Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * CFQ, or complete fairness queueing, disk scheduler. |
3 | * | |
4 | * Based on ideas from a previously unfinished io | |
5 | * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli. | |
6 | * | |
0fe23479 | 7 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
1da177e4 | 8 | */ |
1da177e4 | 9 | #include <linux/module.h> |
1cc9be68 AV |
10 | #include <linux/blkdev.h> |
11 | #include <linux/elevator.h> | |
1da177e4 LT |
12 | #include <linux/hash.h> |
13 | #include <linux/rbtree.h> | |
22e2c507 | 14 | #include <linux/ioprio.h> |
1da177e4 LT |
15 | |
16 | /* | |
17 | * tunables | |
18 | */ | |
64100099 | 19 | static const int cfq_quantum = 4; /* max queue in one round of service */ |
64100099 AV |
20 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
21 | static const int cfq_back_max = 16 * 1024; /* maximum backwards seek, in KiB */ | |
22 | static const int cfq_back_penalty = 2; /* penalty of a backwards seek */ | |
1da177e4 | 23 | |
64100099 | 24 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 25 | static int cfq_slice_async = HZ / 25; |
64100099 | 26 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 27 | static int cfq_slice_idle = HZ / 125; |
22e2c507 JA |
28 | |
29 | #define CFQ_IDLE_GRACE (HZ / 10) | |
30 | #define CFQ_SLICE_SCALE (5) | |
31 | ||
32 | #define CFQ_KEY_ASYNC (0) | |
22e2c507 | 33 | |
1da177e4 LT |
34 | /* |
35 | * for the hash of cfqq inside the cfqd | |
36 | */ | |
37 | #define CFQ_QHASH_SHIFT 6 | |
38 | #define CFQ_QHASH_ENTRIES (1 << CFQ_QHASH_SHIFT) | |
39 | #define list_entry_qhash(entry) hlist_entry((entry), struct cfq_queue, cfq_hash) | |
40 | ||
1da177e4 LT |
41 | #define list_entry_cfqq(ptr) list_entry((ptr), struct cfq_queue, cfq_list) |
42 | ||
5e705374 JA |
43 | #define RQ_CIC(rq) ((struct cfq_io_context*)(rq)->elevator_private) |
44 | #define RQ_CFQQ(rq) ((rq)->elevator_private2) | |
1da177e4 | 45 | |
e18b890b CL |
46 | static struct kmem_cache *cfq_pool; |
47 | static struct kmem_cache *cfq_ioc_pool; | |
1da177e4 | 48 | |
4050cf16 | 49 | static DEFINE_PER_CPU(unsigned long, ioc_count); |
334e94de AV |
50 | static struct completion *ioc_gone; |
51 | ||
22e2c507 JA |
52 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
53 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
54 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
55 | ||
3b18152c JA |
56 | #define ASYNC (0) |
57 | #define SYNC (1) | |
58 | ||
59 | #define cfq_cfqq_dispatched(cfqq) \ | |
60 | ((cfqq)->on_dispatch[ASYNC] + (cfqq)->on_dispatch[SYNC]) | |
61 | ||
62 | #define cfq_cfqq_class_sync(cfqq) ((cfqq)->key != CFQ_KEY_ASYNC) | |
63 | ||
64 | #define cfq_cfqq_sync(cfqq) \ | |
65 | (cfq_cfqq_class_sync(cfqq) || (cfqq)->on_dispatch[SYNC]) | |
22e2c507 | 66 | |
206dc69b JA |
67 | #define sample_valid(samples) ((samples) > 80) |
68 | ||
22e2c507 JA |
69 | /* |
70 | * Per block device queue structure | |
71 | */ | |
1da177e4 | 72 | struct cfq_data { |
22e2c507 JA |
73 | request_queue_t *queue; |
74 | ||
75 | /* | |
76 | * rr list of queues with requests and the count of them | |
77 | */ | |
78 | struct list_head rr_list[CFQ_PRIO_LISTS]; | |
79 | struct list_head busy_rr; | |
80 | struct list_head cur_rr; | |
81 | struct list_head idle_rr; | |
82 | unsigned int busy_queues; | |
83 | ||
22e2c507 JA |
84 | /* |
85 | * cfqq lookup hash | |
86 | */ | |
1da177e4 | 87 | struct hlist_head *cfq_hash; |
1da177e4 | 88 | |
22e2c507 | 89 | int rq_in_driver; |
25776e35 | 90 | int hw_tag; |
1da177e4 | 91 | |
22e2c507 JA |
92 | /* |
93 | * idle window management | |
94 | */ | |
95 | struct timer_list idle_slice_timer; | |
96 | struct work_struct unplug_work; | |
1da177e4 | 97 | |
22e2c507 JA |
98 | struct cfq_queue *active_queue; |
99 | struct cfq_io_context *active_cic; | |
100 | int cur_prio, cur_end_prio; | |
101 | unsigned int dispatch_slice; | |
102 | ||
103 | struct timer_list idle_class_timer; | |
1da177e4 LT |
104 | |
105 | sector_t last_sector; | |
22e2c507 | 106 | unsigned long last_end_request; |
1da177e4 | 107 | |
1da177e4 LT |
108 | /* |
109 | * tunables, see top of file | |
110 | */ | |
111 | unsigned int cfq_quantum; | |
22e2c507 | 112 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
113 | unsigned int cfq_back_penalty; |
114 | unsigned int cfq_back_max; | |
22e2c507 JA |
115 | unsigned int cfq_slice[2]; |
116 | unsigned int cfq_slice_async_rq; | |
117 | unsigned int cfq_slice_idle; | |
d9ff4187 AV |
118 | |
119 | struct list_head cic_list; | |
1da177e4 LT |
120 | }; |
121 | ||
22e2c507 JA |
122 | /* |
123 | * Per process-grouping structure | |
124 | */ | |
1da177e4 LT |
125 | struct cfq_queue { |
126 | /* reference count */ | |
127 | atomic_t ref; | |
128 | /* parent cfq_data */ | |
129 | struct cfq_data *cfqd; | |
22e2c507 | 130 | /* cfqq lookup hash */ |
1da177e4 LT |
131 | struct hlist_node cfq_hash; |
132 | /* hash key */ | |
22e2c507 | 133 | unsigned int key; |
981a7973 | 134 | /* member of the rr/busy/cur/idle cfqd list */ |
1da177e4 LT |
135 | struct list_head cfq_list; |
136 | /* sorted list of pending requests */ | |
137 | struct rb_root sort_list; | |
138 | /* if fifo isn't expired, next request to serve */ | |
5e705374 | 139 | struct request *next_rq; |
1da177e4 LT |
140 | /* requests queued in sort_list */ |
141 | int queued[2]; | |
142 | /* currently allocated requests */ | |
143 | int allocated[2]; | |
374f84ac JA |
144 | /* pending metadata requests */ |
145 | int meta_pending; | |
1da177e4 | 146 | /* fifo list of requests in sort_list */ |
22e2c507 | 147 | struct list_head fifo; |
1da177e4 | 148 | |
22e2c507 JA |
149 | unsigned long slice_start; |
150 | unsigned long slice_end; | |
151 | unsigned long slice_left; | |
1da177e4 | 152 | |
3b18152c JA |
153 | /* number of requests that are on the dispatch list */ |
154 | int on_dispatch[2]; | |
22e2c507 JA |
155 | |
156 | /* io prio of this group */ | |
157 | unsigned short ioprio, org_ioprio; | |
158 | unsigned short ioprio_class, org_ioprio_class; | |
159 | ||
3b18152c JA |
160 | /* various state flags, see below */ |
161 | unsigned int flags; | |
1da177e4 LT |
162 | }; |
163 | ||
3b18152c JA |
164 | enum cfqq_state_flags { |
165 | CFQ_CFQQ_FLAG_on_rr = 0, | |
166 | CFQ_CFQQ_FLAG_wait_request, | |
167 | CFQ_CFQQ_FLAG_must_alloc, | |
168 | CFQ_CFQQ_FLAG_must_alloc_slice, | |
169 | CFQ_CFQQ_FLAG_must_dispatch, | |
170 | CFQ_CFQQ_FLAG_fifo_expire, | |
171 | CFQ_CFQQ_FLAG_idle_window, | |
172 | CFQ_CFQQ_FLAG_prio_changed, | |
53b03744 | 173 | CFQ_CFQQ_FLAG_queue_new, |
3b18152c JA |
174 | }; |
175 | ||
176 | #define CFQ_CFQQ_FNS(name) \ | |
177 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
178 | { \ | |
179 | cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ | |
180 | } \ | |
181 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
182 | { \ | |
183 | cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ | |
184 | } \ | |
185 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
186 | { \ | |
187 | return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ | |
188 | } | |
189 | ||
190 | CFQ_CFQQ_FNS(on_rr); | |
191 | CFQ_CFQQ_FNS(wait_request); | |
192 | CFQ_CFQQ_FNS(must_alloc); | |
193 | CFQ_CFQQ_FNS(must_alloc_slice); | |
194 | CFQ_CFQQ_FNS(must_dispatch); | |
195 | CFQ_CFQQ_FNS(fifo_expire); | |
196 | CFQ_CFQQ_FNS(idle_window); | |
197 | CFQ_CFQQ_FNS(prio_changed); | |
53b03744 | 198 | CFQ_CFQQ_FNS(queue_new); |
3b18152c JA |
199 | #undef CFQ_CFQQ_FNS |
200 | ||
3b18152c | 201 | static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short); |
5e705374 | 202 | static void cfq_dispatch_insert(request_queue_t *, struct request *); |
6f325a13 | 203 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, gfp_t gfp_mask); |
1da177e4 | 204 | |
99f95e52 AM |
205 | /* |
206 | * scheduler run of queue, if there are requests pending and no one in the | |
207 | * driver that will restart queueing | |
208 | */ | |
209 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
210 | { | |
7b14e3b5 | 211 | if (cfqd->busy_queues) |
99f95e52 AM |
212 | kblockd_schedule_work(&cfqd->unplug_work); |
213 | } | |
214 | ||
215 | static int cfq_queue_empty(request_queue_t *q) | |
216 | { | |
217 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
218 | ||
b4878f24 | 219 | return !cfqd->busy_queues; |
99f95e52 AM |
220 | } |
221 | ||
7749a8d4 | 222 | static inline pid_t cfq_queue_pid(struct task_struct *task, int rw, int is_sync) |
206dc69b | 223 | { |
7749a8d4 JA |
224 | /* |
225 | * Use the per-process queue, for read requests and syncronous writes | |
226 | */ | |
227 | if (!(rw & REQ_RW) || is_sync) | |
206dc69b JA |
228 | return task->pid; |
229 | ||
230 | return CFQ_KEY_ASYNC; | |
231 | } | |
232 | ||
1da177e4 | 233 | /* |
5e705374 | 234 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 235 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 236 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 237 | */ |
5e705374 JA |
238 | static struct request * |
239 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
240 | { |
241 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 242 | unsigned long back_max; |
e8a99053 AM |
243 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
244 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
245 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 246 | |
5e705374 JA |
247 | if (rq1 == NULL || rq1 == rq2) |
248 | return rq2; | |
249 | if (rq2 == NULL) | |
250 | return rq1; | |
9c2c38a1 | 251 | |
5e705374 JA |
252 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
253 | return rq1; | |
254 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
255 | return rq2; | |
374f84ac JA |
256 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
257 | return rq1; | |
258 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
259 | return rq2; | |
1da177e4 | 260 | |
5e705374 JA |
261 | s1 = rq1->sector; |
262 | s2 = rq2->sector; | |
1da177e4 LT |
263 | |
264 | last = cfqd->last_sector; | |
265 | ||
1da177e4 LT |
266 | /* |
267 | * by definition, 1KiB is 2 sectors | |
268 | */ | |
269 | back_max = cfqd->cfq_back_max * 2; | |
270 | ||
271 | /* | |
272 | * Strict one way elevator _except_ in the case where we allow | |
273 | * short backward seeks which are biased as twice the cost of a | |
274 | * similar forward seek. | |
275 | */ | |
276 | if (s1 >= last) | |
277 | d1 = s1 - last; | |
278 | else if (s1 + back_max >= last) | |
279 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
280 | else | |
e8a99053 | 281 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
282 | |
283 | if (s2 >= last) | |
284 | d2 = s2 - last; | |
285 | else if (s2 + back_max >= last) | |
286 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
287 | else | |
e8a99053 | 288 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
289 | |
290 | /* Found required data */ | |
e8a99053 AM |
291 | |
292 | /* | |
293 | * By doing switch() on the bit mask "wrap" we avoid having to | |
294 | * check two variables for all permutations: --> faster! | |
295 | */ | |
296 | switch (wrap) { | |
5e705374 | 297 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 298 | if (d1 < d2) |
5e705374 | 299 | return rq1; |
e8a99053 | 300 | else if (d2 < d1) |
5e705374 | 301 | return rq2; |
e8a99053 AM |
302 | else { |
303 | if (s1 >= s2) | |
5e705374 | 304 | return rq1; |
e8a99053 | 305 | else |
5e705374 | 306 | return rq2; |
e8a99053 | 307 | } |
1da177e4 | 308 | |
e8a99053 | 309 | case CFQ_RQ2_WRAP: |
5e705374 | 310 | return rq1; |
e8a99053 | 311 | case CFQ_RQ1_WRAP: |
5e705374 JA |
312 | return rq2; |
313 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
314 | default: |
315 | /* | |
316 | * Since both rqs are wrapped, | |
317 | * start with the one that's further behind head | |
318 | * (--> only *one* back seek required), | |
319 | * since back seek takes more time than forward. | |
320 | */ | |
321 | if (s1 <= s2) | |
5e705374 | 322 | return rq1; |
1da177e4 | 323 | else |
5e705374 | 324 | return rq2; |
1da177e4 LT |
325 | } |
326 | } | |
327 | ||
328 | /* | |
329 | * would be nice to take fifo expire time into account as well | |
330 | */ | |
5e705374 JA |
331 | static struct request * |
332 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
333 | struct request *last) | |
1da177e4 | 334 | { |
21183b07 JA |
335 | struct rb_node *rbnext = rb_next(&last->rb_node); |
336 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 337 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 338 | |
21183b07 | 339 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
340 | |
341 | if (rbprev) | |
5e705374 | 342 | prev = rb_entry_rq(rbprev); |
1da177e4 | 343 | |
21183b07 | 344 | if (rbnext) |
5e705374 | 345 | next = rb_entry_rq(rbnext); |
21183b07 JA |
346 | else { |
347 | rbnext = rb_first(&cfqq->sort_list); | |
348 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 349 | next = rb_entry_rq(rbnext); |
21183b07 | 350 | } |
1da177e4 | 351 | |
21183b07 | 352 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
353 | } |
354 | ||
22e2c507 | 355 | static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted) |
1da177e4 | 356 | { |
22e2c507 | 357 | struct cfq_data *cfqd = cfqq->cfqd; |
53b03744 | 358 | struct list_head *list; |
1da177e4 | 359 | |
3b18152c | 360 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 361 | |
22e2c507 | 362 | list_del(&cfqq->cfq_list); |
1da177e4 | 363 | |
22e2c507 JA |
364 | if (cfq_class_rt(cfqq)) |
365 | list = &cfqd->cur_rr; | |
366 | else if (cfq_class_idle(cfqq)) | |
367 | list = &cfqd->idle_rr; | |
368 | else { | |
369 | /* | |
370 | * if cfqq has requests in flight, don't allow it to be | |
371 | * found in cfq_set_active_queue before it has finished them. | |
372 | * this is done to increase fairness between a process that | |
373 | * has lots of io pending vs one that only generates one | |
374 | * sporadically or synchronously | |
375 | */ | |
3b18152c | 376 | if (cfq_cfqq_dispatched(cfqq)) |
22e2c507 JA |
377 | list = &cfqd->busy_rr; |
378 | else | |
379 | list = &cfqd->rr_list[cfqq->ioprio]; | |
1da177e4 LT |
380 | } |
381 | ||
22e2c507 | 382 | /* |
53b03744 JA |
383 | * If this queue was preempted or is new (never been serviced), let |
384 | * it be added first for fairness but beind other new queues. | |
385 | * Otherwise, just add to the back of the list. | |
22e2c507 | 386 | */ |
53b03744 JA |
387 | if (preempted || cfq_cfqq_queue_new(cfqq)) { |
388 | struct list_head *n = list; | |
389 | struct cfq_queue *__cfqq; | |
b52a8348 | 390 | |
53b03744 JA |
391 | while (n->next != list) { |
392 | __cfqq = list_entry_cfqq(n->next); | |
393 | if (!cfq_cfqq_queue_new(__cfqq)) | |
394 | break; | |
1da177e4 | 395 | |
53b03744 JA |
396 | n = n->next; |
397 | } | |
1da177e4 | 398 | |
53b03744 | 399 | list = n; |
1da177e4 LT |
400 | } |
401 | ||
53b03744 | 402 | list_add_tail(&cfqq->cfq_list, list); |
1da177e4 LT |
403 | } |
404 | ||
405 | /* | |
406 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 407 | * the pending list according to last request service |
1da177e4 LT |
408 | */ |
409 | static inline void | |
b4878f24 | 410 | cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 411 | { |
3b18152c JA |
412 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
413 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
414 | cfqd->busy_queues++; |
415 | ||
b4878f24 | 416 | cfq_resort_rr_list(cfqq, 0); |
1da177e4 LT |
417 | } |
418 | ||
419 | static inline void | |
420 | cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
421 | { | |
3b18152c JA |
422 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
423 | cfq_clear_cfqq_on_rr(cfqq); | |
981a7973 | 424 | list_del_init(&cfqq->cfq_list); |
1da177e4 LT |
425 | |
426 | BUG_ON(!cfqd->busy_queues); | |
427 | cfqd->busy_queues--; | |
428 | } | |
429 | ||
430 | /* | |
431 | * rb tree support functions | |
432 | */ | |
5e705374 | 433 | static inline void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 434 | { |
5e705374 | 435 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 436 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 437 | const int sync = rq_is_sync(rq); |
1da177e4 | 438 | |
b4878f24 JA |
439 | BUG_ON(!cfqq->queued[sync]); |
440 | cfqq->queued[sync]--; | |
1da177e4 | 441 | |
5e705374 | 442 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 443 | |
dd67d051 | 444 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 445 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
446 | } |
447 | ||
5e705374 | 448 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 449 | { |
5e705374 | 450 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 451 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 452 | struct request *__alias; |
1da177e4 | 453 | |
5380a101 | 454 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
455 | |
456 | /* | |
457 | * looks a little odd, but the first insert might return an alias. | |
458 | * if that happens, put the alias on the dispatch list | |
459 | */ | |
21183b07 | 460 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 461 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
462 | |
463 | if (!cfq_cfqq_on_rr(cfqq)) | |
464 | cfq_add_cfqq_rr(cfqd, cfqq); | |
1da177e4 LT |
465 | } |
466 | ||
467 | static inline void | |
5e705374 | 468 | cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 469 | { |
5380a101 JA |
470 | elv_rb_del(&cfqq->sort_list, rq); |
471 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 472 | cfq_add_rq_rb(rq); |
1da177e4 LT |
473 | } |
474 | ||
206dc69b JA |
475 | static struct request * |
476 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 477 | { |
206dc69b | 478 | struct task_struct *tsk = current; |
7749a8d4 | 479 | pid_t key = cfq_queue_pid(tsk, bio_data_dir(bio), bio_sync(bio)); |
206dc69b | 480 | struct cfq_queue *cfqq; |
1da177e4 | 481 | |
206dc69b | 482 | cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio); |
89850f7e JA |
483 | if (cfqq) { |
484 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
485 | ||
21183b07 | 486 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 487 | } |
1da177e4 | 488 | |
1da177e4 LT |
489 | return NULL; |
490 | } | |
491 | ||
b4878f24 | 492 | static void cfq_activate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 493 | { |
22e2c507 | 494 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 495 | |
b4878f24 | 496 | cfqd->rq_in_driver++; |
25776e35 JA |
497 | |
498 | /* | |
499 | * If the depth is larger 1, it really could be queueing. But lets | |
500 | * make the mark a little higher - idling could still be good for | |
501 | * low queueing, and a low queueing number could also just indicate | |
502 | * a SCSI mid layer like behaviour where limit+1 is often seen. | |
503 | */ | |
504 | if (!cfqd->hw_tag && cfqd->rq_in_driver > 4) | |
505 | cfqd->hw_tag = 1; | |
1da177e4 LT |
506 | } |
507 | ||
b4878f24 | 508 | static void cfq_deactivate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 509 | { |
b4878f24 JA |
510 | struct cfq_data *cfqd = q->elevator->elevator_data; |
511 | ||
512 | WARN_ON(!cfqd->rq_in_driver); | |
513 | cfqd->rq_in_driver--; | |
1da177e4 LT |
514 | } |
515 | ||
b4878f24 | 516 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 517 | { |
5e705374 | 518 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 519 | |
5e705374 JA |
520 | if (cfqq->next_rq == rq) |
521 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 522 | |
b4878f24 | 523 | list_del_init(&rq->queuelist); |
5e705374 | 524 | cfq_del_rq_rb(rq); |
374f84ac JA |
525 | |
526 | if (rq_is_meta(rq)) { | |
527 | WARN_ON(!cfqq->meta_pending); | |
528 | cfqq->meta_pending--; | |
529 | } | |
1da177e4 LT |
530 | } |
531 | ||
532 | static int | |
533 | cfq_merge(request_queue_t *q, struct request **req, struct bio *bio) | |
534 | { | |
535 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
536 | struct request *__rq; | |
1da177e4 | 537 | |
206dc69b | 538 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 539 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
540 | *req = __rq; |
541 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
542 | } |
543 | ||
544 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
545 | } |
546 | ||
21183b07 JA |
547 | static void cfq_merged_request(request_queue_t *q, struct request *req, |
548 | int type) | |
1da177e4 | 549 | { |
21183b07 | 550 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 551 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 552 | |
5e705374 | 553 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 554 | } |
1da177e4 LT |
555 | } |
556 | ||
557 | static void | |
558 | cfq_merged_requests(request_queue_t *q, struct request *rq, | |
559 | struct request *next) | |
560 | { | |
22e2c507 JA |
561 | /* |
562 | * reposition in fifo if next is older than rq | |
563 | */ | |
564 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
565 | time_before(next->start_time, rq->start_time)) | |
566 | list_move(&rq->queuelist, &next->queuelist); | |
567 | ||
b4878f24 | 568 | cfq_remove_request(next); |
22e2c507 JA |
569 | } |
570 | ||
da775265 JA |
571 | static int cfq_allow_merge(request_queue_t *q, struct request *rq, |
572 | struct bio *bio) | |
573 | { | |
574 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
575 | const int rw = bio_data_dir(bio); | |
576 | struct cfq_queue *cfqq; | |
577 | pid_t key; | |
578 | ||
579 | /* | |
719d3402 | 580 | * Disallow merge, if bio and rq aren't both sync or async |
da775265 | 581 | */ |
719d3402 | 582 | if (!!bio_sync(bio) != !!rq_is_sync(rq)) |
da775265 JA |
583 | return 0; |
584 | ||
585 | /* | |
719d3402 JA |
586 | * Lookup the cfqq that this bio will be queued with. Allow |
587 | * merge only if rq is queued there. | |
da775265 | 588 | */ |
719d3402 | 589 | key = cfq_queue_pid(current, rw, bio_sync(bio)); |
da775265 | 590 | cfqq = cfq_find_cfq_hash(cfqd, key, current->ioprio); |
719d3402 JA |
591 | |
592 | if (cfqq == RQ_CFQQ(rq)) | |
593 | return 1; | |
da775265 JA |
594 | |
595 | return 1; | |
596 | } | |
597 | ||
22e2c507 JA |
598 | static inline void |
599 | __cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
600 | { | |
601 | if (cfqq) { | |
602 | /* | |
603 | * stop potential idle class queues waiting service | |
604 | */ | |
605 | del_timer(&cfqd->idle_class_timer); | |
606 | ||
607 | cfqq->slice_start = jiffies; | |
608 | cfqq->slice_end = 0; | |
609 | cfqq->slice_left = 0; | |
3b18152c JA |
610 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
611 | cfq_clear_cfqq_fifo_expire(cfqq); | |
22e2c507 JA |
612 | } |
613 | ||
614 | cfqd->active_queue = cfqq; | |
615 | } | |
616 | ||
7b14e3b5 JA |
617 | /* |
618 | * current cfqq expired its slice (or was too idle), select new one | |
619 | */ | |
620 | static void | |
621 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
622 | int preempted) | |
623 | { | |
624 | unsigned long now = jiffies; | |
625 | ||
626 | if (cfq_cfqq_wait_request(cfqq)) | |
627 | del_timer(&cfqd->idle_slice_timer); | |
628 | ||
53b03744 | 629 | if (!preempted && !cfq_cfqq_dispatched(cfqq)) |
7b14e3b5 | 630 | cfq_schedule_dispatch(cfqd); |
7b14e3b5 JA |
631 | |
632 | cfq_clear_cfqq_must_dispatch(cfqq); | |
633 | cfq_clear_cfqq_wait_request(cfqq); | |
53b03744 | 634 | cfq_clear_cfqq_queue_new(cfqq); |
7b14e3b5 JA |
635 | |
636 | /* | |
637 | * store what was left of this slice, if the queue idled out | |
638 | * or was preempted | |
639 | */ | |
640 | if (time_after(cfqq->slice_end, now)) | |
641 | cfqq->slice_left = cfqq->slice_end - now; | |
642 | else | |
643 | cfqq->slice_left = 0; | |
644 | ||
645 | if (cfq_cfqq_on_rr(cfqq)) | |
646 | cfq_resort_rr_list(cfqq, preempted); | |
647 | ||
648 | if (cfqq == cfqd->active_queue) | |
649 | cfqd->active_queue = NULL; | |
650 | ||
651 | if (cfqd->active_cic) { | |
652 | put_io_context(cfqd->active_cic->ioc); | |
653 | cfqd->active_cic = NULL; | |
654 | } | |
655 | ||
656 | cfqd->dispatch_slice = 0; | |
657 | } | |
658 | ||
659 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted) | |
660 | { | |
661 | struct cfq_queue *cfqq = cfqd->active_queue; | |
662 | ||
663 | if (cfqq) | |
664 | __cfq_slice_expired(cfqd, cfqq, preempted); | |
665 | } | |
666 | ||
22e2c507 JA |
667 | /* |
668 | * 0 | |
669 | * 0,1 | |
670 | * 0,1,2 | |
671 | * 0,1,2,3 | |
672 | * 0,1,2,3,4 | |
673 | * 0,1,2,3,4,5 | |
674 | * 0,1,2,3,4,5,6 | |
675 | * 0,1,2,3,4,5,6,7 | |
676 | */ | |
677 | static int cfq_get_next_prio_level(struct cfq_data *cfqd) | |
678 | { | |
679 | int prio, wrap; | |
680 | ||
681 | prio = -1; | |
682 | wrap = 0; | |
683 | do { | |
684 | int p; | |
685 | ||
686 | for (p = cfqd->cur_prio; p <= cfqd->cur_end_prio; p++) { | |
687 | if (!list_empty(&cfqd->rr_list[p])) { | |
688 | prio = p; | |
689 | break; | |
690 | } | |
691 | } | |
692 | ||
693 | if (prio != -1) | |
694 | break; | |
695 | cfqd->cur_prio = 0; | |
696 | if (++cfqd->cur_end_prio == CFQ_PRIO_LISTS) { | |
697 | cfqd->cur_end_prio = 0; | |
698 | if (wrap) | |
699 | break; | |
700 | wrap = 1; | |
1da177e4 | 701 | } |
22e2c507 JA |
702 | } while (1); |
703 | ||
704 | if (unlikely(prio == -1)) | |
705 | return -1; | |
706 | ||
707 | BUG_ON(prio >= CFQ_PRIO_LISTS); | |
708 | ||
709 | list_splice_init(&cfqd->rr_list[prio], &cfqd->cur_rr); | |
710 | ||
711 | cfqd->cur_prio = prio + 1; | |
712 | if (cfqd->cur_prio > cfqd->cur_end_prio) { | |
713 | cfqd->cur_end_prio = cfqd->cur_prio; | |
714 | cfqd->cur_prio = 0; | |
715 | } | |
716 | if (cfqd->cur_end_prio == CFQ_PRIO_LISTS) { | |
717 | cfqd->cur_prio = 0; | |
718 | cfqd->cur_end_prio = 0; | |
1da177e4 LT |
719 | } |
720 | ||
22e2c507 JA |
721 | return prio; |
722 | } | |
723 | ||
3b18152c | 724 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
22e2c507 | 725 | { |
7b14e3b5 | 726 | struct cfq_queue *cfqq = NULL; |
22e2c507 | 727 | |
89850f7e JA |
728 | if (!list_empty(&cfqd->cur_rr) || cfq_get_next_prio_level(cfqd) != -1) { |
729 | /* | |
730 | * if current list is non-empty, grab first entry. if it is | |
731 | * empty, get next prio level and grab first entry then if any | |
732 | * are spliced | |
733 | */ | |
22e2c507 | 734 | cfqq = list_entry_cfqq(cfqd->cur_rr.next); |
89850f7e JA |
735 | } else if (!list_empty(&cfqd->busy_rr)) { |
736 | /* | |
737 | * If no new queues are available, check if the busy list has | |
738 | * some before falling back to idle io. | |
739 | */ | |
e0de0206 | 740 | cfqq = list_entry_cfqq(cfqd->busy_rr.next); |
89850f7e JA |
741 | } else if (!list_empty(&cfqd->idle_rr)) { |
742 | /* | |
743 | * if we have idle queues and no rt or be queues had pending | |
744 | * requests, either allow immediate service if the grace period | |
745 | * has passed or arm the idle grace timer | |
746 | */ | |
22e2c507 JA |
747 | unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE; |
748 | ||
749 | if (time_after_eq(jiffies, end)) | |
750 | cfqq = list_entry_cfqq(cfqd->idle_rr.next); | |
751 | else | |
752 | mod_timer(&cfqd->idle_class_timer, end); | |
753 | } | |
754 | ||
755 | __cfq_set_active_queue(cfqd, cfqq); | |
3b18152c | 756 | return cfqq; |
22e2c507 JA |
757 | } |
758 | ||
caaa5f9f JA |
759 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (128 * 1024)) |
760 | ||
22e2c507 JA |
761 | static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
762 | ||
763 | { | |
206dc69b | 764 | struct cfq_io_context *cic; |
7b14e3b5 JA |
765 | unsigned long sl; |
766 | ||
dd67d051 | 767 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
768 | WARN_ON(cfqq != cfqd->active_queue); |
769 | ||
770 | /* | |
771 | * idle is disabled, either manually or by past process history | |
772 | */ | |
773 | if (!cfqd->cfq_slice_idle) | |
774 | return 0; | |
3b18152c | 775 | if (!cfq_cfqq_idle_window(cfqq)) |
22e2c507 JA |
776 | return 0; |
777 | /* | |
778 | * task has exited, don't wait | |
779 | */ | |
206dc69b JA |
780 | cic = cfqd->active_cic; |
781 | if (!cic || !cic->ioc->task) | |
22e2c507 JA |
782 | return 0; |
783 | ||
3b18152c JA |
784 | cfq_mark_cfqq_must_dispatch(cfqq); |
785 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 786 | |
7b14e3b5 | 787 | sl = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle); |
206dc69b JA |
788 | |
789 | /* | |
790 | * we don't want to idle for seeks, but we do want to allow | |
791 | * fair distribution of slice time for a process doing back-to-back | |
792 | * seeks. so allow a little bit of time for him to submit a new rq | |
793 | */ | |
caaa5f9f | 794 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
44eb1231 | 795 | sl = min(sl, msecs_to_jiffies(2)); |
206dc69b | 796 | |
7b14e3b5 | 797 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
22e2c507 | 798 | return 1; |
1da177e4 LT |
799 | } |
800 | ||
5e705374 | 801 | static void cfq_dispatch_insert(request_queue_t *q, struct request *rq) |
1da177e4 LT |
802 | { |
803 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
5e705374 | 804 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 805 | |
5380a101 JA |
806 | cfq_remove_request(rq); |
807 | cfqq->on_dispatch[rq_is_sync(rq)]++; | |
808 | elv_dispatch_sort(q, rq); | |
fd61af03 JA |
809 | |
810 | rq = list_entry(q->queue_head.prev, struct request, queuelist); | |
811 | cfqd->last_sector = rq->sector + rq->nr_sectors; | |
1da177e4 LT |
812 | } |
813 | ||
814 | /* | |
815 | * return expired entry, or NULL to just start from scratch in rbtree | |
816 | */ | |
5e705374 | 817 | static inline struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
818 | { |
819 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 820 | struct request *rq; |
89850f7e | 821 | int fifo; |
1da177e4 | 822 | |
3b18152c | 823 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 824 | return NULL; |
89850f7e JA |
825 | if (list_empty(&cfqq->fifo)) |
826 | return NULL; | |
1da177e4 | 827 | |
89850f7e JA |
828 | fifo = cfq_cfqq_class_sync(cfqq); |
829 | rq = rq_entry_fifo(cfqq->fifo.next); | |
1da177e4 | 830 | |
89850f7e JA |
831 | if (time_after(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) { |
832 | cfq_mark_cfqq_fifo_expire(cfqq); | |
833 | return rq; | |
1da177e4 LT |
834 | } |
835 | ||
836 | return NULL; | |
837 | } | |
838 | ||
839 | /* | |
3b18152c JA |
840 | * Scale schedule slice based on io priority. Use the sync time slice only |
841 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
842 | * io only, should not get full sync slice length. | |
1da177e4 | 843 | */ |
22e2c507 JA |
844 | static inline int |
845 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
846 | { | |
847 | const int base_slice = cfqd->cfq_slice[cfq_cfqq_sync(cfqq)]; | |
848 | ||
849 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); | |
850 | ||
851 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - cfqq->ioprio)); | |
852 | } | |
853 | ||
1da177e4 | 854 | static inline void |
22e2c507 | 855 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 856 | { |
22e2c507 JA |
857 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; |
858 | } | |
1da177e4 | 859 | |
22e2c507 JA |
860 | static inline int |
861 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
862 | { | |
863 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 864 | |
22e2c507 | 865 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 866 | |
22e2c507 | 867 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
868 | } |
869 | ||
22e2c507 JA |
870 | /* |
871 | * get next queue for service | |
872 | */ | |
1b5ed5e1 | 873 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 874 | { |
22e2c507 | 875 | unsigned long now = jiffies; |
1da177e4 | 876 | struct cfq_queue *cfqq; |
1da177e4 | 877 | |
22e2c507 JA |
878 | cfqq = cfqd->active_queue; |
879 | if (!cfqq) | |
880 | goto new_queue; | |
1da177e4 | 881 | |
22e2c507 JA |
882 | /* |
883 | * slice has expired | |
884 | */ | |
3b18152c JA |
885 | if (!cfq_cfqq_must_dispatch(cfqq) && time_after(now, cfqq->slice_end)) |
886 | goto expire; | |
1da177e4 | 887 | |
22e2c507 JA |
888 | /* |
889 | * if queue has requests, dispatch one. if not, check if | |
890 | * enough slice is left to wait for one | |
891 | */ | |
dd67d051 | 892 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 893 | goto keep_queue; |
caaa5f9f JA |
894 | else if (cfq_cfqq_dispatched(cfqq)) { |
895 | cfqq = NULL; | |
896 | goto keep_queue; | |
897 | } else if (cfq_cfqq_class_sync(cfqq)) { | |
22e2c507 JA |
898 | if (cfq_arm_slice_timer(cfqd, cfqq)) |
899 | return NULL; | |
900 | } | |
901 | ||
3b18152c | 902 | expire: |
22e2c507 | 903 | cfq_slice_expired(cfqd, 0); |
3b18152c JA |
904 | new_queue: |
905 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 906 | keep_queue: |
3b18152c | 907 | return cfqq; |
22e2c507 JA |
908 | } |
909 | ||
910 | static int | |
911 | __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
912 | int max_dispatch) | |
913 | { | |
914 | int dispatched = 0; | |
915 | ||
dd67d051 | 916 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
917 | |
918 | do { | |
5e705374 | 919 | struct request *rq; |
1da177e4 LT |
920 | |
921 | /* | |
22e2c507 | 922 | * follow expired path, else get first next available |
1da177e4 | 923 | */ |
5e705374 JA |
924 | if ((rq = cfq_check_fifo(cfqq)) == NULL) |
925 | rq = cfqq->next_rq; | |
22e2c507 JA |
926 | |
927 | /* | |
928 | * finally, insert request into driver dispatch list | |
929 | */ | |
5e705374 | 930 | cfq_dispatch_insert(cfqd->queue, rq); |
1da177e4 | 931 | |
22e2c507 JA |
932 | cfqd->dispatch_slice++; |
933 | dispatched++; | |
1da177e4 | 934 | |
22e2c507 | 935 | if (!cfqd->active_cic) { |
5e705374 JA |
936 | atomic_inc(&RQ_CIC(rq)->ioc->refcount); |
937 | cfqd->active_cic = RQ_CIC(rq); | |
22e2c507 | 938 | } |
1da177e4 | 939 | |
dd67d051 | 940 | if (RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 JA |
941 | break; |
942 | ||
943 | } while (dispatched < max_dispatch); | |
944 | ||
945 | /* | |
caaa5f9f | 946 | * if slice end isn't set yet, set it. |
22e2c507 JA |
947 | */ |
948 | if (!cfqq->slice_end) | |
949 | cfq_set_prio_slice(cfqd, cfqq); | |
950 | ||
951 | /* | |
952 | * expire an async queue immediately if it has used up its slice. idle | |
953 | * queue always expire after 1 dispatch round. | |
954 | */ | |
955 | if ((!cfq_cfqq_sync(cfqq) && | |
956 | cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
caaa5f9f JA |
957 | cfq_class_idle(cfqq) || |
958 | !cfq_cfqq_idle_window(cfqq)) | |
22e2c507 JA |
959 | cfq_slice_expired(cfqd, 0); |
960 | ||
961 | return dispatched; | |
962 | } | |
963 | ||
1b5ed5e1 TH |
964 | static int |
965 | cfq_forced_dispatch_cfqqs(struct list_head *list) | |
966 | { | |
1b5ed5e1 | 967 | struct cfq_queue *cfqq, *next; |
caaa5f9f | 968 | int dispatched; |
1b5ed5e1 | 969 | |
caaa5f9f | 970 | dispatched = 0; |
1b5ed5e1 | 971 | list_for_each_entry_safe(cfqq, next, list, cfq_list) { |
5e705374 JA |
972 | while (cfqq->next_rq) { |
973 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1b5ed5e1 TH |
974 | dispatched++; |
975 | } | |
976 | BUG_ON(!list_empty(&cfqq->fifo)); | |
977 | } | |
caaa5f9f | 978 | |
1b5ed5e1 TH |
979 | return dispatched; |
980 | } | |
981 | ||
982 | static int | |
983 | cfq_forced_dispatch(struct cfq_data *cfqd) | |
984 | { | |
985 | int i, dispatched = 0; | |
986 | ||
987 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
988 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->rr_list[i]); | |
989 | ||
990 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->busy_rr); | |
991 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->cur_rr); | |
992 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->idle_rr); | |
993 | ||
994 | cfq_slice_expired(cfqd, 0); | |
995 | ||
996 | BUG_ON(cfqd->busy_queues); | |
997 | ||
998 | return dispatched; | |
999 | } | |
1000 | ||
22e2c507 | 1001 | static int |
b4878f24 | 1002 | cfq_dispatch_requests(request_queue_t *q, int force) |
22e2c507 JA |
1003 | { |
1004 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
caaa5f9f JA |
1005 | struct cfq_queue *cfqq, *prev_cfqq; |
1006 | int dispatched; | |
22e2c507 JA |
1007 | |
1008 | if (!cfqd->busy_queues) | |
1009 | return 0; | |
1010 | ||
1b5ed5e1 TH |
1011 | if (unlikely(force)) |
1012 | return cfq_forced_dispatch(cfqd); | |
1013 | ||
caaa5f9f JA |
1014 | dispatched = 0; |
1015 | prev_cfqq = NULL; | |
1016 | while ((cfqq = cfq_select_queue(cfqd)) != NULL) { | |
b4878f24 JA |
1017 | int max_dispatch; |
1018 | ||
caaa5f9f JA |
1019 | /* |
1020 | * Don't repeat dispatch from the previous queue. | |
1021 | */ | |
1022 | if (prev_cfqq == cfqq) | |
1023 | break; | |
1024 | ||
3b18152c JA |
1025 | cfq_clear_cfqq_must_dispatch(cfqq); |
1026 | cfq_clear_cfqq_wait_request(cfqq); | |
22e2c507 JA |
1027 | del_timer(&cfqd->idle_slice_timer); |
1028 | ||
1b5ed5e1 TH |
1029 | max_dispatch = cfqd->cfq_quantum; |
1030 | if (cfq_class_idle(cfqq)) | |
1031 | max_dispatch = 1; | |
1da177e4 | 1032 | |
caaa5f9f JA |
1033 | dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); |
1034 | ||
1035 | /* | |
1036 | * If the dispatch cfqq has idling enabled and is still | |
1037 | * the active queue, break out. | |
1038 | */ | |
1039 | if (cfq_cfqq_idle_window(cfqq) && cfqd->active_queue) | |
1040 | break; | |
1041 | ||
1042 | prev_cfqq = cfqq; | |
1da177e4 LT |
1043 | } |
1044 | ||
caaa5f9f | 1045 | return dispatched; |
1da177e4 LT |
1046 | } |
1047 | ||
1da177e4 | 1048 | /* |
5e705374 JA |
1049 | * task holds one reference to the queue, dropped when task exits. each rq |
1050 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1051 | * |
1052 | * queue lock must be held here. | |
1053 | */ | |
1054 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1055 | { | |
22e2c507 JA |
1056 | struct cfq_data *cfqd = cfqq->cfqd; |
1057 | ||
1058 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1059 | |
1060 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1061 | return; | |
1062 | ||
1063 | BUG_ON(rb_first(&cfqq->sort_list)); | |
22e2c507 | 1064 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1065 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1066 | |
7b14e3b5 | 1067 | if (unlikely(cfqd->active_queue == cfqq)) |
3b18152c | 1068 | __cfq_slice_expired(cfqd, cfqq, 0); |
22e2c507 | 1069 | |
1da177e4 LT |
1070 | /* |
1071 | * it's on the empty list and still hashed | |
1072 | */ | |
1073 | list_del(&cfqq->cfq_list); | |
1074 | hlist_del(&cfqq->cfq_hash); | |
1075 | kmem_cache_free(cfq_pool, cfqq); | |
1076 | } | |
1077 | ||
1ea25ecb | 1078 | static struct cfq_queue * |
3b18152c JA |
1079 | __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio, |
1080 | const int hashval) | |
1da177e4 LT |
1081 | { |
1082 | struct hlist_head *hash_list = &cfqd->cfq_hash[hashval]; | |
206dc69b JA |
1083 | struct hlist_node *entry; |
1084 | struct cfq_queue *__cfqq; | |
1da177e4 | 1085 | |
206dc69b | 1086 | hlist_for_each_entry(__cfqq, entry, hash_list, cfq_hash) { |
b0a6916b | 1087 | const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio); |
1da177e4 | 1088 | |
206dc69b | 1089 | if (__cfqq->key == key && (__p == prio || !prio)) |
1da177e4 LT |
1090 | return __cfqq; |
1091 | } | |
1092 | ||
1093 | return NULL; | |
1094 | } | |
1095 | ||
1096 | static struct cfq_queue * | |
3b18152c | 1097 | cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio) |
1da177e4 | 1098 | { |
3b18152c | 1099 | return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT)); |
1da177e4 LT |
1100 | } |
1101 | ||
e2d74ac0 | 1102 | static void cfq_free_io_context(struct io_context *ioc) |
1da177e4 | 1103 | { |
22e2c507 | 1104 | struct cfq_io_context *__cic; |
e2d74ac0 JA |
1105 | struct rb_node *n; |
1106 | int freed = 0; | |
1da177e4 | 1107 | |
e2d74ac0 JA |
1108 | while ((n = rb_first(&ioc->cic_root)) != NULL) { |
1109 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1110 | rb_erase(&__cic->rb_node, &ioc->cic_root); | |
22e2c507 | 1111 | kmem_cache_free(cfq_ioc_pool, __cic); |
334e94de | 1112 | freed++; |
1da177e4 LT |
1113 | } |
1114 | ||
4050cf16 JA |
1115 | elv_ioc_count_mod(ioc_count, -freed); |
1116 | ||
1117 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) | |
334e94de | 1118 | complete(ioc_gone); |
1da177e4 LT |
1119 | } |
1120 | ||
89850f7e | 1121 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1122 | { |
89850f7e JA |
1123 | if (unlikely(cfqq == cfqd->active_queue)) |
1124 | __cfq_slice_expired(cfqd, cfqq, 0); | |
22e2c507 | 1125 | |
89850f7e JA |
1126 | cfq_put_queue(cfqq); |
1127 | } | |
22e2c507 | 1128 | |
89850f7e JA |
1129 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1130 | struct cfq_io_context *cic) | |
1131 | { | |
fc46379d JA |
1132 | list_del_init(&cic->queue_list); |
1133 | smp_wmb(); | |
1134 | cic->key = NULL; | |
1135 | ||
12a05732 | 1136 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1137 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1138 | cic->cfqq[ASYNC] = NULL; |
1139 | } | |
1140 | ||
1141 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1142 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1143 | cic->cfqq[SYNC] = NULL; |
1144 | } | |
89850f7e JA |
1145 | } |
1146 | ||
1147 | ||
1148 | /* | |
1149 | * Called with interrupts disabled | |
1150 | */ | |
1151 | static void cfq_exit_single_io_context(struct cfq_io_context *cic) | |
1152 | { | |
1153 | struct cfq_data *cfqd = cic->key; | |
1154 | ||
89850f7e JA |
1155 | if (cfqd) { |
1156 | request_queue_t *q = cfqd->queue; | |
1157 | ||
fc46379d | 1158 | spin_lock_irq(q->queue_lock); |
89850f7e | 1159 | __cfq_exit_single_io_context(cfqd, cic); |
fc46379d | 1160 | spin_unlock_irq(q->queue_lock); |
89850f7e | 1161 | } |
1da177e4 LT |
1162 | } |
1163 | ||
e2d74ac0 | 1164 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1165 | { |
22e2c507 | 1166 | struct cfq_io_context *__cic; |
e2d74ac0 | 1167 | struct rb_node *n; |
22e2c507 | 1168 | |
1da177e4 LT |
1169 | /* |
1170 | * put the reference this task is holding to the various queues | |
1171 | */ | |
e2d74ac0 JA |
1172 | |
1173 | n = rb_first(&ioc->cic_root); | |
1174 | while (n != NULL) { | |
1175 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1176 | ||
22e2c507 | 1177 | cfq_exit_single_io_context(__cic); |
e2d74ac0 | 1178 | n = rb_next(n); |
1da177e4 | 1179 | } |
1da177e4 LT |
1180 | } |
1181 | ||
22e2c507 | 1182 | static struct cfq_io_context * |
8267e268 | 1183 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1184 | { |
b5deef90 | 1185 | struct cfq_io_context *cic; |
1da177e4 | 1186 | |
b5deef90 | 1187 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask, cfqd->queue->node); |
1da177e4 | 1188 | if (cic) { |
553698f9 | 1189 | memset(cic, 0, sizeof(*cic)); |
22e2c507 | 1190 | cic->last_end_request = jiffies; |
553698f9 | 1191 | INIT_LIST_HEAD(&cic->queue_list); |
22e2c507 JA |
1192 | cic->dtor = cfq_free_io_context; |
1193 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1194 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1195 | } |
1196 | ||
1197 | return cic; | |
1198 | } | |
1199 | ||
22e2c507 JA |
1200 | static void cfq_init_prio_data(struct cfq_queue *cfqq) |
1201 | { | |
1202 | struct task_struct *tsk = current; | |
1203 | int ioprio_class; | |
1204 | ||
3b18152c | 1205 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1206 | return; |
1207 | ||
1208 | ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio); | |
1209 | switch (ioprio_class) { | |
1210 | default: | |
1211 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1212 | case IOPRIO_CLASS_NONE: | |
1213 | /* | |
1214 | * no prio set, place us in the middle of the BE classes | |
1215 | */ | |
1216 | cfqq->ioprio = task_nice_ioprio(tsk); | |
1217 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1218 | break; | |
1219 | case IOPRIO_CLASS_RT: | |
1220 | cfqq->ioprio = task_ioprio(tsk); | |
1221 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1222 | break; | |
1223 | case IOPRIO_CLASS_BE: | |
1224 | cfqq->ioprio = task_ioprio(tsk); | |
1225 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1226 | break; | |
1227 | case IOPRIO_CLASS_IDLE: | |
1228 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1229 | cfqq->ioprio = 7; | |
3b18152c | 1230 | cfq_clear_cfqq_idle_window(cfqq); |
22e2c507 JA |
1231 | break; |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * keep track of original prio settings in case we have to temporarily | |
1236 | * elevate the priority of this queue | |
1237 | */ | |
1238 | cfqq->org_ioprio = cfqq->ioprio; | |
1239 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
1240 | ||
3b18152c | 1241 | if (cfq_cfqq_on_rr(cfqq)) |
22e2c507 JA |
1242 | cfq_resort_rr_list(cfqq, 0); |
1243 | ||
3b18152c | 1244 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1245 | } |
1246 | ||
478a82b0 | 1247 | static inline void changed_ioprio(struct cfq_io_context *cic) |
22e2c507 | 1248 | { |
478a82b0 AV |
1249 | struct cfq_data *cfqd = cic->key; |
1250 | struct cfq_queue *cfqq; | |
c1b707d2 | 1251 | unsigned long flags; |
35e6077c | 1252 | |
caaa5f9f JA |
1253 | if (unlikely(!cfqd)) |
1254 | return; | |
1255 | ||
c1b707d2 | 1256 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1257 | |
1258 | cfqq = cic->cfqq[ASYNC]; | |
1259 | if (cfqq) { | |
1260 | struct cfq_queue *new_cfqq; | |
1261 | new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC, cic->ioc->task, | |
1262 | GFP_ATOMIC); | |
1263 | if (new_cfqq) { | |
1264 | cic->cfqq[ASYNC] = new_cfqq; | |
1265 | cfq_put_queue(cfqq); | |
1266 | } | |
22e2c507 | 1267 | } |
caaa5f9f JA |
1268 | |
1269 | cfqq = cic->cfqq[SYNC]; | |
1270 | if (cfqq) | |
1271 | cfq_mark_cfqq_prio_changed(cfqq); | |
1272 | ||
c1b707d2 | 1273 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1274 | } |
1275 | ||
fc46379d | 1276 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1277 | { |
a6a0763a | 1278 | struct cfq_io_context *cic; |
e2d74ac0 | 1279 | struct rb_node *n; |
a6a0763a | 1280 | |
fc46379d | 1281 | ioc->ioprio_changed = 0; |
a6a0763a | 1282 | |
e2d74ac0 JA |
1283 | n = rb_first(&ioc->cic_root); |
1284 | while (n != NULL) { | |
1285 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
3793c65c | 1286 | |
478a82b0 | 1287 | changed_ioprio(cic); |
e2d74ac0 JA |
1288 | n = rb_next(n); |
1289 | } | |
22e2c507 JA |
1290 | } |
1291 | ||
1292 | static struct cfq_queue * | |
6f325a13 | 1293 | cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, |
8267e268 | 1294 | gfp_t gfp_mask) |
22e2c507 JA |
1295 | { |
1296 | const int hashval = hash_long(key, CFQ_QHASH_SHIFT); | |
1297 | struct cfq_queue *cfqq, *new_cfqq = NULL; | |
6f325a13 | 1298 | unsigned short ioprio; |
22e2c507 JA |
1299 | |
1300 | retry: | |
6f325a13 | 1301 | ioprio = tsk->ioprio; |
3b18152c | 1302 | cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval); |
22e2c507 JA |
1303 | |
1304 | if (!cfqq) { | |
1305 | if (new_cfqq) { | |
1306 | cfqq = new_cfqq; | |
1307 | new_cfqq = NULL; | |
1308 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1309 | /* |
1310 | * Inform the allocator of the fact that we will | |
1311 | * just repeat this allocation if it fails, to allow | |
1312 | * the allocator to do whatever it needs to attempt to | |
1313 | * free memory. | |
1314 | */ | |
22e2c507 | 1315 | spin_unlock_irq(cfqd->queue->queue_lock); |
b5deef90 | 1316 | new_cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask|__GFP_NOFAIL, cfqd->queue->node); |
22e2c507 JA |
1317 | spin_lock_irq(cfqd->queue->queue_lock); |
1318 | goto retry; | |
1319 | } else { | |
b5deef90 | 1320 | cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask, cfqd->queue->node); |
22e2c507 JA |
1321 | if (!cfqq) |
1322 | goto out; | |
1323 | } | |
1324 | ||
1325 | memset(cfqq, 0, sizeof(*cfqq)); | |
1326 | ||
1327 | INIT_HLIST_NODE(&cfqq->cfq_hash); | |
1328 | INIT_LIST_HEAD(&cfqq->cfq_list); | |
22e2c507 JA |
1329 | INIT_LIST_HEAD(&cfqq->fifo); |
1330 | ||
1331 | cfqq->key = key; | |
1332 | hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]); | |
1333 | atomic_set(&cfqq->ref, 0); | |
1334 | cfqq->cfqd = cfqd; | |
22e2c507 JA |
1335 | /* |
1336 | * set ->slice_left to allow preemption for a new process | |
1337 | */ | |
1338 | cfqq->slice_left = 2 * cfqd->cfq_slice_idle; | |
caaa5f9f | 1339 | cfq_mark_cfqq_idle_window(cfqq); |
3b18152c | 1340 | cfq_mark_cfqq_prio_changed(cfqq); |
53b03744 | 1341 | cfq_mark_cfqq_queue_new(cfqq); |
3b18152c | 1342 | cfq_init_prio_data(cfqq); |
22e2c507 JA |
1343 | } |
1344 | ||
1345 | if (new_cfqq) | |
1346 | kmem_cache_free(cfq_pool, new_cfqq); | |
1347 | ||
1348 | atomic_inc(&cfqq->ref); | |
1349 | out: | |
1350 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1351 | return cfqq; | |
1352 | } | |
1353 | ||
dbecf3ab OH |
1354 | static void |
1355 | cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic) | |
1356 | { | |
fc46379d | 1357 | WARN_ON(!list_empty(&cic->queue_list)); |
dbecf3ab | 1358 | rb_erase(&cic->rb_node, &ioc->cic_root); |
dbecf3ab | 1359 | kmem_cache_free(cfq_ioc_pool, cic); |
4050cf16 | 1360 | elv_ioc_count_dec(ioc_count); |
dbecf3ab OH |
1361 | } |
1362 | ||
e2d74ac0 JA |
1363 | static struct cfq_io_context * |
1364 | cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc) | |
1365 | { | |
dbecf3ab | 1366 | struct rb_node *n; |
e2d74ac0 | 1367 | struct cfq_io_context *cic; |
be3b0753 | 1368 | void *k, *key = cfqd; |
e2d74ac0 | 1369 | |
dbecf3ab OH |
1370 | restart: |
1371 | n = ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1372 | while (n) { |
1373 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1374 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1375 | k = cic->key; | |
1376 | if (unlikely(!k)) { | |
dbecf3ab OH |
1377 | cfq_drop_dead_cic(ioc, cic); |
1378 | goto restart; | |
1379 | } | |
e2d74ac0 | 1380 | |
be3b0753 | 1381 | if (key < k) |
e2d74ac0 | 1382 | n = n->rb_left; |
be3b0753 | 1383 | else if (key > k) |
e2d74ac0 JA |
1384 | n = n->rb_right; |
1385 | else | |
1386 | return cic; | |
1387 | } | |
1388 | ||
1389 | return NULL; | |
1390 | } | |
1391 | ||
1392 | static inline void | |
1393 | cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, | |
1394 | struct cfq_io_context *cic) | |
1395 | { | |
dbecf3ab OH |
1396 | struct rb_node **p; |
1397 | struct rb_node *parent; | |
e2d74ac0 | 1398 | struct cfq_io_context *__cic; |
0261d688 | 1399 | unsigned long flags; |
be3b0753 | 1400 | void *k; |
e2d74ac0 | 1401 | |
e2d74ac0 JA |
1402 | cic->ioc = ioc; |
1403 | cic->key = cfqd; | |
1404 | ||
dbecf3ab OH |
1405 | restart: |
1406 | parent = NULL; | |
1407 | p = &ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1408 | while (*p) { |
1409 | parent = *p; | |
1410 | __cic = rb_entry(parent, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1411 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1412 | k = __cic->key; | |
1413 | if (unlikely(!k)) { | |
be33c3a6 | 1414 | cfq_drop_dead_cic(ioc, __cic); |
dbecf3ab OH |
1415 | goto restart; |
1416 | } | |
e2d74ac0 | 1417 | |
be3b0753 | 1418 | if (cic->key < k) |
e2d74ac0 | 1419 | p = &(*p)->rb_left; |
be3b0753 | 1420 | else if (cic->key > k) |
e2d74ac0 JA |
1421 | p = &(*p)->rb_right; |
1422 | else | |
1423 | BUG(); | |
1424 | } | |
1425 | ||
1426 | rb_link_node(&cic->rb_node, parent, p); | |
1427 | rb_insert_color(&cic->rb_node, &ioc->cic_root); | |
fc46379d | 1428 | |
0261d688 | 1429 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
e2d74ac0 | 1430 | list_add(&cic->queue_list, &cfqd->cic_list); |
0261d688 | 1431 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
e2d74ac0 JA |
1432 | } |
1433 | ||
1da177e4 LT |
1434 | /* |
1435 | * Setup general io context and cfq io context. There can be several cfq | |
1436 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1437 | * than one device managed by cfq. |
1da177e4 LT |
1438 | */ |
1439 | static struct cfq_io_context * | |
e2d74ac0 | 1440 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1441 | { |
22e2c507 | 1442 | struct io_context *ioc = NULL; |
1da177e4 | 1443 | struct cfq_io_context *cic; |
1da177e4 | 1444 | |
22e2c507 | 1445 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1446 | |
b5deef90 | 1447 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1448 | if (!ioc) |
1449 | return NULL; | |
1450 | ||
e2d74ac0 JA |
1451 | cic = cfq_cic_rb_lookup(cfqd, ioc); |
1452 | if (cic) | |
1453 | goto out; | |
1da177e4 | 1454 | |
e2d74ac0 JA |
1455 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1456 | if (cic == NULL) | |
1457 | goto err; | |
1da177e4 | 1458 | |
e2d74ac0 | 1459 | cfq_cic_link(cfqd, ioc, cic); |
1da177e4 | 1460 | out: |
fc46379d JA |
1461 | smp_read_barrier_depends(); |
1462 | if (unlikely(ioc->ioprio_changed)) | |
1463 | cfq_ioc_set_ioprio(ioc); | |
1464 | ||
1da177e4 LT |
1465 | return cic; |
1466 | err: | |
1467 | put_io_context(ioc); | |
1468 | return NULL; | |
1469 | } | |
1470 | ||
22e2c507 JA |
1471 | static void |
1472 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1473 | { |
22e2c507 | 1474 | unsigned long elapsed, ttime; |
1da177e4 | 1475 | |
22e2c507 JA |
1476 | /* |
1477 | * if this context already has stuff queued, thinktime is from | |
1478 | * last queue not last end | |
1479 | */ | |
1480 | #if 0 | |
1481 | if (time_after(cic->last_end_request, cic->last_queue)) | |
1482 | elapsed = jiffies - cic->last_end_request; | |
1483 | else | |
1484 | elapsed = jiffies - cic->last_queue; | |
1485 | #else | |
1486 | elapsed = jiffies - cic->last_end_request; | |
1487 | #endif | |
1da177e4 | 1488 | |
22e2c507 | 1489 | ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); |
db3b5848 | 1490 | |
22e2c507 JA |
1491 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1492 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1493 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1494 | } | |
1da177e4 | 1495 | |
206dc69b | 1496 | static void |
bb37b94c | 1497 | cfq_update_io_seektime(struct cfq_io_context *cic, struct request *rq) |
206dc69b JA |
1498 | { |
1499 | sector_t sdist; | |
1500 | u64 total; | |
1501 | ||
5e705374 JA |
1502 | if (cic->last_request_pos < rq->sector) |
1503 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1504 | else |
5e705374 | 1505 | sdist = cic->last_request_pos - rq->sector; |
206dc69b JA |
1506 | |
1507 | /* | |
1508 | * Don't allow the seek distance to get too large from the | |
1509 | * odd fragment, pagein, etc | |
1510 | */ | |
1511 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1512 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1513 | else | |
1514 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1515 | ||
1516 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1517 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1518 | total = cic->seek_total + (cic->seek_samples/2); | |
1519 | do_div(total, cic->seek_samples); | |
1520 | cic->seek_mean = (sector_t)total; | |
1521 | } | |
1da177e4 | 1522 | |
22e2c507 JA |
1523 | /* |
1524 | * Disable idle window if the process thinks too long or seeks so much that | |
1525 | * it doesn't matter | |
1526 | */ | |
1527 | static void | |
1528 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1529 | struct cfq_io_context *cic) | |
1530 | { | |
3b18152c | 1531 | int enable_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 1532 | |
caaa5f9f JA |
1533 | if (!cic->ioc->task || !cfqd->cfq_slice_idle || |
1534 | (cfqd->hw_tag && CIC_SEEKY(cic))) | |
22e2c507 JA |
1535 | enable_idle = 0; |
1536 | else if (sample_valid(cic->ttime_samples)) { | |
1537 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1538 | enable_idle = 0; | |
1539 | else | |
1540 | enable_idle = 1; | |
1da177e4 LT |
1541 | } |
1542 | ||
3b18152c JA |
1543 | if (enable_idle) |
1544 | cfq_mark_cfqq_idle_window(cfqq); | |
1545 | else | |
1546 | cfq_clear_cfqq_idle_window(cfqq); | |
22e2c507 | 1547 | } |
1da177e4 | 1548 | |
22e2c507 JA |
1549 | |
1550 | /* | |
1551 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1552 | * no or if we aren't sure, a 1 will cause a preempt. | |
1553 | */ | |
1554 | static int | |
1555 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1556 | struct request *rq) |
22e2c507 JA |
1557 | { |
1558 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1559 | ||
1560 | if (cfq_class_idle(new_cfqq)) | |
1561 | return 0; | |
1562 | ||
1563 | if (!cfqq) | |
caaa5f9f | 1564 | return 0; |
22e2c507 JA |
1565 | |
1566 | if (cfq_class_idle(cfqq)) | |
1567 | return 1; | |
3b18152c | 1568 | if (!cfq_cfqq_wait_request(new_cfqq)) |
22e2c507 JA |
1569 | return 0; |
1570 | /* | |
1571 | * if it doesn't have slice left, forget it | |
1572 | */ | |
1573 | if (new_cfqq->slice_left < cfqd->cfq_slice_idle) | |
1574 | return 0; | |
374f84ac JA |
1575 | /* |
1576 | * if the new request is sync, but the currently running queue is | |
1577 | * not, let the sync request have priority. | |
1578 | */ | |
5e705374 | 1579 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1580 | return 1; |
374f84ac JA |
1581 | /* |
1582 | * So both queues are sync. Let the new request get disk time if | |
1583 | * it's a metadata request and the current queue is doing regular IO. | |
1584 | */ | |
1585 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1586 | return 1; | |
22e2c507 JA |
1587 | |
1588 | return 0; | |
1589 | } | |
1590 | ||
1591 | /* | |
1592 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1593 | * let it have half of its nominal slice. | |
1594 | */ | |
1595 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1596 | { | |
bf572256 | 1597 | cfq_slice_expired(cfqd, 1); |
22e2c507 JA |
1598 | |
1599 | if (!cfqq->slice_left) | |
1600 | cfqq->slice_left = cfq_prio_to_slice(cfqd, cfqq) / 2; | |
1601 | ||
bf572256 JA |
1602 | /* |
1603 | * Put the new queue at the front of the of the current list, | |
1604 | * so we know that it will be selected next. | |
1605 | */ | |
1606 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
1607 | list_move(&cfqq->cfq_list, &cfqd->cur_rr); | |
1608 | ||
22e2c507 | 1609 | cfqq->slice_end = cfqq->slice_left + jiffies; |
22e2c507 JA |
1610 | } |
1611 | ||
22e2c507 | 1612 | /* |
5e705374 | 1613 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1614 | * something we should do about it |
1615 | */ | |
1616 | static void | |
5e705374 JA |
1617 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1618 | struct request *rq) | |
22e2c507 | 1619 | { |
5e705374 | 1620 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1621 | |
374f84ac JA |
1622 | if (rq_is_meta(rq)) |
1623 | cfqq->meta_pending++; | |
1624 | ||
21183b07 | 1625 | /* |
5380a101 | 1626 | * check if this request is a better next-serve candidate)) { |
21183b07 | 1627 | */ |
5e705374 JA |
1628 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); |
1629 | BUG_ON(!cfqq->next_rq); | |
21183b07 | 1630 | |
9c2c38a1 JA |
1631 | /* |
1632 | * we never wait for an async request and we don't allow preemption | |
1633 | * of an async request. so just return early | |
1634 | */ | |
5e705374 | 1635 | if (!rq_is_sync(rq)) { |
12e9fddd JA |
1636 | /* |
1637 | * sync process issued an async request, if it's waiting | |
1638 | * then expire it and kick rq handling. | |
1639 | */ | |
1640 | if (cic == cfqd->active_cic && | |
1641 | del_timer(&cfqd->idle_slice_timer)) { | |
1642 | cfq_slice_expired(cfqd, 0); | |
dc72ef4a | 1643 | blk_start_queueing(cfqd->queue); |
12e9fddd | 1644 | } |
9c2c38a1 | 1645 | return; |
12e9fddd | 1646 | } |
22e2c507 | 1647 | |
9c2c38a1 | 1648 | cfq_update_io_thinktime(cfqd, cic); |
bb37b94c | 1649 | cfq_update_io_seektime(cic, rq); |
9c2c38a1 JA |
1650 | cfq_update_idle_window(cfqd, cfqq, cic); |
1651 | ||
1652 | cic->last_queue = jiffies; | |
5e705374 | 1653 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
22e2c507 JA |
1654 | |
1655 | if (cfqq == cfqd->active_queue) { | |
1656 | /* | |
1657 | * if we are waiting for a request for this queue, let it rip | |
1658 | * immediately and flag that we must not expire this queue | |
1659 | * just now | |
1660 | */ | |
3b18152c JA |
1661 | if (cfq_cfqq_wait_request(cfqq)) { |
1662 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 | 1663 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1664 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1665 | } |
5e705374 | 1666 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1667 | /* |
1668 | * not the active queue - expire current slice if it is | |
1669 | * idle and has expired it's mean thinktime or this new queue | |
1670 | * has some old slice time left and is of higher priority | |
1671 | */ | |
1672 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1673 | cfq_mark_cfqq_must_dispatch(cfqq); |
dc72ef4a | 1674 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1675 | } |
1da177e4 LT |
1676 | } |
1677 | ||
b4878f24 | 1678 | static void cfq_insert_request(request_queue_t *q, struct request *rq) |
1da177e4 | 1679 | { |
b4878f24 | 1680 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1681 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1682 | |
1683 | cfq_init_prio_data(cfqq); | |
1da177e4 | 1684 | |
5e705374 | 1685 | cfq_add_rq_rb(rq); |
1da177e4 | 1686 | |
22e2c507 JA |
1687 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1688 | ||
5e705374 | 1689 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1690 | } |
1691 | ||
1da177e4 LT |
1692 | static void cfq_completed_request(request_queue_t *q, struct request *rq) |
1693 | { | |
5e705374 | 1694 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1695 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1696 | const int sync = rq_is_sync(rq); |
b4878f24 | 1697 | unsigned long now; |
1da177e4 | 1698 | |
b4878f24 | 1699 | now = jiffies; |
1da177e4 | 1700 | |
b4878f24 JA |
1701 | WARN_ON(!cfqd->rq_in_driver); |
1702 | WARN_ON(!cfqq->on_dispatch[sync]); | |
1703 | cfqd->rq_in_driver--; | |
1704 | cfqq->on_dispatch[sync]--; | |
1da177e4 | 1705 | |
b4878f24 JA |
1706 | if (!cfq_class_idle(cfqq)) |
1707 | cfqd->last_end_request = now; | |
3b18152c | 1708 | |
53b03744 JA |
1709 | if (!cfq_cfqq_dispatched(cfqq) && cfq_cfqq_on_rr(cfqq)) |
1710 | cfq_resort_rr_list(cfqq, 0); | |
1da177e4 | 1711 | |
caaa5f9f | 1712 | if (sync) |
5e705374 | 1713 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1714 | |
1715 | /* | |
1716 | * If this is the active queue, check if it needs to be expired, | |
1717 | * or if we want to idle in case it has no pending requests. | |
1718 | */ | |
1719 | if (cfqd->active_queue == cfqq) { | |
1720 | if (time_after(now, cfqq->slice_end)) | |
1721 | cfq_slice_expired(cfqd, 0); | |
dd67d051 | 1722 | else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
caaa5f9f JA |
1723 | if (!cfq_arm_slice_timer(cfqd, cfqq)) |
1724 | cfq_schedule_dispatch(cfqd); | |
1725 | } | |
1726 | } | |
1da177e4 LT |
1727 | } |
1728 | ||
22e2c507 JA |
1729 | /* |
1730 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
1731 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
1732 | */ | |
1733 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 1734 | { |
22e2c507 JA |
1735 | const int ioprio_class = cfqq->ioprio_class; |
1736 | const int ioprio = cfqq->ioprio; | |
1da177e4 | 1737 | |
22e2c507 JA |
1738 | if (has_fs_excl()) { |
1739 | /* | |
1740 | * boost idle prio on transactions that would lock out other | |
1741 | * users of the filesystem | |
1742 | */ | |
1743 | if (cfq_class_idle(cfqq)) | |
1744 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1745 | if (cfqq->ioprio > IOPRIO_NORM) | |
1746 | cfqq->ioprio = IOPRIO_NORM; | |
1747 | } else { | |
1748 | /* | |
1749 | * check if we need to unboost the queue | |
1750 | */ | |
1751 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
1752 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
1753 | if (cfqq->ioprio != cfqq->org_ioprio) | |
1754 | cfqq->ioprio = cfqq->org_ioprio; | |
1755 | } | |
1da177e4 | 1756 | |
22e2c507 JA |
1757 | /* |
1758 | * refile between round-robin lists if we moved the priority class | |
1759 | */ | |
1760 | if ((ioprio_class != cfqq->ioprio_class || ioprio != cfqq->ioprio) && | |
3b18152c | 1761 | cfq_cfqq_on_rr(cfqq)) |
22e2c507 JA |
1762 | cfq_resort_rr_list(cfqq, 0); |
1763 | } | |
1da177e4 | 1764 | |
89850f7e | 1765 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 1766 | { |
3b18152c | 1767 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 1768 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 1769 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 1770 | return ELV_MQUEUE_MUST; |
3b18152c | 1771 | } |
1da177e4 | 1772 | |
22e2c507 | 1773 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
1774 | } |
1775 | ||
cb78b285 | 1776 | static int cfq_may_queue(request_queue_t *q, int rw) |
22e2c507 JA |
1777 | { |
1778 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1779 | struct task_struct *tsk = current; | |
1780 | struct cfq_queue *cfqq; | |
7749a8d4 JA |
1781 | unsigned int key; |
1782 | ||
1783 | key = cfq_queue_pid(tsk, rw, rw & REQ_RW_SYNC); | |
22e2c507 JA |
1784 | |
1785 | /* | |
1786 | * don't force setup of a queue from here, as a call to may_queue | |
1787 | * does not necessarily imply that a request actually will be queued. | |
1788 | * so just lookup a possibly existing queue, or return 'may queue' | |
1789 | * if that fails | |
1790 | */ | |
7749a8d4 | 1791 | cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio); |
22e2c507 JA |
1792 | if (cfqq) { |
1793 | cfq_init_prio_data(cfqq); | |
1794 | cfq_prio_boost(cfqq); | |
1795 | ||
89850f7e | 1796 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
1797 | } |
1798 | ||
1799 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
1800 | } |
1801 | ||
1da177e4 LT |
1802 | /* |
1803 | * queue lock held here | |
1804 | */ | |
bb37b94c | 1805 | static void cfq_put_request(struct request *rq) |
1da177e4 | 1806 | { |
5e705374 | 1807 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1808 | |
5e705374 | 1809 | if (cfqq) { |
22e2c507 | 1810 | const int rw = rq_data_dir(rq); |
1da177e4 | 1811 | |
22e2c507 JA |
1812 | BUG_ON(!cfqq->allocated[rw]); |
1813 | cfqq->allocated[rw]--; | |
1da177e4 | 1814 | |
5e705374 | 1815 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 1816 | |
1da177e4 | 1817 | rq->elevator_private = NULL; |
5e705374 | 1818 | rq->elevator_private2 = NULL; |
1da177e4 | 1819 | |
1da177e4 LT |
1820 | cfq_put_queue(cfqq); |
1821 | } | |
1822 | } | |
1823 | ||
1824 | /* | |
22e2c507 | 1825 | * Allocate cfq data structures associated with this request. |
1da177e4 | 1826 | */ |
22e2c507 | 1827 | static int |
cb78b285 | 1828 | cfq_set_request(request_queue_t *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
1829 | { |
1830 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3b18152c | 1831 | struct task_struct *tsk = current; |
1da177e4 LT |
1832 | struct cfq_io_context *cic; |
1833 | const int rw = rq_data_dir(rq); | |
7749a8d4 JA |
1834 | const int is_sync = rq_is_sync(rq); |
1835 | pid_t key = cfq_queue_pid(tsk, rw, is_sync); | |
22e2c507 | 1836 | struct cfq_queue *cfqq; |
1da177e4 LT |
1837 | unsigned long flags; |
1838 | ||
1839 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1840 | ||
e2d74ac0 | 1841 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 1842 | |
1da177e4 LT |
1843 | spin_lock_irqsave(q->queue_lock, flags); |
1844 | ||
22e2c507 JA |
1845 | if (!cic) |
1846 | goto queue_fail; | |
1847 | ||
12a05732 | 1848 | if (!cic->cfqq[is_sync]) { |
6f325a13 | 1849 | cfqq = cfq_get_queue(cfqd, key, tsk, gfp_mask); |
22e2c507 JA |
1850 | if (!cfqq) |
1851 | goto queue_fail; | |
1da177e4 | 1852 | |
12a05732 | 1853 | cic->cfqq[is_sync] = cfqq; |
22e2c507 | 1854 | } else |
12a05732 | 1855 | cfqq = cic->cfqq[is_sync]; |
1da177e4 LT |
1856 | |
1857 | cfqq->allocated[rw]++; | |
3b18152c | 1858 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 1859 | atomic_inc(&cfqq->ref); |
1da177e4 | 1860 | |
5e705374 | 1861 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 1862 | |
5e705374 JA |
1863 | rq->elevator_private = cic; |
1864 | rq->elevator_private2 = cfqq; | |
1865 | return 0; | |
1da177e4 | 1866 | |
22e2c507 JA |
1867 | queue_fail: |
1868 | if (cic) | |
1869 | put_io_context(cic->ioc); | |
89850f7e | 1870 | |
3b18152c | 1871 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
1872 | spin_unlock_irqrestore(q->queue_lock, flags); |
1873 | return 1; | |
1874 | } | |
1875 | ||
65f27f38 | 1876 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 1877 | { |
65f27f38 DH |
1878 | struct cfq_data *cfqd = |
1879 | container_of(work, struct cfq_data, unplug_work); | |
1880 | request_queue_t *q = cfqd->queue; | |
22e2c507 JA |
1881 | unsigned long flags; |
1882 | ||
1883 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 1884 | blk_start_queueing(q); |
22e2c507 JA |
1885 | spin_unlock_irqrestore(q->queue_lock, flags); |
1886 | } | |
1887 | ||
1888 | /* | |
1889 | * Timer running if the active_queue is currently idling inside its time slice | |
1890 | */ | |
1891 | static void cfq_idle_slice_timer(unsigned long data) | |
1892 | { | |
1893 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1894 | struct cfq_queue *cfqq; | |
1895 | unsigned long flags; | |
1896 | ||
1897 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1898 | ||
1899 | if ((cfqq = cfqd->active_queue) != NULL) { | |
1900 | unsigned long now = jiffies; | |
1901 | ||
1902 | /* | |
1903 | * expired | |
1904 | */ | |
1905 | if (time_after(now, cfqq->slice_end)) | |
1906 | goto expire; | |
1907 | ||
1908 | /* | |
1909 | * only expire and reinvoke request handler, if there are | |
1910 | * other queues with pending requests | |
1911 | */ | |
caaa5f9f | 1912 | if (!cfqd->busy_queues) |
22e2c507 | 1913 | goto out_cont; |
22e2c507 JA |
1914 | |
1915 | /* | |
1916 | * not expired and it has a request pending, let it dispatch | |
1917 | */ | |
dd67d051 | 1918 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 1919 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
1920 | goto out_kick; |
1921 | } | |
1922 | } | |
1923 | expire: | |
1924 | cfq_slice_expired(cfqd, 0); | |
1925 | out_kick: | |
3b18152c | 1926 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
1927 | out_cont: |
1928 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1929 | } | |
1930 | ||
1931 | /* | |
1932 | * Timer running if an idle class queue is waiting for service | |
1933 | */ | |
1934 | static void cfq_idle_class_timer(unsigned long data) | |
1935 | { | |
1936 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1937 | unsigned long flags, end; | |
1938 | ||
1939 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1940 | ||
1941 | /* | |
1942 | * race with a non-idle queue, reset timer | |
1943 | */ | |
1944 | end = cfqd->last_end_request + CFQ_IDLE_GRACE; | |
ae818a38 JA |
1945 | if (!time_after_eq(jiffies, end)) |
1946 | mod_timer(&cfqd->idle_class_timer, end); | |
1947 | else | |
3b18152c | 1948 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
1949 | |
1950 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1951 | } | |
1952 | ||
3b18152c JA |
1953 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
1954 | { | |
1955 | del_timer_sync(&cfqd->idle_slice_timer); | |
1956 | del_timer_sync(&cfqd->idle_class_timer); | |
1957 | blk_sync_queue(cfqd->queue); | |
1958 | } | |
22e2c507 | 1959 | |
1da177e4 LT |
1960 | static void cfq_exit_queue(elevator_t *e) |
1961 | { | |
22e2c507 | 1962 | struct cfq_data *cfqd = e->elevator_data; |
d9ff4187 | 1963 | request_queue_t *q = cfqd->queue; |
22e2c507 | 1964 | |
3b18152c | 1965 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 1966 | |
d9ff4187 | 1967 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 1968 | |
d9ff4187 AV |
1969 | if (cfqd->active_queue) |
1970 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); | |
e2d74ac0 JA |
1971 | |
1972 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
1973 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
1974 | struct cfq_io_context, | |
1975 | queue_list); | |
89850f7e JA |
1976 | |
1977 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 1978 | } |
e2d74ac0 | 1979 | |
d9ff4187 | 1980 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
1981 | |
1982 | cfq_shutdown_timer_wq(cfqd); | |
1983 | ||
a90d742e AV |
1984 | kfree(cfqd->cfq_hash); |
1985 | kfree(cfqd); | |
1da177e4 LT |
1986 | } |
1987 | ||
bb37b94c | 1988 | static void *cfq_init_queue(request_queue_t *q) |
1da177e4 LT |
1989 | { |
1990 | struct cfq_data *cfqd; | |
1991 | int i; | |
1992 | ||
b5deef90 | 1993 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node); |
1da177e4 | 1994 | if (!cfqd) |
bc1c1169 | 1995 | return NULL; |
1da177e4 LT |
1996 | |
1997 | memset(cfqd, 0, sizeof(*cfqd)); | |
22e2c507 JA |
1998 | |
1999 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
2000 | INIT_LIST_HEAD(&cfqd->rr_list[i]); | |
2001 | ||
2002 | INIT_LIST_HEAD(&cfqd->busy_rr); | |
2003 | INIT_LIST_HEAD(&cfqd->cur_rr); | |
2004 | INIT_LIST_HEAD(&cfqd->idle_rr); | |
d9ff4187 | 2005 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2006 | |
b5deef90 | 2007 | cfqd->cfq_hash = kmalloc_node(sizeof(struct hlist_head) * CFQ_QHASH_ENTRIES, GFP_KERNEL, q->node); |
1da177e4 | 2008 | if (!cfqd->cfq_hash) |
5e705374 | 2009 | goto out_free; |
1da177e4 | 2010 | |
1da177e4 LT |
2011 | for (i = 0; i < CFQ_QHASH_ENTRIES; i++) |
2012 | INIT_HLIST_HEAD(&cfqd->cfq_hash[i]); | |
2013 | ||
1da177e4 | 2014 | cfqd->queue = q; |
1da177e4 | 2015 | |
22e2c507 JA |
2016 | init_timer(&cfqd->idle_slice_timer); |
2017 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2018 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2019 | ||
2020 | init_timer(&cfqd->idle_class_timer); | |
2021 | cfqd->idle_class_timer.function = cfq_idle_class_timer; | |
2022 | cfqd->idle_class_timer.data = (unsigned long) cfqd; | |
2023 | ||
65f27f38 | 2024 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2025 | |
1da177e4 | 2026 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2027 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2028 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2029 | cfqd->cfq_back_max = cfq_back_max; |
2030 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2031 | cfqd->cfq_slice[0] = cfq_slice_async; |
2032 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2033 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2034 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
3b18152c | 2035 | |
bc1c1169 | 2036 | return cfqd; |
5e705374 | 2037 | out_free: |
1da177e4 | 2038 | kfree(cfqd); |
bc1c1169 | 2039 | return NULL; |
1da177e4 LT |
2040 | } |
2041 | ||
2042 | static void cfq_slab_kill(void) | |
2043 | { | |
1da177e4 LT |
2044 | if (cfq_pool) |
2045 | kmem_cache_destroy(cfq_pool); | |
2046 | if (cfq_ioc_pool) | |
2047 | kmem_cache_destroy(cfq_ioc_pool); | |
2048 | } | |
2049 | ||
2050 | static int __init cfq_slab_setup(void) | |
2051 | { | |
1da177e4 LT |
2052 | cfq_pool = kmem_cache_create("cfq_pool", sizeof(struct cfq_queue), 0, 0, |
2053 | NULL, NULL); | |
2054 | if (!cfq_pool) | |
2055 | goto fail; | |
2056 | ||
2057 | cfq_ioc_pool = kmem_cache_create("cfq_ioc_pool", | |
2058 | sizeof(struct cfq_io_context), 0, 0, NULL, NULL); | |
2059 | if (!cfq_ioc_pool) | |
2060 | goto fail; | |
2061 | ||
2062 | return 0; | |
2063 | fail: | |
2064 | cfq_slab_kill(); | |
2065 | return -ENOMEM; | |
2066 | } | |
2067 | ||
1da177e4 LT |
2068 | /* |
2069 | * sysfs parts below --> | |
2070 | */ | |
1da177e4 LT |
2071 | |
2072 | static ssize_t | |
2073 | cfq_var_show(unsigned int var, char *page) | |
2074 | { | |
2075 | return sprintf(page, "%d\n", var); | |
2076 | } | |
2077 | ||
2078 | static ssize_t | |
2079 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2080 | { | |
2081 | char *p = (char *) page; | |
2082 | ||
2083 | *var = simple_strtoul(p, &p, 10); | |
2084 | return count; | |
2085 | } | |
2086 | ||
1da177e4 | 2087 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
3d1ab40f | 2088 | static ssize_t __FUNC(elevator_t *e, char *page) \ |
1da177e4 | 2089 | { \ |
3d1ab40f | 2090 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2091 | unsigned int __data = __VAR; \ |
2092 | if (__CONV) \ | |
2093 | __data = jiffies_to_msecs(__data); \ | |
2094 | return cfq_var_show(__data, (page)); \ | |
2095 | } | |
2096 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2097 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2098 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2099 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2100 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2101 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2102 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2103 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2104 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2105 | #undef SHOW_FUNCTION |
2106 | ||
2107 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
3d1ab40f | 2108 | static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \ |
1da177e4 | 2109 | { \ |
3d1ab40f | 2110 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2111 | unsigned int __data; \ |
2112 | int ret = cfq_var_store(&__data, (page), count); \ | |
2113 | if (__data < (MIN)) \ | |
2114 | __data = (MIN); \ | |
2115 | else if (__data > (MAX)) \ | |
2116 | __data = (MAX); \ | |
2117 | if (__CONV) \ | |
2118 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2119 | else \ | |
2120 | *(__PTR) = __data; \ | |
2121 | return ret; \ | |
2122 | } | |
2123 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
22e2c507 JA |
2124 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1); |
2125 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1); | |
e572ec7e AV |
2126 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
2127 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0); | |
22e2c507 JA |
2128 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2129 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2130 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
2131 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0); | |
1da177e4 LT |
2132 | #undef STORE_FUNCTION |
2133 | ||
e572ec7e AV |
2134 | #define CFQ_ATTR(name) \ |
2135 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2136 | ||
2137 | static struct elv_fs_entry cfq_attrs[] = { | |
2138 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2139 | CFQ_ATTR(fifo_expire_sync), |
2140 | CFQ_ATTR(fifo_expire_async), | |
2141 | CFQ_ATTR(back_seek_max), | |
2142 | CFQ_ATTR(back_seek_penalty), | |
2143 | CFQ_ATTR(slice_sync), | |
2144 | CFQ_ATTR(slice_async), | |
2145 | CFQ_ATTR(slice_async_rq), | |
2146 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2147 | __ATTR_NULL |
1da177e4 LT |
2148 | }; |
2149 | ||
1da177e4 LT |
2150 | static struct elevator_type iosched_cfq = { |
2151 | .ops = { | |
2152 | .elevator_merge_fn = cfq_merge, | |
2153 | .elevator_merged_fn = cfq_merged_request, | |
2154 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2155 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2156 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2157 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2158 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2159 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2160 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2161 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2162 | .elevator_former_req_fn = elv_rb_former_request, |
2163 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2164 | .elevator_set_req_fn = cfq_set_request, |
2165 | .elevator_put_req_fn = cfq_put_request, | |
2166 | .elevator_may_queue_fn = cfq_may_queue, | |
2167 | .elevator_init_fn = cfq_init_queue, | |
2168 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2169 | .trim = cfq_free_io_context, |
1da177e4 | 2170 | }, |
3d1ab40f | 2171 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2172 | .elevator_name = "cfq", |
2173 | .elevator_owner = THIS_MODULE, | |
2174 | }; | |
2175 | ||
2176 | static int __init cfq_init(void) | |
2177 | { | |
2178 | int ret; | |
2179 | ||
22e2c507 JA |
2180 | /* |
2181 | * could be 0 on HZ < 1000 setups | |
2182 | */ | |
2183 | if (!cfq_slice_async) | |
2184 | cfq_slice_async = 1; | |
2185 | if (!cfq_slice_idle) | |
2186 | cfq_slice_idle = 1; | |
2187 | ||
1da177e4 LT |
2188 | if (cfq_slab_setup()) |
2189 | return -ENOMEM; | |
2190 | ||
2191 | ret = elv_register(&iosched_cfq); | |
22e2c507 JA |
2192 | if (ret) |
2193 | cfq_slab_kill(); | |
1da177e4 | 2194 | |
1da177e4 LT |
2195 | return ret; |
2196 | } | |
2197 | ||
2198 | static void __exit cfq_exit(void) | |
2199 | { | |
6e9a4738 | 2200 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2201 | elv_unregister(&iosched_cfq); |
334e94de | 2202 | ioc_gone = &all_gone; |
fba82272 OH |
2203 | /* ioc_gone's update must be visible before reading ioc_count */ |
2204 | smp_wmb(); | |
4050cf16 | 2205 | if (elv_ioc_count_read(ioc_count)) |
fba82272 | 2206 | wait_for_completion(ioc_gone); |
334e94de | 2207 | synchronize_rcu(); |
83521d3e | 2208 | cfq_slab_kill(); |
1da177e4 LT |
2209 | } |
2210 | ||
2211 | module_init(cfq_init); | |
2212 | module_exit(cfq_exit); | |
2213 | ||
2214 | MODULE_AUTHOR("Jens Axboe"); | |
2215 | MODULE_LICENSE("GPL"); | |
2216 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |