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 | /** |
Martin K. Petersen | e475bba | 2009-06-16 08:23:52 +0200 | [diff] [blame] | 99 | * blk_set_default_limits - reset limits to default values |
Randy Dunlap | f740f5c | 2009-06-19 09:18:32 +0200 | [diff] [blame] | 100 | * @lim: the queue_limits structure to reset |
Martin K. Petersen | e475bba | 2009-06-16 08:23:52 +0200 | [diff] [blame] | 101 | * |
| 102 | * Description: |
| 103 | * Returns a queue_limit struct to its default state. Can be used by |
| 104 | * stacking drivers like DM that stage table swaps and reuse an |
| 105 | * existing device queue. |
| 106 | */ |
| 107 | void blk_set_default_limits(struct queue_limits *lim) |
| 108 | { |
| 109 | lim->max_phys_segments = MAX_PHYS_SEGMENTS; |
| 110 | lim->max_hw_segments = MAX_HW_SEGMENTS; |
| 111 | lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK; |
| 112 | lim->max_segment_size = MAX_SEGMENT_SIZE; |
| 113 | lim->max_sectors = lim->max_hw_sectors = SAFE_MAX_SECTORS; |
| 114 | lim->logical_block_size = lim->physical_block_size = lim->io_min = 512; |
Martin K. Petersen | 3a02c8e | 2009-06-18 09:56:03 +0200 | [diff] [blame] | 115 | lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT); |
Martin K. Petersen | e475bba | 2009-06-16 08:23:52 +0200 | [diff] [blame] | 116 | lim->alignment_offset = 0; |
| 117 | lim->io_opt = 0; |
| 118 | lim->misaligned = 0; |
| 119 | lim->no_cluster = 0; |
| 120 | } |
| 121 | EXPORT_SYMBOL(blk_set_default_limits); |
| 122 | |
| 123 | /** |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 124 | * blk_queue_make_request - define an alternate make_request function for a device |
| 125 | * @q: the request queue for the device to be affected |
| 126 | * @mfn: the alternate make_request function |
| 127 | * |
| 128 | * Description: |
| 129 | * The normal way for &struct bios to be passed to a device |
| 130 | * driver is for them to be collected into requests on a request |
| 131 | * queue, and then to allow the device driver to select requests |
| 132 | * off that queue when it is ready. This works well for many block |
| 133 | * devices. However some block devices (typically virtual devices |
| 134 | * such as md or lvm) do not benefit from the processing on the |
| 135 | * request queue, and are served best by having the requests passed |
| 136 | * directly to them. This can be achieved by providing a function |
| 137 | * to blk_queue_make_request(). |
| 138 | * |
| 139 | * Caveat: |
| 140 | * The driver that does this *must* be able to deal appropriately |
| 141 | * with buffers in "highmemory". This can be accomplished by either calling |
| 142 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling |
| 143 | * blk_queue_bounce() to create a buffer in normal memory. |
| 144 | **/ |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 145 | 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] | 146 | { |
| 147 | /* |
| 148 | * set defaults |
| 149 | */ |
| 150 | q->nr_requests = BLKDEV_MAX_RQ; |
Milan Broz | 0e435ac | 2008-12-03 12:55:08 +0100 | [diff] [blame] | 151 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 152 | q->make_request_fn = mfn; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 153 | blk_queue_dma_alignment(q, 511); |
| 154 | blk_queue_congestion_threshold(q); |
| 155 | q->nr_batching = BLK_BATCH_REQ; |
| 156 | |
| 157 | q->unplug_thresh = 4; /* hmm */ |
| 158 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ |
| 159 | if (q->unplug_delay == 0) |
| 160 | q->unplug_delay = 1; |
| 161 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 162 | q->unplug_timer.function = blk_unplug_timeout; |
| 163 | q->unplug_timer.data = (unsigned long)q; |
| 164 | |
Martin K. Petersen | e475bba | 2009-06-16 08:23:52 +0200 | [diff] [blame] | 165 | blk_set_default_limits(&q->limits); |
| 166 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 167 | /* |
Jens Axboe | a4e7d46 | 2009-07-28 09:07:29 +0200 | [diff] [blame] | 168 | * If the caller didn't supply a lock, fall back to our embedded |
| 169 | * per-queue locks |
| 170 | */ |
| 171 | if (!q->queue_lock) |
| 172 | q->queue_lock = &q->__queue_lock; |
| 173 | |
| 174 | /* |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 175 | * by default assume old behaviour and bounce for any highmem page |
| 176 | */ |
| 177 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); |
| 178 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 179 | EXPORT_SYMBOL(blk_queue_make_request); |
| 180 | |
| 181 | /** |
| 182 | * blk_queue_bounce_limit - set bounce buffer limit for queue |
Tejun Heo | cd0aca2 | 2009-04-15 22:10:25 +0900 | [diff] [blame] | 183 | * @q: the request queue for the device |
| 184 | * @dma_mask: the maximum address the device can handle |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 185 | * |
| 186 | * Description: |
| 187 | * Different hardware can have different requirements as to what pages |
| 188 | * it can do I/O directly to. A low level driver can call |
| 189 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce |
Tejun Heo | cd0aca2 | 2009-04-15 22:10:25 +0900 | [diff] [blame] | 190 | * buffers for doing I/O to pages residing above @dma_mask. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 191 | **/ |
Tejun Heo | cd0aca2 | 2009-04-15 22:10:25 +0900 | [diff] [blame] | 192 | void blk_queue_bounce_limit(struct request_queue *q, u64 dma_mask) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 193 | { |
Tejun Heo | cd0aca2 | 2009-04-15 22:10:25 +0900 | [diff] [blame] | 194 | unsigned long b_pfn = dma_mask >> PAGE_SHIFT; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 195 | int dma = 0; |
| 196 | |
| 197 | q->bounce_gfp = GFP_NOIO; |
| 198 | #if BITS_PER_LONG == 64 |
Tejun Heo | cd0aca2 | 2009-04-15 22:10:25 +0900 | [diff] [blame] | 199 | /* |
| 200 | * Assume anything <= 4GB can be handled by IOMMU. Actually |
| 201 | * some IOMMUs can handle everything, but I don't know of a |
| 202 | * way to test this here. |
| 203 | */ |
| 204 | if (b_pfn < (min_t(u64, 0xffffffffUL, BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 205 | dma = 1; |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 206 | q->limits.bounce_pfn = max_low_pfn; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 207 | #else |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 208 | if (b_pfn < blk_max_low_pfn) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 209 | dma = 1; |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 210 | q->limits.bounce_pfn = b_pfn; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 211 | #endif |
| 212 | if (dma) { |
| 213 | init_emergency_isa_pool(); |
| 214 | q->bounce_gfp = GFP_NOIO | GFP_DMA; |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 215 | q->limits.bounce_pfn = b_pfn; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 216 | } |
| 217 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 218 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
| 219 | |
| 220 | /** |
| 221 | * blk_queue_max_sectors - set max sectors for a request for this queue |
| 222 | * @q: the request queue for the device |
| 223 | * @max_sectors: max sectors in the usual 512b unit |
| 224 | * |
| 225 | * Description: |
| 226 | * Enables a low level driver to set an upper limit on the size of |
| 227 | * received requests. |
| 228 | **/ |
| 229 | void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors) |
| 230 | { |
| 231 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { |
| 232 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 233 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 234 | __func__, max_sectors); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 235 | } |
| 236 | |
| 237 | if (BLK_DEF_MAX_SECTORS > max_sectors) |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 238 | q->limits.max_hw_sectors = q->limits.max_sectors = max_sectors; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 239 | else { |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 240 | q->limits.max_sectors = BLK_DEF_MAX_SECTORS; |
| 241 | q->limits.max_hw_sectors = max_sectors; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 242 | } |
| 243 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 244 | EXPORT_SYMBOL(blk_queue_max_sectors); |
| 245 | |
Martin K. Petersen | ae03bf6 | 2009-05-22 17:17:50 -0400 | [diff] [blame] | 246 | void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_sectors) |
| 247 | { |
| 248 | if (BLK_DEF_MAX_SECTORS > max_sectors) |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 249 | q->limits.max_hw_sectors = BLK_DEF_MAX_SECTORS; |
Martin K. Petersen | ae03bf6 | 2009-05-22 17:17:50 -0400 | [diff] [blame] | 250 | else |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 251 | q->limits.max_hw_sectors = max_sectors; |
Martin K. Petersen | ae03bf6 | 2009-05-22 17:17:50 -0400 | [diff] [blame] | 252 | } |
| 253 | EXPORT_SYMBOL(blk_queue_max_hw_sectors); |
| 254 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 255 | /** |
| 256 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue |
| 257 | * @q: the request queue for the device |
| 258 | * @max_segments: max number of segments |
| 259 | * |
| 260 | * Description: |
| 261 | * Enables a low level driver to set an upper limit on the number of |
| 262 | * physical data segments in a request. This would be the largest sized |
| 263 | * scatter list the driver could handle. |
| 264 | **/ |
| 265 | void blk_queue_max_phys_segments(struct request_queue *q, |
| 266 | unsigned short max_segments) |
| 267 | { |
| 268 | if (!max_segments) { |
| 269 | max_segments = 1; |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 270 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 271 | __func__, max_segments); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 272 | } |
| 273 | |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 274 | q->limits.max_phys_segments = max_segments; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 275 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 276 | EXPORT_SYMBOL(blk_queue_max_phys_segments); |
| 277 | |
| 278 | /** |
| 279 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue |
| 280 | * @q: the request queue for the device |
| 281 | * @max_segments: max number of segments |
| 282 | * |
| 283 | * Description: |
| 284 | * Enables a low level driver to set an upper limit on the number of |
| 285 | * 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] | 286 | * address/length pairs the host adapter can actually give at once |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 287 | * to the device. |
| 288 | **/ |
| 289 | void blk_queue_max_hw_segments(struct request_queue *q, |
| 290 | unsigned short max_segments) |
| 291 | { |
| 292 | if (!max_segments) { |
| 293 | max_segments = 1; |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 294 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 295 | __func__, max_segments); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 296 | } |
| 297 | |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 298 | q->limits.max_hw_segments = max_segments; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 299 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 300 | EXPORT_SYMBOL(blk_queue_max_hw_segments); |
| 301 | |
| 302 | /** |
| 303 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg |
| 304 | * @q: the request queue for the device |
| 305 | * @max_size: max size of segment in bytes |
| 306 | * |
| 307 | * Description: |
| 308 | * Enables a low level driver to set an upper limit on the size of a |
| 309 | * coalesced segment |
| 310 | **/ |
| 311 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) |
| 312 | { |
| 313 | if (max_size < PAGE_CACHE_SIZE) { |
| 314 | max_size = PAGE_CACHE_SIZE; |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 315 | printk(KERN_INFO "%s: set to minimum %d\n", |
| 316 | __func__, max_size); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 317 | } |
| 318 | |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 319 | q->limits.max_segment_size = max_size; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 320 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 321 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
| 322 | |
| 323 | /** |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 324 | * blk_queue_logical_block_size - set logical block size for the queue |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 325 | * @q: the request queue for the device |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 326 | * @size: the logical block size, in bytes |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 327 | * |
| 328 | * Description: |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 329 | * This should be set to the lowest possible block size that the |
| 330 | * storage device can address. The default of 512 covers most |
| 331 | * hardware. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 332 | **/ |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 333 | 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] | 334 | { |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 335 | q->limits.logical_block_size = size; |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 336 | |
| 337 | if (q->limits.physical_block_size < size) |
| 338 | q->limits.physical_block_size = size; |
| 339 | |
| 340 | if (q->limits.io_min < q->limits.physical_block_size) |
| 341 | q->limits.io_min = q->limits.physical_block_size; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 342 | } |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 343 | EXPORT_SYMBOL(blk_queue_logical_block_size); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 344 | |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 345 | /** |
| 346 | * blk_queue_physical_block_size - set physical block size for the queue |
| 347 | * @q: the request queue for the device |
| 348 | * @size: the physical block size, in bytes |
| 349 | * |
| 350 | * Description: |
| 351 | * This should be set to the lowest possible sector size that the |
| 352 | * hardware can operate on without reverting to read-modify-write |
| 353 | * operations. |
| 354 | */ |
| 355 | void blk_queue_physical_block_size(struct request_queue *q, unsigned short size) |
| 356 | { |
| 357 | q->limits.physical_block_size = size; |
| 358 | |
| 359 | if (q->limits.physical_block_size < q->limits.logical_block_size) |
| 360 | q->limits.physical_block_size = q->limits.logical_block_size; |
| 361 | |
| 362 | if (q->limits.io_min < q->limits.physical_block_size) |
| 363 | q->limits.io_min = q->limits.physical_block_size; |
| 364 | } |
| 365 | EXPORT_SYMBOL(blk_queue_physical_block_size); |
| 366 | |
| 367 | /** |
| 368 | * blk_queue_alignment_offset - set physical block alignment offset |
| 369 | * @q: the request queue for the device |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 370 | * @offset: alignment offset in bytes |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 371 | * |
| 372 | * Description: |
| 373 | * Some devices are naturally misaligned to compensate for things like |
| 374 | * the legacy DOS partition table 63-sector offset. Low-level drivers |
| 375 | * should call this function for devices whose first sector is not |
| 376 | * naturally aligned. |
| 377 | */ |
| 378 | void blk_queue_alignment_offset(struct request_queue *q, unsigned int offset) |
| 379 | { |
| 380 | q->limits.alignment_offset = |
| 381 | offset & (q->limits.physical_block_size - 1); |
| 382 | q->limits.misaligned = 0; |
| 383 | } |
| 384 | EXPORT_SYMBOL(blk_queue_alignment_offset); |
| 385 | |
| 386 | /** |
| 387 | * blk_queue_io_min - set minimum request size for the queue |
| 388 | * @q: the request queue for the device |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 389 | * @min: smallest I/O size in bytes |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 390 | * |
| 391 | * Description: |
| 392 | * Some devices have an internal block size bigger than the reported |
| 393 | * hardware sector size. This function can be used to signal the |
| 394 | * smallest I/O the device can perform without incurring a performance |
| 395 | * penalty. |
| 396 | */ |
| 397 | void blk_queue_io_min(struct request_queue *q, unsigned int min) |
| 398 | { |
| 399 | q->limits.io_min = min; |
| 400 | |
| 401 | if (q->limits.io_min < q->limits.logical_block_size) |
| 402 | q->limits.io_min = q->limits.logical_block_size; |
| 403 | |
| 404 | if (q->limits.io_min < q->limits.physical_block_size) |
| 405 | q->limits.io_min = q->limits.physical_block_size; |
| 406 | } |
| 407 | EXPORT_SYMBOL(blk_queue_io_min); |
| 408 | |
| 409 | /** |
| 410 | * blk_queue_io_opt - set optimal request size for the queue |
| 411 | * @q: the request queue for the device |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 412 | * @opt: optimal request size in bytes |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 413 | * |
| 414 | * Description: |
| 415 | * Drivers can call this function to set the preferred I/O request |
| 416 | * size for devices that report such a value. |
| 417 | */ |
| 418 | void blk_queue_io_opt(struct request_queue *q, unsigned int opt) |
| 419 | { |
| 420 | q->limits.io_opt = opt; |
| 421 | } |
| 422 | EXPORT_SYMBOL(blk_queue_io_opt); |
| 423 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 424 | /* |
| 425 | * Returns the minimum that is _not_ zero, unless both are zero. |
| 426 | */ |
| 427 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) |
| 428 | |
| 429 | /** |
| 430 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers |
| 431 | * @t: the stacking driver (top) |
| 432 | * @b: the underlying device (bottom) |
| 433 | **/ |
| 434 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) |
| 435 | { |
| 436 | /* zero is "infinity" */ |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 437 | t->limits.max_sectors = min_not_zero(queue_max_sectors(t), |
| 438 | queue_max_sectors(b)); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 439 | |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 440 | t->limits.max_hw_sectors = min_not_zero(queue_max_hw_sectors(t), |
| 441 | queue_max_hw_sectors(b)); |
| 442 | |
| 443 | t->limits.seg_boundary_mask = min_not_zero(queue_segment_boundary(t), |
| 444 | queue_segment_boundary(b)); |
| 445 | |
| 446 | t->limits.max_phys_segments = min_not_zero(queue_max_phys_segments(t), |
| 447 | queue_max_phys_segments(b)); |
| 448 | |
| 449 | t->limits.max_hw_segments = min_not_zero(queue_max_hw_segments(t), |
| 450 | queue_max_hw_segments(b)); |
| 451 | |
| 452 | t->limits.max_segment_size = min_not_zero(queue_max_segment_size(t), |
| 453 | queue_max_segment_size(b)); |
| 454 | |
| 455 | t->limits.logical_block_size = max(queue_logical_block_size(t), |
| 456 | queue_logical_block_size(b)); |
| 457 | |
Neil Brown | e7e72bf | 2008-05-14 16:05:54 -0700 | [diff] [blame] | 458 | if (!t->queue_lock) |
| 459 | WARN_ON_ONCE(1); |
| 460 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { |
| 461 | unsigned long flags; |
| 462 | spin_lock_irqsave(t->queue_lock, flags); |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 463 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); |
Neil Brown | e7e72bf | 2008-05-14 16:05:54 -0700 | [diff] [blame] | 464 | spin_unlock_irqrestore(t->queue_lock, flags); |
| 465 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 466 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 467 | EXPORT_SYMBOL(blk_queue_stack_limits); |
| 468 | |
| 469 | /** |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 470 | * blk_stack_limits - adjust queue_limits for stacked devices |
| 471 | * @t: the stacking driver limits (top) |
Martin K. Petersen | 77634f3 | 2009-06-09 06:23:22 +0200 | [diff] [blame] | 472 | * @b: the underlying queue limits (bottom) |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 473 | * @offset: offset to beginning of data within component device |
| 474 | * |
| 475 | * Description: |
| 476 | * Merges two queue_limit structs. Returns 0 if alignment didn't |
| 477 | * change. Returns -1 if adding the bottom device caused |
| 478 | * misalignment. |
| 479 | */ |
| 480 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, |
| 481 | sector_t offset) |
| 482 | { |
| 483 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
| 484 | 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] | 485 | 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] | 486 | |
| 487 | t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask, |
| 488 | b->seg_boundary_mask); |
| 489 | |
| 490 | t->max_phys_segments = min_not_zero(t->max_phys_segments, |
| 491 | b->max_phys_segments); |
| 492 | |
| 493 | t->max_hw_segments = min_not_zero(t->max_hw_segments, |
| 494 | b->max_hw_segments); |
| 495 | |
| 496 | t->max_segment_size = min_not_zero(t->max_segment_size, |
| 497 | b->max_segment_size); |
| 498 | |
| 499 | t->logical_block_size = max(t->logical_block_size, |
| 500 | b->logical_block_size); |
| 501 | |
| 502 | t->physical_block_size = max(t->physical_block_size, |
| 503 | b->physical_block_size); |
| 504 | |
| 505 | t->io_min = max(t->io_min, b->io_min); |
| 506 | t->no_cluster |= b->no_cluster; |
| 507 | |
| 508 | /* Bottom device offset aligned? */ |
| 509 | if (offset && |
| 510 | (offset & (b->physical_block_size - 1)) != b->alignment_offset) { |
| 511 | t->misaligned = 1; |
| 512 | return -1; |
| 513 | } |
| 514 | |
| 515 | /* If top has no alignment offset, inherit from bottom */ |
| 516 | if (!t->alignment_offset) |
| 517 | t->alignment_offset = |
| 518 | b->alignment_offset & (b->physical_block_size - 1); |
| 519 | |
| 520 | /* Top device aligned on logical block boundary? */ |
| 521 | if (t->alignment_offset & (t->logical_block_size - 1)) { |
| 522 | t->misaligned = 1; |
| 523 | return -1; |
| 524 | } |
| 525 | |
| 526 | return 0; |
| 527 | } |
Mike Snitzer | 5d85d32 | 2009-05-28 11:04:53 +0200 | [diff] [blame] | 528 | EXPORT_SYMBOL(blk_stack_limits); |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 529 | |
| 530 | /** |
| 531 | * disk_stack_limits - adjust queue limits for stacked drivers |
Martin K. Petersen | 77634f3 | 2009-06-09 06:23:22 +0200 | [diff] [blame] | 532 | * @disk: MD/DM gendisk (top) |
Martin K. Petersen | c72758f | 2009-05-22 17:17:53 -0400 | [diff] [blame] | 533 | * @bdev: the underlying block device (bottom) |
| 534 | * @offset: offset to beginning of data within component device |
| 535 | * |
| 536 | * Description: |
| 537 | * Merges the limits for two queues. Returns 0 if alignment |
| 538 | * didn't change. Returns -1 if adding the bottom device caused |
| 539 | * misalignment. |
| 540 | */ |
| 541 | void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, |
| 542 | sector_t offset) |
| 543 | { |
| 544 | struct request_queue *t = disk->queue; |
| 545 | struct request_queue *b = bdev_get_queue(bdev); |
| 546 | |
| 547 | offset += get_start_sect(bdev) << 9; |
| 548 | |
| 549 | if (blk_stack_limits(&t->limits, &b->limits, offset) < 0) { |
| 550 | char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; |
| 551 | |
| 552 | disk_name(disk, 0, top); |
| 553 | bdevname(bdev, bottom); |
| 554 | |
| 555 | printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", |
| 556 | top, bottom); |
| 557 | } |
| 558 | |
| 559 | if (!t->queue_lock) |
| 560 | WARN_ON_ONCE(1); |
| 561 | else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { |
| 562 | unsigned long flags; |
| 563 | |
| 564 | spin_lock_irqsave(t->queue_lock, flags); |
| 565 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) |
| 566 | queue_flag_clear(QUEUE_FLAG_CLUSTER, t); |
| 567 | spin_unlock_irqrestore(t->queue_lock, flags); |
| 568 | } |
| 569 | } |
| 570 | EXPORT_SYMBOL(disk_stack_limits); |
| 571 | |
| 572 | /** |
Tejun Heo | e3790c7 | 2008-03-04 11:18:17 +0100 | [diff] [blame] | 573 | * blk_queue_dma_pad - set pad mask |
| 574 | * @q: the request queue for the device |
| 575 | * @mask: pad mask |
| 576 | * |
FUJITA Tomonori | 27f8221 | 2008-07-04 09:30:03 +0200 | [diff] [blame] | 577 | * Set dma pad mask. |
Tejun Heo | e3790c7 | 2008-03-04 11:18:17 +0100 | [diff] [blame] | 578 | * |
FUJITA Tomonori | 27f8221 | 2008-07-04 09:30:03 +0200 | [diff] [blame] | 579 | * Appending pad buffer to a request modifies the last entry of a |
| 580 | * scatter list such that it includes the pad buffer. |
Tejun Heo | e3790c7 | 2008-03-04 11:18:17 +0100 | [diff] [blame] | 581 | **/ |
| 582 | void blk_queue_dma_pad(struct request_queue *q, unsigned int mask) |
| 583 | { |
| 584 | q->dma_pad_mask = mask; |
| 585 | } |
| 586 | EXPORT_SYMBOL(blk_queue_dma_pad); |
| 587 | |
| 588 | /** |
FUJITA Tomonori | 27f8221 | 2008-07-04 09:30:03 +0200 | [diff] [blame] | 589 | * blk_queue_update_dma_pad - update pad mask |
| 590 | * @q: the request queue for the device |
| 591 | * @mask: pad mask |
| 592 | * |
| 593 | * Update dma pad mask. |
| 594 | * |
| 595 | * Appending pad buffer to a request modifies the last entry of a |
| 596 | * scatter list such that it includes the pad buffer. |
| 597 | **/ |
| 598 | void blk_queue_update_dma_pad(struct request_queue *q, unsigned int mask) |
| 599 | { |
| 600 | if (mask > q->dma_pad_mask) |
| 601 | q->dma_pad_mask = mask; |
| 602 | } |
| 603 | EXPORT_SYMBOL(blk_queue_update_dma_pad); |
| 604 | |
| 605 | /** |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 606 | * blk_queue_dma_drain - Set up a drain buffer for excess dma. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 607 | * @q: the request queue for the device |
Tejun Heo | 2fb98e8 | 2008-02-19 11:36:53 +0100 | [diff] [blame] | 608 | * @dma_drain_needed: fn which returns non-zero if drain is necessary |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 609 | * @buf: physically contiguous buffer |
| 610 | * @size: size of the buffer in bytes |
| 611 | * |
| 612 | * Some devices have excess DMA problems and can't simply discard (or |
| 613 | * zero fill) the unwanted piece of the transfer. They have to have a |
| 614 | * real area of memory to transfer it into. The use case for this is |
| 615 | * ATAPI devices in DMA mode. If the packet command causes a transfer |
| 616 | * bigger than the transfer size some HBAs will lock up if there |
| 617 | * aren't DMA elements to contain the excess transfer. What this API |
| 618 | * does is adjust the queue so that the buf is always appended |
| 619 | * silently to the scatterlist. |
| 620 | * |
| 621 | * Note: This routine adjusts max_hw_segments to make room for |
| 622 | * appending the drain buffer. If you call |
| 623 | * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after |
| 624 | * calling this routine, you must set the limit to one fewer than your |
| 625 | * device can support otherwise there won't be room for the drain |
| 626 | * buffer. |
| 627 | */ |
Harvey Harrison | 448da4d | 2008-03-04 11:30:18 +0100 | [diff] [blame] | 628 | int blk_queue_dma_drain(struct request_queue *q, |
Tejun Heo | 2fb98e8 | 2008-02-19 11:36:53 +0100 | [diff] [blame] | 629 | dma_drain_needed_fn *dma_drain_needed, |
| 630 | void *buf, unsigned int size) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 631 | { |
Martin K. Petersen | ae03bf6 | 2009-05-22 17:17:50 -0400 | [diff] [blame] | 632 | 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] | 633 | return -EINVAL; |
| 634 | /* make room for appending the drain */ |
Martin K. Petersen | ae03bf6 | 2009-05-22 17:17:50 -0400 | [diff] [blame] | 635 | blk_queue_max_hw_segments(q, queue_max_hw_segments(q) - 1); |
| 636 | blk_queue_max_phys_segments(q, queue_max_phys_segments(q) - 1); |
Tejun Heo | 2fb98e8 | 2008-02-19 11:36:53 +0100 | [diff] [blame] | 637 | q->dma_drain_needed = dma_drain_needed; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 638 | q->dma_drain_buffer = buf; |
| 639 | q->dma_drain_size = size; |
| 640 | |
| 641 | return 0; |
| 642 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 643 | EXPORT_SYMBOL_GPL(blk_queue_dma_drain); |
| 644 | |
| 645 | /** |
| 646 | * blk_queue_segment_boundary - set boundary rules for segment merging |
| 647 | * @q: the request queue for the device |
| 648 | * @mask: the memory boundary mask |
| 649 | **/ |
| 650 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) |
| 651 | { |
| 652 | if (mask < PAGE_CACHE_SIZE - 1) { |
| 653 | mask = PAGE_CACHE_SIZE - 1; |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 654 | printk(KERN_INFO "%s: set to minimum %lx\n", |
| 655 | __func__, mask); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 656 | } |
| 657 | |
Martin K. Petersen | 025146e | 2009-05-22 17:17:51 -0400 | [diff] [blame] | 658 | q->limits.seg_boundary_mask = mask; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 659 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 660 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
| 661 | |
| 662 | /** |
| 663 | * blk_queue_dma_alignment - set dma length and memory alignment |
| 664 | * @q: the request queue for the device |
| 665 | * @mask: alignment mask |
| 666 | * |
| 667 | * description: |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 668 | * set required memory and length alignment for direct dma transactions. |
Alan Cox | 8feb4d2 | 2009-04-01 15:01:39 +0100 | [diff] [blame] | 669 | * this is used when building direct io requests for the queue. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 670 | * |
| 671 | **/ |
| 672 | void blk_queue_dma_alignment(struct request_queue *q, int mask) |
| 673 | { |
| 674 | q->dma_alignment = mask; |
| 675 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 676 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
| 677 | |
| 678 | /** |
| 679 | * blk_queue_update_dma_alignment - update dma length and memory alignment |
| 680 | * @q: the request queue for the device |
| 681 | * @mask: alignment mask |
| 682 | * |
| 683 | * description: |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 684 | * update required memory and length alignment for direct dma transactions. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 685 | * If the requested alignment is larger than the current alignment, then |
| 686 | * the current queue alignment is updated to the new value, otherwise it |
| 687 | * is left alone. The design of this is to allow multiple objects |
| 688 | * (driver, device, transport etc) to set their respective |
| 689 | * alignments without having them interfere. |
| 690 | * |
| 691 | **/ |
| 692 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) |
| 693 | { |
| 694 | BUG_ON(mask > PAGE_SIZE); |
| 695 | |
| 696 | if (mask > q->dma_alignment) |
| 697 | q->dma_alignment = mask; |
| 698 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 699 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); |
| 700 | |
Harvey Harrison | aeb3d3a | 2008-08-28 09:27:42 +0200 | [diff] [blame] | 701 | static int __init blk_settings_init(void) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 702 | { |
| 703 | blk_max_low_pfn = max_low_pfn - 1; |
| 704 | blk_max_pfn = max_pfn - 1; |
| 705 | return 0; |
| 706 | } |
| 707 | subsys_initcall(blk_settings_init); |