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> |
eea8f41c | 17 | #include <linux/blk-cgroup.h> |
6e736be7 | 18 | #include "blk.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 | |
e48453c3 AA |
70 | /* blkio-related constants */ |
71 | #define CFQ_WEIGHT_MIN 10 | |
72 | #define CFQ_WEIGHT_MAX 1000 | |
73 | #define CFQ_WEIGHT_DEFAULT 500 | |
74 | ||
c5869807 TH |
75 | struct cfq_ttime { |
76 | unsigned long last_end_request; | |
77 | ||
78 | unsigned long ttime_total; | |
79 | unsigned long ttime_samples; | |
80 | unsigned long ttime_mean; | |
81 | }; | |
82 | ||
cc09e299 JA |
83 | /* |
84 | * Most of our rbtree usage is for sorting with min extraction, so | |
85 | * if we cache the leftmost node we don't have to walk down the tree | |
86 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
87 | * move this into the elevator for the rq sorting as well. | |
88 | */ | |
89 | struct cfq_rb_root { | |
90 | struct rb_root rb; | |
91 | struct rb_node *left; | |
aa6f6a3d | 92 | unsigned count; |
1fa8f6d6 | 93 | u64 min_vdisktime; |
f5f2b6ce | 94 | struct cfq_ttime ttime; |
cc09e299 | 95 | }; |
f5f2b6ce SL |
96 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
97 | .ttime = {.last_end_request = jiffies,},} | |
cc09e299 | 98 | |
6118b70b JA |
99 | /* |
100 | * Per process-grouping structure | |
101 | */ | |
102 | struct cfq_queue { | |
103 | /* reference count */ | |
30d7b944 | 104 | int ref; |
6118b70b JA |
105 | /* various state flags, see below */ |
106 | unsigned int flags; | |
107 | /* parent cfq_data */ | |
108 | struct cfq_data *cfqd; | |
109 | /* service_tree member */ | |
110 | struct rb_node rb_node; | |
111 | /* service_tree key */ | |
112 | unsigned long rb_key; | |
113 | /* prio tree member */ | |
114 | struct rb_node p_node; | |
115 | /* prio tree root we belong to, if any */ | |
116 | struct rb_root *p_root; | |
117 | /* sorted list of pending requests */ | |
118 | struct rb_root sort_list; | |
119 | /* if fifo isn't expired, next request to serve */ | |
120 | struct request *next_rq; | |
121 | /* requests queued in sort_list */ | |
122 | int queued[2]; | |
123 | /* currently allocated requests */ | |
124 | int allocated[2]; | |
125 | /* fifo list of requests in sort_list */ | |
126 | struct list_head fifo; | |
127 | ||
dae739eb VG |
128 | /* time when queue got scheduled in to dispatch first request. */ |
129 | unsigned long dispatch_start; | |
f75edf2d | 130 | unsigned int allocated_slice; |
c4081ba5 | 131 | unsigned int slice_dispatch; |
dae739eb VG |
132 | /* time when first request from queue completed and slice started. */ |
133 | unsigned long slice_start; | |
6118b70b JA |
134 | unsigned long slice_end; |
135 | long slice_resid; | |
6118b70b | 136 | |
65299a3b CH |
137 | /* pending priority requests */ |
138 | int prio_pending; | |
6118b70b JA |
139 | /* number of requests that are on the dispatch list or inside driver */ |
140 | int dispatched; | |
141 | ||
142 | /* io prio of this group */ | |
143 | unsigned short ioprio, org_ioprio; | |
4aede84b | 144 | unsigned short ioprio_class; |
6118b70b | 145 | |
c4081ba5 RK |
146 | pid_t pid; |
147 | ||
3dde36dd | 148 | u32 seek_history; |
b2c18e1e JM |
149 | sector_t last_request_pos; |
150 | ||
aa6f6a3d | 151 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 152 | struct cfq_queue *new_cfqq; |
cdb16e8f | 153 | struct cfq_group *cfqg; |
c4e7893e VG |
154 | /* Number of sectors dispatched from queue in single dispatch round */ |
155 | unsigned long nr_sectors; | |
6118b70b JA |
156 | }; |
157 | ||
c0324a02 | 158 | /* |
718eee05 | 159 | * First index in the service_trees. |
c0324a02 CZ |
160 | * IDLE is handled separately, so it has negative index |
161 | */ | |
3bf10fea | 162 | enum wl_class_t { |
c0324a02 | 163 | BE_WORKLOAD = 0, |
615f0259 VG |
164 | RT_WORKLOAD = 1, |
165 | IDLE_WORKLOAD = 2, | |
b4627321 | 166 | CFQ_PRIO_NR, |
c0324a02 CZ |
167 | }; |
168 | ||
718eee05 CZ |
169 | /* |
170 | * Second index in the service_trees. | |
171 | */ | |
172 | enum wl_type_t { | |
173 | ASYNC_WORKLOAD = 0, | |
174 | SYNC_NOIDLE_WORKLOAD = 1, | |
175 | SYNC_WORKLOAD = 2 | |
176 | }; | |
177 | ||
155fead9 TH |
178 | struct cfqg_stats { |
179 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
180 | /* total bytes transferred */ | |
181 | struct blkg_rwstat service_bytes; | |
182 | /* total IOs serviced, post merge */ | |
183 | struct blkg_rwstat serviced; | |
184 | /* number of ios merged */ | |
185 | struct blkg_rwstat merged; | |
186 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
187 | struct blkg_rwstat service_time; | |
188 | /* total time spent waiting in scheduler queue in ns */ | |
189 | struct blkg_rwstat wait_time; | |
190 | /* number of IOs queued up */ | |
191 | struct blkg_rwstat queued; | |
192 | /* total sectors transferred */ | |
193 | struct blkg_stat sectors; | |
194 | /* total disk time and nr sectors dispatched by this group */ | |
195 | struct blkg_stat time; | |
196 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
197 | /* time not charged to this cgroup */ | |
198 | struct blkg_stat unaccounted_time; | |
199 | /* sum of number of ios queued across all samples */ | |
200 | struct blkg_stat avg_queue_size_sum; | |
201 | /* count of samples taken for average */ | |
202 | struct blkg_stat avg_queue_size_samples; | |
203 | /* how many times this group has been removed from service tree */ | |
204 | struct blkg_stat dequeue; | |
205 | /* total time spent waiting for it to be assigned a timeslice. */ | |
206 | struct blkg_stat group_wait_time; | |
3c798398 | 207 | /* time spent idling for this blkcg_gq */ |
155fead9 TH |
208 | struct blkg_stat idle_time; |
209 | /* total time with empty current active q with other requests queued */ | |
210 | struct blkg_stat empty_time; | |
211 | /* fields after this shouldn't be cleared on stat reset */ | |
212 | uint64_t start_group_wait_time; | |
213 | uint64_t start_idle_time; | |
214 | uint64_t start_empty_time; | |
215 | uint16_t flags; | |
216 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
217 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
218 | }; | |
219 | ||
e48453c3 AA |
220 | /* Per-cgroup data */ |
221 | struct cfq_group_data { | |
222 | /* must be the first member */ | |
81437648 | 223 | struct blkcg_policy_data cpd; |
e48453c3 AA |
224 | |
225 | unsigned int weight; | |
226 | unsigned int leaf_weight; | |
227 | }; | |
228 | ||
cdb16e8f VG |
229 | /* This is per cgroup per device grouping structure */ |
230 | struct cfq_group { | |
f95a04af TH |
231 | /* must be the first member */ |
232 | struct blkg_policy_data pd; | |
233 | ||
1fa8f6d6 VG |
234 | /* group service_tree member */ |
235 | struct rb_node rb_node; | |
236 | ||
237 | /* group service_tree key */ | |
238 | u64 vdisktime; | |
e71357e1 | 239 | |
7918ffb5 TH |
240 | /* |
241 | * The number of active cfqgs and sum of their weights under this | |
242 | * cfqg. This covers this cfqg's leaf_weight and all children's | |
243 | * weights, but does not cover weights of further descendants. | |
244 | * | |
245 | * If a cfqg is on the service tree, it's active. An active cfqg | |
246 | * also activates its parent and contributes to the children_weight | |
247 | * of the parent. | |
248 | */ | |
249 | int nr_active; | |
250 | unsigned int children_weight; | |
251 | ||
1d3650f7 TH |
252 | /* |
253 | * vfraction is the fraction of vdisktime that the tasks in this | |
254 | * cfqg are entitled to. This is determined by compounding the | |
255 | * ratios walking up from this cfqg to the root. | |
256 | * | |
257 | * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all | |
258 | * vfractions on a service tree is approximately 1. The sum may | |
259 | * deviate a bit due to rounding errors and fluctuations caused by | |
260 | * cfqgs entering and leaving the service tree. | |
261 | */ | |
262 | unsigned int vfraction; | |
263 | ||
e71357e1 TH |
264 | /* |
265 | * There are two weights - (internal) weight is the weight of this | |
266 | * cfqg against the sibling cfqgs. leaf_weight is the wight of | |
267 | * this cfqg against the child cfqgs. For the root cfqg, both | |
268 | * weights are kept in sync for backward compatibility. | |
269 | */ | |
25bc6b07 | 270 | unsigned int weight; |
8184f93e | 271 | unsigned int new_weight; |
3381cb8d | 272 | unsigned int dev_weight; |
1fa8f6d6 | 273 | |
e71357e1 TH |
274 | unsigned int leaf_weight; |
275 | unsigned int new_leaf_weight; | |
276 | unsigned int dev_leaf_weight; | |
277 | ||
1fa8f6d6 VG |
278 | /* number of cfqq currently on this group */ |
279 | int nr_cfqq; | |
280 | ||
cdb16e8f | 281 | /* |
4495a7d4 | 282 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
283 | * create the array for each prio class but at run time it is used |
284 | * only for RT and BE class and slot for IDLE class remains unused. | |
285 | * This is primarily done to avoid confusion and a gcc warning. | |
286 | */ | |
287 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
288 | /* | |
289 | * rr lists of queues with requests. We maintain service trees for | |
290 | * RT and BE classes. These trees are subdivided in subclasses | |
291 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
292 | * class there is no subclassification and all the cfq queues go on | |
293 | * a single tree service_tree_idle. | |
cdb16e8f VG |
294 | * Counts are embedded in the cfq_rb_root |
295 | */ | |
296 | struct cfq_rb_root service_trees[2][3]; | |
297 | struct cfq_rb_root service_tree_idle; | |
dae739eb | 298 | |
4d2ceea4 VG |
299 | unsigned long saved_wl_slice; |
300 | enum wl_type_t saved_wl_type; | |
301 | enum wl_class_t saved_wl_class; | |
4eef3049 | 302 | |
80bdf0c7 VG |
303 | /* number of requests that are on the dispatch list or inside driver */ |
304 | int dispatched; | |
7700fc4f | 305 | struct cfq_ttime ttime; |
0b39920b | 306 | struct cfqg_stats stats; /* stats for this cfqg */ |
60a83707 TH |
307 | |
308 | /* async queue for each priority case */ | |
309 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
310 | struct cfq_queue *async_idle_cfqq; | |
311 | ||
cdb16e8f | 312 | }; |
718eee05 | 313 | |
c5869807 TH |
314 | struct cfq_io_cq { |
315 | struct io_cq icq; /* must be the first member */ | |
316 | struct cfq_queue *cfqq[2]; | |
317 | struct cfq_ttime ttime; | |
598971bf TH |
318 | int ioprio; /* the current ioprio */ |
319 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
f4da8072 | 320 | uint64_t blkcg_serial_nr; /* the current blkcg serial */ |
598971bf | 321 | #endif |
c5869807 TH |
322 | }; |
323 | ||
22e2c507 JA |
324 | /* |
325 | * Per block device queue structure | |
326 | */ | |
1da177e4 | 327 | struct cfq_data { |
165125e1 | 328 | struct request_queue *queue; |
1fa8f6d6 VG |
329 | /* Root service tree for cfq_groups */ |
330 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 331 | struct cfq_group *root_group; |
22e2c507 | 332 | |
c0324a02 CZ |
333 | /* |
334 | * The priority currently being served | |
22e2c507 | 335 | */ |
4d2ceea4 VG |
336 | enum wl_class_t serving_wl_class; |
337 | enum wl_type_t serving_wl_type; | |
718eee05 | 338 | unsigned long workload_expires; |
cdb16e8f | 339 | struct cfq_group *serving_group; |
a36e71f9 JA |
340 | |
341 | /* | |
342 | * Each priority tree is sorted by next_request position. These | |
343 | * trees are used when determining if two or more queues are | |
344 | * interleaving requests (see cfq_close_cooperator). | |
345 | */ | |
346 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
347 | ||
22e2c507 | 348 | unsigned int busy_queues; |
ef8a41df | 349 | unsigned int busy_sync_queues; |
22e2c507 | 350 | |
53c583d2 CZ |
351 | int rq_in_driver; |
352 | int rq_in_flight[2]; | |
45333d5a AC |
353 | |
354 | /* | |
355 | * queue-depth detection | |
356 | */ | |
357 | int rq_queued; | |
25776e35 | 358 | int hw_tag; |
e459dd08 CZ |
359 | /* |
360 | * hw_tag can be | |
361 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
362 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
363 | * 0 => no NCQ | |
364 | */ | |
365 | int hw_tag_est_depth; | |
366 | unsigned int hw_tag_samples; | |
1da177e4 | 367 | |
22e2c507 JA |
368 | /* |
369 | * idle window management | |
370 | */ | |
371 | struct timer_list idle_slice_timer; | |
23e018a1 | 372 | struct work_struct unplug_work; |
1da177e4 | 373 | |
22e2c507 | 374 | struct cfq_queue *active_queue; |
c5869807 | 375 | struct cfq_io_cq *active_cic; |
22e2c507 | 376 | |
6d048f53 | 377 | sector_t last_position; |
1da177e4 | 378 | |
1da177e4 LT |
379 | /* |
380 | * tunables, see top of file | |
381 | */ | |
382 | unsigned int cfq_quantum; | |
22e2c507 | 383 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
384 | unsigned int cfq_back_penalty; |
385 | unsigned int cfq_back_max; | |
22e2c507 JA |
386 | unsigned int cfq_slice[2]; |
387 | unsigned int cfq_slice_async_rq; | |
388 | unsigned int cfq_slice_idle; | |
80bdf0c7 | 389 | unsigned int cfq_group_idle; |
963b72fc | 390 | unsigned int cfq_latency; |
5bf14c07 | 391 | unsigned int cfq_target_latency; |
d9ff4187 | 392 | |
6118b70b JA |
393 | /* |
394 | * Fallback dummy cfqq for extreme OOM conditions | |
395 | */ | |
396 | struct cfq_queue oom_cfqq; | |
365722bb | 397 | |
573412b2 | 398 | unsigned long last_delayed_sync; |
1da177e4 LT |
399 | }; |
400 | ||
25fb5169 | 401 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
60a83707 | 402 | static void cfq_put_queue(struct cfq_queue *cfqq); |
25fb5169 | 403 | |
34b98d03 | 404 | static struct cfq_rb_root *st_for(struct cfq_group *cfqg, |
3bf10fea | 405 | enum wl_class_t class, |
65b32a57 | 406 | enum wl_type_t type) |
c0324a02 | 407 | { |
1fa8f6d6 VG |
408 | if (!cfqg) |
409 | return NULL; | |
410 | ||
3bf10fea | 411 | if (class == IDLE_WORKLOAD) |
cdb16e8f | 412 | return &cfqg->service_tree_idle; |
c0324a02 | 413 | |
3bf10fea | 414 | return &cfqg->service_trees[class][type]; |
c0324a02 CZ |
415 | } |
416 | ||
3b18152c | 417 | enum cfqq_state_flags { |
b0b8d749 JA |
418 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
419 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 420 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 421 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
422 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
423 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
424 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 425 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 426 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 427 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 428 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 429 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 430 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
431 | }; |
432 | ||
433 | #define CFQ_CFQQ_FNS(name) \ | |
434 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
435 | { \ | |
fe094d98 | 436 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
437 | } \ |
438 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
439 | { \ | |
fe094d98 | 440 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
441 | } \ |
442 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
443 | { \ | |
fe094d98 | 444 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
445 | } |
446 | ||
447 | CFQ_CFQQ_FNS(on_rr); | |
448 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 449 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 450 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
451 | CFQ_CFQQ_FNS(fifo_expire); |
452 | CFQ_CFQQ_FNS(idle_window); | |
453 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 454 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 455 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 456 | CFQ_CFQQ_FNS(coop); |
ae54abed | 457 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 458 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 459 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
460 | #undef CFQ_CFQQ_FNS |
461 | ||
629ed0b1 | 462 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
2ce4d50f | 463 | |
155fead9 TH |
464 | /* cfqg stats flags */ |
465 | enum cfqg_stats_flags { | |
466 | CFQG_stats_waiting = 0, | |
467 | CFQG_stats_idling, | |
468 | CFQG_stats_empty, | |
629ed0b1 TH |
469 | }; |
470 | ||
155fead9 TH |
471 | #define CFQG_FLAG_FNS(name) \ |
472 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ | |
629ed0b1 | 473 | { \ |
155fead9 | 474 | stats->flags |= (1 << CFQG_stats_##name); \ |
629ed0b1 | 475 | } \ |
155fead9 | 476 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 477 | { \ |
155fead9 | 478 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
629ed0b1 | 479 | } \ |
155fead9 | 480 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 481 | { \ |
155fead9 | 482 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
629ed0b1 TH |
483 | } \ |
484 | ||
155fead9 TH |
485 | CFQG_FLAG_FNS(waiting) |
486 | CFQG_FLAG_FNS(idling) | |
487 | CFQG_FLAG_FNS(empty) | |
488 | #undef CFQG_FLAG_FNS | |
629ed0b1 TH |
489 | |
490 | /* This should be called with the queue_lock held. */ | |
155fead9 | 491 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
629ed0b1 TH |
492 | { |
493 | unsigned long long now; | |
494 | ||
155fead9 | 495 | if (!cfqg_stats_waiting(stats)) |
629ed0b1 TH |
496 | return; |
497 | ||
498 | now = sched_clock(); | |
499 | if (time_after64(now, stats->start_group_wait_time)) | |
500 | blkg_stat_add(&stats->group_wait_time, | |
501 | now - stats->start_group_wait_time); | |
155fead9 | 502 | cfqg_stats_clear_waiting(stats); |
629ed0b1 TH |
503 | } |
504 | ||
505 | /* This should be called with the queue_lock held. */ | |
155fead9 TH |
506 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
507 | struct cfq_group *curr_cfqg) | |
629ed0b1 | 508 | { |
155fead9 | 509 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 510 | |
155fead9 | 511 | if (cfqg_stats_waiting(stats)) |
629ed0b1 | 512 | return; |
155fead9 | 513 | if (cfqg == curr_cfqg) |
629ed0b1 | 514 | return; |
155fead9 TH |
515 | stats->start_group_wait_time = sched_clock(); |
516 | cfqg_stats_mark_waiting(stats); | |
629ed0b1 TH |
517 | } |
518 | ||
519 | /* This should be called with the queue_lock held. */ | |
155fead9 | 520 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
629ed0b1 TH |
521 | { |
522 | unsigned long long now; | |
523 | ||
155fead9 | 524 | if (!cfqg_stats_empty(stats)) |
629ed0b1 TH |
525 | return; |
526 | ||
527 | now = sched_clock(); | |
528 | if (time_after64(now, stats->start_empty_time)) | |
529 | blkg_stat_add(&stats->empty_time, | |
530 | now - stats->start_empty_time); | |
155fead9 | 531 | cfqg_stats_clear_empty(stats); |
629ed0b1 TH |
532 | } |
533 | ||
155fead9 | 534 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
629ed0b1 | 535 | { |
155fead9 | 536 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
629ed0b1 TH |
537 | } |
538 | ||
155fead9 | 539 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
629ed0b1 | 540 | { |
155fead9 | 541 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 542 | |
4d5e80a7 | 543 | if (blkg_rwstat_total(&stats->queued)) |
629ed0b1 TH |
544 | return; |
545 | ||
546 | /* | |
547 | * group is already marked empty. This can happen if cfqq got new | |
548 | * request in parent group and moved to this group while being added | |
549 | * to service tree. Just ignore the event and move on. | |
550 | */ | |
155fead9 | 551 | if (cfqg_stats_empty(stats)) |
629ed0b1 TH |
552 | return; |
553 | ||
554 | stats->start_empty_time = sched_clock(); | |
155fead9 | 555 | cfqg_stats_mark_empty(stats); |
629ed0b1 TH |
556 | } |
557 | ||
155fead9 | 558 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 559 | { |
155fead9 | 560 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 561 | |
155fead9 | 562 | if (cfqg_stats_idling(stats)) { |
629ed0b1 TH |
563 | unsigned long long now = sched_clock(); |
564 | ||
565 | if (time_after64(now, stats->start_idle_time)) | |
566 | blkg_stat_add(&stats->idle_time, | |
567 | now - stats->start_idle_time); | |
155fead9 | 568 | cfqg_stats_clear_idling(stats); |
629ed0b1 TH |
569 | } |
570 | } | |
571 | ||
155fead9 | 572 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 573 | { |
155fead9 | 574 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 575 | |
155fead9 | 576 | BUG_ON(cfqg_stats_idling(stats)); |
629ed0b1 TH |
577 | |
578 | stats->start_idle_time = sched_clock(); | |
155fead9 | 579 | cfqg_stats_mark_idling(stats); |
629ed0b1 TH |
580 | } |
581 | ||
155fead9 | 582 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
629ed0b1 | 583 | { |
155fead9 | 584 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
585 | |
586 | blkg_stat_add(&stats->avg_queue_size_sum, | |
4d5e80a7 | 587 | blkg_rwstat_total(&stats->queued)); |
629ed0b1 | 588 | blkg_stat_add(&stats->avg_queue_size_samples, 1); |
155fead9 | 589 | cfqg_stats_update_group_wait_time(stats); |
629ed0b1 TH |
590 | } |
591 | ||
592 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
593 | ||
f48ec1d7 TH |
594 | static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { } |
595 | static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } | |
596 | static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } | |
597 | static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } | |
598 | static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } | |
599 | static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } | |
600 | static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } | |
629ed0b1 TH |
601 | |
602 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
603 | ||
604 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
2ce4d50f | 605 | |
4ceab71b JA |
606 | static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd) |
607 | { | |
608 | return pd ? container_of(pd, struct cfq_group, pd) : NULL; | |
609 | } | |
610 | ||
611 | static struct cfq_group_data | |
612 | *cpd_to_cfqgd(struct blkcg_policy_data *cpd) | |
613 | { | |
81437648 | 614 | return cpd ? container_of(cpd, struct cfq_group_data, cpd) : NULL; |
4ceab71b JA |
615 | } |
616 | ||
617 | static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg) | |
618 | { | |
619 | return pd_to_blkg(&cfqg->pd); | |
620 | } | |
621 | ||
ffea73fc TH |
622 | static struct blkcg_policy blkcg_policy_cfq; |
623 | ||
624 | static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg) | |
625 | { | |
626 | return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq)); | |
627 | } | |
628 | ||
e48453c3 AA |
629 | static struct cfq_group_data *blkcg_to_cfqgd(struct blkcg *blkcg) |
630 | { | |
631 | return cpd_to_cfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_cfq)); | |
632 | } | |
633 | ||
d02f7aa8 | 634 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) |
7918ffb5 | 635 | { |
d02f7aa8 | 636 | struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent; |
7918ffb5 | 637 | |
d02f7aa8 | 638 | return pblkg ? blkg_to_cfqg(pblkg) : NULL; |
7918ffb5 TH |
639 | } |
640 | ||
eb7d8c07 TH |
641 | static inline void cfqg_get(struct cfq_group *cfqg) |
642 | { | |
643 | return blkg_get(cfqg_to_blkg(cfqg)); | |
644 | } | |
645 | ||
646 | static inline void cfqg_put(struct cfq_group *cfqg) | |
647 | { | |
648 | return blkg_put(cfqg_to_blkg(cfqg)); | |
649 | } | |
650 | ||
54e7ed12 TH |
651 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \ |
652 | char __pbuf[128]; \ | |
653 | \ | |
654 | blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \ | |
b226e5c4 VG |
655 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \ |
656 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
657 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
54e7ed12 TH |
658 | __pbuf, ##args); \ |
659 | } while (0) | |
660 | ||
661 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \ | |
662 | char __pbuf[128]; \ | |
663 | \ | |
664 | blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \ | |
665 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \ | |
666 | } while (0) | |
2868ef7b | 667 | |
155fead9 TH |
668 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
669 | struct cfq_group *curr_cfqg, int rw) | |
2ce4d50f | 670 | { |
155fead9 TH |
671 | blkg_rwstat_add(&cfqg->stats.queued, rw, 1); |
672 | cfqg_stats_end_empty_time(&cfqg->stats); | |
673 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); | |
2ce4d50f TH |
674 | } |
675 | ||
155fead9 TH |
676 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
677 | unsigned long time, unsigned long unaccounted_time) | |
2ce4d50f | 678 | { |
155fead9 | 679 | blkg_stat_add(&cfqg->stats.time, time); |
629ed0b1 | 680 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
155fead9 | 681 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
629ed0b1 | 682 | #endif |
2ce4d50f TH |
683 | } |
684 | ||
155fead9 | 685 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) |
2ce4d50f | 686 | { |
155fead9 | 687 | blkg_rwstat_add(&cfqg->stats.queued, rw, -1); |
2ce4d50f TH |
688 | } |
689 | ||
155fead9 | 690 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) |
2ce4d50f | 691 | { |
155fead9 | 692 | blkg_rwstat_add(&cfqg->stats.merged, rw, 1); |
2ce4d50f TH |
693 | } |
694 | ||
155fead9 TH |
695 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, |
696 | uint64_t bytes, int rw) | |
2ce4d50f | 697 | { |
155fead9 TH |
698 | blkg_stat_add(&cfqg->stats.sectors, bytes >> 9); |
699 | blkg_rwstat_add(&cfqg->stats.serviced, rw, 1); | |
700 | blkg_rwstat_add(&cfqg->stats.service_bytes, rw, bytes); | |
2ce4d50f TH |
701 | } |
702 | ||
155fead9 TH |
703 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
704 | uint64_t start_time, uint64_t io_start_time, int rw) | |
2ce4d50f | 705 | { |
155fead9 | 706 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 707 | unsigned long long now = sched_clock(); |
629ed0b1 TH |
708 | |
709 | if (time_after64(now, io_start_time)) | |
710 | blkg_rwstat_add(&stats->service_time, rw, now - io_start_time); | |
711 | if (time_after64(io_start_time, start_time)) | |
712 | blkg_rwstat_add(&stats->wait_time, rw, | |
713 | io_start_time - start_time); | |
2ce4d50f TH |
714 | } |
715 | ||
689665af TH |
716 | /* @stats = 0 */ |
717 | static void cfqg_stats_reset(struct cfqg_stats *stats) | |
155fead9 | 718 | { |
155fead9 TH |
719 | /* queued stats shouldn't be cleared */ |
720 | blkg_rwstat_reset(&stats->service_bytes); | |
721 | blkg_rwstat_reset(&stats->serviced); | |
722 | blkg_rwstat_reset(&stats->merged); | |
723 | blkg_rwstat_reset(&stats->service_time); | |
724 | blkg_rwstat_reset(&stats->wait_time); | |
725 | blkg_stat_reset(&stats->time); | |
726 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
727 | blkg_stat_reset(&stats->unaccounted_time); | |
728 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
729 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
730 | blkg_stat_reset(&stats->dequeue); | |
731 | blkg_stat_reset(&stats->group_wait_time); | |
732 | blkg_stat_reset(&stats->idle_time); | |
733 | blkg_stat_reset(&stats->empty_time); | |
734 | #endif | |
735 | } | |
736 | ||
0b39920b | 737 | /* @to += @from */ |
e6269c44 | 738 | static void cfqg_stats_add_aux(struct cfqg_stats *to, struct cfqg_stats *from) |
0b39920b TH |
739 | { |
740 | /* queued stats shouldn't be cleared */ | |
e6269c44 TH |
741 | blkg_rwstat_add_aux(&to->service_bytes, &from->service_bytes); |
742 | blkg_rwstat_add_aux(&to->serviced, &from->serviced); | |
743 | blkg_rwstat_add_aux(&to->merged, &from->merged); | |
744 | blkg_rwstat_add_aux(&to->service_time, &from->service_time); | |
745 | blkg_rwstat_add_aux(&to->wait_time, &from->wait_time); | |
746 | blkg_stat_add_aux(&from->time, &from->time); | |
0b39920b | 747 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
e6269c44 TH |
748 | blkg_stat_add_aux(&to->unaccounted_time, &from->unaccounted_time); |
749 | blkg_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum); | |
750 | blkg_stat_add_aux(&to->avg_queue_size_samples, &from->avg_queue_size_samples); | |
751 | blkg_stat_add_aux(&to->dequeue, &from->dequeue); | |
752 | blkg_stat_add_aux(&to->group_wait_time, &from->group_wait_time); | |
753 | blkg_stat_add_aux(&to->idle_time, &from->idle_time); | |
754 | blkg_stat_add_aux(&to->empty_time, &from->empty_time); | |
0b39920b TH |
755 | #endif |
756 | } | |
757 | ||
758 | /* | |
e6269c44 | 759 | * Transfer @cfqg's stats to its parent's aux counts so that the ancestors' |
0b39920b TH |
760 | * recursive stats can still account for the amount used by this cfqg after |
761 | * it's gone. | |
762 | */ | |
763 | static void cfqg_stats_xfer_dead(struct cfq_group *cfqg) | |
764 | { | |
765 | struct cfq_group *parent = cfqg_parent(cfqg); | |
766 | ||
767 | lockdep_assert_held(cfqg_to_blkg(cfqg)->q->queue_lock); | |
768 | ||
769 | if (unlikely(!parent)) | |
770 | return; | |
771 | ||
e6269c44 | 772 | cfqg_stats_add_aux(&parent->stats, &cfqg->stats); |
0b39920b | 773 | cfqg_stats_reset(&cfqg->stats); |
0b39920b TH |
774 | } |
775 | ||
eb7d8c07 TH |
776 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
777 | ||
d02f7aa8 | 778 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) { return NULL; } |
eb7d8c07 TH |
779 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
780 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
781 | ||
7b679138 | 782 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
b226e5c4 VG |
783 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \ |
784 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
785 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
786 | ##args) | |
4495a7d4 | 787 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 | 788 | |
155fead9 TH |
789 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
790 | struct cfq_group *curr_cfqg, int rw) { } | |
791 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, | |
792 | unsigned long time, unsigned long unaccounted_time) { } | |
793 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) { } | |
794 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) { } | |
795 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, | |
796 | uint64_t bytes, int rw) { } | |
797 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, | |
798 | uint64_t start_time, uint64_t io_start_time, int rw) { } | |
2ce4d50f | 799 | |
eb7d8c07 TH |
800 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
801 | ||
7b679138 JA |
802 | #define cfq_log(cfqd, fmt, args...) \ |
803 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
804 | ||
615f0259 VG |
805 | /* Traverses through cfq group service trees */ |
806 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
807 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
808 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
809 | : &cfqg->service_tree_idle; \ | |
810 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
811 | (i == IDLE_WORKLOAD && j == 0); \ | |
812 | j++, st = i < IDLE_WORKLOAD ? \ | |
813 | &cfqg->service_trees[i][j]: NULL) \ | |
814 | ||
f5f2b6ce SL |
815 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
816 | struct cfq_ttime *ttime, bool group_idle) | |
817 | { | |
818 | unsigned long slice; | |
819 | if (!sample_valid(ttime->ttime_samples)) | |
820 | return false; | |
821 | if (group_idle) | |
822 | slice = cfqd->cfq_group_idle; | |
823 | else | |
824 | slice = cfqd->cfq_slice_idle; | |
825 | return ttime->ttime_mean > slice; | |
826 | } | |
615f0259 | 827 | |
02b35081 VG |
828 | static inline bool iops_mode(struct cfq_data *cfqd) |
829 | { | |
830 | /* | |
831 | * If we are not idling on queues and it is a NCQ drive, parallel | |
832 | * execution of requests is on and measuring time is not possible | |
833 | * in most of the cases until and unless we drive shallower queue | |
834 | * depths and that becomes a performance bottleneck. In such cases | |
835 | * switch to start providing fairness in terms of number of IOs. | |
836 | */ | |
837 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
838 | return true; | |
839 | else | |
840 | return false; | |
841 | } | |
842 | ||
3bf10fea | 843 | static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq) |
c0324a02 CZ |
844 | { |
845 | if (cfq_class_idle(cfqq)) | |
846 | return IDLE_WORKLOAD; | |
847 | if (cfq_class_rt(cfqq)) | |
848 | return RT_WORKLOAD; | |
849 | return BE_WORKLOAD; | |
850 | } | |
851 | ||
718eee05 CZ |
852 | |
853 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
854 | { | |
855 | if (!cfq_cfqq_sync(cfqq)) | |
856 | return ASYNC_WORKLOAD; | |
857 | if (!cfq_cfqq_idle_window(cfqq)) | |
858 | return SYNC_NOIDLE_WORKLOAD; | |
859 | return SYNC_WORKLOAD; | |
860 | } | |
861 | ||
3bf10fea | 862 | static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class, |
58ff82f3 VG |
863 | struct cfq_data *cfqd, |
864 | struct cfq_group *cfqg) | |
c0324a02 | 865 | { |
3bf10fea | 866 | if (wl_class == IDLE_WORKLOAD) |
cdb16e8f | 867 | return cfqg->service_tree_idle.count; |
c0324a02 | 868 | |
34b98d03 VG |
869 | return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count + |
870 | cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count + | |
871 | cfqg->service_trees[wl_class][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
872 | } |
873 | ||
f26bd1f0 VG |
874 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
875 | struct cfq_group *cfqg) | |
876 | { | |
34b98d03 VG |
877 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count + |
878 | cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
f26bd1f0 VG |
879 | } |
880 | ||
165125e1 | 881 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 882 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
2da8de0b | 883 | struct cfq_io_cq *cic, struct bio *bio); |
91fac317 | 884 | |
c5869807 TH |
885 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
886 | { | |
887 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
888 | return container_of(icq, struct cfq_io_cq, icq); | |
889 | } | |
890 | ||
47fdd4ca TH |
891 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
892 | struct io_context *ioc) | |
893 | { | |
894 | if (ioc) | |
895 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
896 | return NULL; | |
897 | } | |
898 | ||
c5869807 | 899 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 900 | { |
a6151c3a | 901 | return cic->cfqq[is_sync]; |
91fac317 VT |
902 | } |
903 | ||
c5869807 TH |
904 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
905 | bool is_sync) | |
91fac317 | 906 | { |
a6151c3a | 907 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
908 | } |
909 | ||
c5869807 | 910 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 911 | { |
c5869807 | 912 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
913 | } |
914 | ||
91fac317 VT |
915 | /* |
916 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
917 | * set (in which case it could also be direct WRITE). | |
918 | */ | |
a6151c3a | 919 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 920 | { |
7b6d91da | 921 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 922 | } |
1da177e4 | 923 | |
99f95e52 AM |
924 | /* |
925 | * scheduler run of queue, if there are requests pending and no one in the | |
926 | * driver that will restart queueing | |
927 | */ | |
23e018a1 | 928 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 929 | { |
7b679138 JA |
930 | if (cfqd->busy_queues) { |
931 | cfq_log(cfqd, "schedule dispatch"); | |
59c3d45e | 932 | kblockd_schedule_work(&cfqd->unplug_work); |
7b679138 | 933 | } |
99f95e52 AM |
934 | } |
935 | ||
44f7c160 JA |
936 | /* |
937 | * Scale schedule slice based on io priority. Use the sync time slice only | |
938 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
939 | * io only, should not get full sync slice length. | |
940 | */ | |
a6151c3a | 941 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 942 | unsigned short prio) |
44f7c160 | 943 | { |
d9e7620e | 944 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 945 | |
d9e7620e JA |
946 | WARN_ON(prio >= IOPRIO_BE_NR); |
947 | ||
948 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
949 | } | |
44f7c160 | 950 | |
d9e7620e JA |
951 | static inline int |
952 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
953 | { | |
954 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
955 | } |
956 | ||
1d3650f7 TH |
957 | /** |
958 | * cfqg_scale_charge - scale disk time charge according to cfqg weight | |
959 | * @charge: disk time being charged | |
960 | * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT | |
961 | * | |
962 | * Scale @charge according to @vfraction, which is in range (0, 1]. The | |
963 | * scaling is inversely proportional. | |
964 | * | |
965 | * scaled = charge / vfraction | |
966 | * | |
967 | * The result is also in fixed point w/ CFQ_SERVICE_SHIFT. | |
968 | */ | |
969 | static inline u64 cfqg_scale_charge(unsigned long charge, | |
970 | unsigned int vfraction) | |
25bc6b07 | 971 | { |
1d3650f7 | 972 | u64 c = charge << CFQ_SERVICE_SHIFT; /* make it fixed point */ |
25bc6b07 | 973 | |
1d3650f7 TH |
974 | /* charge / vfraction */ |
975 | c <<= CFQ_SERVICE_SHIFT; | |
976 | do_div(c, vfraction); | |
977 | return c; | |
25bc6b07 VG |
978 | } |
979 | ||
980 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
981 | { | |
982 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
983 | if (delta > 0) | |
984 | min_vdisktime = vdisktime; | |
985 | ||
986 | return min_vdisktime; | |
987 | } | |
988 | ||
989 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
990 | { | |
991 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
992 | if (delta < 0) | |
993 | min_vdisktime = vdisktime; | |
994 | ||
995 | return min_vdisktime; | |
996 | } | |
997 | ||
998 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
999 | { | |
25bc6b07 VG |
1000 | struct cfq_group *cfqg; |
1001 | ||
25bc6b07 VG |
1002 | if (st->left) { |
1003 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
1004 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
1005 | cfqg->vdisktime); | |
25bc6b07 | 1006 | } |
25bc6b07 VG |
1007 | } |
1008 | ||
5db5d642 CZ |
1009 | /* |
1010 | * get averaged number of queues of RT/BE priority. | |
1011 | * average is updated, with a formula that gives more weight to higher numbers, | |
1012 | * to quickly follows sudden increases and decrease slowly | |
1013 | */ | |
1014 | ||
58ff82f3 VG |
1015 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
1016 | struct cfq_group *cfqg, bool rt) | |
5869619c | 1017 | { |
5db5d642 CZ |
1018 | unsigned min_q, max_q; |
1019 | unsigned mult = cfq_hist_divisor - 1; | |
1020 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 1021 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 1022 | |
58ff82f3 VG |
1023 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
1024 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
1025 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 1026 | cfq_hist_divisor; |
58ff82f3 VG |
1027 | return cfqg->busy_queues_avg[rt]; |
1028 | } | |
1029 | ||
1030 | static inline unsigned | |
1031 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1032 | { | |
41cad6ab | 1033 | return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT; |
5db5d642 CZ |
1034 | } |
1035 | ||
c553f8e3 | 1036 | static inline unsigned |
ba5bd520 | 1037 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 1038 | { |
5db5d642 CZ |
1039 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
1040 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
1041 | /* |
1042 | * interested queues (we consider only the ones with the same | |
1043 | * priority class in the cfq group) | |
1044 | */ | |
1045 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
1046 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
1047 | unsigned sync_slice = cfqd->cfq_slice[1]; |
1048 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
1049 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
1050 | ||
1051 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
1052 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
1053 | /* scale low_slice according to IO priority | |
1054 | * and sync vs async */ | |
1055 | unsigned low_slice = | |
1056 | min(slice, base_low_slice * slice / sync_slice); | |
1057 | /* the adapted slice value is scaled to fit all iqs | |
1058 | * into the target latency */ | |
58ff82f3 | 1059 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
1060 | low_slice); |
1061 | } | |
1062 | } | |
c553f8e3 SL |
1063 | return slice; |
1064 | } | |
1065 | ||
1066 | static inline void | |
1067 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1068 | { | |
ba5bd520 | 1069 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 1070 | |
dae739eb | 1071 | cfqq->slice_start = jiffies; |
5db5d642 | 1072 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 1073 | cfqq->allocated_slice = slice; |
7b679138 | 1074 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
1075 | } |
1076 | ||
1077 | /* | |
1078 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
1079 | * isn't valid until the first request from the dispatch is activated | |
1080 | * and the slice time set. | |
1081 | */ | |
a6151c3a | 1082 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
1083 | { |
1084 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 1085 | return false; |
44f7c160 | 1086 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 1087 | return false; |
44f7c160 | 1088 | |
c1e44756 | 1089 | return true; |
44f7c160 JA |
1090 | } |
1091 | ||
1da177e4 | 1092 | /* |
5e705374 | 1093 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 1094 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 1095 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 1096 | */ |
5e705374 | 1097 | static struct request * |
cf7c25cf | 1098 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 1099 | { |
cf7c25cf | 1100 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 1101 | unsigned long back_max; |
e8a99053 AM |
1102 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
1103 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
1104 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 1105 | |
5e705374 JA |
1106 | if (rq1 == NULL || rq1 == rq2) |
1107 | return rq2; | |
1108 | if (rq2 == NULL) | |
1109 | return rq1; | |
9c2c38a1 | 1110 | |
229836bd NK |
1111 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
1112 | return rq_is_sync(rq1) ? rq1 : rq2; | |
1113 | ||
65299a3b CH |
1114 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
1115 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 1116 | |
83096ebf TH |
1117 | s1 = blk_rq_pos(rq1); |
1118 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 1119 | |
1da177e4 LT |
1120 | /* |
1121 | * by definition, 1KiB is 2 sectors | |
1122 | */ | |
1123 | back_max = cfqd->cfq_back_max * 2; | |
1124 | ||
1125 | /* | |
1126 | * Strict one way elevator _except_ in the case where we allow | |
1127 | * short backward seeks which are biased as twice the cost of a | |
1128 | * similar forward seek. | |
1129 | */ | |
1130 | if (s1 >= last) | |
1131 | d1 = s1 - last; | |
1132 | else if (s1 + back_max >= last) | |
1133 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
1134 | else | |
e8a99053 | 1135 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
1136 | |
1137 | if (s2 >= last) | |
1138 | d2 = s2 - last; | |
1139 | else if (s2 + back_max >= last) | |
1140 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
1141 | else | |
e8a99053 | 1142 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
1143 | |
1144 | /* Found required data */ | |
e8a99053 AM |
1145 | |
1146 | /* | |
1147 | * By doing switch() on the bit mask "wrap" we avoid having to | |
1148 | * check two variables for all permutations: --> faster! | |
1149 | */ | |
1150 | switch (wrap) { | |
5e705374 | 1151 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 1152 | if (d1 < d2) |
5e705374 | 1153 | return rq1; |
e8a99053 | 1154 | else if (d2 < d1) |
5e705374 | 1155 | return rq2; |
e8a99053 AM |
1156 | else { |
1157 | if (s1 >= s2) | |
5e705374 | 1158 | return rq1; |
e8a99053 | 1159 | else |
5e705374 | 1160 | return rq2; |
e8a99053 | 1161 | } |
1da177e4 | 1162 | |
e8a99053 | 1163 | case CFQ_RQ2_WRAP: |
5e705374 | 1164 | return rq1; |
e8a99053 | 1165 | case CFQ_RQ1_WRAP: |
5e705374 JA |
1166 | return rq2; |
1167 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
1168 | default: |
1169 | /* | |
1170 | * Since both rqs are wrapped, | |
1171 | * start with the one that's further behind head | |
1172 | * (--> only *one* back seek required), | |
1173 | * since back seek takes more time than forward. | |
1174 | */ | |
1175 | if (s1 <= s2) | |
5e705374 | 1176 | return rq1; |
1da177e4 | 1177 | else |
5e705374 | 1178 | return rq2; |
1da177e4 LT |
1179 | } |
1180 | } | |
1181 | ||
498d3aa2 JA |
1182 | /* |
1183 | * The below is leftmost cache rbtree addon | |
1184 | */ | |
0871714e | 1185 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 1186 | { |
615f0259 VG |
1187 | /* Service tree is empty */ |
1188 | if (!root->count) | |
1189 | return NULL; | |
1190 | ||
cc09e299 JA |
1191 | if (!root->left) |
1192 | root->left = rb_first(&root->rb); | |
1193 | ||
0871714e JA |
1194 | if (root->left) |
1195 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
1196 | ||
1197 | return NULL; | |
cc09e299 JA |
1198 | } |
1199 | ||
1fa8f6d6 VG |
1200 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
1201 | { | |
1202 | if (!root->left) | |
1203 | root->left = rb_first(&root->rb); | |
1204 | ||
1205 | if (root->left) | |
1206 | return rb_entry_cfqg(root->left); | |
1207 | ||
1208 | return NULL; | |
1209 | } | |
1210 | ||
a36e71f9 JA |
1211 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
1212 | { | |
1213 | rb_erase(n, root); | |
1214 | RB_CLEAR_NODE(n); | |
1215 | } | |
1216 | ||
cc09e299 JA |
1217 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
1218 | { | |
1219 | if (root->left == n) | |
1220 | root->left = NULL; | |
a36e71f9 | 1221 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 1222 | --root->count; |
cc09e299 JA |
1223 | } |
1224 | ||
1da177e4 LT |
1225 | /* |
1226 | * would be nice to take fifo expire time into account as well | |
1227 | */ | |
5e705374 JA |
1228 | static struct request * |
1229 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1230 | struct request *last) | |
1da177e4 | 1231 | { |
21183b07 JA |
1232 | struct rb_node *rbnext = rb_next(&last->rb_node); |
1233 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 1234 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 1235 | |
21183b07 | 1236 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
1237 | |
1238 | if (rbprev) | |
5e705374 | 1239 | prev = rb_entry_rq(rbprev); |
1da177e4 | 1240 | |
21183b07 | 1241 | if (rbnext) |
5e705374 | 1242 | next = rb_entry_rq(rbnext); |
21183b07 JA |
1243 | else { |
1244 | rbnext = rb_first(&cfqq->sort_list); | |
1245 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 1246 | next = rb_entry_rq(rbnext); |
21183b07 | 1247 | } |
1da177e4 | 1248 | |
cf7c25cf | 1249 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
1250 | } |
1251 | ||
d9e7620e JA |
1252 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
1253 | struct cfq_queue *cfqq) | |
1da177e4 | 1254 | { |
d9e7620e JA |
1255 | /* |
1256 | * just an approximation, should be ok. | |
1257 | */ | |
cdb16e8f | 1258 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 1259 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
1260 | } |
1261 | ||
1fa8f6d6 VG |
1262 | static inline s64 |
1263 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1264 | { | |
1265 | return cfqg->vdisktime - st->min_vdisktime; | |
1266 | } | |
1267 | ||
1268 | static void | |
1269 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1270 | { | |
1271 | struct rb_node **node = &st->rb.rb_node; | |
1272 | struct rb_node *parent = NULL; | |
1273 | struct cfq_group *__cfqg; | |
1274 | s64 key = cfqg_key(st, cfqg); | |
1275 | int left = 1; | |
1276 | ||
1277 | while (*node != NULL) { | |
1278 | parent = *node; | |
1279 | __cfqg = rb_entry_cfqg(parent); | |
1280 | ||
1281 | if (key < cfqg_key(st, __cfqg)) | |
1282 | node = &parent->rb_left; | |
1283 | else { | |
1284 | node = &parent->rb_right; | |
1285 | left = 0; | |
1286 | } | |
1287 | } | |
1288 | ||
1289 | if (left) | |
1290 | st->left = &cfqg->rb_node; | |
1291 | ||
1292 | rb_link_node(&cfqg->rb_node, parent, node); | |
1293 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
1294 | } | |
1295 | ||
7b5af5cf TM |
1296 | /* |
1297 | * This has to be called only on activation of cfqg | |
1298 | */ | |
1fa8f6d6 | 1299 | static void |
8184f93e JT |
1300 | cfq_update_group_weight(struct cfq_group *cfqg) |
1301 | { | |
3381cb8d | 1302 | if (cfqg->new_weight) { |
8184f93e | 1303 | cfqg->weight = cfqg->new_weight; |
3381cb8d | 1304 | cfqg->new_weight = 0; |
8184f93e | 1305 | } |
e15693ef TM |
1306 | } |
1307 | ||
1308 | static void | |
1309 | cfq_update_group_leaf_weight(struct cfq_group *cfqg) | |
1310 | { | |
1311 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
e71357e1 TH |
1312 | |
1313 | if (cfqg->new_leaf_weight) { | |
1314 | cfqg->leaf_weight = cfqg->new_leaf_weight; | |
1315 | cfqg->new_leaf_weight = 0; | |
1316 | } | |
8184f93e JT |
1317 | } |
1318 | ||
1319 | static void | |
1320 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1321 | { | |
1d3650f7 | 1322 | unsigned int vfr = 1 << CFQ_SERVICE_SHIFT; /* start with 1 */ |
7918ffb5 | 1323 | struct cfq_group *pos = cfqg; |
1d3650f7 | 1324 | struct cfq_group *parent; |
7918ffb5 TH |
1325 | bool propagate; |
1326 | ||
1327 | /* add to the service tree */ | |
8184f93e JT |
1328 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
1329 | ||
7b5af5cf TM |
1330 | /* |
1331 | * Update leaf_weight. We cannot update weight at this point | |
1332 | * because cfqg might already have been activated and is | |
1333 | * contributing its current weight to the parent's child_weight. | |
1334 | */ | |
e15693ef | 1335 | cfq_update_group_leaf_weight(cfqg); |
8184f93e | 1336 | __cfq_group_service_tree_add(st, cfqg); |
7918ffb5 TH |
1337 | |
1338 | /* | |
1d3650f7 TH |
1339 | * Activate @cfqg and calculate the portion of vfraction @cfqg is |
1340 | * entitled to. vfraction is calculated by walking the tree | |
1341 | * towards the root calculating the fraction it has at each level. | |
1342 | * The compounded ratio is how much vfraction @cfqg owns. | |
1343 | * | |
1344 | * Start with the proportion tasks in this cfqg has against active | |
1345 | * children cfqgs - its leaf_weight against children_weight. | |
7918ffb5 TH |
1346 | */ |
1347 | propagate = !pos->nr_active++; | |
1348 | pos->children_weight += pos->leaf_weight; | |
1d3650f7 | 1349 | vfr = vfr * pos->leaf_weight / pos->children_weight; |
7918ffb5 | 1350 | |
1d3650f7 TH |
1351 | /* |
1352 | * Compound ->weight walking up the tree. Both activation and | |
1353 | * vfraction calculation are done in the same loop. Propagation | |
1354 | * stops once an already activated node is met. vfraction | |
1355 | * calculation should always continue to the root. | |
1356 | */ | |
d02f7aa8 | 1357 | while ((parent = cfqg_parent(pos))) { |
1d3650f7 | 1358 | if (propagate) { |
e15693ef | 1359 | cfq_update_group_weight(pos); |
1d3650f7 TH |
1360 | propagate = !parent->nr_active++; |
1361 | parent->children_weight += pos->weight; | |
1362 | } | |
1363 | vfr = vfr * pos->weight / parent->children_weight; | |
7918ffb5 TH |
1364 | pos = parent; |
1365 | } | |
1d3650f7 TH |
1366 | |
1367 | cfqg->vfraction = max_t(unsigned, vfr, 1); | |
8184f93e JT |
1368 | } |
1369 | ||
1370 | static void | |
1371 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
1372 | { |
1373 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1374 | struct cfq_group *__cfqg; | |
1375 | struct rb_node *n; | |
1376 | ||
1377 | cfqg->nr_cfqq++; | |
760701bf | 1378 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
1379 | return; |
1380 | ||
1381 | /* | |
1382 | * Currently put the group at the end. Later implement something | |
1383 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 1384 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
1385 | */ |
1386 | n = rb_last(&st->rb); | |
1387 | if (n) { | |
1388 | __cfqg = rb_entry_cfqg(n); | |
1389 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
1390 | } else | |
1391 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
1392 | cfq_group_service_tree_add(st, cfqg); |
1393 | } | |
1fa8f6d6 | 1394 | |
8184f93e JT |
1395 | static void |
1396 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1397 | { | |
7918ffb5 TH |
1398 | struct cfq_group *pos = cfqg; |
1399 | bool propagate; | |
1400 | ||
1401 | /* | |
1402 | * Undo activation from cfq_group_service_tree_add(). Deactivate | |
1403 | * @cfqg and propagate deactivation upwards. | |
1404 | */ | |
1405 | propagate = !--pos->nr_active; | |
1406 | pos->children_weight -= pos->leaf_weight; | |
1407 | ||
1408 | while (propagate) { | |
d02f7aa8 | 1409 | struct cfq_group *parent = cfqg_parent(pos); |
7918ffb5 TH |
1410 | |
1411 | /* @pos has 0 nr_active at this point */ | |
1412 | WARN_ON_ONCE(pos->children_weight); | |
1d3650f7 | 1413 | pos->vfraction = 0; |
7918ffb5 TH |
1414 | |
1415 | if (!parent) | |
1416 | break; | |
1417 | ||
1418 | propagate = !--parent->nr_active; | |
1419 | parent->children_weight -= pos->weight; | |
1420 | pos = parent; | |
1421 | } | |
1422 | ||
1423 | /* remove from the service tree */ | |
8184f93e JT |
1424 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1425 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
1426 | } |
1427 | ||
1428 | static void | |
8184f93e | 1429 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
1430 | { |
1431 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1432 | ||
1433 | BUG_ON(cfqg->nr_cfqq < 1); | |
1434 | cfqg->nr_cfqq--; | |
25bc6b07 | 1435 | |
1fa8f6d6 VG |
1436 | /* If there are other cfq queues under this group, don't delete it */ |
1437 | if (cfqg->nr_cfqq) | |
1438 | return; | |
1439 | ||
2868ef7b | 1440 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 1441 | cfq_group_service_tree_del(st, cfqg); |
4d2ceea4 | 1442 | cfqg->saved_wl_slice = 0; |
155fead9 | 1443 | cfqg_stats_update_dequeue(cfqg); |
dae739eb VG |
1444 | } |
1445 | ||
167400d3 JT |
1446 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
1447 | unsigned int *unaccounted_time) | |
dae739eb | 1448 | { |
f75edf2d | 1449 | unsigned int slice_used; |
dae739eb VG |
1450 | |
1451 | /* | |
1452 | * Queue got expired before even a single request completed or | |
1453 | * got expired immediately after first request completion. | |
1454 | */ | |
1455 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
1456 | /* | |
1457 | * Also charge the seek time incurred to the group, otherwise | |
1458 | * if there are mutiple queues in the group, each can dispatch | |
1459 | * a single request on seeky media and cause lots of seek time | |
1460 | * and group will never know it. | |
1461 | */ | |
1462 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
1463 | 1); | |
1464 | } else { | |
1465 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
1466 | if (slice_used > cfqq->allocated_slice) { |
1467 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 1468 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
1469 | } |
1470 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
1471 | *unaccounted_time += cfqq->slice_start - | |
1472 | cfqq->dispatch_start; | |
dae739eb VG |
1473 | } |
1474 | ||
dae739eb VG |
1475 | return slice_used; |
1476 | } | |
1477 | ||
1478 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1479 | struct cfq_queue *cfqq) |
dae739eb VG |
1480 | { |
1481 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 1482 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1483 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1484 | - cfqg->service_tree_idle.count; | |
1d3650f7 | 1485 | unsigned int vfr; |
f26bd1f0 VG |
1486 | |
1487 | BUG_ON(nr_sync < 0); | |
167400d3 | 1488 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1489 | |
02b35081 VG |
1490 | if (iops_mode(cfqd)) |
1491 | charge = cfqq->slice_dispatch; | |
1492 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1493 | charge = cfqq->allocated_slice; | |
dae739eb | 1494 | |
1d3650f7 TH |
1495 | /* |
1496 | * Can't update vdisktime while on service tree and cfqg->vfraction | |
1497 | * is valid only while on it. Cache vfr, leave the service tree, | |
1498 | * update vdisktime and go back on. The re-addition to the tree | |
1499 | * will also update the weights as necessary. | |
1500 | */ | |
1501 | vfr = cfqg->vfraction; | |
8184f93e | 1502 | cfq_group_service_tree_del(st, cfqg); |
1d3650f7 | 1503 | cfqg->vdisktime += cfqg_scale_charge(charge, vfr); |
8184f93e | 1504 | cfq_group_service_tree_add(st, cfqg); |
dae739eb VG |
1505 | |
1506 | /* This group is being expired. Save the context */ | |
1507 | if (time_after(cfqd->workload_expires, jiffies)) { | |
4d2ceea4 | 1508 | cfqg->saved_wl_slice = cfqd->workload_expires |
dae739eb | 1509 | - jiffies; |
4d2ceea4 VG |
1510 | cfqg->saved_wl_type = cfqd->serving_wl_type; |
1511 | cfqg->saved_wl_class = cfqd->serving_wl_class; | |
dae739eb | 1512 | } else |
4d2ceea4 | 1513 | cfqg->saved_wl_slice = 0; |
2868ef7b VG |
1514 | |
1515 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1516 | st->min_vdisktime); | |
fd16d263 JP |
1517 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1518 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1519 | used_sl, cfqq->slice_dispatch, charge, | |
1520 | iops_mode(cfqd), cfqq->nr_sectors); | |
155fead9 TH |
1521 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
1522 | cfqg_stats_set_start_empty_time(cfqg); | |
1fa8f6d6 VG |
1523 | } |
1524 | ||
f51b802c TH |
1525 | /** |
1526 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1527 | * @cfqg: cfq_group to initialize | |
1528 | * | |
1529 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1530 | * is enabled or not. | |
1531 | */ | |
1532 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1533 | { | |
1534 | struct cfq_rb_root *st; | |
1535 | int i, j; | |
1536 | ||
1537 | for_each_cfqg_st(cfqg, i, j, st) | |
1538 | *st = CFQ_RB_ROOT; | |
1539 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1540 | ||
1541 | cfqg->ttime.last_end_request = jiffies; | |
1542 | } | |
1543 | ||
25fb5169 | 1544 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
90d3839b PZ |
1545 | static void cfqg_stats_init(struct cfqg_stats *stats) |
1546 | { | |
1547 | blkg_rwstat_init(&stats->service_bytes); | |
1548 | blkg_rwstat_init(&stats->serviced); | |
1549 | blkg_rwstat_init(&stats->merged); | |
1550 | blkg_rwstat_init(&stats->service_time); | |
1551 | blkg_rwstat_init(&stats->wait_time); | |
1552 | blkg_rwstat_init(&stats->queued); | |
1553 | ||
1554 | blkg_stat_init(&stats->sectors); | |
1555 | blkg_stat_init(&stats->time); | |
1556 | ||
1557 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
1558 | blkg_stat_init(&stats->unaccounted_time); | |
1559 | blkg_stat_init(&stats->avg_queue_size_sum); | |
1560 | blkg_stat_init(&stats->avg_queue_size_samples); | |
1561 | blkg_stat_init(&stats->dequeue); | |
1562 | blkg_stat_init(&stats->group_wait_time); | |
1563 | blkg_stat_init(&stats->idle_time); | |
1564 | blkg_stat_init(&stats->empty_time); | |
1565 | #endif | |
1566 | } | |
1567 | ||
e4a9bde9 TH |
1568 | static struct blkcg_policy_data *cfq_cpd_alloc(gfp_t gfp) |
1569 | { | |
1570 | struct cfq_group_data *cgd; | |
1571 | ||
1572 | cgd = kzalloc(sizeof(*cgd), GFP_KERNEL); | |
1573 | if (!cgd) | |
1574 | return NULL; | |
1575 | return &cgd->cpd; | |
1576 | } | |
1577 | ||
81437648 | 1578 | static void cfq_cpd_init(struct blkcg_policy_data *cpd) |
e48453c3 | 1579 | { |
81437648 | 1580 | struct cfq_group_data *cgd = cpd_to_cfqgd(cpd); |
e48453c3 | 1581 | |
81437648 | 1582 | if (cpd_to_blkcg(cpd) == &blkcg_root) { |
e48453c3 AA |
1583 | cgd->weight = 2 * CFQ_WEIGHT_DEFAULT; |
1584 | cgd->leaf_weight = 2 * CFQ_WEIGHT_DEFAULT; | |
1585 | } else { | |
1586 | cgd->weight = CFQ_WEIGHT_DEFAULT; | |
1587 | cgd->leaf_weight = CFQ_WEIGHT_DEFAULT; | |
1588 | } | |
1589 | } | |
1590 | ||
e4a9bde9 TH |
1591 | static void cfq_cpd_free(struct blkcg_policy_data *cpd) |
1592 | { | |
1593 | kfree(cpd_to_cfqgd(cpd)); | |
1594 | } | |
1595 | ||
001bea73 TH |
1596 | static struct blkg_policy_data *cfq_pd_alloc(gfp_t gfp, int node) |
1597 | { | |
b2ce2643 TH |
1598 | struct cfq_group *cfqg; |
1599 | ||
1600 | cfqg = kzalloc_node(sizeof(*cfqg), gfp, node); | |
1601 | if (!cfqg) | |
1602 | return NULL; | |
1603 | ||
1604 | cfq_init_cfqg_base(cfqg); | |
1605 | cfqg_stats_init(&cfqg->stats); | |
b2ce2643 TH |
1606 | |
1607 | return &cfqg->pd; | |
001bea73 TH |
1608 | } |
1609 | ||
a9520cd6 | 1610 | static void cfq_pd_init(struct blkg_policy_data *pd) |
f469a7b4 | 1611 | { |
a9520cd6 TH |
1612 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
1613 | struct cfq_group_data *cgd = blkcg_to_cfqgd(pd->blkg->blkcg); | |
25fb5169 | 1614 | |
e48453c3 AA |
1615 | cfqg->weight = cgd->weight; |
1616 | cfqg->leaf_weight = cgd->leaf_weight; | |
25fb5169 VG |
1617 | } |
1618 | ||
a9520cd6 | 1619 | static void cfq_pd_offline(struct blkg_policy_data *pd) |
0b39920b | 1620 | { |
a9520cd6 | 1621 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
60a83707 TH |
1622 | int i; |
1623 | ||
1624 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
1625 | if (cfqg->async_cfqq[0][i]) | |
1626 | cfq_put_queue(cfqg->async_cfqq[0][i]); | |
1627 | if (cfqg->async_cfqq[1][i]) | |
1628 | cfq_put_queue(cfqg->async_cfqq[1][i]); | |
1629 | } | |
1630 | ||
1631 | if (cfqg->async_idle_cfqq) | |
1632 | cfq_put_queue(cfqg->async_idle_cfqq); | |
1633 | ||
0b39920b TH |
1634 | /* |
1635 | * @blkg is going offline and will be ignored by | |
1636 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
1637 | * that they don't get lost. If IOs complete after this point, the | |
1638 | * stats for them will be lost. Oh well... | |
1639 | */ | |
60a83707 | 1640 | cfqg_stats_xfer_dead(cfqg); |
0b39920b TH |
1641 | } |
1642 | ||
001bea73 TH |
1643 | static void cfq_pd_free(struct blkg_policy_data *pd) |
1644 | { | |
1645 | return kfree(pd); | |
1646 | } | |
1647 | ||
a9520cd6 | 1648 | static void cfq_pd_reset_stats(struct blkg_policy_data *pd) |
689665af | 1649 | { |
a9520cd6 | 1650 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
689665af TH |
1651 | |
1652 | cfqg_stats_reset(&cfqg->stats); | |
25fb5169 VG |
1653 | } |
1654 | ||
ae118896 TH |
1655 | static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd, |
1656 | struct blkcg *blkcg) | |
25fb5169 | 1657 | { |
ae118896 | 1658 | struct blkcg_gq *blkg; |
f469a7b4 | 1659 | |
ae118896 TH |
1660 | blkg = blkg_lookup(blkcg, cfqd->queue); |
1661 | if (likely(blkg)) | |
1662 | return blkg_to_cfqg(blkg); | |
1663 | return NULL; | |
25fb5169 VG |
1664 | } |
1665 | ||
1666 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1667 | { | |
25fb5169 | 1668 | cfqq->cfqg = cfqg; |
b1c35769 | 1669 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1670 | cfqg_get(cfqg); |
b1c35769 VG |
1671 | } |
1672 | ||
f95a04af TH |
1673 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, |
1674 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1675 | { |
f95a04af | 1676 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
3381cb8d TH |
1677 | |
1678 | if (!cfqg->dev_weight) | |
60c2bc2d | 1679 | return 0; |
f95a04af | 1680 | return __blkg_prfill_u64(sf, pd, cfqg->dev_weight); |
60c2bc2d TH |
1681 | } |
1682 | ||
2da8ca82 | 1683 | static int cfqg_print_weight_device(struct seq_file *sf, void *v) |
60c2bc2d | 1684 | { |
2da8ca82 TH |
1685 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1686 | cfqg_prfill_weight_device, &blkcg_policy_cfq, | |
1687 | 0, false); | |
60c2bc2d TH |
1688 | return 0; |
1689 | } | |
1690 | ||
e71357e1 TH |
1691 | static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf, |
1692 | struct blkg_policy_data *pd, int off) | |
1693 | { | |
1694 | struct cfq_group *cfqg = pd_to_cfqg(pd); | |
1695 | ||
1696 | if (!cfqg->dev_leaf_weight) | |
1697 | return 0; | |
1698 | return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight); | |
1699 | } | |
1700 | ||
2da8ca82 | 1701 | static int cfqg_print_leaf_weight_device(struct seq_file *sf, void *v) |
e71357e1 | 1702 | { |
2da8ca82 TH |
1703 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1704 | cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq, | |
1705 | 0, false); | |
e71357e1 TH |
1706 | return 0; |
1707 | } | |
1708 | ||
2da8ca82 | 1709 | static int cfq_print_weight(struct seq_file *sf, void *v) |
60c2bc2d | 1710 | { |
e48453c3 | 1711 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
9470e4a6 JA |
1712 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); |
1713 | unsigned int val = 0; | |
e48453c3 | 1714 | |
9470e4a6 JA |
1715 | if (cgd) |
1716 | val = cgd->weight; | |
1717 | ||
1718 | seq_printf(sf, "%u\n", val); | |
60c2bc2d TH |
1719 | return 0; |
1720 | } | |
1721 | ||
2da8ca82 | 1722 | static int cfq_print_leaf_weight(struct seq_file *sf, void *v) |
e71357e1 | 1723 | { |
e48453c3 | 1724 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
9470e4a6 JA |
1725 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); |
1726 | unsigned int val = 0; | |
1727 | ||
1728 | if (cgd) | |
1729 | val = cgd->leaf_weight; | |
e48453c3 | 1730 | |
9470e4a6 | 1731 | seq_printf(sf, "%u\n", val); |
e71357e1 TH |
1732 | return 0; |
1733 | } | |
1734 | ||
451af504 TH |
1735 | static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of, |
1736 | char *buf, size_t nbytes, loff_t off, | |
1737 | bool is_leaf_weight) | |
60c2bc2d | 1738 | { |
451af504 | 1739 | struct blkcg *blkcg = css_to_blkcg(of_css(of)); |
60c2bc2d | 1740 | struct blkg_conf_ctx ctx; |
3381cb8d | 1741 | struct cfq_group *cfqg; |
e48453c3 | 1742 | struct cfq_group_data *cfqgd; |
60c2bc2d TH |
1743 | int ret; |
1744 | ||
3c798398 | 1745 | ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx); |
60c2bc2d TH |
1746 | if (ret) |
1747 | return ret; | |
1748 | ||
1749 | ret = -EINVAL; | |
3381cb8d | 1750 | cfqg = blkg_to_cfqg(ctx.blkg); |
e48453c3 | 1751 | cfqgd = blkcg_to_cfqgd(blkcg); |
ae994ea9 JA |
1752 | if (!cfqg || !cfqgd) |
1753 | goto err; | |
1754 | ||
a2b1693b | 1755 | if (!ctx.v || (ctx.v >= CFQ_WEIGHT_MIN && ctx.v <= CFQ_WEIGHT_MAX)) { |
e71357e1 TH |
1756 | if (!is_leaf_weight) { |
1757 | cfqg->dev_weight = ctx.v; | |
e48453c3 | 1758 | cfqg->new_weight = ctx.v ?: cfqgd->weight; |
e71357e1 TH |
1759 | } else { |
1760 | cfqg->dev_leaf_weight = ctx.v; | |
e48453c3 | 1761 | cfqg->new_leaf_weight = ctx.v ?: cfqgd->leaf_weight; |
e71357e1 | 1762 | } |
60c2bc2d TH |
1763 | ret = 0; |
1764 | } | |
1765 | ||
ae994ea9 | 1766 | err: |
60c2bc2d | 1767 | blkg_conf_finish(&ctx); |
451af504 | 1768 | return ret ?: nbytes; |
60c2bc2d TH |
1769 | } |
1770 | ||
451af504 TH |
1771 | static ssize_t cfqg_set_weight_device(struct kernfs_open_file *of, |
1772 | char *buf, size_t nbytes, loff_t off) | |
e71357e1 | 1773 | { |
451af504 | 1774 | return __cfqg_set_weight_device(of, buf, nbytes, off, false); |
e71357e1 TH |
1775 | } |
1776 | ||
451af504 TH |
1777 | static ssize_t cfqg_set_leaf_weight_device(struct kernfs_open_file *of, |
1778 | char *buf, size_t nbytes, loff_t off) | |
e71357e1 | 1779 | { |
451af504 | 1780 | return __cfqg_set_weight_device(of, buf, nbytes, off, true); |
e71357e1 TH |
1781 | } |
1782 | ||
182446d0 TH |
1783 | static int __cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft, |
1784 | u64 val, bool is_leaf_weight) | |
60c2bc2d | 1785 | { |
182446d0 | 1786 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 1787 | struct blkcg_gq *blkg; |
e48453c3 | 1788 | struct cfq_group_data *cfqgd; |
ae994ea9 | 1789 | int ret = 0; |
60c2bc2d | 1790 | |
3381cb8d | 1791 | if (val < CFQ_WEIGHT_MIN || val > CFQ_WEIGHT_MAX) |
60c2bc2d TH |
1792 | return -EINVAL; |
1793 | ||
1794 | spin_lock_irq(&blkcg->lock); | |
e48453c3 | 1795 | cfqgd = blkcg_to_cfqgd(blkcg); |
ae994ea9 JA |
1796 | if (!cfqgd) { |
1797 | ret = -EINVAL; | |
1798 | goto out; | |
1799 | } | |
e71357e1 TH |
1800 | |
1801 | if (!is_leaf_weight) | |
e48453c3 | 1802 | cfqgd->weight = val; |
e71357e1 | 1803 | else |
e48453c3 | 1804 | cfqgd->leaf_weight = val; |
60c2bc2d | 1805 | |
b67bfe0d | 1806 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
3381cb8d | 1807 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
60c2bc2d | 1808 | |
e71357e1 TH |
1809 | if (!cfqg) |
1810 | continue; | |
1811 | ||
1812 | if (!is_leaf_weight) { | |
1813 | if (!cfqg->dev_weight) | |
e48453c3 | 1814 | cfqg->new_weight = cfqgd->weight; |
e71357e1 TH |
1815 | } else { |
1816 | if (!cfqg->dev_leaf_weight) | |
e48453c3 | 1817 | cfqg->new_leaf_weight = cfqgd->leaf_weight; |
e71357e1 | 1818 | } |
60c2bc2d TH |
1819 | } |
1820 | ||
ae994ea9 | 1821 | out: |
60c2bc2d | 1822 | spin_unlock_irq(&blkcg->lock); |
ae994ea9 | 1823 | return ret; |
60c2bc2d TH |
1824 | } |
1825 | ||
182446d0 TH |
1826 | static int cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft, |
1827 | u64 val) | |
e71357e1 | 1828 | { |
182446d0 | 1829 | return __cfq_set_weight(css, cft, val, false); |
e71357e1 TH |
1830 | } |
1831 | ||
182446d0 TH |
1832 | static int cfq_set_leaf_weight(struct cgroup_subsys_state *css, |
1833 | struct cftype *cft, u64 val) | |
e71357e1 | 1834 | { |
182446d0 | 1835 | return __cfq_set_weight(css, cft, val, true); |
e71357e1 TH |
1836 | } |
1837 | ||
2da8ca82 | 1838 | static int cfqg_print_stat(struct seq_file *sf, void *v) |
5bc4afb1 | 1839 | { |
2da8ca82 TH |
1840 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat, |
1841 | &blkcg_policy_cfq, seq_cft(sf)->private, false); | |
5bc4afb1 TH |
1842 | return 0; |
1843 | } | |
1844 | ||
2da8ca82 | 1845 | static int cfqg_print_rwstat(struct seq_file *sf, void *v) |
5bc4afb1 | 1846 | { |
2da8ca82 TH |
1847 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat, |
1848 | &blkcg_policy_cfq, seq_cft(sf)->private, true); | |
5bc4afb1 TH |
1849 | return 0; |
1850 | } | |
1851 | ||
43114018 TH |
1852 | static u64 cfqg_prfill_stat_recursive(struct seq_file *sf, |
1853 | struct blkg_policy_data *pd, int off) | |
1854 | { | |
e6269c44 | 1855 | u64 sum = blkg_stat_recursive_sum(pd, off); |
43114018 TH |
1856 | |
1857 | return __blkg_prfill_u64(sf, pd, sum); | |
1858 | } | |
1859 | ||
1860 | static u64 cfqg_prfill_rwstat_recursive(struct seq_file *sf, | |
1861 | struct blkg_policy_data *pd, int off) | |
1862 | { | |
e6269c44 | 1863 | struct blkg_rwstat sum = blkg_rwstat_recursive_sum(pd, off); |
43114018 TH |
1864 | |
1865 | return __blkg_prfill_rwstat(sf, pd, &sum); | |
1866 | } | |
1867 | ||
2da8ca82 | 1868 | static int cfqg_print_stat_recursive(struct seq_file *sf, void *v) |
43114018 | 1869 | { |
2da8ca82 TH |
1870 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1871 | cfqg_prfill_stat_recursive, &blkcg_policy_cfq, | |
1872 | seq_cft(sf)->private, false); | |
43114018 TH |
1873 | return 0; |
1874 | } | |
1875 | ||
2da8ca82 | 1876 | static int cfqg_print_rwstat_recursive(struct seq_file *sf, void *v) |
43114018 | 1877 | { |
2da8ca82 TH |
1878 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1879 | cfqg_prfill_rwstat_recursive, &blkcg_policy_cfq, | |
1880 | seq_cft(sf)->private, true); | |
43114018 TH |
1881 | return 0; |
1882 | } | |
1883 | ||
60c2bc2d | 1884 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
f95a04af TH |
1885 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, |
1886 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1887 | { |
f95a04af | 1888 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
155fead9 | 1889 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
60c2bc2d TH |
1890 | u64 v = 0; |
1891 | ||
1892 | if (samples) { | |
155fead9 | 1893 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
f3cff25f | 1894 | v = div64_u64(v, samples); |
60c2bc2d | 1895 | } |
f95a04af | 1896 | __blkg_prfill_u64(sf, pd, v); |
60c2bc2d TH |
1897 | return 0; |
1898 | } | |
1899 | ||
1900 | /* print avg_queue_size */ | |
2da8ca82 | 1901 | static int cfqg_print_avg_queue_size(struct seq_file *sf, void *v) |
60c2bc2d | 1902 | { |
2da8ca82 TH |
1903 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1904 | cfqg_prfill_avg_queue_size, &blkcg_policy_cfq, | |
1905 | 0, false); | |
60c2bc2d TH |
1906 | return 0; |
1907 | } | |
1908 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1909 | ||
1910 | static struct cftype cfq_blkcg_files[] = { | |
1d3650f7 | 1911 | /* on root, weight is mapped to leaf_weight */ |
60c2bc2d TH |
1912 | { |
1913 | .name = "weight_device", | |
1d3650f7 | 1914 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 1915 | .seq_show = cfqg_print_leaf_weight_device, |
451af504 | 1916 | .write = cfqg_set_leaf_weight_device, |
60c2bc2d TH |
1917 | }, |
1918 | { | |
1919 | .name = "weight", | |
1d3650f7 | 1920 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 1921 | .seq_show = cfq_print_leaf_weight, |
1d3650f7 | 1922 | .write_u64 = cfq_set_leaf_weight, |
60c2bc2d | 1923 | }, |
e71357e1 | 1924 | |
1d3650f7 | 1925 | /* no such mapping necessary for !roots */ |
60c2bc2d TH |
1926 | { |
1927 | .name = "weight_device", | |
1d3650f7 | 1928 | .flags = CFTYPE_NOT_ON_ROOT, |
2da8ca82 | 1929 | .seq_show = cfqg_print_weight_device, |
451af504 | 1930 | .write = cfqg_set_weight_device, |
60c2bc2d TH |
1931 | }, |
1932 | { | |
1933 | .name = "weight", | |
1d3650f7 | 1934 | .flags = CFTYPE_NOT_ON_ROOT, |
2da8ca82 | 1935 | .seq_show = cfq_print_weight, |
3381cb8d | 1936 | .write_u64 = cfq_set_weight, |
60c2bc2d | 1937 | }, |
e71357e1 | 1938 | |
e71357e1 TH |
1939 | { |
1940 | .name = "leaf_weight_device", | |
2da8ca82 | 1941 | .seq_show = cfqg_print_leaf_weight_device, |
451af504 | 1942 | .write = cfqg_set_leaf_weight_device, |
e71357e1 TH |
1943 | }, |
1944 | { | |
1945 | .name = "leaf_weight", | |
2da8ca82 | 1946 | .seq_show = cfq_print_leaf_weight, |
e71357e1 TH |
1947 | .write_u64 = cfq_set_leaf_weight, |
1948 | }, | |
1949 | ||
43114018 | 1950 | /* statistics, covers only the tasks in the cfqg */ |
60c2bc2d TH |
1951 | { |
1952 | .name = "time", | |
5bc4afb1 | 1953 | .private = offsetof(struct cfq_group, stats.time), |
2da8ca82 | 1954 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
1955 | }, |
1956 | { | |
1957 | .name = "sectors", | |
5bc4afb1 | 1958 | .private = offsetof(struct cfq_group, stats.sectors), |
2da8ca82 | 1959 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
1960 | }, |
1961 | { | |
1962 | .name = "io_service_bytes", | |
5bc4afb1 | 1963 | .private = offsetof(struct cfq_group, stats.service_bytes), |
2da8ca82 | 1964 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
1965 | }, |
1966 | { | |
1967 | .name = "io_serviced", | |
5bc4afb1 | 1968 | .private = offsetof(struct cfq_group, stats.serviced), |
2da8ca82 | 1969 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
1970 | }, |
1971 | { | |
1972 | .name = "io_service_time", | |
5bc4afb1 | 1973 | .private = offsetof(struct cfq_group, stats.service_time), |
2da8ca82 | 1974 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
1975 | }, |
1976 | { | |
1977 | .name = "io_wait_time", | |
5bc4afb1 | 1978 | .private = offsetof(struct cfq_group, stats.wait_time), |
2da8ca82 | 1979 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
1980 | }, |
1981 | { | |
1982 | .name = "io_merged", | |
5bc4afb1 | 1983 | .private = offsetof(struct cfq_group, stats.merged), |
2da8ca82 | 1984 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
1985 | }, |
1986 | { | |
1987 | .name = "io_queued", | |
5bc4afb1 | 1988 | .private = offsetof(struct cfq_group, stats.queued), |
2da8ca82 | 1989 | .seq_show = cfqg_print_rwstat, |
60c2bc2d | 1990 | }, |
43114018 TH |
1991 | |
1992 | /* the same statictics which cover the cfqg and its descendants */ | |
1993 | { | |
1994 | .name = "time_recursive", | |
1995 | .private = offsetof(struct cfq_group, stats.time), | |
2da8ca82 | 1996 | .seq_show = cfqg_print_stat_recursive, |
43114018 TH |
1997 | }, |
1998 | { | |
1999 | .name = "sectors_recursive", | |
2000 | .private = offsetof(struct cfq_group, stats.sectors), | |
2da8ca82 | 2001 | .seq_show = cfqg_print_stat_recursive, |
43114018 TH |
2002 | }, |
2003 | { | |
2004 | .name = "io_service_bytes_recursive", | |
2005 | .private = offsetof(struct cfq_group, stats.service_bytes), | |
2da8ca82 | 2006 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2007 | }, |
2008 | { | |
2009 | .name = "io_serviced_recursive", | |
2010 | .private = offsetof(struct cfq_group, stats.serviced), | |
2da8ca82 | 2011 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2012 | }, |
2013 | { | |
2014 | .name = "io_service_time_recursive", | |
2015 | .private = offsetof(struct cfq_group, stats.service_time), | |
2da8ca82 | 2016 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2017 | }, |
2018 | { | |
2019 | .name = "io_wait_time_recursive", | |
2020 | .private = offsetof(struct cfq_group, stats.wait_time), | |
2da8ca82 | 2021 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2022 | }, |
2023 | { | |
2024 | .name = "io_merged_recursive", | |
2025 | .private = offsetof(struct cfq_group, stats.merged), | |
2da8ca82 | 2026 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2027 | }, |
2028 | { | |
2029 | .name = "io_queued_recursive", | |
2030 | .private = offsetof(struct cfq_group, stats.queued), | |
2da8ca82 | 2031 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 | 2032 | }, |
60c2bc2d TH |
2033 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
2034 | { | |
2035 | .name = "avg_queue_size", | |
2da8ca82 | 2036 | .seq_show = cfqg_print_avg_queue_size, |
60c2bc2d TH |
2037 | }, |
2038 | { | |
2039 | .name = "group_wait_time", | |
5bc4afb1 | 2040 | .private = offsetof(struct cfq_group, stats.group_wait_time), |
2da8ca82 | 2041 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2042 | }, |
2043 | { | |
2044 | .name = "idle_time", | |
5bc4afb1 | 2045 | .private = offsetof(struct cfq_group, stats.idle_time), |
2da8ca82 | 2046 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2047 | }, |
2048 | { | |
2049 | .name = "empty_time", | |
5bc4afb1 | 2050 | .private = offsetof(struct cfq_group, stats.empty_time), |
2da8ca82 | 2051 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2052 | }, |
2053 | { | |
2054 | .name = "dequeue", | |
5bc4afb1 | 2055 | .private = offsetof(struct cfq_group, stats.dequeue), |
2da8ca82 | 2056 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2057 | }, |
2058 | { | |
2059 | .name = "unaccounted_time", | |
5bc4afb1 | 2060 | .private = offsetof(struct cfq_group, stats.unaccounted_time), |
2da8ca82 | 2061 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2062 | }, |
2063 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
2064 | { } /* terminate */ | |
2065 | }; | |
25fb5169 | 2066 | #else /* GROUP_IOSCHED */ |
ae118896 TH |
2067 | static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd, |
2068 | struct blkcg *blkcg) | |
25fb5169 | 2069 | { |
f51b802c | 2070 | return cfqd->root_group; |
25fb5169 | 2071 | } |
7f1dc8a2 | 2072 | |
25fb5169 VG |
2073 | static inline void |
2074 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
2075 | cfqq->cfqg = cfqg; | |
2076 | } | |
2077 | ||
2078 | #endif /* GROUP_IOSCHED */ | |
2079 | ||
498d3aa2 | 2080 | /* |
c0324a02 | 2081 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
2082 | * requests waiting to be processed. It is sorted in the order that |
2083 | * we will service the queues. | |
2084 | */ | |
a36e71f9 | 2085 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2086 | bool add_front) |
d9e7620e | 2087 | { |
0871714e JA |
2088 | struct rb_node **p, *parent; |
2089 | struct cfq_queue *__cfqq; | |
d9e7620e | 2090 | unsigned long rb_key; |
34b98d03 | 2091 | struct cfq_rb_root *st; |
498d3aa2 | 2092 | int left; |
dae739eb | 2093 | int new_cfqq = 1; |
ae30c286 | 2094 | |
34b98d03 | 2095 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq)); |
0871714e JA |
2096 | if (cfq_class_idle(cfqq)) { |
2097 | rb_key = CFQ_IDLE_DELAY; | |
34b98d03 | 2098 | parent = rb_last(&st->rb); |
0871714e JA |
2099 | if (parent && parent != &cfqq->rb_node) { |
2100 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2101 | rb_key += __cfqq->rb_key; | |
2102 | } else | |
2103 | rb_key += jiffies; | |
2104 | } else if (!add_front) { | |
b9c8946b JA |
2105 | /* |
2106 | * Get our rb key offset. Subtract any residual slice | |
2107 | * value carried from last service. A negative resid | |
2108 | * count indicates slice overrun, and this should position | |
2109 | * the next service time further away in the tree. | |
2110 | */ | |
edd75ffd | 2111 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 2112 | rb_key -= cfqq->slice_resid; |
edd75ffd | 2113 | cfqq->slice_resid = 0; |
48e025e6 CZ |
2114 | } else { |
2115 | rb_key = -HZ; | |
34b98d03 | 2116 | __cfqq = cfq_rb_first(st); |
48e025e6 CZ |
2117 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
2118 | } | |
1da177e4 | 2119 | |
d9e7620e | 2120 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 2121 | new_cfqq = 0; |
99f9628a | 2122 | /* |
d9e7620e | 2123 | * same position, nothing more to do |
99f9628a | 2124 | */ |
34b98d03 | 2125 | if (rb_key == cfqq->rb_key && cfqq->service_tree == st) |
d9e7620e | 2126 | return; |
1da177e4 | 2127 | |
aa6f6a3d CZ |
2128 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
2129 | cfqq->service_tree = NULL; | |
1da177e4 | 2130 | } |
d9e7620e | 2131 | |
498d3aa2 | 2132 | left = 1; |
0871714e | 2133 | parent = NULL; |
34b98d03 VG |
2134 | cfqq->service_tree = st; |
2135 | p = &st->rb.rb_node; | |
d9e7620e JA |
2136 | while (*p) { |
2137 | parent = *p; | |
2138 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2139 | ||
0c534e0a | 2140 | /* |
c0324a02 | 2141 | * sort by key, that represents service time. |
0c534e0a | 2142 | */ |
c0324a02 | 2143 | if (time_before(rb_key, __cfqq->rb_key)) |
1f23f121 | 2144 | p = &parent->rb_left; |
c0324a02 | 2145 | else { |
1f23f121 | 2146 | p = &parent->rb_right; |
cc09e299 | 2147 | left = 0; |
c0324a02 | 2148 | } |
d9e7620e JA |
2149 | } |
2150 | ||
cc09e299 | 2151 | if (left) |
34b98d03 | 2152 | st->left = &cfqq->rb_node; |
cc09e299 | 2153 | |
d9e7620e JA |
2154 | cfqq->rb_key = rb_key; |
2155 | rb_link_node(&cfqq->rb_node, parent, p); | |
34b98d03 VG |
2156 | rb_insert_color(&cfqq->rb_node, &st->rb); |
2157 | st->count++; | |
20359f27 | 2158 | if (add_front || !new_cfqq) |
dae739eb | 2159 | return; |
8184f93e | 2160 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
2161 | } |
2162 | ||
a36e71f9 | 2163 | static struct cfq_queue * |
f2d1f0ae JA |
2164 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
2165 | sector_t sector, struct rb_node **ret_parent, | |
2166 | struct rb_node ***rb_link) | |
a36e71f9 | 2167 | { |
a36e71f9 JA |
2168 | struct rb_node **p, *parent; |
2169 | struct cfq_queue *cfqq = NULL; | |
2170 | ||
2171 | parent = NULL; | |
2172 | p = &root->rb_node; | |
2173 | while (*p) { | |
2174 | struct rb_node **n; | |
2175 | ||
2176 | parent = *p; | |
2177 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
2178 | ||
2179 | /* | |
2180 | * Sort strictly based on sector. Smallest to the left, | |
2181 | * largest to the right. | |
2182 | */ | |
2e46e8b2 | 2183 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 2184 | n = &(*p)->rb_right; |
2e46e8b2 | 2185 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
2186 | n = &(*p)->rb_left; |
2187 | else | |
2188 | break; | |
2189 | p = n; | |
3ac6c9f8 | 2190 | cfqq = NULL; |
a36e71f9 JA |
2191 | } |
2192 | ||
2193 | *ret_parent = parent; | |
2194 | if (rb_link) | |
2195 | *rb_link = p; | |
3ac6c9f8 | 2196 | return cfqq; |
a36e71f9 JA |
2197 | } |
2198 | ||
2199 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2200 | { | |
a36e71f9 JA |
2201 | struct rb_node **p, *parent; |
2202 | struct cfq_queue *__cfqq; | |
2203 | ||
f2d1f0ae JA |
2204 | if (cfqq->p_root) { |
2205 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2206 | cfqq->p_root = NULL; | |
2207 | } | |
a36e71f9 JA |
2208 | |
2209 | if (cfq_class_idle(cfqq)) | |
2210 | return; | |
2211 | if (!cfqq->next_rq) | |
2212 | return; | |
2213 | ||
f2d1f0ae | 2214 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
2215 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
2216 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
2217 | if (!__cfqq) { |
2218 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
2219 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
2220 | } else | |
2221 | cfqq->p_root = NULL; | |
a36e71f9 JA |
2222 | } |
2223 | ||
498d3aa2 JA |
2224 | /* |
2225 | * Update cfqq's position in the service tree. | |
2226 | */ | |
edd75ffd | 2227 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 2228 | { |
6d048f53 JA |
2229 | /* |
2230 | * Resorting requires the cfqq to be on the RR list already. | |
2231 | */ | |
a36e71f9 | 2232 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 2233 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
2234 | cfq_prio_tree_add(cfqd, cfqq); |
2235 | } | |
6d048f53 JA |
2236 | } |
2237 | ||
1da177e4 LT |
2238 | /* |
2239 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 2240 | * the pending list according to last request service |
1da177e4 | 2241 | */ |
febffd61 | 2242 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2243 | { |
7b679138 | 2244 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
2245 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
2246 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 2247 | cfqd->busy_queues++; |
ef8a41df SL |
2248 | if (cfq_cfqq_sync(cfqq)) |
2249 | cfqd->busy_sync_queues++; | |
1da177e4 | 2250 | |
edd75ffd | 2251 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
2252 | } |
2253 | ||
498d3aa2 JA |
2254 | /* |
2255 | * Called when the cfqq no longer has requests pending, remove it from | |
2256 | * the service tree. | |
2257 | */ | |
febffd61 | 2258 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2259 | { |
7b679138 | 2260 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
2261 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
2262 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 2263 | |
aa6f6a3d CZ |
2264 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
2265 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
2266 | cfqq->service_tree = NULL; | |
2267 | } | |
f2d1f0ae JA |
2268 | if (cfqq->p_root) { |
2269 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2270 | cfqq->p_root = NULL; | |
2271 | } | |
d9e7620e | 2272 | |
8184f93e | 2273 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
2274 | BUG_ON(!cfqd->busy_queues); |
2275 | cfqd->busy_queues--; | |
ef8a41df SL |
2276 | if (cfq_cfqq_sync(cfqq)) |
2277 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
2278 | } |
2279 | ||
2280 | /* | |
2281 | * rb tree support functions | |
2282 | */ | |
febffd61 | 2283 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 2284 | { |
5e705374 | 2285 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 2286 | const int sync = rq_is_sync(rq); |
1da177e4 | 2287 | |
b4878f24 JA |
2288 | BUG_ON(!cfqq->queued[sync]); |
2289 | cfqq->queued[sync]--; | |
1da177e4 | 2290 | |
5e705374 | 2291 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 2292 | |
f04a6424 VG |
2293 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
2294 | /* | |
2295 | * Queue will be deleted from service tree when we actually | |
2296 | * expire it later. Right now just remove it from prio tree | |
2297 | * as it is empty. | |
2298 | */ | |
2299 | if (cfqq->p_root) { | |
2300 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2301 | cfqq->p_root = NULL; | |
2302 | } | |
2303 | } | |
1da177e4 LT |
2304 | } |
2305 | ||
5e705374 | 2306 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 2307 | { |
5e705374 | 2308 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2309 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 2310 | struct request *prev; |
1da177e4 | 2311 | |
5380a101 | 2312 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 2313 | |
796d5116 | 2314 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
2315 | |
2316 | if (!cfq_cfqq_on_rr(cfqq)) | |
2317 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
2318 | |
2319 | /* | |
2320 | * check if this request is a better next-serve candidate | |
2321 | */ | |
a36e71f9 | 2322 | prev = cfqq->next_rq; |
cf7c25cf | 2323 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
2324 | |
2325 | /* | |
2326 | * adjust priority tree position, if ->next_rq changes | |
2327 | */ | |
2328 | if (prev != cfqq->next_rq) | |
2329 | cfq_prio_tree_add(cfqd, cfqq); | |
2330 | ||
5044eed4 | 2331 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
2332 | } |
2333 | ||
febffd61 | 2334 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 2335 | { |
5380a101 JA |
2336 | elv_rb_del(&cfqq->sort_list, rq); |
2337 | cfqq->queued[rq_is_sync(rq)]--; | |
155fead9 | 2338 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
5e705374 | 2339 | cfq_add_rq_rb(rq); |
155fead9 TH |
2340 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
2341 | rq->cmd_flags); | |
1da177e4 LT |
2342 | } |
2343 | ||
206dc69b JA |
2344 | static struct request * |
2345 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 2346 | { |
206dc69b | 2347 | struct task_struct *tsk = current; |
c5869807 | 2348 | struct cfq_io_cq *cic; |
206dc69b | 2349 | struct cfq_queue *cfqq; |
1da177e4 | 2350 | |
4ac845a2 | 2351 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2352 | if (!cic) |
2353 | return NULL; | |
2354 | ||
2355 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
f73a1c7d KO |
2356 | if (cfqq) |
2357 | return elv_rb_find(&cfqq->sort_list, bio_end_sector(bio)); | |
1da177e4 | 2358 | |
1da177e4 LT |
2359 | return NULL; |
2360 | } | |
2361 | ||
165125e1 | 2362 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2363 | { |
22e2c507 | 2364 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 2365 | |
53c583d2 | 2366 | cfqd->rq_in_driver++; |
7b679138 | 2367 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 2368 | cfqd->rq_in_driver); |
25776e35 | 2369 | |
5b93629b | 2370 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
2371 | } |
2372 | ||
165125e1 | 2373 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2374 | { |
b4878f24 JA |
2375 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2376 | ||
53c583d2 CZ |
2377 | WARN_ON(!cfqd->rq_in_driver); |
2378 | cfqd->rq_in_driver--; | |
7b679138 | 2379 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 2380 | cfqd->rq_in_driver); |
1da177e4 LT |
2381 | } |
2382 | ||
b4878f24 | 2383 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 2384 | { |
5e705374 | 2385 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 2386 | |
5e705374 JA |
2387 | if (cfqq->next_rq == rq) |
2388 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 2389 | |
b4878f24 | 2390 | list_del_init(&rq->queuelist); |
5e705374 | 2391 | cfq_del_rq_rb(rq); |
374f84ac | 2392 | |
45333d5a | 2393 | cfqq->cfqd->rq_queued--; |
155fead9 | 2394 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
65299a3b CH |
2395 | if (rq->cmd_flags & REQ_PRIO) { |
2396 | WARN_ON(!cfqq->prio_pending); | |
2397 | cfqq->prio_pending--; | |
b53d1ed7 | 2398 | } |
1da177e4 LT |
2399 | } |
2400 | ||
165125e1 JA |
2401 | static int cfq_merge(struct request_queue *q, struct request **req, |
2402 | struct bio *bio) | |
1da177e4 LT |
2403 | { |
2404 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2405 | struct request *__rq; | |
1da177e4 | 2406 | |
206dc69b | 2407 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 2408 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
2409 | *req = __rq; |
2410 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
2411 | } |
2412 | ||
2413 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
2414 | } |
2415 | ||
165125e1 | 2416 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 2417 | int type) |
1da177e4 | 2418 | { |
21183b07 | 2419 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 2420 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 2421 | |
5e705374 | 2422 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 2423 | } |
1da177e4 LT |
2424 | } |
2425 | ||
812d4026 DS |
2426 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
2427 | struct bio *bio) | |
2428 | { | |
155fead9 | 2429 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_rw); |
812d4026 DS |
2430 | } |
2431 | ||
1da177e4 | 2432 | static void |
165125e1 | 2433 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
2434 | struct request *next) |
2435 | { | |
cf7c25cf | 2436 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
2437 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2438 | ||
22e2c507 JA |
2439 | /* |
2440 | * reposition in fifo if next is older than rq | |
2441 | */ | |
2442 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
8b4922d3 | 2443 | time_before(next->fifo_time, rq->fifo_time) && |
3d106fba | 2444 | cfqq == RQ_CFQQ(next)) { |
22e2c507 | 2445 | list_move(&rq->queuelist, &next->queuelist); |
8b4922d3 | 2446 | rq->fifo_time = next->fifo_time; |
30996f40 | 2447 | } |
22e2c507 | 2448 | |
cf7c25cf CZ |
2449 | if (cfqq->next_rq == next) |
2450 | cfqq->next_rq = rq; | |
b4878f24 | 2451 | cfq_remove_request(next); |
155fead9 | 2452 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
4a0b75c7 SL |
2453 | |
2454 | cfqq = RQ_CFQQ(next); | |
2455 | /* | |
2456 | * all requests of this queue are merged to other queues, delete it | |
2457 | * from the service tree. If it's the active_queue, | |
2458 | * cfq_dispatch_requests() will choose to expire it or do idle | |
2459 | */ | |
2460 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
2461 | cfqq != cfqd->active_queue) | |
2462 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
2463 | } |
2464 | ||
165125e1 | 2465 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
2466 | struct bio *bio) |
2467 | { | |
2468 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 2469 | struct cfq_io_cq *cic; |
da775265 | 2470 | struct cfq_queue *cfqq; |
da775265 JA |
2471 | |
2472 | /* | |
ec8acb69 | 2473 | * Disallow merge of a sync bio into an async request. |
da775265 | 2474 | */ |
91fac317 | 2475 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 2476 | return false; |
da775265 JA |
2477 | |
2478 | /* | |
f1a4f4d3 | 2479 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 2480 | * merge only if rq is queued there. |
f1a4f4d3 | 2481 | */ |
07c2bd37 TH |
2482 | cic = cfq_cic_lookup(cfqd, current->io_context); |
2483 | if (!cic) | |
2484 | return false; | |
719d3402 | 2485 | |
91fac317 | 2486 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 2487 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
2488 | } |
2489 | ||
812df48d DS |
2490 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2491 | { | |
2492 | del_timer(&cfqd->idle_slice_timer); | |
155fead9 | 2493 | cfqg_stats_update_idle_time(cfqq->cfqg); |
812df48d DS |
2494 | } |
2495 | ||
febffd61 JA |
2496 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
2497 | struct cfq_queue *cfqq) | |
22e2c507 JA |
2498 | { |
2499 | if (cfqq) { | |
3bf10fea | 2500 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d", |
4d2ceea4 | 2501 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
155fead9 | 2502 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
62a37f6b JT |
2503 | cfqq->slice_start = 0; |
2504 | cfqq->dispatch_start = jiffies; | |
2505 | cfqq->allocated_slice = 0; | |
2506 | cfqq->slice_end = 0; | |
2507 | cfqq->slice_dispatch = 0; | |
2508 | cfqq->nr_sectors = 0; | |
2509 | ||
2510 | cfq_clear_cfqq_wait_request(cfqq); | |
2511 | cfq_clear_cfqq_must_dispatch(cfqq); | |
2512 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
2513 | cfq_clear_cfqq_fifo_expire(cfqq); | |
2514 | cfq_mark_cfqq_slice_new(cfqq); | |
2515 | ||
2516 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
2517 | } |
2518 | ||
2519 | cfqd->active_queue = cfqq; | |
2520 | } | |
2521 | ||
7b14e3b5 JA |
2522 | /* |
2523 | * current cfqq expired its slice (or was too idle), select new one | |
2524 | */ | |
2525 | static void | |
2526 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 2527 | bool timed_out) |
7b14e3b5 | 2528 | { |
7b679138 JA |
2529 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
2530 | ||
7b14e3b5 | 2531 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 2532 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 2533 | |
7b14e3b5 | 2534 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 2535 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 2536 | |
ae54abed SL |
2537 | /* |
2538 | * If this cfqq is shared between multiple processes, check to | |
2539 | * make sure that those processes are still issuing I/Os within | |
2540 | * the mean seek distance. If not, it may be time to break the | |
2541 | * queues apart again. | |
2542 | */ | |
2543 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
2544 | cfq_mark_cfqq_split_coop(cfqq); | |
2545 | ||
7b14e3b5 | 2546 | /* |
6084cdda | 2547 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 2548 | */ |
c553f8e3 SL |
2549 | if (timed_out) { |
2550 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 2551 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
2552 | else |
2553 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
2554 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
2555 | } | |
7b14e3b5 | 2556 | |
e5ff082e | 2557 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 2558 | |
f04a6424 VG |
2559 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
2560 | cfq_del_cfqq_rr(cfqd, cfqq); | |
2561 | ||
edd75ffd | 2562 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
2563 | |
2564 | if (cfqq == cfqd->active_queue) | |
2565 | cfqd->active_queue = NULL; | |
2566 | ||
2567 | if (cfqd->active_cic) { | |
11a3122f | 2568 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
2569 | cfqd->active_cic = NULL; |
2570 | } | |
7b14e3b5 JA |
2571 | } |
2572 | ||
e5ff082e | 2573 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
2574 | { |
2575 | struct cfq_queue *cfqq = cfqd->active_queue; | |
2576 | ||
2577 | if (cfqq) | |
e5ff082e | 2578 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
2579 | } |
2580 | ||
498d3aa2 JA |
2581 | /* |
2582 | * Get next queue for service. Unless we have a queue preemption, | |
2583 | * we'll simply select the first cfqq in the service tree. | |
2584 | */ | |
6d048f53 | 2585 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 2586 | { |
34b98d03 VG |
2587 | struct cfq_rb_root *st = st_for(cfqd->serving_group, |
2588 | cfqd->serving_wl_class, cfqd->serving_wl_type); | |
d9e7620e | 2589 | |
f04a6424 VG |
2590 | if (!cfqd->rq_queued) |
2591 | return NULL; | |
2592 | ||
1fa8f6d6 | 2593 | /* There is nothing to dispatch */ |
34b98d03 | 2594 | if (!st) |
1fa8f6d6 | 2595 | return NULL; |
34b98d03 | 2596 | if (RB_EMPTY_ROOT(&st->rb)) |
c0324a02 | 2597 | return NULL; |
34b98d03 | 2598 | return cfq_rb_first(st); |
6d048f53 JA |
2599 | } |
2600 | ||
f04a6424 VG |
2601 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
2602 | { | |
25fb5169 | 2603 | struct cfq_group *cfqg; |
f04a6424 VG |
2604 | struct cfq_queue *cfqq; |
2605 | int i, j; | |
2606 | struct cfq_rb_root *st; | |
2607 | ||
2608 | if (!cfqd->rq_queued) | |
2609 | return NULL; | |
2610 | ||
25fb5169 VG |
2611 | cfqg = cfq_get_next_cfqg(cfqd); |
2612 | if (!cfqg) | |
2613 | return NULL; | |
2614 | ||
f04a6424 VG |
2615 | for_each_cfqg_st(cfqg, i, j, st) |
2616 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
2617 | return cfqq; | |
2618 | return NULL; | |
2619 | } | |
2620 | ||
498d3aa2 JA |
2621 | /* |
2622 | * Get and set a new active queue for service. | |
2623 | */ | |
a36e71f9 JA |
2624 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
2625 | struct cfq_queue *cfqq) | |
6d048f53 | 2626 | { |
e00ef799 | 2627 | if (!cfqq) |
a36e71f9 | 2628 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 2629 | |
22e2c507 | 2630 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 2631 | return cfqq; |
22e2c507 JA |
2632 | } |
2633 | ||
d9e7620e JA |
2634 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
2635 | struct request *rq) | |
2636 | { | |
83096ebf TH |
2637 | if (blk_rq_pos(rq) >= cfqd->last_position) |
2638 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 2639 | else |
83096ebf | 2640 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
2641 | } |
2642 | ||
b2c18e1e | 2643 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 2644 | struct request *rq) |
6d048f53 | 2645 | { |
e9ce335d | 2646 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
2647 | } |
2648 | ||
a36e71f9 JA |
2649 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
2650 | struct cfq_queue *cur_cfqq) | |
2651 | { | |
f2d1f0ae | 2652 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
2653 | struct rb_node *parent, *node; |
2654 | struct cfq_queue *__cfqq; | |
2655 | sector_t sector = cfqd->last_position; | |
2656 | ||
2657 | if (RB_EMPTY_ROOT(root)) | |
2658 | return NULL; | |
2659 | ||
2660 | /* | |
2661 | * First, if we find a request starting at the end of the last | |
2662 | * request, choose it. | |
2663 | */ | |
f2d1f0ae | 2664 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
2665 | if (__cfqq) |
2666 | return __cfqq; | |
2667 | ||
2668 | /* | |
2669 | * If the exact sector wasn't found, the parent of the NULL leaf | |
2670 | * will contain the closest sector. | |
2671 | */ | |
2672 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 2673 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2674 | return __cfqq; |
2675 | ||
2e46e8b2 | 2676 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
2677 | node = rb_next(&__cfqq->p_node); |
2678 | else | |
2679 | node = rb_prev(&__cfqq->p_node); | |
2680 | if (!node) | |
2681 | return NULL; | |
2682 | ||
2683 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 2684 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2685 | return __cfqq; |
2686 | ||
2687 | return NULL; | |
2688 | } | |
2689 | ||
2690 | /* | |
2691 | * cfqd - obvious | |
2692 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
2693 | * closely cooperating with itself. | |
2694 | * | |
2695 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
2696 | * one request, and that cfqd->last_position reflects a position on the disk | |
2697 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
2698 | * assumption. | |
2699 | */ | |
2700 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 2701 | struct cfq_queue *cur_cfqq) |
6d048f53 | 2702 | { |
a36e71f9 JA |
2703 | struct cfq_queue *cfqq; |
2704 | ||
39c01b21 DS |
2705 | if (cfq_class_idle(cur_cfqq)) |
2706 | return NULL; | |
e6c5bc73 JM |
2707 | if (!cfq_cfqq_sync(cur_cfqq)) |
2708 | return NULL; | |
2709 | if (CFQQ_SEEKY(cur_cfqq)) | |
2710 | return NULL; | |
2711 | ||
b9d8f4c7 GJ |
2712 | /* |
2713 | * Don't search priority tree if it's the only queue in the group. | |
2714 | */ | |
2715 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
2716 | return NULL; | |
2717 | ||
6d048f53 | 2718 | /* |
d9e7620e JA |
2719 | * We should notice if some of the queues are cooperating, eg |
2720 | * working closely on the same area of the disk. In that case, | |
2721 | * we can group them together and don't waste time idling. | |
6d048f53 | 2722 | */ |
a36e71f9 JA |
2723 | cfqq = cfqq_close(cfqd, cur_cfqq); |
2724 | if (!cfqq) | |
2725 | return NULL; | |
2726 | ||
8682e1f1 VG |
2727 | /* If new queue belongs to different cfq_group, don't choose it */ |
2728 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
2729 | return NULL; | |
2730 | ||
df5fe3e8 JM |
2731 | /* |
2732 | * It only makes sense to merge sync queues. | |
2733 | */ | |
2734 | if (!cfq_cfqq_sync(cfqq)) | |
2735 | return NULL; | |
e6c5bc73 JM |
2736 | if (CFQQ_SEEKY(cfqq)) |
2737 | return NULL; | |
df5fe3e8 | 2738 | |
c0324a02 CZ |
2739 | /* |
2740 | * Do not merge queues of different priority classes | |
2741 | */ | |
2742 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
2743 | return NULL; | |
2744 | ||
a36e71f9 | 2745 | return cfqq; |
6d048f53 JA |
2746 | } |
2747 | ||
a6d44e98 CZ |
2748 | /* |
2749 | * Determine whether we should enforce idle window for this queue. | |
2750 | */ | |
2751 | ||
2752 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2753 | { | |
3bf10fea | 2754 | enum wl_class_t wl_class = cfqq_class(cfqq); |
34b98d03 | 2755 | struct cfq_rb_root *st = cfqq->service_tree; |
a6d44e98 | 2756 | |
34b98d03 VG |
2757 | BUG_ON(!st); |
2758 | BUG_ON(!st->count); | |
f04a6424 | 2759 | |
b6508c16 VG |
2760 | if (!cfqd->cfq_slice_idle) |
2761 | return false; | |
2762 | ||
a6d44e98 | 2763 | /* We never do for idle class queues. */ |
3bf10fea | 2764 | if (wl_class == IDLE_WORKLOAD) |
a6d44e98 CZ |
2765 | return false; |
2766 | ||
2767 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
2768 | if (cfq_cfqq_idle_window(cfqq) && |
2769 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
2770 | return true; |
2771 | ||
2772 | /* | |
2773 | * Otherwise, we do only if they are the last ones | |
2774 | * in their service tree. | |
2775 | */ | |
34b98d03 VG |
2776 | if (st->count == 1 && cfq_cfqq_sync(cfqq) && |
2777 | !cfq_io_thinktime_big(cfqd, &st->ttime, false)) | |
c1e44756 | 2778 | return true; |
34b98d03 | 2779 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count); |
c1e44756 | 2780 | return false; |
a6d44e98 CZ |
2781 | } |
2782 | ||
6d048f53 | 2783 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2784 | { |
1792669c | 2785 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 2786 | struct cfq_io_cq *cic; |
80bdf0c7 | 2787 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 2788 | |
a68bbddb | 2789 | /* |
f7d7b7a7 JA |
2790 | * SSD device without seek penalty, disable idling. But only do so |
2791 | * for devices that support queuing, otherwise we still have a problem | |
2792 | * with sync vs async workloads. | |
a68bbddb | 2793 | */ |
f7d7b7a7 | 2794 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
2795 | return; |
2796 | ||
dd67d051 | 2797 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2798 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2799 | |
2800 | /* | |
2801 | * idle is disabled, either manually or by past process history | |
2802 | */ | |
80bdf0c7 VG |
2803 | if (!cfq_should_idle(cfqd, cfqq)) { |
2804 | /* no queue idling. Check for group idling */ | |
2805 | if (cfqd->cfq_group_idle) | |
2806 | group_idle = cfqd->cfq_group_idle; | |
2807 | else | |
2808 | return; | |
2809 | } | |
6d048f53 | 2810 | |
7b679138 | 2811 | /* |
8e550632 | 2812 | * still active requests from this queue, don't idle |
7b679138 | 2813 | */ |
8e550632 | 2814 | if (cfqq->dispatched) |
7b679138 JA |
2815 | return; |
2816 | ||
22e2c507 JA |
2817 | /* |
2818 | * task has exited, don't wait | |
2819 | */ | |
206dc69b | 2820 | cic = cfqd->active_cic; |
f6e8d01b | 2821 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
2822 | return; |
2823 | ||
355b659c CZ |
2824 | /* |
2825 | * If our average think time is larger than the remaining time | |
2826 | * slice, then don't idle. This avoids overrunning the allotted | |
2827 | * time slice. | |
2828 | */ | |
383cd721 SL |
2829 | if (sample_valid(cic->ttime.ttime_samples) && |
2830 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 2831 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 2832 | cic->ttime.ttime_mean); |
355b659c | 2833 | return; |
b1ffe737 | 2834 | } |
355b659c | 2835 | |
80bdf0c7 VG |
2836 | /* There are other queues in the group, don't do group idle */ |
2837 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
2838 | return; | |
2839 | ||
3b18152c | 2840 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2841 | |
80bdf0c7 VG |
2842 | if (group_idle) |
2843 | sl = cfqd->cfq_group_idle; | |
2844 | else | |
2845 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2846 | |
7b14e3b5 | 2847 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
155fead9 | 2848 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
80bdf0c7 VG |
2849 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
2850 | group_idle ? 1 : 0); | |
1da177e4 LT |
2851 | } |
2852 | ||
498d3aa2 JA |
2853 | /* |
2854 | * Move request from internal lists to the request queue dispatch list. | |
2855 | */ | |
165125e1 | 2856 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2857 | { |
3ed9a296 | 2858 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2859 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2860 | |
7b679138 JA |
2861 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2862 | ||
06d21886 | 2863 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2864 | cfq_remove_request(rq); |
6d048f53 | 2865 | cfqq->dispatched++; |
80bdf0c7 | 2866 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2867 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2868 | |
53c583d2 | 2869 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2870 | cfqq->nr_sectors += blk_rq_sectors(rq); |
155fead9 | 2871 | cfqg_stats_update_dispatch(cfqq->cfqg, blk_rq_bytes(rq), rq->cmd_flags); |
1da177e4 LT |
2872 | } |
2873 | ||
2874 | /* | |
2875 | * return expired entry, or NULL to just start from scratch in rbtree | |
2876 | */ | |
febffd61 | 2877 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 2878 | { |
30996f40 | 2879 | struct request *rq = NULL; |
1da177e4 | 2880 | |
3b18152c | 2881 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 2882 | return NULL; |
cb887411 JA |
2883 | |
2884 | cfq_mark_cfqq_fifo_expire(cfqq); | |
2885 | ||
89850f7e JA |
2886 | if (list_empty(&cfqq->fifo)) |
2887 | return NULL; | |
1da177e4 | 2888 | |
89850f7e | 2889 | rq = rq_entry_fifo(cfqq->fifo.next); |
8b4922d3 | 2890 | if (time_before(jiffies, rq->fifo_time)) |
7b679138 | 2891 | rq = NULL; |
1da177e4 | 2892 | |
30996f40 | 2893 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 2894 | return rq; |
1da177e4 LT |
2895 | } |
2896 | ||
22e2c507 JA |
2897 | static inline int |
2898 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2899 | { | |
2900 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2901 | |
22e2c507 | 2902 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2903 | |
b9f8ce05 | 2904 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
2905 | } |
2906 | ||
df5fe3e8 JM |
2907 | /* |
2908 | * Must be called with the queue_lock held. | |
2909 | */ | |
2910 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2911 | { | |
2912 | int process_refs, io_refs; | |
2913 | ||
2914 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 2915 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
2916 | BUG_ON(process_refs < 0); |
2917 | return process_refs; | |
2918 | } | |
2919 | ||
2920 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2921 | { | |
e6c5bc73 | 2922 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2923 | struct cfq_queue *__cfqq; |
2924 | ||
c10b61f0 JM |
2925 | /* |
2926 | * If there are no process references on the new_cfqq, then it is | |
2927 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2928 | * chain may have dropped their last reference (not just their | |
2929 | * last process reference). | |
2930 | */ | |
2931 | if (!cfqq_process_refs(new_cfqq)) | |
2932 | return; | |
2933 | ||
df5fe3e8 JM |
2934 | /* Avoid a circular list and skip interim queue merges */ |
2935 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2936 | if (__cfqq == cfqq) | |
2937 | return; | |
2938 | new_cfqq = __cfqq; | |
2939 | } | |
2940 | ||
2941 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2942 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2943 | /* |
2944 | * If the process for the cfqq has gone away, there is no | |
2945 | * sense in merging the queues. | |
2946 | */ | |
c10b61f0 | 2947 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2948 | return; |
2949 | ||
e6c5bc73 JM |
2950 | /* |
2951 | * Merge in the direction of the lesser amount of work. | |
2952 | */ | |
e6c5bc73 JM |
2953 | if (new_process_refs >= process_refs) { |
2954 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2955 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2956 | } else { |
2957 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2958 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2959 | } |
df5fe3e8 JM |
2960 | } |
2961 | ||
6d816ec7 | 2962 | static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd, |
3bf10fea | 2963 | struct cfq_group *cfqg, enum wl_class_t wl_class) |
718eee05 CZ |
2964 | { |
2965 | struct cfq_queue *queue; | |
2966 | int i; | |
2967 | bool key_valid = false; | |
2968 | unsigned long lowest_key = 0; | |
2969 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2970 | ||
65b32a57 VG |
2971 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2972 | /* select the one with lowest rb_key */ | |
34b98d03 | 2973 | queue = cfq_rb_first(st_for(cfqg, wl_class, i)); |
718eee05 CZ |
2974 | if (queue && |
2975 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2976 | lowest_key = queue->rb_key; | |
2977 | cur_best = i; | |
2978 | key_valid = true; | |
2979 | } | |
2980 | } | |
2981 | ||
2982 | return cur_best; | |
2983 | } | |
2984 | ||
6d816ec7 VG |
2985 | static void |
2986 | choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
718eee05 | 2987 | { |
718eee05 CZ |
2988 | unsigned slice; |
2989 | unsigned count; | |
cdb16e8f | 2990 | struct cfq_rb_root *st; |
58ff82f3 | 2991 | unsigned group_slice; |
4d2ceea4 | 2992 | enum wl_class_t original_class = cfqd->serving_wl_class; |
1fa8f6d6 | 2993 | |
718eee05 | 2994 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2995 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 2996 | cfqd->serving_wl_class = RT_WORKLOAD; |
58ff82f3 | 2997 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 2998 | cfqd->serving_wl_class = BE_WORKLOAD; |
718eee05 | 2999 | else { |
4d2ceea4 | 3000 | cfqd->serving_wl_class = IDLE_WORKLOAD; |
718eee05 CZ |
3001 | cfqd->workload_expires = jiffies + 1; |
3002 | return; | |
3003 | } | |
3004 | ||
4d2ceea4 | 3005 | if (original_class != cfqd->serving_wl_class) |
e4ea0c16 SL |
3006 | goto new_workload; |
3007 | ||
718eee05 CZ |
3008 | /* |
3009 | * For RT and BE, we have to choose also the type | |
3010 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
3011 | * expiration time | |
3012 | */ | |
34b98d03 | 3013 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 3014 | count = st->count; |
718eee05 CZ |
3015 | |
3016 | /* | |
65b32a57 | 3017 | * check workload expiration, and that we still have other queues ready |
718eee05 | 3018 | */ |
65b32a57 | 3019 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
3020 | return; |
3021 | ||
e4ea0c16 | 3022 | new_workload: |
718eee05 | 3023 | /* otherwise select new workload type */ |
6d816ec7 | 3024 | cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg, |
4d2ceea4 | 3025 | cfqd->serving_wl_class); |
34b98d03 | 3026 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 3027 | count = st->count; |
718eee05 CZ |
3028 | |
3029 | /* | |
3030 | * the workload slice is computed as a fraction of target latency | |
3031 | * proportional to the number of queues in that workload, over | |
3032 | * all the queues in the same priority class | |
3033 | */ | |
58ff82f3 VG |
3034 | group_slice = cfq_group_slice(cfqd, cfqg); |
3035 | ||
3036 | slice = group_slice * count / | |
4d2ceea4 VG |
3037 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class], |
3038 | cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd, | |
3bf10fea | 3039 | cfqg)); |
718eee05 | 3040 | |
4d2ceea4 | 3041 | if (cfqd->serving_wl_type == ASYNC_WORKLOAD) { |
f26bd1f0 VG |
3042 | unsigned int tmp; |
3043 | ||
3044 | /* | |
3045 | * Async queues are currently system wide. Just taking | |
3046 | * proportion of queues with-in same group will lead to higher | |
3047 | * async ratio system wide as generally root group is going | |
3048 | * to have higher weight. A more accurate thing would be to | |
3049 | * calculate system wide asnc/sync ratio. | |
3050 | */ | |
5bf14c07 TM |
3051 | tmp = cfqd->cfq_target_latency * |
3052 | cfqg_busy_async_queues(cfqd, cfqg); | |
f26bd1f0 VG |
3053 | tmp = tmp/cfqd->busy_queues; |
3054 | slice = min_t(unsigned, slice, tmp); | |
3055 | ||
718eee05 CZ |
3056 | /* async workload slice is scaled down according to |
3057 | * the sync/async slice ratio. */ | |
3058 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 3059 | } else |
718eee05 CZ |
3060 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
3061 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
3062 | ||
3063 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 3064 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
3065 | cfqd->workload_expires = jiffies + slice; |
3066 | } | |
3067 | ||
1fa8f6d6 VG |
3068 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
3069 | { | |
3070 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 3071 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
3072 | |
3073 | if (RB_EMPTY_ROOT(&st->rb)) | |
3074 | return NULL; | |
25bc6b07 | 3075 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
3076 | update_min_vdisktime(st); |
3077 | return cfqg; | |
1fa8f6d6 VG |
3078 | } |
3079 | ||
cdb16e8f VG |
3080 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
3081 | { | |
1fa8f6d6 VG |
3082 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
3083 | ||
3084 | cfqd->serving_group = cfqg; | |
dae739eb VG |
3085 | |
3086 | /* Restore the workload type data */ | |
4d2ceea4 VG |
3087 | if (cfqg->saved_wl_slice) { |
3088 | cfqd->workload_expires = jiffies + cfqg->saved_wl_slice; | |
3089 | cfqd->serving_wl_type = cfqg->saved_wl_type; | |
3090 | cfqd->serving_wl_class = cfqg->saved_wl_class; | |
66ae2919 GJ |
3091 | } else |
3092 | cfqd->workload_expires = jiffies - 1; | |
3093 | ||
6d816ec7 | 3094 | choose_wl_class_and_type(cfqd, cfqg); |
cdb16e8f VG |
3095 | } |
3096 | ||
22e2c507 | 3097 | /* |
498d3aa2 JA |
3098 | * Select a queue for service. If we have a current active queue, |
3099 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 3100 | */ |
1b5ed5e1 | 3101 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 3102 | { |
a36e71f9 | 3103 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 3104 | |
22e2c507 JA |
3105 | cfqq = cfqd->active_queue; |
3106 | if (!cfqq) | |
3107 | goto new_queue; | |
1da177e4 | 3108 | |
f04a6424 VG |
3109 | if (!cfqd->rq_queued) |
3110 | return NULL; | |
c244bb50 VG |
3111 | |
3112 | /* | |
3113 | * We were waiting for group to get backlogged. Expire the queue | |
3114 | */ | |
3115 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3116 | goto expire; | |
3117 | ||
22e2c507 | 3118 | /* |
6d048f53 | 3119 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 3120 | */ |
7667aa06 VG |
3121 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
3122 | /* | |
3123 | * If slice had not expired at the completion of last request | |
3124 | * we might not have turned on wait_busy flag. Don't expire | |
3125 | * the queue yet. Allow the group to get backlogged. | |
3126 | * | |
3127 | * The very fact that we have used the slice, that means we | |
3128 | * have been idling all along on this queue and it should be | |
3129 | * ok to wait for this request to complete. | |
3130 | */ | |
82bbbf28 VG |
3131 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
3132 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
3133 | cfqq = NULL; | |
7667aa06 | 3134 | goto keep_queue; |
82bbbf28 | 3135 | } else |
80bdf0c7 | 3136 | goto check_group_idle; |
7667aa06 | 3137 | } |
1da177e4 | 3138 | |
22e2c507 | 3139 | /* |
6d048f53 JA |
3140 | * The active queue has requests and isn't expired, allow it to |
3141 | * dispatch. | |
22e2c507 | 3142 | */ |
dd67d051 | 3143 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3144 | goto keep_queue; |
6d048f53 | 3145 | |
a36e71f9 JA |
3146 | /* |
3147 | * If another queue has a request waiting within our mean seek | |
3148 | * distance, let it run. The expire code will check for close | |
3149 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 3150 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 3151 | */ |
b3b6d040 | 3152 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
3153 | if (new_cfqq) { |
3154 | if (!cfqq->new_cfqq) | |
3155 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 3156 | goto expire; |
df5fe3e8 | 3157 | } |
a36e71f9 | 3158 | |
6d048f53 JA |
3159 | /* |
3160 | * No requests pending. If the active queue still has requests in | |
3161 | * flight or is idling for a new request, allow either of these | |
3162 | * conditions to happen (or time out) before selecting a new queue. | |
3163 | */ | |
80bdf0c7 VG |
3164 | if (timer_pending(&cfqd->idle_slice_timer)) { |
3165 | cfqq = NULL; | |
3166 | goto keep_queue; | |
3167 | } | |
3168 | ||
8e1ac665 SL |
3169 | /* |
3170 | * This is a deep seek queue, but the device is much faster than | |
3171 | * the queue can deliver, don't idle | |
3172 | **/ | |
3173 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
3174 | (cfq_cfqq_slice_new(cfqq) || | |
3175 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
3176 | cfq_clear_cfqq_deep(cfqq); | |
3177 | cfq_clear_cfqq_idle_window(cfqq); | |
3178 | } | |
3179 | ||
80bdf0c7 VG |
3180 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
3181 | cfqq = NULL; | |
3182 | goto keep_queue; | |
3183 | } | |
3184 | ||
3185 | /* | |
3186 | * If group idle is enabled and there are requests dispatched from | |
3187 | * this group, wait for requests to complete. | |
3188 | */ | |
3189 | check_group_idle: | |
7700fc4f SL |
3190 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
3191 | cfqq->cfqg->dispatched && | |
3192 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
3193 | cfqq = NULL; |
3194 | goto keep_queue; | |
22e2c507 JA |
3195 | } |
3196 | ||
3b18152c | 3197 | expire: |
e5ff082e | 3198 | cfq_slice_expired(cfqd, 0); |
3b18152c | 3199 | new_queue: |
718eee05 CZ |
3200 | /* |
3201 | * Current queue expired. Check if we have to switch to a new | |
3202 | * service tree | |
3203 | */ | |
3204 | if (!new_cfqq) | |
cdb16e8f | 3205 | cfq_choose_cfqg(cfqd); |
718eee05 | 3206 | |
a36e71f9 | 3207 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 3208 | keep_queue: |
3b18152c | 3209 | return cfqq; |
22e2c507 JA |
3210 | } |
3211 | ||
febffd61 | 3212 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
3213 | { |
3214 | int dispatched = 0; | |
3215 | ||
3216 | while (cfqq->next_rq) { | |
3217 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
3218 | dispatched++; | |
3219 | } | |
3220 | ||
3221 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
3222 | |
3223 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 3224 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
3225 | return dispatched; |
3226 | } | |
3227 | ||
498d3aa2 JA |
3228 | /* |
3229 | * Drain our current requests. Used for barriers and when switching | |
3230 | * io schedulers on-the-fly. | |
3231 | */ | |
d9e7620e | 3232 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 3233 | { |
0871714e | 3234 | struct cfq_queue *cfqq; |
d9e7620e | 3235 | int dispatched = 0; |
cdb16e8f | 3236 | |
3440c49f | 3237 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 3238 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
3239 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
3240 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 3241 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 3242 | } |
1b5ed5e1 | 3243 | |
1b5ed5e1 TH |
3244 | BUG_ON(cfqd->busy_queues); |
3245 | ||
6923715a | 3246 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
3247 | return dispatched; |
3248 | } | |
3249 | ||
abc3c744 SL |
3250 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
3251 | struct cfq_queue *cfqq) | |
3252 | { | |
3253 | /* the queue hasn't finished any request, can't estimate */ | |
3254 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 3255 | return true; |
abc3c744 SL |
3256 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
3257 | cfqq->slice_end)) | |
c1e44756 | 3258 | return true; |
abc3c744 | 3259 | |
c1e44756 | 3260 | return false; |
abc3c744 SL |
3261 | } |
3262 | ||
0b182d61 | 3263 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 3264 | { |
2f5cb738 | 3265 | unsigned int max_dispatch; |
22e2c507 | 3266 | |
5ad531db JA |
3267 | /* |
3268 | * Drain async requests before we start sync IO | |
3269 | */ | |
53c583d2 | 3270 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 3271 | return false; |
5ad531db | 3272 | |
2f5cb738 JA |
3273 | /* |
3274 | * If this is an async queue and we have sync IO in flight, let it wait | |
3275 | */ | |
53c583d2 | 3276 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 3277 | return false; |
2f5cb738 | 3278 | |
abc3c744 | 3279 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
3280 | if (cfq_class_idle(cfqq)) |
3281 | max_dispatch = 1; | |
b4878f24 | 3282 | |
2f5cb738 JA |
3283 | /* |
3284 | * Does this cfqq already have too much IO in flight? | |
3285 | */ | |
3286 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 3287 | bool promote_sync = false; |
2f5cb738 JA |
3288 | /* |
3289 | * idle queue must always only have a single IO in flight | |
3290 | */ | |
3ed9a296 | 3291 | if (cfq_class_idle(cfqq)) |
0b182d61 | 3292 | return false; |
3ed9a296 | 3293 | |
ef8a41df | 3294 | /* |
c4ade94f LS |
3295 | * If there is only one sync queue |
3296 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
3297 | * queue no dispatch limit. The reason is a sync queue can |
3298 | * preempt async queue, limiting the sync queue doesn't make | |
3299 | * sense. This is useful for aiostress test. | |
3300 | */ | |
c4ade94f LS |
3301 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
3302 | promote_sync = true; | |
ef8a41df | 3303 | |
2f5cb738 JA |
3304 | /* |
3305 | * We have other queues, don't allow more IO from this one | |
3306 | */ | |
ef8a41df SL |
3307 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
3308 | !promote_sync) | |
0b182d61 | 3309 | return false; |
9ede209e | 3310 | |
365722bb | 3311 | /* |
474b18cc | 3312 | * Sole queue user, no limit |
365722bb | 3313 | */ |
ef8a41df | 3314 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
3315 | max_dispatch = -1; |
3316 | else | |
3317 | /* | |
3318 | * Normally we start throttling cfqq when cfq_quantum/2 | |
3319 | * requests have been dispatched. But we can drive | |
3320 | * deeper queue depths at the beginning of slice | |
3321 | * subjected to upper limit of cfq_quantum. | |
3322 | * */ | |
3323 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
3324 | } |
3325 | ||
3326 | /* | |
3327 | * Async queues must wait a bit before being allowed dispatch. | |
3328 | * We also ramp up the dispatch depth gradually for async IO, | |
3329 | * based on the last sync IO we serviced | |
3330 | */ | |
963b72fc | 3331 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 3332 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 3333 | unsigned int depth; |
365722bb | 3334 | |
61f0c1dc | 3335 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
3336 | if (!depth && !cfqq->dispatched) |
3337 | depth = 1; | |
8e296755 JA |
3338 | if (depth < max_dispatch) |
3339 | max_dispatch = depth; | |
2f5cb738 | 3340 | } |
3ed9a296 | 3341 | |
0b182d61 JA |
3342 | /* |
3343 | * If we're below the current max, allow a dispatch | |
3344 | */ | |
3345 | return cfqq->dispatched < max_dispatch; | |
3346 | } | |
3347 | ||
3348 | /* | |
3349 | * Dispatch a request from cfqq, moving them to the request queue | |
3350 | * dispatch list. | |
3351 | */ | |
3352 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3353 | { | |
3354 | struct request *rq; | |
3355 | ||
3356 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
3357 | ||
3358 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
3359 | return false; | |
3360 | ||
3361 | /* | |
3362 | * follow expired path, else get first next available | |
3363 | */ | |
3364 | rq = cfq_check_fifo(cfqq); | |
3365 | if (!rq) | |
3366 | rq = cfqq->next_rq; | |
3367 | ||
3368 | /* | |
3369 | * insert request into driver dispatch list | |
3370 | */ | |
3371 | cfq_dispatch_insert(cfqd->queue, rq); | |
3372 | ||
3373 | if (!cfqd->active_cic) { | |
c5869807 | 3374 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 3375 | |
c5869807 | 3376 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
3377 | cfqd->active_cic = cic; |
3378 | } | |
3379 | ||
3380 | return true; | |
3381 | } | |
3382 | ||
3383 | /* | |
3384 | * Find the cfqq that we need to service and move a request from that to the | |
3385 | * dispatch list | |
3386 | */ | |
3387 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
3388 | { | |
3389 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3390 | struct cfq_queue *cfqq; | |
3391 | ||
3392 | if (!cfqd->busy_queues) | |
3393 | return 0; | |
3394 | ||
3395 | if (unlikely(force)) | |
3396 | return cfq_forced_dispatch(cfqd); | |
3397 | ||
3398 | cfqq = cfq_select_queue(cfqd); | |
3399 | if (!cfqq) | |
8e296755 JA |
3400 | return 0; |
3401 | ||
2f5cb738 | 3402 | /* |
0b182d61 | 3403 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 3404 | */ |
0b182d61 JA |
3405 | if (!cfq_dispatch_request(cfqd, cfqq)) |
3406 | return 0; | |
3407 | ||
2f5cb738 | 3408 | cfqq->slice_dispatch++; |
b029195d | 3409 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 3410 | |
2f5cb738 JA |
3411 | /* |
3412 | * expire an async queue immediately if it has used up its slice. idle | |
3413 | * queue always expire after 1 dispatch round. | |
3414 | */ | |
3415 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
3416 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
3417 | cfq_class_idle(cfqq))) { | |
3418 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 3419 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
3420 | } |
3421 | ||
b217a903 | 3422 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 3423 | return 1; |
1da177e4 LT |
3424 | } |
3425 | ||
1da177e4 | 3426 | /* |
5e705374 JA |
3427 | * task holds one reference to the queue, dropped when task exits. each rq |
3428 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 3429 | * |
b1c35769 | 3430 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
3431 | * queue lock must be held here. |
3432 | */ | |
3433 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
3434 | { | |
22e2c507 | 3435 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 3436 | struct cfq_group *cfqg; |
22e2c507 | 3437 | |
30d7b944 | 3438 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 3439 | |
30d7b944 SL |
3440 | cfqq->ref--; |
3441 | if (cfqq->ref) | |
1da177e4 LT |
3442 | return; |
3443 | ||
7b679138 | 3444 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 3445 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 3446 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 3447 | cfqg = cfqq->cfqg; |
1da177e4 | 3448 | |
28f95cbc | 3449 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 3450 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 3451 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 3452 | } |
22e2c507 | 3453 | |
f04a6424 | 3454 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 3455 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 3456 | cfqg_put(cfqg); |
1da177e4 LT |
3457 | } |
3458 | ||
d02a2c07 | 3459 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 3460 | { |
df5fe3e8 JM |
3461 | struct cfq_queue *__cfqq, *next; |
3462 | ||
df5fe3e8 JM |
3463 | /* |
3464 | * If this queue was scheduled to merge with another queue, be | |
3465 | * sure to drop the reference taken on that queue (and others in | |
3466 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
3467 | */ | |
3468 | __cfqq = cfqq->new_cfqq; | |
3469 | while (__cfqq) { | |
3470 | if (__cfqq == cfqq) { | |
3471 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
3472 | break; | |
3473 | } | |
3474 | next = __cfqq->new_cfqq; | |
3475 | cfq_put_queue(__cfqq); | |
3476 | __cfqq = next; | |
3477 | } | |
d02a2c07 SL |
3478 | } |
3479 | ||
3480 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3481 | { | |
3482 | if (unlikely(cfqq == cfqd->active_queue)) { | |
3483 | __cfq_slice_expired(cfqd, cfqq, 0); | |
3484 | cfq_schedule_dispatch(cfqd); | |
3485 | } | |
3486 | ||
3487 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 3488 | |
89850f7e JA |
3489 | cfq_put_queue(cfqq); |
3490 | } | |
22e2c507 | 3491 | |
9b84cacd TH |
3492 | static void cfq_init_icq(struct io_cq *icq) |
3493 | { | |
3494 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
3495 | ||
3496 | cic->ttime.last_end_request = jiffies; | |
3497 | } | |
3498 | ||
c5869807 | 3499 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 3500 | { |
c5869807 | 3501 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 3502 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 3503 | |
563180a4 TH |
3504 | if (cic_to_cfqq(cic, false)) { |
3505 | cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, false)); | |
3506 | cic_set_cfqq(cic, NULL, false); | |
12a05732 AV |
3507 | } |
3508 | ||
563180a4 TH |
3509 | if (cic_to_cfqq(cic, true)) { |
3510 | cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, true)); | |
3511 | cic_set_cfqq(cic, NULL, true); | |
12a05732 | 3512 | } |
89850f7e JA |
3513 | } |
3514 | ||
abede6da | 3515 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
3516 | { |
3517 | struct task_struct *tsk = current; | |
3518 | int ioprio_class; | |
3519 | ||
3b18152c | 3520 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
3521 | return; |
3522 | ||
598971bf | 3523 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 3524 | switch (ioprio_class) { |
fe094d98 JA |
3525 | default: |
3526 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
3527 | case IOPRIO_CLASS_NONE: | |
3528 | /* | |
6d63c275 | 3529 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
3530 | */ |
3531 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 3532 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
3533 | break; |
3534 | case IOPRIO_CLASS_RT: | |
598971bf | 3535 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3536 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
3537 | break; | |
3538 | case IOPRIO_CLASS_BE: | |
598971bf | 3539 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3540 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3541 | break; | |
3542 | case IOPRIO_CLASS_IDLE: | |
3543 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
3544 | cfqq->ioprio = 7; | |
3545 | cfq_clear_cfqq_idle_window(cfqq); | |
3546 | break; | |
22e2c507 JA |
3547 | } |
3548 | ||
3549 | /* | |
3550 | * keep track of original prio settings in case we have to temporarily | |
3551 | * elevate the priority of this queue | |
3552 | */ | |
3553 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 3554 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
3555 | } |
3556 | ||
598971bf | 3557 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 3558 | { |
598971bf | 3559 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 3560 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 3561 | struct cfq_queue *cfqq; |
35e6077c | 3562 | |
598971bf TH |
3563 | /* |
3564 | * Check whether ioprio has changed. The condition may trigger | |
3565 | * spuriously on a newly created cic but there's no harm. | |
3566 | */ | |
3567 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
3568 | return; |
3569 | ||
563180a4 | 3570 | cfqq = cic_to_cfqq(cic, false); |
caaa5f9f | 3571 | if (cfqq) { |
563180a4 | 3572 | cfq_put_queue(cfqq); |
2da8de0b | 3573 | cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio); |
563180a4 | 3574 | cic_set_cfqq(cic, cfqq, false); |
22e2c507 | 3575 | } |
caaa5f9f | 3576 | |
563180a4 | 3577 | cfqq = cic_to_cfqq(cic, true); |
caaa5f9f JA |
3578 | if (cfqq) |
3579 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
3580 | |
3581 | cic->ioprio = ioprio; | |
22e2c507 JA |
3582 | } |
3583 | ||
d5036d77 | 3584 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 3585 | pid_t pid, bool is_sync) |
d5036d77 JA |
3586 | { |
3587 | RB_CLEAR_NODE(&cfqq->rb_node); | |
3588 | RB_CLEAR_NODE(&cfqq->p_node); | |
3589 | INIT_LIST_HEAD(&cfqq->fifo); | |
3590 | ||
30d7b944 | 3591 | cfqq->ref = 0; |
d5036d77 JA |
3592 | cfqq->cfqd = cfqd; |
3593 | ||
3594 | cfq_mark_cfqq_prio_changed(cfqq); | |
3595 | ||
3596 | if (is_sync) { | |
3597 | if (!cfq_class_idle(cfqq)) | |
3598 | cfq_mark_cfqq_idle_window(cfqq); | |
3599 | cfq_mark_cfqq_sync(cfqq); | |
3600 | } | |
3601 | cfqq->pid = pid; | |
3602 | } | |
3603 | ||
24610333 | 3604 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
598971bf | 3605 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 3606 | { |
bca4b914 | 3607 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
60a83707 | 3608 | struct cfq_queue *cfqq; |
f4da8072 | 3609 | uint64_t serial_nr; |
24610333 | 3610 | |
598971bf | 3611 | rcu_read_lock(); |
f4da8072 | 3612 | serial_nr = bio_blkcg(bio)->css.serial_nr; |
598971bf | 3613 | rcu_read_unlock(); |
24610333 | 3614 | |
598971bf TH |
3615 | /* |
3616 | * Check whether blkcg has changed. The condition may trigger | |
3617 | * spuriously on a newly created cic but there's no harm. | |
3618 | */ | |
f4da8072 | 3619 | if (unlikely(!cfqd) || likely(cic->blkcg_serial_nr == serial_nr)) |
598971bf | 3620 | return; |
24610333 | 3621 | |
60a83707 TH |
3622 | /* |
3623 | * Drop reference to queues. New queues will be assigned in new | |
3624 | * group upon arrival of fresh requests. | |
3625 | */ | |
3626 | cfqq = cic_to_cfqq(cic, false); | |
3627 | if (cfqq) { | |
3628 | cfq_log_cfqq(cfqd, cfqq, "changed cgroup"); | |
3629 | cic_set_cfqq(cic, NULL, false); | |
3630 | cfq_put_queue(cfqq); | |
3631 | } | |
3632 | ||
3633 | cfqq = cic_to_cfqq(cic, true); | |
3634 | if (cfqq) { | |
3635 | cfq_log_cfqq(cfqd, cfqq, "changed cgroup"); | |
3636 | cic_set_cfqq(cic, NULL, true); | |
3637 | cfq_put_queue(cfqq); | |
24610333 | 3638 | } |
598971bf | 3639 | |
f4da8072 | 3640 | cic->blkcg_serial_nr = serial_nr; |
24610333 | 3641 | } |
598971bf TH |
3642 | #else |
3643 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } | |
24610333 VG |
3644 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
3645 | ||
c2dea2d1 | 3646 | static struct cfq_queue ** |
60a83707 | 3647 | cfq_async_queue_prio(struct cfq_group *cfqg, int ioprio_class, int ioprio) |
c2dea2d1 | 3648 | { |
fe094d98 | 3649 | switch (ioprio_class) { |
c2dea2d1 | 3650 | case IOPRIO_CLASS_RT: |
60a83707 | 3651 | return &cfqg->async_cfqq[0][ioprio]; |
598971bf TH |
3652 | case IOPRIO_CLASS_NONE: |
3653 | ioprio = IOPRIO_NORM; | |
3654 | /* fall through */ | |
c2dea2d1 | 3655 | case IOPRIO_CLASS_BE: |
60a83707 | 3656 | return &cfqg->async_cfqq[1][ioprio]; |
c2dea2d1 | 3657 | case IOPRIO_CLASS_IDLE: |
60a83707 | 3658 | return &cfqg->async_idle_cfqq; |
c2dea2d1 VT |
3659 | default: |
3660 | BUG(); | |
3661 | } | |
3662 | } | |
3663 | ||
15c31be4 | 3664 | static struct cfq_queue * |
abede6da | 3665 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
2da8de0b | 3666 | struct bio *bio) |
15c31be4 | 3667 | { |
c6ce1943 JM |
3668 | int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
3669 | int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
d4aad7ff | 3670 | struct cfq_queue **async_cfqq = NULL; |
4ebc1c61 | 3671 | struct cfq_queue *cfqq; |
322731ed TH |
3672 | struct cfq_group *cfqg; |
3673 | ||
3674 | rcu_read_lock(); | |
ae118896 | 3675 | cfqg = cfq_lookup_cfqg(cfqd, bio_blkcg(bio)); |
322731ed TH |
3676 | if (!cfqg) { |
3677 | cfqq = &cfqd->oom_cfqq; | |
3678 | goto out; | |
3679 | } | |
15c31be4 | 3680 | |
c2dea2d1 | 3681 | if (!is_sync) { |
c6ce1943 JM |
3682 | if (!ioprio_valid(cic->ioprio)) { |
3683 | struct task_struct *tsk = current; | |
3684 | ioprio = task_nice_ioprio(tsk); | |
3685 | ioprio_class = task_nice_ioclass(tsk); | |
3686 | } | |
60a83707 | 3687 | async_cfqq = cfq_async_queue_prio(cfqg, ioprio_class, ioprio); |
c2dea2d1 | 3688 | cfqq = *async_cfqq; |
4ebc1c61 TH |
3689 | if (cfqq) |
3690 | goto out; | |
c2dea2d1 VT |
3691 | } |
3692 | ||
d4aad7ff TH |
3693 | cfqq = kmem_cache_alloc_node(cfq_pool, GFP_NOWAIT | __GFP_ZERO, |
3694 | cfqd->queue->node); | |
3695 | if (!cfqq) { | |
3696 | cfqq = &cfqd->oom_cfqq; | |
3697 | goto out; | |
3698 | } | |
3699 | ||
3700 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
3701 | cfq_init_prio_data(cfqq, cic); | |
3702 | cfq_link_cfqq_cfqg(cfqq, cfqg); | |
3703 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | |
15c31be4 | 3704 | |
d4aad7ff TH |
3705 | if (async_cfqq) { |
3706 | /* a new async queue is created, pin and remember */ | |
30d7b944 | 3707 | cfqq->ref++; |
c2dea2d1 | 3708 | *async_cfqq = cfqq; |
15c31be4 | 3709 | } |
4ebc1c61 | 3710 | out: |
30d7b944 | 3711 | cfqq->ref++; |
322731ed | 3712 | rcu_read_unlock(); |
15c31be4 JA |
3713 | return cfqq; |
3714 | } | |
3715 | ||
22e2c507 | 3716 | static void |
383cd721 | 3717 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 3718 | { |
383cd721 SL |
3719 | unsigned long elapsed = jiffies - ttime->last_end_request; |
3720 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 3721 | |
383cd721 SL |
3722 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
3723 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
3724 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
3725 | } | |
3726 | ||
3727 | static void | |
3728 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3729 | struct cfq_io_cq *cic) |
383cd721 | 3730 | { |
f5f2b6ce | 3731 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3732 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3733 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3734 | cfqd->cfq_slice_idle); | |
3735 | } | |
7700fc4f SL |
3736 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3737 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3738 | #endif | |
22e2c507 | 3739 | } |
1da177e4 | 3740 | |
206dc69b | 3741 | static void |
b2c18e1e | 3742 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3743 | struct request *rq) |
206dc69b | 3744 | { |
3dde36dd | 3745 | sector_t sdist = 0; |
41647e7a | 3746 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3747 | if (cfqq->last_request_pos) { |
3748 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3749 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3750 | else | |
3751 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3752 | } | |
206dc69b | 3753 | |
3dde36dd | 3754 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3755 | if (blk_queue_nonrot(cfqd->queue)) |
3756 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3757 | else | |
3758 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3759 | } |
1da177e4 | 3760 | |
22e2c507 JA |
3761 | /* |
3762 | * Disable idle window if the process thinks too long or seeks so much that | |
3763 | * it doesn't matter | |
3764 | */ | |
3765 | static void | |
3766 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3767 | struct cfq_io_cq *cic) |
22e2c507 | 3768 | { |
7b679138 | 3769 | int old_idle, enable_idle; |
1be92f2f | 3770 | |
0871714e JA |
3771 | /* |
3772 | * Don't idle for async or idle io prio class | |
3773 | */ | |
3774 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3775 | return; |
3776 | ||
c265a7f4 | 3777 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3778 | |
76280aff CZ |
3779 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3780 | cfq_mark_cfqq_deep(cfqq); | |
3781 | ||
749ef9f8 CZ |
3782 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
3783 | enable_idle = 0; | |
f6e8d01b | 3784 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
3785 | !cfqd->cfq_slice_idle || |
3786 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3787 | enable_idle = 0; |
383cd721 SL |
3788 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3789 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3790 | enable_idle = 0; |
3791 | else | |
3792 | enable_idle = 1; | |
1da177e4 LT |
3793 | } |
3794 | ||
7b679138 JA |
3795 | if (old_idle != enable_idle) { |
3796 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3797 | if (enable_idle) | |
3798 | cfq_mark_cfqq_idle_window(cfqq); | |
3799 | else | |
3800 | cfq_clear_cfqq_idle_window(cfqq); | |
3801 | } | |
22e2c507 | 3802 | } |
1da177e4 | 3803 | |
22e2c507 JA |
3804 | /* |
3805 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3806 | * no or if we aren't sure, a 1 will cause a preempt. | |
3807 | */ | |
a6151c3a | 3808 | static bool |
22e2c507 | 3809 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3810 | struct request *rq) |
22e2c507 | 3811 | { |
6d048f53 | 3812 | struct cfq_queue *cfqq; |
22e2c507 | 3813 | |
6d048f53 JA |
3814 | cfqq = cfqd->active_queue; |
3815 | if (!cfqq) | |
a6151c3a | 3816 | return false; |
22e2c507 | 3817 | |
6d048f53 | 3818 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3819 | return false; |
22e2c507 JA |
3820 | |
3821 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3822 | return true; |
1e3335de | 3823 | |
875feb63 DS |
3824 | /* |
3825 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3826 | */ | |
3827 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3828 | return false; | |
3829 | ||
374f84ac JA |
3830 | /* |
3831 | * if the new request is sync, but the currently running queue is | |
3832 | * not, let the sync request have priority. | |
3833 | */ | |
5e705374 | 3834 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3835 | return true; |
1e3335de | 3836 | |
8682e1f1 VG |
3837 | if (new_cfqq->cfqg != cfqq->cfqg) |
3838 | return false; | |
3839 | ||
3840 | if (cfq_slice_used(cfqq)) | |
3841 | return true; | |
3842 | ||
3843 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
4d2ceea4 | 3844 | if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 VG |
3845 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
3846 | new_cfqq->service_tree->count == 2 && | |
3847 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3848 | return true; | |
3849 | ||
b53d1ed7 JA |
3850 | /* |
3851 | * So both queues are sync. Let the new request get disk time if | |
3852 | * it's a metadata request and the current queue is doing regular IO. | |
3853 | */ | |
65299a3b | 3854 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
3855 | return true; |
3856 | ||
3a9a3f6c DS |
3857 | /* |
3858 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3859 | */ | |
3860 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3861 | return true; |
3a9a3f6c | 3862 | |
d2d59e18 SL |
3863 | /* An idle queue should not be idle now for some reason */ |
3864 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
3865 | return true; | |
3866 | ||
1e3335de | 3867 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3868 | return false; |
1e3335de JA |
3869 | |
3870 | /* | |
3871 | * if this request is as-good as one we would expect from the | |
3872 | * current cfqq, let it preempt | |
3873 | */ | |
e9ce335d | 3874 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3875 | return true; |
1e3335de | 3876 | |
a6151c3a | 3877 | return false; |
22e2c507 JA |
3878 | } |
3879 | ||
3880 | /* | |
3881 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3882 | * let it have half of its nominal slice. | |
3883 | */ | |
3884 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3885 | { | |
df0793ab SL |
3886 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
3887 | ||
7b679138 | 3888 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 3889 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3890 | |
f8ae6e3e SL |
3891 | /* |
3892 | * workload type is changed, don't save slice, otherwise preempt | |
3893 | * doesn't happen | |
3894 | */ | |
df0793ab | 3895 | if (old_type != cfqq_type(cfqq)) |
4d2ceea4 | 3896 | cfqq->cfqg->saved_wl_slice = 0; |
f8ae6e3e | 3897 | |
bf572256 JA |
3898 | /* |
3899 | * Put the new queue at the front of the of the current list, | |
3900 | * so we know that it will be selected next. | |
3901 | */ | |
3902 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3903 | |
3904 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3905 | |
62a37f6b JT |
3906 | cfqq->slice_end = 0; |
3907 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3908 | } |
3909 | ||
22e2c507 | 3910 | /* |
5e705374 | 3911 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3912 | * something we should do about it |
3913 | */ | |
3914 | static void | |
5e705374 JA |
3915 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3916 | struct request *rq) | |
22e2c507 | 3917 | { |
c5869807 | 3918 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3919 | |
45333d5a | 3920 | cfqd->rq_queued++; |
65299a3b CH |
3921 | if (rq->cmd_flags & REQ_PRIO) |
3922 | cfqq->prio_pending++; | |
374f84ac | 3923 | |
383cd721 | 3924 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3925 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3926 | cfq_update_idle_window(cfqd, cfqq, cic); |
3927 | ||
b2c18e1e | 3928 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3929 | |
3930 | if (cfqq == cfqd->active_queue) { | |
3931 | /* | |
b029195d JA |
3932 | * Remember that we saw a request from this process, but |
3933 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3934 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3935 | * and merging. If the request is already larger than a single |
3936 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3937 | * merging is already done. Ditto for a busy system that |
3938 | * has other work pending, don't risk delaying until the | |
3939 | * idle timer unplug to continue working. | |
22e2c507 | 3940 | */ |
d6ceb25e | 3941 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3942 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3943 | cfqd->busy_queues > 1) { | |
812df48d | 3944 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3945 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3946 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3947 | } else { |
155fead9 | 3948 | cfqg_stats_update_idle_time(cfqq->cfqg); |
bf791937 | 3949 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3950 | } |
d6ceb25e | 3951 | } |
5e705374 | 3952 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3953 | /* |
3954 | * not the active queue - expire current slice if it is | |
3955 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3956 | * has some old slice time left and is of higher priority or |
3957 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3958 | */ |
3959 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3960 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3961 | } |
1da177e4 LT |
3962 | } |
3963 | ||
165125e1 | 3964 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3965 | { |
b4878f24 | 3966 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3967 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3968 | |
7b679138 | 3969 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 3970 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 3971 | |
8b4922d3 | 3972 | rq->fifo_time = jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]; |
22e2c507 | 3973 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3974 | cfq_add_rq_rb(rq); |
155fead9 TH |
3975 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
3976 | rq->cmd_flags); | |
5e705374 | 3977 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3978 | } |
3979 | ||
45333d5a AC |
3980 | /* |
3981 | * Update hw_tag based on peak queue depth over 50 samples under | |
3982 | * sufficient load. | |
3983 | */ | |
3984 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3985 | { | |
1a1238a7 SL |
3986 | struct cfq_queue *cfqq = cfqd->active_queue; |
3987 | ||
53c583d2 CZ |
3988 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3989 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3990 | |
3991 | if (cfqd->hw_tag == 1) | |
3992 | return; | |
45333d5a AC |
3993 | |
3994 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3995 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3996 | return; |
3997 | ||
1a1238a7 SL |
3998 | /* |
3999 | * If active queue hasn't enough requests and can idle, cfq might not | |
4000 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
4001 | * case | |
4002 | */ | |
4003 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
4004 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 4005 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
4006 | return; |
4007 | ||
45333d5a AC |
4008 | if (cfqd->hw_tag_samples++ < 50) |
4009 | return; | |
4010 | ||
e459dd08 | 4011 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
4012 | cfqd->hw_tag = 1; |
4013 | else | |
4014 | cfqd->hw_tag = 0; | |
45333d5a AC |
4015 | } |
4016 | ||
7667aa06 VG |
4017 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
4018 | { | |
c5869807 | 4019 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 4020 | |
02a8f01b JT |
4021 | /* If the queue already has requests, don't wait */ |
4022 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
4023 | return false; | |
4024 | ||
7667aa06 VG |
4025 | /* If there are other queues in the group, don't wait */ |
4026 | if (cfqq->cfqg->nr_cfqq > 1) | |
4027 | return false; | |
4028 | ||
7700fc4f SL |
4029 | /* the only queue in the group, but think time is big */ |
4030 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
4031 | return false; | |
4032 | ||
7667aa06 VG |
4033 | if (cfq_slice_used(cfqq)) |
4034 | return true; | |
4035 | ||
4036 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
4037 | if (cic && sample_valid(cic->ttime.ttime_samples) |
4038 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
4039 | return true; |
4040 | ||
4041 | /* | |
4042 | * If think times is less than a jiffy than ttime_mean=0 and above | |
4043 | * will not be true. It might happen that slice has not expired yet | |
4044 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
4045 | * case where think time is less than a jiffy, mark the queue wait | |
4046 | * busy if only 1 jiffy is left in the slice. | |
4047 | */ | |
4048 | if (cfqq->slice_end - jiffies == 1) | |
4049 | return true; | |
4050 | ||
4051 | return false; | |
4052 | } | |
4053 | ||
165125e1 | 4054 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 4055 | { |
5e705374 | 4056 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 4057 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 4058 | const int sync = rq_is_sync(rq); |
b4878f24 | 4059 | unsigned long now; |
1da177e4 | 4060 | |
b4878f24 | 4061 | now = jiffies; |
33659ebb CH |
4062 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
4063 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 4064 | |
45333d5a AC |
4065 | cfq_update_hw_tag(cfqd); |
4066 | ||
53c583d2 | 4067 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 4068 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 4069 | cfqd->rq_in_driver--; |
6d048f53 | 4070 | cfqq->dispatched--; |
80bdf0c7 | 4071 | (RQ_CFQG(rq))->dispatched--; |
155fead9 TH |
4072 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
4073 | rq_io_start_time_ns(rq), rq->cmd_flags); | |
1da177e4 | 4074 | |
53c583d2 | 4075 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 4076 | |
365722bb | 4077 | if (sync) { |
34b98d03 | 4078 | struct cfq_rb_root *st; |
f5f2b6ce | 4079 | |
383cd721 | 4080 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
4081 | |
4082 | if (cfq_cfqq_on_rr(cfqq)) | |
34b98d03 | 4083 | st = cfqq->service_tree; |
f5f2b6ce | 4084 | else |
34b98d03 VG |
4085 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), |
4086 | cfqq_type(cfqq)); | |
4087 | ||
4088 | st->ttime.last_end_request = now; | |
573412b2 CZ |
4089 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
4090 | cfqd->last_delayed_sync = now; | |
365722bb | 4091 | } |
caaa5f9f | 4092 | |
7700fc4f SL |
4093 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4094 | cfqq->cfqg->ttime.last_end_request = now; | |
4095 | #endif | |
4096 | ||
caaa5f9f JA |
4097 | /* |
4098 | * If this is the active queue, check if it needs to be expired, | |
4099 | * or if we want to idle in case it has no pending requests. | |
4100 | */ | |
4101 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
4102 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
4103 | ||
44f7c160 JA |
4104 | if (cfq_cfqq_slice_new(cfqq)) { |
4105 | cfq_set_prio_slice(cfqd, cfqq); | |
4106 | cfq_clear_cfqq_slice_new(cfqq); | |
4107 | } | |
f75edf2d VG |
4108 | |
4109 | /* | |
7667aa06 VG |
4110 | * Should we wait for next request to come in before we expire |
4111 | * the queue. | |
f75edf2d | 4112 | */ |
7667aa06 | 4113 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
4114 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
4115 | if (!cfqd->cfq_slice_idle) | |
4116 | extend_sl = cfqd->cfq_group_idle; | |
4117 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 4118 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 4119 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
4120 | } |
4121 | ||
a36e71f9 | 4122 | /* |
8e550632 CZ |
4123 | * Idling is not enabled on: |
4124 | * - expired queues | |
4125 | * - idle-priority queues | |
4126 | * - async queues | |
4127 | * - queues with still some requests queued | |
4128 | * - when there is a close cooperator | |
a36e71f9 | 4129 | */ |
0871714e | 4130 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 4131 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
4132 | else if (sync && cfqq_empty && |
4133 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 4134 | cfq_arm_slice_timer(cfqd); |
8e550632 | 4135 | } |
caaa5f9f | 4136 | } |
6d048f53 | 4137 | |
53c583d2 | 4138 | if (!cfqd->rq_in_driver) |
23e018a1 | 4139 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
4140 | } |
4141 | ||
89850f7e | 4142 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 4143 | { |
1b379d8d | 4144 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 4145 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 4146 | return ELV_MQUEUE_MUST; |
3b18152c | 4147 | } |
1da177e4 | 4148 | |
22e2c507 | 4149 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
4150 | } |
4151 | ||
165125e1 | 4152 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
4153 | { |
4154 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
4155 | struct task_struct *tsk = current; | |
c5869807 | 4156 | struct cfq_io_cq *cic; |
22e2c507 JA |
4157 | struct cfq_queue *cfqq; |
4158 | ||
4159 | /* | |
4160 | * don't force setup of a queue from here, as a call to may_queue | |
4161 | * does not necessarily imply that a request actually will be queued. | |
4162 | * so just lookup a possibly existing queue, or return 'may queue' | |
4163 | * if that fails | |
4164 | */ | |
4ac845a2 | 4165 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
4166 | if (!cic) |
4167 | return ELV_MQUEUE_MAY; | |
4168 | ||
b0b78f81 | 4169 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 4170 | if (cfqq) { |
abede6da | 4171 | cfq_init_prio_data(cfqq, cic); |
22e2c507 | 4172 | |
89850f7e | 4173 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
4174 | } |
4175 | ||
4176 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
4177 | } |
4178 | ||
1da177e4 LT |
4179 | /* |
4180 | * queue lock held here | |
4181 | */ | |
bb37b94c | 4182 | static void cfq_put_request(struct request *rq) |
1da177e4 | 4183 | { |
5e705374 | 4184 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 4185 | |
5e705374 | 4186 | if (cfqq) { |
22e2c507 | 4187 | const int rw = rq_data_dir(rq); |
1da177e4 | 4188 | |
22e2c507 JA |
4189 | BUG_ON(!cfqq->allocated[rw]); |
4190 | cfqq->allocated[rw]--; | |
1da177e4 | 4191 | |
7f1dc8a2 | 4192 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 4193 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
4194 | rq->elv.priv[0] = NULL; |
4195 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 4196 | |
1da177e4 LT |
4197 | cfq_put_queue(cfqq); |
4198 | } | |
4199 | } | |
4200 | ||
df5fe3e8 | 4201 | static struct cfq_queue * |
c5869807 | 4202 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
4203 | struct cfq_queue *cfqq) |
4204 | { | |
4205 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
4206 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 4207 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
4208 | cfq_put_queue(cfqq); |
4209 | return cic_to_cfqq(cic, 1); | |
4210 | } | |
4211 | ||
e6c5bc73 JM |
4212 | /* |
4213 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
4214 | * was the last process referring to said cfqq. | |
4215 | */ | |
4216 | static struct cfq_queue * | |
c5869807 | 4217 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
4218 | { |
4219 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
4220 | cfqq->pid = current->pid; |
4221 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 4222 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
4223 | return cfqq; |
4224 | } | |
4225 | ||
4226 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
4227 | |
4228 | cfq_put_cooperator(cfqq); | |
4229 | ||
e6c5bc73 JM |
4230 | cfq_put_queue(cfqq); |
4231 | return NULL; | |
4232 | } | |
1da177e4 | 4233 | /* |
22e2c507 | 4234 | * Allocate cfq data structures associated with this request. |
1da177e4 | 4235 | */ |
22e2c507 | 4236 | static int |
852c788f TH |
4237 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
4238 | gfp_t gfp_mask) | |
1da177e4 LT |
4239 | { |
4240 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 4241 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 4242 | const int rw = rq_data_dir(rq); |
a6151c3a | 4243 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 4244 | struct cfq_queue *cfqq; |
1da177e4 | 4245 | |
216284c3 | 4246 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 4247 | |
598971bf TH |
4248 | check_ioprio_changed(cic, bio); |
4249 | check_blkcg_changed(cic, bio); | |
e6c5bc73 | 4250 | new_queue: |
91fac317 | 4251 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 4252 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
bce6133b TH |
4253 | if (cfqq) |
4254 | cfq_put_queue(cfqq); | |
2da8de0b | 4255 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio); |
91fac317 | 4256 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 4257 | } else { |
e6c5bc73 JM |
4258 | /* |
4259 | * If the queue was seeky for too long, break it apart. | |
4260 | */ | |
ae54abed | 4261 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
4262 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
4263 | cfqq = split_cfqq(cic, cfqq); | |
4264 | if (!cfqq) | |
4265 | goto new_queue; | |
4266 | } | |
4267 | ||
df5fe3e8 JM |
4268 | /* |
4269 | * Check to see if this queue is scheduled to merge with | |
4270 | * another, closely cooperating queue. The merging of | |
4271 | * queues happens here as it must be done in process context. | |
4272 | * The reference on new_cfqq was taken in merge_cfqqs. | |
4273 | */ | |
4274 | if (cfqq->new_cfqq) | |
4275 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 4276 | } |
1da177e4 LT |
4277 | |
4278 | cfqq->allocated[rw]++; | |
1da177e4 | 4279 | |
6fae9c25 | 4280 | cfqq->ref++; |
eb7d8c07 | 4281 | cfqg_get(cfqq->cfqg); |
a612fddf | 4282 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 4283 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 4284 | spin_unlock_irq(q->queue_lock); |
5e705374 | 4285 | return 0; |
1da177e4 LT |
4286 | } |
4287 | ||
65f27f38 | 4288 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 4289 | { |
65f27f38 | 4290 | struct cfq_data *cfqd = |
23e018a1 | 4291 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 4292 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4293 | |
40bb54d1 | 4294 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 4295 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 4296 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
4297 | } |
4298 | ||
4299 | /* | |
4300 | * Timer running if the active_queue is currently idling inside its time slice | |
4301 | */ | |
4302 | static void cfq_idle_slice_timer(unsigned long data) | |
4303 | { | |
4304 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
4305 | struct cfq_queue *cfqq; | |
4306 | unsigned long flags; | |
3c6bd2f8 | 4307 | int timed_out = 1; |
22e2c507 | 4308 | |
7b679138 JA |
4309 | cfq_log(cfqd, "idle timer fired"); |
4310 | ||
22e2c507 JA |
4311 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
4312 | ||
fe094d98 JA |
4313 | cfqq = cfqd->active_queue; |
4314 | if (cfqq) { | |
3c6bd2f8 JA |
4315 | timed_out = 0; |
4316 | ||
b029195d JA |
4317 | /* |
4318 | * We saw a request before the queue expired, let it through | |
4319 | */ | |
4320 | if (cfq_cfqq_must_dispatch(cfqq)) | |
4321 | goto out_kick; | |
4322 | ||
22e2c507 JA |
4323 | /* |
4324 | * expired | |
4325 | */ | |
44f7c160 | 4326 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
4327 | goto expire; |
4328 | ||
4329 | /* | |
4330 | * only expire and reinvoke request handler, if there are | |
4331 | * other queues with pending requests | |
4332 | */ | |
caaa5f9f | 4333 | if (!cfqd->busy_queues) |
22e2c507 | 4334 | goto out_cont; |
22e2c507 JA |
4335 | |
4336 | /* | |
4337 | * not expired and it has a request pending, let it dispatch | |
4338 | */ | |
75e50984 | 4339 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 4340 | goto out_kick; |
76280aff CZ |
4341 | |
4342 | /* | |
4343 | * Queue depth flag is reset only when the idle didn't succeed | |
4344 | */ | |
4345 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
4346 | } |
4347 | expire: | |
e5ff082e | 4348 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 4349 | out_kick: |
23e018a1 | 4350 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
4351 | out_cont: |
4352 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
4353 | } | |
4354 | ||
3b18152c JA |
4355 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
4356 | { | |
4357 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 4358 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 4359 | } |
22e2c507 | 4360 | |
b374d18a | 4361 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 4362 | { |
22e2c507 | 4363 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 4364 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4365 | |
3b18152c | 4366 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 4367 | |
d9ff4187 | 4368 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 4369 | |
d9ff4187 | 4370 | if (cfqd->active_queue) |
e5ff082e | 4371 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 4372 | |
03aa264a TH |
4373 | spin_unlock_irq(q->queue_lock); |
4374 | ||
a90d742e AV |
4375 | cfq_shutdown_timer_wq(cfqd); |
4376 | ||
ffea73fc TH |
4377 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4378 | blkcg_deactivate_policy(q, &blkcg_policy_cfq); | |
4379 | #else | |
f51b802c | 4380 | kfree(cfqd->root_group); |
2abae55f | 4381 | #endif |
56edf7d7 | 4382 | kfree(cfqd); |
1da177e4 LT |
4383 | } |
4384 | ||
d50235b7 | 4385 | static int cfq_init_queue(struct request_queue *q, struct elevator_type *e) |
1da177e4 LT |
4386 | { |
4387 | struct cfq_data *cfqd; | |
3c798398 | 4388 | struct blkcg_gq *blkg __maybe_unused; |
a2b1693b | 4389 | int i, ret; |
d50235b7 JM |
4390 | struct elevator_queue *eq; |
4391 | ||
4392 | eq = elevator_alloc(q, e); | |
4393 | if (!eq) | |
4394 | return -ENOMEM; | |
1da177e4 | 4395 | |
c1b511eb | 4396 | cfqd = kzalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node); |
d50235b7 JM |
4397 | if (!cfqd) { |
4398 | kobject_put(&eq->kobj); | |
b2fab5ac | 4399 | return -ENOMEM; |
d50235b7 JM |
4400 | } |
4401 | eq->elevator_data = cfqd; | |
80b15c73 | 4402 | |
f51b802c | 4403 | cfqd->queue = q; |
d50235b7 JM |
4404 | spin_lock_irq(q->queue_lock); |
4405 | q->elevator = eq; | |
4406 | spin_unlock_irq(q->queue_lock); | |
f51b802c | 4407 | |
1fa8f6d6 VG |
4408 | /* Init root service tree */ |
4409 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
4410 | ||
f51b802c | 4411 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 4412 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4413 | ret = blkcg_activate_policy(q, &blkcg_policy_cfq); |
a2b1693b TH |
4414 | if (ret) |
4415 | goto out_free; | |
f51b802c | 4416 | |
a2b1693b | 4417 | cfqd->root_group = blkg_to_cfqg(q->root_blkg); |
f51b802c | 4418 | #else |
a2b1693b | 4419 | ret = -ENOMEM; |
f51b802c TH |
4420 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
4421 | GFP_KERNEL, cfqd->queue->node); | |
a2b1693b TH |
4422 | if (!cfqd->root_group) |
4423 | goto out_free; | |
5624a4e4 | 4424 | |
a2b1693b TH |
4425 | cfq_init_cfqg_base(cfqd->root_group); |
4426 | #endif | |
3381cb8d | 4427 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_DEFAULT; |
e71357e1 | 4428 | cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_DEFAULT; |
5624a4e4 | 4429 | |
26a2ac00 JA |
4430 | /* |
4431 | * Not strictly needed (since RB_ROOT just clears the node and we | |
4432 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
4433 | * to add magic to the rb code | |
4434 | */ | |
4435 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
4436 | cfqd->prio_trees[i] = RB_ROOT; | |
4437 | ||
6118b70b | 4438 | /* |
d4aad7ff | 4439 | * Our fallback cfqq if cfq_get_queue() runs into OOM issues. |
6118b70b | 4440 | * Grab a permanent reference to it, so that the normal code flow |
f51b802c TH |
4441 | * will not attempt to free it. oom_cfqq is linked to root_group |
4442 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
4443 | * the reference from linking right away. | |
6118b70b JA |
4444 | */ |
4445 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 4446 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
4447 | |
4448 | spin_lock_irq(q->queue_lock); | |
f51b802c | 4449 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 4450 | cfqg_put(cfqd->root_group); |
1adaf3dd | 4451 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 4452 | |
22e2c507 JA |
4453 | init_timer(&cfqd->idle_slice_timer); |
4454 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
4455 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
4456 | ||
23e018a1 | 4457 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 4458 | |
1da177e4 | 4459 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
4460 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
4461 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
4462 | cfqd->cfq_back_max = cfq_back_max; |
4463 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
4464 | cfqd->cfq_slice[0] = cfq_slice_async; |
4465 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
5bf14c07 | 4466 | cfqd->cfq_target_latency = cfq_target_latency; |
22e2c507 | 4467 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
0bb97947 | 4468 | cfqd->cfq_slice_idle = cfq_slice_idle; |
80bdf0c7 | 4469 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 4470 | cfqd->cfq_latency = 1; |
e459dd08 | 4471 | cfqd->hw_tag = -1; |
edc71131 CZ |
4472 | /* |
4473 | * we optimistically start assuming sync ops weren't delayed in last | |
4474 | * second, in order to have larger depth for async operations. | |
4475 | */ | |
573412b2 | 4476 | cfqd->last_delayed_sync = jiffies - HZ; |
b2fab5ac | 4477 | return 0; |
a2b1693b TH |
4478 | |
4479 | out_free: | |
4480 | kfree(cfqd); | |
d50235b7 | 4481 | kobject_put(&eq->kobj); |
a2b1693b | 4482 | return ret; |
1da177e4 LT |
4483 | } |
4484 | ||
0bb97947 JA |
4485 | static void cfq_registered_queue(struct request_queue *q) |
4486 | { | |
4487 | struct elevator_queue *e = q->elevator; | |
4488 | struct cfq_data *cfqd = e->elevator_data; | |
4489 | ||
4490 | /* | |
4491 | * Default to IOPS mode with no idling for SSDs | |
4492 | */ | |
4493 | if (blk_queue_nonrot(q)) | |
4494 | cfqd->cfq_slice_idle = 0; | |
4495 | } | |
4496 | ||
1da177e4 LT |
4497 | /* |
4498 | * sysfs parts below --> | |
4499 | */ | |
1da177e4 LT |
4500 | static ssize_t |
4501 | cfq_var_show(unsigned int var, char *page) | |
4502 | { | |
176167ad | 4503 | return sprintf(page, "%u\n", var); |
1da177e4 LT |
4504 | } |
4505 | ||
4506 | static ssize_t | |
4507 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
4508 | { | |
4509 | char *p = (char *) page; | |
4510 | ||
4511 | *var = simple_strtoul(p, &p, 10); | |
4512 | return count; | |
4513 | } | |
4514 | ||
1da177e4 | 4515 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 4516 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 4517 | { \ |
3d1ab40f | 4518 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4519 | unsigned int __data = __VAR; \ |
4520 | if (__CONV) \ | |
4521 | __data = jiffies_to_msecs(__data); \ | |
4522 | return cfq_var_show(__data, (page)); \ | |
4523 | } | |
4524 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
4525 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
4526 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
4527 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
4528 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 4529 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 4530 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
4531 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4532 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4533 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4534 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
5bf14c07 | 4535 | SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1); |
1da177e4 LT |
4536 | #undef SHOW_FUNCTION |
4537 | ||
4538 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 4539 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4540 | { \ |
3d1ab40f | 4541 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4542 | unsigned int __data; \ |
4543 | int ret = cfq_var_store(&__data, (page), count); \ | |
4544 | if (__data < (MIN)) \ | |
4545 | __data = (MIN); \ | |
4546 | else if (__data > (MAX)) \ | |
4547 | __data = (MAX); \ | |
4548 | if (__CONV) \ | |
4549 | *(__PTR) = msecs_to_jiffies(__data); \ | |
4550 | else \ | |
4551 | *(__PTR) = __data; \ | |
4552 | return ret; \ | |
4553 | } | |
4554 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4555 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4556 | UINT_MAX, 1); | |
4557 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4558 | UINT_MAX, 1); | |
e572ec7e | 4559 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4560 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4561 | UINT_MAX, 0); | |
22e2c507 | 4562 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4563 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4564 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4565 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4566 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4567 | UINT_MAX, 0); | |
963b72fc | 4568 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
5bf14c07 | 4569 | STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1); |
1da177e4 LT |
4570 | #undef STORE_FUNCTION |
4571 | ||
e572ec7e AV |
4572 | #define CFQ_ATTR(name) \ |
4573 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4574 | ||
4575 | static struct elv_fs_entry cfq_attrs[] = { | |
4576 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4577 | CFQ_ATTR(fifo_expire_sync), |
4578 | CFQ_ATTR(fifo_expire_async), | |
4579 | CFQ_ATTR(back_seek_max), | |
4580 | CFQ_ATTR(back_seek_penalty), | |
4581 | CFQ_ATTR(slice_sync), | |
4582 | CFQ_ATTR(slice_async), | |
4583 | CFQ_ATTR(slice_async_rq), | |
4584 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 4585 | CFQ_ATTR(group_idle), |
963b72fc | 4586 | CFQ_ATTR(low_latency), |
5bf14c07 | 4587 | CFQ_ATTR(target_latency), |
e572ec7e | 4588 | __ATTR_NULL |
1da177e4 LT |
4589 | }; |
4590 | ||
1da177e4 LT |
4591 | static struct elevator_type iosched_cfq = { |
4592 | .ops = { | |
4593 | .elevator_merge_fn = cfq_merge, | |
4594 | .elevator_merged_fn = cfq_merged_request, | |
4595 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 4596 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 4597 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4598 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4599 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4600 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4601 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4602 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4603 | .elevator_former_req_fn = elv_rb_former_request, |
4604 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 4605 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 4606 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
4607 | .elevator_set_req_fn = cfq_set_request, |
4608 | .elevator_put_req_fn = cfq_put_request, | |
4609 | .elevator_may_queue_fn = cfq_may_queue, | |
4610 | .elevator_init_fn = cfq_init_queue, | |
4611 | .elevator_exit_fn = cfq_exit_queue, | |
0bb97947 | 4612 | .elevator_registered_fn = cfq_registered_queue, |
1da177e4 | 4613 | }, |
3d3c2379 TH |
4614 | .icq_size = sizeof(struct cfq_io_cq), |
4615 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 4616 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 4617 | .elevator_name = "cfq", |
1da177e4 LT |
4618 | .elevator_owner = THIS_MODULE, |
4619 | }; | |
4620 | ||
3e252066 | 4621 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4622 | static struct blkcg_policy blkcg_policy_cfq = { |
f9fcc2d3 TH |
4623 | .cftypes = cfq_blkcg_files, |
4624 | ||
e4a9bde9 | 4625 | .cpd_alloc_fn = cfq_cpd_alloc, |
e48453c3 | 4626 | .cpd_init_fn = cfq_cpd_init, |
e4a9bde9 TH |
4627 | .cpd_free_fn = cfq_cpd_free, |
4628 | ||
001bea73 | 4629 | .pd_alloc_fn = cfq_pd_alloc, |
f9fcc2d3 | 4630 | .pd_init_fn = cfq_pd_init, |
0b39920b | 4631 | .pd_offline_fn = cfq_pd_offline, |
001bea73 | 4632 | .pd_free_fn = cfq_pd_free, |
f9fcc2d3 | 4633 | .pd_reset_stats_fn = cfq_pd_reset_stats, |
3e252066 | 4634 | }; |
3e252066 VG |
4635 | #endif |
4636 | ||
1da177e4 LT |
4637 | static int __init cfq_init(void) |
4638 | { | |
3d3c2379 TH |
4639 | int ret; |
4640 | ||
22e2c507 JA |
4641 | /* |
4642 | * could be 0 on HZ < 1000 setups | |
4643 | */ | |
4644 | if (!cfq_slice_async) | |
4645 | cfq_slice_async = 1; | |
4646 | if (!cfq_slice_idle) | |
4647 | cfq_slice_idle = 1; | |
4648 | ||
80bdf0c7 VG |
4649 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4650 | if (!cfq_group_idle) | |
4651 | cfq_group_idle = 1; | |
8bd435b3 | 4652 | |
3c798398 | 4653 | ret = blkcg_policy_register(&blkcg_policy_cfq); |
8bd435b3 TH |
4654 | if (ret) |
4655 | return ret; | |
ffea73fc TH |
4656 | #else |
4657 | cfq_group_idle = 0; | |
4658 | #endif | |
8bd435b3 | 4659 | |
fd794956 | 4660 | ret = -ENOMEM; |
3d3c2379 TH |
4661 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4662 | if (!cfq_pool) | |
8bd435b3 | 4663 | goto err_pol_unreg; |
1da177e4 | 4664 | |
3d3c2379 | 4665 | ret = elv_register(&iosched_cfq); |
8bd435b3 TH |
4666 | if (ret) |
4667 | goto err_free_pool; | |
3d3c2379 | 4668 | |
2fdd82bd | 4669 | return 0; |
8bd435b3 TH |
4670 | |
4671 | err_free_pool: | |
4672 | kmem_cache_destroy(cfq_pool); | |
4673 | err_pol_unreg: | |
ffea73fc | 4674 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4675 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4676 | #endif |
8bd435b3 | 4677 | return ret; |
1da177e4 LT |
4678 | } |
4679 | ||
4680 | static void __exit cfq_exit(void) | |
4681 | { | |
ffea73fc | 4682 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4683 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4684 | #endif |
1da177e4 | 4685 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4686 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4687 | } |
4688 | ||
4689 | module_init(cfq_init); | |
4690 | module_exit(cfq_exit); | |
4691 | ||
4692 | MODULE_AUTHOR("Jens Axboe"); | |
4693 | MODULE_LICENSE("GPL"); | |
4694 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |