]>
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 | 86 | unsigned int busy_queues; |
3a9a3f6c DS |
87 | /* |
88 | * Used to track any pending rt requests so we can pre-empt current | |
89 | * non-RT cfqq in service when this value is non-zero. | |
90 | */ | |
91 | unsigned int busy_rt_queues; | |
22e2c507 | 92 | |
22e2c507 | 93 | int rq_in_driver; |
3ed9a296 | 94 | int sync_flight; |
45333d5a AC |
95 | |
96 | /* | |
97 | * queue-depth detection | |
98 | */ | |
99 | int rq_queued; | |
25776e35 | 100 | int hw_tag; |
45333d5a AC |
101 | int hw_tag_samples; |
102 | int rq_in_driver_peak; | |
1da177e4 | 103 | |
22e2c507 JA |
104 | /* |
105 | * idle window management | |
106 | */ | |
107 | struct timer_list idle_slice_timer; | |
108 | struct work_struct unplug_work; | |
1da177e4 | 109 | |
22e2c507 JA |
110 | struct cfq_queue *active_queue; |
111 | struct cfq_io_context *active_cic; | |
22e2c507 | 112 | |
c2dea2d1 VT |
113 | /* |
114 | * async queue for each priority case | |
115 | */ | |
116 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
117 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 118 | |
6d048f53 | 119 | sector_t last_position; |
22e2c507 | 120 | unsigned long last_end_request; |
1da177e4 | 121 | |
1da177e4 LT |
122 | /* |
123 | * tunables, see top of file | |
124 | */ | |
125 | unsigned int cfq_quantum; | |
22e2c507 | 126 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
127 | unsigned int cfq_back_penalty; |
128 | unsigned int cfq_back_max; | |
22e2c507 JA |
129 | unsigned int cfq_slice[2]; |
130 | unsigned int cfq_slice_async_rq; | |
131 | unsigned int cfq_slice_idle; | |
d9ff4187 AV |
132 | |
133 | struct list_head cic_list; | |
1da177e4 LT |
134 | }; |
135 | ||
22e2c507 JA |
136 | /* |
137 | * Per process-grouping structure | |
138 | */ | |
1da177e4 LT |
139 | struct cfq_queue { |
140 | /* reference count */ | |
141 | atomic_t ref; | |
be754d2c RK |
142 | /* various state flags, see below */ |
143 | unsigned int flags; | |
1da177e4 LT |
144 | /* parent cfq_data */ |
145 | struct cfq_data *cfqd; | |
d9e7620e JA |
146 | /* service_tree member */ |
147 | struct rb_node rb_node; | |
148 | /* service_tree key */ | |
149 | unsigned long rb_key; | |
1da177e4 LT |
150 | /* sorted list of pending requests */ |
151 | struct rb_root sort_list; | |
152 | /* if fifo isn't expired, next request to serve */ | |
5e705374 | 153 | struct request *next_rq; |
1da177e4 LT |
154 | /* requests queued in sort_list */ |
155 | int queued[2]; | |
156 | /* currently allocated requests */ | |
157 | int allocated[2]; | |
158 | /* fifo list of requests in sort_list */ | |
22e2c507 | 159 | struct list_head fifo; |
1da177e4 | 160 | |
22e2c507 | 161 | unsigned long slice_end; |
c5b680f3 | 162 | long slice_resid; |
2f5cb738 | 163 | unsigned int slice_dispatch; |
1da177e4 | 164 | |
be754d2c RK |
165 | /* pending metadata requests */ |
166 | int meta_pending; | |
6d048f53 JA |
167 | /* number of requests that are on the dispatch list or inside driver */ |
168 | int dispatched; | |
22e2c507 JA |
169 | |
170 | /* io prio of this group */ | |
171 | unsigned short ioprio, org_ioprio; | |
172 | unsigned short ioprio_class, org_ioprio_class; | |
173 | ||
7b679138 | 174 | pid_t pid; |
1da177e4 LT |
175 | }; |
176 | ||
3b18152c | 177 | enum cfqq_state_flags { |
b0b8d749 JA |
178 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
179 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
180 | CFQ_CFQQ_FLAG_must_alloc, /* must be allowed rq alloc */ | |
181 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ | |
b0b8d749 JA |
182 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
183 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
184 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 185 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 186 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
3b18152c JA |
187 | }; |
188 | ||
189 | #define CFQ_CFQQ_FNS(name) \ | |
190 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
191 | { \ | |
fe094d98 | 192 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
193 | } \ |
194 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
195 | { \ | |
fe094d98 | 196 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
197 | } \ |
198 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
199 | { \ | |
fe094d98 | 200 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
201 | } |
202 | ||
203 | CFQ_CFQQ_FNS(on_rr); | |
204 | CFQ_CFQQ_FNS(wait_request); | |
205 | CFQ_CFQQ_FNS(must_alloc); | |
206 | CFQ_CFQQ_FNS(must_alloc_slice); | |
3b18152c JA |
207 | CFQ_CFQQ_FNS(fifo_expire); |
208 | CFQ_CFQQ_FNS(idle_window); | |
209 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 210 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 211 | CFQ_CFQQ_FNS(sync); |
3b18152c JA |
212 | #undef CFQ_CFQQ_FNS |
213 | ||
7b679138 JA |
214 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
215 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
216 | #define cfq_log(cfqd, fmt, args...) \ | |
217 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
218 | ||
165125e1 | 219 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
91fac317 | 220 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, int, |
fd0928df | 221 | struct io_context *, gfp_t); |
4ac845a2 | 222 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
91fac317 VT |
223 | struct io_context *); |
224 | ||
225 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, | |
226 | int is_sync) | |
227 | { | |
228 | return cic->cfqq[!!is_sync]; | |
229 | } | |
230 | ||
231 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
232 | struct cfq_queue *cfqq, int is_sync) | |
233 | { | |
234 | cic->cfqq[!!is_sync] = cfqq; | |
235 | } | |
236 | ||
237 | /* | |
238 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
239 | * set (in which case it could also be direct WRITE). | |
240 | */ | |
241 | static inline int cfq_bio_sync(struct bio *bio) | |
242 | { | |
243 | if (bio_data_dir(bio) == READ || bio_sync(bio)) | |
244 | return 1; | |
245 | ||
246 | return 0; | |
247 | } | |
1da177e4 | 248 | |
99f95e52 AM |
249 | /* |
250 | * scheduler run of queue, if there are requests pending and no one in the | |
251 | * driver that will restart queueing | |
252 | */ | |
253 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
254 | { | |
7b679138 JA |
255 | if (cfqd->busy_queues) { |
256 | cfq_log(cfqd, "schedule dispatch"); | |
18887ad9 | 257 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 258 | } |
99f95e52 AM |
259 | } |
260 | ||
165125e1 | 261 | static int cfq_queue_empty(struct request_queue *q) |
99f95e52 AM |
262 | { |
263 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
264 | ||
b4878f24 | 265 | return !cfqd->busy_queues; |
99f95e52 AM |
266 | } |
267 | ||
44f7c160 JA |
268 | /* |
269 | * Scale schedule slice based on io priority. Use the sync time slice only | |
270 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
271 | * io only, should not get full sync slice length. | |
272 | */ | |
d9e7620e JA |
273 | static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync, |
274 | unsigned short prio) | |
44f7c160 | 275 | { |
d9e7620e | 276 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 277 | |
d9e7620e JA |
278 | WARN_ON(prio >= IOPRIO_BE_NR); |
279 | ||
280 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
281 | } | |
44f7c160 | 282 | |
d9e7620e JA |
283 | static inline int |
284 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
285 | { | |
286 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
287 | } |
288 | ||
289 | static inline void | |
290 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
291 | { | |
292 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; | |
7b679138 | 293 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
294 | } |
295 | ||
296 | /* | |
297 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
298 | * isn't valid until the first request from the dispatch is activated | |
299 | * and the slice time set. | |
300 | */ | |
301 | static inline int cfq_slice_used(struct cfq_queue *cfqq) | |
302 | { | |
303 | if (cfq_cfqq_slice_new(cfqq)) | |
304 | return 0; | |
305 | if (time_before(jiffies, cfqq->slice_end)) | |
306 | return 0; | |
307 | ||
308 | return 1; | |
309 | } | |
310 | ||
1da177e4 | 311 | /* |
5e705374 | 312 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 313 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 314 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 315 | */ |
5e705374 JA |
316 | static struct request * |
317 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
318 | { |
319 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 320 | unsigned long back_max; |
e8a99053 AM |
321 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
322 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
323 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 324 | |
5e705374 JA |
325 | if (rq1 == NULL || rq1 == rq2) |
326 | return rq2; | |
327 | if (rq2 == NULL) | |
328 | return rq1; | |
9c2c38a1 | 329 | |
5e705374 JA |
330 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
331 | return rq1; | |
332 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
333 | return rq2; | |
374f84ac JA |
334 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
335 | return rq1; | |
336 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
337 | return rq2; | |
1da177e4 | 338 | |
5e705374 JA |
339 | s1 = rq1->sector; |
340 | s2 = rq2->sector; | |
1da177e4 | 341 | |
6d048f53 | 342 | last = cfqd->last_position; |
1da177e4 | 343 | |
1da177e4 LT |
344 | /* |
345 | * by definition, 1KiB is 2 sectors | |
346 | */ | |
347 | back_max = cfqd->cfq_back_max * 2; | |
348 | ||
349 | /* | |
350 | * Strict one way elevator _except_ in the case where we allow | |
351 | * short backward seeks which are biased as twice the cost of a | |
352 | * similar forward seek. | |
353 | */ | |
354 | if (s1 >= last) | |
355 | d1 = s1 - last; | |
356 | else if (s1 + back_max >= last) | |
357 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
358 | else | |
e8a99053 | 359 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
360 | |
361 | if (s2 >= last) | |
362 | d2 = s2 - last; | |
363 | else if (s2 + back_max >= last) | |
364 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
365 | else | |
e8a99053 | 366 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
367 | |
368 | /* Found required data */ | |
e8a99053 AM |
369 | |
370 | /* | |
371 | * By doing switch() on the bit mask "wrap" we avoid having to | |
372 | * check two variables for all permutations: --> faster! | |
373 | */ | |
374 | switch (wrap) { | |
5e705374 | 375 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 376 | if (d1 < d2) |
5e705374 | 377 | return rq1; |
e8a99053 | 378 | else if (d2 < d1) |
5e705374 | 379 | return rq2; |
e8a99053 AM |
380 | else { |
381 | if (s1 >= s2) | |
5e705374 | 382 | return rq1; |
e8a99053 | 383 | else |
5e705374 | 384 | return rq2; |
e8a99053 | 385 | } |
1da177e4 | 386 | |
e8a99053 | 387 | case CFQ_RQ2_WRAP: |
5e705374 | 388 | return rq1; |
e8a99053 | 389 | case CFQ_RQ1_WRAP: |
5e705374 JA |
390 | return rq2; |
391 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
392 | default: |
393 | /* | |
394 | * Since both rqs are wrapped, | |
395 | * start with the one that's further behind head | |
396 | * (--> only *one* back seek required), | |
397 | * since back seek takes more time than forward. | |
398 | */ | |
399 | if (s1 <= s2) | |
5e705374 | 400 | return rq1; |
1da177e4 | 401 | else |
5e705374 | 402 | return rq2; |
1da177e4 LT |
403 | } |
404 | } | |
405 | ||
498d3aa2 JA |
406 | /* |
407 | * The below is leftmost cache rbtree addon | |
408 | */ | |
0871714e | 409 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 JA |
410 | { |
411 | if (!root->left) | |
412 | root->left = rb_first(&root->rb); | |
413 | ||
0871714e JA |
414 | if (root->left) |
415 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
416 | ||
417 | return NULL; | |
cc09e299 JA |
418 | } |
419 | ||
420 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) | |
421 | { | |
422 | if (root->left == n) | |
423 | root->left = NULL; | |
424 | ||
425 | rb_erase(n, &root->rb); | |
426 | RB_CLEAR_NODE(n); | |
427 | } | |
428 | ||
1da177e4 LT |
429 | /* |
430 | * would be nice to take fifo expire time into account as well | |
431 | */ | |
5e705374 JA |
432 | static struct request * |
433 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
434 | struct request *last) | |
1da177e4 | 435 | { |
21183b07 JA |
436 | struct rb_node *rbnext = rb_next(&last->rb_node); |
437 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 438 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 439 | |
21183b07 | 440 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
441 | |
442 | if (rbprev) | |
5e705374 | 443 | prev = rb_entry_rq(rbprev); |
1da177e4 | 444 | |
21183b07 | 445 | if (rbnext) |
5e705374 | 446 | next = rb_entry_rq(rbnext); |
21183b07 JA |
447 | else { |
448 | rbnext = rb_first(&cfqq->sort_list); | |
449 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 450 | next = rb_entry_rq(rbnext); |
21183b07 | 451 | } |
1da177e4 | 452 | |
21183b07 | 453 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
454 | } |
455 | ||
d9e7620e JA |
456 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
457 | struct cfq_queue *cfqq) | |
1da177e4 | 458 | { |
d9e7620e JA |
459 | /* |
460 | * just an approximation, should be ok. | |
461 | */ | |
67e6b49e JA |
462 | return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
463 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); | |
d9e7620e JA |
464 | } |
465 | ||
498d3aa2 JA |
466 | /* |
467 | * The cfqd->service_tree holds all pending cfq_queue's that have | |
468 | * requests waiting to be processed. It is sorted in the order that | |
469 | * we will service the queues. | |
470 | */ | |
d9e7620e | 471 | static void cfq_service_tree_add(struct cfq_data *cfqd, |
edd75ffd | 472 | struct cfq_queue *cfqq, int add_front) |
d9e7620e | 473 | { |
0871714e JA |
474 | struct rb_node **p, *parent; |
475 | struct cfq_queue *__cfqq; | |
d9e7620e | 476 | unsigned long rb_key; |
498d3aa2 | 477 | int left; |
d9e7620e | 478 | |
0871714e JA |
479 | if (cfq_class_idle(cfqq)) { |
480 | rb_key = CFQ_IDLE_DELAY; | |
481 | parent = rb_last(&cfqd->service_tree.rb); | |
482 | if (parent && parent != &cfqq->rb_node) { | |
483 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
484 | rb_key += __cfqq->rb_key; | |
485 | } else | |
486 | rb_key += jiffies; | |
487 | } else if (!add_front) { | |
edd75ffd JA |
488 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
489 | rb_key += cfqq->slice_resid; | |
490 | cfqq->slice_resid = 0; | |
491 | } else | |
492 | rb_key = 0; | |
1da177e4 | 493 | |
d9e7620e | 494 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
99f9628a | 495 | /* |
d9e7620e | 496 | * same position, nothing more to do |
99f9628a | 497 | */ |
d9e7620e JA |
498 | if (rb_key == cfqq->rb_key) |
499 | return; | |
1da177e4 | 500 | |
cc09e299 | 501 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); |
1da177e4 | 502 | } |
d9e7620e | 503 | |
498d3aa2 | 504 | left = 1; |
0871714e JA |
505 | parent = NULL; |
506 | p = &cfqd->service_tree.rb.rb_node; | |
d9e7620e | 507 | while (*p) { |
67060e37 | 508 | struct rb_node **n; |
cc09e299 | 509 | |
d9e7620e JA |
510 | parent = *p; |
511 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
512 | ||
0c534e0a JA |
513 | /* |
514 | * sort RT queues first, we always want to give | |
67060e37 JA |
515 | * preference to them. IDLE queues goes to the back. |
516 | * after that, sort on the next service time. | |
0c534e0a JA |
517 | */ |
518 | if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq)) | |
67060e37 | 519 | n = &(*p)->rb_left; |
0c534e0a | 520 | else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq)) |
67060e37 JA |
521 | n = &(*p)->rb_right; |
522 | else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq)) | |
523 | n = &(*p)->rb_left; | |
524 | else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq)) | |
525 | n = &(*p)->rb_right; | |
0c534e0a | 526 | else if (rb_key < __cfqq->rb_key) |
67060e37 JA |
527 | n = &(*p)->rb_left; |
528 | else | |
529 | n = &(*p)->rb_right; | |
530 | ||
531 | if (n == &(*p)->rb_right) | |
cc09e299 | 532 | left = 0; |
67060e37 JA |
533 | |
534 | p = n; | |
d9e7620e JA |
535 | } |
536 | ||
cc09e299 JA |
537 | if (left) |
538 | cfqd->service_tree.left = &cfqq->rb_node; | |
539 | ||
d9e7620e JA |
540 | cfqq->rb_key = rb_key; |
541 | rb_link_node(&cfqq->rb_node, parent, p); | |
cc09e299 | 542 | rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb); |
1da177e4 LT |
543 | } |
544 | ||
498d3aa2 JA |
545 | /* |
546 | * Update cfqq's position in the service tree. | |
547 | */ | |
edd75ffd | 548 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 549 | { |
6d048f53 JA |
550 | /* |
551 | * Resorting requires the cfqq to be on the RR list already. | |
552 | */ | |
498d3aa2 | 553 | if (cfq_cfqq_on_rr(cfqq)) |
edd75ffd | 554 | cfq_service_tree_add(cfqd, cfqq, 0); |
6d048f53 JA |
555 | } |
556 | ||
1da177e4 LT |
557 | /* |
558 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 559 | * the pending list according to last request service |
1da177e4 | 560 | */ |
febffd61 | 561 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 562 | { |
7b679138 | 563 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
564 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
565 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 566 | cfqd->busy_queues++; |
3a9a3f6c DS |
567 | if (cfq_class_rt(cfqq)) |
568 | cfqd->busy_rt_queues++; | |
1da177e4 | 569 | |
edd75ffd | 570 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
571 | } |
572 | ||
498d3aa2 JA |
573 | /* |
574 | * Called when the cfqq no longer has requests pending, remove it from | |
575 | * the service tree. | |
576 | */ | |
febffd61 | 577 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 578 | { |
7b679138 | 579 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
580 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
581 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 582 | |
cc09e299 JA |
583 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) |
584 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); | |
d9e7620e | 585 | |
1da177e4 LT |
586 | BUG_ON(!cfqd->busy_queues); |
587 | cfqd->busy_queues--; | |
3a9a3f6c DS |
588 | if (cfq_class_rt(cfqq)) |
589 | cfqd->busy_rt_queues--; | |
1da177e4 LT |
590 | } |
591 | ||
592 | /* | |
593 | * rb tree support functions | |
594 | */ | |
febffd61 | 595 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 596 | { |
5e705374 | 597 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 598 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 599 | const int sync = rq_is_sync(rq); |
1da177e4 | 600 | |
b4878f24 JA |
601 | BUG_ON(!cfqq->queued[sync]); |
602 | cfqq->queued[sync]--; | |
1da177e4 | 603 | |
5e705374 | 604 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 605 | |
dd67d051 | 606 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 607 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
608 | } |
609 | ||
5e705374 | 610 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 611 | { |
5e705374 | 612 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 613 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 614 | struct request *__alias; |
1da177e4 | 615 | |
5380a101 | 616 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
617 | |
618 | /* | |
619 | * looks a little odd, but the first insert might return an alias. | |
620 | * if that happens, put the alias on the dispatch list | |
621 | */ | |
21183b07 | 622 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 623 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
624 | |
625 | if (!cfq_cfqq_on_rr(cfqq)) | |
626 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
627 | |
628 | /* | |
629 | * check if this request is a better next-serve candidate | |
630 | */ | |
631 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); | |
632 | BUG_ON(!cfqq->next_rq); | |
1da177e4 LT |
633 | } |
634 | ||
febffd61 | 635 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 636 | { |
5380a101 JA |
637 | elv_rb_del(&cfqq->sort_list, rq); |
638 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 639 | cfq_add_rq_rb(rq); |
1da177e4 LT |
640 | } |
641 | ||
206dc69b JA |
642 | static struct request * |
643 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 644 | { |
206dc69b | 645 | struct task_struct *tsk = current; |
91fac317 | 646 | struct cfq_io_context *cic; |
206dc69b | 647 | struct cfq_queue *cfqq; |
1da177e4 | 648 | |
4ac845a2 | 649 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
650 | if (!cic) |
651 | return NULL; | |
652 | ||
653 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
654 | if (cfqq) { |
655 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
656 | ||
21183b07 | 657 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 658 | } |
1da177e4 | 659 | |
1da177e4 LT |
660 | return NULL; |
661 | } | |
662 | ||
165125e1 | 663 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 664 | { |
22e2c507 | 665 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 666 | |
b4878f24 | 667 | cfqd->rq_in_driver++; |
7b679138 JA |
668 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
669 | cfqd->rq_in_driver); | |
25776e35 | 670 | |
6d048f53 | 671 | cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors; |
1da177e4 LT |
672 | } |
673 | ||
165125e1 | 674 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 675 | { |
b4878f24 JA |
676 | struct cfq_data *cfqd = q->elevator->elevator_data; |
677 | ||
678 | WARN_ON(!cfqd->rq_in_driver); | |
679 | cfqd->rq_in_driver--; | |
7b679138 JA |
680 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
681 | cfqd->rq_in_driver); | |
1da177e4 LT |
682 | } |
683 | ||
b4878f24 | 684 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 685 | { |
5e705374 | 686 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 687 | |
5e705374 JA |
688 | if (cfqq->next_rq == rq) |
689 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 690 | |
b4878f24 | 691 | list_del_init(&rq->queuelist); |
5e705374 | 692 | cfq_del_rq_rb(rq); |
374f84ac | 693 | |
45333d5a | 694 | cfqq->cfqd->rq_queued--; |
374f84ac JA |
695 | if (rq_is_meta(rq)) { |
696 | WARN_ON(!cfqq->meta_pending); | |
697 | cfqq->meta_pending--; | |
698 | } | |
1da177e4 LT |
699 | } |
700 | ||
165125e1 JA |
701 | static int cfq_merge(struct request_queue *q, struct request **req, |
702 | struct bio *bio) | |
1da177e4 LT |
703 | { |
704 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
705 | struct request *__rq; | |
1da177e4 | 706 | |
206dc69b | 707 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 708 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
709 | *req = __rq; |
710 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
711 | } |
712 | ||
713 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
714 | } |
715 | ||
165125e1 | 716 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 717 | int type) |
1da177e4 | 718 | { |
21183b07 | 719 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 720 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 721 | |
5e705374 | 722 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 723 | } |
1da177e4 LT |
724 | } |
725 | ||
726 | static void | |
165125e1 | 727 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
728 | struct request *next) |
729 | { | |
22e2c507 JA |
730 | /* |
731 | * reposition in fifo if next is older than rq | |
732 | */ | |
733 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
734 | time_before(next->start_time, rq->start_time)) | |
735 | list_move(&rq->queuelist, &next->queuelist); | |
736 | ||
b4878f24 | 737 | cfq_remove_request(next); |
22e2c507 JA |
738 | } |
739 | ||
165125e1 | 740 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
741 | struct bio *bio) |
742 | { | |
743 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
91fac317 | 744 | struct cfq_io_context *cic; |
da775265 | 745 | struct cfq_queue *cfqq; |
da775265 JA |
746 | |
747 | /* | |
ec8acb69 | 748 | * Disallow merge of a sync bio into an async request. |
da775265 | 749 | */ |
91fac317 | 750 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
da775265 JA |
751 | return 0; |
752 | ||
753 | /* | |
719d3402 JA |
754 | * Lookup the cfqq that this bio will be queued with. Allow |
755 | * merge only if rq is queued there. | |
da775265 | 756 | */ |
4ac845a2 | 757 | cic = cfq_cic_lookup(cfqd, current->io_context); |
91fac317 VT |
758 | if (!cic) |
759 | return 0; | |
719d3402 | 760 | |
91fac317 | 761 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
719d3402 JA |
762 | if (cfqq == RQ_CFQQ(rq)) |
763 | return 1; | |
da775265 | 764 | |
ec8acb69 | 765 | return 0; |
da775265 JA |
766 | } |
767 | ||
febffd61 JA |
768 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
769 | struct cfq_queue *cfqq) | |
22e2c507 JA |
770 | { |
771 | if (cfqq) { | |
7b679138 | 772 | cfq_log_cfqq(cfqd, cfqq, "set_active"); |
22e2c507 | 773 | cfqq->slice_end = 0; |
2f5cb738 JA |
774 | cfqq->slice_dispatch = 0; |
775 | ||
2f5cb738 | 776 | cfq_clear_cfqq_wait_request(cfqq); |
3b18152c JA |
777 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
778 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 779 | cfq_mark_cfqq_slice_new(cfqq); |
2f5cb738 JA |
780 | |
781 | del_timer(&cfqd->idle_slice_timer); | |
22e2c507 JA |
782 | } |
783 | ||
784 | cfqd->active_queue = cfqq; | |
785 | } | |
786 | ||
7b14e3b5 JA |
787 | /* |
788 | * current cfqq expired its slice (or was too idle), select new one | |
789 | */ | |
790 | static void | |
791 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
6084cdda | 792 | int timed_out) |
7b14e3b5 | 793 | { |
7b679138 JA |
794 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
795 | ||
7b14e3b5 JA |
796 | if (cfq_cfqq_wait_request(cfqq)) |
797 | del_timer(&cfqd->idle_slice_timer); | |
798 | ||
7b14e3b5 JA |
799 | cfq_clear_cfqq_wait_request(cfqq); |
800 | ||
801 | /* | |
6084cdda | 802 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 803 | */ |
7b679138 | 804 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) { |
c5b680f3 | 805 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b679138 JA |
806 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
807 | } | |
7b14e3b5 | 808 | |
edd75ffd | 809 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
810 | |
811 | if (cfqq == cfqd->active_queue) | |
812 | cfqd->active_queue = NULL; | |
813 | ||
814 | if (cfqd->active_cic) { | |
815 | put_io_context(cfqd->active_cic->ioc); | |
816 | cfqd->active_cic = NULL; | |
817 | } | |
7b14e3b5 JA |
818 | } |
819 | ||
6084cdda | 820 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out) |
7b14e3b5 JA |
821 | { |
822 | struct cfq_queue *cfqq = cfqd->active_queue; | |
823 | ||
824 | if (cfqq) | |
6084cdda | 825 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
826 | } |
827 | ||
498d3aa2 JA |
828 | /* |
829 | * Get next queue for service. Unless we have a queue preemption, | |
830 | * we'll simply select the first cfqq in the service tree. | |
831 | */ | |
6d048f53 | 832 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 833 | { |
edd75ffd JA |
834 | if (RB_EMPTY_ROOT(&cfqd->service_tree.rb)) |
835 | return NULL; | |
d9e7620e | 836 | |
0871714e | 837 | return cfq_rb_first(&cfqd->service_tree); |
6d048f53 JA |
838 | } |
839 | ||
498d3aa2 JA |
840 | /* |
841 | * Get and set a new active queue for service. | |
842 | */ | |
6d048f53 JA |
843 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
844 | { | |
845 | struct cfq_queue *cfqq; | |
846 | ||
d9e7620e | 847 | cfqq = cfq_get_next_queue(cfqd); |
22e2c507 | 848 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 849 | return cfqq; |
22e2c507 JA |
850 | } |
851 | ||
d9e7620e JA |
852 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
853 | struct request *rq) | |
854 | { | |
855 | if (rq->sector >= cfqd->last_position) | |
856 | return rq->sector - cfqd->last_position; | |
857 | else | |
858 | return cfqd->last_position - rq->sector; | |
859 | } | |
860 | ||
6d048f53 JA |
861 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq) |
862 | { | |
863 | struct cfq_io_context *cic = cfqd->active_cic; | |
864 | ||
865 | if (!sample_valid(cic->seek_samples)) | |
866 | return 0; | |
867 | ||
868 | return cfq_dist_from_last(cfqd, rq) <= cic->seek_mean; | |
869 | } | |
870 | ||
d9e7620e JA |
871 | static int cfq_close_cooperator(struct cfq_data *cfq_data, |
872 | struct cfq_queue *cfqq) | |
6d048f53 | 873 | { |
6d048f53 | 874 | /* |
d9e7620e JA |
875 | * We should notice if some of the queues are cooperating, eg |
876 | * working closely on the same area of the disk. In that case, | |
877 | * we can group them together and don't waste time idling. | |
6d048f53 | 878 | */ |
d9e7620e | 879 | return 0; |
6d048f53 JA |
880 | } |
881 | ||
882 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (8 * 1024)) | |
caaa5f9f | 883 | |
6d048f53 | 884 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 885 | { |
1792669c | 886 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 887 | struct cfq_io_context *cic; |
7b14e3b5 JA |
888 | unsigned long sl; |
889 | ||
a68bbddb | 890 | /* |
f7d7b7a7 JA |
891 | * SSD device without seek penalty, disable idling. But only do so |
892 | * for devices that support queuing, otherwise we still have a problem | |
893 | * with sync vs async workloads. | |
a68bbddb | 894 | */ |
f7d7b7a7 | 895 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
896 | return; |
897 | ||
dd67d051 | 898 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 899 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
900 | |
901 | /* | |
902 | * idle is disabled, either manually or by past process history | |
903 | */ | |
6d048f53 JA |
904 | if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq)) |
905 | return; | |
906 | ||
7b679138 JA |
907 | /* |
908 | * still requests with the driver, don't idle | |
909 | */ | |
910 | if (cfqd->rq_in_driver) | |
911 | return; | |
912 | ||
22e2c507 JA |
913 | /* |
914 | * task has exited, don't wait | |
915 | */ | |
206dc69b | 916 | cic = cfqd->active_cic; |
66dac98e | 917 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
6d048f53 JA |
918 | return; |
919 | ||
920 | /* | |
921 | * See if this prio level has a good candidate | |
922 | */ | |
1afba045 JA |
923 | if (cfq_close_cooperator(cfqd, cfqq) && |
924 | (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2)) | |
6d048f53 | 925 | return; |
22e2c507 | 926 | |
3b18152c | 927 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 928 | |
206dc69b JA |
929 | /* |
930 | * we don't want to idle for seeks, but we do want to allow | |
931 | * fair distribution of slice time for a process doing back-to-back | |
932 | * seeks. so allow a little bit of time for him to submit a new rq | |
933 | */ | |
6d048f53 | 934 | sl = cfqd->cfq_slice_idle; |
caaa5f9f | 935 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
d9e7620e | 936 | sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT)); |
206dc69b | 937 | |
7b14e3b5 | 938 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
7b679138 | 939 | cfq_log(cfqd, "arm_idle: %lu", sl); |
1da177e4 LT |
940 | } |
941 | ||
498d3aa2 JA |
942 | /* |
943 | * Move request from internal lists to the request queue dispatch list. | |
944 | */ | |
165125e1 | 945 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 946 | { |
3ed9a296 | 947 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 948 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 949 | |
7b679138 JA |
950 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
951 | ||
5380a101 | 952 | cfq_remove_request(rq); |
6d048f53 | 953 | cfqq->dispatched++; |
5380a101 | 954 | elv_dispatch_sort(q, rq); |
3ed9a296 JA |
955 | |
956 | if (cfq_cfqq_sync(cfqq)) | |
957 | cfqd->sync_flight++; | |
1da177e4 LT |
958 | } |
959 | ||
960 | /* | |
961 | * return expired entry, or NULL to just start from scratch in rbtree | |
962 | */ | |
febffd61 | 963 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
964 | { |
965 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 966 | struct request *rq; |
89850f7e | 967 | int fifo; |
1da177e4 | 968 | |
3b18152c | 969 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 970 | return NULL; |
cb887411 JA |
971 | |
972 | cfq_mark_cfqq_fifo_expire(cfqq); | |
973 | ||
89850f7e JA |
974 | if (list_empty(&cfqq->fifo)) |
975 | return NULL; | |
1da177e4 | 976 | |
6d048f53 | 977 | fifo = cfq_cfqq_sync(cfqq); |
89850f7e | 978 | rq = rq_entry_fifo(cfqq->fifo.next); |
1da177e4 | 979 | |
6d048f53 | 980 | if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) |
7b679138 | 981 | rq = NULL; |
1da177e4 | 982 | |
7b679138 | 983 | cfq_log_cfqq(cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 984 | return rq; |
1da177e4 LT |
985 | } |
986 | ||
22e2c507 JA |
987 | static inline int |
988 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
989 | { | |
990 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 991 | |
22e2c507 | 992 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 993 | |
22e2c507 | 994 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
995 | } |
996 | ||
22e2c507 | 997 | /* |
498d3aa2 JA |
998 | * Select a queue for service. If we have a current active queue, |
999 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 1000 | */ |
1b5ed5e1 | 1001 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 1002 | { |
1da177e4 | 1003 | struct cfq_queue *cfqq; |
1da177e4 | 1004 | |
22e2c507 JA |
1005 | cfqq = cfqd->active_queue; |
1006 | if (!cfqq) | |
1007 | goto new_queue; | |
1da177e4 | 1008 | |
22e2c507 | 1009 | /* |
6d048f53 | 1010 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 1011 | */ |
6d048f53 | 1012 | if (cfq_slice_used(cfqq)) |
3b18152c | 1013 | goto expire; |
1da177e4 | 1014 | |
3a9a3f6c DS |
1015 | /* |
1016 | * If we have a RT cfqq waiting, then we pre-empt the current non-rt | |
1017 | * cfqq. | |
1018 | */ | |
1019 | if (!cfq_class_rt(cfqq) && cfqd->busy_rt_queues) { | |
1020 | /* | |
1021 | * We simulate this as cfqq timed out so that it gets to bank | |
1022 | * the remaining of its time slice. | |
1023 | */ | |
1024 | cfq_log_cfqq(cfqd, cfqq, "preempt"); | |
1025 | cfq_slice_expired(cfqd, 1); | |
1026 | goto new_queue; | |
1027 | } | |
1028 | ||
22e2c507 | 1029 | /* |
6d048f53 JA |
1030 | * The active queue has requests and isn't expired, allow it to |
1031 | * dispatch. | |
22e2c507 | 1032 | */ |
dd67d051 | 1033 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 1034 | goto keep_queue; |
6d048f53 JA |
1035 | |
1036 | /* | |
1037 | * No requests pending. If the active queue still has requests in | |
1038 | * flight or is idling for a new request, allow either of these | |
1039 | * conditions to happen (or time out) before selecting a new queue. | |
1040 | */ | |
cc197479 JA |
1041 | if (timer_pending(&cfqd->idle_slice_timer) || |
1042 | (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) { | |
caaa5f9f JA |
1043 | cfqq = NULL; |
1044 | goto keep_queue; | |
22e2c507 JA |
1045 | } |
1046 | ||
3b18152c | 1047 | expire: |
6084cdda | 1048 | cfq_slice_expired(cfqd, 0); |
3b18152c JA |
1049 | new_queue: |
1050 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 1051 | keep_queue: |
3b18152c | 1052 | return cfqq; |
22e2c507 JA |
1053 | } |
1054 | ||
febffd61 | 1055 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
1056 | { |
1057 | int dispatched = 0; | |
1058 | ||
1059 | while (cfqq->next_rq) { | |
1060 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1061 | dispatched++; | |
1062 | } | |
1063 | ||
1064 | BUG_ON(!list_empty(&cfqq->fifo)); | |
1065 | return dispatched; | |
1066 | } | |
1067 | ||
498d3aa2 JA |
1068 | /* |
1069 | * Drain our current requests. Used for barriers and when switching | |
1070 | * io schedulers on-the-fly. | |
1071 | */ | |
d9e7620e | 1072 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 1073 | { |
0871714e | 1074 | struct cfq_queue *cfqq; |
d9e7620e | 1075 | int dispatched = 0; |
1b5ed5e1 | 1076 | |
0871714e | 1077 | while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL) |
d9e7620e | 1078 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
1b5ed5e1 | 1079 | |
6084cdda | 1080 | cfq_slice_expired(cfqd, 0); |
1b5ed5e1 TH |
1081 | |
1082 | BUG_ON(cfqd->busy_queues); | |
1083 | ||
7b679138 | 1084 | cfq_log(cfqd, "forced_dispatch=%d\n", dispatched); |
1b5ed5e1 TH |
1085 | return dispatched; |
1086 | } | |
1087 | ||
2f5cb738 JA |
1088 | /* |
1089 | * Dispatch a request from cfqq, moving them to the request queue | |
1090 | * dispatch list. | |
1091 | */ | |
1092 | static void cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1093 | { | |
1094 | struct request *rq; | |
1095 | ||
1096 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
1097 | ||
1098 | /* | |
1099 | * follow expired path, else get first next available | |
1100 | */ | |
1101 | rq = cfq_check_fifo(cfqq); | |
1102 | if (!rq) | |
1103 | rq = cfqq->next_rq; | |
1104 | ||
1105 | /* | |
1106 | * insert request into driver dispatch list | |
1107 | */ | |
1108 | cfq_dispatch_insert(cfqd->queue, rq); | |
1109 | ||
1110 | if (!cfqd->active_cic) { | |
1111 | struct cfq_io_context *cic = RQ_CIC(rq); | |
1112 | ||
1113 | atomic_inc(&cic->ioc->refcount); | |
1114 | cfqd->active_cic = cic; | |
1115 | } | |
1116 | } | |
1117 | ||
1118 | /* | |
1119 | * Find the cfqq that we need to service and move a request from that to the | |
1120 | * dispatch list | |
1121 | */ | |
165125e1 | 1122 | static int cfq_dispatch_requests(struct request_queue *q, int force) |
22e2c507 JA |
1123 | { |
1124 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
6d048f53 | 1125 | struct cfq_queue *cfqq; |
2f5cb738 | 1126 | unsigned int max_dispatch; |
22e2c507 JA |
1127 | |
1128 | if (!cfqd->busy_queues) | |
1129 | return 0; | |
1130 | ||
1b5ed5e1 TH |
1131 | if (unlikely(force)) |
1132 | return cfq_forced_dispatch(cfqd); | |
1133 | ||
2f5cb738 JA |
1134 | cfqq = cfq_select_queue(cfqd); |
1135 | if (!cfqq) | |
1136 | return 0; | |
1137 | ||
1138 | /* | |
1139 | * If this is an async queue and we have sync IO in flight, let it wait | |
1140 | */ | |
1141 | if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq)) | |
1142 | return 0; | |
1143 | ||
1144 | max_dispatch = cfqd->cfq_quantum; | |
1145 | if (cfq_class_idle(cfqq)) | |
1146 | max_dispatch = 1; | |
b4878f24 | 1147 | |
2f5cb738 JA |
1148 | /* |
1149 | * Does this cfqq already have too much IO in flight? | |
1150 | */ | |
1151 | if (cfqq->dispatched >= max_dispatch) { | |
1152 | /* | |
1153 | * idle queue must always only have a single IO in flight | |
1154 | */ | |
3ed9a296 | 1155 | if (cfq_class_idle(cfqq)) |
2f5cb738 | 1156 | return 0; |
3ed9a296 | 1157 | |
2f5cb738 JA |
1158 | /* |
1159 | * We have other queues, don't allow more IO from this one | |
1160 | */ | |
1161 | if (cfqd->busy_queues > 1) | |
1162 | return 0; | |
9ede209e | 1163 | |
2f5cb738 JA |
1164 | /* |
1165 | * we are the only queue, allow up to 4 times of 'quantum' | |
1166 | */ | |
1167 | if (cfqq->dispatched >= 4 * max_dispatch) | |
1168 | return 0; | |
1169 | } | |
3ed9a296 | 1170 | |
2f5cb738 JA |
1171 | /* |
1172 | * Dispatch a request from this cfqq | |
1173 | */ | |
1174 | cfq_dispatch_request(cfqd, cfqq); | |
1175 | cfqq->slice_dispatch++; | |
22e2c507 | 1176 | |
2f5cb738 JA |
1177 | /* |
1178 | * expire an async queue immediately if it has used up its slice. idle | |
1179 | * queue always expire after 1 dispatch round. | |
1180 | */ | |
1181 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
1182 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
1183 | cfq_class_idle(cfqq))) { | |
1184 | cfqq->slice_end = jiffies + 1; | |
1185 | cfq_slice_expired(cfqd, 0); | |
1da177e4 LT |
1186 | } |
1187 | ||
2f5cb738 JA |
1188 | cfq_log(cfqd, "dispatched a request"); |
1189 | return 1; | |
1da177e4 LT |
1190 | } |
1191 | ||
1da177e4 | 1192 | /* |
5e705374 JA |
1193 | * task holds one reference to the queue, dropped when task exits. each rq |
1194 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1195 | * |
1196 | * queue lock must be held here. | |
1197 | */ | |
1198 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1199 | { | |
22e2c507 JA |
1200 | struct cfq_data *cfqd = cfqq->cfqd; |
1201 | ||
1202 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1203 | |
1204 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1205 | return; | |
1206 | ||
7b679138 | 1207 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 1208 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 1209 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1210 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1211 | |
28f95cbc | 1212 | if (unlikely(cfqd->active_queue == cfqq)) { |
6084cdda | 1213 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1214 | cfq_schedule_dispatch(cfqd); |
1215 | } | |
22e2c507 | 1216 | |
1da177e4 LT |
1217 | kmem_cache_free(cfq_pool, cfqq); |
1218 | } | |
1219 | ||
d6de8be7 JA |
1220 | /* |
1221 | * Must always be called with the rcu_read_lock() held | |
1222 | */ | |
07416d29 JA |
1223 | static void |
1224 | __call_for_each_cic(struct io_context *ioc, | |
1225 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1226 | { | |
1227 | struct cfq_io_context *cic; | |
1228 | struct hlist_node *n; | |
1229 | ||
1230 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) | |
1231 | func(ioc, cic); | |
1232 | } | |
1233 | ||
4ac845a2 | 1234 | /* |
34e6bbf2 | 1235 | * Call func for each cic attached to this ioc. |
4ac845a2 | 1236 | */ |
34e6bbf2 | 1237 | static void |
4ac845a2 JA |
1238 | call_for_each_cic(struct io_context *ioc, |
1239 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1da177e4 | 1240 | { |
4ac845a2 | 1241 | rcu_read_lock(); |
07416d29 | 1242 | __call_for_each_cic(ioc, func); |
4ac845a2 | 1243 | rcu_read_unlock(); |
34e6bbf2 FC |
1244 | } |
1245 | ||
1246 | static void cfq_cic_free_rcu(struct rcu_head *head) | |
1247 | { | |
1248 | struct cfq_io_context *cic; | |
1249 | ||
1250 | cic = container_of(head, struct cfq_io_context, rcu_head); | |
1251 | ||
1252 | kmem_cache_free(cfq_ioc_pool, cic); | |
1253 | elv_ioc_count_dec(ioc_count); | |
1254 | ||
9a11b4ed JA |
1255 | if (ioc_gone) { |
1256 | /* | |
1257 | * CFQ scheduler is exiting, grab exit lock and check | |
1258 | * the pending io context count. If it hits zero, | |
1259 | * complete ioc_gone and set it back to NULL | |
1260 | */ | |
1261 | spin_lock(&ioc_gone_lock); | |
1262 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) { | |
1263 | complete(ioc_gone); | |
1264 | ioc_gone = NULL; | |
1265 | } | |
1266 | spin_unlock(&ioc_gone_lock); | |
1267 | } | |
34e6bbf2 | 1268 | } |
4ac845a2 | 1269 | |
34e6bbf2 FC |
1270 | static void cfq_cic_free(struct cfq_io_context *cic) |
1271 | { | |
1272 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | |
4ac845a2 JA |
1273 | } |
1274 | ||
1275 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) | |
1276 | { | |
1277 | unsigned long flags; | |
1278 | ||
1279 | BUG_ON(!cic->dead_key); | |
1280 | ||
1281 | spin_lock_irqsave(&ioc->lock, flags); | |
1282 | radix_tree_delete(&ioc->radix_root, cic->dead_key); | |
ffc4e759 | 1283 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1284 | spin_unlock_irqrestore(&ioc->lock, flags); |
1285 | ||
34e6bbf2 | 1286 | cfq_cic_free(cic); |
4ac845a2 JA |
1287 | } |
1288 | ||
d6de8be7 JA |
1289 | /* |
1290 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. | |
1291 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), | |
1292 | * and ->trim() which is called with the task lock held | |
1293 | */ | |
4ac845a2 JA |
1294 | static void cfq_free_io_context(struct io_context *ioc) |
1295 | { | |
4ac845a2 | 1296 | /* |
34e6bbf2 FC |
1297 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
1298 | * so no more cic's are allowed to be linked into this ioc. So it | |
1299 | * should be ok to iterate over the known list, we will see all cic's | |
1300 | * since no new ones are added. | |
4ac845a2 | 1301 | */ |
07416d29 | 1302 | __call_for_each_cic(ioc, cic_free_func); |
1da177e4 LT |
1303 | } |
1304 | ||
89850f7e | 1305 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1306 | { |
28f95cbc | 1307 | if (unlikely(cfqq == cfqd->active_queue)) { |
6084cdda | 1308 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1309 | cfq_schedule_dispatch(cfqd); |
1310 | } | |
22e2c507 | 1311 | |
89850f7e JA |
1312 | cfq_put_queue(cfqq); |
1313 | } | |
22e2c507 | 1314 | |
89850f7e JA |
1315 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1316 | struct cfq_io_context *cic) | |
1317 | { | |
4faa3c81 FC |
1318 | struct io_context *ioc = cic->ioc; |
1319 | ||
fc46379d | 1320 | list_del_init(&cic->queue_list); |
4ac845a2 JA |
1321 | |
1322 | /* | |
1323 | * Make sure key == NULL is seen for dead queues | |
1324 | */ | |
fc46379d | 1325 | smp_wmb(); |
4ac845a2 | 1326 | cic->dead_key = (unsigned long) cic->key; |
fc46379d JA |
1327 | cic->key = NULL; |
1328 | ||
4faa3c81 FC |
1329 | if (ioc->ioc_data == cic) |
1330 | rcu_assign_pointer(ioc->ioc_data, NULL); | |
1331 | ||
12a05732 | 1332 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1333 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1334 | cic->cfqq[ASYNC] = NULL; |
1335 | } | |
1336 | ||
1337 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1338 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1339 | cic->cfqq[SYNC] = NULL; |
1340 | } | |
89850f7e JA |
1341 | } |
1342 | ||
4ac845a2 JA |
1343 | static void cfq_exit_single_io_context(struct io_context *ioc, |
1344 | struct cfq_io_context *cic) | |
89850f7e JA |
1345 | { |
1346 | struct cfq_data *cfqd = cic->key; | |
1347 | ||
89850f7e | 1348 | if (cfqd) { |
165125e1 | 1349 | struct request_queue *q = cfqd->queue; |
4ac845a2 | 1350 | unsigned long flags; |
89850f7e | 1351 | |
4ac845a2 | 1352 | spin_lock_irqsave(q->queue_lock, flags); |
62c1fe9d JA |
1353 | |
1354 | /* | |
1355 | * Ensure we get a fresh copy of the ->key to prevent | |
1356 | * race between exiting task and queue | |
1357 | */ | |
1358 | smp_read_barrier_depends(); | |
1359 | if (cic->key) | |
1360 | __cfq_exit_single_io_context(cfqd, cic); | |
1361 | ||
4ac845a2 | 1362 | spin_unlock_irqrestore(q->queue_lock, flags); |
89850f7e | 1363 | } |
1da177e4 LT |
1364 | } |
1365 | ||
498d3aa2 JA |
1366 | /* |
1367 | * The process that ioc belongs to has exited, we need to clean up | |
1368 | * and put the internal structures we have that belongs to that process. | |
1369 | */ | |
e2d74ac0 | 1370 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1371 | { |
4ac845a2 | 1372 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
1da177e4 LT |
1373 | } |
1374 | ||
22e2c507 | 1375 | static struct cfq_io_context * |
8267e268 | 1376 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1377 | { |
b5deef90 | 1378 | struct cfq_io_context *cic; |
1da177e4 | 1379 | |
94f6030c CL |
1380 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
1381 | cfqd->queue->node); | |
1da177e4 | 1382 | if (cic) { |
22e2c507 | 1383 | cic->last_end_request = jiffies; |
553698f9 | 1384 | INIT_LIST_HEAD(&cic->queue_list); |
ffc4e759 | 1385 | INIT_HLIST_NODE(&cic->cic_list); |
22e2c507 JA |
1386 | cic->dtor = cfq_free_io_context; |
1387 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1388 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1389 | } |
1390 | ||
1391 | return cic; | |
1392 | } | |
1393 | ||
fd0928df | 1394 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
1395 | { |
1396 | struct task_struct *tsk = current; | |
1397 | int ioprio_class; | |
1398 | ||
3b18152c | 1399 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1400 | return; |
1401 | ||
fd0928df | 1402 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 1403 | switch (ioprio_class) { |
fe094d98 JA |
1404 | default: |
1405 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1406 | case IOPRIO_CLASS_NONE: | |
1407 | /* | |
6d63c275 | 1408 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
1409 | */ |
1410 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 1411 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
1412 | break; |
1413 | case IOPRIO_CLASS_RT: | |
1414 | cfqq->ioprio = task_ioprio(ioc); | |
1415 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1416 | break; | |
1417 | case IOPRIO_CLASS_BE: | |
1418 | cfqq->ioprio = task_ioprio(ioc); | |
1419 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1420 | break; | |
1421 | case IOPRIO_CLASS_IDLE: | |
1422 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1423 | cfqq->ioprio = 7; | |
1424 | cfq_clear_cfqq_idle_window(cfqq); | |
1425 | break; | |
22e2c507 JA |
1426 | } |
1427 | ||
1428 | /* | |
1429 | * keep track of original prio settings in case we have to temporarily | |
1430 | * elevate the priority of this queue | |
1431 | */ | |
1432 | cfqq->org_ioprio = cfqq->ioprio; | |
1433 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 1434 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1435 | } |
1436 | ||
febffd61 | 1437 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
22e2c507 | 1438 | { |
478a82b0 AV |
1439 | struct cfq_data *cfqd = cic->key; |
1440 | struct cfq_queue *cfqq; | |
c1b707d2 | 1441 | unsigned long flags; |
35e6077c | 1442 | |
caaa5f9f JA |
1443 | if (unlikely(!cfqd)) |
1444 | return; | |
1445 | ||
c1b707d2 | 1446 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1447 | |
1448 | cfqq = cic->cfqq[ASYNC]; | |
1449 | if (cfqq) { | |
1450 | struct cfq_queue *new_cfqq; | |
fd0928df | 1451 | new_cfqq = cfq_get_queue(cfqd, ASYNC, cic->ioc, GFP_ATOMIC); |
caaa5f9f JA |
1452 | if (new_cfqq) { |
1453 | cic->cfqq[ASYNC] = new_cfqq; | |
1454 | cfq_put_queue(cfqq); | |
1455 | } | |
22e2c507 | 1456 | } |
caaa5f9f JA |
1457 | |
1458 | cfqq = cic->cfqq[SYNC]; | |
1459 | if (cfqq) | |
1460 | cfq_mark_cfqq_prio_changed(cfqq); | |
1461 | ||
c1b707d2 | 1462 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1463 | } |
1464 | ||
fc46379d | 1465 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1466 | { |
4ac845a2 | 1467 | call_for_each_cic(ioc, changed_ioprio); |
fc46379d | 1468 | ioc->ioprio_changed = 0; |
22e2c507 JA |
1469 | } |
1470 | ||
1471 | static struct cfq_queue * | |
15c31be4 | 1472 | cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync, |
fd0928df | 1473 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 1474 | { |
22e2c507 | 1475 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 1476 | struct cfq_io_context *cic; |
22e2c507 JA |
1477 | |
1478 | retry: | |
4ac845a2 | 1479 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
1480 | /* cic always exists here */ |
1481 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 JA |
1482 | |
1483 | if (!cfqq) { | |
1484 | if (new_cfqq) { | |
1485 | cfqq = new_cfqq; | |
1486 | new_cfqq = NULL; | |
1487 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1488 | /* |
1489 | * Inform the allocator of the fact that we will | |
1490 | * just repeat this allocation if it fails, to allow | |
1491 | * the allocator to do whatever it needs to attempt to | |
1492 | * free memory. | |
1493 | */ | |
22e2c507 | 1494 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c CL |
1495 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
1496 | gfp_mask | __GFP_NOFAIL | __GFP_ZERO, | |
1497 | cfqd->queue->node); | |
22e2c507 JA |
1498 | spin_lock_irq(cfqd->queue->queue_lock); |
1499 | goto retry; | |
1500 | } else { | |
94f6030c CL |
1501 | cfqq = kmem_cache_alloc_node(cfq_pool, |
1502 | gfp_mask | __GFP_ZERO, | |
1503 | cfqd->queue->node); | |
22e2c507 JA |
1504 | if (!cfqq) |
1505 | goto out; | |
1506 | } | |
1507 | ||
d9e7620e | 1508 | RB_CLEAR_NODE(&cfqq->rb_node); |
22e2c507 JA |
1509 | INIT_LIST_HEAD(&cfqq->fifo); |
1510 | ||
22e2c507 JA |
1511 | atomic_set(&cfqq->ref, 0); |
1512 | cfqq->cfqd = cfqd; | |
c5b680f3 | 1513 | |
3b18152c | 1514 | cfq_mark_cfqq_prio_changed(cfqq); |
91fac317 | 1515 | |
fd0928df | 1516 | cfq_init_prio_data(cfqq, ioc); |
0871714e JA |
1517 | |
1518 | if (is_sync) { | |
1519 | if (!cfq_class_idle(cfqq)) | |
1520 | cfq_mark_cfqq_idle_window(cfqq); | |
1521 | cfq_mark_cfqq_sync(cfqq); | |
1522 | } | |
7b679138 JA |
1523 | cfqq->pid = current->pid; |
1524 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | |
22e2c507 JA |
1525 | } |
1526 | ||
1527 | if (new_cfqq) | |
1528 | kmem_cache_free(cfq_pool, new_cfqq); | |
1529 | ||
22e2c507 JA |
1530 | out: |
1531 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1532 | return cfqq; | |
1533 | } | |
1534 | ||
c2dea2d1 VT |
1535 | static struct cfq_queue ** |
1536 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
1537 | { | |
fe094d98 | 1538 | switch (ioprio_class) { |
c2dea2d1 VT |
1539 | case IOPRIO_CLASS_RT: |
1540 | return &cfqd->async_cfqq[0][ioprio]; | |
1541 | case IOPRIO_CLASS_BE: | |
1542 | return &cfqd->async_cfqq[1][ioprio]; | |
1543 | case IOPRIO_CLASS_IDLE: | |
1544 | return &cfqd->async_idle_cfqq; | |
1545 | default: | |
1546 | BUG(); | |
1547 | } | |
1548 | } | |
1549 | ||
15c31be4 | 1550 | static struct cfq_queue * |
fd0928df | 1551 | cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc, |
15c31be4 JA |
1552 | gfp_t gfp_mask) |
1553 | { | |
fd0928df JA |
1554 | const int ioprio = task_ioprio(ioc); |
1555 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 1556 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
1557 | struct cfq_queue *cfqq = NULL; |
1558 | ||
c2dea2d1 VT |
1559 | if (!is_sync) { |
1560 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
1561 | cfqq = *async_cfqq; | |
1562 | } | |
1563 | ||
0a0836a0 | 1564 | if (!cfqq) { |
fd0928df | 1565 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
0a0836a0 ON |
1566 | if (!cfqq) |
1567 | return NULL; | |
1568 | } | |
15c31be4 JA |
1569 | |
1570 | /* | |
1571 | * pin the queue now that it's allocated, scheduler exit will prune it | |
1572 | */ | |
c2dea2d1 | 1573 | if (!is_sync && !(*async_cfqq)) { |
15c31be4 | 1574 | atomic_inc(&cfqq->ref); |
c2dea2d1 | 1575 | *async_cfqq = cfqq; |
15c31be4 JA |
1576 | } |
1577 | ||
1578 | atomic_inc(&cfqq->ref); | |
1579 | return cfqq; | |
1580 | } | |
1581 | ||
498d3aa2 JA |
1582 | /* |
1583 | * We drop cfq io contexts lazily, so we may find a dead one. | |
1584 | */ | |
dbecf3ab | 1585 | static void |
4ac845a2 JA |
1586 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
1587 | struct cfq_io_context *cic) | |
dbecf3ab | 1588 | { |
4ac845a2 JA |
1589 | unsigned long flags; |
1590 | ||
fc46379d | 1591 | WARN_ON(!list_empty(&cic->queue_list)); |
597bc485 | 1592 | |
4ac845a2 JA |
1593 | spin_lock_irqsave(&ioc->lock, flags); |
1594 | ||
4faa3c81 | 1595 | BUG_ON(ioc->ioc_data == cic); |
597bc485 | 1596 | |
4ac845a2 | 1597 | radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd); |
ffc4e759 | 1598 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1599 | spin_unlock_irqrestore(&ioc->lock, flags); |
1600 | ||
1601 | cfq_cic_free(cic); | |
dbecf3ab OH |
1602 | } |
1603 | ||
e2d74ac0 | 1604 | static struct cfq_io_context * |
4ac845a2 | 1605 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
e2d74ac0 | 1606 | { |
e2d74ac0 | 1607 | struct cfq_io_context *cic; |
d6de8be7 | 1608 | unsigned long flags; |
4ac845a2 | 1609 | void *k; |
e2d74ac0 | 1610 | |
91fac317 VT |
1611 | if (unlikely(!ioc)) |
1612 | return NULL; | |
1613 | ||
d6de8be7 JA |
1614 | rcu_read_lock(); |
1615 | ||
597bc485 JA |
1616 | /* |
1617 | * we maintain a last-hit cache, to avoid browsing over the tree | |
1618 | */ | |
4ac845a2 | 1619 | cic = rcu_dereference(ioc->ioc_data); |
d6de8be7 JA |
1620 | if (cic && cic->key == cfqd) { |
1621 | rcu_read_unlock(); | |
597bc485 | 1622 | return cic; |
d6de8be7 | 1623 | } |
597bc485 | 1624 | |
4ac845a2 | 1625 | do { |
4ac845a2 JA |
1626 | cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd); |
1627 | rcu_read_unlock(); | |
1628 | if (!cic) | |
1629 | break; | |
be3b0753 OH |
1630 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1631 | k = cic->key; | |
1632 | if (unlikely(!k)) { | |
4ac845a2 | 1633 | cfq_drop_dead_cic(cfqd, ioc, cic); |
d6de8be7 | 1634 | rcu_read_lock(); |
4ac845a2 | 1635 | continue; |
dbecf3ab | 1636 | } |
e2d74ac0 | 1637 | |
d6de8be7 | 1638 | spin_lock_irqsave(&ioc->lock, flags); |
4ac845a2 | 1639 | rcu_assign_pointer(ioc->ioc_data, cic); |
d6de8be7 | 1640 | spin_unlock_irqrestore(&ioc->lock, flags); |
4ac845a2 JA |
1641 | break; |
1642 | } while (1); | |
e2d74ac0 | 1643 | |
4ac845a2 | 1644 | return cic; |
e2d74ac0 JA |
1645 | } |
1646 | ||
4ac845a2 JA |
1647 | /* |
1648 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup | |
1649 | * the process specific cfq io context when entered from the block layer. | |
1650 | * Also adds the cic to a per-cfqd list, used when this queue is removed. | |
1651 | */ | |
febffd61 JA |
1652 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
1653 | struct cfq_io_context *cic, gfp_t gfp_mask) | |
e2d74ac0 | 1654 | { |
0261d688 | 1655 | unsigned long flags; |
4ac845a2 | 1656 | int ret; |
e2d74ac0 | 1657 | |
4ac845a2 JA |
1658 | ret = radix_tree_preload(gfp_mask); |
1659 | if (!ret) { | |
1660 | cic->ioc = ioc; | |
1661 | cic->key = cfqd; | |
e2d74ac0 | 1662 | |
4ac845a2 JA |
1663 | spin_lock_irqsave(&ioc->lock, flags); |
1664 | ret = radix_tree_insert(&ioc->radix_root, | |
1665 | (unsigned long) cfqd, cic); | |
ffc4e759 JA |
1666 | if (!ret) |
1667 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); | |
4ac845a2 | 1668 | spin_unlock_irqrestore(&ioc->lock, flags); |
e2d74ac0 | 1669 | |
4ac845a2 JA |
1670 | radix_tree_preload_end(); |
1671 | ||
1672 | if (!ret) { | |
1673 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1674 | list_add(&cic->queue_list, &cfqd->cic_list); | |
1675 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1676 | } | |
e2d74ac0 JA |
1677 | } |
1678 | ||
4ac845a2 JA |
1679 | if (ret) |
1680 | printk(KERN_ERR "cfq: cic link failed!\n"); | |
fc46379d | 1681 | |
4ac845a2 | 1682 | return ret; |
e2d74ac0 JA |
1683 | } |
1684 | ||
1da177e4 LT |
1685 | /* |
1686 | * Setup general io context and cfq io context. There can be several cfq | |
1687 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1688 | * than one device managed by cfq. |
1da177e4 LT |
1689 | */ |
1690 | static struct cfq_io_context * | |
e2d74ac0 | 1691 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1692 | { |
22e2c507 | 1693 | struct io_context *ioc = NULL; |
1da177e4 | 1694 | struct cfq_io_context *cic; |
1da177e4 | 1695 | |
22e2c507 | 1696 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1697 | |
b5deef90 | 1698 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1699 | if (!ioc) |
1700 | return NULL; | |
1701 | ||
4ac845a2 | 1702 | cic = cfq_cic_lookup(cfqd, ioc); |
e2d74ac0 JA |
1703 | if (cic) |
1704 | goto out; | |
1da177e4 | 1705 | |
e2d74ac0 JA |
1706 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1707 | if (cic == NULL) | |
1708 | goto err; | |
1da177e4 | 1709 | |
4ac845a2 JA |
1710 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) |
1711 | goto err_free; | |
1712 | ||
1da177e4 | 1713 | out: |
fc46379d JA |
1714 | smp_read_barrier_depends(); |
1715 | if (unlikely(ioc->ioprio_changed)) | |
1716 | cfq_ioc_set_ioprio(ioc); | |
1717 | ||
1da177e4 | 1718 | return cic; |
4ac845a2 JA |
1719 | err_free: |
1720 | cfq_cic_free(cic); | |
1da177e4 LT |
1721 | err: |
1722 | put_io_context(ioc); | |
1723 | return NULL; | |
1724 | } | |
1725 | ||
22e2c507 JA |
1726 | static void |
1727 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1728 | { |
aaf1228d JA |
1729 | unsigned long elapsed = jiffies - cic->last_end_request; |
1730 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 1731 | |
22e2c507 JA |
1732 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1733 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1734 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1735 | } | |
1da177e4 | 1736 | |
206dc69b | 1737 | static void |
6d048f53 JA |
1738 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic, |
1739 | struct request *rq) | |
206dc69b JA |
1740 | { |
1741 | sector_t sdist; | |
1742 | u64 total; | |
1743 | ||
5e705374 JA |
1744 | if (cic->last_request_pos < rq->sector) |
1745 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1746 | else |
5e705374 | 1747 | sdist = cic->last_request_pos - rq->sector; |
206dc69b JA |
1748 | |
1749 | /* | |
1750 | * Don't allow the seek distance to get too large from the | |
1751 | * odd fragment, pagein, etc | |
1752 | */ | |
1753 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1754 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1755 | else | |
1756 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1757 | ||
1758 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1759 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1760 | total = cic->seek_total + (cic->seek_samples/2); | |
1761 | do_div(total, cic->seek_samples); | |
1762 | cic->seek_mean = (sector_t)total; | |
1763 | } | |
1da177e4 | 1764 | |
22e2c507 JA |
1765 | /* |
1766 | * Disable idle window if the process thinks too long or seeks so much that | |
1767 | * it doesn't matter | |
1768 | */ | |
1769 | static void | |
1770 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1771 | struct cfq_io_context *cic) | |
1772 | { | |
7b679138 | 1773 | int old_idle, enable_idle; |
1be92f2f | 1774 | |
0871714e JA |
1775 | /* |
1776 | * Don't idle for async or idle io prio class | |
1777 | */ | |
1778 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
1779 | return; |
1780 | ||
c265a7f4 | 1781 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 1782 | |
66dac98e | 1783 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
caaa5f9f | 1784 | (cfqd->hw_tag && CIC_SEEKY(cic))) |
22e2c507 JA |
1785 | enable_idle = 0; |
1786 | else if (sample_valid(cic->ttime_samples)) { | |
1787 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1788 | enable_idle = 0; | |
1789 | else | |
1790 | enable_idle = 1; | |
1da177e4 LT |
1791 | } |
1792 | ||
7b679138 JA |
1793 | if (old_idle != enable_idle) { |
1794 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
1795 | if (enable_idle) | |
1796 | cfq_mark_cfqq_idle_window(cfqq); | |
1797 | else | |
1798 | cfq_clear_cfqq_idle_window(cfqq); | |
1799 | } | |
22e2c507 | 1800 | } |
1da177e4 | 1801 | |
22e2c507 JA |
1802 | /* |
1803 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1804 | * no or if we aren't sure, a 1 will cause a preempt. | |
1805 | */ | |
1806 | static int | |
1807 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1808 | struct request *rq) |
22e2c507 | 1809 | { |
6d048f53 | 1810 | struct cfq_queue *cfqq; |
22e2c507 | 1811 | |
6d048f53 JA |
1812 | cfqq = cfqd->active_queue; |
1813 | if (!cfqq) | |
22e2c507 JA |
1814 | return 0; |
1815 | ||
6d048f53 JA |
1816 | if (cfq_slice_used(cfqq)) |
1817 | return 1; | |
1818 | ||
1819 | if (cfq_class_idle(new_cfqq)) | |
caaa5f9f | 1820 | return 0; |
22e2c507 JA |
1821 | |
1822 | if (cfq_class_idle(cfqq)) | |
1823 | return 1; | |
1e3335de | 1824 | |
374f84ac JA |
1825 | /* |
1826 | * if the new request is sync, but the currently running queue is | |
1827 | * not, let the sync request have priority. | |
1828 | */ | |
5e705374 | 1829 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1830 | return 1; |
1e3335de | 1831 | |
374f84ac JA |
1832 | /* |
1833 | * So both queues are sync. Let the new request get disk time if | |
1834 | * it's a metadata request and the current queue is doing regular IO. | |
1835 | */ | |
1836 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1837 | return 1; | |
22e2c507 | 1838 | |
3a9a3f6c DS |
1839 | /* |
1840 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
1841 | */ | |
1842 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
1843 | return 1; | |
1844 | ||
1e3335de JA |
1845 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
1846 | return 0; | |
1847 | ||
1848 | /* | |
1849 | * if this request is as-good as one we would expect from the | |
1850 | * current cfqq, let it preempt | |
1851 | */ | |
6d048f53 | 1852 | if (cfq_rq_close(cfqd, rq)) |
1e3335de JA |
1853 | return 1; |
1854 | ||
22e2c507 JA |
1855 | return 0; |
1856 | } | |
1857 | ||
1858 | /* | |
1859 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1860 | * let it have half of its nominal slice. | |
1861 | */ | |
1862 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1863 | { | |
7b679138 | 1864 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
6084cdda | 1865 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 1866 | |
bf572256 JA |
1867 | /* |
1868 | * Put the new queue at the front of the of the current list, | |
1869 | * so we know that it will be selected next. | |
1870 | */ | |
1871 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
1872 | |
1873 | cfq_service_tree_add(cfqd, cfqq, 1); | |
bf572256 | 1874 | |
44f7c160 JA |
1875 | cfqq->slice_end = 0; |
1876 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
1877 | } |
1878 | ||
22e2c507 | 1879 | /* |
5e705374 | 1880 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1881 | * something we should do about it |
1882 | */ | |
1883 | static void | |
5e705374 JA |
1884 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1885 | struct request *rq) | |
22e2c507 | 1886 | { |
5e705374 | 1887 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1888 | |
45333d5a | 1889 | cfqd->rq_queued++; |
374f84ac JA |
1890 | if (rq_is_meta(rq)) |
1891 | cfqq->meta_pending++; | |
1892 | ||
9c2c38a1 | 1893 | cfq_update_io_thinktime(cfqd, cic); |
6d048f53 | 1894 | cfq_update_io_seektime(cfqd, cic, rq); |
9c2c38a1 JA |
1895 | cfq_update_idle_window(cfqd, cfqq, cic); |
1896 | ||
5e705374 | 1897 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
22e2c507 JA |
1898 | |
1899 | if (cfqq == cfqd->active_queue) { | |
1900 | /* | |
1901 | * if we are waiting for a request for this queue, let it rip | |
1902 | * immediately and flag that we must not expire this queue | |
1903 | * just now | |
1904 | */ | |
3b18152c | 1905 | if (cfq_cfqq_wait_request(cfqq)) { |
22e2c507 | 1906 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1907 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1908 | } |
5e705374 | 1909 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1910 | /* |
1911 | * not the active queue - expire current slice if it is | |
1912 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
1913 | * has some old slice time left and is of higher priority or |
1914 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
1915 | */ |
1916 | cfq_preempt_queue(cfqd, cfqq); | |
dc72ef4a | 1917 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1918 | } |
1da177e4 LT |
1919 | } |
1920 | ||
165125e1 | 1921 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1922 | { |
b4878f24 | 1923 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1924 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1925 | |
7b679138 | 1926 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
fd0928df | 1927 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
1da177e4 | 1928 | |
5e705374 | 1929 | cfq_add_rq_rb(rq); |
1da177e4 | 1930 | |
22e2c507 JA |
1931 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1932 | ||
5e705374 | 1933 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1934 | } |
1935 | ||
45333d5a AC |
1936 | /* |
1937 | * Update hw_tag based on peak queue depth over 50 samples under | |
1938 | * sufficient load. | |
1939 | */ | |
1940 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
1941 | { | |
1942 | if (cfqd->rq_in_driver > cfqd->rq_in_driver_peak) | |
1943 | cfqd->rq_in_driver_peak = cfqd->rq_in_driver; | |
1944 | ||
1945 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
1946 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) | |
1947 | return; | |
1948 | ||
1949 | if (cfqd->hw_tag_samples++ < 50) | |
1950 | return; | |
1951 | ||
1952 | if (cfqd->rq_in_driver_peak >= CFQ_HW_QUEUE_MIN) | |
1953 | cfqd->hw_tag = 1; | |
1954 | else | |
1955 | cfqd->hw_tag = 0; | |
1956 | ||
1957 | cfqd->hw_tag_samples = 0; | |
1958 | cfqd->rq_in_driver_peak = 0; | |
1959 | } | |
1960 | ||
165125e1 | 1961 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1962 | { |
5e705374 | 1963 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1964 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1965 | const int sync = rq_is_sync(rq); |
b4878f24 | 1966 | unsigned long now; |
1da177e4 | 1967 | |
b4878f24 | 1968 | now = jiffies; |
7b679138 | 1969 | cfq_log_cfqq(cfqd, cfqq, "complete"); |
1da177e4 | 1970 | |
45333d5a AC |
1971 | cfq_update_hw_tag(cfqd); |
1972 | ||
b4878f24 | 1973 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 1974 | WARN_ON(!cfqq->dispatched); |
b4878f24 | 1975 | cfqd->rq_in_driver--; |
6d048f53 | 1976 | cfqq->dispatched--; |
1da177e4 | 1977 | |
3ed9a296 JA |
1978 | if (cfq_cfqq_sync(cfqq)) |
1979 | cfqd->sync_flight--; | |
1980 | ||
b4878f24 JA |
1981 | if (!cfq_class_idle(cfqq)) |
1982 | cfqd->last_end_request = now; | |
3b18152c | 1983 | |
caaa5f9f | 1984 | if (sync) |
5e705374 | 1985 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1986 | |
1987 | /* | |
1988 | * If this is the active queue, check if it needs to be expired, | |
1989 | * or if we want to idle in case it has no pending requests. | |
1990 | */ | |
1991 | if (cfqd->active_queue == cfqq) { | |
44f7c160 JA |
1992 | if (cfq_cfqq_slice_new(cfqq)) { |
1993 | cfq_set_prio_slice(cfqd, cfqq); | |
1994 | cfq_clear_cfqq_slice_new(cfqq); | |
1995 | } | |
0871714e | 1996 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
6084cdda | 1997 | cfq_slice_expired(cfqd, 1); |
aeb6fafb JA |
1998 | else if (sync && !rq_noidle(rq) && |
1999 | RB_EMPTY_ROOT(&cfqq->sort_list)) { | |
6d048f53 | 2000 | cfq_arm_slice_timer(cfqd); |
aeb6fafb | 2001 | } |
caaa5f9f | 2002 | } |
6d048f53 JA |
2003 | |
2004 | if (!cfqd->rq_in_driver) | |
2005 | cfq_schedule_dispatch(cfqd); | |
1da177e4 LT |
2006 | } |
2007 | ||
22e2c507 JA |
2008 | /* |
2009 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
2010 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
2011 | */ | |
2012 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 2013 | { |
22e2c507 JA |
2014 | if (has_fs_excl()) { |
2015 | /* | |
2016 | * boost idle prio on transactions that would lock out other | |
2017 | * users of the filesystem | |
2018 | */ | |
2019 | if (cfq_class_idle(cfqq)) | |
2020 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
2021 | if (cfqq->ioprio > IOPRIO_NORM) | |
2022 | cfqq->ioprio = IOPRIO_NORM; | |
2023 | } else { | |
2024 | /* | |
2025 | * check if we need to unboost the queue | |
2026 | */ | |
2027 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
2028 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
2029 | if (cfqq->ioprio != cfqq->org_ioprio) | |
2030 | cfqq->ioprio = cfqq->org_ioprio; | |
2031 | } | |
22e2c507 | 2032 | } |
1da177e4 | 2033 | |
89850f7e | 2034 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 2035 | { |
3b18152c | 2036 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 2037 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 2038 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 2039 | return ELV_MQUEUE_MUST; |
3b18152c | 2040 | } |
1da177e4 | 2041 | |
22e2c507 | 2042 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
2043 | } |
2044 | ||
165125e1 | 2045 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
2046 | { |
2047 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2048 | struct task_struct *tsk = current; | |
91fac317 | 2049 | struct cfq_io_context *cic; |
22e2c507 JA |
2050 | struct cfq_queue *cfqq; |
2051 | ||
2052 | /* | |
2053 | * don't force setup of a queue from here, as a call to may_queue | |
2054 | * does not necessarily imply that a request actually will be queued. | |
2055 | * so just lookup a possibly existing queue, or return 'may queue' | |
2056 | * if that fails | |
2057 | */ | |
4ac845a2 | 2058 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2059 | if (!cic) |
2060 | return ELV_MQUEUE_MAY; | |
2061 | ||
2062 | cfqq = cic_to_cfqq(cic, rw & REQ_RW_SYNC); | |
22e2c507 | 2063 | if (cfqq) { |
fd0928df | 2064 | cfq_init_prio_data(cfqq, cic->ioc); |
22e2c507 JA |
2065 | cfq_prio_boost(cfqq); |
2066 | ||
89850f7e | 2067 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
2068 | } |
2069 | ||
2070 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
2071 | } |
2072 | ||
1da177e4 LT |
2073 | /* |
2074 | * queue lock held here | |
2075 | */ | |
bb37b94c | 2076 | static void cfq_put_request(struct request *rq) |
1da177e4 | 2077 | { |
5e705374 | 2078 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2079 | |
5e705374 | 2080 | if (cfqq) { |
22e2c507 | 2081 | const int rw = rq_data_dir(rq); |
1da177e4 | 2082 | |
22e2c507 JA |
2083 | BUG_ON(!cfqq->allocated[rw]); |
2084 | cfqq->allocated[rw]--; | |
1da177e4 | 2085 | |
5e705374 | 2086 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 2087 | |
1da177e4 | 2088 | rq->elevator_private = NULL; |
5e705374 | 2089 | rq->elevator_private2 = NULL; |
1da177e4 | 2090 | |
1da177e4 LT |
2091 | cfq_put_queue(cfqq); |
2092 | } | |
2093 | } | |
2094 | ||
2095 | /* | |
22e2c507 | 2096 | * Allocate cfq data structures associated with this request. |
1da177e4 | 2097 | */ |
22e2c507 | 2098 | static int |
165125e1 | 2099 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
2100 | { |
2101 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2102 | struct cfq_io_context *cic; | |
2103 | const int rw = rq_data_dir(rq); | |
7749a8d4 | 2104 | const int is_sync = rq_is_sync(rq); |
22e2c507 | 2105 | struct cfq_queue *cfqq; |
1da177e4 LT |
2106 | unsigned long flags; |
2107 | ||
2108 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2109 | ||
e2d74ac0 | 2110 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 2111 | |
1da177e4 LT |
2112 | spin_lock_irqsave(q->queue_lock, flags); |
2113 | ||
22e2c507 JA |
2114 | if (!cic) |
2115 | goto queue_fail; | |
2116 | ||
91fac317 VT |
2117 | cfqq = cic_to_cfqq(cic, is_sync); |
2118 | if (!cfqq) { | |
fd0928df | 2119 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
91fac317 | 2120 | |
22e2c507 JA |
2121 | if (!cfqq) |
2122 | goto queue_fail; | |
1da177e4 | 2123 | |
91fac317 VT |
2124 | cic_set_cfqq(cic, cfqq, is_sync); |
2125 | } | |
1da177e4 LT |
2126 | |
2127 | cfqq->allocated[rw]++; | |
3b18152c | 2128 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 2129 | atomic_inc(&cfqq->ref); |
1da177e4 | 2130 | |
5e705374 | 2131 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 2132 | |
5e705374 JA |
2133 | rq->elevator_private = cic; |
2134 | rq->elevator_private2 = cfqq; | |
2135 | return 0; | |
1da177e4 | 2136 | |
22e2c507 JA |
2137 | queue_fail: |
2138 | if (cic) | |
2139 | put_io_context(cic->ioc); | |
89850f7e | 2140 | |
3b18152c | 2141 | cfq_schedule_dispatch(cfqd); |
1da177e4 | 2142 | spin_unlock_irqrestore(q->queue_lock, flags); |
7b679138 | 2143 | cfq_log(cfqd, "set_request fail"); |
1da177e4 LT |
2144 | return 1; |
2145 | } | |
2146 | ||
65f27f38 | 2147 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 2148 | { |
65f27f38 DH |
2149 | struct cfq_data *cfqd = |
2150 | container_of(work, struct cfq_data, unplug_work); | |
165125e1 | 2151 | struct request_queue *q = cfqd->queue; |
22e2c507 JA |
2152 | unsigned long flags; |
2153 | ||
2154 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 2155 | blk_start_queueing(q); |
22e2c507 JA |
2156 | spin_unlock_irqrestore(q->queue_lock, flags); |
2157 | } | |
2158 | ||
2159 | /* | |
2160 | * Timer running if the active_queue is currently idling inside its time slice | |
2161 | */ | |
2162 | static void cfq_idle_slice_timer(unsigned long data) | |
2163 | { | |
2164 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
2165 | struct cfq_queue *cfqq; | |
2166 | unsigned long flags; | |
3c6bd2f8 | 2167 | int timed_out = 1; |
22e2c507 | 2168 | |
7b679138 JA |
2169 | cfq_log(cfqd, "idle timer fired"); |
2170 | ||
22e2c507 JA |
2171 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
2172 | ||
fe094d98 JA |
2173 | cfqq = cfqd->active_queue; |
2174 | if (cfqq) { | |
3c6bd2f8 JA |
2175 | timed_out = 0; |
2176 | ||
22e2c507 JA |
2177 | /* |
2178 | * expired | |
2179 | */ | |
44f7c160 | 2180 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
2181 | goto expire; |
2182 | ||
2183 | /* | |
2184 | * only expire and reinvoke request handler, if there are | |
2185 | * other queues with pending requests | |
2186 | */ | |
caaa5f9f | 2187 | if (!cfqd->busy_queues) |
22e2c507 | 2188 | goto out_cont; |
22e2c507 JA |
2189 | |
2190 | /* | |
2191 | * not expired and it has a request pending, let it dispatch | |
2192 | */ | |
75e50984 | 2193 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2194 | goto out_kick; |
22e2c507 JA |
2195 | } |
2196 | expire: | |
6084cdda | 2197 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 2198 | out_kick: |
3b18152c | 2199 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
2200 | out_cont: |
2201 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
2202 | } | |
2203 | ||
3b18152c JA |
2204 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
2205 | { | |
2206 | del_timer_sync(&cfqd->idle_slice_timer); | |
64d01dc9 | 2207 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 2208 | } |
22e2c507 | 2209 | |
c2dea2d1 VT |
2210 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
2211 | { | |
2212 | int i; | |
2213 | ||
2214 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
2215 | if (cfqd->async_cfqq[0][i]) | |
2216 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
2217 | if (cfqd->async_cfqq[1][i]) | |
2218 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 2219 | } |
2389d1ef ON |
2220 | |
2221 | if (cfqd->async_idle_cfqq) | |
2222 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
2223 | } |
2224 | ||
b374d18a | 2225 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 2226 | { |
22e2c507 | 2227 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 2228 | struct request_queue *q = cfqd->queue; |
22e2c507 | 2229 | |
3b18152c | 2230 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 2231 | |
d9ff4187 | 2232 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 2233 | |
d9ff4187 | 2234 | if (cfqd->active_queue) |
6084cdda | 2235 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
2236 | |
2237 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
2238 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
2239 | struct cfq_io_context, | |
2240 | queue_list); | |
89850f7e JA |
2241 | |
2242 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 2243 | } |
e2d74ac0 | 2244 | |
c2dea2d1 | 2245 | cfq_put_async_queues(cfqd); |
15c31be4 | 2246 | |
d9ff4187 | 2247 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
2248 | |
2249 | cfq_shutdown_timer_wq(cfqd); | |
2250 | ||
a90d742e | 2251 | kfree(cfqd); |
1da177e4 LT |
2252 | } |
2253 | ||
165125e1 | 2254 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
2255 | { |
2256 | struct cfq_data *cfqd; | |
1da177e4 | 2257 | |
94f6030c | 2258 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
1da177e4 | 2259 | if (!cfqd) |
bc1c1169 | 2260 | return NULL; |
1da177e4 | 2261 | |
cc09e299 | 2262 | cfqd->service_tree = CFQ_RB_ROOT; |
d9ff4187 | 2263 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2264 | |
1da177e4 | 2265 | cfqd->queue = q; |
1da177e4 | 2266 | |
22e2c507 JA |
2267 | init_timer(&cfqd->idle_slice_timer); |
2268 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2269 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2270 | ||
65f27f38 | 2271 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2272 | |
b70c864d | 2273 | cfqd->last_end_request = jiffies; |
1da177e4 | 2274 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2275 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2276 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2277 | cfqd->cfq_back_max = cfq_back_max; |
2278 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2279 | cfqd->cfq_slice[0] = cfq_slice_async; |
2280 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2281 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2282 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
45333d5a | 2283 | cfqd->hw_tag = 1; |
3b18152c | 2284 | |
bc1c1169 | 2285 | return cfqd; |
1da177e4 LT |
2286 | } |
2287 | ||
2288 | static void cfq_slab_kill(void) | |
2289 | { | |
d6de8be7 JA |
2290 | /* |
2291 | * Caller already ensured that pending RCU callbacks are completed, | |
2292 | * so we should have no busy allocations at this point. | |
2293 | */ | |
1da177e4 LT |
2294 | if (cfq_pool) |
2295 | kmem_cache_destroy(cfq_pool); | |
2296 | if (cfq_ioc_pool) | |
2297 | kmem_cache_destroy(cfq_ioc_pool); | |
2298 | } | |
2299 | ||
2300 | static int __init cfq_slab_setup(void) | |
2301 | { | |
0a31bd5f | 2302 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
1da177e4 LT |
2303 | if (!cfq_pool) |
2304 | goto fail; | |
2305 | ||
34e6bbf2 | 2306 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
1da177e4 LT |
2307 | if (!cfq_ioc_pool) |
2308 | goto fail; | |
2309 | ||
2310 | return 0; | |
2311 | fail: | |
2312 | cfq_slab_kill(); | |
2313 | return -ENOMEM; | |
2314 | } | |
2315 | ||
1da177e4 LT |
2316 | /* |
2317 | * sysfs parts below --> | |
2318 | */ | |
1da177e4 LT |
2319 | static ssize_t |
2320 | cfq_var_show(unsigned int var, char *page) | |
2321 | { | |
2322 | return sprintf(page, "%d\n", var); | |
2323 | } | |
2324 | ||
2325 | static ssize_t | |
2326 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2327 | { | |
2328 | char *p = (char *) page; | |
2329 | ||
2330 | *var = simple_strtoul(p, &p, 10); | |
2331 | return count; | |
2332 | } | |
2333 | ||
1da177e4 | 2334 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 2335 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 2336 | { \ |
3d1ab40f | 2337 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2338 | unsigned int __data = __VAR; \ |
2339 | if (__CONV) \ | |
2340 | __data = jiffies_to_msecs(__data); \ | |
2341 | return cfq_var_show(__data, (page)); \ | |
2342 | } | |
2343 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2344 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2345 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2346 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2347 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2348 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2349 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2350 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2351 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2352 | #undef SHOW_FUNCTION |
2353 | ||
2354 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 2355 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 2356 | { \ |
3d1ab40f | 2357 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2358 | unsigned int __data; \ |
2359 | int ret = cfq_var_store(&__data, (page), count); \ | |
2360 | if (__data < (MIN)) \ | |
2361 | __data = (MIN); \ | |
2362 | else if (__data > (MAX)) \ | |
2363 | __data = (MAX); \ | |
2364 | if (__CONV) \ | |
2365 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2366 | else \ | |
2367 | *(__PTR) = __data; \ | |
2368 | return ret; \ | |
2369 | } | |
2370 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
2371 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
2372 | UINT_MAX, 1); | |
2373 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
2374 | UINT_MAX, 1); | |
e572ec7e | 2375 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
2376 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
2377 | UINT_MAX, 0); | |
22e2c507 JA |
2378 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2379 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2380 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
2381 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
2382 | UINT_MAX, 0); | |
1da177e4 LT |
2383 | #undef STORE_FUNCTION |
2384 | ||
e572ec7e AV |
2385 | #define CFQ_ATTR(name) \ |
2386 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2387 | ||
2388 | static struct elv_fs_entry cfq_attrs[] = { | |
2389 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2390 | CFQ_ATTR(fifo_expire_sync), |
2391 | CFQ_ATTR(fifo_expire_async), | |
2392 | CFQ_ATTR(back_seek_max), | |
2393 | CFQ_ATTR(back_seek_penalty), | |
2394 | CFQ_ATTR(slice_sync), | |
2395 | CFQ_ATTR(slice_async), | |
2396 | CFQ_ATTR(slice_async_rq), | |
2397 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2398 | __ATTR_NULL |
1da177e4 LT |
2399 | }; |
2400 | ||
1da177e4 LT |
2401 | static struct elevator_type iosched_cfq = { |
2402 | .ops = { | |
2403 | .elevator_merge_fn = cfq_merge, | |
2404 | .elevator_merged_fn = cfq_merged_request, | |
2405 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2406 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2407 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2408 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2409 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2410 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2411 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2412 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2413 | .elevator_former_req_fn = elv_rb_former_request, |
2414 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2415 | .elevator_set_req_fn = cfq_set_request, |
2416 | .elevator_put_req_fn = cfq_put_request, | |
2417 | .elevator_may_queue_fn = cfq_may_queue, | |
2418 | .elevator_init_fn = cfq_init_queue, | |
2419 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2420 | .trim = cfq_free_io_context, |
1da177e4 | 2421 | }, |
3d1ab40f | 2422 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2423 | .elevator_name = "cfq", |
2424 | .elevator_owner = THIS_MODULE, | |
2425 | }; | |
2426 | ||
2427 | static int __init cfq_init(void) | |
2428 | { | |
22e2c507 JA |
2429 | /* |
2430 | * could be 0 on HZ < 1000 setups | |
2431 | */ | |
2432 | if (!cfq_slice_async) | |
2433 | cfq_slice_async = 1; | |
2434 | if (!cfq_slice_idle) | |
2435 | cfq_slice_idle = 1; | |
2436 | ||
1da177e4 LT |
2437 | if (cfq_slab_setup()) |
2438 | return -ENOMEM; | |
2439 | ||
2fdd82bd | 2440 | elv_register(&iosched_cfq); |
1da177e4 | 2441 | |
2fdd82bd | 2442 | return 0; |
1da177e4 LT |
2443 | } |
2444 | ||
2445 | static void __exit cfq_exit(void) | |
2446 | { | |
6e9a4738 | 2447 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2448 | elv_unregister(&iosched_cfq); |
334e94de | 2449 | ioc_gone = &all_gone; |
fba82272 OH |
2450 | /* ioc_gone's update must be visible before reading ioc_count */ |
2451 | smp_wmb(); | |
d6de8be7 JA |
2452 | |
2453 | /* | |
2454 | * this also protects us from entering cfq_slab_kill() with | |
2455 | * pending RCU callbacks | |
2456 | */ | |
4050cf16 | 2457 | if (elv_ioc_count_read(ioc_count)) |
9a11b4ed | 2458 | wait_for_completion(&all_gone); |
83521d3e | 2459 | cfq_slab_kill(); |
1da177e4 LT |
2460 | } |
2461 | ||
2462 | module_init(cfq_init); | |
2463 | module_exit(cfq_exit); | |
2464 | ||
2465 | MODULE_AUTHOR("Jens Axboe"); | |
2466 | MODULE_LICENSE("GPL"); | |
2467 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |