Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 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 | |
| 11 | #include "blk.h" |
| 12 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 13 | unsigned long blk_max_low_pfn; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 14 | EXPORT_SYMBOL(blk_max_low_pfn); |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 15 | |
| 16 | unsigned long blk_max_pfn; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 17 | |
| 18 | /** |
| 19 | * blk_queue_prep_rq - set a prepare_request function for queue |
| 20 | * @q: queue |
| 21 | * @pfn: prepare_request function |
| 22 | * |
| 23 | * It's possible for a queue to register a prepare_request callback which |
| 24 | * is invoked before the request is handed to the request_fn. The goal of |
| 25 | * the function is to prepare a request for I/O, it can be used to build a |
| 26 | * cdb from the request data for instance. |
| 27 | * |
| 28 | */ |
| 29 | void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) |
| 30 | { |
| 31 | q->prep_rq_fn = pfn; |
| 32 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 33 | EXPORT_SYMBOL(blk_queue_prep_rq); |
| 34 | |
| 35 | /** |
David Woodhouse | fb2dce8 | 2008-08-05 18:01:53 +0100 | [diff] [blame] | 36 | * blk_queue_set_discard - set a discard_sectors function for queue |
| 37 | * @q: queue |
| 38 | * @dfn: prepare_discard function |
| 39 | * |
| 40 | * It's possible for a queue to register a discard callback which is used |
| 41 | * to transform a discard request into the appropriate type for the |
| 42 | * hardware. If none is registered, then discard requests are failed |
| 43 | * with %EOPNOTSUPP. |
| 44 | * |
| 45 | */ |
| 46 | void blk_queue_set_discard(struct request_queue *q, prepare_discard_fn *dfn) |
| 47 | { |
| 48 | q->prepare_discard_fn = dfn; |
| 49 | } |
| 50 | EXPORT_SYMBOL(blk_queue_set_discard); |
| 51 | |
| 52 | /** |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 53 | * blk_queue_merge_bvec - set a merge_bvec function for queue |
| 54 | * @q: queue |
| 55 | * @mbfn: merge_bvec_fn |
| 56 | * |
| 57 | * Usually queues have static limitations on the max sectors or segments that |
| 58 | * we can put in a request. Stacking drivers may have some settings that |
| 59 | * are dynamic, and thus we have to query the queue whether it is ok to |
| 60 | * add a new bio_vec to a bio at a given offset or not. If the block device |
| 61 | * has such limitations, it needs to register a merge_bvec_fn to control |
| 62 | * the size of bio's sent to it. Note that a block device *must* allow a |
| 63 | * single page to be added to an empty bio. The block device driver may want |
| 64 | * to use the bio_split() function to deal with these bio's. By default |
| 65 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are |
| 66 | * honored. |
| 67 | */ |
| 68 | void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) |
| 69 | { |
| 70 | q->merge_bvec_fn = mbfn; |
| 71 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 72 | EXPORT_SYMBOL(blk_queue_merge_bvec); |
| 73 | |
| 74 | void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) |
| 75 | { |
| 76 | q->softirq_done_fn = fn; |
| 77 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 78 | EXPORT_SYMBOL(blk_queue_softirq_done); |
| 79 | |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 80 | void blk_queue_rq_timeout(struct request_queue *q, unsigned int timeout) |
| 81 | { |
| 82 | q->rq_timeout = timeout; |
| 83 | } |
| 84 | EXPORT_SYMBOL_GPL(blk_queue_rq_timeout); |
| 85 | |
| 86 | void blk_queue_rq_timed_out(struct request_queue *q, rq_timed_out_fn *fn) |
| 87 | { |
| 88 | q->rq_timed_out_fn = fn; |
| 89 | } |
| 90 | EXPORT_SYMBOL_GPL(blk_queue_rq_timed_out); |
| 91 | |
Kiyoshi Ueda | ef9e3fa | 2008-10-01 16:12:15 +0200 | [diff] [blame] | 92 | void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn) |
| 93 | { |
| 94 | q->lld_busy_fn = fn; |
| 95 | } |
| 96 | EXPORT_SYMBOL_GPL(blk_queue_lld_busy); |
| 97 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 98 | /** |
| 99 | * blk_queue_make_request - define an alternate make_request function for a device |
| 100 | * @q: the request queue for the device to be affected |
| 101 | * @mfn: the alternate make_request function |
| 102 | * |
| 103 | * Description: |
| 104 | * The normal way for &struct bios to be passed to a device |
| 105 | * driver is for them to be collected into requests on a request |
| 106 | * queue, and then to allow the device driver to select requests |
| 107 | * off that queue when it is ready. This works well for many block |
| 108 | * devices. However some block devices (typically virtual devices |
| 109 | * such as md or lvm) do not benefit from the processing on the |
| 110 | * request queue, and are served best by having the requests passed |
| 111 | * directly to them. This can be achieved by providing a function |
| 112 | * to blk_queue_make_request(). |
| 113 | * |
| 114 | * Caveat: |
| 115 | * The driver that does this *must* be able to deal appropriately |
| 116 | * with buffers in "highmemory". This can be accomplished by either calling |
| 117 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling |
| 118 | * blk_queue_bounce() to create a buffer in normal memory. |
| 119 | **/ |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 120 | void blk_queue_make_request(struct request_queue *q, make_request_fn *mfn) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 121 | { |
| 122 | /* |
| 123 | * set defaults |
| 124 | */ |
| 125 | q->nr_requests = BLKDEV_MAX_RQ; |
| 126 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); |
| 127 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); |
Milan Broz | 0e435ac | 2008-12-03 12:55:08 +0100 | [diff] [blame] | 128 | blk_queue_segment_boundary(q, BLK_SEG_BOUNDARY_MASK); |
| 129 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); |
| 130 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 131 | q->make_request_fn = mfn; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 132 | blk_queue_max_sectors(q, SAFE_MAX_SECTORS); |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 133 | blk_queue_logical_block_size(q, 512); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 134 | blk_queue_dma_alignment(q, 511); |
| 135 | blk_queue_congestion_threshold(q); |
| 136 | q->nr_batching = BLK_BATCH_REQ; |
| 137 | |
| 138 | q->unplug_thresh = 4; /* hmm */ |
| 139 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ |
| 140 | if (q->unplug_delay == 0) |
| 141 | q->unplug_delay = 1; |
| 142 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 143 | q->unplug_timer.function = blk_unplug_timeout; |
| 144 | q->unplug_timer.data = (unsigned long)q; |
| 145 | |
| 146 | /* |
| 147 | * by default assume old behaviour and bounce for any highmem page |
| 148 | */ |
| 149 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); |
| 150 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 151 | EXPORT_SYMBOL(blk_queue_make_request); |
| 152 | |
| 153 | /** |
| 154 | * blk_queue_bounce_limit - set bounce buffer limit for queue |
Tejun Heo | cd0aca2 | 2009-04-15 22:10:25 +0900 | [diff] [blame] | 155 | * @q: the request queue for the device |
| 156 | * @dma_mask: the maximum address the device can handle |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 157 | * |
| 158 | * Description: |
| 159 | * Different hardware can have different requirements as to what pages |
| 160 | * it can do I/O directly to. A low level driver can call |
| 161 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce |
Tejun Heo | cd0aca2 | 2009-04-15 22:10:25 +0900 | [diff] [blame] | 162 | * buffers for doing I/O to pages residing above @dma_mask. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 163 | **/ |
Tejun Heo | cd0aca2 | 2009-04-15 22:10:25 +0900 | [diff] [blame] | 164 | void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 165 | { |
Tejun Heo | cd0aca2 | 2009-04-15 22:10:25 +0900 | [diff] [blame] | 166 | unsigned long b_pfn = dma_mask >> PAGE_SHIFT; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 167 | int dma = 0; |
| 168 | |
| 169 | q->bounce_gfp = GFP_NOIO; |
| 170 | #if BITS_PER_LONG == 64 |
Tejun Heo | cd0aca2 | 2009-04-15 22:10:25 +0900 | [diff] [blame] | 171 | /* |
| 172 | * Assume anything <= 4GB can be handled by IOMMU. Actually |
| 173 | * some IOMMUs can handle everything, but I don't know of a |
| 174 | * way to test this here. |
| 175 | */ |
| 176 | if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 177 | dma = 1; |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 178 | q->limits.bounce_pfn = max_low_pfn; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 179 | #else |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 180 | if (b_pfn < blk_max_low_pfn) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 181 | dma = 1; |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 182 | q->limits.bounce_pfn = b_pfn; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 183 | #endif |
| 184 | if (dma) { |
| 185 | init_emergency_isa_pool(); |
| 186 | q->bounce_gfp = GFP_NOIO | GFP_DMA; |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 187 | q->limits.bounce_pfn = b_pfn; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 188 | } |
| 189 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 190 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
| 191 | |
| 192 | /** |
| 193 | * blk_queue_max_sectors - set max sectors for a request for this queue |
| 194 | * @q: the request queue for the device |
| 195 | * @max_sectors: max sectors in the usual 512b unit |
| 196 | * |
| 197 | * Description: |
| 198 | * Enables a low level driver to set an upper limit on the size of |
| 199 | * received requests. |
| 200 | **/ |
| 201 | void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors) |
| 202 | { |
| 203 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { |
| 204 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 205 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 206 | __func__, max_sectors); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 207 | } |
| 208 | |
| 209 | if (BLK_DEF_MAX_SECTORS > max_sectors) |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 210 | q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 211 | else { |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 212 | q->limits.max_sectors = BLK_DEF_MAX_SECTORS; |
| 213 | q->limits.max_hw_sectors = max_sectors; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 214 | } |
| 215 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 216 | EXPORT_SYMBOL(blk_queue_max_sectors); |
| 217 | |
Martin K. Petersen | ae03bf6 | 2009-05-22 17:17:50 -0400 | [diff] [blame] | 218 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors) |
| 219 | { |
| 220 | if (BLK_DEF_MAX_SECTORS > max_sectors) |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 221 | q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS; |
Martin K. Petersen | ae03bf6 | 2009-05-22 17:17:50 -0400 | [diff] [blame] | 222 | else |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 223 | q->limits.max_hw_sectors = max_sectors; |
Martin K. Petersen | ae03bf6 | 2009-05-22 17:17:50 -0400 | [diff] [blame] | 224 | } |
| 225 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); |
| 226 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 227 | /** |
| 228 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue |
| 229 | * @q: the request queue for the device |
| 230 | * @max_segments: max number of segments |
| 231 | * |
| 232 | * Description: |
| 233 | * Enables a low level driver to set an upper limit on the number of |
| 234 | * physical data segments in a request. This would be the largest sized |
| 235 | * scatter list the driver could handle. |
| 236 | **/ |
| 237 | void blk_queue_max_phys_segments(struct request_queue *q, |
| 238 | unsigned short max_segments) |
| 239 | { |
| 240 | if (!max_segments) { |
| 241 | max_segments = 1; |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 242 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 243 | __func__, max_segments); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 244 | } |
| 245 | |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 246 | q->limits.max_phys_segments = max_segments; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 247 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 248 | EXPORT_SYMBOL(blk_queue_max_phys_segments); |
| 249 | |
| 250 | /** |
| 251 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue |
| 252 | * @q: the request queue for the device |
| 253 | * @max_segments: max number of segments |
| 254 | * |
| 255 | * Description: |
| 256 | * Enables a low level driver to set an upper limit on the number of |
| 257 | * hw data segments in a request. This would be the largest number of |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 258 | * address/length pairs the host adapter can actually give at once |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 259 | * to the device. |
| 260 | **/ |
| 261 | void blk_queue_max_hw_segments(struct request_queue *q, |
| 262 | unsigned short max_segments) |
| 263 | { |
| 264 | if (!max_segments) { |
| 265 | max_segments = 1; |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 266 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 267 | __func__, max_segments); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 268 | } |
| 269 | |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 270 | q->limits.max_hw_segments = max_segments; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 271 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 272 | EXPORT_SYMBOL(blk_queue_max_hw_segments); |
| 273 | |
| 274 | /** |
| 275 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg |
| 276 | * @q: the request queue for the device |
| 277 | * @max_size: max size of segment in bytes |
| 278 | * |
| 279 | * Description: |
| 280 | * Enables a low level driver to set an upper limit on the size of a |
| 281 | * coalesced segment |
| 282 | **/ |
| 283 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) |
| 284 | { |
| 285 | if (max_size < PAGE_CACHE_SIZE) { |
| 286 | max_size = PAGE_CACHE_SIZE; |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 287 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 288 | __func__, max_size); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 289 | } |
| 290 | |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 291 | q->limits.max_segment_size = max_size; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 292 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 293 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
| 294 | |
| 295 | /** |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 296 | * blk_queue_logical_block_size - set logical block size for the queue |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 297 | * @q: the request queue for the device |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 298 | * @size: the logical block size, in bytes |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 299 | * |
| 300 | * Description: |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 301 | * This should be set to the lowest possible block size that the |
| 302 | * storage device can address. The default of 512 covers most |
| 303 | * hardware. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 304 | **/ |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 305 | void blk_queue_logical_block_size(struct request_queue *q, unsigned short size) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 306 | { |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 307 | q->limits.logical_block_size = size; |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 308 | |
| 309 | if (q->limits.physical_block_size < size) |
| 310 | q->limits.physical_block_size = size; |
| 311 | |
| 312 | if (q->limits.io_min < q->limits.physical_block_size) |
| 313 | q->limits.io_min = q->limits.physical_block_size; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 314 | } |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 315 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 316 | |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 317 | /** |
| 318 | * blk_queue_physical_block_size - set physical block size for the queue |
| 319 | * @q: the request queue for the device |
| 320 | * @size: the physical block size, in bytes |
| 321 | * |
| 322 | * Description: |
| 323 | * This should be set to the lowest possible sector size that the |
| 324 | * hardware can operate on without reverting to read-modify-write |
| 325 | * operations. |
| 326 | */ |
| 327 | void blk_queue_physical_block_size(struct request_queue *q, unsigned short size) |
| 328 | { |
| 329 | q->limits.physical_block_size = size; |
| 330 | |
| 331 | if (q->limits.physical_block_size < q->limits.logical_block_size) |
| 332 | q->limits.physical_block_size = q->limits.logical_block_size; |
| 333 | |
| 334 | if (q->limits.io_min < q->limits.physical_block_size) |
| 335 | q->limits.io_min = q->limits.physical_block_size; |
| 336 | } |
| 337 | EXPORT_SYMBOL(blk_queue_physical_block_size); |
| 338 | |
| 339 | /** |
| 340 | * blk_queue_alignment_offset - set physical block alignment offset |
| 341 | * @q: the request queue for the device |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 342 | * @offset: alignment offset in bytes |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 343 | * |
| 344 | * Description: |
| 345 | * Some devices are naturally misaligned to compensate for things like |
| 346 | * the legacy DOS partition table 63-sector offset. Low-level drivers |
| 347 | * should call this function for devices whose first sector is not |
| 348 | * naturally aligned. |
| 349 | */ |
| 350 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) |
| 351 | { |
| 352 | q->limits.alignment_offset = |
| 353 | offset & (q->limits.physical_block_size - 1); |
| 354 | q->limits.misaligned = 0; |
| 355 | } |
| 356 | EXPORT_SYMBOL(blk_queue_alignment_offset); |
| 357 | |
| 358 | /** |
| 359 | * blk_queue_io_min - set minimum request size for the queue |
| 360 | * @q: the request queue for the device |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 361 | * @min: smallest I/O size in bytes |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 362 | * |
| 363 | * Description: |
| 364 | * Some devices have an internal block size bigger than the reported |
| 365 | * hardware sector size. This function can be used to signal the |
| 366 | * smallest I/O the device can perform without incurring a performance |
| 367 | * penalty. |
| 368 | */ |
| 369 | void blk_queue_io_min(struct request_queue *q, unsigned int min) |
| 370 | { |
| 371 | q->limits.io_min = min; |
| 372 | |
| 373 | if (q->limits.io_min < q->limits.logical_block_size) |
| 374 | q->limits.io_min = q->limits.logical_block_size; |
| 375 | |
| 376 | if (q->limits.io_min < q->limits.physical_block_size) |
| 377 | q->limits.io_min = q->limits.physical_block_size; |
| 378 | } |
| 379 | EXPORT_SYMBOL(blk_queue_io_min); |
| 380 | |
| 381 | /** |
| 382 | * blk_queue_io_opt - set optimal request size for the queue |
| 383 | * @q: the request queue for the device |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 384 | * @opt: optimal request size in bytes |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 385 | * |
| 386 | * Description: |
| 387 | * Drivers can call this function to set the preferred I/O request |
| 388 | * size for devices that report such a value. |
| 389 | */ |
| 390 | void blk_queue_io_opt(struct request_queue *q, unsigned int opt) |
| 391 | { |
| 392 | q->limits.io_opt = opt; |
| 393 | } |
| 394 | EXPORT_SYMBOL(blk_queue_io_opt); |
| 395 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 396 | /* |
| 397 | * Returns the minimum that is _not_ zero, unless both are zero. |
| 398 | */ |
| 399 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) |
| 400 | |
| 401 | /** |
| 402 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers |
| 403 | * @t: the stacking driver (top) |
| 404 | * @b: the underlying device (bottom) |
| 405 | **/ |
| 406 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) |
| 407 | { |
| 408 | /* zero is "infinity" */ |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 409 | t->limits.max_sectors = min_not_zero(queue_max_sectors(t), |
| 410 | queue_max_sectors(b)); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 411 | |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 412 | t->limits.max_hw_sectors = min_not_zero(queue_max_hw_sectors(t), |
| 413 | queue_max_hw_sectors(b)); |
| 414 | |
| 415 | t->limits.seg_boundary_mask = min_not_zero(queue_segment_boundary(t), |
| 416 | queue_segment_boundary(b)); |
| 417 | |
| 418 | t->limits.max_phys_segments = min_not_zero(queue_max_phys_segments(t), |
| 419 | queue_max_phys_segments(b)); |
| 420 | |
| 421 | t->limits.max_hw_segments = min_not_zero(queue_max_hw_segments(t), |
| 422 | queue_max_hw_segments(b)); |
| 423 | |
| 424 | t->limits.max_segment_size = min_not_zero(queue_max_segment_size(t), |
| 425 | queue_max_segment_size(b)); |
| 426 | |
| 427 | t->limits.logical_block_size = max(queue_logical_block_size(t), |
| 428 | queue_logical_block_size(b)); |
| 429 | |
Neil Brown | e7e72bf | 2008-05-14 16:05:54 -0700 | [diff] [blame] | 430 | if (!t->queue_lock) |
| 431 | WARN_ON_ONCE(1); |
| 432 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { |
| 433 | unsigned long flags; |
| 434 | spin_lock_irqsave(t->queue_lock, flags); |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 435 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); |
Neil Brown | e7e72bf | 2008-05-14 16:05:54 -0700 | [diff] [blame] | 436 | spin_unlock_irqrestore(t->queue_lock, flags); |
| 437 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 438 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 439 | EXPORT_SYMBOL(blk_queue_stack_limits); |
| 440 | |
| 441 | /** |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 442 | * blk_stack_limits - adjust queue_limits for stacked devices |
| 443 | * @t: the stacking driver limits (top) |
Martin K. Petersen | 77634f3 | 2009-06-09 06:23:22 +0200 | [diff] [blame] | 444 | * @b: the underlying queue limits (bottom) |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 445 | * @offset: offset to beginning of data within component device |
| 446 | * |
| 447 | * Description: |
| 448 | * Merges two queue_limit structs. Returns 0 if alignment didn't |
| 449 | * change. Returns -1 if adding the bottom device caused |
| 450 | * misalignment. |
| 451 | */ |
| 452 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, |
| 453 | sector_t offset) |
| 454 | { |
| 455 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
| 456 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); |
Martin K. Petersen | 77634f3 | 2009-06-09 06:23:22 +0200 | [diff] [blame] | 457 | t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn); |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 458 | |
| 459 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, |
| 460 | b->seg_boundary_mask); |
| 461 | |
| 462 | t->max_phys_segments = min_not_zero(t->max_phys_segments, |
| 463 | b->max_phys_segments); |
| 464 | |
| 465 | t->max_hw_segments = min_not_zero(t->max_hw_segments, |
| 466 | b->max_hw_segments); |
| 467 | |
| 468 | t->max_segment_size = min_not_zero(t->max_segment_size, |
| 469 | b->max_segment_size); |
| 470 | |
| 471 | t->logical_block_size = max(t->logical_block_size, |
| 472 | b->logical_block_size); |
| 473 | |
| 474 | t->physical_block_size = max(t->physical_block_size, |
| 475 | b->physical_block_size); |
| 476 | |
| 477 | t->io_min = max(t->io_min, b->io_min); |
| 478 | t->no_cluster |= b->no_cluster; |
| 479 | |
| 480 | /* Bottom device offset aligned? */ |
| 481 | if (offset && |
| 482 | (offset & (b->physical_block_size - 1)) != b->alignment_offset) { |
| 483 | t->misaligned = 1; |
| 484 | return -1; |
| 485 | } |
| 486 | |
| 487 | /* If top has no alignment offset, inherit from bottom */ |
| 488 | if (!t->alignment_offset) |
| 489 | t->alignment_offset = |
| 490 | b->alignment_offset & (b->physical_block_size - 1); |
| 491 | |
| 492 | /* Top device aligned on logical block boundary? */ |
| 493 | if (t->alignment_offset & (t->logical_block_size - 1)) { |
| 494 | t->misaligned = 1; |
| 495 | return -1; |
| 496 | } |
| 497 | |
| 498 | return 0; |
| 499 | } |
Mike Snitzer | 5d85d32 | 2009-05-28 11:04:53 +0200 | [diff] [blame] | 500 | EXPORT_SYMBOL(blk_stack_limits); |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 501 | |
| 502 | /** |
| 503 | * disk_stack_limits - adjust queue limits for stacked drivers |
Martin K. Petersen | 77634f3 | 2009-06-09 06:23:22 +0200 | [diff] [blame] | 504 | * @disk: MD/DM gendisk (top) |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 505 | * @bdev: the underlying block device (bottom) |
| 506 | * @offset: offset to beginning of data within component device |
| 507 | * |
| 508 | * Description: |
| 509 | * Merges the limits for two queues. Returns 0 if alignment |
| 510 | * didn't change. Returns -1 if adding the bottom device caused |
| 511 | * misalignment. |
| 512 | */ |
| 513 | void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, |
| 514 | sector_t offset) |
| 515 | { |
| 516 | struct request_queue *t = disk->queue; |
| 517 | struct request_queue *b = bdev_get_queue(bdev); |
| 518 | |
| 519 | offset += get_start_sect(bdev) << 9; |
| 520 | |
| 521 | if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) { |
| 522 | char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; |
| 523 | |
| 524 | disk_name(disk, 0, top); |
| 525 | bdevname(bdev, bottom); |
| 526 | |
| 527 | printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", |
| 528 | top, bottom); |
| 529 | } |
| 530 | |
| 531 | if (!t->queue_lock) |
| 532 | WARN_ON_ONCE(1); |
| 533 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { |
| 534 | unsigned long flags; |
| 535 | |
| 536 | spin_lock_irqsave(t->queue_lock, flags); |
| 537 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) |
| 538 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); |
| 539 | spin_unlock_irqrestore(t->queue_lock, flags); |
| 540 | } |
| 541 | } |
| 542 | EXPORT_SYMBOL(disk_stack_limits); |
| 543 | |
| 544 | /** |
Tejun Heo | e3790c7 | 2008-03-04 11:18:17 +0100 | [diff] [blame] | 545 | * blk_queue_dma_pad - set pad mask |
| 546 | * @q: the request queue for the device |
| 547 | * @mask: pad mask |
| 548 | * |
FUJITA Tomonori | 27f8221 | 2008-07-04 09:30:03 +0200 | [diff] [blame] | 549 | * Set dma pad mask. |
Tejun Heo | e3790c7 | 2008-03-04 11:18:17 +0100 | [diff] [blame] | 550 | * |
FUJITA Tomonori | 27f8221 | 2008-07-04 09:30:03 +0200 | [diff] [blame] | 551 | * Appending pad buffer to a request modifies the last entry of a |
| 552 | * scatter list such that it includes the pad buffer. |
Tejun Heo | e3790c7 | 2008-03-04 11:18:17 +0100 | [diff] [blame] | 553 | **/ |
| 554 | void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) |
| 555 | { |
| 556 | q->dma_pad_mask = mask; |
| 557 | } |
| 558 | EXPORT_SYMBOL(blk_queue_dma_pad); |
| 559 | |
| 560 | /** |
FUJITA Tomonori | 27f8221 | 2008-07-04 09:30:03 +0200 | [diff] [blame] | 561 | * blk_queue_update_dma_pad - update pad mask |
| 562 | * @q: the request queue for the device |
| 563 | * @mask: pad mask |
| 564 | * |
| 565 | * Update dma pad mask. |
| 566 | * |
| 567 | * Appending pad buffer to a request modifies the last entry of a |
| 568 | * scatter list such that it includes the pad buffer. |
| 569 | **/ |
| 570 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) |
| 571 | { |
| 572 | if (mask > q->dma_pad_mask) |
| 573 | q->dma_pad_mask = mask; |
| 574 | } |
| 575 | EXPORT_SYMBOL(blk_queue_update_dma_pad); |
| 576 | |
| 577 | /** |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 578 | * blk_queue_dma_drain - Set up a drain buffer for excess dma. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 579 | * @q: the request queue for the device |
Tejun Heo | 2fb98e8 | 2008-02-19 11:36:53 +0100 | [diff] [blame] | 580 | * @dma_drain_needed: fn which returns non-zero if drain is necessary |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 581 | * @buf: physically contiguous buffer |
| 582 | * @size: size of the buffer in bytes |
| 583 | * |
| 584 | * Some devices have excess DMA problems and can't simply discard (or |
| 585 | * zero fill) the unwanted piece of the transfer. They have to have a |
| 586 | * real area of memory to transfer it into. The use case for this is |
| 587 | * ATAPI devices in DMA mode. If the packet command causes a transfer |
| 588 | * bigger than the transfer size some HBAs will lock up if there |
| 589 | * aren't DMA elements to contain the excess transfer. What this API |
| 590 | * does is adjust the queue so that the buf is always appended |
| 591 | * silently to the scatterlist. |
| 592 | * |
| 593 | * Note: This routine adjusts max_hw_segments to make room for |
| 594 | * appending the drain buffer. If you call |
| 595 | * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after |
| 596 | * calling this routine, you must set the limit to one fewer than your |
| 597 | * device can support otherwise there won't be room for the drain |
| 598 | * buffer. |
| 599 | */ |
Harvey Harrison | 448da4d | 2008-03-04 11:30:18 +0100 | [diff] [blame] | 600 | int blk_queue_dma_drain(struct request_queue *q, |
Tejun Heo | 2fb98e8 | 2008-02-19 11:36:53 +0100 | [diff] [blame] | 601 | dma_drain_needed_fn *dma_drain_needed, |
| 602 | void *buf, unsigned int size) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 603 | { |
Martin K. Petersen | ae03bf6 | 2009-05-22 17:17:50 -0400 | [diff] [blame] | 604 | if (queue_max_hw_segments(q) < 2 || queue_max_phys_segments(q) < 2) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 605 | return -EINVAL; |
| 606 | /* make room for appending the drain */ |
Martin K. Petersen | ae03bf6 | 2009-05-22 17:17:50 -0400 | [diff] [blame] | 607 | blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1); |
| 608 | blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1); |
Tejun Heo | 2fb98e8 | 2008-02-19 11:36:53 +0100 | [diff] [blame] | 609 | q->dma_drain_needed = dma_drain_needed; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 610 | q->dma_drain_buffer = buf; |
| 611 | q->dma_drain_size = size; |
| 612 | |
| 613 | return 0; |
| 614 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 615 | EXPORT_SYMBOL_GPL(blk_queue_dma_drain); |
| 616 | |
| 617 | /** |
| 618 | * blk_queue_segment_boundary - set boundary rules for segment merging |
| 619 | * @q: the request queue for the device |
| 620 | * @mask: the memory boundary mask |
| 621 | **/ |
| 622 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) |
| 623 | { |
| 624 | if (mask < PAGE_CACHE_SIZE - 1) { |
| 625 | mask = PAGE_CACHE_SIZE - 1; |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 626 | printk(KERN_INFO "%s: set to minimum %lx\n", |
| 627 | __func__, mask); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 628 | } |
| 629 | |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 630 | q->limits.seg_boundary_mask = mask; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 631 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 632 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
| 633 | |
| 634 | /** |
| 635 | * blk_queue_dma_alignment - set dma length and memory alignment |
| 636 | * @q: the request queue for the device |
| 637 | * @mask: alignment mask |
| 638 | * |
| 639 | * description: |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 640 | * set required memory and length alignment for direct dma transactions. |
Alan Cox | 8feb4d2 | 2009-04-01 15:01:39 +0100 | [diff] [blame] | 641 | * this is used when building direct io requests for the queue. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 642 | * |
| 643 | **/ |
| 644 | void blk_queue_dma_alignment(struct request_queue *q, int mask) |
| 645 | { |
| 646 | q->dma_alignment = mask; |
| 647 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 648 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
| 649 | |
| 650 | /** |
| 651 | * blk_queue_update_dma_alignment - update dma length and memory alignment |
| 652 | * @q: the request queue for the device |
| 653 | * @mask: alignment mask |
| 654 | * |
| 655 | * description: |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 656 | * update required memory and length alignment for direct dma transactions. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 657 | * If the requested alignment is larger than the current alignment, then |
| 658 | * the current queue alignment is updated to the new value, otherwise it |
| 659 | * is left alone. The design of this is to allow multiple objects |
| 660 | * (driver, device, transport etc) to set their respective |
| 661 | * alignments without having them interfere. |
| 662 | * |
| 663 | **/ |
| 664 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) |
| 665 | { |
| 666 | BUG_ON(mask > PAGE_SIZE); |
| 667 | |
| 668 | if (mask > q->dma_alignment) |
| 669 | q->dma_alignment = mask; |
| 670 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 671 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); |
| 672 | |
Harvey Harrison | aeb3d3a | 2008-08-28 09:27:42 +0200 | [diff] [blame] | 673 | static int __init blk_settings_init(void) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 674 | { |
| 675 | blk_max_low_pfn = max_low_pfn - 1; |
| 676 | blk_max_pfn = max_pfn - 1; |
| 677 | return 0; |
| 678 | } |
| 679 | subsys_initcall(blk_settings_init); |