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> |
5a0e3ad6 | 10 | #include <linux/slab.h> |
1cc9be68 AV |
11 | #include <linux/blkdev.h> |
12 | #include <linux/elevator.h> | |
ad5ebd2f | 13 | #include <linux/jiffies.h> |
1da177e4 | 14 | #include <linux/rbtree.h> |
22e2c507 | 15 | #include <linux/ioprio.h> |
7b679138 | 16 | #include <linux/blktrace_api.h> |
6e736be7 | 17 | #include "blk.h" |
629ed0b1 | 18 | #include "blk-cgroup.h" |
1da177e4 LT |
19 | |
20 | /* | |
21 | * tunables | |
22 | */ | |
fe094d98 | 23 | /* max queue in one round of service */ |
abc3c744 | 24 | static const int cfq_quantum = 8; |
64100099 | 25 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
26 | /* maximum backwards seek, in KiB */ |
27 | static const int cfq_back_max = 16 * 1024; | |
28 | /* penalty of a backwards seek */ | |
29 | static const int cfq_back_penalty = 2; | |
64100099 | 30 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 31 | static int cfq_slice_async = HZ / 25; |
64100099 | 32 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 33 | static int cfq_slice_idle = HZ / 125; |
80bdf0c7 | 34 | static int cfq_group_idle = HZ / 125; |
5db5d642 CZ |
35 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
36 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 37 | |
d9e7620e | 38 | /* |
0871714e | 39 | * offset from end of service tree |
d9e7620e | 40 | */ |
0871714e | 41 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
42 | |
43 | /* | |
44 | * below this threshold, we consider thinktime immediate | |
45 | */ | |
46 | #define CFQ_MIN_TT (2) | |
47 | ||
22e2c507 | 48 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 49 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 50 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 51 | |
3dde36dd | 52 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 53 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 54 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 55 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 56 | |
a612fddf TH |
57 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
58 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
59 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 60 | |
e18b890b | 61 | static struct kmem_cache *cfq_pool; |
1da177e4 | 62 | |
22e2c507 JA |
63 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
64 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
65 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
66 | ||
206dc69b | 67 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 68 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 69 | |
c5869807 TH |
70 | struct cfq_ttime { |
71 | unsigned long last_end_request; | |
72 | ||
73 | unsigned long ttime_total; | |
74 | unsigned long ttime_samples; | |
75 | unsigned long ttime_mean; | |
76 | }; | |
77 | ||
cc09e299 JA |
78 | /* |
79 | * Most of our rbtree usage is for sorting with min extraction, so | |
80 | * if we cache the leftmost node we don't have to walk down the tree | |
81 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
82 | * move this into the elevator for the rq sorting as well. | |
83 | */ | |
84 | struct cfq_rb_root { | |
85 | struct rb_root rb; | |
86 | struct rb_node *left; | |
aa6f6a3d | 87 | unsigned count; |
73e9ffdd | 88 | unsigned total_weight; |
1fa8f6d6 | 89 | u64 min_vdisktime; |
f5f2b6ce | 90 | struct cfq_ttime ttime; |
cc09e299 | 91 | }; |
f5f2b6ce SL |
92 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
93 | .ttime = {.last_end_request = jiffies,},} | |
cc09e299 | 94 | |
6118b70b JA |
95 | /* |
96 | * Per process-grouping structure | |
97 | */ | |
98 | struct cfq_queue { | |
99 | /* reference count */ | |
30d7b944 | 100 | int ref; |
6118b70b JA |
101 | /* various state flags, see below */ |
102 | unsigned int flags; | |
103 | /* parent cfq_data */ | |
104 | struct cfq_data *cfqd; | |
105 | /* service_tree member */ | |
106 | struct rb_node rb_node; | |
107 | /* service_tree key */ | |
108 | unsigned long rb_key; | |
109 | /* prio tree member */ | |
110 | struct rb_node p_node; | |
111 | /* prio tree root we belong to, if any */ | |
112 | struct rb_root *p_root; | |
113 | /* sorted list of pending requests */ | |
114 | struct rb_root sort_list; | |
115 | /* if fifo isn't expired, next request to serve */ | |
116 | struct request *next_rq; | |
117 | /* requests queued in sort_list */ | |
118 | int queued[2]; | |
119 | /* currently allocated requests */ | |
120 | int allocated[2]; | |
121 | /* fifo list of requests in sort_list */ | |
122 | struct list_head fifo; | |
123 | ||
dae739eb VG |
124 | /* time when queue got scheduled in to dispatch first request. */ |
125 | unsigned long dispatch_start; | |
f75edf2d | 126 | unsigned int allocated_slice; |
c4081ba5 | 127 | unsigned int slice_dispatch; |
dae739eb VG |
128 | /* time when first request from queue completed and slice started. */ |
129 | unsigned long slice_start; | |
6118b70b JA |
130 | unsigned long slice_end; |
131 | long slice_resid; | |
6118b70b | 132 | |
65299a3b CH |
133 | /* pending priority requests */ |
134 | int prio_pending; | |
6118b70b JA |
135 | /* number of requests that are on the dispatch list or inside driver */ |
136 | int dispatched; | |
137 | ||
138 | /* io prio of this group */ | |
139 | unsigned short ioprio, org_ioprio; | |
4aede84b | 140 | unsigned short ioprio_class; |
6118b70b | 141 | |
c4081ba5 RK |
142 | pid_t pid; |
143 | ||
3dde36dd | 144 | u32 seek_history; |
b2c18e1e JM |
145 | sector_t last_request_pos; |
146 | ||
aa6f6a3d | 147 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 148 | struct cfq_queue *new_cfqq; |
cdb16e8f | 149 | struct cfq_group *cfqg; |
c4e7893e VG |
150 | /* Number of sectors dispatched from queue in single dispatch round */ |
151 | unsigned long nr_sectors; | |
6118b70b JA |
152 | }; |
153 | ||
c0324a02 | 154 | /* |
718eee05 | 155 | * First index in the service_trees. |
c0324a02 CZ |
156 | * IDLE is handled separately, so it has negative index |
157 | */ | |
3bf10fea | 158 | enum wl_class_t { |
c0324a02 | 159 | BE_WORKLOAD = 0, |
615f0259 VG |
160 | RT_WORKLOAD = 1, |
161 | IDLE_WORKLOAD = 2, | |
b4627321 | 162 | CFQ_PRIO_NR, |
c0324a02 CZ |
163 | }; |
164 | ||
718eee05 CZ |
165 | /* |
166 | * Second index in the service_trees. | |
167 | */ | |
168 | enum wl_type_t { | |
169 | ASYNC_WORKLOAD = 0, | |
170 | SYNC_NOIDLE_WORKLOAD = 1, | |
171 | SYNC_WORKLOAD = 2 | |
172 | }; | |
173 | ||
155fead9 TH |
174 | struct cfqg_stats { |
175 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
176 | /* total bytes transferred */ | |
177 | struct blkg_rwstat service_bytes; | |
178 | /* total IOs serviced, post merge */ | |
179 | struct blkg_rwstat serviced; | |
180 | /* number of ios merged */ | |
181 | struct blkg_rwstat merged; | |
182 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
183 | struct blkg_rwstat service_time; | |
184 | /* total time spent waiting in scheduler queue in ns */ | |
185 | struct blkg_rwstat wait_time; | |
186 | /* number of IOs queued up */ | |
187 | struct blkg_rwstat queued; | |
188 | /* total sectors transferred */ | |
189 | struct blkg_stat sectors; | |
190 | /* total disk time and nr sectors dispatched by this group */ | |
191 | struct blkg_stat time; | |
192 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
193 | /* time not charged to this cgroup */ | |
194 | struct blkg_stat unaccounted_time; | |
195 | /* sum of number of ios queued across all samples */ | |
196 | struct blkg_stat avg_queue_size_sum; | |
197 | /* count of samples taken for average */ | |
198 | struct blkg_stat avg_queue_size_samples; | |
199 | /* how many times this group has been removed from service tree */ | |
200 | struct blkg_stat dequeue; | |
201 | /* total time spent waiting for it to be assigned a timeslice. */ | |
202 | struct blkg_stat group_wait_time; | |
3c798398 | 203 | /* time spent idling for this blkcg_gq */ |
155fead9 TH |
204 | struct blkg_stat idle_time; |
205 | /* total time with empty current active q with other requests queued */ | |
206 | struct blkg_stat empty_time; | |
207 | /* fields after this shouldn't be cleared on stat reset */ | |
208 | uint64_t start_group_wait_time; | |
209 | uint64_t start_idle_time; | |
210 | uint64_t start_empty_time; | |
211 | uint16_t flags; | |
212 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
213 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
214 | }; | |
215 | ||
cdb16e8f VG |
216 | /* This is per cgroup per device grouping structure */ |
217 | struct cfq_group { | |
f95a04af TH |
218 | /* must be the first member */ |
219 | struct blkg_policy_data pd; | |
220 | ||
1fa8f6d6 VG |
221 | /* group service_tree member */ |
222 | struct rb_node rb_node; | |
223 | ||
224 | /* group service_tree key */ | |
225 | u64 vdisktime; | |
e71357e1 | 226 | |
7918ffb5 TH |
227 | /* |
228 | * The number of active cfqgs and sum of their weights under this | |
229 | * cfqg. This covers this cfqg's leaf_weight and all children's | |
230 | * weights, but does not cover weights of further descendants. | |
231 | * | |
232 | * If a cfqg is on the service tree, it's active. An active cfqg | |
233 | * also activates its parent and contributes to the children_weight | |
234 | * of the parent. | |
235 | */ | |
236 | int nr_active; | |
237 | unsigned int children_weight; | |
238 | ||
e71357e1 TH |
239 | /* |
240 | * There are two weights - (internal) weight is the weight of this | |
241 | * cfqg against the sibling cfqgs. leaf_weight is the wight of | |
242 | * this cfqg against the child cfqgs. For the root cfqg, both | |
243 | * weights are kept in sync for backward compatibility. | |
244 | */ | |
25bc6b07 | 245 | unsigned int weight; |
8184f93e | 246 | unsigned int new_weight; |
3381cb8d | 247 | unsigned int dev_weight; |
1fa8f6d6 | 248 | |
e71357e1 TH |
249 | unsigned int leaf_weight; |
250 | unsigned int new_leaf_weight; | |
251 | unsigned int dev_leaf_weight; | |
252 | ||
1fa8f6d6 VG |
253 | /* number of cfqq currently on this group */ |
254 | int nr_cfqq; | |
255 | ||
cdb16e8f | 256 | /* |
4495a7d4 | 257 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
258 | * create the array for each prio class but at run time it is used |
259 | * only for RT and BE class and slot for IDLE class remains unused. | |
260 | * This is primarily done to avoid confusion and a gcc warning. | |
261 | */ | |
262 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
263 | /* | |
264 | * rr lists of queues with requests. We maintain service trees for | |
265 | * RT and BE classes. These trees are subdivided in subclasses | |
266 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
267 | * class there is no subclassification and all the cfq queues go on | |
268 | * a single tree service_tree_idle. | |
cdb16e8f VG |
269 | * Counts are embedded in the cfq_rb_root |
270 | */ | |
271 | struct cfq_rb_root service_trees[2][3]; | |
272 | struct cfq_rb_root service_tree_idle; | |
dae739eb | 273 | |
4d2ceea4 VG |
274 | unsigned long saved_wl_slice; |
275 | enum wl_type_t saved_wl_type; | |
276 | enum wl_class_t saved_wl_class; | |
4eef3049 | 277 | |
80bdf0c7 VG |
278 | /* number of requests that are on the dispatch list or inside driver */ |
279 | int dispatched; | |
7700fc4f | 280 | struct cfq_ttime ttime; |
155fead9 | 281 | struct cfqg_stats stats; |
cdb16e8f | 282 | }; |
718eee05 | 283 | |
c5869807 TH |
284 | struct cfq_io_cq { |
285 | struct io_cq icq; /* must be the first member */ | |
286 | struct cfq_queue *cfqq[2]; | |
287 | struct cfq_ttime ttime; | |
598971bf TH |
288 | int ioprio; /* the current ioprio */ |
289 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
290 | uint64_t blkcg_id; /* the current blkcg ID */ | |
291 | #endif | |
c5869807 TH |
292 | }; |
293 | ||
22e2c507 JA |
294 | /* |
295 | * Per block device queue structure | |
296 | */ | |
1da177e4 | 297 | struct cfq_data { |
165125e1 | 298 | struct request_queue *queue; |
1fa8f6d6 VG |
299 | /* Root service tree for cfq_groups */ |
300 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 301 | struct cfq_group *root_group; |
22e2c507 | 302 | |
c0324a02 CZ |
303 | /* |
304 | * The priority currently being served | |
22e2c507 | 305 | */ |
4d2ceea4 VG |
306 | enum wl_class_t serving_wl_class; |
307 | enum wl_type_t serving_wl_type; | |
718eee05 | 308 | unsigned long workload_expires; |
cdb16e8f | 309 | struct cfq_group *serving_group; |
a36e71f9 JA |
310 | |
311 | /* | |
312 | * Each priority tree is sorted by next_request position. These | |
313 | * trees are used when determining if two or more queues are | |
314 | * interleaving requests (see cfq_close_cooperator). | |
315 | */ | |
316 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
317 | ||
22e2c507 | 318 | unsigned int busy_queues; |
ef8a41df | 319 | unsigned int busy_sync_queues; |
22e2c507 | 320 | |
53c583d2 CZ |
321 | int rq_in_driver; |
322 | int rq_in_flight[2]; | |
45333d5a AC |
323 | |
324 | /* | |
325 | * queue-depth detection | |
326 | */ | |
327 | int rq_queued; | |
25776e35 | 328 | int hw_tag; |
e459dd08 CZ |
329 | /* |
330 | * hw_tag can be | |
331 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
332 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
333 | * 0 => no NCQ | |
334 | */ | |
335 | int hw_tag_est_depth; | |
336 | unsigned int hw_tag_samples; | |
1da177e4 | 337 | |
22e2c507 JA |
338 | /* |
339 | * idle window management | |
340 | */ | |
341 | struct timer_list idle_slice_timer; | |
23e018a1 | 342 | struct work_struct unplug_work; |
1da177e4 | 343 | |
22e2c507 | 344 | struct cfq_queue *active_queue; |
c5869807 | 345 | struct cfq_io_cq *active_cic; |
22e2c507 | 346 | |
c2dea2d1 VT |
347 | /* |
348 | * async queue for each priority case | |
349 | */ | |
350 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
351 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 352 | |
6d048f53 | 353 | sector_t last_position; |
1da177e4 | 354 | |
1da177e4 LT |
355 | /* |
356 | * tunables, see top of file | |
357 | */ | |
358 | unsigned int cfq_quantum; | |
22e2c507 | 359 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
360 | unsigned int cfq_back_penalty; |
361 | unsigned int cfq_back_max; | |
22e2c507 JA |
362 | unsigned int cfq_slice[2]; |
363 | unsigned int cfq_slice_async_rq; | |
364 | unsigned int cfq_slice_idle; | |
80bdf0c7 | 365 | unsigned int cfq_group_idle; |
963b72fc | 366 | unsigned int cfq_latency; |
5bf14c07 | 367 | unsigned int cfq_target_latency; |
d9ff4187 | 368 | |
6118b70b JA |
369 | /* |
370 | * Fallback dummy cfqq for extreme OOM conditions | |
371 | */ | |
372 | struct cfq_queue oom_cfqq; | |
365722bb | 373 | |
573412b2 | 374 | unsigned long last_delayed_sync; |
1da177e4 LT |
375 | }; |
376 | ||
25fb5169 VG |
377 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
378 | ||
34b98d03 | 379 | static struct cfq_rb_root *st_for(struct cfq_group *cfqg, |
3bf10fea | 380 | enum wl_class_t class, |
65b32a57 | 381 | enum wl_type_t type) |
c0324a02 | 382 | { |
1fa8f6d6 VG |
383 | if (!cfqg) |
384 | return NULL; | |
385 | ||
3bf10fea | 386 | if (class == IDLE_WORKLOAD) |
cdb16e8f | 387 | return &cfqg->service_tree_idle; |
c0324a02 | 388 | |
3bf10fea | 389 | return &cfqg->service_trees[class][type]; |
c0324a02 CZ |
390 | } |
391 | ||
3b18152c | 392 | enum cfqq_state_flags { |
b0b8d749 JA |
393 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
394 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 395 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 396 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
397 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
398 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
399 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 400 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 401 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 402 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 403 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 404 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 405 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
406 | }; |
407 | ||
408 | #define CFQ_CFQQ_FNS(name) \ | |
409 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
410 | { \ | |
fe094d98 | 411 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
412 | } \ |
413 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
414 | { \ | |
fe094d98 | 415 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
416 | } \ |
417 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
418 | { \ | |
fe094d98 | 419 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
420 | } |
421 | ||
422 | CFQ_CFQQ_FNS(on_rr); | |
423 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 424 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 425 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
426 | CFQ_CFQQ_FNS(fifo_expire); |
427 | CFQ_CFQQ_FNS(idle_window); | |
428 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 429 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 430 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 431 | CFQ_CFQQ_FNS(coop); |
ae54abed | 432 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 433 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 434 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
435 | #undef CFQ_CFQQ_FNS |
436 | ||
f95a04af TH |
437 | static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd) |
438 | { | |
439 | return pd ? container_of(pd, struct cfq_group, pd) : NULL; | |
440 | } | |
441 | ||
f95a04af TH |
442 | static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg) |
443 | { | |
444 | return pd_to_blkg(&cfqg->pd); | |
445 | } | |
446 | ||
629ed0b1 | 447 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
2ce4d50f | 448 | |
155fead9 TH |
449 | /* cfqg stats flags */ |
450 | enum cfqg_stats_flags { | |
451 | CFQG_stats_waiting = 0, | |
452 | CFQG_stats_idling, | |
453 | CFQG_stats_empty, | |
629ed0b1 TH |
454 | }; |
455 | ||
155fead9 TH |
456 | #define CFQG_FLAG_FNS(name) \ |
457 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ | |
629ed0b1 | 458 | { \ |
155fead9 | 459 | stats->flags |= (1 << CFQG_stats_##name); \ |
629ed0b1 | 460 | } \ |
155fead9 | 461 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 462 | { \ |
155fead9 | 463 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
629ed0b1 | 464 | } \ |
155fead9 | 465 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 466 | { \ |
155fead9 | 467 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
629ed0b1 TH |
468 | } \ |
469 | ||
155fead9 TH |
470 | CFQG_FLAG_FNS(waiting) |
471 | CFQG_FLAG_FNS(idling) | |
472 | CFQG_FLAG_FNS(empty) | |
473 | #undef CFQG_FLAG_FNS | |
629ed0b1 TH |
474 | |
475 | /* This should be called with the queue_lock held. */ | |
155fead9 | 476 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
629ed0b1 TH |
477 | { |
478 | unsigned long long now; | |
479 | ||
155fead9 | 480 | if (!cfqg_stats_waiting(stats)) |
629ed0b1 TH |
481 | return; |
482 | ||
483 | now = sched_clock(); | |
484 | if (time_after64(now, stats->start_group_wait_time)) | |
485 | blkg_stat_add(&stats->group_wait_time, | |
486 | now - stats->start_group_wait_time); | |
155fead9 | 487 | cfqg_stats_clear_waiting(stats); |
629ed0b1 TH |
488 | } |
489 | ||
490 | /* This should be called with the queue_lock held. */ | |
155fead9 TH |
491 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
492 | struct cfq_group *curr_cfqg) | |
629ed0b1 | 493 | { |
155fead9 | 494 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 495 | |
155fead9 | 496 | if (cfqg_stats_waiting(stats)) |
629ed0b1 | 497 | return; |
155fead9 | 498 | if (cfqg == curr_cfqg) |
629ed0b1 | 499 | return; |
155fead9 TH |
500 | stats->start_group_wait_time = sched_clock(); |
501 | cfqg_stats_mark_waiting(stats); | |
629ed0b1 TH |
502 | } |
503 | ||
504 | /* This should be called with the queue_lock held. */ | |
155fead9 | 505 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
629ed0b1 TH |
506 | { |
507 | unsigned long long now; | |
508 | ||
155fead9 | 509 | if (!cfqg_stats_empty(stats)) |
629ed0b1 TH |
510 | return; |
511 | ||
512 | now = sched_clock(); | |
513 | if (time_after64(now, stats->start_empty_time)) | |
514 | blkg_stat_add(&stats->empty_time, | |
515 | now - stats->start_empty_time); | |
155fead9 | 516 | cfqg_stats_clear_empty(stats); |
629ed0b1 TH |
517 | } |
518 | ||
155fead9 | 519 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
629ed0b1 | 520 | { |
155fead9 | 521 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
629ed0b1 TH |
522 | } |
523 | ||
155fead9 | 524 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
629ed0b1 | 525 | { |
155fead9 | 526 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
527 | |
528 | if (blkg_rwstat_sum(&stats->queued)) | |
529 | return; | |
530 | ||
531 | /* | |
532 | * group is already marked empty. This can happen if cfqq got new | |
533 | * request in parent group and moved to this group while being added | |
534 | * to service tree. Just ignore the event and move on. | |
535 | */ | |
155fead9 | 536 | if (cfqg_stats_empty(stats)) |
629ed0b1 TH |
537 | return; |
538 | ||
539 | stats->start_empty_time = sched_clock(); | |
155fead9 | 540 | cfqg_stats_mark_empty(stats); |
629ed0b1 TH |
541 | } |
542 | ||
155fead9 | 543 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 544 | { |
155fead9 | 545 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 546 | |
155fead9 | 547 | if (cfqg_stats_idling(stats)) { |
629ed0b1 TH |
548 | unsigned long long now = sched_clock(); |
549 | ||
550 | if (time_after64(now, stats->start_idle_time)) | |
551 | blkg_stat_add(&stats->idle_time, | |
552 | now - stats->start_idle_time); | |
155fead9 | 553 | cfqg_stats_clear_idling(stats); |
629ed0b1 TH |
554 | } |
555 | } | |
556 | ||
155fead9 | 557 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 558 | { |
155fead9 | 559 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 560 | |
155fead9 | 561 | BUG_ON(cfqg_stats_idling(stats)); |
629ed0b1 TH |
562 | |
563 | stats->start_idle_time = sched_clock(); | |
155fead9 | 564 | cfqg_stats_mark_idling(stats); |
629ed0b1 TH |
565 | } |
566 | ||
155fead9 | 567 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
629ed0b1 | 568 | { |
155fead9 | 569 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
570 | |
571 | blkg_stat_add(&stats->avg_queue_size_sum, | |
572 | blkg_rwstat_sum(&stats->queued)); | |
573 | blkg_stat_add(&stats->avg_queue_size_samples, 1); | |
155fead9 | 574 | cfqg_stats_update_group_wait_time(stats); |
629ed0b1 TH |
575 | } |
576 | ||
577 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
578 | ||
f48ec1d7 TH |
579 | static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { } |
580 | static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } | |
581 | static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } | |
582 | static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } | |
583 | static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } | |
584 | static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } | |
585 | static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } | |
629ed0b1 TH |
586 | |
587 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
588 | ||
589 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
2ce4d50f | 590 | |
ffea73fc TH |
591 | static struct blkcg_policy blkcg_policy_cfq; |
592 | ||
593 | static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg) | |
594 | { | |
595 | return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq)); | |
596 | } | |
597 | ||
7918ffb5 TH |
598 | /* |
599 | * Determine the parent cfqg for weight calculation. Currently, cfqg | |
600 | * scheduling is flat and the root is the parent of everyone else. | |
601 | */ | |
602 | static inline struct cfq_group *cfqg_flat_parent(struct cfq_group *cfqg) | |
603 | { | |
604 | struct blkcg_gq *blkg = cfqg_to_blkg(cfqg); | |
605 | struct cfq_group *root; | |
606 | ||
607 | while (blkg->parent) | |
608 | blkg = blkg->parent; | |
609 | root = blkg_to_cfqg(blkg); | |
610 | ||
611 | return root != cfqg ? root : NULL; | |
612 | } | |
613 | ||
eb7d8c07 TH |
614 | static inline void cfqg_get(struct cfq_group *cfqg) |
615 | { | |
616 | return blkg_get(cfqg_to_blkg(cfqg)); | |
617 | } | |
618 | ||
619 | static inline void cfqg_put(struct cfq_group *cfqg) | |
620 | { | |
621 | return blkg_put(cfqg_to_blkg(cfqg)); | |
622 | } | |
623 | ||
54e7ed12 TH |
624 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \ |
625 | char __pbuf[128]; \ | |
626 | \ | |
627 | blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \ | |
b226e5c4 VG |
628 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \ |
629 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
630 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
54e7ed12 TH |
631 | __pbuf, ##args); \ |
632 | } while (0) | |
633 | ||
634 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \ | |
635 | char __pbuf[128]; \ | |
636 | \ | |
637 | blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \ | |
638 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \ | |
639 | } while (0) | |
2868ef7b | 640 | |
155fead9 TH |
641 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
642 | struct cfq_group *curr_cfqg, int rw) | |
2ce4d50f | 643 | { |
155fead9 TH |
644 | blkg_rwstat_add(&cfqg->stats.queued, rw, 1); |
645 | cfqg_stats_end_empty_time(&cfqg->stats); | |
646 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); | |
2ce4d50f TH |
647 | } |
648 | ||
155fead9 TH |
649 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
650 | unsigned long time, unsigned long unaccounted_time) | |
2ce4d50f | 651 | { |
155fead9 | 652 | blkg_stat_add(&cfqg->stats.time, time); |
629ed0b1 | 653 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
155fead9 | 654 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
629ed0b1 | 655 | #endif |
2ce4d50f TH |
656 | } |
657 | ||
155fead9 | 658 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) |
2ce4d50f | 659 | { |
155fead9 | 660 | blkg_rwstat_add(&cfqg->stats.queued, rw, -1); |
2ce4d50f TH |
661 | } |
662 | ||
155fead9 | 663 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) |
2ce4d50f | 664 | { |
155fead9 | 665 | blkg_rwstat_add(&cfqg->stats.merged, rw, 1); |
2ce4d50f TH |
666 | } |
667 | ||
155fead9 TH |
668 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, |
669 | uint64_t bytes, int rw) | |
2ce4d50f | 670 | { |
155fead9 TH |
671 | blkg_stat_add(&cfqg->stats.sectors, bytes >> 9); |
672 | blkg_rwstat_add(&cfqg->stats.serviced, rw, 1); | |
673 | blkg_rwstat_add(&cfqg->stats.service_bytes, rw, bytes); | |
2ce4d50f TH |
674 | } |
675 | ||
155fead9 TH |
676 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
677 | uint64_t start_time, uint64_t io_start_time, int rw) | |
2ce4d50f | 678 | { |
155fead9 | 679 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 680 | unsigned long long now = sched_clock(); |
629ed0b1 TH |
681 | |
682 | if (time_after64(now, io_start_time)) | |
683 | blkg_rwstat_add(&stats->service_time, rw, now - io_start_time); | |
684 | if (time_after64(io_start_time, start_time)) | |
685 | blkg_rwstat_add(&stats->wait_time, rw, | |
686 | io_start_time - start_time); | |
2ce4d50f TH |
687 | } |
688 | ||
3c798398 | 689 | static void cfq_pd_reset_stats(struct blkcg_gq *blkg) |
155fead9 TH |
690 | { |
691 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); | |
692 | struct cfqg_stats *stats = &cfqg->stats; | |
693 | ||
694 | /* queued stats shouldn't be cleared */ | |
695 | blkg_rwstat_reset(&stats->service_bytes); | |
696 | blkg_rwstat_reset(&stats->serviced); | |
697 | blkg_rwstat_reset(&stats->merged); | |
698 | blkg_rwstat_reset(&stats->service_time); | |
699 | blkg_rwstat_reset(&stats->wait_time); | |
700 | blkg_stat_reset(&stats->time); | |
701 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
702 | blkg_stat_reset(&stats->unaccounted_time); | |
703 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
704 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
705 | blkg_stat_reset(&stats->dequeue); | |
706 | blkg_stat_reset(&stats->group_wait_time); | |
707 | blkg_stat_reset(&stats->idle_time); | |
708 | blkg_stat_reset(&stats->empty_time); | |
709 | #endif | |
710 | } | |
711 | ||
eb7d8c07 TH |
712 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
713 | ||
7918ffb5 | 714 | static inline struct cfq_group *cfqg_flat_parent(struct cfq_group *cfqg) { return NULL; } |
eb7d8c07 TH |
715 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
716 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
717 | ||
7b679138 | 718 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
b226e5c4 VG |
719 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \ |
720 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
721 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
722 | ##args) | |
4495a7d4 | 723 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 | 724 | |
155fead9 TH |
725 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
726 | struct cfq_group *curr_cfqg, int rw) { } | |
727 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, | |
728 | unsigned long time, unsigned long unaccounted_time) { } | |
729 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) { } | |
730 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) { } | |
731 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, | |
732 | uint64_t bytes, int rw) { } | |
733 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, | |
734 | uint64_t start_time, uint64_t io_start_time, int rw) { } | |
2ce4d50f | 735 | |
eb7d8c07 TH |
736 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
737 | ||
7b679138 JA |
738 | #define cfq_log(cfqd, fmt, args...) \ |
739 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
740 | ||
615f0259 VG |
741 | /* Traverses through cfq group service trees */ |
742 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
743 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
744 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
745 | : &cfqg->service_tree_idle; \ | |
746 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
747 | (i == IDLE_WORKLOAD && j == 0); \ | |
748 | j++, st = i < IDLE_WORKLOAD ? \ | |
749 | &cfqg->service_trees[i][j]: NULL) \ | |
750 | ||
f5f2b6ce SL |
751 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
752 | struct cfq_ttime *ttime, bool group_idle) | |
753 | { | |
754 | unsigned long slice; | |
755 | if (!sample_valid(ttime->ttime_samples)) | |
756 | return false; | |
757 | if (group_idle) | |
758 | slice = cfqd->cfq_group_idle; | |
759 | else | |
760 | slice = cfqd->cfq_slice_idle; | |
761 | return ttime->ttime_mean > slice; | |
762 | } | |
615f0259 | 763 | |
02b35081 VG |
764 | static inline bool iops_mode(struct cfq_data *cfqd) |
765 | { | |
766 | /* | |
767 | * If we are not idling on queues and it is a NCQ drive, parallel | |
768 | * execution of requests is on and measuring time is not possible | |
769 | * in most of the cases until and unless we drive shallower queue | |
770 | * depths and that becomes a performance bottleneck. In such cases | |
771 | * switch to start providing fairness in terms of number of IOs. | |
772 | */ | |
773 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
774 | return true; | |
775 | else | |
776 | return false; | |
777 | } | |
778 | ||
3bf10fea | 779 | static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq) |
c0324a02 CZ |
780 | { |
781 | if (cfq_class_idle(cfqq)) | |
782 | return IDLE_WORKLOAD; | |
783 | if (cfq_class_rt(cfqq)) | |
784 | return RT_WORKLOAD; | |
785 | return BE_WORKLOAD; | |
786 | } | |
787 | ||
718eee05 CZ |
788 | |
789 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
790 | { | |
791 | if (!cfq_cfqq_sync(cfqq)) | |
792 | return ASYNC_WORKLOAD; | |
793 | if (!cfq_cfqq_idle_window(cfqq)) | |
794 | return SYNC_NOIDLE_WORKLOAD; | |
795 | return SYNC_WORKLOAD; | |
796 | } | |
797 | ||
3bf10fea | 798 | static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class, |
58ff82f3 VG |
799 | struct cfq_data *cfqd, |
800 | struct cfq_group *cfqg) | |
c0324a02 | 801 | { |
3bf10fea | 802 | if (wl_class == IDLE_WORKLOAD) |
cdb16e8f | 803 | return cfqg->service_tree_idle.count; |
c0324a02 | 804 | |
34b98d03 VG |
805 | return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count + |
806 | cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count + | |
807 | cfqg->service_trees[wl_class][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
808 | } |
809 | ||
f26bd1f0 VG |
810 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
811 | struct cfq_group *cfqg) | |
812 | { | |
34b98d03 VG |
813 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count + |
814 | cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
f26bd1f0 VG |
815 | } |
816 | ||
165125e1 | 817 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 818 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
abede6da | 819 | struct cfq_io_cq *cic, struct bio *bio, |
4f85cb96 | 820 | gfp_t gfp_mask); |
91fac317 | 821 | |
c5869807 TH |
822 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
823 | { | |
824 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
825 | return container_of(icq, struct cfq_io_cq, icq); | |
826 | } | |
827 | ||
47fdd4ca TH |
828 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
829 | struct io_context *ioc) | |
830 | { | |
831 | if (ioc) | |
832 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
833 | return NULL; | |
834 | } | |
835 | ||
c5869807 | 836 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 837 | { |
a6151c3a | 838 | return cic->cfqq[is_sync]; |
91fac317 VT |
839 | } |
840 | ||
c5869807 TH |
841 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
842 | bool is_sync) | |
91fac317 | 843 | { |
a6151c3a | 844 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
845 | } |
846 | ||
c5869807 | 847 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 848 | { |
c5869807 | 849 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
850 | } |
851 | ||
91fac317 VT |
852 | /* |
853 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
854 | * set (in which case it could also be direct WRITE). | |
855 | */ | |
a6151c3a | 856 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 857 | { |
7b6d91da | 858 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 859 | } |
1da177e4 | 860 | |
99f95e52 AM |
861 | /* |
862 | * scheduler run of queue, if there are requests pending and no one in the | |
863 | * driver that will restart queueing | |
864 | */ | |
23e018a1 | 865 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 866 | { |
7b679138 JA |
867 | if (cfqd->busy_queues) { |
868 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 869 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 870 | } |
99f95e52 AM |
871 | } |
872 | ||
44f7c160 JA |
873 | /* |
874 | * Scale schedule slice based on io priority. Use the sync time slice only | |
875 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
876 | * io only, should not get full sync slice length. | |
877 | */ | |
a6151c3a | 878 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 879 | unsigned short prio) |
44f7c160 | 880 | { |
d9e7620e | 881 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 882 | |
d9e7620e JA |
883 | WARN_ON(prio >= IOPRIO_BE_NR); |
884 | ||
885 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
886 | } | |
44f7c160 | 887 | |
d9e7620e JA |
888 | static inline int |
889 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
890 | { | |
891 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
892 | } |
893 | ||
25bc6b07 VG |
894 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
895 | { | |
896 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
897 | ||
3381cb8d | 898 | d = d * CFQ_WEIGHT_DEFAULT; |
25bc6b07 VG |
899 | do_div(d, cfqg->weight); |
900 | return d; | |
901 | } | |
902 | ||
903 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
904 | { | |
905 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
906 | if (delta > 0) | |
907 | min_vdisktime = vdisktime; | |
908 | ||
909 | return min_vdisktime; | |
910 | } | |
911 | ||
912 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
913 | { | |
914 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
915 | if (delta < 0) | |
916 | min_vdisktime = vdisktime; | |
917 | ||
918 | return min_vdisktime; | |
919 | } | |
920 | ||
921 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
922 | { | |
25bc6b07 VG |
923 | struct cfq_group *cfqg; |
924 | ||
25bc6b07 VG |
925 | if (st->left) { |
926 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
927 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
928 | cfqg->vdisktime); | |
25bc6b07 | 929 | } |
25bc6b07 VG |
930 | } |
931 | ||
5db5d642 CZ |
932 | /* |
933 | * get averaged number of queues of RT/BE priority. | |
934 | * average is updated, with a formula that gives more weight to higher numbers, | |
935 | * to quickly follows sudden increases and decrease slowly | |
936 | */ | |
937 | ||
58ff82f3 VG |
938 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
939 | struct cfq_group *cfqg, bool rt) | |
5869619c | 940 | { |
5db5d642 CZ |
941 | unsigned min_q, max_q; |
942 | unsigned mult = cfq_hist_divisor - 1; | |
943 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 944 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 945 | |
58ff82f3 VG |
946 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
947 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
948 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 949 | cfq_hist_divisor; |
58ff82f3 VG |
950 | return cfqg->busy_queues_avg[rt]; |
951 | } | |
952 | ||
953 | static inline unsigned | |
954 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
955 | { | |
956 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
957 | ||
5bf14c07 | 958 | return cfqd->cfq_target_latency * cfqg->weight / st->total_weight; |
5db5d642 CZ |
959 | } |
960 | ||
c553f8e3 | 961 | static inline unsigned |
ba5bd520 | 962 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 963 | { |
5db5d642 CZ |
964 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
965 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
966 | /* |
967 | * interested queues (we consider only the ones with the same | |
968 | * priority class in the cfq group) | |
969 | */ | |
970 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
971 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
972 | unsigned sync_slice = cfqd->cfq_slice[1]; |
973 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
974 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
975 | ||
976 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
977 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
978 | /* scale low_slice according to IO priority | |
979 | * and sync vs async */ | |
980 | unsigned low_slice = | |
981 | min(slice, base_low_slice * slice / sync_slice); | |
982 | /* the adapted slice value is scaled to fit all iqs | |
983 | * into the target latency */ | |
58ff82f3 | 984 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
985 | low_slice); |
986 | } | |
987 | } | |
c553f8e3 SL |
988 | return slice; |
989 | } | |
990 | ||
991 | static inline void | |
992 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
993 | { | |
ba5bd520 | 994 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 995 | |
dae739eb | 996 | cfqq->slice_start = jiffies; |
5db5d642 | 997 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 998 | cfqq->allocated_slice = slice; |
7b679138 | 999 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
1000 | } |
1001 | ||
1002 | /* | |
1003 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
1004 | * isn't valid until the first request from the dispatch is activated | |
1005 | * and the slice time set. | |
1006 | */ | |
a6151c3a | 1007 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
1008 | { |
1009 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 1010 | return false; |
44f7c160 | 1011 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 1012 | return false; |
44f7c160 | 1013 | |
c1e44756 | 1014 | return true; |
44f7c160 JA |
1015 | } |
1016 | ||
1da177e4 | 1017 | /* |
5e705374 | 1018 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 1019 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 1020 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 1021 | */ |
5e705374 | 1022 | static struct request * |
cf7c25cf | 1023 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 1024 | { |
cf7c25cf | 1025 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 1026 | unsigned long back_max; |
e8a99053 AM |
1027 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
1028 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
1029 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 1030 | |
5e705374 JA |
1031 | if (rq1 == NULL || rq1 == rq2) |
1032 | return rq2; | |
1033 | if (rq2 == NULL) | |
1034 | return rq1; | |
9c2c38a1 | 1035 | |
229836bd NK |
1036 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
1037 | return rq_is_sync(rq1) ? rq1 : rq2; | |
1038 | ||
65299a3b CH |
1039 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
1040 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 1041 | |
83096ebf TH |
1042 | s1 = blk_rq_pos(rq1); |
1043 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 1044 | |
1da177e4 LT |
1045 | /* |
1046 | * by definition, 1KiB is 2 sectors | |
1047 | */ | |
1048 | back_max = cfqd->cfq_back_max * 2; | |
1049 | ||
1050 | /* | |
1051 | * Strict one way elevator _except_ in the case where we allow | |
1052 | * short backward seeks which are biased as twice the cost of a | |
1053 | * similar forward seek. | |
1054 | */ | |
1055 | if (s1 >= last) | |
1056 | d1 = s1 - last; | |
1057 | else if (s1 + back_max >= last) | |
1058 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
1059 | else | |
e8a99053 | 1060 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
1061 | |
1062 | if (s2 >= last) | |
1063 | d2 = s2 - last; | |
1064 | else if (s2 + back_max >= last) | |
1065 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
1066 | else | |
e8a99053 | 1067 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
1068 | |
1069 | /* Found required data */ | |
e8a99053 AM |
1070 | |
1071 | /* | |
1072 | * By doing switch() on the bit mask "wrap" we avoid having to | |
1073 | * check two variables for all permutations: --> faster! | |
1074 | */ | |
1075 | switch (wrap) { | |
5e705374 | 1076 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 1077 | if (d1 < d2) |
5e705374 | 1078 | return rq1; |
e8a99053 | 1079 | else if (d2 < d1) |
5e705374 | 1080 | return rq2; |
e8a99053 AM |
1081 | else { |
1082 | if (s1 >= s2) | |
5e705374 | 1083 | return rq1; |
e8a99053 | 1084 | else |
5e705374 | 1085 | return rq2; |
e8a99053 | 1086 | } |
1da177e4 | 1087 | |
e8a99053 | 1088 | case CFQ_RQ2_WRAP: |
5e705374 | 1089 | return rq1; |
e8a99053 | 1090 | case CFQ_RQ1_WRAP: |
5e705374 JA |
1091 | return rq2; |
1092 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
1093 | default: |
1094 | /* | |
1095 | * Since both rqs are wrapped, | |
1096 | * start with the one that's further behind head | |
1097 | * (--> only *one* back seek required), | |
1098 | * since back seek takes more time than forward. | |
1099 | */ | |
1100 | if (s1 <= s2) | |
5e705374 | 1101 | return rq1; |
1da177e4 | 1102 | else |
5e705374 | 1103 | return rq2; |
1da177e4 LT |
1104 | } |
1105 | } | |
1106 | ||
498d3aa2 JA |
1107 | /* |
1108 | * The below is leftmost cache rbtree addon | |
1109 | */ | |
0871714e | 1110 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 1111 | { |
615f0259 VG |
1112 | /* Service tree is empty */ |
1113 | if (!root->count) | |
1114 | return NULL; | |
1115 | ||
cc09e299 JA |
1116 | if (!root->left) |
1117 | root->left = rb_first(&root->rb); | |
1118 | ||
0871714e JA |
1119 | if (root->left) |
1120 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
1121 | ||
1122 | return NULL; | |
cc09e299 JA |
1123 | } |
1124 | ||
1fa8f6d6 VG |
1125 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
1126 | { | |
1127 | if (!root->left) | |
1128 | root->left = rb_first(&root->rb); | |
1129 | ||
1130 | if (root->left) | |
1131 | return rb_entry_cfqg(root->left); | |
1132 | ||
1133 | return NULL; | |
1134 | } | |
1135 | ||
a36e71f9 JA |
1136 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
1137 | { | |
1138 | rb_erase(n, root); | |
1139 | RB_CLEAR_NODE(n); | |
1140 | } | |
1141 | ||
cc09e299 JA |
1142 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
1143 | { | |
1144 | if (root->left == n) | |
1145 | root->left = NULL; | |
a36e71f9 | 1146 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 1147 | --root->count; |
cc09e299 JA |
1148 | } |
1149 | ||
1da177e4 LT |
1150 | /* |
1151 | * would be nice to take fifo expire time into account as well | |
1152 | */ | |
5e705374 JA |
1153 | static struct request * |
1154 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1155 | struct request *last) | |
1da177e4 | 1156 | { |
21183b07 JA |
1157 | struct rb_node *rbnext = rb_next(&last->rb_node); |
1158 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 1159 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 1160 | |
21183b07 | 1161 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
1162 | |
1163 | if (rbprev) | |
5e705374 | 1164 | prev = rb_entry_rq(rbprev); |
1da177e4 | 1165 | |
21183b07 | 1166 | if (rbnext) |
5e705374 | 1167 | next = rb_entry_rq(rbnext); |
21183b07 JA |
1168 | else { |
1169 | rbnext = rb_first(&cfqq->sort_list); | |
1170 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 1171 | next = rb_entry_rq(rbnext); |
21183b07 | 1172 | } |
1da177e4 | 1173 | |
cf7c25cf | 1174 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
1175 | } |
1176 | ||
d9e7620e JA |
1177 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
1178 | struct cfq_queue *cfqq) | |
1da177e4 | 1179 | { |
d9e7620e JA |
1180 | /* |
1181 | * just an approximation, should be ok. | |
1182 | */ | |
cdb16e8f | 1183 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 1184 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
1185 | } |
1186 | ||
1fa8f6d6 VG |
1187 | static inline s64 |
1188 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1189 | { | |
1190 | return cfqg->vdisktime - st->min_vdisktime; | |
1191 | } | |
1192 | ||
1193 | static void | |
1194 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1195 | { | |
1196 | struct rb_node **node = &st->rb.rb_node; | |
1197 | struct rb_node *parent = NULL; | |
1198 | struct cfq_group *__cfqg; | |
1199 | s64 key = cfqg_key(st, cfqg); | |
1200 | int left = 1; | |
1201 | ||
1202 | while (*node != NULL) { | |
1203 | parent = *node; | |
1204 | __cfqg = rb_entry_cfqg(parent); | |
1205 | ||
1206 | if (key < cfqg_key(st, __cfqg)) | |
1207 | node = &parent->rb_left; | |
1208 | else { | |
1209 | node = &parent->rb_right; | |
1210 | left = 0; | |
1211 | } | |
1212 | } | |
1213 | ||
1214 | if (left) | |
1215 | st->left = &cfqg->rb_node; | |
1216 | ||
1217 | rb_link_node(&cfqg->rb_node, parent, node); | |
1218 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
1219 | } | |
1220 | ||
1221 | static void | |
8184f93e JT |
1222 | cfq_update_group_weight(struct cfq_group *cfqg) |
1223 | { | |
1224 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
e71357e1 | 1225 | |
3381cb8d | 1226 | if (cfqg->new_weight) { |
8184f93e | 1227 | cfqg->weight = cfqg->new_weight; |
3381cb8d | 1228 | cfqg->new_weight = 0; |
8184f93e | 1229 | } |
e71357e1 TH |
1230 | |
1231 | if (cfqg->new_leaf_weight) { | |
1232 | cfqg->leaf_weight = cfqg->new_leaf_weight; | |
1233 | cfqg->new_leaf_weight = 0; | |
1234 | } | |
8184f93e JT |
1235 | } |
1236 | ||
1237 | static void | |
1238 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1239 | { | |
7918ffb5 TH |
1240 | struct cfq_group *pos = cfqg; |
1241 | bool propagate; | |
1242 | ||
1243 | /* add to the service tree */ | |
8184f93e JT |
1244 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
1245 | ||
1246 | cfq_update_group_weight(cfqg); | |
1247 | __cfq_group_service_tree_add(st, cfqg); | |
1248 | st->total_weight += cfqg->weight; | |
7918ffb5 TH |
1249 | |
1250 | /* | |
1251 | * Activate @cfqg and propagate activation upwards until we meet an | |
1252 | * already activated node or reach root. | |
1253 | */ | |
1254 | propagate = !pos->nr_active++; | |
1255 | pos->children_weight += pos->leaf_weight; | |
1256 | ||
1257 | while (propagate) { | |
1258 | struct cfq_group *parent = cfqg_flat_parent(pos); | |
1259 | ||
1260 | if (!parent) | |
1261 | break; | |
1262 | ||
1263 | propagate = !parent->nr_active++; | |
1264 | parent->children_weight += pos->weight; | |
1265 | pos = parent; | |
1266 | } | |
8184f93e JT |
1267 | } |
1268 | ||
1269 | static void | |
1270 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
1271 | { |
1272 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1273 | struct cfq_group *__cfqg; | |
1274 | struct rb_node *n; | |
1275 | ||
1276 | cfqg->nr_cfqq++; | |
760701bf | 1277 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
1278 | return; |
1279 | ||
1280 | /* | |
1281 | * Currently put the group at the end. Later implement something | |
1282 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 1283 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
1284 | */ |
1285 | n = rb_last(&st->rb); | |
1286 | if (n) { | |
1287 | __cfqg = rb_entry_cfqg(n); | |
1288 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
1289 | } else | |
1290 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
1291 | cfq_group_service_tree_add(st, cfqg); |
1292 | } | |
1fa8f6d6 | 1293 | |
8184f93e JT |
1294 | static void |
1295 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1296 | { | |
7918ffb5 TH |
1297 | struct cfq_group *pos = cfqg; |
1298 | bool propagate; | |
1299 | ||
1300 | /* | |
1301 | * Undo activation from cfq_group_service_tree_add(). Deactivate | |
1302 | * @cfqg and propagate deactivation upwards. | |
1303 | */ | |
1304 | propagate = !--pos->nr_active; | |
1305 | pos->children_weight -= pos->leaf_weight; | |
1306 | ||
1307 | while (propagate) { | |
1308 | struct cfq_group *parent = cfqg_flat_parent(pos); | |
1309 | ||
1310 | /* @pos has 0 nr_active at this point */ | |
1311 | WARN_ON_ONCE(pos->children_weight); | |
1312 | ||
1313 | if (!parent) | |
1314 | break; | |
1315 | ||
1316 | propagate = !--parent->nr_active; | |
1317 | parent->children_weight -= pos->weight; | |
1318 | pos = parent; | |
1319 | } | |
1320 | ||
1321 | /* remove from the service tree */ | |
8184f93e JT |
1322 | st->total_weight -= cfqg->weight; |
1323 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) | |
1324 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
1325 | } |
1326 | ||
1327 | static void | |
8184f93e | 1328 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
1329 | { |
1330 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1331 | ||
1332 | BUG_ON(cfqg->nr_cfqq < 1); | |
1333 | cfqg->nr_cfqq--; | |
25bc6b07 | 1334 | |
1fa8f6d6 VG |
1335 | /* If there are other cfq queues under this group, don't delete it */ |
1336 | if (cfqg->nr_cfqq) | |
1337 | return; | |
1338 | ||
2868ef7b | 1339 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 1340 | cfq_group_service_tree_del(st, cfqg); |
4d2ceea4 | 1341 | cfqg->saved_wl_slice = 0; |
155fead9 | 1342 | cfqg_stats_update_dequeue(cfqg); |
dae739eb VG |
1343 | } |
1344 | ||
167400d3 JT |
1345 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
1346 | unsigned int *unaccounted_time) | |
dae739eb | 1347 | { |
f75edf2d | 1348 | unsigned int slice_used; |
dae739eb VG |
1349 | |
1350 | /* | |
1351 | * Queue got expired before even a single request completed or | |
1352 | * got expired immediately after first request completion. | |
1353 | */ | |
1354 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
1355 | /* | |
1356 | * Also charge the seek time incurred to the group, otherwise | |
1357 | * if there are mutiple queues in the group, each can dispatch | |
1358 | * a single request on seeky media and cause lots of seek time | |
1359 | * and group will never know it. | |
1360 | */ | |
1361 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
1362 | 1); | |
1363 | } else { | |
1364 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
1365 | if (slice_used > cfqq->allocated_slice) { |
1366 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 1367 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
1368 | } |
1369 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
1370 | *unaccounted_time += cfqq->slice_start - | |
1371 | cfqq->dispatch_start; | |
dae739eb VG |
1372 | } |
1373 | ||
dae739eb VG |
1374 | return slice_used; |
1375 | } | |
1376 | ||
1377 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1378 | struct cfq_queue *cfqq) |
dae739eb VG |
1379 | { |
1380 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 1381 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1382 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1383 | - cfqg->service_tree_idle.count; | |
1384 | ||
1385 | BUG_ON(nr_sync < 0); | |
167400d3 | 1386 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1387 | |
02b35081 VG |
1388 | if (iops_mode(cfqd)) |
1389 | charge = cfqq->slice_dispatch; | |
1390 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1391 | charge = cfqq->allocated_slice; | |
dae739eb VG |
1392 | |
1393 | /* Can't update vdisktime while group is on service tree */ | |
8184f93e | 1394 | cfq_group_service_tree_del(st, cfqg); |
02b35081 | 1395 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
8184f93e JT |
1396 | /* If a new weight was requested, update now, off tree */ |
1397 | cfq_group_service_tree_add(st, cfqg); | |
dae739eb VG |
1398 | |
1399 | /* This group is being expired. Save the context */ | |
1400 | if (time_after(cfqd->workload_expires, jiffies)) { | |
4d2ceea4 | 1401 | cfqg->saved_wl_slice = cfqd->workload_expires |
dae739eb | 1402 | - jiffies; |
4d2ceea4 VG |
1403 | cfqg->saved_wl_type = cfqd->serving_wl_type; |
1404 | cfqg->saved_wl_class = cfqd->serving_wl_class; | |
dae739eb | 1405 | } else |
4d2ceea4 | 1406 | cfqg->saved_wl_slice = 0; |
2868ef7b VG |
1407 | |
1408 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1409 | st->min_vdisktime); | |
fd16d263 JP |
1410 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1411 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1412 | used_sl, cfqq->slice_dispatch, charge, | |
1413 | iops_mode(cfqd), cfqq->nr_sectors); | |
155fead9 TH |
1414 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
1415 | cfqg_stats_set_start_empty_time(cfqg); | |
1fa8f6d6 VG |
1416 | } |
1417 | ||
f51b802c TH |
1418 | /** |
1419 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1420 | * @cfqg: cfq_group to initialize | |
1421 | * | |
1422 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1423 | * is enabled or not. | |
1424 | */ | |
1425 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1426 | { | |
1427 | struct cfq_rb_root *st; | |
1428 | int i, j; | |
1429 | ||
1430 | for_each_cfqg_st(cfqg, i, j, st) | |
1431 | *st = CFQ_RB_ROOT; | |
1432 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1433 | ||
1434 | cfqg->ttime.last_end_request = jiffies; | |
1435 | } | |
1436 | ||
25fb5169 | 1437 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 1438 | static void cfq_pd_init(struct blkcg_gq *blkg) |
f469a7b4 | 1439 | { |
0381411e | 1440 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
25fb5169 | 1441 | |
f51b802c | 1442 | cfq_init_cfqg_base(cfqg); |
3381cb8d | 1443 | cfqg->weight = blkg->blkcg->cfq_weight; |
e71357e1 | 1444 | cfqg->leaf_weight = blkg->blkcg->cfq_leaf_weight; |
25fb5169 VG |
1445 | } |
1446 | ||
1447 | /* | |
3e59cf9d VG |
1448 | * Search for the cfq group current task belongs to. request_queue lock must |
1449 | * be held. | |
25fb5169 | 1450 | */ |
cd1604fa | 1451 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
3c798398 | 1452 | struct blkcg *blkcg) |
25fb5169 | 1453 | { |
f469a7b4 | 1454 | struct request_queue *q = cfqd->queue; |
cd1604fa | 1455 | struct cfq_group *cfqg = NULL; |
25fb5169 | 1456 | |
3c798398 TH |
1457 | /* avoid lookup for the common case where there's no blkcg */ |
1458 | if (blkcg == &blkcg_root) { | |
cd1604fa TH |
1459 | cfqg = cfqd->root_group; |
1460 | } else { | |
3c798398 | 1461 | struct blkcg_gq *blkg; |
f469a7b4 | 1462 | |
3c96cb32 | 1463 | blkg = blkg_lookup_create(blkcg, q); |
cd1604fa | 1464 | if (!IS_ERR(blkg)) |
0381411e | 1465 | cfqg = blkg_to_cfqg(blkg); |
cd1604fa | 1466 | } |
f469a7b4 | 1467 | |
25fb5169 VG |
1468 | return cfqg; |
1469 | } | |
1470 | ||
1471 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1472 | { | |
1473 | /* Currently, all async queues are mapped to root group */ | |
1474 | if (!cfq_cfqq_sync(cfqq)) | |
f51b802c | 1475 | cfqg = cfqq->cfqd->root_group; |
25fb5169 VG |
1476 | |
1477 | cfqq->cfqg = cfqg; | |
b1c35769 | 1478 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1479 | cfqg_get(cfqg); |
b1c35769 VG |
1480 | } |
1481 | ||
f95a04af TH |
1482 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, |
1483 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1484 | { |
f95a04af | 1485 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
3381cb8d TH |
1486 | |
1487 | if (!cfqg->dev_weight) | |
60c2bc2d | 1488 | return 0; |
f95a04af | 1489 | return __blkg_prfill_u64(sf, pd, cfqg->dev_weight); |
60c2bc2d TH |
1490 | } |
1491 | ||
3381cb8d TH |
1492 | static int cfqg_print_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1493 | struct seq_file *sf) | |
60c2bc2d | 1494 | { |
3c798398 TH |
1495 | blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), |
1496 | cfqg_prfill_weight_device, &blkcg_policy_cfq, 0, | |
60c2bc2d TH |
1497 | false); |
1498 | return 0; | |
1499 | } | |
1500 | ||
e71357e1 TH |
1501 | static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf, |
1502 | struct blkg_policy_data *pd, int off) | |
1503 | { | |
1504 | struct cfq_group *cfqg = pd_to_cfqg(pd); | |
1505 | ||
1506 | if (!cfqg->dev_leaf_weight) | |
1507 | return 0; | |
1508 | return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight); | |
1509 | } | |
1510 | ||
1511 | static int cfqg_print_leaf_weight_device(struct cgroup *cgrp, | |
1512 | struct cftype *cft, | |
1513 | struct seq_file *sf) | |
1514 | { | |
1515 | blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), | |
1516 | cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq, 0, | |
1517 | false); | |
1518 | return 0; | |
1519 | } | |
1520 | ||
3381cb8d TH |
1521 | static int cfq_print_weight(struct cgroup *cgrp, struct cftype *cft, |
1522 | struct seq_file *sf) | |
60c2bc2d | 1523 | { |
3c798398 | 1524 | seq_printf(sf, "%u\n", cgroup_to_blkcg(cgrp)->cfq_weight); |
60c2bc2d TH |
1525 | return 0; |
1526 | } | |
1527 | ||
e71357e1 TH |
1528 | static int cfq_print_leaf_weight(struct cgroup *cgrp, struct cftype *cft, |
1529 | struct seq_file *sf) | |
1530 | { | |
1531 | seq_printf(sf, "%u\n", | |
1532 | cgroup_to_blkcg(cgrp)->cfq_leaf_weight); | |
1533 | return 0; | |
1534 | } | |
1535 | ||
1536 | static int __cfqg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, | |
1537 | const char *buf, bool is_leaf_weight) | |
60c2bc2d | 1538 | { |
3c798398 | 1539 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
60c2bc2d | 1540 | struct blkg_conf_ctx ctx; |
3381cb8d | 1541 | struct cfq_group *cfqg; |
60c2bc2d TH |
1542 | int ret; |
1543 | ||
3c798398 | 1544 | ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx); |
60c2bc2d TH |
1545 | if (ret) |
1546 | return ret; | |
1547 | ||
1548 | ret = -EINVAL; | |
3381cb8d | 1549 | cfqg = blkg_to_cfqg(ctx.blkg); |
a2b1693b | 1550 | if (!ctx.v || (ctx.v >= CFQ_WEIGHT_MIN && ctx.v <= CFQ_WEIGHT_MAX)) { |
e71357e1 TH |
1551 | if (!is_leaf_weight) { |
1552 | cfqg->dev_weight = ctx.v; | |
1553 | cfqg->new_weight = ctx.v ?: blkcg->cfq_weight; | |
1554 | } else { | |
1555 | cfqg->dev_leaf_weight = ctx.v; | |
1556 | cfqg->new_leaf_weight = ctx.v ?: blkcg->cfq_leaf_weight; | |
1557 | } | |
60c2bc2d TH |
1558 | ret = 0; |
1559 | } | |
1560 | ||
1561 | blkg_conf_finish(&ctx); | |
1562 | return ret; | |
1563 | } | |
1564 | ||
e71357e1 TH |
1565 | static int cfqg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1566 | const char *buf) | |
1567 | { | |
1568 | return __cfqg_set_weight_device(cgrp, cft, buf, false); | |
1569 | } | |
1570 | ||
1571 | static int cfqg_set_leaf_weight_device(struct cgroup *cgrp, struct cftype *cft, | |
1572 | const char *buf) | |
1573 | { | |
1574 | return __cfqg_set_weight_device(cgrp, cft, buf, true); | |
1575 | } | |
1576 | ||
1577 | static int __cfq_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val, | |
1578 | bool is_leaf_weight) | |
60c2bc2d | 1579 | { |
3c798398 TH |
1580 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
1581 | struct blkcg_gq *blkg; | |
60c2bc2d TH |
1582 | struct hlist_node *n; |
1583 | ||
3381cb8d | 1584 | if (val < CFQ_WEIGHT_MIN || val > CFQ_WEIGHT_MAX) |
60c2bc2d TH |
1585 | return -EINVAL; |
1586 | ||
1587 | spin_lock_irq(&blkcg->lock); | |
e71357e1 TH |
1588 | |
1589 | if (!is_leaf_weight) | |
1590 | blkcg->cfq_weight = val; | |
1591 | else | |
1592 | blkcg->cfq_leaf_weight = val; | |
60c2bc2d TH |
1593 | |
1594 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
3381cb8d | 1595 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
60c2bc2d | 1596 | |
e71357e1 TH |
1597 | if (!cfqg) |
1598 | continue; | |
1599 | ||
1600 | if (!is_leaf_weight) { | |
1601 | if (!cfqg->dev_weight) | |
1602 | cfqg->new_weight = blkcg->cfq_weight; | |
1603 | } else { | |
1604 | if (!cfqg->dev_leaf_weight) | |
1605 | cfqg->new_leaf_weight = blkcg->cfq_leaf_weight; | |
1606 | } | |
60c2bc2d TH |
1607 | } |
1608 | ||
1609 | spin_unlock_irq(&blkcg->lock); | |
1610 | return 0; | |
1611 | } | |
1612 | ||
e71357e1 TH |
1613 | static int cfq_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) |
1614 | { | |
1615 | return __cfq_set_weight(cgrp, cft, val, false); | |
1616 | } | |
1617 | ||
1618 | static int cfq_set_leaf_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1619 | { | |
1620 | return __cfq_set_weight(cgrp, cft, val, true); | |
1621 | } | |
1622 | ||
5bc4afb1 TH |
1623 | static int cfqg_print_stat(struct cgroup *cgrp, struct cftype *cft, |
1624 | struct seq_file *sf) | |
1625 | { | |
3c798398 | 1626 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
5bc4afb1 | 1627 | |
3c798398 | 1628 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_stat, &blkcg_policy_cfq, |
5bc4afb1 TH |
1629 | cft->private, false); |
1630 | return 0; | |
1631 | } | |
1632 | ||
1633 | static int cfqg_print_rwstat(struct cgroup *cgrp, struct cftype *cft, | |
1634 | struct seq_file *sf) | |
1635 | { | |
3c798398 | 1636 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
5bc4afb1 | 1637 | |
3c798398 | 1638 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_rwstat, &blkcg_policy_cfq, |
5bc4afb1 TH |
1639 | cft->private, true); |
1640 | return 0; | |
1641 | } | |
1642 | ||
60c2bc2d | 1643 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
f95a04af TH |
1644 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, |
1645 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1646 | { |
f95a04af | 1647 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
155fead9 | 1648 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
60c2bc2d TH |
1649 | u64 v = 0; |
1650 | ||
1651 | if (samples) { | |
155fead9 | 1652 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
60c2bc2d TH |
1653 | do_div(v, samples); |
1654 | } | |
f95a04af | 1655 | __blkg_prfill_u64(sf, pd, v); |
60c2bc2d TH |
1656 | return 0; |
1657 | } | |
1658 | ||
1659 | /* print avg_queue_size */ | |
155fead9 TH |
1660 | static int cfqg_print_avg_queue_size(struct cgroup *cgrp, struct cftype *cft, |
1661 | struct seq_file *sf) | |
60c2bc2d | 1662 | { |
3c798398 | 1663 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
60c2bc2d | 1664 | |
155fead9 | 1665 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_avg_queue_size, |
3c798398 | 1666 | &blkcg_policy_cfq, 0, false); |
60c2bc2d TH |
1667 | return 0; |
1668 | } | |
1669 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1670 | ||
1671 | static struct cftype cfq_blkcg_files[] = { | |
1672 | { | |
1673 | .name = "weight_device", | |
3381cb8d TH |
1674 | .read_seq_string = cfqg_print_weight_device, |
1675 | .write_string = cfqg_set_weight_device, | |
60c2bc2d TH |
1676 | .max_write_len = 256, |
1677 | }, | |
1678 | { | |
1679 | .name = "weight", | |
3381cb8d TH |
1680 | .read_seq_string = cfq_print_weight, |
1681 | .write_u64 = cfq_set_weight, | |
60c2bc2d | 1682 | }, |
e71357e1 TH |
1683 | |
1684 | /* on root, leaf_weight is mapped to weight */ | |
1685 | { | |
1686 | .name = "leaf_weight_device", | |
1687 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1688 | .read_seq_string = cfqg_print_weight_device, | |
1689 | .write_string = cfqg_set_weight_device, | |
1690 | .max_write_len = 256, | |
1691 | }, | |
1692 | { | |
1693 | .name = "leaf_weight", | |
1694 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1695 | .read_seq_string = cfq_print_weight, | |
1696 | .write_u64 = cfq_set_weight, | |
1697 | }, | |
1698 | ||
1699 | /* no such mapping necessary for !roots */ | |
1700 | { | |
1701 | .name = "leaf_weight_device", | |
1702 | .flags = CFTYPE_NOT_ON_ROOT, | |
1703 | .read_seq_string = cfqg_print_leaf_weight_device, | |
1704 | .write_string = cfqg_set_leaf_weight_device, | |
1705 | .max_write_len = 256, | |
1706 | }, | |
1707 | { | |
1708 | .name = "leaf_weight", | |
1709 | .flags = CFTYPE_NOT_ON_ROOT, | |
1710 | .read_seq_string = cfq_print_leaf_weight, | |
1711 | .write_u64 = cfq_set_leaf_weight, | |
1712 | }, | |
1713 | ||
60c2bc2d TH |
1714 | { |
1715 | .name = "time", | |
5bc4afb1 TH |
1716 | .private = offsetof(struct cfq_group, stats.time), |
1717 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1718 | }, |
1719 | { | |
1720 | .name = "sectors", | |
5bc4afb1 TH |
1721 | .private = offsetof(struct cfq_group, stats.sectors), |
1722 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1723 | }, |
1724 | { | |
1725 | .name = "io_service_bytes", | |
5bc4afb1 TH |
1726 | .private = offsetof(struct cfq_group, stats.service_bytes), |
1727 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1728 | }, |
1729 | { | |
1730 | .name = "io_serviced", | |
5bc4afb1 TH |
1731 | .private = offsetof(struct cfq_group, stats.serviced), |
1732 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1733 | }, |
1734 | { | |
1735 | .name = "io_service_time", | |
5bc4afb1 TH |
1736 | .private = offsetof(struct cfq_group, stats.service_time), |
1737 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1738 | }, |
1739 | { | |
1740 | .name = "io_wait_time", | |
5bc4afb1 TH |
1741 | .private = offsetof(struct cfq_group, stats.wait_time), |
1742 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1743 | }, |
1744 | { | |
1745 | .name = "io_merged", | |
5bc4afb1 TH |
1746 | .private = offsetof(struct cfq_group, stats.merged), |
1747 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1748 | }, |
1749 | { | |
1750 | .name = "io_queued", | |
5bc4afb1 TH |
1751 | .private = offsetof(struct cfq_group, stats.queued), |
1752 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1753 | }, |
1754 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
1755 | { | |
1756 | .name = "avg_queue_size", | |
155fead9 | 1757 | .read_seq_string = cfqg_print_avg_queue_size, |
60c2bc2d TH |
1758 | }, |
1759 | { | |
1760 | .name = "group_wait_time", | |
5bc4afb1 TH |
1761 | .private = offsetof(struct cfq_group, stats.group_wait_time), |
1762 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1763 | }, |
1764 | { | |
1765 | .name = "idle_time", | |
5bc4afb1 TH |
1766 | .private = offsetof(struct cfq_group, stats.idle_time), |
1767 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1768 | }, |
1769 | { | |
1770 | .name = "empty_time", | |
5bc4afb1 TH |
1771 | .private = offsetof(struct cfq_group, stats.empty_time), |
1772 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1773 | }, |
1774 | { | |
1775 | .name = "dequeue", | |
5bc4afb1 TH |
1776 | .private = offsetof(struct cfq_group, stats.dequeue), |
1777 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1778 | }, |
1779 | { | |
1780 | .name = "unaccounted_time", | |
5bc4afb1 TH |
1781 | .private = offsetof(struct cfq_group, stats.unaccounted_time), |
1782 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1783 | }, |
1784 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1785 | { } /* terminate */ | |
1786 | }; | |
25fb5169 | 1787 | #else /* GROUP_IOSCHED */ |
cd1604fa | 1788 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
3c798398 | 1789 | struct blkcg *blkcg) |
25fb5169 | 1790 | { |
f51b802c | 1791 | return cfqd->root_group; |
25fb5169 | 1792 | } |
7f1dc8a2 | 1793 | |
25fb5169 VG |
1794 | static inline void |
1795 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1796 | cfqq->cfqg = cfqg; | |
1797 | } | |
1798 | ||
1799 | #endif /* GROUP_IOSCHED */ | |
1800 | ||
498d3aa2 | 1801 | /* |
c0324a02 | 1802 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1803 | * requests waiting to be processed. It is sorted in the order that |
1804 | * we will service the queues. | |
1805 | */ | |
a36e71f9 | 1806 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1807 | bool add_front) |
d9e7620e | 1808 | { |
0871714e JA |
1809 | struct rb_node **p, *parent; |
1810 | struct cfq_queue *__cfqq; | |
d9e7620e | 1811 | unsigned long rb_key; |
34b98d03 | 1812 | struct cfq_rb_root *st; |
498d3aa2 | 1813 | int left; |
dae739eb | 1814 | int new_cfqq = 1; |
ae30c286 | 1815 | |
34b98d03 | 1816 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq)); |
0871714e JA |
1817 | if (cfq_class_idle(cfqq)) { |
1818 | rb_key = CFQ_IDLE_DELAY; | |
34b98d03 | 1819 | parent = rb_last(&st->rb); |
0871714e JA |
1820 | if (parent && parent != &cfqq->rb_node) { |
1821 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1822 | rb_key += __cfqq->rb_key; | |
1823 | } else | |
1824 | rb_key += jiffies; | |
1825 | } else if (!add_front) { | |
b9c8946b JA |
1826 | /* |
1827 | * Get our rb key offset. Subtract any residual slice | |
1828 | * value carried from last service. A negative resid | |
1829 | * count indicates slice overrun, and this should position | |
1830 | * the next service time further away in the tree. | |
1831 | */ | |
edd75ffd | 1832 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1833 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1834 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1835 | } else { |
1836 | rb_key = -HZ; | |
34b98d03 | 1837 | __cfqq = cfq_rb_first(st); |
48e025e6 CZ |
1838 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1839 | } | |
1da177e4 | 1840 | |
d9e7620e | 1841 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1842 | new_cfqq = 0; |
99f9628a | 1843 | /* |
d9e7620e | 1844 | * same position, nothing more to do |
99f9628a | 1845 | */ |
34b98d03 | 1846 | if (rb_key == cfqq->rb_key && cfqq->service_tree == st) |
d9e7620e | 1847 | return; |
1da177e4 | 1848 | |
aa6f6a3d CZ |
1849 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1850 | cfqq->service_tree = NULL; | |
1da177e4 | 1851 | } |
d9e7620e | 1852 | |
498d3aa2 | 1853 | left = 1; |
0871714e | 1854 | parent = NULL; |
34b98d03 VG |
1855 | cfqq->service_tree = st; |
1856 | p = &st->rb.rb_node; | |
d9e7620e JA |
1857 | while (*p) { |
1858 | parent = *p; | |
1859 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1860 | ||
0c534e0a | 1861 | /* |
c0324a02 | 1862 | * sort by key, that represents service time. |
0c534e0a | 1863 | */ |
c0324a02 | 1864 | if (time_before(rb_key, __cfqq->rb_key)) |
1f23f121 | 1865 | p = &parent->rb_left; |
c0324a02 | 1866 | else { |
1f23f121 | 1867 | p = &parent->rb_right; |
cc09e299 | 1868 | left = 0; |
c0324a02 | 1869 | } |
d9e7620e JA |
1870 | } |
1871 | ||
cc09e299 | 1872 | if (left) |
34b98d03 | 1873 | st->left = &cfqq->rb_node; |
cc09e299 | 1874 | |
d9e7620e JA |
1875 | cfqq->rb_key = rb_key; |
1876 | rb_link_node(&cfqq->rb_node, parent, p); | |
34b98d03 VG |
1877 | rb_insert_color(&cfqq->rb_node, &st->rb); |
1878 | st->count++; | |
20359f27 | 1879 | if (add_front || !new_cfqq) |
dae739eb | 1880 | return; |
8184f93e | 1881 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1882 | } |
1883 | ||
a36e71f9 | 1884 | static struct cfq_queue * |
f2d1f0ae JA |
1885 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1886 | sector_t sector, struct rb_node **ret_parent, | |
1887 | struct rb_node ***rb_link) | |
a36e71f9 | 1888 | { |
a36e71f9 JA |
1889 | struct rb_node **p, *parent; |
1890 | struct cfq_queue *cfqq = NULL; | |
1891 | ||
1892 | parent = NULL; | |
1893 | p = &root->rb_node; | |
1894 | while (*p) { | |
1895 | struct rb_node **n; | |
1896 | ||
1897 | parent = *p; | |
1898 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1899 | ||
1900 | /* | |
1901 | * Sort strictly based on sector. Smallest to the left, | |
1902 | * largest to the right. | |
1903 | */ | |
2e46e8b2 | 1904 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1905 | n = &(*p)->rb_right; |
2e46e8b2 | 1906 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1907 | n = &(*p)->rb_left; |
1908 | else | |
1909 | break; | |
1910 | p = n; | |
3ac6c9f8 | 1911 | cfqq = NULL; |
a36e71f9 JA |
1912 | } |
1913 | ||
1914 | *ret_parent = parent; | |
1915 | if (rb_link) | |
1916 | *rb_link = p; | |
3ac6c9f8 | 1917 | return cfqq; |
a36e71f9 JA |
1918 | } |
1919 | ||
1920 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1921 | { | |
a36e71f9 JA |
1922 | struct rb_node **p, *parent; |
1923 | struct cfq_queue *__cfqq; | |
1924 | ||
f2d1f0ae JA |
1925 | if (cfqq->p_root) { |
1926 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1927 | cfqq->p_root = NULL; | |
1928 | } | |
a36e71f9 JA |
1929 | |
1930 | if (cfq_class_idle(cfqq)) | |
1931 | return; | |
1932 | if (!cfqq->next_rq) | |
1933 | return; | |
1934 | ||
f2d1f0ae | 1935 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1936 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1937 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1938 | if (!__cfqq) { |
1939 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1940 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1941 | } else | |
1942 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1943 | } |
1944 | ||
498d3aa2 JA |
1945 | /* |
1946 | * Update cfqq's position in the service tree. | |
1947 | */ | |
edd75ffd | 1948 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1949 | { |
6d048f53 JA |
1950 | /* |
1951 | * Resorting requires the cfqq to be on the RR list already. | |
1952 | */ | |
a36e71f9 | 1953 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1954 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1955 | cfq_prio_tree_add(cfqd, cfqq); |
1956 | } | |
6d048f53 JA |
1957 | } |
1958 | ||
1da177e4 LT |
1959 | /* |
1960 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1961 | * the pending list according to last request service |
1da177e4 | 1962 | */ |
febffd61 | 1963 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1964 | { |
7b679138 | 1965 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1966 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1967 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 1968 | cfqd->busy_queues++; |
ef8a41df SL |
1969 | if (cfq_cfqq_sync(cfqq)) |
1970 | cfqd->busy_sync_queues++; | |
1da177e4 | 1971 | |
edd75ffd | 1972 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1973 | } |
1974 | ||
498d3aa2 JA |
1975 | /* |
1976 | * Called when the cfqq no longer has requests pending, remove it from | |
1977 | * the service tree. | |
1978 | */ | |
febffd61 | 1979 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1980 | { |
7b679138 | 1981 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1982 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1983 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1984 | |
aa6f6a3d CZ |
1985 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1986 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1987 | cfqq->service_tree = NULL; | |
1988 | } | |
f2d1f0ae JA |
1989 | if (cfqq->p_root) { |
1990 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1991 | cfqq->p_root = NULL; | |
1992 | } | |
d9e7620e | 1993 | |
8184f93e | 1994 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1995 | BUG_ON(!cfqd->busy_queues); |
1996 | cfqd->busy_queues--; | |
ef8a41df SL |
1997 | if (cfq_cfqq_sync(cfqq)) |
1998 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
1999 | } |
2000 | ||
2001 | /* | |
2002 | * rb tree support functions | |
2003 | */ | |
febffd61 | 2004 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 2005 | { |
5e705374 | 2006 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 2007 | const int sync = rq_is_sync(rq); |
1da177e4 | 2008 | |
b4878f24 JA |
2009 | BUG_ON(!cfqq->queued[sync]); |
2010 | cfqq->queued[sync]--; | |
1da177e4 | 2011 | |
5e705374 | 2012 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 2013 | |
f04a6424 VG |
2014 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
2015 | /* | |
2016 | * Queue will be deleted from service tree when we actually | |
2017 | * expire it later. Right now just remove it from prio tree | |
2018 | * as it is empty. | |
2019 | */ | |
2020 | if (cfqq->p_root) { | |
2021 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2022 | cfqq->p_root = NULL; | |
2023 | } | |
2024 | } | |
1da177e4 LT |
2025 | } |
2026 | ||
5e705374 | 2027 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 2028 | { |
5e705374 | 2029 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2030 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 2031 | struct request *prev; |
1da177e4 | 2032 | |
5380a101 | 2033 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 2034 | |
796d5116 | 2035 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
2036 | |
2037 | if (!cfq_cfqq_on_rr(cfqq)) | |
2038 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
2039 | |
2040 | /* | |
2041 | * check if this request is a better next-serve candidate | |
2042 | */ | |
a36e71f9 | 2043 | prev = cfqq->next_rq; |
cf7c25cf | 2044 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
2045 | |
2046 | /* | |
2047 | * adjust priority tree position, if ->next_rq changes | |
2048 | */ | |
2049 | if (prev != cfqq->next_rq) | |
2050 | cfq_prio_tree_add(cfqd, cfqq); | |
2051 | ||
5044eed4 | 2052 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
2053 | } |
2054 | ||
febffd61 | 2055 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 2056 | { |
5380a101 JA |
2057 | elv_rb_del(&cfqq->sort_list, rq); |
2058 | cfqq->queued[rq_is_sync(rq)]--; | |
155fead9 | 2059 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
5e705374 | 2060 | cfq_add_rq_rb(rq); |
155fead9 TH |
2061 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
2062 | rq->cmd_flags); | |
1da177e4 LT |
2063 | } |
2064 | ||
206dc69b JA |
2065 | static struct request * |
2066 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 2067 | { |
206dc69b | 2068 | struct task_struct *tsk = current; |
c5869807 | 2069 | struct cfq_io_cq *cic; |
206dc69b | 2070 | struct cfq_queue *cfqq; |
1da177e4 | 2071 | |
4ac845a2 | 2072 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2073 | if (!cic) |
2074 | return NULL; | |
2075 | ||
2076 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
2077 | if (cfqq) { |
2078 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
2079 | ||
21183b07 | 2080 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 2081 | } |
1da177e4 | 2082 | |
1da177e4 LT |
2083 | return NULL; |
2084 | } | |
2085 | ||
165125e1 | 2086 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2087 | { |
22e2c507 | 2088 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 2089 | |
53c583d2 | 2090 | cfqd->rq_in_driver++; |
7b679138 | 2091 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 2092 | cfqd->rq_in_driver); |
25776e35 | 2093 | |
5b93629b | 2094 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
2095 | } |
2096 | ||
165125e1 | 2097 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2098 | { |
b4878f24 JA |
2099 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2100 | ||
53c583d2 CZ |
2101 | WARN_ON(!cfqd->rq_in_driver); |
2102 | cfqd->rq_in_driver--; | |
7b679138 | 2103 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 2104 | cfqd->rq_in_driver); |
1da177e4 LT |
2105 | } |
2106 | ||
b4878f24 | 2107 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 2108 | { |
5e705374 | 2109 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 2110 | |
5e705374 JA |
2111 | if (cfqq->next_rq == rq) |
2112 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 2113 | |
b4878f24 | 2114 | list_del_init(&rq->queuelist); |
5e705374 | 2115 | cfq_del_rq_rb(rq); |
374f84ac | 2116 | |
45333d5a | 2117 | cfqq->cfqd->rq_queued--; |
155fead9 | 2118 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
65299a3b CH |
2119 | if (rq->cmd_flags & REQ_PRIO) { |
2120 | WARN_ON(!cfqq->prio_pending); | |
2121 | cfqq->prio_pending--; | |
b53d1ed7 | 2122 | } |
1da177e4 LT |
2123 | } |
2124 | ||
165125e1 JA |
2125 | static int cfq_merge(struct request_queue *q, struct request **req, |
2126 | struct bio *bio) | |
1da177e4 LT |
2127 | { |
2128 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2129 | struct request *__rq; | |
1da177e4 | 2130 | |
206dc69b | 2131 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 2132 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
2133 | *req = __rq; |
2134 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
2135 | } |
2136 | ||
2137 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
2138 | } |
2139 | ||
165125e1 | 2140 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 2141 | int type) |
1da177e4 | 2142 | { |
21183b07 | 2143 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 2144 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 2145 | |
5e705374 | 2146 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 2147 | } |
1da177e4 LT |
2148 | } |
2149 | ||
812d4026 DS |
2150 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
2151 | struct bio *bio) | |
2152 | { | |
155fead9 | 2153 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_rw); |
812d4026 DS |
2154 | } |
2155 | ||
1da177e4 | 2156 | static void |
165125e1 | 2157 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
2158 | struct request *next) |
2159 | { | |
cf7c25cf | 2160 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
2161 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2162 | ||
22e2c507 JA |
2163 | /* |
2164 | * reposition in fifo if next is older than rq | |
2165 | */ | |
2166 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
3d106fba SL |
2167 | time_before(rq_fifo_time(next), rq_fifo_time(rq)) && |
2168 | cfqq == RQ_CFQQ(next)) { | |
22e2c507 | 2169 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
2170 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
2171 | } | |
22e2c507 | 2172 | |
cf7c25cf CZ |
2173 | if (cfqq->next_rq == next) |
2174 | cfqq->next_rq = rq; | |
b4878f24 | 2175 | cfq_remove_request(next); |
155fead9 | 2176 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
4a0b75c7 SL |
2177 | |
2178 | cfqq = RQ_CFQQ(next); | |
2179 | /* | |
2180 | * all requests of this queue are merged to other queues, delete it | |
2181 | * from the service tree. If it's the active_queue, | |
2182 | * cfq_dispatch_requests() will choose to expire it or do idle | |
2183 | */ | |
2184 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
2185 | cfqq != cfqd->active_queue) | |
2186 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
2187 | } |
2188 | ||
165125e1 | 2189 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
2190 | struct bio *bio) |
2191 | { | |
2192 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 2193 | struct cfq_io_cq *cic; |
da775265 | 2194 | struct cfq_queue *cfqq; |
da775265 JA |
2195 | |
2196 | /* | |
ec8acb69 | 2197 | * Disallow merge of a sync bio into an async request. |
da775265 | 2198 | */ |
91fac317 | 2199 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 2200 | return false; |
da775265 JA |
2201 | |
2202 | /* | |
f1a4f4d3 | 2203 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 2204 | * merge only if rq is queued there. |
f1a4f4d3 | 2205 | */ |
07c2bd37 TH |
2206 | cic = cfq_cic_lookup(cfqd, current->io_context); |
2207 | if (!cic) | |
2208 | return false; | |
719d3402 | 2209 | |
91fac317 | 2210 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 2211 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
2212 | } |
2213 | ||
812df48d DS |
2214 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2215 | { | |
2216 | del_timer(&cfqd->idle_slice_timer); | |
155fead9 | 2217 | cfqg_stats_update_idle_time(cfqq->cfqg); |
812df48d DS |
2218 | } |
2219 | ||
febffd61 JA |
2220 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
