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 | * | |
7 | * Copyright (C) 2003 Jens Axboe <axboe@suse.de> | |
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 AV |
19 | static const int cfq_quantum = 4; /* max queue in one round of service */ |
20 | static const int cfq_queued = 8; /* minimum rq allocate limit per-queue*/ | |
21 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; | |
22 | static const int cfq_back_max = 16 * 1024; /* maximum backwards seek, in KiB */ | |
23 | static const int cfq_back_penalty = 2; /* penalty of a backwards seek */ | |
1da177e4 | 24 | |
64100099 | 25 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 26 | static int cfq_slice_async = HZ / 25; |
64100099 | 27 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 28 | static int cfq_slice_idle = HZ / 125; |
22e2c507 JA |
29 | |
30 | #define CFQ_IDLE_GRACE (HZ / 10) | |
31 | #define CFQ_SLICE_SCALE (5) | |
32 | ||
33 | #define CFQ_KEY_ASYNC (0) | |
22e2c507 | 34 | |
3793c65c | 35 | static DEFINE_SPINLOCK(cfq_exit_lock); |
a6a0763a | 36 | |
1da177e4 LT |
37 | /* |
38 | * for the hash of cfqq inside the cfqd | |
39 | */ | |
40 | #define CFQ_QHASH_SHIFT 6 | |
41 | #define CFQ_QHASH_ENTRIES (1 << CFQ_QHASH_SHIFT) | |
42 | #define list_entry_qhash(entry) hlist_entry((entry), struct cfq_queue, cfq_hash) | |
43 | ||
1da177e4 LT |
44 | #define list_entry_cfqq(ptr) list_entry((ptr), struct cfq_queue, cfq_list) |
45 | ||
5e705374 JA |
46 | #define RQ_CIC(rq) ((struct cfq_io_context*)(rq)->elevator_private) |
47 | #define RQ_CFQQ(rq) ((rq)->elevator_private2) | |
1da177e4 | 48 | |
1da177e4 LT |
49 | static kmem_cache_t *cfq_pool; |
50 | static kmem_cache_t *cfq_ioc_pool; | |
51 | ||
334e94de AV |
52 | static atomic_t ioc_count = ATOMIC_INIT(0); |
53 | static struct completion *ioc_gone; | |
54 | ||
22e2c507 JA |
55 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
56 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
57 | #define cfq_class_be(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_BE) | |
58 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) | |
59 | ||
3b18152c JA |
60 | #define ASYNC (0) |
61 | #define SYNC (1) | |
62 | ||
63 | #define cfq_cfqq_dispatched(cfqq) \ | |
64 | ((cfqq)->on_dispatch[ASYNC] + (cfqq)->on_dispatch[SYNC]) | |
65 | ||
66 | #define cfq_cfqq_class_sync(cfqq) ((cfqq)->key != CFQ_KEY_ASYNC) | |
67 | ||
68 | #define cfq_cfqq_sync(cfqq) \ | |
69 | (cfq_cfqq_class_sync(cfqq) || (cfqq)->on_dispatch[SYNC]) | |
22e2c507 | 70 | |
206dc69b JA |
71 | #define sample_valid(samples) ((samples) > 80) |
72 | ||
22e2c507 JA |
73 | /* |
74 | * Per block device queue structure | |
75 | */ | |
1da177e4 | 76 | struct cfq_data { |
22e2c507 JA |
77 | request_queue_t *queue; |
78 | ||
79 | /* | |
80 | * rr list of queues with requests and the count of them | |
81 | */ | |
82 | struct list_head rr_list[CFQ_PRIO_LISTS]; | |
83 | struct list_head busy_rr; | |
84 | struct list_head cur_rr; | |
85 | struct list_head idle_rr; | |
86 | unsigned int busy_queues; | |
87 | ||
88 | /* | |
89 | * non-ordered list of empty cfqq's | |
90 | */ | |
1da177e4 LT |
91 | struct list_head empty_list; |
92 | ||
22e2c507 JA |
93 | /* |
94 | * cfqq lookup hash | |
95 | */ | |
1da177e4 | 96 | struct hlist_head *cfq_hash; |
1da177e4 | 97 | |
22e2c507 | 98 | int rq_in_driver; |
25776e35 | 99 | int hw_tag; |
1da177e4 | 100 | |
22e2c507 JA |
101 | /* |
102 | * schedule slice state info | |
103 | */ | |
104 | /* | |
105 | * idle window management | |
106 | */ | |
107 | struct timer_list idle_slice_timer; | |
108 | struct work_struct unplug_work; | |
1da177e4 | 109 | |
22e2c507 JA |
110 | struct cfq_queue *active_queue; |
111 | struct cfq_io_context *active_cic; | |
112 | int cur_prio, cur_end_prio; | |
113 | unsigned int dispatch_slice; | |
114 | ||
115 | struct timer_list idle_class_timer; | |
1da177e4 LT |
116 | |
117 | sector_t last_sector; | |
22e2c507 | 118 | unsigned long last_end_request; |
1da177e4 | 119 | |
22e2c507 | 120 | unsigned int rq_starved; |
1da177e4 LT |
121 | |
122 | /* | |
123 | * tunables, see top of file | |
124 | */ | |
125 | unsigned int cfq_quantum; | |
126 | unsigned int cfq_queued; | |
22e2c507 | 127 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
128 | unsigned int cfq_back_penalty; |
129 | unsigned int cfq_back_max; | |
22e2c507 JA |
130 | unsigned int cfq_slice[2]; |
131 | unsigned int cfq_slice_async_rq; | |
132 | unsigned int cfq_slice_idle; | |
d9ff4187 AV |
133 | |
134 | struct list_head cic_list; | |
1da177e4 LT |
135 | }; |
136 | ||
22e2c507 JA |
137 | /* |
138 | * Per process-grouping structure | |
139 | */ | |
1da177e4 LT |
140 | struct cfq_queue { |
141 | /* reference count */ | |
142 | atomic_t ref; | |
143 | /* parent cfq_data */ | |
144 | struct cfq_data *cfqd; | |
22e2c507 | 145 | /* cfqq lookup hash */ |
1da177e4 LT |
146 | struct hlist_node cfq_hash; |
147 | /* hash key */ | |
22e2c507 | 148 | unsigned int key; |
1da177e4 LT |
149 | /* on either rr or empty list of cfqd */ |
150 | struct list_head cfq_list; | |
151 | /* sorted list of pending requests */ | |
152 | struct rb_root sort_list; | |
153 | /* if fifo isn't expired, next request to serve */ | |
5e705374 | 154 | struct request *next_rq; |
1da177e4 LT |
155 | /* requests queued in sort_list */ |
156 | int queued[2]; | |
157 | /* currently allocated requests */ | |
158 | int allocated[2]; | |
159 | /* fifo list of requests in sort_list */ | |
22e2c507 | 160 | struct list_head fifo; |
1da177e4 | 161 | |
22e2c507 JA |
162 | unsigned long slice_start; |
163 | unsigned long slice_end; | |
164 | unsigned long slice_left; | |
165 | unsigned long service_last; | |
1da177e4 | 166 | |
3b18152c JA |
167 | /* number of requests that are on the dispatch list */ |
168 | int on_dispatch[2]; | |
22e2c507 JA |
169 | |
170 | /* io prio of this group */ | |
171 | unsigned short ioprio, org_ioprio; | |
172 | unsigned short ioprio_class, org_ioprio_class; | |
173 | ||
3b18152c JA |
174 | /* various state flags, see below */ |
175 | unsigned int flags; | |
1da177e4 LT |
176 | }; |
177 | ||
3b18152c JA |
178 | enum cfqq_state_flags { |
179 | CFQ_CFQQ_FLAG_on_rr = 0, | |
180 | CFQ_CFQQ_FLAG_wait_request, | |
181 | CFQ_CFQQ_FLAG_must_alloc, | |
182 | CFQ_CFQQ_FLAG_must_alloc_slice, | |
183 | CFQ_CFQQ_FLAG_must_dispatch, | |
184 | CFQ_CFQQ_FLAG_fifo_expire, | |
185 | CFQ_CFQQ_FLAG_idle_window, | |
186 | CFQ_CFQQ_FLAG_prio_changed, | |
3b18152c JA |
187 | }; |
188 | ||
189 | #define CFQ_CFQQ_FNS(name) \ | |
190 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
191 | { \ | |
192 | cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ | |
193 | } \ | |
194 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
195 | { \ | |
196 | cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ | |
197 | } \ | |
198 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
199 | { \ | |
200 | return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ | |
201 | } | |
202 | ||
203 | CFQ_CFQQ_FNS(on_rr); | |
204 | CFQ_CFQQ_FNS(wait_request); | |
205 | CFQ_CFQQ_FNS(must_alloc); | |
206 | CFQ_CFQQ_FNS(must_alloc_slice); | |
207 | CFQ_CFQQ_FNS(must_dispatch); | |
208 | CFQ_CFQQ_FNS(fifo_expire); | |
209 | CFQ_CFQQ_FNS(idle_window); | |
210 | CFQ_CFQQ_FNS(prio_changed); | |
3b18152c JA |
211 | #undef CFQ_CFQQ_FNS |
212 | ||
3b18152c | 213 | static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short); |
5e705374 | 214 | static void cfq_dispatch_insert(request_queue_t *, struct request *); |
6f325a13 | 215 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, gfp_t gfp_mask); |
1da177e4 | 216 | |
99f95e52 AM |
217 | /* |
218 | * scheduler run of queue, if there are requests pending and no one in the | |
219 | * driver that will restart queueing | |
220 | */ | |
221 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
222 | { | |
7b14e3b5 | 223 | if (cfqd->busy_queues) |
99f95e52 AM |
224 | kblockd_schedule_work(&cfqd->unplug_work); |
225 | } | |
226 | ||
227 | static int cfq_queue_empty(request_queue_t *q) | |
228 | { | |
229 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
230 | ||
b4878f24 | 231 | return !cfqd->busy_queues; |
99f95e52 AM |
232 | } |
233 | ||
206dc69b JA |
234 | static inline pid_t cfq_queue_pid(struct task_struct *task, int rw) |
235 | { | |
b31dc66a | 236 | if (rw == READ || rw == WRITE_SYNC) |
206dc69b JA |
237 | return task->pid; |
238 | ||
239 | return CFQ_KEY_ASYNC; | |
240 | } | |
241 | ||
1da177e4 | 242 | /* |
5e705374 | 243 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 244 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 245 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 246 | */ |
5e705374 JA |
247 | static struct request * |
248 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
249 | { |
250 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 251 | unsigned long back_max; |
e8a99053 AM |
252 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
253 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
254 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 255 | |
5e705374 JA |
256 | if (rq1 == NULL || rq1 == rq2) |
257 | return rq2; | |
258 | if (rq2 == NULL) | |
259 | return rq1; | |
9c2c38a1 | 260 | |
5e705374 JA |
261 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
262 | return rq1; | |
263 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
264 | return rq2; | |
1da177e4 | 265 | |
5e705374 JA |
266 | s1 = rq1->sector; |
267 | s2 = rq2->sector; | |
1da177e4 LT |
268 | |
269 | last = cfqd->last_sector; | |
270 | ||
1da177e4 LT |
271 | /* |
272 | * by definition, 1KiB is 2 sectors | |
273 | */ | |
274 | back_max = cfqd->cfq_back_max * 2; | |
275 | ||
276 | /* | |
277 | * Strict one way elevator _except_ in the case where we allow | |
278 | * short backward seeks which are biased as twice the cost of a | |
279 | * similar forward seek. | |
280 | */ | |
281 | if (s1 >= last) | |
282 | d1 = s1 - last; | |
283 | else if (s1 + back_max >= last) | |
284 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
285 | else | |
e8a99053 | 286 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
287 | |
288 | if (s2 >= last) | |
289 | d2 = s2 - last; | |
290 | else if (s2 + back_max >= last) | |
291 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
292 | else | |
e8a99053 | 293 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
294 | |
295 | /* Found required data */ | |
e8a99053 AM |
296 | |
297 | /* | |
298 | * By doing switch() on the bit mask "wrap" we avoid having to | |
299 | * check two variables for all permutations: --> faster! | |
300 | */ | |
301 | switch (wrap) { | |
5e705374 | 302 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 303 | if (d1 < d2) |
5e705374 | 304 | return rq1; |
e8a99053 | 305 | else if (d2 < d1) |
5e705374 | 306 | return rq2; |
e8a99053 AM |
307 | else { |
308 | if (s1 >= s2) | |
5e705374 | 309 | return rq1; |
e8a99053 | 310 | else |
5e705374 | 311 | return rq2; |
e8a99053 | 312 | } |
1da177e4 | 313 | |
e8a99053 | 314 | case CFQ_RQ2_WRAP: |
5e705374 | 315 | return rq1; |
e8a99053 | 316 | case CFQ_RQ1_WRAP: |
5e705374 JA |
317 | return rq2; |
318 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
319 | default: |
320 | /* | |
321 | * Since both rqs are wrapped, | |
322 | * start with the one that's further behind head | |
323 | * (--> only *one* back seek required), | |
324 | * since back seek takes more time than forward. | |
325 | */ | |
326 | if (s1 <= s2) | |
5e705374 | 327 | return rq1; |
1da177e4 | 328 | else |
5e705374 | 329 | return rq2; |
1da177e4 LT |
330 | } |
331 | } | |
332 | ||
333 | /* | |
334 | * would be nice to take fifo expire time into account as well | |
335 | */ | |
5e705374 JA |
336 | static struct request * |
337 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
338 | struct request *last) | |
1da177e4 | 339 | { |
21183b07 JA |
340 | struct rb_node *rbnext = rb_next(&last->rb_node); |
341 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 342 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 343 | |
21183b07 | 344 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
345 | |
346 | if (rbprev) | |
5e705374 | 347 | prev = rb_entry_rq(rbprev); |
1da177e4 | 348 | |
21183b07 | 349 | if (rbnext) |
5e705374 | 350 | next = rb_entry_rq(rbnext); |
21183b07 JA |
351 | else { |
352 | rbnext = rb_first(&cfqq->sort_list); | |
353 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 354 | next = rb_entry_rq(rbnext); |
21183b07 | 355 | } |
1da177e4 | 356 | |
21183b07 | 357 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
358 | } |
359 | ||
22e2c507 | 360 | static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted) |
1da177e4 | 361 | { |
22e2c507 JA |
362 | struct cfq_data *cfqd = cfqq->cfqd; |
363 | struct list_head *list, *entry; | |
1da177e4 | 364 | |
3b18152c | 365 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 366 | |
22e2c507 | 367 | list_del(&cfqq->cfq_list); |
1da177e4 | 368 | |
22e2c507 JA |
369 | if (cfq_class_rt(cfqq)) |
370 | list = &cfqd->cur_rr; | |
371 | else if (cfq_class_idle(cfqq)) | |
372 | list = &cfqd->idle_rr; | |
373 | else { | |
374 | /* | |
375 | * if cfqq has requests in flight, don't allow it to be | |
376 | * found in cfq_set_active_queue before it has finished them. | |
377 | * this is done to increase fairness between a process that | |
378 | * has lots of io pending vs one that only generates one | |
379 | * sporadically or synchronously | |
380 | */ | |
3b18152c | 381 | if (cfq_cfqq_dispatched(cfqq)) |
22e2c507 JA |
382 | list = &cfqd->busy_rr; |
383 | else | |
384 | list = &cfqd->rr_list[cfqq->ioprio]; | |
1da177e4 LT |
385 | } |
386 | ||
22e2c507 JA |
387 | /* |
388 | * if queue was preempted, just add to front to be fair. busy_rr | |
b52a8348 | 389 | * isn't sorted, but insert at the back for fairness. |
22e2c507 JA |
390 | */ |
391 | if (preempted || list == &cfqd->busy_rr) { | |
b52a8348 JA |
392 | if (preempted) |
393 | list = list->prev; | |
394 | ||
395 | list_add_tail(&cfqq->cfq_list, list); | |
1da177e4 | 396 | return; |
22e2c507 | 397 | } |
1da177e4 LT |
398 | |
399 | /* | |
22e2c507 | 400 | * sort by when queue was last serviced |
1da177e4 | 401 | */ |
22e2c507 JA |
402 | entry = list; |
403 | while ((entry = entry->prev) != list) { | |
1da177e4 LT |
404 | struct cfq_queue *__cfqq = list_entry_cfqq(entry); |
405 | ||
22e2c507 JA |
406 | if (!__cfqq->service_last) |
407 | break; | |
408 | if (time_before(__cfqq->service_last, cfqq->service_last)) | |
1da177e4 | 409 | break; |
1da177e4 LT |
410 | } |
411 | ||
412 | list_add(&cfqq->cfq_list, entry); | |
413 | } | |
414 | ||
415 | /* | |
416 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 417 | * the pending list according to last request service |
1da177e4 LT |
418 | */ |
419 | static inline void | |
b4878f24 | 420 | cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 421 | { |
3b18152c JA |
422 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
423 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
424 | cfqd->busy_queues++; |
425 | ||
b4878f24 | 426 | cfq_resort_rr_list(cfqq, 0); |
1da177e4 LT |
427 | } |
428 | ||
429 | static inline void | |
430 | cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
431 | { | |
3b18152c JA |
432 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
433 | cfq_clear_cfqq_on_rr(cfqq); | |
22e2c507 | 434 | list_move(&cfqq->cfq_list, &cfqd->empty_list); |
1da177e4 LT |
435 | |
436 | BUG_ON(!cfqd->busy_queues); | |
437 | cfqd->busy_queues--; | |
438 | } | |
439 | ||
440 | /* | |
441 | * rb tree support functions | |
442 | */ | |
5e705374 | 443 | static inline void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 444 | { |
5e705374 | 445 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 446 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 447 | const int sync = rq_is_sync(rq); |
1da177e4 | 448 | |
b4878f24 JA |
449 | BUG_ON(!cfqq->queued[sync]); |
450 | cfqq->queued[sync]--; | |
1da177e4 | 451 | |
5e705374 | 452 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 453 | |
dd67d051 | 454 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 455 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
456 | } |
457 | ||
5e705374 | 458 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 459 | { |
5e705374 | 460 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 461 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 462 | struct request *__alias; |
1da177e4 | 463 | |
5380a101 | 464 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
465 | |
466 | /* | |
467 | * looks a little odd, but the first insert might return an alias. | |
468 | * if that happens, put the alias on the dispatch list | |
469 | */ | |
21183b07 | 470 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 471 | cfq_dispatch_insert(cfqd->queue, __alias); |
1da177e4 LT |
472 | } |
473 | ||
474 | static inline void | |
5e705374 | 475 | cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 476 | { |
5380a101 JA |
477 | elv_rb_del(&cfqq->sort_list, rq); |
478 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 479 | cfq_add_rq_rb(rq); |
1da177e4 LT |
480 | } |
481 | ||
206dc69b JA |
482 | static struct request * |
483 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 484 | { |
206dc69b JA |
485 | struct task_struct *tsk = current; |
486 | pid_t key = cfq_queue_pid(tsk, bio_data_dir(bio)); | |
21183b07 | 487 | sector_t sector = bio->bi_sector + bio_sectors(bio); |
206dc69b | 488 | struct cfq_queue *cfqq; |
1da177e4 | 489 | |
206dc69b | 490 | cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio); |
21183b07 JA |
491 | if (cfqq) |
492 | return elv_rb_find(&cfqq->sort_list, sector); | |
1da177e4 | 493 | |
1da177e4 LT |
494 | return NULL; |
495 | } | |
496 | ||
b4878f24 | 497 | static void cfq_activate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 498 | { |
22e2c507 | 499 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 500 | |
b4878f24 | 501 | cfqd->rq_in_driver++; |
25776e35 JA |
502 | |
503 | /* | |
504 | * If the depth is larger 1, it really could be queueing. But lets | |
505 | * make the mark a little higher - idling could still be good for | |
506 | * low queueing, and a low queueing number could also just indicate | |
507 | * a SCSI mid layer like behaviour where limit+1 is often seen. | |
508 | */ | |
509 | if (!cfqd->hw_tag && cfqd->rq_in_driver > 4) | |
510 | cfqd->hw_tag = 1; | |
1da177e4 LT |
511 | } |
512 | ||
b4878f24 | 513 | static void cfq_deactivate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 514 | { |
b4878f24 JA |
515 | struct cfq_data *cfqd = q->elevator->elevator_data; |
516 | ||
517 | WARN_ON(!cfqd->rq_in_driver); | |
518 | cfqd->rq_in_driver--; | |
1da177e4 LT |
519 | } |
520 | ||
b4878f24 | 521 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 522 | { |
5e705374 | 523 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 524 | |
5e705374 JA |
525 | if (cfqq->next_rq == rq) |
526 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 527 | |
b4878f24 | 528 | list_del_init(&rq->queuelist); |
5e705374 | 529 | cfq_del_rq_rb(rq); |
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 | ||
571 | static inline void | |
572 | __cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
573 | { | |
574 | if (cfqq) { | |
575 | /* | |
576 | * stop potential idle class queues waiting service | |
577 | */ | |
578 | del_timer(&cfqd->idle_class_timer); | |
579 | ||
580 | cfqq->slice_start = jiffies; | |
581 | cfqq->slice_end = 0; | |
582 | cfqq->slice_left = 0; | |
3b18152c JA |
583 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
584 | cfq_clear_cfqq_fifo_expire(cfqq); | |
22e2c507 JA |
585 | } |
586 | ||
587 | cfqd->active_queue = cfqq; | |
588 | } | |
589 | ||
7b14e3b5 JA |
590 | /* |
591 | * current cfqq expired its slice (or was too idle), select new one | |
592 | */ | |
593 | static void | |
594 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
595 | int preempted) | |
596 | { | |
597 | unsigned long now = jiffies; | |
598 | ||
599 | if (cfq_cfqq_wait_request(cfqq)) | |
600 | del_timer(&cfqd->idle_slice_timer); | |
601 | ||
602 | if (!preempted && !cfq_cfqq_dispatched(cfqq)) { | |
603 | cfqq->service_last = now; | |
604 | cfq_schedule_dispatch(cfqd); | |
605 | } | |
606 | ||
607 | cfq_clear_cfqq_must_dispatch(cfqq); | |
608 | cfq_clear_cfqq_wait_request(cfqq); | |
609 | ||
610 | /* | |
611 | * store what was left of this slice, if the queue idled out | |
612 | * or was preempted | |
613 | */ | |
614 | if (time_after(cfqq->slice_end, now)) | |
615 | cfqq->slice_left = cfqq->slice_end - now; | |
616 | else | |
617 | cfqq->slice_left = 0; | |
618 | ||
619 | if (cfq_cfqq_on_rr(cfqq)) | |
620 | cfq_resort_rr_list(cfqq, preempted); | |
621 | ||
622 | if (cfqq == cfqd->active_queue) | |
623 | cfqd->active_queue = NULL; | |
624 | ||
625 | if (cfqd->active_cic) { | |
626 | put_io_context(cfqd->active_cic->ioc); | |
627 | cfqd->active_cic = NULL; | |
628 | } | |
629 | ||
630 | cfqd->dispatch_slice = 0; | |
631 | } | |
632 | ||
633 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted) | |
634 | { | |
635 | struct cfq_queue *cfqq = cfqd->active_queue; | |
636 | ||
637 | if (cfqq) | |
638 | __cfq_slice_expired(cfqd, cfqq, preempted); | |
639 | } | |
640 | ||
22e2c507 JA |
641 | /* |
642 | * 0 | |
643 | * 0,1 | |
644 | * 0,1,2 | |
645 | * 0,1,2,3 | |
646 | * 0,1,2,3,4 | |
647 | * 0,1,2,3,4,5 | |
648 | * 0,1,2,3,4,5,6 | |
649 | * 0,1,2,3,4,5,6,7 | |
650 | */ | |
651 | static int cfq_get_next_prio_level(struct cfq_data *cfqd) | |
652 | { | |
653 | int prio, wrap; | |
654 | ||
655 | prio = -1; | |
656 | wrap = 0; | |
657 | do { | |
658 | int p; | |
659 | ||
660 | for (p = cfqd->cur_prio; p <= cfqd->cur_end_prio; p++) { | |
661 | if (!list_empty(&cfqd->rr_list[p])) { | |
662 | prio = p; | |
663 | break; | |
664 | } | |
665 | } | |
666 | ||
667 | if (prio != -1) | |
668 | break; | |
669 | cfqd->cur_prio = 0; | |
670 | if (++cfqd->cur_end_prio == CFQ_PRIO_LISTS) { | |
671 | cfqd->cur_end_prio = 0; | |
672 | if (wrap) | |
673 | break; | |
674 | wrap = 1; | |
1da177e4 | 675 | } |
22e2c507 JA |
676 | } while (1); |
677 | ||
678 | if (unlikely(prio == -1)) | |
679 | return -1; | |
680 | ||
681 | BUG_ON(prio >= CFQ_PRIO_LISTS); | |
682 | ||
683 | list_splice_init(&cfqd->rr_list[prio], &cfqd->cur_rr); | |
684 | ||
685 | cfqd->cur_prio = prio + 1; | |
686 | if (cfqd->cur_prio > cfqd->cur_end_prio) { | |
687 | cfqd->cur_end_prio = cfqd->cur_prio; | |
688 | cfqd->cur_prio = 0; | |
689 | } | |
690 | if (cfqd->cur_end_prio == CFQ_PRIO_LISTS) { | |
691 | cfqd->cur_prio = 0; | |
692 | cfqd->cur_end_prio = 0; | |
1da177e4 LT |
693 | } |
694 | ||
22e2c507 JA |
695 | return prio; |
696 | } | |
697 | ||
3b18152c | 698 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
22e2c507 | 699 | { |
7b14e3b5 | 700 | struct cfq_queue *cfqq = NULL; |
22e2c507 JA |
701 | |
702 | /* | |
703 | * if current list is non-empty, grab first entry. if it is empty, | |
704 | * get next prio level and grab first entry then if any are spliced | |
705 | */ | |
706 | if (!list_empty(&cfqd->cur_rr) || cfq_get_next_prio_level(cfqd) != -1) | |
707 | cfqq = list_entry_cfqq(cfqd->cur_rr.next); | |
708 | ||
e0de0206 JA |
709 | /* |
710 | * If no new queues are available, check if the busy list has some | |
711 | * before falling back to idle io. | |
712 | */ | |
713 | if (!cfqq && !list_empty(&cfqd->busy_rr)) | |
714 | cfqq = list_entry_cfqq(cfqd->busy_rr.next); | |
715 | ||
22e2c507 JA |
716 | /* |
717 | * if we have idle queues and no rt or be queues had pending | |
718 | * requests, either allow immediate service if the grace period | |
719 | * has passed or arm the idle grace timer | |
720 | */ | |
721 | if (!cfqq && !list_empty(&cfqd->idle_rr)) { | |
722 | unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE; | |
723 | ||
724 | if (time_after_eq(jiffies, end)) | |
725 | cfqq = list_entry_cfqq(cfqd->idle_rr.next); | |
726 | else | |
727 | mod_timer(&cfqd->idle_class_timer, end); | |
728 | } | |
729 | ||
730 | __cfq_set_active_queue(cfqd, cfqq); | |
3b18152c | 731 | return cfqq; |
22e2c507 JA |
732 | } |
733 | ||
caaa5f9f JA |
734 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (128 * 1024)) |
735 | ||
22e2c507 JA |
736 | static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
737 | ||
738 | { | |
206dc69b | 739 | struct cfq_io_context *cic; |
7b14e3b5 JA |
740 | unsigned long sl; |
741 | ||
dd67d051 | 742 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
743 | WARN_ON(cfqq != cfqd->active_queue); |
744 | ||
745 | /* | |
746 | * idle is disabled, either manually or by past process history | |
747 | */ | |
748 | if (!cfqd->cfq_slice_idle) | |
749 | return 0; | |
3b18152c | 750 | if (!cfq_cfqq_idle_window(cfqq)) |
22e2c507 JA |
751 | return 0; |
752 | /* | |
753 | * task has exited, don't wait | |
754 | */ | |
206dc69b JA |
755 | cic = cfqd->active_cic; |
756 | if (!cic || !cic->ioc->task) | |
22e2c507 JA |
757 | return 0; |
758 | ||
3b18152c JA |
759 | cfq_mark_cfqq_must_dispatch(cfqq); |
760 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 761 | |
7b14e3b5 | 762 | sl = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle); |
206dc69b JA |
763 | |
764 | /* | |
765 | * we don't want to idle for seeks, but we do want to allow | |
766 | * fair distribution of slice time for a process doing back-to-back | |
767 | * seeks. so allow a little bit of time for him to submit a new rq | |
768 | */ | |
caaa5f9f | 769 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
44eb1231 | 770 | sl = min(sl, msecs_to_jiffies(2)); |
206dc69b | 771 | |
7b14e3b5 | 772 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
22e2c507 | 773 | return 1; |
1da177e4 LT |
774 | } |
775 | ||
5e705374 | 776 | static void cfq_dispatch_insert(request_queue_t *q, struct request *rq) |
1da177e4 LT |
777 | { |
778 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
5e705374 | 779 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 780 | |
5380a101 JA |
781 | cfq_remove_request(rq); |
782 | cfqq->on_dispatch[rq_is_sync(rq)]++; | |
783 | elv_dispatch_sort(q, rq); | |
fd61af03 JA |
784 | |
785 | rq = list_entry(q->queue_head.prev, struct request, queuelist); | |
786 | cfqd->last_sector = rq->sector + rq->nr_sectors; | |
1da177e4 LT |
787 | } |
788 | ||
789 | /* | |
790 | * return expired entry, or NULL to just start from scratch in rbtree | |
791 | */ | |
5e705374 | 792 | static inline struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
793 | { |
794 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 795 | struct request *rq; |
1da177e4 | 796 | |
3b18152c | 797 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 LT |
798 | return NULL; |
799 | ||
22e2c507 | 800 | if (!list_empty(&cfqq->fifo)) { |
3b18152c | 801 | int fifo = cfq_cfqq_class_sync(cfqq); |
1da177e4 | 802 | |
5e705374 | 803 | rq = rq_entry_fifo(cfqq->fifo.next); |
22e2c507 | 804 | if (time_after(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) { |
3b18152c | 805 | cfq_mark_cfqq_fifo_expire(cfqq); |
5e705374 | 806 | return rq; |
22e2c507 | 807 | } |
1da177e4 LT |
808 | } |
809 | ||
810 | return NULL; | |
811 | } | |
812 | ||
813 | /* | |
3b18152c JA |
814 | * Scale schedule slice based on io priority. Use the sync time slice only |
815 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
816 | * io only, should not get full sync slice length. | |
1da177e4 | 817 | */ |
22e2c507 JA |
818 | static inline int |
819 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
820 | { | |
821 | const int base_slice = cfqd->cfq_slice[cfq_cfqq_sync(cfqq)]; | |
822 | ||
823 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); | |
824 | ||
825 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - cfqq->ioprio)); | |
826 | } | |
827 | ||
1da177e4 | 828 | static inline void |
22e2c507 | 829 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 830 | { |
22e2c507 JA |
831 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; |
832 | } | |
1da177e4 | 833 | |
22e2c507 JA |
834 | static inline int |
835 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
836 | { | |
837 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 838 | |
22e2c507 | 839 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 840 | |
22e2c507 | 841 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
842 | } |
843 | ||
22e2c507 JA |
844 | /* |
845 | * get next queue for service | |
846 | */ | |
1b5ed5e1 | 847 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 848 | { |
22e2c507 | 849 | unsigned long now = jiffies; |
1da177e4 | 850 | struct cfq_queue *cfqq; |
1da177e4 | 851 | |
22e2c507 JA |
852 | cfqq = cfqd->active_queue; |
853 | if (!cfqq) | |
854 | goto new_queue; | |
1da177e4 | 855 | |
22e2c507 JA |
856 | /* |
857 | * slice has expired | |
858 | */ | |
3b18152c JA |
859 | if (!cfq_cfqq_must_dispatch(cfqq) && time_after(now, cfqq->slice_end)) |
860 | goto expire; | |
1da177e4 | 861 | |
22e2c507 JA |
862 | /* |
863 | * if queue has requests, dispatch one. if not, check if | |
864 | * enough slice is left to wait for one | |
865 | */ | |
dd67d051 | 866 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 867 | goto keep_queue; |
caaa5f9f JA |
868 | else if (cfq_cfqq_dispatched(cfqq)) { |
869 | cfqq = NULL; | |
870 | goto keep_queue; | |
871 | } else if (cfq_cfqq_class_sync(cfqq)) { | |
22e2c507 JA |
872 | if (cfq_arm_slice_timer(cfqd, cfqq)) |
873 | return NULL; | |
874 | } | |
875 | ||
3b18152c | 876 | expire: |
22e2c507 | 877 | cfq_slice_expired(cfqd, 0); |
3b18152c JA |
878 | new_queue: |
879 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 880 | keep_queue: |
3b18152c | 881 | return cfqq; |
22e2c507 JA |
882 | } |
883 | ||
884 | static int | |
885 | __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
886 | int max_dispatch) | |
887 | { | |
888 | int dispatched = 0; | |
889 | ||
dd67d051 | 890 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
891 | |
892 | do { | |
5e705374 | 893 | struct request *rq; |
1da177e4 LT |
894 | |
895 | /* | |
22e2c507 | 896 | * follow expired path, else get first next available |
1da177e4 | 897 | */ |
5e705374 JA |
898 | if ((rq = cfq_check_fifo(cfqq)) == NULL) |
899 | rq = cfqq->next_rq; | |
22e2c507 JA |
900 | |
901 | /* | |
902 | * finally, insert request into driver dispatch list | |
903 | */ | |
5e705374 | 904 | cfq_dispatch_insert(cfqd->queue, rq); |
1da177e4 | 905 | |
22e2c507 JA |
906 | cfqd->dispatch_slice++; |
907 | dispatched++; | |
1da177e4 | 908 | |
22e2c507 | 909 | if (!cfqd->active_cic) { |
5e705374 JA |
910 | atomic_inc(&RQ_CIC(rq)->ioc->refcount); |
911 | cfqd->active_cic = RQ_CIC(rq); | |
22e2c507 | 912 | } |
1da177e4 | 913 | |
dd67d051 | 914 | if (RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 JA |
915 | break; |
916 | ||
917 | } while (dispatched < max_dispatch); | |
918 | ||
919 | /* | |
caaa5f9f | 920 | * if slice end isn't set yet, set it. |
22e2c507 JA |
921 | */ |
922 | if (!cfqq->slice_end) | |
923 | cfq_set_prio_slice(cfqd, cfqq); | |
924 | ||
925 | /* | |
926 | * expire an async queue immediately if it has used up its slice. idle | |
927 | * queue always expire after 1 dispatch round. | |
928 | */ | |
929 | if ((!cfq_cfqq_sync(cfqq) && | |
930 | cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
caaa5f9f JA |
931 | cfq_class_idle(cfqq) || |
932 | !cfq_cfqq_idle_window(cfqq)) | |
22e2c507 JA |
933 | cfq_slice_expired(cfqd, 0); |
934 | ||
935 | return dispatched; | |
936 | } | |
937 | ||
1b5ed5e1 TH |
938 | static int |
939 | cfq_forced_dispatch_cfqqs(struct list_head *list) | |
940 | { | |
1b5ed5e1 | 941 | struct cfq_queue *cfqq, *next; |
caaa5f9f | 942 | int dispatched; |
1b5ed5e1 | 943 | |
caaa5f9f | 944 | dispatched = 0; |
1b5ed5e1 | 945 | list_for_each_entry_safe(cfqq, next, list, cfq_list) { |
5e705374 JA |
946 | while (cfqq->next_rq) { |
947 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1b5ed5e1 TH |
948 | dispatched++; |
949 | } | |
950 | BUG_ON(!list_empty(&cfqq->fifo)); | |
951 | } | |
caaa5f9f | 952 | |
1b5ed5e1 TH |
953 | return dispatched; |
954 | } | |
955 | ||
956 | static int | |
957 | cfq_forced_dispatch(struct cfq_data *cfqd) | |
958 | { | |
959 | int i, dispatched = 0; | |
960 | ||
961 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
962 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->rr_list[i]); | |
963 | ||
964 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->busy_rr); | |
965 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->cur_rr); | |
966 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->idle_rr); | |
967 | ||
968 | cfq_slice_expired(cfqd, 0); | |
969 | ||
970 | BUG_ON(cfqd->busy_queues); | |
971 | ||
972 | return dispatched; | |
973 | } | |
974 | ||
22e2c507 | 975 | static int |
b4878f24 | 976 | cfq_dispatch_requests(request_queue_t *q, int force) |
22e2c507 JA |
977 | { |
978 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
caaa5f9f JA |
979 | struct cfq_queue *cfqq, *prev_cfqq; |
980 | int dispatched; | |
22e2c507 JA |
981 | |
982 | if (!cfqd->busy_queues) | |
983 | return 0; | |
984 | ||
1b5ed5e1 TH |
985 | if (unlikely(force)) |
986 | return cfq_forced_dispatch(cfqd); | |
987 | ||
caaa5f9f JA |
988 | dispatched = 0; |
989 | prev_cfqq = NULL; | |
990 | while ((cfqq = cfq_select_queue(cfqd)) != NULL) { | |
b4878f24 JA |
991 | int max_dispatch; |
992 | ||
caaa5f9f JA |
993 | /* |
994 | * Don't repeat dispatch from the previous queue. | |
995 | */ | |
996 | if (prev_cfqq == cfqq) | |
997 | break; | |
998 | ||
3b18152c JA |
999 | cfq_clear_cfqq_must_dispatch(cfqq); |
1000 | cfq_clear_cfqq_wait_request(cfqq); | |
22e2c507 JA |
1001 | del_timer(&cfqd->idle_slice_timer); |
1002 | ||
1b5ed5e1 TH |
1003 | max_dispatch = cfqd->cfq_quantum; |
1004 | if (cfq_class_idle(cfqq)) | |
1005 | max_dispatch = 1; | |
1da177e4 | 1006 | |
caaa5f9f JA |
1007 | dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); |
1008 | ||
1009 | /* | |
1010 | * If the dispatch cfqq has idling enabled and is still | |
1011 | * the active queue, break out. | |
1012 | */ | |
1013 | if (cfq_cfqq_idle_window(cfqq) && cfqd->active_queue) | |
1014 | break; | |
1015 | ||
1016 | prev_cfqq = cfqq; | |
1da177e4 LT |
1017 | } |
1018 | ||
caaa5f9f | 1019 | return dispatched; |
1da177e4 LT |
1020 | } |
1021 | ||
1da177e4 | 1022 | /* |
5e705374 JA |
1023 | * task holds one reference to the queue, dropped when task exits. each rq |
1024 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1025 | * |
1026 | * queue lock must be held here. | |
1027 | */ | |
1028 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1029 | { | |
22e2c507 JA |
1030 | struct cfq_data *cfqd = cfqq->cfqd; |
1031 | ||
1032 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1033 | |
1034 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1035 | return; | |
1036 | ||
1037 | BUG_ON(rb_first(&cfqq->sort_list)); | |
22e2c507 | 1038 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1039 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1040 | |
7b14e3b5 | 1041 | if (unlikely(cfqd->active_queue == cfqq)) |
3b18152c | 1042 | __cfq_slice_expired(cfqd, cfqq, 0); |
22e2c507 | 1043 | |
1da177e4 LT |
1044 | /* |
1045 | * it's on the empty list and still hashed | |
1046 | */ | |
1047 | list_del(&cfqq->cfq_list); | |
1048 | hlist_del(&cfqq->cfq_hash); | |
1049 | kmem_cache_free(cfq_pool, cfqq); | |
1050 | } | |
1051 | ||
1052 | static inline struct cfq_queue * | |
3b18152c JA |
1053 | __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio, |
1054 | const int hashval) | |
1da177e4 LT |
1055 | { |
1056 | struct hlist_head *hash_list = &cfqd->cfq_hash[hashval]; | |
206dc69b JA |
1057 | struct hlist_node *entry; |
1058 | struct cfq_queue *__cfqq; | |
1da177e4 | 1059 | |
206dc69b | 1060 | hlist_for_each_entry(__cfqq, entry, hash_list, cfq_hash) { |
b0a6916b | 1061 | const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio); |
1da177e4 | 1062 | |
206dc69b | 1063 | if (__cfqq->key == key && (__p == prio || !prio)) |
1da177e4 LT |
1064 | return __cfqq; |
1065 | } | |
1066 | ||
1067 | return NULL; | |
1068 | } | |
1069 | ||
1070 | static struct cfq_queue * | |
3b18152c | 1071 | cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio) |
1da177e4 | 1072 | { |
3b18152c | 1073 | return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT)); |
1da177e4 LT |
1074 | } |
1075 | ||
e2d74ac0 | 1076 | static void cfq_free_io_context(struct io_context *ioc) |
1da177e4 | 1077 | { |
22e2c507 | 1078 | struct cfq_io_context *__cic; |
e2d74ac0 JA |
1079 | struct rb_node *n; |
1080 | int freed = 0; | |
1da177e4 | 1081 | |
e2d74ac0 JA |
1082 | while ((n = rb_first(&ioc->cic_root)) != NULL) { |
1083 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1084 | rb_erase(&__cic->rb_node, &ioc->cic_root); | |
22e2c507 | 1085 | kmem_cache_free(cfq_ioc_pool, __cic); |
334e94de | 1086 | freed++; |
1da177e4 LT |
1087 | } |
1088 | ||
334e94de AV |
1089 | if (atomic_sub_and_test(freed, &ioc_count) && ioc_gone) |
1090 | complete(ioc_gone); | |
1da177e4 LT |
1091 | } |
1092 | ||
e17a9489 AV |
1093 | static void cfq_trim(struct io_context *ioc) |
1094 | { | |
1095 | ioc->set_ioprio = NULL; | |
e2d74ac0 | 1096 | cfq_free_io_context(ioc); |
e17a9489 AV |
1097 | } |
1098 | ||
22e2c507 JA |
1099 | /* |
1100 | * Called with interrupts disabled | |
1101 | */ | |
1102 | static void cfq_exit_single_io_context(struct cfq_io_context *cic) | |
1da177e4 | 1103 | { |
478a82b0 | 1104 | struct cfq_data *cfqd = cic->key; |
d9ff4187 AV |
1105 | request_queue_t *q; |
1106 | ||
1107 | if (!cfqd) | |
1108 | return; | |
1109 | ||
1110 | q = cfqd->queue; | |
22e2c507 JA |
1111 | |
1112 | WARN_ON(!irqs_disabled()); | |
1113 | ||
1114 | spin_lock(q->queue_lock); | |
1115 | ||
12a05732 AV |
1116 | if (cic->cfqq[ASYNC]) { |
1117 | if (unlikely(cic->cfqq[ASYNC] == cfqd->active_queue)) | |
1118 | __cfq_slice_expired(cfqd, cic->cfqq[ASYNC], 0); | |
1119 | cfq_put_queue(cic->cfqq[ASYNC]); | |
1120 | cic->cfqq[ASYNC] = NULL; | |
1121 | } | |
1122 | ||
1123 | if (cic->cfqq[SYNC]) { | |
1124 | if (unlikely(cic->cfqq[SYNC] == cfqd->active_queue)) | |
1125 | __cfq_slice_expired(cfqd, cic->cfqq[SYNC], 0); | |
1126 | cfq_put_queue(cic->cfqq[SYNC]); | |
1127 | cic->cfqq[SYNC] = NULL; | |
1128 | } | |
22e2c507 | 1129 | |
478a82b0 | 1130 | cic->key = NULL; |
d9ff4187 | 1131 | list_del_init(&cic->queue_list); |
22e2c507 | 1132 | spin_unlock(q->queue_lock); |
1da177e4 LT |
1133 | } |
1134 | ||
e2d74ac0 | 1135 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1136 | { |
22e2c507 | 1137 | struct cfq_io_context *__cic; |
1da177e4 | 1138 | unsigned long flags; |
e2d74ac0 | 1139 | struct rb_node *n; |
22e2c507 | 1140 | |
1da177e4 LT |
1141 | /* |
1142 | * put the reference this task is holding to the various queues | |
1143 | */ | |
3793c65c | 1144 | spin_lock_irqsave(&cfq_exit_lock, flags); |
e2d74ac0 JA |
1145 | |
1146 | n = rb_first(&ioc->cic_root); | |
1147 | while (n != NULL) { | |
1148 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1149 | ||
22e2c507 | 1150 | cfq_exit_single_io_context(__cic); |
e2d74ac0 | 1151 | n = rb_next(n); |
1da177e4 LT |
1152 | } |
1153 | ||
3793c65c | 1154 | spin_unlock_irqrestore(&cfq_exit_lock, flags); |
1da177e4 LT |
1155 | } |
1156 | ||
22e2c507 | 1157 | static struct cfq_io_context * |
8267e268 | 1158 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1159 | { |
22e2c507 | 1160 | struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_mask); |
1da177e4 LT |
1161 | |
1162 | if (cic) { | |
553698f9 | 1163 | memset(cic, 0, sizeof(*cic)); |
22e2c507 | 1164 | cic->last_end_request = jiffies; |
553698f9 | 1165 | INIT_LIST_HEAD(&cic->queue_list); |
22e2c507 JA |
1166 | cic->dtor = cfq_free_io_context; |
1167 | cic->exit = cfq_exit_io_context; | |
334e94de | 1168 | atomic_inc(&ioc_count); |
1da177e4 LT |
1169 | } |
1170 | ||
1171 | return cic; | |
1172 | } | |
1173 | ||
22e2c507 JA |
1174 | static void cfq_init_prio_data(struct cfq_queue *cfqq) |
1175 | { | |
1176 | struct task_struct *tsk = current; | |
1177 | int ioprio_class; | |
1178 | ||
3b18152c | 1179 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1180 | return; |
1181 | ||
1182 | ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio); | |
1183 | switch (ioprio_class) { | |
1184 | default: | |
1185 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1186 | case IOPRIO_CLASS_NONE: | |
1187 | /* | |
1188 | * no prio set, place us in the middle of the BE classes | |
1189 | */ | |
1190 | cfqq->ioprio = task_nice_ioprio(tsk); | |
1191 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1192 | break; | |
1193 | case IOPRIO_CLASS_RT: | |
1194 | cfqq->ioprio = task_ioprio(tsk); | |
1195 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1196 | break; | |
1197 | case IOPRIO_CLASS_BE: | |
1198 | cfqq->ioprio = task_ioprio(tsk); | |
1199 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1200 | break; | |
1201 | case IOPRIO_CLASS_IDLE: | |
1202 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1203 | cfqq->ioprio = 7; | |
3b18152c | 1204 | cfq_clear_cfqq_idle_window(cfqq); |
22e2c507 JA |
1205 | break; |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * keep track of original prio settings in case we have to temporarily | |
1210 | * elevate the priority of this queue | |
1211 | */ | |
1212 | cfqq->org_ioprio = cfqq->ioprio; | |
1213 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
1214 | ||
3b18152c | 1215 | if (cfq_cfqq_on_rr(cfqq)) |
22e2c507 JA |
1216 | cfq_resort_rr_list(cfqq, 0); |
1217 | ||
3b18152c | 1218 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1219 | } |
1220 | ||
478a82b0 | 1221 | static inline void changed_ioprio(struct cfq_io_context *cic) |
22e2c507 | 1222 | { |
478a82b0 AV |
1223 | struct cfq_data *cfqd = cic->key; |
1224 | struct cfq_queue *cfqq; | |
35e6077c | 1225 | |
caaa5f9f JA |
1226 | if (unlikely(!cfqd)) |
1227 | return; | |
1228 | ||
1229 | spin_lock(cfqd->queue->queue_lock); | |
1230 | ||
1231 | cfqq = cic->cfqq[ASYNC]; | |
1232 | if (cfqq) { | |
1233 | struct cfq_queue *new_cfqq; | |
1234 | new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC, cic->ioc->task, | |
1235 | GFP_ATOMIC); | |
1236 | if (new_cfqq) { | |
1237 | cic->cfqq[ASYNC] = new_cfqq; | |
1238 | cfq_put_queue(cfqq); | |
1239 | } | |
22e2c507 | 1240 | } |
caaa5f9f JA |
1241 | |
1242 | cfqq = cic->cfqq[SYNC]; | |
1243 | if (cfqq) | |
1244 | cfq_mark_cfqq_prio_changed(cfqq); | |
1245 | ||
1246 | spin_unlock(cfqd->queue->queue_lock); | |
22e2c507 JA |
1247 | } |
1248 | ||
1249 | /* | |
1250 | * callback from sys_ioprio_set, irqs are disabled | |
1251 | */ | |
1252 | static int cfq_ioc_set_ioprio(struct io_context *ioc, unsigned int ioprio) | |
1253 | { | |
a6a0763a | 1254 | struct cfq_io_context *cic; |
e2d74ac0 | 1255 | struct rb_node *n; |
a6a0763a | 1256 | |
3793c65c | 1257 | spin_lock(&cfq_exit_lock); |
a6a0763a | 1258 | |
e2d74ac0 JA |
1259 | n = rb_first(&ioc->cic_root); |
1260 | while (n != NULL) { | |
1261 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
3793c65c | 1262 | |
478a82b0 | 1263 | changed_ioprio(cic); |
e2d74ac0 JA |
1264 | n = rb_next(n); |
1265 | } | |
22e2c507 | 1266 | |
3793c65c | 1267 | spin_unlock(&cfq_exit_lock); |
a6a0763a | 1268 | |
22e2c507 JA |
1269 | return 0; |
1270 | } | |
1271 | ||
1272 | static struct cfq_queue * | |
6f325a13 | 1273 | cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, |
8267e268 | 1274 | gfp_t gfp_mask) |
22e2c507 JA |
1275 | { |
1276 | const int hashval = hash_long(key, CFQ_QHASH_SHIFT); | |
1277 | struct cfq_queue *cfqq, *new_cfqq = NULL; | |
6f325a13 | 1278 | unsigned short ioprio; |
22e2c507 JA |
1279 | |
1280 | retry: | |
6f325a13 | 1281 | ioprio = tsk->ioprio; |
3b18152c | 1282 | cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval); |
22e2c507 JA |
1283 | |
1284 | if (!cfqq) { | |
1285 | if (new_cfqq) { | |
1286 | cfqq = new_cfqq; | |
1287 | new_cfqq = NULL; | |
1288 | } else if (gfp_mask & __GFP_WAIT) { | |
1289 | spin_unlock_irq(cfqd->queue->queue_lock); | |
1290 | new_cfqq = kmem_cache_alloc(cfq_pool, gfp_mask); | |
1291 | spin_lock_irq(cfqd->queue->queue_lock); | |
1292 | goto retry; | |
1293 | } else { | |
1294 | cfqq = kmem_cache_alloc(cfq_pool, gfp_mask); | |
1295 | if (!cfqq) | |
1296 | goto out; | |
1297 | } | |
1298 | ||
1299 | memset(cfqq, 0, sizeof(*cfqq)); | |
1300 | ||
1301 | INIT_HLIST_NODE(&cfqq->cfq_hash); | |
1302 | INIT_LIST_HEAD(&cfqq->cfq_list); | |
22e2c507 JA |
1303 | INIT_LIST_HEAD(&cfqq->fifo); |
1304 | ||
1305 | cfqq->key = key; | |
1306 | hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]); | |
1307 | atomic_set(&cfqq->ref, 0); | |
1308 | cfqq->cfqd = cfqd; | |
22e2c507 JA |
1309 | cfqq->service_last = 0; |
1310 | /* | |
1311 | * set ->slice_left to allow preemption for a new process | |
1312 | */ | |
1313 | cfqq->slice_left = 2 * cfqd->cfq_slice_idle; | |
caaa5f9f | 1314 | cfq_mark_cfqq_idle_window(cfqq); |
3b18152c JA |
1315 | cfq_mark_cfqq_prio_changed(cfqq); |
1316 | cfq_init_prio_data(cfqq); | |
22e2c507 JA |
1317 | } |
1318 | ||
1319 | if (new_cfqq) | |
1320 | kmem_cache_free(cfq_pool, new_cfqq); | |
1321 | ||
1322 | atomic_inc(&cfqq->ref); | |
1323 | out: | |
1324 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1325 | return cfqq; | |
1326 | } | |
1327 | ||
dbecf3ab OH |
1328 | static void |
1329 | cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic) | |
1330 | { | |
3793c65c | 1331 | spin_lock(&cfq_exit_lock); |
dbecf3ab | 1332 | rb_erase(&cic->rb_node, &ioc->cic_root); |
3793c65c JA |
1333 | list_del_init(&cic->queue_list); |
1334 | spin_unlock(&cfq_exit_lock); | |
dbecf3ab OH |
1335 | kmem_cache_free(cfq_ioc_pool, cic); |
1336 | atomic_dec(&ioc_count); | |
1337 | } | |
1338 | ||
e2d74ac0 JA |
1339 | static struct cfq_io_context * |
1340 | cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc) | |
1341 | { | |
dbecf3ab | 1342 | struct rb_node *n; |
e2d74ac0 | 1343 | struct cfq_io_context *cic; |
be3b0753 | 1344 | void *k, *key = cfqd; |
e2d74ac0 | 1345 | |
dbecf3ab OH |
1346 | restart: |
1347 | n = ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1348 | while (n) { |
1349 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1350 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1351 | k = cic->key; | |
1352 | if (unlikely(!k)) { | |
dbecf3ab OH |
1353 | cfq_drop_dead_cic(ioc, cic); |
1354 | goto restart; | |
1355 | } | |
e2d74ac0 | 1356 | |
be3b0753 | 1357 | if (key < k) |
e2d74ac0 | 1358 | n = n->rb_left; |
be3b0753 | 1359 | else if (key > k) |
e2d74ac0 JA |
1360 | n = n->rb_right; |
1361 | else | |
1362 | return cic; | |
1363 | } | |
1364 | ||
1365 | return NULL; | |
1366 | } | |
1367 | ||
1368 | static inline void | |
1369 | cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, | |
1370 | struct cfq_io_context *cic) | |
1371 | { | |
dbecf3ab OH |
1372 | struct rb_node **p; |
1373 | struct rb_node *parent; | |
e2d74ac0 | 1374 | struct cfq_io_context *__cic; |
be3b0753 | 1375 | void *k; |
e2d74ac0 | 1376 | |
e2d74ac0 JA |
1377 | cic->ioc = ioc; |
1378 | cic->key = cfqd; | |
1379 | ||
1380 | ioc->set_ioprio = cfq_ioc_set_ioprio; | |
dbecf3ab OH |
1381 | restart: |
1382 | parent = NULL; | |
1383 | p = &ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1384 | while (*p) { |
1385 | parent = *p; | |
1386 | __cic = rb_entry(parent, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1387 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1388 | k = __cic->key; | |
1389 | if (unlikely(!k)) { | |
be33c3a6 | 1390 | cfq_drop_dead_cic(ioc, __cic); |
dbecf3ab OH |
1391 | goto restart; |
1392 | } | |
e2d74ac0 | 1393 | |
be3b0753 | 1394 | if (cic->key < k) |
e2d74ac0 | 1395 | p = &(*p)->rb_left; |
be3b0753 | 1396 | else if (cic->key > k) |
e2d74ac0 JA |
1397 | p = &(*p)->rb_right; |
1398 | else | |
1399 | BUG(); | |
1400 | } | |
1401 | ||
3793c65c | 1402 | spin_lock(&cfq_exit_lock); |
e2d74ac0 JA |
1403 | rb_link_node(&cic->rb_node, parent, p); |
1404 | rb_insert_color(&cic->rb_node, &ioc->cic_root); | |
1405 | list_add(&cic->queue_list, &cfqd->cic_list); | |
3793c65c | 1406 | spin_unlock(&cfq_exit_lock); |
e2d74ac0 JA |
1407 | } |
1408 | ||
1da177e4 LT |
1409 | /* |
1410 | * Setup general io context and cfq io context. There can be several cfq | |
1411 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1412 | * than one device managed by cfq. |
1da177e4 LT |
1413 | */ |
1414 | static struct cfq_io_context * | |
e2d74ac0 | 1415 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1416 | { |
22e2c507 | 1417 | struct io_context *ioc = NULL; |
1da177e4 | 1418 | struct cfq_io_context *cic; |
1da177e4 | 1419 | |
22e2c507 | 1420 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1421 | |
22e2c507 | 1422 | ioc = get_io_context(gfp_mask); |
1da177e4 LT |
1423 | if (!ioc) |
1424 | return NULL; | |
1425 | ||
e2d74ac0 JA |
1426 | cic = cfq_cic_rb_lookup(cfqd, ioc); |
1427 | if (cic) | |
1428 | goto out; | |
1da177e4 | 1429 | |
e2d74ac0 JA |
1430 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1431 | if (cic == NULL) | |
1432 | goto err; | |
1da177e4 | 1433 | |
e2d74ac0 | 1434 | cfq_cic_link(cfqd, ioc, cic); |
1da177e4 | 1435 | out: |
1da177e4 LT |
1436 | return cic; |
1437 | err: | |
1438 | put_io_context(ioc); | |
1439 | return NULL; | |
1440 | } | |
1441 | ||
22e2c507 JA |
1442 | static void |
1443 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1444 | { |
22e2c507 | 1445 | unsigned long elapsed, ttime; |
1da177e4 | 1446 | |
22e2c507 JA |
1447 | /* |
1448 | * if this context already has stuff queued, thinktime is from | |
1449 | * last queue not last end | |
1450 | */ | |
1451 | #if 0 | |
1452 | if (time_after(cic->last_end_request, cic->last_queue)) | |
1453 | elapsed = jiffies - cic->last_end_request; | |
1454 | else | |
1455 | elapsed = jiffies - cic->last_queue; | |
1456 | #else | |
1457 | elapsed = jiffies - cic->last_end_request; | |
1458 | #endif | |
1da177e4 | 1459 | |
22e2c507 | 1460 | ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); |
db3b5848 | 1461 | |
22e2c507 JA |
1462 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1463 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1464 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1465 | } | |
1da177e4 | 1466 | |
206dc69b JA |
1467 | static void |
1468 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic, | |
5e705374 | 1469 | struct request *rq) |
206dc69b JA |
1470 | { |
1471 | sector_t sdist; | |
1472 | u64 total; | |
1473 | ||
5e705374 JA |
1474 | if (cic->last_request_pos < rq->sector) |
1475 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1476 | else |
5e705374 | 1477 | sdist = cic->last_request_pos - rq->sector; |
206dc69b JA |
1478 | |
1479 | /* | |
1480 | * Don't allow the seek distance to get too large from the | |
1481 | * odd fragment, pagein, etc | |
1482 | */ | |
1483 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1484 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1485 | else | |
1486 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1487 | ||
1488 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1489 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1490 | total = cic->seek_total + (cic->seek_samples/2); | |
1491 | do_div(total, cic->seek_samples); | |
1492 | cic->seek_mean = (sector_t)total; | |
1493 | } | |
1da177e4 | 1494 | |
22e2c507 JA |
1495 | /* |
1496 | * Disable idle window if the process thinks too long or seeks so much that | |
1497 | * it doesn't matter | |
1498 | */ | |
1499 | static void | |
1500 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1501 | struct cfq_io_context *cic) | |
1502 | { | |
3b18152c | 1503 | int enable_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 1504 | |
caaa5f9f JA |
1505 | if (!cic->ioc->task || !cfqd->cfq_slice_idle || |
1506 | (cfqd->hw_tag && CIC_SEEKY(cic))) | |
22e2c507 JA |
1507 | enable_idle = 0; |
1508 | else if (sample_valid(cic->ttime_samples)) { | |
1509 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1510 | enable_idle = 0; | |
1511 | else | |
1512 | enable_idle = 1; | |
1da177e4 LT |
1513 | } |
1514 | ||
3b18152c JA |
1515 | if (enable_idle) |
1516 | cfq_mark_cfqq_idle_window(cfqq); | |
1517 | else | |
1518 | cfq_clear_cfqq_idle_window(cfqq); | |
22e2c507 | 1519 | } |
1da177e4 | 1520 | |
22e2c507 JA |
1521 | |
1522 | /* | |
1523 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1524 | * no or if we aren't sure, a 1 will cause a preempt. | |
1525 | */ | |
1526 | static int | |
1527 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1528 | struct request *rq) |
22e2c507 JA |
1529 | { |
1530 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1531 | ||
1532 | if (cfq_class_idle(new_cfqq)) | |
1533 | return 0; | |
1534 | ||
1535 | if (!cfqq) | |
caaa5f9f | 1536 | return 0; |
22e2c507 JA |
1537 | |
1538 | if (cfq_class_idle(cfqq)) | |
1539 | return 1; | |
3b18152c | 1540 | if (!cfq_cfqq_wait_request(new_cfqq)) |
22e2c507 JA |
1541 | return 0; |
1542 | /* | |
1543 | * if it doesn't have slice left, forget it | |
1544 | */ | |
1545 | if (new_cfqq->slice_left < cfqd->cfq_slice_idle) | |
1546 | return 0; | |
5e705374 | 1547 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 JA |
1548 | return 1; |
1549 | ||
1550 | return 0; | |
1551 | } | |
1552 | ||
1553 | /* | |
1554 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1555 | * let it have half of its nominal slice. | |
1556 | */ | |
1557 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1558 | { | |
1559 | struct cfq_queue *__cfqq, *next; | |
1560 | ||
1561 | list_for_each_entry_safe(__cfqq, next, &cfqd->cur_rr, cfq_list) | |
1562 | cfq_resort_rr_list(__cfqq, 1); | |
1563 | ||
1564 | if (!cfqq->slice_left) | |
1565 | cfqq->slice_left = cfq_prio_to_slice(cfqd, cfqq) / 2; | |
1566 | ||
1567 | cfqq->slice_end = cfqq->slice_left + jiffies; | |
caaa5f9f | 1568 | cfq_slice_expired(cfqd, 1); |
22e2c507 JA |
1569 | __cfq_set_active_queue(cfqd, cfqq); |
1570 | } | |
1571 | ||
1572 | /* | |
1573 | * should really be a ll_rw_blk.c helper | |
1574 | */ | |
1575 | static void cfq_start_queueing(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1576 | { | |
1577 | request_queue_t *q = cfqd->queue; | |
1578 | ||
1579 | if (!blk_queue_plugged(q)) | |
1580 | q->request_fn(q); | |
1581 | else | |
1582 | __generic_unplug_device(q); | |
1583 | } | |
1584 | ||
1585 | /* | |
5e705374 | 1586 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1587 | * something we should do about it |
1588 | */ | |
1589 | static void | |
5e705374 JA |
1590 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1591 | struct request *rq) | |
22e2c507 | 1592 | { |
5e705374 | 1593 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1594 | |
21183b07 | 1595 | /* |
5380a101 | 1596 | * check if this request is a better next-serve candidate)) { |
21183b07 | 1597 | */ |
5e705374 JA |
1598 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); |
1599 | BUG_ON(!cfqq->next_rq); | |
21183b07 | 1600 | |
9c2c38a1 JA |
1601 | /* |
1602 | * we never wait for an async request and we don't allow preemption | |
1603 | * of an async request. so just return early | |
1604 | */ | |
5e705374 | 1605 | if (!rq_is_sync(rq)) { |
12e9fddd JA |
1606 | /* |
1607 | * sync process issued an async request, if it's waiting | |
1608 | * then expire it and kick rq handling. | |
1609 | */ | |
1610 | if (cic == cfqd->active_cic && | |
1611 | del_timer(&cfqd->idle_slice_timer)) { | |
1612 | cfq_slice_expired(cfqd, 0); | |
1613 | cfq_start_queueing(cfqd, cfqq); | |
1614 | } | |
9c2c38a1 | 1615 | return; |
12e9fddd | 1616 | } |
22e2c507 | 1617 | |
9c2c38a1 | 1618 | cfq_update_io_thinktime(cfqd, cic); |
5e705374 | 1619 | cfq_update_io_seektime(cfqd, cic, rq); |
9c2c38a1 JA |
1620 | cfq_update_idle_window(cfqd, cfqq, cic); |
1621 | ||
1622 | cic->last_queue = jiffies; | |
5e705374 | 1623 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
22e2c507 JA |
1624 | |
1625 | if (cfqq == cfqd->active_queue) { | |
1626 | /* | |
1627 | * if we are waiting for a request for this queue, let it rip | |
1628 | * immediately and flag that we must not expire this queue | |
1629 | * just now | |
1630 | */ | |
3b18152c JA |
1631 | if (cfq_cfqq_wait_request(cfqq)) { |
1632 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 JA |
1633 | del_timer(&cfqd->idle_slice_timer); |
1634 | cfq_start_queueing(cfqd, cfqq); | |
1635 | } | |
5e705374 | 1636 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1637 | /* |
1638 | * not the active queue - expire current slice if it is | |
1639 | * idle and has expired it's mean thinktime or this new queue | |
1640 | * has some old slice time left and is of higher priority | |
1641 | */ | |
1642 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1643 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
1644 | cfq_start_queueing(cfqd, cfqq); |
1645 | } | |
1da177e4 LT |
1646 | } |
1647 | ||
b4878f24 | 1648 | static void cfq_insert_request(request_queue_t *q, struct request *rq) |
1da177e4 | 1649 | { |
b4878f24 | 1650 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1651 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1652 | |
1653 | cfq_init_prio_data(cfqq); | |
1da177e4 | 1654 | |
5e705374 | 1655 | cfq_add_rq_rb(rq); |
1da177e4 | 1656 | |
21183b07 JA |
1657 | if (!cfq_cfqq_on_rr(cfqq)) |
1658 | cfq_add_cfqq_rr(cfqd, cfqq); | |
1659 | ||
22e2c507 JA |
1660 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1661 | ||
5e705374 | 1662 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1663 | } |
1664 | ||
1da177e4 LT |
1665 | static void cfq_completed_request(request_queue_t *q, struct request *rq) |
1666 | { | |
5e705374 | 1667 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1668 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1669 | const int sync = rq_is_sync(rq); |
b4878f24 | 1670 | unsigned long now; |
1da177e4 | 1671 | |
b4878f24 | 1672 | now = jiffies; |
1da177e4 | 1673 | |
b4878f24 JA |
1674 | WARN_ON(!cfqd->rq_in_driver); |
1675 | WARN_ON(!cfqq->on_dispatch[sync]); | |
1676 | cfqd->rq_in_driver--; | |
1677 | cfqq->on_dispatch[sync]--; | |
1da177e4 | 1678 | |
b4878f24 JA |
1679 | if (!cfq_class_idle(cfqq)) |
1680 | cfqd->last_end_request = now; | |
3b18152c | 1681 | |
b4878f24 JA |
1682 | if (!cfq_cfqq_dispatched(cfqq)) { |
1683 | if (cfq_cfqq_on_rr(cfqq)) { | |
1684 | cfqq->service_last = now; | |
1685 | cfq_resort_rr_list(cfqq, 0); | |
1686 | } | |
1da177e4 LT |
1687 | } |
1688 | ||
caaa5f9f | 1689 | if (sync) |
5e705374 | 1690 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1691 | |
1692 | /* | |
1693 | * If this is the active queue, check if it needs to be expired, | |
1694 | * or if we want to idle in case it has no pending requests. | |
1695 | */ | |
1696 | if (cfqd->active_queue == cfqq) { | |
1697 | if (time_after(now, cfqq->slice_end)) | |
1698 | cfq_slice_expired(cfqd, 0); | |
dd67d051 | 1699 | else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
caaa5f9f JA |
1700 | if (!cfq_arm_slice_timer(cfqd, cfqq)) |
1701 | cfq_schedule_dispatch(cfqd); | |
1702 | } | |
1703 | } | |
1da177e4 LT |
1704 | } |
1705 | ||
22e2c507 JA |
1706 | /* |
1707 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
1708 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
1709 | */ | |
1710 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 1711 | { |
22e2c507 JA |
1712 | const int ioprio_class = cfqq->ioprio_class; |
1713 | const int ioprio = cfqq->ioprio; | |
1da177e4 | 1714 | |
22e2c507 JA |
1715 | if (has_fs_excl()) { |
1716 | /* | |
1717 | * boost idle prio on transactions that would lock out other | |
1718 | * users of the filesystem | |
1719 | */ | |
1720 | if (cfq_class_idle(cfqq)) | |
1721 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1722 | if (cfqq->ioprio > IOPRIO_NORM) | |
1723 | cfqq->ioprio = IOPRIO_NORM; | |
1724 | } else { | |
1725 | /* | |
1726 | * check if we need to unboost the queue | |
1727 | */ | |
1728 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
1729 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
1730 | if (cfqq->ioprio != cfqq->org_ioprio) | |
1731 | cfqq->ioprio = cfqq->org_ioprio; | |
1732 | } | |
1da177e4 | 1733 | |
22e2c507 JA |
1734 | /* |
1735 | * refile between round-robin lists if we moved the priority class | |
1736 | */ | |
1737 | if ((ioprio_class != cfqq->ioprio_class || ioprio != cfqq->ioprio) && | |
3b18152c | 1738 | cfq_cfqq_on_rr(cfqq)) |
22e2c507 JA |
1739 | cfq_resort_rr_list(cfqq, 0); |
1740 | } | |
1da177e4 | 1741 | |
22e2c507 JA |
1742 | static inline int |
1743 | __cfq_may_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1744 | struct task_struct *task, int rw) | |
1745 | { | |
3b18152c | 1746 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 1747 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 1748 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 1749 | return ELV_MQUEUE_MUST; |
3b18152c | 1750 | } |
1da177e4 | 1751 | |
22e2c507 | 1752 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
1753 | } |
1754 | ||
1755 | static int cfq_may_queue(request_queue_t *q, int rw, struct bio *bio) | |
1756 | { | |
1757 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1758 | struct task_struct *tsk = current; | |
1759 | struct cfq_queue *cfqq; | |
1760 | ||
1761 | /* | |
1762 | * don't force setup of a queue from here, as a call to may_queue | |
1763 | * does not necessarily imply that a request actually will be queued. | |
1764 | * so just lookup a possibly existing queue, or return 'may queue' | |
1765 | * if that fails | |
1766 | */ | |
3b18152c | 1767 | cfqq = cfq_find_cfq_hash(cfqd, cfq_queue_pid(tsk, rw), tsk->ioprio); |
22e2c507 JA |
1768 | if (cfqq) { |
1769 | cfq_init_prio_data(cfqq); | |
1770 | cfq_prio_boost(cfqq); | |
1771 | ||
1772 | return __cfq_may_queue(cfqd, cfqq, tsk, rw); | |
1773 | } | |
1774 | ||
1775 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
1776 | } |
1777 | ||
1778 | static void cfq_check_waiters(request_queue_t *q, struct cfq_queue *cfqq) | |
1779 | { | |
22e2c507 | 1780 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1da177e4 | 1781 | |
271f18f1 JA |
1782 | if (unlikely(cfqd->rq_starved)) { |
1783 | struct request_list *rl = &q->rq; | |
1784 | ||
22e2c507 JA |
1785 | smp_mb(); |
1786 | if (waitqueue_active(&rl->wait[READ])) | |
1787 | wake_up(&rl->wait[READ]); | |
22e2c507 JA |
1788 | if (waitqueue_active(&rl->wait[WRITE])) |
1789 | wake_up(&rl->wait[WRITE]); | |
1790 | } | |
1da177e4 LT |
1791 | } |
1792 | ||
1793 | /* | |
1794 | * queue lock held here | |
1795 | */ | |
1796 | static void cfq_put_request(request_queue_t *q, struct request *rq) | |
1797 | { | |
5e705374 | 1798 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1799 | |
5e705374 | 1800 | if (cfqq) { |
22e2c507 | 1801 | const int rw = rq_data_dir(rq); |
1da177e4 | 1802 | |
22e2c507 JA |
1803 | BUG_ON(!cfqq->allocated[rw]); |
1804 | cfqq->allocated[rw]--; | |
1da177e4 | 1805 | |
5e705374 | 1806 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 1807 | |
1da177e4 | 1808 | rq->elevator_private = NULL; |
5e705374 | 1809 | rq->elevator_private2 = NULL; |
1da177e4 | 1810 | |
1da177e4 LT |
1811 | cfq_check_waiters(q, cfqq); |
1812 | cfq_put_queue(cfqq); | |
1813 | } | |
1814 | } | |
1815 | ||
1816 | /* | |
22e2c507 | 1817 | * Allocate cfq data structures associated with this request. |
1da177e4 | 1818 | */ |
22e2c507 JA |
1819 | static int |
1820 | cfq_set_request(request_queue_t *q, struct request *rq, struct bio *bio, | |
8267e268 | 1821 | gfp_t gfp_mask) |
1da177e4 LT |
1822 | { |
1823 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3b18152c | 1824 | struct task_struct *tsk = current; |
1da177e4 LT |
1825 | struct cfq_io_context *cic; |
1826 | const int rw = rq_data_dir(rq); | |
3b18152c | 1827 | pid_t key = cfq_queue_pid(tsk, rw); |
22e2c507 | 1828 | struct cfq_queue *cfqq; |
1da177e4 | 1829 | unsigned long flags; |
12a05732 | 1830 | int is_sync = key != CFQ_KEY_ASYNC; |
1da177e4 LT |
1831 | |
1832 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1833 | ||
e2d74ac0 | 1834 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 1835 | |
1da177e4 LT |
1836 | spin_lock_irqsave(q->queue_lock, flags); |
1837 | ||
22e2c507 JA |
1838 | if (!cic) |
1839 | goto queue_fail; | |
1840 | ||
12a05732 | 1841 | if (!cic->cfqq[is_sync]) { |
6f325a13 | 1842 | cfqq = cfq_get_queue(cfqd, key, tsk, gfp_mask); |
22e2c507 JA |
1843 | if (!cfqq) |
1844 | goto queue_fail; | |
1da177e4 | 1845 | |
12a05732 | 1846 | cic->cfqq[is_sync] = cfqq; |
22e2c507 | 1847 | } else |
12a05732 | 1848 | cfqq = cic->cfqq[is_sync]; |
1da177e4 LT |
1849 | |
1850 | cfqq->allocated[rw]++; | |
3b18152c | 1851 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 JA |
1852 | cfqd->rq_starved = 0; |
1853 | atomic_inc(&cfqq->ref); | |
1da177e4 | 1854 | |
5e705374 | 1855 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 1856 | |
5e705374 JA |
1857 | rq->elevator_private = cic; |
1858 | rq->elevator_private2 = cfqq; | |
1859 | return 0; | |
1da177e4 | 1860 | |
22e2c507 JA |
1861 | queue_fail: |
1862 | if (cic) | |
1863 | put_io_context(cic->ioc); | |
1864 | /* | |
1865 | * mark us rq allocation starved. we need to kickstart the process | |
1866 | * ourselves if there are no pending requests that can do it for us. | |
1867 | * that would be an extremely rare OOM situation | |
1868 | */ | |
1869 | cfqd->rq_starved = 1; | |
3b18152c | 1870 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
1871 | spin_unlock_irqrestore(q->queue_lock, flags); |
1872 | return 1; | |
1873 | } | |
1874 | ||
22e2c507 JA |
1875 | static void cfq_kick_queue(void *data) |
1876 | { | |
1877 | request_queue_t *q = data; | |
1878 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1879 | unsigned long flags; | |
1880 | ||
1881 | spin_lock_irqsave(q->queue_lock, flags); | |
1882 | ||
1883 | if (cfqd->rq_starved) { | |
1884 | struct request_list *rl = &q->rq; | |
1885 | ||
1886 | /* | |
1887 | * we aren't guaranteed to get a request after this, but we | |
1888 | * have to be opportunistic | |
1889 | */ | |
1890 | smp_mb(); | |
1891 | if (waitqueue_active(&rl->wait[READ])) | |
1892 | wake_up(&rl->wait[READ]); | |
1893 | if (waitqueue_active(&rl->wait[WRITE])) | |
1894 | wake_up(&rl->wait[WRITE]); | |
1895 | } | |
1896 | ||
1897 | blk_remove_plug(q); | |
1898 | q->request_fn(q); | |
1899 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1900 | } | |
1901 | ||
1902 | /* | |
1903 | * Timer running if the active_queue is currently idling inside its time slice | |
1904 | */ | |
1905 | static void cfq_idle_slice_timer(unsigned long data) | |
1906 | { | |
1907 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1908 | struct cfq_queue *cfqq; | |
1909 | unsigned long flags; | |
1910 | ||
1911 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1912 | ||
1913 | if ((cfqq = cfqd->active_queue) != NULL) { | |
1914 | unsigned long now = jiffies; | |
1915 | ||
1916 | /* | |
1917 | * expired | |
1918 | */ | |
1919 | if (time_after(now, cfqq->slice_end)) | |
1920 | goto expire; | |
1921 | ||
1922 | /* | |
1923 | * only expire and reinvoke request handler, if there are | |
1924 | * other queues with pending requests | |
1925 | */ | |
caaa5f9f | 1926 | if (!cfqd->busy_queues) |
22e2c507 | 1927 | goto out_cont; |
22e2c507 JA |
1928 | |
1929 | /* | |
1930 | * not expired and it has a request pending, let it dispatch | |
1931 | */ | |
dd67d051 | 1932 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 1933 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
1934 | goto out_kick; |
1935 | } | |
1936 | } | |
1937 | expire: | |
1938 | cfq_slice_expired(cfqd, 0); | |
1939 | out_kick: | |
3b18152c | 1940 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
1941 | out_cont: |
1942 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1943 | } | |
1944 | ||
1945 | /* | |
1946 | * Timer running if an idle class queue is waiting for service | |
1947 | */ | |
1948 | static void cfq_idle_class_timer(unsigned long data) | |
1949 | { | |
1950 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1951 | unsigned long flags, end; | |
1952 | ||
1953 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1954 | ||
1955 | /* | |
1956 | * race with a non-idle queue, reset timer | |
1957 | */ | |
1958 | end = cfqd->last_end_request + CFQ_IDLE_GRACE; | |
ae818a38 JA |
1959 | if (!time_after_eq(jiffies, end)) |
1960 | mod_timer(&cfqd->idle_class_timer, end); | |
1961 | else | |
3b18152c | 1962 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
1963 | |
1964 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1965 | } | |
1966 | ||
3b18152c JA |
1967 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
1968 | { | |
1969 | del_timer_sync(&cfqd->idle_slice_timer); | |
1970 | del_timer_sync(&cfqd->idle_class_timer); | |
1971 | blk_sync_queue(cfqd->queue); | |
1972 | } | |
22e2c507 | 1973 | |
1da177e4 LT |
1974 | static void cfq_exit_queue(elevator_t *e) |
1975 | { | |
22e2c507 | 1976 | struct cfq_data *cfqd = e->elevator_data; |
d9ff4187 | 1977 | request_queue_t *q = cfqd->queue; |
22e2c507 | 1978 | |
3b18152c | 1979 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 1980 | |
3793c65c | 1981 | spin_lock(&cfq_exit_lock); |
d9ff4187 | 1982 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 1983 | |
d9ff4187 AV |
1984 | if (cfqd->active_queue) |
1985 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); | |
e2d74ac0 JA |
1986 | |
1987 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
1988 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
1989 | struct cfq_io_context, | |
1990 | queue_list); | |
1991 | if (cic->cfqq[ASYNC]) { | |
1992 | cfq_put_queue(cic->cfqq[ASYNC]); | |
1993 | cic->cfqq[ASYNC] = NULL; | |
1994 | } | |
1995 | if (cic->cfqq[SYNC]) { | |
1996 | cfq_put_queue(cic->cfqq[SYNC]); | |
1997 | cic->cfqq[SYNC] = NULL; | |
1998 | } | |
1999 | cic->key = NULL; | |
2000 | list_del_init(&cic->queue_list); | |
2001 | } | |
e2d74ac0 | 2002 | |
d9ff4187 | 2003 | spin_unlock_irq(q->queue_lock); |
3793c65c | 2004 | spin_unlock(&cfq_exit_lock); |
a90d742e AV |
2005 | |
2006 | cfq_shutdown_timer_wq(cfqd); | |
2007 | ||
a90d742e AV |
2008 | kfree(cfqd->cfq_hash); |
2009 | kfree(cfqd); | |
1da177e4 LT |
2010 | } |
2011 | ||
bc1c1169 | 2012 | static void *cfq_init_queue(request_queue_t *q, elevator_t *e) |
1da177e4 LT |
2013 | { |
2014 | struct cfq_data *cfqd; | |
2015 | int i; | |
2016 | ||
2017 | cfqd = kmalloc(sizeof(*cfqd), GFP_KERNEL); | |
2018 | if (!cfqd) | |
bc1c1169 | 2019 | return NULL; |
1da177e4 LT |
2020 | |
2021 | memset(cfqd, 0, sizeof(*cfqd)); | |
22e2c507 JA |
2022 | |
2023 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
2024 | INIT_LIST_HEAD(&cfqd->rr_list[i]); | |
2025 | ||
2026 | INIT_LIST_HEAD(&cfqd->busy_rr); | |
2027 | INIT_LIST_HEAD(&cfqd->cur_rr); | |
2028 | INIT_LIST_HEAD(&cfqd->idle_rr); | |
1da177e4 | 2029 | INIT_LIST_HEAD(&cfqd->empty_list); |
d9ff4187 | 2030 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2031 | |
1da177e4 LT |
2032 | cfqd->cfq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_QHASH_ENTRIES, GFP_KERNEL); |
2033 | if (!cfqd->cfq_hash) | |
5e705374 | 2034 | goto out_free; |
1da177e4 | 2035 | |
1da177e4 LT |
2036 | for (i = 0; i < CFQ_QHASH_ENTRIES; i++) |
2037 | INIT_HLIST_HEAD(&cfqd->cfq_hash[i]); | |
2038 | ||
1da177e4 | 2039 | cfqd->queue = q; |
1da177e4 | 2040 | |
22e2c507 JA |
2041 | init_timer(&cfqd->idle_slice_timer); |
2042 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2043 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2044 | ||
2045 | init_timer(&cfqd->idle_class_timer); | |
2046 | cfqd->idle_class_timer.function = cfq_idle_class_timer; | |
2047 | cfqd->idle_class_timer.data = (unsigned long) cfqd; | |
2048 | ||
2049 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue, q); | |
2050 | ||
1da177e4 LT |
2051 | cfqd->cfq_queued = cfq_queued; |
2052 | cfqd->cfq_quantum = cfq_quantum; | |
22e2c507 JA |
2053 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2054 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2055 | cfqd->cfq_back_max = cfq_back_max; |
2056 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2057 | cfqd->cfq_slice[0] = cfq_slice_async; |
2058 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2059 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2060 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
3b18152c | 2061 | |
bc1c1169 | 2062 | return cfqd; |
5e705374 | 2063 | out_free: |
1da177e4 | 2064 | kfree(cfqd); |
bc1c1169 | 2065 | return NULL; |
1da177e4 LT |
2066 | } |
2067 | ||
2068 | static void cfq_slab_kill(void) | |
2069 | { | |
1da177e4 LT |
2070 | if (cfq_pool) |
2071 | kmem_cache_destroy(cfq_pool); | |
2072 | if (cfq_ioc_pool) | |
2073 | kmem_cache_destroy(cfq_ioc_pool); | |
2074 | } | |
2075 | ||
2076 | static int __init cfq_slab_setup(void) | |
2077 | { | |
1da177e4 LT |
2078 | cfq_pool = kmem_cache_create("cfq_pool", sizeof(struct cfq_queue), 0, 0, |
2079 | NULL, NULL); | |
2080 | if (!cfq_pool) | |
2081 | goto fail; | |
2082 | ||
2083 | cfq_ioc_pool = kmem_cache_create("cfq_ioc_pool", | |
2084 | sizeof(struct cfq_io_context), 0, 0, NULL, NULL); | |
2085 | if (!cfq_ioc_pool) | |
2086 | goto fail; | |
2087 | ||
2088 | return 0; | |
2089 | fail: | |
2090 | cfq_slab_kill(); | |
2091 | return -ENOMEM; | |
2092 | } | |
2093 | ||
1da177e4 LT |
2094 | /* |
2095 | * sysfs parts below --> | |
2096 | */ | |
1da177e4 LT |
2097 | |
2098 | static ssize_t | |
2099 | cfq_var_show(unsigned int var, char *page) | |
2100 | { | |
2101 | return sprintf(page, "%d\n", var); | |
2102 | } | |
2103 | ||
2104 | static ssize_t | |
2105 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2106 | { | |
2107 | char *p = (char *) page; | |
2108 | ||
2109 | *var = simple_strtoul(p, &p, 10); | |
2110 | return count; | |
2111 | } | |
2112 | ||
1da177e4 | 2113 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
3d1ab40f | 2114 | static ssize_t __FUNC(elevator_t *e, char *page) \ |
1da177e4 | 2115 | { \ |
3d1ab40f | 2116 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2117 | unsigned int __data = __VAR; \ |
2118 | if (__CONV) \ | |
2119 | __data = jiffies_to_msecs(__data); \ | |
2120 | return cfq_var_show(__data, (page)); \ | |
2121 | } | |
2122 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
2123 | SHOW_FUNCTION(cfq_queued_show, cfqd->cfq_queued, 0); | |
22e2c507 JA |
2124 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2125 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2126 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2127 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2128 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2129 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2130 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2131 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2132 | #undef SHOW_FUNCTION |
2133 | ||
2134 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
3d1ab40f | 2135 | static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \ |
1da177e4 | 2136 | { \ |
3d1ab40f | 2137 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2138 | unsigned int __data; \ |
2139 | int ret = cfq_var_store(&__data, (page), count); \ | |
2140 | if (__data < (MIN)) \ | |
2141 | __data = (MIN); \ | |
2142 | else if (__data > (MAX)) \ | |
2143 | __data = (MAX); \ | |
2144 | if (__CONV) \ | |
2145 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2146 | else \ | |
2147 | *(__PTR) = __data; \ | |
2148 | return ret; \ | |
2149 | } | |
2150 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
2151 | STORE_FUNCTION(cfq_queued_store, &cfqd->cfq_queued, 1, UINT_MAX, 0); | |
22e2c507 JA |
2152 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1); |
2153 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1); | |
e572ec7e AV |
2154 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
2155 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0); | |
22e2c507 JA |
2156 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2157 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2158 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
2159 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0); | |
1da177e4 LT |
2160 | #undef STORE_FUNCTION |
2161 | ||
e572ec7e AV |
2162 | #define CFQ_ATTR(name) \ |
2163 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2164 | ||
2165 | static struct elv_fs_entry cfq_attrs[] = { | |
2166 | CFQ_ATTR(quantum), | |
2167 | CFQ_ATTR(queued), | |
2168 | CFQ_ATTR(fifo_expire_sync), | |
2169 | CFQ_ATTR(fifo_expire_async), | |
2170 | CFQ_ATTR(back_seek_max), | |
2171 | CFQ_ATTR(back_seek_penalty), | |
2172 | CFQ_ATTR(slice_sync), | |
2173 | CFQ_ATTR(slice_async), | |
2174 | CFQ_ATTR(slice_async_rq), | |
2175 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2176 | __ATTR_NULL |
1da177e4 LT |
2177 | }; |
2178 | ||
1da177e4 LT |
2179 | static struct elevator_type iosched_cfq = { |
2180 | .ops = { | |
2181 | .elevator_merge_fn = cfq_merge, | |
2182 | .elevator_merged_fn = cfq_merged_request, | |
2183 | .elevator_merge_req_fn = cfq_merged_requests, | |
b4878f24 | 2184 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2185 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2186 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2187 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2188 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2189 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2190 | .elevator_former_req_fn = elv_rb_former_request, |
2191 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2192 | .elevator_set_req_fn = cfq_set_request, |
2193 | .elevator_put_req_fn = cfq_put_request, | |
2194 | .elevator_may_queue_fn = cfq_may_queue, | |
2195 | .elevator_init_fn = cfq_init_queue, | |
2196 | .elevator_exit_fn = cfq_exit_queue, | |
e17a9489 | 2197 | .trim = cfq_trim, |
1da177e4 | 2198 | }, |
3d1ab40f | 2199 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2200 | .elevator_name = "cfq", |
2201 | .elevator_owner = THIS_MODULE, | |
2202 | }; | |
2203 | ||
2204 | static int __init cfq_init(void) | |
2205 | { | |
2206 | int ret; | |
2207 | ||
22e2c507 JA |
2208 | /* |
2209 | * could be 0 on HZ < 1000 setups | |
2210 | */ | |
2211 | if (!cfq_slice_async) | |
2212 | cfq_slice_async = 1; | |
2213 | if (!cfq_slice_idle) | |
2214 | cfq_slice_idle = 1; | |
2215 | ||
1da177e4 LT |
2216 | if (cfq_slab_setup()) |
2217 | return -ENOMEM; | |
2218 | ||
2219 | ret = elv_register(&iosched_cfq); | |
22e2c507 JA |
2220 | if (ret) |
2221 | cfq_slab_kill(); | |
1da177e4 | 2222 | |
1da177e4 LT |
2223 | return ret; |
2224 | } | |
2225 | ||
2226 | static void __exit cfq_exit(void) | |
2227 | { | |
334e94de | 2228 | DECLARE_COMPLETION(all_gone); |
1da177e4 | 2229 | elv_unregister(&iosched_cfq); |
334e94de | 2230 | ioc_gone = &all_gone; |
fba82272 OH |
2231 | /* ioc_gone's update must be visible before reading ioc_count */ |
2232 | smp_wmb(); | |
334e94de | 2233 | if (atomic_read(&ioc_count)) |
fba82272 | 2234 | wait_for_completion(ioc_gone); |
334e94de | 2235 | synchronize_rcu(); |
83521d3e | 2236 | cfq_slab_kill(); |
1da177e4 LT |
2237 | } |
2238 | ||
2239 | module_init(cfq_init); | |
2240 | module_exit(cfq_exit); | |
2241 | ||
2242 | MODULE_AUTHOR("Jens Axboe"); | |
2243 | MODULE_LICENSE("GPL"); | |
2244 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |