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