]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * CFQ, or complete fairness queueing, disk scheduler. |
3 | * | |
4 | * Based on ideas from a previously unfinished io | |
5 | * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli. | |
6 | * | |
0fe23479 | 7 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
1da177e4 | 8 | */ |
1da177e4 | 9 | #include <linux/module.h> |
5a0e3ad6 | 10 | #include <linux/slab.h> |
1cc9be68 AV |
11 | #include <linux/blkdev.h> |
12 | #include <linux/elevator.h> | |
ad5ebd2f | 13 | #include <linux/jiffies.h> |
1da177e4 | 14 | #include <linux/rbtree.h> |
22e2c507 | 15 | #include <linux/ioprio.h> |
7b679138 | 16 | #include <linux/blktrace_api.h> |
6e736be7 | 17 | #include "blk.h" |
629ed0b1 | 18 | #include "blk-cgroup.h" |
1da177e4 LT |
19 | |
20 | /* | |
21 | * tunables | |
22 | */ | |
fe094d98 | 23 | /* max queue in one round of service */ |
abc3c744 | 24 | static const int cfq_quantum = 8; |
64100099 | 25 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
26 | /* maximum backwards seek, in KiB */ |
27 | static const int cfq_back_max = 16 * 1024; | |
28 | /* penalty of a backwards seek */ | |
29 | static const int cfq_back_penalty = 2; | |
64100099 | 30 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 31 | static int cfq_slice_async = HZ / 25; |
64100099 | 32 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 33 | static int cfq_slice_idle = HZ / 125; |
80bdf0c7 | 34 | static int cfq_group_idle = HZ / 125; |
5db5d642 CZ |
35 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
36 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 37 | |
d9e7620e | 38 | /* |
0871714e | 39 | * offset from end of service tree |
d9e7620e | 40 | */ |
0871714e | 41 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
42 | |
43 | /* | |
44 | * below this threshold, we consider thinktime immediate | |
45 | */ | |
46 | #define CFQ_MIN_TT (2) | |
47 | ||
22e2c507 | 48 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 49 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 50 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 51 | |
3dde36dd | 52 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 53 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 54 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 55 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 56 | |
a612fddf TH |
57 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
58 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
59 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 60 | |
e18b890b | 61 | static struct kmem_cache *cfq_pool; |
1da177e4 | 62 | |
22e2c507 JA |
63 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
64 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
65 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
66 | ||
206dc69b | 67 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 68 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 69 | |
c5869807 TH |
70 | struct cfq_ttime { |
71 | unsigned long last_end_request; | |
72 | ||
73 | unsigned long ttime_total; | |
74 | unsigned long ttime_samples; | |
75 | unsigned long ttime_mean; | |
76 | }; | |
77 | ||
cc09e299 JA |
78 | /* |
79 | * Most of our rbtree usage is for sorting with min extraction, so | |
80 | * if we cache the leftmost node we don't have to walk down the tree | |
81 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
82 | * move this into the elevator for the rq sorting as well. | |
83 | */ | |
84 | struct cfq_rb_root { | |
85 | struct rb_root rb; | |
86 | struct rb_node *left; | |
aa6f6a3d | 87 | unsigned count; |
73e9ffdd | 88 | unsigned total_weight; |
1fa8f6d6 | 89 | u64 min_vdisktime; |
f5f2b6ce | 90 | struct cfq_ttime ttime; |
cc09e299 | 91 | }; |
f5f2b6ce SL |
92 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
93 | .ttime = {.last_end_request = jiffies,},} | |
cc09e299 | 94 | |
6118b70b JA |
95 | /* |
96 | * Per process-grouping structure | |
97 | */ | |
98 | struct cfq_queue { | |
99 | /* reference count */ | |
30d7b944 | 100 | int ref; |
6118b70b JA |
101 | /* various state flags, see below */ |
102 | unsigned int flags; | |
103 | /* parent cfq_data */ | |
104 | struct cfq_data *cfqd; | |
105 | /* service_tree member */ | |
106 | struct rb_node rb_node; | |
107 | /* service_tree key */ | |
108 | unsigned long rb_key; | |
109 | /* prio tree member */ | |
110 | struct rb_node p_node; | |
111 | /* prio tree root we belong to, if any */ | |
112 | struct rb_root *p_root; | |
113 | /* sorted list of pending requests */ | |
114 | struct rb_root sort_list; | |
115 | /* if fifo isn't expired, next request to serve */ | |
116 | struct request *next_rq; | |
117 | /* requests queued in sort_list */ | |
118 | int queued[2]; | |
119 | /* currently allocated requests */ | |
120 | int allocated[2]; | |
121 | /* fifo list of requests in sort_list */ | |
122 | struct list_head fifo; | |
123 | ||
dae739eb VG |
124 | /* time when queue got scheduled in to dispatch first request. */ |
125 | unsigned long dispatch_start; | |
f75edf2d | 126 | unsigned int allocated_slice; |
c4081ba5 | 127 | unsigned int slice_dispatch; |
dae739eb VG |
128 | /* time when first request from queue completed and slice started. */ |
129 | unsigned long slice_start; | |
6118b70b JA |
130 | unsigned long slice_end; |
131 | long slice_resid; | |
6118b70b | 132 | |
65299a3b CH |
133 | /* pending priority requests */ |
134 | int prio_pending; | |
6118b70b JA |
135 | /* number of requests that are on the dispatch list or inside driver */ |
136 | int dispatched; | |
137 | ||
138 | /* io prio of this group */ | |
139 | unsigned short ioprio, org_ioprio; | |
4aede84b | 140 | unsigned short ioprio_class; |
6118b70b | 141 | |
c4081ba5 RK |
142 | pid_t pid; |
143 | ||
3dde36dd | 144 | u32 seek_history; |
b2c18e1e JM |
145 | sector_t last_request_pos; |
146 | ||
aa6f6a3d | 147 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 148 | struct cfq_queue *new_cfqq; |
cdb16e8f | 149 | struct cfq_group *cfqg; |
c4e7893e VG |
150 | /* Number of sectors dispatched from queue in single dispatch round */ |
151 | unsigned long nr_sectors; | |
6118b70b JA |
152 | }; |
153 | ||
c0324a02 | 154 | /* |
718eee05 | 155 | * First index in the service_trees. |
c0324a02 CZ |
156 | * IDLE is handled separately, so it has negative index |
157 | */ | |
3bf10fea | 158 | enum wl_class_t { |
c0324a02 | 159 | BE_WORKLOAD = 0, |
615f0259 VG |
160 | RT_WORKLOAD = 1, |
161 | IDLE_WORKLOAD = 2, | |
b4627321 | 162 | CFQ_PRIO_NR, |
c0324a02 CZ |
163 | }; |
164 | ||
718eee05 CZ |
165 | /* |
166 | * Second index in the service_trees. | |
167 | */ | |
168 | enum wl_type_t { | |
169 | ASYNC_WORKLOAD = 0, | |
170 | SYNC_NOIDLE_WORKLOAD = 1, | |
171 | SYNC_WORKLOAD = 2 | |
172 | }; | |
173 | ||
155fead9 TH |
174 | struct cfqg_stats { |
175 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
176 | /* total bytes transferred */ | |
177 | struct blkg_rwstat service_bytes; | |
178 | /* total IOs serviced, post merge */ | |
179 | struct blkg_rwstat serviced; | |
180 | /* number of ios merged */ | |
181 | struct blkg_rwstat merged; | |
182 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
183 | struct blkg_rwstat service_time; | |
184 | /* total time spent waiting in scheduler queue in ns */ | |
185 | struct blkg_rwstat wait_time; | |
186 | /* number of IOs queued up */ | |
187 | struct blkg_rwstat queued; | |
188 | /* total sectors transferred */ | |
189 | struct blkg_stat sectors; | |
190 | /* total disk time and nr sectors dispatched by this group */ | |
191 | struct blkg_stat time; | |
192 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
193 | /* time not charged to this cgroup */ | |
194 | struct blkg_stat unaccounted_time; | |
195 | /* sum of number of ios queued across all samples */ | |
196 | struct blkg_stat avg_queue_size_sum; | |
197 | /* count of samples taken for average */ | |
198 | struct blkg_stat avg_queue_size_samples; | |
199 | /* how many times this group has been removed from service tree */ | |
200 | struct blkg_stat dequeue; | |
201 | /* total time spent waiting for it to be assigned a timeslice. */ | |
202 | struct blkg_stat group_wait_time; | |
3c798398 | 203 | /* time spent idling for this blkcg_gq */ |
155fead9 TH |
204 | struct blkg_stat idle_time; |
205 | /* total time with empty current active q with other requests queued */ | |
206 | struct blkg_stat empty_time; | |
207 | /* fields after this shouldn't be cleared on stat reset */ | |
208 | uint64_t start_group_wait_time; | |
209 | uint64_t start_idle_time; | |
210 | uint64_t start_empty_time; | |
211 | uint16_t flags; | |
212 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
213 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
214 | }; | |
215 | ||
cdb16e8f VG |
216 | /* This is per cgroup per device grouping structure */ |
217 | struct cfq_group { | |
f95a04af TH |
218 | /* must be the first member */ |
219 | struct blkg_policy_data pd; | |
220 | ||
1fa8f6d6 VG |
221 | /* group service_tree member */ |
222 | struct rb_node rb_node; | |
223 | ||
224 | /* group service_tree key */ | |
225 | u64 vdisktime; | |
e71357e1 TH |
226 | |
227 | /* | |
228 | * There are two weights - (internal) weight is the weight of this | |
229 | * cfqg against the sibling cfqgs. leaf_weight is the wight of | |
230 | * this cfqg against the child cfqgs. For the root cfqg, both | |
231 | * weights are kept in sync for backward compatibility. | |
232 | */ | |
25bc6b07 | 233 | unsigned int weight; |
8184f93e | 234 | unsigned int new_weight; |
3381cb8d | 235 | unsigned int dev_weight; |
1fa8f6d6 | 236 | |
e71357e1 TH |
237 | unsigned int leaf_weight; |
238 | unsigned int new_leaf_weight; | |
239 | unsigned int dev_leaf_weight; | |
240 | ||
1fa8f6d6 VG |
241 | /* number of cfqq currently on this group */ |
242 | int nr_cfqq; | |
243 | ||
cdb16e8f | 244 | /* |
4495a7d4 | 245 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
246 | * create the array for each prio class but at run time it is used |
247 | * only for RT and BE class and slot for IDLE class remains unused. | |
248 | * This is primarily done to avoid confusion and a gcc warning. | |
249 | */ | |
250 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
251 | /* | |
252 | * rr lists of queues with requests. We maintain service trees for | |
253 | * RT and BE classes. These trees are subdivided in subclasses | |
254 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
255 | * class there is no subclassification and all the cfq queues go on | |
256 | * a single tree service_tree_idle. | |
cdb16e8f VG |
257 | * Counts are embedded in the cfq_rb_root |
258 | */ | |
259 | struct cfq_rb_root service_trees[2][3]; | |
260 | struct cfq_rb_root service_tree_idle; | |
dae739eb | 261 | |
4d2ceea4 VG |
262 | unsigned long saved_wl_slice; |
263 | enum wl_type_t saved_wl_type; | |
264 | enum wl_class_t saved_wl_class; | |
4eef3049 | 265 | |
80bdf0c7 VG |
266 | /* number of requests that are on the dispatch list or inside driver */ |
267 | int dispatched; | |
7700fc4f | 268 | struct cfq_ttime ttime; |
155fead9 | 269 | struct cfqg_stats stats; |
cdb16e8f | 270 | }; |
718eee05 | 271 | |
c5869807 TH |
272 | struct cfq_io_cq { |
273 | struct io_cq icq; /* must be the first member */ | |
274 | struct cfq_queue *cfqq[2]; | |
275 | struct cfq_ttime ttime; | |
598971bf TH |
276 | int ioprio; /* the current ioprio */ |
277 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
278 | uint64_t blkcg_id; /* the current blkcg ID */ | |
279 | #endif | |
c5869807 TH |
280 | }; |
281 | ||
22e2c507 JA |
282 | /* |
283 | * Per block device queue structure | |
284 | */ | |
1da177e4 | 285 | struct cfq_data { |
165125e1 | 286 | struct request_queue *queue; |
1fa8f6d6 VG |
287 | /* Root service tree for cfq_groups */ |
288 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 289 | struct cfq_group *root_group; |
22e2c507 | 290 | |
c0324a02 CZ |
291 | /* |
292 | * The priority currently being served | |
22e2c507 | 293 | */ |
4d2ceea4 VG |
294 | enum wl_class_t serving_wl_class; |
295 | enum wl_type_t serving_wl_type; | |
718eee05 | 296 | unsigned long workload_expires; |
cdb16e8f | 297 | struct cfq_group *serving_group; |
a36e71f9 JA |
298 | |
299 | /* | |
300 | * Each priority tree is sorted by next_request position. These | |
301 | * trees are used when determining if two or more queues are | |
302 | * interleaving requests (see cfq_close_cooperator). | |
303 | */ | |
304 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
305 | ||
22e2c507 | 306 | unsigned int busy_queues; |
ef8a41df | 307 | unsigned int busy_sync_queues; |
22e2c507 | 308 | |
53c583d2 CZ |
309 | int rq_in_driver; |
310 | int rq_in_flight[2]; | |
45333d5a AC |
311 | |
312 | /* | |
313 | * queue-depth detection | |
314 | */ | |
315 | int rq_queued; | |
25776e35 | 316 | int hw_tag; |
e459dd08 CZ |
317 | /* |
318 | * hw_tag can be | |
319 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
320 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
321 | * 0 => no NCQ | |
322 | */ | |
323 | int hw_tag_est_depth; | |
324 | unsigned int hw_tag_samples; | |
1da177e4 | 325 | |
22e2c507 JA |
326 | /* |
327 | * idle window management | |
328 | */ | |
329 | struct timer_list idle_slice_timer; | |
23e018a1 | 330 | struct work_struct unplug_work; |
1da177e4 | 331 | |
22e2c507 | 332 | struct cfq_queue *active_queue; |
c5869807 | 333 | struct cfq_io_cq *active_cic; |
22e2c507 | 334 | |
c2dea2d1 VT |
335 | /* |
336 | * async queue for each priority case | |
337 | */ | |
338 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
339 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 340 | |
6d048f53 | 341 | sector_t last_position; |
1da177e4 | 342 | |
1da177e4 LT |
343 | /* |
344 | * tunables, see top of file | |
345 | */ | |
346 | unsigned int cfq_quantum; | |
22e2c507 | 347 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
348 | unsigned int cfq_back_penalty; |
349 | unsigned int cfq_back_max; | |
22e2c507 JA |
350 | unsigned int cfq_slice[2]; |
351 | unsigned int cfq_slice_async_rq; | |
352 | unsigned int cfq_slice_idle; | |
80bdf0c7 | 353 | unsigned int cfq_group_idle; |
963b72fc | 354 | unsigned int cfq_latency; |
5bf14c07 | 355 | unsigned int cfq_target_latency; |
d9ff4187 | 356 | |
6118b70b JA |
357 | /* |
358 | * Fallback dummy cfqq for extreme OOM conditions | |
359 | */ | |
360 | struct cfq_queue oom_cfqq; | |
365722bb | 361 | |
573412b2 | 362 | unsigned long last_delayed_sync; |
1da177e4 LT |
363 | }; |
364 | ||
25fb5169 VG |
365 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
366 | ||
34b98d03 | 367 | static struct cfq_rb_root *st_for(struct cfq_group *cfqg, |
3bf10fea | 368 | enum wl_class_t class, |
65b32a57 | 369 | enum wl_type_t type) |
c0324a02 | 370 | { |
1fa8f6d6 VG |
371 | if (!cfqg) |
372 | return NULL; | |
373 | ||
3bf10fea | 374 | if (class == IDLE_WORKLOAD) |
cdb16e8f | 375 | return &cfqg->service_tree_idle; |
c0324a02 | 376 | |
3bf10fea | 377 | return &cfqg->service_trees[class][type]; |
c0324a02 CZ |
378 | } |
379 | ||
3b18152c | 380 | enum cfqq_state_flags { |
b0b8d749 JA |
381 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
382 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 383 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 384 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
385 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
386 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
387 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 388 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 389 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 390 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 391 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 392 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 393 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
394 | }; |
395 | ||
396 | #define CFQ_CFQQ_FNS(name) \ | |
397 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
398 | { \ | |
fe094d98 | 399 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
400 | } \ |
401 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
402 | { \ | |
fe094d98 | 403 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
404 | } \ |
405 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
406 | { \ | |
fe094d98 | 407 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
408 | } |
409 | ||
410 | CFQ_CFQQ_FNS(on_rr); | |
411 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 412 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 413 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
414 | CFQ_CFQQ_FNS(fifo_expire); |
415 | CFQ_CFQQ_FNS(idle_window); | |
416 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 417 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 418 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 419 | CFQ_CFQQ_FNS(coop); |
ae54abed | 420 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 421 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 422 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
423 | #undef CFQ_CFQQ_FNS |
424 | ||
f95a04af TH |
425 | static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd) |
426 | { | |
427 | return pd ? container_of(pd, struct cfq_group, pd) : NULL; | |
428 | } | |
429 | ||
f95a04af TH |
430 | static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg) |
431 | { | |
432 | return pd_to_blkg(&cfqg->pd); | |
433 | } | |
434 | ||
629ed0b1 | 435 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
2ce4d50f | 436 | |
155fead9 TH |
437 | /* cfqg stats flags */ |
438 | enum cfqg_stats_flags { | |
439 | CFQG_stats_waiting = 0, | |
440 | CFQG_stats_idling, | |
441 | CFQG_stats_empty, | |
629ed0b1 TH |
442 | }; |
443 | ||
155fead9 TH |
444 | #define CFQG_FLAG_FNS(name) \ |
445 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ | |
629ed0b1 | 446 | { \ |
155fead9 | 447 | stats->flags |= (1 << CFQG_stats_##name); \ |
629ed0b1 | 448 | } \ |
155fead9 | 449 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 450 | { \ |
155fead9 | 451 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
629ed0b1 | 452 | } \ |
155fead9 | 453 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 454 | { \ |
155fead9 | 455 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
629ed0b1 TH |
456 | } \ |
457 | ||
155fead9 TH |
458 | CFQG_FLAG_FNS(waiting) |
459 | CFQG_FLAG_FNS(idling) | |
460 | CFQG_FLAG_FNS(empty) | |
461 | #undef CFQG_FLAG_FNS | |
629ed0b1 TH |
462 | |
463 | /* This should be called with the queue_lock held. */ | |
155fead9 | 464 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
629ed0b1 TH |
465 | { |
466 | unsigned long long now; | |
467 | ||
155fead9 | 468 | if (!cfqg_stats_waiting(stats)) |
629ed0b1 TH |
469 | return; |
470 | ||
471 | now = sched_clock(); | |
472 | if (time_after64(now, stats->start_group_wait_time)) | |
473 | blkg_stat_add(&stats->group_wait_time, | |
474 | now - stats->start_group_wait_time); | |
155fead9 | 475 | cfqg_stats_clear_waiting(stats); |
629ed0b1 TH |
476 | } |
477 | ||
478 | /* This should be called with the queue_lock held. */ | |
155fead9 TH |
479 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
480 | struct cfq_group *curr_cfqg) | |
629ed0b1 | 481 | { |
155fead9 | 482 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 483 | |
155fead9 | 484 | if (cfqg_stats_waiting(stats)) |
629ed0b1 | 485 | return; |
155fead9 | 486 | if (cfqg == curr_cfqg) |
629ed0b1 | 487 | return; |
155fead9 TH |
488 | stats->start_group_wait_time = sched_clock(); |
489 | cfqg_stats_mark_waiting(stats); | |
629ed0b1 TH |
490 | } |
491 | ||
492 | /* This should be called with the queue_lock held. */ | |
155fead9 | 493 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
629ed0b1 TH |
494 | { |
495 | unsigned long long now; | |
496 | ||
155fead9 | 497 | if (!cfqg_stats_empty(stats)) |
629ed0b1 TH |
498 | return; |
499 | ||
500 | now = sched_clock(); | |
501 | if (time_after64(now, stats->start_empty_time)) | |
502 | blkg_stat_add(&stats->empty_time, | |
503 | now - stats->start_empty_time); | |
155fead9 | 504 | cfqg_stats_clear_empty(stats); |
629ed0b1 TH |
505 | } |
506 | ||
155fead9 | 507 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
629ed0b1 | 508 | { |
155fead9 | 509 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
629ed0b1 TH |
510 | } |
511 | ||
155fead9 | 512 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
629ed0b1 | 513 | { |
155fead9 | 514 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
515 | |
516 | if (blkg_rwstat_sum(&stats->queued)) | |
517 | return; | |
518 | ||
519 | /* | |
520 | * group is already marked empty. This can happen if cfqq got new | |
521 | * request in parent group and moved to this group while being added | |
522 | * to service tree. Just ignore the event and move on. | |
523 | */ | |
155fead9 | 524 | if (cfqg_stats_empty(stats)) |
629ed0b1 TH |
525 | return; |
526 | ||
527 | stats->start_empty_time = sched_clock(); | |
155fead9 | 528 | cfqg_stats_mark_empty(stats); |
629ed0b1 TH |
529 | } |
530 | ||
155fead9 | 531 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 532 | { |
155fead9 | 533 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 534 | |
155fead9 | 535 | if (cfqg_stats_idling(stats)) { |
629ed0b1 TH |
536 | unsigned long long now = sched_clock(); |
537 | ||
538 | if (time_after64(now, stats->start_idle_time)) | |
539 | blkg_stat_add(&stats->idle_time, | |
540 | now - stats->start_idle_time); | |
155fead9 | 541 | cfqg_stats_clear_idling(stats); |
629ed0b1 TH |
542 | } |
543 | } | |
544 | ||
155fead9 | 545 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 546 | { |
155fead9 | 547 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 548 | |
155fead9 | 549 | BUG_ON(cfqg_stats_idling(stats)); |
629ed0b1 TH |
550 | |
551 | stats->start_idle_time = sched_clock(); | |
155fead9 | 552 | cfqg_stats_mark_idling(stats); |
629ed0b1 TH |
553 | } |
554 | ||
155fead9 | 555 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
629ed0b1 | 556 | { |
155fead9 | 557 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
558 | |
559 | blkg_stat_add(&stats->avg_queue_size_sum, | |
560 | blkg_rwstat_sum(&stats->queued)); | |
561 | blkg_stat_add(&stats->avg_queue_size_samples, 1); | |
155fead9 | 562 | cfqg_stats_update_group_wait_time(stats); |
629ed0b1 TH |
563 | } |
564 | ||
565 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
566 | ||
f48ec1d7 TH |
567 | static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { } |
568 | static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } | |
569 | static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } | |
570 | static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } | |
571 | static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } | |
572 | static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } | |
573 | static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } | |
629ed0b1 TH |
574 | |
575 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
576 | ||
577 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
2ce4d50f | 578 | |
ffea73fc TH |
579 | static struct blkcg_policy blkcg_policy_cfq; |
580 | ||
581 | static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg) | |
582 | { | |
583 | return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq)); | |
584 | } | |
585 | ||
eb7d8c07 TH |
586 | static inline void cfqg_get(struct cfq_group *cfqg) |
587 | { | |
588 | return blkg_get(cfqg_to_blkg(cfqg)); | |
589 | } | |
590 | ||
591 | static inline void cfqg_put(struct cfq_group *cfqg) | |
592 | { | |
593 | return blkg_put(cfqg_to_blkg(cfqg)); | |
594 | } | |
595 | ||
54e7ed12 TH |
596 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \ |
597 | char __pbuf[128]; \ | |
598 | \ | |
599 | blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \ | |
b226e5c4 VG |
600 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \ |
601 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
602 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
54e7ed12 TH |
603 | __pbuf, ##args); \ |
604 | } while (0) | |
605 | ||
606 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \ | |
607 | char __pbuf[128]; \ | |
608 | \ | |
609 | blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \ | |
610 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \ | |
611 | } while (0) | |
2868ef7b | 612 | |
155fead9 TH |
613 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
614 | struct cfq_group *curr_cfqg, int rw) | |
2ce4d50f | 615 | { |
155fead9 TH |
616 | blkg_rwstat_add(&cfqg->stats.queued, rw, 1); |
617 | cfqg_stats_end_empty_time(&cfqg->stats); | |
618 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); | |
2ce4d50f TH |
619 | } |
620 | ||
155fead9 TH |
621 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
622 | unsigned long time, unsigned long unaccounted_time) | |
2ce4d50f | 623 | { |
155fead9 | 624 | blkg_stat_add(&cfqg->stats.time, time); |
629ed0b1 | 625 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
155fead9 | 626 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
629ed0b1 | 627 | #endif |
2ce4d50f TH |
628 | } |
629 | ||
155fead9 | 630 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) |
2ce4d50f | 631 | { |
155fead9 | 632 | blkg_rwstat_add(&cfqg->stats.queued, rw, -1); |
2ce4d50f TH |
633 | } |
634 | ||
155fead9 | 635 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) |
2ce4d50f | 636 | { |
155fead9 | 637 | blkg_rwstat_add(&cfqg->stats.merged, rw, 1); |
2ce4d50f TH |
638 | } |
639 | ||
155fead9 TH |
640 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, |
641 | uint64_t bytes, int rw) | |
2ce4d50f | 642 | { |
155fead9 TH |
643 | blkg_stat_add(&cfqg->stats.sectors, bytes >> 9); |
644 | blkg_rwstat_add(&cfqg->stats.serviced, rw, 1); | |
645 | blkg_rwstat_add(&cfqg->stats.service_bytes, rw, bytes); | |
2ce4d50f TH |
646 | } |
647 | ||
155fead9 TH |
648 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
649 | uint64_t start_time, uint64_t io_start_time, int rw) | |
2ce4d50f | 650 | { |
155fead9 | 651 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 652 | unsigned long long now = sched_clock(); |
629ed0b1 TH |
653 | |
654 | if (time_after64(now, io_start_time)) | |
655 | blkg_rwstat_add(&stats->service_time, rw, now - io_start_time); | |
656 | if (time_after64(io_start_time, start_time)) | |
657 | blkg_rwstat_add(&stats->wait_time, rw, | |
658 | io_start_time - start_time); | |
2ce4d50f TH |
659 | } |
660 | ||
3c798398 | 661 | static void cfq_pd_reset_stats(struct blkcg_gq *blkg) |
155fead9 TH |
662 | { |
663 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); | |
664 | struct cfqg_stats *stats = &cfqg->stats; | |
665 | ||
666 | /* queued stats shouldn't be cleared */ | |
667 | blkg_rwstat_reset(&stats->service_bytes); | |
668 | blkg_rwstat_reset(&stats->serviced); | |
669 | blkg_rwstat_reset(&stats->merged); | |
670 | blkg_rwstat_reset(&stats->service_time); | |
671 | blkg_rwstat_reset(&stats->wait_time); | |
672 | blkg_stat_reset(&stats->time); | |
673 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
674 | blkg_stat_reset(&stats->unaccounted_time); | |
675 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
676 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
677 | blkg_stat_reset(&stats->dequeue); | |
678 | blkg_stat_reset(&stats->group_wait_time); | |
679 | blkg_stat_reset(&stats->idle_time); | |
680 | blkg_stat_reset(&stats->empty_time); | |
681 | #endif | |
682 | } | |
683 | ||
eb7d8c07 TH |
684 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
685 | ||
eb7d8c07 TH |
686 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
687 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
688 | ||
7b679138 | 689 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
b226e5c4 VG |
690 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \ |
691 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
692 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
693 | ##args) | |
4495a7d4 | 694 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 | 695 | |
155fead9 TH |
696 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
697 | struct cfq_group *curr_cfqg, int rw) { } | |
698 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, | |
699 | unsigned long time, unsigned long unaccounted_time) { } | |
700 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) { } | |
701 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) { } | |
702 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, | |
703 | uint64_t bytes, int rw) { } | |
704 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, | |
705 | uint64_t start_time, uint64_t io_start_time, int rw) { } | |
2ce4d50f | 706 | |
eb7d8c07 TH |
707 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
708 | ||
7b679138 JA |
709 | #define cfq_log(cfqd, fmt, args...) \ |
710 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
711 | ||
615f0259 VG |
712 | /* Traverses through cfq group service trees */ |
713 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
714 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
715 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
716 | : &cfqg->service_tree_idle; \ | |
717 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
718 | (i == IDLE_WORKLOAD && j == 0); \ | |
719 | j++, st = i < IDLE_WORKLOAD ? \ | |
720 | &cfqg->service_trees[i][j]: NULL) \ | |
721 | ||
f5f2b6ce SL |
722 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
723 | struct cfq_ttime *ttime, bool group_idle) | |
724 | { | |
725 | unsigned long slice; | |
726 | if (!sample_valid(ttime->ttime_samples)) | |
727 | return false; | |
728 | if (group_idle) | |
729 | slice = cfqd->cfq_group_idle; | |
730 | else | |
731 | slice = cfqd->cfq_slice_idle; | |
732 | return ttime->ttime_mean > slice; | |
733 | } | |
615f0259 | 734 | |
02b35081 VG |
735 | static inline bool iops_mode(struct cfq_data *cfqd) |
736 | { | |
737 | /* | |
738 | * If we are not idling on queues and it is a NCQ drive, parallel | |
739 | * execution of requests is on and measuring time is not possible | |
740 | * in most of the cases until and unless we drive shallower queue | |
741 | * depths and that becomes a performance bottleneck. In such cases | |
742 | * switch to start providing fairness in terms of number of IOs. | |
743 | */ | |
744 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
745 | return true; | |
746 | else | |
747 | return false; | |
748 | } | |
749 | ||
3bf10fea | 750 | static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq) |
c0324a02 CZ |
751 | { |
752 | if (cfq_class_idle(cfqq)) | |
753 | return IDLE_WORKLOAD; | |
754 | if (cfq_class_rt(cfqq)) | |
755 | return RT_WORKLOAD; | |
756 | return BE_WORKLOAD; | |
757 | } | |
758 | ||
718eee05 CZ |
759 | |
760 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
761 | { | |
762 | if (!cfq_cfqq_sync(cfqq)) | |
763 | return ASYNC_WORKLOAD; | |
764 | if (!cfq_cfqq_idle_window(cfqq)) | |
765 | return SYNC_NOIDLE_WORKLOAD; | |
766 | return SYNC_WORKLOAD; | |
767 | } | |
768 | ||
3bf10fea | 769 | static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class, |
58ff82f3 VG |
770 | struct cfq_data *cfqd, |
771 | struct cfq_group *cfqg) | |
c0324a02 | 772 | { |
3bf10fea | 773 | if (wl_class == IDLE_WORKLOAD) |
cdb16e8f | 774 | return cfqg->service_tree_idle.count; |
c0324a02 | 775 | |
34b98d03 VG |
776 | return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count + |
777 | cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count + | |
778 | cfqg->service_trees[wl_class][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
779 | } |
780 | ||
f26bd1f0 VG |
781 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
782 | struct cfq_group *cfqg) | |
783 | { | |
34b98d03 VG |
784 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count + |
785 | cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
f26bd1f0 VG |
786 | } |
787 | ||
165125e1 | 788 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 789 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
abede6da | 790 | struct cfq_io_cq *cic, struct bio *bio, |
4f85cb96 | 791 | gfp_t gfp_mask); |
91fac317 | 792 | |
c5869807 TH |
793 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
794 | { | |
795 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
796 | return container_of(icq, struct cfq_io_cq, icq); | |
797 | } | |
798 | ||
47fdd4ca TH |
799 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
800 | struct io_context *ioc) | |
801 | { | |
802 | if (ioc) | |
803 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
804 | return NULL; | |
805 | } | |
806 | ||
c5869807 | 807 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 808 | { |
a6151c3a | 809 | return cic->cfqq[is_sync]; |
91fac317 VT |
810 | } |
811 | ||
c5869807 TH |
812 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
813 | bool is_sync) | |
91fac317 | 814 | { |
a6151c3a | 815 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
816 | } |
817 | ||
c5869807 | 818 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 819 | { |
c5869807 | 820 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
821 | } |
822 | ||
91fac317 VT |
823 | /* |
824 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
825 | * set (in which case it could also be direct WRITE). | |
826 | */ | |
a6151c3a | 827 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 828 | { |
7b6d91da | 829 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 830 | } |
1da177e4 | 831 | |
99f95e52 AM |
832 | /* |
833 | * scheduler run of queue, if there are requests pending and no one in the | |
834 | * driver that will restart queueing | |
835 | */ | |
23e018a1 | 836 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 837 | { |
7b679138 JA |
838 | if (cfqd->busy_queues) { |
839 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 840 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 841 | } |
99f95e52 AM |
842 | } |
843 | ||
44f7c160 JA |
844 | /* |
845 | * Scale schedule slice based on io priority. Use the sync time slice only | |
846 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
847 | * io only, should not get full sync slice length. | |
848 | */ | |
a6151c3a | 849 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 850 | unsigned short prio) |
44f7c160 | 851 | { |
d9e7620e | 852 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 853 | |
d9e7620e JA |
854 | WARN_ON(prio >= IOPRIO_BE_NR); |
855 | ||
856 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
857 | } | |
44f7c160 | 858 | |
d9e7620e JA |
859 | static inline int |
860 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
861 | { | |
862 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
863 | } |
864 | ||
25bc6b07 VG |
865 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
866 | { | |
867 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
868 | ||
3381cb8d | 869 | d = d * CFQ_WEIGHT_DEFAULT; |
25bc6b07 VG |
870 | do_div(d, cfqg->weight); |
871 | return d; | |
872 | } | |
873 | ||
874 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
875 | { | |
876 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
877 | if (delta > 0) | |
878 | min_vdisktime = vdisktime; | |
879 | ||
880 | return min_vdisktime; | |
881 | } | |
882 | ||
883 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
884 | { | |
885 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
886 | if (delta < 0) | |
887 | min_vdisktime = vdisktime; | |
888 | ||
889 | return min_vdisktime; | |
890 | } | |
891 | ||
892 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
893 | { | |
25bc6b07 VG |
894 | struct cfq_group *cfqg; |
895 | ||
25bc6b07 VG |
896 | if (st->left) { |
897 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
898 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
899 | cfqg->vdisktime); | |
25bc6b07 | 900 | } |
25bc6b07 VG |
901 | } |
902 | ||
5db5d642 CZ |
903 | /* |
904 | * get averaged number of queues of RT/BE priority. | |
905 | * average is updated, with a formula that gives more weight to higher numbers, | |
906 | * to quickly follows sudden increases and decrease slowly | |
907 | */ | |
908 | ||
58ff82f3 VG |
909 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
910 | struct cfq_group *cfqg, bool rt) | |
5869619c | 911 | { |
5db5d642 CZ |
912 | unsigned min_q, max_q; |
913 | unsigned mult = cfq_hist_divisor - 1; | |
914 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 915 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 916 | |
58ff82f3 VG |
917 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
918 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
919 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 920 | cfq_hist_divisor; |
58ff82f3 VG |
921 | return cfqg->busy_queues_avg[rt]; |
922 | } | |
923 | ||
924 | static inline unsigned | |
925 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
926 | { | |
927 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
928 | ||
5bf14c07 | 929 | return cfqd->cfq_target_latency * cfqg->weight / st->total_weight; |
5db5d642 CZ |
930 | } |
931 | ||
c553f8e3 | 932 | static inline unsigned |
ba5bd520 | 933 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 934 | { |
5db5d642 CZ |
935 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
936 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
937 | /* |
938 | * interested queues (we consider only the ones with the same | |
939 | * priority class in the cfq group) | |
940 | */ | |
941 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
942 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
943 | unsigned sync_slice = cfqd->cfq_slice[1]; |
944 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
945 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
946 | ||
947 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
948 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
949 | /* scale low_slice according to IO priority | |
950 | * and sync vs async */ | |
951 | unsigned low_slice = | |
952 | min(slice, base_low_slice * slice / sync_slice); | |
953 | /* the adapted slice value is scaled to fit all iqs | |
954 | * into the target latency */ | |
58ff82f3 | 955 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
956 | low_slice); |
957 | } | |
958 | } | |
c553f8e3 SL |
959 | return slice; |
960 | } | |
961 | ||
962 | static inline void | |
963 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
964 | { | |
ba5bd520 | 965 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 966 | |
dae739eb | 967 | cfqq->slice_start = jiffies; |
5db5d642 | 968 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 969 | cfqq->allocated_slice = slice; |
7b679138 | 970 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
971 | } |
972 | ||
973 | /* | |
974 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
975 | * isn't valid until the first request from the dispatch is activated | |
976 | * and the slice time set. | |
977 | */ | |
a6151c3a | 978 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
979 | { |
980 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 981 | return false; |
44f7c160 | 982 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 983 | return false; |
44f7c160 | 984 | |
c1e44756 | 985 | return true; |
44f7c160 JA |
986 | } |
987 | ||
1da177e4 | 988 | /* |
5e705374 | 989 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 990 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 991 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 992 | */ |
5e705374 | 993 | static struct request * |
cf7c25cf | 994 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 995 | { |
cf7c25cf | 996 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 997 | unsigned long back_max; |
e8a99053 AM |
998 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
999 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
1000 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 1001 | |
5e705374 JA |
1002 | if (rq1 == NULL || rq1 == rq2) |
1003 | return rq2; | |
1004 | if (rq2 == NULL) | |
1005 | return rq1; | |
9c2c38a1 | 1006 | |
229836bd NK |
1007 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
1008 | return rq_is_sync(rq1) ? rq1 : rq2; | |
1009 | ||
65299a3b CH |
1010 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
1011 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 1012 | |
83096ebf TH |
1013 | s1 = blk_rq_pos(rq1); |
1014 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 1015 | |
1da177e4 LT |
1016 | /* |
1017 | * by definition, 1KiB is 2 sectors | |
1018 | */ | |
1019 | back_max = cfqd->cfq_back_max * 2; | |
1020 | ||
1021 | /* | |
1022 | * Strict one way elevator _except_ in the case where we allow | |
1023 | * short backward seeks which are biased as twice the cost of a | |
1024 | * similar forward seek. | |
1025 | */ | |
1026 | if (s1 >= last) | |
1027 | d1 = s1 - last; | |
1028 | else if (s1 + back_max >= last) | |
1029 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
1030 | else | |
e8a99053 | 1031 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
1032 | |
1033 | if (s2 >= last) | |
1034 | d2 = s2 - last; | |
1035 | else if (s2 + back_max >= last) | |
1036 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
1037 | else | |
e8a99053 | 1038 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
1039 | |
1040 | /* Found required data */ | |
e8a99053 AM |
1041 | |
1042 | /* | |
1043 | * By doing switch() on the bit mask "wrap" we avoid having to | |
1044 | * check two variables for all permutations: --> faster! | |
1045 | */ | |
1046 | switch (wrap) { | |
5e705374 | 1047 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 1048 | if (d1 < d2) |
5e705374 | 1049 | return rq1; |
e8a99053 | 1050 | else if (d2 < d1) |
5e705374 | 1051 | return rq2; |
e8a99053 AM |
1052 | else { |
1053 | if (s1 >= s2) | |
5e705374 | 1054 | return rq1; |
e8a99053 | 1055 | else |
5e705374 | 1056 | return rq2; |
e8a99053 | 1057 | } |
1da177e4 | 1058 | |
e8a99053 | 1059 | case CFQ_RQ2_WRAP: |
5e705374 | 1060 | return rq1; |
e8a99053 | 1061 | case CFQ_RQ1_WRAP: |
5e705374 JA |
1062 | return rq2; |
1063 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
1064 | default: |
1065 | /* | |
1066 | * Since both rqs are wrapped, | |
1067 | * start with the one that's further behind head | |
1068 | * (--> only *one* back seek required), | |
1069 | * since back seek takes more time than forward. | |
1070 | */ | |
1071 | if (s1 <= s2) | |
5e705374 | 1072 | return rq1; |
1da177e4 | 1073 | else |
5e705374 | 1074 | return rq2; |
1da177e4 LT |
1075 | } |
1076 | } | |
1077 | ||
498d3aa2 JA |
1078 | /* |
1079 | * The below is leftmost cache rbtree addon | |
1080 | */ | |
0871714e | 1081 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 1082 | { |
615f0259 VG |
1083 | /* Service tree is empty */ |
1084 | if (!root->count) | |
1085 | return NULL; | |
1086 | ||
cc09e299 JA |
1087 | if (!root->left) |
1088 | root->left = rb_first(&root->rb); | |
1089 | ||
0871714e JA |
1090 | if (root->left) |
1091 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
1092 | ||
1093 | return NULL; | |
cc09e299 JA |
1094 | } |
1095 | ||
1fa8f6d6 VG |
1096 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
1097 | { | |
1098 | if (!root->left) | |
1099 | root->left = rb_first(&root->rb); | |
1100 | ||
1101 | if (root->left) | |
1102 | return rb_entry_cfqg(root->left); | |
1103 | ||
1104 | return NULL; | |
1105 | } | |
1106 | ||
a36e71f9 JA |
1107 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
1108 | { | |
1109 | rb_erase(n, root); | |
1110 | RB_CLEAR_NODE(n); | |
1111 | } | |
1112 | ||
cc09e299 JA |
1113 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
1114 | { | |
1115 | if (root->left == n) | |
1116 | root->left = NULL; | |
a36e71f9 | 1117 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 1118 | --root->count; |
cc09e299 JA |
1119 | } |
1120 | ||
1da177e4 LT |
1121 | /* |
1122 | * would be nice to take fifo expire time into account as well | |
1123 | */ | |
5e705374 JA |
1124 | static struct request * |
1125 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1126 | struct request *last) | |
1da177e4 | 1127 | { |
21183b07 JA |
1128 | struct rb_node *rbnext = rb_next(&last->rb_node); |
1129 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 1130 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 1131 | |
21183b07 | 1132 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
1133 | |
1134 | if (rbprev) | |
5e705374 | 1135 | prev = rb_entry_rq(rbprev); |
1da177e4 | 1136 | |
21183b07 | 1137 | if (rbnext) |
5e705374 | 1138 | next = rb_entry_rq(rbnext); |
21183b07 JA |
1139 | else { |
1140 | rbnext = rb_first(&cfqq->sort_list); | |
1141 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 1142 | next = rb_entry_rq(rbnext); |
21183b07 | 1143 | } |
1da177e4 | 1144 | |
cf7c25cf | 1145 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
1146 | } |
1147 | ||
d9e7620e JA |
1148 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
1149 | struct cfq_queue *cfqq) | |
1da177e4 | 1150 | { |
d9e7620e JA |
1151 | /* |
1152 | * just an approximation, should be ok. | |
1153 | */ | |
cdb16e8f | 1154 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 1155 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
1156 | } |
1157 | ||
1fa8f6d6 VG |
1158 | static inline s64 |
1159 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1160 | { | |
1161 | return cfqg->vdisktime - st->min_vdisktime; | |
1162 | } | |
1163 | ||
1164 | static void | |
1165 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1166 | { | |
1167 | struct rb_node **node = &st->rb.rb_node; | |
1168 | struct rb_node *parent = NULL; | |
1169 | struct cfq_group *__cfqg; | |
1170 | s64 key = cfqg_key(st, cfqg); | |
1171 | int left = 1; | |
1172 | ||
1173 | while (*node != NULL) { | |
1174 | parent = *node; | |
1175 | __cfqg = rb_entry_cfqg(parent); | |
1176 | ||
1177 | if (key < cfqg_key(st, __cfqg)) | |
1178 | node = &parent->rb_left; | |
1179 | else { | |
1180 | node = &parent->rb_right; | |
1181 | left = 0; | |
1182 | } | |
1183 | } | |
1184 | ||
1185 | if (left) | |
1186 | st->left = &cfqg->rb_node; | |
1187 | ||
1188 | rb_link_node(&cfqg->rb_node, parent, node); | |
1189 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
1190 | } | |
1191 | ||
1192 | static void | |
8184f93e JT |
1193 | cfq_update_group_weight(struct cfq_group *cfqg) |
1194 | { | |
1195 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
e71357e1 | 1196 | |
3381cb8d | 1197 | if (cfqg->new_weight) { |
8184f93e | 1198 | cfqg->weight = cfqg->new_weight; |
3381cb8d | 1199 | cfqg->new_weight = 0; |
8184f93e | 1200 | } |
e71357e1 TH |
1201 | |
1202 | if (cfqg->new_leaf_weight) { | |
1203 | cfqg->leaf_weight = cfqg->new_leaf_weight; | |
1204 | cfqg->new_leaf_weight = 0; | |
1205 | } | |
8184f93e JT |
1206 | } |
1207 | ||
1208 | static void | |
1209 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1210 | { | |
1211 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
1212 | ||
1213 | cfq_update_group_weight(cfqg); | |
1214 | __cfq_group_service_tree_add(st, cfqg); | |
1215 | st->total_weight += cfqg->weight; | |
1216 | } | |
1217 | ||
1218 | static void | |
1219 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
1220 | { |
1221 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1222 | struct cfq_group *__cfqg; | |
1223 | struct rb_node *n; | |
1224 | ||
1225 | cfqg->nr_cfqq++; | |
760701bf | 1226 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
1227 | return; |
1228 | ||
1229 | /* | |
1230 | * Currently put the group at the end. Later implement something | |
1231 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 1232 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
1233 | */ |
1234 | n = rb_last(&st->rb); | |
1235 | if (n) { | |
1236 | __cfqg = rb_entry_cfqg(n); | |
1237 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
1238 | } else | |
1239 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
1240 | cfq_group_service_tree_add(st, cfqg); |
1241 | } | |
1fa8f6d6 | 1242 | |
8184f93e JT |
1243 | static void |
1244 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1245 | { | |
1246 | st->total_weight -= cfqg->weight; | |
1247 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) | |
1248 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
1249 | } |
1250 | ||
1251 | static void | |
8184f93e | 1252 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
1253 | { |
1254 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1255 | ||
1256 | BUG_ON(cfqg->nr_cfqq < 1); | |
1257 | cfqg->nr_cfqq--; | |
25bc6b07 | 1258 | |
1fa8f6d6 VG |
1259 | /* If there are other cfq queues under this group, don't delete it */ |
1260 | if (cfqg->nr_cfqq) | |
1261 | return; | |
1262 | ||
2868ef7b | 1263 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 1264 | cfq_group_service_tree_del(st, cfqg); |
4d2ceea4 | 1265 | cfqg->saved_wl_slice = 0; |
155fead9 | 1266 | cfqg_stats_update_dequeue(cfqg); |
dae739eb VG |
1267 | } |
1268 | ||
167400d3 JT |
1269 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
1270 | unsigned int *unaccounted_time) | |
dae739eb | 1271 | { |
f75edf2d | 1272 | unsigned int slice_used; |
dae739eb VG |
1273 | |
1274 | /* | |
1275 | * Queue got expired before even a single request completed or | |
1276 | * got expired immediately after first request completion. | |
1277 | */ | |
1278 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
1279 | /* | |
1280 | * Also charge the seek time incurred to the group, otherwise | |
1281 | * if there are mutiple queues in the group, each can dispatch | |
1282 | * a single request on seeky media and cause lots of seek time | |
1283 | * and group will never know it. | |
1284 | */ | |
1285 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
1286 | 1); | |
1287 | } else { | |
1288 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
1289 | if (slice_used > cfqq->allocated_slice) { |
1290 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 1291 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
1292 | } |
1293 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
1294 | *unaccounted_time += cfqq->slice_start - | |
1295 | cfqq->dispatch_start; | |
dae739eb VG |
1296 | } |
1297 | ||
dae739eb VG |
1298 | return slice_used; |
1299 | } | |
1300 | ||
1301 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1302 | struct cfq_queue *cfqq) |
dae739eb VG |
1303 | { |
1304 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 1305 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1306 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1307 | - cfqg->service_tree_idle.count; | |
1308 | ||
1309 | BUG_ON(nr_sync < 0); | |
167400d3 | 1310 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1311 | |
02b35081 VG |
1312 | if (iops_mode(cfqd)) |
1313 | charge = cfqq->slice_dispatch; | |
1314 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1315 | charge = cfqq->allocated_slice; | |
dae739eb VG |
1316 | |
1317 | /* Can't update vdisktime while group is on service tree */ | |
8184f93e | 1318 | cfq_group_service_tree_del(st, cfqg); |
02b35081 | 1319 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
8184f93e JT |
1320 | /* If a new weight was requested, update now, off tree */ |
1321 | cfq_group_service_tree_add(st, cfqg); | |
dae739eb VG |
1322 | |
1323 | /* This group is being expired. Save the context */ | |
1324 | if (time_after(cfqd->workload_expires, jiffies)) { | |
4d2ceea4 | 1325 | cfqg->saved_wl_slice = cfqd->workload_expires |
dae739eb | 1326 | - jiffies; |
4d2ceea4 VG |
1327 | cfqg->saved_wl_type = cfqd->serving_wl_type; |
1328 | cfqg->saved_wl_class = cfqd->serving_wl_class; | |
dae739eb | 1329 | } else |
4d2ceea4 | 1330 | cfqg->saved_wl_slice = 0; |
2868ef7b VG |
1331 | |
1332 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1333 | st->min_vdisktime); | |
fd16d263 JP |
1334 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1335 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1336 | used_sl, cfqq->slice_dispatch, charge, | |
1337 | iops_mode(cfqd), cfqq->nr_sectors); | |
155fead9 TH |
1338 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
1339 | cfqg_stats_set_start_empty_time(cfqg); | |
1fa8f6d6 VG |
1340 | } |
1341 | ||
f51b802c TH |
1342 | /** |
1343 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1344 | * @cfqg: cfq_group to initialize | |
1345 | * | |
1346 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1347 | * is enabled or not. | |
1348 | */ | |
1349 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1350 | { | |
1351 | struct cfq_rb_root *st; | |
1352 | int i, j; | |
1353 | ||
1354 | for_each_cfqg_st(cfqg, i, j, st) | |
1355 | *st = CFQ_RB_ROOT; | |
1356 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1357 | ||
1358 | cfqg->ttime.last_end_request = jiffies; | |
1359 | } | |
1360 | ||
25fb5169 | 1361 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 1362 | static void cfq_pd_init(struct blkcg_gq *blkg) |
f469a7b4 | 1363 | { |
0381411e | 1364 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
25fb5169 | 1365 | |
f51b802c | 1366 | cfq_init_cfqg_base(cfqg); |
3381cb8d | 1367 | cfqg->weight = blkg->blkcg->cfq_weight; |
e71357e1 | 1368 | cfqg->leaf_weight = blkg->blkcg->cfq_leaf_weight; |
25fb5169 VG |
1369 | } |
1370 | ||
1371 | /* | |
3e59cf9d VG |
1372 | * Search for the cfq group current task belongs to. request_queue lock must |
1373 | * be held. | |
25fb5169 | 1374 | */ |
cd1604fa | 1375 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
3c798398 | 1376 | struct blkcg *blkcg) |
25fb5169 | 1377 | { |
f469a7b4 | 1378 | struct request_queue *q = cfqd->queue; |
cd1604fa | 1379 | struct cfq_group *cfqg = NULL; |
25fb5169 | 1380 | |
3c798398 TH |
1381 | /* avoid lookup for the common case where there's no blkcg */ |
1382 | if (blkcg == &blkcg_root) { | |
cd1604fa TH |
1383 | cfqg = cfqd->root_group; |
1384 | } else { | |
3c798398 | 1385 | struct blkcg_gq *blkg; |
f469a7b4 | 1386 | |
3c96cb32 | 1387 | blkg = blkg_lookup_create(blkcg, q); |
cd1604fa | 1388 | if (!IS_ERR(blkg)) |
0381411e | 1389 | cfqg = blkg_to_cfqg(blkg); |
cd1604fa | 1390 | } |
f469a7b4 | 1391 | |
25fb5169 VG |
1392 | return cfqg; |
1393 | } | |
1394 | ||
1395 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1396 | { | |
1397 | /* Currently, all async queues are mapped to root group */ | |
1398 | if (!cfq_cfqq_sync(cfqq)) | |
f51b802c | 1399 | cfqg = cfqq->cfqd->root_group; |
25fb5169 VG |
1400 | |
1401 | cfqq->cfqg = cfqg; | |
b1c35769 | 1402 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1403 | cfqg_get(cfqg); |
b1c35769 VG |
1404 | } |
1405 | ||
f95a04af TH |
1406 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, |
1407 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1408 | { |
f95a04af | 1409 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
3381cb8d TH |
1410 | |
1411 | if (!cfqg->dev_weight) | |
60c2bc2d | 1412 | return 0; |
f95a04af | 1413 | return __blkg_prfill_u64(sf, pd, cfqg->dev_weight); |
60c2bc2d TH |
1414 | } |
1415 | ||
3381cb8d TH |
1416 | static int cfqg_print_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1417 | struct seq_file *sf) | |
60c2bc2d | 1418 | { |
3c798398 TH |
1419 | blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), |
1420 | cfqg_prfill_weight_device, &blkcg_policy_cfq, 0, | |
60c2bc2d TH |
1421 | false); |
1422 | return 0; | |
1423 | } | |
1424 | ||
e71357e1 TH |
1425 | static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf, |
1426 | struct blkg_policy_data *pd, int off) | |
1427 | { | |
1428 | struct cfq_group *cfqg = pd_to_cfqg(pd); | |
1429 | ||
1430 | if (!cfqg->dev_leaf_weight) | |
1431 | return 0; | |
1432 | return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight); | |
1433 | } | |
1434 | ||
1435 | static int cfqg_print_leaf_weight_device(struct cgroup *cgrp, | |
1436 | struct cftype *cft, | |
1437 | struct seq_file *sf) | |
1438 | { | |
1439 | blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), | |
1440 | cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq, 0, | |
1441 | false); | |
1442 | return 0; | |
1443 | } | |
1444 | ||
3381cb8d TH |
1445 | static int cfq_print_weight(struct cgroup *cgrp, struct cftype *cft, |
1446 | struct seq_file *sf) | |
60c2bc2d | 1447 | { |
3c798398 | 1448 | seq_printf(sf, "%u\n", cgroup_to_blkcg(cgrp)->cfq_weight); |
60c2bc2d TH |
1449 | return 0; |
1450 | } | |
1451 | ||
e71357e1 TH |
1452 | static int cfq_print_leaf_weight(struct cgroup *cgrp, struct cftype *cft, |
1453 | struct seq_file *sf) | |
1454 | { | |
1455 | seq_printf(sf, "%u\n", | |
1456 | cgroup_to_blkcg(cgrp)->cfq_leaf_weight); | |
1457 | return 0; | |
1458 | } | |
1459 | ||
1460 | static int __cfqg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, | |
1461 | const char *buf, bool is_leaf_weight) | |
60c2bc2d | 1462 | { |
3c798398 | 1463 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
60c2bc2d | 1464 | struct blkg_conf_ctx ctx; |
3381cb8d | 1465 | struct cfq_group *cfqg; |
60c2bc2d TH |
1466 | int ret; |
1467 | ||
3c798398 | 1468 | ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx); |
60c2bc2d TH |
1469 | if (ret) |
1470 | return ret; | |
1471 | ||
1472 | ret = -EINVAL; | |
3381cb8d | 1473 | cfqg = blkg_to_cfqg(ctx.blkg); |
a2b1693b | 1474 | if (!ctx.v || (ctx.v >= CFQ_WEIGHT_MIN && ctx.v <= CFQ_WEIGHT_MAX)) { |
e71357e1 TH |
1475 | if (!is_leaf_weight) { |
1476 | cfqg->dev_weight = ctx.v; | |
1477 | cfqg->new_weight = ctx.v ?: blkcg->cfq_weight; | |
1478 | } else { | |
1479 | cfqg->dev_leaf_weight = ctx.v; | |
1480 | cfqg->new_leaf_weight = ctx.v ?: blkcg->cfq_leaf_weight; | |
1481 | } | |
60c2bc2d TH |
1482 | ret = 0; |
1483 | } | |
1484 | ||
1485 | blkg_conf_finish(&ctx); | |
1486 | return ret; | |
1487 | } | |
1488 | ||
e71357e1 TH |
1489 | static int cfqg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1490 | const char *buf) | |
1491 | { | |
1492 | return __cfqg_set_weight_device(cgrp, cft, buf, false); | |
1493 | } | |
1494 | ||
1495 | static int cfqg_set_leaf_weight_device(struct cgroup *cgrp, struct cftype *cft, | |
1496 | const char *buf) | |
1497 | { | |
1498 | return __cfqg_set_weight_device(cgrp, cft, buf, true); | |
1499 | } | |
1500 | ||
1501 | static int __cfq_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val, | |
1502 | bool is_leaf_weight) | |
60c2bc2d | 1503 | { |
3c798398 TH |
1504 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
1505 | struct blkcg_gq *blkg; | |
60c2bc2d TH |
1506 | struct hlist_node *n; |
1507 | ||
3381cb8d | 1508 | if (val < CFQ_WEIGHT_MIN || val > CFQ_WEIGHT_MAX) |
60c2bc2d TH |
1509 | return -EINVAL; |
1510 | ||
1511 | spin_lock_irq(&blkcg->lock); | |
e71357e1 TH |
1512 | |
1513 | if (!is_leaf_weight) | |
1514 | blkcg->cfq_weight = val; | |
1515 | else | |
1516 | blkcg->cfq_leaf_weight = val; | |
60c2bc2d TH |
1517 | |
1518 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { | |
3381cb8d | 1519 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
60c2bc2d | 1520 | |
e71357e1 TH |
1521 | if (!cfqg) |
1522 | continue; | |
1523 | ||
1524 | if (!is_leaf_weight) { | |
1525 | if (!cfqg->dev_weight) | |
1526 | cfqg->new_weight = blkcg->cfq_weight; | |
1527 | } else { | |
1528 | if (!cfqg->dev_leaf_weight) | |
1529 | cfqg->new_leaf_weight = blkcg->cfq_leaf_weight; | |
1530 | } | |
60c2bc2d TH |
1531 | } |
1532 | ||
1533 | spin_unlock_irq(&blkcg->lock); | |
1534 | return 0; | |
1535 | } | |
1536 | ||
e71357e1 TH |
1537 | static int cfq_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) |
1538 | { | |
1539 | return __cfq_set_weight(cgrp, cft, val, false); | |
1540 | } | |
1541 | ||
1542 | static int cfq_set_leaf_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1543 | { | |
1544 | return __cfq_set_weight(cgrp, cft, val, true); | |
1545 | } | |
1546 | ||
5bc4afb1 TH |
1547 | static int cfqg_print_stat(struct cgroup *cgrp, struct cftype *cft, |
1548 | struct seq_file *sf) | |
1549 | { | |
3c798398 | 1550 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
5bc4afb1 | 1551 | |
3c798398 | 1552 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_stat, &blkcg_policy_cfq, |
5bc4afb1 TH |
1553 | cft->private, false); |
1554 | return 0; | |
1555 | } | |
1556 | ||
1557 | static int cfqg_print_rwstat(struct cgroup *cgrp, struct cftype *cft, | |
1558 | struct seq_file *sf) | |
1559 | { | |
3c798398 | 1560 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
5bc4afb1 | 1561 | |
3c798398 | 1562 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_rwstat, &blkcg_policy_cfq, |
5bc4afb1 TH |
1563 | cft->private, true); |
1564 | return 0; | |
1565 | } | |
1566 | ||
60c2bc2d | 1567 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
f95a04af TH |
1568 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, |
1569 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1570 | { |
f95a04af | 1571 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
155fead9 | 1572 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
60c2bc2d TH |
1573 | u64 v = 0; |
1574 | ||
1575 | if (samples) { | |
155fead9 | 1576 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
60c2bc2d TH |
1577 | do_div(v, samples); |
1578 | } | |
f95a04af | 1579 | __blkg_prfill_u64(sf, pd, v); |
60c2bc2d TH |
1580 | return 0; |
1581 | } | |
1582 | ||
1583 | /* print avg_queue_size */ | |
155fead9 TH |
1584 | static int cfqg_print_avg_queue_size(struct cgroup *cgrp, struct cftype *cft, |
1585 | struct seq_file *sf) | |
60c2bc2d | 1586 | { |
3c798398 | 1587 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
60c2bc2d | 1588 | |
155fead9 | 1589 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_avg_queue_size, |
3c798398 | 1590 | &blkcg_policy_cfq, 0, false); |
60c2bc2d TH |
1591 | return 0; |
1592 | } | |
1593 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1594 | ||
1595 | static struct cftype cfq_blkcg_files[] = { | |
1596 | { | |
1597 | .name = "weight_device", | |
3381cb8d TH |
1598 | .read_seq_string = cfqg_print_weight_device, |
1599 | .write_string = cfqg_set_weight_device, | |
60c2bc2d TH |
1600 | .max_write_len = 256, |
1601 | }, | |
1602 | { | |
1603 | .name = "weight", | |
3381cb8d TH |
1604 | .read_seq_string = cfq_print_weight, |
1605 | .write_u64 = cfq_set_weight, | |
60c2bc2d | 1606 | }, |
e71357e1 TH |
1607 | |
1608 | /* on root, leaf_weight is mapped to weight */ | |
1609 | { | |
1610 | .name = "leaf_weight_device", | |
1611 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1612 | .read_seq_string = cfqg_print_weight_device, | |
1613 | .write_string = cfqg_set_weight_device, | |
1614 | .max_write_len = 256, | |
1615 | }, | |
1616 | { | |
1617 | .name = "leaf_weight", | |
1618 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1619 | .read_seq_string = cfq_print_weight, | |
1620 | .write_u64 = cfq_set_weight, | |
1621 | }, | |
1622 | ||
1623 | /* no such mapping necessary for !roots */ | |
1624 | { | |
1625 | .name = "leaf_weight_device", | |
1626 | .flags = CFTYPE_NOT_ON_ROOT, | |
1627 | .read_seq_string = cfqg_print_leaf_weight_device, | |
1628 | .write_string = cfqg_set_leaf_weight_device, | |
1629 | .max_write_len = 256, | |
1630 | }, | |
1631 | { | |
1632 | .name = "leaf_weight", | |
1633 | .flags = CFTYPE_NOT_ON_ROOT, | |
1634 | .read_seq_string = cfq_print_leaf_weight, | |
1635 | .write_u64 = cfq_set_leaf_weight, | |
1636 | }, | |
1637 | ||
60c2bc2d TH |
1638 | { |
1639 | .name = "time", | |
5bc4afb1 TH |
1640 | .private = offsetof(struct cfq_group, stats.time), |
1641 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1642 | }, |
1643 | { | |
1644 | .name = "sectors", | |
5bc4afb1 TH |
1645 | .private = offsetof(struct cfq_group, stats.sectors), |
1646 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1647 | }, |
1648 | { | |
1649 | .name = "io_service_bytes", | |
5bc4afb1 TH |
1650 | .private = offsetof(struct cfq_group, stats.service_bytes), |
1651 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1652 | }, |
1653 | { | |
1654 | .name = "io_serviced", | |
5bc4afb1 TH |
1655 | .private = offsetof(struct cfq_group, stats.serviced), |
1656 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1657 | }, |
1658 | { | |
1659 | .name = "io_service_time", | |
5bc4afb1 TH |
1660 | .private = offsetof(struct cfq_group, stats.service_time), |
1661 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1662 | }, |
1663 | { | |
1664 | .name = "io_wait_time", | |
5bc4afb1 TH |
1665 | .private = offsetof(struct cfq_group, stats.wait_time), |
1666 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1667 | }, |
1668 | { | |
1669 | .name = "io_merged", | |
5bc4afb1 TH |
1670 | .private = offsetof(struct cfq_group, stats.merged), |
1671 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1672 | }, |
1673 | { | |
1674 | .name = "io_queued", | |
5bc4afb1 TH |
1675 | .private = offsetof(struct cfq_group, stats.queued), |
1676 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1677 | }, |
1678 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
1679 | { | |
1680 | .name = "avg_queue_size", | |
155fead9 | 1681 | .read_seq_string = cfqg_print_avg_queue_size, |
60c2bc2d TH |
1682 | }, |
1683 | { | |
1684 | .name = "group_wait_time", | |
5bc4afb1 TH |
1685 | .private = offsetof(struct cfq_group, stats.group_wait_time), |
1686 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1687 | }, |
1688 | { | |
1689 | .name = "idle_time", | |
5bc4afb1 TH |
1690 | .private = offsetof(struct cfq_group, stats.idle_time), |
1691 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1692 | }, |
1693 | { | |
1694 | .name = "empty_time", | |
5bc4afb1 TH |
1695 | .private = offsetof(struct cfq_group, stats.empty_time), |
1696 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1697 | }, |
1698 | { | |
1699 | .name = "dequeue", | |
5bc4afb1 TH |
1700 | .private = offsetof(struct cfq_group, stats.dequeue), |
1701 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1702 | }, |
1703 | { | |
1704 | .name = "unaccounted_time", | |
5bc4afb1 TH |
1705 | .private = offsetof(struct cfq_group, stats.unaccounted_time), |
1706 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1707 | }, |
1708 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1709 | { } /* terminate */ | |
1710 | }; | |
25fb5169 | 1711 | #else /* GROUP_IOSCHED */ |
cd1604fa | 1712 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
3c798398 | 1713 | struct blkcg *blkcg) |
25fb5169 | 1714 | { |
f51b802c | 1715 | return cfqd->root_group; |
25fb5169 | 1716 | } |
7f1dc8a2 | 1717 | |
25fb5169 VG |
1718 | static inline void |
1719 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1720 | cfqq->cfqg = cfqg; | |
1721 | } | |
1722 | ||
1723 | #endif /* GROUP_IOSCHED */ | |
1724 | ||
498d3aa2 | 1725 | /* |
c0324a02 | 1726 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1727 | * requests waiting to be processed. It is sorted in the order that |
1728 | * we will service the queues. | |
1729 | */ | |
a36e71f9 | 1730 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1731 | bool add_front) |
d9e7620e | 1732 | { |
0871714e JA |
1733 | struct rb_node **p, *parent; |
1734 | struct cfq_queue *__cfqq; | |
d9e7620e | 1735 | unsigned long rb_key; |
34b98d03 | 1736 | struct cfq_rb_root *st; |
498d3aa2 | 1737 | int left; |
dae739eb | 1738 | int new_cfqq = 1; |
ae30c286 | 1739 | |
34b98d03 | 1740 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq)); |
0871714e JA |
1741 | if (cfq_class_idle(cfqq)) { |
1742 | rb_key = CFQ_IDLE_DELAY; | |
34b98d03 | 1743 | parent = rb_last(&st->rb); |
0871714e JA |
1744 | if (parent && parent != &cfqq->rb_node) { |
1745 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1746 | rb_key += __cfqq->rb_key; | |
1747 | } else | |
1748 | rb_key += jiffies; | |
1749 | } else if (!add_front) { | |
b9c8946b JA |
1750 | /* |
1751 | * Get our rb key offset. Subtract any residual slice | |
1752 | * value carried from last service. A negative resid | |
1753 | * count indicates slice overrun, and this should position | |
1754 | * the next service time further away in the tree. | |
1755 | */ | |
edd75ffd | 1756 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1757 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1758 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1759 | } else { |
1760 | rb_key = -HZ; | |
34b98d03 | 1761 | __cfqq = cfq_rb_first(st); |
48e025e6 CZ |
1762 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1763 | } | |
1da177e4 | 1764 | |
d9e7620e | 1765 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1766 | new_cfqq = 0; |
99f9628a | 1767 | /* |
d9e7620e | 1768 | * same position, nothing more to do |
99f9628a | 1769 | */ |
34b98d03 | 1770 | if (rb_key == cfqq->rb_key && cfqq->service_tree == st) |
d9e7620e | 1771 | return; |
1da177e4 | 1772 | |
aa6f6a3d CZ |
1773 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1774 | cfqq->service_tree = NULL; | |
1da177e4 | 1775 | } |
d9e7620e | 1776 | |
498d3aa2 | 1777 | left = 1; |
0871714e | 1778 | parent = NULL; |
34b98d03 VG |
1779 | cfqq->service_tree = st; |
1780 | p = &st->rb.rb_node; | |
d9e7620e JA |
1781 | while (*p) { |
1782 | parent = *p; | |
1783 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1784 | ||
0c534e0a | 1785 | /* |
c0324a02 | 1786 | * sort by key, that represents service time. |
0c534e0a | 1787 | */ |
c0324a02 | 1788 | if (time_before(rb_key, __cfqq->rb_key)) |
1f23f121 | 1789 | p = &parent->rb_left; |
c0324a02 | 1790 | else { |
1f23f121 | 1791 | p = &parent->rb_right; |
cc09e299 | 1792 | left = 0; |
c0324a02 | 1793 | } |
d9e7620e JA |
1794 | } |
1795 | ||
cc09e299 | 1796 | if (left) |
34b98d03 | 1797 | st->left = &cfqq->rb_node; |
cc09e299 | 1798 | |
d9e7620e JA |
1799 | cfqq->rb_key = rb_key; |
1800 | rb_link_node(&cfqq->rb_node, parent, p); | |
34b98d03 VG |
1801 | rb_insert_color(&cfqq->rb_node, &st->rb); |
1802 | st->count++; | |
20359f27 | 1803 | if (add_front || !new_cfqq) |
dae739eb | 1804 | return; |
8184f93e | 1805 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1806 | } |
1807 | ||
a36e71f9 | 1808 | static struct cfq_queue * |
f2d1f0ae JA |
1809 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1810 | sector_t sector, struct rb_node **ret_parent, | |
1811 | struct rb_node ***rb_link) | |
a36e71f9 | 1812 | { |
a36e71f9 JA |
1813 | struct rb_node **p, *parent; |
1814 | struct cfq_queue *cfqq = NULL; | |
1815 | ||
1816 | parent = NULL; | |
1817 | p = &root->rb_node; | |
1818 | while (*p) { | |
1819 | struct rb_node **n; | |
1820 | ||
1821 | parent = *p; | |
1822 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1823 | ||
1824 | /* | |
1825 | * Sort strictly based on sector. Smallest to the left, | |
1826 | * largest to the right. | |
1827 | */ | |
2e46e8b2 | 1828 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1829 | n = &(*p)->rb_right; |
2e46e8b2 | 1830 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1831 | n = &(*p)->rb_left; |
1832 | else | |
1833 | break; | |
1834 | p = n; | |
3ac6c9f8 | 1835 | cfqq = NULL; |
a36e71f9 JA |
1836 | } |
1837 | ||
1838 | *ret_parent = parent; | |
1839 | if (rb_link) | |
1840 | *rb_link = p; | |
3ac6c9f8 | 1841 | return cfqq; |
a36e71f9 JA |
1842 | } |
1843 | ||
1844 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1845 | { | |
a36e71f9 JA |
1846 | struct rb_node **p, *parent; |
1847 | struct cfq_queue *__cfqq; | |
1848 | ||
f2d1f0ae JA |
1849 | if (cfqq->p_root) { |
1850 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1851 | cfqq->p_root = NULL; | |
1852 | } | |
a36e71f9 JA |
1853 | |
1854 | if (cfq_class_idle(cfqq)) | |
1855 | return; | |
1856 | if (!cfqq->next_rq) | |
1857 | return; | |
1858 | ||
f2d1f0ae | 1859 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1860 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1861 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1862 | if (!__cfqq) { |
1863 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1864 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1865 | } else | |
1866 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1867 | } |
1868 | ||
498d3aa2 JA |
1869 | /* |
1870 | * Update cfqq's position in the service tree. | |
1871 | */ | |
edd75ffd | 1872 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1873 | { |
6d048f53 JA |
1874 | /* |
1875 | * Resorting requires the cfqq to be on the RR list already. | |
1876 | */ | |
a36e71f9 | 1877 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1878 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1879 | cfq_prio_tree_add(cfqd, cfqq); |
1880 | } | |
6d048f53 JA |
1881 | } |
1882 | ||
1da177e4 LT |
1883 | /* |
1884 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1885 | * the pending list according to last request service |
1da177e4 | 1886 | */ |
febffd61 | 1887 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1888 | { |
7b679138 | 1889 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1890 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1891 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 1892 | cfqd->busy_queues++; |
ef8a41df SL |
1893 | if (cfq_cfqq_sync(cfqq)) |
1894 | cfqd->busy_sync_queues++; | |
1da177e4 | 1895 | |
edd75ffd | 1896 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1897 | } |
1898 | ||
498d3aa2 JA |
1899 | /* |
1900 | * Called when the cfqq no longer has requests pending, remove it from | |
1901 | * the service tree. | |
1902 | */ | |
febffd61 | 1903 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1904 | { |
7b679138 | 1905 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1906 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1907 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1908 | |
aa6f6a3d CZ |
1909 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1910 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1911 | cfqq->service_tree = NULL; | |
1912 | } | |
f2d1f0ae JA |
1913 | if (cfqq->p_root) { |
1914 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1915 | cfqq->p_root = NULL; | |
1916 | } | |
d9e7620e | 1917 | |
8184f93e | 1918 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1919 | BUG_ON(!cfqd->busy_queues); |
1920 | cfqd->busy_queues--; | |
ef8a41df SL |
1921 | if (cfq_cfqq_sync(cfqq)) |
1922 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
1923 | } |
1924 | ||
1925 | /* | |
1926 | * rb tree support functions | |
1927 | */ | |
febffd61 | 1928 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1929 | { |
5e705374 | 1930 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1931 | const int sync = rq_is_sync(rq); |
1da177e4 | 1932 | |
b4878f24 JA |
1933 | BUG_ON(!cfqq->queued[sync]); |
1934 | cfqq->queued[sync]--; | |
1da177e4 | 1935 | |
5e705374 | 1936 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1937 | |
f04a6424 VG |
1938 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1939 | /* | |
1940 | * Queue will be deleted from service tree when we actually | |
1941 | * expire it later. Right now just remove it from prio tree | |
1942 | * as it is empty. | |
1943 | */ | |
1944 | if (cfqq->p_root) { | |
1945 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1946 | cfqq->p_root = NULL; | |
1947 | } | |
1948 | } | |
1da177e4 LT |
1949 | } |
1950 | ||
5e705374 | 1951 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1952 | { |
5e705374 | 1953 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1954 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 1955 | struct request *prev; |
1da177e4 | 1956 | |
5380a101 | 1957 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 1958 | |
796d5116 | 1959 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
1960 | |
1961 | if (!cfq_cfqq_on_rr(cfqq)) | |
1962 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1963 | |
1964 | /* | |
1965 | * check if this request is a better next-serve candidate | |
1966 | */ | |
a36e71f9 | 1967 | prev = cfqq->next_rq; |
cf7c25cf | 1968 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1969 | |
1970 | /* | |
1971 | * adjust priority tree position, if ->next_rq changes | |
1972 | */ | |
1973 | if (prev != cfqq->next_rq) | |
1974 | cfq_prio_tree_add(cfqd, cfqq); | |
1975 | ||
5044eed4 | 1976 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1977 | } |
1978 | ||
febffd61 | 1979 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1980 | { |
5380a101 JA |
1981 | elv_rb_del(&cfqq->sort_list, rq); |
1982 | cfqq->queued[rq_is_sync(rq)]--; | |
155fead9 | 1983 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
5e705374 | 1984 | cfq_add_rq_rb(rq); |
155fead9 TH |
1985 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
1986 | rq->cmd_flags); | |
1da177e4 LT |
1987 | } |
1988 | ||
206dc69b JA |
1989 | static struct request * |
1990 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1991 | { |
206dc69b | 1992 | struct task_struct *tsk = current; |
c5869807 | 1993 | struct cfq_io_cq *cic; |
206dc69b | 1994 | struct cfq_queue *cfqq; |
1da177e4 | 1995 | |
4ac845a2 | 1996 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1997 | if (!cic) |
1998 | return NULL; | |
1999 | ||
2000 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
2001 | if (cfqq) { |
2002 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
2003 | ||
21183b07 | 2004 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 2005 | } |
1da177e4 | 2006 | |
1da177e4 LT |
2007 | return NULL; |
2008 | } | |
2009 | ||
165125e1 | 2010 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2011 | { |
22e2c507 | 2012 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 2013 | |
53c583d2 | 2014 | cfqd->rq_in_driver++; |
7b679138 | 2015 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 2016 | cfqd->rq_in_driver); |
25776e35 | 2017 | |
5b93629b | 2018 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
2019 | } |
2020 | ||
165125e1 | 2021 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2022 | { |
b4878f24 JA |
2023 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2024 | ||
53c583d2 CZ |
2025 | WARN_ON(!cfqd->rq_in_driver); |
2026 | cfqd->rq_in_driver--; | |
7b679138 | 2027 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 2028 | cfqd->rq_in_driver); |
1da177e4 LT |
2029 | } |
2030 | ||
b4878f24 | 2031 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 2032 | { |
5e705374 | 2033 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 2034 | |
5e705374 JA |
2035 | if (cfqq->next_rq == rq) |
2036 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 2037 | |
b4878f24 | 2038 | list_del_init(&rq->queuelist); |
5e705374 | 2039 | cfq_del_rq_rb(rq); |
374f84ac | 2040 | |
45333d5a | 2041 | cfqq->cfqd->rq_queued--; |
155fead9 | 2042 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
65299a3b CH |
2043 | if (rq->cmd_flags & REQ_PRIO) { |
2044 | WARN_ON(!cfqq->prio_pending); | |
2045 | cfqq->prio_pending--; | |
b53d1ed7 | 2046 | } |
1da177e4 LT |
2047 | } |
2048 | ||
165125e1 JA |
2049 | static int cfq_merge(struct request_queue *q, struct request **req, |
2050 | struct bio *bio) | |
1da177e4 LT |
2051 | { |
2052 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2053 | struct request *__rq; | |
1da177e4 | 2054 | |
206dc69b | 2055 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 2056 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
2057 | *req = __rq; |
2058 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
2059 | } |
2060 | ||
2061 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
2062 | } |
2063 | ||
165125e1 | 2064 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 2065 | int type) |
1da177e4 | 2066 | { |
21183b07 | 2067 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 2068 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 2069 | |
5e705374 | 2070 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 2071 | } |
1da177e4 LT |
2072 | } |
2073 | ||
812d4026 DS |
2074 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
2075 | struct bio *bio) | |
2076 | { | |
155fead9 | 2077 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_rw); |
812d4026 DS |
2078 | } |
2079 | ||
1da177e4 | 2080 | static void |
165125e1 | 2081 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
2082 | struct request *next) |
2083 | { | |
cf7c25cf | 2084 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
2085 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2086 | ||
22e2c507 JA |
2087 | /* |
2088 | * reposition in fifo if next is older than rq | |
2089 | */ | |
2090 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
3d106fba SL |
2091 | time_before(rq_fifo_time(next), rq_fifo_time(rq)) && |
2092 | cfqq == RQ_CFQQ(next)) { | |
22e2c507 | 2093 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
2094 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
2095 | } | |
22e2c507 | 2096 | |
cf7c25cf CZ |
2097 | if (cfqq->next_rq == next) |
2098 | cfqq->next_rq = rq; | |
b4878f24 | 2099 | cfq_remove_request(next); |
155fead9 | 2100 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
4a0b75c7 SL |
2101 | |
2102 | cfqq = RQ_CFQQ(next); | |
2103 | /* | |
2104 | * all requests of this queue are merged to other queues, delete it | |
2105 | * from the service tree. If it's the active_queue, | |
2106 | * cfq_dispatch_requests() will choose to expire it or do idle | |
2107 | */ | |
2108 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
2109 | cfqq != cfqd->active_queue) | |
2110 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
2111 | } |
2112 | ||
165125e1 | 2113 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
2114 | struct bio *bio) |
2115 | { | |
2116 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 2117 | struct cfq_io_cq *cic; |
da775265 | 2118 | struct cfq_queue *cfqq; |
da775265 JA |
2119 | |
2120 | /* | |
ec8acb69 | 2121 | * Disallow merge of a sync bio into an async request. |
da775265 | 2122 | */ |
91fac317 | 2123 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 2124 | return false; |
da775265 JA |
2125 | |
2126 | /* | |
f1a4f4d3 | 2127 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 2128 | * merge only if rq is queued there. |
f1a4f4d3 | 2129 | */ |
07c2bd37 TH |
2130 | cic = cfq_cic_lookup(cfqd, current->io_context); |
2131 | if (!cic) | |
2132 | return false; | |
719d3402 | 2133 | |
91fac317 | 2134 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 2135 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
2136 | } |
2137 | ||
812df48d DS |
2138 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2139 | { | |
2140 | del_timer(&cfqd->idle_slice_timer); | |
155fead9 | 2141 | cfqg_stats_update_idle_time(cfqq->cfqg); |
812df48d DS |
2142 | } |
2143 | ||
febffd61 JA |
2144 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
2145 | struct cfq_queue *cfqq) | |
22e2c507 JA |
2146 | { |
2147 | if (cfqq) { | |
3bf10fea | 2148 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d", |
4d2ceea4 | 2149 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
155fead9 | 2150 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
62a37f6b JT |
2151 | cfqq->slice_start = 0; |
2152 | cfqq->dispatch_start = jiffies; | |
2153 | cfqq->allocated_slice = 0; | |
2154 | cfqq->slice_end = 0; | |
2155 | cfqq->slice_dispatch = 0; | |
2156 | cfqq->nr_sectors = 0; | |
2157 | ||
2158 | cfq_clear_cfqq_wait_request(cfqq); | |
2159 | cfq_clear_cfqq_must_dispatch(cfqq); | |
2160 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
2161 | cfq_clear_cfqq_fifo_expire(cfqq); | |
2162 | cfq_mark_cfqq_slice_new(cfqq); | |
2163 | ||
2164 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
2165 | } |
2166 | ||
2167 | cfqd->active_queue = cfqq; | |
2168 | } | |
2169 | ||
7b14e3b5 JA |
2170 | /* |
2171 | * current cfqq expired its slice (or was too idle), select new one | |
2172 | */ | |
2173 | static void | |
2174 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 2175 | bool timed_out) |
7b14e3b5 | 2176 | { |
7b679138 JA |
2177 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
2178 | ||
7b14e3b5 | 2179 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 2180 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 2181 | |
7b14e3b5 | 2182 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 2183 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 2184 | |
ae54abed SL |
2185 | /* |
2186 | * If this cfqq is shared between multiple processes, check to | |
2187 | * make sure that those processes are still issuing I/Os within | |
2188 | * the mean seek distance. If not, it may be time to break the | |
2189 | * queues apart again. | |
2190 | */ | |
2191 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
2192 | cfq_mark_cfqq_split_coop(cfqq); | |
2193 | ||
7b14e3b5 | 2194 | /* |
6084cdda | 2195 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 2196 | */ |
c553f8e3 SL |
2197 | if (timed_out) { |
2198 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 2199 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
2200 | else |
2201 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
2202 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
2203 | } | |
7b14e3b5 | 2204 | |
e5ff082e | 2205 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 2206 | |
f04a6424 VG |
2207 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
2208 | cfq_del_cfqq_rr(cfqd, cfqq); | |
2209 | ||
edd75ffd | 2210 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
2211 | |
2212 | if (cfqq == cfqd->active_queue) | |
2213 | cfqd->active_queue = NULL; | |
2214 | ||
2215 | if (cfqd->active_cic) { | |
11a3122f | 2216 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
2217 | cfqd->active_cic = NULL; |
2218 | } | |
7b14e3b5 JA |
2219 | } |
2220 | ||
e5ff082e | 2221 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
2222 | { |
2223 | struct cfq_queue *cfqq = cfqd->active_queue; | |
2224 | ||
2225 | if (cfqq) | |
e5ff082e | 2226 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
2227 | } |
2228 | ||
498d3aa2 JA |
2229 | /* |
2230 | * Get next queue for service. Unless we have a queue preemption, | |
2231 | * we'll simply select the first cfqq in the service tree. | |
2232 | */ | |
6d048f53 | 2233 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 2234 | { |
34b98d03 VG |
2235 | struct cfq_rb_root *st = st_for(cfqd->serving_group, |
2236 | cfqd->serving_wl_class, cfqd->serving_wl_type); | |
d9e7620e | 2237 | |
f04a6424 VG |
2238 | if (!cfqd->rq_queued) |
2239 | return NULL; | |
2240 | ||
1fa8f6d6 | 2241 | /* There is nothing to dispatch */ |
34b98d03 | 2242 | if (!st) |
1fa8f6d6 | 2243 | return NULL; |
34b98d03 | 2244 | if (RB_EMPTY_ROOT(&st->rb)) |
c0324a02 | 2245 | return NULL; |
34b98d03 | 2246 | return cfq_rb_first(st); |
6d048f53 JA |
2247 | } |
2248 | ||
f04a6424 VG |
2249 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
2250 | { | |
25fb5169 | 2251 | struct cfq_group *cfqg; |
f04a6424 VG |
2252 | struct cfq_queue *cfqq; |
2253 | int i, j; | |
2254 | struct cfq_rb_root *st; | |
2255 | ||
2256 | if (!cfqd->rq_queued) | |
2257 | return NULL; | |
2258 | ||
25fb5169 VG |
2259 | cfqg = cfq_get_next_cfqg(cfqd); |
2260 | if (!cfqg) | |
2261 | return NULL; | |
2262 | ||
f04a6424 VG |
2263 | for_each_cfqg_st(cfqg, i, j, st) |
2264 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
2265 | return cfqq; | |
2266 | return NULL; | |
2267 | } | |
2268 | ||
498d3aa2 JA |
2269 | /* |
2270 | * Get and set a new active queue for service. | |
2271 | */ | |
a36e71f9 JA |
2272 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
2273 | struct cfq_queue *cfqq) | |
6d048f53 | 2274 | { |
e00ef799 | 2275 | if (!cfqq) |
a36e71f9 | 2276 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 2277 | |
22e2c507 | 2278 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 2279 | return cfqq; |
22e2c507 JA |
2280 | } |
2281 | ||
d9e7620e JA |
2282 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
2283 | struct request *rq) | |
2284 | { | |
83096ebf TH |
2285 | if (blk_rq_pos(rq) >= cfqd->last_position) |
2286 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 2287 | else |
83096ebf | 2288 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
2289 | } |
2290 | ||
b2c18e1e | 2291 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 2292 | struct request *rq) |
6d048f53 | 2293 | { |
e9ce335d | 2294 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
2295 | } |
2296 | ||
a36e71f9 JA |
2297 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
2298 | struct cfq_queue *cur_cfqq) | |
2299 | { | |
f2d1f0ae | 2300 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
2301 | struct rb_node *parent, *node; |
2302 | struct cfq_queue *__cfqq; | |
2303 | sector_t sector = cfqd->last_position; | |
2304 | ||
2305 | if (RB_EMPTY_ROOT(root)) | |
2306 | return NULL; | |
2307 | ||
2308 | /* | |
2309 | * First, if we find a request starting at the end of the last | |
2310 | * request, choose it. | |
2311 | */ | |
f2d1f0ae | 2312 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
2313 | if (__cfqq) |
2314 | return __cfqq; | |
2315 | ||
2316 | /* | |
2317 | * If the exact sector wasn't found, the parent of the NULL leaf | |
2318 | * will contain the closest sector. | |
2319 | */ | |
2320 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 2321 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2322 | return __cfqq; |
2323 | ||
2e46e8b2 | 2324 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
2325 | node = rb_next(&__cfqq->p_node); |
2326 | else | |
2327 | node = rb_prev(&__cfqq->p_node); | |
2328 | if (!node) | |
2329 | return NULL; | |
2330 | ||
2331 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 2332 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2333 | return __cfqq; |
2334 | ||
2335 | return NULL; | |
2336 | } | |
2337 | ||
2338 | /* | |
2339 | * cfqd - obvious | |
2340 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
2341 | * closely cooperating with itself. | |
2342 | * | |
2343 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
2344 | * one request, and that cfqd->last_position reflects a position on the disk | |
2345 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
2346 | * assumption. | |
2347 | */ | |
2348 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 2349 | struct cfq_queue *cur_cfqq) |
6d048f53 | 2350 | { |
a36e71f9 JA |
2351 | struct cfq_queue *cfqq; |
2352 | ||
39c01b21 DS |
2353 | if (cfq_class_idle(cur_cfqq)) |
2354 | return NULL; | |
e6c5bc73 JM |
2355 | if (!cfq_cfqq_sync(cur_cfqq)) |
2356 | return NULL; | |
2357 | if (CFQQ_SEEKY(cur_cfqq)) | |
2358 | return NULL; | |
2359 | ||
b9d8f4c7 GJ |
2360 | /* |
2361 | * Don't search priority tree if it's the only queue in the group. | |
2362 | */ | |
2363 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
2364 | return NULL; | |
2365 | ||
6d048f53 | 2366 | /* |
d9e7620e JA |
2367 | * We should notice if some of the queues are cooperating, eg |
2368 | * working closely on the same area of the disk. In that case, | |
2369 | * we can group them together and don't waste time idling. | |
6d048f53 | 2370 | */ |
a36e71f9 JA |
2371 | cfqq = cfqq_close(cfqd, cur_cfqq); |
2372 | if (!cfqq) | |
2373 | return NULL; | |
2374 | ||
8682e1f1 VG |
2375 | /* If new queue belongs to different cfq_group, don't choose it */ |
2376 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
2377 | return NULL; | |
2378 | ||
df5fe3e8 JM |
2379 | /* |
2380 | * It only makes sense to merge sync queues. | |
2381 | */ | |
2382 | if (!cfq_cfqq_sync(cfqq)) | |
2383 | return NULL; | |
e6c5bc73 JM |
2384 | if (CFQQ_SEEKY(cfqq)) |
2385 | return NULL; | |
df5fe3e8 | 2386 | |
c0324a02 CZ |
2387 | /* |
2388 | * Do not merge queues of different priority classes | |
2389 | */ | |
2390 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
2391 | return NULL; | |
2392 | ||
a36e71f9 | 2393 | return cfqq; |
6d048f53 JA |
2394 | } |
2395 | ||
a6d44e98 CZ |
2396 | /* |
2397 | * Determine whether we should enforce idle window for this queue. | |
2398 | */ | |
2399 | ||
2400 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2401 | { | |
3bf10fea | 2402 | enum wl_class_t wl_class = cfqq_class(cfqq); |
34b98d03 | 2403 | struct cfq_rb_root *st = cfqq->service_tree; |
a6d44e98 | 2404 | |
34b98d03 VG |
2405 | BUG_ON(!st); |
2406 | BUG_ON(!st->count); | |
f04a6424 | 2407 | |
b6508c16 VG |
2408 | if (!cfqd->cfq_slice_idle) |
2409 | return false; | |
2410 | ||
a6d44e98 | 2411 | /* We never do for idle class queues. */ |
3bf10fea | 2412 | if (wl_class == IDLE_WORKLOAD) |
a6d44e98 CZ |
2413 | return false; |
2414 | ||
2415 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
2416 | if (cfq_cfqq_idle_window(cfqq) && |
2417 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
2418 | return true; |
2419 | ||
2420 | /* | |
2421 | * Otherwise, we do only if they are the last ones | |
2422 | * in their service tree. | |
2423 | */ | |
34b98d03 VG |
2424 | if (st->count == 1 && cfq_cfqq_sync(cfqq) && |
2425 | !cfq_io_thinktime_big(cfqd, &st->ttime, false)) | |
c1e44756 | 2426 | return true; |
34b98d03 | 2427 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count); |
c1e44756 | 2428 | return false; |
a6d44e98 CZ |
2429 | } |
2430 | ||
6d048f53 | 2431 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2432 | { |
1792669c | 2433 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 2434 | struct cfq_io_cq *cic; |
80bdf0c7 | 2435 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 2436 | |
a68bbddb | 2437 | /* |
f7d7b7a7 JA |
2438 | * SSD device without seek penalty, disable idling. But only do so |
2439 | * for devices that support queuing, otherwise we still have a problem | |
2440 | * with sync vs async workloads. | |
a68bbddb | 2441 | */ |
f7d7b7a7 | 2442 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
2443 | return; |
2444 | ||
dd67d051 | 2445 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2446 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2447 | |
2448 | /* | |
2449 | * idle is disabled, either manually or by past process history | |
2450 | */ | |
80bdf0c7 VG |
2451 | if (!cfq_should_idle(cfqd, cfqq)) { |
2452 | /* no queue idling. Check for group idling */ | |
2453 | if (cfqd->cfq_group_idle) | |
2454 | group_idle = cfqd->cfq_group_idle; | |
2455 | else | |
2456 | return; | |
2457 | } | |
6d048f53 | 2458 | |
7b679138 | 2459 | /* |
8e550632 | 2460 | * still active requests from this queue, don't idle |
7b679138 | 2461 | */ |
8e550632 | 2462 | if (cfqq->dispatched) |
7b679138 JA |
2463 | return; |
2464 | ||
22e2c507 JA |
2465 | /* |
2466 | * task has exited, don't wait | |
2467 | */ | |
206dc69b | 2468 | cic = cfqd->active_cic; |
f6e8d01b | 2469 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
2470 | return; |
2471 | ||
355b659c CZ |
2472 | /* |
2473 | * If our average think time is larger than the remaining time | |
2474 | * slice, then don't idle. This avoids overrunning the allotted | |
2475 | * time slice. | |
2476 | */ | |
383cd721 SL |
2477 | if (sample_valid(cic->ttime.ttime_samples) && |
2478 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 2479 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 2480 | cic->ttime.ttime_mean); |
355b659c | 2481 | return; |
b1ffe737 | 2482 | } |
355b659c | 2483 | |
80bdf0c7 VG |
2484 | /* There are other queues in the group, don't do group idle */ |
2485 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
2486 | return; | |
2487 | ||
3b18152c | 2488 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2489 | |
80bdf0c7 VG |
2490 | if (group_idle) |
2491 | sl = cfqd->cfq_group_idle; | |
2492 | else | |
2493 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2494 | |
7b14e3b5 | 2495 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
155fead9 | 2496 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
80bdf0c7 VG |
2497 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
2498 | group_idle ? 1 : 0); | |
1da177e4 LT |
2499 | } |
2500 | ||
498d3aa2 JA |
2501 | /* |
2502 | * Move request from internal lists to the request queue dispatch list. | |
2503 | */ | |
165125e1 | 2504 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2505 | { |
3ed9a296 | 2506 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2507 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2508 | |
7b679138 JA |
2509 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2510 | ||
06d21886 | 2511 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2512 | cfq_remove_request(rq); |
6d048f53 | 2513 | cfqq->dispatched++; |
80bdf0c7 | 2514 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2515 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2516 | |
53c583d2 | 2517 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2518 | cfqq->nr_sectors += blk_rq_sectors(rq); |
155fead9 | 2519 | cfqg_stats_update_dispatch(cfqq->cfqg, blk_rq_bytes(rq), rq->cmd_flags); |
1da177e4 LT |
2520 | } |
2521 | ||
2522 | /* | |
2523 | * return expired entry, or NULL to just start from scratch in rbtree | |
2524 | */ | |
febffd61 | 2525 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 2526 | { |
30996f40 | 2527 | struct request *rq = NULL; |
1da177e4 | 2528 | |
3b18152c | 2529 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 2530 | return NULL; |
cb887411 JA |
2531 | |
2532 | cfq_mark_cfqq_fifo_expire(cfqq); | |
2533 | ||
89850f7e JA |
2534 | if (list_empty(&cfqq->fifo)) |
2535 | return NULL; | |
1da177e4 | 2536 | |
89850f7e | 2537 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 2538 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 2539 | rq = NULL; |
1da177e4 | 2540 | |
30996f40 | 2541 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 2542 | return rq; |
1da177e4 LT |
2543 | } |
2544 | ||
22e2c507 JA |
2545 | static inline int |
2546 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2547 | { | |
2548 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2549 | |
22e2c507 | 2550 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2551 | |
b9f8ce05 | 2552 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
2553 | } |
2554 | ||
df5fe3e8 JM |
2555 | /* |
2556 | * Must be called with the queue_lock held. | |
2557 | */ | |
2558 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2559 | { | |
2560 | int process_refs, io_refs; | |
2561 | ||
2562 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 2563 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
2564 | BUG_ON(process_refs < 0); |
2565 | return process_refs; | |
2566 | } | |
2567 | ||
2568 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2569 | { | |
e6c5bc73 | 2570 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2571 | struct cfq_queue *__cfqq; |
2572 | ||
c10b61f0 JM |
2573 | /* |
2574 | * If there are no process references on the new_cfqq, then it is | |
2575 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2576 | * chain may have dropped their last reference (not just their | |
2577 | * last process reference). | |
2578 | */ | |
2579 | if (!cfqq_process_refs(new_cfqq)) | |
2580 | return; | |
2581 | ||
df5fe3e8 JM |
2582 | /* Avoid a circular list and skip interim queue merges */ |
2583 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2584 | if (__cfqq == cfqq) | |
2585 | return; | |
2586 | new_cfqq = __cfqq; | |
2587 | } | |
2588 | ||
2589 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2590 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2591 | /* |
2592 | * If the process for the cfqq has gone away, there is no | |
2593 | * sense in merging the queues. | |
2594 | */ | |
c10b61f0 | 2595 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2596 | return; |
2597 | ||
e6c5bc73 JM |
2598 | /* |
2599 | * Merge in the direction of the lesser amount of work. | |
2600 | */ | |
e6c5bc73 JM |
2601 | if (new_process_refs >= process_refs) { |
2602 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2603 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2604 | } else { |
2605 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2606 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2607 | } |
df5fe3e8 JM |
2608 | } |
2609 | ||
6d816ec7 | 2610 | static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd, |
3bf10fea | 2611 | struct cfq_group *cfqg, enum wl_class_t wl_class) |
718eee05 CZ |
2612 | { |
2613 | struct cfq_queue *queue; | |
2614 | int i; | |
2615 | bool key_valid = false; | |
2616 | unsigned long lowest_key = 0; | |
2617 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2618 | ||
65b32a57 VG |
2619 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2620 | /* select the one with lowest rb_key */ | |
34b98d03 | 2621 | queue = cfq_rb_first(st_for(cfqg, wl_class, i)); |
718eee05 CZ |
2622 | if (queue && |
2623 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2624 | lowest_key = queue->rb_key; | |
2625 | cur_best = i; | |
2626 | key_valid = true; | |
2627 | } | |
2628 | } | |
2629 | ||
2630 | return cur_best; | |
2631 | } | |
2632 | ||
6d816ec7 VG |
2633 | static void |
2634 | choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
718eee05 | 2635 | { |
718eee05 CZ |
2636 | unsigned slice; |
2637 | unsigned count; | |
cdb16e8f | 2638 | struct cfq_rb_root *st; |
58ff82f3 | 2639 | unsigned group_slice; |
4d2ceea4 | 2640 | enum wl_class_t original_class = cfqd->serving_wl_class; |
1fa8f6d6 | 2641 | |
718eee05 | 2642 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2643 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 2644 | cfqd->serving_wl_class = RT_WORKLOAD; |
58ff82f3 | 2645 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 2646 | cfqd->serving_wl_class = BE_WORKLOAD; |
718eee05 | 2647 | else { |
4d2ceea4 | 2648 | cfqd->serving_wl_class = IDLE_WORKLOAD; |
718eee05 CZ |
2649 | cfqd->workload_expires = jiffies + 1; |
2650 | return; | |
2651 | } | |
2652 | ||
4d2ceea4 | 2653 | if (original_class != cfqd->serving_wl_class) |
e4ea0c16 SL |
2654 | goto new_workload; |
2655 | ||
718eee05 CZ |
2656 | /* |
2657 | * For RT and BE, we have to choose also the type | |
2658 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2659 | * expiration time | |
2660 | */ | |
34b98d03 | 2661 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 2662 | count = st->count; |
718eee05 CZ |
2663 | |
2664 | /* | |
65b32a57 | 2665 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2666 | */ |
65b32a57 | 2667 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2668 | return; |
2669 | ||
e4ea0c16 | 2670 | new_workload: |
718eee05 | 2671 | /* otherwise select new workload type */ |
6d816ec7 | 2672 | cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg, |
4d2ceea4 | 2673 | cfqd->serving_wl_class); |
34b98d03 | 2674 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 2675 | count = st->count; |
718eee05 CZ |
2676 | |
2677 | /* | |
2678 | * the workload slice is computed as a fraction of target latency | |
2679 | * proportional to the number of queues in that workload, over | |
2680 | * all the queues in the same priority class | |
2681 | */ | |
58ff82f3 VG |
2682 | group_slice = cfq_group_slice(cfqd, cfqg); |
2683 | ||
2684 | slice = group_slice * count / | |
4d2ceea4 VG |
2685 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class], |
2686 | cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd, | |
3bf10fea | 2687 | cfqg)); |
718eee05 | 2688 | |
4d2ceea4 | 2689 | if (cfqd->serving_wl_type == ASYNC_WORKLOAD) { |
f26bd1f0 VG |
2690 | unsigned int tmp; |
2691 | ||
2692 | /* | |
2693 | * Async queues are currently system wide. Just taking | |
2694 | * proportion of queues with-in same group will lead to higher | |
2695 | * async ratio system wide as generally root group is going | |
2696 | * to have higher weight. A more accurate thing would be to | |
2697 | * calculate system wide asnc/sync ratio. | |
2698 | */ | |
5bf14c07 TM |
2699 | tmp = cfqd->cfq_target_latency * |
2700 | cfqg_busy_async_queues(cfqd, cfqg); | |
f26bd1f0 VG |
2701 | tmp = tmp/cfqd->busy_queues; |
2702 | slice = min_t(unsigned, slice, tmp); | |
2703 | ||
718eee05 CZ |
2704 | /* async workload slice is scaled down according to |
2705 | * the sync/async slice ratio. */ | |
2706 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2707 | } else |
718eee05 CZ |
2708 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2709 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2710 | ||
2711 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2712 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
2713 | cfqd->workload_expires = jiffies + slice; |
2714 | } | |
2715 | ||
1fa8f6d6 VG |
2716 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2717 | { | |
2718 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2719 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2720 | |
2721 | if (RB_EMPTY_ROOT(&st->rb)) | |
2722 | return NULL; | |
25bc6b07 | 2723 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
2724 | update_min_vdisktime(st); |
2725 | return cfqg; | |
1fa8f6d6 VG |
2726 | } |
2727 | ||
cdb16e8f VG |
2728 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2729 | { | |
1fa8f6d6 VG |
2730 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2731 | ||
2732 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2733 | |
2734 | /* Restore the workload type data */ | |
4d2ceea4 VG |
2735 | if (cfqg->saved_wl_slice) { |
2736 | cfqd->workload_expires = jiffies + cfqg->saved_wl_slice; | |
2737 | cfqd->serving_wl_type = cfqg->saved_wl_type; | |
2738 | cfqd->serving_wl_class = cfqg->saved_wl_class; | |
66ae2919 GJ |
2739 | } else |
2740 | cfqd->workload_expires = jiffies - 1; | |
2741 | ||
6d816ec7 | 2742 | choose_wl_class_and_type(cfqd, cfqg); |
cdb16e8f VG |
2743 | } |
2744 | ||
22e2c507 | 2745 | /* |
498d3aa2 JA |
2746 | * Select a queue for service. If we have a current active queue, |
2747 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2748 | */ |
1b5ed5e1 | 2749 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2750 | { |
a36e71f9 | 2751 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2752 | |
22e2c507 JA |
2753 | cfqq = cfqd->active_queue; |
2754 | if (!cfqq) | |
2755 | goto new_queue; | |
1da177e4 | 2756 | |
f04a6424 VG |
2757 | if (!cfqd->rq_queued) |
2758 | return NULL; | |
c244bb50 VG |
2759 | |
2760 | /* | |
2761 | * We were waiting for group to get backlogged. Expire the queue | |
2762 | */ | |
2763 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2764 | goto expire; | |
2765 | ||
22e2c507 | 2766 | /* |
6d048f53 | 2767 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2768 | */ |
7667aa06 VG |
2769 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2770 | /* | |
2771 | * If slice had not expired at the completion of last request | |
2772 | * we might not have turned on wait_busy flag. Don't expire | |
2773 | * the queue yet. Allow the group to get backlogged. | |
2774 | * | |
2775 | * The very fact that we have used the slice, that means we | |
2776 | * have been idling all along on this queue and it should be | |
2777 | * ok to wait for this request to complete. | |
2778 | */ | |
82bbbf28 VG |
2779 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2780 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2781 | cfqq = NULL; | |
7667aa06 | 2782 | goto keep_queue; |
82bbbf28 | 2783 | } else |
80bdf0c7 | 2784 | goto check_group_idle; |
7667aa06 | 2785 | } |
1da177e4 | 2786 | |
22e2c507 | 2787 | /* |
6d048f53 JA |
2788 | * The active queue has requests and isn't expired, allow it to |
2789 | * dispatch. | |
22e2c507 | 2790 | */ |
dd67d051 | 2791 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2792 | goto keep_queue; |
6d048f53 | 2793 | |
a36e71f9 JA |
2794 | /* |
2795 | * If another queue has a request waiting within our mean seek | |
2796 | * distance, let it run. The expire code will check for close | |
2797 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2798 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2799 | */ |
b3b6d040 | 2800 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2801 | if (new_cfqq) { |
2802 | if (!cfqq->new_cfqq) | |
2803 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2804 | goto expire; |
df5fe3e8 | 2805 | } |
a36e71f9 | 2806 | |
6d048f53 JA |
2807 | /* |
2808 | * No requests pending. If the active queue still has requests in | |
2809 | * flight or is idling for a new request, allow either of these | |
2810 | * conditions to happen (or time out) before selecting a new queue. | |
2811 | */ | |
80bdf0c7 VG |
2812 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2813 | cfqq = NULL; | |
2814 | goto keep_queue; | |
2815 | } | |
2816 | ||
8e1ac665 SL |
2817 | /* |
2818 | * This is a deep seek queue, but the device is much faster than | |
2819 | * the queue can deliver, don't idle | |
2820 | **/ | |
2821 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
2822 | (cfq_cfqq_slice_new(cfqq) || | |
2823 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
2824 | cfq_clear_cfqq_deep(cfqq); | |
2825 | cfq_clear_cfqq_idle_window(cfqq); | |
2826 | } | |
2827 | ||
80bdf0c7 VG |
2828 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2829 | cfqq = NULL; | |
2830 | goto keep_queue; | |
2831 | } | |
2832 | ||
2833 | /* | |
2834 | * If group idle is enabled and there are requests dispatched from | |
2835 | * this group, wait for requests to complete. | |
2836 | */ | |
2837 | check_group_idle: | |
7700fc4f SL |
2838 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
2839 | cfqq->cfqg->dispatched && | |
2840 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
2841 | cfqq = NULL; |
2842 | goto keep_queue; | |
22e2c507 JA |
2843 | } |
2844 | ||
3b18152c | 2845 | expire: |
e5ff082e | 2846 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2847 | new_queue: |
718eee05 CZ |
2848 | /* |
2849 | * Current queue expired. Check if we have to switch to a new | |
2850 | * service tree | |
2851 | */ | |
2852 | if (!new_cfqq) | |
cdb16e8f | 2853 | cfq_choose_cfqg(cfqd); |
718eee05 | 2854 | |
a36e71f9 | 2855 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2856 | keep_queue: |
3b18152c | 2857 | return cfqq; |
22e2c507 JA |
2858 | } |
2859 | ||
febffd61 | 2860 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2861 | { |
2862 | int dispatched = 0; | |
2863 | ||
2864 | while (cfqq->next_rq) { | |
2865 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2866 | dispatched++; | |
2867 | } | |
2868 | ||
2869 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2870 | |
2871 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 2872 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
2873 | return dispatched; |
2874 | } | |
2875 | ||
498d3aa2 JA |
2876 | /* |
2877 | * Drain our current requests. Used for barriers and when switching | |
2878 | * io schedulers on-the-fly. | |
2879 | */ | |
d9e7620e | 2880 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2881 | { |
0871714e | 2882 | struct cfq_queue *cfqq; |
d9e7620e | 2883 | int dispatched = 0; |
cdb16e8f | 2884 | |
3440c49f | 2885 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 2886 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
2887 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2888 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 2889 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 2890 | } |
1b5ed5e1 | 2891 | |
1b5ed5e1 TH |
2892 | BUG_ON(cfqd->busy_queues); |
2893 | ||
6923715a | 2894 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2895 | return dispatched; |
2896 | } | |
2897 | ||
abc3c744 SL |
2898 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2899 | struct cfq_queue *cfqq) | |
2900 | { | |
2901 | /* the queue hasn't finished any request, can't estimate */ | |
2902 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 2903 | return true; |
abc3c744 SL |
2904 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
2905 | cfqq->slice_end)) | |
c1e44756 | 2906 | return true; |
abc3c744 | 2907 | |
c1e44756 | 2908 | return false; |
abc3c744 SL |
2909 | } |
2910 | ||
0b182d61 | 2911 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2912 | { |
2f5cb738 | 2913 | unsigned int max_dispatch; |
22e2c507 | 2914 | |
5ad531db JA |
2915 | /* |
2916 | * Drain async requests before we start sync IO | |
2917 | */ | |
53c583d2 | 2918 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 2919 | return false; |
5ad531db | 2920 | |
2f5cb738 JA |
2921 | /* |
2922 | * If this is an async queue and we have sync IO in flight, let it wait | |
2923 | */ | |
53c583d2 | 2924 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 2925 | return false; |
2f5cb738 | 2926 | |
abc3c744 | 2927 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
2928 | if (cfq_class_idle(cfqq)) |
2929 | max_dispatch = 1; | |
b4878f24 | 2930 | |
2f5cb738 JA |
2931 | /* |
2932 | * Does this cfqq already have too much IO in flight? | |
2933 | */ | |
2934 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 2935 | bool promote_sync = false; |
2f5cb738 JA |
2936 | /* |
2937 | * idle queue must always only have a single IO in flight | |
2938 | */ | |
3ed9a296 | 2939 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2940 | return false; |
3ed9a296 | 2941 | |
ef8a41df | 2942 | /* |
c4ade94f LS |
2943 | * If there is only one sync queue |
2944 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
2945 | * queue no dispatch limit. The reason is a sync queue can |
2946 | * preempt async queue, limiting the sync queue doesn't make | |
2947 | * sense. This is useful for aiostress test. | |
2948 | */ | |
c4ade94f LS |
2949 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
2950 | promote_sync = true; | |
ef8a41df | 2951 | |
2f5cb738 JA |
2952 | /* |
2953 | * We have other queues, don't allow more IO from this one | |
2954 | */ | |
ef8a41df SL |
2955 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
2956 | !promote_sync) | |
0b182d61 | 2957 | return false; |
9ede209e | 2958 | |
365722bb | 2959 | /* |
474b18cc | 2960 | * Sole queue user, no limit |
365722bb | 2961 | */ |
ef8a41df | 2962 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
2963 | max_dispatch = -1; |
2964 | else | |
2965 | /* | |
2966 | * Normally we start throttling cfqq when cfq_quantum/2 | |
2967 | * requests have been dispatched. But we can drive | |
2968 | * deeper queue depths at the beginning of slice | |
2969 | * subjected to upper limit of cfq_quantum. | |
2970 | * */ | |
2971 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
2972 | } |
2973 | ||
2974 | /* | |
2975 | * Async queues must wait a bit before being allowed dispatch. | |
2976 | * We also ramp up the dispatch depth gradually for async IO, | |
2977 | * based on the last sync IO we serviced | |
2978 | */ | |
963b72fc | 2979 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 2980 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 2981 | unsigned int depth; |
365722bb | 2982 | |
61f0c1dc | 2983 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2984 | if (!depth && !cfqq->dispatched) |
2985 | depth = 1; | |
8e296755 JA |
2986 | if (depth < max_dispatch) |
2987 | max_dispatch = depth; | |
2f5cb738 | 2988 | } |
3ed9a296 | 2989 | |
0b182d61 JA |
2990 | /* |
2991 | * If we're below the current max, allow a dispatch | |
2992 | */ | |
2993 | return cfqq->dispatched < max_dispatch; | |
2994 | } | |
2995 | ||
2996 | /* | |
2997 | * Dispatch a request from cfqq, moving them to the request queue | |
2998 | * dispatch list. | |
2999 | */ | |
3000 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3001 | { | |
3002 | struct request *rq; | |
3003 | ||
3004 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
3005 | ||
3006 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
3007 | return false; | |
3008 | ||
3009 | /* | |
3010 | * follow expired path, else get first next available | |
3011 | */ | |
3012 | rq = cfq_check_fifo(cfqq); | |
3013 | if (!rq) | |
3014 | rq = cfqq->next_rq; | |
3015 | ||
3016 | /* | |
3017 | * insert request into driver dispatch list | |
3018 | */ | |
3019 | cfq_dispatch_insert(cfqd->queue, rq); | |
3020 | ||
3021 | if (!cfqd->active_cic) { | |
c5869807 | 3022 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 3023 | |
c5869807 | 3024 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
3025 | cfqd->active_cic = cic; |
3026 | } | |
3027 | ||
3028 | return true; | |
3029 | } | |
3030 | ||
3031 | /* | |
3032 | * Find the cfqq that we need to service and move a request from that to the | |
3033 | * dispatch list | |
3034 | */ | |
3035 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
3036 | { | |
3037 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3038 | struct cfq_queue *cfqq; | |
3039 | ||
3040 | if (!cfqd->busy_queues) | |
3041 | return 0; | |
3042 | ||
3043 | if (unlikely(force)) | |
3044 | return cfq_forced_dispatch(cfqd); | |
3045 | ||
3046 | cfqq = cfq_select_queue(cfqd); | |
3047 | if (!cfqq) | |
8e296755 JA |
3048 | return 0; |
3049 | ||
2f5cb738 | 3050 | /* |
0b182d61 | 3051 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 3052 | */ |
0b182d61 JA |
3053 | if (!cfq_dispatch_request(cfqd, cfqq)) |
3054 | return 0; | |
3055 | ||
2f5cb738 | 3056 | cfqq->slice_dispatch++; |
b029195d | 3057 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 3058 | |
2f5cb738 JA |
3059 | /* |
3060 | * expire an async queue immediately if it has used up its slice. idle | |
3061 | * queue always expire after 1 dispatch round. | |
3062 | */ | |
3063 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
3064 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
3065 | cfq_class_idle(cfqq))) { | |
3066 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 3067 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
3068 | } |
3069 | ||
b217a903 | 3070 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 3071 | return 1; |
1da177e4 LT |
3072 | } |
3073 | ||
1da177e4 | 3074 | /* |
5e705374 JA |
3075 | * task holds one reference to the queue, dropped when task exits. each rq |
3076 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 3077 | * |
b1c35769 | 3078 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
3079 | * queue lock must be held here. |
3080 | */ | |
3081 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
3082 | { | |
22e2c507 | 3083 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 3084 | struct cfq_group *cfqg; |
22e2c507 | 3085 | |
30d7b944 | 3086 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 3087 | |
30d7b944 SL |
3088 | cfqq->ref--; |
3089 | if (cfqq->ref) | |
1da177e4 LT |
3090 | return; |
3091 | ||
7b679138 | 3092 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 3093 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 3094 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 3095 | cfqg = cfqq->cfqg; |
1da177e4 | 3096 | |
28f95cbc | 3097 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 3098 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 3099 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 3100 | } |
22e2c507 | 3101 | |
f04a6424 | 3102 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 3103 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 3104 | cfqg_put(cfqg); |
1da177e4 LT |
3105 | } |
3106 | ||
d02a2c07 | 3107 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 3108 | { |
df5fe3e8 JM |
3109 | struct cfq_queue *__cfqq, *next; |
3110 | ||
df5fe3e8 JM |
3111 | /* |
3112 | * If this queue was scheduled to merge with another queue, be | |
3113 | * sure to drop the reference taken on that queue (and others in | |
3114 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
3115 | */ | |
3116 | __cfqq = cfqq->new_cfqq; | |
3117 | while (__cfqq) { | |
3118 | if (__cfqq == cfqq) { | |
3119 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
3120 | break; | |
3121 | } | |
3122 | next = __cfqq->new_cfqq; | |
3123 | cfq_put_queue(__cfqq); | |
3124 | __cfqq = next; | |
3125 | } | |
d02a2c07 SL |
3126 | } |
3127 | ||
3128 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3129 | { | |
3130 | if (unlikely(cfqq == cfqd->active_queue)) { | |
3131 | __cfq_slice_expired(cfqd, cfqq, 0); | |
3132 | cfq_schedule_dispatch(cfqd); | |
3133 | } | |
3134 | ||
3135 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 3136 | |
89850f7e JA |
3137 | cfq_put_queue(cfqq); |
3138 | } | |
22e2c507 | 3139 | |
9b84cacd TH |
3140 | static void cfq_init_icq(struct io_cq *icq) |
3141 | { | |
3142 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
3143 | ||
3144 | cic->ttime.last_end_request = jiffies; | |
3145 | } | |
3146 | ||
c5869807 | 3147 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 3148 | { |
c5869807 | 3149 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 3150 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 3151 | |
ff6657c6 JA |
3152 | if (cic->cfqq[BLK_RW_ASYNC]) { |
3153 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
3154 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
3155 | } |
3156 | ||
ff6657c6 JA |
3157 | if (cic->cfqq[BLK_RW_SYNC]) { |
3158 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
3159 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 3160 | } |
89850f7e JA |
3161 | } |
3162 | ||
abede6da | 3163 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
3164 | { |
3165 | struct task_struct *tsk = current; | |
3166 | int ioprio_class; | |
3167 | ||
3b18152c | 3168 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
3169 | return; |
3170 | ||
598971bf | 3171 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 3172 | switch (ioprio_class) { |
fe094d98 JA |
3173 | default: |
3174 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
3175 | case IOPRIO_CLASS_NONE: | |
3176 | /* | |
6d63c275 | 3177 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
3178 | */ |
3179 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 3180 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
3181 | break; |
3182 | case IOPRIO_CLASS_RT: | |
598971bf | 3183 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3184 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
3185 | break; | |
3186 | case IOPRIO_CLASS_BE: | |
598971bf | 3187 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3188 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3189 | break; | |
3190 | case IOPRIO_CLASS_IDLE: | |
3191 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
3192 | cfqq->ioprio = 7; | |
3193 | cfq_clear_cfqq_idle_window(cfqq); | |
3194 | break; | |
22e2c507 JA |
3195 | } |
3196 | ||
3197 | /* | |
3198 | * keep track of original prio settings in case we have to temporarily | |
3199 | * elevate the priority of this queue | |
3200 | */ | |
3201 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 3202 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
3203 | } |
3204 | ||
598971bf | 3205 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 3206 | { |
598971bf | 3207 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 3208 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 3209 | struct cfq_queue *cfqq; |
35e6077c | 3210 | |
598971bf TH |
3211 | /* |
3212 | * Check whether ioprio has changed. The condition may trigger | |
3213 | * spuriously on a newly created cic but there's no harm. | |
3214 | */ | |
3215 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
3216 | return; |
3217 | ||
ff6657c6 | 3218 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
3219 | if (cfqq) { |
3220 | struct cfq_queue *new_cfqq; | |
abede6da TH |
3221 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio, |
3222 | GFP_ATOMIC); | |
caaa5f9f | 3223 | if (new_cfqq) { |
ff6657c6 | 3224 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
3225 | cfq_put_queue(cfqq); |
3226 | } | |
22e2c507 | 3227 | } |
caaa5f9f | 3228 | |
ff6657c6 | 3229 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
3230 | if (cfqq) |
3231 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
3232 | |
3233 | cic->ioprio = ioprio; | |
22e2c507 JA |
3234 | } |
3235 | ||
d5036d77 | 3236 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 3237 | pid_t pid, bool is_sync) |
d5036d77 JA |
3238 | { |
3239 | RB_CLEAR_NODE(&cfqq->rb_node); | |
3240 | RB_CLEAR_NODE(&cfqq->p_node); | |
3241 | INIT_LIST_HEAD(&cfqq->fifo); | |
3242 | ||
30d7b944 | 3243 | cfqq->ref = 0; |
d5036d77 JA |
3244 | cfqq->cfqd = cfqd; |
3245 | ||
3246 | cfq_mark_cfqq_prio_changed(cfqq); | |
3247 | ||
3248 | if (is_sync) { | |
3249 | if (!cfq_class_idle(cfqq)) | |
3250 | cfq_mark_cfqq_idle_window(cfqq); | |
3251 | cfq_mark_cfqq_sync(cfqq); | |
3252 | } | |
3253 | cfqq->pid = pid; | |
3254 | } | |
3255 | ||
24610333 | 3256 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
598971bf | 3257 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 3258 | { |
bca4b914 | 3259 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
598971bf TH |
3260 | struct cfq_queue *sync_cfqq; |
3261 | uint64_t id; | |
24610333 | 3262 | |
598971bf | 3263 | rcu_read_lock(); |
3c798398 | 3264 | id = bio_blkcg(bio)->id; |
598971bf | 3265 | rcu_read_unlock(); |
24610333 | 3266 | |
598971bf TH |
3267 | /* |
3268 | * Check whether blkcg has changed. The condition may trigger | |
3269 | * spuriously on a newly created cic but there's no harm. | |
3270 | */ | |
3271 | if (unlikely(!cfqd) || likely(cic->blkcg_id == id)) | |
3272 | return; | |
24610333 | 3273 | |
598971bf | 3274 | sync_cfqq = cic_to_cfqq(cic, 1); |
24610333 VG |
3275 | if (sync_cfqq) { |
3276 | /* | |
3277 | * Drop reference to sync queue. A new sync queue will be | |
3278 | * assigned in new group upon arrival of a fresh request. | |
3279 | */ | |
3280 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
3281 | cic_set_cfqq(cic, NULL, 1); | |
3282 | cfq_put_queue(sync_cfqq); | |
3283 | } | |
598971bf TH |
3284 | |
3285 | cic->blkcg_id = id; | |
24610333 | 3286 | } |
598971bf TH |
3287 | #else |
3288 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } | |
24610333 VG |
3289 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
3290 | ||
22e2c507 | 3291 | static struct cfq_queue * |
abede6da TH |
3292 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
3293 | struct bio *bio, gfp_t gfp_mask) | |
22e2c507 | 3294 | { |
3c798398 | 3295 | struct blkcg *blkcg; |
22e2c507 | 3296 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
cdb16e8f | 3297 | struct cfq_group *cfqg; |
22e2c507 JA |
3298 | |
3299 | retry: | |
2a7f1244 TH |
3300 | rcu_read_lock(); |
3301 | ||
3c798398 | 3302 | blkcg = bio_blkcg(bio); |
cd1604fa | 3303 | cfqg = cfq_lookup_create_cfqg(cfqd, blkcg); |
91fac317 | 3304 | cfqq = cic_to_cfqq(cic, is_sync); |
22e2c507 | 3305 | |
6118b70b JA |
3306 | /* |
3307 | * Always try a new alloc if we fell back to the OOM cfqq | |
3308 | * originally, since it should just be a temporary situation. | |
3309 | */ | |
3310 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
3311 | cfqq = NULL; | |
22e2c507 JA |
3312 | if (new_cfqq) { |
3313 | cfqq = new_cfqq; | |
3314 | new_cfqq = NULL; | |
3315 | } else if (gfp_mask & __GFP_WAIT) { | |
2a7f1244 | 3316 | rcu_read_unlock(); |
22e2c507 | 3317 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c | 3318 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 3319 | gfp_mask | __GFP_ZERO, |
94f6030c | 3320 | cfqd->queue->node); |
22e2c507 | 3321 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
3322 | if (new_cfqq) |
3323 | goto retry; | |
22e2c507 | 3324 | } else { |
94f6030c CL |
3325 | cfqq = kmem_cache_alloc_node(cfq_pool, |
3326 | gfp_mask | __GFP_ZERO, | |
3327 | cfqd->queue->node); | |
22e2c507 JA |
3328 | } |
3329 | ||
6118b70b JA |
3330 | if (cfqq) { |
3331 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
abede6da | 3332 | cfq_init_prio_data(cfqq, cic); |
cdb16e8f | 3333 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
3334 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
3335 | } else | |
3336 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
3337 | } |
3338 | ||
3339 | if (new_cfqq) | |
3340 | kmem_cache_free(cfq_pool, new_cfqq); | |
3341 | ||
2a7f1244 | 3342 | rcu_read_unlock(); |
22e2c507 JA |
3343 | return cfqq; |
3344 | } | |
3345 | ||
c2dea2d1 VT |
3346 | static struct cfq_queue ** |
3347 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
3348 | { | |
fe094d98 | 3349 | switch (ioprio_class) { |
c2dea2d1 VT |
3350 | case IOPRIO_CLASS_RT: |
3351 | return &cfqd->async_cfqq[0][ioprio]; | |
598971bf TH |
3352 | case IOPRIO_CLASS_NONE: |
3353 | ioprio = IOPRIO_NORM; | |
3354 | /* fall through */ | |
c2dea2d1 VT |
3355 | case IOPRIO_CLASS_BE: |
3356 | return &cfqd->async_cfqq[1][ioprio]; | |
3357 | case IOPRIO_CLASS_IDLE: | |
3358 | return &cfqd->async_idle_cfqq; | |
3359 | default: | |
3360 | BUG(); | |
3361 | } | |
3362 | } | |
3363 | ||
15c31be4 | 3364 | static struct cfq_queue * |
abede6da | 3365 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
4f85cb96 | 3366 | struct bio *bio, gfp_t gfp_mask) |
15c31be4 | 3367 | { |
598971bf TH |
3368 | const int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
3369 | const int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
c2dea2d1 | 3370 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
3371 | struct cfq_queue *cfqq = NULL; |
3372 | ||
c2dea2d1 VT |
3373 | if (!is_sync) { |
3374 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
3375 | cfqq = *async_cfqq; | |
3376 | } | |
3377 | ||
6118b70b | 3378 | if (!cfqq) |
abede6da | 3379 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, cic, bio, gfp_mask); |
15c31be4 JA |
3380 | |
3381 | /* | |
3382 | * pin the queue now that it's allocated, scheduler exit will prune it | |
3383 | */ | |
c2dea2d1 | 3384 | if (!is_sync && !(*async_cfqq)) { |
30d7b944 | 3385 | cfqq->ref++; |
c2dea2d1 | 3386 | *async_cfqq = cfqq; |
15c31be4 JA |
3387 | } |
3388 | ||
30d7b944 | 3389 | cfqq->ref++; |
15c31be4 JA |
3390 | return cfqq; |
3391 | } | |
3392 | ||
22e2c507 | 3393 | static void |
383cd721 | 3394 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 3395 | { |
383cd721 SL |
3396 | unsigned long elapsed = jiffies - ttime->last_end_request; |
3397 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 3398 | |
383cd721 SL |
3399 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
3400 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
3401 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
3402 | } | |
3403 | ||
3404 | static void | |
3405 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3406 | struct cfq_io_cq *cic) |
383cd721 | 3407 | { |
f5f2b6ce | 3408 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3409 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3410 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3411 | cfqd->cfq_slice_idle); | |
3412 | } | |
7700fc4f SL |
3413 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3414 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3415 | #endif | |
22e2c507 | 3416 | } |
1da177e4 | 3417 | |
206dc69b | 3418 | static void |
b2c18e1e | 3419 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3420 | struct request *rq) |
206dc69b | 3421 | { |
3dde36dd | 3422 | sector_t sdist = 0; |
41647e7a | 3423 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3424 | if (cfqq->last_request_pos) { |
3425 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3426 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3427 | else | |
3428 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3429 | } | |
206dc69b | 3430 | |
3dde36dd | 3431 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3432 | if (blk_queue_nonrot(cfqd->queue)) |
3433 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3434 | else | |
3435 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3436 | } |
1da177e4 | 3437 | |
22e2c507 JA |
3438 | /* |
3439 | * Disable idle window if the process thinks too long or seeks so much that | |
3440 | * it doesn't matter | |
3441 | */ | |
3442 | static void | |
3443 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3444 | struct cfq_io_cq *cic) |
22e2c507 | 3445 | { |
7b679138 | 3446 | int old_idle, enable_idle; |
1be92f2f | 3447 | |
0871714e JA |
3448 | /* |
3449 | * Don't idle for async or idle io prio class | |
3450 | */ | |
3451 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3452 | return; |
3453 | ||
c265a7f4 | 3454 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3455 | |
76280aff CZ |
3456 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3457 | cfq_mark_cfqq_deep(cfqq); | |
3458 | ||
749ef9f8 CZ |
3459 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
3460 | enable_idle = 0; | |
f6e8d01b | 3461 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
3462 | !cfqd->cfq_slice_idle || |
3463 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3464 | enable_idle = 0; |
383cd721 SL |
3465 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3466 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3467 | enable_idle = 0; |
3468 | else | |
3469 | enable_idle = 1; | |
1da177e4 LT |
3470 | } |
3471 | ||
7b679138 JA |
3472 | if (old_idle != enable_idle) { |
3473 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3474 | if (enable_idle) | |
3475 | cfq_mark_cfqq_idle_window(cfqq); | |
3476 | else | |
3477 | cfq_clear_cfqq_idle_window(cfqq); | |
3478 | } | |
22e2c507 | 3479 | } |
1da177e4 | 3480 | |
22e2c507 JA |
3481 | /* |
3482 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3483 | * no or if we aren't sure, a 1 will cause a preempt. | |
3484 | */ | |
a6151c3a | 3485 | static bool |
22e2c507 | 3486 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3487 | struct request *rq) |
22e2c507 | 3488 | { |
6d048f53 | 3489 | struct cfq_queue *cfqq; |
22e2c507 | 3490 | |
6d048f53 JA |
3491 | cfqq = cfqd->active_queue; |
3492 | if (!cfqq) | |
a6151c3a | 3493 | return false; |
22e2c507 | 3494 | |
6d048f53 | 3495 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3496 | return false; |
22e2c507 JA |
3497 | |
3498 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3499 | return true; |
1e3335de | 3500 | |
875feb63 DS |
3501 | /* |
3502 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3503 | */ | |
3504 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3505 | return false; | |
3506 | ||
374f84ac JA |
3507 | /* |
3508 | * if the new request is sync, but the currently running queue is | |
3509 | * not, let the sync request have priority. | |
3510 | */ | |
5e705374 | 3511 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3512 | return true; |
1e3335de | 3513 | |
8682e1f1 VG |
3514 | if (new_cfqq->cfqg != cfqq->cfqg) |
3515 | return false; | |
3516 | ||
3517 | if (cfq_slice_used(cfqq)) | |
3518 | return true; | |
3519 | ||
3520 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
4d2ceea4 | 3521 | if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 VG |
3522 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
3523 | new_cfqq->service_tree->count == 2 && | |
3524 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3525 | return true; | |
3526 | ||
b53d1ed7 JA |
3527 | /* |
3528 | * So both queues are sync. Let the new request get disk time if | |
3529 | * it's a metadata request and the current queue is doing regular IO. | |
3530 | */ | |
65299a3b | 3531 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
3532 | return true; |
3533 | ||
3a9a3f6c DS |
3534 | /* |
3535 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3536 | */ | |
3537 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3538 | return true; |
3a9a3f6c | 3539 | |
d2d59e18 SL |
3540 | /* An idle queue should not be idle now for some reason */ |
3541 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
3542 | return true; | |
3543 | ||
1e3335de | 3544 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3545 | return false; |
1e3335de JA |
3546 | |
3547 | /* | |
3548 | * if this request is as-good as one we would expect from the | |
3549 | * current cfqq, let it preempt | |
3550 | */ | |
e9ce335d | 3551 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3552 | return true; |
1e3335de | 3553 | |
a6151c3a | 3554 | return false; |
22e2c507 JA |
3555 | } |
3556 | ||
3557 | /* | |
3558 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3559 | * let it have half of its nominal slice. | |
3560 | */ | |
3561 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3562 | { | |
df0793ab SL |
3563 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
3564 | ||
7b679138 | 3565 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 3566 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3567 | |
f8ae6e3e SL |
3568 | /* |
3569 | * workload type is changed, don't save slice, otherwise preempt | |
3570 | * doesn't happen | |
3571 | */ | |
df0793ab | 3572 | if (old_type != cfqq_type(cfqq)) |
4d2ceea4 | 3573 | cfqq->cfqg->saved_wl_slice = 0; |
f8ae6e3e | 3574 | |
bf572256 JA |
3575 | /* |
3576 | * Put the new queue at the front of the of the current list, | |
3577 | * so we know that it will be selected next. | |
3578 | */ | |
3579 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3580 | |
3581 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3582 | |
62a37f6b JT |
3583 | cfqq->slice_end = 0; |
3584 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3585 | } |
3586 | ||
22e2c507 | 3587 | /* |
5e705374 | 3588 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3589 | * something we should do about it |
3590 | */ | |
3591 | static void | |
5e705374 JA |
3592 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3593 | struct request *rq) | |
22e2c507 | 3594 | { |
c5869807 | 3595 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3596 | |
45333d5a | 3597 | cfqd->rq_queued++; |
65299a3b CH |
3598 | if (rq->cmd_flags & REQ_PRIO) |
3599 | cfqq->prio_pending++; | |
374f84ac | 3600 | |
383cd721 | 3601 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3602 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3603 | cfq_update_idle_window(cfqd, cfqq, cic); |
3604 | ||
b2c18e1e | 3605 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3606 | |
3607 | if (cfqq == cfqd->active_queue) { | |
3608 | /* | |
b029195d JA |
3609 | * Remember that we saw a request from this process, but |
3610 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3611 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3612 | * and merging. If the request is already larger than a single |
3613 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3614 | * merging is already done. Ditto for a busy system that |
3615 | * has other work pending, don't risk delaying until the | |
3616 | * idle timer unplug to continue working. | |
22e2c507 | 3617 | */ |
d6ceb25e | 3618 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3619 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3620 | cfqd->busy_queues > 1) { | |
812df48d | 3621 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3622 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3623 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3624 | } else { |
155fead9 | 3625 | cfqg_stats_update_idle_time(cfqq->cfqg); |
bf791937 | 3626 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3627 | } |
d6ceb25e | 3628 | } |
5e705374 | 3629 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3630 | /* |
3631 | * not the active queue - expire current slice if it is | |
3632 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3633 | * has some old slice time left and is of higher priority or |
3634 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3635 | */ |
3636 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3637 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3638 | } |
1da177e4 LT |
3639 | } |
3640 | ||
165125e1 | 3641 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3642 | { |
b4878f24 | 3643 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3644 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3645 | |
7b679138 | 3646 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 3647 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 3648 | |
30996f40 | 3649 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3650 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3651 | cfq_add_rq_rb(rq); |
155fead9 TH |
3652 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
3653 | rq->cmd_flags); | |
5e705374 | 3654 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3655 | } |
3656 | ||
45333d5a AC |
3657 | /* |
3658 | * Update hw_tag based on peak queue depth over 50 samples under | |
3659 | * sufficient load. | |
3660 | */ | |
3661 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3662 | { | |
1a1238a7 SL |
3663 | struct cfq_queue *cfqq = cfqd->active_queue; |
3664 | ||
53c583d2 CZ |
3665 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3666 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3667 | |
3668 | if (cfqd->hw_tag == 1) | |
3669 | return; | |
45333d5a AC |
3670 | |
3671 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3672 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3673 | return; |
3674 | ||
1a1238a7 SL |
3675 | /* |
3676 | * If active queue hasn't enough requests and can idle, cfq might not | |
3677 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3678 | * case | |
3679 | */ | |
3680 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3681 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3682 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3683 | return; |
3684 | ||
45333d5a AC |
3685 | if (cfqd->hw_tag_samples++ < 50) |
3686 | return; | |
3687 | ||
e459dd08 | 3688 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3689 | cfqd->hw_tag = 1; |
3690 | else | |
3691 | cfqd->hw_tag = 0; | |
45333d5a AC |
3692 | } |
3693 | ||
7667aa06 VG |
3694 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3695 | { | |
c5869807 | 3696 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 3697 | |
02a8f01b JT |
3698 | /* If the queue already has requests, don't wait */ |
3699 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3700 | return false; | |
3701 | ||
7667aa06 VG |
3702 | /* If there are other queues in the group, don't wait */ |
3703 | if (cfqq->cfqg->nr_cfqq > 1) | |
3704 | return false; | |
3705 | ||
7700fc4f SL |
3706 | /* the only queue in the group, but think time is big */ |
3707 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
3708 | return false; | |
3709 | ||
7667aa06 VG |
3710 | if (cfq_slice_used(cfqq)) |
3711 | return true; | |
3712 | ||
3713 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
3714 | if (cic && sample_valid(cic->ttime.ttime_samples) |
3715 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
3716 | return true; |
3717 | ||
3718 | /* | |
3719 | * If think times is less than a jiffy than ttime_mean=0 and above | |
3720 | * will not be true. It might happen that slice has not expired yet | |
3721 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
3722 | * case where think time is less than a jiffy, mark the queue wait | |
3723 | * busy if only 1 jiffy is left in the slice. | |
3724 | */ | |
3725 | if (cfqq->slice_end - jiffies == 1) | |
3726 | return true; | |
3727 | ||
3728 | return false; | |
3729 | } | |
3730 | ||
165125e1 | 3731 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3732 | { |
5e705374 | 3733 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3734 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3735 | const int sync = rq_is_sync(rq); |
b4878f24 | 3736 | unsigned long now; |
1da177e4 | 3737 | |
b4878f24 | 3738 | now = jiffies; |
33659ebb CH |
3739 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3740 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 3741 | |
45333d5a AC |
3742 | cfq_update_hw_tag(cfqd); |
3743 | ||
53c583d2 | 3744 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 3745 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 3746 | cfqd->rq_in_driver--; |
6d048f53 | 3747 | cfqq->dispatched--; |
80bdf0c7 | 3748 | (RQ_CFQG(rq))->dispatched--; |
155fead9 TH |
3749 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
3750 | rq_io_start_time_ns(rq), rq->cmd_flags); | |
1da177e4 | 3751 | |
53c583d2 | 3752 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 3753 | |
365722bb | 3754 | if (sync) { |
34b98d03 | 3755 | struct cfq_rb_root *st; |
f5f2b6ce | 3756 | |
383cd721 | 3757 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
3758 | |
3759 | if (cfq_cfqq_on_rr(cfqq)) | |
34b98d03 | 3760 | st = cfqq->service_tree; |
f5f2b6ce | 3761 | else |
34b98d03 VG |
3762 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), |
3763 | cfqq_type(cfqq)); | |
3764 | ||
3765 | st->ttime.last_end_request = now; | |
573412b2 CZ |
3766 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3767 | cfqd->last_delayed_sync = now; | |
365722bb | 3768 | } |
caaa5f9f | 3769 | |
7700fc4f SL |
3770 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3771 | cfqq->cfqg->ttime.last_end_request = now; | |
3772 | #endif | |
3773 | ||
caaa5f9f JA |
3774 | /* |
3775 | * If this is the active queue, check if it needs to be expired, | |
3776 | * or if we want to idle in case it has no pending requests. | |
3777 | */ | |
3778 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3779 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3780 | ||
44f7c160 JA |
3781 | if (cfq_cfqq_slice_new(cfqq)) { |
3782 | cfq_set_prio_slice(cfqd, cfqq); | |
3783 | cfq_clear_cfqq_slice_new(cfqq); | |
3784 | } | |
f75edf2d VG |
3785 | |
3786 | /* | |
7667aa06 VG |
3787 | * Should we wait for next request to come in before we expire |
3788 | * the queue. | |
f75edf2d | 3789 | */ |
7667aa06 | 3790 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
3791 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3792 | if (!cfqd->cfq_slice_idle) | |
3793 | extend_sl = cfqd->cfq_group_idle; | |
3794 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 3795 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 3796 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
3797 | } |
3798 | ||
a36e71f9 | 3799 | /* |
8e550632 CZ |
3800 | * Idling is not enabled on: |
3801 | * - expired queues | |
3802 | * - idle-priority queues | |
3803 | * - async queues | |
3804 | * - queues with still some requests queued | |
3805 | * - when there is a close cooperator | |
a36e71f9 | 3806 | */ |
0871714e | 3807 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 3808 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3809 | else if (sync && cfqq_empty && |
3810 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 3811 | cfq_arm_slice_timer(cfqd); |
8e550632 | 3812 | } |
caaa5f9f | 3813 | } |
6d048f53 | 3814 | |
53c583d2 | 3815 | if (!cfqd->rq_in_driver) |
23e018a1 | 3816 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3817 | } |
3818 | ||
89850f7e | 3819 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3820 | { |
1b379d8d | 3821 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3822 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3823 | return ELV_MQUEUE_MUST; |
3b18152c | 3824 | } |
1da177e4 | 3825 | |
22e2c507 | 3826 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3827 | } |
3828 | ||
165125e1 | 3829 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3830 | { |
3831 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3832 | struct task_struct *tsk = current; | |
c5869807 | 3833 | struct cfq_io_cq *cic; |
22e2c507 JA |
3834 | struct cfq_queue *cfqq; |
3835 | ||
3836 | /* | |
3837 | * don't force setup of a queue from here, as a call to may_queue | |
3838 | * does not necessarily imply that a request actually will be queued. | |
3839 | * so just lookup a possibly existing queue, or return 'may queue' | |
3840 | * if that fails | |
3841 | */ | |
4ac845a2 | 3842 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3843 | if (!cic) |
3844 | return ELV_MQUEUE_MAY; | |
3845 | ||
b0b78f81 | 3846 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3847 | if (cfqq) { |
abede6da | 3848 | cfq_init_prio_data(cfqq, cic); |
22e2c507 | 3849 | |
89850f7e | 3850 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3851 | } |
3852 | ||
3853 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3854 | } |
3855 | ||
1da177e4 LT |
3856 | /* |
3857 | * queue lock held here | |
3858 | */ | |
bb37b94c | 3859 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3860 | { |
5e705374 | 3861 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3862 | |
5e705374 | 3863 | if (cfqq) { |
22e2c507 | 3864 | const int rw = rq_data_dir(rq); |
1da177e4 | 3865 | |
22e2c507 JA |
3866 | BUG_ON(!cfqq->allocated[rw]); |
3867 | cfqq->allocated[rw]--; | |
1da177e4 | 3868 | |
7f1dc8a2 | 3869 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 3870 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
3871 | rq->elv.priv[0] = NULL; |
3872 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 3873 | |
1da177e4 LT |
3874 | cfq_put_queue(cfqq); |
3875 | } | |
3876 | } | |
3877 | ||
df5fe3e8 | 3878 | static struct cfq_queue * |
c5869807 | 3879 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
3880 | struct cfq_queue *cfqq) |
3881 | { | |
3882 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3883 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3884 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3885 | cfq_put_queue(cfqq); |
3886 | return cic_to_cfqq(cic, 1); | |
3887 | } | |
3888 | ||
e6c5bc73 JM |
3889 | /* |
3890 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3891 | * was the last process referring to said cfqq. | |
3892 | */ | |
3893 | static struct cfq_queue * | |
c5869807 | 3894 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
3895 | { |
3896 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
3897 | cfqq->pid = current->pid; |
3898 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 3899 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
3900 | return cfqq; |
3901 | } | |
3902 | ||
3903 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
3904 | |
3905 | cfq_put_cooperator(cfqq); | |
3906 | ||
e6c5bc73 JM |
3907 | cfq_put_queue(cfqq); |
3908 | return NULL; | |
3909 | } | |
1da177e4 | 3910 | /* |
22e2c507 | 3911 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3912 | */ |
22e2c507 | 3913 | static int |
852c788f TH |
3914 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
3915 | gfp_t gfp_mask) | |
1da177e4 LT |
3916 | { |
3917 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 3918 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 3919 | const int rw = rq_data_dir(rq); |
a6151c3a | 3920 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3921 | struct cfq_queue *cfqq; |
1da177e4 LT |
3922 | |
3923 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3924 | ||
216284c3 | 3925 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 3926 | |
598971bf TH |
3927 | check_ioprio_changed(cic, bio); |
3928 | check_blkcg_changed(cic, bio); | |
e6c5bc73 | 3929 | new_queue: |
91fac317 | 3930 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3931 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
abede6da | 3932 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio, gfp_mask); |
91fac317 | 3933 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3934 | } else { |
e6c5bc73 JM |
3935 | /* |
3936 | * If the queue was seeky for too long, break it apart. | |
3937 | */ | |
ae54abed | 3938 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
3939 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
3940 | cfqq = split_cfqq(cic, cfqq); | |
3941 | if (!cfqq) | |
3942 | goto new_queue; | |
3943 | } | |
3944 | ||
df5fe3e8 JM |
3945 | /* |
3946 | * Check to see if this queue is scheduled to merge with | |
3947 | * another, closely cooperating queue. The merging of | |
3948 | * queues happens here as it must be done in process context. | |
3949 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3950 | */ | |
3951 | if (cfqq->new_cfqq) | |
3952 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3953 | } |
1da177e4 LT |
3954 | |
3955 | cfqq->allocated[rw]++; | |
1da177e4 | 3956 | |
6fae9c25 | 3957 | cfqq->ref++; |
eb7d8c07 | 3958 | cfqg_get(cfqq->cfqg); |
a612fddf | 3959 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 3960 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 3961 | spin_unlock_irq(q->queue_lock); |
5e705374 | 3962 | return 0; |
1da177e4 LT |
3963 | } |
3964 | ||
65f27f38 | 3965 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3966 | { |
65f27f38 | 3967 | struct cfq_data *cfqd = |
23e018a1 | 3968 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3969 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3970 | |
40bb54d1 | 3971 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 3972 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3973 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3974 | } |
3975 | ||
3976 | /* | |
3977 | * Timer running if the active_queue is currently idling inside its time slice | |
3978 | */ | |
3979 | static void cfq_idle_slice_timer(unsigned long data) | |
3980 | { | |
3981 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3982 | struct cfq_queue *cfqq; | |
3983 | unsigned long flags; | |
3c6bd2f8 | 3984 | int timed_out = 1; |
22e2c507 | 3985 | |
7b679138 JA |
3986 | cfq_log(cfqd, "idle timer fired"); |
3987 | ||
22e2c507 JA |
3988 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3989 | ||
fe094d98 JA |
3990 | cfqq = cfqd->active_queue; |
3991 | if (cfqq) { | |
3c6bd2f8 JA |
3992 | timed_out = 0; |
3993 | ||
b029195d JA |
3994 | /* |
3995 | * We saw a request before the queue expired, let it through | |
3996 | */ | |
3997 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3998 | goto out_kick; | |
3999 | ||
22e2c507 JA |
4000 | /* |
4001 | * expired | |
4002 | */ | |
44f7c160 | 4003 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
4004 | goto expire; |
4005 | ||
4006 | /* | |
4007 | * only expire and reinvoke request handler, if there are | |
4008 | * other queues with pending requests | |
4009 | */ | |
caaa5f9f | 4010 | if (!cfqd->busy_queues) |
22e2c507 | 4011 | goto out_cont; |
22e2c507 JA |
4012 | |
4013 | /* | |
4014 | * not expired and it has a request pending, let it dispatch | |
4015 | */ | |
75e50984 | 4016 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 4017 | goto out_kick; |
76280aff CZ |
4018 | |
4019 | /* | |
4020 | * Queue depth flag is reset only when the idle didn't succeed | |
4021 | */ | |
4022 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
4023 | } |
4024 | expire: | |
e5ff082e | 4025 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 4026 | out_kick: |
23e018a1 | 4027 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
4028 | out_cont: |
4029 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
4030 | } | |
4031 | ||
3b18152c JA |
4032 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
4033 | { | |
4034 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 4035 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 4036 | } |
22e2c507 | 4037 | |
c2dea2d1 VT |
4038 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
4039 | { | |
4040 | int i; | |
4041 | ||
4042 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
4043 | if (cfqd->async_cfqq[0][i]) | |
4044 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
4045 | if (cfqd->async_cfqq[1][i]) | |
4046 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 4047 | } |
2389d1ef ON |
4048 | |
4049 | if (cfqd->async_idle_cfqq) | |
4050 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
4051 | } |
4052 | ||
b374d18a | 4053 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 4054 | { |
22e2c507 | 4055 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 4056 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4057 | |
3b18152c | 4058 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 4059 | |
d9ff4187 | 4060 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 4061 | |
d9ff4187 | 4062 | if (cfqd->active_queue) |
e5ff082e | 4063 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 4064 | |
c2dea2d1 | 4065 | cfq_put_async_queues(cfqd); |
03aa264a TH |
4066 | |
4067 | spin_unlock_irq(q->queue_lock); | |
4068 | ||
a90d742e AV |
4069 | cfq_shutdown_timer_wq(cfqd); |
4070 | ||
ffea73fc TH |
4071 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4072 | blkcg_deactivate_policy(q, &blkcg_policy_cfq); | |
4073 | #else | |
f51b802c | 4074 | kfree(cfqd->root_group); |
2abae55f | 4075 | #endif |
56edf7d7 | 4076 | kfree(cfqd); |
1da177e4 LT |
4077 | } |
4078 | ||
b2fab5ac | 4079 | static int cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
4080 | { |
4081 | struct cfq_data *cfqd; | |
3c798398 | 4082 | struct blkcg_gq *blkg __maybe_unused; |
a2b1693b | 4083 | int i, ret; |
1da177e4 | 4084 | |
94f6030c | 4085 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
a73f730d | 4086 | if (!cfqd) |
b2fab5ac | 4087 | return -ENOMEM; |
80b15c73 | 4088 | |
f51b802c TH |
4089 | cfqd->queue = q; |
4090 | q->elevator->elevator_data = cfqd; | |
4091 | ||
1fa8f6d6 VG |
4092 | /* Init root service tree */ |
4093 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
4094 | ||
f51b802c | 4095 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 4096 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4097 | ret = blkcg_activate_policy(q, &blkcg_policy_cfq); |
a2b1693b TH |
4098 | if (ret) |
4099 | goto out_free; | |
f51b802c | 4100 | |
a2b1693b | 4101 | cfqd->root_group = blkg_to_cfqg(q->root_blkg); |
f51b802c | 4102 | #else |
a2b1693b | 4103 | ret = -ENOMEM; |
f51b802c TH |
4104 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
4105 | GFP_KERNEL, cfqd->queue->node); | |
a2b1693b TH |
4106 | if (!cfqd->root_group) |
4107 | goto out_free; | |
5624a4e4 | 4108 | |
a2b1693b TH |
4109 | cfq_init_cfqg_base(cfqd->root_group); |
4110 | #endif | |
3381cb8d | 4111 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_DEFAULT; |
e71357e1 | 4112 | cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_DEFAULT; |
5624a4e4 | 4113 | |
26a2ac00 JA |
4114 | /* |
4115 | * Not strictly needed (since RB_ROOT just clears the node and we | |
4116 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
4117 | * to add magic to the rb code | |
4118 | */ | |
4119 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
4120 | cfqd->prio_trees[i] = RB_ROOT; | |
4121 | ||
6118b70b JA |
4122 | /* |
4123 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
4124 | * Grab a permanent reference to it, so that the normal code flow | |
f51b802c TH |
4125 | * will not attempt to free it. oom_cfqq is linked to root_group |
4126 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
4127 | * the reference from linking right away. | |
6118b70b JA |
4128 | */ |
4129 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 4130 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
4131 | |
4132 | spin_lock_irq(q->queue_lock); | |
f51b802c | 4133 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 4134 | cfqg_put(cfqd->root_group); |
1adaf3dd | 4135 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 4136 | |
22e2c507 JA |
4137 | init_timer(&cfqd->idle_slice_timer); |
4138 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
4139 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
4140 | ||
23e018a1 | 4141 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 4142 | |
1da177e4 | 4143 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
4144 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
4145 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
4146 | cfqd->cfq_back_max = cfq_back_max; |
4147 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
4148 | cfqd->cfq_slice[0] = cfq_slice_async; |
4149 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
5bf14c07 | 4150 | cfqd->cfq_target_latency = cfq_target_latency; |
22e2c507 JA |
4151 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
4152 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
80bdf0c7 | 4153 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 4154 | cfqd->cfq_latency = 1; |
e459dd08 | 4155 | cfqd->hw_tag = -1; |
edc71131 CZ |
4156 | /* |
4157 | * we optimistically start assuming sync ops weren't delayed in last | |
4158 | * second, in order to have larger depth for async operations. | |
4159 | */ | |
573412b2 | 4160 | cfqd->last_delayed_sync = jiffies - HZ; |
b2fab5ac | 4161 | return 0; |
a2b1693b TH |
4162 | |
4163 | out_free: | |
4164 | kfree(cfqd); | |
4165 | return ret; | |
1da177e4 LT |
4166 | } |
4167 | ||
1da177e4 LT |
4168 | /* |
4169 | * sysfs parts below --> | |
4170 | */ | |
1da177e4 LT |
4171 | static ssize_t |
4172 | cfq_var_show(unsigned int var, char *page) | |
4173 | { | |
4174 | return sprintf(page, "%d\n", var); | |
4175 | } | |
4176 | ||
4177 | static ssize_t | |
4178 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
4179 | { | |
4180 | char *p = (char *) page; | |
4181 | ||
4182 | *var = simple_strtoul(p, &p, 10); | |
4183 | return count; | |
4184 | } | |
4185 | ||
1da177e4 | 4186 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 4187 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 4188 | { \ |
3d1ab40f | 4189 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4190 | unsigned int __data = __VAR; \ |
4191 | if (__CONV) \ | |
4192 | __data = jiffies_to_msecs(__data); \ | |
4193 | return cfq_var_show(__data, (page)); \ | |
4194 | } | |
4195 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
4196 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
4197 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
4198 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
4199 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 4200 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 4201 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
4202 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4203 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4204 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4205 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
5bf14c07 | 4206 | SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1); |
1da177e4 LT |
4207 | #undef SHOW_FUNCTION |
4208 | ||
4209 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 4210 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4211 | { \ |
3d1ab40f | 4212 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4213 | unsigned int __data; \ |
4214 | int ret = cfq_var_store(&__data, (page), count); \ | |
4215 | if (__data < (MIN)) \ | |
4216 | __data = (MIN); \ | |
4217 | else if (__data > (MAX)) \ | |
4218 | __data = (MAX); \ | |
4219 | if (__CONV) \ | |
4220 | *(__PTR) = msecs_to_jiffies(__data); \ | |
4221 | else \ | |
4222 | *(__PTR) = __data; \ | |
4223 | return ret; \ | |
4224 | } | |
4225 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4226 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4227 | UINT_MAX, 1); | |
4228 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4229 | UINT_MAX, 1); | |
e572ec7e | 4230 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4231 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4232 | UINT_MAX, 0); | |
22e2c507 | 4233 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4234 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4235 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4236 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4237 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4238 | UINT_MAX, 0); | |
963b72fc | 4239 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
5bf14c07 | 4240 | STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1); |
1da177e4 LT |
4241 | #undef STORE_FUNCTION |
4242 | ||
e572ec7e AV |
4243 | #define CFQ_ATTR(name) \ |
4244 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4245 | ||
4246 | static struct elv_fs_entry cfq_attrs[] = { | |
4247 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4248 | CFQ_ATTR(fifo_expire_sync), |
4249 | CFQ_ATTR(fifo_expire_async), | |
4250 | CFQ_ATTR(back_seek_max), | |
4251 | CFQ_ATTR(back_seek_penalty), | |
4252 | CFQ_ATTR(slice_sync), | |
4253 | CFQ_ATTR(slice_async), | |
4254 | CFQ_ATTR(slice_async_rq), | |
4255 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 4256 | CFQ_ATTR(group_idle), |
963b72fc | 4257 | CFQ_ATTR(low_latency), |
5bf14c07 | 4258 | CFQ_ATTR(target_latency), |
e572ec7e | 4259 | __ATTR_NULL |
1da177e4 LT |
4260 | }; |
4261 | ||
1da177e4 LT |
4262 | static struct elevator_type iosched_cfq = { |
4263 | .ops = { | |
4264 | .elevator_merge_fn = cfq_merge, | |
4265 | .elevator_merged_fn = cfq_merged_request, | |
4266 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 4267 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 4268 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4269 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4270 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4271 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4272 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4273 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4274 | .elevator_former_req_fn = elv_rb_former_request, |
4275 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 4276 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 4277 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
4278 | .elevator_set_req_fn = cfq_set_request, |
4279 | .elevator_put_req_fn = cfq_put_request, | |
4280 | .elevator_may_queue_fn = cfq_may_queue, | |
4281 | .elevator_init_fn = cfq_init_queue, | |
4282 | .elevator_exit_fn = cfq_exit_queue, | |
4283 | }, | |
3d3c2379 TH |
4284 | .icq_size = sizeof(struct cfq_io_cq), |
4285 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 4286 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 4287 | .elevator_name = "cfq", |
1da177e4 LT |
4288 | .elevator_owner = THIS_MODULE, |
4289 | }; | |
4290 | ||
3e252066 | 4291 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4292 | static struct blkcg_policy blkcg_policy_cfq = { |
f9fcc2d3 TH |
4293 | .pd_size = sizeof(struct cfq_group), |
4294 | .cftypes = cfq_blkcg_files, | |
4295 | ||
4296 | .pd_init_fn = cfq_pd_init, | |
4297 | .pd_reset_stats_fn = cfq_pd_reset_stats, | |
3e252066 | 4298 | }; |
3e252066 VG |
4299 | #endif |
4300 | ||
1da177e4 LT |
4301 | static int __init cfq_init(void) |
4302 | { | |
3d3c2379 TH |
4303 | int ret; |
4304 | ||
22e2c507 JA |
4305 | /* |
4306 | * could be 0 on HZ < 1000 setups | |
4307 | */ | |
4308 | if (!cfq_slice_async) | |
4309 | cfq_slice_async = 1; | |
4310 | if (!cfq_slice_idle) | |
4311 | cfq_slice_idle = 1; | |
4312 | ||
80bdf0c7 VG |
4313 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4314 | if (!cfq_group_idle) | |
4315 | cfq_group_idle = 1; | |
8bd435b3 | 4316 | |
3c798398 | 4317 | ret = blkcg_policy_register(&blkcg_policy_cfq); |
8bd435b3 TH |
4318 | if (ret) |
4319 | return ret; | |
ffea73fc TH |
4320 | #else |
4321 | cfq_group_idle = 0; | |
4322 | #endif | |
8bd435b3 | 4323 | |
fd794956 | 4324 | ret = -ENOMEM; |
3d3c2379 TH |
4325 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4326 | if (!cfq_pool) | |
8bd435b3 | 4327 | goto err_pol_unreg; |
1da177e4 | 4328 | |
3d3c2379 | 4329 | ret = elv_register(&iosched_cfq); |
8bd435b3 TH |
4330 | if (ret) |
4331 | goto err_free_pool; | |
3d3c2379 | 4332 | |
2fdd82bd | 4333 | return 0; |
8bd435b3 TH |
4334 | |
4335 | err_free_pool: | |
4336 | kmem_cache_destroy(cfq_pool); | |
4337 | err_pol_unreg: | |
ffea73fc | 4338 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4339 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4340 | #endif |
8bd435b3 | 4341 | return ret; |
1da177e4 LT |
4342 | } |
4343 | ||
4344 | static void __exit cfq_exit(void) | |
4345 | { | |
ffea73fc | 4346 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4347 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4348 | #endif |
1da177e4 | 4349 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4350 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4351 | } |
4352 | ||
4353 | module_init(cfq_init); | |
4354 | module_exit(cfq_exit); | |
4355 | ||
4356 | MODULE_AUTHOR("Jens Axboe"); | |
4357 | MODULE_LICENSE("GPL"); | |
4358 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |