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_end; |
150 | unsigned long slice_left; | |
99f9628a | 151 | unsigned long service_last; |
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 | 164 | enum cfqq_state_flags { |
b0b8d749 JA |
165 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
166 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
167 | CFQ_CFQQ_FLAG_must_alloc, /* must be allowed rq alloc */ | |
168 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ | |
169 | CFQ_CFQQ_FLAG_must_dispatch, /* must dispatch, even if expired */ | |
170 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ | |
171 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
172 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
173 | CFQ_CFQQ_FLAG_queue_new, /* queue never been serviced */ | |
44f7c160 | 174 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
3b18152c JA |
175 | }; |
176 | ||
177 | #define CFQ_CFQQ_FNS(name) \ | |
178 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
179 | { \ | |
180 | cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ | |
181 | } \ | |
182 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
183 | { \ | |
184 | cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ | |
185 | } \ | |
186 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
187 | { \ | |
188 | return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ | |
189 | } | |
190 | ||
191 | CFQ_CFQQ_FNS(on_rr); | |
192 | CFQ_CFQQ_FNS(wait_request); | |
193 | CFQ_CFQQ_FNS(must_alloc); | |
194 | CFQ_CFQQ_FNS(must_alloc_slice); | |
195 | CFQ_CFQQ_FNS(must_dispatch); | |
196 | CFQ_CFQQ_FNS(fifo_expire); | |
197 | CFQ_CFQQ_FNS(idle_window); | |
198 | CFQ_CFQQ_FNS(prio_changed); | |
53b03744 | 199 | CFQ_CFQQ_FNS(queue_new); |
44f7c160 | 200 | CFQ_CFQQ_FNS(slice_new); |
3b18152c JA |
201 | #undef CFQ_CFQQ_FNS |
202 | ||
3b18152c | 203 | static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short); |
5e705374 | 204 | static void cfq_dispatch_insert(request_queue_t *, struct request *); |
6f325a13 | 205 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, gfp_t gfp_mask); |
1da177e4 | 206 | |
99f95e52 AM |
207 | /* |
208 | * scheduler run of queue, if there are requests pending and no one in the | |
209 | * driver that will restart queueing | |
210 | */ | |
211 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
212 | { | |
7b14e3b5 | 213 | if (cfqd->busy_queues) |
99f95e52 AM |
214 | kblockd_schedule_work(&cfqd->unplug_work); |
215 | } | |
216 | ||
217 | static int cfq_queue_empty(request_queue_t *q) | |
218 | { | |
219 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
220 | ||
b4878f24 | 221 | return !cfqd->busy_queues; |
99f95e52 AM |
222 | } |
223 | ||
7749a8d4 | 224 | static inline pid_t cfq_queue_pid(struct task_struct *task, int rw, int is_sync) |
206dc69b | 225 | { |
7749a8d4 JA |
226 | /* |
227 | * Use the per-process queue, for read requests and syncronous writes | |
228 | */ | |
229 | if (!(rw & REQ_RW) || is_sync) | |
206dc69b JA |
230 | return task->pid; |
231 | ||
232 | return CFQ_KEY_ASYNC; | |
233 | } | |
234 | ||
44f7c160 JA |
235 | /* |
236 | * Scale schedule slice based on io priority. Use the sync time slice only | |
237 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
238 | * io only, should not get full sync slice length. | |
239 | */ | |
240 | static inline int | |
241 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
242 | { | |
243 | const int base_slice = cfqd->cfq_slice[cfq_cfqq_sync(cfqq)]; | |
244 | ||
245 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); | |
246 | ||
247 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - cfqq->ioprio)); | |
248 | } | |
249 | ||
250 | static inline void | |
251 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
252 | { | |
253 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; | |
254 | } | |
255 | ||
256 | /* | |
257 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
258 | * isn't valid until the first request from the dispatch is activated | |
259 | * and the slice time set. | |
260 | */ | |
261 | static inline int cfq_slice_used(struct cfq_queue *cfqq) | |
262 | { | |
263 | if (cfq_cfqq_slice_new(cfqq)) | |
264 | return 0; | |
265 | if (time_before(jiffies, cfqq->slice_end)) | |
266 | return 0; | |
267 | ||
268 | return 1; | |
269 | } | |
270 | ||
1da177e4 | 271 | /* |
5e705374 | 272 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 273 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 274 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 275 | */ |
5e705374 JA |
276 | static struct request * |
277 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
278 | { |
279 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 280 | unsigned long back_max; |
e8a99053 AM |
281 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
282 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
283 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 284 | |
5e705374 JA |
285 | if (rq1 == NULL || rq1 == rq2) |
286 | return rq2; | |
287 | if (rq2 == NULL) | |
288 | return rq1; | |
9c2c38a1 | 289 | |
5e705374 JA |
290 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
291 | return rq1; | |
292 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
293 | return rq2; | |
374f84ac JA |
294 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
295 | return rq1; | |
296 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
297 | return rq2; | |
1da177e4 | 298 | |
5e705374 JA |
299 | s1 = rq1->sector; |
300 | s2 = rq2->sector; | |
1da177e4 LT |
301 | |
302 | last = cfqd->last_sector; | |
303 | ||
1da177e4 LT |
304 | /* |
305 | * by definition, 1KiB is 2 sectors | |
306 | */ | |
307 | back_max = cfqd->cfq_back_max * 2; | |
308 | ||
309 | /* | |
310 | * Strict one way elevator _except_ in the case where we allow | |
311 | * short backward seeks which are biased as twice the cost of a | |
312 | * similar forward seek. | |
313 | */ | |
314 | if (s1 >= last) | |
315 | d1 = s1 - last; | |
316 | else if (s1 + back_max >= last) | |
317 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
318 | else | |
e8a99053 | 319 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
320 | |
321 | if (s2 >= last) | |
322 | d2 = s2 - last; | |
323 | else if (s2 + back_max >= last) | |
324 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
325 | else | |
e8a99053 | 326 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
327 | |
328 | /* Found required data */ | |
e8a99053 AM |
329 | |
330 | /* | |
331 | * By doing switch() on the bit mask "wrap" we avoid having to | |
332 | * check two variables for all permutations: --> faster! | |
333 | */ | |
334 | switch (wrap) { | |
5e705374 | 335 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 336 | if (d1 < d2) |
5e705374 | 337 | return rq1; |
e8a99053 | 338 | else if (d2 < d1) |
5e705374 | 339 | return rq2; |
e8a99053 AM |
340 | else { |
341 | if (s1 >= s2) | |
5e705374 | 342 | return rq1; |
e8a99053 | 343 | else |
5e705374 | 344 | return rq2; |
e8a99053 | 345 | } |
1da177e4 | 346 | |
e8a99053 | 347 | case CFQ_RQ2_WRAP: |
5e705374 | 348 | return rq1; |
e8a99053 | 349 | case CFQ_RQ1_WRAP: |
5e705374 JA |
350 | return rq2; |
351 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
352 | default: |
353 | /* | |
354 | * Since both rqs are wrapped, | |
355 | * start with the one that's further behind head | |
356 | * (--> only *one* back seek required), | |
357 | * since back seek takes more time than forward. | |
358 | */ | |
359 | if (s1 <= s2) | |
5e705374 | 360 | return rq1; |
1da177e4 | 361 | else |
5e705374 | 362 | return rq2; |
1da177e4 LT |
363 | } |
364 | } | |
365 | ||
366 | /* | |
367 | * would be nice to take fifo expire time into account as well | |
368 | */ | |
5e705374 JA |
369 | static struct request * |
370 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
371 | struct request *last) | |
1da177e4 | 372 | { |
21183b07 JA |
373 | struct rb_node *rbnext = rb_next(&last->rb_node); |
374 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 375 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 376 | |
21183b07 | 377 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
378 | |
379 | if (rbprev) | |
5e705374 | 380 | prev = rb_entry_rq(rbprev); |
1da177e4 | 381 | |
21183b07 | 382 | if (rbnext) |
5e705374 | 383 | next = rb_entry_rq(rbnext); |
21183b07 JA |
384 | else { |
385 | rbnext = rb_first(&cfqq->sort_list); | |
386 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 387 | next = rb_entry_rq(rbnext); |
21183b07 | 388 | } |
1da177e4 | 389 | |
21183b07 | 390 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
391 | } |
392 | ||
22e2c507 | 393 | static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted) |
1da177e4 | 394 | { |
22e2c507 | 395 | struct cfq_data *cfqd = cfqq->cfqd; |
99f9628a JA |
396 | struct list_head *list, *n; |
397 | struct cfq_queue *__cfqq; | |
1da177e4 | 398 | |
98e41c7d JA |
399 | /* |
400 | * Resorting requires the cfqq to be on the RR list already. | |
401 | */ | |
402 | if (!cfq_cfqq_on_rr(cfqq)) | |
403 | return; | |
1da177e4 | 404 | |
22e2c507 | 405 | list_del(&cfqq->cfq_list); |
1da177e4 | 406 | |
22e2c507 JA |
407 | if (cfq_class_rt(cfqq)) |
408 | list = &cfqd->cur_rr; | |
409 | else if (cfq_class_idle(cfqq)) | |
410 | list = &cfqd->idle_rr; | |
411 | else { | |
412 | /* | |
413 | * if cfqq has requests in flight, don't allow it to be | |
414 | * found in cfq_set_active_queue before it has finished them. | |
415 | * this is done to increase fairness between a process that | |
416 | * has lots of io pending vs one that only generates one | |
417 | * sporadically or synchronously | |
418 | */ | |
3b18152c | 419 | if (cfq_cfqq_dispatched(cfqq)) |
22e2c507 JA |
420 | list = &cfqd->busy_rr; |
421 | else | |
422 | list = &cfqd->rr_list[cfqq->ioprio]; | |
1da177e4 LT |
423 | } |
424 | ||
53b03744 | 425 | if (preempted || cfq_cfqq_queue_new(cfqq)) { |
99f9628a JA |
426 | /* |
427 | * If this queue was preempted or is new (never been serviced), | |
428 | * let it be added first for fairness but beind other new | |
429 | * queues. | |
430 | */ | |
431 | n = list; | |
53b03744 JA |
432 | while (n->next != list) { |
433 | __cfqq = list_entry_cfqq(n->next); | |
434 | if (!cfq_cfqq_queue_new(__cfqq)) | |
435 | break; | |
1da177e4 | 436 | |
53b03744 JA |
437 | n = n->next; |
438 | } | |
99f9628a JA |
439 | list_add_tail(&cfqq->cfq_list, n); |
440 | } else if (!cfq_cfqq_class_sync(cfqq)) { | |
441 | /* | |
442 | * async queue always goes to the end. this wont be overly | |
443 | * unfair to writes, as the sort of the sync queue wont be | |
444 | * allowed to pass the async queue again. | |
445 | */ | |
446 | list_add_tail(&cfqq->cfq_list, list); | |
447 | } else { | |
448 | /* | |
449 | * sort by last service, but don't cross a new or async | |
450 | * queue. we don't cross a new queue because it hasn't been | |
451 | * service before, and we don't cross an async queue because | |
452 | * it gets added to the end on expire. | |
453 | */ | |
454 | n = list; | |
455 | while ((n = n->prev) != list) { | |
456 | struct cfq_queue *__cfqq = list_entry_cfqq(n); | |
1da177e4 | 457 | |
99f9628a JA |
458 | if (!cfq_cfqq_class_sync(cfqq) || !__cfqq->service_last) |
459 | break; | |
460 | if (time_before(__cfqq->service_last, cfqq->service_last)) | |
461 | break; | |
462 | } | |
463 | list_add(&cfqq->cfq_list, n); | |
1da177e4 | 464 | } |
1da177e4 LT |
465 | } |
466 | ||
467 | /* | |
468 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 469 | * the pending list according to last request service |
1da177e4 LT |
470 | */ |
471 | static inline void | |
b4878f24 | 472 | cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 473 | { |
3b18152c JA |
474 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
475 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
476 | cfqd->busy_queues++; |
477 | ||
b4878f24 | 478 | cfq_resort_rr_list(cfqq, 0); |
1da177e4 LT |
479 | } |
480 | ||
481 | static inline void | |
482 | cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
483 | { | |
3b18152c JA |
484 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
485 | cfq_clear_cfqq_on_rr(cfqq); | |
981a7973 | 486 | list_del_init(&cfqq->cfq_list); |
1da177e4 LT |
487 | |
488 | BUG_ON(!cfqd->busy_queues); | |
489 | cfqd->busy_queues--; | |
490 | } | |
491 | ||
492 | /* | |
493 | * rb tree support functions | |
494 | */ | |
5e705374 | 495 | static inline void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 496 | { |
5e705374 | 497 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 498 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 499 | const int sync = rq_is_sync(rq); |
1da177e4 | 500 | |
b4878f24 JA |
501 | BUG_ON(!cfqq->queued[sync]); |
502 | cfqq->queued[sync]--; | |
1da177e4 | 503 | |
5e705374 | 504 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 505 | |
dd67d051 | 506 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 507 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
508 | } |
509 | ||
5e705374 | 510 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 511 | { |
5e705374 | 512 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 513 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 514 | struct request *__alias; |
1da177e4 | 515 | |
5380a101 | 516 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
517 | |
518 | /* | |
519 | * looks a little odd, but the first insert might return an alias. | |
520 | * if that happens, put the alias on the dispatch list | |
521 | */ | |
21183b07 | 522 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 523 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
524 | |
525 | if (!cfq_cfqq_on_rr(cfqq)) | |
526 | cfq_add_cfqq_rr(cfqd, cfqq); | |
1da177e4 LT |
527 | } |
528 | ||
529 | static inline void | |
5e705374 | 530 | cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 531 | { |
5380a101 JA |
532 | elv_rb_del(&cfqq->sort_list, rq); |
533 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 534 | cfq_add_rq_rb(rq); |
1da177e4 LT |
535 | } |
536 | ||
206dc69b JA |
537 | static struct request * |
538 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 539 | { |
206dc69b | 540 | struct task_struct *tsk = current; |
7749a8d4 | 541 | pid_t key = cfq_queue_pid(tsk, bio_data_dir(bio), bio_sync(bio)); |
206dc69b | 542 | struct cfq_queue *cfqq; |
1da177e4 | 543 | |
206dc69b | 544 | cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio); |
89850f7e JA |
545 | if (cfqq) { |
546 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
547 | ||
21183b07 | 548 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 549 | } |
1da177e4 | 550 | |
1da177e4 LT |
551 | return NULL; |
552 | } | |
553 | ||
b4878f24 | 554 | static void cfq_activate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 555 | { |
22e2c507 | 556 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 557 | |
b4878f24 | 558 | cfqd->rq_in_driver++; |
25776e35 JA |
559 | |
560 | /* | |
561 | * If the depth is larger 1, it really could be queueing. But lets | |
562 | * make the mark a little higher - idling could still be good for | |
563 | * low queueing, and a low queueing number could also just indicate | |
564 | * a SCSI mid layer like behaviour where limit+1 is often seen. | |
565 | */ | |
566 | if (!cfqd->hw_tag && cfqd->rq_in_driver > 4) | |
567 | cfqd->hw_tag = 1; | |
1da177e4 LT |
568 | } |
569 | ||
b4878f24 | 570 | static void cfq_deactivate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 571 | { |
b4878f24 JA |
572 | struct cfq_data *cfqd = q->elevator->elevator_data; |
573 | ||
574 | WARN_ON(!cfqd->rq_in_driver); | |
575 | cfqd->rq_in_driver--; | |
1da177e4 LT |
576 | } |
577 | ||
b4878f24 | 578 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 579 | { |
5e705374 | 580 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 581 | |
5e705374 JA |
582 | if (cfqq->next_rq == rq) |
583 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 584 | |
b4878f24 | 585 | list_del_init(&rq->queuelist); |
5e705374 | 586 | cfq_del_rq_rb(rq); |
374f84ac JA |
587 | |
588 | if (rq_is_meta(rq)) { | |
589 | WARN_ON(!cfqq->meta_pending); | |
590 | cfqq->meta_pending--; | |
591 | } | |
1da177e4 LT |
592 | } |
593 | ||
594 | static int | |
595 | cfq_merge(request_queue_t *q, struct request **req, struct bio *bio) | |
596 | { | |
597 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
598 | struct request *__rq; | |
1da177e4 | 599 | |
206dc69b | 600 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 601 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
602 | *req = __rq; |
603 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
604 | } |
605 | ||
606 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
607 | } |
608 | ||
21183b07 JA |
609 | static void cfq_merged_request(request_queue_t *q, struct request *req, |
610 | int type) | |
1da177e4 | 611 | { |
21183b07 | 612 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 613 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 614 | |
5e705374 | 615 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 616 | } |
1da177e4 LT |
617 | } |
618 | ||
619 | static void | |
620 | cfq_merged_requests(request_queue_t *q, struct request *rq, | |
621 | struct request *next) | |
622 | { | |
22e2c507 JA |
623 | /* |
624 | * reposition in fifo if next is older than rq | |
625 | */ | |
626 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
627 | time_before(next->start_time, rq->start_time)) | |
628 | list_move(&rq->queuelist, &next->queuelist); | |
629 | ||
b4878f24 | 630 | cfq_remove_request(next); |
22e2c507 JA |
631 | } |
632 | ||
da775265 JA |
633 | static int cfq_allow_merge(request_queue_t *q, struct request *rq, |
634 | struct bio *bio) | |
635 | { | |
636 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
637 | const int rw = bio_data_dir(bio); | |
638 | struct cfq_queue *cfqq; | |
639 | pid_t key; | |
640 | ||
641 | /* | |
ec8acb69 | 642 | * Disallow merge of a sync bio into an async request. |
da775265 | 643 | */ |
ec8acb69 | 644 | if ((bio_data_dir(bio) == READ || bio_sync(bio)) && !rq_is_sync(rq)) |
da775265 JA |
645 | return 0; |
646 | ||
647 | /* | |
719d3402 JA |
648 | * Lookup the cfqq that this bio will be queued with. Allow |
649 | * merge only if rq is queued there. | |
da775265 | 650 | */ |
719d3402 | 651 | key = cfq_queue_pid(current, rw, bio_sync(bio)); |
da775265 | 652 | cfqq = cfq_find_cfq_hash(cfqd, key, current->ioprio); |
719d3402 JA |
653 | |
654 | if (cfqq == RQ_CFQQ(rq)) | |
655 | return 1; | |
da775265 | 656 | |
ec8acb69 | 657 | return 0; |
da775265 JA |
658 | } |
659 | ||
22e2c507 JA |
660 | static inline void |
661 | __cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
662 | { | |
663 | if (cfqq) { | |
664 | /* | |
665 | * stop potential idle class queues waiting service | |
666 | */ | |
667 | del_timer(&cfqd->idle_class_timer); | |
668 | ||
22e2c507 JA |
669 | cfqq->slice_end = 0; |
670 | cfqq->slice_left = 0; | |
3b18152c JA |
671 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
672 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 673 | cfq_mark_cfqq_slice_new(cfqq); |
22e2c507 JA |
674 | } |
675 | ||
676 | cfqd->active_queue = cfqq; | |
677 | } | |
678 | ||
7b14e3b5 JA |
679 | /* |
680 | * current cfqq expired its slice (or was too idle), select new one | |
681 | */ | |
682 | static void | |
683 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
684 | int preempted) | |
685 | { | |
686 | unsigned long now = jiffies; | |
687 | ||
688 | if (cfq_cfqq_wait_request(cfqq)) | |
689 | del_timer(&cfqd->idle_slice_timer); | |
690 | ||
53b03744 | 691 | if (!preempted && !cfq_cfqq_dispatched(cfqq)) |
7b14e3b5 | 692 | cfq_schedule_dispatch(cfqd); |
7b14e3b5 JA |
693 | |
694 | cfq_clear_cfqq_must_dispatch(cfqq); | |
695 | cfq_clear_cfqq_wait_request(cfqq); | |
53b03744 | 696 | cfq_clear_cfqq_queue_new(cfqq); |
7b14e3b5 JA |
697 | |
698 | /* | |
699 | * store what was left of this slice, if the queue idled out | |
700 | * or was preempted | |
701 | */ | |
44f7c160 | 702 | if (cfq_slice_used(cfqq)) |
7b14e3b5 JA |
703 | cfqq->slice_left = cfqq->slice_end - now; |
704 | else | |
705 | cfqq->slice_left = 0; | |
706 | ||
98e41c7d | 707 | cfq_resort_rr_list(cfqq, preempted); |
7b14e3b5 JA |
708 | |
709 | if (cfqq == cfqd->active_queue) | |
710 | cfqd->active_queue = NULL; | |
711 | ||
712 | if (cfqd->active_cic) { | |
713 | put_io_context(cfqd->active_cic->ioc); | |
714 | cfqd->active_cic = NULL; | |
715 | } | |
716 | ||
717 | cfqd->dispatch_slice = 0; | |
718 | } | |
719 | ||
720 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted) | |
721 | { | |
722 | struct cfq_queue *cfqq = cfqd->active_queue; | |
723 | ||
724 | if (cfqq) | |
725 | __cfq_slice_expired(cfqd, cfqq, preempted); | |
726 | } | |
727 | ||
22e2c507 JA |
728 | /* |
729 | * 0 | |
730 | * 0,1 | |
731 | * 0,1,2 | |
732 | * 0,1,2,3 | |
733 | * 0,1,2,3,4 | |
734 | * 0,1,2,3,4,5 | |
735 | * 0,1,2,3,4,5,6 | |
736 | * 0,1,2,3,4,5,6,7 | |
737 | */ | |
738 | static int cfq_get_next_prio_level(struct cfq_data *cfqd) | |
739 | { | |
740 | int prio, wrap; | |
741 | ||
742 | prio = -1; | |
743 | wrap = 0; | |
744 | do { | |
745 | int p; | |
746 | ||
747 | for (p = cfqd->cur_prio; p <= cfqd->cur_end_prio; p++) { | |
748 | if (!list_empty(&cfqd->rr_list[p])) { | |
749 | prio = p; | |
750 | break; | |
751 | } | |
752 | } | |
753 | ||
754 | if (prio != -1) | |
755 | break; | |
756 | cfqd->cur_prio = 0; | |
757 | if (++cfqd->cur_end_prio == CFQ_PRIO_LISTS) { | |
758 | cfqd->cur_end_prio = 0; | |
759 | if (wrap) | |
760 | break; | |
761 | wrap = 1; | |
1da177e4 | 762 | } |
22e2c507 JA |
763 | } while (1); |
764 | ||
765 | if (unlikely(prio == -1)) | |
766 | return -1; | |
767 | ||
768 | BUG_ON(prio >= CFQ_PRIO_LISTS); | |
769 | ||
770 | list_splice_init(&cfqd->rr_list[prio], &cfqd->cur_rr); | |
771 | ||
772 | cfqd->cur_prio = prio + 1; | |
773 | if (cfqd->cur_prio > cfqd->cur_end_prio) { | |
774 | cfqd->cur_end_prio = cfqd->cur_prio; | |
775 | cfqd->cur_prio = 0; | |
776 | } | |
777 | if (cfqd->cur_end_prio == CFQ_PRIO_LISTS) { | |
778 | cfqd->cur_prio = 0; | |
779 | cfqd->cur_end_prio = 0; | |
1da177e4 LT |
780 | } |
781 | ||
22e2c507 JA |
782 | return prio; |
783 | } | |
784 | ||
3b18152c | 785 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
22e2c507 | 786 | { |
7b14e3b5 | 787 | struct cfq_queue *cfqq = NULL; |
22e2c507 | 788 | |
89850f7e JA |
789 | if (!list_empty(&cfqd->cur_rr) || cfq_get_next_prio_level(cfqd) != -1) { |
790 | /* | |
791 | * if current list is non-empty, grab first entry. if it is | |
792 | * empty, get next prio level and grab first entry then if any | |
793 | * are spliced | |
794 | */ | |
22e2c507 | 795 | cfqq = list_entry_cfqq(cfqd->cur_rr.next); |
89850f7e JA |
796 | } else if (!list_empty(&cfqd->busy_rr)) { |
797 | /* | |
798 | * If no new queues are available, check if the busy list has | |
799 | * some before falling back to idle io. | |
800 | */ | |
e0de0206 | 801 | cfqq = list_entry_cfqq(cfqd->busy_rr.next); |
89850f7e JA |
802 | } else if (!list_empty(&cfqd->idle_rr)) { |
803 | /* | |
804 | * if we have idle queues and no rt or be queues had pending | |
805 | * requests, either allow immediate service if the grace period | |
806 | * has passed or arm the idle grace timer | |
807 | */ | |
22e2c507 JA |
808 | unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE; |
809 | ||
810 | if (time_after_eq(jiffies, end)) | |
811 | cfqq = list_entry_cfqq(cfqd->idle_rr.next); | |
812 | else | |
813 | mod_timer(&cfqd->idle_class_timer, end); | |
814 | } | |
815 | ||
816 | __cfq_set_active_queue(cfqd, cfqq); | |
3b18152c | 817 | return cfqq; |
22e2c507 JA |
818 | } |
819 | ||
caaa5f9f JA |
820 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (128 * 1024)) |
821 | ||
22e2c507 JA |
822 | static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
823 | ||
824 | { | |
206dc69b | 825 | struct cfq_io_context *cic; |
7b14e3b5 JA |
826 | unsigned long sl; |
827 | ||
dd67d051 | 828 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
829 | WARN_ON(cfqq != cfqd->active_queue); |
830 | ||
831 | /* | |
832 | * idle is disabled, either manually or by past process history | |
833 | */ | |
834 | if (!cfqd->cfq_slice_idle) | |
835 | return 0; | |
3b18152c | 836 | if (!cfq_cfqq_idle_window(cfqq)) |
22e2c507 JA |
837 | return 0; |
838 | /* | |
839 | * task has exited, don't wait | |
840 | */ | |
206dc69b JA |
841 | cic = cfqd->active_cic; |
842 | if (!cic || !cic->ioc->task) | |
22e2c507 JA |
843 | return 0; |
844 | ||
3b18152c JA |
845 | cfq_mark_cfqq_must_dispatch(cfqq); |
846 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 847 | |
7b14e3b5 | 848 | sl = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle); |
206dc69b JA |
849 | |
850 | /* | |
851 | * we don't want to idle for seeks, but we do want to allow | |
852 | * fair distribution of slice time for a process doing back-to-back | |
853 | * seeks. so allow a little bit of time for him to submit a new rq | |
854 | */ | |
caaa5f9f | 855 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
44eb1231 | 856 | sl = min(sl, msecs_to_jiffies(2)); |
206dc69b | 857 | |
7b14e3b5 | 858 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
22e2c507 | 859 | return 1; |
1da177e4 LT |
860 | } |
861 | ||
5e705374 | 862 | static void cfq_dispatch_insert(request_queue_t *q, struct request *rq) |
1da177e4 LT |
863 | { |
864 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
5e705374 | 865 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 866 | |
5380a101 JA |
867 | cfq_remove_request(rq); |
868 | cfqq->on_dispatch[rq_is_sync(rq)]++; | |
869 | elv_dispatch_sort(q, rq); | |
fd61af03 JA |
870 | |
871 | rq = list_entry(q->queue_head.prev, struct request, queuelist); | |
872 | cfqd->last_sector = rq->sector + rq->nr_sectors; | |
1da177e4 LT |
873 | } |
874 | ||
875 | /* | |
876 | * return expired entry, or NULL to just start from scratch in rbtree | |
877 | */ | |
5e705374 | 878 | static inline struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
879 | { |
880 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 881 | struct request *rq; |
89850f7e | 882 | int fifo; |
1da177e4 | 883 | |
3b18152c | 884 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 885 | return NULL; |
89850f7e JA |
886 | if (list_empty(&cfqq->fifo)) |
887 | return NULL; | |
1da177e4 | 888 | |
89850f7e JA |
889 | fifo = cfq_cfqq_class_sync(cfqq); |
890 | rq = rq_entry_fifo(cfqq->fifo.next); | |
1da177e4 | 891 | |
89850f7e JA |
892 | if (time_after(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) { |
893 | cfq_mark_cfqq_fifo_expire(cfqq); | |
894 | return rq; | |
1da177e4 LT |
895 | } |
896 | ||
897 | return NULL; | |
898 | } | |
899 | ||
22e2c507 JA |
900 | static inline int |
901 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
902 | { | |
903 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 904 | |
22e2c507 | 905 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 906 | |
22e2c507 | 907 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
908 | } |
909 | ||
22e2c507 JA |
910 | /* |
911 | * get next queue for service | |
912 | */ | |
1b5ed5e1 | 913 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 914 | { |
1da177e4 | 915 | struct cfq_queue *cfqq; |
1da177e4 | 916 | |
22e2c507 JA |
917 | cfqq = cfqd->active_queue; |
918 | if (!cfqq) | |
919 | goto new_queue; | |
1da177e4 | 920 | |
22e2c507 JA |
921 | /* |
922 | * slice has expired | |
923 | */ | |
44f7c160 | 924 | if (!cfq_cfqq_must_dispatch(cfqq) && cfq_slice_used(cfqq)) |
3b18152c | 925 | goto expire; |
1da177e4 | 926 | |
22e2c507 JA |
927 | /* |
928 | * if queue has requests, dispatch one. if not, check if | |
929 | * enough slice is left to wait for one | |
930 | */ | |
dd67d051 | 931 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 932 | goto keep_queue; |
44f7c160 | 933 | else if (cfq_cfqq_slice_new(cfqq) || cfq_cfqq_dispatched(cfqq)) { |
caaa5f9f JA |
934 | cfqq = NULL; |
935 | goto keep_queue; | |
936 | } else if (cfq_cfqq_class_sync(cfqq)) { | |
22e2c507 JA |
937 | if (cfq_arm_slice_timer(cfqd, cfqq)) |
938 | return NULL; | |
939 | } | |
940 | ||
3b18152c | 941 | expire: |
22e2c507 | 942 | cfq_slice_expired(cfqd, 0); |
3b18152c JA |
943 | new_queue: |
944 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 945 | keep_queue: |
3b18152c | 946 | return cfqq; |
22e2c507 JA |
947 | } |
948 | ||
949 | static int | |
950 | __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
951 | int max_dispatch) | |
952 | { | |
953 | int dispatched = 0; | |
954 | ||
dd67d051 | 955 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
956 | |
957 | do { | |
5e705374 | 958 | struct request *rq; |
1da177e4 LT |
959 | |
960 | /* | |
22e2c507 | 961 | * follow expired path, else get first next available |
1da177e4 | 962 | */ |
5e705374 JA |
963 | if ((rq = cfq_check_fifo(cfqq)) == NULL) |
964 | rq = cfqq->next_rq; | |
22e2c507 JA |
965 | |
966 | /* | |
967 | * finally, insert request into driver dispatch list | |
968 | */ | |
5e705374 | 969 | cfq_dispatch_insert(cfqd->queue, rq); |
1da177e4 | 970 | |
22e2c507 JA |
971 | cfqd->dispatch_slice++; |
972 | dispatched++; | |
1da177e4 | 973 | |
22e2c507 | 974 | if (!cfqd->active_cic) { |
5e705374 JA |
975 | atomic_inc(&RQ_CIC(rq)->ioc->refcount); |
976 | cfqd->active_cic = RQ_CIC(rq); | |
22e2c507 | 977 | } |
1da177e4 | 978 | |
dd67d051 | 979 | if (RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 JA |
980 | break; |
981 | ||
982 | } while (dispatched < max_dispatch); | |
983 | ||
22e2c507 JA |
984 | /* |
985 | * expire an async queue immediately if it has used up its slice. idle | |
986 | * queue always expire after 1 dispatch round. | |
987 | */ | |
988 | if ((!cfq_cfqq_sync(cfqq) && | |
989 | cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
44f7c160 JA |
990 | cfq_class_idle(cfqq)) { |
991 | cfqq->slice_end = jiffies + 1; | |
22e2c507 | 992 | cfq_slice_expired(cfqd, 0); |
44f7c160 | 993 | } |
22e2c507 JA |
994 | |
995 | return dispatched; | |
996 | } | |
997 | ||
1b5ed5e1 TH |
998 | static int |
999 | cfq_forced_dispatch_cfqqs(struct list_head *list) | |
1000 | { | |
1b5ed5e1 | 1001 | struct cfq_queue *cfqq, *next; |
caaa5f9f | 1002 | int dispatched; |
1b5ed5e1 | 1003 | |
caaa5f9f | 1004 | dispatched = 0; |
1b5ed5e1 | 1005 | list_for_each_entry_safe(cfqq, next, list, cfq_list) { |
5e705374 JA |
1006 | while (cfqq->next_rq) { |
1007 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1b5ed5e1 TH |
1008 | dispatched++; |
1009 | } | |
1010 | BUG_ON(!list_empty(&cfqq->fifo)); | |
1011 | } | |
caaa5f9f | 1012 | |
1b5ed5e1 TH |
1013 | return dispatched; |
1014 | } | |
1015 | ||
1016 | static int | |
1017 | cfq_forced_dispatch(struct cfq_data *cfqd) | |
1018 | { | |
1019 | int i, dispatched = 0; | |
1020 | ||
1021 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
1022 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->rr_list[i]); | |
1023 | ||
1024 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->busy_rr); | |
1025 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->cur_rr); | |
1026 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->idle_rr); | |
1027 | ||
1028 | cfq_slice_expired(cfqd, 0); | |
1029 | ||
1030 | BUG_ON(cfqd->busy_queues); | |
1031 | ||
1032 | return dispatched; | |
1033 | } | |
1034 | ||
22e2c507 | 1035 | static int |
b4878f24 | 1036 | cfq_dispatch_requests(request_queue_t *q, int force) |
22e2c507 JA |
1037 | { |
1038 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
caaa5f9f JA |
1039 | struct cfq_queue *cfqq, *prev_cfqq; |
1040 | int dispatched; | |
22e2c507 JA |
1041 | |
1042 | if (!cfqd->busy_queues) | |
1043 | return 0; | |
1044 | ||
1b5ed5e1 TH |
1045 | if (unlikely(force)) |
1046 | return cfq_forced_dispatch(cfqd); | |
1047 | ||
caaa5f9f JA |
1048 | dispatched = 0; |
1049 | prev_cfqq = NULL; | |
1050 | while ((cfqq = cfq_select_queue(cfqd)) != NULL) { | |
b4878f24 JA |
1051 | int max_dispatch; |
1052 | ||
caaa5f9f JA |
1053 | /* |
1054 | * Don't repeat dispatch from the previous queue. | |
1055 | */ | |
1056 | if (prev_cfqq == cfqq) | |
1057 | break; | |
1058 | ||
3b18152c JA |
1059 | cfq_clear_cfqq_must_dispatch(cfqq); |
1060 | cfq_clear_cfqq_wait_request(cfqq); | |
22e2c507 JA |
1061 | del_timer(&cfqd->idle_slice_timer); |
1062 | ||
1b5ed5e1 TH |
1063 | max_dispatch = cfqd->cfq_quantum; |
1064 | if (cfq_class_idle(cfqq)) | |
1065 | max_dispatch = 1; | |
1da177e4 | 1066 | |
caaa5f9f JA |
1067 | dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); |
1068 | ||
1069 | /* | |
1070 | * If the dispatch cfqq has idling enabled and is still | |
1071 | * the active queue, break out. | |
1072 | */ | |
1073 | if (cfq_cfqq_idle_window(cfqq) && cfqd->active_queue) | |
1074 | break; | |
1075 | ||
1076 | prev_cfqq = cfqq; | |
1da177e4 LT |
1077 | } |
1078 | ||
caaa5f9f | 1079 | return dispatched; |
1da177e4 LT |
1080 | } |
1081 | ||
1da177e4 | 1082 | /* |
5e705374 JA |
1083 | * task holds one reference to the queue, dropped when task exits. each rq |
1084 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1085 | * |
1086 | * queue lock must be held here. | |
1087 | */ | |
1088 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1089 | { | |
22e2c507 JA |
1090 | struct cfq_data *cfqd = cfqq->cfqd; |
1091 | ||
1092 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1093 | |
1094 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1095 | return; | |
1096 | ||
1097 | BUG_ON(rb_first(&cfqq->sort_list)); | |
22e2c507 | 1098 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1099 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1100 | |
7b14e3b5 | 1101 | if (unlikely(cfqd->active_queue == cfqq)) |
3b18152c | 1102 | __cfq_slice_expired(cfqd, cfqq, 0); |
22e2c507 | 1103 | |
1da177e4 LT |
1104 | /* |
1105 | * it's on the empty list and still hashed | |
1106 | */ | |
1107 | list_del(&cfqq->cfq_list); | |
1108 | hlist_del(&cfqq->cfq_hash); | |
1109 | kmem_cache_free(cfq_pool, cfqq); | |
1110 | } | |
1111 | ||
1ea25ecb | 1112 | static struct cfq_queue * |
3b18152c JA |
1113 | __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio, |
1114 | const int hashval) | |
1da177e4 LT |
1115 | { |
1116 | struct hlist_head *hash_list = &cfqd->cfq_hash[hashval]; | |
206dc69b JA |
1117 | struct hlist_node *entry; |
1118 | struct cfq_queue *__cfqq; | |
1da177e4 | 1119 | |
206dc69b | 1120 | hlist_for_each_entry(__cfqq, entry, hash_list, cfq_hash) { |
b0a6916b | 1121 | const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio); |
1da177e4 | 1122 | |
206dc69b | 1123 | if (__cfqq->key == key && (__p == prio || !prio)) |
1da177e4 LT |
1124 | return __cfqq; |
1125 | } | |
1126 | ||
1127 | return NULL; | |
1128 | } | |
1129 | ||
1130 | static struct cfq_queue * | |
3b18152c | 1131 | cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio) |
1da177e4 | 1132 | { |
3b18152c | 1133 | return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT)); |
1da177e4 LT |
1134 | } |
1135 | ||
e2d74ac0 | 1136 | static void cfq_free_io_context(struct io_context *ioc) |
1da177e4 | 1137 | { |
22e2c507 | 1138 | struct cfq_io_context *__cic; |
e2d74ac0 JA |
1139 | struct rb_node *n; |
1140 | int freed = 0; | |
1da177e4 | 1141 | |
e2d74ac0 JA |
1142 | while ((n = rb_first(&ioc->cic_root)) != NULL) { |
1143 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1144 | rb_erase(&__cic->rb_node, &ioc->cic_root); | |
22e2c507 | 1145 | kmem_cache_free(cfq_ioc_pool, __cic); |
334e94de | 1146 | freed++; |
1da177e4 LT |
1147 | } |
1148 | ||
4050cf16 JA |
1149 | elv_ioc_count_mod(ioc_count, -freed); |
1150 | ||
1151 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) | |
334e94de | 1152 | complete(ioc_gone); |
1da177e4 LT |
1153 | } |
1154 | ||
89850f7e | 1155 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1156 | { |
89850f7e JA |
1157 | if (unlikely(cfqq == cfqd->active_queue)) |
1158 | __cfq_slice_expired(cfqd, cfqq, 0); | |
22e2c507 | 1159 | |
89850f7e JA |
1160 | cfq_put_queue(cfqq); |
1161 | } | |
22e2c507 | 1162 | |
89850f7e JA |
1163 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1164 | struct cfq_io_context *cic) | |
1165 | { | |
fc46379d JA |
1166 | list_del_init(&cic->queue_list); |
1167 | smp_wmb(); | |
1168 | cic->key = NULL; | |
1169 | ||
12a05732 | 1170 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1171 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1172 | cic->cfqq[ASYNC] = NULL; |
1173 | } | |
1174 | ||
1175 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1176 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1177 | cic->cfqq[SYNC] = NULL; |
1178 | } | |
89850f7e JA |
1179 | } |
1180 | ||
1181 | ||
1182 | /* | |
1183 | * Called with interrupts disabled | |
1184 | */ | |
1185 | static void cfq_exit_single_io_context(struct cfq_io_context *cic) | |
1186 | { | |
1187 | struct cfq_data *cfqd = cic->key; | |
1188 | ||
89850f7e JA |
1189 | if (cfqd) { |
1190 | request_queue_t *q = cfqd->queue; | |
1191 | ||
fc46379d | 1192 | spin_lock_irq(q->queue_lock); |
89850f7e | 1193 | __cfq_exit_single_io_context(cfqd, cic); |
fc46379d | 1194 | spin_unlock_irq(q->queue_lock); |
89850f7e | 1195 | } |
1da177e4 LT |
1196 | } |
1197 | ||
e2d74ac0 | 1198 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1199 | { |
22e2c507 | 1200 | struct cfq_io_context *__cic; |
e2d74ac0 | 1201 | struct rb_node *n; |
22e2c507 | 1202 | |
1da177e4 LT |
1203 | /* |
1204 | * put the reference this task is holding to the various queues | |
1205 | */ | |
e2d74ac0 JA |
1206 | |
1207 | n = rb_first(&ioc->cic_root); | |
1208 | while (n != NULL) { | |
1209 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1210 | ||
22e2c507 | 1211 | cfq_exit_single_io_context(__cic); |
e2d74ac0 | 1212 | n = rb_next(n); |
1da177e4 | 1213 | } |
1da177e4 LT |
1214 | } |
1215 | ||
22e2c507 | 1216 | static struct cfq_io_context * |
8267e268 | 1217 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1218 | { |
b5deef90 | 1219 | struct cfq_io_context *cic; |
1da177e4 | 1220 | |
b5deef90 | 1221 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask, cfqd->queue->node); |
1da177e4 | 1222 | if (cic) { |
553698f9 | 1223 | memset(cic, 0, sizeof(*cic)); |
22e2c507 | 1224 | cic->last_end_request = jiffies; |
553698f9 | 1225 | INIT_LIST_HEAD(&cic->queue_list); |
22e2c507 JA |
1226 | cic->dtor = cfq_free_io_context; |
1227 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1228 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1229 | } |
1230 | ||
1231 | return cic; | |
1232 | } | |
1233 | ||
22e2c507 JA |
1234 | static void cfq_init_prio_data(struct cfq_queue *cfqq) |
1235 | { | |
1236 | struct task_struct *tsk = current; | |
1237 | int ioprio_class; | |
1238 | ||
3b18152c | 1239 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1240 | return; |
1241 | ||
1242 | ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio); | |
1243 | switch (ioprio_class) { | |
1244 | default: | |
1245 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1246 | case IOPRIO_CLASS_NONE: | |
1247 | /* | |
1248 | * no prio set, place us in the middle of the BE classes | |
1249 | */ | |
1250 | cfqq->ioprio = task_nice_ioprio(tsk); | |
1251 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1252 | break; | |
1253 | case IOPRIO_CLASS_RT: | |
1254 | cfqq->ioprio = task_ioprio(tsk); | |
1255 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1256 | break; | |
1257 | case IOPRIO_CLASS_BE: | |
1258 | cfqq->ioprio = task_ioprio(tsk); | |
1259 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1260 | break; | |
1261 | case IOPRIO_CLASS_IDLE: | |
1262 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1263 | cfqq->ioprio = 7; | |
3b18152c | 1264 | cfq_clear_cfqq_idle_window(cfqq); |
22e2c507 JA |
1265 | break; |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * keep track of original prio settings in case we have to temporarily | |
1270 | * elevate the priority of this queue | |
1271 | */ | |
1272 | cfqq->org_ioprio = cfqq->ioprio; | |
1273 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
1274 | ||
98e41c7d | 1275 | cfq_resort_rr_list(cfqq, 0); |
3b18152c | 1276 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1277 | } |
1278 | ||
478a82b0 | 1279 | static inline void changed_ioprio(struct cfq_io_context *cic) |
22e2c507 | 1280 | { |
478a82b0 AV |
1281 | struct cfq_data *cfqd = cic->key; |
1282 | struct cfq_queue *cfqq; | |
c1b707d2 | 1283 | unsigned long flags; |
35e6077c | 1284 | |
caaa5f9f JA |
1285 | if (unlikely(!cfqd)) |
1286 | return; | |
1287 | ||
c1b707d2 | 1288 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1289 | |
1290 | cfqq = cic->cfqq[ASYNC]; | |
1291 | if (cfqq) { | |
1292 | struct cfq_queue *new_cfqq; | |
1293 | new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC, cic->ioc->task, | |
1294 | GFP_ATOMIC); | |
1295 | if (new_cfqq) { | |
1296 | cic->cfqq[ASYNC] = new_cfqq; | |
1297 | cfq_put_queue(cfqq); | |
1298 | } | |
22e2c507 | 1299 | } |
caaa5f9f JA |
1300 | |
1301 | cfqq = cic->cfqq[SYNC]; | |
1302 | if (cfqq) | |
1303 | cfq_mark_cfqq_prio_changed(cfqq); | |
1304 | ||
c1b707d2 | 1305 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1306 | } |
1307 | ||
fc46379d | 1308 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1309 | { |
a6a0763a | 1310 | struct cfq_io_context *cic; |
e2d74ac0 | 1311 | struct rb_node *n; |
a6a0763a | 1312 | |
fc46379d | 1313 | ioc->ioprio_changed = 0; |
a6a0763a | 1314 | |
e2d74ac0 JA |
1315 | n = rb_first(&ioc->cic_root); |
1316 | while (n != NULL) { | |
1317 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
3793c65c | 1318 | |
478a82b0 | 1319 | changed_ioprio(cic); |
e2d74ac0 JA |
1320 | n = rb_next(n); |
1321 | } | |
22e2c507 JA |
1322 | } |
1323 | ||
1324 | static struct cfq_queue * | |
6f325a13 | 1325 | cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, |
8267e268 | 1326 | gfp_t gfp_mask) |
22e2c507 JA |
1327 | { |
1328 | const int hashval = hash_long(key, CFQ_QHASH_SHIFT); | |
1329 | struct cfq_queue *cfqq, *new_cfqq = NULL; | |
6f325a13 | 1330 | unsigned short ioprio; |
22e2c507 JA |
1331 | |
1332 | retry: | |
6f325a13 | 1333 | ioprio = tsk->ioprio; |
3b18152c | 1334 | cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval); |
22e2c507 JA |
1335 | |
1336 | if (!cfqq) { | |
1337 | if (new_cfqq) { | |
1338 | cfqq = new_cfqq; | |
1339 | new_cfqq = NULL; | |
1340 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1341 | /* |
1342 | * Inform the allocator of the fact that we will | |
1343 | * just repeat this allocation if it fails, to allow | |
1344 | * the allocator to do whatever it needs to attempt to | |
1345 | * free memory. | |
1346 | */ | |
22e2c507 | 1347 | spin_unlock_irq(cfqd->queue->queue_lock); |
b5deef90 | 1348 | new_cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask|__GFP_NOFAIL, cfqd->queue->node); |
22e2c507 JA |
1349 | spin_lock_irq(cfqd->queue->queue_lock); |
1350 | goto retry; | |
1351 | } else { | |
b5deef90 | 1352 | cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask, cfqd->queue->node); |
22e2c507 JA |
1353 | if (!cfqq) |
1354 | goto out; | |
1355 | } | |
1356 | ||
1357 | memset(cfqq, 0, sizeof(*cfqq)); | |
1358 | ||
1359 | INIT_HLIST_NODE(&cfqq->cfq_hash); | |
1360 | INIT_LIST_HEAD(&cfqq->cfq_list); | |
22e2c507 JA |
1361 | INIT_LIST_HEAD(&cfqq->fifo); |
1362 | ||
1363 | cfqq->key = key; | |
1364 | hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]); | |
1365 | atomic_set(&cfqq->ref, 0); | |
1366 | cfqq->cfqd = cfqd; | |
22e2c507 JA |
1367 | /* |
1368 | * set ->slice_left to allow preemption for a new process | |
1369 | */ | |
1370 | cfqq->slice_left = 2 * cfqd->cfq_slice_idle; | |
caaa5f9f | 1371 | cfq_mark_cfqq_idle_window(cfqq); |
3b18152c | 1372 | cfq_mark_cfqq_prio_changed(cfqq); |
53b03744 | 1373 | cfq_mark_cfqq_queue_new(cfqq); |
3b18152c | 1374 | cfq_init_prio_data(cfqq); |
22e2c507 JA |
1375 | } |
1376 | ||
1377 | if (new_cfqq) | |
1378 | kmem_cache_free(cfq_pool, new_cfqq); | |
1379 | ||
1380 | atomic_inc(&cfqq->ref); | |
1381 | out: | |
1382 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1383 | return cfqq; | |
1384 | } | |
1385 | ||
dbecf3ab OH |
1386 | static void |
1387 | cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic) | |
1388 | { | |
fc46379d | 1389 | WARN_ON(!list_empty(&cic->queue_list)); |
dbecf3ab | 1390 | rb_erase(&cic->rb_node, &ioc->cic_root); |
dbecf3ab | 1391 | kmem_cache_free(cfq_ioc_pool, cic); |
4050cf16 | 1392 | elv_ioc_count_dec(ioc_count); |
dbecf3ab OH |
1393 | } |
1394 | ||
e2d74ac0 JA |
1395 | static struct cfq_io_context * |
1396 | cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc) | |
1397 | { | |
dbecf3ab | 1398 | struct rb_node *n; |
e2d74ac0 | 1399 | struct cfq_io_context *cic; |
be3b0753 | 1400 | void *k, *key = cfqd; |
e2d74ac0 | 1401 | |
dbecf3ab OH |
1402 | restart: |
1403 | n = ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1404 | while (n) { |
1405 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1406 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1407 | k = cic->key; | |
1408 | if (unlikely(!k)) { | |
dbecf3ab OH |
1409 | cfq_drop_dead_cic(ioc, cic); |
1410 | goto restart; | |
1411 | } | |
e2d74ac0 | 1412 | |
be3b0753 | 1413 | if (key < k) |
e2d74ac0 | 1414 | n = n->rb_left; |
be3b0753 | 1415 | else if (key > k) |
e2d74ac0 JA |
1416 | n = n->rb_right; |
1417 | else | |
1418 | return cic; | |
1419 | } | |
1420 | ||
1421 | return NULL; | |
1422 | } | |
1423 | ||
1424 | static inline void | |
1425 | cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, | |
1426 | struct cfq_io_context *cic) | |
1427 | { | |
dbecf3ab OH |
1428 | struct rb_node **p; |
1429 | struct rb_node *parent; | |
e2d74ac0 | 1430 | struct cfq_io_context *__cic; |
0261d688 | 1431 | unsigned long flags; |
be3b0753 | 1432 | void *k; |
e2d74ac0 | 1433 | |
e2d74ac0 JA |
1434 | cic->ioc = ioc; |
1435 | cic->key = cfqd; | |
1436 | ||
dbecf3ab OH |
1437 | restart: |
1438 | parent = NULL; | |
1439 | p = &ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1440 | while (*p) { |
1441 | parent = *p; | |
1442 | __cic = rb_entry(parent, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1443 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1444 | k = __cic->key; | |
1445 | if (unlikely(!k)) { | |
be33c3a6 | 1446 | cfq_drop_dead_cic(ioc, __cic); |
dbecf3ab OH |
1447 | goto restart; |
1448 | } | |
e2d74ac0 | 1449 | |
be3b0753 | 1450 | if (cic->key < k) |
e2d74ac0 | 1451 | p = &(*p)->rb_left; |
be3b0753 | 1452 | else if (cic->key > k) |
e2d74ac0 JA |
1453 | p = &(*p)->rb_right; |
1454 | else | |
1455 | BUG(); | |
1456 | } | |
1457 | ||
1458 | rb_link_node(&cic->rb_node, parent, p); | |
1459 | rb_insert_color(&cic->rb_node, &ioc->cic_root); | |
fc46379d | 1460 | |
0261d688 | 1461 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
e2d74ac0 | 1462 | list_add(&cic->queue_list, &cfqd->cic_list); |
0261d688 | 1463 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
e2d74ac0 JA |
1464 | } |
1465 | ||
1da177e4 LT |
1466 | /* |
1467 | * Setup general io context and cfq io context. There can be several cfq | |
1468 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1469 | * than one device managed by cfq. |
1da177e4 LT |
1470 | */ |
1471 | static struct cfq_io_context * | |
e2d74ac0 | 1472 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1473 | { |
22e2c507 | 1474 | struct io_context *ioc = NULL; |
1da177e4 | 1475 | struct cfq_io_context *cic; |
1da177e4 | 1476 | |
22e2c507 | 1477 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1478 | |
b5deef90 | 1479 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1480 | if (!ioc) |
1481 | return NULL; | |
1482 | ||
e2d74ac0 JA |
1483 | cic = cfq_cic_rb_lookup(cfqd, ioc); |
1484 | if (cic) | |
1485 | goto out; | |
1da177e4 | 1486 | |
e2d74ac0 JA |
1487 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1488 | if (cic == NULL) | |
1489 | goto err; | |
1da177e4 | 1490 | |
e2d74ac0 | 1491 | cfq_cic_link(cfqd, ioc, cic); |
1da177e4 | 1492 | out: |
fc46379d JA |
1493 | smp_read_barrier_depends(); |
1494 | if (unlikely(ioc->ioprio_changed)) | |
1495 | cfq_ioc_set_ioprio(ioc); | |
1496 | ||
1da177e4 LT |
1497 | return cic; |
1498 | err: | |
1499 | put_io_context(ioc); | |
1500 | return NULL; | |
1501 | } | |
1502 | ||
22e2c507 JA |
1503 | static void |
1504 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1505 | { |
aaf1228d JA |
1506 | unsigned long elapsed = jiffies - cic->last_end_request; |
1507 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 1508 | |
22e2c507 JA |
1509 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1510 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1511 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1512 | } | |
1da177e4 | 1513 | |
206dc69b | 1514 | static void |
bb37b94c | 1515 | cfq_update_io_seektime(struct cfq_io_context *cic, struct request *rq) |
206dc69b JA |
1516 | { |
1517 | sector_t sdist; | |
1518 | u64 total; | |
1519 | ||
5e705374 JA |
1520 | if (cic->last_request_pos < rq->sector) |
1521 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1522 | else |
5e705374 | 1523 | sdist = cic->last_request_pos - rq->sector; |
206dc69b JA |
1524 | |
1525 | /* | |
1526 | * Don't allow the seek distance to get too large from the | |
1527 | * odd fragment, pagein, etc | |
1528 | */ | |
1529 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1530 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1531 | else | |
1532 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1533 | ||
1534 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1535 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1536 | total = cic->seek_total + (cic->seek_samples/2); | |
1537 | do_div(total, cic->seek_samples); | |
1538 | cic->seek_mean = (sector_t)total; | |
1539 | } | |
1da177e4 | 1540 | |
22e2c507 JA |
1541 | /* |
1542 | * Disable idle window if the process thinks too long or seeks so much that | |
1543 | * it doesn't matter | |
1544 | */ | |
1545 | static void | |
1546 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1547 | struct cfq_io_context *cic) | |
1548 | { | |
3b18152c | 1549 | int enable_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 1550 | |
caaa5f9f JA |
1551 | if (!cic->ioc->task || !cfqd->cfq_slice_idle || |
1552 | (cfqd->hw_tag && CIC_SEEKY(cic))) | |
22e2c507 JA |
1553 | enable_idle = 0; |
1554 | else if (sample_valid(cic->ttime_samples)) { | |
1555 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1556 | enable_idle = 0; | |
1557 | else | |
1558 | enable_idle = 1; | |
1da177e4 LT |
1559 | } |
1560 | ||
3b18152c JA |
1561 | if (enable_idle) |
1562 | cfq_mark_cfqq_idle_window(cfqq); | |
1563 | else | |
1564 | cfq_clear_cfqq_idle_window(cfqq); | |
22e2c507 | 1565 | } |
1da177e4 | 1566 | |
22e2c507 JA |
1567 | |
1568 | /* | |
1569 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1570 | * no or if we aren't sure, a 1 will cause a preempt. | |
1571 | */ | |
1572 | static int | |
1573 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1574 | struct request *rq) |
22e2c507 JA |
1575 | { |
1576 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1577 | ||
1578 | if (cfq_class_idle(new_cfqq)) | |
1579 | return 0; | |
1580 | ||
1581 | if (!cfqq) | |
caaa5f9f | 1582 | return 0; |
22e2c507 JA |
1583 | |
1584 | if (cfq_class_idle(cfqq)) | |
1585 | return 1; | |
3b18152c | 1586 | if (!cfq_cfqq_wait_request(new_cfqq)) |
22e2c507 JA |
1587 | return 0; |
1588 | /* | |
1589 | * if it doesn't have slice left, forget it | |
1590 | */ | |
1591 | if (new_cfqq->slice_left < cfqd->cfq_slice_idle) | |
1592 | return 0; | |
374f84ac JA |
1593 | /* |
1594 | * if the new request is sync, but the currently running queue is | |
1595 | * not, let the sync request have priority. | |
1596 | */ | |
5e705374 | 1597 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1598 | return 1; |
374f84ac JA |
1599 | /* |
1600 | * So both queues are sync. Let the new request get disk time if | |
1601 | * it's a metadata request and the current queue is doing regular IO. | |
1602 | */ | |
1603 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1604 | return 1; | |
22e2c507 JA |
1605 | |
1606 | return 0; | |
1607 | } | |
1608 | ||
1609 | /* | |
1610 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1611 | * let it have half of its nominal slice. | |
1612 | */ | |
1613 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1614 | { | |
bf572256 | 1615 | cfq_slice_expired(cfqd, 1); |
22e2c507 JA |
1616 | |
1617 | if (!cfqq->slice_left) | |
1618 | cfqq->slice_left = cfq_prio_to_slice(cfqd, cfqq) / 2; | |
1619 | ||
bf572256 JA |
1620 | /* |
1621 | * Put the new queue at the front of the of the current list, | |
1622 | * so we know that it will be selected next. | |
1623 | */ | |
1624 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
1625 | list_move(&cfqq->cfq_list, &cfqd->cur_rr); | |
1626 | ||
44f7c160 JA |
1627 | cfqq->slice_end = 0; |
1628 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
1629 | } |
1630 | ||
22e2c507 | 1631 | /* |
5e705374 | 1632 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1633 | * something we should do about it |
1634 | */ | |
1635 | static void | |
5e705374 JA |
1636 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1637 | struct request *rq) | |
22e2c507 | 1638 | { |
5e705374 | 1639 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1640 | |
374f84ac JA |
1641 | if (rq_is_meta(rq)) |
1642 | cfqq->meta_pending++; | |
1643 | ||
21183b07 | 1644 | /* |
5380a101 | 1645 | * check if this request is a better next-serve candidate)) { |
21183b07 | 1646 | */ |
5e705374 JA |
1647 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); |
1648 | BUG_ON(!cfqq->next_rq); | |
21183b07 | 1649 | |
9c2c38a1 JA |
1650 | /* |
1651 | * we never wait for an async request and we don't allow preemption | |
1652 | * of an async request. so just return early | |
1653 | */ | |
5e705374 | 1654 | if (!rq_is_sync(rq)) { |
12e9fddd JA |
1655 | /* |
1656 | * sync process issued an async request, if it's waiting | |
1657 | * then expire it and kick rq handling. | |
1658 | */ | |
1659 | if (cic == cfqd->active_cic && | |
1660 | del_timer(&cfqd->idle_slice_timer)) { | |
1661 | cfq_slice_expired(cfqd, 0); | |
dc72ef4a | 1662 | blk_start_queueing(cfqd->queue); |
12e9fddd | 1663 | } |
9c2c38a1 | 1664 | return; |
12e9fddd | 1665 | } |
22e2c507 | 1666 | |
9c2c38a1 | 1667 | cfq_update_io_thinktime(cfqd, cic); |
bb37b94c | 1668 | cfq_update_io_seektime(cic, rq); |
9c2c38a1 JA |
1669 | cfq_update_idle_window(cfqd, cfqq, cic); |
1670 | ||
5e705374 | 1671 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
22e2c507 JA |
1672 | |
1673 | if (cfqq == cfqd->active_queue) { | |
1674 | /* | |
1675 | * if we are waiting for a request for this queue, let it rip | |
1676 | * immediately and flag that we must not expire this queue | |
1677 | * just now | |
1678 | */ | |
3b18152c JA |
1679 | if (cfq_cfqq_wait_request(cfqq)) { |
1680 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 | 1681 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1682 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1683 | } |
5e705374 | 1684 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1685 | /* |
1686 | * not the active queue - expire current slice if it is | |
1687 | * idle and has expired it's mean thinktime or this new queue | |
1688 | * has some old slice time left and is of higher priority | |
1689 | */ | |
1690 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1691 | cfq_mark_cfqq_must_dispatch(cfqq); |
dc72ef4a | 1692 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1693 | } |
1da177e4 LT |
1694 | } |
1695 | ||
b4878f24 | 1696 | static void cfq_insert_request(request_queue_t *q, struct request *rq) |
1da177e4 | 1697 | { |
b4878f24 | 1698 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1699 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1700 | |
1701 | cfq_init_prio_data(cfqq); | |
1da177e4 | 1702 | |
5e705374 | 1703 | cfq_add_rq_rb(rq); |
1da177e4 | 1704 | |
22e2c507 JA |
1705 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1706 | ||
5e705374 | 1707 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1708 | } |
1709 | ||
1da177e4 LT |
1710 | static void cfq_completed_request(request_queue_t *q, struct request *rq) |
1711 | { | |
5e705374 | 1712 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1713 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1714 | const int sync = rq_is_sync(rq); |
b4878f24 | 1715 | unsigned long now; |
1da177e4 | 1716 | |
b4878f24 | 1717 | now = jiffies; |
1da177e4 | 1718 | |
b4878f24 JA |
1719 | WARN_ON(!cfqd->rq_in_driver); |
1720 | WARN_ON(!cfqq->on_dispatch[sync]); | |
1721 | cfqd->rq_in_driver--; | |
1722 | cfqq->on_dispatch[sync]--; | |
99f9628a | 1723 | cfqq->service_last = now; |
1da177e4 | 1724 | |
b4878f24 JA |
1725 | if (!cfq_class_idle(cfqq)) |
1726 | cfqd->last_end_request = now; | |
3b18152c | 1727 | |
98e41c7d | 1728 | cfq_resort_rr_list(cfqq, 0); |
1da177e4 | 1729 | |
caaa5f9f | 1730 | if (sync) |
5e705374 | 1731 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1732 | |
1733 | /* | |
1734 | * If this is the active queue, check if it needs to be expired, | |
1735 | * or if we want to idle in case it has no pending requests. | |
1736 | */ | |
1737 | if (cfqd->active_queue == cfqq) { | |
44f7c160 JA |
1738 | if (cfq_cfqq_slice_new(cfqq)) { |
1739 | cfq_set_prio_slice(cfqd, cfqq); | |
1740 | cfq_clear_cfqq_slice_new(cfqq); | |
1741 | } | |
1742 | if (cfq_slice_used(cfqq)) | |
caaa5f9f | 1743 | cfq_slice_expired(cfqd, 0); |
dd67d051 | 1744 | else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
caaa5f9f JA |
1745 | if (!cfq_arm_slice_timer(cfqd, cfqq)) |
1746 | cfq_schedule_dispatch(cfqd); | |
1747 | } | |
1748 | } | |
1da177e4 LT |
1749 | } |
1750 | ||
22e2c507 JA |
1751 | /* |
1752 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
1753 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
1754 | */ | |
1755 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 1756 | { |
22e2c507 JA |
1757 | const int ioprio_class = cfqq->ioprio_class; |
1758 | const int ioprio = cfqq->ioprio; | |
1da177e4 | 1759 | |
22e2c507 JA |
1760 | if (has_fs_excl()) { |
1761 | /* | |
1762 | * boost idle prio on transactions that would lock out other | |
1763 | * users of the filesystem | |
1764 | */ | |
1765 | if (cfq_class_idle(cfqq)) | |
1766 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1767 | if (cfqq->ioprio > IOPRIO_NORM) | |
1768 | cfqq->ioprio = IOPRIO_NORM; | |
1769 | } else { | |
1770 | /* | |
1771 | * check if we need to unboost the queue | |
1772 | */ | |
1773 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
1774 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
1775 | if (cfqq->ioprio != cfqq->org_ioprio) | |
1776 | cfqq->ioprio = cfqq->org_ioprio; | |
1777 | } | |
1da177e4 | 1778 | |
22e2c507 JA |
1779 | /* |
1780 | * refile between round-robin lists if we moved the priority class | |
1781 | */ | |
98e41c7d | 1782 | if ((ioprio_class != cfqq->ioprio_class || ioprio != cfqq->ioprio)) |
22e2c507 JA |
1783 | cfq_resort_rr_list(cfqq, 0); |
1784 | } | |
1da177e4 | 1785 | |
89850f7e | 1786 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 1787 | { |
3b18152c | 1788 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 1789 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 1790 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 1791 | return ELV_MQUEUE_MUST; |
3b18152c | 1792 | } |
1da177e4 | 1793 | |
22e2c507 | 1794 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
1795 | } |
1796 | ||
cb78b285 | 1797 | static int cfq_may_queue(request_queue_t *q, int rw) |
22e2c507 JA |
1798 | { |
1799 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1800 | struct task_struct *tsk = current; | |
1801 | struct cfq_queue *cfqq; | |
7749a8d4 JA |
1802 | unsigned int key; |
1803 | ||
1804 | key = cfq_queue_pid(tsk, rw, rw & REQ_RW_SYNC); | |
22e2c507 JA |
1805 | |
1806 | /* | |
1807 | * don't force setup of a queue from here, as a call to may_queue | |
1808 | * does not necessarily imply that a request actually will be queued. | |
1809 | * so just lookup a possibly existing queue, or return 'may queue' | |
1810 | * if that fails | |
1811 | */ | |
7749a8d4 | 1812 | cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio); |
22e2c507 JA |
1813 | if (cfqq) { |
1814 | cfq_init_prio_data(cfqq); | |
1815 | cfq_prio_boost(cfqq); | |
1816 | ||
89850f7e | 1817 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
1818 | } |
1819 | ||
1820 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
1821 | } |
1822 | ||
1da177e4 LT |
1823 | /* |
1824 | * queue lock held here | |
1825 | */ | |
bb37b94c | 1826 | static void cfq_put_request(struct request *rq) |
1da177e4 | 1827 | { |
5e705374 | 1828 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1829 | |
5e705374 | 1830 | if (cfqq) { |
22e2c507 | 1831 | const int rw = rq_data_dir(rq); |
1da177e4 | 1832 | |
22e2c507 JA |
1833 | BUG_ON(!cfqq->allocated[rw]); |
1834 | cfqq->allocated[rw]--; | |
1da177e4 | 1835 | |
5e705374 | 1836 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 1837 | |
1da177e4 | 1838 | rq->elevator_private = NULL; |
5e705374 | 1839 | rq->elevator_private2 = NULL; |
1da177e4 | 1840 | |
1da177e4 LT |
1841 | cfq_put_queue(cfqq); |
1842 | } | |
1843 | } | |
1844 | ||
1845 | /* | |
22e2c507 | 1846 | * Allocate cfq data structures associated with this request. |
1da177e4 | 1847 | */ |
22e2c507 | 1848 | static int |
cb78b285 | 1849 | cfq_set_request(request_queue_t *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
1850 | { |
1851 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3b18152c | 1852 | struct task_struct *tsk = current; |
1da177e4 LT |
1853 | struct cfq_io_context *cic; |
1854 | const int rw = rq_data_dir(rq); | |
7749a8d4 JA |
1855 | const int is_sync = rq_is_sync(rq); |
1856 | pid_t key = cfq_queue_pid(tsk, rw, is_sync); | |
22e2c507 | 1857 | struct cfq_queue *cfqq; |
1da177e4 LT |
1858 | unsigned long flags; |
1859 | ||
1860 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1861 | ||
e2d74ac0 | 1862 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 1863 | |
1da177e4 LT |
1864 | spin_lock_irqsave(q->queue_lock, flags); |
1865 | ||
22e2c507 JA |
1866 | if (!cic) |
1867 | goto queue_fail; | |
1868 | ||
12a05732 | 1869 | if (!cic->cfqq[is_sync]) { |
6f325a13 | 1870 | cfqq = cfq_get_queue(cfqd, key, tsk, gfp_mask); |
22e2c507 JA |
1871 | if (!cfqq) |
1872 | goto queue_fail; | |
1da177e4 | 1873 | |
12a05732 | 1874 | cic->cfqq[is_sync] = cfqq; |
22e2c507 | 1875 | } else |
12a05732 | 1876 | cfqq = cic->cfqq[is_sync]; |
1da177e4 LT |
1877 | |
1878 | cfqq->allocated[rw]++; | |
3b18152c | 1879 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 1880 | atomic_inc(&cfqq->ref); |
1da177e4 | 1881 | |
5e705374 | 1882 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 1883 | |
5e705374 JA |
1884 | rq->elevator_private = cic; |
1885 | rq->elevator_private2 = cfqq; | |
1886 | return 0; | |
1da177e4 | 1887 | |
22e2c507 JA |
1888 | queue_fail: |
1889 | if (cic) | |
1890 | put_io_context(cic->ioc); | |
89850f7e | 1891 | |
3b18152c | 1892 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
1893 | spin_unlock_irqrestore(q->queue_lock, flags); |
1894 | return 1; | |
1895 | } | |
1896 | ||
65f27f38 | 1897 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 1898 | { |
65f27f38 DH |
1899 | struct cfq_data *cfqd = |
1900 | container_of(work, struct cfq_data, unplug_work); | |
1901 | request_queue_t *q = cfqd->queue; | |
22e2c507 JA |
1902 | unsigned long flags; |
1903 | ||
1904 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 1905 | blk_start_queueing(q); |
22e2c507 JA |
1906 | spin_unlock_irqrestore(q->queue_lock, flags); |
1907 | } | |
1908 | ||
1909 | /* | |
1910 | * Timer running if the active_queue is currently idling inside its time slice | |
1911 | */ | |
1912 | static void cfq_idle_slice_timer(unsigned long data) | |
1913 | { | |
1914 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1915 | struct cfq_queue *cfqq; | |
1916 | unsigned long flags; | |
1917 | ||
1918 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1919 | ||
1920 | if ((cfqq = cfqd->active_queue) != NULL) { | |
22e2c507 JA |
1921 | /* |
1922 | * expired | |
1923 | */ | |
44f7c160 | 1924 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
1925 | goto expire; |
1926 | ||
1927 | /* | |
1928 | * only expire and reinvoke request handler, if there are | |
1929 | * other queues with pending requests | |
1930 | */ | |
caaa5f9f | 1931 | if (!cfqd->busy_queues) |
22e2c507 | 1932 | goto out_cont; |
22e2c507 JA |
1933 | |
1934 | /* | |
1935 | * not expired and it has a request pending, let it dispatch | |
1936 | */ | |
dd67d051 | 1937 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 1938 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
1939 | goto out_kick; |
1940 | } | |
1941 | } | |
1942 | expire: | |
1943 | cfq_slice_expired(cfqd, 0); | |
1944 | out_kick: | |
3b18152c | 1945 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
1946 | out_cont: |
1947 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1948 | } | |
1949 | ||
1950 | /* | |
1951 | * Timer running if an idle class queue is waiting for service | |
1952 | */ | |
1953 | static void cfq_idle_class_timer(unsigned long data) | |
1954 | { | |
1955 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1956 | unsigned long flags, end; | |
1957 | ||
1958 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1959 | ||
1960 | /* | |
1961 | * race with a non-idle queue, reset timer | |
1962 | */ | |
1963 | end = cfqd->last_end_request + CFQ_IDLE_GRACE; | |
ae818a38 JA |
1964 | if (!time_after_eq(jiffies, end)) |
1965 | mod_timer(&cfqd->idle_class_timer, end); | |
1966 | else | |
3b18152c | 1967 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
1968 | |
1969 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1970 | } | |
1971 | ||
3b18152c JA |
1972 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
1973 | { | |
1974 | del_timer_sync(&cfqd->idle_slice_timer); | |
1975 | del_timer_sync(&cfqd->idle_class_timer); | |
1976 | blk_sync_queue(cfqd->queue); | |
1977 | } | |
22e2c507 | 1978 | |
1da177e4 LT |
1979 | static void cfq_exit_queue(elevator_t *e) |
1980 | { | |
22e2c507 | 1981 | struct cfq_data *cfqd = e->elevator_data; |
d9ff4187 | 1982 | request_queue_t *q = cfqd->queue; |
22e2c507 | 1983 | |
3b18152c | 1984 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 1985 | |
d9ff4187 | 1986 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 1987 | |
d9ff4187 AV |
1988 | if (cfqd->active_queue) |
1989 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); | |
e2d74ac0 JA |
1990 | |
1991 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
1992 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
1993 | struct cfq_io_context, | |
1994 | queue_list); | |
89850f7e JA |
1995 | |
1996 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 1997 | } |
e2d74ac0 | 1998 | |
d9ff4187 | 1999 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
2000 | |
2001 | cfq_shutdown_timer_wq(cfqd); | |
2002 | ||
a90d742e AV |
2003 | kfree(cfqd->cfq_hash); |
2004 | kfree(cfqd); | |
1da177e4 LT |
2005 | } |
2006 | ||
bb37b94c | 2007 | static void *cfq_init_queue(request_queue_t *q) |
1da177e4 LT |
2008 | { |
2009 | struct cfq_data *cfqd; | |
2010 | int i; | |
2011 | ||
b5deef90 | 2012 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node); |
1da177e4 | 2013 | if (!cfqd) |
bc1c1169 | 2014 | return NULL; |
1da177e4 LT |
2015 | |
2016 | memset(cfqd, 0, sizeof(*cfqd)); | |
22e2c507 JA |
2017 | |
2018 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
2019 | INIT_LIST_HEAD(&cfqd->rr_list[i]); | |
2020 | ||
2021 | INIT_LIST_HEAD(&cfqd->busy_rr); | |
2022 | INIT_LIST_HEAD(&cfqd->cur_rr); | |
2023 | INIT_LIST_HEAD(&cfqd->idle_rr); | |
d9ff4187 | 2024 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2025 | |
b5deef90 | 2026 | cfqd->cfq_hash = kmalloc_node(sizeof(struct hlist_head) * CFQ_QHASH_ENTRIES, GFP_KERNEL, q->node); |
1da177e4 | 2027 | if (!cfqd->cfq_hash) |
5e705374 | 2028 | goto out_free; |
1da177e4 | 2029 | |
1da177e4 LT |
2030 | for (i = 0; i < CFQ_QHASH_ENTRIES; i++) |
2031 | INIT_HLIST_HEAD(&cfqd->cfq_hash[i]); | |
2032 | ||
1da177e4 | 2033 | cfqd->queue = q; |
1da177e4 | 2034 | |
22e2c507 JA |
2035 | init_timer(&cfqd->idle_slice_timer); |
2036 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2037 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2038 | ||
2039 | init_timer(&cfqd->idle_class_timer); | |
2040 | cfqd->idle_class_timer.function = cfq_idle_class_timer; | |
2041 | cfqd->idle_class_timer.data = (unsigned long) cfqd; | |
2042 | ||
65f27f38 | 2043 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2044 | |
1da177e4 | 2045 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2046 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2047 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2048 | cfqd->cfq_back_max = cfq_back_max; |
2049 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2050 | cfqd->cfq_slice[0] = cfq_slice_async; |
2051 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2052 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2053 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
3b18152c | 2054 | |
bc1c1169 | 2055 | return cfqd; |
5e705374 | 2056 | out_free: |
1da177e4 | 2057 | kfree(cfqd); |
bc1c1169 | 2058 | return NULL; |
1da177e4 LT |
2059 | } |
2060 | ||
2061 | static void cfq_slab_kill(void) | |
2062 | { | |
1da177e4 LT |
2063 | if (cfq_pool) |
2064 | kmem_cache_destroy(cfq_pool); | |
2065 | if (cfq_ioc_pool) | |
2066 | kmem_cache_destroy(cfq_ioc_pool); | |
2067 | } | |
2068 | ||
2069 | static int __init cfq_slab_setup(void) | |
2070 | { | |
1da177e4 LT |
2071 | cfq_pool = kmem_cache_create("cfq_pool", sizeof(struct cfq_queue), 0, 0, |
2072 | NULL, NULL); | |
2073 | if (!cfq_pool) | |
2074 | goto fail; | |
2075 | ||
2076 | cfq_ioc_pool = kmem_cache_create("cfq_ioc_pool", | |
2077 | sizeof(struct cfq_io_context), 0, 0, NULL, NULL); | |
2078 | if (!cfq_ioc_pool) | |
2079 | goto fail; | |
2080 | ||
2081 | return 0; | |
2082 | fail: | |
2083 | cfq_slab_kill(); | |
2084 | return -ENOMEM; | |
2085 | } | |
2086 | ||
1da177e4 LT |
2087 | /* |
2088 | * sysfs parts below --> | |
2089 | */ | |
1da177e4 LT |
2090 | |
2091 | static ssize_t | |
2092 | cfq_var_show(unsigned int var, char *page) | |
2093 | { | |
2094 | return sprintf(page, "%d\n", var); | |
2095 | } | |
2096 | ||
2097 | static ssize_t | |
2098 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2099 | { | |
2100 | char *p = (char *) page; | |
2101 | ||
2102 | *var = simple_strtoul(p, &p, 10); | |
2103 | return count; | |
2104 | } | |
2105 | ||
1da177e4 | 2106 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
3d1ab40f | 2107 | static ssize_t __FUNC(elevator_t *e, char *page) \ |
1da177e4 | 2108 | { \ |
3d1ab40f | 2109 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2110 | unsigned int __data = __VAR; \ |
2111 | if (__CONV) \ | |
2112 | __data = jiffies_to_msecs(__data); \ | |
2113 | return cfq_var_show(__data, (page)); \ | |
2114 | } | |
2115 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2116 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2117 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2118 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2119 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2120 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2121 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2122 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2123 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2124 | #undef SHOW_FUNCTION |
2125 | ||
2126 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
3d1ab40f | 2127 | static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \ |
1da177e4 | 2128 | { \ |
3d1ab40f | 2129 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2130 | unsigned int __data; \ |
2131 | int ret = cfq_var_store(&__data, (page), count); \ | |
2132 | if (__data < (MIN)) \ | |
2133 | __data = (MIN); \ | |
2134 | else if (__data > (MAX)) \ | |
2135 | __data = (MAX); \ | |
2136 | if (__CONV) \ | |
2137 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2138 | else \ | |
2139 | *(__PTR) = __data; \ | |
2140 | return ret; \ | |
2141 | } | |
2142 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
22e2c507 JA |
2143 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1); |
2144 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1); | |
e572ec7e AV |
2145 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
2146 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0); | |
22e2c507 JA |
2147 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2148 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2149 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
2150 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0); | |
1da177e4 LT |
2151 | #undef STORE_FUNCTION |
2152 | ||
e572ec7e AV |
2153 | #define CFQ_ATTR(name) \ |
2154 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2155 | ||
2156 | static struct elv_fs_entry cfq_attrs[] = { | |
2157 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2158 | CFQ_ATTR(fifo_expire_sync), |
2159 | CFQ_ATTR(fifo_expire_async), | |
2160 | CFQ_ATTR(back_seek_max), | |
2161 | CFQ_ATTR(back_seek_penalty), | |
2162 | CFQ_ATTR(slice_sync), | |
2163 | CFQ_ATTR(slice_async), | |
2164 | CFQ_ATTR(slice_async_rq), | |
2165 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2166 | __ATTR_NULL |
1da177e4 LT |
2167 | }; |
2168 | ||
1da177e4 LT |
2169 | static struct elevator_type iosched_cfq = { |
2170 | .ops = { | |
2171 | .elevator_merge_fn = cfq_merge, | |
2172 | .elevator_merged_fn = cfq_merged_request, | |
2173 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2174 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2175 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2176 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2177 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2178 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2179 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2180 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2181 | .elevator_former_req_fn = elv_rb_former_request, |
2182 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2183 | .elevator_set_req_fn = cfq_set_request, |
2184 | .elevator_put_req_fn = cfq_put_request, | |
2185 | .elevator_may_queue_fn = cfq_may_queue, | |
2186 | .elevator_init_fn = cfq_init_queue, | |
2187 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2188 | .trim = cfq_free_io_context, |
1da177e4 | 2189 | }, |
3d1ab40f | 2190 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2191 | .elevator_name = "cfq", |
2192 | .elevator_owner = THIS_MODULE, | |
2193 | }; | |
2194 | ||
2195 | static int __init cfq_init(void) | |
2196 | { | |
2197 | int ret; | |
2198 | ||
22e2c507 JA |
2199 | /* |
2200 | * could be 0 on HZ < 1000 setups | |
2201 | */ | |
2202 | if (!cfq_slice_async) | |
2203 | cfq_slice_async = 1; | |
2204 | if (!cfq_slice_idle) | |
2205 | cfq_slice_idle = 1; | |
2206 | ||
1da177e4 LT |
2207 | if (cfq_slab_setup()) |
2208 | return -ENOMEM; | |
2209 | ||
2210 | ret = elv_register(&iosched_cfq); | |
22e2c507 JA |
2211 | if (ret) |
2212 | cfq_slab_kill(); | |
1da177e4 | 2213 | |
1da177e4 LT |
2214 | return ret; |
2215 | } | |
2216 | ||
2217 | static void __exit cfq_exit(void) | |
2218 | { | |
6e9a4738 | 2219 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2220 | elv_unregister(&iosched_cfq); |
334e94de | 2221 | ioc_gone = &all_gone; |
fba82272 OH |
2222 | /* ioc_gone's update must be visible before reading ioc_count */ |
2223 | smp_wmb(); | |
4050cf16 | 2224 | if (elv_ioc_count_read(ioc_count)) |
fba82272 | 2225 | wait_for_completion(ioc_gone); |
334e94de | 2226 | synchronize_rcu(); |
83521d3e | 2227 | cfq_slab_kill(); |
1da177e4 LT |
2228 | } |
2229 | ||
2230 | module_init(cfq_init); | |
2231 | module_exit(cfq_exit); | |
2232 | ||
2233 | MODULE_AUTHOR("Jens Axboe"); | |
2234 | MODULE_LICENSE("GPL"); | |
2235 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |