xref: /linux/block/blk-settings.c (revision 27258e448eb301cf89e351df87aa8cb916653bf2)
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 
12 #include "blk.h"
13 
14 unsigned long blk_max_low_pfn;
15 EXPORT_SYMBOL(blk_max_low_pfn);
16 
17 unsigned long blk_max_pfn;
18 
19 /**
20  * blk_queue_prep_rq - set a prepare_request function for queue
21  * @q:		queue
22  * @pfn:	prepare_request function
23  *
24  * It's possible for a queue to register a prepare_request callback which
25  * is invoked before the request is handed to the request_fn. The goal of
26  * the function is to prepare a request for I/O, it can be used to build a
27  * cdb from the request data for instance.
28  *
29  */
30 void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
31 {
32 	q->prep_rq_fn = pfn;
33 }
34 EXPORT_SYMBOL(blk_queue_prep_rq);
35 
36 /**
37  * blk_queue_set_discard - set a discard_sectors function for queue
38  * @q:		queue
39  * @dfn:	prepare_discard function
40  *
41  * It's possible for a queue to register a discard callback which is used
42  * to transform a discard request into the appropriate type for the
43  * hardware. If none is registered, then discard requests are failed
44  * with %EOPNOTSUPP.
45  *
46  */
47 void blk_queue_set_discard(struct request_queue *q, prepare_discard_fn *dfn)
48 {
49 	q->prepare_discard_fn = dfn;
50 }
51 EXPORT_SYMBOL(blk_queue_set_discard);
52 
53 /**
54  * blk_queue_merge_bvec - set a merge_bvec function for queue
55  * @q:		queue
56  * @mbfn:	merge_bvec_fn
57  *
58  * Usually queues have static limitations on the max sectors or segments that
59  * we can put in a request. Stacking drivers may have some settings that
60  * are dynamic, and thus we have to query the queue whether it is ok to
61  * add a new bio_vec to a bio at a given offset or not. If the block device
62  * has such limitations, it needs to register a merge_bvec_fn to control
63  * the size of bio's sent to it. Note that a block device *must* allow a
64  * single page to be added to an empty bio. The block device driver may want
65  * to use the bio_split() function to deal with these bio's. By default
66  * no merge_bvec_fn is defined for a queue, and only the fixed limits are
67  * honored.
68  */
69 void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
70 {
71 	q->merge_bvec_fn = mbfn;
72 }
73 EXPORT_SYMBOL(blk_queue_merge_bvec);
74 
75 void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
76 {
77 	q->softirq_done_fn = fn;
78 }
79 EXPORT_SYMBOL(blk_queue_softirq_done);
80 
81 void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
82 {
83 	q->rq_timeout = timeout;
84 }
85 EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
86 
87 void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
88 {
89 	q->rq_timed_out_fn = fn;
90 }
91 EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
92 
93 void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
94 {
95 	q->lld_busy_fn = fn;
96 }
97 EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
98 
99 /**
100  * blk_set_default_limits - reset limits to default values
101  * @lim:  the queue_limits structure to reset
102  *
103  * Description:
104  *   Returns a queue_limit struct to its default state.  Can be used by
105  *   stacking drivers like DM that stage table swaps and reuse an
106  *   existing device queue.
107  */
108 void blk_set_default_limits(struct queue_limits *lim)
109 {
110 	lim->max_phys_segments = MAX_PHYS_SEGMENTS;
111 	lim->max_hw_segments = MAX_HW_SEGMENTS;
112 	lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
113 	lim->max_segment_size = MAX_SEGMENT_SIZE;
114 	lim->max_sectors = lim->max_hw_sectors = SAFE_MAX_SECTORS;
115 	lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
116 	lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
117 	lim->alignment_offset = 0;
118 	lim->io_opt = 0;
119 	lim->misaligned = 0;
120 	lim->no_cluster = 0;
121 }
122 EXPORT_SYMBOL(blk_set_default_limits);
123 
124 /**
125  * blk_queue_make_request - define an alternate make_request function for a device
126  * @q:  the request queue for the device to be affected
127  * @mfn: the alternate make_request function
128  *
129  * Description:
130  *    The normal way for &struct bios to be passed to a device
131  *    driver is for them to be collected into requests on a request
132  *    queue, and then to allow the device driver to select requests
133  *    off that queue when it is ready.  This works well for many block
134  *    devices. However some block devices (typically virtual devices
135  *    such as md or lvm) do not benefit from the processing on the
136  *    request queue, and are served best by having the requests passed
137  *    directly to them.  This can be achieved by providing a function
138  *    to blk_queue_make_request().
139  *
140  * Caveat:
141  *    The driver that does this *must* be able to deal appropriately
142  *    with buffers in "highmemory". This can be accomplished by either calling
143  *    __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
144  *    blk_queue_bounce() to create a buffer in normal memory.
145  **/
146 void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
147 {
148 	/*
149 	 * set defaults
150 	 */
151 	q->nr_requests = BLKDEV_MAX_RQ;
152 
153 	q->make_request_fn = mfn;
154 	blk_queue_dma_alignment(q, 511);
155 	blk_queue_congestion_threshold(q);
156 	q->nr_batching = BLK_BATCH_REQ;
157 
158 	q->unplug_thresh = 4;		/* hmm */
159 	q->unplug_delay = (3 * HZ) / 1000;	/* 3 milliseconds */
160 	if (q->unplug_delay == 0)
161 		q->unplug_delay = 1;
162 
163 	q->unplug_timer.function = blk_unplug_timeout;
164 	q->unplug_timer.data = (unsigned long)q;
165 
166 	blk_set_default_limits(&q->limits);
167 
168 	/*
169 	 * If the caller didn't supply a lock, fall back to our embedded
170 	 * per-queue locks
171 	 */
172 	if (!q->queue_lock)
173 		q->queue_lock = &q->__queue_lock;
174 
175 	/*
176 	 * by default assume old behaviour and bounce for any highmem page
177 	 */
178 	blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
179 }
180 EXPORT_SYMBOL(blk_queue_make_request);
181 
182 /**
183  * blk_queue_bounce_limit - set bounce buffer limit for queue
184  * @q: the request queue for the device
185  * @dma_mask: the maximum address the device can handle
186  *
187  * Description:
188  *    Different hardware can have different requirements as to what pages
189  *    it can do I/O directly to. A low level driver can call
190  *    blk_queue_bounce_limit to have lower memory pages allocated as bounce
191  *    buffers for doing I/O to pages residing above @dma_mask.
192  **/
193 void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
194 {
195 	unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
196 	int dma = 0;
197 
198 	q->bounce_gfp = GFP_NOIO;
199 #if BITS_PER_LONG == 64
200 	/*
201 	 * Assume anything <= 4GB can be handled by IOMMU.  Actually
202 	 * some IOMMUs can handle everything, but I don't know of a
203 	 * way to test this here.
204 	 */
205 	if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
206 		dma = 1;
207 	q->limits.bounce_pfn = max_low_pfn;
208 #else
209 	if (b_pfn < blk_max_low_pfn)
210 		dma = 1;
211 	q->limits.bounce_pfn = b_pfn;
212 #endif
213 	if (dma) {
214 		init_emergency_isa_pool();
215 		q->bounce_gfp = GFP_NOIO | GFP_DMA;
216 		q->limits.bounce_pfn = b_pfn;
217 	}
218 }
219 EXPORT_SYMBOL(blk_queue_bounce_limit);
220 
221 /**
222  * blk_queue_max_sectors - set max sectors for a request for this queue
223  * @q:  the request queue for the device
224  * @max_sectors:  max sectors in the usual 512b unit
225  *
226  * Description:
227  *    Enables a low level driver to set an upper limit on the size of
228  *    received requests.
229  **/
230 void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
231 {
232 	if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
233 		max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
234 		printk(KERN_INFO "%s: set to minimum %d\n",
235 		       __func__, max_sectors);
236 	}
237 
238 	if (BLK_DEF_MAX_SECTORS > max_sectors)
239 		q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors;
240 	else {
241 		q->limits.max_sectors = BLK_DEF_MAX_SECTORS;
242 		q->limits.max_hw_sectors = max_sectors;
243 	}
244 }
245 EXPORT_SYMBOL(blk_queue_max_sectors);
246 
247 void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors)
248 {
249 	if (BLK_DEF_MAX_SECTORS > max_sectors)
250 		q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS;
251 	else
252 		q->limits.max_hw_sectors = max_sectors;
253 }
254 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
255 
256 /**
257  * blk_queue_max_phys_segments - set max phys segments for a request for this queue
258  * @q:  the request queue for the device
259  * @max_segments:  max number of segments
260  *
261  * Description:
262  *    Enables a low level driver to set an upper limit on the number of
263  *    physical data segments in a request.  This would be the largest sized
264  *    scatter list the driver could handle.
265  **/
266 void blk_queue_max_phys_segments(struct request_queue *q,
267 				 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_phys_segments = max_segments;
276 }
277 EXPORT_SYMBOL(blk_queue_max_phys_segments);
278 
279 /**
280  * blk_queue_max_hw_segments - set max hw segments for a request for this queue
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  *    hw data segments in a request.  This would be the largest number of
287  *    address/length pairs the host adapter can actually give at once
288  *    to the device.
289  **/
290 void blk_queue_max_hw_segments(struct request_queue *q,
291 			       unsigned short max_segments)
292 {
293 	if (!max_segments) {
294 		max_segments = 1;
295 		printk(KERN_INFO "%s: set to minimum %d\n",
296 		       __func__, max_segments);
297 	}
298 
299 	q->limits.max_hw_segments = max_segments;
300 }
301 EXPORT_SYMBOL(blk_queue_max_hw_segments);
302 
303 /**
304  * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
305  * @q:  the request queue for the device
306  * @max_size:  max size of segment in bytes
307  *
308  * Description:
309  *    Enables a low level driver to set an upper limit on the size of a
310  *    coalesced segment
311  **/
312 void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
313 {
314 	if (max_size < PAGE_CACHE_SIZE) {
315 		max_size = PAGE_CACHE_SIZE;
316 		printk(KERN_INFO "%s: set to minimum %d\n",
317 		       __func__, max_size);
318 	}
319 
320 	q->limits.max_segment_size = max_size;
321 }
322 EXPORT_SYMBOL(blk_queue_max_segment_size);
323 
324 /**
325  * blk_queue_logical_block_size - set logical block size for the queue
326  * @q:  the request queue for the device
327  * @size:  the logical block size, in bytes
328  *
329  * Description:
330  *   This should be set to the lowest possible block size that the
331  *   storage device can address.  The default of 512 covers most
332  *   hardware.
333  **/
334 void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
335 {
336 	q->limits.logical_block_size = size;
337 
338 	if (q->limits.physical_block_size < size)
339 		q->limits.physical_block_size = size;
340 
341 	if (q->limits.io_min < q->limits.physical_block_size)
342 		q->limits.io_min = q->limits.physical_block_size;
343 }
344 EXPORT_SYMBOL(blk_queue_logical_block_size);
345 
346 /**
347  * blk_queue_physical_block_size - set physical block size for the queue
348  * @q:  the request queue for the device
349  * @size:  the physical block size, in bytes
350  *
351  * Description:
352  *   This should be set to the lowest possible sector size that the
353  *   hardware can operate on without reverting to read-modify-write
354  *   operations.
355  */
356 void blk_queue_physical_block_size(struct request_queue *q, unsigned short size)
357 {
358 	q->limits.physical_block_size = size;
359 
360 	if (q->limits.physical_block_size < q->limits.logical_block_size)
361 		q->limits.physical_block_size = q->limits.logical_block_size;
362 
363 	if (q->limits.io_min < q->limits.physical_block_size)
364 		q->limits.io_min = q->limits.physical_block_size;
365 }
366 EXPORT_SYMBOL(blk_queue_physical_block_size);
367 
368 /**
369  * blk_queue_alignment_offset - set physical block alignment offset
370  * @q:	the request queue for the device
371  * @offset: alignment offset in bytes
372  *
373  * Description:
374  *   Some devices are naturally misaligned to compensate for things like
375  *   the legacy DOS partition table 63-sector offset.  Low-level drivers
376  *   should call this function for devices whose first sector is not
377  *   naturally aligned.
378  */
379 void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
380 {
381 	q->limits.alignment_offset =
382 		offset & (q->limits.physical_block_size - 1);
383 	q->limits.misaligned = 0;
384 }
385 EXPORT_SYMBOL(blk_queue_alignment_offset);
386 
387 /**
388  * blk_limits_io_min - set minimum request size for a device
389  * @limits: the queue limits
390  * @min:  smallest I/O size in bytes
391  *
392  * Description:
393  *   Some devices have an internal block size bigger than the reported
394  *   hardware sector size.  This function can be used to signal the
395  *   smallest I/O the device can perform without incurring a performance
396  *   penalty.
397  */
398 void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
399 {
400 	limits->io_min = min;
401 
402 	if (limits->io_min < limits->logical_block_size)
403 		limits->io_min = limits->logical_block_size;
404 
405 	if (limits->io_min < limits->physical_block_size)
406 		limits->io_min = limits->physical_block_size;
407 }
408 EXPORT_SYMBOL(blk_limits_io_min);
409 
410 /**
411  * blk_queue_io_min - set minimum request size for the queue
412  * @q:	the request queue for the device
413  * @min:  smallest I/O size in bytes
414  *
415  * Description:
416  *   Storage devices may report a granularity or preferred minimum I/O
417  *   size which is the smallest request the device can perform without
418  *   incurring a performance penalty.  For disk drives this is often the
419  *   physical block size.  For RAID arrays it is often the stripe chunk
420  *   size.  A properly aligned multiple of minimum_io_size is the
421  *   preferred request size for workloads where a high number of I/O
422  *   operations is desired.
423  */
424 void blk_queue_io_min(struct request_queue *q, unsigned int min)
425 {
426 	blk_limits_io_min(&q->limits, min);
427 }
428 EXPORT_SYMBOL(blk_queue_io_min);
429 
430 /**
431  * blk_limits_io_opt - set optimal request size for a device
432  * @limits: the queue limits
433  * @opt:  smallest I/O size in bytes
434  *
435  * Description:
436  *   Storage devices may report an optimal I/O size, which is the
437  *   device's preferred unit for sustained I/O.  This is rarely reported
438  *   for disk drives.  For RAID arrays it is usually the stripe width or
439  *   the internal track size.  A properly aligned multiple of
440  *   optimal_io_size is the preferred request size for workloads where
441  *   sustained throughput is desired.
442  */
443 void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
444 {
445 	limits->io_opt = opt;
446 }
447 EXPORT_SYMBOL(blk_limits_io_opt);
448 
449 /**
450  * blk_queue_io_opt - set optimal request size for the queue
451  * @q:	the request queue for the device
452  * @opt:  optimal request size in bytes
453  *
454  * Description:
455  *   Storage devices may report an optimal I/O size, which is the
456  *   device's preferred unit for sustained I/O.  This is rarely reported
457  *   for disk drives.  For RAID arrays it is usually the stripe width or
458  *   the internal track size.  A properly aligned multiple of
459  *   optimal_io_size is the preferred request size for workloads where
460  *   sustained throughput is desired.
461  */
462 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
463 {
464 	blk_limits_io_opt(&q->limits, opt);
465 }
466 EXPORT_SYMBOL(blk_queue_io_opt);
467 
468 /*
469  * Returns the minimum that is _not_ zero, unless both are zero.
470  */
471 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
472 
473 /**
474  * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
475  * @t:	the stacking driver (top)
476  * @b:  the underlying device (bottom)
477  **/
478 void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
479 {
480 	blk_stack_limits(&t->limits, &b->limits, 0);
481 
482 	if (!t->queue_lock)
483 		WARN_ON_ONCE(1);
484 	else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
485 		unsigned long flags;
486 		spin_lock_irqsave(t->queue_lock, flags);
487 		queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
488 		spin_unlock_irqrestore(t->queue_lock, flags);
489 	}
490 }
491 EXPORT_SYMBOL(blk_queue_stack_limits);
492 
493 /**
494  * blk_stack_limits - adjust queue_limits for stacked devices
495  * @t:	the stacking driver limits (top)
496  * @b:  the underlying queue limits (bottom)
497  * @offset:  offset to beginning of data within component device
498  *
499  * Description:
500  *    Merges two queue_limit structs.  Returns 0 if alignment didn't
501  *    change.  Returns -1 if adding the bottom device caused
502  *    misalignment.
503  */
504 int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
505 		     sector_t offset)
506 {
507 	t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
508 	t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
509 	t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
510 
511 	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
512 					    b->seg_boundary_mask);
513 
514 	t->max_phys_segments = min_not_zero(t->max_phys_segments,
515 					    b->max_phys_segments);
516 
517 	t->max_hw_segments = min_not_zero(t->max_hw_segments,
518 					  b->max_hw_segments);
519 
520 	t->max_segment_size = min_not_zero(t->max_segment_size,
521 					   b->max_segment_size);
522 
523 	t->logical_block_size = max(t->logical_block_size,
524 				    b->logical_block_size);
525 
526 	t->physical_block_size = max(t->physical_block_size,
527 				     b->physical_block_size);
528 
529 	t->io_min = max(t->io_min, b->io_min);
530 	t->no_cluster |= b->no_cluster;
531 
532 	/* Bottom device offset aligned? */
533 	if (offset &&
534 	    (offset & (b->physical_block_size - 1)) != b->alignment_offset) {
535 		t->misaligned = 1;
536 		return -1;
537 	}
538 
539 	/* If top has no alignment offset, inherit from bottom */
540 	if (!t->alignment_offset)
541 		t->alignment_offset =
542 			b->alignment_offset & (b->physical_block_size - 1);
543 
544 	/* Top device aligned on logical block boundary? */
545 	if (t->alignment_offset & (t->logical_block_size - 1)) {
546 		t->misaligned = 1;
547 		return -1;
548 	}
549 
550 	/* Find lcm() of optimal I/O size */
551 	if (t->io_opt && b->io_opt)
552 		t->io_opt = (t->io_opt * b->io_opt) / gcd(t->io_opt, b->io_opt);
553 	else if (b->io_opt)
554 		t->io_opt = b->io_opt;
555 
556 	/* Verify that optimal I/O size is a multiple of io_min */
557 	if (t->io_min && t->io_opt % t->io_min)
558 		return -1;
559 
560 	return 0;
561 }
562 EXPORT_SYMBOL(blk_stack_limits);
563 
564 /**
565  * disk_stack_limits - adjust queue limits for stacked drivers
566  * @disk:  MD/DM gendisk (top)
567  * @bdev:  the underlying block device (bottom)
568  * @offset:  offset to beginning of data within component device
569  *
570  * Description:
571  *    Merges the limits for two queues.  Returns 0 if alignment
572  *    didn't change.  Returns -1 if adding the bottom device caused
573  *    misalignment.
574  */
575 void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
576 		       sector_t offset)
577 {
578 	struct request_queue *t = disk->queue;
579 	struct request_queue *b = bdev_get_queue(bdev);
580 
581 	offset += get_start_sect(bdev) << 9;
582 
583 	if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) {
584 		char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
585 
586 		disk_name(disk, 0, top);
587 		bdevname(bdev, bottom);
588 
589 		printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
590 		       top, bottom);
591 	}
592 
593 	if (!t->queue_lock)
594 		WARN_ON_ONCE(1);
595 	else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
596 		unsigned long flags;
597 
598 		spin_lock_irqsave(t->queue_lock, flags);
599 		if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
600 			queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
601 		spin_unlock_irqrestore(t->queue_lock, flags);
602 	}
603 }
604 EXPORT_SYMBOL(disk_stack_limits);
605 
606 /**
607  * blk_queue_dma_pad - set pad mask
608  * @q:     the request queue for the device
609  * @mask:  pad mask
610  *
611  * Set dma pad mask.
612  *
613  * Appending pad buffer to a request modifies the last entry of a
614  * scatter list such that it includes the pad buffer.
615  **/
616 void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
617 {
618 	q->dma_pad_mask = mask;
619 }
620 EXPORT_SYMBOL(blk_queue_dma_pad);
621 
622 /**
623  * blk_queue_update_dma_pad - update pad mask
624  * @q:     the request queue for the device
625  * @mask:  pad mask
626  *
627  * Update dma pad mask.
628  *
629  * Appending pad buffer to a request modifies the last entry of a
630  * scatter list such that it includes the pad buffer.
631  **/
632 void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
633 {
634 	if (mask > q->dma_pad_mask)
635 		q->dma_pad_mask = mask;
636 }
637 EXPORT_SYMBOL(blk_queue_update_dma_pad);
638 
639 /**
640  * blk_queue_dma_drain - Set up a drain buffer for excess dma.
641  * @q:  the request queue for the device
642  * @dma_drain_needed: fn which returns non-zero if drain is necessary
643  * @buf:	physically contiguous buffer
644  * @size:	size of the buffer in bytes
645  *
646  * Some devices have excess DMA problems and can't simply discard (or
647  * zero fill) the unwanted piece of the transfer.  They have to have a
648  * real area of memory to transfer it into.  The use case for this is
649  * ATAPI devices in DMA mode.  If the packet command causes a transfer
650  * bigger than the transfer size some HBAs will lock up if there
651  * aren't DMA elements to contain the excess transfer.  What this API
652  * does is adjust the queue so that the buf is always appended
653  * silently to the scatterlist.
654  *
655  * Note: This routine adjusts max_hw_segments to make room for
656  * appending the drain buffer.  If you call
657  * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
658  * calling this routine, you must set the limit to one fewer than your
659  * device can support otherwise there won't be room for the drain
660  * buffer.
661  */
662 int blk_queue_dma_drain(struct request_queue *q,
663 			       dma_drain_needed_fn *dma_drain_needed,
664 			       void *buf, unsigned int size)
665 {
666 	if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2)
667 		return -EINVAL;
668 	/* make room for appending the drain */
669 	blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1);
670 	blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1);
671 	q->dma_drain_needed = dma_drain_needed;
672 	q->dma_drain_buffer = buf;
673 	q->dma_drain_size = size;
674 
675 	return 0;
676 }
677 EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
678 
679 /**
680  * blk_queue_segment_boundary - set boundary rules for segment merging
681  * @q:  the request queue for the device
682  * @mask:  the memory boundary mask
683  **/
684 void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
685 {
686 	if (mask < PAGE_CACHE_SIZE - 1) {
687 		mask = PAGE_CACHE_SIZE - 1;
688 		printk(KERN_INFO "%s: set to minimum %lx\n",
689 		       __func__, mask);
690 	}
691 
692 	q->limits.seg_boundary_mask = mask;
693 }
694 EXPORT_SYMBOL(blk_queue_segment_boundary);
695 
696 /**
697  * blk_queue_dma_alignment - set dma length and memory alignment
698  * @q:     the request queue for the device
699  * @mask:  alignment mask
700  *
701  * description:
702  *    set required memory and length alignment for direct dma transactions.
703  *    this is used when building direct io requests for the queue.
704  *
705  **/
706 void blk_queue_dma_alignment(struct request_queue *q, int mask)
707 {
708 	q->dma_alignment = mask;
709 }
710 EXPORT_SYMBOL(blk_queue_dma_alignment);
711 
712 /**
713  * blk_queue_update_dma_alignment - update dma length and memory alignment
714  * @q:     the request queue for the device
715  * @mask:  alignment mask
716  *
717  * description:
718  *    update required memory and length alignment for direct dma transactions.
719  *    If the requested alignment is larger than the current alignment, then
720  *    the current queue alignment is updated to the new value, otherwise it
721  *    is left alone.  The design of this is to allow multiple objects
722  *    (driver, device, transport etc) to set their respective
723  *    alignments without having them interfere.
724  *
725  **/
726 void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
727 {
728 	BUG_ON(mask > PAGE_SIZE);
729 
730 	if (mask > q->dma_alignment)
731 		q->dma_alignment = mask;
732 }
733 EXPORT_SYMBOL(blk_queue_update_dma_alignment);
734 
735 static int __init blk_settings_init(void)
736 {
737 	blk_max_low_pfn = max_low_pfn - 1;
738 	blk_max_pfn = max_pfn - 1;
739 	return 0;
740 }
741 subsys_initcall(blk_settings_init);
742