]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/drivers/block/ll_rw_blk.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
6 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
7 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
8 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000 | |
9 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 | |
10 | */ | |
11 | ||
12 | /* | |
13 | * This handles all read/write requests to block devices | |
14 | */ | |
15 | #include <linux/config.h> | |
16 | #include <linux/kernel.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/backing-dev.h> | |
19 | #include <linux/bio.h> | |
20 | #include <linux/blkdev.h> | |
21 | #include <linux/highmem.h> | |
22 | #include <linux/mm.h> | |
23 | #include <linux/kernel_stat.h> | |
24 | #include <linux/string.h> | |
25 | #include <linux/init.h> | |
26 | #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */ | |
27 | #include <linux/completion.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/swap.h> | |
30 | #include <linux/writeback.h> | |
1946089a | 31 | #include <linux/blkdev.h> |
1da177e4 LT |
32 | |
33 | /* | |
34 | * for max sense size | |
35 | */ | |
36 | #include <scsi/scsi_cmnd.h> | |
37 | ||
38 | static void blk_unplug_work(void *data); | |
39 | static void blk_unplug_timeout(unsigned long data); | |
93d17d3d | 40 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io); |
1da177e4 LT |
41 | |
42 | /* | |
43 | * For the allocated request tables | |
44 | */ | |
45 | static kmem_cache_t *request_cachep; | |
46 | ||
47 | /* | |
48 | * For queue allocation | |
49 | */ | |
50 | static kmem_cache_t *requestq_cachep; | |
51 | ||
52 | /* | |
53 | * For io context allocations | |
54 | */ | |
55 | static kmem_cache_t *iocontext_cachep; | |
56 | ||
57 | static wait_queue_head_t congestion_wqh[2] = { | |
58 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
59 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
60 | }; | |
61 | ||
62 | /* | |
63 | * Controlling structure to kblockd | |
64 | */ | |
65 | static struct workqueue_struct *kblockd_workqueue; | |
66 | ||
67 | unsigned long blk_max_low_pfn, blk_max_pfn; | |
68 | ||
69 | EXPORT_SYMBOL(blk_max_low_pfn); | |
70 | EXPORT_SYMBOL(blk_max_pfn); | |
71 | ||
72 | /* Amount of time in which a process may batch requests */ | |
73 | #define BLK_BATCH_TIME (HZ/50UL) | |
74 | ||
75 | /* Number of requests a "batching" process may submit */ | |
76 | #define BLK_BATCH_REQ 32 | |
77 | ||
78 | /* | |
79 | * Return the threshold (number of used requests) at which the queue is | |
80 | * considered to be congested. It include a little hysteresis to keep the | |
81 | * context switch rate down. | |
82 | */ | |
83 | static inline int queue_congestion_on_threshold(struct request_queue *q) | |
84 | { | |
85 | return q->nr_congestion_on; | |
86 | } | |
87 | ||
88 | /* | |
89 | * The threshold at which a queue is considered to be uncongested | |
90 | */ | |
91 | static inline int queue_congestion_off_threshold(struct request_queue *q) | |
92 | { | |
93 | return q->nr_congestion_off; | |
94 | } | |
95 | ||
96 | static void blk_queue_congestion_threshold(struct request_queue *q) | |
97 | { | |
98 | int nr; | |
99 | ||
100 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
101 | if (nr > q->nr_requests) | |
102 | nr = q->nr_requests; | |
103 | q->nr_congestion_on = nr; | |
104 | ||
105 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
106 | if (nr < 1) | |
107 | nr = 1; | |
108 | q->nr_congestion_off = nr; | |
109 | } | |
110 | ||
111 | /* | |
112 | * A queue has just exitted congestion. Note this in the global counter of | |
113 | * congested queues, and wake up anyone who was waiting for requests to be | |
114 | * put back. | |
115 | */ | |
116 | static void clear_queue_congested(request_queue_t *q, int rw) | |
117 | { | |
118 | enum bdi_state bit; | |
119 | wait_queue_head_t *wqh = &congestion_wqh[rw]; | |
120 | ||
121 | bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested; | |
122 | clear_bit(bit, &q->backing_dev_info.state); | |
123 | smp_mb__after_clear_bit(); | |
124 | if (waitqueue_active(wqh)) | |
125 | wake_up(wqh); | |
126 | } | |
127 | ||
128 | /* | |
129 | * A queue has just entered congestion. Flag that in the queue's VM-visible | |
130 | * state flags and increment the global gounter of congested queues. | |
131 | */ | |
132 | static void set_queue_congested(request_queue_t *q, int rw) | |
133 | { | |
134 | enum bdi_state bit; | |
135 | ||
136 | bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested; | |
137 | set_bit(bit, &q->backing_dev_info.state); | |
138 | } | |
139 | ||
140 | /** | |
141 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
142 | * @bdev: device | |
143 | * | |
144 | * Locates the passed device's request queue and returns the address of its | |
145 | * backing_dev_info | |
146 | * | |
147 | * Will return NULL if the request queue cannot be located. | |
148 | */ | |
149 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
150 | { | |
151 | struct backing_dev_info *ret = NULL; | |
152 | request_queue_t *q = bdev_get_queue(bdev); | |
153 | ||
154 | if (q) | |
155 | ret = &q->backing_dev_info; | |
156 | return ret; | |
157 | } | |
158 | ||
159 | EXPORT_SYMBOL(blk_get_backing_dev_info); | |
160 | ||
161 | void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data) | |
162 | { | |
163 | q->activity_fn = fn; | |
164 | q->activity_data = data; | |
165 | } | |
166 | ||
167 | EXPORT_SYMBOL(blk_queue_activity_fn); | |
168 | ||
169 | /** | |
170 | * blk_queue_prep_rq - set a prepare_request function for queue | |
171 | * @q: queue | |
172 | * @pfn: prepare_request function | |
173 | * | |
174 | * It's possible for a queue to register a prepare_request callback which | |
175 | * is invoked before the request is handed to the request_fn. The goal of | |
176 | * the function is to prepare a request for I/O, it can be used to build a | |
177 | * cdb from the request data for instance. | |
178 | * | |
179 | */ | |
180 | void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn) | |
181 | { | |
182 | q->prep_rq_fn = pfn; | |
183 | } | |
184 | ||
185 | EXPORT_SYMBOL(blk_queue_prep_rq); | |
186 | ||
187 | /** | |
188 | * blk_queue_merge_bvec - set a merge_bvec function for queue | |
189 | * @q: queue | |
190 | * @mbfn: merge_bvec_fn | |
191 | * | |
192 | * Usually queues have static limitations on the max sectors or segments that | |
193 | * we can put in a request. Stacking drivers may have some settings that | |
194 | * are dynamic, and thus we have to query the queue whether it is ok to | |
195 | * add a new bio_vec to a bio at a given offset or not. If the block device | |
196 | * has such limitations, it needs to register a merge_bvec_fn to control | |
197 | * the size of bio's sent to it. Note that a block device *must* allow a | |
198 | * single page to be added to an empty bio. The block device driver may want | |
199 | * to use the bio_split() function to deal with these bio's. By default | |
200 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | |
201 | * honored. | |
202 | */ | |
203 | void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn) | |
204 | { | |
205 | q->merge_bvec_fn = mbfn; | |
206 | } | |
207 | ||
208 | EXPORT_SYMBOL(blk_queue_merge_bvec); | |
209 | ||
210 | /** | |
211 | * blk_queue_make_request - define an alternate make_request function for a device | |
212 | * @q: the request queue for the device to be affected | |
213 | * @mfn: the alternate make_request function | |
214 | * | |
215 | * Description: | |
216 | * The normal way for &struct bios to be passed to a device | |
217 | * driver is for them to be collected into requests on a request | |
218 | * queue, and then to allow the device driver to select requests | |
219 | * off that queue when it is ready. This works well for many block | |
220 | * devices. However some block devices (typically virtual devices | |
221 | * such as md or lvm) do not benefit from the processing on the | |
222 | * request queue, and are served best by having the requests passed | |
223 | * directly to them. This can be achieved by providing a function | |
224 | * to blk_queue_make_request(). | |
225 | * | |
226 | * Caveat: | |
227 | * The driver that does this *must* be able to deal appropriately | |
228 | * with buffers in "highmemory". This can be accomplished by either calling | |
229 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
230 | * blk_queue_bounce() to create a buffer in normal memory. | |
231 | **/ | |
232 | void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn) | |
233 | { | |
234 | /* | |
235 | * set defaults | |
236 | */ | |
237 | q->nr_requests = BLKDEV_MAX_RQ; | |
309c0a1d SM |
238 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); |
239 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
1da177e4 LT |
240 | q->make_request_fn = mfn; |
241 | q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
242 | q->backing_dev_info.state = 0; | |
243 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
244 | blk_queue_max_sectors(q, MAX_SECTORS); | |
245 | blk_queue_hardsect_size(q, 512); | |
246 | blk_queue_dma_alignment(q, 511); | |
247 | blk_queue_congestion_threshold(q); | |
248 | q->nr_batching = BLK_BATCH_REQ; | |
249 | ||
250 | q->unplug_thresh = 4; /* hmm */ | |
251 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ | |
252 | if (q->unplug_delay == 0) | |
253 | q->unplug_delay = 1; | |
254 | ||
255 | INIT_WORK(&q->unplug_work, blk_unplug_work, q); | |
256 | ||
257 | q->unplug_timer.function = blk_unplug_timeout; | |
258 | q->unplug_timer.data = (unsigned long)q; | |
259 | ||
260 | /* | |
261 | * by default assume old behaviour and bounce for any highmem page | |
262 | */ | |
263 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
264 | ||
265 | blk_queue_activity_fn(q, NULL, NULL); | |
266 | ||
267 | INIT_LIST_HEAD(&q->drain_list); | |
268 | } | |
269 | ||
270 | EXPORT_SYMBOL(blk_queue_make_request); | |
271 | ||
272 | static inline void rq_init(request_queue_t *q, struct request *rq) | |
273 | { | |
274 | INIT_LIST_HEAD(&rq->queuelist); | |
275 | ||
276 | rq->errors = 0; | |
277 | rq->rq_status = RQ_ACTIVE; | |
278 | rq->bio = rq->biotail = NULL; | |
22e2c507 | 279 | rq->ioprio = 0; |
1da177e4 LT |
280 | rq->buffer = NULL; |
281 | rq->ref_count = 1; | |
282 | rq->q = q; | |
283 | rq->waiting = NULL; | |
284 | rq->special = NULL; | |
285 | rq->data_len = 0; | |
286 | rq->data = NULL; | |
df46b9a4 | 287 | rq->nr_phys_segments = 0; |
1da177e4 LT |
288 | rq->sense = NULL; |
289 | rq->end_io = NULL; | |
290 | rq->end_io_data = NULL; | |
291 | } | |
292 | ||
293 | /** | |
294 | * blk_queue_ordered - does this queue support ordered writes | |
295 | * @q: the request queue | |
296 | * @flag: see below | |
297 | * | |
298 | * Description: | |
299 | * For journalled file systems, doing ordered writes on a commit | |
300 | * block instead of explicitly doing wait_on_buffer (which is bad | |
301 | * for performance) can be a big win. Block drivers supporting this | |
302 | * feature should call this function and indicate so. | |
303 | * | |
304 | **/ | |
305 | void blk_queue_ordered(request_queue_t *q, int flag) | |
306 | { | |
307 | switch (flag) { | |
308 | case QUEUE_ORDERED_NONE: | |
309 | if (q->flush_rq) | |
310 | kmem_cache_free(request_cachep, q->flush_rq); | |
311 | q->flush_rq = NULL; | |
312 | q->ordered = flag; | |
313 | break; | |
314 | case QUEUE_ORDERED_TAG: | |
315 | q->ordered = flag; | |
316 | break; | |
317 | case QUEUE_ORDERED_FLUSH: | |
318 | q->ordered = flag; | |
319 | if (!q->flush_rq) | |
320 | q->flush_rq = kmem_cache_alloc(request_cachep, | |
321 | GFP_KERNEL); | |
322 | break; | |
323 | default: | |
324 | printk("blk_queue_ordered: bad value %d\n", flag); | |
325 | break; | |
326 | } | |
327 | } | |
328 | ||
329 | EXPORT_SYMBOL(blk_queue_ordered); | |
330 | ||
331 | /** | |
332 | * blk_queue_issue_flush_fn - set function for issuing a flush | |
333 | * @q: the request queue | |
334 | * @iff: the function to be called issuing the flush | |
335 | * | |
336 | * Description: | |
337 | * If a driver supports issuing a flush command, the support is notified | |
338 | * to the block layer by defining it through this call. | |
339 | * | |
340 | **/ | |
341 | void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff) | |
342 | { | |
343 | q->issue_flush_fn = iff; | |
344 | } | |
345 | ||
346 | EXPORT_SYMBOL(blk_queue_issue_flush_fn); | |
347 | ||
348 | /* | |
349 | * Cache flushing for ordered writes handling | |
350 | */ | |
351 | static void blk_pre_flush_end_io(struct request *flush_rq) | |
352 | { | |
353 | struct request *rq = flush_rq->end_io_data; | |
354 | request_queue_t *q = rq->q; | |
355 | ||
356 | rq->flags |= REQ_BAR_PREFLUSH; | |
357 | ||
358 | if (!flush_rq->errors) | |
359 | elv_requeue_request(q, rq); | |
360 | else { | |
361 | q->end_flush_fn(q, flush_rq); | |
362 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | |
363 | q->request_fn(q); | |
364 | } | |
365 | } | |
366 | ||
367 | static void blk_post_flush_end_io(struct request *flush_rq) | |
368 | { | |
369 | struct request *rq = flush_rq->end_io_data; | |
370 | request_queue_t *q = rq->q; | |
371 | ||
372 | rq->flags |= REQ_BAR_POSTFLUSH; | |
373 | ||
374 | q->end_flush_fn(q, flush_rq); | |
375 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | |
376 | q->request_fn(q); | |
377 | } | |
378 | ||
379 | struct request *blk_start_pre_flush(request_queue_t *q, struct request *rq) | |
380 | { | |
381 | struct request *flush_rq = q->flush_rq; | |
382 | ||
383 | BUG_ON(!blk_barrier_rq(rq)); | |
384 | ||
385 | if (test_and_set_bit(QUEUE_FLAG_FLUSH, &q->queue_flags)) | |
386 | return NULL; | |
387 | ||
388 | rq_init(q, flush_rq); | |
389 | flush_rq->elevator_private = NULL; | |
390 | flush_rq->flags = REQ_BAR_FLUSH; | |
391 | flush_rq->rq_disk = rq->rq_disk; | |
392 | flush_rq->rl = NULL; | |
393 | ||
394 | /* | |
395 | * prepare_flush returns 0 if no flush is needed, just mark both | |
396 | * pre and post flush as done in that case | |
397 | */ | |
398 | if (!q->prepare_flush_fn(q, flush_rq)) { | |
399 | rq->flags |= REQ_BAR_PREFLUSH | REQ_BAR_POSTFLUSH; | |
400 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | |
401 | return rq; | |
402 | } | |
403 | ||
404 | /* | |
405 | * some drivers dequeue requests right away, some only after io | |
406 | * completion. make sure the request is dequeued. | |
407 | */ | |
408 | if (!list_empty(&rq->queuelist)) | |
409 | blkdev_dequeue_request(rq); | |
410 | ||
411 | elv_deactivate_request(q, rq); | |
412 | ||
413 | flush_rq->end_io_data = rq; | |
414 | flush_rq->end_io = blk_pre_flush_end_io; | |
415 | ||
416 | __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0); | |
417 | return flush_rq; | |
418 | } | |
419 | ||
420 | static void blk_start_post_flush(request_queue_t *q, struct request *rq) | |
421 | { | |
422 | struct request *flush_rq = q->flush_rq; | |
423 | ||
424 | BUG_ON(!blk_barrier_rq(rq)); | |
425 | ||
426 | rq_init(q, flush_rq); | |
427 | flush_rq->elevator_private = NULL; | |
428 | flush_rq->flags = REQ_BAR_FLUSH; | |
429 | flush_rq->rq_disk = rq->rq_disk; | |
430 | flush_rq->rl = NULL; | |
431 | ||
432 | if (q->prepare_flush_fn(q, flush_rq)) { | |
433 | flush_rq->end_io_data = rq; | |
434 | flush_rq->end_io = blk_post_flush_end_io; | |
435 | ||
436 | __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0); | |
437 | q->request_fn(q); | |
438 | } | |
439 | } | |
440 | ||
441 | static inline int blk_check_end_barrier(request_queue_t *q, struct request *rq, | |
442 | int sectors) | |
443 | { | |
444 | if (sectors > rq->nr_sectors) | |
445 | sectors = rq->nr_sectors; | |
446 | ||
447 | rq->nr_sectors -= sectors; | |
448 | return rq->nr_sectors; | |
449 | } | |
450 | ||
451 | static int __blk_complete_barrier_rq(request_queue_t *q, struct request *rq, | |
452 | int sectors, int queue_locked) | |
453 | { | |
454 | if (q->ordered != QUEUE_ORDERED_FLUSH) | |
455 | return 0; | |
456 | if (!blk_fs_request(rq) || !blk_barrier_rq(rq)) | |
457 | return 0; | |
458 | if (blk_barrier_postflush(rq)) | |
459 | return 0; | |
460 | ||
461 | if (!blk_check_end_barrier(q, rq, sectors)) { | |
462 | unsigned long flags = 0; | |
463 | ||
464 | if (!queue_locked) | |
465 | spin_lock_irqsave(q->queue_lock, flags); | |
466 | ||
467 | blk_start_post_flush(q, rq); | |
468 | ||
469 | if (!queue_locked) | |
470 | spin_unlock_irqrestore(q->queue_lock, flags); | |
471 | } | |
472 | ||
473 | return 1; | |
474 | } | |
475 | ||
476 | /** | |
477 | * blk_complete_barrier_rq - complete possible barrier request | |
478 | * @q: the request queue for the device | |
479 | * @rq: the request | |
480 | * @sectors: number of sectors to complete | |
481 | * | |
482 | * Description: | |
483 | * Used in driver end_io handling to determine whether to postpone | |
484 | * completion of a barrier request until a post flush has been done. This | |
485 | * is the unlocked variant, used if the caller doesn't already hold the | |
486 | * queue lock. | |
487 | **/ | |
488 | int blk_complete_barrier_rq(request_queue_t *q, struct request *rq, int sectors) | |
489 | { | |
490 | return __blk_complete_barrier_rq(q, rq, sectors, 0); | |
491 | } | |
492 | EXPORT_SYMBOL(blk_complete_barrier_rq); | |
493 | ||
494 | /** | |
495 | * blk_complete_barrier_rq_locked - complete possible barrier request | |
496 | * @q: the request queue for the device | |
497 | * @rq: the request | |
498 | * @sectors: number of sectors to complete | |
499 | * | |
500 | * Description: | |
501 | * See blk_complete_barrier_rq(). This variant must be used if the caller | |
502 | * holds the queue lock. | |
503 | **/ | |
504 | int blk_complete_barrier_rq_locked(request_queue_t *q, struct request *rq, | |
505 | int sectors) | |
506 | { | |
507 | return __blk_complete_barrier_rq(q, rq, sectors, 1); | |
508 | } | |
509 | EXPORT_SYMBOL(blk_complete_barrier_rq_locked); | |
510 | ||
511 | /** | |
512 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
513 | * @q: the request queue for the device | |
514 | * @dma_addr: bus address limit | |
515 | * | |
516 | * Description: | |
517 | * Different hardware can have different requirements as to what pages | |
518 | * it can do I/O directly to. A low level driver can call | |
519 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
520 | * buffers for doing I/O to pages residing above @page. By default | |
521 | * the block layer sets this to the highest numbered "low" memory page. | |
522 | **/ | |
523 | void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr) | |
524 | { | |
525 | unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT; | |
526 | ||
527 | /* | |
528 | * set appropriate bounce gfp mask -- unfortunately we don't have a | |
529 | * full 4GB zone, so we have to resort to low memory for any bounces. | |
530 | * ISA has its own < 16MB zone. | |
531 | */ | |
532 | if (bounce_pfn < blk_max_low_pfn) { | |
533 | BUG_ON(dma_addr < BLK_BOUNCE_ISA); | |
534 | init_emergency_isa_pool(); | |
535 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
536 | } else | |
537 | q->bounce_gfp = GFP_NOIO; | |
538 | ||
539 | q->bounce_pfn = bounce_pfn; | |
540 | } | |
541 | ||
542 | EXPORT_SYMBOL(blk_queue_bounce_limit); | |
543 | ||
544 | /** | |
545 | * blk_queue_max_sectors - set max sectors for a request for this queue | |
546 | * @q: the request queue for the device | |
547 | * @max_sectors: max sectors in the usual 512b unit | |
548 | * | |
549 | * Description: | |
550 | * Enables a low level driver to set an upper limit on the size of | |
551 | * received requests. | |
552 | **/ | |
553 | void blk_queue_max_sectors(request_queue_t *q, unsigned short max_sectors) | |
554 | { | |
555 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { | |
556 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
557 | printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors); | |
558 | } | |
559 | ||
560 | q->max_sectors = q->max_hw_sectors = max_sectors; | |
561 | } | |
562 | ||
563 | EXPORT_SYMBOL(blk_queue_max_sectors); | |
564 | ||
565 | /** | |
566 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue | |
567 | * @q: the request queue for the device | |
568 | * @max_segments: max number of segments | |
569 | * | |
570 | * Description: | |
571 | * Enables a low level driver to set an upper limit on the number of | |
572 | * physical data segments in a request. This would be the largest sized | |
573 | * scatter list the driver could handle. | |
574 | **/ | |
575 | void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments) | |
576 | { | |
577 | if (!max_segments) { | |
578 | max_segments = 1; | |
579 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
580 | } | |
581 | ||
582 | q->max_phys_segments = max_segments; | |
583 | } | |
584 | ||
585 | EXPORT_SYMBOL(blk_queue_max_phys_segments); | |
586 | ||
587 | /** | |
588 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue | |
589 | * @q: the request queue for the device | |
590 | * @max_segments: max number of segments | |
591 | * | |
592 | * Description: | |
593 | * Enables a low level driver to set an upper limit on the number of | |
594 | * hw data segments in a request. This would be the largest number of | |
595 | * address/length pairs the host adapter can actually give as once | |
596 | * to the device. | |
597 | **/ | |
598 | void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments) | |
599 | { | |
600 | if (!max_segments) { | |
601 | max_segments = 1; | |
602 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
603 | } | |
604 | ||
605 | q->max_hw_segments = max_segments; | |
606 | } | |
607 | ||
608 | EXPORT_SYMBOL(blk_queue_max_hw_segments); | |
609 | ||
610 | /** | |
611 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
612 | * @q: the request queue for the device | |
613 | * @max_size: max size of segment in bytes | |
614 | * | |
615 | * Description: | |
616 | * Enables a low level driver to set an upper limit on the size of a | |
617 | * coalesced segment | |
618 | **/ | |
619 | void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size) | |
620 | { | |
621 | if (max_size < PAGE_CACHE_SIZE) { | |
622 | max_size = PAGE_CACHE_SIZE; | |
623 | printk("%s: set to minimum %d\n", __FUNCTION__, max_size); | |
624 | } | |
625 | ||
626 | q->max_segment_size = max_size; | |
627 | } | |
628 | ||
629 | EXPORT_SYMBOL(blk_queue_max_segment_size); | |
630 | ||
631 | /** | |
632 | * blk_queue_hardsect_size - set hardware sector size for the queue | |
633 | * @q: the request queue for the device | |
634 | * @size: the hardware sector size, in bytes | |
635 | * | |
636 | * Description: | |
637 | * This should typically be set to the lowest possible sector size | |
638 | * that the hardware can operate on (possible without reverting to | |
639 | * even internal read-modify-write operations). Usually the default | |
640 | * of 512 covers most hardware. | |
641 | **/ | |
642 | void blk_queue_hardsect_size(request_queue_t *q, unsigned short size) | |
643 | { | |
644 | q->hardsect_size = size; | |
645 | } | |
646 | ||
647 | EXPORT_SYMBOL(blk_queue_hardsect_size); | |
648 | ||
649 | /* | |
650 | * Returns the minimum that is _not_ zero, unless both are zero. | |
651 | */ | |
652 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
653 | ||
654 | /** | |
655 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
656 | * @t: the stacking driver (top) | |
657 | * @b: the underlying device (bottom) | |
658 | **/ | |
659 | void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b) | |
660 | { | |
661 | /* zero is "infinity" */ | |
662 | t->max_sectors = t->max_hw_sectors = | |
663 | min_not_zero(t->max_sectors,b->max_sectors); | |
664 | ||
665 | t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments); | |
666 | t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments); | |
667 | t->max_segment_size = min(t->max_segment_size,b->max_segment_size); | |
668 | t->hardsect_size = max(t->hardsect_size,b->hardsect_size); | |
669 | } | |
670 | ||
671 | EXPORT_SYMBOL(blk_queue_stack_limits); | |
672 | ||
673 | /** | |
674 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
675 | * @q: the request queue for the device | |
676 | * @mask: the memory boundary mask | |
677 | **/ | |
678 | void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask) | |
679 | { | |
680 | if (mask < PAGE_CACHE_SIZE - 1) { | |
681 | mask = PAGE_CACHE_SIZE - 1; | |
682 | printk("%s: set to minimum %lx\n", __FUNCTION__, mask); | |
683 | } | |
684 | ||
685 | q->seg_boundary_mask = mask; | |
686 | } | |
687 | ||
688 | EXPORT_SYMBOL(blk_queue_segment_boundary); | |
689 | ||
690 | /** | |
691 | * blk_queue_dma_alignment - set dma length and memory alignment | |
692 | * @q: the request queue for the device | |
693 | * @mask: alignment mask | |
694 | * | |
695 | * description: | |
696 | * set required memory and length aligment for direct dma transactions. | |
697 | * this is used when buiding direct io requests for the queue. | |
698 | * | |
699 | **/ | |
700 | void blk_queue_dma_alignment(request_queue_t *q, int mask) | |
701 | { | |
702 | q->dma_alignment = mask; | |
703 | } | |
704 | ||
705 | EXPORT_SYMBOL(blk_queue_dma_alignment); | |
706 | ||
707 | /** | |
708 | * blk_queue_find_tag - find a request by its tag and queue | |
709 | * | |
710 | * @q: The request queue for the device | |
711 | * @tag: The tag of the request | |
712 | * | |
713 | * Notes: | |
714 | * Should be used when a device returns a tag and you want to match | |
715 | * it with a request. | |
716 | * | |
717 | * no locks need be held. | |
718 | **/ | |
719 | struct request *blk_queue_find_tag(request_queue_t *q, int tag) | |
720 | { | |
721 | struct blk_queue_tag *bqt = q->queue_tags; | |
722 | ||
ba025082 | 723 | if (unlikely(bqt == NULL || tag >= bqt->real_max_depth)) |
1da177e4 LT |
724 | return NULL; |
725 | ||
726 | return bqt->tag_index[tag]; | |
727 | } | |
728 | ||
729 | EXPORT_SYMBOL(blk_queue_find_tag); | |
730 | ||
731 | /** | |
732 | * __blk_queue_free_tags - release tag maintenance info | |
733 | * @q: the request queue for the device | |
734 | * | |
735 | * Notes: | |
736 | * blk_cleanup_queue() will take care of calling this function, if tagging | |
737 | * has been used. So there's no need to call this directly. | |
738 | **/ | |
739 | static void __blk_queue_free_tags(request_queue_t *q) | |
740 | { | |
741 | struct blk_queue_tag *bqt = q->queue_tags; | |
742 | ||
743 | if (!bqt) | |
744 | return; | |
745 | ||
746 | if (atomic_dec_and_test(&bqt->refcnt)) { | |
747 | BUG_ON(bqt->busy); | |
748 | BUG_ON(!list_empty(&bqt->busy_list)); | |
749 | ||
750 | kfree(bqt->tag_index); | |
751 | bqt->tag_index = NULL; | |
752 | ||
753 | kfree(bqt->tag_map); | |
754 | bqt->tag_map = NULL; | |
755 | ||
756 | kfree(bqt); | |
757 | } | |
758 | ||
759 | q->queue_tags = NULL; | |
760 | q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED); | |
761 | } | |
762 | ||
763 | /** | |
764 | * blk_queue_free_tags - release tag maintenance info | |
765 | * @q: the request queue for the device | |
766 | * | |
767 | * Notes: | |
768 | * This is used to disabled tagged queuing to a device, yet leave | |
769 | * queue in function. | |
770 | **/ | |
771 | void blk_queue_free_tags(request_queue_t *q) | |
772 | { | |
773 | clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
774 | } | |
775 | ||
776 | EXPORT_SYMBOL(blk_queue_free_tags); | |
777 | ||
778 | static int | |
779 | init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth) | |
780 | { | |
1da177e4 LT |
781 | struct request **tag_index; |
782 | unsigned long *tag_map; | |
fa72b903 | 783 | int nr_ulongs; |
1da177e4 LT |
784 | |
785 | if (depth > q->nr_requests * 2) { | |
786 | depth = q->nr_requests * 2; | |
787 | printk(KERN_ERR "%s: adjusted depth to %d\n", | |
788 | __FUNCTION__, depth); | |
789 | } | |
790 | ||
791 | tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC); | |
792 | if (!tag_index) | |
793 | goto fail; | |
794 | ||
f7d37d02 | 795 | nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG; |
fa72b903 | 796 | tag_map = kmalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC); |
1da177e4 LT |
797 | if (!tag_map) |
798 | goto fail; | |
799 | ||
800 | memset(tag_index, 0, depth * sizeof(struct request *)); | |
fa72b903 | 801 | memset(tag_map, 0, nr_ulongs * sizeof(unsigned long)); |
ba025082 | 802 | tags->real_max_depth = depth; |
1da177e4 | 803 | tags->max_depth = depth; |
1da177e4 LT |
804 | tags->tag_index = tag_index; |
805 | tags->tag_map = tag_map; | |
806 | ||
1da177e4 LT |
807 | return 0; |
808 | fail: | |
809 | kfree(tag_index); | |
810 | return -ENOMEM; | |
811 | } | |
812 | ||
813 | /** | |
814 | * blk_queue_init_tags - initialize the queue tag info | |
815 | * @q: the request queue for the device | |
816 | * @depth: the maximum queue depth supported | |
817 | * @tags: the tag to use | |
818 | **/ | |
819 | int blk_queue_init_tags(request_queue_t *q, int depth, | |
820 | struct blk_queue_tag *tags) | |
821 | { | |
822 | int rc; | |
823 | ||
824 | BUG_ON(tags && q->queue_tags && tags != q->queue_tags); | |
825 | ||
826 | if (!tags && !q->queue_tags) { | |
827 | tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC); | |
828 | if (!tags) | |
829 | goto fail; | |
830 | ||
831 | if (init_tag_map(q, tags, depth)) | |
832 | goto fail; | |
833 | ||
834 | INIT_LIST_HEAD(&tags->busy_list); | |
835 | tags->busy = 0; | |
836 | atomic_set(&tags->refcnt, 1); | |
837 | } else if (q->queue_tags) { | |
838 | if ((rc = blk_queue_resize_tags(q, depth))) | |
839 | return rc; | |
840 | set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
841 | return 0; | |
842 | } else | |
843 | atomic_inc(&tags->refcnt); | |
844 | ||
845 | /* | |
846 | * assign it, all done | |
847 | */ | |
848 | q->queue_tags = tags; | |
849 | q->queue_flags |= (1 << QUEUE_FLAG_QUEUED); | |
850 | return 0; | |
851 | fail: | |
852 | kfree(tags); | |
853 | return -ENOMEM; | |
854 | } | |
855 | ||
856 | EXPORT_SYMBOL(blk_queue_init_tags); | |
857 | ||
858 | /** | |
859 | * blk_queue_resize_tags - change the queueing depth | |
860 | * @q: the request queue for the device | |
861 | * @new_depth: the new max command queueing depth | |
862 | * | |
863 | * Notes: | |
864 | * Must be called with the queue lock held. | |
865 | **/ | |
866 | int blk_queue_resize_tags(request_queue_t *q, int new_depth) | |
867 | { | |
868 | struct blk_queue_tag *bqt = q->queue_tags; | |
869 | struct request **tag_index; | |
870 | unsigned long *tag_map; | |
fa72b903 | 871 | int max_depth, nr_ulongs; |
1da177e4 LT |
872 | |
873 | if (!bqt) | |
874 | return -ENXIO; | |
875 | ||
ba025082 TH |
876 | /* |
877 | * if we already have large enough real_max_depth. just | |
878 | * adjust max_depth. *NOTE* as requests with tag value | |
879 | * between new_depth and real_max_depth can be in-flight, tag | |
880 | * map can not be shrunk blindly here. | |
881 | */ | |
882 | if (new_depth <= bqt->real_max_depth) { | |
883 | bqt->max_depth = new_depth; | |
884 | return 0; | |
885 | } | |
886 | ||
1da177e4 LT |
887 | /* |
888 | * save the old state info, so we can copy it back | |
889 | */ | |
890 | tag_index = bqt->tag_index; | |
891 | tag_map = bqt->tag_map; | |
ba025082 | 892 | max_depth = bqt->real_max_depth; |
1da177e4 LT |
893 | |
894 | if (init_tag_map(q, bqt, new_depth)) | |
895 | return -ENOMEM; | |
896 | ||
897 | memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); | |
f7d37d02 | 898 | nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG; |
fa72b903 | 899 | memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long)); |
1da177e4 LT |
900 | |
901 | kfree(tag_index); | |
902 | kfree(tag_map); | |
903 | return 0; | |
904 | } | |
905 | ||
906 | EXPORT_SYMBOL(blk_queue_resize_tags); | |
907 | ||
908 | /** | |
909 | * blk_queue_end_tag - end tag operations for a request | |
910 | * @q: the request queue for the device | |
911 | * @rq: the request that has completed | |
912 | * | |
913 | * Description: | |
914 | * Typically called when end_that_request_first() returns 0, meaning | |
915 | * all transfers have been done for a request. It's important to call | |
916 | * this function before end_that_request_last(), as that will put the | |
917 | * request back on the free list thus corrupting the internal tag list. | |
918 | * | |
919 | * Notes: | |
920 | * queue lock must be held. | |
921 | **/ | |
922 | void blk_queue_end_tag(request_queue_t *q, struct request *rq) | |
923 | { | |
924 | struct blk_queue_tag *bqt = q->queue_tags; | |
925 | int tag = rq->tag; | |
926 | ||
927 | BUG_ON(tag == -1); | |
928 | ||
ba025082 | 929 | if (unlikely(tag >= bqt->real_max_depth)) |
040c928c TH |
930 | /* |
931 | * This can happen after tag depth has been reduced. | |
932 | * FIXME: how about a warning or info message here? | |
933 | */ | |
1da177e4 LT |
934 | return; |
935 | ||
936 | if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) { | |
040c928c TH |
937 | printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n", |
938 | __FUNCTION__, tag); | |
1da177e4 LT |
939 | return; |
940 | } | |
941 | ||
942 | list_del_init(&rq->queuelist); | |
943 | rq->flags &= ~REQ_QUEUED; | |
944 | rq->tag = -1; | |
945 | ||
946 | if (unlikely(bqt->tag_index[tag] == NULL)) | |
040c928c TH |
947 | printk(KERN_ERR "%s: tag %d is missing\n", |
948 | __FUNCTION__, tag); | |
1da177e4 LT |
949 | |
950 | bqt->tag_index[tag] = NULL; | |
951 | bqt->busy--; | |
952 | } | |
953 | ||
954 | EXPORT_SYMBOL(blk_queue_end_tag); | |
955 | ||
956 | /** | |
957 | * blk_queue_start_tag - find a free tag and assign it | |
958 | * @q: the request queue for the device | |
959 | * @rq: the block request that needs tagging | |
960 | * | |
961 | * Description: | |
962 | * This can either be used as a stand-alone helper, or possibly be | |
963 | * assigned as the queue &prep_rq_fn (in which case &struct request | |
964 | * automagically gets a tag assigned). Note that this function | |
965 | * assumes that any type of request can be queued! if this is not | |
966 | * true for your device, you must check the request type before | |
967 | * calling this function. The request will also be removed from | |
968 | * the request queue, so it's the drivers responsibility to readd | |
969 | * it if it should need to be restarted for some reason. | |
970 | * | |
971 | * Notes: | |
972 | * queue lock must be held. | |
973 | **/ | |
974 | int blk_queue_start_tag(request_queue_t *q, struct request *rq) | |
975 | { | |
976 | struct blk_queue_tag *bqt = q->queue_tags; | |
2bf0fdad | 977 | int tag; |
1da177e4 LT |
978 | |
979 | if (unlikely((rq->flags & REQ_QUEUED))) { | |
980 | printk(KERN_ERR | |
040c928c TH |
981 | "%s: request %p for device [%s] already tagged %d", |
982 | __FUNCTION__, rq, | |
983 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); | |
1da177e4 LT |
984 | BUG(); |
985 | } | |
986 | ||
2bf0fdad TH |
987 | tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth); |
988 | if (tag >= bqt->max_depth) | |
989 | return 1; | |
1da177e4 | 990 | |
1da177e4 LT |
991 | __set_bit(tag, bqt->tag_map); |
992 | ||
993 | rq->flags |= REQ_QUEUED; | |
994 | rq->tag = tag; | |
995 | bqt->tag_index[tag] = rq; | |
996 | blkdev_dequeue_request(rq); | |
997 | list_add(&rq->queuelist, &bqt->busy_list); | |
998 | bqt->busy++; | |
999 | return 0; | |
1000 | } | |
1001 | ||
1002 | EXPORT_SYMBOL(blk_queue_start_tag); | |
1003 | ||
1004 | /** | |
1005 | * blk_queue_invalidate_tags - invalidate all pending tags | |
1006 | * @q: the request queue for the device | |
1007 | * | |
1008 | * Description: | |
1009 | * Hardware conditions may dictate a need to stop all pending requests. | |
1010 | * In this case, we will safely clear the block side of the tag queue and | |
1011 | * readd all requests to the request queue in the right order. | |
1012 | * | |
1013 | * Notes: | |
1014 | * queue lock must be held. | |
1015 | **/ | |
1016 | void blk_queue_invalidate_tags(request_queue_t *q) | |
1017 | { | |
1018 | struct blk_queue_tag *bqt = q->queue_tags; | |
1019 | struct list_head *tmp, *n; | |
1020 | struct request *rq; | |
1021 | ||
1022 | list_for_each_safe(tmp, n, &bqt->busy_list) { | |
1023 | rq = list_entry_rq(tmp); | |
1024 | ||
1025 | if (rq->tag == -1) { | |
040c928c TH |
1026 | printk(KERN_ERR |
1027 | "%s: bad tag found on list\n", __FUNCTION__); | |
1da177e4 LT |
1028 | list_del_init(&rq->queuelist); |
1029 | rq->flags &= ~REQ_QUEUED; | |
1030 | } else | |
1031 | blk_queue_end_tag(q, rq); | |
1032 | ||
1033 | rq->flags &= ~REQ_STARTED; | |
1034 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1035 | } | |
1036 | } | |
1037 | ||
1038 | EXPORT_SYMBOL(blk_queue_invalidate_tags); | |
1039 | ||
1040 | static char *rq_flags[] = { | |
1041 | "REQ_RW", | |
1042 | "REQ_FAILFAST", | |
1043 | "REQ_SOFTBARRIER", | |
1044 | "REQ_HARDBARRIER", | |
1045 | "REQ_CMD", | |
1046 | "REQ_NOMERGE", | |
1047 | "REQ_STARTED", | |
1048 | "REQ_DONTPREP", | |
1049 | "REQ_QUEUED", | |
1050 | "REQ_PC", | |
1051 | "REQ_BLOCK_PC", | |
1052 | "REQ_SENSE", | |
1053 | "REQ_FAILED", | |
1054 | "REQ_QUIET", | |
1055 | "REQ_SPECIAL", | |
1056 | "REQ_DRIVE_CMD", | |
1057 | "REQ_DRIVE_TASK", | |
1058 | "REQ_DRIVE_TASKFILE", | |
1059 | "REQ_PREEMPT", | |
1060 | "REQ_PM_SUSPEND", | |
1061 | "REQ_PM_RESUME", | |
1062 | "REQ_PM_SHUTDOWN", | |
1063 | }; | |
1064 | ||
1065 | void blk_dump_rq_flags(struct request *rq, char *msg) | |
1066 | { | |
1067 | int bit; | |
1068 | ||
1069 | printk("%s: dev %s: flags = ", msg, | |
1070 | rq->rq_disk ? rq->rq_disk->disk_name : "?"); | |
1071 | bit = 0; | |
1072 | do { | |
1073 | if (rq->flags & (1 << bit)) | |
1074 | printk("%s ", rq_flags[bit]); | |
1075 | bit++; | |
1076 | } while (bit < __REQ_NR_BITS); | |
1077 | ||
1078 | printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector, | |
1079 | rq->nr_sectors, | |
1080 | rq->current_nr_sectors); | |
1081 | printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len); | |
1082 | ||
1083 | if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) { | |
1084 | printk("cdb: "); | |
1085 | for (bit = 0; bit < sizeof(rq->cmd); bit++) | |
1086 | printk("%02x ", rq->cmd[bit]); | |
1087 | printk("\n"); | |
1088 | } | |
1089 | } | |
1090 | ||
1091 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
1092 | ||
1093 | void blk_recount_segments(request_queue_t *q, struct bio *bio) | |
1094 | { | |
1095 | struct bio_vec *bv, *bvprv = NULL; | |
1096 | int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster; | |
1097 | int high, highprv = 1; | |
1098 | ||
1099 | if (unlikely(!bio->bi_io_vec)) | |
1100 | return; | |
1101 | ||
1102 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1103 | hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0; | |
1104 | bio_for_each_segment(bv, bio, i) { | |
1105 | /* | |
1106 | * the trick here is making sure that a high page is never | |
1107 | * considered part of another segment, since that might | |
1108 | * change with the bounce page. | |
1109 | */ | |
1110 | high = page_to_pfn(bv->bv_page) >= q->bounce_pfn; | |
1111 | if (high || highprv) | |
1112 | goto new_hw_segment; | |
1113 | if (cluster) { | |
1114 | if (seg_size + bv->bv_len > q->max_segment_size) | |
1115 | goto new_segment; | |
1116 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv)) | |
1117 | goto new_segment; | |
1118 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv)) | |
1119 | goto new_segment; | |
1120 | if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) | |
1121 | goto new_hw_segment; | |
1122 | ||
1123 | seg_size += bv->bv_len; | |
1124 | hw_seg_size += bv->bv_len; | |
1125 | bvprv = bv; | |
1126 | continue; | |
1127 | } | |
1128 | new_segment: | |
1129 | if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) && | |
1130 | !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) { | |
1131 | hw_seg_size += bv->bv_len; | |
1132 | } else { | |
1133 | new_hw_segment: | |
1134 | if (hw_seg_size > bio->bi_hw_front_size) | |
1135 | bio->bi_hw_front_size = hw_seg_size; | |
1136 | hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len; | |
1137 | nr_hw_segs++; | |
1138 | } | |
1139 | ||
1140 | nr_phys_segs++; | |
1141 | bvprv = bv; | |
1142 | seg_size = bv->bv_len; | |
1143 | highprv = high; | |
1144 | } | |
1145 | if (hw_seg_size > bio->bi_hw_back_size) | |
1146 | bio->bi_hw_back_size = hw_seg_size; | |
1147 | if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size) | |
1148 | bio->bi_hw_front_size = hw_seg_size; | |
1149 | bio->bi_phys_segments = nr_phys_segs; | |
1150 | bio->bi_hw_segments = nr_hw_segs; | |
1151 | bio->bi_flags |= (1 << BIO_SEG_VALID); | |
1152 | } | |
1153 | ||
1154 | ||
93d17d3d | 1155 | static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio, |
1da177e4 LT |
1156 | struct bio *nxt) |
1157 | { | |
1158 | if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) | |
1159 | return 0; | |
1160 | ||
1161 | if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt))) | |
1162 | return 0; | |
1163 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) | |
1164 | return 0; | |
1165 | ||
1166 | /* | |
1167 | * bio and nxt are contigous in memory, check if the queue allows | |
1168 | * these two to be merged into one | |
1169 | */ | |
1170 | if (BIO_SEG_BOUNDARY(q, bio, nxt)) | |
1171 | return 1; | |
1172 | ||
1173 | return 0; | |
1174 | } | |
1175 | ||
93d17d3d | 1176 | static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio, |
1da177e4 LT |
1177 | struct bio *nxt) |
1178 | { | |
1179 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1180 | blk_recount_segments(q, bio); | |
1181 | if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID))) | |
1182 | blk_recount_segments(q, nxt); | |
1183 | if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) || | |
1184 | BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size)) | |
1185 | return 0; | |
1186 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) | |
1187 | return 0; | |
1188 | ||
1189 | return 1; | |
1190 | } | |
1191 | ||
1da177e4 LT |
1192 | /* |
1193 | * map a request to scatterlist, return number of sg entries setup. Caller | |
1194 | * must make sure sg can hold rq->nr_phys_segments entries | |
1195 | */ | |
1196 | int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg) | |
1197 | { | |
1198 | struct bio_vec *bvec, *bvprv; | |
1199 | struct bio *bio; | |
1200 | int nsegs, i, cluster; | |
1201 | ||
1202 | nsegs = 0; | |
1203 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1204 | ||
1205 | /* | |
1206 | * for each bio in rq | |
1207 | */ | |
1208 | bvprv = NULL; | |
1209 | rq_for_each_bio(bio, rq) { | |
1210 | /* | |
1211 | * for each segment in bio | |
1212 | */ | |
1213 | bio_for_each_segment(bvec, bio, i) { | |
1214 | int nbytes = bvec->bv_len; | |
1215 | ||
1216 | if (bvprv && cluster) { | |
1217 | if (sg[nsegs - 1].length + nbytes > q->max_segment_size) | |
1218 | goto new_segment; | |
1219 | ||
1220 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) | |
1221 | goto new_segment; | |
1222 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) | |
1223 | goto new_segment; | |
1224 | ||
1225 | sg[nsegs - 1].length += nbytes; | |
1226 | } else { | |
1227 | new_segment: | |
1228 | memset(&sg[nsegs],0,sizeof(struct scatterlist)); | |
1229 | sg[nsegs].page = bvec->bv_page; | |
1230 | sg[nsegs].length = nbytes; | |
1231 | sg[nsegs].offset = bvec->bv_offset; | |
1232 | ||
1233 | nsegs++; | |
1234 | } | |
1235 | bvprv = bvec; | |
1236 | } /* segments in bio */ | |
1237 | } /* bios in rq */ | |
1238 | ||
1239 | return nsegs; | |
1240 | } | |
1241 | ||
1242 | EXPORT_SYMBOL(blk_rq_map_sg); | |
1243 | ||
1244 | /* | |
1245 | * the standard queue merge functions, can be overridden with device | |
1246 | * specific ones if so desired | |
1247 | */ | |
1248 | ||
1249 | static inline int ll_new_mergeable(request_queue_t *q, | |
1250 | struct request *req, | |
1251 | struct bio *bio) | |
1252 | { | |
1253 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1254 | ||
1255 | if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
1256 | req->flags |= REQ_NOMERGE; | |
1257 | if (req == q->last_merge) | |
1258 | q->last_merge = NULL; | |
1259 | return 0; | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * A hw segment is just getting larger, bump just the phys | |
1264 | * counter. | |
1265 | */ | |
1266 | req->nr_phys_segments += nr_phys_segs; | |
1267 | return 1; | |
1268 | } | |
1269 | ||
1270 | static inline int ll_new_hw_segment(request_queue_t *q, | |
1271 | struct request *req, | |
1272 | struct bio *bio) | |
1273 | { | |
1274 | int nr_hw_segs = bio_hw_segments(q, bio); | |
1275 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1276 | ||
1277 | if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments | |
1278 | || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
1279 | req->flags |= REQ_NOMERGE; | |
1280 | if (req == q->last_merge) | |
1281 | q->last_merge = NULL; | |
1282 | return 0; | |
1283 | } | |
1284 | ||
1285 | /* | |
1286 | * This will form the start of a new hw segment. Bump both | |
1287 | * counters. | |
1288 | */ | |
1289 | req->nr_hw_segments += nr_hw_segs; | |
1290 | req->nr_phys_segments += nr_phys_segs; | |
1291 | return 1; | |
1292 | } | |
1293 | ||
1294 | static int ll_back_merge_fn(request_queue_t *q, struct request *req, | |
1295 | struct bio *bio) | |
1296 | { | |
1297 | int len; | |
1298 | ||
1299 | if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) { | |
1300 | req->flags |= REQ_NOMERGE; | |
1301 | if (req == q->last_merge) | |
1302 | q->last_merge = NULL; | |
1303 | return 0; | |
1304 | } | |
1305 | if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID))) | |
1306 | blk_recount_segments(q, req->biotail); | |
1307 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1308 | blk_recount_segments(q, bio); | |
1309 | len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size; | |
1310 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) && | |
1311 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1312 | int mergeable = ll_new_mergeable(q, req, bio); | |
1313 | ||
1314 | if (mergeable) { | |
1315 | if (req->nr_hw_segments == 1) | |
1316 | req->bio->bi_hw_front_size = len; | |
1317 | if (bio->bi_hw_segments == 1) | |
1318 | bio->bi_hw_back_size = len; | |
1319 | } | |
1320 | return mergeable; | |
1321 | } | |
1322 | ||
1323 | return ll_new_hw_segment(q, req, bio); | |
1324 | } | |
1325 | ||
1326 | static int ll_front_merge_fn(request_queue_t *q, struct request *req, | |
1327 | struct bio *bio) | |
1328 | { | |
1329 | int len; | |
1330 | ||
1331 | if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) { | |
1332 | req->flags |= REQ_NOMERGE; | |
1333 | if (req == q->last_merge) | |
1334 | q->last_merge = NULL; | |
1335 | return 0; | |
1336 | } | |
1337 | len = bio->bi_hw_back_size + req->bio->bi_hw_front_size; | |
1338 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1339 | blk_recount_segments(q, bio); | |
1340 | if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID))) | |
1341 | blk_recount_segments(q, req->bio); | |
1342 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) && | |
1343 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1344 | int mergeable = ll_new_mergeable(q, req, bio); | |
1345 | ||
1346 | if (mergeable) { | |
1347 | if (bio->bi_hw_segments == 1) | |
1348 | bio->bi_hw_front_size = len; | |
1349 | if (req->nr_hw_segments == 1) | |
1350 | req->biotail->bi_hw_back_size = len; | |
1351 | } | |
1352 | return mergeable; | |
1353 | } | |
1354 | ||
1355 | return ll_new_hw_segment(q, req, bio); | |
1356 | } | |
1357 | ||
1358 | static int ll_merge_requests_fn(request_queue_t *q, struct request *req, | |
1359 | struct request *next) | |
1360 | { | |
dfa1a553 ND |
1361 | int total_phys_segments; |
1362 | int total_hw_segments; | |
1da177e4 LT |
1363 | |
1364 | /* | |
1365 | * First check if the either of the requests are re-queued | |
1366 | * requests. Can't merge them if they are. | |
1367 | */ | |
1368 | if (req->special || next->special) | |
1369 | return 0; | |
1370 | ||
1371 | /* | |
dfa1a553 | 1372 | * Will it become too large? |
1da177e4 LT |
1373 | */ |
1374 | if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) | |
1375 | return 0; | |
1376 | ||
1377 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
1378 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) | |
1379 | total_phys_segments--; | |
1380 | ||
1381 | if (total_phys_segments > q->max_phys_segments) | |
1382 | return 0; | |
1383 | ||
1384 | total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; | |
1385 | if (blk_hw_contig_segment(q, req->biotail, next->bio)) { | |
1386 | int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size; | |
1387 | /* | |
1388 | * propagate the combined length to the end of the requests | |
1389 | */ | |
1390 | if (req->nr_hw_segments == 1) | |
1391 | req->bio->bi_hw_front_size = len; | |
1392 | if (next->nr_hw_segments == 1) | |
1393 | next->biotail->bi_hw_back_size = len; | |
1394 | total_hw_segments--; | |
1395 | } | |
1396 | ||
1397 | if (total_hw_segments > q->max_hw_segments) | |
1398 | return 0; | |
1399 | ||
1400 | /* Merge is OK... */ | |
1401 | req->nr_phys_segments = total_phys_segments; | |
1402 | req->nr_hw_segments = total_hw_segments; | |
1403 | return 1; | |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * "plug" the device if there are no outstanding requests: this will | |
1408 | * force the transfer to start only after we have put all the requests | |
1409 | * on the list. | |
1410 | * | |
1411 | * This is called with interrupts off and no requests on the queue and | |
1412 | * with the queue lock held. | |
1413 | */ | |
1414 | void blk_plug_device(request_queue_t *q) | |
1415 | { | |
1416 | WARN_ON(!irqs_disabled()); | |
1417 | ||
1418 | /* | |
1419 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
1420 | * which will restart the queueing | |
1421 | */ | |
1422 | if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)) | |
1423 | return; | |
1424 | ||
1425 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | |
1426 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); | |
1427 | } | |
1428 | ||
1429 | EXPORT_SYMBOL(blk_plug_device); | |
1430 | ||
1431 | /* | |
1432 | * remove the queue from the plugged list, if present. called with | |
1433 | * queue lock held and interrupts disabled. | |
1434 | */ | |
1435 | int blk_remove_plug(request_queue_t *q) | |
1436 | { | |
1437 | WARN_ON(!irqs_disabled()); | |
1438 | ||
1439 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | |
1440 | return 0; | |
1441 | ||
1442 | del_timer(&q->unplug_timer); | |
1443 | return 1; | |
1444 | } | |
1445 | ||
1446 | EXPORT_SYMBOL(blk_remove_plug); | |
1447 | ||
1448 | /* | |
1449 | * remove the plug and let it rip.. | |
1450 | */ | |
1451 | void __generic_unplug_device(request_queue_t *q) | |
1452 | { | |
fde6ad22 | 1453 | if (unlikely(test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))) |
1da177e4 LT |
1454 | return; |
1455 | ||
1456 | if (!blk_remove_plug(q)) | |
1457 | return; | |
1458 | ||
22e2c507 | 1459 | q->request_fn(q); |
1da177e4 LT |
1460 | } |
1461 | EXPORT_SYMBOL(__generic_unplug_device); | |
1462 | ||
1463 | /** | |
1464 | * generic_unplug_device - fire a request queue | |
1465 | * @q: The &request_queue_t in question | |
1466 | * | |
1467 | * Description: | |
1468 | * Linux uses plugging to build bigger requests queues before letting | |
1469 | * the device have at them. If a queue is plugged, the I/O scheduler | |
1470 | * is still adding and merging requests on the queue. Once the queue | |
1471 | * gets unplugged, the request_fn defined for the queue is invoked and | |
1472 | * transfers started. | |
1473 | **/ | |
1474 | void generic_unplug_device(request_queue_t *q) | |
1475 | { | |
1476 | spin_lock_irq(q->queue_lock); | |
1477 | __generic_unplug_device(q); | |
1478 | spin_unlock_irq(q->queue_lock); | |
1479 | } | |
1480 | EXPORT_SYMBOL(generic_unplug_device); | |
1481 | ||
1482 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
1483 | struct page *page) | |
1484 | { | |
1485 | request_queue_t *q = bdi->unplug_io_data; | |
1486 | ||
1487 | /* | |
1488 | * devices don't necessarily have an ->unplug_fn defined | |
1489 | */ | |
1490 | if (q->unplug_fn) | |
1491 | q->unplug_fn(q); | |
1492 | } | |
1493 | ||
1494 | static void blk_unplug_work(void *data) | |
1495 | { | |
1496 | request_queue_t *q = data; | |
1497 | ||
1498 | q->unplug_fn(q); | |
1499 | } | |
1500 | ||
1501 | static void blk_unplug_timeout(unsigned long data) | |
1502 | { | |
1503 | request_queue_t *q = (request_queue_t *)data; | |
1504 | ||
1505 | kblockd_schedule_work(&q->unplug_work); | |
1506 | } | |
1507 | ||
1508 | /** | |
1509 | * blk_start_queue - restart a previously stopped queue | |
1510 | * @q: The &request_queue_t in question | |
1511 | * | |
1512 | * Description: | |
1513 | * blk_start_queue() will clear the stop flag on the queue, and call | |
1514 | * the request_fn for the queue if it was in a stopped state when | |
1515 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
1516 | **/ | |
1517 | void blk_start_queue(request_queue_t *q) | |
1518 | { | |
1519 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | |
1520 | ||
1521 | /* | |
1522 | * one level of recursion is ok and is much faster than kicking | |
1523 | * the unplug handling | |
1524 | */ | |
1525 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1526 | q->request_fn(q); | |
1527 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1528 | } else { | |
1529 | blk_plug_device(q); | |
1530 | kblockd_schedule_work(&q->unplug_work); | |
1531 | } | |
1532 | } | |
1533 | ||
1534 | EXPORT_SYMBOL(blk_start_queue); | |
1535 | ||
1536 | /** | |
1537 | * blk_stop_queue - stop a queue | |
1538 | * @q: The &request_queue_t in question | |
1539 | * | |
1540 | * Description: | |
1541 | * The Linux block layer assumes that a block driver will consume all | |
1542 | * entries on the request queue when the request_fn strategy is called. | |
1543 | * Often this will not happen, because of hardware limitations (queue | |
1544 | * depth settings). If a device driver gets a 'queue full' response, | |
1545 | * or if it simply chooses not to queue more I/O at one point, it can | |
1546 | * call this function to prevent the request_fn from being called until | |
1547 | * the driver has signalled it's ready to go again. This happens by calling | |
1548 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
1549 | **/ | |
1550 | void blk_stop_queue(request_queue_t *q) | |
1551 | { | |
1552 | blk_remove_plug(q); | |
1553 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | |
1554 | } | |
1555 | EXPORT_SYMBOL(blk_stop_queue); | |
1556 | ||
1557 | /** | |
1558 | * blk_sync_queue - cancel any pending callbacks on a queue | |
1559 | * @q: the queue | |
1560 | * | |
1561 | * Description: | |
1562 | * The block layer may perform asynchronous callback activity | |
1563 | * on a queue, such as calling the unplug function after a timeout. | |
1564 | * A block device may call blk_sync_queue to ensure that any | |
1565 | * such activity is cancelled, thus allowing it to release resources | |
1566 | * the the callbacks might use. The caller must already have made sure | |
1567 | * that its ->make_request_fn will not re-add plugging prior to calling | |
1568 | * this function. | |
1569 | * | |
1570 | */ | |
1571 | void blk_sync_queue(struct request_queue *q) | |
1572 | { | |
1573 | del_timer_sync(&q->unplug_timer); | |
1574 | kblockd_flush(); | |
1575 | } | |
1576 | EXPORT_SYMBOL(blk_sync_queue); | |
1577 | ||
1578 | /** | |
1579 | * blk_run_queue - run a single device queue | |
1580 | * @q: The queue to run | |
1581 | */ | |
1582 | void blk_run_queue(struct request_queue *q) | |
1583 | { | |
1584 | unsigned long flags; | |
1585 | ||
1586 | spin_lock_irqsave(q->queue_lock, flags); | |
1587 | blk_remove_plug(q); | |
a2997382 KC |
1588 | if (!elv_queue_empty(q)) |
1589 | q->request_fn(q); | |
1da177e4 LT |
1590 | spin_unlock_irqrestore(q->queue_lock, flags); |
1591 | } | |
1592 | EXPORT_SYMBOL(blk_run_queue); | |
1593 | ||
1594 | /** | |
1595 | * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed | |
1596 | * @q: the request queue to be released | |
1597 | * | |
1598 | * Description: | |
1599 | * blk_cleanup_queue is the pair to blk_init_queue() or | |
1600 | * blk_queue_make_request(). It should be called when a request queue is | |
1601 | * being released; typically when a block device is being de-registered. | |
1602 | * Currently, its primary task it to free all the &struct request | |
1603 | * structures that were allocated to the queue and the queue itself. | |
1604 | * | |
1605 | * Caveat: | |
1606 | * Hopefully the low level driver will have finished any | |
1607 | * outstanding requests first... | |
1608 | **/ | |
1609 | void blk_cleanup_queue(request_queue_t * q) | |
1610 | { | |
1611 | struct request_list *rl = &q->rq; | |
1612 | ||
1613 | if (!atomic_dec_and_test(&q->refcnt)) | |
1614 | return; | |
1615 | ||
1616 | if (q->elevator) | |
1617 | elevator_exit(q->elevator); | |
1618 | ||
1619 | blk_sync_queue(q); | |
1620 | ||
1621 | if (rl->rq_pool) | |
1622 | mempool_destroy(rl->rq_pool); | |
1623 | ||
1624 | if (q->queue_tags) | |
1625 | __blk_queue_free_tags(q); | |
1626 | ||
1627 | blk_queue_ordered(q, QUEUE_ORDERED_NONE); | |
1628 | ||
1629 | kmem_cache_free(requestq_cachep, q); | |
1630 | } | |
1631 | ||
1632 | EXPORT_SYMBOL(blk_cleanup_queue); | |
1633 | ||
1634 | static int blk_init_free_list(request_queue_t *q) | |
1635 | { | |
1636 | struct request_list *rl = &q->rq; | |
1637 | ||
1638 | rl->count[READ] = rl->count[WRITE] = 0; | |
1639 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
1640 | init_waitqueue_head(&rl->wait[READ]); | |
1641 | init_waitqueue_head(&rl->wait[WRITE]); | |
1642 | init_waitqueue_head(&rl->drain); | |
1643 | ||
1946089a CL |
1644 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
1645 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
1646 | |
1647 | if (!rl->rq_pool) | |
1648 | return -ENOMEM; | |
1649 | ||
1650 | return 0; | |
1651 | } | |
1652 | ||
1653 | static int __make_request(request_queue_t *, struct bio *); | |
1654 | ||
1655 | request_queue_t *blk_alloc_queue(int gfp_mask) | |
1656 | { | |
1946089a CL |
1657 | return blk_alloc_queue_node(gfp_mask, -1); |
1658 | } | |
1659 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 1660 | |
1946089a CL |
1661 | request_queue_t *blk_alloc_queue_node(int gfp_mask, int node_id) |
1662 | { | |
1663 | request_queue_t *q; | |
1664 | ||
1665 | q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id); | |
1da177e4 LT |
1666 | if (!q) |
1667 | return NULL; | |
1668 | ||
1669 | memset(q, 0, sizeof(*q)); | |
1670 | init_timer(&q->unplug_timer); | |
1671 | atomic_set(&q->refcnt, 1); | |
1672 | ||
1673 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; | |
1674 | q->backing_dev_info.unplug_io_data = q; | |
1675 | ||
1676 | return q; | |
1677 | } | |
1946089a | 1678 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
1679 | |
1680 | /** | |
1681 | * blk_init_queue - prepare a request queue for use with a block device | |
1682 | * @rfn: The function to be called to process requests that have been | |
1683 | * placed on the queue. | |
1684 | * @lock: Request queue spin lock | |
1685 | * | |
1686 | * Description: | |
1687 | * If a block device wishes to use the standard request handling procedures, | |
1688 | * which sorts requests and coalesces adjacent requests, then it must | |
1689 | * call blk_init_queue(). The function @rfn will be called when there | |
1690 | * are requests on the queue that need to be processed. If the device | |
1691 | * supports plugging, then @rfn may not be called immediately when requests | |
1692 | * are available on the queue, but may be called at some time later instead. | |
1693 | * Plugged queues are generally unplugged when a buffer belonging to one | |
1694 | * of the requests on the queue is needed, or due to memory pressure. | |
1695 | * | |
1696 | * @rfn is not required, or even expected, to remove all requests off the | |
1697 | * queue, but only as many as it can handle at a time. If it does leave | |
1698 | * requests on the queue, it is responsible for arranging that the requests | |
1699 | * get dealt with eventually. | |
1700 | * | |
1701 | * The queue spin lock must be held while manipulating the requests on the | |
1702 | * request queue. | |
1703 | * | |
1704 | * Function returns a pointer to the initialized request queue, or NULL if | |
1705 | * it didn't succeed. | |
1706 | * | |
1707 | * Note: | |
1708 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
1709 | * when the block device is deactivated (such as at module unload). | |
1710 | **/ | |
1946089a | 1711 | |
1da177e4 LT |
1712 | request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1713 | { | |
1946089a CL |
1714 | return blk_init_queue_node(rfn, lock, -1); |
1715 | } | |
1716 | EXPORT_SYMBOL(blk_init_queue); | |
1717 | ||
1718 | request_queue_t * | |
1719 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) | |
1720 | { | |
1721 | request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id); | |
1da177e4 LT |
1722 | |
1723 | if (!q) | |
1724 | return NULL; | |
1725 | ||
1946089a | 1726 | q->node = node_id; |
1da177e4 LT |
1727 | if (blk_init_free_list(q)) |
1728 | goto out_init; | |
1729 | ||
152587de JA |
1730 | /* |
1731 | * if caller didn't supply a lock, they get per-queue locking with | |
1732 | * our embedded lock | |
1733 | */ | |
1734 | if (!lock) { | |
1735 | spin_lock_init(&q->__queue_lock); | |
1736 | lock = &q->__queue_lock; | |
1737 | } | |
1738 | ||
1da177e4 LT |
1739 | q->request_fn = rfn; |
1740 | q->back_merge_fn = ll_back_merge_fn; | |
1741 | q->front_merge_fn = ll_front_merge_fn; | |
1742 | q->merge_requests_fn = ll_merge_requests_fn; | |
1743 | q->prep_rq_fn = NULL; | |
1744 | q->unplug_fn = generic_unplug_device; | |
1745 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); | |
1746 | q->queue_lock = lock; | |
1747 | ||
1748 | blk_queue_segment_boundary(q, 0xffffffff); | |
1749 | ||
1750 | blk_queue_make_request(q, __make_request); | |
1751 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
1752 | ||
1753 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
1754 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
1755 | ||
1756 | /* | |
1757 | * all done | |
1758 | */ | |
1759 | if (!elevator_init(q, NULL)) { | |
1760 | blk_queue_congestion_threshold(q); | |
1761 | return q; | |
1762 | } | |
1763 | ||
1764 | blk_cleanup_queue(q); | |
1765 | out_init: | |
1766 | kmem_cache_free(requestq_cachep, q); | |
1767 | return NULL; | |
1768 | } | |
1946089a | 1769 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 LT |
1770 | |
1771 | int blk_get_queue(request_queue_t *q) | |
1772 | { | |
fde6ad22 | 1773 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
1da177e4 LT |
1774 | atomic_inc(&q->refcnt); |
1775 | return 0; | |
1776 | } | |
1777 | ||
1778 | return 1; | |
1779 | } | |
1780 | ||
1781 | EXPORT_SYMBOL(blk_get_queue); | |
1782 | ||
1783 | static inline void blk_free_request(request_queue_t *q, struct request *rq) | |
1784 | { | |
1785 | elv_put_request(q, rq); | |
1786 | mempool_free(rq, q->rq.rq_pool); | |
1787 | } | |
1788 | ||
22e2c507 JA |
1789 | static inline struct request * |
1790 | blk_alloc_request(request_queue_t *q, int rw, struct bio *bio, int gfp_mask) | |
1da177e4 LT |
1791 | { |
1792 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
1793 | ||
1794 | if (!rq) | |
1795 | return NULL; | |
1796 | ||
1797 | /* | |
1798 | * first three bits are identical in rq->flags and bio->bi_rw, | |
1799 | * see bio.h and blkdev.h | |
1800 | */ | |
1801 | rq->flags = rw; | |
1802 | ||
22e2c507 | 1803 | if (!elv_set_request(q, rq, bio, gfp_mask)) |
1da177e4 LT |
1804 | return rq; |
1805 | ||
1806 | mempool_free(rq, q->rq.rq_pool); | |
1807 | return NULL; | |
1808 | } | |
1809 | ||
1810 | /* | |
1811 | * ioc_batching returns true if the ioc is a valid batching request and | |
1812 | * should be given priority access to a request. | |
1813 | */ | |
1814 | static inline int ioc_batching(request_queue_t *q, struct io_context *ioc) | |
1815 | { | |
1816 | if (!ioc) | |
1817 | return 0; | |
1818 | ||
1819 | /* | |
1820 | * Make sure the process is able to allocate at least 1 request | |
1821 | * even if the batch times out, otherwise we could theoretically | |
1822 | * lose wakeups. | |
1823 | */ | |
1824 | return ioc->nr_batch_requests == q->nr_batching || | |
1825 | (ioc->nr_batch_requests > 0 | |
1826 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
1827 | } | |
1828 | ||
1829 | /* | |
1830 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
1831 | * will cause the process to be a "batcher" on all queues in the system. This | |
1832 | * is the behaviour we want though - once it gets a wakeup it should be given | |
1833 | * a nice run. | |
1834 | */ | |
93d17d3d | 1835 | static void ioc_set_batching(request_queue_t *q, struct io_context *ioc) |
1da177e4 LT |
1836 | { |
1837 | if (!ioc || ioc_batching(q, ioc)) | |
1838 | return; | |
1839 | ||
1840 | ioc->nr_batch_requests = q->nr_batching; | |
1841 | ioc->last_waited = jiffies; | |
1842 | } | |
1843 | ||
1844 | static void __freed_request(request_queue_t *q, int rw) | |
1845 | { | |
1846 | struct request_list *rl = &q->rq; | |
1847 | ||
1848 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
1849 | clear_queue_congested(q, rw); | |
1850 | ||
1851 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
1852 | if (waitqueue_active(&rl->wait[rw])) |
1853 | wake_up(&rl->wait[rw]); | |
1854 | ||
1855 | blk_clear_queue_full(q, rw); | |
1856 | } | |
1857 | } | |
1858 | ||
1859 | /* | |
1860 | * A request has just been released. Account for it, update the full and | |
1861 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
1862 | */ | |
1863 | static void freed_request(request_queue_t *q, int rw) | |
1864 | { | |
1865 | struct request_list *rl = &q->rq; | |
1866 | ||
1867 | rl->count[rw]--; | |
1868 | ||
1869 | __freed_request(q, rw); | |
1870 | ||
1871 | if (unlikely(rl->starved[rw ^ 1])) | |
1872 | __freed_request(q, rw ^ 1); | |
1873 | ||
1874 | if (!rl->count[READ] && !rl->count[WRITE]) { | |
1875 | smp_mb(); | |
1876 | if (unlikely(waitqueue_active(&rl->drain))) | |
1877 | wake_up(&rl->drain); | |
1878 | } | |
1879 | } | |
1880 | ||
1881 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
1882 | /* | |
d6344532 NP |
1883 | * Get a free request, queue_lock must be held. |
1884 | * Returns NULL on failure, with queue_lock held. | |
1885 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 1886 | */ |
22e2c507 JA |
1887 | static struct request *get_request(request_queue_t *q, int rw, struct bio *bio, |
1888 | int gfp_mask) | |
1da177e4 LT |
1889 | { |
1890 | struct request *rq = NULL; | |
1891 | struct request_list *rl = &q->rq; | |
fb3cc432 | 1892 | struct io_context *ioc = current_io_context(GFP_ATOMIC); |
1da177e4 LT |
1893 | |
1894 | if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) | |
1895 | goto out; | |
1896 | ||
1da177e4 LT |
1897 | if (rl->count[rw]+1 >= q->nr_requests) { |
1898 | /* | |
1899 | * The queue will fill after this allocation, so set it as | |
1900 | * full, and mark this process as "batching". This process | |
1901 | * will be allowed to complete a batch of requests, others | |
1902 | * will be blocked. | |
1903 | */ | |
1904 | if (!blk_queue_full(q, rw)) { | |
1905 | ioc_set_batching(q, ioc); | |
1906 | blk_set_queue_full(q, rw); | |
1907 | } | |
1908 | } | |
1909 | ||
22e2c507 | 1910 | switch (elv_may_queue(q, rw, bio)) { |
1da177e4 LT |
1911 | case ELV_MQUEUE_NO: |
1912 | goto rq_starved; | |
1913 | case ELV_MQUEUE_MAY: | |
1914 | break; | |
1915 | case ELV_MQUEUE_MUST: | |
1916 | goto get_rq; | |
1917 | } | |
1918 | ||
1919 | if (blk_queue_full(q, rw) && !ioc_batching(q, ioc)) { | |
1920 | /* | |
1921 | * The queue is full and the allocating process is not a | |
1922 | * "batcher", and not exempted by the IO scheduler | |
1923 | */ | |
1da177e4 LT |
1924 | goto out; |
1925 | } | |
1926 | ||
1927 | get_rq: | |
082cf69e JA |
1928 | /* |
1929 | * Only allow batching queuers to allocate up to 50% over the defined | |
1930 | * limit of requests, otherwise we could have thousands of requests | |
1931 | * allocated with any setting of ->nr_requests | |
1932 | */ | |
fd782a4a | 1933 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 1934 | goto out; |
fd782a4a | 1935 | |
1da177e4 LT |
1936 | rl->count[rw]++; |
1937 | rl->starved[rw] = 0; | |
1938 | if (rl->count[rw] >= queue_congestion_on_threshold(q)) | |
1939 | set_queue_congested(q, rw); | |
1940 | spin_unlock_irq(q->queue_lock); | |
1941 | ||
22e2c507 | 1942 | rq = blk_alloc_request(q, rw, bio, gfp_mask); |
1da177e4 LT |
1943 | if (!rq) { |
1944 | /* | |
1945 | * Allocation failed presumably due to memory. Undo anything | |
1946 | * we might have messed up. | |
1947 | * | |
1948 | * Allocating task should really be put onto the front of the | |
1949 | * wait queue, but this is pretty rare. | |
1950 | */ | |
1951 | spin_lock_irq(q->queue_lock); | |
1952 | freed_request(q, rw); | |
1953 | ||
1954 | /* | |
1955 | * in the very unlikely event that allocation failed and no | |
1956 | * requests for this direction was pending, mark us starved | |
1957 | * so that freeing of a request in the other direction will | |
1958 | * notice us. another possible fix would be to split the | |
1959 | * rq mempool into READ and WRITE | |
1960 | */ | |
1961 | rq_starved: | |
1962 | if (unlikely(rl->count[rw] == 0)) | |
1963 | rl->starved[rw] = 1; | |
1964 | ||
1da177e4 LT |
1965 | goto out; |
1966 | } | |
1967 | ||
1968 | if (ioc_batching(q, ioc)) | |
1969 | ioc->nr_batch_requests--; | |
1970 | ||
1971 | rq_init(q, rq); | |
1972 | rq->rl = rl; | |
1973 | out: | |
1da177e4 LT |
1974 | return rq; |
1975 | } | |
1976 | ||
1977 | /* | |
1978 | * No available requests for this queue, unplug the device and wait for some | |
1979 | * requests to become available. | |
d6344532 NP |
1980 | * |
1981 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 1982 | */ |
22e2c507 JA |
1983 | static struct request *get_request_wait(request_queue_t *q, int rw, |
1984 | struct bio *bio) | |
1da177e4 | 1985 | { |
1da177e4 LT |
1986 | struct request *rq; |
1987 | ||
450991bc NP |
1988 | rq = get_request(q, rw, bio, GFP_NOIO); |
1989 | while (!rq) { | |
1990 | DEFINE_WAIT(wait); | |
1da177e4 LT |
1991 | struct request_list *rl = &q->rq; |
1992 | ||
1993 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
1994 | TASK_UNINTERRUPTIBLE); | |
1995 | ||
22e2c507 | 1996 | rq = get_request(q, rw, bio, GFP_NOIO); |
1da177e4 LT |
1997 | |
1998 | if (!rq) { | |
1999 | struct io_context *ioc; | |
2000 | ||
d6344532 NP |
2001 | __generic_unplug_device(q); |
2002 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
2003 | io_schedule(); |
2004 | ||
2005 | /* | |
2006 | * After sleeping, we become a "batching" process and | |
2007 | * will be able to allocate at least one request, and | |
2008 | * up to a big batch of them for a small period time. | |
2009 | * See ioc_batching, ioc_set_batching | |
2010 | */ | |
fb3cc432 | 2011 | ioc = current_io_context(GFP_NOIO); |
1da177e4 | 2012 | ioc_set_batching(q, ioc); |
d6344532 NP |
2013 | |
2014 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
2015 | } |
2016 | finish_wait(&rl->wait[rw], &wait); | |
450991bc | 2017 | } |
1da177e4 LT |
2018 | |
2019 | return rq; | |
2020 | } | |
2021 | ||
2022 | struct request *blk_get_request(request_queue_t *q, int rw, int gfp_mask) | |
2023 | { | |
2024 | struct request *rq; | |
2025 | ||
2026 | BUG_ON(rw != READ && rw != WRITE); | |
2027 | ||
d6344532 NP |
2028 | spin_lock_irq(q->queue_lock); |
2029 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 2030 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 2031 | } else { |
22e2c507 | 2032 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
2033 | if (!rq) |
2034 | spin_unlock_irq(q->queue_lock); | |
2035 | } | |
2036 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
2037 | |
2038 | return rq; | |
2039 | } | |
1da177e4 LT |
2040 | EXPORT_SYMBOL(blk_get_request); |
2041 | ||
2042 | /** | |
2043 | * blk_requeue_request - put a request back on queue | |
2044 | * @q: request queue where request should be inserted | |
2045 | * @rq: request to be inserted | |
2046 | * | |
2047 | * Description: | |
2048 | * Drivers often keep queueing requests until the hardware cannot accept | |
2049 | * more, when that condition happens we need to put the request back | |
2050 | * on the queue. Must be called with queue lock held. | |
2051 | */ | |
2052 | void blk_requeue_request(request_queue_t *q, struct request *rq) | |
2053 | { | |
2054 | if (blk_rq_tagged(rq)) | |
2055 | blk_queue_end_tag(q, rq); | |
2056 | ||
2057 | elv_requeue_request(q, rq); | |
2058 | } | |
2059 | ||
2060 | EXPORT_SYMBOL(blk_requeue_request); | |
2061 | ||
2062 | /** | |
2063 | * blk_insert_request - insert a special request in to a request queue | |
2064 | * @q: request queue where request should be inserted | |
2065 | * @rq: request to be inserted | |
2066 | * @at_head: insert request at head or tail of queue | |
2067 | * @data: private data | |
1da177e4 LT |
2068 | * |
2069 | * Description: | |
2070 | * Many block devices need to execute commands asynchronously, so they don't | |
2071 | * block the whole kernel from preemption during request execution. This is | |
2072 | * accomplished normally by inserting aritficial requests tagged as | |
2073 | * REQ_SPECIAL in to the corresponding request queue, and letting them be | |
2074 | * scheduled for actual execution by the request queue. | |
2075 | * | |
2076 | * We have the option of inserting the head or the tail of the queue. | |
2077 | * Typically we use the tail for new ioctls and so forth. We use the head | |
2078 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
2079 | * host that is unable to accept a particular command. | |
2080 | */ | |
2081 | void blk_insert_request(request_queue_t *q, struct request *rq, | |
867d1191 | 2082 | int at_head, void *data) |
1da177e4 | 2083 | { |
867d1191 | 2084 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
2085 | unsigned long flags; |
2086 | ||
2087 | /* | |
2088 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
2089 | * must not attempt merges on this) and that it acts as a soft | |
2090 | * barrier | |
2091 | */ | |
2092 | rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER; | |
2093 | ||
2094 | rq->special = data; | |
2095 | ||
2096 | spin_lock_irqsave(q->queue_lock, flags); | |
2097 | ||
2098 | /* | |
2099 | * If command is tagged, release the tag | |
2100 | */ | |
867d1191 TH |
2101 | if (blk_rq_tagged(rq)) |
2102 | blk_queue_end_tag(q, rq); | |
1da177e4 | 2103 | |
867d1191 TH |
2104 | drive_stat_acct(rq, rq->nr_sectors, 1); |
2105 | __elv_add_request(q, rq, where, 0); | |
1da177e4 | 2106 | |
1da177e4 LT |
2107 | if (blk_queue_plugged(q)) |
2108 | __generic_unplug_device(q); | |
2109 | else | |
2110 | q->request_fn(q); | |
2111 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2112 | } | |
2113 | ||
2114 | EXPORT_SYMBOL(blk_insert_request); | |
2115 | ||
2116 | /** | |
2117 | * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage | |
2118 | * @q: request queue where request should be inserted | |
73747aed | 2119 | * @rq: request structure to fill |
1da177e4 LT |
2120 | * @ubuf: the user buffer |
2121 | * @len: length of user data | |
2122 | * | |
2123 | * Description: | |
2124 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2125 | * a kernel bounce buffer is used. | |
2126 | * | |
2127 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2128 | * still in process context. | |
2129 | * | |
2130 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2131 | * before being submitted to the device, as pages mapped may be out of | |
2132 | * reach. It's the callers responsibility to make sure this happens. The | |
2133 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2134 | * unmapping. | |
2135 | */ | |
dd1cab95 JA |
2136 | int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf, |
2137 | unsigned int len) | |
1da177e4 LT |
2138 | { |
2139 | unsigned long uaddr; | |
1da177e4 | 2140 | struct bio *bio; |
dd1cab95 | 2141 | int reading; |
1da177e4 LT |
2142 | |
2143 | if (len > (q->max_sectors << 9)) | |
dd1cab95 JA |
2144 | return -EINVAL; |
2145 | if (!len || !ubuf) | |
2146 | return -EINVAL; | |
1da177e4 | 2147 | |
dd1cab95 | 2148 | reading = rq_data_dir(rq) == READ; |
1da177e4 LT |
2149 | |
2150 | /* | |
2151 | * if alignment requirement is satisfied, map in user pages for | |
2152 | * direct dma. else, set up kernel bounce buffers | |
2153 | */ | |
2154 | uaddr = (unsigned long) ubuf; | |
2155 | if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q))) | |
dd1cab95 | 2156 | bio = bio_map_user(q, NULL, uaddr, len, reading); |
1da177e4 | 2157 | else |
dd1cab95 | 2158 | bio = bio_copy_user(q, uaddr, len, reading); |
1da177e4 LT |
2159 | |
2160 | if (!IS_ERR(bio)) { | |
2161 | rq->bio = rq->biotail = bio; | |
2162 | blk_rq_bio_prep(q, rq, bio); | |
2163 | ||
2164 | rq->buffer = rq->data = NULL; | |
2165 | rq->data_len = len; | |
dd1cab95 | 2166 | return 0; |
1da177e4 LT |
2167 | } |
2168 | ||
2169 | /* | |
2170 | * bio is the err-ptr | |
2171 | */ | |
dd1cab95 | 2172 | return PTR_ERR(bio); |
1da177e4 LT |
2173 | } |
2174 | ||
2175 | EXPORT_SYMBOL(blk_rq_map_user); | |
2176 | ||
f1970baf JB |
2177 | /** |
2178 | * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage | |
2179 | * @q: request queue where request should be inserted | |
2180 | * @rq: request to map data to | |
2181 | * @iov: pointer to the iovec | |
2182 | * @iov_count: number of elements in the iovec | |
2183 | * | |
2184 | * Description: | |
2185 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2186 | * a kernel bounce buffer is used. | |
2187 | * | |
2188 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2189 | * still in process context. | |
2190 | * | |
2191 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2192 | * before being submitted to the device, as pages mapped may be out of | |
2193 | * reach. It's the callers responsibility to make sure this happens. The | |
2194 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2195 | * unmapping. | |
2196 | */ | |
2197 | int blk_rq_map_user_iov(request_queue_t *q, struct request *rq, | |
2198 | struct sg_iovec *iov, int iov_count) | |
2199 | { | |
2200 | struct bio *bio; | |
2201 | ||
2202 | if (!iov || iov_count <= 0) | |
2203 | return -EINVAL; | |
2204 | ||
2205 | /* we don't allow misaligned data like bio_map_user() does. If the | |
2206 | * user is using sg, they're expected to know the alignment constraints | |
2207 | * and respect them accordingly */ | |
2208 | bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ); | |
2209 | if (IS_ERR(bio)) | |
2210 | return PTR_ERR(bio); | |
2211 | ||
2212 | rq->bio = rq->biotail = bio; | |
2213 | blk_rq_bio_prep(q, rq, bio); | |
2214 | rq->buffer = rq->data = NULL; | |
2215 | rq->data_len = bio->bi_size; | |
2216 | return 0; | |
2217 | } | |
2218 | ||
2219 | EXPORT_SYMBOL(blk_rq_map_user_iov); | |
2220 | ||
1da177e4 LT |
2221 | /** |
2222 | * blk_rq_unmap_user - unmap a request with user data | |
73747aed | 2223 | * @bio: bio to be unmapped |
1da177e4 LT |
2224 | * @ulen: length of user buffer |
2225 | * | |
2226 | * Description: | |
73747aed | 2227 | * Unmap a bio previously mapped by blk_rq_map_user(). |
1da177e4 | 2228 | */ |
dd1cab95 | 2229 | int blk_rq_unmap_user(struct bio *bio, unsigned int ulen) |
1da177e4 LT |
2230 | { |
2231 | int ret = 0; | |
2232 | ||
2233 | if (bio) { | |
2234 | if (bio_flagged(bio, BIO_USER_MAPPED)) | |
2235 | bio_unmap_user(bio); | |
2236 | else | |
2237 | ret = bio_uncopy_user(bio); | |
2238 | } | |
2239 | ||
dd1cab95 | 2240 | return 0; |
1da177e4 LT |
2241 | } |
2242 | ||
2243 | EXPORT_SYMBOL(blk_rq_unmap_user); | |
2244 | ||
df46b9a4 MC |
2245 | /** |
2246 | * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage | |
2247 | * @q: request queue where request should be inserted | |
73747aed | 2248 | * @rq: request to fill |
df46b9a4 MC |
2249 | * @kbuf: the kernel buffer |
2250 | * @len: length of user data | |
73747aed | 2251 | * @gfp_mask: memory allocation flags |
df46b9a4 | 2252 | */ |
dd1cab95 JA |
2253 | int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf, |
2254 | unsigned int len, unsigned int gfp_mask) | |
df46b9a4 | 2255 | { |
df46b9a4 MC |
2256 | struct bio *bio; |
2257 | ||
2258 | if (len > (q->max_sectors << 9)) | |
dd1cab95 JA |
2259 | return -EINVAL; |
2260 | if (!len || !kbuf) | |
2261 | return -EINVAL; | |
df46b9a4 MC |
2262 | |
2263 | bio = bio_map_kern(q, kbuf, len, gfp_mask); | |
dd1cab95 JA |
2264 | if (IS_ERR(bio)) |
2265 | return PTR_ERR(bio); | |
df46b9a4 | 2266 | |
dd1cab95 JA |
2267 | if (rq_data_dir(rq) == WRITE) |
2268 | bio->bi_rw |= (1 << BIO_RW); | |
df46b9a4 | 2269 | |
dd1cab95 JA |
2270 | rq->bio = rq->biotail = bio; |
2271 | blk_rq_bio_prep(q, rq, bio); | |
df46b9a4 | 2272 | |
dd1cab95 JA |
2273 | rq->buffer = rq->data = NULL; |
2274 | rq->data_len = len; | |
2275 | return 0; | |
df46b9a4 MC |
2276 | } |
2277 | ||
2278 | EXPORT_SYMBOL(blk_rq_map_kern); | |
2279 | ||
73747aed CH |
2280 | /** |
2281 | * blk_execute_rq_nowait - insert a request into queue for execution | |
2282 | * @q: queue to insert the request in | |
2283 | * @bd_disk: matching gendisk | |
2284 | * @rq: request to insert | |
2285 | * @at_head: insert request at head or tail of queue | |
2286 | * @done: I/O completion handler | |
2287 | * | |
2288 | * Description: | |
2289 | * Insert a fully prepared request at the back of the io scheduler queue | |
2290 | * for execution. Don't wait for completion. | |
2291 | */ | |
f1970baf JB |
2292 | void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk, |
2293 | struct request *rq, int at_head, | |
2294 | void (*done)(struct request *)) | |
2295 | { | |
2296 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | |
2297 | ||
2298 | rq->rq_disk = bd_disk; | |
2299 | rq->flags |= REQ_NOMERGE; | |
2300 | rq->end_io = done; | |
2301 | elv_add_request(q, rq, where, 1); | |
2302 | generic_unplug_device(q); | |
2303 | } | |
2304 | ||
1da177e4 LT |
2305 | /** |
2306 | * blk_execute_rq - insert a request into queue for execution | |
2307 | * @q: queue to insert the request in | |
2308 | * @bd_disk: matching gendisk | |
2309 | * @rq: request to insert | |
994ca9a1 | 2310 | * @at_head: insert request at head or tail of queue |
1da177e4 LT |
2311 | * |
2312 | * Description: | |
2313 | * Insert a fully prepared request at the back of the io scheduler queue | |
73747aed | 2314 | * for execution and wait for completion. |
1da177e4 LT |
2315 | */ |
2316 | int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk, | |
994ca9a1 | 2317 | struct request *rq, int at_head) |
1da177e4 LT |
2318 | { |
2319 | DECLARE_COMPLETION(wait); | |
2320 | char sense[SCSI_SENSE_BUFFERSIZE]; | |
2321 | int err = 0; | |
2322 | ||
1da177e4 LT |
2323 | /* |
2324 | * we need an extra reference to the request, so we can look at | |
2325 | * it after io completion | |
2326 | */ | |
2327 | rq->ref_count++; | |
2328 | ||
2329 | if (!rq->sense) { | |
2330 | memset(sense, 0, sizeof(sense)); | |
2331 | rq->sense = sense; | |
2332 | rq->sense_len = 0; | |
2333 | } | |
2334 | ||
1da177e4 | 2335 | rq->waiting = &wait; |
994ca9a1 | 2336 | blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq); |
1da177e4 LT |
2337 | wait_for_completion(&wait); |
2338 | rq->waiting = NULL; | |
2339 | ||
2340 | if (rq->errors) | |
2341 | err = -EIO; | |
2342 | ||
2343 | return err; | |
2344 | } | |
2345 | ||
2346 | EXPORT_SYMBOL(blk_execute_rq); | |
2347 | ||
2348 | /** | |
2349 | * blkdev_issue_flush - queue a flush | |
2350 | * @bdev: blockdev to issue flush for | |
2351 | * @error_sector: error sector | |
2352 | * | |
2353 | * Description: | |
2354 | * Issue a flush for the block device in question. Caller can supply | |
2355 | * room for storing the error offset in case of a flush error, if they | |
2356 | * wish to. Caller must run wait_for_completion() on its own. | |
2357 | */ | |
2358 | int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector) | |
2359 | { | |
2360 | request_queue_t *q; | |
2361 | ||
2362 | if (bdev->bd_disk == NULL) | |
2363 | return -ENXIO; | |
2364 | ||
2365 | q = bdev_get_queue(bdev); | |
2366 | if (!q) | |
2367 | return -ENXIO; | |
2368 | if (!q->issue_flush_fn) | |
2369 | return -EOPNOTSUPP; | |
2370 | ||
2371 | return q->issue_flush_fn(q, bdev->bd_disk, error_sector); | |
2372 | } | |
2373 | ||
2374 | EXPORT_SYMBOL(blkdev_issue_flush); | |
2375 | ||
93d17d3d | 2376 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io) |
1da177e4 LT |
2377 | { |
2378 | int rw = rq_data_dir(rq); | |
2379 | ||
2380 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
2381 | return; | |
2382 | ||
2383 | if (rw == READ) { | |
2384 | __disk_stat_add(rq->rq_disk, read_sectors, nr_sectors); | |
2385 | if (!new_io) | |
2386 | __disk_stat_inc(rq->rq_disk, read_merges); | |
2387 | } else if (rw == WRITE) { | |
2388 | __disk_stat_add(rq->rq_disk, write_sectors, nr_sectors); | |
2389 | if (!new_io) | |
2390 | __disk_stat_inc(rq->rq_disk, write_merges); | |
2391 | } | |
2392 | if (new_io) { | |
2393 | disk_round_stats(rq->rq_disk); | |
2394 | rq->rq_disk->in_flight++; | |
2395 | } | |
2396 | } | |
2397 | ||
2398 | /* | |
2399 | * add-request adds a request to the linked list. | |
2400 | * queue lock is held and interrupts disabled, as we muck with the | |
2401 | * request queue list. | |
2402 | */ | |
2403 | static inline void add_request(request_queue_t * q, struct request * req) | |
2404 | { | |
2405 | drive_stat_acct(req, req->nr_sectors, 1); | |
2406 | ||
2407 | if (q->activity_fn) | |
2408 | q->activity_fn(q->activity_data, rq_data_dir(req)); | |
2409 | ||
2410 | /* | |
2411 | * elevator indicated where it wants this request to be | |
2412 | * inserted at elevator_merge time | |
2413 | */ | |
2414 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
2415 | } | |
2416 | ||
2417 | /* | |
2418 | * disk_round_stats() - Round off the performance stats on a struct | |
2419 | * disk_stats. | |
2420 | * | |
2421 | * The average IO queue length and utilisation statistics are maintained | |
2422 | * by observing the current state of the queue length and the amount of | |
2423 | * time it has been in this state for. | |
2424 | * | |
2425 | * Normally, that accounting is done on IO completion, but that can result | |
2426 | * in more than a second's worth of IO being accounted for within any one | |
2427 | * second, leading to >100% utilisation. To deal with that, we call this | |
2428 | * function to do a round-off before returning the results when reading | |
2429 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
2430 | * the current jiffies and restarts the counters again. | |
2431 | */ | |
2432 | void disk_round_stats(struct gendisk *disk) | |
2433 | { | |
2434 | unsigned long now = jiffies; | |
2435 | ||
2436 | __disk_stat_add(disk, time_in_queue, | |
2437 | disk->in_flight * (now - disk->stamp)); | |
2438 | disk->stamp = now; | |
2439 | ||
2440 | if (disk->in_flight) | |
2441 | __disk_stat_add(disk, io_ticks, (now - disk->stamp_idle)); | |
2442 | disk->stamp_idle = now; | |
2443 | } | |
2444 | ||
2445 | /* | |
2446 | * queue lock must be held | |
2447 | */ | |
2448 | static void __blk_put_request(request_queue_t *q, struct request *req) | |
2449 | { | |
2450 | struct request_list *rl = req->rl; | |
2451 | ||
2452 | if (unlikely(!q)) | |
2453 | return; | |
2454 | if (unlikely(--req->ref_count)) | |
2455 | return; | |
2456 | ||
2457 | req->rq_status = RQ_INACTIVE; | |
1da177e4 LT |
2458 | req->rl = NULL; |
2459 | ||
2460 | /* | |
2461 | * Request may not have originated from ll_rw_blk. if not, | |
2462 | * it didn't come out of our reserved rq pools | |
2463 | */ | |
2464 | if (rl) { | |
2465 | int rw = rq_data_dir(req); | |
2466 | ||
2467 | elv_completed_request(q, req); | |
2468 | ||
2469 | BUG_ON(!list_empty(&req->queuelist)); | |
2470 | ||
2471 | blk_free_request(q, req); | |
2472 | freed_request(q, rw); | |
2473 | } | |
2474 | } | |
2475 | ||
2476 | void blk_put_request(struct request *req) | |
2477 | { | |
2478 | /* | |
2479 | * if req->rl isn't set, this request didnt originate from the | |
2480 | * block layer, so it's safe to just disregard it | |
2481 | */ | |
2482 | if (req->rl) { | |
2483 | unsigned long flags; | |
2484 | request_queue_t *q = req->q; | |
2485 | ||
2486 | spin_lock_irqsave(q->queue_lock, flags); | |
2487 | __blk_put_request(q, req); | |
2488 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2489 | } | |
2490 | } | |
2491 | ||
2492 | EXPORT_SYMBOL(blk_put_request); | |
2493 | ||
2494 | /** | |
2495 | * blk_end_sync_rq - executes a completion event on a request | |
2496 | * @rq: request to complete | |
2497 | */ | |
2498 | void blk_end_sync_rq(struct request *rq) | |
2499 | { | |
2500 | struct completion *waiting = rq->waiting; | |
2501 | ||
2502 | rq->waiting = NULL; | |
2503 | __blk_put_request(rq->q, rq); | |
2504 | ||
2505 | /* | |
2506 | * complete last, if this is a stack request the process (and thus | |
2507 | * the rq pointer) could be invalid right after this complete() | |
2508 | */ | |
2509 | complete(waiting); | |
2510 | } | |
2511 | EXPORT_SYMBOL(blk_end_sync_rq); | |
2512 | ||
2513 | /** | |
2514 | * blk_congestion_wait - wait for a queue to become uncongested | |
2515 | * @rw: READ or WRITE | |
2516 | * @timeout: timeout in jiffies | |
2517 | * | |
2518 | * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion. | |
2519 | * If no queues are congested then just wait for the next request to be | |
2520 | * returned. | |
2521 | */ | |
2522 | long blk_congestion_wait(int rw, long timeout) | |
2523 | { | |
2524 | long ret; | |
2525 | DEFINE_WAIT(wait); | |
2526 | wait_queue_head_t *wqh = &congestion_wqh[rw]; | |
2527 | ||
2528 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
2529 | ret = io_schedule_timeout(timeout); | |
2530 | finish_wait(wqh, &wait); | |
2531 | return ret; | |
2532 | } | |
2533 | ||
2534 | EXPORT_SYMBOL(blk_congestion_wait); | |
2535 | ||
2536 | /* | |
2537 | * Has to be called with the request spinlock acquired | |
2538 | */ | |
2539 | static int attempt_merge(request_queue_t *q, struct request *req, | |
2540 | struct request *next) | |
2541 | { | |
2542 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
2543 | return 0; | |
2544 | ||
2545 | /* | |
2546 | * not contigious | |
2547 | */ | |
2548 | if (req->sector + req->nr_sectors != next->sector) | |
2549 | return 0; | |
2550 | ||
2551 | if (rq_data_dir(req) != rq_data_dir(next) | |
2552 | || req->rq_disk != next->rq_disk | |
2553 | || next->waiting || next->special) | |
2554 | return 0; | |
2555 | ||
2556 | /* | |
2557 | * If we are allowed to merge, then append bio list | |
2558 | * from next to rq and release next. merge_requests_fn | |
2559 | * will have updated segment counts, update sector | |
2560 | * counts here. | |
2561 | */ | |
2562 | if (!q->merge_requests_fn(q, req, next)) | |
2563 | return 0; | |
2564 | ||
2565 | /* | |
2566 | * At this point we have either done a back merge | |
2567 | * or front merge. We need the smaller start_time of | |
2568 | * the merged requests to be the current request | |
2569 | * for accounting purposes. | |
2570 | */ | |
2571 | if (time_after(req->start_time, next->start_time)) | |
2572 | req->start_time = next->start_time; | |
2573 | ||
2574 | req->biotail->bi_next = next->bio; | |
2575 | req->biotail = next->biotail; | |
2576 | ||
2577 | req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors; | |
2578 | ||
2579 | elv_merge_requests(q, req, next); | |
2580 | ||
2581 | if (req->rq_disk) { | |
2582 | disk_round_stats(req->rq_disk); | |
2583 | req->rq_disk->in_flight--; | |
2584 | } | |
2585 | ||
22e2c507 JA |
2586 | req->ioprio = ioprio_best(req->ioprio, next->ioprio); |
2587 | ||
1da177e4 LT |
2588 | __blk_put_request(q, next); |
2589 | return 1; | |
2590 | } | |
2591 | ||
2592 | static inline int attempt_back_merge(request_queue_t *q, struct request *rq) | |
2593 | { | |
2594 | struct request *next = elv_latter_request(q, rq); | |
2595 | ||
2596 | if (next) | |
2597 | return attempt_merge(q, rq, next); | |
2598 | ||
2599 | return 0; | |
2600 | } | |
2601 | ||
2602 | static inline int attempt_front_merge(request_queue_t *q, struct request *rq) | |
2603 | { | |
2604 | struct request *prev = elv_former_request(q, rq); | |
2605 | ||
2606 | if (prev) | |
2607 | return attempt_merge(q, prev, rq); | |
2608 | ||
2609 | return 0; | |
2610 | } | |
2611 | ||
2612 | /** | |
2613 | * blk_attempt_remerge - attempt to remerge active head with next request | |
2614 | * @q: The &request_queue_t belonging to the device | |
2615 | * @rq: The head request (usually) | |
2616 | * | |
2617 | * Description: | |
2618 | * For head-active devices, the queue can easily be unplugged so quickly | |
2619 | * that proper merging is not done on the front request. This may hurt | |
2620 | * performance greatly for some devices. The block layer cannot safely | |
2621 | * do merging on that first request for these queues, but the driver can | |
2622 | * call this function and make it happen any way. Only the driver knows | |
2623 | * when it is safe to do so. | |
2624 | **/ | |
2625 | void blk_attempt_remerge(request_queue_t *q, struct request *rq) | |
2626 | { | |
2627 | unsigned long flags; | |
2628 | ||
2629 | spin_lock_irqsave(q->queue_lock, flags); | |
2630 | attempt_back_merge(q, rq); | |
2631 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2632 | } | |
2633 | ||
2634 | EXPORT_SYMBOL(blk_attempt_remerge); | |
2635 | ||
1da177e4 LT |
2636 | static int __make_request(request_queue_t *q, struct bio *bio) |
2637 | { | |
450991bc | 2638 | struct request *req; |
4a534f93 | 2639 | int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync; |
22e2c507 | 2640 | unsigned short prio; |
1da177e4 LT |
2641 | sector_t sector; |
2642 | ||
2643 | sector = bio->bi_sector; | |
2644 | nr_sectors = bio_sectors(bio); | |
2645 | cur_nr_sectors = bio_cur_sectors(bio); | |
22e2c507 | 2646 | prio = bio_prio(bio); |
1da177e4 LT |
2647 | |
2648 | rw = bio_data_dir(bio); | |
4a534f93 | 2649 | sync = bio_sync(bio); |
1da177e4 LT |
2650 | |
2651 | /* | |
2652 | * low level driver can indicate that it wants pages above a | |
2653 | * certain limit bounced to low memory (ie for highmem, or even | |
2654 | * ISA dma in theory) | |
2655 | */ | |
2656 | blk_queue_bounce(q, &bio); | |
2657 | ||
2658 | spin_lock_prefetch(q->queue_lock); | |
2659 | ||
2660 | barrier = bio_barrier(bio); | |
fde6ad22 | 2661 | if (unlikely(barrier) && (q->ordered == QUEUE_ORDERED_NONE)) { |
1da177e4 LT |
2662 | err = -EOPNOTSUPP; |
2663 | goto end_io; | |
2664 | } | |
2665 | ||
1da177e4 LT |
2666 | spin_lock_irq(q->queue_lock); |
2667 | ||
450991bc | 2668 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
2669 | goto get_rq; |
2670 | ||
2671 | el_ret = elv_merge(q, &req, bio); | |
2672 | switch (el_ret) { | |
2673 | case ELEVATOR_BACK_MERGE: | |
2674 | BUG_ON(!rq_mergeable(req)); | |
2675 | ||
2676 | if (!q->back_merge_fn(q, req, bio)) | |
2677 | break; | |
2678 | ||
2679 | req->biotail->bi_next = bio; | |
2680 | req->biotail = bio; | |
2681 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
22e2c507 | 2682 | req->ioprio = ioprio_best(req->ioprio, prio); |
1da177e4 LT |
2683 | drive_stat_acct(req, nr_sectors, 0); |
2684 | if (!attempt_back_merge(q, req)) | |
2685 | elv_merged_request(q, req); | |
2686 | goto out; | |
2687 | ||
2688 | case ELEVATOR_FRONT_MERGE: | |
2689 | BUG_ON(!rq_mergeable(req)); | |
2690 | ||
2691 | if (!q->front_merge_fn(q, req, bio)) | |
2692 | break; | |
2693 | ||
2694 | bio->bi_next = req->bio; | |
2695 | req->bio = bio; | |
2696 | ||
2697 | /* | |
2698 | * may not be valid. if the low level driver said | |
2699 | * it didn't need a bounce buffer then it better | |
2700 | * not touch req->buffer either... | |
2701 | */ | |
2702 | req->buffer = bio_data(bio); | |
2703 | req->current_nr_sectors = cur_nr_sectors; | |
2704 | req->hard_cur_sectors = cur_nr_sectors; | |
2705 | req->sector = req->hard_sector = sector; | |
2706 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
22e2c507 | 2707 | req->ioprio = ioprio_best(req->ioprio, prio); |
1da177e4 LT |
2708 | drive_stat_acct(req, nr_sectors, 0); |
2709 | if (!attempt_front_merge(q, req)) | |
2710 | elv_merged_request(q, req); | |
2711 | goto out; | |
2712 | ||
450991bc | 2713 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ |
1da177e4 | 2714 | default: |
450991bc | 2715 | ; |
1da177e4 LT |
2716 | } |
2717 | ||
450991bc | 2718 | get_rq: |
1da177e4 | 2719 | /* |
450991bc | 2720 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 2721 | * Returns with the queue unlocked. |
450991bc | 2722 | */ |
450991bc | 2723 | req = get_request_wait(q, rw, bio); |
d6344532 | 2724 | |
450991bc NP |
2725 | /* |
2726 | * After dropping the lock and possibly sleeping here, our request | |
2727 | * may now be mergeable after it had proven unmergeable (above). | |
2728 | * We don't worry about that case for efficiency. It won't happen | |
2729 | * often, and the elevators are able to handle it. | |
1da177e4 | 2730 | */ |
1da177e4 LT |
2731 | |
2732 | req->flags |= REQ_CMD; | |
2733 | ||
2734 | /* | |
2735 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
2736 | */ | |
2737 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
2738 | req->flags |= REQ_FAILFAST; | |
2739 | ||
2740 | /* | |
2741 | * REQ_BARRIER implies no merging, but lets make it explicit | |
2742 | */ | |
fde6ad22 | 2743 | if (unlikely(barrier)) |
1da177e4 LT |
2744 | req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE); |
2745 | ||
2746 | req->errors = 0; | |
2747 | req->hard_sector = req->sector = sector; | |
2748 | req->hard_nr_sectors = req->nr_sectors = nr_sectors; | |
2749 | req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors; | |
2750 | req->nr_phys_segments = bio_phys_segments(q, bio); | |
2751 | req->nr_hw_segments = bio_hw_segments(q, bio); | |
2752 | req->buffer = bio_data(bio); /* see ->buffer comment above */ | |
2753 | req->waiting = NULL; | |
2754 | req->bio = req->biotail = bio; | |
22e2c507 | 2755 | req->ioprio = prio; |
1da177e4 LT |
2756 | req->rq_disk = bio->bi_bdev->bd_disk; |
2757 | req->start_time = jiffies; | |
2758 | ||
450991bc NP |
2759 | spin_lock_irq(q->queue_lock); |
2760 | if (elv_queue_empty(q)) | |
2761 | blk_plug_device(q); | |
1da177e4 LT |
2762 | add_request(q, req); |
2763 | out: | |
4a534f93 | 2764 | if (sync) |
1da177e4 LT |
2765 | __generic_unplug_device(q); |
2766 | ||
2767 | spin_unlock_irq(q->queue_lock); | |
2768 | return 0; | |
2769 | ||
2770 | end_io: | |
2771 | bio_endio(bio, nr_sectors << 9, err); | |
2772 | return 0; | |
2773 | } | |
2774 | ||
2775 | /* | |
2776 | * If bio->bi_dev is a partition, remap the location | |
2777 | */ | |
2778 | static inline void blk_partition_remap(struct bio *bio) | |
2779 | { | |
2780 | struct block_device *bdev = bio->bi_bdev; | |
2781 | ||
2782 | if (bdev != bdev->bd_contains) { | |
2783 | struct hd_struct *p = bdev->bd_part; | |
2784 | ||
22e2c507 | 2785 | switch (bio_data_dir(bio)) { |
1da177e4 LT |
2786 | case READ: |
2787 | p->read_sectors += bio_sectors(bio); | |
2788 | p->reads++; | |
2789 | break; | |
2790 | case WRITE: | |
2791 | p->write_sectors += bio_sectors(bio); | |
2792 | p->writes++; | |
2793 | break; | |
2794 | } | |
2795 | bio->bi_sector += p->start_sect; | |
2796 | bio->bi_bdev = bdev->bd_contains; | |
2797 | } | |
2798 | } | |
2799 | ||
2800 | void blk_finish_queue_drain(request_queue_t *q) | |
2801 | { | |
2802 | struct request_list *rl = &q->rq; | |
2803 | struct request *rq; | |
22e2c507 | 2804 | int requeued = 0; |
1da177e4 LT |
2805 | |
2806 | spin_lock_irq(q->queue_lock); | |
2807 | clear_bit(QUEUE_FLAG_DRAIN, &q->queue_flags); | |
2808 | ||
2809 | while (!list_empty(&q->drain_list)) { | |
2810 | rq = list_entry_rq(q->drain_list.next); | |
2811 | ||
2812 | list_del_init(&rq->queuelist); | |
22e2c507 JA |
2813 | elv_requeue_request(q, rq); |
2814 | requeued++; | |
1da177e4 LT |
2815 | } |
2816 | ||
22e2c507 JA |
2817 | if (requeued) |
2818 | q->request_fn(q); | |
2819 | ||
1da177e4 LT |
2820 | spin_unlock_irq(q->queue_lock); |
2821 | ||
2822 | wake_up(&rl->wait[0]); | |
2823 | wake_up(&rl->wait[1]); | |
2824 | wake_up(&rl->drain); | |
2825 | } | |
2826 | ||
2827 | static int wait_drain(request_queue_t *q, struct request_list *rl, int dispatch) | |
2828 | { | |
2829 | int wait = rl->count[READ] + rl->count[WRITE]; | |
2830 | ||
2831 | if (dispatch) | |
2832 | wait += !list_empty(&q->queue_head); | |
2833 | ||
2834 | return wait; | |
2835 | } | |
2836 | ||
2837 | /* | |
2838 | * We rely on the fact that only requests allocated through blk_alloc_request() | |
2839 | * have io scheduler private data structures associated with them. Any other | |
2840 | * type of request (allocated on stack or through kmalloc()) should not go | |
2841 | * to the io scheduler core, but be attached to the queue head instead. | |
2842 | */ | |
2843 | void blk_wait_queue_drained(request_queue_t *q, int wait_dispatch) | |
2844 | { | |
2845 | struct request_list *rl = &q->rq; | |
2846 | DEFINE_WAIT(wait); | |
2847 | ||
2848 | spin_lock_irq(q->queue_lock); | |
2849 | set_bit(QUEUE_FLAG_DRAIN, &q->queue_flags); | |
2850 | ||
2851 | while (wait_drain(q, rl, wait_dispatch)) { | |
2852 | prepare_to_wait(&rl->drain, &wait, TASK_UNINTERRUPTIBLE); | |
2853 | ||
2854 | if (wait_drain(q, rl, wait_dispatch)) { | |
2855 | __generic_unplug_device(q); | |
2856 | spin_unlock_irq(q->queue_lock); | |
2857 | io_schedule(); | |
2858 | spin_lock_irq(q->queue_lock); | |
2859 | } | |
2860 | ||
2861 | finish_wait(&rl->drain, &wait); | |
2862 | } | |
2863 | ||
2864 | spin_unlock_irq(q->queue_lock); | |
2865 | } | |
2866 | ||
2867 | /* | |
2868 | * block waiting for the io scheduler being started again. | |
2869 | */ | |
2870 | static inline void block_wait_queue_running(request_queue_t *q) | |
2871 | { | |
2872 | DEFINE_WAIT(wait); | |
2873 | ||
fde6ad22 | 2874 | while (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) { |
1da177e4 LT |
2875 | struct request_list *rl = &q->rq; |
2876 | ||
2877 | prepare_to_wait_exclusive(&rl->drain, &wait, | |
2878 | TASK_UNINTERRUPTIBLE); | |
2879 | ||
2880 | /* | |
2881 | * re-check the condition. avoids using prepare_to_wait() | |
2882 | * in the fast path (queue is running) | |
2883 | */ | |
2884 | if (test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)) | |
2885 | io_schedule(); | |
2886 | ||
2887 | finish_wait(&rl->drain, &wait); | |
2888 | } | |
2889 | } | |
2890 | ||
2891 | static void handle_bad_sector(struct bio *bio) | |
2892 | { | |
2893 | char b[BDEVNAME_SIZE]; | |
2894 | ||
2895 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
2896 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
2897 | bdevname(bio->bi_bdev, b), | |
2898 | bio->bi_rw, | |
2899 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
2900 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
2901 | ||
2902 | set_bit(BIO_EOF, &bio->bi_flags); | |
2903 | } | |
2904 | ||
2905 | /** | |
2906 | * generic_make_request: hand a buffer to its device driver for I/O | |
2907 | * @bio: The bio describing the location in memory and on the device. | |
2908 | * | |
2909 | * generic_make_request() is used to make I/O requests of block | |
2910 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
2911 | * to be done. | |
2912 | * | |
2913 | * generic_make_request() does not return any status. The | |
2914 | * success/failure status of the request, along with notification of | |
2915 | * completion, is delivered asynchronously through the bio->bi_end_io | |
2916 | * function described (one day) else where. | |
2917 | * | |
2918 | * The caller of generic_make_request must make sure that bi_io_vec | |
2919 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
2920 | * set to describe the device address, and the | |
2921 | * bi_end_io and optionally bi_private are set to describe how | |
2922 | * completion notification should be signaled. | |
2923 | * | |
2924 | * generic_make_request and the drivers it calls may use bi_next if this | |
2925 | * bio happens to be merged with someone else, and may change bi_dev and | |
2926 | * bi_sector for remaps as it sees fit. So the values of these fields | |
2927 | * should NOT be depended on after the call to generic_make_request. | |
2928 | */ | |
2929 | void generic_make_request(struct bio *bio) | |
2930 | { | |
2931 | request_queue_t *q; | |
2932 | sector_t maxsector; | |
2933 | int ret, nr_sectors = bio_sectors(bio); | |
2934 | ||
2935 | might_sleep(); | |
2936 | /* Test device or partition size, when known. */ | |
2937 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
2938 | if (maxsector) { | |
2939 | sector_t sector = bio->bi_sector; | |
2940 | ||
2941 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
2942 | /* | |
2943 | * This may well happen - the kernel calls bread() | |
2944 | * without checking the size of the device, e.g., when | |
2945 | * mounting a device. | |
2946 | */ | |
2947 | handle_bad_sector(bio); | |
2948 | goto end_io; | |
2949 | } | |
2950 | } | |
2951 | ||
2952 | /* | |
2953 | * Resolve the mapping until finished. (drivers are | |
2954 | * still free to implement/resolve their own stacking | |
2955 | * by explicitly returning 0) | |
2956 | * | |
2957 | * NOTE: we don't repeat the blk_size check for each new device. | |
2958 | * Stacking drivers are expected to know what they are doing. | |
2959 | */ | |
2960 | do { | |
2961 | char b[BDEVNAME_SIZE]; | |
2962 | ||
2963 | q = bdev_get_queue(bio->bi_bdev); | |
2964 | if (!q) { | |
2965 | printk(KERN_ERR | |
2966 | "generic_make_request: Trying to access " | |
2967 | "nonexistent block-device %s (%Lu)\n", | |
2968 | bdevname(bio->bi_bdev, b), | |
2969 | (long long) bio->bi_sector); | |
2970 | end_io: | |
2971 | bio_endio(bio, bio->bi_size, -EIO); | |
2972 | break; | |
2973 | } | |
2974 | ||
2975 | if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) { | |
2976 | printk("bio too big device %s (%u > %u)\n", | |
2977 | bdevname(bio->bi_bdev, b), | |
2978 | bio_sectors(bio), | |
2979 | q->max_hw_sectors); | |
2980 | goto end_io; | |
2981 | } | |
2982 | ||
fde6ad22 | 2983 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
2984 | goto end_io; |
2985 | ||
2986 | block_wait_queue_running(q); | |
2987 | ||
2988 | /* | |
2989 | * If this device has partitions, remap block n | |
2990 | * of partition p to block n+start(p) of the disk. | |
2991 | */ | |
2992 | blk_partition_remap(bio); | |
2993 | ||
2994 | ret = q->make_request_fn(q, bio); | |
2995 | } while (ret); | |
2996 | } | |
2997 | ||
2998 | EXPORT_SYMBOL(generic_make_request); | |
2999 | ||
3000 | /** | |
3001 | * submit_bio: submit a bio to the block device layer for I/O | |
3002 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
3003 | * @bio: The &struct bio which describes the I/O | |
3004 | * | |
3005 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
3006 | * uses that function to do most of the work. Both are fairly rough | |
3007 | * interfaces, @bio must be presetup and ready for I/O. | |
3008 | * | |
3009 | */ | |
3010 | void submit_bio(int rw, struct bio *bio) | |
3011 | { | |
3012 | int count = bio_sectors(bio); | |
3013 | ||
3014 | BIO_BUG_ON(!bio->bi_size); | |
3015 | BIO_BUG_ON(!bio->bi_io_vec); | |
22e2c507 | 3016 | bio->bi_rw |= rw; |
1da177e4 LT |
3017 | if (rw & WRITE) |
3018 | mod_page_state(pgpgout, count); | |
3019 | else | |
3020 | mod_page_state(pgpgin, count); | |
3021 | ||
3022 | if (unlikely(block_dump)) { | |
3023 | char b[BDEVNAME_SIZE]; | |
3024 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
3025 | current->comm, current->pid, | |
3026 | (rw & WRITE) ? "WRITE" : "READ", | |
3027 | (unsigned long long)bio->bi_sector, | |
3028 | bdevname(bio->bi_bdev,b)); | |
3029 | } | |
3030 | ||
3031 | generic_make_request(bio); | |
3032 | } | |
3033 | ||
3034 | EXPORT_SYMBOL(submit_bio); | |
3035 | ||
93d17d3d | 3036 | static void blk_recalc_rq_segments(struct request *rq) |
1da177e4 LT |
3037 | { |
3038 | struct bio *bio, *prevbio = NULL; | |
3039 | int nr_phys_segs, nr_hw_segs; | |
3040 | unsigned int phys_size, hw_size; | |
3041 | request_queue_t *q = rq->q; | |
3042 | ||
3043 | if (!rq->bio) | |
3044 | return; | |
3045 | ||
3046 | phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0; | |
3047 | rq_for_each_bio(bio, rq) { | |
3048 | /* Force bio hw/phys segs to be recalculated. */ | |
3049 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3050 | ||
3051 | nr_phys_segs += bio_phys_segments(q, bio); | |
3052 | nr_hw_segs += bio_hw_segments(q, bio); | |
3053 | if (prevbio) { | |
3054 | int pseg = phys_size + prevbio->bi_size + bio->bi_size; | |
3055 | int hseg = hw_size + prevbio->bi_size + bio->bi_size; | |
3056 | ||
3057 | if (blk_phys_contig_segment(q, prevbio, bio) && | |
3058 | pseg <= q->max_segment_size) { | |
3059 | nr_phys_segs--; | |
3060 | phys_size += prevbio->bi_size + bio->bi_size; | |
3061 | } else | |
3062 | phys_size = 0; | |
3063 | ||
3064 | if (blk_hw_contig_segment(q, prevbio, bio) && | |
3065 | hseg <= q->max_segment_size) { | |
3066 | nr_hw_segs--; | |
3067 | hw_size += prevbio->bi_size + bio->bi_size; | |
3068 | } else | |
3069 | hw_size = 0; | |
3070 | } | |
3071 | prevbio = bio; | |
3072 | } | |
3073 | ||
3074 | rq->nr_phys_segments = nr_phys_segs; | |
3075 | rq->nr_hw_segments = nr_hw_segs; | |
3076 | } | |
3077 | ||
93d17d3d | 3078 | static void blk_recalc_rq_sectors(struct request *rq, int nsect) |
1da177e4 LT |
3079 | { |
3080 | if (blk_fs_request(rq)) { | |
3081 | rq->hard_sector += nsect; | |
3082 | rq->hard_nr_sectors -= nsect; | |
3083 | ||
3084 | /* | |
3085 | * Move the I/O submission pointers ahead if required. | |
3086 | */ | |
3087 | if ((rq->nr_sectors >= rq->hard_nr_sectors) && | |
3088 | (rq->sector <= rq->hard_sector)) { | |
3089 | rq->sector = rq->hard_sector; | |
3090 | rq->nr_sectors = rq->hard_nr_sectors; | |
3091 | rq->hard_cur_sectors = bio_cur_sectors(rq->bio); | |
3092 | rq->current_nr_sectors = rq->hard_cur_sectors; | |
3093 | rq->buffer = bio_data(rq->bio); | |
3094 | } | |
3095 | ||
3096 | /* | |
3097 | * if total number of sectors is less than the first segment | |
3098 | * size, something has gone terribly wrong | |
3099 | */ | |
3100 | if (rq->nr_sectors < rq->current_nr_sectors) { | |
3101 | printk("blk: request botched\n"); | |
3102 | rq->nr_sectors = rq->current_nr_sectors; | |
3103 | } | |
3104 | } | |
3105 | } | |
3106 | ||
3107 | static int __end_that_request_first(struct request *req, int uptodate, | |
3108 | int nr_bytes) | |
3109 | { | |
3110 | int total_bytes, bio_nbytes, error, next_idx = 0; | |
3111 | struct bio *bio; | |
3112 | ||
3113 | /* | |
3114 | * extend uptodate bool to allow < 0 value to be direct io error | |
3115 | */ | |
3116 | error = 0; | |
3117 | if (end_io_error(uptodate)) | |
3118 | error = !uptodate ? -EIO : uptodate; | |
3119 | ||
3120 | /* | |
3121 | * for a REQ_BLOCK_PC request, we want to carry any eventual | |
3122 | * sense key with us all the way through | |
3123 | */ | |
3124 | if (!blk_pc_request(req)) | |
3125 | req->errors = 0; | |
3126 | ||
3127 | if (!uptodate) { | |
3128 | if (blk_fs_request(req) && !(req->flags & REQ_QUIET)) | |
3129 | printk("end_request: I/O error, dev %s, sector %llu\n", | |
3130 | req->rq_disk ? req->rq_disk->disk_name : "?", | |
3131 | (unsigned long long)req->sector); | |
3132 | } | |
3133 | ||
3134 | total_bytes = bio_nbytes = 0; | |
3135 | while ((bio = req->bio) != NULL) { | |
3136 | int nbytes; | |
3137 | ||
3138 | if (nr_bytes >= bio->bi_size) { | |
3139 | req->bio = bio->bi_next; | |
3140 | nbytes = bio->bi_size; | |
3141 | bio_endio(bio, nbytes, error); | |
3142 | next_idx = 0; | |
3143 | bio_nbytes = 0; | |
3144 | } else { | |
3145 | int idx = bio->bi_idx + next_idx; | |
3146 | ||
3147 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
3148 | blk_dump_rq_flags(req, "__end_that"); | |
3149 | printk("%s: bio idx %d >= vcnt %d\n", | |
3150 | __FUNCTION__, | |
3151 | bio->bi_idx, bio->bi_vcnt); | |
3152 | break; | |
3153 | } | |
3154 | ||
3155 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
3156 | BIO_BUG_ON(nbytes > bio->bi_size); | |
3157 | ||
3158 | /* | |
3159 | * not a complete bvec done | |
3160 | */ | |
3161 | if (unlikely(nbytes > nr_bytes)) { | |
3162 | bio_nbytes += nr_bytes; | |
3163 | total_bytes += nr_bytes; | |
3164 | break; | |
3165 | } | |
3166 | ||
3167 | /* | |
3168 | * advance to the next vector | |
3169 | */ | |
3170 | next_idx++; | |
3171 | bio_nbytes += nbytes; | |
3172 | } | |
3173 | ||
3174 | total_bytes += nbytes; | |
3175 | nr_bytes -= nbytes; | |
3176 | ||
3177 | if ((bio = req->bio)) { | |
3178 | /* | |
3179 | * end more in this run, or just return 'not-done' | |
3180 | */ | |
3181 | if (unlikely(nr_bytes <= 0)) | |
3182 | break; | |
3183 | } | |
3184 | } | |
3185 | ||
3186 | /* | |
3187 | * completely done | |
3188 | */ | |
3189 | if (!req->bio) | |
3190 | return 0; | |
3191 | ||
3192 | /* | |
3193 | * if the request wasn't completed, update state | |
3194 | */ | |
3195 | if (bio_nbytes) { | |
3196 | bio_endio(bio, bio_nbytes, error); | |
3197 | bio->bi_idx += next_idx; | |
3198 | bio_iovec(bio)->bv_offset += nr_bytes; | |
3199 | bio_iovec(bio)->bv_len -= nr_bytes; | |
3200 | } | |
3201 | ||
3202 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
3203 | blk_recalc_rq_segments(req); | |
3204 | return 1; | |
3205 | } | |
3206 | ||
3207 | /** | |
3208 | * end_that_request_first - end I/O on a request | |
3209 | * @req: the request being processed | |
3210 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3211 | * @nr_sectors: number of sectors to end I/O on | |
3212 | * | |
3213 | * Description: | |
3214 | * Ends I/O on a number of sectors attached to @req, and sets it up | |
3215 | * for the next range of segments (if any) in the cluster. | |
3216 | * | |
3217 | * Return: | |
3218 | * 0 - we are done with this request, call end_that_request_last() | |
3219 | * 1 - still buffers pending for this request | |
3220 | **/ | |
3221 | int end_that_request_first(struct request *req, int uptodate, int nr_sectors) | |
3222 | { | |
3223 | return __end_that_request_first(req, uptodate, nr_sectors << 9); | |
3224 | } | |
3225 | ||
3226 | EXPORT_SYMBOL(end_that_request_first); | |
3227 | ||
3228 | /** | |
3229 | * end_that_request_chunk - end I/O on a request | |
3230 | * @req: the request being processed | |
3231 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3232 | * @nr_bytes: number of bytes to complete | |
3233 | * | |
3234 | * Description: | |
3235 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
3236 | * for the next range of segments (if any). Like end_that_request_first(), | |
3237 | * but deals with bytes instead of sectors. | |
3238 | * | |
3239 | * Return: | |
3240 | * 0 - we are done with this request, call end_that_request_last() | |
3241 | * 1 - still buffers pending for this request | |
3242 | **/ | |
3243 | int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes) | |
3244 | { | |
3245 | return __end_that_request_first(req, uptodate, nr_bytes); | |
3246 | } | |
3247 | ||
3248 | EXPORT_SYMBOL(end_that_request_chunk); | |
3249 | ||
3250 | /* | |
3251 | * queue lock must be held | |
3252 | */ | |
3253 | void end_that_request_last(struct request *req) | |
3254 | { | |
3255 | struct gendisk *disk = req->rq_disk; | |
3256 | ||
3257 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
3258 | laptop_io_completion(); | |
3259 | ||
3260 | if (disk && blk_fs_request(req)) { | |
3261 | unsigned long duration = jiffies - req->start_time; | |
3262 | switch (rq_data_dir(req)) { | |
3263 | case WRITE: | |
3264 | __disk_stat_inc(disk, writes); | |
3265 | __disk_stat_add(disk, write_ticks, duration); | |
3266 | break; | |
3267 | case READ: | |
3268 | __disk_stat_inc(disk, reads); | |
3269 | __disk_stat_add(disk, read_ticks, duration); | |
3270 | break; | |
3271 | } | |
3272 | disk_round_stats(disk); | |
3273 | disk->in_flight--; | |
3274 | } | |
3275 | if (req->end_io) | |
3276 | req->end_io(req); | |
3277 | else | |
3278 | __blk_put_request(req->q, req); | |
3279 | } | |
3280 | ||
3281 | EXPORT_SYMBOL(end_that_request_last); | |
3282 | ||
3283 | void end_request(struct request *req, int uptodate) | |
3284 | { | |
3285 | if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) { | |
3286 | add_disk_randomness(req->rq_disk); | |
3287 | blkdev_dequeue_request(req); | |
3288 | end_that_request_last(req); | |
3289 | } | |
3290 | } | |
3291 | ||
3292 | EXPORT_SYMBOL(end_request); | |
3293 | ||
3294 | void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio) | |
3295 | { | |
3296 | /* first three bits are identical in rq->flags and bio->bi_rw */ | |
3297 | rq->flags |= (bio->bi_rw & 7); | |
3298 | ||
3299 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
3300 | rq->nr_hw_segments = bio_hw_segments(q, bio); | |
3301 | rq->current_nr_sectors = bio_cur_sectors(bio); | |
3302 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
3303 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
3304 | rq->buffer = bio_data(bio); | |
3305 | ||
3306 | rq->bio = rq->biotail = bio; | |
3307 | } | |
3308 | ||
3309 | EXPORT_SYMBOL(blk_rq_bio_prep); | |
3310 | ||
3311 | int kblockd_schedule_work(struct work_struct *work) | |
3312 | { | |
3313 | return queue_work(kblockd_workqueue, work); | |
3314 | } | |
3315 | ||
3316 | EXPORT_SYMBOL(kblockd_schedule_work); | |
3317 | ||
3318 | void kblockd_flush(void) | |
3319 | { | |
3320 | flush_workqueue(kblockd_workqueue); | |
3321 | } | |
3322 | EXPORT_SYMBOL(kblockd_flush); | |
3323 | ||
3324 | int __init blk_dev_init(void) | |
3325 | { | |
3326 | kblockd_workqueue = create_workqueue("kblockd"); | |
3327 | if (!kblockd_workqueue) | |
3328 | panic("Failed to create kblockd\n"); | |
3329 | ||
3330 | request_cachep = kmem_cache_create("blkdev_requests", | |
3331 | sizeof(struct request), 0, SLAB_PANIC, NULL, NULL); | |
3332 | ||
3333 | requestq_cachep = kmem_cache_create("blkdev_queue", | |
3334 | sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL); | |
3335 | ||
3336 | iocontext_cachep = kmem_cache_create("blkdev_ioc", | |
3337 | sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL); | |
3338 | ||
3339 | blk_max_low_pfn = max_low_pfn; | |
3340 | blk_max_pfn = max_pfn; | |
3341 | ||
3342 | return 0; | |
3343 | } | |
3344 | ||
3345 | /* | |
3346 | * IO Context helper functions | |
3347 | */ | |
3348 | void put_io_context(struct io_context *ioc) | |
3349 | { | |
3350 | if (ioc == NULL) | |
3351 | return; | |
3352 | ||
3353 | BUG_ON(atomic_read(&ioc->refcount) == 0); | |
3354 | ||
3355 | if (atomic_dec_and_test(&ioc->refcount)) { | |
3356 | if (ioc->aic && ioc->aic->dtor) | |
3357 | ioc->aic->dtor(ioc->aic); | |
3358 | if (ioc->cic && ioc->cic->dtor) | |
3359 | ioc->cic->dtor(ioc->cic); | |
3360 | ||
3361 | kmem_cache_free(iocontext_cachep, ioc); | |
3362 | } | |
3363 | } | |
3364 | EXPORT_SYMBOL(put_io_context); | |
3365 | ||
3366 | /* Called by the exitting task */ | |
3367 | void exit_io_context(void) | |
3368 | { | |
3369 | unsigned long flags; | |
3370 | struct io_context *ioc; | |
3371 | ||
3372 | local_irq_save(flags); | |
22e2c507 | 3373 | task_lock(current); |
1da177e4 LT |
3374 | ioc = current->io_context; |
3375 | current->io_context = NULL; | |
22e2c507 JA |
3376 | ioc->task = NULL; |
3377 | task_unlock(current); | |
1da177e4 LT |
3378 | local_irq_restore(flags); |
3379 | ||
3380 | if (ioc->aic && ioc->aic->exit) | |
3381 | ioc->aic->exit(ioc->aic); | |
3382 | if (ioc->cic && ioc->cic->exit) | |
3383 | ioc->cic->exit(ioc->cic); | |
3384 | ||
3385 | put_io_context(ioc); | |
3386 | } | |
3387 | ||
3388 | /* | |
3389 | * If the current task has no IO context then create one and initialise it. | |
fb3cc432 | 3390 | * Otherwise, return its existing IO context. |
1da177e4 | 3391 | * |
fb3cc432 NP |
3392 | * This returned IO context doesn't have a specifically elevated refcount, |
3393 | * but since the current task itself holds a reference, the context can be | |
3394 | * used in general code, so long as it stays within `current` context. | |
1da177e4 | 3395 | */ |
fb3cc432 | 3396 | struct io_context *current_io_context(int gfp_flags) |
1da177e4 LT |
3397 | { |
3398 | struct task_struct *tsk = current; | |
1da177e4 LT |
3399 | struct io_context *ret; |
3400 | ||
1da177e4 | 3401 | ret = tsk->io_context; |
fb3cc432 NP |
3402 | if (likely(ret)) |
3403 | return ret; | |
1da177e4 LT |
3404 | |
3405 | ret = kmem_cache_alloc(iocontext_cachep, gfp_flags); | |
3406 | if (ret) { | |
3407 | atomic_set(&ret->refcount, 1); | |
22e2c507 JA |
3408 | ret->task = current; |
3409 | ret->set_ioprio = NULL; | |
1da177e4 LT |
3410 | ret->last_waited = jiffies; /* doesn't matter... */ |
3411 | ret->nr_batch_requests = 0; /* because this is 0 */ | |
3412 | ret->aic = NULL; | |
3413 | ret->cic = NULL; | |
fb3cc432 NP |
3414 | tsk->io_context = ret; |
3415 | } | |
1da177e4 | 3416 | |
fb3cc432 NP |
3417 | return ret; |
3418 | } | |
3419 | EXPORT_SYMBOL(current_io_context); | |
1da177e4 | 3420 | |
fb3cc432 NP |
3421 | /* |
3422 | * If the current task has no IO context then create one and initialise it. | |
3423 | * If it does have a context, take a ref on it. | |
3424 | * | |
3425 | * This is always called in the context of the task which submitted the I/O. | |
3426 | */ | |
3427 | struct io_context *get_io_context(int gfp_flags) | |
3428 | { | |
3429 | struct io_context *ret; | |
3430 | ret = current_io_context(gfp_flags); | |
3431 | if (likely(ret)) | |
1da177e4 | 3432 | atomic_inc(&ret->refcount); |
1da177e4 LT |
3433 | return ret; |
3434 | } | |
3435 | EXPORT_SYMBOL(get_io_context); | |
3436 | ||
3437 | void copy_io_context(struct io_context **pdst, struct io_context **psrc) | |
3438 | { | |
3439 | struct io_context *src = *psrc; | |
3440 | struct io_context *dst = *pdst; | |
3441 | ||
3442 | if (src) { | |
3443 | BUG_ON(atomic_read(&src->refcount) == 0); | |
3444 | atomic_inc(&src->refcount); | |
3445 | put_io_context(dst); | |
3446 | *pdst = src; | |
3447 | } | |
3448 | } | |
3449 | EXPORT_SYMBOL(copy_io_context); | |
3450 | ||
3451 | void swap_io_context(struct io_context **ioc1, struct io_context **ioc2) | |
3452 | { | |
3453 | struct io_context *temp; | |
3454 | temp = *ioc1; | |
3455 | *ioc1 = *ioc2; | |
3456 | *ioc2 = temp; | |
3457 | } | |
3458 | EXPORT_SYMBOL(swap_io_context); | |
3459 | ||
3460 | /* | |
3461 | * sysfs parts below | |
3462 | */ | |
3463 | struct queue_sysfs_entry { | |
3464 | struct attribute attr; | |
3465 | ssize_t (*show)(struct request_queue *, char *); | |
3466 | ssize_t (*store)(struct request_queue *, const char *, size_t); | |
3467 | }; | |
3468 | ||
3469 | static ssize_t | |
3470 | queue_var_show(unsigned int var, char *page) | |
3471 | { | |
3472 | return sprintf(page, "%d\n", var); | |
3473 | } | |
3474 | ||
3475 | static ssize_t | |
3476 | queue_var_store(unsigned long *var, const char *page, size_t count) | |
3477 | { | |
3478 | char *p = (char *) page; | |
3479 | ||
3480 | *var = simple_strtoul(p, &p, 10); | |
3481 | return count; | |
3482 | } | |
3483 | ||
3484 | static ssize_t queue_requests_show(struct request_queue *q, char *page) | |
3485 | { | |
3486 | return queue_var_show(q->nr_requests, (page)); | |
3487 | } | |
3488 | ||
3489 | static ssize_t | |
3490 | queue_requests_store(struct request_queue *q, const char *page, size_t count) | |
3491 | { | |
3492 | struct request_list *rl = &q->rq; | |
3493 | ||
3494 | int ret = queue_var_store(&q->nr_requests, page, count); | |
3495 | if (q->nr_requests < BLKDEV_MIN_RQ) | |
3496 | q->nr_requests = BLKDEV_MIN_RQ; | |
3497 | blk_queue_congestion_threshold(q); | |
3498 | ||
3499 | if (rl->count[READ] >= queue_congestion_on_threshold(q)) | |
3500 | set_queue_congested(q, READ); | |
3501 | else if (rl->count[READ] < queue_congestion_off_threshold(q)) | |
3502 | clear_queue_congested(q, READ); | |
3503 | ||
3504 | if (rl->count[WRITE] >= queue_congestion_on_threshold(q)) | |
3505 | set_queue_congested(q, WRITE); | |
3506 | else if (rl->count[WRITE] < queue_congestion_off_threshold(q)) | |
3507 | clear_queue_congested(q, WRITE); | |
3508 | ||
3509 | if (rl->count[READ] >= q->nr_requests) { | |
3510 | blk_set_queue_full(q, READ); | |
3511 | } else if (rl->count[READ]+1 <= q->nr_requests) { | |
3512 | blk_clear_queue_full(q, READ); | |
3513 | wake_up(&rl->wait[READ]); | |
3514 | } | |
3515 | ||
3516 | if (rl->count[WRITE] >= q->nr_requests) { | |
3517 | blk_set_queue_full(q, WRITE); | |
3518 | } else if (rl->count[WRITE]+1 <= q->nr_requests) { | |
3519 | blk_clear_queue_full(q, WRITE); | |
3520 | wake_up(&rl->wait[WRITE]); | |
3521 | } | |
3522 | return ret; | |
3523 | } | |
3524 | ||
3525 | static ssize_t queue_ra_show(struct request_queue *q, char *page) | |
3526 | { | |
3527 | int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | |
3528 | ||
3529 | return queue_var_show(ra_kb, (page)); | |
3530 | } | |
3531 | ||
3532 | static ssize_t | |
3533 | queue_ra_store(struct request_queue *q, const char *page, size_t count) | |
3534 | { | |
3535 | unsigned long ra_kb; | |
3536 | ssize_t ret = queue_var_store(&ra_kb, page, count); | |
3537 | ||
3538 | spin_lock_irq(q->queue_lock); | |
3539 | if (ra_kb > (q->max_sectors >> 1)) | |
3540 | ra_kb = (q->max_sectors >> 1); | |
3541 | ||
3542 | q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); | |
3543 | spin_unlock_irq(q->queue_lock); | |
3544 | ||
3545 | return ret; | |
3546 | } | |
3547 | ||
3548 | static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) | |
3549 | { | |
3550 | int max_sectors_kb = q->max_sectors >> 1; | |
3551 | ||
3552 | return queue_var_show(max_sectors_kb, (page)); | |
3553 | } | |
3554 | ||
3555 | static ssize_t | |
3556 | queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) | |
3557 | { | |
3558 | unsigned long max_sectors_kb, | |
3559 | max_hw_sectors_kb = q->max_hw_sectors >> 1, | |
3560 | page_kb = 1 << (PAGE_CACHE_SHIFT - 10); | |
3561 | ssize_t ret = queue_var_store(&max_sectors_kb, page, count); | |
3562 | int ra_kb; | |
3563 | ||
3564 | if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) | |
3565 | return -EINVAL; | |
3566 | /* | |
3567 | * Take the queue lock to update the readahead and max_sectors | |
3568 | * values synchronously: | |
3569 | */ | |
3570 | spin_lock_irq(q->queue_lock); | |
3571 | /* | |
3572 | * Trim readahead window as well, if necessary: | |
3573 | */ | |
3574 | ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | |
3575 | if (ra_kb > max_sectors_kb) | |
3576 | q->backing_dev_info.ra_pages = | |
3577 | max_sectors_kb >> (PAGE_CACHE_SHIFT - 10); | |
3578 | ||
3579 | q->max_sectors = max_sectors_kb << 1; | |
3580 | spin_unlock_irq(q->queue_lock); | |
3581 | ||
3582 | return ret; | |
3583 | } | |
3584 | ||
3585 | static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) | |
3586 | { | |
3587 | int max_hw_sectors_kb = q->max_hw_sectors >> 1; | |
3588 | ||
3589 | return queue_var_show(max_hw_sectors_kb, (page)); | |
3590 | } | |
3591 | ||
3592 | ||
3593 | static struct queue_sysfs_entry queue_requests_entry = { | |
3594 | .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, | |
3595 | .show = queue_requests_show, | |
3596 | .store = queue_requests_store, | |
3597 | }; | |
3598 | ||
3599 | static struct queue_sysfs_entry queue_ra_entry = { | |
3600 | .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, | |
3601 | .show = queue_ra_show, | |
3602 | .store = queue_ra_store, | |
3603 | }; | |
3604 | ||
3605 | static struct queue_sysfs_entry queue_max_sectors_entry = { | |
3606 | .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, | |
3607 | .show = queue_max_sectors_show, | |
3608 | .store = queue_max_sectors_store, | |
3609 | }; | |
3610 | ||
3611 | static struct queue_sysfs_entry queue_max_hw_sectors_entry = { | |
3612 | .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, | |
3613 | .show = queue_max_hw_sectors_show, | |
3614 | }; | |
3615 | ||
3616 | static struct queue_sysfs_entry queue_iosched_entry = { | |
3617 | .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, | |
3618 | .show = elv_iosched_show, | |
3619 | .store = elv_iosched_store, | |
3620 | }; | |
3621 | ||
3622 | static struct attribute *default_attrs[] = { | |
3623 | &queue_requests_entry.attr, | |
3624 | &queue_ra_entry.attr, | |
3625 | &queue_max_hw_sectors_entry.attr, | |
3626 | &queue_max_sectors_entry.attr, | |
3627 | &queue_iosched_entry.attr, | |
3628 | NULL, | |
3629 | }; | |
3630 | ||
3631 | #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) | |
3632 | ||
3633 | static ssize_t | |
3634 | queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
3635 | { | |
3636 | struct queue_sysfs_entry *entry = to_queue(attr); | |
3637 | struct request_queue *q; | |
3638 | ||
3639 | q = container_of(kobj, struct request_queue, kobj); | |
3640 | if (!entry->show) | |
6c1852a0 | 3641 | return -EIO; |
1da177e4 LT |
3642 | |
3643 | return entry->show(q, page); | |
3644 | } | |
3645 | ||
3646 | static ssize_t | |
3647 | queue_attr_store(struct kobject *kobj, struct attribute *attr, | |
3648 | const char *page, size_t length) | |
3649 | { | |
3650 | struct queue_sysfs_entry *entry = to_queue(attr); | |
3651 | struct request_queue *q; | |
3652 | ||
3653 | q = container_of(kobj, struct request_queue, kobj); | |
3654 | if (!entry->store) | |
6c1852a0 | 3655 | return -EIO; |
1da177e4 LT |
3656 | |
3657 | return entry->store(q, page, length); | |
3658 | } | |
3659 | ||
3660 | static struct sysfs_ops queue_sysfs_ops = { | |
3661 | .show = queue_attr_show, | |
3662 | .store = queue_attr_store, | |
3663 | }; | |
3664 | ||
93d17d3d | 3665 | static struct kobj_type queue_ktype = { |
1da177e4 LT |
3666 | .sysfs_ops = &queue_sysfs_ops, |
3667 | .default_attrs = default_attrs, | |
3668 | }; | |
3669 | ||
3670 | int blk_register_queue(struct gendisk *disk) | |
3671 | { | |
3672 | int ret; | |
3673 | ||
3674 | request_queue_t *q = disk->queue; | |
3675 | ||
3676 | if (!q || !q->request_fn) | |
3677 | return -ENXIO; | |
3678 | ||
3679 | q->kobj.parent = kobject_get(&disk->kobj); | |
3680 | if (!q->kobj.parent) | |
3681 | return -EBUSY; | |
3682 | ||
3683 | snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue"); | |
3684 | q->kobj.ktype = &queue_ktype; | |
3685 | ||
3686 | ret = kobject_register(&q->kobj); | |
3687 | if (ret < 0) | |
3688 | return ret; | |
3689 | ||
3690 | ret = elv_register_queue(q); | |
3691 | if (ret) { | |
3692 | kobject_unregister(&q->kobj); | |
3693 | return ret; | |
3694 | } | |
3695 | ||
3696 | return 0; | |
3697 | } | |
3698 | ||
3699 | void blk_unregister_queue(struct gendisk *disk) | |
3700 | { | |
3701 | request_queue_t *q = disk->queue; | |
3702 | ||
3703 | if (q && q->request_fn) { | |
3704 | elv_unregister_queue(q); | |
3705 | ||
3706 | kobject_unregister(&q->kobj); | |
3707 | kobject_put(&disk->kobj); | |
3708 | } | |
3709 | } |