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> |
7b679138 | 14 | #include <linux/blktrace_api.h> |
1da177e4 LT |
15 | |
16 | /* | |
17 | * tunables | |
18 | */ | |
fe094d98 JA |
19 | /* max queue in one round of service */ |
20 | static const int cfq_quantum = 4; | |
64100099 | 21 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
22 | /* maximum backwards seek, in KiB */ |
23 | static const int cfq_back_max = 16 * 1024; | |
24 | /* penalty of a backwards seek */ | |
25 | static const int cfq_back_penalty = 2; | |
64100099 | 26 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 27 | static int cfq_slice_async = HZ / 25; |
64100099 | 28 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 29 | static int cfq_slice_idle = HZ / 125; |
22e2c507 | 30 | |
d9e7620e | 31 | /* |
0871714e | 32 | * offset from end of service tree |
d9e7620e | 33 | */ |
0871714e | 34 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
35 | |
36 | /* | |
37 | * below this threshold, we consider thinktime immediate | |
38 | */ | |
39 | #define CFQ_MIN_TT (2) | |
40 | ||
22e2c507 JA |
41 | #define CFQ_SLICE_SCALE (5) |
42 | ||
fe094d98 JA |
43 | #define RQ_CIC(rq) \ |
44 | ((struct cfq_io_context *) (rq)->elevator_private) | |
7b679138 | 45 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private2) |
1da177e4 | 46 | |
e18b890b CL |
47 | static struct kmem_cache *cfq_pool; |
48 | static struct kmem_cache *cfq_ioc_pool; | |
1da177e4 | 49 | |
4050cf16 | 50 | static DEFINE_PER_CPU(unsigned long, ioc_count); |
334e94de | 51 | static struct completion *ioc_gone; |
9a11b4ed | 52 | static DEFINE_SPINLOCK(ioc_gone_lock); |
334e94de | 53 | |
22e2c507 JA |
54 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
55 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
56 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
57 | ||
3b18152c JA |
58 | #define ASYNC (0) |
59 | #define SYNC (1) | |
60 | ||
206dc69b JA |
61 | #define sample_valid(samples) ((samples) > 80) |
62 | ||
cc09e299 JA |
63 | /* |
64 | * Most of our rbtree usage is for sorting with min extraction, so | |
65 | * if we cache the leftmost node we don't have to walk down the tree | |
66 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
67 | * move this into the elevator for the rq sorting as well. | |
68 | */ | |
69 | struct cfq_rb_root { | |
70 | struct rb_root rb; | |
71 | struct rb_node *left; | |
72 | }; | |
73 | #define CFQ_RB_ROOT (struct cfq_rb_root) { RB_ROOT, NULL, } | |
74 | ||
22e2c507 JA |
75 | /* |
76 | * Per block device queue structure | |
77 | */ | |
1da177e4 | 78 | struct cfq_data { |
165125e1 | 79 | struct request_queue *queue; |
22e2c507 JA |
80 | |
81 | /* | |
82 | * rr list of queues with requests and the count of them | |
83 | */ | |
cc09e299 | 84 | struct cfq_rb_root service_tree; |
22e2c507 JA |
85 | unsigned int busy_queues; |
86 | ||
22e2c507 | 87 | int rq_in_driver; |
3ed9a296 | 88 | int sync_flight; |
25776e35 | 89 | int hw_tag; |
1da177e4 | 90 | |
22e2c507 JA |
91 | /* |
92 | * idle window management | |
93 | */ | |
94 | struct timer_list idle_slice_timer; | |
95 | struct work_struct unplug_work; | |
1da177e4 | 96 | |
22e2c507 JA |
97 | struct cfq_queue *active_queue; |
98 | struct cfq_io_context *active_cic; | |
22e2c507 | 99 | |
c2dea2d1 VT |
100 | /* |
101 | * async queue for each priority case | |
102 | */ | |
103 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
104 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 105 | |
6d048f53 | 106 | sector_t last_position; |
22e2c507 | 107 | unsigned long last_end_request; |
1da177e4 | 108 | |
1da177e4 LT |
109 | /* |
110 | * tunables, see top of file | |
111 | */ | |
112 | unsigned int cfq_quantum; | |
22e2c507 | 113 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
114 | unsigned int cfq_back_penalty; |
115 | unsigned int cfq_back_max; | |
22e2c507 JA |
116 | unsigned int cfq_slice[2]; |
117 | unsigned int cfq_slice_async_rq; | |
118 | unsigned int cfq_slice_idle; | |
d9ff4187 AV |
119 | |
120 | struct list_head cic_list; | |
1da177e4 LT |
121 | }; |
122 | ||
22e2c507 JA |
123 | /* |
124 | * Per process-grouping structure | |
125 | */ | |
1da177e4 LT |
126 | struct cfq_queue { |
127 | /* reference count */ | |
128 | atomic_t ref; | |
be754d2c RK |
129 | /* various state flags, see below */ |
130 | unsigned int flags; | |
1da177e4 LT |
131 | /* parent cfq_data */ |
132 | struct cfq_data *cfqd; | |
d9e7620e JA |
133 | /* service_tree member */ |
134 | struct rb_node rb_node; | |
135 | /* service_tree key */ | |
136 | unsigned long rb_key; | |
1da177e4 LT |
137 | /* sorted list of pending requests */ |
138 | struct rb_root sort_list; | |
139 | /* if fifo isn't expired, next request to serve */ | |
5e705374 | 140 | struct request *next_rq; |
1da177e4 LT |
141 | /* requests queued in sort_list */ |
142 | int queued[2]; | |
143 | /* currently allocated requests */ | |
144 | int allocated[2]; | |
145 | /* fifo list of requests in sort_list */ | |
22e2c507 | 146 | struct list_head fifo; |
1da177e4 | 147 | |
22e2c507 | 148 | unsigned long slice_end; |
c5b680f3 | 149 | long slice_resid; |
1da177e4 | 150 | |
be754d2c RK |
151 | /* pending metadata requests */ |
152 | int meta_pending; | |
6d048f53 JA |
153 | /* number of requests that are on the dispatch list or inside driver */ |
154 | int dispatched; | |
22e2c507 JA |
155 | |
156 | /* io prio of this group */ | |
157 | unsigned short ioprio, org_ioprio; | |
158 | unsigned short ioprio_class, org_ioprio_class; | |
159 | ||
7b679138 | 160 | pid_t pid; |
1da177e4 LT |
161 | }; |
162 | ||
3b18152c | 163 | enum cfqq_state_flags { |
b0b8d749 JA |
164 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
165 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
166 | CFQ_CFQQ_FLAG_must_alloc, /* must be allowed rq alloc */ | |
167 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ | |
168 | CFQ_CFQQ_FLAG_must_dispatch, /* must dispatch, even if expired */ | |
169 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ | |
170 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
171 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
172 | CFQ_CFQQ_FLAG_queue_new, /* queue never been serviced */ | |
44f7c160 | 173 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 174 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
3b18152c JA |
175 | }; |
176 | ||
177 | #define CFQ_CFQQ_FNS(name) \ | |
178 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
179 | { \ | |
fe094d98 | 180 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
181 | } \ |
182 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
183 | { \ | |
fe094d98 | 184 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
185 | } \ |
186 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
187 | { \ | |
fe094d98 | 188 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
189 | } |
190 | ||
191 | CFQ_CFQQ_FNS(on_rr); | |
192 | CFQ_CFQQ_FNS(wait_request); | |
193 | CFQ_CFQQ_FNS(must_alloc); | |
194 | CFQ_CFQQ_FNS(must_alloc_slice); | |
195 | CFQ_CFQQ_FNS(must_dispatch); | |
196 | CFQ_CFQQ_FNS(fifo_expire); | |
197 | CFQ_CFQQ_FNS(idle_window); | |
198 | CFQ_CFQQ_FNS(prio_changed); | |
53b03744 | 199 | CFQ_CFQQ_FNS(queue_new); |
44f7c160 | 200 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 201 | CFQ_CFQQ_FNS(sync); |
3b18152c JA |
202 | #undef CFQ_CFQQ_FNS |
203 | ||
7b679138 JA |
204 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
205 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
206 | #define cfq_log(cfqd, fmt, args...) \ | |
207 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
208 | ||
165125e1 | 209 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
91fac317 | 210 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, int, |
fd0928df | 211 | struct io_context *, gfp_t); |
4ac845a2 | 212 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
91fac317 VT |
213 | struct io_context *); |
214 | ||
215 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, | |
216 | int is_sync) | |
217 | { | |
218 | return cic->cfqq[!!is_sync]; | |
219 | } | |
220 | ||
221 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
222 | struct cfq_queue *cfqq, int is_sync) | |
223 | { | |
224 | cic->cfqq[!!is_sync] = cfqq; | |
225 | } | |
226 | ||
227 | /* | |
228 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
229 | * set (in which case it could also be direct WRITE). | |
230 | */ | |
231 | static inline int cfq_bio_sync(struct bio *bio) | |
232 | { | |
233 | if (bio_data_dir(bio) == READ || bio_sync(bio)) | |
234 | return 1; | |
235 | ||
236 | return 0; | |
237 | } | |
1da177e4 | 238 | |
99f95e52 AM |
239 | /* |
240 | * scheduler run of queue, if there are requests pending and no one in the | |
241 | * driver that will restart queueing | |
242 | */ | |
243 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
244 | { | |
7b679138 JA |
245 | if (cfqd->busy_queues) { |
246 | cfq_log(cfqd, "schedule dispatch"); | |
99f95e52 | 247 | kblockd_schedule_work(&cfqd->unplug_work); |
7b679138 | 248 | } |
99f95e52 AM |
249 | } |
250 | ||
165125e1 | 251 | static int cfq_queue_empty(struct request_queue *q) |
99f95e52 AM |
252 | { |
253 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
254 | ||
b4878f24 | 255 | return !cfqd->busy_queues; |
99f95e52 AM |
256 | } |
257 | ||
44f7c160 JA |
258 | /* |
259 | * Scale schedule slice based on io priority. Use the sync time slice only | |
260 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
261 | * io only, should not get full sync slice length. | |
262 | */ | |
d9e7620e JA |
263 | static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync, |
264 | unsigned short prio) | |
44f7c160 | 265 | { |
d9e7620e | 266 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 267 | |
d9e7620e JA |
268 | WARN_ON(prio >= IOPRIO_BE_NR); |
269 | ||
270 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
271 | } | |
44f7c160 | 272 | |
d9e7620e JA |
273 | static inline int |
274 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
275 | { | |
276 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
277 | } |
278 | ||
279 | static inline void | |
280 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
281 | { | |
282 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; | |
7b679138 | 283 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
284 | } |
285 | ||
286 | /* | |
287 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
288 | * isn't valid until the first request from the dispatch is activated | |
289 | * and the slice time set. | |
290 | */ | |
291 | static inline int cfq_slice_used(struct cfq_queue *cfqq) | |
292 | { | |
293 | if (cfq_cfqq_slice_new(cfqq)) | |
294 | return 0; | |
295 | if (time_before(jiffies, cfqq->slice_end)) | |
296 | return 0; | |
297 | ||
298 | return 1; | |
299 | } | |
300 | ||
1da177e4 | 301 | /* |
5e705374 | 302 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 303 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 304 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 305 | */ |
5e705374 JA |
306 | static struct request * |
307 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
308 | { |
309 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 310 | unsigned long back_max; |
e8a99053 AM |
311 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
312 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
313 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 314 | |
5e705374 JA |
315 | if (rq1 == NULL || rq1 == rq2) |
316 | return rq2; | |
317 | if (rq2 == NULL) | |
318 | return rq1; | |
9c2c38a1 | 319 | |
5e705374 JA |
320 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
321 | return rq1; | |
322 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
323 | return rq2; | |
374f84ac JA |
324 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
325 | return rq1; | |
326 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
327 | return rq2; | |
1da177e4 | 328 | |
5e705374 JA |
329 | s1 = rq1->sector; |
330 | s2 = rq2->sector; | |
1da177e4 | 331 | |
6d048f53 | 332 | last = cfqd->last_position; |
1da177e4 | 333 | |
1da177e4 LT |
334 | /* |
335 | * by definition, 1KiB is 2 sectors | |
336 | */ | |
337 | back_max = cfqd->cfq_back_max * 2; | |
338 | ||
339 | /* | |
340 | * Strict one way elevator _except_ in the case where we allow | |
341 | * short backward seeks which are biased as twice the cost of a | |
342 | * similar forward seek. | |
343 | */ | |
344 | if (s1 >= last) | |
345 | d1 = s1 - last; | |
346 | else if (s1 + back_max >= last) | |
347 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
348 | else | |
e8a99053 | 349 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
350 | |
351 | if (s2 >= last) | |
352 | d2 = s2 - last; | |
353 | else if (s2 + back_max >= last) | |
354 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
355 | else | |
e8a99053 | 356 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
357 | |
358 | /* Found required data */ | |
e8a99053 AM |
359 | |
360 | /* | |
361 | * By doing switch() on the bit mask "wrap" we avoid having to | |
362 | * check two variables for all permutations: --> faster! | |
363 | */ | |
364 | switch (wrap) { | |
5e705374 | 365 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 366 | if (d1 < d2) |
5e705374 | 367 | return rq1; |
e8a99053 | 368 | else if (d2 < d1) |
5e705374 | 369 | return rq2; |
e8a99053 AM |
370 | else { |
371 | if (s1 >= s2) | |
5e705374 | 372 | return rq1; |
e8a99053 | 373 | else |
5e705374 | 374 | return rq2; |
e8a99053 | 375 | } |
1da177e4 | 376 | |
e8a99053 | 377 | case CFQ_RQ2_WRAP: |
5e705374 | 378 | return rq1; |
e8a99053 | 379 | case CFQ_RQ1_WRAP: |
5e705374 JA |
380 | return rq2; |
381 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
382 | default: |
383 | /* | |
384 | * Since both rqs are wrapped, | |
385 | * start with the one that's further behind head | |
386 | * (--> only *one* back seek required), | |
387 | * since back seek takes more time than forward. | |
388 | */ | |
389 | if (s1 <= s2) | |
5e705374 | 390 | return rq1; |
1da177e4 | 391 | else |
5e705374 | 392 | return rq2; |
1da177e4 LT |
393 | } |
394 | } | |
395 | ||
498d3aa2 JA |
396 | /* |
397 | * The below is leftmost cache rbtree addon | |
398 | */ | |
0871714e | 399 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 JA |
400 | { |
401 | if (!root->left) | |
402 | root->left = rb_first(&root->rb); | |
403 | ||
0871714e JA |
404 | if (root->left) |
405 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
406 | ||
407 | return NULL; | |
cc09e299 JA |
408 | } |
409 | ||
410 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) | |
411 | { | |
412 | if (root->left == n) | |
413 | root->left = NULL; | |
414 | ||
415 | rb_erase(n, &root->rb); | |
416 | RB_CLEAR_NODE(n); | |
417 | } | |
418 | ||
1da177e4 LT |
419 | /* |
420 | * would be nice to take fifo expire time into account as well | |
421 | */ | |
5e705374 JA |
422 | static struct request * |
423 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
424 | struct request *last) | |
1da177e4 | 425 | { |
21183b07 JA |
426 | struct rb_node *rbnext = rb_next(&last->rb_node); |
427 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 428 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 429 | |
21183b07 | 430 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
431 | |
432 | if (rbprev) | |
5e705374 | 433 | prev = rb_entry_rq(rbprev); |
1da177e4 | 434 | |
21183b07 | 435 | if (rbnext) |
5e705374 | 436 | next = rb_entry_rq(rbnext); |
21183b07 JA |
437 | else { |
438 | rbnext = rb_first(&cfqq->sort_list); | |
439 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 440 | next = rb_entry_rq(rbnext); |
21183b07 | 441 | } |
1da177e4 | 442 | |
21183b07 | 443 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
444 | } |
445 | ||
d9e7620e JA |
446 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
447 | struct cfq_queue *cfqq) | |
1da177e4 | 448 | { |
d9e7620e JA |
449 | /* |
450 | * just an approximation, should be ok. | |
451 | */ | |
67e6b49e JA |
452 | return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
453 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); | |
d9e7620e JA |
454 | } |
455 | ||
498d3aa2 JA |
456 | /* |
457 | * The cfqd->service_tree holds all pending cfq_queue's that have | |
458 | * requests waiting to be processed. It is sorted in the order that | |
459 | * we will service the queues. | |
460 | */ | |
d9e7620e | 461 | static void cfq_service_tree_add(struct cfq_data *cfqd, |
edd75ffd | 462 | struct cfq_queue *cfqq, int add_front) |
d9e7620e | 463 | { |
0871714e JA |
464 | struct rb_node **p, *parent; |
465 | struct cfq_queue *__cfqq; | |
d9e7620e | 466 | unsigned long rb_key; |
498d3aa2 | 467 | int left; |
d9e7620e | 468 | |
0871714e JA |
469 | if (cfq_class_idle(cfqq)) { |
470 | rb_key = CFQ_IDLE_DELAY; | |
471 | parent = rb_last(&cfqd->service_tree.rb); | |
472 | if (parent && parent != &cfqq->rb_node) { | |
473 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
474 | rb_key += __cfqq->rb_key; | |
475 | } else | |
476 | rb_key += jiffies; | |
477 | } else if (!add_front) { | |
edd75ffd JA |
478 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
479 | rb_key += cfqq->slice_resid; | |
480 | cfqq->slice_resid = 0; | |
481 | } else | |
482 | rb_key = 0; | |
1da177e4 | 483 | |
d9e7620e | 484 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
99f9628a | 485 | /* |
d9e7620e | 486 | * same position, nothing more to do |
99f9628a | 487 | */ |
d9e7620e JA |
488 | if (rb_key == cfqq->rb_key) |
489 | return; | |
1da177e4 | 490 | |
cc09e299 | 491 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); |
1da177e4 | 492 | } |
d9e7620e | 493 | |
498d3aa2 | 494 | left = 1; |
0871714e JA |
495 | parent = NULL; |
496 | p = &cfqd->service_tree.rb.rb_node; | |
d9e7620e | 497 | while (*p) { |
67060e37 | 498 | struct rb_node **n; |
cc09e299 | 499 | |
d9e7620e JA |
500 | parent = *p; |
501 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
502 | ||
0c534e0a JA |
503 | /* |
504 | * sort RT queues first, we always want to give | |
67060e37 JA |
505 | * preference to them. IDLE queues goes to the back. |
506 | * after that, sort on the next service time. | |
0c534e0a JA |
507 | */ |
508 | if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq)) | |
67060e37 | 509 | n = &(*p)->rb_left; |
0c534e0a | 510 | else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq)) |
67060e37 JA |
511 | n = &(*p)->rb_right; |
512 | else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq)) | |
513 | n = &(*p)->rb_left; | |
514 | else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq)) | |
515 | n = &(*p)->rb_right; | |
0c534e0a | 516 | else if (rb_key < __cfqq->rb_key) |
67060e37 JA |
517 | n = &(*p)->rb_left; |
518 | else | |
519 | n = &(*p)->rb_right; | |
520 | ||
521 | if (n == &(*p)->rb_right) | |
cc09e299 | 522 | left = 0; |
67060e37 JA |
523 | |
524 | p = n; | |
d9e7620e JA |
525 | } |
526 | ||
cc09e299 JA |
527 | if (left) |
528 | cfqd->service_tree.left = &cfqq->rb_node; | |
529 | ||
d9e7620e JA |
530 | cfqq->rb_key = rb_key; |
531 | rb_link_node(&cfqq->rb_node, parent, p); | |
cc09e299 | 532 | rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb); |
1da177e4 LT |
533 | } |
534 | ||
498d3aa2 JA |
535 | /* |
536 | * Update cfqq's position in the service tree. | |
537 | */ | |
edd75ffd | 538 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 539 | { |
6d048f53 JA |
540 | /* |
541 | * Resorting requires the cfqq to be on the RR list already. | |
542 | */ | |
498d3aa2 | 543 | if (cfq_cfqq_on_rr(cfqq)) |
edd75ffd | 544 | cfq_service_tree_add(cfqd, cfqq, 0); |
6d048f53 JA |
545 | } |
546 | ||
1da177e4 LT |
547 | /* |
548 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 549 | * the pending list according to last request service |
1da177e4 | 550 | */ |
febffd61 | 551 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 552 | { |
7b679138 | 553 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
554 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
555 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
556 | cfqd->busy_queues++; |
557 | ||
edd75ffd | 558 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
559 | } |
560 | ||
498d3aa2 JA |
561 | /* |
562 | * Called when the cfqq no longer has requests pending, remove it from | |
563 | * the service tree. | |
564 | */ | |
febffd61 | 565 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 566 | { |
7b679138 | 567 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
568 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
569 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 570 | |
cc09e299 JA |
571 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) |
572 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); | |
d9e7620e | 573 | |
1da177e4 LT |
574 | BUG_ON(!cfqd->busy_queues); |
575 | cfqd->busy_queues--; | |
576 | } | |
577 | ||
578 | /* | |
579 | * rb tree support functions | |
580 | */ | |
febffd61 | 581 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 582 | { |
5e705374 | 583 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 584 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 585 | const int sync = rq_is_sync(rq); |
1da177e4 | 586 | |
b4878f24 JA |
587 | BUG_ON(!cfqq->queued[sync]); |
588 | cfqq->queued[sync]--; | |
1da177e4 | 589 | |
5e705374 | 590 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 591 | |
dd67d051 | 592 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 593 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
594 | } |
595 | ||
5e705374 | 596 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 597 | { |
5e705374 | 598 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 599 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 600 | struct request *__alias; |
1da177e4 | 601 | |
5380a101 | 602 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
603 | |
604 | /* | |
605 | * looks a little odd, but the first insert might return an alias. | |
606 | * if that happens, put the alias on the dispatch list | |
607 | */ | |
21183b07 | 608 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 609 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
610 | |
611 | if (!cfq_cfqq_on_rr(cfqq)) | |
612 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
613 | |
614 | /* | |
615 | * check if this request is a better next-serve candidate | |
616 | */ | |
617 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); | |
618 | BUG_ON(!cfqq->next_rq); | |
1da177e4 LT |
619 | } |
620 | ||
febffd61 | 621 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 622 | { |
5380a101 JA |
623 | elv_rb_del(&cfqq->sort_list, rq); |
624 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 625 | cfq_add_rq_rb(rq); |
1da177e4 LT |
626 | } |
627 | ||
206dc69b JA |
628 | static struct request * |
629 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 630 | { |
206dc69b | 631 | struct task_struct *tsk = current; |
91fac317 | 632 | struct cfq_io_context *cic; |
206dc69b | 633 | struct cfq_queue *cfqq; |
1da177e4 | 634 | |
4ac845a2 | 635 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
636 | if (!cic) |
637 | return NULL; | |
638 | ||
639 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
640 | if (cfqq) { |
641 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
642 | ||
21183b07 | 643 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 644 | } |
1da177e4 | 645 | |
1da177e4 LT |
646 | return NULL; |
647 | } | |
648 | ||
165125e1 | 649 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 650 | { |
22e2c507 | 651 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 652 | |
b4878f24 | 653 | cfqd->rq_in_driver++; |
7b679138 JA |
654 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
655 | cfqd->rq_in_driver); | |
25776e35 JA |
656 | |
657 | /* | |
658 | * If the depth is larger 1, it really could be queueing. But lets | |
659 | * make the mark a little higher - idling could still be good for | |
660 | * low queueing, and a low queueing number could also just indicate | |
661 | * a SCSI mid layer like behaviour where limit+1 is often seen. | |
662 | */ | |
663 | if (!cfqd->hw_tag && cfqd->rq_in_driver > 4) | |
664 | cfqd->hw_tag = 1; | |
6d048f53 JA |
665 | |
666 | cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors; | |
1da177e4 LT |
667 | } |
668 | ||
165125e1 | 669 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 670 | { |
b4878f24 JA |
671 | struct cfq_data *cfqd = q->elevator->elevator_data; |
672 | ||
673 | WARN_ON(!cfqd->rq_in_driver); | |
674 | cfqd->rq_in_driver--; | |
7b679138 JA |
675 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
676 | cfqd->rq_in_driver); | |
1da177e4 LT |
677 | } |
678 | ||
b4878f24 | 679 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 680 | { |
5e705374 | 681 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 682 | |
5e705374 JA |
683 | if (cfqq->next_rq == rq) |
684 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 685 | |
b4878f24 | 686 | list_del_init(&rq->queuelist); |
5e705374 | 687 | cfq_del_rq_rb(rq); |
374f84ac JA |
688 | |
689 | if (rq_is_meta(rq)) { | |
690 | WARN_ON(!cfqq->meta_pending); | |
691 | cfqq->meta_pending--; | |
692 | } | |
1da177e4 LT |
693 | } |
694 | ||
165125e1 JA |
695 | static int cfq_merge(struct request_queue *q, struct request **req, |
696 | struct bio *bio) | |
1da177e4 LT |
697 | { |
698 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
699 | struct request *__rq; | |
1da177e4 | 700 | |
206dc69b | 701 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 702 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
703 | *req = __rq; |
704 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
705 | } |
706 | ||
707 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
708 | } |
709 | ||
165125e1 | 710 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 711 | int type) |
1da177e4 | 712 | { |
21183b07 | 713 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 714 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 715 | |
5e705374 | 716 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 717 | } |
1da177e4 LT |
718 | } |
719 | ||
720 | static void | |
165125e1 | 721 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
722 | struct request *next) |
723 | { | |
22e2c507 JA |
724 | /* |
725 | * reposition in fifo if next is older than rq | |
726 | */ | |
727 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
728 | time_before(next->start_time, rq->start_time)) | |
729 | list_move(&rq->queuelist, &next->queuelist); | |
730 | ||
b4878f24 | 731 | cfq_remove_request(next); |
22e2c507 JA |
732 | } |
733 | ||
165125e1 | 734 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
735 | struct bio *bio) |
736 | { | |
737 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
91fac317 | 738 | struct cfq_io_context *cic; |
da775265 | 739 | struct cfq_queue *cfqq; |
da775265 JA |
740 | |
741 | /* | |
ec8acb69 | 742 | * Disallow merge of a sync bio into an async request. |
da775265 | 743 | */ |
91fac317 | 744 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
da775265 JA |
745 | return 0; |
746 | ||
747 | /* | |
719d3402 JA |
748 | * Lookup the cfqq that this bio will be queued with. Allow |
749 | * merge only if rq is queued there. | |
da775265 | 750 | */ |
4ac845a2 | 751 | cic = cfq_cic_lookup(cfqd, current->io_context); |
91fac317 VT |
752 | if (!cic) |
753 | return 0; | |
719d3402 | 754 | |
91fac317 | 755 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
719d3402 JA |
756 | if (cfqq == RQ_CFQQ(rq)) |
757 | return 1; | |
da775265 | 758 | |
ec8acb69 | 759 | return 0; |
da775265 JA |
760 | } |
761 | ||
febffd61 JA |
762 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
763 | struct cfq_queue *cfqq) | |
22e2c507 JA |
764 | { |
765 | if (cfqq) { | |
7b679138 | 766 | cfq_log_cfqq(cfqd, cfqq, "set_active"); |
22e2c507 | 767 | cfqq->slice_end = 0; |
3b18152c JA |
768 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
769 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 770 | cfq_mark_cfqq_slice_new(cfqq); |
1afba045 | 771 | cfq_clear_cfqq_queue_new(cfqq); |
22e2c507 JA |
772 | } |
773 | ||
774 | cfqd->active_queue = cfqq; | |
775 | } | |
776 | ||
7b14e3b5 JA |
777 | /* |
778 | * current cfqq expired its slice (or was too idle), select new one | |
779 | */ | |
780 | static void | |
781 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
6084cdda | 782 | int timed_out) |
7b14e3b5 | 783 | { |
7b679138 JA |
784 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
785 | ||
7b14e3b5 JA |
786 | if (cfq_cfqq_wait_request(cfqq)) |
787 | del_timer(&cfqd->idle_slice_timer); | |
788 | ||
7b14e3b5 JA |
789 | cfq_clear_cfqq_must_dispatch(cfqq); |
790 | cfq_clear_cfqq_wait_request(cfqq); | |
791 | ||
792 | /* | |
6084cdda | 793 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 794 | */ |
7b679138 | 795 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) { |
c5b680f3 | 796 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b679138 JA |
797 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
798 | } | |
7b14e3b5 | 799 | |
edd75ffd | 800 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
801 | |
802 | if (cfqq == cfqd->active_queue) | |
803 | cfqd->active_queue = NULL; | |
804 | ||
805 | if (cfqd->active_cic) { | |
806 | put_io_context(cfqd->active_cic->ioc); | |
807 | cfqd->active_cic = NULL; | |
808 | } | |
7b14e3b5 JA |
809 | } |
810 | ||
6084cdda | 811 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out) |
7b14e3b5 JA |
812 | { |
813 | struct cfq_queue *cfqq = cfqd->active_queue; | |
814 | ||
815 | if (cfqq) | |
6084cdda | 816 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
817 | } |
818 | ||
498d3aa2 JA |
819 | /* |
820 | * Get next queue for service. Unless we have a queue preemption, | |
821 | * we'll simply select the first cfqq in the service tree. | |
822 | */ | |
6d048f53 | 823 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 824 | { |
edd75ffd JA |
825 | if (RB_EMPTY_ROOT(&cfqd->service_tree.rb)) |
826 | return NULL; | |
d9e7620e | 827 | |
0871714e | 828 | return cfq_rb_first(&cfqd->service_tree); |
6d048f53 JA |
829 | } |
830 | ||
498d3aa2 JA |
831 | /* |
832 | * Get and set a new active queue for service. | |
833 | */ | |
6d048f53 JA |
834 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
835 | { | |
836 | struct cfq_queue *cfqq; | |
837 | ||
d9e7620e | 838 | cfqq = cfq_get_next_queue(cfqd); |
22e2c507 | 839 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 840 | return cfqq; |
22e2c507 JA |
841 | } |
842 | ||
d9e7620e JA |
843 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
844 | struct request *rq) | |
845 | { | |
846 | if (rq->sector >= cfqd->last_position) | |
847 | return rq->sector - cfqd->last_position; | |
848 | else | |
849 | return cfqd->last_position - rq->sector; | |
850 | } | |
851 | ||
6d048f53 JA |
852 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq) |
853 | { | |
854 | struct cfq_io_context *cic = cfqd->active_cic; | |
855 | ||
856 | if (!sample_valid(cic->seek_samples)) | |
857 | return 0; | |
858 | ||
859 | return cfq_dist_from_last(cfqd, rq) <= cic->seek_mean; | |
860 | } | |
861 | ||
d9e7620e JA |
862 | static int cfq_close_cooperator(struct cfq_data *cfq_data, |
863 | struct cfq_queue *cfqq) | |
6d048f53 | 864 | { |
6d048f53 | 865 | /* |
d9e7620e JA |
866 | * We should notice if some of the queues are cooperating, eg |
867 | * working closely on the same area of the disk. In that case, | |
868 | * we can group them together and don't waste time idling. | |
6d048f53 | 869 | */ |
d9e7620e | 870 | return 0; |
6d048f53 JA |
871 | } |
872 | ||
873 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (8 * 1024)) | |
caaa5f9f | 874 | |
6d048f53 | 875 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 876 | { |
1792669c | 877 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 878 | struct cfq_io_context *cic; |
7b14e3b5 JA |
879 | unsigned long sl; |
880 | ||
dd67d051 | 881 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 882 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
883 | |
884 | /* | |
885 | * idle is disabled, either manually or by past process history | |
886 | */ | |
6d048f53 JA |
887 | if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq)) |
888 | return; | |
889 | ||
7b679138 JA |
890 | /* |
891 | * still requests with the driver, don't idle | |
892 | */ | |
893 | if (cfqd->rq_in_driver) | |
894 | return; | |
895 | ||
22e2c507 JA |
896 | /* |
897 | * task has exited, don't wait | |
898 | */ | |
206dc69b | 899 | cic = cfqd->active_cic; |
66dac98e | 900 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
6d048f53 JA |
901 | return; |
902 | ||
903 | /* | |
904 | * See if this prio level has a good candidate | |
905 | */ | |
1afba045 JA |
906 | if (cfq_close_cooperator(cfqd, cfqq) && |
907 | (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2)) | |
6d048f53 | 908 | return; |
22e2c507 | 909 | |
3b18152c JA |
910 | cfq_mark_cfqq_must_dispatch(cfqq); |
911 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 912 | |
206dc69b JA |
913 | /* |
914 | * we don't want to idle for seeks, but we do want to allow | |
915 | * fair distribution of slice time for a process doing back-to-back | |
916 | * seeks. so allow a little bit of time for him to submit a new rq | |
917 | */ | |
6d048f53 | 918 | sl = cfqd->cfq_slice_idle; |
caaa5f9f | 919 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
d9e7620e | 920 | sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT)); |
206dc69b | 921 | |
7b14e3b5 | 922 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
7b679138 | 923 | cfq_log(cfqd, "arm_idle: %lu", sl); |
1da177e4 LT |
924 | } |
925 | ||
498d3aa2 JA |
926 | /* |
927 | * Move request from internal lists to the request queue dispatch list. | |
928 | */ | |
165125e1 | 929 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 930 | { |
3ed9a296 | 931 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 932 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 933 | |
7b679138 JA |
934 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
935 | ||
5380a101 | 936 | cfq_remove_request(rq); |
6d048f53 | 937 | cfqq->dispatched++; |
5380a101 | 938 | elv_dispatch_sort(q, rq); |
3ed9a296 JA |
939 | |
940 | if (cfq_cfqq_sync(cfqq)) | |
941 | cfqd->sync_flight++; | |
1da177e4 LT |
942 | } |
943 | ||
944 | /* | |
945 | * return expired entry, or NULL to just start from scratch in rbtree | |
946 | */ | |
febffd61 | 947 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
948 | { |
949 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 950 | struct request *rq; |
89850f7e | 951 | int fifo; |
1da177e4 | 952 | |
3b18152c | 953 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 954 | return NULL; |
cb887411 JA |
955 | |
956 | cfq_mark_cfqq_fifo_expire(cfqq); | |
957 | ||
89850f7e JA |
958 | if (list_empty(&cfqq->fifo)) |
959 | return NULL; | |
1da177e4 | 960 | |
6d048f53 | 961 | fifo = cfq_cfqq_sync(cfqq); |
89850f7e | 962 | rq = rq_entry_fifo(cfqq->fifo.next); |
1da177e4 | 963 | |
6d048f53 | 964 | if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) |
7b679138 | 965 | rq = NULL; |
1da177e4 | 966 | |
7b679138 | 967 | cfq_log_cfqq(cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 968 | return rq; |
1da177e4 LT |
969 | } |
970 | ||
22e2c507 JA |
971 | static inline int |
972 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
973 | { | |
974 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 975 | |
22e2c507 | 976 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 977 | |
22e2c507 | 978 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
979 | } |
980 | ||
22e2c507 | 981 | /* |
498d3aa2 JA |
982 | * Select a queue for service. If we have a current active queue, |
983 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 984 | */ |
1b5ed5e1 | 985 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 986 | { |
1da177e4 | 987 | struct cfq_queue *cfqq; |
1da177e4 | 988 | |
22e2c507 JA |
989 | cfqq = cfqd->active_queue; |
990 | if (!cfqq) | |
991 | goto new_queue; | |
1da177e4 | 992 | |
22e2c507 | 993 | /* |
6d048f53 | 994 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 995 | */ |
6d048f53 | 996 | if (cfq_slice_used(cfqq)) |
3b18152c | 997 | goto expire; |
1da177e4 | 998 | |
22e2c507 | 999 | /* |
6d048f53 JA |
1000 | * The active queue has requests and isn't expired, allow it to |
1001 | * dispatch. | |
22e2c507 | 1002 | */ |
dd67d051 | 1003 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 1004 | goto keep_queue; |
6d048f53 JA |
1005 | |
1006 | /* | |
1007 | * No requests pending. If the active queue still has requests in | |
1008 | * flight or is idling for a new request, allow either of these | |
1009 | * conditions to happen (or time out) before selecting a new queue. | |
1010 | */ | |
cc197479 JA |
1011 | if (timer_pending(&cfqd->idle_slice_timer) || |
1012 | (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) { | |
caaa5f9f JA |
1013 | cfqq = NULL; |
1014 | goto keep_queue; | |
22e2c507 JA |
1015 | } |
1016 | ||
3b18152c | 1017 | expire: |
6084cdda | 1018 | cfq_slice_expired(cfqd, 0); |
3b18152c JA |
1019 | new_queue: |
1020 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 1021 | keep_queue: |
3b18152c | 1022 | return cfqq; |
22e2c507 JA |
1023 | } |
1024 | ||
498d3aa2 JA |
1025 | /* |
1026 | * Dispatch some requests from cfqq, moving them to the request queue | |
1027 | * dispatch list. | |
1028 | */ | |
22e2c507 JA |
1029 | static int |
1030 | __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1031 | int max_dispatch) | |
1032 | { | |
1033 | int dispatched = 0; | |
1034 | ||
dd67d051 | 1035 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
1036 | |
1037 | do { | |
5e705374 | 1038 | struct request *rq; |
1da177e4 LT |
1039 | |
1040 | /* | |
22e2c507 | 1041 | * follow expired path, else get first next available |
1da177e4 | 1042 | */ |
fe094d98 JA |
1043 | rq = cfq_check_fifo(cfqq); |
1044 | if (rq == NULL) | |
5e705374 | 1045 | rq = cfqq->next_rq; |
22e2c507 JA |
1046 | |
1047 | /* | |
1048 | * finally, insert request into driver dispatch list | |
1049 | */ | |
5e705374 | 1050 | cfq_dispatch_insert(cfqd->queue, rq); |
1da177e4 | 1051 | |
22e2c507 | 1052 | dispatched++; |
1da177e4 | 1053 | |
22e2c507 | 1054 | if (!cfqd->active_cic) { |
5e705374 JA |
1055 | atomic_inc(&RQ_CIC(rq)->ioc->refcount); |
1056 | cfqd->active_cic = RQ_CIC(rq); | |
22e2c507 | 1057 | } |
1da177e4 | 1058 | |
dd67d051 | 1059 | if (RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 JA |
1060 | break; |
1061 | ||
1062 | } while (dispatched < max_dispatch); | |
1063 | ||
22e2c507 JA |
1064 | /* |
1065 | * expire an async queue immediately if it has used up its slice. idle | |
1066 | * queue always expire after 1 dispatch round. | |
1067 | */ | |
a9938006 | 1068 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
20e493a8 | 1069 | dispatched >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
a9938006 | 1070 | cfq_class_idle(cfqq))) { |
44f7c160 | 1071 | cfqq->slice_end = jiffies + 1; |
6084cdda | 1072 | cfq_slice_expired(cfqd, 0); |
44f7c160 | 1073 | } |
22e2c507 JA |
1074 | |
1075 | return dispatched; | |
1076 | } | |
1077 | ||
febffd61 | 1078 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
1079 | { |
1080 | int dispatched = 0; | |
1081 | ||
1082 | while (cfqq->next_rq) { | |
1083 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1084 | dispatched++; | |
1085 | } | |
1086 | ||
1087 | BUG_ON(!list_empty(&cfqq->fifo)); | |
1088 | return dispatched; | |
1089 | } | |
1090 | ||
498d3aa2 JA |
1091 | /* |
1092 | * Drain our current requests. Used for barriers and when switching | |
1093 | * io schedulers on-the-fly. | |
1094 | */ | |
d9e7620e | 1095 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 1096 | { |
0871714e | 1097 | struct cfq_queue *cfqq; |
d9e7620e | 1098 | int dispatched = 0; |
1b5ed5e1 | 1099 | |
0871714e | 1100 | while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL) |
d9e7620e | 1101 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
1b5ed5e1 | 1102 | |
6084cdda | 1103 | cfq_slice_expired(cfqd, 0); |
1b5ed5e1 TH |
1104 | |
1105 | BUG_ON(cfqd->busy_queues); | |
1106 | ||
7b679138 | 1107 | cfq_log(cfqd, "forced_dispatch=%d\n", dispatched); |
1b5ed5e1 TH |
1108 | return dispatched; |
1109 | } | |
1110 | ||
165125e1 | 1111 | static int cfq_dispatch_requests(struct request_queue *q, int force) |
22e2c507 JA |
1112 | { |
1113 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
6d048f53 | 1114 | struct cfq_queue *cfqq; |
caaa5f9f | 1115 | int dispatched; |
22e2c507 JA |
1116 | |
1117 | if (!cfqd->busy_queues) | |
1118 | return 0; | |
1119 | ||
1b5ed5e1 TH |
1120 | if (unlikely(force)) |
1121 | return cfq_forced_dispatch(cfqd); | |
1122 | ||
caaa5f9f | 1123 | dispatched = 0; |
caaa5f9f | 1124 | while ((cfqq = cfq_select_queue(cfqd)) != NULL) { |
b4878f24 JA |
1125 | int max_dispatch; |
1126 | ||
3ed9a296 JA |
1127 | max_dispatch = cfqd->cfq_quantum; |
1128 | if (cfq_class_idle(cfqq)) | |
1129 | max_dispatch = 1; | |
1130 | ||
1131 | if (cfqq->dispatched >= max_dispatch) { | |
1132 | if (cfqd->busy_queues > 1) | |
6d048f53 | 1133 | break; |
3ed9a296 | 1134 | if (cfqq->dispatched >= 4 * max_dispatch) |
a9938006 JA |
1135 | break; |
1136 | } | |
9ede209e | 1137 | |
3ed9a296 JA |
1138 | if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq)) |
1139 | break; | |
1140 | ||
3b18152c JA |
1141 | cfq_clear_cfqq_must_dispatch(cfqq); |
1142 | cfq_clear_cfqq_wait_request(cfqq); | |
22e2c507 JA |
1143 | del_timer(&cfqd->idle_slice_timer); |
1144 | ||
caaa5f9f | 1145 | dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); |
1da177e4 LT |
1146 | } |
1147 | ||
7b679138 | 1148 | cfq_log(cfqd, "dispatched=%d", dispatched); |
caaa5f9f | 1149 | return dispatched; |
1da177e4 LT |
1150 | } |
1151 | ||
1da177e4 | 1152 | /* |
5e705374 JA |
1153 | * task holds one reference to the queue, dropped when task exits. each rq |
1154 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1155 | * |
1156 | * queue lock must be held here. | |
1157 | */ | |
1158 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1159 | { | |
22e2c507 JA |
1160 | struct cfq_data *cfqd = cfqq->cfqd; |
1161 | ||
1162 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1163 | |
1164 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1165 | return; | |
1166 | ||
7b679138 | 1167 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 1168 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 1169 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1170 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1171 | |
28f95cbc | 1172 | if (unlikely(cfqd->active_queue == cfqq)) { |
6084cdda | 1173 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1174 | cfq_schedule_dispatch(cfqd); |
1175 | } | |
22e2c507 | 1176 | |
1da177e4 LT |
1177 | kmem_cache_free(cfq_pool, cfqq); |
1178 | } | |
1179 | ||
d6de8be7 JA |
1180 | /* |
1181 | * Must always be called with the rcu_read_lock() held | |
1182 | */ | |
07416d29 JA |
1183 | static void |
1184 | __call_for_each_cic(struct io_context *ioc, | |
1185 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1186 | { | |
1187 | struct cfq_io_context *cic; | |
1188 | struct hlist_node *n; | |
1189 | ||
1190 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) | |
1191 | func(ioc, cic); | |
1192 | } | |
1193 | ||
4ac845a2 | 1194 | /* |
34e6bbf2 | 1195 | * Call func for each cic attached to this ioc. |
4ac845a2 | 1196 | */ |
34e6bbf2 | 1197 | static void |
4ac845a2 JA |
1198 | call_for_each_cic(struct io_context *ioc, |
1199 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1da177e4 | 1200 | { |
4ac845a2 | 1201 | rcu_read_lock(); |
07416d29 | 1202 | __call_for_each_cic(ioc, func); |
4ac845a2 | 1203 | rcu_read_unlock(); |
34e6bbf2 FC |
1204 | } |
1205 | ||
1206 | static void cfq_cic_free_rcu(struct rcu_head *head) | |
1207 | { | |
1208 | struct cfq_io_context *cic; | |
1209 | ||
1210 | cic = container_of(head, struct cfq_io_context, rcu_head); | |
1211 | ||
1212 | kmem_cache_free(cfq_ioc_pool, cic); | |
1213 | elv_ioc_count_dec(ioc_count); | |
1214 | ||
9a11b4ed JA |
1215 | if (ioc_gone) { |
1216 | /* | |
1217 | * CFQ scheduler is exiting, grab exit lock and check | |
1218 | * the pending io context count. If it hits zero, | |
1219 | * complete ioc_gone and set it back to NULL | |
1220 | */ | |
1221 | spin_lock(&ioc_gone_lock); | |
1222 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) { | |
1223 | complete(ioc_gone); | |
1224 | ioc_gone = NULL; | |
1225 | } | |
1226 | spin_unlock(&ioc_gone_lock); | |
1227 | } | |
34e6bbf2 | 1228 | } |
4ac845a2 | 1229 | |
34e6bbf2 FC |
1230 | static void cfq_cic_free(struct cfq_io_context *cic) |
1231 | { | |
1232 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | |
4ac845a2 JA |
1233 | } |
1234 | ||
1235 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) | |
1236 | { | |
1237 | unsigned long flags; | |
1238 | ||
1239 | BUG_ON(!cic->dead_key); | |
1240 | ||
1241 | spin_lock_irqsave(&ioc->lock, flags); | |
1242 | radix_tree_delete(&ioc->radix_root, cic->dead_key); | |
ffc4e759 | 1243 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1244 | spin_unlock_irqrestore(&ioc->lock, flags); |
1245 | ||
34e6bbf2 | 1246 | cfq_cic_free(cic); |
4ac845a2 JA |
1247 | } |
1248 | ||
d6de8be7 JA |
1249 | /* |
1250 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. | |
1251 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), | |
1252 | * and ->trim() which is called with the task lock held | |
1253 | */ | |
4ac845a2 JA |
1254 | static void cfq_free_io_context(struct io_context *ioc) |
1255 | { | |
4ac845a2 | 1256 | /* |
34e6bbf2 FC |
1257 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
1258 | * so no more cic's are allowed to be linked into this ioc. So it | |
1259 | * should be ok to iterate over the known list, we will see all cic's | |
1260 | * since no new ones are added. | |
4ac845a2 | 1261 | */ |
07416d29 | 1262 | __call_for_each_cic(ioc, cic_free_func); |
1da177e4 LT |
1263 | } |
1264 | ||
89850f7e | 1265 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1266 | { |
28f95cbc | 1267 | if (unlikely(cfqq == cfqd->active_queue)) { |
6084cdda | 1268 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1269 | cfq_schedule_dispatch(cfqd); |
1270 | } | |
22e2c507 | 1271 | |
89850f7e JA |
1272 | cfq_put_queue(cfqq); |
1273 | } | |
22e2c507 | 1274 | |
89850f7e JA |
1275 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1276 | struct cfq_io_context *cic) | |
1277 | { | |
4faa3c81 FC |
1278 | struct io_context *ioc = cic->ioc; |
1279 | ||
fc46379d | 1280 | list_del_init(&cic->queue_list); |
4ac845a2 JA |
1281 | |
1282 | /* | |
1283 | * Make sure key == NULL is seen for dead queues | |
1284 | */ | |
fc46379d | 1285 | smp_wmb(); |
4ac845a2 | 1286 | cic->dead_key = (unsigned long) cic->key; |
fc46379d JA |
1287 | cic->key = NULL; |
1288 | ||
4faa3c81 FC |
1289 | if (ioc->ioc_data == cic) |
1290 | rcu_assign_pointer(ioc->ioc_data, NULL); | |
1291 | ||
12a05732 | 1292 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1293 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1294 | cic->cfqq[ASYNC] = NULL; |
1295 | } | |
1296 | ||
1297 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1298 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1299 | cic->cfqq[SYNC] = NULL; |
1300 | } | |
89850f7e JA |
1301 | } |
1302 | ||
4ac845a2 JA |
1303 | static void cfq_exit_single_io_context(struct io_context *ioc, |
1304 | struct cfq_io_context *cic) | |
89850f7e JA |
1305 | { |
1306 | struct cfq_data *cfqd = cic->key; | |
1307 | ||
89850f7e | 1308 | if (cfqd) { |
165125e1 | 1309 | struct request_queue *q = cfqd->queue; |
4ac845a2 | 1310 | unsigned long flags; |
89850f7e | 1311 | |
4ac845a2 | 1312 | spin_lock_irqsave(q->queue_lock, flags); |
89850f7e | 1313 | __cfq_exit_single_io_context(cfqd, cic); |
4ac845a2 | 1314 | spin_unlock_irqrestore(q->queue_lock, flags); |
89850f7e | 1315 | } |
1da177e4 LT |
1316 | } |
1317 | ||
498d3aa2 JA |
1318 | /* |
1319 | * The process that ioc belongs to has exited, we need to clean up | |
1320 | * and put the internal structures we have that belongs to that process. | |
1321 | */ | |
e2d74ac0 | 1322 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1323 | { |
4ac845a2 | 1324 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
1da177e4 LT |
1325 | } |
1326 | ||
22e2c507 | 1327 | static struct cfq_io_context * |
8267e268 | 1328 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1329 | { |
b5deef90 | 1330 | struct cfq_io_context *cic; |
1da177e4 | 1331 | |
94f6030c CL |
1332 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
1333 | cfqd->queue->node); | |
1da177e4 | 1334 | if (cic) { |
22e2c507 | 1335 | cic->last_end_request = jiffies; |
553698f9 | 1336 | INIT_LIST_HEAD(&cic->queue_list); |
ffc4e759 | 1337 | INIT_HLIST_NODE(&cic->cic_list); |
22e2c507 JA |
1338 | cic->dtor = cfq_free_io_context; |
1339 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1340 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1341 | } |
1342 | ||
1343 | return cic; | |
1344 | } | |
1345 | ||
fd0928df | 1346 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
1347 | { |
1348 | struct task_struct *tsk = current; | |
1349 | int ioprio_class; | |
1350 | ||
3b18152c | 1351 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1352 | return; |
1353 | ||
fd0928df | 1354 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 1355 | switch (ioprio_class) { |
fe094d98 JA |
1356 | default: |
1357 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1358 | case IOPRIO_CLASS_NONE: | |
1359 | /* | |
6d63c275 | 1360 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
1361 | */ |
1362 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 1363 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
1364 | break; |
1365 | case IOPRIO_CLASS_RT: | |
1366 | cfqq->ioprio = task_ioprio(ioc); | |
1367 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1368 | break; | |
1369 | case IOPRIO_CLASS_BE: | |
1370 | cfqq->ioprio = task_ioprio(ioc); | |
1371 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1372 | break; | |
1373 | case IOPRIO_CLASS_IDLE: | |
1374 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1375 | cfqq->ioprio = 7; | |
1376 | cfq_clear_cfqq_idle_window(cfqq); | |
1377 | break; | |
22e2c507 JA |
1378 | } |
1379 | ||
1380 | /* | |
1381 | * keep track of original prio settings in case we have to temporarily | |
1382 | * elevate the priority of this queue | |
1383 | */ | |
1384 | cfqq->org_ioprio = cfqq->ioprio; | |
1385 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 1386 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1387 | } |
1388 | ||
febffd61 | 1389 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
22e2c507 | 1390 | { |
478a82b0 AV |
1391 | struct cfq_data *cfqd = cic->key; |
1392 | struct cfq_queue *cfqq; | |
c1b707d2 | 1393 | unsigned long flags; |
35e6077c | 1394 | |
caaa5f9f JA |
1395 | if (unlikely(!cfqd)) |
1396 | return; | |
1397 | ||
c1b707d2 | 1398 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1399 | |
1400 | cfqq = cic->cfqq[ASYNC]; | |
1401 | if (cfqq) { | |
1402 | struct cfq_queue *new_cfqq; | |
fd0928df | 1403 | new_cfqq = cfq_get_queue(cfqd, ASYNC, cic->ioc, GFP_ATOMIC); |
caaa5f9f JA |
1404 | if (new_cfqq) { |
1405 | cic->cfqq[ASYNC] = new_cfqq; | |
1406 | cfq_put_queue(cfqq); | |
1407 | } | |
22e2c507 | 1408 | } |
caaa5f9f JA |
1409 | |
1410 | cfqq = cic->cfqq[SYNC]; | |
1411 | if (cfqq) | |
1412 | cfq_mark_cfqq_prio_changed(cfqq); | |
1413 | ||
c1b707d2 | 1414 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1415 | } |
1416 | ||
fc46379d | 1417 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1418 | { |
4ac845a2 | 1419 | call_for_each_cic(ioc, changed_ioprio); |
fc46379d | 1420 | ioc->ioprio_changed = 0; |
22e2c507 JA |
1421 | } |
1422 | ||
1423 | static struct cfq_queue * | |
15c31be4 | 1424 | cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync, |
fd0928df | 1425 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 1426 | { |
22e2c507 | 1427 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 1428 | struct cfq_io_context *cic; |
22e2c507 JA |
1429 | |
1430 | retry: | |
4ac845a2 | 1431 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
1432 | /* cic always exists here */ |
1433 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 JA |
1434 | |
1435 | if (!cfqq) { | |
1436 | if (new_cfqq) { | |
1437 | cfqq = new_cfqq; | |
1438 | new_cfqq = NULL; | |
1439 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1440 | /* |
1441 | * Inform the allocator of the fact that we will | |
1442 | * just repeat this allocation if it fails, to allow | |
1443 | * the allocator to do whatever it needs to attempt to | |
1444 | * free memory. | |
1445 | */ | |
22e2c507 | 1446 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c CL |
1447 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
1448 | gfp_mask | __GFP_NOFAIL | __GFP_ZERO, | |
1449 | cfqd->queue->node); | |
22e2c507 JA |
1450 | spin_lock_irq(cfqd->queue->queue_lock); |
1451 | goto retry; | |
1452 | } else { | |
94f6030c CL |
1453 | cfqq = kmem_cache_alloc_node(cfq_pool, |
1454 | gfp_mask | __GFP_ZERO, | |
1455 | cfqd->queue->node); | |
22e2c507 JA |
1456 | if (!cfqq) |
1457 | goto out; | |
1458 | } | |
1459 | ||
d9e7620e | 1460 | RB_CLEAR_NODE(&cfqq->rb_node); |
22e2c507 JA |
1461 | INIT_LIST_HEAD(&cfqq->fifo); |
1462 | ||
22e2c507 JA |
1463 | atomic_set(&cfqq->ref, 0); |
1464 | cfqq->cfqd = cfqd; | |
c5b680f3 | 1465 | |
3b18152c | 1466 | cfq_mark_cfqq_prio_changed(cfqq); |
53b03744 | 1467 | cfq_mark_cfqq_queue_new(cfqq); |
91fac317 | 1468 | |
fd0928df | 1469 | cfq_init_prio_data(cfqq, ioc); |
0871714e JA |
1470 | |
1471 | if (is_sync) { | |
1472 | if (!cfq_class_idle(cfqq)) | |
1473 | cfq_mark_cfqq_idle_window(cfqq); | |
1474 | cfq_mark_cfqq_sync(cfqq); | |
1475 | } | |
7b679138 JA |
1476 | cfqq->pid = current->pid; |
1477 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | |
22e2c507 JA |
1478 | } |
1479 | ||
1480 | if (new_cfqq) | |
1481 | kmem_cache_free(cfq_pool, new_cfqq); | |
1482 | ||
22e2c507 JA |
1483 | out: |
1484 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1485 | return cfqq; | |
1486 | } | |
1487 | ||
c2dea2d1 VT |
1488 | static struct cfq_queue ** |
1489 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
1490 | { | |
fe094d98 | 1491 | switch (ioprio_class) { |
c2dea2d1 VT |
1492 | case IOPRIO_CLASS_RT: |
1493 | return &cfqd->async_cfqq[0][ioprio]; | |
1494 | case IOPRIO_CLASS_BE: | |
1495 | return &cfqd->async_cfqq[1][ioprio]; | |
1496 | case IOPRIO_CLASS_IDLE: | |
1497 | return &cfqd->async_idle_cfqq; | |
1498 | default: | |
1499 | BUG(); | |
1500 | } | |
1501 | } | |
1502 | ||
15c31be4 | 1503 | static struct cfq_queue * |
fd0928df | 1504 | cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc, |
15c31be4 JA |
1505 | gfp_t gfp_mask) |
1506 | { | |
fd0928df JA |
1507 | const int ioprio = task_ioprio(ioc); |
1508 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 1509 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
1510 | struct cfq_queue *cfqq = NULL; |
1511 | ||
c2dea2d1 VT |
1512 | if (!is_sync) { |
1513 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
1514 | cfqq = *async_cfqq; | |
1515 | } | |
1516 | ||
0a0836a0 | 1517 | if (!cfqq) { |
fd0928df | 1518 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
0a0836a0 ON |
1519 | if (!cfqq) |
1520 | return NULL; | |
1521 | } | |
15c31be4 JA |
1522 | |
1523 | /* | |
1524 | * pin the queue now that it's allocated, scheduler exit will prune it | |
1525 | */ | |
c2dea2d1 | 1526 | if (!is_sync && !(*async_cfqq)) { |
15c31be4 | 1527 | atomic_inc(&cfqq->ref); |
c2dea2d1 | 1528 | *async_cfqq = cfqq; |
15c31be4 JA |
1529 | } |
1530 | ||
1531 | atomic_inc(&cfqq->ref); | |
1532 | return cfqq; | |
1533 | } | |
1534 | ||
498d3aa2 JA |
1535 | /* |
1536 | * We drop cfq io contexts lazily, so we may find a dead one. | |
1537 | */ | |
dbecf3ab | 1538 | static void |
4ac845a2 JA |
1539 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
1540 | struct cfq_io_context *cic) | |
dbecf3ab | 1541 | { |
4ac845a2 JA |
1542 | unsigned long flags; |
1543 | ||
fc46379d | 1544 | WARN_ON(!list_empty(&cic->queue_list)); |
597bc485 | 1545 | |
4ac845a2 JA |
1546 | spin_lock_irqsave(&ioc->lock, flags); |
1547 | ||
4faa3c81 | 1548 | BUG_ON(ioc->ioc_data == cic); |
597bc485 | 1549 | |
4ac845a2 | 1550 | radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd); |
ffc4e759 | 1551 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1552 | spin_unlock_irqrestore(&ioc->lock, flags); |
1553 | ||
1554 | cfq_cic_free(cic); | |
dbecf3ab OH |
1555 | } |
1556 | ||
e2d74ac0 | 1557 | static struct cfq_io_context * |
4ac845a2 | 1558 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
e2d74ac0 | 1559 | { |
e2d74ac0 | 1560 | struct cfq_io_context *cic; |
d6de8be7 | 1561 | unsigned long flags; |
4ac845a2 | 1562 | void *k; |
e2d74ac0 | 1563 | |
91fac317 VT |
1564 | if (unlikely(!ioc)) |
1565 | return NULL; | |
1566 | ||
d6de8be7 JA |
1567 | rcu_read_lock(); |
1568 | ||
597bc485 JA |
1569 | /* |
1570 | * we maintain a last-hit cache, to avoid browsing over the tree | |
1571 | */ | |
4ac845a2 | 1572 | cic = rcu_dereference(ioc->ioc_data); |
d6de8be7 JA |
1573 | if (cic && cic->key == cfqd) { |
1574 | rcu_read_unlock(); | |
597bc485 | 1575 | return cic; |
d6de8be7 | 1576 | } |
597bc485 | 1577 | |
4ac845a2 | 1578 | do { |
4ac845a2 JA |
1579 | cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd); |
1580 | rcu_read_unlock(); | |
1581 | if (!cic) | |
1582 | break; | |
be3b0753 OH |
1583 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1584 | k = cic->key; | |
1585 | if (unlikely(!k)) { | |
4ac845a2 | 1586 | cfq_drop_dead_cic(cfqd, ioc, cic); |
d6de8be7 | 1587 | rcu_read_lock(); |
4ac845a2 | 1588 | continue; |
dbecf3ab | 1589 | } |
e2d74ac0 | 1590 | |
d6de8be7 | 1591 | spin_lock_irqsave(&ioc->lock, flags); |
4ac845a2 | 1592 | rcu_assign_pointer(ioc->ioc_data, cic); |
d6de8be7 | 1593 | spin_unlock_irqrestore(&ioc->lock, flags); |
4ac845a2 JA |
1594 | break; |
1595 | } while (1); | |
e2d74ac0 | 1596 | |
4ac845a2 | 1597 | return cic; |
e2d74ac0 JA |
1598 | } |
1599 | ||
4ac845a2 JA |
1600 | /* |
1601 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup | |
1602 | * the process specific cfq io context when entered from the block layer. | |
1603 | * Also adds the cic to a per-cfqd list, used when this queue is removed. | |
1604 | */ | |
febffd61 JA |
1605 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
1606 | struct cfq_io_context *cic, gfp_t gfp_mask) | |
e2d74ac0 | 1607 | { |
0261d688 | 1608 | unsigned long flags; |
4ac845a2 | 1609 | int ret; |
e2d74ac0 | 1610 | |
4ac845a2 JA |
1611 | ret = radix_tree_preload(gfp_mask); |
1612 | if (!ret) { | |
1613 | cic->ioc = ioc; | |
1614 | cic->key = cfqd; | |
e2d74ac0 | 1615 | |
4ac845a2 JA |
1616 | spin_lock_irqsave(&ioc->lock, flags); |
1617 | ret = radix_tree_insert(&ioc->radix_root, | |
1618 | (unsigned long) cfqd, cic); | |
ffc4e759 JA |
1619 | if (!ret) |
1620 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); | |
4ac845a2 | 1621 | spin_unlock_irqrestore(&ioc->lock, flags); |
e2d74ac0 | 1622 | |
4ac845a2 JA |
1623 | radix_tree_preload_end(); |
1624 | ||
1625 | if (!ret) { | |
1626 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1627 | list_add(&cic->queue_list, &cfqd->cic_list); | |
1628 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1629 | } | |
e2d74ac0 JA |
1630 | } |
1631 | ||
4ac845a2 JA |
1632 | if (ret) |
1633 | printk(KERN_ERR "cfq: cic link failed!\n"); | |
fc46379d | 1634 | |
4ac845a2 | 1635 | return ret; |
e2d74ac0 JA |
1636 | } |
1637 | ||
1da177e4 LT |
1638 | /* |
1639 | * Setup general io context and cfq io context. There can be several cfq | |
1640 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1641 | * than one device managed by cfq. |
1da177e4 LT |
1642 | */ |
1643 | static struct cfq_io_context * | |
e2d74ac0 | 1644 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1645 | { |
22e2c507 | 1646 | struct io_context *ioc = NULL; |
1da177e4 | 1647 | struct cfq_io_context *cic; |
1da177e4 | 1648 | |
22e2c507 | 1649 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1650 | |
b5deef90 | 1651 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1652 | if (!ioc) |
1653 | return NULL; | |
1654 | ||
4ac845a2 | 1655 | cic = cfq_cic_lookup(cfqd, ioc); |
e2d74ac0 JA |
1656 | if (cic) |
1657 | goto out; | |
1da177e4 | 1658 | |
e2d74ac0 JA |
1659 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1660 | if (cic == NULL) | |
1661 | goto err; | |
1da177e4 | 1662 | |
4ac845a2 JA |
1663 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) |
1664 | goto err_free; | |
1665 | ||
1da177e4 | 1666 | out: |
fc46379d JA |
1667 | smp_read_barrier_depends(); |
1668 | if (unlikely(ioc->ioprio_changed)) | |
1669 | cfq_ioc_set_ioprio(ioc); | |
1670 | ||
1da177e4 | 1671 | return cic; |
4ac845a2 JA |
1672 | err_free: |
1673 | cfq_cic_free(cic); | |
1da177e4 LT |
1674 | err: |
1675 | put_io_context(ioc); | |
1676 | return NULL; | |
1677 | } | |
1678 | ||
22e2c507 JA |
1679 | static void |
1680 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1681 | { |
aaf1228d JA |
1682 | unsigned long elapsed = jiffies - cic->last_end_request; |
1683 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 1684 | |
22e2c507 JA |
1685 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1686 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1687 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1688 | } | |
1da177e4 | 1689 | |
206dc69b | 1690 | static void |
6d048f53 JA |
1691 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic, |
1692 | struct request *rq) | |
206dc69b JA |
1693 | { |
1694 | sector_t sdist; | |
1695 | u64 total; | |
1696 | ||
5e705374 JA |
1697 | if (cic->last_request_pos < rq->sector) |
1698 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1699 | else |
5e705374 | 1700 | sdist = cic->last_request_pos - rq->sector; |
206dc69b JA |
1701 | |
1702 | /* | |
1703 | * Don't allow the seek distance to get too large from the | |
1704 | * odd fragment, pagein, etc | |
1705 | */ | |
1706 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1707 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1708 | else | |
1709 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1710 | ||
1711 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1712 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1713 | total = cic->seek_total + (cic->seek_samples/2); | |
1714 | do_div(total, cic->seek_samples); | |
1715 | cic->seek_mean = (sector_t)total; | |
1716 | } | |
1da177e4 | 1717 | |
22e2c507 JA |
1718 | /* |
1719 | * Disable idle window if the process thinks too long or seeks so much that | |
1720 | * it doesn't matter | |
1721 | */ | |
1722 | static void | |
1723 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1724 | struct cfq_io_context *cic) | |
1725 | { | |
7b679138 | 1726 | int old_idle, enable_idle; |
1be92f2f | 1727 | |
0871714e JA |
1728 | /* |
1729 | * Don't idle for async or idle io prio class | |
1730 | */ | |
1731 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
1732 | return; |
1733 | ||
c265a7f4 | 1734 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 1735 | |
66dac98e | 1736 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
caaa5f9f | 1737 | (cfqd->hw_tag && CIC_SEEKY(cic))) |
22e2c507 JA |
1738 | enable_idle = 0; |
1739 | else if (sample_valid(cic->ttime_samples)) { | |
1740 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1741 | enable_idle = 0; | |
1742 | else | |
1743 | enable_idle = 1; | |
1da177e4 LT |
1744 | } |
1745 | ||
7b679138 JA |
1746 | if (old_idle != enable_idle) { |
1747 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
1748 | if (enable_idle) | |
1749 | cfq_mark_cfqq_idle_window(cfqq); | |
1750 | else | |
1751 | cfq_clear_cfqq_idle_window(cfqq); | |
1752 | } | |
22e2c507 | 1753 | } |
1da177e4 | 1754 | |
22e2c507 JA |
1755 | /* |
1756 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1757 | * no or if we aren't sure, a 1 will cause a preempt. | |
1758 | */ | |
1759 | static int | |
1760 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1761 | struct request *rq) |
22e2c507 | 1762 | { |
6d048f53 | 1763 | struct cfq_queue *cfqq; |
22e2c507 | 1764 | |
6d048f53 JA |
1765 | cfqq = cfqd->active_queue; |
1766 | if (!cfqq) | |
22e2c507 JA |
1767 | return 0; |
1768 | ||
6d048f53 JA |
1769 | if (cfq_slice_used(cfqq)) |
1770 | return 1; | |
1771 | ||
1772 | if (cfq_class_idle(new_cfqq)) | |
caaa5f9f | 1773 | return 0; |
22e2c507 JA |
1774 | |
1775 | if (cfq_class_idle(cfqq)) | |
1776 | return 1; | |
1e3335de | 1777 | |
374f84ac JA |
1778 | /* |
1779 | * if the new request is sync, but the currently running queue is | |
1780 | * not, let the sync request have priority. | |
1781 | */ | |
5e705374 | 1782 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1783 | return 1; |
1e3335de | 1784 | |
374f84ac JA |
1785 | /* |
1786 | * So both queues are sync. Let the new request get disk time if | |
1787 | * it's a metadata request and the current queue is doing regular IO. | |
1788 | */ | |
1789 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1790 | return 1; | |
22e2c507 | 1791 | |
1e3335de JA |
1792 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
1793 | return 0; | |
1794 | ||
1795 | /* | |
1796 | * if this request is as-good as one we would expect from the | |
1797 | * current cfqq, let it preempt | |
1798 | */ | |
6d048f53 | 1799 | if (cfq_rq_close(cfqd, rq)) |
1e3335de JA |
1800 | return 1; |
1801 | ||
22e2c507 JA |
1802 | return 0; |
1803 | } | |
1804 | ||
1805 | /* | |
1806 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1807 | * let it have half of its nominal slice. | |
1808 | */ | |
1809 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1810 | { | |
7b679138 | 1811 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
6084cdda | 1812 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 1813 | |
bf572256 JA |
1814 | /* |
1815 | * Put the new queue at the front of the of the current list, | |
1816 | * so we know that it will be selected next. | |
1817 | */ | |
1818 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
1819 | |
1820 | cfq_service_tree_add(cfqd, cfqq, 1); | |
bf572256 | 1821 | |
44f7c160 JA |
1822 | cfqq->slice_end = 0; |
1823 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
1824 | } |
1825 | ||
22e2c507 | 1826 | /* |
5e705374 | 1827 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1828 | * something we should do about it |
1829 | */ | |
1830 | static void | |
5e705374 JA |
1831 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1832 | struct request *rq) | |
22e2c507 | 1833 | { |
5e705374 | 1834 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1835 | |
374f84ac JA |
1836 | if (rq_is_meta(rq)) |
1837 | cfqq->meta_pending++; | |
1838 | ||
9c2c38a1 | 1839 | cfq_update_io_thinktime(cfqd, cic); |
6d048f53 | 1840 | cfq_update_io_seektime(cfqd, cic, rq); |
9c2c38a1 JA |
1841 | cfq_update_idle_window(cfqd, cfqq, cic); |
1842 | ||
5e705374 | 1843 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
22e2c507 JA |
1844 | |
1845 | if (cfqq == cfqd->active_queue) { | |
1846 | /* | |
1847 | * if we are waiting for a request for this queue, let it rip | |
1848 | * immediately and flag that we must not expire this queue | |
1849 | * just now | |
1850 | */ | |
3b18152c JA |
1851 | if (cfq_cfqq_wait_request(cfqq)) { |
1852 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 | 1853 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1854 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1855 | } |
5e705374 | 1856 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1857 | /* |
1858 | * not the active queue - expire current slice if it is | |
1859 | * idle and has expired it's mean thinktime or this new queue | |
1860 | * has some old slice time left and is of higher priority | |
1861 | */ | |
1862 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1863 | cfq_mark_cfqq_must_dispatch(cfqq); |
dc72ef4a | 1864 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1865 | } |
1da177e4 LT |
1866 | } |
1867 | ||
165125e1 | 1868 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1869 | { |
b4878f24 | 1870 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1871 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1872 | |
7b679138 | 1873 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
fd0928df | 1874 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
1da177e4 | 1875 | |
5e705374 | 1876 | cfq_add_rq_rb(rq); |
1da177e4 | 1877 | |
22e2c507 JA |
1878 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1879 | ||
5e705374 | 1880 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1881 | } |
1882 | ||
165125e1 | 1883 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1884 | { |
5e705374 | 1885 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1886 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1887 | const int sync = rq_is_sync(rq); |
b4878f24 | 1888 | unsigned long now; |
1da177e4 | 1889 | |
b4878f24 | 1890 | now = jiffies; |
7b679138 | 1891 | cfq_log_cfqq(cfqd, cfqq, "complete"); |
1da177e4 | 1892 | |
b4878f24 | 1893 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 1894 | WARN_ON(!cfqq->dispatched); |
b4878f24 | 1895 | cfqd->rq_in_driver--; |
6d048f53 | 1896 | cfqq->dispatched--; |
1da177e4 | 1897 | |
3ed9a296 JA |
1898 | if (cfq_cfqq_sync(cfqq)) |
1899 | cfqd->sync_flight--; | |
1900 | ||
b4878f24 JA |
1901 | if (!cfq_class_idle(cfqq)) |
1902 | cfqd->last_end_request = now; | |
3b18152c | 1903 | |
caaa5f9f | 1904 | if (sync) |
5e705374 | 1905 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1906 | |
1907 | /* | |
1908 | * If this is the active queue, check if it needs to be expired, | |
1909 | * or if we want to idle in case it has no pending requests. | |
1910 | */ | |
1911 | if (cfqd->active_queue == cfqq) { | |
44f7c160 JA |
1912 | if (cfq_cfqq_slice_new(cfqq)) { |
1913 | cfq_set_prio_slice(cfqd, cfqq); | |
1914 | cfq_clear_cfqq_slice_new(cfqq); | |
1915 | } | |
0871714e | 1916 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
6084cdda | 1917 | cfq_slice_expired(cfqd, 1); |
6d048f53 JA |
1918 | else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1919 | cfq_arm_slice_timer(cfqd); | |
caaa5f9f | 1920 | } |
6d048f53 JA |
1921 | |
1922 | if (!cfqd->rq_in_driver) | |
1923 | cfq_schedule_dispatch(cfqd); | |
1da177e4 LT |
1924 | } |
1925 | ||
22e2c507 JA |
1926 | /* |
1927 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
1928 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
1929 | */ | |
1930 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 1931 | { |
22e2c507 JA |
1932 | if (has_fs_excl()) { |
1933 | /* | |
1934 | * boost idle prio on transactions that would lock out other | |
1935 | * users of the filesystem | |
1936 | */ | |
1937 | if (cfq_class_idle(cfqq)) | |
1938 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1939 | if (cfqq->ioprio > IOPRIO_NORM) | |
1940 | cfqq->ioprio = IOPRIO_NORM; | |
1941 | } else { | |
1942 | /* | |
1943 | * check if we need to unboost the queue | |
1944 | */ | |
1945 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
1946 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
1947 | if (cfqq->ioprio != cfqq->org_ioprio) | |
1948 | cfqq->ioprio = cfqq->org_ioprio; | |
1949 | } | |
22e2c507 | 1950 | } |
1da177e4 | 1951 | |
89850f7e | 1952 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 1953 | { |
3b18152c | 1954 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 1955 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 1956 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 1957 | return ELV_MQUEUE_MUST; |
3b18152c | 1958 | } |
1da177e4 | 1959 | |
22e2c507 | 1960 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
1961 | } |
1962 | ||
165125e1 | 1963 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
1964 | { |
1965 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1966 | struct task_struct *tsk = current; | |
91fac317 | 1967 | struct cfq_io_context *cic; |
22e2c507 JA |
1968 | struct cfq_queue *cfqq; |
1969 | ||
1970 | /* | |
1971 | * don't force setup of a queue from here, as a call to may_queue | |
1972 | * does not necessarily imply that a request actually will be queued. | |
1973 | * so just lookup a possibly existing queue, or return 'may queue' | |
1974 | * if that fails | |
1975 | */ | |
4ac845a2 | 1976 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1977 | if (!cic) |
1978 | return ELV_MQUEUE_MAY; | |
1979 | ||
1980 | cfqq = cic_to_cfqq(cic, rw & REQ_RW_SYNC); | |
22e2c507 | 1981 | if (cfqq) { |
fd0928df | 1982 | cfq_init_prio_data(cfqq, cic->ioc); |
22e2c507 JA |
1983 | cfq_prio_boost(cfqq); |
1984 | ||
89850f7e | 1985 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
1986 | } |
1987 | ||
1988 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
1989 | } |
1990 | ||
1da177e4 LT |
1991 | /* |
1992 | * queue lock held here | |
1993 | */ | |
bb37b94c | 1994 | static void cfq_put_request(struct request *rq) |
1da177e4 | 1995 | { |
5e705374 | 1996 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1997 | |
5e705374 | 1998 | if (cfqq) { |
22e2c507 | 1999 | const int rw = rq_data_dir(rq); |
1da177e4 | 2000 | |
22e2c507 JA |
2001 | BUG_ON(!cfqq->allocated[rw]); |
2002 | cfqq->allocated[rw]--; | |
1da177e4 | 2003 | |
5e705374 | 2004 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 2005 | |
1da177e4 | 2006 | rq->elevator_private = NULL; |
5e705374 | 2007 | rq->elevator_private2 = NULL; |
1da177e4 | 2008 | |
1da177e4 LT |
2009 | cfq_put_queue(cfqq); |
2010 | } | |
2011 | } | |
2012 | ||
2013 | /* | |
22e2c507 | 2014 | * Allocate cfq data structures associated with this request. |
1da177e4 | 2015 | */ |
22e2c507 | 2016 | static int |
165125e1 | 2017 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
2018 | { |
2019 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2020 | struct cfq_io_context *cic; | |
2021 | const int rw = rq_data_dir(rq); | |
7749a8d4 | 2022 | const int is_sync = rq_is_sync(rq); |
22e2c507 | 2023 | struct cfq_queue *cfqq; |
1da177e4 LT |
2024 | unsigned long flags; |
2025 | ||
2026 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2027 | ||
e2d74ac0 | 2028 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 2029 | |
1da177e4 LT |
2030 | spin_lock_irqsave(q->queue_lock, flags); |
2031 | ||
22e2c507 JA |
2032 | if (!cic) |
2033 | goto queue_fail; | |
2034 | ||
91fac317 VT |
2035 | cfqq = cic_to_cfqq(cic, is_sync); |
2036 | if (!cfqq) { | |
fd0928df | 2037 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
91fac317 | 2038 | |
22e2c507 JA |
2039 | if (!cfqq) |
2040 | goto queue_fail; | |
1da177e4 | 2041 | |
91fac317 VT |
2042 | cic_set_cfqq(cic, cfqq, is_sync); |
2043 | } | |
1da177e4 LT |
2044 | |
2045 | cfqq->allocated[rw]++; | |
3b18152c | 2046 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 2047 | atomic_inc(&cfqq->ref); |
1da177e4 | 2048 | |
5e705374 | 2049 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 2050 | |
5e705374 JA |
2051 | rq->elevator_private = cic; |
2052 | rq->elevator_private2 = cfqq; | |
2053 | return 0; | |
1da177e4 | 2054 | |
22e2c507 JA |
2055 | queue_fail: |
2056 | if (cic) | |
2057 | put_io_context(cic->ioc); | |
89850f7e | 2058 | |
3b18152c | 2059 | cfq_schedule_dispatch(cfqd); |
1da177e4 | 2060 | spin_unlock_irqrestore(q->queue_lock, flags); |
7b679138 | 2061 | cfq_log(cfqd, "set_request fail"); |
1da177e4 LT |
2062 | return 1; |
2063 | } | |
2064 | ||
65f27f38 | 2065 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 2066 | { |
65f27f38 DH |
2067 | struct cfq_data *cfqd = |
2068 | container_of(work, struct cfq_data, unplug_work); | |
165125e1 | 2069 | struct request_queue *q = cfqd->queue; |
22e2c507 JA |
2070 | unsigned long flags; |
2071 | ||
2072 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 2073 | blk_start_queueing(q); |
22e2c507 JA |
2074 | spin_unlock_irqrestore(q->queue_lock, flags); |
2075 | } | |
2076 | ||
2077 | /* | |
2078 | * Timer running if the active_queue is currently idling inside its time slice | |
2079 | */ | |
2080 | static void cfq_idle_slice_timer(unsigned long data) | |
2081 | { | |
2082 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
2083 | struct cfq_queue *cfqq; | |
2084 | unsigned long flags; | |
3c6bd2f8 | 2085 | int timed_out = 1; |
22e2c507 | 2086 | |
7b679138 JA |
2087 | cfq_log(cfqd, "idle timer fired"); |
2088 | ||
22e2c507 JA |
2089 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
2090 | ||
fe094d98 JA |
2091 | cfqq = cfqd->active_queue; |
2092 | if (cfqq) { | |
3c6bd2f8 JA |
2093 | timed_out = 0; |
2094 | ||
22e2c507 JA |
2095 | /* |
2096 | * expired | |
2097 | */ | |
44f7c160 | 2098 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
2099 | goto expire; |
2100 | ||
2101 | /* | |
2102 | * only expire and reinvoke request handler, if there are | |
2103 | * other queues with pending requests | |
2104 | */ | |
caaa5f9f | 2105 | if (!cfqd->busy_queues) |
22e2c507 | 2106 | goto out_cont; |
22e2c507 JA |
2107 | |
2108 | /* | |
2109 | * not expired and it has a request pending, let it dispatch | |
2110 | */ | |
dd67d051 | 2111 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 2112 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
2113 | goto out_kick; |
2114 | } | |
2115 | } | |
2116 | expire: | |
6084cdda | 2117 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 2118 | out_kick: |
3b18152c | 2119 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
2120 | out_cont: |
2121 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
2122 | } | |
2123 | ||
3b18152c JA |
2124 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
2125 | { | |
2126 | del_timer_sync(&cfqd->idle_slice_timer); | |
4310864b | 2127 | kblockd_flush_work(&cfqd->unplug_work); |
3b18152c | 2128 | } |
22e2c507 | 2129 | |
c2dea2d1 VT |
2130 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
2131 | { | |
2132 | int i; | |
2133 | ||
2134 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
2135 | if (cfqd->async_cfqq[0][i]) | |
2136 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
2137 | if (cfqd->async_cfqq[1][i]) | |
2138 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 2139 | } |
2389d1ef ON |
2140 | |
2141 | if (cfqd->async_idle_cfqq) | |
2142 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
2143 | } |
2144 | ||
1da177e4 LT |
2145 | static void cfq_exit_queue(elevator_t *e) |
2146 | { | |
22e2c507 | 2147 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 2148 | struct request_queue *q = cfqd->queue; |
22e2c507 | 2149 | |
3b18152c | 2150 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 2151 | |
d9ff4187 | 2152 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 2153 | |
d9ff4187 | 2154 | if (cfqd->active_queue) |
6084cdda | 2155 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
2156 | |
2157 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
2158 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
2159 | struct cfq_io_context, | |
2160 | queue_list); | |
89850f7e JA |
2161 | |
2162 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 2163 | } |
e2d74ac0 | 2164 | |
c2dea2d1 | 2165 | cfq_put_async_queues(cfqd); |
15c31be4 | 2166 | |
d9ff4187 | 2167 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
2168 | |
2169 | cfq_shutdown_timer_wq(cfqd); | |
2170 | ||
a90d742e | 2171 | kfree(cfqd); |
1da177e4 LT |
2172 | } |
2173 | ||
165125e1 | 2174 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
2175 | { |
2176 | struct cfq_data *cfqd; | |
1da177e4 | 2177 | |
94f6030c | 2178 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
1da177e4 | 2179 | if (!cfqd) |
bc1c1169 | 2180 | return NULL; |
1da177e4 | 2181 | |
cc09e299 | 2182 | cfqd->service_tree = CFQ_RB_ROOT; |
d9ff4187 | 2183 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2184 | |
1da177e4 | 2185 | cfqd->queue = q; |
1da177e4 | 2186 | |
22e2c507 JA |
2187 | init_timer(&cfqd->idle_slice_timer); |
2188 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2189 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2190 | ||
65f27f38 | 2191 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2192 | |
b70c864d | 2193 | cfqd->last_end_request = jiffies; |
1da177e4 | 2194 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2195 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2196 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2197 | cfqd->cfq_back_max = cfq_back_max; |
2198 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2199 | cfqd->cfq_slice[0] = cfq_slice_async; |
2200 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2201 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2202 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
3b18152c | 2203 | |
bc1c1169 | 2204 | return cfqd; |
1da177e4 LT |
2205 | } |
2206 | ||
2207 | static void cfq_slab_kill(void) | |
2208 | { | |
d6de8be7 JA |
2209 | /* |
2210 | * Caller already ensured that pending RCU callbacks are completed, | |
2211 | * so we should have no busy allocations at this point. | |
2212 | */ | |
1da177e4 LT |
2213 | if (cfq_pool) |
2214 | kmem_cache_destroy(cfq_pool); | |
2215 | if (cfq_ioc_pool) | |
2216 | kmem_cache_destroy(cfq_ioc_pool); | |
2217 | } | |
2218 | ||
2219 | static int __init cfq_slab_setup(void) | |
2220 | { | |
0a31bd5f | 2221 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
1da177e4 LT |
2222 | if (!cfq_pool) |
2223 | goto fail; | |
2224 | ||
34e6bbf2 | 2225 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
1da177e4 LT |
2226 | if (!cfq_ioc_pool) |
2227 | goto fail; | |
2228 | ||
2229 | return 0; | |
2230 | fail: | |
2231 | cfq_slab_kill(); | |
2232 | return -ENOMEM; | |
2233 | } | |
2234 | ||
1da177e4 LT |
2235 | /* |
2236 | * sysfs parts below --> | |
2237 | */ | |
1da177e4 LT |
2238 | static ssize_t |
2239 | cfq_var_show(unsigned int var, char *page) | |
2240 | { | |
2241 | return sprintf(page, "%d\n", var); | |
2242 | } | |
2243 | ||
2244 | static ssize_t | |
2245 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2246 | { | |
2247 | char *p = (char *) page; | |
2248 | ||
2249 | *var = simple_strtoul(p, &p, 10); | |
2250 | return count; | |
2251 | } | |
2252 | ||
1da177e4 | 2253 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
3d1ab40f | 2254 | static ssize_t __FUNC(elevator_t *e, char *page) \ |
1da177e4 | 2255 | { \ |
3d1ab40f | 2256 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2257 | unsigned int __data = __VAR; \ |
2258 | if (__CONV) \ | |
2259 | __data = jiffies_to_msecs(__data); \ | |
2260 | return cfq_var_show(__data, (page)); \ | |
2261 | } | |
2262 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2263 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2264 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2265 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2266 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2267 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2268 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2269 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2270 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2271 | #undef SHOW_FUNCTION |
2272 | ||
2273 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
3d1ab40f | 2274 | static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \ |
1da177e4 | 2275 | { \ |
3d1ab40f | 2276 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2277 | unsigned int __data; \ |
2278 | int ret = cfq_var_store(&__data, (page), count); \ | |
2279 | if (__data < (MIN)) \ | |
2280 | __data = (MIN); \ | |
2281 | else if (__data > (MAX)) \ | |
2282 | __data = (MAX); \ | |
2283 | if (__CONV) \ | |
2284 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2285 | else \ | |
2286 | *(__PTR) = __data; \ | |
2287 | return ret; \ | |
2288 | } | |
2289 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
2290 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
2291 | UINT_MAX, 1); | |
2292 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
2293 | UINT_MAX, 1); | |
e572ec7e | 2294 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
2295 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
2296 | UINT_MAX, 0); | |
22e2c507 JA |
2297 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2298 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2299 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
2300 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
2301 | UINT_MAX, 0); | |
1da177e4 LT |
2302 | #undef STORE_FUNCTION |
2303 | ||
e572ec7e AV |
2304 | #define CFQ_ATTR(name) \ |
2305 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2306 | ||
2307 | static struct elv_fs_entry cfq_attrs[] = { | |
2308 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2309 | CFQ_ATTR(fifo_expire_sync), |
2310 | CFQ_ATTR(fifo_expire_async), | |
2311 | CFQ_ATTR(back_seek_max), | |
2312 | CFQ_ATTR(back_seek_penalty), | |
2313 | CFQ_ATTR(slice_sync), | |
2314 | CFQ_ATTR(slice_async), | |
2315 | CFQ_ATTR(slice_async_rq), | |
2316 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2317 | __ATTR_NULL |
1da177e4 LT |
2318 | }; |
2319 | ||
1da177e4 LT |
2320 | static struct elevator_type iosched_cfq = { |
2321 | .ops = { | |
2322 | .elevator_merge_fn = cfq_merge, | |
2323 | .elevator_merged_fn = cfq_merged_request, | |
2324 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2325 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2326 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2327 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2328 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2329 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2330 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2331 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2332 | .elevator_former_req_fn = elv_rb_former_request, |
2333 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2334 | .elevator_set_req_fn = cfq_set_request, |
2335 | .elevator_put_req_fn = cfq_put_request, | |
2336 | .elevator_may_queue_fn = cfq_may_queue, | |
2337 | .elevator_init_fn = cfq_init_queue, | |
2338 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2339 | .trim = cfq_free_io_context, |
1da177e4 | 2340 | }, |
3d1ab40f | 2341 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2342 | .elevator_name = "cfq", |
2343 | .elevator_owner = THIS_MODULE, | |
2344 | }; | |
2345 | ||
2346 | static int __init cfq_init(void) | |
2347 | { | |
22e2c507 JA |
2348 | /* |
2349 | * could be 0 on HZ < 1000 setups | |
2350 | */ | |
2351 | if (!cfq_slice_async) | |
2352 | cfq_slice_async = 1; | |
2353 | if (!cfq_slice_idle) | |
2354 | cfq_slice_idle = 1; | |
2355 | ||
1da177e4 LT |
2356 | if (cfq_slab_setup()) |
2357 | return -ENOMEM; | |
2358 | ||
2fdd82bd | 2359 | elv_register(&iosched_cfq); |
1da177e4 | 2360 | |
2fdd82bd | 2361 | return 0; |
1da177e4 LT |
2362 | } |
2363 | ||
2364 | static void __exit cfq_exit(void) | |
2365 | { | |
6e9a4738 | 2366 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2367 | elv_unregister(&iosched_cfq); |
334e94de | 2368 | ioc_gone = &all_gone; |
fba82272 OH |
2369 | /* ioc_gone's update must be visible before reading ioc_count */ |
2370 | smp_wmb(); | |
d6de8be7 JA |
2371 | |
2372 | /* | |
2373 | * this also protects us from entering cfq_slab_kill() with | |
2374 | * pending RCU callbacks | |
2375 | */ | |
4050cf16 | 2376 | if (elv_ioc_count_read(ioc_count)) |
9a11b4ed | 2377 | wait_for_completion(&all_gone); |
83521d3e | 2378 | cfq_slab_kill(); |
1da177e4 LT |
2379 | } |
2380 | ||
2381 | module_init(cfq_init); | |
2382 | module_exit(cfq_exit); | |
2383 | ||
2384 | MODULE_AUTHOR("Jens Axboe"); | |
2385 | MODULE_LICENSE("GPL"); | |
2386 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |