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 | 41 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 42 | #define CFQ_HW_QUEUE_MIN (5) |
22e2c507 | 43 | |
fe094d98 JA |
44 | #define RQ_CIC(rq) \ |
45 | ((struct cfq_io_context *) (rq)->elevator_private) | |
7b679138 | 46 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private2) |
1da177e4 | 47 | |
e18b890b CL |
48 | static struct kmem_cache *cfq_pool; |
49 | static struct kmem_cache *cfq_ioc_pool; | |
1da177e4 | 50 | |
4050cf16 | 51 | static DEFINE_PER_CPU(unsigned long, ioc_count); |
334e94de | 52 | static struct completion *ioc_gone; |
9a11b4ed | 53 | static DEFINE_SPINLOCK(ioc_gone_lock); |
334e94de | 54 | |
22e2c507 JA |
55 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
56 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
57 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
58 | ||
3b18152c JA |
59 | #define ASYNC (0) |
60 | #define SYNC (1) | |
61 | ||
206dc69b JA |
62 | #define sample_valid(samples) ((samples) > 80) |
63 | ||
cc09e299 JA |
64 | /* |
65 | * Most of our rbtree usage is for sorting with min extraction, so | |
66 | * if we cache the leftmost node we don't have to walk down the tree | |
67 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
68 | * move this into the elevator for the rq sorting as well. | |
69 | */ | |
70 | struct cfq_rb_root { | |
71 | struct rb_root rb; | |
72 | struct rb_node *left; | |
73 | }; | |
74 | #define CFQ_RB_ROOT (struct cfq_rb_root) { RB_ROOT, NULL, } | |
75 | ||
22e2c507 JA |
76 | /* |
77 | * Per block device queue structure | |
78 | */ | |
1da177e4 | 79 | struct cfq_data { |
165125e1 | 80 | struct request_queue *queue; |
22e2c507 JA |
81 | |
82 | /* | |
83 | * rr list of queues with requests and the count of them | |
84 | */ | |
cc09e299 | 85 | struct cfq_rb_root service_tree; |
22e2c507 JA |
86 | unsigned int busy_queues; |
87 | ||
22e2c507 | 88 | int rq_in_driver; |
3ed9a296 | 89 | int sync_flight; |
45333d5a AC |
90 | |
91 | /* | |
92 | * queue-depth detection | |
93 | */ | |
94 | int rq_queued; | |
25776e35 | 95 | int hw_tag; |
45333d5a AC |
96 | int hw_tag_samples; |
97 | int rq_in_driver_peak; | |
1da177e4 | 98 | |
22e2c507 JA |
99 | /* |
100 | * idle window management | |
101 | */ | |
102 | struct timer_list idle_slice_timer; | |
103 | struct work_struct unplug_work; | |
1da177e4 | 104 | |
22e2c507 JA |
105 | struct cfq_queue *active_queue; |
106 | struct cfq_io_context *active_cic; | |
22e2c507 | 107 | |
c2dea2d1 VT |
108 | /* |
109 | * async queue for each priority case | |
110 | */ | |
111 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
112 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 113 | |
6d048f53 | 114 | sector_t last_position; |
22e2c507 | 115 | unsigned long last_end_request; |
1da177e4 | 116 | |
1da177e4 LT |
117 | /* |
118 | * tunables, see top of file | |
119 | */ | |
120 | unsigned int cfq_quantum; | |
22e2c507 | 121 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
122 | unsigned int cfq_back_penalty; |
123 | unsigned int cfq_back_max; | |
22e2c507 JA |
124 | unsigned int cfq_slice[2]; |
125 | unsigned int cfq_slice_async_rq; | |
126 | unsigned int cfq_slice_idle; | |
d9ff4187 AV |
127 | |
128 | struct list_head cic_list; | |
1da177e4 LT |
129 | }; |
130 | ||
22e2c507 JA |
131 | /* |
132 | * Per process-grouping structure | |
133 | */ | |
1da177e4 LT |
134 | struct cfq_queue { |
135 | /* reference count */ | |
136 | atomic_t ref; | |
be754d2c RK |
137 | /* various state flags, see below */ |
138 | unsigned int flags; | |
1da177e4 LT |
139 | /* parent cfq_data */ |
140 | struct cfq_data *cfqd; | |
d9e7620e JA |
141 | /* service_tree member */ |
142 | struct rb_node rb_node; | |
143 | /* service_tree key */ | |
144 | unsigned long rb_key; | |
1da177e4 LT |
145 | /* sorted list of pending requests */ |
146 | struct rb_root sort_list; | |
147 | /* if fifo isn't expired, next request to serve */ | |
5e705374 | 148 | struct request *next_rq; |
1da177e4 LT |
149 | /* requests queued in sort_list */ |
150 | int queued[2]; | |
151 | /* currently allocated requests */ | |
152 | int allocated[2]; | |
153 | /* fifo list of requests in sort_list */ | |
22e2c507 | 154 | struct list_head fifo; |
1da177e4 | 155 | |
22e2c507 | 156 | unsigned long slice_end; |
c5b680f3 | 157 | long slice_resid; |
1da177e4 | 158 | |
be754d2c RK |
159 | /* pending metadata requests */ |
160 | int meta_pending; | |
6d048f53 JA |
161 | /* number of requests that are on the dispatch list or inside driver */ |
162 | int dispatched; | |
22e2c507 JA |
163 | |
164 | /* io prio of this group */ | |
165 | unsigned short ioprio, org_ioprio; | |
166 | unsigned short ioprio_class, org_ioprio_class; | |
167 | ||
7b679138 | 168 | pid_t pid; |
1da177e4 LT |
169 | }; |
170 | ||
3b18152c | 171 | enum cfqq_state_flags { |
b0b8d749 JA |
172 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
173 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
174 | CFQ_CFQQ_FLAG_must_alloc, /* must be allowed rq alloc */ | |
175 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ | |
176 | CFQ_CFQQ_FLAG_must_dispatch, /* must dispatch, even if expired */ | |
177 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ | |
178 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
179 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
180 | CFQ_CFQQ_FLAG_queue_new, /* queue never been serviced */ | |
44f7c160 | 181 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 182 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
3b18152c JA |
183 | }; |
184 | ||
185 | #define CFQ_CFQQ_FNS(name) \ | |
186 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
187 | { \ | |
fe094d98 | 188 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
189 | } \ |
190 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
191 | { \ | |
fe094d98 | 192 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
193 | } \ |
194 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
195 | { \ | |
fe094d98 | 196 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
197 | } |
198 | ||
199 | CFQ_CFQQ_FNS(on_rr); | |
200 | CFQ_CFQQ_FNS(wait_request); | |
201 | CFQ_CFQQ_FNS(must_alloc); | |
202 | CFQ_CFQQ_FNS(must_alloc_slice); | |
203 | CFQ_CFQQ_FNS(must_dispatch); | |
204 | CFQ_CFQQ_FNS(fifo_expire); | |
205 | CFQ_CFQQ_FNS(idle_window); | |
206 | CFQ_CFQQ_FNS(prio_changed); | |
53b03744 | 207 | CFQ_CFQQ_FNS(queue_new); |
44f7c160 | 208 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 209 | CFQ_CFQQ_FNS(sync); |
3b18152c JA |
210 | #undef CFQ_CFQQ_FNS |
211 | ||
7b679138 JA |
212 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
213 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
214 | #define cfq_log(cfqd, fmt, args...) \ | |
215 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
216 | ||
165125e1 | 217 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
91fac317 | 218 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, int, |
fd0928df | 219 | struct io_context *, gfp_t); |
4ac845a2 | 220 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
91fac317 VT |
221 | struct io_context *); |
222 | ||
223 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, | |
224 | int is_sync) | |
225 | { | |
226 | return cic->cfqq[!!is_sync]; | |
227 | } | |
228 | ||
229 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
230 | struct cfq_queue *cfqq, int is_sync) | |
231 | { | |
232 | cic->cfqq[!!is_sync] = cfqq; | |
233 | } | |
234 | ||
235 | /* | |
236 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
237 | * set (in which case it could also be direct WRITE). | |
238 | */ | |
239 | static inline int cfq_bio_sync(struct bio *bio) | |
240 | { | |
241 | if (bio_data_dir(bio) == READ || bio_sync(bio)) | |
242 | return 1; | |
243 | ||
244 | return 0; | |
245 | } | |
1da177e4 | 246 | |
99f95e52 AM |
247 | /* |
248 | * scheduler run of queue, if there are requests pending and no one in the | |
249 | * driver that will restart queueing | |
250 | */ | |
251 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
252 | { | |
7b679138 JA |
253 | if (cfqd->busy_queues) { |
254 | cfq_log(cfqd, "schedule dispatch"); | |
18887ad9 | 255 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 256 | } |
99f95e52 AM |
257 | } |
258 | ||
165125e1 | 259 | static int cfq_queue_empty(struct request_queue *q) |
99f95e52 AM |
260 | { |
261 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
262 | ||
b4878f24 | 263 | return !cfqd->busy_queues; |
99f95e52 AM |
264 | } |
265 | ||
44f7c160 JA |
266 | /* |
267 | * Scale schedule slice based on io priority. Use the sync time slice only | |
268 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
269 | * io only, should not get full sync slice length. | |
270 | */ | |
d9e7620e JA |
271 | static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync, |
272 | unsigned short prio) | |
44f7c160 | 273 | { |
d9e7620e | 274 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 275 | |
d9e7620e JA |
276 | WARN_ON(prio >= IOPRIO_BE_NR); |
277 | ||
278 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
279 | } | |
44f7c160 | 280 | |
d9e7620e JA |
281 | static inline int |
282 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
283 | { | |
284 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
285 | } |
286 | ||
287 | static inline void | |
288 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
289 | { | |
290 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; | |
7b679138 | 291 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
292 | } |
293 | ||
294 | /* | |
295 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
296 | * isn't valid until the first request from the dispatch is activated | |
297 | * and the slice time set. | |
298 | */ | |
299 | static inline int cfq_slice_used(struct cfq_queue *cfqq) | |
300 | { | |
301 | if (cfq_cfqq_slice_new(cfqq)) | |
302 | return 0; | |
303 | if (time_before(jiffies, cfqq->slice_end)) | |
304 | return 0; | |
305 | ||
306 | return 1; | |
307 | } | |
308 | ||
1da177e4 | 309 | /* |
5e705374 | 310 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 311 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 312 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 313 | */ |
5e705374 JA |
314 | static struct request * |
315 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
316 | { |
317 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 318 | unsigned long back_max; |
e8a99053 AM |
319 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
320 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
321 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 322 | |
5e705374 JA |
323 | if (rq1 == NULL || rq1 == rq2) |
324 | return rq2; | |
325 | if (rq2 == NULL) | |
326 | return rq1; | |
9c2c38a1 | 327 | |
5e705374 JA |
328 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
329 | return rq1; | |
330 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
331 | return rq2; | |
374f84ac JA |
332 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
333 | return rq1; | |
334 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
335 | return rq2; | |
1da177e4 | 336 | |
5e705374 JA |
337 | s1 = rq1->sector; |
338 | s2 = rq2->sector; | |
1da177e4 | 339 | |
6d048f53 | 340 | last = cfqd->last_position; |
1da177e4 | 341 | |
1da177e4 LT |
342 | /* |
343 | * by definition, 1KiB is 2 sectors | |
344 | */ | |
345 | back_max = cfqd->cfq_back_max * 2; | |
346 | ||
347 | /* | |
348 | * Strict one way elevator _except_ in the case where we allow | |
349 | * short backward seeks which are biased as twice the cost of a | |
350 | * similar forward seek. | |
351 | */ | |
352 | if (s1 >= last) | |
353 | d1 = s1 - last; | |
354 | else if (s1 + back_max >= last) | |
355 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
356 | else | |
e8a99053 | 357 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
358 | |
359 | if (s2 >= last) | |
360 | d2 = s2 - last; | |
361 | else if (s2 + back_max >= last) | |
362 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
363 | else | |
e8a99053 | 364 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
365 | |
366 | /* Found required data */ | |
e8a99053 AM |
367 | |
368 | /* | |
369 | * By doing switch() on the bit mask "wrap" we avoid having to | |
370 | * check two variables for all permutations: --> faster! | |
371 | */ | |
372 | switch (wrap) { | |
5e705374 | 373 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 374 | if (d1 < d2) |
5e705374 | 375 | return rq1; |
e8a99053 | 376 | else if (d2 < d1) |
5e705374 | 377 | return rq2; |
e8a99053 AM |
378 | else { |
379 | if (s1 >= s2) | |
5e705374 | 380 | return rq1; |
e8a99053 | 381 | else |
5e705374 | 382 | return rq2; |
e8a99053 | 383 | } |
1da177e4 | 384 | |
e8a99053 | 385 | case CFQ_RQ2_WRAP: |
5e705374 | 386 | return rq1; |
e8a99053 | 387 | case CFQ_RQ1_WRAP: |
5e705374 JA |
388 | return rq2; |
389 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
390 | default: |
391 | /* | |
392 | * Since both rqs are wrapped, | |
393 | * start with the one that's further behind head | |
394 | * (--> only *one* back seek required), | |
395 | * since back seek takes more time than forward. | |
396 | */ | |
397 | if (s1 <= s2) | |
5e705374 | 398 | return rq1; |
1da177e4 | 399 | else |
5e705374 | 400 | return rq2; |
1da177e4 LT |
401 | } |
402 | } | |
403 | ||
498d3aa2 JA |
404 | /* |
405 | * The below is leftmost cache rbtree addon | |
406 | */ | |
0871714e | 407 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 JA |
408 | { |
409 | if (!root->left) | |
410 | root->left = rb_first(&root->rb); | |
411 | ||
0871714e JA |
412 | if (root->left) |
413 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
414 | ||
415 | return NULL; | |
cc09e299 JA |
416 | } |
417 | ||
418 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) | |
419 | { | |
420 | if (root->left == n) | |
421 | root->left = NULL; | |
422 | ||
423 | rb_erase(n, &root->rb); | |
424 | RB_CLEAR_NODE(n); | |
425 | } | |
426 | ||
1da177e4 LT |
427 | /* |
428 | * would be nice to take fifo expire time into account as well | |
429 | */ | |
5e705374 JA |
430 | static struct request * |
431 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
432 | struct request *last) | |
1da177e4 | 433 | { |
21183b07 JA |
434 | struct rb_node *rbnext = rb_next(&last->rb_node); |
435 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 436 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 437 | |
21183b07 | 438 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
439 | |
440 | if (rbprev) | |
5e705374 | 441 | prev = rb_entry_rq(rbprev); |
1da177e4 | 442 | |
21183b07 | 443 | if (rbnext) |
5e705374 | 444 | next = rb_entry_rq(rbnext); |
21183b07 JA |
445 | else { |
446 | rbnext = rb_first(&cfqq->sort_list); | |
447 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 448 | next = rb_entry_rq(rbnext); |
21183b07 | 449 | } |
1da177e4 | 450 | |
21183b07 | 451 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
452 | } |
453 | ||
d9e7620e JA |
454 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
455 | struct cfq_queue *cfqq) | |
1da177e4 | 456 | { |
d9e7620e JA |
457 | /* |
458 | * just an approximation, should be ok. | |
459 | */ | |
67e6b49e JA |
460 | return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
461 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); | |
d9e7620e JA |
462 | } |
463 | ||
498d3aa2 JA |
464 | /* |
465 | * The cfqd->service_tree holds all pending cfq_queue's that have | |
466 | * requests waiting to be processed. It is sorted in the order that | |
467 | * we will service the queues. | |
468 | */ | |
d9e7620e | 469 | static void cfq_service_tree_add(struct cfq_data *cfqd, |
edd75ffd | 470 | struct cfq_queue *cfqq, int add_front) |
d9e7620e | 471 | { |
0871714e JA |
472 | struct rb_node **p, *parent; |
473 | struct cfq_queue *__cfqq; | |
d9e7620e | 474 | unsigned long rb_key; |
498d3aa2 | 475 | int left; |
d9e7620e | 476 | |
0871714e JA |
477 | if (cfq_class_idle(cfqq)) { |
478 | rb_key = CFQ_IDLE_DELAY; | |
479 | parent = rb_last(&cfqd->service_tree.rb); | |
480 | if (parent && parent != &cfqq->rb_node) { | |
481 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
482 | rb_key += __cfqq->rb_key; | |
483 | } else | |
484 | rb_key += jiffies; | |
485 | } else if (!add_front) { | |
edd75ffd JA |
486 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
487 | rb_key += cfqq->slice_resid; | |
488 | cfqq->slice_resid = 0; | |
489 | } else | |
490 | rb_key = 0; | |
1da177e4 | 491 | |
d9e7620e | 492 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
99f9628a | 493 | /* |
d9e7620e | 494 | * same position, nothing more to do |
99f9628a | 495 | */ |
d9e7620e JA |
496 | if (rb_key == cfqq->rb_key) |
497 | return; | |
1da177e4 | 498 | |
cc09e299 | 499 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); |
1da177e4 | 500 | } |
d9e7620e | 501 | |
498d3aa2 | 502 | left = 1; |
0871714e JA |
503 | parent = NULL; |
504 | p = &cfqd->service_tree.rb.rb_node; | |
d9e7620e | 505 | while (*p) { |
67060e37 | 506 | struct rb_node **n; |
cc09e299 | 507 | |
d9e7620e JA |
508 | parent = *p; |
509 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
510 | ||
0c534e0a JA |
511 | /* |
512 | * sort RT queues first, we always want to give | |
67060e37 JA |
513 | * preference to them. IDLE queues goes to the back. |
514 | * after that, sort on the next service time. | |
0c534e0a JA |
515 | */ |
516 | if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq)) | |
67060e37 | 517 | n = &(*p)->rb_left; |
0c534e0a | 518 | else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq)) |
67060e37 JA |
519 | n = &(*p)->rb_right; |
520 | else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq)) | |
521 | n = &(*p)->rb_left; | |
522 | else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq)) | |
523 | n = &(*p)->rb_right; | |
0c534e0a | 524 | else if (rb_key < __cfqq->rb_key) |
67060e37 JA |
525 | n = &(*p)->rb_left; |
526 | else | |
527 | n = &(*p)->rb_right; | |
528 | ||
529 | if (n == &(*p)->rb_right) | |
cc09e299 | 530 | left = 0; |
67060e37 JA |
531 | |
532 | p = n; | |
d9e7620e JA |
533 | } |
534 | ||
cc09e299 JA |
535 | if (left) |
536 | cfqd->service_tree.left = &cfqq->rb_node; | |
537 | ||
d9e7620e JA |
538 | cfqq->rb_key = rb_key; |
539 | rb_link_node(&cfqq->rb_node, parent, p); | |
cc09e299 | 540 | rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb); |
1da177e4 LT |
541 | } |
542 | ||
498d3aa2 JA |
543 | /* |
544 | * Update cfqq's position in the service tree. | |
545 | */ | |
edd75ffd | 546 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 547 | { |
6d048f53 JA |
548 | /* |
549 | * Resorting requires the cfqq to be on the RR list already. | |
550 | */ | |
498d3aa2 | 551 | if (cfq_cfqq_on_rr(cfqq)) |
edd75ffd | 552 | cfq_service_tree_add(cfqd, cfqq, 0); |
6d048f53 JA |
553 | } |
554 | ||
1da177e4 LT |
555 | /* |
556 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 557 | * the pending list according to last request service |
1da177e4 | 558 | */ |
febffd61 | 559 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 560 | { |
7b679138 | 561 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
562 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
563 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
564 | cfqd->busy_queues++; |
565 | ||
edd75ffd | 566 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
567 | } |
568 | ||
498d3aa2 JA |
569 | /* |
570 | * Called when the cfqq no longer has requests pending, remove it from | |
571 | * the service tree. | |
572 | */ | |
febffd61 | 573 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 574 | { |
7b679138 | 575 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
576 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
577 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 578 | |
cc09e299 JA |
579 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) |
580 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); | |
d9e7620e | 581 | |
1da177e4 LT |
582 | BUG_ON(!cfqd->busy_queues); |
583 | cfqd->busy_queues--; | |
584 | } | |
585 | ||
586 | /* | |
587 | * rb tree support functions | |
588 | */ | |
febffd61 | 589 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 590 | { |
5e705374 | 591 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 592 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 593 | const int sync = rq_is_sync(rq); |
1da177e4 | 594 | |
b4878f24 JA |
595 | BUG_ON(!cfqq->queued[sync]); |
596 | cfqq->queued[sync]--; | |
1da177e4 | 597 | |
5e705374 | 598 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 599 | |
dd67d051 | 600 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 601 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
602 | } |
603 | ||
5e705374 | 604 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 605 | { |
5e705374 | 606 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 607 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 608 | struct request *__alias; |
1da177e4 | 609 | |
5380a101 | 610 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
611 | |
612 | /* | |
613 | * looks a little odd, but the first insert might return an alias. | |
614 | * if that happens, put the alias on the dispatch list | |
615 | */ | |
21183b07 | 616 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 617 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
618 | |
619 | if (!cfq_cfqq_on_rr(cfqq)) | |
620 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
621 | |
622 | /* | |
623 | * check if this request is a better next-serve candidate | |
624 | */ | |
625 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); | |
626 | BUG_ON(!cfqq->next_rq); | |
1da177e4 LT |
627 | } |
628 | ||
febffd61 | 629 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 630 | { |
5380a101 JA |
631 | elv_rb_del(&cfqq->sort_list, rq); |
632 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 633 | cfq_add_rq_rb(rq); |
1da177e4 LT |
634 | } |
635 | ||
206dc69b JA |
636 | static struct request * |
637 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 638 | { |
206dc69b | 639 | struct task_struct *tsk = current; |
91fac317 | 640 | struct cfq_io_context *cic; |
206dc69b | 641 | struct cfq_queue *cfqq; |
1da177e4 | 642 | |
4ac845a2 | 643 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
644 | if (!cic) |
645 | return NULL; | |
646 | ||
647 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
648 | if (cfqq) { |
649 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
650 | ||
21183b07 | 651 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 652 | } |
1da177e4 | 653 | |
1da177e4 LT |
654 | return NULL; |
655 | } | |
656 | ||
165125e1 | 657 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 658 | { |
22e2c507 | 659 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 660 | |
b4878f24 | 661 | cfqd->rq_in_driver++; |
7b679138 JA |
662 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
663 | cfqd->rq_in_driver); | |
25776e35 | 664 | |
6d048f53 | 665 | cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors; |
1da177e4 LT |
666 | } |
667 | ||
165125e1 | 668 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 669 | { |
b4878f24 JA |
670 | struct cfq_data *cfqd = q->elevator->elevator_data; |
671 | ||
672 | WARN_ON(!cfqd->rq_in_driver); | |
673 | cfqd->rq_in_driver--; | |
7b679138 JA |
674 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
675 | cfqd->rq_in_driver); | |
1da177e4 LT |
676 | } |
677 | ||
b4878f24 | 678 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 679 | { |
5e705374 | 680 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 681 | |
5e705374 JA |
682 | if (cfqq->next_rq == rq) |
683 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 684 | |
b4878f24 | 685 | list_del_init(&rq->queuelist); |
5e705374 | 686 | cfq_del_rq_rb(rq); |
374f84ac | 687 | |
45333d5a | 688 | cfqq->cfqd->rq_queued--; |
374f84ac JA |
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 | ||
a68bbddb | 881 | /* |
f7d7b7a7 JA |
882 | * SSD device without seek penalty, disable idling. But only do so |
883 | * for devices that support queuing, otherwise we still have a problem | |
884 | * with sync vs async workloads. | |
a68bbddb | 885 | */ |
f7d7b7a7 | 886 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
887 | return; |
888 | ||
dd67d051 | 889 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 890 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
891 | |
892 | /* | |
893 | * idle is disabled, either manually or by past process history | |
894 | */ | |
6d048f53 JA |
895 | if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq)) |
896 | return; | |
897 | ||
7b679138 JA |
898 | /* |
899 | * still requests with the driver, don't idle | |
900 | */ | |
901 | if (cfqd->rq_in_driver) | |
902 | return; | |
903 | ||
22e2c507 JA |
904 | /* |
905 | * task has exited, don't wait | |
906 | */ | |
206dc69b | 907 | cic = cfqd->active_cic; |
66dac98e | 908 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
6d048f53 JA |
909 | return; |
910 | ||
911 | /* | |
912 | * See if this prio level has a good candidate | |
913 | */ | |
1afba045 JA |
914 | if (cfq_close_cooperator(cfqd, cfqq) && |
915 | (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2)) | |
6d048f53 | 916 | return; |
22e2c507 | 917 | |
3b18152c JA |
918 | cfq_mark_cfqq_must_dispatch(cfqq); |
919 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 920 | |
206dc69b JA |
921 | /* |
922 | * we don't want to idle for seeks, but we do want to allow | |
923 | * fair distribution of slice time for a process doing back-to-back | |
924 | * seeks. so allow a little bit of time for him to submit a new rq | |
925 | */ | |
6d048f53 | 926 | sl = cfqd->cfq_slice_idle; |
caaa5f9f | 927 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
d9e7620e | 928 | sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT)); |
206dc69b | 929 | |
7b14e3b5 | 930 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
7b679138 | 931 | cfq_log(cfqd, "arm_idle: %lu", sl); |
1da177e4 LT |
932 | } |
933 | ||
498d3aa2 JA |
934 | /* |
935 | * Move request from internal lists to the request queue dispatch list. | |
936 | */ | |
165125e1 | 937 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 938 | { |
3ed9a296 | 939 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 940 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 941 | |
7b679138 JA |
942 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
943 | ||
5380a101 | 944 | cfq_remove_request(rq); |
6d048f53 | 945 | cfqq->dispatched++; |
5380a101 | 946 | elv_dispatch_sort(q, rq); |
3ed9a296 JA |
947 | |
948 | if (cfq_cfqq_sync(cfqq)) | |
949 | cfqd->sync_flight++; | |
1da177e4 LT |
950 | } |
951 | ||
952 | /* | |
953 | * return expired entry, or NULL to just start from scratch in rbtree | |
954 | */ | |
febffd61 | 955 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
956 | { |
957 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 958 | struct request *rq; |
89850f7e | 959 | int fifo; |
1da177e4 | 960 | |
3b18152c | 961 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 962 | return NULL; |
cb887411 JA |
963 | |
964 | cfq_mark_cfqq_fifo_expire(cfqq); | |
965 | ||
89850f7e JA |
966 | if (list_empty(&cfqq->fifo)) |
967 | return NULL; | |
1da177e4 | 968 | |
6d048f53 | 969 | fifo = cfq_cfqq_sync(cfqq); |
89850f7e | 970 | rq = rq_entry_fifo(cfqq->fifo.next); |
1da177e4 | 971 | |
6d048f53 | 972 | if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) |
7b679138 | 973 | rq = NULL; |
1da177e4 | 974 | |
7b679138 | 975 | cfq_log_cfqq(cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 976 | return rq; |
1da177e4 LT |
977 | } |
978 | ||
22e2c507 JA |
979 | static inline int |
980 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
981 | { | |
982 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 983 | |
22e2c507 | 984 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 985 | |
22e2c507 | 986 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
987 | } |
988 | ||
22e2c507 | 989 | /* |
498d3aa2 JA |
990 | * Select a queue for service. If we have a current active queue, |
991 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 992 | */ |
1b5ed5e1 | 993 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 994 | { |
1da177e4 | 995 | struct cfq_queue *cfqq; |
1da177e4 | 996 | |
22e2c507 JA |
997 | cfqq = cfqd->active_queue; |
998 | if (!cfqq) | |
999 | goto new_queue; | |
1da177e4 | 1000 | |
22e2c507 | 1001 | /* |
6d048f53 | 1002 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 1003 | */ |
6d048f53 | 1004 | if (cfq_slice_used(cfqq)) |
3b18152c | 1005 | goto expire; |
1da177e4 | 1006 | |
22e2c507 | 1007 | /* |
6d048f53 JA |
1008 | * The active queue has requests and isn't expired, allow it to |
1009 | * dispatch. | |
22e2c507 | 1010 | */ |
dd67d051 | 1011 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 1012 | goto keep_queue; |
6d048f53 JA |
1013 | |
1014 | /* | |
1015 | * No requests pending. If the active queue still has requests in | |
1016 | * flight or is idling for a new request, allow either of these | |
1017 | * conditions to happen (or time out) before selecting a new queue. | |
1018 | */ | |
cc197479 JA |
1019 | if (timer_pending(&cfqd->idle_slice_timer) || |
1020 | (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) { | |
caaa5f9f JA |
1021 | cfqq = NULL; |
1022 | goto keep_queue; | |
22e2c507 JA |
1023 | } |
1024 | ||
3b18152c | 1025 | expire: |
6084cdda | 1026 | cfq_slice_expired(cfqd, 0); |
3b18152c JA |
1027 | new_queue: |
1028 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 1029 | keep_queue: |
3b18152c | 1030 | return cfqq; |
22e2c507 JA |
1031 | } |
1032 | ||
498d3aa2 JA |
1033 | /* |
1034 | * Dispatch some requests from cfqq, moving them to the request queue | |
1035 | * dispatch list. | |
1036 | */ | |
22e2c507 JA |
1037 | static int |
1038 | __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1039 | int max_dispatch) | |
1040 | { | |
1041 | int dispatched = 0; | |
1042 | ||
dd67d051 | 1043 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
1044 | |
1045 | do { | |
5e705374 | 1046 | struct request *rq; |
1da177e4 LT |
1047 | |
1048 | /* | |
22e2c507 | 1049 | * follow expired path, else get first next available |
1da177e4 | 1050 | */ |
fe094d98 JA |
1051 | rq = cfq_check_fifo(cfqq); |
1052 | if (rq == NULL) | |
5e705374 | 1053 | rq = cfqq->next_rq; |
22e2c507 JA |
1054 | |
1055 | /* | |
1056 | * finally, insert request into driver dispatch list | |
1057 | */ | |
5e705374 | 1058 | cfq_dispatch_insert(cfqd->queue, rq); |
1da177e4 | 1059 | |
22e2c507 | 1060 | dispatched++; |
1da177e4 | 1061 | |
22e2c507 | 1062 | if (!cfqd->active_cic) { |
5e705374 JA |
1063 | atomic_inc(&RQ_CIC(rq)->ioc->refcount); |
1064 | cfqd->active_cic = RQ_CIC(rq); | |
22e2c507 | 1065 | } |
1da177e4 | 1066 | |
dd67d051 | 1067 | if (RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 JA |
1068 | break; |
1069 | ||
1070 | } while (dispatched < max_dispatch); | |
1071 | ||
22e2c507 JA |
1072 | /* |
1073 | * expire an async queue immediately if it has used up its slice. idle | |
1074 | * queue always expire after 1 dispatch round. | |
1075 | */ | |
a9938006 | 1076 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
20e493a8 | 1077 | dispatched >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
a9938006 | 1078 | cfq_class_idle(cfqq))) { |
44f7c160 | 1079 | cfqq->slice_end = jiffies + 1; |
6084cdda | 1080 | cfq_slice_expired(cfqd, 0); |
44f7c160 | 1081 | } |
22e2c507 JA |
1082 | |
1083 | return dispatched; | |
1084 | } | |
1085 | ||
febffd61 | 1086 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
1087 | { |
1088 | int dispatched = 0; | |
1089 | ||
1090 | while (cfqq->next_rq) { | |
1091 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1092 | dispatched++; | |
1093 | } | |
1094 | ||
1095 | BUG_ON(!list_empty(&cfqq->fifo)); | |
1096 | return dispatched; | |
1097 | } | |
1098 | ||
498d3aa2 JA |
1099 | /* |
1100 | * Drain our current requests. Used for barriers and when switching | |
1101 | * io schedulers on-the-fly. | |
1102 | */ | |
d9e7620e | 1103 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 1104 | { |
0871714e | 1105 | struct cfq_queue *cfqq; |
d9e7620e | 1106 | int dispatched = 0; |
1b5ed5e1 | 1107 | |
0871714e | 1108 | while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL) |
d9e7620e | 1109 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
1b5ed5e1 | 1110 | |
6084cdda | 1111 | cfq_slice_expired(cfqd, 0); |
1b5ed5e1 TH |
1112 | |
1113 | BUG_ON(cfqd->busy_queues); | |
1114 | ||
7b679138 | 1115 | cfq_log(cfqd, "forced_dispatch=%d\n", dispatched); |
1b5ed5e1 TH |
1116 | return dispatched; |
1117 | } | |
1118 | ||
165125e1 | 1119 | static int cfq_dispatch_requests(struct request_queue *q, int force) |
22e2c507 JA |
1120 | { |
1121 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
6d048f53 | 1122 | struct cfq_queue *cfqq; |
caaa5f9f | 1123 | int dispatched; |
22e2c507 JA |
1124 | |
1125 | if (!cfqd->busy_queues) | |
1126 | return 0; | |
1127 | ||
1b5ed5e1 TH |
1128 | if (unlikely(force)) |
1129 | return cfq_forced_dispatch(cfqd); | |
1130 | ||
caaa5f9f | 1131 | dispatched = 0; |
caaa5f9f | 1132 | while ((cfqq = cfq_select_queue(cfqd)) != NULL) { |
b4878f24 JA |
1133 | int max_dispatch; |
1134 | ||
3ed9a296 JA |
1135 | max_dispatch = cfqd->cfq_quantum; |
1136 | if (cfq_class_idle(cfqq)) | |
1137 | max_dispatch = 1; | |
1138 | ||
1139 | if (cfqq->dispatched >= max_dispatch) { | |
1140 | if (cfqd->busy_queues > 1) | |
6d048f53 | 1141 | break; |
3ed9a296 | 1142 | if (cfqq->dispatched >= 4 * max_dispatch) |
a9938006 JA |
1143 | break; |
1144 | } | |
9ede209e | 1145 | |
3ed9a296 JA |
1146 | if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq)) |
1147 | break; | |
1148 | ||
3b18152c JA |
1149 | cfq_clear_cfqq_must_dispatch(cfqq); |
1150 | cfq_clear_cfqq_wait_request(cfqq); | |
22e2c507 JA |
1151 | del_timer(&cfqd->idle_slice_timer); |
1152 | ||
caaa5f9f | 1153 | dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); |
1da177e4 LT |
1154 | } |
1155 | ||
7b679138 | 1156 | cfq_log(cfqd, "dispatched=%d", dispatched); |
caaa5f9f | 1157 | return dispatched; |
1da177e4 LT |
1158 | } |
1159 | ||
1da177e4 | 1160 | /* |
5e705374 JA |
1161 | * task holds one reference to the queue, dropped when task exits. each rq |
1162 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1163 | * |
1164 | * queue lock must be held here. | |
1165 | */ | |
1166 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1167 | { | |
22e2c507 JA |
1168 | struct cfq_data *cfqd = cfqq->cfqd; |
1169 | ||
1170 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1171 | |
1172 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1173 | return; | |
1174 | ||
7b679138 | 1175 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 1176 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 1177 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1178 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1179 | |
28f95cbc | 1180 | if (unlikely(cfqd->active_queue == cfqq)) { |
6084cdda | 1181 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1182 | cfq_schedule_dispatch(cfqd); |
1183 | } | |
22e2c507 | 1184 | |
1da177e4 LT |
1185 | kmem_cache_free(cfq_pool, cfqq); |
1186 | } | |
1187 | ||
d6de8be7 JA |
1188 | /* |
1189 | * Must always be called with the rcu_read_lock() held | |
1190 | */ | |
07416d29 JA |
1191 | static void |
1192 | __call_for_each_cic(struct io_context *ioc, | |
1193 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1194 | { | |
1195 | struct cfq_io_context *cic; | |
1196 | struct hlist_node *n; | |
1197 | ||
1198 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) | |
1199 | func(ioc, cic); | |
1200 | } | |
1201 | ||
4ac845a2 | 1202 | /* |
34e6bbf2 | 1203 | * Call func for each cic attached to this ioc. |
4ac845a2 | 1204 | */ |
34e6bbf2 | 1205 | static void |
4ac845a2 JA |
1206 | call_for_each_cic(struct io_context *ioc, |
1207 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1da177e4 | 1208 | { |
4ac845a2 | 1209 | rcu_read_lock(); |
07416d29 | 1210 | __call_for_each_cic(ioc, func); |
4ac845a2 | 1211 | rcu_read_unlock(); |
34e6bbf2 FC |
1212 | } |
1213 | ||
1214 | static void cfq_cic_free_rcu(struct rcu_head *head) | |
1215 | { | |
1216 | struct cfq_io_context *cic; | |
1217 | ||
1218 | cic = container_of(head, struct cfq_io_context, rcu_head); | |
1219 | ||
1220 | kmem_cache_free(cfq_ioc_pool, cic); | |
1221 | elv_ioc_count_dec(ioc_count); | |
1222 | ||
9a11b4ed JA |
1223 | if (ioc_gone) { |
1224 | /* | |
1225 | * CFQ scheduler is exiting, grab exit lock and check | |
1226 | * the pending io context count. If it hits zero, | |
1227 | * complete ioc_gone and set it back to NULL | |
1228 | */ | |
1229 | spin_lock(&ioc_gone_lock); | |
1230 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) { | |
1231 | complete(ioc_gone); | |
1232 | ioc_gone = NULL; | |
1233 | } | |
1234 | spin_unlock(&ioc_gone_lock); | |
1235 | } | |
34e6bbf2 | 1236 | } |
4ac845a2 | 1237 | |
34e6bbf2 FC |
1238 | static void cfq_cic_free(struct cfq_io_context *cic) |
1239 | { | |
1240 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | |
4ac845a2 JA |
1241 | } |
1242 | ||
1243 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) | |
1244 | { | |
1245 | unsigned long flags; | |
1246 | ||
1247 | BUG_ON(!cic->dead_key); | |
1248 | ||
1249 | spin_lock_irqsave(&ioc->lock, flags); | |
1250 | radix_tree_delete(&ioc->radix_root, cic->dead_key); | |
ffc4e759 | 1251 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1252 | spin_unlock_irqrestore(&ioc->lock, flags); |
1253 | ||
34e6bbf2 | 1254 | cfq_cic_free(cic); |
4ac845a2 JA |
1255 | } |
1256 | ||
d6de8be7 JA |
1257 | /* |
1258 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. | |
1259 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), | |
1260 | * and ->trim() which is called with the task lock held | |
1261 | */ | |
4ac845a2 JA |
1262 | static void cfq_free_io_context(struct io_context *ioc) |
1263 | { | |
4ac845a2 | 1264 | /* |
34e6bbf2 FC |
1265 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
1266 | * so no more cic's are allowed to be linked into this ioc. So it | |
1267 | * should be ok to iterate over the known list, we will see all cic's | |
1268 | * since no new ones are added. | |
4ac845a2 | 1269 | */ |
07416d29 | 1270 | __call_for_each_cic(ioc, cic_free_func); |
1da177e4 LT |
1271 | } |
1272 | ||
89850f7e | 1273 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1274 | { |
28f95cbc | 1275 | if (unlikely(cfqq == cfqd->active_queue)) { |
6084cdda | 1276 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1277 | cfq_schedule_dispatch(cfqd); |
1278 | } | |
22e2c507 | 1279 | |
89850f7e JA |
1280 | cfq_put_queue(cfqq); |
1281 | } | |
22e2c507 | 1282 | |
89850f7e JA |
1283 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1284 | struct cfq_io_context *cic) | |
1285 | { | |
4faa3c81 FC |
1286 | struct io_context *ioc = cic->ioc; |
1287 | ||
fc46379d | 1288 | list_del_init(&cic->queue_list); |
4ac845a2 JA |
1289 | |
1290 | /* | |
1291 | * Make sure key == NULL is seen for dead queues | |
1292 | */ | |
fc46379d | 1293 | smp_wmb(); |
4ac845a2 | 1294 | cic->dead_key = (unsigned long) cic->key; |
fc46379d JA |
1295 | cic->key = NULL; |
1296 | ||
4faa3c81 FC |
1297 | if (ioc->ioc_data == cic) |
1298 | rcu_assign_pointer(ioc->ioc_data, NULL); | |
1299 | ||
12a05732 | 1300 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1301 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1302 | cic->cfqq[ASYNC] = NULL; |
1303 | } | |
1304 | ||
1305 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1306 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1307 | cic->cfqq[SYNC] = NULL; |
1308 | } | |
89850f7e JA |
1309 | } |
1310 | ||
4ac845a2 JA |
1311 | static void cfq_exit_single_io_context(struct io_context *ioc, |
1312 | struct cfq_io_context *cic) | |
89850f7e JA |
1313 | { |
1314 | struct cfq_data *cfqd = cic->key; | |
1315 | ||
89850f7e | 1316 | if (cfqd) { |
165125e1 | 1317 | struct request_queue *q = cfqd->queue; |
4ac845a2 | 1318 | unsigned long flags; |
89850f7e | 1319 | |
4ac845a2 | 1320 | spin_lock_irqsave(q->queue_lock, flags); |
89850f7e | 1321 | __cfq_exit_single_io_context(cfqd, cic); |
4ac845a2 | 1322 | spin_unlock_irqrestore(q->queue_lock, flags); |
89850f7e | 1323 | } |
1da177e4 LT |
1324 | } |
1325 | ||
498d3aa2 JA |
1326 | /* |
1327 | * The process that ioc belongs to has exited, we need to clean up | |
1328 | * and put the internal structures we have that belongs to that process. | |
1329 | */ | |
e2d74ac0 | 1330 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1331 | { |
4ac845a2 | 1332 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
1da177e4 LT |
1333 | } |
1334 | ||
22e2c507 | 1335 | static struct cfq_io_context * |
8267e268 | 1336 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1337 | { |
b5deef90 | 1338 | struct cfq_io_context *cic; |
1da177e4 | 1339 | |
94f6030c CL |
1340 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
1341 | cfqd->queue->node); | |
1da177e4 | 1342 | if (cic) { |
22e2c507 | 1343 | cic->last_end_request = jiffies; |
553698f9 | 1344 | INIT_LIST_HEAD(&cic->queue_list); |
ffc4e759 | 1345 | INIT_HLIST_NODE(&cic->cic_list); |
22e2c507 JA |
1346 | cic->dtor = cfq_free_io_context; |
1347 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1348 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1349 | } |
1350 | ||
1351 | return cic; | |
1352 | } | |
1353 | ||
fd0928df | 1354 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
1355 | { |
1356 | struct task_struct *tsk = current; | |
1357 | int ioprio_class; | |
1358 | ||
3b18152c | 1359 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1360 | return; |
1361 | ||
fd0928df | 1362 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 1363 | switch (ioprio_class) { |
fe094d98 JA |
1364 | default: |
1365 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1366 | case IOPRIO_CLASS_NONE: | |
1367 | /* | |
6d63c275 | 1368 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
1369 | */ |
1370 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 1371 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
1372 | break; |
1373 | case IOPRIO_CLASS_RT: | |
1374 | cfqq->ioprio = task_ioprio(ioc); | |
1375 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1376 | break; | |
1377 | case IOPRIO_CLASS_BE: | |
1378 | cfqq->ioprio = task_ioprio(ioc); | |
1379 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1380 | break; | |
1381 | case IOPRIO_CLASS_IDLE: | |
1382 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1383 | cfqq->ioprio = 7; | |
1384 | cfq_clear_cfqq_idle_window(cfqq); | |
1385 | break; | |
22e2c507 JA |
1386 | } |
1387 | ||
1388 | /* | |
1389 | * keep track of original prio settings in case we have to temporarily | |
1390 | * elevate the priority of this queue | |
1391 | */ | |
1392 | cfqq->org_ioprio = cfqq->ioprio; | |
1393 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 1394 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1395 | } |
1396 | ||
febffd61 | 1397 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
22e2c507 | 1398 | { |
478a82b0 AV |
1399 | struct cfq_data *cfqd = cic->key; |
1400 | struct cfq_queue *cfqq; | |
c1b707d2 | 1401 | unsigned long flags; |
35e6077c | 1402 | |
caaa5f9f JA |
1403 | if (unlikely(!cfqd)) |
1404 | return; | |
1405 | ||
c1b707d2 | 1406 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1407 | |
1408 | cfqq = cic->cfqq[ASYNC]; | |
1409 | if (cfqq) { | |
1410 | struct cfq_queue *new_cfqq; | |
fd0928df | 1411 | new_cfqq = cfq_get_queue(cfqd, ASYNC, cic->ioc, GFP_ATOMIC); |
caaa5f9f JA |
1412 | if (new_cfqq) { |
1413 | cic->cfqq[ASYNC] = new_cfqq; | |
1414 | cfq_put_queue(cfqq); | |
1415 | } | |
22e2c507 | 1416 | } |
caaa5f9f JA |
1417 | |
1418 | cfqq = cic->cfqq[SYNC]; | |
1419 | if (cfqq) | |
1420 | cfq_mark_cfqq_prio_changed(cfqq); | |
1421 | ||
c1b707d2 | 1422 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1423 | } |
1424 | ||
fc46379d | 1425 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1426 | { |
4ac845a2 | 1427 | call_for_each_cic(ioc, changed_ioprio); |
fc46379d | 1428 | ioc->ioprio_changed = 0; |
22e2c507 JA |
1429 | } |
1430 | ||
1431 | static struct cfq_queue * | |
15c31be4 | 1432 | cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync, |
fd0928df | 1433 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 1434 | { |
22e2c507 | 1435 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 1436 | struct cfq_io_context *cic; |
22e2c507 JA |
1437 | |
1438 | retry: | |
4ac845a2 | 1439 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
1440 | /* cic always exists here */ |
1441 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 JA |
1442 | |
1443 | if (!cfqq) { | |
1444 | if (new_cfqq) { | |
1445 | cfqq = new_cfqq; | |
1446 | new_cfqq = NULL; | |
1447 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1448 | /* |
1449 | * Inform the allocator of the fact that we will | |
1450 | * just repeat this allocation if it fails, to allow | |
1451 | * the allocator to do whatever it needs to attempt to | |
1452 | * free memory. | |
1453 | */ | |
22e2c507 | 1454 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c CL |
1455 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
1456 | gfp_mask | __GFP_NOFAIL | __GFP_ZERO, | |
1457 | cfqd->queue->node); | |
22e2c507 JA |
1458 | spin_lock_irq(cfqd->queue->queue_lock); |
1459 | goto retry; | |
1460 | } else { | |
94f6030c CL |
1461 | cfqq = kmem_cache_alloc_node(cfq_pool, |
1462 | gfp_mask | __GFP_ZERO, | |
1463 | cfqd->queue->node); | |
22e2c507 JA |
1464 | if (!cfqq) |
1465 | goto out; | |
1466 | } | |
1467 | ||
d9e7620e | 1468 | RB_CLEAR_NODE(&cfqq->rb_node); |
22e2c507 JA |
1469 | INIT_LIST_HEAD(&cfqq->fifo); |
1470 | ||
22e2c507 JA |
1471 | atomic_set(&cfqq->ref, 0); |
1472 | cfqq->cfqd = cfqd; | |
c5b680f3 | 1473 | |
3b18152c | 1474 | cfq_mark_cfqq_prio_changed(cfqq); |
53b03744 | 1475 | cfq_mark_cfqq_queue_new(cfqq); |
91fac317 | 1476 | |
fd0928df | 1477 | cfq_init_prio_data(cfqq, ioc); |
0871714e JA |
1478 | |
1479 | if (is_sync) { | |
1480 | if (!cfq_class_idle(cfqq)) | |
1481 | cfq_mark_cfqq_idle_window(cfqq); | |
1482 | cfq_mark_cfqq_sync(cfqq); | |
1483 | } | |
7b679138 JA |
1484 | cfqq->pid = current->pid; |
1485 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | |
22e2c507 JA |
1486 | } |
1487 | ||
1488 | if (new_cfqq) | |
1489 | kmem_cache_free(cfq_pool, new_cfqq); | |
1490 | ||
22e2c507 JA |
1491 | out: |
1492 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1493 | return cfqq; | |
1494 | } | |
1495 | ||
c2dea2d1 VT |
1496 | static struct cfq_queue ** |
1497 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
1498 | { | |
fe094d98 | 1499 | switch (ioprio_class) { |
c2dea2d1 VT |
1500 | case IOPRIO_CLASS_RT: |
1501 | return &cfqd->async_cfqq[0][ioprio]; | |
1502 | case IOPRIO_CLASS_BE: | |
1503 | return &cfqd->async_cfqq[1][ioprio]; | |
1504 | case IOPRIO_CLASS_IDLE: | |
1505 | return &cfqd->async_idle_cfqq; | |
1506 | default: | |
1507 | BUG(); | |
1508 | } | |
1509 | } | |
1510 | ||
15c31be4 | 1511 | static struct cfq_queue * |
fd0928df | 1512 | cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc, |
15c31be4 JA |
1513 | gfp_t gfp_mask) |
1514 | { | |
fd0928df JA |
1515 | const int ioprio = task_ioprio(ioc); |
1516 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 1517 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
1518 | struct cfq_queue *cfqq = NULL; |
1519 | ||
c2dea2d1 VT |
1520 | if (!is_sync) { |
1521 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
1522 | cfqq = *async_cfqq; | |
1523 | } | |
1524 | ||
0a0836a0 | 1525 | if (!cfqq) { |
fd0928df | 1526 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
0a0836a0 ON |
1527 | if (!cfqq) |
1528 | return NULL; | |
1529 | } | |
15c31be4 JA |
1530 | |
1531 | /* | |
1532 | * pin the queue now that it's allocated, scheduler exit will prune it | |
1533 | */ | |
c2dea2d1 | 1534 | if (!is_sync && !(*async_cfqq)) { |
15c31be4 | 1535 | atomic_inc(&cfqq->ref); |
c2dea2d1 | 1536 | *async_cfqq = cfqq; |
15c31be4 JA |
1537 | } |
1538 | ||
1539 | atomic_inc(&cfqq->ref); | |
1540 | return cfqq; | |
1541 | } | |
1542 | ||
498d3aa2 JA |
1543 | /* |
1544 | * We drop cfq io contexts lazily, so we may find a dead one. | |
1545 | */ | |
dbecf3ab | 1546 | static void |
4ac845a2 JA |
1547 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
1548 | struct cfq_io_context *cic) | |
dbecf3ab | 1549 | { |
4ac845a2 JA |
1550 | unsigned long flags; |
1551 | ||
fc46379d | 1552 | WARN_ON(!list_empty(&cic->queue_list)); |
597bc485 | 1553 | |
4ac845a2 JA |
1554 | spin_lock_irqsave(&ioc->lock, flags); |
1555 | ||
4faa3c81 | 1556 | BUG_ON(ioc->ioc_data == cic); |
597bc485 | 1557 | |
4ac845a2 | 1558 | radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd); |
ffc4e759 | 1559 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1560 | spin_unlock_irqrestore(&ioc->lock, flags); |
1561 | ||
1562 | cfq_cic_free(cic); | |
dbecf3ab OH |
1563 | } |
1564 | ||
e2d74ac0 | 1565 | static struct cfq_io_context * |
4ac845a2 | 1566 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
e2d74ac0 | 1567 | { |
e2d74ac0 | 1568 | struct cfq_io_context *cic; |
d6de8be7 | 1569 | unsigned long flags; |
4ac845a2 | 1570 | void *k; |
e2d74ac0 | 1571 | |
91fac317 VT |
1572 | if (unlikely(!ioc)) |
1573 | return NULL; | |
1574 | ||
d6de8be7 JA |
1575 | rcu_read_lock(); |
1576 | ||
597bc485 JA |
1577 | /* |
1578 | * we maintain a last-hit cache, to avoid browsing over the tree | |
1579 | */ | |
4ac845a2 | 1580 | cic = rcu_dereference(ioc->ioc_data); |
d6de8be7 JA |
1581 | if (cic && cic->key == cfqd) { |
1582 | rcu_read_unlock(); | |
597bc485 | 1583 | return cic; |
d6de8be7 | 1584 | } |
597bc485 | 1585 | |
4ac845a2 | 1586 | do { |
4ac845a2 JA |
1587 | cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd); |
1588 | rcu_read_unlock(); | |
1589 | if (!cic) | |
1590 | break; | |
be3b0753 OH |
1591 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1592 | k = cic->key; | |
1593 | if (unlikely(!k)) { | |
4ac845a2 | 1594 | cfq_drop_dead_cic(cfqd, ioc, cic); |
d6de8be7 | 1595 | rcu_read_lock(); |
4ac845a2 | 1596 | continue; |
dbecf3ab | 1597 | } |
e2d74ac0 | 1598 | |
d6de8be7 | 1599 | spin_lock_irqsave(&ioc->lock, flags); |
4ac845a2 | 1600 | rcu_assign_pointer(ioc->ioc_data, cic); |
d6de8be7 | 1601 | spin_unlock_irqrestore(&ioc->lock, flags); |
4ac845a2 JA |
1602 | break; |
1603 | } while (1); | |
e2d74ac0 | 1604 | |
4ac845a2 | 1605 | return cic; |
e2d74ac0 JA |
1606 | } |
1607 | ||
4ac845a2 JA |
1608 | /* |
1609 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup | |
1610 | * the process specific cfq io context when entered from the block layer. | |
1611 | * Also adds the cic to a per-cfqd list, used when this queue is removed. | |
1612 | */ | |
febffd61 JA |
1613 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
1614 | struct cfq_io_context *cic, gfp_t gfp_mask) | |
e2d74ac0 | 1615 | { |
0261d688 | 1616 | unsigned long flags; |
4ac845a2 | 1617 | int ret; |
e2d74ac0 | 1618 | |
4ac845a2 JA |
1619 | ret = radix_tree_preload(gfp_mask); |
1620 | if (!ret) { | |
1621 | cic->ioc = ioc; | |
1622 | cic->key = cfqd; | |
e2d74ac0 | 1623 | |
4ac845a2 JA |
1624 | spin_lock_irqsave(&ioc->lock, flags); |
1625 | ret = radix_tree_insert(&ioc->radix_root, | |
1626 | (unsigned long) cfqd, cic); | |
ffc4e759 JA |
1627 | if (!ret) |
1628 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); | |
4ac845a2 | 1629 | spin_unlock_irqrestore(&ioc->lock, flags); |
e2d74ac0 | 1630 | |
4ac845a2 JA |
1631 | radix_tree_preload_end(); |
1632 | ||
1633 | if (!ret) { | |
1634 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1635 | list_add(&cic->queue_list, &cfqd->cic_list); | |
1636 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1637 | } | |
e2d74ac0 JA |
1638 | } |
1639 | ||
4ac845a2 JA |
1640 | if (ret) |
1641 | printk(KERN_ERR "cfq: cic link failed!\n"); | |
fc46379d | 1642 | |
4ac845a2 | 1643 | return ret; |
e2d74ac0 JA |
1644 | } |
1645 | ||
1da177e4 LT |
1646 | /* |
1647 | * Setup general io context and cfq io context. There can be several cfq | |
1648 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1649 | * than one device managed by cfq. |
1da177e4 LT |
1650 | */ |
1651 | static struct cfq_io_context * | |
e2d74ac0 | 1652 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1653 | { |
22e2c507 | 1654 | struct io_context *ioc = NULL; |
1da177e4 | 1655 | struct cfq_io_context *cic; |
1da177e4 | 1656 | |
22e2c507 | 1657 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1658 | |
b5deef90 | 1659 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1660 | if (!ioc) |
1661 | return NULL; | |
1662 | ||
4ac845a2 | 1663 | cic = cfq_cic_lookup(cfqd, ioc); |
e2d74ac0 JA |
1664 | if (cic) |
1665 | goto out; | |
1da177e4 | 1666 | |
e2d74ac0 JA |
1667 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1668 | if (cic == NULL) | |
1669 | goto err; | |
1da177e4 | 1670 | |
4ac845a2 JA |
1671 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) |
1672 | goto err_free; | |
1673 | ||
1da177e4 | 1674 | out: |
fc46379d JA |
1675 | smp_read_barrier_depends(); |
1676 | if (unlikely(ioc->ioprio_changed)) | |
1677 | cfq_ioc_set_ioprio(ioc); | |
1678 | ||
1da177e4 | 1679 | return cic; |
4ac845a2 JA |
1680 | err_free: |
1681 | cfq_cic_free(cic); | |
1da177e4 LT |
1682 | err: |
1683 | put_io_context(ioc); | |
1684 | return NULL; | |
1685 | } | |
1686 | ||
22e2c507 JA |
1687 | static void |
1688 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1689 | { |
aaf1228d JA |
1690 | unsigned long elapsed = jiffies - cic->last_end_request; |
1691 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 1692 | |
22e2c507 JA |
1693 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1694 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1695 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1696 | } | |
1da177e4 | 1697 | |
206dc69b | 1698 | static void |
6d048f53 JA |
1699 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic, |
1700 | struct request *rq) | |
206dc69b JA |
1701 | { |
1702 | sector_t sdist; | |
1703 | u64 total; | |
1704 | ||
5e705374 JA |
1705 | if (cic->last_request_pos < rq->sector) |
1706 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1707 | else |
5e705374 | 1708 | sdist = cic->last_request_pos - rq->sector; |
206dc69b JA |
1709 | |
1710 | /* | |
1711 | * Don't allow the seek distance to get too large from the | |
1712 | * odd fragment, pagein, etc | |
1713 | */ | |
1714 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1715 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1716 | else | |
1717 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1718 | ||
1719 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1720 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1721 | total = cic->seek_total + (cic->seek_samples/2); | |
1722 | do_div(total, cic->seek_samples); | |
1723 | cic->seek_mean = (sector_t)total; | |
1724 | } | |
1da177e4 | 1725 | |
22e2c507 JA |
1726 | /* |
1727 | * Disable idle window if the process thinks too long or seeks so much that | |
1728 | * it doesn't matter | |
1729 | */ | |
1730 | static void | |
1731 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1732 | struct cfq_io_context *cic) | |
1733 | { | |
7b679138 | 1734 | int old_idle, enable_idle; |
1be92f2f | 1735 | |
0871714e JA |
1736 | /* |
1737 | * Don't idle for async or idle io prio class | |
1738 | */ | |
1739 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
1740 | return; |
1741 | ||
c265a7f4 | 1742 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 1743 | |
66dac98e | 1744 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
caaa5f9f | 1745 | (cfqd->hw_tag && CIC_SEEKY(cic))) |
22e2c507 JA |
1746 | enable_idle = 0; |
1747 | else if (sample_valid(cic->ttime_samples)) { | |
1748 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1749 | enable_idle = 0; | |
1750 | else | |
1751 | enable_idle = 1; | |
1da177e4 LT |
1752 | } |
1753 | ||
7b679138 JA |
1754 | if (old_idle != enable_idle) { |
1755 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
1756 | if (enable_idle) | |
1757 | cfq_mark_cfqq_idle_window(cfqq); | |
1758 | else | |
1759 | cfq_clear_cfqq_idle_window(cfqq); | |
1760 | } | |
22e2c507 | 1761 | } |
1da177e4 | 1762 | |
22e2c507 JA |
1763 | /* |
1764 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1765 | * no or if we aren't sure, a 1 will cause a preempt. | |
1766 | */ | |
1767 | static int | |
1768 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1769 | struct request *rq) |
22e2c507 | 1770 | { |
6d048f53 | 1771 | struct cfq_queue *cfqq; |
22e2c507 | 1772 | |
6d048f53 JA |
1773 | cfqq = cfqd->active_queue; |
1774 | if (!cfqq) | |
22e2c507 JA |
1775 | return 0; |
1776 | ||
6d048f53 JA |
1777 | if (cfq_slice_used(cfqq)) |
1778 | return 1; | |
1779 | ||
1780 | if (cfq_class_idle(new_cfqq)) | |
caaa5f9f | 1781 | return 0; |
22e2c507 JA |
1782 | |
1783 | if (cfq_class_idle(cfqq)) | |
1784 | return 1; | |
1e3335de | 1785 | |
374f84ac JA |
1786 | /* |
1787 | * if the new request is sync, but the currently running queue is | |
1788 | * not, let the sync request have priority. | |
1789 | */ | |
5e705374 | 1790 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1791 | return 1; |
1e3335de | 1792 | |
374f84ac JA |
1793 | /* |
1794 | * So both queues are sync. Let the new request get disk time if | |
1795 | * it's a metadata request and the current queue is doing regular IO. | |
1796 | */ | |
1797 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1798 | return 1; | |
22e2c507 | 1799 | |
1e3335de JA |
1800 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
1801 | return 0; | |
1802 | ||
1803 | /* | |
1804 | * if this request is as-good as one we would expect from the | |
1805 | * current cfqq, let it preempt | |
1806 | */ | |
6d048f53 | 1807 | if (cfq_rq_close(cfqd, rq)) |
1e3335de JA |
1808 | return 1; |
1809 | ||
22e2c507 JA |
1810 | return 0; |
1811 | } | |
1812 | ||
1813 | /* | |
1814 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1815 | * let it have half of its nominal slice. | |
1816 | */ | |
1817 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1818 | { | |
7b679138 | 1819 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
6084cdda | 1820 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 1821 | |
bf572256 JA |
1822 | /* |
1823 | * Put the new queue at the front of the of the current list, | |
1824 | * so we know that it will be selected next. | |
1825 | */ | |
1826 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
1827 | |
1828 | cfq_service_tree_add(cfqd, cfqq, 1); | |
bf572256 | 1829 | |
44f7c160 JA |
1830 | cfqq->slice_end = 0; |
1831 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
1832 | } |
1833 | ||
22e2c507 | 1834 | /* |
5e705374 | 1835 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1836 | * something we should do about it |
1837 | */ | |
1838 | static void | |
5e705374 JA |
1839 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1840 | struct request *rq) | |
22e2c507 | 1841 | { |
5e705374 | 1842 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1843 | |
45333d5a | 1844 | cfqd->rq_queued++; |
374f84ac JA |
1845 | if (rq_is_meta(rq)) |
1846 | cfqq->meta_pending++; | |
1847 | ||
9c2c38a1 | 1848 | cfq_update_io_thinktime(cfqd, cic); |
6d048f53 | 1849 | cfq_update_io_seektime(cfqd, cic, rq); |
9c2c38a1 JA |
1850 | cfq_update_idle_window(cfqd, cfqq, cic); |
1851 | ||
5e705374 | 1852 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
22e2c507 JA |
1853 | |
1854 | if (cfqq == cfqd->active_queue) { | |
1855 | /* | |
1856 | * if we are waiting for a request for this queue, let it rip | |
1857 | * immediately and flag that we must not expire this queue | |
1858 | * just now | |
1859 | */ | |
3b18152c JA |
1860 | if (cfq_cfqq_wait_request(cfqq)) { |
1861 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 | 1862 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1863 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1864 | } |
5e705374 | 1865 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1866 | /* |
1867 | * not the active queue - expire current slice if it is | |
1868 | * idle and has expired it's mean thinktime or this new queue | |
1869 | * has some old slice time left and is of higher priority | |
1870 | */ | |
1871 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1872 | cfq_mark_cfqq_must_dispatch(cfqq); |
dc72ef4a | 1873 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1874 | } |
1da177e4 LT |
1875 | } |
1876 | ||
165125e1 | 1877 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1878 | { |
b4878f24 | 1879 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1880 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1881 | |
7b679138 | 1882 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
fd0928df | 1883 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
1da177e4 | 1884 | |
5e705374 | 1885 | cfq_add_rq_rb(rq); |
1da177e4 | 1886 | |
22e2c507 JA |
1887 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1888 | ||
5e705374 | 1889 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1890 | } |
1891 | ||
45333d5a AC |
1892 | /* |
1893 | * Update hw_tag based on peak queue depth over 50 samples under | |
1894 | * sufficient load. | |
1895 | */ | |
1896 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
1897 | { | |
1898 | if (cfqd->rq_in_driver > cfqd->rq_in_driver_peak) | |
1899 | cfqd->rq_in_driver_peak = cfqd->rq_in_driver; | |
1900 | ||
1901 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
1902 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) | |
1903 | return; | |
1904 | ||
1905 | if (cfqd->hw_tag_samples++ < 50) | |
1906 | return; | |
1907 | ||
1908 | if (cfqd->rq_in_driver_peak >= CFQ_HW_QUEUE_MIN) | |
1909 | cfqd->hw_tag = 1; | |
1910 | else | |
1911 | cfqd->hw_tag = 0; | |
1912 | ||
1913 | cfqd->hw_tag_samples = 0; | |
1914 | cfqd->rq_in_driver_peak = 0; | |
1915 | } | |
1916 | ||
165125e1 | 1917 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1918 | { |
5e705374 | 1919 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1920 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1921 | const int sync = rq_is_sync(rq); |
b4878f24 | 1922 | unsigned long now; |
1da177e4 | 1923 | |
b4878f24 | 1924 | now = jiffies; |
7b679138 | 1925 | cfq_log_cfqq(cfqd, cfqq, "complete"); |
1da177e4 | 1926 | |
45333d5a AC |
1927 | cfq_update_hw_tag(cfqd); |
1928 | ||
b4878f24 | 1929 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 1930 | WARN_ON(!cfqq->dispatched); |
b4878f24 | 1931 | cfqd->rq_in_driver--; |
6d048f53 | 1932 | cfqq->dispatched--; |
1da177e4 | 1933 | |
3ed9a296 JA |
1934 | if (cfq_cfqq_sync(cfqq)) |
1935 | cfqd->sync_flight--; | |
1936 | ||
b4878f24 JA |
1937 | if (!cfq_class_idle(cfqq)) |
1938 | cfqd->last_end_request = now; | |
3b18152c | 1939 | |
caaa5f9f | 1940 | if (sync) |
5e705374 | 1941 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1942 | |
1943 | /* | |
1944 | * If this is the active queue, check if it needs to be expired, | |
1945 | * or if we want to idle in case it has no pending requests. | |
1946 | */ | |
1947 | if (cfqd->active_queue == cfqq) { | |
44f7c160 JA |
1948 | if (cfq_cfqq_slice_new(cfqq)) { |
1949 | cfq_set_prio_slice(cfqd, cfqq); | |
1950 | cfq_clear_cfqq_slice_new(cfqq); | |
1951 | } | |
0871714e | 1952 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
6084cdda | 1953 | cfq_slice_expired(cfqd, 1); |
6d048f53 JA |
1954 | else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1955 | cfq_arm_slice_timer(cfqd); | |
caaa5f9f | 1956 | } |
6d048f53 JA |
1957 | |
1958 | if (!cfqd->rq_in_driver) | |
1959 | cfq_schedule_dispatch(cfqd); | |
1da177e4 LT |
1960 | } |
1961 | ||
22e2c507 JA |
1962 | /* |
1963 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
1964 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
1965 | */ | |
1966 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 1967 | { |
22e2c507 JA |
1968 | if (has_fs_excl()) { |
1969 | /* | |
1970 | * boost idle prio on transactions that would lock out other | |
1971 | * users of the filesystem | |
1972 | */ | |
1973 | if (cfq_class_idle(cfqq)) | |
1974 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1975 | if (cfqq->ioprio > IOPRIO_NORM) | |
1976 | cfqq->ioprio = IOPRIO_NORM; | |
1977 | } else { | |
1978 | /* | |
1979 | * check if we need to unboost the queue | |
1980 | */ | |
1981 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
1982 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
1983 | if (cfqq->ioprio != cfqq->org_ioprio) | |
1984 | cfqq->ioprio = cfqq->org_ioprio; | |
1985 | } | |
22e2c507 | 1986 | } |
1da177e4 | 1987 | |
89850f7e | 1988 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 1989 | { |
3b18152c | 1990 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 1991 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 1992 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 1993 | return ELV_MQUEUE_MUST; |
3b18152c | 1994 | } |
1da177e4 | 1995 | |
22e2c507 | 1996 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
1997 | } |
1998 | ||
165125e1 | 1999 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
2000 | { |
2001 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2002 | struct task_struct *tsk = current; | |
91fac317 | 2003 | struct cfq_io_context *cic; |
22e2c507 JA |
2004 | struct cfq_queue *cfqq; |
2005 | ||
2006 | /* | |
2007 | * don't force setup of a queue from here, as a call to may_queue | |
2008 | * does not necessarily imply that a request actually will be queued. | |
2009 | * so just lookup a possibly existing queue, or return 'may queue' | |
2010 | * if that fails | |
2011 | */ | |
4ac845a2 | 2012 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2013 | if (!cic) |
2014 | return ELV_MQUEUE_MAY; | |
2015 | ||
2016 | cfqq = cic_to_cfqq(cic, rw & REQ_RW_SYNC); | |
22e2c507 | 2017 | if (cfqq) { |
fd0928df | 2018 | cfq_init_prio_data(cfqq, cic->ioc); |
22e2c507 JA |
2019 | cfq_prio_boost(cfqq); |
2020 | ||
89850f7e | 2021 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
2022 | } |
2023 | ||
2024 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
2025 | } |
2026 | ||
1da177e4 LT |
2027 | /* |
2028 | * queue lock held here | |
2029 | */ | |
bb37b94c | 2030 | static void cfq_put_request(struct request *rq) |
1da177e4 | 2031 | { |
5e705374 | 2032 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2033 | |
5e705374 | 2034 | if (cfqq) { |
22e2c507 | 2035 | const int rw = rq_data_dir(rq); |
1da177e4 | 2036 | |
22e2c507 JA |
2037 | BUG_ON(!cfqq->allocated[rw]); |
2038 | cfqq->allocated[rw]--; | |
1da177e4 | 2039 | |
5e705374 | 2040 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 2041 | |
1da177e4 | 2042 | rq->elevator_private = NULL; |
5e705374 | 2043 | rq->elevator_private2 = NULL; |
1da177e4 | 2044 | |
1da177e4 LT |
2045 | cfq_put_queue(cfqq); |
2046 | } | |
2047 | } | |
2048 | ||
2049 | /* | |
22e2c507 | 2050 | * Allocate cfq data structures associated with this request. |
1da177e4 | 2051 | */ |
22e2c507 | 2052 | static int |
165125e1 | 2053 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
2054 | { |
2055 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2056 | struct cfq_io_context *cic; | |
2057 | const int rw = rq_data_dir(rq); | |
7749a8d4 | 2058 | const int is_sync = rq_is_sync(rq); |
22e2c507 | 2059 | struct cfq_queue *cfqq; |
1da177e4 LT |
2060 | unsigned long flags; |
2061 | ||
2062 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2063 | ||
e2d74ac0 | 2064 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 2065 | |
1da177e4 LT |
2066 | spin_lock_irqsave(q->queue_lock, flags); |
2067 | ||
22e2c507 JA |
2068 | if (!cic) |
2069 | goto queue_fail; | |
2070 | ||
91fac317 VT |
2071 | cfqq = cic_to_cfqq(cic, is_sync); |
2072 | if (!cfqq) { | |
fd0928df | 2073 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
91fac317 | 2074 | |
22e2c507 JA |
2075 | if (!cfqq) |
2076 | goto queue_fail; | |
1da177e4 | 2077 | |
91fac317 VT |
2078 | cic_set_cfqq(cic, cfqq, is_sync); |
2079 | } | |
1da177e4 LT |
2080 | |
2081 | cfqq->allocated[rw]++; | |
3b18152c | 2082 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 2083 | atomic_inc(&cfqq->ref); |
1da177e4 | 2084 | |
5e705374 | 2085 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 2086 | |
5e705374 JA |
2087 | rq->elevator_private = cic; |
2088 | rq->elevator_private2 = cfqq; | |
2089 | return 0; | |
1da177e4 | 2090 | |
22e2c507 JA |
2091 | queue_fail: |
2092 | if (cic) | |
2093 | put_io_context(cic->ioc); | |
89850f7e | 2094 | |
3b18152c | 2095 | cfq_schedule_dispatch(cfqd); |
1da177e4 | 2096 | spin_unlock_irqrestore(q->queue_lock, flags); |
7b679138 | 2097 | cfq_log(cfqd, "set_request fail"); |
1da177e4 LT |
2098 | return 1; |
2099 | } | |
2100 | ||
65f27f38 | 2101 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 2102 | { |
65f27f38 DH |
2103 | struct cfq_data *cfqd = |
2104 | container_of(work, struct cfq_data, unplug_work); | |
165125e1 | 2105 | struct request_queue *q = cfqd->queue; |
22e2c507 JA |
2106 | unsigned long flags; |
2107 | ||
2108 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 2109 | blk_start_queueing(q); |
22e2c507 JA |
2110 | spin_unlock_irqrestore(q->queue_lock, flags); |
2111 | } | |
2112 | ||
2113 | /* | |
2114 | * Timer running if the active_queue is currently idling inside its time slice | |
2115 | */ | |
2116 | static void cfq_idle_slice_timer(unsigned long data) | |
2117 | { | |
2118 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
2119 | struct cfq_queue *cfqq; | |
2120 | unsigned long flags; | |
3c6bd2f8 | 2121 | int timed_out = 1; |
22e2c507 | 2122 | |
7b679138 JA |
2123 | cfq_log(cfqd, "idle timer fired"); |
2124 | ||
22e2c507 JA |
2125 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
2126 | ||
fe094d98 JA |
2127 | cfqq = cfqd->active_queue; |
2128 | if (cfqq) { | |
3c6bd2f8 JA |
2129 | timed_out = 0; |
2130 | ||
22e2c507 JA |
2131 | /* |
2132 | * expired | |
2133 | */ | |
44f7c160 | 2134 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
2135 | goto expire; |
2136 | ||
2137 | /* | |
2138 | * only expire and reinvoke request handler, if there are | |
2139 | * other queues with pending requests | |
2140 | */ | |
caaa5f9f | 2141 | if (!cfqd->busy_queues) |
22e2c507 | 2142 | goto out_cont; |
22e2c507 JA |
2143 | |
2144 | /* | |
2145 | * not expired and it has a request pending, let it dispatch | |
2146 | */ | |
dd67d051 | 2147 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 2148 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
2149 | goto out_kick; |
2150 | } | |
2151 | } | |
2152 | expire: | |
6084cdda | 2153 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 2154 | out_kick: |
3b18152c | 2155 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
2156 | out_cont: |
2157 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
2158 | } | |
2159 | ||
3b18152c JA |
2160 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
2161 | { | |
2162 | del_timer_sync(&cfqd->idle_slice_timer); | |
4310864b | 2163 | kblockd_flush_work(&cfqd->unplug_work); |
3b18152c | 2164 | } |
22e2c507 | 2165 | |
c2dea2d1 VT |
2166 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
2167 | { | |
2168 | int i; | |
2169 | ||
2170 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
2171 | if (cfqd->async_cfqq[0][i]) | |
2172 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
2173 | if (cfqd->async_cfqq[1][i]) | |
2174 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 2175 | } |
2389d1ef ON |
2176 | |
2177 | if (cfqd->async_idle_cfqq) | |
2178 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
2179 | } |
2180 | ||
1da177e4 LT |
2181 | static void cfq_exit_queue(elevator_t *e) |
2182 | { | |
22e2c507 | 2183 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 2184 | struct request_queue *q = cfqd->queue; |
22e2c507 | 2185 | |
3b18152c | 2186 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 2187 | |
d9ff4187 | 2188 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 2189 | |
d9ff4187 | 2190 | if (cfqd->active_queue) |
6084cdda | 2191 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
2192 | |
2193 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
2194 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
2195 | struct cfq_io_context, | |
2196 | queue_list); | |
89850f7e JA |
2197 | |
2198 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 2199 | } |
e2d74ac0 | 2200 | |
c2dea2d1 | 2201 | cfq_put_async_queues(cfqd); |
15c31be4 | 2202 | |
d9ff4187 | 2203 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
2204 | |
2205 | cfq_shutdown_timer_wq(cfqd); | |
2206 | ||
a90d742e | 2207 | kfree(cfqd); |
1da177e4 LT |
2208 | } |
2209 | ||
165125e1 | 2210 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
2211 | { |
2212 | struct cfq_data *cfqd; | |
1da177e4 | 2213 | |
94f6030c | 2214 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
1da177e4 | 2215 | if (!cfqd) |
bc1c1169 | 2216 | return NULL; |
1da177e4 | 2217 | |
cc09e299 | 2218 | cfqd->service_tree = CFQ_RB_ROOT; |
d9ff4187 | 2219 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2220 | |
1da177e4 | 2221 | cfqd->queue = q; |
1da177e4 | 2222 | |
22e2c507 JA |
2223 | init_timer(&cfqd->idle_slice_timer); |
2224 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2225 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2226 | ||
65f27f38 | 2227 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2228 | |
b70c864d | 2229 | cfqd->last_end_request = jiffies; |
1da177e4 | 2230 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2231 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2232 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2233 | cfqd->cfq_back_max = cfq_back_max; |
2234 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2235 | cfqd->cfq_slice[0] = cfq_slice_async; |
2236 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2237 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2238 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
45333d5a | 2239 | cfqd->hw_tag = 1; |
3b18152c | 2240 | |
bc1c1169 | 2241 | return cfqd; |
1da177e4 LT |
2242 | } |
2243 | ||
2244 | static void cfq_slab_kill(void) | |
2245 | { | |
d6de8be7 JA |
2246 | /* |
2247 | * Caller already ensured that pending RCU callbacks are completed, | |
2248 | * so we should have no busy allocations at this point. | |
2249 | */ | |
1da177e4 LT |
2250 | if (cfq_pool) |
2251 | kmem_cache_destroy(cfq_pool); | |
2252 | if (cfq_ioc_pool) | |
2253 | kmem_cache_destroy(cfq_ioc_pool); | |
2254 | } | |
2255 | ||
2256 | static int __init cfq_slab_setup(void) | |
2257 | { | |
0a31bd5f | 2258 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
1da177e4 LT |
2259 | if (!cfq_pool) |
2260 | goto fail; | |
2261 | ||
34e6bbf2 | 2262 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
1da177e4 LT |
2263 | if (!cfq_ioc_pool) |
2264 | goto fail; | |
2265 | ||
2266 | return 0; | |
2267 | fail: | |
2268 | cfq_slab_kill(); | |
2269 | return -ENOMEM; | |
2270 | } | |
2271 | ||
1da177e4 LT |
2272 | /* |
2273 | * sysfs parts below --> | |
2274 | */ | |
1da177e4 LT |
2275 | static ssize_t |
2276 | cfq_var_show(unsigned int var, char *page) | |
2277 | { | |
2278 | return sprintf(page, "%d\n", var); | |
2279 | } | |
2280 | ||
2281 | static ssize_t | |
2282 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2283 | { | |
2284 | char *p = (char *) page; | |
2285 | ||
2286 | *var = simple_strtoul(p, &p, 10); | |
2287 | return count; | |
2288 | } | |
2289 | ||
1da177e4 | 2290 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
3d1ab40f | 2291 | static ssize_t __FUNC(elevator_t *e, char *page) \ |
1da177e4 | 2292 | { \ |
3d1ab40f | 2293 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2294 | unsigned int __data = __VAR; \ |
2295 | if (__CONV) \ | |
2296 | __data = jiffies_to_msecs(__data); \ | |
2297 | return cfq_var_show(__data, (page)); \ | |
2298 | } | |
2299 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2300 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2301 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2302 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2303 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2304 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2305 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2306 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2307 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2308 | #undef SHOW_FUNCTION |
2309 | ||
2310 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
3d1ab40f | 2311 | static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \ |
1da177e4 | 2312 | { \ |
3d1ab40f | 2313 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2314 | unsigned int __data; \ |
2315 | int ret = cfq_var_store(&__data, (page), count); \ | |
2316 | if (__data < (MIN)) \ | |
2317 | __data = (MIN); \ | |
2318 | else if (__data > (MAX)) \ | |
2319 | __data = (MAX); \ | |
2320 | if (__CONV) \ | |
2321 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2322 | else \ | |
2323 | *(__PTR) = __data; \ | |
2324 | return ret; \ | |
2325 | } | |
2326 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
2327 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
2328 | UINT_MAX, 1); | |
2329 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
2330 | UINT_MAX, 1); | |
e572ec7e | 2331 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
2332 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
2333 | UINT_MAX, 0); | |
22e2c507 JA |
2334 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2335 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2336 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
2337 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
2338 | UINT_MAX, 0); | |
1da177e4 LT |
2339 | #undef STORE_FUNCTION |
2340 | ||
e572ec7e AV |
2341 | #define CFQ_ATTR(name) \ |
2342 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2343 | ||
2344 | static struct elv_fs_entry cfq_attrs[] = { | |
2345 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2346 | CFQ_ATTR(fifo_expire_sync), |
2347 | CFQ_ATTR(fifo_expire_async), | |
2348 | CFQ_ATTR(back_seek_max), | |
2349 | CFQ_ATTR(back_seek_penalty), | |
2350 | CFQ_ATTR(slice_sync), | |
2351 | CFQ_ATTR(slice_async), | |
2352 | CFQ_ATTR(slice_async_rq), | |
2353 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2354 | __ATTR_NULL |
1da177e4 LT |
2355 | }; |
2356 | ||
1da177e4 LT |
2357 | static struct elevator_type iosched_cfq = { |
2358 | .ops = { | |
2359 | .elevator_merge_fn = cfq_merge, | |
2360 | .elevator_merged_fn = cfq_merged_request, | |
2361 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2362 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2363 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2364 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2365 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2366 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2367 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2368 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2369 | .elevator_former_req_fn = elv_rb_former_request, |
2370 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2371 | .elevator_set_req_fn = cfq_set_request, |
2372 | .elevator_put_req_fn = cfq_put_request, | |
2373 | .elevator_may_queue_fn = cfq_may_queue, | |
2374 | .elevator_init_fn = cfq_init_queue, | |
2375 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2376 | .trim = cfq_free_io_context, |
1da177e4 | 2377 | }, |
3d1ab40f | 2378 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2379 | .elevator_name = "cfq", |
2380 | .elevator_owner = THIS_MODULE, | |
2381 | }; | |
2382 | ||
2383 | static int __init cfq_init(void) | |
2384 | { | |
22e2c507 JA |
2385 | /* |
2386 | * could be 0 on HZ < 1000 setups | |
2387 | */ | |
2388 | if (!cfq_slice_async) | |
2389 | cfq_slice_async = 1; | |
2390 | if (!cfq_slice_idle) | |
2391 | cfq_slice_idle = 1; | |
2392 | ||
1da177e4 LT |
2393 | if (cfq_slab_setup()) |
2394 | return -ENOMEM; | |
2395 | ||
2fdd82bd | 2396 | elv_register(&iosched_cfq); |
1da177e4 | 2397 | |
2fdd82bd | 2398 | return 0; |
1da177e4 LT |
2399 | } |
2400 | ||
2401 | static void __exit cfq_exit(void) | |
2402 | { | |
6e9a4738 | 2403 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2404 | elv_unregister(&iosched_cfq); |
334e94de | 2405 | ioc_gone = &all_gone; |
fba82272 OH |
2406 | /* ioc_gone's update must be visible before reading ioc_count */ |
2407 | smp_wmb(); | |
d6de8be7 JA |
2408 | |
2409 | /* | |
2410 | * this also protects us from entering cfq_slab_kill() with | |
2411 | * pending RCU callbacks | |
2412 | */ | |
4050cf16 | 2413 | if (elv_ioc_count_read(ioc_count)) |
9a11b4ed | 2414 | wait_for_completion(&all_gone); |
83521d3e | 2415 | cfq_slab_kill(); |
1da177e4 LT |
2416 | } |
2417 | ||
2418 | module_init(cfq_init); | |
2419 | module_exit(cfq_exit); | |
2420 | ||
2421 | MODULE_AUTHOR("Jens Axboe"); | |
2422 | MODULE_LICENSE("GPL"); | |
2423 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |