]>
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 | 28 | |
d9e7620e JA |
29 | /* |
30 | * grace period before allowing idle class to get disk access | |
31 | */ | |
22e2c507 | 32 | #define CFQ_IDLE_GRACE (HZ / 10) |
d9e7620e JA |
33 | |
34 | /* | |
35 | * below this threshold, we consider thinktime immediate | |
36 | */ | |
37 | #define CFQ_MIN_TT (2) | |
38 | ||
22e2c507 JA |
39 | #define CFQ_SLICE_SCALE (5) |
40 | ||
41 | #define CFQ_KEY_ASYNC (0) | |
22e2c507 | 42 | |
1da177e4 LT |
43 | /* |
44 | * for the hash of cfqq inside the cfqd | |
45 | */ | |
46 | #define CFQ_QHASH_SHIFT 6 | |
47 | #define CFQ_QHASH_ENTRIES (1 << CFQ_QHASH_SHIFT) | |
48 | #define list_entry_qhash(entry) hlist_entry((entry), struct cfq_queue, cfq_hash) | |
49 | ||
1da177e4 LT |
50 | #define list_entry_cfqq(ptr) list_entry((ptr), struct cfq_queue, cfq_list) |
51 | ||
5e705374 JA |
52 | #define RQ_CIC(rq) ((struct cfq_io_context*)(rq)->elevator_private) |
53 | #define RQ_CFQQ(rq) ((rq)->elevator_private2) | |
1da177e4 | 54 | |
e18b890b CL |
55 | static struct kmem_cache *cfq_pool; |
56 | static struct kmem_cache *cfq_ioc_pool; | |
1da177e4 | 57 | |
4050cf16 | 58 | static DEFINE_PER_CPU(unsigned long, ioc_count); |
334e94de AV |
59 | static struct completion *ioc_gone; |
60 | ||
22e2c507 JA |
61 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
62 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
63 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
64 | ||
3b18152c JA |
65 | #define ASYNC (0) |
66 | #define SYNC (1) | |
67 | ||
6d048f53 | 68 | #define cfq_cfqq_sync(cfqq) ((cfqq)->key != CFQ_KEY_ASYNC) |
22e2c507 | 69 | |
206dc69b JA |
70 | #define sample_valid(samples) ((samples) > 80) |
71 | ||
cc09e299 JA |
72 | /* |
73 | * Most of our rbtree usage is for sorting with min extraction, so | |
74 | * if we cache the leftmost node we don't have to walk down the tree | |
75 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
76 | * move this into the elevator for the rq sorting as well. | |
77 | */ | |
78 | struct cfq_rb_root { | |
79 | struct rb_root rb; | |
80 | struct rb_node *left; | |
81 | }; | |
82 | #define CFQ_RB_ROOT (struct cfq_rb_root) { RB_ROOT, NULL, } | |
83 | ||
22e2c507 JA |
84 | /* |
85 | * Per block device queue structure | |
86 | */ | |
1da177e4 | 87 | struct cfq_data { |
22e2c507 JA |
88 | request_queue_t *queue; |
89 | ||
90 | /* | |
91 | * rr list of queues with requests and the count of them | |
92 | */ | |
cc09e299 | 93 | struct cfq_rb_root service_tree; |
22e2c507 | 94 | struct list_head cur_rr; |
22e2c507 JA |
95 | unsigned int busy_queues; |
96 | ||
22e2c507 JA |
97 | /* |
98 | * cfqq lookup hash | |
99 | */ | |
1da177e4 | 100 | struct hlist_head *cfq_hash; |
1da177e4 | 101 | |
22e2c507 | 102 | int rq_in_driver; |
25776e35 | 103 | int hw_tag; |
1da177e4 | 104 | |
22e2c507 JA |
105 | /* |
106 | * idle window management | |
107 | */ | |
108 | struct timer_list idle_slice_timer; | |
109 | struct work_struct unplug_work; | |
1da177e4 | 110 | |
22e2c507 JA |
111 | struct cfq_queue *active_queue; |
112 | struct cfq_io_context *active_cic; | |
22e2c507 JA |
113 | unsigned int dispatch_slice; |
114 | ||
115 | struct timer_list idle_class_timer; | |
1da177e4 | 116 | |
6d048f53 | 117 | sector_t last_position; |
22e2c507 | 118 | unsigned long last_end_request; |
1da177e4 | 119 | |
1da177e4 LT |
120 | /* |
121 | * tunables, see top of file | |
122 | */ | |
123 | unsigned int cfq_quantum; | |
22e2c507 | 124 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
125 | unsigned int cfq_back_penalty; |
126 | unsigned int cfq_back_max; | |
22e2c507 JA |
127 | unsigned int cfq_slice[2]; |
128 | unsigned int cfq_slice_async_rq; | |
129 | unsigned int cfq_slice_idle; | |
d9ff4187 AV |
130 | |
131 | struct list_head cic_list; | |
6d048f53 JA |
132 | |
133 | sector_t new_seek_mean; | |
134 | u64 new_seek_total; | |
1da177e4 LT |
135 | }; |
136 | ||
22e2c507 JA |
137 | /* |
138 | * Per process-grouping structure | |
139 | */ | |
1da177e4 LT |
140 | struct cfq_queue { |
141 | /* reference count */ | |
142 | atomic_t ref; | |
143 | /* parent cfq_data */ | |
144 | struct cfq_data *cfqd; | |
22e2c507 | 145 | /* cfqq lookup hash */ |
1da177e4 LT |
146 | struct hlist_node cfq_hash; |
147 | /* hash key */ | |
22e2c507 | 148 | unsigned int key; |
981a7973 | 149 | /* member of the rr/busy/cur/idle cfqd list */ |
1da177e4 | 150 | struct list_head cfq_list; |
d9e7620e JA |
151 | /* service_tree member */ |
152 | struct rb_node rb_node; | |
153 | /* service_tree key */ | |
154 | unsigned long rb_key; | |
1da177e4 LT |
155 | /* sorted list of pending requests */ |
156 | struct rb_root sort_list; | |
157 | /* if fifo isn't expired, next request to serve */ | |
5e705374 | 158 | struct request *next_rq; |
1da177e4 LT |
159 | /* requests queued in sort_list */ |
160 | int queued[2]; | |
161 | /* currently allocated requests */ | |
162 | int allocated[2]; | |
374f84ac JA |
163 | /* pending metadata requests */ |
164 | int meta_pending; | |
1da177e4 | 165 | /* fifo list of requests in sort_list */ |
22e2c507 | 166 | struct list_head fifo; |
1da177e4 | 167 | |
22e2c507 | 168 | unsigned long slice_end; |
c5b680f3 | 169 | long slice_resid; |
1da177e4 | 170 | |
6d048f53 JA |
171 | /* number of requests that are on the dispatch list or inside driver */ |
172 | int dispatched; | |
22e2c507 JA |
173 | |
174 | /* io prio of this group */ | |
175 | unsigned short ioprio, org_ioprio; | |
176 | unsigned short ioprio_class, org_ioprio_class; | |
177 | ||
3b18152c JA |
178 | /* various state flags, see below */ |
179 | unsigned int flags; | |
6d048f53 JA |
180 | |
181 | sector_t last_request_pos; | |
1da177e4 LT |
182 | }; |
183 | ||
3b18152c | 184 | enum cfqq_state_flags { |
b0b8d749 JA |
185 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
186 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
187 | CFQ_CFQQ_FLAG_must_alloc, /* must be allowed rq alloc */ | |
188 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ | |
189 | CFQ_CFQQ_FLAG_must_dispatch, /* must dispatch, even if expired */ | |
190 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ | |
191 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
192 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
193 | CFQ_CFQQ_FLAG_queue_new, /* queue never been serviced */ | |
44f7c160 | 194 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
3b18152c JA |
195 | }; |
196 | ||
197 | #define CFQ_CFQQ_FNS(name) \ | |
198 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
199 | { \ | |
200 | cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ | |
201 | } \ | |
202 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
203 | { \ | |
204 | cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ | |
205 | } \ | |
206 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
207 | { \ | |
208 | return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ | |
209 | } | |
210 | ||
211 | CFQ_CFQQ_FNS(on_rr); | |
212 | CFQ_CFQQ_FNS(wait_request); | |
213 | CFQ_CFQQ_FNS(must_alloc); | |
214 | CFQ_CFQQ_FNS(must_alloc_slice); | |
215 | CFQ_CFQQ_FNS(must_dispatch); | |
216 | CFQ_CFQQ_FNS(fifo_expire); | |
217 | CFQ_CFQQ_FNS(idle_window); | |
218 | CFQ_CFQQ_FNS(prio_changed); | |
53b03744 | 219 | CFQ_CFQQ_FNS(queue_new); |
44f7c160 | 220 | CFQ_CFQQ_FNS(slice_new); |
3b18152c JA |
221 | #undef CFQ_CFQQ_FNS |
222 | ||
3b18152c | 223 | static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short); |
5e705374 | 224 | static void cfq_dispatch_insert(request_queue_t *, struct request *); |
498d3aa2 | 225 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, unsigned int, struct task_struct *, gfp_t); |
1da177e4 | 226 | |
99f95e52 AM |
227 | /* |
228 | * scheduler run of queue, if there are requests pending and no one in the | |
229 | * driver that will restart queueing | |
230 | */ | |
231 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
232 | { | |
7b14e3b5 | 233 | if (cfqd->busy_queues) |
99f95e52 AM |
234 | kblockd_schedule_work(&cfqd->unplug_work); |
235 | } | |
236 | ||
237 | static int cfq_queue_empty(request_queue_t *q) | |
238 | { | |
239 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
240 | ||
b4878f24 | 241 | return !cfqd->busy_queues; |
99f95e52 AM |
242 | } |
243 | ||
7749a8d4 | 244 | static inline pid_t cfq_queue_pid(struct task_struct *task, int rw, int is_sync) |
206dc69b | 245 | { |
7749a8d4 JA |
246 | /* |
247 | * Use the per-process queue, for read requests and syncronous writes | |
248 | */ | |
249 | if (!(rw & REQ_RW) || is_sync) | |
206dc69b JA |
250 | return task->pid; |
251 | ||
252 | return CFQ_KEY_ASYNC; | |
253 | } | |
254 | ||
44f7c160 JA |
255 | /* |
256 | * Scale schedule slice based on io priority. Use the sync time slice only | |
257 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
258 | * io only, should not get full sync slice length. | |
259 | */ | |
d9e7620e JA |
260 | static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync, |
261 | unsigned short prio) | |
44f7c160 | 262 | { |
d9e7620e | 263 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 264 | |
d9e7620e JA |
265 | WARN_ON(prio >= IOPRIO_BE_NR); |
266 | ||
267 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
268 | } | |
44f7c160 | 269 | |
d9e7620e JA |
270 | static inline int |
271 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
272 | { | |
273 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
274 | } |
275 | ||
276 | static inline void | |
277 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
278 | { | |
279 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; | |
280 | } | |
281 | ||
282 | /* | |
283 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
284 | * isn't valid until the first request from the dispatch is activated | |
285 | * and the slice time set. | |
286 | */ | |
287 | static inline int cfq_slice_used(struct cfq_queue *cfqq) | |
288 | { | |
289 | if (cfq_cfqq_slice_new(cfqq)) | |
290 | return 0; | |
291 | if (time_before(jiffies, cfqq->slice_end)) | |
292 | return 0; | |
293 | ||
294 | return 1; | |
295 | } | |
296 | ||
1da177e4 | 297 | /* |
5e705374 | 298 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 299 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 300 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 301 | */ |
5e705374 JA |
302 | static struct request * |
303 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
304 | { |
305 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 306 | unsigned long back_max; |
e8a99053 AM |
307 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
308 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
309 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 310 | |
5e705374 JA |
311 | if (rq1 == NULL || rq1 == rq2) |
312 | return rq2; | |
313 | if (rq2 == NULL) | |
314 | return rq1; | |
9c2c38a1 | 315 | |
5e705374 JA |
316 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
317 | return rq1; | |
318 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
319 | return rq2; | |
374f84ac JA |
320 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
321 | return rq1; | |
322 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
323 | return rq2; | |
1da177e4 | 324 | |
5e705374 JA |
325 | s1 = rq1->sector; |
326 | s2 = rq2->sector; | |
1da177e4 | 327 | |
6d048f53 | 328 | last = cfqd->last_position; |
1da177e4 | 329 | |
1da177e4 LT |
330 | /* |
331 | * by definition, 1KiB is 2 sectors | |
332 | */ | |
333 | back_max = cfqd->cfq_back_max * 2; | |
334 | ||
335 | /* | |
336 | * Strict one way elevator _except_ in the case where we allow | |
337 | * short backward seeks which are biased as twice the cost of a | |
338 | * similar forward seek. | |
339 | */ | |
340 | if (s1 >= last) | |
341 | d1 = s1 - last; | |
342 | else if (s1 + back_max >= last) | |
343 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
344 | else | |
e8a99053 | 345 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
346 | |
347 | if (s2 >= last) | |
348 | d2 = s2 - last; | |
349 | else if (s2 + back_max >= last) | |
350 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
351 | else | |
e8a99053 | 352 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
353 | |
354 | /* Found required data */ | |
e8a99053 AM |
355 | |
356 | /* | |
357 | * By doing switch() on the bit mask "wrap" we avoid having to | |
358 | * check two variables for all permutations: --> faster! | |
359 | */ | |
360 | switch (wrap) { | |
5e705374 | 361 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 362 | if (d1 < d2) |
5e705374 | 363 | return rq1; |
e8a99053 | 364 | else if (d2 < d1) |
5e705374 | 365 | return rq2; |
e8a99053 AM |
366 | else { |
367 | if (s1 >= s2) | |
5e705374 | 368 | return rq1; |
e8a99053 | 369 | else |
5e705374 | 370 | return rq2; |
e8a99053 | 371 | } |
1da177e4 | 372 | |
e8a99053 | 373 | case CFQ_RQ2_WRAP: |
5e705374 | 374 | return rq1; |
e8a99053 | 375 | case CFQ_RQ1_WRAP: |
5e705374 JA |
376 | return rq2; |
377 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
378 | default: |
379 | /* | |
380 | * Since both rqs are wrapped, | |
381 | * start with the one that's further behind head | |
382 | * (--> only *one* back seek required), | |
383 | * since back seek takes more time than forward. | |
384 | */ | |
385 | if (s1 <= s2) | |
5e705374 | 386 | return rq1; |
1da177e4 | 387 | else |
5e705374 | 388 | return rq2; |
1da177e4 LT |
389 | } |
390 | } | |
391 | ||
498d3aa2 JA |
392 | /* |
393 | * The below is leftmost cache rbtree addon | |
394 | */ | |
cc09e299 JA |
395 | static struct rb_node *cfq_rb_first(struct cfq_rb_root *root) |
396 | { | |
397 | if (!root->left) | |
398 | root->left = rb_first(&root->rb); | |
399 | ||
400 | return root->left; | |
401 | } | |
402 | ||
403 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) | |
404 | { | |
405 | if (root->left == n) | |
406 | root->left = NULL; | |
407 | ||
408 | rb_erase(n, &root->rb); | |
409 | RB_CLEAR_NODE(n); | |
410 | } | |
411 | ||
1da177e4 LT |
412 | /* |
413 | * would be nice to take fifo expire time into account as well | |
414 | */ | |
5e705374 JA |
415 | static struct request * |
416 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
417 | struct request *last) | |
1da177e4 | 418 | { |
21183b07 JA |
419 | struct rb_node *rbnext = rb_next(&last->rb_node); |
420 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 421 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 422 | |
21183b07 | 423 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
424 | |
425 | if (rbprev) | |
5e705374 | 426 | prev = rb_entry_rq(rbprev); |
1da177e4 | 427 | |
21183b07 | 428 | if (rbnext) |
5e705374 | 429 | next = rb_entry_rq(rbnext); |
21183b07 JA |
430 | else { |
431 | rbnext = rb_first(&cfqq->sort_list); | |
432 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 433 | next = rb_entry_rq(rbnext); |
21183b07 | 434 | } |
1da177e4 | 435 | |
21183b07 | 436 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
437 | } |
438 | ||
d9e7620e JA |
439 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
440 | struct cfq_queue *cfqq) | |
1da177e4 | 441 | { |
d9e7620e JA |
442 | /* |
443 | * just an approximation, should be ok. | |
444 | */ | |
67e6b49e JA |
445 | return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
446 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); | |
d9e7620e JA |
447 | } |
448 | ||
498d3aa2 JA |
449 | /* |
450 | * The cfqd->service_tree holds all pending cfq_queue's that have | |
451 | * requests waiting to be processed. It is sorted in the order that | |
452 | * we will service the queues. | |
453 | */ | |
d9e7620e JA |
454 | static void cfq_service_tree_add(struct cfq_data *cfqd, |
455 | struct cfq_queue *cfqq) | |
456 | { | |
cc09e299 | 457 | struct rb_node **p = &cfqd->service_tree.rb.rb_node; |
d9e7620e | 458 | struct rb_node *parent = NULL; |
d9e7620e | 459 | unsigned long rb_key; |
498d3aa2 | 460 | int left; |
d9e7620e JA |
461 | |
462 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; | |
463 | rb_key += cfqq->slice_resid; | |
464 | cfqq->slice_resid = 0; | |
1da177e4 | 465 | |
d9e7620e | 466 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
99f9628a | 467 | /* |
d9e7620e | 468 | * same position, nothing more to do |
99f9628a | 469 | */ |
d9e7620e JA |
470 | if (rb_key == cfqq->rb_key) |
471 | return; | |
1da177e4 | 472 | |
cc09e299 | 473 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); |
1da177e4 | 474 | } |
d9e7620e | 475 | |
498d3aa2 | 476 | left = 1; |
d9e7620e | 477 | while (*p) { |
cc09e299 | 478 | struct cfq_queue *__cfqq; |
67060e37 | 479 | struct rb_node **n; |
cc09e299 | 480 | |
d9e7620e JA |
481 | parent = *p; |
482 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
483 | ||
0c534e0a JA |
484 | /* |
485 | * sort RT queues first, we always want to give | |
67060e37 JA |
486 | * preference to them. IDLE queues goes to the back. |
487 | * after that, sort on the next service time. | |
0c534e0a JA |
488 | */ |
489 | if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq)) | |
67060e37 | 490 | n = &(*p)->rb_left; |
0c534e0a | 491 | else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq)) |
67060e37 JA |
492 | n = &(*p)->rb_right; |
493 | else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq)) | |
494 | n = &(*p)->rb_left; | |
495 | else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq)) | |
496 | n = &(*p)->rb_right; | |
0c534e0a | 497 | else if (rb_key < __cfqq->rb_key) |
67060e37 JA |
498 | n = &(*p)->rb_left; |
499 | else | |
500 | n = &(*p)->rb_right; | |
501 | ||
502 | if (n == &(*p)->rb_right) | |
cc09e299 | 503 | left = 0; |
67060e37 JA |
504 | |
505 | p = n; | |
d9e7620e JA |
506 | } |
507 | ||
cc09e299 JA |
508 | if (left) |
509 | cfqd->service_tree.left = &cfqq->rb_node; | |
510 | ||
d9e7620e JA |
511 | cfqq->rb_key = rb_key; |
512 | rb_link_node(&cfqq->rb_node, parent, p); | |
cc09e299 | 513 | rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb); |
1da177e4 LT |
514 | } |
515 | ||
498d3aa2 JA |
516 | /* |
517 | * Update cfqq's position in the service tree. | |
518 | */ | |
6d048f53 JA |
519 | static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted) |
520 | { | |
6d048f53 JA |
521 | /* |
522 | * Resorting requires the cfqq to be on the RR list already. | |
523 | */ | |
498d3aa2 JA |
524 | if (cfq_cfqq_on_rr(cfqq)) |
525 | cfq_service_tree_add(cfqq->cfqd, cfqq); | |
6d048f53 JA |
526 | } |
527 | ||
1da177e4 LT |
528 | /* |
529 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 530 | * the pending list according to last request service |
1da177e4 LT |
531 | */ |
532 | static inline void | |
b4878f24 | 533 | cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 534 | { |
3b18152c JA |
535 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
536 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
537 | cfqd->busy_queues++; |
538 | ||
b4878f24 | 539 | cfq_resort_rr_list(cfqq, 0); |
1da177e4 LT |
540 | } |
541 | ||
498d3aa2 JA |
542 | /* |
543 | * Called when the cfqq no longer has requests pending, remove it from | |
544 | * the service tree. | |
545 | */ | |
1da177e4 LT |
546 | static inline void |
547 | cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
548 | { | |
3b18152c JA |
549 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
550 | cfq_clear_cfqq_on_rr(cfqq); | |
981a7973 | 551 | list_del_init(&cfqq->cfq_list); |
1da177e4 | 552 | |
cc09e299 JA |
553 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) |
554 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); | |
d9e7620e | 555 | |
1da177e4 LT |
556 | BUG_ON(!cfqd->busy_queues); |
557 | cfqd->busy_queues--; | |
558 | } | |
559 | ||
560 | /* | |
561 | * rb tree support functions | |
562 | */ | |
5e705374 | 563 | static inline void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 564 | { |
5e705374 | 565 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 566 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 567 | const int sync = rq_is_sync(rq); |
1da177e4 | 568 | |
b4878f24 JA |
569 | BUG_ON(!cfqq->queued[sync]); |
570 | cfqq->queued[sync]--; | |
1da177e4 | 571 | |
5e705374 | 572 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 573 | |
dd67d051 | 574 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 575 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
576 | } |
577 | ||
5e705374 | 578 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 579 | { |
5e705374 | 580 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 581 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 582 | struct request *__alias; |
1da177e4 | 583 | |
5380a101 | 584 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
585 | |
586 | /* | |
587 | * looks a little odd, but the first insert might return an alias. | |
588 | * if that happens, put the alias on the dispatch list | |
589 | */ | |
21183b07 | 590 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 591 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
592 | |
593 | if (!cfq_cfqq_on_rr(cfqq)) | |
594 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
595 | |
596 | /* | |
597 | * check if this request is a better next-serve candidate | |
598 | */ | |
599 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); | |
600 | BUG_ON(!cfqq->next_rq); | |
1da177e4 LT |
601 | } |
602 | ||
603 | static inline void | |
5e705374 | 604 | cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 605 | { |
5380a101 JA |
606 | elv_rb_del(&cfqq->sort_list, rq); |
607 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 608 | cfq_add_rq_rb(rq); |
1da177e4 LT |
609 | } |
610 | ||
206dc69b JA |
611 | static struct request * |
612 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 613 | { |
206dc69b | 614 | struct task_struct *tsk = current; |
7749a8d4 | 615 | pid_t key = cfq_queue_pid(tsk, bio_data_dir(bio), bio_sync(bio)); |
206dc69b | 616 | struct cfq_queue *cfqq; |
1da177e4 | 617 | |
206dc69b | 618 | cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio); |
89850f7e JA |
619 | if (cfqq) { |
620 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
621 | ||
21183b07 | 622 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 623 | } |
1da177e4 | 624 | |
1da177e4 LT |
625 | return NULL; |
626 | } | |
627 | ||
b4878f24 | 628 | static void cfq_activate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 629 | { |
22e2c507 | 630 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 631 | |
b4878f24 | 632 | cfqd->rq_in_driver++; |
25776e35 JA |
633 | |
634 | /* | |
635 | * If the depth is larger 1, it really could be queueing. But lets | |
636 | * make the mark a little higher - idling could still be good for | |
637 | * low queueing, and a low queueing number could also just indicate | |
638 | * a SCSI mid layer like behaviour where limit+1 is often seen. | |
639 | */ | |
640 | if (!cfqd->hw_tag && cfqd->rq_in_driver > 4) | |
641 | cfqd->hw_tag = 1; | |
6d048f53 JA |
642 | |
643 | cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors; | |
1da177e4 LT |
644 | } |
645 | ||
b4878f24 | 646 | static void cfq_deactivate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 647 | { |
b4878f24 JA |
648 | struct cfq_data *cfqd = q->elevator->elevator_data; |
649 | ||
650 | WARN_ON(!cfqd->rq_in_driver); | |
651 | cfqd->rq_in_driver--; | |
1da177e4 LT |
652 | } |
653 | ||
b4878f24 | 654 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 655 | { |
5e705374 | 656 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 657 | |
5e705374 JA |
658 | if (cfqq->next_rq == rq) |
659 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 660 | |
b4878f24 | 661 | list_del_init(&rq->queuelist); |
5e705374 | 662 | cfq_del_rq_rb(rq); |
374f84ac JA |
663 | |
664 | if (rq_is_meta(rq)) { | |
665 | WARN_ON(!cfqq->meta_pending); | |
666 | cfqq->meta_pending--; | |
667 | } | |
1da177e4 LT |
668 | } |
669 | ||
498d3aa2 | 670 | static int cfq_merge(request_queue_t *q, struct request **req, struct bio *bio) |
1da177e4 LT |
671 | { |
672 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
673 | struct request *__rq; | |
1da177e4 | 674 | |
206dc69b | 675 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 676 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
677 | *req = __rq; |
678 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
679 | } |
680 | ||
681 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
682 | } |
683 | ||
21183b07 JA |
684 | static void cfq_merged_request(request_queue_t *q, struct request *req, |
685 | int type) | |
1da177e4 | 686 | { |
21183b07 | 687 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 688 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 689 | |
5e705374 | 690 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 691 | } |
1da177e4 LT |
692 | } |
693 | ||
694 | static void | |
695 | cfq_merged_requests(request_queue_t *q, struct request *rq, | |
696 | struct request *next) | |
697 | { | |
22e2c507 JA |
698 | /* |
699 | * reposition in fifo if next is older than rq | |
700 | */ | |
701 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
702 | time_before(next->start_time, rq->start_time)) | |
703 | list_move(&rq->queuelist, &next->queuelist); | |
704 | ||
b4878f24 | 705 | cfq_remove_request(next); |
22e2c507 JA |
706 | } |
707 | ||
da775265 JA |
708 | static int cfq_allow_merge(request_queue_t *q, struct request *rq, |
709 | struct bio *bio) | |
710 | { | |
711 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
712 | const int rw = bio_data_dir(bio); | |
713 | struct cfq_queue *cfqq; | |
714 | pid_t key; | |
715 | ||
716 | /* | |
ec8acb69 | 717 | * Disallow merge of a sync bio into an async request. |
da775265 | 718 | */ |
ec8acb69 | 719 | if ((bio_data_dir(bio) == READ || bio_sync(bio)) && !rq_is_sync(rq)) |
da775265 JA |
720 | return 0; |
721 | ||
722 | /* | |
719d3402 JA |
723 | * Lookup the cfqq that this bio will be queued with. Allow |
724 | * merge only if rq is queued there. | |
da775265 | 725 | */ |
719d3402 | 726 | key = cfq_queue_pid(current, rw, bio_sync(bio)); |
da775265 | 727 | cfqq = cfq_find_cfq_hash(cfqd, key, current->ioprio); |
719d3402 JA |
728 | |
729 | if (cfqq == RQ_CFQQ(rq)) | |
730 | return 1; | |
da775265 | 731 | |
ec8acb69 | 732 | return 0; |
da775265 JA |
733 | } |
734 | ||
22e2c507 JA |
735 | static inline void |
736 | __cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
737 | { | |
738 | if (cfqq) { | |
739 | /* | |
740 | * stop potential idle class queues waiting service | |
741 | */ | |
742 | del_timer(&cfqd->idle_class_timer); | |
743 | ||
22e2c507 | 744 | cfqq->slice_end = 0; |
3b18152c JA |
745 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
746 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 747 | cfq_mark_cfqq_slice_new(cfqq); |
1afba045 | 748 | cfq_clear_cfqq_queue_new(cfqq); |
22e2c507 JA |
749 | } |
750 | ||
751 | cfqd->active_queue = cfqq; | |
752 | } | |
753 | ||
7b14e3b5 JA |
754 | /* |
755 | * current cfqq expired its slice (or was too idle), select new one | |
756 | */ | |
757 | static void | |
758 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
3c6bd2f8 | 759 | int preempted, int timed_out) |
7b14e3b5 | 760 | { |
7b14e3b5 JA |
761 | if (cfq_cfqq_wait_request(cfqq)) |
762 | del_timer(&cfqd->idle_slice_timer); | |
763 | ||
7b14e3b5 JA |
764 | cfq_clear_cfqq_must_dispatch(cfqq); |
765 | cfq_clear_cfqq_wait_request(cfqq); | |
766 | ||
767 | /* | |
768 | * store what was left of this slice, if the queue idled out | |
769 | * or was preempted | |
770 | */ | |
3c6bd2f8 | 771 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) |
c5b680f3 | 772 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b14e3b5 | 773 | |
98e41c7d | 774 | cfq_resort_rr_list(cfqq, preempted); |
7b14e3b5 JA |
775 | |
776 | if (cfqq == cfqd->active_queue) | |
777 | cfqd->active_queue = NULL; | |
778 | ||
779 | if (cfqd->active_cic) { | |
780 | put_io_context(cfqd->active_cic->ioc); | |
781 | cfqd->active_cic = NULL; | |
782 | } | |
783 | ||
784 | cfqd->dispatch_slice = 0; | |
785 | } | |
786 | ||
3c6bd2f8 JA |
787 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted, |
788 | int timed_out) | |
7b14e3b5 JA |
789 | { |
790 | struct cfq_queue *cfqq = cfqd->active_queue; | |
791 | ||
792 | if (cfqq) | |
3c6bd2f8 | 793 | __cfq_slice_expired(cfqd, cfqq, preempted, timed_out); |
7b14e3b5 JA |
794 | } |
795 | ||
498d3aa2 JA |
796 | /* |
797 | * Get next queue for service. Unless we have a queue preemption, | |
798 | * we'll simply select the first cfqq in the service tree. | |
799 | */ | |
6d048f53 | 800 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 801 | { |
7b14e3b5 | 802 | struct cfq_queue *cfqq = NULL; |
22e2c507 | 803 | |
d9e7620e | 804 | if (!list_empty(&cfqd->cur_rr)) { |
89850f7e | 805 | /* |
d9e7620e | 806 | * if current list is non-empty, grab first entry. |
89850f7e | 807 | */ |
d9e7620e | 808 | cfqq = list_entry_cfqq(cfqd->cur_rr.next); |
cc09e299 JA |
809 | } else if (!RB_EMPTY_ROOT(&cfqd->service_tree.rb)) { |
810 | struct rb_node *n = cfq_rb_first(&cfqd->service_tree); | |
d9e7620e JA |
811 | |
812 | cfqq = rb_entry(n, struct cfq_queue, rb_node); | |
67060e37 | 813 | if (cfq_class_idle(cfqq)) { |
498d3aa2 JA |
814 | unsigned long end; |
815 | ||
67060e37 JA |
816 | /* |
817 | * if we have idle queues and no rt or be queues had | |
818 | * pending requests, either allow immediate service if | |
819 | * the grace period has passed or arm the idle grace | |
820 | * timer | |
821 | */ | |
822 | end = cfqd->last_end_request + CFQ_IDLE_GRACE; | |
823 | if (time_before(jiffies, end)) { | |
824 | mod_timer(&cfqd->idle_class_timer, end); | |
825 | cfqq = NULL; | |
826 | } | |
827 | } | |
22e2c507 JA |
828 | } |
829 | ||
6d048f53 JA |
830 | return cfqq; |
831 | } | |
832 | ||
498d3aa2 JA |
833 | /* |
834 | * Get and set a new active queue for service. | |
835 | */ | |
6d048f53 JA |
836 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
837 | { | |
838 | struct cfq_queue *cfqq; | |
839 | ||
d9e7620e | 840 | cfqq = cfq_get_next_queue(cfqd); |
22e2c507 | 841 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 842 | return cfqq; |
22e2c507 JA |
843 | } |
844 | ||
d9e7620e JA |
845 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
846 | struct request *rq) | |
847 | { | |
848 | if (rq->sector >= cfqd->last_position) | |
849 | return rq->sector - cfqd->last_position; | |
850 | else | |
851 | return cfqd->last_position - rq->sector; | |
852 | } | |
853 | ||
6d048f53 JA |
854 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq) |
855 | { | |
856 | struct cfq_io_context *cic = cfqd->active_cic; | |
857 | ||
858 | if (!sample_valid(cic->seek_samples)) | |
859 | return 0; | |
860 | ||
861 | return cfq_dist_from_last(cfqd, rq) <= cic->seek_mean; | |
862 | } | |
863 | ||
d9e7620e JA |
864 | static int cfq_close_cooperator(struct cfq_data *cfq_data, |
865 | struct cfq_queue *cfqq) | |
6d048f53 | 866 | { |
6d048f53 | 867 | /* |
d9e7620e JA |
868 | * We should notice if some of the queues are cooperating, eg |
869 | * working closely on the same area of the disk. In that case, | |
870 | * we can group them together and don't waste time idling. | |
6d048f53 | 871 | */ |
d9e7620e | 872 | return 0; |
6d048f53 JA |
873 | } |
874 | ||
875 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (8 * 1024)) | |
caaa5f9f | 876 | |
6d048f53 | 877 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 878 | { |
1792669c | 879 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 880 | struct cfq_io_context *cic; |
7b14e3b5 JA |
881 | unsigned long sl; |
882 | ||
dd67d051 | 883 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 884 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
885 | |
886 | /* | |
887 | * idle is disabled, either manually or by past process history | |
888 | */ | |
6d048f53 JA |
889 | if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq)) |
890 | return; | |
891 | ||
22e2c507 JA |
892 | /* |
893 | * task has exited, don't wait | |
894 | */ | |
206dc69b JA |
895 | cic = cfqd->active_cic; |
896 | if (!cic || !cic->ioc->task) | |
6d048f53 JA |
897 | return; |
898 | ||
899 | /* | |
900 | * See if this prio level has a good candidate | |
901 | */ | |
1afba045 JA |
902 | if (cfq_close_cooperator(cfqd, cfqq) && |
903 | (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2)) | |
6d048f53 | 904 | return; |
22e2c507 | 905 | |
3b18152c JA |
906 | cfq_mark_cfqq_must_dispatch(cfqq); |
907 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 908 | |
206dc69b JA |
909 | /* |
910 | * we don't want to idle for seeks, but we do want to allow | |
911 | * fair distribution of slice time for a process doing back-to-back | |
912 | * seeks. so allow a little bit of time for him to submit a new rq | |
913 | */ | |
6d048f53 | 914 | sl = cfqd->cfq_slice_idle; |
caaa5f9f | 915 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
d9e7620e | 916 | sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT)); |
206dc69b | 917 | |
7b14e3b5 | 918 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
1da177e4 LT |
919 | } |
920 | ||
498d3aa2 JA |
921 | /* |
922 | * Move request from internal lists to the request queue dispatch list. | |
923 | */ | |
5e705374 | 924 | static void cfq_dispatch_insert(request_queue_t *q, struct request *rq) |
1da177e4 | 925 | { |
5e705374 | 926 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 927 | |
5380a101 | 928 | cfq_remove_request(rq); |
6d048f53 | 929 | cfqq->dispatched++; |
5380a101 | 930 | elv_dispatch_sort(q, rq); |
1da177e4 LT |
931 | } |
932 | ||
933 | /* | |
934 | * return expired entry, or NULL to just start from scratch in rbtree | |
935 | */ | |
5e705374 | 936 | static inline struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
937 | { |
938 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 939 | struct request *rq; |
89850f7e | 940 | int fifo; |
1da177e4 | 941 | |
3b18152c | 942 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 943 | return NULL; |
cb887411 JA |
944 | |
945 | cfq_mark_cfqq_fifo_expire(cfqq); | |
946 | ||
89850f7e JA |
947 | if (list_empty(&cfqq->fifo)) |
948 | return NULL; | |
1da177e4 | 949 | |
6d048f53 | 950 | fifo = cfq_cfqq_sync(cfqq); |
89850f7e | 951 | rq = rq_entry_fifo(cfqq->fifo.next); |
1da177e4 | 952 | |
6d048f53 JA |
953 | if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) |
954 | return NULL; | |
1da177e4 | 955 | |
6d048f53 | 956 | return rq; |
1da177e4 LT |
957 | } |
958 | ||
22e2c507 JA |
959 | static inline int |
960 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
961 | { | |
962 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 963 | |
22e2c507 | 964 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 965 | |
22e2c507 | 966 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
967 | } |
968 | ||
22e2c507 | 969 | /* |
498d3aa2 JA |
970 | * Select a queue for service. If we have a current active queue, |
971 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 972 | */ |
1b5ed5e1 | 973 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 974 | { |
1da177e4 | 975 | struct cfq_queue *cfqq; |
1da177e4 | 976 | |
22e2c507 JA |
977 | cfqq = cfqd->active_queue; |
978 | if (!cfqq) | |
979 | goto new_queue; | |
1da177e4 | 980 | |
22e2c507 | 981 | /* |
6d048f53 | 982 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 983 | */ |
6d048f53 | 984 | if (cfq_slice_used(cfqq)) |
3b18152c | 985 | goto expire; |
1da177e4 | 986 | |
22e2c507 | 987 | /* |
6d048f53 JA |
988 | * The active queue has requests and isn't expired, allow it to |
989 | * dispatch. | |
22e2c507 | 990 | */ |
dd67d051 | 991 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 992 | goto keep_queue; |
6d048f53 JA |
993 | |
994 | /* | |
995 | * No requests pending. If the active queue still has requests in | |
996 | * flight or is idling for a new request, allow either of these | |
997 | * conditions to happen (or time out) before selecting a new queue. | |
998 | */ | |
999 | if (cfqq->dispatched || timer_pending(&cfqd->idle_slice_timer)) { | |
caaa5f9f JA |
1000 | cfqq = NULL; |
1001 | goto keep_queue; | |
22e2c507 JA |
1002 | } |
1003 | ||
3b18152c | 1004 | expire: |
3c6bd2f8 | 1005 | cfq_slice_expired(cfqd, 0, 0); |
3b18152c JA |
1006 | new_queue: |
1007 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 1008 | keep_queue: |
3b18152c | 1009 | return cfqq; |
22e2c507 JA |
1010 | } |
1011 | ||
498d3aa2 JA |
1012 | /* |
1013 | * Dispatch some requests from cfqq, moving them to the request queue | |
1014 | * dispatch list. | |
1015 | */ | |
22e2c507 JA |
1016 | static int |
1017 | __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1018 | int max_dispatch) | |
1019 | { | |
1020 | int dispatched = 0; | |
1021 | ||
dd67d051 | 1022 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
1023 | |
1024 | do { | |
5e705374 | 1025 | struct request *rq; |
1da177e4 LT |
1026 | |
1027 | /* | |
22e2c507 | 1028 | * follow expired path, else get first next available |
1da177e4 | 1029 | */ |
5e705374 JA |
1030 | if ((rq = cfq_check_fifo(cfqq)) == NULL) |
1031 | rq = cfqq->next_rq; | |
22e2c507 JA |
1032 | |
1033 | /* | |
1034 | * finally, insert request into driver dispatch list | |
1035 | */ | |
5e705374 | 1036 | cfq_dispatch_insert(cfqd->queue, rq); |
1da177e4 | 1037 | |
22e2c507 JA |
1038 | cfqd->dispatch_slice++; |
1039 | dispatched++; | |
1da177e4 | 1040 | |
22e2c507 | 1041 | if (!cfqd->active_cic) { |
5e705374 JA |
1042 | atomic_inc(&RQ_CIC(rq)->ioc->refcount); |
1043 | cfqd->active_cic = RQ_CIC(rq); | |
22e2c507 | 1044 | } |
1da177e4 | 1045 | |
dd67d051 | 1046 | if (RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 JA |
1047 | break; |
1048 | ||
1049 | } while (dispatched < max_dispatch); | |
1050 | ||
22e2c507 JA |
1051 | /* |
1052 | * expire an async queue immediately if it has used up its slice. idle | |
1053 | * queue always expire after 1 dispatch round. | |
1054 | */ | |
a9938006 | 1055 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
22e2c507 | 1056 | cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
a9938006 | 1057 | cfq_class_idle(cfqq))) { |
44f7c160 | 1058 | cfqq->slice_end = jiffies + 1; |
3c6bd2f8 | 1059 | cfq_slice_expired(cfqd, 0, 0); |
44f7c160 | 1060 | } |
22e2c507 JA |
1061 | |
1062 | return dispatched; | |
1063 | } | |
1064 | ||
d9e7620e JA |
1065 | static inline int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
1066 | { | |
1067 | int dispatched = 0; | |
1068 | ||
1069 | while (cfqq->next_rq) { | |
1070 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1071 | dispatched++; | |
1072 | } | |
1073 | ||
1074 | BUG_ON(!list_empty(&cfqq->fifo)); | |
1075 | return dispatched; | |
1076 | } | |
1077 | ||
1078 | static int cfq_forced_dispatch_cfqqs(struct list_head *list) | |
1b5ed5e1 | 1079 | { |
1b5ed5e1 | 1080 | struct cfq_queue *cfqq, *next; |
caaa5f9f | 1081 | int dispatched; |
1b5ed5e1 | 1082 | |
caaa5f9f | 1083 | dispatched = 0; |
d9e7620e JA |
1084 | list_for_each_entry_safe(cfqq, next, list, cfq_list) |
1085 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); | |
caaa5f9f | 1086 | |
1b5ed5e1 TH |
1087 | return dispatched; |
1088 | } | |
1089 | ||
498d3aa2 JA |
1090 | /* |
1091 | * Drain our current requests. Used for barriers and when switching | |
1092 | * io schedulers on-the-fly. | |
1093 | */ | |
d9e7620e | 1094 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 1095 | { |
d9e7620e JA |
1096 | int dispatched = 0; |
1097 | struct rb_node *n; | |
1098 | ||
cc09e299 | 1099 | while ((n = cfq_rb_first(&cfqd->service_tree)) != NULL) { |
d9e7620e | 1100 | struct cfq_queue *cfqq = rb_entry(n, struct cfq_queue, rb_node); |
1b5ed5e1 | 1101 | |
d9e7620e JA |
1102 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
1103 | } | |
1b5ed5e1 | 1104 | |
1b5ed5e1 | 1105 | dispatched += cfq_forced_dispatch_cfqqs(&cfqd->cur_rr); |
1b5ed5e1 | 1106 | |
3c6bd2f8 | 1107 | cfq_slice_expired(cfqd, 0, 0); |
1b5ed5e1 TH |
1108 | |
1109 | BUG_ON(cfqd->busy_queues); | |
1110 | ||
1111 | return dispatched; | |
1112 | } | |
1113 | ||
d9e7620e | 1114 | static int cfq_dispatch_requests(request_queue_t *q, int force) |
22e2c507 JA |
1115 | { |
1116 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
6d048f53 | 1117 | struct cfq_queue *cfqq; |
caaa5f9f | 1118 | int dispatched; |
22e2c507 JA |
1119 | |
1120 | if (!cfqd->busy_queues) | |
1121 | return 0; | |
1122 | ||
1b5ed5e1 TH |
1123 | if (unlikely(force)) |
1124 | return cfq_forced_dispatch(cfqd); | |
1125 | ||
caaa5f9f | 1126 | dispatched = 0; |
caaa5f9f | 1127 | while ((cfqq = cfq_select_queue(cfqd)) != NULL) { |
b4878f24 JA |
1128 | int max_dispatch; |
1129 | ||
a9938006 | 1130 | if (cfqd->busy_queues > 1) { |
a9938006 JA |
1131 | /* |
1132 | * So we have dispatched before in this round, if the | |
1133 | * next queue has idling enabled (must be sync), don't | |
6d048f53 | 1134 | * allow it service until the previous have completed. |
a9938006 | 1135 | */ |
6d048f53 JA |
1136 | if (cfqd->rq_in_driver && cfq_cfqq_idle_window(cfqq) && |
1137 | dispatched) | |
1138 | break; | |
1139 | if (cfqq->dispatched >= cfqd->cfq_quantum) | |
a9938006 JA |
1140 | break; |
1141 | } | |
9ede209e | 1142 | |
3b18152c JA |
1143 | cfq_clear_cfqq_must_dispatch(cfqq); |
1144 | cfq_clear_cfqq_wait_request(cfqq); | |
22e2c507 JA |
1145 | del_timer(&cfqd->idle_slice_timer); |
1146 | ||
1b5ed5e1 TH |
1147 | max_dispatch = cfqd->cfq_quantum; |
1148 | if (cfq_class_idle(cfqq)) | |
1149 | max_dispatch = 1; | |
1da177e4 | 1150 | |
caaa5f9f | 1151 | dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); |
1da177e4 LT |
1152 | } |
1153 | ||
caaa5f9f | 1154 | return dispatched; |
1da177e4 LT |
1155 | } |
1156 | ||
1da177e4 | 1157 | /* |
5e705374 JA |
1158 | * task holds one reference to the queue, dropped when task exits. each rq |
1159 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1160 | * |
1161 | * queue lock must be held here. | |
1162 | */ | |
1163 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1164 | { | |
22e2c507 JA |
1165 | struct cfq_data *cfqd = cfqq->cfqd; |
1166 | ||
1167 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1168 | |
1169 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1170 | return; | |
1171 | ||
1172 | BUG_ON(rb_first(&cfqq->sort_list)); | |
22e2c507 | 1173 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1174 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1175 | |
28f95cbc | 1176 | if (unlikely(cfqd->active_queue == cfqq)) { |
3c6bd2f8 | 1177 | __cfq_slice_expired(cfqd, cfqq, 0, 0); |
28f95cbc JA |
1178 | cfq_schedule_dispatch(cfqd); |
1179 | } | |
22e2c507 | 1180 | |
1da177e4 LT |
1181 | /* |
1182 | * it's on the empty list and still hashed | |
1183 | */ | |
1da177e4 LT |
1184 | hlist_del(&cfqq->cfq_hash); |
1185 | kmem_cache_free(cfq_pool, cfqq); | |
1186 | } | |
1187 | ||
1ea25ecb | 1188 | static struct cfq_queue * |
3b18152c JA |
1189 | __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio, |
1190 | const int hashval) | |
1da177e4 LT |
1191 | { |
1192 | struct hlist_head *hash_list = &cfqd->cfq_hash[hashval]; | |
206dc69b JA |
1193 | struct hlist_node *entry; |
1194 | struct cfq_queue *__cfqq; | |
1da177e4 | 1195 | |
206dc69b | 1196 | hlist_for_each_entry(__cfqq, entry, hash_list, cfq_hash) { |
b0a6916b | 1197 | const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio); |
1da177e4 | 1198 | |
206dc69b | 1199 | if (__cfqq->key == key && (__p == prio || !prio)) |
1da177e4 LT |
1200 | return __cfqq; |
1201 | } | |
1202 | ||
1203 | return NULL; | |
1204 | } | |
1205 | ||
1206 | static struct cfq_queue * | |
3b18152c | 1207 | cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio) |
1da177e4 | 1208 | { |
3b18152c | 1209 | return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT)); |
1da177e4 LT |
1210 | } |
1211 | ||
e2d74ac0 | 1212 | static void cfq_free_io_context(struct io_context *ioc) |
1da177e4 | 1213 | { |
22e2c507 | 1214 | struct cfq_io_context *__cic; |
e2d74ac0 JA |
1215 | struct rb_node *n; |
1216 | int freed = 0; | |
1da177e4 | 1217 | |
e2d74ac0 JA |
1218 | while ((n = rb_first(&ioc->cic_root)) != NULL) { |
1219 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1220 | rb_erase(&__cic->rb_node, &ioc->cic_root); | |
22e2c507 | 1221 | kmem_cache_free(cfq_ioc_pool, __cic); |
334e94de | 1222 | freed++; |
1da177e4 LT |
1223 | } |
1224 | ||
4050cf16 JA |
1225 | elv_ioc_count_mod(ioc_count, -freed); |
1226 | ||
1227 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) | |
334e94de | 1228 | complete(ioc_gone); |
1da177e4 LT |
1229 | } |
1230 | ||
89850f7e | 1231 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1232 | { |
28f95cbc | 1233 | if (unlikely(cfqq == cfqd->active_queue)) { |
3c6bd2f8 | 1234 | __cfq_slice_expired(cfqd, cfqq, 0, 0); |
28f95cbc JA |
1235 | cfq_schedule_dispatch(cfqd); |
1236 | } | |
22e2c507 | 1237 | |
89850f7e JA |
1238 | cfq_put_queue(cfqq); |
1239 | } | |
22e2c507 | 1240 | |
89850f7e JA |
1241 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1242 | struct cfq_io_context *cic) | |
1243 | { | |
fc46379d JA |
1244 | list_del_init(&cic->queue_list); |
1245 | smp_wmb(); | |
1246 | cic->key = NULL; | |
1247 | ||
12a05732 | 1248 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1249 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1250 | cic->cfqq[ASYNC] = NULL; |
1251 | } | |
1252 | ||
1253 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1254 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1255 | cic->cfqq[SYNC] = NULL; |
1256 | } | |
89850f7e JA |
1257 | } |
1258 | ||
89850f7e JA |
1259 | static void cfq_exit_single_io_context(struct cfq_io_context *cic) |
1260 | { | |
1261 | struct cfq_data *cfqd = cic->key; | |
1262 | ||
89850f7e JA |
1263 | if (cfqd) { |
1264 | request_queue_t *q = cfqd->queue; | |
1265 | ||
fc46379d | 1266 | spin_lock_irq(q->queue_lock); |
89850f7e | 1267 | __cfq_exit_single_io_context(cfqd, cic); |
fc46379d | 1268 | spin_unlock_irq(q->queue_lock); |
89850f7e | 1269 | } |
1da177e4 LT |
1270 | } |
1271 | ||
498d3aa2 JA |
1272 | /* |
1273 | * The process that ioc belongs to has exited, we need to clean up | |
1274 | * and put the internal structures we have that belongs to that process. | |
1275 | */ | |
e2d74ac0 | 1276 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1277 | { |
22e2c507 | 1278 | struct cfq_io_context *__cic; |
e2d74ac0 | 1279 | struct rb_node *n; |
22e2c507 | 1280 | |
1da177e4 LT |
1281 | /* |
1282 | * put the reference this task is holding to the various queues | |
1283 | */ | |
e2d74ac0 JA |
1284 | |
1285 | n = rb_first(&ioc->cic_root); | |
1286 | while (n != NULL) { | |
1287 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1288 | ||
22e2c507 | 1289 | cfq_exit_single_io_context(__cic); |
e2d74ac0 | 1290 | n = rb_next(n); |
1da177e4 | 1291 | } |
1da177e4 LT |
1292 | } |
1293 | ||
22e2c507 | 1294 | static struct cfq_io_context * |
8267e268 | 1295 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1296 | { |
b5deef90 | 1297 | struct cfq_io_context *cic; |
1da177e4 | 1298 | |
b5deef90 | 1299 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask, cfqd->queue->node); |
1da177e4 | 1300 | if (cic) { |
553698f9 | 1301 | memset(cic, 0, sizeof(*cic)); |
22e2c507 | 1302 | cic->last_end_request = jiffies; |
553698f9 | 1303 | INIT_LIST_HEAD(&cic->queue_list); |
22e2c507 JA |
1304 | cic->dtor = cfq_free_io_context; |
1305 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1306 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1307 | } |
1308 | ||
1309 | return cic; | |
1310 | } | |
1311 | ||
22e2c507 JA |
1312 | static void cfq_init_prio_data(struct cfq_queue *cfqq) |
1313 | { | |
1314 | struct task_struct *tsk = current; | |
1315 | int ioprio_class; | |
1316 | ||
3b18152c | 1317 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1318 | return; |
1319 | ||
1320 | ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio); | |
1321 | switch (ioprio_class) { | |
1322 | default: | |
1323 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1324 | case IOPRIO_CLASS_NONE: | |
1325 | /* | |
1326 | * no prio set, place us in the middle of the BE classes | |
1327 | */ | |
1328 | cfqq->ioprio = task_nice_ioprio(tsk); | |
1329 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1330 | break; | |
1331 | case IOPRIO_CLASS_RT: | |
1332 | cfqq->ioprio = task_ioprio(tsk); | |
1333 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1334 | break; | |
1335 | case IOPRIO_CLASS_BE: | |
1336 | cfqq->ioprio = task_ioprio(tsk); | |
1337 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1338 | break; | |
1339 | case IOPRIO_CLASS_IDLE: | |
1340 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1341 | cfqq->ioprio = 7; | |
3b18152c | 1342 | cfq_clear_cfqq_idle_window(cfqq); |
22e2c507 JA |
1343 | break; |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * keep track of original prio settings in case we have to temporarily | |
1348 | * elevate the priority of this queue | |
1349 | */ | |
1350 | cfqq->org_ioprio = cfqq->ioprio; | |
1351 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 1352 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1353 | } |
1354 | ||
478a82b0 | 1355 | static inline void changed_ioprio(struct cfq_io_context *cic) |
22e2c507 | 1356 | { |
478a82b0 AV |
1357 | struct cfq_data *cfqd = cic->key; |
1358 | struct cfq_queue *cfqq; | |
c1b707d2 | 1359 | unsigned long flags; |
35e6077c | 1360 | |
caaa5f9f JA |
1361 | if (unlikely(!cfqd)) |
1362 | return; | |
1363 | ||
c1b707d2 | 1364 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1365 | |
1366 | cfqq = cic->cfqq[ASYNC]; | |
1367 | if (cfqq) { | |
1368 | struct cfq_queue *new_cfqq; | |
1369 | new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC, cic->ioc->task, | |
1370 | GFP_ATOMIC); | |
1371 | if (new_cfqq) { | |
1372 | cic->cfqq[ASYNC] = new_cfqq; | |
1373 | cfq_put_queue(cfqq); | |
1374 | } | |
22e2c507 | 1375 | } |
caaa5f9f JA |
1376 | |
1377 | cfqq = cic->cfqq[SYNC]; | |
1378 | if (cfqq) | |
1379 | cfq_mark_cfqq_prio_changed(cfqq); | |
1380 | ||
c1b707d2 | 1381 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1382 | } |
1383 | ||
fc46379d | 1384 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1385 | { |
a6a0763a | 1386 | struct cfq_io_context *cic; |
e2d74ac0 | 1387 | struct rb_node *n; |
a6a0763a | 1388 | |
fc46379d | 1389 | ioc->ioprio_changed = 0; |
a6a0763a | 1390 | |
e2d74ac0 JA |
1391 | n = rb_first(&ioc->cic_root); |
1392 | while (n != NULL) { | |
1393 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
3793c65c | 1394 | |
478a82b0 | 1395 | changed_ioprio(cic); |
e2d74ac0 JA |
1396 | n = rb_next(n); |
1397 | } | |
22e2c507 JA |
1398 | } |
1399 | ||
1400 | static struct cfq_queue * | |
6f325a13 | 1401 | cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, |
8267e268 | 1402 | gfp_t gfp_mask) |
22e2c507 JA |
1403 | { |
1404 | const int hashval = hash_long(key, CFQ_QHASH_SHIFT); | |
1405 | struct cfq_queue *cfqq, *new_cfqq = NULL; | |
6f325a13 | 1406 | unsigned short ioprio; |
22e2c507 JA |
1407 | |
1408 | retry: | |
6f325a13 | 1409 | ioprio = tsk->ioprio; |
3b18152c | 1410 | cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval); |
22e2c507 JA |
1411 | |
1412 | if (!cfqq) { | |
1413 | if (new_cfqq) { | |
1414 | cfqq = new_cfqq; | |
1415 | new_cfqq = NULL; | |
1416 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1417 | /* |
1418 | * Inform the allocator of the fact that we will | |
1419 | * just repeat this allocation if it fails, to allow | |
1420 | * the allocator to do whatever it needs to attempt to | |
1421 | * free memory. | |
1422 | */ | |
22e2c507 | 1423 | spin_unlock_irq(cfqd->queue->queue_lock); |
b5deef90 | 1424 | new_cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask|__GFP_NOFAIL, cfqd->queue->node); |
22e2c507 JA |
1425 | spin_lock_irq(cfqd->queue->queue_lock); |
1426 | goto retry; | |
1427 | } else { | |
b5deef90 | 1428 | cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask, cfqd->queue->node); |
22e2c507 JA |
1429 | if (!cfqq) |
1430 | goto out; | |
1431 | } | |
1432 | ||
1433 | memset(cfqq, 0, sizeof(*cfqq)); | |
1434 | ||
1435 | INIT_HLIST_NODE(&cfqq->cfq_hash); | |
1436 | INIT_LIST_HEAD(&cfqq->cfq_list); | |
d9e7620e | 1437 | RB_CLEAR_NODE(&cfqq->rb_node); |
22e2c507 JA |
1438 | INIT_LIST_HEAD(&cfqq->fifo); |
1439 | ||
1440 | cfqq->key = key; | |
1441 | hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]); | |
1442 | atomic_set(&cfqq->ref, 0); | |
1443 | cfqq->cfqd = cfqd; | |
c5b680f3 | 1444 | |
a9938006 JA |
1445 | if (key != CFQ_KEY_ASYNC) |
1446 | cfq_mark_cfqq_idle_window(cfqq); | |
1447 | ||
3b18152c | 1448 | cfq_mark_cfqq_prio_changed(cfqq); |
53b03744 | 1449 | cfq_mark_cfqq_queue_new(cfqq); |
3b18152c | 1450 | cfq_init_prio_data(cfqq); |
22e2c507 JA |
1451 | } |
1452 | ||
1453 | if (new_cfqq) | |
1454 | kmem_cache_free(cfq_pool, new_cfqq); | |
1455 | ||
1456 | atomic_inc(&cfqq->ref); | |
1457 | out: | |
1458 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1459 | return cfqq; | |
1460 | } | |
1461 | ||
498d3aa2 JA |
1462 | /* |
1463 | * We drop cfq io contexts lazily, so we may find a dead one. | |
1464 | */ | |
dbecf3ab OH |
1465 | static void |
1466 | cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic) | |
1467 | { | |
fc46379d | 1468 | WARN_ON(!list_empty(&cic->queue_list)); |
dbecf3ab | 1469 | rb_erase(&cic->rb_node, &ioc->cic_root); |
dbecf3ab | 1470 | kmem_cache_free(cfq_ioc_pool, cic); |
4050cf16 | 1471 | elv_ioc_count_dec(ioc_count); |
dbecf3ab OH |
1472 | } |
1473 | ||
e2d74ac0 JA |
1474 | static struct cfq_io_context * |
1475 | cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc) | |
1476 | { | |
dbecf3ab | 1477 | struct rb_node *n; |
e2d74ac0 | 1478 | struct cfq_io_context *cic; |
be3b0753 | 1479 | void *k, *key = cfqd; |
e2d74ac0 | 1480 | |
dbecf3ab OH |
1481 | restart: |
1482 | n = ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1483 | while (n) { |
1484 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1485 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1486 | k = cic->key; | |
1487 | if (unlikely(!k)) { | |
dbecf3ab OH |
1488 | cfq_drop_dead_cic(ioc, cic); |
1489 | goto restart; | |
1490 | } | |
e2d74ac0 | 1491 | |
be3b0753 | 1492 | if (key < k) |
e2d74ac0 | 1493 | n = n->rb_left; |
be3b0753 | 1494 | else if (key > k) |
e2d74ac0 JA |
1495 | n = n->rb_right; |
1496 | else | |
1497 | return cic; | |
1498 | } | |
1499 | ||
1500 | return NULL; | |
1501 | } | |
1502 | ||
1503 | static inline void | |
1504 | cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, | |
1505 | struct cfq_io_context *cic) | |
1506 | { | |
dbecf3ab OH |
1507 | struct rb_node **p; |
1508 | struct rb_node *parent; | |
e2d74ac0 | 1509 | struct cfq_io_context *__cic; |
0261d688 | 1510 | unsigned long flags; |
be3b0753 | 1511 | void *k; |
e2d74ac0 | 1512 | |
e2d74ac0 JA |
1513 | cic->ioc = ioc; |
1514 | cic->key = cfqd; | |
1515 | ||
dbecf3ab OH |
1516 | restart: |
1517 | parent = NULL; | |
1518 | p = &ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1519 | while (*p) { |
1520 | parent = *p; | |
1521 | __cic = rb_entry(parent, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1522 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1523 | k = __cic->key; | |
1524 | if (unlikely(!k)) { | |
be33c3a6 | 1525 | cfq_drop_dead_cic(ioc, __cic); |
dbecf3ab OH |
1526 | goto restart; |
1527 | } | |
e2d74ac0 | 1528 | |
be3b0753 | 1529 | if (cic->key < k) |
e2d74ac0 | 1530 | p = &(*p)->rb_left; |
be3b0753 | 1531 | else if (cic->key > k) |
e2d74ac0 JA |
1532 | p = &(*p)->rb_right; |
1533 | else | |
1534 | BUG(); | |
1535 | } | |
1536 | ||
1537 | rb_link_node(&cic->rb_node, parent, p); | |
1538 | rb_insert_color(&cic->rb_node, &ioc->cic_root); | |
fc46379d | 1539 | |
0261d688 | 1540 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
e2d74ac0 | 1541 | list_add(&cic->queue_list, &cfqd->cic_list); |
0261d688 | 1542 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
e2d74ac0 JA |
1543 | } |
1544 | ||
1da177e4 LT |
1545 | /* |
1546 | * Setup general io context and cfq io context. There can be several cfq | |
1547 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1548 | * than one device managed by cfq. |
1da177e4 LT |
1549 | */ |
1550 | static struct cfq_io_context * | |
e2d74ac0 | 1551 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1552 | { |
22e2c507 | 1553 | struct io_context *ioc = NULL; |
1da177e4 | 1554 | struct cfq_io_context *cic; |
1da177e4 | 1555 | |
22e2c507 | 1556 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1557 | |
b5deef90 | 1558 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1559 | if (!ioc) |
1560 | return NULL; | |
1561 | ||
e2d74ac0 JA |
1562 | cic = cfq_cic_rb_lookup(cfqd, ioc); |
1563 | if (cic) | |
1564 | goto out; | |
1da177e4 | 1565 | |
e2d74ac0 JA |
1566 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1567 | if (cic == NULL) | |
1568 | goto err; | |
1da177e4 | 1569 | |
e2d74ac0 | 1570 | cfq_cic_link(cfqd, ioc, cic); |
1da177e4 | 1571 | out: |
fc46379d JA |
1572 | smp_read_barrier_depends(); |
1573 | if (unlikely(ioc->ioprio_changed)) | |
1574 | cfq_ioc_set_ioprio(ioc); | |
1575 | ||
1da177e4 LT |
1576 | return cic; |
1577 | err: | |
1578 | put_io_context(ioc); | |
1579 | return NULL; | |
1580 | } | |
1581 | ||
22e2c507 JA |
1582 | static void |
1583 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1584 | { |
aaf1228d JA |
1585 | unsigned long elapsed = jiffies - cic->last_end_request; |
1586 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 1587 | |
22e2c507 JA |
1588 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1589 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1590 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1591 | } | |
1da177e4 | 1592 | |
206dc69b | 1593 | static void |
6d048f53 JA |
1594 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic, |
1595 | struct request *rq) | |
206dc69b JA |
1596 | { |
1597 | sector_t sdist; | |
1598 | u64 total; | |
1599 | ||
5e705374 JA |
1600 | if (cic->last_request_pos < rq->sector) |
1601 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1602 | else |
5e705374 | 1603 | sdist = cic->last_request_pos - rq->sector; |
206dc69b | 1604 | |
6d048f53 JA |
1605 | if (!cic->seek_samples) { |
1606 | cfqd->new_seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1607 | cfqd->new_seek_mean = cfqd->new_seek_total / 256; | |
1608 | } | |
1609 | ||
206dc69b JA |
1610 | /* |
1611 | * Don't allow the seek distance to get too large from the | |
1612 | * odd fragment, pagein, etc | |
1613 | */ | |
1614 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1615 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1616 | else | |
1617 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1618 | ||
1619 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1620 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1621 | total = cic->seek_total + (cic->seek_samples/2); | |
1622 | do_div(total, cic->seek_samples); | |
1623 | cic->seek_mean = (sector_t)total; | |
1624 | } | |
1da177e4 | 1625 | |
22e2c507 JA |
1626 | /* |
1627 | * Disable idle window if the process thinks too long or seeks so much that | |
1628 | * it doesn't matter | |
1629 | */ | |
1630 | static void | |
1631 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1632 | struct cfq_io_context *cic) | |
1633 | { | |
3b18152c | 1634 | int enable_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 1635 | |
caaa5f9f JA |
1636 | if (!cic->ioc->task || !cfqd->cfq_slice_idle || |
1637 | (cfqd->hw_tag && CIC_SEEKY(cic))) | |
22e2c507 JA |
1638 | enable_idle = 0; |
1639 | else if (sample_valid(cic->ttime_samples)) { | |
1640 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1641 | enable_idle = 0; | |
1642 | else | |
1643 | enable_idle = 1; | |
1da177e4 LT |
1644 | } |
1645 | ||
3b18152c JA |
1646 | if (enable_idle) |
1647 | cfq_mark_cfqq_idle_window(cfqq); | |
1648 | else | |
1649 | cfq_clear_cfqq_idle_window(cfqq); | |
22e2c507 | 1650 | } |
1da177e4 | 1651 | |
22e2c507 JA |
1652 | /* |
1653 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1654 | * no or if we aren't sure, a 1 will cause a preempt. | |
1655 | */ | |
1656 | static int | |
1657 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1658 | struct request *rq) |
22e2c507 | 1659 | { |
6d048f53 | 1660 | struct cfq_queue *cfqq; |
22e2c507 | 1661 | |
6d048f53 JA |
1662 | cfqq = cfqd->active_queue; |
1663 | if (!cfqq) | |
22e2c507 JA |
1664 | return 0; |
1665 | ||
6d048f53 JA |
1666 | if (cfq_slice_used(cfqq)) |
1667 | return 1; | |
1668 | ||
1669 | if (cfq_class_idle(new_cfqq)) | |
caaa5f9f | 1670 | return 0; |
22e2c507 JA |
1671 | |
1672 | if (cfq_class_idle(cfqq)) | |
1673 | return 1; | |
1e3335de | 1674 | |
374f84ac JA |
1675 | /* |
1676 | * if the new request is sync, but the currently running queue is | |
1677 | * not, let the sync request have priority. | |
1678 | */ | |
5e705374 | 1679 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1680 | return 1; |
1e3335de | 1681 | |
374f84ac JA |
1682 | /* |
1683 | * So both queues are sync. Let the new request get disk time if | |
1684 | * it's a metadata request and the current queue is doing regular IO. | |
1685 | */ | |
1686 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1687 | return 1; | |
22e2c507 | 1688 | |
1e3335de JA |
1689 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
1690 | return 0; | |
1691 | ||
1692 | /* | |
1693 | * if this request is as-good as one we would expect from the | |
1694 | * current cfqq, let it preempt | |
1695 | */ | |
6d048f53 | 1696 | if (cfq_rq_close(cfqd, rq)) |
1e3335de JA |
1697 | return 1; |
1698 | ||
22e2c507 JA |
1699 | return 0; |
1700 | } | |
1701 | ||
1702 | /* | |
1703 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1704 | * let it have half of its nominal slice. | |
1705 | */ | |
1706 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1707 | { | |
3c6bd2f8 | 1708 | cfq_slice_expired(cfqd, 1, 1); |
22e2c507 | 1709 | |
bf572256 JA |
1710 | /* |
1711 | * Put the new queue at the front of the of the current list, | |
1712 | * so we know that it will be selected next. | |
1713 | */ | |
1714 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
d9e7620e JA |
1715 | list_del_init(&cfqq->cfq_list); |
1716 | list_add(&cfqq->cfq_list, &cfqd->cur_rr); | |
bf572256 | 1717 | |
44f7c160 JA |
1718 | cfqq->slice_end = 0; |
1719 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
1720 | } |
1721 | ||
22e2c507 | 1722 | /* |
5e705374 | 1723 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1724 | * something we should do about it |
1725 | */ | |
1726 | static void | |
5e705374 JA |
1727 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1728 | struct request *rq) | |
22e2c507 | 1729 | { |
5e705374 | 1730 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1731 | |
374f84ac JA |
1732 | if (rq_is_meta(rq)) |
1733 | cfqq->meta_pending++; | |
1734 | ||
9c2c38a1 | 1735 | cfq_update_io_thinktime(cfqd, cic); |
6d048f53 | 1736 | cfq_update_io_seektime(cfqd, cic, rq); |
9c2c38a1 JA |
1737 | cfq_update_idle_window(cfqd, cfqq, cic); |
1738 | ||
5e705374 | 1739 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
6d048f53 | 1740 | cfqq->last_request_pos = cic->last_request_pos; |
22e2c507 JA |
1741 | |
1742 | if (cfqq == cfqd->active_queue) { | |
1743 | /* | |
1744 | * if we are waiting for a request for this queue, let it rip | |
1745 | * immediately and flag that we must not expire this queue | |
1746 | * just now | |
1747 | */ | |
3b18152c JA |
1748 | if (cfq_cfqq_wait_request(cfqq)) { |
1749 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 | 1750 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1751 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1752 | } |
5e705374 | 1753 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1754 | /* |
1755 | * not the active queue - expire current slice if it is | |
1756 | * idle and has expired it's mean thinktime or this new queue | |
1757 | * has some old slice time left and is of higher priority | |
1758 | */ | |
1759 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1760 | cfq_mark_cfqq_must_dispatch(cfqq); |
dc72ef4a | 1761 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1762 | } |
1da177e4 LT |
1763 | } |
1764 | ||
b4878f24 | 1765 | static void cfq_insert_request(request_queue_t *q, struct request *rq) |
1da177e4 | 1766 | { |
b4878f24 | 1767 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1768 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1769 | |
1770 | cfq_init_prio_data(cfqq); | |
1da177e4 | 1771 | |
5e705374 | 1772 | cfq_add_rq_rb(rq); |
1da177e4 | 1773 | |
22e2c507 JA |
1774 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1775 | ||
5e705374 | 1776 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1777 | } |
1778 | ||
1da177e4 LT |
1779 | static void cfq_completed_request(request_queue_t *q, struct request *rq) |
1780 | { | |
5e705374 | 1781 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1782 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1783 | const int sync = rq_is_sync(rq); |
b4878f24 | 1784 | unsigned long now; |
1da177e4 | 1785 | |
b4878f24 | 1786 | now = jiffies; |
1da177e4 | 1787 | |
b4878f24 | 1788 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 1789 | WARN_ON(!cfqq->dispatched); |
b4878f24 | 1790 | cfqd->rq_in_driver--; |
6d048f53 | 1791 | cfqq->dispatched--; |
1da177e4 | 1792 | |
b4878f24 JA |
1793 | if (!cfq_class_idle(cfqq)) |
1794 | cfqd->last_end_request = now; | |
3b18152c | 1795 | |
caaa5f9f | 1796 | if (sync) |
5e705374 | 1797 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1798 | |
1799 | /* | |
1800 | * If this is the active queue, check if it needs to be expired, | |
1801 | * or if we want to idle in case it has no pending requests. | |
1802 | */ | |
1803 | if (cfqd->active_queue == cfqq) { | |
44f7c160 JA |
1804 | if (cfq_cfqq_slice_new(cfqq)) { |
1805 | cfq_set_prio_slice(cfqd, cfqq); | |
1806 | cfq_clear_cfqq_slice_new(cfqq); | |
1807 | } | |
1808 | if (cfq_slice_used(cfqq)) | |
3c6bd2f8 | 1809 | cfq_slice_expired(cfqd, 0, 1); |
6d048f53 JA |
1810 | else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1811 | cfq_arm_slice_timer(cfqd); | |
caaa5f9f | 1812 | } |
6d048f53 JA |
1813 | |
1814 | if (!cfqd->rq_in_driver) | |
1815 | cfq_schedule_dispatch(cfqd); | |
1da177e4 LT |
1816 | } |
1817 | ||
22e2c507 JA |
1818 | /* |
1819 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
1820 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
1821 | */ | |
1822 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 1823 | { |
22e2c507 JA |
1824 | if (has_fs_excl()) { |
1825 | /* | |
1826 | * boost idle prio on transactions that would lock out other | |
1827 | * users of the filesystem | |
1828 | */ | |
1829 | if (cfq_class_idle(cfqq)) | |
1830 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1831 | if (cfqq->ioprio > IOPRIO_NORM) | |
1832 | cfqq->ioprio = IOPRIO_NORM; | |
1833 | } else { | |
1834 | /* | |
1835 | * check if we need to unboost the queue | |
1836 | */ | |
1837 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
1838 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
1839 | if (cfqq->ioprio != cfqq->org_ioprio) | |
1840 | cfqq->ioprio = cfqq->org_ioprio; | |
1841 | } | |
22e2c507 | 1842 | } |
1da177e4 | 1843 | |
89850f7e | 1844 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 1845 | { |
3b18152c | 1846 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 1847 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 1848 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 1849 | return ELV_MQUEUE_MUST; |
3b18152c | 1850 | } |
1da177e4 | 1851 | |
22e2c507 | 1852 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
1853 | } |
1854 | ||
cb78b285 | 1855 | static int cfq_may_queue(request_queue_t *q, int rw) |
22e2c507 JA |
1856 | { |
1857 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1858 | struct task_struct *tsk = current; | |
1859 | struct cfq_queue *cfqq; | |
7749a8d4 JA |
1860 | unsigned int key; |
1861 | ||
1862 | key = cfq_queue_pid(tsk, rw, rw & REQ_RW_SYNC); | |
22e2c507 JA |
1863 | |
1864 | /* | |
1865 | * don't force setup of a queue from here, as a call to may_queue | |
1866 | * does not necessarily imply that a request actually will be queued. | |
1867 | * so just lookup a possibly existing queue, or return 'may queue' | |
1868 | * if that fails | |
1869 | */ | |
7749a8d4 | 1870 | cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio); |
22e2c507 JA |
1871 | if (cfqq) { |
1872 | cfq_init_prio_data(cfqq); | |
1873 | cfq_prio_boost(cfqq); | |
1874 | ||
89850f7e | 1875 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
1876 | } |
1877 | ||
1878 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
1879 | } |
1880 | ||
1da177e4 LT |
1881 | /* |
1882 | * queue lock held here | |
1883 | */ | |
bb37b94c | 1884 | static void cfq_put_request(struct request *rq) |
1da177e4 | 1885 | { |
5e705374 | 1886 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1887 | |
5e705374 | 1888 | if (cfqq) { |
22e2c507 | 1889 | const int rw = rq_data_dir(rq); |
1da177e4 | 1890 | |
22e2c507 JA |
1891 | BUG_ON(!cfqq->allocated[rw]); |
1892 | cfqq->allocated[rw]--; | |
1da177e4 | 1893 | |
5e705374 | 1894 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 1895 | |
1da177e4 | 1896 | rq->elevator_private = NULL; |
5e705374 | 1897 | rq->elevator_private2 = NULL; |
1da177e4 | 1898 | |
1da177e4 LT |
1899 | cfq_put_queue(cfqq); |
1900 | } | |
1901 | } | |
1902 | ||
1903 | /* | |
22e2c507 | 1904 | * Allocate cfq data structures associated with this request. |
1da177e4 | 1905 | */ |
22e2c507 | 1906 | static int |
cb78b285 | 1907 | cfq_set_request(request_queue_t *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
1908 | { |
1909 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3b18152c | 1910 | struct task_struct *tsk = current; |
1da177e4 LT |
1911 | struct cfq_io_context *cic; |
1912 | const int rw = rq_data_dir(rq); | |
7749a8d4 JA |
1913 | const int is_sync = rq_is_sync(rq); |
1914 | pid_t key = cfq_queue_pid(tsk, rw, is_sync); | |
22e2c507 | 1915 | struct cfq_queue *cfqq; |
1da177e4 LT |
1916 | unsigned long flags; |
1917 | ||
1918 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1919 | ||
e2d74ac0 | 1920 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 1921 | |
1da177e4 LT |
1922 | spin_lock_irqsave(q->queue_lock, flags); |
1923 | ||
22e2c507 JA |
1924 | if (!cic) |
1925 | goto queue_fail; | |
1926 | ||
12a05732 | 1927 | if (!cic->cfqq[is_sync]) { |
6f325a13 | 1928 | cfqq = cfq_get_queue(cfqd, key, tsk, gfp_mask); |
22e2c507 JA |
1929 | if (!cfqq) |
1930 | goto queue_fail; | |
1da177e4 | 1931 | |
12a05732 | 1932 | cic->cfqq[is_sync] = cfqq; |
22e2c507 | 1933 | } else |
12a05732 | 1934 | cfqq = cic->cfqq[is_sync]; |
1da177e4 LT |
1935 | |
1936 | cfqq->allocated[rw]++; | |
3b18152c | 1937 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 1938 | atomic_inc(&cfqq->ref); |
1da177e4 | 1939 | |
5e705374 | 1940 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 1941 | |
5e705374 JA |
1942 | rq->elevator_private = cic; |
1943 | rq->elevator_private2 = cfqq; | |
1944 | return 0; | |
1da177e4 | 1945 | |
22e2c507 JA |
1946 | queue_fail: |
1947 | if (cic) | |
1948 | put_io_context(cic->ioc); | |
89850f7e | 1949 | |
3b18152c | 1950 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
1951 | spin_unlock_irqrestore(q->queue_lock, flags); |
1952 | return 1; | |
1953 | } | |
1954 | ||
65f27f38 | 1955 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 1956 | { |
65f27f38 DH |
1957 | struct cfq_data *cfqd = |
1958 | container_of(work, struct cfq_data, unplug_work); | |
1959 | request_queue_t *q = cfqd->queue; | |
22e2c507 JA |
1960 | unsigned long flags; |
1961 | ||
1962 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 1963 | blk_start_queueing(q); |
22e2c507 JA |
1964 | spin_unlock_irqrestore(q->queue_lock, flags); |
1965 | } | |
1966 | ||
1967 | /* | |
1968 | * Timer running if the active_queue is currently idling inside its time slice | |
1969 | */ | |
1970 | static void cfq_idle_slice_timer(unsigned long data) | |
1971 | { | |
1972 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1973 | struct cfq_queue *cfqq; | |
1974 | unsigned long flags; | |
3c6bd2f8 | 1975 | int timed_out = 1; |
22e2c507 JA |
1976 | |
1977 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1978 | ||
1979 | if ((cfqq = cfqd->active_queue) != NULL) { | |
3c6bd2f8 JA |
1980 | timed_out = 0; |
1981 | ||
22e2c507 JA |
1982 | /* |
1983 | * expired | |
1984 | */ | |
44f7c160 | 1985 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
1986 | goto expire; |
1987 | ||
1988 | /* | |
1989 | * only expire and reinvoke request handler, if there are | |
1990 | * other queues with pending requests | |
1991 | */ | |
caaa5f9f | 1992 | if (!cfqd->busy_queues) |
22e2c507 | 1993 | goto out_cont; |
22e2c507 JA |
1994 | |
1995 | /* | |
1996 | * not expired and it has a request pending, let it dispatch | |
1997 | */ | |
dd67d051 | 1998 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 1999 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
2000 | goto out_kick; |
2001 | } | |
2002 | } | |
2003 | expire: | |
3c6bd2f8 | 2004 | cfq_slice_expired(cfqd, 0, timed_out); |
22e2c507 | 2005 | out_kick: |
3b18152c | 2006 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
2007 | out_cont: |
2008 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
2009 | } | |
2010 | ||
2011 | /* | |
2012 | * Timer running if an idle class queue is waiting for service | |
2013 | */ | |
2014 | static void cfq_idle_class_timer(unsigned long data) | |
2015 | { | |
2016 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
2017 | unsigned long flags, end; | |
2018 | ||
2019 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
2020 | ||
2021 | /* | |
2022 | * race with a non-idle queue, reset timer | |
2023 | */ | |
2024 | end = cfqd->last_end_request + CFQ_IDLE_GRACE; | |
ae818a38 JA |
2025 | if (!time_after_eq(jiffies, end)) |
2026 | mod_timer(&cfqd->idle_class_timer, end); | |
2027 | else | |
3b18152c | 2028 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
2029 | |
2030 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
2031 | } | |
2032 | ||
3b18152c JA |
2033 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
2034 | { | |
2035 | del_timer_sync(&cfqd->idle_slice_timer); | |
2036 | del_timer_sync(&cfqd->idle_class_timer); | |
2037 | blk_sync_queue(cfqd->queue); | |
2038 | } | |
22e2c507 | 2039 | |
1da177e4 LT |
2040 | static void cfq_exit_queue(elevator_t *e) |
2041 | { | |
22e2c507 | 2042 | struct cfq_data *cfqd = e->elevator_data; |
d9ff4187 | 2043 | request_queue_t *q = cfqd->queue; |
22e2c507 | 2044 | |
3b18152c | 2045 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 2046 | |
d9ff4187 | 2047 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 2048 | |
d9ff4187 | 2049 | if (cfqd->active_queue) |
3c6bd2f8 | 2050 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0, 0); |
e2d74ac0 JA |
2051 | |
2052 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
2053 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
2054 | struct cfq_io_context, | |
2055 | queue_list); | |
89850f7e JA |
2056 | |
2057 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 2058 | } |
e2d74ac0 | 2059 | |
d9ff4187 | 2060 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
2061 | |
2062 | cfq_shutdown_timer_wq(cfqd); | |
2063 | ||
a90d742e AV |
2064 | kfree(cfqd->cfq_hash); |
2065 | kfree(cfqd); | |
1da177e4 LT |
2066 | } |
2067 | ||
bb37b94c | 2068 | static void *cfq_init_queue(request_queue_t *q) |
1da177e4 LT |
2069 | { |
2070 | struct cfq_data *cfqd; | |
2071 | int i; | |
2072 | ||
b5deef90 | 2073 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node); |
1da177e4 | 2074 | if (!cfqd) |
bc1c1169 | 2075 | return NULL; |
1da177e4 LT |
2076 | |
2077 | memset(cfqd, 0, sizeof(*cfqd)); | |
22e2c507 | 2078 | |
cc09e299 | 2079 | cfqd->service_tree = CFQ_RB_ROOT; |
22e2c507 | 2080 | INIT_LIST_HEAD(&cfqd->cur_rr); |
d9ff4187 | 2081 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2082 | |
b5deef90 | 2083 | cfqd->cfq_hash = kmalloc_node(sizeof(struct hlist_head) * CFQ_QHASH_ENTRIES, GFP_KERNEL, q->node); |
1da177e4 | 2084 | if (!cfqd->cfq_hash) |
5e705374 | 2085 | goto out_free; |
1da177e4 | 2086 | |
1da177e4 LT |
2087 | for (i = 0; i < CFQ_QHASH_ENTRIES; i++) |
2088 | INIT_HLIST_HEAD(&cfqd->cfq_hash[i]); | |
2089 | ||
1da177e4 | 2090 | cfqd->queue = q; |
1da177e4 | 2091 | |
22e2c507 JA |
2092 | init_timer(&cfqd->idle_slice_timer); |
2093 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2094 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2095 | ||
2096 | init_timer(&cfqd->idle_class_timer); | |
2097 | cfqd->idle_class_timer.function = cfq_idle_class_timer; | |
2098 | cfqd->idle_class_timer.data = (unsigned long) cfqd; | |
2099 | ||
65f27f38 | 2100 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2101 | |
1da177e4 | 2102 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2103 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2104 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2105 | cfqd->cfq_back_max = cfq_back_max; |
2106 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2107 | cfqd->cfq_slice[0] = cfq_slice_async; |
2108 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2109 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2110 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
3b18152c | 2111 | |
bc1c1169 | 2112 | return cfqd; |
5e705374 | 2113 | out_free: |
1da177e4 | 2114 | kfree(cfqd); |
bc1c1169 | 2115 | return NULL; |
1da177e4 LT |
2116 | } |
2117 | ||
2118 | static void cfq_slab_kill(void) | |
2119 | { | |
1da177e4 LT |
2120 | if (cfq_pool) |
2121 | kmem_cache_destroy(cfq_pool); | |
2122 | if (cfq_ioc_pool) | |
2123 | kmem_cache_destroy(cfq_ioc_pool); | |
2124 | } | |
2125 | ||
2126 | static int __init cfq_slab_setup(void) | |
2127 | { | |
1da177e4 LT |
2128 | cfq_pool = kmem_cache_create("cfq_pool", sizeof(struct cfq_queue), 0, 0, |
2129 | NULL, NULL); | |
2130 | if (!cfq_pool) | |
2131 | goto fail; | |
2132 | ||
2133 | cfq_ioc_pool = kmem_cache_create("cfq_ioc_pool", | |
2134 | sizeof(struct cfq_io_context), 0, 0, NULL, NULL); | |
2135 | if (!cfq_ioc_pool) | |
2136 | goto fail; | |
2137 | ||
2138 | return 0; | |
2139 | fail: | |
2140 | cfq_slab_kill(); | |
2141 | return -ENOMEM; | |
2142 | } | |
2143 | ||
1da177e4 LT |
2144 | /* |
2145 | * sysfs parts below --> | |
2146 | */ | |
1da177e4 LT |
2147 | static ssize_t |
2148 | cfq_var_show(unsigned int var, char *page) | |
2149 | { | |
2150 | return sprintf(page, "%d\n", var); | |
2151 | } | |
2152 | ||
2153 | static ssize_t | |
2154 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2155 | { | |
2156 | char *p = (char *) page; | |
2157 | ||
2158 | *var = simple_strtoul(p, &p, 10); | |
2159 | return count; | |
2160 | } | |
2161 | ||
1da177e4 | 2162 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
3d1ab40f | 2163 | static ssize_t __FUNC(elevator_t *e, char *page) \ |
1da177e4 | 2164 | { \ |
3d1ab40f | 2165 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2166 | unsigned int __data = __VAR; \ |
2167 | if (__CONV) \ | |
2168 | __data = jiffies_to_msecs(__data); \ | |
2169 | return cfq_var_show(__data, (page)); \ | |
2170 | } | |
2171 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2172 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2173 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2174 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2175 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2176 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2177 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2178 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2179 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2180 | #undef SHOW_FUNCTION |
2181 | ||
2182 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
3d1ab40f | 2183 | static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \ |
1da177e4 | 2184 | { \ |
3d1ab40f | 2185 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2186 | unsigned int __data; \ |
2187 | int ret = cfq_var_store(&__data, (page), count); \ | |
2188 | if (__data < (MIN)) \ | |
2189 | __data = (MIN); \ | |
2190 | else if (__data > (MAX)) \ | |
2191 | __data = (MAX); \ | |
2192 | if (__CONV) \ | |
2193 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2194 | else \ | |
2195 | *(__PTR) = __data; \ | |
2196 | return ret; \ | |
2197 | } | |
2198 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
22e2c507 JA |
2199 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1); |
2200 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1); | |
e572ec7e AV |
2201 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
2202 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0); | |
22e2c507 JA |
2203 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2204 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2205 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
2206 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0); | |
1da177e4 LT |
2207 | #undef STORE_FUNCTION |
2208 | ||
e572ec7e AV |
2209 | #define CFQ_ATTR(name) \ |
2210 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2211 | ||
2212 | static struct elv_fs_entry cfq_attrs[] = { | |
2213 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2214 | CFQ_ATTR(fifo_expire_sync), |
2215 | CFQ_ATTR(fifo_expire_async), | |
2216 | CFQ_ATTR(back_seek_max), | |
2217 | CFQ_ATTR(back_seek_penalty), | |
2218 | CFQ_ATTR(slice_sync), | |
2219 | CFQ_ATTR(slice_async), | |
2220 | CFQ_ATTR(slice_async_rq), | |
2221 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2222 | __ATTR_NULL |
1da177e4 LT |
2223 | }; |
2224 | ||
1da177e4 LT |
2225 | static struct elevator_type iosched_cfq = { |
2226 | .ops = { | |
2227 | .elevator_merge_fn = cfq_merge, | |
2228 | .elevator_merged_fn = cfq_merged_request, | |
2229 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2230 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2231 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2232 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2233 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2234 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2235 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2236 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2237 | .elevator_former_req_fn = elv_rb_former_request, |
2238 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2239 | .elevator_set_req_fn = cfq_set_request, |
2240 | .elevator_put_req_fn = cfq_put_request, | |
2241 | .elevator_may_queue_fn = cfq_may_queue, | |
2242 | .elevator_init_fn = cfq_init_queue, | |
2243 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2244 | .trim = cfq_free_io_context, |
1da177e4 | 2245 | }, |
3d1ab40f | 2246 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2247 | .elevator_name = "cfq", |
2248 | .elevator_owner = THIS_MODULE, | |
2249 | }; | |
2250 | ||
2251 | static int __init cfq_init(void) | |
2252 | { | |
2253 | int ret; | |
2254 | ||
22e2c507 JA |
2255 | /* |
2256 | * could be 0 on HZ < 1000 setups | |
2257 | */ | |
2258 | if (!cfq_slice_async) | |
2259 | cfq_slice_async = 1; | |
2260 | if (!cfq_slice_idle) | |
2261 | cfq_slice_idle = 1; | |
2262 | ||
1da177e4 LT |
2263 | if (cfq_slab_setup()) |
2264 | return -ENOMEM; | |
2265 | ||
2266 | ret = elv_register(&iosched_cfq); | |
22e2c507 JA |
2267 | if (ret) |
2268 | cfq_slab_kill(); | |
1da177e4 | 2269 | |
1da177e4 LT |
2270 | return ret; |
2271 | } | |
2272 | ||
2273 | static void __exit cfq_exit(void) | |
2274 | { | |
6e9a4738 | 2275 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2276 | elv_unregister(&iosched_cfq); |
334e94de | 2277 | ioc_gone = &all_gone; |
fba82272 OH |
2278 | /* ioc_gone's update must be visible before reading ioc_count */ |
2279 | smp_wmb(); | |
4050cf16 | 2280 | if (elv_ioc_count_read(ioc_count)) |
fba82272 | 2281 | wait_for_completion(ioc_gone); |
334e94de | 2282 | synchronize_rcu(); |
83521d3e | 2283 | cfq_slab_kill(); |
1da177e4 LT |
2284 | } |
2285 | ||
2286 | module_init(cfq_init); | |
2287 | module_exit(cfq_exit); | |
2288 | ||
2289 | MODULE_AUTHOR("Jens Axboe"); | |
2290 | MODULE_LICENSE("GPL"); | |
2291 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |