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 | |
155fead9 TH |
180 | /* number of ios merged */ |
181 | struct blkg_rwstat merged; | |
182 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
183 | struct blkg_rwstat service_time; | |
184 | /* total time spent waiting in scheduler queue in ns */ | |
185 | struct blkg_rwstat wait_time; | |
186 | /* number of IOs queued up */ | |
187 | struct blkg_rwstat queued; | |
188 | /* total sectors transferred */ | |
189 | struct blkg_stat sectors; | |
190 | /* total disk time and nr sectors dispatched by this group */ | |
191 | struct blkg_stat time; | |
192 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
193 | /* time not charged to this cgroup */ | |
194 | struct blkg_stat unaccounted_time; | |
195 | /* sum of number of ios queued across all samples */ | |
196 | struct blkg_stat avg_queue_size_sum; | |
197 | /* count of samples taken for average */ | |
198 | struct blkg_stat avg_queue_size_samples; | |
199 | /* how many times this group has been removed from service tree */ | |
200 | struct blkg_stat dequeue; | |
201 | /* total time spent waiting for it to be assigned a timeslice. */ | |
202 | struct blkg_stat group_wait_time; | |
3c798398 | 203 | /* time spent idling for this blkcg_gq */ |
155fead9 TH |
204 | struct blkg_stat idle_time; |
205 | /* total time with empty current active q with other requests queued */ | |
206 | struct blkg_stat empty_time; | |
207 | /* fields after this shouldn't be cleared on stat reset */ | |
208 | uint64_t start_group_wait_time; | |
209 | uint64_t start_idle_time; | |
210 | uint64_t start_empty_time; | |
211 | uint16_t flags; | |
212 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
213 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
214 | }; | |
215 | ||
e48453c3 AA |
216 | /* Per-cgroup data */ |
217 | struct cfq_group_data { | |
218 | /* must be the first member */ | |
81437648 | 219 | struct blkcg_policy_data cpd; |
e48453c3 AA |
220 | |
221 | unsigned int weight; | |
222 | unsigned int leaf_weight; | |
223 | }; | |
224 | ||
cdb16e8f VG |
225 | /* This is per cgroup per device grouping structure */ |
226 | struct cfq_group { | |
f95a04af TH |
227 | /* must be the first member */ |
228 | struct blkg_policy_data pd; | |
229 | ||
1fa8f6d6 VG |
230 | /* group service_tree member */ |
231 | struct rb_node rb_node; | |
232 | ||
233 | /* group service_tree key */ | |
234 | u64 vdisktime; | |
e71357e1 | 235 | |
7918ffb5 TH |
236 | /* |
237 | * The number of active cfqgs and sum of their weights under this | |
238 | * cfqg. This covers this cfqg's leaf_weight and all children's | |
239 | * weights, but does not cover weights of further descendants. | |
240 | * | |
241 | * If a cfqg is on the service tree, it's active. An active cfqg | |
242 | * also activates its parent and contributes to the children_weight | |
243 | * of the parent. | |
244 | */ | |
245 | int nr_active; | |
246 | unsigned int children_weight; | |
247 | ||
1d3650f7 TH |
248 | /* |
249 | * vfraction is the fraction of vdisktime that the tasks in this | |
250 | * cfqg are entitled to. This is determined by compounding the | |
251 | * ratios walking up from this cfqg to the root. | |
252 | * | |
253 | * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all | |
254 | * vfractions on a service tree is approximately 1. The sum may | |
255 | * deviate a bit due to rounding errors and fluctuations caused by | |
256 | * cfqgs entering and leaving the service tree. | |
257 | */ | |
258 | unsigned int vfraction; | |
259 | ||
e71357e1 TH |
260 | /* |
261 | * There are two weights - (internal) weight is the weight of this | |
262 | * cfqg against the sibling cfqgs. leaf_weight is the wight of | |
263 | * this cfqg against the child cfqgs. For the root cfqg, both | |
264 | * weights are kept in sync for backward compatibility. | |
265 | */ | |
25bc6b07 | 266 | unsigned int weight; |
8184f93e | 267 | unsigned int new_weight; |
3381cb8d | 268 | unsigned int dev_weight; |
1fa8f6d6 | 269 | |
e71357e1 TH |
270 | unsigned int leaf_weight; |
271 | unsigned int new_leaf_weight; | |
272 | unsigned int dev_leaf_weight; | |
273 | ||
1fa8f6d6 VG |
274 | /* number of cfqq currently on this group */ |
275 | int nr_cfqq; | |
276 | ||
cdb16e8f | 277 | /* |
4495a7d4 | 278 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
279 | * create the array for each prio class but at run time it is used |
280 | * only for RT and BE class and slot for IDLE class remains unused. | |
281 | * This is primarily done to avoid confusion and a gcc warning. | |
282 | */ | |
283 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
284 | /* | |
285 | * rr lists of queues with requests. We maintain service trees for | |
286 | * RT and BE classes. These trees are subdivided in subclasses | |
287 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
288 | * class there is no subclassification and all the cfq queues go on | |
289 | * a single tree service_tree_idle. | |
cdb16e8f VG |
290 | * Counts are embedded in the cfq_rb_root |
291 | */ | |
292 | struct cfq_rb_root service_trees[2][3]; | |
293 | struct cfq_rb_root service_tree_idle; | |
dae739eb | 294 | |
4d2ceea4 VG |
295 | unsigned long saved_wl_slice; |
296 | enum wl_type_t saved_wl_type; | |
297 | enum wl_class_t saved_wl_class; | |
4eef3049 | 298 | |
80bdf0c7 VG |
299 | /* number of requests that are on the dispatch list or inside driver */ |
300 | int dispatched; | |
7700fc4f | 301 | struct cfq_ttime ttime; |
0b39920b | 302 | struct cfqg_stats stats; /* stats for this cfqg */ |
60a83707 TH |
303 | |
304 | /* async queue for each priority case */ | |
305 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
306 | struct cfq_queue *async_idle_cfqq; | |
307 | ||
cdb16e8f | 308 | }; |
718eee05 | 309 | |
c5869807 TH |
310 | struct cfq_io_cq { |
311 | struct io_cq icq; /* must be the first member */ | |
312 | struct cfq_queue *cfqq[2]; | |
313 | struct cfq_ttime ttime; | |
598971bf TH |
314 | int ioprio; /* the current ioprio */ |
315 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
f4da8072 | 316 | uint64_t blkcg_serial_nr; /* the current blkcg serial */ |
598971bf | 317 | #endif |
c5869807 TH |
318 | }; |
319 | ||
22e2c507 JA |
320 | /* |
321 | * Per block device queue structure | |
322 | */ | |
1da177e4 | 323 | struct cfq_data { |
165125e1 | 324 | struct request_queue *queue; |
1fa8f6d6 VG |
325 | /* Root service tree for cfq_groups */ |
326 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 327 | struct cfq_group *root_group; |
22e2c507 | 328 | |
c0324a02 CZ |
329 | /* |
330 | * The priority currently being served | |
22e2c507 | 331 | */ |
4d2ceea4 VG |
332 | enum wl_class_t serving_wl_class; |
333 | enum wl_type_t serving_wl_type; | |
718eee05 | 334 | unsigned long workload_expires; |
cdb16e8f | 335 | struct cfq_group *serving_group; |
a36e71f9 JA |
336 | |
337 | /* | |
338 | * Each priority tree is sorted by next_request position. These | |
339 | * trees are used when determining if two or more queues are | |
340 | * interleaving requests (see cfq_close_cooperator). | |
341 | */ | |
342 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
343 | ||
22e2c507 | 344 | unsigned int busy_queues; |
ef8a41df | 345 | unsigned int busy_sync_queues; |
22e2c507 | 346 | |
53c583d2 CZ |
347 | int rq_in_driver; |
348 | int rq_in_flight[2]; | |
45333d5a AC |
349 | |
350 | /* | |
351 | * queue-depth detection | |
352 | */ | |
353 | int rq_queued; | |
25776e35 | 354 | int hw_tag; |
e459dd08 CZ |
355 | /* |
356 | * hw_tag can be | |
357 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
358 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
359 | * 0 => no NCQ | |
360 | */ | |
361 | int hw_tag_est_depth; | |
362 | unsigned int hw_tag_samples; | |
1da177e4 | 363 | |
22e2c507 JA |
364 | /* |
365 | * idle window management | |
366 | */ | |
367 | struct timer_list idle_slice_timer; | |
23e018a1 | 368 | struct work_struct unplug_work; |
1da177e4 | 369 | |
22e2c507 | 370 | struct cfq_queue *active_queue; |
c5869807 | 371 | struct cfq_io_cq *active_cic; |
22e2c507 | 372 | |
6d048f53 | 373 | sector_t last_position; |
1da177e4 | 374 | |
1da177e4 LT |
375 | /* |
376 | * tunables, see top of file | |
377 | */ | |
378 | unsigned int cfq_quantum; | |
22e2c507 | 379 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
380 | unsigned int cfq_back_penalty; |
381 | unsigned int cfq_back_max; | |
22e2c507 JA |
382 | unsigned int cfq_slice[2]; |
383 | unsigned int cfq_slice_async_rq; | |
384 | unsigned int cfq_slice_idle; | |
80bdf0c7 | 385 | unsigned int cfq_group_idle; |
963b72fc | 386 | unsigned int cfq_latency; |
5bf14c07 | 387 | unsigned int cfq_target_latency; |
d9ff4187 | 388 | |
6118b70b JA |
389 | /* |
390 | * Fallback dummy cfqq for extreme OOM conditions | |
391 | */ | |
392 | struct cfq_queue oom_cfqq; | |
365722bb | 393 | |
573412b2 | 394 | unsigned long last_delayed_sync; |
1da177e4 LT |
395 | }; |
396 | ||
25fb5169 | 397 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
60a83707 | 398 | static void cfq_put_queue(struct cfq_queue *cfqq); |
25fb5169 | 399 | |
34b98d03 | 400 | static struct cfq_rb_root *st_for(struct cfq_group *cfqg, |
3bf10fea | 401 | enum wl_class_t class, |
65b32a57 | 402 | enum wl_type_t type) |
c0324a02 | 403 | { |
1fa8f6d6 VG |
404 | if (!cfqg) |
405 | return NULL; | |
406 | ||
3bf10fea | 407 | if (class == IDLE_WORKLOAD) |
cdb16e8f | 408 | return &cfqg->service_tree_idle; |
c0324a02 | 409 | |
3bf10fea | 410 | return &cfqg->service_trees[class][type]; |
c0324a02 CZ |
411 | } |
412 | ||
3b18152c | 413 | enum cfqq_state_flags { |
b0b8d749 JA |
414 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
415 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 416 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 417 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
418 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
419 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
420 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 421 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 422 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 423 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 424 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 425 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 426 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
427 | }; |
428 | ||
429 | #define CFQ_CFQQ_FNS(name) \ | |
430 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
431 | { \ | |
fe094d98 | 432 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
433 | } \ |
434 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
435 | { \ | |
fe094d98 | 436 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
437 | } \ |
438 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
439 | { \ | |
fe094d98 | 440 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
441 | } |
442 | ||
443 | CFQ_CFQQ_FNS(on_rr); | |
444 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 445 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 446 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
447 | CFQ_CFQQ_FNS(fifo_expire); |
448 | CFQ_CFQQ_FNS(idle_window); | |
449 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 450 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 451 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 452 | CFQ_CFQQ_FNS(coop); |
ae54abed | 453 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 454 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 455 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
456 | #undef CFQ_CFQQ_FNS |
457 | ||
629ed0b1 | 458 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
2ce4d50f | 459 | |
155fead9 TH |
460 | /* cfqg stats flags */ |
461 | enum cfqg_stats_flags { | |
462 | CFQG_stats_waiting = 0, | |
463 | CFQG_stats_idling, | |
464 | CFQG_stats_empty, | |
629ed0b1 TH |
465 | }; |
466 | ||
155fead9 TH |
467 | #define CFQG_FLAG_FNS(name) \ |
468 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ | |
629ed0b1 | 469 | { \ |
155fead9 | 470 | stats->flags |= (1 << CFQG_stats_##name); \ |
629ed0b1 | 471 | } \ |
155fead9 | 472 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 473 | { \ |
155fead9 | 474 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
629ed0b1 | 475 | } \ |
155fead9 | 476 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 477 | { \ |
155fead9 | 478 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
629ed0b1 TH |
479 | } \ |
480 | ||
155fead9 TH |
481 | CFQG_FLAG_FNS(waiting) |
482 | CFQG_FLAG_FNS(idling) | |
483 | CFQG_FLAG_FNS(empty) | |
484 | #undef CFQG_FLAG_FNS | |
629ed0b1 TH |
485 | |
486 | /* This should be called with the queue_lock held. */ | |
155fead9 | 487 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
629ed0b1 TH |
488 | { |
489 | unsigned long long now; | |
490 | ||
155fead9 | 491 | if (!cfqg_stats_waiting(stats)) |
629ed0b1 TH |
492 | return; |
493 | ||
494 | now = sched_clock(); | |
495 | if (time_after64(now, stats->start_group_wait_time)) | |
496 | blkg_stat_add(&stats->group_wait_time, | |
497 | now - stats->start_group_wait_time); | |
155fead9 | 498 | cfqg_stats_clear_waiting(stats); |
629ed0b1 TH |
499 | } |
500 | ||
501 | /* This should be called with the queue_lock held. */ | |
155fead9 TH |
502 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
503 | struct cfq_group *curr_cfqg) | |
629ed0b1 | 504 | { |
155fead9 | 505 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 506 | |
155fead9 | 507 | if (cfqg_stats_waiting(stats)) |
629ed0b1 | 508 | return; |
155fead9 | 509 | if (cfqg == curr_cfqg) |
629ed0b1 | 510 | return; |
155fead9 TH |
511 | stats->start_group_wait_time = sched_clock(); |
512 | cfqg_stats_mark_waiting(stats); | |
629ed0b1 TH |
513 | } |
514 | ||
515 | /* This should be called with the queue_lock held. */ | |
155fead9 | 516 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
629ed0b1 TH |
517 | { |
518 | unsigned long long now; | |
519 | ||
155fead9 | 520 | if (!cfqg_stats_empty(stats)) |
629ed0b1 TH |
521 | return; |
522 | ||
523 | now = sched_clock(); | |
524 | if (time_after64(now, stats->start_empty_time)) | |
525 | blkg_stat_add(&stats->empty_time, | |
526 | now - stats->start_empty_time); | |
155fead9 | 527 | cfqg_stats_clear_empty(stats); |
629ed0b1 TH |
528 | } |
529 | ||
155fead9 | 530 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
629ed0b1 | 531 | { |
155fead9 | 532 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
629ed0b1 TH |
533 | } |
534 | ||
155fead9 | 535 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
629ed0b1 | 536 | { |
155fead9 | 537 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 538 | |
4d5e80a7 | 539 | if (blkg_rwstat_total(&stats->queued)) |
629ed0b1 TH |
540 | return; |
541 | ||
542 | /* | |
543 | * group is already marked empty. This can happen if cfqq got new | |
544 | * request in parent group and moved to this group while being added | |
545 | * to service tree. Just ignore the event and move on. | |
546 | */ | |
155fead9 | 547 | if (cfqg_stats_empty(stats)) |
629ed0b1 TH |
548 | return; |
549 | ||
550 | stats->start_empty_time = sched_clock(); | |
155fead9 | 551 | cfqg_stats_mark_empty(stats); |
629ed0b1 TH |
552 | } |
553 | ||
155fead9 | 554 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 555 | { |
155fead9 | 556 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 557 | |
155fead9 | 558 | if (cfqg_stats_idling(stats)) { |
629ed0b1 TH |
559 | unsigned long long now = sched_clock(); |
560 | ||
561 | if (time_after64(now, stats->start_idle_time)) | |
562 | blkg_stat_add(&stats->idle_time, | |
563 | now - stats->start_idle_time); | |
155fead9 | 564 | cfqg_stats_clear_idling(stats); |
629ed0b1 TH |
565 | } |
566 | } | |
567 | ||
155fead9 | 568 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 569 | { |
155fead9 | 570 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 571 | |
155fead9 | 572 | BUG_ON(cfqg_stats_idling(stats)); |
629ed0b1 TH |
573 | |
574 | stats->start_idle_time = sched_clock(); | |
155fead9 | 575 | cfqg_stats_mark_idling(stats); |
629ed0b1 TH |
576 | } |
577 | ||
155fead9 | 578 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
629ed0b1 | 579 | { |
155fead9 | 580 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
581 | |
582 | blkg_stat_add(&stats->avg_queue_size_sum, | |
4d5e80a7 | 583 | blkg_rwstat_total(&stats->queued)); |
629ed0b1 | 584 | blkg_stat_add(&stats->avg_queue_size_samples, 1); |
155fead9 | 585 | cfqg_stats_update_group_wait_time(stats); |
629ed0b1 TH |
586 | } |
587 | ||
588 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
589 | ||
f48ec1d7 TH |
590 | static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { } |
591 | static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } | |
592 | static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } | |
593 | static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } | |
594 | static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } | |
595 | static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } | |
596 | static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } | |
629ed0b1 TH |
597 | |
598 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
599 | ||
600 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
2ce4d50f | 601 | |
4ceab71b JA |
602 | static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd) |
603 | { | |
604 | return pd ? container_of(pd, struct cfq_group, pd) : NULL; | |
605 | } | |
606 | ||
607 | static struct cfq_group_data | |
608 | *cpd_to_cfqgd(struct blkcg_policy_data *cpd) | |
609 | { | |
81437648 | 610 | return cpd ? container_of(cpd, struct cfq_group_data, cpd) : NULL; |
4ceab71b JA |
611 | } |
612 | ||
613 | static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg) | |
614 | { | |
615 | return pd_to_blkg(&cfqg->pd); | |
616 | } | |
617 | ||
ffea73fc TH |
618 | static struct blkcg_policy blkcg_policy_cfq; |
619 | ||
620 | static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg) | |
621 | { | |
622 | return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq)); | |
623 | } | |
624 | ||
e48453c3 AA |
625 | static struct cfq_group_data *blkcg_to_cfqgd(struct blkcg *blkcg) |
626 | { | |
627 | return cpd_to_cfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_cfq)); | |
628 | } | |
629 | ||
d02f7aa8 | 630 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) |
7918ffb5 | 631 | { |
d02f7aa8 | 632 | struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent; |
7918ffb5 | 633 | |
d02f7aa8 | 634 | return pblkg ? blkg_to_cfqg(pblkg) : NULL; |
7918ffb5 TH |
635 | } |
636 | ||
eb7d8c07 TH |
637 | static inline void cfqg_get(struct cfq_group *cfqg) |
638 | { | |
639 | return blkg_get(cfqg_to_blkg(cfqg)); | |
640 | } | |
641 | ||
642 | static inline void cfqg_put(struct cfq_group *cfqg) | |
643 | { | |
644 | return blkg_put(cfqg_to_blkg(cfqg)); | |
645 | } | |
646 | ||
54e7ed12 TH |
647 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \ |
648 | char __pbuf[128]; \ | |
649 | \ | |
650 | blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \ | |
b226e5c4 VG |
651 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \ |
652 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
653 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
54e7ed12 TH |
654 | __pbuf, ##args); \ |
655 | } while (0) | |
656 | ||
657 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \ | |
658 | char __pbuf[128]; \ | |
659 | \ | |
660 | blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \ | |
661 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \ | |
662 | } while (0) | |
2868ef7b | 663 | |
155fead9 TH |
664 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
665 | struct cfq_group *curr_cfqg, int rw) | |
2ce4d50f | 666 | { |
155fead9 TH |
667 | blkg_rwstat_add(&cfqg->stats.queued, rw, 1); |
668 | cfqg_stats_end_empty_time(&cfqg->stats); | |
669 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); | |
2ce4d50f TH |
670 | } |
671 | ||
155fead9 TH |
672 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
673 | unsigned long time, unsigned long unaccounted_time) | |
2ce4d50f | 674 | { |
155fead9 | 675 | blkg_stat_add(&cfqg->stats.time, time); |
629ed0b1 | 676 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
155fead9 | 677 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
629ed0b1 | 678 | #endif |
2ce4d50f TH |
679 | } |
680 | ||
155fead9 | 681 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) |
2ce4d50f | 682 | { |
155fead9 | 683 | blkg_rwstat_add(&cfqg->stats.queued, rw, -1); |
2ce4d50f TH |
684 | } |
685 | ||
155fead9 | 686 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) |
2ce4d50f | 687 | { |
155fead9 | 688 | blkg_rwstat_add(&cfqg->stats.merged, rw, 1); |
2ce4d50f TH |
689 | } |
690 | ||
155fead9 TH |
691 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, |
692 | uint64_t bytes, int rw) | |
2ce4d50f | 693 | { |
155fead9 | 694 | blkg_stat_add(&cfqg->stats.sectors, bytes >> 9); |
2ce4d50f TH |
695 | } |
696 | ||
155fead9 TH |
697 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
698 | uint64_t start_time, uint64_t io_start_time, int rw) | |
2ce4d50f | 699 | { |
155fead9 | 700 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 701 | unsigned long long now = sched_clock(); |
629ed0b1 TH |
702 | |
703 | if (time_after64(now, io_start_time)) | |
704 | blkg_rwstat_add(&stats->service_time, rw, now - io_start_time); | |
705 | if (time_after64(io_start_time, start_time)) | |
706 | blkg_rwstat_add(&stats->wait_time, rw, | |
707 | io_start_time - start_time); | |
2ce4d50f TH |
708 | } |
709 | ||
689665af TH |
710 | /* @stats = 0 */ |
711 | static void cfqg_stats_reset(struct cfqg_stats *stats) | |
155fead9 | 712 | { |
155fead9 | 713 | /* queued stats shouldn't be cleared */ |
155fead9 TH |
714 | blkg_rwstat_reset(&stats->merged); |
715 | blkg_rwstat_reset(&stats->service_time); | |
716 | blkg_rwstat_reset(&stats->wait_time); | |
717 | blkg_stat_reset(&stats->time); | |
718 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
719 | blkg_stat_reset(&stats->unaccounted_time); | |
720 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
721 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
722 | blkg_stat_reset(&stats->dequeue); | |
723 | blkg_stat_reset(&stats->group_wait_time); | |
724 | blkg_stat_reset(&stats->idle_time); | |
725 | blkg_stat_reset(&stats->empty_time); | |
726 | #endif | |
727 | } | |
728 | ||
0b39920b | 729 | /* @to += @from */ |
e6269c44 | 730 | static void cfqg_stats_add_aux(struct cfqg_stats *to, struct cfqg_stats *from) |
0b39920b TH |
731 | { |
732 | /* queued stats shouldn't be cleared */ | |
e6269c44 TH |
733 | blkg_rwstat_add_aux(&to->merged, &from->merged); |
734 | blkg_rwstat_add_aux(&to->service_time, &from->service_time); | |
735 | blkg_rwstat_add_aux(&to->wait_time, &from->wait_time); | |
736 | blkg_stat_add_aux(&from->time, &from->time); | |
0b39920b | 737 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
e6269c44 TH |
738 | blkg_stat_add_aux(&to->unaccounted_time, &from->unaccounted_time); |
739 | blkg_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum); | |
740 | blkg_stat_add_aux(&to->avg_queue_size_samples, &from->avg_queue_size_samples); | |
741 | blkg_stat_add_aux(&to->dequeue, &from->dequeue); | |
742 | blkg_stat_add_aux(&to->group_wait_time, &from->group_wait_time); | |
743 | blkg_stat_add_aux(&to->idle_time, &from->idle_time); | |
744 | blkg_stat_add_aux(&to->empty_time, &from->empty_time); | |
0b39920b TH |
745 | #endif |
746 | } | |
747 | ||
748 | /* | |
e6269c44 | 749 | * Transfer @cfqg's stats to its parent's aux counts so that the ancestors' |
0b39920b TH |
750 | * recursive stats can still account for the amount used by this cfqg after |
751 | * it's gone. | |
752 | */ | |
753 | static void cfqg_stats_xfer_dead(struct cfq_group *cfqg) | |
754 | { | |
755 | struct cfq_group *parent = cfqg_parent(cfqg); | |
756 | ||
757 | lockdep_assert_held(cfqg_to_blkg(cfqg)->q->queue_lock); | |
758 | ||
759 | if (unlikely(!parent)) | |
760 | return; | |
761 | ||
e6269c44 | 762 | cfqg_stats_add_aux(&parent->stats, &cfqg->stats); |
0b39920b | 763 | cfqg_stats_reset(&cfqg->stats); |
0b39920b TH |
764 | } |
765 | ||
eb7d8c07 TH |
766 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
767 | ||
d02f7aa8 | 768 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) { return NULL; } |
eb7d8c07 TH |
769 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
770 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
771 | ||
7b679138 | 772 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
b226e5c4 VG |
773 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \ |
774 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
775 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
776 | ##args) | |
4495a7d4 | 777 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 | 778 | |
155fead9 TH |
779 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
780 | struct cfq_group *curr_cfqg, int rw) { } | |
781 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, | |
782 | unsigned long time, unsigned long unaccounted_time) { } | |
783 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) { } | |
784 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) { } | |
785 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, | |
786 | uint64_t bytes, int rw) { } | |
787 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, | |
788 | uint64_t start_time, uint64_t io_start_time, int rw) { } | |
2ce4d50f | 789 | |
eb7d8c07 TH |
790 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
791 | ||
7b679138 JA |
792 | #define cfq_log(cfqd, fmt, args...) \ |
793 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
794 | ||
615f0259 VG |
795 | /* Traverses through cfq group service trees */ |
796 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
797 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
798 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
799 | : &cfqg->service_tree_idle; \ | |
800 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
801 | (i == IDLE_WORKLOAD && j == 0); \ | |
802 | j++, st = i < IDLE_WORKLOAD ? \ | |
803 | &cfqg->service_trees[i][j]: NULL) \ | |
804 | ||
f5f2b6ce SL |
805 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
806 | struct cfq_ttime *ttime, bool group_idle) | |
807 | { | |
808 | unsigned long slice; | |
809 | if (!sample_valid(ttime->ttime_samples)) | |
810 | return false; | |
811 | if (group_idle) | |
812 | slice = cfqd->cfq_group_idle; | |
813 | else | |
814 | slice = cfqd->cfq_slice_idle; | |
815 | return ttime->ttime_mean > slice; | |
816 | } | |
615f0259 | 817 | |
02b35081 VG |
818 | static inline bool iops_mode(struct cfq_data *cfqd) |
819 | { | |
820 | /* | |
821 | * If we are not idling on queues and it is a NCQ drive, parallel | |
822 | * execution of requests is on and measuring time is not possible | |
823 | * in most of the cases until and unless we drive shallower queue | |
824 | * depths and that becomes a performance bottleneck. In such cases | |
825 | * switch to start providing fairness in terms of number of IOs. | |
826 | */ | |
827 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
828 | return true; | |
829 | else | |
830 | return false; | |
831 | } | |
832 | ||
3bf10fea | 833 | static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq) |
c0324a02 CZ |
834 | { |
835 | if (cfq_class_idle(cfqq)) | |
836 | return IDLE_WORKLOAD; | |
837 | if (cfq_class_rt(cfqq)) | |
838 | return RT_WORKLOAD; | |
839 | return BE_WORKLOAD; | |
840 | } | |
841 | ||
718eee05 CZ |
842 | |
843 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
844 | { | |
845 | if (!cfq_cfqq_sync(cfqq)) | |
846 | return ASYNC_WORKLOAD; | |
847 | if (!cfq_cfqq_idle_window(cfqq)) | |
848 | return SYNC_NOIDLE_WORKLOAD; | |
849 | return SYNC_WORKLOAD; | |
850 | } | |
851 | ||
3bf10fea | 852 | static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class, |
58ff82f3 VG |
853 | struct cfq_data *cfqd, |
854 | struct cfq_group *cfqg) | |
c0324a02 | 855 | { |
3bf10fea | 856 | if (wl_class == IDLE_WORKLOAD) |
cdb16e8f | 857 | return cfqg->service_tree_idle.count; |
c0324a02 | 858 | |
34b98d03 VG |
859 | return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count + |
860 | cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count + | |
861 | cfqg->service_trees[wl_class][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
862 | } |
863 | ||
f26bd1f0 VG |
864 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
865 | struct cfq_group *cfqg) | |
866 | { | |
34b98d03 VG |
867 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count + |
868 | cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
f26bd1f0 VG |
869 | } |
870 | ||
165125e1 | 871 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 872 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
2da8de0b | 873 | struct cfq_io_cq *cic, struct bio *bio); |
91fac317 | 874 | |
c5869807 TH |
875 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
876 | { | |
877 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
878 | return container_of(icq, struct cfq_io_cq, icq); | |
879 | } | |
880 | ||
47fdd4ca TH |
881 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
882 | struct io_context *ioc) | |
883 | { | |
884 | if (ioc) | |
885 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
886 | return NULL; | |
887 | } | |
888 | ||
c5869807 | 889 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 890 | { |
a6151c3a | 891 | return cic->cfqq[is_sync]; |
91fac317 VT |
892 | } |
893 | ||
c5869807 TH |
894 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
895 | bool is_sync) | |
91fac317 | 896 | { |
a6151c3a | 897 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
898 | } |
899 | ||
c5869807 | 900 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 901 | { |
c5869807 | 902 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
903 | } |
904 | ||
91fac317 VT |
905 | /* |
906 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
907 | * set (in which case it could also be direct WRITE). | |
908 | */ | |
a6151c3a | 909 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 910 | { |
7b6d91da | 911 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 912 | } |
1da177e4 | 913 | |
99f95e52 AM |
914 | /* |
915 | * scheduler run of queue, if there are requests pending and no one in the | |
916 | * driver that will restart queueing | |
917 | */ | |
23e018a1 | 918 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 919 | { |
7b679138 JA |
920 | if (cfqd->busy_queues) { |
921 | cfq_log(cfqd, "schedule dispatch"); | |
59c3d45e | 922 | kblockd_schedule_work(&cfqd->unplug_work); |
7b679138 | 923 | } |
99f95e52 AM |
924 | } |
925 | ||
44f7c160 JA |
926 | /* |
927 | * Scale schedule slice based on io priority. Use the sync time slice only | |
928 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
929 | * io only, should not get full sync slice length. | |
930 | */ | |
a6151c3a | 931 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 932 | unsigned short prio) |
44f7c160 | 933 | { |
d9e7620e | 934 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 935 | |
d9e7620e JA |
936 | WARN_ON(prio >= IOPRIO_BE_NR); |
937 | ||
938 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
939 | } | |
44f7c160 | 940 | |
d9e7620e JA |
941 | static inline int |
942 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
943 | { | |
944 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
945 | } |
946 | ||
1d3650f7 TH |
947 | /** |
948 | * cfqg_scale_charge - scale disk time charge according to cfqg weight | |
949 | * @charge: disk time being charged | |
950 | * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT | |
951 | * | |
952 | * Scale @charge according to @vfraction, which is in range (0, 1]. The | |
953 | * scaling is inversely proportional. | |
954 | * | |
955 | * scaled = charge / vfraction | |
956 | * | |
957 | * The result is also in fixed point w/ CFQ_SERVICE_SHIFT. | |
958 | */ | |
959 | static inline u64 cfqg_scale_charge(unsigned long charge, | |
960 | unsigned int vfraction) | |
25bc6b07 | 961 | { |
1d3650f7 | 962 | u64 c = charge << CFQ_SERVICE_SHIFT; /* make it fixed point */ |
25bc6b07 | 963 | |
1d3650f7 TH |
964 | /* charge / vfraction */ |
965 | c <<= CFQ_SERVICE_SHIFT; | |
966 | do_div(c, vfraction); | |
967 | return c; | |
25bc6b07 VG |
968 | } |
969 | ||
970 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
971 | { | |
972 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
973 | if (delta > 0) | |
974 | min_vdisktime = vdisktime; | |
975 | ||
976 | return min_vdisktime; | |
977 | } | |
978 | ||
979 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
980 | { | |
981 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
982 | if (delta < 0) | |
983 | min_vdisktime = vdisktime; | |
984 | ||
985 | return min_vdisktime; | |
986 | } | |
987 | ||
988 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
989 | { | |
25bc6b07 VG |
990 | struct cfq_group *cfqg; |
991 | ||
25bc6b07 VG |
992 | if (st->left) { |
993 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
994 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
995 | cfqg->vdisktime); | |
25bc6b07 | 996 | } |
25bc6b07 VG |
997 | } |
998 | ||
5db5d642 CZ |
999 | /* |
1000 | * get averaged number of queues of RT/BE priority. | |
1001 | * average is updated, with a formula that gives more weight to higher numbers, | |
1002 | * to quickly follows sudden increases and decrease slowly | |
1003 | */ | |
1004 | ||
58ff82f3 VG |
1005 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
1006 | struct cfq_group *cfqg, bool rt) | |
5869619c | 1007 | { |
5db5d642 CZ |
1008 | unsigned min_q, max_q; |
1009 | unsigned mult = cfq_hist_divisor - 1; | |
1010 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 1011 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 1012 | |
58ff82f3 VG |
1013 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
1014 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
1015 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 1016 | cfq_hist_divisor; |
58ff82f3 VG |
1017 | return cfqg->busy_queues_avg[rt]; |
1018 | } | |
1019 | ||
1020 | static inline unsigned | |
1021 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1022 | { | |
41cad6ab | 1023 | return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT; |
5db5d642 CZ |
1024 | } |
1025 | ||
c553f8e3 | 1026 | static inline unsigned |
ba5bd520 | 1027 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 1028 | { |
5db5d642 CZ |
1029 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
1030 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
1031 | /* |
1032 | * interested queues (we consider only the ones with the same | |
1033 | * priority class in the cfq group) | |
1034 | */ | |
1035 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
1036 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
1037 | unsigned sync_slice = cfqd->cfq_slice[1]; |
1038 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
1039 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
1040 | ||
1041 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
1042 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
1043 | /* scale low_slice according to IO priority | |
1044 | * and sync vs async */ | |
1045 | unsigned low_slice = | |
1046 | min(slice, base_low_slice * slice / sync_slice); | |
1047 | /* the adapted slice value is scaled to fit all iqs | |
1048 | * into the target latency */ | |
58ff82f3 | 1049 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
1050 | low_slice); |
1051 | } | |
1052 | } | |
c553f8e3 SL |
1053 | return slice; |
1054 | } | |
1055 | ||
1056 | static inline void | |
1057 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1058 | { | |
ba5bd520 | 1059 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 1060 | |
dae739eb | 1061 | cfqq->slice_start = jiffies; |
5db5d642 | 1062 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 1063 | cfqq->allocated_slice = slice; |
7b679138 | 1064 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
1065 | } |
1066 | ||
1067 | /* | |
1068 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
1069 | * isn't valid until the first request from the dispatch is activated | |
1070 | * and the slice time set. | |
1071 | */ | |
a6151c3a | 1072 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
1073 | { |
1074 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 1075 | return false; |
44f7c160 | 1076 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 1077 | return false; |
44f7c160 | 1078 | |
c1e44756 | 1079 | return true; |
44f7c160 JA |
1080 | } |
1081 | ||
1da177e4 | 1082 | /* |
5e705374 | 1083 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 1084 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 1085 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 1086 | */ |
5e705374 | 1087 | static struct request * |
cf7c25cf | 1088 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 1089 | { |
cf7c25cf | 1090 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 1091 | unsigned long back_max; |
e8a99053 AM |
1092 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
1093 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
1094 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 1095 | |
5e705374 JA |
1096 | if (rq1 == NULL || rq1 == rq2) |
1097 | return rq2; | |
1098 | if (rq2 == NULL) | |
1099 | return rq1; | |
9c2c38a1 | 1100 | |
229836bd NK |
1101 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
1102 | return rq_is_sync(rq1) ? rq1 : rq2; | |
1103 | ||
65299a3b CH |
1104 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
1105 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 1106 | |
83096ebf TH |
1107 | s1 = blk_rq_pos(rq1); |
1108 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 1109 | |
1da177e4 LT |
1110 | /* |
1111 | * by definition, 1KiB is 2 sectors | |
1112 | */ | |
1113 | back_max = cfqd->cfq_back_max * 2; | |
1114 | ||
1115 | /* | |
1116 | * Strict one way elevator _except_ in the case where we allow | |
1117 | * short backward seeks which are biased as twice the cost of a | |
1118 | * similar forward seek. | |
1119 | */ | |
1120 | if (s1 >= last) | |
1121 | d1 = s1 - last; | |
1122 | else if (s1 + back_max >= last) | |
1123 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
1124 | else | |
e8a99053 | 1125 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
1126 | |
1127 | if (s2 >= last) | |
1128 | d2 = s2 - last; | |
1129 | else if (s2 + back_max >= last) | |
1130 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
1131 | else | |
e8a99053 | 1132 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
1133 | |
1134 | /* Found required data */ | |
e8a99053 AM |
1135 | |
1136 | /* | |
1137 | * By doing switch() on the bit mask "wrap" we avoid having to | |
1138 | * check two variables for all permutations: --> faster! | |
1139 | */ | |
1140 | switch (wrap) { | |
5e705374 | 1141 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 1142 | if (d1 < d2) |
5e705374 | 1143 | return rq1; |
e8a99053 | 1144 | else if (d2 < d1) |
5e705374 | 1145 | return rq2; |
e8a99053 AM |
1146 | else { |
1147 | if (s1 >= s2) | |
5e705374 | 1148 | return rq1; |
e8a99053 | 1149 | else |
5e705374 | 1150 | return rq2; |
e8a99053 | 1151 | } |
1da177e4 | 1152 | |
e8a99053 | 1153 | case CFQ_RQ2_WRAP: |
5e705374 | 1154 | return rq1; |
e8a99053 | 1155 | case CFQ_RQ1_WRAP: |
5e705374 JA |
1156 | return rq2; |
1157 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
1158 | default: |
1159 | /* | |
1160 | * Since both rqs are wrapped, | |
1161 | * start with the one that's further behind head | |
1162 | * (--> only *one* back seek required), | |
1163 | * since back seek takes more time than forward. | |
1164 | */ | |
1165 | if (s1 <= s2) | |
5e705374 | 1166 | return rq1; |
1da177e4 | 1167 | else |
5e705374 | 1168 | return rq2; |
1da177e4 LT |
1169 | } |
1170 | } | |
1171 | ||
498d3aa2 JA |
1172 | /* |
1173 | * The below is leftmost cache rbtree addon | |
1174 | */ | |
0871714e | 1175 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 1176 | { |
615f0259 VG |
1177 | /* Service tree is empty */ |
1178 | if (!root->count) | |
1179 | return NULL; | |
1180 | ||
cc09e299 JA |
1181 | if (!root->left) |
1182 | root->left = rb_first(&root->rb); | |
1183 | ||
0871714e JA |
1184 | if (root->left) |
1185 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
1186 | ||
1187 | return NULL; | |
cc09e299 JA |
1188 | } |
1189 | ||
1fa8f6d6 VG |
1190 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
1191 | { | |
1192 | if (!root->left) | |
1193 | root->left = rb_first(&root->rb); | |
1194 | ||
1195 | if (root->left) | |
1196 | return rb_entry_cfqg(root->left); | |
1197 | ||
1198 | return NULL; | |
1199 | } | |
1200 | ||
a36e71f9 JA |
1201 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
1202 | { | |
1203 | rb_erase(n, root); | |
1204 | RB_CLEAR_NODE(n); | |
1205 | } | |
1206 | ||
cc09e299 JA |
1207 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
1208 | { | |
1209 | if (root->left == n) | |
1210 | root->left = NULL; | |
a36e71f9 | 1211 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 1212 | --root->count; |
cc09e299 JA |
1213 | } |
1214 | ||
1da177e4 LT |
1215 | /* |
1216 | * would be nice to take fifo expire time into account as well | |
1217 | */ | |
5e705374 JA |
1218 | static struct request * |
1219 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1220 | struct request *last) | |
1da177e4 | 1221 | { |
21183b07 JA |
1222 | struct rb_node *rbnext = rb_next(&last->rb_node); |
1223 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 1224 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 1225 | |
21183b07 | 1226 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
1227 | |
1228 | if (rbprev) | |
5e705374 | 1229 | prev = rb_entry_rq(rbprev); |
1da177e4 | 1230 | |
21183b07 | 1231 | if (rbnext) |
5e705374 | 1232 | next = rb_entry_rq(rbnext); |
21183b07 JA |
1233 | else { |
1234 | rbnext = rb_first(&cfqq->sort_list); | |
1235 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 1236 | next = rb_entry_rq(rbnext); |
21183b07 | 1237 | } |
1da177e4 | 1238 | |
cf7c25cf | 1239 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
1240 | } |
1241 | ||
d9e7620e JA |
1242 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
1243 | struct cfq_queue *cfqq) | |
1da177e4 | 1244 | { |
d9e7620e JA |
1245 | /* |
1246 | * just an approximation, should be ok. | |
1247 | */ | |
cdb16e8f | 1248 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 1249 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
1250 | } |
1251 | ||
1fa8f6d6 VG |
1252 | static inline s64 |
1253 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1254 | { | |
1255 | return cfqg->vdisktime - st->min_vdisktime; | |
1256 | } | |
1257 | ||
1258 | static void | |
1259 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1260 | { | |
1261 | struct rb_node **node = &st->rb.rb_node; | |
1262 | struct rb_node *parent = NULL; | |
1263 | struct cfq_group *__cfqg; | |
1264 | s64 key = cfqg_key(st, cfqg); | |
1265 | int left = 1; | |
1266 | ||
1267 | while (*node != NULL) { | |
1268 | parent = *node; | |
1269 | __cfqg = rb_entry_cfqg(parent); | |
1270 | ||
1271 | if (key < cfqg_key(st, __cfqg)) | |
1272 | node = &parent->rb_left; | |
1273 | else { | |
1274 | node = &parent->rb_right; | |
1275 | left = 0; | |
1276 | } | |
1277 | } | |
1278 | ||
1279 | if (left) | |
1280 | st->left = &cfqg->rb_node; | |
1281 | ||
1282 | rb_link_node(&cfqg->rb_node, parent, node); | |
1283 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
1284 | } | |
1285 | ||
7b5af5cf TM |
1286 | /* |
1287 | * This has to be called only on activation of cfqg | |
1288 | */ | |
1fa8f6d6 | 1289 | static void |
8184f93e JT |
1290 | cfq_update_group_weight(struct cfq_group *cfqg) |
1291 | { | |
3381cb8d | 1292 | if (cfqg->new_weight) { |
8184f93e | 1293 | cfqg->weight = cfqg->new_weight; |
3381cb8d | 1294 | cfqg->new_weight = 0; |
8184f93e | 1295 | } |
e15693ef TM |
1296 | } |
1297 | ||
1298 | static void | |
1299 | cfq_update_group_leaf_weight(struct cfq_group *cfqg) | |
1300 | { | |
1301 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
e71357e1 TH |
1302 | |
1303 | if (cfqg->new_leaf_weight) { | |
1304 | cfqg->leaf_weight = cfqg->new_leaf_weight; | |
1305 | cfqg->new_leaf_weight = 0; | |
1306 | } | |
8184f93e JT |
1307 | } |
1308 | ||
1309 | static void | |
1310 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1311 | { | |
1d3650f7 | 1312 | unsigned int vfr = 1 << CFQ_SERVICE_SHIFT; /* start with 1 */ |
7918ffb5 | 1313 | struct cfq_group *pos = cfqg; |
1d3650f7 | 1314 | struct cfq_group *parent; |
7918ffb5 TH |
1315 | bool propagate; |
1316 | ||
1317 | /* add to the service tree */ | |
8184f93e JT |
1318 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
1319 | ||
7b5af5cf TM |
1320 | /* |
1321 | * Update leaf_weight. We cannot update weight at this point | |
1322 | * because cfqg might already have been activated and is | |
1323 | * contributing its current weight to the parent's child_weight. | |
1324 | */ | |
e15693ef | 1325 | cfq_update_group_leaf_weight(cfqg); |
8184f93e | 1326 | __cfq_group_service_tree_add(st, cfqg); |
7918ffb5 TH |
1327 | |
1328 | /* | |
1d3650f7 TH |
1329 | * Activate @cfqg and calculate the portion of vfraction @cfqg is |
1330 | * entitled to. vfraction is calculated by walking the tree | |
1331 | * towards the root calculating the fraction it has at each level. | |
1332 | * The compounded ratio is how much vfraction @cfqg owns. | |
1333 | * | |
1334 | * Start with the proportion tasks in this cfqg has against active | |
1335 | * children cfqgs - its leaf_weight against children_weight. | |
7918ffb5 TH |
1336 | */ |
1337 | propagate = !pos->nr_active++; | |
1338 | pos->children_weight += pos->leaf_weight; | |
1d3650f7 | 1339 | vfr = vfr * pos->leaf_weight / pos->children_weight; |
7918ffb5 | 1340 | |
1d3650f7 TH |
1341 | /* |
1342 | * Compound ->weight walking up the tree. Both activation and | |
1343 | * vfraction calculation are done in the same loop. Propagation | |
1344 | * stops once an already activated node is met. vfraction | |
1345 | * calculation should always continue to the root. | |
1346 | */ | |
d02f7aa8 | 1347 | while ((parent = cfqg_parent(pos))) { |
1d3650f7 | 1348 | if (propagate) { |
e15693ef | 1349 | cfq_update_group_weight(pos); |
1d3650f7 TH |
1350 | propagate = !parent->nr_active++; |
1351 | parent->children_weight += pos->weight; | |
1352 | } | |
1353 | vfr = vfr * pos->weight / parent->children_weight; | |
7918ffb5 TH |
1354 | pos = parent; |
1355 | } | |
1d3650f7 TH |
1356 | |
1357 | cfqg->vfraction = max_t(unsigned, vfr, 1); | |
8184f93e JT |
1358 | } |
1359 | ||
1360 | static void | |
1361 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
1362 | { |
1363 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1364 | struct cfq_group *__cfqg; | |
1365 | struct rb_node *n; | |
1366 | ||
1367 | cfqg->nr_cfqq++; | |
760701bf | 1368 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
1369 | return; |
1370 | ||
1371 | /* | |
1372 | * Currently put the group at the end. Later implement something | |
1373 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 1374 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
1375 | */ |
1376 | n = rb_last(&st->rb); | |
1377 | if (n) { | |
1378 | __cfqg = rb_entry_cfqg(n); | |
1379 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
1380 | } else | |
1381 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
1382 | cfq_group_service_tree_add(st, cfqg); |
1383 | } | |
1fa8f6d6 | 1384 | |
8184f93e JT |
1385 | static void |
1386 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1387 | { | |
7918ffb5 TH |
1388 | struct cfq_group *pos = cfqg; |
1389 | bool propagate; | |
1390 | ||
1391 | /* | |
1392 | * Undo activation from cfq_group_service_tree_add(). Deactivate | |
1393 | * @cfqg and propagate deactivation upwards. | |
1394 | */ | |
1395 | propagate = !--pos->nr_active; | |
1396 | pos->children_weight -= pos->leaf_weight; | |
1397 | ||
1398 | while (propagate) { | |
d02f7aa8 | 1399 | struct cfq_group *parent = cfqg_parent(pos); |
7918ffb5 TH |
1400 | |
1401 | /* @pos has 0 nr_active at this point */ | |
1402 | WARN_ON_ONCE(pos->children_weight); | |
1d3650f7 | 1403 | pos->vfraction = 0; |
7918ffb5 TH |
1404 | |
1405 | if (!parent) | |
1406 | break; | |
1407 | ||
1408 | propagate = !--parent->nr_active; | |
1409 | parent->children_weight -= pos->weight; | |
1410 | pos = parent; | |
1411 | } | |
1412 | ||
1413 | /* remove from the service tree */ | |
8184f93e JT |
1414 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1415 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
1416 | } |
1417 | ||
1418 | static void | |
8184f93e | 1419 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
1420 | { |
1421 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1422 | ||
1423 | BUG_ON(cfqg->nr_cfqq < 1); | |
1424 | cfqg->nr_cfqq--; | |
25bc6b07 | 1425 | |
1fa8f6d6 VG |
1426 | /* If there are other cfq queues under this group, don't delete it */ |
1427 | if (cfqg->nr_cfqq) | |
1428 | return; | |
1429 | ||
2868ef7b | 1430 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 1431 | cfq_group_service_tree_del(st, cfqg); |
4d2ceea4 | 1432 | cfqg->saved_wl_slice = 0; |
155fead9 | 1433 | cfqg_stats_update_dequeue(cfqg); |
dae739eb VG |
1434 | } |
1435 | ||
167400d3 JT |
1436 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
1437 | unsigned int *unaccounted_time) | |
dae739eb | 1438 | { |
f75edf2d | 1439 | unsigned int slice_used; |
dae739eb VG |
1440 | |
1441 | /* | |
1442 | * Queue got expired before even a single request completed or | |
1443 | * got expired immediately after first request completion. | |
1444 | */ | |
1445 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
1446 | /* | |
1447 | * Also charge the seek time incurred to the group, otherwise | |
1448 | * if there are mutiple queues in the group, each can dispatch | |
1449 | * a single request on seeky media and cause lots of seek time | |
1450 | * and group will never know it. | |
1451 | */ | |
1452 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
1453 | 1); | |
1454 | } else { | |
1455 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
1456 | if (slice_used > cfqq->allocated_slice) { |
1457 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 1458 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
1459 | } |
1460 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
1461 | *unaccounted_time += cfqq->slice_start - | |
1462 | cfqq->dispatch_start; | |
dae739eb VG |
1463 | } |
1464 | ||
dae739eb VG |
1465 | return slice_used; |
1466 | } | |
1467 | ||
1468 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1469 | struct cfq_queue *cfqq) |
dae739eb VG |
1470 | { |
1471 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 1472 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1473 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1474 | - cfqg->service_tree_idle.count; | |
1d3650f7 | 1475 | unsigned int vfr; |
f26bd1f0 VG |
1476 | |
1477 | BUG_ON(nr_sync < 0); | |
167400d3 | 1478 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1479 | |
02b35081 VG |
1480 | if (iops_mode(cfqd)) |
1481 | charge = cfqq->slice_dispatch; | |
1482 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1483 | charge = cfqq->allocated_slice; | |
dae739eb | 1484 | |
1d3650f7 TH |
1485 | /* |
1486 | * Can't update vdisktime while on service tree and cfqg->vfraction | |
1487 | * is valid only while on it. Cache vfr, leave the service tree, | |
1488 | * update vdisktime and go back on. The re-addition to the tree | |
1489 | * will also update the weights as necessary. | |
1490 | */ | |
1491 | vfr = cfqg->vfraction; | |
8184f93e | 1492 | cfq_group_service_tree_del(st, cfqg); |
1d3650f7 | 1493 | cfqg->vdisktime += cfqg_scale_charge(charge, vfr); |
8184f93e | 1494 | cfq_group_service_tree_add(st, cfqg); |
dae739eb VG |
1495 | |
1496 | /* This group is being expired. Save the context */ | |
1497 | if (time_after(cfqd->workload_expires, jiffies)) { | |
4d2ceea4 | 1498 | cfqg->saved_wl_slice = cfqd->workload_expires |
dae739eb | 1499 | - jiffies; |
4d2ceea4 VG |
1500 | cfqg->saved_wl_type = cfqd->serving_wl_type; |
1501 | cfqg->saved_wl_class = cfqd->serving_wl_class; | |
dae739eb | 1502 | } else |
4d2ceea4 | 1503 | cfqg->saved_wl_slice = 0; |
2868ef7b VG |
1504 | |
1505 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1506 | st->min_vdisktime); | |
fd16d263 JP |
1507 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1508 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1509 | used_sl, cfqq->slice_dispatch, charge, | |
1510 | iops_mode(cfqd), cfqq->nr_sectors); | |
155fead9 TH |
1511 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
1512 | cfqg_stats_set_start_empty_time(cfqg); | |
1fa8f6d6 VG |
1513 | } |
1514 | ||
f51b802c TH |
1515 | /** |
1516 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1517 | * @cfqg: cfq_group to initialize | |
1518 | * | |
1519 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1520 | * is enabled or not. | |
1521 | */ | |
1522 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1523 | { | |
1524 | struct cfq_rb_root *st; | |
1525 | int i, j; | |
1526 | ||
1527 | for_each_cfqg_st(cfqg, i, j, st) | |
1528 | *st = CFQ_RB_ROOT; | |
1529 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1530 | ||
1531 | cfqg->ttime.last_end_request = jiffies; | |
1532 | } | |
1533 | ||
25fb5169 | 1534 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
24bdb8ef | 1535 | static void cfqg_stats_exit(struct cfqg_stats *stats) |
90d3839b | 1536 | { |
24bdb8ef TH |
1537 | blkg_rwstat_exit(&stats->merged); |
1538 | blkg_rwstat_exit(&stats->service_time); | |
1539 | blkg_rwstat_exit(&stats->wait_time); | |
1540 | blkg_rwstat_exit(&stats->queued); | |
90d3839b | 1541 | |
24bdb8ef TH |
1542 | blkg_stat_exit(&stats->sectors); |
1543 | blkg_stat_exit(&stats->time); | |
1544 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
1545 | blkg_stat_exit(&stats->unaccounted_time); | |
1546 | blkg_stat_exit(&stats->avg_queue_size_sum); | |
1547 | blkg_stat_exit(&stats->avg_queue_size_samples); | |
1548 | blkg_stat_exit(&stats->dequeue); | |
1549 | blkg_stat_exit(&stats->group_wait_time); | |
1550 | blkg_stat_exit(&stats->idle_time); | |
1551 | blkg_stat_exit(&stats->empty_time); | |
1552 | #endif | |
1553 | } | |
1554 | ||
1555 | static int cfqg_stats_init(struct cfqg_stats *stats, gfp_t gfp) | |
1556 | { | |
77ea7338 | 1557 | if (blkg_rwstat_init(&stats->merged, gfp) || |
24bdb8ef TH |
1558 | blkg_rwstat_init(&stats->service_time, gfp) || |
1559 | blkg_rwstat_init(&stats->wait_time, gfp) || | |
1560 | blkg_rwstat_init(&stats->queued, gfp) || | |
1561 | ||
1562 | blkg_stat_init(&stats->sectors, gfp) || | |
1563 | blkg_stat_init(&stats->time, gfp)) | |
1564 | goto err; | |
90d3839b PZ |
1565 | |
1566 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
24bdb8ef TH |
1567 | if (blkg_stat_init(&stats->unaccounted_time, gfp) || |
1568 | blkg_stat_init(&stats->avg_queue_size_sum, gfp) || | |
1569 | blkg_stat_init(&stats->avg_queue_size_samples, gfp) || | |
1570 | blkg_stat_init(&stats->dequeue, gfp) || | |
1571 | blkg_stat_init(&stats->group_wait_time, gfp) || | |
1572 | blkg_stat_init(&stats->idle_time, gfp) || | |
1573 | blkg_stat_init(&stats->empty_time, gfp)) | |
1574 | goto err; | |
90d3839b | 1575 | #endif |
24bdb8ef TH |
1576 | return 0; |
1577 | err: | |
1578 | cfqg_stats_exit(stats); | |
1579 | return -ENOMEM; | |
90d3839b PZ |
1580 | } |
1581 | ||
e4a9bde9 TH |
1582 | static struct blkcg_policy_data *cfq_cpd_alloc(gfp_t gfp) |
1583 | { | |
1584 | struct cfq_group_data *cgd; | |
1585 | ||
1586 | cgd = kzalloc(sizeof(*cgd), GFP_KERNEL); | |
1587 | if (!cgd) | |
1588 | return NULL; | |
1589 | return &cgd->cpd; | |
1590 | } | |
1591 | ||
81437648 | 1592 | static void cfq_cpd_init(struct blkcg_policy_data *cpd) |
e48453c3 | 1593 | { |
81437648 | 1594 | struct cfq_group_data *cgd = cpd_to_cfqgd(cpd); |
e48453c3 | 1595 | |
81437648 | 1596 | if (cpd_to_blkcg(cpd) == &blkcg_root) { |
e48453c3 AA |
1597 | cgd->weight = 2 * CFQ_WEIGHT_DEFAULT; |
1598 | cgd->leaf_weight = 2 * CFQ_WEIGHT_DEFAULT; | |
1599 | } else { | |
1600 | cgd->weight = CFQ_WEIGHT_DEFAULT; | |
1601 | cgd->leaf_weight = CFQ_WEIGHT_DEFAULT; | |
1602 | } | |
1603 | } | |
1604 | ||
e4a9bde9 TH |
1605 | static void cfq_cpd_free(struct blkcg_policy_data *cpd) |
1606 | { | |
1607 | kfree(cpd_to_cfqgd(cpd)); | |
1608 | } | |
1609 | ||
001bea73 TH |
1610 | static struct blkg_policy_data *cfq_pd_alloc(gfp_t gfp, int node) |
1611 | { | |
b2ce2643 TH |
1612 | struct cfq_group *cfqg; |
1613 | ||
1614 | cfqg = kzalloc_node(sizeof(*cfqg), gfp, node); | |
1615 | if (!cfqg) | |
1616 | return NULL; | |
1617 | ||
1618 | cfq_init_cfqg_base(cfqg); | |
24bdb8ef TH |
1619 | if (cfqg_stats_init(&cfqg->stats, gfp)) { |
1620 | kfree(cfqg); | |
1621 | return NULL; | |
1622 | } | |
b2ce2643 TH |
1623 | |
1624 | return &cfqg->pd; | |
001bea73 TH |
1625 | } |
1626 | ||
a9520cd6 | 1627 | static void cfq_pd_init(struct blkg_policy_data *pd) |
f469a7b4 | 1628 | { |
a9520cd6 TH |
1629 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
1630 | struct cfq_group_data *cgd = blkcg_to_cfqgd(pd->blkg->blkcg); | |
25fb5169 | 1631 | |
e48453c3 AA |
1632 | cfqg->weight = cgd->weight; |
1633 | cfqg->leaf_weight = cgd->leaf_weight; | |
25fb5169 VG |
1634 | } |
1635 | ||
a9520cd6 | 1636 | static void cfq_pd_offline(struct blkg_policy_data *pd) |
0b39920b | 1637 | { |
a9520cd6 | 1638 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
60a83707 TH |
1639 | int i; |
1640 | ||
1641 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
1642 | if (cfqg->async_cfqq[0][i]) | |
1643 | cfq_put_queue(cfqg->async_cfqq[0][i]); | |
1644 | if (cfqg->async_cfqq[1][i]) | |
1645 | cfq_put_queue(cfqg->async_cfqq[1][i]); | |
1646 | } | |
1647 | ||
1648 | if (cfqg->async_idle_cfqq) | |
1649 | cfq_put_queue(cfqg->async_idle_cfqq); | |
1650 | ||
0b39920b TH |
1651 | /* |
1652 | * @blkg is going offline and will be ignored by | |
1653 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
1654 | * that they don't get lost. If IOs complete after this point, the | |
1655 | * stats for them will be lost. Oh well... | |
1656 | */ | |
60a83707 | 1657 | cfqg_stats_xfer_dead(cfqg); |
0b39920b TH |
1658 | } |
1659 | ||
001bea73 TH |
1660 | static void cfq_pd_free(struct blkg_policy_data *pd) |
1661 | { | |
24bdb8ef TH |
1662 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
1663 | ||
1664 | cfqg_stats_exit(&cfqg->stats); | |
1665 | return kfree(cfqg); | |
001bea73 TH |
1666 | } |
1667 | ||
a9520cd6 | 1668 | static void cfq_pd_reset_stats(struct blkg_policy_data *pd) |
689665af | 1669 | { |
a9520cd6 | 1670 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
689665af TH |
1671 | |
1672 | cfqg_stats_reset(&cfqg->stats); | |
25fb5169 VG |
1673 | } |
1674 | ||
ae118896 TH |
1675 | static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd, |
1676 | struct blkcg *blkcg) | |
25fb5169 | 1677 | { |
ae118896 | 1678 | struct blkcg_gq *blkg; |
f469a7b4 | 1679 | |
ae118896 TH |
1680 | blkg = blkg_lookup(blkcg, cfqd->queue); |
1681 | if (likely(blkg)) | |
1682 | return blkg_to_cfqg(blkg); | |
1683 | return NULL; | |
25fb5169 VG |
1684 | } |
1685 | ||
1686 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1687 | { | |
25fb5169 | 1688 | cfqq->cfqg = cfqg; |
b1c35769 | 1689 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1690 | cfqg_get(cfqg); |
b1c35769 VG |
1691 | } |
1692 | ||
f95a04af TH |
1693 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, |
1694 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1695 | { |
f95a04af | 1696 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
3381cb8d TH |
1697 | |
1698 | if (!cfqg->dev_weight) | |
60c2bc2d | 1699 | return 0; |
f95a04af | 1700 | return __blkg_prfill_u64(sf, pd, cfqg->dev_weight); |
60c2bc2d TH |
1701 | } |
1702 | ||
2da8ca82 | 1703 | static int cfqg_print_weight_device(struct seq_file *sf, void *v) |
60c2bc2d | 1704 | { |
2da8ca82 TH |
1705 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1706 | cfqg_prfill_weight_device, &blkcg_policy_cfq, | |
1707 | 0, false); | |
60c2bc2d TH |
1708 | return 0; |
1709 | } | |
1710 | ||
e71357e1 TH |
1711 | static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf, |
1712 | struct blkg_policy_data *pd, int off) | |
1713 | { | |
1714 | struct cfq_group *cfqg = pd_to_cfqg(pd); | |
1715 | ||
1716 | if (!cfqg->dev_leaf_weight) | |
1717 | return 0; | |
1718 | return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight); | |
1719 | } | |
1720 | ||
2da8ca82 | 1721 | static int cfqg_print_leaf_weight_device(struct seq_file *sf, void *v) |
e71357e1 | 1722 | { |
2da8ca82 TH |
1723 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1724 | cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq, | |
1725 | 0, false); | |
e71357e1 TH |
1726 | return 0; |
1727 | } | |
1728 | ||
2da8ca82 | 1729 | static int cfq_print_weight(struct seq_file *sf, void *v) |
60c2bc2d | 1730 | { |
e48453c3 | 1731 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
9470e4a6 JA |
1732 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); |
1733 | unsigned int val = 0; | |
e48453c3 | 1734 | |
9470e4a6 JA |
1735 | if (cgd) |
1736 | val = cgd->weight; | |
1737 | ||
1738 | seq_printf(sf, "%u\n", val); | |
60c2bc2d TH |
1739 | return 0; |
1740 | } | |
1741 | ||
2da8ca82 | 1742 | static int cfq_print_leaf_weight(struct seq_file *sf, void *v) |
e71357e1 | 1743 | { |
e48453c3 | 1744 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
9470e4a6 JA |
1745 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); |
1746 | unsigned int val = 0; | |
1747 | ||
1748 | if (cgd) | |
1749 | val = cgd->leaf_weight; | |
e48453c3 | 1750 | |
9470e4a6 | 1751 | seq_printf(sf, "%u\n", val); |
e71357e1 TH |
1752 | return 0; |
1753 | } | |
1754 | ||
451af504 TH |
1755 | static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of, |
1756 | char *buf, size_t nbytes, loff_t off, | |
1757 | bool is_leaf_weight) | |
60c2bc2d | 1758 | { |
451af504 | 1759 | struct blkcg *blkcg = css_to_blkcg(of_css(of)); |
60c2bc2d | 1760 | struct blkg_conf_ctx ctx; |
3381cb8d | 1761 | struct cfq_group *cfqg; |
e48453c3 | 1762 | struct cfq_group_data *cfqgd; |
60c2bc2d TH |
1763 | int ret; |
1764 | ||
3c798398 | 1765 | ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx); |
60c2bc2d TH |
1766 | if (ret) |
1767 | return ret; | |
1768 | ||
1769 | ret = -EINVAL; | |
3381cb8d | 1770 | cfqg = blkg_to_cfqg(ctx.blkg); |
e48453c3 | 1771 | cfqgd = blkcg_to_cfqgd(blkcg); |
ae994ea9 JA |
1772 | if (!cfqg || !cfqgd) |
1773 | goto err; | |
1774 | ||
a2b1693b | 1775 | if (!ctx.v || (ctx.v >= CFQ_WEIGHT_MIN && ctx.v <= CFQ_WEIGHT_MAX)) { |
e71357e1 TH |
1776 | if (!is_leaf_weight) { |
1777 | cfqg->dev_weight = ctx.v; | |
e48453c3 | 1778 | cfqg->new_weight = ctx.v ?: cfqgd->weight; |
e71357e1 TH |
1779 | } else { |
1780 | cfqg->dev_leaf_weight = ctx.v; | |
e48453c3 | 1781 | cfqg->new_leaf_weight = ctx.v ?: cfqgd->leaf_weight; |
e71357e1 | 1782 | } |
60c2bc2d TH |
1783 | ret = 0; |
1784 | } | |
1785 | ||
ae994ea9 | 1786 | err: |
60c2bc2d | 1787 | blkg_conf_finish(&ctx); |
451af504 | 1788 | return ret ?: nbytes; |
60c2bc2d TH |
1789 | } |
1790 | ||
451af504 TH |
1791 | static ssize_t cfqg_set_weight_device(struct kernfs_open_file *of, |
1792 | char *buf, size_t nbytes, loff_t off) | |
e71357e1 | 1793 | { |
451af504 | 1794 | return __cfqg_set_weight_device(of, buf, nbytes, off, false); |
e71357e1 TH |
1795 | } |
1796 | ||
451af504 TH |
1797 | static ssize_t cfqg_set_leaf_weight_device(struct kernfs_open_file *of, |
1798 | char *buf, size_t nbytes, loff_t off) | |
e71357e1 | 1799 | { |
451af504 | 1800 | return __cfqg_set_weight_device(of, buf, nbytes, off, true); |
e71357e1 TH |
1801 | } |
1802 | ||
182446d0 TH |
1803 | static int __cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft, |
1804 | u64 val, bool is_leaf_weight) | |
60c2bc2d | 1805 | { |
182446d0 | 1806 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 1807 | struct blkcg_gq *blkg; |
e48453c3 | 1808 | struct cfq_group_data *cfqgd; |
ae994ea9 | 1809 | int ret = 0; |
60c2bc2d | 1810 | |
3381cb8d | 1811 | if (val < CFQ_WEIGHT_MIN || val > CFQ_WEIGHT_MAX) |
60c2bc2d TH |
1812 | return -EINVAL; |
1813 | ||
1814 | spin_lock_irq(&blkcg->lock); | |
e48453c3 | 1815 | cfqgd = blkcg_to_cfqgd(blkcg); |
ae994ea9 JA |
1816 | if (!cfqgd) { |
1817 | ret = -EINVAL; | |
1818 | goto out; | |
1819 | } | |
e71357e1 TH |
1820 | |
1821 | if (!is_leaf_weight) | |
e48453c3 | 1822 | cfqgd->weight = val; |
e71357e1 | 1823 | else |
e48453c3 | 1824 | cfqgd->leaf_weight = val; |
60c2bc2d | 1825 | |
b67bfe0d | 1826 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
3381cb8d | 1827 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
60c2bc2d | 1828 | |
e71357e1 TH |
1829 | if (!cfqg) |
1830 | continue; | |
1831 | ||
1832 | if (!is_leaf_weight) { | |
1833 | if (!cfqg->dev_weight) | |
e48453c3 | 1834 | cfqg->new_weight = cfqgd->weight; |
e71357e1 TH |
1835 | } else { |
1836 | if (!cfqg->dev_leaf_weight) | |
e48453c3 | 1837 | cfqg->new_leaf_weight = cfqgd->leaf_weight; |
e71357e1 | 1838 | } |
60c2bc2d TH |
1839 | } |
1840 | ||
ae994ea9 | 1841 | out: |
60c2bc2d | 1842 | spin_unlock_irq(&blkcg->lock); |
ae994ea9 | 1843 | return ret; |
60c2bc2d TH |
1844 | } |
1845 | ||
182446d0 TH |
1846 | static int cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft, |
1847 | u64 val) | |
e71357e1 | 1848 | { |
182446d0 | 1849 | return __cfq_set_weight(css, cft, val, false); |
e71357e1 TH |
1850 | } |
1851 | ||
182446d0 TH |
1852 | static int cfq_set_leaf_weight(struct cgroup_subsys_state *css, |
1853 | struct cftype *cft, u64 val) | |
e71357e1 | 1854 | { |
182446d0 | 1855 | return __cfq_set_weight(css, cft, val, true); |
e71357e1 TH |
1856 | } |
1857 | ||
2da8ca82 | 1858 | static int cfqg_print_stat(struct seq_file *sf, void *v) |
5bc4afb1 | 1859 | { |
2da8ca82 TH |
1860 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat, |
1861 | &blkcg_policy_cfq, seq_cft(sf)->private, false); | |
5bc4afb1 TH |
1862 | return 0; |
1863 | } | |
1864 | ||
2da8ca82 | 1865 | static int cfqg_print_rwstat(struct seq_file *sf, void *v) |
5bc4afb1 | 1866 | { |
2da8ca82 TH |
1867 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat, |
1868 | &blkcg_policy_cfq, seq_cft(sf)->private, true); | |
5bc4afb1 TH |
1869 | return 0; |
1870 | } | |
1871 | ||
43114018 TH |
1872 | static u64 cfqg_prfill_stat_recursive(struct seq_file *sf, |
1873 | struct blkg_policy_data *pd, int off) | |
1874 | { | |
f12c74ca TH |
1875 | u64 sum = blkg_stat_recursive_sum(pd_to_blkg(pd), |
1876 | &blkcg_policy_cfq, off); | |
43114018 TH |
1877 | return __blkg_prfill_u64(sf, pd, sum); |
1878 | } | |
1879 | ||
1880 | static u64 cfqg_prfill_rwstat_recursive(struct seq_file *sf, | |
1881 | struct blkg_policy_data *pd, int off) | |
1882 | { | |
f12c74ca TH |
1883 | struct blkg_rwstat sum = blkg_rwstat_recursive_sum(pd_to_blkg(pd), |
1884 | &blkcg_policy_cfq, off); | |
43114018 TH |
1885 | return __blkg_prfill_rwstat(sf, pd, &sum); |
1886 | } | |
1887 | ||
2da8ca82 | 1888 | static int cfqg_print_stat_recursive(struct seq_file *sf, void *v) |
43114018 | 1889 | { |
2da8ca82 TH |
1890 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1891 | cfqg_prfill_stat_recursive, &blkcg_policy_cfq, | |
1892 | seq_cft(sf)->private, false); | |
43114018 TH |
1893 | return 0; |
1894 | } | |
1895 | ||
2da8ca82 | 1896 | static int cfqg_print_rwstat_recursive(struct seq_file *sf, void *v) |
43114018 | 1897 | { |
2da8ca82 TH |
1898 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1899 | cfqg_prfill_rwstat_recursive, &blkcg_policy_cfq, | |
1900 | seq_cft(sf)->private, true); | |
43114018 TH |
1901 | return 0; |
1902 | } | |
1903 | ||
60c2bc2d | 1904 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
f95a04af TH |
1905 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, |
1906 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1907 | { |
f95a04af | 1908 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
155fead9 | 1909 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
60c2bc2d TH |
1910 | u64 v = 0; |
1911 | ||
1912 | if (samples) { | |
155fead9 | 1913 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
f3cff25f | 1914 | v = div64_u64(v, samples); |
60c2bc2d | 1915 | } |
f95a04af | 1916 | __blkg_prfill_u64(sf, pd, v); |
60c2bc2d TH |
1917 | return 0; |
1918 | } | |
1919 | ||
1920 | /* print avg_queue_size */ | |
2da8ca82 | 1921 | static int cfqg_print_avg_queue_size(struct seq_file *sf, void *v) |
60c2bc2d | 1922 | { |
2da8ca82 TH |
1923 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1924 | cfqg_prfill_avg_queue_size, &blkcg_policy_cfq, | |
1925 | 0, false); | |
60c2bc2d TH |
1926 | return 0; |
1927 | } | |
1928 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1929 | ||
1930 | static struct cftype cfq_blkcg_files[] = { | |
1d3650f7 | 1931 | /* on root, weight is mapped to leaf_weight */ |
60c2bc2d TH |
1932 | { |
1933 | .name = "weight_device", | |
1d3650f7 | 1934 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 1935 | .seq_show = cfqg_print_leaf_weight_device, |
451af504 | 1936 | .write = cfqg_set_leaf_weight_device, |
60c2bc2d TH |
1937 | }, |
1938 | { | |
1939 | .name = "weight", | |
1d3650f7 | 1940 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 1941 | .seq_show = cfq_print_leaf_weight, |
1d3650f7 | 1942 | .write_u64 = cfq_set_leaf_weight, |
60c2bc2d | 1943 | }, |
e71357e1 | 1944 | |
1d3650f7 | 1945 | /* no such mapping necessary for !roots */ |
60c2bc2d TH |
1946 | { |
1947 | .name = "weight_device", | |
1d3650f7 | 1948 | .flags = CFTYPE_NOT_ON_ROOT, |
2da8ca82 | 1949 | .seq_show = cfqg_print_weight_device, |
451af504 | 1950 | .write = cfqg_set_weight_device, |
60c2bc2d TH |
1951 | }, |
1952 | { | |
1953 | .name = "weight", | |
1d3650f7 | 1954 | .flags = CFTYPE_NOT_ON_ROOT, |
2da8ca82 | 1955 | .seq_show = cfq_print_weight, |
3381cb8d | 1956 | .write_u64 = cfq_set_weight, |
60c2bc2d | 1957 | }, |
e71357e1 | 1958 | |
e71357e1 TH |
1959 | { |
1960 | .name = "leaf_weight_device", | |
2da8ca82 | 1961 | .seq_show = cfqg_print_leaf_weight_device, |
451af504 | 1962 | .write = cfqg_set_leaf_weight_device, |
e71357e1 TH |
1963 | }, |
1964 | { | |
1965 | .name = "leaf_weight", | |
2da8ca82 | 1966 | .seq_show = cfq_print_leaf_weight, |
e71357e1 TH |
1967 | .write_u64 = cfq_set_leaf_weight, |
1968 | }, | |
1969 | ||
43114018 | 1970 | /* statistics, covers only the tasks in the cfqg */ |
60c2bc2d TH |
1971 | { |
1972 | .name = "time", | |
5bc4afb1 | 1973 | .private = offsetof(struct cfq_group, stats.time), |
2da8ca82 | 1974 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
1975 | }, |
1976 | { | |
1977 | .name = "sectors", | |
5bc4afb1 | 1978 | .private = offsetof(struct cfq_group, stats.sectors), |
2da8ca82 | 1979 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
1980 | }, |
1981 | { | |
1982 | .name = "io_service_bytes", | |
77ea7338 TH |
1983 | .private = (unsigned long)&blkcg_policy_cfq, |
1984 | .seq_show = blkg_print_stat_bytes, | |
60c2bc2d TH |
1985 | }, |
1986 | { | |
1987 | .name = "io_serviced", | |
77ea7338 TH |
1988 | .private = (unsigned long)&blkcg_policy_cfq, |
1989 | .seq_show = blkg_print_stat_ios, | |
60c2bc2d TH |
1990 | }, |
1991 | { | |
1992 | .name = "io_service_time", | |
5bc4afb1 | 1993 | .private = offsetof(struct cfq_group, stats.service_time), |
2da8ca82 | 1994 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
1995 | }, |
1996 | { | |
1997 | .name = "io_wait_time", | |
5bc4afb1 | 1998 | .private = offsetof(struct cfq_group, stats.wait_time), |
2da8ca82 | 1999 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2000 | }, |
2001 | { | |
2002 | .name = "io_merged", | |
5bc4afb1 | 2003 | .private = offsetof(struct cfq_group, stats.merged), |
2da8ca82 | 2004 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2005 | }, |
2006 | { | |
2007 | .name = "io_queued", | |
5bc4afb1 | 2008 | .private = offsetof(struct cfq_group, stats.queued), |
2da8ca82 | 2009 | .seq_show = cfqg_print_rwstat, |
60c2bc2d | 2010 | }, |
43114018 TH |
2011 | |
2012 | /* the same statictics which cover the cfqg and its descendants */ | |
2013 | { | |
2014 | .name = "time_recursive", | |
2015 | .private = offsetof(struct cfq_group, stats.time), | |
2da8ca82 | 2016 | .seq_show = cfqg_print_stat_recursive, |
43114018 TH |
2017 | }, |
2018 | { | |
2019 | .name = "sectors_recursive", | |
2020 | .private = offsetof(struct cfq_group, stats.sectors), | |
2da8ca82 | 2021 | .seq_show = cfqg_print_stat_recursive, |
43114018 TH |
2022 | }, |
2023 | { | |
2024 | .name = "io_service_bytes_recursive", | |
77ea7338 TH |
2025 | .private = (unsigned long)&blkcg_policy_cfq, |
2026 | .seq_show = blkg_print_stat_bytes_recursive, | |
43114018 TH |
2027 | }, |
2028 | { | |
2029 | .name = "io_serviced_recursive", | |
77ea7338 TH |
2030 | .private = (unsigned long)&blkcg_policy_cfq, |
2031 | .seq_show = blkg_print_stat_ios_recursive, | |
43114018 TH |
2032 | }, |
2033 | { | |
2034 | .name = "io_service_time_recursive", | |
2035 | .private = offsetof(struct cfq_group, stats.service_time), | |
2da8ca82 | 2036 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2037 | }, |
2038 | { | |
2039 | .name = "io_wait_time_recursive", | |
2040 | .private = offsetof(struct cfq_group, stats.wait_time), | |
2da8ca82 | 2041 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2042 | }, |
2043 | { | |
2044 | .name = "io_merged_recursive", | |
2045 | .private = offsetof(struct cfq_group, stats.merged), | |
2da8ca82 | 2046 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2047 | }, |
2048 | { | |
2049 | .name = "io_queued_recursive", | |
2050 | .private = offsetof(struct cfq_group, stats.queued), | |
2da8ca82 | 2051 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 | 2052 | }, |
60c2bc2d TH |
2053 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
2054 | { | |
2055 | .name = "avg_queue_size", | |
2da8ca82 | 2056 | .seq_show = cfqg_print_avg_queue_size, |
60c2bc2d TH |
2057 | }, |
2058 | { | |
2059 | .name = "group_wait_time", | |
5bc4afb1 | 2060 | .private = offsetof(struct cfq_group, stats.group_wait_time), |
2da8ca82 | 2061 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2062 | }, |
2063 | { | |
2064 | .name = "idle_time", | |
5bc4afb1 | 2065 | .private = offsetof(struct cfq_group, stats.idle_time), |
2da8ca82 | 2066 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2067 | }, |
2068 | { | |
2069 | .name = "empty_time", | |
5bc4afb1 | 2070 | .private = offsetof(struct cfq_group, stats.empty_time), |
2da8ca82 | 2071 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2072 | }, |
2073 | { | |
2074 | .name = "dequeue", | |
5bc4afb1 | 2075 | .private = offsetof(struct cfq_group, stats.dequeue), |
2da8ca82 | 2076 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2077 | }, |
2078 | { | |
2079 | .name = "unaccounted_time", | |
5bc4afb1 | 2080 | .private = offsetof(struct cfq_group, stats.unaccounted_time), |
2da8ca82 | 2081 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2082 | }, |
2083 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
2084 | { } /* terminate */ | |
2085 | }; | |
25fb5169 | 2086 | #else /* GROUP_IOSCHED */ |
ae118896 TH |
2087 | static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd, |
2088 | struct blkcg *blkcg) | |
25fb5169 | 2089 | { |
f51b802c | 2090 | return cfqd->root_group; |
25fb5169 | 2091 | } |
7f1dc8a2 | 2092 | |
25fb5169 VG |
2093 | static inline void |
2094 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
2095 | cfqq->cfqg = cfqg; | |
2096 | } | |
2097 | ||
2098 | #endif /* GROUP_IOSCHED */ | |
2099 | ||
498d3aa2 | 2100 | /* |
c0324a02 | 2101 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
2102 | * requests waiting to be processed. It is sorted in the order that |
2103 | * we will service the queues. | |
2104 | */ | |
a36e71f9 | 2105 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2106 | bool add_front) |
d9e7620e | 2107 | { |
0871714e JA |
2108 | struct rb_node **p, *parent; |
2109 | struct cfq_queue *__cfqq; | |
d9e7620e | 2110 | unsigned long rb_key; |
34b98d03 | 2111 | struct cfq_rb_root *st; |
498d3aa2 | 2112 | int left; |
dae739eb | 2113 | int new_cfqq = 1; |
ae30c286 | 2114 | |
34b98d03 | 2115 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq)); |
0871714e JA |
2116 | if (cfq_class_idle(cfqq)) { |
2117 | rb_key = CFQ_IDLE_DELAY; | |
34b98d03 | 2118 | parent = rb_last(&st->rb); |
0871714e JA |
2119 | if (parent && parent != &cfqq->rb_node) { |
2120 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2121 | rb_key += __cfqq->rb_key; | |
2122 | } else | |
2123 | rb_key += jiffies; | |
2124 | } else if (!add_front) { | |
b9c8946b JA |
2125 | /* |
2126 | * Get our rb key offset. Subtract any residual slice | |
2127 | * value carried from last service. A negative resid | |
2128 | * count indicates slice overrun, and this should position | |
2129 | * the next service time further away in the tree. | |
2130 | */ | |
edd75ffd | 2131 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 2132 | rb_key -= cfqq->slice_resid; |
edd75ffd | 2133 | cfqq->slice_resid = 0; |
48e025e6 CZ |
2134 | } else { |
2135 | rb_key = -HZ; | |
34b98d03 | 2136 | __cfqq = cfq_rb_first(st); |
48e025e6 CZ |
2137 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
2138 | } | |
1da177e4 | 2139 | |
d9e7620e | 2140 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 2141 | new_cfqq = 0; |
99f9628a | 2142 | /* |
d9e7620e | 2143 | * same position, nothing more to do |
99f9628a | 2144 | */ |
34b98d03 | 2145 | if (rb_key == cfqq->rb_key && cfqq->service_tree == st) |
d9e7620e | 2146 | return; |
1da177e4 | 2147 | |
aa6f6a3d CZ |
2148 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
2149 | cfqq->service_tree = NULL; | |
1da177e4 | 2150 | } |
d9e7620e | 2151 | |
498d3aa2 | 2152 | left = 1; |
0871714e | 2153 | parent = NULL; |
34b98d03 VG |
2154 | cfqq->service_tree = st; |
2155 | p = &st->rb.rb_node; | |
d9e7620e JA |
2156 | while (*p) { |
2157 | parent = *p; | |
2158 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2159 | ||
0c534e0a | 2160 | /* |
c0324a02 | 2161 | * sort by key, that represents service time. |
0c534e0a | 2162 | */ |
c0324a02 | 2163 | if (time_before(rb_key, __cfqq->rb_key)) |
1f23f121 | 2164 | p = &parent->rb_left; |
c0324a02 | 2165 | else { |
1f23f121 | 2166 | p = &parent->rb_right; |
cc09e299 | 2167 | left = 0; |
c0324a02 | 2168 | } |
d9e7620e JA |
2169 | } |
2170 | ||
cc09e299 | 2171 | if (left) |
34b98d03 | 2172 | st->left = &cfqq->rb_node; |
cc09e299 | 2173 | |
d9e7620e JA |
2174 | cfqq->rb_key = rb_key; |
2175 | rb_link_node(&cfqq->rb_node, parent, p); | |
34b98d03 VG |
2176 | rb_insert_color(&cfqq->rb_node, &st->rb); |
2177 | st->count++; | |
20359f27 | 2178 | if (add_front || !new_cfqq) |
dae739eb | 2179 | return; |
8184f93e | 2180 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
2181 | } |
2182 | ||
a36e71f9 | 2183 | static struct cfq_queue * |
f2d1f0ae JA |
2184 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
2185 | sector_t sector, struct rb_node **ret_parent, | |
2186 | struct rb_node ***rb_link) | |
a36e71f9 | 2187 | { |
a36e71f9 JA |
2188 | struct rb_node **p, *parent; |
2189 | struct cfq_queue *cfqq = NULL; | |
2190 | ||
2191 | parent = NULL; | |
2192 | p = &root->rb_node; | |
2193 | while (*p) { | |
2194 | struct rb_node **n; | |
2195 | ||
2196 | parent = *p; | |
2197 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
2198 | ||
2199 | /* | |
2200 | * Sort strictly based on sector. Smallest to the left, | |
2201 | * largest to the right. | |
2202 | */ | |
2e46e8b2 | 2203 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 2204 | n = &(*p)->rb_right; |
2e46e8b2 | 2205 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
2206 | n = &(*p)->rb_left; |
2207 | else | |
2208 | break; | |
2209 | p = n; | |
3ac6c9f8 | 2210 | cfqq = NULL; |
a36e71f9 JA |
2211 | } |
2212 | ||
2213 | *ret_parent = parent; | |
2214 | if (rb_link) | |
2215 | *rb_link = p; | |
3ac6c9f8 | 2216 | return cfqq; |
a36e71f9 JA |
2217 | } |
2218 | ||
2219 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2220 | { | |
a36e71f9 JA |
2221 | struct rb_node **p, *parent; |
2222 | struct cfq_queue *__cfqq; | |
2223 | ||
f2d1f0ae JA |
2224 | if (cfqq->p_root) { |
2225 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2226 | cfqq->p_root = NULL; | |
2227 | } | |
a36e71f9 JA |
2228 | |
2229 | if (cfq_class_idle(cfqq)) | |
2230 | return; | |
2231 | if (!cfqq->next_rq) | |
2232 | return; | |
2233 | ||
f2d1f0ae | 2234 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
2235 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
2236 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
2237 | if (!__cfqq) { |
2238 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
2239 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
2240 | } else | |
2241 | cfqq->p_root = NULL; | |
a36e71f9 JA |
2242 | } |
2243 | ||
498d3aa2 JA |
2244 | /* |
2245 | * Update cfqq's position in the service tree. | |
2246 | */ | |
edd75ffd | 2247 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 2248 | { |
6d048f53 JA |
2249 | /* |
2250 | * Resorting requires the cfqq to be on the RR list already. | |
2251 | */ | |
a36e71f9 | 2252 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 2253 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
2254 | cfq_prio_tree_add(cfqd, cfqq); |
2255 | } | |
6d048f53 JA |
2256 | } |
2257 | ||
1da177e4 LT |
2258 | /* |
2259 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 2260 | * the pending list according to last request service |
1da177e4 | 2261 | */ |
febffd61 | 2262 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2263 | { |
7b679138 | 2264 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
2265 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
2266 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 2267 | cfqd->busy_queues++; |
ef8a41df SL |
2268 | if (cfq_cfqq_sync(cfqq)) |
2269 | cfqd->busy_sync_queues++; | |
1da177e4 | 2270 | |
edd75ffd | 2271 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
2272 | } |
2273 | ||
498d3aa2 JA |
2274 | /* |
2275 | * Called when the cfqq no longer has requests pending, remove it from | |
2276 | * the service tree. | |
2277 | */ | |
febffd61 | 2278 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2279 | { |
7b679138 | 2280 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
2281 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
2282 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 2283 | |
aa6f6a3d CZ |
2284 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
2285 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
2286 | cfqq->service_tree = NULL; | |
2287 | } | |
f2d1f0ae JA |
2288 | if (cfqq->p_root) { |
2289 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2290 | cfqq->p_root = NULL; | |
2291 | } | |
d9e7620e | 2292 | |
8184f93e | 2293 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
2294 | BUG_ON(!cfqd->busy_queues); |
2295 | cfqd->busy_queues--; | |
ef8a41df SL |
2296 | if (cfq_cfqq_sync(cfqq)) |
2297 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
2298 | } |
2299 | ||
2300 | /* | |
2301 | * rb tree support functions | |
2302 | */ | |
febffd61 | 2303 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 2304 | { |
5e705374 | 2305 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 2306 | const int sync = rq_is_sync(rq); |
1da177e4 | 2307 | |
b4878f24 JA |
2308 | BUG_ON(!cfqq->queued[sync]); |
2309 | cfqq->queued[sync]--; | |
1da177e4 | 2310 | |
5e705374 | 2311 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 2312 | |
f04a6424 VG |
2313 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
2314 | /* | |
2315 | * Queue will be deleted from service tree when we actually | |
2316 | * expire it later. Right now just remove it from prio tree | |
2317 | * as it is empty. | |
2318 | */ | |
2319 | if (cfqq->p_root) { | |
2320 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2321 | cfqq->p_root = NULL; | |
2322 | } | |
2323 | } | |
1da177e4 LT |
2324 | } |
2325 | ||
5e705374 | 2326 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 2327 | { |
5e705374 | 2328 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2329 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 2330 | struct request *prev; |
1da177e4 | 2331 | |
5380a101 | 2332 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 2333 | |
796d5116 | 2334 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
2335 | |
2336 | if (!cfq_cfqq_on_rr(cfqq)) | |
2337 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
2338 | |
2339 | /* | |
2340 | * check if this request is a better next-serve candidate | |
2341 | */ | |
a36e71f9 | 2342 | prev = cfqq->next_rq; |
cf7c25cf | 2343 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
2344 | |
2345 | /* | |
2346 | * adjust priority tree position, if ->next_rq changes | |
2347 | */ | |
2348 | if (prev != cfqq->next_rq) | |
2349 | cfq_prio_tree_add(cfqd, cfqq); | |
2350 | ||
5044eed4 | 2351 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
2352 | } |
2353 | ||
febffd61 | 2354 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 2355 | { |
5380a101 JA |
2356 | elv_rb_del(&cfqq->sort_list, rq); |
2357 | cfqq->queued[rq_is_sync(rq)]--; | |
155fead9 | 2358 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
5e705374 | 2359 | cfq_add_rq_rb(rq); |
155fead9 TH |
2360 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
2361 | rq->cmd_flags); | |
1da177e4 LT |
2362 | } |
2363 | ||
206dc69b JA |
2364 | static struct request * |
2365 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 2366 | { |
206dc69b | 2367 | struct task_struct *tsk = current; |
c5869807 | 2368 | struct cfq_io_cq *cic; |
206dc69b | 2369 | struct cfq_queue *cfqq; |
1da177e4 | 2370 | |
4ac845a2 | 2371 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2372 | if (!cic) |
2373 | return NULL; | |
2374 | ||
2375 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
f73a1c7d KO |
2376 | if (cfqq) |
2377 | return elv_rb_find(&cfqq->sort_list, bio_end_sector(bio)); | |
1da177e4 | 2378 | |
1da177e4 LT |
2379 | return NULL; |
2380 | } | |
2381 | ||
165125e1 | 2382 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2383 | { |
22e2c507 | 2384 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 2385 | |
53c583d2 | 2386 | cfqd->rq_in_driver++; |
7b679138 | 2387 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 2388 | cfqd->rq_in_driver); |
25776e35 | 2389 | |
5b93629b | 2390 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
2391 | } |
2392 | ||
165125e1 | 2393 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2394 | { |
b4878f24 JA |
2395 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2396 | ||
53c583d2 CZ |
2397 | WARN_ON(!cfqd->rq_in_driver); |
2398 | cfqd->rq_in_driver--; | |
7b679138 | 2399 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 2400 | cfqd->rq_in_driver); |
1da177e4 LT |
2401 | } |
2402 | ||
b4878f24 | 2403 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 2404 | { |
5e705374 | 2405 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 2406 | |
5e705374 JA |
2407 | if (cfqq->next_rq == rq) |
2408 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 2409 | |
b4878f24 | 2410 | list_del_init(&rq->queuelist); |
5e705374 | 2411 | cfq_del_rq_rb(rq); |
374f84ac | 2412 | |
45333d5a | 2413 | cfqq->cfqd->rq_queued--; |
155fead9 | 2414 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
65299a3b CH |
2415 | if (rq->cmd_flags & REQ_PRIO) { |
2416 | WARN_ON(!cfqq->prio_pending); | |
2417 | cfqq->prio_pending--; | |
b53d1ed7 | 2418 | } |
1da177e4 LT |
2419 | } |
2420 | ||
165125e1 JA |
2421 | static int cfq_merge(struct request_queue *q, struct request **req, |
2422 | struct bio *bio) | |
1da177e4 LT |
2423 | { |
2424 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2425 | struct request *__rq; | |
1da177e4 | 2426 | |
206dc69b | 2427 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 2428 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
2429 | *req = __rq; |
2430 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
2431 | } |
2432 | ||
2433 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
2434 | } |
2435 | ||
165125e1 | 2436 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 2437 | int type) |
1da177e4 | 2438 | { |
21183b07 | 2439 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 2440 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 2441 | |
5e705374 | 2442 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 2443 | } |
1da177e4 LT |
2444 | } |
2445 | ||
812d4026 DS |
2446 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
2447 | struct bio *bio) | |
2448 | { | |
155fead9 | 2449 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_rw); |
812d4026 DS |
2450 | } |
2451 | ||
1da177e4 | 2452 | static void |
165125e1 | 2453 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
2454 | struct request *next) |
2455 | { | |
cf7c25cf | 2456 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
2457 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2458 | ||
22e2c507 JA |
2459 | /* |
2460 | * reposition in fifo if next is older than rq | |
2461 | */ | |
2462 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
8b4922d3 | 2463 | time_before(next->fifo_time, rq->fifo_time) && |
3d106fba | 2464 | cfqq == RQ_CFQQ(next)) { |
22e2c507 | 2465 | list_move(&rq->queuelist, &next->queuelist); |
8b4922d3 | 2466 | rq->fifo_time = next->fifo_time; |
30996f40 | 2467 | } |
22e2c507 | 2468 | |
cf7c25cf CZ |
2469 | if (cfqq->next_rq == next) |
2470 | cfqq->next_rq = rq; | |
b4878f24 | 2471 | cfq_remove_request(next); |
155fead9 | 2472 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
4a0b75c7 SL |
2473 | |
2474 | cfqq = RQ_CFQQ(next); | |
2475 | /* | |
2476 | * all requests of this queue are merged to other queues, delete it | |
2477 | * from the service tree. If it's the active_queue, | |
2478 | * cfq_dispatch_requests() will choose to expire it or do idle | |
2479 | */ | |
2480 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
2481 | cfqq != cfqd->active_queue) | |
2482 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
2483 | } |
2484 | ||
165125e1 | 2485 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
2486 | struct bio *bio) |
2487 | { | |
2488 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 2489 | struct cfq_io_cq *cic; |
da775265 | 2490 | struct cfq_queue *cfqq; |
da775265 JA |
2491 | |
2492 | /* | |
ec8acb69 | 2493 | * Disallow merge of a sync bio into an async request. |
da775265 | 2494 | */ |
91fac317 | 2495 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 2496 | return false; |
da775265 JA |
2497 | |
2498 | /* | |
f1a4f4d3 | 2499 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 2500 | * merge only if rq is queued there. |
f1a4f4d3 | 2501 | */ |
07c2bd37 TH |
2502 | cic = cfq_cic_lookup(cfqd, current->io_context); |
2503 | if (!cic) | |
2504 | return false; | |
719d3402 | 2505 | |
91fac317 | 2506 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 2507 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
2508 | } |
2509 | ||
812df48d DS |
2510 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2511 | { | |
2512 | del_timer(&cfqd->idle_slice_timer); | |
155fead9 | 2513 | cfqg_stats_update_idle_time(cfqq->cfqg); |
812df48d DS |
2514 | } |
2515 | ||
febffd61 JA |
2516 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
2517 | struct cfq_queue *cfqq) | |
22e2c507 JA |
2518 | { |
2519 | if (cfqq) { | |
3bf10fea | 2520 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d", |
4d2ceea4 | 2521 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
155fead9 | 2522 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
62a37f6b JT |
2523 | cfqq->slice_start = 0; |
2524 | cfqq->dispatch_start = jiffies; | |
2525 | cfqq->allocated_slice = 0; | |
2526 | cfqq->slice_end = 0; | |
2527 | cfqq->slice_dispatch = 0; | |
2528 | cfqq->nr_sectors = 0; | |
2529 | ||
2530 | cfq_clear_cfqq_wait_request(cfqq); | |
2531 | cfq_clear_cfqq_must_dispatch(cfqq); | |
2532 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
2533 | cfq_clear_cfqq_fifo_expire(cfqq); | |
2534 | cfq_mark_cfqq_slice_new(cfqq); | |
2535 | ||
2536 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
2537 | } |
2538 | ||
2539 | cfqd->active_queue = cfqq; | |
2540 | } | |
2541 | ||
7b14e3b5 JA |
2542 | /* |
2543 | * current cfqq expired its slice (or was too idle), select new one | |
2544 | */ | |
2545 | static void | |
2546 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 2547 | bool timed_out) |
7b14e3b5 | 2548 | { |
7b679138 JA |
2549 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
2550 | ||
7b14e3b5 | 2551 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 2552 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 2553 | |
7b14e3b5 | 2554 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 2555 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 2556 | |
ae54abed SL |
2557 | /* |
2558 | * If this cfqq is shared between multiple processes, check to | |
2559 | * make sure that those processes are still issuing I/Os within | |
2560 | * the mean seek distance. If not, it may be time to break the | |
2561 | * queues apart again. | |
2562 | */ | |
2563 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
2564 | cfq_mark_cfqq_split_coop(cfqq); | |
2565 | ||
7b14e3b5 | 2566 | /* |
6084cdda | 2567 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 2568 | */ |
c553f8e3 SL |
2569 | if (timed_out) { |
2570 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 2571 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
2572 | else |
2573 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
2574 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
2575 | } | |
7b14e3b5 | 2576 | |
e5ff082e | 2577 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 2578 | |
f04a6424 VG |
2579 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
2580 | cfq_del_cfqq_rr(cfqd, cfqq); | |
2581 | ||
edd75ffd | 2582 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
2583 | |
2584 | if (cfqq == cfqd->active_queue) | |
2585 | cfqd->active_queue = NULL; | |
2586 | ||
2587 | if (cfqd->active_cic) { | |
11a3122f | 2588 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
2589 | cfqd->active_cic = NULL; |
2590 | } | |
7b14e3b5 JA |
2591 | } |
2592 | ||
e5ff082e | 2593 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
2594 | { |
2595 | struct cfq_queue *cfqq = cfqd->active_queue; | |
2596 | ||
2597 | if (cfqq) | |
e5ff082e | 2598 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
2599 | } |
2600 | ||
498d3aa2 JA |
2601 | /* |
2602 | * Get next queue for service. Unless we have a queue preemption, | |
2603 | * we'll simply select the first cfqq in the service tree. | |
2604 | */ | |
6d048f53 | 2605 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 2606 | { |
34b98d03 VG |
2607 | struct cfq_rb_root *st = st_for(cfqd->serving_group, |
2608 | cfqd->serving_wl_class, cfqd->serving_wl_type); | |
d9e7620e | 2609 | |
f04a6424 VG |
2610 | if (!cfqd->rq_queued) |
2611 | return NULL; | |
2612 | ||
1fa8f6d6 | 2613 | /* There is nothing to dispatch */ |
34b98d03 | 2614 | if (!st) |
1fa8f6d6 | 2615 | return NULL; |
34b98d03 | 2616 | if (RB_EMPTY_ROOT(&st->rb)) |
c0324a02 | 2617 | return NULL; |
34b98d03 | 2618 | return cfq_rb_first(st); |
6d048f53 JA |
2619 | } |
2620 | ||
f04a6424 VG |
2621 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
2622 | { | |
25fb5169 | 2623 | struct cfq_group *cfqg; |
f04a6424 VG |
2624 | struct cfq_queue *cfqq; |
2625 | int i, j; | |
2626 | struct cfq_rb_root *st; | |
2627 | ||
2628 | if (!cfqd->rq_queued) | |
2629 | return NULL; | |
2630 | ||
25fb5169 VG |
2631 | cfqg = cfq_get_next_cfqg(cfqd); |
2632 | if (!cfqg) | |
2633 | return NULL; | |
2634 | ||
f04a6424 VG |
2635 | for_each_cfqg_st(cfqg, i, j, st) |
2636 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
2637 | return cfqq; | |
2638 | return NULL; | |
2639 | } | |
2640 | ||
498d3aa2 JA |
2641 | /* |
2642 | * Get and set a new active queue for service. | |
2643 | */ | |
a36e71f9 JA |
2644 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
2645 | struct cfq_queue *cfqq) | |
6d048f53 | 2646 | { |
e00ef799 | 2647 | if (!cfqq) |
a36e71f9 | 2648 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 2649 | |
22e2c507 | 2650 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 2651 | return cfqq; |
22e2c507 JA |
2652 | } |
2653 | ||
d9e7620e JA |
2654 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
2655 | struct request *rq) | |
2656 | { | |
83096ebf TH |
2657 | if (blk_rq_pos(rq) >= cfqd->last_position) |
2658 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 2659 | else |
83096ebf | 2660 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
2661 | } |
2662 | ||
b2c18e1e | 2663 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 2664 | struct request *rq) |
6d048f53 | 2665 | { |
e9ce335d | 2666 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
2667 | } |
2668 | ||
a36e71f9 JA |
2669 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
2670 | struct cfq_queue *cur_cfqq) | |
2671 | { | |
f2d1f0ae | 2672 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
2673 | struct rb_node *parent, *node; |
2674 | struct cfq_queue *__cfqq; | |
2675 | sector_t sector = cfqd->last_position; | |
2676 | ||
2677 | if (RB_EMPTY_ROOT(root)) | |
2678 | return NULL; | |
2679 | ||
2680 | /* | |
2681 | * First, if we find a request starting at the end of the last | |
2682 | * request, choose it. | |
2683 | */ | |
f2d1f0ae | 2684 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
2685 | if (__cfqq) |
2686 | return __cfqq; | |
2687 | ||
2688 | /* | |
2689 | * If the exact sector wasn't found, the parent of the NULL leaf | |
2690 | * will contain the closest sector. | |
2691 | */ | |
2692 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 2693 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2694 | return __cfqq; |
2695 | ||
2e46e8b2 | 2696 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
2697 | node = rb_next(&__cfqq->p_node); |
2698 | else | |
2699 | node = rb_prev(&__cfqq->p_node); | |
2700 | if (!node) | |
2701 | return NULL; | |
2702 | ||
2703 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 2704 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2705 | return __cfqq; |
2706 | ||
2707 | return NULL; | |
2708 | } | |
2709 | ||
2710 | /* | |
2711 | * cfqd - obvious | |
2712 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
2713 | * closely cooperating with itself. | |
2714 | * | |
2715 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
2716 | * one request, and that cfqd->last_position reflects a position on the disk | |
2717 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
2718 | * assumption. | |
2719 | */ | |
2720 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 2721 | struct cfq_queue *cur_cfqq) |
6d048f53 | 2722 | { |
a36e71f9 JA |
2723 | struct cfq_queue *cfqq; |
2724 | ||
39c01b21 DS |
2725 | if (cfq_class_idle(cur_cfqq)) |
2726 | return NULL; | |
e6c5bc73 JM |
2727 | if (!cfq_cfqq_sync(cur_cfqq)) |
2728 | return NULL; | |
2729 | if (CFQQ_SEEKY(cur_cfqq)) | |
2730 | return NULL; | |
2731 | ||
b9d8f4c7 GJ |
2732 | /* |
2733 | * Don't search priority tree if it's the only queue in the group. | |
2734 | */ | |
2735 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
2736 | return NULL; | |
2737 | ||
6d048f53 | 2738 | /* |
d9e7620e JA |
2739 | * We should notice if some of the queues are cooperating, eg |
2740 | * working closely on the same area of the disk. In that case, | |
2741 | * we can group them together and don't waste time idling. | |
6d048f53 | 2742 | */ |
a36e71f9 JA |
2743 | cfqq = cfqq_close(cfqd, cur_cfqq); |
2744 | if (!cfqq) | |
2745 | return NULL; | |
2746 | ||
8682e1f1 VG |
2747 | /* If new queue belongs to different cfq_group, don't choose it */ |
2748 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
2749 | return NULL; | |
2750 | ||
df5fe3e8 JM |
2751 | /* |
2752 | * It only makes sense to merge sync queues. | |
2753 | */ | |
2754 | if (!cfq_cfqq_sync(cfqq)) | |
2755 | return NULL; | |
e6c5bc73 JM |
2756 | if (CFQQ_SEEKY(cfqq)) |
2757 | return NULL; | |
df5fe3e8 | 2758 | |
c0324a02 CZ |
2759 | /* |
2760 | * Do not merge queues of different priority classes | |
2761 | */ | |
2762 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
2763 | return NULL; | |
2764 | ||
a36e71f9 | 2765 | return cfqq; |
6d048f53 JA |
2766 | } |
2767 | ||
a6d44e98 CZ |
2768 | /* |
2769 | * Determine whether we should enforce idle window for this queue. | |
2770 | */ | |
2771 | ||
2772 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2773 | { | |
3bf10fea | 2774 | enum wl_class_t wl_class = cfqq_class(cfqq); |
34b98d03 | 2775 | struct cfq_rb_root *st = cfqq->service_tree; |
a6d44e98 | 2776 | |
34b98d03 VG |
2777 | BUG_ON(!st); |
2778 | BUG_ON(!st->count); | |
f04a6424 | 2779 | |
b6508c16 VG |
2780 | if (!cfqd->cfq_slice_idle) |
2781 | return false; | |
2782 | ||
a6d44e98 | 2783 | /* We never do for idle class queues. */ |
3bf10fea | 2784 | if (wl_class == IDLE_WORKLOAD) |
a6d44e98 CZ |
2785 | return false; |
2786 | ||
2787 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
2788 | if (cfq_cfqq_idle_window(cfqq) && |
2789 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
2790 | return true; |
2791 | ||
2792 | /* | |
2793 | * Otherwise, we do only if they are the last ones | |
2794 | * in their service tree. | |
2795 | */ | |
34b98d03 VG |
2796 | if (st->count == 1 && cfq_cfqq_sync(cfqq) && |
2797 | !cfq_io_thinktime_big(cfqd, &st->ttime, false)) | |
c1e44756 | 2798 | return true; |
34b98d03 | 2799 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count); |
c1e44756 | 2800 | return false; |
a6d44e98 CZ |
2801 | } |
2802 | ||
6d048f53 | 2803 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2804 | { |
1792669c | 2805 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 2806 | struct cfq_io_cq *cic; |
80bdf0c7 | 2807 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 2808 | |
a68bbddb | 2809 | /* |
f7d7b7a7 JA |
2810 | * SSD device without seek penalty, disable idling. But only do so |
2811 | * for devices that support queuing, otherwise we still have a problem | |
2812 | * with sync vs async workloads. | |
a68bbddb | 2813 | */ |
f7d7b7a7 | 2814 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
2815 | return; |
2816 | ||
dd67d051 | 2817 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2818 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2819 | |
2820 | /* | |
2821 | * idle is disabled, either manually or by past process history | |
2822 | */ | |
80bdf0c7 VG |
2823 | if (!cfq_should_idle(cfqd, cfqq)) { |
2824 | /* no queue idling. Check for group idling */ | |
2825 | if (cfqd->cfq_group_idle) | |
2826 | group_idle = cfqd->cfq_group_idle; | |
2827 | else | |
2828 | return; | |
2829 | } | |
6d048f53 | 2830 | |
7b679138 | 2831 | /* |
8e550632 | 2832 | * still active requests from this queue, don't idle |
7b679138 | 2833 | */ |
8e550632 | 2834 | if (cfqq->dispatched) |
7b679138 JA |
2835 | return; |
2836 | ||
22e2c507 JA |
2837 | /* |
2838 | * task has exited, don't wait | |
2839 | */ | |
206dc69b | 2840 | cic = cfqd->active_cic; |
f6e8d01b | 2841 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
2842 | return; |
2843 | ||
355b659c CZ |
2844 | /* |
2845 | * If our average think time is larger than the remaining time | |
2846 | * slice, then don't idle. This avoids overrunning the allotted | |
2847 | * time slice. | |
2848 | */ | |
383cd721 SL |
2849 | if (sample_valid(cic->ttime.ttime_samples) && |
2850 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 2851 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 2852 | cic->ttime.ttime_mean); |
355b659c | 2853 | return; |
b1ffe737 | 2854 | } |
355b659c | 2855 | |
80bdf0c7 VG |
2856 | /* There are other queues in the group, don't do group idle */ |
2857 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
2858 | return; | |
2859 | ||
3b18152c | 2860 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2861 | |
80bdf0c7 VG |
2862 | if (group_idle) |
2863 | sl = cfqd->cfq_group_idle; | |
2864 | else | |
2865 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2866 | |
7b14e3b5 | 2867 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
155fead9 | 2868 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
80bdf0c7 VG |
2869 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
2870 | group_idle ? 1 : 0); | |
1da177e4 LT |
2871 | } |
2872 | ||
498d3aa2 JA |
2873 | /* |
2874 | * Move request from internal lists to the request queue dispatch list. | |
2875 | */ | |
165125e1 | 2876 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2877 | { |
3ed9a296 | 2878 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2879 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2880 | |
7b679138 JA |
2881 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2882 | ||
06d21886 | 2883 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2884 | cfq_remove_request(rq); |
6d048f53 | 2885 | cfqq->dispatched++; |
80bdf0c7 | 2886 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2887 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2888 | |
53c583d2 | 2889 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2890 | cfqq->nr_sectors += blk_rq_sectors(rq); |
155fead9 | 2891 | cfqg_stats_update_dispatch(cfqq->cfqg, blk_rq_bytes(rq), rq->cmd_flags); |
1da177e4 LT |
2892 | } |
2893 | ||
2894 | /* | |
2895 | * return expired entry, or NULL to just start from scratch in rbtree | |
2896 | */ | |
febffd61 | 2897 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 2898 | { |
30996f40 | 2899 | struct request *rq = NULL; |
1da177e4 | 2900 | |
3b18152c | 2901 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 2902 | return NULL; |
cb887411 JA |
2903 | |
2904 | cfq_mark_cfqq_fifo_expire(cfqq); | |
2905 | ||
89850f7e JA |
2906 | if (list_empty(&cfqq->fifo)) |
2907 | return NULL; | |
1da177e4 | 2908 | |
89850f7e | 2909 | rq = rq_entry_fifo(cfqq->fifo.next); |
8b4922d3 | 2910 | if (time_before(jiffies, rq->fifo_time)) |
7b679138 | 2911 | rq = NULL; |
1da177e4 | 2912 | |
30996f40 | 2913 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 2914 | return rq; |
1da177e4 LT |
2915 | } |
2916 | ||
22e2c507 JA |
2917 | static inline int |
2918 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2919 | { | |
2920 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2921 | |
22e2c507 | 2922 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2923 | |
b9f8ce05 | 2924 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
2925 | } |
2926 | ||
df5fe3e8 JM |
2927 | /* |
2928 | * Must be called with the queue_lock held. | |
2929 | */ | |
2930 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2931 | { | |
2932 | int process_refs, io_refs; | |
2933 | ||
2934 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 2935 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
2936 | BUG_ON(process_refs < 0); |
2937 | return process_refs; | |
2938 | } | |
2939 | ||
2940 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2941 | { | |
e6c5bc73 | 2942 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2943 | struct cfq_queue *__cfqq; |
2944 | ||
c10b61f0 JM |
2945 | /* |
2946 | * If there are no process references on the new_cfqq, then it is | |
2947 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2948 | * chain may have dropped their last reference (not just their | |
2949 | * last process reference). | |
2950 | */ | |
2951 | if (!cfqq_process_refs(new_cfqq)) | |
2952 | return; | |
2953 | ||
df5fe3e8 JM |
2954 | /* Avoid a circular list and skip interim queue merges */ |
2955 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2956 | if (__cfqq == cfqq) | |
2957 | return; | |
2958 | new_cfqq = __cfqq; | |
2959 | } | |
2960 | ||
2961 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2962 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2963 | /* |
2964 | * If the process for the cfqq has gone away, there is no | |
2965 | * sense in merging the queues. | |
2966 | */ | |
c10b61f0 | 2967 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2968 | return; |
2969 | ||
e6c5bc73 JM |
2970 | /* |
2971 | * Merge in the direction of the lesser amount of work. | |
2972 | */ | |
e6c5bc73 JM |
2973 | if (new_process_refs >= process_refs) { |
2974 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2975 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2976 | } else { |
2977 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2978 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2979 | } |
df5fe3e8 JM |
2980 | } |
2981 | ||
6d816ec7 | 2982 | static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd, |
3bf10fea | 2983 | struct cfq_group *cfqg, enum wl_class_t wl_class) |
718eee05 CZ |
2984 | { |
2985 | struct cfq_queue *queue; | |
2986 | int i; | |
2987 | bool key_valid = false; | |
2988 | unsigned long lowest_key = 0; | |
2989 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2990 | ||
65b32a57 VG |
2991 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2992 | /* select the one with lowest rb_key */ | |
34b98d03 | 2993 | queue = cfq_rb_first(st_for(cfqg, wl_class, i)); |
718eee05 CZ |
2994 | if (queue && |
2995 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2996 | lowest_key = queue->rb_key; | |
2997 | cur_best = i; | |
2998 | key_valid = true; | |
2999 | } | |
3000 | } | |
3001 | ||
3002 | return cur_best; | |
3003 | } | |
3004 | ||
6d816ec7 VG |
3005 | static void |
3006 | choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
718eee05 | 3007 | { |
718eee05 CZ |
3008 | unsigned slice; |
3009 | unsigned count; | |
cdb16e8f | 3010 | struct cfq_rb_root *st; |
58ff82f3 | 3011 | unsigned group_slice; |
4d2ceea4 | 3012 | enum wl_class_t original_class = cfqd->serving_wl_class; |
1fa8f6d6 | 3013 | |
718eee05 | 3014 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 3015 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 3016 | cfqd->serving_wl_class = RT_WORKLOAD; |
58ff82f3 | 3017 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 3018 | cfqd->serving_wl_class = BE_WORKLOAD; |
718eee05 | 3019 | else { |
4d2ceea4 | 3020 | cfqd->serving_wl_class = IDLE_WORKLOAD; |
718eee05 CZ |
3021 | cfqd->workload_expires = jiffies + 1; |
3022 | return; | |
3023 | } | |
3024 | ||
4d2ceea4 | 3025 | if (original_class != cfqd->serving_wl_class) |
e4ea0c16 SL |
3026 | goto new_workload; |
3027 | ||
718eee05 CZ |
3028 | /* |
3029 | * For RT and BE, we have to choose also the type | |
3030 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
3031 | * expiration time | |
3032 | */ | |
34b98d03 | 3033 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 3034 | count = st->count; |
718eee05 CZ |
3035 | |
3036 | /* | |
65b32a57 | 3037 | * check workload expiration, and that we still have other queues ready |
718eee05 | 3038 | */ |
65b32a57 | 3039 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
3040 | return; |
3041 | ||
e4ea0c16 | 3042 | new_workload: |
718eee05 | 3043 | /* otherwise select new workload type */ |
6d816ec7 | 3044 | cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg, |
4d2ceea4 | 3045 | cfqd->serving_wl_class); |
34b98d03 | 3046 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 3047 | count = st->count; |
718eee05 CZ |
3048 | |
3049 | /* | |
3050 | * the workload slice is computed as a fraction of target latency | |
3051 | * proportional to the number of queues in that workload, over | |
3052 | * all the queues in the same priority class | |
3053 | */ | |
58ff82f3 VG |
3054 | group_slice = cfq_group_slice(cfqd, cfqg); |
3055 | ||
3056 | slice = group_slice * count / | |
4d2ceea4 VG |
3057 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class], |
3058 | cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd, | |
3bf10fea | 3059 | cfqg)); |
718eee05 | 3060 | |
4d2ceea4 | 3061 | if (cfqd->serving_wl_type == ASYNC_WORKLOAD) { |
f26bd1f0 VG |
3062 | unsigned int tmp; |
3063 | ||
3064 | /* | |
3065 | * Async queues are currently system wide. Just taking | |
3066 | * proportion of queues with-in same group will lead to higher | |
3067 | * async ratio system wide as generally root group is going | |
3068 | * to have higher weight. A more accurate thing would be to | |
3069 | * calculate system wide asnc/sync ratio. | |
3070 | */ | |
5bf14c07 TM |
3071 | tmp = cfqd->cfq_target_latency * |
3072 | cfqg_busy_async_queues(cfqd, cfqg); | |
f26bd1f0 VG |
3073 | tmp = tmp/cfqd->busy_queues; |
3074 | slice = min_t(unsigned, slice, tmp); | |
3075 | ||
718eee05 CZ |
3076 | /* async workload slice is scaled down according to |
3077 | * the sync/async slice ratio. */ | |
3078 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 3079 | } else |
718eee05 CZ |
3080 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
3081 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
3082 | ||
3083 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 3084 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
3085 | cfqd->workload_expires = jiffies + slice; |
3086 | } | |
3087 | ||
1fa8f6d6 VG |
3088 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
3089 | { | |
3090 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 3091 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
3092 | |
3093 | if (RB_EMPTY_ROOT(&st->rb)) | |
3094 | return NULL; | |
25bc6b07 | 3095 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
3096 | update_min_vdisktime(st); |
3097 | return cfqg; | |
1fa8f6d6 VG |
3098 | } |
3099 | ||
cdb16e8f VG |
3100 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
3101 | { | |
1fa8f6d6 VG |
3102 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
3103 | ||
3104 | cfqd->serving_group = cfqg; | |
dae739eb VG |
3105 | |
3106 | /* Restore the workload type data */ | |
4d2ceea4 VG |
3107 | if (cfqg->saved_wl_slice) { |
3108 | cfqd->workload_expires = jiffies + cfqg->saved_wl_slice; | |
3109 | cfqd->serving_wl_type = cfqg->saved_wl_type; | |
3110 | cfqd->serving_wl_class = cfqg->saved_wl_class; | |
66ae2919 GJ |
3111 | } else |
3112 | cfqd->workload_expires = jiffies - 1; | |
3113 | ||
6d816ec7 | 3114 | choose_wl_class_and_type(cfqd, cfqg); |
cdb16e8f VG |
3115 | } |
3116 | ||
22e2c507 | 3117 | /* |
498d3aa2 JA |
3118 | * Select a queue for service. If we have a current active queue, |
3119 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 3120 | */ |
1b5ed5e1 | 3121 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 3122 | { |
a36e71f9 | 3123 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 3124 | |
22e2c507 JA |
3125 | cfqq = cfqd->active_queue; |
3126 | if (!cfqq) | |
3127 | goto new_queue; | |
1da177e4 | 3128 | |
f04a6424 VG |
3129 | if (!cfqd->rq_queued) |
3130 | return NULL; | |
c244bb50 VG |
3131 | |
3132 | /* | |
3133 | * We were waiting for group to get backlogged. Expire the queue | |
3134 | */ | |
3135 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3136 | goto expire; | |
3137 | ||
22e2c507 | 3138 | /* |
6d048f53 | 3139 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 3140 | */ |
7667aa06 VG |
3141 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
3142 | /* | |
3143 | * If slice had not expired at the completion of last request | |
3144 | * we might not have turned on wait_busy flag. Don't expire | |
3145 | * the queue yet. Allow the group to get backlogged. | |
3146 | * | |
3147 | * The very fact that we have used the slice, that means we | |
3148 | * have been idling all along on this queue and it should be | |
3149 | * ok to wait for this request to complete. | |
3150 | */ | |
82bbbf28 VG |
3151 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
3152 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
3153 | cfqq = NULL; | |
7667aa06 | 3154 | goto keep_queue; |
82bbbf28 | 3155 | } else |
80bdf0c7 | 3156 | goto check_group_idle; |
7667aa06 | 3157 | } |
1da177e4 | 3158 | |
22e2c507 | 3159 | /* |
6d048f53 JA |
3160 | * The active queue has requests and isn't expired, allow it to |
3161 | * dispatch. | |
22e2c507 | 3162 | */ |
dd67d051 | 3163 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3164 | goto keep_queue; |
6d048f53 | 3165 | |
a36e71f9 JA |
3166 | /* |
3167 | * If another queue has a request waiting within our mean seek | |
3168 | * distance, let it run. The expire code will check for close | |
3169 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 3170 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 3171 | */ |
b3b6d040 | 3172 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
3173 | if (new_cfqq) { |
3174 | if (!cfqq->new_cfqq) | |
3175 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 3176 | goto expire; |
df5fe3e8 | 3177 | } |
a36e71f9 | 3178 | |
6d048f53 JA |
3179 | /* |
3180 | * No requests pending. If the active queue still has requests in | |
3181 | * flight or is idling for a new request, allow either of these | |
3182 | * conditions to happen (or time out) before selecting a new queue. | |
3183 | */ | |
80bdf0c7 VG |
3184 | if (timer_pending(&cfqd->idle_slice_timer)) { |
3185 | cfqq = NULL; | |
3186 | goto keep_queue; | |
3187 | } | |
3188 | ||
8e1ac665 SL |
3189 | /* |
3190 | * This is a deep seek queue, but the device is much faster than | |
3191 | * the queue can deliver, don't idle | |
3192 | **/ | |
3193 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
3194 | (cfq_cfqq_slice_new(cfqq) || | |
3195 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
3196 | cfq_clear_cfqq_deep(cfqq); | |
3197 | cfq_clear_cfqq_idle_window(cfqq); | |
3198 | } | |
3199 | ||
80bdf0c7 VG |
3200 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
3201 | cfqq = NULL; | |
3202 | goto keep_queue; | |
3203 | } | |
3204 | ||
3205 | /* | |
3206 | * If group idle is enabled and there are requests dispatched from | |
3207 | * this group, wait for requests to complete. | |
3208 | */ | |
3209 | check_group_idle: | |
7700fc4f SL |
3210 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
3211 | cfqq->cfqg->dispatched && | |
3212 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
3213 | cfqq = NULL; |
3214 | goto keep_queue; | |
22e2c507 JA |
3215 | } |
3216 | ||
3b18152c | 3217 | expire: |
e5ff082e | 3218 | cfq_slice_expired(cfqd, 0); |
3b18152c | 3219 | new_queue: |
718eee05 CZ |
3220 | /* |
3221 | * Current queue expired. Check if we have to switch to a new | |
3222 | * service tree | |
3223 | */ | |
3224 | if (!new_cfqq) | |
cdb16e8f | 3225 | cfq_choose_cfqg(cfqd); |
718eee05 | 3226 | |
a36e71f9 | 3227 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 3228 | keep_queue: |
3b18152c | 3229 | return cfqq; |
22e2c507 JA |
3230 | } |
3231 | ||
febffd61 | 3232 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
3233 | { |
3234 | int dispatched = 0; | |
3235 | ||
3236 | while (cfqq->next_rq) { | |
3237 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
3238 | dispatched++; | |
3239 | } | |
3240 | ||
3241 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
3242 | |
3243 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 3244 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
3245 | return dispatched; |
3246 | } | |
3247 | ||
498d3aa2 JA |
3248 | /* |
3249 | * Drain our current requests. Used for barriers and when switching | |
3250 | * io schedulers on-the-fly. | |
3251 | */ | |
d9e7620e | 3252 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 3253 | { |
0871714e | 3254 | struct cfq_queue *cfqq; |
d9e7620e | 3255 | int dispatched = 0; |
cdb16e8f | 3256 | |
3440c49f | 3257 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 3258 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
3259 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
3260 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 3261 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 3262 | } |
1b5ed5e1 | 3263 | |
1b5ed5e1 TH |
3264 | BUG_ON(cfqd->busy_queues); |
3265 | ||
6923715a | 3266 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
3267 | return dispatched; |
3268 | } | |
3269 | ||
abc3c744 SL |
3270 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
3271 | struct cfq_queue *cfqq) | |
3272 | { | |
3273 | /* the queue hasn't finished any request, can't estimate */ | |
3274 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 3275 | return true; |
abc3c744 SL |
3276 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
3277 | cfqq->slice_end)) | |
c1e44756 | 3278 | return true; |
abc3c744 | 3279 | |
c1e44756 | 3280 | return false; |
abc3c744 SL |
3281 | } |
3282 | ||
0b182d61 | 3283 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 3284 | { |
2f5cb738 | 3285 | unsigned int max_dispatch; |
22e2c507 | 3286 | |
5ad531db JA |
3287 | /* |
3288 | * Drain async requests before we start sync IO | |
3289 | */ | |
53c583d2 | 3290 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 3291 | return false; |
5ad531db | 3292 | |
2f5cb738 JA |
3293 | /* |
3294 | * If this is an async queue and we have sync IO in flight, let it wait | |
3295 | */ | |
53c583d2 | 3296 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 3297 | return false; |
2f5cb738 | 3298 | |
abc3c744 | 3299 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
3300 | if (cfq_class_idle(cfqq)) |
3301 | max_dispatch = 1; | |
b4878f24 | 3302 | |
2f5cb738 JA |
3303 | /* |
3304 | * Does this cfqq already have too much IO in flight? | |
3305 | */ | |
3306 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 3307 | bool promote_sync = false; |
2f5cb738 JA |
3308 | /* |
3309 | * idle queue must always only have a single IO in flight | |
3310 | */ | |
3ed9a296 | 3311 | if (cfq_class_idle(cfqq)) |
0b182d61 | 3312 | return false; |
3ed9a296 | 3313 | |
ef8a41df | 3314 | /* |
c4ade94f LS |
3315 | * If there is only one sync queue |
3316 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
3317 | * queue no dispatch limit. The reason is a sync queue can |
3318 | * preempt async queue, limiting the sync queue doesn't make | |
3319 | * sense. This is useful for aiostress test. | |
3320 | */ | |
c4ade94f LS |
3321 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
3322 | promote_sync = true; | |
ef8a41df | 3323 | |
2f5cb738 JA |
3324 | /* |
3325 | * We have other queues, don't allow more IO from this one | |
3326 | */ | |
ef8a41df SL |
3327 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
3328 | !promote_sync) | |
0b182d61 | 3329 | return false; |
9ede209e | 3330 | |
365722bb | 3331 | /* |
474b18cc | 3332 | * Sole queue user, no limit |
365722bb | 3333 | */ |
ef8a41df | 3334 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
3335 | max_dispatch = -1; |
3336 | else | |
3337 | /* | |
3338 | * Normally we start throttling cfqq when cfq_quantum/2 | |
3339 | * requests have been dispatched. But we can drive | |
3340 | * deeper queue depths at the beginning of slice | |
3341 | * subjected to upper limit of cfq_quantum. | |
3342 | * */ | |
3343 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
3344 | } |
3345 | ||
3346 | /* | |
3347 | * Async queues must wait a bit before being allowed dispatch. | |
3348 | * We also ramp up the dispatch depth gradually for async IO, | |
3349 | * based on the last sync IO we serviced | |
3350 | */ | |
963b72fc | 3351 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 3352 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 3353 | unsigned int depth; |
365722bb | 3354 | |
61f0c1dc | 3355 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
3356 | if (!depth && !cfqq->dispatched) |
3357 | depth = 1; | |
8e296755 JA |
3358 | if (depth < max_dispatch) |
3359 | max_dispatch = depth; | |
2f5cb738 | 3360 | } |
3ed9a296 | 3361 | |
0b182d61 JA |
3362 | /* |
3363 | * If we're below the current max, allow a dispatch | |
3364 | */ | |
3365 | return cfqq->dispatched < max_dispatch; | |
3366 | } | |
3367 | ||
3368 | /* | |
3369 | * Dispatch a request from cfqq, moving them to the request queue | |
3370 | * dispatch list. | |
3371 | */ | |
3372 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3373 | { | |
3374 | struct request *rq; | |
3375 | ||
3376 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
3377 | ||
3378 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
3379 | return false; | |
3380 | ||
3381 | /* | |
3382 | * follow expired path, else get first next available | |
3383 | */ | |
3384 | rq = cfq_check_fifo(cfqq); | |
3385 | if (!rq) | |
3386 | rq = cfqq->next_rq; | |
3387 | ||
3388 | /* | |
3389 | * insert request into driver dispatch list | |
3390 | */ | |
3391 | cfq_dispatch_insert(cfqd->queue, rq); | |
3392 | ||
3393 | if (!cfqd->active_cic) { | |
c5869807 | 3394 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 3395 | |
c5869807 | 3396 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
3397 | cfqd->active_cic = cic; |
3398 | } | |
3399 | ||
3400 | return true; | |
3401 | } | |
3402 | ||
3403 | /* | |
3404 | * Find the cfqq that we need to service and move a request from that to the | |
3405 | * dispatch list | |
3406 | */ | |
3407 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
3408 | { | |
3409 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3410 | struct cfq_queue *cfqq; | |
3411 | ||
3412 | if (!cfqd->busy_queues) | |
3413 | return 0; | |
3414 | ||
3415 | if (unlikely(force)) | |
3416 | return cfq_forced_dispatch(cfqd); | |
3417 | ||
3418 | cfqq = cfq_select_queue(cfqd); | |
3419 | if (!cfqq) | |
8e296755 JA |
3420 | return 0; |
3421 | ||
2f5cb738 | 3422 | /* |
0b182d61 | 3423 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 3424 | */ |
0b182d61 JA |
3425 | if (!cfq_dispatch_request(cfqd, cfqq)) |
3426 | return 0; | |
3427 | ||
2f5cb738 | 3428 | cfqq->slice_dispatch++; |
b029195d | 3429 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 3430 | |
2f5cb738 JA |
3431 | /* |
3432 | * expire an async queue immediately if it has used up its slice. idle | |
3433 | * queue always expire after 1 dispatch round. | |
3434 | */ | |
3435 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
3436 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
3437 | cfq_class_idle(cfqq))) { | |
3438 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 3439 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
3440 | } |
3441 | ||
b217a903 | 3442 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 3443 | return 1; |
1da177e4 LT |
3444 | } |
3445 | ||
1da177e4 | 3446 | /* |
5e705374 JA |
3447 | * task holds one reference to the queue, dropped when task exits. each rq |
3448 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 3449 | * |
b1c35769 | 3450 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
3451 | * queue lock must be held here. |
3452 | */ | |
3453 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
3454 | { | |
22e2c507 | 3455 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 3456 | struct cfq_group *cfqg; |
22e2c507 | 3457 | |
30d7b944 | 3458 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 3459 | |
30d7b944 SL |
3460 | cfqq->ref--; |
3461 | if (cfqq->ref) | |
1da177e4 LT |
3462 | return; |
3463 | ||
7b679138 | 3464 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 3465 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 3466 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 3467 | cfqg = cfqq->cfqg; |
1da177e4 | 3468 | |
28f95cbc | 3469 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 3470 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 3471 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 3472 | } |
22e2c507 | 3473 | |
f04a6424 | 3474 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 3475 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 3476 | cfqg_put(cfqg); |
1da177e4 LT |
3477 | } |
3478 | ||
d02a2c07 | 3479 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 3480 | { |
df5fe3e8 JM |
3481 | struct cfq_queue *__cfqq, *next; |
3482 | ||
df5fe3e8 JM |
3483 | /* |
3484 | * If this queue was scheduled to merge with another queue, be | |
3485 | * sure to drop the reference taken on that queue (and others in | |
3486 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
3487 | */ | |
3488 | __cfqq = cfqq->new_cfqq; | |
3489 | while (__cfqq) { | |
3490 | if (__cfqq == cfqq) { | |
3491 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
3492 | break; | |
3493 | } | |
3494 | next = __cfqq->new_cfqq; | |
3495 | cfq_put_queue(__cfqq); | |
3496 | __cfqq = next; | |
3497 | } | |
d02a2c07 SL |
3498 | } |
3499 | ||
3500 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3501 | { | |
3502 | if (unlikely(cfqq == cfqd->active_queue)) { | |
3503 | __cfq_slice_expired(cfqd, cfqq, 0); | |
3504 | cfq_schedule_dispatch(cfqd); | |
3505 | } | |
3506 | ||
3507 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 3508 | |
89850f7e JA |
3509 | cfq_put_queue(cfqq); |
3510 | } | |
22e2c507 | 3511 | |
9b84cacd TH |
3512 | static void cfq_init_icq(struct io_cq *icq) |
3513 | { | |
3514 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
3515 | ||
3516 | cic->ttime.last_end_request = jiffies; | |
3517 | } | |
3518 | ||
c5869807 | 3519 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 3520 | { |
c5869807 | 3521 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 3522 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 3523 | |
563180a4 TH |
3524 | if (cic_to_cfqq(cic, false)) { |
3525 | cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, false)); | |
3526 | cic_set_cfqq(cic, NULL, false); | |
12a05732 AV |
3527 | } |
3528 | ||
563180a4 TH |
3529 | if (cic_to_cfqq(cic, true)) { |
3530 | cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, true)); | |
3531 | cic_set_cfqq(cic, NULL, true); | |
12a05732 | 3532 | } |
89850f7e JA |
3533 | } |
3534 | ||
abede6da | 3535 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
3536 | { |
3537 | struct task_struct *tsk = current; | |
3538 | int ioprio_class; | |
3539 | ||
3b18152c | 3540 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
3541 | return; |
3542 | ||
598971bf | 3543 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 3544 | switch (ioprio_class) { |
fe094d98 JA |
3545 | default: |
3546 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
3547 | case IOPRIO_CLASS_NONE: | |
3548 | /* | |
6d63c275 | 3549 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
3550 | */ |
3551 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 3552 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
3553 | break; |
3554 | case IOPRIO_CLASS_RT: | |
598971bf | 3555 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3556 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
3557 | break; | |
3558 | case IOPRIO_CLASS_BE: | |
598971bf | 3559 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3560 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3561 | break; | |
3562 | case IOPRIO_CLASS_IDLE: | |
3563 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
3564 | cfqq->ioprio = 7; | |
3565 | cfq_clear_cfqq_idle_window(cfqq); | |
3566 | break; | |
22e2c507 JA |
3567 | } |
3568 | ||
3569 | /* | |
3570 | * keep track of original prio settings in case we have to temporarily | |
3571 | * elevate the priority of this queue | |
3572 | */ | |
3573 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 3574 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
3575 | } |
3576 | ||
598971bf | 3577 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 3578 | { |
598971bf | 3579 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 3580 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 3581 | struct cfq_queue *cfqq; |
35e6077c | 3582 | |
598971bf TH |
3583 | /* |
3584 | * Check whether ioprio has changed. The condition may trigger | |
3585 | * spuriously on a newly created cic but there's no harm. | |
3586 | */ | |
3587 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
3588 | return; |
3589 | ||
563180a4 | 3590 | cfqq = cic_to_cfqq(cic, false); |
caaa5f9f | 3591 | if (cfqq) { |
563180a4 | 3592 | cfq_put_queue(cfqq); |
2da8de0b | 3593 | cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio); |
563180a4 | 3594 | cic_set_cfqq(cic, cfqq, false); |
22e2c507 | 3595 | } |
caaa5f9f | 3596 | |
563180a4 | 3597 | cfqq = cic_to_cfqq(cic, true); |
caaa5f9f JA |
3598 | if (cfqq) |
3599 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
3600 | |
3601 | cic->ioprio = ioprio; | |
22e2c507 JA |
3602 | } |
3603 | ||
d5036d77 | 3604 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 3605 | pid_t pid, bool is_sync) |
d5036d77 JA |
3606 | { |
3607 | RB_CLEAR_NODE(&cfqq->rb_node); | |
3608 | RB_CLEAR_NODE(&cfqq->p_node); | |
3609 | INIT_LIST_HEAD(&cfqq->fifo); | |
3610 | ||
30d7b944 | 3611 | cfqq->ref = 0; |
d5036d77 JA |
3612 | cfqq->cfqd = cfqd; |
3613 | ||
3614 | cfq_mark_cfqq_prio_changed(cfqq); | |
3615 | ||
3616 | if (is_sync) { | |
3617 | if (!cfq_class_idle(cfqq)) | |
3618 | cfq_mark_cfqq_idle_window(cfqq); | |
3619 | cfq_mark_cfqq_sync(cfqq); | |
3620 | } | |
3621 | cfqq->pid = pid; | |
3622 | } | |
3623 | ||
24610333 | 3624 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
598971bf | 3625 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 3626 | { |
bca4b914 | 3627 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
60a83707 | 3628 | struct cfq_queue *cfqq; |
f4da8072 | 3629 | uint64_t serial_nr; |
24610333 | 3630 | |
598971bf | 3631 | rcu_read_lock(); |
f4da8072 | 3632 | serial_nr = bio_blkcg(bio)->css.serial_nr; |
598971bf | 3633 | rcu_read_unlock(); |
24610333 | 3634 | |
598971bf TH |
3635 | /* |
3636 | * Check whether blkcg has changed. The condition may trigger | |
3637 | * spuriously on a newly created cic but there's no harm. | |
3638 | */ | |
f4da8072 | 3639 | if (unlikely(!cfqd) || likely(cic->blkcg_serial_nr == serial_nr)) |
598971bf | 3640 | return; |
24610333 | 3641 | |
60a83707 TH |
3642 | /* |
3643 | * Drop reference to queues. New queues will be assigned in new | |
3644 | * group upon arrival of fresh requests. | |
3645 | */ | |
3646 | cfqq = cic_to_cfqq(cic, false); | |
3647 | if (cfqq) { | |
3648 | cfq_log_cfqq(cfqd, cfqq, "changed cgroup"); | |
3649 | cic_set_cfqq(cic, NULL, false); | |
3650 | cfq_put_queue(cfqq); | |
3651 | } | |
3652 | ||
3653 | cfqq = cic_to_cfqq(cic, true); | |
3654 | if (cfqq) { | |
3655 | cfq_log_cfqq(cfqd, cfqq, "changed cgroup"); | |
3656 | cic_set_cfqq(cic, NULL, true); | |
3657 | cfq_put_queue(cfqq); | |
24610333 | 3658 | } |
598971bf | 3659 | |
f4da8072 | 3660 | cic->blkcg_serial_nr = serial_nr; |
24610333 | 3661 | } |
598971bf TH |
3662 | #else |
3663 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } | |
24610333 VG |
3664 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
3665 | ||
c2dea2d1 | 3666 | static struct cfq_queue ** |
60a83707 | 3667 | cfq_async_queue_prio(struct cfq_group *cfqg, int ioprio_class, int ioprio) |
c2dea2d1 | 3668 | { |
fe094d98 | 3669 | switch (ioprio_class) { |
c2dea2d1 | 3670 | case IOPRIO_CLASS_RT: |
60a83707 | 3671 | return &cfqg->async_cfqq[0][ioprio]; |
598971bf TH |
3672 | case IOPRIO_CLASS_NONE: |
3673 | ioprio = IOPRIO_NORM; | |
3674 | /* fall through */ | |
c2dea2d1 | 3675 | case IOPRIO_CLASS_BE: |
60a83707 | 3676 | return &cfqg->async_cfqq[1][ioprio]; |
c2dea2d1 | 3677 | case IOPRIO_CLASS_IDLE: |
60a83707 | 3678 | return &cfqg->async_idle_cfqq; |
c2dea2d1 VT |
3679 | default: |
3680 | BUG(); | |
3681 | } | |
3682 | } | |
3683 | ||
15c31be4 | 3684 | static struct cfq_queue * |
abede6da | 3685 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
2da8de0b | 3686 | struct bio *bio) |
15c31be4 | 3687 | { |
c6ce1943 JM |
3688 | int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
3689 | int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
d4aad7ff | 3690 | struct cfq_queue **async_cfqq = NULL; |
4ebc1c61 | 3691 | struct cfq_queue *cfqq; |
322731ed TH |
3692 | struct cfq_group *cfqg; |
3693 | ||
3694 | rcu_read_lock(); | |
ae118896 | 3695 | cfqg = cfq_lookup_cfqg(cfqd, bio_blkcg(bio)); |
322731ed TH |
3696 | if (!cfqg) { |
3697 | cfqq = &cfqd->oom_cfqq; | |
3698 | goto out; | |
3699 | } | |
15c31be4 | 3700 | |
c2dea2d1 | 3701 | if (!is_sync) { |
c6ce1943 JM |
3702 | if (!ioprio_valid(cic->ioprio)) { |
3703 | struct task_struct *tsk = current; | |
3704 | ioprio = task_nice_ioprio(tsk); | |
3705 | ioprio_class = task_nice_ioclass(tsk); | |
3706 | } | |
60a83707 | 3707 | async_cfqq = cfq_async_queue_prio(cfqg, ioprio_class, ioprio); |
c2dea2d1 | 3708 | cfqq = *async_cfqq; |
4ebc1c61 TH |
3709 | if (cfqq) |
3710 | goto out; | |
c2dea2d1 VT |
3711 | } |
3712 | ||
d4aad7ff TH |
3713 | cfqq = kmem_cache_alloc_node(cfq_pool, GFP_NOWAIT | __GFP_ZERO, |
3714 | cfqd->queue->node); | |
3715 | if (!cfqq) { | |
3716 | cfqq = &cfqd->oom_cfqq; | |
3717 | goto out; | |
3718 | } | |
3719 | ||
3720 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
3721 | cfq_init_prio_data(cfqq, cic); | |
3722 | cfq_link_cfqq_cfqg(cfqq, cfqg); | |
3723 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | |
15c31be4 | 3724 | |
d4aad7ff TH |
3725 | if (async_cfqq) { |
3726 | /* a new async queue is created, pin and remember */ | |
30d7b944 | 3727 | cfqq->ref++; |
c2dea2d1 | 3728 | *async_cfqq = cfqq; |
15c31be4 | 3729 | } |
4ebc1c61 | 3730 | out: |
30d7b944 | 3731 | cfqq->ref++; |
322731ed | 3732 | rcu_read_unlock(); |
15c31be4 JA |
3733 | return cfqq; |
3734 | } | |
3735 | ||
22e2c507 | 3736 | static void |
383cd721 | 3737 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 3738 | { |
383cd721 SL |
3739 | unsigned long elapsed = jiffies - ttime->last_end_request; |
3740 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 3741 | |
383cd721 SL |
3742 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
3743 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
3744 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
3745 | } | |
3746 | ||
3747 | static void | |
3748 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3749 | struct cfq_io_cq *cic) |
383cd721 | 3750 | { |
f5f2b6ce | 3751 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3752 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3753 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3754 | cfqd->cfq_slice_idle); | |
3755 | } | |
7700fc4f SL |
3756 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3757 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3758 | #endif | |
22e2c507 | 3759 | } |
1da177e4 | 3760 | |
206dc69b | 3761 | static void |
b2c18e1e | 3762 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3763 | struct request *rq) |
206dc69b | 3764 | { |
3dde36dd | 3765 | sector_t sdist = 0; |
41647e7a | 3766 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3767 | if (cfqq->last_request_pos) { |
3768 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3769 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3770 | else | |
3771 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3772 | } | |
206dc69b | 3773 | |
3dde36dd | 3774 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3775 | if (blk_queue_nonrot(cfqd->queue)) |
3776 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3777 | else | |
3778 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3779 | } |
1da177e4 | 3780 | |
22e2c507 JA |
3781 | /* |
3782 | * Disable idle window if the process thinks too long or seeks so much that | |
3783 | * it doesn't matter | |
3784 | */ | |
3785 | static void | |
3786 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3787 | struct cfq_io_cq *cic) |
22e2c507 | 3788 | { |
7b679138 | 3789 | int old_idle, enable_idle; |
1be92f2f | 3790 | |
0871714e JA |
3791 | /* |
3792 | * Don't idle for async or idle io prio class | |
3793 | */ | |
3794 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3795 | return; |
3796 | ||
c265a7f4 | 3797 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3798 | |
76280aff CZ |
3799 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3800 | cfq_mark_cfqq_deep(cfqq); | |
3801 | ||
749ef9f8 CZ |
3802 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
3803 | enable_idle = 0; | |
f6e8d01b | 3804 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
3805 | !cfqd->cfq_slice_idle || |
3806 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3807 | enable_idle = 0; |
383cd721 SL |
3808 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3809 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3810 | enable_idle = 0; |
3811 | else | |
3812 | enable_idle = 1; | |
1da177e4 LT |
3813 | } |
3814 | ||
7b679138 JA |
3815 | if (old_idle != enable_idle) { |
3816 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3817 | if (enable_idle) | |
3818 | cfq_mark_cfqq_idle_window(cfqq); | |
3819 | else | |
3820 | cfq_clear_cfqq_idle_window(cfqq); | |
3821 | } | |
22e2c507 | 3822 | } |
1da177e4 | 3823 | |
22e2c507 JA |
3824 | /* |
3825 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3826 | * no or if we aren't sure, a 1 will cause a preempt. | |
3827 | */ | |
a6151c3a | 3828 | static bool |
22e2c507 | 3829 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3830 | struct request *rq) |
22e2c507 | 3831 | { |
6d048f53 | 3832 | struct cfq_queue *cfqq; |
22e2c507 | 3833 | |
6d048f53 JA |
3834 | cfqq = cfqd->active_queue; |
3835 | if (!cfqq) | |
a6151c3a | 3836 | return false; |
22e2c507 | 3837 | |
6d048f53 | 3838 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3839 | return false; |
22e2c507 JA |
3840 | |
3841 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3842 | return true; |
1e3335de | 3843 | |
875feb63 DS |
3844 | /* |
3845 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3846 | */ | |
3847 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3848 | return false; | |
3849 | ||
374f84ac JA |
3850 | /* |
3851 | * if the new request is sync, but the currently running queue is | |
3852 | * not, let the sync request have priority. | |
3853 | */ | |
5e705374 | 3854 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3855 | return true; |
1e3335de | 3856 | |
8682e1f1 VG |
3857 | if (new_cfqq->cfqg != cfqq->cfqg) |
3858 | return false; | |
3859 | ||
3860 | if (cfq_slice_used(cfqq)) | |
3861 | return true; | |
3862 | ||
3863 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
4d2ceea4 | 3864 | if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 VG |
3865 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
3866 | new_cfqq->service_tree->count == 2 && | |
3867 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3868 | return true; | |
3869 | ||
b53d1ed7 JA |
3870 | /* |
3871 | * So both queues are sync. Let the new request get disk time if | |
3872 | * it's a metadata request and the current queue is doing regular IO. | |
3873 | */ | |
65299a3b | 3874 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
3875 | return true; |
3876 | ||
3a9a3f6c DS |
3877 | /* |
3878 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3879 | */ | |
3880 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3881 | return true; |
3a9a3f6c | 3882 | |
d2d59e18 SL |
3883 | /* An idle queue should not be idle now for some reason */ |
3884 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
3885 | return true; | |
3886 | ||
1e3335de | 3887 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3888 | return false; |
1e3335de JA |
3889 | |
3890 | /* | |
3891 | * if this request is as-good as one we would expect from the | |
3892 | * current cfqq, let it preempt | |
3893 | */ | |
e9ce335d | 3894 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3895 | return true; |
1e3335de | 3896 | |
a6151c3a | 3897 | return false; |
22e2c507 JA |
3898 | } |
3899 | ||
3900 | /* | |
3901 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3902 | * let it have half of its nominal slice. | |
3903 | */ | |
3904 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3905 | { | |
df0793ab SL |
3906 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
3907 | ||
7b679138 | 3908 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 3909 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3910 | |
f8ae6e3e SL |
3911 | /* |
3912 | * workload type is changed, don't save slice, otherwise preempt | |
3913 | * doesn't happen | |
3914 | */ | |
df0793ab | 3915 | if (old_type != cfqq_type(cfqq)) |
4d2ceea4 | 3916 | cfqq->cfqg->saved_wl_slice = 0; |
f8ae6e3e | 3917 | |
bf572256 JA |
3918 | /* |
3919 | * Put the new queue at the front of the of the current list, | |
3920 | * so we know that it will be selected next. | |
3921 | */ | |
3922 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3923 | |
3924 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3925 | |
62a37f6b JT |
3926 | cfqq->slice_end = 0; |
3927 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3928 | } |
3929 | ||
22e2c507 | 3930 | /* |
5e705374 | 3931 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3932 | * something we should do about it |
3933 | */ | |
3934 | static void | |
5e705374 JA |
3935 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3936 | struct request *rq) | |
22e2c507 | 3937 | { |
c5869807 | 3938 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3939 | |
45333d5a | 3940 | cfqd->rq_queued++; |
65299a3b CH |
3941 | if (rq->cmd_flags & REQ_PRIO) |
3942 | cfqq->prio_pending++; | |
374f84ac | 3943 | |
383cd721 | 3944 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3945 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3946 | cfq_update_idle_window(cfqd, cfqq, cic); |
3947 | ||
b2c18e1e | 3948 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3949 | |
3950 | if (cfqq == cfqd->active_queue) { | |
3951 | /* | |
b029195d JA |
3952 | * Remember that we saw a request from this process, but |
3953 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3954 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3955 | * and merging. If the request is already larger than a single |
3956 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3957 | * merging is already done. Ditto for a busy system that |
3958 | * has other work pending, don't risk delaying until the | |
3959 | * idle timer unplug to continue working. | |
22e2c507 | 3960 | */ |
d6ceb25e | 3961 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3962 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3963 | cfqd->busy_queues > 1) { | |
812df48d | 3964 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3965 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3966 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3967 | } else { |
155fead9 | 3968 | cfqg_stats_update_idle_time(cfqq->cfqg); |
bf791937 | 3969 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3970 | } |
d6ceb25e | 3971 | } |
5e705374 | 3972 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3973 | /* |
3974 | * not the active queue - expire current slice if it is | |
3975 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3976 | * has some old slice time left and is of higher priority or |
3977 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3978 | */ |
3979 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3980 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3981 | } |
1da177e4 LT |
3982 | } |
3983 | ||
165125e1 | 3984 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3985 | { |
b4878f24 | 3986 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3987 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3988 | |
7b679138 | 3989 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 3990 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 3991 | |
8b4922d3 | 3992 | rq->fifo_time = jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]; |
22e2c507 | 3993 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3994 | cfq_add_rq_rb(rq); |
155fead9 TH |
3995 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
3996 | rq->cmd_flags); | |
5e705374 | 3997 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3998 | } |
3999 | ||
45333d5a AC |
4000 | /* |
4001 | * Update hw_tag based on peak queue depth over 50 samples under | |
4002 | * sufficient load. | |
4003 | */ | |
4004 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
4005 | { | |
1a1238a7 SL |
4006 | struct cfq_queue *cfqq = cfqd->active_queue; |
4007 | ||
53c583d2 CZ |
4008 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
4009 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
4010 | |
4011 | if (cfqd->hw_tag == 1) | |
4012 | return; | |
45333d5a AC |
4013 | |
4014 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 4015 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
4016 | return; |
4017 | ||
1a1238a7 SL |
4018 | /* |
4019 | * If active queue hasn't enough requests and can idle, cfq might not | |
4020 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
4021 | * case | |
4022 | */ | |
4023 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
4024 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 4025 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
4026 | return; |
4027 | ||
45333d5a AC |
4028 | if (cfqd->hw_tag_samples++ < 50) |
4029 | return; | |
4030 | ||
e459dd08 | 4031 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
4032 | cfqd->hw_tag = 1; |
4033 | else | |
4034 | cfqd->hw_tag = 0; | |
45333d5a AC |
4035 | } |
4036 | ||
7667aa06 VG |
4037 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
4038 | { | |
c5869807 | 4039 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 4040 | |
02a8f01b JT |
4041 | /* If the queue already has requests, don't wait */ |
4042 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
4043 | return false; | |
4044 | ||
7667aa06 VG |
4045 | /* If there are other queues in the group, don't wait */ |
4046 | if (cfqq->cfqg->nr_cfqq > 1) | |
4047 | return false; | |
4048 | ||
7700fc4f SL |
4049 | /* the only queue in the group, but think time is big */ |
4050 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
4051 | return false; | |
4052 | ||
7667aa06 VG |
4053 | if (cfq_slice_used(cfqq)) |
4054 | return true; | |
4055 | ||
4056 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
4057 | if (cic && sample_valid(cic->ttime.ttime_samples) |
4058 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
4059 | return true; |
4060 | ||
4061 | /* | |
4062 | * If think times is less than a jiffy than ttime_mean=0 and above | |
4063 | * will not be true. It might happen that slice has not expired yet | |
4064 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
4065 | * case where think time is less than a jiffy, mark the queue wait | |
4066 | * busy if only 1 jiffy is left in the slice. | |
4067 | */ | |
4068 | if (cfqq->slice_end - jiffies == 1) | |
4069 | return true; | |
4070 | ||
4071 | return false; | |
4072 | } | |
4073 | ||
165125e1 | 4074 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 4075 | { |
5e705374 | 4076 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 4077 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 4078 | const int sync = rq_is_sync(rq); |
b4878f24 | 4079 | unsigned long now; |
1da177e4 | 4080 | |
b4878f24 | 4081 | now = jiffies; |
33659ebb CH |
4082 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
4083 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 4084 | |
45333d5a AC |
4085 | cfq_update_hw_tag(cfqd); |
4086 | ||
53c583d2 | 4087 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 4088 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 4089 | cfqd->rq_in_driver--; |
6d048f53 | 4090 | cfqq->dispatched--; |
80bdf0c7 | 4091 | (RQ_CFQG(rq))->dispatched--; |
155fead9 TH |
4092 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
4093 | rq_io_start_time_ns(rq), rq->cmd_flags); | |
1da177e4 | 4094 | |
53c583d2 | 4095 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 4096 | |
365722bb | 4097 | if (sync) { |
34b98d03 | 4098 | struct cfq_rb_root *st; |
f5f2b6ce | 4099 | |
383cd721 | 4100 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
4101 | |
4102 | if (cfq_cfqq_on_rr(cfqq)) | |
34b98d03 | 4103 | st = cfqq->service_tree; |
f5f2b6ce | 4104 | else |
34b98d03 VG |
4105 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), |
4106 | cfqq_type(cfqq)); | |
4107 | ||
4108 | st->ttime.last_end_request = now; | |
573412b2 CZ |
4109 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
4110 | cfqd->last_delayed_sync = now; | |
365722bb | 4111 | } |
caaa5f9f | 4112 | |
7700fc4f SL |
4113 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4114 | cfqq->cfqg->ttime.last_end_request = now; | |
4115 | #endif | |
4116 | ||
caaa5f9f JA |
4117 | /* |
4118 | * If this is the active queue, check if it needs to be expired, | |
4119 | * or if we want to idle in case it has no pending requests. | |
4120 | */ | |
4121 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
4122 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
4123 | ||
44f7c160 JA |
4124 | if (cfq_cfqq_slice_new(cfqq)) { |
4125 | cfq_set_prio_slice(cfqd, cfqq); | |
4126 | cfq_clear_cfqq_slice_new(cfqq); | |
4127 | } | |
f75edf2d VG |
4128 | |
4129 | /* | |
7667aa06 VG |
4130 | * Should we wait for next request to come in before we expire |
4131 | * the queue. | |
f75edf2d | 4132 | */ |
7667aa06 | 4133 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
4134 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
4135 | if (!cfqd->cfq_slice_idle) | |
4136 | extend_sl = cfqd->cfq_group_idle; | |
4137 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 4138 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 4139 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
4140 | } |
4141 | ||
a36e71f9 | 4142 | /* |
8e550632 CZ |
4143 | * Idling is not enabled on: |
4144 | * - expired queues | |
4145 | * - idle-priority queues | |
4146 | * - async queues | |
4147 | * - queues with still some requests queued | |
4148 | * - when there is a close cooperator | |
a36e71f9 | 4149 | */ |
0871714e | 4150 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 4151 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
4152 | else if (sync && cfqq_empty && |
4153 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 4154 | cfq_arm_slice_timer(cfqd); |
8e550632 | 4155 | } |
caaa5f9f | 4156 | } |
6d048f53 | 4157 | |
53c583d2 | 4158 | if (!cfqd->rq_in_driver) |
23e018a1 | 4159 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
4160 | } |
4161 | ||
89850f7e | 4162 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 4163 | { |
1b379d8d | 4164 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 4165 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 4166 | return ELV_MQUEUE_MUST; |
3b18152c | 4167 | } |
1da177e4 | 4168 | |
22e2c507 | 4169 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
4170 | } |
4171 | ||
165125e1 | 4172 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
4173 | { |
4174 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
4175 | struct task_struct *tsk = current; | |
c5869807 | 4176 | struct cfq_io_cq *cic; |
22e2c507 JA |
4177 | struct cfq_queue *cfqq; |
4178 | ||
4179 | /* | |
4180 | * don't force setup of a queue from here, as a call to may_queue | |
4181 | * does not necessarily imply that a request actually will be queued. | |
4182 | * so just lookup a possibly existing queue, or return 'may queue' | |
4183 | * if that fails | |
4184 | */ | |
4ac845a2 | 4185 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
4186 | if (!cic) |
4187 | return ELV_MQUEUE_MAY; | |
4188 | ||
b0b78f81 | 4189 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 4190 | if (cfqq) { |
abede6da | 4191 | cfq_init_prio_data(cfqq, cic); |
22e2c507 | 4192 | |
89850f7e | 4193 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
4194 | } |
4195 | ||
4196 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
4197 | } |
4198 | ||
1da177e4 LT |
4199 | /* |
4200 | * queue lock held here | |
4201 | */ | |
bb37b94c | 4202 | static void cfq_put_request(struct request *rq) |
1da177e4 | 4203 | { |
5e705374 | 4204 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 4205 | |
5e705374 | 4206 | if (cfqq) { |
22e2c507 | 4207 | const int rw = rq_data_dir(rq); |
1da177e4 | 4208 | |
22e2c507 JA |
4209 | BUG_ON(!cfqq->allocated[rw]); |
4210 | cfqq->allocated[rw]--; | |
1da177e4 | 4211 | |
7f1dc8a2 | 4212 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 4213 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
4214 | rq->elv.priv[0] = NULL; |
4215 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 4216 | |
1da177e4 LT |
4217 | cfq_put_queue(cfqq); |
4218 | } | |
4219 | } | |
4220 | ||
df5fe3e8 | 4221 | static struct cfq_queue * |
c5869807 | 4222 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
4223 | struct cfq_queue *cfqq) |
4224 | { | |
4225 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
4226 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 4227 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
4228 | cfq_put_queue(cfqq); |
4229 | return cic_to_cfqq(cic, 1); | |
4230 | } | |
4231 | ||
e6c5bc73 JM |
4232 | /* |
4233 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
4234 | * was the last process referring to said cfqq. | |
4235 | */ | |
4236 | static struct cfq_queue * | |
c5869807 | 4237 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
4238 | { |
4239 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
4240 | cfqq->pid = current->pid; |
4241 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 4242 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
4243 | return cfqq; |
4244 | } | |
4245 | ||
4246 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
4247 | |
4248 | cfq_put_cooperator(cfqq); | |
4249 | ||
e6c5bc73 JM |
4250 | cfq_put_queue(cfqq); |
4251 | return NULL; | |
4252 | } | |
1da177e4 | 4253 | /* |
22e2c507 | 4254 | * Allocate cfq data structures associated with this request. |
1da177e4 | 4255 | */ |
22e2c507 | 4256 | static int |
852c788f TH |
4257 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
4258 | gfp_t gfp_mask) | |
1da177e4 LT |
4259 | { |
4260 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 4261 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 4262 | const int rw = rq_data_dir(rq); |
a6151c3a | 4263 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 4264 | struct cfq_queue *cfqq; |
1da177e4 | 4265 | |
216284c3 | 4266 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 4267 | |
598971bf TH |
4268 | check_ioprio_changed(cic, bio); |
4269 | check_blkcg_changed(cic, bio); | |
e6c5bc73 | 4270 | new_queue: |
91fac317 | 4271 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 4272 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
bce6133b TH |
4273 | if (cfqq) |
4274 | cfq_put_queue(cfqq); | |
2da8de0b | 4275 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio); |
91fac317 | 4276 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 4277 | } else { |
e6c5bc73 JM |
4278 | /* |
4279 | * If the queue was seeky for too long, break it apart. | |
4280 | */ | |
ae54abed | 4281 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
4282 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
4283 | cfqq = split_cfqq(cic, cfqq); | |
4284 | if (!cfqq) | |
4285 | goto new_queue; | |
4286 | } | |
4287 | ||
df5fe3e8 JM |
4288 | /* |
4289 | * Check to see if this queue is scheduled to merge with | |
4290 | * another, closely cooperating queue. The merging of | |
4291 | * queues happens here as it must be done in process context. | |
4292 | * The reference on new_cfqq was taken in merge_cfqqs. | |
4293 | */ | |
4294 | if (cfqq->new_cfqq) | |
4295 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 4296 | } |
1da177e4 LT |
4297 | |
4298 | cfqq->allocated[rw]++; | |
1da177e4 | 4299 | |
6fae9c25 | 4300 | cfqq->ref++; |
eb7d8c07 | 4301 | cfqg_get(cfqq->cfqg); |
a612fddf | 4302 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 4303 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 4304 | spin_unlock_irq(q->queue_lock); |
5e705374 | 4305 | return 0; |
1da177e4 LT |
4306 | } |
4307 | ||
65f27f38 | 4308 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 4309 | { |
65f27f38 | 4310 | struct cfq_data *cfqd = |
23e018a1 | 4311 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 4312 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4313 | |
40bb54d1 | 4314 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 4315 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 4316 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
4317 | } |
4318 | ||
4319 | /* | |
4320 | * Timer running if the active_queue is currently idling inside its time slice | |
4321 | */ | |
4322 | static void cfq_idle_slice_timer(unsigned long data) | |
4323 | { | |
4324 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
4325 | struct cfq_queue *cfqq; | |
4326 | unsigned long flags; | |
3c6bd2f8 | 4327 | int timed_out = 1; |
22e2c507 | 4328 | |
7b679138 JA |
4329 | cfq_log(cfqd, "idle timer fired"); |
4330 | ||
22e2c507 JA |
4331 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
4332 | ||
fe094d98 JA |
4333 | cfqq = cfqd->active_queue; |
4334 | if (cfqq) { | |
3c6bd2f8 JA |
4335 | timed_out = 0; |
4336 | ||
b029195d JA |
4337 | /* |
4338 | * We saw a request before the queue expired, let it through | |
4339 | */ | |
4340 | if (cfq_cfqq_must_dispatch(cfqq)) | |
4341 | goto out_kick; | |
4342 | ||
22e2c507 JA |
4343 | /* |
4344 | * expired | |
4345 | */ | |
44f7c160 | 4346 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
4347 | goto expire; |
4348 | ||
4349 | /* | |
4350 | * only expire and reinvoke request handler, if there are | |
4351 | * other queues with pending requests | |
4352 | */ | |
caaa5f9f | 4353 | if (!cfqd->busy_queues) |
22e2c507 | 4354 | goto out_cont; |
22e2c507 JA |
4355 | |
4356 | /* | |
4357 | * not expired and it has a request pending, let it dispatch | |
4358 | */ | |
75e50984 | 4359 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 4360 | goto out_kick; |
76280aff CZ |
4361 | |
4362 | /* | |
4363 | * Queue depth flag is reset only when the idle didn't succeed | |
4364 | */ | |
4365 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
4366 | } |
4367 | expire: | |
e5ff082e | 4368 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 4369 | out_kick: |
23e018a1 | 4370 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
4371 | out_cont: |
4372 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
4373 | } | |
4374 | ||
3b18152c JA |
4375 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
4376 | { | |
4377 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 4378 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 4379 | } |
22e2c507 | 4380 | |
b374d18a | 4381 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 4382 | { |
22e2c507 | 4383 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 4384 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4385 | |
3b18152c | 4386 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 4387 | |
d9ff4187 | 4388 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 4389 | |
d9ff4187 | 4390 | if (cfqd->active_queue) |
e5ff082e | 4391 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 4392 | |
03aa264a TH |
4393 | spin_unlock_irq(q->queue_lock); |
4394 | ||
a90d742e AV |
4395 | cfq_shutdown_timer_wq(cfqd); |
4396 | ||
ffea73fc TH |
4397 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4398 | blkcg_deactivate_policy(q, &blkcg_policy_cfq); | |
4399 | #else | |
f51b802c | 4400 | kfree(cfqd->root_group); |
2abae55f | 4401 | #endif |
56edf7d7 | 4402 | kfree(cfqd); |
1da177e4 LT |
4403 | } |
4404 | ||
d50235b7 | 4405 | static int cfq_init_queue(struct request_queue *q, struct elevator_type *e) |
1da177e4 LT |
4406 | { |
4407 | struct cfq_data *cfqd; | |
3c798398 | 4408 | struct blkcg_gq *blkg __maybe_unused; |
a2b1693b | 4409 | int i, ret; |
d50235b7 JM |
4410 | struct elevator_queue *eq; |
4411 | ||
4412 | eq = elevator_alloc(q, e); | |
4413 | if (!eq) | |
4414 | return -ENOMEM; | |
1da177e4 | 4415 | |
c1b511eb | 4416 | cfqd = kzalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node); |
d50235b7 JM |
4417 | if (!cfqd) { |
4418 | kobject_put(&eq->kobj); | |
b2fab5ac | 4419 | return -ENOMEM; |
d50235b7 JM |
4420 | } |
4421 | eq->elevator_data = cfqd; | |
80b15c73 | 4422 | |
f51b802c | 4423 | cfqd->queue = q; |
d50235b7 JM |
4424 | spin_lock_irq(q->queue_lock); |
4425 | q->elevator = eq; | |
4426 | spin_unlock_irq(q->queue_lock); | |
f51b802c | 4427 | |
1fa8f6d6 VG |
4428 | /* Init root service tree */ |
4429 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
4430 | ||
f51b802c | 4431 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 4432 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4433 | ret = blkcg_activate_policy(q, &blkcg_policy_cfq); |
a2b1693b TH |
4434 | if (ret) |
4435 | goto out_free; | |
f51b802c | 4436 | |
a2b1693b | 4437 | cfqd->root_group = blkg_to_cfqg(q->root_blkg); |
f51b802c | 4438 | #else |
a2b1693b | 4439 | ret = -ENOMEM; |
f51b802c TH |
4440 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
4441 | GFP_KERNEL, cfqd->queue->node); | |
a2b1693b TH |
4442 | if (!cfqd->root_group) |
4443 | goto out_free; | |
5624a4e4 | 4444 | |
a2b1693b TH |
4445 | cfq_init_cfqg_base(cfqd->root_group); |
4446 | #endif | |
3381cb8d | 4447 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_DEFAULT; |
e71357e1 | 4448 | cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_DEFAULT; |
5624a4e4 | 4449 | |
26a2ac00 JA |
4450 | /* |
4451 | * Not strictly needed (since RB_ROOT just clears the node and we | |
4452 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
4453 | * to add magic to the rb code | |
4454 | */ | |
4455 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
4456 | cfqd->prio_trees[i] = RB_ROOT; | |
4457 | ||
6118b70b | 4458 | /* |
d4aad7ff | 4459 | * Our fallback cfqq if cfq_get_queue() runs into OOM issues. |
6118b70b | 4460 | * Grab a permanent reference to it, so that the normal code flow |
f51b802c TH |
4461 | * will not attempt to free it. oom_cfqq is linked to root_group |
4462 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
4463 | * the reference from linking right away. | |
6118b70b JA |
4464 | */ |
4465 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 4466 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
4467 | |
4468 | spin_lock_irq(q->queue_lock); | |
f51b802c | 4469 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 4470 | cfqg_put(cfqd->root_group); |
1adaf3dd | 4471 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 4472 | |
22e2c507 JA |
4473 | init_timer(&cfqd->idle_slice_timer); |
4474 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
4475 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
4476 | ||
23e018a1 | 4477 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 4478 | |
1da177e4 | 4479 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
4480 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
4481 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
4482 | cfqd->cfq_back_max = cfq_back_max; |
4483 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
4484 | cfqd->cfq_slice[0] = cfq_slice_async; |
4485 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
5bf14c07 | 4486 | cfqd->cfq_target_latency = cfq_target_latency; |
22e2c507 | 4487 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
0bb97947 | 4488 | cfqd->cfq_slice_idle = cfq_slice_idle; |
80bdf0c7 | 4489 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 4490 | cfqd->cfq_latency = 1; |
e459dd08 | 4491 | cfqd->hw_tag = -1; |
edc71131 CZ |
4492 | /* |
4493 | * we optimistically start assuming sync ops weren't delayed in last | |
4494 | * second, in order to have larger depth for async operations. | |
4495 | */ | |
573412b2 | 4496 | cfqd->last_delayed_sync = jiffies - HZ; |
b2fab5ac | 4497 | return 0; |
a2b1693b TH |
4498 | |
4499 | out_free: | |
4500 | kfree(cfqd); | |
d50235b7 | 4501 | kobject_put(&eq->kobj); |
a2b1693b | 4502 | return ret; |
1da177e4 LT |
4503 | } |
4504 | ||
0bb97947 JA |
4505 | static void cfq_registered_queue(struct request_queue *q) |
4506 | { | |
4507 | struct elevator_queue *e = q->elevator; | |
4508 | struct cfq_data *cfqd = e->elevator_data; | |
4509 | ||
4510 | /* | |
4511 | * Default to IOPS mode with no idling for SSDs | |
4512 | */ | |
4513 | if (blk_queue_nonrot(q)) | |
4514 | cfqd->cfq_slice_idle = 0; | |
4515 | } | |
4516 | ||
1da177e4 LT |
4517 | /* |
4518 | * sysfs parts below --> | |
4519 | */ | |
1da177e4 LT |
4520 | static ssize_t |
4521 | cfq_var_show(unsigned int var, char *page) | |
4522 | { | |
176167ad | 4523 | return sprintf(page, "%u\n", var); |
1da177e4 LT |
4524 | } |
4525 | ||
4526 | static ssize_t | |
4527 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
4528 | { | |
4529 | char *p = (char *) page; | |
4530 | ||
4531 | *var = simple_strtoul(p, &p, 10); | |
4532 | return count; | |
4533 | } | |
4534 | ||
1da177e4 | 4535 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 4536 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 4537 | { \ |
3d1ab40f | 4538 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4539 | unsigned int __data = __VAR; \ |
4540 | if (__CONV) \ | |
4541 | __data = jiffies_to_msecs(__data); \ | |
4542 | return cfq_var_show(__data, (page)); \ | |
4543 | } | |
4544 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
4545 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
4546 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
4547 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
4548 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 4549 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 4550 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
4551 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4552 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4553 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4554 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
5bf14c07 | 4555 | SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1); |
1da177e4 LT |
4556 | #undef SHOW_FUNCTION |
4557 | ||
4558 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 4559 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4560 | { \ |
3d1ab40f | 4561 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4562 | unsigned int __data; \ |
4563 | int ret = cfq_var_store(&__data, (page), count); \ | |
4564 | if (__data < (MIN)) \ | |
4565 | __data = (MIN); \ | |
4566 | else if (__data > (MAX)) \ | |
4567 | __data = (MAX); \ | |
4568 | if (__CONV) \ | |
4569 | *(__PTR) = msecs_to_jiffies(__data); \ | |
4570 | else \ | |
4571 | *(__PTR) = __data; \ | |
4572 | return ret; \ | |
4573 | } | |
4574 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4575 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4576 | UINT_MAX, 1); | |
4577 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4578 | UINT_MAX, 1); | |
e572ec7e | 4579 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4580 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4581 | UINT_MAX, 0); | |
22e2c507 | 4582 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4583 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4584 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4585 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4586 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4587 | UINT_MAX, 0); | |
963b72fc | 4588 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
5bf14c07 | 4589 | STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1); |
1da177e4 LT |
4590 | #undef STORE_FUNCTION |
4591 | ||
e572ec7e AV |
4592 | #define CFQ_ATTR(name) \ |
4593 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4594 | ||
4595 | static struct elv_fs_entry cfq_attrs[] = { | |
4596 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4597 | CFQ_ATTR(fifo_expire_sync), |
4598 | CFQ_ATTR(fifo_expire_async), | |
4599 | CFQ_ATTR(back_seek_max), | |
4600 | CFQ_ATTR(back_seek_penalty), | |
4601 | CFQ_ATTR(slice_sync), | |
4602 | CFQ_ATTR(slice_async), | |
4603 | CFQ_ATTR(slice_async_rq), | |
4604 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 4605 | CFQ_ATTR(group_idle), |
963b72fc | 4606 | CFQ_ATTR(low_latency), |
5bf14c07 | 4607 | CFQ_ATTR(target_latency), |
e572ec7e | 4608 | __ATTR_NULL |
1da177e4 LT |
4609 | }; |
4610 | ||
1da177e4 LT |
4611 | static struct elevator_type iosched_cfq = { |
4612 | .ops = { | |
4613 | .elevator_merge_fn = cfq_merge, | |
4614 | .elevator_merged_fn = cfq_merged_request, | |
4615 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 4616 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 4617 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4618 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4619 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4620 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4621 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4622 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4623 | .elevator_former_req_fn = elv_rb_former_request, |
4624 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 4625 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 4626 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
4627 | .elevator_set_req_fn = cfq_set_request, |
4628 | .elevator_put_req_fn = cfq_put_request, | |
4629 | .elevator_may_queue_fn = cfq_may_queue, | |
4630 | .elevator_init_fn = cfq_init_queue, | |
4631 | .elevator_exit_fn = cfq_exit_queue, | |
0bb97947 | 4632 | .elevator_registered_fn = cfq_registered_queue, |
1da177e4 | 4633 | }, |
3d3c2379 TH |
4634 | .icq_size = sizeof(struct cfq_io_cq), |
4635 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 4636 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 4637 | .elevator_name = "cfq", |
1da177e4 LT |
4638 | .elevator_owner = THIS_MODULE, |
4639 | }; | |
4640 | ||
3e252066 | 4641 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4642 | static struct blkcg_policy blkcg_policy_cfq = { |
f9fcc2d3 TH |
4643 | .cftypes = cfq_blkcg_files, |
4644 | ||
e4a9bde9 | 4645 | .cpd_alloc_fn = cfq_cpd_alloc, |
e48453c3 | 4646 | .cpd_init_fn = cfq_cpd_init, |
e4a9bde9 TH |
4647 | .cpd_free_fn = cfq_cpd_free, |
4648 | ||
001bea73 | 4649 | .pd_alloc_fn = cfq_pd_alloc, |
f9fcc2d3 | 4650 | .pd_init_fn = cfq_pd_init, |
0b39920b | 4651 | .pd_offline_fn = cfq_pd_offline, |
001bea73 | 4652 | .pd_free_fn = cfq_pd_free, |
f9fcc2d3 | 4653 | .pd_reset_stats_fn = cfq_pd_reset_stats, |
3e252066 | 4654 | }; |
3e252066 VG |
4655 | #endif |
4656 | ||
1da177e4 LT |
4657 | static int __init cfq_init(void) |
4658 | { | |
3d3c2379 TH |
4659 | int ret; |
4660 | ||
22e2c507 JA |
4661 | /* |
4662 | * could be 0 on HZ < 1000 setups | |
4663 | */ | |
4664 | if (!cfq_slice_async) | |
4665 | cfq_slice_async = 1; | |
4666 | if (!cfq_slice_idle) | |
4667 | cfq_slice_idle = 1; | |
4668 | ||
80bdf0c7 VG |
4669 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4670 | if (!cfq_group_idle) | |
4671 | cfq_group_idle = 1; | |
8bd435b3 | 4672 | |
3c798398 | 4673 | ret = blkcg_policy_register(&blkcg_policy_cfq); |
8bd435b3 TH |
4674 | if (ret) |
4675 | return ret; | |
ffea73fc TH |
4676 | #else |
4677 | cfq_group_idle = 0; | |
4678 | #endif | |
8bd435b3 | 4679 | |
fd794956 | 4680 | ret = -ENOMEM; |
3d3c2379 TH |
4681 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4682 | if (!cfq_pool) | |
8bd435b3 | 4683 | goto err_pol_unreg; |
1da177e4 | 4684 | |
3d3c2379 | 4685 | ret = elv_register(&iosched_cfq); |
8bd435b3 TH |
4686 | if (ret) |
4687 | goto err_free_pool; | |
3d3c2379 | 4688 | |
2fdd82bd | 4689 | return 0; |
8bd435b3 TH |
4690 | |
4691 | err_free_pool: | |
4692 | kmem_cache_destroy(cfq_pool); | |
4693 | err_pol_unreg: | |
ffea73fc | 4694 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4695 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4696 | #endif |
8bd435b3 | 4697 | return ret; |
1da177e4 LT |
4698 | } |
4699 | ||
4700 | static void __exit cfq_exit(void) | |
4701 | { | |
ffea73fc | 4702 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4703 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4704 | #endif |
1da177e4 | 4705 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4706 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4707 | } |
4708 | ||
4709 | module_init(cfq_init); | |
4710 | module_exit(cfq_exit); | |
4711 | ||
4712 | MODULE_AUTHOR("Jens Axboe"); | |
4713 | MODULE_LICENSE("GPL"); | |
4714 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |