xref: /openbmc/linux/block/blk-settings.c (revision a01822e9)
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 
20 unsigned long blk_max_low_pfn;
21 EXPORT_SYMBOL(blk_max_low_pfn);
22 
23 unsigned long blk_max_pfn;
24 
25 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
26 {
27 	q->rq_timeout = timeout;
28 }
29 EXPORT_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  */
38 void 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 }
63 EXPORT_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  */
73 void 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 }
87 EXPORT_SYMBOL(blk_set_stacking_limits);
88 
89 /**
90  * blk_queue_bounce_limit - set bounce buffer limit for queue
91  * @q: the request queue for the device
92  * @max_addr: the maximum address the device can handle
93  *
94  * Description:
95  *    Different hardware can have different requirements as to what pages
96  *    it can do I/O directly to. A low level driver can call
97  *    blk_queue_bounce_limit to have lower memory pages allocated as bounce
98  *    buffers for doing I/O to pages residing above @max_addr.
99  **/
100 void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
101 {
102 	unsigned long b_pfn = max_addr >> PAGE_SHIFT;
103 	int dma = 0;
104 
105 	q->bounce_gfp = GFP_NOIO;
106 #if BITS_PER_LONG == 64
107 	/*
108 	 * Assume anything <= 4GB can be handled by IOMMU.  Actually
109 	 * some IOMMUs can handle everything, but I don't know of a
110 	 * way to test this here.
111 	 */
112 	if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
113 		dma = 1;
114 	q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
115 #else
116 	if (b_pfn < blk_max_low_pfn)
117 		dma = 1;
118 	q->limits.bounce_pfn = b_pfn;
119 #endif
120 	if (dma) {
121 		init_emergency_isa_pool();
122 		q->bounce_gfp = GFP_NOIO | GFP_DMA;
123 		q->limits.bounce_pfn = b_pfn;
124 	}
125 }
126 EXPORT_SYMBOL(blk_queue_bounce_limit);
127 
128 /**
129  * blk_queue_max_hw_sectors - set max sectors for a request for this queue
130  * @q:  the request queue for the device
131  * @max_hw_sectors:  max hardware sectors in the usual 512b unit
132  *
133  * Description:
134  *    Enables a low level driver to set a hard upper limit,
135  *    max_hw_sectors, on the size of requests.  max_hw_sectors is set by
136  *    the device driver based upon the capabilities of the I/O
137  *    controller.
138  *
139  *    max_dev_sectors is a hard limit imposed by the storage device for
140  *    READ/WRITE requests. It is set by the disk driver.
141  *
142  *    max_sectors is a soft limit imposed by the block layer for
143  *    filesystem type requests.  This value can be overridden on a
144  *    per-device basis in /sys/block/<device>/queue/max_sectors_kb.
145  *    The soft limit can not exceed max_hw_sectors.
146  **/
147 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
148 {
149 	struct queue_limits *limits = &q->limits;
150 	unsigned int max_sectors;
151 
152 	if ((max_hw_sectors << 9) < PAGE_SIZE) {
153 		max_hw_sectors = 1 << (PAGE_SHIFT - 9);
154 		printk(KERN_INFO "%s: set to minimum %d\n",
155 		       __func__, max_hw_sectors);
156 	}
157 
158 	limits->max_hw_sectors = max_hw_sectors;
159 	max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
160 	max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
161 	limits->max_sectors = max_sectors;
162 	q->backing_dev_info->io_pages = max_sectors >> (PAGE_SHIFT - 9);
163 }
164 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
165 
166 /**
167  * blk_queue_chunk_sectors - set size of the chunk for this queue
168  * @q:  the request queue for the device
169  * @chunk_sectors:  chunk sectors in the usual 512b unit
170  *
171  * Description:
172  *    If a driver doesn't want IOs to cross a given chunk size, it can set
173  *    this limit and prevent merging across chunks. Note that the chunk size
174  *    must currently be a power-of-2 in sectors. Also note that the block
175  *    layer must accept a page worth of data at any offset. So if the
176  *    crossing of chunks is a hard limitation in the driver, it must still be
177  *    prepared to split single page bios.
178  **/
179 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
180 {
181 	BUG_ON(!is_power_of_2(chunk_sectors));
182 	q->limits.chunk_sectors = chunk_sectors;
183 }
184 EXPORT_SYMBOL(blk_queue_chunk_sectors);
185 
186 /**
187  * blk_queue_max_discard_sectors - set max sectors for a single discard
188  * @q:  the request queue for the device
189  * @max_discard_sectors: maximum number of sectors to discard
190  **/
191 void blk_queue_max_discard_sectors(struct request_queue *q,
192 		unsigned int max_discard_sectors)
193 {
194 	q->limits.max_hw_discard_sectors = max_discard_sectors;
195 	q->limits.max_discard_sectors = max_discard_sectors;
196 }
197 EXPORT_SYMBOL(blk_queue_max_discard_sectors);
198 
199 /**
200  * blk_queue_max_write_same_sectors - set max sectors for a single write same
201  * @q:  the request queue for the device
202  * @max_write_same_sectors: maximum number of sectors to write per command
203  **/
204 void blk_queue_max_write_same_sectors(struct request_queue *q,
205 				      unsigned int max_write_same_sectors)
206 {
207 	q->limits.max_write_same_sectors = max_write_same_sectors;
208 }
209 EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
210 
211 /**
212  * blk_queue_max_write_zeroes_sectors - set max sectors for a single
213  *                                      write zeroes
214  * @q:  the request queue for the device
215  * @max_write_zeroes_sectors: maximum number of sectors to write per command
216  **/
217 void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
218 		unsigned int max_write_zeroes_sectors)
219 {
220 	q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
221 }
222 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
223 
224 /**
225  * blk_queue_max_segments - set max hw segments for a request for this queue
226  * @q:  the request queue for the device
227  * @max_segments:  max number of segments
228  *
229  * Description:
230  *    Enables a low level driver to set an upper limit on the number of
231  *    hw data segments in a request.
232  **/
233 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
234 {
235 	if (!max_segments) {
236 		max_segments = 1;
237 		printk(KERN_INFO "%s: set to minimum %d\n",
238 		       __func__, max_segments);
239 	}
240 
241 	q->limits.max_segments = max_segments;
242 }
243 EXPORT_SYMBOL(blk_queue_max_segments);
244 
245 /**
246  * blk_queue_max_discard_segments - set max segments for discard requests
247  * @q:  the request queue for the device
248  * @max_segments:  max number of segments
249  *
250  * Description:
251  *    Enables a low level driver to set an upper limit on the number of
252  *    segments in a discard request.
253  **/
254 void blk_queue_max_discard_segments(struct request_queue *q,
255 		unsigned short max_segments)
256 {
257 	q->limits.max_discard_segments = max_segments;
258 }
259 EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
260 
261 /**
262  * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
263  * @q:  the request queue for the device
264  * @max_size:  max size of segment in bytes
265  *
266  * Description:
267  *    Enables a low level driver to set an upper limit on the size of a
268  *    coalesced segment
269  **/
270 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
271 {
272 	if (max_size < PAGE_SIZE) {
273 		max_size = PAGE_SIZE;
274 		printk(KERN_INFO "%s: set to minimum %d\n",
275 		       __func__, max_size);
276 	}
277 
278 	/* see blk_queue_virt_boundary() for the explanation */
279 	WARN_ON_ONCE(q->limits.virt_boundary_mask);
280 
281 	q->limits.max_segment_size = max_size;
282 }
283 EXPORT_SYMBOL(blk_queue_max_segment_size);
284 
285 /**
286  * blk_queue_logical_block_size - set logical block size for the queue
287  * @q:  the request queue for the device
288  * @size:  the logical block size, in bytes
289  *
290  * Description:
291  *   This should be set to the lowest possible block size that the
292  *   storage device can address.  The default of 512 covers most
293  *   hardware.
294  **/
295 void blk_queue_logical_block_size(struct request_queue *q, unsigned int size)
296 {
297 	q->limits.logical_block_size = size;
298 
299 	if (q->limits.physical_block_size < size)
300 		q->limits.physical_block_size = size;
301 
302 	if (q->limits.io_min < q->limits.physical_block_size)
303 		q->limits.io_min = q->limits.physical_block_size;
304 }
305 EXPORT_SYMBOL(blk_queue_logical_block_size);
306 
307 /**
308  * blk_queue_physical_block_size - set physical block size for the queue
309  * @q:  the request queue for the device
310  * @size:  the physical block size, in bytes
311  *
312  * Description:
313  *   This should be set to the lowest possible sector size that the
314  *   hardware can operate on without reverting to read-modify-write
315  *   operations.
316  */
317 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
318 {
319 	q->limits.physical_block_size = size;
320 
321 	if (q->limits.physical_block_size < q->limits.logical_block_size)
322 		q->limits.physical_block_size = q->limits.logical_block_size;
323 
324 	if (q->limits.io_min < q->limits.physical_block_size)
325 		q->limits.io_min = q->limits.physical_block_size;
326 }
327 EXPORT_SYMBOL(blk_queue_physical_block_size);
328 
329 /**
330  * blk_queue_alignment_offset - set physical block alignment offset
331  * @q:	the request queue for the device
332  * @offset: alignment offset in bytes
333  *
334  * Description:
335  *   Some devices are naturally misaligned to compensate for things like
336  *   the legacy DOS partition table 63-sector offset.  Low-level drivers
337  *   should call this function for devices whose first sector is not
338  *   naturally aligned.
339  */
340 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
341 {
342 	q->limits.alignment_offset =
343 		offset & (q->limits.physical_block_size - 1);
344 	q->limits.misaligned = 0;
345 }
346 EXPORT_SYMBOL(blk_queue_alignment_offset);
347 
348 /**
349  * blk_limits_io_min - set minimum request size for a device
350  * @limits: the queue limits
351  * @min:  smallest I/O size in bytes
352  *
353  * Description:
354  *   Some devices have an internal block size bigger than the reported
355  *   hardware sector size.  This function can be used to signal the
356  *   smallest I/O the device can perform without incurring a performance
357  *   penalty.
358  */
359 void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
360 {
361 	limits->io_min = min;
362 
363 	if (limits->io_min < limits->logical_block_size)
364 		limits->io_min = limits->logical_block_size;
365 
366 	if (limits->io_min < limits->physical_block_size)
367 		limits->io_min = limits->physical_block_size;
368 }
369 EXPORT_SYMBOL(blk_limits_io_min);
370 
371 /**
372  * blk_queue_io_min - set minimum request size for the queue
373  * @q:	the request queue for the device
374  * @min:  smallest I/O size in bytes
375  *
376  * Description:
377  *   Storage devices may report a granularity or preferred minimum I/O
378  *   size which is the smallest request the device can perform without
379  *   incurring a performance penalty.  For disk drives this is often the
380  *   physical block size.  For RAID arrays it is often the stripe chunk
381  *   size.  A properly aligned multiple of minimum_io_size is the
382  *   preferred request size for workloads where a high number of I/O
383  *   operations is desired.
384  */
385 void blk_queue_io_min(struct request_queue *q, unsigned int min)
386 {
387 	blk_limits_io_min(&q->limits, min);
388 }
389 EXPORT_SYMBOL(blk_queue_io_min);
390 
391 /**
392  * blk_limits_io_opt - set optimal request size for a device
393  * @limits: the queue limits
394  * @opt:  smallest I/O size in bytes
395  *
396  * Description:
397  *   Storage devices may report an optimal I/O size, which is the
398  *   device's preferred unit for sustained I/O.  This is rarely reported
399  *   for disk drives.  For RAID arrays it is usually the stripe width or
400  *   the internal track size.  A properly aligned multiple of
401  *   optimal_io_size is the preferred request size for workloads where
402  *   sustained throughput is desired.
403  */
404 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
405 {
406 	limits->io_opt = opt;
407 }
408 EXPORT_SYMBOL(blk_limits_io_opt);
409 
410 /**
411  * blk_queue_io_opt - set optimal request size for the queue
412  * @q:	the request queue for the device
413  * @opt:  optimal request size in bytes
414  *
415  * Description:
416  *   Storage devices may report an optimal I/O size, which is the
417  *   device's preferred unit for sustained I/O.  This is rarely reported
418  *   for disk drives.  For RAID arrays it is usually the stripe width or
419  *   the internal track size.  A properly aligned multiple of
420  *   optimal_io_size is the preferred request size for workloads where
421  *   sustained throughput is desired.
422  */
423 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
424 {
425 	blk_limits_io_opt(&q->limits, opt);
426 }
427 EXPORT_SYMBOL(blk_queue_io_opt);
428 
429 /**
430  * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
431  * @t:	the stacking driver (top)
432  * @b:  the underlying device (bottom)
433  **/
434 void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
435 {
436 	blk_stack_limits(&t->limits, &b->limits, 0);
437 }
438 EXPORT_SYMBOL(blk_queue_stack_limits);
439 
440 /**
441  * blk_stack_limits - adjust queue_limits for stacked devices
442  * @t:	the stacking driver limits (top device)
443  * @b:  the underlying queue limits (bottom, component device)
444  * @start:  first data sector within component device
445  *
446  * Description:
447  *    This function is used by stacking drivers like MD and DM to ensure
448  *    that all component devices have compatible block sizes and
449  *    alignments.  The stacking driver must provide a queue_limits
450  *    struct (top) and then iteratively call the stacking function for
451  *    all component (bottom) devices.  The stacking function will
452  *    attempt to combine the values and ensure proper alignment.
453  *
454  *    Returns 0 if the top and bottom queue_limits are compatible.  The
455  *    top device's block sizes and alignment offsets may be adjusted to
456  *    ensure alignment with the bottom device. If no compatible sizes
457  *    and alignments exist, -1 is returned and the resulting top
458  *    queue_limits will have the misaligned flag set to indicate that
459  *    the alignment_offset is undefined.
460  */
461 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
462 		     sector_t start)
463 {
464 	unsigned int top, bottom, alignment, ret = 0;
465 
466 	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
467 	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
468 	t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
469 	t->max_write_same_sectors = min(t->max_write_same_sectors,
470 					b->max_write_same_sectors);
471 	t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
472 					b->max_write_zeroes_sectors);
473 	t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
474 
475 	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
476 					    b->seg_boundary_mask);
477 	t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
478 					    b->virt_boundary_mask);
479 
480 	t->max_segments = min_not_zero(t->max_segments, b->max_segments);
481 	t->max_discard_segments = min_not_zero(t->max_discard_segments,
482 					       b->max_discard_segments);
483 	t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
484 						 b->max_integrity_segments);
485 
486 	t->max_segment_size = min_not_zero(t->max_segment_size,
487 					   b->max_segment_size);
488 
489 	t->misaligned |= b->misaligned;
490 
491 	alignment = queue_limit_alignment_offset(b, start);
492 
493 	/* Bottom device has different alignment.  Check that it is
494 	 * compatible with the current top alignment.
495 	 */
496 	if (t->alignment_offset != alignment) {
497 
498 		top = max(t->physical_block_size, t->io_min)
499 			+ t->alignment_offset;
500 		bottom = max(b->physical_block_size, b->io_min) + alignment;
501 
502 		/* Verify that top and bottom intervals line up */
503 		if (max(top, bottom) % min(top, bottom)) {
504 			t->misaligned = 1;
505 			ret = -1;
506 		}
507 	}
508 
509 	t->logical_block_size = max(t->logical_block_size,
510 				    b->logical_block_size);
511 
512 	t->physical_block_size = max(t->physical_block_size,
513 				     b->physical_block_size);
514 
515 	t->io_min = max(t->io_min, b->io_min);
516 	t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
517 
518 	/* Physical block size a multiple of the logical block size? */
519 	if (t->physical_block_size & (t->logical_block_size - 1)) {
520 		t->physical_block_size = t->logical_block_size;
521 		t->misaligned = 1;
522 		ret = -1;
523 	}
524 
525 	/* Minimum I/O a multiple of the physical block size? */
526 	if (t->io_min & (t->physical_block_size - 1)) {
527 		t->io_min = t->physical_block_size;
528 		t->misaligned = 1;
529 		ret = -1;
530 	}
531 
532 	/* Optimal I/O a multiple of the physical block size? */
533 	if (t->io_opt & (t->physical_block_size - 1)) {
534 		t->io_opt = 0;
535 		t->misaligned = 1;
536 		ret = -1;
537 	}
538 
539 	t->raid_partial_stripes_expensive =
540 		max(t->raid_partial_stripes_expensive,
541 		    b->raid_partial_stripes_expensive);
542 
543 	/* Find lowest common alignment_offset */
544 	t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
545 		% max(t->physical_block_size, t->io_min);
546 
547 	/* Verify that new alignment_offset is on a logical block boundary */
548 	if (t->alignment_offset & (t->logical_block_size - 1)) {
549 		t->misaligned = 1;
550 		ret = -1;
551 	}
552 
553 	/* Discard alignment and granularity */
554 	if (b->discard_granularity) {
555 		alignment = queue_limit_discard_alignment(b, start);
556 
557 		if (t->discard_granularity != 0 &&
558 		    t->discard_alignment != alignment) {
559 			top = t->discard_granularity + t->discard_alignment;
560 			bottom = b->discard_granularity + alignment;
561 
562 			/* Verify that top and bottom intervals line up */
563 			if ((max(top, bottom) % min(top, bottom)) != 0)
564 				t->discard_misaligned = 1;
565 		}
566 
567 		t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
568 						      b->max_discard_sectors);
569 		t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
570 							 b->max_hw_discard_sectors);
571 		t->discard_granularity = max(t->discard_granularity,
572 					     b->discard_granularity);
573 		t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
574 			t->discard_granularity;
575 	}
576 
577 	if (b->chunk_sectors)
578 		t->chunk_sectors = min_not_zero(t->chunk_sectors,
579 						b->chunk_sectors);
580 
581 	return ret;
582 }
583 EXPORT_SYMBOL(blk_stack_limits);
584 
585 /**
586  * bdev_stack_limits - adjust queue limits for stacked drivers
587  * @t:	the stacking driver limits (top device)
588  * @bdev:  the component block_device (bottom)
589  * @start:  first data sector within component device
590  *
591  * Description:
592  *    Merges queue limits for a top device and a block_device.  Returns
593  *    0 if alignment didn't change.  Returns -1 if adding the bottom
594  *    device caused misalignment.
595  */
596 int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
597 		      sector_t start)
598 {
599 	struct request_queue *bq = bdev_get_queue(bdev);
600 
601 	start += get_start_sect(bdev);
602 
603 	return blk_stack_limits(t, &bq->limits, start);
604 }
605 EXPORT_SYMBOL(bdev_stack_limits);
606 
607 /**
608  * disk_stack_limits - adjust queue limits for stacked drivers
609  * @disk:  MD/DM gendisk (top)
610  * @bdev:  the underlying block device (bottom)
611  * @offset:  offset to beginning of data within component device
612  *
613  * Description:
614  *    Merges the limits for a top level gendisk and a bottom level
615  *    block_device.
616  */
617 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
618 		       sector_t offset)
619 {
620 	struct request_queue *t = disk->queue;
621 
622 	if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
623 		char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
624 
625 		disk_name(disk, 0, top);
626 		bdevname(bdev, bottom);
627 
628 		printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
629 		       top, bottom);
630 	}
631 
632 	t->backing_dev_info->io_pages =
633 		t->limits.max_sectors >> (PAGE_SHIFT - 9);
634 }
635 EXPORT_SYMBOL(disk_stack_limits);
636 
637 /**
638  * blk_queue_update_dma_pad - update pad mask
639  * @q:     the request queue for the device
640  * @mask:  pad mask
641  *
642  * Update dma pad mask.
643  *
644  * Appending pad buffer to a request modifies the last entry of a
645  * scatter list such that it includes the pad buffer.
646  **/
647 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
648 {
649 	if (mask > q->dma_pad_mask)
650 		q->dma_pad_mask = mask;
651 }
652 EXPORT_SYMBOL(blk_queue_update_dma_pad);
653 
654 /**
655  * blk_queue_dma_drain - Set up a drain buffer for excess dma.
656  * @q:  the request queue for the device
657  * @dma_drain_needed: fn which returns non-zero if drain is necessary
658  * @buf:	physically contiguous buffer
659  * @size:	size of the buffer in bytes
660  *
661  * Some devices have excess DMA problems and can't simply discard (or
662  * zero fill) the unwanted piece of the transfer.  They have to have a
663  * real area of memory to transfer it into.  The use case for this is
664  * ATAPI devices in DMA mode.  If the packet command causes a transfer
665  * bigger than the transfer size some HBAs will lock up if there
666  * aren't DMA elements to contain the excess transfer.  What this API
667  * does is adjust the queue so that the buf is always appended
668  * silently to the scatterlist.
669  *
670  * Note: This routine adjusts max_hw_segments to make room for appending
671  * the drain buffer.  If you call blk_queue_max_segments() after calling
672  * this routine, you must set the limit to one fewer than your device
673  * can support otherwise there won't be room for the drain buffer.
674  */
675 int blk_queue_dma_drain(struct request_queue *q,
676 			       dma_drain_needed_fn *dma_drain_needed,
677 			       void *buf, unsigned int size)
678 {
679 	if (queue_max_segments(q) < 2)
680 		return -EINVAL;
681 	/* make room for appending the drain */
682 	blk_queue_max_segments(q, queue_max_segments(q) - 1);
683 	q->dma_drain_needed = dma_drain_needed;
684 	q->dma_drain_buffer = buf;
685 	q->dma_drain_size = size;
686 
687 	return 0;
688 }
689 EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
690 
691 /**
692  * blk_queue_segment_boundary - set boundary rules for segment merging
693  * @q:  the request queue for the device
694  * @mask:  the memory boundary mask
695  **/
696 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
697 {
698 	if (mask < PAGE_SIZE - 1) {
699 		mask = PAGE_SIZE - 1;
700 		printk(KERN_INFO "%s: set to minimum %lx\n",
701 		       __func__, mask);
702 	}
703 
704 	q->limits.seg_boundary_mask = mask;
705 }
706 EXPORT_SYMBOL(blk_queue_segment_boundary);
707 
708 /**
709  * blk_queue_virt_boundary - set boundary rules for bio merging
710  * @q:  the request queue for the device
711  * @mask:  the memory boundary mask
712  **/
713 void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
714 {
715 	q->limits.virt_boundary_mask = mask;
716 
717 	/*
718 	 * Devices that require a virtual boundary do not support scatter/gather
719 	 * I/O natively, but instead require a descriptor list entry for each
720 	 * page (which might not be idential to the Linux PAGE_SIZE).  Because
721 	 * of that they are not limited by our notion of "segment size".
722 	 */
723 	if (mask)
724 		q->limits.max_segment_size = UINT_MAX;
725 }
726 EXPORT_SYMBOL(blk_queue_virt_boundary);
727 
728 /**
729  * blk_queue_dma_alignment - set dma length and memory alignment
730  * @q:     the request queue for the device
731  * @mask:  alignment mask
732  *
733  * description:
734  *    set required memory and length alignment for direct dma transactions.
735  *    this is used when building direct io requests for the queue.
736  *
737  **/
738 void blk_queue_dma_alignment(struct request_queue *q, int mask)
739 {
740 	q->dma_alignment = mask;
741 }
742 EXPORT_SYMBOL(blk_queue_dma_alignment);
743 
744 /**
745  * blk_queue_update_dma_alignment - update dma length and memory alignment
746  * @q:     the request queue for the device
747  * @mask:  alignment mask
748  *
749  * description:
750  *    update required memory and length alignment for direct dma transactions.
751  *    If the requested alignment is larger than the current alignment, then
752  *    the current queue alignment is updated to the new value, otherwise it
753  *    is left alone.  The design of this is to allow multiple objects
754  *    (driver, device, transport etc) to set their respective
755  *    alignments without having them interfere.
756  *
757  **/
758 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
759 {
760 	BUG_ON(mask > PAGE_SIZE);
761 
762 	if (mask > q->dma_alignment)
763 		q->dma_alignment = mask;
764 }
765 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
766 
767 /**
768  * blk_set_queue_depth - tell the block layer about the device queue depth
769  * @q:		the request queue for the device
770  * @depth:		queue depth
771  *
772  */
773 void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
774 {
775 	q->queue_depth = depth;
776 	rq_qos_queue_depth_changed(q);
777 }
778 EXPORT_SYMBOL(blk_set_queue_depth);
779 
780 /**
781  * blk_queue_write_cache - configure queue's write cache
782  * @q:		the request queue for the device
783  * @wc:		write back cache on or off
784  * @fua:	device supports FUA writes, if true
785  *
786  * Tell the block layer about the write cache of @q.
787  */
788 void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
789 {
790 	if (wc)
791 		blk_queue_flag_set(QUEUE_FLAG_WC, q);
792 	else
793 		blk_queue_flag_clear(QUEUE_FLAG_WC, q);
794 	if (fua)
795 		blk_queue_flag_set(QUEUE_FLAG_FUA, q);
796 	else
797 		blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
798 
799 	wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
800 }
801 EXPORT_SYMBOL_GPL(blk_queue_write_cache);
802 
803 /**
804  * blk_queue_required_elevator_features - Set a queue required elevator features
805  * @q:		the request queue for the target device
806  * @features:	Required elevator features OR'ed together
807  *
808  * Tell the block layer that for the device controlled through @q, only the
809  * only elevators that can be used are those that implement at least the set of
810  * features specified by @features.
811  */
812 void blk_queue_required_elevator_features(struct request_queue *q,
813 					  unsigned int features)
814 {
815 	q->required_elevator_features = features;
816 }
817 EXPORT_SYMBOL_GPL(blk_queue_required_elevator_features);
818 
819 /**
820  * blk_queue_can_use_dma_map_merging - configure queue for merging segments.
821  * @q:		the request queue for the device
822  * @dev:	the device pointer for dma
823  *
824  * Tell the block layer about merging the segments by dma map of @q.
825  */
826 bool blk_queue_can_use_dma_map_merging(struct request_queue *q,
827 				       struct device *dev)
828 {
829 	unsigned long boundary = dma_get_merge_boundary(dev);
830 
831 	if (!boundary)
832 		return false;
833 
834 	/* No need to update max_segment_size. see blk_queue_virt_boundary() */
835 	blk_queue_virt_boundary(q, boundary);
836 
837 	return true;
838 }
839 EXPORT_SYMBOL_GPL(blk_queue_can_use_dma_map_merging);
840 
841 static int __init blk_settings_init(void)
842 {
843 	blk_max_low_pfn = max_low_pfn - 1;
844 	blk_max_pfn = max_pfn - 1;
845 	return 0;
846 }
847 subsys_initcall(blk_settings_init);
848