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 | */ | |
fe094d98 JA |
18 | /* max queue in one round of service */ |
19 | static const int cfq_quantum = 4; | |
64100099 | 20 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
21 | /* maximum backwards seek, in KiB */ |
22 | static const int cfq_back_max = 16 * 1024; | |
23 | /* penalty of a backwards seek */ | |
24 | static const int cfq_back_penalty = 2; | |
64100099 | 25 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 26 | static int cfq_slice_async = HZ / 25; |
64100099 | 27 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 28 | static int cfq_slice_idle = HZ / 125; |
22e2c507 | 29 | |
d9e7620e | 30 | /* |
0871714e | 31 | * offset from end of service tree |
d9e7620e | 32 | */ |
0871714e | 33 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
34 | |
35 | /* | |
36 | * below this threshold, we consider thinktime immediate | |
37 | */ | |
38 | #define CFQ_MIN_TT (2) | |
39 | ||
22e2c507 JA |
40 | #define CFQ_SLICE_SCALE (5) |
41 | ||
fe094d98 JA |
42 | #define RQ_CIC(rq) \ |
43 | ((struct cfq_io_context *) (rq)->elevator_private) | |
5e705374 | 44 | #define RQ_CFQQ(rq) ((rq)->elevator_private2) |
1da177e4 | 45 | |
e18b890b CL |
46 | static struct kmem_cache *cfq_pool; |
47 | static struct kmem_cache *cfq_ioc_pool; | |
1da177e4 | 48 | |
4050cf16 | 49 | static DEFINE_PER_CPU(unsigned long, ioc_count); |
334e94de | 50 | static struct completion *ioc_gone; |
9a11b4ed | 51 | static DEFINE_SPINLOCK(ioc_gone_lock); |
334e94de | 52 | |
22e2c507 JA |
53 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
54 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
55 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
56 | ||
3b18152c JA |
57 | #define ASYNC (0) |
58 | #define SYNC (1) | |
59 | ||
206dc69b JA |
60 | #define sample_valid(samples) ((samples) > 80) |
61 | ||
cc09e299 JA |
62 | /* |
63 | * Most of our rbtree usage is for sorting with min extraction, so | |
64 | * if we cache the leftmost node we don't have to walk down the tree | |
65 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
66 | * move this into the elevator for the rq sorting as well. | |
67 | */ | |
68 | struct cfq_rb_root { | |
69 | struct rb_root rb; | |
70 | struct rb_node *left; | |
71 | }; | |
72 | #define CFQ_RB_ROOT (struct cfq_rb_root) { RB_ROOT, NULL, } | |
73 | ||
22e2c507 JA |
74 | /* |
75 | * Per block device queue structure | |
76 | */ | |
1da177e4 | 77 | struct cfq_data { |
165125e1 | 78 | struct request_queue *queue; |
22e2c507 JA |
79 | |
80 | /* | |
81 | * rr list of queues with requests and the count of them | |
82 | */ | |
cc09e299 | 83 | struct cfq_rb_root service_tree; |
22e2c507 JA |
84 | unsigned int busy_queues; |
85 | ||
22e2c507 | 86 | int rq_in_driver; |
3ed9a296 | 87 | int sync_flight; |
25776e35 | 88 | int hw_tag; |
1da177e4 | 89 | |
22e2c507 JA |
90 | /* |
91 | * idle window management | |
92 | */ | |
93 | struct timer_list idle_slice_timer; | |
94 | struct work_struct unplug_work; | |
1da177e4 | 95 | |
22e2c507 JA |
96 | struct cfq_queue *active_queue; |
97 | struct cfq_io_context *active_cic; | |
22e2c507 | 98 | |
c2dea2d1 VT |
99 | /* |
100 | * async queue for each priority case | |
101 | */ | |
102 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
103 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 104 | |
6d048f53 | 105 | sector_t last_position; |
22e2c507 | 106 | unsigned long last_end_request; |
1da177e4 | 107 | |
1da177e4 LT |
108 | /* |
109 | * tunables, see top of file | |
110 | */ | |
111 | unsigned int cfq_quantum; | |
22e2c507 | 112 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
113 | unsigned int cfq_back_penalty; |
114 | unsigned int cfq_back_max; | |
22e2c507 JA |
115 | unsigned int cfq_slice[2]; |
116 | unsigned int cfq_slice_async_rq; | |
117 | unsigned int cfq_slice_idle; | |
d9ff4187 AV |
118 | |
119 | struct list_head cic_list; | |
1da177e4 LT |
120 | }; |
121 | ||
22e2c507 JA |
122 | /* |
123 | * Per process-grouping structure | |
124 | */ | |
1da177e4 LT |
125 | struct cfq_queue { |
126 | /* reference count */ | |
127 | atomic_t ref; | |
be754d2c RK |
128 | /* various state flags, see below */ |
129 | unsigned int flags; | |
1da177e4 LT |
130 | /* parent cfq_data */ |
131 | struct cfq_data *cfqd; | |
d9e7620e JA |
132 | /* service_tree member */ |
133 | struct rb_node rb_node; | |
134 | /* service_tree key */ | |
135 | unsigned long rb_key; | |
1da177e4 LT |
136 | /* sorted list of pending requests */ |
137 | struct rb_root sort_list; | |
138 | /* if fifo isn't expired, next request to serve */ | |
5e705374 | 139 | struct request *next_rq; |
1da177e4 LT |
140 | /* requests queued in sort_list */ |
141 | int queued[2]; | |
142 | /* currently allocated requests */ | |
143 | int allocated[2]; | |
144 | /* fifo list of requests in sort_list */ | |
22e2c507 | 145 | struct list_head fifo; |
1da177e4 | 146 | |
22e2c507 | 147 | unsigned long slice_end; |
c5b680f3 | 148 | long slice_resid; |
1da177e4 | 149 | |
be754d2c RK |
150 | /* pending metadata requests */ |
151 | int meta_pending; | |
6d048f53 JA |
152 | /* number of requests that are on the dispatch list or inside driver */ |
153 | int dispatched; | |
22e2c507 JA |
154 | |
155 | /* io prio of this group */ | |
156 | unsigned short ioprio, org_ioprio; | |
157 | unsigned short ioprio_class, org_ioprio_class; | |
158 | ||
1da177e4 LT |
159 | }; |
160 | ||
3b18152c | 161 | enum cfqq_state_flags { |
b0b8d749 JA |
162 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
163 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
164 | CFQ_CFQQ_FLAG_must_alloc, /* must be allowed rq alloc */ | |
165 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ | |
166 | CFQ_CFQQ_FLAG_must_dispatch, /* must dispatch, even if expired */ | |
167 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ | |
168 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
169 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
170 | CFQ_CFQQ_FLAG_queue_new, /* queue never been serviced */ | |
44f7c160 | 171 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 172 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
3b18152c JA |
173 | }; |
174 | ||
175 | #define CFQ_CFQQ_FNS(name) \ | |
176 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
177 | { \ | |
fe094d98 | 178 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
179 | } \ |
180 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
181 | { \ | |
fe094d98 | 182 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
183 | } \ |
184 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
185 | { \ | |
fe094d98 | 186 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
187 | } |
188 | ||
189 | CFQ_CFQQ_FNS(on_rr); | |
190 | CFQ_CFQQ_FNS(wait_request); | |
191 | CFQ_CFQQ_FNS(must_alloc); | |
192 | CFQ_CFQQ_FNS(must_alloc_slice); | |
193 | CFQ_CFQQ_FNS(must_dispatch); | |
194 | CFQ_CFQQ_FNS(fifo_expire); | |
195 | CFQ_CFQQ_FNS(idle_window); | |
196 | CFQ_CFQQ_FNS(prio_changed); | |
53b03744 | 197 | CFQ_CFQQ_FNS(queue_new); |
44f7c160 | 198 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 199 | CFQ_CFQQ_FNS(sync); |
3b18152c JA |
200 | #undef CFQ_CFQQ_FNS |
201 | ||
165125e1 | 202 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
91fac317 | 203 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, int, |
fd0928df | 204 | struct io_context *, gfp_t); |
4ac845a2 | 205 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
91fac317 VT |
206 | struct io_context *); |
207 | ||
208 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, | |
209 | int is_sync) | |
210 | { | |
211 | return cic->cfqq[!!is_sync]; | |
212 | } | |
213 | ||
214 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
215 | struct cfq_queue *cfqq, int is_sync) | |
216 | { | |
217 | cic->cfqq[!!is_sync] = cfqq; | |
218 | } | |
219 | ||
220 | /* | |
221 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
222 | * set (in which case it could also be direct WRITE). | |
223 | */ | |
224 | static inline int cfq_bio_sync(struct bio *bio) | |
225 | { | |
226 | if (bio_data_dir(bio) == READ || bio_sync(bio)) | |
227 | return 1; | |
228 | ||
229 | return 0; | |
230 | } | |
1da177e4 | 231 | |
99f95e52 AM |
232 | /* |
233 | * scheduler run of queue, if there are requests pending and no one in the | |
234 | * driver that will restart queueing | |
235 | */ | |
236 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
237 | { | |
7b14e3b5 | 238 | if (cfqd->busy_queues) |
99f95e52 AM |
239 | kblockd_schedule_work(&cfqd->unplug_work); |
240 | } | |
241 | ||
165125e1 | 242 | static int cfq_queue_empty(struct request_queue *q) |
99f95e52 AM |
243 | { |
244 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
245 | ||
b4878f24 | 246 | return !cfqd->busy_queues; |
99f95e52 AM |
247 | } |
248 | ||
44f7c160 JA |
249 | /* |
250 | * Scale schedule slice based on io priority. Use the sync time slice only | |
251 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
252 | * io only, should not get full sync slice length. | |
253 | */ | |
d9e7620e JA |
254 | static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync, |
255 | unsigned short prio) | |
44f7c160 | 256 | { |
d9e7620e | 257 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 258 | |
d9e7620e JA |
259 | WARN_ON(prio >= IOPRIO_BE_NR); |
260 | ||
261 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
262 | } | |
44f7c160 | 263 | |
d9e7620e JA |
264 | static inline int |
265 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
266 | { | |
267 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
268 | } |
269 | ||
270 | static inline void | |
271 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
272 | { | |
273 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; | |
274 | } | |
275 | ||
276 | /* | |
277 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
278 | * isn't valid until the first request from the dispatch is activated | |
279 | * and the slice time set. | |
280 | */ | |
281 | static inline int cfq_slice_used(struct cfq_queue *cfqq) | |
282 | { | |
283 | if (cfq_cfqq_slice_new(cfqq)) | |
284 | return 0; | |
285 | if (time_before(jiffies, cfqq->slice_end)) | |
286 | return 0; | |
287 | ||
288 | return 1; | |
289 | } | |
290 | ||
1da177e4 | 291 | /* |
5e705374 | 292 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 293 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 294 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 295 | */ |
5e705374 JA |
296 | static struct request * |
297 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
298 | { |
299 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 300 | unsigned long back_max; |
e8a99053 AM |
301 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
302 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
303 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 304 | |
5e705374 JA |
305 | if (rq1 == NULL || rq1 == rq2) |
306 | return rq2; | |
307 | if (rq2 == NULL) | |
308 | return rq1; | |
9c2c38a1 | 309 | |
5e705374 JA |
310 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
311 | return rq1; | |
312 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
313 | return rq2; | |
374f84ac JA |
314 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
315 | return rq1; | |
316 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
317 | return rq2; | |
1da177e4 | 318 | |
5e705374 JA |
319 | s1 = rq1->sector; |
320 | s2 = rq2->sector; | |
1da177e4 | 321 | |
6d048f53 | 322 | last = cfqd->last_position; |
1da177e4 | 323 | |
1da177e4 LT |
324 | /* |
325 | * by definition, 1KiB is 2 sectors | |
326 | */ | |
327 | back_max = cfqd->cfq_back_max * 2; | |
328 | ||
329 | /* | |
330 | * Strict one way elevator _except_ in the case where we allow | |
331 | * short backward seeks which are biased as twice the cost of a | |
332 | * similar forward seek. | |
333 | */ | |
334 | if (s1 >= last) | |
335 | d1 = s1 - last; | |
336 | else if (s1 + back_max >= last) | |
337 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
338 | else | |
e8a99053 | 339 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
340 | |
341 | if (s2 >= last) | |
342 | d2 = s2 - last; | |
343 | else if (s2 + back_max >= last) | |
344 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
345 | else | |
e8a99053 | 346 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
347 | |
348 | /* Found required data */ | |
e8a99053 AM |
349 | |
350 | /* | |
351 | * By doing switch() on the bit mask "wrap" we avoid having to | |
352 | * check two variables for all permutations: --> faster! | |
353 | */ | |
354 | switch (wrap) { | |
5e705374 | 355 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 356 | if (d1 < d2) |
5e705374 | 357 | return rq1; |
e8a99053 | 358 | else if (d2 < d1) |
5e705374 | 359 | return rq2; |
e8a99053 AM |
360 | else { |
361 | if (s1 >= s2) | |
5e705374 | 362 | return rq1; |
e8a99053 | 363 | else |
5e705374 | 364 | return rq2; |
e8a99053 | 365 | } |
1da177e4 | 366 | |
e8a99053 | 367 | case CFQ_RQ2_WRAP: |
5e705374 | 368 | return rq1; |
e8a99053 | 369 | case CFQ_RQ1_WRAP: |
5e705374 JA |
370 | return rq2; |
371 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
372 | default: |
373 | /* | |
374 | * Since both rqs are wrapped, | |
375 | * start with the one that's further behind head | |
376 | * (--> only *one* back seek required), | |
377 | * since back seek takes more time than forward. | |
378 | */ | |
379 | if (s1 <= s2) | |
5e705374 | 380 | return rq1; |
1da177e4 | 381 | else |
5e705374 | 382 | return rq2; |
1da177e4 LT |
383 | } |
384 | } | |
385 | ||
498d3aa2 JA |
386 | /* |
387 | * The below is leftmost cache rbtree addon | |
388 | */ | |
0871714e | 389 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 JA |
390 | { |
391 | if (!root->left) | |
392 | root->left = rb_first(&root->rb); | |
393 | ||
0871714e JA |
394 | if (root->left) |
395 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
396 | ||
397 | return NULL; | |
cc09e299 JA |
398 | } |
399 | ||
400 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) | |
401 | { | |
402 | if (root->left == n) | |
403 | root->left = NULL; | |
404 | ||
405 | rb_erase(n, &root->rb); | |
406 | RB_CLEAR_NODE(n); | |
407 | } | |
408 | ||
1da177e4 LT |
409 | /* |
410 | * would be nice to take fifo expire time into account as well | |
411 | */ | |
5e705374 JA |
412 | static struct request * |
413 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
414 | struct request *last) | |
1da177e4 | 415 | { |
21183b07 JA |
416 | struct rb_node *rbnext = rb_next(&last->rb_node); |
417 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 418 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 419 | |
21183b07 | 420 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
421 | |
422 | if (rbprev) | |
5e705374 | 423 | prev = rb_entry_rq(rbprev); |
1da177e4 | 424 | |
21183b07 | 425 | if (rbnext) |
5e705374 | 426 | next = rb_entry_rq(rbnext); |
21183b07 JA |
427 | else { |
428 | rbnext = rb_first(&cfqq->sort_list); | |
429 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 430 | next = rb_entry_rq(rbnext); |
21183b07 | 431 | } |
1da177e4 | 432 | |
21183b07 | 433 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
434 | } |
435 | ||
d9e7620e JA |
436 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
437 | struct cfq_queue *cfqq) | |
1da177e4 | 438 | { |
d9e7620e JA |
439 | /* |
440 | * just an approximation, should be ok. | |
441 | */ | |
67e6b49e JA |
442 | return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
443 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); | |
d9e7620e JA |
444 | } |
445 | ||
498d3aa2 JA |
446 | /* |
447 | * The cfqd->service_tree holds all pending cfq_queue's that have | |
448 | * requests waiting to be processed. It is sorted in the order that | |
449 | * we will service the queues. | |
450 | */ | |
d9e7620e | 451 | static void cfq_service_tree_add(struct cfq_data *cfqd, |
edd75ffd | 452 | struct cfq_queue *cfqq, int add_front) |
d9e7620e | 453 | { |
0871714e JA |
454 | struct rb_node **p, *parent; |
455 | struct cfq_queue *__cfqq; | |
d9e7620e | 456 | unsigned long rb_key; |
498d3aa2 | 457 | int left; |
d9e7620e | 458 | |
0871714e JA |
459 | if (cfq_class_idle(cfqq)) { |
460 | rb_key = CFQ_IDLE_DELAY; | |
461 | parent = rb_last(&cfqd->service_tree.rb); | |
462 | if (parent && parent != &cfqq->rb_node) { | |
463 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
464 | rb_key += __cfqq->rb_key; | |
465 | } else | |
466 | rb_key += jiffies; | |
467 | } else if (!add_front) { | |
edd75ffd JA |
468 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
469 | rb_key += cfqq->slice_resid; | |
470 | cfqq->slice_resid = 0; | |
471 | } else | |
472 | rb_key = 0; | |
1da177e4 | 473 | |
d9e7620e | 474 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
99f9628a | 475 | /* |
d9e7620e | 476 | * same position, nothing more to do |
99f9628a | 477 | */ |
d9e7620e JA |
478 | if (rb_key == cfqq->rb_key) |
479 | return; | |
1da177e4 | 480 | |
cc09e299 | 481 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); |
1da177e4 | 482 | } |
d9e7620e | 483 | |
498d3aa2 | 484 | left = 1; |
0871714e JA |
485 | parent = NULL; |
486 | p = &cfqd->service_tree.rb.rb_node; | |
d9e7620e | 487 | while (*p) { |
67060e37 | 488 | struct rb_node **n; |
cc09e299 | 489 | |
d9e7620e JA |
490 | parent = *p; |
491 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
492 | ||
0c534e0a JA |
493 | /* |
494 | * sort RT queues first, we always want to give | |
67060e37 JA |
495 | * preference to them. IDLE queues goes to the back. |
496 | * after that, sort on the next service time. | |
0c534e0a JA |
497 | */ |
498 | if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq)) | |
67060e37 | 499 | n = &(*p)->rb_left; |
0c534e0a | 500 | else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq)) |
67060e37 JA |
501 | n = &(*p)->rb_right; |
502 | else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq)) | |
503 | n = &(*p)->rb_left; | |
504 | else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq)) | |
505 | n = &(*p)->rb_right; | |
0c534e0a | 506 | else if (rb_key < __cfqq->rb_key) |
67060e37 JA |
507 | n = &(*p)->rb_left; |
508 | else | |
509 | n = &(*p)->rb_right; | |
510 | ||
511 | if (n == &(*p)->rb_right) | |
cc09e299 | 512 | left = 0; |
67060e37 JA |
513 | |
514 | p = n; | |
d9e7620e JA |
515 | } |
516 | ||
cc09e299 JA |
517 | if (left) |
518 | cfqd->service_tree.left = &cfqq->rb_node; | |
519 | ||
d9e7620e JA |
520 | cfqq->rb_key = rb_key; |
521 | rb_link_node(&cfqq->rb_node, parent, p); | |
cc09e299 | 522 | rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb); |
1da177e4 LT |
523 | } |
524 | ||
498d3aa2 JA |
525 | /* |
526 | * Update cfqq's position in the service tree. | |
527 | */ | |
edd75ffd | 528 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 529 | { |
6d048f53 JA |
530 | /* |
531 | * Resorting requires the cfqq to be on the RR list already. | |
532 | */ | |
498d3aa2 | 533 | if (cfq_cfqq_on_rr(cfqq)) |
edd75ffd | 534 | cfq_service_tree_add(cfqd, cfqq, 0); |
6d048f53 JA |
535 | } |
536 | ||
1da177e4 LT |
537 | /* |
538 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 539 | * the pending list according to last request service |
1da177e4 | 540 | */ |
febffd61 | 541 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 542 | { |
3b18152c JA |
543 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
544 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
545 | cfqd->busy_queues++; |
546 | ||
edd75ffd | 547 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
548 | } |
549 | ||
498d3aa2 JA |
550 | /* |
551 | * Called when the cfqq no longer has requests pending, remove it from | |
552 | * the service tree. | |
553 | */ | |
febffd61 | 554 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 555 | { |
3b18152c JA |
556 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
557 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 558 | |
cc09e299 JA |
559 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) |
560 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); | |
d9e7620e | 561 | |
1da177e4 LT |
562 | BUG_ON(!cfqd->busy_queues); |
563 | cfqd->busy_queues--; | |
564 | } | |
565 | ||
566 | /* | |
567 | * rb tree support functions | |
568 | */ | |
febffd61 | 569 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 570 | { |
5e705374 | 571 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 572 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 573 | const int sync = rq_is_sync(rq); |
1da177e4 | 574 | |
b4878f24 JA |
575 | BUG_ON(!cfqq->queued[sync]); |
576 | cfqq->queued[sync]--; | |
1da177e4 | 577 | |
5e705374 | 578 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 579 | |
dd67d051 | 580 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 581 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
582 | } |
583 | ||
5e705374 | 584 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 585 | { |
5e705374 | 586 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 587 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 588 | struct request *__alias; |
1da177e4 | 589 | |
5380a101 | 590 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
591 | |
592 | /* | |
593 | * looks a little odd, but the first insert might return an alias. | |
594 | * if that happens, put the alias on the dispatch list | |
595 | */ | |
21183b07 | 596 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 597 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
598 | |
599 | if (!cfq_cfqq_on_rr(cfqq)) | |
600 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
601 | |
602 | /* | |
603 | * check if this request is a better next-serve candidate | |
604 | */ | |
605 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); | |
606 | BUG_ON(!cfqq->next_rq); | |
1da177e4 LT |
607 | } |
608 | ||
febffd61 | 609 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 610 | { |
5380a101 JA |
611 | elv_rb_del(&cfqq->sort_list, rq); |
612 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 613 | cfq_add_rq_rb(rq); |
1da177e4 LT |
614 | } |
615 | ||
206dc69b JA |
616 | static struct request * |
617 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 618 | { |
206dc69b | 619 | struct task_struct *tsk = current; |
91fac317 | 620 | struct cfq_io_context *cic; |
206dc69b | 621 | struct cfq_queue *cfqq; |
1da177e4 | 622 | |
4ac845a2 | 623 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
624 | if (!cic) |
625 | return NULL; | |
626 | ||
627 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
628 | if (cfqq) { |
629 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
630 | ||
21183b07 | 631 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 632 | } |
1da177e4 | 633 | |
1da177e4 LT |
634 | return NULL; |
635 | } | |
636 | ||
165125e1 | 637 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 638 | { |
22e2c507 | 639 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 640 | |
b4878f24 | 641 | cfqd->rq_in_driver++; |
25776e35 JA |
642 | |
643 | /* | |
644 | * If the depth is larger 1, it really could be queueing. But lets | |
645 | * make the mark a little higher - idling could still be good for | |
646 | * low queueing, and a low queueing number could also just indicate | |
647 | * a SCSI mid layer like behaviour where limit+1 is often seen. | |
648 | */ | |
649 | if (!cfqd->hw_tag && cfqd->rq_in_driver > 4) | |
650 | cfqd->hw_tag = 1; | |
6d048f53 JA |
651 | |
652 | cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors; | |
1da177e4 LT |
653 | } |
654 | ||
165125e1 | 655 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 656 | { |
b4878f24 JA |
657 | struct cfq_data *cfqd = q->elevator->elevator_data; |
658 | ||
659 | WARN_ON(!cfqd->rq_in_driver); | |
660 | cfqd->rq_in_driver--; | |
1da177e4 LT |
661 | } |
662 | ||
b4878f24 | 663 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 664 | { |
5e705374 | 665 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 666 | |
5e705374 JA |
667 | if (cfqq->next_rq == rq) |
668 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 669 | |
b4878f24 | 670 | list_del_init(&rq->queuelist); |
5e705374 | 671 | cfq_del_rq_rb(rq); |
374f84ac JA |
672 | |
673 | if (rq_is_meta(rq)) { | |
674 | WARN_ON(!cfqq->meta_pending); | |
675 | cfqq->meta_pending--; | |
676 | } | |
1da177e4 LT |
677 | } |
678 | ||
165125e1 JA |
679 | static int cfq_merge(struct request_queue *q, struct request **req, |
680 | struct bio *bio) | |
1da177e4 LT |
681 | { |
682 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
683 | struct request *__rq; | |
1da177e4 | 684 | |
206dc69b | 685 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 686 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
687 | *req = __rq; |
688 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
689 | } |
690 | ||
691 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
692 | } |
693 | ||
165125e1 | 694 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 695 | int type) |
1da177e4 | 696 | { |
21183b07 | 697 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 698 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 699 | |
5e705374 | 700 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 701 | } |
1da177e4 LT |
702 | } |
703 | ||
704 | static void | |
165125e1 | 705 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
706 | struct request *next) |
707 | { | |
22e2c507 JA |
708 | /* |
709 | * reposition in fifo if next is older than rq | |
710 | */ | |
711 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
712 | time_before(next->start_time, rq->start_time)) | |
713 | list_move(&rq->queuelist, &next->queuelist); | |
714 | ||
b4878f24 | 715 | cfq_remove_request(next); |
22e2c507 JA |
716 | } |
717 | ||
165125e1 | 718 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
719 | struct bio *bio) |
720 | { | |
721 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
91fac317 | 722 | struct cfq_io_context *cic; |
da775265 | 723 | struct cfq_queue *cfqq; |
da775265 JA |
724 | |
725 | /* | |
ec8acb69 | 726 | * Disallow merge of a sync bio into an async request. |
da775265 | 727 | */ |
91fac317 | 728 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
da775265 JA |
729 | return 0; |
730 | ||
731 | /* | |
719d3402 JA |
732 | * Lookup the cfqq that this bio will be queued with. Allow |
733 | * merge only if rq is queued there. | |
da775265 | 734 | */ |
4ac845a2 | 735 | cic = cfq_cic_lookup(cfqd, current->io_context); |
91fac317 VT |
736 | if (!cic) |
737 | return 0; | |
719d3402 | 738 | |
91fac317 | 739 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
719d3402 JA |
740 | if (cfqq == RQ_CFQQ(rq)) |
741 | return 1; | |
da775265 | 742 | |
ec8acb69 | 743 | return 0; |
da775265 JA |
744 | } |
745 | ||
febffd61 JA |
746 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
747 | struct cfq_queue *cfqq) | |
22e2c507 JA |
748 | { |
749 | if (cfqq) { | |
22e2c507 | 750 | cfqq->slice_end = 0; |
3b18152c JA |
751 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
752 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 753 | cfq_mark_cfqq_slice_new(cfqq); |
1afba045 | 754 | cfq_clear_cfqq_queue_new(cfqq); |
22e2c507 JA |
755 | } |
756 | ||
757 | cfqd->active_queue = cfqq; | |
758 | } | |
759 | ||
7b14e3b5 JA |
760 | /* |
761 | * current cfqq expired its slice (or was too idle), select new one | |
762 | */ | |
763 | static void | |
764 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
6084cdda | 765 | int timed_out) |
7b14e3b5 | 766 | { |
7b14e3b5 JA |
767 | if (cfq_cfqq_wait_request(cfqq)) |
768 | del_timer(&cfqd->idle_slice_timer); | |
769 | ||
7b14e3b5 JA |
770 | cfq_clear_cfqq_must_dispatch(cfqq); |
771 | cfq_clear_cfqq_wait_request(cfqq); | |
772 | ||
773 | /* | |
6084cdda | 774 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 775 | */ |
3c6bd2f8 | 776 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) |
c5b680f3 | 777 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b14e3b5 | 778 | |
edd75ffd | 779 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
780 | |
781 | if (cfqq == cfqd->active_queue) | |
782 | cfqd->active_queue = NULL; | |
783 | ||
784 | if (cfqd->active_cic) { | |
785 | put_io_context(cfqd->active_cic->ioc); | |
786 | cfqd->active_cic = NULL; | |
787 | } | |
7b14e3b5 JA |
788 | } |
789 | ||
6084cdda | 790 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out) |
7b14e3b5 JA |
791 | { |
792 | struct cfq_queue *cfqq = cfqd->active_queue; | |
793 | ||
794 | if (cfqq) | |
6084cdda | 795 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
796 | } |
797 | ||
498d3aa2 JA |
798 | /* |
799 | * Get next queue for service. Unless we have a queue preemption, | |
800 | * we'll simply select the first cfqq in the service tree. | |
801 | */ | |
6d048f53 | 802 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 803 | { |
edd75ffd JA |
804 | if (RB_EMPTY_ROOT(&cfqd->service_tree.rb)) |
805 | return NULL; | |
d9e7620e | 806 | |
0871714e | 807 | return cfq_rb_first(&cfqd->service_tree); |
6d048f53 JA |
808 | } |
809 | ||
498d3aa2 JA |
810 | /* |
811 | * Get and set a new active queue for service. | |
812 | */ | |
6d048f53 JA |
813 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
814 | { | |
815 | struct cfq_queue *cfqq; | |
816 | ||
d9e7620e | 817 | cfqq = cfq_get_next_queue(cfqd); |
22e2c507 | 818 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 819 | return cfqq; |
22e2c507 JA |
820 | } |
821 | ||
d9e7620e JA |
822 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
823 | struct request *rq) | |
824 | { | |
825 | if (rq->sector >= cfqd->last_position) | |
826 | return rq->sector - cfqd->last_position; | |
827 | else | |
828 | return cfqd->last_position - rq->sector; | |
829 | } | |
830 | ||
6d048f53 JA |
831 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq) |
832 | { | |
833 | struct cfq_io_context *cic = cfqd->active_cic; | |
834 | ||
835 | if (!sample_valid(cic->seek_samples)) | |
836 | return 0; | |
837 | ||
838 | return cfq_dist_from_last(cfqd, rq) <= cic->seek_mean; | |
839 | } | |
840 | ||
d9e7620e JA |
841 | static int cfq_close_cooperator(struct cfq_data *cfq_data, |
842 | struct cfq_queue *cfqq) | |
6d048f53 | 843 | { |
6d048f53 | 844 | /* |
d9e7620e JA |
845 | * We should notice if some of the queues are cooperating, eg |
846 | * working closely on the same area of the disk. In that case, | |
847 | * we can group them together and don't waste time idling. | |
6d048f53 | 848 | */ |
d9e7620e | 849 | return 0; |
6d048f53 JA |
850 | } |
851 | ||
852 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (8 * 1024)) | |
caaa5f9f | 853 | |
6d048f53 | 854 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 855 | { |
1792669c | 856 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 857 | struct cfq_io_context *cic; |
7b14e3b5 JA |
858 | unsigned long sl; |
859 | ||
dd67d051 | 860 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 861 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
862 | |
863 | /* | |
864 | * idle is disabled, either manually or by past process history | |
865 | */ | |
6d048f53 JA |
866 | if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq)) |
867 | return; | |
868 | ||
22e2c507 JA |
869 | /* |
870 | * task has exited, don't wait | |
871 | */ | |
206dc69b | 872 | cic = cfqd->active_cic; |
66dac98e | 873 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
6d048f53 JA |
874 | return; |
875 | ||
876 | /* | |
877 | * See if this prio level has a good candidate | |
878 | */ | |
1afba045 JA |
879 | if (cfq_close_cooperator(cfqd, cfqq) && |
880 | (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2)) | |
6d048f53 | 881 | return; |
22e2c507 | 882 | |
3b18152c JA |
883 | cfq_mark_cfqq_must_dispatch(cfqq); |
884 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 885 | |
206dc69b JA |
886 | /* |
887 | * we don't want to idle for seeks, but we do want to allow | |
888 | * fair distribution of slice time for a process doing back-to-back | |
889 | * seeks. so allow a little bit of time for him to submit a new rq | |
890 | */ | |
6d048f53 | 891 | sl = cfqd->cfq_slice_idle; |
caaa5f9f | 892 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
d9e7620e | 893 | sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT)); |
206dc69b | 894 | |
7b14e3b5 | 895 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
1da177e4 LT |
896 | } |
897 | ||
498d3aa2 JA |
898 | /* |
899 | * Move request from internal lists to the request queue dispatch list. | |
900 | */ | |
165125e1 | 901 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 902 | { |
3ed9a296 | 903 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 904 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 905 | |
5380a101 | 906 | cfq_remove_request(rq); |
6d048f53 | 907 | cfqq->dispatched++; |
5380a101 | 908 | elv_dispatch_sort(q, rq); |
3ed9a296 JA |
909 | |
910 | if (cfq_cfqq_sync(cfqq)) | |
911 | cfqd->sync_flight++; | |
1da177e4 LT |
912 | } |
913 | ||
914 | /* | |
915 | * return expired entry, or NULL to just start from scratch in rbtree | |
916 | */ | |
febffd61 | 917 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
918 | { |
919 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 920 | struct request *rq; |
89850f7e | 921 | int fifo; |
1da177e4 | 922 | |
3b18152c | 923 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 924 | return NULL; |
cb887411 JA |
925 | |
926 | cfq_mark_cfqq_fifo_expire(cfqq); | |
927 | ||
89850f7e JA |
928 | if (list_empty(&cfqq->fifo)) |
929 | return NULL; | |
1da177e4 | 930 | |
6d048f53 | 931 | fifo = cfq_cfqq_sync(cfqq); |
89850f7e | 932 | rq = rq_entry_fifo(cfqq->fifo.next); |
1da177e4 | 933 | |
6d048f53 JA |
934 | if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) |
935 | return NULL; | |
1da177e4 | 936 | |
6d048f53 | 937 | return rq; |
1da177e4 LT |
938 | } |
939 | ||
22e2c507 JA |
940 | static inline int |
941 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
942 | { | |
943 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 944 | |
22e2c507 | 945 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 946 | |
22e2c507 | 947 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
948 | } |
949 | ||
22e2c507 | 950 | /* |
498d3aa2 JA |
951 | * Select a queue for service. If we have a current active queue, |
952 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 953 | */ |
1b5ed5e1 | 954 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 955 | { |
1da177e4 | 956 | struct cfq_queue *cfqq; |
1da177e4 | 957 | |
22e2c507 JA |
958 | cfqq = cfqd->active_queue; |
959 | if (!cfqq) | |
960 | goto new_queue; | |
1da177e4 | 961 | |
22e2c507 | 962 | /* |
6d048f53 | 963 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 964 | */ |
6d048f53 | 965 | if (cfq_slice_used(cfqq)) |
3b18152c | 966 | goto expire; |
1da177e4 | 967 | |
22e2c507 | 968 | /* |
6d048f53 JA |
969 | * The active queue has requests and isn't expired, allow it to |
970 | * dispatch. | |
22e2c507 | 971 | */ |
dd67d051 | 972 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 973 | goto keep_queue; |
6d048f53 JA |
974 | |
975 | /* | |
976 | * No requests pending. If the active queue still has requests in | |
977 | * flight or is idling for a new request, allow either of these | |
978 | * conditions to happen (or time out) before selecting a new queue. | |
979 | */ | |
cc197479 JA |
980 | if (timer_pending(&cfqd->idle_slice_timer) || |
981 | (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) { | |
caaa5f9f JA |
982 | cfqq = NULL; |
983 | goto keep_queue; | |
22e2c507 JA |
984 | } |
985 | ||
3b18152c | 986 | expire: |
6084cdda | 987 | cfq_slice_expired(cfqd, 0); |
3b18152c JA |
988 | new_queue: |
989 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 990 | keep_queue: |
3b18152c | 991 | return cfqq; |
22e2c507 JA |
992 | } |
993 | ||
498d3aa2 JA |
994 | /* |
995 | * Dispatch some requests from cfqq, moving them to the request queue | |
996 | * dispatch list. | |
997 | */ | |
22e2c507 JA |
998 | static int |
999 | __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1000 | int max_dispatch) | |
1001 | { | |
1002 | int dispatched = 0; | |
1003 | ||
dd67d051 | 1004 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
1005 | |
1006 | do { | |
5e705374 | 1007 | struct request *rq; |
1da177e4 LT |
1008 | |
1009 | /* | |
22e2c507 | 1010 | * follow expired path, else get first next available |
1da177e4 | 1011 | */ |
fe094d98 JA |
1012 | rq = cfq_check_fifo(cfqq); |
1013 | if (rq == NULL) | |
5e705374 | 1014 | rq = cfqq->next_rq; |
22e2c507 JA |
1015 | |
1016 | /* | |
1017 | * finally, insert request into driver dispatch list | |
1018 | */ | |
5e705374 | 1019 | cfq_dispatch_insert(cfqd->queue, rq); |
1da177e4 | 1020 | |
22e2c507 | 1021 | dispatched++; |
1da177e4 | 1022 | |
22e2c507 | 1023 | if (!cfqd->active_cic) { |
5e705374 JA |
1024 | atomic_inc(&RQ_CIC(rq)->ioc->refcount); |
1025 | cfqd->active_cic = RQ_CIC(rq); | |
22e2c507 | 1026 | } |
1da177e4 | 1027 | |
dd67d051 | 1028 | if (RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 JA |
1029 | break; |
1030 | ||
1031 | } while (dispatched < max_dispatch); | |
1032 | ||
22e2c507 JA |
1033 | /* |
1034 | * expire an async queue immediately if it has used up its slice. idle | |
1035 | * queue always expire after 1 dispatch round. | |
1036 | */ | |
a9938006 | 1037 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
20e493a8 | 1038 | dispatched >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
a9938006 | 1039 | cfq_class_idle(cfqq))) { |
44f7c160 | 1040 | cfqq->slice_end = jiffies + 1; |
6084cdda | 1041 | cfq_slice_expired(cfqd, 0); |
44f7c160 | 1042 | } |
22e2c507 JA |
1043 | |
1044 | return dispatched; | |
1045 | } | |
1046 | ||
febffd61 | 1047 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
1048 | { |
1049 | int dispatched = 0; | |
1050 | ||
1051 | while (cfqq->next_rq) { | |
1052 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1053 | dispatched++; | |
1054 | } | |
1055 | ||
1056 | BUG_ON(!list_empty(&cfqq->fifo)); | |
1057 | return dispatched; | |
1058 | } | |
1059 | ||
498d3aa2 JA |
1060 | /* |
1061 | * Drain our current requests. Used for barriers and when switching | |
1062 | * io schedulers on-the-fly. | |
1063 | */ | |
d9e7620e | 1064 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 1065 | { |
0871714e | 1066 | struct cfq_queue *cfqq; |
d9e7620e | 1067 | int dispatched = 0; |
1b5ed5e1 | 1068 | |
0871714e | 1069 | while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL) |
d9e7620e | 1070 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
1b5ed5e1 | 1071 | |
6084cdda | 1072 | cfq_slice_expired(cfqd, 0); |
1b5ed5e1 TH |
1073 | |
1074 | BUG_ON(cfqd->busy_queues); | |
1075 | ||
1076 | return dispatched; | |
1077 | } | |
1078 | ||
165125e1 | 1079 | static int cfq_dispatch_requests(struct request_queue *q, int force) |
22e2c507 JA |
1080 | { |
1081 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
6d048f53 | 1082 | struct cfq_queue *cfqq; |
caaa5f9f | 1083 | int dispatched; |
22e2c507 JA |
1084 | |
1085 | if (!cfqd->busy_queues) | |
1086 | return 0; | |
1087 | ||
1b5ed5e1 TH |
1088 | if (unlikely(force)) |
1089 | return cfq_forced_dispatch(cfqd); | |
1090 | ||
caaa5f9f | 1091 | dispatched = 0; |
caaa5f9f | 1092 | while ((cfqq = cfq_select_queue(cfqd)) != NULL) { |
b4878f24 JA |
1093 | int max_dispatch; |
1094 | ||
3ed9a296 JA |
1095 | max_dispatch = cfqd->cfq_quantum; |
1096 | if (cfq_class_idle(cfqq)) | |
1097 | max_dispatch = 1; | |
1098 | ||
1099 | if (cfqq->dispatched >= max_dispatch) { | |
1100 | if (cfqd->busy_queues > 1) | |
6d048f53 | 1101 | break; |
3ed9a296 | 1102 | if (cfqq->dispatched >= 4 * max_dispatch) |
a9938006 JA |
1103 | break; |
1104 | } | |
9ede209e | 1105 | |
3ed9a296 JA |
1106 | if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq)) |
1107 | break; | |
1108 | ||
3b18152c JA |
1109 | cfq_clear_cfqq_must_dispatch(cfqq); |
1110 | cfq_clear_cfqq_wait_request(cfqq); | |
22e2c507 JA |
1111 | del_timer(&cfqd->idle_slice_timer); |
1112 | ||
caaa5f9f | 1113 | dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); |
1da177e4 LT |
1114 | } |
1115 | ||
caaa5f9f | 1116 | return dispatched; |
1da177e4 LT |
1117 | } |
1118 | ||
1da177e4 | 1119 | /* |
5e705374 JA |
1120 | * task holds one reference to the queue, dropped when task exits. each rq |
1121 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1122 | * |
1123 | * queue lock must be held here. | |
1124 | */ | |
1125 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1126 | { | |
22e2c507 JA |
1127 | struct cfq_data *cfqd = cfqq->cfqd; |
1128 | ||
1129 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1130 | |
1131 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1132 | return; | |
1133 | ||
1134 | BUG_ON(rb_first(&cfqq->sort_list)); | |
22e2c507 | 1135 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1136 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1137 | |
28f95cbc | 1138 | if (unlikely(cfqd->active_queue == cfqq)) { |
6084cdda | 1139 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1140 | cfq_schedule_dispatch(cfqd); |
1141 | } | |
22e2c507 | 1142 | |
1da177e4 LT |
1143 | kmem_cache_free(cfq_pool, cfqq); |
1144 | } | |
1145 | ||
d6de8be7 JA |
1146 | /* |
1147 | * Must always be called with the rcu_read_lock() held | |
1148 | */ | |
07416d29 JA |
1149 | static void |
1150 | __call_for_each_cic(struct io_context *ioc, | |
1151 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1152 | { | |
1153 | struct cfq_io_context *cic; | |
1154 | struct hlist_node *n; | |
1155 | ||
1156 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) | |
1157 | func(ioc, cic); | |
1158 | } | |
1159 | ||
4ac845a2 | 1160 | /* |
34e6bbf2 | 1161 | * Call func for each cic attached to this ioc. |
4ac845a2 | 1162 | */ |
34e6bbf2 | 1163 | static void |
4ac845a2 JA |
1164 | call_for_each_cic(struct io_context *ioc, |
1165 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1da177e4 | 1166 | { |
4ac845a2 | 1167 | rcu_read_lock(); |
07416d29 | 1168 | __call_for_each_cic(ioc, func); |
4ac845a2 | 1169 | rcu_read_unlock(); |
34e6bbf2 FC |
1170 | } |
1171 | ||
1172 | static void cfq_cic_free_rcu(struct rcu_head *head) | |
1173 | { | |
1174 | struct cfq_io_context *cic; | |
1175 | ||
1176 | cic = container_of(head, struct cfq_io_context, rcu_head); | |
1177 | ||
1178 | kmem_cache_free(cfq_ioc_pool, cic); | |
1179 | elv_ioc_count_dec(ioc_count); | |
1180 | ||
9a11b4ed JA |
1181 | if (ioc_gone) { |
1182 | /* | |
1183 | * CFQ scheduler is exiting, grab exit lock and check | |
1184 | * the pending io context count. If it hits zero, | |
1185 | * complete ioc_gone and set it back to NULL | |
1186 | */ | |
1187 | spin_lock(&ioc_gone_lock); | |
1188 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) { | |
1189 | complete(ioc_gone); | |
1190 | ioc_gone = NULL; | |
1191 | } | |
1192 | spin_unlock(&ioc_gone_lock); | |
1193 | } | |
34e6bbf2 | 1194 | } |
4ac845a2 | 1195 | |
34e6bbf2 FC |
1196 | static void cfq_cic_free(struct cfq_io_context *cic) |
1197 | { | |
1198 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | |
4ac845a2 JA |
1199 | } |
1200 | ||
1201 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) | |
1202 | { | |
1203 | unsigned long flags; | |
1204 | ||
1205 | BUG_ON(!cic->dead_key); | |
1206 | ||
1207 | spin_lock_irqsave(&ioc->lock, flags); | |
1208 | radix_tree_delete(&ioc->radix_root, cic->dead_key); | |
ffc4e759 | 1209 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1210 | spin_unlock_irqrestore(&ioc->lock, flags); |
1211 | ||
34e6bbf2 | 1212 | cfq_cic_free(cic); |
4ac845a2 JA |
1213 | } |
1214 | ||
d6de8be7 JA |
1215 | /* |
1216 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. | |
1217 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), | |
1218 | * and ->trim() which is called with the task lock held | |
1219 | */ | |
4ac845a2 JA |
1220 | static void cfq_free_io_context(struct io_context *ioc) |
1221 | { | |
4ac845a2 | 1222 | /* |
34e6bbf2 FC |
1223 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
1224 | * so no more cic's are allowed to be linked into this ioc. So it | |
1225 | * should be ok to iterate over the known list, we will see all cic's | |
1226 | * since no new ones are added. | |
4ac845a2 | 1227 | */ |
07416d29 | 1228 | __call_for_each_cic(ioc, cic_free_func); |
1da177e4 LT |
1229 | } |
1230 | ||
89850f7e | 1231 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1232 | { |
28f95cbc | 1233 | if (unlikely(cfqq == cfqd->active_queue)) { |
6084cdda | 1234 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1235 | cfq_schedule_dispatch(cfqd); |
1236 | } | |
22e2c507 | 1237 | |
89850f7e JA |
1238 | cfq_put_queue(cfqq); |
1239 | } | |
22e2c507 | 1240 | |
89850f7e JA |
1241 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1242 | struct cfq_io_context *cic) | |
1243 | { | |
4faa3c81 FC |
1244 | struct io_context *ioc = cic->ioc; |
1245 | ||
fc46379d | 1246 | list_del_init(&cic->queue_list); |
4ac845a2 JA |
1247 | |
1248 | /* | |
1249 | * Make sure key == NULL is seen for dead queues | |
1250 | */ | |
fc46379d | 1251 | smp_wmb(); |
4ac845a2 | 1252 | cic->dead_key = (unsigned long) cic->key; |
fc46379d JA |
1253 | cic->key = NULL; |
1254 | ||
4faa3c81 FC |
1255 | if (ioc->ioc_data == cic) |
1256 | rcu_assign_pointer(ioc->ioc_data, NULL); | |
1257 | ||
12a05732 | 1258 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1259 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1260 | cic->cfqq[ASYNC] = NULL; |
1261 | } | |
1262 | ||
1263 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1264 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1265 | cic->cfqq[SYNC] = NULL; |
1266 | } | |
89850f7e JA |
1267 | } |
1268 | ||
4ac845a2 JA |
1269 | static void cfq_exit_single_io_context(struct io_context *ioc, |
1270 | struct cfq_io_context *cic) | |
89850f7e JA |
1271 | { |
1272 | struct cfq_data *cfqd = cic->key; | |
1273 | ||
89850f7e | 1274 | if (cfqd) { |
165125e1 | 1275 | struct request_queue *q = cfqd->queue; |
4ac845a2 | 1276 | unsigned long flags; |
89850f7e | 1277 | |
4ac845a2 | 1278 | spin_lock_irqsave(q->queue_lock, flags); |
89850f7e | 1279 | __cfq_exit_single_io_context(cfqd, cic); |
4ac845a2 | 1280 | spin_unlock_irqrestore(q->queue_lock, flags); |
89850f7e | 1281 | } |
1da177e4 LT |
1282 | } |
1283 | ||
498d3aa2 JA |
1284 | /* |
1285 | * The process that ioc belongs to has exited, we need to clean up | |
1286 | * and put the internal structures we have that belongs to that process. | |
1287 | */ | |
e2d74ac0 | 1288 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1289 | { |
4ac845a2 | 1290 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
1da177e4 LT |
1291 | } |
1292 | ||
22e2c507 | 1293 | static struct cfq_io_context * |
8267e268 | 1294 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1295 | { |
b5deef90 | 1296 | struct cfq_io_context *cic; |
1da177e4 | 1297 | |
94f6030c CL |
1298 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
1299 | cfqd->queue->node); | |
1da177e4 | 1300 | if (cic) { |
22e2c507 | 1301 | cic->last_end_request = jiffies; |
553698f9 | 1302 | INIT_LIST_HEAD(&cic->queue_list); |
ffc4e759 | 1303 | INIT_HLIST_NODE(&cic->cic_list); |
22e2c507 JA |
1304 | cic->dtor = cfq_free_io_context; |
1305 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1306 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1307 | } |
1308 | ||
1309 | return cic; | |
1310 | } | |
1311 | ||
fd0928df | 1312 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
1313 | { |
1314 | struct task_struct *tsk = current; | |
1315 | int ioprio_class; | |
1316 | ||
3b18152c | 1317 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1318 | return; |
1319 | ||
fd0928df | 1320 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 1321 | switch (ioprio_class) { |
fe094d98 JA |
1322 | default: |
1323 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1324 | case IOPRIO_CLASS_NONE: | |
1325 | /* | |
6d63c275 | 1326 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
1327 | */ |
1328 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 1329 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
1330 | break; |
1331 | case IOPRIO_CLASS_RT: | |
1332 | cfqq->ioprio = task_ioprio(ioc); | |
1333 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1334 | break; | |
1335 | case IOPRIO_CLASS_BE: | |
1336 | cfqq->ioprio = task_ioprio(ioc); | |
1337 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1338 | break; | |
1339 | case IOPRIO_CLASS_IDLE: | |
1340 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1341 | cfqq->ioprio = 7; | |
1342 | cfq_clear_cfqq_idle_window(cfqq); | |
1343 | break; | |
22e2c507 JA |
1344 | } |
1345 | ||
1346 | /* | |
1347 | * keep track of original prio settings in case we have to temporarily | |
1348 | * elevate the priority of this queue | |
1349 | */ | |
1350 | cfqq->org_ioprio = cfqq->ioprio; | |
1351 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 1352 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1353 | } |
1354 | ||
febffd61 | 1355 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
22e2c507 | 1356 | { |
478a82b0 AV |
1357 | struct cfq_data *cfqd = cic->key; |
1358 | struct cfq_queue *cfqq; | |
c1b707d2 | 1359 | unsigned long flags; |
35e6077c | 1360 | |
caaa5f9f JA |
1361 | if (unlikely(!cfqd)) |
1362 | return; | |
1363 | ||
c1b707d2 | 1364 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1365 | |
1366 | cfqq = cic->cfqq[ASYNC]; | |
1367 | if (cfqq) { | |
1368 | struct cfq_queue *new_cfqq; | |
fd0928df | 1369 | new_cfqq = cfq_get_queue(cfqd, ASYNC, cic->ioc, GFP_ATOMIC); |
caaa5f9f JA |
1370 | if (new_cfqq) { |
1371 | cic->cfqq[ASYNC] = new_cfqq; | |
1372 | cfq_put_queue(cfqq); | |
1373 | } | |
22e2c507 | 1374 | } |
caaa5f9f JA |
1375 | |
1376 | cfqq = cic->cfqq[SYNC]; | |
1377 | if (cfqq) | |
1378 | cfq_mark_cfqq_prio_changed(cfqq); | |
1379 | ||
c1b707d2 | 1380 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1381 | } |
1382 | ||
fc46379d | 1383 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1384 | { |
4ac845a2 | 1385 | call_for_each_cic(ioc, changed_ioprio); |
fc46379d | 1386 | ioc->ioprio_changed = 0; |
22e2c507 JA |
1387 | } |
1388 | ||
1389 | static struct cfq_queue * | |
15c31be4 | 1390 | cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync, |
fd0928df | 1391 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 1392 | { |
22e2c507 | 1393 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 1394 | struct cfq_io_context *cic; |
22e2c507 JA |
1395 | |
1396 | retry: | |
4ac845a2 | 1397 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
1398 | /* cic always exists here */ |
1399 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 JA |
1400 | |
1401 | if (!cfqq) { | |
1402 | if (new_cfqq) { | |
1403 | cfqq = new_cfqq; | |
1404 | new_cfqq = NULL; | |
1405 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1406 | /* |
1407 | * Inform the allocator of the fact that we will | |
1408 | * just repeat this allocation if it fails, to allow | |
1409 | * the allocator to do whatever it needs to attempt to | |
1410 | * free memory. | |
1411 | */ | |
22e2c507 | 1412 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c CL |
1413 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
1414 | gfp_mask | __GFP_NOFAIL | __GFP_ZERO, | |
1415 | cfqd->queue->node); | |
22e2c507 JA |
1416 | spin_lock_irq(cfqd->queue->queue_lock); |
1417 | goto retry; | |
1418 | } else { | |
94f6030c CL |
1419 | cfqq = kmem_cache_alloc_node(cfq_pool, |
1420 | gfp_mask | __GFP_ZERO, | |
1421 | cfqd->queue->node); | |
22e2c507 JA |
1422 | if (!cfqq) |
1423 | goto out; | |
1424 | } | |
1425 | ||
d9e7620e | 1426 | RB_CLEAR_NODE(&cfqq->rb_node); |
22e2c507 JA |
1427 | INIT_LIST_HEAD(&cfqq->fifo); |
1428 | ||
22e2c507 JA |
1429 | atomic_set(&cfqq->ref, 0); |
1430 | cfqq->cfqd = cfqd; | |
c5b680f3 | 1431 | |
3b18152c | 1432 | cfq_mark_cfqq_prio_changed(cfqq); |
53b03744 | 1433 | cfq_mark_cfqq_queue_new(cfqq); |
91fac317 | 1434 | |
fd0928df | 1435 | cfq_init_prio_data(cfqq, ioc); |
0871714e JA |
1436 | |
1437 | if (is_sync) { | |
1438 | if (!cfq_class_idle(cfqq)) | |
1439 | cfq_mark_cfqq_idle_window(cfqq); | |
1440 | cfq_mark_cfqq_sync(cfqq); | |
1441 | } | |
22e2c507 JA |
1442 | } |
1443 | ||
1444 | if (new_cfqq) | |
1445 | kmem_cache_free(cfq_pool, new_cfqq); | |
1446 | ||
22e2c507 JA |
1447 | out: |
1448 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1449 | return cfqq; | |
1450 | } | |
1451 | ||
c2dea2d1 VT |
1452 | static struct cfq_queue ** |
1453 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
1454 | { | |
fe094d98 | 1455 | switch (ioprio_class) { |
c2dea2d1 VT |
1456 | case IOPRIO_CLASS_RT: |
1457 | return &cfqd->async_cfqq[0][ioprio]; | |
1458 | case IOPRIO_CLASS_BE: | |
1459 | return &cfqd->async_cfqq[1][ioprio]; | |
1460 | case IOPRIO_CLASS_IDLE: | |
1461 | return &cfqd->async_idle_cfqq; | |
1462 | default: | |
1463 | BUG(); | |
1464 | } | |
1465 | } | |
1466 | ||
15c31be4 | 1467 | static struct cfq_queue * |
fd0928df | 1468 | cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc, |
15c31be4 JA |
1469 | gfp_t gfp_mask) |
1470 | { | |
fd0928df JA |
1471 | const int ioprio = task_ioprio(ioc); |
1472 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 1473 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
1474 | struct cfq_queue *cfqq = NULL; |
1475 | ||
c2dea2d1 VT |
1476 | if (!is_sync) { |
1477 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
1478 | cfqq = *async_cfqq; | |
1479 | } | |
1480 | ||
0a0836a0 | 1481 | if (!cfqq) { |
fd0928df | 1482 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
0a0836a0 ON |
1483 | if (!cfqq) |
1484 | return NULL; | |
1485 | } | |
15c31be4 JA |
1486 | |
1487 | /* | |
1488 | * pin the queue now that it's allocated, scheduler exit will prune it | |
1489 | */ | |
c2dea2d1 | 1490 | if (!is_sync && !(*async_cfqq)) { |
15c31be4 | 1491 | atomic_inc(&cfqq->ref); |
c2dea2d1 | 1492 | *async_cfqq = cfqq; |
15c31be4 JA |
1493 | } |
1494 | ||
1495 | atomic_inc(&cfqq->ref); | |
1496 | return cfqq; | |
1497 | } | |
1498 | ||
498d3aa2 JA |
1499 | /* |
1500 | * We drop cfq io contexts lazily, so we may find a dead one. | |
1501 | */ | |
dbecf3ab | 1502 | static void |
4ac845a2 JA |
1503 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
1504 | struct cfq_io_context *cic) | |
dbecf3ab | 1505 | { |
4ac845a2 JA |
1506 | unsigned long flags; |
1507 | ||
fc46379d | 1508 | WARN_ON(!list_empty(&cic->queue_list)); |
597bc485 | 1509 | |
4ac845a2 JA |
1510 | spin_lock_irqsave(&ioc->lock, flags); |
1511 | ||
4faa3c81 | 1512 | BUG_ON(ioc->ioc_data == cic); |
597bc485 | 1513 | |
4ac845a2 | 1514 | radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd); |
ffc4e759 | 1515 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1516 | spin_unlock_irqrestore(&ioc->lock, flags); |
1517 | ||
1518 | cfq_cic_free(cic); | |
dbecf3ab OH |
1519 | } |
1520 | ||
e2d74ac0 | 1521 | static struct cfq_io_context * |
4ac845a2 | 1522 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
e2d74ac0 | 1523 | { |
e2d74ac0 | 1524 | struct cfq_io_context *cic; |
d6de8be7 | 1525 | unsigned long flags; |
4ac845a2 | 1526 | void *k; |
e2d74ac0 | 1527 | |
91fac317 VT |
1528 | if (unlikely(!ioc)) |
1529 | return NULL; | |
1530 | ||
d6de8be7 JA |
1531 | rcu_read_lock(); |
1532 | ||
597bc485 JA |
1533 | /* |
1534 | * we maintain a last-hit cache, to avoid browsing over the tree | |
1535 | */ | |
4ac845a2 | 1536 | cic = rcu_dereference(ioc->ioc_data); |
d6de8be7 JA |
1537 | if (cic && cic->key == cfqd) { |
1538 | rcu_read_unlock(); | |
597bc485 | 1539 | return cic; |
d6de8be7 | 1540 | } |
597bc485 | 1541 | |
4ac845a2 | 1542 | do { |
4ac845a2 JA |
1543 | cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd); |
1544 | rcu_read_unlock(); | |
1545 | if (!cic) | |
1546 | break; | |
be3b0753 OH |
1547 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1548 | k = cic->key; | |
1549 | if (unlikely(!k)) { | |
4ac845a2 | 1550 | cfq_drop_dead_cic(cfqd, ioc, cic); |
d6de8be7 | 1551 | rcu_read_lock(); |
4ac845a2 | 1552 | continue; |
dbecf3ab | 1553 | } |
e2d74ac0 | 1554 | |
d6de8be7 | 1555 | spin_lock_irqsave(&ioc->lock, flags); |
4ac845a2 | 1556 | rcu_assign_pointer(ioc->ioc_data, cic); |
d6de8be7 | 1557 | spin_unlock_irqrestore(&ioc->lock, flags); |
4ac845a2 JA |
1558 | break; |
1559 | } while (1); | |
e2d74ac0 | 1560 | |
4ac845a2 | 1561 | return cic; |
e2d74ac0 JA |
1562 | } |
1563 | ||
4ac845a2 JA |
1564 | /* |
1565 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup | |
1566 | * the process specific cfq io context when entered from the block layer. | |
1567 | * Also adds the cic to a per-cfqd list, used when this queue is removed. | |
1568 | */ | |
febffd61 JA |
1569 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
1570 | struct cfq_io_context *cic, gfp_t gfp_mask) | |
e2d74ac0 | 1571 | { |
0261d688 | 1572 | unsigned long flags; |
4ac845a2 | 1573 | int ret; |
e2d74ac0 | 1574 | |
4ac845a2 JA |
1575 | ret = radix_tree_preload(gfp_mask); |
1576 | if (!ret) { | |
1577 | cic->ioc = ioc; | |
1578 | cic->key = cfqd; | |
e2d74ac0 | 1579 | |
4ac845a2 JA |
1580 | spin_lock_irqsave(&ioc->lock, flags); |
1581 | ret = radix_tree_insert(&ioc->radix_root, | |
1582 | (unsigned long) cfqd, cic); | |
ffc4e759 JA |
1583 | if (!ret) |
1584 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); | |
4ac845a2 | 1585 | spin_unlock_irqrestore(&ioc->lock, flags); |
e2d74ac0 | 1586 | |
4ac845a2 JA |
1587 | radix_tree_preload_end(); |
1588 | ||
1589 | if (!ret) { | |
1590 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1591 | list_add(&cic->queue_list, &cfqd->cic_list); | |
1592 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1593 | } | |
e2d74ac0 JA |
1594 | } |
1595 | ||
4ac845a2 JA |
1596 | if (ret) |
1597 | printk(KERN_ERR "cfq: cic link failed!\n"); | |
fc46379d | 1598 | |
4ac845a2 | 1599 | return ret; |
e2d74ac0 JA |
1600 | } |
1601 | ||
1da177e4 LT |
1602 | /* |
1603 | * Setup general io context and cfq io context. There can be several cfq | |
1604 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1605 | * than one device managed by cfq. |
1da177e4 LT |
1606 | */ |
1607 | static struct cfq_io_context * | |
e2d74ac0 | 1608 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1609 | { |
22e2c507 | 1610 | struct io_context *ioc = NULL; |
1da177e4 | 1611 | struct cfq_io_context *cic; |
1da177e4 | 1612 | |
22e2c507 | 1613 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1614 | |
b5deef90 | 1615 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1616 | if (!ioc) |
1617 | return NULL; | |
1618 | ||
4ac845a2 | 1619 | cic = cfq_cic_lookup(cfqd, ioc); |
e2d74ac0 JA |
1620 | if (cic) |
1621 | goto out; | |
1da177e4 | 1622 | |
e2d74ac0 JA |
1623 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1624 | if (cic == NULL) | |
1625 | goto err; | |
1da177e4 | 1626 | |
4ac845a2 JA |
1627 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) |
1628 | goto err_free; | |
1629 | ||
1da177e4 | 1630 | out: |
fc46379d JA |
1631 | smp_read_barrier_depends(); |
1632 | if (unlikely(ioc->ioprio_changed)) | |
1633 | cfq_ioc_set_ioprio(ioc); | |
1634 | ||
1da177e4 | 1635 | return cic; |
4ac845a2 JA |
1636 | err_free: |
1637 | cfq_cic_free(cic); | |
1da177e4 LT |
1638 | err: |
1639 | put_io_context(ioc); | |
1640 | return NULL; | |
1641 | } | |
1642 | ||
22e2c507 JA |
1643 | static void |
1644 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1645 | { |
aaf1228d JA |
1646 | unsigned long elapsed = jiffies - cic->last_end_request; |
1647 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 1648 | |
22e2c507 JA |
1649 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1650 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1651 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1652 | } | |
1da177e4 | 1653 | |
206dc69b | 1654 | static void |
6d048f53 JA |
1655 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic, |
1656 | struct request *rq) | |
206dc69b JA |
1657 | { |
1658 | sector_t sdist; | |
1659 | u64 total; | |
1660 | ||
5e705374 JA |
1661 | if (cic->last_request_pos < rq->sector) |
1662 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1663 | else |
5e705374 | 1664 | sdist = cic->last_request_pos - rq->sector; |
206dc69b JA |
1665 | |
1666 | /* | |
1667 | * Don't allow the seek distance to get too large from the | |
1668 | * odd fragment, pagein, etc | |
1669 | */ | |
1670 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1671 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1672 | else | |
1673 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1674 | ||
1675 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1676 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1677 | total = cic->seek_total + (cic->seek_samples/2); | |
1678 | do_div(total, cic->seek_samples); | |
1679 | cic->seek_mean = (sector_t)total; | |
1680 | } | |
1da177e4 | 1681 | |
22e2c507 JA |
1682 | /* |
1683 | * Disable idle window if the process thinks too long or seeks so much that | |
1684 | * it doesn't matter | |
1685 | */ | |
1686 | static void | |
1687 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1688 | struct cfq_io_context *cic) | |
1689 | { | |
1be92f2f JA |
1690 | int enable_idle; |
1691 | ||
0871714e JA |
1692 | /* |
1693 | * Don't idle for async or idle io prio class | |
1694 | */ | |
1695 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
1696 | return; |
1697 | ||
1698 | enable_idle = cfq_cfqq_idle_window(cfqq); | |
1da177e4 | 1699 | |
66dac98e | 1700 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
caaa5f9f | 1701 | (cfqd->hw_tag && CIC_SEEKY(cic))) |
22e2c507 JA |
1702 | enable_idle = 0; |
1703 | else if (sample_valid(cic->ttime_samples)) { | |
1704 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1705 | enable_idle = 0; | |
1706 | else | |
1707 | enable_idle = 1; | |
1da177e4 LT |
1708 | } |
1709 | ||
3b18152c JA |
1710 | if (enable_idle) |
1711 | cfq_mark_cfqq_idle_window(cfqq); | |
1712 | else | |
1713 | cfq_clear_cfqq_idle_window(cfqq); | |
22e2c507 | 1714 | } |
1da177e4 | 1715 | |
22e2c507 JA |
1716 | /* |
1717 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1718 | * no or if we aren't sure, a 1 will cause a preempt. | |
1719 | */ | |
1720 | static int | |
1721 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1722 | struct request *rq) |
22e2c507 | 1723 | { |
6d048f53 | 1724 | struct cfq_queue *cfqq; |
22e2c507 | 1725 | |
6d048f53 JA |
1726 | cfqq = cfqd->active_queue; |
1727 | if (!cfqq) | |
22e2c507 JA |
1728 | return 0; |
1729 | ||
6d048f53 JA |
1730 | if (cfq_slice_used(cfqq)) |
1731 | return 1; | |
1732 | ||
1733 | if (cfq_class_idle(new_cfqq)) | |
caaa5f9f | 1734 | return 0; |
22e2c507 JA |
1735 | |
1736 | if (cfq_class_idle(cfqq)) | |
1737 | return 1; | |
1e3335de | 1738 | |
374f84ac JA |
1739 | /* |
1740 | * if the new request is sync, but the currently running queue is | |
1741 | * not, let the sync request have priority. | |
1742 | */ | |
5e705374 | 1743 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1744 | return 1; |
1e3335de | 1745 | |
374f84ac JA |
1746 | /* |
1747 | * So both queues are sync. Let the new request get disk time if | |
1748 | * it's a metadata request and the current queue is doing regular IO. | |
1749 | */ | |
1750 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1751 | return 1; | |
22e2c507 | 1752 | |
1e3335de JA |
1753 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
1754 | return 0; | |
1755 | ||
1756 | /* | |
1757 | * if this request is as-good as one we would expect from the | |
1758 | * current cfqq, let it preempt | |
1759 | */ | |
6d048f53 | 1760 | if (cfq_rq_close(cfqd, rq)) |
1e3335de JA |
1761 | return 1; |
1762 | ||
22e2c507 JA |
1763 | return 0; |
1764 | } | |
1765 | ||
1766 | /* | |
1767 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1768 | * let it have half of its nominal slice. | |
1769 | */ | |
1770 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1771 | { | |
6084cdda | 1772 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 1773 | |
bf572256 JA |
1774 | /* |
1775 | * Put the new queue at the front of the of the current list, | |
1776 | * so we know that it will be selected next. | |
1777 | */ | |
1778 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
1779 | |
1780 | cfq_service_tree_add(cfqd, cfqq, 1); | |
bf572256 | 1781 | |
44f7c160 JA |
1782 | cfqq->slice_end = 0; |
1783 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
1784 | } |
1785 | ||
22e2c507 | 1786 | /* |
5e705374 | 1787 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1788 | * something we should do about it |
1789 | */ | |
1790 | static void | |
5e705374 JA |
1791 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1792 | struct request *rq) | |
22e2c507 | 1793 | { |
5e705374 | 1794 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1795 | |
374f84ac JA |
1796 | if (rq_is_meta(rq)) |
1797 | cfqq->meta_pending++; | |
1798 | ||
9c2c38a1 | 1799 | cfq_update_io_thinktime(cfqd, cic); |
6d048f53 | 1800 | cfq_update_io_seektime(cfqd, cic, rq); |
9c2c38a1 JA |
1801 | cfq_update_idle_window(cfqd, cfqq, cic); |
1802 | ||
5e705374 | 1803 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
22e2c507 JA |
1804 | |
1805 | if (cfqq == cfqd->active_queue) { | |
1806 | /* | |
1807 | * if we are waiting for a request for this queue, let it rip | |
1808 | * immediately and flag that we must not expire this queue | |
1809 | * just now | |
1810 | */ | |
3b18152c JA |
1811 | if (cfq_cfqq_wait_request(cfqq)) { |
1812 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 | 1813 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1814 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1815 | } |
5e705374 | 1816 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1817 | /* |
1818 | * not the active queue - expire current slice if it is | |
1819 | * idle and has expired it's mean thinktime or this new queue | |
1820 | * has some old slice time left and is of higher priority | |
1821 | */ | |
1822 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1823 | cfq_mark_cfqq_must_dispatch(cfqq); |
dc72ef4a | 1824 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1825 | } |
1da177e4 LT |
1826 | } |
1827 | ||
165125e1 | 1828 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1829 | { |
b4878f24 | 1830 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1831 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1832 | |
fd0928df | 1833 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
1da177e4 | 1834 | |
5e705374 | 1835 | cfq_add_rq_rb(rq); |
1da177e4 | 1836 | |
22e2c507 JA |
1837 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1838 | ||
5e705374 | 1839 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1840 | } |
1841 | ||
165125e1 | 1842 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1843 | { |
5e705374 | 1844 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1845 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1846 | const int sync = rq_is_sync(rq); |
b4878f24 | 1847 | unsigned long now; |
1da177e4 | 1848 | |
b4878f24 | 1849 | now = jiffies; |
1da177e4 | 1850 | |
b4878f24 | 1851 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 1852 | WARN_ON(!cfqq->dispatched); |
b4878f24 | 1853 | cfqd->rq_in_driver--; |
6d048f53 | 1854 | cfqq->dispatched--; |
1da177e4 | 1855 | |
3ed9a296 JA |
1856 | if (cfq_cfqq_sync(cfqq)) |
1857 | cfqd->sync_flight--; | |
1858 | ||
b4878f24 JA |
1859 | if (!cfq_class_idle(cfqq)) |
1860 | cfqd->last_end_request = now; | |
3b18152c | 1861 | |
caaa5f9f | 1862 | if (sync) |
5e705374 | 1863 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1864 | |
1865 | /* | |
1866 | * If this is the active queue, check if it needs to be expired, | |
1867 | * or if we want to idle in case it has no pending requests. | |
1868 | */ | |
1869 | if (cfqd->active_queue == cfqq) { | |
44f7c160 JA |
1870 | if (cfq_cfqq_slice_new(cfqq)) { |
1871 | cfq_set_prio_slice(cfqd, cfqq); | |
1872 | cfq_clear_cfqq_slice_new(cfqq); | |
1873 | } | |
0871714e | 1874 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
6084cdda | 1875 | cfq_slice_expired(cfqd, 1); |
6d048f53 JA |
1876 | else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1877 | cfq_arm_slice_timer(cfqd); | |
caaa5f9f | 1878 | } |
6d048f53 JA |
1879 | |
1880 | if (!cfqd->rq_in_driver) | |
1881 | cfq_schedule_dispatch(cfqd); | |
1da177e4 LT |
1882 | } |
1883 | ||
22e2c507 JA |
1884 | /* |
1885 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
1886 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
1887 | */ | |
1888 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 1889 | { |
22e2c507 JA |
1890 | if (has_fs_excl()) { |
1891 | /* | |
1892 | * boost idle prio on transactions that would lock out other | |
1893 | * users of the filesystem | |
1894 | */ | |
1895 | if (cfq_class_idle(cfqq)) | |
1896 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1897 | if (cfqq->ioprio > IOPRIO_NORM) | |
1898 | cfqq->ioprio = IOPRIO_NORM; | |
1899 | } else { | |
1900 | /* | |
1901 | * check if we need to unboost the queue | |
1902 | */ | |
1903 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
1904 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
1905 | if (cfqq->ioprio != cfqq->org_ioprio) | |
1906 | cfqq->ioprio = cfqq->org_ioprio; | |
1907 | } | |
22e2c507 | 1908 | } |
1da177e4 | 1909 | |
89850f7e | 1910 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 1911 | { |
3b18152c | 1912 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 1913 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 1914 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 1915 | return ELV_MQUEUE_MUST; |
3b18152c | 1916 | } |
1da177e4 | 1917 | |
22e2c507 | 1918 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
1919 | } |
1920 | ||
165125e1 | 1921 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
1922 | { |
1923 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1924 | struct task_struct *tsk = current; | |
91fac317 | 1925 | struct cfq_io_context *cic; |
22e2c507 JA |
1926 | struct cfq_queue *cfqq; |
1927 | ||
1928 | /* | |
1929 | * don't force setup of a queue from here, as a call to may_queue | |
1930 | * does not necessarily imply that a request actually will be queued. | |
1931 | * so just lookup a possibly existing queue, or return 'may queue' | |
1932 | * if that fails | |
1933 | */ | |
4ac845a2 | 1934 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1935 | if (!cic) |
1936 | return ELV_MQUEUE_MAY; | |
1937 | ||
1938 | cfqq = cic_to_cfqq(cic, rw & REQ_RW_SYNC); | |
22e2c507 | 1939 | if (cfqq) { |
fd0928df | 1940 | cfq_init_prio_data(cfqq, cic->ioc); |
22e2c507 JA |
1941 | cfq_prio_boost(cfqq); |
1942 | ||
89850f7e | 1943 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
1944 | } |
1945 | ||
1946 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
1947 | } |
1948 | ||
1da177e4 LT |
1949 | /* |
1950 | * queue lock held here | |
1951 | */ | |
bb37b94c | 1952 | static void cfq_put_request(struct request *rq) |
1da177e4 | 1953 | { |
5e705374 | 1954 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1955 | |
5e705374 | 1956 | if (cfqq) { |
22e2c507 | 1957 | const int rw = rq_data_dir(rq); |
1da177e4 | 1958 | |
22e2c507 JA |
1959 | BUG_ON(!cfqq->allocated[rw]); |
1960 | cfqq->allocated[rw]--; | |
1da177e4 | 1961 | |
5e705374 | 1962 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 1963 | |
1da177e4 | 1964 | rq->elevator_private = NULL; |
5e705374 | 1965 | rq->elevator_private2 = NULL; |
1da177e4 | 1966 | |
1da177e4 LT |
1967 | cfq_put_queue(cfqq); |
1968 | } | |
1969 | } | |
1970 | ||
1971 | /* | |
22e2c507 | 1972 | * Allocate cfq data structures associated with this request. |
1da177e4 | 1973 | */ |
22e2c507 | 1974 | static int |
165125e1 | 1975 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
1976 | { |
1977 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1978 | struct cfq_io_context *cic; | |
1979 | const int rw = rq_data_dir(rq); | |
7749a8d4 | 1980 | const int is_sync = rq_is_sync(rq); |
22e2c507 | 1981 | struct cfq_queue *cfqq; |
1da177e4 LT |
1982 | unsigned long flags; |
1983 | ||
1984 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1985 | ||
e2d74ac0 | 1986 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 1987 | |
1da177e4 LT |
1988 | spin_lock_irqsave(q->queue_lock, flags); |
1989 | ||
22e2c507 JA |
1990 | if (!cic) |
1991 | goto queue_fail; | |
1992 | ||
91fac317 VT |
1993 | cfqq = cic_to_cfqq(cic, is_sync); |
1994 | if (!cfqq) { | |
fd0928df | 1995 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
91fac317 | 1996 | |
22e2c507 JA |
1997 | if (!cfqq) |
1998 | goto queue_fail; | |
1da177e4 | 1999 | |
91fac317 VT |
2000 | cic_set_cfqq(cic, cfqq, is_sync); |
2001 | } | |
1da177e4 LT |
2002 | |
2003 | cfqq->allocated[rw]++; | |
3b18152c | 2004 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 2005 | atomic_inc(&cfqq->ref); |
1da177e4 | 2006 | |
5e705374 | 2007 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 2008 | |
5e705374 JA |
2009 | rq->elevator_private = cic; |
2010 | rq->elevator_private2 = cfqq; | |
2011 | return 0; | |
1da177e4 | 2012 | |
22e2c507 JA |
2013 | queue_fail: |
2014 | if (cic) | |
2015 | put_io_context(cic->ioc); | |
89850f7e | 2016 | |
3b18152c | 2017 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
2018 | spin_unlock_irqrestore(q->queue_lock, flags); |
2019 | return 1; | |
2020 | } | |
2021 | ||
65f27f38 | 2022 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 2023 | { |
65f27f38 DH |
2024 | struct cfq_data *cfqd = |
2025 | container_of(work, struct cfq_data, unplug_work); | |
165125e1 | 2026 | struct request_queue *q = cfqd->queue; |
22e2c507 JA |
2027 | unsigned long flags; |
2028 | ||
2029 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 2030 | blk_start_queueing(q); |
22e2c507 JA |
2031 | spin_unlock_irqrestore(q->queue_lock, flags); |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * Timer running if the active_queue is currently idling inside its time slice | |
2036 | */ | |
2037 | static void cfq_idle_slice_timer(unsigned long data) | |
2038 | { | |
2039 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
2040 | struct cfq_queue *cfqq; | |
2041 | unsigned long flags; | |
3c6bd2f8 | 2042 | int timed_out = 1; |
22e2c507 JA |
2043 | |
2044 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
2045 | ||
fe094d98 JA |
2046 | cfqq = cfqd->active_queue; |
2047 | if (cfqq) { | |
3c6bd2f8 JA |
2048 | timed_out = 0; |
2049 | ||
22e2c507 JA |
2050 | /* |
2051 | * expired | |
2052 | */ | |
44f7c160 | 2053 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
2054 | goto expire; |
2055 | ||
2056 | /* | |
2057 | * only expire and reinvoke request handler, if there are | |
2058 | * other queues with pending requests | |
2059 | */ | |
caaa5f9f | 2060 | if (!cfqd->busy_queues) |
22e2c507 | 2061 | goto out_cont; |
22e2c507 JA |
2062 | |
2063 | /* | |
2064 | * not expired and it has a request pending, let it dispatch | |
2065 | */ | |
dd67d051 | 2066 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 2067 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
2068 | goto out_kick; |
2069 | } | |
2070 | } | |
2071 | expire: | |
6084cdda | 2072 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 2073 | out_kick: |
3b18152c | 2074 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
2075 | out_cont: |
2076 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
2077 | } | |
2078 | ||
3b18152c JA |
2079 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
2080 | { | |
2081 | del_timer_sync(&cfqd->idle_slice_timer); | |
4310864b | 2082 | kblockd_flush_work(&cfqd->unplug_work); |
3b18152c | 2083 | } |
22e2c507 | 2084 | |
c2dea2d1 VT |
2085 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
2086 | { | |
2087 | int i; | |
2088 | ||
2089 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
2090 | if (cfqd->async_cfqq[0][i]) | |
2091 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
2092 | if (cfqd->async_cfqq[1][i]) | |
2093 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 2094 | } |
2389d1ef ON |
2095 | |
2096 | if (cfqd->async_idle_cfqq) | |
2097 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
2098 | } |
2099 | ||
1da177e4 LT |
2100 | static void cfq_exit_queue(elevator_t *e) |
2101 | { | |
22e2c507 | 2102 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 2103 | struct request_queue *q = cfqd->queue; |
22e2c507 | 2104 | |
3b18152c | 2105 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 2106 | |
d9ff4187 | 2107 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 2108 | |
d9ff4187 | 2109 | if (cfqd->active_queue) |
6084cdda | 2110 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
2111 | |
2112 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
2113 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
2114 | struct cfq_io_context, | |
2115 | queue_list); | |
89850f7e JA |
2116 | |
2117 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 2118 | } |
e2d74ac0 | 2119 | |
c2dea2d1 | 2120 | cfq_put_async_queues(cfqd); |
15c31be4 | 2121 | |
d9ff4187 | 2122 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
2123 | |
2124 | cfq_shutdown_timer_wq(cfqd); | |
2125 | ||
a90d742e | 2126 | kfree(cfqd); |
1da177e4 LT |
2127 | } |
2128 | ||
165125e1 | 2129 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
2130 | { |
2131 | struct cfq_data *cfqd; | |
1da177e4 | 2132 | |
94f6030c | 2133 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
1da177e4 | 2134 | if (!cfqd) |
bc1c1169 | 2135 | return NULL; |
1da177e4 | 2136 | |
cc09e299 | 2137 | cfqd->service_tree = CFQ_RB_ROOT; |
d9ff4187 | 2138 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2139 | |
1da177e4 | 2140 | cfqd->queue = q; |
1da177e4 | 2141 | |
22e2c507 JA |
2142 | init_timer(&cfqd->idle_slice_timer); |
2143 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2144 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2145 | ||
65f27f38 | 2146 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2147 | |
b70c864d | 2148 | cfqd->last_end_request = jiffies; |
1da177e4 | 2149 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2150 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2151 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2152 | cfqd->cfq_back_max = cfq_back_max; |
2153 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2154 | cfqd->cfq_slice[0] = cfq_slice_async; |
2155 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2156 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2157 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
3b18152c | 2158 | |
bc1c1169 | 2159 | return cfqd; |
1da177e4 LT |
2160 | } |
2161 | ||
2162 | static void cfq_slab_kill(void) | |
2163 | { | |
d6de8be7 JA |
2164 | /* |
2165 | * Caller already ensured that pending RCU callbacks are completed, | |
2166 | * so we should have no busy allocations at this point. | |
2167 | */ | |
1da177e4 LT |
2168 | if (cfq_pool) |
2169 | kmem_cache_destroy(cfq_pool); | |
2170 | if (cfq_ioc_pool) | |
2171 | kmem_cache_destroy(cfq_ioc_pool); | |
2172 | } | |
2173 | ||
2174 | static int __init cfq_slab_setup(void) | |
2175 | { | |
0a31bd5f | 2176 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
1da177e4 LT |
2177 | if (!cfq_pool) |
2178 | goto fail; | |
2179 | ||
34e6bbf2 | 2180 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
1da177e4 LT |
2181 | if (!cfq_ioc_pool) |
2182 | goto fail; | |
2183 | ||
2184 | return 0; | |
2185 | fail: | |
2186 | cfq_slab_kill(); | |
2187 | return -ENOMEM; | |
2188 | } | |
2189 | ||
1da177e4 LT |
2190 | /* |
2191 | * sysfs parts below --> | |
2192 | */ | |
1da177e4 LT |
2193 | static ssize_t |
2194 | cfq_var_show(unsigned int var, char *page) | |
2195 | { | |
2196 | return sprintf(page, "%d\n", var); | |
2197 | } | |
2198 | ||
2199 | static ssize_t | |
2200 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2201 | { | |
2202 | char *p = (char *) page; | |
2203 | ||
2204 | *var = simple_strtoul(p, &p, 10); | |
2205 | return count; | |
2206 | } | |
2207 | ||
1da177e4 | 2208 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
3d1ab40f | 2209 | static ssize_t __FUNC(elevator_t *e, char *page) \ |
1da177e4 | 2210 | { \ |
3d1ab40f | 2211 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2212 | unsigned int __data = __VAR; \ |
2213 | if (__CONV) \ | |
2214 | __data = jiffies_to_msecs(__data); \ | |
2215 | return cfq_var_show(__data, (page)); \ | |
2216 | } | |
2217 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2218 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2219 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2220 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2221 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2222 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2223 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2224 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2225 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2226 | #undef SHOW_FUNCTION |
2227 | ||
2228 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
3d1ab40f | 2229 | static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \ |
1da177e4 | 2230 | { \ |
3d1ab40f | 2231 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2232 | unsigned int __data; \ |
2233 | int ret = cfq_var_store(&__data, (page), count); \ | |
2234 | if (__data < (MIN)) \ | |
2235 | __data = (MIN); \ | |
2236 | else if (__data > (MAX)) \ | |
2237 | __data = (MAX); \ | |
2238 | if (__CONV) \ | |
2239 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2240 | else \ | |
2241 | *(__PTR) = __data; \ | |
2242 | return ret; \ | |
2243 | } | |
2244 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
2245 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
2246 | UINT_MAX, 1); | |
2247 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
2248 | UINT_MAX, 1); | |
e572ec7e | 2249 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
2250 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
2251 | UINT_MAX, 0); | |
22e2c507 JA |
2252 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2253 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2254 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
2255 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
2256 | UINT_MAX, 0); | |
1da177e4 LT |
2257 | #undef STORE_FUNCTION |
2258 | ||
e572ec7e AV |
2259 | #define CFQ_ATTR(name) \ |
2260 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2261 | ||
2262 | static struct elv_fs_entry cfq_attrs[] = { | |
2263 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2264 | CFQ_ATTR(fifo_expire_sync), |
2265 | CFQ_ATTR(fifo_expire_async), | |
2266 | CFQ_ATTR(back_seek_max), | |
2267 | CFQ_ATTR(back_seek_penalty), | |
2268 | CFQ_ATTR(slice_sync), | |
2269 | CFQ_ATTR(slice_async), | |
2270 | CFQ_ATTR(slice_async_rq), | |
2271 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2272 | __ATTR_NULL |
1da177e4 LT |
2273 | }; |
2274 | ||
1da177e4 LT |
2275 | static struct elevator_type iosched_cfq = { |
2276 | .ops = { | |
2277 | .elevator_merge_fn = cfq_merge, | |
2278 | .elevator_merged_fn = cfq_merged_request, | |
2279 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2280 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2281 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2282 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2283 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2284 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2285 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2286 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2287 | .elevator_former_req_fn = elv_rb_former_request, |
2288 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2289 | .elevator_set_req_fn = cfq_set_request, |
2290 | .elevator_put_req_fn = cfq_put_request, | |
2291 | .elevator_may_queue_fn = cfq_may_queue, | |
2292 | .elevator_init_fn = cfq_init_queue, | |
2293 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2294 | .trim = cfq_free_io_context, |
1da177e4 | 2295 | }, |
3d1ab40f | 2296 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2297 | .elevator_name = "cfq", |
2298 | .elevator_owner = THIS_MODULE, | |
2299 | }; | |
2300 | ||
2301 | static int __init cfq_init(void) | |
2302 | { | |
22e2c507 JA |
2303 | /* |
2304 | * could be 0 on HZ < 1000 setups | |
2305 | */ | |
2306 | if (!cfq_slice_async) | |
2307 | cfq_slice_async = 1; | |
2308 | if (!cfq_slice_idle) | |
2309 | cfq_slice_idle = 1; | |
2310 | ||
1da177e4 LT |
2311 | if (cfq_slab_setup()) |
2312 | return -ENOMEM; | |
2313 | ||
2fdd82bd | 2314 | elv_register(&iosched_cfq); |
1da177e4 | 2315 | |
2fdd82bd | 2316 | return 0; |
1da177e4 LT |
2317 | } |
2318 | ||
2319 | static void __exit cfq_exit(void) | |
2320 | { | |
6e9a4738 | 2321 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2322 | elv_unregister(&iosched_cfq); |
334e94de | 2323 | ioc_gone = &all_gone; |
fba82272 OH |
2324 | /* ioc_gone's update must be visible before reading ioc_count */ |
2325 | smp_wmb(); | |
d6de8be7 JA |
2326 | |
2327 | /* | |
2328 | * this also protects us from entering cfq_slab_kill() with | |
2329 | * pending RCU callbacks | |
2330 | */ | |
4050cf16 | 2331 | if (elv_ioc_count_read(ioc_count)) |
9a11b4ed | 2332 | wait_for_completion(&all_gone); |
83521d3e | 2333 | cfq_slab_kill(); |
1da177e4 LT |
2334 | } |
2335 | ||
2336 | module_init(cfq_init); | |
2337 | module_exit(cfq_exit); | |
2338 | ||
2339 | MODULE_AUTHOR("Jens Axboe"); | |
2340 | MODULE_LICENSE("GPL"); | |
2341 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |