diff options
Diffstat (limited to 'block/deadline-iosched.c')
-rw-r--r-- | block/deadline-iosched.c | 878 |
1 files changed, 878 insertions, 0 deletions
diff --git a/block/deadline-iosched.c b/block/deadline-iosched.c new file mode 100644 index 000000000000..7929471d7df7 --- /dev/null +++ b/block/deadline-iosched.c | |||
@@ -0,0 +1,878 @@ | |||
1 | /* | ||
2 | * linux/drivers/block/deadline-iosched.c | ||
3 | * | ||
4 | * Deadline i/o scheduler. | ||
5 | * | ||
6 | * Copyright (C) 2002 Jens Axboe <axboe@suse.de> | ||
7 | */ | ||
8 | #include <linux/kernel.h> | ||
9 | #include <linux/fs.h> | ||
10 | #include <linux/blkdev.h> | ||
11 | #include <linux/elevator.h> | ||
12 | #include <linux/bio.h> | ||
13 | #include <linux/config.h> | ||
14 | #include <linux/module.h> | ||
15 | #include <linux/slab.h> | ||
16 | #include <linux/init.h> | ||
17 | #include <linux/compiler.h> | ||
18 | #include <linux/hash.h> | ||
19 | #include <linux/rbtree.h> | ||
20 | |||
21 | /* | ||
22 | * See Documentation/block/deadline-iosched.txt | ||
23 | */ | ||
24 | static int read_expire = HZ / 2; /* max time before a read is submitted. */ | ||
25 | static int write_expire = 5 * HZ; /* ditto for writes, these limits are SOFT! */ | ||
26 | static int writes_starved = 2; /* max times reads can starve a write */ | ||
27 | static int fifo_batch = 16; /* # of sequential requests treated as one | ||
28 | by the above parameters. For throughput. */ | ||
29 | |||
30 | static const int deadline_hash_shift = 5; | ||
31 | #define DL_HASH_BLOCK(sec) ((sec) >> 3) | ||
32 | #define DL_HASH_FN(sec) (hash_long(DL_HASH_BLOCK((sec)), deadline_hash_shift)) | ||
33 | #define DL_HASH_ENTRIES (1 << deadline_hash_shift) | ||
34 | #define rq_hash_key(rq) ((rq)->sector + (rq)->nr_sectors) | ||
35 | #define list_entry_hash(ptr) list_entry((ptr), struct deadline_rq, hash) | ||
36 | #define ON_HASH(drq) (drq)->on_hash | ||
37 | |||
38 | struct deadline_data { | ||
39 | /* | ||
40 | * run time data | ||
41 | */ | ||
42 | |||
43 | /* | ||
44 | * requests (deadline_rq s) are present on both sort_list and fifo_list | ||
45 | */ | ||
46 | struct rb_root sort_list[2]; | ||
47 | struct list_head fifo_list[2]; | ||
48 | |||
49 | /* | ||
50 | * next in sort order. read, write or both are NULL | ||
51 | */ | ||
52 | struct deadline_rq *next_drq[2]; | ||
53 | struct list_head *hash; /* request hash */ | ||
54 | unsigned int batching; /* number of sequential requests made */ | ||
55 | sector_t last_sector; /* head position */ | ||
56 | unsigned int starved; /* times reads have starved writes */ | ||
57 | |||
58 | /* | ||
59 | * settings that change how the i/o scheduler behaves | ||
60 | */ | ||
61 | int fifo_expire[2]; | ||
62 | int fifo_batch; | ||
63 | int writes_starved; | ||
64 | int front_merges; | ||
65 | |||
66 | mempool_t *drq_pool; | ||
67 | }; | ||
68 | |||
69 | /* | ||
70 | * pre-request data. | ||
71 | */ | ||
72 | struct deadline_rq { | ||
73 | /* | ||
74 | * rbtree index, key is the starting offset | ||
75 | */ | ||
76 | struct rb_node rb_node; | ||
77 | sector_t rb_key; | ||
78 | |||
79 | struct request *request; | ||
80 | |||
81 | /* | ||
82 | * request hash, key is the ending offset (for back merge lookup) | ||
83 | */ | ||
84 | struct list_head hash; | ||
85 | char on_hash; | ||
86 | |||
87 | /* | ||
88 | * expire fifo | ||
89 | */ | ||
90 | struct list_head fifo; | ||
91 | unsigned long expires; | ||
92 | }; | ||
93 | |||
94 | static void deadline_move_request(struct deadline_data *dd, struct deadline_rq *drq); | ||
95 | |||
96 | static kmem_cache_t *drq_pool; | ||
97 | |||
98 | #define RQ_DATA(rq) ((struct deadline_rq *) (rq)->elevator_private) | ||
99 | |||
100 | /* | ||
101 | * the back merge hash support functions | ||
102 | */ | ||
103 | static inline void __deadline_del_drq_hash(struct deadline_rq *drq) | ||
104 | { | ||
105 | drq->on_hash = 0; | ||
106 | list_del_init(&drq->hash); | ||
107 | } | ||
108 | |||
109 | static inline void deadline_del_drq_hash(struct deadline_rq *drq) | ||
110 | { | ||
111 | if (ON_HASH(drq)) | ||
112 | __deadline_del_drq_hash(drq); | ||
113 | } | ||
114 | |||
115 | static inline void | ||
116 | deadline_add_drq_hash(struct deadline_data *dd, struct deadline_rq *drq) | ||
117 | { | ||
118 | struct request *rq = drq->request; | ||
119 | |||
120 | BUG_ON(ON_HASH(drq)); | ||
121 | |||
122 | drq->on_hash = 1; | ||
123 | list_add(&drq->hash, &dd->hash[DL_HASH_FN(rq_hash_key(rq))]); | ||
124 | } | ||
125 | |||
126 | /* | ||
127 | * move hot entry to front of chain | ||
128 | */ | ||
129 | static inline void | ||
130 | deadline_hot_drq_hash(struct deadline_data *dd, struct deadline_rq *drq) | ||
131 | { | ||
132 | struct request *rq = drq->request; | ||
133 | struct list_head *head = &dd->hash[DL_HASH_FN(rq_hash_key(rq))]; | ||
134 | |||
135 | if (ON_HASH(drq) && drq->hash.prev != head) { | ||
136 | list_del(&drq->hash); | ||
137 | list_add(&drq->hash, head); | ||
138 | } | ||
139 | } | ||
140 | |||
141 | static struct request * | ||
142 | deadline_find_drq_hash(struct deadline_data *dd, sector_t offset) | ||
143 | { | ||
144 | struct list_head *hash_list = &dd->hash[DL_HASH_FN(offset)]; | ||
145 | struct list_head *entry, *next = hash_list->next; | ||
146 | |||
147 | while ((entry = next) != hash_list) { | ||
148 | struct deadline_rq *drq = list_entry_hash(entry); | ||
149 | struct request *__rq = drq->request; | ||
150 | |||
151 | next = entry->next; | ||
152 | |||
153 | BUG_ON(!ON_HASH(drq)); | ||
154 | |||
155 | if (!rq_mergeable(__rq)) { | ||
156 | __deadline_del_drq_hash(drq); | ||
157 | continue; | ||
158 | } | ||
159 | |||
160 | if (rq_hash_key(__rq) == offset) | ||
161 | return __rq; | ||
162 | } | ||
163 | |||
164 | return NULL; | ||
165 | } | ||
166 | |||
167 | /* | ||
168 | * rb tree support functions | ||
169 | */ | ||
170 | #define RB_NONE (2) | ||
171 | #define RB_EMPTY(root) ((root)->rb_node == NULL) | ||
172 | #define ON_RB(node) ((node)->rb_color != RB_NONE) | ||
173 | #define RB_CLEAR(node) ((node)->rb_color = RB_NONE) | ||
174 | #define rb_entry_drq(node) rb_entry((node), struct deadline_rq, rb_node) | ||
175 | #define DRQ_RB_ROOT(dd, drq) (&(dd)->sort_list[rq_data_dir((drq)->request)]) | ||
176 | #define rq_rb_key(rq) (rq)->sector | ||
177 | |||
178 | static struct deadline_rq * | ||
179 | __deadline_add_drq_rb(struct deadline_data *dd, struct deadline_rq *drq) | ||
180 | { | ||
181 | struct rb_node **p = &DRQ_RB_ROOT(dd, drq)->rb_node; | ||
182 | struct rb_node *parent = NULL; | ||
183 | struct deadline_rq *__drq; | ||
184 | |||
185 | while (*p) { | ||
186 | parent = *p; | ||
187 | __drq = rb_entry_drq(parent); | ||
188 | |||
189 | if (drq->rb_key < __drq->rb_key) | ||
190 | p = &(*p)->rb_left; | ||
191 | else if (drq->rb_key > __drq->rb_key) | ||
192 | p = &(*p)->rb_right; | ||
193 | else | ||
194 | return __drq; | ||
195 | } | ||
196 | |||
197 | rb_link_node(&drq->rb_node, parent, p); | ||
198 | return NULL; | ||
199 | } | ||
200 | |||
201 | static void | ||
202 | deadline_add_drq_rb(struct deadline_data *dd, struct deadline_rq *drq) | ||
203 | { | ||
204 | struct deadline_rq *__alias; | ||
205 | |||
206 | drq->rb_key = rq_rb_key(drq->request); | ||
207 | |||
208 | retry: | ||
209 | __alias = __deadline_add_drq_rb(dd, drq); | ||
210 | if (!__alias) { | ||
211 | rb_insert_color(&drq->rb_node, DRQ_RB_ROOT(dd, drq)); | ||
212 | return; | ||
213 | } | ||
214 | |||
215 | deadline_move_request(dd, __alias); | ||
216 | goto retry; | ||
217 | } | ||
218 | |||
219 | static inline void | ||
220 | deadline_del_drq_rb(struct deadline_data *dd, struct deadline_rq *drq) | ||
221 | { | ||
222 | const int data_dir = rq_data_dir(drq->request); | ||
223 | |||
224 | if (dd->next_drq[data_dir] == drq) { | ||
225 | struct rb_node *rbnext = rb_next(&drq->rb_node); | ||
226 | |||
227 | dd->next_drq[data_dir] = NULL; | ||
228 | if (rbnext) | ||
229 | dd->next_drq[data_dir] = rb_entry_drq(rbnext); | ||
230 | } | ||
231 | |||
232 | BUG_ON(!ON_RB(&drq->rb_node)); | ||
233 | rb_erase(&drq->rb_node, DRQ_RB_ROOT(dd, drq)); | ||
234 | RB_CLEAR(&drq->rb_node); | ||
235 | } | ||
236 | |||
237 | static struct request * | ||
238 | deadline_find_drq_rb(struct deadline_data *dd, sector_t sector, int data_dir) | ||
239 | { | ||
240 | struct rb_node *n = dd->sort_list[data_dir].rb_node; | ||
241 | struct deadline_rq *drq; | ||
242 | |||
243 | while (n) { | ||
244 | drq = rb_entry_drq(n); | ||
245 | |||
246 | if (sector < drq->rb_key) | ||
247 | n = n->rb_left; | ||
248 | else if (sector > drq->rb_key) | ||
249 | n = n->rb_right; | ||
250 | else | ||
251 | return drq->request; | ||
252 | } | ||
253 | |||
254 | return NULL; | ||
255 | } | ||
256 | |||
257 | /* | ||
258 | * deadline_find_first_drq finds the first (lowest sector numbered) request | ||
259 | * for the specified data_dir. Used to sweep back to the start of the disk | ||
260 | * (1-way elevator) after we process the last (highest sector) request. | ||
261 | */ | ||
262 | static struct deadline_rq * | ||
263 | deadline_find_first_drq(struct deadline_data *dd, int data_dir) | ||
264 | { | ||
265 | struct rb_node *n = dd->sort_list[data_dir].rb_node; | ||
266 | |||
267 | for (;;) { | ||
268 | if (n->rb_left == NULL) | ||
269 | return rb_entry_drq(n); | ||
270 | |||
271 | n = n->rb_left; | ||
272 | } | ||
273 | } | ||
274 | |||
275 | /* | ||
276 | * add drq to rbtree and fifo | ||
277 | */ | ||
278 | static void | ||
279 | deadline_add_request(struct request_queue *q, struct request *rq) | ||
280 | { | ||
281 | struct deadline_data *dd = q->elevator->elevator_data; | ||
282 | struct deadline_rq *drq = RQ_DATA(rq); | ||
283 | |||
284 | const int data_dir = rq_data_dir(drq->request); | ||
285 | |||
286 | deadline_add_drq_rb(dd, drq); | ||
287 | /* | ||
288 | * set expire time (only used for reads) and add to fifo list | ||
289 | */ | ||
290 | drq->expires = jiffies + dd->fifo_expire[data_dir]; | ||
291 | list_add_tail(&drq->fifo, &dd->fifo_list[data_dir]); | ||
292 | |||
293 | if (rq_mergeable(rq)) | ||
294 | deadline_add_drq_hash(dd, drq); | ||
295 | } | ||
296 | |||
297 | /* | ||
298 | * remove rq from rbtree, fifo, and hash | ||
299 | */ | ||
300 | static void deadline_remove_request(request_queue_t *q, struct request *rq) | ||
301 | { | ||
302 | struct deadline_rq *drq = RQ_DATA(rq); | ||
303 | struct deadline_data *dd = q->elevator->elevator_data; | ||
304 | |||
305 | list_del_init(&drq->fifo); | ||
306 | deadline_del_drq_rb(dd, drq); | ||
307 | deadline_del_drq_hash(drq); | ||
308 | } | ||
309 | |||
310 | static int | ||
311 | deadline_merge(request_queue_t *q, struct request **req, struct bio *bio) | ||
312 | { | ||
313 | struct deadline_data *dd = q->elevator->elevator_data; | ||
314 | struct request *__rq; | ||
315 | int ret; | ||
316 | |||
317 | /* | ||
318 | * see if the merge hash can satisfy a back merge | ||
319 | */ | ||
320 | __rq = deadline_find_drq_hash(dd, bio->bi_sector); | ||
321 | if (__rq) { | ||
322 | BUG_ON(__rq->sector + __rq->nr_sectors != bio->bi_sector); | ||
323 | |||
324 | if (elv_rq_merge_ok(__rq, bio)) { | ||
325 | ret = ELEVATOR_BACK_MERGE; | ||
326 | goto out; | ||
327 | } | ||
328 | } | ||
329 | |||
330 | /* | ||
331 | * check for front merge | ||
332 | */ | ||
333 | if (dd->front_merges) { | ||
334 | sector_t rb_key = bio->bi_sector + bio_sectors(bio); | ||
335 | |||
336 | __rq = deadline_find_drq_rb(dd, rb_key, bio_data_dir(bio)); | ||
337 | if (__rq) { | ||
338 | BUG_ON(rb_key != rq_rb_key(__rq)); | ||
339 | |||
340 | if (elv_rq_merge_ok(__rq, bio)) { | ||
341 | ret = ELEVATOR_FRONT_MERGE; | ||
342 | goto out; | ||
343 | } | ||
344 | } | ||
345 | } | ||
346 | |||
347 | return ELEVATOR_NO_MERGE; | ||
348 | out: | ||
349 | if (ret) | ||
350 | deadline_hot_drq_hash(dd, RQ_DATA(__rq)); | ||
351 | *req = __rq; | ||
352 | return ret; | ||
353 | } | ||
354 | |||
355 | static void deadline_merged_request(request_queue_t *q, struct request *req) | ||
356 | { | ||
357 | struct deadline_data *dd = q->elevator->elevator_data; | ||
358 | struct deadline_rq *drq = RQ_DATA(req); | ||
359 | |||
360 | /* | ||
361 | * hash always needs to be repositioned, key is end sector | ||
362 | */ | ||
363 | deadline_del_drq_hash(drq); | ||
364 | deadline_add_drq_hash(dd, drq); | ||
365 | |||
366 | /* | ||
367 | * if the merge was a front merge, we need to reposition request | ||
368 | */ | ||
369 | if (rq_rb_key(req) != drq->rb_key) { | ||
370 | deadline_del_drq_rb(dd, drq); | ||
371 | deadline_add_drq_rb(dd, drq); | ||
372 | } | ||
373 | } | ||
374 | |||
375 | static void | ||
376 | deadline_merged_requests(request_queue_t *q, struct request *req, | ||
377 | struct request *next) | ||
378 | { | ||
379 | struct deadline_data *dd = q->elevator->elevator_data; | ||
380 | struct deadline_rq *drq = RQ_DATA(req); | ||
381 | struct deadline_rq *dnext = RQ_DATA(next); | ||
382 | |||
383 | BUG_ON(!drq); | ||
384 | BUG_ON(!dnext); | ||
385 | |||
386 | /* | ||
387 | * reposition drq (this is the merged request) in hash, and in rbtree | ||
388 | * in case of a front merge | ||
389 | */ | ||
390 | deadline_del_drq_hash(drq); | ||
391 | deadline_add_drq_hash(dd, drq); | ||
392 | |||
393 | if (rq_rb_key(req) != drq->rb_key) { | ||
394 | deadline_del_drq_rb(dd, drq); | ||
395 | deadline_add_drq_rb(dd, drq); | ||
396 | } | ||
397 | |||
398 | /* | ||
399 | * if dnext expires before drq, assign its expire time to drq | ||
400 | * and move into dnext position (dnext will be deleted) in fifo | ||
401 | */ | ||
402 | if (!list_empty(&drq->fifo) && !list_empty(&dnext->fifo)) { | ||
403 | if (time_before(dnext->expires, drq->expires)) { | ||
404 | list_move(&drq->fifo, &dnext->fifo); | ||
405 | drq->expires = dnext->expires; | ||
406 | } | ||
407 | } | ||
408 | |||
409 | /* | ||
410 | * kill knowledge of next, this one is a goner | ||
411 | */ | ||
412 | deadline_remove_request(q, next); | ||
413 | } | ||
414 | |||
415 | /* | ||
416 | * move request from sort list to dispatch queue. | ||
417 | */ | ||
418 | static inline void | ||
419 | deadline_move_to_dispatch(struct deadline_data *dd, struct deadline_rq *drq) | ||
420 | { | ||
421 | request_queue_t *q = drq->request->q; | ||
422 | |||
423 | deadline_remove_request(q, drq->request); | ||
424 | elv_dispatch_add_tail(q, drq->request); | ||
425 | } | ||
426 | |||
427 | /* | ||
428 | * move an entry to dispatch queue | ||
429 | */ | ||
430 | static void | ||
431 | deadline_move_request(struct deadline_data *dd, struct deadline_rq *drq) | ||
432 | { | ||
433 | const int data_dir = rq_data_dir(drq->request); | ||
434 | struct rb_node *rbnext = rb_next(&drq->rb_node); | ||
435 | |||
436 | dd->next_drq[READ] = NULL; | ||
437 | dd->next_drq[WRITE] = NULL; | ||
438 | |||
439 | if (rbnext) | ||
440 | dd->next_drq[data_dir] = rb_entry_drq(rbnext); | ||
441 | |||
442 | dd->last_sector = drq->request->sector + drq->request->nr_sectors; | ||
443 | |||
444 | /* | ||
445 | * take it off the sort and fifo list, move | ||
446 | * to dispatch queue | ||
447 | */ | ||
448 | deadline_move_to_dispatch(dd, drq); | ||
449 | } | ||
450 | |||
451 | #define list_entry_fifo(ptr) list_entry((ptr), struct deadline_rq, fifo) | ||
452 | |||
453 | /* | ||
454 | * deadline_check_fifo returns 0 if there are no expired reads on the fifo, | ||
455 | * 1 otherwise. Requires !list_empty(&dd->fifo_list[data_dir]) | ||
456 | */ | ||
457 | static inline int deadline_check_fifo(struct deadline_data *dd, int ddir) | ||
458 | { | ||
459 | struct deadline_rq *drq = list_entry_fifo(dd->fifo_list[ddir].next); | ||
460 | |||
461 | /* | ||
462 | * drq is expired! | ||
463 | */ | ||
464 | if (time_after(jiffies, drq->expires)) | ||
465 | return 1; | ||
466 | |||
467 | return 0; | ||
468 | } | ||
469 | |||
470 | /* | ||
471 | * deadline_dispatch_requests selects the best request according to | ||
472 | * read/write expire, fifo_batch, etc | ||
473 | */ | ||
474 | static int deadline_dispatch_requests(request_queue_t *q, int force) | ||
475 | { | ||
476 | struct deadline_data *dd = q->elevator->elevator_data; | ||
477 | const int reads = !list_empty(&dd->fifo_list[READ]); | ||
478 | const int writes = !list_empty(&dd->fifo_list[WRITE]); | ||
479 | struct deadline_rq *drq; | ||
480 | int data_dir; | ||
481 | |||
482 | /* | ||
483 | * batches are currently reads XOR writes | ||
484 | */ | ||
485 | if (dd->next_drq[WRITE]) | ||
486 | drq = dd->next_drq[WRITE]; | ||
487 | else | ||
488 | drq = dd->next_drq[READ]; | ||
489 | |||
490 | if (drq) { | ||
491 | /* we have a "next request" */ | ||
492 | |||
493 | if (dd->last_sector != drq->request->sector) | ||
494 | /* end the batch on a non sequential request */ | ||
495 | dd->batching += dd->fifo_batch; | ||
496 | |||
497 | if (dd->batching < dd->fifo_batch) | ||
498 | /* we are still entitled to batch */ | ||
499 | goto dispatch_request; | ||
500 | } | ||
501 | |||
502 | /* | ||
503 | * at this point we are not running a batch. select the appropriate | ||
504 | * data direction (read / write) | ||
505 | */ | ||
506 | |||
507 | if (reads) { | ||
508 | BUG_ON(RB_EMPTY(&dd->sort_list[READ])); | ||
509 | |||
510 | if (writes && (dd->starved++ >= dd->writes_starved)) | ||
511 | goto dispatch_writes; | ||
512 | |||
513 | data_dir = READ; | ||
514 | |||
515 | goto dispatch_find_request; | ||
516 | } | ||
517 | |||
518 | /* | ||
519 | * there are either no reads or writes have been starved | ||
520 | */ | ||
521 | |||
522 | if (writes) { | ||
523 | dispatch_writes: | ||
524 | BUG_ON(RB_EMPTY(&dd->sort_list[WRITE])); | ||
525 | |||
526 | dd->starved = 0; | ||
527 | |||
528 | data_dir = WRITE; | ||
529 | |||
530 | goto dispatch_find_request; | ||
531 | } | ||
532 | |||
533 | return 0; | ||
534 | |||
535 | dispatch_find_request: | ||
536 | /* | ||
537 | * we are not running a batch, find best request for selected data_dir | ||
538 | */ | ||
539 | if (deadline_check_fifo(dd, data_dir)) { | ||
540 | /* An expired request exists - satisfy it */ | ||
541 | dd->batching = 0; | ||
542 | drq = list_entry_fifo(dd->fifo_list[data_dir].next); | ||
543 | |||
544 | } else if (dd->next_drq[data_dir]) { | ||
545 | /* | ||
546 | * The last req was the same dir and we have a next request in | ||
547 | * sort order. No expired requests so continue on from here. | ||
548 | */ | ||
549 | drq = dd->next_drq[data_dir]; | ||
550 | } else { | ||
551 | /* | ||
552 | * The last req was the other direction or we have run out of | ||
553 | * higher-sectored requests. Go back to the lowest sectored | ||
554 | * request (1 way elevator) and start a new batch. | ||
555 | */ | ||
556 | dd->batching = 0; | ||
557 | drq = deadline_find_first_drq(dd, data_dir); | ||
558 | } | ||
559 | |||
560 | dispatch_request: | ||
561 | /* | ||
562 | * drq is the selected appropriate request. | ||
563 | */ | ||
564 | dd->batching++; | ||
565 | deadline_move_request(dd, drq); | ||
566 | |||
567 | return 1; | ||
568 | } | ||
569 | |||
570 | static int deadline_queue_empty(request_queue_t *q) | ||
571 | { | ||
572 | struct deadline_data *dd = q->elevator->elevator_data; | ||
573 | |||
574 | return list_empty(&dd->fifo_list[WRITE]) | ||
575 | && list_empty(&dd->fifo_list[READ]); | ||
576 | } | ||
577 | |||
578 | static struct request * | ||
579 | deadline_former_request(request_queue_t *q, struct request *rq) | ||
580 | { | ||
581 | struct deadline_rq *drq = RQ_DATA(rq); | ||
582 | struct rb_node *rbprev = rb_prev(&drq->rb_node); | ||
583 | |||
584 | if (rbprev) | ||
585 | return rb_entry_drq(rbprev)->request; | ||
586 | |||
587 | return NULL; | ||
588 | } | ||
589 | |||
590 | static struct request * | ||
591 | deadline_latter_request(request_queue_t *q, struct request *rq) | ||
592 | { | ||
593 | struct deadline_rq *drq = RQ_DATA(rq); | ||
594 | struct rb_node *rbnext = rb_next(&drq->rb_node); | ||
595 | |||
596 | if (rbnext) | ||
597 | return rb_entry_drq(rbnext)->request; | ||
598 | |||
599 | return NULL; | ||
600 | } | ||
601 | |||
602 | static void deadline_exit_queue(elevator_t *e) | ||
603 | { | ||
604 | struct deadline_data *dd = e->elevator_data; | ||
605 | |||
606 | BUG_ON(!list_empty(&dd->fifo_list[READ])); | ||
607 | BUG_ON(!list_empty(&dd->fifo_list[WRITE])); | ||
608 | |||
609 | mempool_destroy(dd->drq_pool); | ||
610 | kfree(dd->hash); | ||
611 | kfree(dd); | ||
612 | } | ||
613 | |||
614 | /* | ||
615 | * initialize elevator private data (deadline_data), and alloc a drq for | ||
616 | * each request on the free lists | ||
617 | */ | ||
618 | static int deadline_init_queue(request_queue_t *q, elevator_t *e) | ||
619 | { | ||
620 | struct deadline_data *dd; | ||
621 | int i; | ||
622 | |||
623 | if (!drq_pool) | ||
624 | return -ENOMEM; | ||
625 | |||
626 | dd = kmalloc_node(sizeof(*dd), GFP_KERNEL, q->node); | ||
627 | if (!dd) | ||
628 | return -ENOMEM; | ||
629 | memset(dd, 0, sizeof(*dd)); | ||
630 | |||
631 | dd->hash = kmalloc_node(sizeof(struct list_head)*DL_HASH_ENTRIES, | ||
632 | GFP_KERNEL, q->node); | ||
633 | if (!dd->hash) { | ||
634 | kfree(dd); | ||
635 | return -ENOMEM; | ||
636 | } | ||
637 | |||
638 | dd->drq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, | ||
639 | mempool_free_slab, drq_pool, q->node); | ||
640 | if (!dd->drq_pool) { | ||
641 | kfree(dd->hash); | ||
642 | kfree(dd); | ||
643 | return -ENOMEM; | ||
644 | } | ||
645 | |||
646 | for (i = 0; i < DL_HASH_ENTRIES; i++) | ||
647 | INIT_LIST_HEAD(&dd->hash[i]); | ||
648 | |||
649 | INIT_LIST_HEAD(&dd->fifo_list[READ]); | ||
650 | INIT_LIST_HEAD(&dd->fifo_list[WRITE]); | ||
651 | dd->sort_list[READ] = RB_ROOT; | ||
652 | dd->sort_list[WRITE] = RB_ROOT; | ||
653 | dd->fifo_expire[READ] = read_expire; | ||
654 | dd->fifo_expire[WRITE] = write_expire; | ||
655 | dd->writes_starved = writes_starved; | ||
656 | dd->front_merges = 1; | ||
657 | dd->fifo_batch = fifo_batch; | ||
658 | e->elevator_data = dd; | ||
659 | return 0; | ||
660 | } | ||
661 | |||
662 | static void deadline_put_request(request_queue_t *q, struct request *rq) | ||
663 | { | ||
664 | struct deadline_data *dd = q->elevator->elevator_data; | ||
665 | struct deadline_rq *drq = RQ_DATA(rq); | ||
666 | |||
667 | mempool_free(drq, dd->drq_pool); | ||
668 | rq->elevator_private = NULL; | ||
669 | } | ||
670 | |||
671 | static int | ||
672 | deadline_set_request(request_queue_t *q, struct request *rq, struct bio *bio, | ||
673 | gfp_t gfp_mask) | ||
674 | { | ||
675 | struct deadline_data *dd = q->elevator->elevator_data; | ||
676 | struct deadline_rq *drq; | ||
677 | |||
678 | drq = mempool_alloc(dd->drq_pool, gfp_mask); | ||
679 | if (drq) { | ||
680 | memset(drq, 0, sizeof(*drq)); | ||
681 | RB_CLEAR(&drq->rb_node); | ||
682 | drq->request = rq; | ||
683 | |||
684 | INIT_LIST_HEAD(&drq->hash); | ||
685 | drq->on_hash = 0; | ||
686 | |||
687 | INIT_LIST_HEAD(&drq->fifo); | ||
688 | |||
689 | rq->elevator_private = drq; | ||
690 | return 0; | ||
691 | } | ||
692 | |||
693 | return 1; | ||
694 | } | ||
695 | |||
696 | /* | ||
697 | * sysfs parts below | ||
698 | */ | ||
699 | struct deadline_fs_entry { | ||
700 | struct attribute attr; | ||
701 | ssize_t (*show)(struct deadline_data *, char *); | ||
702 | ssize_t (*store)(struct deadline_data *, const char *, size_t); | ||
703 | }; | ||
704 | |||
705 | static ssize_t | ||
706 | deadline_var_show(int var, char *page) | ||
707 | { | ||
708 | return sprintf(page, "%d\n", var); | ||
709 | } | ||
710 | |||
711 | static ssize_t | ||
712 | deadline_var_store(int *var, const char *page, size_t count) | ||
713 | { | ||
714 | char *p = (char *) page; | ||
715 | |||
716 | *var = simple_strtol(p, &p, 10); | ||
717 | return count; | ||
718 | } | ||
719 | |||
720 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ | ||
721 | static ssize_t __FUNC(struct deadline_data *dd, char *page) \ | ||
722 | { \ | ||
723 | int __data = __VAR; \ | ||
724 | if (__CONV) \ | ||
725 | __data = jiffies_to_msecs(__data); \ | ||
726 | return deadline_var_show(__data, (page)); \ | ||
727 | } | ||
728 | SHOW_FUNCTION(deadline_readexpire_show, dd->fifo_expire[READ], 1); | ||
729 | SHOW_FUNCTION(deadline_writeexpire_show, dd->fifo_expire[WRITE], 1); | ||
730 | SHOW_FUNCTION(deadline_writesstarved_show, dd->writes_starved, 0); | ||
731 | SHOW_FUNCTION(deadline_frontmerges_show, dd->front_merges, 0); | ||
732 | SHOW_FUNCTION(deadline_fifobatch_show, dd->fifo_batch, 0); | ||
733 | #undef SHOW_FUNCTION | ||
734 | |||
735 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | ||
736 | static ssize_t __FUNC(struct deadline_data *dd, const char *page, size_t count) \ | ||
737 | { \ | ||
738 | int __data; \ | ||
739 | int ret = deadline_var_store(&__data, (page), count); \ | ||
740 | if (__data < (MIN)) \ | ||
741 | __data = (MIN); \ | ||
742 | else if (__data > (MAX)) \ | ||
743 | __data = (MAX); \ | ||
744 | if (__CONV) \ | ||
745 | *(__PTR) = msecs_to_jiffies(__data); \ | ||
746 | else \ | ||
747 | *(__PTR) = __data; \ | ||
748 | return ret; \ | ||
749 | } | ||
750 | STORE_FUNCTION(deadline_readexpire_store, &dd->fifo_expire[READ], 0, INT_MAX, 1); | ||
751 | STORE_FUNCTION(deadline_writeexpire_store, &dd->fifo_expire[WRITE], 0, INT_MAX, 1); | ||
752 | STORE_FUNCTION(deadline_writesstarved_store, &dd->writes_starved, INT_MIN, INT_MAX, 0); | ||
753 | STORE_FUNCTION(deadline_frontmerges_store, &dd->front_merges, 0, 1, 0); | ||
754 | STORE_FUNCTION(deadline_fifobatch_store, &dd->fifo_batch, 0, INT_MAX, 0); | ||
755 | #undef STORE_FUNCTION | ||
756 | |||
757 | static struct deadline_fs_entry deadline_readexpire_entry = { | ||
758 | .attr = {.name = "read_expire", .mode = S_IRUGO | S_IWUSR }, | ||
759 | .show = deadline_readexpire_show, | ||
760 | .store = deadline_readexpire_store, | ||
761 | }; | ||
762 | static struct deadline_fs_entry deadline_writeexpire_entry = { | ||
763 | .attr = {.name = "write_expire", .mode = S_IRUGO | S_IWUSR }, | ||
764 | .show = deadline_writeexpire_show, | ||
765 | .store = deadline_writeexpire_store, | ||
766 | }; | ||
767 | static struct deadline_fs_entry deadline_writesstarved_entry = { | ||
768 | .attr = {.name = "writes_starved", .mode = S_IRUGO | S_IWUSR }, | ||
769 | .show = deadline_writesstarved_show, | ||
770 | .store = deadline_writesstarved_store, | ||
771 | }; | ||
772 | static struct deadline_fs_entry deadline_frontmerges_entry = { | ||
773 | .attr = {.name = "front_merges", .mode = S_IRUGO | S_IWUSR }, | ||
774 | .show = deadline_frontmerges_show, | ||
775 | .store = deadline_frontmerges_store, | ||
776 | }; | ||
777 | static struct deadline_fs_entry deadline_fifobatch_entry = { | ||
778 | .attr = {.name = "fifo_batch", .mode = S_IRUGO | S_IWUSR }, | ||
779 | .show = deadline_fifobatch_show, | ||
780 | .store = deadline_fifobatch_store, | ||
781 | }; | ||
782 | |||
783 | static struct attribute *default_attrs[] = { | ||
784 | &deadline_readexpire_entry.attr, | ||
785 | &deadline_writeexpire_entry.attr, | ||
786 | &deadline_writesstarved_entry.attr, | ||
787 | &deadline_frontmerges_entry.attr, | ||
788 | &deadline_fifobatch_entry.attr, | ||
789 | NULL, | ||
790 | }; | ||
791 | |||
792 | #define to_deadline(atr) container_of((atr), struct deadline_fs_entry, attr) | ||
793 | |||
794 | static ssize_t | ||
795 | deadline_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | ||
796 | { | ||
797 | elevator_t *e = container_of(kobj, elevator_t, kobj); | ||
798 | struct deadline_fs_entry *entry = to_deadline(attr); | ||
799 | |||
800 | if (!entry->show) | ||
801 | return -EIO; | ||
802 | |||
803 | return entry->show(e->elevator_data, page); | ||
804 | } | ||
805 | |||
806 | static ssize_t | ||
807 | deadline_attr_store(struct kobject *kobj, struct attribute *attr, | ||
808 | const char *page, size_t length) | ||
809 | { | ||
810 | elevator_t *e = container_of(kobj, elevator_t, kobj); | ||
811 | struct deadline_fs_entry *entry = to_deadline(attr); | ||
812 | |||
813 | if (!entry->store) | ||
814 | return -EIO; | ||
815 | |||
816 | return entry->store(e->elevator_data, page, length); | ||
817 | } | ||
818 | |||
819 | static struct sysfs_ops deadline_sysfs_ops = { | ||
820 | .show = deadline_attr_show, | ||
821 | .store = deadline_attr_store, | ||
822 | }; | ||
823 | |||
824 | static struct kobj_type deadline_ktype = { | ||
825 | .sysfs_ops = &deadline_sysfs_ops, | ||
826 | .default_attrs = default_attrs, | ||
827 | }; | ||
828 | |||
829 | static struct elevator_type iosched_deadline = { | ||
830 | .ops = { | ||
831 | .elevator_merge_fn = deadline_merge, | ||
832 | .elevator_merged_fn = deadline_merged_request, | ||
833 | .elevator_merge_req_fn = deadline_merged_requests, | ||
834 | .elevator_dispatch_fn = deadline_dispatch_requests, | ||
835 | .elevator_add_req_fn = deadline_add_request, | ||
836 | .elevator_queue_empty_fn = deadline_queue_empty, | ||
837 | .elevator_former_req_fn = deadline_former_request, | ||
838 | .elevator_latter_req_fn = deadline_latter_request, | ||
839 | .elevator_set_req_fn = deadline_set_request, | ||
840 | .elevator_put_req_fn = deadline_put_request, | ||
841 | .elevator_init_fn = deadline_init_queue, | ||
842 | .elevator_exit_fn = deadline_exit_queue, | ||
843 | }, | ||
844 | |||
845 | .elevator_ktype = &deadline_ktype, | ||
846 | .elevator_name = "deadline", | ||
847 | .elevator_owner = THIS_MODULE, | ||
848 | }; | ||
849 | |||
850 | static int __init deadline_init(void) | ||
851 | { | ||
852 | int ret; | ||
853 | |||
854 | drq_pool = kmem_cache_create("deadline_drq", sizeof(struct deadline_rq), | ||
855 | 0, 0, NULL, NULL); | ||
856 | |||
857 | if (!drq_pool) | ||
858 | return -ENOMEM; | ||
859 | |||
860 | ret = elv_register(&iosched_deadline); | ||
861 | if (ret) | ||
862 | kmem_cache_destroy(drq_pool); | ||
863 | |||
864 | return ret; | ||
865 | } | ||
866 | |||
867 | static void __exit deadline_exit(void) | ||
868 | { | ||
869 | kmem_cache_destroy(drq_pool); | ||
870 | elv_unregister(&iosched_deadline); | ||
871 | } | ||
872 | |||
873 | module_init(deadline_init); | ||
874 | module_exit(deadline_exit); | ||
875 | |||
876 | MODULE_AUTHOR("Jens Axboe"); | ||
877 | MODULE_LICENSE("GPL"); | ||
878 | MODULE_DESCRIPTION("deadline IO scheduler"); | ||