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 | 12 | #include <linux/rbtree.h> |
22e2c507 | 13 | #include <linux/ioprio.h> |
1da177e4 LT |
14 | |
15 | /* | |
16 | * tunables | |
17 | */ | |
64100099 | 18 | static const int cfq_quantum = 4; /* max queue in one round of service */ |
64100099 AV |
19 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
20 | static const int cfq_back_max = 16 * 1024; /* maximum backwards seek, in KiB */ | |
21 | static const int cfq_back_penalty = 2; /* penalty of a backwards seek */ | |
1da177e4 | 22 | |
64100099 | 23 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 24 | static int cfq_slice_async = HZ / 25; |
64100099 | 25 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 26 | static int cfq_slice_idle = HZ / 125; |
22e2c507 | 27 | |
d9e7620e JA |
28 | /* |
29 | * grace period before allowing idle class to get disk access | |
30 | */ | |
22e2c507 | 31 | #define CFQ_IDLE_GRACE (HZ / 10) |
d9e7620e JA |
32 | |
33 | /* | |
34 | * below this threshold, we consider thinktime immediate | |
35 | */ | |
36 | #define CFQ_MIN_TT (2) | |
37 | ||
22e2c507 JA |
38 | #define CFQ_SLICE_SCALE (5) |
39 | ||
5e705374 JA |
40 | #define RQ_CIC(rq) ((struct cfq_io_context*)(rq)->elevator_private) |
41 | #define RQ_CFQQ(rq) ((rq)->elevator_private2) | |
1da177e4 | 42 | |
e18b890b CL |
43 | static struct kmem_cache *cfq_pool; |
44 | static struct kmem_cache *cfq_ioc_pool; | |
1da177e4 | 45 | |
4050cf16 | 46 | static DEFINE_PER_CPU(unsigned long, ioc_count); |
334e94de AV |
47 | static struct completion *ioc_gone; |
48 | ||
22e2c507 JA |
49 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
50 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
51 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
52 | ||
3b18152c JA |
53 | #define ASYNC (0) |
54 | #define SYNC (1) | |
55 | ||
206dc69b JA |
56 | #define sample_valid(samples) ((samples) > 80) |
57 | ||
cc09e299 JA |
58 | /* |
59 | * Most of our rbtree usage is for sorting with min extraction, so | |
60 | * if we cache the leftmost node we don't have to walk down the tree | |
61 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
62 | * move this into the elevator for the rq sorting as well. | |
63 | */ | |
64 | struct cfq_rb_root { | |
65 | struct rb_root rb; | |
66 | struct rb_node *left; | |
67 | }; | |
68 | #define CFQ_RB_ROOT (struct cfq_rb_root) { RB_ROOT, NULL, } | |
69 | ||
22e2c507 JA |
70 | /* |
71 | * Per block device queue structure | |
72 | */ | |
1da177e4 | 73 | struct cfq_data { |
22e2c507 JA |
74 | request_queue_t *queue; |
75 | ||
76 | /* | |
77 | * rr list of queues with requests and the count of them | |
78 | */ | |
cc09e299 | 79 | struct cfq_rb_root service_tree; |
22e2c507 JA |
80 | unsigned int busy_queues; |
81 | ||
22e2c507 | 82 | int rq_in_driver; |
3ed9a296 | 83 | int sync_flight; |
25776e35 | 84 | int hw_tag; |
1da177e4 | 85 | |
22e2c507 JA |
86 | /* |
87 | * idle window management | |
88 | */ | |
89 | struct timer_list idle_slice_timer; | |
90 | struct work_struct unplug_work; | |
1da177e4 | 91 | |
22e2c507 JA |
92 | struct cfq_queue *active_queue; |
93 | struct cfq_io_context *active_cic; | |
22e2c507 JA |
94 | |
95 | struct timer_list idle_class_timer; | |
1da177e4 | 96 | |
6d048f53 | 97 | sector_t last_position; |
22e2c507 | 98 | unsigned long last_end_request; |
1da177e4 | 99 | |
1da177e4 LT |
100 | /* |
101 | * tunables, see top of file | |
102 | */ | |
103 | unsigned int cfq_quantum; | |
22e2c507 | 104 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
105 | unsigned int cfq_back_penalty; |
106 | unsigned int cfq_back_max; | |
22e2c507 JA |
107 | unsigned int cfq_slice[2]; |
108 | unsigned int cfq_slice_async_rq; | |
109 | unsigned int cfq_slice_idle; | |
d9ff4187 AV |
110 | |
111 | struct list_head cic_list; | |
6d048f53 JA |
112 | |
113 | sector_t new_seek_mean; | |
114 | u64 new_seek_total; | |
1da177e4 LT |
115 | }; |
116 | ||
22e2c507 JA |
117 | /* |
118 | * Per process-grouping structure | |
119 | */ | |
1da177e4 LT |
120 | struct cfq_queue { |
121 | /* reference count */ | |
122 | atomic_t ref; | |
123 | /* parent cfq_data */ | |
124 | struct cfq_data *cfqd; | |
d9e7620e JA |
125 | /* service_tree member */ |
126 | struct rb_node rb_node; | |
127 | /* service_tree key */ | |
128 | unsigned long rb_key; | |
1da177e4 LT |
129 | /* sorted list of pending requests */ |
130 | struct rb_root sort_list; | |
131 | /* if fifo isn't expired, next request to serve */ | |
5e705374 | 132 | struct request *next_rq; |
1da177e4 LT |
133 | /* requests queued in sort_list */ |
134 | int queued[2]; | |
135 | /* currently allocated requests */ | |
136 | int allocated[2]; | |
374f84ac JA |
137 | /* pending metadata requests */ |
138 | int meta_pending; | |
1da177e4 | 139 | /* fifo list of requests in sort_list */ |
22e2c507 | 140 | struct list_head fifo; |
1da177e4 | 141 | |
22e2c507 | 142 | unsigned long slice_end; |
c5b680f3 | 143 | long slice_resid; |
1da177e4 | 144 | |
6d048f53 JA |
145 | /* number of requests that are on the dispatch list or inside driver */ |
146 | int dispatched; | |
22e2c507 JA |
147 | |
148 | /* io prio of this group */ | |
149 | unsigned short ioprio, org_ioprio; | |
150 | unsigned short ioprio_class, org_ioprio_class; | |
151 | ||
3b18152c JA |
152 | /* various state flags, see below */ |
153 | unsigned int flags; | |
6d048f53 JA |
154 | |
155 | sector_t last_request_pos; | |
1da177e4 LT |
156 | }; |
157 | ||
3b18152c | 158 | enum cfqq_state_flags { |
b0b8d749 JA |
159 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
160 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
161 | CFQ_CFQQ_FLAG_must_alloc, /* must be allowed rq alloc */ | |
162 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ | |
163 | CFQ_CFQQ_FLAG_must_dispatch, /* must dispatch, even if expired */ | |
164 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ | |
165 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
166 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
167 | CFQ_CFQQ_FLAG_queue_new, /* queue never been serviced */ | |
44f7c160 | 168 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 169 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
3b18152c JA |
170 | }; |
171 | ||
172 | #define CFQ_CFQQ_FNS(name) \ | |
173 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
174 | { \ | |
175 | cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ | |
176 | } \ | |
177 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
178 | { \ | |
179 | cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ | |
180 | } \ | |
181 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
182 | { \ | |
183 | return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ | |
184 | } | |
185 | ||
186 | CFQ_CFQQ_FNS(on_rr); | |
187 | CFQ_CFQQ_FNS(wait_request); | |
188 | CFQ_CFQQ_FNS(must_alloc); | |
189 | CFQ_CFQQ_FNS(must_alloc_slice); | |
190 | CFQ_CFQQ_FNS(must_dispatch); | |
191 | CFQ_CFQQ_FNS(fifo_expire); | |
192 | CFQ_CFQQ_FNS(idle_window); | |
193 | CFQ_CFQQ_FNS(prio_changed); | |
53b03744 | 194 | CFQ_CFQQ_FNS(queue_new); |
44f7c160 | 195 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 196 | CFQ_CFQQ_FNS(sync); |
3b18152c JA |
197 | #undef CFQ_CFQQ_FNS |
198 | ||
5e705374 | 199 | static void cfq_dispatch_insert(request_queue_t *, struct request *); |
91fac317 VT |
200 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, int, |
201 | struct task_struct *, gfp_t); | |
202 | static struct cfq_io_context *cfq_cic_rb_lookup(struct cfq_data *, | |
203 | struct io_context *); | |
204 | ||
205 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, | |
206 | int is_sync) | |
207 | { | |
208 | return cic->cfqq[!!is_sync]; | |
209 | } | |
210 | ||
211 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
212 | struct cfq_queue *cfqq, int is_sync) | |
213 | { | |
214 | cic->cfqq[!!is_sync] = cfqq; | |
215 | } | |
216 | ||
217 | /* | |
218 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
219 | * set (in which case it could also be direct WRITE). | |
220 | */ | |
221 | static inline int cfq_bio_sync(struct bio *bio) | |
222 | { | |
223 | if (bio_data_dir(bio) == READ || bio_sync(bio)) | |
224 | return 1; | |
225 | ||
226 | return 0; | |
227 | } | |
1da177e4 | 228 | |
99f95e52 AM |
229 | /* |
230 | * scheduler run of queue, if there are requests pending and no one in the | |
231 | * driver that will restart queueing | |
232 | */ | |
233 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
234 | { | |
7b14e3b5 | 235 | if (cfqd->busy_queues) |
99f95e52 AM |
236 | kblockd_schedule_work(&cfqd->unplug_work); |
237 | } | |
238 | ||
239 | static int cfq_queue_empty(request_queue_t *q) | |
240 | { | |
241 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
242 | ||
b4878f24 | 243 | return !cfqd->busy_queues; |
99f95e52 AM |
244 | } |
245 | ||
44f7c160 JA |
246 | /* |
247 | * Scale schedule slice based on io priority. Use the sync time slice only | |
248 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
249 | * io only, should not get full sync slice length. | |
250 | */ | |
d9e7620e JA |
251 | static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync, |
252 | unsigned short prio) | |
44f7c160 | 253 | { |
d9e7620e | 254 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 255 | |
d9e7620e JA |
256 | WARN_ON(prio >= IOPRIO_BE_NR); |
257 | ||
258 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
259 | } | |
44f7c160 | 260 | |
d9e7620e JA |
261 | static inline int |
262 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
263 | { | |
264 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
265 | } |
266 | ||
267 | static inline void | |
268 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
269 | { | |
270 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; | |
271 | } | |
272 | ||
273 | /* | |
274 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
275 | * isn't valid until the first request from the dispatch is activated | |
276 | * and the slice time set. | |
277 | */ | |
278 | static inline int cfq_slice_used(struct cfq_queue *cfqq) | |
279 | { | |
280 | if (cfq_cfqq_slice_new(cfqq)) | |
281 | return 0; | |
282 | if (time_before(jiffies, cfqq->slice_end)) | |
283 | return 0; | |
284 | ||
285 | return 1; | |
286 | } | |
287 | ||
1da177e4 | 288 | /* |
5e705374 | 289 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 290 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 291 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 292 | */ |
5e705374 JA |
293 | static struct request * |
294 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
295 | { |
296 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 297 | unsigned long back_max; |
e8a99053 AM |
298 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
299 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
300 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 301 | |
5e705374 JA |
302 | if (rq1 == NULL || rq1 == rq2) |
303 | return rq2; | |
304 | if (rq2 == NULL) | |
305 | return rq1; | |
9c2c38a1 | 306 | |
5e705374 JA |
307 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
308 | return rq1; | |
309 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
310 | return rq2; | |
374f84ac JA |
311 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
312 | return rq1; | |
313 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
314 | return rq2; | |
1da177e4 | 315 | |
5e705374 JA |
316 | s1 = rq1->sector; |
317 | s2 = rq2->sector; | |
1da177e4 | 318 | |
6d048f53 | 319 | last = cfqd->last_position; |
1da177e4 | 320 | |
1da177e4 LT |
321 | /* |
322 | * by definition, 1KiB is 2 sectors | |
323 | */ | |
324 | back_max = cfqd->cfq_back_max * 2; | |
325 | ||
326 | /* | |
327 | * Strict one way elevator _except_ in the case where we allow | |
328 | * short backward seeks which are biased as twice the cost of a | |
329 | * similar forward seek. | |
330 | */ | |
331 | if (s1 >= last) | |
332 | d1 = s1 - last; | |
333 | else if (s1 + back_max >= last) | |
334 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
335 | else | |
e8a99053 | 336 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
337 | |
338 | if (s2 >= last) | |
339 | d2 = s2 - last; | |
340 | else if (s2 + back_max >= last) | |
341 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
342 | else | |
e8a99053 | 343 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
344 | |
345 | /* Found required data */ | |
e8a99053 AM |
346 | |
347 | /* | |
348 | * By doing switch() on the bit mask "wrap" we avoid having to | |
349 | * check two variables for all permutations: --> faster! | |
350 | */ | |
351 | switch (wrap) { | |
5e705374 | 352 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 353 | if (d1 < d2) |
5e705374 | 354 | return rq1; |
e8a99053 | 355 | else if (d2 < d1) |
5e705374 | 356 | return rq2; |
e8a99053 AM |
357 | else { |
358 | if (s1 >= s2) | |
5e705374 | 359 | return rq1; |
e8a99053 | 360 | else |
5e705374 | 361 | return rq2; |
e8a99053 | 362 | } |
1da177e4 | 363 | |
e8a99053 | 364 | case CFQ_RQ2_WRAP: |
5e705374 | 365 | return rq1; |
e8a99053 | 366 | case CFQ_RQ1_WRAP: |
5e705374 JA |
367 | return rq2; |
368 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
369 | default: |
370 | /* | |
371 | * Since both rqs are wrapped, | |
372 | * start with the one that's further behind head | |
373 | * (--> only *one* back seek required), | |
374 | * since back seek takes more time than forward. | |
375 | */ | |
376 | if (s1 <= s2) | |
5e705374 | 377 | return rq1; |
1da177e4 | 378 | else |
5e705374 | 379 | return rq2; |
1da177e4 LT |
380 | } |
381 | } | |
382 | ||
498d3aa2 JA |
383 | /* |
384 | * The below is leftmost cache rbtree addon | |
385 | */ | |
cc09e299 JA |
386 | static struct rb_node *cfq_rb_first(struct cfq_rb_root *root) |
387 | { | |
388 | if (!root->left) | |
389 | root->left = rb_first(&root->rb); | |
390 | ||
391 | return root->left; | |
392 | } | |
393 | ||
394 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) | |
395 | { | |
396 | if (root->left == n) | |
397 | root->left = NULL; | |
398 | ||
399 | rb_erase(n, &root->rb); | |
400 | RB_CLEAR_NODE(n); | |
401 | } | |
402 | ||
1da177e4 LT |
403 | /* |
404 | * would be nice to take fifo expire time into account as well | |
405 | */ | |
5e705374 JA |
406 | static struct request * |
407 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
408 | struct request *last) | |
1da177e4 | 409 | { |
21183b07 JA |
410 | struct rb_node *rbnext = rb_next(&last->rb_node); |
411 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 412 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 413 | |
21183b07 | 414 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
415 | |
416 | if (rbprev) | |
5e705374 | 417 | prev = rb_entry_rq(rbprev); |
1da177e4 | 418 | |
21183b07 | 419 | if (rbnext) |
5e705374 | 420 | next = rb_entry_rq(rbnext); |
21183b07 JA |
421 | else { |
422 | rbnext = rb_first(&cfqq->sort_list); | |
423 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 424 | next = rb_entry_rq(rbnext); |
21183b07 | 425 | } |
1da177e4 | 426 | |
21183b07 | 427 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
428 | } |
429 | ||
d9e7620e JA |
430 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
431 | struct cfq_queue *cfqq) | |
1da177e4 | 432 | { |
d9e7620e JA |
433 | /* |
434 | * just an approximation, should be ok. | |
435 | */ | |
67e6b49e JA |
436 | return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
437 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); | |
d9e7620e JA |
438 | } |
439 | ||
498d3aa2 JA |
440 | /* |
441 | * The cfqd->service_tree holds all pending cfq_queue's that have | |
442 | * requests waiting to be processed. It is sorted in the order that | |
443 | * we will service the queues. | |
444 | */ | |
d9e7620e | 445 | static void cfq_service_tree_add(struct cfq_data *cfqd, |
edd75ffd | 446 | struct cfq_queue *cfqq, int add_front) |
d9e7620e | 447 | { |
cc09e299 | 448 | struct rb_node **p = &cfqd->service_tree.rb.rb_node; |
d9e7620e | 449 | struct rb_node *parent = NULL; |
d9e7620e | 450 | unsigned long rb_key; |
498d3aa2 | 451 | int left; |
d9e7620e | 452 | |
edd75ffd JA |
453 | if (!add_front) { |
454 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; | |
455 | rb_key += cfqq->slice_resid; | |
456 | cfqq->slice_resid = 0; | |
457 | } else | |
458 | rb_key = 0; | |
1da177e4 | 459 | |
d9e7620e | 460 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
99f9628a | 461 | /* |
d9e7620e | 462 | * same position, nothing more to do |
99f9628a | 463 | */ |
d9e7620e JA |
464 | if (rb_key == cfqq->rb_key) |
465 | return; | |
1da177e4 | 466 | |
cc09e299 | 467 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); |
1da177e4 | 468 | } |
d9e7620e | 469 | |
498d3aa2 | 470 | left = 1; |
d9e7620e | 471 | while (*p) { |
cc09e299 | 472 | struct cfq_queue *__cfqq; |
67060e37 | 473 | struct rb_node **n; |
cc09e299 | 474 | |
d9e7620e JA |
475 | parent = *p; |
476 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
477 | ||
0c534e0a JA |
478 | /* |
479 | * sort RT queues first, we always want to give | |
67060e37 JA |
480 | * preference to them. IDLE queues goes to the back. |
481 | * after that, sort on the next service time. | |
0c534e0a JA |
482 | */ |
483 | if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq)) | |
67060e37 | 484 | n = &(*p)->rb_left; |
0c534e0a | 485 | else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq)) |
67060e37 JA |
486 | n = &(*p)->rb_right; |
487 | else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq)) | |
488 | n = &(*p)->rb_left; | |
489 | else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq)) | |
490 | n = &(*p)->rb_right; | |
0c534e0a | 491 | else if (rb_key < __cfqq->rb_key) |
67060e37 JA |
492 | n = &(*p)->rb_left; |
493 | else | |
494 | n = &(*p)->rb_right; | |
495 | ||
496 | if (n == &(*p)->rb_right) | |
cc09e299 | 497 | left = 0; |
67060e37 JA |
498 | |
499 | p = n; | |
d9e7620e JA |
500 | } |
501 | ||
cc09e299 JA |
502 | if (left) |
503 | cfqd->service_tree.left = &cfqq->rb_node; | |
504 | ||
d9e7620e JA |
505 | cfqq->rb_key = rb_key; |
506 | rb_link_node(&cfqq->rb_node, parent, p); | |
cc09e299 | 507 | rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb); |
1da177e4 LT |
508 | } |
509 | ||
498d3aa2 JA |
510 | /* |
511 | * Update cfqq's position in the service tree. | |
512 | */ | |
edd75ffd | 513 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 514 | { |
6d048f53 JA |
515 | /* |
516 | * Resorting requires the cfqq to be on the RR list already. | |
517 | */ | |
498d3aa2 | 518 | if (cfq_cfqq_on_rr(cfqq)) |
edd75ffd | 519 | cfq_service_tree_add(cfqd, cfqq, 0); |
6d048f53 JA |
520 | } |
521 | ||
1da177e4 LT |
522 | /* |
523 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 524 | * the pending list according to last request service |
1da177e4 LT |
525 | */ |
526 | static inline void | |
b4878f24 | 527 | cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 528 | { |
3b18152c JA |
529 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
530 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
531 | cfqd->busy_queues++; |
532 | ||
edd75ffd | 533 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
534 | } |
535 | ||
498d3aa2 JA |
536 | /* |
537 | * Called when the cfqq no longer has requests pending, remove it from | |
538 | * the service tree. | |
539 | */ | |
1da177e4 LT |
540 | static inline void |
541 | cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
542 | { | |
3b18152c JA |
543 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
544 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 545 | |
cc09e299 JA |
546 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) |
547 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); | |
d9e7620e | 548 | |
1da177e4 LT |
549 | BUG_ON(!cfqd->busy_queues); |
550 | cfqd->busy_queues--; | |
551 | } | |
552 | ||
553 | /* | |
554 | * rb tree support functions | |
555 | */ | |
5e705374 | 556 | static inline void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 557 | { |
5e705374 | 558 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 559 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 560 | const int sync = rq_is_sync(rq); |
1da177e4 | 561 | |
b4878f24 JA |
562 | BUG_ON(!cfqq->queued[sync]); |
563 | cfqq->queued[sync]--; | |
1da177e4 | 564 | |
5e705374 | 565 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 566 | |
dd67d051 | 567 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 568 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
569 | } |
570 | ||
5e705374 | 571 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 572 | { |
5e705374 | 573 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 574 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 575 | struct request *__alias; |
1da177e4 | 576 | |
5380a101 | 577 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
578 | |
579 | /* | |
580 | * looks a little odd, but the first insert might return an alias. | |
581 | * if that happens, put the alias on the dispatch list | |
582 | */ | |
21183b07 | 583 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 584 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
585 | |
586 | if (!cfq_cfqq_on_rr(cfqq)) | |
587 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
588 | |
589 | /* | |
590 | * check if this request is a better next-serve candidate | |
591 | */ | |
592 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); | |
593 | BUG_ON(!cfqq->next_rq); | |
1da177e4 LT |
594 | } |
595 | ||
596 | static inline void | |
5e705374 | 597 | cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 598 | { |
5380a101 JA |
599 | elv_rb_del(&cfqq->sort_list, rq); |
600 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 601 | cfq_add_rq_rb(rq); |
1da177e4 LT |
602 | } |
603 | ||
206dc69b JA |
604 | static struct request * |
605 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 606 | { |
206dc69b | 607 | struct task_struct *tsk = current; |
91fac317 | 608 | struct cfq_io_context *cic; |
206dc69b | 609 | struct cfq_queue *cfqq; |
1da177e4 | 610 | |
91fac317 VT |
611 | cic = cfq_cic_rb_lookup(cfqd, tsk->io_context); |
612 | if (!cic) | |
613 | return NULL; | |
614 | ||
615 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
616 | if (cfqq) { |
617 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
618 | ||
21183b07 | 619 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 620 | } |
1da177e4 | 621 | |
1da177e4 LT |
622 | return NULL; |
623 | } | |
624 | ||
b4878f24 | 625 | static void cfq_activate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 626 | { |
22e2c507 | 627 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 628 | |
b4878f24 | 629 | cfqd->rq_in_driver++; |
25776e35 JA |
630 | |
631 | /* | |
632 | * If the depth is larger 1, it really could be queueing. But lets | |
633 | * make the mark a little higher - idling could still be good for | |
634 | * low queueing, and a low queueing number could also just indicate | |
635 | * a SCSI mid layer like behaviour where limit+1 is often seen. | |
636 | */ | |
637 | if (!cfqd->hw_tag && cfqd->rq_in_driver > 4) | |
638 | cfqd->hw_tag = 1; | |
6d048f53 JA |
639 | |
640 | cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors; | |
1da177e4 LT |
641 | } |
642 | ||
b4878f24 | 643 | static void cfq_deactivate_request(request_queue_t *q, struct request *rq) |
1da177e4 | 644 | { |
b4878f24 JA |
645 | struct cfq_data *cfqd = q->elevator->elevator_data; |
646 | ||
647 | WARN_ON(!cfqd->rq_in_driver); | |
648 | cfqd->rq_in_driver--; | |
1da177e4 LT |
649 | } |
650 | ||
b4878f24 | 651 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 652 | { |
5e705374 | 653 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 654 | |
5e705374 JA |
655 | if (cfqq->next_rq == rq) |
656 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 657 | |
b4878f24 | 658 | list_del_init(&rq->queuelist); |
5e705374 | 659 | cfq_del_rq_rb(rq); |
374f84ac JA |
660 | |
661 | if (rq_is_meta(rq)) { | |
662 | WARN_ON(!cfqq->meta_pending); | |
663 | cfqq->meta_pending--; | |
664 | } | |
1da177e4 LT |
665 | } |
666 | ||
498d3aa2 | 667 | static int cfq_merge(request_queue_t *q, struct request **req, struct bio *bio) |
1da177e4 LT |
668 | { |
669 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
670 | struct request *__rq; | |
1da177e4 | 671 | |
206dc69b | 672 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 673 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
674 | *req = __rq; |
675 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
676 | } |
677 | ||
678 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
679 | } |
680 | ||
21183b07 JA |
681 | static void cfq_merged_request(request_queue_t *q, struct request *req, |
682 | int type) | |
1da177e4 | 683 | { |
21183b07 | 684 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 685 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 686 | |
5e705374 | 687 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 688 | } |
1da177e4 LT |
689 | } |
690 | ||
691 | static void | |
692 | cfq_merged_requests(request_queue_t *q, struct request *rq, | |
693 | struct request *next) | |
694 | { | |
22e2c507 JA |
695 | /* |
696 | * reposition in fifo if next is older than rq | |
697 | */ | |
698 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
699 | time_before(next->start_time, rq->start_time)) | |
700 | list_move(&rq->queuelist, &next->queuelist); | |
701 | ||
b4878f24 | 702 | cfq_remove_request(next); |
22e2c507 JA |
703 | } |
704 | ||
da775265 JA |
705 | static int cfq_allow_merge(request_queue_t *q, struct request *rq, |
706 | struct bio *bio) | |
707 | { | |
708 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
91fac317 | 709 | struct cfq_io_context *cic; |
da775265 | 710 | struct cfq_queue *cfqq; |
da775265 JA |
711 | |
712 | /* | |
ec8acb69 | 713 | * Disallow merge of a sync bio into an async request. |
da775265 | 714 | */ |
91fac317 | 715 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
da775265 JA |
716 | return 0; |
717 | ||
718 | /* | |
719d3402 JA |
719 | * Lookup the cfqq that this bio will be queued with. Allow |
720 | * merge only if rq is queued there. | |
da775265 | 721 | */ |
91fac317 VT |
722 | cic = cfq_cic_rb_lookup(cfqd, current->io_context); |
723 | if (!cic) | |
724 | return 0; | |
719d3402 | 725 | |
91fac317 | 726 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
719d3402 JA |
727 | if (cfqq == RQ_CFQQ(rq)) |
728 | return 1; | |
da775265 | 729 | |
ec8acb69 | 730 | return 0; |
da775265 JA |
731 | } |
732 | ||
22e2c507 JA |
733 | static inline void |
734 | __cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
735 | { | |
736 | if (cfqq) { | |
737 | /* | |
738 | * stop potential idle class queues waiting service | |
739 | */ | |
740 | del_timer(&cfqd->idle_class_timer); | |
741 | ||
22e2c507 | 742 | cfqq->slice_end = 0; |
3b18152c JA |
743 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
744 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 745 | cfq_mark_cfqq_slice_new(cfqq); |
1afba045 | 746 | cfq_clear_cfqq_queue_new(cfqq); |
22e2c507 JA |
747 | } |
748 | ||
749 | cfqd->active_queue = cfqq; | |
750 | } | |
751 | ||
7b14e3b5 JA |
752 | /* |
753 | * current cfqq expired its slice (or was too idle), select new one | |
754 | */ | |
755 | static void | |
756 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
6084cdda | 757 | int timed_out) |
7b14e3b5 | 758 | { |
7b14e3b5 JA |
759 | if (cfq_cfqq_wait_request(cfqq)) |
760 | del_timer(&cfqd->idle_slice_timer); | |
761 | ||
7b14e3b5 JA |
762 | cfq_clear_cfqq_must_dispatch(cfqq); |
763 | cfq_clear_cfqq_wait_request(cfqq); | |
764 | ||
765 | /* | |
6084cdda | 766 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 767 | */ |
3c6bd2f8 | 768 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) |
c5b680f3 | 769 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b14e3b5 | 770 | |
edd75ffd | 771 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
772 | |
773 | if (cfqq == cfqd->active_queue) | |
774 | cfqd->active_queue = NULL; | |
775 | ||
776 | if (cfqd->active_cic) { | |
777 | put_io_context(cfqd->active_cic->ioc); | |
778 | cfqd->active_cic = NULL; | |
779 | } | |
7b14e3b5 JA |
780 | } |
781 | ||
6084cdda | 782 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out) |
7b14e3b5 JA |
783 | { |
784 | struct cfq_queue *cfqq = cfqd->active_queue; | |
785 | ||
786 | if (cfqq) | |
6084cdda | 787 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
788 | } |
789 | ||
498d3aa2 JA |
790 | /* |
791 | * Get next queue for service. Unless we have a queue preemption, | |
792 | * we'll simply select the first cfqq in the service tree. | |
793 | */ | |
6d048f53 | 794 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 795 | { |
edd75ffd JA |
796 | struct cfq_queue *cfqq; |
797 | struct rb_node *n; | |
22e2c507 | 798 | |
edd75ffd JA |
799 | if (RB_EMPTY_ROOT(&cfqd->service_tree.rb)) |
800 | return NULL; | |
d9e7620e | 801 | |
edd75ffd JA |
802 | n = cfq_rb_first(&cfqd->service_tree); |
803 | cfqq = rb_entry(n, struct cfq_queue, rb_node); | |
498d3aa2 | 804 | |
edd75ffd JA |
805 | if (cfq_class_idle(cfqq)) { |
806 | unsigned long end; | |
807 | ||
808 | /* | |
809 | * if we have idle queues and no rt or be queues had | |
810 | * pending requests, either allow immediate service if | |
811 | * the grace period has passed or arm the idle grace | |
812 | * timer | |
813 | */ | |
814 | end = cfqd->last_end_request + CFQ_IDLE_GRACE; | |
815 | if (time_before(jiffies, end)) { | |
816 | mod_timer(&cfqd->idle_class_timer, end); | |
817 | cfqq = NULL; | |
67060e37 | 818 | } |
22e2c507 JA |
819 | } |
820 | ||
6d048f53 JA |
821 | return cfqq; |
822 | } | |
823 | ||
498d3aa2 JA |
824 | /* |
825 | * Get and set a new active queue for service. | |
826 | */ | |
6d048f53 JA |
827 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
828 | { | |
829 | struct cfq_queue *cfqq; | |
830 | ||
d9e7620e | 831 | cfqq = cfq_get_next_queue(cfqd); |
22e2c507 | 832 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 833 | return cfqq; |
22e2c507 JA |
834 | } |
835 | ||
d9e7620e JA |
836 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
837 | struct request *rq) | |
838 | { | |
839 | if (rq->sector >= cfqd->last_position) | |
840 | return rq->sector - cfqd->last_position; | |
841 | else | |
842 | return cfqd->last_position - rq->sector; | |
843 | } | |
844 | ||
6d048f53 JA |
845 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq) |
846 | { | |
847 | struct cfq_io_context *cic = cfqd->active_cic; | |
848 | ||
849 | if (!sample_valid(cic->seek_samples)) | |
850 | return 0; | |
851 | ||
852 | return cfq_dist_from_last(cfqd, rq) <= cic->seek_mean; | |
853 | } | |
854 | ||
d9e7620e JA |
855 | static int cfq_close_cooperator(struct cfq_data *cfq_data, |
856 | struct cfq_queue *cfqq) | |
6d048f53 | 857 | { |
6d048f53 | 858 | /* |
d9e7620e JA |
859 | * We should notice if some of the queues are cooperating, eg |
860 | * working closely on the same area of the disk. In that case, | |
861 | * we can group them together and don't waste time idling. | |
6d048f53 | 862 | */ |
d9e7620e | 863 | return 0; |
6d048f53 JA |
864 | } |
865 | ||
866 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (8 * 1024)) | |
caaa5f9f | 867 | |
6d048f53 | 868 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 869 | { |
1792669c | 870 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 871 | struct cfq_io_context *cic; |
7b14e3b5 JA |
872 | unsigned long sl; |
873 | ||
dd67d051 | 874 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 875 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
876 | |
877 | /* | |
878 | * idle is disabled, either manually or by past process history | |
879 | */ | |
6d048f53 JA |
880 | if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq)) |
881 | return; | |
882 | ||
22e2c507 JA |
883 | /* |
884 | * task has exited, don't wait | |
885 | */ | |
206dc69b JA |
886 | cic = cfqd->active_cic; |
887 | if (!cic || !cic->ioc->task) | |
6d048f53 JA |
888 | return; |
889 | ||
890 | /* | |
891 | * See if this prio level has a good candidate | |
892 | */ | |
1afba045 JA |
893 | if (cfq_close_cooperator(cfqd, cfqq) && |
894 | (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2)) | |
6d048f53 | 895 | return; |
22e2c507 | 896 | |
3b18152c JA |
897 | cfq_mark_cfqq_must_dispatch(cfqq); |
898 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 899 | |
206dc69b JA |
900 | /* |
901 | * we don't want to idle for seeks, but we do want to allow | |
902 | * fair distribution of slice time for a process doing back-to-back | |
903 | * seeks. so allow a little bit of time for him to submit a new rq | |
904 | */ | |
6d048f53 | 905 | sl = cfqd->cfq_slice_idle; |
caaa5f9f | 906 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
d9e7620e | 907 | sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT)); |
206dc69b | 908 | |
7b14e3b5 | 909 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
1da177e4 LT |
910 | } |
911 | ||
498d3aa2 JA |
912 | /* |
913 | * Move request from internal lists to the request queue dispatch list. | |
914 | */ | |
5e705374 | 915 | static void cfq_dispatch_insert(request_queue_t *q, struct request *rq) |
1da177e4 | 916 | { |
3ed9a296 | 917 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 918 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 919 | |
5380a101 | 920 | cfq_remove_request(rq); |
6d048f53 | 921 | cfqq->dispatched++; |
5380a101 | 922 | elv_dispatch_sort(q, rq); |
3ed9a296 JA |
923 | |
924 | if (cfq_cfqq_sync(cfqq)) | |
925 | cfqd->sync_flight++; | |
1da177e4 LT |
926 | } |
927 | ||
928 | /* | |
929 | * return expired entry, or NULL to just start from scratch in rbtree | |
930 | */ | |
5e705374 | 931 | static inline struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
932 | { |
933 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 934 | struct request *rq; |
89850f7e | 935 | int fifo; |
1da177e4 | 936 | |
3b18152c | 937 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 938 | return NULL; |
cb887411 JA |
939 | |
940 | cfq_mark_cfqq_fifo_expire(cfqq); | |
941 | ||
89850f7e JA |
942 | if (list_empty(&cfqq->fifo)) |
943 | return NULL; | |
1da177e4 | 944 | |
6d048f53 | 945 | fifo = cfq_cfqq_sync(cfqq); |
89850f7e | 946 | rq = rq_entry_fifo(cfqq->fifo.next); |
1da177e4 | 947 | |
6d048f53 JA |
948 | if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) |
949 | return NULL; | |
1da177e4 | 950 | |
6d048f53 | 951 | return rq; |
1da177e4 LT |
952 | } |
953 | ||
22e2c507 JA |
954 | static inline int |
955 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
956 | { | |
957 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 958 | |
22e2c507 | 959 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 960 | |
22e2c507 | 961 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
962 | } |
963 | ||
22e2c507 | 964 | /* |
498d3aa2 JA |
965 | * Select a queue for service. If we have a current active queue, |
966 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 967 | */ |
1b5ed5e1 | 968 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 969 | { |
1da177e4 | 970 | struct cfq_queue *cfqq; |
1da177e4 | 971 | |
22e2c507 JA |
972 | cfqq = cfqd->active_queue; |
973 | if (!cfqq) | |
974 | goto new_queue; | |
1da177e4 | 975 | |
22e2c507 | 976 | /* |
6d048f53 | 977 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 978 | */ |
6d048f53 | 979 | if (cfq_slice_used(cfqq)) |
3b18152c | 980 | goto expire; |
1da177e4 | 981 | |
22e2c507 | 982 | /* |
6d048f53 JA |
983 | * The active queue has requests and isn't expired, allow it to |
984 | * dispatch. | |
22e2c507 | 985 | */ |
dd67d051 | 986 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 987 | goto keep_queue; |
6d048f53 JA |
988 | |
989 | /* | |
990 | * No requests pending. If the active queue still has requests in | |
991 | * flight or is idling for a new request, allow either of these | |
992 | * conditions to happen (or time out) before selecting a new queue. | |
993 | */ | |
cc197479 JA |
994 | if (timer_pending(&cfqd->idle_slice_timer) || |
995 | (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) { | |
caaa5f9f JA |
996 | cfqq = NULL; |
997 | goto keep_queue; | |
22e2c507 JA |
998 | } |
999 | ||
3b18152c | 1000 | expire: |
6084cdda | 1001 | cfq_slice_expired(cfqd, 0); |
3b18152c JA |
1002 | new_queue: |
1003 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 1004 | keep_queue: |
3b18152c | 1005 | return cfqq; |
22e2c507 JA |
1006 | } |
1007 | ||
498d3aa2 JA |
1008 | /* |
1009 | * Dispatch some requests from cfqq, moving them to the request queue | |
1010 | * dispatch list. | |
1011 | */ | |
22e2c507 JA |
1012 | static int |
1013 | __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1014 | int max_dispatch) | |
1015 | { | |
1016 | int dispatched = 0; | |
1017 | ||
dd67d051 | 1018 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
1019 | |
1020 | do { | |
5e705374 | 1021 | struct request *rq; |
1da177e4 LT |
1022 | |
1023 | /* | |
22e2c507 | 1024 | * follow expired path, else get first next available |
1da177e4 | 1025 | */ |
5e705374 JA |
1026 | if ((rq = cfq_check_fifo(cfqq)) == NULL) |
1027 | rq = cfqq->next_rq; | |
22e2c507 JA |
1028 | |
1029 | /* | |
1030 | * finally, insert request into driver dispatch list | |
1031 | */ | |
5e705374 | 1032 | cfq_dispatch_insert(cfqd->queue, rq); |
1da177e4 | 1033 | |
22e2c507 | 1034 | dispatched++; |
1da177e4 | 1035 | |
22e2c507 | 1036 | if (!cfqd->active_cic) { |
5e705374 JA |
1037 | atomic_inc(&RQ_CIC(rq)->ioc->refcount); |
1038 | cfqd->active_cic = RQ_CIC(rq); | |
22e2c507 | 1039 | } |
1da177e4 | 1040 | |
dd67d051 | 1041 | if (RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 JA |
1042 | break; |
1043 | ||
1044 | } while (dispatched < max_dispatch); | |
1045 | ||
22e2c507 JA |
1046 | /* |
1047 | * expire an async queue immediately if it has used up its slice. idle | |
1048 | * queue always expire after 1 dispatch round. | |
1049 | */ | |
a9938006 | 1050 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
20e493a8 | 1051 | dispatched >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
a9938006 | 1052 | cfq_class_idle(cfqq))) { |
44f7c160 | 1053 | cfqq->slice_end = jiffies + 1; |
6084cdda | 1054 | cfq_slice_expired(cfqd, 0); |
44f7c160 | 1055 | } |
22e2c507 JA |
1056 | |
1057 | return dispatched; | |
1058 | } | |
1059 | ||
d9e7620e JA |
1060 | static inline int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
1061 | { | |
1062 | int dispatched = 0; | |
1063 | ||
1064 | while (cfqq->next_rq) { | |
1065 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1066 | dispatched++; | |
1067 | } | |
1068 | ||
1069 | BUG_ON(!list_empty(&cfqq->fifo)); | |
1070 | return dispatched; | |
1071 | } | |
1072 | ||
498d3aa2 JA |
1073 | /* |
1074 | * Drain our current requests. Used for barriers and when switching | |
1075 | * io schedulers on-the-fly. | |
1076 | */ | |
d9e7620e | 1077 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 1078 | { |
d9e7620e JA |
1079 | int dispatched = 0; |
1080 | struct rb_node *n; | |
1081 | ||
cc09e299 | 1082 | while ((n = cfq_rb_first(&cfqd->service_tree)) != NULL) { |
d9e7620e | 1083 | struct cfq_queue *cfqq = rb_entry(n, struct cfq_queue, rb_node); |
1b5ed5e1 | 1084 | |
d9e7620e JA |
1085 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
1086 | } | |
1b5ed5e1 | 1087 | |
6084cdda | 1088 | cfq_slice_expired(cfqd, 0); |
1b5ed5e1 TH |
1089 | |
1090 | BUG_ON(cfqd->busy_queues); | |
1091 | ||
1092 | return dispatched; | |
1093 | } | |
1094 | ||
d9e7620e | 1095 | static int cfq_dispatch_requests(request_queue_t *q, int force) |
22e2c507 JA |
1096 | { |
1097 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
6d048f53 | 1098 | struct cfq_queue *cfqq; |
caaa5f9f | 1099 | int dispatched; |
22e2c507 JA |
1100 | |
1101 | if (!cfqd->busy_queues) | |
1102 | return 0; | |
1103 | ||
1b5ed5e1 TH |
1104 | if (unlikely(force)) |
1105 | return cfq_forced_dispatch(cfqd); | |
1106 | ||
caaa5f9f | 1107 | dispatched = 0; |
caaa5f9f | 1108 | while ((cfqq = cfq_select_queue(cfqd)) != NULL) { |
b4878f24 JA |
1109 | int max_dispatch; |
1110 | ||
3ed9a296 JA |
1111 | max_dispatch = cfqd->cfq_quantum; |
1112 | if (cfq_class_idle(cfqq)) | |
1113 | max_dispatch = 1; | |
1114 | ||
1115 | if (cfqq->dispatched >= max_dispatch) { | |
1116 | if (cfqd->busy_queues > 1) | |
6d048f53 | 1117 | break; |
3ed9a296 | 1118 | if (cfqq->dispatched >= 4 * max_dispatch) |
a9938006 JA |
1119 | break; |
1120 | } | |
9ede209e | 1121 | |
3ed9a296 JA |
1122 | if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq)) |
1123 | break; | |
1124 | ||
3b18152c JA |
1125 | cfq_clear_cfqq_must_dispatch(cfqq); |
1126 | cfq_clear_cfqq_wait_request(cfqq); | |
22e2c507 JA |
1127 | del_timer(&cfqd->idle_slice_timer); |
1128 | ||
caaa5f9f | 1129 | dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); |
1da177e4 LT |
1130 | } |
1131 | ||
caaa5f9f | 1132 | return dispatched; |
1da177e4 LT |
1133 | } |
1134 | ||
1da177e4 | 1135 | /* |
5e705374 JA |
1136 | * task holds one reference to the queue, dropped when task exits. each rq |
1137 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1138 | * |
1139 | * queue lock must be held here. | |
1140 | */ | |
1141 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1142 | { | |
22e2c507 JA |
1143 | struct cfq_data *cfqd = cfqq->cfqd; |
1144 | ||
1145 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1146 | |
1147 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1148 | return; | |
1149 | ||
1150 | BUG_ON(rb_first(&cfqq->sort_list)); | |
22e2c507 | 1151 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1152 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1153 | |
28f95cbc | 1154 | if (unlikely(cfqd->active_queue == cfqq)) { |
6084cdda | 1155 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1156 | cfq_schedule_dispatch(cfqd); |
1157 | } | |
22e2c507 | 1158 | |
1da177e4 LT |
1159 | kmem_cache_free(cfq_pool, cfqq); |
1160 | } | |
1161 | ||
e2d74ac0 | 1162 | static void cfq_free_io_context(struct io_context *ioc) |
1da177e4 | 1163 | { |
22e2c507 | 1164 | struct cfq_io_context *__cic; |
e2d74ac0 JA |
1165 | struct rb_node *n; |
1166 | int freed = 0; | |
1da177e4 | 1167 | |
e2d74ac0 JA |
1168 | while ((n = rb_first(&ioc->cic_root)) != NULL) { |
1169 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1170 | rb_erase(&__cic->rb_node, &ioc->cic_root); | |
22e2c507 | 1171 | kmem_cache_free(cfq_ioc_pool, __cic); |
334e94de | 1172 | freed++; |
1da177e4 LT |
1173 | } |
1174 | ||
4050cf16 JA |
1175 | elv_ioc_count_mod(ioc_count, -freed); |
1176 | ||
1177 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) | |
334e94de | 1178 | complete(ioc_gone); |
1da177e4 LT |
1179 | } |
1180 | ||
89850f7e | 1181 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1182 | { |
28f95cbc | 1183 | if (unlikely(cfqq == cfqd->active_queue)) { |
6084cdda | 1184 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1185 | cfq_schedule_dispatch(cfqd); |
1186 | } | |
22e2c507 | 1187 | |
89850f7e JA |
1188 | cfq_put_queue(cfqq); |
1189 | } | |
22e2c507 | 1190 | |
89850f7e JA |
1191 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1192 | struct cfq_io_context *cic) | |
1193 | { | |
fc46379d JA |
1194 | list_del_init(&cic->queue_list); |
1195 | smp_wmb(); | |
1196 | cic->key = NULL; | |
1197 | ||
12a05732 | 1198 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1199 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1200 | cic->cfqq[ASYNC] = NULL; |
1201 | } | |
1202 | ||
1203 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1204 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1205 | cic->cfqq[SYNC] = NULL; |
1206 | } | |
89850f7e JA |
1207 | } |
1208 | ||
89850f7e JA |
1209 | static void cfq_exit_single_io_context(struct cfq_io_context *cic) |
1210 | { | |
1211 | struct cfq_data *cfqd = cic->key; | |
1212 | ||
89850f7e JA |
1213 | if (cfqd) { |
1214 | request_queue_t *q = cfqd->queue; | |
1215 | ||
fc46379d | 1216 | spin_lock_irq(q->queue_lock); |
89850f7e | 1217 | __cfq_exit_single_io_context(cfqd, cic); |
fc46379d | 1218 | spin_unlock_irq(q->queue_lock); |
89850f7e | 1219 | } |
1da177e4 LT |
1220 | } |
1221 | ||
498d3aa2 JA |
1222 | /* |
1223 | * The process that ioc belongs to has exited, we need to clean up | |
1224 | * and put the internal structures we have that belongs to that process. | |
1225 | */ | |
e2d74ac0 | 1226 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1227 | { |
22e2c507 | 1228 | struct cfq_io_context *__cic; |
e2d74ac0 | 1229 | struct rb_node *n; |
22e2c507 | 1230 | |
1da177e4 LT |
1231 | /* |
1232 | * put the reference this task is holding to the various queues | |
1233 | */ | |
e2d74ac0 JA |
1234 | |
1235 | n = rb_first(&ioc->cic_root); | |
1236 | while (n != NULL) { | |
1237 | __cic = rb_entry(n, struct cfq_io_context, rb_node); | |
1238 | ||
22e2c507 | 1239 | cfq_exit_single_io_context(__cic); |
e2d74ac0 | 1240 | n = rb_next(n); |
1da177e4 | 1241 | } |
1da177e4 LT |
1242 | } |
1243 | ||
22e2c507 | 1244 | static struct cfq_io_context * |
8267e268 | 1245 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1246 | { |
b5deef90 | 1247 | struct cfq_io_context *cic; |
1da177e4 | 1248 | |
b5deef90 | 1249 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask, cfqd->queue->node); |
1da177e4 | 1250 | if (cic) { |
553698f9 | 1251 | memset(cic, 0, sizeof(*cic)); |
22e2c507 | 1252 | cic->last_end_request = jiffies; |
553698f9 | 1253 | INIT_LIST_HEAD(&cic->queue_list); |
22e2c507 JA |
1254 | cic->dtor = cfq_free_io_context; |
1255 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1256 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1257 | } |
1258 | ||
1259 | return cic; | |
1260 | } | |
1261 | ||
22e2c507 JA |
1262 | static void cfq_init_prio_data(struct cfq_queue *cfqq) |
1263 | { | |
1264 | struct task_struct *tsk = current; | |
1265 | int ioprio_class; | |
1266 | ||
3b18152c | 1267 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1268 | return; |
1269 | ||
1270 | ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio); | |
1271 | switch (ioprio_class) { | |
1272 | default: | |
1273 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1274 | case IOPRIO_CLASS_NONE: | |
1275 | /* | |
1276 | * no prio set, place us in the middle of the BE classes | |
1277 | */ | |
1278 | cfqq->ioprio = task_nice_ioprio(tsk); | |
1279 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1280 | break; | |
1281 | case IOPRIO_CLASS_RT: | |
1282 | cfqq->ioprio = task_ioprio(tsk); | |
1283 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1284 | break; | |
1285 | case IOPRIO_CLASS_BE: | |
1286 | cfqq->ioprio = task_ioprio(tsk); | |
1287 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1288 | break; | |
1289 | case IOPRIO_CLASS_IDLE: | |
1290 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1291 | cfqq->ioprio = 7; | |
3b18152c | 1292 | cfq_clear_cfqq_idle_window(cfqq); |
22e2c507 JA |
1293 | break; |
1294 | } | |
1295 | ||
1296 | /* | |
1297 | * keep track of original prio settings in case we have to temporarily | |
1298 | * elevate the priority of this queue | |
1299 | */ | |
1300 | cfqq->org_ioprio = cfqq->ioprio; | |
1301 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 1302 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1303 | } |
1304 | ||
478a82b0 | 1305 | static inline void changed_ioprio(struct cfq_io_context *cic) |
22e2c507 | 1306 | { |
478a82b0 AV |
1307 | struct cfq_data *cfqd = cic->key; |
1308 | struct cfq_queue *cfqq; | |
c1b707d2 | 1309 | unsigned long flags; |
35e6077c | 1310 | |
caaa5f9f JA |
1311 | if (unlikely(!cfqd)) |
1312 | return; | |
1313 | ||
c1b707d2 | 1314 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1315 | |
1316 | cfqq = cic->cfqq[ASYNC]; | |
1317 | if (cfqq) { | |
1318 | struct cfq_queue *new_cfqq; | |
91fac317 | 1319 | new_cfqq = cfq_get_queue(cfqd, ASYNC, cic->ioc->task, |
caaa5f9f JA |
1320 | GFP_ATOMIC); |
1321 | if (new_cfqq) { | |
1322 | cic->cfqq[ASYNC] = new_cfqq; | |
1323 | cfq_put_queue(cfqq); | |
1324 | } | |
22e2c507 | 1325 | } |
caaa5f9f JA |
1326 | |
1327 | cfqq = cic->cfqq[SYNC]; | |
1328 | if (cfqq) | |
1329 | cfq_mark_cfqq_prio_changed(cfqq); | |
1330 | ||
c1b707d2 | 1331 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1332 | } |
1333 | ||
fc46379d | 1334 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1335 | { |
a6a0763a | 1336 | struct cfq_io_context *cic; |
e2d74ac0 | 1337 | struct rb_node *n; |
a6a0763a | 1338 | |
fc46379d | 1339 | ioc->ioprio_changed = 0; |
a6a0763a | 1340 | |
e2d74ac0 JA |
1341 | n = rb_first(&ioc->cic_root); |
1342 | while (n != NULL) { | |
1343 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
3793c65c | 1344 | |
478a82b0 | 1345 | changed_ioprio(cic); |
e2d74ac0 JA |
1346 | n = rb_next(n); |
1347 | } | |
22e2c507 JA |
1348 | } |
1349 | ||
1350 | static struct cfq_queue * | |
91fac317 | 1351 | cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct task_struct *tsk, |
8267e268 | 1352 | gfp_t gfp_mask) |
22e2c507 | 1353 | { |
22e2c507 | 1354 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 1355 | struct cfq_io_context *cic; |
22e2c507 JA |
1356 | |
1357 | retry: | |
91fac317 VT |
1358 | cic = cfq_cic_rb_lookup(cfqd, tsk->io_context); |
1359 | /* cic always exists here */ | |
1360 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 JA |
1361 | |
1362 | if (!cfqq) { | |
1363 | if (new_cfqq) { | |
1364 | cfqq = new_cfqq; | |
1365 | new_cfqq = NULL; | |
1366 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1367 | /* |
1368 | * Inform the allocator of the fact that we will | |
1369 | * just repeat this allocation if it fails, to allow | |
1370 | * the allocator to do whatever it needs to attempt to | |
1371 | * free memory. | |
1372 | */ | |
22e2c507 | 1373 | spin_unlock_irq(cfqd->queue->queue_lock); |
b5deef90 | 1374 | new_cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask|__GFP_NOFAIL, cfqd->queue->node); |
22e2c507 JA |
1375 | spin_lock_irq(cfqd->queue->queue_lock); |
1376 | goto retry; | |
1377 | } else { | |
b5deef90 | 1378 | cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask, cfqd->queue->node); |
22e2c507 JA |
1379 | if (!cfqq) |
1380 | goto out; | |
1381 | } | |
1382 | ||
1383 | memset(cfqq, 0, sizeof(*cfqq)); | |
1384 | ||
d9e7620e | 1385 | RB_CLEAR_NODE(&cfqq->rb_node); |
22e2c507 JA |
1386 | INIT_LIST_HEAD(&cfqq->fifo); |
1387 | ||
22e2c507 JA |
1388 | atomic_set(&cfqq->ref, 0); |
1389 | cfqq->cfqd = cfqd; | |
c5b680f3 | 1390 | |
91fac317 | 1391 | if (is_sync) { |
a9938006 | 1392 | cfq_mark_cfqq_idle_window(cfqq); |
91fac317 VT |
1393 | cfq_mark_cfqq_sync(cfqq); |
1394 | } | |
a9938006 | 1395 | |
3b18152c | 1396 | cfq_mark_cfqq_prio_changed(cfqq); |
53b03744 | 1397 | cfq_mark_cfqq_queue_new(cfqq); |
91fac317 | 1398 | |
3b18152c | 1399 | cfq_init_prio_data(cfqq); |
22e2c507 JA |
1400 | } |
1401 | ||
1402 | if (new_cfqq) | |
1403 | kmem_cache_free(cfq_pool, new_cfqq); | |
1404 | ||
1405 | atomic_inc(&cfqq->ref); | |
1406 | out: | |
1407 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1408 | return cfqq; | |
1409 | } | |
1410 | ||
498d3aa2 JA |
1411 | /* |
1412 | * We drop cfq io contexts lazily, so we may find a dead one. | |
1413 | */ | |
dbecf3ab OH |
1414 | static void |
1415 | cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic) | |
1416 | { | |
fc46379d | 1417 | WARN_ON(!list_empty(&cic->queue_list)); |
dbecf3ab | 1418 | rb_erase(&cic->rb_node, &ioc->cic_root); |
dbecf3ab | 1419 | kmem_cache_free(cfq_ioc_pool, cic); |
4050cf16 | 1420 | elv_ioc_count_dec(ioc_count); |
dbecf3ab OH |
1421 | } |
1422 | ||
e2d74ac0 JA |
1423 | static struct cfq_io_context * |
1424 | cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc) | |
1425 | { | |
dbecf3ab | 1426 | struct rb_node *n; |
e2d74ac0 | 1427 | struct cfq_io_context *cic; |
be3b0753 | 1428 | void *k, *key = cfqd; |
e2d74ac0 | 1429 | |
91fac317 VT |
1430 | if (unlikely(!ioc)) |
1431 | return NULL; | |
1432 | ||
dbecf3ab OH |
1433 | restart: |
1434 | n = ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1435 | while (n) { |
1436 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1437 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1438 | k = cic->key; | |
1439 | if (unlikely(!k)) { | |
dbecf3ab OH |
1440 | cfq_drop_dead_cic(ioc, cic); |
1441 | goto restart; | |
1442 | } | |
e2d74ac0 | 1443 | |
be3b0753 | 1444 | if (key < k) |
e2d74ac0 | 1445 | n = n->rb_left; |
be3b0753 | 1446 | else if (key > k) |
e2d74ac0 JA |
1447 | n = n->rb_right; |
1448 | else | |
1449 | return cic; | |
1450 | } | |
1451 | ||
1452 | return NULL; | |
1453 | } | |
1454 | ||
1455 | static inline void | |
1456 | cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, | |
1457 | struct cfq_io_context *cic) | |
1458 | { | |
dbecf3ab OH |
1459 | struct rb_node **p; |
1460 | struct rb_node *parent; | |
e2d74ac0 | 1461 | struct cfq_io_context *__cic; |
0261d688 | 1462 | unsigned long flags; |
be3b0753 | 1463 | void *k; |
e2d74ac0 | 1464 | |
e2d74ac0 JA |
1465 | cic->ioc = ioc; |
1466 | cic->key = cfqd; | |
1467 | ||
dbecf3ab OH |
1468 | restart: |
1469 | parent = NULL; | |
1470 | p = &ioc->cic_root.rb_node; | |
e2d74ac0 JA |
1471 | while (*p) { |
1472 | parent = *p; | |
1473 | __cic = rb_entry(parent, struct cfq_io_context, rb_node); | |
be3b0753 OH |
1474 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1475 | k = __cic->key; | |
1476 | if (unlikely(!k)) { | |
be33c3a6 | 1477 | cfq_drop_dead_cic(ioc, __cic); |
dbecf3ab OH |
1478 | goto restart; |
1479 | } | |
e2d74ac0 | 1480 | |
be3b0753 | 1481 | if (cic->key < k) |
e2d74ac0 | 1482 | p = &(*p)->rb_left; |
be3b0753 | 1483 | else if (cic->key > k) |
e2d74ac0 JA |
1484 | p = &(*p)->rb_right; |
1485 | else | |
1486 | BUG(); | |
1487 | } | |
1488 | ||
1489 | rb_link_node(&cic->rb_node, parent, p); | |
1490 | rb_insert_color(&cic->rb_node, &ioc->cic_root); | |
fc46379d | 1491 | |
0261d688 | 1492 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
e2d74ac0 | 1493 | list_add(&cic->queue_list, &cfqd->cic_list); |
0261d688 | 1494 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
e2d74ac0 JA |
1495 | } |
1496 | ||
1da177e4 LT |
1497 | /* |
1498 | * Setup general io context and cfq io context. There can be several cfq | |
1499 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1500 | * than one device managed by cfq. |
1da177e4 LT |
1501 | */ |
1502 | static struct cfq_io_context * | |
e2d74ac0 | 1503 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1504 | { |
22e2c507 | 1505 | struct io_context *ioc = NULL; |
1da177e4 | 1506 | struct cfq_io_context *cic; |
1da177e4 | 1507 | |
22e2c507 | 1508 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1509 | |
b5deef90 | 1510 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1511 | if (!ioc) |
1512 | return NULL; | |
1513 | ||
e2d74ac0 JA |
1514 | cic = cfq_cic_rb_lookup(cfqd, ioc); |
1515 | if (cic) | |
1516 | goto out; | |
1da177e4 | 1517 | |
e2d74ac0 JA |
1518 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1519 | if (cic == NULL) | |
1520 | goto err; | |
1da177e4 | 1521 | |
e2d74ac0 | 1522 | cfq_cic_link(cfqd, ioc, cic); |
1da177e4 | 1523 | out: |
fc46379d JA |
1524 | smp_read_barrier_depends(); |
1525 | if (unlikely(ioc->ioprio_changed)) | |
1526 | cfq_ioc_set_ioprio(ioc); | |
1527 | ||
1da177e4 LT |
1528 | return cic; |
1529 | err: | |
1530 | put_io_context(ioc); | |
1531 | return NULL; | |
1532 | } | |
1533 | ||
22e2c507 JA |
1534 | static void |
1535 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1536 | { |
aaf1228d JA |
1537 | unsigned long elapsed = jiffies - cic->last_end_request; |
1538 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 1539 | |
22e2c507 JA |
1540 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1541 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1542 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1543 | } | |
1da177e4 | 1544 | |
206dc69b | 1545 | static void |
6d048f53 JA |
1546 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic, |
1547 | struct request *rq) | |
206dc69b JA |
1548 | { |
1549 | sector_t sdist; | |
1550 | u64 total; | |
1551 | ||
5e705374 JA |
1552 | if (cic->last_request_pos < rq->sector) |
1553 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1554 | else |
5e705374 | 1555 | sdist = cic->last_request_pos - rq->sector; |
206dc69b | 1556 | |
6d048f53 JA |
1557 | if (!cic->seek_samples) { |
1558 | cfqd->new_seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1559 | cfqd->new_seek_mean = cfqd->new_seek_total / 256; | |
1560 | } | |
1561 | ||
206dc69b JA |
1562 | /* |
1563 | * Don't allow the seek distance to get too large from the | |
1564 | * odd fragment, pagein, etc | |
1565 | */ | |
1566 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1567 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1568 | else | |
1569 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1570 | ||
1571 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1572 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1573 | total = cic->seek_total + (cic->seek_samples/2); | |
1574 | do_div(total, cic->seek_samples); | |
1575 | cic->seek_mean = (sector_t)total; | |
1576 | } | |
1da177e4 | 1577 | |
22e2c507 JA |
1578 | /* |
1579 | * Disable idle window if the process thinks too long or seeks so much that | |
1580 | * it doesn't matter | |
1581 | */ | |
1582 | static void | |
1583 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1584 | struct cfq_io_context *cic) | |
1585 | { | |
1be92f2f JA |
1586 | int enable_idle; |
1587 | ||
1588 | if (!cfq_cfqq_sync(cfqq)) | |
1589 | return; | |
1590 | ||
1591 | enable_idle = cfq_cfqq_idle_window(cfqq); | |
1da177e4 | 1592 | |
caaa5f9f JA |
1593 | if (!cic->ioc->task || !cfqd->cfq_slice_idle || |
1594 | (cfqd->hw_tag && CIC_SEEKY(cic))) | |
22e2c507 JA |
1595 | enable_idle = 0; |
1596 | else if (sample_valid(cic->ttime_samples)) { | |
1597 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1598 | enable_idle = 0; | |
1599 | else | |
1600 | enable_idle = 1; | |
1da177e4 LT |
1601 | } |
1602 | ||
3b18152c JA |
1603 | if (enable_idle) |
1604 | cfq_mark_cfqq_idle_window(cfqq); | |
1605 | else | |
1606 | cfq_clear_cfqq_idle_window(cfqq); | |
22e2c507 | 1607 | } |
1da177e4 | 1608 | |
22e2c507 JA |
1609 | /* |
1610 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1611 | * no or if we aren't sure, a 1 will cause a preempt. | |
1612 | */ | |
1613 | static int | |
1614 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1615 | struct request *rq) |
22e2c507 | 1616 | { |
6d048f53 | 1617 | struct cfq_queue *cfqq; |
22e2c507 | 1618 | |
6d048f53 JA |
1619 | cfqq = cfqd->active_queue; |
1620 | if (!cfqq) | |
22e2c507 JA |
1621 | return 0; |
1622 | ||
6d048f53 JA |
1623 | if (cfq_slice_used(cfqq)) |
1624 | return 1; | |
1625 | ||
1626 | if (cfq_class_idle(new_cfqq)) | |
caaa5f9f | 1627 | return 0; |
22e2c507 JA |
1628 | |
1629 | if (cfq_class_idle(cfqq)) | |
1630 | return 1; | |
1e3335de | 1631 | |
374f84ac JA |
1632 | /* |
1633 | * if the new request is sync, but the currently running queue is | |
1634 | * not, let the sync request have priority. | |
1635 | */ | |
5e705374 | 1636 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1637 | return 1; |
1e3335de | 1638 | |
374f84ac JA |
1639 | /* |
1640 | * So both queues are sync. Let the new request get disk time if | |
1641 | * it's a metadata request and the current queue is doing regular IO. | |
1642 | */ | |
1643 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1644 | return 1; | |
22e2c507 | 1645 | |
1e3335de JA |
1646 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
1647 | return 0; | |
1648 | ||
1649 | /* | |
1650 | * if this request is as-good as one we would expect from the | |
1651 | * current cfqq, let it preempt | |
1652 | */ | |
6d048f53 | 1653 | if (cfq_rq_close(cfqd, rq)) |
1e3335de JA |
1654 | return 1; |
1655 | ||
22e2c507 JA |
1656 | return 0; |
1657 | } | |
1658 | ||
1659 | /* | |
1660 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1661 | * let it have half of its nominal slice. | |
1662 | */ | |
1663 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1664 | { | |
6084cdda | 1665 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 1666 | |
bf572256 JA |
1667 | /* |
1668 | * Put the new queue at the front of the of the current list, | |
1669 | * so we know that it will be selected next. | |
1670 | */ | |
1671 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
1672 | |
1673 | cfq_service_tree_add(cfqd, cfqq, 1); | |
bf572256 | 1674 | |
44f7c160 JA |
1675 | cfqq->slice_end = 0; |
1676 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
1677 | } |
1678 | ||
22e2c507 | 1679 | /* |
5e705374 | 1680 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1681 | * something we should do about it |
1682 | */ | |
1683 | static void | |
5e705374 JA |
1684 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1685 | struct request *rq) | |
22e2c507 | 1686 | { |
5e705374 | 1687 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1688 | |
374f84ac JA |
1689 | if (rq_is_meta(rq)) |
1690 | cfqq->meta_pending++; | |
1691 | ||
9c2c38a1 | 1692 | cfq_update_io_thinktime(cfqd, cic); |
6d048f53 | 1693 | cfq_update_io_seektime(cfqd, cic, rq); |
9c2c38a1 JA |
1694 | cfq_update_idle_window(cfqd, cfqq, cic); |
1695 | ||
5e705374 | 1696 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
6d048f53 | 1697 | cfqq->last_request_pos = cic->last_request_pos; |
22e2c507 JA |
1698 | |
1699 | if (cfqq == cfqd->active_queue) { | |
1700 | /* | |
1701 | * if we are waiting for a request for this queue, let it rip | |
1702 | * immediately and flag that we must not expire this queue | |
1703 | * just now | |
1704 | */ | |
3b18152c JA |
1705 | if (cfq_cfqq_wait_request(cfqq)) { |
1706 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 | 1707 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1708 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1709 | } |
5e705374 | 1710 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1711 | /* |
1712 | * not the active queue - expire current slice if it is | |
1713 | * idle and has expired it's mean thinktime or this new queue | |
1714 | * has some old slice time left and is of higher priority | |
1715 | */ | |
1716 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1717 | cfq_mark_cfqq_must_dispatch(cfqq); |
dc72ef4a | 1718 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1719 | } |
1da177e4 LT |
1720 | } |
1721 | ||
b4878f24 | 1722 | static void cfq_insert_request(request_queue_t *q, struct request *rq) |
1da177e4 | 1723 | { |
b4878f24 | 1724 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1725 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1726 | |
1727 | cfq_init_prio_data(cfqq); | |
1da177e4 | 1728 | |
5e705374 | 1729 | cfq_add_rq_rb(rq); |
1da177e4 | 1730 | |
22e2c507 JA |
1731 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1732 | ||
5e705374 | 1733 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1734 | } |
1735 | ||
1da177e4 LT |
1736 | static void cfq_completed_request(request_queue_t *q, struct request *rq) |
1737 | { | |
5e705374 | 1738 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1739 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1740 | const int sync = rq_is_sync(rq); |
b4878f24 | 1741 | unsigned long now; |
1da177e4 | 1742 | |
b4878f24 | 1743 | now = jiffies; |
1da177e4 | 1744 | |
b4878f24 | 1745 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 1746 | WARN_ON(!cfqq->dispatched); |
b4878f24 | 1747 | cfqd->rq_in_driver--; |
6d048f53 | 1748 | cfqq->dispatched--; |
1da177e4 | 1749 | |
3ed9a296 JA |
1750 | if (cfq_cfqq_sync(cfqq)) |
1751 | cfqd->sync_flight--; | |
1752 | ||
b4878f24 JA |
1753 | if (!cfq_class_idle(cfqq)) |
1754 | cfqd->last_end_request = now; | |
3b18152c | 1755 | |
caaa5f9f | 1756 | if (sync) |
5e705374 | 1757 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1758 | |
1759 | /* | |
1760 | * If this is the active queue, check if it needs to be expired, | |
1761 | * or if we want to idle in case it has no pending requests. | |
1762 | */ | |
1763 | if (cfqd->active_queue == cfqq) { | |
44f7c160 JA |
1764 | if (cfq_cfqq_slice_new(cfqq)) { |
1765 | cfq_set_prio_slice(cfqd, cfqq); | |
1766 | cfq_clear_cfqq_slice_new(cfqq); | |
1767 | } | |
1768 | if (cfq_slice_used(cfqq)) | |
6084cdda | 1769 | cfq_slice_expired(cfqd, 1); |
6d048f53 JA |
1770 | else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1771 | cfq_arm_slice_timer(cfqd); | |
caaa5f9f | 1772 | } |
6d048f53 JA |
1773 | |
1774 | if (!cfqd->rq_in_driver) | |
1775 | cfq_schedule_dispatch(cfqd); | |
1da177e4 LT |
1776 | } |
1777 | ||
22e2c507 JA |
1778 | /* |
1779 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
1780 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
1781 | */ | |
1782 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 1783 | { |
22e2c507 JA |
1784 | if (has_fs_excl()) { |
1785 | /* | |
1786 | * boost idle prio on transactions that would lock out other | |
1787 | * users of the filesystem | |
1788 | */ | |
1789 | if (cfq_class_idle(cfqq)) | |
1790 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1791 | if (cfqq->ioprio > IOPRIO_NORM) | |
1792 | cfqq->ioprio = IOPRIO_NORM; | |
1793 | } else { | |
1794 | /* | |
1795 | * check if we need to unboost the queue | |
1796 | */ | |
1797 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
1798 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
1799 | if (cfqq->ioprio != cfqq->org_ioprio) | |
1800 | cfqq->ioprio = cfqq->org_ioprio; | |
1801 | } | |
22e2c507 | 1802 | } |
1da177e4 | 1803 | |
89850f7e | 1804 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 1805 | { |
3b18152c | 1806 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 1807 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 1808 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 1809 | return ELV_MQUEUE_MUST; |
3b18152c | 1810 | } |
1da177e4 | 1811 | |
22e2c507 | 1812 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
1813 | } |
1814 | ||
cb78b285 | 1815 | static int cfq_may_queue(request_queue_t *q, int rw) |
22e2c507 JA |
1816 | { |
1817 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1818 | struct task_struct *tsk = current; | |
91fac317 | 1819 | struct cfq_io_context *cic; |
22e2c507 JA |
1820 | struct cfq_queue *cfqq; |
1821 | ||
1822 | /* | |
1823 | * don't force setup of a queue from here, as a call to may_queue | |
1824 | * does not necessarily imply that a request actually will be queued. | |
1825 | * so just lookup a possibly existing queue, or return 'may queue' | |
1826 | * if that fails | |
1827 | */ | |
91fac317 VT |
1828 | cic = cfq_cic_rb_lookup(cfqd, tsk->io_context); |
1829 | if (!cic) | |
1830 | return ELV_MQUEUE_MAY; | |
1831 | ||
1832 | cfqq = cic_to_cfqq(cic, rw & REQ_RW_SYNC); | |
22e2c507 JA |
1833 | if (cfqq) { |
1834 | cfq_init_prio_data(cfqq); | |
1835 | cfq_prio_boost(cfqq); | |
1836 | ||
89850f7e | 1837 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
1838 | } |
1839 | ||
1840 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
1841 | } |
1842 | ||
1da177e4 LT |
1843 | /* |
1844 | * queue lock held here | |
1845 | */ | |
bb37b94c | 1846 | static void cfq_put_request(struct request *rq) |
1da177e4 | 1847 | { |
5e705374 | 1848 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1849 | |
5e705374 | 1850 | if (cfqq) { |
22e2c507 | 1851 | const int rw = rq_data_dir(rq); |
1da177e4 | 1852 | |
22e2c507 JA |
1853 | BUG_ON(!cfqq->allocated[rw]); |
1854 | cfqq->allocated[rw]--; | |
1da177e4 | 1855 | |
5e705374 | 1856 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 1857 | |
1da177e4 | 1858 | rq->elevator_private = NULL; |
5e705374 | 1859 | rq->elevator_private2 = NULL; |
1da177e4 | 1860 | |
1da177e4 LT |
1861 | cfq_put_queue(cfqq); |
1862 | } | |
1863 | } | |
1864 | ||
1865 | /* | |
22e2c507 | 1866 | * Allocate cfq data structures associated with this request. |
1da177e4 | 1867 | */ |
22e2c507 | 1868 | static int |
cb78b285 | 1869 | cfq_set_request(request_queue_t *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
1870 | { |
1871 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3b18152c | 1872 | struct task_struct *tsk = current; |
1da177e4 LT |
1873 | struct cfq_io_context *cic; |
1874 | const int rw = rq_data_dir(rq); | |
7749a8d4 | 1875 | const int is_sync = rq_is_sync(rq); |
22e2c507 | 1876 | struct cfq_queue *cfqq; |
1da177e4 LT |
1877 | unsigned long flags; |
1878 | ||
1879 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1880 | ||
e2d74ac0 | 1881 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 1882 | |
1da177e4 LT |
1883 | spin_lock_irqsave(q->queue_lock, flags); |
1884 | ||
22e2c507 JA |
1885 | if (!cic) |
1886 | goto queue_fail; | |
1887 | ||
91fac317 VT |
1888 | cfqq = cic_to_cfqq(cic, is_sync); |
1889 | if (!cfqq) { | |
1890 | cfqq = cfq_get_queue(cfqd, is_sync, tsk, gfp_mask); | |
1891 | ||
22e2c507 JA |
1892 | if (!cfqq) |
1893 | goto queue_fail; | |
1da177e4 | 1894 | |
91fac317 VT |
1895 | cic_set_cfqq(cic, cfqq, is_sync); |
1896 | } | |
1da177e4 LT |
1897 | |
1898 | cfqq->allocated[rw]++; | |
3b18152c | 1899 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 1900 | atomic_inc(&cfqq->ref); |
1da177e4 | 1901 | |
5e705374 | 1902 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 1903 | |
5e705374 JA |
1904 | rq->elevator_private = cic; |
1905 | rq->elevator_private2 = cfqq; | |
1906 | return 0; | |
1da177e4 | 1907 | |
22e2c507 JA |
1908 | queue_fail: |
1909 | if (cic) | |
1910 | put_io_context(cic->ioc); | |
89850f7e | 1911 | |
3b18152c | 1912 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
1913 | spin_unlock_irqrestore(q->queue_lock, flags); |
1914 | return 1; | |
1915 | } | |
1916 | ||
65f27f38 | 1917 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 1918 | { |
65f27f38 DH |
1919 | struct cfq_data *cfqd = |
1920 | container_of(work, struct cfq_data, unplug_work); | |
1921 | request_queue_t *q = cfqd->queue; | |
22e2c507 JA |
1922 | unsigned long flags; |
1923 | ||
1924 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 1925 | blk_start_queueing(q); |
22e2c507 JA |
1926 | spin_unlock_irqrestore(q->queue_lock, flags); |
1927 | } | |
1928 | ||
1929 | /* | |
1930 | * Timer running if the active_queue is currently idling inside its time slice | |
1931 | */ | |
1932 | static void cfq_idle_slice_timer(unsigned long data) | |
1933 | { | |
1934 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1935 | struct cfq_queue *cfqq; | |
1936 | unsigned long flags; | |
3c6bd2f8 | 1937 | int timed_out = 1; |
22e2c507 JA |
1938 | |
1939 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1940 | ||
1941 | if ((cfqq = cfqd->active_queue) != NULL) { | |
3c6bd2f8 JA |
1942 | timed_out = 0; |
1943 | ||
22e2c507 JA |
1944 | /* |
1945 | * expired | |
1946 | */ | |
44f7c160 | 1947 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
1948 | goto expire; |
1949 | ||
1950 | /* | |
1951 | * only expire and reinvoke request handler, if there are | |
1952 | * other queues with pending requests | |
1953 | */ | |
caaa5f9f | 1954 | if (!cfqd->busy_queues) |
22e2c507 | 1955 | goto out_cont; |
22e2c507 JA |
1956 | |
1957 | /* | |
1958 | * not expired and it has a request pending, let it dispatch | |
1959 | */ | |
dd67d051 | 1960 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 1961 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
1962 | goto out_kick; |
1963 | } | |
1964 | } | |
1965 | expire: | |
6084cdda | 1966 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 1967 | out_kick: |
3b18152c | 1968 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
1969 | out_cont: |
1970 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1971 | } | |
1972 | ||
1973 | /* | |
1974 | * Timer running if an idle class queue is waiting for service | |
1975 | */ | |
1976 | static void cfq_idle_class_timer(unsigned long data) | |
1977 | { | |
1978 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
1979 | unsigned long flags, end; | |
1980 | ||
1981 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1982 | ||
1983 | /* | |
1984 | * race with a non-idle queue, reset timer | |
1985 | */ | |
1986 | end = cfqd->last_end_request + CFQ_IDLE_GRACE; | |
ae818a38 JA |
1987 | if (!time_after_eq(jiffies, end)) |
1988 | mod_timer(&cfqd->idle_class_timer, end); | |
1989 | else | |
3b18152c | 1990 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
1991 | |
1992 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1993 | } | |
1994 | ||
3b18152c JA |
1995 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
1996 | { | |
1997 | del_timer_sync(&cfqd->idle_slice_timer); | |
1998 | del_timer_sync(&cfqd->idle_class_timer); | |
1999 | blk_sync_queue(cfqd->queue); | |
2000 | } | |
22e2c507 | 2001 | |
1da177e4 LT |
2002 | static void cfq_exit_queue(elevator_t *e) |
2003 | { | |
22e2c507 | 2004 | struct cfq_data *cfqd = e->elevator_data; |
d9ff4187 | 2005 | request_queue_t *q = cfqd->queue; |
22e2c507 | 2006 | |
3b18152c | 2007 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 2008 | |
d9ff4187 | 2009 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 2010 | |
d9ff4187 | 2011 | if (cfqd->active_queue) |
6084cdda | 2012 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
2013 | |
2014 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
2015 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
2016 | struct cfq_io_context, | |
2017 | queue_list); | |
89850f7e JA |
2018 | |
2019 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 2020 | } |
e2d74ac0 | 2021 | |
d9ff4187 | 2022 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
2023 | |
2024 | cfq_shutdown_timer_wq(cfqd); | |
2025 | ||
a90d742e | 2026 | kfree(cfqd); |
1da177e4 LT |
2027 | } |
2028 | ||
bb37b94c | 2029 | static void *cfq_init_queue(request_queue_t *q) |
1da177e4 LT |
2030 | { |
2031 | struct cfq_data *cfqd; | |
1da177e4 | 2032 | |
b5deef90 | 2033 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node); |
1da177e4 | 2034 | if (!cfqd) |
bc1c1169 | 2035 | return NULL; |
1da177e4 LT |
2036 | |
2037 | memset(cfqd, 0, sizeof(*cfqd)); | |
22e2c507 | 2038 | |
cc09e299 | 2039 | cfqd->service_tree = CFQ_RB_ROOT; |
d9ff4187 | 2040 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2041 | |
1da177e4 | 2042 | cfqd->queue = q; |
1da177e4 | 2043 | |
22e2c507 JA |
2044 | init_timer(&cfqd->idle_slice_timer); |
2045 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2046 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2047 | ||
2048 | init_timer(&cfqd->idle_class_timer); | |
2049 | cfqd->idle_class_timer.function = cfq_idle_class_timer; | |
2050 | cfqd->idle_class_timer.data = (unsigned long) cfqd; | |
2051 | ||
65f27f38 | 2052 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2053 | |
1da177e4 | 2054 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2055 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2056 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2057 | cfqd->cfq_back_max = cfq_back_max; |
2058 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2059 | cfqd->cfq_slice[0] = cfq_slice_async; |
2060 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2061 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2062 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
3b18152c | 2063 | |
bc1c1169 | 2064 | return cfqd; |
1da177e4 LT |
2065 | } |
2066 | ||
2067 | static void cfq_slab_kill(void) | |
2068 | { | |
1da177e4 LT |
2069 | if (cfq_pool) |
2070 | kmem_cache_destroy(cfq_pool); | |
2071 | if (cfq_ioc_pool) | |
2072 | kmem_cache_destroy(cfq_ioc_pool); | |
2073 | } | |
2074 | ||
2075 | static int __init cfq_slab_setup(void) | |
2076 | { | |
1da177e4 LT |
2077 | cfq_pool = kmem_cache_create("cfq_pool", sizeof(struct cfq_queue), 0, 0, |
2078 | NULL, NULL); | |
2079 | if (!cfq_pool) | |
2080 | goto fail; | |
2081 | ||
2082 | cfq_ioc_pool = kmem_cache_create("cfq_ioc_pool", | |
2083 | sizeof(struct cfq_io_context), 0, 0, NULL, NULL); | |
2084 | if (!cfq_ioc_pool) | |
2085 | goto fail; | |
2086 | ||
2087 | return 0; | |
2088 | fail: | |
2089 | cfq_slab_kill(); | |
2090 | return -ENOMEM; | |
2091 | } | |
2092 | ||
1da177e4 LT |
2093 | /* |
2094 | * sysfs parts below --> | |
2095 | */ | |
1da177e4 LT |
2096 | static ssize_t |
2097 | cfq_var_show(unsigned int var, char *page) | |
2098 | { | |
2099 | return sprintf(page, "%d\n", var); | |
2100 | } | |
2101 | ||
2102 | static ssize_t | |
2103 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2104 | { | |
2105 | char *p = (char *) page; | |
2106 | ||
2107 | *var = simple_strtoul(p, &p, 10); | |
2108 | return count; | |
2109 | } | |
2110 | ||
1da177e4 | 2111 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
3d1ab40f | 2112 | static ssize_t __FUNC(elevator_t *e, char *page) \ |
1da177e4 | 2113 | { \ |
3d1ab40f | 2114 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2115 | unsigned int __data = __VAR; \ |
2116 | if (__CONV) \ | |
2117 | __data = jiffies_to_msecs(__data); \ | |
2118 | return cfq_var_show(__data, (page)); \ | |
2119 | } | |
2120 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2121 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2122 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2123 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2124 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2125 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2126 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2127 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2128 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2129 | #undef SHOW_FUNCTION |
2130 | ||
2131 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
3d1ab40f | 2132 | static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \ |
1da177e4 | 2133 | { \ |
3d1ab40f | 2134 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2135 | unsigned int __data; \ |
2136 | int ret = cfq_var_store(&__data, (page), count); \ | |
2137 | if (__data < (MIN)) \ | |
2138 | __data = (MIN); \ | |
2139 | else if (__data > (MAX)) \ | |
2140 | __data = (MAX); \ | |
2141 | if (__CONV) \ | |
2142 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2143 | else \ | |
2144 | *(__PTR) = __data; \ | |
2145 | return ret; \ | |
2146 | } | |
2147 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
22e2c507 JA |
2148 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1); |
2149 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1); | |
e572ec7e AV |
2150 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
2151 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0); | |
22e2c507 JA |
2152 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2153 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2154 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
2155 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0); | |
1da177e4 LT |
2156 | #undef STORE_FUNCTION |
2157 | ||
e572ec7e AV |
2158 | #define CFQ_ATTR(name) \ |
2159 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2160 | ||
2161 | static struct elv_fs_entry cfq_attrs[] = { | |
2162 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2163 | CFQ_ATTR(fifo_expire_sync), |
2164 | CFQ_ATTR(fifo_expire_async), | |
2165 | CFQ_ATTR(back_seek_max), | |
2166 | CFQ_ATTR(back_seek_penalty), | |
2167 | CFQ_ATTR(slice_sync), | |
2168 | CFQ_ATTR(slice_async), | |
2169 | CFQ_ATTR(slice_async_rq), | |
2170 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2171 | __ATTR_NULL |
1da177e4 LT |
2172 | }; |
2173 | ||
1da177e4 LT |
2174 | static struct elevator_type iosched_cfq = { |
2175 | .ops = { | |
2176 | .elevator_merge_fn = cfq_merge, | |
2177 | .elevator_merged_fn = cfq_merged_request, | |
2178 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2179 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2180 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2181 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2182 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2183 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2184 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2185 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2186 | .elevator_former_req_fn = elv_rb_former_request, |
2187 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2188 | .elevator_set_req_fn = cfq_set_request, |
2189 | .elevator_put_req_fn = cfq_put_request, | |
2190 | .elevator_may_queue_fn = cfq_may_queue, | |
2191 | .elevator_init_fn = cfq_init_queue, | |
2192 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2193 | .trim = cfq_free_io_context, |
1da177e4 | 2194 | }, |
3d1ab40f | 2195 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2196 | .elevator_name = "cfq", |
2197 | .elevator_owner = THIS_MODULE, | |
2198 | }; | |
2199 | ||
2200 | static int __init cfq_init(void) | |
2201 | { | |
2202 | int ret; | |
2203 | ||
22e2c507 JA |
2204 | /* |
2205 | * could be 0 on HZ < 1000 setups | |
2206 | */ | |
2207 | if (!cfq_slice_async) | |
2208 | cfq_slice_async = 1; | |
2209 | if (!cfq_slice_idle) | |
2210 | cfq_slice_idle = 1; | |
2211 | ||
1da177e4 LT |
2212 | if (cfq_slab_setup()) |
2213 | return -ENOMEM; | |
2214 | ||
2215 | ret = elv_register(&iosched_cfq); | |
22e2c507 JA |
2216 | if (ret) |
2217 | cfq_slab_kill(); | |
1da177e4 | 2218 | |
1da177e4 LT |
2219 | return ret; |
2220 | } | |
2221 | ||
2222 | static void __exit cfq_exit(void) | |
2223 | { | |
6e9a4738 | 2224 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2225 | elv_unregister(&iosched_cfq); |
334e94de | 2226 | ioc_gone = &all_gone; |
fba82272 OH |
2227 | /* ioc_gone's update must be visible before reading ioc_count */ |
2228 | smp_wmb(); | |
4050cf16 | 2229 | if (elv_ioc_count_read(ioc_count)) |
fba82272 | 2230 | wait_for_completion(ioc_gone); |
334e94de | 2231 | synchronize_rcu(); |
83521d3e | 2232 | cfq_slab_kill(); |
1da177e4 LT |
2233 | } |
2234 | ||
2235 | module_init(cfq_init); | |
2236 | module_exit(cfq_exit); | |
2237 | ||
2238 | MODULE_AUTHOR("Jens Axboe"); | |
2239 | MODULE_LICENSE("GPL"); | |
2240 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |