blob: d52d4adc440b29e8d428a9f05d59413d2bb421c9 [file] [log] [blame]
Jens Axboe86db1e22008-01-29 14:53:40 +01001/*
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 */
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -040010#include <linux/gcd.h>
Randy Dunlapad5ebd22009-11-11 13:47:45 +010011#include <linux/jiffies.h>
Jens Axboe86db1e22008-01-29 14:53:40 +010012
13#include "blk.h"
14
Jens Axboe6728cb02008-01-31 13:03:55 +010015unsigned long blk_max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010016EXPORT_SYMBOL(blk_max_low_pfn);
Jens Axboe6728cb02008-01-31 13:03:55 +010017
18unsigned long blk_max_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010019
20/**
21 * blk_queue_prep_rq - set a prepare_request function for queue
22 * @q: queue
23 * @pfn: prepare_request function
24 *
25 * It's possible for a queue to register a prepare_request callback which
26 * is invoked before the request is handed to the request_fn. The goal of
27 * the function is to prepare a request for I/O, it can be used to build a
28 * cdb from the request data for instance.
29 *
30 */
31void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
32{
33 q->prep_rq_fn = pfn;
34}
Jens Axboe86db1e22008-01-29 14:53:40 +010035EXPORT_SYMBOL(blk_queue_prep_rq);
36
37/**
38 * blk_queue_merge_bvec - set a merge_bvec function for queue
39 * @q: queue
40 * @mbfn: merge_bvec_fn
41 *
42 * Usually queues have static limitations on the max sectors or segments that
43 * we can put in a request. Stacking drivers may have some settings that
44 * are dynamic, and thus we have to query the queue whether it is ok to
45 * add a new bio_vec to a bio at a given offset or not. If the block device
46 * has such limitations, it needs to register a merge_bvec_fn to control
47 * the size of bio's sent to it. Note that a block device *must* allow a
48 * single page to be added to an empty bio. The block device driver may want
49 * to use the bio_split() function to deal with these bio's. By default
50 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
51 * honored.
52 */
53void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
54{
55 q->merge_bvec_fn = mbfn;
56}
Jens Axboe86db1e22008-01-29 14:53:40 +010057EXPORT_SYMBOL(blk_queue_merge_bvec);
58
59void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
60{
61 q->softirq_done_fn = fn;
62}
Jens Axboe86db1e22008-01-29 14:53:40 +010063EXPORT_SYMBOL(blk_queue_softirq_done);
64
Jens Axboe242f9dc2008-09-14 05:55:09 -070065void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
66{
67 q->rq_timeout = timeout;
68}
69EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
70
71void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
72{
73 q->rq_timed_out_fn = fn;
74}
75EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
76
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +020077void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
78{
79 q->lld_busy_fn = fn;
80}
81EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
82
Jens Axboe86db1e22008-01-29 14:53:40 +010083/**
Martin K. Petersene475bba2009-06-16 08:23:52 +020084 * blk_set_default_limits - reset limits to default values
Randy Dunlapf740f5ca2009-06-19 09:18:32 +020085 * @lim: the queue_limits structure to reset
Martin K. Petersene475bba2009-06-16 08:23:52 +020086 *
87 * Description:
88 * Returns a queue_limit struct to its default state. Can be used by
89 * stacking drivers like DM that stage table swaps and reuse an
90 * existing device queue.
91 */
92void blk_set_default_limits(struct queue_limits *lim)
93{
94 lim->max_phys_segments = MAX_PHYS_SEGMENTS;
95 lim->max_hw_segments = MAX_HW_SEGMENTS;
96 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
97 lim->max_segment_size = MAX_SEGMENT_SIZE;
Martin K. Petersen5dee2472009-09-21 21:46:05 +020098 lim->max_sectors = BLK_DEF_MAX_SECTORS;
99 lim->max_hw_sectors = INT_MAX;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100100 lim->max_discard_sectors = 0;
101 lim->discard_granularity = 0;
102 lim->discard_alignment = 0;
103 lim->discard_misaligned = 0;
Martin K. Petersen98262f22009-12-03 09:24:48 +0100104 lim->discard_zeroes_data = -1;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200105 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
Martin K. Petersen3a02c8e2009-06-18 09:56:03 +0200106 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200107 lim->alignment_offset = 0;
108 lim->io_opt = 0;
109 lim->misaligned = 0;
110 lim->no_cluster = 0;
111}
112EXPORT_SYMBOL(blk_set_default_limits);
113
114/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100115 * blk_queue_make_request - define an alternate make_request function for a device
116 * @q: the request queue for the device to be affected
117 * @mfn: the alternate make_request function
118 *
119 * Description:
120 * The normal way for &struct bios to be passed to a device
121 * driver is for them to be collected into requests on a request
122 * queue, and then to allow the device driver to select requests
123 * off that queue when it is ready. This works well for many block
124 * devices. However some block devices (typically virtual devices
125 * such as md or lvm) do not benefit from the processing on the
126 * request queue, and are served best by having the requests passed
127 * directly to them. This can be achieved by providing a function
128 * to blk_queue_make_request().
129 *
130 * Caveat:
131 * The driver that does this *must* be able to deal appropriately
132 * with buffers in "highmemory". This can be accomplished by either calling
133 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
134 * blk_queue_bounce() to create a buffer in normal memory.
135 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100136void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100137{
138 /*
139 * set defaults
140 */
141 q->nr_requests = BLKDEV_MAX_RQ;
Milan Broz0e435ac2008-12-03 12:55:08 +0100142
Jens Axboe86db1e22008-01-29 14:53:40 +0100143 q->make_request_fn = mfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100144 blk_queue_dma_alignment(q, 511);
145 blk_queue_congestion_threshold(q);
146 q->nr_batching = BLK_BATCH_REQ;
147
148 q->unplug_thresh = 4; /* hmm */
Randy Dunlapad5ebd22009-11-11 13:47:45 +0100149 q->unplug_delay = msecs_to_jiffies(3); /* 3 milliseconds */
Jens Axboe86db1e22008-01-29 14:53:40 +0100150 if (q->unplug_delay == 0)
151 q->unplug_delay = 1;
152
Jens Axboe86db1e22008-01-29 14:53:40 +0100153 q->unplug_timer.function = blk_unplug_timeout;
154 q->unplug_timer.data = (unsigned long)q;
155
Martin K. Petersene475bba2009-06-16 08:23:52 +0200156 blk_set_default_limits(&q->limits);
Martin K. Petersen80ddf242009-09-18 22:54:37 +0200157 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200158
Jens Axboe86db1e22008-01-29 14:53:40 +0100159 /*
Jens Axboea4e7d462009-07-28 09:07:29 +0200160 * If the caller didn't supply a lock, fall back to our embedded
161 * per-queue locks
162 */
163 if (!q->queue_lock)
164 q->queue_lock = &q->__queue_lock;
165
166 /*
Jens Axboe86db1e22008-01-29 14:53:40 +0100167 * by default assume old behaviour and bounce for any highmem page
168 */
169 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
170}
Jens Axboe86db1e22008-01-29 14:53:40 +0100171EXPORT_SYMBOL(blk_queue_make_request);
172
173/**
174 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900175 * @q: the request queue for the device
176 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100177 *
178 * Description:
179 * Different hardware can have different requirements as to what pages
180 * it can do I/O directly to. A low level driver can call
181 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Tejun Heocd0aca22009-04-15 22:10:25 +0900182 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100183 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900184void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100185{
Tejun Heocd0aca22009-04-15 22:10:25 +0900186 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100187 int dma = 0;
188
189 q->bounce_gfp = GFP_NOIO;
190#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900191 /*
192 * Assume anything <= 4GB can be handled by IOMMU. Actually
193 * some IOMMUs can handle everything, but I don't know of a
194 * way to test this here.
195 */
196 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100197 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400198 q->limits.bounce_pfn = max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100199#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100200 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100201 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400202 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100203#endif
204 if (dma) {
205 init_emergency_isa_pool();
206 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400207 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100208 }
209}
Jens Axboe86db1e22008-01-29 14:53:40 +0100210EXPORT_SYMBOL(blk_queue_bounce_limit);
211
212/**
213 * blk_queue_max_sectors - set max sectors for a request for this queue
214 * @q: the request queue for the device
215 * @max_sectors: max sectors in the usual 512b unit
216 *
217 * Description:
218 * Enables a low level driver to set an upper limit on the size of
219 * received requests.
220 **/
221void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
222{
223 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
224 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700225 printk(KERN_INFO "%s: set to minimum %d\n",
226 __func__, max_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100227 }
228
229 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400230 q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100231 else {
Martin K. Petersen025146e2009-05-22 17:17:51 -0400232 q->limits.max_sectors = BLK_DEF_MAX_SECTORS;
233 q->limits.max_hw_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100234 }
235}
Jens Axboe86db1e22008-01-29 14:53:40 +0100236EXPORT_SYMBOL(blk_queue_max_sectors);
237
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400238void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors)
239{
240 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400241 q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400242 else
Martin K. Petersen025146e2009-05-22 17:17:51 -0400243 q->limits.max_hw_sectors = max_sectors;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400244}
245EXPORT_SYMBOL(blk_queue_max_hw_sectors);
246
Jens Axboe86db1e22008-01-29 14:53:40 +0100247/**
Christoph Hellwig67efc922009-09-30 13:54:20 +0200248 * blk_queue_max_discard_sectors - set max sectors for a single discard
249 * @q: the request queue for the device
Randy Dunlapc7ebf062009-10-12 08:20:47 +0200250 * @max_discard_sectors: maximum number of sectors to discard
Christoph Hellwig67efc922009-09-30 13:54:20 +0200251 **/
252void blk_queue_max_discard_sectors(struct request_queue *q,
253 unsigned int max_discard_sectors)
254{
255 q->limits.max_discard_sectors = max_discard_sectors;
256}
257EXPORT_SYMBOL(blk_queue_max_discard_sectors);
258
259/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100260 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
261 * @q: the request queue for the device
262 * @max_segments: max number of segments
263 *
264 * Description:
265 * Enables a low level driver to set an upper limit on the number of
266 * physical data segments in a request. This would be the largest sized
267 * scatter list the driver could handle.
268 **/
269void blk_queue_max_phys_segments(struct request_queue *q,
270 unsigned short max_segments)
271{
272 if (!max_segments) {
273 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700274 printk(KERN_INFO "%s: set to minimum %d\n",
275 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100276 }
277
Martin K. Petersen025146e2009-05-22 17:17:51 -0400278 q->limits.max_phys_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100279}
Jens Axboe86db1e22008-01-29 14:53:40 +0100280EXPORT_SYMBOL(blk_queue_max_phys_segments);
281
282/**
283 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
284 * @q: the request queue for the device
285 * @max_segments: max number of segments
286 *
287 * Description:
288 * Enables a low level driver to set an upper limit on the number of
289 * hw data segments in a request. This would be the largest number of
Randy Dunlap710027a2008-08-19 20:13:11 +0200290 * address/length pairs the host adapter can actually give at once
Jens Axboe86db1e22008-01-29 14:53:40 +0100291 * to the device.
292 **/
293void blk_queue_max_hw_segments(struct request_queue *q,
294 unsigned short max_segments)
295{
296 if (!max_segments) {
297 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700298 printk(KERN_INFO "%s: set to minimum %d\n",
299 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100300 }
301
Martin K. Petersen025146e2009-05-22 17:17:51 -0400302 q->limits.max_hw_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100303}
Jens Axboe86db1e22008-01-29 14:53:40 +0100304EXPORT_SYMBOL(blk_queue_max_hw_segments);
305
306/**
307 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
308 * @q: the request queue for the device
309 * @max_size: max size of segment in bytes
310 *
311 * Description:
312 * Enables a low level driver to set an upper limit on the size of a
313 * coalesced segment
314 **/
315void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
316{
317 if (max_size < PAGE_CACHE_SIZE) {
318 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700319 printk(KERN_INFO "%s: set to minimum %d\n",
320 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100321 }
322
Martin K. Petersen025146e2009-05-22 17:17:51 -0400323 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100324}
Jens Axboe86db1e22008-01-29 14:53:40 +0100325EXPORT_SYMBOL(blk_queue_max_segment_size);
326
327/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400328 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100329 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400330 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100331 *
332 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400333 * This should be set to the lowest possible block size that the
334 * storage device can address. The default of 512 covers most
335 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100336 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400337void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100338{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400339 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400340
341 if (q->limits.physical_block_size < size)
342 q->limits.physical_block_size = size;
343
344 if (q->limits.io_min < q->limits.physical_block_size)
345 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100346}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400347EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100348
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400349/**
350 * blk_queue_physical_block_size - set physical block size for the queue
351 * @q: the request queue for the device
352 * @size: the physical block size, in bytes
353 *
354 * Description:
355 * This should be set to the lowest possible sector size that the
356 * hardware can operate on without reverting to read-modify-write
357 * operations.
358 */
359void blk_queue_physical_block_size(struct request_queue *q, unsigned short size)
360{
361 q->limits.physical_block_size = size;
362
363 if (q->limits.physical_block_size < q->limits.logical_block_size)
364 q->limits.physical_block_size = q->limits.logical_block_size;
365
366 if (q->limits.io_min < q->limits.physical_block_size)
367 q->limits.io_min = q->limits.physical_block_size;
368}
369EXPORT_SYMBOL(blk_queue_physical_block_size);
370
371/**
372 * blk_queue_alignment_offset - set physical block alignment offset
373 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700374 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400375 *
376 * Description:
377 * Some devices are naturally misaligned to compensate for things like
378 * the legacy DOS partition table 63-sector offset. Low-level drivers
379 * should call this function for devices whose first sector is not
380 * naturally aligned.
381 */
382void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
383{
384 q->limits.alignment_offset =
385 offset & (q->limits.physical_block_size - 1);
386 q->limits.misaligned = 0;
387}
388EXPORT_SYMBOL(blk_queue_alignment_offset);
389
390/**
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400391 * blk_limits_io_min - set minimum request size for a device
392 * @limits: the queue limits
393 * @min: smallest I/O size in bytes
394 *
395 * Description:
396 * Some devices have an internal block size bigger than the reported
397 * hardware sector size. This function can be used to signal the
398 * smallest I/O the device can perform without incurring a performance
399 * penalty.
400 */
401void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
402{
403 limits->io_min = min;
404
405 if (limits->io_min < limits->logical_block_size)
406 limits->io_min = limits->logical_block_size;
407
408 if (limits->io_min < limits->physical_block_size)
409 limits->io_min = limits->physical_block_size;
410}
411EXPORT_SYMBOL(blk_limits_io_min);
412
413/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400414 * blk_queue_io_min - set minimum request size for the queue
415 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700416 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400417 *
418 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400419 * Storage devices may report a granularity or preferred minimum I/O
420 * size which is the smallest request the device can perform without
421 * incurring a performance penalty. For disk drives this is often the
422 * physical block size. For RAID arrays it is often the stripe chunk
423 * size. A properly aligned multiple of minimum_io_size is the
424 * preferred request size for workloads where a high number of I/O
425 * operations is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400426 */
427void blk_queue_io_min(struct request_queue *q, unsigned int min)
428{
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400429 blk_limits_io_min(&q->limits, min);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400430}
431EXPORT_SYMBOL(blk_queue_io_min);
432
433/**
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200434 * blk_limits_io_opt - set optimal request size for a device
435 * @limits: the queue limits
436 * @opt: smallest I/O size in bytes
437 *
438 * Description:
439 * Storage devices may report an optimal I/O size, which is the
440 * device's preferred unit for sustained I/O. This is rarely reported
441 * for disk drives. For RAID arrays it is usually the stripe width or
442 * the internal track size. A properly aligned multiple of
443 * optimal_io_size is the preferred request size for workloads where
444 * sustained throughput is desired.
445 */
446void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
447{
448 limits->io_opt = opt;
449}
450EXPORT_SYMBOL(blk_limits_io_opt);
451
452/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400453 * blk_queue_io_opt - set optimal request size for the queue
454 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700455 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400456 *
457 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400458 * Storage devices may report an optimal I/O size, which is the
459 * device's preferred unit for sustained I/O. This is rarely reported
460 * for disk drives. For RAID arrays it is usually the stripe width or
461 * the internal track size. A properly aligned multiple of
462 * optimal_io_size is the preferred request size for workloads where
463 * sustained throughput is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400464 */
465void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
466{
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200467 blk_limits_io_opt(&q->limits, opt);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400468}
469EXPORT_SYMBOL(blk_queue_io_opt);
470
Jens Axboe86db1e22008-01-29 14:53:40 +0100471/*
472 * Returns the minimum that is _not_ zero, unless both are zero.
473 */
474#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
475
476/**
477 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
478 * @t: the stacking driver (top)
479 * @b: the underlying device (bottom)
480 **/
481void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
482{
Martin K. Petersenfef24662009-07-31 11:49:10 -0400483 blk_stack_limits(&t->limits, &b->limits, 0);
Martin K. Petersen025146e2009-05-22 17:17:51 -0400484
Neil Browne7e72bf2008-05-14 16:05:54 -0700485 if (!t->queue_lock)
486 WARN_ON_ONCE(1);
487 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
488 unsigned long flags;
489 spin_lock_irqsave(t->queue_lock, flags);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200490 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
Neil Browne7e72bf2008-05-14 16:05:54 -0700491 spin_unlock_irqrestore(t->queue_lock, flags);
492 }
Jens Axboe86db1e22008-01-29 14:53:40 +0100493}
Jens Axboe86db1e22008-01-29 14:53:40 +0100494EXPORT_SYMBOL(blk_queue_stack_limits);
495
Martin K. Petersen86b37282009-11-10 11:50:21 +0100496static unsigned int lcm(unsigned int a, unsigned int b)
497{
498 if (a && b)
499 return (a * b) / gcd(a, b);
500 else if (b)
501 return b;
502
503 return a;
504}
505
Jens Axboe86db1e22008-01-29 14:53:40 +0100506/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400507 * blk_stack_limits - adjust queue_limits for stacked devices
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100508 * @t: the stacking driver limits (top device)
509 * @b: the underlying queue limits (bottom, component device)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400510 * @offset: offset to beginning of data within component device
511 *
512 * Description:
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100513 * This function is used by stacking drivers like MD and DM to ensure
514 * that all component devices have compatible block sizes and
515 * alignments. The stacking driver must provide a queue_limits
516 * struct (top) and then iteratively call the stacking function for
517 * all component (bottom) devices. The stacking function will
518 * attempt to combine the values and ensure proper alignment.
519 *
520 * Returns 0 if the top and bottom queue_limits are compatible. The
521 * top device's block sizes and alignment offsets may be adjusted to
522 * ensure alignment with the bottom device. If no compatible sizes
523 * and alignments exist, -1 is returned and the resulting top
524 * queue_limits will have the misaligned flag set to indicate that
525 * the alignment_offset is undefined.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400526 */
527int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
528 sector_t offset)
529{
Martin K. Petersen9504e082009-12-21 15:55:51 +0100530 sector_t alignment;
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100531 unsigned int top, bottom;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100532
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400533 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
534 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200535 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400536
537 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
538 b->seg_boundary_mask);
539
540 t->max_phys_segments = min_not_zero(t->max_phys_segments,
541 b->max_phys_segments);
542
543 t->max_hw_segments = min_not_zero(t->max_hw_segments,
544 b->max_hw_segments);
545
546 t->max_segment_size = min_not_zero(t->max_segment_size,
547 b->max_segment_size);
548
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100549 alignment = queue_limit_alignment_offset(b, offset);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100550
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100551 /* Bottom device has different alignment. Check that it is
552 * compatible with the current top alignment.
553 */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100554 if (t->alignment_offset != alignment) {
555
556 top = max(t->physical_block_size, t->io_min)
557 + t->alignment_offset;
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100558 bottom = max(b->physical_block_size, b->io_min) + alignment;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100559
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100560 /* Verify that top and bottom intervals line up */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100561 if (max(top, bottom) & (min(top, bottom) - 1))
562 t->misaligned = 1;
563 }
564
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400565 t->logical_block_size = max(t->logical_block_size,
566 b->logical_block_size);
567
568 t->physical_block_size = max(t->physical_block_size,
569 b->physical_block_size);
570
571 t->io_min = max(t->io_min, b->io_min);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100572 t->io_opt = lcm(t->io_opt, b->io_opt);
573
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400574 t->no_cluster |= b->no_cluster;
Martin K. Petersen98262f22009-12-03 09:24:48 +0100575 t->discard_zeroes_data &= b->discard_zeroes_data;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400576
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100577 /* Physical block size a multiple of the logical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100578 if (t->physical_block_size & (t->logical_block_size - 1)) {
579 t->physical_block_size = t->logical_block_size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400580 t->misaligned = 1;
Martin K. Petersen86b37282009-11-10 11:50:21 +0100581 }
582
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100583 /* Minimum I/O a multiple of the physical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100584 if (t->io_min & (t->physical_block_size - 1)) {
585 t->io_min = t->physical_block_size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400586 t->misaligned = 1;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400587 }
588
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100589 /* Optimal I/O a multiple of the physical block size? */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100590 if (t->io_opt & (t->physical_block_size - 1)) {
591 t->io_opt = 0;
592 t->misaligned = 1;
593 }
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400594
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100595 /* Find lowest common alignment_offset */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100596 t->alignment_offset = lcm(t->alignment_offset, alignment)
597 & (max(t->physical_block_size, t->io_min) - 1);
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400598
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100599 /* Verify that new alignment_offset is on a logical block boundary */
Martin K. Petersen9504e082009-12-21 15:55:51 +0100600 if (t->alignment_offset & (t->logical_block_size - 1))
601 t->misaligned = 1;
602
603 /* Discard alignment and granularity */
604 if (b->discard_granularity) {
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100605 unsigned int granularity = b->discard_granularity;
606 offset &= granularity - 1;
Martin K. Petersen9504e082009-12-21 15:55:51 +0100607
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100608 alignment = (granularity + b->discard_alignment - offset)
609 & (granularity - 1);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100610
611 if (t->discard_granularity != 0 &&
612 t->discard_alignment != alignment) {
613 top = t->discard_granularity + t->discard_alignment;
614 bottom = b->discard_granularity + alignment;
615
616 /* Verify that top and bottom intervals line up */
617 if (max(top, bottom) & (min(top, bottom) - 1))
618 t->discard_misaligned = 1;
619 }
620
Martin K. Petersen81744ee2009-12-29 08:35:35 +0100621 t->max_discard_sectors = min_not_zero(t->max_discard_sectors,
622 b->max_discard_sectors);
Martin K. Petersen9504e082009-12-21 15:55:51 +0100623 t->discard_granularity = max(t->discard_granularity,
624 b->discard_granularity);
625 t->discard_alignment = lcm(t->discard_alignment, alignment) &
626 (t->discard_granularity - 1);
627 }
628
629 return t->misaligned ? -1 : 0;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400630}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200631EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400632
633/**
634 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200635 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400636 * @bdev: the underlying block device (bottom)
637 * @offset: offset to beginning of data within component device
638 *
639 * Description:
640 * Merges the limits for two queues. Returns 0 if alignment
641 * didn't change. Returns -1 if adding the bottom device caused
642 * misalignment.
643 */
644void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
645 sector_t offset)
646{
647 struct request_queue *t = disk->queue;
648 struct request_queue *b = bdev_get_queue(bdev);
649
650 offset += get_start_sect(bdev) << 9;
651
652 if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) {
653 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
654
655 disk_name(disk, 0, top);
656 bdevname(bdev, bottom);
657
658 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
659 top, bottom);
660 }
661
662 if (!t->queue_lock)
663 WARN_ON_ONCE(1);
664 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
665 unsigned long flags;
666
667 spin_lock_irqsave(t->queue_lock, flags);
668 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
669 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
670 spin_unlock_irqrestore(t->queue_lock, flags);
671 }
672}
673EXPORT_SYMBOL(disk_stack_limits);
674
675/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100676 * blk_queue_dma_pad - set pad mask
677 * @q: the request queue for the device
678 * @mask: pad mask
679 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200680 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100681 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200682 * Appending pad buffer to a request modifies the last entry of a
683 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100684 **/
685void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
686{
687 q->dma_pad_mask = mask;
688}
689EXPORT_SYMBOL(blk_queue_dma_pad);
690
691/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200692 * blk_queue_update_dma_pad - update pad mask
693 * @q: the request queue for the device
694 * @mask: pad mask
695 *
696 * Update dma pad mask.
697 *
698 * Appending pad buffer to a request modifies the last entry of a
699 * scatter list such that it includes the pad buffer.
700 **/
701void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
702{
703 if (mask > q->dma_pad_mask)
704 q->dma_pad_mask = mask;
705}
706EXPORT_SYMBOL(blk_queue_update_dma_pad);
707
708/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100709 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100710 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100711 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100712 * @buf: physically contiguous buffer
713 * @size: size of the buffer in bytes
714 *
715 * Some devices have excess DMA problems and can't simply discard (or
716 * zero fill) the unwanted piece of the transfer. They have to have a
717 * real area of memory to transfer it into. The use case for this is
718 * ATAPI devices in DMA mode. If the packet command causes a transfer
719 * bigger than the transfer size some HBAs will lock up if there
720 * aren't DMA elements to contain the excess transfer. What this API
721 * does is adjust the queue so that the buf is always appended
722 * silently to the scatterlist.
723 *
724 * Note: This routine adjusts max_hw_segments to make room for
725 * appending the drain buffer. If you call
726 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
727 * calling this routine, you must set the limit to one fewer than your
728 * device can support otherwise there won't be room for the drain
729 * buffer.
730 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100731int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100732 dma_drain_needed_fn *dma_drain_needed,
733 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100734{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400735 if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100736 return -EINVAL;
737 /* make room for appending the drain */
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400738 blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1);
739 blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100740 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100741 q->dma_drain_buffer = buf;
742 q->dma_drain_size = size;
743
744 return 0;
745}
Jens Axboe86db1e22008-01-29 14:53:40 +0100746EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
747
748/**
749 * blk_queue_segment_boundary - set boundary rules for segment merging
750 * @q: the request queue for the device
751 * @mask: the memory boundary mask
752 **/
753void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
754{
755 if (mask < PAGE_CACHE_SIZE - 1) {
756 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700757 printk(KERN_INFO "%s: set to minimum %lx\n",
758 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100759 }
760
Martin K. Petersen025146e2009-05-22 17:17:51 -0400761 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100762}
Jens Axboe86db1e22008-01-29 14:53:40 +0100763EXPORT_SYMBOL(blk_queue_segment_boundary);
764
765/**
766 * blk_queue_dma_alignment - set dma length and memory alignment
767 * @q: the request queue for the device
768 * @mask: alignment mask
769 *
770 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200771 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100772 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100773 *
774 **/
775void blk_queue_dma_alignment(struct request_queue *q, int mask)
776{
777 q->dma_alignment = mask;
778}
Jens Axboe86db1e22008-01-29 14:53:40 +0100779EXPORT_SYMBOL(blk_queue_dma_alignment);
780
781/**
782 * blk_queue_update_dma_alignment - update dma length and memory alignment
783 * @q: the request queue for the device
784 * @mask: alignment mask
785 *
786 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200787 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100788 * If the requested alignment is larger than the current alignment, then
789 * the current queue alignment is updated to the new value, otherwise it
790 * is left alone. The design of this is to allow multiple objects
791 * (driver, device, transport etc) to set their respective
792 * alignments without having them interfere.
793 *
794 **/
795void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
796{
797 BUG_ON(mask > PAGE_SIZE);
798
799 if (mask > q->dma_alignment)
800 q->dma_alignment = mask;
801}
Jens Axboe86db1e22008-01-29 14:53:40 +0100802EXPORT_SYMBOL(blk_queue_update_dma_alignment);
803
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200804static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100805{
806 blk_max_low_pfn = max_low_pfn - 1;
807 blk_max_pfn = max_pfn - 1;
808 return 0;
809}
810subsys_initcall(blk_settings_init);