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 */ | |
182 | CFQ_CFQQ_FLAG_must_dispatch, /* must dispatch, even if expired */ | |
183 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ | |
184 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
185 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
186 | CFQ_CFQQ_FLAG_queue_new, /* queue never been serviced */ | |
44f7c160 | 187 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 188 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
3b18152c JA |
189 | }; |
190 | ||
191 | #define CFQ_CFQQ_FNS(name) \ | |
192 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
193 | { \ | |
fe094d98 | 194 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
195 | } \ |
196 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
197 | { \ | |
fe094d98 | 198 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
199 | } \ |
200 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
201 | { \ | |
fe094d98 | 202 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
203 | } |
204 | ||
205 | CFQ_CFQQ_FNS(on_rr); | |
206 | CFQ_CFQQ_FNS(wait_request); | |
207 | CFQ_CFQQ_FNS(must_alloc); | |
208 | CFQ_CFQQ_FNS(must_alloc_slice); | |
209 | CFQ_CFQQ_FNS(must_dispatch); | |
210 | CFQ_CFQQ_FNS(fifo_expire); | |
211 | CFQ_CFQQ_FNS(idle_window); | |
212 | CFQ_CFQQ_FNS(prio_changed); | |
53b03744 | 213 | CFQ_CFQQ_FNS(queue_new); |
44f7c160 | 214 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 215 | CFQ_CFQQ_FNS(sync); |
3b18152c JA |
216 | #undef CFQ_CFQQ_FNS |
217 | ||
7b679138 JA |
218 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
219 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
220 | #define cfq_log(cfqd, fmt, args...) \ | |
221 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
222 | ||
165125e1 | 223 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
91fac317 | 224 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, int, |
fd0928df | 225 | struct io_context *, gfp_t); |
4ac845a2 | 226 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
91fac317 VT |
227 | struct io_context *); |
228 | ||
229 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, | |
230 | int is_sync) | |
231 | { | |
232 | return cic->cfqq[!!is_sync]; | |
233 | } | |
234 | ||
235 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
236 | struct cfq_queue *cfqq, int is_sync) | |
237 | { | |
238 | cic->cfqq[!!is_sync] = cfqq; | |
239 | } | |
240 | ||
241 | /* | |
242 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
243 | * set (in which case it could also be direct WRITE). | |
244 | */ | |
245 | static inline int cfq_bio_sync(struct bio *bio) | |
246 | { | |
247 | if (bio_data_dir(bio) == READ || bio_sync(bio)) | |
248 | return 1; | |
249 | ||
250 | return 0; | |
251 | } | |
1da177e4 | 252 | |
99f95e52 AM |
253 | /* |
254 | * scheduler run of queue, if there are requests pending and no one in the | |
255 | * driver that will restart queueing | |
256 | */ | |
257 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
258 | { | |
7b679138 JA |
259 | if (cfqd->busy_queues) { |
260 | cfq_log(cfqd, "schedule dispatch"); | |
18887ad9 | 261 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 262 | } |
99f95e52 AM |
263 | } |
264 | ||
165125e1 | 265 | static int cfq_queue_empty(struct request_queue *q) |
99f95e52 AM |
266 | { |
267 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
268 | ||
b4878f24 | 269 | return !cfqd->busy_queues; |
99f95e52 AM |
270 | } |
271 | ||
44f7c160 JA |
272 | /* |
273 | * Scale schedule slice based on io priority. Use the sync time slice only | |
274 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
275 | * io only, should not get full sync slice length. | |
276 | */ | |
d9e7620e JA |
277 | static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync, |
278 | unsigned short prio) | |
44f7c160 | 279 | { |
d9e7620e | 280 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 281 | |
d9e7620e JA |
282 | WARN_ON(prio >= IOPRIO_BE_NR); |
283 | ||
284 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
285 | } | |
44f7c160 | 286 | |
d9e7620e JA |
287 | static inline int |
288 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
289 | { | |
290 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
291 | } |
292 | ||
293 | static inline void | |
294 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
295 | { | |
296 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; | |
7b679138 | 297 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
298 | } |
299 | ||
300 | /* | |
301 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
302 | * isn't valid until the first request from the dispatch is activated | |
303 | * and the slice time set. | |
304 | */ | |
305 | static inline int cfq_slice_used(struct cfq_queue *cfqq) | |
306 | { | |
307 | if (cfq_cfqq_slice_new(cfqq)) | |
308 | return 0; | |
309 | if (time_before(jiffies, cfqq->slice_end)) | |
310 | return 0; | |
311 | ||
312 | return 1; | |
313 | } | |
314 | ||
1da177e4 | 315 | /* |
5e705374 | 316 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 317 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 318 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 319 | */ |
5e705374 JA |
320 | static struct request * |
321 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
322 | { |
323 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 324 | unsigned long back_max; |
e8a99053 AM |
325 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
326 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
327 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 328 | |
5e705374 JA |
329 | if (rq1 == NULL || rq1 == rq2) |
330 | return rq2; | |
331 | if (rq2 == NULL) | |
332 | return rq1; | |
9c2c38a1 | 333 | |
5e705374 JA |
334 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
335 | return rq1; | |
336 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
337 | return rq2; | |
374f84ac JA |
338 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
339 | return rq1; | |
340 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
341 | return rq2; | |
1da177e4 | 342 | |
5e705374 JA |
343 | s1 = rq1->sector; |
344 | s2 = rq2->sector; | |
1da177e4 | 345 | |
6d048f53 | 346 | last = cfqd->last_position; |
1da177e4 | 347 | |
1da177e4 LT |
348 | /* |
349 | * by definition, 1KiB is 2 sectors | |
350 | */ | |
351 | back_max = cfqd->cfq_back_max * 2; | |
352 | ||
353 | /* | |
354 | * Strict one way elevator _except_ in the case where we allow | |
355 | * short backward seeks which are biased as twice the cost of a | |
356 | * similar forward seek. | |
357 | */ | |
358 | if (s1 >= last) | |
359 | d1 = s1 - last; | |
360 | else if (s1 + back_max >= last) | |
361 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
362 | else | |
e8a99053 | 363 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
364 | |
365 | if (s2 >= last) | |
366 | d2 = s2 - last; | |
367 | else if (s2 + back_max >= last) | |
368 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
369 | else | |
e8a99053 | 370 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
371 | |
372 | /* Found required data */ | |
e8a99053 AM |
373 | |
374 | /* | |
375 | * By doing switch() on the bit mask "wrap" we avoid having to | |
376 | * check two variables for all permutations: --> faster! | |
377 | */ | |
378 | switch (wrap) { | |
5e705374 | 379 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 380 | if (d1 < d2) |
5e705374 | 381 | return rq1; |
e8a99053 | 382 | else if (d2 < d1) |
5e705374 | 383 | return rq2; |
e8a99053 AM |
384 | else { |
385 | if (s1 >= s2) | |
5e705374 | 386 | return rq1; |
e8a99053 | 387 | else |
5e705374 | 388 | return rq2; |
e8a99053 | 389 | } |
1da177e4 | 390 | |
e8a99053 | 391 | case CFQ_RQ2_WRAP: |
5e705374 | 392 | return rq1; |
e8a99053 | 393 | case CFQ_RQ1_WRAP: |
5e705374 JA |
394 | return rq2; |
395 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
396 | default: |
397 | /* | |
398 | * Since both rqs are wrapped, | |
399 | * start with the one that's further behind head | |
400 | * (--> only *one* back seek required), | |
401 | * since back seek takes more time than forward. | |
402 | */ | |
403 | if (s1 <= s2) | |
5e705374 | 404 | return rq1; |
1da177e4 | 405 | else |
5e705374 | 406 | return rq2; |
1da177e4 LT |
407 | } |
408 | } | |
409 | ||
498d3aa2 JA |
410 | /* |
411 | * The below is leftmost cache rbtree addon | |
412 | */ | |
0871714e | 413 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 JA |
414 | { |
415 | if (!root->left) | |
416 | root->left = rb_first(&root->rb); | |
417 | ||
0871714e JA |
418 | if (root->left) |
419 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
420 | ||
421 | return NULL; | |
cc09e299 JA |
422 | } |
423 | ||
424 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) | |
425 | { | |
426 | if (root->left == n) | |
427 | root->left = NULL; | |
428 | ||
429 | rb_erase(n, &root->rb); | |
430 | RB_CLEAR_NODE(n); | |
431 | } | |
432 | ||
1da177e4 LT |
433 | /* |
434 | * would be nice to take fifo expire time into account as well | |
435 | */ | |
5e705374 JA |
436 | static struct request * |
437 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
438 | struct request *last) | |
1da177e4 | 439 | { |
21183b07 JA |
440 | struct rb_node *rbnext = rb_next(&last->rb_node); |
441 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 442 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 443 | |
21183b07 | 444 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
445 | |
446 | if (rbprev) | |
5e705374 | 447 | prev = rb_entry_rq(rbprev); |
1da177e4 | 448 | |
21183b07 | 449 | if (rbnext) |
5e705374 | 450 | next = rb_entry_rq(rbnext); |
21183b07 JA |
451 | else { |
452 | rbnext = rb_first(&cfqq->sort_list); | |
453 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 454 | next = rb_entry_rq(rbnext); |
21183b07 | 455 | } |
1da177e4 | 456 | |
21183b07 | 457 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
458 | } |
459 | ||
d9e7620e JA |
460 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
461 | struct cfq_queue *cfqq) | |
1da177e4 | 462 | { |
d9e7620e JA |
463 | /* |
464 | * just an approximation, should be ok. | |
465 | */ | |
67e6b49e JA |
466 | return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
467 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); | |
d9e7620e JA |
468 | } |
469 | ||
498d3aa2 JA |
470 | /* |
471 | * The cfqd->service_tree holds all pending cfq_queue's that have | |
472 | * requests waiting to be processed. It is sorted in the order that | |
473 | * we will service the queues. | |
474 | */ | |
d9e7620e | 475 | static void cfq_service_tree_add(struct cfq_data *cfqd, |
edd75ffd | 476 | struct cfq_queue *cfqq, int add_front) |
d9e7620e | 477 | { |
0871714e JA |
478 | struct rb_node **p, *parent; |
479 | struct cfq_queue *__cfqq; | |
d9e7620e | 480 | unsigned long rb_key; |
498d3aa2 | 481 | int left; |
d9e7620e | 482 | |
0871714e JA |
483 | if (cfq_class_idle(cfqq)) { |
484 | rb_key = CFQ_IDLE_DELAY; | |
485 | parent = rb_last(&cfqd->service_tree.rb); | |
486 | if (parent && parent != &cfqq->rb_node) { | |
487 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
488 | rb_key += __cfqq->rb_key; | |
489 | } else | |
490 | rb_key += jiffies; | |
491 | } else if (!add_front) { | |
edd75ffd JA |
492 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
493 | rb_key += cfqq->slice_resid; | |
494 | cfqq->slice_resid = 0; | |
495 | } else | |
496 | rb_key = 0; | |
1da177e4 | 497 | |
d9e7620e | 498 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
99f9628a | 499 | /* |
d9e7620e | 500 | * same position, nothing more to do |
99f9628a | 501 | */ |
d9e7620e JA |
502 | if (rb_key == cfqq->rb_key) |
503 | return; | |
1da177e4 | 504 | |
cc09e299 | 505 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); |
1da177e4 | 506 | } |
d9e7620e | 507 | |
498d3aa2 | 508 | left = 1; |
0871714e JA |
509 | parent = NULL; |
510 | p = &cfqd->service_tree.rb.rb_node; | |
d9e7620e | 511 | while (*p) { |
67060e37 | 512 | struct rb_node **n; |
cc09e299 | 513 | |
d9e7620e JA |
514 | parent = *p; |
515 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
516 | ||
0c534e0a JA |
517 | /* |
518 | * sort RT queues first, we always want to give | |
67060e37 JA |
519 | * preference to them. IDLE queues goes to the back. |
520 | * after that, sort on the next service time. | |
0c534e0a JA |
521 | */ |
522 | if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq)) | |
67060e37 | 523 | n = &(*p)->rb_left; |
0c534e0a | 524 | else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq)) |
67060e37 JA |
525 | n = &(*p)->rb_right; |
526 | else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq)) | |
527 | n = &(*p)->rb_left; | |
528 | else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq)) | |
529 | n = &(*p)->rb_right; | |
0c534e0a | 530 | else if (rb_key < __cfqq->rb_key) |
67060e37 JA |
531 | n = &(*p)->rb_left; |
532 | else | |
533 | n = &(*p)->rb_right; | |
534 | ||
535 | if (n == &(*p)->rb_right) | |
cc09e299 | 536 | left = 0; |
67060e37 JA |
537 | |
538 | p = n; | |
d9e7620e JA |
539 | } |
540 | ||
cc09e299 JA |
541 | if (left) |
542 | cfqd->service_tree.left = &cfqq->rb_node; | |
543 | ||
d9e7620e JA |
544 | cfqq->rb_key = rb_key; |
545 | rb_link_node(&cfqq->rb_node, parent, p); | |
cc09e299 | 546 | rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb); |
1da177e4 LT |
547 | } |
548 | ||
498d3aa2 JA |
549 | /* |
550 | * Update cfqq's position in the service tree. | |
551 | */ | |
edd75ffd | 552 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 553 | { |
6d048f53 JA |
554 | /* |
555 | * Resorting requires the cfqq to be on the RR list already. | |
556 | */ | |
498d3aa2 | 557 | if (cfq_cfqq_on_rr(cfqq)) |
edd75ffd | 558 | cfq_service_tree_add(cfqd, cfqq, 0); |
6d048f53 JA |
559 | } |
560 | ||
1da177e4 LT |
561 | /* |
562 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 563 | * the pending list according to last request service |
1da177e4 | 564 | */ |
febffd61 | 565 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 566 | { |
7b679138 | 567 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
568 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
569 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 570 | cfqd->busy_queues++; |
3a9a3f6c DS |
571 | if (cfq_class_rt(cfqq)) |
572 | cfqd->busy_rt_queues++; | |
1da177e4 | 573 | |
edd75ffd | 574 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
575 | } |
576 | ||
498d3aa2 JA |
577 | /* |
578 | * Called when the cfqq no longer has requests pending, remove it from | |
579 | * the service tree. | |
580 | */ | |
febffd61 | 581 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 582 | { |
7b679138 | 583 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
584 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
585 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 586 | |
cc09e299 JA |
587 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) |
588 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); | |
d9e7620e | 589 | |
1da177e4 LT |
590 | BUG_ON(!cfqd->busy_queues); |
591 | cfqd->busy_queues--; | |
3a9a3f6c DS |
592 | if (cfq_class_rt(cfqq)) |
593 | cfqd->busy_rt_queues--; | |
1da177e4 LT |
594 | } |
595 | ||
596 | /* | |
597 | * rb tree support functions | |
598 | */ | |
febffd61 | 599 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 600 | { |
5e705374 | 601 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 602 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 603 | const int sync = rq_is_sync(rq); |
1da177e4 | 604 | |
b4878f24 JA |
605 | BUG_ON(!cfqq->queued[sync]); |
606 | cfqq->queued[sync]--; | |
1da177e4 | 607 | |
5e705374 | 608 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 609 | |
dd67d051 | 610 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 611 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
612 | } |
613 | ||
5e705374 | 614 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 615 | { |
5e705374 | 616 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 617 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 618 | struct request *__alias; |
1da177e4 | 619 | |
5380a101 | 620 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
621 | |
622 | /* | |
623 | * looks a little odd, but the first insert might return an alias. | |
624 | * if that happens, put the alias on the dispatch list | |
625 | */ | |
21183b07 | 626 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 627 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
628 | |
629 | if (!cfq_cfqq_on_rr(cfqq)) | |
630 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
631 | |
632 | /* | |
633 | * check if this request is a better next-serve candidate | |
634 | */ | |
635 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); | |
636 | BUG_ON(!cfqq->next_rq); | |
1da177e4 LT |
637 | } |
638 | ||
febffd61 | 639 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 640 | { |
5380a101 JA |
641 | elv_rb_del(&cfqq->sort_list, rq); |
642 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 643 | cfq_add_rq_rb(rq); |
1da177e4 LT |
644 | } |
645 | ||
206dc69b JA |
646 | static struct request * |
647 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 648 | { |
206dc69b | 649 | struct task_struct *tsk = current; |
91fac317 | 650 | struct cfq_io_context *cic; |
206dc69b | 651 | struct cfq_queue *cfqq; |
1da177e4 | 652 | |
4ac845a2 | 653 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
654 | if (!cic) |
655 | return NULL; | |
656 | ||
657 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
658 | if (cfqq) { |
659 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
660 | ||
21183b07 | 661 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 662 | } |
1da177e4 | 663 | |
1da177e4 LT |
664 | return NULL; |
665 | } | |
666 | ||
165125e1 | 667 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 668 | { |
22e2c507 | 669 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 670 | |
b4878f24 | 671 | cfqd->rq_in_driver++; |
7b679138 JA |
672 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
673 | cfqd->rq_in_driver); | |
25776e35 | 674 | |
6d048f53 | 675 | cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors; |
1da177e4 LT |
676 | } |
677 | ||
165125e1 | 678 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 679 | { |
b4878f24 JA |
680 | struct cfq_data *cfqd = q->elevator->elevator_data; |
681 | ||
682 | WARN_ON(!cfqd->rq_in_driver); | |
683 | cfqd->rq_in_driver--; | |
7b679138 JA |
684 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
685 | cfqd->rq_in_driver); | |
1da177e4 LT |
686 | } |
687 | ||
b4878f24 | 688 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 689 | { |
5e705374 | 690 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 691 | |
5e705374 JA |
692 | if (cfqq->next_rq == rq) |
693 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 694 | |
b4878f24 | 695 | list_del_init(&rq->queuelist); |
5e705374 | 696 | cfq_del_rq_rb(rq); |
374f84ac | 697 | |
45333d5a | 698 | cfqq->cfqd->rq_queued--; |
374f84ac JA |
699 | if (rq_is_meta(rq)) { |
700 | WARN_ON(!cfqq->meta_pending); | |
701 | cfqq->meta_pending--; | |
702 | } | |
1da177e4 LT |
703 | } |
704 | ||
165125e1 JA |
705 | static int cfq_merge(struct request_queue *q, struct request **req, |
706 | struct bio *bio) | |
1da177e4 LT |
707 | { |
708 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
709 | struct request *__rq; | |
1da177e4 | 710 | |
206dc69b | 711 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 712 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
713 | *req = __rq; |
714 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
715 | } |
716 | ||
717 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
718 | } |
719 | ||
165125e1 | 720 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 721 | int type) |
1da177e4 | 722 | { |
21183b07 | 723 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 724 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 725 | |
5e705374 | 726 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 727 | } |
1da177e4 LT |
728 | } |
729 | ||
730 | static void | |
165125e1 | 731 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
732 | struct request *next) |
733 | { | |
22e2c507 JA |
734 | /* |
735 | * reposition in fifo if next is older than rq | |
736 | */ | |
737 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
738 | time_before(next->start_time, rq->start_time)) | |
739 | list_move(&rq->queuelist, &next->queuelist); | |
740 | ||
b4878f24 | 741 | cfq_remove_request(next); |
22e2c507 JA |
742 | } |
743 | ||
165125e1 | 744 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
745 | struct bio *bio) |
746 | { | |
747 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
91fac317 | 748 | struct cfq_io_context *cic; |
da775265 | 749 | struct cfq_queue *cfqq; |
da775265 JA |
750 | |
751 | /* | |
ec8acb69 | 752 | * Disallow merge of a sync bio into an async request. |
da775265 | 753 | */ |
91fac317 | 754 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
da775265 JA |
755 | return 0; |
756 | ||
757 | /* | |
719d3402 JA |
758 | * Lookup the cfqq that this bio will be queued with. Allow |
759 | * merge only if rq is queued there. | |
da775265 | 760 | */ |
4ac845a2 | 761 | cic = cfq_cic_lookup(cfqd, current->io_context); |
91fac317 VT |
762 | if (!cic) |
763 | return 0; | |
719d3402 | 764 | |
91fac317 | 765 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
719d3402 JA |
766 | if (cfqq == RQ_CFQQ(rq)) |
767 | return 1; | |
da775265 | 768 | |
ec8acb69 | 769 | return 0; |
da775265 JA |
770 | } |
771 | ||
febffd61 JA |
772 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
773 | struct cfq_queue *cfqq) | |
22e2c507 JA |
774 | { |
775 | if (cfqq) { | |
7b679138 | 776 | cfq_log_cfqq(cfqd, cfqq, "set_active"); |
22e2c507 | 777 | cfqq->slice_end = 0; |
2f5cb738 JA |
778 | cfqq->slice_dispatch = 0; |
779 | ||
780 | cfq_clear_cfqq_must_dispatch(cfqq); | |
781 | cfq_clear_cfqq_wait_request(cfqq); | |
3b18152c JA |
782 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
783 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 784 | cfq_mark_cfqq_slice_new(cfqq); |
1afba045 | 785 | cfq_clear_cfqq_queue_new(cfqq); |
2f5cb738 JA |
786 | |
787 | del_timer(&cfqd->idle_slice_timer); | |
22e2c507 JA |
788 | } |
789 | ||
790 | cfqd->active_queue = cfqq; | |
791 | } | |
792 | ||
7b14e3b5 JA |
793 | /* |
794 | * current cfqq expired its slice (or was too idle), select new one | |
795 | */ | |
796 | static void | |
797 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
6084cdda | 798 | int timed_out) |
7b14e3b5 | 799 | { |
7b679138 JA |
800 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
801 | ||
7b14e3b5 JA |
802 | if (cfq_cfqq_wait_request(cfqq)) |
803 | del_timer(&cfqd->idle_slice_timer); | |
804 | ||
7b14e3b5 JA |
805 | cfq_clear_cfqq_must_dispatch(cfqq); |
806 | cfq_clear_cfqq_wait_request(cfqq); | |
807 | ||
808 | /* | |
6084cdda | 809 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 810 | */ |
7b679138 | 811 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) { |
c5b680f3 | 812 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b679138 JA |
813 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
814 | } | |
7b14e3b5 | 815 | |
edd75ffd | 816 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
817 | |
818 | if (cfqq == cfqd->active_queue) | |
819 | cfqd->active_queue = NULL; | |
820 | ||
821 | if (cfqd->active_cic) { | |
822 | put_io_context(cfqd->active_cic->ioc); | |
823 | cfqd->active_cic = NULL; | |
824 | } | |
7b14e3b5 JA |
825 | } |
826 | ||
6084cdda | 827 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out) |
7b14e3b5 JA |
828 | { |
829 | struct cfq_queue *cfqq = cfqd->active_queue; | |
830 | ||
831 | if (cfqq) | |
6084cdda | 832 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
833 | } |
834 | ||
498d3aa2 JA |
835 | /* |
836 | * Get next queue for service. Unless we have a queue preemption, | |
837 | * we'll simply select the first cfqq in the service tree. | |
838 | */ | |
6d048f53 | 839 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 840 | { |
edd75ffd JA |
841 | if (RB_EMPTY_ROOT(&cfqd->service_tree.rb)) |
842 | return NULL; | |
d9e7620e | 843 | |
0871714e | 844 | return cfq_rb_first(&cfqd->service_tree); |
6d048f53 JA |
845 | } |
846 | ||
498d3aa2 JA |
847 | /* |
848 | * Get and set a new active queue for service. | |
849 | */ | |
6d048f53 JA |
850 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
851 | { | |
852 | struct cfq_queue *cfqq; | |
853 | ||
d9e7620e | 854 | cfqq = cfq_get_next_queue(cfqd); |
22e2c507 | 855 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 856 | return cfqq; |
22e2c507 JA |
857 | } |
858 | ||
d9e7620e JA |
859 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
860 | struct request *rq) | |
861 | { | |
862 | if (rq->sector >= cfqd->last_position) | |
863 | return rq->sector - cfqd->last_position; | |
864 | else | |
865 | return cfqd->last_position - rq->sector; | |
866 | } | |
867 | ||
6d048f53 JA |
868 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq) |
869 | { | |
870 | struct cfq_io_context *cic = cfqd->active_cic; | |
871 | ||
872 | if (!sample_valid(cic->seek_samples)) | |
873 | return 0; | |
874 | ||
875 | return cfq_dist_from_last(cfqd, rq) <= cic->seek_mean; | |
876 | } | |
877 | ||
d9e7620e JA |
878 | static int cfq_close_cooperator(struct cfq_data *cfq_data, |
879 | struct cfq_queue *cfqq) | |
6d048f53 | 880 | { |
6d048f53 | 881 | /* |
d9e7620e JA |
882 | * We should notice if some of the queues are cooperating, eg |
883 | * working closely on the same area of the disk. In that case, | |
884 | * we can group them together and don't waste time idling. | |
6d048f53 | 885 | */ |
d9e7620e | 886 | return 0; |
6d048f53 JA |
887 | } |
888 | ||
889 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (8 * 1024)) | |
caaa5f9f | 890 | |
6d048f53 | 891 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 892 | { |
1792669c | 893 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 894 | struct cfq_io_context *cic; |
7b14e3b5 JA |
895 | unsigned long sl; |
896 | ||
a68bbddb | 897 | /* |
f7d7b7a7 JA |
898 | * SSD device without seek penalty, disable idling. But only do so |
899 | * for devices that support queuing, otherwise we still have a problem | |
900 | * with sync vs async workloads. | |
a68bbddb | 901 | */ |
f7d7b7a7 | 902 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
903 | return; |
904 | ||
dd67d051 | 905 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 906 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
907 | |
908 | /* | |
909 | * idle is disabled, either manually or by past process history | |
910 | */ | |
6d048f53 JA |
911 | if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq)) |
912 | return; | |
913 | ||
7b679138 JA |
914 | /* |
915 | * still requests with the driver, don't idle | |
916 | */ | |
917 | if (cfqd->rq_in_driver) | |
918 | return; | |
919 | ||
22e2c507 JA |
920 | /* |
921 | * task has exited, don't wait | |
922 | */ | |
206dc69b | 923 | cic = cfqd->active_cic; |
66dac98e | 924 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
6d048f53 JA |
925 | return; |
926 | ||
927 | /* | |
928 | * See if this prio level has a good candidate | |
929 | */ | |
1afba045 JA |
930 | if (cfq_close_cooperator(cfqd, cfqq) && |
931 | (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2)) | |
6d048f53 | 932 | return; |
22e2c507 | 933 | |
3b18152c JA |
934 | cfq_mark_cfqq_must_dispatch(cfqq); |
935 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 936 | |
206dc69b JA |
937 | /* |
938 | * we don't want to idle for seeks, but we do want to allow | |
939 | * fair distribution of slice time for a process doing back-to-back | |
940 | * seeks. so allow a little bit of time for him to submit a new rq | |
941 | */ | |
6d048f53 | 942 | sl = cfqd->cfq_slice_idle; |
caaa5f9f | 943 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
d9e7620e | 944 | sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT)); |
206dc69b | 945 | |
7b14e3b5 | 946 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
7b679138 | 947 | cfq_log(cfqd, "arm_idle: %lu", sl); |
1da177e4 LT |
948 | } |
949 | ||
498d3aa2 JA |
950 | /* |
951 | * Move request from internal lists to the request queue dispatch list. | |
952 | */ | |
165125e1 | 953 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 954 | { |
3ed9a296 | 955 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 956 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 957 | |
7b679138 JA |
958 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
959 | ||
5380a101 | 960 | cfq_remove_request(rq); |
6d048f53 | 961 | cfqq->dispatched++; |
5380a101 | 962 | elv_dispatch_sort(q, rq); |
3ed9a296 JA |
963 | |
964 | if (cfq_cfqq_sync(cfqq)) | |
965 | cfqd->sync_flight++; | |
1da177e4 LT |
966 | } |
967 | ||
968 | /* | |
969 | * return expired entry, or NULL to just start from scratch in rbtree | |
970 | */ | |
febffd61 | 971 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
972 | { |
973 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 974 | struct request *rq; |
89850f7e | 975 | int fifo; |
1da177e4 | 976 | |
3b18152c | 977 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 978 | return NULL; |
cb887411 JA |
979 | |
980 | cfq_mark_cfqq_fifo_expire(cfqq); | |
981 | ||
89850f7e JA |
982 | if (list_empty(&cfqq->fifo)) |
983 | return NULL; | |
1da177e4 | 984 | |
6d048f53 | 985 | fifo = cfq_cfqq_sync(cfqq); |
89850f7e | 986 | rq = rq_entry_fifo(cfqq->fifo.next); |
1da177e4 | 987 | |
6d048f53 | 988 | if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) |
7b679138 | 989 | rq = NULL; |
1da177e4 | 990 | |
7b679138 | 991 | cfq_log_cfqq(cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 992 | return rq; |
1da177e4 LT |
993 | } |
994 | ||
22e2c507 JA |
995 | static inline int |
996 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
997 | { | |
998 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 999 | |
22e2c507 | 1000 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 1001 | |
22e2c507 | 1002 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
1003 | } |
1004 | ||
22e2c507 | 1005 | /* |
498d3aa2 JA |
1006 | * Select a queue for service. If we have a current active queue, |
1007 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 1008 | */ |
1b5ed5e1 | 1009 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 1010 | { |
1da177e4 | 1011 | struct cfq_queue *cfqq; |
1da177e4 | 1012 | |
22e2c507 JA |
1013 | cfqq = cfqd->active_queue; |
1014 | if (!cfqq) | |
1015 | goto new_queue; | |
1da177e4 | 1016 | |
22e2c507 | 1017 | /* |
6d048f53 | 1018 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 1019 | */ |
6d048f53 | 1020 | if (cfq_slice_used(cfqq)) |
3b18152c | 1021 | goto expire; |
1da177e4 | 1022 | |
3a9a3f6c DS |
1023 | /* |
1024 | * If we have a RT cfqq waiting, then we pre-empt the current non-rt | |
1025 | * cfqq. | |
1026 | */ | |
1027 | if (!cfq_class_rt(cfqq) && cfqd->busy_rt_queues) { | |
1028 | /* | |
1029 | * We simulate this as cfqq timed out so that it gets to bank | |
1030 | * the remaining of its time slice. | |
1031 | */ | |
1032 | cfq_log_cfqq(cfqd, cfqq, "preempt"); | |
1033 | cfq_slice_expired(cfqd, 1); | |
1034 | goto new_queue; | |
1035 | } | |
1036 | ||
22e2c507 | 1037 | /* |
6d048f53 JA |
1038 | * The active queue has requests and isn't expired, allow it to |
1039 | * dispatch. | |
22e2c507 | 1040 | */ |
dd67d051 | 1041 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 1042 | goto keep_queue; |
6d048f53 JA |
1043 | |
1044 | /* | |
1045 | * No requests pending. If the active queue still has requests in | |
1046 | * flight or is idling for a new request, allow either of these | |
1047 | * conditions to happen (or time out) before selecting a new queue. | |
1048 | */ | |
cc197479 JA |
1049 | if (timer_pending(&cfqd->idle_slice_timer) || |
1050 | (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) { | |
caaa5f9f JA |
1051 | cfqq = NULL; |
1052 | goto keep_queue; | |
22e2c507 JA |
1053 | } |
1054 | ||
3b18152c | 1055 | expire: |
6084cdda | 1056 | cfq_slice_expired(cfqd, 0); |
3b18152c JA |
1057 | new_queue: |
1058 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 1059 | keep_queue: |
3b18152c | 1060 | return cfqq; |
22e2c507 JA |
1061 | } |
1062 | ||
febffd61 | 1063 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
1064 | { |
1065 | int dispatched = 0; | |
1066 | ||
1067 | while (cfqq->next_rq) { | |
1068 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1069 | dispatched++; | |
1070 | } | |
1071 | ||
1072 | BUG_ON(!list_empty(&cfqq->fifo)); | |
1073 | return dispatched; | |
1074 | } | |
1075 | ||
498d3aa2 JA |
1076 | /* |
1077 | * Drain our current requests. Used for barriers and when switching | |
1078 | * io schedulers on-the-fly. | |
1079 | */ | |
d9e7620e | 1080 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 1081 | { |
0871714e | 1082 | struct cfq_queue *cfqq; |
d9e7620e | 1083 | int dispatched = 0; |
1b5ed5e1 | 1084 | |
0871714e | 1085 | while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL) |
d9e7620e | 1086 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
1b5ed5e1 | 1087 | |
6084cdda | 1088 | cfq_slice_expired(cfqd, 0); |
1b5ed5e1 TH |
1089 | |
1090 | BUG_ON(cfqd->busy_queues); | |
1091 | ||
7b679138 | 1092 | cfq_log(cfqd, "forced_dispatch=%d\n", dispatched); |
1b5ed5e1 TH |
1093 | return dispatched; |
1094 | } | |
1095 | ||
2f5cb738 JA |
1096 | /* |
1097 | * Dispatch a request from cfqq, moving them to the request queue | |
1098 | * dispatch list. | |
1099 | */ | |
1100 | static void cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1101 | { | |
1102 | struct request *rq; | |
1103 | ||
1104 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
1105 | ||
1106 | /* | |
1107 | * follow expired path, else get first next available | |
1108 | */ | |
1109 | rq = cfq_check_fifo(cfqq); | |
1110 | if (!rq) | |
1111 | rq = cfqq->next_rq; | |
1112 | ||
1113 | /* | |
1114 | * insert request into driver dispatch list | |
1115 | */ | |
1116 | cfq_dispatch_insert(cfqd->queue, rq); | |
1117 | ||
1118 | if (!cfqd->active_cic) { | |
1119 | struct cfq_io_context *cic = RQ_CIC(rq); | |
1120 | ||
1121 | atomic_inc(&cic->ioc->refcount); | |
1122 | cfqd->active_cic = cic; | |
1123 | } | |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * Find the cfqq that we need to service and move a request from that to the | |
1128 | * dispatch list | |
1129 | */ | |
165125e1 | 1130 | static int cfq_dispatch_requests(struct request_queue *q, int force) |
22e2c507 JA |
1131 | { |
1132 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
6d048f53 | 1133 | struct cfq_queue *cfqq; |
2f5cb738 | 1134 | unsigned int max_dispatch; |
22e2c507 JA |
1135 | |
1136 | if (!cfqd->busy_queues) | |
1137 | return 0; | |
1138 | ||
1b5ed5e1 TH |
1139 | if (unlikely(force)) |
1140 | return cfq_forced_dispatch(cfqd); | |
1141 | ||
2f5cb738 JA |
1142 | cfqq = cfq_select_queue(cfqd); |
1143 | if (!cfqq) | |
1144 | return 0; | |
1145 | ||
1146 | /* | |
1147 | * If this is an async queue and we have sync IO in flight, let it wait | |
1148 | */ | |
1149 | if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq)) | |
1150 | return 0; | |
1151 | ||
1152 | max_dispatch = cfqd->cfq_quantum; | |
1153 | if (cfq_class_idle(cfqq)) | |
1154 | max_dispatch = 1; | |
b4878f24 | 1155 | |
2f5cb738 JA |
1156 | /* |
1157 | * Does this cfqq already have too much IO in flight? | |
1158 | */ | |
1159 | if (cfqq->dispatched >= max_dispatch) { | |
1160 | /* | |
1161 | * idle queue must always only have a single IO in flight | |
1162 | */ | |
3ed9a296 | 1163 | if (cfq_class_idle(cfqq)) |
2f5cb738 | 1164 | return 0; |
3ed9a296 | 1165 | |
2f5cb738 JA |
1166 | /* |
1167 | * We have other queues, don't allow more IO from this one | |
1168 | */ | |
1169 | if (cfqd->busy_queues > 1) | |
1170 | return 0; | |
9ede209e | 1171 | |
2f5cb738 JA |
1172 | /* |
1173 | * we are the only queue, allow up to 4 times of 'quantum' | |
1174 | */ | |
1175 | if (cfqq->dispatched >= 4 * max_dispatch) | |
1176 | return 0; | |
1177 | } | |
3ed9a296 | 1178 | |
2f5cb738 JA |
1179 | /* |
1180 | * Dispatch a request from this cfqq | |
1181 | */ | |
1182 | cfq_dispatch_request(cfqd, cfqq); | |
1183 | cfqq->slice_dispatch++; | |
22e2c507 | 1184 | |
2f5cb738 JA |
1185 | /* |
1186 | * expire an async queue immediately if it has used up its slice. idle | |
1187 | * queue always expire after 1 dispatch round. | |
1188 | */ | |
1189 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
1190 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
1191 | cfq_class_idle(cfqq))) { | |
1192 | cfqq->slice_end = jiffies + 1; | |
1193 | cfq_slice_expired(cfqd, 0); | |
1da177e4 LT |
1194 | } |
1195 | ||
2f5cb738 JA |
1196 | cfq_log(cfqd, "dispatched a request"); |
1197 | return 1; | |
1da177e4 LT |
1198 | } |
1199 | ||
1da177e4 | 1200 | /* |
5e705374 JA |
1201 | * task holds one reference to the queue, dropped when task exits. each rq |
1202 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1203 | * |
1204 | * queue lock must be held here. | |
1205 | */ | |
1206 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1207 | { | |
22e2c507 JA |
1208 | struct cfq_data *cfqd = cfqq->cfqd; |
1209 | ||
1210 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1211 | |
1212 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1213 | return; | |
1214 | ||
7b679138 | 1215 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 1216 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 1217 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1218 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1219 | |
28f95cbc | 1220 | if (unlikely(cfqd->active_queue == cfqq)) { |
6084cdda | 1221 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1222 | cfq_schedule_dispatch(cfqd); |
1223 | } | |
22e2c507 | 1224 | |
1da177e4 LT |
1225 | kmem_cache_free(cfq_pool, cfqq); |
1226 | } | |
1227 | ||
d6de8be7 JA |
1228 | /* |
1229 | * Must always be called with the rcu_read_lock() held | |
1230 | */ | |
07416d29 JA |
1231 | static void |
1232 | __call_for_each_cic(struct io_context *ioc, | |
1233 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1234 | { | |
1235 | struct cfq_io_context *cic; | |
1236 | struct hlist_node *n; | |
1237 | ||
1238 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) | |
1239 | func(ioc, cic); | |
1240 | } | |
1241 | ||
4ac845a2 | 1242 | /* |
34e6bbf2 | 1243 | * Call func for each cic attached to this ioc. |
4ac845a2 | 1244 | */ |
34e6bbf2 | 1245 | static void |
4ac845a2 JA |
1246 | call_for_each_cic(struct io_context *ioc, |
1247 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1da177e4 | 1248 | { |
4ac845a2 | 1249 | rcu_read_lock(); |
07416d29 | 1250 | __call_for_each_cic(ioc, func); |
4ac845a2 | 1251 | rcu_read_unlock(); |
34e6bbf2 FC |
1252 | } |
1253 | ||
1254 | static void cfq_cic_free_rcu(struct rcu_head *head) | |
1255 | { | |
1256 | struct cfq_io_context *cic; | |
1257 | ||
1258 | cic = container_of(head, struct cfq_io_context, rcu_head); | |
1259 | ||
1260 | kmem_cache_free(cfq_ioc_pool, cic); | |
1261 | elv_ioc_count_dec(ioc_count); | |
1262 | ||
9a11b4ed JA |
1263 | if (ioc_gone) { |
1264 | /* | |
1265 | * CFQ scheduler is exiting, grab exit lock and check | |
1266 | * the pending io context count. If it hits zero, | |
1267 | * complete ioc_gone and set it back to NULL | |
1268 | */ | |
1269 | spin_lock(&ioc_gone_lock); | |
1270 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) { | |
1271 | complete(ioc_gone); | |
1272 | ioc_gone = NULL; | |
1273 | } | |
1274 | spin_unlock(&ioc_gone_lock); | |
1275 | } | |
34e6bbf2 | 1276 | } |
4ac845a2 | 1277 | |
34e6bbf2 FC |
1278 | static void cfq_cic_free(struct cfq_io_context *cic) |
1279 | { | |
1280 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | |
4ac845a2 JA |
1281 | } |
1282 | ||
1283 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) | |
1284 | { | |
1285 | unsigned long flags; | |
1286 | ||
1287 | BUG_ON(!cic->dead_key); | |
1288 | ||
1289 | spin_lock_irqsave(&ioc->lock, flags); | |
1290 | radix_tree_delete(&ioc->radix_root, cic->dead_key); | |
ffc4e759 | 1291 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1292 | spin_unlock_irqrestore(&ioc->lock, flags); |
1293 | ||
34e6bbf2 | 1294 | cfq_cic_free(cic); |
4ac845a2 JA |
1295 | } |
1296 | ||
d6de8be7 JA |
1297 | /* |
1298 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. | |
1299 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), | |
1300 | * and ->trim() which is called with the task lock held | |
1301 | */ | |
4ac845a2 JA |
1302 | static void cfq_free_io_context(struct io_context *ioc) |
1303 | { | |
4ac845a2 | 1304 | /* |
34e6bbf2 FC |
1305 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
1306 | * so no more cic's are allowed to be linked into this ioc. So it | |
1307 | * should be ok to iterate over the known list, we will see all cic's | |
1308 | * since no new ones are added. | |
4ac845a2 | 1309 | */ |
07416d29 | 1310 | __call_for_each_cic(ioc, cic_free_func); |
1da177e4 LT |
1311 | } |
1312 | ||
89850f7e | 1313 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1314 | { |
28f95cbc | 1315 | if (unlikely(cfqq == cfqd->active_queue)) { |
6084cdda | 1316 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1317 | cfq_schedule_dispatch(cfqd); |
1318 | } | |
22e2c507 | 1319 | |
89850f7e JA |
1320 | cfq_put_queue(cfqq); |
1321 | } | |
22e2c507 | 1322 | |
89850f7e JA |
1323 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1324 | struct cfq_io_context *cic) | |
1325 | { | |
4faa3c81 FC |
1326 | struct io_context *ioc = cic->ioc; |
1327 | ||
fc46379d | 1328 | list_del_init(&cic->queue_list); |
4ac845a2 JA |
1329 | |
1330 | /* | |
1331 | * Make sure key == NULL is seen for dead queues | |
1332 | */ | |
fc46379d | 1333 | smp_wmb(); |
4ac845a2 | 1334 | cic->dead_key = (unsigned long) cic->key; |
fc46379d JA |
1335 | cic->key = NULL; |
1336 | ||
4faa3c81 FC |
1337 | if (ioc->ioc_data == cic) |
1338 | rcu_assign_pointer(ioc->ioc_data, NULL); | |
1339 | ||
12a05732 | 1340 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1341 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1342 | cic->cfqq[ASYNC] = NULL; |
1343 | } | |
1344 | ||
1345 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1346 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1347 | cic->cfqq[SYNC] = NULL; |
1348 | } | |
89850f7e JA |
1349 | } |
1350 | ||
4ac845a2 JA |
1351 | static void cfq_exit_single_io_context(struct io_context *ioc, |
1352 | struct cfq_io_context *cic) | |
89850f7e JA |
1353 | { |
1354 | struct cfq_data *cfqd = cic->key; | |
1355 | ||
89850f7e | 1356 | if (cfqd) { |
165125e1 | 1357 | struct request_queue *q = cfqd->queue; |
4ac845a2 | 1358 | unsigned long flags; |
89850f7e | 1359 | |
4ac845a2 | 1360 | spin_lock_irqsave(q->queue_lock, flags); |
62c1fe9d JA |
1361 | |
1362 | /* | |
1363 | * Ensure we get a fresh copy of the ->key to prevent | |
1364 | * race between exiting task and queue | |
1365 | */ | |
1366 | smp_read_barrier_depends(); | |
1367 | if (cic->key) | |
1368 | __cfq_exit_single_io_context(cfqd, cic); | |
1369 | ||
4ac845a2 | 1370 | spin_unlock_irqrestore(q->queue_lock, flags); |
89850f7e | 1371 | } |
1da177e4 LT |
1372 | } |
1373 | ||
498d3aa2 JA |
1374 | /* |
1375 | * The process that ioc belongs to has exited, we need to clean up | |
1376 | * and put the internal structures we have that belongs to that process. | |
1377 | */ | |
e2d74ac0 | 1378 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1379 | { |
4ac845a2 | 1380 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
1da177e4 LT |
1381 | } |
1382 | ||
22e2c507 | 1383 | static struct cfq_io_context * |
8267e268 | 1384 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1385 | { |
b5deef90 | 1386 | struct cfq_io_context *cic; |
1da177e4 | 1387 | |
94f6030c CL |
1388 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
1389 | cfqd->queue->node); | |
1da177e4 | 1390 | if (cic) { |
22e2c507 | 1391 | cic->last_end_request = jiffies; |
553698f9 | 1392 | INIT_LIST_HEAD(&cic->queue_list); |
ffc4e759 | 1393 | INIT_HLIST_NODE(&cic->cic_list); |
22e2c507 JA |
1394 | cic->dtor = cfq_free_io_context; |
1395 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1396 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1397 | } |
1398 | ||
1399 | return cic; | |
1400 | } | |
1401 | ||
fd0928df | 1402 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
1403 | { |
1404 | struct task_struct *tsk = current; | |
1405 | int ioprio_class; | |
1406 | ||
3b18152c | 1407 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1408 | return; |
1409 | ||
fd0928df | 1410 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 1411 | switch (ioprio_class) { |
fe094d98 JA |
1412 | default: |
1413 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1414 | case IOPRIO_CLASS_NONE: | |
1415 | /* | |
6d63c275 | 1416 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
1417 | */ |
1418 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 1419 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
1420 | break; |
1421 | case IOPRIO_CLASS_RT: | |
1422 | cfqq->ioprio = task_ioprio(ioc); | |
1423 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1424 | break; | |
1425 | case IOPRIO_CLASS_BE: | |
1426 | cfqq->ioprio = task_ioprio(ioc); | |
1427 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1428 | break; | |
1429 | case IOPRIO_CLASS_IDLE: | |
1430 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1431 | cfqq->ioprio = 7; | |
1432 | cfq_clear_cfqq_idle_window(cfqq); | |
1433 | break; | |
22e2c507 JA |
1434 | } |
1435 | ||
1436 | /* | |
1437 | * keep track of original prio settings in case we have to temporarily | |
1438 | * elevate the priority of this queue | |
1439 | */ | |
1440 | cfqq->org_ioprio = cfqq->ioprio; | |
1441 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 1442 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1443 | } |
1444 | ||
febffd61 | 1445 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
22e2c507 | 1446 | { |
478a82b0 AV |
1447 | struct cfq_data *cfqd = cic->key; |
1448 | struct cfq_queue *cfqq; | |
c1b707d2 | 1449 | unsigned long flags; |
35e6077c | 1450 | |
caaa5f9f JA |
1451 | if (unlikely(!cfqd)) |
1452 | return; | |
1453 | ||
c1b707d2 | 1454 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1455 | |
1456 | cfqq = cic->cfqq[ASYNC]; | |
1457 | if (cfqq) { | |
1458 | struct cfq_queue *new_cfqq; | |
fd0928df | 1459 | new_cfqq = cfq_get_queue(cfqd, ASYNC, cic->ioc, GFP_ATOMIC); |
caaa5f9f JA |
1460 | if (new_cfqq) { |
1461 | cic->cfqq[ASYNC] = new_cfqq; | |
1462 | cfq_put_queue(cfqq); | |
1463 | } | |
22e2c507 | 1464 | } |
caaa5f9f JA |
1465 | |
1466 | cfqq = cic->cfqq[SYNC]; | |
1467 | if (cfqq) | |
1468 | cfq_mark_cfqq_prio_changed(cfqq); | |
1469 | ||
c1b707d2 | 1470 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1471 | } |
1472 | ||
fc46379d | 1473 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1474 | { |
4ac845a2 | 1475 | call_for_each_cic(ioc, changed_ioprio); |
fc46379d | 1476 | ioc->ioprio_changed = 0; |
22e2c507 JA |
1477 | } |
1478 | ||
1479 | static struct cfq_queue * | |
15c31be4 | 1480 | cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync, |
fd0928df | 1481 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 1482 | { |
22e2c507 | 1483 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 1484 | struct cfq_io_context *cic; |
22e2c507 JA |
1485 | |
1486 | retry: | |
4ac845a2 | 1487 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
1488 | /* cic always exists here */ |
1489 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 JA |
1490 | |
1491 | if (!cfqq) { | |
1492 | if (new_cfqq) { | |
1493 | cfqq = new_cfqq; | |
1494 | new_cfqq = NULL; | |
1495 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1496 | /* |
1497 | * Inform the allocator of the fact that we will | |
1498 | * just repeat this allocation if it fails, to allow | |
1499 | * the allocator to do whatever it needs to attempt to | |
1500 | * free memory. | |
1501 | */ | |
22e2c507 | 1502 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c CL |
1503 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
1504 | gfp_mask | __GFP_NOFAIL | __GFP_ZERO, | |
1505 | cfqd->queue->node); | |
22e2c507 JA |
1506 | spin_lock_irq(cfqd->queue->queue_lock); |
1507 | goto retry; | |
1508 | } else { | |
94f6030c CL |
1509 | cfqq = kmem_cache_alloc_node(cfq_pool, |
1510 | gfp_mask | __GFP_ZERO, | |
1511 | cfqd->queue->node); | |
22e2c507 JA |
1512 | if (!cfqq) |
1513 | goto out; | |
1514 | } | |
1515 | ||
d9e7620e | 1516 | RB_CLEAR_NODE(&cfqq->rb_node); |
22e2c507 JA |
1517 | INIT_LIST_HEAD(&cfqq->fifo); |
1518 | ||
22e2c507 JA |
1519 | atomic_set(&cfqq->ref, 0); |
1520 | cfqq->cfqd = cfqd; | |
c5b680f3 | 1521 | |
3b18152c | 1522 | cfq_mark_cfqq_prio_changed(cfqq); |
53b03744 | 1523 | cfq_mark_cfqq_queue_new(cfqq); |
91fac317 | 1524 | |
fd0928df | 1525 | cfq_init_prio_data(cfqq, ioc); |
0871714e JA |
1526 | |
1527 | if (is_sync) { | |
1528 | if (!cfq_class_idle(cfqq)) | |
1529 | cfq_mark_cfqq_idle_window(cfqq); | |
1530 | cfq_mark_cfqq_sync(cfqq); | |
1531 | } | |
7b679138 JA |
1532 | cfqq->pid = current->pid; |
1533 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | |
22e2c507 JA |
1534 | } |
1535 | ||
1536 | if (new_cfqq) | |
1537 | kmem_cache_free(cfq_pool, new_cfqq); | |
1538 | ||
22e2c507 JA |
1539 | out: |
1540 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1541 | return cfqq; | |
1542 | } | |
1543 | ||
c2dea2d1 VT |
1544 | static struct cfq_queue ** |
1545 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
1546 | { | |
fe094d98 | 1547 | switch (ioprio_class) { |
c2dea2d1 VT |
1548 | case IOPRIO_CLASS_RT: |
1549 | return &cfqd->async_cfqq[0][ioprio]; | |
1550 | case IOPRIO_CLASS_BE: | |
1551 | return &cfqd->async_cfqq[1][ioprio]; | |
1552 | case IOPRIO_CLASS_IDLE: | |
1553 | return &cfqd->async_idle_cfqq; | |
1554 | default: | |
1555 | BUG(); | |
1556 | } | |
1557 | } | |
1558 | ||
15c31be4 | 1559 | static struct cfq_queue * |
fd0928df | 1560 | cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc, |
15c31be4 JA |
1561 | gfp_t gfp_mask) |
1562 | { | |
fd0928df JA |
1563 | const int ioprio = task_ioprio(ioc); |
1564 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 1565 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
1566 | struct cfq_queue *cfqq = NULL; |
1567 | ||
c2dea2d1 VT |
1568 | if (!is_sync) { |
1569 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
1570 | cfqq = *async_cfqq; | |
1571 | } | |
1572 | ||
0a0836a0 | 1573 | if (!cfqq) { |
fd0928df | 1574 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
0a0836a0 ON |
1575 | if (!cfqq) |
1576 | return NULL; | |
1577 | } | |
15c31be4 JA |
1578 | |
1579 | /* | |
1580 | * pin the queue now that it's allocated, scheduler exit will prune it | |
1581 | */ | |
c2dea2d1 | 1582 | if (!is_sync && !(*async_cfqq)) { |
15c31be4 | 1583 | atomic_inc(&cfqq->ref); |
c2dea2d1 | 1584 | *async_cfqq = cfqq; |
15c31be4 JA |
1585 | } |
1586 | ||
1587 | atomic_inc(&cfqq->ref); | |
1588 | return cfqq; | |
1589 | } | |
1590 | ||
498d3aa2 JA |
1591 | /* |
1592 | * We drop cfq io contexts lazily, so we may find a dead one. | |
1593 | */ | |
dbecf3ab | 1594 | static void |
4ac845a2 JA |
1595 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
1596 | struct cfq_io_context *cic) | |
dbecf3ab | 1597 | { |
4ac845a2 JA |
1598 | unsigned long flags; |
1599 | ||
fc46379d | 1600 | WARN_ON(!list_empty(&cic->queue_list)); |
597bc485 | 1601 | |
4ac845a2 JA |
1602 | spin_lock_irqsave(&ioc->lock, flags); |
1603 | ||
4faa3c81 | 1604 | BUG_ON(ioc->ioc_data == cic); |
597bc485 | 1605 | |
4ac845a2 | 1606 | radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd); |
ffc4e759 | 1607 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1608 | spin_unlock_irqrestore(&ioc->lock, flags); |
1609 | ||
1610 | cfq_cic_free(cic); | |
dbecf3ab OH |
1611 | } |
1612 | ||
e2d74ac0 | 1613 | static struct cfq_io_context * |
4ac845a2 | 1614 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
e2d74ac0 | 1615 | { |
e2d74ac0 | 1616 | struct cfq_io_context *cic; |
d6de8be7 | 1617 | unsigned long flags; |
4ac845a2 | 1618 | void *k; |
e2d74ac0 | 1619 | |
91fac317 VT |
1620 | if (unlikely(!ioc)) |
1621 | return NULL; | |
1622 | ||
d6de8be7 JA |
1623 | rcu_read_lock(); |
1624 | ||
597bc485 JA |
1625 | /* |
1626 | * we maintain a last-hit cache, to avoid browsing over the tree | |
1627 | */ | |
4ac845a2 | 1628 | cic = rcu_dereference(ioc->ioc_data); |
d6de8be7 JA |
1629 | if (cic && cic->key == cfqd) { |
1630 | rcu_read_unlock(); | |
597bc485 | 1631 | return cic; |
d6de8be7 | 1632 | } |
597bc485 | 1633 | |
4ac845a2 | 1634 | do { |
4ac845a2 JA |
1635 | cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd); |
1636 | rcu_read_unlock(); | |
1637 | if (!cic) | |
1638 | break; | |
be3b0753 OH |
1639 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1640 | k = cic->key; | |
1641 | if (unlikely(!k)) { | |
4ac845a2 | 1642 | cfq_drop_dead_cic(cfqd, ioc, cic); |
d6de8be7 | 1643 | rcu_read_lock(); |
4ac845a2 | 1644 | continue; |
dbecf3ab | 1645 | } |
e2d74ac0 | 1646 | |
d6de8be7 | 1647 | spin_lock_irqsave(&ioc->lock, flags); |
4ac845a2 | 1648 | rcu_assign_pointer(ioc->ioc_data, cic); |
d6de8be7 | 1649 | spin_unlock_irqrestore(&ioc->lock, flags); |
4ac845a2 JA |
1650 | break; |
1651 | } while (1); | |
e2d74ac0 | 1652 | |
4ac845a2 | 1653 | return cic; |
e2d74ac0 JA |
1654 | } |
1655 | ||
4ac845a2 JA |
1656 | /* |
1657 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup | |
1658 | * the process specific cfq io context when entered from the block layer. | |
1659 | * Also adds the cic to a per-cfqd list, used when this queue is removed. | |
1660 | */ | |
febffd61 JA |
1661 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
1662 | struct cfq_io_context *cic, gfp_t gfp_mask) | |
e2d74ac0 | 1663 | { |
0261d688 | 1664 | unsigned long flags; |
4ac845a2 | 1665 | int ret; |
e2d74ac0 | 1666 | |
4ac845a2 JA |
1667 | ret = radix_tree_preload(gfp_mask); |
1668 | if (!ret) { | |
1669 | cic->ioc = ioc; | |
1670 | cic->key = cfqd; | |
e2d74ac0 | 1671 | |
4ac845a2 JA |
1672 | spin_lock_irqsave(&ioc->lock, flags); |
1673 | ret = radix_tree_insert(&ioc->radix_root, | |
1674 | (unsigned long) cfqd, cic); | |
ffc4e759 JA |
1675 | if (!ret) |
1676 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); | |
4ac845a2 | 1677 | spin_unlock_irqrestore(&ioc->lock, flags); |
e2d74ac0 | 1678 | |
4ac845a2 JA |
1679 | radix_tree_preload_end(); |
1680 | ||
1681 | if (!ret) { | |
1682 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1683 | list_add(&cic->queue_list, &cfqd->cic_list); | |
1684 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1685 | } | |
e2d74ac0 JA |
1686 | } |
1687 | ||
4ac845a2 JA |
1688 | if (ret) |
1689 | printk(KERN_ERR "cfq: cic link failed!\n"); | |
fc46379d | 1690 | |
4ac845a2 | 1691 | return ret; |
e2d74ac0 JA |
1692 | } |
1693 | ||
1da177e4 LT |
1694 | /* |
1695 | * Setup general io context and cfq io context. There can be several cfq | |
1696 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1697 | * than one device managed by cfq. |
1da177e4 LT |
1698 | */ |
1699 | static struct cfq_io_context * | |
e2d74ac0 | 1700 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1701 | { |
22e2c507 | 1702 | struct io_context *ioc = NULL; |
1da177e4 | 1703 | struct cfq_io_context *cic; |
1da177e4 | 1704 | |
22e2c507 | 1705 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1706 | |
b5deef90 | 1707 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1708 | if (!ioc) |
1709 | return NULL; | |
1710 | ||
4ac845a2 | 1711 | cic = cfq_cic_lookup(cfqd, ioc); |
e2d74ac0 JA |
1712 | if (cic) |
1713 | goto out; | |
1da177e4 | 1714 | |
e2d74ac0 JA |
1715 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1716 | if (cic == NULL) | |
1717 | goto err; | |
1da177e4 | 1718 | |
4ac845a2 JA |
1719 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) |
1720 | goto err_free; | |
1721 | ||
1da177e4 | 1722 | out: |
fc46379d JA |
1723 | smp_read_barrier_depends(); |
1724 | if (unlikely(ioc->ioprio_changed)) | |
1725 | cfq_ioc_set_ioprio(ioc); | |
1726 | ||
1da177e4 | 1727 | return cic; |
4ac845a2 JA |
1728 | err_free: |
1729 | cfq_cic_free(cic); | |
1da177e4 LT |
1730 | err: |
1731 | put_io_context(ioc); | |
1732 | return NULL; | |
1733 | } | |
1734 | ||
22e2c507 JA |
1735 | static void |
1736 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1737 | { |
aaf1228d JA |
1738 | unsigned long elapsed = jiffies - cic->last_end_request; |
1739 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 1740 | |
22e2c507 JA |
1741 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1742 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1743 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1744 | } | |
1da177e4 | 1745 | |
206dc69b | 1746 | static void |
6d048f53 JA |
1747 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic, |
1748 | struct request *rq) | |
206dc69b JA |
1749 | { |
1750 | sector_t sdist; | |
1751 | u64 total; | |
1752 | ||
5e705374 JA |
1753 | if (cic->last_request_pos < rq->sector) |
1754 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1755 | else |
5e705374 | 1756 | sdist = cic->last_request_pos - rq->sector; |
206dc69b JA |
1757 | |
1758 | /* | |
1759 | * Don't allow the seek distance to get too large from the | |
1760 | * odd fragment, pagein, etc | |
1761 | */ | |
1762 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1763 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1764 | else | |
1765 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1766 | ||
1767 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1768 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1769 | total = cic->seek_total + (cic->seek_samples/2); | |
1770 | do_div(total, cic->seek_samples); | |
1771 | cic->seek_mean = (sector_t)total; | |
1772 | } | |
1da177e4 | 1773 | |
22e2c507 JA |
1774 | /* |
1775 | * Disable idle window if the process thinks too long or seeks so much that | |
1776 | * it doesn't matter | |
1777 | */ | |
1778 | static void | |
1779 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1780 | struct cfq_io_context *cic) | |
1781 | { | |
7b679138 | 1782 | int old_idle, enable_idle; |
1be92f2f | 1783 | |
0871714e JA |
1784 | /* |
1785 | * Don't idle for async or idle io prio class | |
1786 | */ | |
1787 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
1788 | return; |
1789 | ||
c265a7f4 | 1790 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 1791 | |
66dac98e | 1792 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
caaa5f9f | 1793 | (cfqd->hw_tag && CIC_SEEKY(cic))) |
22e2c507 JA |
1794 | enable_idle = 0; |
1795 | else if (sample_valid(cic->ttime_samples)) { | |
1796 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1797 | enable_idle = 0; | |
1798 | else | |
1799 | enable_idle = 1; | |
1da177e4 LT |
1800 | } |
1801 | ||
7b679138 JA |
1802 | if (old_idle != enable_idle) { |
1803 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
1804 | if (enable_idle) | |
1805 | cfq_mark_cfqq_idle_window(cfqq); | |
1806 | else | |
1807 | cfq_clear_cfqq_idle_window(cfqq); | |
1808 | } | |
22e2c507 | 1809 | } |
1da177e4 | 1810 | |
22e2c507 JA |
1811 | /* |
1812 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1813 | * no or if we aren't sure, a 1 will cause a preempt. | |
1814 | */ | |
1815 | static int | |
1816 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1817 | struct request *rq) |
22e2c507 | 1818 | { |
6d048f53 | 1819 | struct cfq_queue *cfqq; |
22e2c507 | 1820 | |
6d048f53 JA |
1821 | cfqq = cfqd->active_queue; |
1822 | if (!cfqq) | |
22e2c507 JA |
1823 | return 0; |
1824 | ||
6d048f53 JA |
1825 | if (cfq_slice_used(cfqq)) |
1826 | return 1; | |
1827 | ||
1828 | if (cfq_class_idle(new_cfqq)) | |
caaa5f9f | 1829 | return 0; |
22e2c507 JA |
1830 | |
1831 | if (cfq_class_idle(cfqq)) | |
1832 | return 1; | |
1e3335de | 1833 | |
374f84ac JA |
1834 | /* |
1835 | * if the new request is sync, but the currently running queue is | |
1836 | * not, let the sync request have priority. | |
1837 | */ | |
5e705374 | 1838 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1839 | return 1; |
1e3335de | 1840 | |
374f84ac JA |
1841 | /* |
1842 | * So both queues are sync. Let the new request get disk time if | |
1843 | * it's a metadata request and the current queue is doing regular IO. | |
1844 | */ | |
1845 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1846 | return 1; | |
22e2c507 | 1847 | |
3a9a3f6c DS |
1848 | /* |
1849 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
1850 | */ | |
1851 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
1852 | return 1; | |
1853 | ||
1e3335de JA |
1854 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
1855 | return 0; | |
1856 | ||
1857 | /* | |
1858 | * if this request is as-good as one we would expect from the | |
1859 | * current cfqq, let it preempt | |
1860 | */ | |
6d048f53 | 1861 | if (cfq_rq_close(cfqd, rq)) |
1e3335de JA |
1862 | return 1; |
1863 | ||
22e2c507 JA |
1864 | return 0; |
1865 | } | |
1866 | ||
1867 | /* | |
1868 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1869 | * let it have half of its nominal slice. | |
1870 | */ | |
1871 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1872 | { | |
7b679138 | 1873 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
6084cdda | 1874 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 1875 | |
bf572256 JA |
1876 | /* |
1877 | * Put the new queue at the front of the of the current list, | |
1878 | * so we know that it will be selected next. | |
1879 | */ | |
1880 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
1881 | |
1882 | cfq_service_tree_add(cfqd, cfqq, 1); | |
bf572256 | 1883 | |
44f7c160 JA |
1884 | cfqq->slice_end = 0; |
1885 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
1886 | } |
1887 | ||
22e2c507 | 1888 | /* |
5e705374 | 1889 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1890 | * something we should do about it |
1891 | */ | |
1892 | static void | |
5e705374 JA |
1893 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1894 | struct request *rq) | |
22e2c507 | 1895 | { |
5e705374 | 1896 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1897 | |
45333d5a | 1898 | cfqd->rq_queued++; |
374f84ac JA |
1899 | if (rq_is_meta(rq)) |
1900 | cfqq->meta_pending++; | |
1901 | ||
9c2c38a1 | 1902 | cfq_update_io_thinktime(cfqd, cic); |
6d048f53 | 1903 | cfq_update_io_seektime(cfqd, cic, rq); |
9c2c38a1 JA |
1904 | cfq_update_idle_window(cfqd, cfqq, cic); |
1905 | ||
5e705374 | 1906 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
22e2c507 JA |
1907 | |
1908 | if (cfqq == cfqd->active_queue) { | |
1909 | /* | |
1910 | * if we are waiting for a request for this queue, let it rip | |
1911 | * immediately and flag that we must not expire this queue | |
1912 | * just now | |
1913 | */ | |
3b18152c JA |
1914 | if (cfq_cfqq_wait_request(cfqq)) { |
1915 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 | 1916 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1917 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1918 | } |
5e705374 | 1919 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1920 | /* |
1921 | * not the active queue - expire current slice if it is | |
1922 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
1923 | * has some old slice time left and is of higher priority or |
1924 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
1925 | */ |
1926 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1927 | cfq_mark_cfqq_must_dispatch(cfqq); |
dc72ef4a | 1928 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1929 | } |
1da177e4 LT |
1930 | } |
1931 | ||
165125e1 | 1932 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1933 | { |
b4878f24 | 1934 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1935 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1936 | |
7b679138 | 1937 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
fd0928df | 1938 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
1da177e4 | 1939 | |
5e705374 | 1940 | cfq_add_rq_rb(rq); |
1da177e4 | 1941 | |
22e2c507 JA |
1942 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1943 | ||
5e705374 | 1944 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1945 | } |
1946 | ||
45333d5a AC |
1947 | /* |
1948 | * Update hw_tag based on peak queue depth over 50 samples under | |
1949 | * sufficient load. | |
1950 | */ | |
1951 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
1952 | { | |
1953 | if (cfqd->rq_in_driver > cfqd->rq_in_driver_peak) | |
1954 | cfqd->rq_in_driver_peak = cfqd->rq_in_driver; | |
1955 | ||
1956 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
1957 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) | |
1958 | return; | |
1959 | ||
1960 | if (cfqd->hw_tag_samples++ < 50) | |
1961 | return; | |
1962 | ||
1963 | if (cfqd->rq_in_driver_peak >= CFQ_HW_QUEUE_MIN) | |
1964 | cfqd->hw_tag = 1; | |
1965 | else | |
1966 | cfqd->hw_tag = 0; | |
1967 | ||
1968 | cfqd->hw_tag_samples = 0; | |
1969 | cfqd->rq_in_driver_peak = 0; | |
1970 | } | |
1971 | ||
165125e1 | 1972 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1973 | { |
5e705374 | 1974 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1975 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1976 | const int sync = rq_is_sync(rq); |
b4878f24 | 1977 | unsigned long now; |
1da177e4 | 1978 | |
b4878f24 | 1979 | now = jiffies; |
7b679138 | 1980 | cfq_log_cfqq(cfqd, cfqq, "complete"); |
1da177e4 | 1981 | |
45333d5a AC |
1982 | cfq_update_hw_tag(cfqd); |
1983 | ||
b4878f24 | 1984 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 1985 | WARN_ON(!cfqq->dispatched); |
b4878f24 | 1986 | cfqd->rq_in_driver--; |
6d048f53 | 1987 | cfqq->dispatched--; |
1da177e4 | 1988 | |
3ed9a296 JA |
1989 | if (cfq_cfqq_sync(cfqq)) |
1990 | cfqd->sync_flight--; | |
1991 | ||
b4878f24 JA |
1992 | if (!cfq_class_idle(cfqq)) |
1993 | cfqd->last_end_request = now; | |
3b18152c | 1994 | |
caaa5f9f | 1995 | if (sync) |
5e705374 | 1996 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1997 | |
1998 | /* | |
1999 | * If this is the active queue, check if it needs to be expired, | |
2000 | * or if we want to idle in case it has no pending requests. | |
2001 | */ | |
2002 | if (cfqd->active_queue == cfqq) { | |
44f7c160 JA |
2003 | if (cfq_cfqq_slice_new(cfqq)) { |
2004 | cfq_set_prio_slice(cfqd, cfqq); | |
2005 | cfq_clear_cfqq_slice_new(cfqq); | |
2006 | } | |
0871714e | 2007 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
6084cdda | 2008 | cfq_slice_expired(cfqd, 1); |
aeb6fafb JA |
2009 | else if (sync && !rq_noidle(rq) && |
2010 | RB_EMPTY_ROOT(&cfqq->sort_list)) { | |
6d048f53 | 2011 | cfq_arm_slice_timer(cfqd); |
aeb6fafb | 2012 | } |
caaa5f9f | 2013 | } |
6d048f53 JA |
2014 | |
2015 | if (!cfqd->rq_in_driver) | |
2016 | cfq_schedule_dispatch(cfqd); | |
1da177e4 LT |
2017 | } |
2018 | ||
22e2c507 JA |
2019 | /* |
2020 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
2021 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
2022 | */ | |
2023 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 2024 | { |
22e2c507 JA |
2025 | if (has_fs_excl()) { |
2026 | /* | |
2027 | * boost idle prio on transactions that would lock out other | |
2028 | * users of the filesystem | |
2029 | */ | |
2030 | if (cfq_class_idle(cfqq)) | |
2031 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
2032 | if (cfqq->ioprio > IOPRIO_NORM) | |
2033 | cfqq->ioprio = IOPRIO_NORM; | |
2034 | } else { | |
2035 | /* | |
2036 | * check if we need to unboost the queue | |
2037 | */ | |
2038 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
2039 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
2040 | if (cfqq->ioprio != cfqq->org_ioprio) | |
2041 | cfqq->ioprio = cfqq->org_ioprio; | |
2042 | } | |
22e2c507 | 2043 | } |
1da177e4 | 2044 | |
89850f7e | 2045 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 2046 | { |
3b18152c | 2047 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 2048 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 2049 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 2050 | return ELV_MQUEUE_MUST; |
3b18152c | 2051 | } |
1da177e4 | 2052 | |
22e2c507 | 2053 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
2054 | } |
2055 | ||
165125e1 | 2056 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
2057 | { |
2058 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2059 | struct task_struct *tsk = current; | |
91fac317 | 2060 | struct cfq_io_context *cic; |
22e2c507 JA |
2061 | struct cfq_queue *cfqq; |
2062 | ||
2063 | /* | |
2064 | * don't force setup of a queue from here, as a call to may_queue | |
2065 | * does not necessarily imply that a request actually will be queued. | |
2066 | * so just lookup a possibly existing queue, or return 'may queue' | |
2067 | * if that fails | |
2068 | */ | |
4ac845a2 | 2069 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2070 | if (!cic) |
2071 | return ELV_MQUEUE_MAY; | |
2072 | ||
2073 | cfqq = cic_to_cfqq(cic, rw & REQ_RW_SYNC); | |
22e2c507 | 2074 | if (cfqq) { |
fd0928df | 2075 | cfq_init_prio_data(cfqq, cic->ioc); |
22e2c507 JA |
2076 | cfq_prio_boost(cfqq); |
2077 | ||
89850f7e | 2078 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
2079 | } |
2080 | ||
2081 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
2082 | } |
2083 | ||
1da177e4 LT |
2084 | /* |
2085 | * queue lock held here | |
2086 | */ | |
bb37b94c | 2087 | static void cfq_put_request(struct request *rq) |
1da177e4 | 2088 | { |
5e705374 | 2089 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2090 | |
5e705374 | 2091 | if (cfqq) { |
22e2c507 | 2092 | const int rw = rq_data_dir(rq); |
1da177e4 | 2093 | |
22e2c507 JA |
2094 | BUG_ON(!cfqq->allocated[rw]); |
2095 | cfqq->allocated[rw]--; | |
1da177e4 | 2096 | |
5e705374 | 2097 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 2098 | |
1da177e4 | 2099 | rq->elevator_private = NULL; |
5e705374 | 2100 | rq->elevator_private2 = NULL; |
1da177e4 | 2101 | |
1da177e4 LT |
2102 | cfq_put_queue(cfqq); |
2103 | } | |
2104 | } | |
2105 | ||
2106 | /* | |
22e2c507 | 2107 | * Allocate cfq data structures associated with this request. |
1da177e4 | 2108 | */ |
22e2c507 | 2109 | static int |
165125e1 | 2110 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
2111 | { |
2112 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2113 | struct cfq_io_context *cic; | |
2114 | const int rw = rq_data_dir(rq); | |
7749a8d4 | 2115 | const int is_sync = rq_is_sync(rq); |
22e2c507 | 2116 | struct cfq_queue *cfqq; |
1da177e4 LT |
2117 | unsigned long flags; |
2118 | ||
2119 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2120 | ||
e2d74ac0 | 2121 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 2122 | |
1da177e4 LT |
2123 | spin_lock_irqsave(q->queue_lock, flags); |
2124 | ||
22e2c507 JA |
2125 | if (!cic) |
2126 | goto queue_fail; | |
2127 | ||
91fac317 VT |
2128 | cfqq = cic_to_cfqq(cic, is_sync); |
2129 | if (!cfqq) { | |
fd0928df | 2130 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
91fac317 | 2131 | |
22e2c507 JA |
2132 | if (!cfqq) |
2133 | goto queue_fail; | |
1da177e4 | 2134 | |
91fac317 VT |
2135 | cic_set_cfqq(cic, cfqq, is_sync); |
2136 | } | |
1da177e4 LT |
2137 | |
2138 | cfqq->allocated[rw]++; | |
3b18152c | 2139 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 2140 | atomic_inc(&cfqq->ref); |
1da177e4 | 2141 | |
5e705374 | 2142 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 2143 | |
5e705374 JA |
2144 | rq->elevator_private = cic; |
2145 | rq->elevator_private2 = cfqq; | |
2146 | return 0; | |
1da177e4 | 2147 | |
22e2c507 JA |
2148 | queue_fail: |
2149 | if (cic) | |
2150 | put_io_context(cic->ioc); | |
89850f7e | 2151 | |
3b18152c | 2152 | cfq_schedule_dispatch(cfqd); |
1da177e4 | 2153 | spin_unlock_irqrestore(q->queue_lock, flags); |
7b679138 | 2154 | cfq_log(cfqd, "set_request fail"); |
1da177e4 LT |
2155 | return 1; |
2156 | } | |
2157 | ||
65f27f38 | 2158 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 2159 | { |
65f27f38 DH |
2160 | struct cfq_data *cfqd = |
2161 | container_of(work, struct cfq_data, unplug_work); | |
165125e1 | 2162 | struct request_queue *q = cfqd->queue; |
22e2c507 JA |
2163 | unsigned long flags; |
2164 | ||
2165 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 2166 | blk_start_queueing(q); |
22e2c507 JA |
2167 | spin_unlock_irqrestore(q->queue_lock, flags); |
2168 | } | |
2169 | ||
2170 | /* | |
2171 | * Timer running if the active_queue is currently idling inside its time slice | |
2172 | */ | |
2173 | static void cfq_idle_slice_timer(unsigned long data) | |
2174 | { | |
2175 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
2176 | struct cfq_queue *cfqq; | |
2177 | unsigned long flags; | |
3c6bd2f8 | 2178 | int timed_out = 1; |
22e2c507 | 2179 | |
7b679138 JA |
2180 | cfq_log(cfqd, "idle timer fired"); |
2181 | ||
22e2c507 JA |
2182 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
2183 | ||
fe094d98 JA |
2184 | cfqq = cfqd->active_queue; |
2185 | if (cfqq) { | |
3c6bd2f8 JA |
2186 | timed_out = 0; |
2187 | ||
22e2c507 JA |
2188 | /* |
2189 | * expired | |
2190 | */ | |
44f7c160 | 2191 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
2192 | goto expire; |
2193 | ||
2194 | /* | |
2195 | * only expire and reinvoke request handler, if there are | |
2196 | * other queues with pending requests | |
2197 | */ | |
caaa5f9f | 2198 | if (!cfqd->busy_queues) |
22e2c507 | 2199 | goto out_cont; |
22e2c507 JA |
2200 | |
2201 | /* | |
2202 | * not expired and it has a request pending, let it dispatch | |
2203 | */ | |
dd67d051 | 2204 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 2205 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
2206 | goto out_kick; |
2207 | } | |
2208 | } | |
2209 | expire: | |
6084cdda | 2210 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 2211 | out_kick: |
3b18152c | 2212 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
2213 | out_cont: |
2214 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
2215 | } | |
2216 | ||
3b18152c JA |
2217 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
2218 | { | |
2219 | del_timer_sync(&cfqd->idle_slice_timer); | |
64d01dc9 | 2220 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 2221 | } |
22e2c507 | 2222 | |
c2dea2d1 VT |
2223 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
2224 | { | |
2225 | int i; | |
2226 | ||
2227 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
2228 | if (cfqd->async_cfqq[0][i]) | |
2229 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
2230 | if (cfqd->async_cfqq[1][i]) | |
2231 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 2232 | } |
2389d1ef ON |
2233 | |
2234 | if (cfqd->async_idle_cfqq) | |
2235 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
2236 | } |
2237 | ||
b374d18a | 2238 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 2239 | { |
22e2c507 | 2240 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 2241 | struct request_queue *q = cfqd->queue; |
22e2c507 | 2242 | |
3b18152c | 2243 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 2244 | |
d9ff4187 | 2245 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 2246 | |
d9ff4187 | 2247 | if (cfqd->active_queue) |
6084cdda | 2248 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
2249 | |
2250 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
2251 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
2252 | struct cfq_io_context, | |
2253 | queue_list); | |
89850f7e JA |
2254 | |
2255 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 2256 | } |
e2d74ac0 | 2257 | |
c2dea2d1 | 2258 | cfq_put_async_queues(cfqd); |
15c31be4 | 2259 | |
d9ff4187 | 2260 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
2261 | |
2262 | cfq_shutdown_timer_wq(cfqd); | |
2263 | ||
a90d742e | 2264 | kfree(cfqd); |
1da177e4 LT |
2265 | } |
2266 | ||
165125e1 | 2267 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
2268 | { |
2269 | struct cfq_data *cfqd; | |
1da177e4 | 2270 | |
94f6030c | 2271 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
1da177e4 | 2272 | if (!cfqd) |
bc1c1169 | 2273 | return NULL; |
1da177e4 | 2274 | |
cc09e299 | 2275 | cfqd->service_tree = CFQ_RB_ROOT; |
d9ff4187 | 2276 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2277 | |
1da177e4 | 2278 | cfqd->queue = q; |
1da177e4 | 2279 | |
22e2c507 JA |
2280 | init_timer(&cfqd->idle_slice_timer); |
2281 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2282 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2283 | ||
65f27f38 | 2284 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2285 | |
b70c864d | 2286 | cfqd->last_end_request = jiffies; |
1da177e4 | 2287 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2288 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2289 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2290 | cfqd->cfq_back_max = cfq_back_max; |
2291 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2292 | cfqd->cfq_slice[0] = cfq_slice_async; |
2293 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2294 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2295 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
45333d5a | 2296 | cfqd->hw_tag = 1; |
3b18152c | 2297 | |
bc1c1169 | 2298 | return cfqd; |
1da177e4 LT |
2299 | } |
2300 | ||
2301 | static void cfq_slab_kill(void) | |
2302 | { | |
d6de8be7 JA |
2303 | /* |
2304 | * Caller already ensured that pending RCU callbacks are completed, | |
2305 | * so we should have no busy allocations at this point. | |
2306 | */ | |
1da177e4 LT |
2307 | if (cfq_pool) |
2308 | kmem_cache_destroy(cfq_pool); | |
2309 | if (cfq_ioc_pool) | |
2310 | kmem_cache_destroy(cfq_ioc_pool); | |
2311 | } | |
2312 | ||
2313 | static int __init cfq_slab_setup(void) | |
2314 | { | |
0a31bd5f | 2315 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
1da177e4 LT |
2316 | if (!cfq_pool) |
2317 | goto fail; | |
2318 | ||
34e6bbf2 | 2319 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
1da177e4 LT |
2320 | if (!cfq_ioc_pool) |
2321 | goto fail; | |
2322 | ||
2323 | return 0; | |
2324 | fail: | |
2325 | cfq_slab_kill(); | |
2326 | return -ENOMEM; | |
2327 | } | |
2328 | ||
1da177e4 LT |
2329 | /* |
2330 | * sysfs parts below --> | |
2331 | */ | |
1da177e4 LT |
2332 | static ssize_t |
2333 | cfq_var_show(unsigned int var, char *page) | |
2334 | { | |
2335 | return sprintf(page, "%d\n", var); | |
2336 | } | |
2337 | ||
2338 | static ssize_t | |
2339 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2340 | { | |
2341 | char *p = (char *) page; | |
2342 | ||
2343 | *var = simple_strtoul(p, &p, 10); | |
2344 | return count; | |
2345 | } | |
2346 | ||
1da177e4 | 2347 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 2348 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 2349 | { \ |
3d1ab40f | 2350 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2351 | unsigned int __data = __VAR; \ |
2352 | if (__CONV) \ | |
2353 | __data = jiffies_to_msecs(__data); \ | |
2354 | return cfq_var_show(__data, (page)); \ | |
2355 | } | |
2356 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2357 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2358 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2359 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2360 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2361 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2362 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2363 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2364 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2365 | #undef SHOW_FUNCTION |
2366 | ||
2367 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 2368 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 2369 | { \ |
3d1ab40f | 2370 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2371 | unsigned int __data; \ |
2372 | int ret = cfq_var_store(&__data, (page), count); \ | |
2373 | if (__data < (MIN)) \ | |
2374 | __data = (MIN); \ | |
2375 | else if (__data > (MAX)) \ | |
2376 | __data = (MAX); \ | |
2377 | if (__CONV) \ | |
2378 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2379 | else \ | |
2380 | *(__PTR) = __data; \ | |
2381 | return ret; \ | |
2382 | } | |
2383 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
2384 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
2385 | UINT_MAX, 1); | |
2386 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
2387 | UINT_MAX, 1); | |
e572ec7e | 2388 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
2389 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
2390 | UINT_MAX, 0); | |
22e2c507 JA |
2391 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2392 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2393 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
2394 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
2395 | UINT_MAX, 0); | |
1da177e4 LT |
2396 | #undef STORE_FUNCTION |
2397 | ||
e572ec7e AV |
2398 | #define CFQ_ATTR(name) \ |
2399 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2400 | ||
2401 | static struct elv_fs_entry cfq_attrs[] = { | |
2402 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2403 | CFQ_ATTR(fifo_expire_sync), |
2404 | CFQ_ATTR(fifo_expire_async), | |
2405 | CFQ_ATTR(back_seek_max), | |
2406 | CFQ_ATTR(back_seek_penalty), | |
2407 | CFQ_ATTR(slice_sync), | |
2408 | CFQ_ATTR(slice_async), | |
2409 | CFQ_ATTR(slice_async_rq), | |
2410 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2411 | __ATTR_NULL |
1da177e4 LT |
2412 | }; |
2413 | ||
1da177e4 LT |
2414 | static struct elevator_type iosched_cfq = { |
2415 | .ops = { | |
2416 | .elevator_merge_fn = cfq_merge, | |
2417 | .elevator_merged_fn = cfq_merged_request, | |
2418 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2419 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2420 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2421 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2422 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2423 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2424 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2425 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2426 | .elevator_former_req_fn = elv_rb_former_request, |
2427 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2428 | .elevator_set_req_fn = cfq_set_request, |
2429 | .elevator_put_req_fn = cfq_put_request, | |
2430 | .elevator_may_queue_fn = cfq_may_queue, | |
2431 | .elevator_init_fn = cfq_init_queue, | |
2432 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2433 | .trim = cfq_free_io_context, |
1da177e4 | 2434 | }, |
3d1ab40f | 2435 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2436 | .elevator_name = "cfq", |
2437 | .elevator_owner = THIS_MODULE, | |
2438 | }; | |
2439 | ||
2440 | static int __init cfq_init(void) | |
2441 | { | |
22e2c507 JA |
2442 | /* |
2443 | * could be 0 on HZ < 1000 setups | |
2444 | */ | |
2445 | if (!cfq_slice_async) | |
2446 | cfq_slice_async = 1; | |
2447 | if (!cfq_slice_idle) | |
2448 | cfq_slice_idle = 1; | |
2449 | ||
1da177e4 LT |
2450 | if (cfq_slab_setup()) |
2451 | return -ENOMEM; | |
2452 | ||
2fdd82bd | 2453 | elv_register(&iosched_cfq); |
1da177e4 | 2454 | |
2fdd82bd | 2455 | return 0; |
1da177e4 LT |
2456 | } |
2457 | ||
2458 | static void __exit cfq_exit(void) | |
2459 | { | |
6e9a4738 | 2460 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2461 | elv_unregister(&iosched_cfq); |
334e94de | 2462 | ioc_gone = &all_gone; |
fba82272 OH |
2463 | /* ioc_gone's update must be visible before reading ioc_count */ |
2464 | smp_wmb(); | |
d6de8be7 JA |
2465 | |
2466 | /* | |
2467 | * this also protects us from entering cfq_slab_kill() with | |
2468 | * pending RCU callbacks | |
2469 | */ | |
4050cf16 | 2470 | if (elv_ioc_count_read(ioc_count)) |
9a11b4ed | 2471 | wait_for_completion(&all_gone); |
83521d3e | 2472 | cfq_slab_kill(); |
1da177e4 LT |
2473 | } |
2474 | ||
2475 | module_init(cfq_init); | |
2476 | module_exit(cfq_exit); | |
2477 | ||
2478 | MODULE_AUTHOR("Jens Axboe"); | |
2479 | MODULE_LICENSE("GPL"); | |
2480 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |