xref: /openbmc/linux/block/blk-settings.c (revision a8da474e)
1 /*
2  * Functions related to setting various queue properties from drivers
3  */
4 #include <linux/kernel.h>
5 #include <linux/module.h>
6 #include <linux/init.h>
7 #include <linux/bio.h>
8 #include <linux/blkdev.h>
9 #include <linux/bootmem.h>	/* for max_pfn/max_low_pfn */
10 #include <linux/gcd.h>
11 #include <linux/lcm.h>
12 #include <linux/jiffies.h>
13 #include <linux/gfp.h>
14 
15 #include "blk.h"
16 
17 unsigned long blk_max_low_pfn;
18 EXPORT_SYMBOL(blk_max_low_pfn);
19 
20 unsigned long blk_max_pfn;
21 
22 /**
23  * blk_queue_prep_rq - set a prepare_request function for queue
24  * @q:		queue
25  * @pfn:	prepare_request function
26  *
27  * It's possible for a queue to register a prepare_request callback which
28  * is invoked before the request is handed to the request_fn. The goal of
29  * the function is to prepare a request for I/O, it can be used to build a
30  * cdb from the request data for instance.
31  *
32  */
33 void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
34 {
35 	q->prep_rq_fn = pfn;
36 }
37 EXPORT_SYMBOL(blk_queue_prep_rq);
38 
39 /**
40  * blk_queue_unprep_rq - set an unprepare_request function for queue
41  * @q:		queue
42  * @ufn:	unprepare_request function
43  *
44  * It's possible for a queue to register an unprepare_request callback
45  * which is invoked before the request is finally completed. The goal
46  * of the function is to deallocate any data that was allocated in the
47  * prepare_request callback.
48  *
49  */
50 void blk_queue_unprep_rq(struct request_queue *q, unprep_rq_fn *ufn)
51 {
52 	q->unprep_rq_fn = ufn;
53 }
54 EXPORT_SYMBOL(blk_queue_unprep_rq);
55 
56 void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
57 {
58 	q->softirq_done_fn = fn;
59 }
60 EXPORT_SYMBOL(blk_queue_softirq_done);
61 
62 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
63 {
64 	q->rq_timeout = timeout;
65 }
66 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
67 
68 void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
69 {
70 	q->rq_timed_out_fn = fn;
71 }
72 EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
73 
74 void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
75 {
76 	q->lld_busy_fn = fn;
77 }
78 EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
79 
80 /**
81  * blk_set_default_limits - reset limits to default values
82  * @lim:  the queue_limits structure to reset
83  *
84  * Description:
85  *   Returns a queue_limit struct to its default state.
86  */
87 void blk_set_default_limits(struct queue_limits *lim)
88 {
89 	lim->max_segments = BLK_MAX_SEGMENTS;
90 	lim->max_integrity_segments = 0;
91 	lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
92 	lim->virt_boundary_mask = 0;
93 	lim->max_segment_size = BLK_MAX_SEGMENT_SIZE;
94 	lim->max_sectors = lim->max_hw_sectors = BLK_SAFE_MAX_SECTORS;
95 	lim->chunk_sectors = 0;
96 	lim->max_write_same_sectors = 0;
97 	lim->max_discard_sectors = 0;
98 	lim->max_hw_discard_sectors = 0;
99 	lim->discard_granularity = 0;
100 	lim->discard_alignment = 0;
101 	lim->discard_misaligned = 0;
102 	lim->discard_zeroes_data = 0;
103 	lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
104 	lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
105 	lim->alignment_offset = 0;
106 	lim->io_opt = 0;
107 	lim->misaligned = 0;
108 	lim->cluster = 1;
109 }
110 EXPORT_SYMBOL(blk_set_default_limits);
111 
112 /**
113  * blk_set_stacking_limits - set default limits for stacking devices
114  * @lim:  the queue_limits structure to reset
115  *
116  * Description:
117  *   Returns a queue_limit struct to its default state. Should be used
118  *   by stacking drivers like DM that have no internal limits.
119  */
120 void blk_set_stacking_limits(struct queue_limits *lim)
121 {
122 	blk_set_default_limits(lim);
123 
124 	/* Inherit limits from component devices */
125 	lim->discard_zeroes_data = 1;
126 	lim->max_segments = USHRT_MAX;
127 	lim->max_hw_sectors = UINT_MAX;
128 	lim->max_segment_size = UINT_MAX;
129 	lim->max_sectors = UINT_MAX;
130 	lim->max_write_same_sectors = UINT_MAX;
131 }
132 EXPORT_SYMBOL(blk_set_stacking_limits);
133 
134 /**
135  * blk_queue_make_request - define an alternate make_request function for a device
136  * @q:  the request queue for the device to be affected
137  * @mfn: the alternate make_request function
138  *
139  * Description:
140  *    The normal way for &struct bios to be passed to a device
141  *    driver is for them to be collected into requests on a request
142  *    queue, and then to allow the device driver to select requests
143  *    off that queue when it is ready.  This works well for many block
144  *    devices. However some block devices (typically virtual devices
145  *    such as md or lvm) do not benefit from the processing on the
146  *    request queue, and are served best by having the requests passed
147  *    directly to them.  This can be achieved by providing a function
148  *    to blk_queue_make_request().
149  *
150  * Caveat:
151  *    The driver that does this *must* be able to deal appropriately
152  *    with buffers in "highmemory". This can be accomplished by either calling
153  *    __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
154  *    blk_queue_bounce() to create a buffer in normal memory.
155  **/
156 void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
157 {
158 	/*
159 	 * set defaults
160 	 */
161 	q->nr_requests = BLKDEV_MAX_RQ;
162 
163 	q->make_request_fn = mfn;
164 	blk_queue_dma_alignment(q, 511);
165 	blk_queue_congestion_threshold(q);
166 	q->nr_batching = BLK_BATCH_REQ;
167 
168 	blk_set_default_limits(&q->limits);
169 
170 	/*
171 	 * by default assume old behaviour and bounce for any highmem page
172 	 */
173 	blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
174 }
175 EXPORT_SYMBOL(blk_queue_make_request);
176 
177 /**
178  * blk_queue_bounce_limit - set bounce buffer limit for queue
179  * @q: the request queue for the device
180  * @max_addr: the maximum address the device can handle
181  *
182  * Description:
183  *    Different hardware can have different requirements as to what pages
184  *    it can do I/O directly to. A low level driver can call
185  *    blk_queue_bounce_limit to have lower memory pages allocated as bounce
186  *    buffers for doing I/O to pages residing above @max_addr.
187  **/
188 void blk_queue_bounce_limit(struct request_queue *q, u64 max_addr)
189 {
190 	unsigned long b_pfn = max_addr >> PAGE_SHIFT;
191 	int dma = 0;
192 
193 	q->bounce_gfp = GFP_NOIO;
194 #if BITS_PER_LONG == 64
195 	/*
196 	 * Assume anything <= 4GB can be handled by IOMMU.  Actually
197 	 * some IOMMUs can handle everything, but I don't know of a
198 	 * way to test this here.
199 	 */
200 	if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
201 		dma = 1;
202 	q->limits.bounce_pfn = max(max_low_pfn, b_pfn);
203 #else
204 	if (b_pfn < blk_max_low_pfn)
205 		dma = 1;
206 	q->limits.bounce_pfn = b_pfn;
207 #endif
208 	if (dma) {
209 		init_emergency_isa_pool();
210 		q->bounce_gfp = GFP_NOIO | GFP_DMA;
211 		q->limits.bounce_pfn = b_pfn;
212 	}
213 }
214 EXPORT_SYMBOL(blk_queue_bounce_limit);
215 
216 /**
217  * blk_limits_max_hw_sectors - set hard and soft limit of max sectors for request
218  * @limits: the queue limits
219  * @max_hw_sectors:  max hardware sectors in the usual 512b unit
220  *
221  * Description:
222  *    Enables a low level driver to set a hard upper limit,
223  *    max_hw_sectors, on the size of requests.  max_hw_sectors is set by
224  *    the device driver based upon the capabilities of the I/O
225  *    controller.
226  *
227  *    max_sectors is a soft limit imposed by the block layer for
228  *    filesystem type requests.  This value can be overridden on a
229  *    per-device basis in /sys/block/<device>/queue/max_sectors_kb.
230  *    The soft limit can not exceed max_hw_sectors.
231  **/
232 void blk_limits_max_hw_sectors(struct queue_limits *limits, unsigned int max_hw_sectors)
233 {
234 	if ((max_hw_sectors << 9) < PAGE_CACHE_SIZE) {
235 		max_hw_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
236 		printk(KERN_INFO "%s: set to minimum %d\n",
237 		       __func__, max_hw_sectors);
238 	}
239 
240 	limits->max_hw_sectors = max_hw_sectors;
241 	limits->max_sectors = min_t(unsigned int, max_hw_sectors,
242 				    BLK_DEF_MAX_SECTORS);
243 }
244 EXPORT_SYMBOL(blk_limits_max_hw_sectors);
245 
246 /**
247  * blk_queue_max_hw_sectors - set max sectors for a request for this queue
248  * @q:  the request queue for the device
249  * @max_hw_sectors:  max hardware sectors in the usual 512b unit
250  *
251  * Description:
252  *    See description for blk_limits_max_hw_sectors().
253  **/
254 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
255 {
256 	blk_limits_max_hw_sectors(&q->limits, max_hw_sectors);
257 }
258 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
259 
260 /**
261  * blk_queue_chunk_sectors - set size of the chunk for this queue
262  * @q:  the request queue for the device
263  * @chunk_sectors:  chunk sectors in the usual 512b unit
264  *
265  * Description:
266  *    If a driver doesn't want IOs to cross a given chunk size, it can set
267  *    this limit and prevent merging across chunks. Note that the chunk size
268  *    must currently be a power-of-2 in sectors. Also note that the block
269  *    layer must accept a page worth of data at any offset. So if the
270  *    crossing of chunks is a hard limitation in the driver, it must still be
271  *    prepared to split single page bios.
272  **/
273 void blk_queue_chunk_sectors(struct request_queue *q, unsigned int chunk_sectors)
274 {
275 	BUG_ON(!is_power_of_2(chunk_sectors));
276 	q->limits.chunk_sectors = chunk_sectors;
277 }
278 EXPORT_SYMBOL(blk_queue_chunk_sectors);
279 
280 /**
281  * blk_queue_max_discard_sectors - set max sectors for a single discard
282  * @q:  the request queue for the device
283  * @max_discard_sectors: maximum number of sectors to discard
284  **/
285 void blk_queue_max_discard_sectors(struct request_queue *q,
286 		unsigned int max_discard_sectors)
287 {
288 	q->limits.max_hw_discard_sectors = max_discard_sectors;
289 	q->limits.max_discard_sectors = max_discard_sectors;
290 }
291 EXPORT_SYMBOL(blk_queue_max_discard_sectors);
292 
293 /**
294  * blk_queue_max_write_same_sectors - set max sectors for a single write same
295  * @q:  the request queue for the device
296  * @max_write_same_sectors: maximum number of sectors to write per command
297  **/
298 void blk_queue_max_write_same_sectors(struct request_queue *q,
299 				      unsigned int max_write_same_sectors)
300 {
301 	q->limits.max_write_same_sectors = max_write_same_sectors;
302 }
303 EXPORT_SYMBOL(blk_queue_max_write_same_sectors);
304 
305 /**
306  * blk_queue_max_segments - set max hw segments for a request for this queue
307  * @q:  the request queue for the device
308  * @max_segments:  max number of segments
309  *
310  * Description:
311  *    Enables a low level driver to set an upper limit on the number of
312  *    hw data segments in a request.
313  **/
314 void blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
315 {
316 	if (!max_segments) {
317 		max_segments = 1;
318 		printk(KERN_INFO "%s: set to minimum %d\n",
319 		       __func__, max_segments);
320 	}
321 
322 	q->limits.max_segments = max_segments;
323 }
324 EXPORT_SYMBOL(blk_queue_max_segments);
325 
326 /**
327  * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
328  * @q:  the request queue for the device
329  * @max_size:  max size of segment in bytes
330  *
331  * Description:
332  *    Enables a low level driver to set an upper limit on the size of a
333  *    coalesced segment
334  **/
335 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
336 {
337 	if (max_size < PAGE_CACHE_SIZE) {
338 		max_size = PAGE_CACHE_SIZE;
339 		printk(KERN_INFO "%s: set to minimum %d\n",
340 		       __func__, max_size);
341 	}
342 
343 	q->limits.max_segment_size = max_size;
344 }
345 EXPORT_SYMBOL(blk_queue_max_segment_size);
346 
347 /**
348  * blk_queue_logical_block_size - set logical block size for the queue
349  * @q:  the request queue for the device
350  * @size:  the logical block size, in bytes
351  *
352  * Description:
353  *   This should be set to the lowest possible block size that the
354  *   storage device can address.  The default of 512 covers most
355  *   hardware.
356  **/
357 void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
358 {
359 	q->limits.logical_block_size = size;
360 
361 	if (q->limits.physical_block_size < size)
362 		q->limits.physical_block_size = size;
363 
364 	if (q->limits.io_min < q->limits.physical_block_size)
365 		q->limits.io_min = q->limits.physical_block_size;
366 }
367 EXPORT_SYMBOL(blk_queue_logical_block_size);
368 
369 /**
370  * blk_queue_physical_block_size - set physical block size for the queue
371  * @q:  the request queue for the device
372  * @size:  the physical block size, in bytes
373  *
374  * Description:
375  *   This should be set to the lowest possible sector size that the
376  *   hardware can operate on without reverting to read-modify-write
377  *   operations.
378  */
379 void blk_queue_physical_block_size(struct request_queue *q, unsigned int size)
380 {
381 	q->limits.physical_block_size = size;
382 
383 	if (q->limits.physical_block_size < q->limits.logical_block_size)
384 		q->limits.physical_block_size = q->limits.logical_block_size;
385 
386 	if (q->limits.io_min < q->limits.physical_block_size)
387 		q->limits.io_min = q->limits.physical_block_size;
388 }
389 EXPORT_SYMBOL(blk_queue_physical_block_size);
390 
391 /**
392  * blk_queue_alignment_offset - set physical block alignment offset
393  * @q:	the request queue for the device
394  * @offset: alignment offset in bytes
395  *
396  * Description:
397  *   Some devices are naturally misaligned to compensate for things like
398  *   the legacy DOS partition table 63-sector offset.  Low-level drivers
399  *   should call this function for devices whose first sector is not
400  *   naturally aligned.
401  */
402 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
403 {
404 	q->limits.alignment_offset =
405 		offset & (q->limits.physical_block_size - 1);
406 	q->limits.misaligned = 0;
407 }
408 EXPORT_SYMBOL(blk_queue_alignment_offset);
409 
410 /**
411  * blk_limits_io_min - set minimum request size for a device
412  * @limits: the queue limits
413  * @min:  smallest I/O size in bytes
414  *
415  * Description:
416  *   Some devices have an internal block size bigger than the reported
417  *   hardware sector size.  This function can be used to signal the
418  *   smallest I/O the device can perform without incurring a performance
419  *   penalty.
420  */
421 void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
422 {
423 	limits->io_min = min;
424 
425 	if (limits->io_min < limits->logical_block_size)
426 		limits->io_min = limits->logical_block_size;
427 
428 	if (limits->io_min < limits->physical_block_size)
429 		limits->io_min = limits->physical_block_size;
430 }
431 EXPORT_SYMBOL(blk_limits_io_min);
432 
433 /**
434  * blk_queue_io_min - set minimum request size for the queue
435  * @q:	the request queue for the device
436  * @min:  smallest I/O size in bytes
437  *
438  * Description:
439  *   Storage devices may report a granularity or preferred minimum I/O
440  *   size which is the smallest request the device can perform without
441  *   incurring a performance penalty.  For disk drives this is often the
442  *   physical block size.  For RAID arrays it is often the stripe chunk
443  *   size.  A properly aligned multiple of minimum_io_size is the
444  *   preferred request size for workloads where a high number of I/O
445  *   operations is desired.
446  */
447 void blk_queue_io_min(struct request_queue *q, unsigned int min)
448 {
449 	blk_limits_io_min(&q->limits, min);
450 }
451 EXPORT_SYMBOL(blk_queue_io_min);
452 
453 /**
454  * blk_limits_io_opt - set optimal request size for a device
455  * @limits: the queue limits
456  * @opt:  smallest I/O size in bytes
457  *
458  * Description:
459  *   Storage devices may report an optimal I/O size, which is the
460  *   device's preferred unit for sustained I/O.  This is rarely reported
461  *   for disk drives.  For RAID arrays it is usually the stripe width or
462  *   the internal track size.  A properly aligned multiple of
463  *   optimal_io_size is the preferred request size for workloads where
464  *   sustained throughput is desired.
465  */
466 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
467 {
468 	limits->io_opt = opt;
469 }
470 EXPORT_SYMBOL(blk_limits_io_opt);
471 
472 /**
473  * blk_queue_io_opt - set optimal request size for the queue
474  * @q:	the request queue for the device
475  * @opt:  optimal request size in bytes
476  *
477  * Description:
478  *   Storage devices may report an optimal I/O size, which is the
479  *   device's preferred unit for sustained I/O.  This is rarely reported
480  *   for disk drives.  For RAID arrays it is usually the stripe width or
481  *   the internal track size.  A properly aligned multiple of
482  *   optimal_io_size is the preferred request size for workloads where
483  *   sustained throughput is desired.
484  */
485 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
486 {
487 	blk_limits_io_opt(&q->limits, opt);
488 }
489 EXPORT_SYMBOL(blk_queue_io_opt);
490 
491 /**
492  * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
493  * @t:	the stacking driver (top)
494  * @b:  the underlying device (bottom)
495  **/
496 void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
497 {
498 	blk_stack_limits(&t->limits, &b->limits, 0);
499 }
500 EXPORT_SYMBOL(blk_queue_stack_limits);
501 
502 /**
503  * blk_stack_limits - adjust queue_limits for stacked devices
504  * @t:	the stacking driver limits (top device)
505  * @b:  the underlying queue limits (bottom, component device)
506  * @start:  first data sector within component device
507  *
508  * Description:
509  *    This function is used by stacking drivers like MD and DM to ensure
510  *    that all component devices have compatible block sizes and
511  *    alignments.  The stacking driver must provide a queue_limits
512  *    struct (top) and then iteratively call the stacking function for
513  *    all component (bottom) devices.  The stacking function will
514  *    attempt to combine the values and ensure proper alignment.
515  *
516  *    Returns 0 if the top and bottom queue_limits are compatible.  The
517  *    top device's block sizes and alignment offsets may be adjusted to
518  *    ensure alignment with the bottom device. If no compatible sizes
519  *    and alignments exist, -1 is returned and the resulting top
520  *    queue_limits will have the misaligned flag set to indicate that
521  *    the alignment_offset is undefined.
522  */
523 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
524 		     sector_t start)
525 {
526 	unsigned int top, bottom, alignment, ret = 0;
527 
528 	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
529 	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
530 	t->max_write_same_sectors = min(t->max_write_same_sectors,
531 					b->max_write_same_sectors);
532 	t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
533 
534 	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
535 					    b->seg_boundary_mask);
536 	t->virt_boundary_mask = min_not_zero(t->virt_boundary_mask,
537 					    b->virt_boundary_mask);
538 
539 	t->max_segments = min_not_zero(t->max_segments, b->max_segments);
540 	t->max_integrity_segments = min_not_zero(t->max_integrity_segments,
541 						 b->max_integrity_segments);
542 
543 	t->max_segment_size = min_not_zero(t->max_segment_size,
544 					   b->max_segment_size);
545 
546 	t->misaligned |= b->misaligned;
547 
548 	alignment = queue_limit_alignment_offset(b, start);
549 
550 	/* Bottom device has different alignment.  Check that it is
551 	 * compatible with the current top alignment.
552 	 */
553 	if (t->alignment_offset != alignment) {
554 
555 		top = max(t->physical_block_size, t->io_min)
556 			+ t->alignment_offset;
557 		bottom = max(b->physical_block_size, b->io_min) + alignment;
558 
559 		/* Verify that top and bottom intervals line up */
560 		if (max(top, bottom) % min(top, bottom)) {
561 			t->misaligned = 1;
562 			ret = -1;
563 		}
564 	}
565 
566 	t->logical_block_size = max(t->logical_block_size,
567 				    b->logical_block_size);
568 
569 	t->physical_block_size = max(t->physical_block_size,
570 				     b->physical_block_size);
571 
572 	t->io_min = max(t->io_min, b->io_min);
573 	t->io_opt = lcm_not_zero(t->io_opt, b->io_opt);
574 
575 	t->cluster &= b->cluster;
576 	t->discard_zeroes_data &= b->discard_zeroes_data;
577 
578 	/* Physical block size a multiple of the logical block size? */
579 	if (t->physical_block_size & (t->logical_block_size - 1)) {
580 		t->physical_block_size = t->logical_block_size;
581 		t->misaligned = 1;
582 		ret = -1;
583 	}
584 
585 	/* Minimum I/O a multiple of the physical block size? */
586 	if (t->io_min & (t->physical_block_size - 1)) {
587 		t->io_min = t->physical_block_size;
588 		t->misaligned = 1;
589 		ret = -1;
590 	}
591 
592 	/* Optimal I/O a multiple of the physical block size? */
593 	if (t->io_opt & (t->physical_block_size - 1)) {
594 		t->io_opt = 0;
595 		t->misaligned = 1;
596 		ret = -1;
597 	}
598 
599 	t->raid_partial_stripes_expensive =
600 		max(t->raid_partial_stripes_expensive,
601 		    b->raid_partial_stripes_expensive);
602 
603 	/* Find lowest common alignment_offset */
604 	t->alignment_offset = lcm_not_zero(t->alignment_offset, alignment)
605 		% max(t->physical_block_size, t->io_min);
606 
607 	/* Verify that new alignment_offset is on a logical block boundary */
608 	if (t->alignment_offset & (t->logical_block_size - 1)) {
609 		t->misaligned = 1;
610 		ret = -1;
611 	}
612 
613 	/* Discard alignment and granularity */
614 	if (b->discard_granularity) {
615 		alignment = queue_limit_discard_alignment(b, start);
616 
617 		if (t->discard_granularity != 0 &&
618 		    t->discard_alignment != alignment) {
619 			top = t->discard_granularity + t->discard_alignment;
620 			bottom = b->discard_granularity + alignment;
621 
622 			/* Verify that top and bottom intervals line up */
623 			if ((max(top, bottom) % min(top, bottom)) != 0)
624 				t->discard_misaligned = 1;
625 		}
626 
627 		t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
628 						      b->max_discard_sectors);
629 		t->max_hw_discard_sectors = min_not_zero(t->max_hw_discard_sectors,
630 							 b->max_hw_discard_sectors);
631 		t->discard_granularity = max(t->discard_granularity,
632 					     b->discard_granularity);
633 		t->discard_alignment = lcm_not_zero(t->discard_alignment, alignment) %
634 			t->discard_granularity;
635 	}
636 
637 	return ret;
638 }
639 EXPORT_SYMBOL(blk_stack_limits);
640 
641 /**
642  * bdev_stack_limits - adjust queue limits for stacked drivers
643  * @t:	the stacking driver limits (top device)
644  * @bdev:  the component block_device (bottom)
645  * @start:  first data sector within component device
646  *
647  * Description:
648  *    Merges queue limits for a top device and a block_device.  Returns
649  *    0 if alignment didn't change.  Returns -1 if adding the bottom
650  *    device caused misalignment.
651  */
652 int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
653 		      sector_t start)
654 {
655 	struct request_queue *bq = bdev_get_queue(bdev);
656 
657 	start += get_start_sect(bdev);
658 
659 	return blk_stack_limits(t, &bq->limits, start);
660 }
661 EXPORT_SYMBOL(bdev_stack_limits);
662 
663 /**
664  * disk_stack_limits - adjust queue limits for stacked drivers
665  * @disk:  MD/DM gendisk (top)
666  * @bdev:  the underlying block device (bottom)
667  * @offset:  offset to beginning of data within component device
668  *
669  * Description:
670  *    Merges the limits for a top level gendisk and a bottom level
671  *    block_device.
672  */
673 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
674 		       sector_t offset)
675 {
676 	struct request_queue *t = disk->queue;
677 
678 	if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) {
679 		char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
680 
681 		disk_name(disk, 0, top);
682 		bdevname(bdev, bottom);
683 
684 		printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
685 		       top, bottom);
686 	}
687 }
688 EXPORT_SYMBOL(disk_stack_limits);
689 
690 /**
691  * blk_queue_dma_pad - set pad mask
692  * @q:     the request queue for the device
693  * @mask:  pad mask
694  *
695  * Set dma pad mask.
696  *
697  * Appending pad buffer to a request modifies the last entry of a
698  * scatter list such that it includes the pad buffer.
699  **/
700 void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
701 {
702 	q->dma_pad_mask = mask;
703 }
704 EXPORT_SYMBOL(blk_queue_dma_pad);
705 
706 /**
707  * blk_queue_update_dma_pad - update pad mask
708  * @q:     the request queue for the device
709  * @mask:  pad mask
710  *
711  * Update dma pad mask.
712  *
713  * Appending pad buffer to a request modifies the last entry of a
714  * scatter list such that it includes the pad buffer.
715  **/
716 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
717 {
718 	if (mask > q->dma_pad_mask)
719 		q->dma_pad_mask = mask;
720 }
721 EXPORT_SYMBOL(blk_queue_update_dma_pad);
722 
723 /**
724  * blk_queue_dma_drain - Set up a drain buffer for excess dma.
725  * @q:  the request queue for the device
726  * @dma_drain_needed: fn which returns non-zero if drain is necessary
727  * @buf:	physically contiguous buffer
728  * @size:	size of the buffer in bytes
729  *
730  * Some devices have excess DMA problems and can't simply discard (or
731  * zero fill) the unwanted piece of the transfer.  They have to have a
732  * real area of memory to transfer it into.  The use case for this is
733  * ATAPI devices in DMA mode.  If the packet command causes a transfer
734  * bigger than the transfer size some HBAs will lock up if there
735  * aren't DMA elements to contain the excess transfer.  What this API
736  * does is adjust the queue so that the buf is always appended
737  * silently to the scatterlist.
738  *
739  * Note: This routine adjusts max_hw_segments to make room for appending
740  * the drain buffer.  If you call blk_queue_max_segments() after calling
741  * this routine, you must set the limit to one fewer than your device
742  * can support otherwise there won't be room for the drain buffer.
743  */
744 int blk_queue_dma_drain(struct request_queue *q,
745 			       dma_drain_needed_fn *dma_drain_needed,
746 			       void *buf, unsigned int size)
747 {
748 	if (queue_max_segments(q) < 2)
749 		return -EINVAL;
750 	/* make room for appending the drain */
751 	blk_queue_max_segments(q, queue_max_segments(q) - 1);
752 	q->dma_drain_needed = dma_drain_needed;
753 	q->dma_drain_buffer = buf;
754 	q->dma_drain_size = size;
755 
756 	return 0;
757 }
758 EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
759 
760 /**
761  * blk_queue_segment_boundary - set boundary rules for segment merging
762  * @q:  the request queue for the device
763  * @mask:  the memory boundary mask
764  **/
765 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
766 {
767 	if (mask < PAGE_CACHE_SIZE - 1) {
768 		mask = PAGE_CACHE_SIZE - 1;
769 		printk(KERN_INFO "%s: set to minimum %lx\n",
770 		       __func__, mask);
771 	}
772 
773 	q->limits.seg_boundary_mask = mask;
774 }
775 EXPORT_SYMBOL(blk_queue_segment_boundary);
776 
777 /**
778  * blk_queue_virt_boundary - set boundary rules for bio merging
779  * @q:  the request queue for the device
780  * @mask:  the memory boundary mask
781  **/
782 void blk_queue_virt_boundary(struct request_queue *q, unsigned long mask)
783 {
784 	q->limits.virt_boundary_mask = mask;
785 }
786 EXPORT_SYMBOL(blk_queue_virt_boundary);
787 
788 /**
789  * blk_queue_dma_alignment - set dma length and memory alignment
790  * @q:     the request queue for the device
791  * @mask:  alignment mask
792  *
793  * description:
794  *    set required memory and length alignment for direct dma transactions.
795  *    this is used when building direct io requests for the queue.
796  *
797  **/
798 void blk_queue_dma_alignment(struct request_queue *q, int mask)
799 {
800 	q->dma_alignment = mask;
801 }
802 EXPORT_SYMBOL(blk_queue_dma_alignment);
803 
804 /**
805  * blk_queue_update_dma_alignment - update dma length and memory alignment
806  * @q:     the request queue for the device
807  * @mask:  alignment mask
808  *
809  * description:
810  *    update required memory and length alignment for direct dma transactions.
811  *    If the requested alignment is larger than the current alignment, then
812  *    the current queue alignment is updated to the new value, otherwise it
813  *    is left alone.  The design of this is to allow multiple objects
814  *    (driver, device, transport etc) to set their respective
815  *    alignments without having them interfere.
816  *
817  **/
818 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
819 {
820 	BUG_ON(mask > PAGE_SIZE);
821 
822 	if (mask > q->dma_alignment)
823 		q->dma_alignment = mask;
824 }
825 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
826 
827 /**
828  * blk_queue_flush - configure queue's cache flush capability
829  * @q:		the request queue for the device
830  * @flush:	0, REQ_FLUSH or REQ_FLUSH | REQ_FUA
831  *
832  * Tell block layer cache flush capability of @q.  If it supports
833  * flushing, REQ_FLUSH should be set.  If it supports bypassing
834  * write cache for individual writes, REQ_FUA should be set.
835  */
836 void blk_queue_flush(struct request_queue *q, unsigned int flush)
837 {
838 	WARN_ON_ONCE(flush & ~(REQ_FLUSH | REQ_FUA));
839 
840 	if (WARN_ON_ONCE(!(flush & REQ_FLUSH) && (flush & REQ_FUA)))
841 		flush &= ~REQ_FUA;
842 
843 	q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
844 }
845 EXPORT_SYMBOL_GPL(blk_queue_flush);
846 
847 void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
848 {
849 	q->flush_not_queueable = !queueable;
850 }
851 EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
852 
853 static int __init blk_settings_init(void)
854 {
855 	blk_max_low_pfn = max_low_pfn - 1;
856 	blk_max_pfn = max_pfn - 1;
857 	return 0;
858 }
859 subsys_initcall(blk_settings_init);
860