Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * CFQ, or complete fairness queueing, disk scheduler. |
3 | * | |
4 | * Based on ideas from a previously unfinished io | |
5 | * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli. | |
6 | * | |
0fe23479 | 7 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
1da177e4 | 8 | */ |
1da177e4 | 9 | #include <linux/module.h> |
5a0e3ad6 | 10 | #include <linux/slab.h> |
1cc9be68 AV |
11 | #include <linux/blkdev.h> |
12 | #include <linux/elevator.h> | |
ad5ebd2f | 13 | #include <linux/jiffies.h> |
1da177e4 | 14 | #include <linux/rbtree.h> |
22e2c507 | 15 | #include <linux/ioprio.h> |
7b679138 | 16 | #include <linux/blktrace_api.h> |
6e736be7 | 17 | #include "blk.h" |
629ed0b1 | 18 | #include "blk-cgroup.h" |
1da177e4 LT |
19 | |
20 | /* | |
21 | * tunables | |
22 | */ | |
fe094d98 | 23 | /* max queue in one round of service */ |
abc3c744 | 24 | static const int cfq_quantum = 8; |
64100099 | 25 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
26 | /* maximum backwards seek, in KiB */ |
27 | static const int cfq_back_max = 16 * 1024; | |
28 | /* penalty of a backwards seek */ | |
29 | static const int cfq_back_penalty = 2; | |
64100099 | 30 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 31 | static int cfq_slice_async = HZ / 25; |
64100099 | 32 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 33 | static int cfq_slice_idle = HZ / 125; |
80bdf0c7 | 34 | static int cfq_group_idle = HZ / 125; |
5db5d642 CZ |
35 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
36 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 37 | |
d9e7620e | 38 | /* |
0871714e | 39 | * offset from end of service tree |
d9e7620e | 40 | */ |
0871714e | 41 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
42 | |
43 | /* | |
44 | * below this threshold, we consider thinktime immediate | |
45 | */ | |
46 | #define CFQ_MIN_TT (2) | |
47 | ||
22e2c507 | 48 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 49 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 50 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 51 | |
3dde36dd | 52 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 53 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 54 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 55 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 56 | |
a612fddf TH |
57 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
58 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
59 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 60 | |
e18b890b | 61 | static struct kmem_cache *cfq_pool; |
1da177e4 | 62 | |
22e2c507 JA |
63 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
64 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
65 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
66 | ||
206dc69b | 67 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 68 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 69 | |
c5869807 TH |
70 | struct cfq_ttime { |
71 | unsigned long last_end_request; | |
72 | ||
73 | unsigned long ttime_total; | |
74 | unsigned long ttime_samples; | |
75 | unsigned long ttime_mean; | |
76 | }; | |
77 | ||
cc09e299 JA |
78 | /* |
79 | * Most of our rbtree usage is for sorting with min extraction, so | |
80 | * if we cache the leftmost node we don't have to walk down the tree | |
81 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
82 | * move this into the elevator for the rq sorting as well. | |
83 | */ | |
84 | struct cfq_rb_root { | |
85 | struct rb_root rb; | |
86 | struct rb_node *left; | |
aa6f6a3d | 87 | unsigned count; |
73e9ffdd | 88 | unsigned total_weight; |
1fa8f6d6 | 89 | u64 min_vdisktime; |
f5f2b6ce | 90 | struct cfq_ttime ttime; |
cc09e299 | 91 | }; |
f5f2b6ce SL |
92 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
93 | .ttime = {.last_end_request = jiffies,},} | |
cc09e299 | 94 | |
6118b70b JA |
95 | /* |
96 | * Per process-grouping structure | |
97 | */ | |
98 | struct cfq_queue { | |
99 | /* reference count */ | |
30d7b944 | 100 | int ref; |
6118b70b JA |
101 | /* various state flags, see below */ |
102 | unsigned int flags; | |
103 | /* parent cfq_data */ | |
104 | struct cfq_data *cfqd; | |
105 | /* service_tree member */ | |
106 | struct rb_node rb_node; | |
107 | /* service_tree key */ | |
108 | unsigned long rb_key; | |
109 | /* prio tree member */ | |
110 | struct rb_node p_node; | |
111 | /* prio tree root we belong to, if any */ | |
112 | struct rb_root *p_root; | |
113 | /* sorted list of pending requests */ | |
114 | struct rb_root sort_list; | |
115 | /* if fifo isn't expired, next request to serve */ | |
116 | struct request *next_rq; | |
117 | /* requests queued in sort_list */ | |
118 | int queued[2]; | |
119 | /* currently allocated requests */ | |
120 | int allocated[2]; | |
121 | /* fifo list of requests in sort_list */ | |
122 | struct list_head fifo; | |
123 | ||
dae739eb VG |
124 | /* time when queue got scheduled in to dispatch first request. */ |
125 | unsigned long dispatch_start; | |
f75edf2d | 126 | unsigned int allocated_slice; |
c4081ba5 | 127 | unsigned int slice_dispatch; |
dae739eb VG |
128 | /* time when first request from queue completed and slice started. */ |
129 | unsigned long slice_start; | |
6118b70b JA |
130 | unsigned long slice_end; |
131 | long slice_resid; | |
6118b70b | 132 | |
65299a3b CH |
133 | /* pending priority requests */ |
134 | int prio_pending; | |
6118b70b JA |
135 | /* number of requests that are on the dispatch list or inside driver */ |
136 | int dispatched; | |
137 | ||
138 | /* io prio of this group */ | |
139 | unsigned short ioprio, org_ioprio; | |
4aede84b | 140 | unsigned short ioprio_class; |
6118b70b | 141 | |
c4081ba5 RK |
142 | pid_t pid; |
143 | ||
3dde36dd | 144 | u32 seek_history; |
b2c18e1e JM |
145 | sector_t last_request_pos; |
146 | ||
aa6f6a3d | 147 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 148 | struct cfq_queue *new_cfqq; |
cdb16e8f | 149 | struct cfq_group *cfqg; |
c4e7893e VG |
150 | /* Number of sectors dispatched from queue in single dispatch round */ |
151 | unsigned long nr_sectors; | |
6118b70b JA |
152 | }; |
153 | ||
c0324a02 | 154 | /* |
718eee05 | 155 | * First index in the service_trees. |
c0324a02 CZ |
156 | * IDLE is handled separately, so it has negative index |
157 | */ | |
3bf10fea | 158 | enum wl_class_t { |
c0324a02 | 159 | BE_WORKLOAD = 0, |
615f0259 VG |
160 | RT_WORKLOAD = 1, |
161 | IDLE_WORKLOAD = 2, | |
b4627321 | 162 | CFQ_PRIO_NR, |
c0324a02 CZ |
163 | }; |
164 | ||
718eee05 CZ |
165 | /* |
166 | * Second index in the service_trees. | |
167 | */ | |
168 | enum wl_type_t { | |
169 | ASYNC_WORKLOAD = 0, | |
170 | SYNC_NOIDLE_WORKLOAD = 1, | |
171 | SYNC_WORKLOAD = 2 | |
172 | }; | |
173 | ||
155fead9 TH |
174 | struct cfqg_stats { |
175 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
176 | /* total bytes transferred */ | |
177 | struct blkg_rwstat service_bytes; | |
178 | /* total IOs serviced, post merge */ | |
179 | struct blkg_rwstat serviced; | |
180 | /* number of ios merged */ | |
181 | struct blkg_rwstat merged; | |
182 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
183 | struct blkg_rwstat service_time; | |
184 | /* total time spent waiting in scheduler queue in ns */ | |
185 | struct blkg_rwstat wait_time; | |
186 | /* number of IOs queued up */ | |
187 | struct blkg_rwstat queued; | |
188 | /* total sectors transferred */ | |
189 | struct blkg_stat sectors; | |
190 | /* total disk time and nr sectors dispatched by this group */ | |
191 | struct blkg_stat time; | |
192 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
193 | /* time not charged to this cgroup */ | |
194 | struct blkg_stat unaccounted_time; | |
195 | /* sum of number of ios queued across all samples */ | |
196 | struct blkg_stat avg_queue_size_sum; | |
197 | /* count of samples taken for average */ | |
198 | struct blkg_stat avg_queue_size_samples; | |
199 | /* how many times this group has been removed from service tree */ | |
200 | struct blkg_stat dequeue; | |
201 | /* total time spent waiting for it to be assigned a timeslice. */ | |
202 | struct blkg_stat group_wait_time; | |
3c798398 | 203 | /* time spent idling for this blkcg_gq */ |
155fead9 TH |
204 | struct blkg_stat idle_time; |
205 | /* total time with empty current active q with other requests queued */ | |
206 | struct blkg_stat empty_time; | |
207 | /* fields after this shouldn't be cleared on stat reset */ | |
208 | uint64_t start_group_wait_time; | |
209 | uint64_t start_idle_time; | |
210 | uint64_t start_empty_time; | |
211 | uint16_t flags; | |
212 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
213 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
214 | }; | |
215 | ||
cdb16e8f VG |
216 | /* This is per cgroup per device grouping structure */ |
217 | struct cfq_group { | |
f95a04af TH |
218 | /* must be the first member */ |
219 | struct blkg_policy_data pd; | |
220 | ||
1fa8f6d6 VG |
221 | /* group service_tree member */ |
222 | struct rb_node rb_node; | |
223 | ||
224 | /* group service_tree key */ | |
225 | u64 vdisktime; | |
25bc6b07 | 226 | unsigned int weight; |
8184f93e | 227 | unsigned int new_weight; |
3381cb8d | 228 | unsigned int dev_weight; |
1fa8f6d6 VG |
229 | |
230 | /* number of cfqq currently on this group */ | |
231 | int nr_cfqq; | |
232 | ||
cdb16e8f | 233 | /* |
4495a7d4 | 234 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
235 | * create the array for each prio class but at run time it is used |
236 | * only for RT and BE class and slot for IDLE class remains unused. | |
237 | * This is primarily done to avoid confusion and a gcc warning. | |
238 | */ | |
239 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
240 | /* | |
241 | * rr lists of queues with requests. We maintain service trees for | |
242 | * RT and BE classes. These trees are subdivided in subclasses | |
243 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
244 | * class there is no subclassification and all the cfq queues go on | |
245 | * a single tree service_tree_idle. | |
cdb16e8f VG |
246 | * Counts are embedded in the cfq_rb_root |
247 | */ | |
248 | struct cfq_rb_root service_trees[2][3]; | |
249 | struct cfq_rb_root service_tree_idle; | |
dae739eb | 250 | |
4d2ceea4 VG |
251 | unsigned long saved_wl_slice; |
252 | enum wl_type_t saved_wl_type; | |
253 | enum wl_class_t saved_wl_class; | |
4eef3049 | 254 | |
80bdf0c7 VG |
255 | /* number of requests that are on the dispatch list or inside driver */ |
256 | int dispatched; | |
7700fc4f | 257 | struct cfq_ttime ttime; |
155fead9 | 258 | struct cfqg_stats stats; |
cdb16e8f | 259 | }; |
718eee05 | 260 | |
c5869807 TH |
261 | struct cfq_io_cq { |
262 | struct io_cq icq; /* must be the first member */ | |
263 | struct cfq_queue *cfqq[2]; | |
264 | struct cfq_ttime ttime; | |
598971bf TH |
265 | int ioprio; /* the current ioprio */ |
266 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
267 | uint64_t blkcg_id; /* the current blkcg ID */ | |
268 | #endif | |
c5869807 TH |
269 | }; |
270 | ||
22e2c507 JA |
271 | /* |
272 | * Per block device queue structure | |
273 | */ | |
1da177e4 | 274 | struct cfq_data { |
165125e1 | 275 | struct request_queue *queue; |
1fa8f6d6 VG |
276 | /* Root service tree for cfq_groups */ |
277 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 278 | struct cfq_group *root_group; |
22e2c507 | 279 | |
c0324a02 CZ |
280 | /* |
281 | * The priority currently being served | |
22e2c507 | 282 | */ |
4d2ceea4 VG |
283 | enum wl_class_t serving_wl_class; |
284 | enum wl_type_t serving_wl_type; | |
718eee05 | 285 | unsigned long workload_expires; |
cdb16e8f | 286 | struct cfq_group *serving_group; |
a36e71f9 JA |
287 | |
288 | /* | |
289 | * Each priority tree is sorted by next_request position. These | |
290 | * trees are used when determining if two or more queues are | |
291 | * interleaving requests (see cfq_close_cooperator). | |
292 | */ | |
293 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
294 | ||
22e2c507 | 295 | unsigned int busy_queues; |
ef8a41df | 296 | unsigned int busy_sync_queues; |
22e2c507 | 297 | |
53c583d2 CZ |
298 | int rq_in_driver; |
299 | int rq_in_flight[2]; | |
45333d5a AC |
300 | |
301 | /* | |
302 | * queue-depth detection | |
303 | */ | |
304 | int rq_queued; | |
25776e35 | 305 | int hw_tag; |
e459dd08 CZ |
306 | /* |
307 | * hw_tag can be | |
308 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
309 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
310 | * 0 => no NCQ | |
311 | */ | |
312 | int hw_tag_est_depth; | |
313 | unsigned int hw_tag_samples; | |
1da177e4 | 314 | |
22e2c507 JA |
315 | /* |
316 | * idle window management | |
317 | */ | |
318 | struct timer_list idle_slice_timer; | |
23e018a1 | 319 | struct work_struct unplug_work; |
1da177e4 | 320 | |
22e2c507 | 321 | struct cfq_queue *active_queue; |
c5869807 | 322 | struct cfq_io_cq *active_cic; |
22e2c507 | 323 | |
c2dea2d1 VT |
324 | /* |
325 | * async queue for each priority case | |
326 | */ | |
327 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
328 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 329 | |
6d048f53 | 330 | sector_t last_position; |
1da177e4 | 331 | |
1da177e4 LT |
332 | /* |
333 | * tunables, see top of file | |
334 | */ | |
335 | unsigned int cfq_quantum; | |
22e2c507 | 336 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
337 | unsigned int cfq_back_penalty; |
338 | unsigned int cfq_back_max; | |
22e2c507 JA |
339 | unsigned int cfq_slice[2]; |
340 | unsigned int cfq_slice_async_rq; | |
341 | unsigned int cfq_slice_idle; | |
80bdf0c7 | 342 | unsigned int cfq_group_idle; |
963b72fc | 343 | unsigned int cfq_latency; |
5bf14c07 | 344 | unsigned int cfq_target_latency; |
d9ff4187 | 345 | |
6118b70b JA |
346 | /* |
347 | * Fallback dummy cfqq for extreme OOM conditions | |
348 | */ | |
349 | struct cfq_queue oom_cfqq; | |
365722bb | 350 | |
573412b2 | 351 | unsigned long last_delayed_sync; |
1da177e4 LT |
352 | }; |
353 | ||
25fb5169 VG |
354 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
355 | ||
34b98d03 | 356 | static struct cfq_rb_root *st_for(struct cfq_group *cfqg, |
3bf10fea | 357 | enum wl_class_t class, |
65b32a57 | 358 | enum wl_type_t type) |
c0324a02 | 359 | { |
1fa8f6d6 VG |
360 | if (!cfqg) |
361 | return NULL; | |
362 | ||
3bf10fea | 363 | if (class == IDLE_WORKLOAD) |
cdb16e8f | 364 | return &cfqg->service_tree_idle; |
c0324a02 | 365 | |
3bf10fea | 366 | return &cfqg->service_trees[class][type]; |
c0324a02 CZ |
367 | } |
368 | ||
3b18152c | 369 | enum cfqq_state_flags { |
b0b8d749 JA |
370 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
371 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 372 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 373 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
374 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
375 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
376 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 377 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 378 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 379 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 380 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 381 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 382 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
383 | }; |
384 | ||
385 | #define CFQ_CFQQ_FNS(name) \ | |
386 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
387 | { \ | |
fe094d98 | 388 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
389 | } \ |
390 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
391 | { \ | |
fe094d98 | 392 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
393 | } \ |
394 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
395 | { \ | |
fe094d98 | 396 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
397 | } |
398 | ||
399 | CFQ_CFQQ_FNS(on_rr); | |
400 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 401 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 402 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
403 | CFQ_CFQQ_FNS(fifo_expire); |
404 | CFQ_CFQQ_FNS(idle_window); | |
405 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 406 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 407 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 408 | CFQ_CFQQ_FNS(coop); |
ae54abed | 409 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 410 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 411 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
412 | #undef CFQ_CFQQ_FNS |
413 | ||
f95a04af TH |
414 | static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd) |
415 | { | |
416 | return pd ? container_of(pd, struct cfq_group, pd) : NULL; | |
417 | } | |
418 | ||
f95a04af TH |
419 | static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg) |
420 | { | |
421 | return pd_to_blkg(&cfqg->pd); | |
422 | } | |
423 | ||
629ed0b1 | 424 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
2ce4d50f | 425 | |
155fead9 TH |
426 | /* cfqg stats flags */ |
427 | enum cfqg_stats_flags { | |
428 | CFQG_stats_waiting = 0, | |
429 | CFQG_stats_idling, | |
430 | CFQG_stats_empty, | |
629ed0b1 TH |
431 | }; |
432 | ||
155fead9 TH |
433 | #define CFQG_FLAG_FNS(name) \ |
434 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ | |
629ed0b1 | 435 | { \ |
155fead9 | 436 | stats->flags |= (1 << CFQG_stats_##name); \ |
629ed0b1 | 437 | } \ |
155fead9 | 438 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 439 | { \ |
155fead9 | 440 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
629ed0b1 | 441 | } \ |
155fead9 | 442 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 443 | { \ |
155fead9 | 444 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
629ed0b1 TH |
445 | } \ |
446 | ||
155fead9 TH |
447 | CFQG_FLAG_FNS(waiting) |
448 | CFQG_FLAG_FNS(idling) | |
449 | CFQG_FLAG_FNS(empty) | |
450 | #undef CFQG_FLAG_FNS | |
629ed0b1 TH |
451 | |
452 | /* This should be called with the queue_lock held. */ | |
155fead9 | 453 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
629ed0b1 TH |
454 | { |
455 | unsigned long long now; | |
456 | ||
155fead9 | 457 | if (!cfqg_stats_waiting(stats)) |
629ed0b1 TH |
458 | return; |
459 | ||
460 | now = sched_clock(); | |
461 | if (time_after64(now, stats->start_group_wait_time)) | |
462 | blkg_stat_add(&stats->group_wait_time, | |
463 | now - stats->start_group_wait_time); | |
155fead9 | 464 | cfqg_stats_clear_waiting(stats); |
629ed0b1 TH |
465 | } |
466 | ||
467 | /* This should be called with the queue_lock held. */ | |
155fead9 TH |
468 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
469 | struct cfq_group *curr_cfqg) | |
629ed0b1 | 470 | { |
155fead9 | 471 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 472 | |
155fead9 | 473 | if (cfqg_stats_waiting(stats)) |
629ed0b1 | 474 | return; |
155fead9 | 475 | if (cfqg == curr_cfqg) |
629ed0b1 | 476 | return; |
155fead9 TH |
477 | stats->start_group_wait_time = sched_clock(); |
478 | cfqg_stats_mark_waiting(stats); | |
629ed0b1 TH |
479 | } |
480 | ||
481 | /* This should be called with the queue_lock held. */ | |
155fead9 | 482 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
629ed0b1 TH |
483 | { |
484 | unsigned long long now; | |
485 | ||
155fead9 | 486 | if (!cfqg_stats_empty(stats)) |
629ed0b1 TH |
487 | return; |
488 | ||
489 | now = sched_clock(); | |
490 | if (time_after64(now, stats->start_empty_time)) | |
491 | blkg_stat_add(&stats->empty_time, | |
492 | now - stats->start_empty_time); | |
155fead9 | 493 | cfqg_stats_clear_empty(stats); |
629ed0b1 TH |
494 | } |
495 | ||
155fead9 | 496 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
629ed0b1 | 497 | { |
155fead9 | 498 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
629ed0b1 TH |
499 | } |
500 | ||
155fead9 | 501 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
629ed0b1 | 502 | { |
155fead9 | 503 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
504 | |
505 | if (blkg_rwstat_sum(&stats->queued)) | |
506 | return; | |
507 | ||
508 | /* | |
509 | * group is already marked empty. This can happen if cfqq got new | |
510 | * request in parent group and moved to this group while being added | |
511 | * to service tree. Just ignore the event and move on. | |
512 | */ | |
155fead9 | 513 | if (cfqg_stats_empty(stats)) |
629ed0b1 TH |
514 | return; |
515 | ||
516 | stats->start_empty_time = sched_clock(); | |
155fead9 | 517 | cfqg_stats_mark_empty(stats); |
629ed0b1 TH |
518 | } |
519 | ||
155fead9 | 520 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 521 | { |
155fead9 | 522 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 523 | |
155fead9 | 524 | if (cfqg_stats_idling(stats)) { |
629ed0b1 TH |
525 | unsigned long long now = sched_clock(); |
526 | ||
527 | if (time_after64(now, stats->start_idle_time)) | |
528 | blkg_stat_add(&stats->idle_time, | |
529 | now - stats->start_idle_time); | |
155fead9 | 530 | cfqg_stats_clear_idling(stats); |
629ed0b1 TH |
531 | } |
532 | } | |
533 | ||
155fead9 | 534 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 535 | { |
155fead9 | 536 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 537 | |
155fead9 | 538 | BUG_ON(cfqg_stats_idling(stats)); |
629ed0b1 TH |
539 | |
540 | stats->start_idle_time = sched_clock(); | |
155fead9 | 541 | cfqg_stats_mark_idling(stats); |
629ed0b1 TH |
542 | } |
543 | ||
155fead9 | 544 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
629ed0b1 | 545 | { |
155fead9 | 546 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
547 | |
548 | blkg_stat_add(&stats->avg_queue_size_sum, | |
549 | blkg_rwstat_sum(&stats->queued)); | |
550 | blkg_stat_add(&stats->avg_queue_size_samples, 1); | |
155fead9 | 551 | cfqg_stats_update_group_wait_time(stats); |
629ed0b1 TH |
552 | } |
553 | ||
554 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
555 | ||
f48ec1d7 TH |
556 | static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { } |
557 | static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } | |
558 | static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } | |
559 | static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } | |
560 | static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } | |
561 | static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } | |
562 | static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } | |
629ed0b1 TH |
563 | |
564 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
565 | ||
566 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
2ce4d50f | 567 | |
ffea73fc TH |
568 | static struct blkcg_policy blkcg_policy_cfq; |
569 | ||
570 | static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg) | |
571 | { | |
572 | return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq)); | |
573 | } | |
574 | ||
eb7d8c07 TH |
575 | static inline void cfqg_get(struct cfq_group *cfqg) |
576 | { | |
577 | return blkg_get(cfqg_to_blkg(cfqg)); | |
578 | } | |
579 | ||
580 | static inline void cfqg_put(struct cfq_group *cfqg) | |
581 | { | |
582 | return blkg_put(cfqg_to_blkg(cfqg)); | |
583 | } | |
584 | ||
54e7ed12 TH |
585 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \ |
586 | char __pbuf[128]; \ | |
587 | \ | |
588 | blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \ | |
2868ef7b | 589 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ |
54e7ed12 TH |
590 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ |
591 | __pbuf, ##args); \ | |
592 | } while (0) | |
593 | ||
594 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \ | |
595 | char __pbuf[128]; \ | |
596 | \ | |
597 | blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \ | |
598 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \ | |
599 | } while (0) | |
2868ef7b | 600 | |
155fead9 TH |
601 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
602 | struct cfq_group *curr_cfqg, int rw) | |
2ce4d50f | 603 | { |
155fead9 TH |
604 | blkg_rwstat_add(&cfqg->stats.queued, rw, 1); |
605 | cfqg_stats_end_empty_time(&cfqg->stats); | |
606 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); | |
2ce4d50f TH |
607 | } |
608 | ||
155fead9 TH |
609 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
610 | unsigned long time, unsigned long unaccounted_time) | |
2ce4d50f | 611 | { |
155fead9 | 612 | blkg_stat_add(&cfqg->stats.time, time); |
629ed0b1 | 613 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
155fead9 | 614 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
629ed0b1 | 615 | #endif |
2ce4d50f TH |
616 | } |
617 | ||
155fead9 | 618 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) |
2ce4d50f | 619 | { |
155fead9 | 620 | blkg_rwstat_add(&cfqg->stats.queued, rw, -1); |
2ce4d50f TH |
621 | } |
622 | ||
155fead9 | 623 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) |
2ce4d50f | 624 | { |
155fead9 | 625 | blkg_rwstat_add(&cfqg->stats.merged, rw, 1); |
2ce4d50f TH |
626 | } |
627 | ||
155fead9 TH |
628 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, |
629 | uint64_t bytes, int rw) | |
2ce4d50f | 630 | { |
155fead9 TH |
631 | blkg_stat_add(&cfqg->stats.sectors, bytes >> 9); |
632 | blkg_rwstat_add(&cfqg->stats.serviced, rw, 1); | |
633 | blkg_rwstat_add(&cfqg->stats.service_bytes, rw, bytes); | |
2ce4d50f TH |
634 | } |
635 | ||
155fead9 TH |
636 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
637 | uint64_t start_time, uint64_t io_start_time, int rw) | |
2ce4d50f | 638 | { |
155fead9 | 639 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 640 | unsigned long long now = sched_clock(); |
629ed0b1 TH |
641 | |
642 | if (time_after64(now, io_start_time)) | |
643 | blkg_rwstat_add(&stats->service_time, rw, now - io_start_time); | |
644 | if (time_after64(io_start_time, start_time)) | |
645 | blkg_rwstat_add(&stats->wait_time, rw, | |
646 | io_start_time - start_time); | |
2ce4d50f TH |
647 | } |
648 | ||
3c798398 | 649 | static void cfq_pd_reset_stats(struct blkcg_gq *blkg) |
155fead9 TH |
650 | { |
651 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); | |
652 | struct cfqg_stats *stats = &cfqg->stats; | |
653 | ||
654 | /* queued stats shouldn't be cleared */ | |
655 | blkg_rwstat_reset(&stats->service_bytes); | |
656 | blkg_rwstat_reset(&stats->serviced); | |
657 | blkg_rwstat_reset(&stats->merged); | |
658 | blkg_rwstat_reset(&stats->service_time); | |
659 | blkg_rwstat_reset(&stats->wait_time); | |
660 | blkg_stat_reset(&stats->time); | |
661 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
662 | blkg_stat_reset(&stats->unaccounted_time); | |
663 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
664 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
665 | blkg_stat_reset(&stats->dequeue); | |
666 | blkg_stat_reset(&stats->group_wait_time); | |
667 | blkg_stat_reset(&stats->idle_time); | |
668 | blkg_stat_reset(&stats->empty_time); | |
669 | #endif | |
670 | } | |
671 | ||
eb7d8c07 TH |
672 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
673 | ||
eb7d8c07 TH |
674 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
675 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
676 | ||
7b679138 JA |
677 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
678 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
4495a7d4 | 679 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 | 680 | |
155fead9 TH |
681 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
682 | struct cfq_group *curr_cfqg, int rw) { } | |
683 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, | |
684 | unsigned long time, unsigned long unaccounted_time) { } | |
685 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) { } | |
686 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) { } | |
687 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, | |
688 | uint64_t bytes, int rw) { } | |
689 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, | |
690 | uint64_t start_time, uint64_t io_start_time, int rw) { } | |
2ce4d50f | 691 | |
eb7d8c07 TH |
692 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
693 | ||
7b679138 JA |
694 | #define cfq_log(cfqd, fmt, args...) \ |
695 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
696 | ||
615f0259 VG |
697 | /* Traverses through cfq group service trees */ |
698 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
699 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
700 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
701 | : &cfqg->service_tree_idle; \ | |
702 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
703 | (i == IDLE_WORKLOAD && j == 0); \ | |
704 | j++, st = i < IDLE_WORKLOAD ? \ | |
705 | &cfqg->service_trees[i][j]: NULL) \ | |
706 | ||
f5f2b6ce SL |
707 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
708 | struct cfq_ttime *ttime, bool group_idle) | |
709 | { | |
710 | unsigned long slice; | |
711 | if (!sample_valid(ttime->ttime_samples)) | |
712 | return false; | |
713 | if (group_idle) | |
714 | slice = cfqd->cfq_group_idle; | |
715 | else | |
716 | slice = cfqd->cfq_slice_idle; | |
717 | return ttime->ttime_mean > slice; | |
718 | } | |
615f0259 | 719 | |
02b35081 VG |
720 | static inline bool iops_mode(struct cfq_data *cfqd) |
721 | { | |
722 | /* | |
723 | * If we are not idling on queues and it is a NCQ drive, parallel | |
724 | * execution of requests is on and measuring time is not possible | |
725 | * in most of the cases until and unless we drive shallower queue | |
726 | * depths and that becomes a performance bottleneck. In such cases | |
727 | * switch to start providing fairness in terms of number of IOs. | |
728 | */ | |
729 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
730 | return true; | |
731 | else | |
732 | return false; | |
733 | } | |
734 | ||
3bf10fea | 735 | static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq) |
c0324a02 CZ |
736 | { |
737 | if (cfq_class_idle(cfqq)) | |
738 | return IDLE_WORKLOAD; | |
739 | if (cfq_class_rt(cfqq)) | |
740 | return RT_WORKLOAD; | |
741 | return BE_WORKLOAD; | |
742 | } | |
743 | ||
718eee05 CZ |
744 | |
745 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
746 | { | |
747 | if (!cfq_cfqq_sync(cfqq)) | |
748 | return ASYNC_WORKLOAD; | |
749 | if (!cfq_cfqq_idle_window(cfqq)) | |
750 | return SYNC_NOIDLE_WORKLOAD; | |
751 | return SYNC_WORKLOAD; | |
752 | } | |
753 | ||
3bf10fea | 754 | static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class, |
58ff82f3 VG |
755 | struct cfq_data *cfqd, |
756 | struct cfq_group *cfqg) | |
c0324a02 | 757 | { |
3bf10fea | 758 | if (wl_class == IDLE_WORKLOAD) |
cdb16e8f | 759 | return cfqg->service_tree_idle.count; |
c0324a02 | 760 | |
34b98d03 VG |
761 | return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count + |
762 | cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count + | |
763 | cfqg->service_trees[wl_class][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
764 | } |
765 | ||
f26bd1f0 VG |
766 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
767 | struct cfq_group *cfqg) | |
768 | { | |
34b98d03 VG |
769 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count + |
770 | cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
f26bd1f0 VG |
771 | } |
772 | ||
165125e1 | 773 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 774 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
abede6da | 775 | struct cfq_io_cq *cic, struct bio *bio, |
4f85cb96 | 776 | gfp_t gfp_mask); |
91fac317 | 777 | |
c5869807 TH |
778 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
779 | { | |
780 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
781 | return container_of(icq, struct cfq_io_cq, icq); | |
782 | } | |
783 | ||
47fdd4ca TH |
784 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
785 | struct io_context *ioc) | |
786 | { | |
787 | if (ioc) | |
788 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
789 | return NULL; | |
790 | } | |
791 | ||
c5869807 | 792 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 793 | { |
a6151c3a | 794 | return cic->cfqq[is_sync]; |
91fac317 VT |
795 | } |
796 | ||
c5869807 TH |
797 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
798 | bool is_sync) | |
91fac317 | 799 | { |
a6151c3a | 800 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
801 | } |
802 | ||
c5869807 | 803 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 804 | { |
c5869807 | 805 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
806 | } |
807 | ||
91fac317 VT |
808 | /* |
809 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
810 | * set (in which case it could also be direct WRITE). | |
811 | */ | |
a6151c3a | 812 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 813 | { |
7b6d91da | 814 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 815 | } |
1da177e4 | 816 | |
99f95e52 AM |
817 | /* |
818 | * scheduler run of queue, if there are requests pending and no one in the | |
819 | * driver that will restart queueing | |
820 | */ | |
23e018a1 | 821 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 822 | { |
7b679138 JA |
823 | if (cfqd->busy_queues) { |
824 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 825 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 826 | } |
99f95e52 AM |
827 | } |
828 | ||
44f7c160 JA |
829 | /* |
830 | * Scale schedule slice based on io priority. Use the sync time slice only | |
831 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
832 | * io only, should not get full sync slice length. | |
833 | */ | |
a6151c3a | 834 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 835 | unsigned short prio) |
44f7c160 | 836 | { |
d9e7620e | 837 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 838 | |
d9e7620e JA |
839 | WARN_ON(prio >= IOPRIO_BE_NR); |
840 | ||
841 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
842 | } | |
44f7c160 | 843 | |
d9e7620e JA |
844 | static inline int |
845 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
846 | { | |
847 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
848 | } |
849 | ||
25bc6b07 VG |
850 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
851 | { | |
852 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
853 | ||
3381cb8d | 854 | d = d * CFQ_WEIGHT_DEFAULT; |
25bc6b07 VG |
855 | do_div(d, cfqg->weight); |
856 | return d; | |
857 | } | |
858 | ||
859 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
860 | { | |
861 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
862 | if (delta > 0) | |
863 | min_vdisktime = vdisktime; | |
864 | ||
865 | return min_vdisktime; | |
866 | } | |
867 | ||
868 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
869 | { | |
870 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
871 | if (delta < 0) | |
872 | min_vdisktime = vdisktime; | |
873 | ||
874 | return min_vdisktime; | |
875 | } | |
876 | ||
877 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
878 | { | |
25bc6b07 VG |
879 | struct cfq_group *cfqg; |
880 | ||
25bc6b07 VG |
881 | if (st->left) { |
882 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
883 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
884 | cfqg->vdisktime); | |
25bc6b07 | 885 | } |
25bc6b07 VG |
886 | } |
887 | ||
5db5d642 CZ |
888 | /* |
889 | * get averaged number of queues of RT/BE priority. | |
890 | * average is updated, with a formula that gives more weight to higher numbers, | |
891 | * to quickly follows sudden increases and decrease slowly | |
892 | */ | |
893 | ||
58ff82f3 VG |
894 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
895 | struct cfq_group *cfqg, bool rt) | |
5869619c | 896 | { |
5db5d642 CZ |
897 | unsigned min_q, max_q; |
898 | unsigned mult = cfq_hist_divisor - 1; | |
899 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 900 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 901 | |
58ff82f3 VG |
902 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
903 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
904 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 905 | cfq_hist_divisor; |
58ff82f3 VG |
906 | return cfqg->busy_queues_avg[rt]; |
907 | } | |
908 | ||
909 | static inline unsigned | |
910 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
911 | { | |
912 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
913 | ||
5bf14c07 | 914 | return cfqd->cfq_target_latency * cfqg->weight / st->total_weight; |
5db5d642 CZ |
915 | } |
916 | ||
c553f8e3 | 917 | static inline unsigned |
ba5bd520 | 918 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 919 | { |
5db5d642 CZ |
920 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
921 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
922 | /* |
923 | * interested queues (we consider only the ones with the same | |
924 | * priority class in the cfq group) | |
925 | */ | |
926 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
927 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
928 | unsigned sync_slice = cfqd->cfq_slice[1]; |
929 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
930 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
931 | ||
932 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
933 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
934 | /* scale low_slice according to IO priority | |
935 | * and sync vs async */ | |
936 | unsigned low_slice = | |
937 | min(slice, base_low_slice * slice / sync_slice); | |
938 | /* the adapted slice value is scaled to fit all iqs | |
939 | * into the target latency */ | |
58ff82f3 | 940 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
941 | low_slice); |
942 | } | |
943 | } | |
c553f8e3 SL |
944 | return slice; |
945 | } | |
946 | ||
947 | static inline void | |
948 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
949 | { | |
ba5bd520 | 950 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 951 | |
dae739eb | 952 | cfqq->slice_start = jiffies; |
5db5d642 | 953 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 954 | cfqq->allocated_slice = slice; |
7b679138 | 955 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
956 | } |
957 | ||
958 | /* | |
959 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
960 | * isn't valid until the first request from the dispatch is activated | |
961 | * and the slice time set. | |
962 | */ | |
a6151c3a | 963 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
964 | { |
965 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 966 | return false; |
44f7c160 | 967 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 968 | return false; |
44f7c160 | 969 | |
c1e44756 | 970 | return true; |
44f7c160 JA |
971 | } |
972 | ||
1da177e4 | 973 | /* |
5e705374 | 974 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 975 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 976 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 977 | */ |
5e705374 | 978 | static struct request * |
cf7c25cf | 979 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 980 | { |
cf7c25cf | 981 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 982 | unsigned long back_max; |
e8a99053 AM |
983 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
984 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
985 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 986 | |
5e705374 JA |
987 | if (rq1 == NULL || rq1 == rq2) |
988 | return rq2; | |
989 | if (rq2 == NULL) | |
990 | return rq1; | |
9c2c38a1 | 991 | |
229836bd NK |
992 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
993 | return rq_is_sync(rq1) ? rq1 : rq2; | |
994 | ||
65299a3b CH |
995 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
996 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 997 | |
83096ebf TH |
998 | s1 = blk_rq_pos(rq1); |
999 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 1000 | |
1da177e4 LT |
1001 | /* |
1002 | * by definition, 1KiB is 2 sectors | |
1003 | */ | |
1004 | back_max = cfqd->cfq_back_max * 2; | |
1005 | ||
1006 | /* | |
1007 | * Strict one way elevator _except_ in the case where we allow | |
1008 | * short backward seeks which are biased as twice the cost of a | |
1009 | * similar forward seek. | |
1010 | */ | |
1011 | if (s1 >= last) | |
1012 | d1 = s1 - last; | |
1013 | else if (s1 + back_max >= last) | |
1014 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
1015 | else | |
e8a99053 | 1016 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
1017 | |
1018 | if (s2 >= last) | |
1019 | d2 = s2 - last; | |
1020 | else if (s2 + back_max >= last) | |
1021 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
1022 | else | |
e8a99053 | 1023 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
1024 | |
1025 | /* Found required data */ | |
e8a99053 AM |
1026 | |
1027 | /* | |
1028 | * By doing switch() on the bit mask "wrap" we avoid having to | |
