]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Deadline i/o scheduler. |
3 | * | |
0fe23479 | 4 | * Copyright (C) 2002 Jens Axboe <[email protected]> |
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]; | |
4fb72f76 | 36 | |
1da177e4 LT |
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 | |
4fb72f76 AC |
56 | static inline struct rb_root * |
57 | deadline_rb_root(struct deadline_data *dd, struct request *rq) | |
58 | { | |
59 | return &dd->sort_list[rq_data_dir(rq)]; | |
60 | } | |
1da177e4 | 61 | |
5d1a5366 AC |
62 | /* |
63 | * get the request after `rq' in sector-sorted order | |
64 | */ | |
65 | static inline struct request * | |
66 | deadline_latter_request(struct request *rq) | |
67 | { | |
68 | struct rb_node *node = rb_next(&rq->rb_node); | |
69 | ||
70 | if (node) | |
71 | return rb_entry_rq(node); | |
72 | ||
73 | return NULL; | |
74 | } | |
75 | ||
1da177e4 | 76 | static void |
8840faa1 | 77 | deadline_add_rq_rb(struct deadline_data *dd, struct request *rq) |
1da177e4 | 78 | { |
4fb72f76 | 79 | struct rb_root *root = deadline_rb_root(dd, rq); |
1da177e4 | 80 | |
796d5116 | 81 | elv_rb_add(root, rq); |
1da177e4 LT |
82 | } |
83 | ||
84 | static inline void | |
8840faa1 | 85 | deadline_del_rq_rb(struct deadline_data *dd, struct request *rq) |
1da177e4 | 86 | { |
b8aca35a | 87 | const int data_dir = rq_data_dir(rq); |
1da177e4 | 88 | |
5d1a5366 AC |
89 | if (dd->next_rq[data_dir] == rq) |
90 | dd->next_rq[data_dir] = deadline_latter_request(rq); | |
1da177e4 | 91 | |
4fb72f76 | 92 | elv_rb_del(deadline_rb_root(dd, rq), rq); |
1da177e4 LT |
93 | } |
94 | ||
95 | /* | |
8840faa1 | 96 | * add rq to rbtree and fifo |
1da177e4 | 97 | */ |
b4878f24 | 98 | static void |
1da177e4 LT |
99 | deadline_add_request(struct request_queue *q, struct request *rq) |
100 | { | |
101 | struct deadline_data *dd = q->elevator->elevator_data; | |
8840faa1 | 102 | const int data_dir = rq_data_dir(rq); |
1da177e4 | 103 | |
8840faa1 | 104 | deadline_add_rq_rb(dd, rq); |
9817064b | 105 | |
1da177e4 | 106 | /* |
4fb72f76 | 107 | * set expire time and add to fifo list |
1da177e4 | 108 | */ |
8b4922d3 | 109 | rq->fifo_time = jiffies + dd->fifo_expire[data_dir]; |
8840faa1 | 110 | list_add_tail(&rq->queuelist, &dd->fifo_list[data_dir]); |
1da177e4 LT |
111 | } |
112 | ||
113 | /* | |
9817064b | 114 | * remove rq from rbtree and fifo. |
1da177e4 | 115 | */ |
165125e1 | 116 | static void deadline_remove_request(struct request_queue *q, struct request *rq) |
1da177e4 | 117 | { |
b4878f24 | 118 | struct deadline_data *dd = q->elevator->elevator_data; |
1da177e4 | 119 | |
8840faa1 JA |
120 | rq_fifo_clear(rq); |
121 | deadline_del_rq_rb(dd, rq); | |
1da177e4 LT |
122 | } |
123 | ||
124 | static int | |
165125e1 | 125 | deadline_merge(struct request_queue *q, struct request **req, struct bio *bio) |
1da177e4 LT |
126 | { |
127 | struct deadline_data *dd = q->elevator->elevator_data; | |
128 | struct request *__rq; | |
129 | int ret; | |
130 | ||
1da177e4 LT |
131 | /* |
132 | * check for front merge | |
133 | */ | |
134 | if (dd->front_merges) { | |
f73a1c7d | 135 | sector_t sector = bio_end_sector(bio); |
1da177e4 | 136 | |
b8aca35a | 137 | __rq = elv_rb_find(&dd->sort_list[bio_data_dir(bio)], sector); |
1da177e4 | 138 | if (__rq) { |
83096ebf | 139 | BUG_ON(sector != blk_rq_pos(__rq)); |
1da177e4 LT |
140 | |
141 | if (elv_rq_merge_ok(__rq, bio)) { | |
142 | ret = ELEVATOR_FRONT_MERGE; | |
143 | goto out; | |
144 | } | |
145 | } | |
146 | } | |
147 | ||
148 | return ELEVATOR_NO_MERGE; | |
149 | out: | |
1da177e4 LT |
150 | *req = __rq; |
151 | return ret; | |
152 | } | |
153 | ||
165125e1 JA |
154 | static void deadline_merged_request(struct request_queue *q, |
155 | struct request *req, int type) | |
1da177e4 LT |
156 | { |
157 | struct deadline_data *dd = q->elevator->elevator_data; | |
1da177e4 | 158 | |
1da177e4 LT |
159 | /* |
160 | * if the merge was a front merge, we need to reposition request | |
161 | */ | |
b8aca35a | 162 | if (type == ELEVATOR_FRONT_MERGE) { |
4fb72f76 | 163 | elv_rb_del(deadline_rb_root(dd, req), req); |
8840faa1 | 164 | deadline_add_rq_rb(dd, req); |
1da177e4 | 165 | } |
1da177e4 LT |
166 | } |
167 | ||
168 | static void | |
165125e1 | 169 | deadline_merged_requests(struct request_queue *q, struct request *req, |
1da177e4 LT |
170 | struct request *next) |
171 | { | |
1da177e4 | 172 | /* |
8840faa1 JA |
173 | * if next expires before rq, assign its expire time to rq |
174 | * and move into next position (next will be deleted) in fifo | |
1da177e4 | 175 | */ |
8840faa1 | 176 | if (!list_empty(&req->queuelist) && !list_empty(&next->queuelist)) { |
8b4922d3 | 177 | if (time_before(next->fifo_time, req->fifo_time)) { |
8840faa1 | 178 | list_move(&req->queuelist, &next->queuelist); |
8b4922d3 | 179 | req->fifo_time = next->fifo_time; |
1da177e4 LT |
180 | } |
181 | } | |
182 | ||
183 | /* | |
184 | * kill knowledge of next, this one is a goner | |
185 | */ | |
186 | deadline_remove_request(q, next); | |
187 | } | |
188 | ||
189 | /* | |
190 | * move request from sort list to dispatch queue. | |
191 | */ | |
192 | static inline void | |
8840faa1 | 193 | deadline_move_to_dispatch(struct deadline_data *dd, struct request *rq) |
1da177e4 | 194 | { |
165125e1 | 195 | struct request_queue *q = rq->q; |
1da177e4 | 196 | |
8840faa1 JA |
197 | deadline_remove_request(q, rq); |
198 | elv_dispatch_add_tail(q, rq); | |
1da177e4 LT |
199 | } |
200 | ||
201 | /* | |
202 | * move an entry to dispatch queue | |
203 | */ | |
204 | static void | |
8840faa1 | 205 | deadline_move_request(struct deadline_data *dd, struct request *rq) |
1da177e4 | 206 | { |
b8aca35a | 207 | const int data_dir = rq_data_dir(rq); |
1da177e4 | 208 | |
8840faa1 JA |
209 | dd->next_rq[READ] = NULL; |
210 | dd->next_rq[WRITE] = NULL; | |
5d1a5366 | 211 | dd->next_rq[data_dir] = deadline_latter_request(rq); |
1da177e4 | 212 | |
4fb72f76 | 213 | dd->last_sector = rq_end_sector(rq); |
1da177e4 LT |
214 | |
215 | /* | |
216 | * take it off the sort and fifo list, move | |
217 | * to dispatch queue | |
218 | */ | |
8840faa1 | 219 | deadline_move_to_dispatch(dd, rq); |
1da177e4 LT |
220 | } |
221 | ||
1da177e4 | 222 | /* |
4fb72f76 | 223 | * deadline_check_fifo returns 0 if there are no expired requests on the fifo, |
1da177e4 LT |
224 | * 1 otherwise. Requires !list_empty(&dd->fifo_list[data_dir]) |
225 | */ | |
226 | static inline int deadline_check_fifo(struct deadline_data *dd, int ddir) | |
227 | { | |
8840faa1 | 228 | struct request *rq = rq_entry_fifo(dd->fifo_list[ddir].next); |
1da177e4 LT |
229 | |
230 | /* | |
8840faa1 | 231 | * rq is expired! |
1da177e4 | 232 | */ |
8b4922d3 | 233 | if (time_after_eq(jiffies, rq->fifo_time)) |
1da177e4 LT |
234 | return 1; |
235 | ||
236 | return 0; | |
237 | } | |
238 | ||
239 | /* | |
240 | * deadline_dispatch_requests selects the best request according to | |
241 | * read/write expire, fifo_batch, etc | |
242 | */ | |
165125e1 | 243 | static int deadline_dispatch_requests(struct request_queue *q, int force) |
1da177e4 | 244 | { |
b4878f24 | 245 | struct deadline_data *dd = q->elevator->elevator_data; |
1da177e4 LT |
246 | const int reads = !list_empty(&dd->fifo_list[READ]); |
247 | const int writes = !list_empty(&dd->fifo_list[WRITE]); | |
8840faa1 | 248 | struct request *rq; |
4b0dc07e | 249 | int data_dir; |
1da177e4 LT |
250 | |
251 | /* | |
252 | * batches are currently reads XOR writes | |
253 | */ | |
8840faa1 JA |
254 | if (dd->next_rq[WRITE]) |
255 | rq = dd->next_rq[WRITE]; | |
9d5c1e1b | 256 | else |
8840faa1 | 257 | rq = dd->next_rq[READ]; |
1da177e4 | 258 | |
63de428b AC |
259 | if (rq && dd->batching < dd->fifo_batch) |
260 | /* we have a next request are still entitled to batch */ | |
261 | goto dispatch_request; | |
1da177e4 LT |
262 | |
263 | /* | |
264 | * at this point we are not running a batch. select the appropriate | |
265 | * data direction (read / write) | |
266 | */ | |
267 | ||
268 | if (reads) { | |
dd67d051 | 269 | BUG_ON(RB_EMPTY_ROOT(&dd->sort_list[READ])); |
1da177e4 LT |
270 | |
271 | if (writes && (dd->starved++ >= dd->writes_starved)) | |
272 | goto dispatch_writes; | |
273 | ||
274 | data_dir = READ; | |
1da177e4 LT |
275 | |
276 | goto dispatch_find_request; | |
277 | } | |
278 | ||
279 | /* | |
280 | * there are either no reads or writes have been starved | |
281 | */ | |
282 | ||
283 | if (writes) { | |
284 | dispatch_writes: | |
dd67d051 | 285 | BUG_ON(RB_EMPTY_ROOT(&dd->sort_list[WRITE])); |
1da177e4 LT |
286 | |
287 | dd->starved = 0; | |
288 | ||
289 | data_dir = WRITE; | |
1da177e4 LT |
290 | |
291 | goto dispatch_find_request; | |
292 | } | |
293 | ||
294 | return 0; | |
295 | ||
296 | dispatch_find_request: | |
297 | /* | |
298 | * we are not running a batch, find best request for selected data_dir | |
299 | */ | |
6f5d8aa6 AC |
300 | if (deadline_check_fifo(dd, data_dir) || !dd->next_rq[data_dir]) { |
301 | /* | |
302 | * A deadline has expired, the last request was in the other | |
303 | * direction, or we have run out of higher-sectored requests. | |
304 | * Start again from the request with the earliest expiry time. | |
305 | */ | |
8840faa1 | 306 | rq = rq_entry_fifo(dd->fifo_list[data_dir].next); |
6f5d8aa6 | 307 | } else { |
1da177e4 LT |
308 | /* |
309 | * The last req was the same dir and we have a next request in | |
310 | * sort order. No expired requests so continue on from here. | |
311 | */ | |
8840faa1 | 312 | rq = dd->next_rq[data_dir]; |
1da177e4 LT |
313 | } |
314 | ||
dfb3d72a AC |
315 | dd->batching = 0; |
316 | ||
1da177e4 LT |
317 | dispatch_request: |
318 | /* | |
8840faa1 | 319 | * rq is the selected appropriate request. |
1da177e4 LT |
320 | */ |
321 | dd->batching++; | |
8840faa1 | 322 | deadline_move_request(dd, rq); |
1da177e4 LT |
323 | |
324 | return 1; | |
325 | } | |
326 | ||
b374d18a | 327 | static void deadline_exit_queue(struct elevator_queue *e) |
1da177e4 LT |
328 | { |
329 | struct deadline_data *dd = e->elevator_data; | |
330 | ||
331 | BUG_ON(!list_empty(&dd->fifo_list[READ])); | |
332 | BUG_ON(!list_empty(&dd->fifo_list[WRITE])); | |
333 | ||
1da177e4 LT |
334 | kfree(dd); |
335 | } | |
336 | ||
337 | /* | |
8840faa1 | 338 | * initialize elevator private data (deadline_data). |
1da177e4 | 339 | */ |
d50235b7 | 340 | static int deadline_init_queue(struct request_queue *q, struct elevator_type *e) |
1da177e4 LT |
341 | { |
342 | struct deadline_data *dd; | |
d50235b7 JM |
343 | struct elevator_queue *eq; |
344 | ||
345 | eq = elevator_alloc(q, e); | |
346 | if (!eq) | |
347 | return -ENOMEM; | |
1da177e4 | 348 | |
c1b511eb | 349 | dd = kzalloc_node(sizeof(*dd), GFP_KERNEL, q->node); |
d50235b7 JM |
350 | if (!dd) { |
351 | kobject_put(&eq->kobj); | |
b2fab5ac | 352 | return -ENOMEM; |
d50235b7 JM |
353 | } |
354 | eq->elevator_data = dd; | |
1da177e4 | 355 | |
1da177e4 LT |
356 | INIT_LIST_HEAD(&dd->fifo_list[READ]); |
357 | INIT_LIST_HEAD(&dd->fifo_list[WRITE]); | |
358 | dd->sort_list[READ] = RB_ROOT; | |
359 | dd->sort_list[WRITE] = RB_ROOT; | |
1da177e4 LT |
360 | dd->fifo_expire[READ] = read_expire; |
361 | dd->fifo_expire[WRITE] = write_expire; | |
362 | dd->writes_starved = writes_starved; | |
363 | dd->front_merges = 1; | |
364 | dd->fifo_batch = fifo_batch; | |
b2fab5ac | 365 | |
d50235b7 JM |
366 | spin_lock_irq(q->queue_lock); |
367 | q->elevator = eq; | |
368 | spin_unlock_irq(q->queue_lock); | |
b2fab5ac | 369 | return 0; |
1da177e4 LT |
370 | } |
371 | ||
1da177e4 LT |
372 | /* |
373 | * sysfs parts below | |
374 | */ | |
1da177e4 LT |
375 | |
376 | static ssize_t | |
377 | deadline_var_show(int var, char *page) | |
378 | { | |
379 | return sprintf(page, "%d\n", var); | |
380 | } | |
381 | ||
382 | static ssize_t | |
383 | deadline_var_store(int *var, const char *page, size_t count) | |
384 | { | |
385 | char *p = (char *) page; | |
386 | ||
387 | *var = simple_strtol(p, &p, 10); | |
388 | return count; | |
389 | } | |
390 | ||
391 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ | |
b374d18a | 392 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 393 | { \ |
3d1ab40f AV |
394 | struct deadline_data *dd = e->elevator_data; \ |
395 | int __data = __VAR; \ | |
1da177e4 LT |
396 | if (__CONV) \ |
397 | __data = jiffies_to_msecs(__data); \ | |
398 | return deadline_var_show(__data, (page)); \ | |
399 | } | |
e572ec7e AV |
400 | SHOW_FUNCTION(deadline_read_expire_show, dd->fifo_expire[READ], 1); |
401 | SHOW_FUNCTION(deadline_write_expire_show, dd->fifo_expire[WRITE], 1); | |
402 | SHOW_FUNCTION(deadline_writes_starved_show, dd->writes_starved, 0); | |
403 | SHOW_FUNCTION(deadline_front_merges_show, dd->front_merges, 0); | |
404 | SHOW_FUNCTION(deadline_fifo_batch_show, dd->fifo_batch, 0); | |
1da177e4 LT |
405 | #undef SHOW_FUNCTION |
406 | ||
407 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 408 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 409 | { \ |
3d1ab40f | 410 | struct deadline_data *dd = e->elevator_data; \ |
1da177e4 LT |
411 | int __data; \ |
412 | int ret = deadline_var_store(&__data, (page), count); \ | |
413 | if (__data < (MIN)) \ | |
414 | __data = (MIN); \ | |
415 | else if (__data > (MAX)) \ | |
416 | __data = (MAX); \ | |
417 | if (__CONV) \ | |
418 | *(__PTR) = msecs_to_jiffies(__data); \ | |
419 | else \ | |
420 | *(__PTR) = __data; \ | |
421 | return ret; \ | |
422 | } | |
e572ec7e AV |
423 | STORE_FUNCTION(deadline_read_expire_store, &dd->fifo_expire[READ], 0, INT_MAX, 1); |
424 | STORE_FUNCTION(deadline_write_expire_store, &dd->fifo_expire[WRITE], 0, INT_MAX, 1); | |
425 | STORE_FUNCTION(deadline_writes_starved_store, &dd->writes_starved, INT_MIN, INT_MAX, 0); | |
426 | STORE_FUNCTION(deadline_front_merges_store, &dd->front_merges, 0, 1, 0); | |
427 | STORE_FUNCTION(deadline_fifo_batch_store, &dd->fifo_batch, 0, INT_MAX, 0); | |
1da177e4 LT |
428 | #undef STORE_FUNCTION |
429 | ||
e572ec7e AV |
430 | #define DD_ATTR(name) \ |
431 | __ATTR(name, S_IRUGO|S_IWUSR, deadline_##name##_show, \ | |
432 | deadline_##name##_store) | |
433 | ||
434 | static struct elv_fs_entry deadline_attrs[] = { | |
435 | DD_ATTR(read_expire), | |
436 | DD_ATTR(write_expire), | |
437 | DD_ATTR(writes_starved), | |
438 | DD_ATTR(front_merges), | |
439 | DD_ATTR(fifo_batch), | |
440 | __ATTR_NULL | |
1da177e4 LT |
441 | }; |
442 | ||
1da177e4 LT |
443 | static struct elevator_type iosched_deadline = { |
444 | .ops = { | |
445 | .elevator_merge_fn = deadline_merge, | |
446 | .elevator_merged_fn = deadline_merged_request, | |
447 | .elevator_merge_req_fn = deadline_merged_requests, | |
b4878f24 JA |
448 | .elevator_dispatch_fn = deadline_dispatch_requests, |
449 | .elevator_add_req_fn = deadline_add_request, | |
b8aca35a JA |
450 | .elevator_former_req_fn = elv_rb_former_request, |
451 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
452 | .elevator_init_fn = deadline_init_queue, |
453 | .elevator_exit_fn = deadline_exit_queue, | |
454 | }, | |
455 | ||
3d1ab40f | 456 | .elevator_attrs = deadline_attrs, |
1da177e4 LT |
457 | .elevator_name = "deadline", |
458 | .elevator_owner = THIS_MODULE, | |
459 | }; | |
460 | ||
461 | static int __init deadline_init(void) | |
462 | { | |
3d3c2379 | 463 | return elv_register(&iosched_deadline); |
1da177e4 LT |
464 | } |
465 | ||
466 | static void __exit deadline_exit(void) | |
467 | { | |
1da177e4 LT |
468 | elv_unregister(&iosched_deadline); |
469 | } | |
470 | ||
471 | module_init(deadline_init); | |
472 | module_exit(deadline_exit); | |
473 | ||
474 | MODULE_AUTHOR("Jens Axboe"); | |
475 | MODULE_LICENSE("GPL"); | |
476 | MODULE_DESCRIPTION("deadline IO scheduler"); |