Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/drivers/block/ll_rw_blk.c |
| 3 | * |
| 4 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 5 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics |
| 6 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE |
| 7 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> |
| 8 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000 |
| 9 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
| 10 | */ |
| 11 | |
| 12 | /* |
| 13 | * This handles all read/write requests to block devices |
| 14 | */ |
| 15 | #include <linux/config.h> |
| 16 | #include <linux/kernel.h> |
| 17 | #include <linux/module.h> |
| 18 | #include <linux/backing-dev.h> |
| 19 | #include <linux/bio.h> |
| 20 | #include <linux/blkdev.h> |
| 21 | #include <linux/highmem.h> |
| 22 | #include <linux/mm.h> |
| 23 | #include <linux/kernel_stat.h> |
| 24 | #include <linux/string.h> |
| 25 | #include <linux/init.h> |
| 26 | #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */ |
| 27 | #include <linux/completion.h> |
| 28 | #include <linux/slab.h> |
| 29 | #include <linux/swap.h> |
| 30 | #include <linux/writeback.h> |
| 31 | |
| 32 | /* |
| 33 | * for max sense size |
| 34 | */ |
| 35 | #include <scsi/scsi_cmnd.h> |
| 36 | |
| 37 | static void blk_unplug_work(void *data); |
| 38 | static void blk_unplug_timeout(unsigned long data); |
| 39 | |
| 40 | /* |
| 41 | * For the allocated request tables |
| 42 | */ |
| 43 | static kmem_cache_t *request_cachep; |
| 44 | |
| 45 | /* |
| 46 | * For queue allocation |
| 47 | */ |
| 48 | static kmem_cache_t *requestq_cachep; |
| 49 | |
| 50 | /* |
| 51 | * For io context allocations |
| 52 | */ |
| 53 | static kmem_cache_t *iocontext_cachep; |
| 54 | |
| 55 | static wait_queue_head_t congestion_wqh[2] = { |
| 56 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), |
| 57 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) |
| 58 | }; |
| 59 | |
| 60 | /* |
| 61 | * Controlling structure to kblockd |
| 62 | */ |
| 63 | static struct workqueue_struct *kblockd_workqueue; |
| 64 | |
| 65 | unsigned long blk_max_low_pfn, blk_max_pfn; |
| 66 | |
| 67 | EXPORT_SYMBOL(blk_max_low_pfn); |
| 68 | EXPORT_SYMBOL(blk_max_pfn); |
| 69 | |
| 70 | /* Amount of time in which a process may batch requests */ |
| 71 | #define BLK_BATCH_TIME (HZ/50UL) |
| 72 | |
| 73 | /* Number of requests a "batching" process may submit */ |
| 74 | #define BLK_BATCH_REQ 32 |
| 75 | |
| 76 | /* |
| 77 | * Return the threshold (number of used requests) at which the queue is |
| 78 | * considered to be congested. It include a little hysteresis to keep the |
| 79 | * context switch rate down. |
| 80 | */ |
| 81 | static inline int queue_congestion_on_threshold(struct request_queue *q) |
| 82 | { |
| 83 | return q->nr_congestion_on; |
| 84 | } |
| 85 | |
| 86 | /* |
| 87 | * The threshold at which a queue is considered to be uncongested |
| 88 | */ |
| 89 | static inline int queue_congestion_off_threshold(struct request_queue *q) |
| 90 | { |
| 91 | return q->nr_congestion_off; |
| 92 | } |
| 93 | |
| 94 | static void blk_queue_congestion_threshold(struct request_queue *q) |
| 95 | { |
| 96 | int nr; |
| 97 | |
| 98 | nr = q->nr_requests - (q->nr_requests / 8) + 1; |
| 99 | if (nr > q->nr_requests) |
| 100 | nr = q->nr_requests; |
| 101 | q->nr_congestion_on = nr; |
| 102 | |
| 103 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; |
| 104 | if (nr < 1) |
| 105 | nr = 1; |
| 106 | q->nr_congestion_off = nr; |
| 107 | } |
| 108 | |
| 109 | /* |
| 110 | * A queue has just exitted congestion. Note this in the global counter of |
| 111 | * congested queues, and wake up anyone who was waiting for requests to be |
| 112 | * put back. |
| 113 | */ |
| 114 | static void clear_queue_congested(request_queue_t *q, int rw) |
| 115 | { |
| 116 | enum bdi_state bit; |
| 117 | wait_queue_head_t *wqh = &congestion_wqh[rw]; |
| 118 | |
| 119 | bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested; |
| 120 | clear_bit(bit, &q->backing_dev_info.state); |
| 121 | smp_mb__after_clear_bit(); |
| 122 | if (waitqueue_active(wqh)) |
| 123 | wake_up(wqh); |
| 124 | } |
| 125 | |
| 126 | /* |
| 127 | * A queue has just entered congestion. Flag that in the queue's VM-visible |
| 128 | * state flags and increment the global gounter of congested queues. |
| 129 | */ |
| 130 | static void set_queue_congested(request_queue_t *q, int rw) |
| 131 | { |
| 132 | enum bdi_state bit; |
| 133 | |
| 134 | bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested; |
| 135 | set_bit(bit, &q->backing_dev_info.state); |
| 136 | } |
| 137 | |
| 138 | /** |
| 139 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info |
| 140 | * @bdev: device |
| 141 | * |
| 142 | * Locates the passed device's request queue and returns the address of its |
| 143 | * backing_dev_info |
| 144 | * |
| 145 | * Will return NULL if the request queue cannot be located. |
| 146 | */ |
| 147 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) |
| 148 | { |
| 149 | struct backing_dev_info *ret = NULL; |
| 150 | request_queue_t *q = bdev_get_queue(bdev); |
| 151 | |
| 152 | if (q) |
| 153 | ret = &q->backing_dev_info; |
| 154 | return ret; |
| 155 | } |
| 156 | |
| 157 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
| 158 | |
| 159 | void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data) |
| 160 | { |
| 161 | q->activity_fn = fn; |
| 162 | q->activity_data = data; |
| 163 | } |
| 164 | |
| 165 | EXPORT_SYMBOL(blk_queue_activity_fn); |
| 166 | |
| 167 | /** |
| 168 | * blk_queue_prep_rq - set a prepare_request function for queue |
| 169 | * @q: queue |
| 170 | * @pfn: prepare_request function |
| 171 | * |
| 172 | * It's possible for a queue to register a prepare_request callback which |
| 173 | * is invoked before the request is handed to the request_fn. The goal of |
| 174 | * the function is to prepare a request for I/O, it can be used to build a |
| 175 | * cdb from the request data for instance. |
| 176 | * |
| 177 | */ |
| 178 | void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn) |
| 179 | { |
| 180 | q->prep_rq_fn = pfn; |
| 181 | } |
| 182 | |
| 183 | EXPORT_SYMBOL(blk_queue_prep_rq); |
| 184 | |
| 185 | /** |
| 186 | * blk_queue_merge_bvec - set a merge_bvec function for queue |
| 187 | * @q: queue |
| 188 | * @mbfn: merge_bvec_fn |
| 189 | * |
| 190 | * Usually queues have static limitations on the max sectors or segments that |
| 191 | * we can put in a request. Stacking drivers may have some settings that |
| 192 | * are dynamic, and thus we have to query the queue whether it is ok to |
| 193 | * add a new bio_vec to a bio at a given offset or not. If the block device |
| 194 | * has such limitations, it needs to register a merge_bvec_fn to control |
| 195 | * the size of bio's sent to it. Note that a block device *must* allow a |
| 196 | * single page to be added to an empty bio. The block device driver may want |
| 197 | * to use the bio_split() function to deal with these bio's. By default |
| 198 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are |
| 199 | * honored. |
| 200 | */ |
| 201 | void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn) |
| 202 | { |
| 203 | q->merge_bvec_fn = mbfn; |
| 204 | } |
| 205 | |
| 206 | EXPORT_SYMBOL(blk_queue_merge_bvec); |
| 207 | |
| 208 | /** |
| 209 | * blk_queue_make_request - define an alternate make_request function for a device |
| 210 | * @q: the request queue for the device to be affected |
| 211 | * @mfn: the alternate make_request function |
| 212 | * |
| 213 | * Description: |
| 214 | * The normal way for &struct bios to be passed to a device |
| 215 | * driver is for them to be collected into requests on a request |
| 216 | * queue, and then to allow the device driver to select requests |
| 217 | * off that queue when it is ready. This works well for many block |
| 218 | * devices. However some block devices (typically virtual devices |
| 219 | * such as md or lvm) do not benefit from the processing on the |
| 220 | * request queue, and are served best by having the requests passed |
| 221 | * directly to them. This can be achieved by providing a function |
| 222 | * to blk_queue_make_request(). |
| 223 | * |
| 224 | * Caveat: |
| 225 | * The driver that does this *must* be able to deal appropriately |
| 226 | * with buffers in "highmemory". This can be accomplished by either calling |
| 227 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling |
| 228 | * blk_queue_bounce() to create a buffer in normal memory. |
| 229 | **/ |
| 230 | void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn) |
| 231 | { |
| 232 | /* |
| 233 | * set defaults |
| 234 | */ |
| 235 | q->nr_requests = BLKDEV_MAX_RQ; |
| 236 | q->max_phys_segments = MAX_PHYS_SEGMENTS; |
| 237 | q->max_hw_segments = MAX_HW_SEGMENTS; |
| 238 | q->make_request_fn = mfn; |
| 239 | q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; |
| 240 | q->backing_dev_info.state = 0; |
| 241 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; |
| 242 | blk_queue_max_sectors(q, MAX_SECTORS); |
| 243 | blk_queue_hardsect_size(q, 512); |
| 244 | blk_queue_dma_alignment(q, 511); |
| 245 | blk_queue_congestion_threshold(q); |
| 246 | q->nr_batching = BLK_BATCH_REQ; |
| 247 | |
| 248 | q->unplug_thresh = 4; /* hmm */ |
| 249 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ |
| 250 | if (q->unplug_delay == 0) |
| 251 | q->unplug_delay = 1; |
| 252 | |
| 253 | INIT_WORK(&q->unplug_work, blk_unplug_work, q); |
| 254 | |
| 255 | q->unplug_timer.function = blk_unplug_timeout; |
| 256 | q->unplug_timer.data = (unsigned long)q; |
| 257 | |
| 258 | /* |
| 259 | * by default assume old behaviour and bounce for any highmem page |
| 260 | */ |
| 261 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); |
| 262 | |
| 263 | blk_queue_activity_fn(q, NULL, NULL); |
| 264 | |
| 265 | INIT_LIST_HEAD(&q->drain_list); |
| 266 | } |
| 267 | |
| 268 | EXPORT_SYMBOL(blk_queue_make_request); |
| 269 | |
| 270 | static inline void rq_init(request_queue_t *q, struct request *rq) |
| 271 | { |
| 272 | INIT_LIST_HEAD(&rq->queuelist); |
| 273 | |
| 274 | rq->errors = 0; |
| 275 | rq->rq_status = RQ_ACTIVE; |
| 276 | rq->bio = rq->biotail = NULL; |
| 277 | rq->buffer = NULL; |
| 278 | rq->ref_count = 1; |
| 279 | rq->q = q; |
| 280 | rq->waiting = NULL; |
| 281 | rq->special = NULL; |
| 282 | rq->data_len = 0; |
| 283 | rq->data = NULL; |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 284 | rq->nr_phys_segments = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | rq->sense = NULL; |
| 286 | rq->end_io = NULL; |
| 287 | rq->end_io_data = NULL; |
| 288 | } |
| 289 | |
| 290 | /** |
| 291 | * blk_queue_ordered - does this queue support ordered writes |
| 292 | * @q: the request queue |
| 293 | * @flag: see below |
| 294 | * |
| 295 | * Description: |
| 296 | * For journalled file systems, doing ordered writes on a commit |
| 297 | * block instead of explicitly doing wait_on_buffer (which is bad |
| 298 | * for performance) can be a big win. Block drivers supporting this |
| 299 | * feature should call this function and indicate so. |
| 300 | * |
| 301 | **/ |
| 302 | void blk_queue_ordered(request_queue_t *q, int flag) |
| 303 | { |
| 304 | switch (flag) { |
| 305 | case QUEUE_ORDERED_NONE: |
| 306 | if (q->flush_rq) |
| 307 | kmem_cache_free(request_cachep, q->flush_rq); |
| 308 | q->flush_rq = NULL; |
| 309 | q->ordered = flag; |
| 310 | break; |
| 311 | case QUEUE_ORDERED_TAG: |
| 312 | q->ordered = flag; |
| 313 | break; |
| 314 | case QUEUE_ORDERED_FLUSH: |
| 315 | q->ordered = flag; |
| 316 | if (!q->flush_rq) |
| 317 | q->flush_rq = kmem_cache_alloc(request_cachep, |
| 318 | GFP_KERNEL); |
| 319 | break; |
| 320 | default: |
| 321 | printk("blk_queue_ordered: bad value %d\n", flag); |
| 322 | break; |
| 323 | } |
| 324 | } |
| 325 | |
| 326 | EXPORT_SYMBOL(blk_queue_ordered); |
| 327 | |
| 328 | /** |
| 329 | * blk_queue_issue_flush_fn - set function for issuing a flush |
| 330 | * @q: the request queue |
| 331 | * @iff: the function to be called issuing the flush |
| 332 | * |
| 333 | * Description: |
| 334 | * If a driver supports issuing a flush command, the support is notified |
| 335 | * to the block layer by defining it through this call. |
| 336 | * |
| 337 | **/ |
| 338 | void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff) |
| 339 | { |
| 340 | q->issue_flush_fn = iff; |
| 341 | } |
| 342 | |
| 343 | EXPORT_SYMBOL(blk_queue_issue_flush_fn); |
| 344 | |
| 345 | /* |
| 346 | * Cache flushing for ordered writes handling |
| 347 | */ |
| 348 | static void blk_pre_flush_end_io(struct request *flush_rq) |
| 349 | { |
| 350 | struct request *rq = flush_rq->end_io_data; |
| 351 | request_queue_t *q = rq->q; |
| 352 | |
| 353 | rq->flags |= REQ_BAR_PREFLUSH; |
| 354 | |
| 355 | if (!flush_rq->errors) |
| 356 | elv_requeue_request(q, rq); |
| 357 | else { |
| 358 | q->end_flush_fn(q, flush_rq); |
| 359 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); |
| 360 | q->request_fn(q); |
| 361 | } |
| 362 | } |
| 363 | |
| 364 | static void blk_post_flush_end_io(struct request *flush_rq) |
| 365 | { |
| 366 | struct request *rq = flush_rq->end_io_data; |
| 367 | request_queue_t *q = rq->q; |
| 368 | |
| 369 | rq->flags |= REQ_BAR_POSTFLUSH; |
| 370 | |
| 371 | q->end_flush_fn(q, flush_rq); |
| 372 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); |
| 373 | q->request_fn(q); |
| 374 | } |
| 375 | |
| 376 | struct request *blk_start_pre_flush(request_queue_t *q, struct request *rq) |
| 377 | { |
| 378 | struct request *flush_rq = q->flush_rq; |
| 379 | |
| 380 | BUG_ON(!blk_barrier_rq(rq)); |
| 381 | |
| 382 | if (test_and_set_bit(QUEUE_FLAG_FLUSH, &q->queue_flags)) |
| 383 | return NULL; |
| 384 | |
| 385 | rq_init(q, flush_rq); |
| 386 | flush_rq->elevator_private = NULL; |
| 387 | flush_rq->flags = REQ_BAR_FLUSH; |
| 388 | flush_rq->rq_disk = rq->rq_disk; |
| 389 | flush_rq->rl = NULL; |
| 390 | |
| 391 | /* |
| 392 | * prepare_flush returns 0 if no flush is needed, just mark both |
| 393 | * pre and post flush as done in that case |
| 394 | */ |
| 395 | if (!q->prepare_flush_fn(q, flush_rq)) { |
| 396 | rq->flags |= REQ_BAR_PREFLUSH | REQ_BAR_POSTFLUSH; |
| 397 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); |
| 398 | return rq; |
| 399 | } |
| 400 | |
| 401 | /* |
| 402 | * some drivers dequeue requests right away, some only after io |
| 403 | * completion. make sure the request is dequeued. |
| 404 | */ |
| 405 | if (!list_empty(&rq->queuelist)) |
| 406 | blkdev_dequeue_request(rq); |
| 407 | |
| 408 | elv_deactivate_request(q, rq); |
| 409 | |
| 410 | flush_rq->end_io_data = rq; |
| 411 | flush_rq->end_io = blk_pre_flush_end_io; |
| 412 | |
| 413 | __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0); |
| 414 | return flush_rq; |
| 415 | } |
| 416 | |
| 417 | static void blk_start_post_flush(request_queue_t *q, struct request *rq) |
| 418 | { |
| 419 | struct request *flush_rq = q->flush_rq; |
| 420 | |
| 421 | BUG_ON(!blk_barrier_rq(rq)); |
| 422 | |
| 423 | rq_init(q, flush_rq); |
| 424 | flush_rq->elevator_private = NULL; |
| 425 | flush_rq->flags = REQ_BAR_FLUSH; |
| 426 | flush_rq->rq_disk = rq->rq_disk; |
| 427 | flush_rq->rl = NULL; |
| 428 | |
| 429 | if (q->prepare_flush_fn(q, flush_rq)) { |
| 430 | flush_rq->end_io_data = rq; |
| 431 | flush_rq->end_io = blk_post_flush_end_io; |
| 432 | |
| 433 | __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0); |
| 434 | q->request_fn(q); |
| 435 | } |
| 436 | } |
| 437 | |
| 438 | static inline int blk_check_end_barrier(request_queue_t *q, struct request *rq, |
| 439 | int sectors) |
| 440 | { |
| 441 | if (sectors > rq->nr_sectors) |
| 442 | sectors = rq->nr_sectors; |
| 443 | |
| 444 | rq->nr_sectors -= sectors; |
| 445 | return rq->nr_sectors; |
| 446 | } |
| 447 | |
| 448 | static int __blk_complete_barrier_rq(request_queue_t *q, struct request *rq, |
| 449 | int sectors, int queue_locked) |
| 450 | { |
| 451 | if (q->ordered != QUEUE_ORDERED_FLUSH) |
| 452 | return 0; |
| 453 | if (!blk_fs_request(rq) || !blk_barrier_rq(rq)) |
| 454 | return 0; |
| 455 | if (blk_barrier_postflush(rq)) |
| 456 | return 0; |
| 457 | |
| 458 | if (!blk_check_end_barrier(q, rq, sectors)) { |
| 459 | unsigned long flags = 0; |
| 460 | |
| 461 | if (!queue_locked) |
| 462 | spin_lock_irqsave(q->queue_lock, flags); |
| 463 | |
| 464 | blk_start_post_flush(q, rq); |
| 465 | |
| 466 | if (!queue_locked) |
| 467 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 468 | } |
| 469 | |
| 470 | return 1; |
| 471 | } |
| 472 | |
| 473 | /** |
| 474 | * blk_complete_barrier_rq - complete possible barrier request |
| 475 | * @q: the request queue for the device |
| 476 | * @rq: the request |
| 477 | * @sectors: number of sectors to complete |
| 478 | * |
| 479 | * Description: |
| 480 | * Used in driver end_io handling to determine whether to postpone |
| 481 | * completion of a barrier request until a post flush has been done. This |
| 482 | * is the unlocked variant, used if the caller doesn't already hold the |
| 483 | * queue lock. |
| 484 | **/ |
| 485 | int blk_complete_barrier_rq(request_queue_t *q, struct request *rq, int sectors) |
| 486 | { |
| 487 | return __blk_complete_barrier_rq(q, rq, sectors, 0); |
| 488 | } |
| 489 | EXPORT_SYMBOL(blk_complete_barrier_rq); |
| 490 | |
| 491 | /** |
| 492 | * blk_complete_barrier_rq_locked - complete possible barrier request |
| 493 | * @q: the request queue for the device |
| 494 | * @rq: the request |
| 495 | * @sectors: number of sectors to complete |
| 496 | * |
| 497 | * Description: |
| 498 | * See blk_complete_barrier_rq(). This variant must be used if the caller |
| 499 | * holds the queue lock. |
| 500 | **/ |
| 501 | int blk_complete_barrier_rq_locked(request_queue_t *q, struct request *rq, |
| 502 | int sectors) |
| 503 | { |
| 504 | return __blk_complete_barrier_rq(q, rq, sectors, 1); |
| 505 | } |
| 506 | EXPORT_SYMBOL(blk_complete_barrier_rq_locked); |
| 507 | |
| 508 | /** |
| 509 | * blk_queue_bounce_limit - set bounce buffer limit for queue |
| 510 | * @q: the request queue for the device |
| 511 | * @dma_addr: bus address limit |
| 512 | * |
| 513 | * Description: |
| 514 | * Different hardware can have different requirements as to what pages |
| 515 | * it can do I/O directly to. A low level driver can call |
| 516 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce |
| 517 | * buffers for doing I/O to pages residing above @page. By default |
| 518 | * the block layer sets this to the highest numbered "low" memory page. |
| 519 | **/ |
| 520 | void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr) |
| 521 | { |
| 522 | unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT; |
| 523 | |
| 524 | /* |
| 525 | * set appropriate bounce gfp mask -- unfortunately we don't have a |
| 526 | * full 4GB zone, so we have to resort to low memory for any bounces. |
| 527 | * ISA has its own < 16MB zone. |
| 528 | */ |
| 529 | if (bounce_pfn < blk_max_low_pfn) { |
| 530 | BUG_ON(dma_addr < BLK_BOUNCE_ISA); |
| 531 | init_emergency_isa_pool(); |
| 532 | q->bounce_gfp = GFP_NOIO | GFP_DMA; |
| 533 | } else |
| 534 | q->bounce_gfp = GFP_NOIO; |
| 535 | |
| 536 | q->bounce_pfn = bounce_pfn; |
| 537 | } |
| 538 | |
| 539 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
| 540 | |
| 541 | /** |
| 542 | * blk_queue_max_sectors - set max sectors for a request for this queue |
| 543 | * @q: the request queue for the device |
| 544 | * @max_sectors: max sectors in the usual 512b unit |
| 545 | * |
| 546 | * Description: |
| 547 | * Enables a low level driver to set an upper limit on the size of |
| 548 | * received requests. |
| 549 | **/ |
| 550 | void blk_queue_max_sectors(request_queue_t *q, unsigned short max_sectors) |
| 551 | { |
| 552 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { |
| 553 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); |
| 554 | printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors); |
| 555 | } |
| 556 | |
| 557 | q->max_sectors = q->max_hw_sectors = max_sectors; |
| 558 | } |
| 559 | |
| 560 | EXPORT_SYMBOL(blk_queue_max_sectors); |
| 561 | |
| 562 | /** |
| 563 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue |
| 564 | * @q: the request queue for the device |
| 565 | * @max_segments: max number of segments |
| 566 | * |
| 567 | * Description: |
| 568 | * Enables a low level driver to set an upper limit on the number of |
| 569 | * physical data segments in a request. This would be the largest sized |
| 570 | * scatter list the driver could handle. |
| 571 | **/ |
| 572 | void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments) |
| 573 | { |
| 574 | if (!max_segments) { |
| 575 | max_segments = 1; |
| 576 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); |
| 577 | } |
| 578 | |
| 579 | q->max_phys_segments = max_segments; |
| 580 | } |
| 581 | |
| 582 | EXPORT_SYMBOL(blk_queue_max_phys_segments); |
| 583 | |
| 584 | /** |
| 585 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue |
| 586 | * @q: the request queue for the device |
| 587 | * @max_segments: max number of segments |
| 588 | * |
| 589 | * Description: |
| 590 | * Enables a low level driver to set an upper limit on the number of |
| 591 | * hw data segments in a request. This would be the largest number of |
| 592 | * address/length pairs the host adapter can actually give as once |
| 593 | * to the device. |
| 594 | **/ |
| 595 | void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments) |
| 596 | { |
| 597 | if (!max_segments) { |
| 598 | max_segments = 1; |
| 599 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); |
| 600 | } |
| 601 | |
| 602 | q->max_hw_segments = max_segments; |
| 603 | } |
| 604 | |
| 605 | EXPORT_SYMBOL(blk_queue_max_hw_segments); |
| 606 | |
| 607 | /** |
| 608 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg |
| 609 | * @q: the request queue for the device |
| 610 | * @max_size: max size of segment in bytes |
| 611 | * |
| 612 | * Description: |
| 613 | * Enables a low level driver to set an upper limit on the size of a |
| 614 | * coalesced segment |
| 615 | **/ |
| 616 | void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size) |
| 617 | { |
| 618 | if (max_size < PAGE_CACHE_SIZE) { |
| 619 | max_size = PAGE_CACHE_SIZE; |
| 620 | printk("%s: set to minimum %d\n", __FUNCTION__, max_size); |
| 621 | } |
| 622 | |
| 623 | q->max_segment_size = max_size; |
| 624 | } |
| 625 | |
| 626 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
| 627 | |
| 628 | /** |
| 629 | * blk_queue_hardsect_size - set hardware sector size for the queue |
| 630 | * @q: the request queue for the device |
| 631 | * @size: the hardware sector size, in bytes |
| 632 | * |
| 633 | * Description: |
| 634 | * This should typically be set to the lowest possible sector size |
| 635 | * that the hardware can operate on (possible without reverting to |
| 636 | * even internal read-modify-write operations). Usually the default |
| 637 | * of 512 covers most hardware. |
| 638 | **/ |
| 639 | void blk_queue_hardsect_size(request_queue_t *q, unsigned short size) |
| 640 | { |
| 641 | q->hardsect_size = size; |
| 642 | } |
| 643 | |
| 644 | EXPORT_SYMBOL(blk_queue_hardsect_size); |
| 645 | |
| 646 | /* |
| 647 | * Returns the minimum that is _not_ zero, unless both are zero. |
| 648 | */ |
| 649 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) |
| 650 | |
| 651 | /** |
| 652 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers |
| 653 | * @t: the stacking driver (top) |
| 654 | * @b: the underlying device (bottom) |
| 655 | **/ |
| 656 | void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b) |
| 657 | { |
| 658 | /* zero is "infinity" */ |
| 659 | t->max_sectors = t->max_hw_sectors = |
| 660 | min_not_zero(t->max_sectors,b->max_sectors); |
| 661 | |
| 662 | t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments); |
| 663 | t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments); |
| 664 | t->max_segment_size = min(t->max_segment_size,b->max_segment_size); |
| 665 | t->hardsect_size = max(t->hardsect_size,b->hardsect_size); |
| 666 | } |
| 667 | |
| 668 | EXPORT_SYMBOL(blk_queue_stack_limits); |
| 669 | |
| 670 | /** |
| 671 | * blk_queue_segment_boundary - set boundary rules for segment merging |
| 672 | * @q: the request queue for the device |
| 673 | * @mask: the memory boundary mask |
| 674 | **/ |
| 675 | void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask) |
| 676 | { |
| 677 | if (mask < PAGE_CACHE_SIZE - 1) { |
| 678 | mask = PAGE_CACHE_SIZE - 1; |
| 679 | printk("%s: set to minimum %lx\n", __FUNCTION__, mask); |
| 680 | } |
| 681 | |
| 682 | q->seg_boundary_mask = mask; |
| 683 | } |
| 684 | |
| 685 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
| 686 | |
| 687 | /** |
| 688 | * blk_queue_dma_alignment - set dma length and memory alignment |
| 689 | * @q: the request queue for the device |
| 690 | * @mask: alignment mask |
| 691 | * |
| 692 | * description: |
| 693 | * set required memory and length aligment for direct dma transactions. |
| 694 | * this is used when buiding direct io requests for the queue. |
| 695 | * |
| 696 | **/ |
| 697 | void blk_queue_dma_alignment(request_queue_t *q, int mask) |
| 698 | { |
| 699 | q->dma_alignment = mask; |
| 700 | } |
| 701 | |
| 702 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
| 703 | |
| 704 | /** |
| 705 | * blk_queue_find_tag - find a request by its tag and queue |
| 706 | * |
| 707 | * @q: The request queue for the device |
| 708 | * @tag: The tag of the request |
| 709 | * |
| 710 | * Notes: |
| 711 | * Should be used when a device returns a tag and you want to match |
| 712 | * it with a request. |
| 713 | * |
| 714 | * no locks need be held. |
| 715 | **/ |
| 716 | struct request *blk_queue_find_tag(request_queue_t *q, int tag) |
| 717 | { |
| 718 | struct blk_queue_tag *bqt = q->queue_tags; |
| 719 | |
| 720 | if (unlikely(bqt == NULL || tag >= bqt->real_max_depth)) |
| 721 | return NULL; |
| 722 | |
| 723 | return bqt->tag_index[tag]; |
| 724 | } |
| 725 | |
| 726 | EXPORT_SYMBOL(blk_queue_find_tag); |
| 727 | |
| 728 | /** |
| 729 | * __blk_queue_free_tags - release tag maintenance info |
| 730 | * @q: the request queue for the device |
| 731 | * |
| 732 | * Notes: |
| 733 | * blk_cleanup_queue() will take care of calling this function, if tagging |
| 734 | * has been used. So there's no need to call this directly. |
| 735 | **/ |
| 736 | static void __blk_queue_free_tags(request_queue_t *q) |
| 737 | { |
| 738 | struct blk_queue_tag *bqt = q->queue_tags; |
| 739 | |
| 740 | if (!bqt) |
| 741 | return; |
| 742 | |
| 743 | if (atomic_dec_and_test(&bqt->refcnt)) { |
| 744 | BUG_ON(bqt->busy); |
| 745 | BUG_ON(!list_empty(&bqt->busy_list)); |
| 746 | |
| 747 | kfree(bqt->tag_index); |
| 748 | bqt->tag_index = NULL; |
| 749 | |
| 750 | kfree(bqt->tag_map); |
| 751 | bqt->tag_map = NULL; |
| 752 | |
| 753 | kfree(bqt); |
| 754 | } |
| 755 | |
| 756 | q->queue_tags = NULL; |
| 757 | q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED); |
| 758 | } |
| 759 | |
| 760 | /** |
| 761 | * blk_queue_free_tags - release tag maintenance info |
| 762 | * @q: the request queue for the device |
| 763 | * |
| 764 | * Notes: |
| 765 | * This is used to disabled tagged queuing to a device, yet leave |
| 766 | * queue in function. |
| 767 | **/ |
| 768 | void blk_queue_free_tags(request_queue_t *q) |
| 769 | { |
| 770 | clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); |
| 771 | } |
| 772 | |
| 773 | EXPORT_SYMBOL(blk_queue_free_tags); |
| 774 | |
| 775 | static int |
| 776 | init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth) |
| 777 | { |
| 778 | int bits, i; |
| 779 | struct request **tag_index; |
| 780 | unsigned long *tag_map; |
| 781 | |
| 782 | if (depth > q->nr_requests * 2) { |
| 783 | depth = q->nr_requests * 2; |
| 784 | printk(KERN_ERR "%s: adjusted depth to %d\n", |
| 785 | __FUNCTION__, depth); |
| 786 | } |
| 787 | |
| 788 | tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC); |
| 789 | if (!tag_index) |
| 790 | goto fail; |
| 791 | |
| 792 | bits = (depth / BLK_TAGS_PER_LONG) + 1; |
| 793 | tag_map = kmalloc(bits * sizeof(unsigned long), GFP_ATOMIC); |
| 794 | if (!tag_map) |
| 795 | goto fail; |
| 796 | |
| 797 | memset(tag_index, 0, depth * sizeof(struct request *)); |
| 798 | memset(tag_map, 0, bits * sizeof(unsigned long)); |
| 799 | tags->max_depth = depth; |
| 800 | tags->real_max_depth = bits * BITS_PER_LONG; |
| 801 | tags->tag_index = tag_index; |
| 802 | tags->tag_map = tag_map; |
| 803 | |
| 804 | /* |
| 805 | * set the upper bits if the depth isn't a multiple of the word size |
| 806 | */ |
| 807 | for (i = depth; i < bits * BLK_TAGS_PER_LONG; i++) |
| 808 | __set_bit(i, tag_map); |
| 809 | |
| 810 | return 0; |
| 811 | fail: |
| 812 | kfree(tag_index); |
| 813 | return -ENOMEM; |
| 814 | } |
| 815 | |
| 816 | /** |
| 817 | * blk_queue_init_tags - initialize the queue tag info |
| 818 | * @q: the request queue for the device |
| 819 | * @depth: the maximum queue depth supported |
| 820 | * @tags: the tag to use |
| 821 | **/ |
| 822 | int blk_queue_init_tags(request_queue_t *q, int depth, |
| 823 | struct blk_queue_tag *tags) |
| 824 | { |
| 825 | int rc; |
| 826 | |
| 827 | BUG_ON(tags && q->queue_tags && tags != q->queue_tags); |
| 828 | |
| 829 | if (!tags && !q->queue_tags) { |
| 830 | tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC); |
| 831 | if (!tags) |
| 832 | goto fail; |
| 833 | |
| 834 | if (init_tag_map(q, tags, depth)) |
| 835 | goto fail; |
| 836 | |
| 837 | INIT_LIST_HEAD(&tags->busy_list); |
| 838 | tags->busy = 0; |
| 839 | atomic_set(&tags->refcnt, 1); |
| 840 | } else if (q->queue_tags) { |
| 841 | if ((rc = blk_queue_resize_tags(q, depth))) |
| 842 | return rc; |
| 843 | set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); |
| 844 | return 0; |
| 845 | } else |
| 846 | atomic_inc(&tags->refcnt); |
| 847 | |
| 848 | /* |
| 849 | * assign it, all done |
| 850 | */ |
| 851 | q->queue_tags = tags; |
| 852 | q->queue_flags |= (1 << QUEUE_FLAG_QUEUED); |
| 853 | return 0; |
| 854 | fail: |
| 855 | kfree(tags); |
| 856 | return -ENOMEM; |
| 857 | } |
| 858 | |
| 859 | EXPORT_SYMBOL(blk_queue_init_tags); |
| 860 | |
| 861 | /** |
| 862 | * blk_queue_resize_tags - change the queueing depth |
| 863 | * @q: the request queue for the device |
| 864 | * @new_depth: the new max command queueing depth |
| 865 | * |
| 866 | * Notes: |
| 867 | * Must be called with the queue lock held. |
| 868 | **/ |
| 869 | int blk_queue_resize_tags(request_queue_t *q, int new_depth) |
| 870 | { |
| 871 | struct blk_queue_tag *bqt = q->queue_tags; |
| 872 | struct request **tag_index; |
| 873 | unsigned long *tag_map; |
| 874 | int bits, max_depth; |
| 875 | |
| 876 | if (!bqt) |
| 877 | return -ENXIO; |
| 878 | |
| 879 | /* |
| 880 | * don't bother sizing down |
| 881 | */ |
| 882 | if (new_depth <= bqt->real_max_depth) { |
| 883 | bqt->max_depth = new_depth; |
| 884 | return 0; |
| 885 | } |
| 886 | |
| 887 | /* |
| 888 | * save the old state info, so we can copy it back |
| 889 | */ |
| 890 | tag_index = bqt->tag_index; |
| 891 | tag_map = bqt->tag_map; |
| 892 | max_depth = bqt->real_max_depth; |
| 893 | |
| 894 | if (init_tag_map(q, bqt, new_depth)) |
| 895 | return -ENOMEM; |
| 896 | |
| 897 | memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); |
| 898 | bits = max_depth / BLK_TAGS_PER_LONG; |
| 899 | memcpy(bqt->tag_map, tag_map, bits * sizeof(unsigned long)); |
| 900 | |
| 901 | kfree(tag_index); |
| 902 | kfree(tag_map); |
| 903 | return 0; |
| 904 | } |
| 905 | |
| 906 | EXPORT_SYMBOL(blk_queue_resize_tags); |
| 907 | |
| 908 | /** |
| 909 | * blk_queue_end_tag - end tag operations for a request |
| 910 | * @q: the request queue for the device |
| 911 | * @rq: the request that has completed |
| 912 | * |
| 913 | * Description: |
| 914 | * Typically called when end_that_request_first() returns 0, meaning |
| 915 | * all transfers have been done for a request. It's important to call |
| 916 | * this function before end_that_request_last(), as that will put the |
| 917 | * request back on the free list thus corrupting the internal tag list. |
| 918 | * |
| 919 | * Notes: |
| 920 | * queue lock must be held. |
| 921 | **/ |
| 922 | void blk_queue_end_tag(request_queue_t *q, struct request *rq) |
| 923 | { |
| 924 | struct blk_queue_tag *bqt = q->queue_tags; |
| 925 | int tag = rq->tag; |
| 926 | |
| 927 | BUG_ON(tag == -1); |
| 928 | |
| 929 | if (unlikely(tag >= bqt->real_max_depth)) |
| 930 | return; |
| 931 | |
| 932 | if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) { |
| 933 | printk("attempt to clear non-busy tag (%d)\n", tag); |
| 934 | return; |
| 935 | } |
| 936 | |
| 937 | list_del_init(&rq->queuelist); |
| 938 | rq->flags &= ~REQ_QUEUED; |
| 939 | rq->tag = -1; |
| 940 | |
| 941 | if (unlikely(bqt->tag_index[tag] == NULL)) |
| 942 | printk("tag %d is missing\n", tag); |
| 943 | |
| 944 | bqt->tag_index[tag] = NULL; |
| 945 | bqt->busy--; |
| 946 | } |
| 947 | |
| 948 | EXPORT_SYMBOL(blk_queue_end_tag); |
| 949 | |
| 950 | /** |
| 951 | * blk_queue_start_tag - find a free tag and assign it |
| 952 | * @q: the request queue for the device |
| 953 | * @rq: the block request that needs tagging |
| 954 | * |
| 955 | * Description: |
| 956 | * This can either be used as a stand-alone helper, or possibly be |
| 957 | * assigned as the queue &prep_rq_fn (in which case &struct request |
| 958 | * automagically gets a tag assigned). Note that this function |
| 959 | * assumes that any type of request can be queued! if this is not |
| 960 | * true for your device, you must check the request type before |
| 961 | * calling this function. The request will also be removed from |
| 962 | * the request queue, so it's the drivers responsibility to readd |
| 963 | * it if it should need to be restarted for some reason. |
| 964 | * |
| 965 | * Notes: |
| 966 | * queue lock must be held. |
| 967 | **/ |
| 968 | int blk_queue_start_tag(request_queue_t *q, struct request *rq) |
| 969 | { |
| 970 | struct blk_queue_tag *bqt = q->queue_tags; |
| 971 | unsigned long *map = bqt->tag_map; |
| 972 | int tag = 0; |
| 973 | |
| 974 | if (unlikely((rq->flags & REQ_QUEUED))) { |
| 975 | printk(KERN_ERR |
| 976 | "request %p for device [%s] already tagged %d", |
| 977 | rq, rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); |
| 978 | BUG(); |
| 979 | } |
| 980 | |
| 981 | for (map = bqt->tag_map; *map == -1UL; map++) { |
| 982 | tag += BLK_TAGS_PER_LONG; |
| 983 | |
| 984 | if (tag >= bqt->max_depth) |
| 985 | return 1; |
| 986 | } |
| 987 | |
| 988 | tag += ffz(*map); |
| 989 | __set_bit(tag, bqt->tag_map); |
| 990 | |
| 991 | rq->flags |= REQ_QUEUED; |
| 992 | rq->tag = tag; |
| 993 | bqt->tag_index[tag] = rq; |
| 994 | blkdev_dequeue_request(rq); |
| 995 | list_add(&rq->queuelist, &bqt->busy_list); |
| 996 | bqt->busy++; |
| 997 | return 0; |
| 998 | } |
| 999 | |
| 1000 | EXPORT_SYMBOL(blk_queue_start_tag); |
| 1001 | |
| 1002 | /** |
| 1003 | * blk_queue_invalidate_tags - invalidate all pending tags |
| 1004 | * @q: the request queue for the device |
| 1005 | * |
| 1006 | * Description: |
| 1007 | * Hardware conditions may dictate a need to stop all pending requests. |
| 1008 | * In this case, we will safely clear the block side of the tag queue and |
| 1009 | * readd all requests to the request queue in the right order. |
| 1010 | * |
| 1011 | * Notes: |
| 1012 | * queue lock must be held. |
| 1013 | **/ |
| 1014 | void blk_queue_invalidate_tags(request_queue_t *q) |
| 1015 | { |
| 1016 | struct blk_queue_tag *bqt = q->queue_tags; |
| 1017 | struct list_head *tmp, *n; |
| 1018 | struct request *rq; |
| 1019 | |
| 1020 | list_for_each_safe(tmp, n, &bqt->busy_list) { |
| 1021 | rq = list_entry_rq(tmp); |
| 1022 | |
| 1023 | if (rq->tag == -1) { |
| 1024 | printk("bad tag found on list\n"); |
| 1025 | list_del_init(&rq->queuelist); |
| 1026 | rq->flags &= ~REQ_QUEUED; |
| 1027 | } else |
| 1028 | blk_queue_end_tag(q, rq); |
| 1029 | |
| 1030 | rq->flags &= ~REQ_STARTED; |
| 1031 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); |
| 1032 | } |
| 1033 | } |
| 1034 | |
| 1035 | EXPORT_SYMBOL(blk_queue_invalidate_tags); |
| 1036 | |
| 1037 | static char *rq_flags[] = { |
| 1038 | "REQ_RW", |
| 1039 | "REQ_FAILFAST", |
| 1040 | "REQ_SOFTBARRIER", |
| 1041 | "REQ_HARDBARRIER", |
| 1042 | "REQ_CMD", |
| 1043 | "REQ_NOMERGE", |
| 1044 | "REQ_STARTED", |
| 1045 | "REQ_DONTPREP", |
| 1046 | "REQ_QUEUED", |
| 1047 | "REQ_PC", |
| 1048 | "REQ_BLOCK_PC", |
| 1049 | "REQ_SENSE", |
| 1050 | "REQ_FAILED", |
| 1051 | "REQ_QUIET", |
| 1052 | "REQ_SPECIAL", |
| 1053 | "REQ_DRIVE_CMD", |
| 1054 | "REQ_DRIVE_TASK", |
| 1055 | "REQ_DRIVE_TASKFILE", |
| 1056 | "REQ_PREEMPT", |
| 1057 | "REQ_PM_SUSPEND", |
| 1058 | "REQ_PM_RESUME", |
| 1059 | "REQ_PM_SHUTDOWN", |
| 1060 | }; |
| 1061 | |
| 1062 | void blk_dump_rq_flags(struct request *rq, char *msg) |
| 1063 | { |
| 1064 | int bit; |
| 1065 | |
| 1066 | printk("%s: dev %s: flags = ", msg, |
| 1067 | rq->rq_disk ? rq->rq_disk->disk_name : "?"); |
| 1068 | bit = 0; |
| 1069 | do { |
| 1070 | if (rq->flags & (1 << bit)) |
| 1071 | printk("%s ", rq_flags[bit]); |
| 1072 | bit++; |
| 1073 | } while (bit < __REQ_NR_BITS); |
| 1074 | |
| 1075 | printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector, |
| 1076 | rq->nr_sectors, |
| 1077 | rq->current_nr_sectors); |
| 1078 | printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len); |
| 1079 | |
| 1080 | if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) { |
| 1081 | printk("cdb: "); |
| 1082 | for (bit = 0; bit < sizeof(rq->cmd); bit++) |
| 1083 | printk("%02x ", rq->cmd[bit]); |
| 1084 | printk("\n"); |
| 1085 | } |
| 1086 | } |
| 1087 | |
| 1088 | EXPORT_SYMBOL(blk_dump_rq_flags); |
| 1089 | |
| 1090 | void blk_recount_segments(request_queue_t *q, struct bio *bio) |
| 1091 | { |
| 1092 | struct bio_vec *bv, *bvprv = NULL; |
| 1093 | int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster; |
| 1094 | int high, highprv = 1; |
| 1095 | |
| 1096 | if (unlikely(!bio->bi_io_vec)) |
| 1097 | return; |
| 1098 | |
| 1099 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); |
| 1100 | hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0; |
| 1101 | bio_for_each_segment(bv, bio, i) { |
| 1102 | /* |
| 1103 | * the trick here is making sure that a high page is never |
| 1104 | * considered part of another segment, since that might |
| 1105 | * change with the bounce page. |
| 1106 | */ |
| 1107 | high = page_to_pfn(bv->bv_page) >= q->bounce_pfn; |
| 1108 | if (high || highprv) |
| 1109 | goto new_hw_segment; |
| 1110 | if (cluster) { |
| 1111 | if (seg_size + bv->bv_len > q->max_segment_size) |
| 1112 | goto new_segment; |
| 1113 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv)) |
| 1114 | goto new_segment; |
| 1115 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv)) |
| 1116 | goto new_segment; |
| 1117 | if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) |
| 1118 | goto new_hw_segment; |
| 1119 | |
| 1120 | seg_size += bv->bv_len; |
| 1121 | hw_seg_size += bv->bv_len; |
| 1122 | bvprv = bv; |
| 1123 | continue; |
| 1124 | } |
| 1125 | new_segment: |
| 1126 | if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) && |
| 1127 | !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) { |
| 1128 | hw_seg_size += bv->bv_len; |
| 1129 | } else { |
| 1130 | new_hw_segment: |
| 1131 | if (hw_seg_size > bio->bi_hw_front_size) |
| 1132 | bio->bi_hw_front_size = hw_seg_size; |
| 1133 | hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len; |
| 1134 | nr_hw_segs++; |
| 1135 | } |
| 1136 | |
| 1137 | nr_phys_segs++; |
| 1138 | bvprv = bv; |
| 1139 | seg_size = bv->bv_len; |
| 1140 | highprv = high; |
| 1141 | } |
| 1142 | if (hw_seg_size > bio->bi_hw_back_size) |
| 1143 | bio->bi_hw_back_size = hw_seg_size; |
| 1144 | if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size) |
| 1145 | bio->bi_hw_front_size = hw_seg_size; |
| 1146 | bio->bi_phys_segments = nr_phys_segs; |
| 1147 | bio->bi_hw_segments = nr_hw_segs; |
| 1148 | bio->bi_flags |= (1 << BIO_SEG_VALID); |
| 1149 | } |
| 1150 | |
| 1151 | |
| 1152 | int blk_phys_contig_segment(request_queue_t *q, struct bio *bio, |
| 1153 | struct bio *nxt) |
| 1154 | { |
| 1155 | if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) |
| 1156 | return 0; |
| 1157 | |
| 1158 | if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt))) |
| 1159 | return 0; |
| 1160 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) |
| 1161 | return 0; |
| 1162 | |
| 1163 | /* |
| 1164 | * bio and nxt are contigous in memory, check if the queue allows |
| 1165 | * these two to be merged into one |
| 1166 | */ |
| 1167 | if (BIO_SEG_BOUNDARY(q, bio, nxt)) |
| 1168 | return 1; |
| 1169 | |
| 1170 | return 0; |
| 1171 | } |
| 1172 | |
| 1173 | EXPORT_SYMBOL(blk_phys_contig_segment); |
| 1174 | |
| 1175 | int blk_hw_contig_segment(request_queue_t *q, struct bio *bio, |
| 1176 | struct bio *nxt) |
| 1177 | { |
| 1178 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
| 1179 | blk_recount_segments(q, bio); |
| 1180 | if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID))) |
| 1181 | blk_recount_segments(q, nxt); |
| 1182 | if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) || |
| 1183 | BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size)) |
| 1184 | return 0; |
| 1185 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) |
| 1186 | return 0; |
| 1187 | |
| 1188 | return 1; |
| 1189 | } |
| 1190 | |
| 1191 | EXPORT_SYMBOL(blk_hw_contig_segment); |
| 1192 | |
| 1193 | /* |
| 1194 | * map a request to scatterlist, return number of sg entries setup. Caller |
| 1195 | * must make sure sg can hold rq->nr_phys_segments entries |
| 1196 | */ |
| 1197 | int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg) |
| 1198 | { |
| 1199 | struct bio_vec *bvec, *bvprv; |
| 1200 | struct bio *bio; |
| 1201 | int nsegs, i, cluster; |
| 1202 | |
| 1203 | nsegs = 0; |
| 1204 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); |
| 1205 | |
| 1206 | /* |
| 1207 | * for each bio in rq |
| 1208 | */ |
| 1209 | bvprv = NULL; |
| 1210 | rq_for_each_bio(bio, rq) { |
| 1211 | /* |
| 1212 | * for each segment in bio |
| 1213 | */ |
| 1214 | bio_for_each_segment(bvec, bio, i) { |
| 1215 | int nbytes = bvec->bv_len; |
| 1216 | |
| 1217 | if (bvprv && cluster) { |
| 1218 | if (sg[nsegs - 1].length + nbytes > q->max_segment_size) |
| 1219 | goto new_segment; |
| 1220 | |
| 1221 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) |
| 1222 | goto new_segment; |
| 1223 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) |
| 1224 | goto new_segment; |
| 1225 | |
| 1226 | sg[nsegs - 1].length += nbytes; |
| 1227 | } else { |
| 1228 | new_segment: |
| 1229 | memset(&sg[nsegs],0,sizeof(struct scatterlist)); |
| 1230 | sg[nsegs].page = bvec->bv_page; |
| 1231 | sg[nsegs].length = nbytes; |
| 1232 | sg[nsegs].offset = bvec->bv_offset; |
| 1233 | |
| 1234 | nsegs++; |
| 1235 | } |
| 1236 | bvprv = bvec; |
| 1237 | } /* segments in bio */ |
| 1238 | } /* bios in rq */ |
| 1239 | |
| 1240 | return nsegs; |
| 1241 | } |
| 1242 | |
| 1243 | EXPORT_SYMBOL(blk_rq_map_sg); |
| 1244 | |
| 1245 | /* |
| 1246 | * the standard queue merge functions, can be overridden with device |
| 1247 | * specific ones if so desired |
| 1248 | */ |
| 1249 | |
| 1250 | static inline int ll_new_mergeable(request_queue_t *q, |
| 1251 | struct request *req, |
| 1252 | struct bio *bio) |
| 1253 | { |
| 1254 | int nr_phys_segs = bio_phys_segments(q, bio); |
| 1255 | |
| 1256 | if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { |
| 1257 | req->flags |= REQ_NOMERGE; |
| 1258 | if (req == q->last_merge) |
| 1259 | q->last_merge = NULL; |
| 1260 | return 0; |
| 1261 | } |
| 1262 | |
| 1263 | /* |
| 1264 | * A hw segment is just getting larger, bump just the phys |
| 1265 | * counter. |
| 1266 | */ |
| 1267 | req->nr_phys_segments += nr_phys_segs; |
| 1268 | return 1; |
| 1269 | } |
| 1270 | |
| 1271 | static inline int ll_new_hw_segment(request_queue_t *q, |
| 1272 | struct request *req, |
| 1273 | struct bio *bio) |
| 1274 | { |
| 1275 | int nr_hw_segs = bio_hw_segments(q, bio); |
| 1276 | int nr_phys_segs = bio_phys_segments(q, bio); |
| 1277 | |
| 1278 | if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments |
| 1279 | || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { |
| 1280 | req->flags |= REQ_NOMERGE; |
| 1281 | if (req == q->last_merge) |
| 1282 | q->last_merge = NULL; |
| 1283 | return 0; |
| 1284 | } |
| 1285 | |
| 1286 | /* |
| 1287 | * This will form the start of a new hw segment. Bump both |
| 1288 | * counters. |
| 1289 | */ |
| 1290 | req->nr_hw_segments += nr_hw_segs; |
| 1291 | req->nr_phys_segments += nr_phys_segs; |
| 1292 | return 1; |
| 1293 | } |
| 1294 | |
| 1295 | static int ll_back_merge_fn(request_queue_t *q, struct request *req, |
| 1296 | struct bio *bio) |
| 1297 | { |
| 1298 | int len; |
| 1299 | |
| 1300 | if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) { |
| 1301 | req->flags |= REQ_NOMERGE; |
| 1302 | if (req == q->last_merge) |
| 1303 | q->last_merge = NULL; |
| 1304 | return 0; |
| 1305 | } |
| 1306 | if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID))) |
| 1307 | blk_recount_segments(q, req->biotail); |
| 1308 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
| 1309 | blk_recount_segments(q, bio); |
| 1310 | len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size; |
| 1311 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) && |
| 1312 | !BIOVEC_VIRT_OVERSIZE(len)) { |
| 1313 | int mergeable = ll_new_mergeable(q, req, bio); |
| 1314 | |
| 1315 | if (mergeable) { |
| 1316 | if (req->nr_hw_segments == 1) |
| 1317 | req->bio->bi_hw_front_size = len; |
| 1318 | if (bio->bi_hw_segments == 1) |
| 1319 | bio->bi_hw_back_size = len; |
| 1320 | } |
| 1321 | return mergeable; |
| 1322 | } |
| 1323 | |
| 1324 | return ll_new_hw_segment(q, req, bio); |
| 1325 | } |
| 1326 | |
| 1327 | static int ll_front_merge_fn(request_queue_t *q, struct request *req, |
| 1328 | struct bio *bio) |
| 1329 | { |
| 1330 | int len; |
| 1331 | |
| 1332 | if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) { |
| 1333 | req->flags |= REQ_NOMERGE; |
| 1334 | if (req == q->last_merge) |
| 1335 | q->last_merge = NULL; |
| 1336 | return 0; |
| 1337 | } |
| 1338 | len = bio->bi_hw_back_size + req->bio->bi_hw_front_size; |
| 1339 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
| 1340 | blk_recount_segments(q, bio); |
| 1341 | if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID))) |
| 1342 | blk_recount_segments(q, req->bio); |
| 1343 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) && |
| 1344 | !BIOVEC_VIRT_OVERSIZE(len)) { |
| 1345 | int mergeable = ll_new_mergeable(q, req, bio); |
| 1346 | |
| 1347 | if (mergeable) { |
| 1348 | if (bio->bi_hw_segments == 1) |
| 1349 | bio->bi_hw_front_size = len; |
| 1350 | if (req->nr_hw_segments == 1) |
| 1351 | req->biotail->bi_hw_back_size = len; |
| 1352 | } |
| 1353 | return mergeable; |
| 1354 | } |
| 1355 | |
| 1356 | return ll_new_hw_segment(q, req, bio); |
| 1357 | } |
| 1358 | |
| 1359 | static int ll_merge_requests_fn(request_queue_t *q, struct request *req, |
| 1360 | struct request *next) |
| 1361 | { |
| 1362 | int total_phys_segments = req->nr_phys_segments +next->nr_phys_segments; |
| 1363 | int total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; |
| 1364 | |
| 1365 | /* |
| 1366 | * First check if the either of the requests are re-queued |
| 1367 | * requests. Can't merge them if they are. |
| 1368 | */ |
| 1369 | if (req->special || next->special) |
| 1370 | return 0; |
| 1371 | |
| 1372 | /* |
| 1373 | * Will it become to large? |
| 1374 | */ |
| 1375 | if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) |
| 1376 | return 0; |
| 1377 | |
| 1378 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; |
| 1379 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) |
| 1380 | total_phys_segments--; |
| 1381 | |
| 1382 | if (total_phys_segments > q->max_phys_segments) |
| 1383 | return 0; |
| 1384 | |
| 1385 | total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; |
| 1386 | if (blk_hw_contig_segment(q, req->biotail, next->bio)) { |
| 1387 | int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size; |
| 1388 | /* |
| 1389 | * propagate the combined length to the end of the requests |
| 1390 | */ |
| 1391 | if (req->nr_hw_segments == 1) |
| 1392 | req->bio->bi_hw_front_size = len; |
| 1393 | if (next->nr_hw_segments == 1) |
| 1394 | next->biotail->bi_hw_back_size = len; |
| 1395 | total_hw_segments--; |
| 1396 | } |
| 1397 | |
| 1398 | if (total_hw_segments > q->max_hw_segments) |
| 1399 | return 0; |
| 1400 | |
| 1401 | /* Merge is OK... */ |
| 1402 | req->nr_phys_segments = total_phys_segments; |
| 1403 | req->nr_hw_segments = total_hw_segments; |
| 1404 | return 1; |
| 1405 | } |
| 1406 | |
| 1407 | /* |
| 1408 | * "plug" the device if there are no outstanding requests: this will |
| 1409 | * force the transfer to start only after we have put all the requests |
| 1410 | * on the list. |
| 1411 | * |
| 1412 | * This is called with interrupts off and no requests on the queue and |
| 1413 | * with the queue lock held. |
| 1414 | */ |
| 1415 | void blk_plug_device(request_queue_t *q) |
| 1416 | { |
| 1417 | WARN_ON(!irqs_disabled()); |
| 1418 | |
| 1419 | /* |
| 1420 | * don't plug a stopped queue, it must be paired with blk_start_queue() |
| 1421 | * which will restart the queueing |
| 1422 | */ |
| 1423 | if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)) |
| 1424 | return; |
| 1425 | |
| 1426 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) |
| 1427 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
| 1428 | } |
| 1429 | |
| 1430 | EXPORT_SYMBOL(blk_plug_device); |
| 1431 | |
| 1432 | /* |
| 1433 | * remove the queue from the plugged list, if present. called with |
| 1434 | * queue lock held and interrupts disabled. |
| 1435 | */ |
| 1436 | int blk_remove_plug(request_queue_t *q) |
| 1437 | { |
| 1438 | WARN_ON(!irqs_disabled()); |
| 1439 | |
| 1440 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) |
| 1441 | return 0; |
| 1442 | |
| 1443 | del_timer(&q->unplug_timer); |
| 1444 | return 1; |
| 1445 | } |
| 1446 | |
| 1447 | EXPORT_SYMBOL(blk_remove_plug); |
| 1448 | |
| 1449 | /* |
| 1450 | * remove the plug and let it rip.. |
| 1451 | */ |
| 1452 | void __generic_unplug_device(request_queue_t *q) |
| 1453 | { |
| 1454 | if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)) |
| 1455 | return; |
| 1456 | |
| 1457 | if (!blk_remove_plug(q)) |
| 1458 | return; |
| 1459 | |
| 1460 | /* |
| 1461 | * was plugged, fire request_fn if queue has stuff to do |
| 1462 | */ |
| 1463 | if (elv_next_request(q)) |
| 1464 | q->request_fn(q); |
| 1465 | } |
| 1466 | EXPORT_SYMBOL(__generic_unplug_device); |
| 1467 | |
| 1468 | /** |
| 1469 | * generic_unplug_device - fire a request queue |
| 1470 | * @q: The &request_queue_t in question |
| 1471 | * |
| 1472 | * Description: |
| 1473 | * Linux uses plugging to build bigger requests queues before letting |
| 1474 | * the device have at them. If a queue is plugged, the I/O scheduler |
| 1475 | * is still adding and merging requests on the queue. Once the queue |
| 1476 | * gets unplugged, the request_fn defined for the queue is invoked and |
| 1477 | * transfers started. |
| 1478 | **/ |
| 1479 | void generic_unplug_device(request_queue_t *q) |
| 1480 | { |
| 1481 | spin_lock_irq(q->queue_lock); |
| 1482 | __generic_unplug_device(q); |
| 1483 | spin_unlock_irq(q->queue_lock); |
| 1484 | } |
| 1485 | EXPORT_SYMBOL(generic_unplug_device); |
| 1486 | |
| 1487 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, |
| 1488 | struct page *page) |
| 1489 | { |
| 1490 | request_queue_t *q = bdi->unplug_io_data; |
| 1491 | |
| 1492 | /* |
| 1493 | * devices don't necessarily have an ->unplug_fn defined |
| 1494 | */ |
| 1495 | if (q->unplug_fn) |
| 1496 | q->unplug_fn(q); |
| 1497 | } |
| 1498 | |
| 1499 | static void blk_unplug_work(void *data) |
| 1500 | { |
| 1501 | request_queue_t *q = data; |
| 1502 | |
| 1503 | q->unplug_fn(q); |
| 1504 | } |
| 1505 | |
| 1506 | static void blk_unplug_timeout(unsigned long data) |
| 1507 | { |
| 1508 | request_queue_t *q = (request_queue_t *)data; |
| 1509 | |
| 1510 | kblockd_schedule_work(&q->unplug_work); |
| 1511 | } |
| 1512 | |
| 1513 | /** |
| 1514 | * blk_start_queue - restart a previously stopped queue |
| 1515 | * @q: The &request_queue_t in question |
| 1516 | * |
| 1517 | * Description: |
| 1518 | * blk_start_queue() will clear the stop flag on the queue, and call |
| 1519 | * the request_fn for the queue if it was in a stopped state when |
| 1520 | * entered. Also see blk_stop_queue(). Queue lock must be held. |
| 1521 | **/ |
| 1522 | void blk_start_queue(request_queue_t *q) |
| 1523 | { |
| 1524 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); |
| 1525 | |
| 1526 | /* |
| 1527 | * one level of recursion is ok and is much faster than kicking |
| 1528 | * the unplug handling |
| 1529 | */ |
| 1530 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { |
| 1531 | q->request_fn(q); |
| 1532 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); |
| 1533 | } else { |
| 1534 | blk_plug_device(q); |
| 1535 | kblockd_schedule_work(&q->unplug_work); |
| 1536 | } |
| 1537 | } |
| 1538 | |
| 1539 | EXPORT_SYMBOL(blk_start_queue); |
| 1540 | |
| 1541 | /** |
| 1542 | * blk_stop_queue - stop a queue |
| 1543 | * @q: The &request_queue_t in question |
| 1544 | * |
| 1545 | * Description: |
| 1546 | * The Linux block layer assumes that a block driver will consume all |
| 1547 | * entries on the request queue when the request_fn strategy is called. |
| 1548 | * Often this will not happen, because of hardware limitations (queue |
| 1549 | * depth settings). If a device driver gets a 'queue full' response, |
| 1550 | * or if it simply chooses not to queue more I/O at one point, it can |
| 1551 | * call this function to prevent the request_fn from being called until |
| 1552 | * the driver has signalled it's ready to go again. This happens by calling |
| 1553 | * blk_start_queue() to restart queue operations. Queue lock must be held. |
| 1554 | **/ |
| 1555 | void blk_stop_queue(request_queue_t *q) |
| 1556 | { |
| 1557 | blk_remove_plug(q); |
| 1558 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); |
| 1559 | } |
| 1560 | EXPORT_SYMBOL(blk_stop_queue); |
| 1561 | |
| 1562 | /** |
| 1563 | * blk_sync_queue - cancel any pending callbacks on a queue |
| 1564 | * @q: the queue |
| 1565 | * |
| 1566 | * Description: |
| 1567 | * The block layer may perform asynchronous callback activity |
| 1568 | * on a queue, such as calling the unplug function after a timeout. |
| 1569 | * A block device may call blk_sync_queue to ensure that any |
| 1570 | * such activity is cancelled, thus allowing it to release resources |
| 1571 | * the the callbacks might use. The caller must already have made sure |
| 1572 | * that its ->make_request_fn will not re-add plugging prior to calling |
| 1573 | * this function. |
| 1574 | * |
| 1575 | */ |
| 1576 | void blk_sync_queue(struct request_queue *q) |
| 1577 | { |
| 1578 | del_timer_sync(&q->unplug_timer); |
| 1579 | kblockd_flush(); |
| 1580 | } |
| 1581 | EXPORT_SYMBOL(blk_sync_queue); |
| 1582 | |
| 1583 | /** |
| 1584 | * blk_run_queue - run a single device queue |
| 1585 | * @q: The queue to run |
| 1586 | */ |
| 1587 | void blk_run_queue(struct request_queue *q) |
| 1588 | { |
| 1589 | unsigned long flags; |
| 1590 | |
| 1591 | spin_lock_irqsave(q->queue_lock, flags); |
| 1592 | blk_remove_plug(q); |
Ken Chen | a299738 | 2005-04-16 15:25:43 -0700 | [diff] [blame] | 1593 | if (!elv_queue_empty(q)) |
| 1594 | q->request_fn(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1595 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 1596 | } |
| 1597 | EXPORT_SYMBOL(blk_run_queue); |
| 1598 | |
| 1599 | /** |
| 1600 | * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed |
| 1601 | * @q: the request queue to be released |
| 1602 | * |
| 1603 | * Description: |
| 1604 | * blk_cleanup_queue is the pair to blk_init_queue() or |
| 1605 | * blk_queue_make_request(). It should be called when a request queue is |
| 1606 | * being released; typically when a block device is being de-registered. |
| 1607 | * Currently, its primary task it to free all the &struct request |
| 1608 | * structures that were allocated to the queue and the queue itself. |
| 1609 | * |
| 1610 | * Caveat: |
| 1611 | * Hopefully the low level driver will have finished any |
| 1612 | * outstanding requests first... |
| 1613 | **/ |
| 1614 | void blk_cleanup_queue(request_queue_t * q) |
| 1615 | { |
| 1616 | struct request_list *rl = &q->rq; |
| 1617 | |
| 1618 | if (!atomic_dec_and_test(&q->refcnt)) |
| 1619 | return; |
| 1620 | |
| 1621 | if (q->elevator) |
| 1622 | elevator_exit(q->elevator); |
| 1623 | |
| 1624 | blk_sync_queue(q); |
| 1625 | |
| 1626 | if (rl->rq_pool) |
| 1627 | mempool_destroy(rl->rq_pool); |
| 1628 | |
| 1629 | if (q->queue_tags) |
| 1630 | __blk_queue_free_tags(q); |
| 1631 | |
| 1632 | blk_queue_ordered(q, QUEUE_ORDERED_NONE); |
| 1633 | |
| 1634 | kmem_cache_free(requestq_cachep, q); |
| 1635 | } |
| 1636 | |
| 1637 | EXPORT_SYMBOL(blk_cleanup_queue); |
| 1638 | |
| 1639 | static int blk_init_free_list(request_queue_t *q) |
| 1640 | { |
| 1641 | struct request_list *rl = &q->rq; |
| 1642 | |
| 1643 | rl->count[READ] = rl->count[WRITE] = 0; |
| 1644 | rl->starved[READ] = rl->starved[WRITE] = 0; |
| 1645 | init_waitqueue_head(&rl->wait[READ]); |
| 1646 | init_waitqueue_head(&rl->wait[WRITE]); |
| 1647 | init_waitqueue_head(&rl->drain); |
| 1648 | |
| 1649 | rl->rq_pool = mempool_create(BLKDEV_MIN_RQ, mempool_alloc_slab, mempool_free_slab, request_cachep); |
| 1650 | |
| 1651 | if (!rl->rq_pool) |
| 1652 | return -ENOMEM; |
| 1653 | |
| 1654 | return 0; |
| 1655 | } |
| 1656 | |
| 1657 | static int __make_request(request_queue_t *, struct bio *); |
| 1658 | |
| 1659 | request_queue_t *blk_alloc_queue(int gfp_mask) |
| 1660 | { |
| 1661 | request_queue_t *q = kmem_cache_alloc(requestq_cachep, gfp_mask); |
| 1662 | |
| 1663 | if (!q) |
| 1664 | return NULL; |
| 1665 | |
| 1666 | memset(q, 0, sizeof(*q)); |
| 1667 | init_timer(&q->unplug_timer); |
| 1668 | atomic_set(&q->refcnt, 1); |
| 1669 | |
| 1670 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
| 1671 | q->backing_dev_info.unplug_io_data = q; |
| 1672 | |
| 1673 | return q; |
| 1674 | } |
| 1675 | |
| 1676 | EXPORT_SYMBOL(blk_alloc_queue); |
| 1677 | |
| 1678 | /** |
| 1679 | * blk_init_queue - prepare a request queue for use with a block device |
| 1680 | * @rfn: The function to be called to process requests that have been |
| 1681 | * placed on the queue. |
| 1682 | * @lock: Request queue spin lock |
| 1683 | * |
| 1684 | * Description: |
| 1685 | * If a block device wishes to use the standard request handling procedures, |
| 1686 | * which sorts requests and coalesces adjacent requests, then it must |
| 1687 | * call blk_init_queue(). The function @rfn will be called when there |
| 1688 | * are requests on the queue that need to be processed. If the device |
| 1689 | * supports plugging, then @rfn may not be called immediately when requests |
| 1690 | * are available on the queue, but may be called at some time later instead. |
| 1691 | * Plugged queues are generally unplugged when a buffer belonging to one |
| 1692 | * of the requests on the queue is needed, or due to memory pressure. |
| 1693 | * |
| 1694 | * @rfn is not required, or even expected, to remove all requests off the |
| 1695 | * queue, but only as many as it can handle at a time. If it does leave |
| 1696 | * requests on the queue, it is responsible for arranging that the requests |
| 1697 | * get dealt with eventually. |
| 1698 | * |
| 1699 | * The queue spin lock must be held while manipulating the requests on the |
| 1700 | * request queue. |
| 1701 | * |
| 1702 | * Function returns a pointer to the initialized request queue, or NULL if |
| 1703 | * it didn't succeed. |
| 1704 | * |
| 1705 | * Note: |
| 1706 | * blk_init_queue() must be paired with a blk_cleanup_queue() call |
| 1707 | * when the block device is deactivated (such as at module unload). |
| 1708 | **/ |
| 1709 | request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
| 1710 | { |
| 1711 | request_queue_t *q = blk_alloc_queue(GFP_KERNEL); |
| 1712 | |
| 1713 | if (!q) |
| 1714 | return NULL; |
| 1715 | |
| 1716 | if (blk_init_free_list(q)) |
| 1717 | goto out_init; |
| 1718 | |
| 152587d | 2005-04-12 16:22:06 -0500 | [diff] [blame] | 1719 | /* |
| 1720 | * if caller didn't supply a lock, they get per-queue locking with |
| 1721 | * our embedded lock |
| 1722 | */ |
| 1723 | if (!lock) { |
| 1724 | spin_lock_init(&q->__queue_lock); |
| 1725 | lock = &q->__queue_lock; |
| 1726 | } |
| 1727 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1728 | q->request_fn = rfn; |
| 1729 | q->back_merge_fn = ll_back_merge_fn; |
| 1730 | q->front_merge_fn = ll_front_merge_fn; |
| 1731 | q->merge_requests_fn = ll_merge_requests_fn; |
| 1732 | q->prep_rq_fn = NULL; |
| 1733 | q->unplug_fn = generic_unplug_device; |
| 1734 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); |
| 1735 | q->queue_lock = lock; |
| 1736 | |
| 1737 | blk_queue_segment_boundary(q, 0xffffffff); |
| 1738 | |
| 1739 | blk_queue_make_request(q, __make_request); |
| 1740 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); |
| 1741 | |
| 1742 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); |
| 1743 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); |
| 1744 | |
| 1745 | /* |
| 1746 | * all done |
| 1747 | */ |
| 1748 | if (!elevator_init(q, NULL)) { |
| 1749 | blk_queue_congestion_threshold(q); |
| 1750 | return q; |
| 1751 | } |
| 1752 | |
| 1753 | blk_cleanup_queue(q); |
| 1754 | out_init: |
| 1755 | kmem_cache_free(requestq_cachep, q); |
| 1756 | return NULL; |
| 1757 | } |
| 1758 | |
| 1759 | EXPORT_SYMBOL(blk_init_queue); |
| 1760 | |
| 1761 | int blk_get_queue(request_queue_t *q) |
| 1762 | { |
| 1763 | if (!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { |
| 1764 | atomic_inc(&q->refcnt); |
| 1765 | return 0; |
| 1766 | } |
| 1767 | |
| 1768 | return 1; |
| 1769 | } |
| 1770 | |
| 1771 | EXPORT_SYMBOL(blk_get_queue); |
| 1772 | |
| 1773 | static inline void blk_free_request(request_queue_t *q, struct request *rq) |
| 1774 | { |
| 1775 | elv_put_request(q, rq); |
| 1776 | mempool_free(rq, q->rq.rq_pool); |
| 1777 | } |
| 1778 | |
| 1779 | static inline struct request *blk_alloc_request(request_queue_t *q, int rw, |
| 1780 | int gfp_mask) |
| 1781 | { |
| 1782 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); |
| 1783 | |
| 1784 | if (!rq) |
| 1785 | return NULL; |
| 1786 | |
| 1787 | /* |
| 1788 | * first three bits are identical in rq->flags and bio->bi_rw, |
| 1789 | * see bio.h and blkdev.h |
| 1790 | */ |
| 1791 | rq->flags = rw; |
| 1792 | |
| 1793 | if (!elv_set_request(q, rq, gfp_mask)) |
| 1794 | return rq; |
| 1795 | |
| 1796 | mempool_free(rq, q->rq.rq_pool); |
| 1797 | return NULL; |
| 1798 | } |
| 1799 | |
| 1800 | /* |
| 1801 | * ioc_batching returns true if the ioc is a valid batching request and |
| 1802 | * should be given priority access to a request. |
| 1803 | */ |
| 1804 | static inline int ioc_batching(request_queue_t *q, struct io_context *ioc) |
| 1805 | { |
| 1806 | if (!ioc) |
| 1807 | return 0; |
| 1808 | |
| 1809 | /* |
| 1810 | * Make sure the process is able to allocate at least 1 request |
| 1811 | * even if the batch times out, otherwise we could theoretically |
| 1812 | * lose wakeups. |
| 1813 | */ |
| 1814 | return ioc->nr_batch_requests == q->nr_batching || |
| 1815 | (ioc->nr_batch_requests > 0 |
| 1816 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); |
| 1817 | } |
| 1818 | |
| 1819 | /* |
| 1820 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This |
| 1821 | * will cause the process to be a "batcher" on all queues in the system. This |
| 1822 | * is the behaviour we want though - once it gets a wakeup it should be given |
| 1823 | * a nice run. |
| 1824 | */ |
| 1825 | void ioc_set_batching(request_queue_t *q, struct io_context *ioc) |
| 1826 | { |
| 1827 | if (!ioc || ioc_batching(q, ioc)) |
| 1828 | return; |
| 1829 | |
| 1830 | ioc->nr_batch_requests = q->nr_batching; |
| 1831 | ioc->last_waited = jiffies; |
| 1832 | } |
| 1833 | |
| 1834 | static void __freed_request(request_queue_t *q, int rw) |
| 1835 | { |
| 1836 | struct request_list *rl = &q->rq; |
| 1837 | |
| 1838 | if (rl->count[rw] < queue_congestion_off_threshold(q)) |
| 1839 | clear_queue_congested(q, rw); |
| 1840 | |
| 1841 | if (rl->count[rw] + 1 <= q->nr_requests) { |
| 1842 | smp_mb(); |
| 1843 | if (waitqueue_active(&rl->wait[rw])) |
| 1844 | wake_up(&rl->wait[rw]); |
| 1845 | |
| 1846 | blk_clear_queue_full(q, rw); |
| 1847 | } |
| 1848 | } |
| 1849 | |
| 1850 | /* |
| 1851 | * A request has just been released. Account for it, update the full and |
| 1852 | * congestion status, wake up any waiters. Called under q->queue_lock. |
| 1853 | */ |
| 1854 | static void freed_request(request_queue_t *q, int rw) |
| 1855 | { |
| 1856 | struct request_list *rl = &q->rq; |
| 1857 | |
| 1858 | rl->count[rw]--; |
| 1859 | |
| 1860 | __freed_request(q, rw); |
| 1861 | |
| 1862 | if (unlikely(rl->starved[rw ^ 1])) |
| 1863 | __freed_request(q, rw ^ 1); |
| 1864 | |
| 1865 | if (!rl->count[READ] && !rl->count[WRITE]) { |
| 1866 | smp_mb(); |
| 1867 | if (unlikely(waitqueue_active(&rl->drain))) |
| 1868 | wake_up(&rl->drain); |
| 1869 | } |
| 1870 | } |
| 1871 | |
| 1872 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) |
| 1873 | /* |
| 1874 | * Get a free request, queue_lock must not be held |
| 1875 | */ |
| 1876 | static struct request *get_request(request_queue_t *q, int rw, int gfp_mask) |
| 1877 | { |
| 1878 | struct request *rq = NULL; |
| 1879 | struct request_list *rl = &q->rq; |
| 1880 | struct io_context *ioc = get_io_context(gfp_mask); |
| 1881 | |
| 1882 | if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) |
| 1883 | goto out; |
| 1884 | |
| 1885 | spin_lock_irq(q->queue_lock); |
| 1886 | if (rl->count[rw]+1 >= q->nr_requests) { |
| 1887 | /* |
| 1888 | * The queue will fill after this allocation, so set it as |
| 1889 | * full, and mark this process as "batching". This process |
| 1890 | * will be allowed to complete a batch of requests, others |
| 1891 | * will be blocked. |
| 1892 | */ |
| 1893 | if (!blk_queue_full(q, rw)) { |
| 1894 | ioc_set_batching(q, ioc); |
| 1895 | blk_set_queue_full(q, rw); |
| 1896 | } |
| 1897 | } |
| 1898 | |
| 1899 | switch (elv_may_queue(q, rw)) { |
| 1900 | case ELV_MQUEUE_NO: |
| 1901 | goto rq_starved; |
| 1902 | case ELV_MQUEUE_MAY: |
| 1903 | break; |
| 1904 | case ELV_MQUEUE_MUST: |
| 1905 | goto get_rq; |
| 1906 | } |
| 1907 | |
| 1908 | if (blk_queue_full(q, rw) && !ioc_batching(q, ioc)) { |
| 1909 | /* |
| 1910 | * The queue is full and the allocating process is not a |
| 1911 | * "batcher", and not exempted by the IO scheduler |
| 1912 | */ |
| 1913 | spin_unlock_irq(q->queue_lock); |
| 1914 | goto out; |
| 1915 | } |
| 1916 | |
| 1917 | get_rq: |
| 1918 | rl->count[rw]++; |
| 1919 | rl->starved[rw] = 0; |
| 1920 | if (rl->count[rw] >= queue_congestion_on_threshold(q)) |
| 1921 | set_queue_congested(q, rw); |
| 1922 | spin_unlock_irq(q->queue_lock); |
| 1923 | |
| 1924 | rq = blk_alloc_request(q, rw, gfp_mask); |
| 1925 | if (!rq) { |
| 1926 | /* |
| 1927 | * Allocation failed presumably due to memory. Undo anything |
| 1928 | * we might have messed up. |
| 1929 | * |
| 1930 | * Allocating task should really be put onto the front of the |
| 1931 | * wait queue, but this is pretty rare. |
| 1932 | */ |
| 1933 | spin_lock_irq(q->queue_lock); |
| 1934 | freed_request(q, rw); |
| 1935 | |
| 1936 | /* |
| 1937 | * in the very unlikely event that allocation failed and no |
| 1938 | * requests for this direction was pending, mark us starved |
| 1939 | * so that freeing of a request in the other direction will |
| 1940 | * notice us. another possible fix would be to split the |
| 1941 | * rq mempool into READ and WRITE |
| 1942 | */ |
| 1943 | rq_starved: |
| 1944 | if (unlikely(rl->count[rw] == 0)) |
| 1945 | rl->starved[rw] = 1; |
| 1946 | |
| 1947 | spin_unlock_irq(q->queue_lock); |
| 1948 | goto out; |
| 1949 | } |
| 1950 | |
| 1951 | if (ioc_batching(q, ioc)) |
| 1952 | ioc->nr_batch_requests--; |
| 1953 | |
| 1954 | rq_init(q, rq); |
| 1955 | rq->rl = rl; |
| 1956 | out: |
| 1957 | put_io_context(ioc); |
| 1958 | return rq; |
| 1959 | } |
| 1960 | |
| 1961 | /* |
| 1962 | * No available requests for this queue, unplug the device and wait for some |
| 1963 | * requests to become available. |
| 1964 | */ |
| 1965 | static struct request *get_request_wait(request_queue_t *q, int rw) |
| 1966 | { |
| 1967 | DEFINE_WAIT(wait); |
| 1968 | struct request *rq; |
| 1969 | |
| 1970 | generic_unplug_device(q); |
| 1971 | do { |
| 1972 | struct request_list *rl = &q->rq; |
| 1973 | |
| 1974 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, |
| 1975 | TASK_UNINTERRUPTIBLE); |
| 1976 | |
| 1977 | rq = get_request(q, rw, GFP_NOIO); |
| 1978 | |
| 1979 | if (!rq) { |
| 1980 | struct io_context *ioc; |
| 1981 | |
| 1982 | io_schedule(); |
| 1983 | |
| 1984 | /* |
| 1985 | * After sleeping, we become a "batching" process and |
| 1986 | * will be able to allocate at least one request, and |
| 1987 | * up to a big batch of them for a small period time. |
| 1988 | * See ioc_batching, ioc_set_batching |
| 1989 | */ |
| 1990 | ioc = get_io_context(GFP_NOIO); |
| 1991 | ioc_set_batching(q, ioc); |
| 1992 | put_io_context(ioc); |
| 1993 | } |
| 1994 | finish_wait(&rl->wait[rw], &wait); |
| 1995 | } while (!rq); |
| 1996 | |
| 1997 | return rq; |
| 1998 | } |
| 1999 | |
| 2000 | struct request *blk_get_request(request_queue_t *q, int rw, int gfp_mask) |
| 2001 | { |
| 2002 | struct request *rq; |
| 2003 | |
| 2004 | BUG_ON(rw != READ && rw != WRITE); |
| 2005 | |
| 2006 | if (gfp_mask & __GFP_WAIT) |
| 2007 | rq = get_request_wait(q, rw); |
| 2008 | else |
| 2009 | rq = get_request(q, rw, gfp_mask); |
| 2010 | |
| 2011 | return rq; |
| 2012 | } |
| 2013 | |
| 2014 | EXPORT_SYMBOL(blk_get_request); |
| 2015 | |
| 2016 | /** |
| 2017 | * blk_requeue_request - put a request back on queue |
| 2018 | * @q: request queue where request should be inserted |
| 2019 | * @rq: request to be inserted |
| 2020 | * |
| 2021 | * Description: |
| 2022 | * Drivers often keep queueing requests until the hardware cannot accept |
| 2023 | * more, when that condition happens we need to put the request back |
| 2024 | * on the queue. Must be called with queue lock held. |
| 2025 | */ |
| 2026 | void blk_requeue_request(request_queue_t *q, struct request *rq) |
| 2027 | { |
| 2028 | if (blk_rq_tagged(rq)) |
| 2029 | blk_queue_end_tag(q, rq); |
| 2030 | |
| 2031 | elv_requeue_request(q, rq); |
| 2032 | } |
| 2033 | |
| 2034 | EXPORT_SYMBOL(blk_requeue_request); |
| 2035 | |
| 2036 | /** |
| 2037 | * blk_insert_request - insert a special request in to a request queue |
| 2038 | * @q: request queue where request should be inserted |
| 2039 | * @rq: request to be inserted |
| 2040 | * @at_head: insert request at head or tail of queue |
| 2041 | * @data: private data |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2042 | * |
| 2043 | * Description: |
| 2044 | * Many block devices need to execute commands asynchronously, so they don't |
| 2045 | * block the whole kernel from preemption during request execution. This is |
| 2046 | * accomplished normally by inserting aritficial requests tagged as |
| 2047 | * REQ_SPECIAL in to the corresponding request queue, and letting them be |
| 2048 | * scheduled for actual execution by the request queue. |
| 2049 | * |
| 2050 | * We have the option of inserting the head or the tail of the queue. |
| 2051 | * Typically we use the tail for new ioctls and so forth. We use the head |
| 2052 | * of the queue for things like a QUEUE_FULL message from a device, or a |
| 2053 | * host that is unable to accept a particular command. |
| 2054 | */ |
| 2055 | void blk_insert_request(request_queue_t *q, struct request *rq, |
Tejun Heo | 867d119 | 2005-04-24 02:06:05 -0500 | [diff] [blame] | 2056 | int at_head, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2057 | { |
Tejun Heo | 867d119 | 2005-04-24 02:06:05 -0500 | [diff] [blame] | 2058 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2059 | unsigned long flags; |
| 2060 | |
| 2061 | /* |
| 2062 | * tell I/O scheduler that this isn't a regular read/write (ie it |
| 2063 | * must not attempt merges on this) and that it acts as a soft |
| 2064 | * barrier |
| 2065 | */ |
| 2066 | rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER; |
| 2067 | |
| 2068 | rq->special = data; |
| 2069 | |
| 2070 | spin_lock_irqsave(q->queue_lock, flags); |
| 2071 | |
| 2072 | /* |
| 2073 | * If command is tagged, release the tag |
| 2074 | */ |
Tejun Heo | 867d119 | 2005-04-24 02:06:05 -0500 | [diff] [blame] | 2075 | if (blk_rq_tagged(rq)) |
| 2076 | blk_queue_end_tag(q, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2077 | |
Tejun Heo | 867d119 | 2005-04-24 02:06:05 -0500 | [diff] [blame] | 2078 | drive_stat_acct(rq, rq->nr_sectors, 1); |
| 2079 | __elv_add_request(q, rq, where, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2080 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2081 | if (blk_queue_plugged(q)) |
| 2082 | __generic_unplug_device(q); |
| 2083 | else |
| 2084 | q->request_fn(q); |
| 2085 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 2086 | } |
| 2087 | |
| 2088 | EXPORT_SYMBOL(blk_insert_request); |
| 2089 | |
| 2090 | /** |
| 2091 | * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage |
| 2092 | * @q: request queue where request should be inserted |
| 2093 | * @rw: READ or WRITE data |
| 2094 | * @ubuf: the user buffer |
| 2095 | * @len: length of user data |
| 2096 | * |
| 2097 | * Description: |
| 2098 | * Data will be mapped directly for zero copy io, if possible. Otherwise |
| 2099 | * a kernel bounce buffer is used. |
| 2100 | * |
| 2101 | * A matching blk_rq_unmap_user() must be issued at the end of io, while |
| 2102 | * still in process context. |
| 2103 | * |
| 2104 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() |
| 2105 | * before being submitted to the device, as pages mapped may be out of |
| 2106 | * reach. It's the callers responsibility to make sure this happens. The |
| 2107 | * original bio must be passed back in to blk_rq_unmap_user() for proper |
| 2108 | * unmapping. |
| 2109 | */ |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2110 | int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf, |
| 2111 | unsigned int len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2112 | { |
| 2113 | unsigned long uaddr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2114 | struct bio *bio; |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2115 | int reading; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2116 | |
| 2117 | if (len > (q->max_sectors << 9)) |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2118 | return -EINVAL; |
| 2119 | if (!len || !ubuf) |
| 2120 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2121 | |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2122 | reading = rq_data_dir(rq) == READ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2123 | |
| 2124 | /* |
| 2125 | * if alignment requirement is satisfied, map in user pages for |
| 2126 | * direct dma. else, set up kernel bounce buffers |
| 2127 | */ |
| 2128 | uaddr = (unsigned long) ubuf; |
| 2129 | if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q))) |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2130 | bio = bio_map_user(q, NULL, uaddr, len, reading); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2131 | else |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2132 | bio = bio_copy_user(q, uaddr, len, reading); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2133 | |
| 2134 | if (!IS_ERR(bio)) { |
| 2135 | rq->bio = rq->biotail = bio; |
| 2136 | blk_rq_bio_prep(q, rq, bio); |
| 2137 | |
| 2138 | rq->buffer = rq->data = NULL; |
| 2139 | rq->data_len = len; |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2140 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2141 | } |
| 2142 | |
| 2143 | /* |
| 2144 | * bio is the err-ptr |
| 2145 | */ |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2146 | return PTR_ERR(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2147 | } |
| 2148 | |
| 2149 | EXPORT_SYMBOL(blk_rq_map_user); |
| 2150 | |
| 2151 | /** |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame^] | 2152 | * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage |
| 2153 | * @q: request queue where request should be inserted |
| 2154 | * @rq: request to map data to |
| 2155 | * @iov: pointer to the iovec |
| 2156 | * @iov_count: number of elements in the iovec |
| 2157 | * |
| 2158 | * Description: |
| 2159 | * Data will be mapped directly for zero copy io, if possible. Otherwise |
| 2160 | * a kernel bounce buffer is used. |
| 2161 | * |
| 2162 | * A matching blk_rq_unmap_user() must be issued at the end of io, while |
| 2163 | * still in process context. |
| 2164 | * |
| 2165 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() |
| 2166 | * before being submitted to the device, as pages mapped may be out of |
| 2167 | * reach. It's the callers responsibility to make sure this happens. The |
| 2168 | * original bio must be passed back in to blk_rq_unmap_user() for proper |
| 2169 | * unmapping. |
| 2170 | */ |
| 2171 | int blk_rq_map_user_iov(request_queue_t *q, struct request *rq, |
| 2172 | struct sg_iovec *iov, int iov_count) |
| 2173 | { |
| 2174 | struct bio *bio; |
| 2175 | |
| 2176 | if (!iov || iov_count <= 0) |
| 2177 | return -EINVAL; |
| 2178 | |
| 2179 | /* we don't allow misaligned data like bio_map_user() does. If the |
| 2180 | * user is using sg, they're expected to know the alignment constraints |
| 2181 | * and respect them accordingly */ |
| 2182 | bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ); |
| 2183 | if (IS_ERR(bio)) |
| 2184 | return PTR_ERR(bio); |
| 2185 | |
| 2186 | rq->bio = rq->biotail = bio; |
| 2187 | blk_rq_bio_prep(q, rq, bio); |
| 2188 | rq->buffer = rq->data = NULL; |
| 2189 | rq->data_len = bio->bi_size; |
| 2190 | return 0; |
| 2191 | } |
| 2192 | |
| 2193 | EXPORT_SYMBOL(blk_rq_map_user_iov); |
| 2194 | |
| 2195 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2196 | * blk_rq_unmap_user - unmap a request with user data |
| 2197 | * @rq: request to be unmapped |
| 2198 | * @bio: bio for the request |
| 2199 | * @ulen: length of user buffer |
| 2200 | * |
| 2201 | * Description: |
| 2202 | * Unmap a request previously mapped by blk_rq_map_user(). |
| 2203 | */ |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2204 | int blk_rq_unmap_user(struct bio *bio, unsigned int ulen) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2205 | { |
| 2206 | int ret = 0; |
| 2207 | |
| 2208 | if (bio) { |
| 2209 | if (bio_flagged(bio, BIO_USER_MAPPED)) |
| 2210 | bio_unmap_user(bio); |
| 2211 | else |
| 2212 | ret = bio_uncopy_user(bio); |
| 2213 | } |
| 2214 | |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2215 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2216 | } |
| 2217 | |
| 2218 | EXPORT_SYMBOL(blk_rq_unmap_user); |
| 2219 | |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 2220 | /** |
| 2221 | * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage |
| 2222 | * @q: request queue where request should be inserted |
| 2223 | * @rw: READ or WRITE data |
| 2224 | * @kbuf: the kernel buffer |
| 2225 | * @len: length of user data |
| 2226 | */ |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2227 | int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf, |
| 2228 | unsigned int len, unsigned int gfp_mask) |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 2229 | { |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 2230 | struct bio *bio; |
| 2231 | |
| 2232 | if (len > (q->max_sectors << 9)) |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2233 | return -EINVAL; |
| 2234 | if (!len || !kbuf) |
| 2235 | return -EINVAL; |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 2236 | |
| 2237 | bio = bio_map_kern(q, kbuf, len, gfp_mask); |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2238 | if (IS_ERR(bio)) |
| 2239 | return PTR_ERR(bio); |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 2240 | |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2241 | if (rq_data_dir(rq) == WRITE) |
| 2242 | bio->bi_rw |= (1 << BIO_RW); |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 2243 | |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2244 | rq->bio = rq->biotail = bio; |
| 2245 | blk_rq_bio_prep(q, rq, bio); |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 2246 | |
Jens Axboe | dd1cab9 | 2005-06-20 14:06:01 +0200 | [diff] [blame] | 2247 | rq->buffer = rq->data = NULL; |
| 2248 | rq->data_len = len; |
| 2249 | return 0; |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 2250 | } |
| 2251 | |
| 2252 | EXPORT_SYMBOL(blk_rq_map_kern); |
| 2253 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame^] | 2254 | void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk, |
| 2255 | struct request *rq, int at_head, |
| 2256 | void (*done)(struct request *)) |
| 2257 | { |
| 2258 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
| 2259 | |
| 2260 | rq->rq_disk = bd_disk; |
| 2261 | rq->flags |= REQ_NOMERGE; |
| 2262 | rq->end_io = done; |
| 2263 | elv_add_request(q, rq, where, 1); |
| 2264 | generic_unplug_device(q); |
| 2265 | } |
| 2266 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2267 | /** |
| 2268 | * blk_execute_rq - insert a request into queue for execution |
| 2269 | * @q: queue to insert the request in |
| 2270 | * @bd_disk: matching gendisk |
| 2271 | * @rq: request to insert |
| 2272 | * |
| 2273 | * Description: |
| 2274 | * Insert a fully prepared request at the back of the io scheduler queue |
| 2275 | * for execution. |
| 2276 | */ |
| 2277 | int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk, |
| 2278 | struct request *rq) |
| 2279 | { |
| 2280 | DECLARE_COMPLETION(wait); |
| 2281 | char sense[SCSI_SENSE_BUFFERSIZE]; |
| 2282 | int err = 0; |
| 2283 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2284 | /* |
| 2285 | * we need an extra reference to the request, so we can look at |
| 2286 | * it after io completion |
| 2287 | */ |
| 2288 | rq->ref_count++; |
| 2289 | |
| 2290 | if (!rq->sense) { |
| 2291 | memset(sense, 0, sizeof(sense)); |
| 2292 | rq->sense = sense; |
| 2293 | rq->sense_len = 0; |
| 2294 | } |
| 2295 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2296 | rq->waiting = &wait; |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame^] | 2297 | blk_execute_rq_nowait(q, bd_disk, rq, 0, blk_end_sync_rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2298 | wait_for_completion(&wait); |
| 2299 | rq->waiting = NULL; |
| 2300 | |
| 2301 | if (rq->errors) |
| 2302 | err = -EIO; |
| 2303 | |
| 2304 | return err; |
| 2305 | } |
| 2306 | |
| 2307 | EXPORT_SYMBOL(blk_execute_rq); |
| 2308 | |
| 2309 | /** |
| 2310 | * blkdev_issue_flush - queue a flush |
| 2311 | * @bdev: blockdev to issue flush for |
| 2312 | * @error_sector: error sector |
| 2313 | * |
| 2314 | * Description: |
| 2315 | * Issue a flush for the block device in question. Caller can supply |
| 2316 | * room for storing the error offset in case of a flush error, if they |
| 2317 | * wish to. Caller must run wait_for_completion() on its own. |
| 2318 | */ |
| 2319 | int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector) |
| 2320 | { |
| 2321 | request_queue_t *q; |
| 2322 | |
| 2323 | if (bdev->bd_disk == NULL) |
| 2324 | return -ENXIO; |
| 2325 | |
| 2326 | q = bdev_get_queue(bdev); |
| 2327 | if (!q) |
| 2328 | return -ENXIO; |
| 2329 | if (!q->issue_flush_fn) |
| 2330 | return -EOPNOTSUPP; |
| 2331 | |
| 2332 | return q->issue_flush_fn(q, bdev->bd_disk, error_sector); |
| 2333 | } |
| 2334 | |
| 2335 | EXPORT_SYMBOL(blkdev_issue_flush); |
| 2336 | |
| 2337 | /** |
| 2338 | * blkdev_scsi_issue_flush_fn - issue flush for SCSI devices |
| 2339 | * @q: device queue |
| 2340 | * @disk: gendisk |
| 2341 | * @error_sector: error offset |
| 2342 | * |
| 2343 | * Description: |
| 2344 | * Devices understanding the SCSI command set, can use this function as |
| 2345 | * a helper for issuing a cache flush. Note: driver is required to store |
| 2346 | * the error offset (in case of error flushing) in ->sector of struct |
| 2347 | * request. |
| 2348 | */ |
| 2349 | int blkdev_scsi_issue_flush_fn(request_queue_t *q, struct gendisk *disk, |
| 2350 | sector_t *error_sector) |
| 2351 | { |
| 2352 | struct request *rq = blk_get_request(q, WRITE, __GFP_WAIT); |
| 2353 | int ret; |
| 2354 | |
| 2355 | rq->flags |= REQ_BLOCK_PC | REQ_SOFTBARRIER; |
| 2356 | rq->sector = 0; |
| 2357 | memset(rq->cmd, 0, sizeof(rq->cmd)); |
| 2358 | rq->cmd[0] = 0x35; |
| 2359 | rq->cmd_len = 12; |
| 2360 | rq->data = NULL; |
| 2361 | rq->data_len = 0; |
| 2362 | rq->timeout = 60 * HZ; |
| 2363 | |
| 2364 | ret = blk_execute_rq(q, disk, rq); |
| 2365 | |
| 2366 | if (ret && error_sector) |
| 2367 | *error_sector = rq->sector; |
| 2368 | |
| 2369 | blk_put_request(rq); |
| 2370 | return ret; |
| 2371 | } |
| 2372 | |
| 2373 | EXPORT_SYMBOL(blkdev_scsi_issue_flush_fn); |
| 2374 | |
| 2375 | void drive_stat_acct(struct request *rq, int nr_sectors, int new_io) |
| 2376 | { |
| 2377 | int rw = rq_data_dir(rq); |
| 2378 | |
| 2379 | if (!blk_fs_request(rq) || !rq->rq_disk) |
| 2380 | return; |
| 2381 | |
| 2382 | if (rw == READ) { |
| 2383 | __disk_stat_add(rq->rq_disk, read_sectors, nr_sectors); |
| 2384 | if (!new_io) |
| 2385 | __disk_stat_inc(rq->rq_disk, read_merges); |
| 2386 | } else if (rw == WRITE) { |
| 2387 | __disk_stat_add(rq->rq_disk, write_sectors, nr_sectors); |
| 2388 | if (!new_io) |
| 2389 | __disk_stat_inc(rq->rq_disk, write_merges); |
| 2390 | } |
| 2391 | if (new_io) { |
| 2392 | disk_round_stats(rq->rq_disk); |
| 2393 | rq->rq_disk->in_flight++; |
| 2394 | } |
| 2395 | } |
| 2396 | |
| 2397 | /* |
| 2398 | * add-request adds a request to the linked list. |
| 2399 | * queue lock is held and interrupts disabled, as we muck with the |
| 2400 | * request queue list. |
| 2401 | */ |
| 2402 | static inline void add_request(request_queue_t * q, struct request * req) |
| 2403 | { |
| 2404 | drive_stat_acct(req, req->nr_sectors, 1); |
| 2405 | |
| 2406 | if (q->activity_fn) |
| 2407 | q->activity_fn(q->activity_data, rq_data_dir(req)); |
| 2408 | |
| 2409 | /* |
| 2410 | * elevator indicated where it wants this request to be |
| 2411 | * inserted at elevator_merge time |
| 2412 | */ |
| 2413 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); |
| 2414 | } |
| 2415 | |
| 2416 | /* |
| 2417 | * disk_round_stats() - Round off the performance stats on a struct |
| 2418 | * disk_stats. |
| 2419 | * |
| 2420 | * The average IO queue length and utilisation statistics are maintained |
| 2421 | * by observing the current state of the queue length and the amount of |
| 2422 | * time it has been in this state for. |
| 2423 | * |
| 2424 | * Normally, that accounting is done on IO completion, but that can result |
| 2425 | * in more than a second's worth of IO being accounted for within any one |
| 2426 | * second, leading to >100% utilisation. To deal with that, we call this |
| 2427 | * function to do a round-off before returning the results when reading |
| 2428 | * /proc/diskstats. This accounts immediately for all queue usage up to |
| 2429 | * the current jiffies and restarts the counters again. |
| 2430 | */ |
| 2431 | void disk_round_stats(struct gendisk *disk) |
| 2432 | { |
| 2433 | unsigned long now = jiffies; |
| 2434 | |
| 2435 | __disk_stat_add(disk, time_in_queue, |
| 2436 | disk->in_flight * (now - disk->stamp)); |
| 2437 | disk->stamp = now; |
| 2438 | |
| 2439 | if (disk->in_flight) |
| 2440 | __disk_stat_add(disk, io_ticks, (now - disk->stamp_idle)); |
| 2441 | disk->stamp_idle = now; |
| 2442 | } |
| 2443 | |
| 2444 | /* |
| 2445 | * queue lock must be held |
| 2446 | */ |
| 2447 | static void __blk_put_request(request_queue_t *q, struct request *req) |
| 2448 | { |
| 2449 | struct request_list *rl = req->rl; |
| 2450 | |
| 2451 | if (unlikely(!q)) |
| 2452 | return; |
| 2453 | if (unlikely(--req->ref_count)) |
| 2454 | return; |
| 2455 | |
| 2456 | req->rq_status = RQ_INACTIVE; |
| 2457 | req->q = NULL; |
| 2458 | req->rl = NULL; |
| 2459 | |
| 2460 | /* |
| 2461 | * Request may not have originated from ll_rw_blk. if not, |
| 2462 | * it didn't come out of our reserved rq pools |
| 2463 | */ |
| 2464 | if (rl) { |
| 2465 | int rw = rq_data_dir(req); |
| 2466 | |
| 2467 | elv_completed_request(q, req); |
| 2468 | |
| 2469 | BUG_ON(!list_empty(&req->queuelist)); |
| 2470 | |
| 2471 | blk_free_request(q, req); |
| 2472 | freed_request(q, rw); |
| 2473 | } |
| 2474 | } |
| 2475 | |
| 2476 | void blk_put_request(struct request *req) |
| 2477 | { |
| 2478 | /* |
| 2479 | * if req->rl isn't set, this request didnt originate from the |
| 2480 | * block layer, so it's safe to just disregard it |
| 2481 | */ |
| 2482 | if (req->rl) { |
| 2483 | unsigned long flags; |
| 2484 | request_queue_t *q = req->q; |
| 2485 | |
| 2486 | spin_lock_irqsave(q->queue_lock, flags); |
| 2487 | __blk_put_request(q, req); |
| 2488 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 2489 | } |
| 2490 | } |
| 2491 | |
| 2492 | EXPORT_SYMBOL(blk_put_request); |
| 2493 | |
| 2494 | /** |
| 2495 | * blk_end_sync_rq - executes a completion event on a request |
| 2496 | * @rq: request to complete |
| 2497 | */ |
| 2498 | void blk_end_sync_rq(struct request *rq) |
| 2499 | { |
| 2500 | struct completion *waiting = rq->waiting; |
| 2501 | |
| 2502 | rq->waiting = NULL; |
| 2503 | __blk_put_request(rq->q, rq); |
| 2504 | |
| 2505 | /* |
| 2506 | * complete last, if this is a stack request the process (and thus |
| 2507 | * the rq pointer) could be invalid right after this complete() |
| 2508 | */ |
| 2509 | complete(waiting); |
| 2510 | } |
| 2511 | EXPORT_SYMBOL(blk_end_sync_rq); |
| 2512 | |
| 2513 | /** |
| 2514 | * blk_congestion_wait - wait for a queue to become uncongested |
| 2515 | * @rw: READ or WRITE |
| 2516 | * @timeout: timeout in jiffies |
| 2517 | * |
| 2518 | * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion. |
| 2519 | * If no queues are congested then just wait for the next request to be |
| 2520 | * returned. |
| 2521 | */ |
| 2522 | long blk_congestion_wait(int rw, long timeout) |
| 2523 | { |
| 2524 | long ret; |
| 2525 | DEFINE_WAIT(wait); |
| 2526 | wait_queue_head_t *wqh = &congestion_wqh[rw]; |
| 2527 | |
| 2528 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); |
| 2529 | ret = io_schedule_timeout(timeout); |
| 2530 | finish_wait(wqh, &wait); |
| 2531 | return ret; |
| 2532 | } |
| 2533 | |
| 2534 | EXPORT_SYMBOL(blk_congestion_wait); |
| 2535 | |
| 2536 | /* |
| 2537 | * Has to be called with the request spinlock acquired |
| 2538 | */ |
| 2539 | static int attempt_merge(request_queue_t *q, struct request *req, |
| 2540 | struct request *next) |
| 2541 | { |
| 2542 | if (!rq_mergeable(req) || !rq_mergeable(next)) |
| 2543 | return 0; |
| 2544 | |
| 2545 | /* |
| 2546 | * not contigious |
| 2547 | */ |
| 2548 | if (req->sector + req->nr_sectors != next->sector) |
| 2549 | return 0; |
| 2550 | |
| 2551 | if (rq_data_dir(req) != rq_data_dir(next) |
| 2552 | || req->rq_disk != next->rq_disk |
| 2553 | || next->waiting || next->special) |
| 2554 | return 0; |
| 2555 | |
| 2556 | /* |
| 2557 | * If we are allowed to merge, then append bio list |
| 2558 | * from next to rq and release next. merge_requests_fn |
| 2559 | * will have updated segment counts, update sector |
| 2560 | * counts here. |
| 2561 | */ |
| 2562 | if (!q->merge_requests_fn(q, req, next)) |
| 2563 | return 0; |
| 2564 | |
| 2565 | /* |
| 2566 | * At this point we have either done a back merge |
| 2567 | * or front merge. We need the smaller start_time of |
| 2568 | * the merged requests to be the current request |
| 2569 | * for accounting purposes. |
| 2570 | */ |
| 2571 | if (time_after(req->start_time, next->start_time)) |
| 2572 | req->start_time = next->start_time; |
| 2573 | |
| 2574 | req->biotail->bi_next = next->bio; |
| 2575 | req->biotail = next->biotail; |
| 2576 | |
| 2577 | req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors; |
| 2578 | |
| 2579 | elv_merge_requests(q, req, next); |
| 2580 | |
| 2581 | if (req->rq_disk) { |
| 2582 | disk_round_stats(req->rq_disk); |
| 2583 | req->rq_disk->in_flight--; |
| 2584 | } |
| 2585 | |
| 2586 | __blk_put_request(q, next); |
| 2587 | return 1; |
| 2588 | } |
| 2589 | |
| 2590 | static inline int attempt_back_merge(request_queue_t *q, struct request *rq) |
| 2591 | { |
| 2592 | struct request *next = elv_latter_request(q, rq); |
| 2593 | |
| 2594 | if (next) |
| 2595 | return attempt_merge(q, rq, next); |
| 2596 | |
| 2597 | return 0; |
| 2598 | } |
| 2599 | |
| 2600 | static inline int attempt_front_merge(request_queue_t *q, struct request *rq) |
| 2601 | { |
| 2602 | struct request *prev = elv_former_request(q, rq); |
| 2603 | |
| 2604 | if (prev) |
| 2605 | return attempt_merge(q, prev, rq); |
| 2606 | |
| 2607 | return 0; |
| 2608 | } |
| 2609 | |
| 2610 | /** |
| 2611 | * blk_attempt_remerge - attempt to remerge active head with next request |
| 2612 | * @q: The &request_queue_t belonging to the device |
| 2613 | * @rq: The head request (usually) |
| 2614 | * |
| 2615 | * Description: |
| 2616 | * For head-active devices, the queue can easily be unplugged so quickly |
| 2617 | * that proper merging is not done on the front request. This may hurt |
| 2618 | * performance greatly for some devices. The block layer cannot safely |
| 2619 | * do merging on that first request for these queues, but the driver can |
| 2620 | * call this function and make it happen any way. Only the driver knows |
| 2621 | * when it is safe to do so. |
| 2622 | **/ |
| 2623 | void blk_attempt_remerge(request_queue_t *q, struct request *rq) |
| 2624 | { |
| 2625 | unsigned long flags; |
| 2626 | |
| 2627 | spin_lock_irqsave(q->queue_lock, flags); |
| 2628 | attempt_back_merge(q, rq); |
| 2629 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 2630 | } |
| 2631 | |
| 2632 | EXPORT_SYMBOL(blk_attempt_remerge); |
| 2633 | |
| 2634 | /* |
| 2635 | * Non-locking blk_attempt_remerge variant. |
| 2636 | */ |
| 2637 | void __blk_attempt_remerge(request_queue_t *q, struct request *rq) |
| 2638 | { |
| 2639 | attempt_back_merge(q, rq); |
| 2640 | } |
| 2641 | |
| 2642 | EXPORT_SYMBOL(__blk_attempt_remerge); |
| 2643 | |
| 2644 | static int __make_request(request_queue_t *q, struct bio *bio) |
| 2645 | { |
| 2646 | struct request *req, *freereq = NULL; |
Jens Axboe | 4a534f9 | 2005-04-16 15:25:40 -0700 | [diff] [blame] | 2647 | int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2648 | sector_t sector; |
| 2649 | |
| 2650 | sector = bio->bi_sector; |
| 2651 | nr_sectors = bio_sectors(bio); |
| 2652 | cur_nr_sectors = bio_cur_sectors(bio); |
| 2653 | |
| 2654 | rw = bio_data_dir(bio); |
Jens Axboe | 4a534f9 | 2005-04-16 15:25:40 -0700 | [diff] [blame] | 2655 | sync = bio_sync(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2656 | |
| 2657 | /* |
| 2658 | * low level driver can indicate that it wants pages above a |
| 2659 | * certain limit bounced to low memory (ie for highmem, or even |
| 2660 | * ISA dma in theory) |
| 2661 | */ |
| 2662 | blk_queue_bounce(q, &bio); |
| 2663 | |
| 2664 | spin_lock_prefetch(q->queue_lock); |
| 2665 | |
| 2666 | barrier = bio_barrier(bio); |
| 2667 | if (barrier && (q->ordered == QUEUE_ORDERED_NONE)) { |
| 2668 | err = -EOPNOTSUPP; |
| 2669 | goto end_io; |
| 2670 | } |
| 2671 | |
| 2672 | again: |
| 2673 | spin_lock_irq(q->queue_lock); |
| 2674 | |
| 2675 | if (elv_queue_empty(q)) { |
| 2676 | blk_plug_device(q); |
| 2677 | goto get_rq; |
| 2678 | } |
| 2679 | if (barrier) |
| 2680 | goto get_rq; |
| 2681 | |
| 2682 | el_ret = elv_merge(q, &req, bio); |
| 2683 | switch (el_ret) { |
| 2684 | case ELEVATOR_BACK_MERGE: |
| 2685 | BUG_ON(!rq_mergeable(req)); |
| 2686 | |
| 2687 | if (!q->back_merge_fn(q, req, bio)) |
| 2688 | break; |
| 2689 | |
| 2690 | req->biotail->bi_next = bio; |
| 2691 | req->biotail = bio; |
| 2692 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; |
| 2693 | drive_stat_acct(req, nr_sectors, 0); |
| 2694 | if (!attempt_back_merge(q, req)) |
| 2695 | elv_merged_request(q, req); |
| 2696 | goto out; |
| 2697 | |
| 2698 | case ELEVATOR_FRONT_MERGE: |
| 2699 | BUG_ON(!rq_mergeable(req)); |
| 2700 | |
| 2701 | if (!q->front_merge_fn(q, req, bio)) |
| 2702 | break; |
| 2703 | |
| 2704 | bio->bi_next = req->bio; |
| 2705 | req->bio = bio; |
| 2706 | |
| 2707 | /* |
| 2708 | * may not be valid. if the low level driver said |
| 2709 | * it didn't need a bounce buffer then it better |
| 2710 | * not touch req->buffer either... |
| 2711 | */ |
| 2712 | req->buffer = bio_data(bio); |
| 2713 | req->current_nr_sectors = cur_nr_sectors; |
| 2714 | req->hard_cur_sectors = cur_nr_sectors; |
| 2715 | req->sector = req->hard_sector = sector; |
| 2716 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; |
| 2717 | drive_stat_acct(req, nr_sectors, 0); |
| 2718 | if (!attempt_front_merge(q, req)) |
| 2719 | elv_merged_request(q, req); |
| 2720 | goto out; |
| 2721 | |
| 2722 | /* |
| 2723 | * elevator says don't/can't merge. get new request |
| 2724 | */ |
| 2725 | case ELEVATOR_NO_MERGE: |
| 2726 | break; |
| 2727 | |
| 2728 | default: |
| 2729 | printk("elevator returned crap (%d)\n", el_ret); |
| 2730 | BUG(); |
| 2731 | } |
| 2732 | |
| 2733 | /* |
| 2734 | * Grab a free request from the freelist - if that is empty, check |
| 2735 | * if we are doing read ahead and abort instead of blocking for |
| 2736 | * a free slot. |
| 2737 | */ |
| 2738 | get_rq: |
| 2739 | if (freereq) { |
| 2740 | req = freereq; |
| 2741 | freereq = NULL; |
| 2742 | } else { |
| 2743 | spin_unlock_irq(q->queue_lock); |
| 2744 | if ((freereq = get_request(q, rw, GFP_ATOMIC)) == NULL) { |
| 2745 | /* |
| 2746 | * READA bit set |
| 2747 | */ |
| 2748 | err = -EWOULDBLOCK; |
| 2749 | if (bio_rw_ahead(bio)) |
| 2750 | goto end_io; |
| 2751 | |
| 2752 | freereq = get_request_wait(q, rw); |
| 2753 | } |
| 2754 | goto again; |
| 2755 | } |
| 2756 | |
| 2757 | req->flags |= REQ_CMD; |
| 2758 | |
| 2759 | /* |
| 2760 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) |
| 2761 | */ |
| 2762 | if (bio_rw_ahead(bio) || bio_failfast(bio)) |
| 2763 | req->flags |= REQ_FAILFAST; |
| 2764 | |
| 2765 | /* |
| 2766 | * REQ_BARRIER implies no merging, but lets make it explicit |
| 2767 | */ |
| 2768 | if (barrier) |
| 2769 | req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE); |
| 2770 | |
| 2771 | req->errors = 0; |
| 2772 | req->hard_sector = req->sector = sector; |
| 2773 | req->hard_nr_sectors = req->nr_sectors = nr_sectors; |
| 2774 | req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors; |
| 2775 | req->nr_phys_segments = bio_phys_segments(q, bio); |
| 2776 | req->nr_hw_segments = bio_hw_segments(q, bio); |
| 2777 | req->buffer = bio_data(bio); /* see ->buffer comment above */ |
| 2778 | req->waiting = NULL; |
| 2779 | req->bio = req->biotail = bio; |
| 2780 | req->rq_disk = bio->bi_bdev->bd_disk; |
| 2781 | req->start_time = jiffies; |
| 2782 | |
| 2783 | add_request(q, req); |
| 2784 | out: |
| 2785 | if (freereq) |
| 2786 | __blk_put_request(q, freereq); |
Jens Axboe | 4a534f9 | 2005-04-16 15:25:40 -0700 | [diff] [blame] | 2787 | if (sync) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2788 | __generic_unplug_device(q); |
| 2789 | |
| 2790 | spin_unlock_irq(q->queue_lock); |
| 2791 | return 0; |
| 2792 | |
| 2793 | end_io: |
| 2794 | bio_endio(bio, nr_sectors << 9, err); |
| 2795 | return 0; |
| 2796 | } |
| 2797 | |
| 2798 | /* |
| 2799 | * If bio->bi_dev is a partition, remap the location |
| 2800 | */ |
| 2801 | static inline void blk_partition_remap(struct bio *bio) |
| 2802 | { |
| 2803 | struct block_device *bdev = bio->bi_bdev; |
| 2804 | |
| 2805 | if (bdev != bdev->bd_contains) { |
| 2806 | struct hd_struct *p = bdev->bd_part; |
| 2807 | |
| 2808 | switch (bio->bi_rw) { |
| 2809 | case READ: |
| 2810 | p->read_sectors += bio_sectors(bio); |
| 2811 | p->reads++; |
| 2812 | break; |
| 2813 | case WRITE: |
| 2814 | p->write_sectors += bio_sectors(bio); |
| 2815 | p->writes++; |
| 2816 | break; |
| 2817 | } |
| 2818 | bio->bi_sector += p->start_sect; |
| 2819 | bio->bi_bdev = bdev->bd_contains; |
| 2820 | } |
| 2821 | } |
| 2822 | |
| 2823 | void blk_finish_queue_drain(request_queue_t *q) |
| 2824 | { |
| 2825 | struct request_list *rl = &q->rq; |
| 2826 | struct request *rq; |
| 2827 | |
| 2828 | spin_lock_irq(q->queue_lock); |
| 2829 | clear_bit(QUEUE_FLAG_DRAIN, &q->queue_flags); |
| 2830 | |
| 2831 | while (!list_empty(&q->drain_list)) { |
| 2832 | rq = list_entry_rq(q->drain_list.next); |
| 2833 | |
| 2834 | list_del_init(&rq->queuelist); |
| 2835 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 1); |
| 2836 | } |
| 2837 | |
| 2838 | spin_unlock_irq(q->queue_lock); |
| 2839 | |
| 2840 | wake_up(&rl->wait[0]); |
| 2841 | wake_up(&rl->wait[1]); |
| 2842 | wake_up(&rl->drain); |
| 2843 | } |
| 2844 | |
| 2845 | static int wait_drain(request_queue_t *q, struct request_list *rl, int dispatch) |
| 2846 | { |
| 2847 | int wait = rl->count[READ] + rl->count[WRITE]; |
| 2848 | |
| 2849 | if (dispatch) |
| 2850 | wait += !list_empty(&q->queue_head); |
| 2851 | |
| 2852 | return wait; |
| 2853 | } |
| 2854 | |
| 2855 | /* |
| 2856 | * We rely on the fact that only requests allocated through blk_alloc_request() |
| 2857 | * have io scheduler private data structures associated with them. Any other |
| 2858 | * type of request (allocated on stack or through kmalloc()) should not go |
| 2859 | * to the io scheduler core, but be attached to the queue head instead. |
| 2860 | */ |
| 2861 | void blk_wait_queue_drained(request_queue_t *q, int wait_dispatch) |
| 2862 | { |
| 2863 | struct request_list *rl = &q->rq; |
| 2864 | DEFINE_WAIT(wait); |
| 2865 | |
| 2866 | spin_lock_irq(q->queue_lock); |
| 2867 | set_bit(QUEUE_FLAG_DRAIN, &q->queue_flags); |
| 2868 | |
| 2869 | while (wait_drain(q, rl, wait_dispatch)) { |
| 2870 | prepare_to_wait(&rl->drain, &wait, TASK_UNINTERRUPTIBLE); |
| 2871 | |
| 2872 | if (wait_drain(q, rl, wait_dispatch)) { |
| 2873 | __generic_unplug_device(q); |
| 2874 | spin_unlock_irq(q->queue_lock); |
| 2875 | io_schedule(); |
| 2876 | spin_lock_irq(q->queue_lock); |
| 2877 | } |
| 2878 | |
| 2879 | finish_wait(&rl->drain, &wait); |
| 2880 | } |
| 2881 | |
| 2882 | spin_unlock_irq(q->queue_lock); |
| 2883 | } |
| 2884 | |
| 2885 | /* |
| 2886 | * block waiting for the io scheduler being started again. |
| 2887 | */ |
| 2888 | static inline void block_wait_queue_running(request_queue_t *q) |
| 2889 | { |
| 2890 | DEFINE_WAIT(wait); |
| 2891 | |
| 2892 | while (test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)) { |
| 2893 | struct request_list *rl = &q->rq; |
| 2894 | |
| 2895 | prepare_to_wait_exclusive(&rl->drain, &wait, |
| 2896 | TASK_UNINTERRUPTIBLE); |
| 2897 | |
| 2898 | /* |
| 2899 | * re-check the condition. avoids using prepare_to_wait() |
| 2900 | * in the fast path (queue is running) |
| 2901 | */ |
| 2902 | if (test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)) |
| 2903 | io_schedule(); |
| 2904 | |
| 2905 | finish_wait(&rl->drain, &wait); |
| 2906 | } |
| 2907 | } |
| 2908 | |
| 2909 | static void handle_bad_sector(struct bio *bio) |
| 2910 | { |
| 2911 | char b[BDEVNAME_SIZE]; |
| 2912 | |
| 2913 | printk(KERN_INFO "attempt to access beyond end of device\n"); |
| 2914 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", |
| 2915 | bdevname(bio->bi_bdev, b), |
| 2916 | bio->bi_rw, |
| 2917 | (unsigned long long)bio->bi_sector + bio_sectors(bio), |
| 2918 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); |
| 2919 | |
| 2920 | set_bit(BIO_EOF, &bio->bi_flags); |
| 2921 | } |
| 2922 | |
| 2923 | /** |
| 2924 | * generic_make_request: hand a buffer to its device driver for I/O |
| 2925 | * @bio: The bio describing the location in memory and on the device. |
| 2926 | * |
| 2927 | * generic_make_request() is used to make I/O requests of block |
| 2928 | * devices. It is passed a &struct bio, which describes the I/O that needs |
| 2929 | * to be done. |
| 2930 | * |
| 2931 | * generic_make_request() does not return any status. The |
| 2932 | * success/failure status of the request, along with notification of |
| 2933 | * completion, is delivered asynchronously through the bio->bi_end_io |
| 2934 | * function described (one day) else where. |
| 2935 | * |
| 2936 | * The caller of generic_make_request must make sure that bi_io_vec |
| 2937 | * are set to describe the memory buffer, and that bi_dev and bi_sector are |
| 2938 | * set to describe the device address, and the |
| 2939 | * bi_end_io and optionally bi_private are set to describe how |
| 2940 | * completion notification should be signaled. |
| 2941 | * |
| 2942 | * generic_make_request and the drivers it calls may use bi_next if this |
| 2943 | * bio happens to be merged with someone else, and may change bi_dev and |
| 2944 | * bi_sector for remaps as it sees fit. So the values of these fields |
| 2945 | * should NOT be depended on after the call to generic_make_request. |
| 2946 | */ |
| 2947 | void generic_make_request(struct bio *bio) |
| 2948 | { |
| 2949 | request_queue_t *q; |
| 2950 | sector_t maxsector; |
| 2951 | int ret, nr_sectors = bio_sectors(bio); |
| 2952 | |
| 2953 | might_sleep(); |
| 2954 | /* Test device or partition size, when known. */ |
| 2955 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; |
| 2956 | if (maxsector) { |
| 2957 | sector_t sector = bio->bi_sector; |
| 2958 | |
| 2959 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { |
| 2960 | /* |
| 2961 | * This may well happen - the kernel calls bread() |
| 2962 | * without checking the size of the device, e.g., when |
| 2963 | * mounting a device. |
| 2964 | */ |
| 2965 | handle_bad_sector(bio); |
| 2966 | goto end_io; |
| 2967 | } |
| 2968 | } |
| 2969 | |
| 2970 | /* |
| 2971 | * Resolve the mapping until finished. (drivers are |
| 2972 | * still free to implement/resolve their own stacking |
| 2973 | * by explicitly returning 0) |
| 2974 | * |
| 2975 | * NOTE: we don't repeat the blk_size check for each new device. |
| 2976 | * Stacking drivers are expected to know what they are doing. |
| 2977 | */ |
| 2978 | do { |
| 2979 | char b[BDEVNAME_SIZE]; |
| 2980 | |
| 2981 | q = bdev_get_queue(bio->bi_bdev); |
| 2982 | if (!q) { |
| 2983 | printk(KERN_ERR |
| 2984 | "generic_make_request: Trying to access " |
| 2985 | "nonexistent block-device %s (%Lu)\n", |
| 2986 | bdevname(bio->bi_bdev, b), |
| 2987 | (long long) bio->bi_sector); |
| 2988 | end_io: |
| 2989 | bio_endio(bio, bio->bi_size, -EIO); |
| 2990 | break; |
| 2991 | } |
| 2992 | |
| 2993 | if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) { |
| 2994 | printk("bio too big device %s (%u > %u)\n", |
| 2995 | bdevname(bio->bi_bdev, b), |
| 2996 | bio_sectors(bio), |
| 2997 | q->max_hw_sectors); |
| 2998 | goto end_io; |
| 2999 | } |
| 3000 | |
| 3001 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) |
| 3002 | goto end_io; |
| 3003 | |
| 3004 | block_wait_queue_running(q); |
| 3005 | |
| 3006 | /* |
| 3007 | * If this device has partitions, remap block n |
| 3008 | * of partition p to block n+start(p) of the disk. |
| 3009 | */ |
| 3010 | blk_partition_remap(bio); |
| 3011 | |
| 3012 | ret = q->make_request_fn(q, bio); |
| 3013 | } while (ret); |
| 3014 | } |
| 3015 | |
| 3016 | EXPORT_SYMBOL(generic_make_request); |
| 3017 | |
| 3018 | /** |
| 3019 | * submit_bio: submit a bio to the block device layer for I/O |
| 3020 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
| 3021 | * @bio: The &struct bio which describes the I/O |
| 3022 | * |
| 3023 | * submit_bio() is very similar in purpose to generic_make_request(), and |
| 3024 | * uses that function to do most of the work. Both are fairly rough |
| 3025 | * interfaces, @bio must be presetup and ready for I/O. |
| 3026 | * |
| 3027 | */ |
| 3028 | void submit_bio(int rw, struct bio *bio) |
| 3029 | { |
| 3030 | int count = bio_sectors(bio); |
| 3031 | |
| 3032 | BIO_BUG_ON(!bio->bi_size); |
| 3033 | BIO_BUG_ON(!bio->bi_io_vec); |
| 3034 | bio->bi_rw = rw; |
| 3035 | if (rw & WRITE) |
| 3036 | mod_page_state(pgpgout, count); |
| 3037 | else |
| 3038 | mod_page_state(pgpgin, count); |
| 3039 | |
| 3040 | if (unlikely(block_dump)) { |
| 3041 | char b[BDEVNAME_SIZE]; |
| 3042 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", |
| 3043 | current->comm, current->pid, |
| 3044 | (rw & WRITE) ? "WRITE" : "READ", |
| 3045 | (unsigned long long)bio->bi_sector, |
| 3046 | bdevname(bio->bi_bdev,b)); |
| 3047 | } |
| 3048 | |
| 3049 | generic_make_request(bio); |
| 3050 | } |
| 3051 | |
| 3052 | EXPORT_SYMBOL(submit_bio); |
| 3053 | |
| 3054 | void blk_recalc_rq_segments(struct request *rq) |
| 3055 | { |
| 3056 | struct bio *bio, *prevbio = NULL; |
| 3057 | int nr_phys_segs, nr_hw_segs; |
| 3058 | unsigned int phys_size, hw_size; |
| 3059 | request_queue_t *q = rq->q; |
| 3060 | |
| 3061 | if (!rq->bio) |
| 3062 | return; |
| 3063 | |
| 3064 | phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0; |
| 3065 | rq_for_each_bio(bio, rq) { |
| 3066 | /* Force bio hw/phys segs to be recalculated. */ |
| 3067 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
| 3068 | |
| 3069 | nr_phys_segs += bio_phys_segments(q, bio); |
| 3070 | nr_hw_segs += bio_hw_segments(q, bio); |
| 3071 | if (prevbio) { |
| 3072 | int pseg = phys_size + prevbio->bi_size + bio->bi_size; |
| 3073 | int hseg = hw_size + prevbio->bi_size + bio->bi_size; |
| 3074 | |
| 3075 | if (blk_phys_contig_segment(q, prevbio, bio) && |
| 3076 | pseg <= q->max_segment_size) { |
| 3077 | nr_phys_segs--; |
| 3078 | phys_size += prevbio->bi_size + bio->bi_size; |
| 3079 | } else |
| 3080 | phys_size = 0; |
| 3081 | |
| 3082 | if (blk_hw_contig_segment(q, prevbio, bio) && |
| 3083 | hseg <= q->max_segment_size) { |
| 3084 | nr_hw_segs--; |
| 3085 | hw_size += prevbio->bi_size + bio->bi_size; |
| 3086 | } else |
| 3087 | hw_size = 0; |
| 3088 | } |
| 3089 | prevbio = bio; |
| 3090 | } |
| 3091 | |
| 3092 | rq->nr_phys_segments = nr_phys_segs; |
| 3093 | rq->nr_hw_segments = nr_hw_segs; |
| 3094 | } |
| 3095 | |
| 3096 | void blk_recalc_rq_sectors(struct request *rq, int nsect) |
| 3097 | { |
| 3098 | if (blk_fs_request(rq)) { |
| 3099 | rq->hard_sector += nsect; |
| 3100 | rq->hard_nr_sectors -= nsect; |
| 3101 | |
| 3102 | /* |
| 3103 | * Move the I/O submission pointers ahead if required. |
| 3104 | */ |
| 3105 | if ((rq->nr_sectors >= rq->hard_nr_sectors) && |
| 3106 | (rq->sector <= rq->hard_sector)) { |
| 3107 | rq->sector = rq->hard_sector; |
| 3108 | rq->nr_sectors = rq->hard_nr_sectors; |
| 3109 | rq->hard_cur_sectors = bio_cur_sectors(rq->bio); |
| 3110 | rq->current_nr_sectors = rq->hard_cur_sectors; |
| 3111 | rq->buffer = bio_data(rq->bio); |
| 3112 | } |
| 3113 | |
| 3114 | /* |
| 3115 | * if total number of sectors is less than the first segment |
| 3116 | * size, something has gone terribly wrong |
| 3117 | */ |
| 3118 | if (rq->nr_sectors < rq->current_nr_sectors) { |
| 3119 | printk("blk: request botched\n"); |
| 3120 | rq->nr_sectors = rq->current_nr_sectors; |
| 3121 | } |
| 3122 | } |
| 3123 | } |
| 3124 | |
| 3125 | static int __end_that_request_first(struct request *req, int uptodate, |
| 3126 | int nr_bytes) |
| 3127 | { |
| 3128 | int total_bytes, bio_nbytes, error, next_idx = 0; |
| 3129 | struct bio *bio; |
| 3130 | |
| 3131 | /* |
| 3132 | * extend uptodate bool to allow < 0 value to be direct io error |
| 3133 | */ |
| 3134 | error = 0; |
| 3135 | if (end_io_error(uptodate)) |
| 3136 | error = !uptodate ? -EIO : uptodate; |
| 3137 | |
| 3138 | /* |
| 3139 | * for a REQ_BLOCK_PC request, we want to carry any eventual |
| 3140 | * sense key with us all the way through |
| 3141 | */ |
| 3142 | if (!blk_pc_request(req)) |
| 3143 | req->errors = 0; |
| 3144 | |
| 3145 | if (!uptodate) { |
| 3146 | if (blk_fs_request(req) && !(req->flags & REQ_QUIET)) |
| 3147 | printk("end_request: I/O error, dev %s, sector %llu\n", |
| 3148 | req->rq_disk ? req->rq_disk->disk_name : "?", |
| 3149 | (unsigned long long)req->sector); |
| 3150 | } |
| 3151 | |
| 3152 | total_bytes = bio_nbytes = 0; |
| 3153 | while ((bio = req->bio) != NULL) { |
| 3154 | int nbytes; |
| 3155 | |
| 3156 | if (nr_bytes >= bio->bi_size) { |
| 3157 | req->bio = bio->bi_next; |
| 3158 | nbytes = bio->bi_size; |
| 3159 | bio_endio(bio, nbytes, error); |
| 3160 | next_idx = 0; |
| 3161 | bio_nbytes = 0; |
| 3162 | } else { |
| 3163 | int idx = bio->bi_idx + next_idx; |
| 3164 | |
| 3165 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { |
| 3166 | blk_dump_rq_flags(req, "__end_that"); |
| 3167 | printk("%s: bio idx %d >= vcnt %d\n", |
| 3168 | __FUNCTION__, |
| 3169 | bio->bi_idx, bio->bi_vcnt); |
| 3170 | break; |
| 3171 | } |
| 3172 | |
| 3173 | nbytes = bio_iovec_idx(bio, idx)->bv_len; |
| 3174 | BIO_BUG_ON(nbytes > bio->bi_size); |
| 3175 | |
| 3176 | /* |
| 3177 | * not a complete bvec done |
| 3178 | */ |
| 3179 | if (unlikely(nbytes > nr_bytes)) { |
| 3180 | bio_nbytes += nr_bytes; |
| 3181 | total_bytes += nr_bytes; |
| 3182 | break; |
| 3183 | } |
| 3184 | |
| 3185 | /* |
| 3186 | * advance to the next vector |
| 3187 | */ |
| 3188 | next_idx++; |
| 3189 | bio_nbytes += nbytes; |
| 3190 | } |
| 3191 | |
| 3192 | total_bytes += nbytes; |
| 3193 | nr_bytes -= nbytes; |
| 3194 | |
| 3195 | if ((bio = req->bio)) { |
| 3196 | /* |
| 3197 | * end more in this run, or just return 'not-done' |
| 3198 | */ |
| 3199 | if (unlikely(nr_bytes <= 0)) |
| 3200 | break; |
| 3201 | } |
| 3202 | } |
| 3203 | |
| 3204 | /* |
| 3205 | * completely done |
| 3206 | */ |
| 3207 | if (!req->bio) |
| 3208 | return 0; |
| 3209 | |
| 3210 | /* |
| 3211 | * if the request wasn't completed, update state |
| 3212 | */ |
| 3213 | if (bio_nbytes) { |
| 3214 | bio_endio(bio, bio_nbytes, error); |
| 3215 | bio->bi_idx += next_idx; |
| 3216 | bio_iovec(bio)->bv_offset += nr_bytes; |
| 3217 | bio_iovec(bio)->bv_len -= nr_bytes; |
| 3218 | } |
| 3219 | |
| 3220 | blk_recalc_rq_sectors(req, total_bytes >> 9); |
| 3221 | blk_recalc_rq_segments(req); |
| 3222 | return 1; |
| 3223 | } |
| 3224 | |
| 3225 | /** |
| 3226 | * end_that_request_first - end I/O on a request |
| 3227 | * @req: the request being processed |
| 3228 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error |
| 3229 | * @nr_sectors: number of sectors to end I/O on |
| 3230 | * |
| 3231 | * Description: |
| 3232 | * Ends I/O on a number of sectors attached to @req, and sets it up |
| 3233 | * for the next range of segments (if any) in the cluster. |
| 3234 | * |
| 3235 | * Return: |
| 3236 | * 0 - we are done with this request, call end_that_request_last() |
| 3237 | * 1 - still buffers pending for this request |
| 3238 | **/ |
| 3239 | int end_that_request_first(struct request *req, int uptodate, int nr_sectors) |
| 3240 | { |
| 3241 | return __end_that_request_first(req, uptodate, nr_sectors << 9); |
| 3242 | } |
| 3243 | |
| 3244 | EXPORT_SYMBOL(end_that_request_first); |
| 3245 | |
| 3246 | /** |
| 3247 | * end_that_request_chunk - end I/O on a request |
| 3248 | * @req: the request being processed |
| 3249 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error |
| 3250 | * @nr_bytes: number of bytes to complete |
| 3251 | * |
| 3252 | * Description: |
| 3253 | * Ends I/O on a number of bytes attached to @req, and sets it up |
| 3254 | * for the next range of segments (if any). Like end_that_request_first(), |
| 3255 | * but deals with bytes instead of sectors. |
| 3256 | * |
| 3257 | * Return: |
| 3258 | * 0 - we are done with this request, call end_that_request_last() |
| 3259 | * 1 - still buffers pending for this request |
| 3260 | **/ |
| 3261 | int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes) |
| 3262 | { |
| 3263 | return __end_that_request_first(req, uptodate, nr_bytes); |
| 3264 | } |
| 3265 | |
| 3266 | EXPORT_SYMBOL(end_that_request_chunk); |
| 3267 | |
| 3268 | /* |
| 3269 | * queue lock must be held |
| 3270 | */ |
| 3271 | void end_that_request_last(struct request *req) |
| 3272 | { |
| 3273 | struct gendisk *disk = req->rq_disk; |
| 3274 | |
| 3275 | if (unlikely(laptop_mode) && blk_fs_request(req)) |
| 3276 | laptop_io_completion(); |
| 3277 | |
| 3278 | if (disk && blk_fs_request(req)) { |
| 3279 | unsigned long duration = jiffies - req->start_time; |
| 3280 | switch (rq_data_dir(req)) { |
| 3281 | case WRITE: |
| 3282 | __disk_stat_inc(disk, writes); |
| 3283 | __disk_stat_add(disk, write_ticks, duration); |
| 3284 | break; |
| 3285 | case READ: |
| 3286 | __disk_stat_inc(disk, reads); |
| 3287 | __disk_stat_add(disk, read_ticks, duration); |
| 3288 | break; |
| 3289 | } |
| 3290 | disk_round_stats(disk); |
| 3291 | disk->in_flight--; |
| 3292 | } |
| 3293 | if (req->end_io) |
| 3294 | req->end_io(req); |
| 3295 | else |
| 3296 | __blk_put_request(req->q, req); |
| 3297 | } |
| 3298 | |
| 3299 | EXPORT_SYMBOL(end_that_request_last); |
| 3300 | |
| 3301 | void end_request(struct request *req, int uptodate) |
| 3302 | { |
| 3303 | if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) { |
| 3304 | add_disk_randomness(req->rq_disk); |
| 3305 | blkdev_dequeue_request(req); |
| 3306 | end_that_request_last(req); |
| 3307 | } |
| 3308 | } |
| 3309 | |
| 3310 | EXPORT_SYMBOL(end_request); |
| 3311 | |
| 3312 | void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio) |
| 3313 | { |
| 3314 | /* first three bits are identical in rq->flags and bio->bi_rw */ |
| 3315 | rq->flags |= (bio->bi_rw & 7); |
| 3316 | |
| 3317 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
| 3318 | rq->nr_hw_segments = bio_hw_segments(q, bio); |
| 3319 | rq->current_nr_sectors = bio_cur_sectors(bio); |
| 3320 | rq->hard_cur_sectors = rq->current_nr_sectors; |
| 3321 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); |
| 3322 | rq->buffer = bio_data(bio); |
| 3323 | |
| 3324 | rq->bio = rq->biotail = bio; |
| 3325 | } |
| 3326 | |
| 3327 | EXPORT_SYMBOL(blk_rq_bio_prep); |
| 3328 | |
| 3329 | int kblockd_schedule_work(struct work_struct *work) |
| 3330 | { |
| 3331 | return queue_work(kblockd_workqueue, work); |
| 3332 | } |
| 3333 | |
| 3334 | EXPORT_SYMBOL(kblockd_schedule_work); |
| 3335 | |
| 3336 | void kblockd_flush(void) |
| 3337 | { |
| 3338 | flush_workqueue(kblockd_workqueue); |
| 3339 | } |
| 3340 | EXPORT_SYMBOL(kblockd_flush); |
| 3341 | |
| 3342 | int __init blk_dev_init(void) |
| 3343 | { |
| 3344 | kblockd_workqueue = create_workqueue("kblockd"); |
| 3345 | if (!kblockd_workqueue) |
| 3346 | panic("Failed to create kblockd\n"); |
| 3347 | |
| 3348 | request_cachep = kmem_cache_create("blkdev_requests", |
| 3349 | sizeof(struct request), 0, SLAB_PANIC, NULL, NULL); |
| 3350 | |
| 3351 | requestq_cachep = kmem_cache_create("blkdev_queue", |
| 3352 | sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL); |
| 3353 | |
| 3354 | iocontext_cachep = kmem_cache_create("blkdev_ioc", |
| 3355 | sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL); |
| 3356 | |
| 3357 | blk_max_low_pfn = max_low_pfn; |
| 3358 | blk_max_pfn = max_pfn; |
| 3359 | |
| 3360 | return 0; |
| 3361 | } |
| 3362 | |
| 3363 | /* |
| 3364 | * IO Context helper functions |
| 3365 | */ |
| 3366 | void put_io_context(struct io_context *ioc) |
| 3367 | { |
| 3368 | if (ioc == NULL) |
| 3369 | return; |
| 3370 | |
| 3371 | BUG_ON(atomic_read(&ioc->refcount) == 0); |
| 3372 | |
| 3373 | if (atomic_dec_and_test(&ioc->refcount)) { |
| 3374 | if (ioc->aic && ioc->aic->dtor) |
| 3375 | ioc->aic->dtor(ioc->aic); |
| 3376 | if (ioc->cic && ioc->cic->dtor) |
| 3377 | ioc->cic->dtor(ioc->cic); |
| 3378 | |
| 3379 | kmem_cache_free(iocontext_cachep, ioc); |
| 3380 | } |
| 3381 | } |
| 3382 | EXPORT_SYMBOL(put_io_context); |
| 3383 | |
| 3384 | /* Called by the exitting task */ |
| 3385 | void exit_io_context(void) |
| 3386 | { |
| 3387 | unsigned long flags; |
| 3388 | struct io_context *ioc; |
| 3389 | |
| 3390 | local_irq_save(flags); |
| 3391 | ioc = current->io_context; |
| 3392 | current->io_context = NULL; |
| 3393 | local_irq_restore(flags); |
| 3394 | |
| 3395 | if (ioc->aic && ioc->aic->exit) |
| 3396 | ioc->aic->exit(ioc->aic); |
| 3397 | if (ioc->cic && ioc->cic->exit) |
| 3398 | ioc->cic->exit(ioc->cic); |
| 3399 | |
| 3400 | put_io_context(ioc); |
| 3401 | } |
| 3402 | |
| 3403 | /* |
| 3404 | * If the current task has no IO context then create one and initialise it. |
| 3405 | * If it does have a context, take a ref on it. |
| 3406 | * |
| 3407 | * This is always called in the context of the task which submitted the I/O. |
| 3408 | * But weird things happen, so we disable local interrupts to ensure exclusive |
| 3409 | * access to *current. |
| 3410 | */ |
| 3411 | struct io_context *get_io_context(int gfp_flags) |
| 3412 | { |
| 3413 | struct task_struct *tsk = current; |
| 3414 | unsigned long flags; |
| 3415 | struct io_context *ret; |
| 3416 | |
| 3417 | local_irq_save(flags); |
| 3418 | ret = tsk->io_context; |
| 3419 | if (ret) |
| 3420 | goto out; |
| 3421 | |
| 3422 | local_irq_restore(flags); |
| 3423 | |
| 3424 | ret = kmem_cache_alloc(iocontext_cachep, gfp_flags); |
| 3425 | if (ret) { |
| 3426 | atomic_set(&ret->refcount, 1); |
| 3427 | ret->pid = tsk->pid; |
| 3428 | ret->last_waited = jiffies; /* doesn't matter... */ |
| 3429 | ret->nr_batch_requests = 0; /* because this is 0 */ |
| 3430 | ret->aic = NULL; |
| 3431 | ret->cic = NULL; |
| 3432 | spin_lock_init(&ret->lock); |
| 3433 | |
| 3434 | local_irq_save(flags); |
| 3435 | |
| 3436 | /* |
| 3437 | * very unlikely, someone raced with us in setting up the task |
| 3438 | * io context. free new context and just grab a reference. |
| 3439 | */ |
| 3440 | if (!tsk->io_context) |
| 3441 | tsk->io_context = ret; |
| 3442 | else { |
| 3443 | kmem_cache_free(iocontext_cachep, ret); |
| 3444 | ret = tsk->io_context; |
| 3445 | } |
| 3446 | |
| 3447 | out: |
| 3448 | atomic_inc(&ret->refcount); |
| 3449 | local_irq_restore(flags); |
| 3450 | } |
| 3451 | |
| 3452 | return ret; |
| 3453 | } |
| 3454 | EXPORT_SYMBOL(get_io_context); |
| 3455 | |
| 3456 | void copy_io_context(struct io_context **pdst, struct io_context **psrc) |
| 3457 | { |
| 3458 | struct io_context *src = *psrc; |
| 3459 | struct io_context *dst = *pdst; |
| 3460 | |
| 3461 | if (src) { |
| 3462 | BUG_ON(atomic_read(&src->refcount) == 0); |
| 3463 | atomic_inc(&src->refcount); |
| 3464 | put_io_context(dst); |
| 3465 | *pdst = src; |
| 3466 | } |
| 3467 | } |
| 3468 | EXPORT_SYMBOL(copy_io_context); |
| 3469 | |
| 3470 | void swap_io_context(struct io_context **ioc1, struct io_context **ioc2) |
| 3471 | { |
| 3472 | struct io_context *temp; |
| 3473 | temp = *ioc1; |
| 3474 | *ioc1 = *ioc2; |
| 3475 | *ioc2 = temp; |
| 3476 | } |
| 3477 | EXPORT_SYMBOL(swap_io_context); |
| 3478 | |
| 3479 | /* |
| 3480 | * sysfs parts below |
| 3481 | */ |
| 3482 | struct queue_sysfs_entry { |
| 3483 | struct attribute attr; |
| 3484 | ssize_t (*show)(struct request_queue *, char *); |
| 3485 | ssize_t (*store)(struct request_queue *, const char *, size_t); |
| 3486 | }; |
| 3487 | |
| 3488 | static ssize_t |
| 3489 | queue_var_show(unsigned int var, char *page) |
| 3490 | { |
| 3491 | return sprintf(page, "%d\n", var); |
| 3492 | } |
| 3493 | |
| 3494 | static ssize_t |
| 3495 | queue_var_store(unsigned long *var, const char *page, size_t count) |
| 3496 | { |
| 3497 | char *p = (char *) page; |
| 3498 | |
| 3499 | *var = simple_strtoul(p, &p, 10); |
| 3500 | return count; |
| 3501 | } |
| 3502 | |
| 3503 | static ssize_t queue_requests_show(struct request_queue *q, char *page) |
| 3504 | { |
| 3505 | return queue_var_show(q->nr_requests, (page)); |
| 3506 | } |
| 3507 | |
| 3508 | static ssize_t |
| 3509 | queue_requests_store(struct request_queue *q, const char *page, size_t count) |
| 3510 | { |
| 3511 | struct request_list *rl = &q->rq; |
| 3512 | |
| 3513 | int ret = queue_var_store(&q->nr_requests, page, count); |
| 3514 | if (q->nr_requests < BLKDEV_MIN_RQ) |
| 3515 | q->nr_requests = BLKDEV_MIN_RQ; |
| 3516 | blk_queue_congestion_threshold(q); |
| 3517 | |
| 3518 | if (rl->count[READ] >= queue_congestion_on_threshold(q)) |
| 3519 | set_queue_congested(q, READ); |
| 3520 | else if (rl->count[READ] < queue_congestion_off_threshold(q)) |
| 3521 | clear_queue_congested(q, READ); |
| 3522 | |
| 3523 | if (rl->count[WRITE] >= queue_congestion_on_threshold(q)) |
| 3524 | set_queue_congested(q, WRITE); |
| 3525 | else if (rl->count[WRITE] < queue_congestion_off_threshold(q)) |
| 3526 | clear_queue_congested(q, WRITE); |
| 3527 | |
| 3528 | if (rl->count[READ] >= q->nr_requests) { |
| 3529 | blk_set_queue_full(q, READ); |
| 3530 | } else if (rl->count[READ]+1 <= q->nr_requests) { |
| 3531 | blk_clear_queue_full(q, READ); |
| 3532 | wake_up(&rl->wait[READ]); |
| 3533 | } |
| 3534 | |
| 3535 | if (rl->count[WRITE] >= q->nr_requests) { |
| 3536 | blk_set_queue_full(q, WRITE); |
| 3537 | } else if (rl->count[WRITE]+1 <= q->nr_requests) { |
| 3538 | blk_clear_queue_full(q, WRITE); |
| 3539 | wake_up(&rl->wait[WRITE]); |
| 3540 | } |
| 3541 | return ret; |
| 3542 | } |
| 3543 | |
| 3544 | static ssize_t queue_ra_show(struct request_queue *q, char *page) |
| 3545 | { |
| 3546 | int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); |
| 3547 | |
| 3548 | return queue_var_show(ra_kb, (page)); |
| 3549 | } |
| 3550 | |
| 3551 | static ssize_t |
| 3552 | queue_ra_store(struct request_queue *q, const char *page, size_t count) |
| 3553 | { |
| 3554 | unsigned long ra_kb; |
| 3555 | ssize_t ret = queue_var_store(&ra_kb, page, count); |
| 3556 | |
| 3557 | spin_lock_irq(q->queue_lock); |
| 3558 | if (ra_kb > (q->max_sectors >> 1)) |
| 3559 | ra_kb = (q->max_sectors >> 1); |
| 3560 | |
| 3561 | q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); |
| 3562 | spin_unlock_irq(q->queue_lock); |
| 3563 | |
| 3564 | return ret; |
| 3565 | } |
| 3566 | |
| 3567 | static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) |
| 3568 | { |
| 3569 | int max_sectors_kb = q->max_sectors >> 1; |
| 3570 | |
| 3571 | return queue_var_show(max_sectors_kb, (page)); |
| 3572 | } |
| 3573 | |
| 3574 | static ssize_t |
| 3575 | queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) |
| 3576 | { |
| 3577 | unsigned long max_sectors_kb, |
| 3578 | max_hw_sectors_kb = q->max_hw_sectors >> 1, |
| 3579 | page_kb = 1 << (PAGE_CACHE_SHIFT - 10); |
| 3580 | ssize_t ret = queue_var_store(&max_sectors_kb, page, count); |
| 3581 | int ra_kb; |
| 3582 | |
| 3583 | if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) |
| 3584 | return -EINVAL; |
| 3585 | /* |
| 3586 | * Take the queue lock to update the readahead and max_sectors |
| 3587 | * values synchronously: |
| 3588 | */ |
| 3589 | spin_lock_irq(q->queue_lock); |
| 3590 | /* |
| 3591 | * Trim readahead window as well, if necessary: |
| 3592 | */ |
| 3593 | ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); |
| 3594 | if (ra_kb > max_sectors_kb) |
| 3595 | q->backing_dev_info.ra_pages = |
| 3596 | max_sectors_kb >> (PAGE_CACHE_SHIFT - 10); |
| 3597 | |
| 3598 | q->max_sectors = max_sectors_kb << 1; |
| 3599 | spin_unlock_irq(q->queue_lock); |
| 3600 | |
| 3601 | return ret; |
| 3602 | } |
| 3603 | |
| 3604 | static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) |
| 3605 | { |
| 3606 | int max_hw_sectors_kb = q->max_hw_sectors >> 1; |
| 3607 | |
| 3608 | return queue_var_show(max_hw_sectors_kb, (page)); |
| 3609 | } |
| 3610 | |
| 3611 | |
| 3612 | static struct queue_sysfs_entry queue_requests_entry = { |
| 3613 | .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, |
| 3614 | .show = queue_requests_show, |
| 3615 | .store = queue_requests_store, |
| 3616 | }; |
| 3617 | |
| 3618 | static struct queue_sysfs_entry queue_ra_entry = { |
| 3619 | .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, |
| 3620 | .show = queue_ra_show, |
| 3621 | .store = queue_ra_store, |
| 3622 | }; |
| 3623 | |
| 3624 | static struct queue_sysfs_entry queue_max_sectors_entry = { |
| 3625 | .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, |
| 3626 | .show = queue_max_sectors_show, |
| 3627 | .store = queue_max_sectors_store, |
| 3628 | }; |
| 3629 | |
| 3630 | static struct queue_sysfs_entry queue_max_hw_sectors_entry = { |
| 3631 | .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, |
| 3632 | .show = queue_max_hw_sectors_show, |
| 3633 | }; |
| 3634 | |
| 3635 | static struct queue_sysfs_entry queue_iosched_entry = { |
| 3636 | .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, |
| 3637 | .show = elv_iosched_show, |
| 3638 | .store = elv_iosched_store, |
| 3639 | }; |
| 3640 | |
| 3641 | static struct attribute *default_attrs[] = { |
| 3642 | &queue_requests_entry.attr, |
| 3643 | &queue_ra_entry.attr, |
| 3644 | &queue_max_hw_sectors_entry.attr, |
| 3645 | &queue_max_sectors_entry.attr, |
| 3646 | &queue_iosched_entry.attr, |
| 3647 | NULL, |
| 3648 | }; |
| 3649 | |
| 3650 | #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) |
| 3651 | |
| 3652 | static ssize_t |
| 3653 | queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) |
| 3654 | { |
| 3655 | struct queue_sysfs_entry *entry = to_queue(attr); |
| 3656 | struct request_queue *q; |
| 3657 | |
| 3658 | q = container_of(kobj, struct request_queue, kobj); |
| 3659 | if (!entry->show) |
| 3660 | return 0; |
| 3661 | |
| 3662 | return entry->show(q, page); |
| 3663 | } |
| 3664 | |
| 3665 | static ssize_t |
| 3666 | queue_attr_store(struct kobject *kobj, struct attribute *attr, |
| 3667 | const char *page, size_t length) |
| 3668 | { |
| 3669 | struct queue_sysfs_entry *entry = to_queue(attr); |
| 3670 | struct request_queue *q; |
| 3671 | |
| 3672 | q = container_of(kobj, struct request_queue, kobj); |
| 3673 | if (!entry->store) |
| 3674 | return -EINVAL; |
| 3675 | |
| 3676 | return entry->store(q, page, length); |
| 3677 | } |
| 3678 | |
| 3679 | static struct sysfs_ops queue_sysfs_ops = { |
| 3680 | .show = queue_attr_show, |
| 3681 | .store = queue_attr_store, |
| 3682 | }; |
| 3683 | |
| 3684 | struct kobj_type queue_ktype = { |
| 3685 | .sysfs_ops = &queue_sysfs_ops, |
| 3686 | .default_attrs = default_attrs, |
| 3687 | }; |
| 3688 | |
| 3689 | int blk_register_queue(struct gendisk *disk) |
| 3690 | { |
| 3691 | int ret; |
| 3692 | |
| 3693 | request_queue_t *q = disk->queue; |
| 3694 | |
| 3695 | if (!q || !q->request_fn) |
| 3696 | return -ENXIO; |
| 3697 | |
| 3698 | q->kobj.parent = kobject_get(&disk->kobj); |
| 3699 | if (!q->kobj.parent) |
| 3700 | return -EBUSY; |
| 3701 | |
| 3702 | snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue"); |
| 3703 | q->kobj.ktype = &queue_ktype; |
| 3704 | |
| 3705 | ret = kobject_register(&q->kobj); |
| 3706 | if (ret < 0) |
| 3707 | return ret; |
| 3708 | |
| 3709 | ret = elv_register_queue(q); |
| 3710 | if (ret) { |
| 3711 | kobject_unregister(&q->kobj); |
| 3712 | return ret; |
| 3713 | } |
| 3714 | |
| 3715 | return 0; |
| 3716 | } |
| 3717 | |
| 3718 | void blk_unregister_queue(struct gendisk *disk) |
| 3719 | { |
| 3720 | request_queue_t *q = disk->queue; |
| 3721 | |
| 3722 | if (q && q->request_fn) { |
| 3723 | elv_unregister_queue(q); |
| 3724 | |
| 3725 | kobject_unregister(&q->kobj); |
| 3726 | kobject_put(&disk->kobj); |
| 3727 | } |
| 3728 | } |