1029 | * check two variables for all permutations: --> faster! | |
1030 | */ | |
1031 | switch (wrap) { | |
5e705374 | 1032 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 1033 | if (d1 < d2) |
5e705374 | 1034 | return rq1; |
e8a99053 | 1035 | else if (d2 < d1) |
5e705374 | 1036 | return rq2; |
e8a99053 AM |
1037 | else { |
1038 | if (s1 >= s2) | |
5e705374 | 1039 | return rq1; |
e8a99053 | 1040 | else |
5e705374 | 1041 | return rq2; |
e8a99053 | 1042 | } |
1da177e4 | 1043 | |
e8a99053 | 1044 | case CFQ_RQ2_WRAP: |
5e705374 | 1045 | return rq1; |
e8a99053 | 1046 | case CFQ_RQ1_WRAP: |
5e705374 JA |
1047 | return rq2; |
1048 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
1049 | default: |
1050 | /* | |
1051 | * Since both rqs are wrapped, | |
1052 | * start with the one that's further behind head | |
1053 | * (--> only *one* back seek required), | |
1054 | * since back seek takes more time than forward. | |
1055 | */ | |
1056 | if (s1 <= s2) | |
5e705374 | 1057 | return rq1; |
1da177e4 | 1058 | else |
5e705374 | 1059 | return rq2; |
1da177e4 LT |
1060 | } |
1061 | } | |
1062 | ||
498d3aa2 JA |
1063 | /* |
1064 | * The below is leftmost cache rbtree addon | |
1065 | */ | |
0871714e | 1066 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 1067 | { |
615f0259 VG |
1068 | /* Service tree is empty */ |
1069 | if (!root->count) | |
1070 | return NULL; | |
1071 | ||
cc09e299 JA |
1072 | if (!root->left) |
1073 | root->left = rb_first(&root->rb); | |
1074 | ||
0871714e JA |
1075 | if (root->left) |
1076 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
1077 | ||
1078 | return NULL; | |
cc09e299 JA |
1079 | } |
1080 | ||
1fa8f6d6 VG |
1081 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
1082 | { | |
1083 | if (!root->left) | |
1084 | root->left = rb_first(&root->rb); | |
1085 | ||
1086 | if (root->left) | |
1087 | return rb_entry_cfqg(root->left); | |
1088 | ||
1089 | return NULL; | |
1090 | } | |
1091 | ||
a36e71f9 JA |
1092 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
1093 | { | |
1094 | rb_erase(n, root); | |
1095 | RB_CLEAR_NODE(n); | |
1096 | } | |
1097 | ||
cc09e299 JA |
1098 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
1099 | { | |
1100 | if (root->left == n) | |
1101 | root->left = NULL; | |
a36e71f9 | 1102 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 1103 | --root->count; |
cc09e299 JA |
1104 | } |
1105 | ||
1da177e4 LT |
1106 | /* |
1107 | * would be nice to take fifo expire time into account as well | |
1108 | */ | |
5e705374 JA |
1109 | static struct request * |
1110 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1111 | struct request *last) | |
1da177e4 | 1112 | { |
21183b07 JA |
1113 | struct rb_node *rbnext = rb_next(&last->rb_node); |
1114 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 1115 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 1116 | |
21183b07 | 1117 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
1118 | |
1119 | if (rbprev) | |
5e705374 | 1120 | prev = rb_entry_rq(rbprev); |
1da177e4 | 1121 | |
21183b07 | 1122 | if (rbnext) |
5e705374 | 1123 | next = rb_entry_rq(rbnext); |
21183b07 JA |
1124 | else { |
1125 | rbnext = rb_first(&cfqq->sort_list); | |
1126 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 1127 | next = rb_entry_rq(rbnext); |
21183b07 | 1128 | } |
1da177e4 | 1129 | |
cf7c25cf | 1130 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
1131 | } |
1132 | ||
d9e7620e JA |
1133 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
1134 | struct cfq_queue *cfqq) | |
1da177e4 | 1135 | { |
d9e7620e JA |
1136 | /* |
1137 | * just an approximation, should be ok. | |
1138 | */ | |
cdb16e8f | 1139 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 1140 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
1141 | } |
1142 | ||
1fa8f6d6 VG |
1143 | static inline s64 |
1144 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1145 | { | |
1146 | return cfqg->vdisktime - st->min_vdisktime; | |
1147 | } | |
1148 | ||
1149 | static void | |
1150 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1151 | { | |
1152 | struct rb_node **node = &st->rb.rb_node; | |
1153 | struct rb_node *parent = NULL; | |
1154 | struct cfq_group *__cfqg; | |
1155 | s64 key = cfqg_key(st, cfqg); | |
1156 | int left = 1; | |
1157 | ||
1158 | while (*node != NULL) { | |
1159 | parent = *node; | |
1160 | __cfqg = rb_entry_cfqg(parent); | |
1161 | ||
1162 | if (key < cfqg_key(st, __cfqg)) | |
1163 | node = &parent->rb_left; | |
1164 | else { | |
1165 | node = &parent->rb_right; | |
1166 | left = 0; | |
1167 | } | |
1168 | } | |
1169 | ||
1170 | if (left) | |
1171 | st->left = &cfqg->rb_node; | |
1172 | ||
1173 | rb_link_node(&cfqg->rb_node, parent, node); | |
1174 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
1175 | } | |
1176 | ||
1177 | static void | |
8184f93e JT |
1178 | cfq_update_group_weight(struct cfq_group *cfqg) |
1179 | { | |
1180 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
3381cb8d | 1181 | if (cfqg->new_weight) { |
8184f93e | 1182 | cfqg->weight = cfqg->new_weight; |
3381cb8d | 1183 | cfqg->new_weight = 0; |
8184f93e JT |
1184 | } |
1185 | } | |
1186 | ||
1187 | static void | |
1188 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1189 | { | |
1190 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
1191 | ||
1192 | cfq_update_group_weight(cfqg); | |
1193 | __cfq_group_service_tree_add(st, cfqg); | |
1194 | st->total_weight += cfqg->weight; | |
1195 | } | |
1196 | ||
1197 | static void | |
1198 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
1199 | { |
1200 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1201 | struct cfq_group *__cfqg; | |
1202 | struct rb_node *n; | |
1203 | ||
1204 | cfqg->nr_cfqq++; | |
760701bf | 1205 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
1206 | return; |
1207 | ||
1208 | /* | |
1209 | * Currently put the group at the end. Later implement something | |
1210 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 1211 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
1212 | */ |
1213 | n = rb_last(&st->rb); | |
1214 | if (n) { | |
1215 | __cfqg = rb_entry_cfqg(n); | |
1216 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
1217 | } else | |
1218 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
1219 | cfq_group_service_tree_add(st, cfqg); |
1220 | } | |
1fa8f6d6 | 1221 | |
8184f93e JT |
1222 | static void |
1223 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1224 | { | |
1225 | st->total_weight -= cfqg->weight; | |
1226 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) | |
1227 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
1228 | } |
1229 | ||
1230 | static void | |
8184f93e | 1231 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
1232 | { |
1233 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1234 | ||
1235 | BUG_ON(cfqg->nr_cfqq < 1); | |
1236 | cfqg->nr_cfqq--; | |
25bc6b07 | 1237 | |
1fa8f6d6 VG |
1238 | /* If there are other cfq queues under this group, don't delete it */ |
1239 | if (cfqg->nr_cfqq) | |
1240 | return; | |
1241 | ||
2868ef7b | 1242 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 1243 | cfq_group_service_tree_del(st, cfqg); |
4d2ceea4 | 1244 | cfqg->saved_wl_slice = 0; |
155fead9 | 1245 | cfqg_stats_update_dequeue(cfqg); |
dae739eb VG |
1246 | } |
1247 | ||
167400d3 JT |
1248 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
1249 | unsigned int *unaccounted_time) | |
dae739eb | 1250 | { |
f75edf2d | 1251 | unsigned int slice_used; |
dae739eb VG |
1252 | |
1253 | /* | |
1254 | * Queue got expired before even a single request completed or | |
1255 | * got expired immediately after first request completion. | |
1256 | */ | |
1257 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
1258 | /* | |
1259 | * Also charge the seek time incurred to the group, otherwise | |
1260 | * if there are mutiple queues in the group, each can dispatch | |
1261 | * a single request on seeky media and cause lots of seek time | |
1262 | * and group will never know it. | |
1263 | */ | |
1264 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
1265 | 1); | |
1266 | } else { | |
1267 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
1268 | if (slice_used > cfqq->allocated_slice) { |
1269 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 1270 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
1271 | } |
1272 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
1273 | *unaccounted_time += cfqq->slice_start - | |
1274 | cfqq->dispatch_start; | |
dae739eb VG |
1275 | } |
1276 | ||
dae739eb VG |
1277 | return slice_used; |
1278 | } | |
1279 | ||
1280 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1281 | struct cfq_queue *cfqq) |
dae739eb VG |
1282 | { |
1283 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 1284 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1285 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1286 | - cfqg->service_tree_idle.count; | |
1287 | ||
1288 | BUG_ON(nr_sync < 0); | |
167400d3 | 1289 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1290 | |
02b35081 VG |
1291 | if (iops_mode(cfqd)) |
1292 | charge = cfqq->slice_dispatch; | |
1293 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1294 | charge = cfqq->allocated_slice; | |
dae739eb VG |
1295 | |
1296 | /* Can't update vdisktime while group is on service tree */ | |
8184f93e | 1297 | cfq_group_service_tree_del(st, cfqg); |
02b35081 | 1298 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
8184f93e JT |
1299 | /* If a new weight was requested, update now, off tree */ |
1300 | cfq_group_service_tree_add(st, cfqg); | |
dae739eb VG |
1301 | |
1302 | /* This group is being expired. Save the context */ | |
1303 | if (time_after(cfqd->workload_expires, jiffies)) { | |
4d2ceea4 | 1304 | cfqg->saved_wl_slice = cfqd->workload_expires |
dae739eb | 1305 | - jiffies; |
4d2ceea4 VG |
1306 | cfqg->saved_wl_type = cfqd->serving_wl_type; |
1307 | cfqg->saved_wl_class = cfqd->serving_wl_class; | |
dae739eb | 1308 | } else |
4d2ceea4 | 1309 | cfqg->saved_wl_slice = 0; |
2868ef7b VG |
1310 | |
1311 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1312 | st->min_vdisktime); | |
fd16d263 JP |
1313 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1314 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1315 | used_sl, cfqq->slice_dispatch, charge, | |
1316 | iops_mode(cfqd), cfqq->nr_sectors); | |
155fead9 TH |
1317 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
1318 | cfqg_stats_set_start_empty_time(cfqg); | |
1fa8f6d6 VG |
1319 | } |
1320 | ||
f51b802c TH |
1321 | /** |
1322 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1323 | * @cfqg: cfq_group to initialize | |
1324 | * | |
1325 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1326 | * is enabled or not. | |
1327 | */ | |
1328 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1329 | { | |
1330 | struct cfq_rb_root *st; | |
1331 | int i, j; | |
1332 | ||
1333 | for_each_cfqg_st(cfqg, i, j, st) | |
1334 | *st = CFQ_RB_ROOT; | |
1335 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1336 | ||
1337 | cfqg->ttime.last_end_request = jiffies; | |
1338 | } | |
1339 | ||
25fb5169 | 1340 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 1341 | static void cfq_pd_init(struct blkcg_gq *blkg) |
f469a7b4 | 1342 | { |
0381411e | 1343 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
25fb5169 | 1344 | |
f51b802c | 1345 | cfq_init_cfqg_base(cfqg); |
3381cb8d | 1346 | cfqg->weight = blkg->blkcg->cfq_weight; |
25fb5169 VG |
1347 | } |
1348 | ||
1349 | /* | |
3e59cf9d VG |
1350 | * Search for the cfq group current task belongs to. request_queue lock must |
1351 | * be held. | |
25fb5169 | 1352 | */ |
cd1604fa | 1353 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
3c798398 | 1354 | struct blkcg *blkcg) |
25fb5169 | 1355 | { |
f469a7b4 | 1356 | struct request_queue *q = cfqd->queue; |
cd1604fa | 1357 | struct cfq_group *cfqg = NULL; |
25fb5169 | 1358 | |
3c798398 TH |
1359 | /* avoid lookup for the common case where there's no blkcg */ |
1360 | if (blkcg == &blkcg_root) { | |
cd1604fa TH |
1361 | cfqg = cfqd->root_group; |
1362 | } else { | |
3c798398 | 1363 | struct blkcg_gq *blkg; |
f469a7b4 | 1364 | |
3c96cb32 | 1365 | blkg = blkg_lookup_create(blkcg, q); |
cd1604fa | 1366 | if (!IS_ERR(blkg)) |
0381411e | 1367 | cfqg = blkg_to_cfqg(blkg); |
cd1604fa | 1368 | } |
f469a7b4 | 1369 | |
25fb5169 VG |
1370 | return cfqg; |
1371 | } | |
1372 | ||
1373 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1374 | { | |
1375 | /* Currently, all async queues are mapped to root group */ | |
1376 | if (!cfq_cfqq_sync(cfqq)) | |
f51b802c | 1377 | cfqg = cfqq->cfqd->root_group; |
25fb5169 VG |
1378 | |
1379 | cfqq->cfqg = cfqg; | |
b1c35769 | 1380 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1381 | cfqg_get(cfqg); |
b1c35769 VG |
1382 | } |
1383 | ||
f95a04af TH |
1384 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, |
1385 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1386 | { |
f95a04af | 1387 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
3381cb8d TH |
1388 | |
1389 | if (!cfqg->dev_weight) | |
60c2bc2d | 1390 | return 0; |
f95a04af | 1391 | return __blkg_prfill_u64(sf, pd, cfqg->dev_weight); |
60c2bc2d TH |
1392 | } |
1393 | ||
3381cb8d TH |
1394 | static int cfqg_print_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1395 | struct seq_file *sf) | |
60c2bc2d | 1396 | { |
3c798398 TH |
1397 | blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), |
1398 | cfqg_prfill_weight_device, &blkcg_policy_cfq, 0, | |
60c2bc2d TH |
1399 | false); |
1400 | return 0; | |
1401 | } | |
1402 | ||
3381cb8d TH |
1403 | static int cfq_print_weight(struct cgroup *cgrp, struct cftype *cft, |
1404 | struct seq_file *sf) | |
60c2bc2d | 1405 | { |
3c798398 | 1406 | seq_printf(sf, "%u\n", cgroup_to_blkcg(cgrp)->cfq_weight); |
60c2bc2d TH |
1407 | return 0; |
1408 | } | |
1409 | ||
3381cb8d TH |
1410 | static int cfqg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1411 | const char *buf) | |
60c2bc2d | 1412 | { |
3c798398 | 1413 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
60c2bc2d | 1414 | struct blkg_conf_ctx ctx; |
3381cb8d | 1415 | struct cfq_group *cfqg; |
60c2bc2d TH |
1416 | int ret; |
1417 | ||
3c798398 | 1418 | ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx); |
60c2bc2d TH |
1419 | if (ret) |
1420 | return ret; | |
1421 | ||
1422 | ret = -EINVAL; | |
3381cb8d | 1423 | cfqg = blkg_to_cfqg(ctx.blkg); |
a2b1693b | 1424 | if (!ctx.v || (ctx.v >= CFQ_WEIGHT_MIN && ctx.v <= CFQ_WEIGHT_MAX)) { |
3381cb8d TH |
1425 | cfqg->dev_weight = ctx.v; |
1426 | cfqg->new_weight = cfqg->dev_weight ?: blkcg->cfq_weight; | |
60c2bc2d TH |
1427 | ret = 0; |
1428 | } | |
1429 | ||
1430 | blkg_conf_finish(&ctx); | |
1431 | return ret; | |
1432 | } | |
1433 | ||
3381cb8d | 1434 | static int cfq_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) |
60c2bc2d | 1435 | { |
3c798398 TH |
1436 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
1437 | struct blkcg_gq *blkg; | |
60c2bc2d TH |
1438 | struct hlist_node *n; |
1439 | ||
3381cb8d | 1440 | if (val < CFQ_WEIGHT_MIN || val > CFQ_WEIGHT_MAX) |
60c2bc2d TH |
1441 | return -EINVAL; |
1442 | ||
1443 | spin_lock_irq(&blkcg->lock); | |
3381cb8d | 1444 | blkcg->cfq_weight = (unsigned int)val; |
60c2bc2d TH |
1445 | |
1446 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
3381cb8d | 1447 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
60c2bc2d | 1448 | |
3381cb8d TH |
1449 | if (cfqg && !cfqg->dev_weight) |
1450 | cfqg->new_weight = blkcg->cfq_weight; | |
60c2bc2d TH |
1451 | } |
1452 | ||
1453 | spin_unlock_irq(&blkcg->lock); | |
1454 | return 0; | |
1455 | } | |
1456 | ||
5bc4afb1 TH |
1457 | static int cfqg_print_stat(struct cgroup *cgrp, struct cftype *cft, |
1458 | struct seq_file *sf) | |
1459 | { | |
3c798398 | 1460 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
5bc4afb1 | 1461 | |
3c798398 | 1462 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_stat, &blkcg_policy_cfq, |
5bc4afb1 TH |
1463 | cft->private, false); |
1464 | return 0; | |
1465 | } | |
1466 | ||
1467 | static int cfqg_print_rwstat(struct cgroup *cgrp, struct cftype *cft, | |
1468 | struct seq_file *sf) | |
1469 | { | |
3c798398 | 1470 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
5bc4afb1 | 1471 | |
3c798398 | 1472 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_rwstat, &blkcg_policy_cfq, |
5bc4afb1 TH |
1473 | cft->private, true); |
1474 | return 0; | |
1475 | } | |
1476 | ||
60c2bc2d | 1477 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
f95a04af TH |
1478 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, |
1479 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1480 | { |
f95a04af | 1481 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
155fead9 | 1482 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
60c2bc2d TH |
1483 | u64 v = 0; |
1484 | ||
1485 | if (samples) { | |
155fead9 | 1486 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
60c2bc2d TH |
1487 | do_div(v, samples); |
1488 | } | |
f95a04af | 1489 | __blkg_prfill_u64(sf, pd, v); |
60c2bc2d TH |
1490 | return 0; |
1491 | } | |
1492 | ||
1493 | /* print avg_queue_size */ | |
155fead9 TH |
1494 | static int cfqg_print_avg_queue_size(struct cgroup *cgrp, struct cftype *cft, |
1495 | struct seq_file *sf) | |
60c2bc2d | 1496 | { |
3c798398 | 1497 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
60c2bc2d | 1498 | |
155fead9 | 1499 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_avg_queue_size, |
3c798398 | 1500 | &blkcg_policy_cfq, 0, false); |
60c2bc2d TH |
1501 | return 0; |
1502 | } | |
1503 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1504 | ||
1505 | static struct cftype cfq_blkcg_files[] = { | |
1506 | { | |
1507 | .name = "weight_device", | |
3381cb8d TH |
1508 | .read_seq_string = cfqg_print_weight_device, |
1509 | .write_string = cfqg_set_weight_device, | |
60c2bc2d TH |
1510 | .max_write_len = 256, |
1511 | }, | |
1512 | { | |
1513 | .name = "weight", | |
3381cb8d TH |
1514 | .read_seq_string = cfq_print_weight, |
1515 | .write_u64 = cfq_set_weight, | |
60c2bc2d TH |
1516 | }, |
1517 | { | |
1518 | .name = "time", | |
5bc4afb1 TH |
1519 | .private = offsetof(struct cfq_group, stats.time), |
1520 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1521 | }, |
1522 | { | |
1523 | .name = "sectors", | |
5bc4afb1 TH |
1524 | .private = offsetof(struct cfq_group, stats.sectors), |
1525 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1526 | }, |
1527 | { | |
1528 | .name = "io_service_bytes", | |
5bc4afb1 TH |
1529 | .private = offsetof(struct cfq_group, stats.service_bytes), |
1530 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1531 | }, |
1532 | { | |
1533 | .name = "io_serviced", | |
5bc4afb1 TH |
1534 | .private = offsetof(struct cfq_group, stats.serviced), |
1535 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1536 | }, |
1537 | { | |
1538 | .name = "io_service_time", | |
5bc4afb1 TH |
1539 | .private = offsetof(struct cfq_group, stats.service_time), |
1540 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1541 | }, |
1542 | { | |
1543 | .name = "io_wait_time", | |
5bc4afb1 TH |
1544 | .private = offsetof(struct cfq_group, stats.wait_time), |
1545 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1546 | }, |
1547 | { | |
1548 | .name = "io_merged", | |
5bc4afb1 TH |
1549 | .private = offsetof(struct cfq_group, stats.merged), |
1550 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1551 | }, |
1552 | { | |
1553 | .name = "io_queued", | |
5bc4afb1 TH |
1554 | .private = offsetof(struct cfq_group, stats.queued), |
1555 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1556 | }, |
1557 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
1558 | { | |
1559 | .name = "avg_queue_size", | |
155fead9 | 1560 | .read_seq_string = cfqg_print_avg_queue_size, |
60c2bc2d TH |
1561 | }, |
1562 | { | |
1563 | .name = "group_wait_time", | |
5bc4afb1 TH |
1564 | .private = offsetof(struct cfq_group, stats.group_wait_time), |
1565 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1566 | }, |
1567 | { | |
1568 | .name = "idle_time", | |
5bc4afb1 TH |
1569 | .private = offsetof(struct cfq_group, stats.idle_time), |
1570 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1571 | }, |
1572 | { | |
1573 | .name = "empty_time", | |
5bc4afb1 TH |
1574 | .private = offsetof(struct cfq_group, stats.empty_time), |
1575 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1576 | }, |
1577 | { | |
1578 | .name = "dequeue", | |
5bc4afb1 TH |
1579 | .private = offsetof(struct cfq_group, stats.dequeue), |
1580 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1581 | }, |
1582 | { | |
1583 | .name = "unaccounted_time", | |
5bc4afb1 TH |
1584 | .private = offsetof(struct cfq_group, stats.unaccounted_time), |
1585 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1586 | }, |
1587 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1588 | { } /* terminate */ | |
1589 | }; | |
25fb5169 | 1590 | #else /* GROUP_IOSCHED */ |
cd1604fa | 1591 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
3c798398 | 1592 | struct blkcg *blkcg) |
25fb5169 | 1593 | { |
f51b802c | 1594 | return cfqd->root_group; |
25fb5169 | 1595 | } |
7f1dc8a2 | 1596 | |
25fb5169 VG |
1597 | static inline void |
1598 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1599 | cfqq->cfqg = cfqg; | |
1600 | } | |
1601 | ||
1602 | #endif /* GROUP_IOSCHED */ | |
1603 | ||
498d3aa2 | 1604 | /* |
c0324a02 | 1605 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1606 | * requests waiting to be processed. It is sorted in the order that |
1607 | * we will service the queues. | |
1608 | */ | |
a36e71f9 | 1609 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1610 | bool add_front) |
d9e7620e | 1611 | { |
0871714e JA |
1612 | struct rb_node **p, *parent; |
1613 | struct cfq_queue *__cfqq; | |
d9e7620e | 1614 | unsigned long rb_key; |
34b98d03 | 1615 | struct cfq_rb_root *st; |
498d3aa2 | 1616 | int left; |
dae739eb | 1617 | int new_cfqq = 1; |
ae30c286 | 1618 | |
34b98d03 | 1619 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq)); |
0871714e JA |
1620 | if (cfq_class_idle(cfqq)) { |
1621 | rb_key = CFQ_IDLE_DELAY; | |
34b98d03 | 1622 | parent = rb_last(&st->rb); |
0871714e JA |
1623 | if (parent && parent != &cfqq->rb_node) { |
1624 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1625 | rb_key += __cfqq->rb_key; | |
1626 | } else | |
1627 | rb_key += jiffies; | |
1628 | } else if (!add_front) { | |
b9c8946b JA |
1629 | /* |
1630 | * Get our rb key offset. Subtract any residual slice | |
1631 | * value carried from last service. A negative resid | |
1632 | * count indicates slice overrun, and this should position | |
1633 | * the next service time further away in the tree. | |
1634 | */ | |
edd75ffd | 1635 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1636 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1637 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1638 | } else { |
1639 | rb_key = -HZ; | |
34b98d03 | 1640 | __cfqq = cfq_rb_first(st); |
48e025e6 CZ |
1641 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1642 | } | |
1da177e4 | 1643 | |
d9e7620e | 1644 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1645 | new_cfqq = 0; |
99f9628a | 1646 | /* |
d9e7620e | 1647 | * same position, nothing more to do |
99f9628a | 1648 | */ |
34b98d03 | 1649 | if (rb_key == cfqq->rb_key && cfqq->service_tree == st) |
d9e7620e | 1650 | return; |
1da177e4 | 1651 | |
aa6f6a3d CZ |
1652 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1653 | cfqq->service_tree = NULL; | |
1da177e4 | 1654 | } |
d9e7620e | 1655 | |
498d3aa2 | 1656 | left = 1; |
0871714e | 1657 | parent = NULL; |
34b98d03 VG |
1658 | cfqq->service_tree = st; |
1659 | p = &st->rb.rb_node; | |
d9e7620e JA |
1660 | while (*p) { |
1661 | parent = *p; | |
1662 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1663 | ||
0c534e0a | 1664 | /* |
c0324a02 | 1665 | * sort by key, that represents service time. |
0c534e0a | 1666 | */ |
c0324a02 | 1667 | if (time_before(rb_key, __cfqq->rb_key)) |
1f23f121 | 1668 | p = &parent->rb_left; |
c0324a02 | 1669 | else { |
1f23f121 | 1670 | p = &parent->rb_right; |
cc09e299 | 1671 | left = 0; |
c0324a02 | 1672 | } |
d9e7620e JA |
1673 | } |
1674 | ||
cc09e299 | 1675 | if (left) |
34b98d03 | 1676 | st->left = &cfqq->rb_node; |
cc09e299 | 1677 | |
d9e7620e JA |
1678 | cfqq->rb_key = rb_key; |
1679 | rb_link_node(&cfqq->rb_node, parent, p); | |
34b98d03 VG |
1680 | rb_insert_color(&cfqq->rb_node, &st->rb); |
1681 | st->count++; | |
20359f27 | 1682 | if (add_front || !new_cfqq) |
dae739eb | 1683 | return; |
8184f93e | 1684 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1685 | } |
1686 | ||
a36e71f9 | 1687 | static struct cfq_queue * |
f2d1f0ae JA |
1688 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1689 | sector_t sector, struct rb_node **ret_parent, | |
1690 | struct rb_node ***rb_link) | |
a36e71f9 | 1691 | { |
a36e71f9 JA |
1692 | struct rb_node **p, *parent; |
1693 | struct cfq_queue *cfqq = NULL; | |
1694 | ||
1695 | parent = NULL; | |
1696 | p = &root->rb_node; | |
1697 | while (*p) { | |
1698 | struct rb_node **n; | |
1699 | ||
1700 | parent = *p; | |
1701 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1702 | ||
1703 | /* | |
1704 | * Sort strictly based on sector. Smallest to the left, | |
1705 | * largest to the right. | |
1706 | */ | |
2e46e8b2 | 1707 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1708 | n = &(*p)->rb_right; |
2e46e8b2 | 1709 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1710 | n = &(*p)->rb_left; |
1711 | else | |
1712 | break; | |
1713 | p = n; | |
3ac6c9f8 | 1714 | cfqq = NULL; |
a36e71f9 JA |
1715 | } |
1716 | ||
1717 | *ret_parent = parent; | |
1718 | if (rb_link) | |
1719 | *rb_link = p; | |
3ac6c9f8 | 1720 | return cfqq; |
a36e71f9 JA |
1721 | } |
1722 | ||
1723 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1724 | { | |
a36e71f9 JA |
1725 | struct rb_node **p, *parent; |
1726 | struct cfq_queue *__cfqq; | |
1727 | ||
f2d1f0ae JA |
1728 | if (cfqq->p_root) { |
1729 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1730 | cfqq->p_root = NULL; | |
1731 | } | |
a36e71f9 JA |
1732 | |
1733 | if (cfq_class_idle(cfqq)) | |
1734 | return; | |
1735 | if (!cfqq->next_rq) | |
1736 | return; | |
1737 | ||
f2d1f0ae | 1738 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1739 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1740 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1741 | if (!__cfqq) { |
1742 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1743 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1744 | } else | |
1745 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1746 | } |
1747 | ||
498d3aa2 JA |
1748 | /* |
1749 | * Update cfqq's position in the service tree. | |
1750 | */ | |
edd75ffd | 1751 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1752 | { |
6d048f53 JA |
1753 | /* |
1754 | * Resorting requires the cfqq to be on the RR list already. | |
1755 | */ | |
a36e71f9 | 1756 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1757 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1758 | cfq_prio_tree_add(cfqd, cfqq); |
1759 | } | |
6d048f53 JA |
1760 | } |
1761 | ||
1da177e4 LT |
1762 | /* |
1763 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1764 | * the pending list according to last request service |
1da177e4 | 1765 | */ |
febffd61 | 1766 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1767 | { |
7b679138 | 1768 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1769 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1770 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 1771 | cfqd->busy_queues++; |
ef8a41df SL |
1772 | if (cfq_cfqq_sync(cfqq)) |
1773 | cfqd->busy_sync_queues++; | |
1da177e4 | 1774 | |
edd75ffd | 1775 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1776 | } |
1777 | ||
498d3aa2 JA |
1778 | /* |
1779 | * Called when the cfqq no longer has requests pending, remove it from | |
1780 | * the service tree. | |
1781 | */ | |
febffd61 | 1782 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1783 | { |
7b679138 | 1784 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1785 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1786 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1787 | |
aa6f6a3d CZ |
1788 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1789 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1790 | cfqq->service_tree = NULL; | |
1791 | } | |
f2d1f0ae JA |
1792 | if (cfqq->p_root) { |
1793 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1794 | cfqq->p_root = NULL; | |
1795 | } | |
d9e7620e | 1796 | |
8184f93e | 1797 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1798 | BUG_ON(!cfqd->busy_queues); |
1799 | cfqd->busy_queues--; | |
ef8a41df SL |
1800 | if (cfq_cfqq_sync(cfqq)) |
1801 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
1802 | } |
1803 | ||
1804 | /* | |
1805 | * rb tree support functions | |
1806 | */ | |
febffd61 | 1807 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1808 | { |
5e705374 | 1809 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1810 | const int sync = rq_is_sync(rq); |
1da177e4 | 1811 | |
b4878f24 JA |
1812 | BUG_ON(!cfqq->queued[sync]); |
1813 | cfqq->queued[sync]--; | |
1da177e4 | 1814 | |
5e705374 | 1815 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1816 | |
f04a6424 VG |
1817 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1818 | /* | |
1819 | * Queue will be deleted from service tree when we actually | |
1820 | * expire it later. Right now just remove it from prio tree | |
1821 | * as it is empty. | |
1822 | */ | |
1823 | if (cfqq->p_root) { | |
1824 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1825 | cfqq->p_root = NULL; | |
1826 | } | |
1827 | } | |
1da177e4 LT |
1828 | } |
1829 | ||
5e705374 | 1830 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1831 | { |
5e705374 | 1832 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1833 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 1834 | struct request *prev; |
1da177e4 | 1835 | |
5380a101 | 1836 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 1837 | |
796d5116 | 1838 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
1839 | |
1840 | if (!cfq_cfqq_on_rr(cfqq)) | |
1841 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1842 | |
1843 | /* | |
1844 | * check if this request is a better next-serve candidate | |
1845 | */ | |
a36e71f9 | 1846 | prev = cfqq->next_rq; |
cf7c25cf | 1847 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1848 | |
1849 | /* | |
1850 | * adjust priority tree position, if ->next_rq changes | |
1851 | */ | |
1852 | if (prev != cfqq->next_rq) | |
1853 | cfq_prio_tree_add(cfqd, cfqq); | |
1854 | ||
5044eed4 | 1855 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1856 | } |
1857 | ||
febffd61 | 1858 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1859 | { |
5380a101 JA |
1860 | elv_rb_del(&cfqq->sort_list, rq); |
1861 | cfqq->queued[rq_is_sync(rq)]--; | |
155fead9 | 1862 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
5e705374 | 1863 | cfq_add_rq_rb(rq); |
155fead9 TH |
1864 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
1865 | rq->cmd_flags); | |
1da177e4 LT |
1866 | } |
1867 | ||
206dc69b JA |
1868 | static struct request * |
1869 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1870 | { |
206dc69b | 1871 | struct task_struct *tsk = current; |
c5869807 | 1872 | struct cfq_io_cq *cic; |
206dc69b | 1873 | struct cfq_queue *cfqq; |
1da177e4 | 1874 | |
4ac845a2 | 1875 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1876 | if (!cic) |
1877 | return NULL; | |
1878 | ||
1879 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
1880 | if (cfqq) { |
1881 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
1882 | ||
21183b07 | 1883 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 1884 | } |
1da177e4 | 1885 | |
1da177e4 LT |
1886 | return NULL; |
1887 | } | |
1888 | ||
165125e1 | 1889 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1890 | { |
22e2c507 | 1891 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 1892 | |
53c583d2 | 1893 | cfqd->rq_in_driver++; |
7b679138 | 1894 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 1895 | cfqd->rq_in_driver); |
25776e35 | 1896 | |
5b93629b | 1897 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
1898 | } |
1899 | ||
165125e1 | 1900 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1901 | { |
b4878f24 JA |
1902 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1903 | ||
53c583d2 CZ |
1904 | WARN_ON(!cfqd->rq_in_driver); |
1905 | cfqd->rq_in_driver--; | |
7b679138 | 1906 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 1907 | cfqd->rq_in_driver); |
1da177e4 LT |
1908 | } |
1909 | ||
b4878f24 | 1910 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 1911 | { |
5e705374 | 1912 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 1913 | |
5e705374 JA |
1914 | if (cfqq->next_rq == rq) |
1915 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 1916 | |
b4878f24 | 1917 | list_del_init(&rq->queuelist); |
5e705374 | 1918 | cfq_del_rq_rb(rq); |
374f84ac | 1919 | |
45333d5a | 1920 | cfqq->cfqd->rq_queued--; |
155fead9 | 1921 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
65299a3b CH |
1922 | if (rq->cmd_flags & REQ_PRIO) { |
1923 | WARN_ON(!cfqq->prio_pending); | |
1924 | cfqq->prio_pending--; | |
b53d1ed7 | 1925 | } |
1da177e4 LT |
1926 | } |
1927 | ||
165125e1 JA |
1928 | static int cfq_merge(struct request_queue *q, struct request **req, |
1929 | struct bio *bio) | |
1da177e4 LT |
1930 | { |
1931 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1932 | struct request *__rq; | |
1da177e4 | 1933 | |
206dc69b | 1934 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 1935 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
1936 | *req = __rq; |
1937 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
1938 | } |
1939 | ||
1940 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
1941 | } |
1942 | ||
165125e1 | 1943 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 1944 | int type) |
1da177e4 | 1945 | { |
21183b07 | 1946 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 1947 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 1948 | |
5e705374 | 1949 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 1950 | } |
1da177e4 LT |
1951 | } |
1952 | ||
812d4026 DS |
1953 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
1954 | struct bio *bio) | |
1955 | { | |
155fead9 | 1956 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_rw); |
812d4026 DS |
1957 | } |
1958 | ||
1da177e4 | 1959 | static void |
165125e1 | 1960 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
1961 | struct request *next) |
1962 | { | |
cf7c25cf | 1963 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
1964 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1965 | ||
22e2c507 JA |
1966 | /* |
1967 | * reposition in fifo if next is older than rq | |
1968 | */ | |
1969 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
3d106fba SL |
1970 | time_before(rq_fifo_time(next), rq_fifo_time(rq)) && |
1971 | cfqq == RQ_CFQQ(next)) { | |
22e2c507 | 1972 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
1973 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1974 | } | |
22e2c507 | 1975 | |
cf7c25cf CZ |
1976 | if (cfqq->next_rq == next) |
1977 | cfqq->next_rq = rq; | |
b4878f24 | 1978 | cfq_remove_request(next); |
155fead9 | 1979 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
4a0b75c7 SL |
1980 | |
1981 | cfqq = RQ_CFQQ(next); | |
1982 | /* | |
1983 | * all requests of this queue are merged to other queues, delete it | |
1984 | * from the service tree. If it's the active_queue, | |
1985 | * cfq_dispatch_requests() will choose to expire it or do idle | |
1986 | */ | |
1987 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
1988 | cfqq != cfqd->active_queue) | |
1989 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
1990 | } |
1991 | ||
165125e1 | 1992 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
1993 | struct bio *bio) |
1994 | { | |
1995 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 1996 | struct cfq_io_cq *cic; |
da775265 | 1997 | struct cfq_queue *cfqq; |
da775265 JA |
1998 | |
1999 | /* | |
ec8acb69 | 2000 | * Disallow merge of a sync bio into an async request. |
da775265 | 2001 | */ |
91fac317 | 2002 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 2003 | return false; |
da775265 JA |
2004 | |
2005 | /* | |
f1a4f4d3 | 2006 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 2007 | * merge only if rq is queued there. |
f1a4f4d3 | 2008 | */ |
07c2bd37 TH |
2009 | cic = cfq_cic_lookup(cfqd, current->io_context); |
2010 | if (!cic) | |
2011 | return false; | |
719d3402 | 2012 | |
91fac317 | 2013 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 2014 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
2015 | } |
2016 | ||
812df48d DS |
2017 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2018 | { | |
2019 | del_timer(&cfqd->idle_slice_timer); | |
155fead9 | 2020 | cfqg_stats_update_idle_time(cfqq->cfqg); |
812df48d DS |
2021 | } |
2022 | ||
febffd61 JA |
2023 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
2024 | struct cfq_queue *cfqq) | |
22e2c507 JA |
2025 | { |
2026 | if (cfqq) { | |
3bf10fea | 2027 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d", |
4d2ceea4 | 2028 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
155fead9 | 2029 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
62a37f6b JT |
2030 | cfqq->slice_start = 0; |
2031 | cfqq->dispatch_start = jiffies; | |
2032 | cfqq->allocated_slice = 0; | |
2033 | cfqq->slice_end = 0; | |
2034 | cfqq->slice_dispatch = 0; | |
2035 | cfqq->nr_sectors = 0; | |
2036 | ||
2037 | cfq_clear_cfqq_wait_request(cfqq); | |
2038 | cfq_clear_cfqq_must_dispatch(cfqq); | |
2039 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
2040 | cfq_clear_cfqq_fifo_expire(cfqq); | |
2041 | cfq_mark_cfqq_slice_new(cfqq); | |
2042 | ||
2043 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
2044 | } |
2045 | ||
2046 | cfqd->active_queue = cfqq; | |
2047 | } | |
2048 | ||
7b14e3b5 JA |
2049 | /* |
2050 | * current cfqq expired its slice (or was too idle), select new one | |
2051 | */ | |
2052 | static void | |
2053 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 2054 | bool timed_out) |
7b14e3b5 | 2055 | { |
7b679138 JA |
2056 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
2057 | ||
7b14e3b5 | 2058 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 2059 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 2060 | |
7b14e3b5 | 2061 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 2062 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 2063 | |
ae54abed SL |
2064 | /* |
2065 | * If this cfqq is shared between multiple processes, check to | |
2066 | * make sure that those processes are still issuing I/Os within | |
2067 | * the mean seek distance. If not, it may be time to break the | |
2068 | * queues apart again. | |
2069 | */ | |
2070 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
2071 | cfq_mark_cfqq_split_coop(cfqq); | |
2072 | ||
7b14e3b5 | 2073 | /* |
6084cdda | 2074 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 2075 | */ |
c553f8e3 SL |
2076 | if (timed_out) { |
2077 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 2078 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
2079 | else |
2080 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
2081 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
2082 | } | |
7b14e3b5 | 2083 | |
e5ff082e | 2084 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 2085 | |
f04a6424 VG |
2086 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
2087 | cfq_del_cfqq_rr(cfqd, cfqq); | |
2088 | ||
edd75ffd | 2089 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
2090 | |
2091 | if (cfqq == cfqd->active_queue) | |
2092 | cfqd->active_queue = NULL; | |
2093 | ||
2094 | if (cfqd->active_cic) { | |
11a3122f | 2095 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
2096 | cfqd->active_cic = NULL; |
2097 | } | |
7b14e3b5 JA |
2098 | } |
2099 | ||
e5ff082e | 2100 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
2101 | { |
2102 | struct cfq_queue *cfqq = cfqd->active_queue; | |
2103 | ||
2104 | if (cfqq) | |
e5ff082e | 2105 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
2106 | } |
2107 | ||
498d3aa2 JA |
2108 | /* |
2109 | * Get next queue for service. Unless we have a queue preemption, | |
2110 | * we'll simply select the first cfqq in the service tree. | |
2111 | */ | |
6d048f53 | 2112 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 2113 | { |
34b98d03 VG |
2114 | struct cfq_rb_root *st = st_for(cfqd->serving_group, |
2115 | cfqd->serving_wl_class, cfqd->serving_wl_type); | |
d9e7620e | 2116 | |
f04a6424 VG |
2117 | if (!cfqd->rq_queued) |
2118 | return NULL; | |
2119 | ||
1fa8f6d6 | 2120 | /* There is nothing to dispatch */ |
34b98d03 | 2121 | if (!st) |
1fa8f6d6 | 2122 | return NULL; |
34b98d03 | 2123 | if (RB_EMPTY_ROOT(&st->rb)) |
c0324a02 | 2124 | return NULL; |
34b98d03 | 2125 | return cfq_rb_first(st); |
6d048f53 JA |
2126 | } |
2127 | ||
f04a6424 VG |
2128 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
2129 | { | |
25fb5169 | 2130 | struct cfq_group *cfqg; |
f04a6424 VG |
2131 | struct cfq_queue *cfqq; |
2132 | int i, j; | |
2133 | struct cfq_rb_root *st; | |
2134 | ||
2135 | if (!cfqd->rq_queued) | |
2136 | return NULL; | |
2137 | ||
25fb5169 VG |
2138 | cfqg = cfq_get_next_cfqg(cfqd); |
2139 | if (!cfqg) | |
2140 | return NULL; | |
2141 | ||
f04a6424 VG |
2142 | for_each_cfqg_st(cfqg, i, j, st) |
2143 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
2144 | return cfqq; | |
2145 | return NULL; | |
2146 | } | |
2147 | ||
498d3aa2 JA |
2148 | /* |
2149 | * Get and set a new active queue for service. | |
2150 | */ | |
a36e71f9 JA |
2151 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
2152 | struct cfq_queue *cfqq) | |
6d048f53 | 2153 | { |
e00ef799 | 2154 | if (!cfqq) |
a36e71f9 | 2155 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 2156 | |
22e2c507 | 2157 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 2158 | return cfqq; |
22e2c507 JA |
2159 | } |
2160 | ||
d9e7620e JA |
2161 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
2162 | struct request *rq) | |
2163 | { | |
83096ebf TH |
2164 | if (blk_rq_pos(rq) >= cfqd->last_position) |
2165 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 2166 | else |
83096ebf | 2167 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
2168 | } |
2169 | ||
b2c18e1e | 2170 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 2171 | struct request *rq) |
6d048f53 | 2172 | { |
e9ce335d | 2173 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
2174 | } |
2175 | ||
a36e71f9 JA |
2176 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
2177 | struct cfq_queue *cur_cfqq) | |
2178 | { | |
f2d1f0ae | 2179 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
2180 | struct rb_node *parent, *node; |
2181 | struct cfq_queue *__cfqq; | |
2182 | sector_t sector = cfqd->last_position; | |
2183 | ||
2184 | if (RB_EMPTY_ROOT(root)) | |
2185 | return NULL; | |
2186 | ||
2187 | /* | |
2188 | * First, if we find a request starting at the end of the last | |
2189 | * request, choose it. | |
2190 | */ | |
f2d1f0ae | 2191 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
2192 | if (__cfqq) |
2193 | return __cfqq; | |
2194 | ||
2195 | /* | |
2196 | * If the exact sector wasn't found, the parent of the NULL leaf | |
2197 | * will contain the closest sector. | |
2198 | */ | |
2199 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 2200 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2201 | return __cfqq; |
2202 | ||
2e46e8b2 | 2203 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
2204 | node = rb_next(&__cfqq->p_node); |
2205 | else | |
2206 | node = rb_prev(&__cfqq->p_node); | |
2207 | if (!node) | |
2208 | return NULL; | |
2209 | ||
2210 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 2211 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2212 | return __cfqq; |
2213 | ||
2214 | return NULL; | |
2215 | } | |
2216 | ||
2217 | /* | |
2218 | * cfqd - obvious | |
2219 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
2220 | * closely cooperating with itself. | |
2221 | * | |
2222 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
2223 | * one request, and that cfqd->last_position reflects a position on the disk | |
2224 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
2225 | * assumption. | |
2226 | */ | |
2227 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 2228 | struct cfq_queue *cur_cfqq) |
6d048f53 | 2229 | { |
a36e71f9 JA |
2230 | struct cfq_queue *cfqq; |
2231 | ||
39c01b21 DS |
2232 | if (cfq_class_idle(cur_cfqq)) |
2233 | return NULL; | |
e6c5bc73 JM |
2234 | if (!cfq_cfqq_sync(cur_cfqq)) |
2235 | return NULL; | |
2236 | if (CFQQ_SEEKY(cur_cfqq)) | |
2237 | return NULL; | |
2238 | ||
b9d8f4c7 GJ |
2239 | /* |
2240 | * Don't search priority tree if it's the only queue in the group. | |
2241 | */ | |
2242 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
2243 | return NULL; | |
2244 | ||
6d048f53 | 2245 | /* |
d9e7620e JA |
2246 | * We should notice if some of the queues are cooperating, eg |
2247 | * working closely on the same area of the disk. In that case, | |
2248 | * we can group them together and don't waste time idling. | |
6d048f53 | 2249 | */ |
a36e71f9 JA |
2250 | cfqq = cfqq_close(cfqd, cur_cfqq); |
2251 | if (!cfqq) | |
2252 | return NULL; | |
2253 | ||
8682e1f1 VG |
2254 | /* If new queue belongs to different cfq_group, don't choose it */ |
2255 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
2256 | return NULL; | |
2257 | ||
df5fe3e8 JM |
2258 | /* |
2259 | * It only makes sense to merge sync queues. | |
2260 | */ | |
2261 | if (!cfq_cfqq_sync(cfqq)) | |
2262 | return NULL; | |
e6c5bc73 JM |
2263 | if (CFQQ_SEEKY(cfqq)) |
2264 | return NULL; | |
df5fe3e8 | 2265 | |
c0324a02 CZ |
2266 | /* |
2267 | * Do not merge queues of different priority classes | |
2268 | */ | |
2269 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
2270 | return NULL; | |
2271 | ||
a36e71f9 | 2272 | return cfqq; |
6d048f53 JA |
2273 | } |
2274 | ||
a6d44e98 CZ |
2275 | /* |
2276 | * Determine whether we should enforce idle window for this queue. | |
2277 | */ | |
2278 | ||
2279 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2280 | { | |
3bf10fea | 2281 | enum wl_class_t wl_class = cfqq_class(cfqq); |
34b98d03 | 2282 | struct cfq_rb_root *st = cfqq->service_tree; |
a6d44e98 | 2283 | |
34b98d03 VG |
2284 | BUG_ON(!st); |
2285 | BUG_ON(!st->count); | |
f04a6424 | 2286 | |
b6508c16 VG |
2287 | if (!cfqd->cfq_slice_idle) |
2288 | return false; | |
2289 | ||
a6d44e98 | 2290 | /* We never do for idle class queues. */ |
3bf10fea | 2291 | if (wl_class == IDLE_WORKLOAD) |
a6d44e98 CZ |
2292 | return false; |
2293 | ||
2294 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
2295 | if (cfq_cfqq_idle_window(cfqq) && |
2296 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
2297 | return true; |
2298 | ||
2299 | /* | |
2300 | * Otherwise, we do only if they are the last ones | |
2301 | * in their service tree. | |
2302 | */ | |
34b98d03 VG |
2303 | if (st->count == 1 && cfq_cfqq_sync(cfqq) && |
2304 | !cfq_io_thinktime_big(cfqd, &st->ttime, false)) | |
c1e44756 | 2305 | return true; |
34b98d03 | 2306 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count); |
c1e44756 | 2307 | return false; |
a6d44e98 CZ |
2308 | } |
2309 | ||
6d048f53 | 2310 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2311 | { |
1792669c | 2312 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 2313 | struct cfq_io_cq *cic; |
80bdf0c7 | 2314 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 2315 | |
a68bbddb | 2316 | /* |
f7d7b7a7 JA |
2317 | * SSD device without seek penalty, disable idling. But only do so |
2318 | * for devices that support queuing, otherwise we still have a problem | |
2319 | * with sync vs async workloads. | |
a68bbddb | 2320 | */ |
f7d7b7a7 | 2321 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
2322 | return; |
2323 | ||
dd67d051 | 2324 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2325 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2326 | |
2327 | /* | |
2328 | * idle is disabled, either manually or by past process history | |
2329 | */ | |
80bdf0c7 VG |
2330 | if (!cfq_should_idle(cfqd, cfqq)) { |
2331 | /* no queue idling. Check for group idling */ | |
2332 | if (cfqd->cfq_group_idle) | |
2333 | group_idle = cfqd->cfq_group_idle; | |
2334 | else | |
2335 | return; | |
2336 | } | |
6d048f53 | 2337 | |
7b679138 | 2338 | /* |
8e550632 | 2339 | * still active requests from this queue, don't idle |
7b679138 | 2340 | */ |
8e550632 | 2341 | if (cfqq->dispatched) |
7b679138 JA |
2342 | return; |
2343 | ||
22e2c507 JA |
2344 | /* |
2345 | * task has exited, don't wait | |
2346 | */ | |
206dc69b | 2347 | cic = cfqd->active_cic; |
f6e8d01b | 2348 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
2349 | return; |
2350 | ||
355b659c CZ |
2351 | /* |
2352 | * If our average think time is larger than the remaining time | |
2353 | * slice, then don't idle. This avoids overrunning the allotted | |
2354 | * time slice. | |
2355 | */ | |
383cd721 SL |
2356 | if (sample_valid(cic->ttime.ttime_samples) && |
2357 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 2358 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 2359 | cic->ttime.ttime_mean); |
355b659c | 2360 | return; |
b1ffe737 | 2361 | } |
355b659c | 2362 | |
80bdf0c7 VG |
2363 | /* There are other queues in the group, don't do group idle */ |
2364 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
2365 | return; | |
2366 | ||
3b18152c | 2367 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2368 | |
80bdf0c7 VG |
2369 | if (group_idle) |
2370 | sl = cfqd->cfq_group_idle; | |
2371 | else | |
2372 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2373 | |
7b14e3b5 | 2374 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
155fead9 | 2375 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
80bdf0c7 VG |
2376 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
2377 | group_idle ? 1 : 0); | |
1da177e4 LT |
2378 | } |
2379 | ||
498d3aa2 JA |
2380 | /* |
2381 | * Move request from internal lists to the request queue dispatch list. | |
2382 | */ | |
165125e1 | 2383 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2384 | { |
3ed9a296 | 2385 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2386 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2387 | |
7b679138 JA |
2388 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2389 | ||
06d21886 | 2390 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2391 | cfq_remove_request(rq); |
6d048f53 | 2392 | cfqq->dispatched++; |
80bdf0c7 | 2393 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2394 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2395 | |
53c583d2 | 2396 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2397 | cfqq->nr_sectors += blk_rq_sectors(rq); |
155fead9 | 2398 | cfqg_stats_update_dispatch(cfqq->cfqg, blk_rq_bytes(rq), rq->cmd_flags); |
1da177e4 LT |
2399 | } |
2400 | ||
2401 | /* | |
2402 | * return expired entry, or NULL to just start from scratch in rbtree | |
2403 | */ | |
febffd61 | 2404 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 2405 | { |
30996f40 | 2406 | struct request *rq = NULL; |
1da177e4 | 2407 | |
3b18152c | 2408 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 2409 | return NULL; |
cb887411 JA |
2410 | |
2411 | cfq_mark_cfqq_fifo_expire(cfqq); | |
2412 | ||
89850f7e JA |
2413 | if (list_empty(&cfqq->fifo)) |
2414 | return NULL; | |
1da177e4 | 2415 | |
89850f7e | 2416 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 2417 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 2418 | rq = NULL; |
1da177e4 | 2419 | |
30996f40 | 2420 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 2421 | return rq; |
1da177e4 LT |
2422 | } |
2423 | ||
22e2c507 JA |
2424 | static inline int |
2425 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2426 | { | |
2427 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2428 | |
22e2c507 | 2429 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2430 | |
b9f8ce05 | 2431 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
2432 | } |
2433 | ||
df5fe3e8 JM |
2434 | /* |
2435 | * Must be called with the queue_lock held. | |
2436 | */ | |
2437 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2438 | { | |
2439 | int process_refs, io_refs; | |
2440 | ||
2441 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 2442 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
2443 | BUG_ON(process_refs < 0); |
2444 | return process_refs; | |
2445 | } | |
2446 | ||
2447 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2448 | { | |
e6c5bc73 | 2449 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2450 | struct cfq_queue *__cfqq; |
2451 | ||
c10b61f0 JM |
2452 | /* |
2453 | * If there are no process references on the new_cfqq, then it is | |
2454 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2455 | * chain may have dropped their last reference (not just their | |
2456 | * last process reference). | |
2457 | */ | |
2458 | if (!cfqq_process_refs(new_cfqq)) | |
2459 | return; | |
2460 | ||
df5fe3e8 JM |
2461 | /* Avoid a circular list and skip interim queue merges */ |
2462 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2463 | if (__cfqq == cfqq) | |
2464 | return; | |
2465 | new_cfqq = __cfqq; | |
2466 | } | |
2467 | ||
2468 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2469 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2470 | /* |
2471 | * If the process for the cfqq has gone away, there is no | |
2472 | * sense in merging the queues. | |
2473 | */ | |
c10b61f0 | 2474 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2475 | return; |
2476 | ||
e6c5bc73 JM |
2477 | /* |
2478 | * Merge in the direction of the lesser amount of work. | |
2479 | */ | |
e6c5bc73 JM |
2480 | if (new_process_refs >= process_refs) { |
2481 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2482 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2483 | } else { |
2484 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2485 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2486 | } |
df5fe3e8 JM |
2487 | } |
2488 | ||
6d816ec7 | 2489 | static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd, |
3bf10fea | 2490 | struct cfq_group *cfqg, enum wl_class_t wl_class) |
718eee05 CZ |
2491 | { |
2492 | struct cfq_queue *queue; | |
2493 | int i; | |
2494 | bool key_valid = false; | |
2495 | unsigned long lowest_key = 0; | |
2496 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2497 | ||
65b32a57 VG |
2498 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2499 | /* select the one with lowest rb_key */ | |
34b98d03 | 2500 | queue = cfq_rb_first(st_for(cfqg, wl_class, i)); |
718eee05 CZ |
2501 | if (queue && |
2502 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2503 | lowest_key = queue->rb_key; | |
2504 | cur_best = i; | |
2505 | key_valid = true; | |
2506 | } | |
2507 | } | |
2508 | ||
2509 | return cur_best; | |
2510 | } | |
2511 | ||
6d816ec7 VG |
2512 | static void |
2513 | choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
718eee05 | 2514 | { |
718eee05 CZ |
2515 | unsigned slice; |
2516 | unsigned count; | |
cdb16e8f | 2517 | struct cfq_rb_root *st; |
58ff82f3 | 2518 | unsigned group_slice; |
4d2ceea4 | 2519 | enum wl_class_t original_class = cfqd->serving_wl_class; |
1fa8f6d6 | 2520 | |
718eee05 | 2521 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2522 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 2523 | cfqd->serving_wl_class = RT_WORKLOAD; |
58ff82f3 | 2524 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 2525 | cfqd->serving_wl_class = BE_WORKLOAD; |
718eee05 | 2526 | else { |
4d2ceea4 | 2527 | cfqd->serving_wl_class = IDLE_WORKLOAD; |
718eee05 CZ |
2528 | cfqd->workload_expires = jiffies + 1; |
2529 | return; | |
2530 | } | |
2531 | ||
4d2ceea4 | 2532 | if (original_class != cfqd->serving_wl_class) |
e4ea0c16 SL |
2533 | goto new_workload; |
2534 | ||
718eee05 CZ |
2535 | /* |
2536 | * For RT and BE, we have to choose also the type | |
2537 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2538 | * expiration time | |
2539 | */ | |
34b98d03 | 2540 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 2541 | count = st->count; |
718eee05 CZ |
2542 | |
2543 | /* | |
65b32a57 | 2544 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2545 | */ |
65b32a57 | 2546 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2547 | return; |
2548 | ||
e4ea0c16 | 2549 | new_workload: |
718eee05 | 2550 | /* otherwise select new workload type */ |
6d816ec7 | 2551 | cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg, |
4d2ceea4 | 2552 | cfqd->serving_wl_class); |
34b98d03 | 2553 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 2554 | count = st->count; |
718eee05 CZ |
2555 | |
2556 | /* | |
2557 | * the workload slice is computed as a fraction of target latency | |
2558 | * proportional to the number of queues in that workload, over | |
2559 | * all the queues in the same priority class | |
2560 | */ | |
58ff82f3 VG |
2561 | group_slice = cfq_group_slice(cfqd, cfqg); |
2562 | ||
2563 | slice = group_slice * count / | |
4d2ceea4 VG |
2564 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class], |
2565 | cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd, | |
3bf10fea | 2566 | cfqg)); |
718eee05 | 2567 | |
4d2ceea4 | 2568 | if (cfqd->serving_wl_type == ASYNC_WORKLOAD) { |
f26bd1f0 VG |
2569 | unsigned int tmp; |
2570 | ||
2571 | /* | |
2572 | * Async queues are currently system wide. Just taking | |
2573 | * proportion of queues with-in same group will lead to higher | |
2574 | * async ratio system wide as generally root group is going | |
2575 | * to have higher weight. A more accurate thing would be to | |
2576 | * calculate system wide asnc/sync ratio. | |
2577 | */ | |
5bf14c07 TM |
2578 | tmp = cfqd->cfq_target_latency * |
2579 | cfqg_busy_async_queues(cfqd, cfqg); | |
f26bd1f0 VG |
2580 | tmp = tmp/cfqd->busy_queues; |
2581 | slice = min_t(unsigned, slice, tmp); | |
2582 | ||
718eee05 CZ |
2583 | /* async workload slice is scaled down according to |
2584 | * the sync/async slice ratio. */ | |
2585 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2586 | } else |
718eee05 CZ |
2587 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2588 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2589 | ||
2590 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2591 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
2592 | cfqd->workload_expires = jiffies + slice; |
2593 | } | |
2594 | ||
1fa8f6d6 VG |
2595 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2596 | { | |
2597 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2598 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2599 | |
2600 | if (RB_EMPTY_ROOT(&st->rb)) | |
2601 | return NULL; | |
25bc6b07 | 2602 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
2603 | update_min_vdisktime(st); |
2604 | return cfqg; | |
1fa8f6d6 VG |
2605 | } |
2606 | ||
cdb16e8f VG |
2607 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2608 | { | |
1fa8f6d6 VG |
2609 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2610 | ||
2611 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2612 | |
2613 | /* Restore the workload type data */ | |
4d2ceea4 VG |
2614 | if (cfqg->saved_wl_slice) { |
2615 | cfqd->workload_expires = jiffies + cfqg->saved_wl_slice; | |
2616 | cfqd->serving_wl_type = cfqg->saved_wl_type; | |
2617 | cfqd->serving_wl_class = cfqg->saved_wl_class; | |
66ae2919 GJ |
2618 | } else |
2619 | cfqd->workload_expires = jiffies - 1; | |
2620 | ||
6d816ec7 | 2621 | choose_wl_class_and_type(cfqd, cfqg); |
cdb16e8f VG |
2622 | } |
2623 | ||
22e2c507 | 2624 | /* |
498d3aa2 JA |
2625 | * Select a queue for service. If we have a current active queue, |
2626 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2627 | */ |
1b5ed5e1 | 2628 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2629 | { |
a36e71f9 | 2630 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2631 | |
22e2c507 JA |
2632 | cfqq = cfqd->active_queue; |
2633 | if (!cfqq) | |
2634 | goto new_queue; | |
1da177e4 | 2635 | |
f04a6424 VG |
2636 | if (!cfqd->rq_queued) |
2637 | return NULL; | |
c244bb50 VG |
2638 | |
2639 | /* | |
2640 | * We were waiting for group to get backlogged. Expire the queue | |
2641 | */ | |
2642 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2643 | goto expire; | |
2644 | ||
22e2c507 | 2645 | /* |
6d048f53 | 2646 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2647 | */ |
7667aa06 VG |
2648 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2649 | /* | |
2650 | * If slice had not expired at the completion of last request | |
2651 | * we might not have turned on wait_busy flag. Don't expire | |
2652 | * the queue yet. Allow the group to get backlogged. | |
2653 | * | |
2654 | * The very fact that we have used the slice, that means we | |
2655 | * have been idling all along on this queue and it should be | |
2656 | * ok to wait for this request to complete. | |
2657 | */ | |
82bbbf28 VG |
2658 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2659 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2660 | cfqq = NULL; | |
7667aa06 | 2661 | goto keep_queue; |
82bbbf28 | 2662 | } else |
80bdf0c7 | 2663 | goto check_group_idle; |
7667aa06 | 2664 | } |
1da177e4 | 2665 | |
22e2c507 | 2666 | /* |
6d048f53 JA |
2667 | * The active queue has requests and isn't expired, allow it to |
2668 | * dispatch. | |
22e2c507 | 2669 | */ |
dd67d051 | 2670 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2671 | goto keep_queue; |
6d048f53 | 2672 | |
a36e71f9 JA |
2673 | /* |
2674 | * If another queue has a request waiting within our mean seek | |
2675 | * distance, let it run. The expire code will check for close | |
2676 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2677 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2678 | */ |
b3b6d040 | 2679 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2680 | if (new_cfqq) { |
2681 | if (!cfqq->new_cfqq) | |
2682 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2683 | goto expire; |
df5fe3e8 | 2684 | } |
a36e71f9 | 2685 | |
6d048f53 JA |
2686 | /* |
2687 | * No requests pending. If the active queue still has requests in | |
2688 | * flight or is idling for a new request, allow either of these | |
2689 | * conditions to happen (or time out) before selecting a new queue. | |
2690 | */ | |
80bdf0c7 VG |
2691 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2692 | cfqq = NULL; | |
2693 | goto keep_queue; | |
2694 | } | |
2695 | ||
8e1ac665 SL |
2696 | /* |
2697 | * This is a deep seek queue, but the device is much faster than | |
2698 | * the queue can deliver, don't idle | |
2699 | **/ | |
2700 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
2701 | (cfq_cfqq_slice_new(cfqq) || | |
2702 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
2703 | cfq_clear_cfqq_deep(cfqq); | |
2704 | cfq_clear_cfqq_idle_window(cfqq); | |
2705 | } | |
2706 | ||
80bdf0c7 VG |
2707 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2708 | cfqq = NULL; | |
2709 | goto keep_queue; | |
2710 | } | |
2711 | ||
2712 | /* | |
2713 | * If group idle is enabled and there are requests dispatched from | |
2714 | * this group, wait for requests to complete. | |
2715 | */ | |
2716 | check_group_idle: | |
7700fc4f SL |
2717 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
2718 | cfqq->cfqg->dispatched && | |
2719 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
2720 | cfqq = NULL; |
2721 | goto keep_queue; | |
22e2c507 JA |
2722 | } |
2723 | ||
3b18152c | 2724 | expire: |
e5ff082e | 2725 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2726 | new_queue: |
718eee05 CZ |
2727 | /* |
2728 | * Current queue expired. Check if we have to switch to a new | |
2729 | * service tree | |
2730 | */ | |
2731 | if (!new_cfqq) | |
cdb16e8f | 2732 | cfq_choose_cfqg(cfqd); |
718eee05 | 2733 | |
a36e71f9 | 2734 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2735 | keep_queue: |
3b18152c | 2736 | return cfqq; |
22e2c507 JA |
2737 | } |
2738 | ||
febffd61 | 2739 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2740 | { |
2741 | int dispatched = 0; | |
2742 | ||
2743 | while (cfqq->next_rq) { | |
2744 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2745 | dispatched++; | |
2746 | } | |
2747 | ||
2748 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2749 | |
2750 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 2751 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
2752 | return dispatched; |
2753 | } | |
2754 | ||
498d3aa2 JA |
2755 | /* |
2756 | * Drain our current requests. Used for barriers and when switching | |
2757 | * io schedulers on-the-fly. | |
2758 | */ | |
d9e7620e | 2759 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2760 | { |
0871714e | 2761 | struct cfq_queue *cfqq; |
d9e7620e | 2762 | int dispatched = 0; |
cdb16e8f | 2763 | |
3440c49f | 2764 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 2765 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
2766 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2767 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 2768 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 2769 | } |
1b5ed5e1 | 2770 | |
1b5ed5e1 TH |
2771 | BUG_ON(cfqd->busy_queues); |
2772 | ||
6923715a | 2773 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2774 | return dispatched; |
2775 | } | |
2776 | ||
abc3c744 SL |
2777 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2778 | struct cfq_queue *cfqq) | |
2779 | { | |
2780 | /* the queue hasn't finished any request, can't estimate */ | |
2781 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 2782 | return true; |
abc3c744 SL |
2783 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
2784 | cfqq->slice_end)) | |
c1e44756 | 2785 | return true; |
abc3c744 | 2786 | |
c1e44756 | 2787 | return false; |
abc3c744 SL |
2788 | } |
2789 | ||
0b182d61 | 2790 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2791 | { |
2f5cb738 | 2792 | unsigned int max_dispatch; |
22e2c507 | 2793 | |
5ad531db JA |
2794 | /* |
2795 | * Drain async requests before we start sync IO | |
2796 | */ | |
53c583d2 | 2797 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 2798 | return false; |
5ad531db | 2799 | |
2f5cb738 JA |
2800 | /* |
2801 | * If this is an async queue and we have sync IO in flight, let it wait | |
2802 | */ | |
53c583d2 | 2803 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 2804 | return false; |
2f5cb738 | 2805 | |
abc3c744 | 2806 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
2807 | if (cfq_class_idle(cfqq)) |
2808 | max_dispatch = 1; | |
b4878f24 | 2809 | |
2f5cb738 JA |
2810 | /* |
2811 | * Does this cfqq already have too much IO in flight? | |
2812 | */ | |
2813 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 2814 | bool promote_sync = false; |
2f5cb738 JA |
2815 | /* |
2816 | * idle queue must always only have a single IO in flight | |
2817 | */ | |
3ed9a296 | 2818 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2819 | return false; |
3ed9a296 | 2820 | |
ef8a41df | 2821 | /* |
c4ade94f LS |
2822 | * If there is only one sync queue |
2823 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
2824 | * queue no dispatch limit. The reason is a sync queue can |
2825 | * preempt async queue, limiting the sync queue doesn't make | |
2826 | * sense. This is useful for aiostress test. | |
2827 | */ | |
c4ade94f LS |
2828 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
2829 | promote_sync = true; | |
ef8a41df | 2830 | |
2f5cb738 JA |
2831 | /* |
2832 | * We have other queues, don't allow more IO from this one | |
2833 | */ | |
ef8a41df SL |
2834 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
2835 | !promote_sync) | |
0b182d61 | 2836 | return false; |
9ede209e | 2837 | |
365722bb | 2838 | /* |
474b18cc | 2839 | * Sole queue user, no limit |
365722bb | 2840 | */ |
ef8a41df | 2841 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
2842 | max_dispatch = -1; |
2843 | else | |
2844 | /* | |
2845 | * Normally we start throttling cfqq when cfq_quantum/2 | |
2846 | * requests have been dispatched. But we can drive | |
2847 | * deeper queue depths at the beginning of slice | |
2848 | * subjected to upper limit of cfq_quantum. | |
2849 | * */ | |
2850 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
2851 | } |
2852 | ||
2853 | /* | |
2854 | * Async queues must wait a bit before being allowed dispatch. | |
2855 | * We also ramp up the dispatch depth gradually for async IO, | |
2856 | * based on the last sync IO we serviced | |
2857 | */ | |
963b72fc | 2858 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 2859 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 2860 | unsigned int depth; |
365722bb | 2861 | |
61f0c1dc | 2862 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2863 | if (!depth && !cfqq->dispatched) |
2864 | depth = 1; | |
8e296755 JA |
2865 | if (depth < max_dispatch) |
2866 | max_dispatch = depth; | |
2f5cb738 | 2867 | } |
3ed9a296 | 2868 | |
0b182d61 JA |
2869 | /* |
2870 | * If we're below the current max, allow a dispatch | |
2871 | */ | |
2872 | return cfqq->dispatched < max_dispatch; | |
2873 | } | |
2874 | ||
2875 | /* | |
2876 | * Dispatch a request from cfqq, moving them to the request queue | |
2877 | * dispatch list. | |
2878 | */ | |
2879 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2880 | { | |
2881 | struct request *rq; | |
2882 | ||
2883 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
2884 | ||
2885 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
2886 | return false; | |
2887 | ||
2888 | /* | |
2889 | * follow expired path, else get first next available | |
2890 | */ | |
2891 | rq = cfq_check_fifo(cfqq); | |
2892 | if (!rq) | |
2893 | rq = cfqq->next_rq; | |
2894 | ||
2895 | /* | |
2896 | * insert request into driver dispatch list | |
2897 | */ | |
2898 | cfq_dispatch_insert(cfqd->queue, rq); | |
2899 | ||
2900 | if (!cfqd->active_cic) { | |
c5869807 | 2901 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 2902 | |
c5869807 | 2903 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
2904 | cfqd->active_cic = cic; |
2905 | } | |
2906 | ||
2907 | return true; | |
2908 | } | |
2909 | ||
2910 | /* | |
2911 | * Find the cfqq that we need to service and move a request from that to the | |
2912 | * dispatch list | |
2913 | */ | |
2914 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
2915 | { | |
2916 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2917 | struct cfq_queue *cfqq; | |
2918 | ||
2919 | if (!cfqd->busy_queues) | |
2920 | return 0; | |
2921 | ||
2922 | if (unlikely(force)) | |
2923 | return cfq_forced_dispatch(cfqd); | |
2924 | ||
2925 | cfqq = cfq_select_queue(cfqd); | |
2926 | if (!cfqq) | |
8e296755 JA |
2927 | return 0; |
2928 | ||
2f5cb738 | 2929 | /* |
0b182d61 | 2930 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 2931 | */ |
0b182d61 JA |
2932 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2933 | return 0; | |
2934 | ||
2f5cb738 | 2935 | cfqq->slice_dispatch++; |
b029195d | 2936 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 2937 | |
2f5cb738 JA |
2938 | /* |
2939 | * expire an async queue immediately if it has used up its slice. idle | |
2940 | * queue always expire after 1 dispatch round. | |
2941 | */ | |
2942 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
2943 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
2944 | cfq_class_idle(cfqq))) { | |
2945 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 2946 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
2947 | } |
2948 | ||
b217a903 | 2949 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 2950 | return 1; |
1da177e4 LT |
2951 | } |
2952 | ||
1da177e4 | 2953 | /* |
5e705374 JA |
2954 | * task holds one reference to the queue, dropped when task exits. each rq |
2955 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 2956 | * |
b1c35769 | 2957 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
2958 | * queue lock must be held here. |
2959 | */ | |
2960 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
2961 | { | |
22e2c507 | 2962 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 2963 | struct cfq_group *cfqg; |
22e2c507 | 2964 | |
30d7b944 | 2965 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 2966 | |
30d7b944 SL |
2967 | cfqq->ref--; |
2968 | if (cfqq->ref) | |
1da177e4 LT |
2969 | return; |
2970 | ||
7b679138 | 2971 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 2972 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 2973 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 2974 | cfqg = cfqq->cfqg; |
1da177e4 | 2975 | |
28f95cbc | 2976 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 2977 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2978 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2979 | } |
22e2c507 | 2980 | |
f04a6424 | 2981 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 2982 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 2983 | cfqg_put(cfqg); |
1da177e4 LT |
2984 | } |
2985 | ||
d02a2c07 | 2986 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 2987 | { |
df5fe3e8 JM |
2988 | struct cfq_queue *__cfqq, *next; |
2989 | ||
df5fe3e8 JM |
2990 | /* |
2991 | * If this queue was scheduled to merge with another queue, be | |
2992 | * sure to drop the reference taken on that queue (and others in | |
2993 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
2994 | */ | |
2995 | __cfqq = cfqq->new_cfqq; | |
2996 | while (__cfqq) { | |
2997 | if (__cfqq == cfqq) { | |
2998 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
2999 | break; | |
3000 | } | |
3001 | next = __cfqq->new_cfqq; | |
3002 | cfq_put_queue(__cfqq); | |
3003 | __cfqq = next; | |
3004 | } | |
d02a2c07 SL |
3005 | } |
3006 | ||
3007 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3008 | { | |
3009 | if (unlikely(cfqq == cfqd->active_queue)) { | |
3010 | __cfq_slice_expired(cfqd, cfqq, 0); | |
3011 | cfq_schedule_dispatch(cfqd); | |
3012 | } | |
3013 | ||
3014 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 3015 | |
89850f7e JA |
3016 | cfq_put_queue(cfqq); |
3017 | } | |
22e2c507 | 3018 | |
9b84cacd TH |
3019 | static void cfq_init_icq(struct io_cq *icq) |
3020 | { | |
3021 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
3022 | ||
3023 | cic->ttime.last_end_request = jiffies; | |
3024 | } | |
3025 | ||
c5869807 | 3026 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 3027 | { |
c5869807 | 3028 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 3029 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 3030 | |
ff6657c6 JA |
3031 | if (cic->cfqq[BLK_RW_ASYNC]) { |
3032 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
3033 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
3034 | } |
3035 | ||
ff6657c6 JA |
3036 | if (cic->cfqq[BLK_RW_SYNC]) { |
3037 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
3038 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 3039 | } |
89850f7e JA |
3040 | } |
3041 | ||
abede6da | 3042 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
3043 | { |
3044 | struct task_struct *tsk = current; | |
3045 | int ioprio_class; | |
3046 | ||
3b18152c | 3047 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
3048 | return; |
3049 | ||
598971bf | 3050 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 3051 | switch (ioprio_class) { |
fe094d98 JA |
3052 | default: |
3053 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
3054 | case IOPRIO_CLASS_NONE: | |
3055 | /* | |
6d63c275 | 3056 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
3057 | */ |
3058 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 3059 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
3060 | break; |
3061 | case IOPRIO_CLASS_RT: | |
598971bf | 3062 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3063 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
3064 | break; | |
3065 | case IOPRIO_CLASS_BE: | |
598971bf | 3066 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3067 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3068 | break; | |
3069 | case IOPRIO_CLASS_IDLE: | |
3070 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
3071 | cfqq->ioprio = 7; | |
3072 | cfq_clear_cfqq_idle_window(cfqq); | |
3073 | break; | |
22e2c507 JA |
3074 | } |
3075 | ||
3076 | /* | |
3077 | * keep track of original prio settings in case we have to temporarily | |
3078 | * elevate the priority of this queue | |
3079 | */ | |
3080 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 3081 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
3082 | } |
3083 | ||
598971bf | 3084 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 3085 | { |
598971bf | 3086 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 3087 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 3088 | struct cfq_queue *cfqq; |
35e6077c | 3089 | |
598971bf TH |
3090 | /* |
3091 | * Check whether ioprio has changed. The condition may trigger | |
3092 | * spuriously on a newly created cic but there's no harm. | |
3093 | */ | |
3094 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
3095 | return; |
3096 | ||
ff6657c6 | 3097 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
3098 | if (cfqq) { |
3099 | struct cfq_queue *new_cfqq; | |
abede6da TH |
3100 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio, |
3101 | GFP_ATOMIC); | |
caaa5f9f | 3102 | if (new_cfqq) { |
ff6657c6 | 3103 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
3104 | cfq_put_queue(cfqq); |
3105 | } | |
22e2c507 | 3106 | } |
caaa5f9f | 3107 | |
ff6657c6 | 3108 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
3109 | if (cfqq) |
3110 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
3111 | |
3112 | cic->ioprio = ioprio; | |
22e2c507 JA |
3113 | } |
3114 | ||
d5036d77 | 3115 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 3116 | pid_t pid, bool is_sync) |
d5036d77 JA |
3117 | { |
3118 | RB_CLEAR_NODE(&cfqq->rb_node); | |
3119 | RB_CLEAR_NODE(&cfqq->p_node); | |
3120 | INIT_LIST_HEAD(&cfqq->fifo); | |
3121 | ||
30d7b944 | 3122 | cfqq->ref = 0; |
d5036d77 JA |
3123 | cfqq->cfqd = cfqd; |
3124 | ||
3125 | cfq_mark_cfqq_prio_changed(cfqq); | |
3126 | ||
3127 | if (is_sync) { | |
3128 | if (!cfq_class_idle(cfqq)) | |
3129 | cfq_mark_cfqq_idle_window(cfqq); | |
3130 | cfq_mark_cfqq_sync(cfqq); | |
3131 | } | |
3132 | cfqq->pid = pid; | |
3133 | } | |
3134 | ||
24610333 | 3135 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
598971bf | 3136 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 3137 | { |
bca4b914 | 3138 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
598971bf TH |
3139 | struct cfq_queue *sync_cfqq; |
3140 | uint64_t id; | |
24610333 | 3141 | |
598971bf | 3142 | rcu_read_lock(); |
3c798398 | 3143 | id = bio_blkcg(bio)->id; |
598971bf | 3144 | rcu_read_unlock(); |
24610333 | 3145 | |
598971bf TH |
3146 | /* |
3147 | * Check whether blkcg has changed. The condition may trigger | |
3148 | * spuriously on a newly created cic but there's no harm. | |
3149 | */ | |
3150 | if (unlikely(!cfqd) || likely(cic->blkcg_id == id)) | |
3151 | return; | |
24610333 | 3152 | |
598971bf | 3153 | sync_cfqq = cic_to_cfqq(cic, 1); |
24610333 VG |
3154 | if (sync_cfqq) { |
3155 | /* | |
3156 | * Drop reference to sync queue. A new sync queue will be | |
3157 | * assigned in new group upon arrival of a fresh request. | |
3158 | */ | |
3159 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
3160 | cic_set_cfqq(cic, NULL, 1); | |
3161 | cfq_put_queue(sync_cfqq); | |
3162 | } | |
598971bf TH |
3163 | |
3164 | cic->blkcg_id = id; | |
24610333 | 3165 | } |
598971bf TH |
3166 | #else |
3167 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } | |
24610333 VG |
3168 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
3169 | ||
22e2c507 | 3170 | static struct cfq_queue * |
abede6da TH |
3171 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
3172 | struct bio *bio, gfp_t gfp_mask) | |
22e2c507 | 3173 | { |
3c798398 | 3174 | struct blkcg *blkcg; |
22e2c507 | 3175 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
cdb16e8f | 3176 | struct cfq_group *cfqg; |
22e2c507 JA |
3177 | |
3178 | retry: | |
2a7f1244 TH |
3179 | rcu_read_lock(); |
3180 | ||
3c798398 | 3181 | blkcg = bio_blkcg(bio); |
cd1604fa | 3182 | cfqg = cfq_lookup_create_cfqg(cfqd, blkcg); |
91fac317 | 3183 | cfqq = cic_to_cfqq(cic, is_sync); |
22e2c507 | 3184 | |
6118b70b JA |
3185 | /* |
3186 | * Always try a new alloc if we fell back to the OOM cfqq | |
3187 | * originally, since it should just be a temporary situation. | |
3188 | */ | |
3189 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
3190 | cfqq = NULL; | |
22e2c507 JA |
3191 | if (new_cfqq) { |
3192 | cfqq = new_cfqq; | |
3193 | new_cfqq = NULL; | |
3194 | } else if (gfp_mask & __GFP_WAIT) { | |
2a7f1244 | 3195 | rcu_read_unlock(); |
22e2c507 | 3196 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c | 3197 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 3198 | gfp_mask | __GFP_ZERO, |
94f6030c | 3199 | cfqd->queue->node); |
22e2c507 | 3200 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
3201 | if (new_cfqq) |
3202 | goto retry; | |
22e2c507 | 3203 | } else { |
94f6030c CL |
3204 | cfqq = kmem_cache_alloc_node(cfq_pool, |
3205 | gfp_mask | __GFP_ZERO, | |
3206 | cfqd->queue->node); | |
22e2c507 JA |
3207 | } |
3208 | ||
6118b70b JA |
3209 | if (cfqq) { |
3210 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
abede6da | 3211 | cfq_init_prio_data(cfqq, cic); |
cdb16e8f | 3212 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
3213 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
3214 | } else | |
3215 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
3216 | } |
3217 | ||
3218 | if (new_cfqq) | |
3219 | kmem_cache_free(cfq_pool, new_cfqq); | |
3220 | ||
2a7f1244 | 3221 | rcu_read_unlock(); |
22e2c507 JA |
3222 | return cfqq; |
3223 | } | |
3224 | ||
c2dea2d1 VT |
3225 | static struct cfq_queue ** |
3226 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
3227 | { | |
fe094d98 | 3228 | switch (ioprio_class) { |
c2dea2d1 VT |
3229 | case IOPRIO_CLASS_RT: |
3230 | return &cfqd->async_cfqq[0][ioprio]; | |
598971bf TH |
3231 | case IOPRIO_CLASS_NONE: |
3232 | ioprio = IOPRIO_NORM; | |
3233 | /* fall through */ | |
c2dea2d1 VT |
3234 | case IOPRIO_CLASS_BE: |
3235 | return &cfqd->async_cfqq[1][ioprio]; | |
3236 | case IOPRIO_CLASS_IDLE: | |
3237 | return &cfqd->async_idle_cfqq; | |
3238 | default: | |
3239 | BUG(); | |
3240 | } | |
3241 | } | |
3242 | ||
15c31be4 | 3243 | static struct cfq_queue * |
abede6da | 3244 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
4f85cb96 | 3245 | struct bio *bio, gfp_t gfp_mask) |
15c31be4 | 3246 | { |
598971bf TH |
3247 | const int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
3248 | const int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
c2dea2d1 | 3249 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
3250 | struct cfq_queue *cfqq = NULL; |
3251 | ||
c2dea2d1 VT |
3252 | if (!is_sync) { |
3253 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
3254 | cfqq = *async_cfqq; | |
3255 | } | |
3256 | ||
6118b70b | 3257 | if (!cfqq) |
abede6da | 3258 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, cic, bio, gfp_mask); |
15c31be4 JA |
3259 | |
3260 | /* | |
3261 | * pin the queue now that it's allocated, scheduler exit will prune it | |
3262 | */ | |
c2dea2d1 | 3263 | if (!is_sync && !(*async_cfqq)) { |
30d7b944 | 3264 | cfqq->ref++; |
c2dea2d1 | 3265 | *async_cfqq = cfqq; |
15c31be4 JA |
3266 | } |
3267 | ||
30d7b944 | 3268 | cfqq->ref++; |
15c31be4 JA |
3269 | return cfqq; |
3270 | } | |
3271 | ||
22e2c507 | 3272 | static void |
383cd721 | 3273 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 3274 | { |
383cd721 SL |
3275 | unsigned long elapsed = jiffies - ttime->last_end_request; |
3276 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 3277 | |
383cd721 SL |
3278 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
3279 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
3280 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
3281 | } | |
3282 | ||
3283 | static void | |
3284 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3285 | struct cfq_io_cq *cic) |
383cd721 | 3286 | { |
f5f2b6ce | 3287 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3288 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3289 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3290 | cfqd->cfq_slice_idle); | |
3291 | } | |
7700fc4f SL |
3292 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3293 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3294 | #endif | |
22e2c507 | 3295 | } |
1da177e4 | 3296 | |
206dc69b | 3297 | static void |
b2c18e1e | 3298 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3299 | struct request *rq) |
206dc69b | 3300 | { |
3dde36dd | 3301 | sector_t sdist = 0; |
41647e7a | 3302 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3303 | if (cfqq->last_request_pos) { |
3304 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3305 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3306 | else | |
3307 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3308 | } | |
206dc69b | 3309 | |
3dde36dd | 3310 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3311 | if (blk_queue_nonrot(cfqd->queue)) |
3312 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3313 | else | |
3314 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3315 | } |
1da177e4 | 3316 | |
22e2c507 JA |
3317 | /* |
3318 | * Disable idle window if the process thinks too long or seeks so much that | |
3319 | * it doesn't matter | |
3320 | */ | |
3321 | static void | |
3322 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3323 | struct cfq_io_cq *cic) |
22e2c507 | 3324 | { |
7b679138 | 3325 | int old_idle, enable_idle; |
1be92f2f | 3326 | |
0871714e JA |
3327 | /* |
3328 | * Don't idle for async or idle io prio class | |
3329 | */ | |
3330 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3331 | return; |
3332 | ||
c265a7f4 | 3333 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3334 | |
76280aff CZ |
3335 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3336 | cfq_mark_cfqq_deep(cfqq); | |
3337 | ||
749ef9f8 CZ |
3338 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
3339 | enable_idle = 0; | |
f6e8d01b | 3340 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
3341 | !cfqd->cfq_slice_idle || |
3342 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3343 | enable_idle = 0; |
383cd721 SL |
3344 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3345 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3346 | enable_idle = 0; |
3347 | else | |
3348 | enable_idle = 1; | |
1da177e4 LT |
3349 | } |
3350 | ||
7b679138 JA |
3351 | if (old_idle != enable_idle) { |
3352 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3353 | if (enable_idle) | |
3354 | cfq_mark_cfqq_idle_window(cfqq); | |
3355 | else | |
3356 | cfq_clear_cfqq_idle_window(cfqq); | |
3357 | } | |
22e2c507 | 3358 | } |
1da177e4 | 3359 | |
22e2c507 JA |
3360 | /* |
3361 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3362 | * no or if we aren't sure, a 1 will cause a preempt. | |
3363 | */ | |
a6151c3a | 3364 | static bool |
22e2c507 | 3365 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3366 | struct request *rq) |
22e2c507 | 3367 | { |
6d048f53 | 3368 | struct cfq_queue *cfqq; |
22e2c507 | 3369 | |
6d048f53 JA |
3370 | cfqq = cfqd->active_queue; |
3371 | if (!cfqq) | |
a6151c3a | 3372 | return false; |
22e2c507 | 3373 | |
6d048f53 | 3374 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3375 | return false; |
22e2c507 JA |
3376 | |
3377 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3378 | return true; |
1e3335de | 3379 | |
875feb63 DS |
3380 | /* |
3381 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3382 | */ | |
3383 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3384 | return false; | |
3385 | ||
374f84ac JA |
3386 | /* |
3387 | * if the new request is sync, but the currently running queue is | |
3388 | * not, let the sync request have priority. | |
3389 | */ | |
5e705374 | 3390 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3391 | return true; |
1e3335de | 3392 | |
8682e1f1 VG |
3393 | if (new_cfqq->cfqg != cfqq->cfqg) |
3394 | return false; | |
3395 | ||
3396 | if (cfq_slice_used(cfqq)) | |
3397 | return true; | |
3398 | ||
3399 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
4d2ceea4 | 3400 | if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 VG |
3401 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
3402 | new_cfqq->service_tree->count == 2 && | |
3403 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3404 | return true; | |
3405 | ||
b53d1ed7 JA |
3406 | /* |
3407 | * So both queues are sync. Let the new request get disk time if | |
3408 | * it's a metadata request and the current queue is doing regular IO. | |
3409 | */ | |
65299a3b | 3410 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
3411 | return true; |
3412 | ||
3a9a3f6c DS |
3413 | /* |
3414 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3415 | */ | |
3416 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3417 | return true; |
3a9a3f6c | 3418 | |
d2d59e18 SL |
3419 | /* An idle queue should not be idle now for some reason */ |
3420 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
3421 | return true; | |
3422 | ||
1e3335de | 3423 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3424 | return false; |
1e3335de JA |
3425 | |
3426 | /* | |
3427 | * if this request is as-good as one we would expect from the | |
3428 | * current cfqq, let it preempt | |
3429 | */ | |
e9ce335d | 3430 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3431 | return true; |
1e3335de | 3432 | |
a6151c3a | 3433 | return false; |
22e2c507 JA |
3434 | } |
3435 | ||
3436 | /* | |
3437 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3438 | * let it have half of its nominal slice. | |
3439 | */ | |
3440 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3441 | { | |
df0793ab SL |
3442 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
3443 | ||
7b679138 | 3444 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 3445 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3446 | |
f8ae6e3e SL |
3447 | /* |
3448 | * workload type is changed, don't save slice, otherwise preempt | |
3449 | * doesn't happen | |
3450 | */ | |
df0793ab | 3451 | if (old_type != cfqq_type(cfqq)) |
4d2ceea4 | 3452 | cfqq->cfqg->saved_wl_slice = 0; |
f8ae6e3e | 3453 | |
bf572256 JA |
3454 | /* |
3455 | * Put the new queue at the front of the of the current list, | |
3456 | * so we know that it will be selected next. | |
3457 | */ | |
3458 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3459 | |
3460 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3461 | |
62a37f6b JT |
3462 | cfqq->slice_end = 0; |
3463 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3464 | } |
3465 | ||
22e2c507 | 3466 | /* |
5e705374 | 3467 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3468 | * something we should do about it |
3469 | */ | |
3470 | static void | |
5e705374 JA |
3471 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3472 | struct request *rq) | |
22e2c507 | 3473 | { |
c5869807 | 3474 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3475 | |
45333d5a | 3476 | cfqd->rq_queued++; |
65299a3b CH |
3477 | if (rq->cmd_flags & REQ_PRIO) |
3478 | cfqq->prio_pending++; | |
374f84ac | 3479 | |
383cd721 | 3480 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3481 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3482 | cfq_update_idle_window(cfqd, cfqq, cic); |
3483 | ||
b2c18e1e | 3484 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3485 | |
3486 | if (cfqq == cfqd->active_queue) { | |
3487 | /* | |
b029195d JA |
3488 | * Remember that we saw a request from this process, but |
3489 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3490 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3491 | * and merging. If the request is already larger than a single |
3492 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3493 | * merging is already done. Ditto for a busy system that |
3494 | * has other work pending, don't risk delaying until the | |
3495 | * idle timer unplug to continue working. | |
22e2c507 | 3496 | */ |
d6ceb25e | 3497 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3498 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3499 | cfqd->busy_queues > 1) { | |
812df48d | 3500 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3501 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3502 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3503 | } else { |
155fead9 | 3504 | cfqg_stats_update_idle_time(cfqq->cfqg); |
bf791937 | 3505 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3506 | } |
d6ceb25e | 3507 | } |
5e705374 | 3508 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3509 | /* |
3510 | * not the active queue - expire current slice if it is | |
3511 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3512 | * has some old slice time left and is of higher priority or |
3513 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3514 | */ |
3515 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3516 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3517 | } |
1da177e4 LT |
3518 | } |
3519 | ||
165125e1 | 3520 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3521 | { |
b4878f24 | 3522 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3523 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3524 | |
7b679138 | 3525 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 3526 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 3527 | |
30996f40 | 3528 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3529 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3530 | cfq_add_rq_rb(rq); |
155fead9 TH |
3531 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
3532 | rq->cmd_flags); | |
5e705374 | 3533 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3534 | } |
3535 | ||
45333d5a AC |
3536 | /* |
3537 | * Update hw_tag based on peak queue depth over 50 samples under | |
3538 | * sufficient load. | |
3539 | */ | |
3540 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3541 | { | |
1a1238a7 SL |
3542 | struct cfq_queue *cfqq = cfqd->active_queue; |
3543 | ||
53c583d2 CZ |
3544 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3545 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3546 | |
3547 | if (cfqd->hw_tag == 1) | |
3548 | return; | |
45333d5a AC |
3549 | |
3550 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3551 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3552 | return; |
3553 | ||
1a1238a7 SL |
3554 | /* |
3555 | * If active queue hasn't enough requests and can idle, cfq might not | |
3556 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3557 | * case | |
3558 | */ | |
3559 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3560 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3561 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3562 | return; |
3563 | ||
45333d5a AC |
3564 | if (cfqd->hw_tag_samples++ < 50) |
3565 | return; | |
3566 | ||
e459dd08 | 3567 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3568 | cfqd->hw_tag = 1; |
3569 | else | |
3570 | cfqd->hw_tag = 0; | |
45333d5a AC |
3571 | } |
3572 | ||
7667aa06 VG |
3573 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3574 | { | |
c5869807 | 3575 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 3576 | |
02a8f01b JT |
3577 | /* If the queue already has requests, don't wait */ |
3578 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3579 | return false; | |
3580 | ||
7667aa06 VG |
3581 | /* If there are other queues in the group, don't wait */ |
3582 | if (cfqq->cfqg->nr_cfqq > 1) | |
3583 | return false; | |
3584 | ||
7700fc4f SL |
3585 | /* the only queue in the group, but think time is big */ |
3586 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
3587 | return false; | |
3588 | ||
7667aa06 VG |
3589 | if (cfq_slice_used(cfqq)) |
3590 | return true; | |
3591 | ||
3592 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
3593 | if (cic && sample_valid(cic->ttime.ttime_samples) |
3594 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
3595 | return true; |
3596 | ||
3597 | /* | |
3598 | * If think times is less than a jiffy than ttime_mean=0 and above | |
3599 | * will not be true. It might happen that slice has not expired yet | |
3600 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
3601 | * case where think time is less than a jiffy, mark the queue wait | |
3602 | * busy if only 1 jiffy is left in the slice. | |
3603 | */ | |
3604 | if (cfqq->slice_end - jiffies == 1) | |
3605 | return true; | |
3606 | ||
3607 | return false; | |
3608 | } | |
3609 | ||
165125e1 | 3610 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3611 | { |
5e705374 | 3612 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3613 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3614 | const int sync = rq_is_sync(rq); |
b4878f24 | 3615 | unsigned long now; |
1da177e4 | 3616 | |
b4878f24 | 3617 | now = jiffies; |
33659ebb CH |
3618 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3619 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 3620 | |
45333d5a AC |
3621 | cfq_update_hw_tag(cfqd); |
3622 | ||
53c583d2 | 3623 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 3624 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 3625 | cfqd->rq_in_driver--; |
6d048f53 | 3626 | cfqq->dispatched--; |
80bdf0c7 | 3627 | (RQ_CFQG(rq))->dispatched--; |
155fead9 TH |
3628 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
3629 | rq_io_start_time_ns(rq), rq->cmd_flags); | |
1da177e4 | 3630 | |
53c583d2 | 3631 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 3632 | |
365722bb | 3633 | if (sync) { |
34b98d03 | 3634 | struct cfq_rb_root *st; |
f5f2b6ce | 3635 | |
383cd721 | 3636 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
3637 | |
3638 | if (cfq_cfqq_on_rr(cfqq)) | |
34b98d03 | 3639 | st = cfqq->service_tree; |
f5f2b6ce | 3640 | else |
34b98d03 VG |
3641 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), |
3642 | cfqq_type(cfqq)); | |
3643 | ||
3644 | st->ttime.last_end_request = now; | |
573412b2 CZ |
3645 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3646 | cfqd->last_delayed_sync = now; | |
365722bb | 3647 | } |
caaa5f9f | 3648 | |
7700fc4f SL |
3649 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3650 | cfqq->cfqg->ttime.last_end_request = now; | |
3651 | #endif | |
3652 | ||
caaa5f9f JA |
3653 | /* |
3654 | * If this is the active queue, check if it needs to be expired, | |
3655 | * or if we want to idle in case it has no pending requests. | |
3656 | */ | |
3657 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3658 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3659 | ||
44f7c160 JA |
3660 | if (cfq_cfqq_slice_new(cfqq)) { |
3661 | cfq_set_prio_slice(cfqd, cfqq); | |
3662 | cfq_clear_cfqq_slice_new(cfqq); | |
3663 | } | |
f75edf2d VG |
3664 | |
3665 | /* | |
7667aa06 VG |
3666 | * Should we wait for next request to come in before we expire |
3667 | * the queue. | |
f75edf2d | 3668 | */ |
7667aa06 | 3669 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
3670 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3671 | if (!cfqd->cfq_slice_idle) | |
3672 | extend_sl = cfqd->cfq_group_idle; | |
3673 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 3674 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 3675 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
3676 | } |
3677 | ||
a36e71f9 | 3678 | /* |
8e550632 CZ |
3679 | * Idling is not enabled on: |
3680 | * - expired queues | |
3681 | * - idle-priority queues | |
3682 | * - async queues | |
3683 | * - queues with still some requests queued | |
3684 | * - when there is a close cooperator | |
a36e71f9 | 3685 | */ |
0871714e | 3686 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 3687 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3688 | else if (sync && cfqq_empty && |
3689 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 3690 | cfq_arm_slice_timer(cfqd); |
8e550632 | 3691 | } |
caaa5f9f | 3692 | } |
6d048f53 | 3693 | |
53c583d2 | 3694 | if (!cfqd->rq_in_driver) |
23e018a1 | 3695 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3696 | } |
3697 | ||
89850f7e | 3698 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3699 | { |
1b379d8d | 3700 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3701 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3702 | return ELV_MQUEUE_MUST; |
3b18152c | 3703 | } |
1da177e4 | 3704 | |
22e2c507 | 3705 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3706 | } |
3707 | ||
165125e1 | 3708 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3709 | { |
3710 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3711 | struct task_struct *tsk = current; | |
c5869807 | 3712 | struct cfq_io_cq *cic; |
22e2c507 JA |
3713 | struct cfq_queue *cfqq; |
3714 | ||
3715 | /* | |
3716 | * don't force setup of a queue from here, as a call to may_queue | |
3717 | * does not necessarily imply that a request actually will be queued. | |
3718 | * so just lookup a possibly existing queue, or return 'may queue' | |
3719 | * if that fails | |
3720 | */ | |
4ac845a2 | 3721 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3722 | if (!cic) |
3723 | return ELV_MQUEUE_MAY; | |
3724 | ||
b0b78f81 | 3725 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3726 | if (cfqq) { |
abede6da | 3727 | cfq_init_prio_data(cfqq, cic); |
22e2c507 | 3728 | |
89850f7e | 3729 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3730 | } |
3731 | ||
3732 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3733 | } |
3734 | ||
1da177e4 LT |
3735 | /* |
3736 | * queue lock held here | |
3737 | */ | |
bb37b94c | 3738 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3739 | { |
5e705374 | 3740 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3741 | |
5e705374 | 3742 | if (cfqq) { |
22e2c507 | 3743 | const int rw = rq_data_dir(rq); |
1da177e4 | 3744 | |
22e2c507 JA |
3745 | BUG_ON(!cfqq->allocated[rw]); |
3746 | cfqq->allocated[rw]--; | |
1da177e4 | 3747 | |
7f1dc8a2 | 3748 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 3749 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
3750 | rq->elv.priv[0] = NULL; |
3751 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 3752 | |
1da177e4 LT |
3753 | cfq_put_queue(cfqq); |
3754 | } | |
3755 | } | |
3756 | ||
df5fe3e8 | 3757 | static struct cfq_queue * |
c5869807 | 3758 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
3759 | struct cfq_queue *cfqq) |
3760 | { | |
3761 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3762 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3763 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3764 | cfq_put_queue(cfqq); |
3765 | return cic_to_cfqq(cic, 1); | |
3766 | } | |
3767 | ||
e6c5bc73 JM |
3768 | /* |
3769 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3770 | * was the last process referring to said cfqq. | |
3771 | */ | |
3772 | static struct cfq_queue * | |
c5869807 | 3773 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
3774 | { |
3775 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
3776 | cfqq->pid = current->pid; |
3777 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 3778 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
3779 | return cfqq; |
3780 | } | |
3781 | ||
3782 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
3783 | |
3784 | cfq_put_cooperator(cfqq); | |
3785 | ||
e6c5bc73 JM |
3786 | cfq_put_queue(cfqq); |
3787 | return NULL; | |
3788 | } | |
1da177e4 | 3789 | /* |
22e2c507 | 3790 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3791 | */ |
22e2c507 | 3792 | static int |
852c788f TH |
3793 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
3794 | gfp_t gfp_mask) | |
1da177e4 LT |
3795 | { |
3796 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 3797 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 3798 | const int rw = rq_data_dir(rq); |
a6151c3a | 3799 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3800 | struct cfq_queue *cfqq; |
1da177e4 LT |
3801 | |
3802 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3803 | ||
216284c3 | 3804 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 3805 | |
598971bf TH |
3806 | check_ioprio_changed(cic, bio); |
3807 | check_blkcg_changed(cic, bio); | |
e6c5bc73 | 3808 | new_queue: |
91fac317 | 3809 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3810 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
abede6da | 3811 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio, gfp_mask); |
91fac317 | 3812 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3813 | } else { |
e6c5bc73 JM |
3814 | /* |
3815 | * If the queue was seeky for too long, break it apart. | |
3816 | */ | |
ae54abed | 3817 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
3818 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
3819 | cfqq = split_cfqq(cic, cfqq); | |
3820 | if (!cfqq) | |
3821 | goto new_queue; | |
3822 | } | |
3823 | ||
df5fe3e8 JM |
3824 | /* |
3825 | * Check to see if this queue is scheduled to merge with | |
3826 | * another, closely cooperating queue. The merging of | |
3827 | * queues happens here as it must be done in process context. | |
3828 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3829 | */ | |
3830 | if (cfqq->new_cfqq) | |
3831 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3832 | } |
1da177e4 LT |
3833 | |
3834 | cfqq->allocated[rw]++; | |
1da177e4 | 3835 | |
6fae9c25 | 3836 | cfqq->ref++; |
eb7d8c07 | 3837 | cfqg_get(cfqq->cfqg); |
a612fddf | 3838 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 3839 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 3840 | spin_unlock_irq(q->queue_lock); |
5e705374 | 3841 | return 0; |
1da177e4 LT |
3842 | } |
3843 | ||
65f27f38 | 3844 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3845 | { |
65f27f38 | 3846 | struct cfq_data *cfqd = |
23e018a1 | 3847 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3848 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3849 | |
40bb54d1 | 3850 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 3851 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3852 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3853 | } |
3854 | ||
3855 | /* | |
3856 | * Timer running if the active_queue is currently idling inside its time slice | |
3857 | */ | |
3858 | static void cfq_idle_slice_timer(unsigned long data) | |
3859 | { | |
3860 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3861 | struct cfq_queue *cfqq; | |
3862 | unsigned long flags; | |
3c6bd2f8 | 3863 | int timed_out = 1; |
22e2c507 | 3864 | |
7b679138 JA |
3865 | cfq_log(cfqd, "idle timer fired"); |
3866 | ||
22e2c507 JA |
3867 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3868 | ||
fe094d98 JA |
3869 | cfqq = cfqd->active_queue; |
3870 | if (cfqq) { | |
3c6bd2f8 JA |
3871 | timed_out = 0; |
3872 | ||
b029195d JA |
3873 | /* |
3874 | * We saw a request before the queue expired, let it through | |
3875 | */ | |
3876 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3877 | goto out_kick; | |
3878 | ||
22e2c507 JA |
3879 | /* |
3880 | * expired | |
3881 | */ | |
44f7c160 | 3882 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
3883 | goto expire; |
3884 | ||
3885 | /* | |
3886 | * only expire and reinvoke request handler, if there are | |
3887 | * other queues with pending requests | |
3888 | */ | |
caaa5f9f | 3889 | if (!cfqd->busy_queues) |
22e2c507 | 3890 | goto out_cont; |
22e2c507 JA |
3891 | |
3892 | /* | |
3893 | * not expired and it has a request pending, let it dispatch | |
3894 | */ | |
75e50984 | 3895 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3896 | goto out_kick; |
76280aff CZ |
3897 | |
3898 | /* | |
3899 | * Queue depth flag is reset only when the idle didn't succeed | |
3900 | */ | |
3901 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
3902 | } |
3903 | expire: | |
e5ff082e | 3904 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 3905 | out_kick: |
23e018a1 | 3906 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
3907 | out_cont: |
3908 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3909 | } | |
3910 | ||
3b18152c JA |
3911 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3912 | { | |
3913 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 3914 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 3915 | } |
22e2c507 | 3916 | |
c2dea2d1 VT |
3917 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3918 | { | |
3919 | int i; | |
3920 | ||
3921 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
3922 | if (cfqd->async_cfqq[0][i]) | |
3923 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
3924 | if (cfqd->async_cfqq[1][i]) | |
3925 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 3926 | } |
2389d1ef ON |
3927 | |
3928 | if (cfqd->async_idle_cfqq) | |
3929 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
3930 | } |
3931 | ||
b374d18a | 3932 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 3933 | { |
22e2c507 | 3934 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 3935 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3936 | |
3b18152c | 3937 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 3938 | |
d9ff4187 | 3939 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3940 | |
d9ff4187 | 3941 | if (cfqd->active_queue) |
e5ff082e | 3942 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 3943 | |
c2dea2d1 | 3944 | cfq_put_async_queues(cfqd); |
03aa264a TH |
3945 | |
3946 | spin_unlock_irq(q->queue_lock); | |
3947 | ||
a90d742e AV |
3948 | cfq_shutdown_timer_wq(cfqd); |
3949 | ||
ffea73fc TH |
3950 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3951 | blkcg_deactivate_policy(q, &blkcg_policy_cfq); | |
3952 | #else | |
f51b802c | 3953 | kfree(cfqd->root_group); |
2abae55f | 3954 | #endif |
56edf7d7 | 3955 | kfree(cfqd); |
1da177e4 LT |
3956 | } |
3957 | ||
b2fab5ac | 3958 | static int cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
3959 | { |
3960 | struct cfq_data *cfqd; | |
3c798398 | 3961 | struct blkcg_gq *blkg __maybe_unused; |
a2b1693b | 3962 | int i, ret; |
1da177e4 | 3963 | |
94f6030c | 3964 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
a73f730d | 3965 | if (!cfqd) |
b2fab5ac | 3966 | return -ENOMEM; |
80b15c73 | 3967 | |
f51b802c TH |
3968 | cfqd->queue = q; |
3969 | q->elevator->elevator_data = cfqd; | |
3970 | ||
1fa8f6d6 VG |
3971 | /* Init root service tree */ |
3972 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
3973 | ||
f51b802c | 3974 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 3975 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 3976 | ret = blkcg_activate_policy(q, &blkcg_policy_cfq); |
a2b1693b TH |
3977 | if (ret) |
3978 | goto out_free; | |
f51b802c | 3979 | |
a2b1693b | 3980 | cfqd->root_group = blkg_to_cfqg(q->root_blkg); |
f51b802c | 3981 | #else |
a2b1693b | 3982 | ret = -ENOMEM; |
f51b802c TH |
3983 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
3984 | GFP_KERNEL, cfqd->queue->node); | |
a2b1693b TH |
3985 | if (!cfqd->root_group) |
3986 | goto out_free; | |
5624a4e4 | 3987 | |
a2b1693b TH |
3988 | cfq_init_cfqg_base(cfqd->root_group); |
3989 | #endif | |
3381cb8d | 3990 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_DEFAULT; |
5624a4e4 | 3991 | |
26a2ac00 JA |
3992 | /* |
3993 | * Not strictly needed (since RB_ROOT just clears the node and we | |
3994 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
3995 | * to add magic to the rb code | |
3996 | */ | |
3997 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
3998 | cfqd->prio_trees[i] = RB_ROOT; | |
3999 | ||
6118b70b JA |
4000 | /* |
4001 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
4002 | * Grab a permanent reference to it, so that the normal code flow | |
f51b802c TH |
4003 | * will not attempt to free it. oom_cfqq is linked to root_group |
4004 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
4005 | * the reference from linking right away. | |
6118b70b JA |
4006 | */ |
4007 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 4008 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
4009 | |
4010 | spin_lock_irq(q->queue_lock); | |
f51b802c | 4011 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 4012 | cfqg_put(cfqd->root_group); |
1adaf3dd | 4013 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 4014 | |
22e2c507 JA |
4015 | init_timer(&cfqd->idle_slice_timer); |
4016 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
4017 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
4018 | ||
23e018a1 | 4019 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 4020 | |
1da177e4 | 4021 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
4022 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
4023 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
4024 | cfqd->cfq_back_max = cfq_back_max; |
4025 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
4026 | cfqd->cfq_slice[0] = cfq_slice_async; |
4027 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
5bf14c07 | 4028 | cfqd->cfq_target_latency = cfq_target_latency; |
22e2c507 JA |
4029 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
4030 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
80bdf0c7 | 4031 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 4032 | cfqd->cfq_latency = 1; |
e459dd08 | 4033 | cfqd->hw_tag = -1; |
edc71131 CZ |
4034 | /* |
4035 | * we optimistically start assuming sync ops weren't delayed in last | |
4036 | * second, in order to have larger depth for async operations. | |
4037 | */ | |
573412b2 | 4038 | cfqd->last_delayed_sync = jiffies - HZ; |
b2fab5ac | 4039 | return 0; |
a2b1693b TH |
4040 | |
4041 | out_free: | |
4042 | kfree(cfqd); | |
4043 | return ret; | |
1da177e4 LT |
4044 | } |
4045 | ||
1da177e4 LT |
4046 | /* |
4047 | * sysfs parts below --> | |
4048 | */ | |
1da177e4 LT |
4049 | static ssize_t |
4050 | cfq_var_show(unsigned int var, char *page) | |
4051 | { | |
4052 | return sprintf(page, "%d\n", var); | |
4053 | } | |
4054 | ||
4055 | static ssize_t | |
4056 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
4057 | { | |
4058 | char *p = (char *) page; | |
4059 | ||
4060 | *var = simple_strtoul(p, &p, 10); | |
4061 | return count; | |
4062 | } | |
4063 | ||
1da177e4 | 4064 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 4065 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 4066 | { \ |
3d1ab40f | 4067 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4068 | unsigned int __data = __VAR; \ |
4069 | if (__CONV) \ | |
4070 | __data = jiffies_to_msecs(__data); \ | |
4071 | return cfq_var_show(__data, (page)); \ | |
4072 | } | |
4073 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
4074 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
4075 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
4076 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
4077 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 4078 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 4079 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
4080 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4081 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4082 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4083 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
5bf14c07 | 4084 | SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1); |
1da177e4 LT |
4085 | #undef SHOW_FUNCTION |
4086 | ||
4087 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 4088 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4089 | { \ |
3d1ab40f | 4090 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4091 | unsigned int __data; \ |
4092 | int ret = cfq_var_store(&__data, (page), count); \ | |
4093 | if (__data < (MIN)) \ | |
4094 | __data = (MIN); \ | |
4095 | else if (__data > (MAX)) \ | |
4096 | __data = (MAX); \ | |
4097 | if (__CONV) \ | |
4098 | *(__PTR) = msecs_to_jiffies(__data); \ | |
4099 | else \ | |
4100 | *(__PTR) = __data; \ | |
4101 | return ret; \ | |
4102 | } | |
4103 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4104 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4105 | UINT_MAX, 1); | |
4106 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4107 | UINT_MAX, 1); | |
e572ec7e | 4108 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4109 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4110 | UINT_MAX, 0); | |
22e2c507 | 4111 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4112 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4113 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4114 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4115 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4116 | UINT_MAX, 0); | |
963b72fc | 4117 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
5bf14c07 | 4118 | STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1); |
1da177e4 LT |
4119 | #undef STORE_FUNCTION |
4120 | ||
e572ec7e AV |
4121 | #define CFQ_ATTR(name) \ |
4122 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4123 | ||
4124 | static struct elv_fs_entry cfq_attrs[] = { | |
4125 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4126 | CFQ_ATTR(fifo_expire_sync), |
4127 | CFQ_ATTR(fifo_expire_async), | |
4128 | CFQ_ATTR(back_seek_max), | |
4129 | CFQ_ATTR(back_seek_penalty), | |
4130 | CFQ_ATTR(slice_sync), | |
4131 | CFQ_ATTR(slice_async), | |
4132 | CFQ_ATTR(slice_async_rq), | |
4133 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 4134 | CFQ_ATTR(group_idle), |
963b72fc | 4135 | CFQ_ATTR(low_latency), |
5bf14c07 | 4136 | CFQ_ATTR(target_latency), |
e572ec7e | 4137 | __ATTR_NULL |
1da177e4 LT |
4138 | }; |
4139 | ||
1da177e4 LT |
4140 | static struct elevator_type iosched_cfq = { |
4141 | .ops = { | |
4142 | .elevator_merge_fn = cfq_merge, | |
4143 | .elevator_merged_fn = cfq_merged_request, | |
4144 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 4145 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 4146 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4147 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4148 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4149 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4150 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4151 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4152 | .elevator_former_req_fn = elv_rb_former_request, |
4153 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 4154 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 4155 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
4156 | .elevator_set_req_fn = cfq_set_request, |
4157 | .elevator_put_req_fn = cfq_put_request, | |
4158 | .elevator_may_queue_fn = cfq_may_queue, | |
4159 | .elevator_init_fn = cfq_init_queue, | |
4160 | .elevator_exit_fn = cfq_exit_queue, | |
4161 | }, | |
3d3c2379 TH |
4162 | .icq_size = sizeof(struct cfq_io_cq), |
4163 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 4164 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 4165 | .elevator_name = "cfq", |
1da177e4 LT |
4166 | .elevator_owner = THIS_MODULE, |
4167 | }; | |
4168 | ||
3e252066 | 4169 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4170 | static struct blkcg_policy blkcg_policy_cfq = { |
f9fcc2d3 TH |
4171 | .pd_size = sizeof(struct cfq_group), |
4172 | .cftypes = cfq_blkcg_files, | |
4173 | ||
4174 | .pd_init_fn = cfq_pd_init, | |
4175 | .pd_reset_stats_fn = cfq_pd_reset_stats, | |
3e252066 | 4176 | }; |
3e252066 VG |
4177 | #endif |
4178 | ||
1da177e4 LT |
4179 | static int __init cfq_init(void) |
4180 | { | |
3d3c2379 TH |
4181 | int ret; |
4182 | ||
22e2c507 JA |
4183 | /* |
4184 | * could be 0 on HZ < 1000 setups | |
4185 | */ | |
4186 | if (!cfq_slice_async) | |
4187 | cfq_slice_async = 1; | |
4188 | if (!cfq_slice_idle) | |
4189 | cfq_slice_idle = 1; | |
4190 | ||
80bdf0c7 VG |
4191 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4192 | if (!cfq_group_idle) | |
4193 | cfq_group_idle = 1; | |
8bd435b3 | 4194 | |
3c798398 | 4195 | ret = blkcg_policy_register(&blkcg_policy_cfq); |
8bd435b3 TH |
4196 | if (ret) |
4197 | return ret; | |
ffea73fc TH |
4198 | #else |
4199 | cfq_group_idle = 0; | |
4200 | #endif | |
8bd435b3 | 4201 | |
fd794956 | 4202 | ret = -ENOMEM; |
3d3c2379 TH |
4203 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4204 | if (!cfq_pool) | |
8bd435b3 | 4205 | goto err_pol_unreg; |
1da177e4 | 4206 | |
3d3c2379 | 4207 | ret = elv_register(&iosched_cfq); |
8bd435b3 TH |
4208 | if (ret) |
4209 | goto err_free_pool; | |
3d3c2379 | 4210 | |
2fdd82bd | 4211 | return 0; |
8bd435b3 TH |
4212 | |
4213 | err_free_pool: | |
4214 | kmem_cache_destroy(cfq_pool); | |
4215 | err_pol_unreg: | |
ffea73fc | 4216 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4217 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4218 | #endif |
8bd435b3 | 4219 | return ret; |
1da177e4 LT |
4220 | } |
4221 | ||
4222 | static void __exit cfq_exit(void) | |
4223 | { | |
ffea73fc | 4224 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4225 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4226 | #endif |
1da177e4 | 4227 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4228 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4229 | } |
4230 | ||
4231 | module_init(cfq_init); | |
4232 | module_exit(cfq_exit); | |
4233 | ||
4234 | MODULE_AUTHOR("Jens Axboe"); | |
4235 | MODULE_LICENSE("GPL"); | |
4236 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |