]>
Commit | Line | Data |
---|---|---|
1 | /* | |
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 | */ | |
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> | |
28 | #include <linux/task_io_accounting_ops.h> | |
29 | #include <linux/interrupt.h> | |
30 | #include <linux/cpu.h> | |
31 | #include <linux/blktrace_api.h> | |
32 | #include <linux/fault-inject.h> | |
33 | #include <linux/scatterlist.h> | |
34 | ||
35 | /* | |
36 | * for max sense size | |
37 | */ | |
38 | #include <scsi/scsi_cmnd.h> | |
39 | ||
40 | static void blk_unplug_work(struct work_struct *work); | |
41 | static void blk_unplug_timeout(unsigned long data); | |
42 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io); | |
43 | static void init_request_from_bio(struct request *req, struct bio *bio); | |
44 | static int __make_request(struct request_queue *q, struct bio *bio); | |
45 | static struct io_context *current_io_context(gfp_t gfp_flags, int node); | |
46 | static void blk_recalc_rq_segments(struct request *rq); | |
47 | static void blk_rq_bio_prep(struct request_queue *q, struct request *rq, | |
48 | struct bio *bio); | |
49 | ||
50 | /* | |
51 | * For the allocated request tables | |
52 | */ | |
53 | static struct kmem_cache *request_cachep; | |
54 | ||
55 | /* | |
56 | * For queue allocation | |
57 | */ | |
58 | static struct kmem_cache *requestq_cachep; | |
59 | ||
60 | /* | |
61 | * For io context allocations | |
62 | */ | |
63 | static struct kmem_cache *iocontext_cachep; | |
64 | ||
65 | /* | |
66 | * Controlling structure to kblockd | |
67 | */ | |
68 | static struct workqueue_struct *kblockd_workqueue; | |
69 | ||
70 | unsigned long blk_max_low_pfn, blk_max_pfn; | |
71 | ||
72 | EXPORT_SYMBOL(blk_max_low_pfn); | |
73 | EXPORT_SYMBOL(blk_max_pfn); | |
74 | ||
75 | static DEFINE_PER_CPU(struct list_head, blk_cpu_done); | |
76 | ||
77 | /* Amount of time in which a process may batch requests */ | |
78 | #define BLK_BATCH_TIME (HZ/50UL) | |
79 | ||
80 | /* Number of requests a "batching" process may submit */ | |
81 | #define BLK_BATCH_REQ 32 | |
82 | ||
83 | /* | |
84 | * Return the threshold (number of used requests) at which the queue is | |
85 | * considered to be congested. It include a little hysteresis to keep the | |
86 | * context switch rate down. | |
87 | */ | |
88 | static inline int queue_congestion_on_threshold(struct request_queue *q) | |
89 | { | |
90 | return q->nr_congestion_on; | |
91 | } | |
92 | ||
93 | /* | |
94 | * The threshold at which a queue is considered to be uncongested | |
95 | */ | |
96 | static inline int queue_congestion_off_threshold(struct request_queue *q) | |
97 | { | |
98 | return q->nr_congestion_off; | |
99 | } | |
100 | ||
101 | static void blk_queue_congestion_threshold(struct request_queue *q) | |
102 | { | |
103 | int nr; | |
104 | ||
105 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
106 | if (nr > q->nr_requests) | |
107 | nr = q->nr_requests; | |
108 | q->nr_congestion_on = nr; | |
109 | ||
110 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
111 | if (nr < 1) | |
112 | nr = 1; | |
113 | q->nr_congestion_off = nr; | |
114 | } | |
115 | ||
116 | /** | |
117 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
118 | * @bdev: device | |
119 | * | |
120 | * Locates the passed device's request queue and returns the address of its | |
121 | * backing_dev_info | |
122 | * | |
123 | * Will return NULL if the request queue cannot be located. | |
124 | */ | |
125 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
126 | { | |
127 | struct backing_dev_info *ret = NULL; | |
128 | struct request_queue *q = bdev_get_queue(bdev); | |
129 | ||
130 | if (q) | |
131 | ret = &q->backing_dev_info; | |
132 | return ret; | |
133 | } | |
134 | EXPORT_SYMBOL(blk_get_backing_dev_info); | |
135 | ||
136 | /** | |
137 | * blk_queue_prep_rq - set a prepare_request function for queue | |
138 | * @q: queue | |
139 | * @pfn: prepare_request function | |
140 | * | |
141 | * It's possible for a queue to register a prepare_request callback which | |
142 | * is invoked before the request is handed to the request_fn. The goal of | |
143 | * the function is to prepare a request for I/O, it can be used to build a | |
144 | * cdb from the request data for instance. | |
145 | * | |
146 | */ | |
147 | void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) | |
148 | { | |
149 | q->prep_rq_fn = pfn; | |
150 | } | |
151 | ||
152 | EXPORT_SYMBOL(blk_queue_prep_rq); | |
153 | ||
154 | /** | |
155 | * blk_queue_merge_bvec - set a merge_bvec function for queue | |
156 | * @q: queue | |
157 | * @mbfn: merge_bvec_fn | |
158 | * | |
159 | * Usually queues have static limitations on the max sectors or segments that | |
160 | * we can put in a request. Stacking drivers may have some settings that | |
161 | * are dynamic, and thus we have to query the queue whether it is ok to | |
162 | * add a new bio_vec to a bio at a given offset or not. If the block device | |
163 | * has such limitations, it needs to register a merge_bvec_fn to control | |
164 | * the size of bio's sent to it. Note that a block device *must* allow a | |
165 | * single page to be added to an empty bio. The block device driver may want | |
166 | * to use the bio_split() function to deal with these bio's. By default | |
167 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | |
168 | * honored. | |
169 | */ | |
170 | void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) | |
171 | { | |
172 | q->merge_bvec_fn = mbfn; | |
173 | } | |
174 | ||
175 | EXPORT_SYMBOL(blk_queue_merge_bvec); | |
176 | ||
177 | void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) | |
178 | { | |
179 | q->softirq_done_fn = fn; | |
180 | } | |
181 | ||
182 | EXPORT_SYMBOL(blk_queue_softirq_done); | |
183 | ||
184 | /** | |
185 | * blk_queue_make_request - define an alternate make_request function for a device | |
186 | * @q: the request queue for the device to be affected | |
187 | * @mfn: the alternate make_request function | |
188 | * | |
189 | * Description: | |
190 | * The normal way for &struct bios to be passed to a device | |
191 | * driver is for them to be collected into requests on a request | |
192 | * queue, and then to allow the device driver to select requests | |
193 | * off that queue when it is ready. This works well for many block | |
194 | * devices. However some block devices (typically virtual devices | |
195 | * such as md or lvm) do not benefit from the processing on the | |
196 | * request queue, and are served best by having the requests passed | |
197 | * directly to them. This can be achieved by providing a function | |
198 | * to blk_queue_make_request(). | |
199 | * | |
200 | * Caveat: | |
201 | * The driver that does this *must* be able to deal appropriately | |
202 | * with buffers in "highmemory". This can be accomplished by either calling | |
203 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
204 | * blk_queue_bounce() to create a buffer in normal memory. | |
205 | **/ | |
206 | void blk_queue_make_request(struct request_queue * q, make_request_fn * mfn) | |
207 | { | |
208 | /* | |
209 | * set defaults | |
210 | */ | |
211 | q->nr_requests = BLKDEV_MAX_RQ; | |
212 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
213 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
214 | q->make_request_fn = mfn; | |
215 | q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
216 | q->backing_dev_info.state = 0; | |
217 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
218 | blk_queue_max_sectors(q, SAFE_MAX_SECTORS); | |
219 | blk_queue_hardsect_size(q, 512); | |
220 | blk_queue_dma_alignment(q, 511); | |
221 | blk_queue_congestion_threshold(q); | |
222 | q->nr_batching = BLK_BATCH_REQ; | |
223 | ||
224 | q->unplug_thresh = 4; /* hmm */ | |
225 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ | |
226 | if (q->unplug_delay == 0) | |
227 | q->unplug_delay = 1; | |
228 | ||
229 | INIT_WORK(&q->unplug_work, blk_unplug_work); | |
230 | ||
231 | q->unplug_timer.function = blk_unplug_timeout; | |
232 | q->unplug_timer.data = (unsigned long)q; | |
233 | ||
234 | /* | |
235 | * by default assume old behaviour and bounce for any highmem page | |
236 | */ | |
237 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
238 | } | |
239 | ||
240 | EXPORT_SYMBOL(blk_queue_make_request); | |
241 | ||
242 | static void rq_init(struct request_queue *q, struct request *rq) | |
243 | { | |
244 | INIT_LIST_HEAD(&rq->queuelist); | |
245 | INIT_LIST_HEAD(&rq->donelist); | |
246 | ||
247 | rq->errors = 0; | |
248 | rq->bio = rq->biotail = NULL; | |
249 | INIT_HLIST_NODE(&rq->hash); | |
250 | RB_CLEAR_NODE(&rq->rb_node); | |
251 | rq->ioprio = 0; | |
252 | rq->buffer = NULL; | |
253 | rq->ref_count = 1; | |
254 | rq->q = q; | |
255 | rq->special = NULL; | |
256 | rq->data_len = 0; | |
257 | rq->data = NULL; | |
258 | rq->nr_phys_segments = 0; | |
259 | rq->sense = NULL; | |
260 | rq->end_io = NULL; | |
261 | rq->end_io_data = NULL; | |
262 | rq->completion_data = NULL; | |
263 | rq->next_rq = NULL; | |
264 | } | |
265 | ||
266 | /** | |
267 | * blk_queue_ordered - does this queue support ordered writes | |
268 | * @q: the request queue | |
269 | * @ordered: one of QUEUE_ORDERED_* | |
270 | * @prepare_flush_fn: rq setup helper for cache flush ordered writes | |
271 | * | |
272 | * Description: | |
273 | * For journalled file systems, doing ordered writes on a commit | |
274 | * block instead of explicitly doing wait_on_buffer (which is bad | |
275 | * for performance) can be a big win. Block drivers supporting this | |
276 | * feature should call this function and indicate so. | |
277 | * | |
278 | **/ | |
279 | int blk_queue_ordered(struct request_queue *q, unsigned ordered, | |
280 | prepare_flush_fn *prepare_flush_fn) | |
281 | { | |
282 | if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) && | |
283 | prepare_flush_fn == NULL) { | |
284 | printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n"); | |
285 | return -EINVAL; | |
286 | } | |
287 | ||
288 | if (ordered != QUEUE_ORDERED_NONE && | |
289 | ordered != QUEUE_ORDERED_DRAIN && | |
290 | ordered != QUEUE_ORDERED_DRAIN_FLUSH && | |
291 | ordered != QUEUE_ORDERED_DRAIN_FUA && | |
292 | ordered != QUEUE_ORDERED_TAG && | |
293 | ordered != QUEUE_ORDERED_TAG_FLUSH && | |
294 | ordered != QUEUE_ORDERED_TAG_FUA) { | |
295 | printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered); | |
296 | return -EINVAL; | |
297 | } | |
298 | ||
299 | q->ordered = ordered; | |
300 | q->next_ordered = ordered; | |
301 | q->prepare_flush_fn = prepare_flush_fn; | |
302 | ||
303 | return 0; | |
304 | } | |
305 | ||
306 | EXPORT_SYMBOL(blk_queue_ordered); | |
307 | ||
308 | /* | |
309 | * Cache flushing for ordered writes handling | |
310 | */ | |
311 | inline unsigned blk_ordered_cur_seq(struct request_queue *q) | |
312 | { | |
313 | if (!q->ordseq) | |
314 | return 0; | |
315 | return 1 << ffz(q->ordseq); | |
316 | } | |
317 | ||
318 | unsigned blk_ordered_req_seq(struct request *rq) | |
319 | { | |
320 | struct request_queue *q = rq->q; | |
321 | ||
322 | BUG_ON(q->ordseq == 0); | |
323 | ||
324 | if (rq == &q->pre_flush_rq) | |
325 | return QUEUE_ORDSEQ_PREFLUSH; | |
326 | if (rq == &q->bar_rq) | |
327 | return QUEUE_ORDSEQ_BAR; | |
328 | if (rq == &q->post_flush_rq) | |
329 | return QUEUE_ORDSEQ_POSTFLUSH; | |
330 | ||
331 | /* | |
332 | * !fs requests don't need to follow barrier ordering. Always | |
333 | * put them at the front. This fixes the following deadlock. | |
334 | * | |
335 | * http://thread.gmane.org/gmane.linux.kernel/537473 | |
336 | */ | |
337 | if (!blk_fs_request(rq)) | |
338 | return QUEUE_ORDSEQ_DRAIN; | |
339 | ||
340 | if ((rq->cmd_flags & REQ_ORDERED_COLOR) == | |
341 | (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR)) | |
342 | return QUEUE_ORDSEQ_DRAIN; | |
343 | else | |
344 | return QUEUE_ORDSEQ_DONE; | |
345 | } | |
346 | ||
347 | void blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error) | |
348 | { | |
349 | struct request *rq; | |
350 | int uptodate; | |
351 | ||
352 | if (error && !q->orderr) | |
353 | q->orderr = error; | |
354 | ||
355 | BUG_ON(q->ordseq & seq); | |
356 | q->ordseq |= seq; | |
357 | ||
358 | if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE) | |
359 | return; | |
360 | ||
361 | /* | |
362 | * Okay, sequence complete. | |
363 | */ | |
364 | uptodate = 1; | |
365 | if (q->orderr) | |
366 | uptodate = q->orderr; | |
367 | ||
368 | q->ordseq = 0; | |
369 | rq = q->orig_bar_rq; | |
370 | ||
371 | end_that_request_first(rq, uptodate, rq->hard_nr_sectors); | |
372 | end_that_request_last(rq, uptodate); | |
373 | } | |
374 | ||
375 | static void pre_flush_end_io(struct request *rq, int error) | |
376 | { | |
377 | elv_completed_request(rq->q, rq); | |
378 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error); | |
379 | } | |
380 | ||
381 | static void bar_end_io(struct request *rq, int error) | |
382 | { | |
383 | elv_completed_request(rq->q, rq); | |
384 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error); | |
385 | } | |
386 | ||
387 | static void post_flush_end_io(struct request *rq, int error) | |
388 | { | |
389 | elv_completed_request(rq->q, rq); | |
390 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error); | |
391 | } | |
392 | ||
393 | static void queue_flush(struct request_queue *q, unsigned which) | |
394 | { | |
395 | struct request *rq; | |
396 | rq_end_io_fn *end_io; | |
397 | ||
398 | if (which == QUEUE_ORDERED_PREFLUSH) { | |
399 | rq = &q->pre_flush_rq; | |
400 | end_io = pre_flush_end_io; | |
401 | } else { | |
402 | rq = &q->post_flush_rq; | |
403 | end_io = post_flush_end_io; | |
404 | } | |
405 | ||
406 | rq->cmd_flags = REQ_HARDBARRIER; | |
407 | rq_init(q, rq); | |
408 | rq->elevator_private = NULL; | |
409 | rq->elevator_private2 = NULL; | |
410 | rq->rq_disk = q->bar_rq.rq_disk; | |
411 | rq->end_io = end_io; | |
412 | q->prepare_flush_fn(q, rq); | |
413 | ||
414 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); | |
415 | } | |
416 | ||
417 | static inline struct request *start_ordered(struct request_queue *q, | |
418 | struct request *rq) | |
419 | { | |
420 | q->orderr = 0; | |
421 | q->ordered = q->next_ordered; | |
422 | q->ordseq |= QUEUE_ORDSEQ_STARTED; | |
423 | ||
424 | /* | |
425 | * Prep proxy barrier request. | |
426 | */ | |
427 | blkdev_dequeue_request(rq); | |
428 | q->orig_bar_rq = rq; | |
429 | rq = &q->bar_rq; | |
430 | rq->cmd_flags = 0; | |
431 | rq_init(q, rq); | |
432 | if (bio_data_dir(q->orig_bar_rq->bio) == WRITE) | |
433 | rq->cmd_flags |= REQ_RW; | |
434 | if (q->ordered & QUEUE_ORDERED_FUA) | |
435 | rq->cmd_flags |= REQ_FUA; | |
436 | rq->elevator_private = NULL; | |
437 | rq->elevator_private2 = NULL; | |
438 | init_request_from_bio(rq, q->orig_bar_rq->bio); | |
439 | rq->end_io = bar_end_io; | |
440 | ||
441 | /* | |
442 | * Queue ordered sequence. As we stack them at the head, we | |
443 | * need to queue in reverse order. Note that we rely on that | |
444 | * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs | |
445 | * request gets inbetween ordered sequence. If this request is | |
446 | * an empty barrier, we don't need to do a postflush ever since | |
447 | * there will be no data written between the pre and post flush. | |
448 | * Hence a single flush will suffice. | |
449 | */ | |
450 | if ((q->ordered & QUEUE_ORDERED_POSTFLUSH) && !blk_empty_barrier(rq)) | |
451 | queue_flush(q, QUEUE_ORDERED_POSTFLUSH); | |
452 | else | |
453 | q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH; | |
454 | ||
455 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); | |
456 | ||
457 | if (q->ordered & QUEUE_ORDERED_PREFLUSH) { | |
458 | queue_flush(q, QUEUE_ORDERED_PREFLUSH); | |
459 | rq = &q->pre_flush_rq; | |
460 | } else | |
461 | q->ordseq |= QUEUE_ORDSEQ_PREFLUSH; | |
462 | ||
463 | if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0) | |
464 | q->ordseq |= QUEUE_ORDSEQ_DRAIN; | |
465 | else | |
466 | rq = NULL; | |
467 | ||
468 | return rq; | |
469 | } | |
470 | ||
471 | int blk_do_ordered(struct request_queue *q, struct request **rqp) | |
472 | { | |
473 | struct request *rq = *rqp; | |
474 | const int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq); | |
475 | ||
476 | if (!q->ordseq) { | |
477 | if (!is_barrier) | |
478 | return 1; | |
479 | ||
480 | if (q->next_ordered != QUEUE_ORDERED_NONE) { | |
481 | *rqp = start_ordered(q, rq); | |
482 | return 1; | |
483 | } else { | |
484 | /* | |
485 | * This can happen when the queue switches to | |
486 | * ORDERED_NONE while this request is on it. | |
487 | */ | |
488 | blkdev_dequeue_request(rq); | |
489 | end_that_request_first(rq, -EOPNOTSUPP, | |
490 | rq->hard_nr_sectors); | |
491 | end_that_request_last(rq, -EOPNOTSUPP); | |
492 | *rqp = NULL; | |
493 | return 0; | |
494 | } | |
495 | } | |
496 | ||
497 | /* | |
498 | * Ordered sequence in progress | |
499 | */ | |
500 | ||
501 | /* Special requests are not subject to ordering rules. */ | |
502 | if (!blk_fs_request(rq) && | |
503 | rq != &q->pre_flush_rq && rq != &q->post_flush_rq) | |
504 | return 1; | |
505 | ||
506 | if (q->ordered & QUEUE_ORDERED_TAG) { | |
507 | /* Ordered by tag. Blocking the next barrier is enough. */ | |
508 | if (is_barrier && rq != &q->bar_rq) | |
509 | *rqp = NULL; | |
510 | } else { | |
511 | /* Ordered by draining. Wait for turn. */ | |
512 | WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q)); | |
513 | if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q)) | |
514 | *rqp = NULL; | |
515 | } | |
516 | ||
517 | return 1; | |
518 | } | |
519 | ||
520 | static void req_bio_endio(struct request *rq, struct bio *bio, | |
521 | unsigned int nbytes, int error) | |
522 | { | |
523 | struct request_queue *q = rq->q; | |
524 | ||
525 | if (&q->bar_rq != rq) { | |
526 | if (error) | |
527 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
528 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
529 | error = -EIO; | |
530 | ||
531 | if (unlikely(nbytes > bio->bi_size)) { | |
532 | printk("%s: want %u bytes done, only %u left\n", | |
533 | __FUNCTION__, nbytes, bio->bi_size); | |
534 | nbytes = bio->bi_size; | |
535 | } | |
536 | ||
537 | bio->bi_size -= nbytes; | |
538 | bio->bi_sector += (nbytes >> 9); | |
539 | if (bio->bi_size == 0) | |
540 | bio_endio(bio, error); | |
541 | } else { | |
542 | ||
543 | /* | |
544 | * Okay, this is the barrier request in progress, just | |
545 | * record the error; | |
546 | */ | |
547 | if (error && !q->orderr) | |
548 | q->orderr = error; | |
549 | } | |
550 | } | |
551 | ||
552 | /** | |
553 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
554 | * @q: the request queue for the device | |
555 | * @dma_addr: bus address limit | |
556 | * | |
557 | * Description: | |
558 | * Different hardware can have different requirements as to what pages | |
559 | * it can do I/O directly to. A low level driver can call | |
560 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
561 | * buffers for doing I/O to pages residing above @page. | |
562 | **/ | |
563 | void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr) | |
564 | { | |
565 | unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT; | |
566 | int dma = 0; | |
567 | ||
568 | q->bounce_gfp = GFP_NOIO; | |
569 | #if BITS_PER_LONG == 64 | |
570 | /* Assume anything <= 4GB can be handled by IOMMU. | |
571 | Actually some IOMMUs can handle everything, but I don't | |
572 | know of a way to test this here. */ | |
573 | if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) | |
574 | dma = 1; | |
575 | q->bounce_pfn = max_low_pfn; | |
576 | #else | |
577 | if (bounce_pfn < blk_max_low_pfn) | |
578 | dma = 1; | |
579 | q->bounce_pfn = bounce_pfn; | |
580 | #endif | |
581 | if (dma) { | |
582 | init_emergency_isa_pool(); | |
583 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
584 | q->bounce_pfn = bounce_pfn; | |
585 | } | |
586 | } | |
587 | ||
588 | EXPORT_SYMBOL(blk_queue_bounce_limit); | |
589 | ||
590 | /** | |
591 | * blk_queue_max_sectors - set max sectors for a request for this queue | |
592 | * @q: the request queue for the device | |
593 | * @max_sectors: max sectors in the usual 512b unit | |
594 | * | |
595 | * Description: | |
596 | * Enables a low level driver to set an upper limit on the size of | |
597 | * received requests. | |
598 | **/ | |
599 | void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors) | |
600 | { | |
601 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { | |
602 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
603 | printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors); | |
604 | } | |
605 | ||
606 | if (BLK_DEF_MAX_SECTORS > max_sectors) | |
607 | q->max_hw_sectors = q->max_sectors = max_sectors; | |
608 | else { | |
609 | q->max_sectors = BLK_DEF_MAX_SECTORS; | |
610 | q->max_hw_sectors = max_sectors; | |
611 | } | |
612 | } | |
613 | ||
614 | EXPORT_SYMBOL(blk_queue_max_sectors); | |
615 | ||
616 | /** | |
617 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue | |
618 | * @q: the request queue for the device | |
619 | * @max_segments: max number of segments | |
620 | * | |
621 | * Description: | |
622 | * Enables a low level driver to set an upper limit on the number of | |
623 | * physical data segments in a request. This would be the largest sized | |
624 | * scatter list the driver could handle. | |
625 | **/ | |
626 | void blk_queue_max_phys_segments(struct request_queue *q, | |
627 | unsigned short max_segments) | |
628 | { | |
629 | if (!max_segments) { | |
630 | max_segments = 1; | |
631 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
632 | } | |
633 | ||
634 | q->max_phys_segments = max_segments; | |
635 | } | |
636 | ||
637 | EXPORT_SYMBOL(blk_queue_max_phys_segments); | |
638 | ||
639 | /** | |
640 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue | |
641 | * @q: the request queue for the device | |
642 | * @max_segments: max number of segments | |
643 | * | |
644 | * Description: | |
645 | * Enables a low level driver to set an upper limit on the number of | |
646 | * hw data segments in a request. This would be the largest number of | |
647 | * address/length pairs the host adapter can actually give as once | |
648 | * to the device. | |
649 | **/ | |
650 | void blk_queue_max_hw_segments(struct request_queue *q, | |
651 | unsigned short max_segments) | |
652 | { | |
653 | if (!max_segments) { | |
654 | max_segments = 1; | |
655 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
656 | } | |
657 | ||
658 | q->max_hw_segments = max_segments; | |
659 | } | |
660 | ||
661 | EXPORT_SYMBOL(blk_queue_max_hw_segments); | |
662 | ||
663 | /** | |
664 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
665 | * @q: the request queue for the device | |
666 | * @max_size: max size of segment in bytes | |
667 | * | |
668 | * Description: | |
669 | * Enables a low level driver to set an upper limit on the size of a | |
670 | * coalesced segment | |
671 | **/ | |
672 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) | |
673 | { | |
674 | if (max_size < PAGE_CACHE_SIZE) { | |
675 | max_size = PAGE_CACHE_SIZE; | |
676 | printk("%s: set to minimum %d\n", __FUNCTION__, max_size); | |
677 | } | |
678 | ||
679 | q->max_segment_size = max_size; | |
680 | } | |
681 | ||
682 | EXPORT_SYMBOL(blk_queue_max_segment_size); | |
683 | ||
684 | /** | |
685 | * blk_queue_hardsect_size - set hardware sector size for the queue | |
686 | * @q: the request queue for the device | |
687 | * @size: the hardware sector size, in bytes | |
688 | * | |
689 | * Description: | |
690 | * This should typically be set to the lowest possible sector size | |
691 | * that the hardware can operate on (possible without reverting to | |
692 | * even internal read-modify-write operations). Usually the default | |
693 | * of 512 covers most hardware. | |
694 | **/ | |
695 | void blk_queue_hardsect_size(struct request_queue *q, unsigned short size) | |
696 | { | |
697 | q->hardsect_size = size; | |
698 | } | |
699 | ||
700 | EXPORT_SYMBOL(blk_queue_hardsect_size); | |
701 | ||
702 | /* | |
703 | * Returns the minimum that is _not_ zero, unless both are zero. | |
704 | */ | |
705 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
706 | ||
707 | /** | |
708 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
709 | * @t: the stacking driver (top) | |
710 | * @b: the underlying device (bottom) | |
711 | **/ | |
712 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) | |
713 | { | |
714 | /* zero is "infinity" */ | |
715 | t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors); | |
716 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors); | |
717 | ||
718 | t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments); | |
719 | t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments); | |
720 | t->max_segment_size = min(t->max_segment_size,b->max_segment_size); | |
721 | t->hardsect_size = max(t->hardsect_size,b->hardsect_size); | |
722 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) | |
723 | clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags); | |
724 | } | |
725 | ||
726 | EXPORT_SYMBOL(blk_queue_stack_limits); | |
727 | ||
728 | /** | |
729 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
730 | * @q: the request queue for the device | |
731 | * @mask: the memory boundary mask | |
732 | **/ | |
733 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) | |
734 | { | |
735 | if (mask < PAGE_CACHE_SIZE - 1) { | |
736 | mask = PAGE_CACHE_SIZE - 1; | |
737 | printk("%s: set to minimum %lx\n", __FUNCTION__, mask); | |
738 | } | |
739 | ||
740 | q->seg_boundary_mask = mask; | |
741 | } | |
742 | ||
743 | EXPORT_SYMBOL(blk_queue_segment_boundary); | |
744 | ||
745 | /** | |
746 | * blk_queue_dma_alignment - set dma length and memory alignment | |
747 | * @q: the request queue for the device | |
748 | * @mask: alignment mask | |
749 | * | |
750 | * description: | |
751 | * set required memory and length aligment for direct dma transactions. | |
752 | * this is used when buiding direct io requests for the queue. | |
753 | * | |
754 | **/ | |
755 | void blk_queue_dma_alignment(struct request_queue *q, int mask) | |
756 | { | |
757 | q->dma_alignment = mask; | |
758 | } | |
759 | ||
760 | EXPORT_SYMBOL(blk_queue_dma_alignment); | |
761 | ||
762 | /** | |
763 | * blk_queue_find_tag - find a request by its tag and queue | |
764 | * @q: The request queue for the device | |
765 | * @tag: The tag of the request | |
766 | * | |
767 | * Notes: | |
768 | * Should be used when a device returns a tag and you want to match | |
769 | * it with a request. | |
770 | * | |
771 | * no locks need be held. | |
772 | **/ | |
773 | struct request *blk_queue_find_tag(struct request_queue *q, int tag) | |
774 | { | |
775 | return blk_map_queue_find_tag(q->queue_tags, tag); | |
776 | } | |
777 | ||
778 | EXPORT_SYMBOL(blk_queue_find_tag); | |
779 | ||
780 | /** | |
781 | * __blk_free_tags - release a given set of tag maintenance info | |
782 | * @bqt: the tag map to free | |
783 | * | |
784 | * Tries to free the specified @bqt@. Returns true if it was | |
785 | * actually freed and false if there are still references using it | |
786 | */ | |
787 | static int __blk_free_tags(struct blk_queue_tag *bqt) | |
788 | { | |
789 | int retval; | |
790 | ||
791 | retval = atomic_dec_and_test(&bqt->refcnt); | |
792 | if (retval) { | |
793 | BUG_ON(bqt->busy); | |
794 | BUG_ON(!list_empty(&bqt->busy_list)); | |
795 | ||
796 | kfree(bqt->tag_index); | |
797 | bqt->tag_index = NULL; | |
798 | ||
799 | kfree(bqt->tag_map); | |
800 | bqt->tag_map = NULL; | |
801 | ||
802 | kfree(bqt); | |
803 | ||
804 | } | |
805 | ||
806 | return retval; | |
807 | } | |
808 | ||
809 | /** | |
810 | * __blk_queue_free_tags - release tag maintenance info | |
811 | * @q: the request queue for the device | |
812 | * | |
813 | * Notes: | |
814 | * blk_cleanup_queue() will take care of calling this function, if tagging | |
815 | * has been used. So there's no need to call this directly. | |
816 | **/ | |
817 | static void __blk_queue_free_tags(struct request_queue *q) | |
818 | { | |
819 | struct blk_queue_tag *bqt = q->queue_tags; | |
820 | ||
821 | if (!bqt) | |
822 | return; | |
823 | ||
824 | __blk_free_tags(bqt); | |
825 | ||
826 | q->queue_tags = NULL; | |
827 | q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED); | |
828 | } | |
829 | ||
830 | ||
831 | /** | |
832 | * blk_free_tags - release a given set of tag maintenance info | |
833 | * @bqt: the tag map to free | |
834 | * | |
835 | * For externally managed @bqt@ frees the map. Callers of this | |
836 | * function must guarantee to have released all the queues that | |
837 | * might have been using this tag map. | |
838 | */ | |
839 | void blk_free_tags(struct blk_queue_tag *bqt) | |
840 | { | |
841 | if (unlikely(!__blk_free_tags(bqt))) | |
842 | BUG(); | |
843 | } | |
844 | EXPORT_SYMBOL(blk_free_tags); | |
845 | ||
846 | /** | |
847 | * blk_queue_free_tags - release tag maintenance info | |
848 | * @q: the request queue for the device | |
849 | * | |
850 | * Notes: | |
851 | * This is used to disabled tagged queuing to a device, yet leave | |
852 | * queue in function. | |
853 | **/ | |
854 | void blk_queue_free_tags(struct request_queue *q) | |
855 | { | |
856 | clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
857 | } | |
858 | ||
859 | EXPORT_SYMBOL(blk_queue_free_tags); | |
860 | ||
861 | static int | |
862 | init_tag_map(struct request_queue *q, struct blk_queue_tag *tags, int depth) | |
863 | { | |
864 | struct request **tag_index; | |
865 | unsigned long *tag_map; | |
866 | int nr_ulongs; | |
867 | ||
868 | if (q && depth > q->nr_requests * 2) { | |
869 | depth = q->nr_requests * 2; | |
870 | printk(KERN_ERR "%s: adjusted depth to %d\n", | |
871 | __FUNCTION__, depth); | |
872 | } | |
873 | ||
874 | tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC); | |
875 | if (!tag_index) | |
876 | goto fail; | |
877 | ||
878 | nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG; | |
879 | tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC); | |
880 | if (!tag_map) | |
881 | goto fail; | |
882 | ||
883 | tags->real_max_depth = depth; | |
884 | tags->max_depth = depth; | |
885 | tags->tag_index = tag_index; | |
886 | tags->tag_map = tag_map; | |
887 | ||
888 | return 0; | |
889 | fail: | |
890 | kfree(tag_index); | |
891 | return -ENOMEM; | |
892 | } | |
893 | ||
894 | static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q, | |
895 | int depth) | |
896 | { | |
897 | struct blk_queue_tag *tags; | |
898 | ||
899 | tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC); | |
900 | if (!tags) | |
901 | goto fail; | |
902 | ||
903 | if (init_tag_map(q, tags, depth)) | |
904 | goto fail; | |
905 | ||
906 | INIT_LIST_HEAD(&tags->busy_list); | |
907 | tags->busy = 0; | |
908 | atomic_set(&tags->refcnt, 1); | |
909 | return tags; | |
910 | fail: | |
911 | kfree(tags); | |
912 | return NULL; | |
913 | } | |
914 | ||
915 | /** | |
916 | * blk_init_tags - initialize the tag info for an external tag map | |
917 | * @depth: the maximum queue depth supported | |
918 | * @tags: the tag to use | |
919 | **/ | |
920 | struct blk_queue_tag *blk_init_tags(int depth) | |
921 | { | |
922 | return __blk_queue_init_tags(NULL, depth); | |
923 | } | |
924 | EXPORT_SYMBOL(blk_init_tags); | |
925 | ||
926 | /** | |
927 | * blk_queue_init_tags - initialize the queue tag info | |
928 | * @q: the request queue for the device | |
929 | * @depth: the maximum queue depth supported | |
930 | * @tags: the tag to use | |
931 | **/ | |
932 | int blk_queue_init_tags(struct request_queue *q, int depth, | |
933 | struct blk_queue_tag *tags) | |
934 | { | |
935 | int rc; | |
936 | ||
937 | BUG_ON(tags && q->queue_tags && tags != q->queue_tags); | |
938 | ||
939 | if (!tags && !q->queue_tags) { | |
940 | tags = __blk_queue_init_tags(q, depth); | |
941 | ||
942 | if (!tags) | |
943 | goto fail; | |
944 | } else if (q->queue_tags) { | |
945 | if ((rc = blk_queue_resize_tags(q, depth))) | |
946 | return rc; | |
947 | set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
948 | return 0; | |
949 | } else | |
950 | atomic_inc(&tags->refcnt); | |
951 | ||
952 | /* | |
953 | * assign it, all done | |
954 | */ | |
955 | q->queue_tags = tags; | |
956 | q->queue_flags |= (1 << QUEUE_FLAG_QUEUED); | |
957 | return 0; | |
958 | fail: | |
959 | kfree(tags); | |
960 | return -ENOMEM; | |
961 | } | |
962 | ||
963 | EXPORT_SYMBOL(blk_queue_init_tags); | |
964 | ||
965 | /** | |
966 | * blk_queue_resize_tags - change the queueing depth | |
967 | * @q: the request queue for the device | |
968 | * @new_depth: the new max command queueing depth | |
969 | * | |
970 | * Notes: | |
971 | * Must be called with the queue lock held. | |
972 | **/ | |
973 | int blk_queue_resize_tags(struct request_queue *q, int new_depth) | |
974 | { | |
975 | struct blk_queue_tag *bqt = q->queue_tags; | |
976 | struct request **tag_index; | |
977 | unsigned long *tag_map; | |
978 | int max_depth, nr_ulongs; | |
979 | ||
980 | if (!bqt) | |
981 | return -ENXIO; | |
982 | ||
983 | /* | |
984 | * if we already have large enough real_max_depth. just | |
985 | * adjust max_depth. *NOTE* as requests with tag value | |
986 | * between new_depth and real_max_depth can be in-flight, tag | |
987 | * map can not be shrunk blindly here. | |
988 | */ | |
989 | if (new_depth <= bqt->real_max_depth) { | |
990 | bqt->max_depth = new_depth; | |
991 | return 0; | |
992 | } | |
993 | ||
994 | /* | |
995 | * Currently cannot replace a shared tag map with a new | |
996 | * one, so error out if this is the case | |
997 | */ | |
998 | if (atomic_read(&bqt->refcnt) != 1) | |
999 | return -EBUSY; | |
1000 | ||
1001 | /* | |
1002 | * save the old state info, so we can copy it back | |
1003 | */ | |
1004 | tag_index = bqt->tag_index; | |
1005 | tag_map = bqt->tag_map; | |
1006 | max_depth = bqt->real_max_depth; | |
1007 | ||
1008 | if (init_tag_map(q, bqt, new_depth)) | |
1009 | return -ENOMEM; | |
1010 | ||
1011 | memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); | |
1012 | nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG; | |
1013 | memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long)); | |
1014 | ||
1015 | kfree(tag_index); | |
1016 | kfree(tag_map); | |
1017 | return 0; | |
1018 | } | |
1019 | ||
1020 | EXPORT_SYMBOL(blk_queue_resize_tags); | |
1021 | ||
1022 | /** | |
1023 | * blk_queue_end_tag - end tag operations for a request | |
1024 | * @q: the request queue for the device | |
1025 | * @rq: the request that has completed | |
1026 | * | |
1027 | * Description: | |
1028 | * Typically called when end_that_request_first() returns 0, meaning | |
1029 | * all transfers have been done for a request. It's important to call | |
1030 | * this function before end_that_request_last(), as that will put the | |
1031 | * request back on the free list thus corrupting the internal tag list. | |
1032 | * | |
1033 | * Notes: | |
1034 | * queue lock must be held. | |
1035 | **/ | |
1036 | void blk_queue_end_tag(struct request_queue *q, struct request *rq) | |
1037 | { | |
1038 | struct blk_queue_tag *bqt = q->queue_tags; | |
1039 | int tag = rq->tag; | |
1040 | ||
1041 | BUG_ON(tag == -1); | |
1042 | ||
1043 | if (unlikely(tag >= bqt->real_max_depth)) | |
1044 | /* | |
1045 | * This can happen after tag depth has been reduced. | |
1046 | * FIXME: how about a warning or info message here? | |
1047 | */ | |
1048 | return; | |
1049 | ||
1050 | list_del_init(&rq->queuelist); | |
1051 | rq->cmd_flags &= ~REQ_QUEUED; | |
1052 | rq->tag = -1; | |
1053 | ||
1054 | if (unlikely(bqt->tag_index[tag] == NULL)) | |
1055 | printk(KERN_ERR "%s: tag %d is missing\n", | |
1056 | __FUNCTION__, tag); | |
1057 | ||
1058 | bqt->tag_index[tag] = NULL; | |
1059 | ||
1060 | /* | |
1061 | * We use test_and_clear_bit's memory ordering properties here. | |
1062 | * The tag_map bit acts as a lock for tag_index[bit], so we need | |
1063 | * a barrer before clearing the bit (precisely: release semantics). | |
1064 | * Could use clear_bit_unlock when it is merged. | |
1065 | */ | |
1066 | if (unlikely(!test_and_clear_bit(tag, bqt->tag_map))) { | |
1067 | printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n", | |
1068 | __FUNCTION__, tag); | |
1069 | return; | |
1070 | } | |
1071 | ||
1072 | bqt->busy--; | |
1073 | } | |
1074 | ||
1075 | EXPORT_SYMBOL(blk_queue_end_tag); | |
1076 | ||
1077 | /** | |
1078 | * blk_queue_start_tag - find a free tag and assign it | |
1079 | * @q: the request queue for the device | |
1080 | * @rq: the block request that needs tagging | |
1081 | * | |
1082 | * Description: | |
1083 | * This can either be used as a stand-alone helper, or possibly be | |
1084 | * assigned as the queue &prep_rq_fn (in which case &struct request | |
1085 | * automagically gets a tag assigned). Note that this function | |
1086 | * assumes that any type of request can be queued! if this is not | |
1087 | * true for your device, you must check the request type before | |
1088 | * calling this function. The request will also be removed from | |
1089 | * the request queue, so it's the drivers responsibility to readd | |
1090 | * it if it should need to be restarted for some reason. | |
1091 | * | |
1092 | * Notes: | |
1093 | * queue lock must be held. | |
1094 | **/ | |
1095 | int blk_queue_start_tag(struct request_queue *q, struct request *rq) | |
1096 | { | |
1097 | struct blk_queue_tag *bqt = q->queue_tags; | |
1098 | int tag; | |
1099 | ||
1100 | if (unlikely((rq->cmd_flags & REQ_QUEUED))) { | |
1101 | printk(KERN_ERR | |
1102 | "%s: request %p for device [%s] already tagged %d", | |
1103 | __FUNCTION__, rq, | |
1104 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); | |
1105 | BUG(); | |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * Protect against shared tag maps, as we may not have exclusive | |
1110 | * access to the tag map. | |
1111 | */ | |
1112 | do { | |
1113 | tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth); | |
1114 | if (tag >= bqt->max_depth) | |
1115 | return 1; | |
1116 | ||
1117 | } while (test_and_set_bit(tag, bqt->tag_map)); | |
1118 | /* | |
1119 | * We rely on test_and_set_bit providing lock memory ordering semantics | |
1120 | * (could use test_and_set_bit_lock when it is merged). | |
1121 | */ | |
1122 | ||
1123 | rq->cmd_flags |= REQ_QUEUED; | |
1124 | rq->tag = tag; | |
1125 | bqt->tag_index[tag] = rq; | |
1126 | blkdev_dequeue_request(rq); | |
1127 | list_add(&rq->queuelist, &bqt->busy_list); | |
1128 | bqt->busy++; | |
1129 | return 0; | |
1130 | } | |
1131 | ||
1132 | EXPORT_SYMBOL(blk_queue_start_tag); | |
1133 | ||
1134 | /** | |
1135 | * blk_queue_invalidate_tags - invalidate all pending tags | |
1136 | * @q: the request queue for the device | |
1137 | * | |
1138 | * Description: | |
1139 | * Hardware conditions may dictate a need to stop all pending requests. | |
1140 | * In this case, we will safely clear the block side of the tag queue and | |
1141 | * readd all requests to the request queue in the right order. | |
1142 | * | |
1143 | * Notes: | |
1144 | * queue lock must be held. | |
1145 | **/ | |
1146 | void blk_queue_invalidate_tags(struct request_queue *q) | |
1147 | { | |
1148 | struct blk_queue_tag *bqt = q->queue_tags; | |
1149 | struct list_head *tmp, *n; | |
1150 | struct request *rq; | |
1151 | ||
1152 | list_for_each_safe(tmp, n, &bqt->busy_list) { | |
1153 | rq = list_entry_rq(tmp); | |
1154 | ||
1155 | if (rq->tag == -1) { | |
1156 | printk(KERN_ERR | |
1157 | "%s: bad tag found on list\n", __FUNCTION__); | |
1158 | list_del_init(&rq->queuelist); | |
1159 | rq->cmd_flags &= ~REQ_QUEUED; | |
1160 | } else | |
1161 | blk_queue_end_tag(q, rq); | |
1162 | ||
1163 | rq->cmd_flags &= ~REQ_STARTED; | |
1164 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1165 | } | |
1166 | } | |
1167 | ||
1168 | EXPORT_SYMBOL(blk_queue_invalidate_tags); | |
1169 | ||
1170 | void blk_dump_rq_flags(struct request *rq, char *msg) | |
1171 | { | |
1172 | int bit; | |
1173 | ||
1174 | printk("%s: dev %s: type=%x, flags=%x\n", msg, | |
1175 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, | |
1176 | rq->cmd_flags); | |
1177 | ||
1178 | printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector, | |
1179 | rq->nr_sectors, | |
1180 | rq->current_nr_sectors); | |
1181 | printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len); | |
1182 | ||
1183 | if (blk_pc_request(rq)) { | |
1184 | printk("cdb: "); | |
1185 | for (bit = 0; bit < sizeof(rq->cmd); bit++) | |
1186 | printk("%02x ", rq->cmd[bit]); | |
1187 | printk("\n"); | |
1188 | } | |
1189 | } | |
1190 | ||
1191 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
1192 | ||
1193 | void blk_recount_segments(struct request_queue *q, struct bio *bio) | |
1194 | { | |
1195 | struct request rq; | |
1196 | struct bio *nxt = bio->bi_next; | |
1197 | rq.q = q; | |
1198 | rq.bio = rq.biotail = bio; | |
1199 | bio->bi_next = NULL; | |
1200 | blk_recalc_rq_segments(&rq); | |
1201 | bio->bi_next = nxt; | |
1202 | bio->bi_phys_segments = rq.nr_phys_segments; | |
1203 | bio->bi_hw_segments = rq.nr_hw_segments; | |
1204 | bio->bi_flags |= (1 << BIO_SEG_VALID); | |
1205 | } | |
1206 | EXPORT_SYMBOL(blk_recount_segments); | |
1207 | ||
1208 | static void blk_recalc_rq_segments(struct request *rq) | |
1209 | { | |
1210 | int nr_phys_segs; | |
1211 | int nr_hw_segs; | |
1212 | unsigned int phys_size; | |
1213 | unsigned int hw_size; | |
1214 | struct bio_vec *bv, *bvprv = NULL; | |
1215 | int seg_size; | |
1216 | int hw_seg_size; | |
1217 | int cluster; | |
1218 | struct req_iterator iter; | |
1219 | int high, highprv = 1; | |
1220 | struct request_queue *q = rq->q; | |
1221 | ||
1222 | if (!rq->bio) | |
1223 | return; | |
1224 | ||
1225 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1226 | hw_seg_size = seg_size = 0; | |
1227 | phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0; | |
1228 | rq_for_each_segment(bv, rq, iter) { | |
1229 | /* | |
1230 | * the trick here is making sure that a high page is never | |
1231 | * considered part of another segment, since that might | |
1232 | * change with the bounce page. | |
1233 | */ | |
1234 | high = page_to_pfn(bv->bv_page) > q->bounce_pfn; | |
1235 | if (high || highprv) | |
1236 | goto new_hw_segment; | |
1237 | if (cluster) { | |
1238 | if (seg_size + bv->bv_len > q->max_segment_size) | |
1239 | goto new_segment; | |
1240 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv)) | |
1241 | goto new_segment; | |
1242 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv)) | |
1243 | goto new_segment; | |
1244 | if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) | |
1245 | goto new_hw_segment; | |
1246 | ||
1247 | seg_size += bv->bv_len; | |
1248 | hw_seg_size += bv->bv_len; | |
1249 | bvprv = bv; | |
1250 | continue; | |
1251 | } | |
1252 | new_segment: | |
1253 | if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) && | |
1254 | !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) | |
1255 | hw_seg_size += bv->bv_len; | |
1256 | else { | |
1257 | new_hw_segment: | |
1258 | if (nr_hw_segs == 1 && | |
1259 | hw_seg_size > rq->bio->bi_hw_front_size) | |
1260 | rq->bio->bi_hw_front_size = hw_seg_size; | |
1261 | hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len; | |
1262 | nr_hw_segs++; | |
1263 | } | |
1264 | ||
1265 | nr_phys_segs++; | |
1266 | bvprv = bv; | |
1267 | seg_size = bv->bv_len; | |
1268 | highprv = high; | |
1269 | } | |
1270 | ||
1271 | if (nr_hw_segs == 1 && | |
1272 | hw_seg_size > rq->bio->bi_hw_front_size) | |
1273 | rq->bio->bi_hw_front_size = hw_seg_size; | |
1274 | if (hw_seg_size > rq->biotail->bi_hw_back_size) | |
1275 | rq->biotail->bi_hw_back_size = hw_seg_size; | |
1276 | rq->nr_phys_segments = nr_phys_segs; | |
1277 | rq->nr_hw_segments = nr_hw_segs; | |
1278 | } | |
1279 | ||
1280 | static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio, | |
1281 | struct bio *nxt) | |
1282 | { | |
1283 | if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) | |
1284 | return 0; | |
1285 | ||
1286 | if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt))) | |
1287 | return 0; | |
1288 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) | |
1289 | return 0; | |
1290 | ||
1291 | /* | |
1292 | * bio and nxt are contigous in memory, check if the queue allows | |
1293 | * these two to be merged into one | |
1294 | */ | |
1295 | if (BIO_SEG_BOUNDARY(q, bio, nxt)) | |
1296 | return 1; | |
1297 | ||
1298 | return 0; | |
1299 | } | |
1300 | ||
1301 | static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio, | |
1302 | struct bio *nxt) | |
1303 | { | |
1304 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1305 | blk_recount_segments(q, bio); | |
1306 | if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID))) | |
1307 | blk_recount_segments(q, nxt); | |
1308 | if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) || | |
1309 | BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size)) | |
1310 | return 0; | |
1311 | if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size) | |
1312 | return 0; | |
1313 | ||
1314 | return 1; | |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * map a request to scatterlist, return number of sg entries setup. Caller | |
1319 | * must make sure sg can hold rq->nr_phys_segments entries | |
1320 | */ | |
1321 | int blk_rq_map_sg(struct request_queue *q, struct request *rq, | |
1322 | struct scatterlist *sglist) | |
1323 | { | |
1324 | struct bio_vec *bvec, *bvprv; | |
1325 | struct req_iterator iter; | |
1326 | struct scatterlist *sg; | |
1327 | int nsegs, cluster; | |
1328 | ||
1329 | nsegs = 0; | |
1330 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1331 | ||
1332 | /* | |
1333 | * for each bio in rq | |
1334 | */ | |
1335 | bvprv = NULL; | |
1336 | sg = NULL; | |
1337 | rq_for_each_segment(bvec, rq, iter) { | |
1338 | int nbytes = bvec->bv_len; | |
1339 | ||
1340 | if (bvprv && cluster) { | |
1341 | if (sg->length + nbytes > q->max_segment_size) | |
1342 | goto new_segment; | |
1343 | ||
1344 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) | |
1345 | goto new_segment; | |
1346 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) | |
1347 | goto new_segment; | |
1348 | ||
1349 | sg->length += nbytes; | |
1350 | } else { | |
1351 | new_segment: | |
1352 | if (!sg) | |
1353 | sg = sglist; | |
1354 | else | |
1355 | sg = sg_next(sg); | |
1356 | ||
1357 | memset(sg, 0, sizeof(*sg)); | |
1358 | sg->page = bvec->bv_page; | |
1359 | sg->length = nbytes; | |
1360 | sg->offset = bvec->bv_offset; | |
1361 | nsegs++; | |
1362 | } | |
1363 | bvprv = bvec; | |
1364 | } /* segments in rq */ | |
1365 | ||
1366 | return nsegs; | |
1367 | } | |
1368 | ||
1369 | EXPORT_SYMBOL(blk_rq_map_sg); | |
1370 | ||
1371 | /* | |
1372 | * the standard queue merge functions, can be overridden with device | |
1373 | * specific ones if so desired | |
1374 | */ | |
1375 | ||
1376 | static inline int ll_new_mergeable(struct request_queue *q, | |
1377 | struct request *req, | |
1378 | struct bio *bio) | |
1379 | { | |
1380 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1381 | ||
1382 | if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
1383 | req->cmd_flags |= REQ_NOMERGE; | |
1384 | if (req == q->last_merge) | |
1385 | q->last_merge = NULL; | |
1386 | return 0; | |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * A hw segment is just getting larger, bump just the phys | |
1391 | * counter. | |
1392 | */ | |
1393 | req->nr_phys_segments += nr_phys_segs; | |
1394 | return 1; | |
1395 | } | |
1396 | ||
1397 | static inline int ll_new_hw_segment(struct request_queue *q, | |
1398 | struct request *req, | |
1399 | struct bio *bio) | |
1400 | { | |
1401 | int nr_hw_segs = bio_hw_segments(q, bio); | |
1402 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1403 | ||
1404 | if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments | |
1405 | || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
1406 | req->cmd_flags |= REQ_NOMERGE; | |
1407 | if (req == q->last_merge) | |
1408 | q->last_merge = NULL; | |
1409 | return 0; | |
1410 | } | |
1411 | ||
1412 | /* | |
1413 | * This will form the start of a new hw segment. Bump both | |
1414 | * counters. | |
1415 | */ | |
1416 | req->nr_hw_segments += nr_hw_segs; | |
1417 | req->nr_phys_segments += nr_phys_segs; | |
1418 | return 1; | |
1419 | } | |
1420 | ||
1421 | static int ll_back_merge_fn(struct request_queue *q, struct request *req, | |
1422 | struct bio *bio) | |
1423 | { | |
1424 | unsigned short max_sectors; | |
1425 | int len; | |
1426 | ||
1427 | if (unlikely(blk_pc_request(req))) | |
1428 | max_sectors = q->max_hw_sectors; | |
1429 | else | |
1430 | max_sectors = q->max_sectors; | |
1431 | ||
1432 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | |
1433 | req->cmd_flags |= REQ_NOMERGE; | |
1434 | if (req == q->last_merge) | |
1435 | q->last_merge = NULL; | |
1436 | return 0; | |
1437 | } | |
1438 | if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID))) | |
1439 | blk_recount_segments(q, req->biotail); | |
1440 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1441 | blk_recount_segments(q, bio); | |
1442 | len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size; | |
1443 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) && | |
1444 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1445 | int mergeable = ll_new_mergeable(q, req, bio); | |
1446 | ||
1447 | if (mergeable) { | |
1448 | if (req->nr_hw_segments == 1) | |
1449 | req->bio->bi_hw_front_size = len; | |
1450 | if (bio->bi_hw_segments == 1) | |
1451 | bio->bi_hw_back_size = len; | |
1452 | } | |
1453 | return mergeable; | |
1454 | } | |
1455 | ||
1456 | return ll_new_hw_segment(q, req, bio); | |
1457 | } | |
1458 | ||
1459 | static int ll_front_merge_fn(struct request_queue *q, struct request *req, | |
1460 | struct bio *bio) | |
1461 | { | |
1462 | unsigned short max_sectors; | |
1463 | int len; | |
1464 | ||
1465 | if (unlikely(blk_pc_request(req))) | |
1466 | max_sectors = q->max_hw_sectors; | |
1467 | else | |
1468 | max_sectors = q->max_sectors; | |
1469 | ||
1470 | ||
1471 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | |
1472 | req->cmd_flags |= REQ_NOMERGE; | |
1473 | if (req == q->last_merge) | |
1474 | q->last_merge = NULL; | |
1475 | return 0; | |
1476 | } | |
1477 | len = bio->bi_hw_back_size + req->bio->bi_hw_front_size; | |
1478 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1479 | blk_recount_segments(q, bio); | |
1480 | if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID))) | |
1481 | blk_recount_segments(q, req->bio); | |
1482 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) && | |
1483 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1484 | int mergeable = ll_new_mergeable(q, req, bio); | |
1485 | ||
1486 | if (mergeable) { | |
1487 | if (bio->bi_hw_segments == 1) | |
1488 | bio->bi_hw_front_size = len; | |
1489 | if (req->nr_hw_segments == 1) | |
1490 | req->biotail->bi_hw_back_size = len; | |
1491 | } | |
1492 | return mergeable; | |
1493 | } | |
1494 | ||
1495 | return ll_new_hw_segment(q, req, bio); | |
1496 | } | |
1497 | ||
1498 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, | |
1499 | struct request *next) | |
1500 | { | |
1501 | int total_phys_segments; | |
1502 | int total_hw_segments; | |
1503 | ||
1504 | /* | |
1505 | * First check if the either of the requests are re-queued | |
1506 | * requests. Can't merge them if they are. | |
1507 | */ | |
1508 | if (req->special || next->special) | |
1509 | return 0; | |
1510 | ||
1511 | /* | |
1512 | * Will it become too large? | |
1513 | */ | |
1514 | if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) | |
1515 | return 0; | |
1516 | ||
1517 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
1518 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) | |
1519 | total_phys_segments--; | |
1520 | ||
1521 | if (total_phys_segments > q->max_phys_segments) | |
1522 | return 0; | |
1523 | ||
1524 | total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; | |
1525 | if (blk_hw_contig_segment(q, req->biotail, next->bio)) { | |
1526 | int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size; | |
1527 | /* | |
1528 | * propagate the combined length to the end of the requests | |
1529 | */ | |
1530 | if (req->nr_hw_segments == 1) | |
1531 | req->bio->bi_hw_front_size = len; | |
1532 | if (next->nr_hw_segments == 1) | |
1533 | next->biotail->bi_hw_back_size = len; | |
1534 | total_hw_segments--; | |
1535 | } | |
1536 | ||
1537 | if (total_hw_segments > q->max_hw_segments) | |
1538 | return 0; | |
1539 | ||
1540 | /* Merge is OK... */ | |
1541 | req->nr_phys_segments = total_phys_segments; | |
1542 | req->nr_hw_segments = total_hw_segments; | |
1543 | return 1; | |
1544 | } | |
1545 | ||
1546 | /* | |
1547 | * "plug" the device if there are no outstanding requests: this will | |
1548 | * force the transfer to start only after we have put all the requests | |
1549 | * on the list. | |
1550 | * | |
1551 | * This is called with interrupts off and no requests on the queue and | |
1552 | * with the queue lock held. | |
1553 | */ | |
1554 | void blk_plug_device(struct request_queue *q) | |
1555 | { | |
1556 | WARN_ON(!irqs_disabled()); | |
1557 | ||
1558 | /* | |
1559 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
1560 | * which will restart the queueing | |
1561 | */ | |
1562 | if (blk_queue_stopped(q)) | |
1563 | return; | |
1564 | ||
1565 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) { | |
1566 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); | |
1567 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); | |
1568 | } | |
1569 | } | |
1570 | ||
1571 | EXPORT_SYMBOL(blk_plug_device); | |
1572 | ||
1573 | /* | |
1574 | * remove the queue from the plugged list, if present. called with | |
1575 | * queue lock held and interrupts disabled. | |
1576 | */ | |
1577 | int blk_remove_plug(struct request_queue *q) | |
1578 | { | |
1579 | WARN_ON(!irqs_disabled()); | |
1580 | ||
1581 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | |
1582 | return 0; | |
1583 | ||
1584 | del_timer(&q->unplug_timer); | |
1585 | return 1; | |
1586 | } | |
1587 | ||
1588 | EXPORT_SYMBOL(blk_remove_plug); | |
1589 | ||
1590 | /* | |
1591 | * remove the plug and let it rip.. | |
1592 | */ | |
1593 | void __generic_unplug_device(struct request_queue *q) | |
1594 | { | |
1595 | if (unlikely(blk_queue_stopped(q))) | |
1596 | return; | |
1597 | ||
1598 | if (!blk_remove_plug(q)) | |
1599 | return; | |
1600 | ||
1601 | q->request_fn(q); | |
1602 | } | |
1603 | EXPORT_SYMBOL(__generic_unplug_device); | |
1604 | ||
1605 | /** | |
1606 | * generic_unplug_device - fire a request queue | |
1607 | * @q: The &struct request_queue in question | |
1608 | * | |
1609 | * Description: | |
1610 | * Linux uses plugging to build bigger requests queues before letting | |
1611 | * the device have at them. If a queue is plugged, the I/O scheduler | |
1612 | * is still adding and merging requests on the queue. Once the queue | |
1613 | * gets unplugged, the request_fn defined for the queue is invoked and | |
1614 | * transfers started. | |
1615 | **/ | |
1616 | void generic_unplug_device(struct request_queue *q) | |
1617 | { | |
1618 | spin_lock_irq(q->queue_lock); | |
1619 | __generic_unplug_device(q); | |
1620 | spin_unlock_irq(q->queue_lock); | |
1621 | } | |
1622 | EXPORT_SYMBOL(generic_unplug_device); | |
1623 | ||
1624 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
1625 | struct page *page) | |
1626 | { | |
1627 | struct request_queue *q = bdi->unplug_io_data; | |
1628 | ||
1629 | /* | |
1630 | * devices don't necessarily have an ->unplug_fn defined | |
1631 | */ | |
1632 | if (q->unplug_fn) { | |
1633 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, | |
1634 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1635 | ||
1636 | q->unplug_fn(q); | |
1637 | } | |
1638 | } | |
1639 | ||
1640 | static void blk_unplug_work(struct work_struct *work) | |
1641 | { | |
1642 | struct request_queue *q = | |
1643 | container_of(work, struct request_queue, unplug_work); | |
1644 | ||
1645 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, | |
1646 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1647 | ||
1648 | q->unplug_fn(q); | |
1649 | } | |
1650 | ||
1651 | static void blk_unplug_timeout(unsigned long data) | |
1652 | { | |
1653 | struct request_queue *q = (struct request_queue *)data; | |
1654 | ||
1655 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, | |
1656 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1657 | ||
1658 | kblockd_schedule_work(&q->unplug_work); | |
1659 | } | |
1660 | ||
1661 | /** | |
1662 | * blk_start_queue - restart a previously stopped queue | |
1663 | * @q: The &struct request_queue in question | |
1664 | * | |
1665 | * Description: | |
1666 | * blk_start_queue() will clear the stop flag on the queue, and call | |
1667 | * the request_fn for the queue if it was in a stopped state when | |
1668 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
1669 | **/ | |
1670 | void blk_start_queue(struct request_queue *q) | |
1671 | { | |
1672 | WARN_ON(!irqs_disabled()); | |
1673 | ||
1674 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | |
1675 | ||
1676 | /* | |
1677 | * one level of recursion is ok and is much faster than kicking | |
1678 | * the unplug handling | |
1679 | */ | |
1680 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1681 | q->request_fn(q); | |
1682 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1683 | } else { | |
1684 | blk_plug_device(q); | |
1685 | kblockd_schedule_work(&q->unplug_work); | |
1686 | } | |
1687 | } | |
1688 | ||
1689 | EXPORT_SYMBOL(blk_start_queue); | |
1690 | ||
1691 | /** | |
1692 | * blk_stop_queue - stop a queue | |
1693 | * @q: The &struct request_queue in question | |
1694 | * | |
1695 | * Description: | |
1696 | * The Linux block layer assumes that a block driver will consume all | |
1697 | * entries on the request queue when the request_fn strategy is called. | |
1698 | * Often this will not happen, because of hardware limitations (queue | |
1699 | * depth settings). If a device driver gets a 'queue full' response, | |
1700 | * or if it simply chooses not to queue more I/O at one point, it can | |
1701 | * call this function to prevent the request_fn from being called until | |
1702 | * the driver has signalled it's ready to go again. This happens by calling | |
1703 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
1704 | **/ | |
1705 | void blk_stop_queue(struct request_queue *q) | |
1706 | { | |
1707 | blk_remove_plug(q); | |
1708 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | |
1709 | } | |
1710 | EXPORT_SYMBOL(blk_stop_queue); | |
1711 | ||
1712 | /** | |
1713 | * blk_sync_queue - cancel any pending callbacks on a queue | |
1714 | * @q: the queue | |
1715 | * | |
1716 | * Description: | |
1717 | * The block layer may perform asynchronous callback activity | |
1718 | * on a queue, such as calling the unplug function after a timeout. | |
1719 | * A block device may call blk_sync_queue to ensure that any | |
1720 | * such activity is cancelled, thus allowing it to release resources | |
1721 | * that the callbacks might use. The caller must already have made sure | |
1722 | * that its ->make_request_fn will not re-add plugging prior to calling | |
1723 | * this function. | |
1724 | * | |
1725 | */ | |
1726 | void blk_sync_queue(struct request_queue *q) | |
1727 | { | |
1728 | del_timer_sync(&q->unplug_timer); | |
1729 | } | |
1730 | EXPORT_SYMBOL(blk_sync_queue); | |
1731 | ||
1732 | /** | |
1733 | * blk_run_queue - run a single device queue | |
1734 | * @q: The queue to run | |
1735 | */ | |
1736 | void blk_run_queue(struct request_queue *q) | |
1737 | { | |
1738 | unsigned long flags; | |
1739 | ||
1740 | spin_lock_irqsave(q->queue_lock, flags); | |
1741 | blk_remove_plug(q); | |
1742 | ||
1743 | /* | |
1744 | * Only recurse once to avoid overrunning the stack, let the unplug | |
1745 | * handling reinvoke the handler shortly if we already got there. | |
1746 | */ | |
1747 | if (!elv_queue_empty(q)) { | |
1748 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1749 | q->request_fn(q); | |
1750 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1751 | } else { | |
1752 | blk_plug_device(q); | |
1753 | kblockd_schedule_work(&q->unplug_work); | |
1754 | } | |
1755 | } | |
1756 | ||
1757 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1758 | } | |
1759 | EXPORT_SYMBOL(blk_run_queue); | |
1760 | ||
1761 | /** | |
1762 | * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed | |
1763 | * @kobj: the kobj belonging of the request queue to be released | |
1764 | * | |
1765 | * Description: | |
1766 | * blk_cleanup_queue is the pair to blk_init_queue() or | |
1767 | * blk_queue_make_request(). It should be called when a request queue is | |
1768 | * being released; typically when a block device is being de-registered. | |
1769 | * Currently, its primary task it to free all the &struct request | |
1770 | * structures that were allocated to the queue and the queue itself. | |
1771 | * | |
1772 | * Caveat: | |
1773 | * Hopefully the low level driver will have finished any | |
1774 | * outstanding requests first... | |
1775 | **/ | |
1776 | static void blk_release_queue(struct kobject *kobj) | |
1777 | { | |
1778 | struct request_queue *q = | |
1779 | container_of(kobj, struct request_queue, kobj); | |
1780 | struct request_list *rl = &q->rq; | |
1781 | ||
1782 | blk_sync_queue(q); | |
1783 | ||
1784 | if (rl->rq_pool) | |
1785 | mempool_destroy(rl->rq_pool); | |
1786 | ||
1787 | if (q->queue_tags) | |
1788 | __blk_queue_free_tags(q); | |
1789 | ||
1790 | blk_trace_shutdown(q); | |
1791 | ||
1792 | bdi_destroy(&q->backing_dev_info); | |
1793 | kmem_cache_free(requestq_cachep, q); | |
1794 | } | |
1795 | ||
1796 | void blk_put_queue(struct request_queue *q) | |
1797 | { | |
1798 | kobject_put(&q->kobj); | |
1799 | } | |
1800 | EXPORT_SYMBOL(blk_put_queue); | |
1801 | ||
1802 | void blk_cleanup_queue(struct request_queue * q) | |
1803 | { | |
1804 | mutex_lock(&q->sysfs_lock); | |
1805 | set_bit(QUEUE_FLAG_DEAD, &q->queue_flags); | |
1806 | mutex_unlock(&q->sysfs_lock); | |
1807 | ||
1808 | if (q->elevator) | |
1809 | elevator_exit(q->elevator); | |
1810 | ||
1811 | blk_put_queue(q); | |
1812 | } | |
1813 | ||
1814 | EXPORT_SYMBOL(blk_cleanup_queue); | |
1815 | ||
1816 | static int blk_init_free_list(struct request_queue *q) | |
1817 | { | |
1818 | struct request_list *rl = &q->rq; | |
1819 | ||
1820 | rl->count[READ] = rl->count[WRITE] = 0; | |
1821 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
1822 | rl->elvpriv = 0; | |
1823 | init_waitqueue_head(&rl->wait[READ]); | |
1824 | init_waitqueue_head(&rl->wait[WRITE]); | |
1825 | ||
1826 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, | |
1827 | mempool_free_slab, request_cachep, q->node); | |
1828 | ||
1829 | if (!rl->rq_pool) | |
1830 | return -ENOMEM; | |
1831 | ||
1832 | return 0; | |
1833 | } | |
1834 | ||
1835 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) | |
1836 | { | |
1837 | return blk_alloc_queue_node(gfp_mask, -1); | |
1838 | } | |
1839 | EXPORT_SYMBOL(blk_alloc_queue); | |
1840 | ||
1841 | static struct kobj_type queue_ktype; | |
1842 | ||
1843 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) | |
1844 | { | |
1845 | struct request_queue *q; | |
1846 | int err; | |
1847 | ||
1848 | q = kmem_cache_alloc_node(requestq_cachep, | |
1849 | gfp_mask | __GFP_ZERO, node_id); | |
1850 | if (!q) | |
1851 | return NULL; | |
1852 | ||
1853 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; | |
1854 | q->backing_dev_info.unplug_io_data = q; | |
1855 | err = bdi_init(&q->backing_dev_info); | |
1856 | if (err) { | |
1857 | kmem_cache_free(requestq_cachep, q); | |
1858 | return NULL; | |
1859 | } | |
1860 | ||
1861 | init_timer(&q->unplug_timer); | |
1862 | ||
1863 | kobject_set_name(&q->kobj, "%s", "queue"); | |
1864 | q->kobj.ktype = &queue_ktype; | |
1865 | kobject_init(&q->kobj); | |
1866 | ||
1867 | mutex_init(&q->sysfs_lock); | |
1868 | ||
1869 | return q; | |
1870 | } | |
1871 | EXPORT_SYMBOL(blk_alloc_queue_node); | |
1872 | ||
1873 | /** | |
1874 | * blk_init_queue - prepare a request queue for use with a block device | |
1875 | * @rfn: The function to be called to process requests that have been | |
1876 | * placed on the queue. | |
1877 | * @lock: Request queue spin lock | |
1878 | * | |
1879 | * Description: | |
1880 | * If a block device wishes to use the standard request handling procedures, | |
1881 | * which sorts requests and coalesces adjacent requests, then it must | |
1882 | * call blk_init_queue(). The function @rfn will be called when there | |
1883 | * are requests on the queue that need to be processed. If the device | |
1884 | * supports plugging, then @rfn may not be called immediately when requests | |
1885 | * are available on the queue, but may be called at some time later instead. | |
1886 | * Plugged queues are generally unplugged when a buffer belonging to one | |
1887 | * of the requests on the queue is needed, or due to memory pressure. | |
1888 | * | |
1889 | * @rfn is not required, or even expected, to remove all requests off the | |
1890 | * queue, but only as many as it can handle at a time. If it does leave | |
1891 | * requests on the queue, it is responsible for arranging that the requests | |
1892 | * get dealt with eventually. | |
1893 | * | |
1894 | * The queue spin lock must be held while manipulating the requests on the | |
1895 | * request queue; this lock will be taken also from interrupt context, so irq | |
1896 | * disabling is needed for it. | |
1897 | * | |
1898 | * Function returns a pointer to the initialized request queue, or NULL if | |
1899 | * it didn't succeed. | |
1900 | * | |
1901 | * Note: | |
1902 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
1903 | * when the block device is deactivated (such as at module unload). | |
1904 | **/ | |
1905 | ||
1906 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) | |
1907 | { | |
1908 | return blk_init_queue_node(rfn, lock, -1); | |
1909 | } | |
1910 | EXPORT_SYMBOL(blk_init_queue); | |
1911 | ||
1912 | struct request_queue * | |
1913 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) | |
1914 | { | |
1915 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); | |
1916 | ||
1917 | if (!q) | |
1918 | return NULL; | |
1919 | ||
1920 | q->node = node_id; | |
1921 | if (blk_init_free_list(q)) { | |
1922 | kmem_cache_free(requestq_cachep, q); | |
1923 | return NULL; | |
1924 | } | |
1925 | ||
1926 | /* | |
1927 | * if caller didn't supply a lock, they get per-queue locking with | |
1928 | * our embedded lock | |
1929 | */ | |
1930 | if (!lock) { | |
1931 | spin_lock_init(&q->__queue_lock); | |
1932 | lock = &q->__queue_lock; | |
1933 | } | |
1934 | ||
1935 | q->request_fn = rfn; | |
1936 | q->prep_rq_fn = NULL; | |
1937 | q->unplug_fn = generic_unplug_device; | |
1938 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); | |
1939 | q->queue_lock = lock; | |
1940 | ||
1941 | blk_queue_segment_boundary(q, 0xffffffff); | |
1942 | ||
1943 | blk_queue_make_request(q, __make_request); | |
1944 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
1945 | ||
1946 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
1947 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
1948 | ||
1949 | q->sg_reserved_size = INT_MAX; | |
1950 | ||
1951 | /* | |
1952 | * all done | |
1953 | */ | |
1954 | if (!elevator_init(q, NULL)) { | |
1955 | blk_queue_congestion_threshold(q); | |
1956 | return q; | |
1957 | } | |
1958 | ||
1959 | blk_put_queue(q); | |
1960 | return NULL; | |
1961 | } | |
1962 | EXPORT_SYMBOL(blk_init_queue_node); | |
1963 | ||
1964 | int blk_get_queue(struct request_queue *q) | |
1965 | { | |
1966 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { | |
1967 | kobject_get(&q->kobj); | |
1968 | return 0; | |
1969 | } | |
1970 | ||
1971 | return 1; | |
1972 | } | |
1973 | ||
1974 | EXPORT_SYMBOL(blk_get_queue); | |
1975 | ||
1976 | static inline void blk_free_request(struct request_queue *q, struct request *rq) | |
1977 | { | |
1978 | if (rq->cmd_flags & REQ_ELVPRIV) | |
1979 | elv_put_request(q, rq); | |
1980 | mempool_free(rq, q->rq.rq_pool); | |
1981 | } | |
1982 | ||
1983 | static struct request * | |
1984 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) | |
1985 | { | |
1986 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
1987 | ||
1988 | if (!rq) | |
1989 | return NULL; | |
1990 | ||
1991 | /* | |
1992 | * first three bits are identical in rq->cmd_flags and bio->bi_rw, | |
1993 | * see bio.h and blkdev.h | |
1994 | */ | |
1995 | rq->cmd_flags = rw | REQ_ALLOCED; | |
1996 | ||
1997 | if (priv) { | |
1998 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { | |
1999 | mempool_free(rq, q->rq.rq_pool); | |
2000 | return NULL; | |
2001 | } | |
2002 | rq->cmd_flags |= REQ_ELVPRIV; | |
2003 | } | |
2004 | ||
2005 | return rq; | |
2006 | } | |
2007 | ||
2008 | /* | |
2009 | * ioc_batching returns true if the ioc is a valid batching request and | |
2010 | * should be given priority access to a request. | |
2011 | */ | |
2012 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) | |
2013 | { | |
2014 | if (!ioc) | |
2015 | return 0; | |
2016 | ||
2017 | /* | |
2018 | * Make sure the process is able to allocate at least 1 request | |
2019 | * even if the batch times out, otherwise we could theoretically | |
2020 | * lose wakeups. | |
2021 | */ | |
2022 | return ioc->nr_batch_requests == q->nr_batching || | |
2023 | (ioc->nr_batch_requests > 0 | |
2024 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
2025 | } | |
2026 | ||
2027 | /* | |
2028 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
2029 | * will cause the process to be a "batcher" on all queues in the system. This | |
2030 | * is the behaviour we want though - once it gets a wakeup it should be given | |
2031 | * a nice run. | |
2032 | */ | |
2033 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) | |
2034 | { | |
2035 | if (!ioc || ioc_batching(q, ioc)) | |
2036 | return; | |
2037 | ||
2038 | ioc->nr_batch_requests = q->nr_batching; | |
2039 | ioc->last_waited = jiffies; | |
2040 | } | |
2041 | ||
2042 | static void __freed_request(struct request_queue *q, int rw) | |
2043 | { | |
2044 | struct request_list *rl = &q->rq; | |
2045 | ||
2046 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
2047 | blk_clear_queue_congested(q, rw); | |
2048 | ||
2049 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
2050 | if (waitqueue_active(&rl->wait[rw])) | |
2051 | wake_up(&rl->wait[rw]); | |
2052 | ||
2053 | blk_clear_queue_full(q, rw); | |
2054 | } | |
2055 | } | |
2056 | ||
2057 | /* | |
2058 | * A request has just been released. Account for it, update the full and | |
2059 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
2060 | */ | |
2061 | static void freed_request(struct request_queue *q, int rw, int priv) | |
2062 | { | |
2063 | struct request_list *rl = &q->rq; | |
2064 | ||
2065 | rl->count[rw]--; | |
2066 | if (priv) | |
2067 | rl->elvpriv--; | |
2068 | ||
2069 | __freed_request(q, rw); | |
2070 | ||
2071 | if (unlikely(rl->starved[rw ^ 1])) | |
2072 | __freed_request(q, rw ^ 1); | |
2073 | } | |
2074 | ||
2075 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
2076 | /* | |
2077 | * Get a free request, queue_lock must be held. | |
2078 | * Returns NULL on failure, with queue_lock held. | |
2079 | * Returns !NULL on success, with queue_lock *not held*. | |
2080 | */ | |
2081 | static struct request *get_request(struct request_queue *q, int rw_flags, | |
2082 | struct bio *bio, gfp_t gfp_mask) | |
2083 | { | |
2084 | struct request *rq = NULL; | |
2085 | struct request_list *rl = &q->rq; | |
2086 | struct io_context *ioc = NULL; | |
2087 | const int rw = rw_flags & 0x01; | |
2088 | int may_queue, priv; | |
2089 | ||
2090 | may_queue = elv_may_queue(q, rw_flags); | |
2091 | if (may_queue == ELV_MQUEUE_NO) | |
2092 | goto rq_starved; | |
2093 | ||
2094 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
2095 | if (rl->count[rw]+1 >= q->nr_requests) { | |
2096 | ioc = current_io_context(GFP_ATOMIC, q->node); | |
2097 | /* | |
2098 | * The queue will fill after this allocation, so set | |
2099 | * it as full, and mark this process as "batching". | |
2100 | * This process will be allowed to complete a batch of | |
2101 | * requests, others will be blocked. | |
2102 | */ | |
2103 | if (!blk_queue_full(q, rw)) { | |
2104 | ioc_set_batching(q, ioc); | |
2105 | blk_set_queue_full(q, rw); | |
2106 | } else { | |
2107 | if (may_queue != ELV_MQUEUE_MUST | |
2108 | && !ioc_batching(q, ioc)) { | |
2109 | /* | |
2110 | * The queue is full and the allocating | |
2111 | * process is not a "batcher", and not | |
2112 | * exempted by the IO scheduler | |
2113 | */ | |
2114 | goto out; | |
2115 | } | |
2116 | } | |
2117 | } | |
2118 | blk_set_queue_congested(q, rw); | |
2119 | } | |
2120 | ||
2121 | /* | |
2122 | * Only allow batching queuers to allocate up to 50% over the defined | |
2123 | * limit of requests, otherwise we could have thousands of requests | |
2124 | * allocated with any setting of ->nr_requests | |
2125 | */ | |
2126 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) | |
2127 | goto out; | |
2128 | ||
2129 | rl->count[rw]++; | |
2130 | rl->starved[rw] = 0; | |
2131 | ||
2132 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); | |
2133 | if (priv) | |
2134 | rl->elvpriv++; | |
2135 | ||
2136 | spin_unlock_irq(q->queue_lock); | |
2137 | ||
2138 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); | |
2139 | if (unlikely(!rq)) { | |
2140 | /* | |
2141 | * Allocation failed presumably due to memory. Undo anything | |
2142 | * we might have messed up. | |
2143 | * | |
2144 | * Allocating task should really be put onto the front of the | |
2145 | * wait queue, but this is pretty rare. | |
2146 | */ | |
2147 | spin_lock_irq(q->queue_lock); | |
2148 | freed_request(q, rw, priv); | |
2149 | ||
2150 | /* | |
2151 | * in the very unlikely event that allocation failed and no | |
2152 | * requests for this direction was pending, mark us starved | |
2153 | * so that freeing of a request in the other direction will | |
2154 | * notice us. another possible fix would be to split the | |
2155 | * rq mempool into READ and WRITE | |
2156 | */ | |
2157 | rq_starved: | |
2158 | if (unlikely(rl->count[rw] == 0)) | |
2159 | rl->starved[rw] = 1; | |
2160 | ||
2161 | goto out; | |
2162 | } | |
2163 | ||
2164 | /* | |
2165 | * ioc may be NULL here, and ioc_batching will be false. That's | |
2166 | * OK, if the queue is under the request limit then requests need | |
2167 | * not count toward the nr_batch_requests limit. There will always | |
2168 | * be some limit enforced by BLK_BATCH_TIME. | |
2169 | */ | |
2170 | if (ioc_batching(q, ioc)) | |
2171 | ioc->nr_batch_requests--; | |
2172 | ||
2173 | rq_init(q, rq); | |
2174 | ||
2175 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); | |
2176 | out: | |
2177 | return rq; | |
2178 | } | |
2179 | ||
2180 | /* | |
2181 | * No available requests for this queue, unplug the device and wait for some | |
2182 | * requests to become available. | |
2183 | * | |
2184 | * Called with q->queue_lock held, and returns with it unlocked. | |
2185 | */ | |
2186 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, | |
2187 | struct bio *bio) | |
2188 | { | |
2189 | const int rw = rw_flags & 0x01; | |
2190 | struct request *rq; | |
2191 | ||
2192 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
2193 | while (!rq) { | |
2194 | DEFINE_WAIT(wait); | |
2195 | struct request_list *rl = &q->rq; | |
2196 | ||
2197 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
2198 | TASK_UNINTERRUPTIBLE); | |
2199 | ||
2200 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
2201 | ||
2202 | if (!rq) { | |
2203 | struct io_context *ioc; | |
2204 | ||
2205 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); | |
2206 | ||
2207 | __generic_unplug_device(q); | |
2208 | spin_unlock_irq(q->queue_lock); | |
2209 | io_schedule(); | |
2210 | ||
2211 | /* | |
2212 | * After sleeping, we become a "batching" process and | |
2213 | * will be able to allocate at least one request, and | |
2214 | * up to a big batch of them for a small period time. | |
2215 | * See ioc_batching, ioc_set_batching | |
2216 | */ | |
2217 | ioc = current_io_context(GFP_NOIO, q->node); | |
2218 | ioc_set_batching(q, ioc); | |
2219 | ||
2220 | spin_lock_irq(q->queue_lock); | |
2221 | } | |
2222 | finish_wait(&rl->wait[rw], &wait); | |
2223 | } | |
2224 | ||
2225 | return rq; | |
2226 | } | |
2227 | ||
2228 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) | |
2229 | { | |
2230 | struct request *rq; | |
2231 | ||
2232 | BUG_ON(rw != READ && rw != WRITE); | |
2233 | ||
2234 | spin_lock_irq(q->queue_lock); | |
2235 | if (gfp_mask & __GFP_WAIT) { | |
2236 | rq = get_request_wait(q, rw, NULL); | |
2237 | } else { | |
2238 | rq = get_request(q, rw, NULL, gfp_mask); | |
2239 | if (!rq) | |
2240 | spin_unlock_irq(q->queue_lock); | |
2241 | } | |
2242 | /* q->queue_lock is unlocked at this point */ | |
2243 | ||
2244 | return rq; | |
2245 | } | |
2246 | EXPORT_SYMBOL(blk_get_request); | |
2247 | ||
2248 | /** | |
2249 | * blk_start_queueing - initiate dispatch of requests to device | |
2250 | * @q: request queue to kick into gear | |
2251 | * | |
2252 | * This is basically a helper to remove the need to know whether a queue | |
2253 | * is plugged or not if someone just wants to initiate dispatch of requests | |
2254 | * for this queue. | |
2255 | * | |
2256 | * The queue lock must be held with interrupts disabled. | |
2257 | */ | |
2258 | void blk_start_queueing(struct request_queue *q) | |
2259 | { | |
2260 | if (!blk_queue_plugged(q)) | |
2261 | q->request_fn(q); | |
2262 | else | |
2263 | __generic_unplug_device(q); | |
2264 | } | |
2265 | EXPORT_SYMBOL(blk_start_queueing); | |
2266 | ||
2267 | /** | |
2268 | * blk_requeue_request - put a request back on queue | |
2269 | * @q: request queue where request should be inserted | |
2270 | * @rq: request to be inserted | |
2271 | * | |
2272 | * Description: | |
2273 | * Drivers often keep queueing requests until the hardware cannot accept | |
2274 | * more, when that condition happens we need to put the request back | |
2275 | * on the queue. Must be called with queue lock held. | |
2276 | */ | |
2277 | void blk_requeue_request(struct request_queue *q, struct request *rq) | |
2278 | { | |
2279 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); | |
2280 | ||
2281 | if (blk_rq_tagged(rq)) | |
2282 | blk_queue_end_tag(q, rq); | |
2283 | ||
2284 | elv_requeue_request(q, rq); | |
2285 | } | |
2286 | ||
2287 | EXPORT_SYMBOL(blk_requeue_request); | |
2288 | ||
2289 | /** | |
2290 | * blk_insert_request - insert a special request in to a request queue | |
2291 | * @q: request queue where request should be inserted | |
2292 | * @rq: request to be inserted | |
2293 | * @at_head: insert request at head or tail of queue | |
2294 | * @data: private data | |
2295 | * | |
2296 | * Description: | |
2297 | * Many block devices need to execute commands asynchronously, so they don't | |
2298 | * block the whole kernel from preemption during request execution. This is | |
2299 | * accomplished normally by inserting aritficial requests tagged as | |
2300 | * REQ_SPECIAL in to the corresponding request queue, and letting them be | |
2301 | * scheduled for actual execution by the request queue. | |
2302 | * | |
2303 | * We have the option of inserting the head or the tail of the queue. | |
2304 | * Typically we use the tail for new ioctls and so forth. We use the head | |
2305 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
2306 | * host that is unable to accept a particular command. | |
2307 | */ | |
2308 | void blk_insert_request(struct request_queue *q, struct request *rq, | |
2309 | int at_head, void *data) | |
2310 | { | |
2311 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | |
2312 | unsigned long flags; | |
2313 | ||
2314 | /* | |
2315 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
2316 | * must not attempt merges on this) and that it acts as a soft | |
2317 | * barrier | |
2318 | */ | |
2319 | rq->cmd_type = REQ_TYPE_SPECIAL; | |
2320 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
2321 | ||
2322 | rq->special = data; | |
2323 | ||
2324 | spin_lock_irqsave(q->queue_lock, flags); | |
2325 | ||
2326 | /* | |
2327 | * If command is tagged, release the tag | |
2328 | */ | |
2329 | if (blk_rq_tagged(rq)) | |
2330 | blk_queue_end_tag(q, rq); | |
2331 | ||
2332 | drive_stat_acct(rq, rq->nr_sectors, 1); | |
2333 | __elv_add_request(q, rq, where, 0); | |
2334 | blk_start_queueing(q); | |
2335 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2336 | } | |
2337 | ||
2338 | EXPORT_SYMBOL(blk_insert_request); | |
2339 | ||
2340 | static int __blk_rq_unmap_user(struct bio *bio) | |
2341 | { | |
2342 | int ret = 0; | |
2343 | ||
2344 | if (bio) { | |
2345 | if (bio_flagged(bio, BIO_USER_MAPPED)) | |
2346 | bio_unmap_user(bio); | |
2347 | else | |
2348 | ret = bio_uncopy_user(bio); | |
2349 | } | |
2350 | ||
2351 | return ret; | |
2352 | } | |
2353 | ||
2354 | int blk_rq_append_bio(struct request_queue *q, struct request *rq, | |
2355 | struct bio *bio) | |
2356 | { | |
2357 | if (!rq->bio) | |
2358 | blk_rq_bio_prep(q, rq, bio); | |
2359 | else if (!ll_back_merge_fn(q, rq, bio)) | |
2360 | return -EINVAL; | |
2361 | else { | |
2362 | rq->biotail->bi_next = bio; | |
2363 | rq->biotail = bio; | |
2364 | ||
2365 | rq->data_len += bio->bi_size; | |
2366 | } | |
2367 | return 0; | |
2368 | } | |
2369 | EXPORT_SYMBOL(blk_rq_append_bio); | |
2370 | ||
2371 | static int __blk_rq_map_user(struct request_queue *q, struct request *rq, | |
2372 | void __user *ubuf, unsigned int len) | |
2373 | { | |
2374 | unsigned long uaddr; | |
2375 | struct bio *bio, *orig_bio; | |
2376 | int reading, ret; | |
2377 | ||
2378 | reading = rq_data_dir(rq) == READ; | |
2379 | ||
2380 | /* | |
2381 | * if alignment requirement is satisfied, map in user pages for | |
2382 | * direct dma. else, set up kernel bounce buffers | |
2383 | */ | |
2384 | uaddr = (unsigned long) ubuf; | |
2385 | if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q))) | |
2386 | bio = bio_map_user(q, NULL, uaddr, len, reading); | |
2387 | else | |
2388 | bio = bio_copy_user(q, uaddr, len, reading); | |
2389 | ||
2390 | if (IS_ERR(bio)) | |
2391 | return PTR_ERR(bio); | |
2392 | ||
2393 | orig_bio = bio; | |
2394 | blk_queue_bounce(q, &bio); | |
2395 | ||
2396 | /* | |
2397 | * We link the bounce buffer in and could have to traverse it | |
2398 | * later so we have to get a ref to prevent it from being freed | |
2399 | */ | |
2400 | bio_get(bio); | |
2401 | ||
2402 | ret = blk_rq_append_bio(q, rq, bio); | |
2403 | if (!ret) | |
2404 | return bio->bi_size; | |
2405 | ||
2406 | /* if it was boucned we must call the end io function */ | |
2407 | bio_endio(bio, 0); | |
2408 | __blk_rq_unmap_user(orig_bio); | |
2409 | bio_put(bio); | |
2410 | return ret; | |
2411 | } | |
2412 | ||
2413 | /** | |
2414 | * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage | |
2415 | * @q: request queue where request should be inserted | |
2416 | * @rq: request structure to fill | |
2417 | * @ubuf: the user buffer | |
2418 | * @len: length of user data | |
2419 | * | |
2420 | * Description: | |
2421 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2422 | * a kernel bounce buffer is used. | |
2423 | * | |
2424 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2425 | * still in process context. | |
2426 | * | |
2427 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2428 | * before being submitted to the device, as pages mapped may be out of | |
2429 | * reach. It's the callers responsibility to make sure this happens. The | |
2430 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2431 | * unmapping. | |
2432 | */ | |
2433 | int blk_rq_map_user(struct request_queue *q, struct request *rq, | |
2434 | void __user *ubuf, unsigned long len) | |
2435 | { | |
2436 | unsigned long bytes_read = 0; | |
2437 | struct bio *bio = NULL; | |
2438 | int ret; | |
2439 | ||
2440 | if (len > (q->max_hw_sectors << 9)) | |
2441 | return -EINVAL; | |
2442 | if (!len || !ubuf) | |
2443 | return -EINVAL; | |
2444 | ||
2445 | while (bytes_read != len) { | |
2446 | unsigned long map_len, end, start; | |
2447 | ||
2448 | map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE); | |
2449 | end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1) | |
2450 | >> PAGE_SHIFT; | |
2451 | start = (unsigned long)ubuf >> PAGE_SHIFT; | |
2452 | ||
2453 | /* | |
2454 | * A bad offset could cause us to require BIO_MAX_PAGES + 1 | |
2455 | * pages. If this happens we just lower the requested | |
2456 | * mapping len by a page so that we can fit | |
2457 | */ | |
2458 | if (end - start > BIO_MAX_PAGES) | |
2459 | map_len -= PAGE_SIZE; | |
2460 | ||
2461 | ret = __blk_rq_map_user(q, rq, ubuf, map_len); | |
2462 | if (ret < 0) | |
2463 | goto unmap_rq; | |
2464 | if (!bio) | |
2465 | bio = rq->bio; | |
2466 | bytes_read += ret; | |
2467 | ubuf += ret; | |
2468 | } | |
2469 | ||
2470 | rq->buffer = rq->data = NULL; | |
2471 | return 0; | |
2472 | unmap_rq: | |
2473 | blk_rq_unmap_user(bio); | |
2474 | return ret; | |
2475 | } | |
2476 | ||
2477 | EXPORT_SYMBOL(blk_rq_map_user); | |
2478 | ||
2479 | /** | |
2480 | * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage | |
2481 | * @q: request queue where request should be inserted | |
2482 | * @rq: request to map data to | |
2483 | * @iov: pointer to the iovec | |
2484 | * @iov_count: number of elements in the iovec | |
2485 | * @len: I/O byte count | |
2486 | * | |
2487 | * Description: | |
2488 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2489 | * a kernel bounce buffer is used. | |
2490 | * | |
2491 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2492 | * still in process context. | |
2493 | * | |
2494 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2495 | * before being submitted to the device, as pages mapped may be out of | |
2496 | * reach. It's the callers responsibility to make sure this happens. The | |
2497 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2498 | * unmapping. | |
2499 | */ | |
2500 | int blk_rq_map_user_iov(struct request_queue *q, struct request *rq, | |
2501 | struct sg_iovec *iov, int iov_count, unsigned int len) | |
2502 | { | |
2503 | struct bio *bio; | |
2504 | ||
2505 | if (!iov || iov_count <= 0) | |
2506 | return -EINVAL; | |
2507 | ||
2508 | /* we don't allow misaligned data like bio_map_user() does. If the | |
2509 | * user is using sg, they're expected to know the alignment constraints | |
2510 | * and respect them accordingly */ | |
2511 | bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ); | |
2512 | if (IS_ERR(bio)) | |
2513 | return PTR_ERR(bio); | |
2514 | ||
2515 | if (bio->bi_size != len) { | |
2516 | bio_endio(bio, 0); | |
2517 | bio_unmap_user(bio); | |
2518 | return -EINVAL; | |
2519 | } | |
2520 | ||
2521 | bio_get(bio); | |
2522 | blk_rq_bio_prep(q, rq, bio); | |
2523 | rq->buffer = rq->data = NULL; | |
2524 | return 0; | |
2525 | } | |
2526 | ||
2527 | EXPORT_SYMBOL(blk_rq_map_user_iov); | |
2528 | ||
2529 | /** | |
2530 | * blk_rq_unmap_user - unmap a request with user data | |
2531 | * @bio: start of bio list | |
2532 | * | |
2533 | * Description: | |
2534 | * Unmap a rq previously mapped by blk_rq_map_user(). The caller must | |
2535 | * supply the original rq->bio from the blk_rq_map_user() return, since | |
2536 | * the io completion may have changed rq->bio. | |
2537 | */ | |
2538 | int blk_rq_unmap_user(struct bio *bio) | |
2539 | { | |
2540 | struct bio *mapped_bio; | |
2541 | int ret = 0, ret2; | |
2542 | ||
2543 | while (bio) { | |
2544 | mapped_bio = bio; | |
2545 | if (unlikely(bio_flagged(bio, BIO_BOUNCED))) | |
2546 | mapped_bio = bio->bi_private; | |
2547 | ||
2548 | ret2 = __blk_rq_unmap_user(mapped_bio); | |
2549 | if (ret2 && !ret) | |
2550 | ret = ret2; | |
2551 | ||
2552 | mapped_bio = bio; | |
2553 | bio = bio->bi_next; | |
2554 | bio_put(mapped_bio); | |
2555 | } | |
2556 | ||
2557 | return ret; | |
2558 | } | |
2559 | ||
2560 | EXPORT_SYMBOL(blk_rq_unmap_user); | |
2561 | ||
2562 | /** | |
2563 | * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage | |
2564 | * @q: request queue where request should be inserted | |
2565 | * @rq: request to fill | |
2566 | * @kbuf: the kernel buffer | |
2567 | * @len: length of user data | |
2568 | * @gfp_mask: memory allocation flags | |
2569 | */ | |
2570 | int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf, | |
2571 | unsigned int len, gfp_t gfp_mask) | |
2572 | { | |
2573 | struct bio *bio; | |
2574 | ||
2575 | if (len > (q->max_hw_sectors << 9)) | |
2576 | return -EINVAL; | |
2577 | if (!len || !kbuf) | |
2578 | return -EINVAL; | |
2579 | ||
2580 | bio = bio_map_kern(q, kbuf, len, gfp_mask); | |
2581 | if (IS_ERR(bio)) | |
2582 | return PTR_ERR(bio); | |
2583 | ||
2584 | if (rq_data_dir(rq) == WRITE) | |
2585 | bio->bi_rw |= (1 << BIO_RW); | |
2586 | ||
2587 | blk_rq_bio_prep(q, rq, bio); | |
2588 | blk_queue_bounce(q, &rq->bio); | |
2589 | rq->buffer = rq->data = NULL; | |
2590 | return 0; | |
2591 | } | |
2592 | ||
2593 | EXPORT_SYMBOL(blk_rq_map_kern); | |
2594 | ||
2595 | /** | |
2596 | * blk_execute_rq_nowait - insert a request into queue for execution | |
2597 | * @q: queue to insert the request in | |
2598 | * @bd_disk: matching gendisk | |
2599 | * @rq: request to insert | |
2600 | * @at_head: insert request at head or tail of queue | |
2601 | * @done: I/O completion handler | |
2602 | * | |
2603 | * Description: | |
2604 | * Insert a fully prepared request at the back of the io scheduler queue | |
2605 | * for execution. Don't wait for completion. | |
2606 | */ | |
2607 | void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk, | |
2608 | struct request *rq, int at_head, | |
2609 | rq_end_io_fn *done) | |
2610 | { | |
2611 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | |
2612 | ||
2613 | rq->rq_disk = bd_disk; | |
2614 | rq->cmd_flags |= REQ_NOMERGE; | |
2615 | rq->end_io = done; | |
2616 | WARN_ON(irqs_disabled()); | |
2617 | spin_lock_irq(q->queue_lock); | |
2618 | __elv_add_request(q, rq, where, 1); | |
2619 | __generic_unplug_device(q); | |
2620 | spin_unlock_irq(q->queue_lock); | |
2621 | } | |
2622 | EXPORT_SYMBOL_GPL(blk_execute_rq_nowait); | |
2623 | ||
2624 | /** | |
2625 | * blk_execute_rq - insert a request into queue for execution | |
2626 | * @q: queue to insert the request in | |
2627 | * @bd_disk: matching gendisk | |
2628 | * @rq: request to insert | |
2629 | * @at_head: insert request at head or tail of queue | |
2630 | * | |
2631 | * Description: | |
2632 | * Insert a fully prepared request at the back of the io scheduler queue | |
2633 | * for execution and wait for completion. | |
2634 | */ | |
2635 | int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk, | |
2636 | struct request *rq, int at_head) | |
2637 | { | |
2638 | DECLARE_COMPLETION_ONSTACK(wait); | |
2639 | char sense[SCSI_SENSE_BUFFERSIZE]; | |
2640 | int err = 0; | |
2641 | ||
2642 | /* | |
2643 | * we need an extra reference to the request, so we can look at | |
2644 | * it after io completion | |
2645 | */ | |
2646 | rq->ref_count++; | |
2647 | ||
2648 | if (!rq->sense) { | |
2649 | memset(sense, 0, sizeof(sense)); | |
2650 | rq->sense = sense; | |
2651 | rq->sense_len = 0; | |
2652 | } | |
2653 | ||
2654 | rq->end_io_data = &wait; | |
2655 | blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq); | |
2656 | wait_for_completion(&wait); | |
2657 | ||
2658 | if (rq->errors) | |
2659 | err = -EIO; | |
2660 | ||
2661 | return err; | |
2662 | } | |
2663 | ||
2664 | EXPORT_SYMBOL(blk_execute_rq); | |
2665 | ||
2666 | static void bio_end_empty_barrier(struct bio *bio, int err) | |
2667 | { | |
2668 | if (err) | |
2669 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
2670 | ||
2671 | complete(bio->bi_private); | |
2672 | } | |
2673 | ||
2674 | /** | |
2675 | * blkdev_issue_flush - queue a flush | |
2676 | * @bdev: blockdev to issue flush for | |
2677 | * @error_sector: error sector | |
2678 | * | |
2679 | * Description: | |
2680 | * Issue a flush for the block device in question. Caller can supply | |
2681 | * room for storing the error offset in case of a flush error, if they | |
2682 | * wish to. Caller must run wait_for_completion() on its own. | |
2683 | */ | |
2684 | int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector) | |
2685 | { | |
2686 | DECLARE_COMPLETION_ONSTACK(wait); | |
2687 | struct request_queue *q; | |
2688 | struct bio *bio; | |
2689 | int ret; | |
2690 | ||
2691 | if (bdev->bd_disk == NULL) | |
2692 | return -ENXIO; | |
2693 | ||
2694 | q = bdev_get_queue(bdev); | |
2695 | if (!q) | |
2696 | return -ENXIO; | |
2697 | ||
2698 | bio = bio_alloc(GFP_KERNEL, 0); | |
2699 | if (!bio) | |
2700 | return -ENOMEM; | |
2701 | ||
2702 | bio->bi_end_io = bio_end_empty_barrier; | |
2703 | bio->bi_private = &wait; | |
2704 | bio->bi_bdev = bdev; | |
2705 | submit_bio(1 << BIO_RW_BARRIER, bio); | |
2706 | ||
2707 | wait_for_completion(&wait); | |
2708 | ||
2709 | /* | |
2710 | * The driver must store the error location in ->bi_sector, if | |
2711 | * it supports it. For non-stacked drivers, this should be copied | |
2712 | * from rq->sector. | |
2713 | */ | |
2714 | if (error_sector) | |
2715 | *error_sector = bio->bi_sector; | |
2716 | ||
2717 | ret = 0; | |
2718 | if (!bio_flagged(bio, BIO_UPTODATE)) | |
2719 | ret = -EIO; | |
2720 | ||
2721 | bio_put(bio); | |
2722 | return ret; | |
2723 | } | |
2724 | ||
2725 | EXPORT_SYMBOL(blkdev_issue_flush); | |
2726 | ||
2727 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io) | |
2728 | { | |
2729 | int rw = rq_data_dir(rq); | |
2730 | ||
2731 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
2732 | return; | |
2733 | ||
2734 | if (!new_io) { | |
2735 | __disk_stat_inc(rq->rq_disk, merges[rw]); | |
2736 | } else { | |
2737 | disk_round_stats(rq->rq_disk); | |
2738 | rq->rq_disk->in_flight++; | |
2739 | } | |
2740 | } | |
2741 | ||
2742 | /* | |
2743 | * add-request adds a request to the linked list. | |
2744 | * queue lock is held and interrupts disabled, as we muck with the | |
2745 | * request queue list. | |
2746 | */ | |
2747 | static inline void add_request(struct request_queue * q, struct request * req) | |
2748 | { | |
2749 | drive_stat_acct(req, req->nr_sectors, 1); | |
2750 | ||
2751 | /* | |
2752 | * elevator indicated where it wants this request to be | |
2753 | * inserted at elevator_merge time | |
2754 | */ | |
2755 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
2756 | } | |
2757 | ||
2758 | /* | |
2759 | * disk_round_stats() - Round off the performance stats on a struct | |
2760 | * disk_stats. | |
2761 | * | |
2762 | * The average IO queue length and utilisation statistics are maintained | |
2763 | * by observing the current state of the queue length and the amount of | |
2764 | * time it has been in this state for. | |
2765 | * | |
2766 | * Normally, that accounting is done on IO completion, but that can result | |
2767 | * in more than a second's worth of IO being accounted for within any one | |
2768 | * second, leading to >100% utilisation. To deal with that, we call this | |
2769 | * function to do a round-off before returning the results when reading | |
2770 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
2771 | * the current jiffies and restarts the counters again. | |
2772 | */ | |
2773 | void disk_round_stats(struct gendisk *disk) | |
2774 | { | |
2775 | unsigned long now = jiffies; | |
2776 | ||
2777 | if (now == disk->stamp) | |
2778 | return; | |
2779 | ||
2780 | if (disk->in_flight) { | |
2781 | __disk_stat_add(disk, time_in_queue, | |
2782 | disk->in_flight * (now - disk->stamp)); | |
2783 | __disk_stat_add(disk, io_ticks, (now - disk->stamp)); | |
2784 | } | |
2785 | disk->stamp = now; | |
2786 | } | |
2787 | ||
2788 | EXPORT_SYMBOL_GPL(disk_round_stats); | |
2789 | ||
2790 | /* | |
2791 | * queue lock must be held | |
2792 | */ | |
2793 | void __blk_put_request(struct request_queue *q, struct request *req) | |
2794 | { | |
2795 | if (unlikely(!q)) | |
2796 | return; | |
2797 | if (unlikely(--req->ref_count)) | |
2798 | return; | |
2799 | ||
2800 | elv_completed_request(q, req); | |
2801 | ||
2802 | /* | |
2803 | * Request may not have originated from ll_rw_blk. if not, | |
2804 | * it didn't come out of our reserved rq pools | |
2805 | */ | |
2806 | if (req->cmd_flags & REQ_ALLOCED) { | |
2807 | int rw = rq_data_dir(req); | |
2808 | int priv = req->cmd_flags & REQ_ELVPRIV; | |
2809 | ||
2810 | BUG_ON(!list_empty(&req->queuelist)); | |
2811 | BUG_ON(!hlist_unhashed(&req->hash)); | |
2812 | ||
2813 | blk_free_request(q, req); | |
2814 | freed_request(q, rw, priv); | |
2815 | } | |
2816 | } | |
2817 | ||
2818 | EXPORT_SYMBOL_GPL(__blk_put_request); | |
2819 | ||
2820 | void blk_put_request(struct request *req) | |
2821 | { | |
2822 | unsigned long flags; | |
2823 | struct request_queue *q = req->q; | |
2824 | ||
2825 | /* | |
2826 | * Gee, IDE calls in w/ NULL q. Fix IDE and remove the | |
2827 | * following if (q) test. | |
2828 | */ | |
2829 | if (q) { | |
2830 | spin_lock_irqsave(q->queue_lock, flags); | |
2831 | __blk_put_request(q, req); | |
2832 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2833 | } | |
2834 | } | |
2835 | ||
2836 | EXPORT_SYMBOL(blk_put_request); | |
2837 | ||
2838 | /** | |
2839 | * blk_end_sync_rq - executes a completion event on a request | |
2840 | * @rq: request to complete | |
2841 | * @error: end io status of the request | |
2842 | */ | |
2843 | void blk_end_sync_rq(struct request *rq, int error) | |
2844 | { | |
2845 | struct completion *waiting = rq->end_io_data; | |
2846 | ||
2847 | rq->end_io_data = NULL; | |
2848 | __blk_put_request(rq->q, rq); | |
2849 | ||
2850 | /* | |
2851 | * complete last, if this is a stack request the process (and thus | |
2852 | * the rq pointer) could be invalid right after this complete() | |
2853 | */ | |
2854 | complete(waiting); | |
2855 | } | |
2856 | EXPORT_SYMBOL(blk_end_sync_rq); | |
2857 | ||
2858 | /* | |
2859 | * Has to be called with the request spinlock acquired | |
2860 | */ | |
2861 | static int attempt_merge(struct request_queue *q, struct request *req, | |
2862 | struct request *next) | |
2863 | { | |
2864 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
2865 | return 0; | |
2866 | ||
2867 | /* | |
2868 | * not contiguous | |
2869 | */ | |
2870 | if (req->sector + req->nr_sectors != next->sector) | |
2871 | return 0; | |
2872 | ||
2873 | if (rq_data_dir(req) != rq_data_dir(next) | |
2874 | || req->rq_disk != next->rq_disk | |
2875 | || next->special) | |
2876 | return 0; | |
2877 | ||
2878 | /* | |
2879 | * If we are allowed to merge, then append bio list | |
2880 | * from next to rq and release next. merge_requests_fn | |
2881 | * will have updated segment counts, update sector | |
2882 | * counts here. | |
2883 | */ | |
2884 | if (!ll_merge_requests_fn(q, req, next)) | |
2885 | return 0; | |
2886 | ||
2887 | /* | |
2888 | * At this point we have either done a back merge | |
2889 | * or front merge. We need the smaller start_time of | |
2890 | * the merged requests to be the current request | |
2891 | * for accounting purposes. | |
2892 | */ | |
2893 | if (time_after(req->start_time, next->start_time)) | |
2894 | req->start_time = next->start_time; | |
2895 | ||
2896 | req->biotail->bi_next = next->bio; | |
2897 | req->biotail = next->biotail; | |
2898 | ||
2899 | req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors; | |
2900 | ||
2901 | elv_merge_requests(q, req, next); | |
2902 | ||
2903 | if (req->rq_disk) { | |
2904 | disk_round_stats(req->rq_disk); | |
2905 | req->rq_disk->in_flight--; | |
2906 | } | |
2907 | ||
2908 | req->ioprio = ioprio_best(req->ioprio, next->ioprio); | |
2909 | ||
2910 | __blk_put_request(q, next); | |
2911 | return 1; | |
2912 | } | |
2913 | ||
2914 | static inline int attempt_back_merge(struct request_queue *q, | |
2915 | struct request *rq) | |
2916 | { | |
2917 | struct request *next = elv_latter_request(q, rq); | |
2918 | ||
2919 | if (next) | |
2920 | return attempt_merge(q, rq, next); | |
2921 | ||
2922 | return 0; | |
2923 | } | |
2924 | ||
2925 | static inline int attempt_front_merge(struct request_queue *q, | |
2926 | struct request *rq) | |
2927 | { | |
2928 | struct request *prev = elv_former_request(q, rq); | |
2929 | ||
2930 | if (prev) | |
2931 | return attempt_merge(q, prev, rq); | |
2932 | ||
2933 | return 0; | |
2934 | } | |
2935 | ||
2936 | static void init_request_from_bio(struct request *req, struct bio *bio) | |
2937 | { | |
2938 | req->cmd_type = REQ_TYPE_FS; | |
2939 | ||
2940 | /* | |
2941 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
2942 | */ | |
2943 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
2944 | req->cmd_flags |= REQ_FAILFAST; | |
2945 | ||
2946 | /* | |
2947 | * REQ_BARRIER implies no merging, but lets make it explicit | |
2948 | */ | |
2949 | if (unlikely(bio_barrier(bio))) | |
2950 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); | |
2951 | ||
2952 | if (bio_sync(bio)) | |
2953 | req->cmd_flags |= REQ_RW_SYNC; | |
2954 | if (bio_rw_meta(bio)) | |
2955 | req->cmd_flags |= REQ_RW_META; | |
2956 | ||
2957 | req->errors = 0; | |
2958 | req->hard_sector = req->sector = bio->bi_sector; | |
2959 | req->ioprio = bio_prio(bio); | |
2960 | req->start_time = jiffies; | |
2961 | blk_rq_bio_prep(req->q, req, bio); | |
2962 | } | |
2963 | ||
2964 | static int __make_request(struct request_queue *q, struct bio *bio) | |
2965 | { | |
2966 | struct request *req; | |
2967 | int el_ret, nr_sectors, barrier, err; | |
2968 | const unsigned short prio = bio_prio(bio); | |
2969 | const int sync = bio_sync(bio); | |
2970 | int rw_flags; | |
2971 | ||
2972 | nr_sectors = bio_sectors(bio); | |
2973 | ||
2974 | /* | |
2975 | * low level driver can indicate that it wants pages above a | |
2976 | * certain limit bounced to low memory (ie for highmem, or even | |
2977 | * ISA dma in theory) | |
2978 | */ | |
2979 | blk_queue_bounce(q, &bio); | |
2980 | ||
2981 | barrier = bio_barrier(bio); | |
2982 | if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
2983 | err = -EOPNOTSUPP; | |
2984 | goto end_io; | |
2985 | } | |
2986 | ||
2987 | spin_lock_irq(q->queue_lock); | |
2988 | ||
2989 | if (unlikely(barrier) || elv_queue_empty(q)) | |
2990 | goto get_rq; | |
2991 | ||
2992 | el_ret = elv_merge(q, &req, bio); | |
2993 | switch (el_ret) { | |
2994 | case ELEVATOR_BACK_MERGE: | |
2995 | BUG_ON(!rq_mergeable(req)); | |
2996 | ||
2997 | if (!ll_back_merge_fn(q, req, bio)) | |
2998 | break; | |
2999 | ||
3000 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); | |
3001 | ||
3002 | req->biotail->bi_next = bio; | |
3003 | req->biotail = bio; | |
3004 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
3005 | req->ioprio = ioprio_best(req->ioprio, prio); | |
3006 | drive_stat_acct(req, nr_sectors, 0); | |
3007 | if (!attempt_back_merge(q, req)) | |
3008 | elv_merged_request(q, req, el_ret); | |
3009 | goto out; | |
3010 | ||
3011 | case ELEVATOR_FRONT_MERGE: | |
3012 | BUG_ON(!rq_mergeable(req)); | |
3013 | ||
3014 | if (!ll_front_merge_fn(q, req, bio)) | |
3015 | break; | |
3016 | ||
3017 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); | |
3018 | ||
3019 | bio->bi_next = req->bio; | |
3020 | req->bio = bio; | |
3021 | ||
3022 | /* | |
3023 | * may not be valid. if the low level driver said | |
3024 | * it didn't need a bounce buffer then it better | |
3025 | * not touch req->buffer either... | |
3026 | */ | |
3027 | req->buffer = bio_data(bio); | |
3028 | req->current_nr_sectors = bio_cur_sectors(bio); | |
3029 | req->hard_cur_sectors = req->current_nr_sectors; | |
3030 | req->sector = req->hard_sector = bio->bi_sector; | |
3031 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
3032 | req->ioprio = ioprio_best(req->ioprio, prio); | |
3033 | drive_stat_acct(req, nr_sectors, 0); | |
3034 | if (!attempt_front_merge(q, req)) | |
3035 | elv_merged_request(q, req, el_ret); | |
3036 | goto out; | |
3037 | ||
3038 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
3039 | default: | |
3040 | ; | |
3041 | } | |
3042 | ||
3043 | get_rq: | |
3044 | /* | |
3045 | * This sync check and mask will be re-done in init_request_from_bio(), | |
3046 | * but we need to set it earlier to expose the sync flag to the | |
3047 | * rq allocator and io schedulers. | |
3048 | */ | |
3049 | rw_flags = bio_data_dir(bio); | |
3050 | if (sync) | |
3051 | rw_flags |= REQ_RW_SYNC; | |
3052 | ||
3053 | /* | |
3054 | * Grab a free request. This is might sleep but can not fail. | |
3055 | * Returns with the queue unlocked. | |
3056 | */ | |
3057 | req = get_request_wait(q, rw_flags, bio); | |
3058 | ||
3059 | /* | |
3060 | * After dropping the lock and possibly sleeping here, our request | |
3061 | * may now be mergeable after it had proven unmergeable (above). | |
3062 | * We don't worry about that case for efficiency. It won't happen | |
3063 | * often, and the elevators are able to handle it. | |
3064 | */ | |
3065 | init_request_from_bio(req, bio); | |
3066 | ||
3067 | spin_lock_irq(q->queue_lock); | |
3068 | if (elv_queue_empty(q)) | |
3069 | blk_plug_device(q); | |
3070 | add_request(q, req); | |
3071 | out: | |
3072 | if (sync) | |
3073 | __generic_unplug_device(q); | |
3074 | ||
3075 | spin_unlock_irq(q->queue_lock); | |
3076 | return 0; | |
3077 | ||
3078 | end_io: | |
3079 | bio_endio(bio, err); | |
3080 | return 0; | |
3081 | } | |
3082 | ||
3083 | /* | |
3084 | * If bio->bi_dev is a partition, remap the location | |
3085 | */ | |
3086 | static inline void blk_partition_remap(struct bio *bio) | |
3087 | { | |
3088 | struct block_device *bdev = bio->bi_bdev; | |
3089 | ||
3090 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { | |
3091 | struct hd_struct *p = bdev->bd_part; | |
3092 | const int rw = bio_data_dir(bio); | |
3093 | ||
3094 | p->sectors[rw] += bio_sectors(bio); | |
3095 | p->ios[rw]++; | |
3096 | ||
3097 | bio->bi_sector += p->start_sect; | |
3098 | bio->bi_bdev = bdev->bd_contains; | |
3099 | ||
3100 | blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio, | |
3101 | bdev->bd_dev, bio->bi_sector, | |
3102 | bio->bi_sector - p->start_sect); | |
3103 | } | |
3104 | } | |
3105 | ||
3106 | static void handle_bad_sector(struct bio *bio) | |
3107 | { | |
3108 | char b[BDEVNAME_SIZE]; | |
3109 | ||
3110 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
3111 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
3112 | bdevname(bio->bi_bdev, b), | |
3113 | bio->bi_rw, | |
3114 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
3115 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
3116 | ||
3117 | set_bit(BIO_EOF, &bio->bi_flags); | |
3118 | } | |
3119 | ||
3120 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
3121 | ||
3122 | static DECLARE_FAULT_ATTR(fail_make_request); | |
3123 | ||
3124 | static int __init setup_fail_make_request(char *str) | |
3125 | { | |
3126 | return setup_fault_attr(&fail_make_request, str); | |
3127 | } | |
3128 | __setup("fail_make_request=", setup_fail_make_request); | |
3129 | ||
3130 | static int should_fail_request(struct bio *bio) | |
3131 | { | |
3132 | if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) || | |
3133 | (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail)) | |
3134 | return should_fail(&fail_make_request, bio->bi_size); | |
3135 | ||
3136 | return 0; | |
3137 | } | |
3138 | ||
3139 | static int __init fail_make_request_debugfs(void) | |
3140 | { | |
3141 | return init_fault_attr_dentries(&fail_make_request, | |
3142 | "fail_make_request"); | |
3143 | } | |
3144 | ||
3145 | late_initcall(fail_make_request_debugfs); | |
3146 | ||
3147 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
3148 | ||
3149 | static inline int should_fail_request(struct bio *bio) | |
3150 | { | |
3151 | return 0; | |
3152 | } | |
3153 | ||
3154 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
3155 | ||
3156 | /* | |
3157 | * Check whether this bio extends beyond the end of the device. | |
3158 | */ | |
3159 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
3160 | { | |
3161 | sector_t maxsector; | |
3162 | ||
3163 | if (!nr_sectors) | |
3164 | return 0; | |
3165 | ||
3166 | /* Test device or partition size, when known. */ | |
3167 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
3168 | if (maxsector) { | |
3169 | sector_t sector = bio->bi_sector; | |
3170 | ||
3171 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
3172 | /* | |
3173 | * This may well happen - the kernel calls bread() | |
3174 | * without checking the size of the device, e.g., when | |
3175 | * mounting a device. | |
3176 | */ | |
3177 | handle_bad_sector(bio); | |
3178 | return 1; | |
3179 | } | |
3180 | } | |
3181 | ||
3182 | return 0; | |
3183 | } | |
3184 | ||
3185 | /** | |
3186 | * generic_make_request: hand a buffer to its device driver for I/O | |
3187 | * @bio: The bio describing the location in memory and on the device. | |
3188 | * | |
3189 | * generic_make_request() is used to make I/O requests of block | |
3190 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
3191 | * to be done. | |
3192 | * | |
3193 | * generic_make_request() does not return any status. The | |
3194 | * success/failure status of the request, along with notification of | |
3195 | * completion, is delivered asynchronously through the bio->bi_end_io | |
3196 | * function described (one day) else where. | |
3197 | * | |
3198 | * The caller of generic_make_request must make sure that bi_io_vec | |
3199 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
3200 | * set to describe the device address, and the | |
3201 | * bi_end_io and optionally bi_private are set to describe how | |
3202 | * completion notification should be signaled. | |
3203 | * | |
3204 | * generic_make_request and the drivers it calls may use bi_next if this | |
3205 | * bio happens to be merged with someone else, and may change bi_dev and | |
3206 | * bi_sector for remaps as it sees fit. So the values of these fields | |
3207 | * should NOT be depended on after the call to generic_make_request. | |
3208 | */ | |
3209 | static inline void __generic_make_request(struct bio *bio) | |
3210 | { | |
3211 | struct request_queue *q; | |
3212 | sector_t old_sector; | |
3213 | int ret, nr_sectors = bio_sectors(bio); | |
3214 | dev_t old_dev; | |
3215 | ||
3216 | might_sleep(); | |
3217 | ||
3218 | if (bio_check_eod(bio, nr_sectors)) | |
3219 | goto end_io; | |
3220 | ||
3221 | /* | |
3222 | * Resolve the mapping until finished. (drivers are | |
3223 | * still free to implement/resolve their own stacking | |
3224 | * by explicitly returning 0) | |
3225 | * | |
3226 | * NOTE: we don't repeat the blk_size check for each new device. | |
3227 | * Stacking drivers are expected to know what they are doing. | |
3228 | */ | |
3229 | old_sector = -1; | |
3230 | old_dev = 0; | |
3231 | do { | |
3232 | char b[BDEVNAME_SIZE]; | |
3233 | ||
3234 | q = bdev_get_queue(bio->bi_bdev); | |
3235 | if (!q) { | |
3236 | printk(KERN_ERR | |
3237 | "generic_make_request: Trying to access " | |
3238 | "nonexistent block-device %s (%Lu)\n", | |
3239 | bdevname(bio->bi_bdev, b), | |
3240 | (long long) bio->bi_sector); | |
3241 | end_io: | |
3242 | bio_endio(bio, -EIO); | |
3243 | break; | |
3244 | } | |
3245 | ||
3246 | if (unlikely(nr_sectors > q->max_hw_sectors)) { | |
3247 | printk("bio too big device %s (%u > %u)\n", | |
3248 | bdevname(bio->bi_bdev, b), | |
3249 | bio_sectors(bio), | |
3250 | q->max_hw_sectors); | |
3251 | goto end_io; | |
3252 | } | |
3253 | ||
3254 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) | |
3255 | goto end_io; | |
3256 | ||
3257 | if (should_fail_request(bio)) | |
3258 | goto end_io; | |
3259 | ||
3260 | /* | |
3261 | * If this device has partitions, remap block n | |
3262 | * of partition p to block n+start(p) of the disk. | |
3263 | */ | |
3264 | blk_partition_remap(bio); | |
3265 | ||
3266 | if (old_sector != -1) | |
3267 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, | |
3268 | old_sector); | |
3269 | ||
3270 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); | |
3271 | ||
3272 | old_sector = bio->bi_sector; | |
3273 | old_dev = bio->bi_bdev->bd_dev; | |
3274 | ||
3275 | if (bio_check_eod(bio, nr_sectors)) | |
3276 | goto end_io; | |
3277 | ||
3278 | ret = q->make_request_fn(q, bio); | |
3279 | } while (ret); | |
3280 | } | |
3281 | ||
3282 | /* | |
3283 | * We only want one ->make_request_fn to be active at a time, | |
3284 | * else stack usage with stacked devices could be a problem. | |
3285 | * So use current->bio_{list,tail} to keep a list of requests | |
3286 | * submited by a make_request_fn function. | |
3287 | * current->bio_tail is also used as a flag to say if | |
3288 | * generic_make_request is currently active in this task or not. | |
3289 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
3290 | * then a make_request is active, and new requests should be added | |
3291 | * at the tail | |
3292 | */ | |
3293 | void generic_make_request(struct bio *bio) | |
3294 | { | |
3295 | if (current->bio_tail) { | |
3296 | /* make_request is active */ | |
3297 | *(current->bio_tail) = bio; | |
3298 | bio->bi_next = NULL; | |
3299 | current->bio_tail = &bio->bi_next; | |
3300 | return; | |
3301 | } | |
3302 | /* following loop may be a bit non-obvious, and so deserves some | |
3303 | * explanation. | |
3304 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
3305 | * ensure that) so we have a list with a single bio. | |
3306 | * We pretend that we have just taken it off a longer list, so | |
3307 | * we assign bio_list to the next (which is NULL) and bio_tail | |
3308 | * to &bio_list, thus initialising the bio_list of new bios to be | |
3309 | * added. __generic_make_request may indeed add some more bios | |
3310 | * through a recursive call to generic_make_request. If it | |
3311 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
3312 | * from the top. In this case we really did just take the bio | |
3313 | * of the top of the list (no pretending) and so fixup bio_list and | |
3314 | * bio_tail or bi_next, and call into __generic_make_request again. | |
3315 | * | |
3316 | * The loop was structured like this to make only one call to | |
3317 | * __generic_make_request (which is important as it is large and | |
3318 | * inlined) and to keep the structure simple. | |
3319 | */ | |
3320 | BUG_ON(bio->bi_next); | |
3321 | do { | |
3322 | current->bio_list = bio->bi_next; | |
3323 | if (bio->bi_next == NULL) | |
3324 | current->bio_tail = ¤t->bio_list; | |
3325 | else | |
3326 | bio->bi_next = NULL; | |
3327 | __generic_make_request(bio); | |
3328 | bio = current->bio_list; | |
3329 | } while (bio); | |
3330 | current->bio_tail = NULL; /* deactivate */ | |
3331 | } | |
3332 | ||
3333 | EXPORT_SYMBOL(generic_make_request); | |
3334 | ||
3335 | /** | |
3336 | * submit_bio: submit a bio to the block device layer for I/O | |
3337 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
3338 | * @bio: The &struct bio which describes the I/O | |
3339 | * | |
3340 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
3341 | * uses that function to do most of the work. Both are fairly rough | |
3342 | * interfaces, @bio must be presetup and ready for I/O. | |
3343 | * | |
3344 | */ | |
3345 | void submit_bio(int rw, struct bio *bio) | |
3346 | { | |
3347 | int count = bio_sectors(bio); | |
3348 | ||
3349 | bio->bi_rw |= rw; | |
3350 | ||
3351 | /* | |
3352 | * If it's a regular read/write or a barrier with data attached, | |
3353 | * go through the normal accounting stuff before submission. | |
3354 | */ | |
3355 | if (!bio_empty_barrier(bio)) { | |
3356 | ||
3357 | BIO_BUG_ON(!bio->bi_size); | |
3358 | BIO_BUG_ON(!bio->bi_io_vec); | |
3359 | ||
3360 | if (rw & WRITE) { | |
3361 | count_vm_events(PGPGOUT, count); | |
3362 | } else { | |
3363 | task_io_account_read(bio->bi_size); | |
3364 | count_vm_events(PGPGIN, count); | |
3365 | } | |
3366 | ||
3367 | if (unlikely(block_dump)) { | |
3368 | char b[BDEVNAME_SIZE]; | |
3369 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
3370 | current->comm, current->pid, | |
3371 | (rw & WRITE) ? "WRITE" : "READ", | |
3372 | (unsigned long long)bio->bi_sector, | |
3373 | bdevname(bio->bi_bdev,b)); | |
3374 | } | |
3375 | } | |
3376 | ||
3377 | generic_make_request(bio); | |
3378 | } | |
3379 | ||
3380 | EXPORT_SYMBOL(submit_bio); | |
3381 | ||
3382 | static void blk_recalc_rq_sectors(struct request *rq, int nsect) | |
3383 | { | |
3384 | if (blk_fs_request(rq)) { | |
3385 | rq->hard_sector += nsect; | |
3386 | rq->hard_nr_sectors -= nsect; | |
3387 | ||
3388 | /* | |
3389 | * Move the I/O submission pointers ahead if required. | |
3390 | */ | |
3391 | if ((rq->nr_sectors >= rq->hard_nr_sectors) && | |
3392 | (rq->sector <= rq->hard_sector)) { | |
3393 | rq->sector = rq->hard_sector; | |
3394 | rq->nr_sectors = rq->hard_nr_sectors; | |
3395 | rq->hard_cur_sectors = bio_cur_sectors(rq->bio); | |
3396 | rq->current_nr_sectors = rq->hard_cur_sectors; | |
3397 | rq->buffer = bio_data(rq->bio); | |
3398 | } | |
3399 | ||
3400 | /* | |
3401 | * if total number of sectors is less than the first segment | |
3402 | * size, something has gone terribly wrong | |
3403 | */ | |
3404 | if (rq->nr_sectors < rq->current_nr_sectors) { | |
3405 | printk("blk: request botched\n"); | |
3406 | rq->nr_sectors = rq->current_nr_sectors; | |
3407 | } | |
3408 | } | |
3409 | } | |
3410 | ||
3411 | static int __end_that_request_first(struct request *req, int uptodate, | |
3412 | int nr_bytes) | |
3413 | { | |
3414 | int total_bytes, bio_nbytes, error, next_idx = 0; | |
3415 | struct bio *bio; | |
3416 | ||
3417 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); | |
3418 | ||
3419 | /* | |
3420 | * extend uptodate bool to allow < 0 value to be direct io error | |
3421 | */ | |
3422 | error = 0; | |
3423 | if (end_io_error(uptodate)) | |
3424 | error = !uptodate ? -EIO : uptodate; | |
3425 | ||
3426 | /* | |
3427 | * for a REQ_BLOCK_PC request, we want to carry any eventual | |
3428 | * sense key with us all the way through | |
3429 | */ | |
3430 | if (!blk_pc_request(req)) | |
3431 | req->errors = 0; | |
3432 | ||
3433 | if (!uptodate) { | |
3434 | if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET)) | |
3435 | printk("end_request: I/O error, dev %s, sector %llu\n", | |
3436 | req->rq_disk ? req->rq_disk->disk_name : "?", | |
3437 | (unsigned long long)req->sector); | |
3438 | } | |
3439 | ||
3440 | if (blk_fs_request(req) && req->rq_disk) { | |
3441 | const int rw = rq_data_dir(req); | |
3442 | ||
3443 | disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9); | |
3444 | } | |
3445 | ||
3446 | total_bytes = bio_nbytes = 0; | |
3447 | while ((bio = req->bio) != NULL) { | |
3448 | int nbytes; | |
3449 | ||
3450 | /* | |
3451 | * For an empty barrier request, the low level driver must | |
3452 | * store a potential error location in ->sector. We pass | |
3453 | * that back up in ->bi_sector. | |
3454 | */ | |
3455 | if (blk_empty_barrier(req)) | |
3456 | bio->bi_sector = req->sector; | |
3457 | ||
3458 | if (nr_bytes >= bio->bi_size) { | |
3459 | req->bio = bio->bi_next; | |
3460 | nbytes = bio->bi_size; | |
3461 | req_bio_endio(req, bio, nbytes, error); | |
3462 | next_idx = 0; | |
3463 | bio_nbytes = 0; | |
3464 | } else { | |
3465 | int idx = bio->bi_idx + next_idx; | |
3466 | ||
3467 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
3468 | blk_dump_rq_flags(req, "__end_that"); | |
3469 | printk("%s: bio idx %d >= vcnt %d\n", | |
3470 | __FUNCTION__, | |
3471 | bio->bi_idx, bio->bi_vcnt); | |
3472 | break; | |
3473 | } | |
3474 | ||
3475 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
3476 | BIO_BUG_ON(nbytes > bio->bi_size); | |
3477 | ||
3478 | /* | |
3479 | * not a complete bvec done | |
3480 | */ | |
3481 | if (unlikely(nbytes > nr_bytes)) { | |
3482 | bio_nbytes += nr_bytes; | |
3483 | total_bytes += nr_bytes; | |
3484 | break; | |
3485 | } | |
3486 | ||
3487 | /* | |
3488 | * advance to the next vector | |
3489 | */ | |
3490 | next_idx++; | |
3491 | bio_nbytes += nbytes; | |
3492 | } | |
3493 | ||
3494 | total_bytes += nbytes; | |
3495 | nr_bytes -= nbytes; | |
3496 | ||
3497 | if ((bio = req->bio)) { | |
3498 | /* | |
3499 | * end more in this run, or just return 'not-done' | |
3500 | */ | |
3501 | if (unlikely(nr_bytes <= 0)) | |
3502 | break; | |
3503 | } | |
3504 | } | |
3505 | ||
3506 | /* | |
3507 | * completely done | |
3508 | */ | |
3509 | if (!req->bio) | |
3510 | return 0; | |
3511 | ||
3512 | /* | |
3513 | * if the request wasn't completed, update state | |
3514 | */ | |
3515 | if (bio_nbytes) { | |
3516 | req_bio_endio(req, bio, bio_nbytes, error); | |
3517 | bio->bi_idx += next_idx; | |
3518 | bio_iovec(bio)->bv_offset += nr_bytes; | |
3519 | bio_iovec(bio)->bv_len -= nr_bytes; | |
3520 | } | |
3521 | ||
3522 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
3523 | blk_recalc_rq_segments(req); | |
3524 | return 1; | |
3525 | } | |
3526 | ||
3527 | /** | |
3528 | * end_that_request_first - end I/O on a request | |
3529 | * @req: the request being processed | |
3530 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3531 | * @nr_sectors: number of sectors to end I/O on | |
3532 | * | |
3533 | * Description: | |
3534 | * Ends I/O on a number of sectors attached to @req, and sets it up | |
3535 | * for the next range of segments (if any) in the cluster. | |
3536 | * | |
3537 | * Return: | |
3538 | * 0 - we are done with this request, call end_that_request_last() | |
3539 | * 1 - still buffers pending for this request | |
3540 | **/ | |
3541 | int end_that_request_first(struct request *req, int uptodate, int nr_sectors) | |
3542 | { | |
3543 | return __end_that_request_first(req, uptodate, nr_sectors << 9); | |
3544 | } | |
3545 | ||
3546 | EXPORT_SYMBOL(end_that_request_first); | |
3547 | ||
3548 | /** | |
3549 | * end_that_request_chunk - end I/O on a request | |
3550 | * @req: the request being processed | |
3551 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3552 | * @nr_bytes: number of bytes to complete | |
3553 | * | |
3554 | * Description: | |
3555 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
3556 | * for the next range of segments (if any). Like end_that_request_first(), | |
3557 | * but deals with bytes instead of sectors. | |
3558 | * | |
3559 | * Return: | |
3560 | * 0 - we are done with this request, call end_that_request_last() | |
3561 | * 1 - still buffers pending for this request | |
3562 | **/ | |
3563 | int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes) | |
3564 | { | |
3565 | return __end_that_request_first(req, uptodate, nr_bytes); | |
3566 | } | |
3567 | ||
3568 | EXPORT_SYMBOL(end_that_request_chunk); | |
3569 | ||
3570 | /* | |
3571 | * splice the completion data to a local structure and hand off to | |
3572 | * process_completion_queue() to complete the requests | |
3573 | */ | |
3574 | static void blk_done_softirq(struct softirq_action *h) | |
3575 | { | |
3576 | struct list_head *cpu_list, local_list; | |
3577 | ||
3578 | local_irq_disable(); | |
3579 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
3580 | list_replace_init(cpu_list, &local_list); | |
3581 | local_irq_enable(); | |
3582 | ||
3583 | while (!list_empty(&local_list)) { | |
3584 | struct request *rq = list_entry(local_list.next, struct request, donelist); | |
3585 | ||
3586 | list_del_init(&rq->donelist); | |
3587 | rq->q->softirq_done_fn(rq); | |
3588 | } | |
3589 | } | |
3590 | ||
3591 | static int __cpuinit blk_cpu_notify(struct notifier_block *self, unsigned long action, | |
3592 | void *hcpu) | |
3593 | { | |
3594 | /* | |
3595 | * If a CPU goes away, splice its entries to the current CPU | |
3596 | * and trigger a run of the softirq | |
3597 | */ | |
3598 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { | |
3599 | int cpu = (unsigned long) hcpu; | |
3600 | ||
3601 | local_irq_disable(); | |
3602 | list_splice_init(&per_cpu(blk_cpu_done, cpu), | |
3603 | &__get_cpu_var(blk_cpu_done)); | |
3604 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
3605 | local_irq_enable(); | |
3606 | } | |
3607 | ||
3608 | return NOTIFY_OK; | |
3609 | } | |
3610 | ||
3611 | ||
3612 | static struct notifier_block blk_cpu_notifier __cpuinitdata = { | |
3613 | .notifier_call = blk_cpu_notify, | |
3614 | }; | |
3615 | ||
3616 | /** | |
3617 | * blk_complete_request - end I/O on a request | |
3618 | * @req: the request being processed | |
3619 | * | |
3620 | * Description: | |
3621 | * Ends all I/O on a request. It does not handle partial completions, | |
3622 | * unless the driver actually implements this in its completion callback | |
3623 | * through requeueing. The actual completion happens out-of-order, | |
3624 | * through a softirq handler. The user must have registered a completion | |
3625 | * callback through blk_queue_softirq_done(). | |
3626 | **/ | |
3627 | ||
3628 | void blk_complete_request(struct request *req) | |
3629 | { | |
3630 | struct list_head *cpu_list; | |
3631 | unsigned long flags; | |
3632 | ||
3633 | BUG_ON(!req->q->softirq_done_fn); | |
3634 | ||
3635 | local_irq_save(flags); | |
3636 | ||
3637 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
3638 | list_add_tail(&req->donelist, cpu_list); | |
3639 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
3640 | ||
3641 | local_irq_restore(flags); | |
3642 | } | |
3643 | ||
3644 | EXPORT_SYMBOL(blk_complete_request); | |
3645 | ||
3646 | /* | |
3647 | * queue lock must be held | |
3648 | */ | |
3649 | void end_that_request_last(struct request *req, int uptodate) | |
3650 | { | |
3651 | struct gendisk *disk = req->rq_disk; | |
3652 | int error; | |
3653 | ||
3654 | /* | |
3655 | * extend uptodate bool to allow < 0 value to be direct io error | |
3656 | */ | |
3657 | error = 0; | |
3658 | if (end_io_error(uptodate)) | |
3659 | error = !uptodate ? -EIO : uptodate; | |
3660 | ||
3661 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
3662 | laptop_io_completion(); | |
3663 | ||
3664 | /* | |
3665 | * Account IO completion. bar_rq isn't accounted as a normal | |
3666 | * IO on queueing nor completion. Accounting the containing | |
3667 | * request is enough. | |
3668 | */ | |
3669 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
3670 | unsigned long duration = jiffies - req->start_time; | |
3671 | const int rw = rq_data_dir(req); | |
3672 | ||
3673 | __disk_stat_inc(disk, ios[rw]); | |
3674 | __disk_stat_add(disk, ticks[rw], duration); | |
3675 | disk_round_stats(disk); | |
3676 | disk->in_flight--; | |
3677 | } | |
3678 | if (req->end_io) | |
3679 | req->end_io(req, error); | |
3680 | else | |
3681 | __blk_put_request(req->q, req); | |
3682 | } | |
3683 | ||
3684 | EXPORT_SYMBOL(end_that_request_last); | |
3685 | ||
3686 | static inline void __end_request(struct request *rq, int uptodate, | |
3687 | unsigned int nr_bytes, int dequeue) | |
3688 | { | |
3689 | if (!end_that_request_chunk(rq, uptodate, nr_bytes)) { | |
3690 | if (dequeue) | |
3691 | blkdev_dequeue_request(rq); | |
3692 | add_disk_randomness(rq->rq_disk); | |
3693 | end_that_request_last(rq, uptodate); | |
3694 | } | |
3695 | } | |
3696 | ||
3697 | static unsigned int rq_byte_size(struct request *rq) | |
3698 | { | |
3699 | if (blk_fs_request(rq)) | |
3700 | return rq->hard_nr_sectors << 9; | |
3701 | ||
3702 | return rq->data_len; | |
3703 | } | |
3704 | ||
3705 | /** | |
3706 | * end_queued_request - end all I/O on a queued request | |
3707 | * @rq: the request being processed | |
3708 | * @uptodate: error value or 0/1 uptodate flag | |
3709 | * | |
3710 | * Description: | |
3711 | * Ends all I/O on a request, and removes it from the block layer queues. | |
3712 | * Not suitable for normal IO completion, unless the driver still has | |
3713 | * the request attached to the block layer. | |
3714 | * | |
3715 | **/ | |
3716 | void end_queued_request(struct request *rq, int uptodate) | |
3717 | { | |
3718 | __end_request(rq, uptodate, rq_byte_size(rq), 1); | |
3719 | } | |
3720 | EXPORT_SYMBOL(end_queued_request); | |
3721 | ||
3722 | /** | |
3723 | * end_dequeued_request - end all I/O on a dequeued request | |
3724 | * @rq: the request being processed | |
3725 | * @uptodate: error value or 0/1 uptodate flag | |
3726 | * | |
3727 | * Description: | |
3728 | * Ends all I/O on a request. The request must already have been | |
3729 | * dequeued using blkdev_dequeue_request(), as is normally the case | |
3730 | * for most drivers. | |
3731 | * | |
3732 | **/ | |
3733 | void end_dequeued_request(struct request *rq, int uptodate) | |
3734 | { | |
3735 | __end_request(rq, uptodate, rq_byte_size(rq), 0); | |
3736 | } | |
3737 | EXPORT_SYMBOL(end_dequeued_request); | |
3738 | ||
3739 | ||
3740 | /** | |
3741 | * end_request - end I/O on the current segment of the request | |
3742 | * @rq: the request being processed | |
3743 | * @uptodate: error value or 0/1 uptodate flag | |
3744 | * | |
3745 | * Description: | |
3746 | * Ends I/O on the current segment of a request. If that is the only | |
3747 | * remaining segment, the request is also completed and freed. | |
3748 | * | |
3749 | * This is a remnant of how older block drivers handled IO completions. | |
3750 | * Modern drivers typically end IO on the full request in one go, unless | |
3751 | * they have a residual value to account for. For that case this function | |
3752 | * isn't really useful, unless the residual just happens to be the | |
3753 | * full current segment. In other words, don't use this function in new | |
3754 | * code. Either use end_request_completely(), or the | |
3755 | * end_that_request_chunk() (along with end_that_request_last()) for | |
3756 | * partial completions. | |
3757 | * | |
3758 | **/ | |
3759 | void end_request(struct request *req, int uptodate) | |
3760 | { | |
3761 | __end_request(req, uptodate, req->hard_cur_sectors << 9, 1); | |
3762 | } | |
3763 | EXPORT_SYMBOL(end_request); | |
3764 | ||
3765 | static void blk_rq_bio_prep(struct request_queue *q, struct request *rq, | |
3766 | struct bio *bio) | |
3767 | { | |
3768 | /* first two bits are identical in rq->cmd_flags and bio->bi_rw */ | |
3769 | rq->cmd_flags |= (bio->bi_rw & 3); | |
3770 | ||
3771 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
3772 | rq->nr_hw_segments = bio_hw_segments(q, bio); | |
3773 | rq->current_nr_sectors = bio_cur_sectors(bio); | |
3774 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
3775 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
3776 | rq->buffer = bio_data(bio); | |
3777 | rq->data_len = bio->bi_size; | |
3778 | ||
3779 | rq->bio = rq->biotail = bio; | |
3780 | ||
3781 | if (bio->bi_bdev) | |
3782 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
3783 | } | |
3784 | ||
3785 | int kblockd_schedule_work(struct work_struct *work) | |
3786 | { | |
3787 | return queue_work(kblockd_workqueue, work); | |
3788 | } | |
3789 | ||
3790 | EXPORT_SYMBOL(kblockd_schedule_work); | |
3791 | ||
3792 | void kblockd_flush_work(struct work_struct *work) | |
3793 | { | |
3794 | cancel_work_sync(work); | |
3795 | } | |
3796 | EXPORT_SYMBOL(kblockd_flush_work); | |
3797 | ||
3798 | int __init blk_dev_init(void) | |
3799 | { | |
3800 | int i; | |
3801 | ||
3802 | kblockd_workqueue = create_workqueue("kblockd"); | |
3803 | if (!kblockd_workqueue) | |
3804 | panic("Failed to create kblockd\n"); | |
3805 | ||
3806 | request_cachep = kmem_cache_create("blkdev_requests", | |
3807 | sizeof(struct request), 0, SLAB_PANIC, NULL); | |
3808 | ||
3809 | requestq_cachep = kmem_cache_create("blkdev_queue", | |
3810 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); | |
3811 | ||
3812 | iocontext_cachep = kmem_cache_create("blkdev_ioc", | |
3813 | sizeof(struct io_context), 0, SLAB_PANIC, NULL); | |
3814 | ||
3815 | for_each_possible_cpu(i) | |
3816 | INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i)); | |
3817 | ||
3818 | open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL); | |
3819 | register_hotcpu_notifier(&blk_cpu_notifier); | |
3820 | ||
3821 | blk_max_low_pfn = max_low_pfn - 1; | |
3822 | blk_max_pfn = max_pfn - 1; | |
3823 | ||
3824 | return 0; | |
3825 | } | |
3826 | ||
3827 | /* | |
3828 | * IO Context helper functions | |
3829 | */ | |
3830 | void put_io_context(struct io_context *ioc) | |
3831 | { | |
3832 | if (ioc == NULL) | |
3833 | return; | |
3834 | ||
3835 | BUG_ON(atomic_read(&ioc->refcount) == 0); | |
3836 | ||
3837 | if (atomic_dec_and_test(&ioc->refcount)) { | |
3838 | struct cfq_io_context *cic; | |
3839 | ||
3840 | rcu_read_lock(); | |
3841 | if (ioc->aic && ioc->aic->dtor) | |
3842 | ioc->aic->dtor(ioc->aic); | |
3843 | if (ioc->cic_root.rb_node != NULL) { | |
3844 | struct rb_node *n = rb_first(&ioc->cic_root); | |
3845 | ||
3846 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
3847 | cic->dtor(ioc); | |
3848 | } | |
3849 | rcu_read_unlock(); | |
3850 | ||
3851 | kmem_cache_free(iocontext_cachep, ioc); | |
3852 | } | |
3853 | } | |
3854 | EXPORT_SYMBOL(put_io_context); | |
3855 | ||
3856 | /* Called by the exitting task */ | |
3857 | void exit_io_context(void) | |
3858 | { | |
3859 | struct io_context *ioc; | |
3860 | struct cfq_io_context *cic; | |
3861 | ||
3862 | task_lock(current); | |
3863 | ioc = current->io_context; | |
3864 | current->io_context = NULL; | |
3865 | task_unlock(current); | |
3866 | ||
3867 | ioc->task = NULL; | |
3868 | if (ioc->aic && ioc->aic->exit) | |
3869 | ioc->aic->exit(ioc->aic); | |
3870 | if (ioc->cic_root.rb_node != NULL) { | |
3871 | cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node); | |
3872 | cic->exit(ioc); | |
3873 | } | |
3874 | ||
3875 | put_io_context(ioc); | |
3876 | } | |
3877 | ||
3878 | /* | |
3879 | * If the current task has no IO context then create one and initialise it. | |
3880 | * Otherwise, return its existing IO context. | |
3881 | * | |
3882 | * This returned IO context doesn't have a specifically elevated refcount, | |
3883 | * but since the current task itself holds a reference, the context can be | |
3884 | * used in general code, so long as it stays within `current` context. | |
3885 | */ | |
3886 | static struct io_context *current_io_context(gfp_t gfp_flags, int node) | |
3887 | { | |
3888 | struct task_struct *tsk = current; | |
3889 | struct io_context *ret; | |
3890 | ||
3891 | ret = tsk->io_context; | |
3892 | if (likely(ret)) | |
3893 | return ret; | |
3894 | ||
3895 | ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node); | |
3896 | if (ret) { | |
3897 | atomic_set(&ret->refcount, 1); | |
3898 | ret->task = current; | |
3899 | ret->ioprio_changed = 0; | |
3900 | ret->last_waited = jiffies; /* doesn't matter... */ | |
3901 | ret->nr_batch_requests = 0; /* because this is 0 */ | |
3902 | ret->aic = NULL; | |
3903 | ret->cic_root.rb_node = NULL; | |
3904 | ret->ioc_data = NULL; | |
3905 | /* make sure set_task_ioprio() sees the settings above */ | |
3906 | smp_wmb(); | |
3907 | tsk->io_context = ret; | |
3908 | } | |
3909 | ||
3910 | return ret; | |
3911 | } | |
3912 | ||
3913 | /* | |
3914 | * If the current task has no IO context then create one and initialise it. | |
3915 | * If it does have a context, take a ref on it. | |
3916 | * | |
3917 | * This is always called in the context of the task which submitted the I/O. | |
3918 | */ | |
3919 | struct io_context *get_io_context(gfp_t gfp_flags, int node) | |
3920 | { | |
3921 | struct io_context *ret; | |
3922 | ret = current_io_context(gfp_flags, node); | |
3923 | if (likely(ret)) | |
3924 | atomic_inc(&ret->refcount); | |
3925 | return ret; | |
3926 | } | |
3927 | EXPORT_SYMBOL(get_io_context); | |
3928 | ||
3929 | void copy_io_context(struct io_context **pdst, struct io_context **psrc) | |
3930 | { | |
3931 | struct io_context *src = *psrc; | |
3932 | struct io_context *dst = *pdst; | |
3933 | ||
3934 | if (src) { | |
3935 | BUG_ON(atomic_read(&src->refcount) == 0); | |
3936 | atomic_inc(&src->refcount); | |
3937 | put_io_context(dst); | |
3938 | *pdst = src; | |
3939 | } | |
3940 | } | |
3941 | EXPORT_SYMBOL(copy_io_context); | |
3942 | ||
3943 | void swap_io_context(struct io_context **ioc1, struct io_context **ioc2) | |
3944 | { | |
3945 | struct io_context *temp; | |
3946 | temp = *ioc1; | |
3947 | *ioc1 = *ioc2; | |
3948 | *ioc2 = temp; | |
3949 | } | |
3950 | EXPORT_SYMBOL(swap_io_context); | |
3951 | ||
3952 | /* | |
3953 | * sysfs parts below | |
3954 | */ | |
3955 | struct queue_sysfs_entry { | |
3956 | struct attribute attr; | |
3957 | ssize_t (*show)(struct request_queue *, char *); | |
3958 | ssize_t (*store)(struct request_queue *, const char *, size_t); | |
3959 | }; | |
3960 | ||
3961 | static ssize_t | |
3962 | queue_var_show(unsigned int var, char *page) | |
3963 | { | |
3964 | return sprintf(page, "%d\n", var); | |
3965 | } | |
3966 | ||
3967 | static ssize_t | |
3968 | queue_var_store(unsigned long *var, const char *page, size_t count) | |
3969 | { | |
3970 | char *p = (char *) page; | |
3971 | ||
3972 | *var = simple_strtoul(p, &p, 10); | |
3973 | return count; | |
3974 | } | |
3975 | ||
3976 | static ssize_t queue_requests_show(struct request_queue *q, char *page) | |
3977 | { | |
3978 | return queue_var_show(q->nr_requests, (page)); | |
3979 | } | |
3980 | ||
3981 | static ssize_t | |
3982 | queue_requests_store(struct request_queue *q, const char *page, size_t count) | |
3983 | { | |
3984 | struct request_list *rl = &q->rq; | |
3985 | unsigned long nr; | |
3986 | int ret = queue_var_store(&nr, page, count); | |
3987 | if (nr < BLKDEV_MIN_RQ) | |
3988 | nr = BLKDEV_MIN_RQ; | |
3989 | ||
3990 | spin_lock_irq(q->queue_lock); | |
3991 | q->nr_requests = nr; | |
3992 | blk_queue_congestion_threshold(q); | |
3993 | ||
3994 | if (rl->count[READ] >= queue_congestion_on_threshold(q)) | |
3995 | blk_set_queue_congested(q, READ); | |
3996 | else if (rl->count[READ] < queue_congestion_off_threshold(q)) | |
3997 | blk_clear_queue_congested(q, READ); | |
3998 | ||
3999 | if (rl->count[WRITE] >= queue_congestion_on_threshold(q)) | |
4000 | blk_set_queue_congested(q, WRITE); | |
4001 | else if (rl->count[WRITE] < queue_congestion_off_threshold(q)) | |
4002 | blk_clear_queue_congested(q, WRITE); | |
4003 | ||
4004 | if (rl->count[READ] >= q->nr_requests) { | |
4005 | blk_set_queue_full(q, READ); | |
4006 | } else if (rl->count[READ]+1 <= q->nr_requests) { | |
4007 | blk_clear_queue_full(q, READ); | |
4008 | wake_up(&rl->wait[READ]); | |
4009 | } | |
4010 | ||
4011 | if (rl->count[WRITE] >= q->nr_requests) { | |
4012 | blk_set_queue_full(q, WRITE); | |
4013 | } else if (rl->count[WRITE]+1 <= q->nr_requests) { | |
4014 | blk_clear_queue_full(q, WRITE); | |
4015 | wake_up(&rl->wait[WRITE]); | |
4016 | } | |
4017 | spin_unlock_irq(q->queue_lock); | |
4018 | return ret; | |
4019 | } | |
4020 | ||
4021 | static ssize_t queue_ra_show(struct request_queue *q, char *page) | |
4022 | { | |
4023 | int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | |
4024 | ||
4025 | return queue_var_show(ra_kb, (page)); | |
4026 | } | |
4027 | ||
4028 | static ssize_t | |
4029 | queue_ra_store(struct request_queue *q, const char *page, size_t count) | |
4030 | { | |
4031 | unsigned long ra_kb; | |
4032 | ssize_t ret = queue_var_store(&ra_kb, page, count); | |
4033 | ||
4034 | spin_lock_irq(q->queue_lock); | |
4035 | q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); | |
4036 | spin_unlock_irq(q->queue_lock); | |
4037 | ||
4038 | return ret; | |
4039 | } | |
4040 | ||
4041 | static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) | |
4042 | { | |
4043 | int max_sectors_kb = q->max_sectors >> 1; | |
4044 | ||
4045 | return queue_var_show(max_sectors_kb, (page)); | |
4046 | } | |
4047 | ||
4048 | static ssize_t | |
4049 | queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) | |
4050 | { | |
4051 | unsigned long max_sectors_kb, | |
4052 | max_hw_sectors_kb = q->max_hw_sectors >> 1, | |
4053 | page_kb = 1 << (PAGE_CACHE_SHIFT - 10); | |
4054 | ssize_t ret = queue_var_store(&max_sectors_kb, page, count); | |
4055 | ||
4056 | if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) | |
4057 | return -EINVAL; | |
4058 | /* | |
4059 | * Take the queue lock to update the readahead and max_sectors | |
4060 | * values synchronously: | |
4061 | */ | |
4062 | spin_lock_irq(q->queue_lock); | |
4063 | q->max_sectors = max_sectors_kb << 1; | |
4064 | spin_unlock_irq(q->queue_lock); | |
4065 | ||
4066 | return ret; | |
4067 | } | |
4068 | ||
4069 | static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) | |
4070 | { | |
4071 | int max_hw_sectors_kb = q->max_hw_sectors >> 1; | |
4072 | ||
4073 | return queue_var_show(max_hw_sectors_kb, (page)); | |
4074 | } | |
4075 | ||
4076 | static ssize_t queue_max_segments_show(struct request_queue *q, char *page) | |
4077 | { | |
4078 | return queue_var_show(q->max_phys_segments, page); | |
4079 | } | |
4080 | ||
4081 | static ssize_t queue_max_segments_store(struct request_queue *q, | |
4082 | const char *page, size_t count) | |
4083 | { | |
4084 | unsigned long segments; | |
4085 | ssize_t ret = queue_var_store(&segments, page, count); | |
4086 | ||
4087 | spin_lock_irq(q->queue_lock); | |
4088 | q->max_phys_segments = segments; | |
4089 | spin_unlock_irq(q->queue_lock); | |
4090 | ||
4091 | return ret; | |
4092 | } | |
4093 | static struct queue_sysfs_entry queue_requests_entry = { | |
4094 | .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, | |
4095 | .show = queue_requests_show, | |
4096 | .store = queue_requests_store, | |
4097 | }; | |
4098 | ||
4099 | static struct queue_sysfs_entry queue_ra_entry = { | |
4100 | .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, | |
4101 | .show = queue_ra_show, | |
4102 | .store = queue_ra_store, | |
4103 | }; | |
4104 | ||
4105 | static struct queue_sysfs_entry queue_max_sectors_entry = { | |
4106 | .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, | |
4107 | .show = queue_max_sectors_show, | |
4108 | .store = queue_max_sectors_store, | |
4109 | }; | |
4110 | ||
4111 | static struct queue_sysfs_entry queue_max_hw_sectors_entry = { | |
4112 | .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, | |
4113 | .show = queue_max_hw_sectors_show, | |
4114 | }; | |
4115 | ||
4116 | static struct queue_sysfs_entry queue_max_segments_entry = { | |
4117 | .attr = {.name = "max_segments", .mode = S_IRUGO | S_IWUSR }, | |
4118 | .show = queue_max_segments_show, | |
4119 | .store = queue_max_segments_store, | |
4120 | }; | |
4121 | ||
4122 | static struct queue_sysfs_entry queue_iosched_entry = { | |
4123 | .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, | |
4124 | .show = elv_iosched_show, | |
4125 | .store = elv_iosched_store, | |
4126 | }; | |
4127 | ||
4128 | static struct attribute *default_attrs[] = { | |
4129 | &queue_requests_entry.attr, | |
4130 | &queue_ra_entry.attr, | |
4131 | &queue_max_hw_sectors_entry.attr, | |
4132 | &queue_max_sectors_entry.attr, | |
4133 | &queue_max_segments_entry.attr, | |
4134 | &queue_iosched_entry.attr, | |
4135 | NULL, | |
4136 | }; | |
4137 | ||
4138 | #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) | |
4139 | ||
4140 | static ssize_t | |
4141 | queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4142 | { | |
4143 | struct queue_sysfs_entry *entry = to_queue(attr); | |
4144 | struct request_queue *q = | |
4145 | container_of(kobj, struct request_queue, kobj); | |
4146 | ssize_t res; | |
4147 | ||
4148 | if (!entry->show) | |
4149 | return -EIO; | |
4150 | mutex_lock(&q->sysfs_lock); | |
4151 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { | |
4152 | mutex_unlock(&q->sysfs_lock); | |
4153 | return -ENOENT; | |
4154 | } | |
4155 | res = entry->show(q, page); | |
4156 | mutex_unlock(&q->sysfs_lock); | |
4157 | return res; | |
4158 | } | |
4159 | ||
4160 | static ssize_t | |
4161 | queue_attr_store(struct kobject *kobj, struct attribute *attr, | |
4162 | const char *page, size_t length) | |
4163 | { | |
4164 | struct queue_sysfs_entry *entry = to_queue(attr); | |
4165 | struct request_queue *q = container_of(kobj, struct request_queue, kobj); | |
4166 | ||
4167 | ssize_t res; | |
4168 | ||
4169 | if (!entry->store) | |
4170 | return -EIO; | |
4171 | mutex_lock(&q->sysfs_lock); | |
4172 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { | |
4173 | mutex_unlock(&q->sysfs_lock); | |
4174 | return -ENOENT; | |
4175 | } | |
4176 | res = entry->store(q, page, length); | |
4177 | mutex_unlock(&q->sysfs_lock); | |
4178 | return res; | |
4179 | } | |
4180 | ||
4181 | static struct sysfs_ops queue_sysfs_ops = { | |
4182 | .show = queue_attr_show, | |
4183 | .store = queue_attr_store, | |
4184 | }; | |
4185 | ||
4186 | static struct kobj_type queue_ktype = { | |
4187 | .sysfs_ops = &queue_sysfs_ops, | |
4188 | .default_attrs = default_attrs, | |
4189 | .release = blk_release_queue, | |
4190 | }; | |
4191 | ||
4192 | int blk_register_queue(struct gendisk *disk) | |
4193 | { | |
4194 | int ret; | |
4195 | ||
4196 | struct request_queue *q = disk->queue; | |
4197 | ||
4198 | if (!q || !q->request_fn) | |
4199 | return -ENXIO; | |
4200 | ||
4201 | q->kobj.parent = kobject_get(&disk->kobj); | |
4202 | ||
4203 | ret = kobject_add(&q->kobj); | |
4204 | if (ret < 0) | |
4205 | return ret; | |
4206 | ||
4207 | kobject_uevent(&q->kobj, KOBJ_ADD); | |
4208 | ||
4209 | ret = elv_register_queue(q); | |
4210 | if (ret) { | |
4211 | kobject_uevent(&q->kobj, KOBJ_REMOVE); | |
4212 | kobject_del(&q->kobj); | |
4213 | return ret; | |
4214 | } | |
4215 | ||
4216 | return 0; | |
4217 | } | |
4218 | ||
4219 | void blk_unregister_queue(struct gendisk *disk) | |
4220 | { | |
4221 | struct request_queue *q = disk->queue; | |
4222 | ||
4223 | if (q && q->request_fn) { | |
4224 | elv_unregister_queue(q); | |
4225 | ||
4226 | kobject_uevent(&q->kobj, KOBJ_REMOVE); | |
4227 | kobject_del(&q->kobj); | |
4228 | kobject_put(&disk->kobj); | |
4229 | } | |
4230 | } |