]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 *dispatch; /* driver dispatch queue */ | |
54 | struct list_head *hash; /* request hash */ | |
55 | unsigned int batching; /* number of sequential requests made */ | |
56 | sector_t last_sector; /* head position */ | |
57 | unsigned int starved; /* times reads have starved writes */ | |
58 | ||
59 | /* | |
60 | * settings that change how the i/o scheduler behaves | |
61 | */ | |
62 | int fifo_expire[2]; | |
63 | int fifo_batch; | |
64 | int writes_starved; | |
65 | int front_merges; | |
66 | ||
67 | mempool_t *drq_pool; | |
68 | }; | |
69 | ||
70 | /* | |
71 | * pre-request data. | |
72 | */ | |
73 | struct deadline_rq { | |
74 | /* | |
75 | * rbtree index, key is the starting offset | |
76 | */ | |
77 | struct rb_node rb_node; | |
78 | sector_t rb_key; | |
79 | ||
80 | struct request *request; | |
81 | ||
82 | /* | |
83 | * request hash, key is the ending offset (for back merge lookup) | |
84 | */ | |
85 | struct list_head hash; | |
86 | char on_hash; | |
87 | ||
88 | /* | |
89 | * expire fifo | |
90 | */ | |
91 | struct list_head fifo; | |
92 | unsigned long expires; | |
93 | }; | |
94 | ||
95 | static void deadline_move_request(struct deadline_data *dd, struct deadline_rq *drq); | |
96 | ||
97 | static kmem_cache_t *drq_pool; | |
98 | ||
99 | #define RQ_DATA(rq) ((struct deadline_rq *) (rq)->elevator_private) | |
100 | ||
101 | /* | |
102 | * the back merge hash support functions | |
103 | */ | |
104 | static inline void __deadline_del_drq_hash(struct deadline_rq *drq) | |
105 | { | |
106 | drq->on_hash = 0; | |
107 | list_del_init(&drq->hash); | |
108 | } | |
109 | ||
110 | static inline void deadline_del_drq_hash(struct deadline_rq *drq) | |
111 | { | |
112 | if (ON_HASH(drq)) | |
113 | __deadline_del_drq_hash(drq); | |
114 | } | |
115 | ||
116 | static void | |
117 | deadline_remove_merge_hints(request_queue_t *q, struct deadline_rq *drq) | |
118 | { | |
119 | deadline_del_drq_hash(drq); | |
120 | ||
121 | if (q->last_merge == drq->request) | |
122 | q->last_merge = NULL; | |
123 | } | |
124 | ||
125 | static inline void | |
126 | deadline_add_drq_hash(struct deadline_data *dd, struct deadline_rq *drq) | |
127 | { | |
128 | struct request *rq = drq->request; | |
129 | ||
130 | BUG_ON(ON_HASH(drq)); | |
131 | ||
132 | drq->on_hash = 1; | |
133 | list_add(&drq->hash, &dd->hash[DL_HASH_FN(rq_hash_key(rq))]); | |
134 | } | |
135 | ||
136 | /* | |
137 | * move hot entry to front of chain | |
138 | */ | |
139 | static inline void | |
140 | deadline_hot_drq_hash(struct deadline_data *dd, struct deadline_rq *drq) | |
141 | { | |
142 | struct request *rq = drq->request; | |
143 | struct list_head *head = &dd->hash[DL_HASH_FN(rq_hash_key(rq))]; | |
144 | ||
145 | if (ON_HASH(drq) && drq->hash.prev != head) { | |
146 | list_del(&drq->hash); | |
147 | list_add(&drq->hash, head); | |
148 | } | |
149 | } | |
150 | ||
151 | static struct request * | |
152 | deadline_find_drq_hash(struct deadline_data *dd, sector_t offset) | |
153 | { | |
154 | struct list_head *hash_list = &dd->hash[DL_HASH_FN(offset)]; | |
155 | struct list_head *entry, *next = hash_list->next; | |
156 | ||
157 | while ((entry = next) != hash_list) { | |
158 | struct deadline_rq *drq = list_entry_hash(entry); | |
159 | struct request *__rq = drq->request; | |
160 | ||
161 | next = entry->next; | |
162 | ||
163 | BUG_ON(!ON_HASH(drq)); | |
164 | ||
165 | if (!rq_mergeable(__rq)) { | |
166 | __deadline_del_drq_hash(drq); | |
167 | continue; | |
168 | } | |
169 | ||
170 | if (rq_hash_key(__rq) == offset) | |
171 | return __rq; | |
172 | } | |
173 | ||
174 | return NULL; | |
175 | } | |
176 | ||
177 | /* | |
178 | * rb tree support functions | |
179 | */ | |
180 | #define RB_NONE (2) | |
181 | #define RB_EMPTY(root) ((root)->rb_node == NULL) | |
182 | #define ON_RB(node) ((node)->rb_color != RB_NONE) | |
183 | #define RB_CLEAR(node) ((node)->rb_color = RB_NONE) | |
184 | #define rb_entry_drq(node) rb_entry((node), struct deadline_rq, rb_node) | |
185 | #define DRQ_RB_ROOT(dd, drq) (&(dd)->sort_list[rq_data_dir((drq)->request)]) | |
186 | #define rq_rb_key(rq) (rq)->sector | |
187 | ||
188 | static struct deadline_rq * | |
189 | __deadline_add_drq_rb(struct deadline_data *dd, struct deadline_rq *drq) | |
190 | { | |
191 | struct rb_node **p = &DRQ_RB_ROOT(dd, drq)->rb_node; | |
192 | struct rb_node *parent = NULL; | |
193 | struct deadline_rq *__drq; | |
194 | ||
195 | while (*p) { | |
196 | parent = *p; | |
197 | __drq = rb_entry_drq(parent); | |
198 | ||
199 | if (drq->rb_key < __drq->rb_key) | |
200 | p = &(*p)->rb_left; | |
201 | else if (drq->rb_key > __drq->rb_key) | |
202 | p = &(*p)->rb_right; | |
203 | else | |
204 | return __drq; | |
205 | } | |
206 | ||
207 | rb_link_node(&drq->rb_node, parent, p); | |
208 | return NULL; | |
209 | } | |
210 | ||
211 | static void | |
212 | deadline_add_drq_rb(struct deadline_data *dd, struct deadline_rq *drq) | |
213 | { | |
214 | struct deadline_rq *__alias; | |
215 | ||
216 | drq->rb_key = rq_rb_key(drq->request); | |
217 | ||
218 | retry: | |
219 | __alias = __deadline_add_drq_rb(dd, drq); | |
220 | if (!__alias) { | |
221 | rb_insert_color(&drq->rb_node, DRQ_RB_ROOT(dd, drq)); | |
222 | return; | |
223 | } | |
224 | ||
225 | deadline_move_request(dd, __alias); | |
226 | goto retry; | |
227 | } | |
228 | ||
229 | static inline void | |
230 | deadline_del_drq_rb(struct deadline_data *dd, struct deadline_rq *drq) | |
231 | { | |
232 | const int data_dir = rq_data_dir(drq->request); | |
233 | ||
234 | if (dd->next_drq[data_dir] == drq) { | |
235 | struct rb_node *rbnext = rb_next(&drq->rb_node); | |
236 | ||
237 | dd->next_drq[data_dir] = NULL; | |
238 | if (rbnext) | |
239 | dd->next_drq[data_dir] = rb_entry_drq(rbnext); | |
240 | } | |
241 | ||
242 | if (ON_RB(&drq->rb_node)) { | |
243 | rb_erase(&drq->rb_node, DRQ_RB_ROOT(dd, drq)); | |
244 | RB_CLEAR(&drq->rb_node); | |
245 | } | |
246 | } | |
247 | ||
248 | static struct request * | |
249 | deadline_find_drq_rb(struct deadline_data *dd, sector_t sector, int data_dir) | |
250 | { | |
251 | struct rb_node *n = dd->sort_list[data_dir].rb_node; | |
252 | struct deadline_rq *drq; | |
253 | ||
254 | while (n) { | |
255 | drq = rb_entry_drq(n); | |
256 | ||
257 | if (sector < drq->rb_key) | |
258 | n = n->rb_left; | |
259 | else if (sector > drq->rb_key) | |
260 | n = n->rb_right; | |
261 | else | |
262 | return drq->request; | |
263 | } | |
264 | ||
265 | return NULL; | |
266 | } | |
267 | ||
268 | /* | |
269 | * deadline_find_first_drq finds the first (lowest sector numbered) request | |
270 | * for the specified data_dir. Used to sweep back to the start of the disk | |
271 | * (1-way elevator) after we process the last (highest sector) request. | |
272 | */ | |
273 | static struct deadline_rq * | |
274 | deadline_find_first_drq(struct deadline_data *dd, int data_dir) | |
275 | { | |
276 | struct rb_node *n = dd->sort_list[data_dir].rb_node; | |
277 | ||
278 | for (;;) { | |
279 | if (n->rb_left == NULL) | |
280 | return rb_entry_drq(n); | |
281 | ||
282 | n = n->rb_left; | |
283 | } | |
284 | } | |
285 | ||
286 | /* | |
287 | * add drq to rbtree and fifo | |
288 | */ | |
289 | static inline void | |
290 | deadline_add_request(struct request_queue *q, struct request *rq) | |
291 | { | |
292 | struct deadline_data *dd = q->elevator->elevator_data; | |
293 | struct deadline_rq *drq = RQ_DATA(rq); | |
294 | ||
295 | const int data_dir = rq_data_dir(drq->request); | |
296 | ||
297 | deadline_add_drq_rb(dd, drq); | |
298 | /* | |
299 | * set expire time (only used for reads) and add to fifo list | |
300 | */ | |
301 | drq->expires = jiffies + dd->fifo_expire[data_dir]; | |
302 | list_add_tail(&drq->fifo, &dd->fifo_list[data_dir]); | |
303 | ||
304 | if (rq_mergeable(rq)) { | |
305 | deadline_add_drq_hash(dd, drq); | |
306 | ||
307 | if (!q->last_merge) | |
308 | q->last_merge = rq; | |
309 | } | |
310 | } | |
311 | ||
312 | /* | |
313 | * remove rq from rbtree, fifo, and hash | |
314 | */ | |
315 | static void deadline_remove_request(request_queue_t *q, struct request *rq) | |
316 | { | |
317 | struct deadline_rq *drq = RQ_DATA(rq); | |
318 | ||
319 | if (drq) { | |
320 | struct deadline_data *dd = q->elevator->elevator_data; | |
321 | ||
322 | list_del_init(&drq->fifo); | |
323 | deadline_remove_merge_hints(q, drq); | |
324 | deadline_del_drq_rb(dd, drq); | |
325 | } | |
326 | } | |
327 | ||
328 | static int | |
329 | deadline_merge(request_queue_t *q, struct request **req, struct bio *bio) | |
330 | { | |
331 | struct deadline_data *dd = q->elevator->elevator_data; | |
332 | struct request *__rq; | |
333 | int ret; | |
334 | ||
335 | /* | |
336 | * try last_merge to avoid going to hash | |
337 | */ | |
338 | ret = elv_try_last_merge(q, bio); | |
339 | if (ret != ELEVATOR_NO_MERGE) { | |
340 | __rq = q->last_merge; | |
341 | goto out_insert; | |
342 | } | |
343 | ||
344 | /* | |
345 | * see if the merge hash can satisfy a back merge | |
346 | */ | |
347 | __rq = deadline_find_drq_hash(dd, bio->bi_sector); | |
348 | if (__rq) { | |
349 | BUG_ON(__rq->sector + __rq->nr_sectors != bio->bi_sector); | |
350 | ||
351 | if (elv_rq_merge_ok(__rq, bio)) { | |
352 | ret = ELEVATOR_BACK_MERGE; | |
353 | goto out; | |
354 | } | |
355 | } | |
356 | ||
357 | /* | |
358 | * check for front merge | |
359 | */ | |
360 | if (dd->front_merges) { | |
361 | sector_t rb_key = bio->bi_sector + bio_sectors(bio); | |
362 | ||
363 | __rq = deadline_find_drq_rb(dd, rb_key, bio_data_dir(bio)); | |
364 | if (__rq) { | |
365 | BUG_ON(rb_key != rq_rb_key(__rq)); | |
366 | ||
367 | if (elv_rq_merge_ok(__rq, bio)) { | |
368 | ret = ELEVATOR_FRONT_MERGE; | |
369 | goto out; | |
370 | } | |
371 | } | |
372 | } | |
373 | ||
374 | return ELEVATOR_NO_MERGE; | |
375 | out: | |
376 | q->last_merge = __rq; | |
377 | out_insert: | |
378 | if (ret) | |
379 | deadline_hot_drq_hash(dd, RQ_DATA(__rq)); | |
380 | *req = __rq; | |
381 | return ret; | |
382 | } | |
383 | ||
384 | static void deadline_merged_request(request_queue_t *q, struct request *req) | |
385 | { | |
386 | struct deadline_data *dd = q->elevator->elevator_data; | |
387 | struct deadline_rq *drq = RQ_DATA(req); | |
388 | ||
389 | /* | |
390 | * hash always needs to be repositioned, key is end sector | |
391 | */ | |
392 | deadline_del_drq_hash(drq); | |
393 | deadline_add_drq_hash(dd, drq); | |
394 | ||
395 | /* | |
396 | * if the merge was a front merge, we need to reposition request | |
397 | */ | |
398 | if (rq_rb_key(req) != drq->rb_key) { | |
399 | deadline_del_drq_rb(dd, drq); | |
400 | deadline_add_drq_rb(dd, drq); | |
401 | } | |
402 | ||
403 | q->last_merge = req; | |
404 | } | |
405 | ||
406 | static void | |
407 | deadline_merged_requests(request_queue_t *q, struct request *req, | |
408 | struct request *next) | |
409 | { | |
410 | struct deadline_data *dd = q->elevator->elevator_data; | |
411 | struct deadline_rq *drq = RQ_DATA(req); | |
412 | struct deadline_rq *dnext = RQ_DATA(next); | |
413 | ||
414 | BUG_ON(!drq); | |
415 | BUG_ON(!dnext); | |
416 | ||
417 | /* | |
418 | * reposition drq (this is the merged request) in hash, and in rbtree | |
419 | * in case of a front merge | |
420 | */ | |
421 | deadline_del_drq_hash(drq); | |
422 | deadline_add_drq_hash(dd, drq); | |
423 | ||
424 | if (rq_rb_key(req) != drq->rb_key) { | |
425 | deadline_del_drq_rb(dd, drq); | |
426 | deadline_add_drq_rb(dd, drq); | |
427 | } | |
428 | ||
429 | /* | |
430 | * if dnext expires before drq, assign its expire time to drq | |
431 | * and move into dnext position (dnext will be deleted) in fifo | |
432 | */ | |
433 | if (!list_empty(&drq->fifo) && !list_empty(&dnext->fifo)) { | |
434 | if (time_before(dnext->expires, drq->expires)) { | |
435 | list_move(&drq->fifo, &dnext->fifo); | |
436 | drq->expires = dnext->expires; | |
437 | } | |
438 | } | |
439 | ||
440 | /* | |
441 | * kill knowledge of next, this one is a goner | |
442 | */ | |
443 | deadline_remove_request(q, next); | |
444 | } | |
445 | ||
446 | /* | |
447 | * move request from sort list to dispatch queue. | |
448 | */ | |
449 | static inline void | |
450 | deadline_move_to_dispatch(struct deadline_data *dd, struct deadline_rq *drq) | |
451 | { | |
452 | request_queue_t *q = drq->request->q; | |
453 | ||
454 | deadline_remove_request(q, drq->request); | |
455 | list_add_tail(&drq->request->queuelist, dd->dispatch); | |
456 | } | |
457 | ||
458 | /* | |
459 | * move an entry to dispatch queue | |
460 | */ | |
461 | static void | |
462 | deadline_move_request(struct deadline_data *dd, struct deadline_rq *drq) | |
463 | { | |
464 | const int data_dir = rq_data_dir(drq->request); | |
465 | struct rb_node *rbnext = rb_next(&drq->rb_node); | |
466 | ||
467 | dd->next_drq[READ] = NULL; | |
468 | dd->next_drq[WRITE] = NULL; | |
469 | ||
470 | if (rbnext) | |
471 | dd->next_drq[data_dir] = rb_entry_drq(rbnext); | |
472 | ||
473 | dd->last_sector = drq->request->sector + drq->request->nr_sectors; | |
474 | ||
475 | /* | |
476 | * take it off the sort and fifo list, move | |
477 | * to dispatch queue | |
478 | */ | |
479 | deadline_move_to_dispatch(dd, drq); | |
480 | } | |
481 | ||
482 | #define list_entry_fifo(ptr) list_entry((ptr), struct deadline_rq, fifo) | |
483 | ||
484 | /* | |
485 | * deadline_check_fifo returns 0 if there are no expired reads on the fifo, | |
486 | * 1 otherwise. Requires !list_empty(&dd->fifo_list[data_dir]) | |
487 | */ | |
488 | static inline int deadline_check_fifo(struct deadline_data *dd, int ddir) | |
489 | { | |
490 | struct deadline_rq *drq = list_entry_fifo(dd->fifo_list[ddir].next); | |
491 | ||
492 | /* | |
493 | * drq is expired! | |
494 | */ | |
495 | if (time_after(jiffies, drq->expires)) | |
496 | return 1; | |
497 | ||
498 | return 0; | |
499 | } | |
500 | ||
501 | /* | |
502 | * deadline_dispatch_requests selects the best request according to | |
503 | * read/write expire, fifo_batch, etc | |
504 | */ | |
505 | static int deadline_dispatch_requests(struct deadline_data *dd) | |
506 | { | |
507 | const int reads = !list_empty(&dd->fifo_list[READ]); | |
508 | const int writes = !list_empty(&dd->fifo_list[WRITE]); | |
509 | struct deadline_rq *drq; | |
510 | int data_dir, other_dir; | |
511 | ||
512 | /* | |
513 | * batches are currently reads XOR writes | |
514 | */ | |
515 | drq = NULL; | |
516 | ||
517 | if (dd->next_drq[READ]) | |
518 | drq = dd->next_drq[READ]; | |
519 | ||
520 | if (dd->next_drq[WRITE]) | |
521 | drq = dd->next_drq[WRITE]; | |
522 | ||
523 | if (drq) { | |
524 | /* we have a "next request" */ | |
525 | ||
526 | if (dd->last_sector != drq->request->sector) | |
527 | /* end the batch on a non sequential request */ | |
528 | dd->batching += dd->fifo_batch; | |
529 | ||
530 | if (dd->batching < dd->fifo_batch) | |
531 | /* we are still entitled to batch */ | |
532 | goto dispatch_request; | |
533 | } | |
534 | ||
535 | /* | |
536 | * at this point we are not running a batch. select the appropriate | |
537 | * data direction (read / write) | |
538 | */ | |
539 | ||
540 | if (reads) { | |
541 | BUG_ON(RB_EMPTY(&dd->sort_list[READ])); | |
542 | ||
543 | if (writes && (dd->starved++ >= dd->writes_starved)) | |
544 | goto dispatch_writes; | |
545 | ||
546 | data_dir = READ; | |
547 | other_dir = WRITE; | |
548 | ||
549 | goto dispatch_find_request; | |
550 | } | |
551 | ||
552 | /* | |
553 | * there are either no reads or writes have been starved | |
554 | */ | |
555 | ||
556 | if (writes) { | |
557 | dispatch_writes: | |
558 | BUG_ON(RB_EMPTY(&dd->sort_list[WRITE])); | |
559 | ||
560 | dd->starved = 0; | |
561 | ||
562 | data_dir = WRITE; | |
563 | other_dir = READ; | |
564 | ||
565 | goto dispatch_find_request; | |
566 | } | |
567 | ||
568 | return 0; | |
569 | ||
570 | dispatch_find_request: | |
571 | /* | |
572 | * we are not running a batch, find best request for selected data_dir | |
573 | */ | |
574 | if (deadline_check_fifo(dd, data_dir)) { | |
575 | /* An expired request exists - satisfy it */ | |
576 | dd->batching = 0; | |
577 | drq = list_entry_fifo(dd->fifo_list[data_dir].next); | |
578 | ||
579 | } else if (dd->next_drq[data_dir]) { | |
580 | /* | |
581 | * The last req was the same dir and we have a next request in | |
582 | * sort order. No expired requests so continue on from here. | |
583 | */ | |
584 | drq = dd->next_drq[data_dir]; | |
585 | } else { | |
586 | /* | |
587 | * The last req was the other direction or we have run out of | |
588 | * higher-sectored requests. Go back to the lowest sectored | |
589 | * request (1 way elevator) and start a new batch. | |
590 | */ | |
591 | dd->batching = 0; | |
592 | drq = deadline_find_first_drq(dd, data_dir); | |
593 | } | |
594 | ||
595 | dispatch_request: | |
596 | /* | |
597 | * drq is the selected appropriate request. | |
598 | */ | |
599 | dd->batching++; | |
600 | deadline_move_request(dd, drq); | |
601 | ||
602 | return 1; | |
603 | } | |
604 | ||
605 | static struct request *deadline_next_request(request_queue_t *q) | |
606 | { | |
607 | struct deadline_data *dd = q->elevator->elevator_data; | |
608 | struct request *rq; | |
609 | ||
610 | /* | |
611 | * if there are still requests on the dispatch queue, grab the first one | |
612 | */ | |
613 | if (!list_empty(dd->dispatch)) { | |
614 | dispatch: | |
615 | rq = list_entry_rq(dd->dispatch->next); | |
616 | return rq; | |
617 | } | |
618 | ||
619 | if (deadline_dispatch_requests(dd)) | |
620 | goto dispatch; | |
621 | ||
622 | return NULL; | |
623 | } | |
624 | ||
625 | static void | |
626 | deadline_insert_request(request_queue_t *q, struct request *rq, int where) | |
627 | { | |
628 | struct deadline_data *dd = q->elevator->elevator_data; | |
629 | ||
630 | /* barriers must flush the reorder queue */ | |
631 | if (unlikely(rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER) | |
632 | && where == ELEVATOR_INSERT_SORT)) | |
633 | where = ELEVATOR_INSERT_BACK; | |
634 | ||
635 | switch (where) { | |
636 | case ELEVATOR_INSERT_BACK: | |
637 | while (deadline_dispatch_requests(dd)) | |
638 | ; | |
639 | list_add_tail(&rq->queuelist, dd->dispatch); | |
640 | break; | |
641 | case ELEVATOR_INSERT_FRONT: | |
642 | list_add(&rq->queuelist, dd->dispatch); | |
643 | break; | |
644 | case ELEVATOR_INSERT_SORT: | |
645 | BUG_ON(!blk_fs_request(rq)); | |
646 | deadline_add_request(q, rq); | |
647 | break; | |
648 | default: | |
649 | printk("%s: bad insert point %d\n", __FUNCTION__,where); | |
650 | return; | |
651 | } | |
652 | } | |
653 | ||
654 | static int deadline_queue_empty(request_queue_t *q) | |
655 | { | |
656 | struct deadline_data *dd = q->elevator->elevator_data; | |
657 | ||
658 | if (!list_empty(&dd->fifo_list[WRITE]) | |
659 | || !list_empty(&dd->fifo_list[READ]) | |
660 | || !list_empty(dd->dispatch)) | |
661 | return 0; | |
662 | ||
663 | return 1; | |
664 | } | |
665 | ||
666 | static struct request * | |
667 | deadline_former_request(request_queue_t *q, struct request *rq) | |
668 | { | |
669 | struct deadline_rq *drq = RQ_DATA(rq); | |
670 | struct rb_node *rbprev = rb_prev(&drq->rb_node); | |
671 | ||
672 | if (rbprev) | |
673 | return rb_entry_drq(rbprev)->request; | |
674 | ||
675 | return NULL; | |
676 | } | |
677 | ||
678 | static struct request * | |
679 | deadline_latter_request(request_queue_t *q, struct request *rq) | |
680 | { | |
681 | struct deadline_rq *drq = RQ_DATA(rq); | |
682 | struct rb_node *rbnext = rb_next(&drq->rb_node); | |
683 | ||
684 | if (rbnext) | |
685 | return rb_entry_drq(rbnext)->request; | |
686 | ||
687 | return NULL; | |
688 | } | |
689 | ||
690 | static void deadline_exit_queue(elevator_t *e) | |
691 | { | |
692 | struct deadline_data *dd = e->elevator_data; | |
693 | ||
694 | BUG_ON(!list_empty(&dd->fifo_list[READ])); | |
695 | BUG_ON(!list_empty(&dd->fifo_list[WRITE])); | |
696 | ||
697 | mempool_destroy(dd->drq_pool); | |
698 | kfree(dd->hash); | |
699 | kfree(dd); | |
700 | } | |
701 | ||
702 | /* | |
703 | * initialize elevator private data (deadline_data), and alloc a drq for | |
704 | * each request on the free lists | |
705 | */ | |
706 | static int deadline_init_queue(request_queue_t *q, elevator_t *e) | |
707 | { | |
708 | struct deadline_data *dd; | |
709 | int i; | |
710 | ||
711 | if (!drq_pool) | |
712 | return -ENOMEM; | |
713 | ||
1946089a | 714 | dd = kmalloc_node(sizeof(*dd), GFP_KERNEL, q->node); |
1da177e4 LT |
715 | if (!dd) |
716 | return -ENOMEM; | |
717 | memset(dd, 0, sizeof(*dd)); | |
718 | ||
1946089a CL |
719 | dd->hash = kmalloc_node(sizeof(struct list_head)*DL_HASH_ENTRIES, |
720 | GFP_KERNEL, q->node); | |
1da177e4 LT |
721 | if (!dd->hash) { |
722 | kfree(dd); | |
723 | return -ENOMEM; | |
724 | } | |
725 | ||
1946089a CL |
726 | dd->drq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
727 | mempool_free_slab, drq_pool, q->node); | |
1da177e4 LT |
728 | if (!dd->drq_pool) { |
729 | kfree(dd->hash); | |
730 | kfree(dd); | |
731 | return -ENOMEM; | |
732 | } | |
733 | ||
734 | for (i = 0; i < DL_HASH_ENTRIES; i++) | |
735 | INIT_LIST_HEAD(&dd->hash[i]); | |
736 | ||
737 | INIT_LIST_HEAD(&dd->fifo_list[READ]); | |
738 | INIT_LIST_HEAD(&dd->fifo_list[WRITE]); | |
739 | dd->sort_list[READ] = RB_ROOT; | |
740 | dd->sort_list[WRITE] = RB_ROOT; | |
741 | dd->dispatch = &q->queue_head; | |
742 | dd->fifo_expire[READ] = read_expire; | |
743 | dd->fifo_expire[WRITE] = write_expire; | |
744 | dd->writes_starved = writes_starved; | |
745 | dd->front_merges = 1; | |
746 | dd->fifo_batch = fifo_batch; | |
747 | e->elevator_data = dd; | |
748 | return 0; | |
749 | } | |
750 | ||
751 | static void deadline_put_request(request_queue_t *q, struct request *rq) | |
752 | { | |
753 | struct deadline_data *dd = q->elevator->elevator_data; | |
754 | struct deadline_rq *drq = RQ_DATA(rq); | |
755 | ||
756 | if (drq) { | |
757 | mempool_free(drq, dd->drq_pool); | |
758 | rq->elevator_private = NULL; | |
759 | } | |
760 | } | |
761 | ||
762 | static int | |
763 | deadline_set_request(request_queue_t *q, struct request *rq, int gfp_mask) | |
764 | { | |
765 | struct deadline_data *dd = q->elevator->elevator_data; | |
766 | struct deadline_rq *drq; | |
767 | ||
768 | drq = mempool_alloc(dd->drq_pool, gfp_mask); | |
769 | if (drq) { | |
770 | memset(drq, 0, sizeof(*drq)); | |
771 | RB_CLEAR(&drq->rb_node); | |
772 | drq->request = rq; | |
773 | ||
774 | INIT_LIST_HEAD(&drq->hash); | |
775 | drq->on_hash = 0; | |
776 | ||
777 | INIT_LIST_HEAD(&drq->fifo); | |
778 | ||
779 | rq->elevator_private = drq; | |
780 | return 0; | |
781 | } | |
782 | ||
783 | return 1; | |
784 | } | |
785 | ||
786 | /* | |
787 | * sysfs parts below | |
788 | */ | |
789 | struct deadline_fs_entry { | |
790 | struct attribute attr; | |
791 | ssize_t (*show)(struct deadline_data *, char *); | |
792 | ssize_t (*store)(struct deadline_data *, const char *, size_t); | |
793 | }; | |
794 | ||
795 | static ssize_t | |
796 | deadline_var_show(int var, char *page) | |
797 | { | |
798 | return sprintf(page, "%d\n", var); | |
799 | } | |
800 | ||
801 | static ssize_t | |
802 | deadline_var_store(int *var, const char *page, size_t count) | |
803 | { | |
804 | char *p = (char *) page; | |
805 | ||
806 | *var = simple_strtol(p, &p, 10); | |
807 | return count; | |
808 | } | |
809 | ||
810 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ | |
811 | static ssize_t __FUNC(struct deadline_data *dd, char *page) \ | |
812 | { \ | |
813 | int __data = __VAR; \ | |
814 | if (__CONV) \ | |
815 | __data = jiffies_to_msecs(__data); \ | |
816 | return deadline_var_show(__data, (page)); \ | |
817 | } | |
818 | SHOW_FUNCTION(deadline_readexpire_show, dd->fifo_expire[READ], 1); | |
819 | SHOW_FUNCTION(deadline_writeexpire_show, dd->fifo_expire[WRITE], 1); | |
820 | SHOW_FUNCTION(deadline_writesstarved_show, dd->writes_starved, 0); | |
821 | SHOW_FUNCTION(deadline_frontmerges_show, dd->front_merges, 0); | |
822 | SHOW_FUNCTION(deadline_fifobatch_show, dd->fifo_batch, 0); | |
823 | #undef SHOW_FUNCTION | |
824 | ||
825 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
826 | static ssize_t __FUNC(struct deadline_data *dd, const char *page, size_t count) \ | |
827 | { \ | |
828 | int __data; \ | |
829 | int ret = deadline_var_store(&__data, (page), count); \ | |
830 | if (__data < (MIN)) \ | |
831 | __data = (MIN); \ | |
832 | else if (__data > (MAX)) \ | |
833 | __data = (MAX); \ | |
834 | if (__CONV) \ | |
835 | *(__PTR) = msecs_to_jiffies(__data); \ | |
836 | else \ | |
837 | *(__PTR) = __data; \ | |
838 | return ret; \ | |
839 | } | |
840 | STORE_FUNCTION(deadline_readexpire_store, &dd->fifo_expire[READ], 0, INT_MAX, 1); | |
841 | STORE_FUNCTION(deadline_writeexpire_store, &dd->fifo_expire[WRITE], 0, INT_MAX, 1); | |
842 | STORE_FUNCTION(deadline_writesstarved_store, &dd->writes_starved, INT_MIN, INT_MAX, 0); | |
843 | STORE_FUNCTION(deadline_frontmerges_store, &dd->front_merges, 0, 1, 0); | |
844 | STORE_FUNCTION(deadline_fifobatch_store, &dd->fifo_batch, 0, INT_MAX, 0); | |
845 | #undef STORE_FUNCTION | |
846 | ||
847 | static struct deadline_fs_entry deadline_readexpire_entry = { | |
848 | .attr = {.name = "read_expire", .mode = S_IRUGO | S_IWUSR }, | |
849 | .show = deadline_readexpire_show, | |
850 | .store = deadline_readexpire_store, | |
851 | }; | |
852 | static struct deadline_fs_entry deadline_writeexpire_entry = { | |
853 | .attr = {.name = "write_expire", .mode = S_IRUGO | S_IWUSR }, | |
854 | .show = deadline_writeexpire_show, | |
855 | .store = deadline_writeexpire_store, | |
856 | }; | |
857 | static struct deadline_fs_entry deadline_writesstarved_entry = { | |
858 | .attr = {.name = "writes_starved", .mode = S_IRUGO | S_IWUSR }, | |
859 | .show = deadline_writesstarved_show, | |
860 | .store = deadline_writesstarved_store, | |
861 | }; | |
862 | static struct deadline_fs_entry deadline_frontmerges_entry = { | |
863 | .attr = {.name = "front_merges", .mode = S_IRUGO | S_IWUSR }, | |
864 | .show = deadline_frontmerges_show, | |
865 | .store = deadline_frontmerges_store, | |
866 | }; | |
867 | static struct deadline_fs_entry deadline_fifobatch_entry = { | |
868 | .attr = {.name = "fifo_batch", .mode = S_IRUGO | S_IWUSR }, | |
869 | .show = deadline_fifobatch_show, | |
870 | .store = deadline_fifobatch_store, | |
871 | }; | |
872 | ||
873 | static struct attribute *default_attrs[] = { | |
874 | &deadline_readexpire_entry.attr, | |
875 | &deadline_writeexpire_entry.attr, | |
876 | &deadline_writesstarved_entry.attr, | |
877 | &deadline_frontmerges_entry.attr, | |
878 | &deadline_fifobatch_entry.attr, | |
879 | NULL, | |
880 | }; | |
881 | ||
882 | #define to_deadline(atr) container_of((atr), struct deadline_fs_entry, attr) | |
883 | ||
884 | static ssize_t | |
885 | deadline_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
886 | { | |
887 | elevator_t *e = container_of(kobj, elevator_t, kobj); | |
888 | struct deadline_fs_entry *entry = to_deadline(attr); | |
889 | ||
890 | if (!entry->show) | |
6c1852a0 | 891 | return -EIO; |
1da177e4 LT |
892 | |
893 | return entry->show(e->elevator_data, page); | |
894 | } | |
895 | ||
896 | static ssize_t | |
897 | deadline_attr_store(struct kobject *kobj, struct attribute *attr, | |
898 | const char *page, size_t length) | |
899 | { | |
900 | elevator_t *e = container_of(kobj, elevator_t, kobj); | |
901 | struct deadline_fs_entry *entry = to_deadline(attr); | |
902 | ||
903 | if (!entry->store) | |
6c1852a0 | 904 | return -EIO; |
1da177e4 LT |
905 | |
906 | return entry->store(e->elevator_data, page, length); | |
907 | } | |
908 | ||
909 | static struct sysfs_ops deadline_sysfs_ops = { | |
910 | .show = deadline_attr_show, | |
911 | .store = deadline_attr_store, | |
912 | }; | |
913 | ||
914 | static struct kobj_type deadline_ktype = { | |
915 | .sysfs_ops = &deadline_sysfs_ops, | |
916 | .default_attrs = default_attrs, | |
917 | }; | |
918 | ||
919 | static struct elevator_type iosched_deadline = { | |
920 | .ops = { | |
921 | .elevator_merge_fn = deadline_merge, | |
922 | .elevator_merged_fn = deadline_merged_request, | |
923 | .elevator_merge_req_fn = deadline_merged_requests, | |
924 | .elevator_next_req_fn = deadline_next_request, | |
925 | .elevator_add_req_fn = deadline_insert_request, | |
926 | .elevator_remove_req_fn = deadline_remove_request, | |
927 | .elevator_queue_empty_fn = deadline_queue_empty, | |
928 | .elevator_former_req_fn = deadline_former_request, | |
929 | .elevator_latter_req_fn = deadline_latter_request, | |
930 | .elevator_set_req_fn = deadline_set_request, | |
931 | .elevator_put_req_fn = deadline_put_request, | |
932 | .elevator_init_fn = deadline_init_queue, | |
933 | .elevator_exit_fn = deadline_exit_queue, | |
934 | }, | |
935 | ||
936 | .elevator_ktype = &deadline_ktype, | |
937 | .elevator_name = "deadline", | |
938 | .elevator_owner = THIS_MODULE, | |
939 | }; | |
940 | ||
941 | static int __init deadline_init(void) | |
942 | { | |
943 | int ret; | |
944 | ||
945 | drq_pool = kmem_cache_create("deadline_drq", sizeof(struct deadline_rq), | |
946 | 0, 0, NULL, NULL); | |
947 | ||
948 | if (!drq_pool) | |
949 | return -ENOMEM; | |
950 | ||
951 | ret = elv_register(&iosched_deadline); | |
952 | if (ret) | |
953 | kmem_cache_destroy(drq_pool); | |
954 | ||
955 | return ret; | |
956 | } | |
957 | ||
958 | static void __exit deadline_exit(void) | |
959 | { | |
960 | kmem_cache_destroy(drq_pool); | |
961 | elv_unregister(&iosched_deadline); | |
962 | } | |
963 | ||
964 | module_init(deadline_init); | |
965 | module_exit(deadline_exit); | |
966 | ||
967 | MODULE_AUTHOR("Jens Axboe"); | |
968 | MODULE_LICENSE("GPL"); | |
969 | MODULE_DESCRIPTION("deadline IO scheduler"); |