]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1991, 1992 Linus Torvalds |
3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6728cb0e JA |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
7 | * - July2000 | |
1da177e4 LT |
8 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
9 | */ | |
10 | ||
11 | /* | |
12 | * This handles all read/write requests to block devices | |
13 | */ | |
1da177e4 LT |
14 | #include <linux/kernel.h> |
15 | #include <linux/module.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/blkdev.h> | |
19 | #include <linux/highmem.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/kernel_stat.h> | |
22 | #include <linux/string.h> | |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/completion.h> |
25 | #include <linux/slab.h> | |
26 | #include <linux/swap.h> | |
27 | #include <linux/writeback.h> | |
faccbd4b | 28 | #include <linux/task_io_accounting_ops.h> |
ff856bad JA |
29 | #include <linux/interrupt.h> |
30 | #include <linux/cpu.h> | |
2056a782 | 31 | #include <linux/blktrace_api.h> |
c17bb495 | 32 | #include <linux/fault-inject.h> |
1da177e4 | 33 | |
8324aa91 JA |
34 | #include "blk.h" |
35 | ||
165125e1 | 36 | static int __make_request(struct request_queue *q, struct bio *bio); |
1da177e4 LT |
37 | |
38 | /* | |
39 | * For the allocated request tables | |
40 | */ | |
5ece6c52 | 41 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
42 | |
43 | /* | |
44 | * For queue allocation | |
45 | */ | |
6728cb0e | 46 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 47 | |
1da177e4 LT |
48 | /* |
49 | * Controlling structure to kblockd | |
50 | */ | |
ff856bad | 51 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 52 | |
ff856bad JA |
53 | static DEFINE_PER_CPU(struct list_head, blk_cpu_done); |
54 | ||
26b8256e JA |
55 | static void drive_stat_acct(struct request *rq, int new_io) |
56 | { | |
57 | int rw = rq_data_dir(rq); | |
58 | ||
59 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
60 | return; | |
61 | ||
62 | if (!new_io) { | |
6f2576af | 63 | __all_stat_inc(rq->rq_disk, merges[rw], rq->sector); |
26b8256e | 64 | } else { |
6f2576af | 65 | struct hd_struct *part = get_part(rq->rq_disk, rq->sector); |
26b8256e JA |
66 | disk_round_stats(rq->rq_disk); |
67 | rq->rq_disk->in_flight++; | |
6f2576af JM |
68 | if (part) { |
69 | part_round_stats(part); | |
70 | part->in_flight++; | |
71 | } | |
26b8256e JA |
72 | } |
73 | } | |
74 | ||
8324aa91 | 75 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
76 | { |
77 | int nr; | |
78 | ||
79 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
80 | if (nr > q->nr_requests) | |
81 | nr = q->nr_requests; | |
82 | q->nr_congestion_on = nr; | |
83 | ||
84 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
85 | if (nr < 1) | |
86 | nr = 1; | |
87 | q->nr_congestion_off = nr; | |
88 | } | |
89 | ||
1da177e4 LT |
90 | /** |
91 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
92 | * @bdev: device | |
93 | * | |
94 | * Locates the passed device's request queue and returns the address of its | |
95 | * backing_dev_info | |
96 | * | |
97 | * Will return NULL if the request queue cannot be located. | |
98 | */ | |
99 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
100 | { | |
101 | struct backing_dev_info *ret = NULL; | |
165125e1 | 102 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
103 | |
104 | if (q) | |
105 | ret = &q->backing_dev_info; | |
106 | return ret; | |
107 | } | |
1da177e4 LT |
108 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
109 | ||
63a71386 JA |
110 | /* |
111 | * We can't just memset() the structure, since the allocation path | |
112 | * already stored some information in the request. | |
113 | */ | |
86db1e29 | 114 | void rq_init(struct request_queue *q, struct request *rq) |
1da177e4 LT |
115 | { |
116 | INIT_LIST_HEAD(&rq->queuelist); | |
ff856bad | 117 | INIT_LIST_HEAD(&rq->donelist); |
63a71386 JA |
118 | rq->q = q; |
119 | rq->sector = rq->hard_sector = (sector_t) -1; | |
120 | rq->nr_sectors = rq->hard_nr_sectors = 0; | |
121 | rq->current_nr_sectors = rq->hard_cur_sectors = 0; | |
1da177e4 | 122 | rq->bio = rq->biotail = NULL; |
2e662b65 JA |
123 | INIT_HLIST_NODE(&rq->hash); |
124 | RB_CLEAR_NODE(&rq->rb_node); | |
63a71386 JA |
125 | rq->rq_disk = NULL; |
126 | rq->nr_phys_segments = 0; | |
127 | rq->nr_hw_segments = 0; | |
22e2c507 | 128 | rq->ioprio = 0; |
63a71386 | 129 | rq->special = NULL; |
6b00769f | 130 | rq->raw_data_len = 0; |
1da177e4 | 131 | rq->buffer = NULL; |
63a71386 JA |
132 | rq->tag = -1; |
133 | rq->errors = 0; | |
1da177e4 | 134 | rq->ref_count = 1; |
63a71386 JA |
135 | rq->cmd_len = 0; |
136 | memset(rq->cmd, 0, sizeof(rq->cmd)); | |
1da177e4 | 137 | rq->data_len = 0; |
63a71386 | 138 | rq->sense_len = 0; |
1da177e4 LT |
139 | rq->data = NULL; |
140 | rq->sense = NULL; | |
141 | rq->end_io = NULL; | |
142 | rq->end_io_data = NULL; | |
abae1fde | 143 | rq->next_rq = NULL; |
1da177e4 LT |
144 | } |
145 | ||
5bb23a68 N |
146 | static void req_bio_endio(struct request *rq, struct bio *bio, |
147 | unsigned int nbytes, int error) | |
1da177e4 | 148 | { |
165125e1 | 149 | struct request_queue *q = rq->q; |
797e7dbb | 150 | |
5bb23a68 N |
151 | if (&q->bar_rq != rq) { |
152 | if (error) | |
153 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
154 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
155 | error = -EIO; | |
797e7dbb | 156 | |
5bb23a68 | 157 | if (unlikely(nbytes > bio->bi_size)) { |
6728cb0e | 158 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", |
5bb23a68 N |
159 | __FUNCTION__, nbytes, bio->bi_size); |
160 | nbytes = bio->bi_size; | |
161 | } | |
797e7dbb | 162 | |
5bb23a68 N |
163 | bio->bi_size -= nbytes; |
164 | bio->bi_sector += (nbytes >> 9); | |
165 | if (bio->bi_size == 0) | |
6712ecf8 | 166 | bio_endio(bio, error); |
5bb23a68 N |
167 | } else { |
168 | ||
169 | /* | |
170 | * Okay, this is the barrier request in progress, just | |
171 | * record the error; | |
172 | */ | |
173 | if (error && !q->orderr) | |
174 | q->orderr = error; | |
175 | } | |
1da177e4 | 176 | } |
1da177e4 | 177 | |
1da177e4 LT |
178 | void blk_dump_rq_flags(struct request *rq, char *msg) |
179 | { | |
180 | int bit; | |
181 | ||
6728cb0e | 182 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
183 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
184 | rq->cmd_flags); | |
1da177e4 | 185 | |
6728cb0e JA |
186 | printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n", |
187 | (unsigned long long)rq->sector, | |
188 | rq->nr_sectors, | |
189 | rq->current_nr_sectors); | |
190 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n", | |
191 | rq->bio, rq->biotail, | |
192 | rq->buffer, rq->data, | |
193 | rq->data_len); | |
1da177e4 | 194 | |
4aff5e23 | 195 | if (blk_pc_request(rq)) { |
6728cb0e | 196 | printk(KERN_INFO " cdb: "); |
1da177e4 LT |
197 | for (bit = 0; bit < sizeof(rq->cmd); bit++) |
198 | printk("%02x ", rq->cmd[bit]); | |
199 | printk("\n"); | |
200 | } | |
201 | } | |
1da177e4 LT |
202 | EXPORT_SYMBOL(blk_dump_rq_flags); |
203 | ||
1da177e4 LT |
204 | /* |
205 | * "plug" the device if there are no outstanding requests: this will | |
206 | * force the transfer to start only after we have put all the requests | |
207 | * on the list. | |
208 | * | |
209 | * This is called with interrupts off and no requests on the queue and | |
210 | * with the queue lock held. | |
211 | */ | |
165125e1 | 212 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
213 | { |
214 | WARN_ON(!irqs_disabled()); | |
215 | ||
216 | /* | |
217 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
218 | * which will restart the queueing | |
219 | */ | |
7daac490 | 220 | if (blk_queue_stopped(q)) |
1da177e4 LT |
221 | return; |
222 | ||
2056a782 | 223 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) { |
1da177e4 | 224 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
2056a782 JA |
225 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); |
226 | } | |
1da177e4 | 227 | } |
1da177e4 LT |
228 | EXPORT_SYMBOL(blk_plug_device); |
229 | ||
230 | /* | |
231 | * remove the queue from the plugged list, if present. called with | |
232 | * queue lock held and interrupts disabled. | |
233 | */ | |
165125e1 | 234 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
235 | { |
236 | WARN_ON(!irqs_disabled()); | |
237 | ||
238 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | |
239 | return 0; | |
240 | ||
241 | del_timer(&q->unplug_timer); | |
242 | return 1; | |
243 | } | |
1da177e4 LT |
244 | EXPORT_SYMBOL(blk_remove_plug); |
245 | ||
246 | /* | |
247 | * remove the plug and let it rip.. | |
248 | */ | |
165125e1 | 249 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 250 | { |
7daac490 | 251 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 LT |
252 | return; |
253 | ||
254 | if (!blk_remove_plug(q)) | |
255 | return; | |
256 | ||
22e2c507 | 257 | q->request_fn(q); |
1da177e4 LT |
258 | } |
259 | EXPORT_SYMBOL(__generic_unplug_device); | |
260 | ||
261 | /** | |
262 | * generic_unplug_device - fire a request queue | |
165125e1 | 263 | * @q: The &struct request_queue in question |
1da177e4 LT |
264 | * |
265 | * Description: | |
266 | * Linux uses plugging to build bigger requests queues before letting | |
267 | * the device have at them. If a queue is plugged, the I/O scheduler | |
268 | * is still adding and merging requests on the queue. Once the queue | |
269 | * gets unplugged, the request_fn defined for the queue is invoked and | |
270 | * transfers started. | |
271 | **/ | |
165125e1 | 272 | void generic_unplug_device(struct request_queue *q) |
1da177e4 LT |
273 | { |
274 | spin_lock_irq(q->queue_lock); | |
275 | __generic_unplug_device(q); | |
276 | spin_unlock_irq(q->queue_lock); | |
277 | } | |
278 | EXPORT_SYMBOL(generic_unplug_device); | |
279 | ||
280 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
281 | struct page *page) | |
282 | { | |
165125e1 | 283 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 | 284 | |
2ad8b1ef | 285 | blk_unplug(q); |
1da177e4 LT |
286 | } |
287 | ||
86db1e29 | 288 | void blk_unplug_work(struct work_struct *work) |
1da177e4 | 289 | { |
165125e1 JA |
290 | struct request_queue *q = |
291 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 292 | |
2056a782 JA |
293 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, |
294 | q->rq.count[READ] + q->rq.count[WRITE]); | |
295 | ||
1da177e4 LT |
296 | q->unplug_fn(q); |
297 | } | |
298 | ||
86db1e29 | 299 | void blk_unplug_timeout(unsigned long data) |
1da177e4 | 300 | { |
165125e1 | 301 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 302 | |
2056a782 JA |
303 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, |
304 | q->rq.count[READ] + q->rq.count[WRITE]); | |
305 | ||
1da177e4 LT |
306 | kblockd_schedule_work(&q->unplug_work); |
307 | } | |
308 | ||
2ad8b1ef AB |
309 | void blk_unplug(struct request_queue *q) |
310 | { | |
311 | /* | |
312 | * devices don't necessarily have an ->unplug_fn defined | |
313 | */ | |
314 | if (q->unplug_fn) { | |
315 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, | |
316 | q->rq.count[READ] + q->rq.count[WRITE]); | |
317 | ||
318 | q->unplug_fn(q); | |
319 | } | |
320 | } | |
321 | EXPORT_SYMBOL(blk_unplug); | |
322 | ||
1da177e4 LT |
323 | /** |
324 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 325 | * @q: The &struct request_queue in question |
1da177e4 LT |
326 | * |
327 | * Description: | |
328 | * blk_start_queue() will clear the stop flag on the queue, and call | |
329 | * the request_fn for the queue if it was in a stopped state when | |
330 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
331 | **/ | |
165125e1 | 332 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 333 | { |
a038e253 PBG |
334 | WARN_ON(!irqs_disabled()); |
335 | ||
1da177e4 LT |
336 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); |
337 | ||
338 | /* | |
339 | * one level of recursion is ok and is much faster than kicking | |
340 | * the unplug handling | |
341 | */ | |
342 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
343 | q->request_fn(q); | |
344 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
345 | } else { | |
346 | blk_plug_device(q); | |
347 | kblockd_schedule_work(&q->unplug_work); | |
348 | } | |
349 | } | |
1da177e4 LT |
350 | EXPORT_SYMBOL(blk_start_queue); |
351 | ||
352 | /** | |
353 | * blk_stop_queue - stop a queue | |
165125e1 | 354 | * @q: The &struct request_queue in question |
1da177e4 LT |
355 | * |
356 | * Description: | |
357 | * The Linux block layer assumes that a block driver will consume all | |
358 | * entries on the request queue when the request_fn strategy is called. | |
359 | * Often this will not happen, because of hardware limitations (queue | |
360 | * depth settings). If a device driver gets a 'queue full' response, | |
361 | * or if it simply chooses not to queue more I/O at one point, it can | |
362 | * call this function to prevent the request_fn from being called until | |
363 | * the driver has signalled it's ready to go again. This happens by calling | |
364 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
365 | **/ | |
165125e1 | 366 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
367 | { |
368 | blk_remove_plug(q); | |
369 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | |
370 | } | |
371 | EXPORT_SYMBOL(blk_stop_queue); | |
372 | ||
373 | /** | |
374 | * blk_sync_queue - cancel any pending callbacks on a queue | |
375 | * @q: the queue | |
376 | * | |
377 | * Description: | |
378 | * The block layer may perform asynchronous callback activity | |
379 | * on a queue, such as calling the unplug function after a timeout. | |
380 | * A block device may call blk_sync_queue to ensure that any | |
381 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 382 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
383 | * that its ->make_request_fn will not re-add plugging prior to calling |
384 | * this function. | |
385 | * | |
386 | */ | |
387 | void blk_sync_queue(struct request_queue *q) | |
388 | { | |
389 | del_timer_sync(&q->unplug_timer); | |
abbeb88d | 390 | kblockd_flush_work(&q->unplug_work); |
1da177e4 LT |
391 | } |
392 | EXPORT_SYMBOL(blk_sync_queue); | |
393 | ||
394 | /** | |
395 | * blk_run_queue - run a single device queue | |
396 | * @q: The queue to run | |
397 | */ | |
398 | void blk_run_queue(struct request_queue *q) | |
399 | { | |
400 | unsigned long flags; | |
401 | ||
402 | spin_lock_irqsave(q->queue_lock, flags); | |
403 | blk_remove_plug(q); | |
dac07ec1 JA |
404 | |
405 | /* | |
406 | * Only recurse once to avoid overrunning the stack, let the unplug | |
407 | * handling reinvoke the handler shortly if we already got there. | |
408 | */ | |
409 | if (!elv_queue_empty(q)) { | |
410 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
411 | q->request_fn(q); | |
412 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
413 | } else { | |
414 | blk_plug_device(q); | |
415 | kblockd_schedule_work(&q->unplug_work); | |
416 | } | |
417 | } | |
418 | ||
1da177e4 LT |
419 | spin_unlock_irqrestore(q->queue_lock, flags); |
420 | } | |
421 | EXPORT_SYMBOL(blk_run_queue); | |
422 | ||
165125e1 | 423 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
424 | { |
425 | kobject_put(&q->kobj); | |
426 | } | |
427 | EXPORT_SYMBOL(blk_put_queue); | |
428 | ||
6728cb0e | 429 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc AV |
430 | { |
431 | mutex_lock(&q->sysfs_lock); | |
432 | set_bit(QUEUE_FLAG_DEAD, &q->queue_flags); | |
433 | mutex_unlock(&q->sysfs_lock); | |
434 | ||
435 | if (q->elevator) | |
436 | elevator_exit(q->elevator); | |
437 | ||
438 | blk_put_queue(q); | |
439 | } | |
1da177e4 LT |
440 | EXPORT_SYMBOL(blk_cleanup_queue); |
441 | ||
165125e1 | 442 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
443 | { |
444 | struct request_list *rl = &q->rq; | |
445 | ||
446 | rl->count[READ] = rl->count[WRITE] = 0; | |
447 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
cb98fc8b | 448 | rl->elvpriv = 0; |
1da177e4 LT |
449 | init_waitqueue_head(&rl->wait[READ]); |
450 | init_waitqueue_head(&rl->wait[WRITE]); | |
1da177e4 | 451 | |
1946089a CL |
452 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
453 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
454 | |
455 | if (!rl->rq_pool) | |
456 | return -ENOMEM; | |
457 | ||
458 | return 0; | |
459 | } | |
460 | ||
165125e1 | 461 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 462 | { |
1946089a CL |
463 | return blk_alloc_queue_node(gfp_mask, -1); |
464 | } | |
465 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 466 | |
165125e1 | 467 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 468 | { |
165125e1 | 469 | struct request_queue *q; |
e0bf68dd | 470 | int err; |
1946089a | 471 | |
8324aa91 | 472 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 473 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
474 | if (!q) |
475 | return NULL; | |
476 | ||
e0bf68dd PZ |
477 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
478 | q->backing_dev_info.unplug_io_data = q; | |
479 | err = bdi_init(&q->backing_dev_info); | |
480 | if (err) { | |
8324aa91 | 481 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
482 | return NULL; |
483 | } | |
484 | ||
1da177e4 | 485 | init_timer(&q->unplug_timer); |
483f4afc | 486 | |
8324aa91 | 487 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 488 | |
483f4afc AV |
489 | mutex_init(&q->sysfs_lock); |
490 | ||
1da177e4 LT |
491 | return q; |
492 | } | |
1946089a | 493 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
494 | |
495 | /** | |
496 | * blk_init_queue - prepare a request queue for use with a block device | |
497 | * @rfn: The function to be called to process requests that have been | |
498 | * placed on the queue. | |
499 | * @lock: Request queue spin lock | |
500 | * | |
501 | * Description: | |
502 | * If a block device wishes to use the standard request handling procedures, | |
503 | * which sorts requests and coalesces adjacent requests, then it must | |
504 | * call blk_init_queue(). The function @rfn will be called when there | |
505 | * are requests on the queue that need to be processed. If the device | |
506 | * supports plugging, then @rfn may not be called immediately when requests | |
507 | * are available on the queue, but may be called at some time later instead. | |
508 | * Plugged queues are generally unplugged when a buffer belonging to one | |
509 | * of the requests on the queue is needed, or due to memory pressure. | |
510 | * | |
511 | * @rfn is not required, or even expected, to remove all requests off the | |
512 | * queue, but only as many as it can handle at a time. If it does leave | |
513 | * requests on the queue, it is responsible for arranging that the requests | |
514 | * get dealt with eventually. | |
515 | * | |
516 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
517 | * request queue; this lock will be taken also from interrupt context, so irq |
518 | * disabling is needed for it. | |
1da177e4 LT |
519 | * |
520 | * Function returns a pointer to the initialized request queue, or NULL if | |
521 | * it didn't succeed. | |
522 | * | |
523 | * Note: | |
524 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
525 | * when the block device is deactivated (such as at module unload). | |
526 | **/ | |
1946089a | 527 | |
165125e1 | 528 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 529 | { |
1946089a CL |
530 | return blk_init_queue_node(rfn, lock, -1); |
531 | } | |
532 | EXPORT_SYMBOL(blk_init_queue); | |
533 | ||
165125e1 | 534 | struct request_queue * |
1946089a CL |
535 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
536 | { | |
165125e1 | 537 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
538 | |
539 | if (!q) | |
540 | return NULL; | |
541 | ||
1946089a | 542 | q->node = node_id; |
8669aafd | 543 | if (blk_init_free_list(q)) { |
8324aa91 | 544 | kmem_cache_free(blk_requestq_cachep, q); |
8669aafd AV |
545 | return NULL; |
546 | } | |
1da177e4 | 547 | |
152587de JA |
548 | /* |
549 | * if caller didn't supply a lock, they get per-queue locking with | |
550 | * our embedded lock | |
551 | */ | |
552 | if (!lock) { | |
553 | spin_lock_init(&q->__queue_lock); | |
554 | lock = &q->__queue_lock; | |
555 | } | |
556 | ||
1da177e4 | 557 | q->request_fn = rfn; |
1da177e4 LT |
558 | q->prep_rq_fn = NULL; |
559 | q->unplug_fn = generic_unplug_device; | |
560 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); | |
561 | q->queue_lock = lock; | |
562 | ||
563 | blk_queue_segment_boundary(q, 0xffffffff); | |
564 | ||
565 | blk_queue_make_request(q, __make_request); | |
566 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
567 | ||
568 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
569 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
570 | ||
44ec9542 AS |
571 | q->sg_reserved_size = INT_MAX; |
572 | ||
1da177e4 LT |
573 | /* |
574 | * all done | |
575 | */ | |
576 | if (!elevator_init(q, NULL)) { | |
577 | blk_queue_congestion_threshold(q); | |
578 | return q; | |
579 | } | |
580 | ||
8669aafd | 581 | blk_put_queue(q); |
1da177e4 LT |
582 | return NULL; |
583 | } | |
1946089a | 584 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 585 | |
165125e1 | 586 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 587 | { |
fde6ad22 | 588 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 589 | kobject_get(&q->kobj); |
1da177e4 LT |
590 | return 0; |
591 | } | |
592 | ||
593 | return 1; | |
594 | } | |
1da177e4 LT |
595 | EXPORT_SYMBOL(blk_get_queue); |
596 | ||
165125e1 | 597 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 598 | { |
4aff5e23 | 599 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 600 | elv_put_request(q, rq); |
1da177e4 LT |
601 | mempool_free(rq, q->rq.rq_pool); |
602 | } | |
603 | ||
1ea25ecb | 604 | static struct request * |
165125e1 | 605 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) |
1da177e4 LT |
606 | { |
607 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
608 | ||
609 | if (!rq) | |
610 | return NULL; | |
611 | ||
612 | /* | |
4aff5e23 | 613 | * first three bits are identical in rq->cmd_flags and bio->bi_rw, |
1da177e4 LT |
614 | * see bio.h and blkdev.h |
615 | */ | |
49171e5c | 616 | rq->cmd_flags = rw | REQ_ALLOCED; |
1da177e4 | 617 | |
cb98fc8b | 618 | if (priv) { |
cb78b285 | 619 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
620 | mempool_free(rq, q->rq.rq_pool); |
621 | return NULL; | |
622 | } | |
4aff5e23 | 623 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 624 | } |
1da177e4 | 625 | |
cb98fc8b | 626 | return rq; |
1da177e4 LT |
627 | } |
628 | ||
629 | /* | |
630 | * ioc_batching returns true if the ioc is a valid batching request and | |
631 | * should be given priority access to a request. | |
632 | */ | |
165125e1 | 633 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
634 | { |
635 | if (!ioc) | |
636 | return 0; | |
637 | ||
638 | /* | |
639 | * Make sure the process is able to allocate at least 1 request | |
640 | * even if the batch times out, otherwise we could theoretically | |
641 | * lose wakeups. | |
642 | */ | |
643 | return ioc->nr_batch_requests == q->nr_batching || | |
644 | (ioc->nr_batch_requests > 0 | |
645 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
646 | } | |
647 | ||
648 | /* | |
649 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
650 | * will cause the process to be a "batcher" on all queues in the system. This | |
651 | * is the behaviour we want though - once it gets a wakeup it should be given | |
652 | * a nice run. | |
653 | */ | |
165125e1 | 654 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
655 | { |
656 | if (!ioc || ioc_batching(q, ioc)) | |
657 | return; | |
658 | ||
659 | ioc->nr_batch_requests = q->nr_batching; | |
660 | ioc->last_waited = jiffies; | |
661 | } | |
662 | ||
165125e1 | 663 | static void __freed_request(struct request_queue *q, int rw) |
1da177e4 LT |
664 | { |
665 | struct request_list *rl = &q->rq; | |
666 | ||
667 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
79e2de4b | 668 | blk_clear_queue_congested(q, rw); |
1da177e4 LT |
669 | |
670 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
671 | if (waitqueue_active(&rl->wait[rw])) |
672 | wake_up(&rl->wait[rw]); | |
673 | ||
674 | blk_clear_queue_full(q, rw); | |
675 | } | |
676 | } | |
677 | ||
678 | /* | |
679 | * A request has just been released. Account for it, update the full and | |
680 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
681 | */ | |
165125e1 | 682 | static void freed_request(struct request_queue *q, int rw, int priv) |
1da177e4 LT |
683 | { |
684 | struct request_list *rl = &q->rq; | |
685 | ||
686 | rl->count[rw]--; | |
cb98fc8b TH |
687 | if (priv) |
688 | rl->elvpriv--; | |
1da177e4 LT |
689 | |
690 | __freed_request(q, rw); | |
691 | ||
692 | if (unlikely(rl->starved[rw ^ 1])) | |
693 | __freed_request(q, rw ^ 1); | |
1da177e4 LT |
694 | } |
695 | ||
696 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
697 | /* | |
d6344532 NP |
698 | * Get a free request, queue_lock must be held. |
699 | * Returns NULL on failure, with queue_lock held. | |
700 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 701 | */ |
165125e1 | 702 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 703 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
704 | { |
705 | struct request *rq = NULL; | |
706 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 707 | struct io_context *ioc = NULL; |
7749a8d4 | 708 | const int rw = rw_flags & 0x01; |
88ee5ef1 JA |
709 | int may_queue, priv; |
710 | ||
7749a8d4 | 711 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
712 | if (may_queue == ELV_MQUEUE_NO) |
713 | goto rq_starved; | |
714 | ||
715 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
716 | if (rl->count[rw]+1 >= q->nr_requests) { | |
b5deef90 | 717 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
718 | /* |
719 | * The queue will fill after this allocation, so set | |
720 | * it as full, and mark this process as "batching". | |
721 | * This process will be allowed to complete a batch of | |
722 | * requests, others will be blocked. | |
723 | */ | |
724 | if (!blk_queue_full(q, rw)) { | |
725 | ioc_set_batching(q, ioc); | |
726 | blk_set_queue_full(q, rw); | |
727 | } else { | |
728 | if (may_queue != ELV_MQUEUE_MUST | |
729 | && !ioc_batching(q, ioc)) { | |
730 | /* | |
731 | * The queue is full and the allocating | |
732 | * process is not a "batcher", and not | |
733 | * exempted by the IO scheduler | |
734 | */ | |
735 | goto out; | |
736 | } | |
737 | } | |
1da177e4 | 738 | } |
79e2de4b | 739 | blk_set_queue_congested(q, rw); |
1da177e4 LT |
740 | } |
741 | ||
082cf69e JA |
742 | /* |
743 | * Only allow batching queuers to allocate up to 50% over the defined | |
744 | * limit of requests, otherwise we could have thousands of requests | |
745 | * allocated with any setting of ->nr_requests | |
746 | */ | |
fd782a4a | 747 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 748 | goto out; |
fd782a4a | 749 | |
1da177e4 LT |
750 | rl->count[rw]++; |
751 | rl->starved[rw] = 0; | |
cb98fc8b | 752 | |
64521d1a | 753 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
754 | if (priv) |
755 | rl->elvpriv++; | |
756 | ||
1da177e4 LT |
757 | spin_unlock_irq(q->queue_lock); |
758 | ||
7749a8d4 | 759 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 760 | if (unlikely(!rq)) { |
1da177e4 LT |
761 | /* |
762 | * Allocation failed presumably due to memory. Undo anything | |
763 | * we might have messed up. | |
764 | * | |
765 | * Allocating task should really be put onto the front of the | |
766 | * wait queue, but this is pretty rare. | |
767 | */ | |
768 | spin_lock_irq(q->queue_lock); | |
cb98fc8b | 769 | freed_request(q, rw, priv); |
1da177e4 LT |
770 | |
771 | /* | |
772 | * in the very unlikely event that allocation failed and no | |
773 | * requests for this direction was pending, mark us starved | |
774 | * so that freeing of a request in the other direction will | |
775 | * notice us. another possible fix would be to split the | |
776 | * rq mempool into READ and WRITE | |
777 | */ | |
778 | rq_starved: | |
779 | if (unlikely(rl->count[rw] == 0)) | |
780 | rl->starved[rw] = 1; | |
781 | ||
1da177e4 LT |
782 | goto out; |
783 | } | |
784 | ||
88ee5ef1 JA |
785 | /* |
786 | * ioc may be NULL here, and ioc_batching will be false. That's | |
787 | * OK, if the queue is under the request limit then requests need | |
788 | * not count toward the nr_batch_requests limit. There will always | |
789 | * be some limit enforced by BLK_BATCH_TIME. | |
790 | */ | |
1da177e4 LT |
791 | if (ioc_batching(q, ioc)) |
792 | ioc->nr_batch_requests--; | |
6728cb0e | 793 | |
1da177e4 | 794 | rq_init(q, rq); |
2056a782 JA |
795 | |
796 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); | |
1da177e4 | 797 | out: |
1da177e4 LT |
798 | return rq; |
799 | } | |
800 | ||
801 | /* | |
802 | * No available requests for this queue, unplug the device and wait for some | |
803 | * requests to become available. | |
d6344532 NP |
804 | * |
805 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 806 | */ |
165125e1 | 807 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 808 | struct bio *bio) |
1da177e4 | 809 | { |
7749a8d4 | 810 | const int rw = rw_flags & 0x01; |
1da177e4 LT |
811 | struct request *rq; |
812 | ||
7749a8d4 | 813 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
814 | while (!rq) { |
815 | DEFINE_WAIT(wait); | |
1da177e4 LT |
816 | struct request_list *rl = &q->rq; |
817 | ||
818 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
819 | TASK_UNINTERRUPTIBLE); | |
820 | ||
7749a8d4 | 821 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
1da177e4 LT |
822 | |
823 | if (!rq) { | |
824 | struct io_context *ioc; | |
825 | ||
2056a782 JA |
826 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); |
827 | ||
d6344532 NP |
828 | __generic_unplug_device(q); |
829 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
830 | io_schedule(); |
831 | ||
832 | /* | |
833 | * After sleeping, we become a "batching" process and | |
834 | * will be able to allocate at least one request, and | |
835 | * up to a big batch of them for a small period time. | |
836 | * See ioc_batching, ioc_set_batching | |
837 | */ | |
b5deef90 | 838 | ioc = current_io_context(GFP_NOIO, q->node); |
1da177e4 | 839 | ioc_set_batching(q, ioc); |
d6344532 NP |
840 | |
841 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
842 | } |
843 | finish_wait(&rl->wait[rw], &wait); | |
450991bc | 844 | } |
1da177e4 LT |
845 | |
846 | return rq; | |
847 | } | |
848 | ||
165125e1 | 849 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
850 | { |
851 | struct request *rq; | |
852 | ||
853 | BUG_ON(rw != READ && rw != WRITE); | |
854 | ||
d6344532 NP |
855 | spin_lock_irq(q->queue_lock); |
856 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 857 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 858 | } else { |
22e2c507 | 859 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
860 | if (!rq) |
861 | spin_unlock_irq(q->queue_lock); | |
862 | } | |
863 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
864 | |
865 | return rq; | |
866 | } | |
1da177e4 LT |
867 | EXPORT_SYMBOL(blk_get_request); |
868 | ||
dc72ef4a JA |
869 | /** |
870 | * blk_start_queueing - initiate dispatch of requests to device | |
871 | * @q: request queue to kick into gear | |
872 | * | |
873 | * This is basically a helper to remove the need to know whether a queue | |
874 | * is plugged or not if someone just wants to initiate dispatch of requests | |
875 | * for this queue. | |
876 | * | |
877 | * The queue lock must be held with interrupts disabled. | |
878 | */ | |
165125e1 | 879 | void blk_start_queueing(struct request_queue *q) |
dc72ef4a JA |
880 | { |
881 | if (!blk_queue_plugged(q)) | |
882 | q->request_fn(q); | |
883 | else | |
884 | __generic_unplug_device(q); | |
885 | } | |
886 | EXPORT_SYMBOL(blk_start_queueing); | |
887 | ||
1da177e4 LT |
888 | /** |
889 | * blk_requeue_request - put a request back on queue | |
890 | * @q: request queue where request should be inserted | |
891 | * @rq: request to be inserted | |
892 | * | |
893 | * Description: | |
894 | * Drivers often keep queueing requests until the hardware cannot accept | |
895 | * more, when that condition happens we need to put the request back | |
896 | * on the queue. Must be called with queue lock held. | |
897 | */ | |
165125e1 | 898 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 899 | { |
2056a782 JA |
900 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); |
901 | ||
1da177e4 LT |
902 | if (blk_rq_tagged(rq)) |
903 | blk_queue_end_tag(q, rq); | |
904 | ||
905 | elv_requeue_request(q, rq); | |
906 | } | |
1da177e4 LT |
907 | EXPORT_SYMBOL(blk_requeue_request); |
908 | ||
909 | /** | |
910 | * blk_insert_request - insert a special request in to a request queue | |
911 | * @q: request queue where request should be inserted | |
912 | * @rq: request to be inserted | |
913 | * @at_head: insert request at head or tail of queue | |
914 | * @data: private data | |
1da177e4 LT |
915 | * |
916 | * Description: | |
917 | * Many block devices need to execute commands asynchronously, so they don't | |
918 | * block the whole kernel from preemption during request execution. This is | |
919 | * accomplished normally by inserting aritficial requests tagged as | |
920 | * REQ_SPECIAL in to the corresponding request queue, and letting them be | |
921 | * scheduled for actual execution by the request queue. | |
922 | * | |
923 | * We have the option of inserting the head or the tail of the queue. | |
924 | * Typically we use the tail for new ioctls and so forth. We use the head | |
925 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
926 | * host that is unable to accept a particular command. | |
927 | */ | |
165125e1 | 928 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 929 | int at_head, void *data) |
1da177e4 | 930 | { |
867d1191 | 931 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
932 | unsigned long flags; |
933 | ||
934 | /* | |
935 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
936 | * must not attempt merges on this) and that it acts as a soft | |
937 | * barrier | |
938 | */ | |
4aff5e23 JA |
939 | rq->cmd_type = REQ_TYPE_SPECIAL; |
940 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
941 | |
942 | rq->special = data; | |
943 | ||
944 | spin_lock_irqsave(q->queue_lock, flags); | |
945 | ||
946 | /* | |
947 | * If command is tagged, release the tag | |
948 | */ | |
867d1191 TH |
949 | if (blk_rq_tagged(rq)) |
950 | blk_queue_end_tag(q, rq); | |
1da177e4 | 951 | |
b238b3d4 | 952 | drive_stat_acct(rq, 1); |
867d1191 | 953 | __elv_add_request(q, rq, where, 0); |
dc72ef4a | 954 | blk_start_queueing(q); |
1da177e4 LT |
955 | spin_unlock_irqrestore(q->queue_lock, flags); |
956 | } | |
1da177e4 LT |
957 | EXPORT_SYMBOL(blk_insert_request); |
958 | ||
1da177e4 LT |
959 | /* |
960 | * add-request adds a request to the linked list. | |
961 | * queue lock is held and interrupts disabled, as we muck with the | |
962 | * request queue list. | |
963 | */ | |
6728cb0e | 964 | static inline void add_request(struct request_queue *q, struct request *req) |
1da177e4 | 965 | { |
b238b3d4 | 966 | drive_stat_acct(req, 1); |
1da177e4 | 967 | |
1da177e4 LT |
968 | /* |
969 | * elevator indicated where it wants this request to be | |
970 | * inserted at elevator_merge time | |
971 | */ | |
972 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
973 | } | |
6728cb0e | 974 | |
1da177e4 LT |
975 | /* |
976 | * disk_round_stats() - Round off the performance stats on a struct | |
977 | * disk_stats. | |
978 | * | |
979 | * The average IO queue length and utilisation statistics are maintained | |
980 | * by observing the current state of the queue length and the amount of | |
981 | * time it has been in this state for. | |
982 | * | |
983 | * Normally, that accounting is done on IO completion, but that can result | |
984 | * in more than a second's worth of IO being accounted for within any one | |
985 | * second, leading to >100% utilisation. To deal with that, we call this | |
986 | * function to do a round-off before returning the results when reading | |
987 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
988 | * the current jiffies and restarts the counters again. | |
989 | */ | |
990 | void disk_round_stats(struct gendisk *disk) | |
991 | { | |
992 | unsigned long now = jiffies; | |
993 | ||
b2982649 KC |
994 | if (now == disk->stamp) |
995 | return; | |
1da177e4 | 996 | |
20e5c81f KC |
997 | if (disk->in_flight) { |
998 | __disk_stat_add(disk, time_in_queue, | |
999 | disk->in_flight * (now - disk->stamp)); | |
1000 | __disk_stat_add(disk, io_ticks, (now - disk->stamp)); | |
1001 | } | |
1da177e4 | 1002 | disk->stamp = now; |
1da177e4 | 1003 | } |
3eaf840e JNN |
1004 | EXPORT_SYMBOL_GPL(disk_round_stats); |
1005 | ||
6f2576af JM |
1006 | void part_round_stats(struct hd_struct *part) |
1007 | { | |
1008 | unsigned long now = jiffies; | |
1009 | ||
1010 | if (now == part->stamp) | |
1011 | return; | |
1012 | ||
1013 | if (part->in_flight) { | |
1014 | __part_stat_add(part, time_in_queue, | |
1015 | part->in_flight * (now - part->stamp)); | |
1016 | __part_stat_add(part, io_ticks, (now - part->stamp)); | |
1017 | } | |
1018 | part->stamp = now; | |
1019 | } | |
1020 | ||
1da177e4 LT |
1021 | /* |
1022 | * queue lock must be held | |
1023 | */ | |
165125e1 | 1024 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1025 | { |
1da177e4 LT |
1026 | if (unlikely(!q)) |
1027 | return; | |
1028 | if (unlikely(--req->ref_count)) | |
1029 | return; | |
1030 | ||
8922e16c TH |
1031 | elv_completed_request(q, req); |
1032 | ||
1da177e4 LT |
1033 | /* |
1034 | * Request may not have originated from ll_rw_blk. if not, | |
1035 | * it didn't come out of our reserved rq pools | |
1036 | */ | |
49171e5c | 1037 | if (req->cmd_flags & REQ_ALLOCED) { |
1da177e4 | 1038 | int rw = rq_data_dir(req); |
4aff5e23 | 1039 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1040 | |
1da177e4 | 1041 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1042 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1043 | |
1044 | blk_free_request(q, req); | |
cb98fc8b | 1045 | freed_request(q, rw, priv); |
1da177e4 LT |
1046 | } |
1047 | } | |
6e39b69e MC |
1048 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1049 | ||
1da177e4 LT |
1050 | void blk_put_request(struct request *req) |
1051 | { | |
8922e16c | 1052 | unsigned long flags; |
165125e1 | 1053 | struct request_queue *q = req->q; |
8922e16c | 1054 | |
1da177e4 | 1055 | /* |
8922e16c TH |
1056 | * Gee, IDE calls in w/ NULL q. Fix IDE and remove the |
1057 | * following if (q) test. | |
1da177e4 | 1058 | */ |
8922e16c | 1059 | if (q) { |
1da177e4 LT |
1060 | spin_lock_irqsave(q->queue_lock, flags); |
1061 | __blk_put_request(q, req); | |
1062 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1063 | } | |
1064 | } | |
1da177e4 LT |
1065 | EXPORT_SYMBOL(blk_put_request); |
1066 | ||
86db1e29 | 1067 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1068 | { |
4aff5e23 | 1069 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
1070 | |
1071 | /* | |
1072 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1073 | */ | |
1074 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
4aff5e23 | 1075 | req->cmd_flags |= REQ_FAILFAST; |
52d9e675 TH |
1076 | |
1077 | /* | |
1078 | * REQ_BARRIER implies no merging, but lets make it explicit | |
1079 | */ | |
1080 | if (unlikely(bio_barrier(bio))) | |
4aff5e23 | 1081 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); |
52d9e675 | 1082 | |
b31dc66a | 1083 | if (bio_sync(bio)) |
4aff5e23 | 1084 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
1085 | if (bio_rw_meta(bio)) |
1086 | req->cmd_flags |= REQ_RW_META; | |
b31dc66a | 1087 | |
52d9e675 TH |
1088 | req->errors = 0; |
1089 | req->hard_sector = req->sector = bio->bi_sector; | |
52d9e675 | 1090 | req->ioprio = bio_prio(bio); |
52d9e675 | 1091 | req->start_time = jiffies; |
bc1c56fd | 1092 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1093 | } |
1094 | ||
165125e1 | 1095 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1096 | { |
450991bc | 1097 | struct request *req; |
51da90fc JA |
1098 | int el_ret, nr_sectors, barrier, err; |
1099 | const unsigned short prio = bio_prio(bio); | |
1100 | const int sync = bio_sync(bio); | |
7749a8d4 | 1101 | int rw_flags; |
1da177e4 | 1102 | |
1da177e4 | 1103 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
1104 | |
1105 | /* | |
1106 | * low level driver can indicate that it wants pages above a | |
1107 | * certain limit bounced to low memory (ie for highmem, or even | |
1108 | * ISA dma in theory) | |
1109 | */ | |
1110 | blk_queue_bounce(q, &bio); | |
1111 | ||
1da177e4 | 1112 | barrier = bio_barrier(bio); |
797e7dbb | 1113 | if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) { |
1da177e4 LT |
1114 | err = -EOPNOTSUPP; |
1115 | goto end_io; | |
1116 | } | |
1117 | ||
1da177e4 LT |
1118 | spin_lock_irq(q->queue_lock); |
1119 | ||
450991bc | 1120 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
1121 | goto get_rq; |
1122 | ||
1123 | el_ret = elv_merge(q, &req, bio); | |
1124 | switch (el_ret) { | |
6728cb0e JA |
1125 | case ELEVATOR_BACK_MERGE: |
1126 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1127 | |
6728cb0e JA |
1128 | if (!ll_back_merge_fn(q, req, bio)) |
1129 | break; | |
1da177e4 | 1130 | |
6728cb0e | 1131 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); |
2056a782 | 1132 | |
6728cb0e JA |
1133 | req->biotail->bi_next = bio; |
1134 | req->biotail = bio; | |
1135 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1136 | req->ioprio = ioprio_best(req->ioprio, prio); | |
1137 | drive_stat_acct(req, 0); | |
1138 | if (!attempt_back_merge(q, req)) | |
1139 | elv_merged_request(q, req, el_ret); | |
1140 | goto out; | |
1da177e4 | 1141 | |
6728cb0e JA |
1142 | case ELEVATOR_FRONT_MERGE: |
1143 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1144 | |
6728cb0e JA |
1145 | if (!ll_front_merge_fn(q, req, bio)) |
1146 | break; | |
1da177e4 | 1147 | |
6728cb0e | 1148 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); |
2056a782 | 1149 | |
6728cb0e JA |
1150 | bio->bi_next = req->bio; |
1151 | req->bio = bio; | |
1da177e4 | 1152 | |
6728cb0e JA |
1153 | /* |
1154 | * may not be valid. if the low level driver said | |
1155 | * it didn't need a bounce buffer then it better | |
1156 | * not touch req->buffer either... | |
1157 | */ | |
1158 | req->buffer = bio_data(bio); | |
1159 | req->current_nr_sectors = bio_cur_sectors(bio); | |
1160 | req->hard_cur_sectors = req->current_nr_sectors; | |
1161 | req->sector = req->hard_sector = bio->bi_sector; | |
1162 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1163 | req->ioprio = ioprio_best(req->ioprio, prio); | |
1164 | drive_stat_acct(req, 0); | |
1165 | if (!attempt_front_merge(q, req)) | |
1166 | elv_merged_request(q, req, el_ret); | |
1167 | goto out; | |
1168 | ||
1169 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1170 | default: | |
1171 | ; | |
1da177e4 LT |
1172 | } |
1173 | ||
450991bc | 1174 | get_rq: |
7749a8d4 JA |
1175 | /* |
1176 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1177 | * but we need to set it earlier to expose the sync flag to the | |
1178 | * rq allocator and io schedulers. | |
1179 | */ | |
1180 | rw_flags = bio_data_dir(bio); | |
1181 | if (sync) | |
1182 | rw_flags |= REQ_RW_SYNC; | |
1183 | ||
1da177e4 | 1184 | /* |
450991bc | 1185 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1186 | * Returns with the queue unlocked. |
450991bc | 1187 | */ |
7749a8d4 | 1188 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1189 | |
450991bc NP |
1190 | /* |
1191 | * After dropping the lock and possibly sleeping here, our request | |
1192 | * may now be mergeable after it had proven unmergeable (above). | |
1193 | * We don't worry about that case for efficiency. It won't happen | |
1194 | * often, and the elevators are able to handle it. | |
1da177e4 | 1195 | */ |
52d9e675 | 1196 | init_request_from_bio(req, bio); |
1da177e4 | 1197 | |
450991bc NP |
1198 | spin_lock_irq(q->queue_lock); |
1199 | if (elv_queue_empty(q)) | |
1200 | blk_plug_device(q); | |
1da177e4 LT |
1201 | add_request(q, req); |
1202 | out: | |
4a534f93 | 1203 | if (sync) |
1da177e4 LT |
1204 | __generic_unplug_device(q); |
1205 | ||
1206 | spin_unlock_irq(q->queue_lock); | |
1207 | return 0; | |
1208 | ||
1209 | end_io: | |
6712ecf8 | 1210 | bio_endio(bio, err); |
1da177e4 LT |
1211 | return 0; |
1212 | } | |
1213 | ||
1214 | /* | |
1215 | * If bio->bi_dev is a partition, remap the location | |
1216 | */ | |
1217 | static inline void blk_partition_remap(struct bio *bio) | |
1218 | { | |
1219 | struct block_device *bdev = bio->bi_bdev; | |
1220 | ||
bf2de6f5 | 1221 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1222 | struct hd_struct *p = bdev->bd_part; |
1223 | ||
1da177e4 LT |
1224 | bio->bi_sector += p->start_sect; |
1225 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b AB |
1226 | |
1227 | blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio, | |
1228 | bdev->bd_dev, bio->bi_sector, | |
1229 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1230 | } |
1231 | } | |
1232 | ||
1da177e4 LT |
1233 | static void handle_bad_sector(struct bio *bio) |
1234 | { | |
1235 | char b[BDEVNAME_SIZE]; | |
1236 | ||
1237 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1238 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1239 | bdevname(bio->bi_bdev, b), | |
1240 | bio->bi_rw, | |
1241 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1242 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1243 | ||
1244 | set_bit(BIO_EOF, &bio->bi_flags); | |
1245 | } | |
1246 | ||
c17bb495 AM |
1247 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1248 | ||
1249 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1250 | ||
1251 | static int __init setup_fail_make_request(char *str) | |
1252 | { | |
1253 | return setup_fault_attr(&fail_make_request, str); | |
1254 | } | |
1255 | __setup("fail_make_request=", setup_fail_make_request); | |
1256 | ||
1257 | static int should_fail_request(struct bio *bio) | |
1258 | { | |
1259 | if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) || | |
1260 | (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail)) | |
1261 | return should_fail(&fail_make_request, bio->bi_size); | |
1262 | ||
1263 | return 0; | |
1264 | } | |
1265 | ||
1266 | static int __init fail_make_request_debugfs(void) | |
1267 | { | |
1268 | return init_fault_attr_dentries(&fail_make_request, | |
1269 | "fail_make_request"); | |
1270 | } | |
1271 | ||
1272 | late_initcall(fail_make_request_debugfs); | |
1273 | ||
1274 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1275 | ||
1276 | static inline int should_fail_request(struct bio *bio) | |
1277 | { | |
1278 | return 0; | |
1279 | } | |
1280 | ||
1281 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1282 | ||
c07e2b41 JA |
1283 | /* |
1284 | * Check whether this bio extends beyond the end of the device. | |
1285 | */ | |
1286 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1287 | { | |
1288 | sector_t maxsector; | |
1289 | ||
1290 | if (!nr_sectors) | |
1291 | return 0; | |
1292 | ||
1293 | /* Test device or partition size, when known. */ | |
1294 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1295 | if (maxsector) { | |
1296 | sector_t sector = bio->bi_sector; | |
1297 | ||
1298 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1299 | /* | |
1300 | * This may well happen - the kernel calls bread() | |
1301 | * without checking the size of the device, e.g., when | |
1302 | * mounting a device. | |
1303 | */ | |
1304 | handle_bad_sector(bio); | |
1305 | return 1; | |
1306 | } | |
1307 | } | |
1308 | ||
1309 | return 0; | |
1310 | } | |
1311 | ||
1da177e4 LT |
1312 | /** |
1313 | * generic_make_request: hand a buffer to its device driver for I/O | |
1314 | * @bio: The bio describing the location in memory and on the device. | |
1315 | * | |
1316 | * generic_make_request() is used to make I/O requests of block | |
1317 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1318 | * to be done. | |
1319 | * | |
1320 | * generic_make_request() does not return any status. The | |
1321 | * success/failure status of the request, along with notification of | |
1322 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1323 | * function described (one day) else where. | |
1324 | * | |
1325 | * The caller of generic_make_request must make sure that bi_io_vec | |
1326 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1327 | * set to describe the device address, and the | |
1328 | * bi_end_io and optionally bi_private are set to describe how | |
1329 | * completion notification should be signaled. | |
1330 | * | |
1331 | * generic_make_request and the drivers it calls may use bi_next if this | |
1332 | * bio happens to be merged with someone else, and may change bi_dev and | |
1333 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1334 | * should NOT be depended on after the call to generic_make_request. | |
1335 | */ | |
d89d8796 | 1336 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1337 | { |
165125e1 | 1338 | struct request_queue *q; |
5ddfe969 | 1339 | sector_t old_sector; |
1da177e4 | 1340 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1341 | dev_t old_dev; |
51fd77bd | 1342 | int err = -EIO; |
1da177e4 LT |
1343 | |
1344 | might_sleep(); | |
1da177e4 | 1345 | |
c07e2b41 JA |
1346 | if (bio_check_eod(bio, nr_sectors)) |
1347 | goto end_io; | |
1da177e4 LT |
1348 | |
1349 | /* | |
1350 | * Resolve the mapping until finished. (drivers are | |
1351 | * still free to implement/resolve their own stacking | |
1352 | * by explicitly returning 0) | |
1353 | * | |
1354 | * NOTE: we don't repeat the blk_size check for each new device. | |
1355 | * Stacking drivers are expected to know what they are doing. | |
1356 | */ | |
5ddfe969 | 1357 | old_sector = -1; |
2056a782 | 1358 | old_dev = 0; |
1da177e4 LT |
1359 | do { |
1360 | char b[BDEVNAME_SIZE]; | |
1361 | ||
1362 | q = bdev_get_queue(bio->bi_bdev); | |
1363 | if (!q) { | |
1364 | printk(KERN_ERR | |
1365 | "generic_make_request: Trying to access " | |
1366 | "nonexistent block-device %s (%Lu)\n", | |
1367 | bdevname(bio->bi_bdev, b), | |
1368 | (long long) bio->bi_sector); | |
1369 | end_io: | |
51fd77bd | 1370 | bio_endio(bio, err); |
1da177e4 LT |
1371 | break; |
1372 | } | |
1373 | ||
4fa253f3 | 1374 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
6728cb0e | 1375 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1da177e4 LT |
1376 | bdevname(bio->bi_bdev, b), |
1377 | bio_sectors(bio), | |
1378 | q->max_hw_sectors); | |
1379 | goto end_io; | |
1380 | } | |
1381 | ||
fde6ad22 | 1382 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1383 | goto end_io; |
1384 | ||
c17bb495 AM |
1385 | if (should_fail_request(bio)) |
1386 | goto end_io; | |
1387 | ||
1da177e4 LT |
1388 | /* |
1389 | * If this device has partitions, remap block n | |
1390 | * of partition p to block n+start(p) of the disk. | |
1391 | */ | |
1392 | blk_partition_remap(bio); | |
1393 | ||
5ddfe969 | 1394 | if (old_sector != -1) |
4fa253f3 | 1395 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 1396 | old_sector); |
2056a782 JA |
1397 | |
1398 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); | |
1399 | ||
5ddfe969 | 1400 | old_sector = bio->bi_sector; |
2056a782 JA |
1401 | old_dev = bio->bi_bdev->bd_dev; |
1402 | ||
c07e2b41 JA |
1403 | if (bio_check_eod(bio, nr_sectors)) |
1404 | goto end_io; | |
51fd77bd JA |
1405 | if (bio_empty_barrier(bio) && !q->prepare_flush_fn) { |
1406 | err = -EOPNOTSUPP; | |
1407 | goto end_io; | |
1408 | } | |
5ddfe969 | 1409 | |
1da177e4 LT |
1410 | ret = q->make_request_fn(q, bio); |
1411 | } while (ret); | |
1412 | } | |
1413 | ||
d89d8796 NB |
1414 | /* |
1415 | * We only want one ->make_request_fn to be active at a time, | |
1416 | * else stack usage with stacked devices could be a problem. | |
1417 | * So use current->bio_{list,tail} to keep a list of requests | |
1418 | * submited by a make_request_fn function. | |
1419 | * current->bio_tail is also used as a flag to say if | |
1420 | * generic_make_request is currently active in this task or not. | |
1421 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1422 | * then a make_request is active, and new requests should be added | |
1423 | * at the tail | |
1424 | */ | |
1425 | void generic_make_request(struct bio *bio) | |
1426 | { | |
1427 | if (current->bio_tail) { | |
1428 | /* make_request is active */ | |
1429 | *(current->bio_tail) = bio; | |
1430 | bio->bi_next = NULL; | |
1431 | current->bio_tail = &bio->bi_next; | |
1432 | return; | |
1433 | } | |
1434 | /* following loop may be a bit non-obvious, and so deserves some | |
1435 | * explanation. | |
1436 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1437 | * ensure that) so we have a list with a single bio. | |
1438 | * We pretend that we have just taken it off a longer list, so | |
1439 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1440 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1441 | * added. __generic_make_request may indeed add some more bios | |
1442 | * through a recursive call to generic_make_request. If it | |
1443 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1444 | * from the top. In this case we really did just take the bio | |
1445 | * of the top of the list (no pretending) and so fixup bio_list and | |
1446 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1447 | * | |
1448 | * The loop was structured like this to make only one call to | |
1449 | * __generic_make_request (which is important as it is large and | |
1450 | * inlined) and to keep the structure simple. | |
1451 | */ | |
1452 | BUG_ON(bio->bi_next); | |
1453 | do { | |
1454 | current->bio_list = bio->bi_next; | |
1455 | if (bio->bi_next == NULL) | |
1456 | current->bio_tail = ¤t->bio_list; | |
1457 | else | |
1458 | bio->bi_next = NULL; | |
1459 | __generic_make_request(bio); | |
1460 | bio = current->bio_list; | |
1461 | } while (bio); | |
1462 | current->bio_tail = NULL; /* deactivate */ | |
1463 | } | |
1da177e4 LT |
1464 | EXPORT_SYMBOL(generic_make_request); |
1465 | ||
1466 | /** | |
1467 | * submit_bio: submit a bio to the block device layer for I/O | |
1468 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
1469 | * @bio: The &struct bio which describes the I/O | |
1470 | * | |
1471 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1472 | * uses that function to do most of the work. Both are fairly rough | |
1473 | * interfaces, @bio must be presetup and ready for I/O. | |
1474 | * | |
1475 | */ | |
1476 | void submit_bio(int rw, struct bio *bio) | |
1477 | { | |
1478 | int count = bio_sectors(bio); | |
1479 | ||
22e2c507 | 1480 | bio->bi_rw |= rw; |
1da177e4 | 1481 | |
bf2de6f5 JA |
1482 | /* |
1483 | * If it's a regular read/write or a barrier with data attached, | |
1484 | * go through the normal accounting stuff before submission. | |
1485 | */ | |
1486 | if (!bio_empty_barrier(bio)) { | |
1487 | ||
1488 | BIO_BUG_ON(!bio->bi_size); | |
1489 | BIO_BUG_ON(!bio->bi_io_vec); | |
1490 | ||
1491 | if (rw & WRITE) { | |
1492 | count_vm_events(PGPGOUT, count); | |
1493 | } else { | |
1494 | task_io_account_read(bio->bi_size); | |
1495 | count_vm_events(PGPGIN, count); | |
1496 | } | |
1497 | ||
1498 | if (unlikely(block_dump)) { | |
1499 | char b[BDEVNAME_SIZE]; | |
1500 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
ba25f9dc | 1501 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1502 | (rw & WRITE) ? "WRITE" : "READ", |
1503 | (unsigned long long)bio->bi_sector, | |
6728cb0e | 1504 | bdevname(bio->bi_bdev, b)); |
bf2de6f5 | 1505 | } |
1da177e4 LT |
1506 | } |
1507 | ||
1508 | generic_make_request(bio); | |
1509 | } | |
1da177e4 LT |
1510 | EXPORT_SYMBOL(submit_bio); |
1511 | ||
3bcddeac KU |
1512 | /** |
1513 | * __end_that_request_first - end I/O on a request | |
1514 | * @req: the request being processed | |
5450d3e1 | 1515 | * @error: 0 for success, < 0 for error |
3bcddeac KU |
1516 | * @nr_bytes: number of bytes to complete |
1517 | * | |
1518 | * Description: | |
1519 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
1520 | * for the next range of segments (if any) in the cluster. | |
1521 | * | |
1522 | * Return: | |
1523 | * 0 - we are done with this request, call end_that_request_last() | |
1524 | * 1 - still buffers pending for this request | |
1525 | **/ | |
5450d3e1 | 1526 | static int __end_that_request_first(struct request *req, int error, |
1da177e4 LT |
1527 | int nr_bytes) |
1528 | { | |
5450d3e1 | 1529 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
1530 | struct bio *bio; |
1531 | ||
2056a782 JA |
1532 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); |
1533 | ||
1da177e4 LT |
1534 | /* |
1535 | * for a REQ_BLOCK_PC request, we want to carry any eventual | |
1536 | * sense key with us all the way through | |
1537 | */ | |
1538 | if (!blk_pc_request(req)) | |
1539 | req->errors = 0; | |
1540 | ||
6728cb0e JA |
1541 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { |
1542 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1da177e4 LT |
1543 | req->rq_disk ? req->rq_disk->disk_name : "?", |
1544 | (unsigned long long)req->sector); | |
1545 | } | |
1546 | ||
d72d904a | 1547 | if (blk_fs_request(req) && req->rq_disk) { |
a362357b JA |
1548 | const int rw = rq_data_dir(req); |
1549 | ||
6f2576af JM |
1550 | all_stat_add(req->rq_disk, sectors[rw], |
1551 | nr_bytes >> 9, req->sector); | |
d72d904a JA |
1552 | } |
1553 | ||
1da177e4 LT |
1554 | total_bytes = bio_nbytes = 0; |
1555 | while ((bio = req->bio) != NULL) { | |
1556 | int nbytes; | |
1557 | ||
bf2de6f5 JA |
1558 | /* |
1559 | * For an empty barrier request, the low level driver must | |
1560 | * store a potential error location in ->sector. We pass | |
1561 | * that back up in ->bi_sector. | |
1562 | */ | |
1563 | if (blk_empty_barrier(req)) | |
1564 | bio->bi_sector = req->sector; | |
1565 | ||
1da177e4 LT |
1566 | if (nr_bytes >= bio->bi_size) { |
1567 | req->bio = bio->bi_next; | |
1568 | nbytes = bio->bi_size; | |
5bb23a68 | 1569 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
1570 | next_idx = 0; |
1571 | bio_nbytes = 0; | |
1572 | } else { | |
1573 | int idx = bio->bi_idx + next_idx; | |
1574 | ||
1575 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
1576 | blk_dump_rq_flags(req, "__end_that"); | |
6728cb0e JA |
1577 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
1578 | __FUNCTION__, bio->bi_idx, | |
1579 | bio->bi_vcnt); | |
1da177e4 LT |
1580 | break; |
1581 | } | |
1582 | ||
1583 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1584 | BIO_BUG_ON(nbytes > bio->bi_size); | |
1585 | ||
1586 | /* | |
1587 | * not a complete bvec done | |
1588 | */ | |
1589 | if (unlikely(nbytes > nr_bytes)) { | |
1590 | bio_nbytes += nr_bytes; | |
1591 | total_bytes += nr_bytes; | |
1592 | break; | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * advance to the next vector | |
1597 | */ | |
1598 | next_idx++; | |
1599 | bio_nbytes += nbytes; | |
1600 | } | |
1601 | ||
1602 | total_bytes += nbytes; | |
1603 | nr_bytes -= nbytes; | |
1604 | ||
6728cb0e JA |
1605 | bio = req->bio; |
1606 | if (bio) { | |
1da177e4 LT |
1607 | /* |
1608 | * end more in this run, or just return 'not-done' | |
1609 | */ | |
1610 | if (unlikely(nr_bytes <= 0)) | |
1611 | break; | |
1612 | } | |
1613 | } | |
1614 | ||
1615 | /* | |
1616 | * completely done | |
1617 | */ | |
1618 | if (!req->bio) | |
1619 | return 0; | |
1620 | ||
1621 | /* | |
1622 | * if the request wasn't completed, update state | |
1623 | */ | |
1624 | if (bio_nbytes) { | |
5bb23a68 | 1625 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
1626 | bio->bi_idx += next_idx; |
1627 | bio_iovec(bio)->bv_offset += nr_bytes; | |
1628 | bio_iovec(bio)->bv_len -= nr_bytes; | |
1629 | } | |
1630 | ||
1631 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
1632 | blk_recalc_rq_segments(req); | |
1633 | return 1; | |
1634 | } | |
1635 | ||
ff856bad JA |
1636 | /* |
1637 | * splice the completion data to a local structure and hand off to | |
1638 | * process_completion_queue() to complete the requests | |
1639 | */ | |
1640 | static void blk_done_softirq(struct softirq_action *h) | |
1641 | { | |
626ab0e6 | 1642 | struct list_head *cpu_list, local_list; |
ff856bad JA |
1643 | |
1644 | local_irq_disable(); | |
1645 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
626ab0e6 | 1646 | list_replace_init(cpu_list, &local_list); |
ff856bad JA |
1647 | local_irq_enable(); |
1648 | ||
1649 | while (!list_empty(&local_list)) { | |
6728cb0e | 1650 | struct request *rq; |
ff856bad | 1651 | |
6728cb0e | 1652 | rq = list_entry(local_list.next, struct request, donelist); |
ff856bad JA |
1653 | list_del_init(&rq->donelist); |
1654 | rq->q->softirq_done_fn(rq); | |
1655 | } | |
1656 | } | |
1657 | ||
6728cb0e JA |
1658 | static int __cpuinit blk_cpu_notify(struct notifier_block *self, |
1659 | unsigned long action, void *hcpu) | |
ff856bad JA |
1660 | { |
1661 | /* | |
1662 | * If a CPU goes away, splice its entries to the current CPU | |
1663 | * and trigger a run of the softirq | |
1664 | */ | |
8bb78442 | 1665 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
ff856bad JA |
1666 | int cpu = (unsigned long) hcpu; |
1667 | ||
1668 | local_irq_disable(); | |
1669 | list_splice_init(&per_cpu(blk_cpu_done, cpu), | |
1670 | &__get_cpu_var(blk_cpu_done)); | |
1671 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
1672 | local_irq_enable(); | |
1673 | } | |
1674 | ||
1675 | return NOTIFY_OK; | |
1676 | } | |
1677 | ||
1678 | ||
db47d475 | 1679 | static struct notifier_block blk_cpu_notifier __cpuinitdata = { |
ff856bad JA |
1680 | .notifier_call = blk_cpu_notify, |
1681 | }; | |
1682 | ||
ff856bad JA |
1683 | /** |
1684 | * blk_complete_request - end I/O on a request | |
1685 | * @req: the request being processed | |
1686 | * | |
1687 | * Description: | |
1688 | * Ends all I/O on a request. It does not handle partial completions, | |
d6e05edc | 1689 | * unless the driver actually implements this in its completion callback |
4fa253f3 | 1690 | * through requeueing. The actual completion happens out-of-order, |
ff856bad JA |
1691 | * through a softirq handler. The user must have registered a completion |
1692 | * callback through blk_queue_softirq_done(). | |
1693 | **/ | |
1694 | ||
1695 | void blk_complete_request(struct request *req) | |
1696 | { | |
1697 | struct list_head *cpu_list; | |
1698 | unsigned long flags; | |
1699 | ||
1700 | BUG_ON(!req->q->softirq_done_fn); | |
6728cb0e | 1701 | |
ff856bad JA |
1702 | local_irq_save(flags); |
1703 | ||
1704 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
1705 | list_add_tail(&req->donelist, cpu_list); | |
1706 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
1707 | ||
1708 | local_irq_restore(flags); | |
1709 | } | |
ff856bad | 1710 | EXPORT_SYMBOL(blk_complete_request); |
6728cb0e | 1711 | |
1da177e4 LT |
1712 | /* |
1713 | * queue lock must be held | |
1714 | */ | |
5450d3e1 | 1715 | static void end_that_request_last(struct request *req, int error) |
1da177e4 LT |
1716 | { |
1717 | struct gendisk *disk = req->rq_disk; | |
8ffdc655 | 1718 | |
b8286239 KU |
1719 | if (blk_rq_tagged(req)) |
1720 | blk_queue_end_tag(req->q, req); | |
1721 | ||
1722 | if (blk_queued_rq(req)) | |
1723 | blkdev_dequeue_request(req); | |
1da177e4 LT |
1724 | |
1725 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1726 | laptop_io_completion(); | |
1727 | ||
fd0ff8aa JA |
1728 | /* |
1729 | * Account IO completion. bar_rq isn't accounted as a normal | |
1730 | * IO on queueing nor completion. Accounting the containing | |
1731 | * request is enough. | |
1732 | */ | |
1733 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
1da177e4 | 1734 | unsigned long duration = jiffies - req->start_time; |
a362357b | 1735 | const int rw = rq_data_dir(req); |
6f2576af | 1736 | struct hd_struct *part = get_part(disk, req->sector); |
a362357b | 1737 | |
6f2576af JM |
1738 | __all_stat_inc(disk, ios[rw], req->sector); |
1739 | __all_stat_add(disk, ticks[rw], duration, req->sector); | |
1da177e4 LT |
1740 | disk_round_stats(disk); |
1741 | disk->in_flight--; | |
6f2576af JM |
1742 | if (part) { |
1743 | part_round_stats(part); | |
1744 | part->in_flight--; | |
1745 | } | |
1da177e4 | 1746 | } |
b8286239 | 1747 | |
1da177e4 | 1748 | if (req->end_io) |
8ffdc655 | 1749 | req->end_io(req, error); |
b8286239 KU |
1750 | else { |
1751 | if (blk_bidi_rq(req)) | |
1752 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1753 | ||
1da177e4 | 1754 | __blk_put_request(req->q, req); |
b8286239 | 1755 | } |
1da177e4 LT |
1756 | } |
1757 | ||
a0cd1285 | 1758 | static inline void __end_request(struct request *rq, int uptodate, |
9e6e39f2 | 1759 | unsigned int nr_bytes) |
1da177e4 | 1760 | { |
9e6e39f2 KU |
1761 | int error = 0; |
1762 | ||
1763 | if (uptodate <= 0) | |
1764 | error = uptodate ? uptodate : -EIO; | |
1765 | ||
1766 | __blk_end_request(rq, error, nr_bytes); | |
1da177e4 LT |
1767 | } |
1768 | ||
3b11313a KU |
1769 | /** |
1770 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
1771 | **/ | |
1772 | unsigned int blk_rq_bytes(struct request *rq) | |
a0cd1285 JA |
1773 | { |
1774 | if (blk_fs_request(rq)) | |
1775 | return rq->hard_nr_sectors << 9; | |
1776 | ||
1777 | return rq->data_len; | |
1778 | } | |
3b11313a KU |
1779 | EXPORT_SYMBOL_GPL(blk_rq_bytes); |
1780 | ||
1781 | /** | |
1782 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
1783 | **/ | |
1784 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
1785 | { | |
1786 | if (blk_fs_request(rq)) | |
1787 | return rq->current_nr_sectors << 9; | |
1788 | ||
1789 | if (rq->bio) | |
1790 | return rq->bio->bi_size; | |
1791 | ||
1792 | return rq->data_len; | |
1793 | } | |
1794 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
a0cd1285 JA |
1795 | |
1796 | /** | |
1797 | * end_queued_request - end all I/O on a queued request | |
1798 | * @rq: the request being processed | |
1799 | * @uptodate: error value or 0/1 uptodate flag | |
1800 | * | |
1801 | * Description: | |
1802 | * Ends all I/O on a request, and removes it from the block layer queues. | |
1803 | * Not suitable for normal IO completion, unless the driver still has | |
1804 | * the request attached to the block layer. | |
1805 | * | |
1806 | **/ | |
1807 | void end_queued_request(struct request *rq, int uptodate) | |
1808 | { | |
9e6e39f2 | 1809 | __end_request(rq, uptodate, blk_rq_bytes(rq)); |
a0cd1285 JA |
1810 | } |
1811 | EXPORT_SYMBOL(end_queued_request); | |
1812 | ||
1813 | /** | |
1814 | * end_dequeued_request - end all I/O on a dequeued request | |
1815 | * @rq: the request being processed | |
1816 | * @uptodate: error value or 0/1 uptodate flag | |
1817 | * | |
1818 | * Description: | |
1819 | * Ends all I/O on a request. The request must already have been | |
1820 | * dequeued using blkdev_dequeue_request(), as is normally the case | |
1821 | * for most drivers. | |
1822 | * | |
1823 | **/ | |
1824 | void end_dequeued_request(struct request *rq, int uptodate) | |
1825 | { | |
9e6e39f2 | 1826 | __end_request(rq, uptodate, blk_rq_bytes(rq)); |
a0cd1285 JA |
1827 | } |
1828 | EXPORT_SYMBOL(end_dequeued_request); | |
1829 | ||
1830 | ||
1831 | /** | |
1832 | * end_request - end I/O on the current segment of the request | |
8f731f7d | 1833 | * @req: the request being processed |
a0cd1285 JA |
1834 | * @uptodate: error value or 0/1 uptodate flag |
1835 | * | |
1836 | * Description: | |
1837 | * Ends I/O on the current segment of a request. If that is the only | |
1838 | * remaining segment, the request is also completed and freed. | |
1839 | * | |
1840 | * This is a remnant of how older block drivers handled IO completions. | |
1841 | * Modern drivers typically end IO on the full request in one go, unless | |
1842 | * they have a residual value to account for. For that case this function | |
1843 | * isn't really useful, unless the residual just happens to be the | |
1844 | * full current segment. In other words, don't use this function in new | |
1845 | * code. Either use end_request_completely(), or the | |
1846 | * end_that_request_chunk() (along with end_that_request_last()) for | |
1847 | * partial completions. | |
1848 | * | |
1849 | **/ | |
1850 | void end_request(struct request *req, int uptodate) | |
1851 | { | |
9e6e39f2 | 1852 | __end_request(req, uptodate, req->hard_cur_sectors << 9); |
a0cd1285 | 1853 | } |
1da177e4 LT |
1854 | EXPORT_SYMBOL(end_request); |
1855 | ||
336cdb40 | 1856 | /** |
e19a3ab0 KU |
1857 | * blk_end_io - Generic end_io function to complete a request. |
1858 | * @rq: the request being processed | |
1859 | * @error: 0 for success, < 0 for error | |
e3a04fe3 KU |
1860 | * @nr_bytes: number of bytes to complete @rq |
1861 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
e19a3ab0 KU |
1862 | * @drv_callback: function called between completion of bios in the request |
1863 | * and completion of the request. | |
1864 | * If the callback returns non 0, this helper returns without | |
1865 | * completion of the request. | |
336cdb40 KU |
1866 | * |
1867 | * Description: | |
e3a04fe3 | 1868 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
336cdb40 KU |
1869 | * If @rq has leftover, sets it up for the next range of segments. |
1870 | * | |
1871 | * Return: | |
1872 | * 0 - we are done with this request | |
e19a3ab0 | 1873 | * 1 - this request is not freed yet, it still has pending buffers. |
336cdb40 | 1874 | **/ |
22b13210 JA |
1875 | static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes, |
1876 | unsigned int bidi_bytes, | |
1877 | int (drv_callback)(struct request *)) | |
336cdb40 KU |
1878 | { |
1879 | struct request_queue *q = rq->q; | |
1880 | unsigned long flags = 0UL; | |
336cdb40 KU |
1881 | |
1882 | if (blk_fs_request(rq) || blk_pc_request(rq)) { | |
5450d3e1 | 1883 | if (__end_that_request_first(rq, error, nr_bytes)) |
336cdb40 | 1884 | return 1; |
e3a04fe3 KU |
1885 | |
1886 | /* Bidi request must be completed as a whole */ | |
1887 | if (blk_bidi_rq(rq) && | |
5450d3e1 | 1888 | __end_that_request_first(rq->next_rq, error, bidi_bytes)) |
e3a04fe3 | 1889 | return 1; |
336cdb40 KU |
1890 | } |
1891 | ||
e19a3ab0 KU |
1892 | /* Special feature for tricky drivers */ |
1893 | if (drv_callback && drv_callback(rq)) | |
1894 | return 1; | |
1895 | ||
336cdb40 KU |
1896 | add_disk_randomness(rq->rq_disk); |
1897 | ||
1898 | spin_lock_irqsave(q->queue_lock, flags); | |
b8286239 | 1899 | end_that_request_last(rq, error); |
336cdb40 KU |
1900 | spin_unlock_irqrestore(q->queue_lock, flags); |
1901 | ||
1902 | return 0; | |
1903 | } | |
e19a3ab0 KU |
1904 | |
1905 | /** | |
1906 | * blk_end_request - Helper function for drivers to complete the request. | |
1907 | * @rq: the request being processed | |
1908 | * @error: 0 for success, < 0 for error | |
1909 | * @nr_bytes: number of bytes to complete | |
1910 | * | |
1911 | * Description: | |
1912 | * Ends I/O on a number of bytes attached to @rq. | |
1913 | * If @rq has leftover, sets it up for the next range of segments. | |
1914 | * | |
1915 | * Return: | |
1916 | * 0 - we are done with this request | |
1917 | * 1 - still buffers pending for this request | |
1918 | **/ | |
22b13210 | 1919 | int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 1920 | { |
e3a04fe3 | 1921 | return blk_end_io(rq, error, nr_bytes, 0, NULL); |
e19a3ab0 | 1922 | } |
336cdb40 KU |
1923 | EXPORT_SYMBOL_GPL(blk_end_request); |
1924 | ||
1925 | /** | |
1926 | * __blk_end_request - Helper function for drivers to complete the request. | |
1927 | * @rq: the request being processed | |
1928 | * @error: 0 for success, < 0 for error | |
1929 | * @nr_bytes: number of bytes to complete | |
1930 | * | |
1931 | * Description: | |
1932 | * Must be called with queue lock held unlike blk_end_request(). | |
1933 | * | |
1934 | * Return: | |
1935 | * 0 - we are done with this request | |
1936 | * 1 - still buffers pending for this request | |
1937 | **/ | |
22b13210 | 1938 | int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
336cdb40 | 1939 | { |
336cdb40 | 1940 | if (blk_fs_request(rq) || blk_pc_request(rq)) { |
5450d3e1 | 1941 | if (__end_that_request_first(rq, error, nr_bytes)) |
336cdb40 KU |
1942 | return 1; |
1943 | } | |
1944 | ||
1945 | add_disk_randomness(rq->rq_disk); | |
1946 | ||
b8286239 | 1947 | end_that_request_last(rq, error); |
336cdb40 KU |
1948 | |
1949 | return 0; | |
1950 | } | |
1951 | EXPORT_SYMBOL_GPL(__blk_end_request); | |
1952 | ||
e3a04fe3 KU |
1953 | /** |
1954 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. | |
1955 | * @rq: the bidi request being processed | |
1956 | * @error: 0 for success, < 0 for error | |
1957 | * @nr_bytes: number of bytes to complete @rq | |
1958 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
1959 | * | |
1960 | * Description: | |
1961 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
1962 | * | |
1963 | * Return: | |
1964 | * 0 - we are done with this request | |
1965 | * 1 - still buffers pending for this request | |
1966 | **/ | |
22b13210 JA |
1967 | int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, |
1968 | unsigned int bidi_bytes) | |
e3a04fe3 KU |
1969 | { |
1970 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); | |
1971 | } | |
1972 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
1973 | ||
e19a3ab0 KU |
1974 | /** |
1975 | * blk_end_request_callback - Special helper function for tricky drivers | |
1976 | * @rq: the request being processed | |
1977 | * @error: 0 for success, < 0 for error | |
1978 | * @nr_bytes: number of bytes to complete | |
1979 | * @drv_callback: function called between completion of bios in the request | |
1980 | * and completion of the request. | |
1981 | * If the callback returns non 0, this helper returns without | |
1982 | * completion of the request. | |
1983 | * | |
1984 | * Description: | |
1985 | * Ends I/O on a number of bytes attached to @rq. | |
1986 | * If @rq has leftover, sets it up for the next range of segments. | |
1987 | * | |
1988 | * This special helper function is used only for existing tricky drivers. | |
1989 | * (e.g. cdrom_newpc_intr() of ide-cd) | |
1990 | * This interface will be removed when such drivers are rewritten. | |
1991 | * Don't use this interface in other places anymore. | |
1992 | * | |
1993 | * Return: | |
1994 | * 0 - we are done with this request | |
1995 | * 1 - this request is not freed yet. | |
1996 | * this request still has pending buffers or | |
1997 | * the driver doesn't want to finish this request yet. | |
1998 | **/ | |
22b13210 JA |
1999 | int blk_end_request_callback(struct request *rq, int error, |
2000 | unsigned int nr_bytes, | |
e19a3ab0 KU |
2001 | int (drv_callback)(struct request *)) |
2002 | { | |
e3a04fe3 | 2003 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); |
e19a3ab0 KU |
2004 | } |
2005 | EXPORT_SYMBOL_GPL(blk_end_request_callback); | |
2006 | ||
86db1e29 JA |
2007 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2008 | struct bio *bio) | |
1da177e4 | 2009 | { |
4aff5e23 JA |
2010 | /* first two bits are identical in rq->cmd_flags and bio->bi_rw */ |
2011 | rq->cmd_flags |= (bio->bi_rw & 3); | |
1da177e4 LT |
2012 | |
2013 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
2014 | rq->nr_hw_segments = bio_hw_segments(q, bio); | |
2015 | rq->current_nr_sectors = bio_cur_sectors(bio); | |
2016 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
2017 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
2018 | rq->buffer = bio_data(bio); | |
6b00769f | 2019 | rq->raw_data_len = bio->bi_size; |
0e75f906 | 2020 | rq->data_len = bio->bi_size; |
1da177e4 LT |
2021 | |
2022 | rq->bio = rq->biotail = bio; | |
1da177e4 | 2023 | |
66846572 N |
2024 | if (bio->bi_bdev) |
2025 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2026 | } | |
1da177e4 LT |
2027 | |
2028 | int kblockd_schedule_work(struct work_struct *work) | |
2029 | { | |
2030 | return queue_work(kblockd_workqueue, work); | |
2031 | } | |
1da177e4 LT |
2032 | EXPORT_SYMBOL(kblockd_schedule_work); |
2033 | ||
19a75d83 | 2034 | void kblockd_flush_work(struct work_struct *work) |
1da177e4 | 2035 | { |
28e53bdd | 2036 | cancel_work_sync(work); |
1da177e4 | 2037 | } |
19a75d83 | 2038 | EXPORT_SYMBOL(kblockd_flush_work); |
1da177e4 LT |
2039 | |
2040 | int __init blk_dev_init(void) | |
2041 | { | |
ff856bad JA |
2042 | int i; |
2043 | ||
1da177e4 LT |
2044 | kblockd_workqueue = create_workqueue("kblockd"); |
2045 | if (!kblockd_workqueue) | |
2046 | panic("Failed to create kblockd\n"); | |
2047 | ||
2048 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2049 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2050 | |
8324aa91 | 2051 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2052 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2053 | |
0a945022 | 2054 | for_each_possible_cpu(i) |
ff856bad JA |
2055 | INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i)); |
2056 | ||
2057 | open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL); | |
5a67e4c5 | 2058 | register_hotcpu_notifier(&blk_cpu_notifier); |
ff856bad | 2059 | |
d38ecf93 | 2060 | return 0; |
1da177e4 | 2061 | } |
1da177e4 | 2062 |