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> | |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000 | |
7 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 | |
8 | */ | |
9 | ||
10 | /* | |
11 | * This handles all read/write requests to block devices | |
12 | */ | |
1da177e4 LT |
13 | #include <linux/kernel.h> |
14 | #include <linux/module.h> | |
15 | #include <linux/backing-dev.h> | |
16 | #include <linux/bio.h> | |
17 | #include <linux/blkdev.h> | |
18 | #include <linux/highmem.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/kernel_stat.h> | |
21 | #include <linux/string.h> | |
22 | #include <linux/init.h> | |
23 | #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */ | |
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 LT |
33 | |
34 | /* | |
35 | * for max sense size | |
36 | */ | |
37 | #include <scsi/scsi_cmnd.h> | |
38 | ||
65f27f38 | 39 | static void blk_unplug_work(struct work_struct *work); |
1da177e4 | 40 | static void blk_unplug_timeout(unsigned long data); |
93d17d3d | 41 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io); |
52d9e675 | 42 | static void init_request_from_bio(struct request *req, struct bio *bio); |
165125e1 | 43 | static int __make_request(struct request_queue *q, struct bio *bio); |
b5deef90 | 44 | static struct io_context *current_io_context(gfp_t gfp_flags, int node); |
1da177e4 LT |
45 | |
46 | /* | |
47 | * For the allocated request tables | |
48 | */ | |
e18b890b | 49 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
50 | |
51 | /* | |
52 | * For queue allocation | |
53 | */ | |
e18b890b | 54 | static struct kmem_cache *requestq_cachep; |
1da177e4 LT |
55 | |
56 | /* | |
57 | * For io context allocations | |
58 | */ | |
e18b890b | 59 | static struct kmem_cache *iocontext_cachep; |
1da177e4 | 60 | |
1da177e4 LT |
61 | /* |
62 | * Controlling structure to kblockd | |
63 | */ | |
ff856bad | 64 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 LT |
65 | |
66 | unsigned long blk_max_low_pfn, blk_max_pfn; | |
67 | ||
68 | EXPORT_SYMBOL(blk_max_low_pfn); | |
69 | EXPORT_SYMBOL(blk_max_pfn); | |
70 | ||
ff856bad JA |
71 | static DEFINE_PER_CPU(struct list_head, blk_cpu_done); |
72 | ||
1da177e4 LT |
73 | /* Amount of time in which a process may batch requests */ |
74 | #define BLK_BATCH_TIME (HZ/50UL) | |
75 | ||
76 | /* Number of requests a "batching" process may submit */ | |
77 | #define BLK_BATCH_REQ 32 | |
78 | ||
79 | /* | |
80 | * Return the threshold (number of used requests) at which the queue is | |
81 | * considered to be congested. It include a little hysteresis to keep the | |
82 | * context switch rate down. | |
83 | */ | |
84 | static inline int queue_congestion_on_threshold(struct request_queue *q) | |
85 | { | |
86 | return q->nr_congestion_on; | |
87 | } | |
88 | ||
89 | /* | |
90 | * The threshold at which a queue is considered to be uncongested | |
91 | */ | |
92 | static inline int queue_congestion_off_threshold(struct request_queue *q) | |
93 | { | |
94 | return q->nr_congestion_off; | |
95 | } | |
96 | ||
97 | static void blk_queue_congestion_threshold(struct request_queue *q) | |
98 | { | |
99 | int nr; | |
100 | ||
101 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
102 | if (nr > q->nr_requests) | |
103 | nr = q->nr_requests; | |
104 | q->nr_congestion_on = nr; | |
105 | ||
106 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
107 | if (nr < 1) | |
108 | nr = 1; | |
109 | q->nr_congestion_off = nr; | |
110 | } | |
111 | ||
1da177e4 LT |
112 | /** |
113 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
114 | * @bdev: device | |
115 | * | |
116 | * Locates the passed device's request queue and returns the address of its | |
117 | * backing_dev_info | |
118 | * | |
119 | * Will return NULL if the request queue cannot be located. | |
120 | */ | |
121 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
122 | { | |
123 | struct backing_dev_info *ret = NULL; | |
165125e1 | 124 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
125 | |
126 | if (q) | |
127 | ret = &q->backing_dev_info; | |
128 | return ret; | |
129 | } | |
1da177e4 LT |
130 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
131 | ||
1da177e4 LT |
132 | /** |
133 | * blk_queue_prep_rq - set a prepare_request function for queue | |
134 | * @q: queue | |
135 | * @pfn: prepare_request function | |
136 | * | |
137 | * It's possible for a queue to register a prepare_request callback which | |
138 | * is invoked before the request is handed to the request_fn. The goal of | |
139 | * the function is to prepare a request for I/O, it can be used to build a | |
140 | * cdb from the request data for instance. | |
141 | * | |
142 | */ | |
165125e1 | 143 | void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) |
1da177e4 LT |
144 | { |
145 | q->prep_rq_fn = pfn; | |
146 | } | |
147 | ||
148 | EXPORT_SYMBOL(blk_queue_prep_rq); | |
149 | ||
150 | /** | |
151 | * blk_queue_merge_bvec - set a merge_bvec function for queue | |
152 | * @q: queue | |
153 | * @mbfn: merge_bvec_fn | |
154 | * | |
155 | * Usually queues have static limitations on the max sectors or segments that | |
156 | * we can put in a request. Stacking drivers may have some settings that | |
157 | * are dynamic, and thus we have to query the queue whether it is ok to | |
158 | * add a new bio_vec to a bio at a given offset or not. If the block device | |
159 | * has such limitations, it needs to register a merge_bvec_fn to control | |
160 | * the size of bio's sent to it. Note that a block device *must* allow a | |
161 | * single page to be added to an empty bio. The block device driver may want | |
162 | * to use the bio_split() function to deal with these bio's. By default | |
163 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | |
164 | * honored. | |
165 | */ | |
165125e1 | 166 | void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) |
1da177e4 LT |
167 | { |
168 | q->merge_bvec_fn = mbfn; | |
169 | } | |
170 | ||
171 | EXPORT_SYMBOL(blk_queue_merge_bvec); | |
172 | ||
165125e1 | 173 | void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) |
ff856bad JA |
174 | { |
175 | q->softirq_done_fn = fn; | |
176 | } | |
177 | ||
178 | EXPORT_SYMBOL(blk_queue_softirq_done); | |
179 | ||
1da177e4 LT |
180 | /** |
181 | * blk_queue_make_request - define an alternate make_request function for a device | |
182 | * @q: the request queue for the device to be affected | |
183 | * @mfn: the alternate make_request function | |
184 | * | |
185 | * Description: | |
186 | * The normal way for &struct bios to be passed to a device | |
187 | * driver is for them to be collected into requests on a request | |
188 | * queue, and then to allow the device driver to select requests | |
189 | * off that queue when it is ready. This works well for many block | |
190 | * devices. However some block devices (typically virtual devices | |
191 | * such as md or lvm) do not benefit from the processing on the | |
192 | * request queue, and are served best by having the requests passed | |
193 | * directly to them. This can be achieved by providing a function | |
194 | * to blk_queue_make_request(). | |
195 | * | |
196 | * Caveat: | |
197 | * The driver that does this *must* be able to deal appropriately | |
198 | * with buffers in "highmemory". This can be accomplished by either calling | |
199 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
200 | * blk_queue_bounce() to create a buffer in normal memory. | |
201 | **/ | |
165125e1 | 202 | void blk_queue_make_request(struct request_queue * q, make_request_fn * mfn) |
1da177e4 LT |
203 | { |
204 | /* | |
205 | * set defaults | |
206 | */ | |
207 | q->nr_requests = BLKDEV_MAX_RQ; | |
309c0a1d SM |
208 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); |
209 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
1da177e4 LT |
210 | q->make_request_fn = mfn; |
211 | q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
212 | q->backing_dev_info.state = 0; | |
213 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
defd94b7 | 214 | blk_queue_max_sectors(q, SAFE_MAX_SECTORS); |
1da177e4 LT |
215 | blk_queue_hardsect_size(q, 512); |
216 | blk_queue_dma_alignment(q, 511); | |
217 | blk_queue_congestion_threshold(q); | |
218 | q->nr_batching = BLK_BATCH_REQ; | |
219 | ||
220 | q->unplug_thresh = 4; /* hmm */ | |
221 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ | |
222 | if (q->unplug_delay == 0) | |
223 | q->unplug_delay = 1; | |
224 | ||
65f27f38 | 225 | INIT_WORK(&q->unplug_work, blk_unplug_work); |
1da177e4 LT |
226 | |
227 | q->unplug_timer.function = blk_unplug_timeout; | |
228 | q->unplug_timer.data = (unsigned long)q; | |
229 | ||
230 | /* | |
231 | * by default assume old behaviour and bounce for any highmem page | |
232 | */ | |
233 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
1da177e4 LT |
234 | } |
235 | ||
236 | EXPORT_SYMBOL(blk_queue_make_request); | |
237 | ||
165125e1 | 238 | static void rq_init(struct request_queue *q, struct request *rq) |
1da177e4 LT |
239 | { |
240 | INIT_LIST_HEAD(&rq->queuelist); | |
ff856bad | 241 | INIT_LIST_HEAD(&rq->donelist); |
1da177e4 LT |
242 | |
243 | rq->errors = 0; | |
1da177e4 | 244 | rq->bio = rq->biotail = NULL; |
2e662b65 JA |
245 | INIT_HLIST_NODE(&rq->hash); |
246 | RB_CLEAR_NODE(&rq->rb_node); | |
22e2c507 | 247 | rq->ioprio = 0; |
1da177e4 LT |
248 | rq->buffer = NULL; |
249 | rq->ref_count = 1; | |
250 | rq->q = q; | |
1da177e4 LT |
251 | rq->special = NULL; |
252 | rq->data_len = 0; | |
253 | rq->data = NULL; | |
df46b9a4 | 254 | rq->nr_phys_segments = 0; |
1da177e4 LT |
255 | rq->sense = NULL; |
256 | rq->end_io = NULL; | |
257 | rq->end_io_data = NULL; | |
ff856bad | 258 | rq->completion_data = NULL; |
abae1fde | 259 | rq->next_rq = NULL; |
1da177e4 LT |
260 | } |
261 | ||
262 | /** | |
263 | * blk_queue_ordered - does this queue support ordered writes | |
797e7dbb TH |
264 | * @q: the request queue |
265 | * @ordered: one of QUEUE_ORDERED_* | |
fddfdeaf | 266 | * @prepare_flush_fn: rq setup helper for cache flush ordered writes |
1da177e4 LT |
267 | * |
268 | * Description: | |
269 | * For journalled file systems, doing ordered writes on a commit | |
270 | * block instead of explicitly doing wait_on_buffer (which is bad | |
271 | * for performance) can be a big win. Block drivers supporting this | |
272 | * feature should call this function and indicate so. | |
273 | * | |
274 | **/ | |
165125e1 | 275 | int blk_queue_ordered(struct request_queue *q, unsigned ordered, |
797e7dbb TH |
276 | prepare_flush_fn *prepare_flush_fn) |
277 | { | |
278 | if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) && | |
279 | prepare_flush_fn == NULL) { | |
280 | printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n"); | |
281 | return -EINVAL; | |
282 | } | |
283 | ||
284 | if (ordered != QUEUE_ORDERED_NONE && | |
285 | ordered != QUEUE_ORDERED_DRAIN && | |
286 | ordered != QUEUE_ORDERED_DRAIN_FLUSH && | |
287 | ordered != QUEUE_ORDERED_DRAIN_FUA && | |
288 | ordered != QUEUE_ORDERED_TAG && | |
289 | ordered != QUEUE_ORDERED_TAG_FLUSH && | |
290 | ordered != QUEUE_ORDERED_TAG_FUA) { | |
291 | printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered); | |
292 | return -EINVAL; | |
1da177e4 | 293 | } |
797e7dbb | 294 | |
60481b12 | 295 | q->ordered = ordered; |
797e7dbb TH |
296 | q->next_ordered = ordered; |
297 | q->prepare_flush_fn = prepare_flush_fn; | |
298 | ||
299 | return 0; | |
1da177e4 LT |
300 | } |
301 | ||
302 | EXPORT_SYMBOL(blk_queue_ordered); | |
303 | ||
304 | /** | |
305 | * blk_queue_issue_flush_fn - set function for issuing a flush | |
306 | * @q: the request queue | |
307 | * @iff: the function to be called issuing the flush | |
308 | * | |
309 | * Description: | |
310 | * If a driver supports issuing a flush command, the support is notified | |
311 | * to the block layer by defining it through this call. | |
312 | * | |
313 | **/ | |
165125e1 | 314 | void blk_queue_issue_flush_fn(struct request_queue *q, issue_flush_fn *iff) |
1da177e4 LT |
315 | { |
316 | q->issue_flush_fn = iff; | |
317 | } | |
318 | ||
319 | EXPORT_SYMBOL(blk_queue_issue_flush_fn); | |
320 | ||
321 | /* | |
322 | * Cache flushing for ordered writes handling | |
323 | */ | |
165125e1 | 324 | inline unsigned blk_ordered_cur_seq(struct request_queue *q) |
1da177e4 | 325 | { |
797e7dbb TH |
326 | if (!q->ordseq) |
327 | return 0; | |
328 | return 1 << ffz(q->ordseq); | |
1da177e4 LT |
329 | } |
330 | ||
797e7dbb | 331 | unsigned blk_ordered_req_seq(struct request *rq) |
1da177e4 | 332 | { |
165125e1 | 333 | struct request_queue *q = rq->q; |
1da177e4 | 334 | |
797e7dbb | 335 | BUG_ON(q->ordseq == 0); |
8922e16c | 336 | |
797e7dbb TH |
337 | if (rq == &q->pre_flush_rq) |
338 | return QUEUE_ORDSEQ_PREFLUSH; | |
339 | if (rq == &q->bar_rq) | |
340 | return QUEUE_ORDSEQ_BAR; | |
341 | if (rq == &q->post_flush_rq) | |
342 | return QUEUE_ORDSEQ_POSTFLUSH; | |
1da177e4 | 343 | |
bc90ba09 TH |
344 | /* |
345 | * !fs requests don't need to follow barrier ordering. Always | |
346 | * put them at the front. This fixes the following deadlock. | |
347 | * | |
348 | * http://thread.gmane.org/gmane.linux.kernel/537473 | |
349 | */ | |
350 | if (!blk_fs_request(rq)) | |
351 | return QUEUE_ORDSEQ_DRAIN; | |
352 | ||
4aff5e23 JA |
353 | if ((rq->cmd_flags & REQ_ORDERED_COLOR) == |
354 | (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR)) | |
797e7dbb TH |
355 | return QUEUE_ORDSEQ_DRAIN; |
356 | else | |
357 | return QUEUE_ORDSEQ_DONE; | |
1da177e4 LT |
358 | } |
359 | ||
165125e1 | 360 | void blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error) |
1da177e4 | 361 | { |
797e7dbb TH |
362 | struct request *rq; |
363 | int uptodate; | |
1da177e4 | 364 | |
797e7dbb TH |
365 | if (error && !q->orderr) |
366 | q->orderr = error; | |
1da177e4 | 367 | |
797e7dbb TH |
368 | BUG_ON(q->ordseq & seq); |
369 | q->ordseq |= seq; | |
1da177e4 | 370 | |
797e7dbb TH |
371 | if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE) |
372 | return; | |
1da177e4 LT |
373 | |
374 | /* | |
797e7dbb | 375 | * Okay, sequence complete. |
1da177e4 | 376 | */ |
797e7dbb TH |
377 | rq = q->orig_bar_rq; |
378 | uptodate = q->orderr ? q->orderr : 1; | |
1da177e4 | 379 | |
797e7dbb | 380 | q->ordseq = 0; |
1da177e4 | 381 | |
797e7dbb TH |
382 | end_that_request_first(rq, uptodate, rq->hard_nr_sectors); |
383 | end_that_request_last(rq, uptodate); | |
1da177e4 LT |
384 | } |
385 | ||
797e7dbb | 386 | static void pre_flush_end_io(struct request *rq, int error) |
1da177e4 | 387 | { |
797e7dbb TH |
388 | elv_completed_request(rq->q, rq); |
389 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error); | |
390 | } | |
1da177e4 | 391 | |
797e7dbb TH |
392 | static void bar_end_io(struct request *rq, int error) |
393 | { | |
394 | elv_completed_request(rq->q, rq); | |
395 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error); | |
396 | } | |
1da177e4 | 397 | |
797e7dbb TH |
398 | static void post_flush_end_io(struct request *rq, int error) |
399 | { | |
400 | elv_completed_request(rq->q, rq); | |
401 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error); | |
402 | } | |
1da177e4 | 403 | |
165125e1 | 404 | static void queue_flush(struct request_queue *q, unsigned which) |
797e7dbb TH |
405 | { |
406 | struct request *rq; | |
407 | rq_end_io_fn *end_io; | |
1da177e4 | 408 | |
797e7dbb TH |
409 | if (which == QUEUE_ORDERED_PREFLUSH) { |
410 | rq = &q->pre_flush_rq; | |
411 | end_io = pre_flush_end_io; | |
412 | } else { | |
413 | rq = &q->post_flush_rq; | |
414 | end_io = post_flush_end_io; | |
1da177e4 | 415 | } |
797e7dbb | 416 | |
4aff5e23 | 417 | rq->cmd_flags = REQ_HARDBARRIER; |
797e7dbb | 418 | rq_init(q, rq); |
797e7dbb | 419 | rq->elevator_private = NULL; |
c00895ab | 420 | rq->elevator_private2 = NULL; |
797e7dbb | 421 | rq->rq_disk = q->bar_rq.rq_disk; |
797e7dbb TH |
422 | rq->end_io = end_io; |
423 | q->prepare_flush_fn(q, rq); | |
424 | ||
30e9656c | 425 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); |
1da177e4 LT |
426 | } |
427 | ||
165125e1 | 428 | static inline struct request *start_ordered(struct request_queue *q, |
797e7dbb | 429 | struct request *rq) |
1da177e4 | 430 | { |
797e7dbb TH |
431 | q->bi_size = 0; |
432 | q->orderr = 0; | |
433 | q->ordered = q->next_ordered; | |
434 | q->ordseq |= QUEUE_ORDSEQ_STARTED; | |
435 | ||
436 | /* | |
437 | * Prep proxy barrier request. | |
438 | */ | |
439 | blkdev_dequeue_request(rq); | |
440 | q->orig_bar_rq = rq; | |
441 | rq = &q->bar_rq; | |
4aff5e23 | 442 | rq->cmd_flags = 0; |
797e7dbb | 443 | rq_init(q, rq); |
4aff5e23 JA |
444 | if (bio_data_dir(q->orig_bar_rq->bio) == WRITE) |
445 | rq->cmd_flags |= REQ_RW; | |
446 | rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0; | |
797e7dbb | 447 | rq->elevator_private = NULL; |
c00895ab | 448 | rq->elevator_private2 = NULL; |
797e7dbb TH |
449 | init_request_from_bio(rq, q->orig_bar_rq->bio); |
450 | rq->end_io = bar_end_io; | |
451 | ||
452 | /* | |
453 | * Queue ordered sequence. As we stack them at the head, we | |
454 | * need to queue in reverse order. Note that we rely on that | |
455 | * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs | |
456 | * request gets inbetween ordered sequence. | |
457 | */ | |
458 | if (q->ordered & QUEUE_ORDERED_POSTFLUSH) | |
459 | queue_flush(q, QUEUE_ORDERED_POSTFLUSH); | |
460 | else | |
461 | q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH; | |
462 | ||
30e9656c | 463 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); |
797e7dbb TH |
464 | |
465 | if (q->ordered & QUEUE_ORDERED_PREFLUSH) { | |
466 | queue_flush(q, QUEUE_ORDERED_PREFLUSH); | |
467 | rq = &q->pre_flush_rq; | |
468 | } else | |
469 | q->ordseq |= QUEUE_ORDSEQ_PREFLUSH; | |
1da177e4 | 470 | |
797e7dbb TH |
471 | if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0) |
472 | q->ordseq |= QUEUE_ORDSEQ_DRAIN; | |
473 | else | |
474 | rq = NULL; | |
475 | ||
476 | return rq; | |
1da177e4 LT |
477 | } |
478 | ||
165125e1 | 479 | int blk_do_ordered(struct request_queue *q, struct request **rqp) |
1da177e4 | 480 | { |
9a7a67af | 481 | struct request *rq = *rqp; |
797e7dbb | 482 | int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq); |
1da177e4 | 483 | |
797e7dbb TH |
484 | if (!q->ordseq) { |
485 | if (!is_barrier) | |
486 | return 1; | |
1da177e4 | 487 | |
797e7dbb TH |
488 | if (q->next_ordered != QUEUE_ORDERED_NONE) { |
489 | *rqp = start_ordered(q, rq); | |
490 | return 1; | |
491 | } else { | |
492 | /* | |
493 | * This can happen when the queue switches to | |
494 | * ORDERED_NONE while this request is on it. | |
495 | */ | |
496 | blkdev_dequeue_request(rq); | |
497 | end_that_request_first(rq, -EOPNOTSUPP, | |
498 | rq->hard_nr_sectors); | |
499 | end_that_request_last(rq, -EOPNOTSUPP); | |
500 | *rqp = NULL; | |
501 | return 0; | |
502 | } | |
503 | } | |
1da177e4 | 504 | |
9a7a67af JA |
505 | /* |
506 | * Ordered sequence in progress | |
507 | */ | |
508 | ||
509 | /* Special requests are not subject to ordering rules. */ | |
510 | if (!blk_fs_request(rq) && | |
511 | rq != &q->pre_flush_rq && rq != &q->post_flush_rq) | |
512 | return 1; | |
513 | ||
797e7dbb | 514 | if (q->ordered & QUEUE_ORDERED_TAG) { |
9a7a67af | 515 | /* Ordered by tag. Blocking the next barrier is enough. */ |
797e7dbb TH |
516 | if (is_barrier && rq != &q->bar_rq) |
517 | *rqp = NULL; | |
9a7a67af JA |
518 | } else { |
519 | /* Ordered by draining. Wait for turn. */ | |
520 | WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q)); | |
521 | if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q)) | |
522 | *rqp = NULL; | |
1da177e4 LT |
523 | } |
524 | ||
525 | return 1; | |
526 | } | |
527 | ||
797e7dbb | 528 | static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error) |
1da177e4 | 529 | { |
165125e1 | 530 | struct request_queue *q = bio->bi_private; |
797e7dbb TH |
531 | |
532 | /* | |
533 | * This is dry run, restore bio_sector and size. We'll finish | |
534 | * this request again with the original bi_end_io after an | |
535 | * error occurs or post flush is complete. | |
536 | */ | |
537 | q->bi_size += bytes; | |
538 | ||
539 | if (bio->bi_size) | |
540 | return 1; | |
541 | ||
797e7dbb TH |
542 | /* Reset bio */ |
543 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
544 | bio->bi_size = q->bi_size; | |
545 | bio->bi_sector -= (q->bi_size >> 9); | |
546 | q->bi_size = 0; | |
547 | ||
548 | return 0; | |
1da177e4 | 549 | } |
1da177e4 | 550 | |
1ea25ecb JA |
551 | static int ordered_bio_endio(struct request *rq, struct bio *bio, |
552 | unsigned int nbytes, int error) | |
1da177e4 | 553 | { |
165125e1 | 554 | struct request_queue *q = rq->q; |
797e7dbb TH |
555 | bio_end_io_t *endio; |
556 | void *private; | |
557 | ||
558 | if (&q->bar_rq != rq) | |
559 | return 0; | |
560 | ||
561 | /* | |
562 | * Okay, this is the barrier request in progress, dry finish it. | |
563 | */ | |
564 | if (error && !q->orderr) | |
565 | q->orderr = error; | |
566 | ||
567 | endio = bio->bi_end_io; | |
568 | private = bio->bi_private; | |
569 | bio->bi_end_io = flush_dry_bio_endio; | |
570 | bio->bi_private = q; | |
571 | ||
572 | bio_endio(bio, nbytes, error); | |
573 | ||
574 | bio->bi_end_io = endio; | |
575 | bio->bi_private = private; | |
576 | ||
577 | return 1; | |
1da177e4 | 578 | } |
1da177e4 LT |
579 | |
580 | /** | |
581 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
582 | * @q: the request queue for the device | |
583 | * @dma_addr: bus address limit | |
584 | * | |
585 | * Description: | |
586 | * Different hardware can have different requirements as to what pages | |
587 | * it can do I/O directly to. A low level driver can call | |
588 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
5ee1af9f | 589 | * buffers for doing I/O to pages residing above @page. |
1da177e4 | 590 | **/ |
165125e1 | 591 | void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr) |
1da177e4 LT |
592 | { |
593 | unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT; | |
5ee1af9f AK |
594 | int dma = 0; |
595 | ||
596 | q->bounce_gfp = GFP_NOIO; | |
597 | #if BITS_PER_LONG == 64 | |
598 | /* Assume anything <= 4GB can be handled by IOMMU. | |
599 | Actually some IOMMUs can handle everything, but I don't | |
600 | know of a way to test this here. */ | |
8269730b | 601 | if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) |
5ee1af9f AK |
602 | dma = 1; |
603 | q->bounce_pfn = max_low_pfn; | |
604 | #else | |
605 | if (bounce_pfn < blk_max_low_pfn) | |
606 | dma = 1; | |
607 | q->bounce_pfn = bounce_pfn; | |
608 | #endif | |
609 | if (dma) { | |
1da177e4 LT |
610 | init_emergency_isa_pool(); |
611 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
5ee1af9f AK |
612 | q->bounce_pfn = bounce_pfn; |
613 | } | |
1da177e4 LT |
614 | } |
615 | ||
616 | EXPORT_SYMBOL(blk_queue_bounce_limit); | |
617 | ||
618 | /** | |
619 | * blk_queue_max_sectors - set max sectors for a request for this queue | |
620 | * @q: the request queue for the device | |
621 | * @max_sectors: max sectors in the usual 512b unit | |
622 | * | |
623 | * Description: | |
624 | * Enables a low level driver to set an upper limit on the size of | |
625 | * received requests. | |
626 | **/ | |
165125e1 | 627 | void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors) |
1da177e4 LT |
628 | { |
629 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { | |
630 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
631 | printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors); | |
632 | } | |
633 | ||
defd94b7 MC |
634 | if (BLK_DEF_MAX_SECTORS > max_sectors) |
635 | q->max_hw_sectors = q->max_sectors = max_sectors; | |
636 | else { | |
637 | q->max_sectors = BLK_DEF_MAX_SECTORS; | |
638 | q->max_hw_sectors = max_sectors; | |
639 | } | |
1da177e4 LT |
640 | } |
641 | ||
642 | EXPORT_SYMBOL(blk_queue_max_sectors); | |
643 | ||
644 | /** | |
645 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue | |
646 | * @q: the request queue for the device | |
647 | * @max_segments: max number of segments | |
648 | * | |
649 | * Description: | |
650 | * Enables a low level driver to set an upper limit on the number of | |
651 | * physical data segments in a request. This would be the largest sized | |
652 | * scatter list the driver could handle. | |
653 | **/ | |
165125e1 JA |
654 | void blk_queue_max_phys_segments(struct request_queue *q, |
655 | unsigned short max_segments) | |
1da177e4 LT |
656 | { |
657 | if (!max_segments) { | |
658 | max_segments = 1; | |
659 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
660 | } | |
661 | ||
662 | q->max_phys_segments = max_segments; | |
663 | } | |
664 | ||
665 | EXPORT_SYMBOL(blk_queue_max_phys_segments); | |
666 | ||
667 | /** | |
668 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue | |
669 | * @q: the request queue for the device | |
670 | * @max_segments: max number of segments | |
671 | * | |
672 | * Description: | |
673 | * Enables a low level driver to set an upper limit on the number of | |
674 | * hw data segments in a request. This would be the largest number of | |
675 | * address/length pairs the host adapter can actually give as once | |
676 | * to the device. | |
677 | **/ | |
165125e1 JA |
678 | void blk_queue_max_hw_segments(struct request_queue *q, |
679 | unsigned short max_segments) | |
1da177e4 LT |
680 | { |
681 | if (!max_segments) { | |
682 | max_segments = 1; | |
683 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
684 | } | |
685 | ||
686 | q->max_hw_segments = max_segments; | |
687 | } | |
688 | ||
689 | EXPORT_SYMBOL(blk_queue_max_hw_segments); | |
690 | ||
691 | /** | |
692 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
693 | * @q: the request queue for the device | |
694 | * @max_size: max size of segment in bytes | |
695 | * | |
696 | * Description: | |
697 | * Enables a low level driver to set an upper limit on the size of a | |
698 | * coalesced segment | |
699 | **/ | |
165125e1 | 700 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) |
1da177e4 LT |
701 | { |
702 | if (max_size < PAGE_CACHE_SIZE) { | |
703 | max_size = PAGE_CACHE_SIZE; | |
704 | printk("%s: set to minimum %d\n", __FUNCTION__, max_size); | |
705 | } | |
706 | ||
707 | q->max_segment_size = max_size; | |
708 | } | |
709 | ||
710 | EXPORT_SYMBOL(blk_queue_max_segment_size); | |
711 | ||
712 | /** | |
713 | * blk_queue_hardsect_size - set hardware sector size for the queue | |
714 | * @q: the request queue for the device | |
715 | * @size: the hardware sector size, in bytes | |
716 | * | |
717 | * Description: | |
718 | * This should typically be set to the lowest possible sector size | |
719 | * that the hardware can operate on (possible without reverting to | |
720 | * even internal read-modify-write operations). Usually the default | |
721 | * of 512 covers most hardware. | |
722 | **/ | |
165125e1 | 723 | void blk_queue_hardsect_size(struct request_queue *q, unsigned short size) |
1da177e4 LT |
724 | { |
725 | q->hardsect_size = size; | |
726 | } | |
727 | ||
728 | EXPORT_SYMBOL(blk_queue_hardsect_size); | |
729 | ||
730 | /* | |
731 | * Returns the minimum that is _not_ zero, unless both are zero. | |
732 | */ | |
733 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
734 | ||
735 | /** | |
736 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
737 | * @t: the stacking driver (top) | |
738 | * @b: the underlying device (bottom) | |
739 | **/ | |
165125e1 | 740 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) |
1da177e4 LT |
741 | { |
742 | /* zero is "infinity" */ | |
defd94b7 MC |
743 | t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors); |
744 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors); | |
1da177e4 LT |
745 | |
746 | t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments); | |
747 | t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments); | |
748 | t->max_segment_size = min(t->max_segment_size,b->max_segment_size); | |
749 | t->hardsect_size = max(t->hardsect_size,b->hardsect_size); | |
89e5c8b5 N |
750 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) |
751 | clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags); | |
1da177e4 LT |
752 | } |
753 | ||
754 | EXPORT_SYMBOL(blk_queue_stack_limits); | |
755 | ||
756 | /** | |
757 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
758 | * @q: the request queue for the device | |
759 | * @mask: the memory boundary mask | |
760 | **/ | |
165125e1 | 761 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) |
1da177e4 LT |
762 | { |
763 | if (mask < PAGE_CACHE_SIZE - 1) { | |
764 | mask = PAGE_CACHE_SIZE - 1; | |
765 | printk("%s: set to minimum %lx\n", __FUNCTION__, mask); | |
766 | } | |
767 | ||
768 | q->seg_boundary_mask = mask; | |
769 | } | |
770 | ||
771 | EXPORT_SYMBOL(blk_queue_segment_boundary); | |
772 | ||
773 | /** | |
774 | * blk_queue_dma_alignment - set dma length and memory alignment | |
775 | * @q: the request queue for the device | |
776 | * @mask: alignment mask | |
777 | * | |
778 | * description: | |
779 | * set required memory and length aligment for direct dma transactions. | |
780 | * this is used when buiding direct io requests for the queue. | |
781 | * | |
782 | **/ | |
165125e1 | 783 | void blk_queue_dma_alignment(struct request_queue *q, int mask) |
1da177e4 LT |
784 | { |
785 | q->dma_alignment = mask; | |
786 | } | |
787 | ||
788 | EXPORT_SYMBOL(blk_queue_dma_alignment); | |
789 | ||
790 | /** | |
791 | * blk_queue_find_tag - find a request by its tag and queue | |
1da177e4 LT |
792 | * @q: The request queue for the device |
793 | * @tag: The tag of the request | |
794 | * | |
795 | * Notes: | |
796 | * Should be used when a device returns a tag and you want to match | |
797 | * it with a request. | |
798 | * | |
799 | * no locks need be held. | |
800 | **/ | |
165125e1 | 801 | struct request *blk_queue_find_tag(struct request_queue *q, int tag) |
1da177e4 | 802 | { |
f583f492 | 803 | return blk_map_queue_find_tag(q->queue_tags, tag); |
1da177e4 LT |
804 | } |
805 | ||
806 | EXPORT_SYMBOL(blk_queue_find_tag); | |
807 | ||
808 | /** | |
492dfb48 JB |
809 | * __blk_free_tags - release a given set of tag maintenance info |
810 | * @bqt: the tag map to free | |
1da177e4 | 811 | * |
492dfb48 JB |
812 | * Tries to free the specified @bqt@. Returns true if it was |
813 | * actually freed and false if there are still references using it | |
814 | */ | |
815 | static int __blk_free_tags(struct blk_queue_tag *bqt) | |
1da177e4 | 816 | { |
492dfb48 | 817 | int retval; |
1da177e4 | 818 | |
492dfb48 JB |
819 | retval = atomic_dec_and_test(&bqt->refcnt); |
820 | if (retval) { | |
1da177e4 LT |
821 | BUG_ON(bqt->busy); |
822 | BUG_ON(!list_empty(&bqt->busy_list)); | |
823 | ||
824 | kfree(bqt->tag_index); | |
825 | bqt->tag_index = NULL; | |
826 | ||
827 | kfree(bqt->tag_map); | |
828 | bqt->tag_map = NULL; | |
829 | ||
830 | kfree(bqt); | |
492dfb48 | 831 | |
1da177e4 LT |
832 | } |
833 | ||
492dfb48 JB |
834 | return retval; |
835 | } | |
836 | ||
837 | /** | |
838 | * __blk_queue_free_tags - release tag maintenance info | |
839 | * @q: the request queue for the device | |
840 | * | |
841 | * Notes: | |
842 | * blk_cleanup_queue() will take care of calling this function, if tagging | |
843 | * has been used. So there's no need to call this directly. | |
844 | **/ | |
165125e1 | 845 | static void __blk_queue_free_tags(struct request_queue *q) |
492dfb48 JB |
846 | { |
847 | struct blk_queue_tag *bqt = q->queue_tags; | |
848 | ||
849 | if (!bqt) | |
850 | return; | |
851 | ||
852 | __blk_free_tags(bqt); | |
853 | ||
1da177e4 LT |
854 | q->queue_tags = NULL; |
855 | q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED); | |
856 | } | |
857 | ||
492dfb48 JB |
858 | |
859 | /** | |
860 | * blk_free_tags - release a given set of tag maintenance info | |
861 | * @bqt: the tag map to free | |
862 | * | |
863 | * For externally managed @bqt@ frees the map. Callers of this | |
864 | * function must guarantee to have released all the queues that | |
865 | * might have been using this tag map. | |
866 | */ | |
867 | void blk_free_tags(struct blk_queue_tag *bqt) | |
868 | { | |
869 | if (unlikely(!__blk_free_tags(bqt))) | |
870 | BUG(); | |
871 | } | |
872 | EXPORT_SYMBOL(blk_free_tags); | |
873 | ||
1da177e4 LT |
874 | /** |
875 | * blk_queue_free_tags - release tag maintenance info | |
876 | * @q: the request queue for the device | |
877 | * | |
878 | * Notes: | |
879 | * This is used to disabled tagged queuing to a device, yet leave | |
880 | * queue in function. | |
881 | **/ | |
165125e1 | 882 | void blk_queue_free_tags(struct request_queue *q) |
1da177e4 LT |
883 | { |
884 | clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
885 | } | |
886 | ||
887 | EXPORT_SYMBOL(blk_queue_free_tags); | |
888 | ||
889 | static int | |
165125e1 | 890 | init_tag_map(struct request_queue *q, struct blk_queue_tag *tags, int depth) |
1da177e4 | 891 | { |
1da177e4 LT |
892 | struct request **tag_index; |
893 | unsigned long *tag_map; | |
fa72b903 | 894 | int nr_ulongs; |
1da177e4 | 895 | |
492dfb48 | 896 | if (q && depth > q->nr_requests * 2) { |
1da177e4 LT |
897 | depth = q->nr_requests * 2; |
898 | printk(KERN_ERR "%s: adjusted depth to %d\n", | |
899 | __FUNCTION__, depth); | |
900 | } | |
901 | ||
f68110fc | 902 | tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC); |
1da177e4 LT |
903 | if (!tag_index) |
904 | goto fail; | |
905 | ||
f7d37d02 | 906 | nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG; |
f68110fc | 907 | tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC); |
1da177e4 LT |
908 | if (!tag_map) |
909 | goto fail; | |
910 | ||
ba025082 | 911 | tags->real_max_depth = depth; |
1da177e4 | 912 | tags->max_depth = depth; |
1da177e4 LT |
913 | tags->tag_index = tag_index; |
914 | tags->tag_map = tag_map; | |
915 | ||
1da177e4 LT |
916 | return 0; |
917 | fail: | |
918 | kfree(tag_index); | |
919 | return -ENOMEM; | |
920 | } | |
921 | ||
492dfb48 JB |
922 | static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q, |
923 | int depth) | |
924 | { | |
925 | struct blk_queue_tag *tags; | |
926 | ||
927 | tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC); | |
928 | if (!tags) | |
929 | goto fail; | |
930 | ||
931 | if (init_tag_map(q, tags, depth)) | |
932 | goto fail; | |
933 | ||
934 | INIT_LIST_HEAD(&tags->busy_list); | |
935 | tags->busy = 0; | |
936 | atomic_set(&tags->refcnt, 1); | |
937 | return tags; | |
938 | fail: | |
939 | kfree(tags); | |
940 | return NULL; | |
941 | } | |
942 | ||
943 | /** | |
944 | * blk_init_tags - initialize the tag info for an external tag map | |
945 | * @depth: the maximum queue depth supported | |
946 | * @tags: the tag to use | |
947 | **/ | |
948 | struct blk_queue_tag *blk_init_tags(int depth) | |
949 | { | |
950 | return __blk_queue_init_tags(NULL, depth); | |
951 | } | |
952 | EXPORT_SYMBOL(blk_init_tags); | |
953 | ||
1da177e4 LT |
954 | /** |
955 | * blk_queue_init_tags - initialize the queue tag info | |
956 | * @q: the request queue for the device | |
957 | * @depth: the maximum queue depth supported | |
958 | * @tags: the tag to use | |
959 | **/ | |
165125e1 | 960 | int blk_queue_init_tags(struct request_queue *q, int depth, |
1da177e4 LT |
961 | struct blk_queue_tag *tags) |
962 | { | |
963 | int rc; | |
964 | ||
965 | BUG_ON(tags && q->queue_tags && tags != q->queue_tags); | |
966 | ||
967 | if (!tags && !q->queue_tags) { | |
492dfb48 | 968 | tags = __blk_queue_init_tags(q, depth); |
1da177e4 | 969 | |
492dfb48 | 970 | if (!tags) |
1da177e4 | 971 | goto fail; |
1da177e4 LT |
972 | } else if (q->queue_tags) { |
973 | if ((rc = blk_queue_resize_tags(q, depth))) | |
974 | return rc; | |
975 | set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
976 | return 0; | |
977 | } else | |
978 | atomic_inc(&tags->refcnt); | |
979 | ||
980 | /* | |
981 | * assign it, all done | |
982 | */ | |
983 | q->queue_tags = tags; | |
984 | q->queue_flags |= (1 << QUEUE_FLAG_QUEUED); | |
985 | return 0; | |
986 | fail: | |
987 | kfree(tags); | |
988 | return -ENOMEM; | |
989 | } | |
990 | ||
991 | EXPORT_SYMBOL(blk_queue_init_tags); | |
992 | ||
993 | /** | |
994 | * blk_queue_resize_tags - change the queueing depth | |
995 | * @q: the request queue for the device | |
996 | * @new_depth: the new max command queueing depth | |
997 | * | |
998 | * Notes: | |
999 | * Must be called with the queue lock held. | |
1000 | **/ | |
165125e1 | 1001 | int blk_queue_resize_tags(struct request_queue *q, int new_depth) |
1da177e4 LT |
1002 | { |
1003 | struct blk_queue_tag *bqt = q->queue_tags; | |
1004 | struct request **tag_index; | |
1005 | unsigned long *tag_map; | |
fa72b903 | 1006 | int max_depth, nr_ulongs; |
1da177e4 LT |
1007 | |
1008 | if (!bqt) | |
1009 | return -ENXIO; | |
1010 | ||
ba025082 TH |
1011 | /* |
1012 | * if we already have large enough real_max_depth. just | |
1013 | * adjust max_depth. *NOTE* as requests with tag value | |
1014 | * between new_depth and real_max_depth can be in-flight, tag | |
1015 | * map can not be shrunk blindly here. | |
1016 | */ | |
1017 | if (new_depth <= bqt->real_max_depth) { | |
1018 | bqt->max_depth = new_depth; | |
1019 | return 0; | |
1020 | } | |
1021 | ||
492dfb48 JB |
1022 | /* |
1023 | * Currently cannot replace a shared tag map with a new | |
1024 | * one, so error out if this is the case | |
1025 | */ | |
1026 | if (atomic_read(&bqt->refcnt) != 1) | |
1027 | return -EBUSY; | |
1028 | ||
1da177e4 LT |
1029 | /* |
1030 | * save the old state info, so we can copy it back | |
1031 | */ | |
1032 | tag_index = bqt->tag_index; | |
1033 | tag_map = bqt->tag_map; | |
ba025082 | 1034 | max_depth = bqt->real_max_depth; |
1da177e4 LT |
1035 | |
1036 | if (init_tag_map(q, bqt, new_depth)) | |
1037 | return -ENOMEM; | |
1038 | ||
1039 | memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); | |
f7d37d02 | 1040 | nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG; |
fa72b903 | 1041 | memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long)); |
1da177e4 LT |
1042 | |
1043 | kfree(tag_index); | |
1044 | kfree(tag_map); | |
1045 | return 0; | |
1046 | } | |
1047 | ||
1048 | EXPORT_SYMBOL(blk_queue_resize_tags); | |
1049 | ||
1050 | /** | |
1051 | * blk_queue_end_tag - end tag operations for a request | |
1052 | * @q: the request queue for the device | |
1053 | * @rq: the request that has completed | |
1054 | * | |
1055 | * Description: | |
1056 | * Typically called when end_that_request_first() returns 0, meaning | |
1057 | * all transfers have been done for a request. It's important to call | |
1058 | * this function before end_that_request_last(), as that will put the | |
1059 | * request back on the free list thus corrupting the internal tag list. | |
1060 | * | |
1061 | * Notes: | |
1062 | * queue lock must be held. | |
1063 | **/ | |
165125e1 | 1064 | void blk_queue_end_tag(struct request_queue *q, struct request *rq) |
1da177e4 LT |
1065 | { |
1066 | struct blk_queue_tag *bqt = q->queue_tags; | |
1067 | int tag = rq->tag; | |
1068 | ||
1069 | BUG_ON(tag == -1); | |
1070 | ||
ba025082 | 1071 | if (unlikely(tag >= bqt->real_max_depth)) |
040c928c TH |
1072 | /* |
1073 | * This can happen after tag depth has been reduced. | |
1074 | * FIXME: how about a warning or info message here? | |
1075 | */ | |
1da177e4 LT |
1076 | return; |
1077 | ||
1078 | if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) { | |
040c928c TH |
1079 | printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n", |
1080 | __FUNCTION__, tag); | |
1da177e4 LT |
1081 | return; |
1082 | } | |
1083 | ||
1084 | list_del_init(&rq->queuelist); | |
4aff5e23 | 1085 | rq->cmd_flags &= ~REQ_QUEUED; |
1da177e4 LT |
1086 | rq->tag = -1; |
1087 | ||
1088 | if (unlikely(bqt->tag_index[tag] == NULL)) | |
040c928c TH |
1089 | printk(KERN_ERR "%s: tag %d is missing\n", |
1090 | __FUNCTION__, tag); | |
1da177e4 LT |
1091 | |
1092 | bqt->tag_index[tag] = NULL; | |
1093 | bqt->busy--; | |
1094 | } | |
1095 | ||
1096 | EXPORT_SYMBOL(blk_queue_end_tag); | |
1097 | ||
1098 | /** | |
1099 | * blk_queue_start_tag - find a free tag and assign it | |
1100 | * @q: the request queue for the device | |
1101 | * @rq: the block request that needs tagging | |
1102 | * | |
1103 | * Description: | |
1104 | * This can either be used as a stand-alone helper, or possibly be | |
1105 | * assigned as the queue &prep_rq_fn (in which case &struct request | |
1106 | * automagically gets a tag assigned). Note that this function | |
1107 | * assumes that any type of request can be queued! if this is not | |
1108 | * true for your device, you must check the request type before | |
1109 | * calling this function. The request will also be removed from | |
1110 | * the request queue, so it's the drivers responsibility to readd | |
1111 | * it if it should need to be restarted for some reason. | |
1112 | * | |
1113 | * Notes: | |
1114 | * queue lock must be held. | |
1115 | **/ | |
165125e1 | 1116 | int blk_queue_start_tag(struct request_queue *q, struct request *rq) |
1da177e4 LT |
1117 | { |
1118 | struct blk_queue_tag *bqt = q->queue_tags; | |
2bf0fdad | 1119 | int tag; |
1da177e4 | 1120 | |
4aff5e23 | 1121 | if (unlikely((rq->cmd_flags & REQ_QUEUED))) { |
1da177e4 | 1122 | printk(KERN_ERR |
040c928c TH |
1123 | "%s: request %p for device [%s] already tagged %d", |
1124 | __FUNCTION__, rq, | |
1125 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); | |
1da177e4 LT |
1126 | BUG(); |
1127 | } | |
1128 | ||
059af497 JA |
1129 | /* |
1130 | * Protect against shared tag maps, as we may not have exclusive | |
1131 | * access to the tag map. | |
1132 | */ | |
1133 | do { | |
1134 | tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth); | |
1135 | if (tag >= bqt->max_depth) | |
1136 | return 1; | |
1da177e4 | 1137 | |
059af497 | 1138 | } while (test_and_set_bit(tag, bqt->tag_map)); |
1da177e4 | 1139 | |
4aff5e23 | 1140 | rq->cmd_flags |= REQ_QUEUED; |
1da177e4 LT |
1141 | rq->tag = tag; |
1142 | bqt->tag_index[tag] = rq; | |
1143 | blkdev_dequeue_request(rq); | |
1144 | list_add(&rq->queuelist, &bqt->busy_list); | |
1145 | bqt->busy++; | |
1146 | return 0; | |
1147 | } | |
1148 | ||
1149 | EXPORT_SYMBOL(blk_queue_start_tag); | |
1150 | ||
1151 | /** | |
1152 | * blk_queue_invalidate_tags - invalidate all pending tags | |
1153 | * @q: the request queue for the device | |
1154 | * | |
1155 | * Description: | |
1156 | * Hardware conditions may dictate a need to stop all pending requests. | |
1157 | * In this case, we will safely clear the block side of the tag queue and | |
1158 | * readd all requests to the request queue in the right order. | |
1159 | * | |
1160 | * Notes: | |
1161 | * queue lock must be held. | |
1162 | **/ | |
165125e1 | 1163 | void blk_queue_invalidate_tags(struct request_queue *q) |
1da177e4 LT |
1164 | { |
1165 | struct blk_queue_tag *bqt = q->queue_tags; | |
1166 | struct list_head *tmp, *n; | |
1167 | struct request *rq; | |
1168 | ||
1169 | list_for_each_safe(tmp, n, &bqt->busy_list) { | |
1170 | rq = list_entry_rq(tmp); | |
1171 | ||
1172 | if (rq->tag == -1) { | |
040c928c TH |
1173 | printk(KERN_ERR |
1174 | "%s: bad tag found on list\n", __FUNCTION__); | |
1da177e4 | 1175 | list_del_init(&rq->queuelist); |
4aff5e23 | 1176 | rq->cmd_flags &= ~REQ_QUEUED; |
1da177e4 LT |
1177 | } else |
1178 | blk_queue_end_tag(q, rq); | |
1179 | ||
4aff5e23 | 1180 | rq->cmd_flags &= ~REQ_STARTED; |
1da177e4 LT |
1181 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); |
1182 | } | |
1183 | } | |
1184 | ||
1185 | EXPORT_SYMBOL(blk_queue_invalidate_tags); | |
1186 | ||
1da177e4 LT |
1187 | void blk_dump_rq_flags(struct request *rq, char *msg) |
1188 | { | |
1189 | int bit; | |
1190 | ||
4aff5e23 JA |
1191 | printk("%s: dev %s: type=%x, flags=%x\n", msg, |
1192 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, | |
1193 | rq->cmd_flags); | |
1da177e4 LT |
1194 | |
1195 | printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector, | |
1196 | rq->nr_sectors, | |
1197 | rq->current_nr_sectors); | |
1198 | printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len); | |
1199 | ||
4aff5e23 | 1200 | if (blk_pc_request(rq)) { |
1da177e4 LT |
1201 | printk("cdb: "); |
1202 | for (bit = 0; bit < sizeof(rq->cmd); bit++) | |
1203 | printk("%02x ", rq->cmd[bit]); | |
1204 | printk("\n"); | |
1205 | } | |
1206 | } | |
1207 | ||
1208 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
1209 | ||
165125e1 | 1210 | void blk_recount_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
1211 | { |
1212 | struct bio_vec *bv, *bvprv = NULL; | |
1213 | int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster; | |
1214 | int high, highprv = 1; | |
1215 | ||
1216 | if (unlikely(!bio->bi_io_vec)) | |
1217 | return; | |
1218 | ||
1219 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1220 | hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0; | |
1221 | bio_for_each_segment(bv, bio, i) { | |
1222 | /* | |
1223 | * the trick here is making sure that a high page is never | |
1224 | * considered part of another segment, since that might | |
1225 | * change with the bounce page. | |
1226 | */ | |
f772b3d9 | 1227 | high = page_to_pfn(bv->bv_page) > q->bounce_pfn; |
1da177e4 LT |
1228 | if (high || highprv) |
1229 | goto new_hw_segment; | |
1230 | if (cluster) { | |
1231 | if (seg_size + bv->bv_len > q->max_segment_size) | |
1232 | goto new_segment; | |
1233 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv)) | |
1234 | goto new_segment; | |
1235 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv)) | |
1236 | goto new_segment; | |
1237 | if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) | |
1238 | goto new_hw_segment; | |
1239 | ||
1240 | seg_size += bv->bv_len; | |
1241 | hw_seg_size += bv->bv_len; | |
1242 | bvprv = bv; | |
1243 | continue; | |
1244 | } | |
1245 | new_segment: | |
1246 | if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) && | |
1247 | !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) { | |
1248 | hw_seg_size += bv->bv_len; | |
1249 | } else { | |
1250 | new_hw_segment: | |
1251 | if (hw_seg_size > bio->bi_hw_front_size) | |
1252 | bio->bi_hw_front_size = hw_seg_size; | |
1253 | hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len; | |
1254 | nr_hw_segs++; | |
1255 | } | |
1256 | ||
1257 | nr_phys_segs++; | |
1258 | bvprv = bv; | |
1259 | seg_size = bv->bv_len; | |
1260 | highprv = high; | |
1261 | } | |
1262 | if (hw_seg_size > bio->bi_hw_back_size) | |
1263 | bio->bi_hw_back_size = hw_seg_size; | |
1264 | if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size) | |
1265 | bio->bi_hw_front_size = hw_seg_size; | |
1266 | bio->bi_phys_segments = nr_phys_segs; | |
1267 | bio->bi_hw_segments = nr_hw_segs; | |
1268 | bio->bi_flags |= (1 << BIO_SEG_VALID); | |
1269 | } | |
387bb173 | 1270 | EXPORT_SYMBOL(blk_recount_segments); |
1da177e4 | 1271 | |
165125e1 | 1272 | static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio, |
1da177e4 LT |
1273 | struct bio *nxt) |
1274 | { | |
1275 | if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) | |
1276 | return 0; | |
1277 | ||
1278 | if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt))) | |
1279 | return 0; | |
1280 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) | |
1281 | return 0; | |
1282 | ||
1283 | /* | |
1284 | * bio and nxt are contigous in memory, check if the queue allows | |
1285 | * these two to be merged into one | |
1286 | */ | |
1287 | if (BIO_SEG_BOUNDARY(q, bio, nxt)) | |
1288 | return 1; | |
1289 | ||
1290 | return 0; | |
1291 | } | |
1292 | ||
165125e1 | 1293 | static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio, |
1da177e4 LT |
1294 | struct bio *nxt) |
1295 | { | |
1296 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1297 | blk_recount_segments(q, bio); | |
1298 | if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID))) | |
1299 | blk_recount_segments(q, nxt); | |
1300 | if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) || | |
32eef964 | 1301 | BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size)) |
1da177e4 | 1302 | return 0; |
32eef964 | 1303 | if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size) |
1da177e4 LT |
1304 | return 0; |
1305 | ||
1306 | return 1; | |
1307 | } | |
1308 | ||
1da177e4 LT |
1309 | /* |
1310 | * map a request to scatterlist, return number of sg entries setup. Caller | |
1311 | * must make sure sg can hold rq->nr_phys_segments entries | |
1312 | */ | |
165125e1 JA |
1313 | int blk_rq_map_sg(struct request_queue *q, struct request *rq, |
1314 | struct scatterlist *sg) | |
1da177e4 LT |
1315 | { |
1316 | struct bio_vec *bvec, *bvprv; | |
1317 | struct bio *bio; | |
1318 | int nsegs, i, cluster; | |
1319 | ||
1320 | nsegs = 0; | |
1321 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1322 | ||
1323 | /* | |
1324 | * for each bio in rq | |
1325 | */ | |
1326 | bvprv = NULL; | |
1327 | rq_for_each_bio(bio, rq) { | |
1328 | /* | |
1329 | * for each segment in bio | |
1330 | */ | |
1331 | bio_for_each_segment(bvec, bio, i) { | |
1332 | int nbytes = bvec->bv_len; | |
1333 | ||
1334 | if (bvprv && cluster) { | |
1335 | if (sg[nsegs - 1].length + nbytes > q->max_segment_size) | |
1336 | goto new_segment; | |
1337 | ||
1338 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) | |
1339 | goto new_segment; | |
1340 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) | |
1341 | goto new_segment; | |
1342 | ||
1343 | sg[nsegs - 1].length += nbytes; | |
1344 | } else { | |
1345 | new_segment: | |
1346 | memset(&sg[nsegs],0,sizeof(struct scatterlist)); | |
1347 | sg[nsegs].page = bvec->bv_page; | |
1348 | sg[nsegs].length = nbytes; | |
1349 | sg[nsegs].offset = bvec->bv_offset; | |
1350 | ||
1351 | nsegs++; | |
1352 | } | |
1353 | bvprv = bvec; | |
1354 | } /* segments in bio */ | |
1355 | } /* bios in rq */ | |
1356 | ||
1357 | return nsegs; | |
1358 | } | |
1359 | ||
1360 | EXPORT_SYMBOL(blk_rq_map_sg); | |
1361 | ||
1362 | /* | |
1363 | * the standard queue merge functions, can be overridden with device | |
1364 | * specific ones if so desired | |
1365 | */ | |
1366 | ||
165125e1 | 1367 | static inline int ll_new_mergeable(struct request_queue *q, |
1da177e4 LT |
1368 | struct request *req, |
1369 | struct bio *bio) | |
1370 | { | |
1371 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1372 | ||
1373 | if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
4aff5e23 | 1374 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1375 | if (req == q->last_merge) |
1376 | q->last_merge = NULL; | |
1377 | return 0; | |
1378 | } | |
1379 | ||
1380 | /* | |
1381 | * A hw segment is just getting larger, bump just the phys | |
1382 | * counter. | |
1383 | */ | |
1384 | req->nr_phys_segments += nr_phys_segs; | |
1385 | return 1; | |
1386 | } | |
1387 | ||
165125e1 | 1388 | static inline int ll_new_hw_segment(struct request_queue *q, |
1da177e4 LT |
1389 | struct request *req, |
1390 | struct bio *bio) | |
1391 | { | |
1392 | int nr_hw_segs = bio_hw_segments(q, bio); | |
1393 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1394 | ||
1395 | if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments | |
1396 | || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
4aff5e23 | 1397 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1398 | if (req == q->last_merge) |
1399 | q->last_merge = NULL; | |
1400 | return 0; | |
1401 | } | |
1402 | ||
1403 | /* | |
1404 | * This will form the start of a new hw segment. Bump both | |
1405 | * counters. | |
1406 | */ | |
1407 | req->nr_hw_segments += nr_hw_segs; | |
1408 | req->nr_phys_segments += nr_phys_segs; | |
1409 | return 1; | |
1410 | } | |
1411 | ||
165125e1 | 1412 | int ll_back_merge_fn(struct request_queue *q, struct request *req, struct bio *bio) |
1da177e4 | 1413 | { |
defd94b7 | 1414 | unsigned short max_sectors; |
1da177e4 LT |
1415 | int len; |
1416 | ||
defd94b7 MC |
1417 | if (unlikely(blk_pc_request(req))) |
1418 | max_sectors = q->max_hw_sectors; | |
1419 | else | |
1420 | max_sectors = q->max_sectors; | |
1421 | ||
1422 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | |
4aff5e23 | 1423 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1424 | if (req == q->last_merge) |
1425 | q->last_merge = NULL; | |
1426 | return 0; | |
1427 | } | |
1428 | if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID))) | |
1429 | blk_recount_segments(q, req->biotail); | |
1430 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1431 | blk_recount_segments(q, bio); | |
1432 | len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size; | |
1433 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) && | |
1434 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1435 | int mergeable = ll_new_mergeable(q, req, bio); | |
1436 | ||
1437 | if (mergeable) { | |
1438 | if (req->nr_hw_segments == 1) | |
1439 | req->bio->bi_hw_front_size = len; | |
1440 | if (bio->bi_hw_segments == 1) | |
1441 | bio->bi_hw_back_size = len; | |
1442 | } | |
1443 | return mergeable; | |
1444 | } | |
1445 | ||
1446 | return ll_new_hw_segment(q, req, bio); | |
1447 | } | |
1aa4f24f | 1448 | EXPORT_SYMBOL(ll_back_merge_fn); |
1da177e4 | 1449 | |
165125e1 | 1450 | static int ll_front_merge_fn(struct request_queue *q, struct request *req, |
1da177e4 LT |
1451 | struct bio *bio) |
1452 | { | |
defd94b7 | 1453 | unsigned short max_sectors; |
1da177e4 LT |
1454 | int len; |
1455 | ||
defd94b7 MC |
1456 | if (unlikely(blk_pc_request(req))) |
1457 | max_sectors = q->max_hw_sectors; | |
1458 | else | |
1459 | max_sectors = q->max_sectors; | |
1460 | ||
1461 | ||
1462 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | |
4aff5e23 | 1463 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1464 | if (req == q->last_merge) |
1465 | q->last_merge = NULL; | |
1466 | return 0; | |
1467 | } | |
1468 | len = bio->bi_hw_back_size + req->bio->bi_hw_front_size; | |
1469 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1470 | blk_recount_segments(q, bio); | |
1471 | if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID))) | |
1472 | blk_recount_segments(q, req->bio); | |
1473 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) && | |
1474 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1475 | int mergeable = ll_new_mergeable(q, req, bio); | |
1476 | ||
1477 | if (mergeable) { | |
1478 | if (bio->bi_hw_segments == 1) | |
1479 | bio->bi_hw_front_size = len; | |
1480 | if (req->nr_hw_segments == 1) | |
1481 | req->biotail->bi_hw_back_size = len; | |
1482 | } | |
1483 | return mergeable; | |
1484 | } | |
1485 | ||
1486 | return ll_new_hw_segment(q, req, bio); | |
1487 | } | |
1488 | ||
165125e1 | 1489 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
1da177e4 LT |
1490 | struct request *next) |
1491 | { | |
dfa1a553 ND |
1492 | int total_phys_segments; |
1493 | int total_hw_segments; | |
1da177e4 LT |
1494 | |
1495 | /* | |
1496 | * First check if the either of the requests are re-queued | |
1497 | * requests. Can't merge them if they are. | |
1498 | */ | |
1499 | if (req->special || next->special) | |
1500 | return 0; | |
1501 | ||
1502 | /* | |
dfa1a553 | 1503 | * Will it become too large? |
1da177e4 LT |
1504 | */ |
1505 | if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) | |
1506 | return 0; | |
1507 | ||
1508 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
1509 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) | |
1510 | total_phys_segments--; | |
1511 | ||
1512 | if (total_phys_segments > q->max_phys_segments) | |
1513 | return 0; | |
1514 | ||
1515 | total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; | |
1516 | if (blk_hw_contig_segment(q, req->biotail, next->bio)) { | |
1517 | int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size; | |
1518 | /* | |
1519 | * propagate the combined length to the end of the requests | |
1520 | */ | |
1521 | if (req->nr_hw_segments == 1) | |
1522 | req->bio->bi_hw_front_size = len; | |
1523 | if (next->nr_hw_segments == 1) | |
1524 | next->biotail->bi_hw_back_size = len; | |
1525 | total_hw_segments--; | |
1526 | } | |
1527 | ||
1528 | if (total_hw_segments > q->max_hw_segments) | |
1529 | return 0; | |
1530 | ||
1531 | /* Merge is OK... */ | |
1532 | req->nr_phys_segments = total_phys_segments; | |
1533 | req->nr_hw_segments = total_hw_segments; | |
1534 | return 1; | |
1535 | } | |
1536 | ||
1537 | /* | |
1538 | * "plug" the device if there are no outstanding requests: this will | |
1539 | * force the transfer to start only after we have put all the requests | |
1540 | * on the list. | |
1541 | * | |
1542 | * This is called with interrupts off and no requests on the queue and | |
1543 | * with the queue lock held. | |
1544 | */ | |
165125e1 | 1545 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
1546 | { |
1547 | WARN_ON(!irqs_disabled()); | |
1548 | ||
1549 | /* | |
1550 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
1551 | * which will restart the queueing | |
1552 | */ | |
7daac490 | 1553 | if (blk_queue_stopped(q)) |
1da177e4 LT |
1554 | return; |
1555 | ||
2056a782 | 1556 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) { |
1da177e4 | 1557 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
2056a782 JA |
1558 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); |
1559 | } | |
1da177e4 LT |
1560 | } |
1561 | ||
1562 | EXPORT_SYMBOL(blk_plug_device); | |
1563 | ||
1564 | /* | |
1565 | * remove the queue from the plugged list, if present. called with | |
1566 | * queue lock held and interrupts disabled. | |
1567 | */ | |
165125e1 | 1568 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
1569 | { |
1570 | WARN_ON(!irqs_disabled()); | |
1571 | ||
1572 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | |
1573 | return 0; | |
1574 | ||
1575 | del_timer(&q->unplug_timer); | |
1576 | return 1; | |
1577 | } | |
1578 | ||
1579 | EXPORT_SYMBOL(blk_remove_plug); | |
1580 | ||
1581 | /* | |
1582 | * remove the plug and let it rip.. | |
1583 | */ | |
165125e1 | 1584 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 1585 | { |
7daac490 | 1586 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 LT |
1587 | return; |
1588 | ||
1589 | if (!blk_remove_plug(q)) | |
1590 | return; | |
1591 | ||
22e2c507 | 1592 | q->request_fn(q); |
1da177e4 LT |
1593 | } |
1594 | EXPORT_SYMBOL(__generic_unplug_device); | |
1595 | ||
1596 | /** | |
1597 | * generic_unplug_device - fire a request queue | |
165125e1 | 1598 | * @q: The &struct request_queue in question |
1da177e4 LT |
1599 | * |
1600 | * Description: | |
1601 | * Linux uses plugging to build bigger requests queues before letting | |
1602 | * the device have at them. If a queue is plugged, the I/O scheduler | |
1603 | * is still adding and merging requests on the queue. Once the queue | |
1604 | * gets unplugged, the request_fn defined for the queue is invoked and | |
1605 | * transfers started. | |
1606 | **/ | |
165125e1 | 1607 | void generic_unplug_device(struct request_queue *q) |
1da177e4 LT |
1608 | { |
1609 | spin_lock_irq(q->queue_lock); | |
1610 | __generic_unplug_device(q); | |
1611 | spin_unlock_irq(q->queue_lock); | |
1612 | } | |
1613 | EXPORT_SYMBOL(generic_unplug_device); | |
1614 | ||
1615 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
1616 | struct page *page) | |
1617 | { | |
165125e1 | 1618 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 LT |
1619 | |
1620 | /* | |
1621 | * devices don't necessarily have an ->unplug_fn defined | |
1622 | */ | |
2056a782 JA |
1623 | if (q->unplug_fn) { |
1624 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, | |
1625 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1626 | ||
1da177e4 | 1627 | q->unplug_fn(q); |
2056a782 | 1628 | } |
1da177e4 LT |
1629 | } |
1630 | ||
65f27f38 | 1631 | static void blk_unplug_work(struct work_struct *work) |
1da177e4 | 1632 | { |
165125e1 JA |
1633 | struct request_queue *q = |
1634 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 1635 | |
2056a782 JA |
1636 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, |
1637 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1638 | ||
1da177e4 LT |
1639 | q->unplug_fn(q); |
1640 | } | |
1641 | ||
1642 | static void blk_unplug_timeout(unsigned long data) | |
1643 | { | |
165125e1 | 1644 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 1645 | |
2056a782 JA |
1646 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, |
1647 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1648 | ||
1da177e4 LT |
1649 | kblockd_schedule_work(&q->unplug_work); |
1650 | } | |
1651 | ||
1652 | /** | |
1653 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 1654 | * @q: The &struct request_queue in question |
1da177e4 LT |
1655 | * |
1656 | * Description: | |
1657 | * blk_start_queue() will clear the stop flag on the queue, and call | |
1658 | * the request_fn for the queue if it was in a stopped state when | |
1659 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
1660 | **/ | |
165125e1 | 1661 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 1662 | { |
a038e253 PBG |
1663 | WARN_ON(!irqs_disabled()); |
1664 | ||
1da177e4 LT |
1665 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); |
1666 | ||
1667 | /* | |
1668 | * one level of recursion is ok and is much faster than kicking | |
1669 | * the unplug handling | |
1670 | */ | |
1671 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1672 | q->request_fn(q); | |
1673 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1674 | } else { | |
1675 | blk_plug_device(q); | |
1676 | kblockd_schedule_work(&q->unplug_work); | |
1677 | } | |
1678 | } | |
1679 | ||
1680 | EXPORT_SYMBOL(blk_start_queue); | |
1681 | ||
1682 | /** | |
1683 | * blk_stop_queue - stop a queue | |
165125e1 | 1684 | * @q: The &struct request_queue in question |
1da177e4 LT |
1685 | * |
1686 | * Description: | |
1687 | * The Linux block layer assumes that a block driver will consume all | |
1688 | * entries on the request queue when the request_fn strategy is called. | |
1689 | * Often this will not happen, because of hardware limitations (queue | |
1690 | * depth settings). If a device driver gets a 'queue full' response, | |
1691 | * or if it simply chooses not to queue more I/O at one point, it can | |
1692 | * call this function to prevent the request_fn from being called until | |
1693 | * the driver has signalled it's ready to go again. This happens by calling | |
1694 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
1695 | **/ | |
165125e1 | 1696 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
1697 | { |
1698 | blk_remove_plug(q); | |
1699 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | |
1700 | } | |
1701 | EXPORT_SYMBOL(blk_stop_queue); | |
1702 | ||
1703 | /** | |
1704 | * blk_sync_queue - cancel any pending callbacks on a queue | |
1705 | * @q: the queue | |
1706 | * | |
1707 | * Description: | |
1708 | * The block layer may perform asynchronous callback activity | |
1709 | * on a queue, such as calling the unplug function after a timeout. | |
1710 | * A block device may call blk_sync_queue to ensure that any | |
1711 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 1712 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
1713 | * that its ->make_request_fn will not re-add plugging prior to calling |
1714 | * this function. | |
1715 | * | |
1716 | */ | |
1717 | void blk_sync_queue(struct request_queue *q) | |
1718 | { | |
1719 | del_timer_sync(&q->unplug_timer); | |
1da177e4 LT |
1720 | } |
1721 | EXPORT_SYMBOL(blk_sync_queue); | |
1722 | ||
1723 | /** | |
1724 | * blk_run_queue - run a single device queue | |
1725 | * @q: The queue to run | |
1726 | */ | |
1727 | void blk_run_queue(struct request_queue *q) | |
1728 | { | |
1729 | unsigned long flags; | |
1730 | ||
1731 | spin_lock_irqsave(q->queue_lock, flags); | |
1732 | blk_remove_plug(q); | |
dac07ec1 JA |
1733 | |
1734 | /* | |
1735 | * Only recurse once to avoid overrunning the stack, let the unplug | |
1736 | * handling reinvoke the handler shortly if we already got there. | |
1737 | */ | |
1738 | if (!elv_queue_empty(q)) { | |
1739 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1740 | q->request_fn(q); | |
1741 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1742 | } else { | |
1743 | blk_plug_device(q); | |
1744 | kblockd_schedule_work(&q->unplug_work); | |
1745 | } | |
1746 | } | |
1747 | ||
1da177e4 LT |
1748 | spin_unlock_irqrestore(q->queue_lock, flags); |
1749 | } | |
1750 | EXPORT_SYMBOL(blk_run_queue); | |
1751 | ||
1752 | /** | |
165125e1 | 1753 | * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed |
a580290c | 1754 | * @kobj: the kobj belonging of the request queue to be released |
1da177e4 LT |
1755 | * |
1756 | * Description: | |
1757 | * blk_cleanup_queue is the pair to blk_init_queue() or | |
1758 | * blk_queue_make_request(). It should be called when a request queue is | |
1759 | * being released; typically when a block device is being de-registered. | |
1760 | * Currently, its primary task it to free all the &struct request | |
1761 | * structures that were allocated to the queue and the queue itself. | |
1762 | * | |
1763 | * Caveat: | |
1764 | * Hopefully the low level driver will have finished any | |
1765 | * outstanding requests first... | |
1766 | **/ | |
483f4afc | 1767 | static void blk_release_queue(struct kobject *kobj) |
1da177e4 | 1768 | { |
165125e1 JA |
1769 | struct request_queue *q = |
1770 | container_of(kobj, struct request_queue, kobj); | |
1da177e4 LT |
1771 | struct request_list *rl = &q->rq; |
1772 | ||
1da177e4 LT |
1773 | blk_sync_queue(q); |
1774 | ||
1775 | if (rl->rq_pool) | |
1776 | mempool_destroy(rl->rq_pool); | |
1777 | ||
1778 | if (q->queue_tags) | |
1779 | __blk_queue_free_tags(q); | |
1780 | ||
6c5c9341 | 1781 | blk_trace_shutdown(q); |
2056a782 | 1782 | |
1da177e4 LT |
1783 | kmem_cache_free(requestq_cachep, q); |
1784 | } | |
1785 | ||
165125e1 | 1786 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
1787 | { |
1788 | kobject_put(&q->kobj); | |
1789 | } | |
1790 | EXPORT_SYMBOL(blk_put_queue); | |
1791 | ||
165125e1 | 1792 | void blk_cleanup_queue(struct request_queue * q) |
483f4afc AV |
1793 | { |
1794 | mutex_lock(&q->sysfs_lock); | |
1795 | set_bit(QUEUE_FLAG_DEAD, &q->queue_flags); | |
1796 | mutex_unlock(&q->sysfs_lock); | |
1797 | ||
1798 | if (q->elevator) | |
1799 | elevator_exit(q->elevator); | |
1800 | ||
1801 | blk_put_queue(q); | |
1802 | } | |
1803 | ||
1da177e4 LT |
1804 | EXPORT_SYMBOL(blk_cleanup_queue); |
1805 | ||
165125e1 | 1806 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
1807 | { |
1808 | struct request_list *rl = &q->rq; | |
1809 | ||
1810 | rl->count[READ] = rl->count[WRITE] = 0; | |
1811 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
cb98fc8b | 1812 | rl->elvpriv = 0; |
1da177e4 LT |
1813 | init_waitqueue_head(&rl->wait[READ]); |
1814 | init_waitqueue_head(&rl->wait[WRITE]); | |
1da177e4 | 1815 | |
1946089a CL |
1816 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
1817 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
1818 | |
1819 | if (!rl->rq_pool) | |
1820 | return -ENOMEM; | |
1821 | ||
1822 | return 0; | |
1823 | } | |
1824 | ||
165125e1 | 1825 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 1826 | { |
1946089a CL |
1827 | return blk_alloc_queue_node(gfp_mask, -1); |
1828 | } | |
1829 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 1830 | |
483f4afc AV |
1831 | static struct kobj_type queue_ktype; |
1832 | ||
165125e1 | 1833 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 1834 | { |
165125e1 | 1835 | struct request_queue *q; |
1946089a | 1836 | |
94f6030c CL |
1837 | q = kmem_cache_alloc_node(requestq_cachep, |
1838 | gfp_mask | __GFP_ZERO, node_id); | |
1da177e4 LT |
1839 | if (!q) |
1840 | return NULL; | |
1841 | ||
1da177e4 | 1842 | init_timer(&q->unplug_timer); |
483f4afc AV |
1843 | |
1844 | snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue"); | |
1845 | q->kobj.ktype = &queue_ktype; | |
1846 | kobject_init(&q->kobj); | |
1da177e4 LT |
1847 | |
1848 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; | |
1849 | q->backing_dev_info.unplug_io_data = q; | |
1850 | ||
483f4afc AV |
1851 | mutex_init(&q->sysfs_lock); |
1852 | ||
1da177e4 LT |
1853 | return q; |
1854 | } | |
1946089a | 1855 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
1856 | |
1857 | /** | |
1858 | * blk_init_queue - prepare a request queue for use with a block device | |
1859 | * @rfn: The function to be called to process requests that have been | |
1860 | * placed on the queue. | |
1861 | * @lock: Request queue spin lock | |
1862 | * | |
1863 | * Description: | |
1864 | * If a block device wishes to use the standard request handling procedures, | |
1865 | * which sorts requests and coalesces adjacent requests, then it must | |
1866 | * call blk_init_queue(). The function @rfn will be called when there | |
1867 | * are requests on the queue that need to be processed. If the device | |
1868 | * supports plugging, then @rfn may not be called immediately when requests | |
1869 | * are available on the queue, but may be called at some time later instead. | |
1870 | * Plugged queues are generally unplugged when a buffer belonging to one | |
1871 | * of the requests on the queue is needed, or due to memory pressure. | |
1872 | * | |
1873 | * @rfn is not required, or even expected, to remove all requests off the | |
1874 | * queue, but only as many as it can handle at a time. If it does leave | |
1875 | * requests on the queue, it is responsible for arranging that the requests | |
1876 | * get dealt with eventually. | |
1877 | * | |
1878 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
1879 | * request queue; this lock will be taken also from interrupt context, so irq |
1880 | * disabling is needed for it. | |
1da177e4 LT |
1881 | * |
1882 | * Function returns a pointer to the initialized request queue, or NULL if | |
1883 | * it didn't succeed. | |
1884 | * | |
1885 | * Note: | |
1886 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
1887 | * when the block device is deactivated (such as at module unload). | |
1888 | **/ | |
1946089a | 1889 | |
165125e1 | 1890 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 1891 | { |
1946089a CL |
1892 | return blk_init_queue_node(rfn, lock, -1); |
1893 | } | |
1894 | EXPORT_SYMBOL(blk_init_queue); | |
1895 | ||
165125e1 | 1896 | struct request_queue * |
1946089a CL |
1897 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
1898 | { | |
165125e1 | 1899 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
1900 | |
1901 | if (!q) | |
1902 | return NULL; | |
1903 | ||
1946089a | 1904 | q->node = node_id; |
8669aafd AV |
1905 | if (blk_init_free_list(q)) { |
1906 | kmem_cache_free(requestq_cachep, q); | |
1907 | return NULL; | |
1908 | } | |
1da177e4 | 1909 | |
152587de | 1910 | /* |
1911 | * if caller didn't supply a lock, they get per-queue locking with | |
1912 | * our embedded lock | |
1913 | */ | |
1914 | if (!lock) { | |
1915 | spin_lock_init(&q->__queue_lock); | |
1916 | lock = &q->__queue_lock; | |
1917 | } | |
1918 | ||
1da177e4 | 1919 | q->request_fn = rfn; |
1da177e4 LT |
1920 | q->prep_rq_fn = NULL; |
1921 | q->unplug_fn = generic_unplug_device; | |
1922 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); | |
1923 | q->queue_lock = lock; | |
1924 | ||
1925 | blk_queue_segment_boundary(q, 0xffffffff); | |
1926 | ||
1927 | blk_queue_make_request(q, __make_request); | |
1928 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
1929 | ||
1930 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
1931 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
1932 | ||
44ec9542 AS |
1933 | q->sg_reserved_size = INT_MAX; |
1934 | ||
1da177e4 LT |
1935 | /* |
1936 | * all done | |
1937 | */ | |
1938 | if (!elevator_init(q, NULL)) { | |
1939 | blk_queue_congestion_threshold(q); | |
1940 | return q; | |
1941 | } | |
1942 | ||
8669aafd | 1943 | blk_put_queue(q); |
1da177e4 LT |
1944 | return NULL; |
1945 | } | |
1946089a | 1946 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 1947 | |
165125e1 | 1948 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 1949 | { |
fde6ad22 | 1950 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 1951 | kobject_get(&q->kobj); |
1da177e4 LT |
1952 | return 0; |
1953 | } | |
1954 | ||
1955 | return 1; | |
1956 | } | |
1957 | ||
1958 | EXPORT_SYMBOL(blk_get_queue); | |
1959 | ||
165125e1 | 1960 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1961 | { |
4aff5e23 | 1962 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 1963 | elv_put_request(q, rq); |
1da177e4 LT |
1964 | mempool_free(rq, q->rq.rq_pool); |
1965 | } | |
1966 | ||
1ea25ecb | 1967 | static struct request * |
165125e1 | 1968 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) |
1da177e4 LT |
1969 | { |
1970 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
1971 | ||
1972 | if (!rq) | |
1973 | return NULL; | |
1974 | ||
1975 | /* | |
4aff5e23 | 1976 | * first three bits are identical in rq->cmd_flags and bio->bi_rw, |
1da177e4 LT |
1977 | * see bio.h and blkdev.h |
1978 | */ | |
49171e5c | 1979 | rq->cmd_flags = rw | REQ_ALLOCED; |
1da177e4 | 1980 | |
cb98fc8b | 1981 | if (priv) { |
cb78b285 | 1982 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
1983 | mempool_free(rq, q->rq.rq_pool); |
1984 | return NULL; | |
1985 | } | |
4aff5e23 | 1986 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 1987 | } |
1da177e4 | 1988 | |
cb98fc8b | 1989 | return rq; |
1da177e4 LT |
1990 | } |
1991 | ||
1992 | /* | |
1993 | * ioc_batching returns true if the ioc is a valid batching request and | |
1994 | * should be given priority access to a request. | |
1995 | */ | |
165125e1 | 1996 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
1997 | { |
1998 | if (!ioc) | |
1999 | return 0; | |
2000 | ||
2001 | /* | |
2002 | * Make sure the process is able to allocate at least 1 request | |
2003 | * even if the batch times out, otherwise we could theoretically | |
2004 | * lose wakeups. | |
2005 | */ | |
2006 | return ioc->nr_batch_requests == q->nr_batching || | |
2007 | (ioc->nr_batch_requests > 0 | |
2008 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
2009 | } | |
2010 | ||
2011 | /* | |
2012 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
2013 | * will cause the process to be a "batcher" on all queues in the system. This | |
2014 | * is the behaviour we want though - once it gets a wakeup it should be given | |
2015 | * a nice run. | |
2016 | */ | |
165125e1 | 2017 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
2018 | { |
2019 | if (!ioc || ioc_batching(q, ioc)) | |
2020 | return; | |
2021 | ||
2022 | ioc->nr_batch_requests = q->nr_batching; | |
2023 | ioc->last_waited = jiffies; | |
2024 | } | |
2025 | ||
165125e1 | 2026 | static void __freed_request(struct request_queue *q, int rw) |
1da177e4 LT |
2027 | { |
2028 | struct request_list *rl = &q->rq; | |
2029 | ||
2030 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
79e2de4b | 2031 | blk_clear_queue_congested(q, rw); |
1da177e4 LT |
2032 | |
2033 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
2034 | if (waitqueue_active(&rl->wait[rw])) |
2035 | wake_up(&rl->wait[rw]); | |
2036 | ||
2037 | blk_clear_queue_full(q, rw); | |
2038 | } | |
2039 | } | |
2040 | ||
2041 | /* | |
2042 | * A request has just been released. Account for it, update the full and | |
2043 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
2044 | */ | |
165125e1 | 2045 | static void freed_request(struct request_queue *q, int rw, int priv) |
1da177e4 LT |
2046 | { |
2047 | struct request_list *rl = &q->rq; | |
2048 | ||
2049 | rl->count[rw]--; | |
cb98fc8b TH |
2050 | if (priv) |
2051 | rl->elvpriv--; | |
1da177e4 LT |
2052 | |
2053 | __freed_request(q, rw); | |
2054 | ||
2055 | if (unlikely(rl->starved[rw ^ 1])) | |
2056 | __freed_request(q, rw ^ 1); | |
1da177e4 LT |
2057 | } |
2058 | ||
2059 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
2060 | /* | |
d6344532 NP |
2061 | * Get a free request, queue_lock must be held. |
2062 | * Returns NULL on failure, with queue_lock held. | |
2063 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 2064 | */ |
165125e1 | 2065 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 2066 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
2067 | { |
2068 | struct request *rq = NULL; | |
2069 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 2070 | struct io_context *ioc = NULL; |
7749a8d4 | 2071 | const int rw = rw_flags & 0x01; |
88ee5ef1 JA |
2072 | int may_queue, priv; |
2073 | ||
7749a8d4 | 2074 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
2075 | if (may_queue == ELV_MQUEUE_NO) |
2076 | goto rq_starved; | |
2077 | ||
2078 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
2079 | if (rl->count[rw]+1 >= q->nr_requests) { | |
b5deef90 | 2080 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
2081 | /* |
2082 | * The queue will fill after this allocation, so set | |
2083 | * it as full, and mark this process as "batching". | |
2084 | * This process will be allowed to complete a batch of | |
2085 | * requests, others will be blocked. | |
2086 | */ | |
2087 | if (!blk_queue_full(q, rw)) { | |
2088 | ioc_set_batching(q, ioc); | |
2089 | blk_set_queue_full(q, rw); | |
2090 | } else { | |
2091 | if (may_queue != ELV_MQUEUE_MUST | |
2092 | && !ioc_batching(q, ioc)) { | |
2093 | /* | |
2094 | * The queue is full and the allocating | |
2095 | * process is not a "batcher", and not | |
2096 | * exempted by the IO scheduler | |
2097 | */ | |
2098 | goto out; | |
2099 | } | |
2100 | } | |
1da177e4 | 2101 | } |
79e2de4b | 2102 | blk_set_queue_congested(q, rw); |
1da177e4 LT |
2103 | } |
2104 | ||
082cf69e JA |
2105 | /* |
2106 | * Only allow batching queuers to allocate up to 50% over the defined | |
2107 | * limit of requests, otherwise we could have thousands of requests | |
2108 | * allocated with any setting of ->nr_requests | |
2109 | */ | |
fd782a4a | 2110 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 2111 | goto out; |
fd782a4a | 2112 | |
1da177e4 LT |
2113 | rl->count[rw]++; |
2114 | rl->starved[rw] = 0; | |
cb98fc8b | 2115 | |
64521d1a | 2116 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
2117 | if (priv) |
2118 | rl->elvpriv++; | |
2119 | ||
1da177e4 LT |
2120 | spin_unlock_irq(q->queue_lock); |
2121 | ||
7749a8d4 | 2122 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 2123 | if (unlikely(!rq)) { |
1da177e4 LT |
2124 | /* |
2125 | * Allocation failed presumably due to memory. Undo anything | |
2126 | * we might have messed up. | |
2127 | * | |
2128 | * Allocating task should really be put onto the front of the | |
2129 | * wait queue, but this is pretty rare. | |
2130 | */ | |
2131 | spin_lock_irq(q->queue_lock); | |
cb98fc8b | 2132 | freed_request(q, rw, priv); |
1da177e4 LT |
2133 | |
2134 | /* | |
2135 | * in the very unlikely event that allocation failed and no | |
2136 | * requests for this direction was pending, mark us starved | |
2137 | * so that freeing of a request in the other direction will | |
2138 | * notice us. another possible fix would be to split the | |
2139 | * rq mempool into READ and WRITE | |
2140 | */ | |
2141 | rq_starved: | |
2142 | if (unlikely(rl->count[rw] == 0)) | |
2143 | rl->starved[rw] = 1; | |
2144 | ||
1da177e4 LT |
2145 | goto out; |
2146 | } | |
2147 | ||
88ee5ef1 JA |
2148 | /* |
2149 | * ioc may be NULL here, and ioc_batching will be false. That's | |
2150 | * OK, if the queue is under the request limit then requests need | |
2151 | * not count toward the nr_batch_requests limit. There will always | |
2152 | * be some limit enforced by BLK_BATCH_TIME. | |
2153 | */ | |
1da177e4 LT |
2154 | if (ioc_batching(q, ioc)) |
2155 | ioc->nr_batch_requests--; | |
2156 | ||
2157 | rq_init(q, rq); | |
2056a782 JA |
2158 | |
2159 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); | |
1da177e4 | 2160 | out: |
1da177e4 LT |
2161 | return rq; |
2162 | } | |
2163 | ||
2164 | /* | |
2165 | * No available requests for this queue, unplug the device and wait for some | |
2166 | * requests to become available. | |
d6344532 NP |
2167 | * |
2168 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 2169 | */ |
165125e1 | 2170 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 2171 | struct bio *bio) |
1da177e4 | 2172 | { |
7749a8d4 | 2173 | const int rw = rw_flags & 0x01; |
1da177e4 LT |
2174 | struct request *rq; |
2175 | ||
7749a8d4 | 2176 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
2177 | while (!rq) { |
2178 | DEFINE_WAIT(wait); | |
1da177e4 LT |
2179 | struct request_list *rl = &q->rq; |
2180 | ||
2181 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
2182 | TASK_UNINTERRUPTIBLE); | |
2183 | ||
7749a8d4 | 2184 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
1da177e4 LT |
2185 | |
2186 | if (!rq) { | |
2187 | struct io_context *ioc; | |
2188 | ||
2056a782 JA |
2189 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); |
2190 | ||
d6344532 NP |
2191 | __generic_unplug_device(q); |
2192 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
2193 | io_schedule(); |
2194 | ||
2195 | /* | |
2196 | * After sleeping, we become a "batching" process and | |
2197 | * will be able to allocate at least one request, and | |
2198 | * up to a big batch of them for a small period time. | |
2199 | * See ioc_batching, ioc_set_batching | |
2200 | */ | |
b5deef90 | 2201 | ioc = current_io_context(GFP_NOIO, q->node); |
1da177e4 | 2202 | ioc_set_batching(q, ioc); |
d6344532 NP |
2203 | |
2204 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
2205 | } |
2206 | finish_wait(&rl->wait[rw], &wait); | |
450991bc | 2207 | } |
1da177e4 LT |
2208 | |
2209 | return rq; | |
2210 | } | |
2211 | ||
165125e1 | 2212 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
2213 | { |
2214 | struct request *rq; | |
2215 | ||
2216 | BUG_ON(rw != READ && rw != WRITE); | |
2217 | ||
d6344532 NP |
2218 | spin_lock_irq(q->queue_lock); |
2219 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 2220 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 2221 | } else { |
22e2c507 | 2222 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
2223 | if (!rq) |
2224 | spin_unlock_irq(q->queue_lock); | |
2225 | } | |
2226 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
2227 | |
2228 | return rq; | |
2229 | } | |
1da177e4 LT |
2230 | EXPORT_SYMBOL(blk_get_request); |
2231 | ||
dc72ef4a JA |
2232 | /** |
2233 | * blk_start_queueing - initiate dispatch of requests to device | |
2234 | * @q: request queue to kick into gear | |
2235 | * | |
2236 | * This is basically a helper to remove the need to know whether a queue | |
2237 | * is plugged or not if someone just wants to initiate dispatch of requests | |
2238 | * for this queue. | |
2239 | * | |
2240 | * The queue lock must be held with interrupts disabled. | |
2241 | */ | |
165125e1 | 2242 | void blk_start_queueing(struct request_queue *q) |
dc72ef4a JA |
2243 | { |
2244 | if (!blk_queue_plugged(q)) | |
2245 | q->request_fn(q); | |
2246 | else | |
2247 | __generic_unplug_device(q); | |
2248 | } | |
2249 | EXPORT_SYMBOL(blk_start_queueing); | |
2250 | ||
1da177e4 LT |
2251 | /** |
2252 | * blk_requeue_request - put a request back on queue | |
2253 | * @q: request queue where request should be inserted | |
2254 | * @rq: request to be inserted | |
2255 | * | |
2256 | * Description: | |
2257 | * Drivers often keep queueing requests until the hardware cannot accept | |
2258 | * more, when that condition happens we need to put the request back | |
2259 | * on the queue. Must be called with queue lock held. | |
2260 | */ | |
165125e1 | 2261 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2262 | { |
2056a782 JA |
2263 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); |
2264 | ||
1da177e4 LT |
2265 | if (blk_rq_tagged(rq)) |
2266 | blk_queue_end_tag(q, rq); | |
2267 | ||
2268 | elv_requeue_request(q, rq); | |
2269 | } | |
2270 | ||
2271 | EXPORT_SYMBOL(blk_requeue_request); | |
2272 | ||
2273 | /** | |
2274 | * blk_insert_request - insert a special request in to a request queue | |
2275 | * @q: request queue where request should be inserted | |
2276 | * @rq: request to be inserted | |
2277 | * @at_head: insert request at head or tail of queue | |
2278 | * @data: private data | |
1da177e4 LT |
2279 | * |
2280 | * Description: | |
2281 | * Many block devices need to execute commands asynchronously, so they don't | |
2282 | * block the whole kernel from preemption during request execution. This is | |
2283 | * accomplished normally by inserting aritficial requests tagged as | |
2284 | * REQ_SPECIAL in to the corresponding request queue, and letting them be | |
2285 | * scheduled for actual execution by the request queue. | |
2286 | * | |
2287 | * We have the option of inserting the head or the tail of the queue. | |
2288 | * Typically we use the tail for new ioctls and so forth. We use the head | |
2289 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
2290 | * host that is unable to accept a particular command. | |
2291 | */ | |
165125e1 | 2292 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 2293 | int at_head, void *data) |
1da177e4 | 2294 | { |
867d1191 | 2295 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
2296 | unsigned long flags; |
2297 | ||
2298 | /* | |
2299 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
2300 | * must not attempt merges on this) and that it acts as a soft | |
2301 | * barrier | |
2302 | */ | |
4aff5e23 JA |
2303 | rq->cmd_type = REQ_TYPE_SPECIAL; |
2304 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
2305 | |
2306 | rq->special = data; | |
2307 | ||
2308 | spin_lock_irqsave(q->queue_lock, flags); | |
2309 | ||
2310 | /* | |
2311 | * If command is tagged, release the tag | |
2312 | */ | |
867d1191 TH |
2313 | if (blk_rq_tagged(rq)) |
2314 | blk_queue_end_tag(q, rq); | |
1da177e4 | 2315 | |
867d1191 TH |
2316 | drive_stat_acct(rq, rq->nr_sectors, 1); |
2317 | __elv_add_request(q, rq, where, 0); | |
dc72ef4a | 2318 | blk_start_queueing(q); |
1da177e4 LT |
2319 | spin_unlock_irqrestore(q->queue_lock, flags); |
2320 | } | |
2321 | ||
2322 | EXPORT_SYMBOL(blk_insert_request); | |
2323 | ||
0e75f906 MC |
2324 | static int __blk_rq_unmap_user(struct bio *bio) |
2325 | { | |
2326 | int ret = 0; | |
2327 | ||
2328 | if (bio) { | |
2329 | if (bio_flagged(bio, BIO_USER_MAPPED)) | |
2330 | bio_unmap_user(bio); | |
2331 | else | |
2332 | ret = bio_uncopy_user(bio); | |
2333 | } | |
2334 | ||
2335 | return ret; | |
2336 | } | |
2337 | ||
165125e1 | 2338 | static int __blk_rq_map_user(struct request_queue *q, struct request *rq, |
0e75f906 MC |
2339 | void __user *ubuf, unsigned int len) |
2340 | { | |
2341 | unsigned long uaddr; | |
2342 | struct bio *bio, *orig_bio; | |
2343 | int reading, ret; | |
2344 | ||
2345 | reading = rq_data_dir(rq) == READ; | |
2346 | ||
2347 | /* | |
2348 | * if alignment requirement is satisfied, map in user pages for | |
2349 | * direct dma. else, set up kernel bounce buffers | |
2350 | */ | |
2351 | uaddr = (unsigned long) ubuf; | |
2352 | if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q))) | |
2353 | bio = bio_map_user(q, NULL, uaddr, len, reading); | |
2354 | else | |
2355 | bio = bio_copy_user(q, uaddr, len, reading); | |
2356 | ||
2985259b | 2357 | if (IS_ERR(bio)) |
0e75f906 | 2358 | return PTR_ERR(bio); |
0e75f906 MC |
2359 | |
2360 | orig_bio = bio; | |
2361 | blk_queue_bounce(q, &bio); | |
2985259b | 2362 | |
0e75f906 MC |
2363 | /* |
2364 | * We link the bounce buffer in and could have to traverse it | |
2365 | * later so we have to get a ref to prevent it from being freed | |
2366 | */ | |
2367 | bio_get(bio); | |
2368 | ||
0e75f906 MC |
2369 | if (!rq->bio) |
2370 | blk_rq_bio_prep(q, rq, bio); | |
1aa4f24f | 2371 | else if (!ll_back_merge_fn(q, rq, bio)) { |
0e75f906 | 2372 | ret = -EINVAL; |
0e75f906 MC |
2373 | goto unmap_bio; |
2374 | } else { | |
2375 | rq->biotail->bi_next = bio; | |
2376 | rq->biotail = bio; | |
2377 | ||
0e75f906 MC |
2378 | rq->data_len += bio->bi_size; |
2379 | } | |
0e75f906 MC |
2380 | |
2381 | return bio->bi_size; | |
2382 | ||
2383 | unmap_bio: | |
2384 | /* if it was boucned we must call the end io function */ | |
2385 | bio_endio(bio, bio->bi_size, 0); | |
2386 | __blk_rq_unmap_user(orig_bio); | |
2387 | bio_put(bio); | |
2388 | return ret; | |
2389 | } | |
2390 | ||
1da177e4 LT |
2391 | /** |
2392 | * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage | |
2393 | * @q: request queue where request should be inserted | |
73747aed | 2394 | * @rq: request structure to fill |
1da177e4 LT |
2395 | * @ubuf: the user buffer |
2396 | * @len: length of user data | |
2397 | * | |
2398 | * Description: | |
2399 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2400 | * a kernel bounce buffer is used. | |
2401 | * | |
2402 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2403 | * still in process context. | |
2404 | * | |
2405 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2406 | * before being submitted to the device, as pages mapped may be out of | |
2407 | * reach. It's the callers responsibility to make sure this happens. The | |
2408 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2409 | * unmapping. | |
2410 | */ | |
165125e1 JA |
2411 | int blk_rq_map_user(struct request_queue *q, struct request *rq, |
2412 | void __user *ubuf, unsigned long len) | |
1da177e4 | 2413 | { |
0e75f906 | 2414 | unsigned long bytes_read = 0; |
8e5cfc45 | 2415 | struct bio *bio = NULL; |
0e75f906 | 2416 | int ret; |
1da177e4 | 2417 | |
defd94b7 | 2418 | if (len > (q->max_hw_sectors << 9)) |
dd1cab95 JA |
2419 | return -EINVAL; |
2420 | if (!len || !ubuf) | |
2421 | return -EINVAL; | |
1da177e4 | 2422 | |
0e75f906 MC |
2423 | while (bytes_read != len) { |
2424 | unsigned long map_len, end, start; | |
1da177e4 | 2425 | |
0e75f906 MC |
2426 | map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE); |
2427 | end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1) | |
2428 | >> PAGE_SHIFT; | |
2429 | start = (unsigned long)ubuf >> PAGE_SHIFT; | |
1da177e4 | 2430 | |
0e75f906 MC |
2431 | /* |
2432 | * A bad offset could cause us to require BIO_MAX_PAGES + 1 | |
2433 | * pages. If this happens we just lower the requested | |
2434 | * mapping len by a page so that we can fit | |
2435 | */ | |
2436 | if (end - start > BIO_MAX_PAGES) | |
2437 | map_len -= PAGE_SIZE; | |
1da177e4 | 2438 | |
0e75f906 MC |
2439 | ret = __blk_rq_map_user(q, rq, ubuf, map_len); |
2440 | if (ret < 0) | |
2441 | goto unmap_rq; | |
8e5cfc45 JA |
2442 | if (!bio) |
2443 | bio = rq->bio; | |
0e75f906 MC |
2444 | bytes_read += ret; |
2445 | ubuf += ret; | |
1da177e4 LT |
2446 | } |
2447 | ||
0e75f906 MC |
2448 | rq->buffer = rq->data = NULL; |
2449 | return 0; | |
2450 | unmap_rq: | |
8e5cfc45 | 2451 | blk_rq_unmap_user(bio); |
0e75f906 | 2452 | return ret; |
1da177e4 LT |
2453 | } |
2454 | ||
2455 | EXPORT_SYMBOL(blk_rq_map_user); | |
2456 | ||
f1970baf JB |
2457 | /** |
2458 | * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage | |
2459 | * @q: request queue where request should be inserted | |
2460 | * @rq: request to map data to | |
2461 | * @iov: pointer to the iovec | |
2462 | * @iov_count: number of elements in the iovec | |
af9997e4 | 2463 | * @len: I/O byte count |
f1970baf JB |
2464 | * |
2465 | * Description: | |
2466 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2467 | * a kernel bounce buffer is used. | |
2468 | * | |
2469 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2470 | * still in process context. | |
2471 | * | |
2472 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2473 | * before being submitted to the device, as pages mapped may be out of | |
2474 | * reach. It's the callers responsibility to make sure this happens. The | |
2475 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2476 | * unmapping. | |
2477 | */ | |
165125e1 | 2478 | int blk_rq_map_user_iov(struct request_queue *q, struct request *rq, |
0e75f906 | 2479 | struct sg_iovec *iov, int iov_count, unsigned int len) |
f1970baf JB |
2480 | { |
2481 | struct bio *bio; | |
2482 | ||
2483 | if (!iov || iov_count <= 0) | |
2484 | return -EINVAL; | |
2485 | ||
2486 | /* we don't allow misaligned data like bio_map_user() does. If the | |
2487 | * user is using sg, they're expected to know the alignment constraints | |
2488 | * and respect them accordingly */ | |
2489 | bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ); | |
2490 | if (IS_ERR(bio)) | |
2491 | return PTR_ERR(bio); | |
2492 | ||
0e75f906 MC |
2493 | if (bio->bi_size != len) { |
2494 | bio_endio(bio, bio->bi_size, 0); | |
2495 | bio_unmap_user(bio); | |
2496 | return -EINVAL; | |
2497 | } | |
2498 | ||
2499 | bio_get(bio); | |
f1970baf JB |
2500 | blk_rq_bio_prep(q, rq, bio); |
2501 | rq->buffer = rq->data = NULL; | |
f1970baf JB |
2502 | return 0; |
2503 | } | |
2504 | ||
2505 | EXPORT_SYMBOL(blk_rq_map_user_iov); | |
2506 | ||
1da177e4 LT |
2507 | /** |
2508 | * blk_rq_unmap_user - unmap a request with user data | |
8e5cfc45 | 2509 | * @bio: start of bio list |
1da177e4 LT |
2510 | * |
2511 | * Description: | |
8e5cfc45 JA |
2512 | * Unmap a rq previously mapped by blk_rq_map_user(). The caller must |
2513 | * supply the original rq->bio from the blk_rq_map_user() return, since | |
2514 | * the io completion may have changed rq->bio. | |
1da177e4 | 2515 | */ |
8e5cfc45 | 2516 | int blk_rq_unmap_user(struct bio *bio) |
1da177e4 | 2517 | { |
8e5cfc45 | 2518 | struct bio *mapped_bio; |
48785bb9 | 2519 | int ret = 0, ret2; |
1da177e4 | 2520 | |
8e5cfc45 JA |
2521 | while (bio) { |
2522 | mapped_bio = bio; | |
2523 | if (unlikely(bio_flagged(bio, BIO_BOUNCED))) | |
0e75f906 | 2524 | mapped_bio = bio->bi_private; |
1da177e4 | 2525 | |
48785bb9 JA |
2526 | ret2 = __blk_rq_unmap_user(mapped_bio); |
2527 | if (ret2 && !ret) | |
2528 | ret = ret2; | |
2529 | ||
8e5cfc45 JA |
2530 | mapped_bio = bio; |
2531 | bio = bio->bi_next; | |
2532 | bio_put(mapped_bio); | |
0e75f906 | 2533 | } |
48785bb9 JA |
2534 | |
2535 | return ret; | |
1da177e4 LT |
2536 | } |
2537 | ||
2538 | EXPORT_SYMBOL(blk_rq_unmap_user); | |
2539 | ||
df46b9a4 MC |
2540 | /** |
2541 | * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage | |
2542 | * @q: request queue where request should be inserted | |
73747aed | 2543 | * @rq: request to fill |
df46b9a4 MC |
2544 | * @kbuf: the kernel buffer |
2545 | * @len: length of user data | |
73747aed | 2546 | * @gfp_mask: memory allocation flags |
df46b9a4 | 2547 | */ |
165125e1 | 2548 | int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf, |
8267e268 | 2549 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 | 2550 | { |
df46b9a4 MC |
2551 | struct bio *bio; |
2552 | ||
defd94b7 | 2553 | if (len > (q->max_hw_sectors << 9)) |
dd1cab95 JA |
2554 | return -EINVAL; |
2555 | if (!len || !kbuf) | |
2556 | return -EINVAL; | |
df46b9a4 MC |
2557 | |
2558 | bio = bio_map_kern(q, kbuf, len, gfp_mask); | |
dd1cab95 JA |
2559 | if (IS_ERR(bio)) |
2560 | return PTR_ERR(bio); | |
df46b9a4 | 2561 | |
dd1cab95 JA |
2562 | if (rq_data_dir(rq) == WRITE) |
2563 | bio->bi_rw |= (1 << BIO_RW); | |
df46b9a4 | 2564 | |
dd1cab95 | 2565 | blk_rq_bio_prep(q, rq, bio); |
821de3a2 | 2566 | blk_queue_bounce(q, &rq->bio); |
dd1cab95 | 2567 | rq->buffer = rq->data = NULL; |
dd1cab95 | 2568 | return 0; |
df46b9a4 MC |
2569 | } |
2570 | ||
2571 | EXPORT_SYMBOL(blk_rq_map_kern); | |
2572 | ||
73747aed CH |
2573 | /** |
2574 | * blk_execute_rq_nowait - insert a request into queue for execution | |
2575 | * @q: queue to insert the request in | |
2576 | * @bd_disk: matching gendisk | |
2577 | * @rq: request to insert | |
2578 | * @at_head: insert request at head or tail of queue | |
2579 | * @done: I/O completion handler | |
2580 | * | |
2581 | * Description: | |
2582 | * Insert a fully prepared request at the back of the io scheduler queue | |
2583 | * for execution. Don't wait for completion. | |
2584 | */ | |
165125e1 | 2585 | void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk, |
f1970baf | 2586 | struct request *rq, int at_head, |
8ffdc655 | 2587 | rq_end_io_fn *done) |
f1970baf JB |
2588 | { |
2589 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | |
2590 | ||
2591 | rq->rq_disk = bd_disk; | |
4aff5e23 | 2592 | rq->cmd_flags |= REQ_NOMERGE; |
f1970baf | 2593 | rq->end_io = done; |
4c5d0bbd AM |
2594 | WARN_ON(irqs_disabled()); |
2595 | spin_lock_irq(q->queue_lock); | |
2596 | __elv_add_request(q, rq, where, 1); | |
2597 | __generic_unplug_device(q); | |
2598 | spin_unlock_irq(q->queue_lock); | |
f1970baf | 2599 | } |
6e39b69e MC |
2600 | EXPORT_SYMBOL_GPL(blk_execute_rq_nowait); |
2601 | ||
1da177e4 LT |
2602 | /** |
2603 | * blk_execute_rq - insert a request into queue for execution | |
2604 | * @q: queue to insert the request in | |
2605 | * @bd_disk: matching gendisk | |
2606 | * @rq: request to insert | |
994ca9a1 | 2607 | * @at_head: insert request at head or tail of queue |
1da177e4 LT |
2608 | * |
2609 | * Description: | |
2610 | * Insert a fully prepared request at the back of the io scheduler queue | |
73747aed | 2611 | * for execution and wait for completion. |
1da177e4 | 2612 | */ |
165125e1 | 2613 | int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk, |
994ca9a1 | 2614 | struct request *rq, int at_head) |
1da177e4 | 2615 | { |
60be6b9a | 2616 | DECLARE_COMPLETION_ONSTACK(wait); |
1da177e4 LT |
2617 | char sense[SCSI_SENSE_BUFFERSIZE]; |
2618 | int err = 0; | |
2619 | ||
1da177e4 LT |
2620 | /* |
2621 | * we need an extra reference to the request, so we can look at | |
2622 | * it after io completion | |
2623 | */ | |
2624 | rq->ref_count++; | |
2625 | ||
2626 | if (!rq->sense) { | |
2627 | memset(sense, 0, sizeof(sense)); | |
2628 | rq->sense = sense; | |
2629 | rq->sense_len = 0; | |
2630 | } | |
2631 | ||
c00895ab | 2632 | rq->end_io_data = &wait; |
994ca9a1 | 2633 | blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq); |
1da177e4 | 2634 | wait_for_completion(&wait); |
1da177e4 LT |
2635 | |
2636 | if (rq->errors) | |
2637 | err = -EIO; | |
2638 | ||
2639 | return err; | |
2640 | } | |
2641 | ||
2642 | EXPORT_SYMBOL(blk_execute_rq); | |
2643 | ||
2644 | /** | |
2645 | * blkdev_issue_flush - queue a flush | |
2646 | * @bdev: blockdev to issue flush for | |
2647 | * @error_sector: error sector | |
2648 | * | |
2649 | * Description: | |
2650 | * Issue a flush for the block device in question. Caller can supply | |
2651 | * room for storing the error offset in case of a flush error, if they | |
2652 | * wish to. Caller must run wait_for_completion() on its own. | |
2653 | */ | |
2654 | int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector) | |
2655 | { | |
165125e1 | 2656 | struct request_queue *q; |
1da177e4 LT |
2657 | |
2658 | if (bdev->bd_disk == NULL) | |
2659 | return -ENXIO; | |
2660 | ||
2661 | q = bdev_get_queue(bdev); | |
2662 | if (!q) | |
2663 | return -ENXIO; | |
2664 | if (!q->issue_flush_fn) | |
2665 | return -EOPNOTSUPP; | |
2666 | ||
2667 | return q->issue_flush_fn(q, bdev->bd_disk, error_sector); | |
2668 | } | |
2669 | ||
2670 | EXPORT_SYMBOL(blkdev_issue_flush); | |
2671 | ||
93d17d3d | 2672 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io) |
1da177e4 LT |
2673 | { |
2674 | int rw = rq_data_dir(rq); | |
2675 | ||
2676 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
2677 | return; | |
2678 | ||
d72d904a | 2679 | if (!new_io) { |
a362357b | 2680 | __disk_stat_inc(rq->rq_disk, merges[rw]); |
d72d904a | 2681 | } else { |
1da177e4 LT |
2682 | disk_round_stats(rq->rq_disk); |
2683 | rq->rq_disk->in_flight++; | |
2684 | } | |
2685 | } | |
2686 | ||
2687 | /* | |
2688 | * add-request adds a request to the linked list. | |
2689 | * queue lock is held and interrupts disabled, as we muck with the | |
2690 | * request queue list. | |
2691 | */ | |
165125e1 | 2692 | static inline void add_request(struct request_queue * q, struct request * req) |
1da177e4 LT |
2693 | { |
2694 | drive_stat_acct(req, req->nr_sectors, 1); | |
2695 | ||
1da177e4 LT |
2696 | /* |
2697 | * elevator indicated where it wants this request to be | |
2698 | * inserted at elevator_merge time | |
2699 | */ | |
2700 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
2701 | } | |
2702 | ||
2703 | /* | |
2704 | * disk_round_stats() - Round off the performance stats on a struct | |
2705 | * disk_stats. | |
2706 | * | |
2707 | * The average IO queue length and utilisation statistics are maintained | |
2708 | * by observing the current state of the queue length and the amount of | |
2709 | * time it has been in this state for. | |
2710 | * | |
2711 | * Normally, that accounting is done on IO completion, but that can result | |
2712 | * in more than a second's worth of IO being accounted for within any one | |
2713 | * second, leading to >100% utilisation. To deal with that, we call this | |
2714 | * function to do a round-off before returning the results when reading | |
2715 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
2716 | * the current jiffies and restarts the counters again. | |
2717 | */ | |
2718 | void disk_round_stats(struct gendisk *disk) | |
2719 | { | |
2720 | unsigned long now = jiffies; | |
2721 | ||
b2982649 CK |
2722 | if (now == disk->stamp) |
2723 | return; | |
1da177e4 | 2724 | |
20e5c81f CK |
2725 | if (disk->in_flight) { |
2726 | __disk_stat_add(disk, time_in_queue, | |
2727 | disk->in_flight * (now - disk->stamp)); | |
2728 | __disk_stat_add(disk, io_ticks, (now - disk->stamp)); | |
2729 | } | |
1da177e4 | 2730 | disk->stamp = now; |
1da177e4 LT |
2731 | } |
2732 | ||
3eaf840e JNN |
2733 | EXPORT_SYMBOL_GPL(disk_round_stats); |
2734 | ||
1da177e4 LT |
2735 | /* |
2736 | * queue lock must be held | |
2737 | */ | |
165125e1 | 2738 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 2739 | { |
1da177e4 LT |
2740 | if (unlikely(!q)) |
2741 | return; | |
2742 | if (unlikely(--req->ref_count)) | |
2743 | return; | |
2744 | ||
8922e16c TH |
2745 | elv_completed_request(q, req); |
2746 | ||
1da177e4 LT |
2747 | /* |
2748 | * Request may not have originated from ll_rw_blk. if not, | |
2749 | * it didn't come out of our reserved rq pools | |
2750 | */ | |
49171e5c | 2751 | if (req->cmd_flags & REQ_ALLOCED) { |
1da177e4 | 2752 | int rw = rq_data_dir(req); |
4aff5e23 | 2753 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 2754 | |
1da177e4 | 2755 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 2756 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
2757 | |
2758 | blk_free_request(q, req); | |
cb98fc8b | 2759 | freed_request(q, rw, priv); |
1da177e4 LT |
2760 | } |
2761 | } | |
2762 | ||
6e39b69e MC |
2763 | EXPORT_SYMBOL_GPL(__blk_put_request); |
2764 | ||
1da177e4 LT |
2765 | void blk_put_request(struct request *req) |
2766 | { | |
8922e16c | 2767 | unsigned long flags; |
165125e1 | 2768 | struct request_queue *q = req->q; |
8922e16c | 2769 | |
1da177e4 | 2770 | /* |
8922e16c TH |
2771 | * Gee, IDE calls in w/ NULL q. Fix IDE and remove the |
2772 | * following if (q) test. | |
1da177e4 | 2773 | */ |
8922e16c | 2774 | if (q) { |
1da177e4 LT |
2775 | spin_lock_irqsave(q->queue_lock, flags); |
2776 | __blk_put_request(q, req); | |
2777 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2778 | } | |
2779 | } | |
2780 | ||
2781 | EXPORT_SYMBOL(blk_put_request); | |
2782 | ||
2783 | /** | |
2784 | * blk_end_sync_rq - executes a completion event on a request | |
2785 | * @rq: request to complete | |
fddfdeaf | 2786 | * @error: end io status of the request |
1da177e4 | 2787 | */ |
8ffdc655 | 2788 | void blk_end_sync_rq(struct request *rq, int error) |
1da177e4 | 2789 | { |
c00895ab | 2790 | struct completion *waiting = rq->end_io_data; |
1da177e4 | 2791 | |
c00895ab | 2792 | rq->end_io_data = NULL; |
1da177e4 LT |
2793 | __blk_put_request(rq->q, rq); |
2794 | ||
2795 | /* | |
2796 | * complete last, if this is a stack request the process (and thus | |
2797 | * the rq pointer) could be invalid right after this complete() | |
2798 | */ | |
2799 | complete(waiting); | |
2800 | } | |
2801 | EXPORT_SYMBOL(blk_end_sync_rq); | |
2802 | ||
1da177e4 LT |
2803 | /* |
2804 | * Has to be called with the request spinlock acquired | |
2805 | */ | |
165125e1 | 2806 | static int attempt_merge(struct request_queue *q, struct request *req, |
1da177e4 LT |
2807 | struct request *next) |
2808 | { | |
2809 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
2810 | return 0; | |
2811 | ||
2812 | /* | |
d6e05edc | 2813 | * not contiguous |
1da177e4 LT |
2814 | */ |
2815 | if (req->sector + req->nr_sectors != next->sector) | |
2816 | return 0; | |
2817 | ||
2818 | if (rq_data_dir(req) != rq_data_dir(next) | |
2819 | || req->rq_disk != next->rq_disk | |
c00895ab | 2820 | || next->special) |
1da177e4 LT |
2821 | return 0; |
2822 | ||
2823 | /* | |
2824 | * If we are allowed to merge, then append bio list | |
2825 | * from next to rq and release next. merge_requests_fn | |
2826 | * will have updated segment counts, update sector | |
2827 | * counts here. | |
2828 | */ | |
1aa4f24f | 2829 | if (!ll_merge_requests_fn(q, req, next)) |
1da177e4 LT |
2830 | return 0; |
2831 | ||
2832 | /* | |
2833 | * At this point we have either done a back merge | |
2834 | * or front merge. We need the smaller start_time of | |
2835 | * the merged requests to be the current request | |
2836 | * for accounting purposes. | |
2837 | */ | |
2838 | if (time_after(req->start_time, next->start_time)) | |
2839 | req->start_time = next->start_time; | |
2840 | ||
2841 | req->biotail->bi_next = next->bio; | |
2842 | req->biotail = next->biotail; | |
2843 | ||
2844 | req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors; | |
2845 | ||
2846 | elv_merge_requests(q, req, next); | |
2847 | ||
2848 | if (req->rq_disk) { | |
2849 | disk_round_stats(req->rq_disk); | |
2850 | req->rq_disk->in_flight--; | |
2851 | } | |
2852 | ||
22e2c507 JA |
2853 | req->ioprio = ioprio_best(req->ioprio, next->ioprio); |
2854 | ||
1da177e4 LT |
2855 | __blk_put_request(q, next); |
2856 | return 1; | |
2857 | } | |
2858 | ||
165125e1 JA |
2859 | static inline int attempt_back_merge(struct request_queue *q, |
2860 | struct request *rq) | |
1da177e4 LT |
2861 | { |
2862 | struct request *next = elv_latter_request(q, rq); | |
2863 | ||
2864 | if (next) | |
2865 | return attempt_merge(q, rq, next); | |
2866 | ||
2867 | return 0; | |
2868 | } | |
2869 | ||
165125e1 JA |
2870 | static inline int attempt_front_merge(struct request_queue *q, |
2871 | struct request *rq) | |
1da177e4 LT |
2872 | { |
2873 | struct request *prev = elv_former_request(q, rq); | |
2874 | ||
2875 | if (prev) | |
2876 | return attempt_merge(q, prev, rq); | |
2877 | ||
2878 | return 0; | |
2879 | } | |
2880 | ||
52d9e675 TH |
2881 | static void init_request_from_bio(struct request *req, struct bio *bio) |
2882 | { | |
4aff5e23 | 2883 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
2884 | |
2885 | /* | |
2886 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
2887 | */ | |
2888 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
4aff5e23 | 2889 | req->cmd_flags |= REQ_FAILFAST; |
52d9e675 TH |
2890 | |
2891 | /* | |
2892 | * REQ_BARRIER implies no merging, but lets make it explicit | |
2893 | */ | |
2894 | if (unlikely(bio_barrier(bio))) | |
4aff5e23 | 2895 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); |
52d9e675 | 2896 | |
b31dc66a | 2897 | if (bio_sync(bio)) |
4aff5e23 | 2898 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
2899 | if (bio_rw_meta(bio)) |
2900 | req->cmd_flags |= REQ_RW_META; | |
b31dc66a | 2901 | |
52d9e675 TH |
2902 | req->errors = 0; |
2903 | req->hard_sector = req->sector = bio->bi_sector; | |
2904 | req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio); | |
2905 | req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio); | |
2906 | req->nr_phys_segments = bio_phys_segments(req->q, bio); | |
2907 | req->nr_hw_segments = bio_hw_segments(req->q, bio); | |
2908 | req->buffer = bio_data(bio); /* see ->buffer comment above */ | |
52d9e675 TH |
2909 | req->bio = req->biotail = bio; |
2910 | req->ioprio = bio_prio(bio); | |
2911 | req->rq_disk = bio->bi_bdev->bd_disk; | |
2912 | req->start_time = jiffies; | |
2913 | } | |
2914 | ||
165125e1 | 2915 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 2916 | { |
450991bc | 2917 | struct request *req; |
51da90fc JA |
2918 | int el_ret, nr_sectors, barrier, err; |
2919 | const unsigned short prio = bio_prio(bio); | |
2920 | const int sync = bio_sync(bio); | |
7749a8d4 | 2921 | int rw_flags; |
1da177e4 | 2922 | |
1da177e4 | 2923 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
2924 | |
2925 | /* | |
2926 | * low level driver can indicate that it wants pages above a | |
2927 | * certain limit bounced to low memory (ie for highmem, or even | |
2928 | * ISA dma in theory) | |
2929 | */ | |
2930 | blk_queue_bounce(q, &bio); | |
2931 | ||
1da177e4 | 2932 | barrier = bio_barrier(bio); |
797e7dbb | 2933 | if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) { |
1da177e4 LT |
2934 | err = -EOPNOTSUPP; |
2935 | goto end_io; | |
2936 | } | |
2937 | ||
1da177e4 LT |
2938 | spin_lock_irq(q->queue_lock); |
2939 | ||
450991bc | 2940 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
2941 | goto get_rq; |
2942 | ||
2943 | el_ret = elv_merge(q, &req, bio); | |
2944 | switch (el_ret) { | |
2945 | case ELEVATOR_BACK_MERGE: | |
2946 | BUG_ON(!rq_mergeable(req)); | |
2947 | ||
1aa4f24f | 2948 | if (!ll_back_merge_fn(q, req, bio)) |
1da177e4 LT |
2949 | break; |
2950 | ||
2056a782 JA |
2951 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); |
2952 | ||
1da177e4 LT |
2953 | req->biotail->bi_next = bio; |
2954 | req->biotail = bio; | |
2955 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
22e2c507 | 2956 | req->ioprio = ioprio_best(req->ioprio, prio); |
1da177e4 LT |
2957 | drive_stat_acct(req, nr_sectors, 0); |
2958 | if (!attempt_back_merge(q, req)) | |
2e662b65 | 2959 | elv_merged_request(q, req, el_ret); |
1da177e4 LT |
2960 | goto out; |
2961 | ||
2962 | case ELEVATOR_FRONT_MERGE: | |
2963 | BUG_ON(!rq_mergeable(req)); | |
2964 | ||
1aa4f24f | 2965 | if (!ll_front_merge_fn(q, req, bio)) |
1da177e4 LT |
2966 | break; |
2967 | ||
2056a782 JA |
2968 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); |
2969 | ||
1da177e4 LT |
2970 | bio->bi_next = req->bio; |
2971 | req->bio = bio; | |
2972 | ||
2973 | /* | |
2974 | * may not be valid. if the low level driver said | |
2975 | * it didn't need a bounce buffer then it better | |
2976 | * not touch req->buffer either... | |
2977 | */ | |
2978 | req->buffer = bio_data(bio); | |
51da90fc JA |
2979 | req->current_nr_sectors = bio_cur_sectors(bio); |
2980 | req->hard_cur_sectors = req->current_nr_sectors; | |
2981 | req->sector = req->hard_sector = bio->bi_sector; | |
1da177e4 | 2982 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; |
22e2c507 | 2983 | req->ioprio = ioprio_best(req->ioprio, prio); |
1da177e4 LT |
2984 | drive_stat_acct(req, nr_sectors, 0); |
2985 | if (!attempt_front_merge(q, req)) | |
2e662b65 | 2986 | elv_merged_request(q, req, el_ret); |
1da177e4 LT |
2987 | goto out; |
2988 | ||
450991bc | 2989 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ |
1da177e4 | 2990 | default: |
450991bc | 2991 | ; |
1da177e4 LT |
2992 | } |
2993 | ||
450991bc | 2994 | get_rq: |
7749a8d4 JA |
2995 | /* |
2996 | * This sync check and mask will be re-done in init_request_from_bio(), | |
2997 | * but we need to set it earlier to expose the sync flag to the | |
2998 | * rq allocator and io schedulers. | |
2999 | */ | |
3000 | rw_flags = bio_data_dir(bio); | |
3001 | if (sync) | |
3002 | rw_flags |= REQ_RW_SYNC; | |
3003 | ||
1da177e4 | 3004 | /* |
450991bc | 3005 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 3006 | * Returns with the queue unlocked. |
450991bc | 3007 | */ |
7749a8d4 | 3008 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 3009 | |
450991bc NP |
3010 | /* |
3011 | * After dropping the lock and possibly sleeping here, our request | |
3012 | * may now be mergeable after it had proven unmergeable (above). | |
3013 | * We don't worry about that case for efficiency. It won't happen | |
3014 | * often, and the elevators are able to handle it. | |
1da177e4 | 3015 | */ |
52d9e675 | 3016 | init_request_from_bio(req, bio); |
1da177e4 | 3017 | |
450991bc NP |
3018 | spin_lock_irq(q->queue_lock); |
3019 | if (elv_queue_empty(q)) | |
3020 | blk_plug_device(q); | |
1da177e4 LT |
3021 | add_request(q, req); |
3022 | out: | |
4a534f93 | 3023 | if (sync) |
1da177e4 LT |
3024 | __generic_unplug_device(q); |
3025 | ||
3026 | spin_unlock_irq(q->queue_lock); | |
3027 | return 0; | |
3028 | ||
3029 | end_io: | |
3030 | bio_endio(bio, nr_sectors << 9, err); | |
3031 | return 0; | |
3032 | } | |
3033 | ||
3034 | /* | |
3035 | * If bio->bi_dev is a partition, remap the location | |
3036 | */ | |
3037 | static inline void blk_partition_remap(struct bio *bio) | |
3038 | { | |
3039 | struct block_device *bdev = bio->bi_bdev; | |
3040 | ||
3041 | if (bdev != bdev->bd_contains) { | |
3042 | struct hd_struct *p = bdev->bd_part; | |
a362357b JA |
3043 | const int rw = bio_data_dir(bio); |
3044 | ||
3045 | p->sectors[rw] += bio_sectors(bio); | |
3046 | p->ios[rw]++; | |
1da177e4 | 3047 | |
1da177e4 LT |
3048 | bio->bi_sector += p->start_sect; |
3049 | bio->bi_bdev = bdev->bd_contains; | |
3050 | } | |
3051 | } | |
3052 | ||
1da177e4 LT |
3053 | static void handle_bad_sector(struct bio *bio) |
3054 | { | |
3055 | char b[BDEVNAME_SIZE]; | |
3056 | ||
3057 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
3058 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
3059 | bdevname(bio->bi_bdev, b), | |
3060 | bio->bi_rw, | |
3061 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
3062 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
3063 | ||
3064 | set_bit(BIO_EOF, &bio->bi_flags); | |
3065 | } | |
3066 | ||
c17bb495 AM |
3067 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
3068 | ||
3069 | static DECLARE_FAULT_ATTR(fail_make_request); | |
3070 | ||
3071 | static int __init setup_fail_make_request(char *str) | |
3072 | { | |
3073 | return setup_fault_attr(&fail_make_request, str); | |
3074 | } | |
3075 | __setup("fail_make_request=", setup_fail_make_request); | |
3076 | ||
3077 | static int should_fail_request(struct bio *bio) | |
3078 | { | |
3079 | if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) || | |
3080 | (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail)) | |
3081 | return should_fail(&fail_make_request, bio->bi_size); | |
3082 | ||
3083 | return 0; | |
3084 | } | |
3085 | ||
3086 | static int __init fail_make_request_debugfs(void) | |
3087 | { | |
3088 | return init_fault_attr_dentries(&fail_make_request, | |
3089 | "fail_make_request"); | |
3090 | } | |
3091 | ||
3092 | late_initcall(fail_make_request_debugfs); | |
3093 | ||
3094 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
3095 | ||
3096 | static inline int should_fail_request(struct bio *bio) | |
3097 | { | |
3098 | return 0; | |
3099 | } | |
3100 | ||
3101 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
3102 | ||
1da177e4 LT |
3103 | /** |
3104 | * generic_make_request: hand a buffer to its device driver for I/O | |
3105 | * @bio: The bio describing the location in memory and on the device. | |
3106 | * | |
3107 | * generic_make_request() is used to make I/O requests of block | |
3108 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
3109 | * to be done. | |
3110 | * | |
3111 | * generic_make_request() does not return any status. The | |
3112 | * success/failure status of the request, along with notification of | |
3113 | * completion, is delivered asynchronously through the bio->bi_end_io | |
3114 | * function described (one day) else where. | |
3115 | * | |
3116 | * The caller of generic_make_request must make sure that bi_io_vec | |
3117 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
3118 | * set to describe the device address, and the | |
3119 | * bi_end_io and optionally bi_private are set to describe how | |
3120 | * completion notification should be signaled. | |
3121 | * | |
3122 | * generic_make_request and the drivers it calls may use bi_next if this | |
3123 | * bio happens to be merged with someone else, and may change bi_dev and | |
3124 | * bi_sector for remaps as it sees fit. So the values of these fields | |
3125 | * should NOT be depended on after the call to generic_make_request. | |
3126 | */ | |
d89d8796 | 3127 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 3128 | { |
165125e1 | 3129 | struct request_queue *q; |
1da177e4 | 3130 | sector_t maxsector; |
5ddfe969 | 3131 | sector_t old_sector; |
1da177e4 | 3132 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 3133 | dev_t old_dev; |
1da177e4 LT |
3134 | |
3135 | might_sleep(); | |
3136 | /* Test device or partition size, when known. */ | |
3137 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
3138 | if (maxsector) { | |
3139 | sector_t sector = bio->bi_sector; | |
3140 | ||
3141 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
3142 | /* | |
3143 | * This may well happen - the kernel calls bread() | |
3144 | * without checking the size of the device, e.g., when | |
3145 | * mounting a device. | |
3146 | */ | |
3147 | handle_bad_sector(bio); | |
3148 | goto end_io; | |
3149 | } | |
3150 | } | |
3151 | ||
3152 | /* | |
3153 | * Resolve the mapping until finished. (drivers are | |
3154 | * still free to implement/resolve their own stacking | |
3155 | * by explicitly returning 0) | |
3156 | * | |
3157 | * NOTE: we don't repeat the blk_size check for each new device. | |
3158 | * Stacking drivers are expected to know what they are doing. | |
3159 | */ | |
5ddfe969 | 3160 | old_sector = -1; |
2056a782 | 3161 | old_dev = 0; |
1da177e4 LT |
3162 | do { |
3163 | char b[BDEVNAME_SIZE]; | |
3164 | ||
3165 | q = bdev_get_queue(bio->bi_bdev); | |
3166 | if (!q) { | |
3167 | printk(KERN_ERR | |
3168 | "generic_make_request: Trying to access " | |
3169 | "nonexistent block-device %s (%Lu)\n", | |
3170 | bdevname(bio->bi_bdev, b), | |
3171 | (long long) bio->bi_sector); | |
3172 | end_io: | |
3173 | bio_endio(bio, bio->bi_size, -EIO); | |
3174 | break; | |
3175 | } | |
3176 | ||
3177 | if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) { | |
3178 | printk("bio too big device %s (%u > %u)\n", | |
3179 | bdevname(bio->bi_bdev, b), | |
3180 | bio_sectors(bio), | |
3181 | q->max_hw_sectors); | |
3182 | goto end_io; | |
3183 | } | |
3184 | ||
fde6ad22 | 3185 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
3186 | goto end_io; |
3187 | ||
c17bb495 AM |
3188 | if (should_fail_request(bio)) |
3189 | goto end_io; | |
3190 | ||
1da177e4 LT |
3191 | /* |
3192 | * If this device has partitions, remap block n | |
3193 | * of partition p to block n+start(p) of the disk. | |
3194 | */ | |
3195 | blk_partition_remap(bio); | |
3196 | ||
5ddfe969 | 3197 | if (old_sector != -1) |
2056a782 | 3198 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 3199 | old_sector); |
2056a782 JA |
3200 | |
3201 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); | |
3202 | ||
5ddfe969 | 3203 | old_sector = bio->bi_sector; |
2056a782 JA |
3204 | old_dev = bio->bi_bdev->bd_dev; |
3205 | ||
5ddfe969 N |
3206 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; |
3207 | if (maxsector) { | |
3208 | sector_t sector = bio->bi_sector; | |
3209 | ||
df66b855 AM |
3210 | if (maxsector < nr_sectors || |
3211 | maxsector - nr_sectors < sector) { | |
5ddfe969 | 3212 | /* |
df66b855 AM |
3213 | * This may well happen - partitions are not |
3214 | * checked to make sure they are within the size | |
3215 | * of the whole device. | |
5ddfe969 N |
3216 | */ |
3217 | handle_bad_sector(bio); | |
3218 | goto end_io; | |
3219 | } | |
3220 | } | |
3221 | ||
1da177e4 LT |
3222 | ret = q->make_request_fn(q, bio); |
3223 | } while (ret); | |
3224 | } | |
3225 | ||
d89d8796 NB |
3226 | /* |
3227 | * We only want one ->make_request_fn to be active at a time, | |
3228 | * else stack usage with stacked devices could be a problem. | |
3229 | * So use current->bio_{list,tail} to keep a list of requests | |
3230 | * submited by a make_request_fn function. | |
3231 | * current->bio_tail is also used as a flag to say if | |
3232 | * generic_make_request is currently active in this task or not. | |
3233 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
3234 | * then a make_request is active, and new requests should be added | |
3235 | * at the tail | |
3236 | */ | |
3237 | void generic_make_request(struct bio *bio) | |
3238 | { | |
3239 | if (current->bio_tail) { | |
3240 | /* make_request is active */ | |
3241 | *(current->bio_tail) = bio; | |
3242 | bio->bi_next = NULL; | |
3243 | current->bio_tail = &bio->bi_next; | |
3244 | return; | |
3245 | } | |
3246 | /* following loop may be a bit non-obvious, and so deserves some | |
3247 | * explanation. | |
3248 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
3249 | * ensure that) so we have a list with a single bio. | |
3250 | * We pretend that we have just taken it off a longer list, so | |
3251 | * we assign bio_list to the next (which is NULL) and bio_tail | |
3252 | * to &bio_list, thus initialising the bio_list of new bios to be | |
3253 | * added. __generic_make_request may indeed add some more bios | |
3254 | * through a recursive call to generic_make_request. If it | |
3255 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
3256 | * from the top. In this case we really did just take the bio | |
3257 | * of the top of the list (no pretending) and so fixup bio_list and | |
3258 | * bio_tail or bi_next, and call into __generic_make_request again. | |
3259 | * | |
3260 | * The loop was structured like this to make only one call to | |
3261 | * __generic_make_request (which is important as it is large and | |
3262 | * inlined) and to keep the structure simple. | |
3263 | */ | |
3264 | BUG_ON(bio->bi_next); | |
3265 | do { | |
3266 | current->bio_list = bio->bi_next; | |
3267 | if (bio->bi_next == NULL) | |
3268 | current->bio_tail = ¤t->bio_list; | |
3269 | else | |
3270 | bio->bi_next = NULL; | |
3271 | __generic_make_request(bio); | |
3272 | bio = current->bio_list; | |
3273 | } while (bio); | |
3274 | current->bio_tail = NULL; /* deactivate */ | |
3275 | } | |
3276 | ||
1da177e4 LT |
3277 | EXPORT_SYMBOL(generic_make_request); |
3278 | ||
3279 | /** | |
3280 | * submit_bio: submit a bio to the block device layer for I/O | |
3281 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
3282 | * @bio: The &struct bio which describes the I/O | |
3283 | * | |
3284 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
3285 | * uses that function to do most of the work. Both are fairly rough | |
3286 | * interfaces, @bio must be presetup and ready for I/O. | |
3287 | * | |
3288 | */ | |
3289 | void submit_bio(int rw, struct bio *bio) | |
3290 | { | |
3291 | int count = bio_sectors(bio); | |
3292 | ||
3293 | BIO_BUG_ON(!bio->bi_size); | |
3294 | BIO_BUG_ON(!bio->bi_io_vec); | |
22e2c507 | 3295 | bio->bi_rw |= rw; |
faccbd4b | 3296 | if (rw & WRITE) { |
f8891e5e | 3297 | count_vm_events(PGPGOUT, count); |
faccbd4b AM |
3298 | } else { |
3299 | task_io_account_read(bio->bi_size); | |
f8891e5e | 3300 | count_vm_events(PGPGIN, count); |
faccbd4b | 3301 | } |
1da177e4 LT |
3302 | |
3303 | if (unlikely(block_dump)) { | |
3304 | char b[BDEVNAME_SIZE]; | |
3305 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
3306 | current->comm, current->pid, | |
3307 | (rw & WRITE) ? "WRITE" : "READ", | |
3308 | (unsigned long long)bio->bi_sector, | |
3309 | bdevname(bio->bi_bdev,b)); | |
3310 | } | |
3311 | ||
3312 | generic_make_request(bio); | |
3313 | } | |
3314 | ||
3315 | EXPORT_SYMBOL(submit_bio); | |
3316 | ||
93d17d3d | 3317 | static void blk_recalc_rq_segments(struct request *rq) |
1da177e4 LT |
3318 | { |
3319 | struct bio *bio, *prevbio = NULL; | |
3320 | int nr_phys_segs, nr_hw_segs; | |
3321 | unsigned int phys_size, hw_size; | |
165125e1 | 3322 | struct request_queue *q = rq->q; |
1da177e4 LT |
3323 | |
3324 | if (!rq->bio) | |
3325 | return; | |
3326 | ||
3327 | phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0; | |
3328 | rq_for_each_bio(bio, rq) { | |
3329 | /* Force bio hw/phys segs to be recalculated. */ | |
3330 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3331 | ||
3332 | nr_phys_segs += bio_phys_segments(q, bio); | |
3333 | nr_hw_segs += bio_hw_segments(q, bio); | |
3334 | if (prevbio) { | |
3335 | int pseg = phys_size + prevbio->bi_size + bio->bi_size; | |
3336 | int hseg = hw_size + prevbio->bi_size + bio->bi_size; | |
3337 | ||
3338 | if (blk_phys_contig_segment(q, prevbio, bio) && | |
3339 | pseg <= q->max_segment_size) { | |
3340 | nr_phys_segs--; | |
3341 | phys_size += prevbio->bi_size + bio->bi_size; | |
3342 | } else | |
3343 | phys_size = 0; | |
3344 | ||
3345 | if (blk_hw_contig_segment(q, prevbio, bio) && | |
3346 | hseg <= q->max_segment_size) { | |
3347 | nr_hw_segs--; | |
3348 | hw_size += prevbio->bi_size + bio->bi_size; | |
3349 | } else | |
3350 | hw_size = 0; | |
3351 | } | |
3352 | prevbio = bio; | |
3353 | } | |
3354 | ||
3355 | rq->nr_phys_segments = nr_phys_segs; | |
3356 | rq->nr_hw_segments = nr_hw_segs; | |
3357 | } | |
3358 | ||
93d17d3d | 3359 | static void blk_recalc_rq_sectors(struct request *rq, int nsect) |
1da177e4 LT |
3360 | { |
3361 | if (blk_fs_request(rq)) { | |
3362 | rq->hard_sector += nsect; | |
3363 | rq->hard_nr_sectors -= nsect; | |
3364 | ||
3365 | /* | |
3366 | * Move the I/O submission pointers ahead if required. | |
3367 | */ | |
3368 | if ((rq->nr_sectors >= rq->hard_nr_sectors) && | |
3369 | (rq->sector <= rq->hard_sector)) { | |
3370 | rq->sector = rq->hard_sector; | |
3371 | rq->nr_sectors = rq->hard_nr_sectors; | |
3372 | rq->hard_cur_sectors = bio_cur_sectors(rq->bio); | |
3373 | rq->current_nr_sectors = rq->hard_cur_sectors; | |
3374 | rq->buffer = bio_data(rq->bio); | |
3375 | } | |
3376 | ||
3377 | /* | |
3378 | * if total number of sectors is less than the first segment | |
3379 | * size, something has gone terribly wrong | |
3380 | */ | |
3381 | if (rq->nr_sectors < rq->current_nr_sectors) { | |
3382 | printk("blk: request botched\n"); | |
3383 | rq->nr_sectors = rq->current_nr_sectors; | |
3384 | } | |
3385 | } | |
3386 | } | |
3387 | ||
3388 | static int __end_that_request_first(struct request *req, int uptodate, | |
3389 | int nr_bytes) | |
3390 | { | |
3391 | int total_bytes, bio_nbytes, error, next_idx = 0; | |
3392 | struct bio *bio; | |
3393 | ||
2056a782 JA |
3394 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); |
3395 | ||
1da177e4 LT |
3396 | /* |
3397 | * extend uptodate bool to allow < 0 value to be direct io error | |
3398 | */ | |
3399 | error = 0; | |
3400 | if (end_io_error(uptodate)) | |
3401 | error = !uptodate ? -EIO : uptodate; | |
3402 | ||
3403 | /* | |
3404 | * for a REQ_BLOCK_PC request, we want to carry any eventual | |
3405 | * sense key with us all the way through | |
3406 | */ | |
3407 | if (!blk_pc_request(req)) | |
3408 | req->errors = 0; | |
3409 | ||
3410 | if (!uptodate) { | |
4aff5e23 | 3411 | if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET)) |
1da177e4 LT |
3412 | printk("end_request: I/O error, dev %s, sector %llu\n", |
3413 | req->rq_disk ? req->rq_disk->disk_name : "?", | |
3414 | (unsigned long long)req->sector); | |
3415 | } | |
3416 | ||
d72d904a | 3417 | if (blk_fs_request(req) && req->rq_disk) { |
a362357b JA |
3418 | const int rw = rq_data_dir(req); |
3419 | ||
53e86061 | 3420 | disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9); |
d72d904a JA |
3421 | } |
3422 | ||
1da177e4 LT |
3423 | total_bytes = bio_nbytes = 0; |
3424 | while ((bio = req->bio) != NULL) { | |
3425 | int nbytes; | |
3426 | ||
3427 | if (nr_bytes >= bio->bi_size) { | |
3428 | req->bio = bio->bi_next; | |
3429 | nbytes = bio->bi_size; | |
797e7dbb TH |
3430 | if (!ordered_bio_endio(req, bio, nbytes, error)) |
3431 | bio_endio(bio, nbytes, error); | |
1da177e4 LT |
3432 | next_idx = 0; |
3433 | bio_nbytes = 0; | |
3434 | } else { | |
3435 | int idx = bio->bi_idx + next_idx; | |
3436 | ||
3437 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
3438 | blk_dump_rq_flags(req, "__end_that"); | |
3439 | printk("%s: bio idx %d >= vcnt %d\n", | |
3440 | __FUNCTION__, | |
3441 | bio->bi_idx, bio->bi_vcnt); | |
3442 | break; | |
3443 | } | |
3444 | ||
3445 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
3446 | BIO_BUG_ON(nbytes > bio->bi_size); | |
3447 | ||
3448 | /* | |
3449 | * not a complete bvec done | |
3450 | */ | |
3451 | if (unlikely(nbytes > nr_bytes)) { | |
3452 | bio_nbytes += nr_bytes; | |
3453 | total_bytes += nr_bytes; | |
3454 | break; | |
3455 | } | |
3456 | ||
3457 | /* | |
3458 | * advance to the next vector | |
3459 | */ | |
3460 | next_idx++; | |
3461 | bio_nbytes += nbytes; | |
3462 | } | |
3463 | ||
3464 | total_bytes += nbytes; | |
3465 | nr_bytes -= nbytes; | |
3466 | ||
3467 | if ((bio = req->bio)) { | |
3468 | /* | |
3469 | * end more in this run, or just return 'not-done' | |
3470 | */ | |
3471 | if (unlikely(nr_bytes <= 0)) | |
3472 | break; | |
3473 | } | |
3474 | } | |
3475 | ||
3476 | /* | |
3477 | * completely done | |
3478 | */ | |
3479 | if (!req->bio) | |
3480 | return 0; | |
3481 | ||
3482 | /* | |
3483 | * if the request wasn't completed, update state | |
3484 | */ | |
3485 | if (bio_nbytes) { | |
797e7dbb TH |
3486 | if (!ordered_bio_endio(req, bio, bio_nbytes, error)) |
3487 | bio_endio(bio, bio_nbytes, error); | |
1da177e4 LT |
3488 | bio->bi_idx += next_idx; |
3489 | bio_iovec(bio)->bv_offset += nr_bytes; | |
3490 | bio_iovec(bio)->bv_len -= nr_bytes; | |
3491 | } | |
3492 | ||
3493 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
3494 | blk_recalc_rq_segments(req); | |
3495 | return 1; | |
3496 | } | |
3497 | ||
3498 | /** | |
3499 | * end_that_request_first - end I/O on a request | |
3500 | * @req: the request being processed | |
3501 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3502 | * @nr_sectors: number of sectors to end I/O on | |
3503 | * | |
3504 | * Description: | |
3505 | * Ends I/O on a number of sectors attached to @req, and sets it up | |
3506 | * for the next range of segments (if any) in the cluster. | |
3507 | * | |
3508 | * Return: | |
3509 | * 0 - we are done with this request, call end_that_request_last() | |
3510 | * 1 - still buffers pending for this request | |
3511 | **/ | |
3512 | int end_that_request_first(struct request *req, int uptodate, int nr_sectors) | |
3513 | { | |
3514 | return __end_that_request_first(req, uptodate, nr_sectors << 9); | |
3515 | } | |
3516 | ||
3517 | EXPORT_SYMBOL(end_that_request_first); | |
3518 | ||
3519 | /** | |
3520 | * end_that_request_chunk - end I/O on a request | |
3521 | * @req: the request being processed | |
3522 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3523 | * @nr_bytes: number of bytes to complete | |
3524 | * | |
3525 | * Description: | |
3526 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
3527 | * for the next range of segments (if any). Like end_that_request_first(), | |
3528 | * but deals with bytes instead of sectors. | |
3529 | * | |
3530 | * Return: | |
3531 | * 0 - we are done with this request, call end_that_request_last() | |
3532 | * 1 - still buffers pending for this request | |
3533 | **/ | |
3534 | int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes) | |
3535 | { | |
3536 | return __end_that_request_first(req, uptodate, nr_bytes); | |
3537 | } | |
3538 | ||
3539 | EXPORT_SYMBOL(end_that_request_chunk); | |
3540 | ||
ff856bad JA |
3541 | /* |
3542 | * splice the completion data to a local structure and hand off to | |
3543 | * process_completion_queue() to complete the requests | |
3544 | */ | |
3545 | static void blk_done_softirq(struct softirq_action *h) | |
3546 | { | |
626ab0e6 | 3547 | struct list_head *cpu_list, local_list; |
ff856bad JA |
3548 | |
3549 | local_irq_disable(); | |
3550 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
626ab0e6 | 3551 | list_replace_init(cpu_list, &local_list); |
ff856bad JA |
3552 | local_irq_enable(); |
3553 | ||
3554 | while (!list_empty(&local_list)) { | |
3555 | struct request *rq = list_entry(local_list.next, struct request, donelist); | |
3556 | ||
3557 | list_del_init(&rq->donelist); | |
3558 | rq->q->softirq_done_fn(rq); | |
3559 | } | |
3560 | } | |
3561 | ||
ff856bad JA |
3562 | static int blk_cpu_notify(struct notifier_block *self, unsigned long action, |
3563 | void *hcpu) | |
3564 | { | |
3565 | /* | |
3566 | * If a CPU goes away, splice its entries to the current CPU | |
3567 | * and trigger a run of the softirq | |
3568 | */ | |
8bb78442 | 3569 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
ff856bad JA |
3570 | int cpu = (unsigned long) hcpu; |
3571 | ||
3572 | local_irq_disable(); | |
3573 | list_splice_init(&per_cpu(blk_cpu_done, cpu), | |
3574 | &__get_cpu_var(blk_cpu_done)); | |
3575 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
3576 | local_irq_enable(); | |
3577 | } | |
3578 | ||
3579 | return NOTIFY_OK; | |
3580 | } | |
3581 | ||
3582 | ||
054cc8a2 | 3583 | static struct notifier_block __devinitdata blk_cpu_notifier = { |
ff856bad JA |
3584 | .notifier_call = blk_cpu_notify, |
3585 | }; | |
3586 | ||
ff856bad JA |
3587 | /** |
3588 | * blk_complete_request - end I/O on a request | |
3589 | * @req: the request being processed | |
3590 | * | |
3591 | * Description: | |
3592 | * Ends all I/O on a request. It does not handle partial completions, | |
d6e05edc | 3593 | * unless the driver actually implements this in its completion callback |
ff856bad JA |
3594 | * through requeueing. Theh actual completion happens out-of-order, |
3595 | * through a softirq handler. The user must have registered a completion | |
3596 | * callback through blk_queue_softirq_done(). | |
3597 | **/ | |
3598 | ||
3599 | void blk_complete_request(struct request *req) | |
3600 | { | |
3601 | struct list_head *cpu_list; | |
3602 | unsigned long flags; | |
3603 | ||
3604 | BUG_ON(!req->q->softirq_done_fn); | |
3605 | ||
3606 | local_irq_save(flags); | |
3607 | ||
3608 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
3609 | list_add_tail(&req->donelist, cpu_list); | |
3610 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
3611 | ||
3612 | local_irq_restore(flags); | |
3613 | } | |
3614 | ||
3615 | EXPORT_SYMBOL(blk_complete_request); | |
3616 | ||
1da177e4 LT |
3617 | /* |
3618 | * queue lock must be held | |
3619 | */ | |
8ffdc655 | 3620 | void end_that_request_last(struct request *req, int uptodate) |
1da177e4 LT |
3621 | { |
3622 | struct gendisk *disk = req->rq_disk; | |
8ffdc655 TH |
3623 | int error; |
3624 | ||
3625 | /* | |
3626 | * extend uptodate bool to allow < 0 value to be direct io error | |
3627 | */ | |
3628 | error = 0; | |
3629 | if (end_io_error(uptodate)) | |
3630 | error = !uptodate ? -EIO : uptodate; | |
1da177e4 LT |
3631 | |
3632 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
3633 | laptop_io_completion(); | |
3634 | ||
fd0ff8aa JA |
3635 | /* |
3636 | * Account IO completion. bar_rq isn't accounted as a normal | |
3637 | * IO on queueing nor completion. Accounting the containing | |
3638 | * request is enough. | |
3639 | */ | |
3640 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
1da177e4 | 3641 | unsigned long duration = jiffies - req->start_time; |
a362357b JA |
3642 | const int rw = rq_data_dir(req); |
3643 | ||
3644 | __disk_stat_inc(disk, ios[rw]); | |
3645 | __disk_stat_add(disk, ticks[rw], duration); | |
1da177e4 LT |
3646 | disk_round_stats(disk); |
3647 | disk->in_flight--; | |
3648 | } | |
3649 | if (req->end_io) | |
8ffdc655 | 3650 | req->end_io(req, error); |
1da177e4 LT |
3651 | else |
3652 | __blk_put_request(req->q, req); | |
3653 | } | |
3654 | ||
3655 | EXPORT_SYMBOL(end_that_request_last); | |
3656 | ||
3657 | void end_request(struct request *req, int uptodate) | |
3658 | { | |
3659 | if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) { | |
3660 | add_disk_randomness(req->rq_disk); | |
3661 | blkdev_dequeue_request(req); | |
8ffdc655 | 3662 | end_that_request_last(req, uptodate); |
1da177e4 LT |
3663 | } |
3664 | } | |
3665 | ||
3666 | EXPORT_SYMBOL(end_request); | |
3667 | ||
165125e1 JA |
3668 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
3669 | struct bio *bio) | |
1da177e4 | 3670 | { |
4aff5e23 JA |
3671 | /* first two bits are identical in rq->cmd_flags and bio->bi_rw */ |
3672 | rq->cmd_flags |= (bio->bi_rw & 3); | |
1da177e4 LT |
3673 | |
3674 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
3675 | rq->nr_hw_segments = bio_hw_segments(q, bio); | |
3676 | rq->current_nr_sectors = bio_cur_sectors(bio); | |
3677 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
3678 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
3679 | rq->buffer = bio_data(bio); | |
0e75f906 | 3680 | rq->data_len = bio->bi_size; |
1da177e4 LT |
3681 | |
3682 | rq->bio = rq->biotail = bio; | |
3683 | } | |
3684 | ||
3685 | EXPORT_SYMBOL(blk_rq_bio_prep); | |
3686 | ||
3687 | int kblockd_schedule_work(struct work_struct *work) | |
3688 | { | |
3689 | return queue_work(kblockd_workqueue, work); | |
3690 | } | |
3691 | ||
3692 | EXPORT_SYMBOL(kblockd_schedule_work); | |
3693 | ||
19a75d83 | 3694 | void kblockd_flush_work(struct work_struct *work) |
1da177e4 | 3695 | { |
28e53bdd | 3696 | cancel_work_sync(work); |
1da177e4 | 3697 | } |
19a75d83 | 3698 | EXPORT_SYMBOL(kblockd_flush_work); |
1da177e4 LT |
3699 | |
3700 | int __init blk_dev_init(void) | |
3701 | { | |
ff856bad JA |
3702 | int i; |
3703 | ||
1da177e4 LT |
3704 | kblockd_workqueue = create_workqueue("kblockd"); |
3705 | if (!kblockd_workqueue) | |
3706 | panic("Failed to create kblockd\n"); | |
3707 | ||
3708 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 3709 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 LT |
3710 | |
3711 | requestq_cachep = kmem_cache_create("blkdev_queue", | |
165125e1 | 3712 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 LT |
3713 | |
3714 | iocontext_cachep = kmem_cache_create("blkdev_ioc", | |
20c2df83 | 3715 | sizeof(struct io_context), 0, SLAB_PANIC, NULL); |
1da177e4 | 3716 | |
0a945022 | 3717 | for_each_possible_cpu(i) |
ff856bad JA |
3718 | INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i)); |
3719 | ||
3720 | open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL); | |
5a67e4c5 | 3721 | register_hotcpu_notifier(&blk_cpu_notifier); |
ff856bad | 3722 | |
f772b3d9 VT |
3723 | blk_max_low_pfn = max_low_pfn - 1; |
3724 | blk_max_pfn = max_pfn - 1; | |
1da177e4 LT |
3725 | |
3726 | return 0; | |
3727 | } | |
3728 | ||
3729 | /* | |
3730 | * IO Context helper functions | |
3731 | */ | |
3732 | void put_io_context(struct io_context *ioc) | |
3733 | { | |
3734 | if (ioc == NULL) | |
3735 | return; | |
3736 | ||
3737 | BUG_ON(atomic_read(&ioc->refcount) == 0); | |
3738 | ||
3739 | if (atomic_dec_and_test(&ioc->refcount)) { | |
e2d74ac0 JA |
3740 | struct cfq_io_context *cic; |
3741 | ||
334e94de | 3742 | rcu_read_lock(); |
1da177e4 LT |
3743 | if (ioc->aic && ioc->aic->dtor) |
3744 | ioc->aic->dtor(ioc->aic); | |
e2d74ac0 | 3745 | if (ioc->cic_root.rb_node != NULL) { |
7143dd4b JA |
3746 | struct rb_node *n = rb_first(&ioc->cic_root); |
3747 | ||
3748 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
e2d74ac0 JA |
3749 | cic->dtor(ioc); |
3750 | } | |
334e94de | 3751 | rcu_read_unlock(); |
1da177e4 LT |
3752 | |
3753 | kmem_cache_free(iocontext_cachep, ioc); | |
3754 | } | |
3755 | } | |
3756 | EXPORT_SYMBOL(put_io_context); | |
3757 | ||
3758 | /* Called by the exitting task */ | |
3759 | void exit_io_context(void) | |
3760 | { | |
1da177e4 | 3761 | struct io_context *ioc; |
e2d74ac0 | 3762 | struct cfq_io_context *cic; |
1da177e4 | 3763 | |
22e2c507 | 3764 | task_lock(current); |
1da177e4 LT |
3765 | ioc = current->io_context; |
3766 | current->io_context = NULL; | |
22e2c507 | 3767 | task_unlock(current); |
1da177e4 | 3768 | |
25034d7a | 3769 | ioc->task = NULL; |
1da177e4 LT |
3770 | if (ioc->aic && ioc->aic->exit) |
3771 | ioc->aic->exit(ioc->aic); | |
e2d74ac0 JA |
3772 | if (ioc->cic_root.rb_node != NULL) { |
3773 | cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node); | |
3774 | cic->exit(ioc); | |
3775 | } | |
25034d7a | 3776 | |
1da177e4 LT |
3777 | put_io_context(ioc); |
3778 | } | |
3779 | ||
3780 | /* | |
3781 | * If the current task has no IO context then create one and initialise it. | |
fb3cc432 | 3782 | * Otherwise, return its existing IO context. |
1da177e4 | 3783 | * |
fb3cc432 NP |
3784 | * This returned IO context doesn't have a specifically elevated refcount, |
3785 | * but since the current task itself holds a reference, the context can be | |
3786 | * used in general code, so long as it stays within `current` context. | |
1da177e4 | 3787 | */ |
b5deef90 | 3788 | static struct io_context *current_io_context(gfp_t gfp_flags, int node) |
1da177e4 LT |
3789 | { |
3790 | struct task_struct *tsk = current; | |
1da177e4 LT |
3791 | struct io_context *ret; |
3792 | ||
1da177e4 | 3793 | ret = tsk->io_context; |
fb3cc432 NP |
3794 | if (likely(ret)) |
3795 | return ret; | |
1da177e4 | 3796 | |
b5deef90 | 3797 | ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node); |
1da177e4 LT |
3798 | if (ret) { |
3799 | atomic_set(&ret->refcount, 1); | |
22e2c507 | 3800 | ret->task = current; |
fc46379d | 3801 | ret->ioprio_changed = 0; |
1da177e4 LT |
3802 | ret->last_waited = jiffies; /* doesn't matter... */ |
3803 | ret->nr_batch_requests = 0; /* because this is 0 */ | |
3804 | ret->aic = NULL; | |
e2d74ac0 | 3805 | ret->cic_root.rb_node = NULL; |
4e521c27 | 3806 | ret->ioc_data = NULL; |
9f83e45e ON |
3807 | /* make sure set_task_ioprio() sees the settings above */ |
3808 | smp_wmb(); | |
fb3cc432 NP |
3809 | tsk->io_context = ret; |
3810 | } | |
1da177e4 | 3811 | |
fb3cc432 NP |
3812 | return ret; |
3813 | } | |
1da177e4 | 3814 | |
fb3cc432 NP |
3815 | /* |
3816 | * If the current task has no IO context then create one and initialise it. | |
3817 | * If it does have a context, take a ref on it. | |
3818 | * | |
3819 | * This is always called in the context of the task which submitted the I/O. | |
3820 | */ | |
b5deef90 | 3821 | struct io_context *get_io_context(gfp_t gfp_flags, int node) |
fb3cc432 NP |
3822 | { |
3823 | struct io_context *ret; | |
b5deef90 | 3824 | ret = current_io_context(gfp_flags, node); |
fb3cc432 | 3825 | if (likely(ret)) |
1da177e4 | 3826 | atomic_inc(&ret->refcount); |
1da177e4 LT |
3827 | return ret; |
3828 | } | |
3829 | EXPORT_SYMBOL(get_io_context); | |
3830 | ||
3831 | void copy_io_context(struct io_context **pdst, struct io_context **psrc) | |
3832 | { | |
3833 | struct io_context *src = *psrc; | |
3834 | struct io_context *dst = *pdst; | |
3835 | ||
3836 | if (src) { | |
3837 | BUG_ON(atomic_read(&src->refcount) == 0); | |
3838 | atomic_inc(&src->refcount); | |
3839 | put_io_context(dst); | |
3840 | *pdst = src; | |
3841 | } | |
3842 | } | |
3843 | EXPORT_SYMBOL(copy_io_context); | |
3844 | ||
3845 | void swap_io_context(struct io_context **ioc1, struct io_context **ioc2) | |
3846 | { | |
3847 | struct io_context *temp; | |
3848 | temp = *ioc1; | |
3849 | *ioc1 = *ioc2; | |
3850 | *ioc2 = temp; | |
3851 | } | |
3852 | EXPORT_SYMBOL(swap_io_context); | |
3853 | ||
3854 | /* | |
3855 | * sysfs parts below | |
3856 | */ | |
3857 | struct queue_sysfs_entry { | |
3858 | struct attribute attr; | |
3859 | ssize_t (*show)(struct request_queue *, char *); | |
3860 | ssize_t (*store)(struct request_queue *, const char *, size_t); | |
3861 | }; | |
3862 | ||
3863 | static ssize_t | |
3864 | queue_var_show(unsigned int var, char *page) | |
3865 | { | |
3866 | return sprintf(page, "%d\n", var); | |
3867 | } | |
3868 | ||
3869 | static ssize_t | |
3870 | queue_var_store(unsigned long *var, const char *page, size_t count) | |
3871 | { | |
3872 | char *p = (char *) page; | |
3873 | ||
3874 | *var = simple_strtoul(p, &p, 10); | |
3875 | return count; | |
3876 | } | |
3877 | ||
3878 | static ssize_t queue_requests_show(struct request_queue *q, char *page) | |
3879 | { | |
3880 | return queue_var_show(q->nr_requests, (page)); | |
3881 | } | |
3882 | ||
3883 | static ssize_t | |
3884 | queue_requests_store(struct request_queue *q, const char *page, size_t count) | |
3885 | { | |
3886 | struct request_list *rl = &q->rq; | |
c981ff9f AV |
3887 | unsigned long nr; |
3888 | int ret = queue_var_store(&nr, page, count); | |
3889 | if (nr < BLKDEV_MIN_RQ) | |
3890 | nr = BLKDEV_MIN_RQ; | |
1da177e4 | 3891 | |
c981ff9f AV |
3892 | spin_lock_irq(q->queue_lock); |
3893 | q->nr_requests = nr; | |
1da177e4 LT |
3894 | blk_queue_congestion_threshold(q); |
3895 | ||
3896 | if (rl->count[READ] >= queue_congestion_on_threshold(q)) | |
79e2de4b | 3897 | blk_set_queue_congested(q, READ); |
1da177e4 | 3898 | else if (rl->count[READ] < queue_congestion_off_threshold(q)) |
79e2de4b | 3899 | blk_clear_queue_congested(q, READ); |
1da177e4 LT |
3900 | |
3901 | if (rl->count[WRITE] >= queue_congestion_on_threshold(q)) | |
79e2de4b | 3902 | blk_set_queue_congested(q, WRITE); |
1da177e4 | 3903 | else if (rl->count[WRITE] < queue_congestion_off_threshold(q)) |
79e2de4b | 3904 | blk_clear_queue_congested(q, WRITE); |
1da177e4 LT |
3905 | |
3906 | if (rl->count[READ] >= q->nr_requests) { | |
3907 | blk_set_queue_full(q, READ); | |
3908 | } else if (rl->count[READ]+1 <= q->nr_requests) { | |
3909 | blk_clear_queue_full(q, READ); | |
3910 | wake_up(&rl->wait[READ]); | |
3911 | } | |
3912 | ||
3913 | if (rl->count[WRITE] >= q->nr_requests) { | |
3914 | blk_set_queue_full(q, WRITE); | |
3915 | } else if (rl->count[WRITE]+1 <= q->nr_requests) { | |
3916 | blk_clear_queue_full(q, WRITE); | |
3917 | wake_up(&rl->wait[WRITE]); | |
3918 | } | |
c981ff9f | 3919 | spin_unlock_irq(q->queue_lock); |
1da177e4 LT |
3920 | return ret; |
3921 | } | |
3922 | ||
3923 | static ssize_t queue_ra_show(struct request_queue *q, char *page) | |
3924 | { | |
3925 | int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | |
3926 | ||
3927 | return queue_var_show(ra_kb, (page)); | |
3928 | } | |
3929 | ||
3930 | static ssize_t | |
3931 | queue_ra_store(struct request_queue *q, const char *page, size_t count) | |
3932 | { | |
3933 | unsigned long ra_kb; | |
3934 | ssize_t ret = queue_var_store(&ra_kb, page, count); | |
3935 | ||
3936 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
3937 | q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); |
3938 | spin_unlock_irq(q->queue_lock); | |
3939 | ||
3940 | return ret; | |
3941 | } | |
3942 | ||
3943 | static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) | |
3944 | { | |
3945 | int max_sectors_kb = q->max_sectors >> 1; | |
3946 | ||
3947 | return queue_var_show(max_sectors_kb, (page)); | |
3948 | } | |
3949 | ||
3950 | static ssize_t | |
3951 | queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) | |
3952 | { | |
3953 | unsigned long max_sectors_kb, | |
3954 | max_hw_sectors_kb = q->max_hw_sectors >> 1, | |
3955 | page_kb = 1 << (PAGE_CACHE_SHIFT - 10); | |
3956 | ssize_t ret = queue_var_store(&max_sectors_kb, page, count); | |
3957 | int ra_kb; | |
3958 | ||
3959 | if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) | |
3960 | return -EINVAL; | |
3961 | /* | |
3962 | * Take the queue lock to update the readahead and max_sectors | |
3963 | * values synchronously: | |
3964 | */ | |
3965 | spin_lock_irq(q->queue_lock); | |
3966 | /* | |
3967 | * Trim readahead window as well, if necessary: | |
3968 | */ | |
3969 | ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | |
3970 | if (ra_kb > max_sectors_kb) | |
3971 | q->backing_dev_info.ra_pages = | |
3972 | max_sectors_kb >> (PAGE_CACHE_SHIFT - 10); | |
3973 | ||
3974 | q->max_sectors = max_sectors_kb << 1; | |
3975 | spin_unlock_irq(q->queue_lock); | |
3976 | ||
3977 | return ret; | |
3978 | } | |
3979 | ||
3980 | static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) | |
3981 | { | |
3982 | int max_hw_sectors_kb = q->max_hw_sectors >> 1; | |
3983 | ||
3984 | return queue_var_show(max_hw_sectors_kb, (page)); | |
3985 | } | |
3986 | ||
3987 | ||
3988 | static struct queue_sysfs_entry queue_requests_entry = { | |
3989 | .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, | |
3990 | .show = queue_requests_show, | |
3991 | .store = queue_requests_store, | |
3992 | }; | |
3993 | ||
3994 | static struct queue_sysfs_entry queue_ra_entry = { | |
3995 | .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, | |
3996 | .show = queue_ra_show, | |
3997 | .store = queue_ra_store, | |
3998 | }; | |
3999 | ||
4000 | static struct queue_sysfs_entry queue_max_sectors_entry = { | |
4001 | .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, | |
4002 | .show = queue_max_sectors_show, | |
4003 | .store = queue_max_sectors_store, | |
4004 | }; | |
4005 | ||
4006 | static struct queue_sysfs_entry queue_max_hw_sectors_entry = { | |
4007 | .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, | |
4008 | .show = queue_max_hw_sectors_show, | |
4009 | }; | |
4010 | ||
4011 | static struct queue_sysfs_entry queue_iosched_entry = { | |
4012 | .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, | |
4013 | .show = elv_iosched_show, | |
4014 | .store = elv_iosched_store, | |
4015 | }; | |
4016 | ||
4017 | static struct attribute *default_attrs[] = { | |
4018 | &queue_requests_entry.attr, | |
4019 | &queue_ra_entry.attr, | |
4020 | &queue_max_hw_sectors_entry.attr, | |
4021 | &queue_max_sectors_entry.attr, | |
4022 | &queue_iosched_entry.attr, | |
4023 | NULL, | |
4024 | }; | |
4025 | ||
4026 | #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) | |
4027 | ||
4028 | static ssize_t | |
4029 | queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4030 | { | |
4031 | struct queue_sysfs_entry *entry = to_queue(attr); | |
165125e1 JA |
4032 | struct request_queue *q = |
4033 | container_of(kobj, struct request_queue, kobj); | |
483f4afc | 4034 | ssize_t res; |
1da177e4 | 4035 | |
1da177e4 | 4036 | if (!entry->show) |
6c1852a0 | 4037 | return -EIO; |
483f4afc AV |
4038 | mutex_lock(&q->sysfs_lock); |
4039 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { | |
4040 | mutex_unlock(&q->sysfs_lock); | |
4041 | return -ENOENT; | |
4042 | } | |
4043 | res = entry->show(q, page); | |
4044 | mutex_unlock(&q->sysfs_lock); | |
4045 | return res; | |
1da177e4 LT |
4046 | } |
4047 | ||
4048 | static ssize_t | |
4049 | queue_attr_store(struct kobject *kobj, struct attribute *attr, | |
4050 | const char *page, size_t length) | |
4051 | { | |
4052 | struct queue_sysfs_entry *entry = to_queue(attr); | |
165125e1 | 4053 | struct request_queue *q = container_of(kobj, struct request_queue, kobj); |
483f4afc AV |
4054 | |
4055 | ssize_t res; | |
1da177e4 | 4056 | |
1da177e4 | 4057 | if (!entry->store) |
6c1852a0 | 4058 | return -EIO; |
483f4afc AV |
4059 | mutex_lock(&q->sysfs_lock); |
4060 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { | |
4061 | mutex_unlock(&q->sysfs_lock); | |
4062 | return -ENOENT; | |
4063 | } | |
4064 | res = entry->store(q, page, length); | |
4065 | mutex_unlock(&q->sysfs_lock); | |
4066 | return res; | |
1da177e4 LT |
4067 | } |
4068 | ||
4069 | static struct sysfs_ops queue_sysfs_ops = { | |
4070 | .show = queue_attr_show, | |
4071 | .store = queue_attr_store, | |
4072 | }; | |
4073 | ||
93d17d3d | 4074 | static struct kobj_type queue_ktype = { |
1da177e4 LT |
4075 | .sysfs_ops = &queue_sysfs_ops, |
4076 | .default_attrs = default_attrs, | |
483f4afc | 4077 | .release = blk_release_queue, |
1da177e4 LT |
4078 | }; |
4079 | ||
4080 | int blk_register_queue(struct gendisk *disk) | |
4081 | { | |
4082 | int ret; | |
4083 | ||
165125e1 | 4084 | struct request_queue *q = disk->queue; |
1da177e4 LT |
4085 | |
4086 | if (!q || !q->request_fn) | |
4087 | return -ENXIO; | |
4088 | ||
4089 | q->kobj.parent = kobject_get(&disk->kobj); | |
1da177e4 | 4090 | |
483f4afc | 4091 | ret = kobject_add(&q->kobj); |
1da177e4 LT |
4092 | if (ret < 0) |
4093 | return ret; | |
4094 | ||
483f4afc AV |
4095 | kobject_uevent(&q->kobj, KOBJ_ADD); |
4096 | ||
1da177e4 LT |
4097 | ret = elv_register_queue(q); |
4098 | if (ret) { | |
483f4afc AV |
4099 | kobject_uevent(&q->kobj, KOBJ_REMOVE); |
4100 | kobject_del(&q->kobj); | |
1da177e4 LT |
4101 | return ret; |
4102 | } | |
4103 | ||
4104 | return 0; | |
4105 | } | |
4106 | ||
4107 | void blk_unregister_queue(struct gendisk *disk) | |
4108 | { | |
165125e1 | 4109 | struct request_queue *q = disk->queue; |
1da177e4 LT |
4110 | |
4111 | if (q && q->request_fn) { | |
4112 | elv_unregister_queue(q); | |
4113 | ||
483f4afc AV |
4114 | kobject_uevent(&q->kobj, KOBJ_REMOVE); |
4115 | kobject_del(&q->kobj); | |
1da177e4 LT |
4116 | kobject_put(&disk->kobj); |
4117 | } | |
4118 | } |