Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Deadline i/o scheduler. |
3 | * | |
0fe23479 | 4 | * Copyright (C) 2002 Jens Axboe <axboe@kernel.dk> |
1da177e4 LT |
5 | */ |
6 | #include <linux/kernel.h> | |
7 | #include <linux/fs.h> | |
8 | #include <linux/blkdev.h> | |
9 | #include <linux/elevator.h> | |
10 | #include <linux/bio.h> | |
1da177e4 LT |
11 | #include <linux/module.h> |
12 | #include <linux/slab.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/compiler.h> | |
1da177e4 LT |
15 | #include <linux/rbtree.h> |
16 | ||
17 | /* | |
18 | * See Documentation/block/deadline-iosched.txt | |
19 | */ | |
64100099 AV |
20 | static const int read_expire = HZ / 2; /* max time before a read is submitted. */ |
21 | static const int write_expire = 5 * HZ; /* ditto for writes, these limits are SOFT! */ | |
22 | static const int writes_starved = 2; /* max times reads can starve a write */ | |
23 | static const int fifo_batch = 16; /* # of sequential requests treated as one | |
1da177e4 LT |
24 | by the above parameters. For throughput. */ |
25 | ||
1da177e4 LT |
26 | struct deadline_data { |
27 | /* | |
28 | * run time data | |
29 | */ | |
30 | ||
31 | /* | |
32 | * requests (deadline_rq s) are present on both sort_list and fifo_list | |
33 | */ | |
34 | struct rb_root sort_list[2]; | |
35 | struct list_head fifo_list[2]; | |
36 | ||
37 | /* | |
38 | * next in sort order. read, write or both are NULL | |
39 | */ | |
8840faa1 | 40 | struct request *next_rq[2]; |
1da177e4 LT |
41 | unsigned int batching; /* number of sequential requests made */ |
42 | sector_t last_sector; /* head position */ | |
43 | unsigned int starved; /* times reads have starved writes */ | |
44 | ||
45 | /* | |
46 | * settings that change how the i/o scheduler behaves | |
47 | */ | |
48 | int fifo_expire[2]; | |
49 | int fifo_batch; | |
50 | int writes_starved; | |
51 | int front_merges; | |
1da177e4 LT |
52 | }; |
53 | ||
8840faa1 | 54 | static void deadline_move_request(struct deadline_data *, struct request *); |
1da177e4 | 55 | |
b8aca35a | 56 | #define RQ_RB_ROOT(dd, rq) (&(dd)->sort_list[rq_data_dir((rq))]) |
1da177e4 | 57 | |
5d1a5366 AC |
58 | /* |
59 | * get the request after `rq' in sector-sorted order | |
60 | */ | |
61 | static inline struct request * | |
62 | deadline_latter_request(struct request *rq) | |
63 | { | |
64 | struct rb_node *node = rb_next(&rq->rb_node); | |
65 | ||
66 | if (node) | |
67 | return rb_entry_rq(node); | |
68 | ||
69 | return NULL; | |
70 | } | |
71 | ||
1da177e4 | 72 | static void |
8840faa1 | 73 | deadline_add_rq_rb(struct deadline_data *dd, struct request *rq) |
1da177e4 | 74 | { |
b8aca35a JA |
75 | struct rb_root *root = RQ_RB_ROOT(dd, rq); |
76 | struct request *__alias; | |
1da177e4 LT |
77 | |
78 | retry: | |
b8aca35a JA |
79 | __alias = elv_rb_add(root, rq); |
80 | if (unlikely(__alias)) { | |
8840faa1 | 81 | deadline_move_request(dd, __alias); |
b8aca35a | 82 | goto retry; |
1da177e4 | 83 | } |
1da177e4 LT |
84 | } |
85 | ||
86 | static inline void | |
8840faa1 | 87 | deadline_del_rq_rb(struct deadline_data *dd, struct request *rq) |
1da177e4 | 88 | { |
b8aca35a | 89 | const int data_dir = rq_data_dir(rq); |
1da177e4 | 90 | |
5d1a5366 AC |
91 | if (dd->next_rq[data_dir] == rq) |
92 | dd->next_rq[data_dir] = deadline_latter_request(rq); | |
1da177e4 | 93 | |
b8aca35a | 94 | elv_rb_del(RQ_RB_ROOT(dd, rq), rq); |
1da177e4 LT |
95 | } |
96 | ||
97 | /* | |
8840faa1 | 98 | * add rq to rbtree and fifo |
1da177e4 | 99 | */ |
b4878f24 | 100 | static void |
1da177e4 LT |
101 | deadline_add_request(struct request_queue *q, struct request *rq) |
102 | { | |
103 | struct deadline_data *dd = q->elevator->elevator_data; | |
8840faa1 | 104 | const int data_dir = rq_data_dir(rq); |
1da177e4 | 105 | |
8840faa1 | 106 | deadline_add_rq_rb(dd, rq); |
9817064b | 107 | |
1da177e4 LT |
108 | /* |
109 | * set expire time (only used for reads) and add to fifo list | |
110 | */ | |
8840faa1 JA |
111 | rq_set_fifo_time(rq, jiffies + dd->fifo_expire[data_dir]); |
112 | list_add_tail(&rq->queuelist, &dd->fifo_list[data_dir]); | |
1da177e4 LT |
113 | } |
114 | ||
115 | /* | |
9817064b | 116 | * remove rq from rbtree and fifo. |
1da177e4 | 117 | */ |
165125e1 | 118 | static void deadline_remove_request(struct request_queue *q, struct request *rq) |
1da177e4 | 119 | { |
b4878f24 | 120 | struct deadline_data *dd = q->elevator->elevator_data; |
1da177e4 | 121 | |
8840faa1 JA |
122 | rq_fifo_clear(rq); |
123 | deadline_del_rq_rb(dd, rq); | |
1da177e4 LT |
124 | } |
125 | ||
126 | static int | |
165125e1 | 127 | deadline_merge(struct request_queue *q, struct request **req, struct bio *bio) |
1da177e4 LT |
128 | { |
129 | struct deadline_data *dd = q->elevator->elevator_data; | |
130 | struct request *__rq; | |
131 | int ret; | |
132 | ||
1da177e4 LT |
133 | /* |
134 | * check for front merge | |
135 | */ | |
136 | if (dd->front_merges) { | |
b8aca35a | 137 | sector_t sector = bio->bi_sector + bio_sectors(bio); |
1da177e4 | 138 | |
b8aca35a | 139 | __rq = elv_rb_find(&dd->sort_list[bio_data_dir(bio)], sector); |
1da177e4 | 140 | if (__rq) { |
b8aca35a | 141 | BUG_ON(sector != __rq->sector); |
1da177e4 LT |
142 | |
143 | if (elv_rq_merge_ok(__rq, bio)) { | |
144 | ret = ELEVATOR_FRONT_MERGE; | |
145 | goto out; | |
146 | } | |
147 | } | |
148 | } | |
149 | ||
150 | return ELEVATOR_NO_MERGE; | |
151 | out: | |
1da177e4 LT |
152 | *req = __rq; |
153 | return ret; | |
154 | } | |
155 | ||
165125e1 JA |
156 | static void deadline_merged_request(struct request_queue *q, |
157 | struct request *req, int type) | |
1da177e4 LT |
158 | { |
159 | struct deadline_data *dd = q->elevator->elevator_data; | |
1da177e4 | 160 | |
1da177e4 LT |
161 | /* |
162 | * if the merge was a front merge, we need to reposition request | |
163 | */ | |
b8aca35a JA |
164 | if (type == ELEVATOR_FRONT_MERGE) { |
165 | elv_rb_del(RQ_RB_ROOT(dd, req), req); | |
8840faa1 | 166 | deadline_add_rq_rb(dd, req); |
1da177e4 | 167 | } |
1da177e4 LT |
168 | } |
169 | ||
170 | static void | |
165125e1 | 171 | deadline_merged_requests(struct request_queue *q, struct request *req, |
1da177e4 LT |
172 | struct request *next) |
173 | { | |
1da177e4 | 174 | /* |
8840faa1 JA |
175 | * if next expires before rq, assign its expire time to rq |
176 | * and move into next position (next will be deleted) in fifo | |
1da177e4 | 177 | */ |
8840faa1 JA |
178 | if (!list_empty(&req->queuelist) && !list_empty(&next->queuelist)) { |
179 | if (time_before(rq_fifo_time(next), rq_fifo_time(req))) { | |
180 | list_move(&req->queuelist, &next->queuelist); | |
181 | rq_set_fifo_time(req, rq_fifo_time(next)); | |
1da177e4 LT |
182 | } |
183 | } | |
184 | ||
185 | /* | |
186 | * kill knowledge of next, this one is a goner | |
187 | */ | |
188 | deadline_remove_request(q, next); | |
189 | } | |
190 | ||
191 | /* | |
192 | * move request from sort list to dispatch queue. | |
193 | */ | |
194 | static inline void | |
8840faa1 | 195 | deadline_move_to_dispatch(struct deadline_data *dd, struct request *rq) |
1da177e4 | 196 | { |
165125e1 | 197 | struct request_queue *q = rq->q; |
1da177e4 | 198 | |
8840faa1 JA |
199 | deadline_remove_request(q, rq); |
200 | elv_dispatch_add_tail(q, rq); | |
1da177e4 LT |
201 | } |
202 | ||
203 | /* | |
204 | * move an entry to dispatch queue | |
205 | */ | |
206 | static void | |
8840faa1 | 207 | deadline_move_request(struct deadline_data *dd, struct request *rq) |
1da177e4 | 208 | { |
b8aca35a | 209 | const int data_dir = rq_data_dir(rq); |
1da177e4 | 210 | |
8840faa1 JA |
211 | dd->next_rq[READ] = NULL; |
212 | dd->next_rq[WRITE] = NULL; | |
5d1a5366 | 213 | dd->next_rq[data_dir] = deadline_latter_request(rq); |
1da177e4 | 214 | |
8840faa1 | 215 | dd->last_sector = rq->sector + rq->nr_sectors; |
1da177e4 LT |
216 | |
217 | /* | |
218 | * take it off the sort and fifo list, move | |
219 | * to dispatch queue | |
220 | */ | |
8840faa1 | 221 | deadline_move_to_dispatch(dd, rq); |
1da177e4 LT |
222 | } |
223 | ||
1da177e4 LT |
224 | /* |
225 | * deadline_check_fifo returns 0 if there are no expired reads on the fifo, | |
226 | * 1 otherwise. Requires !list_empty(&dd->fifo_list[data_dir]) | |
227 | */ | |
228 | static inline int deadline_check_fifo(struct deadline_data *dd, int ddir) | |
229 | { | |
8840faa1 | 230 | struct request *rq = rq_entry_fifo(dd->fifo_list[ddir].next); |
1da177e4 LT |
231 | |
232 | /* | |
8840faa1 | 233 | * rq is expired! |
1da177e4 | 234 | */ |
8840faa1 | 235 | if (time_after(jiffies, rq_fifo_time(rq))) |
1da177e4 LT |
236 | return 1; |
237 | ||
238 | return 0; | |
239 | } | |
240 | ||
241 | /* | |
242 | * deadline_dispatch_requests selects the best request according to | |
243 | * read/write expire, fifo_batch, etc | |
244 | */ | |
165125e1 | 245 | static int deadline_dispatch_requests(struct request_queue *q, int force) |
1da177e4 | 246 | { |
b4878f24 | 247 | struct deadline_data *dd = q->elevator->elevator_data; |
1da177e4 LT |
248 | const int reads = !list_empty(&dd->fifo_list[READ]); |
249 | const int writes = !list_empty(&dd->fifo_list[WRITE]); | |
8840faa1 | 250 | struct request *rq; |
4b0dc07e | 251 | int data_dir; |
1da177e4 LT |
252 | |
253 | /* | |
254 | * batches are currently reads XOR writes | |
255 | */ | |
8840faa1 JA |
256 | if (dd->next_rq[WRITE]) |
257 | rq = dd->next_rq[WRITE]; | |
9d5c1e1b | 258 | else |
8840faa1 | 259 | rq = dd->next_rq[READ]; |
1da177e4 | 260 | |
8840faa1 | 261 | if (rq) { |
1da177e4 LT |
262 | /* we have a "next request" */ |
263 | ||
8840faa1 | 264 | if (dd->last_sector != rq->sector) |
1da177e4 LT |
265 | /* end the batch on a non sequential request */ |
266 | dd->batching += dd->fifo_batch; | |
267 | ||
268 | if (dd->batching < dd->fifo_batch) | |
269 | /* we are still entitled to batch */ | |
270 | goto dispatch_request; | |
271 | } | |
272 | ||
273 | /* | |
274 | * at this point we are not running a batch. select the appropriate | |
275 | * data direction (read / write) | |
276 | */ | |
277 | ||
278 | if (reads) { | |
dd67d051 | 279 | BUG_ON(RB_EMPTY_ROOT(&dd->sort_list[READ])); |
1da177e4 LT |
280 | |
281 | if (writes && (dd->starved++ >= dd->writes_starved)) | |
282 | goto dispatch_writes; | |
283 | ||
284 | data_dir = READ; | |
1da177e4 LT |
285 | |
286 | goto dispatch_find_request; | |
287 | } | |
288 | ||
289 | /* | |
290 | * there are either no reads or writes have been starved | |
291 | */ | |
292 | ||
293 | if (writes) { | |
294 | dispatch_writes: | |
dd67d051 | 295 | BUG_ON(RB_EMPTY_ROOT(&dd->sort_list[WRITE])); |
1da177e4 LT |
296 | |
297 | dd->starved = 0; | |
298 | ||
299 | data_dir = WRITE; | |
1da177e4 LT |
300 | |
301 | goto dispatch_find_request; | |
302 | } | |
303 | ||
304 | return 0; | |
305 | ||
306 | dispatch_find_request: | |
307 | /* | |
308 | * we are not running a batch, find best request for selected data_dir | |
309 | */ | |
6f5d8aa6 AC |
310 | if (deadline_check_fifo(dd, data_dir) || !dd->next_rq[data_dir]) { |
311 | /* | |
312 | * A deadline has expired, the last request was in the other | |
313 | * direction, or we have run out of higher-sectored requests. | |
314 | * Start again from the request with the earliest expiry time. | |
315 | */ | |
8840faa1 | 316 | rq = rq_entry_fifo(dd->fifo_list[data_dir].next); |
6f5d8aa6 | 317 | } else { |
1da177e4 LT |
318 | /* |
319 | * The last req was the same dir and we have a next request in | |
320 | * sort order. No expired requests so continue on from here. | |
321 | */ | |
8840faa1 | 322 | rq = dd->next_rq[data_dir]; |
1da177e4 LT |
323 | } |
324 | ||
dfb3d72a AC |
325 | dd->batching = 0; |
326 | ||
1da177e4 LT |
327 | dispatch_request: |
328 | /* | |
8840faa1 | 329 | * rq is the selected appropriate request. |
1da177e4 LT |
330 | */ |
331 | dd->batching++; | |
8840faa1 | 332 | deadline_move_request(dd, rq); |
1da177e4 LT |
333 | |
334 | return 1; | |
335 | } | |
336 | ||
165125e1 | 337 | static int deadline_queue_empty(struct request_queue *q) |
1da177e4 LT |
338 | { |
339 | struct deadline_data *dd = q->elevator->elevator_data; | |
340 | ||
b4878f24 JA |
341 | return list_empty(&dd->fifo_list[WRITE]) |
342 | && list_empty(&dd->fifo_list[READ]); | |
1da177e4 LT |
343 | } |
344 | ||
1da177e4 LT |
345 | static void deadline_exit_queue(elevator_t *e) |
346 | { | |
347 | struct deadline_data *dd = e->elevator_data; | |
348 | ||
349 | BUG_ON(!list_empty(&dd->fifo_list[READ])); | |
350 | BUG_ON(!list_empty(&dd->fifo_list[WRITE])); | |
351 | ||
1da177e4 LT |
352 | kfree(dd); |
353 | } | |
354 | ||
355 | /* | |
8840faa1 | 356 | * initialize elevator private data (deadline_data). |
1da177e4 | 357 | */ |
165125e1 | 358 | static void *deadline_init_queue(struct request_queue *q) |
1da177e4 LT |
359 | { |
360 | struct deadline_data *dd; | |
1da177e4 | 361 | |
94f6030c | 362 | dd = kmalloc_node(sizeof(*dd), GFP_KERNEL | __GFP_ZERO, q->node); |
1da177e4 | 363 | if (!dd) |
bc1c1169 | 364 | return NULL; |
1da177e4 | 365 | |
1da177e4 LT |
366 | INIT_LIST_HEAD(&dd->fifo_list[READ]); |
367 | INIT_LIST_HEAD(&dd->fifo_list[WRITE]); | |
368 | dd->sort_list[READ] = RB_ROOT; | |
369 | dd->sort_list[WRITE] = RB_ROOT; | |
1da177e4 LT |
370 | dd->fifo_expire[READ] = read_expire; |
371 | dd->fifo_expire[WRITE] = write_expire; | |
372 | dd->writes_starved = writes_starved; | |
373 | dd->front_merges = 1; | |
374 | dd->fifo_batch = fifo_batch; | |
bc1c1169 | 375 | return dd; |
1da177e4 LT |
376 | } |
377 | ||
1da177e4 LT |
378 | /* |
379 | * sysfs parts below | |
380 | */ | |
1da177e4 LT |
381 | |
382 | static ssize_t | |
383 | deadline_var_show(int var, char *page) | |
384 | { | |
385 | return sprintf(page, "%d\n", var); | |
386 | } | |
387 | ||
388 | static ssize_t | |
389 | deadline_var_store(int *var, const char *page, size_t count) | |
390 | { | |
391 | char *p = (char *) page; | |
392 | ||
393 | *var = simple_strtol(p, &p, 10); | |
394 | return count; | |
395 | } | |
396 | ||
397 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ | |
3d1ab40f | 398 | static ssize_t __FUNC(elevator_t *e, char *page) \ |
1da177e4 | 399 | { \ |
3d1ab40f AV |
400 | struct deadline_data *dd = e->elevator_data; \ |
401 | int __data = __VAR; \ | |
1da177e4 LT |
402 | if (__CONV) \ |
403 | __data = jiffies_to_msecs(__data); \ | |
404 | return deadline_var_show(__data, (page)); \ | |
405 | } | |
e572ec7e AV |
406 | SHOW_FUNCTION(deadline_read_expire_show, dd->fifo_expire[READ], 1); |
407 | SHOW_FUNCTION(deadline_write_expire_show, dd->fifo_expire[WRITE], 1); | |
408 | SHOW_FUNCTION(deadline_writes_starved_show, dd->writes_starved, 0); | |
409 | SHOW_FUNCTION(deadline_front_merges_show, dd->front_merges, 0); | |
410 | SHOW_FUNCTION(deadline_fifo_batch_show, dd->fifo_batch, 0); | |
1da177e4 LT |
411 | #undef SHOW_FUNCTION |
412 | ||
413 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
3d1ab40f | 414 | static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \ |
1da177e4 | 415 | { \ |
3d1ab40f | 416 | struct deadline_data *dd = e->elevator_data; \ |
1da177e4 LT |
417 | int __data; \ |
418 | int ret = deadline_var_store(&__data, (page), count); \ | |
419 | if (__data < (MIN)) \ | |
420 | __data = (MIN); \ | |
421 | else if (__data > (MAX)) \ | |
422 | __data = (MAX); \ | |
423 | if (__CONV) \ | |
424 | *(__PTR) = msecs_to_jiffies(__data); \ | |
425 | else \ | |
426 | *(__PTR) = __data; \ | |
427 | return ret; \ | |
428 | } | |
e572ec7e AV |
429 | STORE_FUNCTION(deadline_read_expire_store, &dd->fifo_expire[READ], 0, INT_MAX, 1); |
430 | STORE_FUNCTION(deadline_write_expire_store, &dd->fifo_expire[WRITE], 0, INT_MAX, 1); | |
431 | STORE_FUNCTION(deadline_writes_starved_store, &dd->writes_starved, INT_MIN, INT_MAX, 0); | |
432 | STORE_FUNCTION(deadline_front_merges_store, &dd->front_merges, 0, 1, 0); | |
433 | STORE_FUNCTION(deadline_fifo_batch_store, &dd->fifo_batch, 0, INT_MAX, 0); | |
1da177e4 LT |
434 | #undef STORE_FUNCTION |
435 | ||
e572ec7e AV |
436 | #define DD_ATTR(name) \ |
437 | __ATTR(name, S_IRUGO|S_IWUSR, deadline_##name##_show, \ | |
438 | deadline_##name##_store) | |
439 | ||
440 | static struct elv_fs_entry deadline_attrs[] = { | |
441 | DD_ATTR(read_expire), | |
442 | DD_ATTR(write_expire), | |
443 | DD_ATTR(writes_starved), | |
444 | DD_ATTR(front_merges), | |
445 | DD_ATTR(fifo_batch), | |
446 | __ATTR_NULL | |
1da177e4 LT |
447 | }; |
448 | ||
1da177e4 LT |
449 | static struct elevator_type iosched_deadline = { |
450 | .ops = { | |
451 | .elevator_merge_fn = deadline_merge, | |
452 | .elevator_merged_fn = deadline_merged_request, | |
453 | .elevator_merge_req_fn = deadline_merged_requests, | |
b4878f24 JA |
454 | .elevator_dispatch_fn = deadline_dispatch_requests, |
455 | .elevator_add_req_fn = deadline_add_request, | |
1da177e4 | 456 | .elevator_queue_empty_fn = deadline_queue_empty, |
b8aca35a JA |
457 | .elevator_former_req_fn = elv_rb_former_request, |
458 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
459 | .elevator_init_fn = deadline_init_queue, |
460 | .elevator_exit_fn = deadline_exit_queue, | |
461 | }, | |
462 | ||
3d1ab40f | 463 | .elevator_attrs = deadline_attrs, |
1da177e4 LT |
464 | .elevator_name = "deadline", |
465 | .elevator_owner = THIS_MODULE, | |
466 | }; | |
467 | ||
468 | static int __init deadline_init(void) | |
469 | { | |
2fdd82bd AB |
470 | elv_register(&iosched_deadline); |
471 | ||
472 | return 0; | |
1da177e4 LT |
473 | } |
474 | ||
475 | static void __exit deadline_exit(void) | |
476 | { | |
1da177e4 LT |
477 | elv_unregister(&iosched_deadline); |
478 | } | |
479 | ||
480 | module_init(deadline_init); | |
481 | module_exit(deadline_exit); | |
482 | ||
483 | MODULE_AUTHOR("Jens Axboe"); | |
484 | MODULE_LICENSE("GPL"); | |
485 | MODULE_DESCRIPTION("deadline IO scheduler"); |