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> |
e6017571 | 11 | #include <linux/sched/clock.h> |
1cc9be68 AV |
12 | #include <linux/blkdev.h> |
13 | #include <linux/elevator.h> | |
9a7f38c4 | 14 | #include <linux/ktime.h> |
1da177e4 | 15 | #include <linux/rbtree.h> |
22e2c507 | 16 | #include <linux/ioprio.h> |
7b679138 | 17 | #include <linux/blktrace_api.h> |
eea8f41c | 18 | #include <linux/blk-cgroup.h> |
6e736be7 | 19 | #include "blk.h" |
87760e5e | 20 | #include "blk-wbt.h" |
1da177e4 LT |
21 | |
22 | /* | |
23 | * tunables | |
24 | */ | |
fe094d98 | 25 | /* max queue in one round of service */ |
abc3c744 | 26 | static const int cfq_quantum = 8; |
9a7f38c4 | 27 | static const u64 cfq_fifo_expire[2] = { NSEC_PER_SEC / 4, NSEC_PER_SEC / 8 }; |
fe094d98 JA |
28 | /* maximum backwards seek, in KiB */ |
29 | static const int cfq_back_max = 16 * 1024; | |
30 | /* penalty of a backwards seek */ | |
31 | static const int cfq_back_penalty = 2; | |
9a7f38c4 JM |
32 | static const u64 cfq_slice_sync = NSEC_PER_SEC / 10; |
33 | static u64 cfq_slice_async = NSEC_PER_SEC / 25; | |
64100099 | 34 | static const int cfq_slice_async_rq = 2; |
9a7f38c4 JM |
35 | static u64 cfq_slice_idle = NSEC_PER_SEC / 125; |
36 | static u64 cfq_group_idle = NSEC_PER_SEC / 125; | |
37 | static const u64 cfq_target_latency = (u64)NSEC_PER_SEC * 3/10; /* 300 ms */ | |
5db5d642 | 38 | static const int cfq_hist_divisor = 4; |
22e2c507 | 39 | |
d9e7620e | 40 | /* |
5be6b756 | 41 | * offset from end of queue service tree for idle class |
d9e7620e | 42 | */ |
9a7f38c4 | 43 | #define CFQ_IDLE_DELAY (NSEC_PER_SEC / 5) |
5be6b756 HT |
44 | /* offset from end of group service tree under time slice mode */ |
45 | #define CFQ_SLICE_MODE_GROUP_DELAY (NSEC_PER_SEC / 5) | |
46 | /* offset from end of group service under IOPS mode */ | |
47 | #define CFQ_IOPS_MODE_GROUP_DELAY (HZ / 5) | |
d9e7620e JA |
48 | |
49 | /* | |
50 | * below this threshold, we consider thinktime immediate | |
51 | */ | |
9a7f38c4 | 52 | #define CFQ_MIN_TT (2 * NSEC_PER_SEC / HZ) |
d9e7620e | 53 | |
22e2c507 | 54 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 55 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 56 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 57 | |
3dde36dd | 58 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 59 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 60 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 61 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 62 | |
a612fddf TH |
63 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
64 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
65 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 66 | |
e18b890b | 67 | static struct kmem_cache *cfq_pool; |
1da177e4 | 68 | |
22e2c507 JA |
69 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
70 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
71 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
72 | ||
206dc69b | 73 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 74 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 75 | |
e48453c3 | 76 | /* blkio-related constants */ |
3ecca629 TH |
77 | #define CFQ_WEIGHT_LEGACY_MIN 10 |
78 | #define CFQ_WEIGHT_LEGACY_DFL 500 | |
79 | #define CFQ_WEIGHT_LEGACY_MAX 1000 | |
e48453c3 | 80 | |
c5869807 | 81 | struct cfq_ttime { |
9a7f38c4 | 82 | u64 last_end_request; |
c5869807 | 83 | |
9a7f38c4 JM |
84 | u64 ttime_total; |
85 | u64 ttime_mean; | |
c5869807 | 86 | unsigned long ttime_samples; |
c5869807 TH |
87 | }; |
88 | ||
cc09e299 JA |
89 | /* |
90 | * Most of our rbtree usage is for sorting with min extraction, so | |
91 | * if we cache the leftmost node we don't have to walk down the tree | |
92 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
93 | * move this into the elevator for the rq sorting as well. | |
94 | */ | |
95 | struct cfq_rb_root { | |
09663c86 | 96 | struct rb_root_cached rb; |
f0f1a45f | 97 | struct rb_node *rb_rightmost; |
aa6f6a3d | 98 | unsigned count; |
1fa8f6d6 | 99 | u64 min_vdisktime; |
f5f2b6ce | 100 | struct cfq_ttime ttime; |
cc09e299 | 101 | }; |
09663c86 | 102 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT_CACHED, \ |
f0f1a45f | 103 | .rb_rightmost = NULL, \ |
9a7f38c4 | 104 | .ttime = {.last_end_request = ktime_get_ns(),},} |
cc09e299 | 105 | |
6118b70b JA |
106 | /* |
107 | * Per process-grouping structure | |
108 | */ | |
109 | struct cfq_queue { | |
110 | /* reference count */ | |
30d7b944 | 111 | int ref; |
6118b70b JA |
112 | /* various state flags, see below */ |
113 | unsigned int flags; | |
114 | /* parent cfq_data */ | |
115 | struct cfq_data *cfqd; | |
116 | /* service_tree member */ | |
117 | struct rb_node rb_node; | |
118 | /* service_tree key */ | |
9a7f38c4 | 119 | u64 rb_key; |
6118b70b JA |
120 | /* prio tree member */ |
121 | struct rb_node p_node; | |
122 | /* prio tree root we belong to, if any */ | |
123 | struct rb_root *p_root; | |
124 | /* sorted list of pending requests */ | |
125 | struct rb_root sort_list; | |
126 | /* if fifo isn't expired, next request to serve */ | |
127 | struct request *next_rq; | |
128 | /* requests queued in sort_list */ | |
129 | int queued[2]; | |
130 | /* currently allocated requests */ | |
131 | int allocated[2]; | |
132 | /* fifo list of requests in sort_list */ | |
133 | struct list_head fifo; | |
134 | ||
dae739eb | 135 | /* time when queue got scheduled in to dispatch first request. */ |
9a7f38c4 JM |
136 | u64 dispatch_start; |
137 | u64 allocated_slice; | |
138 | u64 slice_dispatch; | |
dae739eb | 139 | /* time when first request from queue completed and slice started. */ |
9a7f38c4 JM |
140 | u64 slice_start; |
141 | u64 slice_end; | |
93fdf147 | 142 | s64 slice_resid; |
6118b70b | 143 | |
65299a3b CH |
144 | /* pending priority requests */ |
145 | int prio_pending; | |
6118b70b JA |
146 | /* number of requests that are on the dispatch list or inside driver */ |
147 | int dispatched; | |
148 | ||
149 | /* io prio of this group */ | |
150 | unsigned short ioprio, org_ioprio; | |
b8269db4 | 151 | unsigned short ioprio_class, org_ioprio_class; |
6118b70b | 152 | |
c4081ba5 RK |
153 | pid_t pid; |
154 | ||
3dde36dd | 155 | u32 seek_history; |
b2c18e1e JM |
156 | sector_t last_request_pos; |
157 | ||
aa6f6a3d | 158 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 159 | struct cfq_queue *new_cfqq; |
cdb16e8f | 160 | struct cfq_group *cfqg; |
c4e7893e VG |
161 | /* Number of sectors dispatched from queue in single dispatch round */ |
162 | unsigned long nr_sectors; | |
6118b70b JA |
163 | }; |
164 | ||
c0324a02 | 165 | /* |
718eee05 | 166 | * First index in the service_trees. |
c0324a02 CZ |
167 | * IDLE is handled separately, so it has negative index |
168 | */ | |
3bf10fea | 169 | enum wl_class_t { |
c0324a02 | 170 | BE_WORKLOAD = 0, |
615f0259 VG |
171 | RT_WORKLOAD = 1, |
172 | IDLE_WORKLOAD = 2, | |
b4627321 | 173 | CFQ_PRIO_NR, |
c0324a02 CZ |
174 | }; |
175 | ||
718eee05 CZ |
176 | /* |
177 | * Second index in the service_trees. | |
178 | */ | |
179 | enum wl_type_t { | |
180 | ASYNC_WORKLOAD = 0, | |
181 | SYNC_NOIDLE_WORKLOAD = 1, | |
182 | SYNC_WORKLOAD = 2 | |
183 | }; | |
184 | ||
155fead9 TH |
185 | struct cfqg_stats { |
186 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
155fead9 TH |
187 | /* number of ios merged */ |
188 | struct blkg_rwstat merged; | |
189 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
190 | struct blkg_rwstat service_time; | |
191 | /* total time spent waiting in scheduler queue in ns */ | |
192 | struct blkg_rwstat wait_time; | |
193 | /* number of IOs queued up */ | |
194 | struct blkg_rwstat queued; | |
155fead9 TH |
195 | /* total disk time and nr sectors dispatched by this group */ |
196 | struct blkg_stat time; | |
197 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
198 | /* time not charged to this cgroup */ | |
199 | struct blkg_stat unaccounted_time; | |
200 | /* sum of number of ios queued across all samples */ | |
201 | struct blkg_stat avg_queue_size_sum; | |
202 | /* count of samples taken for average */ | |
203 | struct blkg_stat avg_queue_size_samples; | |
204 | /* how many times this group has been removed from service tree */ | |
205 | struct blkg_stat dequeue; | |
206 | /* total time spent waiting for it to be assigned a timeslice. */ | |
207 | struct blkg_stat group_wait_time; | |
3c798398 | 208 | /* time spent idling for this blkcg_gq */ |
155fead9 TH |
209 | struct blkg_stat idle_time; |
210 | /* total time with empty current active q with other requests queued */ | |
211 | struct blkg_stat empty_time; | |
212 | /* fields after this shouldn't be cleared on stat reset */ | |
213 | uint64_t start_group_wait_time; | |
214 | uint64_t start_idle_time; | |
215 | uint64_t start_empty_time; | |
216 | uint16_t flags; | |
217 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
218 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
219 | }; | |
220 | ||
e48453c3 AA |
221 | /* Per-cgroup data */ |
222 | struct cfq_group_data { | |
223 | /* must be the first member */ | |
81437648 | 224 | struct blkcg_policy_data cpd; |
e48453c3 AA |
225 | |
226 | unsigned int weight; | |
227 | unsigned int leaf_weight; | |
228 | }; | |
229 | ||
cdb16e8f VG |
230 | /* This is per cgroup per device grouping structure */ |
231 | struct cfq_group { | |
f95a04af TH |
232 | /* must be the first member */ |
233 | struct blkg_policy_data pd; | |
234 | ||
1fa8f6d6 VG |
235 | /* group service_tree member */ |
236 | struct rb_node rb_node; | |
237 | ||
238 | /* group service_tree key */ | |
239 | u64 vdisktime; | |
e71357e1 | 240 | |
7918ffb5 TH |
241 | /* |
242 | * The number of active cfqgs and sum of their weights under this | |
243 | * cfqg. This covers this cfqg's leaf_weight and all children's | |
244 | * weights, but does not cover weights of further descendants. | |
245 | * | |
246 | * If a cfqg is on the service tree, it's active. An active cfqg | |
247 | * also activates its parent and contributes to the children_weight | |
248 | * of the parent. | |
249 | */ | |
250 | int nr_active; | |
251 | unsigned int children_weight; | |
252 | ||
1d3650f7 TH |
253 | /* |
254 | * vfraction is the fraction of vdisktime that the tasks in this | |
255 | * cfqg are entitled to. This is determined by compounding the | |
256 | * ratios walking up from this cfqg to the root. | |
257 | * | |
258 | * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all | |
259 | * vfractions on a service tree is approximately 1. The sum may | |
260 | * deviate a bit due to rounding errors and fluctuations caused by | |
261 | * cfqgs entering and leaving the service tree. | |
262 | */ | |
263 | unsigned int vfraction; | |
264 | ||
e71357e1 TH |
265 | /* |
266 | * There are two weights - (internal) weight is the weight of this | |
267 | * cfqg against the sibling cfqgs. leaf_weight is the wight of | |
268 | * this cfqg against the child cfqgs. For the root cfqg, both | |
269 | * weights are kept in sync for backward compatibility. | |
270 | */ | |
25bc6b07 | 271 | unsigned int weight; |
8184f93e | 272 | unsigned int new_weight; |
3381cb8d | 273 | unsigned int dev_weight; |
1fa8f6d6 | 274 | |
e71357e1 TH |
275 | unsigned int leaf_weight; |
276 | unsigned int new_leaf_weight; | |
277 | unsigned int dev_leaf_weight; | |
278 | ||
1fa8f6d6 VG |
279 | /* number of cfqq currently on this group */ |
280 | int nr_cfqq; | |
281 | ||
cdb16e8f | 282 | /* |
4495a7d4 | 283 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
284 | * create the array for each prio class but at run time it is used |
285 | * only for RT and BE class and slot for IDLE class remains unused. | |
286 | * This is primarily done to avoid confusion and a gcc warning. | |
287 | */ | |
288 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
289 | /* | |
290 | * rr lists of queues with requests. We maintain service trees for | |
291 | * RT and BE classes. These trees are subdivided in subclasses | |
292 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
293 | * class there is no subclassification and all the cfq queues go on | |
294 | * a single tree service_tree_idle. | |
cdb16e8f VG |
295 | * Counts are embedded in the cfq_rb_root |
296 | */ | |
297 | struct cfq_rb_root service_trees[2][3]; | |
298 | struct cfq_rb_root service_tree_idle; | |
dae739eb | 299 | |
9a7f38c4 | 300 | u64 saved_wl_slice; |
4d2ceea4 VG |
301 | enum wl_type_t saved_wl_type; |
302 | enum wl_class_t saved_wl_class; | |
4eef3049 | 303 | |
80bdf0c7 VG |
304 | /* number of requests that are on the dispatch list or inside driver */ |
305 | int dispatched; | |
7700fc4f | 306 | struct cfq_ttime ttime; |
0b39920b | 307 | struct cfqg_stats stats; /* stats for this cfqg */ |
60a83707 TH |
308 | |
309 | /* async queue for each priority case */ | |
310 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
311 | struct cfq_queue *async_idle_cfqq; | |
312 | ||
cdb16e8f | 313 | }; |
718eee05 | 314 | |
c5869807 TH |
315 | struct cfq_io_cq { |
316 | struct io_cq icq; /* must be the first member */ | |
317 | struct cfq_queue *cfqq[2]; | |
318 | struct cfq_ttime ttime; | |
598971bf TH |
319 | int ioprio; /* the current ioprio */ |
320 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
f4da8072 | 321 | uint64_t blkcg_serial_nr; /* the current blkcg serial */ |
598971bf | 322 | #endif |
c5869807 TH |
323 | }; |
324 | ||
22e2c507 JA |
325 | /* |
326 | * Per block device queue structure | |
327 | */ | |
1da177e4 | 328 | struct cfq_data { |
165125e1 | 329 | struct request_queue *queue; |
1fa8f6d6 VG |
330 | /* Root service tree for cfq_groups */ |
331 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 332 | struct cfq_group *root_group; |
22e2c507 | 333 | |
c0324a02 CZ |
334 | /* |
335 | * The priority currently being served | |
22e2c507 | 336 | */ |
4d2ceea4 VG |
337 | enum wl_class_t serving_wl_class; |
338 | enum wl_type_t serving_wl_type; | |
9a7f38c4 | 339 | u64 workload_expires; |
cdb16e8f | 340 | struct cfq_group *serving_group; |
a36e71f9 JA |
341 | |
342 | /* | |
343 | * Each priority tree is sorted by next_request position. These | |
344 | * trees are used when determining if two or more queues are | |
345 | * interleaving requests (see cfq_close_cooperator). | |
346 | */ | |
347 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
348 | ||
22e2c507 | 349 | unsigned int busy_queues; |
ef8a41df | 350 | unsigned int busy_sync_queues; |
22e2c507 | 351 | |
53c583d2 CZ |
352 | int rq_in_driver; |
353 | int rq_in_flight[2]; | |
45333d5a AC |
354 | |
355 | /* | |
356 | * queue-depth detection | |
357 | */ | |
358 | int rq_queued; | |
25776e35 | 359 | int hw_tag; |
e459dd08 CZ |
360 | /* |
361 | * hw_tag can be | |
362 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
363 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
364 | * 0 => no NCQ | |
365 | */ | |
366 | int hw_tag_est_depth; | |
367 | unsigned int hw_tag_samples; | |
1da177e4 | 368 | |
22e2c507 JA |
369 | /* |
370 | * idle window management | |
371 | */ | |
91148325 | 372 | struct hrtimer idle_slice_timer; |
23e018a1 | 373 | struct work_struct unplug_work; |
1da177e4 | 374 | |
22e2c507 | 375 | struct cfq_queue *active_queue; |
c5869807 | 376 | struct cfq_io_cq *active_cic; |
22e2c507 | 377 | |
6d048f53 | 378 | sector_t last_position; |
1da177e4 | 379 | |
1da177e4 LT |
380 | /* |
381 | * tunables, see top of file | |
382 | */ | |
383 | unsigned int cfq_quantum; | |
1da177e4 LT |
384 | unsigned int cfq_back_penalty; |
385 | unsigned int cfq_back_max; | |
22e2c507 | 386 | unsigned int cfq_slice_async_rq; |
963b72fc | 387 | unsigned int cfq_latency; |
9a7f38c4 JM |
388 | u64 cfq_fifo_expire[2]; |
389 | u64 cfq_slice[2]; | |
390 | u64 cfq_slice_idle; | |
391 | u64 cfq_group_idle; | |
392 | u64 cfq_target_latency; | |
d9ff4187 | 393 | |
6118b70b JA |
394 | /* |
395 | * Fallback dummy cfqq for extreme OOM conditions | |
396 | */ | |
397 | struct cfq_queue oom_cfqq; | |
365722bb | 398 | |
9a7f38c4 | 399 | u64 last_delayed_sync; |
1da177e4 LT |
400 | }; |
401 | ||
25fb5169 | 402 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
60a83707 | 403 | static void cfq_put_queue(struct cfq_queue *cfqq); |
25fb5169 | 404 | |
34b98d03 | 405 | static struct cfq_rb_root *st_for(struct cfq_group *cfqg, |
3bf10fea | 406 | enum wl_class_t class, |
65b32a57 | 407 | enum wl_type_t type) |
c0324a02 | 408 | { |
1fa8f6d6 VG |
409 | if (!cfqg) |
410 | return NULL; | |
411 | ||
3bf10fea | 412 | if (class == IDLE_WORKLOAD) |
cdb16e8f | 413 | return &cfqg->service_tree_idle; |
c0324a02 | 414 | |
3bf10fea | 415 | return &cfqg->service_trees[class][type]; |
c0324a02 CZ |
416 | } |
417 | ||
3b18152c | 418 | enum cfqq_state_flags { |
b0b8d749 JA |
419 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
420 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 421 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 422 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
423 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
424 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
425 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 426 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 427 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 428 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 429 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 430 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 431 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
432 | }; |
433 | ||
434 | #define CFQ_CFQQ_FNS(name) \ | |
435 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
436 | { \ | |
fe094d98 | 437 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
438 | } \ |
439 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
440 | { \ | |
fe094d98 | 441 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
442 | } \ |
443 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
444 | { \ | |
fe094d98 | 445 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
446 | } |
447 | ||
448 | CFQ_CFQQ_FNS(on_rr); | |
449 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 450 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 451 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
452 | CFQ_CFQQ_FNS(fifo_expire); |
453 | CFQ_CFQQ_FNS(idle_window); | |
454 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 455 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 456 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 457 | CFQ_CFQQ_FNS(coop); |
ae54abed | 458 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 459 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 460 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
461 | #undef CFQ_CFQQ_FNS |
462 | ||
629ed0b1 | 463 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
2ce4d50f | 464 | |
155fead9 TH |
465 | /* cfqg stats flags */ |
466 | enum cfqg_stats_flags { | |
467 | CFQG_stats_waiting = 0, | |
468 | CFQG_stats_idling, | |
469 | CFQG_stats_empty, | |
629ed0b1 TH |
470 | }; |
471 | ||
155fead9 TH |
472 | #define CFQG_FLAG_FNS(name) \ |
473 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ | |
629ed0b1 | 474 | { \ |
155fead9 | 475 | stats->flags |= (1 << CFQG_stats_##name); \ |
629ed0b1 | 476 | } \ |
155fead9 | 477 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 478 | { \ |
155fead9 | 479 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
629ed0b1 | 480 | } \ |
155fead9 | 481 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 482 | { \ |
155fead9 | 483 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
629ed0b1 TH |
484 | } \ |
485 | ||
155fead9 TH |
486 | CFQG_FLAG_FNS(waiting) |
487 | CFQG_FLAG_FNS(idling) | |
488 | CFQG_FLAG_FNS(empty) | |
489 | #undef CFQG_FLAG_FNS | |
629ed0b1 TH |
490 | |
491 | /* This should be called with the queue_lock held. */ | |
155fead9 | 492 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
629ed0b1 TH |
493 | { |
494 | unsigned long long now; | |
495 | ||
155fead9 | 496 | if (!cfqg_stats_waiting(stats)) |
629ed0b1 TH |
497 | return; |
498 | ||
499 | now = sched_clock(); | |
500 | if (time_after64(now, stats->start_group_wait_time)) | |
501 | blkg_stat_add(&stats->group_wait_time, | |
502 | now - stats->start_group_wait_time); | |
155fead9 | 503 | cfqg_stats_clear_waiting(stats); |
629ed0b1 TH |
504 | } |
505 | ||
506 | /* This should be called with the queue_lock held. */ | |
155fead9 TH |
507 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
508 | struct cfq_group *curr_cfqg) | |
629ed0b1 | 509 | { |
155fead9 | 510 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 511 | |
155fead9 | 512 | if (cfqg_stats_waiting(stats)) |
629ed0b1 | 513 | return; |
155fead9 | 514 | if (cfqg == curr_cfqg) |
629ed0b1 | 515 | return; |
155fead9 TH |
516 | stats->start_group_wait_time = sched_clock(); |
517 | cfqg_stats_mark_waiting(stats); | |
629ed0b1 TH |
518 | } |
519 | ||
520 | /* This should be called with the queue_lock held. */ | |
155fead9 | 521 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
629ed0b1 TH |
522 | { |
523 | unsigned long long now; | |
524 | ||
155fead9 | 525 | if (!cfqg_stats_empty(stats)) |
629ed0b1 TH |
526 | return; |
527 | ||
528 | now = sched_clock(); | |
529 | if (time_after64(now, stats->start_empty_time)) | |
530 | blkg_stat_add(&stats->empty_time, | |
531 | now - stats->start_empty_time); | |
155fead9 | 532 | cfqg_stats_clear_empty(stats); |
629ed0b1 TH |
533 | } |
534 | ||
155fead9 | 535 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
629ed0b1 | 536 | { |
155fead9 | 537 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
629ed0b1 TH |
538 | } |
539 | ||
155fead9 | 540 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
629ed0b1 | 541 | { |
155fead9 | 542 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 543 | |
4d5e80a7 | 544 | if (blkg_rwstat_total(&stats->queued)) |
629ed0b1 TH |
545 | return; |
546 | ||
547 | /* | |
548 | * group is already marked empty. This can happen if cfqq got new | |
549 | * request in parent group and moved to this group while being added | |
550 | * to service tree. Just ignore the event and move on. | |
551 | */ | |
155fead9 | 552 | if (cfqg_stats_empty(stats)) |
629ed0b1 TH |
553 | return; |
554 | ||
555 | stats->start_empty_time = sched_clock(); | |
155fead9 | 556 | cfqg_stats_mark_empty(stats); |
629ed0b1 TH |
557 | } |
558 | ||
155fead9 | 559 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 560 | { |
155fead9 | 561 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 562 | |
155fead9 | 563 | if (cfqg_stats_idling(stats)) { |
629ed0b1 TH |
564 | unsigned long long now = sched_clock(); |
565 | ||
566 | if (time_after64(now, stats->start_idle_time)) | |
567 | blkg_stat_add(&stats->idle_time, | |
568 | now - stats->start_idle_time); | |
155fead9 | 569 | cfqg_stats_clear_idling(stats); |
629ed0b1 TH |
570 | } |
571 | } | |
572 | ||
155fead9 | 573 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 574 | { |
155fead9 | 575 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 576 | |
155fead9 | 577 | BUG_ON(cfqg_stats_idling(stats)); |
629ed0b1 TH |
578 | |
579 | stats->start_idle_time = sched_clock(); | |
155fead9 | 580 | cfqg_stats_mark_idling(stats); |
629ed0b1 TH |
581 | } |
582 | ||
155fead9 | 583 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
629ed0b1 | 584 | { |
155fead9 | 585 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
586 | |
587 | blkg_stat_add(&stats->avg_queue_size_sum, | |
4d5e80a7 | 588 | blkg_rwstat_total(&stats->queued)); |
629ed0b1 | 589 | blkg_stat_add(&stats->avg_queue_size_samples, 1); |
155fead9 | 590 | cfqg_stats_update_group_wait_time(stats); |
629ed0b1 TH |
591 | } |
592 | ||
593 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
594 | ||
f48ec1d7 TH |
595 | static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { } |
596 | static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } | |
597 | static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } | |
598 | static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } | |
599 | static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } | |
600 | static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } | |
601 | static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } | |
629ed0b1 TH |
602 | |
603 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
604 | ||