2221 | struct cfq_queue *cfqq) | |
22e2c507 JA |
2222 | { |
2223 | if (cfqq) { | |
3bf10fea | 2224 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d", |
4d2ceea4 | 2225 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
155fead9 | 2226 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
62a37f6b JT |
2227 | cfqq->slice_start = 0; |
2228 | cfqq->dispatch_start = jiffies; | |
2229 | cfqq->allocated_slice = 0; | |
2230 | cfqq->slice_end = 0; | |
2231 | cfqq->slice_dispatch = 0; | |
2232 | cfqq->nr_sectors = 0; | |
2233 | ||
2234 | cfq_clear_cfqq_wait_request(cfqq); | |
2235 | cfq_clear_cfqq_must_dispatch(cfqq); | |
2236 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
2237 | cfq_clear_cfqq_fifo_expire(cfqq); | |
2238 | cfq_mark_cfqq_slice_new(cfqq); | |
2239 | ||
2240 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
2241 | } |
2242 | ||
2243 | cfqd->active_queue = cfqq; | |
2244 | } | |
2245 | ||
7b14e3b5 JA |
2246 | /* |
2247 | * current cfqq expired its slice (or was too idle), select new one | |
2248 | */ | |
2249 | static void | |
2250 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 2251 | bool timed_out) |
7b14e3b5 | 2252 | { |
7b679138 JA |
2253 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
2254 | ||
7b14e3b5 | 2255 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 2256 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 2257 | |
7b14e3b5 | 2258 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 2259 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 2260 | |
ae54abed SL |
2261 | /* |
2262 | * If this cfqq is shared between multiple processes, check to | |
2263 | * make sure that those processes are still issuing I/Os within | |
2264 | * the mean seek distance. If not, it may be time to break the | |
2265 | * queues apart again. | |
2266 | */ | |
2267 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
2268 | cfq_mark_cfqq_split_coop(cfqq); | |
2269 | ||
7b14e3b5 | 2270 | /* |
6084cdda | 2271 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 2272 | */ |
c553f8e3 SL |
2273 | if (timed_out) { |
2274 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 2275 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
2276 | else |
2277 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
2278 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
2279 | } | |
7b14e3b5 | 2280 | |
e5ff082e | 2281 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 2282 | |
f04a6424 VG |
2283 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
2284 | cfq_del_cfqq_rr(cfqd, cfqq); | |
2285 | ||
edd75ffd | 2286 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
2287 | |
2288 | if (cfqq == cfqd->active_queue) | |
2289 | cfqd->active_queue = NULL; | |
2290 | ||
2291 | if (cfqd->active_cic) { | |
11a3122f | 2292 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
2293 | cfqd->active_cic = NULL; |
2294 | } | |
7b14e3b5 JA |
2295 | } |
2296 | ||
e5ff082e | 2297 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
2298 | { |
2299 | struct cfq_queue *cfqq = cfqd->active_queue; | |
2300 | ||
2301 | if (cfqq) | |
e5ff082e | 2302 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
2303 | } |
2304 | ||
498d3aa2 JA |
2305 | /* |
2306 | * Get next queue for service. Unless we have a queue preemption, | |
2307 | * we'll simply select the first cfqq in the service tree. | |
2308 | */ | |
6d048f53 | 2309 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 2310 | { |
34b98d03 VG |
2311 | struct cfq_rb_root *st = st_for(cfqd->serving_group, |
2312 | cfqd->serving_wl_class, cfqd->serving_wl_type); | |
d9e7620e | 2313 | |
f04a6424 VG |
2314 | if (!cfqd->rq_queued) |
2315 | return NULL; | |
2316 | ||
1fa8f6d6 | 2317 | /* There is nothing to dispatch */ |
34b98d03 | 2318 | if (!st) |
1fa8f6d6 | 2319 | return NULL; |
34b98d03 | 2320 | if (RB_EMPTY_ROOT(&st->rb)) |
c0324a02 | 2321 | return NULL; |
34b98d03 | 2322 | return cfq_rb_first(st); |
6d048f53 JA |
2323 | } |
2324 | ||
f04a6424 VG |
2325 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
2326 | { | |
25fb5169 | 2327 | struct cfq_group *cfqg; |
f04a6424 VG |
2328 | struct cfq_queue *cfqq; |
2329 | int i, j; | |
2330 | struct cfq_rb_root *st; | |
2331 | ||
2332 | if (!cfqd->rq_queued) | |
2333 | return NULL; | |
2334 | ||
25fb5169 VG |
2335 | cfqg = cfq_get_next_cfqg(cfqd); |
2336 | if (!cfqg) | |
2337 | return NULL; | |
2338 | ||
f04a6424 VG |
2339 | for_each_cfqg_st(cfqg, i, j, st) |
2340 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
2341 | return cfqq; | |
2342 | return NULL; | |
2343 | } | |
2344 | ||
498d3aa2 JA |
2345 | /* |
2346 | * Get and set a new active queue for service. | |
2347 | */ | |
a36e71f9 JA |
2348 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
2349 | struct cfq_queue *cfqq) | |
6d048f53 | 2350 | { |
e00ef799 | 2351 | if (!cfqq) |
a36e71f9 | 2352 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 2353 | |
22e2c507 | 2354 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 2355 | return cfqq; |
22e2c507 JA |
2356 | } |
2357 | ||
d9e7620e JA |
2358 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
2359 | struct request *rq) | |
2360 | { | |
83096ebf TH |
2361 | if (blk_rq_pos(rq) >= cfqd->last_position) |
2362 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 2363 | else |
83096ebf | 2364 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
2365 | } |
2366 | ||
b2c18e1e | 2367 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 2368 | struct request *rq) |
6d048f53 | 2369 | { |
e9ce335d | 2370 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
2371 | } |
2372 | ||
a36e71f9 JA |
2373 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
2374 | struct cfq_queue *cur_cfqq) | |
2375 | { | |
f2d1f0ae | 2376 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
2377 | struct rb_node *parent, *node; |
2378 | struct cfq_queue *__cfqq; | |
2379 | sector_t sector = cfqd->last_position; | |
2380 | ||
2381 | if (RB_EMPTY_ROOT(root)) | |
2382 | return NULL; | |
2383 | ||
2384 | /* | |
2385 | * First, if we find a request starting at the end of the last | |
2386 | * request, choose it. | |
2387 | */ | |
f2d1f0ae | 2388 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
2389 | if (__cfqq) |
2390 | return __cfqq; | |
2391 | ||
2392 | /* | |
2393 | * If the exact sector wasn't found, the parent of the NULL leaf | |
2394 | * will contain the closest sector. | |
2395 | */ | |
2396 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 2397 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2398 | return __cfqq; |
2399 | ||
2e46e8b2 | 2400 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
2401 | node = rb_next(&__cfqq->p_node); |
2402 | else | |
2403 | node = rb_prev(&__cfqq->p_node); | |
2404 | if (!node) | |
2405 | return NULL; | |
2406 | ||
2407 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 2408 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2409 | return __cfqq; |
2410 | ||
2411 | return NULL; | |
2412 | } | |
2413 | ||
2414 | /* | |
2415 | * cfqd - obvious | |
2416 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
2417 | * closely cooperating with itself. | |
2418 | * | |
2419 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
2420 | * one request, and that cfqd->last_position reflects a position on the disk | |
2421 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
2422 | * assumption. | |
2423 | */ | |
2424 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 2425 | struct cfq_queue *cur_cfqq) |
6d048f53 | 2426 | { |
a36e71f9 JA |
2427 | struct cfq_queue *cfqq; |
2428 | ||
39c01b21 DS |
2429 | if (cfq_class_idle(cur_cfqq)) |
2430 | return NULL; | |
e6c5bc73 JM |
2431 | if (!cfq_cfqq_sync(cur_cfqq)) |
2432 | return NULL; | |
2433 | if (CFQQ_SEEKY(cur_cfqq)) | |
2434 | return NULL; | |
2435 | ||
b9d8f4c7 GJ |
2436 | /* |
2437 | * Don't search priority tree if it's the only queue in the group. | |
2438 | */ | |
2439 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
2440 | return NULL; | |
2441 | ||
6d048f53 | 2442 | /* |
d9e7620e JA |
2443 | * We should notice if some of the queues are cooperating, eg |
2444 | * working closely on the same area of the disk. In that case, | |
2445 | * we can group them together and don't waste time idling. | |
6d048f53 | 2446 | */ |
a36e71f9 JA |
2447 | cfqq = cfqq_close(cfqd, cur_cfqq); |
2448 | if (!cfqq) | |
2449 | return NULL; | |
2450 | ||
8682e1f1 VG |
2451 | /* If new queue belongs to different cfq_group, don't choose it */ |
2452 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
2453 | return NULL; | |
2454 | ||
df5fe3e8 JM |
2455 | /* |
2456 | * It only makes sense to merge sync queues. | |
2457 | */ | |
2458 | if (!cfq_cfqq_sync(cfqq)) | |
2459 | return NULL; | |
e6c5bc73 JM |
2460 | if (CFQQ_SEEKY(cfqq)) |
2461 | return NULL; | |
df5fe3e8 | 2462 | |
c0324a02 CZ |
2463 | /* |
2464 | * Do not merge queues of different priority classes | |
2465 | */ | |
2466 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
2467 | return NULL; | |
2468 | ||
a36e71f9 | 2469 | return cfqq; |
6d048f53 JA |
2470 | } |
2471 | ||
a6d44e98 CZ |
2472 | /* |
2473 | * Determine whether we should enforce idle window for this queue. | |
2474 | */ | |
2475 | ||
2476 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2477 | { | |
3bf10fea | 2478 | enum wl_class_t wl_class = cfqq_class(cfqq); |
34b98d03 | 2479 | struct cfq_rb_root *st = cfqq->service_tree; |
a6d44e98 | 2480 | |
34b98d03 VG |
2481 | BUG_ON(!st); |
2482 | BUG_ON(!st->count); | |
f04a6424 | 2483 | |
b6508c16 VG |
2484 | if (!cfqd->cfq_slice_idle) |
2485 | return false; | |
2486 | ||
a6d44e98 | 2487 | /* We never do for idle class queues. */ |
3bf10fea | 2488 | if (wl_class == IDLE_WORKLOAD) |
a6d44e98 CZ |
2489 | return false; |
2490 | ||
2491 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
2492 | if (cfq_cfqq_idle_window(cfqq) && |
2493 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
2494 | return true; |
2495 | ||
2496 | /* | |
2497 | * Otherwise, we do only if they are the last ones | |
2498 | * in their service tree. | |
2499 | */ | |
34b98d03 VG |
2500 | if (st->count == 1 && cfq_cfqq_sync(cfqq) && |
2501 | !cfq_io_thinktime_big(cfqd, &st->ttime, false)) | |
c1e44756 | 2502 | return true; |
34b98d03 | 2503 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count); |
c1e44756 | 2504 | return false; |
a6d44e98 CZ |
2505 | } |
2506 | ||
6d048f53 | 2507 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2508 | { |
1792669c | 2509 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 2510 | struct cfq_io_cq *cic; |
80bdf0c7 | 2511 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 2512 | |
a68bbddb | 2513 | /* |
f7d7b7a7 JA |
2514 | * SSD device without seek penalty, disable idling. But only do so |
2515 | * for devices that support queuing, otherwise we still have a problem | |
2516 | * with sync vs async workloads. | |
a68bbddb | 2517 | */ |
f7d7b7a7 | 2518 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
2519 | return; |
2520 | ||
dd67d051 | 2521 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2522 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2523 | |
2524 | /* | |
2525 | * idle is disabled, either manually or by past process history | |
2526 | */ | |
80bdf0c7 VG |
2527 | if (!cfq_should_idle(cfqd, cfqq)) { |
2528 | /* no queue idling. Check for group idling */ | |
2529 | if (cfqd->cfq_group_idle) | |
2530 | group_idle = cfqd->cfq_group_idle; | |
2531 | else | |
2532 | return; | |
2533 | } | |
6d048f53 | 2534 | |
7b679138 | 2535 | /* |
8e550632 | 2536 | * still active requests from this queue, don't idle |
7b679138 | 2537 | */ |
8e550632 | 2538 | if (cfqq->dispatched) |
7b679138 JA |
2539 | return; |
2540 | ||
22e2c507 JA |
2541 | /* |
2542 | * task has exited, don't wait | |
2543 | */ | |
206dc69b | 2544 | cic = cfqd->active_cic; |
f6e8d01b | 2545 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
2546 | return; |
2547 | ||
355b659c CZ |
2548 | /* |
2549 | * If our average think time is larger than the remaining time | |
2550 | * slice, then don't idle. This avoids overrunning the allotted | |
2551 | * time slice. | |
2552 | */ | |
383cd721 SL |
2553 | if (sample_valid(cic->ttime.ttime_samples) && |
2554 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 2555 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 2556 | cic->ttime.ttime_mean); |
355b659c | 2557 | return; |
b1ffe737 | 2558 | } |
355b659c | 2559 | |
80bdf0c7 VG |
2560 | /* There are other queues in the group, don't do group idle */ |
2561 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
2562 | return; | |
2563 | ||
3b18152c | 2564 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2565 | |
80bdf0c7 VG |
2566 | if (group_idle) |
2567 | sl = cfqd->cfq_group_idle; | |
2568 | else | |
2569 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2570 | |
7b14e3b5 | 2571 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
155fead9 | 2572 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
80bdf0c7 VG |
2573 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
2574 | group_idle ? 1 : 0); | |
1da177e4 LT |
2575 | } |
2576 | ||
498d3aa2 JA |
2577 | /* |
2578 | * Move request from internal lists to the request queue dispatch list. | |
2579 | */ | |
165125e1 | 2580 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2581 | { |
3ed9a296 | 2582 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2583 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2584 | |
7b679138 JA |
2585 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2586 | ||
06d21886 | 2587 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2588 | cfq_remove_request(rq); |
6d048f53 | 2589 | cfqq->dispatched++; |
80bdf0c7 | 2590 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2591 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2592 | |
53c583d2 | 2593 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2594 | cfqq->nr_sectors += blk_rq_sectors(rq); |
155fead9 | 2595 | cfqg_stats_update_dispatch(cfqq->cfqg, blk_rq_bytes(rq), rq->cmd_flags); |
1da177e4 LT |
2596 | } |
2597 | ||
2598 | /* | |
2599 | * return expired entry, or NULL to just start from scratch in rbtree | |
2600 | */ | |
febffd61 | 2601 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 2602 | { |
30996f40 | 2603 | struct request *rq = NULL; |
1da177e4 | 2604 | |
3b18152c | 2605 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 2606 | return NULL; |
cb887411 JA |
2607 | |
2608 | cfq_mark_cfqq_fifo_expire(cfqq); | |
2609 | ||
89850f7e JA |
2610 | if (list_empty(&cfqq->fifo)) |
2611 | return NULL; | |
1da177e4 | 2612 | |
89850f7e | 2613 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 2614 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 2615 | rq = NULL; |
1da177e4 | 2616 | |
30996f40 | 2617 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 2618 | return rq; |
1da177e4 LT |
2619 | } |
2620 | ||
22e2c507 JA |
2621 | static inline int |
2622 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2623 | { | |
2624 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2625 | |
22e2c507 | 2626 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2627 | |
b9f8ce05 | 2628 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
2629 | } |
2630 | ||
df5fe3e8 JM |
2631 | /* |
2632 | * Must be called with the queue_lock held. | |
2633 | */ | |
2634 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2635 | { | |
2636 | int process_refs, io_refs; | |
2637 | ||
2638 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 2639 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
2640 | BUG_ON(process_refs < 0); |
2641 | return process_refs; | |
2642 | } | |
2643 | ||
2644 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2645 | { | |
e6c5bc73 | 2646 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2647 | struct cfq_queue *__cfqq; |
2648 | ||
c10b61f0 JM |
2649 | /* |
2650 | * If there are no process references on the new_cfqq, then it is | |
2651 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2652 | * chain may have dropped their last reference (not just their | |
2653 | * last process reference). | |
2654 | */ | |
2655 | if (!cfqq_process_refs(new_cfqq)) | |
2656 | return; | |
2657 | ||
df5fe3e8 JM |
2658 | /* Avoid a circular list and skip interim queue merges */ |
2659 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2660 | if (__cfqq == cfqq) | |
2661 | return; | |
2662 | new_cfqq = __cfqq; | |
2663 | } | |
2664 | ||
2665 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2666 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2667 | /* |
2668 | * If the process for the cfqq has gone away, there is no | |
2669 | * sense in merging the queues. | |
2670 | */ | |
c10b61f0 | 2671 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2672 | return; |
2673 | ||
e6c5bc73 JM |
2674 | /* |
2675 | * Merge in the direction of the lesser amount of work. | |
2676 | */ | |
e6c5bc73 JM |
2677 | if (new_process_refs >= process_refs) { |
2678 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2679 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2680 | } else { |
2681 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2682 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2683 | } |
df5fe3e8 JM |
2684 | } |
2685 | ||
6d816ec7 | 2686 | static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd, |
3bf10fea | 2687 | struct cfq_group *cfqg, enum wl_class_t wl_class) |
718eee05 CZ |
2688 | { |
2689 | struct cfq_queue *queue; | |
2690 | int i; | |
2691 | bool key_valid = false; | |
2692 | unsigned long lowest_key = 0; | |
2693 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2694 | ||
65b32a57 VG |
2695 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2696 | /* select the one with lowest rb_key */ | |
34b98d03 | 2697 | queue = cfq_rb_first(st_for(cfqg, wl_class, i)); |
718eee05 CZ |
2698 | if (queue && |
2699 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2700 | lowest_key = queue->rb_key; | |
2701 | cur_best = i; | |
2702 | key_valid = true; | |
2703 | } | |
2704 | } | |
2705 | ||
2706 | return cur_best; | |
2707 | } | |
2708 | ||
6d816ec7 VG |
2709 | static void |
2710 | choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
718eee05 | 2711 | { |
718eee05 CZ |
2712 | unsigned slice; |
2713 | unsigned count; | |
cdb16e8f | 2714 | struct cfq_rb_root *st; |
58ff82f3 | 2715 | unsigned group_slice; |
4d2ceea4 | 2716 | enum wl_class_t original_class = cfqd->serving_wl_class; |
1fa8f6d6 | 2717 | |
718eee05 | 2718 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2719 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 2720 | cfqd->serving_wl_class = RT_WORKLOAD; |
58ff82f3 | 2721 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 2722 | cfqd->serving_wl_class = BE_WORKLOAD; |
718eee05 | 2723 | else { |
4d2ceea4 | 2724 | cfqd->serving_wl_class = IDLE_WORKLOAD; |
718eee05 CZ |
2725 | cfqd->workload_expires = jiffies + 1; |
2726 | return; | |
2727 | } | |
2728 | ||
4d2ceea4 | 2729 | if (original_class != cfqd->serving_wl_class) |
e4ea0c16 SL |
2730 | goto new_workload; |
2731 | ||
718eee05 CZ |
2732 | /* |
2733 | * For RT and BE, we have to choose also the type | |
2734 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2735 | * expiration time | |
2736 | */ | |
34b98d03 | 2737 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 2738 | count = st->count; |
718eee05 CZ |
2739 | |
2740 | /* | |
65b32a57 | 2741 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2742 | */ |
65b32a57 | 2743 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2744 | return; |
2745 | ||
e4ea0c16 | 2746 | new_workload: |
718eee05 | 2747 | /* otherwise select new workload type */ |
6d816ec7 | 2748 | cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg, |
4d2ceea4 | 2749 | cfqd->serving_wl_class); |
34b98d03 | 2750 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 2751 | count = st->count; |
718eee05 CZ |
2752 | |
2753 | /* | |
2754 | * the workload slice is computed as a fraction of target latency | |
2755 | * proportional to the number of queues in that workload, over | |
2756 | * all the queues in the same priority class | |
2757 | */ | |
58ff82f3 VG |
2758 | group_slice = cfq_group_slice(cfqd, cfqg); |
2759 | ||
2760 | slice = group_slice * count / | |
4d2ceea4 VG |
2761 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class], |
2762 | cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd, | |
3bf10fea | 2763 | cfqg)); |
718eee05 | 2764 | |
4d2ceea4 | 2765 | if (cfqd->serving_wl_type == ASYNC_WORKLOAD) { |
f26bd1f0 VG |
2766 | unsigned int tmp; |
2767 | ||
2768 | /* | |
2769 | * Async queues are currently system wide. Just taking | |
2770 | * proportion of queues with-in same group will lead to higher | |
2771 | * async ratio system wide as generally root group is going | |
2772 | * to have higher weight. A more accurate thing would be to | |
2773 | * calculate system wide asnc/sync ratio. | |
2774 | */ | |
5bf14c07 TM |
2775 | tmp = cfqd->cfq_target_latency * |
2776 | cfqg_busy_async_queues(cfqd, cfqg); | |
f26bd1f0 VG |
2777 | tmp = tmp/cfqd->busy_queues; |
2778 | slice = min_t(unsigned, slice, tmp); | |
2779 | ||
718eee05 CZ |
2780 | /* async workload slice is scaled down according to |
2781 | * the sync/async slice ratio. */ | |
2782 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2783 | } else |
718eee05 CZ |
2784 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2785 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2786 | ||
2787 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2788 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
2789 | cfqd->workload_expires = jiffies + slice; |
2790 | } | |
2791 | ||
1fa8f6d6 VG |
2792 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2793 | { | |
2794 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2795 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2796 | |
2797 | if (RB_EMPTY_ROOT(&st->rb)) | |
2798 | return NULL; | |
25bc6b07 | 2799 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
2800 | update_min_vdisktime(st); |
2801 | return cfqg; | |
1fa8f6d6 VG |
2802 | } |
2803 | ||
cdb16e8f VG |
2804 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2805 | { | |
1fa8f6d6 VG |
2806 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2807 | ||
2808 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2809 | |
2810 | /* Restore the workload type data */ | |
4d2ceea4 VG |
2811 | if (cfqg->saved_wl_slice) { |
2812 | cfqd->workload_expires = jiffies + cfqg->saved_wl_slice; | |
2813 | cfqd->serving_wl_type = cfqg->saved_wl_type; | |
2814 | cfqd->serving_wl_class = cfqg->saved_wl_class; | |
66ae2919 GJ |
2815 | } else |
2816 | cfqd->workload_expires = jiffies - 1; | |
2817 | ||
6d816ec7 | 2818 | choose_wl_class_and_type(cfqd, cfqg); |
cdb16e8f VG |
2819 | } |
2820 | ||
22e2c507 | 2821 | /* |
498d3aa2 JA |
2822 | * Select a queue for service. If we have a current active queue, |
2823 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2824 | */ |
1b5ed5e1 | 2825 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2826 | { |
a36e71f9 | 2827 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2828 | |
22e2c507 JA |
2829 | cfqq = cfqd->active_queue; |
2830 | if (!cfqq) | |
2831 | goto new_queue; | |
1da177e4 | 2832 | |
f04a6424 VG |
2833 | if (!cfqd->rq_queued) |
2834 | return NULL; | |
c244bb50 VG |
2835 | |
2836 | /* | |
2837 | * We were waiting for group to get backlogged. Expire the queue | |
2838 | */ | |
2839 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2840 | goto expire; | |
2841 | ||
22e2c507 | 2842 | /* |
6d048f53 | 2843 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2844 | */ |
7667aa06 VG |
2845 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2846 | /* | |
2847 | * If slice had not expired at the completion of last request | |
2848 | * we might not have turned on wait_busy flag. Don't expire | |
2849 | * the queue yet. Allow the group to get backlogged. | |
2850 | * | |
2851 | * The very fact that we have used the slice, that means we | |
2852 | * have been idling all along on this queue and it should be | |
2853 | * ok to wait for this request to complete. | |
2854 | */ | |
82bbbf28 VG |
2855 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2856 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2857 | cfqq = NULL; | |
7667aa06 | 2858 | goto keep_queue; |
82bbbf28 | 2859 | } else |
80bdf0c7 | 2860 | goto check_group_idle; |
7667aa06 | 2861 | } |
1da177e4 | 2862 | |
22e2c507 | 2863 | /* |
6d048f53 JA |
2864 | * The active queue has requests and isn't expired, allow it to |
2865 | * dispatch. | |
22e2c507 | 2866 | */ |
dd67d051 | 2867 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2868 | goto keep_queue; |
6d048f53 | 2869 | |
a36e71f9 JA |
2870 | /* |
2871 | * If another queue has a request waiting within our mean seek | |
2872 | * distance, let it run. The expire code will check for close | |
2873 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2874 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2875 | */ |
b3b6d040 | 2876 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2877 | if (new_cfqq) { |
2878 | if (!cfqq->new_cfqq) | |
2879 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2880 | goto expire; |
df5fe3e8 | 2881 | } |
a36e71f9 | 2882 | |
6d048f53 JA |
2883 | /* |
2884 | * No requests pending. If the active queue still has requests in | |
2885 | * flight or is idling for a new request, allow either of these | |
2886 | * conditions to happen (or time out) before selecting a new queue. | |
2887 | */ | |
80bdf0c7 VG |
2888 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2889 | cfqq = NULL; | |
2890 | goto keep_queue; | |
2891 | } | |
2892 | ||
8e1ac665 SL |
2893 | /* |
2894 | * This is a deep seek queue, but the device is much faster than | |
2895 | * the queue can deliver, don't idle | |
2896 | **/ | |
2897 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
2898 | (cfq_cfqq_slice_new(cfqq) || | |
2899 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
2900 | cfq_clear_cfqq_deep(cfqq); | |
2901 | cfq_clear_cfqq_idle_window(cfqq); | |
2902 | } | |
2903 | ||
80bdf0c7 VG |
2904 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2905 | cfqq = NULL; | |
2906 | goto keep_queue; | |
2907 | } | |
2908 | ||
2909 | /* | |
2910 | * If group idle is enabled and there are requests dispatched from | |
2911 | * this group, wait for requests to complete. | |
2912 | */ | |
2913 | check_group_idle: | |
7700fc4f SL |
2914 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
2915 | cfqq->cfqg->dispatched && | |
2916 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
2917 | cfqq = NULL; |
2918 | goto keep_queue; | |
22e2c507 JA |
2919 | } |
2920 | ||
3b18152c | 2921 | expire: |
e5ff082e | 2922 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2923 | new_queue: |
718eee05 CZ |
2924 | /* |
2925 | * Current queue expired. Check if we have to switch to a new | |
2926 | * service tree | |
2927 | */ | |
2928 | if (!new_cfqq) | |
cdb16e8f | 2929 | cfq_choose_cfqg(cfqd); |
718eee05 | 2930 | |
a36e71f9 | 2931 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2932 | keep_queue: |
3b18152c | 2933 | return cfqq; |
22e2c507 JA |
2934 | } |
2935 | ||
febffd61 | 2936 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2937 | { |
2938 | int dispatched = 0; | |
2939 | ||
2940 | while (cfqq->next_rq) { | |
2941 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2942 | dispatched++; | |
2943 | } | |
2944 | ||
2945 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2946 | |
2947 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 2948 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
2949 | return dispatched; |
2950 | } | |
2951 | ||
498d3aa2 JA |
2952 | /* |
2953 | * Drain our current requests. Used for barriers and when switching | |
2954 | * io schedulers on-the-fly. | |
2955 | */ | |
d9e7620e | 2956 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2957 | { |
0871714e | 2958 | struct cfq_queue *cfqq; |
d9e7620e | 2959 | int dispatched = 0; |
cdb16e8f | 2960 | |
3440c49f | 2961 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 2962 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
2963 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2964 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 2965 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 2966 | } |
1b5ed5e1 | 2967 | |
1b5ed5e1 TH |
2968 | BUG_ON(cfqd->busy_queues); |
2969 | ||
6923715a | 2970 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2971 | return dispatched; |
2972 | } | |
2973 | ||
abc3c744 SL |
2974 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2975 | struct cfq_queue *cfqq) | |
2976 | { | |
2977 | /* the queue hasn't finished any request, can't estimate */ | |
2978 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 2979 | return true; |
abc3c744 SL |
2980 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
2981 | cfqq->slice_end)) | |
c1e44756 | 2982 | return true; |
abc3c744 | 2983 | |
c1e44756 | 2984 | return false; |
abc3c744 SL |
2985 | } |
2986 | ||
0b182d61 | 2987 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2988 | { |
2f5cb738 | 2989 | unsigned int max_dispatch; |
22e2c507 | 2990 | |
5ad531db JA |
2991 | /* |
2992 | * Drain async requests before we start sync IO | |
2993 | */ | |
53c583d2 | 2994 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 2995 | return false; |
5ad531db | 2996 | |
2f5cb738 JA |
2997 | /* |
2998 | * If this is an async queue and we have sync IO in flight, let it wait | |
2999 | */ | |
53c583d2 | 3000 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 3001 | return false; |
2f5cb738 | 3002 | |
abc3c744 | 3003 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
3004 | if (cfq_class_idle(cfqq)) |
3005 | max_dispatch = 1; | |
b4878f24 | 3006 | |
2f5cb738 JA |
3007 | /* |
3008 | * Does this cfqq already have too much IO in flight? | |
3009 | */ | |
3010 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 3011 | bool promote_sync = false; |
2f5cb738 JA |
3012 | /* |
3013 | * idle queue must always only have a single IO in flight | |
3014 | */ | |
3ed9a296 | 3015 | if (cfq_class_idle(cfqq)) |
0b182d61 | 3016 | return false; |
3ed9a296 | 3017 | |
ef8a41df | 3018 | /* |
c4ade94f LS |
3019 | * If there is only one sync queue |
3020 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
3021 | * queue no dispatch limit. The reason is a sync queue can |
3022 | * preempt async queue, limiting the sync queue doesn't make | |
3023 | * sense. This is useful for aiostress test. | |
3024 | */ | |
c4ade94f LS |
3025 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
3026 | promote_sync = true; | |
ef8a41df | 3027 | |
2f5cb738 JA |
3028 | /* |
3029 | * We have other queues, don't allow more IO from this one | |
3030 | */ | |
ef8a41df SL |
3031 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
3032 | !promote_sync) | |
0b182d61 | 3033 | return false; |
9ede209e | 3034 | |
365722bb | 3035 | /* |
474b18cc | 3036 | * Sole queue user, no limit |
365722bb | 3037 | */ |
ef8a41df | 3038 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
3039 | max_dispatch = -1; |
3040 | else | |
3041 | /* | |
3042 | * Normally we start throttling cfqq when cfq_quantum/2 | |
3043 | * requests have been dispatched. But we can drive | |
3044 | * deeper queue depths at the beginning of slice | |
3045 | * subjected to upper limit of cfq_quantum. | |
3046 | * */ | |
3047 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
3048 | } |
3049 | ||
3050 | /* | |
3051 | * Async queues must wait a bit before being allowed dispatch. | |
3052 | * We also ramp up the dispatch depth gradually for async IO, | |
3053 | * based on the last sync IO we serviced | |
3054 | */ | |
963b72fc | 3055 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 3056 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 3057 | unsigned int depth; |
365722bb | 3058 | |
61f0c1dc | 3059 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
3060 | if (!depth && !cfqq->dispatched) |
3061 | depth = 1; | |
8e296755 JA |
3062 | if (depth < max_dispatch) |
3063 | max_dispatch = depth; | |
2f5cb738 | 3064 | } |
3ed9a296 | 3065 | |
0b182d61 JA |
3066 | /* |
3067 | * If we're below the current max, allow a dispatch | |
3068 | */ | |
3069 | return cfqq->dispatched < max_dispatch; | |
3070 | } | |
3071 | ||
3072 | /* | |
3073 | * Dispatch a request from cfqq, moving them to the request queue | |
3074 | * dispatch list. | |
3075 | */ | |
3076 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3077 | { | |
3078 | struct request *rq; | |
3079 | ||
3080 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
3081 | ||
3082 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
3083 | return false; | |
3084 | ||
3085 | /* | |
3086 | * follow expired path, else get first next available | |
3087 | */ | |
3088 | rq = cfq_check_fifo(cfqq); | |
3089 | if (!rq) | |
3090 | rq = cfqq->next_rq; | |
3091 | ||
3092 | /* | |
3093 | * insert request into driver dispatch list | |
3094 | */ | |
3095 | cfq_dispatch_insert(cfqd->queue, rq); | |
3096 | ||
3097 | if (!cfqd->active_cic) { | |
c5869807 | 3098 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 3099 | |
c5869807 | 3100 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
3101 | cfqd->active_cic = cic; |
3102 | } | |
3103 | ||
3104 | return true; | |
3105 | } | |
3106 | ||
3107 | /* | |
3108 | * Find the cfqq that we need to service and move a request from that to the | |
3109 | * dispatch list | |
3110 | */ | |
3111 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
3112 | { | |
3113 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3114 | struct cfq_queue *cfqq; | |
3115 | ||
3116 | if (!cfqd->busy_queues) | |
3117 | return 0; | |
3118 | ||
3119 | if (unlikely(force)) | |
3120 | return cfq_forced_dispatch(cfqd); | |
3121 | ||
3122 | cfqq = cfq_select_queue(cfqd); | |
3123 | if (!cfqq) | |
8e296755 JA |
3124 | return 0; |
3125 | ||
2f5cb738 | 3126 | /* |
0b182d61 | 3127 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 3128 | */ |
0b182d61 JA |
3129 | if (!cfq_dispatch_request(cfqd, cfqq)) |
3130 | return 0; | |
3131 | ||
2f5cb738 | 3132 | cfqq->slice_dispatch++; |
b029195d | 3133 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 3134 | |
2f5cb738 JA |
3135 | /* |
3136 | * expire an async queue immediately if it has used up its slice. idle | |
3137 | * queue always expire after 1 dispatch round. | |
3138 | */ | |
3139 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
3140 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
3141 | cfq_class_idle(cfqq))) { | |
3142 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 3143 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
3144 | } |
3145 | ||
b217a903 | 3146 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 3147 | return 1; |
1da177e4 LT |
3148 | } |
3149 | ||
1da177e4 | 3150 | /* |
5e705374 JA |
3151 | * task holds one reference to the queue, dropped when task exits. each rq |
3152 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 3153 | * |
b1c35769 | 3154 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
3155 | * queue lock must be held here. |
3156 | */ | |
3157 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
3158 | { | |
22e2c507 | 3159 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 3160 | struct cfq_group *cfqg; |
22e2c507 | 3161 | |
30d7b944 | 3162 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 3163 | |
30d7b944 SL |
3164 | cfqq->ref--; |
3165 | if (cfqq->ref) | |
1da177e4 LT |
3166 | return; |
3167 | ||
7b679138 | 3168 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 3169 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 3170 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 3171 | cfqg = cfqq->cfqg; |
1da177e4 | 3172 | |
28f95cbc | 3173 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 3174 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 3175 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 3176 | } |
22e2c507 | 3177 | |
f04a6424 | 3178 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 3179 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 3180 | cfqg_put(cfqg); |
1da177e4 LT |
3181 | } |
3182 | ||
d02a2c07 | 3183 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 3184 | { |
df5fe3e8 JM |
3185 | struct cfq_queue *__cfqq, *next; |
3186 | ||
df5fe3e8 JM |
3187 | /* |
3188 | * If this queue was scheduled to merge with another queue, be | |
3189 | * sure to drop the reference taken on that queue (and others in | |
3190 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
3191 | */ | |
3192 | __cfqq = cfqq->new_cfqq; | |
3193 | while (__cfqq) { | |
3194 | if (__cfqq == cfqq) { | |
3195 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
3196 | break; | |
3197 | } | |
3198 | next = __cfqq->new_cfqq; | |
3199 | cfq_put_queue(__cfqq); | |
3200 | __cfqq = next; | |
3201 | } | |
d02a2c07 SL |
3202 | } |
3203 | ||
3204 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3205 | { | |
3206 | if (unlikely(cfqq == cfqd->active_queue)) { | |
3207 | __cfq_slice_expired(cfqd, cfqq, 0); | |
3208 | cfq_schedule_dispatch(cfqd); | |
3209 | } | |
3210 | ||
3211 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 3212 | |
89850f7e JA |
3213 | cfq_put_queue(cfqq); |
3214 | } | |
22e2c507 | 3215 | |
9b84cacd TH |
3216 | static void cfq_init_icq(struct io_cq *icq) |
3217 | { | |
3218 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
3219 | ||
3220 | cic->ttime.last_end_request = jiffies; | |
3221 | } | |
3222 | ||
c5869807 | 3223 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 3224 | { |
c5869807 | 3225 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 3226 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 3227 | |
ff6657c6 JA |
3228 | if (cic->cfqq[BLK_RW_ASYNC]) { |
3229 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
3230 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
3231 | } |
3232 | ||
ff6657c6 JA |
3233 | if (cic->cfqq[BLK_RW_SYNC]) { |
3234 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
3235 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 3236 | } |
89850f7e JA |
3237 | } |
3238 | ||
abede6da | 3239 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
3240 | { |
3241 | struct task_struct *tsk = current; | |
3242 | int ioprio_class; | |
3243 | ||
3b18152c | 3244 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
3245 | return; |
3246 | ||
598971bf | 3247 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 3248 | switch (ioprio_class) { |
fe094d98 JA |
3249 | default: |
3250 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
3251 | case IOPRIO_CLASS_NONE: | |
3252 | /* | |
6d63c275 | 3253 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
3254 | */ |
3255 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 3256 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
3257 | break; |
3258 | case IOPRIO_CLASS_RT: | |
598971bf | 3259 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3260 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
3261 | break; | |
3262 | case IOPRIO_CLASS_BE: | |
598971bf | 3263 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3264 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3265 | break; | |
3266 | case IOPRIO_CLASS_IDLE: | |
3267 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
3268 | cfqq->ioprio = 7; | |
3269 | cfq_clear_cfqq_idle_window(cfqq); | |
3270 | break; | |
22e2c507 JA |
3271 | } |
3272 | ||
3273 | /* | |
3274 | * keep track of original prio settings in case we have to temporarily | |
3275 | * elevate the priority of this queue | |
3276 | */ | |
3277 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 3278 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
3279 | } |
3280 | ||
598971bf | 3281 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 3282 | { |
598971bf | 3283 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 3284 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 3285 | struct cfq_queue *cfqq; |
35e6077c | 3286 | |
598971bf TH |
3287 | /* |
3288 | * Check whether ioprio has changed. The condition may trigger | |
3289 | * spuriously on a newly created cic but there's no harm. | |
3290 | */ | |
3291 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
3292 | return; |
3293 | ||
ff6657c6 | 3294 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
3295 | if (cfqq) { |
3296 | struct cfq_queue *new_cfqq; | |
abede6da TH |
3297 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio, |
3298 | GFP_ATOMIC); | |
caaa5f9f | 3299 | if (new_cfqq) { |
ff6657c6 | 3300 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
3301 | cfq_put_queue(cfqq); |
3302 | } | |
22e2c507 | 3303 | } |
caaa5f9f | 3304 | |
ff6657c6 | 3305 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
3306 | if (cfqq) |
3307 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
3308 | |
3309 | cic->ioprio = ioprio; | |
22e2c507 JA |
3310 | } |
3311 | ||
d5036d77 | 3312 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 3313 | pid_t pid, bool is_sync) |
d5036d77 JA |
3314 | { |
3315 | RB_CLEAR_NODE(&cfqq->rb_node); | |
3316 | RB_CLEAR_NODE(&cfqq->p_node); | |
3317 | INIT_LIST_HEAD(&cfqq->fifo); | |
3318 | ||
30d7b944 | 3319 | cfqq->ref = 0; |
d5036d77 JA |
3320 | cfqq->cfqd = cfqd; |
3321 | ||
3322 | cfq_mark_cfqq_prio_changed(cfqq); | |
3323 | ||
3324 | if (is_sync) { | |
3325 | if (!cfq_class_idle(cfqq)) | |
3326 | cfq_mark_cfqq_idle_window(cfqq); | |
3327 | cfq_mark_cfqq_sync(cfqq); | |
3328 | } | |
3329 | cfqq->pid = pid; | |
3330 | } | |
3331 | ||
24610333 | 3332 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
598971bf | 3333 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 3334 | { |
bca4b914 | 3335 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
598971bf TH |
3336 | struct cfq_queue *sync_cfqq; |
3337 | uint64_t id; | |
24610333 | 3338 | |
598971bf | 3339 | rcu_read_lock(); |
3c798398 | 3340 | id = bio_blkcg(bio)->id; |
598971bf | 3341 | rcu_read_unlock(); |
24610333 | 3342 | |
598971bf TH |
3343 | /* |
3344 | * Check whether blkcg has changed. The condition may trigger | |
3345 | * spuriously on a newly created cic but there's no harm. | |
3346 | */ | |
3347 | if (unlikely(!cfqd) || likely(cic->blkcg_id == id)) | |
3348 | return; | |
24610333 | 3349 | |
598971bf | 3350 | sync_cfqq = cic_to_cfqq(cic, 1); |
24610333 VG |
3351 | if (sync_cfqq) { |
3352 | /* | |
3353 | * Drop reference to sync queue. A new sync queue will be | |
3354 | * assigned in new group upon arrival of a fresh request. | |
3355 | */ | |
3356 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
3357 | cic_set_cfqq(cic, NULL, 1); | |
3358 | cfq_put_queue(sync_cfqq); | |
3359 | } | |
598971bf TH |
3360 | |
3361 | cic->blkcg_id = id; | |
24610333 | 3362 | } |
598971bf TH |
3363 | #else |
3364 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } | |
24610333 VG |
3365 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
3366 | ||
22e2c507 | 3367 | static struct cfq_queue * |
abede6da TH |
3368 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
3369 | struct bio *bio, gfp_t gfp_mask) | |
22e2c507 | 3370 | { |
3c798398 | 3371 | struct blkcg *blkcg; |
22e2c507 | 3372 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
cdb16e8f | 3373 | struct cfq_group *cfqg; |
22e2c507 JA |
3374 | |
3375 | retry: | |
2a7f1244 TH |
3376 | rcu_read_lock(); |
3377 | ||
3c798398 | 3378 | blkcg = bio_blkcg(bio); |
cd1604fa | 3379 | cfqg = cfq_lookup_create_cfqg(cfqd, blkcg); |
91fac317 | 3380 | cfqq = cic_to_cfqq(cic, is_sync); |
22e2c507 | 3381 | |
6118b70b JA |
3382 | /* |
3383 | * Always try a new alloc if we fell back to the OOM cfqq | |
3384 | * originally, since it should just be a temporary situation. | |
3385 | */ | |
3386 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
3387 | cfqq = NULL; | |
22e2c507 JA |
3388 | if (new_cfqq) { |
3389 | cfqq = new_cfqq; | |
3390 | new_cfqq = NULL; | |
3391 | } else if (gfp_mask & __GFP_WAIT) { | |
2a7f1244 | 3392 | rcu_read_unlock(); |
22e2c507 | 3393 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c | 3394 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 3395 | gfp_mask | __GFP_ZERO, |
94f6030c | 3396 | cfqd->queue->node); |
22e2c507 | 3397 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
3398 | if (new_cfqq) |
3399 | goto retry; | |
22e2c507 | 3400 | } else { |
94f6030c CL |
3401 | cfqq = kmem_cache_alloc_node(cfq_pool, |
3402 | gfp_mask | __GFP_ZERO, | |
3403 | cfqd->queue->node); | |
22e2c507 JA |
3404 | } |
3405 | ||
6118b70b JA |
3406 | if (cfqq) { |
3407 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
abede6da | 3408 | cfq_init_prio_data(cfqq, cic); |
cdb16e8f | 3409 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
3410 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
3411 | } else | |
3412 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
3413 | } |
3414 | ||
3415 | if (new_cfqq) | |
3416 | kmem_cache_free(cfq_pool, new_cfqq); | |
3417 | ||
2a7f1244 | 3418 | rcu_read_unlock(); |
22e2c507 JA |
3419 | return cfqq; |
3420 | } | |
3421 | ||
c2dea2d1 VT |
3422 | static struct cfq_queue ** |
3423 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
3424 | { | |
fe094d98 | 3425 | switch (ioprio_class) { |
c2dea2d1 VT |
3426 | case IOPRIO_CLASS_RT: |
3427 | return &cfqd->async_cfqq[0][ioprio]; | |
598971bf TH |
3428 | case IOPRIO_CLASS_NONE: |
3429 | ioprio = IOPRIO_NORM; | |
3430 | /* fall through */ | |
c2dea2d1 VT |
3431 | case IOPRIO_CLASS_BE: |
3432 | return &cfqd->async_cfqq[1][ioprio]; | |
3433 | case IOPRIO_CLASS_IDLE: | |
3434 | return &cfqd->async_idle_cfqq; | |
3435 | default: | |
3436 | BUG(); | |
3437 | } | |
3438 | } | |
3439 | ||
15c31be4 | 3440 | static struct cfq_queue * |
abede6da | 3441 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
4f85cb96 | 3442 | struct bio *bio, gfp_t gfp_mask) |
15c31be4 | 3443 | { |
598971bf TH |
3444 | const int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
3445 | const int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
c2dea2d1 | 3446 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
3447 | struct cfq_queue *cfqq = NULL; |
3448 | ||
c2dea2d1 VT |
3449 | if (!is_sync) { |
3450 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
3451 | cfqq = *async_cfqq; | |
3452 | } | |
3453 | ||
6118b70b | 3454 | if (!cfqq) |
abede6da | 3455 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, cic, bio, gfp_mask); |
15c31be4 JA |
3456 | |
3457 | /* | |
3458 | * pin the queue now that it's allocated, scheduler exit will prune it | |
3459 | */ | |
c2dea2d1 | 3460 | if (!is_sync && !(*async_cfqq)) { |
30d7b944 | 3461 | cfqq->ref++; |
c2dea2d1 | 3462 | *async_cfqq = cfqq; |
15c31be4 JA |
3463 | } |
3464 | ||
30d7b944 | 3465 | cfqq->ref++; |
15c31be4 JA |
3466 | return cfqq; |
3467 | } | |
3468 | ||
22e2c507 | 3469 | static void |
383cd721 | 3470 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 3471 | { |
383cd721 SL |
3472 | unsigned long elapsed = jiffies - ttime->last_end_request; |
3473 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 3474 | |
383cd721 SL |
3475 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
3476 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
3477 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
3478 | } | |
3479 | ||
3480 | static void | |
3481 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3482 | struct cfq_io_cq *cic) |
383cd721 | 3483 | { |
f5f2b6ce | 3484 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3485 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3486 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3487 | cfqd->cfq_slice_idle); | |
3488 | } | |
7700fc4f SL |
3489 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3490 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3491 | #endif | |
22e2c507 | 3492 | } |
1da177e4 | 3493 | |
206dc69b | 3494 | static void |
b2c18e1e | 3495 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3496 | struct request *rq) |
206dc69b | 3497 | { |
3dde36dd | 3498 | sector_t sdist = 0; |
41647e7a | 3499 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3500 | if (cfqq->last_request_pos) { |
3501 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3502 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3503 | else | |
3504 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3505 | } | |
206dc69b | 3506 | |
3dde36dd | 3507 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3508 | if (blk_queue_nonrot(cfqd->queue)) |
3509 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3510 | else | |
3511 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3512 | } |
1da177e4 | 3513 | |
22e2c507 JA |
3514 | /* |
3515 | * Disable idle window if the process thinks too long or seeks so much that | |
3516 | * it doesn't matter | |
3517 | */ | |
3518 | static void | |
3519 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3520 | struct cfq_io_cq *cic) |
22e2c507 | 3521 | { |
7b679138 | 3522 | int old_idle, enable_idle; |
1be92f2f | 3523 | |
0871714e JA |
3524 | /* |
3525 | * Don't idle for async or idle io prio class | |
3526 | */ | |
3527 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3528 | return; |
3529 | ||
c265a7f4 | 3530 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3531 | |
76280aff CZ |
3532 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3533 | cfq_mark_cfqq_deep(cfqq); | |
3534 | ||
749ef9f8 CZ |
3535 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
3536 | enable_idle = 0; | |
f6e8d01b | 3537 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
3538 | !cfqd->cfq_slice_idle || |
3539 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3540 | enable_idle = 0; |
383cd721 SL |
3541 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3542 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3543 | enable_idle = 0; |
3544 | else | |
3545 | enable_idle = 1; | |
1da177e4 LT |
3546 | } |
3547 | ||
7b679138 JA |
3548 | if (old_idle != enable_idle) { |
3549 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3550 | if (enable_idle) | |
3551 | cfq_mark_cfqq_idle_window(cfqq); | |
3552 | else | |
3553 | cfq_clear_cfqq_idle_window(cfqq); | |
3554 | } | |
22e2c507 | 3555 | } |
1da177e4 | 3556 | |
22e2c507 JA |
3557 | /* |
3558 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3559 | * no or if we aren't sure, a 1 will cause a preempt. | |
3560 | */ | |
a6151c3a | 3561 | static bool |
22e2c507 | 3562 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3563 | struct request *rq) |
22e2c507 | 3564 | { |
6d048f53 | 3565 | struct cfq_queue *cfqq; |
22e2c507 | 3566 | |
6d048f53 JA |
3567 | cfqq = cfqd->active_queue; |
3568 | if (!cfqq) | |
a6151c3a | 3569 | return false; |
22e2c507 | 3570 | |
6d048f53 | 3571 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3572 | return false; |
22e2c507 JA |
3573 | |
3574 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3575 | return true; |
1e3335de | 3576 | |
875feb63 DS |
3577 | /* |
3578 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3579 | */ | |
3580 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3581 | return false; | |
3582 | ||
374f84ac JA |
3583 | /* |
3584 | * if the new request is sync, but the currently running queue is | |
3585 | * not, let the sync request have priority. | |
3586 | */ | |
5e705374 | 3587 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3588 | return true; |
1e3335de | 3589 | |
8682e1f1 VG |
3590 | if (new_cfqq->cfqg != cfqq->cfqg) |
3591 | return false; | |
3592 | ||
3593 | if (cfq_slice_used(cfqq)) | |
3594 | return true; | |
3595 | ||
3596 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
4d2ceea4 | 3597 | if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 VG |
3598 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
3599 | new_cfqq->service_tree->count == 2 && | |
3600 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3601 | return true; | |
3602 | ||
b53d1ed7 JA |
3603 | /* |
3604 | * So both queues are sync. Let the new request get disk time if | |
3605 | * it's a metadata request and the current queue is doing regular IO. | |
3606 | */ | |
65299a3b | 3607 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
3608 | return true; |
3609 | ||
3a9a3f6c DS |
3610 | /* |
3611 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3612 | */ | |
3613 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3614 | return true; |
3a9a3f6c | 3615 | |
d2d59e18 SL |
3616 | /* An idle queue should not be idle now for some reason */ |
3617 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
3618 | return true; | |
3619 | ||
1e3335de | 3620 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3621 | return false; |
1e3335de JA |
3622 | |
3623 | /* | |
3624 | * if this request is as-good as one we would expect from the | |
3625 | * current cfqq, let it preempt | |
3626 | */ | |
e9ce335d | 3627 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3628 | return true; |
1e3335de | 3629 | |
a6151c3a | 3630 | return false; |
22e2c507 JA |
3631 | } |
3632 | ||
3633 | /* | |
3634 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3635 | * let it have half of its nominal slice. | |
3636 | */ | |
3637 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3638 | { | |
df0793ab SL |
3639 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
3640 | ||
7b679138 | 3641 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 3642 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3643 | |
f8ae6e3e SL |
3644 | /* |
3645 | * workload type is changed, don't save slice, otherwise preempt | |
3646 | * doesn't happen | |
3647 | */ | |
df0793ab | 3648 | if (old_type != cfqq_type(cfqq)) |
4d2ceea4 | 3649 | cfqq->cfqg->saved_wl_slice = 0; |
f8ae6e3e | 3650 | |
bf572256 JA |
3651 | /* |
3652 | * Put the new queue at the front of the of the current list, | |
3653 | * so we know that it will be selected next. | |
3654 | */ | |
3655 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3656 | |
3657 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3658 | |
62a37f6b JT |
3659 | cfqq->slice_end = 0; |
3660 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3661 | } |
3662 | ||
22e2c507 | 3663 | /* |
5e705374 | 3664 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3665 | * something we should do about it |
3666 | */ | |
3667 | static void | |
5e705374 JA |
3668 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3669 | struct request *rq) | |
22e2c507 | 3670 | { |
c5869807 | 3671 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3672 | |
45333d5a | 3673 | cfqd->rq_queued++; |
65299a3b CH |
3674 | if (rq->cmd_flags & REQ_PRIO) |
3675 | cfqq->prio_pending++; | |
374f84ac | 3676 | |
383cd721 | 3677 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3678 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3679 | cfq_update_idle_window(cfqd, cfqq, cic); |
3680 | ||
b2c18e1e | 3681 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3682 | |
3683 | if (cfqq == cfqd->active_queue) { | |
3684 | /* | |
b029195d JA |
3685 | * Remember that we saw a request from this process, but |
3686 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3687 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3688 | * and merging. If the request is already larger than a single |
3689 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3690 | * merging is already done. Ditto for a busy system that |
3691 | * has other work pending, don't risk delaying until the | |
3692 | * idle timer unplug to continue working. | |
22e2c507 | 3693 | */ |
d6ceb25e | 3694 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3695 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3696 | cfqd->busy_queues > 1) { | |
812df48d | 3697 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3698 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3699 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3700 | } else { |
155fead9 | 3701 | cfqg_stats_update_idle_time(cfqq->cfqg); |
bf791937 | 3702 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3703 | } |
d6ceb25e | 3704 | } |
5e705374 | 3705 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3706 | /* |
3707 | * not the active queue - expire current slice if it is | |
3708 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3709 | * has some old slice time left and is of higher priority or |
3710 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3711 | */ |
3712 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3713 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3714 | } |
1da177e4 LT |
3715 | } |
3716 | ||
165125e1 | 3717 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3718 | { |
b4878f24 | 3719 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3720 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3721 | |
7b679138 | 3722 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 3723 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 3724 | |
30996f40 | 3725 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3726 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3727 | cfq_add_rq_rb(rq); |
155fead9 TH |
3728 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
3729 | rq->cmd_flags); | |
5e705374 | 3730 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3731 | } |
3732 | ||
45333d5a AC |
3733 | /* |
3734 | * Update hw_tag based on peak queue depth over 50 samples under | |
3735 | * sufficient load. | |
3736 | */ | |
3737 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3738 | { | |
1a1238a7 SL |
3739 | struct cfq_queue *cfqq = cfqd->active_queue; |
3740 | ||
53c583d2 CZ |
3741 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3742 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3743 | |
3744 | if (cfqd->hw_tag == 1) | |
3745 | return; | |
45333d5a AC |
3746 | |
3747 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3748 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3749 | return; |
3750 | ||
1a1238a7 SL |
3751 | /* |
3752 | * If active queue hasn't enough requests and can idle, cfq might not | |
3753 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3754 | * case | |
3755 | */ | |
3756 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3757 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3758 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3759 | return; |
3760 | ||
45333d5a AC |
3761 | if (cfqd->hw_tag_samples++ < 50) |
3762 | return; | |
3763 | ||
e459dd08 | 3764 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3765 | cfqd->hw_tag = 1; |
3766 | else | |
3767 | cfqd->hw_tag = 0; | |
45333d5a AC |
3768 | } |
3769 | ||
7667aa06 VG |
3770 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3771 | { | |
c5869807 | 3772 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 3773 | |
02a8f01b JT |
3774 | /* If the queue already has requests, don't wait */ |
3775 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3776 | return false; | |
3777 | ||
7667aa06 VG |
3778 | /* If there are other queues in the group, don't wait */ |
3779 | if (cfqq->cfqg->nr_cfqq > 1) | |
3780 | return false; | |
3781 | ||
7700fc4f SL |
3782 | /* the only queue in the group, but think time is big */ |
3783 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
3784 | return false; | |
3785 | ||
7667aa06 VG |
3786 | if (cfq_slice_used(cfqq)) |
3787 | return true; | |
3788 | ||
3789 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
3790 | if (cic && sample_valid(cic->ttime.ttime_samples) |
3791 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
3792 | return true; |
3793 | ||
3794 | /* | |
3795 | * If think times is less than a jiffy than ttime_mean=0 and above | |
3796 | * will not be true. It might happen that slice has not expired yet | |
3797 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
3798 | * case where think time is less than a jiffy, mark the queue wait | |
3799 | * busy if only 1 jiffy is left in the slice. | |
3800 | */ | |
3801 | if (cfqq->slice_end - jiffies == 1) | |
3802 | return true; | |
3803 | ||
3804 | return false; | |
3805 | } | |
3806 | ||
165125e1 | 3807 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3808 | { |
5e705374 | 3809 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3810 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3811 | const int sync = rq_is_sync(rq); |
b4878f24 | 3812 | unsigned long now; |
1da177e4 | 3813 | |
b4878f24 | 3814 | now = jiffies; |
33659ebb CH |
3815 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3816 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 3817 | |
45333d5a AC |
3818 | cfq_update_hw_tag(cfqd); |
3819 | ||
53c583d2 | 3820 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 3821 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 3822 | cfqd->rq_in_driver--; |
6d048f53 | 3823 | cfqq->dispatched--; |
80bdf0c7 | 3824 | (RQ_CFQG(rq))->dispatched--; |
155fead9 TH |
3825 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
3826 | rq_io_start_time_ns(rq), rq->cmd_flags); | |
1da177e4 | 3827 | |
53c583d2 | 3828 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 3829 | |
365722bb | 3830 | if (sync) { |
34b98d03 | 3831 | struct cfq_rb_root *st; |
f5f2b6ce | 3832 | |
383cd721 | 3833 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
3834 | |
3835 | if (cfq_cfqq_on_rr(cfqq)) | |
34b98d03 | 3836 | st = cfqq->service_tree; |
f5f2b6ce | 3837 | else |
34b98d03 VG |
3838 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), |
3839 | cfqq_type(cfqq)); | |
3840 | ||
3841 | st->ttime.last_end_request = now; | |
573412b2 CZ |
3842 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3843 | cfqd->last_delayed_sync = now; | |
365722bb | 3844 | } |
caaa5f9f | 3845 | |
7700fc4f SL |
3846 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3847 | cfqq->cfqg->ttime.last_end_request = now; | |
3848 | #endif | |
3849 | ||
caaa5f9f JA |
3850 | /* |
3851 | * If this is the active queue, check if it needs to be expired, | |
3852 | * or if we want to idle in case it has no pending requests. | |
3853 | */ | |
3854 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3855 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3856 | ||
44f7c160 JA |
3857 | if (cfq_cfqq_slice_new(cfqq)) { |
3858 | cfq_set_prio_slice(cfqd, cfqq); | |
3859 | cfq_clear_cfqq_slice_new(cfqq); | |
3860 | } | |
f75edf2d VG |
3861 | |
3862 | /* | |
7667aa06 VG |
3863 | * Should we wait for next request to come in before we expire |
3864 | * the queue. | |
f75edf2d | 3865 | */ |
7667aa06 | 3866 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
3867 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3868 | if (!cfqd->cfq_slice_idle) | |
3869 | extend_sl = cfqd->cfq_group_idle; | |
3870 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 3871 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 3872 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
3873 | } |
3874 | ||
a36e71f9 | 3875 | /* |
8e550632 CZ |
3876 | * Idling is not enabled on: |
3877 | * - expired queues | |
3878 | * - idle-priority queues | |
3879 | * - async queues | |
3880 | * - queues with still some requests queued | |
3881 | * - when there is a close cooperator | |
a36e71f9 | 3882 | */ |
0871714e | 3883 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 3884 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3885 | else if (sync && cfqq_empty && |
3886 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 3887 | cfq_arm_slice_timer(cfqd); |
8e550632 | 3888 | } |
caaa5f9f | 3889 | } |
6d048f53 | 3890 | |
53c583d2 | 3891 | if (!cfqd->rq_in_driver) |
23e018a1 | 3892 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3893 | } |
3894 | ||
89850f7e | 3895 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3896 | { |
1b379d8d | 3897 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3898 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3899 | return ELV_MQUEUE_MUST; |
3b18152c | 3900 | } |
1da177e4 | 3901 | |
22e2c507 | 3902 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3903 | } |
3904 | ||
165125e1 | 3905 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3906 | { |
3907 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3908 | struct task_struct *tsk = current; | |
c5869807 | 3909 | struct cfq_io_cq *cic; |
22e2c507 JA |
3910 | struct cfq_queue *cfqq; |
3911 | ||
3912 | /* | |
3913 | * don't force setup of a queue from here, as a call to may_queue | |
3914 | * does not necessarily imply that a request actually will be queued. | |
3915 | * so just lookup a possibly existing queue, or return 'may queue' | |
3916 | * if that fails | |
3917 | */ | |
4ac845a2 | 3918 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3919 | if (!cic) |
3920 | return ELV_MQUEUE_MAY; | |
3921 | ||
b0b78f81 | 3922 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3923 | if (cfqq) { |
abede6da | 3924 | cfq_init_prio_data(cfqq, cic); |
22e2c507 | 3925 | |
89850f7e | 3926 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3927 | } |
3928 | ||
3929 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3930 | } |
3931 | ||
1da177e4 LT |
3932 | /* |
3933 | * queue lock held here | |
3934 | */ | |
bb37b94c | 3935 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3936 | { |
5e705374 | 3937 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3938 | |
5e705374 | 3939 | if (cfqq) { |
22e2c507 | 3940 | const int rw = rq_data_dir(rq); |
1da177e4 | 3941 | |
22e2c507 JA |
3942 | BUG_ON(!cfqq->allocated[rw]); |
3943 | cfqq->allocated[rw]--; | |
1da177e4 | 3944 | |
7f1dc8a2 | 3945 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 3946 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
3947 | rq->elv.priv[0] = NULL; |
3948 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 3949 | |
1da177e4 LT |
3950 | cfq_put_queue(cfqq); |
3951 | } | |
3952 | } | |
3953 | ||
df5fe3e8 | 3954 | static struct cfq_queue * |
c5869807 | 3955 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
3956 | struct cfq_queue *cfqq) |
3957 | { | |
3958 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3959 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3960 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3961 | cfq_put_queue(cfqq); |
3962 | return cic_to_cfqq(cic, 1); | |
3963 | } | |
3964 | ||
e6c5bc73 JM |
3965 | /* |
3966 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3967 | * was the last process referring to said cfqq. | |
3968 | */ | |
3969 | static struct cfq_queue * | |
c5869807 | 3970 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
3971 | { |
3972 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
3973 | cfqq->pid = current->pid; |
3974 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 3975 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
3976 | return cfqq; |
3977 | } | |
3978 | ||
3979 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
3980 | |
3981 | cfq_put_cooperator(cfqq); | |
3982 | ||
e6c5bc73 JM |
3983 | cfq_put_queue(cfqq); |
3984 | return NULL; | |
3985 | } | |
1da177e4 | 3986 | /* |
22e2c507 | 3987 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3988 | */ |
22e2c507 | 3989 | static int |
852c788f TH |
3990 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
3991 | gfp_t gfp_mask) | |
1da177e4 LT |
3992 | { |
3993 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 3994 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 3995 | const int rw = rq_data_dir(rq); |
a6151c3a | 3996 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3997 | struct cfq_queue *cfqq; |
1da177e4 LT |
3998 | |
3999 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
4000 | ||
216284c3 | 4001 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 4002 | |
598971bf TH |
4003 | check_ioprio_changed(cic, bio); |
4004 | check_blkcg_changed(cic, bio); | |
e6c5bc73 | 4005 | new_queue: |
91fac317 | 4006 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 4007 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
abede6da | 4008 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio, gfp_mask); |
91fac317 | 4009 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 4010 | } else { |
e6c5bc73 JM |
4011 | /* |
4012 | * If the queue was seeky for too long, break it apart. | |
4013 | */ | |
ae54abed | 4014 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
4015 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
4016 | cfqq = split_cfqq(cic, cfqq); | |
4017 | if (!cfqq) | |
4018 | goto new_queue; | |
4019 | } | |
4020 | ||
df5fe3e8 JM |
4021 | /* |
4022 | * Check to see if this queue is scheduled to merge with | |
4023 | * another, closely cooperating queue. The merging of | |
4024 | * queues happens here as it must be done in process context. | |
4025 | * The reference on new_cfqq was taken in merge_cfqqs. | |
4026 | */ | |
4027 | if (cfqq->new_cfqq) | |
4028 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 4029 | } |
1da177e4 LT |
4030 | |
4031 | cfqq->allocated[rw]++; | |
1da177e4 | 4032 | |
6fae9c25 | 4033 | cfqq->ref++; |
eb7d8c07 | 4034 | cfqg_get(cfqq->cfqg); |
a612fddf | 4035 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 4036 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 4037 | spin_unlock_irq(q->queue_lock); |
5e705374 | 4038 | return 0; |
1da177e4 LT |
4039 | } |
4040 | ||
65f27f38 | 4041 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 4042 | { |
65f27f38 | 4043 | struct cfq_data *cfqd = |
23e018a1 | 4044 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 4045 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4046 | |
40bb54d1 | 4047 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 4048 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 4049 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
4050 | } |
4051 | ||
4052 | /* | |
4053 | * Timer running if the active_queue is currently idling inside its time slice | |
4054 | */ | |
4055 | static void cfq_idle_slice_timer(unsigned long data) | |
4056 | { | |
4057 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
4058 | struct cfq_queue *cfqq; | |
4059 | unsigned long flags; | |
3c6bd2f8 | 4060 | int timed_out = 1; |
22e2c507 | 4061 | |
7b679138 JA |
4062 | cfq_log(cfqd, "idle timer fired"); |
4063 | ||
22e2c507 JA |
4064 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
4065 | ||
fe094d98 JA |
4066 | cfqq = cfqd->active_queue; |
4067 | if (cfqq) { | |
3c6bd2f8 JA |
4068 | timed_out = 0; |
4069 | ||
b029195d JA |
4070 | /* |
4071 | * We saw a request before the queue expired, let it through | |
4072 | */ | |
4073 | if (cfq_cfqq_must_dispatch(cfqq)) | |
4074 | goto out_kick; | |
4075 | ||
22e2c507 JA |
4076 | /* |
4077 | * expired | |
4078 | */ | |
44f7c160 | 4079 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
4080 | goto expire; |
4081 | ||
4082 | /* | |
4083 | * only expire and reinvoke request handler, if there are | |
4084 | * other queues with pending requests | |
4085 | */ | |
caaa5f9f | 4086 | if (!cfqd->busy_queues) |
22e2c507 | 4087 | goto out_cont; |
22e2c507 JA |
4088 | |
4089 | /* | |
4090 | * not expired and it has a request pending, let it dispatch | |
4091 | */ | |
75e50984 | 4092 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 4093 | goto out_kick; |
76280aff CZ |
4094 | |
4095 | /* | |
4096 | * Queue depth flag is reset only when the idle didn't succeed | |
4097 | */ | |
4098 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
4099 | } |
4100 | expire: | |
e5ff082e | 4101 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 4102 | out_kick: |
23e018a1 | 4103 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
4104 | out_cont: |
4105 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
4106 | } | |
4107 | ||
3b18152c JA |
4108 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
4109 | { | |
4110 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 4111 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 4112 | } |
22e2c507 | 4113 | |
c2dea2d1 VT |
4114 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
4115 | { | |
4116 | int i; | |
4117 | ||
4118 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
4119 | if (cfqd->async_cfqq[0][i]) | |
4120 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
4121 | if (cfqd->async_cfqq[1][i]) | |
4122 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 4123 | } |
2389d1ef ON |
4124 | |
4125 | if (cfqd->async_idle_cfqq) | |
4126 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
4127 | } |
4128 | ||
b374d18a | 4129 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 4130 | { |
22e2c507 | 4131 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 4132 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4133 | |
3b18152c | 4134 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 4135 | |
d9ff4187 | 4136 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 4137 | |
d9ff4187 | 4138 | if (cfqd->active_queue) |
e5ff082e | 4139 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 4140 | |
c2dea2d1 | 4141 | cfq_put_async_queues(cfqd); |
03aa264a TH |
4142 | |
4143 | spin_unlock_irq(q->queue_lock); | |
4144 | ||
a90d742e AV |
4145 | cfq_shutdown_timer_wq(cfqd); |
4146 | ||
ffea73fc TH |
4147 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4148 | blkcg_deactivate_policy(q, &blkcg_policy_cfq); | |
4149 | #else | |
f51b802c | 4150 | kfree(cfqd->root_group); |
2abae55f | 4151 | #endif |
56edf7d7 | 4152 | kfree(cfqd); |
1da177e4 LT |
4153 | } |
4154 | ||
b2fab5ac | 4155 | static int cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
4156 | { |
4157 | struct cfq_data *cfqd; | |
3c798398 | 4158 | struct blkcg_gq *blkg __maybe_unused; |
a2b1693b | 4159 | int i, ret; |
1da177e4 | 4160 | |
94f6030c | 4161 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
a73f730d | 4162 | if (!cfqd) |
b2fab5ac | 4163 | return -ENOMEM; |
80b15c73 | 4164 | |
f51b802c TH |
4165 | cfqd->queue = q; |
4166 | q->elevator->elevator_data = cfqd; | |
4167 | ||
1fa8f6d6 VG |
4168 | /* Init root service tree */ |
4169 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
4170 | ||
f51b802c | 4171 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 4172 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4173 | ret = blkcg_activate_policy(q, &blkcg_policy_cfq); |
a2b1693b TH |
4174 | if (ret) |
4175 | goto out_free; | |
f51b802c | 4176 | |
a2b1693b | 4177 | cfqd->root_group = blkg_to_cfqg(q->root_blkg); |
f51b802c | 4178 | #else |
a2b1693b | 4179 | ret = -ENOMEM; |
f51b802c TH |
4180 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
4181 | GFP_KERNEL, cfqd->queue->node); | |
a2b1693b TH |
4182 | if (!cfqd->root_group) |
4183 | goto out_free; | |
5624a4e4 | 4184 | |
a2b1693b TH |
4185 | cfq_init_cfqg_base(cfqd->root_group); |
4186 | #endif | |
3381cb8d | 4187 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_DEFAULT; |
e71357e1 | 4188 | cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_DEFAULT; |
5624a4e4 | 4189 | |
26a2ac00 JA |
4190 | /* |
4191 | * Not strictly needed (since RB_ROOT just clears the node and we | |
4192 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
4193 | * to add magic to the rb code | |
4194 | */ | |
4195 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
4196 | cfqd->prio_trees[i] = RB_ROOT; | |
4197 | ||
6118b70b JA |
4198 | /* |
4199 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
4200 | * Grab a permanent reference to it, so that the normal code flow | |
f51b802c TH |
4201 | * will not attempt to free it. oom_cfqq is linked to root_group |
4202 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
4203 | * the reference from linking right away. | |
6118b70b JA |
4204 | */ |
4205 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 4206 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
4207 | |
4208 | spin_lock_irq(q->queue_lock); | |
f51b802c | 4209 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 4210 | cfqg_put(cfqd->root_group); |
1adaf3dd | 4211 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 4212 | |
22e2c507 JA |
4213 | init_timer(&cfqd->idle_slice_timer); |
4214 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
4215 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
4216 | ||
23e018a1 | 4217 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 4218 | |
1da177e4 | 4219 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
4220 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
4221 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
4222 | cfqd->cfq_back_max = cfq_back_max; |
4223 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
4224 | cfqd->cfq_slice[0] = cfq_slice_async; |
4225 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
5bf14c07 | 4226 | cfqd->cfq_target_latency = cfq_target_latency; |
22e2c507 JA |
4227 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
4228 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
80bdf0c7 | 4229 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 4230 | cfqd->cfq_latency = 1; |
e459dd08 | 4231 | cfqd->hw_tag = -1; |
edc71131 CZ |
4232 | /* |
4233 | * we optimistically start assuming sync ops weren't delayed in last | |
4234 | * second, in order to have larger depth for async operations. | |
4235 | */ | |
573412b2 | 4236 | cfqd->last_delayed_sync = jiffies - HZ; |
b2fab5ac | 4237 | return 0; |
a2b1693b TH |
4238 | |
4239 | out_free: | |
4240 | kfree(cfqd); | |
4241 | return ret; | |
1da177e4 LT |
4242 | } |
4243 | ||
1da177e4 LT |
4244 | /* |
4245 | * sysfs parts below --> | |
4246 | */ | |
1da177e4 LT |
4247 | static ssize_t |
4248 | cfq_var_show(unsigned int var, char *page) | |
4249 | { | |
4250 | return sprintf(page, "%d\n", var); | |
4251 | } | |
4252 | ||
4253 | static ssize_t | |
4254 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
4255 | { | |
4256 | char *p = (char *) page; | |
4257 | ||
4258 | *var = simple_strtoul(p, &p, 10); | |
4259 | return count; | |
4260 | } | |
4261 | ||
1da177e4 | 4262 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 4263 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 4264 | { \ |
3d1ab40f | 4265 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4266 | unsigned int __data = __VAR; \ |
4267 | if (__CONV) \ | |
4268 | __data = jiffies_to_msecs(__data); \ | |
4269 | return cfq_var_show(__data, (page)); \ | |
4270 | } | |
4271 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
4272 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
4273 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
4274 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
4275 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 4276 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 4277 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
4278 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4279 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4280 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4281 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
5bf14c07 | 4282 | SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1); |
1da177e4 LT |
4283 | #undef SHOW_FUNCTION |
4284 | ||
4285 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 4286 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4287 | { \ |
3d1ab40f | 4288 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4289 | unsigned int __data; \ |
4290 | int ret = cfq_var_store(&__data, (page), count); \ | |
4291 | if (__data < (MIN)) \ | |
4292 | __data = (MIN); \ | |
4293 | else if (__data > (MAX)) \ | |
4294 | __data = (MAX); \ | |
4295 | if (__CONV) \ | |
4296 | *(__PTR) = msecs_to_jiffies(__data); \ | |
4297 | else \ | |
4298 | *(__PTR) = __data; \ | |
4299 | return ret; \ | |
4300 | } | |
4301 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4302 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4303 | UINT_MAX, 1); | |
4304 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4305 | UINT_MAX, 1); | |
e572ec7e | 4306 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4307 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4308 | UINT_MAX, 0); | |
22e2c507 | 4309 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4310 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4311 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4312 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4313 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4314 | UINT_MAX, 0); | |
963b72fc | 4315 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
5bf14c07 | 4316 | STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1); |
1da177e4 LT |
4317 | #undef STORE_FUNCTION |
4318 | ||
e572ec7e AV |
4319 | #define CFQ_ATTR(name) \ |
4320 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4321 | ||
4322 | static struct elv_fs_entry cfq_attrs[] = { | |
4323 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4324 | CFQ_ATTR(fifo_expire_sync), |
4325 | CFQ_ATTR(fifo_expire_async), | |
4326 | CFQ_ATTR(back_seek_max), | |
4327 | CFQ_ATTR(back_seek_penalty), | |
4328 | CFQ_ATTR(slice_sync), | |
4329 | CFQ_ATTR(slice_async), | |
4330 | CFQ_ATTR(slice_async_rq), | |
4331 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 4332 | CFQ_ATTR(group_idle), |
963b72fc | 4333 | CFQ_ATTR(low_latency), |
5bf14c07 | 4334 | CFQ_ATTR(target_latency), |
e572ec7e | 4335 | __ATTR_NULL |
1da177e4 LT |
4336 | }; |
4337 | ||
1da177e4 LT |
4338 | static struct elevator_type iosched_cfq = { |
4339 | .ops = { | |
4340 | .elevator_merge_fn = cfq_merge, | |
4341 | .elevator_merged_fn = cfq_merged_request, | |
4342 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 4343 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 4344 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4345 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4346 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4347 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4348 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4349 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4350 | .elevator_former_req_fn = elv_rb_former_request, |
4351 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 4352 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 4353 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
4354 | .elevator_set_req_fn = cfq_set_request, |
4355 | .elevator_put_req_fn = cfq_put_request, | |
4356 | .elevator_may_queue_fn = cfq_may_queue, | |
4357 | .elevator_init_fn = cfq_init_queue, | |
4358 | .elevator_exit_fn = cfq_exit_queue, | |
4359 | }, | |
3d3c2379 TH |
4360 | .icq_size = sizeof(struct cfq_io_cq), |
4361 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 4362 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 4363 | .elevator_name = "cfq", |
1da177e4 LT |
4364 | .elevator_owner = THIS_MODULE, |
4365 | }; | |
4366 | ||
3e252066 | 4367 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4368 | static struct blkcg_policy blkcg_policy_cfq = { |
f9fcc2d3 TH |
4369 | .pd_size = sizeof(struct cfq_group), |
4370 | .cftypes = cfq_blkcg_files, | |
4371 | ||
4372 | .pd_init_fn = cfq_pd_init, | |
4373 | .pd_reset_stats_fn = cfq_pd_reset_stats, | |
3e252066 | 4374 | }; |
3e252066 VG |
4375 | #endif |
4376 | ||
1da177e4 LT |
4377 | static int __init cfq_init(void) |
4378 | { | |
3d3c2379 TH |
4379 | int ret; |
4380 | ||
22e2c507 JA |
4381 | /* |
4382 | * could be 0 on HZ < 1000 setups | |
4383 | */ | |
4384 | if (!cfq_slice_async) | |
4385 | cfq_slice_async = 1; | |
4386 | if (!cfq_slice_idle) | |
4387 | cfq_slice_idle = 1; | |
4388 | ||
80bdf0c7 VG |
4389 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4390 | if (!cfq_group_idle) | |
4391 | cfq_group_idle = 1; | |
8bd435b3 | 4392 | |
3c798398 | 4393 | ret = blkcg_policy_register(&blkcg_policy_cfq); |
8bd435b3 TH |
4394 | if (ret) |
4395 | return ret; | |
ffea73fc TH |
4396 | #else |
4397 | cfq_group_idle = 0; | |
4398 | #endif | |
8bd435b3 | 4399 | |
fd794956 | 4400 | ret = -ENOMEM; |
3d3c2379 TH |
4401 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4402 | if (!cfq_pool) | |
8bd435b3 | 4403 | goto err_pol_unreg; |
1da177e4 | 4404 | |
3d3c2379 | 4405 | ret = elv_register(&iosched_cfq); |
8bd435b3 TH |
4406 | if (ret) |
4407 | goto err_free_pool; | |
3d3c2379 | 4408 | |
2fdd82bd | 4409 | return 0; |
8bd435b3 TH |
4410 | |
4411 | err_free_pool: | |
4412 | kmem_cache_destroy(cfq_pool); | |
4413 | err_pol_unreg: | |
ffea73fc | 4414 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4415 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4416 | #endif |
8bd435b3 | 4417 | return ret; |
1da177e4 LT |
4418 | } |
4419 | ||
4420 | static void __exit cfq_exit(void) | |
4421 | { | |
ffea73fc | 4422 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4423 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4424 | #endif |
1da177e4 | 4425 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4426 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4427 | } |
4428 | ||
4429 | module_init(cfq_init); | |
4430 | module_exit(cfq_exit); | |
4431 | ||
4432 | MODULE_AUTHOR("Jens Axboe"); | |
4433 | MODULE_LICENSE("GPL"); | |
4434 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |