xref: /linux/block/blk-settings.c (revision 4e0ae876f77bc01a7e77724dea57b4b82bd53244)
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/memblock.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 
18 unsigned long blk_max_low_pfn;
19 EXPORT_SYMBOL(blk_max_low_pfn);
20 
21 unsigned long blk_max_pfn;
22 
23 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
24 {
25 	q->rq_timeout = timeout;
26 }
27 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
28 
29 /**
30  * blk_set_default_limits - reset limits to default values
31  * @lim:  the queue_limits structure to reset
32  *
33  * Description:
34  *   Returns a queue_limit struct to its default state.
35  */
36 void blk_set_default_limits(struct queue_limits *lim)
37 {
38 	lim->max_segments = BLK_MAX_SEGMENTS;
39 	lim->max_discard_segments = 1;
40 	lim->max_integrity_segments = 0;
41 	lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
42 	lim->virt_boundary_mask = 0;
43 	lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
44 	lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
45 	lim->max_dev_sectors = 0;
46 	lim->chunk_sectors = 0;
47 	lim->max_write_same_sectors = 0;
48 	lim->max_write_zeroes_sectors = 0;
49 	lim->max_discard_sectors = 0;
50 	lim->max_hw_discard_sectors = 0;
51 	lim->discard_granularity = 0;
52 	lim->discard_alignment = 0;
53 	lim->discard_misaligned = 0;
54 	lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
55 	lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
56 	lim->alignment_offset = 0;
57 	lim->io_opt = 0;
58 	lim->misaligned = 0;
59 	lim->zoned = BLK_ZONED_NONE;
60 }
61 EXPORT_SYMBOL(blk_set_default_limits);
62 
63 /**
64  * blk_set_stacking_limits - set default limits for stacking devices
65  * @lim:  the queue_limits structure to reset
66  *
67  * Description:
68  *   Returns a queue_limit struct to its default state. Should be used
69  *   by stacking drivers like DM that have no internal limits.
70  */
71 void blk_set_stacking_limits(struct queue_limits *lim)
72 {
73 	blk_set_default_limits(lim);
74 
75 	/* Inherit limits from component devices */
76 	lim->max_segments = USHRT_MAX;
77 	lim->max_discard_segments = USHRT_MAX;
78 	lim->max_hw_sectors = UINT_MAX;
79 	lim->max_segment_size = UINT_MAX;
80 	lim->max_sectors = UINT_MAX;
81 	lim->max_dev_sectors = UINT_MAX;
82 	lim->max_write_same_sectors = UINT_MAX;
83 	lim->max_write_zeroes_sectors = UINT_MAX;
84 }
85 EXPORT_SYMBOL(blk_set_stacking_limits);
86 
87 /**
88  * blk_queue_make_request - define an alternate make_request function for a device
89  * @q:  the request queue for the device to be affected
90  * @mfn: the alternate make_request function
91  *
92  * Description:
93  *    The normal way for &struct bios to be passed to a device
94  *    driver is for them to be collected into requests on a request
95  *    queue, and then to allow the device driver to select requests
96  *    off that queue when it is ready.  This works well for many block
97  *    devices. However some block devices (typically virtual devices
98  *    such as md or lvm) do not benefit from the processing on the
99  *    request queue, and are served best by having the requests passed
100  *    directly to them.  This can be achieved by providing a function
101  *    to blk_queue_make_request().
102  *
103  * Caveat:
104  *    The driver that does this *must* be able to deal appropriately
105  *    with buffers in "highmemory". This can be accomplished by either calling
106  *    kmap_atomic() to get a temporary kernel mapping, or by calling
107  *    blk_queue_bounce() to create a buffer in normal memory.
108  **/
109 void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
110 {
111 	/*
112 	 * set defaults
113 	 */
114 	q->nr_requests = BLKDEV_MAX_RQ;
115 
116 	q->make_request_fn = mfn;
117 	blk_queue_dma_alignment(q, 511);
118 
119 	blk_set_default_limits(&q->limits);
120 }
121 EXPORT_SYMBOL(blk_queue_make_request);
122 
123 /**
124  * blk_queue_bounce_limit - set bounce buffer limit for queue
125  * @q: the request queue for the device
126  * @max_addr: the maximum address the device can handle
127  *
128  * Description:
129  *    Different hardware can have different requirements as to what pages
130  *    it can do I/O directly to. A low level driver can call
131  *    blk_queue_bounce_limit to have lower memory pages allocated as bounce
132  *    buffers for doing I/O to pages residing above @max_addr.
133  **/
134 void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
135 {
136 	unsigned long b_pfn = max_addr >> PAGE_SHIFT;
137 	int dma = 0;
138 
139 	q->bounce_gfp = GFP_NOIO;
140 #if BITS_PER_LONG == 64
141 	/*
142 	 * Assume anything <= 4GB can be handled by IOMMU.  Actually
143 	 * some IOMMUs can handle everything, but I don't know of a
144 	 * way to test this here.
145 	 */
146 	if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
147 		dma = 1;
148 	q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
149 #else
150 	if (b_pfn < blk_max_low_pfn)
151 		dma = 1;
152 	q->limits.bounce_pfn = b_pfn;
153 #endif
154 	if (dma) {
155 		init_emergency_isa_pool();
156 		q->bounce_gfp = GFP_NOIO | GFP_DMA;
157 		q->limits.bounce_pfn = b_pfn;
158 	}
159 }
160 EXPORT_SYMBOL(blk_queue_bounce_limit);
161 
162 /**
163  * blk_queue_max_hw_sectors - set max sectors for a request for this queue
164  * @q:  the request queue for the device
165  * @max_hw_sectors:  max hardware sectors in the usual 512b unit
166  *
167  * Description:
168  *    Enables a low level driver to set a hard upper limit,
169  *    max_hw_sectors, on the size of requests.  max_hw_sectors is set by
170  *    the device driver based upon the capabilities of the I/O
171  *    controller.
172  *
173  *    max_dev_sectors is a hard limit imposed by the storage device for
174  *    READ/WRITE requests. It is set by the disk driver.
175  *
176  *    max_sectors is a soft limit imposed by the block layer for
177  *    filesystem type requests.  This value can be overridden on a
178  *    per-device basis in /sys/block/<device>/queue/max_sectors_kb.
179  *    The soft limit can not exceed max_hw_sectors.
180  **/
181 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
182 {
183 	struct queue_limits *limits = &q->limits;
184 	unsigned int max_sectors;
185 
186 	if ((max_hw_sectors << 9) < PAGE_SIZE) {
187 		max_hw_sectors = 1 << (PAGE_SHIFT - 9);
188 		printk(KERN_INFO "%s: set to minimum %d\n",
189 		       __func__, max_hw_sectors);
190 	}
191 
192 	limits->max_hw_sectors = max_hw_sectors;
193 	max_sectors = min_not_zero(max_hw_sectors, limits->max_dev_sectors);
194 	max_sectors = min_t(unsigned int, max_sectors, BLK_DEF_MAX_SECTORS);
195 	limits->max_sectors = max_sectors;
196 	q->backing_dev_info->io_pages = max_sectors >> (PAGE_SHIFT - 9);
197 }
198 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
199 
200 /**
201  * blk_queue_chunk_sectors - set size of the chunk for this queue
202  * @q:  the request queue for the device
203  * @chunk_sectors:  chunk sectors in the usual 512b unit
204  *
205  * Description:
206  *    If a driver doesn't want IOs to cross a given chunk size, it can set
207  *    this limit and prevent merging across chunks. Note that the chunk size
208  *    must currently be a power-of-2 in sectors. Also note that the block
209  *    layer must accept a page worth of data at any offset. So if the
210  *    crossing of chunks is a hard limitation in the driver, it must still be
211  *    prepared to split single page bios.
212  **/
213 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
214 {
215 	BUG_ON(!is_power_of_2(chunk_sectors));
216 	q->limits.chunk_sectors = chunk_sectors;
217 }
218 EXPORT_SYMBOL(blk_queue_chunk_sectors);
219 
220 /**
221  * blk_queue_max_discard_sectors - set max sectors for a single discard
222  * @q:  the request queue for the device
223  * @max_discard_sectors: maximum number of sectors to discard
224  **/
225 void blk_queue_max_discard_sectors(struct request_queue *q,
226 		unsigned int max_discard_sectors)
227 {
228 	q->limits.max_hw_discard_sectors = max_discard_sectors;
229 	q->limits.max_discard_sectors = max_discard_sectors;
230 }
231 EXPORT_SYMBOL(blk_queue_max_discard_sectors);
232 
233 /**
234  * blk_queue_max_write_same_sectors - set max sectors for a single write same
235  * @q:  the request queue for the device
236  * @max_write_same_sectors: maximum number of sectors to write per command
237  **/
238 void blk_queue_max_write_same_sectors(struct request_queue *q,
239 				      unsigned int max_write_same_sectors)
240 {
241 	q->limits.max_write_same_sectors = max_write_same_sectors;
242 }
243 EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
244 
245 /**
246  * blk_queue_max_write_zeroes_sectors - set max sectors for a single
247  *                                      write zeroes
248  * @q:  the request queue for the device
249  * @max_write_zeroes_sectors: maximum number of sectors to write per command
250  **/
251 void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
252 		unsigned int max_write_zeroes_sectors)
253 {
254 	q->limits.max_write_zeroes_sectors = max_write_zeroes_sectors;
255 }
256 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
257 
258 /**
259  * blk_queue_max_segments - set max hw segments for a request for this queue
260  * @q:  the request queue for the device
261  * @max_segments:  max number of segments
262  *
263  * Description:
264  *    Enables a low level driver to set an upper limit on the number of
265  *    hw data segments in a request.
266  **/
267 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
268 {
269 	if (!max_segments) {
270 		max_segments = 1;
271 		printk(KERN_INFO "%s: set to minimum %d\n",
272 		       __func__, max_segments);
273 	}
274 
275 	q->limits.max_segments = max_segments;
276 }
277 EXPORT_SYMBOL(blk_queue_max_segments);
278 
279 /**
280  * blk_queue_max_discard_segments - set max segments for discard requests
281  * @q:  the request queue for the device
282  * @max_segments:  max number of segments
283  *
284  * Description:
285  *    Enables a low level driver to set an upper limit on the number of
286  *    segments in a discard request.
287  **/
288 void blk_queue_max_discard_segments(struct request_queue *q,
289 		unsigned short max_segments)
290 {
291 	q->limits.max_discard_segments = max_segments;
292 }
293 EXPORT_SYMBOL_GPL(blk_queue_max_discard_segments);
294 
295 /**
296  * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
297  * @q:  the request queue for the device
298  * @max_size:  max size of segment in bytes
299  *
300  * Description:
301  *    Enables a low level driver to set an upper limit on the size of a
302  *    coalesced segment
303  **/
304 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
305 {
306 	if (max_size < PAGE_SIZE) {
307 		max_size = PAGE_SIZE;
308 		printk(KERN_INFO "%s: set to minimum %d\n",
309 		       __func__, max_size);
310 	}
311 
312 	q->limits.max_segment_size = max_size;
313 }
314 EXPORT_SYMBOL(blk_queue_max_segment_size);
315 
316 /**
317  * blk_queue_logical_block_size - set logical block size for the queue
318  * @q:  the request queue for the device
319  * @size:  the logical block size, in bytes
320  *
321  * Description:
322  *   This should be set to the lowest possible block size that the
323  *   storage device can address.  The default of 512 covers most
324  *   hardware.
325  **/
326 void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
327 {
328 	q->limits.logical_block_size = size;
329 
330 	if (q->limits.physical_block_size < size)
331 		q->limits.physical_block_size = size;
332 
333 	if (q->limits.io_min < q->limits.physical_block_size)
334 		q->limits.io_min = q->limits.physical_block_size;
335 }
336 EXPORT_SYMBOL(blk_queue_logical_block_size);
337 
338 /**
339  * blk_queue_physical_block_size - set physical block size for the queue
340  * @q:  the request queue for the device
341  * @size:  the physical block size, in bytes
342  *
343  * Description:
344  *   This should be set to the lowest possible sector size that the
345  *   hardware can operate on without reverting to read-modify-write
346  *   operations.
347  */
348 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
349 {
350 	q->limits.physical_block_size = size;
351 
352 	if (q->limits.physical_block_size < q->limits.logical_block_size)
353 		q->limits.physical_block_size = q->limits.logical_block_size;
354 
355 	if (q->limits.io_min < q->limits.physical_block_size)
356 		q->limits.io_min = q->limits.physical_block_size;
357 }
358 EXPORT_SYMBOL(blk_queue_physical_block_size);
359 
360 /**
361  * blk_queue_alignment_offset - set physical block alignment offset
362  * @q:	the request queue for the device
363  * @offset: alignment offset in bytes
364  *
365  * Description:
366  *   Some devices are naturally misaligned to compensate for things like
367  *   the legacy DOS partition table 63-sector offset.  Low-level drivers
368  *   should call this function for devices whose first sector is not
369  *   naturally aligned.
370  */
371 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
372 {
373 	q->limits.alignment_offset =
374 		offset & (q->limits.physical_block_size - 1);
375 	q->limits.misaligned = 0;
376 }
377 EXPORT_SYMBOL(blk_queue_alignment_offset);
378 
379 /**
380  * blk_limits_io_min - set minimum request size for a device
381  * @limits: the queue limits
382  * @min:  smallest I/O size in bytes
383  *
384  * Description:
385  *   Some devices have an internal block size bigger than the reported
386  *   hardware sector size.  This function can be used to signal the
387  *   smallest I/O the device can perform without incurring a performance
388  *   penalty.
389  */
390 void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
391 {
392 	limits->io_min = min;
393 
394 	if (limits->io_min < limits->logical_block_size)
395 		limits->io_min = limits->logical_block_size;
396 
397 	if (limits->io_min < limits->physical_block_size)
398 		limits->io_min = limits->physical_block_size;
399 }
400 EXPORT_SYMBOL(blk_limits_io_min);
401 
402 /**
403  * blk_queue_io_min - set minimum request size for the queue
404  * @q:	the request queue for the device
405  * @min:  smallest I/O size in bytes
406  *
407  * Description:
408  *   Storage devices may report a granularity or preferred minimum I/O
409  *   size which is the smallest request the device can perform without
410  *   incurring a performance penalty.  For disk drives this is often the
411  *   physical block size.  For RAID arrays it is often the stripe chunk
412  *   size.  A properly aligned multiple of minimum_io_size is the
413  *   preferred request size for workloads where a high number of I/O
414  *   operations is desired.
415  */
416 void blk_queue_io_min(struct request_queue *q, unsigned int min)
417 {
418 	blk_limits_io_min(&q->limits, min);
419 }
420 EXPORT_SYMBOL(blk_queue_io_min);
421 
422 /**
423  * blk_limits_io_opt - set optimal request size for a device
424  * @limits: the queue limits
425  * @opt:  smallest I/O size in bytes
426  *
427  * Description:
428  *   Storage devices may report an optimal I/O size, which is the
429  *   device's preferred unit for sustained I/O.  This is rarely reported
430  *   for disk drives.  For RAID arrays it is usually the stripe width or
431  *   the internal track size.  A properly aligned multiple of
432  *   optimal_io_size is the preferred request size for workloads where
433  *   sustained throughput is desired.
434  */
435 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
436 {
437 	limits->io_opt = opt;
438 }
439 EXPORT_SYMBOL(blk_limits_io_opt);
440 
441 /**
442  * blk_queue_io_opt - set optimal request size for the queue
443  * @q:	the request queue for the device
444  * @opt:  optimal request size in bytes
445  *
446  * Description:
447  *   Storage devices may report an optimal I/O size, which is the
448  *   device's preferred unit for sustained I/O.  This is rarely reported
449  *   for disk drives.  For RAID arrays it is usually the stripe width or
450  *   the internal track size.  A properly aligned multiple of
451  *   optimal_io_size is the preferred request size for workloads where
452  *   sustained throughput is desired.
453  */
454 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
455 {
456 	blk_limits_io_opt(&q->limits, opt);
457 }
458 EXPORT_SYMBOL(blk_queue_io_opt);
459 
460 /**
461  * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
462  * @t:	the stacking driver (top)
463  * @b:  the underlying device (bottom)
464  **/
465 void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
466 {
467 	blk_stack_limits(&t->limits, &b->limits, 0);
468 }
469 EXPORT_SYMBOL(blk_queue_stack_limits);
470 
471 /**
472  * blk_stack_limits - adjust queue_limits for stacked devices
473  * @t:	the stacking driver limits (top device)
474  * @b:  the underlying queue limits (bottom, component device)
475  * @start:  first data sector within component device
476  *
477  * Description:
478  *    This function is used by stacking drivers like MD and DM to ensure
479  *    that all component devices have compatible block sizes and
480  *    alignments.  The stacking driver must provide a queue_limits
481  *    struct (top) and then iteratively call the stacking function for
482  *    all component (bottom) devices.  The stacking function will
483  *    attempt to combine the values and ensure proper alignment.
484  *
485  *    Returns 0 if the top and bottom queue_limits are compatible.  The
486  *    top device's block sizes and alignment offsets may be adjusted to
487  *    ensure alignment with the bottom device. If no compatible sizes
488  *    and alignments exist, -1 is returned and the resulting top
489  *    queue_limits will have the misaligned flag set to indicate that
490  *    the alignment_offset is undefined.
491  */
492 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
493 		     sector_t start)
494 {
495 	unsigned int top, bottom, alignment, ret = 0;
496 
497 	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
498 	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
499 	t->max_dev_sectors = min_not_zero(t->max_dev_sectors, b->max_dev_sectors);
500 	t->max_write_same_sectors = min(t->max_write_same_sectors,
501 					b->max_write_same_sectors);
502 	t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
503 					b->max_write_zeroes_sectors);
504 	t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
505 
506 	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
507 					    b->seg_boundary_mask);
508 	t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
509 					    b->virt_boundary_mask);
510 
511 	t->max_segments = min_not_zero(t->max_segments, b->max_segments);
512 	t->max_discard_segments = min_not_zero(t->max_discard_segments,
513 					       b->max_discard_segments);
514 	t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
515 						 b->max_integrity_segments);
516 
517 	t->max_segment_size = min_not_zero(t->max_segment_size,
518 					   b->max_segment_size);
519 
520 	t->misaligned |= b->misaligned;
521 
522 	alignment = queue_limit_alignment_offset(b, start);
523 
524 	/* Bottom device has different alignment.  Check that it is
525 	 * compatible with the current top alignment.
526 	 */
527 	if (t->alignment_offset != alignment) {
528 
529 		top = max(t->physical_block_size, t->io_min)
530 			+ t->alignment_offset;
531 		bottom = max(b->physical_block_size, b->io_min) + alignment;
532 
533 		/* Verify that top and bottom intervals line up */
534 		if (max(top, bottom) % min(top, bottom)) {
535 			t->misaligned = 1;
536 			ret = -1;
537 		}
538 	}
539 
540 	t->logical_block_size = max(t->logical_block_size,
541 				    b->logical_block_size);
542 
543 	t->physical_block_size = max(t->physical_block_size,
544 				     b->physical_block_size);
545 
546 	t->io_min = max(t->io_min, b->io_min);
547 	t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
548 
549 	/* Physical block size a multiple of the logical block size? */
550 	if (t->physical_block_size & (t->logical_block_size - 1)) {
551 		t->physical_block_size = t->logical_block_size;
552 		t->misaligned = 1;
553 		ret = -1;
554 	}
555 
556 	/* Minimum I/O a multiple of the physical block size? */
557 	if (t->io_min & (t->physical_block_size - 1)) {
558 		t->io_min = t->physical_block_size;
559 		t->misaligned = 1;
560 		ret = -1;
561 	}
562 
563 	/* Optimal I/O a multiple of the physical block size? */
564 	if (t->io_opt & (t->physical_block_size - 1)) {
565 		t->io_opt = 0;
566 		t->misaligned = 1;
567 		ret = -1;
568 	}
569 
570 	t->raid_partial_stripes_expensive =
571 		max(t->raid_partial_stripes_expensive,
572 		    b->raid_partial_stripes_expensive);
573 
574 	/* Find lowest common alignment_offset */
575 	t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
576 		% max(t->physical_block_size, t->io_min);
577 
578 	/* Verify that new alignment_offset is on a logical block boundary */
579 	if (t->alignment_offset & (t->logical_block_size - 1)) {
580 		t->misaligned = 1;
581 		ret = -1;
582 	}
583 
584 	/* Discard alignment and granularity */
585 	if (b->discard_granularity) {
586 		alignment = queue_limit_discard_alignment(b, start);
587 
588 		if (t->discard_granularity != 0 &&
589 		    t->discard_alignment != alignment) {
590 			top = t->discard_granularity + t->discard_alignment;
591 			bottom = b->discard_granularity + alignment;
592 
593 			/* Verify that top and bottom intervals line up */
594 			if ((max(top, bottom) % min(top, bottom)) != 0)
595 				t->discard_misaligned = 1;
596 		}
597 
598 		t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
599 						      b->max_discard_sectors);
600 		t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
601 							 b->max_hw_discard_sectors);
602 		t->discard_granularity = max(t->discard_granularity,
603 					     b->discard_granularity);
604 		t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
605 			t->discard_granularity;
606 	}
607 
608 	if (b->chunk_sectors)
609 		t->chunk_sectors = min_not_zero(t->chunk_sectors,
610 						b->chunk_sectors);
611 
612 	return ret;
613 }
614 EXPORT_SYMBOL(blk_stack_limits);
615 
616 /**
617  * bdev_stack_limits - adjust queue limits for stacked drivers
618  * @t:	the stacking driver limits (top device)
619  * @bdev:  the component block_device (bottom)
620  * @start:  first data sector within component device
621  *
622  * Description:
623  *    Merges queue limits for a top device and a block_device.  Returns
624  *    0 if alignment didn't change.  Returns -1 if adding the bottom
625  *    device caused misalignment.
626  */
627 int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
628 		      sector_t start)
629 {
630 	struct request_queue *bq = bdev_get_queue(bdev);
631 
632 	start += get_start_sect(bdev);
633 
634 	return blk_stack_limits(t, &bq->limits, start);
635 }
636 EXPORT_SYMBOL(bdev_stack_limits);
637 
638 /**
639  * disk_stack_limits - adjust queue limits for stacked drivers
640  * @disk:  MD/DM gendisk (top)
641  * @bdev:  the underlying block device (bottom)
642  * @offset:  offset to beginning of data within component device
643  *
644  * Description:
645  *    Merges the limits for a top level gendisk and a bottom level
646  *    block_device.
647  */
648 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
649 		       sector_t offset)
650 {
651 	struct request_queue *t = disk->queue;
652 
653 	if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
654 		char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
655 
656 		disk_name(disk, 0, top);
657 		bdevname(bdev, bottom);
658 
659 		printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
660 		       top, bottom);
661 	}
662 }
663 EXPORT_SYMBOL(disk_stack_limits);
664 
665 /**
666  * blk_queue_dma_pad - set pad mask
667  * @q:     the request queue for the device
668  * @mask:  pad mask
669  *
670  * Set dma pad mask.
671  *
672  * Appending pad buffer to a request modifies the last entry of a
673  * scatter list such that it includes the pad buffer.
674  **/
675 void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
676 {
677 	q->dma_pad_mask = mask;
678 }
679 EXPORT_SYMBOL(blk_queue_dma_pad);
680 
681 /**
682  * blk_queue_update_dma_pad - update pad mask
683  * @q:     the request queue for the device
684  * @mask:  pad mask
685  *
686  * Update dma pad mask.
687  *
688  * Appending pad buffer to a request modifies the last entry of a
689  * scatter list such that it includes the pad buffer.
690  **/
691 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
692 {
693 	if (mask > q->dma_pad_mask)
694 		q->dma_pad_mask = mask;
695 }
696 EXPORT_SYMBOL(blk_queue_update_dma_pad);
697 
698 /**
699  * blk_queue_dma_drain - Set up a drain buffer for excess dma.
700  * @q:  the request queue for the device
701  * @dma_drain_needed: fn which returns non-zero if drain is necessary
702  * @buf:	physically contiguous buffer
703  * @size:	size of the buffer in bytes
704  *
705  * Some devices have excess DMA problems and can't simply discard (or
706  * zero fill) the unwanted piece of the transfer.  They have to have a
707  * real area of memory to transfer it into.  The use case for this is
708  * ATAPI devices in DMA mode.  If the packet command causes a transfer
709  * bigger than the transfer size some HBAs will lock up if there
710  * aren't DMA elements to contain the excess transfer.  What this API
711  * does is adjust the queue so that the buf is always appended
712  * silently to the scatterlist.
713  *
714  * Note: This routine adjusts max_hw_segments to make room for appending
715  * the drain buffer.  If you call blk_queue_max_segments() after calling
716  * this routine, you must set the limit to one fewer than your device
717  * can support otherwise there won't be room for the drain buffer.
718  */
719 int blk_queue_dma_drain(struct request_queue *q,
720 			       dma_drain_needed_fn *dma_drain_needed,
721 			       void *buf, unsigned int size)
722 {
723 	if (queue_max_segments(q) < 2)
724 		return -EINVAL;
725 	/* make room for appending the drain */
726 	blk_queue_max_segments(q, queue_max_segments(q) - 1);
727 	q->dma_drain_needed = dma_drain_needed;
728 	q->dma_drain_buffer = buf;
729 	q->dma_drain_size = size;
730 
731 	return 0;
732 }
733 EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
734 
735 /**
736  * blk_queue_segment_boundary - set boundary rules for segment merging
737  * @q:  the request queue for the device
738  * @mask:  the memory boundary mask
739  **/
740 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
741 {
742 	if (mask < PAGE_SIZE - 1) {
743 		mask = PAGE_SIZE - 1;
744 		printk(KERN_INFO "%s: set to minimum %lx\n",
745 		       __func__, mask);
746 	}
747 
748 	q->limits.seg_boundary_mask = mask;
749 }
750 EXPORT_SYMBOL(blk_queue_segment_boundary);
751 
752 /**
753  * blk_queue_virt_boundary - set boundary rules for bio merging
754  * @q:  the request queue for the device
755  * @mask:  the memory boundary mask
756  **/
757 void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
758 {
759 	q->limits.virt_boundary_mask = mask;
760 }
761 EXPORT_SYMBOL(blk_queue_virt_boundary);
762 
763 /**
764  * blk_queue_dma_alignment - set dma length and memory alignment
765  * @q:     the request queue for the device
766  * @mask:  alignment mask
767  *
768  * description:
769  *    set required memory and length alignment for direct dma transactions.
770  *    this is used when building direct io requests for the queue.
771  *
772  **/
773 void blk_queue_dma_alignment(struct request_queue *q, int mask)
774 {
775 	q->dma_alignment = mask;
776 }
777 EXPORT_SYMBOL(blk_queue_dma_alignment);
778 
779 /**
780  * blk_queue_update_dma_alignment - update dma length and memory alignment
781  * @q:     the request queue for the device
782  * @mask:  alignment mask
783  *
784  * description:
785  *    update required memory and length alignment for direct dma transactions.
786  *    If the requested alignment is larger than the current alignment, then
787  *    the current queue alignment is updated to the new value, otherwise it
788  *    is left alone.  The design of this is to allow multiple objects
789  *    (driver, device, transport etc) to set their respective
790  *    alignments without having them interfere.
791  *
792  **/
793 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
794 {
795 	BUG_ON(mask > PAGE_SIZE);
796 
797 	if (mask > q->dma_alignment)
798 		q->dma_alignment = mask;
799 }
800 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
801 
802 /**
803  * blk_set_queue_depth - tell the block layer about the device queue depth
804  * @q:		the request queue for the device
805  * @depth:		queue depth
806  *
807  */
808 void blk_set_queue_depth(struct request_queue *q, unsigned int depth)
809 {
810 	q->queue_depth = depth;
811 	wbt_set_queue_depth(q, depth);
812 }
813 EXPORT_SYMBOL(blk_set_queue_depth);
814 
815 /**
816  * blk_queue_write_cache - configure queue's write cache
817  * @q:		the request queue for the device
818  * @wc:		write back cache on or off
819  * @fua:	device supports FUA writes, if true
820  *
821  * Tell the block layer about the write cache of @q.
822  */
823 void blk_queue_write_cache(struct request_queue *q, bool wc, bool fua)
824 {
825 	if (wc)
826 		blk_queue_flag_set(QUEUE_FLAG_WC, q);
827 	else
828 		blk_queue_flag_clear(QUEUE_FLAG_WC, q);
829 	if (fua)
830 		blk_queue_flag_set(QUEUE_FLAG_FUA, q);
831 	else
832 		blk_queue_flag_clear(QUEUE_FLAG_FUA, q);
833 
834 	wbt_set_write_cache(q, test_bit(QUEUE_FLAG_WC, &q->queue_flags));
835 }
836 EXPORT_SYMBOL_GPL(blk_queue_write_cache);
837 
838 static int __init blk_settings_init(void)
839 {
840 	blk_max_low_pfn = max_low_pfn - 1;
841 	blk_max_pfn = max_pfn - 1;
842 	return 0;
843 }
844 subsys_initcall(blk_settings_init);
845