605 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
2ce4d50f | 606 | |
4ceab71b JA |
607 | static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd) |
608 | { | |
609 | return pd ? container_of(pd, struct cfq_group, pd) : NULL; | |
610 | } | |
611 | ||
612 | static struct cfq_group_data | |
613 | *cpd_to_cfqgd(struct blkcg_policy_data *cpd) | |
614 | { | |
81437648 | 615 | return cpd ? container_of(cpd, struct cfq_group_data, cpd) : NULL; |
4ceab71b JA |
616 | } |
617 | ||
618 | static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg) | |
619 | { | |
620 | return pd_to_blkg(&cfqg->pd); | |
621 | } | |
622 | ||
ffea73fc TH |
623 | static struct blkcg_policy blkcg_policy_cfq; |
624 | ||
625 | static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg) | |
626 | { | |
627 | return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq)); | |
628 | } | |
629 | ||
e48453c3 AA |
630 | static struct cfq_group_data *blkcg_to_cfqgd(struct blkcg *blkcg) |
631 | { | |
632 | return cpd_to_cfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_cfq)); | |
633 | } | |
634 | ||
d02f7aa8 | 635 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) |
7918ffb5 | 636 | { |
d02f7aa8 | 637 | struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent; |
7918ffb5 | 638 | |
d02f7aa8 | 639 | return pblkg ? blkg_to_cfqg(pblkg) : NULL; |
7918ffb5 TH |
640 | } |
641 | ||
3984aa55 JK |
642 | static inline bool cfqg_is_descendant(struct cfq_group *cfqg, |
643 | struct cfq_group *ancestor) | |
644 | { | |
645 | return cgroup_is_descendant(cfqg_to_blkg(cfqg)->blkcg->css.cgroup, | |
646 | cfqg_to_blkg(ancestor)->blkcg->css.cgroup); | |
647 | } | |
648 | ||
eb7d8c07 TH |
649 | static inline void cfqg_get(struct cfq_group *cfqg) |
650 | { | |
651 | return blkg_get(cfqg_to_blkg(cfqg)); | |
652 | } | |
653 | ||
654 | static inline void cfqg_put(struct cfq_group *cfqg) | |
655 | { | |
656 | return blkg_put(cfqg_to_blkg(cfqg)); | |
657 | } | |
658 | ||
54e7ed12 | 659 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \ |
35fe6d76 SL |
660 | blk_add_cgroup_trace_msg((cfqd)->queue, \ |
661 | cfqg_to_blkg((cfqq)->cfqg)->blkcg, \ | |
662 | "cfq%d%c%c " fmt, (cfqq)->pid, \ | |
b226e5c4 VG |
663 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ |
664 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
35fe6d76 | 665 | ##args); \ |
54e7ed12 TH |
666 | } while (0) |
667 | ||
668 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \ | |
35fe6d76 SL |
669 | blk_add_cgroup_trace_msg((cfqd)->queue, \ |
670 | cfqg_to_blkg(cfqg)->blkcg, fmt, ##args); \ | |
54e7ed12 | 671 | } while (0) |
2868ef7b | 672 | |
155fead9 | 673 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
ef295ecf CH |
674 | struct cfq_group *curr_cfqg, |
675 | unsigned int op) | |
2ce4d50f | 676 | { |
ef295ecf | 677 | blkg_rwstat_add(&cfqg->stats.queued, op, 1); |
155fead9 TH |
678 | cfqg_stats_end_empty_time(&cfqg->stats); |
679 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); | |
2ce4d50f TH |
680 | } |
681 | ||
155fead9 | 682 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
9a7f38c4 | 683 | uint64_t time, unsigned long unaccounted_time) |
2ce4d50f | 684 | { |
155fead9 | 685 | blkg_stat_add(&cfqg->stats.time, time); |
629ed0b1 | 686 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
155fead9 | 687 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
629ed0b1 | 688 | #endif |
2ce4d50f TH |
689 | } |
690 | ||
ef295ecf CH |
691 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, |
692 | unsigned int op) | |
2ce4d50f | 693 | { |
ef295ecf | 694 | blkg_rwstat_add(&cfqg->stats.queued, op, -1); |
2ce4d50f TH |
695 | } |
696 | ||
ef295ecf CH |
697 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, |
698 | unsigned int op) | |
2ce4d50f | 699 | { |
ef295ecf | 700 | blkg_rwstat_add(&cfqg->stats.merged, op, 1); |
2ce4d50f TH |
701 | } |
702 | ||
155fead9 | 703 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
ef295ecf CH |
704 | uint64_t start_time, uint64_t io_start_time, |
705 | unsigned int op) | |
2ce4d50f | 706 | { |
155fead9 | 707 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 708 | unsigned long long now = sched_clock(); |
629ed0b1 TH |
709 | |
710 | if (time_after64(now, io_start_time)) | |
ef295ecf | 711 | blkg_rwstat_add(&stats->service_time, op, now - io_start_time); |
629ed0b1 | 712 | if (time_after64(io_start_time, start_time)) |
ef295ecf | 713 | blkg_rwstat_add(&stats->wait_time, op, |
629ed0b1 | 714 | io_start_time - start_time); |
2ce4d50f TH |
715 | } |
716 | ||
689665af TH |
717 | /* @stats = 0 */ |
718 | static void cfqg_stats_reset(struct cfqg_stats *stats) | |
155fead9 | 719 | { |
155fead9 | 720 | /* queued stats shouldn't be cleared */ |
155fead9 TH |
721 | blkg_rwstat_reset(&stats->merged); |
722 | blkg_rwstat_reset(&stats->service_time); | |
723 | blkg_rwstat_reset(&stats->wait_time); | |
724 | blkg_stat_reset(&stats->time); | |
725 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
726 | blkg_stat_reset(&stats->unaccounted_time); | |
727 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
728 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
729 | blkg_stat_reset(&stats->dequeue); | |
730 | blkg_stat_reset(&stats->group_wait_time); | |
731 | blkg_stat_reset(&stats->idle_time); | |
732 | blkg_stat_reset(&stats->empty_time); | |
733 | #endif | |
734 | } | |
735 | ||
0b39920b | 736 | /* @to += @from */ |
e6269c44 | 737 | static void cfqg_stats_add_aux(struct cfqg_stats *to, struct cfqg_stats *from) |
0b39920b TH |
738 | { |
739 | /* queued stats shouldn't be cleared */ | |
e6269c44 TH |
740 | blkg_rwstat_add_aux(&to->merged, &from->merged); |
741 | blkg_rwstat_add_aux(&to->service_time, &from->service_time); | |
742 | blkg_rwstat_add_aux(&to->wait_time, &from->wait_time); | |
743 | blkg_stat_add_aux(&from->time, &from->time); | |
0b39920b | 744 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
e6269c44 TH |
745 | blkg_stat_add_aux(&to->unaccounted_time, &from->unaccounted_time); |
746 | blkg_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum); | |
747 | blkg_stat_add_aux(&to->avg_queue_size_samples, &from->avg_queue_size_samples); | |
748 | blkg_stat_add_aux(&to->dequeue, &from->dequeue); | |
749 | blkg_stat_add_aux(&to->group_wait_time, &from->group_wait_time); | |
750 | blkg_stat_add_aux(&to->idle_time, &from->idle_time); | |
751 | blkg_stat_add_aux(&to->empty_time, &from->empty_time); | |
0b39920b TH |
752 | #endif |
753 | } | |
754 | ||
755 | /* | |
e6269c44 | 756 | * Transfer @cfqg's stats to its parent's aux counts so that the ancestors' |
0b39920b TH |
757 | * recursive stats can still account for the amount used by this cfqg after |
758 | * it's gone. | |
759 | */ | |
760 | static void cfqg_stats_xfer_dead(struct cfq_group *cfqg) | |
761 | { | |
762 | struct cfq_group *parent = cfqg_parent(cfqg); | |
763 | ||
764 | lockdep_assert_held(cfqg_to_blkg(cfqg)->q->queue_lock); | |
765 | ||
766 | if (unlikely(!parent)) | |
767 | return; | |
768 | ||
e6269c44 | 769 | cfqg_stats_add_aux(&parent->stats, &cfqg->stats); |
0b39920b | 770 | cfqg_stats_reset(&cfqg->stats); |
0b39920b TH |
771 | } |
772 | ||
eb7d8c07 TH |
773 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
774 | ||
d02f7aa8 | 775 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) { return NULL; } |
3984aa55 JK |
776 | static inline bool cfqg_is_descendant(struct cfq_group *cfqg, |
777 | struct cfq_group *ancestor) | |
778 | { | |
779 | return true; | |
780 | } | |
eb7d8c07 TH |
781 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
782 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
783 | ||
7b679138 | 784 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
b226e5c4 VG |
785 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \ |
786 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
787 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
788 | ##args) | |
4495a7d4 | 789 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 | 790 | |
155fead9 | 791 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
ef295ecf | 792 | struct cfq_group *curr_cfqg, unsigned int op) { } |
155fead9 | 793 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
9a7f38c4 | 794 | uint64_t time, unsigned long unaccounted_time) { } |
ef295ecf CH |
795 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, |
796 | unsigned int op) { } | |
797 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, | |
798 | unsigned int op) { } | |
155fead9 | 799 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
ef295ecf CH |
800 | uint64_t start_time, uint64_t io_start_time, |
801 | unsigned int op) { } | |
2ce4d50f | 802 | |
eb7d8c07 TH |
803 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
804 | ||
7b679138 JA |
805 | #define cfq_log(cfqd, fmt, args...) \ |
806 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
807 | ||
615f0259 VG |
808 | /* Traverses through cfq group service trees */ |
809 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
810 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
811 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
812 | : &cfqg->service_tree_idle; \ | |
813 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
814 | (i == IDLE_WORKLOAD && j == 0); \ | |
815 | j++, st = i < IDLE_WORKLOAD ? \ | |
816 | &cfqg->service_trees[i][j]: NULL) \ | |
817 | ||
f5f2b6ce SL |
818 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
819 | struct cfq_ttime *ttime, bool group_idle) | |
820 | { | |
9a7f38c4 | 821 | u64 slice; |
f5f2b6ce SL |
822 | if (!sample_valid(ttime->ttime_samples)) |
823 | return false; | |
824 | if (group_idle) | |
825 | slice = cfqd->cfq_group_idle; | |
826 | else | |
827 | slice = cfqd->cfq_slice_idle; | |
828 | return ttime->ttime_mean > slice; | |
829 | } | |
615f0259 | 830 | |
02b35081 VG |
831 | static inline bool iops_mode(struct cfq_data *cfqd) |
832 | { | |
833 | /* | |
834 | * If we are not idling on queues and it is a NCQ drive, parallel | |
835 | * execution of requests is on and measuring time is not possible | |
836 | * in most of the cases until and unless we drive shallower queue | |
837 | * depths and that becomes a performance bottleneck. In such cases | |
838 | * switch to start providing fairness in terms of number of IOs. | |
839 | */ | |
840 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
841 | return true; | |
842 | else | |
843 | return false; | |
844 | } | |
845 | ||
3bf10fea | 846 | static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq) |
c0324a02 CZ |
847 | { |
848 | if (cfq_class_idle(cfqq)) | |
849 | return IDLE_WORKLOAD; | |
850 | if (cfq_class_rt(cfqq)) | |
851 | return RT_WORKLOAD; | |
852 | return BE_WORKLOAD; | |
853 | } | |
854 | ||
718eee05 CZ |
855 | |
856 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
857 | { | |
858 | if (!cfq_cfqq_sync(cfqq)) | |
859 | return ASYNC_WORKLOAD; | |
860 | if (!cfq_cfqq_idle_window(cfqq)) | |
861 | return SYNC_NOIDLE_WORKLOAD; | |
862 | return SYNC_WORKLOAD; | |
863 | } | |
864 | ||
3bf10fea | 865 | static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class, |
58ff82f3 VG |
866 | struct cfq_data *cfqd, |
867 | struct cfq_group *cfqg) | |
c0324a02 | 868 | { |
3bf10fea | 869 | if (wl_class == IDLE_WORKLOAD) |
cdb16e8f | 870 | return cfqg->service_tree_idle.count; |
c0324a02 | 871 | |
34b98d03 VG |
872 | return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count + |
873 | cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count + | |
874 | cfqg->service_trees[wl_class][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
875 | } |
876 | ||
f26bd1f0 VG |
877 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
878 | struct cfq_group *cfqg) | |
879 | { | |
34b98d03 VG |
880 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count + |
881 | cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
f26bd1f0 VG |
882 | } |
883 | ||
165125e1 | 884 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 885 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
2da8de0b | 886 | struct cfq_io_cq *cic, struct bio *bio); |
91fac317 | 887 | |
c5869807 TH |
888 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
889 | { | |
890 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
891 | return container_of(icq, struct cfq_io_cq, icq); | |
892 | } | |
893 | ||
47fdd4ca TH |
894 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
895 | struct io_context *ioc) | |
896 | { | |
897 | if (ioc) | |
898 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
899 | return NULL; | |
900 | } | |
901 | ||
c5869807 | 902 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 903 | { |
a6151c3a | 904 | return cic->cfqq[is_sync]; |
91fac317 VT |
905 | } |
906 | ||
c5869807 TH |
907 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
908 | bool is_sync) | |
91fac317 | 909 | { |
a6151c3a | 910 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
911 | } |
912 | ||
c5869807 | 913 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 914 | { |
c5869807 | 915 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
916 | } |
917 | ||
99f95e52 AM |
918 | /* |
919 | * scheduler run of queue, if there are requests pending and no one in the | |
920 | * driver that will restart queueing | |
921 | */ | |
23e018a1 | 922 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 923 | { |
7b679138 JA |
924 | if (cfqd->busy_queues) { |
925 | cfq_log(cfqd, "schedule dispatch"); | |
59c3d45e | 926 | kblockd_schedule_work(&cfqd->unplug_work); |
7b679138 | 927 | } |
99f95e52 AM |
928 | } |
929 | ||
44f7c160 JA |
930 | /* |
931 | * Scale schedule slice based on io priority. Use the sync time slice only | |
932 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
933 | * io only, should not get full sync slice length. | |
934 | */ | |
9a7f38c4 | 935 | static inline u64 cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 936 | unsigned short prio) |
44f7c160 | 937 | { |
9a7f38c4 JM |
938 | u64 base_slice = cfqd->cfq_slice[sync]; |
939 | u64 slice = div_u64(base_slice, CFQ_SLICE_SCALE); | |
44f7c160 | 940 | |
d9e7620e JA |
941 | WARN_ON(prio >= IOPRIO_BE_NR); |
942 | ||
9a7f38c4 | 943 | return base_slice + (slice * (4 - prio)); |
d9e7620e | 944 | } |
44f7c160 | 945 | |
9a7f38c4 | 946 | static inline u64 |
d9e7620e JA |
947 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
948 | { | |
949 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
950 | } |
951 | ||
1d3650f7 TH |
952 | /** |
953 | * cfqg_scale_charge - scale disk time charge according to cfqg weight | |
954 | * @charge: disk time being charged | |
955 | * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT | |
956 | * | |
957 | * Scale @charge according to @vfraction, which is in range (0, 1]. The | |
958 | * scaling is inversely proportional. | |
959 | * | |
960 | * scaled = charge / vfraction | |
961 | * | |
962 | * The result is also in fixed point w/ CFQ_SERVICE_SHIFT. | |
963 | */ | |
9a7f38c4 | 964 | static inline u64 cfqg_scale_charge(u64 charge, |
1d3650f7 | 965 | unsigned int vfraction) |
25bc6b07 | 966 | { |
1d3650f7 | 967 | u64 c = charge << CFQ_SERVICE_SHIFT; /* make it fixed point */ |
25bc6b07 | 968 | |
1d3650f7 TH |
969 | /* charge / vfraction */ |
970 | c <<= CFQ_SERVICE_SHIFT; | |
9a7f38c4 | 971 | return div_u64(c, vfraction); |
25bc6b07 VG |
972 | } |
973 | ||
974 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
975 | { | |
976 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
977 | if (delta > 0) | |
978 | min_vdisktime = vdisktime; | |
979 | ||
980 | return min_vdisktime; | |
981 | } | |
982 | ||
25bc6b07 VG |
983 | static void update_min_vdisktime(struct cfq_rb_root *st) |
984 | { | |
09663c86 DB |
985 | if (!RB_EMPTY_ROOT(&st->rb.rb_root)) { |
986 | struct cfq_group *cfqg = rb_entry_cfqg(st->rb.rb_leftmost); | |
25bc6b07 | 987 | |
a6032710 GJ |
988 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
989 | cfqg->vdisktime); | |
25bc6b07 | 990 | } |
25bc6b07 VG |
991 | } |
992 | ||
5db5d642 CZ |
993 | /* |
994 | * get averaged number of queues of RT/BE priority. | |
995 | * average is updated, with a formula that gives more weight to higher numbers, | |
996 | * to quickly follows sudden increases and decrease slowly | |
997 | */ | |
998 | ||
58ff82f3 VG |
999 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
1000 | struct cfq_group *cfqg, bool rt) | |
5869619c | 1001 | { |
5db5d642 CZ |
1002 | unsigned min_q, max_q; |
1003 | unsigned mult = cfq_hist_divisor - 1; | |
1004 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 1005 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 1006 | |
58ff82f3 VG |
1007 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
1008 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
1009 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 1010 | cfq_hist_divisor; |
58ff82f3 VG |
1011 | return cfqg->busy_queues_avg[rt]; |
1012 | } | |
1013 | ||
9a7f38c4 | 1014 | static inline u64 |
58ff82f3 VG |
1015 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1016 | { | |
41cad6ab | 1017 | return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT; |
5db5d642 CZ |
1018 | } |
1019 | ||
9a7f38c4 | 1020 | static inline u64 |
ba5bd520 | 1021 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 1022 | { |
9a7f38c4 | 1023 | u64 slice = cfq_prio_to_slice(cfqd, cfqq); |
5db5d642 | 1024 | if (cfqd->cfq_latency) { |
58ff82f3 VG |
1025 | /* |
1026 | * interested queues (we consider only the ones with the same | |
1027 | * priority class in the cfq group) | |
1028 | */ | |
1029 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
1030 | cfq_class_rt(cfqq)); | |
9a7f38c4 JM |
1031 | u64 sync_slice = cfqd->cfq_slice[1]; |
1032 | u64 expect_latency = sync_slice * iq; | |
1033 | u64 group_slice = cfq_group_slice(cfqd, cfqq->cfqg); | |
58ff82f3 VG |
1034 | |
1035 | if (expect_latency > group_slice) { | |
9a7f38c4 JM |
1036 | u64 base_low_slice = 2 * cfqd->cfq_slice_idle; |
1037 | u64 low_slice; | |
1038 | ||
5db5d642 CZ |
1039 | /* scale low_slice according to IO priority |
1040 | * and sync vs async */ | |
9a7f38c4 JM |
1041 | low_slice = div64_u64(base_low_slice*slice, sync_slice); |
1042 | low_slice = min(slice, low_slice); | |
5db5d642 CZ |
1043 | /* the adapted slice value is scaled to fit all iqs |
1044 | * into the target latency */ | |
9a7f38c4 JM |
1045 | slice = div64_u64(slice*group_slice, expect_latency); |
1046 | slice = max(slice, low_slice); | |
5db5d642 CZ |
1047 | } |
1048 | } | |
c553f8e3 SL |
1049 | return slice; |
1050 | } | |
1051 | ||
1052 | static inline void | |
1053 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1054 | { | |
9a7f38c4 JM |
1055 | u64 slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
1056 | u64 now = ktime_get_ns(); | |
c553f8e3 | 1057 | |
9a7f38c4 JM |
1058 | cfqq->slice_start = now; |
1059 | cfqq->slice_end = now + slice; | |
f75edf2d | 1060 | cfqq->allocated_slice = slice; |
9a7f38c4 | 1061 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%llu", cfqq->slice_end - now); |
44f7c160 JA |
1062 | } |
1063 | ||
1064 | /* | |
1065 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
1066 | * isn't valid until the first request from the dispatch is activated | |
1067 | * and the slice time set. | |
1068 | */ | |
a6151c3a | 1069 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
1070 | { |
1071 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 1072 | return false; |
9a7f38c4 | 1073 | if (ktime_get_ns() < cfqq->slice_end) |
c1e44756 | 1074 | return false; |
44f7c160 | 1075 | |
c1e44756 | 1076 | return true; |
44f7c160 JA |
1077 | } |
1078 | ||
1da177e4 | 1079 | /* |
5e705374 | 1080 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 1081 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 1082 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 1083 | */ |
5e705374 | 1084 | static struct request * |
cf7c25cf | 1085 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 1086 | { |
cf7c25cf | 1087 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 1088 | unsigned long back_max; |
e8a99053 AM |
1089 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
1090 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
1091 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 1092 | |
5e705374 JA |
1093 | if (rq1 == NULL || rq1 == rq2) |
1094 | return rq2; | |
1095 | if (rq2 == NULL) | |
1096 | return rq1; | |
9c2c38a1 | 1097 | |
229836bd NK |
1098 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
1099 | return rq_is_sync(rq1) ? rq1 : rq2; | |
1100 | ||
65299a3b CH |
1101 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
1102 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 1103 | |
83096ebf TH |
1104 | s1 = blk_rq_pos(rq1); |
1105 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 1106 | |
1da177e4 LT |
1107 | /* |
1108 | * by definition, 1KiB is 2 sectors | |
1109 | */ | |
1110 | back_max = cfqd->cfq_back_max * 2; | |
1111 | ||
1112 | /* | |
1113 | * Strict one way elevator _except_ in the case where we allow | |
1114 | * short backward seeks which are biased as twice the cost of a | |
1115 | * similar forward seek. | |
1116 | */ | |
1117 | if (s1 >= last) | |
1118 | d1 = s1 - last; | |
1119 | else if (s1 + back_max >= last) | |
1120 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
1121 | else | |
e8a99053 | 1122 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
1123 | |
1124 | if (s2 >= last) | |
1125 | d2 = s2 - last; | |
1126 | else if (s2 + back_max >= last) | |
1127 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
1128 | else | |
e8a99053 | 1129 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
1130 | |
1131 | /* Found required data */ | |
e8a99053 AM |
1132 | |
1133 | /* | |
1134 | * By doing switch() on the bit mask "wrap" we avoid having to | |
1135 | * check two variables for all permutations: --> faster! | |
1136 | */ | |
1137 | switch (wrap) { | |
5e705374 | 1138 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 1139 | if (d1 < d2) |
5e705374 | 1140 | return rq1; |
e8a99053 | 1141 | else if (d2 < d1) |
5e705374 | 1142 | return rq2; |
e8a99053 AM |
1143 | else { |
1144 | if (s1 >= s2) | |
5e705374 | 1145 | return rq1; |
e8a99053 | 1146 | else |
5e705374 | 1147 | return rq2; |
e8a99053 | 1148 | } |
1da177e4 | 1149 | |
e8a99053 | 1150 | case CFQ_RQ2_WRAP: |
5e705374 | 1151 | return rq1; |
e8a99053 | 1152 | case CFQ_RQ1_WRAP: |
5e705374 JA |
1153 | return rq2; |
1154 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
1155 | default: |
1156 | /* | |
1157 | * Since both rqs are wrapped, | |
1158 | * start with the one that's further behind head | |
1159 | * (--> only *one* back seek required), | |
1160 | * since back seek takes more time than forward. | |
1161 | */ | |
1162 | if (s1 <= s2) | |
5e705374 | 1163 | return rq1; |
1da177e4 | 1164 | else |
5e705374 | 1165 | return rq2; |
1da177e4 LT |
1166 | } |
1167 | } | |
1168 | ||
0871714e | 1169 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 1170 | { |
615f0259 VG |
1171 | /* Service tree is empty */ |
1172 | if (!root->count) | |
1173 | return NULL; | |
1174 | ||
09663c86 | 1175 | return rb_entry(rb_first_cached(&root->rb), struct cfq_queue, rb_node); |
cc09e299 JA |
1176 | } |
1177 | ||
1fa8f6d6 VG |
1178 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
1179 | { | |
09663c86 | 1180 | return rb_entry_cfqg(rb_first_cached(&root->rb)); |
1fa8f6d6 VG |
1181 | } |
1182 | ||
09663c86 | 1183 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
a36e71f9 | 1184 | { |
f0f1a45f DB |
1185 | if (root->rb_rightmost == n) |
1186 | root->rb_rightmost = rb_prev(n); | |
1187 | ||
09663c86 | 1188 | rb_erase_cached(n, &root->rb); |
a36e71f9 | 1189 | RB_CLEAR_NODE(n); |
a36e71f9 | 1190 | |
aa6f6a3d | 1191 | --root->count; |
cc09e299 JA |
1192 | } |
1193 | ||
1da177e4 LT |
1194 | /* |
1195 | * would be nice to take fifo expire time into account as well | |
1196 | */ | |
5e705374 JA |
1197 | static struct request * |
1198 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1199 | struct request *last) | |
1da177e4 | 1200 | { |
21183b07 JA |
1201 | struct rb_node *rbnext = rb_next(&last->rb_node); |
1202 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 1203 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 1204 | |
21183b07 | 1205 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
1206 | |
1207 | if (rbprev) | |
5e705374 | 1208 | prev = rb_entry_rq(rbprev); |
1da177e4 | 1209 | |
21183b07 | 1210 | if (rbnext) |
5e705374 | 1211 | next = rb_entry_rq(rbnext); |
21183b07 JA |
1212 | else { |
1213 | rbnext = rb_first(&cfqq->sort_list); | |
1214 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 1215 | next = rb_entry_rq(rbnext); |
21183b07 | 1216 | } |
1da177e4 | 1217 | |
cf7c25cf | 1218 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
1219 | } |
1220 | ||
9a7f38c4 JM |
1221 | static u64 cfq_slice_offset(struct cfq_data *cfqd, |
1222 | struct cfq_queue *cfqq) | |
1da177e4 | 1223 | { |
d9e7620e JA |
1224 | /* |
1225 | * just an approximation, should be ok. | |
1226 | */ | |
cdb16e8f | 1227 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 1228 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
1229 | } |
1230 | ||
1fa8f6d6 VG |
1231 | static inline s64 |
1232 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1233 | { | |
1234 | return cfqg->vdisktime - st->min_vdisktime; | |
1235 | } | |
1236 | ||
1237 | static void | |
1238 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1239 | { | |
09663c86 | 1240 | struct rb_node **node = &st->rb.rb_root.rb_node; |
1fa8f6d6 VG |
1241 | struct rb_node *parent = NULL; |
1242 | struct cfq_group *__cfqg; | |
1243 | s64 key = cfqg_key(st, cfqg); | |
f0f1a45f | 1244 | bool leftmost = true, rightmost = true; |
1fa8f6d6 VG |
1245 | |
1246 | while (*node != NULL) { | |
1247 | parent = *node; | |
1248 | __cfqg = rb_entry_cfqg(parent); | |
1249 | ||
f0f1a45f | 1250 | if (key < cfqg_key(st, __cfqg)) { |
1fa8f6d6 | 1251 | node = &parent->rb_left; |
f0f1a45f DB |
1252 | rightmost = false; |
1253 | } else { | |
1fa8f6d6 | 1254 | node = &parent->rb_right; |
09663c86 | 1255 | leftmost = false; |
1fa8f6d6 VG |
1256 | } |
1257 | } | |
1258 | ||
f0f1a45f DB |
1259 | if (rightmost) |
1260 | st->rb_rightmost = &cfqg->rb_node; | |
1261 | ||
1fa8f6d6 | 1262 | rb_link_node(&cfqg->rb_node, parent, node); |
09663c86 | 1263 | rb_insert_color_cached(&cfqg->rb_node, &st->rb, leftmost); |
1fa8f6d6 VG |
1264 | } |
1265 | ||
7b5af5cf TM |
1266 | /* |
1267 | * This has to be called only on activation of cfqg | |
1268 | */ | |
1fa8f6d6 | 1269 | static void |
8184f93e JT |
1270 | cfq_update_group_weight(struct cfq_group *cfqg) |
1271 | { | |
3381cb8d | 1272 | if (cfqg->new_weight) { |
8184f93e | 1273 | cfqg->weight = cfqg->new_weight; |
3381cb8d | 1274 | cfqg->new_weight = 0; |
8184f93e | 1275 | } |
e15693ef TM |
1276 | } |
1277 | ||
1278 | static void | |
1279 | cfq_update_group_leaf_weight(struct cfq_group *cfqg) | |
1280 | { | |
1281 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
e71357e1 TH |
1282 | |
1283 | if (cfqg->new_leaf_weight) { | |
1284 | cfqg->leaf_weight = cfqg->new_leaf_weight; | |
1285 | cfqg->new_leaf_weight = 0; | |
1286 | } | |
8184f93e JT |
1287 | } |
1288 | ||
1289 | static void | |
1290 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1291 | { | |
1d3650f7 | 1292 | unsigned int vfr = 1 << CFQ_SERVICE_SHIFT; /* start with 1 */ |
7918ffb5 | 1293 | struct cfq_group *pos = cfqg; |
1d3650f7 | 1294 | struct cfq_group *parent; |
7918ffb5 TH |
1295 | bool propagate; |
1296 | ||
1297 | /* add to the service tree */ | |
8184f93e JT |
1298 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
1299 | ||
7b5af5cf TM |
1300 | /* |
1301 | * Update leaf_weight. We cannot update weight at this point | |
1302 | * because cfqg might already have been activated and is | |
1303 | * contributing its current weight to the parent's child_weight. | |
1304 | */ | |
e15693ef | 1305 | cfq_update_group_leaf_weight(cfqg); |
8184f93e | 1306 | __cfq_group_service_tree_add(st, cfqg); |
7918ffb5 TH |
1307 | |
1308 | /* | |
1d3650f7 TH |
1309 | * Activate @cfqg and calculate the portion of vfraction @cfqg is |
1310 | * entitled to. vfraction is calculated by walking the tree | |
1311 | * towards the root calculating the fraction it has at each level. | |
1312 | * The compounded ratio is how much vfraction @cfqg owns. | |
1313 | * | |
1314 | * Start with the proportion tasks in this cfqg has against active | |
1315 | * children cfqgs - its leaf_weight against children_weight. | |
7918ffb5 TH |
1316 | */ |
1317 | propagate = !pos->nr_active++; | |
1318 | pos->children_weight += pos->leaf_weight; | |
1d3650f7 | 1319 | vfr = vfr * pos->leaf_weight / pos->children_weight; |
7918ffb5 | 1320 | |
1d3650f7 TH |
1321 | /* |
1322 | * Compound ->weight walking up the tree. Both activation and | |
1323 | * vfraction calculation are done in the same loop. Propagation | |
1324 | * stops once an already activated node is met. vfraction | |
1325 | * calculation should always continue to the root. | |
1326 | */ | |
d02f7aa8 | 1327 | while ((parent = cfqg_parent(pos))) { |
1d3650f7 | 1328 | if (propagate) { |
e15693ef | 1329 | cfq_update_group_weight(pos); |
1d3650f7 TH |
1330 | propagate = !parent->nr_active++; |
1331 | parent->children_weight += pos->weight; | |
1332 | } | |
1333 | vfr = vfr * pos->weight / parent->children_weight; | |
7918ffb5 TH |
1334 | pos = parent; |
1335 | } | |
1d3650f7 TH |
1336 | |
1337 | cfqg->vfraction = max_t(unsigned, vfr, 1); | |
8184f93e JT |
1338 | } |
1339 | ||
5be6b756 HT |
1340 | static inline u64 cfq_get_cfqg_vdisktime_delay(struct cfq_data *cfqd) |
1341 | { | |
1342 | if (!iops_mode(cfqd)) | |
1343 | return CFQ_SLICE_MODE_GROUP_DELAY; | |
1344 | else | |
1345 | return CFQ_IOPS_MODE_GROUP_DELAY; | |
1346 | } | |
1347 | ||
8184f93e JT |
1348 | static void |
1349 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
1350 | { |
1351 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1352 | struct cfq_group *__cfqg; | |
1353 | struct rb_node *n; | |
1354 | ||
1355 | cfqg->nr_cfqq++; | |
760701bf | 1356 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
1357 | return; |
1358 | ||
1359 | /* | |
1360 | * Currently put the group at the end. Later implement something | |
1361 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 1362 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 | 1363 | */ |
f0f1a45f | 1364 | n = st->rb_rightmost; |
1fa8f6d6 VG |
1365 | if (n) { |
1366 | __cfqg = rb_entry_cfqg(n); | |
5be6b756 HT |
1367 | cfqg->vdisktime = __cfqg->vdisktime + |
1368 | cfq_get_cfqg_vdisktime_delay(cfqd); | |
1fa8f6d6 VG |
1369 | } else |
1370 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
1371 | cfq_group_service_tree_add(st, cfqg); |
1372 | } | |
1fa8f6d6 | 1373 | |
8184f93e JT |
1374 | static void |
1375 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1376 | { | |
7918ffb5 TH |
1377 | struct cfq_group *pos = cfqg; |
1378 | bool propagate; | |
1379 | ||
1380 | /* | |
1381 | * Undo activation from cfq_group_service_tree_add(). Deactivate | |
1382 | * @cfqg and propagate deactivation upwards. | |
1383 | */ | |
1384 | propagate = !--pos->nr_active; | |
1385 | pos->children_weight -= pos->leaf_weight; | |
1386 | ||
1387 | while (propagate) { | |
d02f7aa8 | 1388 | struct cfq_group *parent = cfqg_parent(pos); |
7918ffb5 TH |
1389 | |
1390 | /* @pos has 0 nr_active at this point */ | |
1391 | WARN_ON_ONCE(pos->children_weight); | |
1d3650f7 | 1392 | pos->vfraction = 0; |
7918ffb5 TH |
1393 | |
1394 | if (!parent) | |
1395 | break; | |
1396 | ||
1397 | propagate = !--parent->nr_active; | |
1398 | parent->children_weight -= pos->weight; | |
1399 | pos = parent; | |
1400 | } | |
1401 | ||
1402 | /* remove from the service tree */ | |
8184f93e JT |
1403 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1404 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
1405 | } |
1406 | ||
1407 | static void | |
8184f93e | 1408 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
1409 | { |
1410 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1411 | ||
1412 | BUG_ON(cfqg->nr_cfqq < 1); | |
1413 | cfqg->nr_cfqq--; | |
25bc6b07 | 1414 | |
1fa8f6d6 VG |
1415 | /* If there are other cfq queues under this group, don't delete it */ |
1416 | if (cfqg->nr_cfqq) | |
1417 | return; | |
1418 | ||
2868ef7b | 1419 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 1420 | cfq_group_service_tree_del(st, cfqg); |
4d2ceea4 | 1421 | cfqg->saved_wl_slice = 0; |
155fead9 | 1422 | cfqg_stats_update_dequeue(cfqg); |
dae739eb VG |
1423 | } |
1424 | ||
9a7f38c4 JM |
1425 | static inline u64 cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
1426 | u64 *unaccounted_time) | |
dae739eb | 1427 | { |
9a7f38c4 JM |
1428 | u64 slice_used; |
1429 | u64 now = ktime_get_ns(); | |
dae739eb VG |
1430 | |
1431 | /* | |
1432 | * Queue got expired before even a single request completed or | |
1433 | * got expired immediately after first request completion. | |
1434 | */ | |
9a7f38c4 | 1435 | if (!cfqq->slice_start || cfqq->slice_start == now) { |
dae739eb VG |
1436 | /* |
1437 | * Also charge the seek time incurred to the group, otherwise | |
1438 | * if there are mutiple queues in the group, each can dispatch | |
1439 | * a single request on seeky media and cause lots of seek time | |
1440 | * and group will never know it. | |
1441 | */ | |
0b31c10c JK |
1442 | slice_used = max_t(u64, (now - cfqq->dispatch_start), |
1443 | jiffies_to_nsecs(1)); | |
dae739eb | 1444 | } else { |
9a7f38c4 | 1445 | slice_used = now - cfqq->slice_start; |
167400d3 JT |
1446 | if (slice_used > cfqq->allocated_slice) { |
1447 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 1448 | slice_used = cfqq->allocated_slice; |
167400d3 | 1449 | } |
9a7f38c4 | 1450 | if (cfqq->slice_start > cfqq->dispatch_start) |
167400d3 JT |
1451 | *unaccounted_time += cfqq->slice_start - |
1452 | cfqq->dispatch_start; | |
dae739eb VG |
1453 | } |
1454 | ||
dae739eb VG |
1455 | return slice_used; |
1456 | } | |
1457 | ||
1458 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1459 | struct cfq_queue *cfqq) |
dae739eb VG |
1460 | { |
1461 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
9a7f38c4 | 1462 | u64 used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1463 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1464 | - cfqg->service_tree_idle.count; | |
1d3650f7 | 1465 | unsigned int vfr; |
9a7f38c4 | 1466 | u64 now = ktime_get_ns(); |
f26bd1f0 VG |
1467 | |
1468 | BUG_ON(nr_sync < 0); | |
167400d3 | 1469 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1470 | |
02b35081 VG |
1471 | if (iops_mode(cfqd)) |
1472 | charge = cfqq->slice_dispatch; | |
1473 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1474 | charge = cfqq->allocated_slice; | |
dae739eb | 1475 | |
1d3650f7 TH |
1476 | /* |
1477 | * Can't update vdisktime while on service tree and cfqg->vfraction | |
1478 | * is valid only while on it. Cache vfr, leave the service tree, | |
1479 | * update vdisktime and go back on. The re-addition to the tree | |
1480 | * will also update the weights as necessary. | |
1481 | */ | |
1482 | vfr = cfqg->vfraction; | |
8184f93e | 1483 | cfq_group_service_tree_del(st, cfqg); |
1d3650f7 | 1484 | cfqg->vdisktime += cfqg_scale_charge(charge, vfr); |
8184f93e | 1485 | cfq_group_service_tree_add(st, cfqg); |
dae739eb VG |
1486 | |
1487 | /* This group is being expired. Save the context */ | |
9a7f38c4 JM |
1488 | if (cfqd->workload_expires > now) { |
1489 | cfqg->saved_wl_slice = cfqd->workload_expires - now; | |
4d2ceea4 VG |
1490 | cfqg->saved_wl_type = cfqd->serving_wl_type; |
1491 | cfqg->saved_wl_class = cfqd->serving_wl_class; | |
dae739eb | 1492 | } else |
4d2ceea4 | 1493 | cfqg->saved_wl_slice = 0; |
2868ef7b VG |
1494 | |
1495 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1496 | st->min_vdisktime); | |
fd16d263 | 1497 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
9a7f38c4 | 1498 | "sl_used=%llu disp=%llu charge=%llu iops=%u sect=%lu", |
fd16d263 JP |
1499 | used_sl, cfqq->slice_dispatch, charge, |
1500 | iops_mode(cfqd), cfqq->nr_sectors); | |
155fead9 TH |
1501 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
1502 | cfqg_stats_set_start_empty_time(cfqg); | |
1fa8f6d6 VG |
1503 | } |
1504 | ||
f51b802c TH |
1505 | /** |
1506 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1507 | * @cfqg: cfq_group to initialize | |
1508 | * | |
1509 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1510 | * is enabled or not. | |
1511 | */ | |
1512 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1513 | { | |
1514 | struct cfq_rb_root *st; | |
1515 | int i, j; | |
1516 | ||
1517 | for_each_cfqg_st(cfqg, i, j, st) | |
1518 | *st = CFQ_RB_ROOT; | |
1519 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1520 | ||
9a7f38c4 | 1521 | cfqg->ttime.last_end_request = ktime_get_ns(); |
f51b802c TH |
1522 | } |
1523 | ||
25fb5169 | 1524 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
69d7fde5 TH |
1525 | static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val, |
1526 | bool on_dfl, bool reset_dev, bool is_leaf_weight); | |
1527 | ||
24bdb8ef | 1528 | static void cfqg_stats_exit(struct cfqg_stats *stats) |
90d3839b | 1529 | { |
24bdb8ef TH |
1530 | blkg_rwstat_exit(&stats->merged); |
1531 | blkg_rwstat_exit(&stats->service_time); | |
1532 | blkg_rwstat_exit(&stats->wait_time); | |
1533 | blkg_rwstat_exit(&stats->queued); | |
24bdb8ef TH |
1534 | blkg_stat_exit(&stats->time); |
1535 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
1536 | blkg_stat_exit(&stats->unaccounted_time); | |
1537 | blkg_stat_exit(&stats->avg_queue_size_sum); | |
1538 | blkg_stat_exit(&stats->avg_queue_size_samples); | |
1539 | blkg_stat_exit(&stats->dequeue); | |
1540 | blkg_stat_exit(&stats->group_wait_time); | |
1541 | blkg_stat_exit(&stats->idle_time); | |
1542 | blkg_stat_exit(&stats->empty_time); | |
1543 | #endif | |
1544 | } | |
1545 | ||
1546 | static int cfqg_stats_init(struct cfqg_stats *stats, gfp_t gfp) | |
1547 | { | |
77ea7338 | 1548 | if (blkg_rwstat_init(&stats->merged, gfp) || |
24bdb8ef TH |
1549 | blkg_rwstat_init(&stats->service_time, gfp) || |
1550 | blkg_rwstat_init(&stats->wait_time, gfp) || | |
1551 | blkg_rwstat_init(&stats->queued, gfp) || | |
24bdb8ef TH |
1552 | blkg_stat_init(&stats->time, gfp)) |
1553 | goto err; | |
90d3839b PZ |
1554 | |
1555 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
24bdb8ef TH |
1556 | if (blkg_stat_init(&stats->unaccounted_time, gfp) || |
1557 | blkg_stat_init(&stats->avg_queue_size_sum, gfp) || | |
1558 | blkg_stat_init(&stats->avg_queue_size_samples, gfp) || | |
1559 | blkg_stat_init(&stats->dequeue, gfp) || | |
1560 | blkg_stat_init(&stats->group_wait_time, gfp) || | |
1561 | blkg_stat_init(&stats->idle_time, gfp) || | |
1562 | blkg_stat_init(&stats->empty_time, gfp)) | |
1563 | goto err; | |
90d3839b | 1564 | #endif |
24bdb8ef TH |
1565 | return 0; |
1566 | err: | |
1567 | cfqg_stats_exit(stats); | |
1568 | return -ENOMEM; | |
90d3839b PZ |
1569 | } |
1570 | ||
e4a9bde9 TH |
1571 | static struct blkcg_policy_data *cfq_cpd_alloc(gfp_t gfp) |
1572 | { | |
1573 | struct cfq_group_data *cgd; | |
1574 | ||
ebc4ff66 | 1575 | cgd = kzalloc(sizeof(*cgd), gfp); |
e4a9bde9 TH |
1576 | if (!cgd) |
1577 | return NULL; | |
1578 | return &cgd->cpd; | |
1579 | } | |
1580 | ||
81437648 | 1581 | static void cfq_cpd_init(struct blkcg_policy_data *cpd) |
e48453c3 | 1582 | { |
81437648 | 1583 | struct cfq_group_data *cgd = cpd_to_cfqgd(cpd); |
9e10a130 | 1584 | unsigned int weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ? |
69d7fde5 | 1585 | CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL; |
e48453c3 | 1586 | |
69d7fde5 TH |
1587 | if (cpd_to_blkcg(cpd) == &blkcg_root) |
1588 | weight *= 2; | |
1589 | ||
1590 | cgd->weight = weight; | |
1591 | cgd->leaf_weight = weight; | |
e48453c3 AA |
1592 | } |
1593 | ||
e4a9bde9 TH |
1594 | static void cfq_cpd_free(struct blkcg_policy_data *cpd) |
1595 | { | |
1596 | kfree(cpd_to_cfqgd(cpd)); | |
1597 | } | |
1598 | ||
69d7fde5 TH |
1599 | static void cfq_cpd_bind(struct blkcg_policy_data *cpd) |
1600 | { | |
1601 | struct blkcg *blkcg = cpd_to_blkcg(cpd); | |
9e10a130 | 1602 | bool on_dfl = cgroup_subsys_on_dfl(io_cgrp_subsys); |
69d7fde5 TH |
1603 | unsigned int weight = on_dfl ? CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL; |
1604 | ||
1605 | if (blkcg == &blkcg_root) | |
1606 | weight *= 2; | |
1607 | ||
1608 | WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, false)); | |
1609 | WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, true)); | |
1610 | } | |
1611 | ||
001bea73 TH |
1612 | static struct blkg_policy_data *cfq_pd_alloc(gfp_t gfp, int node) |
1613 | { | |
b2ce2643 TH |
1614 | struct cfq_group *cfqg; |
1615 | ||
1616 | cfqg = kzalloc_node(sizeof(*cfqg), gfp, node); | |
1617 | if (!cfqg) | |
1618 | return NULL; | |
1619 | ||
1620 | cfq_init_cfqg_base(cfqg); | |
24bdb8ef TH |
1621 | if (cfqg_stats_init(&cfqg->stats, gfp)) { |
1622 | kfree(cfqg); | |
1623 | return NULL; | |
1624 | } | |
b2ce2643 TH |
1625 | |
1626 | return &cfqg->pd; | |
001bea73 TH |
1627 | } |
1628 | ||
a9520cd6 | 1629 | static void cfq_pd_init(struct blkg_policy_data *pd) |
f469a7b4 | 1630 | { |
a9520cd6 TH |
1631 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
1632 | struct cfq_group_data *cgd = blkcg_to_cfqgd(pd->blkg->blkcg); | |
25fb5169 | 1633 | |
e48453c3 AA |
1634 | cfqg->weight = cgd->weight; |
1635 | cfqg->leaf_weight = cgd->leaf_weight; | |
25fb5169 VG |
1636 | } |
1637 | ||
a9520cd6 | 1638 | static void cfq_pd_offline(struct blkg_policy_data *pd) |
0b39920b | 1639 | { |
a9520cd6 | 1640 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
60a83707 TH |
1641 | int i; |
1642 | ||
1643 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
1644 | if (cfqg->async_cfqq[0][i]) | |
1645 | cfq_put_queue(cfqg->async_cfqq[0][i]); | |
1646 | if (cfqg->async_cfqq[1][i]) | |
1647 | cfq_put_queue(cfqg->async_cfqq[1][i]); | |
1648 | } | |
1649 | ||
1650 | if (cfqg->async_idle_cfqq) | |
1651 | cfq_put_queue(cfqg->async_idle_cfqq); | |
1652 | ||
0b39920b TH |
1653 | /* |
1654 | * @blkg is going offline and will be ignored by | |
1655 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
1656 | * that they don't get lost. If IOs complete after this point, the | |
1657 | * stats for them will be lost. Oh well... | |
1658 | */ | |
60a83707 | 1659 | cfqg_stats_xfer_dead(cfqg); |
0b39920b TH |
1660 | } |
1661 | ||
001bea73 TH |
1662 | static void cfq_pd_free(struct blkg_policy_data *pd) |
1663 | { | |
24bdb8ef TH |
1664 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
1665 | ||
1666 | cfqg_stats_exit(&cfqg->stats); | |
1667 | return kfree(cfqg); | |
001bea73 TH |
1668 | } |
1669 | ||
a9520cd6 | 1670 | static void cfq_pd_reset_stats(struct blkg_policy_data *pd) |
689665af | 1671 | { |
a9520cd6 | 1672 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
689665af TH |
1673 | |
1674 | cfqg_stats_reset(&cfqg->stats); | |
25fb5169 VG |
1675 | } |
1676 | ||
ae118896 TH |
1677 | static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd, |
1678 | struct blkcg *blkcg) | |
25fb5169 | 1679 | { |
ae118896 | 1680 | struct blkcg_gq *blkg; |
f469a7b4 | 1681 | |
ae118896 TH |
1682 | blkg = blkg_lookup(blkcg, cfqd->queue); |
1683 | if (likely(blkg)) | |
1684 | return blkg_to_cfqg(blkg); | |
1685 | return NULL; | |
25fb5169 VG |
1686 | } |
1687 | ||
1688 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1689 | { | |
25fb5169 | 1690 | cfqq->cfqg = cfqg; |
b1c35769 | 1691 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1692 | cfqg_get(cfqg); |
b1c35769 VG |
1693 | } |
1694 | ||
f95a04af TH |
1695 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, |
1696 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1697 | { |
f95a04af | 1698 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
3381cb8d TH |
1699 | |
1700 | if (!cfqg->dev_weight) | |
60c2bc2d | 1701 | return 0; |
f95a04af | 1702 | return __blkg_prfill_u64(sf, pd, cfqg->dev_weight); |
60c2bc2d TH |
1703 | } |
1704 | ||
2da8ca82 | 1705 | static int cfqg_print_weight_device(struct seq_file *sf, void *v) |
60c2bc2d | 1706 | { |
2da8ca82 TH |
1707 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1708 | cfqg_prfill_weight_device, &blkcg_policy_cfq, | |
1709 | 0, false); | |
60c2bc2d TH |
1710 | return 0; |
1711 | } | |
1712 | ||
e71357e1 TH |
1713 | static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf, |
1714 | struct blkg_policy_data *pd, int off) | |
1715 | { | |
1716 | struct cfq_group *cfqg = pd_to_cfqg(pd); | |
1717 | ||
1718 | if (!cfqg->dev_leaf_weight) | |
1719 | return 0; | |
1720 | return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight); | |
1721 | } | |
1722 | ||
2da8ca82 | 1723 | static int cfqg_print_leaf_weight_device(struct seq_file *sf, void *v) |
e71357e1 | 1724 | { |
2da8ca82 TH |
1725 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1726 | cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq, | |
1727 | 0, false); | |
e71357e1 TH |
1728 | return 0; |
1729 | } | |
1730 | ||
2da8ca82 | 1731 | static int cfq_print_weight(struct seq_file *sf, void *v) |
60c2bc2d | 1732 | { |
e48453c3 | 1733 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
9470e4a6 JA |
1734 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); |
1735 | unsigned int val = 0; | |
e48453c3 | 1736 | |
9470e4a6 JA |
1737 | if (cgd) |
1738 | val = cgd->weight; | |
1739 | ||
1740 | seq_printf(sf, "%u\n", val); | |
60c2bc2d TH |
1741 | return 0; |
1742 | } | |
1743 | ||
2da8ca82 | 1744 | static int cfq_print_leaf_weight(struct seq_file *sf, void *v) |
e71357e1 | 1745 | { |
e48453c3 | 1746 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
9470e4a6 JA |
1747 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); |
1748 | unsigned int val = 0; | |
1749 | ||
1750 | if (cgd) | |
1751 | val = cgd->leaf_weight; | |
e48453c3 | 1752 | |
9470e4a6 | 1753 | seq_printf(sf, "%u\n", val); |
e71357e1 TH |
1754 | return 0; |
1755 | } | |
1756 | ||
451af504 TH |
1757 | static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of, |
1758 | char *buf, size_t nbytes, loff_t off, | |
2ee867dc | 1759 | bool on_dfl, bool is_leaf_weight) |
60c2bc2d | 1760 | { |
69d7fde5 TH |
1761 | unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN; |
1762 | unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX; | |
451af504 | 1763 | struct blkcg *blkcg = css_to_blkcg(of_css(of)); |
60c2bc2d | 1764 | struct blkg_conf_ctx ctx; |
3381cb8d | 1765 | struct cfq_group *cfqg; |
e48453c3 | 1766 | struct cfq_group_data *cfqgd; |
60c2bc2d | 1767 | int ret; |
36aa9e5f | 1768 | u64 v; |
60c2bc2d | 1769 | |
3c798398 | 1770 | ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx); |
60c2bc2d TH |
1771 | if (ret) |
1772 | return ret; | |
1773 | ||
2ee867dc TH |
1774 | if (sscanf(ctx.body, "%llu", &v) == 1) { |
1775 | /* require "default" on dfl */ | |
1776 | ret = -ERANGE; | |
1777 | if (!v && on_dfl) | |
1778 | goto out_finish; | |
1779 | } else if (!strcmp(strim(ctx.body), "default")) { | |
1780 | v = 0; | |
1781 | } else { | |
1782 | ret = -EINVAL; | |
36aa9e5f | 1783 | goto out_finish; |
2ee867dc | 1784 | } |
36aa9e5f | 1785 | |
3381cb8d | 1786 | cfqg = blkg_to_cfqg(ctx.blkg); |
e48453c3 | 1787 | cfqgd = blkcg_to_cfqgd(blkcg); |
ae994ea9 | 1788 | |
20386ce0 | 1789 | ret = -ERANGE; |
69d7fde5 | 1790 | if (!v || (v >= min && v <= max)) { |
e71357e1 | 1791 | if (!is_leaf_weight) { |
36aa9e5f TH |
1792 | cfqg->dev_weight = v; |
1793 | cfqg->new_weight = v ?: cfqgd->weight; | |
e71357e1 | 1794 | } else { |
36aa9e5f TH |
1795 | cfqg->dev_leaf_weight = v; |
1796 | cfqg->new_leaf_weight = v ?: cfqgd->leaf_weight; | |
e71357e1 | 1797 | } |
60c2bc2d TH |
1798 | ret = 0; |
1799 | } | |
36aa9e5f | 1800 | out_finish: |
60c2bc2d | 1801 | blkg_conf_finish(&ctx); |
451af504 | 1802 | return ret ?: nbytes; |
60c2bc2d TH |
1803 | } |
1804 | ||
451af504 TH |
1805 | static ssize_t cfqg_set_weight_device(struct kernfs_open_file *of, |
1806 | char *buf, size_t nbytes, loff_t off) | |
e71357e1 | 1807 | { |
2ee867dc | 1808 | return __cfqg_set_weight_device(of, buf, nbytes, off, false, false); |
e71357e1 TH |
1809 | } |
1810 | ||
451af504 TH |
1811 | static ssize_t cfqg_set_leaf_weight_device(struct kernfs_open_file *of, |
1812 | char *buf, size_t nbytes, loff_t off) | |
e71357e1 | 1813 | { |
2ee867dc | 1814 | return __cfqg_set_weight_device(of, buf, nbytes, off, false, true); |
e71357e1 TH |
1815 | } |
1816 | ||
dd165eb3 | 1817 | static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val, |
