blob: 66d4aa8799b7517de258e3e334a69372154580a9 [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>
Jens Axboe86db1e22008-01-29 14:53:40 +010011
12#include "blk.h"
13
Jens Axboe6728cb02008-01-31 13:03:55 +010014unsigned long blk_max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010015EXPORT_SYMBOL(blk_max_low_pfn);
Jens Axboe6728cb02008-01-31 13:03:55 +010016
17unsigned long blk_max_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +010018
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 */
30void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
31{
32 q->prep_rq_fn = pfn;
33}
Jens Axboe86db1e22008-01-29 14:53:40 +010034EXPORT_SYMBOL(blk_queue_prep_rq);
35
36/**
37 * blk_queue_merge_bvec - set a merge_bvec function for queue
38 * @q: queue
39 * @mbfn: merge_bvec_fn
40 *
41 * Usually queues have static limitations on the max sectors or segments that
42 * we can put in a request. Stacking drivers may have some settings that
43 * are dynamic, and thus we have to query the queue whether it is ok to
44 * add a new bio_vec to a bio at a given offset or not. If the block device
45 * has such limitations, it needs to register a merge_bvec_fn to control
46 * the size of bio's sent to it. Note that a block device *must* allow a
47 * single page to be added to an empty bio. The block device driver may want
48 * to use the bio_split() function to deal with these bio's. By default
49 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
50 * honored.
51 */
52void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
53{
54 q->merge_bvec_fn = mbfn;
55}
Jens Axboe86db1e22008-01-29 14:53:40 +010056EXPORT_SYMBOL(blk_queue_merge_bvec);
57
58void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
59{
60 q->softirq_done_fn = fn;
61}
Jens Axboe86db1e22008-01-29 14:53:40 +010062EXPORT_SYMBOL(blk_queue_softirq_done);
63
Jens Axboe242f9dc2008-09-14 05:55:09 -070064void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout)
65{
66 q->rq_timeout = timeout;
67}
68EXPORT_SYMBOL_GPL(blk_queue_rq_timeout);
69
70void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn)
71{
72 q->rq_timed_out_fn = fn;
73}
74EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out);
75
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +020076void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn)
77{
78 q->lld_busy_fn = fn;
79}
80EXPORT_SYMBOL_GPL(blk_queue_lld_busy);
81
Jens Axboe86db1e22008-01-29 14:53:40 +010082/**
Martin K. Petersene475bba2009-06-16 08:23:52 +020083 * blk_set_default_limits - reset limits to default values
Randy Dunlapf740f5ca2009-06-19 09:18:32 +020084 * @lim: the queue_limits structure to reset
Martin K. Petersene475bba2009-06-16 08:23:52 +020085 *
86 * Description:
87 * Returns a queue_limit struct to its default state. Can be used by
88 * stacking drivers like DM that stage table swaps and reuse an
89 * existing device queue.
90 */
91void blk_set_default_limits(struct queue_limits *lim)
92{
93 lim->max_phys_segments = MAX_PHYS_SEGMENTS;
94 lim->max_hw_segments = MAX_HW_SEGMENTS;
95 lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
96 lim->max_segment_size = MAX_SEGMENT_SIZE;
Martin K. Petersen5dee2472009-09-21 21:46:05 +020097 lim->max_sectors = BLK_DEF_MAX_SECTORS;
98 lim->max_hw_sectors = INT_MAX;
Christoph Hellwig67efc922009-09-30 13:54:20 +020099 lim->max_discard_sectors = SAFE_MAX_SECTORS;
Martin K. Petersene475bba2009-06-16 08:23:52 +0200100 lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
Martin K. Petersen3a02c8e2009-06-18 09:56:03 +0200101 lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200102 lim->alignment_offset = 0;
103 lim->io_opt = 0;
104 lim->misaligned = 0;
105 lim->no_cluster = 0;
106}
107EXPORT_SYMBOL(blk_set_default_limits);
108
109/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100110 * blk_queue_make_request - define an alternate make_request function for a device
111 * @q: the request queue for the device to be affected
112 * @mfn: the alternate make_request function
113 *
114 * Description:
115 * The normal way for &struct bios to be passed to a device
116 * driver is for them to be collected into requests on a request
117 * queue, and then to allow the device driver to select requests
118 * off that queue when it is ready. This works well for many block
119 * devices. However some block devices (typically virtual devices
120 * such as md or lvm) do not benefit from the processing on the
121 * request queue, and are served best by having the requests passed
122 * directly to them. This can be achieved by providing a function
123 * to blk_queue_make_request().
124 *
125 * Caveat:
126 * The driver that does this *must* be able to deal appropriately
127 * with buffers in "highmemory". This can be accomplished by either calling
128 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
129 * blk_queue_bounce() to create a buffer in normal memory.
130 **/
Jens Axboe6728cb02008-01-31 13:03:55 +0100131void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100132{
133 /*
134 * set defaults
135 */
136 q->nr_requests = BLKDEV_MAX_RQ;
Milan Broz0e435ac2008-12-03 12:55:08 +0100137
Jens Axboe86db1e22008-01-29 14:53:40 +0100138 q->make_request_fn = mfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100139 blk_queue_dma_alignment(q, 511);
140 blk_queue_congestion_threshold(q);
141 q->nr_batching = BLK_BATCH_REQ;
142
143 q->unplug_thresh = 4; /* hmm */
144 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
145 if (q->unplug_delay == 0)
146 q->unplug_delay = 1;
147
Jens Axboe86db1e22008-01-29 14:53:40 +0100148 q->unplug_timer.function = blk_unplug_timeout;
149 q->unplug_timer.data = (unsigned long)q;
150
Martin K. Petersene475bba2009-06-16 08:23:52 +0200151 blk_set_default_limits(&q->limits);
Martin K. Petersen80ddf242009-09-18 22:54:37 +0200152 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Martin K. Petersene475bba2009-06-16 08:23:52 +0200153
Jens Axboe86db1e22008-01-29 14:53:40 +0100154 /*
Jens Axboea4e7d462009-07-28 09:07:29 +0200155 * If the caller didn't supply a lock, fall back to our embedded
156 * per-queue locks
157 */
158 if (!q->queue_lock)
159 q->queue_lock = &q->__queue_lock;
160
161 /*
Jens Axboe86db1e22008-01-29 14:53:40 +0100162 * by default assume old behaviour and bounce for any highmem page
163 */
164 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
165}
Jens Axboe86db1e22008-01-29 14:53:40 +0100166EXPORT_SYMBOL(blk_queue_make_request);
167
168/**
169 * blk_queue_bounce_limit - set bounce buffer limit for queue
Tejun Heocd0aca22009-04-15 22:10:25 +0900170 * @q: the request queue for the device
171 * @dma_mask: the maximum address the device can handle
Jens Axboe86db1e22008-01-29 14:53:40 +0100172 *
173 * Description:
174 * Different hardware can have different requirements as to what pages
175 * it can do I/O directly to. A low level driver can call
176 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Tejun Heocd0aca22009-04-15 22:10:25 +0900177 * buffers for doing I/O to pages residing above @dma_mask.
Jens Axboe86db1e22008-01-29 14:53:40 +0100178 **/
Tejun Heocd0aca22009-04-15 22:10:25 +0900179void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask)
Jens Axboe86db1e22008-01-29 14:53:40 +0100180{
Tejun Heocd0aca22009-04-15 22:10:25 +0900181 unsigned long b_pfn = dma_mask >> PAGE_SHIFT;
Jens Axboe86db1e22008-01-29 14:53:40 +0100182 int dma = 0;
183
184 q->bounce_gfp = GFP_NOIO;
185#if BITS_PER_LONG == 64
Tejun Heocd0aca22009-04-15 22:10:25 +0900186 /*
187 * Assume anything <= 4GB can be handled by IOMMU. Actually
188 * some IOMMUs can handle everything, but I don't know of a
189 * way to test this here.
190 */
191 if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Jens Axboe86db1e22008-01-29 14:53:40 +0100192 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400193 q->limits.bounce_pfn = max_low_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100194#else
Jens Axboe6728cb02008-01-31 13:03:55 +0100195 if (b_pfn < blk_max_low_pfn)
Jens Axboe86db1e22008-01-29 14:53:40 +0100196 dma = 1;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400197 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100198#endif
199 if (dma) {
200 init_emergency_isa_pool();
201 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Martin K. Petersen025146e2009-05-22 17:17:51 -0400202 q->limits.bounce_pfn = b_pfn;
Jens Axboe86db1e22008-01-29 14:53:40 +0100203 }
204}
Jens Axboe86db1e22008-01-29 14:53:40 +0100205EXPORT_SYMBOL(blk_queue_bounce_limit);
206
207/**
208 * blk_queue_max_sectors - set max sectors for a request for this queue
209 * @q: the request queue for the device
210 * @max_sectors: max sectors in the usual 512b unit
211 *
212 * Description:
213 * Enables a low level driver to set an upper limit on the size of
214 * received requests.
215 **/
216void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
217{
218 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
219 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
Harvey Harrison24c03d42008-05-01 04:35:17 -0700220 printk(KERN_INFO "%s: set to minimum %d\n",
221 __func__, max_sectors);
Jens Axboe86db1e22008-01-29 14:53:40 +0100222 }
223
224 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400225 q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100226 else {
Martin K. Petersen025146e2009-05-22 17:17:51 -0400227 q->limits.max_sectors = BLK_DEF_MAX_SECTORS;
228 q->limits.max_hw_sectors = max_sectors;
Jens Axboe86db1e22008-01-29 14:53:40 +0100229 }
230}
Jens Axboe86db1e22008-01-29 14:53:40 +0100231EXPORT_SYMBOL(blk_queue_max_sectors);
232
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400233void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors)
234{
235 if (BLK_DEF_MAX_SECTORS > max_sectors)
Martin K. Petersen025146e2009-05-22 17:17:51 -0400236 q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400237 else
Martin K. Petersen025146e2009-05-22 17:17:51 -0400238 q->limits.max_hw_sectors = max_sectors;
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400239}
240EXPORT_SYMBOL(blk_queue_max_hw_sectors);
241
Jens Axboe86db1e22008-01-29 14:53:40 +0100242/**
Christoph Hellwig67efc922009-09-30 13:54:20 +0200243 * blk_queue_max_discard_sectors - set max sectors for a single discard
244 * @q: the request queue for the device
Randy Dunlapc7ebf062009-10-12 08:20:47 +0200245 * @max_discard_sectors: maximum number of sectors to discard
Christoph Hellwig67efc922009-09-30 13:54:20 +0200246 **/
247void blk_queue_max_discard_sectors(struct request_queue *q,
248 unsigned int max_discard_sectors)
249{
250 q->limits.max_discard_sectors = max_discard_sectors;
251}
252EXPORT_SYMBOL(blk_queue_max_discard_sectors);
253
254/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100255 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
256 * @q: the request queue for the device
257 * @max_segments: max number of segments
258 *
259 * Description:
260 * Enables a low level driver to set an upper limit on the number of
261 * physical data segments in a request. This would be the largest sized
262 * scatter list the driver could handle.
263 **/
264void blk_queue_max_phys_segments(struct request_queue *q,
265 unsigned short max_segments)
266{
267 if (!max_segments) {
268 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700269 printk(KERN_INFO "%s: set to minimum %d\n",
270 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100271 }
272
Martin K. Petersen025146e2009-05-22 17:17:51 -0400273 q->limits.max_phys_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100274}
Jens Axboe86db1e22008-01-29 14:53:40 +0100275EXPORT_SYMBOL(blk_queue_max_phys_segments);
276
277/**
278 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
279 * @q: the request queue for the device
280 * @max_segments: max number of segments
281 *
282 * Description:
283 * Enables a low level driver to set an upper limit on the number of
284 * hw data segments in a request. This would be the largest number of
Randy Dunlap710027a2008-08-19 20:13:11 +0200285 * address/length pairs the host adapter can actually give at once
Jens Axboe86db1e22008-01-29 14:53:40 +0100286 * to the device.
287 **/
288void blk_queue_max_hw_segments(struct request_queue *q,
289 unsigned short max_segments)
290{
291 if (!max_segments) {
292 max_segments = 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700293 printk(KERN_INFO "%s: set to minimum %d\n",
294 __func__, max_segments);
Jens Axboe86db1e22008-01-29 14:53:40 +0100295 }
296
Martin K. Petersen025146e2009-05-22 17:17:51 -0400297 q->limits.max_hw_segments = max_segments;
Jens Axboe86db1e22008-01-29 14:53:40 +0100298}
Jens Axboe86db1e22008-01-29 14:53:40 +0100299EXPORT_SYMBOL(blk_queue_max_hw_segments);
300
301/**
302 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
303 * @q: the request queue for the device
304 * @max_size: max size of segment in bytes
305 *
306 * Description:
307 * Enables a low level driver to set an upper limit on the size of a
308 * coalesced segment
309 **/
310void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
311{
312 if (max_size < PAGE_CACHE_SIZE) {
313 max_size = PAGE_CACHE_SIZE;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700314 printk(KERN_INFO "%s: set to minimum %d\n",
315 __func__, max_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100316 }
317
Martin K. Petersen025146e2009-05-22 17:17:51 -0400318 q->limits.max_segment_size = max_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100319}
Jens Axboe86db1e22008-01-29 14:53:40 +0100320EXPORT_SYMBOL(blk_queue_max_segment_size);
321
322/**
Martin K. Petersene1defc42009-05-22 17:17:49 -0400323 * blk_queue_logical_block_size - set logical block size for the queue
Jens Axboe86db1e22008-01-29 14:53:40 +0100324 * @q: the request queue for the device
Martin K. Petersene1defc42009-05-22 17:17:49 -0400325 * @size: the logical block size, in bytes
Jens Axboe86db1e22008-01-29 14:53:40 +0100326 *
327 * Description:
Martin K. Petersene1defc42009-05-22 17:17:49 -0400328 * This should be set to the lowest possible block size that the
329 * storage device can address. The default of 512 covers most
330 * hardware.
Jens Axboe86db1e22008-01-29 14:53:40 +0100331 **/
Martin K. Petersene1defc42009-05-22 17:17:49 -0400332void blk_queue_logical_block_size(struct request_queue *q, unsigned short size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100333{
Martin K. Petersen025146e2009-05-22 17:17:51 -0400334 q->limits.logical_block_size = size;
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400335
336 if (q->limits.physical_block_size < size)
337 q->limits.physical_block_size = size;
338
339 if (q->limits.io_min < q->limits.physical_block_size)
340 q->limits.io_min = q->limits.physical_block_size;
Jens Axboe86db1e22008-01-29 14:53:40 +0100341}
Martin K. Petersene1defc42009-05-22 17:17:49 -0400342EXPORT_SYMBOL(blk_queue_logical_block_size);
Jens Axboe86db1e22008-01-29 14:53:40 +0100343
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400344/**
345 * blk_queue_physical_block_size - set physical block size for the queue
346 * @q: the request queue for the device
347 * @size: the physical block size, in bytes
348 *
349 * Description:
350 * This should be set to the lowest possible sector size that the
351 * hardware can operate on without reverting to read-modify-write
352 * operations.
353 */
354void blk_queue_physical_block_size(struct request_queue *q, unsigned short size)
355{
356 q->limits.physical_block_size = size;
357
358 if (q->limits.physical_block_size < q->limits.logical_block_size)
359 q->limits.physical_block_size = q->limits.logical_block_size;
360
361 if (q->limits.io_min < q->limits.physical_block_size)
362 q->limits.io_min = q->limits.physical_block_size;
363}
364EXPORT_SYMBOL(blk_queue_physical_block_size);
365
366/**
367 * blk_queue_alignment_offset - set physical block alignment offset
368 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700369 * @offset: alignment offset in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400370 *
371 * Description:
372 * Some devices are naturally misaligned to compensate for things like
373 * the legacy DOS partition table 63-sector offset. Low-level drivers
374 * should call this function for devices whose first sector is not
375 * naturally aligned.
376 */
377void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset)
378{
379 q->limits.alignment_offset =
380 offset & (q->limits.physical_block_size - 1);
381 q->limits.misaligned = 0;
382}
383EXPORT_SYMBOL(blk_queue_alignment_offset);
384
385/**
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400386 * blk_limits_io_min - set minimum request size for a device
387 * @limits: the queue limits
388 * @min: smallest I/O size in bytes
389 *
390 * Description:
391 * Some devices have an internal block size bigger than the reported
392 * hardware sector size. This function can be used to signal the
393 * smallest I/O the device can perform without incurring a performance
394 * penalty.
395 */
396void blk_limits_io_min(struct queue_limits *limits, unsigned int min)
397{
398 limits->io_min = min;
399
400 if (limits->io_min < limits->logical_block_size)
401 limits->io_min = limits->logical_block_size;
402
403 if (limits->io_min < limits->physical_block_size)
404 limits->io_min = limits->physical_block_size;
405}
406EXPORT_SYMBOL(blk_limits_io_min);
407
408/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400409 * blk_queue_io_min - set minimum request size for the queue
410 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700411 * @min: smallest I/O size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400412 *
413 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400414 * Storage devices may report a granularity or preferred minimum I/O
415 * size which is the smallest request the device can perform without
416 * incurring a performance penalty. For disk drives this is often the
417 * physical block size. For RAID arrays it is often the stripe chunk
418 * size. A properly aligned multiple of minimum_io_size is the
419 * preferred request size for workloads where a high number of I/O
420 * operations is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400421 */
422void blk_queue_io_min(struct request_queue *q, unsigned int min)
423{
Martin K. Petersen7c958e32009-07-31 11:49:11 -0400424 blk_limits_io_min(&q->limits, min);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400425}
426EXPORT_SYMBOL(blk_queue_io_min);
427
428/**
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200429 * blk_limits_io_opt - set optimal request size for a device
430 * @limits: the queue limits
431 * @opt: smallest I/O size in bytes
432 *
433 * Description:
434 * Storage devices may report an optimal I/O size, which is the
435 * device's preferred unit for sustained I/O. This is rarely reported
436 * for disk drives. For RAID arrays it is usually the stripe width or
437 * the internal track size. A properly aligned multiple of
438 * optimal_io_size is the preferred request size for workloads where
439 * sustained throughput is desired.
440 */
441void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt)
442{
443 limits->io_opt = opt;
444}
445EXPORT_SYMBOL(blk_limits_io_opt);
446
447/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400448 * blk_queue_io_opt - set optimal request size for the queue
449 * @q: the request queue for the device
Randy Dunlap8ebf9752009-06-11 20:00:41 -0700450 * @opt: optimal request size in bytes
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400451 *
452 * Description:
Martin K. Petersen7e5f5fb2009-07-31 11:49:13 -0400453 * Storage devices may report an optimal I/O size, which is the
454 * device's preferred unit for sustained I/O. This is rarely reported
455 * for disk drives. For RAID arrays it is usually the stripe width or
456 * the internal track size. A properly aligned multiple of
457 * optimal_io_size is the preferred request size for workloads where
458 * sustained throughput is desired.
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400459 */
460void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
461{
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200462 blk_limits_io_opt(&q->limits, opt);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400463}
464EXPORT_SYMBOL(blk_queue_io_opt);
465
Jens Axboe86db1e22008-01-29 14:53:40 +0100466/*
467 * Returns the minimum that is _not_ zero, unless both are zero.
468 */
469#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
470
471/**
472 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
473 * @t: the stacking driver (top)
474 * @b: the underlying device (bottom)
475 **/
476void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
477{
Martin K. Petersenfef24662009-07-31 11:49:10 -0400478 blk_stack_limits(&t->limits, &b->limits, 0);
Martin K. Petersen025146e2009-05-22 17:17:51 -0400479
Neil Browne7e72bf2008-05-14 16:05:54 -0700480 if (!t->queue_lock)
481 WARN_ON_ONCE(1);
482 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
483 unsigned long flags;
484 spin_lock_irqsave(t->queue_lock, flags);
Nick Piggin75ad23b2008-04-29 14:48:33 +0200485 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
Neil Browne7e72bf2008-05-14 16:05:54 -0700486 spin_unlock_irqrestore(t->queue_lock, flags);
487 }
Jens Axboe86db1e22008-01-29 14:53:40 +0100488}
Jens Axboe86db1e22008-01-29 14:53:40 +0100489EXPORT_SYMBOL(blk_queue_stack_limits);
490
491/**
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400492 * blk_stack_limits - adjust queue_limits for stacked devices
493 * @t: the stacking driver limits (top)
Martin K. Petersen77634f32009-06-09 06:23:22 +0200494 * @b: the underlying queue limits (bottom)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400495 * @offset: offset to beginning of data within component device
496 *
497 * Description:
498 * Merges two queue_limit structs. Returns 0 if alignment didn't
499 * change. Returns -1 if adding the bottom device caused
500 * misalignment.
501 */
502int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
503 sector_t offset)
504{
505 t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors);
506 t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors);
Martin K. Petersen77634f32009-06-09 06:23:22 +0200507 t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400508
509 t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
510 b->seg_boundary_mask);
511
512 t->max_phys_segments = min_not_zero(t->max_phys_segments,
513 b->max_phys_segments);
514
515 t->max_hw_segments = min_not_zero(t->max_hw_segments,
516 b->max_hw_segments);
517
518 t->max_segment_size = min_not_zero(t->max_segment_size,
519 b->max_segment_size);
520
521 t->logical_block_size = max(t->logical_block_size,
522 b->logical_block_size);
523
524 t->physical_block_size = max(t->physical_block_size,
525 b->physical_block_size);
526
527 t->io_min = max(t->io_min, b->io_min);
528 t->no_cluster |= b->no_cluster;
529
530 /* Bottom device offset aligned? */
531 if (offset &&
532 (offset & (b->physical_block_size - 1)) != b->alignment_offset) {
533 t->misaligned = 1;
534 return -1;
535 }
536
537 /* If top has no alignment offset, inherit from bottom */
538 if (!t->alignment_offset)
539 t->alignment_offset =
540 b->alignment_offset & (b->physical_block_size - 1);
541
542 /* Top device aligned on logical block boundary? */
543 if (t->alignment_offset & (t->logical_block_size - 1)) {
544 t->misaligned = 1;
545 return -1;
546 }
547
Martin K. Petersen70dd5bf2009-07-31 11:49:12 -0400548 /* Find lcm() of optimal I/O size */
549 if (t->io_opt && b->io_opt)
550 t->io_opt = (t->io_opt * b->io_opt) / gcd(t->io_opt, b->io_opt);
551 else if (b->io_opt)
552 t->io_opt = b->io_opt;
553
554 /* Verify that optimal I/O size is a multiple of io_min */
555 if (t->io_min && t->io_opt % t->io_min)
556 return -1;
557
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400558 return 0;
559}
Mike Snitzer5d85d322009-05-28 11:04:53 +0200560EXPORT_SYMBOL(blk_stack_limits);
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400561
562/**
563 * disk_stack_limits - adjust queue limits for stacked drivers
Martin K. Petersen77634f32009-06-09 06:23:22 +0200564 * @disk: MD/DM gendisk (top)
Martin K. Petersenc72758f2009-05-22 17:17:53 -0400565 * @bdev: the underlying block device (bottom)
566 * @offset: offset to beginning of data within component device
567 *
568 * Description:
569 * Merges the limits for two queues. Returns 0 if alignment
570 * didn't change. Returns -1 if adding the bottom device caused
571 * misalignment.
572 */
573void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
574 sector_t offset)
575{
576 struct request_queue *t = disk->queue;
577 struct request_queue *b = bdev_get_queue(bdev);
578
579 offset += get_start_sect(bdev) << 9;
580
581 if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) {
582 char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE];
583
584 disk_name(disk, 0, top);
585 bdevname(bdev, bottom);
586
587 printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n",
588 top, bottom);
589 }
590
591 if (!t->queue_lock)
592 WARN_ON_ONCE(1);
593 else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) {
594 unsigned long flags;
595
596 spin_lock_irqsave(t->queue_lock, flags);
597 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
598 queue_flag_clear(QUEUE_FLAG_CLUSTER, t);
599 spin_unlock_irqrestore(t->queue_lock, flags);
600 }
601}
602EXPORT_SYMBOL(disk_stack_limits);
603
604/**
Tejun Heoe3790c72008-03-04 11:18:17 +0100605 * blk_queue_dma_pad - set pad mask
606 * @q: the request queue for the device
607 * @mask: pad mask
608 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200609 * Set dma pad mask.
Tejun Heoe3790c72008-03-04 11:18:17 +0100610 *
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200611 * Appending pad buffer to a request modifies the last entry of a
612 * scatter list such that it includes the pad buffer.
Tejun Heoe3790c72008-03-04 11:18:17 +0100613 **/
614void blk_queue_dma_pad(struct request_queue *q, unsigned int mask)
615{
616 q->dma_pad_mask = mask;
617}
618EXPORT_SYMBOL(blk_queue_dma_pad);
619
620/**
FUJITA Tomonori27f82212008-07-04 09:30:03 +0200621 * blk_queue_update_dma_pad - update pad mask
622 * @q: the request queue for the device
623 * @mask: pad mask
624 *
625 * Update dma pad mask.
626 *
627 * Appending pad buffer to a request modifies the last entry of a
628 * scatter list such that it includes the pad buffer.
629 **/
630void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask)
631{
632 if (mask > q->dma_pad_mask)
633 q->dma_pad_mask = mask;
634}
635EXPORT_SYMBOL(blk_queue_update_dma_pad);
636
637/**
Jens Axboe86db1e22008-01-29 14:53:40 +0100638 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
Jens Axboe86db1e22008-01-29 14:53:40 +0100639 * @q: the request queue for the device
Tejun Heo2fb98e82008-02-19 11:36:53 +0100640 * @dma_drain_needed: fn which returns non-zero if drain is necessary
Jens Axboe86db1e22008-01-29 14:53:40 +0100641 * @buf: physically contiguous buffer
642 * @size: size of the buffer in bytes
643 *
644 * Some devices have excess DMA problems and can't simply discard (or
645 * zero fill) the unwanted piece of the transfer. They have to have a
646 * real area of memory to transfer it into. The use case for this is
647 * ATAPI devices in DMA mode. If the packet command causes a transfer
648 * bigger than the transfer size some HBAs will lock up if there
649 * aren't DMA elements to contain the excess transfer. What this API
650 * does is adjust the queue so that the buf is always appended
651 * silently to the scatterlist.
652 *
653 * Note: This routine adjusts max_hw_segments to make room for
654 * appending the drain buffer. If you call
655 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
656 * calling this routine, you must set the limit to one fewer than your
657 * device can support otherwise there won't be room for the drain
658 * buffer.
659 */
Harvey Harrison448da4d2008-03-04 11:30:18 +0100660int blk_queue_dma_drain(struct request_queue *q,
Tejun Heo2fb98e82008-02-19 11:36:53 +0100661 dma_drain_needed_fn *dma_drain_needed,
662 void *buf, unsigned int size)
Jens Axboe86db1e22008-01-29 14:53:40 +0100663{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400664 if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2)
Jens Axboe86db1e22008-01-29 14:53:40 +0100665 return -EINVAL;
666 /* make room for appending the drain */
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400667 blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1);
668 blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1);
Tejun Heo2fb98e82008-02-19 11:36:53 +0100669 q->dma_drain_needed = dma_drain_needed;
Jens Axboe86db1e22008-01-29 14:53:40 +0100670 q->dma_drain_buffer = buf;
671 q->dma_drain_size = size;
672
673 return 0;
674}
Jens Axboe86db1e22008-01-29 14:53:40 +0100675EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
676
677/**
678 * blk_queue_segment_boundary - set boundary rules for segment merging
679 * @q: the request queue for the device
680 * @mask: the memory boundary mask
681 **/
682void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
683{
684 if (mask < PAGE_CACHE_SIZE - 1) {
685 mask = PAGE_CACHE_SIZE - 1;
Harvey Harrison24c03d42008-05-01 04:35:17 -0700686 printk(KERN_INFO "%s: set to minimum %lx\n",
687 __func__, mask);
Jens Axboe86db1e22008-01-29 14:53:40 +0100688 }
689
Martin K. Petersen025146e2009-05-22 17:17:51 -0400690 q->limits.seg_boundary_mask = mask;
Jens Axboe86db1e22008-01-29 14:53:40 +0100691}
Jens Axboe86db1e22008-01-29 14:53:40 +0100692EXPORT_SYMBOL(blk_queue_segment_boundary);
693
694/**
695 * blk_queue_dma_alignment - set dma length and memory alignment
696 * @q: the request queue for the device
697 * @mask: alignment mask
698 *
699 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200700 * set required memory and length alignment for direct dma transactions.
Alan Cox8feb4d22009-04-01 15:01:39 +0100701 * this is used when building direct io requests for the queue.
Jens Axboe86db1e22008-01-29 14:53:40 +0100702 *
703 **/
704void blk_queue_dma_alignment(struct request_queue *q, int mask)
705{
706 q->dma_alignment = mask;
707}
Jens Axboe86db1e22008-01-29 14:53:40 +0100708EXPORT_SYMBOL(blk_queue_dma_alignment);
709
710/**
711 * blk_queue_update_dma_alignment - update dma length and memory alignment
712 * @q: the request queue for the device
713 * @mask: alignment mask
714 *
715 * description:
Randy Dunlap710027a2008-08-19 20:13:11 +0200716 * update required memory and length alignment for direct dma transactions.
Jens Axboe86db1e22008-01-29 14:53:40 +0100717 * If the requested alignment is larger than the current alignment, then
718 * the current queue alignment is updated to the new value, otherwise it
719 * is left alone. The design of this is to allow multiple objects
720 * (driver, device, transport etc) to set their respective
721 * alignments without having them interfere.
722 *
723 **/
724void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
725{
726 BUG_ON(mask > PAGE_SIZE);
727
728 if (mask > q->dma_alignment)
729 q->dma_alignment = mask;
730}
Jens Axboe86db1e22008-01-29 14:53:40 +0100731EXPORT_SYMBOL(blk_queue_update_dma_alignment);
732
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200733static int __init blk_settings_init(void)
Jens Axboe86db1e22008-01-29 14:53:40 +0100734{
735 blk_max_low_pfn = max_low_pfn - 1;
736 blk_max_pfn = max_pfn - 1;
737 return 0;
738}
739subsys_initcall(blk_settings_init);