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 | 19 | |
0381411e TH |
20 | static struct blkio_policy_type blkio_policy_cfq; |
21 | ||
1da177e4 LT |
22 | /* |
23 | * tunables | |
24 | */ | |
fe094d98 | 25 | /* max queue in one round of service */ |
abc3c744 | 26 | static const int cfq_quantum = 8; |
64100099 | 27 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
28 | /* maximum backwards seek, in KiB */ |
29 | static const int cfq_back_max = 16 * 1024; | |
30 | /* penalty of a backwards seek */ | |
31 | static const int cfq_back_penalty = 2; | |
64100099 | 32 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 33 | static int cfq_slice_async = HZ / 25; |
64100099 | 34 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 35 | static int cfq_slice_idle = HZ / 125; |
80bdf0c7 | 36 | static int cfq_group_idle = HZ / 125; |
5db5d642 CZ |
37 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
38 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 39 | |
d9e7620e | 40 | /* |
0871714e | 41 | * offset from end of service tree |
d9e7620e | 42 | */ |
0871714e | 43 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
44 | |
45 | /* | |
46 | * below this threshold, we consider thinktime immediate | |
47 | */ | |
48 | #define CFQ_MIN_TT (2) | |
49 | ||
22e2c507 | 50 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 51 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 52 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 53 | |
3dde36dd | 54 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 55 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 56 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 57 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 58 | |
a612fddf TH |
59 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
60 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
61 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 62 | |
e18b890b | 63 | static struct kmem_cache *cfq_pool; |
1da177e4 | 64 | |
22e2c507 JA |
65 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
66 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
67 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
68 | ||
206dc69b | 69 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 70 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 71 | |
c5869807 TH |
72 | struct cfq_ttime { |
73 | unsigned long last_end_request; | |
74 | ||
75 | unsigned long ttime_total; | |
76 | unsigned long ttime_samples; | |
77 | unsigned long ttime_mean; | |
78 | }; | |
79 | ||
cc09e299 JA |
80 | /* |
81 | * Most of our rbtree usage is for sorting with min extraction, so | |
82 | * if we cache the leftmost node we don't have to walk down the tree | |
83 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
84 | * move this into the elevator for the rq sorting as well. | |
85 | */ | |
86 | struct cfq_rb_root { | |
87 | struct rb_root rb; | |
88 | struct rb_node *left; | |
aa6f6a3d | 89 | unsigned count; |
73e9ffdd | 90 | unsigned total_weight; |
1fa8f6d6 | 91 | u64 min_vdisktime; |
f5f2b6ce | 92 | struct cfq_ttime ttime; |
cc09e299 | 93 | }; |
f5f2b6ce SL |
94 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
95 | .ttime = {.last_end_request = jiffies,},} | |
cc09e299 | 96 | |
6118b70b JA |
97 | /* |
98 | * Per process-grouping structure | |
99 | */ | |
100 | struct cfq_queue { | |
101 | /* reference count */ | |
30d7b944 | 102 | int ref; |
6118b70b JA |
103 | /* various state flags, see below */ |
104 | unsigned int flags; | |
105 | /* parent cfq_data */ | |
106 | struct cfq_data *cfqd; | |
107 | /* service_tree member */ | |
108 | struct rb_node rb_node; | |
109 | /* service_tree key */ | |
110 | unsigned long rb_key; | |
111 | /* prio tree member */ | |
112 | struct rb_node p_node; | |
113 | /* prio tree root we belong to, if any */ | |
114 | struct rb_root *p_root; | |
115 | /* sorted list of pending requests */ | |
116 | struct rb_root sort_list; | |
117 | /* if fifo isn't expired, next request to serve */ | |
118 | struct request *next_rq; | |
119 | /* requests queued in sort_list */ | |
120 | int queued[2]; | |
121 | /* currently allocated requests */ | |
122 | int allocated[2]; | |
123 | /* fifo list of requests in sort_list */ | |
124 | struct list_head fifo; | |
125 | ||
dae739eb VG |
126 | /* time when queue got scheduled in to dispatch first request. */ |
127 | unsigned long dispatch_start; | |
f75edf2d | 128 | unsigned int allocated_slice; |
c4081ba5 | 129 | unsigned int slice_dispatch; |
dae739eb VG |
130 | /* time when first request from queue completed and slice started. */ |
131 | unsigned long slice_start; | |
6118b70b JA |
132 | unsigned long slice_end; |
133 | long slice_resid; | |
6118b70b | 134 | |
65299a3b CH |
135 | /* pending priority requests */ |
136 | int prio_pending; | |
6118b70b JA |
137 | /* number of requests that are on the dispatch list or inside driver */ |
138 | int dispatched; | |
139 | ||
140 | /* io prio of this group */ | |
141 | unsigned short ioprio, org_ioprio; | |
4aede84b | 142 | unsigned short ioprio_class; |
6118b70b | 143 | |
c4081ba5 RK |
144 | pid_t pid; |
145 | ||
3dde36dd | 146 | u32 seek_history; |
b2c18e1e JM |
147 | sector_t last_request_pos; |
148 | ||
aa6f6a3d | 149 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 150 | struct cfq_queue *new_cfqq; |
cdb16e8f | 151 | struct cfq_group *cfqg; |
c4e7893e VG |
152 | /* Number of sectors dispatched from queue in single dispatch round */ |
153 | unsigned long nr_sectors; | |
6118b70b JA |
154 | }; |
155 | ||
c0324a02 | 156 | /* |
718eee05 | 157 | * First index in the service_trees. |
c0324a02 CZ |
158 | * IDLE is handled separately, so it has negative index |
159 | */ | |
160 | enum wl_prio_t { | |
c0324a02 | 161 | BE_WORKLOAD = 0, |
615f0259 VG |
162 | RT_WORKLOAD = 1, |
163 | IDLE_WORKLOAD = 2, | |
b4627321 | 164 | CFQ_PRIO_NR, |
c0324a02 CZ |
165 | }; |
166 | ||
718eee05 CZ |
167 | /* |
168 | * Second index in the service_trees. | |
169 | */ | |
170 | enum wl_type_t { | |
171 | ASYNC_WORKLOAD = 0, | |
172 | SYNC_NOIDLE_WORKLOAD = 1, | |
173 | SYNC_WORKLOAD = 2 | |
174 | }; | |
175 | ||
155fead9 TH |
176 | struct cfqg_stats { |
177 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
178 | /* total bytes transferred */ | |
179 | struct blkg_rwstat service_bytes; | |
180 | /* total IOs serviced, post merge */ | |
181 | struct blkg_rwstat serviced; | |
182 | /* number of ios merged */ | |
183 | struct blkg_rwstat merged; | |
184 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
185 | struct blkg_rwstat service_time; | |
186 | /* total time spent waiting in scheduler queue in ns */ | |
187 | struct blkg_rwstat wait_time; | |
188 | /* number of IOs queued up */ | |
189 | struct blkg_rwstat queued; | |
190 | /* total sectors transferred */ | |
191 | struct blkg_stat sectors; | |
192 | /* total disk time and nr sectors dispatched by this group */ | |
193 | struct blkg_stat time; | |
194 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
195 | /* time not charged to this cgroup */ | |
196 | struct blkg_stat unaccounted_time; | |
197 | /* sum of number of ios queued across all samples */ | |
198 | struct blkg_stat avg_queue_size_sum; | |
199 | /* count of samples taken for average */ | |
200 | struct blkg_stat avg_queue_size_samples; | |
201 | /* how many times this group has been removed from service tree */ | |
202 | struct blkg_stat dequeue; | |
203 | /* total time spent waiting for it to be assigned a timeslice. */ | |
204 | struct blkg_stat group_wait_time; | |
205 | /* time spent idling for this blkio_group */ | |
206 | struct blkg_stat idle_time; | |
207 | /* total time with empty current active q with other requests queued */ | |
208 | struct blkg_stat empty_time; | |
209 | /* fields after this shouldn't be cleared on stat reset */ | |
210 | uint64_t start_group_wait_time; | |
211 | uint64_t start_idle_time; | |
212 | uint64_t start_empty_time; | |
213 | uint16_t flags; | |
214 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
215 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
216 | }; | |
217 | ||
cdb16e8f VG |
218 | /* This is per cgroup per device grouping structure */ |
219 | struct cfq_group { | |
1fa8f6d6 VG |
220 | /* group service_tree member */ |
221 | struct rb_node rb_node; | |
222 | ||
223 | /* group service_tree key */ | |
224 | u64 vdisktime; | |
25bc6b07 | 225 | unsigned int weight; |
8184f93e | 226 | unsigned int new_weight; |
3381cb8d | 227 | unsigned int dev_weight; |
1fa8f6d6 VG |
228 | |
229 | /* number of cfqq currently on this group */ | |
230 | int nr_cfqq; | |
231 | ||
cdb16e8f | 232 | /* |
4495a7d4 | 233 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
234 | * create the array for each prio class but at run time it is used |
235 | * only for RT and BE class and slot for IDLE class remains unused. | |
236 | * This is primarily done to avoid confusion and a gcc warning. | |
237 | */ | |
238 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
239 | /* | |
240 | * rr lists of queues with requests. We maintain service trees for | |
241 | * RT and BE classes. These trees are subdivided in subclasses | |
242 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
243 | * class there is no subclassification and all the cfq queues go on | |
244 | * a single tree service_tree_idle. | |
cdb16e8f VG |
245 | * Counts are embedded in the cfq_rb_root |
246 | */ | |
247 | struct cfq_rb_root service_trees[2][3]; | |
248 | struct cfq_rb_root service_tree_idle; | |
dae739eb VG |
249 | |
250 | unsigned long saved_workload_slice; | |
251 | enum wl_type_t saved_workload; | |
252 | enum wl_prio_t saved_serving_prio; | |
4eef3049 | 253 | |
80bdf0c7 VG |
254 | /* number of requests that are on the dispatch list or inside driver */ |
255 | int dispatched; | |
7700fc4f | 256 | struct cfq_ttime ttime; |
155fead9 | 257 | struct cfqg_stats stats; |
cdb16e8f | 258 | }; |
718eee05 | 259 | |
c5869807 TH |
260 | struct cfq_io_cq { |
261 | struct io_cq icq; /* must be the first member */ | |
262 | struct cfq_queue *cfqq[2]; | |
263 | struct cfq_ttime ttime; | |
598971bf TH |
264 | int ioprio; /* the current ioprio */ |
265 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
266 | uint64_t blkcg_id; /* the current blkcg ID */ | |
267 | #endif | |
c5869807 TH |
268 | }; |
269 | ||
22e2c507 JA |
270 | /* |
271 | * Per block device queue structure | |
272 | */ | |
1da177e4 | 273 | struct cfq_data { |
165125e1 | 274 | struct request_queue *queue; |
1fa8f6d6 VG |
275 | /* Root service tree for cfq_groups */ |
276 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 277 | struct cfq_group *root_group; |
22e2c507 | 278 | |
c0324a02 CZ |
279 | /* |
280 | * The priority currently being served | |
22e2c507 | 281 | */ |
c0324a02 | 282 | enum wl_prio_t serving_prio; |
718eee05 CZ |
283 | enum wl_type_t serving_type; |
284 | unsigned long workload_expires; | |
cdb16e8f | 285 | struct cfq_group *serving_group; |
a36e71f9 JA |
286 | |
287 | /* | |
288 | * Each priority tree is sorted by next_request position. These | |
289 | * trees are used when determining if two or more queues are | |
290 | * interleaving requests (see cfq_close_cooperator). | |
291 | */ | |
292 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
293 | ||
22e2c507 | 294 | unsigned int busy_queues; |
ef8a41df | 295 | unsigned int busy_sync_queues; |
22e2c507 | 296 | |
53c583d2 CZ |
297 | int rq_in_driver; |
298 | int rq_in_flight[2]; | |
45333d5a AC |
299 | |
300 | /* | |
301 | * queue-depth detection | |
302 | */ | |
303 | int rq_queued; | |
25776e35 | 304 | int hw_tag; |
e459dd08 CZ |
305 | /* |
306 | * hw_tag can be | |
307 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
308 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
309 | * 0 => no NCQ | |
310 | */ | |
311 | int hw_tag_est_depth; | |
312 | unsigned int hw_tag_samples; | |
1da177e4 | 313 | |
22e2c507 JA |
314 | /* |
315 | * idle window management | |
316 | */ | |
317 | struct timer_list idle_slice_timer; | |
23e018a1 | 318 | struct work_struct unplug_work; |
1da177e4 | 319 | |
22e2c507 | 320 | struct cfq_queue *active_queue; |
c5869807 | 321 | struct cfq_io_cq *active_cic; |
22e2c507 | 322 | |
c2dea2d1 VT |
323 | /* |
324 | * async queue for each priority case | |
325 | */ | |
326 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
327 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 328 | |
6d048f53 | 329 | sector_t last_position; |
1da177e4 | 330 | |
1da177e4 LT |
331 | /* |
332 | * tunables, see top of file | |
333 | */ | |
334 | unsigned int cfq_quantum; | |
22e2c507 | 335 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
336 | unsigned int cfq_back_penalty; |
337 | unsigned int cfq_back_max; | |
22e2c507 JA |
338 | unsigned int cfq_slice[2]; |
339 | unsigned int cfq_slice_async_rq; | |
340 | unsigned int cfq_slice_idle; | |
80bdf0c7 | 341 | unsigned int cfq_group_idle; |
963b72fc | 342 | unsigned int cfq_latency; |
d9ff4187 | 343 | |
6118b70b JA |
344 | /* |
345 | * Fallback dummy cfqq for extreme OOM conditions | |
346 | */ | |
347 | struct cfq_queue oom_cfqq; | |
365722bb | 348 | |
573412b2 | 349 | unsigned long last_delayed_sync; |
1da177e4 LT |
350 | }; |
351 | ||
25fb5169 VG |
352 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
353 | ||
cdb16e8f VG |
354 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
355 | enum wl_prio_t prio, | |
65b32a57 | 356 | enum wl_type_t type) |
c0324a02 | 357 | { |
1fa8f6d6 VG |
358 | if (!cfqg) |
359 | return NULL; | |
360 | ||
c0324a02 | 361 | if (prio == IDLE_WORKLOAD) |
cdb16e8f | 362 | return &cfqg->service_tree_idle; |
c0324a02 | 363 | |
cdb16e8f | 364 | return &cfqg->service_trees[prio][type]; |
c0324a02 CZ |
365 | } |
366 | ||
3b18152c | 367 | enum cfqq_state_flags { |
b0b8d749 JA |
368 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
369 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 370 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 371 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
372 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
373 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
374 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 375 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 376 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 377 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 378 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 379 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 380 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
381 | }; |
382 | ||
383 | #define CFQ_CFQQ_FNS(name) \ | |
384 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
385 | { \ | |
fe094d98 | 386 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
387 | } \ |
388 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
389 | { \ | |
fe094d98 | 390 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
391 | } \ |
392 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
393 | { \ | |
fe094d98 | 394 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
395 | } |
396 | ||
397 | CFQ_CFQQ_FNS(on_rr); | |
398 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 399 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 400 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
401 | CFQ_CFQQ_FNS(fifo_expire); |
402 | CFQ_CFQQ_FNS(idle_window); | |
403 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 404 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 405 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 406 | CFQ_CFQQ_FNS(coop); |
ae54abed | 407 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 408 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 409 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
410 | #undef CFQ_CFQQ_FNS |
411 | ||
629ed0b1 | 412 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
2ce4d50f | 413 | |
155fead9 TH |
414 | /* cfqg stats flags */ |
415 | enum cfqg_stats_flags { | |
416 | CFQG_stats_waiting = 0, | |
417 | CFQG_stats_idling, | |
418 | CFQG_stats_empty, | |
629ed0b1 TH |
419 | }; |
420 | ||
155fead9 TH |
421 | #define CFQG_FLAG_FNS(name) \ |
422 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ | |
629ed0b1 | 423 | { \ |
155fead9 | 424 | stats->flags |= (1 << CFQG_stats_##name); \ |
629ed0b1 | 425 | } \ |
155fead9 | 426 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 427 | { \ |
155fead9 | 428 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
629ed0b1 | 429 | } \ |
155fead9 | 430 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 431 | { \ |
155fead9 | 432 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
629ed0b1 TH |
433 | } \ |
434 | ||
155fead9 TH |
435 | CFQG_FLAG_FNS(waiting) |
436 | CFQG_FLAG_FNS(idling) | |
437 | CFQG_FLAG_FNS(empty) | |
438 | #undef CFQG_FLAG_FNS | |
629ed0b1 TH |
439 | |
440 | /* This should be called with the queue_lock held. */ | |
155fead9 | 441 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
629ed0b1 TH |
442 | { |
443 | unsigned long long now; | |
444 | ||
155fead9 | 445 | if (!cfqg_stats_waiting(stats)) |
629ed0b1 TH |
446 | return; |
447 | ||
448 | now = sched_clock(); | |
449 | if (time_after64(now, stats->start_group_wait_time)) | |
450 | blkg_stat_add(&stats->group_wait_time, | |
451 | now - stats->start_group_wait_time); | |
155fead9 | 452 | cfqg_stats_clear_waiting(stats); |
629ed0b1 TH |
453 | } |
454 | ||
455 | /* This should be called with the queue_lock held. */ | |
155fead9 TH |
456 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
457 | struct cfq_group *curr_cfqg) | |
629ed0b1 | 458 | { |
155fead9 | 459 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 460 | |
155fead9 | 461 | if (cfqg_stats_waiting(stats)) |
629ed0b1 | 462 | return; |
155fead9 | 463 | if (cfqg == curr_cfqg) |
629ed0b1 | 464 | return; |
155fead9 TH |
465 | stats->start_group_wait_time = sched_clock(); |
466 | cfqg_stats_mark_waiting(stats); | |
629ed0b1 TH |
467 | } |
468 | ||
469 | /* This should be called with the queue_lock held. */ | |
155fead9 | 470 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
629ed0b1 TH |
471 | { |
472 | unsigned long long now; | |
473 | ||
155fead9 | 474 | if (!cfqg_stats_empty(stats)) |
629ed0b1 TH |
475 | return; |
476 | ||
477 | now = sched_clock(); | |
478 | if (time_after64(now, stats->start_empty_time)) | |
479 | blkg_stat_add(&stats->empty_time, | |
480 | now - stats->start_empty_time); | |
155fead9 | 481 | cfqg_stats_clear_empty(stats); |
629ed0b1 TH |
482 | } |
483 | ||
155fead9 | 484 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
629ed0b1 | 485 | { |
155fead9 | 486 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
629ed0b1 TH |
487 | } |
488 | ||
155fead9 | 489 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
629ed0b1 | 490 | { |
155fead9 | 491 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
492 | |
493 | if (blkg_rwstat_sum(&stats->queued)) | |
494 | return; | |
495 | ||
496 | /* | |
497 | * group is already marked empty. This can happen if cfqq got new | |
498 | * request in parent group and moved to this group while being added | |
499 | * to service tree. Just ignore the event and move on. | |
500 | */ | |
155fead9 | 501 | if (cfqg_stats_empty(stats)) |
629ed0b1 TH |
502 | return; |
503 | ||
504 | stats->start_empty_time = sched_clock(); | |
155fead9 | 505 | cfqg_stats_mark_empty(stats); |
629ed0b1 TH |
506 | } |
507 | ||
155fead9 | 508 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 509 | { |
155fead9 | 510 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 511 | |
155fead9 | 512 | if (cfqg_stats_idling(stats)) { |
629ed0b1 TH |
513 | unsigned long long now = sched_clock(); |
514 | ||
515 | if (time_after64(now, stats->start_idle_time)) | |
516 | blkg_stat_add(&stats->idle_time, | |
517 | now - stats->start_idle_time); | |
155fead9 | 518 | cfqg_stats_clear_idling(stats); |
629ed0b1 TH |
519 | } |
520 | } | |
521 | ||
155fead9 | 522 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 523 | { |
155fead9 | 524 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 525 | |
155fead9 | 526 | BUG_ON(cfqg_stats_idling(stats)); |
629ed0b1 TH |
527 | |
528 | stats->start_idle_time = sched_clock(); | |
155fead9 | 529 | cfqg_stats_mark_idling(stats); |
629ed0b1 TH |
530 | } |
531 | ||
155fead9 | 532 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
629ed0b1 | 533 | { |
155fead9 | 534 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
535 | |
536 | blkg_stat_add(&stats->avg_queue_size_sum, | |
537 | blkg_rwstat_sum(&stats->queued)); | |
538 | blkg_stat_add(&stats->avg_queue_size_samples, 1); | |
155fead9 | 539 | cfqg_stats_update_group_wait_time(stats); |
629ed0b1 TH |
540 | } |
541 | ||
542 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
543 | ||
155fead9 TH |
544 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
545 | struct cfq_group *curr_cfqg) { } | |
546 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } | |
547 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } | |
548 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } | |
549 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } | |
550 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } | |
551 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } | |
629ed0b1 TH |
552 | |
553 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
554 | ||
555 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
2ce4d50f | 556 | |
eb7d8c07 TH |
557 | static inline struct cfq_group *blkg_to_cfqg(struct blkio_group *blkg) |
558 | { | |
559 | return blkg_to_pdata(blkg, &blkio_policy_cfq); | |
560 | } | |
561 | ||
562 | static inline struct blkio_group *cfqg_to_blkg(struct cfq_group *cfqg) | |
563 | { | |
aaec55a0 | 564 | return pdata_to_blkg(cfqg); |
eb7d8c07 TH |
565 | } |
566 | ||
567 | static inline void cfqg_get(struct cfq_group *cfqg) | |
568 | { | |
569 | return blkg_get(cfqg_to_blkg(cfqg)); | |
570 | } | |
571 | ||
572 | static inline void cfqg_put(struct cfq_group *cfqg) | |
573 | { | |
574 | return blkg_put(cfqg_to_blkg(cfqg)); | |
575 | } | |
576 | ||
2868ef7b VG |
577 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
578 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ | |
579 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
0381411e | 580 | blkg_path(cfqg_to_blkg((cfqq)->cfqg)), ##args) |
2868ef7b VG |
581 | |
582 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ | |
583 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ | |
0381411e | 584 | blkg_path(cfqg_to_blkg((cfqg))), ##args) \ |
2868ef7b | 585 | |
155fead9 TH |
586 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
587 | struct cfq_group *curr_cfqg, int rw) | |
2ce4d50f | 588 | { |
155fead9 TH |
589 | blkg_rwstat_add(&cfqg->stats.queued, rw, 1); |
590 | cfqg_stats_end_empty_time(&cfqg->stats); | |
591 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); | |
2ce4d50f TH |
592 | } |
593 | ||
155fead9 TH |
594 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
595 | unsigned long time, unsigned long unaccounted_time) | |
2ce4d50f | 596 | { |
155fead9 | 597 | blkg_stat_add(&cfqg->stats.time, time); |
629ed0b1 | 598 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
155fead9 | 599 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
629ed0b1 | 600 | #endif |
2ce4d50f TH |
601 | } |
602 | ||
155fead9 | 603 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) |
2ce4d50f | 604 | { |
155fead9 | 605 | blkg_rwstat_add(&cfqg->stats.queued, rw, -1); |
2ce4d50f TH |
606 | } |
607 | ||
155fead9 | 608 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) |
2ce4d50f | 609 | { |
155fead9 | 610 | blkg_rwstat_add(&cfqg->stats.merged, rw, 1); |
2ce4d50f TH |
611 | } |
612 | ||
155fead9 TH |
613 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, |
614 | uint64_t bytes, int rw) | |
2ce4d50f | 615 | { |
155fead9 TH |
616 | blkg_stat_add(&cfqg->stats.sectors, bytes >> 9); |
617 | blkg_rwstat_add(&cfqg->stats.serviced, rw, 1); | |
618 | blkg_rwstat_add(&cfqg->stats.service_bytes, rw, bytes); | |
2ce4d50f TH |
619 | } |
620 | ||
155fead9 TH |
621 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
622 | uint64_t start_time, uint64_t io_start_time, int rw) | |
2ce4d50f | 623 | { |
155fead9 | 624 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 625 | unsigned long long now = sched_clock(); |
629ed0b1 TH |
626 | |
627 | if (time_after64(now, io_start_time)) | |
628 | blkg_rwstat_add(&stats->service_time, rw, now - io_start_time); | |
629 | if (time_after64(io_start_time, start_time)) | |
630 | blkg_rwstat_add(&stats->wait_time, rw, | |
631 | io_start_time - start_time); | |
2ce4d50f TH |
632 | } |
633 | ||
155fead9 TH |
634 | static void cfqg_stats_reset(struct blkio_group *blkg) |
635 | { | |
636 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); | |
637 | struct cfqg_stats *stats = &cfqg->stats; | |
638 | ||
639 | /* queued stats shouldn't be cleared */ | |
640 | blkg_rwstat_reset(&stats->service_bytes); | |
641 | blkg_rwstat_reset(&stats->serviced); | |
642 | blkg_rwstat_reset(&stats->merged); | |
643 | blkg_rwstat_reset(&stats->service_time); | |
644 | blkg_rwstat_reset(&stats->wait_time); | |
645 | blkg_stat_reset(&stats->time); | |
646 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
647 | blkg_stat_reset(&stats->unaccounted_time); | |
648 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
649 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
650 | blkg_stat_reset(&stats->dequeue); | |
651 | blkg_stat_reset(&stats->group_wait_time); | |
652 | blkg_stat_reset(&stats->idle_time); | |
653 | blkg_stat_reset(&stats->empty_time); | |
654 | #endif | |
655 | } | |
656 | ||
eb7d8c07 TH |
657 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
658 | ||
659 | static inline struct cfq_group *blkg_to_cfqg(struct blkio_group *blkg) { return NULL; } | |
660 | static inline struct blkio_group *cfqg_to_blkg(struct cfq_group *cfqg) { return NULL; } | |
661 | static inline void cfqg_get(struct cfq_group *cfqg) { } | |
662 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
663 | ||
7b679138 JA |
664 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
665 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
4495a7d4 | 666 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 | 667 | |
155fead9 TH |
668 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
669 | struct cfq_group *curr_cfqg, int rw) { } | |
670 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, | |
671 | unsigned long time, unsigned long unaccounted_time) { } | |
672 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) { } | |
673 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) { } | |
674 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, | |
675 | uint64_t bytes, int rw) { } | |
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 | |
eb7d8c07 TH |
679 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
680 | ||
7b679138 JA |
681 | #define cfq_log(cfqd, fmt, args...) \ |
682 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
683 | ||
615f0259 VG |
684 | /* Traverses through cfq group service trees */ |
685 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
686 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
687 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
688 | : &cfqg->service_tree_idle; \ | |
689 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
690 | (i == IDLE_WORKLOAD && j == 0); \ | |
691 | j++, st = i < IDLE_WORKLOAD ? \ | |
692 | &cfqg->service_trees[i][j]: NULL) \ | |
693 | ||
f5f2b6ce SL |
694 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
695 | struct cfq_ttime *ttime, bool group_idle) | |
696 | { | |
697 | unsigned long slice; | |
698 | if (!sample_valid(ttime->ttime_samples)) | |
699 | return false; | |
700 | if (group_idle) | |
701 | slice = cfqd->cfq_group_idle; | |
702 | else | |
703 | slice = cfqd->cfq_slice_idle; | |
704 | return ttime->ttime_mean > slice; | |
705 | } | |
615f0259 | 706 | |
02b35081 VG |
707 | static inline bool iops_mode(struct cfq_data *cfqd) |
708 | { | |
709 | /* | |
710 | * If we are not idling on queues and it is a NCQ drive, parallel | |
711 | * execution of requests is on and measuring time is not possible | |
712 | * in most of the cases until and unless we drive shallower queue | |
713 | * depths and that becomes a performance bottleneck. In such cases | |
714 | * switch to start providing fairness in terms of number of IOs. | |
715 | */ | |
716 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
717 | return true; | |
718 | else | |
719 | return false; | |
720 | } | |
721 | ||
c0324a02 CZ |
722 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
723 | { | |
724 | if (cfq_class_idle(cfqq)) | |
725 | return IDLE_WORKLOAD; | |
726 | if (cfq_class_rt(cfqq)) | |
727 | return RT_WORKLOAD; | |
728 | return BE_WORKLOAD; | |
729 | } | |
730 | ||
718eee05 CZ |
731 | |
732 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
733 | { | |
734 | if (!cfq_cfqq_sync(cfqq)) | |
735 | return ASYNC_WORKLOAD; | |
736 | if (!cfq_cfqq_idle_window(cfqq)) | |
737 | return SYNC_NOIDLE_WORKLOAD; | |
738 | return SYNC_WORKLOAD; | |
739 | } | |
740 | ||
58ff82f3 VG |
741 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
742 | struct cfq_data *cfqd, | |
743 | struct cfq_group *cfqg) | |
c0324a02 CZ |
744 | { |
745 | if (wl == IDLE_WORKLOAD) | |
cdb16e8f | 746 | return cfqg->service_tree_idle.count; |
c0324a02 | 747 | |
cdb16e8f VG |
748 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
749 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count | |
750 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
751 | } |
752 | ||
f26bd1f0 VG |
753 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
754 | struct cfq_group *cfqg) | |
755 | { | |
756 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count | |
757 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
758 | } | |
759 | ||
165125e1 | 760 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 761 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
abede6da | 762 | struct cfq_io_cq *cic, struct bio *bio, |
4f85cb96 | 763 | gfp_t gfp_mask); |
91fac317 | 764 | |
c5869807 TH |
765 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
766 | { | |
767 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
768 | return container_of(icq, struct cfq_io_cq, icq); | |
769 | } | |
770 | ||
47fdd4ca TH |
771 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
772 | struct io_context *ioc) | |
773 | { | |
774 | if (ioc) | |
775 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
776 | return NULL; | |
777 | } | |
778 | ||
c5869807 | 779 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 780 | { |
a6151c3a | 781 | return cic->cfqq[is_sync]; |
91fac317 VT |
782 | } |
783 | ||
c5869807 TH |
784 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
785 | bool is_sync) | |
91fac317 | 786 | { |
a6151c3a | 787 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
788 | } |
789 | ||
c5869807 | 790 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 791 | { |
c5869807 | 792 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
793 | } |
794 | ||
91fac317 VT |
795 | /* |
796 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
797 | * set (in which case it could also be direct WRITE). | |
798 | */ | |
a6151c3a | 799 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 800 | { |
7b6d91da | 801 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 802 | } |
1da177e4 | 803 | |
99f95e52 AM |
804 | /* |
805 | * scheduler run of queue, if there are requests pending and no one in the | |
806 | * driver that will restart queueing | |
807 | */ | |
23e018a1 | 808 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 809 | { |
7b679138 JA |
810 | if (cfqd->busy_queues) { |
811 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 812 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 813 | } |
99f95e52 AM |
814 | } |
815 | ||
44f7c160 JA |
816 | /* |
817 | * Scale schedule slice based on io priority. Use the sync time slice only | |
818 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
819 | * io only, should not get full sync slice length. | |
820 | */ | |
a6151c3a | 821 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 822 | unsigned short prio) |
44f7c160 | 823 | { |
d9e7620e | 824 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 825 | |
d9e7620e JA |
826 | WARN_ON(prio >= IOPRIO_BE_NR); |
827 | ||
828 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
829 | } | |
44f7c160 | 830 | |
d9e7620e JA |
831 | static inline int |
832 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
833 | { | |
834 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
835 | } |
836 | ||
25bc6b07 VG |
837 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
838 | { | |
839 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
840 | ||
3381cb8d | 841 | d = d * CFQ_WEIGHT_DEFAULT; |
25bc6b07 VG |
842 | do_div(d, cfqg->weight); |
843 | return d; | |
844 | } | |
845 | ||
846 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
847 | { | |
848 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
849 | if (delta > 0) | |
850 | min_vdisktime = vdisktime; | |
851 | ||
852 | return min_vdisktime; | |
853 | } | |
854 | ||
855 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
856 | { | |
857 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
858 | if (delta < 0) | |
859 | min_vdisktime = vdisktime; | |
860 | ||
861 | return min_vdisktime; | |
862 | } | |
863 | ||
864 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
865 | { | |
25bc6b07 VG |
866 | struct cfq_group *cfqg; |
867 | ||
25bc6b07 VG |
868 | if (st->left) { |
869 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
870 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
871 | cfqg->vdisktime); | |
25bc6b07 | 872 | } |
25bc6b07 VG |
873 | } |
874 | ||
5db5d642 CZ |
875 | /* |
876 | * get averaged number of queues of RT/BE priority. | |
877 | * average is updated, with a formula that gives more weight to higher numbers, | |
878 | * to quickly follows sudden increases and decrease slowly | |
879 | */ | |
880 | ||
58ff82f3 VG |
881 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
882 | struct cfq_group *cfqg, bool rt) | |
5869619c | 883 | { |
5db5d642 CZ |
884 | unsigned min_q, max_q; |
885 | unsigned mult = cfq_hist_divisor - 1; | |
886 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 887 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 888 | |
58ff82f3 VG |
889 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
890 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
891 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 892 | cfq_hist_divisor; |
58ff82f3 VG |
893 | return cfqg->busy_queues_avg[rt]; |
894 | } | |
895 | ||
896 | static inline unsigned | |
897 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
898 | { | |
899 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
900 | ||
901 | return cfq_target_latency * cfqg->weight / st->total_weight; | |
5db5d642 CZ |
902 | } |
903 | ||
c553f8e3 | 904 | static inline unsigned |
ba5bd520 | 905 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 906 | { |
5db5d642 CZ |
907 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
908 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
909 | /* |
910 | * interested queues (we consider only the ones with the same | |
911 | * priority class in the cfq group) | |
912 | */ | |
913 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
914 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
915 | unsigned sync_slice = cfqd->cfq_slice[1]; |
916 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
917 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
918 | ||
919 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
920 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
921 | /* scale low_slice according to IO priority | |
922 | * and sync vs async */ | |
923 | unsigned low_slice = | |
924 | min(slice, base_low_slice * slice / sync_slice); | |
925 | /* the adapted slice value is scaled to fit all iqs | |
926 | * into the target latency */ | |
58ff82f3 | 927 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
928 | low_slice); |
929 | } | |
930 | } | |
c553f8e3 SL |
931 | return slice; |
932 | } | |
933 | ||
934 | static inline void | |
935 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
936 | { | |
ba5bd520 | 937 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 938 | |
dae739eb | 939 | cfqq->slice_start = jiffies; |
5db5d642 | 940 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 941 | cfqq->allocated_slice = slice; |
7b679138 | 942 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
943 | } |
944 | ||
945 | /* | |
946 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
947 | * isn't valid until the first request from the dispatch is activated | |
948 | * and the slice time set. | |
949 | */ | |
a6151c3a | 950 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
951 | { |
952 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 953 | return false; |
44f7c160 | 954 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 955 | return false; |
44f7c160 | 956 | |
c1e44756 | 957 | return true; |
44f7c160 JA |
958 | } |
959 | ||
1da177e4 | 960 | /* |
5e705374 | 961 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 962 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 963 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 964 | */ |
5e705374 | 965 | static struct request * |
cf7c25cf | 966 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 967 | { |
cf7c25cf | 968 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 969 | unsigned long back_max; |
e8a99053 AM |
970 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
971 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
972 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 973 | |
5e705374 JA |
974 | if (rq1 == NULL || rq1 == rq2) |
975 | return rq2; | |
976 | if (rq2 == NULL) | |
977 | return rq1; | |
9c2c38a1 | 978 | |
229836bd NK |
979 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
980 | return rq_is_sync(rq1) ? rq1 : rq2; | |
981 | ||
65299a3b CH |
982 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
983 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 984 | |
83096ebf TH |
985 | s1 = blk_rq_pos(rq1); |
986 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 987 | |
1da177e4 LT |
988 | /* |
989 | * by definition, 1KiB is 2 sectors | |
990 | */ | |
991 | back_max = cfqd->cfq_back_max * 2; | |
992 | ||
993 | /* | |
994 | * Strict one way elevator _except_ in the case where we allow | |
995 | * short backward seeks which are biased as twice the cost of a | |
996 | * similar forward seek. | |
997 | */ | |
998 | if (s1 >= last) | |
999 | d1 = s1 - last; | |
1000 | else if (s1 + back_max >= last) | |
1001 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
1002 | else | |
e8a99053 | 1003 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
1004 | |
1005 | if (s2 >= last) | |
1006 | d2 = s2 - last; | |
1007 | else if (s2 + back_max >= last) | |
1008 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
1009 | else | |
e8a99053 | 1010 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
1011 | |
1012 | /* Found required data */ | |
e8a99053 AM |
1013 | |
1014 | /* | |
1015 | * By doing switch() on the bit mask "wrap" we avoid having to | |
1016 | * check two variables for all permutations: --> faster! | |
1017 | */ | |
1018 | switch (wrap) { | |
5e705374 | 1019 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 1020 | if (d1 < d2) |
5e705374 | 1021 | return rq1; |
e8a99053 | 1022 | else if (d2 < d1) |
5e705374 | 1023 | return rq2; |
e8a99053 AM |
1024 | else { |
1025 | if (s1 >= s2) | |
5e705374 | 1026 | return rq1; |
e8a99053 | 1027 | else |
5e705374 | 1028 | return rq2; |
e8a99053 | 1029 | } |
1da177e4 | 1030 | |
e8a99053 | 1031 | case CFQ_RQ2_WRAP: |
5e705374 | 1032 | return rq1; |
e8a99053 | 1033 | case CFQ_RQ1_WRAP: |
5e705374 JA |
1034 | return rq2; |
1035 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
1036 | default: |
1037 | /* | |
1038 | * Since both rqs are wrapped, | |
1039 | * start with the one that's further behind head | |
1040 | * (--> only *one* back seek required), | |
1041 | * since back seek takes more time than forward. | |
1042 | */ | |
1043 | if (s1 <= s2) | |
5e705374 | 1044 | return rq1; |
1da177e4 | 1045 | else |
5e705374 | 1046 | return rq2; |
1da177e4 LT |
1047 | } |
1048 | } | |
1049 | ||
498d3aa2 JA |
1050 | /* |
1051 | * The below is leftmost cache rbtree addon | |
1052 | */ | |
0871714e | 1053 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 1054 | { |
615f0259 VG |
1055 | /* Service tree is empty */ |
1056 | if (!root->count) | |
1057 | return NULL; | |
1058 | ||
cc09e299 JA |
1059 | if (!root->left) |
1060 | root->left = rb_first(&root->rb); | |
1061 | ||
0871714e JA |
1062 | if (root->left) |
1063 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
1064 | ||
1065 | return NULL; | |
cc09e299 JA |
1066 | } |
1067 | ||
1fa8f6d6 VG |
1068 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
1069 | { | |
1070 | if (!root->left) | |
1071 | root->left = rb_first(&root->rb); | |
1072 | ||
1073 | if (root->left) | |
1074 | return rb_entry_cfqg(root->left); | |
1075 | ||
1076 | return NULL; | |
1077 | } | |
1078 | ||
a36e71f9 JA |
1079 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
1080 | { | |
1081 | rb_erase(n, root); | |
1082 | RB_CLEAR_NODE(n); | |
1083 | } | |
1084 | ||
cc09e299 JA |
1085 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
1086 | { | |
1087 | if (root->left == n) | |
1088 | root->left = NULL; | |
a36e71f9 | 1089 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 1090 | --root->count; |
cc09e299 JA |
1091 | } |
1092 | ||
1da177e4 LT |
1093 | /* |
1094 | * would be nice to take fifo expire time into account as well | |
1095 | */ | |
5e705374 JA |
1096 | static struct request * |
1097 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1098 | struct request *last) | |
1da177e4 | 1099 | { |
21183b07 JA |
1100 | struct rb_node *rbnext = rb_next(&last->rb_node); |
1101 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 1102 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 1103 | |
21183b07 | 1104 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
1105 | |
1106 | if (rbprev) | |
5e705374 | 1107 | prev = rb_entry_rq(rbprev); |
1da177e4 | 1108 | |
21183b07 | 1109 | if (rbnext) |
5e705374 | 1110 | next = rb_entry_rq(rbnext); |
21183b07 JA |
1111 | else { |
1112 | rbnext = rb_first(&cfqq->sort_list); | |
1113 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 1114 | next = rb_entry_rq(rbnext); |
21183b07 | 1115 | } |
1da177e4 | 1116 | |
cf7c25cf | 1117 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
1118 | } |
1119 | ||
d9e7620e JA |
1120 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
1121 | struct cfq_queue *cfqq) | |
1da177e4 | 1122 | { |
d9e7620e JA |
1123 | /* |
1124 | * just an approximation, should be ok. | |
1125 | */ | |
cdb16e8f | 1126 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 1127 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
1128 | } |
1129 | ||
1fa8f6d6 VG |
1130 | static inline s64 |
1131 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1132 | { | |
1133 | return cfqg->vdisktime - st->min_vdisktime; | |
1134 | } | |
1135 | ||
1136 | static void | |
1137 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1138 | { | |
1139 | struct rb_node **node = &st->rb.rb_node; | |
1140 | struct rb_node *parent = NULL; | |
1141 | struct cfq_group *__cfqg; | |
1142 | s64 key = cfqg_key(st, cfqg); | |
1143 | int left = 1; | |
1144 | ||
1145 | while (*node != NULL) { | |
1146 | parent = *node; | |
1147 | __cfqg = rb_entry_cfqg(parent); | |
1148 | ||
1149 | if (key < cfqg_key(st, __cfqg)) | |
1150 | node = &parent->rb_left; | |
1151 | else { | |
1152 | node = &parent->rb_right; | |
1153 | left = 0; | |
1154 | } | |
1155 | } | |
1156 | ||
1157 | if (left) | |
1158 | st->left = &cfqg->rb_node; | |
1159 | ||
1160 | rb_link_node(&cfqg->rb_node, parent, node); | |
1161 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
1162 | } | |
1163 | ||
1164 | static void | |
8184f93e JT |
1165 | cfq_update_group_weight(struct cfq_group *cfqg) |
1166 | { | |
1167 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
3381cb8d | 1168 | if (cfqg->new_weight) { |
8184f93e | 1169 | cfqg->weight = cfqg->new_weight; |
3381cb8d | 1170 | cfqg->new_weight = 0; |
8184f93e JT |
1171 | } |
1172 | } | |
1173 | ||
1174 | static void | |
1175 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1176 | { | |
1177 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
1178 | ||
1179 | cfq_update_group_weight(cfqg); | |
1180 | __cfq_group_service_tree_add(st, cfqg); | |
1181 | st->total_weight += cfqg->weight; | |
1182 | } | |
1183 | ||
1184 | static void | |
1185 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
1186 | { |
1187 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1188 | struct cfq_group *__cfqg; | |
1189 | struct rb_node *n; | |
1190 | ||
1191 | cfqg->nr_cfqq++; | |
760701bf | 1192 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
1193 | return; |
1194 | ||
1195 | /* | |
1196 | * Currently put the group at the end. Later implement something | |
1197 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 1198 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
1199 | */ |
1200 | n = rb_last(&st->rb); | |
1201 | if (n) { | |
1202 | __cfqg = rb_entry_cfqg(n); | |
1203 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
1204 | } else | |
1205 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
1206 | cfq_group_service_tree_add(st, cfqg); |
1207 | } | |
1fa8f6d6 | 1208 | |
8184f93e JT |
1209 | static void |
1210 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1211 | { | |
1212 | st->total_weight -= cfqg->weight; | |
1213 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) | |
1214 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
1215 | } |
1216 | ||
1217 | static void | |
8184f93e | 1218 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
1219 | { |
1220 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1221 | ||
1222 | BUG_ON(cfqg->nr_cfqq < 1); | |
1223 | cfqg->nr_cfqq--; | |
25bc6b07 | 1224 | |
1fa8f6d6 VG |
1225 | /* If there are other cfq queues under this group, don't delete it */ |
1226 | if (cfqg->nr_cfqq) | |
1227 | return; | |
1228 | ||
2868ef7b | 1229 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 1230 | cfq_group_service_tree_del(st, cfqg); |
dae739eb | 1231 | cfqg->saved_workload_slice = 0; |
155fead9 | 1232 | cfqg_stats_update_dequeue(cfqg); |
dae739eb VG |
1233 | } |
1234 | ||
167400d3 JT |
1235 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
1236 | unsigned int *unaccounted_time) | |
dae739eb | 1237 | { |
f75edf2d | 1238 | unsigned int slice_used; |
dae739eb VG |
1239 | |
1240 | /* | |
1241 | * Queue got expired before even a single request completed or | |
1242 | * got expired immediately after first request completion. | |
1243 | */ | |
1244 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
1245 | /* | |
1246 | * Also charge the seek time incurred to the group, otherwise | |
1247 | * if there are mutiple queues in the group, each can dispatch | |
1248 | * a single request on seeky media and cause lots of seek time | |
1249 | * and group will never know it. | |
1250 | */ | |
1251 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
1252 | 1); | |
1253 | } else { | |
1254 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
1255 | if (slice_used > cfqq->allocated_slice) { |
1256 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 1257 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
1258 | } |
1259 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
1260 | *unaccounted_time += cfqq->slice_start - | |
1261 | cfqq->dispatch_start; | |
dae739eb VG |
1262 | } |
1263 | ||
dae739eb VG |
1264 | return slice_used; |
1265 | } | |
1266 | ||
1267 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1268 | struct cfq_queue *cfqq) |
dae739eb VG |
1269 | { |
1270 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 1271 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1272 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1273 | - cfqg->service_tree_idle.count; | |
1274 | ||
1275 | BUG_ON(nr_sync < 0); | |
167400d3 | 1276 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1277 | |
02b35081 VG |
1278 | if (iops_mode(cfqd)) |
1279 | charge = cfqq->slice_dispatch; | |
1280 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1281 | charge = cfqq->allocated_slice; | |
dae739eb VG |
1282 | |
1283 | /* Can't update vdisktime while group is on service tree */ | |
8184f93e | 1284 | cfq_group_service_tree_del(st, cfqg); |
02b35081 | 1285 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
8184f93e JT |
1286 | /* If a new weight was requested, update now, off tree */ |
1287 | cfq_group_service_tree_add(st, cfqg); | |
dae739eb VG |
1288 | |
1289 | /* This group is being expired. Save the context */ | |
1290 | if (time_after(cfqd->workload_expires, jiffies)) { | |
1291 | cfqg->saved_workload_slice = cfqd->workload_expires | |
1292 | - jiffies; | |
1293 | cfqg->saved_workload = cfqd->serving_type; | |
1294 | cfqg->saved_serving_prio = cfqd->serving_prio; | |
1295 | } else | |
1296 | cfqg->saved_workload_slice = 0; | |
2868ef7b VG |
1297 | |
1298 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1299 | st->min_vdisktime); | |
fd16d263 JP |
1300 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1301 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1302 | used_sl, cfqq->slice_dispatch, charge, | |
1303 | iops_mode(cfqd), cfqq->nr_sectors); | |
155fead9 TH |
1304 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
1305 | cfqg_stats_set_start_empty_time(cfqg); | |
1fa8f6d6 VG |
1306 | } |
1307 | ||
f51b802c TH |
1308 | /** |
1309 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1310 | * @cfqg: cfq_group to initialize | |
1311 | * | |
1312 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1313 | * is enabled or not. | |
1314 | */ | |
1315 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1316 | { | |
1317 | struct cfq_rb_root *st; | |
1318 | int i, j; | |
1319 | ||
1320 | for_each_cfqg_st(cfqg, i, j, st) | |
1321 | *st = CFQ_RB_ROOT; | |
1322 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1323 | ||
1324 | cfqg->ttime.last_end_request = jiffies; | |
1325 | } | |
1326 | ||
25fb5169 | 1327 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
0381411e | 1328 | static void cfq_init_blkio_group(struct blkio_group *blkg) |
f469a7b4 | 1329 | { |
0381411e | 1330 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
25fb5169 | 1331 | |
f51b802c | 1332 | cfq_init_cfqg_base(cfqg); |
3381cb8d | 1333 | cfqg->weight = blkg->blkcg->cfq_weight; |
25fb5169 VG |
1334 | } |
1335 | ||
1336 | /* | |
3e59cf9d VG |
1337 | * Search for the cfq group current task belongs to. request_queue lock must |
1338 | * be held. | |
25fb5169 | 1339 | */ |
cd1604fa TH |
1340 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
1341 | struct blkio_cgroup *blkcg) | |
25fb5169 | 1342 | { |
f469a7b4 | 1343 | struct request_queue *q = cfqd->queue; |
cd1604fa | 1344 | struct cfq_group *cfqg = NULL; |
25fb5169 | 1345 | |
cd1604fa TH |
1346 | /* avoid lookup for the common case where there's no blkio cgroup */ |
1347 | if (blkcg == &blkio_root_cgroup) { | |
1348 | cfqg = cfqd->root_group; | |
1349 | } else { | |
1350 | struct blkio_group *blkg; | |
f469a7b4 | 1351 | |
aaec55a0 | 1352 | blkg = blkg_lookup_create(blkcg, q, false); |
cd1604fa | 1353 | if (!IS_ERR(blkg)) |
0381411e | 1354 | cfqg = blkg_to_cfqg(blkg); |
cd1604fa | 1355 | } |
f469a7b4 | 1356 | |
25fb5169 VG |
1357 | return cfqg; |
1358 | } | |
1359 | ||
1360 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1361 | { | |
1362 | /* Currently, all async queues are mapped to root group */ | |
1363 | if (!cfq_cfqq_sync(cfqq)) | |
f51b802c | 1364 | cfqg = cfqq->cfqd->root_group; |
25fb5169 VG |
1365 | |
1366 | cfqq->cfqg = cfqg; | |
b1c35769 | 1367 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1368 | cfqg_get(cfqg); |
b1c35769 VG |
1369 | } |
1370 | ||
d366e7ec | 1371 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, void *pdata, int off) |
60c2bc2d | 1372 | { |
d366e7ec | 1373 | struct cfq_group *cfqg = pdata; |
3381cb8d TH |
1374 | |
1375 | if (!cfqg->dev_weight) | |
60c2bc2d | 1376 | return 0; |
d366e7ec | 1377 | return __blkg_prfill_u64(sf, pdata, cfqg->dev_weight); |
60c2bc2d TH |
1378 | } |
1379 | ||
3381cb8d TH |
1380 | static int cfqg_print_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1381 | struct seq_file *sf) | |
60c2bc2d TH |
1382 | { |
1383 | blkcg_print_blkgs(sf, cgroup_to_blkio_cgroup(cgrp), | |
3381cb8d | 1384 | cfqg_prfill_weight_device, BLKIO_POLICY_PROP, 0, |
60c2bc2d TH |
1385 | false); |
1386 | return 0; | |
1387 | } | |
1388 | ||
3381cb8d TH |
1389 | static int cfq_print_weight(struct cgroup *cgrp, struct cftype *cft, |
1390 | struct seq_file *sf) | |
60c2bc2d | 1391 | { |
3381cb8d | 1392 | seq_printf(sf, "%u\n", cgroup_to_blkio_cgroup(cgrp)->cfq_weight); |
60c2bc2d TH |
1393 | return 0; |
1394 | } | |
1395 | ||
3381cb8d TH |
1396 | static int cfqg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1397 | const char *buf) | |
60c2bc2d TH |
1398 | { |
1399 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); | |
60c2bc2d | 1400 | struct blkg_conf_ctx ctx; |
3381cb8d | 1401 | struct cfq_group *cfqg; |
60c2bc2d TH |
1402 | int ret; |
1403 | ||
1404 | ret = blkg_conf_prep(blkcg, buf, &ctx); | |
1405 | if (ret) | |
1406 | return ret; | |
1407 | ||
1408 | ret = -EINVAL; | |
3381cb8d TH |
1409 | cfqg = blkg_to_cfqg(ctx.blkg); |
1410 | if (cfqg && (!ctx.v || (ctx.v >= CFQ_WEIGHT_MIN && | |
1411 | ctx.v <= CFQ_WEIGHT_MAX))) { | |
1412 | cfqg->dev_weight = ctx.v; | |
1413 | cfqg->new_weight = cfqg->dev_weight ?: blkcg->cfq_weight; | |
60c2bc2d TH |
1414 | ret = 0; |
1415 | } | |
1416 | ||
1417 | blkg_conf_finish(&ctx); | |
1418 | return ret; | |
1419 | } | |
1420 | ||
3381cb8d | 1421 | static int cfq_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) |
60c2bc2d TH |
1422 | { |
1423 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1424 | struct blkio_group *blkg; | |
1425 | struct hlist_node *n; | |
1426 | ||
3381cb8d | 1427 | if (val < CFQ_WEIGHT_MIN || val > CFQ_WEIGHT_MAX) |
60c2bc2d TH |
1428 | return -EINVAL; |
1429 | ||
1430 | spin_lock_irq(&blkcg->lock); | |
3381cb8d | 1431 | blkcg->cfq_weight = (unsigned int)val; |
60c2bc2d TH |
1432 | |
1433 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
3381cb8d | 1434 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
60c2bc2d | 1435 | |
3381cb8d TH |
1436 | if (cfqg && !cfqg->dev_weight) |
1437 | cfqg->new_weight = blkcg->cfq_weight; | |
60c2bc2d TH |
1438 | } |
1439 | ||
1440 | spin_unlock_irq(&blkcg->lock); | |
1441 | return 0; | |
1442 | } | |
1443 | ||
1444 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
d366e7ec | 1445 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, void *pdata, int off) |
60c2bc2d | 1446 | { |
d366e7ec | 1447 | struct cfq_group *cfqg = pdata; |
155fead9 | 1448 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
60c2bc2d TH |
1449 | u64 v = 0; |
1450 | ||
1451 | if (samples) { | |
155fead9 | 1452 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
60c2bc2d TH |
1453 | do_div(v, samples); |
1454 | } | |
d366e7ec | 1455 | __blkg_prfill_u64(sf, pdata, v); |
60c2bc2d TH |
1456 | return 0; |
1457 | } | |
1458 | ||
1459 | /* print avg_queue_size */ | |
155fead9 TH |
1460 | static int cfqg_print_avg_queue_size(struct cgroup *cgrp, struct cftype *cft, |
1461 | struct seq_file *sf) | |
60c2bc2d TH |
1462 | { |
1463 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); | |
1464 | ||
155fead9 | 1465 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_avg_queue_size, |
60c2bc2d TH |
1466 | BLKIO_POLICY_PROP, 0, false); |
1467 | return 0; | |
1468 | } | |
1469 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1470 | ||
1471 | static struct cftype cfq_blkcg_files[] = { | |
1472 | { | |
1473 | .name = "weight_device", | |
3381cb8d TH |
1474 | .read_seq_string = cfqg_print_weight_device, |
1475 | .write_string = cfqg_set_weight_device, | |
60c2bc2d TH |
1476 | .max_write_len = 256, |
1477 | }, | |
1478 | { | |
1479 | .name = "weight", | |
3381cb8d TH |
1480 | .read_seq_string = cfq_print_weight, |
1481 | .write_u64 = cfq_set_weight, | |
60c2bc2d TH |
1482 | }, |
1483 | { | |
1484 | .name = "time", | |
1485 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1486 | offsetof(struct cfq_group, stats.time)), |
60c2bc2d TH |
1487 | .read_seq_string = blkcg_print_stat, |
1488 | }, | |
1489 | { | |
1490 | .name = "sectors", | |
1491 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1492 | offsetof(struct cfq_group, stats.sectors)), |
41b38b6d | 1493 | .read_seq_string = blkcg_print_stat, |
60c2bc2d TH |
1494 | }, |
1495 | { | |
1496 | .name = "io_service_bytes", | |
1497 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1498 | offsetof(struct cfq_group, stats.service_bytes)), |
41b38b6d | 1499 | .read_seq_string = blkcg_print_rwstat, |
60c2bc2d TH |
1500 | }, |
1501 | { | |
1502 | .name = "io_serviced", | |
1503 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1504 | offsetof(struct cfq_group, stats.serviced)), |
41b38b6d | 1505 | .read_seq_string = blkcg_print_rwstat, |
60c2bc2d TH |
1506 | }, |
1507 | { | |
1508 | .name = "io_service_time", | |
1509 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1510 | offsetof(struct cfq_group, stats.service_time)), |
60c2bc2d TH |
1511 | .read_seq_string = blkcg_print_rwstat, |
1512 | }, | |
1513 | { | |
1514 | .name = "io_wait_time", | |
1515 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1516 | offsetof(struct cfq_group, stats.wait_time)), |
60c2bc2d TH |
1517 | .read_seq_string = blkcg_print_rwstat, |
1518 | }, | |
1519 | { | |
1520 | .name = "io_merged", | |
1521 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1522 | offsetof(struct cfq_group, stats.merged)), |
60c2bc2d TH |
1523 | .read_seq_string = blkcg_print_rwstat, |
1524 | }, | |
1525 | { | |
1526 | .name = "io_queued", | |
1527 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1528 | offsetof(struct cfq_group, stats.queued)), |
60c2bc2d TH |
1529 | .read_seq_string = blkcg_print_rwstat, |
1530 | }, | |
1531 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
1532 | { | |
1533 | .name = "avg_queue_size", | |
155fead9 | 1534 | .read_seq_string = cfqg_print_avg_queue_size, |
60c2bc2d TH |
1535 | }, |
1536 | { | |
1537 | .name = "group_wait_time", | |
1538 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1539 | offsetof(struct cfq_group, stats.group_wait_time)), |
60c2bc2d TH |
1540 | .read_seq_string = blkcg_print_stat, |
1541 | }, | |
1542 | { | |
1543 | .name = "idle_time", | |
1544 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1545 | offsetof(struct cfq_group, stats.idle_time)), |
60c2bc2d TH |
1546 | .read_seq_string = blkcg_print_stat, |
1547 | }, | |
1548 | { | |
1549 | .name = "empty_time", | |
1550 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1551 | offsetof(struct cfq_group, stats.empty_time)), |
60c2bc2d TH |
1552 | .read_seq_string = blkcg_print_stat, |
1553 | }, | |
1554 | { | |
1555 | .name = "dequeue", | |
1556 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1557 | offsetof(struct cfq_group, stats.dequeue)), |
60c2bc2d TH |
1558 | .read_seq_string = blkcg_print_stat, |
1559 | }, | |
1560 | { | |
1561 | .name = "unaccounted_time", | |
1562 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, | |
155fead9 | 1563 | offsetof(struct cfq_group, stats.unaccounted_time)), |
60c2bc2d TH |
1564 | .read_seq_string = blkcg_print_stat, |
1565 | }, | |
1566 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1567 | { } /* terminate */ | |
1568 | }; | |
25fb5169 | 1569 | #else /* GROUP_IOSCHED */ |
cd1604fa TH |
1570 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
1571 | struct blkio_cgroup *blkcg) | |
25fb5169 | 1572 | { |
f51b802c | 1573 | return cfqd->root_group; |
25fb5169 | 1574 | } |
7f1dc8a2 | 1575 | |
25fb5169 VG |
1576 | static inline void |
1577 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1578 | cfqq->cfqg = cfqg; | |
1579 | } | |
1580 | ||
1581 | #endif /* GROUP_IOSCHED */ | |
1582 | ||
498d3aa2 | 1583 | /* |
c0324a02 | 1584 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1585 | * requests waiting to be processed. It is sorted in the order that |
1586 | * we will service the queues. | |
1587 | */ | |
a36e71f9 | 1588 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1589 | bool add_front) |
d9e7620e | 1590 | { |
0871714e JA |
1591 | struct rb_node **p, *parent; |
1592 | struct cfq_queue *__cfqq; | |
d9e7620e | 1593 | unsigned long rb_key; |
c0324a02 | 1594 | struct cfq_rb_root *service_tree; |
498d3aa2 | 1595 | int left; |
dae739eb | 1596 | int new_cfqq = 1; |
ae30c286 | 1597 | |
cdb16e8f | 1598 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
65b32a57 | 1599 | cfqq_type(cfqq)); |
0871714e JA |
1600 | if (cfq_class_idle(cfqq)) { |
1601 | rb_key = CFQ_IDLE_DELAY; | |
aa6f6a3d | 1602 | parent = rb_last(&service_tree->rb); |
0871714e JA |
1603 | if (parent && parent != &cfqq->rb_node) { |
1604 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1605 | rb_key += __cfqq->rb_key; | |
1606 | } else | |
1607 | rb_key += jiffies; | |
1608 | } else if (!add_front) { | |
b9c8946b JA |
1609 | /* |
1610 | * Get our rb key offset. Subtract any residual slice | |
1611 | * value carried from last service. A negative resid | |
1612 | * count indicates slice overrun, and this should position | |
1613 | * the next service time further away in the tree. | |
1614 | */ | |
edd75ffd | 1615 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1616 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1617 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1618 | } else { |
1619 | rb_key = -HZ; | |
aa6f6a3d | 1620 | __cfqq = cfq_rb_first(service_tree); |
48e025e6 CZ |
1621 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1622 | } | |
1da177e4 | 1623 | |
d9e7620e | 1624 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1625 | new_cfqq = 0; |
99f9628a | 1626 | /* |
d9e7620e | 1627 | * same position, nothing more to do |
99f9628a | 1628 | */ |
c0324a02 CZ |
1629 | if (rb_key == cfqq->rb_key && |
1630 | cfqq->service_tree == service_tree) | |
d9e7620e | 1631 | return; |
1da177e4 | 1632 | |
aa6f6a3d CZ |
1633 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1634 | cfqq->service_tree = NULL; | |
1da177e4 | 1635 | } |
d9e7620e | 1636 | |
498d3aa2 | 1637 | left = 1; |
0871714e | 1638 | parent = NULL; |
aa6f6a3d CZ |
1639 | cfqq->service_tree = service_tree; |
1640 | p = &service_tree->rb.rb_node; | |
d9e7620e | 1641 | while (*p) { |
67060e37 | 1642 | struct rb_node **n; |
cc09e299 | 1643 | |
d9e7620e JA |
1644 | parent = *p; |
1645 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1646 | ||
0c534e0a | 1647 | /* |
c0324a02 | 1648 | * sort by key, that represents service time. |
0c534e0a | 1649 | */ |
c0324a02 | 1650 | if (time_before(rb_key, __cfqq->rb_key)) |
67060e37 | 1651 | n = &(*p)->rb_left; |
c0324a02 | 1652 | else { |
67060e37 | 1653 | n = &(*p)->rb_right; |
cc09e299 | 1654 | left = 0; |
c0324a02 | 1655 | } |
67060e37 JA |
1656 | |
1657 | p = n; | |
d9e7620e JA |
1658 | } |
1659 | ||
cc09e299 | 1660 | if (left) |
aa6f6a3d | 1661 | service_tree->left = &cfqq->rb_node; |
cc09e299 | 1662 | |
d9e7620e JA |
1663 | cfqq->rb_key = rb_key; |
1664 | rb_link_node(&cfqq->rb_node, parent, p); | |
aa6f6a3d CZ |
1665 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
1666 | service_tree->count++; | |
20359f27 | 1667 | if (add_front || !new_cfqq) |
dae739eb | 1668 | return; |
8184f93e | 1669 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1670 | } |
1671 | ||
a36e71f9 | 1672 | static struct cfq_queue * |
f2d1f0ae JA |
1673 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1674 | sector_t sector, struct rb_node **ret_parent, | |
1675 | struct rb_node ***rb_link) | |
a36e71f9 | 1676 | { |
a36e71f9 JA |
1677 | struct rb_node **p, *parent; |
1678 | struct cfq_queue *cfqq = NULL; | |
1679 | ||
1680 | parent = NULL; | |
1681 | p = &root->rb_node; | |
1682 | while (*p) { | |
1683 | struct rb_node **n; | |
1684 | ||
1685 | parent = *p; | |
1686 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1687 | ||
1688 | /* | |
1689 | * Sort strictly based on sector. Smallest to the left, | |
1690 | * largest to the right. | |
1691 | */ | |
2e46e8b2 | 1692 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1693 | n = &(*p)->rb_right; |
2e46e8b2 | 1694 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1695 | n = &(*p)->rb_left; |
1696 | else | |
1697 | break; | |
1698 | p = n; | |
3ac6c9f8 | 1699 | cfqq = NULL; |
a36e71f9 JA |
1700 | } |
1701 | ||
1702 | *ret_parent = parent; | |
1703 | if (rb_link) | |
1704 | *rb_link = p; | |
3ac6c9f8 | 1705 | return cfqq; |
a36e71f9 JA |
1706 | } |
1707 | ||
1708 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1709 | { | |
a36e71f9 JA |
1710 | struct rb_node **p, *parent; |
1711 | struct cfq_queue *__cfqq; | |
1712 | ||
f2d1f0ae JA |
1713 | if (cfqq->p_root) { |
1714 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1715 | cfqq->p_root = NULL; | |
1716 | } | |
a36e71f9 JA |
1717 | |
1718 | if (cfq_class_idle(cfqq)) | |
1719 | return; | |
1720 | if (!cfqq->next_rq) | |
1721 | return; | |
1722 | ||
f2d1f0ae | 1723 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1724 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1725 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1726 | if (!__cfqq) { |
1727 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1728 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1729 | } else | |
1730 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1731 | } |
1732 | ||
498d3aa2 JA |
1733 | /* |
1734 | * Update cfqq's position in the service tree. | |
1735 | */ | |
edd75ffd | 1736 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1737 | { |
6d048f53 JA |
1738 | /* |
1739 | * Resorting requires the cfqq to be on the RR list already. | |
1740 | */ | |
a36e71f9 | 1741 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1742 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1743 | cfq_prio_tree_add(cfqd, cfqq); |
1744 | } | |
6d048f53 JA |
1745 | } |
1746 | ||
1da177e4 LT |
1747 | /* |
1748 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1749 | * the pending list according to last request service |
1da177e4 | 1750 | */ |
febffd61 | 1751 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1752 | { |
7b679138 | 1753 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1754 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1755 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 1756 | cfqd->busy_queues++; |
ef8a41df SL |
1757 | if (cfq_cfqq_sync(cfqq)) |
1758 | cfqd->busy_sync_queues++; | |
1da177e4 | 1759 | |
edd75ffd | 1760 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1761 | } |
1762 | ||
498d3aa2 JA |
1763 | /* |
1764 | * Called when the cfqq no longer has requests pending, remove it from | |
1765 | * the service tree. | |
1766 | */ | |
febffd61 | 1767 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1768 | { |
7b679138 | 1769 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1770 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1771 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1772 | |
aa6f6a3d CZ |
1773 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1774 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1775 | cfqq->service_tree = NULL; | |
1776 | } | |
f2d1f0ae JA |
1777 | if (cfqq->p_root) { |
1778 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1779 | cfqq->p_root = NULL; | |
1780 | } | |
d9e7620e | 1781 | |
8184f93e | 1782 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1783 | BUG_ON(!cfqd->busy_queues); |
1784 | cfqd->busy_queues--; | |
ef8a41df SL |
1785 | if (cfq_cfqq_sync(cfqq)) |
1786 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
1787 | } |
1788 | ||
1789 | /* | |
1790 | * rb tree support functions | |
1791 | */ | |
febffd61 | 1792 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1793 | { |
5e705374 | 1794 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1795 | const int sync = rq_is_sync(rq); |
1da177e4 | 1796 | |
b4878f24 JA |
1797 | BUG_ON(!cfqq->queued[sync]); |
1798 | cfqq->queued[sync]--; | |
1da177e4 | 1799 | |
5e705374 | 1800 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1801 | |
f04a6424 VG |
1802 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1803 | /* | |
1804 | * Queue will be deleted from service tree when we actually | |
1805 | * expire it later. Right now just remove it from prio tree | |
1806 | * as it is empty. | |
1807 | */ | |
1808 | if (cfqq->p_root) { | |
1809 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1810 | cfqq->p_root = NULL; | |
1811 | } | |
1812 | } | |
1da177e4 LT |
1813 | } |
1814 | ||
5e705374 | 1815 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1816 | { |
5e705374 | 1817 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1818 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 1819 | struct request *prev; |
1da177e4 | 1820 | |
5380a101 | 1821 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 1822 | |
796d5116 | 1823 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
1824 | |
1825 | if (!cfq_cfqq_on_rr(cfqq)) | |
1826 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1827 | |
1828 | /* | |
1829 | * check if this request is a better next-serve candidate | |
1830 | */ | |
a36e71f9 | 1831 | prev = cfqq->next_rq; |
cf7c25cf | 1832 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1833 | |
1834 | /* | |
1835 | * adjust priority tree position, if ->next_rq changes | |
1836 | */ | |
1837 | if (prev != cfqq->next_rq) | |
1838 | cfq_prio_tree_add(cfqd, cfqq); | |
1839 | ||
5044eed4 | 1840 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1841 | } |
1842 | ||
febffd61 | 1843 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1844 | { |
5380a101 JA |
1845 | elv_rb_del(&cfqq->sort_list, rq); |
1846 | cfqq->queued[rq_is_sync(rq)]--; | |
155fead9 | 1847 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
5e705374 | 1848 | cfq_add_rq_rb(rq); |
155fead9 TH |
1849 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
1850 | rq->cmd_flags); | |
1da177e4 LT |
1851 | } |
1852 | ||
206dc69b JA |
1853 | static struct request * |
1854 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1855 | { |
206dc69b | 1856 | struct task_struct *tsk = current; |
c5869807 | 1857 | struct cfq_io_cq *cic; |
206dc69b | 1858 | struct cfq_queue *cfqq; |
1da177e4 | 1859 | |
4ac845a2 | 1860 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1861 | if (!cic) |
1862 | return NULL; | |
1863 | ||
1864 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
1865 | if (cfqq) { |
1866 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
1867 | ||
21183b07 | 1868 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 1869 | } |
1da177e4 | 1870 | |
1da177e4 LT |
1871 | return NULL; |
1872 | } | |
1873 | ||
165125e1 | 1874 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1875 | { |
22e2c507 | 1876 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 1877 | |
53c583d2 | 1878 | cfqd->rq_in_driver++; |
7b679138 | 1879 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 1880 | cfqd->rq_in_driver); |
25776e35 | 1881 | |
5b93629b | 1882 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
1883 | } |
1884 | ||
165125e1 | 1885 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1886 | { |
b4878f24 JA |
1887 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1888 | ||
53c583d2 CZ |
1889 | WARN_ON(!cfqd->rq_in_driver); |
1890 | cfqd->rq_in_driver--; | |
7b679138 | 1891 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 1892 | cfqd->rq_in_driver); |
1da177e4 LT |
1893 | } |
1894 | ||
b4878f24 | 1895 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 1896 | { |
5e705374 | 1897 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 1898 | |
5e705374 JA |
1899 | if (cfqq->next_rq == rq) |
1900 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 1901 | |
b4878f24 | 1902 | list_del_init(&rq->queuelist); |
5e705374 | 1903 | cfq_del_rq_rb(rq); |
374f84ac | 1904 | |
45333d5a | 1905 | cfqq->cfqd->rq_queued--; |
155fead9 | 1906 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
65299a3b CH |
1907 | if (rq->cmd_flags & REQ_PRIO) { |
1908 | WARN_ON(!cfqq->prio_pending); | |
1909 | cfqq->prio_pending--; | |
b53d1ed7 | 1910 | } |
1da177e4 LT |
1911 | } |
1912 | ||
165125e1 JA |
1913 | static int cfq_merge(struct request_queue *q, struct request **req, |
1914 | struct bio *bio) | |
1da177e4 LT |
1915 | { |
1916 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1917 | struct request *__rq; | |
1da177e4 | 1918 | |
206dc69b | 1919 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 1920 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
1921 | *req = __rq; |
1922 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
1923 | } |
1924 | ||
1925 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
1926 | } |
1927 | ||
165125e1 | 1928 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 1929 | int type) |
1da177e4 | 1930 | { |
21183b07 | 1931 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 1932 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 1933 | |
5e705374 | 1934 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 1935 | } |
1da177e4 LT |
1936 | } |
1937 | ||
812d4026 DS |
1938 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
1939 | struct bio *bio) | |
1940 | { | |
155fead9 | 1941 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_rw); |
812d4026 DS |
1942 | } |
1943 | ||
1da177e4 | 1944 | static void |
165125e1 | 1945 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
1946 | struct request *next) |
1947 | { | |
cf7c25cf | 1948 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
1949 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1950 | ||
22e2c507 JA |
1951 | /* |
1952 | * reposition in fifo if next is older than rq | |
1953 | */ | |
1954 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
30996f40 | 1955 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
22e2c507 | 1956 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
1957 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1958 | } | |
22e2c507 | 1959 | |
cf7c25cf CZ |
1960 | if (cfqq->next_rq == next) |
1961 | cfqq->next_rq = rq; | |
b4878f24 | 1962 | cfq_remove_request(next); |
155fead9 | 1963 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
4a0b75c7 SL |
1964 | |
1965 | cfqq = RQ_CFQQ(next); | |
1966 | /* | |
1967 | * all requests of this queue are merged to other queues, delete it | |
1968 | * from the service tree. If it's the active_queue, | |
1969 | * cfq_dispatch_requests() will choose to expire it or do idle | |
1970 | */ | |
1971 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
1972 | cfqq != cfqd->active_queue) | |
1973 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
1974 | } |
1975 | ||
165125e1 | 1976 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
1977 | struct bio *bio) |
1978 | { | |
1979 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 1980 | struct cfq_io_cq *cic; |
da775265 | 1981 | struct cfq_queue *cfqq; |
da775265 JA |
1982 | |
1983 | /* | |
ec8acb69 | 1984 | * Disallow merge of a sync bio into an async request. |
da775265 | 1985 | */ |
91fac317 | 1986 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 1987 | return false; |
da775265 JA |
1988 | |
1989 | /* | |
f1a4f4d3 | 1990 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 1991 | * merge only if rq is queued there. |
f1a4f4d3 | 1992 | */ |
07c2bd37 TH |
1993 | cic = cfq_cic_lookup(cfqd, current->io_context); |
1994 | if (!cic) | |
1995 | return false; | |
719d3402 | 1996 | |
91fac317 | 1997 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 1998 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
1999 | } |
2000 | ||
812df48d DS |
2001 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2002 | { | |
2003 | del_timer(&cfqd->idle_slice_timer); | |
155fead9 | 2004 | cfqg_stats_update_idle_time(cfqq->cfqg); |
812df48d DS |
2005 | } |
2006 | ||
febffd61 JA |
2007 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
2008 | struct cfq_queue *cfqq) | |
22e2c507 JA |
2009 | { |
2010 | if (cfqq) { | |
b1ffe737 DS |
2011 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", |
2012 | cfqd->serving_prio, cfqd->serving_type); | |
155fead9 | 2013 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
62a37f6b JT |
2014 | cfqq->slice_start = 0; |
2015 | cfqq->dispatch_start = jiffies; | |
2016 | cfqq->allocated_slice = 0; | |
2017 | cfqq->slice_end = 0; | |
2018 | cfqq->slice_dispatch = 0; | |
2019 | cfqq->nr_sectors = 0; | |
2020 | ||
2021 | cfq_clear_cfqq_wait_request(cfqq); | |
2022 | cfq_clear_cfqq_must_dispatch(cfqq); | |
2023 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
2024 | cfq_clear_cfqq_fifo_expire(cfqq); | |
2025 | cfq_mark_cfqq_slice_new(cfqq); | |
2026 | ||
2027 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
2028 | } |
2029 | ||
2030 | cfqd->active_queue = cfqq; | |
2031 | } | |
2032 | ||
7b14e3b5 JA |
2033 | /* |
2034 | * current cfqq expired its slice (or was too idle), select new one | |
2035 | */ | |
2036 | static void | |
2037 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 2038 | bool timed_out) |
7b14e3b5 | 2039 | { |
7b679138 JA |
2040 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
2041 | ||
7b14e3b5 | 2042 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 2043 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 2044 | |
7b14e3b5 | 2045 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 2046 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 2047 | |
ae54abed SL |
2048 | /* |
2049 | * If this cfqq is shared between multiple processes, check to | |
2050 | * make sure that those processes are still issuing I/Os within | |
2051 | * the mean seek distance. If not, it may be time to break the | |
2052 | * queues apart again. | |
2053 | */ | |
2054 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
2055 | cfq_mark_cfqq_split_coop(cfqq); | |
2056 | ||
7b14e3b5 | 2057 | /* |
6084cdda | 2058 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 2059 | */ |
c553f8e3 SL |
2060 | if (timed_out) { |
2061 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 2062 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
2063 | else |
2064 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
2065 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
2066 | } | |
7b14e3b5 | 2067 | |
e5ff082e | 2068 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 2069 | |
f04a6424 VG |
2070 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
2071 | cfq_del_cfqq_rr(cfqd, cfqq); | |
2072 | ||
edd75ffd | 2073 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
2074 | |
2075 | if (cfqq == cfqd->active_queue) | |
2076 | cfqd->active_queue = NULL; | |
2077 | ||
2078 | if (cfqd->active_cic) { | |
11a3122f | 2079 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
2080 | cfqd->active_cic = NULL; |
2081 | } | |
7b14e3b5 JA |
2082 | } |
2083 | ||
e5ff082e | 2084 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
2085 | { |
2086 | struct cfq_queue *cfqq = cfqd->active_queue; | |
2087 | ||
2088 | if (cfqq) | |
e5ff082e | 2089 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
2090 | } |
2091 | ||
498d3aa2 JA |
2092 | /* |
2093 | * Get next queue for service. Unless we have a queue preemption, | |
2094 | * we'll simply select the first cfqq in the service tree. | |
2095 | */ | |
6d048f53 | 2096 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 2097 | { |
c0324a02 | 2098 | struct cfq_rb_root *service_tree = |
cdb16e8f | 2099 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
65b32a57 | 2100 | cfqd->serving_type); |
d9e7620e | 2101 | |
f04a6424 VG |
2102 | if (!cfqd->rq_queued) |
2103 | return NULL; | |
2104 | ||
1fa8f6d6 VG |
2105 | /* There is nothing to dispatch */ |
2106 | if (!service_tree) | |
2107 | return NULL; | |
c0324a02 CZ |
2108 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
2109 | return NULL; | |
2110 | return cfq_rb_first(service_tree); | |
6d048f53 JA |
2111 | } |
2112 | ||
f04a6424 VG |
2113 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
2114 | { | |
25fb5169 | 2115 | struct cfq_group *cfqg; |
f04a6424 VG |
2116 | struct cfq_queue *cfqq; |
2117 | int i, j; | |
2118 | struct cfq_rb_root *st; | |
2119 | ||
2120 | if (!cfqd->rq_queued) | |
2121 | return NULL; | |
2122 | ||
25fb5169 VG |
2123 | cfqg = cfq_get_next_cfqg(cfqd); |
2124 | if (!cfqg) | |
2125 | return NULL; | |
2126 | ||
f04a6424 VG |
2127 | for_each_cfqg_st(cfqg, i, j, st) |
2128 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
2129 | return cfqq; | |
2130 | return NULL; | |
2131 | } | |
2132 | ||
498d3aa2 JA |
2133 | /* |
2134 | * Get and set a new active queue for service. | |
2135 | */ | |
a36e71f9 JA |
2136 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
2137 | struct cfq_queue *cfqq) | |
6d048f53 | 2138 | { |
e00ef799 | 2139 | if (!cfqq) |
a36e71f9 | 2140 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 2141 | |
22e2c507 | 2142 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 2143 | return cfqq; |
22e2c507 JA |
2144 | } |
2145 | ||
d9e7620e JA |
2146 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
2147 | struct request *rq) | |
2148 | { | |
83096ebf TH |
2149 | if (blk_rq_pos(rq) >= cfqd->last_position) |
2150 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 2151 | else |
83096ebf | 2152 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
2153 | } |
2154 | ||
b2c18e1e | 2155 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 2156 | struct request *rq) |
6d048f53 | 2157 | { |
e9ce335d | 2158 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
2159 | } |
2160 | ||
a36e71f9 JA |
2161 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
2162 | struct cfq_queue *cur_cfqq) | |
2163 | { | |
f2d1f0ae | 2164 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
2165 | struct rb_node *parent, *node; |
2166 | struct cfq_queue *__cfqq; | |
2167 | sector_t sector = cfqd->last_position; | |
2168 | ||
2169 | if (RB_EMPTY_ROOT(root)) | |
2170 | return NULL; | |
2171 | ||
2172 | /* | |
2173 | * First, if we find a request starting at the end of the last | |
2174 | * request, choose it. | |
2175 | */ | |
f2d1f0ae | 2176 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
2177 | if (__cfqq) |
2178 | return __cfqq; | |
2179 | ||
2180 | /* | |
2181 | * If the exact sector wasn't found, the parent of the NULL leaf | |
2182 | * will contain the closest sector. | |
2183 | */ | |
2184 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 2185 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2186 | return __cfqq; |
2187 | ||
2e46e8b2 | 2188 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
2189 | node = rb_next(&__cfqq->p_node); |
2190 | else | |
2191 | node = rb_prev(&__cfqq->p_node); | |
2192 | if (!node) | |
2193 | return NULL; | |
2194 | ||
2195 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 2196 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2197 | return __cfqq; |
2198 | ||
2199 | return NULL; | |
2200 | } | |
2201 | ||
2202 | /* | |
2203 | * cfqd - obvious | |
2204 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
2205 | * closely cooperating with itself. | |
2206 | * | |
2207 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
2208 | * one request, and that cfqd->last_position reflects a position on the disk | |
2209 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
2210 | * assumption. | |
2211 | */ | |
2212 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 2213 | struct cfq_queue *cur_cfqq) |
6d048f53 | 2214 | { |
a36e71f9 JA |
2215 | struct cfq_queue *cfqq; |
2216 | ||
39c01b21 DS |
2217 | if (cfq_class_idle(cur_cfqq)) |
2218 | return NULL; | |
e6c5bc73 JM |
2219 | if (!cfq_cfqq_sync(cur_cfqq)) |
2220 | return NULL; | |
2221 | if (CFQQ_SEEKY(cur_cfqq)) | |
2222 | return NULL; | |
2223 | ||
b9d8f4c7 GJ |
2224 | /* |
2225 | * Don't search priority tree if it's the only queue in the group. | |
2226 | */ | |
2227 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
2228 | return NULL; | |
2229 | ||
6d048f53 | 2230 | /* |
d9e7620e JA |
2231 | * We should notice if some of the queues are cooperating, eg |
2232 | * working closely on the same area of the disk. In that case, | |
2233 | * we can group them together and don't waste time idling. | |
6d048f53 | 2234 | */ |
a36e71f9 JA |
2235 | cfqq = cfqq_close(cfqd, cur_cfqq); |
2236 | if (!cfqq) | |
2237 | return NULL; | |
2238 | ||
8682e1f1 VG |
2239 | /* If new queue belongs to different cfq_group, don't choose it */ |
2240 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
2241 | return NULL; | |
2242 | ||
df5fe3e8 JM |
2243 | /* |
2244 | * It only makes sense to merge sync queues. | |
2245 | */ | |
2246 | if (!cfq_cfqq_sync(cfqq)) | |
2247 | return NULL; | |
e6c5bc73 JM |
2248 | if (CFQQ_SEEKY(cfqq)) |
2249 | return NULL; | |
df5fe3e8 | 2250 | |
c0324a02 CZ |
2251 | /* |
2252 | * Do not merge queues of different priority classes | |
2253 | */ | |
2254 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
2255 | return NULL; | |
2256 | ||
a36e71f9 | 2257 | return cfqq; |
6d048f53 JA |
2258 | } |
2259 | ||
a6d44e98 CZ |
2260 | /* |
2261 | * Determine whether we should enforce idle window for this queue. | |
2262 | */ | |
2263 | ||
2264 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2265 | { | |
2266 | enum wl_prio_t prio = cfqq_prio(cfqq); | |
718eee05 | 2267 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
a6d44e98 | 2268 | |
f04a6424 VG |
2269 | BUG_ON(!service_tree); |
2270 | BUG_ON(!service_tree->count); | |
2271 | ||
b6508c16 VG |
2272 | if (!cfqd->cfq_slice_idle) |
2273 | return false; | |
2274 | ||
a6d44e98 CZ |
2275 | /* We never do for idle class queues. */ |
2276 | if (prio == IDLE_WORKLOAD) | |
2277 | return false; | |
2278 | ||
2279 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
2280 | if (cfq_cfqq_idle_window(cfqq) && |
2281 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
2282 | return true; |
2283 | ||
2284 | /* | |
2285 | * Otherwise, we do only if they are the last ones | |
2286 | * in their service tree. | |
2287 | */ | |
f5f2b6ce SL |
2288 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq) && |
2289 | !cfq_io_thinktime_big(cfqd, &service_tree->ttime, false)) | |
c1e44756 | 2290 | return true; |
b1ffe737 DS |
2291 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", |
2292 | service_tree->count); | |
c1e44756 | 2293 | return false; |
a6d44e98 CZ |
2294 | } |
2295 | ||
6d048f53 | 2296 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2297 | { |
1792669c | 2298 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 2299 | struct cfq_io_cq *cic; |
80bdf0c7 | 2300 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 2301 | |
a68bbddb | 2302 | /* |
f7d7b7a7 JA |
2303 | * SSD device without seek penalty, disable idling. But only do so |
2304 | * for devices that support queuing, otherwise we still have a problem | |
2305 | * with sync vs async workloads. | |
a68bbddb | 2306 | */ |
f7d7b7a7 | 2307 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
2308 | return; |
2309 | ||
dd67d051 | 2310 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2311 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2312 | |
2313 | /* | |
2314 | * idle is disabled, either manually or by past process history | |
2315 | */ | |
80bdf0c7 VG |
2316 | if (!cfq_should_idle(cfqd, cfqq)) { |
2317 | /* no queue idling. Check for group idling */ | |
2318 | if (cfqd->cfq_group_idle) | |
2319 | group_idle = cfqd->cfq_group_idle; | |
2320 | else | |
2321 | return; | |
2322 | } | |
6d048f53 | 2323 | |
7b679138 | 2324 | /* |
8e550632 | 2325 | * still active requests from this queue, don't idle |
7b679138 | 2326 | */ |
8e550632 | 2327 | if (cfqq->dispatched) |
7b679138 JA |
2328 | return; |
2329 | ||
22e2c507 JA |
2330 | /* |
2331 | * task has exited, don't wait | |
2332 | */ | |
206dc69b | 2333 | cic = cfqd->active_cic; |
f6e8d01b | 2334 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
2335 | return; |
2336 | ||
355b659c CZ |
2337 | /* |
2338 | * If our average think time is larger than the remaining time | |
2339 | * slice, then don't idle. This avoids overrunning the allotted | |
2340 | * time slice. | |
2341 | */ | |
383cd721 SL |
2342 | if (sample_valid(cic->ttime.ttime_samples) && |
2343 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 2344 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 2345 | cic->ttime.ttime_mean); |
355b659c | 2346 | return; |
b1ffe737 | 2347 | } |
355b659c | 2348 | |
80bdf0c7 VG |
2349 | /* There are other queues in the group, don't do group idle */ |
2350 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
2351 | return; | |
2352 | ||
3b18152c | 2353 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2354 | |
80bdf0c7 VG |
2355 | if (group_idle) |
2356 | sl = cfqd->cfq_group_idle; | |
2357 | else | |
2358 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2359 | |
7b14e3b5 | 2360 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
155fead9 | 2361 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
80bdf0c7 VG |
2362 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
2363 | group_idle ? 1 : 0); | |
1da177e4 LT |
2364 | } |
2365 | ||
498d3aa2 JA |
2366 | /* |
2367 | * Move request from internal lists to the request queue dispatch list. | |
2368 | */ | |
165125e1 | 2369 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2370 | { |
3ed9a296 | 2371 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2372 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2373 | |
7b679138 JA |
2374 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2375 | ||
06d21886 | 2376 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2377 | cfq_remove_request(rq); |
6d048f53 | 2378 | cfqq->dispatched++; |
80bdf0c7 | 2379 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2380 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2381 | |
53c583d2 | 2382 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2383 | cfqq->nr_sectors += blk_rq_sectors(rq); |
155fead9 | 2384 | cfqg_stats_update_dispatch(cfqq->cfqg, blk_rq_bytes(rq), rq->cmd_flags); |
1da177e4 LT |
2385 | } |
2386 | ||
2387 | /* | |
2388 | * return expired entry, or NULL to just start from scratch in rbtree | |
2389 | */ | |
febffd61 | 2390 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 2391 | { |
30996f40 | 2392 | struct request *rq = NULL; |
1da177e4 | 2393 | |
3b18152c | 2394 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 2395 | return NULL; |
cb887411 JA |
2396 | |
2397 | cfq_mark_cfqq_fifo_expire(cfqq); | |
2398 | ||
89850f7e JA |
2399 | if (list_empty(&cfqq->fifo)) |
2400 | return NULL; | |
1da177e4 | 2401 | |
89850f7e | 2402 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 2403 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 2404 | rq = NULL; |
1da177e4 | 2405 | |
30996f40 | 2406 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 2407 | return rq; |
1da177e4 LT |
2408 | } |
2409 | ||
22e2c507 JA |
2410 | static inline int |
2411 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2412 | { | |
2413 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2414 | |
22e2c507 | 2415 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2416 | |
b9f8ce05 | 2417 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
2418 | } |
2419 | ||
df5fe3e8 JM |
2420 | /* |
2421 | * Must be called with the queue_lock held. | |
2422 | */ | |
2423 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2424 | { | |
2425 | int process_refs, io_refs; | |
2426 | ||
2427 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 2428 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
2429 | BUG_ON(process_refs < 0); |
2430 | return process_refs; | |
2431 | } | |
2432 | ||
2433 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2434 | { | |
e6c5bc73 | 2435 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2436 | struct cfq_queue *__cfqq; |
2437 | ||
c10b61f0 JM |
2438 | /* |
2439 | * If there are no process references on the new_cfqq, then it is | |
2440 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2441 | * chain may have dropped their last reference (not just their | |
2442 | * last process reference). | |
2443 | */ | |
2444 | if (!cfqq_process_refs(new_cfqq)) | |
2445 | return; | |
2446 | ||
df5fe3e8 JM |
2447 | /* Avoid a circular list and skip interim queue merges */ |
2448 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2449 | if (__cfqq == cfqq) | |
2450 | return; | |
2451 | new_cfqq = __cfqq; | |
2452 | } | |
2453 | ||
2454 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2455 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2456 | /* |
2457 | * If the process for the cfqq has gone away, there is no | |
2458 | * sense in merging the queues. | |
2459 | */ | |
c10b61f0 | 2460 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2461 | return; |
2462 | ||
e6c5bc73 JM |
2463 | /* |
2464 | * Merge in the direction of the lesser amount of work. | |
2465 | */ | |
e6c5bc73 JM |
2466 | if (new_process_refs >= process_refs) { |
2467 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2468 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2469 | } else { |
2470 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2471 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2472 | } |
df5fe3e8 JM |
2473 | } |
2474 | ||
cdb16e8f | 2475 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
65b32a57 | 2476 | struct cfq_group *cfqg, enum wl_prio_t prio) |
718eee05 CZ |
2477 | { |
2478 | struct cfq_queue *queue; | |
2479 | int i; | |
2480 | bool key_valid = false; | |
2481 | unsigned long lowest_key = 0; | |
2482 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2483 | ||
65b32a57 VG |
2484 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2485 | /* select the one with lowest rb_key */ | |
2486 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); | |
718eee05 CZ |
2487 | if (queue && |
2488 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2489 | lowest_key = queue->rb_key; | |
2490 | cur_best = i; | |
2491 | key_valid = true; | |
2492 | } | |
2493 | } | |
2494 | ||
2495 | return cur_best; | |
2496 | } | |
2497 | ||
cdb16e8f | 2498 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
718eee05 | 2499 | { |
718eee05 CZ |
2500 | unsigned slice; |
2501 | unsigned count; | |
cdb16e8f | 2502 | struct cfq_rb_root *st; |
58ff82f3 | 2503 | unsigned group_slice; |
e4ea0c16 | 2504 | enum wl_prio_t original_prio = cfqd->serving_prio; |
1fa8f6d6 | 2505 | |
718eee05 | 2506 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2507 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
718eee05 | 2508 | cfqd->serving_prio = RT_WORKLOAD; |
58ff82f3 | 2509 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
718eee05 CZ |
2510 | cfqd->serving_prio = BE_WORKLOAD; |
2511 | else { | |
2512 | cfqd->serving_prio = IDLE_WORKLOAD; | |
2513 | cfqd->workload_expires = jiffies + 1; | |
2514 | return; | |
2515 | } | |
2516 | ||
e4ea0c16 SL |
2517 | if (original_prio != cfqd->serving_prio) |
2518 | goto new_workload; | |
2519 | ||
718eee05 CZ |
2520 | /* |
2521 | * For RT and BE, we have to choose also the type | |
2522 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2523 | * expiration time | |
2524 | */ | |
65b32a57 | 2525 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
cdb16e8f | 2526 | count = st->count; |
718eee05 CZ |
2527 | |
2528 | /* | |
65b32a57 | 2529 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2530 | */ |
65b32a57 | 2531 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2532 | return; |
2533 | ||
e4ea0c16 | 2534 | new_workload: |
718eee05 CZ |
2535 | /* otherwise select new workload type */ |
2536 | cfqd->serving_type = | |
65b32a57 VG |
2537 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); |
2538 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); | |
cdb16e8f | 2539 | count = st->count; |
718eee05 CZ |
2540 | |
2541 | /* | |
2542 | * the workload slice is computed as a fraction of target latency | |
2543 | * proportional to the number of queues in that workload, over | |
2544 | * all the queues in the same priority class | |
2545 | */ | |
58ff82f3 VG |
2546 | group_slice = cfq_group_slice(cfqd, cfqg); |
2547 | ||
2548 | slice = group_slice * count / | |
2549 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], | |
2550 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); | |
718eee05 | 2551 | |
f26bd1f0 VG |
2552 | if (cfqd->serving_type == ASYNC_WORKLOAD) { |
2553 | unsigned int tmp; | |
2554 | ||
2555 | /* | |
2556 | * Async queues are currently system wide. Just taking | |
2557 | * proportion of queues with-in same group will lead to higher | |
2558 | * async ratio system wide as generally root group is going | |
2559 | * to have higher weight. A more accurate thing would be to | |
2560 | * calculate system wide asnc/sync ratio. | |
2561 | */ | |
2562 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); | |
2563 | tmp = tmp/cfqd->busy_queues; | |
2564 | slice = min_t(unsigned, slice, tmp); | |
2565 | ||
718eee05 CZ |
2566 | /* async workload slice is scaled down according to |
2567 | * the sync/async slice ratio. */ | |
2568 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2569 | } else |
718eee05 CZ |
2570 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2571 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2572 | ||
2573 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2574 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
2575 | cfqd->workload_expires = jiffies + slice; |
2576 | } | |
2577 | ||
1fa8f6d6 VG |
2578 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2579 | { | |
2580 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2581 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2582 | |
2583 | if (RB_EMPTY_ROOT(&st->rb)) | |
2584 | return NULL; | |
25bc6b07 | 2585 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
2586 | update_min_vdisktime(st); |
2587 | return cfqg; | |
1fa8f6d6 VG |
2588 | } |
2589 | ||
cdb16e8f VG |
2590 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2591 | { | |
1fa8f6d6 VG |
2592 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2593 | ||
2594 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2595 | |
2596 | /* Restore the workload type data */ | |
2597 | if (cfqg->saved_workload_slice) { | |
2598 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; | |
2599 | cfqd->serving_type = cfqg->saved_workload; | |
2600 | cfqd->serving_prio = cfqg->saved_serving_prio; | |
66ae2919 GJ |
2601 | } else |
2602 | cfqd->workload_expires = jiffies - 1; | |
2603 | ||
1fa8f6d6 | 2604 | choose_service_tree(cfqd, cfqg); |
cdb16e8f VG |
2605 | } |
2606 | ||
22e2c507 | 2607 | /* |
498d3aa2 JA |
2608 | * Select a queue for service. If we have a current active queue, |
2609 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2610 | */ |
1b5ed5e1 | 2611 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2612 | { |
a36e71f9 | 2613 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2614 | |
22e2c507 JA |
2615 | cfqq = cfqd->active_queue; |
2616 | if (!cfqq) | |
2617 | goto new_queue; | |
1da177e4 | 2618 | |
f04a6424 VG |
2619 | if (!cfqd->rq_queued) |
2620 | return NULL; | |
c244bb50 VG |
2621 | |
2622 | /* | |
2623 | * We were waiting for group to get backlogged. Expire the queue | |
2624 | */ | |
2625 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2626 | goto expire; | |
2627 | ||
22e2c507 | 2628 | /* |
6d048f53 | 2629 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2630 | */ |
7667aa06 VG |
2631 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2632 | /* | |
2633 | * If slice had not expired at the completion of last request | |
2634 | * we might not have turned on wait_busy flag. Don't expire | |
2635 | * the queue yet. Allow the group to get backlogged. | |
2636 | * | |
2637 | * The very fact that we have used the slice, that means we | |
2638 | * have been idling all along on this queue and it should be | |
2639 | * ok to wait for this request to complete. | |
2640 | */ | |
82bbbf28 VG |
2641 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2642 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2643 | cfqq = NULL; | |
7667aa06 | 2644 | goto keep_queue; |
82bbbf28 | 2645 | } else |
80bdf0c7 | 2646 | goto check_group_idle; |
7667aa06 | 2647 | } |
1da177e4 | 2648 | |
22e2c507 | 2649 | /* |
6d048f53 JA |
2650 | * The active queue has requests and isn't expired, allow it to |
2651 | * dispatch. | |
22e2c507 | 2652 | */ |
dd67d051 | 2653 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2654 | goto keep_queue; |
6d048f53 | 2655 | |
a36e71f9 JA |
2656 | /* |
2657 | * If another queue has a request waiting within our mean seek | |
2658 | * distance, let it run. The expire code will check for close | |
2659 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2660 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2661 | */ |
b3b6d040 | 2662 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2663 | if (new_cfqq) { |
2664 | if (!cfqq->new_cfqq) | |
2665 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2666 | goto expire; |
df5fe3e8 | 2667 | } |
a36e71f9 | 2668 | |
6d048f53 JA |
2669 | /* |
2670 | * No requests pending. If the active queue still has requests in | |
2671 | * flight or is idling for a new request, allow either of these | |
2672 | * conditions to happen (or time out) before selecting a new queue. | |
2673 | */ | |
80bdf0c7 VG |
2674 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2675 | cfqq = NULL; | |
2676 | goto keep_queue; | |
2677 | } | |
2678 | ||
8e1ac665 SL |
2679 | /* |
2680 | * This is a deep seek queue, but the device is much faster than | |
2681 | * the queue can deliver, don't idle | |
2682 | **/ | |
2683 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
2684 | (cfq_cfqq_slice_new(cfqq) || | |
2685 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
2686 | cfq_clear_cfqq_deep(cfqq); | |
2687 | cfq_clear_cfqq_idle_window(cfqq); | |
2688 | } | |
2689 | ||
80bdf0c7 VG |
2690 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2691 | cfqq = NULL; | |
2692 | goto keep_queue; | |
2693 | } | |
2694 | ||
2695 | /* | |
2696 | * If group idle is enabled and there are requests dispatched from | |
2697 | * this group, wait for requests to complete. | |
2698 | */ | |
2699 | check_group_idle: | |
7700fc4f SL |
2700 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
2701 | cfqq->cfqg->dispatched && | |
2702 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
2703 | cfqq = NULL; |
2704 | goto keep_queue; | |
22e2c507 JA |
2705 | } |
2706 | ||
3b18152c | 2707 | expire: |
e5ff082e | 2708 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2709 | new_queue: |
718eee05 CZ |
2710 | /* |
2711 | * Current queue expired. Check if we have to switch to a new | |
2712 | * service tree | |
2713 | */ | |
2714 | if (!new_cfqq) | |
cdb16e8f | 2715 | cfq_choose_cfqg(cfqd); |
718eee05 | 2716 | |
a36e71f9 | 2717 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2718 | keep_queue: |
3b18152c | 2719 | return cfqq; |
22e2c507 JA |
2720 | } |
2721 | ||
febffd61 | 2722 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2723 | { |
2724 | int dispatched = 0; | |
2725 | ||
2726 | while (cfqq->next_rq) { | |
2727 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2728 | dispatched++; | |
2729 | } | |
2730 | ||
2731 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2732 | |
2733 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 2734 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
2735 | return dispatched; |
2736 | } | |
2737 | ||
498d3aa2 JA |
2738 | /* |
2739 | * Drain our current requests. Used for barriers and when switching | |
2740 | * io schedulers on-the-fly. | |
2741 | */ | |
d9e7620e | 2742 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2743 | { |
0871714e | 2744 | struct cfq_queue *cfqq; |
d9e7620e | 2745 | int dispatched = 0; |
cdb16e8f | 2746 | |
3440c49f | 2747 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 2748 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
2749 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2750 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 2751 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 2752 | } |
1b5ed5e1 | 2753 | |
1b5ed5e1 TH |
2754 | BUG_ON(cfqd->busy_queues); |
2755 | ||
6923715a | 2756 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2757 | return dispatched; |
2758 | } | |
2759 | ||
abc3c744 SL |
2760 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2761 | struct cfq_queue *cfqq) | |
2762 | { | |
2763 | /* the queue hasn't finished any request, can't estimate */ | |
2764 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 2765 | return true; |
abc3c744 SL |
2766 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
2767 | cfqq->slice_end)) | |
c1e44756 | 2768 | return true; |
abc3c744 | 2769 | |
c1e44756 | 2770 | return false; |
abc3c744 SL |
2771 | } |
2772 | ||
0b182d61 | 2773 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2774 | { |
2f5cb738 | 2775 | unsigned int max_dispatch; |
22e2c507 | 2776 | |
5ad531db JA |
2777 | /* |
2778 | * Drain async requests before we start sync IO | |
2779 | */ | |
53c583d2 | 2780 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 2781 | return false; |
5ad531db | 2782 | |
2f5cb738 JA |
2783 | /* |
2784 | * If this is an async queue and we have sync IO in flight, let it wait | |
2785 | */ | |
53c583d2 | 2786 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 2787 | return false; |
2f5cb738 | 2788 | |
abc3c744 | 2789 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
2790 | if (cfq_class_idle(cfqq)) |
2791 | max_dispatch = 1; | |
b4878f24 | 2792 | |
2f5cb738 JA |
2793 | /* |
2794 | * Does this cfqq already have too much IO in flight? | |
2795 | */ | |
2796 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 2797 | bool promote_sync = false; |
2f5cb738 JA |
2798 | /* |
2799 | * idle queue must always only have a single IO in flight | |
2800 | */ | |
3ed9a296 | 2801 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2802 | return false; |
3ed9a296 | 2803 | |
ef8a41df | 2804 | /* |
c4ade94f LS |
2805 | * If there is only one sync queue |
2806 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
2807 | * queue no dispatch limit. The reason is a sync queue can |
2808 | * preempt async queue, limiting the sync queue doesn't make | |
2809 | * sense. This is useful for aiostress test. | |
2810 | */ | |
c4ade94f LS |
2811 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
2812 | promote_sync = true; | |
ef8a41df | 2813 | |
2f5cb738 JA |
2814 | /* |
2815 | * We have other queues, don't allow more IO from this one | |
2816 | */ | |
ef8a41df SL |
2817 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
2818 | !promote_sync) | |
0b182d61 | 2819 | return false; |
9ede209e | 2820 | |
365722bb | 2821 | /* |
474b18cc | 2822 | * Sole queue user, no limit |
365722bb | 2823 | */ |
ef8a41df | 2824 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
2825 | max_dispatch = -1; |
2826 | else | |
2827 | /* | |
2828 | * Normally we start throttling cfqq when cfq_quantum/2 | |
2829 | * requests have been dispatched. But we can drive | |
2830 | * deeper queue depths at the beginning of slice | |
2831 | * subjected to upper limit of cfq_quantum. | |
2832 | * */ | |
2833 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
2834 | } |
2835 | ||
2836 | /* | |
2837 | * Async queues must wait a bit before being allowed dispatch. | |
2838 | * We also ramp up the dispatch depth gradually for async IO, | |
2839 | * based on the last sync IO we serviced | |
2840 | */ | |
963b72fc | 2841 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 2842 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 2843 | unsigned int depth; |
365722bb | 2844 | |
61f0c1dc | 2845 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2846 | if (!depth && !cfqq->dispatched) |
2847 | depth = 1; | |
8e296755 JA |
2848 | if (depth < max_dispatch) |
2849 | max_dispatch = depth; | |
2f5cb738 | 2850 | } |
3ed9a296 | 2851 | |
0b182d61 JA |
2852 | /* |
2853 | * If we're below the current max, allow a dispatch | |
2854 | */ | |
2855 | return cfqq->dispatched < max_dispatch; | |
2856 | } | |
2857 | ||
2858 | /* | |
2859 | * Dispatch a request from cfqq, moving them to the request queue | |
2860 | * dispatch list. | |
2861 | */ | |
2862 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2863 | { | |
2864 | struct request *rq; | |
2865 | ||
2866 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
2867 | ||
2868 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
2869 | return false; | |
2870 | ||
2871 | /* | |
2872 | * follow expired path, else get first next available | |
2873 | */ | |
2874 | rq = cfq_check_fifo(cfqq); | |
2875 | if (!rq) | |
2876 | rq = cfqq->next_rq; | |
2877 | ||
2878 | /* | |
2879 | * insert request into driver dispatch list | |
2880 | */ | |
2881 | cfq_dispatch_insert(cfqd->queue, rq); | |
2882 | ||
2883 | if (!cfqd->active_cic) { | |
c5869807 | 2884 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 2885 | |
c5869807 | 2886 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
2887 | cfqd->active_cic = cic; |
2888 | } | |
2889 | ||
2890 | return true; | |
2891 | } | |
2892 | ||
2893 | /* | |
2894 | * Find the cfqq that we need to service and move a request from that to the | |
2895 | * dispatch list | |
2896 | */ | |
2897 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
2898 | { | |
2899 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2900 | struct cfq_queue *cfqq; | |
2901 | ||
2902 | if (!cfqd->busy_queues) | |
2903 | return 0; | |
2904 | ||
2905 | if (unlikely(force)) | |
2906 | return cfq_forced_dispatch(cfqd); | |
2907 | ||
2908 | cfqq = cfq_select_queue(cfqd); | |
2909 | if (!cfqq) | |
8e296755 JA |
2910 | return 0; |
2911 | ||
2f5cb738 | 2912 | /* |
0b182d61 | 2913 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 2914 | */ |
0b182d61 JA |
2915 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2916 | return 0; | |
2917 | ||
2f5cb738 | 2918 | cfqq->slice_dispatch++; |
b029195d | 2919 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 2920 | |
2f5cb738 JA |
2921 | /* |
2922 | * expire an async queue immediately if it has used up its slice. idle | |
2923 | * queue always expire after 1 dispatch round. | |
2924 | */ | |
2925 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
2926 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
2927 | cfq_class_idle(cfqq))) { | |
2928 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 2929 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
2930 | } |
2931 | ||
b217a903 | 2932 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 2933 | return 1; |
1da177e4 LT |
2934 | } |
2935 | ||
1da177e4 | 2936 | /* |
5e705374 JA |
2937 | * task holds one reference to the queue, dropped when task exits. each rq |
2938 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 2939 | * |
b1c35769 | 2940 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
2941 | * queue lock must be held here. |
2942 | */ | |
2943 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
2944 | { | |
22e2c507 | 2945 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 2946 | struct cfq_group *cfqg; |
22e2c507 | 2947 | |
30d7b944 | 2948 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 2949 | |
30d7b944 SL |
2950 | cfqq->ref--; |
2951 | if (cfqq->ref) | |
1da177e4 LT |
2952 | return; |
2953 | ||
7b679138 | 2954 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 2955 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 2956 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 2957 | cfqg = cfqq->cfqg; |
1da177e4 | 2958 | |
28f95cbc | 2959 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 2960 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2961 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2962 | } |
22e2c507 | 2963 | |
f04a6424 | 2964 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 2965 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 2966 | cfqg_put(cfqg); |
1da177e4 LT |
2967 | } |
2968 | ||
d02a2c07 | 2969 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 2970 | { |
df5fe3e8 JM |
2971 | struct cfq_queue *__cfqq, *next; |
2972 | ||
df5fe3e8 JM |
2973 | /* |
2974 | * If this queue was scheduled to merge with another queue, be | |
2975 | * sure to drop the reference taken on that queue (and others in | |
2976 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
2977 | */ | |
2978 | __cfqq = cfqq->new_cfqq; | |
2979 | while (__cfqq) { | |
2980 | if (__cfqq == cfqq) { | |
2981 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
2982 | break; | |
2983 | } | |
2984 | next = __cfqq->new_cfqq; | |
2985 | cfq_put_queue(__cfqq); | |
2986 | __cfqq = next; | |
2987 | } | |
d02a2c07 SL |
2988 | } |
2989 | ||
2990 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2991 | { | |
2992 | if (unlikely(cfqq == cfqd->active_queue)) { | |
2993 | __cfq_slice_expired(cfqd, cfqq, 0); | |
2994 | cfq_schedule_dispatch(cfqd); | |
2995 | } | |
2996 | ||
2997 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 2998 | |
89850f7e JA |
2999 | cfq_put_queue(cfqq); |
3000 | } | |
22e2c507 | 3001 | |
9b84cacd TH |
3002 | static void cfq_init_icq(struct io_cq *icq) |
3003 | { | |
3004 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
3005 | ||
3006 | cic->ttime.last_end_request = jiffies; | |
3007 | } | |
3008 | ||
c5869807 | 3009 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 3010 | { |
c5869807 | 3011 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 3012 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 3013 | |
ff6657c6 JA |
3014 | if (cic->cfqq[BLK_RW_ASYNC]) { |
3015 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
3016 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
3017 | } |
3018 | ||
ff6657c6 JA |
3019 | if (cic->cfqq[BLK_RW_SYNC]) { |
3020 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
3021 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 3022 | } |
89850f7e JA |
3023 | } |
3024 | ||
abede6da | 3025 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
3026 | { |
3027 | struct task_struct *tsk = current; | |
3028 | int ioprio_class; | |
3029 | ||
3b18152c | 3030 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
3031 | return; |
3032 | ||
598971bf | 3033 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 3034 | switch (ioprio_class) { |
fe094d98 JA |
3035 | default: |
3036 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
3037 | case IOPRIO_CLASS_NONE: | |
3038 | /* | |
6d63c275 | 3039 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
3040 | */ |
3041 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 3042 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
3043 | break; |
3044 | case IOPRIO_CLASS_RT: | |
598971bf | 3045 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3046 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
3047 | break; | |
3048 | case IOPRIO_CLASS_BE: | |
598971bf | 3049 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3050 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3051 | break; | |
3052 | case IOPRIO_CLASS_IDLE: | |
3053 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
3054 | cfqq->ioprio = 7; | |
3055 | cfq_clear_cfqq_idle_window(cfqq); | |
3056 | break; | |
22e2c507 JA |
3057 | } |
3058 | ||
3059 | /* | |
3060 | * keep track of original prio settings in case we have to temporarily | |
3061 | * elevate the priority of this queue | |
3062 | */ | |
3063 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 3064 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
3065 | } |
3066 | ||
598971bf | 3067 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 3068 | { |
598971bf | 3069 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 3070 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 3071 | struct cfq_queue *cfqq; |
35e6077c | 3072 | |
598971bf TH |
3073 | /* |
3074 | * Check whether ioprio has changed. The condition may trigger | |
3075 | * spuriously on a newly created cic but there's no harm. | |
3076 | */ | |
3077 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
3078 | return; |
3079 | ||
ff6657c6 | 3080 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
3081 | if (cfqq) { |
3082 | struct cfq_queue *new_cfqq; | |
abede6da TH |
3083 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio, |
3084 | GFP_ATOMIC); | |
caaa5f9f | 3085 | if (new_cfqq) { |
ff6657c6 | 3086 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
3087 | cfq_put_queue(cfqq); |
3088 | } | |
22e2c507 | 3089 | } |
caaa5f9f | 3090 | |
ff6657c6 | 3091 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
3092 | if (cfqq) |
3093 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
3094 | |
3095 | cic->ioprio = ioprio; | |
22e2c507 JA |
3096 | } |
3097 | ||
d5036d77 | 3098 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 3099 | pid_t pid, bool is_sync) |
d5036d77 JA |
3100 | { |
3101 | RB_CLEAR_NODE(&cfqq->rb_node); | |
3102 | RB_CLEAR_NODE(&cfqq->p_node); | |
3103 | INIT_LIST_HEAD(&cfqq->fifo); | |
3104 | ||
30d7b944 | 3105 | cfqq->ref = 0; |
d5036d77 JA |
3106 | cfqq->cfqd = cfqd; |
3107 | ||
3108 | cfq_mark_cfqq_prio_changed(cfqq); | |
3109 | ||
3110 | if (is_sync) { | |
3111 | if (!cfq_class_idle(cfqq)) | |
3112 | cfq_mark_cfqq_idle_window(cfqq); | |
3113 | cfq_mark_cfqq_sync(cfqq); | |
3114 | } | |
3115 | cfqq->pid = pid; | |
3116 | } | |
3117 | ||
24610333 | 3118 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
598971bf | 3119 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 3120 | { |
bca4b914 | 3121 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
598971bf TH |
3122 | struct cfq_queue *sync_cfqq; |
3123 | uint64_t id; | |
24610333 | 3124 | |
598971bf TH |
3125 | rcu_read_lock(); |
3126 | id = bio_blkio_cgroup(bio)->id; | |
3127 | rcu_read_unlock(); | |
24610333 | 3128 | |
598971bf TH |
3129 | /* |
3130 | * Check whether blkcg has changed. The condition may trigger | |
3131 | * spuriously on a newly created cic but there's no harm. | |
3132 | */ | |
3133 | if (unlikely(!cfqd) || likely(cic->blkcg_id == id)) | |
3134 | return; | |
24610333 | 3135 | |
598971bf | 3136 | sync_cfqq = cic_to_cfqq(cic, 1); |
24610333 VG |
3137 | if (sync_cfqq) { |
3138 | /* | |
3139 | * Drop reference to sync queue. A new sync queue will be | |
3140 | * assigned in new group upon arrival of a fresh request. | |
3141 | */ | |
3142 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
3143 | cic_set_cfqq(cic, NULL, 1); | |
3144 | cfq_put_queue(sync_cfqq); | |
3145 | } | |
598971bf TH |
3146 | |
3147 | cic->blkcg_id = id; | |
24610333 | 3148 | } |
598971bf TH |
3149 | #else |
3150 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } | |
24610333 VG |
3151 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
3152 | ||
22e2c507 | 3153 | static struct cfq_queue * |
abede6da TH |
3154 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
3155 | struct bio *bio, gfp_t gfp_mask) | |
22e2c507 | 3156 | { |
0a5a7d0e | 3157 | struct blkio_cgroup *blkcg; |
22e2c507 | 3158 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
cdb16e8f | 3159 | struct cfq_group *cfqg; |
22e2c507 JA |
3160 | |
3161 | retry: | |
2a7f1244 TH |
3162 | rcu_read_lock(); |
3163 | ||
4f85cb96 | 3164 | blkcg = bio_blkio_cgroup(bio); |
cd1604fa | 3165 | cfqg = cfq_lookup_create_cfqg(cfqd, blkcg); |
91fac317 | 3166 | cfqq = cic_to_cfqq(cic, is_sync); |
22e2c507 | 3167 | |
6118b70b JA |
3168 | /* |
3169 | * Always try a new alloc if we fell back to the OOM cfqq | |
3170 | * originally, since it should just be a temporary situation. | |
3171 | */ | |
3172 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
3173 | cfqq = NULL; | |
22e2c507 JA |
3174 | if (new_cfqq) { |
3175 | cfqq = new_cfqq; | |
3176 | new_cfqq = NULL; | |
3177 | } else if (gfp_mask & __GFP_WAIT) { | |
2a7f1244 | 3178 | rcu_read_unlock(); |
22e2c507 | 3179 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c | 3180 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 3181 | gfp_mask | __GFP_ZERO, |
94f6030c | 3182 | cfqd->queue->node); |
22e2c507 | 3183 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
3184 | if (new_cfqq) |
3185 | goto retry; | |
22e2c507 | 3186 | } else { |
94f6030c CL |
3187 | cfqq = kmem_cache_alloc_node(cfq_pool, |
3188 | gfp_mask | __GFP_ZERO, | |
3189 | cfqd->queue->node); | |
22e2c507 JA |
3190 | } |
3191 | ||
6118b70b JA |
3192 | if (cfqq) { |
3193 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
abede6da | 3194 | cfq_init_prio_data(cfqq, cic); |
cdb16e8f | 3195 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
3196 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
3197 | } else | |
3198 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
3199 | } |
3200 | ||
3201 | if (new_cfqq) | |
3202 | kmem_cache_free(cfq_pool, new_cfqq); | |
3203 | ||
2a7f1244 | 3204 | rcu_read_unlock(); |
22e2c507 JA |
3205 | return cfqq; |
3206 | } | |
3207 | ||
c2dea2d1 VT |
3208 | static struct cfq_queue ** |
3209 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
3210 | { | |
fe094d98 | 3211 | switch (ioprio_class) { |
c2dea2d1 VT |
3212 | case IOPRIO_CLASS_RT: |
3213 | return &cfqd->async_cfqq[0][ioprio]; | |
598971bf TH |
3214 | case IOPRIO_CLASS_NONE: |
3215 | ioprio = IOPRIO_NORM; | |
3216 | /* fall through */ | |
c2dea2d1 VT |
3217 | case IOPRIO_CLASS_BE: |
3218 | return &cfqd->async_cfqq[1][ioprio]; | |
3219 | case IOPRIO_CLASS_IDLE: | |
3220 | return &cfqd->async_idle_cfqq; | |
3221 | default: | |
3222 | BUG(); | |
3223 | } | |
3224 | } | |
3225 | ||
15c31be4 | 3226 | static struct cfq_queue * |
abede6da | 3227 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
4f85cb96 | 3228 | struct bio *bio, gfp_t gfp_mask) |
15c31be4 | 3229 | { |
598971bf TH |
3230 | const int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
3231 | const int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
c2dea2d1 | 3232 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
3233 | struct cfq_queue *cfqq = NULL; |
3234 | ||
c2dea2d1 VT |
3235 | if (!is_sync) { |
3236 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
3237 | cfqq = *async_cfqq; | |
3238 | } | |
3239 | ||
6118b70b | 3240 | if (!cfqq) |
abede6da | 3241 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, cic, bio, gfp_mask); |
15c31be4 JA |
3242 | |
3243 | /* | |
3244 | * pin the queue now that it's allocated, scheduler exit will prune it | |
3245 | */ | |
c2dea2d1 | 3246 | if (!is_sync && !(*async_cfqq)) { |
30d7b944 | 3247 | cfqq->ref++; |
c2dea2d1 | 3248 | *async_cfqq = cfqq; |
15c31be4 JA |
3249 | } |
3250 | ||
30d7b944 | 3251 | cfqq->ref++; |
15c31be4 JA |
3252 | return cfqq; |
3253 | } | |
3254 | ||
22e2c507 | 3255 | static void |
383cd721 | 3256 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 3257 | { |
383cd721 SL |
3258 | unsigned long elapsed = jiffies - ttime->last_end_request; |
3259 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 3260 | |
383cd721 SL |
3261 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
3262 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
3263 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
3264 | } | |
3265 | ||
3266 | static void | |
3267 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3268 | struct cfq_io_cq *cic) |
383cd721 | 3269 | { |
f5f2b6ce | 3270 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3271 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3272 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3273 | cfqd->cfq_slice_idle); | |
3274 | } | |
7700fc4f SL |
3275 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3276 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3277 | #endif | |
22e2c507 | 3278 | } |
1da177e4 | 3279 | |
206dc69b | 3280 | static void |
b2c18e1e | 3281 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3282 | struct request *rq) |
206dc69b | 3283 | { |
3dde36dd | 3284 | sector_t sdist = 0; |
41647e7a | 3285 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3286 | if (cfqq->last_request_pos) { |
3287 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3288 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3289 | else | |
3290 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3291 | } | |
206dc69b | 3292 | |
3dde36dd | 3293 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3294 | if (blk_queue_nonrot(cfqd->queue)) |
3295 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3296 | else | |
3297 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3298 | } |
1da177e4 | 3299 | |
22e2c507 JA |
3300 | /* |
3301 | * Disable idle window if the process thinks too long or seeks so much that | |
3302 | * it doesn't matter | |
3303 | */ | |
3304 | static void | |
3305 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3306 | struct cfq_io_cq *cic) |
22e2c507 | 3307 | { |
7b679138 | 3308 | int old_idle, enable_idle; |
1be92f2f | 3309 | |
0871714e JA |
3310 | /* |
3311 | * Don't idle for async or idle io prio class | |
3312 | */ | |
3313 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3314 | return; |
3315 | ||
c265a7f4 | 3316 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3317 | |
76280aff CZ |
3318 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3319 | cfq_mark_cfqq_deep(cfqq); | |
3320 | ||
749ef9f8 CZ |
3321 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
3322 | enable_idle = 0; | |
f6e8d01b | 3323 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
3324 | !cfqd->cfq_slice_idle || |
3325 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3326 | enable_idle = 0; |
383cd721 SL |
3327 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3328 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3329 | enable_idle = 0; |
3330 | else | |
3331 | enable_idle = 1; | |
1da177e4 LT |
3332 | } |
3333 | ||
7b679138 JA |
3334 | if (old_idle != enable_idle) { |
3335 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3336 | if (enable_idle) | |
3337 | cfq_mark_cfqq_idle_window(cfqq); | |
3338 | else | |
3339 | cfq_clear_cfqq_idle_window(cfqq); | |
3340 | } | |
22e2c507 | 3341 | } |
1da177e4 | 3342 | |
22e2c507 JA |
3343 | /* |
3344 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3345 | * no or if we aren't sure, a 1 will cause a preempt. | |
3346 | */ | |
a6151c3a | 3347 | static bool |
22e2c507 | 3348 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3349 | struct request *rq) |
22e2c507 | 3350 | { |
6d048f53 | 3351 | struct cfq_queue *cfqq; |
22e2c507 | 3352 | |
6d048f53 JA |
3353 | cfqq = cfqd->active_queue; |
3354 | if (!cfqq) | |
a6151c3a | 3355 | return false; |
22e2c507 | 3356 | |
6d048f53 | 3357 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3358 | return false; |
22e2c507 JA |
3359 | |
3360 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3361 | return true; |
1e3335de | 3362 | |
875feb63 DS |
3363 | /* |
3364 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3365 | */ | |
3366 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3367 | return false; | |
3368 | ||
374f84ac JA |
3369 | /* |
3370 | * if the new request is sync, but the currently running queue is | |
3371 | * not, let the sync request have priority. | |
3372 | */ | |
5e705374 | 3373 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3374 | return true; |
1e3335de | 3375 | |
8682e1f1 VG |
3376 | if (new_cfqq->cfqg != cfqq->cfqg) |
3377 | return false; | |
3378 | ||
3379 | if (cfq_slice_used(cfqq)) | |
3380 | return true; | |
3381 | ||
3382 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
3383 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && | |
3384 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && | |
3385 | new_cfqq->service_tree->count == 2 && | |
3386 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3387 | return true; | |
3388 | ||
b53d1ed7 JA |
3389 | /* |
3390 | * So both queues are sync. Let the new request get disk time if | |
3391 | * it's a metadata request and the current queue is doing regular IO. | |
3392 | */ | |
65299a3b | 3393 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
3394 | return true; |
3395 | ||
3a9a3f6c DS |
3396 | /* |
3397 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3398 | */ | |
3399 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3400 | return true; |
3a9a3f6c | 3401 | |
d2d59e18 SL |
3402 | /* An idle queue should not be idle now for some reason */ |
3403 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
3404 | return true; | |
3405 | ||
1e3335de | 3406 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3407 | return false; |
1e3335de JA |
3408 | |
3409 | /* | |
3410 | * if this request is as-good as one we would expect from the | |
3411 | * current cfqq, let it preempt | |
3412 | */ | |
e9ce335d | 3413 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3414 | return true; |
1e3335de | 3415 | |
a6151c3a | 3416 | return false; |
22e2c507 JA |
3417 | } |
3418 | ||
3419 | /* | |
3420 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3421 | * let it have half of its nominal slice. | |
3422 | */ | |
3423 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3424 | { | |
df0793ab SL |
3425 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
3426 | ||
7b679138 | 3427 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 3428 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3429 | |
f8ae6e3e SL |
3430 | /* |
3431 | * workload type is changed, don't save slice, otherwise preempt | |
3432 | * doesn't happen | |
3433 | */ | |
df0793ab | 3434 | if (old_type != cfqq_type(cfqq)) |
f8ae6e3e SL |
3435 | cfqq->cfqg->saved_workload_slice = 0; |
3436 | ||
bf572256 JA |
3437 | /* |
3438 | * Put the new queue at the front of the of the current list, | |
3439 | * so we know that it will be selected next. | |
3440 | */ | |
3441 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3442 | |
3443 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3444 | |
62a37f6b JT |
3445 | cfqq->slice_end = 0; |
3446 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3447 | } |
3448 | ||
22e2c507 | 3449 | /* |
5e705374 | 3450 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3451 | * something we should do about it |
3452 | */ | |
3453 | static void | |
5e705374 JA |
3454 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3455 | struct request *rq) | |
22e2c507 | 3456 | { |
c5869807 | 3457 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3458 | |
45333d5a | 3459 | cfqd->rq_queued++; |
65299a3b CH |
3460 | if (rq->cmd_flags & REQ_PRIO) |
3461 | cfqq->prio_pending++; | |
374f84ac | 3462 | |
383cd721 | 3463 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3464 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3465 | cfq_update_idle_window(cfqd, cfqq, cic); |
3466 | ||
b2c18e1e | 3467 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3468 | |
3469 | if (cfqq == cfqd->active_queue) { | |
3470 | /* | |
b029195d JA |
3471 | * Remember that we saw a request from this process, but |
3472 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3473 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3474 | * and merging. If the request is already larger than a single |
3475 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3476 | * merging is already done. Ditto for a busy system that |
3477 | * has other work pending, don't risk delaying until the | |
3478 | * idle timer unplug to continue working. | |
22e2c507 | 3479 | */ |
d6ceb25e | 3480 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3481 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3482 | cfqd->busy_queues > 1) { | |
812df48d | 3483 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3484 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3485 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3486 | } else { |
155fead9 | 3487 | cfqg_stats_update_idle_time(cfqq->cfqg); |
bf791937 | 3488 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3489 | } |
d6ceb25e | 3490 | } |
5e705374 | 3491 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3492 | /* |
3493 | * not the active queue - expire current slice if it is | |
3494 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3495 | * has some old slice time left and is of higher priority or |
3496 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3497 | */ |
3498 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3499 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3500 | } |
1da177e4 LT |
3501 | } |
3502 | ||
165125e1 | 3503 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3504 | { |
b4878f24 | 3505 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3506 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3507 | |
7b679138 | 3508 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 3509 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 3510 | |
30996f40 | 3511 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3512 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3513 | cfq_add_rq_rb(rq); |
155fead9 TH |
3514 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
3515 | rq->cmd_flags); | |
5e705374 | 3516 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3517 | } |
3518 | ||
45333d5a AC |
3519 | /* |
3520 | * Update hw_tag based on peak queue depth over 50 samples under | |
3521 | * sufficient load. | |
3522 | */ | |
3523 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3524 | { | |
1a1238a7 SL |
3525 | struct cfq_queue *cfqq = cfqd->active_queue; |
3526 | ||
53c583d2 CZ |
3527 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3528 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3529 | |
3530 | if (cfqd->hw_tag == 1) | |
3531 | return; | |
45333d5a AC |
3532 | |
3533 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3534 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3535 | return; |
3536 | ||
1a1238a7 SL |
3537 | /* |
3538 | * If active queue hasn't enough requests and can idle, cfq might not | |
3539 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3540 | * case | |
3541 | */ | |
3542 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3543 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3544 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3545 | return; |
3546 | ||
45333d5a AC |
3547 | if (cfqd->hw_tag_samples++ < 50) |
3548 | return; | |
3549 | ||
e459dd08 | 3550 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3551 | cfqd->hw_tag = 1; |
3552 | else | |
3553 | cfqd->hw_tag = 0; | |
45333d5a AC |
3554 | } |
3555 | ||
7667aa06 VG |
3556 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3557 | { | |
c5869807 | 3558 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 3559 | |
02a8f01b JT |
3560 | /* If the queue already has requests, don't wait */ |
3561 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3562 | return false; | |
3563 | ||
7667aa06 VG |
3564 | /* If there are other queues in the group, don't wait */ |
3565 | if (cfqq->cfqg->nr_cfqq > 1) | |
3566 | return false; | |
3567 | ||
7700fc4f SL |
3568 | /* the only queue in the group, but think time is big */ |
3569 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
3570 | return false; | |
3571 | ||
7667aa06 VG |
3572 | if (cfq_slice_used(cfqq)) |
3573 | return true; | |
3574 | ||
3575 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
3576 | if (cic && sample_valid(cic->ttime.ttime_samples) |
3577 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
3578 | return true; |
3579 | ||
3580 | /* | |
3581 | * If think times is less than a jiffy than ttime_mean=0 and above | |
3582 | * will not be true. It might happen that slice has not expired yet | |
3583 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
3584 | * case where think time is less than a jiffy, mark the queue wait | |
3585 | * busy if only 1 jiffy is left in the slice. | |
3586 | */ | |
3587 | if (cfqq->slice_end - jiffies == 1) | |
3588 | return true; | |
3589 | ||
3590 | return false; | |
3591 | } | |
3592 | ||
165125e1 | 3593 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3594 | { |
5e705374 | 3595 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3596 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3597 | const int sync = rq_is_sync(rq); |
b4878f24 | 3598 | unsigned long now; |
1da177e4 | 3599 | |
b4878f24 | 3600 | now = jiffies; |
33659ebb CH |
3601 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3602 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 3603 | |
45333d5a AC |
3604 | cfq_update_hw_tag(cfqd); |
3605 | ||
53c583d2 | 3606 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 3607 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 3608 | cfqd->rq_in_driver--; |
6d048f53 | 3609 | cfqq->dispatched--; |
80bdf0c7 | 3610 | (RQ_CFQG(rq))->dispatched--; |
155fead9 TH |
3611 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
3612 | rq_io_start_time_ns(rq), rq->cmd_flags); | |
1da177e4 | 3613 | |
53c583d2 | 3614 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 3615 | |
365722bb | 3616 | if (sync) { |
f5f2b6ce SL |
3617 | struct cfq_rb_root *service_tree; |
3618 | ||
383cd721 | 3619 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
3620 | |
3621 | if (cfq_cfqq_on_rr(cfqq)) | |
3622 | service_tree = cfqq->service_tree; | |
3623 | else | |
3624 | service_tree = service_tree_for(cfqq->cfqg, | |
3625 | cfqq_prio(cfqq), cfqq_type(cfqq)); | |
3626 | service_tree->ttime.last_end_request = now; | |
573412b2 CZ |
3627 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3628 | cfqd->last_delayed_sync = now; | |
365722bb | 3629 | } |
caaa5f9f | 3630 | |
7700fc4f SL |
3631 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3632 | cfqq->cfqg->ttime.last_end_request = now; | |
3633 | #endif | |
3634 | ||
caaa5f9f JA |
3635 | /* |
3636 | * If this is the active queue, check if it needs to be expired, | |
3637 | * or if we want to idle in case it has no pending requests. | |
3638 | */ | |
3639 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3640 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3641 | ||
44f7c160 JA |
3642 | if (cfq_cfqq_slice_new(cfqq)) { |
3643 | cfq_set_prio_slice(cfqd, cfqq); | |
3644 | cfq_clear_cfqq_slice_new(cfqq); | |
3645 | } | |
f75edf2d VG |
3646 | |
3647 | /* | |
7667aa06 VG |
3648 | * Should we wait for next request to come in before we expire |
3649 | * the queue. | |
f75edf2d | 3650 | */ |
7667aa06 | 3651 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
3652 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3653 | if (!cfqd->cfq_slice_idle) | |
3654 | extend_sl = cfqd->cfq_group_idle; | |
3655 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 3656 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 3657 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
3658 | } |
3659 | ||
a36e71f9 | 3660 | /* |
8e550632 CZ |
3661 | * Idling is not enabled on: |
3662 | * - expired queues | |
3663 | * - idle-priority queues | |
3664 | * - async queues | |
3665 | * - queues with still some requests queued | |
3666 | * - when there is a close cooperator | |
a36e71f9 | 3667 | */ |
0871714e | 3668 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 3669 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3670 | else if (sync && cfqq_empty && |
3671 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 3672 | cfq_arm_slice_timer(cfqd); |
8e550632 | 3673 | } |
caaa5f9f | 3674 | } |
6d048f53 | 3675 | |
53c583d2 | 3676 | if (!cfqd->rq_in_driver) |
23e018a1 | 3677 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3678 | } |
3679 | ||
89850f7e | 3680 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3681 | { |
1b379d8d | 3682 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3683 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3684 | return ELV_MQUEUE_MUST; |
3b18152c | 3685 | } |
1da177e4 | 3686 | |
22e2c507 | 3687 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3688 | } |
3689 | ||
165125e1 | 3690 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3691 | { |
3692 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3693 | struct task_struct *tsk = current; | |
c5869807 | 3694 | struct cfq_io_cq *cic; |
22e2c507 JA |
3695 | struct cfq_queue *cfqq; |
3696 | ||
3697 | /* | |
3698 | * don't force setup of a queue from here, as a call to may_queue | |
3699 | * does not necessarily imply that a request actually will be queued. | |
3700 | * so just lookup a possibly existing queue, or return 'may queue' | |
3701 | * if that fails | |
3702 | */ | |
4ac845a2 | 3703 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3704 | if (!cic) |
3705 | return ELV_MQUEUE_MAY; | |
3706 | ||
b0b78f81 | 3707 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3708 | if (cfqq) { |
abede6da | 3709 | cfq_init_prio_data(cfqq, cic); |
22e2c507 | 3710 | |
89850f7e | 3711 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3712 | } |
3713 | ||
3714 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3715 | } |
3716 | ||
1da177e4 LT |
3717 | /* |
3718 | * queue lock held here | |
3719 | */ | |
bb37b94c | 3720 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3721 | { |
5e705374 | 3722 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3723 | |
5e705374 | 3724 | if (cfqq) { |
22e2c507 | 3725 | const int rw = rq_data_dir(rq); |
1da177e4 | 3726 | |
22e2c507 JA |
3727 | BUG_ON(!cfqq->allocated[rw]); |
3728 | cfqq->allocated[rw]--; | |
1da177e4 | 3729 | |
7f1dc8a2 | 3730 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 3731 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
3732 | rq->elv.priv[0] = NULL; |
3733 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 3734 | |
1da177e4 LT |
3735 | cfq_put_queue(cfqq); |
3736 | } | |
3737 | } | |
3738 | ||
df5fe3e8 | 3739 | static struct cfq_queue * |
c5869807 | 3740 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
3741 | struct cfq_queue *cfqq) |
3742 | { | |
3743 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3744 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3745 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3746 | cfq_put_queue(cfqq); |
3747 | return cic_to_cfqq(cic, 1); | |
3748 | } | |
3749 | ||
e6c5bc73 JM |
3750 | /* |
3751 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3752 | * was the last process referring to said cfqq. | |
3753 | */ | |
3754 | static struct cfq_queue * | |
c5869807 | 3755 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
3756 | { |
3757 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
3758 | cfqq->pid = current->pid; |
3759 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 3760 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
3761 | return cfqq; |
3762 | } | |
3763 | ||
3764 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
3765 | |
3766 | cfq_put_cooperator(cfqq); | |
3767 | ||
e6c5bc73 JM |
3768 | cfq_put_queue(cfqq); |
3769 | return NULL; | |
3770 | } | |
1da177e4 | 3771 | /* |
22e2c507 | 3772 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3773 | */ |
22e2c507 | 3774 | static int |
852c788f TH |
3775 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
3776 | gfp_t gfp_mask) | |
1da177e4 LT |
3777 | { |
3778 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 3779 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 3780 | const int rw = rq_data_dir(rq); |
a6151c3a | 3781 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3782 | struct cfq_queue *cfqq; |
1da177e4 LT |
3783 | |
3784 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3785 | ||
216284c3 | 3786 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 3787 | |
598971bf TH |
3788 | check_ioprio_changed(cic, bio); |
3789 | check_blkcg_changed(cic, bio); | |
e6c5bc73 | 3790 | new_queue: |
91fac317 | 3791 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3792 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
abede6da | 3793 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio, gfp_mask); |
91fac317 | 3794 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3795 | } else { |
e6c5bc73 JM |
3796 | /* |
3797 | * If the queue was seeky for too long, break it apart. | |
3798 | */ | |
ae54abed | 3799 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
3800 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
3801 | cfqq = split_cfqq(cic, cfqq); | |
3802 | if (!cfqq) | |
3803 | goto new_queue; | |
3804 | } | |
3805 | ||
df5fe3e8 JM |
3806 | /* |
3807 | * Check to see if this queue is scheduled to merge with | |
3808 | * another, closely cooperating queue. The merging of | |
3809 | * queues happens here as it must be done in process context. | |
3810 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3811 | */ | |
3812 | if (cfqq->new_cfqq) | |
3813 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3814 | } |
1da177e4 LT |
3815 | |
3816 | cfqq->allocated[rw]++; | |
1da177e4 | 3817 | |
6fae9c25 | 3818 | cfqq->ref++; |
eb7d8c07 | 3819 | cfqg_get(cfqq->cfqg); |
a612fddf | 3820 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 3821 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 3822 | spin_unlock_irq(q->queue_lock); |
5e705374 | 3823 | return 0; |
1da177e4 LT |
3824 | } |
3825 | ||
65f27f38 | 3826 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3827 | { |
65f27f38 | 3828 | struct cfq_data *cfqd = |
23e018a1 | 3829 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3830 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3831 | |
40bb54d1 | 3832 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 3833 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3834 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3835 | } |
3836 | ||
3837 | /* | |
3838 | * Timer running if the active_queue is currently idling inside its time slice | |
3839 | */ | |
3840 | static void cfq_idle_slice_timer(unsigned long data) | |
3841 | { | |
3842 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3843 | struct cfq_queue *cfqq; | |
3844 | unsigned long flags; | |
3c6bd2f8 | 3845 | int timed_out = 1; |
22e2c507 | 3846 | |
7b679138 JA |
3847 | cfq_log(cfqd, "idle timer fired"); |
3848 | ||
22e2c507 JA |
3849 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3850 | ||
fe094d98 JA |
3851 | cfqq = cfqd->active_queue; |
3852 | if (cfqq) { | |
3c6bd2f8 JA |
3853 | timed_out = 0; |
3854 | ||
b029195d JA |
3855 | /* |
3856 | * We saw a request before the queue expired, let it through | |
3857 | */ | |
3858 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3859 | goto out_kick; | |
3860 | ||
22e2c507 JA |
3861 | /* |
3862 | * expired | |
3863 | */ | |
44f7c160 | 3864 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
3865 | goto expire; |
3866 | ||
3867 | /* | |
3868 | * only expire and reinvoke request handler, if there are | |
3869 | * other queues with pending requests | |
3870 | */ | |
caaa5f9f | 3871 | if (!cfqd->busy_queues) |
22e2c507 | 3872 | goto out_cont; |
22e2c507 JA |
3873 | |
3874 | /* | |
3875 | * not expired and it has a request pending, let it dispatch | |
3876 | */ | |
75e50984 | 3877 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3878 | goto out_kick; |
76280aff CZ |
3879 | |
3880 | /* | |
3881 | * Queue depth flag is reset only when the idle didn't succeed | |
3882 | */ | |
3883 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
3884 | } |
3885 | expire: | |
e5ff082e | 3886 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 3887 | out_kick: |
23e018a1 | 3888 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
3889 | out_cont: |
3890 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3891 | } | |
3892 | ||
3b18152c JA |
3893 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3894 | { | |
3895 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 3896 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 3897 | } |
22e2c507 | 3898 | |
c2dea2d1 VT |
3899 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3900 | { | |
3901 | int i; | |
3902 | ||
3903 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
3904 | if (cfqd->async_cfqq[0][i]) | |
3905 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
3906 | if (cfqd->async_cfqq[1][i]) | |
3907 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 3908 | } |
2389d1ef ON |
3909 | |
3910 | if (cfqd->async_idle_cfqq) | |
3911 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
3912 | } |
3913 | ||
b374d18a | 3914 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 3915 | { |
22e2c507 | 3916 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 3917 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3918 | |
3b18152c | 3919 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 3920 | |
d9ff4187 | 3921 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3922 | |
d9ff4187 | 3923 | if (cfqd->active_queue) |
e5ff082e | 3924 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 3925 | |
c2dea2d1 | 3926 | cfq_put_async_queues(cfqd); |
03aa264a TH |
3927 | |
3928 | spin_unlock_irq(q->queue_lock); | |
3929 | ||
a90d742e AV |
3930 | cfq_shutdown_timer_wq(cfqd); |
3931 | ||
f51b802c TH |
3932 | #ifndef CONFIG_CFQ_GROUP_IOSCHED |
3933 | kfree(cfqd->root_group); | |
2abae55f | 3934 | #endif |
e8989fae | 3935 | update_root_blkg_pd(q, BLKIO_POLICY_PROP); |
56edf7d7 | 3936 | kfree(cfqd); |
1da177e4 LT |
3937 | } |
3938 | ||
b2fab5ac | 3939 | static int cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
3940 | { |
3941 | struct cfq_data *cfqd; | |
cd1604fa | 3942 | struct blkio_group *blkg __maybe_unused; |
f51b802c | 3943 | int i; |
1da177e4 | 3944 | |
94f6030c | 3945 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
a73f730d | 3946 | if (!cfqd) |
b2fab5ac | 3947 | return -ENOMEM; |
80b15c73 | 3948 | |
f51b802c TH |
3949 | cfqd->queue = q; |
3950 | q->elevator->elevator_data = cfqd; | |
3951 | ||
1fa8f6d6 VG |
3952 | /* Init root service tree */ |
3953 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
3954 | ||
f51b802c | 3955 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 3956 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
f51b802c TH |
3957 | rcu_read_lock(); |
3958 | spin_lock_irq(q->queue_lock); | |
5624a4e4 | 3959 | |
aaec55a0 | 3960 | blkg = blkg_lookup_create(&blkio_root_cgroup, q, true); |
cd1604fa | 3961 | if (!IS_ERR(blkg)) |
0381411e | 3962 | cfqd->root_group = blkg_to_cfqg(blkg); |
f51b802c TH |
3963 | |
3964 | spin_unlock_irq(q->queue_lock); | |
3965 | rcu_read_unlock(); | |
3966 | #else | |
3967 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), | |
3968 | GFP_KERNEL, cfqd->queue->node); | |
3969 | if (cfqd->root_group) | |
3970 | cfq_init_cfqg_base(cfqd->root_group); | |
3971 | #endif | |
3972 | if (!cfqd->root_group) { | |
5624a4e4 | 3973 | kfree(cfqd); |
b2fab5ac | 3974 | return -ENOMEM; |
5624a4e4 VG |
3975 | } |
3976 | ||
3381cb8d | 3977 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_DEFAULT; |
5624a4e4 | 3978 | |
26a2ac00 JA |
3979 | /* |
3980 | * Not strictly needed (since RB_ROOT just clears the node and we | |
3981 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
3982 | * to add magic to the rb code | |
3983 | */ | |
3984 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
3985 | cfqd->prio_trees[i] = RB_ROOT; | |
3986 | ||
6118b70b JA |
3987 | /* |
3988 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
3989 | * Grab a permanent reference to it, so that the normal code flow | |
f51b802c TH |
3990 | * will not attempt to free it. oom_cfqq is linked to root_group |
3991 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
3992 | * the reference from linking right away. | |
6118b70b JA |
3993 | */ |
3994 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 3995 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
3996 | |
3997 | spin_lock_irq(q->queue_lock); | |
f51b802c | 3998 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 3999 | cfqg_put(cfqd->root_group); |
1adaf3dd | 4000 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 4001 | |
22e2c507 JA |
4002 | init_timer(&cfqd->idle_slice_timer); |
4003 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
4004 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
4005 | ||
23e018a1 | 4006 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 4007 | |
1da177e4 | 4008 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
4009 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
4010 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
4011 | cfqd->cfq_back_max = cfq_back_max; |
4012 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
4013 | cfqd->cfq_slice[0] = cfq_slice_async; |
4014 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
4015 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
4016 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
80bdf0c7 | 4017 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 4018 | cfqd->cfq_latency = 1; |
e459dd08 | 4019 | cfqd->hw_tag = -1; |
edc71131 CZ |
4020 | /* |
4021 | * we optimistically start assuming sync ops weren't delayed in last | |
4022 | * second, in order to have larger depth for async operations. | |
4023 | */ | |
573412b2 | 4024 | cfqd->last_delayed_sync = jiffies - HZ; |
b2fab5ac | 4025 | return 0; |
1da177e4 LT |
4026 | } |
4027 | ||
1da177e4 LT |
4028 | /* |
4029 | * sysfs parts below --> | |
4030 | */ | |
1da177e4 LT |
4031 | static ssize_t |
4032 | cfq_var_show(unsigned int var, char *page) | |
4033 | { | |
4034 | return sprintf(page, "%d\n", var); | |
4035 | } | |
4036 | ||
4037 | static ssize_t | |
4038 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
4039 | { | |
4040 | char *p = (char *) page; | |
4041 | ||
4042 | *var = simple_strtoul(p, &p, 10); | |
4043 | return count; | |
4044 | } | |
4045 | ||
1da177e4 | 4046 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 4047 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 4048 | { \ |
3d1ab40f | 4049 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4050 | unsigned int __data = __VAR; \ |
4051 | if (__CONV) \ | |
4052 | __data = jiffies_to_msecs(__data); \ | |
4053 | return cfq_var_show(__data, (page)); \ | |
4054 | } | |
4055 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
4056 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
4057 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
4058 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
4059 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 4060 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 4061 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
4062 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4063 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4064 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4065 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
1da177e4 LT |
4066 | #undef SHOW_FUNCTION |
4067 | ||
4068 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 4069 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4070 | { \ |
3d1ab40f | 4071 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4072 | unsigned int __data; \ |
4073 | int ret = cfq_var_store(&__data, (page), count); \ | |
4074 | if (__data < (MIN)) \ | |
4075 | __data = (MIN); \ | |
4076 | else if (__data > (MAX)) \ | |
4077 | __data = (MAX); \ | |
4078 | if (__CONV) \ | |
4079 | *(__PTR) = msecs_to_jiffies(__data); \ | |
4080 | else \ | |
4081 | *(__PTR) = __data; \ | |
4082 | return ret; \ | |
4083 | } | |
4084 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4085 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4086 | UINT_MAX, 1); | |
4087 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4088 | UINT_MAX, 1); | |
e572ec7e | 4089 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4090 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4091 | UINT_MAX, 0); | |
22e2c507 | 4092 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4093 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4094 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4095 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4096 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4097 | UINT_MAX, 0); | |
963b72fc | 4098 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
1da177e4 LT |
4099 | #undef STORE_FUNCTION |
4100 | ||
e572ec7e AV |
4101 | #define CFQ_ATTR(name) \ |
4102 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4103 | ||
4104 | static struct elv_fs_entry cfq_attrs[] = { | |
4105 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4106 | CFQ_ATTR(fifo_expire_sync), |
4107 | CFQ_ATTR(fifo_expire_async), | |
4108 | CFQ_ATTR(back_seek_max), | |
4109 | CFQ_ATTR(back_seek_penalty), | |
4110 | CFQ_ATTR(slice_sync), | |
4111 | CFQ_ATTR(slice_async), | |
4112 | CFQ_ATTR(slice_async_rq), | |
4113 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 4114 | CFQ_ATTR(group_idle), |
963b72fc | 4115 | CFQ_ATTR(low_latency), |
e572ec7e | 4116 | __ATTR_NULL |
1da177e4 LT |
4117 | }; |
4118 | ||
1da177e4 LT |
4119 | static struct elevator_type iosched_cfq = { |
4120 | .ops = { | |
4121 | .elevator_merge_fn = cfq_merge, | |
4122 | .elevator_merged_fn = cfq_merged_request, | |
4123 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 4124 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 4125 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4126 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4127 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4128 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4129 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4130 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4131 | .elevator_former_req_fn = elv_rb_former_request, |
4132 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 4133 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 4134 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
4135 | .elevator_set_req_fn = cfq_set_request, |
4136 | .elevator_put_req_fn = cfq_put_request, | |
4137 | .elevator_may_queue_fn = cfq_may_queue, | |
4138 | .elevator_init_fn = cfq_init_queue, | |
4139 | .elevator_exit_fn = cfq_exit_queue, | |
4140 | }, | |
3d3c2379 TH |
4141 | .icq_size = sizeof(struct cfq_io_cq), |
4142 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 4143 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 4144 | .elevator_name = "cfq", |
1da177e4 LT |
4145 | .elevator_owner = THIS_MODULE, |
4146 | }; | |
4147 | ||
3e252066 VG |
4148 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4149 | static struct blkio_policy_type blkio_policy_cfq = { | |
4150 | .ops = { | |
0381411e | 4151 | .blkio_init_group_fn = cfq_init_blkio_group, |
155fead9 | 4152 | .blkio_reset_group_stats_fn = cfqg_stats_reset, |
3e252066 | 4153 | }, |
062a644d | 4154 | .plid = BLKIO_POLICY_PROP, |
0381411e | 4155 | .pdata_size = sizeof(struct cfq_group), |
60c2bc2d | 4156 | .cftypes = cfq_blkcg_files, |
3e252066 | 4157 | }; |
3e252066 VG |
4158 | #endif |
4159 | ||
1da177e4 LT |
4160 | static int __init cfq_init(void) |
4161 | { | |
3d3c2379 TH |
4162 | int ret; |
4163 | ||
22e2c507 JA |
4164 | /* |
4165 | * could be 0 on HZ < 1000 setups | |
4166 | */ | |
4167 | if (!cfq_slice_async) | |
4168 | cfq_slice_async = 1; | |
4169 | if (!cfq_slice_idle) | |
4170 | cfq_slice_idle = 1; | |
4171 | ||
80bdf0c7 VG |
4172 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4173 | if (!cfq_group_idle) | |
4174 | cfq_group_idle = 1; | |
4175 | #else | |
4176 | cfq_group_idle = 0; | |
4177 | #endif | |
3d3c2379 TH |
4178 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4179 | if (!cfq_pool) | |
1da177e4 LT |
4180 | return -ENOMEM; |
4181 | ||
3d3c2379 TH |
4182 | ret = elv_register(&iosched_cfq); |
4183 | if (ret) { | |
4184 | kmem_cache_destroy(cfq_pool); | |
4185 | return ret; | |
4186 | } | |
3d3c2379 | 4187 | |
b95ada55 | 4188 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3e252066 | 4189 | blkio_policy_register(&blkio_policy_cfq); |
b95ada55 | 4190 | #endif |
2fdd82bd | 4191 | return 0; |
1da177e4 LT |
4192 | } |
4193 | ||
4194 | static void __exit cfq_exit(void) | |
4195 | { | |
b95ada55 | 4196 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3e252066 | 4197 | blkio_policy_unregister(&blkio_policy_cfq); |
b95ada55 | 4198 | #endif |
1da177e4 | 4199 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4200 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4201 | } |
4202 | ||
4203 | module_init(cfq_init); | |
4204 | module_exit(cfq_exit); | |
4205 | ||
4206 | MODULE_AUTHOR("Jens Axboe"); | |
4207 | MODULE_LICENSE("GPL"); | |
4208 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |