	return ret;
617 }
618 EXPORT_SYMBOL(blk_stack_limits);
619 
620 /**
621  * bdev_stack_limits - adjust queue limits for stacked drivers
622  * @t:	the stacking driver limits (top device)
623  * @bdev:  the component block_device (bottom)
624  * @start:  first data sector within component device
625  *
626  * Description:
627  *    Merges queue limits for a top device and a block_device.  Returns
628  *    0 if alignment didn't change.  Returns -1 if adding the bottom
629  *    device caused misalignment.
630  */
631 int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
632 		      sector_t start)
633 {
634 	struct request_queue *bq = bdev_get_queue(bdev);
635 
636 	start += get_start_sect(bdev);
637 
638 	return blk_stack_limits(t, &bq->limits, start);
639 }
640 EXPORT_SYMBOL(bdev_stack_limits);
641 
642 /**
643  * disk_stack_limits - adjust queue limits for stacked drivers
644  * @disk:  MD/DM gendisk (top)
645  * @bdev:  the underlying block device (bottom)
646  * @offset:  offset to beginning of data within component device
647  *
648  * Description:
649  *    Merges the limits for a top level gendisk and a bottom level
650  *    block_device.
651  */
652 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
653 		       sector_t offset)
654 {
655 	struct request_queue *t = disk->queue;
656 
657 	if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
658 		char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
659 
660 		disk_name(disk, 0, top);
661 		bdevname(bdev, bottom);
662 
663 		printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
664 		       top, bottom);
665 	}
666 }
667 EXPORT_SYMBOL(disk_stack_limits);
668 
669 /**
670  * blk_queue_update_dma_pad - update pad mask
671  * @q:     the request queue for the device
672  * @mask:  pad mask
673  *
674  * Update dma pad mask.
675  *
676  * Appending pad buffer to a request modifies the last entry of a
677  * scatter list such that it includes the pad buffer.
678  **/
679 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
680 {
681 	if (mask > q->dma_pad_mask)
682 		q->dma_pad_mask = mask;
683 }
684 EXPORT_SYMBOL(blk_queue_update_dma_pad);
685 
686 /**
687  * blk_queue_dma_drain - Set up a drain buffer for excess dma.
688  * @q:  the request queue for the device
689  * @dma_drain_needed: fn which returns non-zero if drain is necessary
690  * @buf:	physically contiguous buffer
691  * @size:	size of the buffer in bytes
692  *
693  * Some devices have excess DMA problems and can't simply discard (or
694  * zero fill) the unwanted piece of the transfer.  They have to have a
695  * real area of memory to transfer it into.  The use case for this is
696  * ATAPI devices in DMA mode.  If the packet command causes a transfer
697  * bigger than the transfer size some HBAs will lock up if there
698  * aren't DMA elements to contain the excess transfer.  What this API
699  * does is adjust the queue so that the buf is always appended
700  * silently to the scatterlist.
701  *
702  * Note: This routine adjusts max_hw_segments to make room for appending
703  * the drain buffer.  If you call blk_queue_max_segments() after calling
704  * this routine, you must set the limit to one fewer than your device
705  * can support otherwise there won't be room for the drain buffer.
706  */
707 int blk_queue_dma_drain(struct request_queue *q,
708 			       dma_drain_needed_fn *dma_drain_needed,
709 			       void *buf, unsigned int size)
710 {
711 	if (queue_max_segments(q) < 2)
712 		return -EINVAL;
713 	/* make room for appending the drain */
714 	blk_queue_max_segments(q, queue_max_segments(q) - 1);
715 	q->dma_drain_needed = dma_drain_needed;
716 	q->dma_drain_buffer = buf;
717 	q->dma_drain_size = size;
718 
719 	return 0;
720 }
721 EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
722 
723 /**
724  * blk_queue_segment_boundary - set boundary rules for segment merging
725  * @q:  the request queue for the device
726  * @mask:  the memory boundary mask
727  **/
728 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
729 {
730 	if (mask < PAGE_SIZE - 1) {
731 		mask = PAGE_SIZE - 1;
732 		printk(KERN_INFO "%s: set to minimum %lx\n",
733 		       __func__, mask);
734 	}
735 
736 	q->limits.seg_boundary_mask = mask;
737 }
738 EXPORT_SYMBOL(blk_queue_segment_boundary);
739 
740 /**
741  * blk_queue_virt_boundary - set boundary rules for bio merging
742  * @q:  the request queue for the device
743  * @mask:  the memory boundary mask
744  **/
745 void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
746 {
747 	q->limits.virt_boundary_mask = mask;
748 
749 	/*
750 	 * Devices that require a virtual boundary do not support scatter/gather
751 	 * I/O natively, but instead require a descriptor list entry for each
752 	 * page (which might not be idential to the Linux PAGE_SIZE).  Because
753 	 * of that they are not limited by our notion of "segment size".
754 	 */
755 	q->limits.max_segment_size = UINT_MAX;
756 }
757 EXPORT_SYMBOL(blk_queue_virt_boundary);
758 
759 /**
760  * blk_queue_dma_alignment - set dma length and memory alignment
761  * @q:     the request queue for the device
762  * @mask:  alignment mask
763  *
764  * description:
765  *    set required memory and length alignment for direct dma transactions.
766  *    this is used when building direct io requests for the queue.
767  *
768  **/
769 void blk_queue_dma_alignment(struct request_queue *q, int mask)
770 {
771 	q->dma_alignment = mask;
772 }
773 EXPORT_SYMBOL(blk_queue_dma_alignment);
774 
775 /**
776  * blk_queue_update_dma_alignment - update dma length and memory alignment
777  * @q:     the request queue for the device
778  * @mask:  alignment mask
779  *
780  * description:
781  *    update required memory and length alignment for direct dma transactions.
782  *    If the requested alignment is larger than the current alignment, then
783  *    the current queue alignment is updated to the new value, otherwise it
784  *    is left alone.  The design of this is to allow multiple objects
785  *    (driver, device, transport etc) to set their respective
786  *    alignments without having them interfere.
787  *
788  **/
789 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
790 {
791 	BUG_ON(mask > PAGE_SIZE);
792 
793 	if (mask > q->dma_alignment)
794 		q->dma_alignment = mask;
795 }
796 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
797 
798 /**
799  * blk_set_queue_depth - tell the block layer about the device queue depth
800  * @q:		the request queue for the device
801  * @depth:		queue depth
802  *
803  */
804 void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
805 {
806 	q->queue_depth = depth;
807 	wbt_set_queue_depth(q, depth);
808 }
809 EXPORT_SYMBOL(blk_set_queue_depth);
810 
811 /**
812  * blk_queue_write_cache - configure queue's write cache
813  * @q:		the request queue for the device
814  * @wc:		write back cache on or off
815  * @fua:	device supports FUA writes, if true
816  *
817  * Tell the block layer about the write cache of @q.
818  */
819 void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
820 {
821 	if (wc)
822 		blk_queue_flag_set(QUEUE_FLAG_WC, q);
823 	else
824 		blk_queue_flag_clear(QUEUE_FLAG_WC, q);
825 	if (fua)
826 		blk_queue_flag_set(QUEUE_FLAG_FUA, q);
827 	else
828 		blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
829 
830 	wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
831 }
832 EXPORT_SYMBOL_GPL(blk_queue_write_cache);
833 
834 static int __init blk_settings_init(void)
835 {
836 	blk_max_low_pfn = max_low_pfn - 1;
837 	blk_max_pfn = max_pfn - 1;
838 	return 0;
839 }
840 subsys_initcall(blk_settings_init);
841