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; |
1da177e4 | 163 | |
be754d2c RK |
164 | /* pending metadata requests */ |
165 | int meta_pending; | |
6d048f53 JA |
166 | /* number of requests that are on the dispatch list or inside driver */ |
167 | int dispatched; | |
22e2c507 JA |
168 | |
169 | /* io prio of this group */ | |
170 | unsigned short ioprio, org_ioprio; | |
171 | unsigned short ioprio_class, org_ioprio_class; | |
172 | ||
7b679138 | 173 | pid_t pid; |
1da177e4 LT |
174 | }; |
175 | ||
3b18152c | 176 | enum cfqq_state_flags { |
b0b8d749 JA |
177 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
178 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
179 | CFQ_CFQQ_FLAG_must_alloc, /* must be allowed rq alloc */ | |
180 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ | |
181 | CFQ_CFQQ_FLAG_must_dispatch, /* must dispatch, even if expired */ | |
182 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ | |
183 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
184 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
185 | CFQ_CFQQ_FLAG_queue_new, /* queue never been serviced */ | |
44f7c160 | 186 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 187 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
3b18152c JA |
188 | }; |
189 | ||
190 | #define CFQ_CFQQ_FNS(name) \ | |
191 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
192 | { \ | |
fe094d98 | 193 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
194 | } \ |
195 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
196 | { \ | |
fe094d98 | 197 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
198 | } \ |
199 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
200 | { \ | |
fe094d98 | 201 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
202 | } |
203 | ||
204 | CFQ_CFQQ_FNS(on_rr); | |
205 | CFQ_CFQQ_FNS(wait_request); | |
206 | CFQ_CFQQ_FNS(must_alloc); | |
207 | CFQ_CFQQ_FNS(must_alloc_slice); | |
208 | CFQ_CFQQ_FNS(must_dispatch); | |
209 | CFQ_CFQQ_FNS(fifo_expire); | |
210 | CFQ_CFQQ_FNS(idle_window); | |
211 | CFQ_CFQQ_FNS(prio_changed); | |
53b03744 | 212 | CFQ_CFQQ_FNS(queue_new); |
44f7c160 | 213 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 214 | CFQ_CFQQ_FNS(sync); |
3b18152c JA |
215 | #undef CFQ_CFQQ_FNS |
216 | ||
7b679138 JA |
217 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
218 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
219 | #define cfq_log(cfqd, fmt, args...) \ | |
220 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
221 | ||
165125e1 | 222 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
91fac317 | 223 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, int, |
fd0928df | 224 | struct io_context *, gfp_t); |
4ac845a2 | 225 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
91fac317 VT |
226 | struct io_context *); |
227 | ||
228 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, | |
229 | int is_sync) | |
230 | { | |
231 | return cic->cfqq[!!is_sync]; | |
232 | } | |
233 | ||
234 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
235 | struct cfq_queue *cfqq, int is_sync) | |
236 | { | |
237 | cic->cfqq[!!is_sync] = cfqq; | |
238 | } | |
239 | ||
240 | /* | |
241 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
242 | * set (in which case it could also be direct WRITE). | |
243 | */ | |
244 | static inline int cfq_bio_sync(struct bio *bio) | |
245 | { | |
246 | if (bio_data_dir(bio) == READ || bio_sync(bio)) | |
247 | return 1; | |
248 | ||
249 | return 0; | |
250 | } | |
1da177e4 | 251 | |
99f95e52 AM |
252 | /* |
253 | * scheduler run of queue, if there are requests pending and no one in the | |
254 | * driver that will restart queueing | |
255 | */ | |
256 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) | |
257 | { | |
7b679138 JA |
258 | if (cfqd->busy_queues) { |
259 | cfq_log(cfqd, "schedule dispatch"); | |
18887ad9 | 260 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 261 | } |
99f95e52 AM |
262 | } |
263 | ||
165125e1 | 264 | static int cfq_queue_empty(struct request_queue *q) |
99f95e52 AM |
265 | { |
266 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
267 | ||
b4878f24 | 268 | return !cfqd->busy_queues; |
99f95e52 AM |
269 | } |
270 | ||
44f7c160 JA |
271 | /* |
272 | * Scale schedule slice based on io priority. Use the sync time slice only | |
273 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
274 | * io only, should not get full sync slice length. | |
275 | */ | |
d9e7620e JA |
276 | static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync, |
277 | unsigned short prio) | |
44f7c160 | 278 | { |
d9e7620e | 279 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 280 | |
d9e7620e JA |
281 | WARN_ON(prio >= IOPRIO_BE_NR); |
282 | ||
283 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
284 | } | |
44f7c160 | 285 | |
d9e7620e JA |
286 | static inline int |
287 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
288 | { | |
289 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
290 | } |
291 | ||
292 | static inline void | |
293 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
294 | { | |
295 | cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; | |
7b679138 | 296 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
297 | } |
298 | ||
299 | /* | |
300 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
301 | * isn't valid until the first request from the dispatch is activated | |
302 | * and the slice time set. | |
303 | */ | |
304 | static inline int cfq_slice_used(struct cfq_queue *cfqq) | |
305 | { | |
306 | if (cfq_cfqq_slice_new(cfqq)) | |
307 | return 0; | |
308 | if (time_before(jiffies, cfqq->slice_end)) | |
309 | return 0; | |
310 | ||
311 | return 1; | |
312 | } | |
313 | ||
1da177e4 | 314 | /* |
5e705374 | 315 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 316 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 317 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 318 | */ |
5e705374 JA |
319 | static struct request * |
320 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2) | |
1da177e4 LT |
321 | { |
322 | sector_t last, s1, s2, d1 = 0, d2 = 0; | |
1da177e4 | 323 | unsigned long back_max; |
e8a99053 AM |
324 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
325 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
326 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 327 | |
5e705374 JA |
328 | if (rq1 == NULL || rq1 == rq2) |
329 | return rq2; | |
330 | if (rq2 == NULL) | |
331 | return rq1; | |
9c2c38a1 | 332 | |
5e705374 JA |
333 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
334 | return rq1; | |
335 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
336 | return rq2; | |
374f84ac JA |
337 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
338 | return rq1; | |
339 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
340 | return rq2; | |
1da177e4 | 341 | |
5e705374 JA |
342 | s1 = rq1->sector; |
343 | s2 = rq2->sector; | |
1da177e4 | 344 | |
6d048f53 | 345 | last = cfqd->last_position; |
1da177e4 | 346 | |
1da177e4 LT |
347 | /* |
348 | * by definition, 1KiB is 2 sectors | |
349 | */ | |
350 | back_max = cfqd->cfq_back_max * 2; | |
351 | ||
352 | /* | |
353 | * Strict one way elevator _except_ in the case where we allow | |
354 | * short backward seeks which are biased as twice the cost of a | |
355 | * similar forward seek. | |
356 | */ | |
357 | if (s1 >= last) | |
358 | d1 = s1 - last; | |
359 | else if (s1 + back_max >= last) | |
360 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
361 | else | |
e8a99053 | 362 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
363 | |
364 | if (s2 >= last) | |
365 | d2 = s2 - last; | |
366 | else if (s2 + back_max >= last) | |
367 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
368 | else | |
e8a99053 | 369 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
370 | |
371 | /* Found required data */ | |
e8a99053 AM |
372 | |
373 | /* | |
374 | * By doing switch() on the bit mask "wrap" we avoid having to | |
375 | * check two variables for all permutations: --> faster! | |
376 | */ | |
377 | switch (wrap) { | |
5e705374 | 378 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 379 | if (d1 < d2) |
5e705374 | 380 | return rq1; |
e8a99053 | 381 | else if (d2 < d1) |
5e705374 | 382 | return rq2; |
e8a99053 AM |
383 | else { |
384 | if (s1 >= s2) | |
5e705374 | 385 | return rq1; |
e8a99053 | 386 | else |
5e705374 | 387 | return rq2; |
e8a99053 | 388 | } |
1da177e4 | 389 | |
e8a99053 | 390 | case CFQ_RQ2_WRAP: |
5e705374 | 391 | return rq1; |
e8a99053 | 392 | case CFQ_RQ1_WRAP: |
5e705374 JA |
393 | return rq2; |
394 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
395 | default: |
396 | /* | |
397 | * Since both rqs are wrapped, | |
398 | * start with the one that's further behind head | |
399 | * (--> only *one* back seek required), | |
400 | * since back seek takes more time than forward. | |
401 | */ | |
402 | if (s1 <= s2) | |
5e705374 | 403 | return rq1; |
1da177e4 | 404 | else |
5e705374 | 405 | return rq2; |
1da177e4 LT |
406 | } |
407 | } | |
408 | ||
498d3aa2 JA |
409 | /* |
410 | * The below is leftmost cache rbtree addon | |
411 | */ | |
0871714e | 412 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 JA |
413 | { |
414 | if (!root->left) | |
415 | root->left = rb_first(&root->rb); | |
416 | ||
0871714e JA |
417 | if (root->left) |
418 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
419 | ||
420 | return NULL; | |
cc09e299 JA |
421 | } |
422 | ||
423 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) | |
424 | { | |
425 | if (root->left == n) | |
426 | root->left = NULL; | |
427 | ||
428 | rb_erase(n, &root->rb); | |
429 | RB_CLEAR_NODE(n); | |
430 | } | |
431 | ||
1da177e4 LT |
432 | /* |
433 | * would be nice to take fifo expire time into account as well | |
434 | */ | |
5e705374 JA |
435 | static struct request * |
436 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
437 | struct request *last) | |
1da177e4 | 438 | { |
21183b07 JA |
439 | struct rb_node *rbnext = rb_next(&last->rb_node); |
440 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 441 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 442 | |
21183b07 | 443 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
444 | |
445 | if (rbprev) | |
5e705374 | 446 | prev = rb_entry_rq(rbprev); |
1da177e4 | 447 | |
21183b07 | 448 | if (rbnext) |
5e705374 | 449 | next = rb_entry_rq(rbnext); |
21183b07 JA |
450 | else { |
451 | rbnext = rb_first(&cfqq->sort_list); | |
452 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 453 | next = rb_entry_rq(rbnext); |
21183b07 | 454 | } |
1da177e4 | 455 | |
21183b07 | 456 | return cfq_choose_req(cfqd, next, prev); |
1da177e4 LT |
457 | } |
458 | ||
d9e7620e JA |
459 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
460 | struct cfq_queue *cfqq) | |
1da177e4 | 461 | { |
d9e7620e JA |
462 | /* |
463 | * just an approximation, should be ok. | |
464 | */ | |
67e6b49e JA |
465 | return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
466 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); | |
d9e7620e JA |
467 | } |
468 | ||
498d3aa2 JA |
469 | /* |
470 | * The cfqd->service_tree holds all pending cfq_queue's that have | |
471 | * requests waiting to be processed. It is sorted in the order that | |
472 | * we will service the queues. | |
473 | */ | |
d9e7620e | 474 | static void cfq_service_tree_add(struct cfq_data *cfqd, |
edd75ffd | 475 | struct cfq_queue *cfqq, int add_front) |
d9e7620e | 476 | { |
0871714e JA |
477 | struct rb_node **p, *parent; |
478 | struct cfq_queue *__cfqq; | |
d9e7620e | 479 | unsigned long rb_key; |
498d3aa2 | 480 | int left; |
d9e7620e | 481 | |
0871714e JA |
482 | if (cfq_class_idle(cfqq)) { |
483 | rb_key = CFQ_IDLE_DELAY; | |
484 | parent = rb_last(&cfqd->service_tree.rb); | |
485 | if (parent && parent != &cfqq->rb_node) { | |
486 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
487 | rb_key += __cfqq->rb_key; | |
488 | } else | |
489 | rb_key += jiffies; | |
490 | } else if (!add_front) { | |
edd75ffd JA |
491 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
492 | rb_key += cfqq->slice_resid; | |
493 | cfqq->slice_resid = 0; | |
494 | } else | |
495 | rb_key = 0; | |
1da177e4 | 496 | |
d9e7620e | 497 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
99f9628a | 498 | /* |
d9e7620e | 499 | * same position, nothing more to do |
99f9628a | 500 | */ |
d9e7620e JA |
501 | if (rb_key == cfqq->rb_key) |
502 | return; | |
1da177e4 | 503 | |
cc09e299 | 504 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); |
1da177e4 | 505 | } |
d9e7620e | 506 | |
498d3aa2 | 507 | left = 1; |
0871714e JA |
508 | parent = NULL; |
509 | p = &cfqd->service_tree.rb.rb_node; | |
d9e7620e | 510 | while (*p) { |
67060e37 | 511 | struct rb_node **n; |
cc09e299 | 512 | |
d9e7620e JA |
513 | parent = *p; |
514 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
515 | ||
0c534e0a JA |
516 | /* |
517 | * sort RT queues first, we always want to give | |
67060e37 JA |
518 | * preference to them. IDLE queues goes to the back. |
519 | * after that, sort on the next service time. | |
0c534e0a JA |
520 | */ |
521 | if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq)) | |
67060e37 | 522 | n = &(*p)->rb_left; |
0c534e0a | 523 | else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq)) |
67060e37 JA |
524 | n = &(*p)->rb_right; |
525 | else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq)) | |
526 | n = &(*p)->rb_left; | |
527 | else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq)) | |
528 | n = &(*p)->rb_right; | |
0c534e0a | 529 | else if (rb_key < __cfqq->rb_key) |
67060e37 JA |
530 | n = &(*p)->rb_left; |
531 | else | |
532 | n = &(*p)->rb_right; | |
533 | ||
534 | if (n == &(*p)->rb_right) | |
cc09e299 | 535 | left = 0; |
67060e37 JA |
536 | |
537 | p = n; | |
d9e7620e JA |
538 | } |
539 | ||
cc09e299 JA |
540 | if (left) |
541 | cfqd->service_tree.left = &cfqq->rb_node; | |
542 | ||
d9e7620e JA |
543 | cfqq->rb_key = rb_key; |
544 | rb_link_node(&cfqq->rb_node, parent, p); | |
cc09e299 | 545 | rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb); |
1da177e4 LT |
546 | } |
547 | ||
498d3aa2 JA |
548 | /* |
549 | * Update cfqq's position in the service tree. | |
550 | */ | |
edd75ffd | 551 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 552 | { |
6d048f53 JA |
553 | /* |
554 | * Resorting requires the cfqq to be on the RR list already. | |
555 | */ | |
498d3aa2 | 556 | if (cfq_cfqq_on_rr(cfqq)) |
edd75ffd | 557 | cfq_service_tree_add(cfqd, cfqq, 0); |
6d048f53 JA |
558 | } |
559 | ||
1da177e4 LT |
560 | /* |
561 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 562 | * the pending list according to last request service |
1da177e4 | 563 | */ |
febffd61 | 564 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 565 | { |
7b679138 | 566 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
567 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
568 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 569 | cfqd->busy_queues++; |
3a9a3f6c DS |
570 | if (cfq_class_rt(cfqq)) |
571 | cfqd->busy_rt_queues++; | |
1da177e4 | 572 | |
edd75ffd | 573 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
574 | } |
575 | ||
498d3aa2 JA |
576 | /* |
577 | * Called when the cfqq no longer has requests pending, remove it from | |
578 | * the service tree. | |
579 | */ | |
febffd61 | 580 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 581 | { |
7b679138 | 582 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
583 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
584 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 585 | |
cc09e299 JA |
586 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) |
587 | cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree); | |
d9e7620e | 588 | |
1da177e4 LT |
589 | BUG_ON(!cfqd->busy_queues); |
590 | cfqd->busy_queues--; | |
3a9a3f6c DS |
591 | if (cfq_class_rt(cfqq)) |
592 | cfqd->busy_rt_queues--; | |
1da177e4 LT |
593 | } |
594 | ||
595 | /* | |
596 | * rb tree support functions | |
597 | */ | |
febffd61 | 598 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 599 | { |
5e705374 | 600 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 601 | struct cfq_data *cfqd = cfqq->cfqd; |
5e705374 | 602 | const int sync = rq_is_sync(rq); |
1da177e4 | 603 | |
b4878f24 JA |
604 | BUG_ON(!cfqq->queued[sync]); |
605 | cfqq->queued[sync]--; | |
1da177e4 | 606 | |
5e705374 | 607 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 608 | |
dd67d051 | 609 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
b4878f24 | 610 | cfq_del_cfqq_rr(cfqd, cfqq); |
1da177e4 LT |
611 | } |
612 | ||
5e705374 | 613 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 614 | { |
5e705374 | 615 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 616 | struct cfq_data *cfqd = cfqq->cfqd; |
21183b07 | 617 | struct request *__alias; |
1da177e4 | 618 | |
5380a101 | 619 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
620 | |
621 | /* | |
622 | * looks a little odd, but the first insert might return an alias. | |
623 | * if that happens, put the alias on the dispatch list | |
624 | */ | |
21183b07 | 625 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 626 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
627 | |
628 | if (!cfq_cfqq_on_rr(cfqq)) | |
629 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
630 | |
631 | /* | |
632 | * check if this request is a better next-serve candidate | |
633 | */ | |
634 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq); | |
635 | BUG_ON(!cfqq->next_rq); | |
1da177e4 LT |
636 | } |
637 | ||
febffd61 | 638 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 639 | { |
5380a101 JA |
640 | elv_rb_del(&cfqq->sort_list, rq); |
641 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 642 | cfq_add_rq_rb(rq); |
1da177e4 LT |
643 | } |
644 | ||
206dc69b JA |
645 | static struct request * |
646 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 647 | { |
206dc69b | 648 | struct task_struct *tsk = current; |
91fac317 | 649 | struct cfq_io_context *cic; |
206dc69b | 650 | struct cfq_queue *cfqq; |
1da177e4 | 651 | |
4ac845a2 | 652 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
653 | if (!cic) |
654 | return NULL; | |
655 | ||
656 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
657 | if (cfqq) { |
658 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
659 | ||
21183b07 | 660 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 661 | } |
1da177e4 | 662 | |
1da177e4 LT |
663 | return NULL; |
664 | } | |
665 | ||
165125e1 | 666 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 667 | { |
22e2c507 | 668 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 669 | |
b4878f24 | 670 | cfqd->rq_in_driver++; |
7b679138 JA |
671 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
672 | cfqd->rq_in_driver); | |
25776e35 | 673 | |
6d048f53 | 674 | cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors; |
1da177e4 LT |
675 | } |
676 | ||
165125e1 | 677 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 678 | { |
b4878f24 JA |
679 | struct cfq_data *cfqd = q->elevator->elevator_data; |
680 | ||
681 | WARN_ON(!cfqd->rq_in_driver); | |
682 | cfqd->rq_in_driver--; | |
7b679138 JA |
683 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
684 | cfqd->rq_in_driver); | |
1da177e4 LT |
685 | } |
686 | ||
b4878f24 | 687 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 688 | { |
5e705374 | 689 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 690 | |
5e705374 JA |
691 | if (cfqq->next_rq == rq) |
692 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 693 | |
b4878f24 | 694 | list_del_init(&rq->queuelist); |
5e705374 | 695 | cfq_del_rq_rb(rq); |
374f84ac | 696 | |
45333d5a | 697 | cfqq->cfqd->rq_queued--; |
374f84ac JA |
698 | if (rq_is_meta(rq)) { |
699 | WARN_ON(!cfqq->meta_pending); | |
700 | cfqq->meta_pending--; | |
701 | } | |
1da177e4 LT |
702 | } |
703 | ||
165125e1 JA |
704 | static int cfq_merge(struct request_queue *q, struct request **req, |
705 | struct bio *bio) | |
1da177e4 LT |
706 | { |
707 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
708 | struct request *__rq; | |
1da177e4 | 709 | |
206dc69b | 710 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 711 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
712 | *req = __rq; |
713 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
714 | } |
715 | ||
716 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
717 | } |
718 | ||
165125e1 | 719 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 720 | int type) |
1da177e4 | 721 | { |
21183b07 | 722 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 723 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 724 | |
5e705374 | 725 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 726 | } |
1da177e4 LT |
727 | } |
728 | ||
729 | static void | |
165125e1 | 730 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
731 | struct request *next) |
732 | { | |
22e2c507 JA |
733 | /* |
734 | * reposition in fifo if next is older than rq | |
735 | */ | |
736 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
737 | time_before(next->start_time, rq->start_time)) | |
738 | list_move(&rq->queuelist, &next->queuelist); | |
739 | ||
b4878f24 | 740 | cfq_remove_request(next); |
22e2c507 JA |
741 | } |
742 | ||
165125e1 | 743 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
744 | struct bio *bio) |
745 | { | |
746 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
91fac317 | 747 | struct cfq_io_context *cic; |
da775265 | 748 | struct cfq_queue *cfqq; |
da775265 JA |
749 | |
750 | /* | |
ec8acb69 | 751 | * Disallow merge of a sync bio into an async request. |
da775265 | 752 | */ |
91fac317 | 753 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
da775265 JA |
754 | return 0; |
755 | ||
756 | /* | |
719d3402 JA |
757 | * Lookup the cfqq that this bio will be queued with. Allow |
758 | * merge only if rq is queued there. | |
da775265 | 759 | */ |
4ac845a2 | 760 | cic = cfq_cic_lookup(cfqd, current->io_context); |
91fac317 VT |
761 | if (!cic) |
762 | return 0; | |
719d3402 | 763 | |
91fac317 | 764 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
719d3402 JA |
765 | if (cfqq == RQ_CFQQ(rq)) |
766 | return 1; | |
da775265 | 767 | |
ec8acb69 | 768 | return 0; |
da775265 JA |
769 | } |
770 | ||
febffd61 JA |
771 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
772 | struct cfq_queue *cfqq) | |
22e2c507 JA |
773 | { |
774 | if (cfqq) { | |
7b679138 | 775 | cfq_log_cfqq(cfqd, cfqq, "set_active"); |
22e2c507 | 776 | cfqq->slice_end = 0; |
3b18152c JA |
777 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
778 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 779 | cfq_mark_cfqq_slice_new(cfqq); |
1afba045 | 780 | cfq_clear_cfqq_queue_new(cfqq); |
22e2c507 JA |
781 | } |
782 | ||
783 | cfqd->active_queue = cfqq; | |
784 | } | |
785 | ||
7b14e3b5 JA |
786 | /* |
787 | * current cfqq expired its slice (or was too idle), select new one | |
788 | */ | |
789 | static void | |
790 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
6084cdda | 791 | int timed_out) |
7b14e3b5 | 792 | { |
7b679138 JA |
793 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
794 | ||
7b14e3b5 JA |
795 | if (cfq_cfqq_wait_request(cfqq)) |
796 | del_timer(&cfqd->idle_slice_timer); | |
797 | ||
7b14e3b5 JA |
798 | cfq_clear_cfqq_must_dispatch(cfqq); |
799 | cfq_clear_cfqq_wait_request(cfqq); | |
800 | ||
801 | /* | |
6084cdda | 802 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 803 | */ |
7b679138 | 804 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) { |
c5b680f3 | 805 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b679138 JA |
806 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
807 | } | |
7b14e3b5 | 808 | |
edd75ffd | 809 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
810 | |
811 | if (cfqq == cfqd->active_queue) | |
812 | cfqd->active_queue = NULL; | |
813 | ||
814 | if (cfqd->active_cic) { | |
815 | put_io_context(cfqd->active_cic->ioc); | |
816 | cfqd->active_cic = NULL; | |
817 | } | |
7b14e3b5 JA |
818 | } |
819 | ||
6084cdda | 820 | static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out) |
7b14e3b5 JA |
821 | { |
822 | struct cfq_queue *cfqq = cfqd->active_queue; | |
823 | ||
824 | if (cfqq) | |
6084cdda | 825 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
826 | } |
827 | ||
498d3aa2 JA |
828 | /* |
829 | * Get next queue for service. Unless we have a queue preemption, | |
830 | * we'll simply select the first cfqq in the service tree. | |
831 | */ | |
6d048f53 | 832 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 833 | { |
edd75ffd JA |
834 | if (RB_EMPTY_ROOT(&cfqd->service_tree.rb)) |
835 | return NULL; | |
d9e7620e | 836 | |
0871714e | 837 | return cfq_rb_first(&cfqd->service_tree); |
6d048f53 JA |
838 | } |
839 | ||
498d3aa2 JA |
840 | /* |
841 | * Get and set a new active queue for service. | |
842 | */ | |
6d048f53 JA |
843 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) |
844 | { | |
845 | struct cfq_queue *cfqq; | |
846 | ||
d9e7620e | 847 | cfqq = cfq_get_next_queue(cfqd); |
22e2c507 | 848 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 849 | return cfqq; |
22e2c507 JA |
850 | } |
851 | ||
d9e7620e JA |
852 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
853 | struct request *rq) | |
854 | { | |
855 | if (rq->sector >= cfqd->last_position) | |
856 | return rq->sector - cfqd->last_position; | |
857 | else | |
858 | return cfqd->last_position - rq->sector; | |
859 | } | |
860 | ||
6d048f53 JA |
861 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq) |
862 | { | |
863 | struct cfq_io_context *cic = cfqd->active_cic; | |
864 | ||
865 | if (!sample_valid(cic->seek_samples)) | |
866 | return 0; | |
867 | ||
868 | return cfq_dist_from_last(cfqd, rq) <= cic->seek_mean; | |
869 | } | |
870 | ||
d9e7620e JA |
871 | static int cfq_close_cooperator(struct cfq_data *cfq_data, |
872 | struct cfq_queue *cfqq) | |
6d048f53 | 873 | { |
6d048f53 | 874 | /* |
d9e7620e JA |
875 | * We should notice if some of the queues are cooperating, eg |
876 | * working closely on the same area of the disk. In that case, | |
877 | * we can group them together and don't waste time idling. | |
6d048f53 | 878 | */ |
d9e7620e | 879 | return 0; |
6d048f53 JA |
880 | } |
881 | ||
882 | #define CIC_SEEKY(cic) ((cic)->seek_mean > (8 * 1024)) | |
caaa5f9f | 883 | |
6d048f53 | 884 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 885 | { |
1792669c | 886 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 887 | struct cfq_io_context *cic; |
7b14e3b5 JA |
888 | unsigned long sl; |
889 | ||
a68bbddb | 890 | /* |
f7d7b7a7 JA |
891 | * SSD device without seek penalty, disable idling. But only do so |
892 | * for devices that support queuing, otherwise we still have a problem | |
893 | * with sync vs async workloads. | |
a68bbddb | 894 | */ |
f7d7b7a7 | 895 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
896 | return; |
897 | ||
dd67d051 | 898 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 899 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
900 | |
901 | /* | |
902 | * idle is disabled, either manually or by past process history | |
903 | */ | |
6d048f53 JA |
904 | if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq)) |
905 | return; | |
906 | ||
7b679138 JA |
907 | /* |
908 | * still requests with the driver, don't idle | |
909 | */ | |
910 | if (cfqd->rq_in_driver) | |
911 | return; | |
912 | ||
22e2c507 JA |
913 | /* |
914 | * task has exited, don't wait | |
915 | */ | |
206dc69b | 916 | cic = cfqd->active_cic; |
66dac98e | 917 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
6d048f53 JA |
918 | return; |
919 | ||
920 | /* | |
921 | * See if this prio level has a good candidate | |
922 | */ | |
1afba045 JA |
923 | if (cfq_close_cooperator(cfqd, cfqq) && |
924 | (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2)) | |
6d048f53 | 925 | return; |
22e2c507 | 926 | |
3b18152c JA |
927 | cfq_mark_cfqq_must_dispatch(cfqq); |
928 | cfq_mark_cfqq_wait_request(cfqq); | |
22e2c507 | 929 | |
206dc69b JA |
930 | /* |
931 | * we don't want to idle for seeks, but we do want to allow | |
932 | * fair distribution of slice time for a process doing back-to-back | |
933 | * seeks. so allow a little bit of time for him to submit a new rq | |
934 | */ | |
6d048f53 | 935 | sl = cfqd->cfq_slice_idle; |
caaa5f9f | 936 | if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic)) |
d9e7620e | 937 | sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT)); |
206dc69b | 938 | |
7b14e3b5 | 939 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
7b679138 | 940 | cfq_log(cfqd, "arm_idle: %lu", sl); |
1da177e4 LT |
941 | } |
942 | ||
498d3aa2 JA |
943 | /* |
944 | * Move request from internal lists to the request queue dispatch list. | |
945 | */ | |
165125e1 | 946 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 947 | { |
3ed9a296 | 948 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 949 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 950 | |
7b679138 JA |
951 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
952 | ||
5380a101 | 953 | cfq_remove_request(rq); |
6d048f53 | 954 | cfqq->dispatched++; |
5380a101 | 955 | elv_dispatch_sort(q, rq); |
3ed9a296 JA |
956 | |
957 | if (cfq_cfqq_sync(cfqq)) | |
958 | cfqd->sync_flight++; | |
1da177e4 LT |
959 | } |
960 | ||
961 | /* | |
962 | * return expired entry, or NULL to just start from scratch in rbtree | |
963 | */ | |
febffd61 | 964 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 LT |
965 | { |
966 | struct cfq_data *cfqd = cfqq->cfqd; | |
22e2c507 | 967 | struct request *rq; |
89850f7e | 968 | int fifo; |
1da177e4 | 969 | |
3b18152c | 970 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 971 | return NULL; |
cb887411 JA |
972 | |
973 | cfq_mark_cfqq_fifo_expire(cfqq); | |
974 | ||
89850f7e JA |
975 | if (list_empty(&cfqq->fifo)) |
976 | return NULL; | |
1da177e4 | 977 | |
6d048f53 | 978 | fifo = cfq_cfqq_sync(cfqq); |
89850f7e | 979 | rq = rq_entry_fifo(cfqq->fifo.next); |
1da177e4 | 980 | |
6d048f53 | 981 | if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) |
7b679138 | 982 | rq = NULL; |
1da177e4 | 983 | |
7b679138 | 984 | cfq_log_cfqq(cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 985 | return rq; |
1da177e4 LT |
986 | } |
987 | ||
22e2c507 JA |
988 | static inline int |
989 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
990 | { | |
991 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 992 | |
22e2c507 | 993 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 994 | |
22e2c507 | 995 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
996 | } |
997 | ||
22e2c507 | 998 | /* |
498d3aa2 JA |
999 | * Select a queue for service. If we have a current active queue, |
1000 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 1001 | */ |
1b5ed5e1 | 1002 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 1003 | { |
1da177e4 | 1004 | struct cfq_queue *cfqq; |
1da177e4 | 1005 | |
22e2c507 JA |
1006 | cfqq = cfqd->active_queue; |
1007 | if (!cfqq) | |
1008 | goto new_queue; | |
1da177e4 | 1009 | |
22e2c507 | 1010 | /* |
6d048f53 | 1011 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 1012 | */ |
6d048f53 | 1013 | if (cfq_slice_used(cfqq)) |
3b18152c | 1014 | goto expire; |
1da177e4 | 1015 | |
3a9a3f6c DS |
1016 | /* |
1017 | * If we have a RT cfqq waiting, then we pre-empt the current non-rt | |
1018 | * cfqq. | |
1019 | */ | |
1020 | if (!cfq_class_rt(cfqq) && cfqd->busy_rt_queues) { | |
1021 | /* | |
1022 | * We simulate this as cfqq timed out so that it gets to bank | |
1023 | * the remaining of its time slice. | |
1024 | */ | |
1025 | cfq_log_cfqq(cfqd, cfqq, "preempt"); | |
1026 | cfq_slice_expired(cfqd, 1); | |
1027 | goto new_queue; | |
1028 | } | |
1029 | ||
22e2c507 | 1030 | /* |
6d048f53 JA |
1031 | * The active queue has requests and isn't expired, allow it to |
1032 | * dispatch. | |
22e2c507 | 1033 | */ |
dd67d051 | 1034 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 1035 | goto keep_queue; |
6d048f53 JA |
1036 | |
1037 | /* | |
1038 | * No requests pending. If the active queue still has requests in | |
1039 | * flight or is idling for a new request, allow either of these | |
1040 | * conditions to happen (or time out) before selecting a new queue. | |
1041 | */ | |
cc197479 JA |
1042 | if (timer_pending(&cfqd->idle_slice_timer) || |
1043 | (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) { | |
caaa5f9f JA |
1044 | cfqq = NULL; |
1045 | goto keep_queue; | |
22e2c507 JA |
1046 | } |
1047 | ||
3b18152c | 1048 | expire: |
6084cdda | 1049 | cfq_slice_expired(cfqd, 0); |
3b18152c JA |
1050 | new_queue: |
1051 | cfqq = cfq_set_active_queue(cfqd); | |
22e2c507 | 1052 | keep_queue: |
3b18152c | 1053 | return cfqq; |
22e2c507 JA |
1054 | } |
1055 | ||
498d3aa2 JA |
1056 | /* |
1057 | * Dispatch some requests from cfqq, moving them to the request queue | |
1058 | * dispatch list. | |
1059 | */ | |
22e2c507 JA |
1060 | static int |
1061 | __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1062 | int max_dispatch) | |
1063 | { | |
1064 | int dispatched = 0; | |
1065 | ||
dd67d051 | 1066 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
22e2c507 JA |
1067 | |
1068 | do { | |
5e705374 | 1069 | struct request *rq; |
1da177e4 LT |
1070 | |
1071 | /* | |
22e2c507 | 1072 | * follow expired path, else get first next available |
1da177e4 | 1073 | */ |
fe094d98 JA |
1074 | rq = cfq_check_fifo(cfqq); |
1075 | if (rq == NULL) | |
5e705374 | 1076 | rq = cfqq->next_rq; |
22e2c507 JA |
1077 | |
1078 | /* | |
1079 | * finally, insert request into driver dispatch list | |
1080 | */ | |
5e705374 | 1081 | cfq_dispatch_insert(cfqd->queue, rq); |
1da177e4 | 1082 | |
22e2c507 | 1083 | dispatched++; |
1da177e4 | 1084 | |
22e2c507 | 1085 | if (!cfqd->active_cic) { |
5e705374 JA |
1086 | atomic_inc(&RQ_CIC(rq)->ioc->refcount); |
1087 | cfqd->active_cic = RQ_CIC(rq); | |
22e2c507 | 1088 | } |
1da177e4 | 1089 | |
dd67d051 | 1090 | if (RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 JA |
1091 | break; |
1092 | ||
3a9a3f6c DS |
1093 | /* |
1094 | * If there is a non-empty RT cfqq waiting for current | |
1095 | * cfqq's timeslice to complete, pre-empt this cfqq | |
1096 | */ | |
1097 | if (!cfq_class_rt(cfqq) && cfqd->busy_rt_queues) | |
1098 | break; | |
1099 | ||
22e2c507 JA |
1100 | } while (dispatched < max_dispatch); |
1101 | ||
22e2c507 JA |
1102 | /* |
1103 | * expire an async queue immediately if it has used up its slice. idle | |
1104 | * queue always expire after 1 dispatch round. | |
1105 | */ | |
a9938006 | 1106 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
20e493a8 | 1107 | dispatched >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
a9938006 | 1108 | cfq_class_idle(cfqq))) { |
44f7c160 | 1109 | cfqq->slice_end = jiffies + 1; |
6084cdda | 1110 | cfq_slice_expired(cfqd, 0); |
44f7c160 | 1111 | } |
22e2c507 JA |
1112 | |
1113 | return dispatched; | |
1114 | } | |
1115 | ||
febffd61 | 1116 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
1117 | { |
1118 | int dispatched = 0; | |
1119 | ||
1120 | while (cfqq->next_rq) { | |
1121 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
1122 | dispatched++; | |
1123 | } | |
1124 | ||
1125 | BUG_ON(!list_empty(&cfqq->fifo)); | |
1126 | return dispatched; | |
1127 | } | |
1128 | ||
498d3aa2 JA |
1129 | /* |
1130 | * Drain our current requests. Used for barriers and when switching | |
1131 | * io schedulers on-the-fly. | |
1132 | */ | |
d9e7620e | 1133 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 1134 | { |
0871714e | 1135 | struct cfq_queue *cfqq; |
d9e7620e | 1136 | int dispatched = 0; |
1b5ed5e1 | 1137 | |
0871714e | 1138 | while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL) |
d9e7620e | 1139 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
1b5ed5e1 | 1140 | |
6084cdda | 1141 | cfq_slice_expired(cfqd, 0); |
1b5ed5e1 TH |
1142 | |
1143 | BUG_ON(cfqd->busy_queues); | |
1144 | ||
7b679138 | 1145 | cfq_log(cfqd, "forced_dispatch=%d\n", dispatched); |
1b5ed5e1 TH |
1146 | return dispatched; |
1147 | } | |
1148 | ||
165125e1 | 1149 | static int cfq_dispatch_requests(struct request_queue *q, int force) |
22e2c507 JA |
1150 | { |
1151 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
6d048f53 | 1152 | struct cfq_queue *cfqq; |
caaa5f9f | 1153 | int dispatched; |
22e2c507 JA |
1154 | |
1155 | if (!cfqd->busy_queues) | |
1156 | return 0; | |
1157 | ||
1b5ed5e1 TH |
1158 | if (unlikely(force)) |
1159 | return cfq_forced_dispatch(cfqd); | |
1160 | ||
caaa5f9f | 1161 | dispatched = 0; |
caaa5f9f | 1162 | while ((cfqq = cfq_select_queue(cfqd)) != NULL) { |
b4878f24 JA |
1163 | int max_dispatch; |
1164 | ||
3ed9a296 JA |
1165 | max_dispatch = cfqd->cfq_quantum; |
1166 | if (cfq_class_idle(cfqq)) | |
1167 | max_dispatch = 1; | |
1168 | ||
30e0dc28 JA |
1169 | if (cfqq->dispatched >= max_dispatch && cfqd->busy_queues > 1) |
1170 | break; | |
9ede209e | 1171 | |
3ed9a296 JA |
1172 | if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq)) |
1173 | break; | |
1174 | ||
3b18152c JA |
1175 | cfq_clear_cfqq_must_dispatch(cfqq); |
1176 | cfq_clear_cfqq_wait_request(cfqq); | |
22e2c507 JA |
1177 | del_timer(&cfqd->idle_slice_timer); |
1178 | ||
caaa5f9f | 1179 | dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); |
1da177e4 LT |
1180 | } |
1181 | ||
7b679138 | 1182 | cfq_log(cfqd, "dispatched=%d", dispatched); |
caaa5f9f | 1183 | return dispatched; |
1da177e4 LT |
1184 | } |
1185 | ||
1da177e4 | 1186 | /* |
5e705374 JA |
1187 | * task holds one reference to the queue, dropped when task exits. each rq |
1188 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 LT |
1189 | * |
1190 | * queue lock must be held here. | |
1191 | */ | |
1192 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
1193 | { | |
22e2c507 JA |
1194 | struct cfq_data *cfqd = cfqq->cfqd; |
1195 | ||
1196 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
1197 | |
1198 | if (!atomic_dec_and_test(&cfqq->ref)) | |
1199 | return; | |
1200 | ||
7b679138 | 1201 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 1202 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 1203 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
3b18152c | 1204 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 1205 | |
28f95cbc | 1206 | if (unlikely(cfqd->active_queue == cfqq)) { |
6084cdda | 1207 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1208 | cfq_schedule_dispatch(cfqd); |
1209 | } | |
22e2c507 | 1210 | |
1da177e4 LT |
1211 | kmem_cache_free(cfq_pool, cfqq); |
1212 | } | |
1213 | ||
d6de8be7 JA |
1214 | /* |
1215 | * Must always be called with the rcu_read_lock() held | |
1216 | */ | |
07416d29 JA |
1217 | static void |
1218 | __call_for_each_cic(struct io_context *ioc, | |
1219 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1220 | { | |
1221 | struct cfq_io_context *cic; | |
1222 | struct hlist_node *n; | |
1223 | ||
1224 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) | |
1225 | func(ioc, cic); | |
1226 | } | |
1227 | ||
4ac845a2 | 1228 | /* |
34e6bbf2 | 1229 | * Call func for each cic attached to this ioc. |
4ac845a2 | 1230 | */ |
34e6bbf2 | 1231 | static void |
4ac845a2 JA |
1232 | call_for_each_cic(struct io_context *ioc, |
1233 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1da177e4 | 1234 | { |
4ac845a2 | 1235 | rcu_read_lock(); |
07416d29 | 1236 | __call_for_each_cic(ioc, func); |
4ac845a2 | 1237 | rcu_read_unlock(); |
34e6bbf2 FC |
1238 | } |
1239 | ||
1240 | static void cfq_cic_free_rcu(struct rcu_head *head) | |
1241 | { | |
1242 | struct cfq_io_context *cic; | |
1243 | ||
1244 | cic = container_of(head, struct cfq_io_context, rcu_head); | |
1245 | ||
1246 | kmem_cache_free(cfq_ioc_pool, cic); | |
1247 | elv_ioc_count_dec(ioc_count); | |
1248 | ||
9a11b4ed JA |
1249 | if (ioc_gone) { |
1250 | /* | |
1251 | * CFQ scheduler is exiting, grab exit lock and check | |
1252 | * the pending io context count. If it hits zero, | |
1253 | * complete ioc_gone and set it back to NULL | |
1254 | */ | |
1255 | spin_lock(&ioc_gone_lock); | |
1256 | if (ioc_gone && !elv_ioc_count_read(ioc_count)) { | |
1257 | complete(ioc_gone); | |
1258 | ioc_gone = NULL; | |
1259 | } | |
1260 | spin_unlock(&ioc_gone_lock); | |
1261 | } | |
34e6bbf2 | 1262 | } |
4ac845a2 | 1263 | |
34e6bbf2 FC |
1264 | static void cfq_cic_free(struct cfq_io_context *cic) |
1265 | { | |
1266 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | |
4ac845a2 JA |
1267 | } |
1268 | ||
1269 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) | |
1270 | { | |
1271 | unsigned long flags; | |
1272 | ||
1273 | BUG_ON(!cic->dead_key); | |
1274 | ||
1275 | spin_lock_irqsave(&ioc->lock, flags); | |
1276 | radix_tree_delete(&ioc->radix_root, cic->dead_key); | |
ffc4e759 | 1277 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1278 | spin_unlock_irqrestore(&ioc->lock, flags); |
1279 | ||
34e6bbf2 | 1280 | cfq_cic_free(cic); |
4ac845a2 JA |
1281 | } |
1282 | ||
d6de8be7 JA |
1283 | /* |
1284 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. | |
1285 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), | |
1286 | * and ->trim() which is called with the task lock held | |
1287 | */ | |
4ac845a2 JA |
1288 | static void cfq_free_io_context(struct io_context *ioc) |
1289 | { | |
4ac845a2 | 1290 | /* |
34e6bbf2 FC |
1291 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
1292 | * so no more cic's are allowed to be linked into this ioc. So it | |
1293 | * should be ok to iterate over the known list, we will see all cic's | |
1294 | * since no new ones are added. | |
4ac845a2 | 1295 | */ |
07416d29 | 1296 | __call_for_each_cic(ioc, cic_free_func); |
1da177e4 LT |
1297 | } |
1298 | ||
89850f7e | 1299 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1300 | { |
28f95cbc | 1301 | if (unlikely(cfqq == cfqd->active_queue)) { |
6084cdda | 1302 | __cfq_slice_expired(cfqd, cfqq, 0); |
28f95cbc JA |
1303 | cfq_schedule_dispatch(cfqd); |
1304 | } | |
22e2c507 | 1305 | |
89850f7e JA |
1306 | cfq_put_queue(cfqq); |
1307 | } | |
22e2c507 | 1308 | |
89850f7e JA |
1309 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
1310 | struct cfq_io_context *cic) | |
1311 | { | |
4faa3c81 FC |
1312 | struct io_context *ioc = cic->ioc; |
1313 | ||
fc46379d | 1314 | list_del_init(&cic->queue_list); |
4ac845a2 JA |
1315 | |
1316 | /* | |
1317 | * Make sure key == NULL is seen for dead queues | |
1318 | */ | |
fc46379d | 1319 | smp_wmb(); |
4ac845a2 | 1320 | cic->dead_key = (unsigned long) cic->key; |
fc46379d JA |
1321 | cic->key = NULL; |
1322 | ||
4faa3c81 FC |
1323 | if (ioc->ioc_data == cic) |
1324 | rcu_assign_pointer(ioc->ioc_data, NULL); | |
1325 | ||
12a05732 | 1326 | if (cic->cfqq[ASYNC]) { |
89850f7e | 1327 | cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]); |
12a05732 AV |
1328 | cic->cfqq[ASYNC] = NULL; |
1329 | } | |
1330 | ||
1331 | if (cic->cfqq[SYNC]) { | |
89850f7e | 1332 | cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]); |
12a05732 AV |
1333 | cic->cfqq[SYNC] = NULL; |
1334 | } | |
89850f7e JA |
1335 | } |
1336 | ||
4ac845a2 JA |
1337 | static void cfq_exit_single_io_context(struct io_context *ioc, |
1338 | struct cfq_io_context *cic) | |
89850f7e JA |
1339 | { |
1340 | struct cfq_data *cfqd = cic->key; | |
1341 | ||
89850f7e | 1342 | if (cfqd) { |
165125e1 | 1343 | struct request_queue *q = cfqd->queue; |
4ac845a2 | 1344 | unsigned long flags; |
89850f7e | 1345 | |
4ac845a2 | 1346 | spin_lock_irqsave(q->queue_lock, flags); |
62c1fe9d JA |
1347 | |
1348 | /* | |
1349 | * Ensure we get a fresh copy of the ->key to prevent | |
1350 | * race between exiting task and queue | |
1351 | */ | |
1352 | smp_read_barrier_depends(); | |
1353 | if (cic->key) | |
1354 | __cfq_exit_single_io_context(cfqd, cic); | |
1355 | ||
4ac845a2 | 1356 | spin_unlock_irqrestore(q->queue_lock, flags); |
89850f7e | 1357 | } |
1da177e4 LT |
1358 | } |
1359 | ||
498d3aa2 JA |
1360 | /* |
1361 | * The process that ioc belongs to has exited, we need to clean up | |
1362 | * and put the internal structures we have that belongs to that process. | |
1363 | */ | |
e2d74ac0 | 1364 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 1365 | { |
4ac845a2 | 1366 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
1da177e4 LT |
1367 | } |
1368 | ||
22e2c507 | 1369 | static struct cfq_io_context * |
8267e268 | 1370 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1371 | { |
b5deef90 | 1372 | struct cfq_io_context *cic; |
1da177e4 | 1373 | |
94f6030c CL |
1374 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
1375 | cfqd->queue->node); | |
1da177e4 | 1376 | if (cic) { |
22e2c507 | 1377 | cic->last_end_request = jiffies; |
553698f9 | 1378 | INIT_LIST_HEAD(&cic->queue_list); |
ffc4e759 | 1379 | INIT_HLIST_NODE(&cic->cic_list); |
22e2c507 JA |
1380 | cic->dtor = cfq_free_io_context; |
1381 | cic->exit = cfq_exit_io_context; | |
4050cf16 | 1382 | elv_ioc_count_inc(ioc_count); |
1da177e4 LT |
1383 | } |
1384 | ||
1385 | return cic; | |
1386 | } | |
1387 | ||
fd0928df | 1388 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
1389 | { |
1390 | struct task_struct *tsk = current; | |
1391 | int ioprio_class; | |
1392 | ||
3b18152c | 1393 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
1394 | return; |
1395 | ||
fd0928df | 1396 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 1397 | switch (ioprio_class) { |
fe094d98 JA |
1398 | default: |
1399 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
1400 | case IOPRIO_CLASS_NONE: | |
1401 | /* | |
6d63c275 | 1402 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
1403 | */ |
1404 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 1405 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
1406 | break; |
1407 | case IOPRIO_CLASS_RT: | |
1408 | cfqq->ioprio = task_ioprio(ioc); | |
1409 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
1410 | break; | |
1411 | case IOPRIO_CLASS_BE: | |
1412 | cfqq->ioprio = task_ioprio(ioc); | |
1413 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
1414 | break; | |
1415 | case IOPRIO_CLASS_IDLE: | |
1416 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
1417 | cfqq->ioprio = 7; | |
1418 | cfq_clear_cfqq_idle_window(cfqq); | |
1419 | break; | |
22e2c507 JA |
1420 | } |
1421 | ||
1422 | /* | |
1423 | * keep track of original prio settings in case we have to temporarily | |
1424 | * elevate the priority of this queue | |
1425 | */ | |
1426 | cfqq->org_ioprio = cfqq->ioprio; | |
1427 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 1428 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
1429 | } |
1430 | ||
febffd61 | 1431 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
22e2c507 | 1432 | { |
478a82b0 AV |
1433 | struct cfq_data *cfqd = cic->key; |
1434 | struct cfq_queue *cfqq; | |
c1b707d2 | 1435 | unsigned long flags; |
35e6077c | 1436 | |
caaa5f9f JA |
1437 | if (unlikely(!cfqd)) |
1438 | return; | |
1439 | ||
c1b707d2 | 1440 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f JA |
1441 | |
1442 | cfqq = cic->cfqq[ASYNC]; | |
1443 | if (cfqq) { | |
1444 | struct cfq_queue *new_cfqq; | |
fd0928df | 1445 | new_cfqq = cfq_get_queue(cfqd, ASYNC, cic->ioc, GFP_ATOMIC); |
caaa5f9f JA |
1446 | if (new_cfqq) { |
1447 | cic->cfqq[ASYNC] = new_cfqq; | |
1448 | cfq_put_queue(cfqq); | |
1449 | } | |
22e2c507 | 1450 | } |
caaa5f9f JA |
1451 | |
1452 | cfqq = cic->cfqq[SYNC]; | |
1453 | if (cfqq) | |
1454 | cfq_mark_cfqq_prio_changed(cfqq); | |
1455 | ||
c1b707d2 | 1456 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
1457 | } |
1458 | ||
fc46379d | 1459 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 1460 | { |
4ac845a2 | 1461 | call_for_each_cic(ioc, changed_ioprio); |
fc46379d | 1462 | ioc->ioprio_changed = 0; |
22e2c507 JA |
1463 | } |
1464 | ||
1465 | static struct cfq_queue * | |
15c31be4 | 1466 | cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync, |
fd0928df | 1467 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 1468 | { |
22e2c507 | 1469 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 1470 | struct cfq_io_context *cic; |
22e2c507 JA |
1471 | |
1472 | retry: | |
4ac845a2 | 1473 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
1474 | /* cic always exists here */ |
1475 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 JA |
1476 | |
1477 | if (!cfqq) { | |
1478 | if (new_cfqq) { | |
1479 | cfqq = new_cfqq; | |
1480 | new_cfqq = NULL; | |
1481 | } else if (gfp_mask & __GFP_WAIT) { | |
89850f7e JA |
1482 | /* |
1483 | * Inform the allocator of the fact that we will | |
1484 | * just repeat this allocation if it fails, to allow | |
1485 | * the allocator to do whatever it needs to attempt to | |
1486 | * free memory. | |
1487 | */ | |
22e2c507 | 1488 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c CL |
1489 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
1490 | gfp_mask | __GFP_NOFAIL | __GFP_ZERO, | |
1491 | cfqd->queue->node); | |
22e2c507 JA |
1492 | spin_lock_irq(cfqd->queue->queue_lock); |
1493 | goto retry; | |
1494 | } else { | |
94f6030c CL |
1495 | cfqq = kmem_cache_alloc_node(cfq_pool, |
1496 | gfp_mask | __GFP_ZERO, | |
1497 | cfqd->queue->node); | |
22e2c507 JA |
1498 | if (!cfqq) |
1499 | goto out; | |
1500 | } | |
1501 | ||
d9e7620e | 1502 | RB_CLEAR_NODE(&cfqq->rb_node); |
22e2c507 JA |
1503 | INIT_LIST_HEAD(&cfqq->fifo); |
1504 | ||
22e2c507 JA |
1505 | atomic_set(&cfqq->ref, 0); |
1506 | cfqq->cfqd = cfqd; | |
c5b680f3 | 1507 | |
3b18152c | 1508 | cfq_mark_cfqq_prio_changed(cfqq); |
53b03744 | 1509 | cfq_mark_cfqq_queue_new(cfqq); |
91fac317 | 1510 | |
fd0928df | 1511 | cfq_init_prio_data(cfqq, ioc); |
0871714e JA |
1512 | |
1513 | if (is_sync) { | |
1514 | if (!cfq_class_idle(cfqq)) | |
1515 | cfq_mark_cfqq_idle_window(cfqq); | |
1516 | cfq_mark_cfqq_sync(cfqq); | |
1517 | } | |
7b679138 JA |
1518 | cfqq->pid = current->pid; |
1519 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | |
22e2c507 JA |
1520 | } |
1521 | ||
1522 | if (new_cfqq) | |
1523 | kmem_cache_free(cfq_pool, new_cfqq); | |
1524 | ||
22e2c507 JA |
1525 | out: |
1526 | WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); | |
1527 | return cfqq; | |
1528 | } | |
1529 | ||
c2dea2d1 VT |
1530 | static struct cfq_queue ** |
1531 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
1532 | { | |
fe094d98 | 1533 | switch (ioprio_class) { |
c2dea2d1 VT |
1534 | case IOPRIO_CLASS_RT: |
1535 | return &cfqd->async_cfqq[0][ioprio]; | |
1536 | case IOPRIO_CLASS_BE: | |
1537 | return &cfqd->async_cfqq[1][ioprio]; | |
1538 | case IOPRIO_CLASS_IDLE: | |
1539 | return &cfqd->async_idle_cfqq; | |
1540 | default: | |
1541 | BUG(); | |
1542 | } | |
1543 | } | |
1544 | ||
15c31be4 | 1545 | static struct cfq_queue * |
fd0928df | 1546 | cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc, |
15c31be4 JA |
1547 | gfp_t gfp_mask) |
1548 | { | |
fd0928df JA |
1549 | const int ioprio = task_ioprio(ioc); |
1550 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 1551 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
1552 | struct cfq_queue *cfqq = NULL; |
1553 | ||
c2dea2d1 VT |
1554 | if (!is_sync) { |
1555 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
1556 | cfqq = *async_cfqq; | |
1557 | } | |
1558 | ||
0a0836a0 | 1559 | if (!cfqq) { |
fd0928df | 1560 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
0a0836a0 ON |
1561 | if (!cfqq) |
1562 | return NULL; | |
1563 | } | |
15c31be4 JA |
1564 | |
1565 | /* | |
1566 | * pin the queue now that it's allocated, scheduler exit will prune it | |
1567 | */ | |
c2dea2d1 | 1568 | if (!is_sync && !(*async_cfqq)) { |
15c31be4 | 1569 | atomic_inc(&cfqq->ref); |
c2dea2d1 | 1570 | *async_cfqq = cfqq; |
15c31be4 JA |
1571 | } |
1572 | ||
1573 | atomic_inc(&cfqq->ref); | |
1574 | return cfqq; | |
1575 | } | |
1576 | ||
498d3aa2 JA |
1577 | /* |
1578 | * We drop cfq io contexts lazily, so we may find a dead one. | |
1579 | */ | |
dbecf3ab | 1580 | static void |
4ac845a2 JA |
1581 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
1582 | struct cfq_io_context *cic) | |
dbecf3ab | 1583 | { |
4ac845a2 JA |
1584 | unsigned long flags; |
1585 | ||
fc46379d | 1586 | WARN_ON(!list_empty(&cic->queue_list)); |
597bc485 | 1587 | |
4ac845a2 JA |
1588 | spin_lock_irqsave(&ioc->lock, flags); |
1589 | ||
4faa3c81 | 1590 | BUG_ON(ioc->ioc_data == cic); |
597bc485 | 1591 | |
4ac845a2 | 1592 | radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd); |
ffc4e759 | 1593 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
1594 | spin_unlock_irqrestore(&ioc->lock, flags); |
1595 | ||
1596 | cfq_cic_free(cic); | |
dbecf3ab OH |
1597 | } |
1598 | ||
e2d74ac0 | 1599 | static struct cfq_io_context * |
4ac845a2 | 1600 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
e2d74ac0 | 1601 | { |
e2d74ac0 | 1602 | struct cfq_io_context *cic; |
d6de8be7 | 1603 | unsigned long flags; |
4ac845a2 | 1604 | void *k; |
e2d74ac0 | 1605 | |
91fac317 VT |
1606 | if (unlikely(!ioc)) |
1607 | return NULL; | |
1608 | ||
d6de8be7 JA |
1609 | rcu_read_lock(); |
1610 | ||
597bc485 JA |
1611 | /* |
1612 | * we maintain a last-hit cache, to avoid browsing over the tree | |
1613 | */ | |
4ac845a2 | 1614 | cic = rcu_dereference(ioc->ioc_data); |
d6de8be7 JA |
1615 | if (cic && cic->key == cfqd) { |
1616 | rcu_read_unlock(); | |
597bc485 | 1617 | return cic; |
d6de8be7 | 1618 | } |
597bc485 | 1619 | |
4ac845a2 | 1620 | do { |
4ac845a2 JA |
1621 | cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd); |
1622 | rcu_read_unlock(); | |
1623 | if (!cic) | |
1624 | break; | |
be3b0753 OH |
1625 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
1626 | k = cic->key; | |
1627 | if (unlikely(!k)) { | |
4ac845a2 | 1628 | cfq_drop_dead_cic(cfqd, ioc, cic); |
d6de8be7 | 1629 | rcu_read_lock(); |
4ac845a2 | 1630 | continue; |
dbecf3ab | 1631 | } |
e2d74ac0 | 1632 | |
d6de8be7 | 1633 | spin_lock_irqsave(&ioc->lock, flags); |
4ac845a2 | 1634 | rcu_assign_pointer(ioc->ioc_data, cic); |
d6de8be7 | 1635 | spin_unlock_irqrestore(&ioc->lock, flags); |
4ac845a2 JA |
1636 | break; |
1637 | } while (1); | |
e2d74ac0 | 1638 | |
4ac845a2 | 1639 | return cic; |
e2d74ac0 JA |
1640 | } |
1641 | ||
4ac845a2 JA |
1642 | /* |
1643 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup | |
1644 | * the process specific cfq io context when entered from the block layer. | |
1645 | * Also adds the cic to a per-cfqd list, used when this queue is removed. | |
1646 | */ | |
febffd61 JA |
1647 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
1648 | struct cfq_io_context *cic, gfp_t gfp_mask) | |
e2d74ac0 | 1649 | { |
0261d688 | 1650 | unsigned long flags; |
4ac845a2 | 1651 | int ret; |
e2d74ac0 | 1652 | |
4ac845a2 JA |
1653 | ret = radix_tree_preload(gfp_mask); |
1654 | if (!ret) { | |
1655 | cic->ioc = ioc; | |
1656 | cic->key = cfqd; | |
e2d74ac0 | 1657 | |
4ac845a2 JA |
1658 | spin_lock_irqsave(&ioc->lock, flags); |
1659 | ret = radix_tree_insert(&ioc->radix_root, | |
1660 | (unsigned long) cfqd, cic); | |
ffc4e759 JA |
1661 | if (!ret) |
1662 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); | |
4ac845a2 | 1663 | spin_unlock_irqrestore(&ioc->lock, flags); |
e2d74ac0 | 1664 | |
4ac845a2 JA |
1665 | radix_tree_preload_end(); |
1666 | ||
1667 | if (!ret) { | |
1668 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1669 | list_add(&cic->queue_list, &cfqd->cic_list); | |
1670 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1671 | } | |
e2d74ac0 JA |
1672 | } |
1673 | ||
4ac845a2 JA |
1674 | if (ret) |
1675 | printk(KERN_ERR "cfq: cic link failed!\n"); | |
fc46379d | 1676 | |
4ac845a2 | 1677 | return ret; |
e2d74ac0 JA |
1678 | } |
1679 | ||
1da177e4 LT |
1680 | /* |
1681 | * Setup general io context and cfq io context. There can be several cfq | |
1682 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 1683 | * than one device managed by cfq. |
1da177e4 LT |
1684 | */ |
1685 | static struct cfq_io_context * | |
e2d74ac0 | 1686 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 1687 | { |
22e2c507 | 1688 | struct io_context *ioc = NULL; |
1da177e4 | 1689 | struct cfq_io_context *cic; |
1da177e4 | 1690 | |
22e2c507 | 1691 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 1692 | |
b5deef90 | 1693 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
1694 | if (!ioc) |
1695 | return NULL; | |
1696 | ||
4ac845a2 | 1697 | cic = cfq_cic_lookup(cfqd, ioc); |
e2d74ac0 JA |
1698 | if (cic) |
1699 | goto out; | |
1da177e4 | 1700 | |
e2d74ac0 JA |
1701 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
1702 | if (cic == NULL) | |
1703 | goto err; | |
1da177e4 | 1704 | |
4ac845a2 JA |
1705 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) |
1706 | goto err_free; | |
1707 | ||
1da177e4 | 1708 | out: |
fc46379d JA |
1709 | smp_read_barrier_depends(); |
1710 | if (unlikely(ioc->ioprio_changed)) | |
1711 | cfq_ioc_set_ioprio(ioc); | |
1712 | ||
1da177e4 | 1713 | return cic; |
4ac845a2 JA |
1714 | err_free: |
1715 | cfq_cic_free(cic); | |
1da177e4 LT |
1716 | err: |
1717 | put_io_context(ioc); | |
1718 | return NULL; | |
1719 | } | |
1720 | ||
22e2c507 JA |
1721 | static void |
1722 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 1723 | { |
aaf1228d JA |
1724 | unsigned long elapsed = jiffies - cic->last_end_request; |
1725 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 1726 | |
22e2c507 JA |
1727 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
1728 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
1729 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
1730 | } | |
1da177e4 | 1731 | |
206dc69b | 1732 | static void |
6d048f53 JA |
1733 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic, |
1734 | struct request *rq) | |
206dc69b JA |
1735 | { |
1736 | sector_t sdist; | |
1737 | u64 total; | |
1738 | ||
5e705374 JA |
1739 | if (cic->last_request_pos < rq->sector) |
1740 | sdist = rq->sector - cic->last_request_pos; | |
206dc69b | 1741 | else |
5e705374 | 1742 | sdist = cic->last_request_pos - rq->sector; |
206dc69b JA |
1743 | |
1744 | /* | |
1745 | * Don't allow the seek distance to get too large from the | |
1746 | * odd fragment, pagein, etc | |
1747 | */ | |
1748 | if (cic->seek_samples <= 60) /* second&third seek */ | |
1749 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024); | |
1750 | else | |
1751 | sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64); | |
1752 | ||
1753 | cic->seek_samples = (7*cic->seek_samples + 256) / 8; | |
1754 | cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8; | |
1755 | total = cic->seek_total + (cic->seek_samples/2); | |
1756 | do_div(total, cic->seek_samples); | |
1757 | cic->seek_mean = (sector_t)total; | |
1758 | } | |
1da177e4 | 1759 | |
22e2c507 JA |
1760 | /* |
1761 | * Disable idle window if the process thinks too long or seeks so much that | |
1762 | * it doesn't matter | |
1763 | */ | |
1764 | static void | |
1765 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1766 | struct cfq_io_context *cic) | |
1767 | { | |
7b679138 | 1768 | int old_idle, enable_idle; |
1be92f2f | 1769 | |
0871714e JA |
1770 | /* |
1771 | * Don't idle for async or idle io prio class | |
1772 | */ | |
1773 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
1774 | return; |
1775 | ||
c265a7f4 | 1776 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 1777 | |
66dac98e | 1778 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
caaa5f9f | 1779 | (cfqd->hw_tag && CIC_SEEKY(cic))) |
22e2c507 JA |
1780 | enable_idle = 0; |
1781 | else if (sample_valid(cic->ttime_samples)) { | |
1782 | if (cic->ttime_mean > cfqd->cfq_slice_idle) | |
1783 | enable_idle = 0; | |
1784 | else | |
1785 | enable_idle = 1; | |
1da177e4 LT |
1786 | } |
1787 | ||
7b679138 JA |
1788 | if (old_idle != enable_idle) { |
1789 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
1790 | if (enable_idle) | |
1791 | cfq_mark_cfqq_idle_window(cfqq); | |
1792 | else | |
1793 | cfq_clear_cfqq_idle_window(cfqq); | |
1794 | } | |
22e2c507 | 1795 | } |
1da177e4 | 1796 | |
22e2c507 JA |
1797 | /* |
1798 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
1799 | * no or if we aren't sure, a 1 will cause a preempt. | |
1800 | */ | |
1801 | static int | |
1802 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, | |
5e705374 | 1803 | struct request *rq) |
22e2c507 | 1804 | { |
6d048f53 | 1805 | struct cfq_queue *cfqq; |
22e2c507 | 1806 | |
6d048f53 JA |
1807 | cfqq = cfqd->active_queue; |
1808 | if (!cfqq) | |
22e2c507 JA |
1809 | return 0; |
1810 | ||
6d048f53 JA |
1811 | if (cfq_slice_used(cfqq)) |
1812 | return 1; | |
1813 | ||
1814 | if (cfq_class_idle(new_cfqq)) | |
caaa5f9f | 1815 | return 0; |
22e2c507 JA |
1816 | |
1817 | if (cfq_class_idle(cfqq)) | |
1818 | return 1; | |
1e3335de | 1819 | |
374f84ac JA |
1820 | /* |
1821 | * if the new request is sync, but the currently running queue is | |
1822 | * not, let the sync request have priority. | |
1823 | */ | |
5e705374 | 1824 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
22e2c507 | 1825 | return 1; |
1e3335de | 1826 | |
374f84ac JA |
1827 | /* |
1828 | * So both queues are sync. Let the new request get disk time if | |
1829 | * it's a metadata request and the current queue is doing regular IO. | |
1830 | */ | |
1831 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
1832 | return 1; | |
22e2c507 | 1833 | |
3a9a3f6c DS |
1834 | /* |
1835 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
1836 | */ | |
1837 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
1838 | return 1; | |
1839 | ||
1e3335de JA |
1840 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
1841 | return 0; | |
1842 | ||
1843 | /* | |
1844 | * if this request is as-good as one we would expect from the | |
1845 | * current cfqq, let it preempt | |
1846 | */ | |
6d048f53 | 1847 | if (cfq_rq_close(cfqd, rq)) |
1e3335de JA |
1848 | return 1; |
1849 | ||
22e2c507 JA |
1850 | return 0; |
1851 | } | |
1852 | ||
1853 | /* | |
1854 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
1855 | * let it have half of its nominal slice. | |
1856 | */ | |
1857 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1858 | { | |
7b679138 | 1859 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
6084cdda | 1860 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 1861 | |
bf572256 JA |
1862 | /* |
1863 | * Put the new queue at the front of the of the current list, | |
1864 | * so we know that it will be selected next. | |
1865 | */ | |
1866 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
1867 | |
1868 | cfq_service_tree_add(cfqd, cfqq, 1); | |
bf572256 | 1869 | |
44f7c160 JA |
1870 | cfqq->slice_end = 0; |
1871 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
1872 | } |
1873 | ||
22e2c507 | 1874 | /* |
5e705374 | 1875 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
1876 | * something we should do about it |
1877 | */ | |
1878 | static void | |
5e705374 JA |
1879 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1880 | struct request *rq) | |
22e2c507 | 1881 | { |
5e705374 | 1882 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 1883 | |
45333d5a | 1884 | cfqd->rq_queued++; |
374f84ac JA |
1885 | if (rq_is_meta(rq)) |
1886 | cfqq->meta_pending++; | |
1887 | ||
9c2c38a1 | 1888 | cfq_update_io_thinktime(cfqd, cic); |
6d048f53 | 1889 | cfq_update_io_seektime(cfqd, cic, rq); |
9c2c38a1 JA |
1890 | cfq_update_idle_window(cfqd, cfqq, cic); |
1891 | ||
5e705374 | 1892 | cic->last_request_pos = rq->sector + rq->nr_sectors; |
22e2c507 JA |
1893 | |
1894 | if (cfqq == cfqd->active_queue) { | |
1895 | /* | |
1896 | * if we are waiting for a request for this queue, let it rip | |
1897 | * immediately and flag that we must not expire this queue | |
1898 | * just now | |
1899 | */ | |
3b18152c JA |
1900 | if (cfq_cfqq_wait_request(cfqq)) { |
1901 | cfq_mark_cfqq_must_dispatch(cfqq); | |
22e2c507 | 1902 | del_timer(&cfqd->idle_slice_timer); |
dc72ef4a | 1903 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1904 | } |
5e705374 | 1905 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
1906 | /* |
1907 | * not the active queue - expire current slice if it is | |
1908 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
1909 | * has some old slice time left and is of higher priority or |
1910 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
1911 | */ |
1912 | cfq_preempt_queue(cfqd, cfqq); | |
3b18152c | 1913 | cfq_mark_cfqq_must_dispatch(cfqq); |
dc72ef4a | 1914 | blk_start_queueing(cfqd->queue); |
22e2c507 | 1915 | } |
1da177e4 LT |
1916 | } |
1917 | ||
165125e1 | 1918 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1919 | { |
b4878f24 | 1920 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1921 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1922 | |
7b679138 | 1923 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
fd0928df | 1924 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
1da177e4 | 1925 | |
5e705374 | 1926 | cfq_add_rq_rb(rq); |
1da177e4 | 1927 | |
22e2c507 JA |
1928 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
1929 | ||
5e705374 | 1930 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
1931 | } |
1932 | ||
45333d5a AC |
1933 | /* |
1934 | * Update hw_tag based on peak queue depth over 50 samples under | |
1935 | * sufficient load. | |
1936 | */ | |
1937 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
1938 | { | |
1939 | if (cfqd->rq_in_driver > cfqd->rq_in_driver_peak) | |
1940 | cfqd->rq_in_driver_peak = cfqd->rq_in_driver; | |
1941 | ||
1942 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
1943 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) | |
1944 | return; | |
1945 | ||
1946 | if (cfqd->hw_tag_samples++ < 50) | |
1947 | return; | |
1948 | ||
1949 | if (cfqd->rq_in_driver_peak >= CFQ_HW_QUEUE_MIN) | |
1950 | cfqd->hw_tag = 1; | |
1951 | else | |
1952 | cfqd->hw_tag = 0; | |
1953 | ||
1954 | cfqd->hw_tag_samples = 0; | |
1955 | cfqd->rq_in_driver_peak = 0; | |
1956 | } | |
1957 | ||
165125e1 | 1958 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1959 | { |
5e705374 | 1960 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 1961 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 1962 | const int sync = rq_is_sync(rq); |
b4878f24 | 1963 | unsigned long now; |
1da177e4 | 1964 | |
b4878f24 | 1965 | now = jiffies; |
7b679138 | 1966 | cfq_log_cfqq(cfqd, cfqq, "complete"); |
1da177e4 | 1967 | |
45333d5a AC |
1968 | cfq_update_hw_tag(cfqd); |
1969 | ||
b4878f24 | 1970 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 1971 | WARN_ON(!cfqq->dispatched); |
b4878f24 | 1972 | cfqd->rq_in_driver--; |
6d048f53 | 1973 | cfqq->dispatched--; |
1da177e4 | 1974 | |
3ed9a296 JA |
1975 | if (cfq_cfqq_sync(cfqq)) |
1976 | cfqd->sync_flight--; | |
1977 | ||
b4878f24 JA |
1978 | if (!cfq_class_idle(cfqq)) |
1979 | cfqd->last_end_request = now; | |
3b18152c | 1980 | |
caaa5f9f | 1981 | if (sync) |
5e705374 | 1982 | RQ_CIC(rq)->last_end_request = now; |
caaa5f9f JA |
1983 | |
1984 | /* | |
1985 | * If this is the active queue, check if it needs to be expired, | |
1986 | * or if we want to idle in case it has no pending requests. | |
1987 | */ | |
1988 | if (cfqd->active_queue == cfqq) { | |
44f7c160 JA |
1989 | if (cfq_cfqq_slice_new(cfqq)) { |
1990 | cfq_set_prio_slice(cfqd, cfqq); | |
1991 | cfq_clear_cfqq_slice_new(cfqq); | |
1992 | } | |
0871714e | 1993 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
6084cdda | 1994 | cfq_slice_expired(cfqd, 1); |
6d048f53 JA |
1995 | else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1996 | cfq_arm_slice_timer(cfqd); | |
caaa5f9f | 1997 | } |
6d048f53 JA |
1998 | |
1999 | if (!cfqd->rq_in_driver) | |
2000 | cfq_schedule_dispatch(cfqd); | |
1da177e4 LT |
2001 | } |
2002 | ||
22e2c507 JA |
2003 | /* |
2004 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
2005 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
2006 | */ | |
2007 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 2008 | { |
22e2c507 JA |
2009 | if (has_fs_excl()) { |
2010 | /* | |
2011 | * boost idle prio on transactions that would lock out other | |
2012 | * users of the filesystem | |
2013 | */ | |
2014 | if (cfq_class_idle(cfqq)) | |
2015 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
2016 | if (cfqq->ioprio > IOPRIO_NORM) | |
2017 | cfqq->ioprio = IOPRIO_NORM; | |
2018 | } else { | |
2019 | /* | |
2020 | * check if we need to unboost the queue | |
2021 | */ | |
2022 | if (cfqq->ioprio_class != cfqq->org_ioprio_class) | |
2023 | cfqq->ioprio_class = cfqq->org_ioprio_class; | |
2024 | if (cfqq->ioprio != cfqq->org_ioprio) | |
2025 | cfqq->ioprio = cfqq->org_ioprio; | |
2026 | } | |
22e2c507 | 2027 | } |
1da177e4 | 2028 | |
89850f7e | 2029 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 2030 | { |
3b18152c | 2031 | if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) && |
99f95e52 | 2032 | !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 2033 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 2034 | return ELV_MQUEUE_MUST; |
3b18152c | 2035 | } |
1da177e4 | 2036 | |
22e2c507 | 2037 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
2038 | } |
2039 | ||
165125e1 | 2040 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
2041 | { |
2042 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2043 | struct task_struct *tsk = current; | |
91fac317 | 2044 | struct cfq_io_context *cic; |
22e2c507 JA |
2045 | struct cfq_queue *cfqq; |
2046 | ||
2047 | /* | |
2048 | * don't force setup of a queue from here, as a call to may_queue | |
2049 | * does not necessarily imply that a request actually will be queued. | |
2050 | * so just lookup a possibly existing queue, or return 'may queue' | |
2051 | * if that fails | |
2052 | */ | |
4ac845a2 | 2053 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2054 | if (!cic) |
2055 | return ELV_MQUEUE_MAY; | |
2056 | ||
2057 | cfqq = cic_to_cfqq(cic, rw & REQ_RW_SYNC); | |
22e2c507 | 2058 | if (cfqq) { |
fd0928df | 2059 | cfq_init_prio_data(cfqq, cic->ioc); |
22e2c507 JA |
2060 | cfq_prio_boost(cfqq); |
2061 | ||
89850f7e | 2062 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
2063 | } |
2064 | ||
2065 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
2066 | } |
2067 | ||
1da177e4 LT |
2068 | /* |
2069 | * queue lock held here | |
2070 | */ | |
bb37b94c | 2071 | static void cfq_put_request(struct request *rq) |
1da177e4 | 2072 | { |
5e705374 | 2073 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2074 | |
5e705374 | 2075 | if (cfqq) { |
22e2c507 | 2076 | const int rw = rq_data_dir(rq); |
1da177e4 | 2077 | |
22e2c507 JA |
2078 | BUG_ON(!cfqq->allocated[rw]); |
2079 | cfqq->allocated[rw]--; | |
1da177e4 | 2080 | |
5e705374 | 2081 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 2082 | |
1da177e4 | 2083 | rq->elevator_private = NULL; |
5e705374 | 2084 | rq->elevator_private2 = NULL; |
1da177e4 | 2085 | |
1da177e4 LT |
2086 | cfq_put_queue(cfqq); |
2087 | } | |
2088 | } | |
2089 | ||
2090 | /* | |
22e2c507 | 2091 | * Allocate cfq data structures associated with this request. |
1da177e4 | 2092 | */ |
22e2c507 | 2093 | static int |
165125e1 | 2094 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
2095 | { |
2096 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2097 | struct cfq_io_context *cic; | |
2098 | const int rw = rq_data_dir(rq); | |
7749a8d4 | 2099 | const int is_sync = rq_is_sync(rq); |
22e2c507 | 2100 | struct cfq_queue *cfqq; |
1da177e4 LT |
2101 | unsigned long flags; |
2102 | ||
2103 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
2104 | ||
e2d74ac0 | 2105 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 2106 | |
1da177e4 LT |
2107 | spin_lock_irqsave(q->queue_lock, flags); |
2108 | ||
22e2c507 JA |
2109 | if (!cic) |
2110 | goto queue_fail; | |
2111 | ||
91fac317 VT |
2112 | cfqq = cic_to_cfqq(cic, is_sync); |
2113 | if (!cfqq) { | |
fd0928df | 2114 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
91fac317 | 2115 | |
22e2c507 JA |
2116 | if (!cfqq) |
2117 | goto queue_fail; | |
1da177e4 | 2118 | |
91fac317 VT |
2119 | cic_set_cfqq(cic, cfqq, is_sync); |
2120 | } | |
1da177e4 LT |
2121 | |
2122 | cfqq->allocated[rw]++; | |
3b18152c | 2123 | cfq_clear_cfqq_must_alloc(cfqq); |
22e2c507 | 2124 | atomic_inc(&cfqq->ref); |
1da177e4 | 2125 | |
5e705374 | 2126 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 2127 | |
5e705374 JA |
2128 | rq->elevator_private = cic; |
2129 | rq->elevator_private2 = cfqq; | |
2130 | return 0; | |
1da177e4 | 2131 | |
22e2c507 JA |
2132 | queue_fail: |
2133 | if (cic) | |
2134 | put_io_context(cic->ioc); | |
89850f7e | 2135 | |
3b18152c | 2136 | cfq_schedule_dispatch(cfqd); |
1da177e4 | 2137 | spin_unlock_irqrestore(q->queue_lock, flags); |
7b679138 | 2138 | cfq_log(cfqd, "set_request fail"); |
1da177e4 LT |
2139 | return 1; |
2140 | } | |
2141 | ||
65f27f38 | 2142 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 2143 | { |
65f27f38 DH |
2144 | struct cfq_data *cfqd = |
2145 | container_of(work, struct cfq_data, unplug_work); | |
165125e1 | 2146 | struct request_queue *q = cfqd->queue; |
22e2c507 JA |
2147 | unsigned long flags; |
2148 | ||
2149 | spin_lock_irqsave(q->queue_lock, flags); | |
dc72ef4a | 2150 | blk_start_queueing(q); |
22e2c507 JA |
2151 | spin_unlock_irqrestore(q->queue_lock, flags); |
2152 | } | |
2153 | ||
2154 | /* | |
2155 | * Timer running if the active_queue is currently idling inside its time slice | |
2156 | */ | |
2157 | static void cfq_idle_slice_timer(unsigned long data) | |
2158 | { | |
2159 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
2160 | struct cfq_queue *cfqq; | |
2161 | unsigned long flags; | |
3c6bd2f8 | 2162 | int timed_out = 1; |
22e2c507 | 2163 | |
7b679138 JA |
2164 | cfq_log(cfqd, "idle timer fired"); |
2165 | ||
22e2c507 JA |
2166 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
2167 | ||
fe094d98 JA |
2168 | cfqq = cfqd->active_queue; |
2169 | if (cfqq) { | |
3c6bd2f8 JA |
2170 | timed_out = 0; |
2171 | ||
22e2c507 JA |
2172 | /* |
2173 | * expired | |
2174 | */ | |
44f7c160 | 2175 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
2176 | goto expire; |
2177 | ||
2178 | /* | |
2179 | * only expire and reinvoke request handler, if there are | |
2180 | * other queues with pending requests | |
2181 | */ | |
caaa5f9f | 2182 | if (!cfqd->busy_queues) |
22e2c507 | 2183 | goto out_cont; |
22e2c507 JA |
2184 | |
2185 | /* | |
2186 | * not expired and it has a request pending, let it dispatch | |
2187 | */ | |
dd67d051 | 2188 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) { |
3b18152c | 2189 | cfq_mark_cfqq_must_dispatch(cfqq); |
22e2c507 JA |
2190 | goto out_kick; |
2191 | } | |
2192 | } | |
2193 | expire: | |
6084cdda | 2194 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 2195 | out_kick: |
3b18152c | 2196 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
2197 | out_cont: |
2198 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
2199 | } | |
2200 | ||
3b18152c JA |
2201 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
2202 | { | |
2203 | del_timer_sync(&cfqd->idle_slice_timer); | |
64d01dc9 | 2204 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 2205 | } |
22e2c507 | 2206 | |
c2dea2d1 VT |
2207 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
2208 | { | |
2209 | int i; | |
2210 | ||
2211 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
2212 | if (cfqd->async_cfqq[0][i]) | |
2213 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
2214 | if (cfqd->async_cfqq[1][i]) | |
2215 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 2216 | } |
2389d1ef ON |
2217 | |
2218 | if (cfqd->async_idle_cfqq) | |
2219 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
2220 | } |
2221 | ||
b374d18a | 2222 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 2223 | { |
22e2c507 | 2224 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 2225 | struct request_queue *q = cfqd->queue; |
22e2c507 | 2226 | |
3b18152c | 2227 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 2228 | |
d9ff4187 | 2229 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 2230 | |
d9ff4187 | 2231 | if (cfqd->active_queue) |
6084cdda | 2232 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
2233 | |
2234 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
2235 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
2236 | struct cfq_io_context, | |
2237 | queue_list); | |
89850f7e JA |
2238 | |
2239 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 2240 | } |
e2d74ac0 | 2241 | |
c2dea2d1 | 2242 | cfq_put_async_queues(cfqd); |
15c31be4 | 2243 | |
d9ff4187 | 2244 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
2245 | |
2246 | cfq_shutdown_timer_wq(cfqd); | |
2247 | ||
a90d742e | 2248 | kfree(cfqd); |
1da177e4 LT |
2249 | } |
2250 | ||
165125e1 | 2251 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
2252 | { |
2253 | struct cfq_data *cfqd; | |
1da177e4 | 2254 | |
94f6030c | 2255 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
1da177e4 | 2256 | if (!cfqd) |
bc1c1169 | 2257 | return NULL; |
1da177e4 | 2258 | |
cc09e299 | 2259 | cfqd->service_tree = CFQ_RB_ROOT; |
d9ff4187 | 2260 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 2261 | |
1da177e4 | 2262 | cfqd->queue = q; |
1da177e4 | 2263 | |
22e2c507 JA |
2264 | init_timer(&cfqd->idle_slice_timer); |
2265 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
2266 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
2267 | ||
65f27f38 | 2268 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 2269 | |
b70c864d | 2270 | cfqd->last_end_request = jiffies; |
1da177e4 | 2271 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
2272 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
2273 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
2274 | cfqd->cfq_back_max = cfq_back_max; |
2275 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
2276 | cfqd->cfq_slice[0] = cfq_slice_async; |
2277 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
2278 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
2279 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
45333d5a | 2280 | cfqd->hw_tag = 1; |
3b18152c | 2281 | |
bc1c1169 | 2282 | return cfqd; |
1da177e4 LT |
2283 | } |
2284 | ||
2285 | static void cfq_slab_kill(void) | |
2286 | { | |
d6de8be7 JA |
2287 | /* |
2288 | * Caller already ensured that pending RCU callbacks are completed, | |
2289 | * so we should have no busy allocations at this point. | |
2290 | */ | |
1da177e4 LT |
2291 | if (cfq_pool) |
2292 | kmem_cache_destroy(cfq_pool); | |
2293 | if (cfq_ioc_pool) | |
2294 | kmem_cache_destroy(cfq_ioc_pool); | |
2295 | } | |
2296 | ||
2297 | static int __init cfq_slab_setup(void) | |
2298 | { | |
0a31bd5f | 2299 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
1da177e4 LT |
2300 | if (!cfq_pool) |
2301 | goto fail; | |
2302 | ||
34e6bbf2 | 2303 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
1da177e4 LT |
2304 | if (!cfq_ioc_pool) |
2305 | goto fail; | |
2306 | ||
2307 | return 0; | |
2308 | fail: | |
2309 | cfq_slab_kill(); | |
2310 | return -ENOMEM; | |
2311 | } | |
2312 | ||
1da177e4 LT |
2313 | /* |
2314 | * sysfs parts below --> | |
2315 | */ | |
1da177e4 LT |
2316 | static ssize_t |
2317 | cfq_var_show(unsigned int var, char *page) | |
2318 | { | |
2319 | return sprintf(page, "%d\n", var); | |
2320 | } | |
2321 | ||
2322 | static ssize_t | |
2323 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
2324 | { | |
2325 | char *p = (char *) page; | |
2326 | ||
2327 | *var = simple_strtoul(p, &p, 10); | |
2328 | return count; | |
2329 | } | |
2330 | ||
1da177e4 | 2331 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 2332 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 2333 | { \ |
3d1ab40f | 2334 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2335 | unsigned int __data = __VAR; \ |
2336 | if (__CONV) \ | |
2337 | __data = jiffies_to_msecs(__data); \ | |
2338 | return cfq_var_show(__data, (page)); \ | |
2339 | } | |
2340 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
2341 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
2342 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
2343 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
2344 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
2345 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
2346 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
2347 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
2348 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
1da177e4 LT |
2349 | #undef SHOW_FUNCTION |
2350 | ||
2351 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 2352 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 2353 | { \ |
3d1ab40f | 2354 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
2355 | unsigned int __data; \ |
2356 | int ret = cfq_var_store(&__data, (page), count); \ | |
2357 | if (__data < (MIN)) \ | |
2358 | __data = (MIN); \ | |
2359 | else if (__data > (MAX)) \ | |
2360 | __data = (MAX); \ | |
2361 | if (__CONV) \ | |
2362 | *(__PTR) = msecs_to_jiffies(__data); \ | |
2363 | else \ | |
2364 | *(__PTR) = __data; \ | |
2365 | return ret; \ | |
2366 | } | |
2367 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
2368 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
2369 | UINT_MAX, 1); | |
2370 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
2371 | UINT_MAX, 1); | |
e572ec7e | 2372 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
2373 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
2374 | UINT_MAX, 0); | |
22e2c507 JA |
2375 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
2376 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
2377 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
2378 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
2379 | UINT_MAX, 0); | |
1da177e4 LT |
2380 | #undef STORE_FUNCTION |
2381 | ||
e572ec7e AV |
2382 | #define CFQ_ATTR(name) \ |
2383 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
2384 | ||
2385 | static struct elv_fs_entry cfq_attrs[] = { | |
2386 | CFQ_ATTR(quantum), | |
e572ec7e AV |
2387 | CFQ_ATTR(fifo_expire_sync), |
2388 | CFQ_ATTR(fifo_expire_async), | |
2389 | CFQ_ATTR(back_seek_max), | |
2390 | CFQ_ATTR(back_seek_penalty), | |
2391 | CFQ_ATTR(slice_sync), | |
2392 | CFQ_ATTR(slice_async), | |
2393 | CFQ_ATTR(slice_async_rq), | |
2394 | CFQ_ATTR(slice_idle), | |
e572ec7e | 2395 | __ATTR_NULL |
1da177e4 LT |
2396 | }; |
2397 | ||
1da177e4 LT |
2398 | static struct elevator_type iosched_cfq = { |
2399 | .ops = { | |
2400 | .elevator_merge_fn = cfq_merge, | |
2401 | .elevator_merged_fn = cfq_merged_request, | |
2402 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 2403 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 2404 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 2405 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 2406 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
2407 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
2408 | .elevator_queue_empty_fn = cfq_queue_empty, | |
2409 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
2410 | .elevator_former_req_fn = elv_rb_former_request, |
2411 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
2412 | .elevator_set_req_fn = cfq_set_request, |
2413 | .elevator_put_req_fn = cfq_put_request, | |
2414 | .elevator_may_queue_fn = cfq_may_queue, | |
2415 | .elevator_init_fn = cfq_init_queue, | |
2416 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 2417 | .trim = cfq_free_io_context, |
1da177e4 | 2418 | }, |
3d1ab40f | 2419 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
2420 | .elevator_name = "cfq", |
2421 | .elevator_owner = THIS_MODULE, | |
2422 | }; | |
2423 | ||
2424 | static int __init cfq_init(void) | |
2425 | { | |
22e2c507 JA |
2426 | /* |
2427 | * could be 0 on HZ < 1000 setups | |
2428 | */ | |
2429 | if (!cfq_slice_async) | |
2430 | cfq_slice_async = 1; | |
2431 | if (!cfq_slice_idle) | |
2432 | cfq_slice_idle = 1; | |
2433 | ||
1da177e4 LT |
2434 | if (cfq_slab_setup()) |
2435 | return -ENOMEM; | |
2436 | ||
2fdd82bd | 2437 | elv_register(&iosched_cfq); |
1da177e4 | 2438 | |
2fdd82bd | 2439 | return 0; |
1da177e4 LT |
2440 | } |
2441 | ||
2442 | static void __exit cfq_exit(void) | |
2443 | { | |
6e9a4738 | 2444 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 2445 | elv_unregister(&iosched_cfq); |
334e94de | 2446 | ioc_gone = &all_gone; |
fba82272 OH |
2447 | /* ioc_gone's update must be visible before reading ioc_count */ |
2448 | smp_wmb(); | |
d6de8be7 JA |
2449 | |
2450 | /* | |
2451 | * this also protects us from entering cfq_slab_kill() with | |
2452 | * pending RCU callbacks | |
2453 | */ | |
4050cf16 | 2454 | if (elv_ioc_count_read(ioc_count)) |
9a11b4ed | 2455 | wait_for_completion(&all_gone); |
83521d3e | 2456 | cfq_slab_kill(); |
1da177e4 LT |
2457 | } |
2458 | ||
2459 | module_init(cfq_init); | |
2460 | module_exit(cfq_exit); | |
2461 | ||
2462 | MODULE_AUTHOR("Jens Axboe"); | |
2463 | MODULE_LICENSE("GPL"); | |
2464 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |