]>
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 | 149 | unsigned long slice_end; |
99f9628a | 150 | unsigned long service_last; |
c5b680f3 | 151 | long slice_resid; |
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; | |
c5b680f3 JA |
254 | cfqq->slice_end += cfqq->slice_resid; |
255 | ||
256 | /* | |
257 | * Don't carry over residual for more than one slice, we only want | |
258 | * to slightly correct the fairness. Carrying over forever would | |
259 | * easily introduce oscillations. | |
260 | */ | |
261 | cfqq->slice_resid = 0; | |
44f7c160 JA |
262 | } |
263 | ||
264 | /* | |
265 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
266 | * isn't valid until the first request from the dispatch is activated | |
267 | * and the slice time set. | |
268 | */ | |
269 | static inline int cfq_slice_used(struct cfq_queue *cfqq) | |
270 | { | |
271 | if (cfq_cfqq_slice_new(cfqq)) | |
272 | return 0; | |
273 | if (time_before(jiffies, cfqq->slice_end)) | |
274 | return 0; | |
275 | ||
276 | return 1; | |
277 | } | |
278 | ||
1da177e4 | 279 | /* |
5e705374 | 280 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 281 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 282 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 283 | */ |
5e705374 JA |
284 | static struct request * |
285 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
286 | { |
287 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 288 | unsigned long back_max; |
e8a99053 AM |
289 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
290 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
291 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 292 | |
5e705374 JA |
293 | if (rq1 == NULL || rq1 == rq2) |
294 | return rq2; | |
295 | if (rq2 == NULL) | |
296 | return rq1; | |
9c2c38a1 | 297 | |
5e705374 JA |
298 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
299 | return rq1; | |
300 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
301 | return rq2; | |
374f84ac JA |
302 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
303 | return rq1; | |
304 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
305 | return rq2; | |
1da177e4 | 306 | |
5e705374 JA |
307 | s1 = rq1->sector; |
308 | s2 = rq2->sector; | |
1da177e4 LT |
309 | |
310 | last = cfqd->last_sector; | |
311 | ||
1da177e4 LT |
312 | /* |
313 | * by definition, 1KiB is 2 sectors | |
314 | */ | |
315 | back_max = cfqd->cfq_back_max * 2; | |
316 | ||
317 | /* | |
318 | * Strict one way elevator _except_ in the case where we allow | |
319 | * short backward seeks which are biased as twice the cost of a | |
320 | * similar forward seek. | |
321 | */ | |
322 | if (s1 >= last) | |
323 | d1 = s1 - last; | |
324 | else if (s1 + back_max >= last) | |
325 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
326 | else | |
e8a99053 | 327 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
328 | |
329 | if (s2 >= last) | |
330 | d2 = s2 - last; | |
331 | else if (s2 + back_max >= last) | |
332 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
333 | else | |
e8a99053 | 334 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
335 | |
336 | /* Found required data */ | |
e8a99053 AM |
337 | |
338 | /* | |
339 | * By doing switch() on the bit mask "wrap" we avoid having to | |
340 | * check two variables for all permutations: --> faster! | |
341 | */ | |
342 | switch (wrap) { | |
5e705374 | 343 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 344 | if (d1 < d2) |
5e705374 | 345 | return rq1; |
e8a99053 | 346 | else if (d2 < d1) |
5e705374 | 347 | return rq2; |
e8a99053 AM |
348 | else { |
349 | if (s1 >= s2) | |
5e705374 | 350 | return rq1; |
e8a99053 | 351 | else |
5e705374 | 352 | return rq2; |
e8a99053 | 353 | } |
1da177e4 | 354 | |
e8a99053 | 355 | case CFQ_RQ2_WRAP: |
5e705374 | 356 | return rq1; |
e8a99053 | 357 | case CFQ_RQ1_WRAP: |
5e705374 JA |
358 | return rq2; |
359 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
360 | default: |
361 | /* | |
362 | * Since both rqs are wrapped, | |
363 | * start with the one that's further behind head | |
364 | * (--> only *one* back seek required), | |
365 | * since back seek takes more time than forward. | |
366 | */ | |
367 | if (s1 <= s2) | |
5e705374 | 368 | return rq1; |
1da177e4 | 369 | else |
5e705374 | 370 | return rq2; |
1da177e4 LT |
371 | } |
372 | } | |
373 | ||
374 | /* | |
375 | * would be nice to take fifo expire time into account as well | |
376 | */ | |
5e705374 JA |
377 | static struct request * |
378 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
379 | struct request *last) | |
1da177e4 | 380 | { |
21183b07 JA |
381 | struct rb_node *rbnext = rb_next(&last->rb_node); |
382 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 383 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 384 | |
21183b07 | 385 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
386 | |
387 | if (rbprev) | |
5e705374 | 388 | prev = rb_entry_rq(rbprev); |
1da177e4 | 389 | |
21183b07 | 390 | if (rbnext) |
5e705374 | 391 | next = rb_entry_rq(rbnext); |
21183b07 JA |
392 | else { |
393 | rbnext = rb_first(&cfqq->sort_list); | |
394 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 395 | next = rb_entry_rq(rbnext); |
21183b07 | 396 | } |
1da177e4 | 397 | |
21183b07 | 398 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
399 | } |
400 | ||
22e2c507 | 401 | static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted) |
1da177e4 | 402 | { |
22e2c507 | 403 | struct cfq_data *cfqd = cfqq->cfqd; |
99f9628a JA |
404 | struct list_head *list, *n; |
405 | struct cfq_queue *__cfqq; | |
1da177e4 | 406 | |
98e41c7d JA |
407 | /* |
408 | * Resorting requires the cfqq to be on the RR list already. | |
409 | */ | |
410 | if (!cfq_cfqq_on_rr(cfqq)) | |
411 | return; | |
1da177e4 | 412 | |
22e2c507 | 413 | list_del(&cfqq->cfq_list); |
1da177e4 | 414 | |
22e2c507 JA |
415 | if (cfq_class_rt(cfqq)) |
416 | list = &cfqd->cur_rr; | |
417 | else if (cfq_class_idle(cfqq)) | |
418 | list = &cfqd->idle_rr; | |
419 | else { | |
420 | /* | |
421 | * if cfqq has requests in flight, don't allow it to be | |
422 | * found in cfq_set_active_queue before it has finished them. | |
423 | * this is done to increase fairness between a process that | |
424 | * has lots of io pending vs one that only generates one | |
425 | * sporadically or synchronously | |
426 | */ | |
3b18152c | 427 | if (cfq_cfqq_dispatched(cfqq)) |
22e2c507 JA |
428 | list = &cfqd->busy_rr; |
429 | else | |
430 | list = &cfqd->rr_list[cfqq->ioprio]; | |
1da177e4 LT |
431 | } |
432 | ||
53b03744 | 433 | if (preempted || cfq_cfqq_queue_new(cfqq)) { |
99f9628a JA |
434 | /* |
435 | * If this queue was preempted or is new (never been serviced), | |
436 | * let it be added first for fairness but beind other new | |
437 | * queues. | |
438 | */ | |
439 | n = list; | |
53b03744 JA |
440 | while (n->next != list) { |
441 | __cfqq = list_entry_cfqq(n->next); | |
442 | if (!cfq_cfqq_queue_new(__cfqq)) | |
443 | break; | |
1da177e4 | 444 | |
53b03744 JA |
445 | n = n->next; |
446 | } | |
99f9628a JA |
447 | list_add_tail(&cfqq->cfq_list, n); |
448 | } else if (!cfq_cfqq_class_sync(cfqq)) { | |
449 | /* | |
450 | * async queue always goes to the end. this wont be overly | |
451 | * unfair to writes, as the sort of the sync queue wont be | |
452 | * allowed to pass the async queue again. | |
453 | */ | |
454 | list_add_tail(&cfqq->cfq_list, list); | |
455 | } else { | |
456 | /* | |
457 | * sort by last service, but don't cross a new or async | |
458 | * queue. we don't cross a new queue because it hasn't been | |
459 | * service before, and we don't cross an async queue because | |
460 | * it gets added to the end on expire. | |
461 | */ | |
462 | n = list; | |
463 | while ((n = n->prev) != list) { | |
464 | struct cfq_queue *__cfqq = list_entry_cfqq(n); | |
1da177e4 | 465 | |
99f9628a JA |
466 | if (!cfq_cfqq_class_sync(cfqq) || !__cfqq->service_last) |
467 | break; | |
468 | if (time_before(__cfqq->service_last, cfqq->service_last)) | |
469 | break; | |
470 | } | |
471 | list_add(&cfqq->cfq_list, n); | |
1da177e4 | 472 | } |
1da177e4 LT |
473 | } |
474 | ||
475 | /* | |
476 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 477 | * the pending list according to last request service |
1da177e4 LT |
478 | */ |
479 | static inline void | |
b4878f24 | 480 | cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 481 | { |
3b18152c JA |
482 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
483 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
484 | cfqd->busy_queues++; |
485 | ||
b4878f24 | 486 | cfq_resort_rr_list(cfqq, 0); |
1da177e4 LT |
487 | } |
488 | ||
489 | static inline void | |
490 | cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
491 | { | |
3b18152c JA |
492 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
493 | cfq_clear_cfqq_on_rr(cfqq); | |
981a7973 | 494 | list_del_init(&cfqq->cfq_list); |
1da177e4 LT |
495 | |
496 | BUG_ON(!cfqd->busy_queues); | |
497 | cfqd->busy_queues--; | |
498 | } | |
499 | ||
500 | /* | |
501 | * rb tree support functions | |
502 | */ | |
5e705374 | 503 | static inline void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 504 | { |
5e705374 | 505 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 506 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 507 | const int sync = rq_is_sync(rq); |
1da177e4 | 508 | |
b4878f24 JA |
509 | BUG_ON(!cfqq->queued[sync]); |
510 | cfqq->queued[sync]--; | |
1da177e4 | 511 | |
5e705374 | 512 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 513 | |
dd67d051 | 514 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 515 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
516 | } |
517 | ||
5e705374 | 518 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 519 | { |
5e705374 | 520 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 521 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 522 | struct request *__alias; |
1da177e4 | 523 | |
5380a101 | 524 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
525 | |
526 | /* | |
527 | * looks a little odd, but the first insert might return an alias. | |
528 | * if that happens, put the alias on the dispatch list | |
529 | */ | |
21183b07 | 530 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 531 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
532 | |
533 | if (!cfq_cfqq_on_rr(cfqq)) | |
534 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
535 | |
536 | /* | |
537 | * check if this request is a better next-serve candidate | |
538 | */ | |
539 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); | |
540 | BUG_ON(!cfqq->next_rq); | |
1da177e4 LT |
541 | } |
542 | ||
543 | static inline void | |
5e705374 | 544 | cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 545 | { |
5380a101 JA |
546 | elv_rb_del(&cfqq->sort_list, rq); |
547 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 548 | cfq_add_rq_rb(rq); |
1da177e4 LT |
549 | } |
550 | ||
206dc69b JA |
551 | static struct request * |
552 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 553 | { |
206dc69b | 554 | struct task_struct *tsk = current; |
7749a8d4 | 555 | pid_t key = cfq_queue_pid(tsk, bio_data_dir(bio), bio_sync(bio)); |
206dc69b | 556 | struct cfq_queue *cfqq; |
1da177e4 | 557 | |
206dc69b | 558 | cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio); |
89850f7e JA |
559 | if (cfqq) { |
560 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
561 | ||
21183b07 | 562 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 563 | } |
1da177e4 | 564 | |
1da177e4 LT |
565 | return NULL; |
566 | } | |
567 | ||
b4878f24 | 568 | static void cfq_activate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 569 | { |
22e2c507 | 570 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 571 | |
b4878f24 | 572 | cfqd->rq_in_driver++; |
25776e35 JA |
573 | |
574 | /* | |
575 | * If the depth is larger 1, it really could be queueing. But lets | |
576 | * make the mark a little higher - idling could still be good for | |
577 | * low queueing, and a low queueing number could also just indicate | |
578 | * a SCSI mid layer like behaviour where limit+1 is often seen. | |
579 | */ | |
580 | if (!cfqd->hw_tag && cfqd->rq_in_driver > 4) | |
581 | cfqd->hw_tag = 1; | |
1da177e4 LT |
582 | } |
583 | ||
b4878f24 | 584 | static void cfq_deactivate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 585 | { |
b4878f24 JA |
586 | struct cfq_data *cfqd = q->elevator->elevator_data; |
587 | ||
588 | WARN_ON(!cfqd->rq_in_driver); | |
589 | cfqd->rq_in_driver--; | |
1da177e4 LT |
590 | } |
591 | ||
b4878f24 | 592 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 593 | { |
5e705374 | 594 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 595 | |
5e705374 JA |
596 | if (cfqq->next_rq == rq) |
597 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 598 | |
b4878f24 | 599 | list_del_init(&rq->queuelist); |
5e705374 | 600 | cfq_del_rq_rb(rq); |
374f84ac JA |
601 | |
602 | if (rq_is_meta(rq)) { | |
603 | WARN_ON(!cfqq->meta_pending); | |
604 | cfqq->meta_pending--; | |
605 | } | |
1da177e4 LT |
606 | } |
607 | ||
608 | static int | |
609 | cfq_merge(request_queue_t *q, struct request **req, struct bio *bio) | |
610 | { | |
611 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
612 | struct request *__rq; | |
1da177e4 | 613 | |
206dc69b | 614 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 615 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
616 | *req = __rq; |
617 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
618 | } |
619 | ||
620 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
621 | } |
622 | ||
21183b07 JA |
623 | static void cfq_merged_request(request_queue_t *q, struct request *req, |
624 | int type) | |
1da177e4 | 625 | { |
21183b07 | 626 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 627 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 628 | |
5e705374 | 629 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 630 | } |
1da177e4 LT |
631 | } |
632 | ||
633 | static void | |
634 | cfq_merged_requests(request_queue_t *q, struct request *rq, | |
635 | struct request *next) | |
636 | { | |
22e2c507 JA |
637 | /* |
638 | * reposition in fifo if next is older than rq | |
639 | */ | |
640 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
641 | time_before(next->start_time, rq->start_time)) | |
642 | list_move(&rq->queuelist, &next->queuelist); | |
643 | ||
b4878f24 | 644 | cfq_remove_request(next); |
22e2c507 JA |
645 | } |
646 | ||
da775265 JA |
647 | static int cfq_allow_merge(request_queue_t *q, struct request *rq, |
648 | struct bio *bio) | |
649 | { | |
650 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
651 | const int rw = bio_data_dir(bio); | |
652 | struct cfq_queue *cfqq; | |
653 | pid_t key; | |
654 | ||
655 | /* | |
ec8acb69 | 656 | * Disallow merge of a sync bio into an async request. |
da775265 | 657 | */ |
ec8acb69 | 658 | if ((bio_data_dir(bio) == READ || bio_sync(bio)) && !rq_is_sync(rq)) |
da775265 JA |
659 | return 0; |
660 | ||
661 | /* | |
719d3402 JA |
662 | * Lookup the cfqq that this bio will be queued with. Allow |
663 | * merge only if rq is queued there. | |
da775265 | 664 | */ |
719d3402 | 665 | key = cfq_queue_pid(current, rw, bio_sync(bio)); |
da775265 | 666 | cfqq = cfq_find_cfq_hash(cfqd, key, current->ioprio); |
719d3402 JA |
667 | |
668 | if (cfqq == RQ_CFQQ(rq)) | |
669 | return 1; | |
da775265 | 670 | |
ec8acb69 | 671 | return 0; |
da775265 JA |
672 | } |
673 | ||
22e2c507 JA |
674 | static inline void |
675 | __cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
676 | { | |
677 | if (cfqq) { | |
678 | /* | |
679 | * stop potential idle class queues waiting service | |
680 | */ | |
681 | del_timer(&cfqd->idle_class_timer); | |
682 | ||
22e2c507 | 683 | cfqq->slice_end = 0; |
3b18152c JA |
684 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
685 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 686 | cfq_mark_cfqq_slice_new(cfqq); |
22e2c507 JA |
687 | } |
688 | ||
689 | cfqd->active_queue = cfqq; | |
690 | } | |
691 | ||
7b14e3b5 JA |
692 | /* |
693 | * current cfqq expired its slice (or was too idle), select new one | |
694 | */ | |
695 | static void | |
696 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
3c6bd2f8 | 697 | int preempted, int timed_out) |
7b14e3b5 | 698 | { |
7b14e3b5 JA |
699 | if (cfq_cfqq_wait_request(cfqq)) |
700 | del_timer(&cfqd->idle_slice_timer); | |
701 | ||
7b14e3b5 JA |
702 | cfq_clear_cfqq_must_dispatch(cfqq); |
703 | cfq_clear_cfqq_wait_request(cfqq); | |
53b03744 | 704 | cfq_clear_cfqq_queue_new(cfqq); |
7b14e3b5 JA |
705 | |
706 | /* | |
707 | * store what was left of this slice, if the queue idled out | |
708 | * or was preempted | |
709 | */ | |
3c6bd2f8 | 710 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) |
c5b680f3 | 711 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b14e3b5 | 712 | |
98e41c7d | 713 | cfq_resort_rr_list(cfqq, preempted); |
7b14e3b5 JA |
714 | |
715 | if (cfqq == cfqd->active_queue) | |
716 | cfqd->active_queue = NULL; | |
717 | ||
718 | if (cfqd->active_cic) { | |
719 | put_io_context(cfqd->active_cic->ioc); | |
720 | cfqd->active_cic = NULL; | |
721 | } | |
722 | ||
723 | cfqd->dispatch_slice = 0; | |
724 | } | |
725 | ||
3c6bd2f8 JA |
726 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted, |
727 | int timed_out) | |
7b14e3b5 JA |
728 | { |
729 | struct cfq_queue *cfqq = cfqd->active_queue; | |
730 | ||
731 | if (cfqq) | |
3c6bd2f8 | 732 | __cfq_slice_expired(cfqd, cfqq, preempted, timed_out); |
7b14e3b5 JA |
733 | } |
734 | ||
22e2c507 JA |
735 | /* |
736 | * 0 | |
737 | * 0,1 | |
738 | * 0,1,2 | |
739 | * 0,1,2,3 | |
740 | * 0,1,2,3,4 | |
741 | * 0,1,2,3,4,5 | |
742 | * 0,1,2,3,4,5,6 | |
743 | * 0,1,2,3,4,5,6,7 | |
744 | */ | |
745 | static int cfq_get_next_prio_level(struct cfq_data *cfqd) | |
746 | { | |
747 | int prio, wrap; | |
748 | ||
749 | prio = -1; | |
750 | wrap = 0; | |
751 | do { | |
752 | int p; | |
753 | ||
754 | for (p = cfqd->cur_prio; p <= cfqd->cur_end_prio; p++) { | |
755 | if (!list_empty(&cfqd->rr_list[p])) { | |
756 | prio = p; | |
757 | break; | |
758 | } | |
759 | } | |
760 | ||
761 | if (prio != -1) | |
762 | break; | |
763 | cfqd->cur_prio = 0; | |
764 | if (++cfqd->cur_end_prio == CFQ_PRIO_LISTS) { | |
765 | cfqd->cur_end_prio = 0; | |
766 | if (wrap) | |
767 | break; | |
768 | wrap = 1; | |
1da177e4 | 769 | } |
22e2c507 JA |
770 | } while (1); |
771 | ||
772 | if (unlikely(prio == -1)) | |
773 | return -1; | |
774 | ||
775 | BUG_ON(prio >= CFQ_PRIO_LISTS); | |
776 | ||
777 | list_splice_init(&cfqd->rr_list[prio], &cfqd->cur_rr); | |
778 | ||
779 | cfqd->cur_prio = prio + 1; | |
780 | if (cfqd->cur_prio > cfqd->cur_end_prio) { | |
781 | cfqd->cur_end_prio = cfqd->cur_prio; | |
782 | cfqd->cur_prio = 0; | |
783 | } | |
784 | if (cfqd->cur_end_prio == CFQ_PRIO_LISTS) { | |
785 | cfqd->cur_prio = 0; | |
786 | cfqd->cur_end_prio = 0; | |
1da177e4 LT |
787 | } |
788 | ||
22e2c507 JA |
789 | return prio; |
790 | } | |
791 | ||
3b18152c | 792 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
22e2c507 | 793 | { |
7b14e3b5 | 794 | struct cfq_queue *cfqq = NULL; |
22e2c507 | 795 | |
89850f7e JA |
796 | if (!list_empty(&cfqd->cur_rr) || cfq_get_next_prio_level(cfqd) != -1) { |
797 | /* | |
798 | * if current list is non-empty, grab first entry. if it is | |
799 | * empty, get next prio level and grab first entry then if any | |
800 | * are spliced | |
801 | */ | |
22e2c507 | 802 | cfqq = list_entry_cfqq(cfqd->cur_rr.next); |
89850f7e JA |
803 | } else if (!list_empty(&cfqd->busy_rr)) { |
804 | /* | |
805 | * If no new queues are available, check if the busy list has | |
806 | * some before falling back to idle io. | |
807 | */ | |
e0de0206 | 808 | cfqq = list_entry_cfqq(cfqd->busy_rr.next); |
89850f7e JA |
809 | } else if (!list_empty(&cfqd->idle_rr)) { |
810 | /* | |
811 | * if we have idle queues and no rt or be queues had pending | |
812 | * requests, either allow immediate service if the grace period | |
813 | * has passed or arm the idle grace timer | |
814 | */ | |
22e2c507 JA |
815 | unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE; |
816 | ||
817 | if (time_after_eq(jiffies, end)) | |
818 | cfqq = list_entry_cfqq(cfqd->idle_rr.next); | |
819 | else | |
820 | mod_timer(&cfqd->idle_class_timer, end); | |
821 | } | |
822 | ||
823 | __cfq_set_active_queue(cfqd, cfqq); | |
3b18152c | 824 | return cfqq; |
22e2c507 JA |
825 | } |
826 | ||
caaa5f9f JA |
827 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (128 * 1024)) |
828 | ||
1792669c | 829 | static int cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 830 | { |
1792669c | 831 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 832 | struct cfq_io_context *cic; |
7b14e3b5 JA |
833 | unsigned long sl; |
834 | ||
dd67d051 | 835 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
836 | |
837 | /* | |
838 | * idle is disabled, either manually or by past process history | |
839 | */ | |
840 | if (!cfqd->cfq_slice_idle) | |
841 | return 0; | |
3b18152c | 842 | if (!cfq_cfqq_idle_window(cfqq)) |
22e2c507 JA |
843 | return 0; |
844 | /* | |
845 | * task has exited, don't wait | |
846 | */ | |
206dc69b JA |
847 | cic = cfqd->active_cic; |
848 | if (!cic || !cic->ioc->task) | |
22e2c507 JA |
849 | return 0; |
850 | ||
3b18152c JA |
851 | cfq_mark_cfqq_must_dispatch(cfqq); |
852 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 853 | |
7b14e3b5 | 854 | sl = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle); |
206dc69b JA |
855 | |
856 | /* | |
857 | * we don't want to idle for seeks, but we do want to allow | |
858 | * fair distribution of slice time for a process doing back-to-back | |
859 | * seeks. so allow a little bit of time for him to submit a new rq | |
860 | */ | |
caaa5f9f | 861 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
44eb1231 | 862 | sl = min(sl, msecs_to_jiffies(2)); |
206dc69b | 863 | |
7b14e3b5 | 864 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
22e2c507 | 865 | return 1; |
1da177e4 LT |
866 | } |
867 | ||
5e705374 | 868 | static void cfq_dispatch_insert(request_queue_t *q, struct request *rq) |
1da177e4 | 869 | { |
5e705374 | 870 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 871 | |
5380a101 JA |
872 | cfq_remove_request(rq); |
873 | cfqq->on_dispatch[rq_is_sync(rq)]++; | |
874 | elv_dispatch_sort(q, rq); | |
1da177e4 LT |
875 | } |
876 | ||
877 | /* | |
878 | * return expired entry, or NULL to just start from scratch in rbtree | |
879 | */ | |
5e705374 | 880 | static inline struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
881 | { |
882 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 883 | struct request *rq; |
89850f7e | 884 | int fifo; |
1da177e4 | 885 | |
3b18152c | 886 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 887 | return NULL; |
cb887411 JA |
888 | |
889 | cfq_mark_cfqq_fifo_expire(cfqq); | |
890 | ||
89850f7e JA |
891 | if (list_empty(&cfqq->fifo)) |
892 | return NULL; | |
1da177e4 | 893 | |
89850f7e JA |
894 | fifo = cfq_cfqq_class_sync(cfqq); |
895 | rq = rq_entry_fifo(cfqq->fifo.next); | |
1da177e4 | 896 | |
cb887411 | 897 | if (time_after(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) |
89850f7e | 898 | return rq; |
1da177e4 LT |
899 | |
900 | return NULL; | |
901 | } | |
902 | ||
22e2c507 JA |
903 | static inline int |
904 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
905 | { | |
906 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 907 | |
22e2c507 | 908 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 909 | |
22e2c507 | 910 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
911 | } |
912 | ||
22e2c507 JA |
913 | /* |
914 | * get next queue for service | |
915 | */ | |
1b5ed5e1 | 916 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 917 | { |
1da177e4 | 918 | struct cfq_queue *cfqq; |
1da177e4 | 919 | |
22e2c507 JA |
920 | cfqq = cfqd->active_queue; |
921 | if (!cfqq) | |
922 | goto new_queue; | |
1da177e4 | 923 | |
22e2c507 JA |
924 | /* |
925 | * slice has expired | |
926 | */ | |
44f7c160 | 927 | if (!cfq_cfqq_must_dispatch(cfqq) && cfq_slice_used(cfqq)) |
3b18152c | 928 | goto expire; |
1da177e4 | 929 | |
22e2c507 JA |
930 | /* |
931 | * if queue has requests, dispatch one. if not, check if | |
932 | * enough slice is left to wait for one | |
933 | */ | |
dd67d051 | 934 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 935 | goto keep_queue; |
44f7c160 | 936 | else if (cfq_cfqq_slice_new(cfqq) || cfq_cfqq_dispatched(cfqq)) { |
caaa5f9f JA |
937 | cfqq = NULL; |
938 | goto keep_queue; | |
939 | } else if (cfq_cfqq_class_sync(cfqq)) { | |
1792669c | 940 | if (cfq_arm_slice_timer(cfqd)) |
22e2c507 JA |
941 | return NULL; |
942 | } | |
943 | ||
3b18152c | 944 | expire: |
3c6bd2f8 | 945 | cfq_slice_expired(cfqd, 0, 0); |
3b18152c JA |
946 | new_queue: |
947 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 948 | keep_queue: |
3b18152c | 949 | return cfqq; |
22e2c507 JA |
950 | } |
951 | ||
952 | static int | |
953 | __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
954 | int max_dispatch) | |
955 | { | |
956 | int dispatched = 0; | |
957 | ||
dd67d051 | 958 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
959 | |
960 | do { | |
5e705374 | 961 | struct request *rq; |
1da177e4 LT |
962 | |
963 | /* | |
22e2c507 | 964 | * follow expired path, else get first next available |
1da177e4 | 965 | */ |
5e705374 JA |
966 | if ((rq = cfq_check_fifo(cfqq)) == NULL) |
967 | rq = cfqq->next_rq; | |
22e2c507 JA |
968 | |
969 | /* | |
970 | * finally, insert request into driver dispatch list | |
971 | */ | |
5e705374 | 972 | cfq_dispatch_insert(cfqd->queue, rq); |
1da177e4 | 973 | |
22e2c507 JA |
974 | cfqd->dispatch_slice++; |
975 | dispatched++; | |
1da177e4 | 976 | |
22e2c507 | 977 | if (!cfqd->active_cic) { |
5e705374 JA |
978 | atomic_inc(&RQ_CIC(rq)->ioc->refcount); |
979 | cfqd->active_cic = RQ_CIC(rq); | |
22e2c507 | 980 | } |
1da177e4 | 981 | |
dd67d051 | 982 | if (RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 JA |
983 | break; |
984 | ||
985 | } while (dispatched < max_dispatch); | |
986 | ||
22e2c507 JA |
987 | /* |
988 | * expire an async queue immediately if it has used up its slice. idle | |
989 | * queue always expire after 1 dispatch round. | |
990 | */ | |
a9938006 | 991 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
22e2c507 | 992 | cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
a9938006 | 993 | cfq_class_idle(cfqq))) { |
44f7c160 | 994 | cfqq->slice_end = jiffies + 1; |
3c6bd2f8 | 995 | cfq_slice_expired(cfqd, 0, 0); |
44f7c160 | 996 | } |
22e2c507 JA |
997 | |
998 | return dispatched; | |
999 | } | |
1000 | ||
1b5ed5e1 TH |
1001 | static int |
1002 | cfq_forced_dispatch_cfqqs(struct list_head *list) | |
1003 | { | |
1b5ed5e1 | 1004 | struct cfq_queue *cfqq, *next; |
caaa5f9f | 1005 | int dispatched; |
1b5ed5e1 | 1006 | |
caaa5f9f | 1007 | dispatched = 0; |
1b5ed5e1 | 1008 | list_for_each_entry_safe(cfqq, next, list, cfq_list) { |
5e705374 JA |
1009 | while (cfqq->next_rq) { |
1010 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1b5ed5e1 TH |
1011 | dispatched++; |
1012 | } | |
1013 | BUG_ON(!list_empty(&cfqq->fifo)); | |
1014 | } | |
caaa5f9f | 1015 | |
1b5ed5e1 TH |
1016 | return dispatched; |
1017 | } | |
1018 | ||
1019 | static int | |
1020 | cfq_forced_dispatch(struct cfq_data *cfqd) | |
1021 | { | |
1022 | int i, dispatched = 0; | |
1023 | ||
1024 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
1025 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->rr_list[i]); | |
1026 | ||
1027 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->busy_rr); | |
1028 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->cur_rr); | |
1029 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->idle_rr); | |
1030 | ||
3c6bd2f8 | 1031 | cfq_slice_expired(cfqd, 0, 0); |
1b5ed5e1 TH |
1032 | |
1033 | BUG_ON(cfqd->busy_queues); | |
1034 | ||
1035 | return dispatched; | |
1036 | } | |
1037 | ||
22e2c507 | 1038 | static int |
b4878f24 | 1039 | cfq_dispatch_requests(request_queue_t *q, int force) |
22e2c507 JA |
1040 | { |
1041 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
caaa5f9f JA |
1042 | struct cfq_queue *cfqq, *prev_cfqq; |
1043 | int dispatched; | |
22e2c507 JA |
1044 | |
1045 | if (!cfqd->busy_queues) | |
1046 | return 0; | |
1047 | ||
1b5ed5e1 TH |
1048 | if (unlikely(force)) |
1049 | return cfq_forced_dispatch(cfqd); | |
1050 | ||
caaa5f9f JA |
1051 | dispatched = 0; |
1052 | prev_cfqq = NULL; | |
1053 | while ((cfqq = cfq_select_queue(cfqd)) != NULL) { | |
b4878f24 JA |
1054 | int max_dispatch; |
1055 | ||
a9938006 JA |
1056 | if (cfqd->busy_queues > 1) { |
1057 | /* | |
1058 | * Don't repeat dispatch from the previous queue. | |
1059 | */ | |
1060 | if (prev_cfqq == cfqq) | |
1061 | break; | |
caaa5f9f | 1062 | |
a9938006 JA |
1063 | /* |
1064 | * So we have dispatched before in this round, if the | |
1065 | * next queue has idling enabled (must be sync), don't | |
1066 | * allow it service until the previous have continued. | |
1067 | */ | |
1068 | if (cfqd->rq_in_driver && cfq_cfqq_idle_window(cfqq)) | |
1069 | break; | |
1070 | } | |
9ede209e | 1071 | |
3b18152c JA |
1072 | cfq_clear_cfqq_must_dispatch(cfqq); |
1073 | cfq_clear_cfqq_wait_request(cfqq); | |
22e2c507 JA |
1074 | del_timer(&cfqd->idle_slice_timer); |
1075 | ||
1b5ed5e1 TH |
1076 | max_dispatch = cfqd->cfq_quantum; |
1077 | if (cfq_class_idle(cfqq)) | |
1078 | max_dispatch = 1; | |
1da177e4 | 1079 | |
caaa5f9f | 1080 | dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); |
caaa5f9f | 1081 | prev_cfqq = cfqq; |
1da177e4 LT |
1082 | } |
1083 | ||
caaa5f9f | 1084 | return dispatched; |
1da177e4 LT |
1085 | } |
1086 | ||
1da177e4 | 1087 | /* |
5e705374 JA |
1088 | * task holds one reference to the queue, dropped when task exits. each rq |
1089 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1090 | * |
1091 | * queue lock must be held here. | |
1092 | */ | |
1093 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1094 | { | |
22e2c507 JA |
1095 | struct cfq_data *cfqd = cfqq->cfqd; |
1096 | ||
1097 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1098 | |
1099 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1100 | return; | |
1101 | ||
1102 | BUG_ON(rb_first(&cfqq->sort_list)); | |
22e2c507 | 1103 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1104 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1105 | |
28f95cbc | 1106 | if (unlikely(cfqd->active_queue == cfqq)) { |
3c6bd2f8 | 1107 | __cfq_slice_expired(cfqd, cfqq, 0, 0); |
28f95cbc JA |
1108 | cfq_schedule_dispatch(cfqd); |
1109 | } | |
22e2c507 | 1110 | |
1da177e4 LT |
1111 | /* |
1112 | * it's on the empty list and still hashed | |
1113 | */ | |
1114 | list_del(&cfqq->cfq_list); | |
1115 | hlist_del(&cfqq->cfq_hash); | |
1116 | kmem_cache_free(cfq_pool, cfqq); | |
1117 | } | |
1118 | ||
1ea25ecb | 1119 | static struct cfq_queue * |
3b18152c JA |
1120 | __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio, |
1121 | const int hashval) | |
1da177e4 LT |
1122 | { |
1123 | struct hlist_head *hash_list = &cfqd->cfq_hash[hashval]; | |
206dc69b JA |
1124 | struct hlist_node *entry; |
1125 | struct cfq_queue *__cfqq; | |
1da177e4 | 1126 | |
206dc69b | 1127 | hlist_for_each_entry(__cfqq, entry, hash_list, cfq_hash) { |
b0a6916b | 1128 | const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio); |
1da177e4 | 1129 | |
206dc69b | 1130 | if (__cfqq->key == key && (__p == prio || !prio)) |
1da177e4 LT |
1131 | return __cfqq; |
1132 | } | |
1133 | ||
1134 | return NULL; | |
1135 | } | |
1136 | ||
1137 | static struct cfq_queue * | |
3b18152c | 1138 | cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio) |
1da177e4 | 1139 | { |
3b18152c | 1140 | return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT)); |
1da177e4 LT |
1141 | } |
1142 | ||
e2d74ac0 | 1143 | static void cfq_free_io_context(struct io_context *ioc) |
1da177e4 | 1144 | { |
22e2c507 | 1145 | struct cfq_io_context *__cic; |
e2d74ac0 JA |
1146 | struct rb_node *n; |
1147 | int freed = 0; | |
1da177e4 | 1148 | |
e2d74ac0 JA |
1149 | while ((n = rb_first(&ioc->cic_root)) != NULL) { |
1150 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1151 | rb_erase(&__cic->rb_node, &ioc->cic_root); | |
22e2c507 | 1152 | kmem_cache_free(cfq_ioc_pool, __cic); |
334e94de | 1153 | freed++; |
1da177e4 LT |
1154 | } |
1155 | ||
4050cf16 JA |
1156 | elv_ioc_count_mod(ioc_count, -freed); |
1157 | ||
1158 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) | |
334e94de | 1159 | complete(ioc_gone); |
1da177e4 LT |
1160 | } |
1161 | ||
89850f7e | 1162 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1163 | { |
28f95cbc | 1164 | if (unlikely(cfqq == cfqd->active_queue)) { |
3c6bd2f8 | 1165 | __cfq_slice_expired(cfqd, cfqq, 0, 0); |
28f95cbc JA |
1166 | cfq_schedule_dispatch(cfqd); |
1167 | } | |
22e2c507 | 1168 | |
89850f7e JA |
1169 | cfq_put_queue(cfqq); |
1170 | } | |
22e2c507 | 1171 | |
89850f7e JA |
1172 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1173 | struct cfq_io_context *cic) | |
1174 | { | |
fc46379d JA |
1175 | list_del_init(&cic->queue_list); |
1176 | smp_wmb(); | |
1177 | cic->key = NULL; | |
1178 | ||
12a05732 | 1179 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1180 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1181 | cic->cfqq[ASYNC] = NULL; |
1182 | } | |
1183 | ||
1184 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1185 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1186 | cic->cfqq[SYNC] = NULL; |
1187 | } | |
89850f7e JA |
1188 | } |
1189 | ||
1190 | ||
1191 | /* | |
1192 | * Called with interrupts disabled | |
1193 | */ | |
1194 | static void cfq_exit_single_io_context(struct cfq_io_context *cic) | |
1195 | { | |
1196 | struct cfq_data *cfqd = cic->key; | |
1197 | ||
89850f7e JA |
1198 | if (cfqd) { |
1199 | request_queue_t *q = cfqd->queue; | |
1200 | ||
fc46379d | 1201 | spin_lock_irq(q->queue_lock); |
89850f7e | 1202 | __cfq_exit_single_io_context(cfqd, cic); |
fc46379d | 1203 | spin_unlock_irq(q->queue_lock); |
89850f7e | 1204 | } |
1da177e4 LT |
1205 | } |
1206 | ||
e2d74ac0 | 1207 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1208 | { |
22e2c507 | 1209 | struct cfq_io_context *__cic; |
e2d74ac0 | 1210 | struct rb_node *n; |
22e2c507 | 1211 | |
1da177e4 LT |
1212 | /* |
1213 | * put the reference this task is holding to the various queues | |
1214 | */ | |
e2d74ac0 JA |
1215 | |
1216 | n = rb_first(&ioc->cic_root); | |
1217 | while (n != NULL) { | |
1218 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1219 | ||
22e2c507 | 1220 | cfq_exit_single_io_context(__cic); |
e2d74ac0 | 1221 | n = rb_next(n); |
1da177e4 | 1222 | } |
1da177e4 LT |
1223 | } |
1224 | ||
22e2c507 | 1225 | static struct cfq_io_context * |
8267e268 | 1226 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1227 | { |
b5deef90 | 1228 | struct cfq_io_context *cic; |
1da177e4 | 1229 | |
b5deef90 | 1230 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask, cfqd->queue->node); |
1da177e4 | 1231 | if (cic) { |
553698f9 | 1232 | memset(cic, 0, sizeof(*cic)); |
22e2c507 | 1233 | cic->last_end_request = jiffies; |
553698f9 | 1234 | INIT_LIST_HEAD(&cic->queue_list); |
22e2c507 JA |
1235 | cic->dtor = cfq_free_io_context; |
1236 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1237 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1238 | } |
1239 | ||
1240 | return cic; | |
1241 | } | |
1242 | ||
22e2c507 JA |
1243 | static void cfq_init_prio_data(struct cfq_queue *cfqq) |
1244 | { | |
1245 | struct task_struct *tsk = current; | |
1246 | int ioprio_class; | |
1247 | ||
3b18152c | 1248 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1249 | return; |
1250 | ||
1251 | ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio); | |
1252 | switch (ioprio_class) { | |
1253 | default: | |
1254 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1255 | case IOPRIO_CLASS_NONE: | |
1256 | /* | |
1257 | * no prio set, place us in the middle of the BE classes | |
1258 | */ | |
1259 | cfqq->ioprio = task_nice_ioprio(tsk); | |
1260 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1261 | break; | |
1262 | case IOPRIO_CLASS_RT: | |
1263 | cfqq->ioprio = task_ioprio(tsk); | |
1264 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1265 | break; | |
1266 | case IOPRIO_CLASS_BE: | |
1267 | cfqq->ioprio = task_ioprio(tsk); | |
1268 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1269 | break; | |
1270 | case IOPRIO_CLASS_IDLE: | |
1271 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1272 | cfqq->ioprio = 7; | |
3b18152c | 1273 | cfq_clear_cfqq_idle_window(cfqq); |
22e2c507 JA |
1274 | break; |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * keep track of original prio settings in case we have to temporarily | |
1279 | * elevate the priority of this queue | |
1280 | */ | |
1281 | cfqq->org_ioprio = cfqq->ioprio; | |
1282 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
1283 | ||
98e41c7d | 1284 | cfq_resort_rr_list(cfqq, 0); |
3b18152c | 1285 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1286 | } |
1287 | ||
478a82b0 | 1288 | static inline void changed_ioprio(struct cfq_io_context *cic) |
22e2c507 | 1289 | { |
478a82b0 AV |
1290 | struct cfq_data *cfqd = cic->key; |
1291 | struct cfq_queue *cfqq; | |
c1b707d2 | 1292 | unsigned long flags; |
35e6077c | 1293 | |
caaa5f9f JA |
1294 | if (unlikely(!cfqd)) |
1295 | return; | |
1296 | ||
c1b707d2 | 1297 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1298 | |
1299 | cfqq = cic->cfqq[ASYNC]; | |
1300 | if (cfqq) { | |
1301 | struct cfq_queue *new_cfqq; | |
1302 | new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC, cic->ioc->task, | |
1303 | GFP_ATOMIC); | |
1304 | if (new_cfqq) { | |
1305 | cic->cfqq[ASYNC] = new_cfqq; | |
1306 | cfq_put_queue(cfqq); | |
1307 | } | |
22e2c507 | 1308 | } |
caaa5f9f JA |
1309 | |
1310 | cfqq = cic->cfqq[SYNC]; | |
1311 | if (cfqq) | |
1312 | cfq_mark_cfqq_prio_changed(cfqq); | |
1313 | ||
c1b707d2 | 1314 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1315 | } |
1316 | ||
fc46379d | 1317 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1318 | { |
a6a0763a | 1319 | struct cfq_io_context *cic; |
e2d74ac0 | 1320 | struct rb_node *n; |
a6a0763a | 1321 | |
fc46379d | 1322 | ioc->ioprio_changed = 0; |
a6a0763a | 1323 | |
e2d74ac0 JA |
1324 | n = rb_first(&ioc->cic_root); |
1325 | while (n != NULL) { | |
1326 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
3793c65c | 1327 | |
478a82b0 | 1328 | changed_ioprio(cic); |
e2d74ac0 JA |
1329 | n = rb_next(n); |
1330 | } | |
22e2c507 JA |
1331 | } |
1332 | ||
1333 | static struct cfq_queue * | |
6f325a13 | 1334 | cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, |
8267e268 | 1335 | gfp_t gfp_mask) |
22e2c507 JA |
1336 | { |
1337 | const int hashval = hash_long(key, CFQ_QHASH_SHIFT); | |
1338 | struct cfq_queue *cfqq, *new_cfqq = NULL; | |
6f325a13 | 1339 | unsigned short ioprio; |
22e2c507 JA |
1340 | |
1341 | retry: | |
6f325a13 | 1342 | ioprio = tsk->ioprio; |
3b18152c | 1343 | cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval); |
22e2c507 JA |
1344 | |
1345 | if (!cfqq) { | |
1346 | if (new_cfqq) { | |
1347 | cfqq = new_cfqq; | |
1348 | new_cfqq = NULL; | |
1349 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1350 | /* |
1351 | * Inform the allocator of the fact that we will | |
1352 | * just repeat this allocation if it fails, to allow | |
1353 | * the allocator to do whatever it needs to attempt to | |
1354 | * free memory. | |
1355 | */ | |
22e2c507 | 1356 | spin_unlock_irq(cfqd->queue->queue_lock); |
b5deef90 | 1357 | new_cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask|__GFP_NOFAIL, cfqd->queue->node); |
22e2c507 JA |
1358 | spin_lock_irq(cfqd->queue->queue_lock); |
1359 | goto retry; | |
1360 | } else { | |
b5deef90 | 1361 | cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask, cfqd->queue->node); |
22e2c507 JA |
1362 | if (!cfqq) |
1363 | goto out; | |
1364 | } | |
1365 | ||
1366 | memset(cfqq, 0, sizeof(*cfqq)); | |
1367 | ||
1368 | INIT_HLIST_NODE(&cfqq->cfq_hash); | |
1369 | INIT_LIST_HEAD(&cfqq->cfq_list); | |
22e2c507 JA |
1370 | INIT_LIST_HEAD(&cfqq->fifo); |
1371 | ||
1372 | cfqq->key = key; | |
1373 | hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]); | |
1374 | atomic_set(&cfqq->ref, 0); | |
1375 | cfqq->cfqd = cfqd; | |
c5b680f3 | 1376 | |
a9938006 JA |
1377 | if (key != CFQ_KEY_ASYNC) |
1378 | cfq_mark_cfqq_idle_window(cfqq); | |
1379 | ||
3b18152c | 1380 | cfq_mark_cfqq_prio_changed(cfqq); |
53b03744 | 1381 | cfq_mark_cfqq_queue_new(cfqq); |
3b18152c | 1382 | cfq_init_prio_data(cfqq); |
22e2c507 JA |
1383 | } |
1384 | ||
1385 | if (new_cfqq) | |
1386 | kmem_cache_free(cfq_pool, new_cfqq); | |
1387 | ||
1388 | atomic_inc(&cfqq->ref); | |
1389 | out: | |
1390 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1391 | return cfqq; | |
1392 | } | |
1393 | ||
dbecf3ab OH |
1394 | static void |
1395 | cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic) | |
1396 | { | |
fc46379d | 1397 | WARN_ON(!list_empty(&cic->queue_list)); |
dbecf3ab | 1398 | rb_erase(&cic->rb_node, &ioc->cic_root); |
dbecf3ab | 1399 | kmem_cache_free(cfq_ioc_pool, cic); |
4050cf16 | 1400 | elv_ioc_count_dec(ioc_count); |
dbecf3ab OH |
1401 | } |
1402 | ||
e2d74ac0 JA |
1403 | static struct cfq_io_context * |
1404 | cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc) | |
1405 | { | |
dbecf3ab | 1406 | struct rb_node *n; |
e2d74ac0 | 1407 | struct cfq_io_context *cic; |
be3b0753 | 1408 | void *k, *key = cfqd; |
e2d74ac0 | 1409 | |
dbecf3ab OH |
1410 | restart: |
1411 | n = ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1412 | while (n) { |
1413 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1414 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1415 | k = cic->key; | |
1416 | if (unlikely(!k)) { | |
dbecf3ab OH |
1417 | cfq_drop_dead_cic(ioc, cic); |
1418 | goto restart; | |
1419 | } | |
e2d74ac0 | 1420 | |
be3b0753 | 1421 | if (key < k) |
e2d74ac0 | 1422 | n = n->rb_left; |
be3b0753 | 1423 | else if (key > k) |
e2d74ac0 JA |
1424 | n = n->rb_right; |
1425 | else | |
1426 | return cic; | |
1427 | } | |
1428 | ||
1429 | return NULL; | |
1430 | } | |
1431 | ||
1432 | static inline void | |
1433 | cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, | |
1434 | struct cfq_io_context *cic) | |
1435 | { | |
dbecf3ab OH |
1436 | struct rb_node **p; |
1437 | struct rb_node *parent; | |
e2d74ac0 | 1438 | struct cfq_io_context *__cic; |
0261d688 | 1439 | unsigned long flags; |
be3b0753 | 1440 | void *k; |
e2d74ac0 | 1441 | |
e2d74ac0 JA |
1442 | cic->ioc = ioc; |
1443 | cic->key = cfqd; | |
1444 | ||
dbecf3ab OH |
1445 | restart: |
1446 | parent = NULL; | |
1447 | p = &ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1448 | while (*p) { |
1449 | parent = *p; | |
1450 | __cic = rb_entry(parent, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1451 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1452 | k = __cic->key; | |
1453 | if (unlikely(!k)) { | |
be33c3a6 | 1454 | cfq_drop_dead_cic(ioc, __cic); |
dbecf3ab OH |
1455 | goto restart; |
1456 | } | |
e2d74ac0 | 1457 | |
be3b0753 | 1458 | if (cic->key < k) |
e2d74ac0 | 1459 | p = &(*p)->rb_left; |
be3b0753 | 1460 | else if (cic->key > k) |
e2d74ac0 JA |
1461 | p = &(*p)->rb_right; |
1462 | else | |
1463 | BUG(); | |
1464 | } | |
1465 | ||
1466 | rb_link_node(&cic->rb_node, parent, p); | |
1467 | rb_insert_color(&cic->rb_node, &ioc->cic_root); | |
fc46379d | 1468 | |
0261d688 | 1469 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
e2d74ac0 | 1470 | list_add(&cic->queue_list, &cfqd->cic_list); |
0261d688 | 1471 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
e2d74ac0 JA |
1472 | } |
1473 | ||
1da177e4 LT |
1474 | /* |
1475 | * Setup general io context and cfq io context. There can be several cfq | |
1476 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1477 | * than one device managed by cfq. |
1da177e4 LT |
1478 | */ |
1479 | static struct cfq_io_context * | |
e2d74ac0 | 1480 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1481 | { |
22e2c507 | 1482 | struct io_context *ioc = NULL; |
1da177e4 | 1483 | struct cfq_io_context *cic; |
1da177e4 | 1484 | |
22e2c507 | 1485 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1486 | |
b5deef90 | 1487 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1488 | if (!ioc) |
1489 | return NULL; | |
1490 | ||
e2d74ac0 JA |
1491 | cic = cfq_cic_rb_lookup(cfqd, ioc); |
1492 | if (cic) | |
1493 | goto out; | |
1da177e4 | 1494 | |
e2d74ac0 JA |
1495 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1496 | if (cic == NULL) | |
1497 | goto err; | |
1da177e4 | 1498 | |
e2d74ac0 | 1499 | cfq_cic_link(cfqd, ioc, cic); |
1da177e4 | 1500 | out: |
fc46379d JA |
1501 | smp_read_barrier_depends(); |
1502 | if (unlikely(ioc->ioprio_changed)) | |
1503 | cfq_ioc_set_ioprio(ioc); | |
1504 | ||
1da177e4 LT |
1505 | return cic; |
1506 | err: | |
1507 | put_io_context(ioc); | |
1508 | return NULL; | |
1509 | } | |
1510 | ||
22e2c507 JA |
1511 | static void |
1512 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1513 | { |
aaf1228d JA |
1514 | unsigned long elapsed = jiffies - cic->last_end_request; |
1515 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 1516 | |
22e2c507 JA |
1517 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1518 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1519 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1520 | } | |
1da177e4 | 1521 | |
206dc69b | 1522 | static void |
bb37b94c | 1523 | cfq_update_io_seektime(struct cfq_io_context *cic, struct request *rq) |
206dc69b JA |
1524 | { |
1525 | sector_t sdist; | |
1526 | u64 total; | |
1527 | ||
5e705374 JA |
1528 | if (cic->last_request_pos < rq->sector) |
1529 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1530 | else |
5e705374 | 1531 | sdist = cic->last_request_pos - rq->sector; |
206dc69b JA |
1532 | |
1533 | /* | |
1534 | * Don't allow the seek distance to get too large from the | |
1535 | * odd fragment, pagein, etc | |
1536 | */ | |
1537 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1538 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1539 | else | |
1540 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1541 | ||
1542 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1543 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1544 | total = cic->seek_total + (cic->seek_samples/2); | |
1545 | do_div(total, cic->seek_samples); | |
1546 | cic->seek_mean = (sector_t)total; | |
1547 | } | |
1da177e4 | 1548 | |
22e2c507 JA |
1549 | /* |
1550 | * Disable idle window if the process thinks too long or seeks so much that | |
1551 | * it doesn't matter | |
1552 | */ | |
1553 | static void | |
1554 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1555 | struct cfq_io_context *cic) | |
1556 | { | |
3b18152c | 1557 | int enable_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 1558 | |
caaa5f9f JA |
1559 | if (!cic->ioc->task || !cfqd->cfq_slice_idle || |
1560 | (cfqd->hw_tag && CIC_SEEKY(cic))) | |
22e2c507 JA |
1561 | enable_idle = 0; |
1562 | else if (sample_valid(cic->ttime_samples)) { | |
1563 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1564 | enable_idle = 0; | |
1565 | else | |
1566 | enable_idle = 1; | |
1da177e4 LT |
1567 | } |
1568 | ||
3b18152c JA |
1569 | if (enable_idle) |
1570 | cfq_mark_cfqq_idle_window(cfqq); | |
1571 | else | |
1572 | cfq_clear_cfqq_idle_window(cfqq); | |
22e2c507 | 1573 | } |
1da177e4 | 1574 | |
22e2c507 JA |
1575 | /* |
1576 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1577 | * no or if we aren't sure, a 1 will cause a preempt. | |
1578 | */ | |
1579 | static int | |
1580 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1581 | struct request *rq) |
22e2c507 JA |
1582 | { |
1583 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1e3335de | 1584 | sector_t dist; |
22e2c507 JA |
1585 | |
1586 | if (cfq_class_idle(new_cfqq)) | |
1587 | return 0; | |
1588 | ||
1589 | if (!cfqq) | |
caaa5f9f | 1590 | return 0; |
22e2c507 JA |
1591 | |
1592 | if (cfq_class_idle(cfqq)) | |
1593 | return 1; | |
1e3335de | 1594 | |
374f84ac JA |
1595 | /* |
1596 | * if the new request is sync, but the currently running queue is | |
1597 | * not, let the sync request have priority. | |
1598 | */ | |
5e705374 | 1599 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1600 | return 1; |
1e3335de | 1601 | |
374f84ac JA |
1602 | /* |
1603 | * So both queues are sync. Let the new request get disk time if | |
1604 | * it's a metadata request and the current queue is doing regular IO. | |
1605 | */ | |
1606 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1607 | return 1; | |
22e2c507 | 1608 | |
1e3335de JA |
1609 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
1610 | return 0; | |
1611 | ||
1612 | /* | |
1613 | * if this request is as-good as one we would expect from the | |
1614 | * current cfqq, let it preempt | |
1615 | */ | |
1616 | if (rq->sector > cfqd->last_sector) | |
1617 | dist = rq->sector - cfqd->last_sector; | |
1618 | else | |
1619 | dist = cfqd->last_sector - rq->sector; | |
1620 | ||
1621 | if (dist <= cfqd->active_cic->seek_mean) | |
1622 | return 1; | |
1623 | ||
22e2c507 JA |
1624 | return 0; |
1625 | } | |
1626 | ||
1627 | /* | |
1628 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1629 | * let it have half of its nominal slice. | |
1630 | */ | |
1631 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1632 | { | |
3c6bd2f8 | 1633 | cfq_slice_expired(cfqd, 1, 1); |
22e2c507 | 1634 | |
bf572256 JA |
1635 | /* |
1636 | * Put the new queue at the front of the of the current list, | |
1637 | * so we know that it will be selected next. | |
1638 | */ | |
1639 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
1640 | list_move(&cfqq->cfq_list, &cfqd->cur_rr); | |
1641 | ||
44f7c160 JA |
1642 | cfqq->slice_end = 0; |
1643 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
1644 | } |
1645 | ||
22e2c507 | 1646 | /* |
5e705374 | 1647 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1648 | * something we should do about it |
1649 | */ | |
1650 | static void | |
5e705374 JA |
1651 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1652 | struct request *rq) | |
22e2c507 | 1653 | { |
5e705374 | 1654 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1655 | |
374f84ac JA |
1656 | if (rq_is_meta(rq)) |
1657 | cfqq->meta_pending++; | |
1658 | ||
9c2c38a1 JA |
1659 | /* |
1660 | * we never wait for an async request and we don't allow preemption | |
1661 | * of an async request. so just return early | |
1662 | */ | |
5e705374 | 1663 | if (!rq_is_sync(rq)) { |
12e9fddd JA |
1664 | /* |
1665 | * sync process issued an async request, if it's waiting | |
1666 | * then expire it and kick rq handling. | |
1667 | */ | |
1668 | if (cic == cfqd->active_cic && | |
1669 | del_timer(&cfqd->idle_slice_timer)) { | |
3c6bd2f8 | 1670 | cfq_slice_expired(cfqd, 0, 0); |
dc72ef4a | 1671 | blk_start_queueing(cfqd->queue); |
12e9fddd | 1672 | } |
9c2c38a1 | 1673 | return; |
12e9fddd | 1674 | } |
22e2c507 | 1675 | |
9c2c38a1 | 1676 | cfq_update_io_thinktime(cfqd, cic); |
bb37b94c | 1677 | cfq_update_io_seektime(cic, rq); |
9c2c38a1 JA |
1678 | cfq_update_idle_window(cfqd, cfqq, cic); |
1679 | ||
5e705374 | 1680 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
22e2c507 JA |
1681 | |
1682 | if (cfqq == cfqd->active_queue) { | |
1683 | /* | |
1684 | * if we are waiting for a request for this queue, let it rip | |
1685 | * immediately and flag that we must not expire this queue | |
1686 | * just now | |
1687 | */ | |
3b18152c JA |
1688 | if (cfq_cfqq_wait_request(cfqq)) { |
1689 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 | 1690 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1691 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1692 | } |
5e705374 | 1693 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1694 | /* |
1695 | * not the active queue - expire current slice if it is | |
1696 | * idle and has expired it's mean thinktime or this new queue | |
1697 | * has some old slice time left and is of higher priority | |
1698 | */ | |
1699 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1700 | cfq_mark_cfqq_must_dispatch(cfqq); |
dc72ef4a | 1701 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1702 | } |
1da177e4 LT |
1703 | } |
1704 | ||
b4878f24 | 1705 | static void cfq_insert_request(request_queue_t *q, struct request *rq) |
1da177e4 | 1706 | { |
b4878f24 | 1707 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1708 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1709 | |
1710 | cfq_init_prio_data(cfqq); | |
1da177e4 | 1711 | |
5e705374 | 1712 | cfq_add_rq_rb(rq); |
1da177e4 | 1713 | |
22e2c507 JA |
1714 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1715 | ||
5e705374 | 1716 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1717 | } |
1718 | ||
1da177e4 LT |
1719 | static void cfq_completed_request(request_queue_t *q, struct request *rq) |
1720 | { | |
5e705374 | 1721 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1722 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1723 | const int sync = rq_is_sync(rq); |
b4878f24 | 1724 | unsigned long now; |
1da177e4 | 1725 | |
b4878f24 | 1726 | now = jiffies; |
1da177e4 | 1727 | |
b4878f24 JA |
1728 | WARN_ON(!cfqd->rq_in_driver); |
1729 | WARN_ON(!cfqq->on_dispatch[sync]); | |
1730 | cfqd->rq_in_driver--; | |
1731 | cfqq->on_dispatch[sync]--; | |
99f9628a | 1732 | cfqq->service_last = now; |
1da177e4 | 1733 | |
1e3335de JA |
1734 | cfqd->last_sector = rq->hard_sector + rq->hard_nr_sectors; |
1735 | ||
b4878f24 JA |
1736 | if (!cfq_class_idle(cfqq)) |
1737 | cfqd->last_end_request = now; | |
3b18152c | 1738 | |
98e41c7d | 1739 | cfq_resort_rr_list(cfqq, 0); |
1da177e4 | 1740 | |
caaa5f9f | 1741 | if (sync) |
5e705374 | 1742 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1743 | |
1744 | /* | |
1745 | * If this is the active queue, check if it needs to be expired, | |
1746 | * or if we want to idle in case it has no pending requests. | |
1747 | */ | |
1748 | if (cfqd->active_queue == cfqq) { | |
44f7c160 JA |
1749 | if (cfq_cfqq_slice_new(cfqq)) { |
1750 | cfq_set_prio_slice(cfqd, cfqq); | |
1751 | cfq_clear_cfqq_slice_new(cfqq); | |
1752 | } | |
1753 | if (cfq_slice_used(cfqq)) | |
3c6bd2f8 | 1754 | cfq_slice_expired(cfqd, 0, 1); |
dd67d051 | 1755 | else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1792669c | 1756 | if (!cfq_arm_slice_timer(cfqd)) |
caaa5f9f JA |
1757 | cfq_schedule_dispatch(cfqd); |
1758 | } | |
1759 | } | |
1da177e4 LT |
1760 | } |
1761 | ||
22e2c507 JA |
1762 | /* |
1763 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
1764 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
1765 | */ | |
1766 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 1767 | { |
22e2c507 JA |
1768 | const int ioprio_class = cfqq->ioprio_class; |
1769 | const int ioprio = cfqq->ioprio; | |
1da177e4 | 1770 | |
22e2c507 JA |
1771 | if (has_fs_excl()) { |
1772 | /* | |
1773 | * boost idle prio on transactions that would lock out other | |
1774 | * users of the filesystem | |
1775 | */ | |
1776 | if (cfq_class_idle(cfqq)) | |
1777 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1778 | if (cfqq->ioprio > IOPRIO_NORM) | |
1779 | cfqq->ioprio = IOPRIO_NORM; | |
1780 | } else { | |
1781 | /* | |
1782 | * check if we need to unboost the queue | |
1783 | */ | |
1784 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
1785 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
1786 | if (cfqq->ioprio != cfqq->org_ioprio) | |
1787 | cfqq->ioprio = cfqq->org_ioprio; | |
1788 | } | |
1da177e4 | 1789 | |
22e2c507 JA |
1790 | /* |
1791 | * refile between round-robin lists if we moved the priority class | |
1792 | */ | |
98e41c7d | 1793 | if ((ioprio_class != cfqq->ioprio_class || ioprio != cfqq->ioprio)) |
22e2c507 JA |
1794 | cfq_resort_rr_list(cfqq, 0); |
1795 | } | |
1da177e4 | 1796 | |
89850f7e | 1797 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 1798 | { |
3b18152c | 1799 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 1800 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 1801 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 1802 | return ELV_MQUEUE_MUST; |
3b18152c | 1803 | } |
1da177e4 | 1804 | |
22e2c507 | 1805 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
1806 | } |
1807 | ||
cb78b285 | 1808 | static int cfq_may_queue(request_queue_t *q, int rw) |
22e2c507 JA |
1809 | { |
1810 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1811 | struct task_struct *tsk = current; | |
1812 | struct cfq_queue *cfqq; | |
7749a8d4 JA |
1813 | unsigned int key; |
1814 | ||
1815 | key = cfq_queue_pid(tsk, rw, rw & REQ_RW_SYNC); | |
22e2c507 JA |
1816 | |
1817 | /* | |
1818 | * don't force setup of a queue from here, as a call to may_queue | |
1819 | * does not necessarily imply that a request actually will be queued. | |
1820 | * so just lookup a possibly existing queue, or return 'may queue' | |
1821 | * if that fails | |
1822 | */ | |
7749a8d4 | 1823 | cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio); |
22e2c507 JA |
1824 | if (cfqq) { |
1825 | cfq_init_prio_data(cfqq); | |
1826 | cfq_prio_boost(cfqq); | |
1827 | ||
89850f7e | 1828 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
1829 | } |
1830 | ||
1831 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
1832 | } |
1833 | ||
1da177e4 LT |
1834 | /* |
1835 | * queue lock held here | |
1836 | */ | |
bb37b94c | 1837 | static void cfq_put_request(struct request *rq) |
1da177e4 | 1838 | { |
5e705374 | 1839 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1840 | |
5e705374 | 1841 | if (cfqq) { |
22e2c507 | 1842 | const int rw = rq_data_dir(rq); |
1da177e4 | 1843 | |
22e2c507 JA |
1844 | BUG_ON(!cfqq->allocated[rw]); |
1845 | cfqq->allocated[rw]--; | |
1da177e4 | 1846 | |
5e705374 | 1847 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 1848 | |
1da177e4 | 1849 | rq->elevator_private = NULL; |
5e705374 | 1850 | rq->elevator_private2 = NULL; |
1da177e4 | 1851 | |
1da177e4 LT |
1852 | cfq_put_queue(cfqq); |
1853 | } | |
1854 | } | |
1855 | ||
1856 | /* | |
22e2c507 | 1857 | * Allocate cfq data structures associated with this request. |
1da177e4 | 1858 | */ |
22e2c507 | 1859 | static int |
cb78b285 | 1860 | cfq_set_request(request_queue_t *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
1861 | { |
1862 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3b18152c | 1863 | struct task_struct *tsk = current; |
1da177e4 LT |
1864 | struct cfq_io_context *cic; |
1865 | const int rw = rq_data_dir(rq); | |
7749a8d4 JA |
1866 | const int is_sync = rq_is_sync(rq); |
1867 | pid_t key = cfq_queue_pid(tsk, rw, is_sync); | |
22e2c507 | 1868 | struct cfq_queue *cfqq; |
1da177e4 LT |
1869 | unsigned long flags; |
1870 | ||
1871 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1872 | ||
e2d74ac0 | 1873 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 1874 | |
1da177e4 LT |
1875 | spin_lock_irqsave(q->queue_lock, flags); |
1876 | ||
22e2c507 JA |
1877 | if (!cic) |
1878 | goto queue_fail; | |
1879 | ||
12a05732 | 1880 | if (!cic->cfqq[is_sync]) { |
6f325a13 | 1881 | cfqq = cfq_get_queue(cfqd, key, tsk, gfp_mask); |
22e2c507 JA |
1882 | if (!cfqq) |
1883 | goto queue_fail; | |
1da177e4 | 1884 | |
12a05732 | 1885 | cic->cfqq[is_sync] = cfqq; |
22e2c507 | 1886 | } else |
12a05732 | 1887 | cfqq = cic->cfqq[is_sync]; |
1da177e4 LT |
1888 | |
1889 | cfqq->allocated[rw]++; | |
3b18152c | 1890 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 1891 | atomic_inc(&cfqq->ref); |
1da177e4 | 1892 | |
5e705374 | 1893 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 1894 | |
5e705374 JA |
1895 | rq->elevator_private = cic; |
1896 | rq->elevator_private2 = cfqq; | |
1897 | return 0; | |
1da177e4 | 1898 | |
22e2c507 JA |
1899 | queue_fail: |
1900 | if (cic) | |
1901 | put_io_context(cic->ioc); | |
89850f7e | 1902 | |
3b18152c | 1903 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
1904 | spin_unlock_irqrestore(q->queue_lock, flags); |
1905 | return 1; | |
1906 | } | |
1907 | ||
65f27f38 | 1908 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 1909 | { |
65f27f38 DH |
1910 | struct cfq_data *cfqd = |
1911 | container_of(work, struct cfq_data, unplug_work); | |
1912 | request_queue_t *q = cfqd->queue; | |
22e2c507 JA |
1913 | unsigned long flags; |
1914 | ||
1915 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 1916 | blk_start_queueing(q); |
22e2c507 JA |
1917 | spin_unlock_irqrestore(q->queue_lock, flags); |
1918 | } | |
1919 | ||
1920 | /* | |
1921 | * Timer running if the active_queue is currently idling inside its time slice | |
1922 | */ | |
1923 | static void cfq_idle_slice_timer(unsigned long data) | |
1924 | { | |
1925 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1926 | struct cfq_queue *cfqq; | |
1927 | unsigned long flags; | |
3c6bd2f8 | 1928 | int timed_out = 1; |
22e2c507 JA |
1929 | |
1930 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1931 | ||
1932 | if ((cfqq = cfqd->active_queue) != NULL) { | |
3c6bd2f8 JA |
1933 | timed_out = 0; |
1934 | ||
22e2c507 JA |
1935 | /* |
1936 | * expired | |
1937 | */ | |
44f7c160 | 1938 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
1939 | goto expire; |
1940 | ||
1941 | /* | |
1942 | * only expire and reinvoke request handler, if there are | |
1943 | * other queues with pending requests | |
1944 | */ | |
caaa5f9f | 1945 | if (!cfqd->busy_queues) |
22e2c507 | 1946 | goto out_cont; |
22e2c507 JA |
1947 | |
1948 | /* | |
1949 | * not expired and it has a request pending, let it dispatch | |
1950 | */ | |
dd67d051 | 1951 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 1952 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
1953 | goto out_kick; |
1954 | } | |
1955 | } | |
1956 | expire: | |
3c6bd2f8 | 1957 | cfq_slice_expired(cfqd, 0, timed_out); |
22e2c507 | 1958 | out_kick: |
3b18152c | 1959 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
1960 | out_cont: |
1961 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1962 | } | |
1963 | ||
1964 | /* | |
1965 | * Timer running if an idle class queue is waiting for service | |
1966 | */ | |
1967 | static void cfq_idle_class_timer(unsigned long data) | |
1968 | { | |
1969 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1970 | unsigned long flags, end; | |
1971 | ||
1972 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1973 | ||
1974 | /* | |
1975 | * race with a non-idle queue, reset timer | |
1976 | */ | |
1977 | end = cfqd->last_end_request + CFQ_IDLE_GRACE; | |
ae818a38 JA |
1978 | if (!time_after_eq(jiffies, end)) |
1979 | mod_timer(&cfqd->idle_class_timer, end); | |
1980 | else | |
3b18152c | 1981 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
1982 | |
1983 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1984 | } | |
1985 | ||
3b18152c JA |
1986 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
1987 | { | |
1988 | del_timer_sync(&cfqd->idle_slice_timer); | |
1989 | del_timer_sync(&cfqd->idle_class_timer); | |
1990 | blk_sync_queue(cfqd->queue); | |
1991 | } | |
22e2c507 | 1992 | |
1da177e4 LT |
1993 | static void cfq_exit_queue(elevator_t *e) |
1994 | { | |
22e2c507 | 1995 | struct cfq_data *cfqd = e->elevator_data; |
d9ff4187 | 1996 | request_queue_t *q = cfqd->queue; |
22e2c507 | 1997 | |
3b18152c | 1998 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 1999 | |
d9ff4187 | 2000 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 2001 | |
d9ff4187 | 2002 | if (cfqd->active_queue) |
3c6bd2f8 | 2003 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0, 0); |
e2d74ac0 JA |
2004 | |
2005 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
2006 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
2007 | struct cfq_io_context, | |
2008 | queue_list); | |
89850f7e JA |
2009 | |
2010 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 2011 | } |
e2d74ac0 | 2012 | |
d9ff4187 | 2013 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
2014 | |
2015 | cfq_shutdown_timer_wq(cfqd); | |
2016 | ||
a90d742e AV |
2017 | kfree(cfqd->cfq_hash); |
2018 | kfree(cfqd); | |
1da177e4 LT |
2019 | } |
2020 | ||
bb37b94c | 2021 | static void *cfq_init_queue(request_queue_t *q) |
1da177e4 LT |
2022 | { |
2023 | struct cfq_data *cfqd; | |
2024 | int i; | |
2025 | ||
b5deef90 | 2026 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node); |
1da177e4 | 2027 | if (!cfqd) |
bc1c1169 | 2028 | return NULL; |
1da177e4 LT |
2029 | |
2030 | memset(cfqd, 0, sizeof(*cfqd)); | |
22e2c507 JA |
2031 | |
2032 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
2033 | INIT_LIST_HEAD(&cfqd->rr_list[i]); | |
2034 | ||
2035 | INIT_LIST_HEAD(&cfqd->busy_rr); | |
2036 | INIT_LIST_HEAD(&cfqd->cur_rr); | |
2037 | INIT_LIST_HEAD(&cfqd->idle_rr); | |
d9ff4187 | 2038 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2039 | |
b5deef90 | 2040 | cfqd->cfq_hash = kmalloc_node(sizeof(struct hlist_head) * CFQ_QHASH_ENTRIES, GFP_KERNEL, q->node); |
1da177e4 | 2041 | if (!cfqd->cfq_hash) |
5e705374 | 2042 | goto out_free; |
1da177e4 | 2043 | |
1da177e4 LT |
2044 | for (i = 0; i < CFQ_QHASH_ENTRIES; i++) |
2045 | INIT_HLIST_HEAD(&cfqd->cfq_hash[i]); | |
2046 | ||
1da177e4 | 2047 | cfqd->queue = q; |
1da177e4 | 2048 | |
22e2c507 JA |
2049 | init_timer(&cfqd->idle_slice_timer); |
2050 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2051 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2052 | ||
2053 | init_timer(&cfqd->idle_class_timer); | |
2054 | cfqd->idle_class_timer.function = cfq_idle_class_timer; | |
2055 | cfqd->idle_class_timer.data = (unsigned long) cfqd; | |
2056 | ||
65f27f38 | 2057 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2058 | |
1da177e4 | 2059 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2060 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2061 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2062 | cfqd->cfq_back_max = cfq_back_max; |
2063 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2064 | cfqd->cfq_slice[0] = cfq_slice_async; |
2065 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2066 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2067 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
3b18152c | 2068 | |
bc1c1169 | 2069 | return cfqd; |
5e705374 | 2070 | out_free: |
1da177e4 | 2071 | kfree(cfqd); |
bc1c1169 | 2072 | return NULL; |
1da177e4 LT |
2073 | } |
2074 | ||
2075 | static void cfq_slab_kill(void) | |
2076 | { | |
1da177e4 LT |
2077 | if (cfq_pool) |
2078 | kmem_cache_destroy(cfq_pool); | |
2079 | if (cfq_ioc_pool) | |
2080 | kmem_cache_destroy(cfq_ioc_pool); | |
2081 | } | |
2082 | ||
2083 | static int __init cfq_slab_setup(void) | |
2084 | { | |
1da177e4 LT |
2085 | cfq_pool = kmem_cache_create("cfq_pool", sizeof(struct cfq_queue), 0, 0, |
2086 | NULL, NULL); | |
2087 | if (!cfq_pool) | |
2088 | goto fail; | |
2089 | ||
2090 | cfq_ioc_pool = kmem_cache_create("cfq_ioc_pool", | |
2091 | sizeof(struct cfq_io_context), 0, 0, NULL, NULL); | |
2092 | if (!cfq_ioc_pool) | |
2093 | goto fail; | |
2094 | ||
2095 | return 0; | |
2096 | fail: | |
2097 | cfq_slab_kill(); | |
2098 | return -ENOMEM; | |
2099 | } | |
2100 | ||
1da177e4 LT |
2101 | /* |
2102 | * sysfs parts below --> | |
2103 | */ | |
1da177e4 LT |
2104 | |
2105 | static ssize_t | |
2106 | cfq_var_show(unsigned int var, char *page) | |
2107 | { | |
2108 | return sprintf(page, "%d\n", var); | |
2109 | } | |
2110 | ||
2111 | static ssize_t | |
2112 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2113 | { | |
2114 | char *p = (char *) page; | |
2115 | ||
2116 | *var = simple_strtoul(p, &p, 10); | |
2117 | return count; | |
2118 | } | |
2119 | ||
1da177e4 | 2120 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
3d1ab40f | 2121 | static ssize_t __FUNC(elevator_t *e, char *page) \ |
1da177e4 | 2122 | { \ |
3d1ab40f | 2123 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2124 | unsigned int __data = __VAR; \ |
2125 | if (__CONV) \ | |
2126 | __data = jiffies_to_msecs(__data); \ | |
2127 | return cfq_var_show(__data, (page)); \ | |
2128 | } | |
2129 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2130 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2131 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2132 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2133 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2134 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2135 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2136 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2137 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2138 | #undef SHOW_FUNCTION |
2139 | ||
2140 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
3d1ab40f | 2141 | static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \ |
1da177e4 | 2142 | { \ |
3d1ab40f | 2143 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2144 | unsigned int __data; \ |
2145 | int ret = cfq_var_store(&__data, (page), count); \ | |
2146 | if (__data < (MIN)) \ | |
2147 | __data = (MIN); \ | |
2148 | else if (__data > (MAX)) \ | |
2149 | __data = (MAX); \ | |
2150 | if (__CONV) \ | |
2151 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2152 | else \ | |
2153 | *(__PTR) = __data; \ | |
2154 | return ret; \ | |
2155 | } | |
2156 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
22e2c507 JA |
2157 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1); |
2158 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1); | |
e572ec7e AV |
2159 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
2160 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0); | |
22e2c507 JA |
2161 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2162 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2163 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
2164 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0); | |
1da177e4 LT |
2165 | #undef STORE_FUNCTION |
2166 | ||
e572ec7e AV |
2167 | #define CFQ_ATTR(name) \ |
2168 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2169 | ||
2170 | static struct elv_fs_entry cfq_attrs[] = { | |
2171 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2172 | CFQ_ATTR(fifo_expire_sync), |
2173 | CFQ_ATTR(fifo_expire_async), | |
2174 | CFQ_ATTR(back_seek_max), | |
2175 | CFQ_ATTR(back_seek_penalty), | |
2176 | CFQ_ATTR(slice_sync), | |
2177 | CFQ_ATTR(slice_async), | |
2178 | CFQ_ATTR(slice_async_rq), | |
2179 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2180 | __ATTR_NULL |
1da177e4 LT |
2181 | }; |
2182 | ||
1da177e4 LT |
2183 | static struct elevator_type iosched_cfq = { |
2184 | .ops = { | |
2185 | .elevator_merge_fn = cfq_merge, | |
2186 | .elevator_merged_fn = cfq_merged_request, | |
2187 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2188 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2189 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2190 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2191 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2192 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2193 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2194 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2195 | .elevator_former_req_fn = elv_rb_former_request, |
2196 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2197 | .elevator_set_req_fn = cfq_set_request, |
2198 | .elevator_put_req_fn = cfq_put_request, | |
2199 | .elevator_may_queue_fn = cfq_may_queue, | |
2200 | .elevator_init_fn = cfq_init_queue, | |
2201 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2202 | .trim = cfq_free_io_context, |
1da177e4 | 2203 | }, |
3d1ab40f | 2204 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2205 | .elevator_name = "cfq", |
2206 | .elevator_owner = THIS_MODULE, | |
2207 | }; | |
2208 | ||
2209 | static int __init cfq_init(void) | |
2210 | { | |
2211 | int ret; | |
2212 | ||
22e2c507 JA |
2213 | /* |
2214 | * could be 0 on HZ < 1000 setups | |
2215 | */ | |
2216 | if (!cfq_slice_async) | |
2217 | cfq_slice_async = 1; | |
2218 | if (!cfq_slice_idle) | |
2219 | cfq_slice_idle = 1; | |
2220 | ||
1da177e4 LT |
2221 | if (cfq_slab_setup()) |
2222 | return -ENOMEM; | |
2223 | ||
2224 | ret = elv_register(&iosched_cfq); | |
22e2c507 JA |
2225 | if (ret) |
2226 | cfq_slab_kill(); | |
1da177e4 | 2227 | |
1da177e4 LT |
2228 | return ret; |
2229 | } | |
2230 | ||
2231 | static void __exit cfq_exit(void) | |
2232 | { | |
6e9a4738 | 2233 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2234 | elv_unregister(&iosched_cfq); |
334e94de | 2235 | ioc_gone = &all_gone; |
fba82272 OH |
2236 | /* ioc_gone's update must be visible before reading ioc_count */ |
2237 | smp_wmb(); | |
4050cf16 | 2238 | if (elv_ioc_count_read(ioc_count)) |
fba82272 | 2239 | wait_for_completion(ioc_gone); |
334e94de | 2240 | synchronize_rcu(); |
83521d3e | 2241 | cfq_slab_kill(); |
1da177e4 LT |
2242 | } |
2243 | ||
2244 | module_init(cfq_init); | |
2245 | module_exit(cfq_exit); | |
2246 | ||
2247 | MODULE_AUTHOR("Jens Axboe"); | |
2248 | MODULE_LICENSE("GPL"); | |
2249 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |