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