69d7fde5 | 1818 | bool on_dfl, bool reset_dev, bool is_leaf_weight) |
60c2bc2d | 1819 | { |
69d7fde5 TH |
1820 | unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN; |
1821 | unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX; | |
182446d0 | 1822 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 1823 | struct blkcg_gq *blkg; |
e48453c3 | 1824 | struct cfq_group_data *cfqgd; |
ae994ea9 | 1825 | int ret = 0; |
60c2bc2d | 1826 | |
69d7fde5 TH |
1827 | if (val < min || val > max) |
1828 | return -ERANGE; | |
60c2bc2d TH |
1829 | |
1830 | spin_lock_irq(&blkcg->lock); | |
e48453c3 | 1831 | cfqgd = blkcg_to_cfqgd(blkcg); |
ae994ea9 JA |
1832 | if (!cfqgd) { |
1833 | ret = -EINVAL; | |
1834 | goto out; | |
1835 | } | |
e71357e1 TH |
1836 | |
1837 | if (!is_leaf_weight) | |
e48453c3 | 1838 | cfqgd->weight = val; |
e71357e1 | 1839 | else |
e48453c3 | 1840 | cfqgd->leaf_weight = val; |
60c2bc2d | 1841 | |
b67bfe0d | 1842 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
3381cb8d | 1843 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
60c2bc2d | 1844 | |
e71357e1 TH |
1845 | if (!cfqg) |
1846 | continue; | |
1847 | ||
1848 | if (!is_leaf_weight) { | |
69d7fde5 TH |
1849 | if (reset_dev) |
1850 | cfqg->dev_weight = 0; | |
e71357e1 | 1851 | if (!cfqg->dev_weight) |
e48453c3 | 1852 | cfqg->new_weight = cfqgd->weight; |
e71357e1 | 1853 | } else { |
69d7fde5 TH |
1854 | if (reset_dev) |
1855 | cfqg->dev_leaf_weight = 0; | |
e71357e1 | 1856 | if (!cfqg->dev_leaf_weight) |
e48453c3 | 1857 | cfqg->new_leaf_weight = cfqgd->leaf_weight; |
e71357e1 | 1858 | } |
60c2bc2d TH |
1859 | } |
1860 | ||
ae994ea9 | 1861 | out: |
60c2bc2d | 1862 | spin_unlock_irq(&blkcg->lock); |
ae994ea9 | 1863 | return ret; |
60c2bc2d TH |
1864 | } |
1865 | ||
182446d0 TH |
1866 | static int cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft, |
1867 | u64 val) | |
e71357e1 | 1868 | { |
69d7fde5 | 1869 | return __cfq_set_weight(css, val, false, false, false); |
e71357e1 TH |
1870 | } |
1871 | ||
182446d0 TH |
1872 | static int cfq_set_leaf_weight(struct cgroup_subsys_state *css, |
1873 | struct cftype *cft, u64 val) | |
e71357e1 | 1874 | { |
69d7fde5 | 1875 | return __cfq_set_weight(css, val, false, false, true); |
e71357e1 TH |
1876 | } |
1877 | ||
2da8ca82 | 1878 | static int cfqg_print_stat(struct seq_file *sf, void *v) |
5bc4afb1 | 1879 | { |
2da8ca82 TH |
1880 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat, |
1881 | &blkcg_policy_cfq, seq_cft(sf)->private, false); | |
5bc4afb1 TH |
1882 | return 0; |
1883 | } | |
1884 | ||
2da8ca82 | 1885 | static int cfqg_print_rwstat(struct seq_file *sf, void *v) |
5bc4afb1 | 1886 | { |
2da8ca82 TH |
1887 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat, |
1888 | &blkcg_policy_cfq, seq_cft(sf)->private, true); | |
5bc4afb1 TH |
1889 | return 0; |
1890 | } | |
1891 | ||
43114018 TH |
1892 | static u64 cfqg_prfill_stat_recursive(struct seq_file *sf, |
1893 | struct blkg_policy_data *pd, int off) | |
1894 | { | |
f12c74ca TH |
1895 | u64 sum = blkg_stat_recursive_sum(pd_to_blkg(pd), |
1896 | &blkcg_policy_cfq, off); | |
43114018 TH |
1897 | return __blkg_prfill_u64(sf, pd, sum); |
1898 | } | |
1899 | ||
1900 | static u64 cfqg_prfill_rwstat_recursive(struct seq_file *sf, | |
1901 | struct blkg_policy_data *pd, int off) | |
1902 | { | |
f12c74ca TH |
1903 | struct blkg_rwstat sum = blkg_rwstat_recursive_sum(pd_to_blkg(pd), |
1904 | &blkcg_policy_cfq, off); | |
43114018 TH |
1905 | return __blkg_prfill_rwstat(sf, pd, &sum); |
1906 | } | |
1907 | ||
2da8ca82 | 1908 | static int cfqg_print_stat_recursive(struct seq_file *sf, void *v) |
43114018 | 1909 | { |
2da8ca82 TH |
1910 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1911 | cfqg_prfill_stat_recursive, &blkcg_policy_cfq, | |
1912 | seq_cft(sf)->private, false); | |
43114018 TH |
1913 | return 0; |
1914 | } | |
1915 | ||
2da8ca82 | 1916 | static int cfqg_print_rwstat_recursive(struct seq_file *sf, void *v) |
43114018 | 1917 | { |
2da8ca82 TH |
1918 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1919 | cfqg_prfill_rwstat_recursive, &blkcg_policy_cfq, | |
1920 | seq_cft(sf)->private, true); | |
43114018 TH |
1921 | return 0; |
1922 | } | |
1923 | ||
702747ca TH |
1924 | static u64 cfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd, |
1925 | int off) | |
1926 | { | |
1927 | u64 sum = blkg_rwstat_total(&pd->blkg->stat_bytes); | |
1928 | ||
1929 | return __blkg_prfill_u64(sf, pd, sum >> 9); | |
1930 | } | |
1931 | ||
1932 | static int cfqg_print_stat_sectors(struct seq_file *sf, void *v) | |
1933 | { | |
1934 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1935 | cfqg_prfill_sectors, &blkcg_policy_cfq, 0, false); | |
1936 | return 0; | |
1937 | } | |
1938 | ||
1939 | static u64 cfqg_prfill_sectors_recursive(struct seq_file *sf, | |
1940 | struct blkg_policy_data *pd, int off) | |
1941 | { | |
1942 | struct blkg_rwstat tmp = blkg_rwstat_recursive_sum(pd->blkg, NULL, | |
1943 | offsetof(struct blkcg_gq, stat_bytes)); | |
1944 | u64 sum = atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) + | |
1945 | atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]); | |
1946 | ||
1947 | return __blkg_prfill_u64(sf, pd, sum >> 9); | |
1948 | } | |
1949 | ||
1950 | static int cfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v) | |
1951 | { | |
1952 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1953 | cfqg_prfill_sectors_recursive, &blkcg_policy_cfq, 0, | |
1954 | false); | |
1955 | return 0; | |
1956 | } | |
1957 | ||
60c2bc2d | 1958 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
f95a04af TH |
1959 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, |
1960 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1961 | { |
f95a04af | 1962 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
155fead9 | 1963 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
60c2bc2d TH |
1964 | u64 v = 0; |
1965 | ||
1966 | if (samples) { | |
155fead9 | 1967 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
f3cff25f | 1968 | v = div64_u64(v, samples); |
60c2bc2d | 1969 | } |
f95a04af | 1970 | __blkg_prfill_u64(sf, pd, v); |
60c2bc2d TH |
1971 | return 0; |
1972 | } | |
1973 | ||
1974 | /* print avg_queue_size */ | |
2da8ca82 | 1975 | static int cfqg_print_avg_queue_size(struct seq_file *sf, void *v) |
60c2bc2d | 1976 | { |
2da8ca82 TH |
1977 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1978 | cfqg_prfill_avg_queue_size, &blkcg_policy_cfq, | |
1979 | 0, false); | |
60c2bc2d TH |
1980 | return 0; |
1981 | } | |
1982 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1983 | ||
880f50e2 | 1984 | static struct cftype cfq_blkcg_legacy_files[] = { |
1d3650f7 | 1985 | /* on root, weight is mapped to leaf_weight */ |
60c2bc2d TH |
1986 | { |
1987 | .name = "weight_device", | |
1d3650f7 | 1988 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 1989 | .seq_show = cfqg_print_leaf_weight_device, |
451af504 | 1990 | .write = cfqg_set_leaf_weight_device, |
60c2bc2d TH |
1991 | }, |
1992 | { | |
1993 | .name = "weight", | |
1d3650f7 | 1994 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 1995 | .seq_show = cfq_print_leaf_weight, |
1d3650f7 | 1996 | .write_u64 = cfq_set_leaf_weight, |
60c2bc2d | 1997 | }, |
e71357e1 | 1998 | |
1d3650f7 | 1999 | /* no such mapping necessary for !roots */ |
60c2bc2d TH |
2000 | { |
2001 | .name = "weight_device", | |
1d3650f7 | 2002 | .flags = CFTYPE_NOT_ON_ROOT, |
2da8ca82 | 2003 | .seq_show = cfqg_print_weight_device, |
451af504 | 2004 | .write = cfqg_set_weight_device, |
60c2bc2d TH |
2005 | }, |
2006 | { | |
2007 | .name = "weight", | |
1d3650f7 | 2008 | .flags = CFTYPE_NOT_ON_ROOT, |
2da8ca82 | 2009 | .seq_show = cfq_print_weight, |
3381cb8d | 2010 | .write_u64 = cfq_set_weight, |
60c2bc2d | 2011 | }, |
e71357e1 | 2012 | |
e71357e1 TH |
2013 | { |
2014 | .name = "leaf_weight_device", | |
2da8ca82 | 2015 | .seq_show = cfqg_print_leaf_weight_device, |
451af504 | 2016 | .write = cfqg_set_leaf_weight_device, |
e71357e1 TH |
2017 | }, |
2018 | { | |
2019 | .name = "leaf_weight", | |
2da8ca82 | 2020 | .seq_show = cfq_print_leaf_weight, |
e71357e1 TH |
2021 | .write_u64 = cfq_set_leaf_weight, |
2022 | }, | |
2023 | ||
43114018 | 2024 | /* statistics, covers only the tasks in the cfqg */ |
60c2bc2d TH |
2025 | { |
2026 | .name = "time", | |
5bc4afb1 | 2027 | .private = offsetof(struct cfq_group, stats.time), |
2da8ca82 | 2028 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2029 | }, |
2030 | { | |
2031 | .name = "sectors", | |
702747ca | 2032 | .seq_show = cfqg_print_stat_sectors, |
60c2bc2d TH |
2033 | }, |
2034 | { | |
2035 | .name = "io_service_bytes", | |
77ea7338 TH |
2036 | .private = (unsigned long)&blkcg_policy_cfq, |
2037 | .seq_show = blkg_print_stat_bytes, | |
60c2bc2d TH |
2038 | }, |
2039 | { | |
2040 | .name = "io_serviced", | |
77ea7338 TH |
2041 | .private = (unsigned long)&blkcg_policy_cfq, |
2042 | .seq_show = blkg_print_stat_ios, | |
60c2bc2d TH |
2043 | }, |
2044 | { | |
2045 | .name = "io_service_time", | |
5bc4afb1 | 2046 | .private = offsetof(struct cfq_group, stats.service_time), |
2da8ca82 | 2047 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2048 | }, |
2049 | { | |
2050 | .name = "io_wait_time", | |
5bc4afb1 | 2051 | .private = offsetof(struct cfq_group, stats.wait_time), |
2da8ca82 | 2052 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2053 | }, |
2054 | { | |
2055 | .name = "io_merged", | |
5bc4afb1 | 2056 | .private = offsetof(struct cfq_group, stats.merged), |
2da8ca82 | 2057 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2058 | }, |
2059 | { | |
2060 | .name = "io_queued", | |
5bc4afb1 | 2061 | .private = offsetof(struct cfq_group, stats.queued), |
2da8ca82 | 2062 | .seq_show = cfqg_print_rwstat, |
60c2bc2d | 2063 | }, |
43114018 TH |
2064 | |
2065 | /* the same statictics which cover the cfqg and its descendants */ | |
2066 | { | |
2067 | .name = "time_recursive", | |
2068 | .private = offsetof(struct cfq_group, stats.time), | |
2da8ca82 | 2069 | .seq_show = cfqg_print_stat_recursive, |
43114018 TH |
2070 | }, |
2071 | { | |
2072 | .name = "sectors_recursive", | |
702747ca | 2073 | .seq_show = cfqg_print_stat_sectors_recursive, |
43114018 TH |
2074 | }, |
2075 | { | |
2076 | .name = "io_service_bytes_recursive", | |
77ea7338 TH |
2077 | .private = (unsigned long)&blkcg_policy_cfq, |
2078 | .seq_show = blkg_print_stat_bytes_recursive, | |
43114018 TH |
2079 | }, |
2080 | { | |
2081 | .name = "io_serviced_recursive", | |
77ea7338 TH |
2082 | .private = (unsigned long)&blkcg_policy_cfq, |
2083 | .seq_show = blkg_print_stat_ios_recursive, | |
43114018 TH |
2084 | }, |
2085 | { | |
2086 | .name = "io_service_time_recursive", | |
2087 | .private = offsetof(struct cfq_group, stats.service_time), | |
2da8ca82 | 2088 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2089 | }, |
2090 | { | |
2091 | .name = "io_wait_time_recursive", | |
2092 | .private = offsetof(struct cfq_group, stats.wait_time), | |
2da8ca82 | 2093 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2094 | }, |
2095 | { | |
2096 | .name = "io_merged_recursive", | |
2097 | .private = offsetof(struct cfq_group, stats.merged), | |
2da8ca82 | 2098 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2099 | }, |
2100 | { | |
2101 | .name = "io_queued_recursive", | |
2102 | .private = offsetof(struct cfq_group, stats.queued), | |
2da8ca82 | 2103 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 | 2104 | }, |
60c2bc2d TH |
2105 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
2106 | { | |
2107 | .name = "avg_queue_size", | |
2da8ca82 | 2108 | .seq_show = cfqg_print_avg_queue_size, |
60c2bc2d TH |
2109 | }, |
2110 | { | |
2111 | .name = "group_wait_time", | |
5bc4afb1 | 2112 | .private = offsetof(struct cfq_group, stats.group_wait_time), |
2da8ca82 | 2113 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2114 | }, |
2115 | { | |
2116 | .name = "idle_time", | |
5bc4afb1 | 2117 | .private = offsetof(struct cfq_group, stats.idle_time), |
2da8ca82 | 2118 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2119 | }, |
2120 | { | |
2121 | .name = "empty_time", | |
5bc4afb1 | 2122 | .private = offsetof(struct cfq_group, stats.empty_time), |
2da8ca82 | 2123 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2124 | }, |
2125 | { | |
2126 | .name = "dequeue", | |
5bc4afb1 | 2127 | .private = offsetof(struct cfq_group, stats.dequeue), |
2da8ca82 | 2128 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2129 | }, |
2130 | { | |
2131 | .name = "unaccounted_time", | |
5bc4afb1 | 2132 | .private = offsetof(struct cfq_group, stats.unaccounted_time), |
2da8ca82 | 2133 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2134 | }, |
2135 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
2136 | { } /* terminate */ | |
2137 | }; | |
2ee867dc TH |
2138 | |
2139 | static int cfq_print_weight_on_dfl(struct seq_file *sf, void *v) | |
2140 | { | |
2141 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
2142 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); | |
2143 | ||
2144 | seq_printf(sf, "default %u\n", cgd->weight); | |
2145 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_weight_device, | |
2146 | &blkcg_policy_cfq, 0, false); | |
2147 | return 0; | |
2148 | } | |
2149 | ||
2150 | static ssize_t cfq_set_weight_on_dfl(struct kernfs_open_file *of, | |
2151 | char *buf, size_t nbytes, loff_t off) | |
2152 | { | |
2153 | char *endp; | |
2154 | int ret; | |
2155 | u64 v; | |
2156 | ||
2157 | buf = strim(buf); | |
2158 | ||
2159 | /* "WEIGHT" or "default WEIGHT" sets the default weight */ | |
2160 | v = simple_strtoull(buf, &endp, 0); | |
2161 | if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) { | |
69d7fde5 | 2162 | ret = __cfq_set_weight(of_css(of), v, true, false, false); |
2ee867dc TH |
2163 | return ret ?: nbytes; |
2164 | } | |
2165 | ||
2166 | /* "MAJ:MIN WEIGHT" */ | |
2167 | return __cfqg_set_weight_device(of, buf, nbytes, off, true, false); | |
2168 | } | |
2169 | ||
2170 | static struct cftype cfq_blkcg_files[] = { | |
2171 | { | |
2172 | .name = "weight", | |
2173 | .flags = CFTYPE_NOT_ON_ROOT, | |
2174 | .seq_show = cfq_print_weight_on_dfl, | |
2175 | .write = cfq_set_weight_on_dfl, | |
2176 | }, | |
2177 | { } /* terminate */ | |
2178 | }; | |
2179 | ||
25fb5169 | 2180 | #else /* GROUP_IOSCHED */ |
ae118896 TH |
2181 | static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd, |
2182 | struct blkcg *blkcg) | |
25fb5169 | 2183 | { |
f51b802c | 2184 | return cfqd->root_group; |
25fb5169 | 2185 | } |
7f1dc8a2 | 2186 | |
25fb5169 VG |
2187 | static inline void |
2188 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
2189 | cfqq->cfqg = cfqg; | |
2190 | } | |
2191 | ||
2192 | #endif /* GROUP_IOSCHED */ | |
2193 | ||
498d3aa2 | 2194 | /* |
c0324a02 | 2195 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
2196 | * requests waiting to be processed. It is sorted in the order that |
2197 | * we will service the queues. | |
2198 | */ | |
a36e71f9 | 2199 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2200 | bool add_front) |
d9e7620e | 2201 | { |
0871714e JA |
2202 | struct rb_node **p, *parent; |
2203 | struct cfq_queue *__cfqq; | |
9a7f38c4 | 2204 | u64 rb_key; |
34b98d03 | 2205 | struct cfq_rb_root *st; |
09663c86 | 2206 | bool leftmost = true; |
dae739eb | 2207 | int new_cfqq = 1; |
9a7f38c4 | 2208 | u64 now = ktime_get_ns(); |
ae30c286 | 2209 | |
34b98d03 | 2210 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq)); |
0871714e JA |
2211 | if (cfq_class_idle(cfqq)) { |
2212 | rb_key = CFQ_IDLE_DELAY; | |
f0f1a45f | 2213 | parent = st->rb_rightmost; |
0871714e JA |
2214 | if (parent && parent != &cfqq->rb_node) { |
2215 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2216 | rb_key += __cfqq->rb_key; | |
2217 | } else | |
9a7f38c4 | 2218 | rb_key += now; |
0871714e | 2219 | } else if (!add_front) { |
b9c8946b JA |
2220 | /* |
2221 | * Get our rb key offset. Subtract any residual slice | |
2222 | * value carried from last service. A negative resid | |
2223 | * count indicates slice overrun, and this should position | |
2224 | * the next service time further away in the tree. | |
2225 | */ | |
9a7f38c4 | 2226 | rb_key = cfq_slice_offset(cfqd, cfqq) + now; |
b9c8946b | 2227 | rb_key -= cfqq->slice_resid; |
edd75ffd | 2228 | cfqq->slice_resid = 0; |
48e025e6 | 2229 | } else { |
9a7f38c4 | 2230 | rb_key = -NSEC_PER_SEC; |
34b98d03 | 2231 | __cfqq = cfq_rb_first(st); |
9a7f38c4 | 2232 | rb_key += __cfqq ? __cfqq->rb_key : now; |
48e025e6 | 2233 | } |
1da177e4 | 2234 | |
d9e7620e | 2235 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 2236 | new_cfqq = 0; |
99f9628a | 2237 | /* |
d9e7620e | 2238 | * same position, nothing more to do |
99f9628a | 2239 | */ |
34b98d03 | 2240 | if (rb_key == cfqq->rb_key && cfqq->service_tree == st) |
d9e7620e | 2241 | return; |
1da177e4 | 2242 | |
aa6f6a3d CZ |
2243 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
2244 | cfqq->service_tree = NULL; | |
1da177e4 | 2245 | } |
d9e7620e | 2246 | |
0871714e | 2247 | parent = NULL; |
34b98d03 | 2248 | cfqq->service_tree = st; |
09663c86 | 2249 | p = &st->rb.rb_root.rb_node; |
d9e7620e JA |
2250 | while (*p) { |
2251 | parent = *p; | |
2252 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2253 | ||
0c534e0a | 2254 | /* |
c0324a02 | 2255 | * sort by key, that represents service time. |
0c534e0a | 2256 | */ |
9a7f38c4 | 2257 | if (rb_key < __cfqq->rb_key) |
1f23f121 | 2258 | p = &parent->rb_left; |
c0324a02 | 2259 | else { |
1f23f121 | 2260 | p = &parent->rb_right; |
09663c86 | 2261 | leftmost = false; |
c0324a02 | 2262 | } |
d9e7620e JA |
2263 | } |
2264 | ||
2265 | cfqq->rb_key = rb_key; | |
2266 | rb_link_node(&cfqq->rb_node, parent, p); | |
09663c86 | 2267 | rb_insert_color_cached(&cfqq->rb_node, &st->rb, leftmost); |
34b98d03 | 2268 | st->count++; |
20359f27 | 2269 | if (add_front || !new_cfqq) |
dae739eb | 2270 | return; |
8184f93e | 2271 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
2272 | } |
2273 | ||
a36e71f9 | 2274 | static struct cfq_queue * |
f2d1f0ae JA |
2275 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
2276 | sector_t sector, struct rb_node **ret_parent, | |
2277 | struct rb_node ***rb_link) | |
a36e71f9 | 2278 | { |
a36e71f9 JA |
2279 | struct rb_node **p, *parent; |
2280 | struct cfq_queue *cfqq = NULL; | |
2281 | ||
2282 | parent = NULL; | |
2283 | p = &root->rb_node; | |
2284 | while (*p) { | |
2285 | struct rb_node **n; | |
2286 | ||
2287 | parent = *p; | |
2288 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
2289 | ||
2290 | /* | |
2291 | * Sort strictly based on sector. Smallest to the left, | |
2292 | * largest to the right. | |
2293 | */ | |
2e46e8b2 | 2294 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 2295 | n = &(*p)->rb_right; |
2e46e8b2 | 2296 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
2297 | n = &(*p)->rb_left; |
2298 | else | |
2299 | break; | |
2300 | p = n; | |
3ac6c9f8 | 2301 | cfqq = NULL; |
a36e71f9 JA |
2302 | } |
2303 | ||
2304 | *ret_parent = parent; | |
2305 | if (rb_link) | |
2306 | *rb_link = p; | |
3ac6c9f8 | 2307 | return cfqq; |
a36e71f9 JA |
2308 | } |
2309 | ||
2310 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2311 | { | |
a36e71f9 JA |
2312 | struct rb_node **p, *parent; |
2313 | struct cfq_queue *__cfqq; | |
2314 | ||
f2d1f0ae JA |
2315 | if (cfqq->p_root) { |
2316 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2317 | cfqq->p_root = NULL; | |
2318 | } | |
a36e71f9 JA |
2319 | |
2320 | if (cfq_class_idle(cfqq)) | |
2321 | return; | |
2322 | if (!cfqq->next_rq) | |
2323 | return; | |
2324 | ||
f2d1f0ae | 2325 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
2326 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
2327 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
2328 | if (!__cfqq) { |
2329 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
2330 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
2331 | } else | |
2332 | cfqq->p_root = NULL; | |
a36e71f9 JA |
2333 | } |
2334 | ||
498d3aa2 JA |
2335 | /* |
2336 | * Update cfqq's position in the service tree. | |
2337 | */ | |
edd75ffd | 2338 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 2339 | { |
6d048f53 JA |
2340 | /* |
2341 | * Resorting requires the cfqq to be on the RR list already. | |
2342 | */ | |
a36e71f9 | 2343 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 2344 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
2345 | cfq_prio_tree_add(cfqd, cfqq); |
2346 | } | |
6d048f53 JA |
2347 | } |
2348 | ||
1da177e4 LT |
2349 | /* |
2350 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 2351 | * the pending list according to last request service |
1da177e4 | 2352 | */ |
febffd61 | 2353 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2354 | { |
7b679138 | 2355 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
2356 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
2357 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 2358 | cfqd->busy_queues++; |
ef8a41df SL |
2359 | if (cfq_cfqq_sync(cfqq)) |
2360 | cfqd->busy_sync_queues++; | |
1da177e4 | 2361 | |
edd75ffd | 2362 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
2363 | } |
2364 | ||
498d3aa2 JA |
2365 | /* |
2366 | * Called when the cfqq no longer has requests pending, remove it from | |
2367 | * the service tree. | |
2368 | */ | |
febffd61 | 2369 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2370 | { |
7b679138 | 2371 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
2372 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
2373 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 2374 | |
aa6f6a3d CZ |
2375 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
2376 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
2377 | cfqq->service_tree = NULL; | |
2378 | } | |
f2d1f0ae JA |
2379 | if (cfqq->p_root) { |
2380 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2381 | cfqq->p_root = NULL; | |
2382 | } | |
d9e7620e | 2383 | |
8184f93e | 2384 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
2385 | BUG_ON(!cfqd->busy_queues); |
2386 | cfqd->busy_queues--; | |
ef8a41df SL |
2387 | if (cfq_cfqq_sync(cfqq)) |
2388 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
2389 | } |
2390 | ||
2391 | /* | |
2392 | * rb tree support functions | |
2393 | */ | |
febffd61 | 2394 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 2395 | { |
5e705374 | 2396 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 2397 | const int sync = rq_is_sync(rq); |
1da177e4 | 2398 | |
b4878f24 JA |
2399 | BUG_ON(!cfqq->queued[sync]); |
2400 | cfqq->queued[sync]--; | |
1da177e4 | 2401 | |
5e705374 | 2402 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 2403 | |
f04a6424 VG |
2404 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
2405 | /* | |
2406 | * Queue will be deleted from service tree when we actually | |
2407 | * expire it later. Right now just remove it from prio tree | |
2408 | * as it is empty. | |
2409 | */ | |
2410 | if (cfqq->p_root) { | |
2411 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2412 | cfqq->p_root = NULL; | |
2413 | } | |
2414 | } | |
1da177e4 LT |
2415 | } |
2416 | ||
5e705374 | 2417 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 2418 | { |
5e705374 | 2419 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2420 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 2421 | struct request *prev; |
1da177e4 | 2422 | |
5380a101 | 2423 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 2424 | |
796d5116 | 2425 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
2426 | |
2427 | if (!cfq_cfqq_on_rr(cfqq)) | |
2428 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
2429 | |
2430 | /* | |
2431 | * check if this request is a better next-serve candidate | |
2432 | */ | |
a36e71f9 | 2433 | prev = cfqq->next_rq; |
cf7c25cf | 2434 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
2435 | |
2436 | /* | |
2437 | * adjust priority tree position, if ->next_rq changes | |
2438 | */ | |
2439 | if (prev != cfqq->next_rq) | |
2440 | cfq_prio_tree_add(cfqd, cfqq); | |
2441 | ||
5044eed4 | 2442 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
2443 | } |
2444 | ||
febffd61 | 2445 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 2446 | { |
5380a101 JA |
2447 | elv_rb_del(&cfqq->sort_list, rq); |
2448 | cfqq->queued[rq_is_sync(rq)]--; | |
ef295ecf | 2449 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
5e705374 | 2450 | cfq_add_rq_rb(rq); |
155fead9 | 2451 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
ef295ecf | 2452 | rq->cmd_flags); |
1da177e4 LT |
2453 | } |
2454 | ||
206dc69b JA |
2455 | static struct request * |
2456 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 2457 | { |
206dc69b | 2458 | struct task_struct *tsk = current; |
c5869807 | 2459 | struct cfq_io_cq *cic; |
206dc69b | 2460 | struct cfq_queue *cfqq; |
1da177e4 | 2461 | |
4ac845a2 | 2462 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2463 | if (!cic) |
2464 | return NULL; | |
2465 | ||
aa39ebd4 | 2466 | cfqq = cic_to_cfqq(cic, op_is_sync(bio->bi_opf)); |
f73a1c7d KO |
2467 | if (cfqq) |
2468 | return elv_rb_find(&cfqq->sort_list, bio_end_sector(bio)); | |
1da177e4 | 2469 | |
1da177e4 LT |
2470 | return NULL; |
2471 | } | |
2472 | ||
165125e1 | 2473 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2474 | { |
22e2c507 | 2475 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 2476 | |
53c583d2 | 2477 | cfqd->rq_in_driver++; |
7b679138 | 2478 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 2479 | cfqd->rq_in_driver); |
25776e35 | 2480 | |
5b93629b | 2481 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
2482 | } |
2483 | ||
165125e1 | 2484 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2485 | { |
b4878f24 JA |
2486 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2487 | ||
53c583d2 CZ |
2488 | WARN_ON(!cfqd->rq_in_driver); |
2489 | cfqd->rq_in_driver--; | |
7b679138 | 2490 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 2491 | cfqd->rq_in_driver); |
1da177e4 LT |
2492 | } |
2493 | ||
b4878f24 | 2494 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 2495 | { |
5e705374 | 2496 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 2497 | |
5e705374 JA |
2498 | if (cfqq->next_rq == rq) |
2499 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 2500 | |
b4878f24 | 2501 | list_del_init(&rq->queuelist); |
5e705374 | 2502 | cfq_del_rq_rb(rq); |
374f84ac | 2503 | |
45333d5a | 2504 | cfqq->cfqd->rq_queued--; |
ef295ecf | 2505 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
65299a3b CH |
2506 | if (rq->cmd_flags & REQ_PRIO) { |
2507 | WARN_ON(!cfqq->prio_pending); | |
2508 | cfqq->prio_pending--; | |
b53d1ed7 | 2509 | } |
1da177e4 LT |
2510 | } |
2511 | ||
34fe7c05 | 2512 | static enum elv_merge cfq_merge(struct request_queue *q, struct request **req, |
165125e1 | 2513 | struct bio *bio) |
1da177e4 LT |
2514 | { |
2515 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2516 | struct request *__rq; | |
1da177e4 | 2517 | |
206dc69b | 2518 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
72ef799b | 2519 | if (__rq && elv_bio_merge_ok(__rq, bio)) { |
9817064b JA |
2520 | *req = __rq; |
2521 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
2522 | } |
2523 | ||
2524 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
2525 | } |
2526 | ||
165125e1 | 2527 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
34fe7c05 | 2528 | enum elv_merge type) |
1da177e4 | 2529 | { |
21183b07 | 2530 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 2531 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 2532 | |
5e705374 | 2533 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 2534 | } |
1da177e4 LT |
2535 | } |
2536 | ||
812d4026 DS |
2537 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
2538 | struct bio *bio) | |
2539 | { | |
ef295ecf | 2540 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_opf); |
812d4026 DS |
2541 | } |
2542 | ||
1da177e4 | 2543 | static void |
165125e1 | 2544 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
2545 | struct request *next) |
2546 | { | |
cf7c25cf | 2547 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
2548 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2549 | ||
22e2c507 JA |
2550 | /* |
2551 | * reposition in fifo if next is older than rq | |
2552 | */ | |
2553 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
9a7f38c4 | 2554 | next->fifo_time < rq->fifo_time && |
3d106fba | 2555 | cfqq == RQ_CFQQ(next)) { |
22e2c507 | 2556 | list_move(&rq->queuelist, &next->queuelist); |
8b4922d3 | 2557 | rq->fifo_time = next->fifo_time; |
30996f40 | 2558 | } |
22e2c507 | 2559 | |
cf7c25cf CZ |
2560 | if (cfqq->next_rq == next) |
2561 | cfqq->next_rq = rq; | |
b4878f24 | 2562 | cfq_remove_request(next); |
ef295ecf | 2563 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
4a0b75c7 SL |
2564 | |
2565 | cfqq = RQ_CFQQ(next); | |
2566 | /* | |
2567 | * all requests of this queue are merged to other queues, delete it | |
2568 | * from the service tree. If it's the active_queue, | |
2569 | * cfq_dispatch_requests() will choose to expire it or do idle | |
2570 | */ | |
2571 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
2572 | cfqq != cfqd->active_queue) | |
2573 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
2574 | } |
2575 | ||
72ef799b TE |
2576 | static int cfq_allow_bio_merge(struct request_queue *q, struct request *rq, |
2577 | struct bio *bio) | |
da775265 JA |
2578 | { |
2579 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
aa39ebd4 | 2580 | bool is_sync = op_is_sync(bio->bi_opf); |
c5869807 | 2581 | struct cfq_io_cq *cic; |
da775265 | 2582 | struct cfq_queue *cfqq; |
da775265 JA |
2583 | |
2584 | /* | |
ec8acb69 | 2585 | * Disallow merge of a sync bio into an async request. |
da775265 | 2586 | */ |
aa39ebd4 | 2587 | if (is_sync && !rq_is_sync(rq)) |
a6151c3a | 2588 | return false; |
da775265 JA |
2589 | |
2590 | /* | |
f1a4f4d3 | 2591 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 2592 | * merge only if rq is queued there. |
f1a4f4d3 | 2593 | */ |
07c2bd37 TH |
2594 | cic = cfq_cic_lookup(cfqd, current->io_context); |
2595 | if (!cic) | |
2596 | return false; | |
719d3402 | 2597 | |
aa39ebd4 | 2598 | cfqq = cic_to_cfqq(cic, is_sync); |
a6151c3a | 2599 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
2600 | } |
2601 | ||
72ef799b TE |
2602 | static int cfq_allow_rq_merge(struct request_queue *q, struct request *rq, |
2603 | struct request *next) | |
2604 | { | |
2605 | return RQ_CFQQ(rq) == RQ_CFQQ(next); | |
2606 | } | |
2607 | ||
812df48d DS |
2608 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2609 | { | |
91148325 | 2610 | hrtimer_try_to_cancel(&cfqd->idle_slice_timer); |
155fead9 | 2611 | cfqg_stats_update_idle_time(cfqq->cfqg); |
812df48d DS |
2612 | } |
2613 | ||
febffd61 JA |
2614 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
2615 | struct cfq_queue *cfqq) | |
22e2c507 JA |
2616 | { |
2617 | if (cfqq) { | |
3bf10fea | 2618 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d", |
4d2ceea4 | 2619 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
155fead9 | 2620 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
62a37f6b | 2621 | cfqq->slice_start = 0; |
9a7f38c4 | 2622 | cfqq->dispatch_start = ktime_get_ns(); |
62a37f6b JT |
2623 | cfqq->allocated_slice = 0; |
2624 | cfqq->slice_end = 0; | |
2625 | cfqq->slice_dispatch = 0; | |
2626 | cfqq->nr_sectors = 0; | |
2627 | ||
2628 | cfq_clear_cfqq_wait_request(cfqq); | |
2629 | cfq_clear_cfqq_must_dispatch(cfqq); | |
2630 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
2631 | cfq_clear_cfqq_fifo_expire(cfqq); | |
2632 | cfq_mark_cfqq_slice_new(cfqq); | |
2633 | ||
2634 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
2635 | } |
2636 | ||
2637 | cfqd->active_queue = cfqq; | |
2638 | } | |
2639 | ||
7b14e3b5 JA |
2640 | /* |
2641 | * current cfqq expired its slice (or was too idle), select new one | |
2642 | */ | |
2643 | static void | |
2644 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 2645 | bool timed_out) |
7b14e3b5 | 2646 | { |
7b679138 JA |
2647 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
2648 | ||
7b14e3b5 | 2649 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 2650 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 2651 | |
7b14e3b5 | 2652 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 2653 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 2654 | |
ae54abed SL |
2655 | /* |
2656 | * If this cfqq is shared between multiple processes, check to | |
2657 | * make sure that those processes are still issuing I/Os within | |
2658 | * the mean seek distance. If not, it may be time to break the | |
2659 | * queues apart again. | |
2660 | */ | |
2661 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
2662 | cfq_mark_cfqq_split_coop(cfqq); | |
2663 | ||
7b14e3b5 | 2664 | /* |
6084cdda | 2665 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 2666 | */ |
c553f8e3 SL |
2667 | if (timed_out) { |
2668 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 2669 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 2670 | else |
9a7f38c4 | 2671 | cfqq->slice_resid = cfqq->slice_end - ktime_get_ns(); |
93fdf147 | 2672 | cfq_log_cfqq(cfqd, cfqq, "resid=%lld", cfqq->slice_resid); |
7b679138 | 2673 | } |
7b14e3b5 | 2674 | |
e5ff082e | 2675 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 2676 | |
f04a6424 VG |
2677 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
2678 | cfq_del_cfqq_rr(cfqd, cfqq); | |
2679 | ||
edd75ffd | 2680 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
2681 | |
2682 | if (cfqq == cfqd->active_queue) | |
2683 | cfqd->active_queue = NULL; | |
2684 | ||
2685 | if (cfqd->active_cic) { | |
11a3122f | 2686 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
2687 | cfqd->active_cic = NULL; |
2688 | } | |
7b14e3b5 JA |
2689 | } |
2690 | ||
e5ff082e | 2691 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
2692 | { |
2693 | struct cfq_queue *cfqq = cfqd->active_queue; | |
2694 | ||
2695 | if (cfqq) | |
e5ff082e | 2696 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
2697 | } |
2698 | ||
498d3aa2 JA |
2699 | /* |
2700 | * Get next queue for service. Unless we have a queue preemption, | |
2701 | * we'll simply select the first cfqq in the service tree. | |
2702 | */ | |
6d048f53 | 2703 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 2704 | { |
34b98d03 VG |
2705 | struct cfq_rb_root *st = st_for(cfqd->serving_group, |
2706 | cfqd->serving_wl_class, cfqd->serving_wl_type); | |
d9e7620e | 2707 | |
f04a6424 VG |
2708 | if (!cfqd->rq_queued) |
2709 | return NULL; | |
2710 | ||
1fa8f6d6 | 2711 | /* There is nothing to dispatch */ |
34b98d03 | 2712 | if (!st) |
1fa8f6d6 | 2713 | return NULL; |
09663c86 | 2714 | if (RB_EMPTY_ROOT(&st->rb.rb_root)) |
c0324a02 | 2715 | return NULL; |
34b98d03 | 2716 | return cfq_rb_first(st); |
6d048f53 JA |
2717 | } |
2718 | ||
f04a6424 VG |
2719 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
2720 | { | |
25fb5169 | 2721 | struct cfq_group *cfqg; |
f04a6424 VG |
2722 | struct cfq_queue *cfqq; |
2723 | int i, j; | |
2724 | struct cfq_rb_root *st; | |
2725 | ||
2726 | if (!cfqd->rq_queued) | |
2727 | return NULL; | |
2728 | ||
25fb5169 VG |
2729 | cfqg = cfq_get_next_cfqg(cfqd); |
2730 | if (!cfqg) | |
2731 | return NULL; | |
2732 | ||
1cf41753 ME |
2733 | for_each_cfqg_st(cfqg, i, j, st) { |
2734 | cfqq = cfq_rb_first(st); | |
2735 | if (cfqq) | |
f04a6424 | 2736 | return cfqq; |
1cf41753 | 2737 | } |
f04a6424 VG |
2738 | return NULL; |
2739 | } | |
2740 | ||
498d3aa2 JA |
2741 | /* |
2742 | * Get and set a new active queue for service. | |
2743 | */ | |
a36e71f9 JA |
2744 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
2745 | struct cfq_queue *cfqq) | |
6d048f53 | 2746 | { |
e00ef799 | 2747 | if (!cfqq) |
a36e71f9 | 2748 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 2749 | |
22e2c507 | 2750 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 2751 | return cfqq; |
22e2c507 JA |
2752 | } |
2753 | ||
d9e7620e JA |
2754 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
2755 | struct request *rq) | |
2756 | { | |
83096ebf TH |
2757 | if (blk_rq_pos(rq) >= cfqd->last_position) |
2758 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 2759 | else |
83096ebf | 2760 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
2761 | } |
2762 | ||
b2c18e1e | 2763 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 2764 | struct request *rq) |
6d048f53 | 2765 | { |
e9ce335d | 2766 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
2767 | } |
2768 | ||
a36e71f9 JA |
2769 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
2770 | struct cfq_queue *cur_cfqq) | |
2771 | { | |
f2d1f0ae | 2772 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
2773 | struct rb_node *parent, *node; |
2774 | struct cfq_queue *__cfqq; | |
2775 | sector_t sector = cfqd->last_position; | |
2776 | ||
2777 | if (RB_EMPTY_ROOT(root)) | |
2778 | return NULL; | |
2779 | ||
2780 | /* | |
2781 | * First, if we find a request starting at the end of the last | |
2782 | * request, choose it. | |
2783 | */ | |
f2d1f0ae | 2784 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
2785 | if (__cfqq) |
2786 | return __cfqq; | |
2787 | ||
2788 | /* | |
2789 | * If the exact sector wasn't found, the parent of the NULL leaf | |
2790 | * will contain the closest sector. | |
2791 | */ | |
2792 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 2793 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2794 | return __cfqq; |
2795 | ||
2e46e8b2 | 2796 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
2797 | node = rb_next(&__cfqq->p_node); |
2798 | else | |
2799 | node = rb_prev(&__cfqq->p_node); | |
2800 | if (!node) | |
2801 | return NULL; | |
2802 | ||
2803 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 2804 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2805 | return __cfqq; |
2806 | ||
2807 | return NULL; | |
2808 | } | |
2809 | ||
2810 | /* | |
2811 | * cfqd - obvious | |
2812 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
2813 | * closely cooperating with itself. | |
2814 | * | |
2815 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
2816 | * one request, and that cfqd->last_position reflects a position on the disk | |
2817 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
2818 | * assumption. | |
2819 | */ | |
2820 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 2821 | struct cfq_queue *cur_cfqq) |
6d048f53 | 2822 | { |
a36e71f9 JA |
2823 | struct cfq_queue *cfqq; |
2824 | ||
39c01b21 DS |
2825 | if (cfq_class_idle(cur_cfqq)) |
2826 | return NULL; | |
e6c5bc73 JM |
2827 | if (!cfq_cfqq_sync(cur_cfqq)) |
2828 | return NULL; | |
2829 | if (CFQQ_SEEKY(cur_cfqq)) | |
2830 | return NULL; | |
2831 | ||
b9d8f4c7 GJ |
2832 | /* |
2833 | * Don't search priority tree if it's the only queue in the group. | |
2834 | */ | |
2835 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
2836 | return NULL; | |
2837 | ||
6d048f53 | 2838 | /* |
d9e7620e JA |
2839 | * We should notice if some of the queues are cooperating, eg |
2840 | * working closely on the same area of the disk. In that case, | |
2841 | * we can group them together and don't waste time idling. | |
6d048f53 | 2842 | */ |
a36e71f9 JA |
2843 | cfqq = cfqq_close(cfqd, cur_cfqq); |
2844 | if (!cfqq) | |
2845 | return NULL; | |
2846 | ||
8682e1f1 VG |
2847 | /* If new queue belongs to different cfq_group, don't choose it */ |
2848 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
2849 | return NULL; | |
2850 | ||
df5fe3e8 JM |
2851 | /* |
2852 | * It only makes sense to merge sync queues. | |
2853 | */ | |
2854 | if (!cfq_cfqq_sync(cfqq)) | |
2855 | return NULL; | |
e6c5bc73 JM |
2856 | if (CFQQ_SEEKY(cfqq)) |
2857 | return NULL; | |
df5fe3e8 | 2858 | |
c0324a02 CZ |
2859 | /* |
2860 | * Do not merge queues of different priority classes | |
2861 | */ | |
2862 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
2863 | return NULL; | |
2864 | ||
a36e71f9 | 2865 | return cfqq; |
6d048f53 JA |
2866 | } |
2867 | ||
a6d44e98 CZ |
2868 | /* |
2869 | * Determine whether we should enforce idle window for this queue. | |
2870 | */ | |
2871 | ||
2872 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2873 | { | |
3bf10fea | 2874 | enum wl_class_t wl_class = cfqq_class(cfqq); |
34b98d03 | 2875 | struct cfq_rb_root *st = cfqq->service_tree; |
a6d44e98 | 2876 | |
34b98d03 VG |
2877 | BUG_ON(!st); |
2878 | BUG_ON(!st->count); | |
f04a6424 | 2879 | |
b6508c16 VG |
2880 | if (!cfqd->cfq_slice_idle) |
2881 | return false; | |
2882 | ||
a6d44e98 | 2883 | /* We never do for idle class queues. */ |
3bf10fea | 2884 | if (wl_class == IDLE_WORKLOAD) |
a6d44e98 CZ |
2885 | return false; |
2886 | ||
2887 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
2888 | if (cfq_cfqq_idle_window(cfqq) && |
2889 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
2890 | return true; |
2891 | ||
2892 | /* | |
2893 | * Otherwise, we do only if they are the last ones | |
2894 | * in their service tree. | |
2895 | */ | |
34b98d03 VG |
2896 | if (st->count == 1 && cfq_cfqq_sync(cfqq) && |
2897 | !cfq_io_thinktime_big(cfqd, &st->ttime, false)) | |
c1e44756 | 2898 | return true; |
34b98d03 | 2899 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count); |
c1e44756 | 2900 | return false; |
a6d44e98 CZ |
2901 | } |
2902 | ||
6d048f53 | 2903 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2904 | { |
1792669c | 2905 | struct cfq_queue *cfqq = cfqd->active_queue; |
e795421e | 2906 | struct cfq_rb_root *st = cfqq->service_tree; |
c5869807 | 2907 | struct cfq_io_cq *cic; |
9a7f38c4 JM |
2908 | u64 sl, group_idle = 0; |
2909 | u64 now = ktime_get_ns(); | |
7b14e3b5 | 2910 | |
a68bbddb | 2911 | /* |
f7d7b7a7 JA |
2912 | * SSD device without seek penalty, disable idling. But only do so |
2913 | * for devices that support queuing, otherwise we still have a problem | |
2914 | * with sync vs async workloads. | |
a68bbddb | 2915 | */ |
b3193bc0 RH |
2916 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag && |
2917 | !cfqd->cfq_group_idle) | |
a68bbddb JA |
2918 | return; |
2919 | ||
dd67d051 | 2920 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2921 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2922 | |
2923 | /* | |
2924 | * idle is disabled, either manually or by past process history | |
2925 | */ | |
80bdf0c7 VG |
2926 | if (!cfq_should_idle(cfqd, cfqq)) { |
2927 | /* no queue idling. Check for group idling */ | |
2928 | if (cfqd->cfq_group_idle) | |
2929 | group_idle = cfqd->cfq_group_idle; | |
2930 | else | |
2931 | return; | |
2932 | } | |
6d048f53 | 2933 | |
7b679138 | 2934 | /* |
8e550632 | 2935 | * still active requests from this queue, don't idle |
7b679138 | 2936 | */ |
8e550632 | 2937 | if (cfqq->dispatched) |
7b679138 JA |
2938 | return; |
2939 | ||
22e2c507 JA |
2940 | /* |
2941 | * task has exited, don't wait | |
2942 | */ | |
206dc69b | 2943 | cic = cfqd->active_cic; |
f6e8d01b | 2944 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
2945 | return; |
2946 | ||
355b659c CZ |
2947 | /* |
2948 | * If our average think time is larger than the remaining time | |
2949 | * slice, then don't idle. This avoids overrunning the allotted | |
2950 | * time slice. | |
2951 | */ | |
383cd721 | 2952 | if (sample_valid(cic->ttime.ttime_samples) && |
9a7f38c4 JM |
2953 | (cfqq->slice_end - now < cic->ttime.ttime_mean)) { |
2954 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%llu", | |
383cd721 | 2955 | cic->ttime.ttime_mean); |
355b659c | 2956 | return; |
b1ffe737 | 2957 | } |
355b659c | 2958 | |
e795421e JK |
2959 | /* |
2960 | * There are other queues in the group or this is the only group and | |
2961 | * it has too big thinktime, don't do group idle. | |
2962 | */ | |
2963 | if (group_idle && | |
2964 | (cfqq->cfqg->nr_cfqq > 1 || | |
2965 | cfq_io_thinktime_big(cfqd, &st->ttime, true))) | |
80bdf0c7 VG |
2966 | return; |
2967 | ||
3b18152c | 2968 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2969 | |
80bdf0c7 VG |
2970 | if (group_idle) |
2971 | sl = cfqd->cfq_group_idle; | |
2972 | else | |
2973 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2974 | |
91148325 JK |
2975 | hrtimer_start(&cfqd->idle_slice_timer, ns_to_ktime(sl), |
2976 | HRTIMER_MODE_REL); | |
155fead9 | 2977 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
9a7f38c4 | 2978 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %llu group_idle: %d", sl, |
80bdf0c7 | 2979 | group_idle ? 1 : 0); |
1da177e4 LT |
2980 | } |
2981 | ||
498d3aa2 JA |
2982 | /* |
2983 | * Move request from internal lists to the request queue dispatch list. | |
2984 | */ | |
165125e1 | 2985 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2986 | { |
3ed9a296 | 2987 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2988 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2989 | |
7b679138 JA |
2990 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2991 | ||
06d21886 | 2992 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2993 | cfq_remove_request(rq); |
6d048f53 | 2994 | cfqq->dispatched++; |
80bdf0c7 | 2995 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2996 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2997 | |
53c583d2 | 2998 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2999 | cfqq->nr_sectors += blk_rq_sectors(rq); |
1da177e4 LT |
3000 | } |
3001 | ||
3002 | /* | |
3003 | * return expired entry, or NULL to just start from scratch in rbtree | |
3004 | */ | |
febffd61 | 3005 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 3006 | { |
30996f40 | 3007 | struct request *rq = NULL; |
1da177e4 | 3008 | |
3b18152c | 3009 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 3010 | return NULL; |
cb887411 JA |
3011 | |
3012 | cfq_mark_cfqq_fifo_expire(cfqq); | |
3013 | ||
89850f7e JA |
3014 | if (list_empty(&cfqq->fifo)) |
3015 | return NULL; | |
1da177e4 | 3016 | |
89850f7e | 3017 | rq = rq_entry_fifo(cfqq->fifo.next); |
9a7f38c4 | 3018 | if (ktime_get_ns() < rq->fifo_time) |
7b679138 | 3019 | rq = NULL; |
1da177e4 | 3020 | |
6d048f53 | 3021 | return rq; |
1da177e4 LT |
3022 | } |
3023 | ||
22e2c507 JA |
3024 | static inline int |
3025 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3026 | { | |
3027 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 3028 | |
22e2c507 | 3029 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 3030 | |
b9f8ce05 | 3031 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
3032 | } |
3033 | ||
df5fe3e8 JM |
3034 | /* |
3035 | * Must be called with the queue_lock held. | |
3036 | */ | |
3037 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
3038 | { | |
3039 | int process_refs, io_refs; | |
3040 | ||
3041 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 3042 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
3043 | BUG_ON(process_refs < 0); |
3044 | return process_refs; | |
3045 | } | |
3046 | ||
3047 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
3048 | { | |
e6c5bc73 | 3049 | int process_refs, new_process_refs; |
df5fe3e8 JM |
3050 | struct cfq_queue *__cfqq; |
3051 | ||
c10b61f0 JM |
3052 | /* |
3053 | * If there are no process references on the new_cfqq, then it is | |
3054 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
3055 | * chain may have dropped their last reference (not just their | |
3056 | * last process reference). | |
3057 | */ | |
3058 | if (!cfqq_process_refs(new_cfqq)) | |
3059 | return; | |
3060 | ||
df5fe3e8 JM |
3061 | /* Avoid a circular list and skip interim queue merges */ |
3062 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
3063 | if (__cfqq == cfqq) | |
3064 | return; | |
3065 | new_cfqq = __cfqq; | |
3066 | } | |
3067 | ||
3068 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 3069 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
3070 | /* |
3071 | * If the process for the cfqq has gone away, there is no | |
3072 | * sense in merging the queues. | |
3073 | */ | |
c10b61f0 | 3074 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
3075 | return; |
3076 | ||
e6c5bc73 JM |
3077 | /* |
3078 | * Merge in the direction of the lesser amount of work. | |
3079 | */ | |
e6c5bc73 JM |
3080 | if (new_process_refs >= process_refs) { |
3081 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 3082 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
3083 | } else { |
3084 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 3085 | cfqq->ref += new_process_refs; |
e6c5bc73 | 3086 | } |
df5fe3e8 JM |
3087 | } |
3088 | ||
6d816ec7 | 3089 | static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd, |
3bf10fea | 3090 | struct cfq_group *cfqg, enum wl_class_t wl_class) |
718eee05 CZ |
3091 | { |
3092 | struct cfq_queue *queue; | |
3093 | int i; | |
3094 | bool key_valid = false; | |
9a7f38c4 | 3095 | u64 lowest_key = 0; |
718eee05 CZ |
3096 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; |
3097 | ||
65b32a57 VG |
3098 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
3099 | /* select the one with lowest rb_key */ | |
34b98d03 | 3100 | queue = cfq_rb_first(st_for(cfqg, wl_class, i)); |
718eee05 | 3101 | if (queue && |
9a7f38c4 | 3102 | (!key_valid || queue->rb_key < lowest_key)) { |
718eee05 CZ |
3103 | lowest_key = queue->rb_key; |
3104 | cur_best = i; | |
3105 | key_valid = true; | |
3106 | } | |
3107 | } | |
3108 | ||
3109 | return cur_best; | |
3110 | } | |
3111 | ||
6d816ec7 VG |
3112 | static void |
3113 | choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
718eee05 | 3114 | { |
9a7f38c4 | 3115 | u64 slice; |
718eee05 | 3116 | unsigned count; |
cdb16e8f | 3117 | struct cfq_rb_root *st; |
9a7f38c4 | 3118 | u64 group_slice; |
4d2ceea4 | 3119 | enum wl_class_t original_class = cfqd->serving_wl_class; |
9a7f38c4 | 3120 | u64 now = ktime_get_ns(); |
1fa8f6d6 | 3121 | |
718eee05 | 3122 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 3123 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 3124 | cfqd->serving_wl_class = RT_WORKLOAD; |
58ff82f3 | 3125 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 3126 | cfqd->serving_wl_class = BE_WORKLOAD; |
718eee05 | 3127 | else { |
4d2ceea4 | 3128 | cfqd->serving_wl_class = IDLE_WORKLOAD; |
9a7f38c4 | 3129 | cfqd->workload_expires = now + jiffies_to_nsecs(1); |
718eee05 CZ |
3130 | return; |
3131 | } | |
3132 | ||
4d2ceea4 | 3133 | if (original_class != cfqd->serving_wl_class) |
e4ea0c16 SL |
3134 | goto new_workload; |
3135 | ||
718eee05 CZ |
3136 | /* |
3137 | * For RT and BE, we have to choose also the type | |
3138 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
3139 | * expiration time | |
3140 | */ | |
34b98d03 | 3141 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 3142 | count = st->count; |
718eee05 CZ |
3143 | |
3144 | /* | |
65b32a57 | 3145 | * check workload expiration, and that we still have other queues ready |
718eee05 | 3146 | */ |
9a7f38c4 | 3147 | if (count && !(now > cfqd->workload_expires)) |
718eee05 CZ |
3148 | return; |
3149 | ||
e4ea0c16 | 3150 | new_workload: |
718eee05 | 3151 | /* otherwise select new workload type */ |
6d816ec7 | 3152 | cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg, |
4d2ceea4 | 3153 | cfqd->serving_wl_class); |
34b98d03 | 3154 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 3155 | count = st->count; |
718eee05 CZ |
3156 | |
3157 | /* | |
3158 | * the workload slice is computed as a fraction of target latency | |
3159 | * proportional to the number of queues in that workload, over | |
3160 | * all the queues in the same priority class | |
3161 | */ | |
58ff82f3 VG |
3162 | group_slice = cfq_group_slice(cfqd, cfqg); |
3163 | ||
9a7f38c4 | 3164 | slice = div_u64(group_slice * count, |
4d2ceea4 VG |
3165 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class], |
3166 | cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd, | |
9a7f38c4 | 3167 | cfqg))); |
718eee05 | 3168 | |
4d2ceea4 | 3169 | if (cfqd->serving_wl_type == ASYNC_WORKLOAD) { |
9a7f38c4 | 3170 | u64 tmp; |
f26bd1f0 VG |
3171 | |
3172 | /* | |
3173 | * Async queues are currently system wide. Just taking | |
3174 | * proportion of queues with-in same group will lead to higher | |
3175 | * async ratio system wide as generally root group is going | |
3176 | * to have higher weight. A more accurate thing would be to | |
3177 | * calculate system wide asnc/sync ratio. | |
3178 | */ | |
5bf14c07 TM |
3179 | tmp = cfqd->cfq_target_latency * |
3180 | cfqg_busy_async_queues(cfqd, cfqg); | |
9a7f38c4 JM |
3181 | tmp = div_u64(tmp, cfqd->busy_queues); |
3182 | slice = min_t(u64, slice, tmp); | |
f26bd1f0 | 3183 | |
718eee05 CZ |
3184 | /* async workload slice is scaled down according to |
3185 | * the sync/async slice ratio. */ | |
9a7f38c4 | 3186 | slice = div64_u64(slice*cfqd->cfq_slice[0], cfqd->cfq_slice[1]); |
f26bd1f0 | 3187 | } else |
718eee05 CZ |
3188 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
3189 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
3190 | ||
9a7f38c4 JM |
3191 | slice = max_t(u64, slice, CFQ_MIN_TT); |
3192 | cfq_log(cfqd, "workload slice:%llu", slice); | |
3193 | cfqd->workload_expires = now + slice; | |
718eee05 CZ |
3194 | } |
3195 | ||
1fa8f6d6 VG |
3196 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
3197 | { | |
3198 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 3199 | struct cfq_group *cfqg; |
1fa8f6d6 | 3200 | |
09663c86 | 3201 | if (RB_EMPTY_ROOT(&st->rb.rb_root)) |
1fa8f6d6 | 3202 | return NULL; |
25bc6b07 | 3203 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
3204 | update_min_vdisktime(st); |
3205 | return cfqg; | |
1fa8f6d6 VG |
3206 | } |
3207 | ||
cdb16e8f VG |
3208 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
3209 | { | |
1fa8f6d6 | 3210 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
9a7f38c4 | 3211 | u64 now = ktime_get_ns(); |
1fa8f6d6 VG |
3212 | |
3213 | cfqd->serving_group = cfqg; | |
dae739eb VG |
3214 | |
3215 | /* Restore the workload type data */ | |
4d2ceea4 | 3216 | if (cfqg->saved_wl_slice) { |
9a7f38c4 | 3217 | cfqd->workload_expires = now + cfqg->saved_wl_slice; |
4d2ceea4 VG |
3218 | cfqd->serving_wl_type = cfqg->saved_wl_type; |
3219 | cfqd->serving_wl_class = cfqg->saved_wl_class; | |
66ae2919 | 3220 | } else |
9a7f38c4 | 3221 | cfqd->workload_expires = now - 1; |
66ae2919 | 3222 | |
6d816ec7 | 3223 | choose_wl_class_and_type(cfqd, cfqg); |
cdb16e8f VG |
3224 | } |
3225 | ||
22e2c507 | 3226 | /* |
498d3aa2 JA |
3227 | * Select a queue for service. If we have a current active queue, |
3228 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 3229 | */ |
1b5ed5e1 | 3230 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 3231 | { |
a36e71f9 | 3232 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
9a7f38c4 | 3233 | u64 now = ktime_get_ns(); |
1da177e4 | 3234 | |
22e2c507 JA |
3235 | cfqq = cfqd->active_queue; |
3236 | if (!cfqq) | |
3237 | goto new_queue; | |
1da177e4 | 3238 | |
f04a6424 VG |
3239 | if (!cfqd->rq_queued) |
3240 | return NULL; | |
c244bb50 VG |
3241 | |
3242 | /* | |
3243 | * We were waiting for group to get backlogged. Expire the queue | |
3244 | */ | |
3245 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3246 | goto expire; | |
3247 | ||
22e2c507 | 3248 | /* |
6d048f53 | 3249 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 3250 | */ |
7667aa06 VG |
3251 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
3252 | /* | |
3253 | * If slice had not expired at the completion of last request | |
3254 | * we might not have turned on wait_busy flag. Don't expire | |
3255 | * the queue yet. Allow the group to get backlogged. | |
3256 | * | |
3257 | * The very fact that we have used the slice, that means we | |
3258 | * have been idling all along on this queue and it should be | |
3259 | * ok to wait for this request to complete. | |
3260 | */ | |
82bbbf28 VG |
3261 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
3262 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
3263 | cfqq = NULL; | |
7667aa06 | 3264 | goto keep_queue; |
82bbbf28 | 3265 | } else |
80bdf0c7 | 3266 | goto check_group_idle; |
7667aa06 | 3267 | } |
1da177e4 | 3268 | |
22e2c507 | 3269 | /* |
6d048f53 JA |
3270 | * The active queue has requests and isn't expired, allow it to |
3271 | * dispatch. | |
22e2c507 | 3272 | */ |
dd67d051 | 3273 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3274 | goto keep_queue; |
6d048f53 | 3275 | |
a36e71f9 JA |
3276 | /* |
3277 | * If another queue has a request waiting within our mean seek | |
3278 | * distance, let it run. The expire code will check for close | |
3279 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 3280 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 3281 | */ |
b3b6d040 | 3282 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
3283 | if (new_cfqq) { |
3284 | if (!cfqq->new_cfqq) | |
3285 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 3286 | goto expire; |
df5fe3e8 | 3287 | } |
a36e71f9 | 3288 | |
6d048f53 JA |
3289 | /* |
3290 | * No requests pending. If the active queue still has requests in | |
3291 | * flight or is idling for a new request, allow either of these | |
3292 | * conditions to happen (or time out) before selecting a new queue. | |
3293 | */ | |
91148325 | 3294 | if (hrtimer_active(&cfqd->idle_slice_timer)) { |
80bdf0c7 VG |
3295 | cfqq = NULL; |
3296 | goto keep_queue; | |
3297 | } | |
3298 | ||
8e1ac665 SL |
3299 | /* |
3300 | * This is a deep seek queue, but the device is much faster than | |
3301 | * the queue can deliver, don't idle | |
3302 | **/ | |
3303 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
3304 | (cfq_cfqq_slice_new(cfqq) || | |
9a7f38c4 | 3305 | (cfqq->slice_end - now > now - cfqq->slice_start))) { |
8e1ac665 SL |
3306 | cfq_clear_cfqq_deep(cfqq); |
3307 | cfq_clear_cfqq_idle_window(cfqq); | |
3308 | } | |
3309 | ||
80bdf0c7 VG |
3310 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
3311 | cfqq = NULL; | |
3312 | goto keep_queue; | |
3313 | } | |
3314 | ||
3315 | /* | |
3316 | * If group idle is enabled and there are requests dispatched from | |
3317 | * this group, wait for requests to complete. | |
3318 | */ | |
3319 | check_group_idle: | |
7700fc4f SL |
3320 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
3321 | cfqq->cfqg->dispatched && | |
3322 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
3323 | cfqq = NULL; |
3324 | goto keep_queue; | |
22e2c507 JA |
3325 | } |
3326 | ||
3b18152c | 3327 | expire: |
e5ff082e | 3328 | cfq_slice_expired(cfqd, 0); |
3b18152c | 3329 | new_queue: |
718eee05 CZ |
3330 | /* |
3331 | * Current queue expired. Check if we have to switch to a new | |
3332 | * service tree | |
3333 | */ | |
3334 | if (!new_cfqq) | |
cdb16e8f | 3335 | cfq_choose_cfqg(cfqd); |
718eee05 | 3336 | |
a36e71f9 | 3337 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 3338 | keep_queue: |
3b18152c | 3339 | return cfqq; |
22e2c507 JA |
3340 | } |
3341 | ||
febffd61 | 3342 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
3343 | { |
3344 | int dispatched = 0; | |
3345 | ||
3346 | while (cfqq->next_rq) { | |
3347 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
3348 | dispatched++; | |
3349 | } | |
3350 | ||
3351 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
3352 | |
3353 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 3354 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
3355 | return dispatched; |
3356 | } | |
3357 | ||
498d3aa2 JA |
3358 | /* |
3359 | * Drain our current requests. Used for barriers and when switching | |
3360 | * io schedulers on-the-fly. | |
3361 | */ | |
d9e7620e | 3362 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 3363 | { |
0871714e | 3364 | struct cfq_queue *cfqq; |
d9e7620e | 3365 | int dispatched = 0; |
cdb16e8f | 3366 | |
3440c49f | 3367 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 3368 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
3369 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
3370 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 3371 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 3372 | } |
1b5ed5e1 | 3373 | |
1b5ed5e1 TH |
3374 | BUG_ON(cfqd->busy_queues); |
3375 | ||
6923715a | 3376 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
3377 | return dispatched; |
3378 | } | |
3379 | ||
abc3c744 SL |
3380 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
3381 | struct cfq_queue *cfqq) | |
3382 | { | |
9a7f38c4 JM |
3383 | u64 now = ktime_get_ns(); |
3384 | ||
abc3c744 SL |
3385 | /* the queue hasn't finished any request, can't estimate */ |
3386 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 3387 | return true; |
9a7f38c4 | 3388 | if (now + cfqd->cfq_slice_idle * cfqq->dispatched > cfqq->slice_end) |
c1e44756 | 3389 | return true; |
abc3c744 | 3390 | |
c1e44756 | 3391 | return false; |
abc3c744 SL |
3392 | } |
3393 | ||
0b182d61 | 3394 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 3395 | { |
2f5cb738 | 3396 | unsigned int max_dispatch; |
22e2c507 | 3397 | |
3932a86b GC |
3398 | if (cfq_cfqq_must_dispatch(cfqq)) |
3399 | return true; | |
3400 | ||
5ad531db JA |
3401 | /* |
3402 | * Drain async requests before we start sync IO | |
3403 | */ | |
53c583d2 | 3404 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 3405 | return false; |
5ad531db | 3406 | |
2f5cb738 JA |
3407 | /* |
3408 | * If this is an async queue and we have sync IO in flight, let it wait | |
3409 | */ | |
53c583d2 | 3410 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 3411 | return false; |
2f5cb738 | 3412 | |
abc3c744 | 3413 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
3414 | if (cfq_class_idle(cfqq)) |
3415 | max_dispatch = 1; | |
b4878f24 | 3416 | |
2f5cb738 JA |
3417 | /* |
3418 | * Does this cfqq already have too much IO in flight? | |
3419 | */ | |
3420 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 3421 | bool promote_sync = false; |
2f5cb738 JA |
3422 | /* |
3423 | * idle queue must always only have a single IO in flight | |
3424 | */ | |
3ed9a296 | 3425 | if (cfq_class_idle(cfqq)) |
0b182d61 | 3426 | return false; |
3ed9a296 | 3427 | |
ef8a41df | 3428 | /* |
c4ade94f LS |
3429 | * If there is only one sync queue |
3430 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
3431 | * queue no dispatch limit. The reason is a sync queue can |
3432 | * preempt async queue, limiting the sync queue doesn't make | |
3433 | * sense. This is useful for aiostress test. | |
3434 | */ | |
c4ade94f LS |
3435 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
3436 | promote_sync = true; | |
ef8a41df | 3437 | |
2f5cb738 JA |
3438 | /* |
3439 | * We have other queues, don't allow more IO from this one | |
3440 | */ | |
ef8a41df SL |
3441 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
3442 | !promote_sync) | |
0b182d61 | 3443 | return false; |
9ede209e | 3444 | |
365722bb | 3445 | /* |
474b18cc | 3446 | * Sole queue user, no limit |
365722bb | 3447 | */ |
ef8a41df | 3448 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
3449 | max_dispatch = -1; |
3450 | else | |
3451 | /* | |
3452 | * Normally we start throttling cfqq when cfq_quantum/2 | |
3453 | * requests have been dispatched. But we can drive | |
3454 | * deeper queue depths at the beginning of slice | |
3455 | * subjected to upper limit of cfq_quantum. | |
3456 | * */ | |
3457 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
3458 | } |
3459 | ||
3460 | /* | |
3461 | * Async queues must wait a bit before being allowed dispatch. | |
3462 | * We also ramp up the dispatch depth gradually for async IO, | |
3463 | * based on the last sync IO we serviced | |
3464 | */ | |
963b72fc | 3465 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
9a7f38c4 | 3466 | u64 last_sync = ktime_get_ns() - cfqd->last_delayed_sync; |
8e296755 | 3467 | unsigned int depth; |
365722bb | 3468 | |
9a7f38c4 | 3469 | depth = div64_u64(last_sync, cfqd->cfq_slice[1]); |
e00c54c3 JA |
3470 | if (!depth && !cfqq->dispatched) |
3471 | depth = 1; | |
8e296755 JA |
3472 | if (depth < max_dispatch) |
3473 | max_dispatch = depth; | |
2f5cb738 | 3474 | } |
3ed9a296 | 3475 | |
0b182d61 JA |
3476 | /* |
3477 | * If we're below the current max, allow a dispatch | |
3478 | */ | |
3479 | return cfqq->dispatched < max_dispatch; | |
3480 | } | |
3481 | ||
3482 | /* | |
3483 | * Dispatch a request from cfqq, moving them to the request queue | |
3484 | * dispatch list. | |
3485 | */ | |
3486 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3487 | { | |
3488 | struct request *rq; | |
3489 | ||
3490 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
3491 | ||
3932a86b GC |
3492 | rq = cfq_check_fifo(cfqq); |
3493 | if (rq) | |
3494 | cfq_mark_cfqq_must_dispatch(cfqq); | |
3495 | ||
0b182d61 JA |
3496 | if (!cfq_may_dispatch(cfqd, cfqq)) |
3497 | return false; | |
3498 | ||
3499 | /* | |
3500 | * follow expired path, else get first next available | |
3501 | */ | |
0b182d61 JA |
3502 | if (!rq) |
3503 | rq = cfqq->next_rq; | |
3932a86b GC |
3504 | else |
3505 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); | |
0b182d61 JA |
3506 | |
3507 | /* | |
3508 | * insert request into driver dispatch list | |
3509 | */ | |
3510 | cfq_dispatch_insert(cfqd->queue, rq); | |
3511 | ||
3512 | if (!cfqd->active_cic) { | |
c5869807 | 3513 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 3514 | |
c5869807 | 3515 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
3516 | cfqd->active_cic = cic; |
3517 | } | |
3518 | ||
3519 | return true; | |
3520 | } | |
3521 | ||
3522 | /* | |
3523 | * Find the cfqq that we need to service and move a request from that to the | |
3524 | * dispatch list | |
3525 | */ | |
3526 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
3527 | { | |
3528 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3529 | struct cfq_queue *cfqq; | |
3530 | ||
3531 | if (!cfqd->busy_queues) | |
3532 | return 0; | |
3533 | ||
3534 | if (unlikely(force)) | |
3535 | return cfq_forced_dispatch(cfqd); | |
3536 | ||
3537 | cfqq = cfq_select_queue(cfqd); | |
3538 | if (!cfqq) | |
8e296755 JA |
3539 | return 0; |
3540 | ||
2f5cb738 | 3541 | /* |
0b182d61 | 3542 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 3543 | */ |
0b182d61 JA |
3544 | if (!cfq_dispatch_request(cfqd, cfqq)) |
3545 | return 0; | |
3546 | ||
2f5cb738 | 3547 | cfqq->slice_dispatch++; |
b029195d | 3548 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 3549 | |
2f5cb738 JA |
3550 | /* |
3551 | * expire an async queue immediately if it has used up its slice. idle | |
3552 | * queue always expire after 1 dispatch round. | |
3553 | */ | |
3554 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
3555 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
3556 | cfq_class_idle(cfqq))) { | |
9a7f38c4 | 3557 | cfqq->slice_end = ktime_get_ns() + 1; |
e5ff082e | 3558 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
3559 | } |
3560 | ||
b217a903 | 3561 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 3562 | return 1; |
1da177e4 LT |
3563 | } |
3564 | ||
1da177e4 | 3565 | /* |
5e705374 JA |
3566 | * task holds one reference to the queue, dropped when task exits. each rq |
3567 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 3568 | * |
b1c35769 | 3569 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
3570 | * queue lock must be held here. |
3571 | */ | |
3572 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
3573 | { | |
22e2c507 | 3574 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 3575 | struct cfq_group *cfqg; |
22e2c507 | 3576 | |
30d7b944 | 3577 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 3578 | |
30d7b944 SL |
3579 | cfqq->ref--; |
3580 | if (cfqq->ref) | |
1da177e4 LT |
3581 | return; |
3582 | ||
7b679138 | 3583 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 3584 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 3585 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 3586 | cfqg = cfqq->cfqg; |
1da177e4 | 3587 | |
28f95cbc | 3588 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 3589 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 3590 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 3591 | } |
22e2c507 | 3592 | |
f04a6424 | 3593 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 3594 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 3595 | cfqg_put(cfqg); |
1da177e4 LT |
3596 | } |
3597 | ||
d02a2c07 | 3598 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 3599 | { |
df5fe3e8 JM |
3600 | struct cfq_queue *__cfqq, *next; |
3601 | ||
df5fe3e8 JM |
3602 | /* |
3603 | * If this queue was scheduled to merge with another queue, be | |
3604 | * sure to drop the reference taken on that queue (and others in | |
3605 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
3606 | */ | |
3607 | __cfqq = cfqq->new_cfqq; | |
3608 | while (__cfqq) { | |
3609 | if (__cfqq == cfqq) { | |
3610 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
3611 | break; | |
3612 | } | |
3613 | next = __cfqq->new_cfqq; | |
3614 | cfq_put_queue(__cfqq); | |
3615 | __cfqq = next; | |
3616 | } | |
d02a2c07 SL |
3617 | } |
3618 | ||
3619 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3620 | { | |
3621 | if (unlikely(cfqq == cfqd->active_queue)) { | |
3622 | __cfq_slice_expired(cfqd, cfqq, 0); | |
3623 | cfq_schedule_dispatch(cfqd); | |
3624 | } | |
3625 | ||
3626 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 3627 | |
89850f7e JA |
3628 | cfq_put_queue(cfqq); |
3629 | } | |
22e2c507 | 3630 | |
9b84cacd TH |
3631 | static void cfq_init_icq(struct io_cq *icq) |
3632 | { | |
3633 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
3634 | ||
9a7f38c4 | 3635 | cic->ttime.last_end_request = ktime_get_ns(); |
9b84cacd TH |
3636 | } |
3637 | ||
c5869807 | 3638 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 3639 | { |
c5869807 | 3640 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 3641 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 3642 | |
563180a4 TH |
3643 | if (cic_to_cfqq(cic, false)) { |
3644 | cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, false)); | |
3645 | cic_set_cfqq(cic, NULL, false); | |
12a05732 AV |
3646 | } |
3647 | ||
563180a4 TH |
3648 | if (cic_to_cfqq(cic, true)) { |
3649 | cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, true)); | |
3650 | cic_set_cfqq(cic, NULL, true); | |
12a05732 | 3651 | } |
89850f7e JA |
3652 | } |
3653 | ||
abede6da | 3654 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
3655 | { |
3656 | struct task_struct *tsk = current; | |
3657 | int ioprio_class; | |
3658 | ||
3b18152c | 3659 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
3660 | return; |
3661 | ||
598971bf | 3662 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 3663 | switch (ioprio_class) { |
fe094d98 JA |
3664 | default: |
3665 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
3666 | case IOPRIO_CLASS_NONE: | |
3667 | /* | |
6d63c275 | 3668 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
3669 | */ |
3670 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 3671 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
3672 | break; |
3673 | case IOPRIO_CLASS_RT: | |
598971bf | 3674 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3675 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
3676 | break; | |
3677 | case IOPRIO_CLASS_BE: | |
598971bf | 3678 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3679 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3680 | break; | |
3681 | case IOPRIO_CLASS_IDLE: | |
3682 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
3683 | cfqq->ioprio = 7; | |
3684 | cfq_clear_cfqq_idle_window(cfqq); | |
3685 | break; | |
22e2c507 JA |
3686 | } |
3687 | ||
3688 | /* | |
3689 | * keep track of original prio settings in case we have to temporarily | |
3690 | * elevate the priority of this queue | |
3691 | */ | |
3692 | cfqq->org_ioprio = cfqq->ioprio; | |
b8269db4 | 3693 | cfqq->org_ioprio_class = cfqq->ioprio_class; |
3b18152c | 3694 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
3695 | } |
3696 | ||
598971bf | 3697 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 3698 | { |
598971bf | 3699 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 3700 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 3701 | struct cfq_queue *cfqq; |
35e6077c | 3702 | |
598971bf TH |
3703 | /* |
3704 | * Check whether ioprio has changed. The condition may trigger | |
3705 | * spuriously on a newly created cic but there's no harm. | |
3706 | */ | |
3707 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
3708 | return; |
3709 | ||
563180a4 | 3710 | cfqq = cic_to_cfqq(cic, false); |
caaa5f9f | 3711 | if (cfqq) { |
563180a4 | 3712 | cfq_put_queue(cfqq); |
2da8de0b | 3713 | cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio); |
563180a4 | 3714 | cic_set_cfqq(cic, cfqq, false); |
22e2c507 | 3715 | } |
caaa5f9f | 3716 | |
563180a4 | 3717 | cfqq = cic_to_cfqq(cic, true); |
caaa5f9f JA |
3718 | if (cfqq) |
3719 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
3720 | |
3721 | cic->ioprio = ioprio; | |
22e2c507 JA |
3722 | } |
3723 | ||
d5036d77 | 3724 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 3725 | pid_t pid, bool is_sync) |
d5036d77 JA |
3726 | { |
3727 | RB_CLEAR_NODE(&cfqq->rb_node); | |
3728 | RB_CLEAR_NODE(&cfqq->p_node); | |
3729 | INIT_LIST_HEAD(&cfqq->fifo); | |
3730 | ||
30d7b944 | 3731 | cfqq->ref = 0; |
d5036d77 JA |
3732 | cfqq->cfqd = cfqd; |
3733 | ||
3734 | cfq_mark_cfqq_prio_changed(cfqq); | |
3735 | ||
3736 | if (is_sync) { | |
3737 | if (!cfq_class_idle(cfqq)) | |
3738 | cfq_mark_cfqq_idle_window(cfqq); | |
3739 | cfq_mark_cfqq_sync(cfqq); | |
3740 | } | |
3741 | cfqq->pid = pid; | |
3742 | } | |
3743 | ||
24610333 | 3744 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
142bbdfc | 3745 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 3746 | { |
bca4b914 | 3747 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
60a83707 | 3748 | struct cfq_queue *cfqq; |
f4da8072 | 3749 | uint64_t serial_nr; |
24610333 | 3750 | |
598971bf | 3751 | rcu_read_lock(); |
f4da8072 | 3752 | serial_nr = bio_blkcg(bio)->css.serial_nr; |
598971bf | 3753 | rcu_read_unlock(); |
24610333 | 3754 | |
598971bf TH |
3755 | /* |
3756 | * Check whether blkcg has changed. The condition may trigger | |
3757 | * spuriously on a newly created cic but there's no harm. | |
3758 | */ | |
f4da8072 | 3759 | if (unlikely(!cfqd) || likely(cic->blkcg_serial_nr == serial_nr)) |
142bbdfc | 3760 | return; |
87760e5e | 3761 | |
60a83707 TH |
3762 | /* |
3763 | * Drop reference to queues. New queues will be assigned in new | |
3764 | * group upon arrival of fresh requests. | |
3765 | */ | |
3766 | cfqq = cic_to_cfqq(cic, false); | |
3767 | if (cfqq) { | |
3768 | cfq_log_cfqq(cfqd, cfqq, "changed cgroup"); | |
3769 | cic_set_cfqq(cic, NULL, false); | |
3770 | cfq_put_queue(cfqq); | |
3771 | } | |
3772 | ||
3773 | cfqq = cic_to_cfqq(cic, true); | |
3774 | if (cfqq) { | |
3775 | cfq_log_cfqq(cfqd, cfqq, "changed cgroup"); | |
3776 | cic_set_cfqq(cic, NULL, true); | |
3777 | cfq_put_queue(cfqq); | |
24610333 | 3778 | } |
598971bf | 3779 | |
f4da8072 | 3780 | cic->blkcg_serial_nr = serial_nr; |
24610333 | 3781 | } |
598971bf | 3782 | #else |
142bbdfc | 3783 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
5d7f5ce1 | 3784 | { |
5d7f5ce1 | 3785 | } |
24610333 VG |
3786 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
3787 | ||
c2dea2d1 | 3788 | static struct cfq_queue ** |
60a83707 | 3789 | cfq_async_queue_prio(struct cfq_group *cfqg, int ioprio_class, int ioprio) |
c2dea2d1 | 3790 | { |
fe094d98 | 3791 | switch (ioprio_class) { |
c2dea2d1 | 3792 | case IOPRIO_CLASS_RT: |
60a83707 | 3793 | return &cfqg->async_cfqq[0][ioprio]; |
598971bf TH |
3794 | case IOPRIO_CLASS_NONE: |
3795 | ioprio = IOPRIO_NORM; | |
3796 | /* fall through */ | |
c2dea2d1 | 3797 | case IOPRIO_CLASS_BE: |
60a83707 | 3798 | return &cfqg->async_cfqq[1][ioprio]; |
c2dea2d1 | 3799 | case IOPRIO_CLASS_IDLE: |
60a83707 | 3800 | return &cfqg->async_idle_cfqq; |
c2dea2d1 VT |
3801 | default: |
3802 | BUG(); | |
3803 | } | |
3804 | } | |
3805 | ||
15c31be4 | 3806 | static struct cfq_queue * |
abede6da | 3807 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
2da8de0b | 3808 | struct bio *bio) |
15c31be4 | 3809 | { |
c6ce1943 JM |
3810 | int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
3811 | int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
d4aad7ff | 3812 | struct cfq_queue **async_cfqq = NULL; |
4ebc1c61 | 3813 | struct cfq_queue *cfqq; |
322731ed TH |
3814 | struct cfq_group *cfqg; |
3815 | ||
3816 | rcu_read_lock(); | |
ae118896 | 3817 | cfqg = cfq_lookup_cfqg(cfqd, bio_blkcg(bio)); |
322731ed TH |
3818 | if (!cfqg) { |
3819 | cfqq = &cfqd->oom_cfqq; | |
3820 | goto out; | |
3821 | } | |
15c31be4 | 3822 | |
c2dea2d1 | 3823 | if (!is_sync) { |
c6ce1943 JM |
3824 | if (!ioprio_valid(cic->ioprio)) { |
3825 | struct task_struct *tsk = current; | |
3826 | ioprio = task_nice_ioprio(tsk); | |
3827 | ioprio_class = task_nice_ioclass(tsk); | |
3828 | } | |
60a83707 | 3829 | async_cfqq = cfq_async_queue_prio(cfqg, ioprio_class, ioprio); |
c2dea2d1 | 3830 | cfqq = *async_cfqq; |
4ebc1c61 TH |
3831 | if (cfqq) |
3832 | goto out; | |
c2dea2d1 VT |
3833 | } |
3834 | ||
e00f4f4d TH |
3835 | cfqq = kmem_cache_alloc_node(cfq_pool, |
3836 | GFP_NOWAIT | __GFP_ZERO | __GFP_NOWARN, | |
d4aad7ff TH |
3837 | cfqd->queue->node); |
3838 | if (!cfqq) { | |
3839 | cfqq = &cfqd->oom_cfqq; | |
3840 | goto out; | |
3841 | } | |
3842 | ||
4d608baa AP |
3843 | /* cfq_init_cfqq() assumes cfqq->ioprio_class is initialized. */ |
3844 | cfqq->ioprio_class = IOPRIO_CLASS_NONE; | |
d4aad7ff TH |
3845 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); |
3846 | cfq_init_prio_data(cfqq, cic); | |
3847 | cfq_link_cfqq_cfqg(cfqq, cfqg); | |
3848 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | |
15c31be4 | 3849 | |
d4aad7ff TH |
3850 | if (async_cfqq) { |
3851 | /* a new async queue is created, pin and remember */ | |
30d7b944 | 3852 | cfqq->ref++; |
c2dea2d1 | 3853 | *async_cfqq = cfqq; |
15c31be4 | 3854 | } |
4ebc1c61 | 3855 | out: |
30d7b944 | 3856 | cfqq->ref++; |
322731ed | 3857 | rcu_read_unlock(); |
15c31be4 JA |
3858 | return cfqq; |
3859 | } | |
3860 | ||
22e2c507 | 3861 | static void |
9a7f38c4 | 3862 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, u64 slice_idle) |
1da177e4 | 3863 | { |
9a7f38c4 | 3864 | u64 elapsed = ktime_get_ns() - ttime->last_end_request; |
383cd721 | 3865 | elapsed = min(elapsed, 2UL * slice_idle); |
db3b5848 | 3866 | |
383cd721 | 3867 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
9a7f38c4 JM |
3868 | ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed, 8); |
3869 | ttime->ttime_mean = div64_ul(ttime->ttime_total + 128, | |
3870 | ttime->ttime_samples); | |
383cd721 SL |
3871 | } |
3872 | ||
3873 | static void | |
3874 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3875 | struct cfq_io_cq *cic) |
383cd721 | 3876 | { |
f5f2b6ce | 3877 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3878 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3879 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3880 | cfqd->cfq_slice_idle); | |
3881 | } | |
7700fc4f SL |
3882 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3883 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3884 | #endif | |
22e2c507 | 3885 | } |
1da177e4 | 3886 | |
206dc69b | 3887 | static void |
b2c18e1e | 3888 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3889 | struct request *rq) |
206dc69b | 3890 | { |
3dde36dd | 3891 | sector_t sdist = 0; |
41647e7a | 3892 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3893 | if (cfqq->last_request_pos) { |
3894 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3895 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3896 | else | |
3897 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3898 | } | |
206dc69b | 3899 | |
3dde36dd | 3900 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3901 | if (blk_queue_nonrot(cfqd->queue)) |
3902 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3903 | else | |
3904 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3905 | } |
1da177e4 | 3906 | |
a2b80967 CH |
3907 | static inline bool req_noidle(struct request *req) |
3908 | { | |
3909 | return req_op(req) == REQ_OP_WRITE && | |
3910 | (req->cmd_flags & (REQ_SYNC | REQ_IDLE)) == REQ_SYNC; | |
3911 | } | |
3912 | ||
22e2c507 JA |
3913 | /* |
3914 | * Disable idle window if the process thinks too long or seeks so much that | |
3915 | * it doesn't matter | |
3916 | */ | |
3917 | static void | |
3918 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3919 | struct cfq_io_cq *cic) |
22e2c507 | 3920 | { |
7b679138 | 3921 | int old_idle, enable_idle; |
1be92f2f | 3922 | |
0871714e JA |
3923 | /* |
3924 | * Don't idle for async or idle io prio class | |
3925 | */ | |
3926 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3927 | return; |
3928 | ||
c265a7f4 | 3929 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3930 | |
76280aff CZ |
3931 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3932 | cfq_mark_cfqq_deep(cfqq); | |
3933 | ||
a2b80967 | 3934 | if (cfqq->next_rq && req_noidle(cfqq->next_rq)) |
749ef9f8 | 3935 | enable_idle = 0; |
f6e8d01b | 3936 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
3937 | !cfqd->cfq_slice_idle || |
3938 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3939 | enable_idle = 0; |
383cd721 SL |
3940 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3941 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3942 | enable_idle = 0; |
3943 | else | |
3944 | enable_idle = 1; | |
1da177e4 LT |
3945 | } |
3946 | ||
7b679138 JA |
3947 | if (old_idle != enable_idle) { |
3948 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3949 | if (enable_idle) | |
3950 | cfq_mark_cfqq_idle_window(cfqq); | |
3951 | else | |
3952 | cfq_clear_cfqq_idle_window(cfqq); | |
3953 | } | |
22e2c507 | 3954 | } |
1da177e4 | 3955 | |
22e2c507 JA |
3956 | /* |
3957 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3958 | * no or if we aren't sure, a 1 will cause a preempt. | |
3959 | */ | |
a6151c3a | 3960 | static bool |
22e2c507 | 3961 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3962 | struct request *rq) |
22e2c507 | 3963 | { |
6d048f53 | 3964 | struct cfq_queue *cfqq; |
22e2c507 | 3965 | |
6d048f53 JA |
3966 | cfqq = cfqd->active_queue; |
3967 | if (!cfqq) | |
a6151c3a | 3968 | return false; |
22e2c507 | 3969 | |
6d048f53 | 3970 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3971 | return false; |
22e2c507 JA |
3972 | |
3973 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3974 | return true; |
1e3335de | 3975 | |
875feb63 DS |
3976 | /* |
3977 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3978 | */ | |
3979 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3980 | return false; | |
3981 | ||
374f84ac JA |
3982 | /* |
3983 | * if the new request is sync, but the currently running queue is | |
3984 | * not, let the sync request have priority. | |
3985 | */ | |
3932a86b | 3986 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) |
a6151c3a | 3987 | return true; |
1e3335de | 3988 | |
3984aa55 JK |
3989 | /* |
3990 | * Treat ancestors of current cgroup the same way as current cgroup. | |
3991 | * For anybody else we disallow preemption to guarantee service | |
3992 | * fairness among cgroups. | |
3993 | */ | |
3994 | if (!cfqg_is_descendant(cfqq->cfqg, new_cfqq->cfqg)) | |
8682e1f1 VG |
3995 | return false; |
3996 | ||
3997 | if (cfq_slice_used(cfqq)) | |
3998 | return true; | |
3999 | ||
6c80731c JK |
4000 | /* |
4001 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
4002 | */ | |
4003 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
4004 | return true; | |
4005 | ||
4006 | WARN_ON_ONCE(cfqq->ioprio_class != new_cfqq->ioprio_class); | |
8682e1f1 | 4007 | /* Allow preemption only if we are idling on sync-noidle tree */ |
4d2ceea4 | 4008 | if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 | 4009 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 VG |
4010 | RB_EMPTY_ROOT(&cfqq->sort_list)) |
4011 | return true; | |
4012 | ||
b53d1ed7 JA |
4013 | /* |
4014 | * So both queues are sync. Let the new request get disk time if | |
4015 | * it's a metadata request and the current queue is doing regular IO. | |
4016 | */ | |
65299a3b | 4017 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
4018 | return true; |
4019 | ||
d2d59e18 SL |
4020 | /* An idle queue should not be idle now for some reason */ |
4021 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
4022 | return true; | |
4023 | ||
1e3335de | 4024 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 4025 | return false; |
1e3335de JA |
4026 | |
4027 | /* | |
4028 | * if this request is as-good as one we would expect from the | |
4029 | * current cfqq, let it preempt | |
4030 | */ | |
e9ce335d | 4031 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 4032 | return true; |
1e3335de | 4033 | |
a6151c3a | 4034 | return false; |
22e2c507 JA |
4035 | } |
4036 | ||
4037 | /* | |
4038 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
4039 | * let it have half of its nominal slice. | |
4040 | */ | |
4041 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
4042 | { | |
df0793ab SL |
4043 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
4044 | ||
7b679138 | 4045 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 4046 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 4047 | |
f8ae6e3e SL |
4048 | /* |
4049 | * workload type is changed, don't save slice, otherwise preempt | |
4050 | * doesn't happen | |
4051 | */ | |
df0793ab | 4052 | if (old_type != cfqq_type(cfqq)) |
4d2ceea4 | 4053 | cfqq->cfqg->saved_wl_slice = 0; |
f8ae6e3e | 4054 | |
bf572256 JA |
4055 | /* |
4056 | * Put the new queue at the front of the of the current list, | |
4057 | * so we know that it will be selected next. | |
4058 | */ | |
4059 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
4060 | |
4061 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 4062 | |
62a37f6b JT |
4063 | cfqq->slice_end = 0; |
4064 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
4065 | } |
4066 | ||
22e2c507 | 4067 | /* |
5e705374 | 4068 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
4069 | * something we should do about it |
4070 | */ | |
4071 | static void | |
5e705374 JA |
4072 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
4073 | struct request *rq) | |
22e2c507 | 4074 | { |
c5869807 | 4075 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 4076 | |
45333d5a | 4077 | cfqd->rq_queued++; |
65299a3b CH |
4078 | if (rq->cmd_flags & REQ_PRIO) |
4079 | cfqq->prio_pending++; | |
374f84ac | 4080 | |
383cd721 | 4081 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 4082 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
4083 | cfq_update_idle_window(cfqd, cfqq, cic); |
4084 | ||
b2c18e1e | 4085 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
4086 | |
4087 | if (cfqq == cfqd->active_queue) { | |
4088 | /* | |
b029195d JA |
4089 | * Remember that we saw a request from this process, but |
4090 | * don't start queuing just yet. Otherwise we risk seeing lots | |
4091 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
4092 | * and merging. If the request is already larger than a single |
4093 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
4094 | * merging is already done. Ditto for a busy system that |
4095 | * has other work pending, don't risk delaying until the | |
4096 | * idle timer unplug to continue working. | |
22e2c507 | 4097 | */ |
d6ceb25e | 4098 | if (cfq_cfqq_wait_request(cfqq)) { |
09cbfeaf | 4099 | if (blk_rq_bytes(rq) > PAGE_SIZE || |
2d870722 | 4100 | cfqd->busy_queues > 1) { |
812df48d | 4101 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 4102 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 4103 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 4104 | } else { |
155fead9 | 4105 | cfqg_stats_update_idle_time(cfqq->cfqg); |
bf791937 | 4106 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 4107 | } |
d6ceb25e | 4108 | } |
5e705374 | 4109 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
4110 | /* |
4111 | * not the active queue - expire current slice if it is | |
4112 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
4113 | * has some old slice time left and is of higher priority or |
4114 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
4115 | */ |
4116 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 4117 | __blk_run_queue(cfqd->queue); |
22e2c507 | 4118 | } |
1da177e4 LT |
4119 | } |
4120 | ||
165125e1 | 4121 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 4122 | { |
b4878f24 | 4123 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 4124 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 4125 | |
7b679138 | 4126 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 4127 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 4128 | |
9a7f38c4 | 4129 | rq->fifo_time = ktime_get_ns() + cfqd->cfq_fifo_expire[rq_is_sync(rq)]; |
22e2c507 | 4130 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 4131 | cfq_add_rq_rb(rq); |
ef295ecf | 4132 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
155fead9 | 4133 | rq->cmd_flags); |
5e705374 | 4134 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
4135 | } |
4136 | ||
45333d5a AC |
4137 | /* |
4138 | * Update hw_tag based on peak queue depth over 50 samples under | |
4139 | * sufficient load. | |
4140 | */ | |
4141 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
4142 | { | |
1a1238a7 SL |
4143 | struct cfq_queue *cfqq = cfqd->active_queue; |
4144 | ||
53c583d2 CZ |
4145 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
4146 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
4147 | |
4148 | if (cfqd->hw_tag == 1) | |
4149 | return; | |
45333d5a AC |
4150 | |
4151 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 4152 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
4153 | return; |
4154 | ||
1a1238a7 SL |
4155 | /* |
4156 | * If active queue hasn't enough requests and can idle, cfq might not | |
4157 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
4158 | * case | |
4159 | */ | |
4160 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
4161 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 4162 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
4163 | return; |
4164 | ||
45333d5a AC |
4165 | if (cfqd->hw_tag_samples++ < 50) |
4166 | return; | |
4167 | ||
e459dd08 | 4168 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
4169 | cfqd->hw_tag = 1; |
4170 | else | |
4171 | cfqd->hw_tag = 0; | |
45333d5a AC |
4172 | } |
4173 | ||
7667aa06 VG |
4174 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
4175 | { | |
c5869807 | 4176 | struct cfq_io_cq *cic = cfqd->active_cic; |
9a7f38c4 | 4177 | u64 now = ktime_get_ns(); |
7667aa06 | 4178 | |
02a8f01b JT |
4179 | /* If the queue already has requests, don't wait */ |
4180 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
4181 | return false; | |
4182 | ||
7667aa06 VG |
4183 | /* If there are other queues in the group, don't wait */ |
4184 | if (cfqq->cfqg->nr_cfqq > 1) | |
4185 | return false; | |
4186 | ||
7700fc4f SL |
4187 | /* the only queue in the group, but think time is big */ |
4188 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
4189 | return false; | |
4190 | ||
7667aa06 VG |
4191 | if (cfq_slice_used(cfqq)) |
4192 | return true; | |
4193 | ||
4194 | /* if slice left is less than think time, wait busy */ | |
383cd721 | 4195 | if (cic && sample_valid(cic->ttime.ttime_samples) |
9a7f38c4 | 4196 | && (cfqq->slice_end - now < cic->ttime.ttime_mean)) |
7667aa06 VG |
4197 | return true; |
4198 | ||
4199 | /* | |
4200 | * If think times is less than a jiffy than ttime_mean=0 and above | |
4201 | * will not be true. It might happen that slice has not expired yet | |
4202 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
4203 | * case where think time is less than a jiffy, mark the queue wait | |
4204 | * busy if only 1 jiffy is left in the slice. | |
4205 | */ | |
9a7f38c4 | 4206 | if (cfqq->slice_end - now <= jiffies_to_nsecs(1)) |
7667aa06 VG |
4207 | return true; |
4208 | ||
4209 | return false; | |
4210 | } | |
4211 | ||
165125e1 | 4212 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 4213 | { |
5e705374 | 4214 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 4215 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 4216 | const int sync = rq_is_sync(rq); |
9a7f38c4 | 4217 | u64 now = ktime_get_ns(); |
1da177e4 | 4218 | |
a2b80967 | 4219 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", req_noidle(rq)); |
1da177e4 | 4220 | |
45333d5a AC |
4221 | cfq_update_hw_tag(cfqd); |
4222 | ||
53c583d2 | 4223 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 4224 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 4225 | cfqd->rq_in_driver--; |
6d048f53 | 4226 | cfqq->dispatched--; |
80bdf0c7 | 4227 | (RQ_CFQG(rq))->dispatched--; |
155fead9 | 4228 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
ef295ecf | 4229 | rq_io_start_time_ns(rq), rq->cmd_flags); |
1da177e4 | 4230 | |
53c583d2 | 4231 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 4232 | |
365722bb | 4233 | if (sync) { |
34b98d03 | 4234 | struct cfq_rb_root *st; |
f5f2b6ce | 4235 | |
383cd721 | 4236 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
4237 | |
4238 | if (cfq_cfqq_on_rr(cfqq)) | |
34b98d03 | 4239 | st = cfqq->service_tree; |
f5f2b6ce | 4240 | else |
34b98d03 VG |
4241 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), |
4242 | cfqq_type(cfqq)); | |
4243 | ||
4244 | st->ttime.last_end_request = now; | |
149321a6 JK |
4245 | /* |
4246 | * We have to do this check in jiffies since start_time is in | |
4247 | * jiffies and it is not trivial to convert to ns. If | |
4248 | * cfq_fifo_expire[1] ever comes close to 1 jiffie, this test | |
4249 | * will become problematic but so far we are fine (the default | |
4250 | * is 128 ms). | |
4251 | */ | |
4252 | if (!time_after(rq->start_time + | |
4253 | nsecs_to_jiffies(cfqd->cfq_fifo_expire[1]), | |
4254 | jiffies)) | |
573412b2 | 4255 | cfqd->last_delayed_sync = now; |
365722bb | 4256 | } |
caaa5f9f | 4257 | |
7700fc4f SL |
4258 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4259 | cfqq->cfqg->ttime.last_end_request = now; | |
4260 | #endif | |
4261 | ||
caaa5f9f JA |
4262 | /* |
4263 | * If this is the active queue, check if it needs to be expired, | |
4264 | * or if we want to idle in case it has no pending requests. | |
4265 | */ | |
4266 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
4267 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
4268 | ||
44f7c160 JA |
4269 | if (cfq_cfqq_slice_new(cfqq)) { |
4270 | cfq_set_prio_slice(cfqd, cfqq); | |
4271 | cfq_clear_cfqq_slice_new(cfqq); | |
4272 | } | |
f75edf2d VG |
4273 | |
4274 | /* | |
7667aa06 VG |
4275 | * Should we wait for next request to come in before we expire |
4276 | * the queue. | |
f75edf2d | 4277 | */ |
7667aa06 | 4278 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
9a7f38c4 | 4279 | u64 extend_sl = cfqd->cfq_slice_idle; |
80bdf0c7 VG |
4280 | if (!cfqd->cfq_slice_idle) |
4281 | extend_sl = cfqd->cfq_group_idle; | |
9a7f38c4 | 4282 | cfqq->slice_end = now + extend_sl; |
f75edf2d | 4283 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 4284 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
4285 | } |
4286 | ||
a36e71f9 | 4287 | /* |
8e550632 CZ |
4288 | * Idling is not enabled on: |
4289 | * - expired queues | |
4290 | * - idle-priority queues | |
4291 | * - async queues | |
4292 | * - queues with still some requests queued | |
4293 | * - when there is a close cooperator | |
a36e71f9 | 4294 | */ |
0871714e | 4295 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 4296 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
4297 | else if (sync && cfqq_empty && |
4298 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 4299 | cfq_arm_slice_timer(cfqd); |
8e550632 | 4300 | } |
caaa5f9f | 4301 | } |
6d048f53 | 4302 | |
53c583d2 | 4303 | if (!cfqd->rq_in_driver) |
23e018a1 | 4304 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
4305 | } |
4306 | ||
ef295ecf | 4307 | static void cfqq_boost_on_prio(struct cfq_queue *cfqq, unsigned int op) |
b8269db4 JA |
4308 | { |
4309 | /* | |
4310 | * If REQ_PRIO is set, boost class and prio level, if it's below | |
4311 | * BE/NORM. If prio is not set, restore the potentially boosted | |
4312 | * class/prio level. | |
4313 | */ | |
ef295ecf | 4314 | if (!(op & REQ_PRIO)) { |
b8269db4 JA |
4315 | cfqq->ioprio_class = cfqq->org_ioprio_class; |
4316 | cfqq->ioprio = cfqq->org_ioprio; | |
4317 | } else { | |
4318 | if (cfq_class_idle(cfqq)) | |
4319 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
4320 | if (cfqq->ioprio > IOPRIO_NORM) | |
4321 | cfqq->ioprio = IOPRIO_NORM; | |
4322 | } | |
4323 | } | |
4324 | ||
89850f7e | 4325 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 4326 | { |
1b379d8d | 4327 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 4328 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 4329 | return ELV_MQUEUE_MUST; |
3b18152c | 4330 | } |
1da177e4 | 4331 | |
22e2c507 | 4332 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
4333 | } |
4334 | ||
ef295ecf | 4335 | static int cfq_may_queue(struct request_queue *q, unsigned int op) |
22e2c507 JA |
4336 | { |
4337 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
4338 | struct task_struct *tsk = current; | |
c5869807 | 4339 | struct cfq_io_cq *cic; |
22e2c507 JA |
4340 | struct cfq_queue *cfqq; |
4341 | ||
4342 | /* | |
4343 | * don't force setup of a queue from here, as a call to may_queue | |
4344 | * does not necessarily imply that a request actually will be queued. | |
4345 | * so just lookup a possibly existing queue, or return 'may queue' | |
4346 | * if that fails | |
4347 | */ | |
4ac845a2 | 4348 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
4349 | if (!cic) |
4350 | return ELV_MQUEUE_MAY; | |
4351 | ||
ef295ecf | 4352 | cfqq = cic_to_cfqq(cic, op_is_sync(op)); |
22e2c507 | 4353 | if (cfqq) { |
abede6da | 4354 | cfq_init_prio_data(cfqq, cic); |
ef295ecf | 4355 | cfqq_boost_on_prio(cfqq, op); |
22e2c507 | 4356 | |
89850f7e | 4357 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
4358 | } |
4359 | ||
4360 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
4361 | } |
4362 | ||
1da177e4 LT |
4363 | /* |
4364 | * queue lock held here | |
4365 | */ | |
bb37b94c | 4366 | static void cfq_put_request(struct request *rq) |
1da177e4 | 4367 | { |
5e705374 | 4368 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 4369 | |
5e705374 | 4370 | if (cfqq) { |
22e2c507 | 4371 | const int rw = rq_data_dir(rq); |
1da177e4 | 4372 | |
22e2c507 JA |
4373 | BUG_ON(!cfqq->allocated[rw]); |
4374 | cfqq->allocated[rw]--; | |
1da177e4 | 4375 | |
7f1dc8a2 | 4376 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 4377 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
4378 | rq->elv.priv[0] = NULL; |
4379 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 4380 | |
1da177e4 LT |
4381 | cfq_put_queue(cfqq); |
4382 | } | |
4383 | } | |
4384 | ||
df5fe3e8 | 4385 | static struct cfq_queue * |
c5869807 | 4386 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
4387 | struct cfq_queue *cfqq) |
4388 | { | |
4389 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
4390 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 4391 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
4392 | cfq_put_queue(cfqq); |
4393 | return cic_to_cfqq(cic, 1); | |
4394 | } | |
4395 | ||
e6c5bc73 JM |
4396 | /* |
4397 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
4398 | * was the last process referring to said cfqq. | |
4399 | */ | |
4400 | static struct cfq_queue * | |
c5869807 | 4401 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
4402 | { |
4403 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
4404 | cfqq->pid = current->pid; |
4405 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 4406 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
4407 | return cfqq; |
4408 | } | |
4409 | ||
4410 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
4411 | |
4412 | cfq_put_cooperator(cfqq); | |
4413 | ||
e6c5bc73 JM |
4414 | cfq_put_queue(cfqq); |
4415 | return NULL; | |
4416 | } | |
1da177e4 | 4417 | /* |
22e2c507 | 4418 | * Allocate cfq data structures associated with this request. |
1da177e4 | 4419 | */ |
22e2c507 | 4420 | static int |
852c788f TH |
4421 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
4422 | gfp_t gfp_mask) | |
1da177e4 LT |
4423 | { |
4424 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 4425 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 4426 | const int rw = rq_data_dir(rq); |
a6151c3a | 4427 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 4428 | struct cfq_queue *cfqq; |
1da177e4 | 4429 | |
216284c3 | 4430 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 4431 | |
598971bf | 4432 | check_ioprio_changed(cic, bio); |
142bbdfc | 4433 | check_blkcg_changed(cic, bio); |
e6c5bc73 | 4434 | new_queue: |
91fac317 | 4435 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 4436 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
bce6133b TH |
4437 | if (cfqq) |
4438 | cfq_put_queue(cfqq); | |
2da8de0b | 4439 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio); |
91fac317 | 4440 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 4441 | } else { |
e6c5bc73 JM |
4442 | /* |
4443 | * If the queue was seeky for too long, break it apart. | |
4444 | */ | |
ae54abed | 4445 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
4446 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
4447 | cfqq = split_cfqq(cic, cfqq); | |
4448 | if (!cfqq) | |
4449 | goto new_queue; | |
4450 | } | |
4451 | ||
df5fe3e8 JM |
4452 | /* |
4453 | * Check to see if this queue is scheduled to merge with | |
4454 | * another, closely cooperating queue. The merging of | |
4455 | * queues happens here as it must be done in process context. | |
4456 | * The reference on new_cfqq was taken in merge_cfqqs. | |
4457 | */ | |
4458 | if (cfqq->new_cfqq) | |
4459 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 4460 | } |
1da177e4 LT |
4461 | |
4462 | cfqq->allocated[rw]++; | |
1da177e4 | 4463 | |
6fae9c25 | 4464 | cfqq->ref++; |
eb7d8c07 | 4465 | cfqg_get(cfqq->cfqg); |
a612fddf | 4466 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 4467 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 4468 | spin_unlock_irq(q->queue_lock); |
5d7f5ce1 | 4469 | |
5e705374 | 4470 | return 0; |
1da177e4 LT |
4471 | } |
4472 | ||
65f27f38 | 4473 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 4474 | { |
65f27f38 | 4475 | struct cfq_data *cfqd = |
23e018a1 | 4476 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 4477 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4478 | |
40bb54d1 | 4479 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 4480 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 4481 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
4482 | } |
4483 | ||
4484 | /* | |
4485 | * Timer running if the active_queue is currently idling inside its time slice | |
4486 | */ | |
91148325 | 4487 | static enum hrtimer_restart cfq_idle_slice_timer(struct hrtimer *timer) |
22e2c507 | 4488 | { |
91148325 JK |
4489 | struct cfq_data *cfqd = container_of(timer, struct cfq_data, |
4490 | idle_slice_timer); | |
22e2c507 JA |
4491 | struct cfq_queue *cfqq; |
4492 | unsigned long flags; | |
3c6bd2f8 | 4493 | int timed_out = 1; |
22e2c507 | 4494 | |
7b679138 JA |
4495 | cfq_log(cfqd, "idle timer fired"); |
4496 | ||
22e2c507 JA |
4497 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
4498 | ||
fe094d98 JA |
4499 | cfqq = cfqd->active_queue; |
4500 | if (cfqq) { | |
3c6bd2f8 JA |
4501 | timed_out = 0; |
4502 | ||
b029195d JA |
4503 | /* |
4504 | * We saw a request before the queue expired, let it through | |
4505 | */ | |
4506 | if (cfq_cfqq_must_dispatch(cfqq)) | |
4507 | goto out_kick; | |
4508 | ||
22e2c507 JA |
4509 | /* |
4510 | * expired | |
4511 | */ | |
44f7c160 | 4512 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
4513 | goto expire; |
4514 | ||
4515 | /* | |
4516 | * only expire and reinvoke request handler, if there are | |
4517 | * other queues with pending requests | |
4518 | */ | |
caaa5f9f | 4519 | if (!cfqd->busy_queues) |
22e2c507 | 4520 | goto out_cont; |
22e2c507 JA |
4521 | |
4522 | /* | |
4523 | * not expired and it has a request pending, let it dispatch | |
4524 | */ | |
75e50984 | 4525 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 4526 | goto out_kick; |
76280aff CZ |
4527 | |
4528 | /* | |
4529 | * Queue depth flag is reset only when the idle didn't succeed | |
4530 | */ | |
4531 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
4532 | } |
4533 | expire: | |
e5ff082e | 4534 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 4535 | out_kick: |
23e018a1 | 4536 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
4537 | out_cont: |
4538 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
91148325 | 4539 | return HRTIMER_NORESTART; |
22e2c507 JA |
4540 | } |
4541 | ||
3b18152c JA |
4542 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
4543 | { | |
91148325 | 4544 | hrtimer_cancel(&cfqd->idle_slice_timer); |
23e018a1 | 4545 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 4546 | } |
22e2c507 | 4547 | |
b374d18a | 4548 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 4549 | { |
22e2c507 | 4550 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 4551 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4552 | |
3b18152c | 4553 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 4554 | |
d9ff4187 | 4555 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 4556 | |
d9ff4187 | 4557 | if (cfqd->active_queue) |
e5ff082e | 4558 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 4559 | |
03aa264a TH |
4560 | spin_unlock_irq(q->queue_lock); |
4561 | ||
a90d742e AV |
4562 | cfq_shutdown_timer_wq(cfqd); |
4563 | ||
ffea73fc TH |
4564 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4565 | blkcg_deactivate_policy(q, &blkcg_policy_cfq); | |
4566 | #else | |
f51b802c | 4567 | kfree(cfqd->root_group); |
2abae55f | 4568 | #endif |
56edf7d7 | 4569 | kfree(cfqd); |
1da177e4 LT |
4570 | } |
4571 | ||
d50235b7 | 4572 | static int cfq_init_queue(struct request_queue *q, struct elevator_type *e) |
1da177e4 LT |
4573 | { |
4574 | struct cfq_data *cfqd; | |
3c798398 | 4575 | struct blkcg_gq *blkg __maybe_unused; |
a2b1693b | 4576 | int i, ret; |
d50235b7 JM |
4577 | struct elevator_queue *eq; |
4578 | ||
4579 | eq = elevator_alloc(q, e); | |
4580 | if (!eq) | |
4581 | return -ENOMEM; | |
1da177e4 | 4582 | |
c1b511eb | 4583 | cfqd = kzalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node); |
d50235b7 JM |
4584 | if (!cfqd) { |
4585 | kobject_put(&eq->kobj); | |
b2fab5ac | 4586 | return -ENOMEM; |
d50235b7 JM |
4587 | } |
4588 | eq->elevator_data = cfqd; | |
80b15c73 | 4589 | |
f51b802c | 4590 | cfqd->queue = q; |
d50235b7 JM |
4591 | spin_lock_irq(q->queue_lock); |
4592 | q->elevator = eq; | |
4593 | spin_unlock_irq(q->queue_lock); | |
f51b802c | 4594 | |
1fa8f6d6 VG |
4595 | /* Init root service tree */ |
4596 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
4597 | ||
f51b802c | 4598 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 4599 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4600 | ret = blkcg_activate_policy(q, &blkcg_policy_cfq); |
a2b1693b TH |
4601 | if (ret) |
4602 | goto out_free; | |
f51b802c | 4603 | |
a2b1693b | 4604 | cfqd->root_group = blkg_to_cfqg(q->root_blkg); |
f51b802c | 4605 | #else |
a2b1693b | 4606 | ret = -ENOMEM; |
f51b802c TH |
4607 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
4608 | GFP_KERNEL, cfqd->queue->node); | |
a2b1693b TH |
4609 | if (!cfqd->root_group) |
4610 | goto out_free; | |
5624a4e4 | 4611 | |
a2b1693b | 4612 | cfq_init_cfqg_base(cfqd->root_group); |
3ecca629 TH |
4613 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_LEGACY_DFL; |
4614 | cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_LEGACY_DFL; | |
69d7fde5 | 4615 | #endif |
5624a4e4 | 4616 | |
26a2ac00 JA |
4617 | /* |
4618 | * Not strictly needed (since RB_ROOT just clears the node and we | |
4619 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
4620 | * to add magic to the rb code | |
4621 | */ | |
4622 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
4623 | cfqd->prio_trees[i] = RB_ROOT; | |
4624 | ||
6118b70b | 4625 | /* |
d4aad7ff | 4626 | * Our fallback cfqq if cfq_get_queue() runs into OOM issues. |
6118b70b | 4627 | * Grab a permanent reference to it, so that the normal code flow |
f51b802c TH |
4628 | * will not attempt to free it. oom_cfqq is linked to root_group |
4629 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
4630 | * the reference from linking right away. | |
6118b70b JA |
4631 | */ |
4632 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 4633 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
4634 | |
4635 | spin_lock_irq(q->queue_lock); | |
f51b802c | 4636 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 4637 | cfqg_put(cfqd->root_group); |
1adaf3dd | 4638 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 4639 | |
91148325 JK |
4640 | hrtimer_init(&cfqd->idle_slice_timer, CLOCK_MONOTONIC, |
4641 | HRTIMER_MODE_REL); | |
22e2c507 | 4642 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; |
22e2c507 | 4643 | |
23e018a1 | 4644 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 4645 | |
1da177e4 | 4646 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
4647 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
4648 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
4649 | cfqd->cfq_back_max = cfq_back_max; |
4650 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
4651 | cfqd->cfq_slice[0] = cfq_slice_async; |
4652 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
5bf14c07 | 4653 | cfqd->cfq_target_latency = cfq_target_latency; |
22e2c507 | 4654 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
0bb97947 | 4655 | cfqd->cfq_slice_idle = cfq_slice_idle; |
80bdf0c7 | 4656 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 4657 | cfqd->cfq_latency = 1; |
e459dd08 | 4658 | cfqd->hw_tag = -1; |
edc71131 CZ |
4659 | /* |
4660 | * we optimistically start assuming sync ops weren't delayed in last | |
4661 | * second, in order to have larger depth for async operations. | |
4662 | */ | |
9a7f38c4 | 4663 | cfqd->last_delayed_sync = ktime_get_ns() - NSEC_PER_SEC; |
b2fab5ac | 4664 | return 0; |
a2b1693b TH |
4665 | |
4666 | out_free: | |
4667 | kfree(cfqd); | |
d50235b7 | 4668 | kobject_put(&eq->kobj); |
a2b1693b | 4669 | return ret; |
1da177e4 LT |
4670 | } |
4671 | ||
0bb97947 JA |
4672 | static void cfq_registered_queue(struct request_queue *q) |
4673 | { | |
4674 | struct elevator_queue *e = q->elevator; | |
4675 | struct cfq_data *cfqd = e->elevator_data; | |
4676 | ||
4677 | /* | |
4678 | * Default to IOPS mode with no idling for SSDs | |
4679 | */ | |
4680 | if (blk_queue_nonrot(q)) | |
4681 | cfqd->cfq_slice_idle = 0; | |
142bbdfc | 4682 | wbt_disable_default(q); |
0bb97947 JA |
4683 | } |
4684 | ||
1da177e4 LT |
4685 | /* |
4686 | * sysfs parts below --> | |
4687 | */ | |
1da177e4 LT |
4688 | static ssize_t |
4689 | cfq_var_show(unsigned int var, char *page) | |
4690 | { | |
176167ad | 4691 | return sprintf(page, "%u\n", var); |
1da177e4 LT |
4692 | } |
4693 | ||
235f8da1 | 4694 | static void |
4695 | cfq_var_store(unsigned int *var, const char *page) | |
1da177e4 LT |
4696 | { |
4697 | char *p = (char *) page; | |
4698 | ||
4699 | *var = simple_strtoul(p, &p, 10); | |
1da177e4 LT |
4700 | } |
4701 | ||
1da177e4 | 4702 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 4703 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 4704 | { \ |
3d1ab40f | 4705 | struct cfq_data *cfqd = e->elevator_data; \ |
9a7f38c4 | 4706 | u64 __data = __VAR; \ |
1da177e4 | 4707 | if (__CONV) \ |
9a7f38c4 | 4708 | __data = div_u64(__data, NSEC_PER_MSEC); \ |
1da177e4 LT |
4709 | return cfq_var_show(__data, (page)); \ |
4710 | } | |
4711 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
4712 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
4713 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
4714 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
4715 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 4716 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 4717 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
4718 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4719 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4720 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4721 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
5bf14c07 | 4722 | SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1); |
1da177e4 LT |
4723 | #undef SHOW_FUNCTION |
4724 | ||
d2d481d0 JM |
4725 | #define USEC_SHOW_FUNCTION(__FUNC, __VAR) \ |
4726 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ | |
4727 | { \ | |
4728 | struct cfq_data *cfqd = e->elevator_data; \ | |
4729 | u64 __data = __VAR; \ | |
4730 | __data = div_u64(__data, NSEC_PER_USEC); \ | |
4731 | return cfq_var_show(__data, (page)); \ | |
4732 | } | |
4733 | USEC_SHOW_FUNCTION(cfq_slice_idle_us_show, cfqd->cfq_slice_idle); | |
4734 | USEC_SHOW_FUNCTION(cfq_group_idle_us_show, cfqd->cfq_group_idle); | |
4735 | USEC_SHOW_FUNCTION(cfq_slice_sync_us_show, cfqd->cfq_slice[1]); | |
4736 | USEC_SHOW_FUNCTION(cfq_slice_async_us_show, cfqd->cfq_slice[0]); | |
4737 | USEC_SHOW_FUNCTION(cfq_target_latency_us_show, cfqd->cfq_target_latency); | |
4738 | #undef USEC_SHOW_FUNCTION | |
4739 | ||
1da177e4 | 4740 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ |
b374d18a | 4741 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4742 | { \ |
3d1ab40f | 4743 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 | 4744 | unsigned int __data; \ |
235f8da1 | 4745 | cfq_var_store(&__data, (page)); \ |
1da177e4 LT |
4746 | if (__data < (MIN)) \ |
4747 | __data = (MIN); \ | |
4748 | else if (__data > (MAX)) \ | |
4749 | __data = (MAX); \ | |
4750 | if (__CONV) \ | |
9a7f38c4 | 4751 | *(__PTR) = (u64)__data * NSEC_PER_MSEC; \ |
1da177e4 LT |
4752 | else \ |
4753 | *(__PTR) = __data; \ | |
235f8da1 | 4754 | return count; \ |
1da177e4 LT |
4755 | } |
4756 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4757 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4758 | UINT_MAX, 1); | |
4759 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4760 | UINT_MAX, 1); | |
e572ec7e | 4761 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4762 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4763 | UINT_MAX, 0); | |
22e2c507 | 4764 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4765 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4766 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4767 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4768 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4769 | UINT_MAX, 0); | |
963b72fc | 4770 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
5bf14c07 | 4771 | STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1); |
1da177e4 LT |
4772 | #undef STORE_FUNCTION |
4773 | ||
d2d481d0 JM |
4774 | #define USEC_STORE_FUNCTION(__FUNC, __PTR, MIN, MAX) \ |
4775 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ | |
4776 | { \ | |
4777 | struct cfq_data *cfqd = e->elevator_data; \ | |
4778 | unsigned int __data; \ | |
235f8da1 | 4779 | cfq_var_store(&__data, (page)); \ |
d2d481d0 JM |
4780 | if (__data < (MIN)) \ |
4781 | __data = (MIN); \ | |
4782 | else if (__data > (MAX)) \ | |
4783 | __data = (MAX); \ | |
4784 | *(__PTR) = (u64)__data * NSEC_PER_USEC; \ | |
235f8da1 | 4785 | return count; \ |
d2d481d0 JM |
4786 | } |
4787 | USEC_STORE_FUNCTION(cfq_slice_idle_us_store, &cfqd->cfq_slice_idle, 0, UINT_MAX); | |
4788 | USEC_STORE_FUNCTION(cfq_group_idle_us_store, &cfqd->cfq_group_idle, 0, UINT_MAX); | |
4789 | USEC_STORE_FUNCTION(cfq_slice_sync_us_store, &cfqd->cfq_slice[1], 1, UINT_MAX); | |
4790 | USEC_STORE_FUNCTION(cfq_slice_async_us_store, &cfqd->cfq_slice[0], 1, UINT_MAX); | |
4791 | USEC_STORE_FUNCTION(cfq_target_latency_us_store, &cfqd->cfq_target_latency, 1, UINT_MAX); | |
4792 | #undef USEC_STORE_FUNCTION | |
4793 | ||
e572ec7e AV |
4794 | #define CFQ_ATTR(name) \ |
4795 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4796 | ||
4797 | static struct elv_fs_entry cfq_attrs[] = { | |
4798 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4799 | CFQ_ATTR(fifo_expire_sync), |
4800 | CFQ_ATTR(fifo_expire_async), | |
4801 | CFQ_ATTR(back_seek_max), | |
4802 | CFQ_ATTR(back_seek_penalty), | |
4803 | CFQ_ATTR(slice_sync), | |
d2d481d0 | 4804 | CFQ_ATTR(slice_sync_us), |
e572ec7e | 4805 | CFQ_ATTR(slice_async), |
d2d481d0 | 4806 | CFQ_ATTR(slice_async_us), |
e572ec7e AV |
4807 | CFQ_ATTR(slice_async_rq), |
4808 | CFQ_ATTR(slice_idle), | |
d2d481d0 | 4809 | CFQ_ATTR(slice_idle_us), |
80bdf0c7 | 4810 | CFQ_ATTR(group_idle), |
d2d481d0 | 4811 | CFQ_ATTR(group_idle_us), |
963b72fc | 4812 | CFQ_ATTR(low_latency), |
5bf14c07 | 4813 | CFQ_ATTR(target_latency), |
d2d481d0 | 4814 | CFQ_ATTR(target_latency_us), |
e572ec7e | 4815 | __ATTR_NULL |
1da177e4 LT |
4816 | }; |
4817 | ||
1da177e4 | 4818 | static struct elevator_type iosched_cfq = { |
c51ca6cf | 4819 | .ops.sq = { |
1da177e4 LT |
4820 | .elevator_merge_fn = cfq_merge, |
4821 | .elevator_merged_fn = cfq_merged_request, | |
4822 | .elevator_merge_req_fn = cfq_merged_requests, | |
72ef799b TE |
4823 | .elevator_allow_bio_merge_fn = cfq_allow_bio_merge, |
4824 | .elevator_allow_rq_merge_fn = cfq_allow_rq_merge, | |
812d4026 | 4825 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4826 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4827 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4828 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4829 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4830 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4831 | .elevator_former_req_fn = elv_rb_former_request, |
4832 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 4833 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 4834 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
4835 | .elevator_set_req_fn = cfq_set_request, |
4836 | .elevator_put_req_fn = cfq_put_request, | |
4837 | .elevator_may_queue_fn = cfq_may_queue, | |
4838 | .elevator_init_fn = cfq_init_queue, | |
4839 | .elevator_exit_fn = cfq_exit_queue, | |
0bb97947 | 4840 | .elevator_registered_fn = cfq_registered_queue, |
1da177e4 | 4841 | }, |
3d3c2379 TH |
4842 | .icq_size = sizeof(struct cfq_io_cq), |
4843 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 4844 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 4845 | .elevator_name = "cfq", |
1da177e4 LT |
4846 | .elevator_owner = THIS_MODULE, |
4847 | }; | |
4848 | ||
3e252066 | 4849 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4850 | static struct blkcg_policy blkcg_policy_cfq = { |
2ee867dc | 4851 | .dfl_cftypes = cfq_blkcg_files, |
880f50e2 | 4852 | .legacy_cftypes = cfq_blkcg_legacy_files, |
f9fcc2d3 | 4853 | |
e4a9bde9 | 4854 | .cpd_alloc_fn = cfq_cpd_alloc, |
e48453c3 | 4855 | .cpd_init_fn = cfq_cpd_init, |
e4a9bde9 | 4856 | .cpd_free_fn = cfq_cpd_free, |
69d7fde5 | 4857 | .cpd_bind_fn = cfq_cpd_bind, |
e4a9bde9 | 4858 | |
001bea73 | 4859 | .pd_alloc_fn = cfq_pd_alloc, |
f9fcc2d3 | 4860 | .pd_init_fn = cfq_pd_init, |
0b39920b | 4861 | .pd_offline_fn = cfq_pd_offline, |
001bea73 | 4862 | .pd_free_fn = cfq_pd_free, |
f9fcc2d3 | 4863 | .pd_reset_stats_fn = cfq_pd_reset_stats, |
3e252066 | 4864 | }; |
3e252066 VG |
4865 | #endif |
4866 | ||
1da177e4 LT |
4867 | static int __init cfq_init(void) |
4868 | { | |
3d3c2379 TH |
4869 | int ret; |
4870 | ||
80bdf0c7 | 4871 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4872 | ret = blkcg_policy_register(&blkcg_policy_cfq); |
8bd435b3 TH |
4873 | if (ret) |
4874 | return ret; | |
ffea73fc TH |
4875 | #else |
4876 | cfq_group_idle = 0; | |
4877 | #endif | |
8bd435b3 | 4878 | |
fd794956 | 4879 | ret = -ENOMEM; |
3d3c2379 TH |
4880 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4881 | if (!cfq_pool) | |
8bd435b3 | 4882 | goto err_pol_unreg; |
1da177e4 | 4883 | |
3d3c2379 | 4884 | ret = elv_register(&iosched_cfq); |
8bd435b3 TH |
4885 | if (ret) |
4886 | goto err_free_pool; | |
3d3c2379 | 4887 | |
2fdd82bd | 4888 | return 0; |
8bd435b3 TH |
4889 | |
4890 | err_free_pool: | |
4891 | kmem_cache_destroy(cfq_pool); | |
4892 | err_pol_unreg: | |
ffea73fc | 4893 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4894 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4895 | #endif |
8bd435b3 | 4896 | return ret; |
1da177e4 LT |
4897 | } |
4898 | ||
4899 | static void __exit cfq_exit(void) | |
4900 | { | |
ffea73fc | 4901 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4902 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4903 | #endif |
1da177e4 | 4904 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4905 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4906 | } |
4907 | ||
4908 | module_init(cfq_init); | |
4909 | module_exit(cfq_exit); | |
4910 | ||
4911 | MODULE_AUTHOR("Jens Axboe"); | |
4912 | MODULE_LICENSE("GPL"); | |
4913 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |