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 | /* |
0871714e | 41 | * offset from end of service tree |
d9e7620e | 42 | */ |
9a7f38c4 | 43 | #define CFQ_IDLE_DELAY (NSEC_PER_SEC / 5) |
d9e7620e JA |
44 | |
45 | /* | |
46 | * below this threshold, we consider thinktime immediate | |
47 | */ | |
9a7f38c4 | 48 | #define CFQ_MIN_TT (2 * NSEC_PER_SEC / HZ) |
d9e7620e | 49 | |
22e2c507 | 50 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 51 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 52 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 53 | |
3dde36dd | 54 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 55 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 56 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 57 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 58 | |
a612fddf TH |
59 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
60 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
61 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 62 | |
e18b890b | 63 | static struct kmem_cache *cfq_pool; |
1da177e4 | 64 | |
22e2c507 JA |
65 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
66 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
67 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
68 | ||
206dc69b | 69 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 70 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 71 | |
e48453c3 | 72 | /* blkio-related constants */ |
3ecca629 TH |
73 | #define CFQ_WEIGHT_LEGACY_MIN 10 |
74 | #define CFQ_WEIGHT_LEGACY_DFL 500 | |
75 | #define CFQ_WEIGHT_LEGACY_MAX 1000 | |
e48453c3 | 76 | |
c5869807 | 77 | struct cfq_ttime { |
9a7f38c4 | 78 | u64 last_end_request; |
c5869807 | 79 | |
9a7f38c4 JM |
80 | u64 ttime_total; |
81 | u64 ttime_mean; | |
c5869807 | 82 | unsigned long ttime_samples; |
c5869807 TH |
83 | }; |
84 | ||
cc09e299 JA |
85 | /* |
86 | * Most of our rbtree usage is for sorting with min extraction, so | |
87 | * if we cache the leftmost node we don't have to walk down the tree | |
88 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
89 | * move this into the elevator for the rq sorting as well. | |
90 | */ | |
91 | struct cfq_rb_root { | |
92 | struct rb_root rb; | |
93 | struct rb_node *left; | |
aa6f6a3d | 94 | unsigned count; |
1fa8f6d6 | 95 | u64 min_vdisktime; |
f5f2b6ce | 96 | struct cfq_ttime ttime; |
cc09e299 | 97 | }; |
f5f2b6ce | 98 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
9a7f38c4 | 99 | .ttime = {.last_end_request = ktime_get_ns(),},} |
cc09e299 | 100 | |
6118b70b JA |
101 | /* |
102 | * Per process-grouping structure | |
103 | */ | |
104 | struct cfq_queue { | |
105 | /* reference count */ | |
30d7b944 | 106 | int ref; |
6118b70b JA |
107 | /* various state flags, see below */ |
108 | unsigned int flags; | |
109 | /* parent cfq_data */ | |
110 | struct cfq_data *cfqd; | |
111 | /* service_tree member */ | |
112 | struct rb_node rb_node; | |
113 | /* service_tree key */ | |
9a7f38c4 | 114 | u64 rb_key; |
6118b70b JA |
115 | /* prio tree member */ |
116 | struct rb_node p_node; | |
117 | /* prio tree root we belong to, if any */ | |
118 | struct rb_root *p_root; | |
119 | /* sorted list of pending requests */ | |
120 | struct rb_root sort_list; | |
121 | /* if fifo isn't expired, next request to serve */ | |
122 | struct request *next_rq; | |
123 | /* requests queued in sort_list */ | |
124 | int queued[2]; | |
125 | /* currently allocated requests */ | |
126 | int allocated[2]; | |
127 | /* fifo list of requests in sort_list */ | |
128 | struct list_head fifo; | |
129 | ||
dae739eb | 130 | /* time when queue got scheduled in to dispatch first request. */ |
9a7f38c4 JM |
131 | u64 dispatch_start; |
132 | u64 allocated_slice; | |
133 | u64 slice_dispatch; | |
dae739eb | 134 | /* time when first request from queue completed and slice started. */ |
9a7f38c4 JM |
135 | u64 slice_start; |
136 | u64 slice_end; | |
93fdf147 | 137 | s64 slice_resid; |
6118b70b | 138 | |
65299a3b CH |
139 | /* pending priority requests */ |
140 | int prio_pending; | |
6118b70b JA |
141 | /* number of requests that are on the dispatch list or inside driver */ |
142 | int dispatched; | |
143 | ||
144 | /* io prio of this group */ | |
145 | unsigned short ioprio, org_ioprio; | |
b8269db4 | 146 | unsigned short ioprio_class, org_ioprio_class; |
6118b70b | 147 | |
c4081ba5 RK |
148 | pid_t pid; |
149 | ||
3dde36dd | 150 | u32 seek_history; |
b2c18e1e JM |
151 | sector_t last_request_pos; |
152 | ||
aa6f6a3d | 153 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 154 | struct cfq_queue *new_cfqq; |
cdb16e8f | 155 | struct cfq_group *cfqg; |
c4e7893e VG |
156 | /* Number of sectors dispatched from queue in single dispatch round */ |
157 | unsigned long nr_sectors; | |
6118b70b JA |
158 | }; |
159 | ||
c0324a02 | 160 | /* |
718eee05 | 161 | * First index in the service_trees. |
c0324a02 CZ |
162 | * IDLE is handled separately, so it has negative index |
163 | */ | |
3bf10fea | 164 | enum wl_class_t { |
c0324a02 | 165 | BE_WORKLOAD = 0, |
615f0259 VG |
166 | RT_WORKLOAD = 1, |
167 | IDLE_WORKLOAD = 2, | |
b4627321 | 168 | CFQ_PRIO_NR, |
c0324a02 CZ |
169 | }; |
170 | ||
718eee05 CZ |
171 | /* |
172 | * Second index in the service_trees. | |
173 | */ | |
174 | enum wl_type_t { | |
175 | ASYNC_WORKLOAD = 0, | |
176 | SYNC_NOIDLE_WORKLOAD = 1, | |
177 | SYNC_WORKLOAD = 2 | |
178 | }; | |
179 | ||
155fead9 TH |
180 | struct cfqg_stats { |
181 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
155fead9 TH |
182 | /* number of ios merged */ |
183 | struct blkg_rwstat merged; | |
184 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
185 | struct blkg_rwstat service_time; | |
186 | /* total time spent waiting in scheduler queue in ns */ | |
187 | struct blkg_rwstat wait_time; | |
188 | /* number of IOs queued up */ | |
189 | struct blkg_rwstat queued; | |
155fead9 TH |
190 | /* total disk time and nr sectors dispatched by this group */ |
191 | struct blkg_stat time; | |
192 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
193 | /* time not charged to this cgroup */ | |
194 | struct blkg_stat unaccounted_time; | |
195 | /* sum of number of ios queued across all samples */ | |
196 | struct blkg_stat avg_queue_size_sum; | |
197 | /* count of samples taken for average */ | |
198 | struct blkg_stat avg_queue_size_samples; | |
199 | /* how many times this group has been removed from service tree */ | |
200 | struct blkg_stat dequeue; | |
201 | /* total time spent waiting for it to be assigned a timeslice. */ | |
202 | struct blkg_stat group_wait_time; | |
3c798398 | 203 | /* time spent idling for this blkcg_gq */ |
155fead9 TH |
204 | struct blkg_stat idle_time; |
205 | /* total time with empty current active q with other requests queued */ | |
206 | struct blkg_stat empty_time; | |
207 | /* fields after this shouldn't be cleared on stat reset */ | |
208 | uint64_t start_group_wait_time; | |
209 | uint64_t start_idle_time; | |
210 | uint64_t start_empty_time; | |
211 | uint16_t flags; | |
212 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
213 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
214 | }; | |
215 | ||
e48453c3 AA |
216 | /* Per-cgroup data */ |
217 | struct cfq_group_data { | |
218 | /* must be the first member */ | |
81437648 | 219 | struct blkcg_policy_data cpd; |
e48453c3 AA |
220 | |
221 | unsigned int weight; | |
222 | unsigned int leaf_weight; | |
223 | }; | |
224 | ||
cdb16e8f VG |
225 | /* This is per cgroup per device grouping structure */ |
226 | struct cfq_group { | |
f95a04af TH |
227 | /* must be the first member */ |
228 | struct blkg_policy_data pd; | |
229 | ||
1fa8f6d6 VG |
230 | /* group service_tree member */ |
231 | struct rb_node rb_node; | |
232 | ||
233 | /* group service_tree key */ | |
234 | u64 vdisktime; | |
e71357e1 | 235 | |
7918ffb5 TH |
236 | /* |
237 | * The number of active cfqgs and sum of their weights under this | |
238 | * cfqg. This covers this cfqg's leaf_weight and all children's | |
239 | * weights, but does not cover weights of further descendants. | |
240 | * | |
241 | * If a cfqg is on the service tree, it's active. An active cfqg | |
242 | * also activates its parent and contributes to the children_weight | |
243 | * of the parent. | |
244 | */ | |
245 | int nr_active; | |
246 | unsigned int children_weight; | |
247 | ||
1d3650f7 TH |
248 | /* |
249 | * vfraction is the fraction of vdisktime that the tasks in this | |
250 | * cfqg are entitled to. This is determined by compounding the | |
251 | * ratios walking up from this cfqg to the root. | |
252 | * | |
253 | * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all | |
254 | * vfractions on a service tree is approximately 1. The sum may | |
255 | * deviate a bit due to rounding errors and fluctuations caused by | |
256 | * cfqgs entering and leaving the service tree. | |
257 | */ | |
258 | unsigned int vfraction; | |
259 | ||
e71357e1 TH |
260 | /* |
261 | * There are two weights - (internal) weight is the weight of this | |
262 | * cfqg against the sibling cfqgs. leaf_weight is the wight of | |
263 | * this cfqg against the child cfqgs. For the root cfqg, both | |
264 | * weights are kept in sync for backward compatibility. | |
265 | */ | |
25bc6b07 | 266 | unsigned int weight; |
8184f93e | 267 | unsigned int new_weight; |
3381cb8d | 268 | unsigned int dev_weight; |
1fa8f6d6 | 269 | |
e71357e1 TH |
270 | unsigned int leaf_weight; |
271 | unsigned int new_leaf_weight; | |
272 | unsigned int dev_leaf_weight; | |
273 | ||
1fa8f6d6 VG |
274 | /* number of cfqq currently on this group */ |
275 | int nr_cfqq; | |
276 | ||
cdb16e8f | 277 | /* |
4495a7d4 | 278 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
279 | * create the array for each prio class but at run time it is used |
280 | * only for RT and BE class and slot for IDLE class remains unused. | |
281 | * This is primarily done to avoid confusion and a gcc warning. | |
282 | */ | |
283 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
284 | /* | |
285 | * rr lists of queues with requests. We maintain service trees for | |
286 | * RT and BE classes. These trees are subdivided in subclasses | |
287 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
288 | * class there is no subclassification and all the cfq queues go on | |
289 | * a single tree service_tree_idle. | |
cdb16e8f VG |
290 | * Counts are embedded in the cfq_rb_root |
291 | */ | |
292 | struct cfq_rb_root service_trees[2][3]; | |
293 | struct cfq_rb_root service_tree_idle; | |
dae739eb | 294 | |
9a7f38c4 | 295 | u64 saved_wl_slice; |
4d2ceea4 VG |
296 | enum wl_type_t saved_wl_type; |
297 | enum wl_class_t saved_wl_class; | |
4eef3049 | 298 | |
80bdf0c7 VG |
299 | /* number of requests that are on the dispatch list or inside driver */ |
300 | int dispatched; | |
7700fc4f | 301 | struct cfq_ttime ttime; |
0b39920b | 302 | struct cfqg_stats stats; /* stats for this cfqg */ |
60a83707 TH |
303 | |
304 | /* async queue for each priority case */ | |
305 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
306 | struct cfq_queue *async_idle_cfqq; | |
307 | ||
cdb16e8f | 308 | }; |
718eee05 | 309 | |
c5869807 TH |
310 | struct cfq_io_cq { |
311 | struct io_cq icq; /* must be the first member */ | |
312 | struct cfq_queue *cfqq[2]; | |
313 | struct cfq_ttime ttime; | |
598971bf TH |
314 | int ioprio; /* the current ioprio */ |
315 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
f4da8072 | 316 | uint64_t blkcg_serial_nr; /* the current blkcg serial */ |
598971bf | 317 | #endif |
c5869807 TH |
318 | }; |
319 | ||
22e2c507 JA |
320 | /* |
321 | * Per block device queue structure | |
322 | */ | |
1da177e4 | 323 | struct cfq_data { |
165125e1 | 324 | struct request_queue *queue; |
1fa8f6d6 VG |
325 | /* Root service tree for cfq_groups */ |
326 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 327 | struct cfq_group *root_group; |
22e2c507 | 328 | |
c0324a02 CZ |
329 | /* |
330 | * The priority currently being served | |
22e2c507 | 331 | */ |
4d2ceea4 VG |
332 | enum wl_class_t serving_wl_class; |
333 | enum wl_type_t serving_wl_type; | |
9a7f38c4 | 334 | u64 workload_expires; |
cdb16e8f | 335 | struct cfq_group *serving_group; |
a36e71f9 JA |
336 | |
337 | /* | |
338 | * Each priority tree is sorted by next_request position. These | |
339 | * trees are used when determining if two or more queues are | |
340 | * interleaving requests (see cfq_close_cooperator). | |
341 | */ | |
342 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
343 | ||
22e2c507 | 344 | unsigned int busy_queues; |
ef8a41df | 345 | unsigned int busy_sync_queues; |
22e2c507 | 346 | |
53c583d2 CZ |
347 | int rq_in_driver; |
348 | int rq_in_flight[2]; | |
45333d5a AC |
349 | |
350 | /* | |
351 | * queue-depth detection | |
352 | */ | |
353 | int rq_queued; | |
25776e35 | 354 | int hw_tag; |
e459dd08 CZ |
355 | /* |
356 | * hw_tag can be | |
357 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
358 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
359 | * 0 => no NCQ | |
360 | */ | |
361 | int hw_tag_est_depth; | |
362 | unsigned int hw_tag_samples; | |
1da177e4 | 363 | |
22e2c507 JA |
364 | /* |
365 | * idle window management | |
366 | */ | |
91148325 | 367 | struct hrtimer idle_slice_timer; |
23e018a1 | 368 | struct work_struct unplug_work; |
1da177e4 | 369 | |
22e2c507 | 370 | struct cfq_queue *active_queue; |
c5869807 | 371 | struct cfq_io_cq *active_cic; |
22e2c507 | 372 | |
6d048f53 | 373 | sector_t last_position; |
1da177e4 | 374 | |
1da177e4 LT |
375 | /* |
376 | * tunables, see top of file | |
377 | */ | |
378 | unsigned int cfq_quantum; | |
1da177e4 LT |
379 | unsigned int cfq_back_penalty; |
380 | unsigned int cfq_back_max; | |
22e2c507 | 381 | unsigned int cfq_slice_async_rq; |
963b72fc | 382 | unsigned int cfq_latency; |
9a7f38c4 JM |
383 | u64 cfq_fifo_expire[2]; |
384 | u64 cfq_slice[2]; | |
385 | u64 cfq_slice_idle; | |
386 | u64 cfq_group_idle; | |
387 | u64 cfq_target_latency; | |
d9ff4187 | 388 | |
6118b70b JA |
389 | /* |
390 | * Fallback dummy cfqq for extreme OOM conditions | |
391 | */ | |
392 | struct cfq_queue oom_cfqq; | |
365722bb | 393 | |
9a7f38c4 | 394 | u64 last_delayed_sync; |
1da177e4 LT |
395 | }; |
396 | ||
25fb5169 | 397 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
60a83707 | 398 | static void cfq_put_queue(struct cfq_queue *cfqq); |
25fb5169 | 399 | |
34b98d03 | 400 | static struct cfq_rb_root *st_for(struct cfq_group *cfqg, |
3bf10fea | 401 | enum wl_class_t class, |
65b32a57 | 402 | enum wl_type_t type) |
c0324a02 | 403 | { |
1fa8f6d6 VG |
404 | if (!cfqg) |
405 | return NULL; | |
406 | ||
3bf10fea | 407 | if (class == IDLE_WORKLOAD) |
cdb16e8f | 408 | return &cfqg->service_tree_idle; |
c0324a02 | 409 | |
3bf10fea | 410 | return &cfqg->service_trees[class][type]; |
c0324a02 CZ |
411 | } |
412 | ||
3b18152c | 413 | enum cfqq_state_flags { |
b0b8d749 JA |
414 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
415 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 416 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 417 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
418 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
419 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
420 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 421 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 422 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 423 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 424 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 425 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 426 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
427 | }; |
428 | ||
429 | #define CFQ_CFQQ_FNS(name) \ | |
430 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
431 | { \ | |
fe094d98 | 432 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
433 | } \ |
434 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
435 | { \ | |
fe094d98 | 436 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
437 | } \ |
438 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
439 | { \ | |
fe094d98 | 440 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
441 | } |
442 | ||
443 | CFQ_CFQQ_FNS(on_rr); | |
444 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 445 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 446 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
447 | CFQ_CFQQ_FNS(fifo_expire); |
448 | CFQ_CFQQ_FNS(idle_window); | |
449 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 450 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 451 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 452 | CFQ_CFQQ_FNS(coop); |
ae54abed | 453 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 454 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 455 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
456 | #undef CFQ_CFQQ_FNS |
457 | ||
629ed0b1 | 458 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
2ce4d50f | 459 | |
155fead9 TH |
460 | /* cfqg stats flags */ |
461 | enum cfqg_stats_flags { | |
462 | CFQG_stats_waiting = 0, | |
463 | CFQG_stats_idling, | |
464 | CFQG_stats_empty, | |
629ed0b1 TH |
465 | }; |
466 | ||
155fead9 TH |
467 | #define CFQG_FLAG_FNS(name) \ |
468 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ | |
629ed0b1 | 469 | { \ |
155fead9 | 470 | stats->flags |= (1 << CFQG_stats_##name); \ |
629ed0b1 | 471 | } \ |
155fead9 | 472 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 473 | { \ |
155fead9 | 474 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
629ed0b1 | 475 | } \ |
155fead9 | 476 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 477 | { \ |
155fead9 | 478 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
629ed0b1 TH |
479 | } \ |
480 | ||
155fead9 TH |
481 | CFQG_FLAG_FNS(waiting) |
482 | CFQG_FLAG_FNS(idling) | |
483 | CFQG_FLAG_FNS(empty) | |
484 | #undef CFQG_FLAG_FNS | |
629ed0b1 TH |
485 | |
486 | /* This should be called with the queue_lock held. */ | |
155fead9 | 487 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
629ed0b1 TH |
488 | { |
489 | unsigned long long now; | |
490 | ||
155fead9 | 491 | if (!cfqg_stats_waiting(stats)) |
629ed0b1 TH |
492 | return; |
493 | ||
494 | now = sched_clock(); | |
495 | if (time_after64(now, stats->start_group_wait_time)) | |
496 | blkg_stat_add(&stats->group_wait_time, | |
497 | now - stats->start_group_wait_time); | |
155fead9 | 498 | cfqg_stats_clear_waiting(stats); |
629ed0b1 TH |
499 | } |
500 | ||
501 | /* This should be called with the queue_lock held. */ | |
155fead9 TH |
502 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
503 | struct cfq_group *curr_cfqg) | |
629ed0b1 | 504 | { |
155fead9 | 505 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 506 | |
155fead9 | 507 | if (cfqg_stats_waiting(stats)) |
629ed0b1 | 508 | return; |
155fead9 | 509 | if (cfqg == curr_cfqg) |
629ed0b1 | 510 | return; |
155fead9 TH |
511 | stats->start_group_wait_time = sched_clock(); |
512 | cfqg_stats_mark_waiting(stats); | |
629ed0b1 TH |
513 | } |
514 | ||
515 | /* This should be called with the queue_lock held. */ | |
155fead9 | 516 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
629ed0b1 TH |
517 | { |
518 | unsigned long long now; | |
519 | ||
155fead9 | 520 | if (!cfqg_stats_empty(stats)) |
629ed0b1 TH |
521 | return; |
522 | ||
523 | now = sched_clock(); | |
524 | if (time_after64(now, stats->start_empty_time)) | |
525 | blkg_stat_add(&stats->empty_time, | |
526 | now - stats->start_empty_time); | |
155fead9 | 527 | cfqg_stats_clear_empty(stats); |
629ed0b1 TH |
528 | } |
529 | ||
155fead9 | 530 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
629ed0b1 | 531 | { |
155fead9 | 532 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
629ed0b1 TH |
533 | } |
534 | ||
155fead9 | 535 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
629ed0b1 | 536 | { |
155fead9 | 537 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 538 | |
4d5e80a7 | 539 | if (blkg_rwstat_total(&stats->queued)) |
629ed0b1 TH |
540 | return; |
541 | ||
542 | /* | |
543 | * group is already marked empty. This can happen if cfqq got new | |
544 | * request in parent group and moved to this group while being added | |
545 | * to service tree. Just ignore the event and move on. | |
546 | */ | |
155fead9 | 547 | if (cfqg_stats_empty(stats)) |
629ed0b1 TH |
548 | return; |
549 | ||
550 | stats->start_empty_time = sched_clock(); | |
155fead9 | 551 | cfqg_stats_mark_empty(stats); |
629ed0b1 TH |
552 | } |
553 | ||
155fead9 | 554 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 555 | { |
155fead9 | 556 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 557 | |
155fead9 | 558 | if (cfqg_stats_idling(stats)) { |
629ed0b1 TH |
559 | unsigned long long now = sched_clock(); |
560 | ||
561 | if (time_after64(now, stats->start_idle_time)) | |
562 | blkg_stat_add(&stats->idle_time, | |
563 | now - stats->start_idle_time); | |
155fead9 | 564 | cfqg_stats_clear_idling(stats); |
629ed0b1 TH |
565 | } |
566 | } | |
567 | ||
155fead9 | 568 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 569 | { |
155fead9 | 570 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 571 | |
155fead9 | 572 | BUG_ON(cfqg_stats_idling(stats)); |
629ed0b1 TH |
573 | |
574 | stats->start_idle_time = sched_clock(); | |
155fead9 | 575 | cfqg_stats_mark_idling(stats); |
629ed0b1 TH |
576 | } |
577 | ||
155fead9 | 578 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
629ed0b1 | 579 | { |
155fead9 | 580 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
581 | |
582 | blkg_stat_add(&stats->avg_queue_size_sum, | |
4d5e80a7 | 583 | blkg_rwstat_total(&stats->queued)); |
629ed0b1 | 584 | blkg_stat_add(&stats->avg_queue_size_samples, 1); |
155fead9 | 585 | cfqg_stats_update_group_wait_time(stats); |
629ed0b1 TH |
586 | } |
587 | ||
588 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
589 | ||
f48ec1d7 TH |
590 | static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { } |
591 | static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } | |
592 | static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } | |
593 | static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } | |
594 | static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } | |
595 | static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } | |
596 | static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } | |
629ed0b1 TH |
597 | |
598 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
599 | ||
600 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
2ce4d50f | 601 | |
4ceab71b JA |
602 | static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd) |
603 | { | |
604 | return pd ? container_of(pd, struct cfq_group, pd) : NULL; | |
605 | } | |
606 | ||
607 | static struct cfq_group_data | |
608 | *cpd_to_cfqgd(struct blkcg_policy_data *cpd) | |
609 | { | |
81437648 | 610 | return cpd ? container_of(cpd, struct cfq_group_data, cpd) : NULL; |
4ceab71b JA |
611 | } |
612 | ||
613 | static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg) | |
614 | { | |
615 | return pd_to_blkg(&cfqg->pd); | |
616 | } | |
617 | ||
ffea73fc TH |
618 | static struct blkcg_policy blkcg_policy_cfq; |
619 | ||
620 | static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg) | |
621 | { | |
622 | return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq)); | |
623 | } | |
624 | ||
e48453c3 AA |
625 | static struct cfq_group_data *blkcg_to_cfqgd(struct blkcg *blkcg) |
626 | { | |
627 | return cpd_to_cfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_cfq)); | |
628 | } | |
629 | ||
d02f7aa8 | 630 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) |
7918ffb5 | 631 | { |
d02f7aa8 | 632 | struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent; |
7918ffb5 | 633 | |
d02f7aa8 | 634 | return pblkg ? blkg_to_cfqg(pblkg) : NULL; |
7918ffb5 TH |
635 | } |
636 | ||
3984aa55 JK |
637 | static inline bool cfqg_is_descendant(struct cfq_group *cfqg, |
638 | struct cfq_group *ancestor) | |
639 | { | |
640 | return cgroup_is_descendant(cfqg_to_blkg(cfqg)->blkcg->css.cgroup, | |
641 | cfqg_to_blkg(ancestor)->blkcg->css.cgroup); | |
642 | } | |
643 | ||
eb7d8c07 TH |
644 | static inline void cfqg_get(struct cfq_group *cfqg) |
645 | { | |
646 | return blkg_get(cfqg_to_blkg(cfqg)); | |
647 | } | |
648 | ||
649 | static inline void cfqg_put(struct cfq_group *cfqg) | |
650 | { | |
651 | return blkg_put(cfqg_to_blkg(cfqg)); | |
652 | } | |
653 | ||
54e7ed12 TH |
654 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \ |
655 | char __pbuf[128]; \ | |
656 | \ | |
657 | blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \ | |
b226e5c4 VG |
658 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \ |
659 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
660 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
54e7ed12 TH |
661 | __pbuf, ##args); \ |
662 | } while (0) | |
663 | ||
664 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \ | |
665 | char __pbuf[128]; \ | |
666 | \ | |
667 | blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \ | |
668 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \ | |
669 | } while (0) | |
2868ef7b | 670 | |
155fead9 | 671 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
ef295ecf CH |
672 | struct cfq_group *curr_cfqg, |
673 | unsigned int op) | |
2ce4d50f | 674 | { |
ef295ecf | 675 | blkg_rwstat_add(&cfqg->stats.queued, op, 1); |
155fead9 TH |
676 | cfqg_stats_end_empty_time(&cfqg->stats); |
677 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); | |
2ce4d50f TH |
678 | } |
679 | ||
155fead9 | 680 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
9a7f38c4 | 681 | uint64_t time, unsigned long unaccounted_time) |
2ce4d50f | 682 | { |
155fead9 | 683 | blkg_stat_add(&cfqg->stats.time, time); |
629ed0b1 | 684 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
155fead9 | 685 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
629ed0b1 | 686 | #endif |
2ce4d50f TH |
687 | } |
688 | ||
ef295ecf CH |
689 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, |
690 | unsigned int op) | |
2ce4d50f | 691 | { |
ef295ecf | 692 | blkg_rwstat_add(&cfqg->stats.queued, op, -1); |
2ce4d50f TH |
693 | } |
694 | ||
ef295ecf CH |
695 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, |
696 | unsigned int op) | |
2ce4d50f | 697 | { |
ef295ecf | 698 | blkg_rwstat_add(&cfqg->stats.merged, op, 1); |
2ce4d50f TH |
699 | } |
700 | ||
155fead9 | 701 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
ef295ecf CH |
702 | uint64_t start_time, uint64_t io_start_time, |
703 | unsigned int op) | |
2ce4d50f | 704 | { |
155fead9 | 705 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 706 | unsigned long long now = sched_clock(); |
629ed0b1 TH |
707 | |
708 | if (time_after64(now, io_start_time)) | |
ef295ecf | 709 | blkg_rwstat_add(&stats->service_time, op, now - io_start_time); |
629ed0b1 | 710 | if (time_after64(io_start_time, start_time)) |
ef295ecf | 711 | blkg_rwstat_add(&stats->wait_time, op, |
629ed0b1 | 712 | io_start_time - start_time); |
2ce4d50f TH |
713 | } |
714 | ||
689665af TH |
715 | /* @stats = 0 */ |
716 | static void cfqg_stats_reset(struct cfqg_stats *stats) | |
155fead9 | 717 | { |
155fead9 | 718 | /* queued stats shouldn't be cleared */ |
155fead9 TH |
719 | blkg_rwstat_reset(&stats->merged); |
720 | blkg_rwstat_reset(&stats->service_time); | |
721 | blkg_rwstat_reset(&stats->wait_time); | |
722 | blkg_stat_reset(&stats->time); | |
723 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
724 | blkg_stat_reset(&stats->unaccounted_time); | |
725 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
726 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
727 | blkg_stat_reset(&stats->dequeue); | |
728 | blkg_stat_reset(&stats->group_wait_time); | |
729 | blkg_stat_reset(&stats->idle_time); | |
730 | blkg_stat_reset(&stats->empty_time); | |
731 | #endif | |
732 | } | |
733 | ||
0b39920b | 734 | /* @to += @from */ |
e6269c44 | 735 | static void cfqg_stats_add_aux(struct cfqg_stats *to, struct cfqg_stats *from) |
0b39920b TH |
736 | { |
737 | /* queued stats shouldn't be cleared */ | |
e6269c44 TH |
738 | blkg_rwstat_add_aux(&to->merged, &from->merged); |
739 | blkg_rwstat_add_aux(&to->service_time, &from->service_time); | |
740 | blkg_rwstat_add_aux(&to->wait_time, &from->wait_time); | |
741 | blkg_stat_add_aux(&from->time, &from->time); | |
0b39920b | 742 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
e6269c44 TH |
743 | blkg_stat_add_aux(&to->unaccounted_time, &from->unaccounted_time); |
744 | blkg_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum); | |
745 | blkg_stat_add_aux(&to->avg_queue_size_samples, &from->avg_queue_size_samples); | |
746 | blkg_stat_add_aux(&to->dequeue, &from->dequeue); | |
747 | blkg_stat_add_aux(&to->group_wait_time, &from->group_wait_time); | |
748 | blkg_stat_add_aux(&to->idle_time, &from->idle_time); | |
749 | blkg_stat_add_aux(&to->empty_time, &from->empty_time); | |
0b39920b TH |
750 | #endif |
751 | } | |
752 | ||
753 | /* | |
e6269c44 | 754 | * Transfer @cfqg's stats to its parent's aux counts so that the ancestors' |
0b39920b TH |
755 | * recursive stats can still account for the amount used by this cfqg after |
756 | * it's gone. | |
757 | */ | |
758 | static void cfqg_stats_xfer_dead(struct cfq_group *cfqg) | |
759 | { | |
760 | struct cfq_group *parent = cfqg_parent(cfqg); | |
761 | ||
762 | lockdep_assert_held(cfqg_to_blkg(cfqg)->q->queue_lock); | |
763 | ||
764 | if (unlikely(!parent)) | |
765 | return; | |
766 | ||
e6269c44 | 767 | cfqg_stats_add_aux(&parent->stats, &cfqg->stats); |
0b39920b | 768 | cfqg_stats_reset(&cfqg->stats); |
0b39920b TH |
769 | } |
770 | ||
eb7d8c07 TH |
771 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
772 | ||
d02f7aa8 | 773 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) { return NULL; } |
3984aa55 JK |
774 | static inline bool cfqg_is_descendant(struct cfq_group *cfqg, |
775 | struct cfq_group *ancestor) | |
776 | { | |
777 | return true; | |
778 | } | |
eb7d8c07 TH |
779 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
780 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
781 | ||
7b679138 | 782 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
b226e5c4 VG |
783 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \ |
784 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
785 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
786 | ##args) | |
4495a7d4 | 787 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 | 788 | |
155fead9 | 789 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
ef295ecf | 790 | struct cfq_group *curr_cfqg, unsigned int op) { } |
155fead9 | 791 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
9a7f38c4 | 792 | uint64_t time, unsigned long unaccounted_time) { } |
ef295ecf CH |
793 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, |
794 | unsigned int op) { } | |
795 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, | |
796 | unsigned int op) { } | |
155fead9 | 797 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
ef295ecf CH |
798 | uint64_t start_time, uint64_t io_start_time, |
799 | unsigned int op) { } | |
2ce4d50f | 800 | |
eb7d8c07 TH |
801 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
802 | ||
7b679138 JA |
803 | #define cfq_log(cfqd, fmt, args...) \ |
804 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
805 | ||
615f0259 VG |
806 | /* Traverses through cfq group service trees */ |
807 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
808 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
809 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
810 | : &cfqg->service_tree_idle; \ | |
811 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
812 | (i == IDLE_WORKLOAD && j == 0); \ | |
813 | j++, st = i < IDLE_WORKLOAD ? \ | |
814 | &cfqg->service_trees[i][j]: NULL) \ | |
815 | ||
f5f2b6ce SL |
816 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
817 | struct cfq_ttime *ttime, bool group_idle) | |
818 | { | |
9a7f38c4 | 819 | u64 slice; |
f5f2b6ce SL |
820 | if (!sample_valid(ttime->ttime_samples)) |
821 | return false; | |
822 | if (group_idle) | |
823 | slice = cfqd->cfq_group_idle; | |
824 | else | |
825 | slice = cfqd->cfq_slice_idle; | |
826 | return ttime->ttime_mean > slice; | |
827 | } | |
615f0259 | 828 | |
02b35081 VG |
829 | static inline bool iops_mode(struct cfq_data *cfqd) |
830 | { | |
831 | /* | |
832 | * If we are not idling on queues and it is a NCQ drive, parallel | |
833 | * execution of requests is on and measuring time is not possible | |
834 | * in most of the cases until and unless we drive shallower queue | |
835 | * depths and that becomes a performance bottleneck. In such cases | |
836 | * switch to start providing fairness in terms of number of IOs. | |
837 | */ | |
838 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
839 | return true; | |
840 | else | |
841 | return false; | |
842 | } | |
843 | ||
3bf10fea | 844 | static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq) |
c0324a02 CZ |
845 | { |
846 | if (cfq_class_idle(cfqq)) | |
847 | return IDLE_WORKLOAD; | |
848 | if (cfq_class_rt(cfqq)) | |
849 | return RT_WORKLOAD; | |
850 | return BE_WORKLOAD; | |
851 | } | |
852 | ||
718eee05 CZ |
853 | |
854 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
855 | { | |
856 | if (!cfq_cfqq_sync(cfqq)) | |
857 | return ASYNC_WORKLOAD; | |
858 | if (!cfq_cfqq_idle_window(cfqq)) | |
859 | return SYNC_NOIDLE_WORKLOAD; | |
860 | return SYNC_WORKLOAD; | |
861 | } | |
862 | ||
3bf10fea | 863 | static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class, |
58ff82f3 VG |
864 | struct cfq_data *cfqd, |
865 | struct cfq_group *cfqg) | |
c0324a02 | 866 | { |
3bf10fea | 867 | if (wl_class == IDLE_WORKLOAD) |
cdb16e8f | 868 | return cfqg->service_tree_idle.count; |
c0324a02 | 869 | |
34b98d03 VG |
870 | return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count + |
871 | cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count + | |
872 | cfqg->service_trees[wl_class][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
873 | } |
874 | ||
f26bd1f0 VG |
875 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
876 | struct cfq_group *cfqg) | |
877 | { | |
34b98d03 VG |
878 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count + |
879 | cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
f26bd1f0 VG |
880 | } |
881 | ||
165125e1 | 882 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 883 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
2da8de0b | 884 | struct cfq_io_cq *cic, struct bio *bio); |
91fac317 | 885 | |
c5869807 TH |
886 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
887 | { | |
888 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
889 | return container_of(icq, struct cfq_io_cq, icq); | |
890 | } | |
891 | ||
47fdd4ca TH |
892 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
893 | struct io_context *ioc) | |
894 | { | |
895 | if (ioc) | |
896 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
897 | return NULL; | |
898 | } | |
899 | ||
c5869807 | 900 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 901 | { |
a6151c3a | 902 | return cic->cfqq[is_sync]; |
91fac317 VT |
903 | } |
904 | ||
c5869807 TH |
905 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
906 | bool is_sync) | |
91fac317 | 907 | { |
a6151c3a | 908 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
909 | } |
910 | ||
c5869807 | 911 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 912 | { |
c5869807 | 913 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
914 | } |
915 | ||
99f95e52 AM |
916 | /* |
917 | * scheduler run of queue, if there are requests pending and no one in the | |
918 | * driver that will restart queueing | |
919 | */ | |
23e018a1 | 920 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 921 | { |
7b679138 JA |
922 | if (cfqd->busy_queues) { |
923 | cfq_log(cfqd, "schedule dispatch"); | |
59c3d45e | 924 | kblockd_schedule_work(&cfqd->unplug_work); |
7b679138 | 925 | } |
99f95e52 AM |
926 | } |
927 | ||
44f7c160 JA |
928 | /* |
929 | * Scale schedule slice based on io priority. Use the sync time slice only | |
930 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
931 | * io only, should not get full sync slice length. | |
932 | */ | |
9a7f38c4 | 933 | static inline u64 cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 934 | unsigned short prio) |
44f7c160 | 935 | { |
9a7f38c4 JM |
936 | u64 base_slice = cfqd->cfq_slice[sync]; |
937 | u64 slice = div_u64(base_slice, CFQ_SLICE_SCALE); | |
44f7c160 | 938 | |
d9e7620e JA |
939 | WARN_ON(prio >= IOPRIO_BE_NR); |
940 | ||
9a7f38c4 | 941 | return base_slice + (slice * (4 - prio)); |
d9e7620e | 942 | } |
44f7c160 | 943 | |
9a7f38c4 | 944 | static inline u64 |
d9e7620e JA |
945 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
946 | { | |
947 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
948 | } |
949 | ||
1d3650f7 TH |
950 | /** |
951 | * cfqg_scale_charge - scale disk time charge according to cfqg weight | |
952 | * @charge: disk time being charged | |
953 | * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT | |
954 | * | |
955 | * Scale @charge according to @vfraction, which is in range (0, 1]. The | |
956 | * scaling is inversely proportional. | |
957 | * | |
958 | * scaled = charge / vfraction | |
959 | * | |
960 | * The result is also in fixed point w/ CFQ_SERVICE_SHIFT. | |
961 | */ | |
9a7f38c4 | 962 | static inline u64 cfqg_scale_charge(u64 charge, |
1d3650f7 | 963 | unsigned int vfraction) |
25bc6b07 | 964 | { |
1d3650f7 | 965 | u64 c = charge << CFQ_SERVICE_SHIFT; /* make it fixed point */ |
25bc6b07 | 966 | |
1d3650f7 TH |
967 | /* charge / vfraction */ |
968 | c <<= CFQ_SERVICE_SHIFT; | |
9a7f38c4 | 969 | return div_u64(c, vfraction); |
25bc6b07 VG |
970 | } |
971 | ||
972 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
973 | { | |
974 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
975 | if (delta > 0) | |
976 | min_vdisktime = vdisktime; | |
977 | ||
978 | return min_vdisktime; | |
979 | } | |
980 | ||
25bc6b07 VG |
981 | static void update_min_vdisktime(struct cfq_rb_root *st) |
982 | { | |
25bc6b07 VG |
983 | struct cfq_group *cfqg; |
984 | ||
25bc6b07 VG |
985 | if (st->left) { |
986 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
987 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
988 | cfqg->vdisktime); | |
25bc6b07 | 989 | } |
25bc6b07 VG |
990 | } |
991 | ||
5db5d642 CZ |
992 | /* |
993 | * get averaged number of queues of RT/BE priority. | |
994 | * average is updated, with a formula that gives more weight to higher numbers, | |
995 | * to quickly follows sudden increases and decrease slowly | |
996 | */ | |
997 | ||
58ff82f3 VG |
998 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
999 | struct cfq_group *cfqg, bool rt) | |
5869619c | 1000 | { |
5db5d642 CZ |
1001 | unsigned min_q, max_q; |
1002 | unsigned mult = cfq_hist_divisor - 1; | |
1003 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 1004 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 1005 | |
58ff82f3 VG |
1006 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
1007 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
1008 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 1009 | cfq_hist_divisor; |
58ff82f3 VG |
1010 | return cfqg->busy_queues_avg[rt]; |
1011 | } | |
1012 | ||
9a7f38c4 | 1013 | static inline u64 |
58ff82f3 VG |
1014 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1015 | { | |
41cad6ab | 1016 | return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT; |
5db5d642 CZ |
1017 | } |
1018 | ||
9a7f38c4 | 1019 | static inline u64 |
ba5bd520 | 1020 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 1021 | { |
9a7f38c4 | 1022 | u64 slice = cfq_prio_to_slice(cfqd, cfqq); |
5db5d642 | 1023 | if (cfqd->cfq_latency) { |
58ff82f3 VG |
1024 | /* |
1025 | * interested queues (we consider only the ones with the same | |
1026 | * priority class in the cfq group) | |
1027 | */ | |
1028 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
1029 | cfq_class_rt(cfqq)); | |
9a7f38c4 JM |
1030 | u64 sync_slice = cfqd->cfq_slice[1]; |
1031 | u64 expect_latency = sync_slice * iq; | |
1032 | u64 group_slice = cfq_group_slice(cfqd, cfqq->cfqg); | |
58ff82f3 VG |
1033 | |
1034 | if (expect_latency > group_slice) { | |
9a7f38c4 JM |
1035 | u64 base_low_slice = 2 * cfqd->cfq_slice_idle; |
1036 | u64 low_slice; | |
1037 | ||
5db5d642 CZ |
1038 | /* scale low_slice according to IO priority |
1039 | * and sync vs async */ | |
9a7f38c4 JM |
1040 | low_slice = div64_u64(base_low_slice*slice, sync_slice); |
1041 | low_slice = min(slice, low_slice); | |
5db5d642 CZ |
1042 | /* the adapted slice value is scaled to fit all iqs |
1043 | * into the target latency */ | |
9a7f38c4 JM |
1044 | slice = div64_u64(slice*group_slice, expect_latency); |
1045 | slice = max(slice, low_slice); | |
5db5d642 CZ |
1046 | } |
1047 | } | |
c553f8e3 SL |
1048 | return slice; |
1049 | } | |
1050 | ||
1051 | static inline void | |
1052 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1053 | { | |
9a7f38c4 JM |
1054 | u64 slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
1055 | u64 now = ktime_get_ns(); | |
c553f8e3 | 1056 | |
9a7f38c4 JM |
1057 | cfqq->slice_start = now; |
1058 | cfqq->slice_end = now + slice; | |
f75edf2d | 1059 | cfqq->allocated_slice = slice; |
9a7f38c4 | 1060 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%llu", cfqq->slice_end - now); |
44f7c160 JA |
1061 | } |
1062 | ||
1063 | /* | |
1064 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
1065 | * isn't valid until the first request from the dispatch is activated | |
1066 | * and the slice time set. | |
1067 | */ | |
a6151c3a | 1068 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
1069 | { |
1070 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 1071 | return false; |
9a7f38c4 | 1072 | if (ktime_get_ns() < cfqq->slice_end) |
c1e44756 | 1073 | return false; |
44f7c160 | 1074 | |
c1e44756 | 1075 | return true; |
44f7c160 JA |
1076 | } |
1077 | ||
1da177e4 | 1078 | /* |
5e705374 | 1079 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 1080 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 1081 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 1082 | */ |
5e705374 | 1083 | static struct request * |
cf7c25cf | 1084 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 1085 | { |
cf7c25cf | 1086 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 1087 | unsigned long back_max; |
e8a99053 AM |
1088 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
1089 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
1090 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 1091 | |
5e705374 JA |
1092 | if (rq1 == NULL || rq1 == rq2) |
1093 | return rq2; | |
1094 | if (rq2 == NULL) | |
1095 | return rq1; | |
9c2c38a1 | 1096 | |
229836bd NK |
1097 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
1098 | return rq_is_sync(rq1) ? rq1 : rq2; | |
1099 | ||
65299a3b CH |
1100 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
1101 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 1102 | |
83096ebf TH |
1103 | s1 = blk_rq_pos(rq1); |
1104 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 1105 | |
1da177e4 LT |
1106 | /* |
1107 | * by definition, 1KiB is 2 sectors | |
1108 | */ | |
1109 | back_max = cfqd->cfq_back_max * 2; | |
1110 | ||
1111 | /* | |
1112 | * Strict one way elevator _except_ in the case where we allow | |
1113 | * short backward seeks which are biased as twice the cost of a | |
1114 | * similar forward seek. | |
1115 | */ | |
1116 | if (s1 >= last) | |
1117 | d1 = s1 - last; | |
1118 | else if (s1 + back_max >= last) | |
1119 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
1120 | else | |
e8a99053 | 1121 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
1122 | |
1123 | if (s2 >= last) | |
1124 | d2 = s2 - last; | |
1125 | else if (s2 + back_max >= last) | |
1126 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
1127 | else | |
e8a99053 | 1128 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
1129 | |
1130 | /* Found required data */ | |
e8a99053 AM |
1131 | |
1132 | /* | |
1133 | * By doing switch() on the bit mask "wrap" we avoid having to | |
1134 | * check two variables for all permutations: --> faster! | |
1135 | */ | |
1136 | switch (wrap) { | |
5e705374 | 1137 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 1138 | if (d1 < d2) |
5e705374 | 1139 | return rq1; |
e8a99053 | 1140 | else if (d2 < d1) |
5e705374 | 1141 | return rq2; |
e8a99053 AM |
1142 | else { |
1143 | if (s1 >= s2) | |
5e705374 | 1144 | return rq1; |
e8a99053 | 1145 | else |
5e705374 | 1146 | return rq2; |
e8a99053 | 1147 | } |
1da177e4 | 1148 | |
e8a99053 | 1149 | case CFQ_RQ2_WRAP: |
5e705374 | 1150 | return rq1; |
e8a99053 | 1151 | case CFQ_RQ1_WRAP: |
5e705374 JA |
1152 | return rq2; |
1153 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
1154 | default: |
1155 | /* | |
1156 | * Since both rqs are wrapped, | |
1157 | * start with the one that's further behind head | |
1158 | * (--> only *one* back seek required), | |
1159 | * since back seek takes more time than forward. | |
1160 | */ | |
1161 | if (s1 <= s2) | |
5e705374 | 1162 | return rq1; |
1da177e4 | 1163 | else |
5e705374 | 1164 | return rq2; |
1da177e4 LT |
1165 | } |
1166 | } | |
1167 | ||
498d3aa2 JA |
1168 | /* |
1169 | * The below is leftmost cache rbtree addon | |
1170 | */ | |
0871714e | 1171 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 1172 | { |
615f0259 VG |
1173 | /* Service tree is empty */ |
1174 | if (!root->count) | |
1175 | return NULL; | |
1176 | ||
cc09e299 JA |
1177 | if (!root->left) |
1178 | root->left = rb_first(&root->rb); | |
1179 | ||
0871714e JA |
1180 | if (root->left) |
1181 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
1182 | ||
1183 | return NULL; | |
cc09e299 JA |
1184 | } |
1185 | ||
1fa8f6d6 VG |
1186 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
1187 | { | |
1188 | if (!root->left) | |
1189 | root->left = rb_first(&root->rb); | |
1190 | ||
1191 | if (root->left) | |
1192 | return rb_entry_cfqg(root->left); | |
1193 | ||
1194 | return NULL; | |
1195 | } | |
1196 | ||
a36e71f9 JA |
1197 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
1198 | { | |
1199 | rb_erase(n, root); | |
1200 | RB_CLEAR_NODE(n); | |
1201 | } | |
1202 | ||
cc09e299 JA |
1203 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
1204 | { | |
1205 | if (root->left == n) | |
1206 | root->left = NULL; | |
a36e71f9 | 1207 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 1208 | --root->count; |
cc09e299 JA |
1209 | } |
1210 | ||
1da177e4 LT |
1211 | /* |
1212 | * would be nice to take fifo expire time into account as well | |
1213 | */ | |
5e705374 JA |
1214 | static struct request * |
1215 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1216 | struct request *last) | |
1da177e4 | 1217 | { |
21183b07 JA |
1218 | struct rb_node *rbnext = rb_next(&last->rb_node); |
1219 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 1220 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 1221 | |
21183b07 | 1222 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
1223 | |
1224 | if (rbprev) | |
5e705374 | 1225 | prev = rb_entry_rq(rbprev); |
1da177e4 | 1226 | |
21183b07 | 1227 | if (rbnext) |
5e705374 | 1228 | next = rb_entry_rq(rbnext); |
21183b07 JA |
1229 | else { |
1230 | rbnext = rb_first(&cfqq->sort_list); | |
1231 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 1232 | next = rb_entry_rq(rbnext); |
21183b07 | 1233 | } |
1da177e4 | 1234 | |
cf7c25cf | 1235 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
1236 | } |
1237 | ||
9a7f38c4 JM |
1238 | static u64 cfq_slice_offset(struct cfq_data *cfqd, |
1239 | struct cfq_queue *cfqq) | |
1da177e4 | 1240 | { |
d9e7620e JA |
1241 | /* |
1242 | * just an approximation, should be ok. | |
1243 | */ | |
cdb16e8f | 1244 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 1245 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
1246 | } |
1247 | ||
1fa8f6d6 VG |
1248 | static inline s64 |
1249 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1250 | { | |
1251 | return cfqg->vdisktime - st->min_vdisktime; | |
1252 | } | |
1253 | ||
1254 | static void | |
1255 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1256 | { | |
1257 | struct rb_node **node = &st->rb.rb_node; | |
1258 | struct rb_node *parent = NULL; | |
1259 | struct cfq_group *__cfqg; | |
1260 | s64 key = cfqg_key(st, cfqg); | |
1261 | int left = 1; | |
1262 | ||
1263 | while (*node != NULL) { | |
1264 | parent = *node; | |
1265 | __cfqg = rb_entry_cfqg(parent); | |
1266 | ||
1267 | if (key < cfqg_key(st, __cfqg)) | |
1268 | node = &parent->rb_left; | |
1269 | else { | |
1270 | node = &parent->rb_right; | |
1271 | left = 0; | |
1272 | } | |
1273 | } | |
1274 | ||
1275 | if (left) | |
1276 | st->left = &cfqg->rb_node; | |
1277 | ||
1278 | rb_link_node(&cfqg->rb_node, parent, node); | |
1279 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
1280 | } | |
1281 | ||
7b5af5cf TM |
1282 | /* |
1283 | * This has to be called only on activation of cfqg | |
1284 | */ | |
1fa8f6d6 | 1285 | static void |
8184f93e JT |
1286 | cfq_update_group_weight(struct cfq_group *cfqg) |
1287 | { | |
3381cb8d | 1288 | if (cfqg->new_weight) { |
8184f93e | 1289 | cfqg->weight = cfqg->new_weight; |
3381cb8d | 1290 | cfqg->new_weight = 0; |
8184f93e | 1291 | } |
e15693ef TM |
1292 | } |
1293 | ||
1294 | static void | |
1295 | cfq_update_group_leaf_weight(struct cfq_group *cfqg) | |
1296 | { | |
1297 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
e71357e1 TH |
1298 | |
1299 | if (cfqg->new_leaf_weight) { | |
1300 | cfqg->leaf_weight = cfqg->new_leaf_weight; | |
1301 | cfqg->new_leaf_weight = 0; | |
1302 | } | |
8184f93e JT |
1303 | } |
1304 | ||
1305 | static void | |
1306 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1307 | { | |
1d3650f7 | 1308 | unsigned int vfr = 1 << CFQ_SERVICE_SHIFT; /* start with 1 */ |
7918ffb5 | 1309 | struct cfq_group *pos = cfqg; |
1d3650f7 | 1310 | struct cfq_group *parent; |
7918ffb5 TH |
1311 | bool propagate; |
1312 | ||
1313 | /* add to the service tree */ | |
8184f93e JT |
1314 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
1315 | ||
7b5af5cf TM |
1316 | /* |
1317 | * Update leaf_weight. We cannot update weight at this point | |
1318 | * because cfqg might already have been activated and is | |
1319 | * contributing its current weight to the parent's child_weight. | |
1320 | */ | |
e15693ef | 1321 | cfq_update_group_leaf_weight(cfqg); |
8184f93e | 1322 | __cfq_group_service_tree_add(st, cfqg); |
7918ffb5 TH |
1323 | |
1324 | /* | |
1d3650f7 TH |
1325 | * Activate @cfqg and calculate the portion of vfraction @cfqg is |
1326 | * entitled to. vfraction is calculated by walking the tree | |
1327 | * towards the root calculating the fraction it has at each level. | |
1328 | * The compounded ratio is how much vfraction @cfqg owns. | |
1329 | * | |
1330 | * Start with the proportion tasks in this cfqg has against active | |
1331 | * children cfqgs - its leaf_weight against children_weight. | |
7918ffb5 TH |
1332 | */ |
1333 | propagate = !pos->nr_active++; | |
1334 | pos->children_weight += pos->leaf_weight; | |
1d3650f7 | 1335 | vfr = vfr * pos->leaf_weight / pos->children_weight; |
7918ffb5 | 1336 | |
1d3650f7 TH |
1337 | /* |
1338 | * Compound ->weight walking up the tree. Both activation and | |
1339 | * vfraction calculation are done in the same loop. Propagation | |
1340 | * stops once an already activated node is met. vfraction | |
1341 | * calculation should always continue to the root. | |
1342 | */ | |
d02f7aa8 | 1343 | while ((parent = cfqg_parent(pos))) { |
1d3650f7 | 1344 | if (propagate) { |
e15693ef | 1345 | cfq_update_group_weight(pos); |
1d3650f7 TH |
1346 | propagate = !parent->nr_active++; |
1347 | parent->children_weight += pos->weight; | |
1348 | } | |
1349 | vfr = vfr * pos->weight / parent->children_weight; | |
7918ffb5 TH |
1350 | pos = parent; |
1351 | } | |
1d3650f7 TH |
1352 | |
1353 | cfqg->vfraction = max_t(unsigned, vfr, 1); | |
8184f93e JT |
1354 | } |
1355 | ||
1356 | static void | |
1357 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
1358 | { |
1359 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1360 | struct cfq_group *__cfqg; | |
1361 | struct rb_node *n; | |
1362 | ||
1363 | cfqg->nr_cfqq++; | |
760701bf | 1364 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
1365 | return; |
1366 | ||
1367 | /* | |
1368 | * Currently put the group at the end. Later implement something | |
1369 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 1370 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
1371 | */ |
1372 | n = rb_last(&st->rb); | |
1373 | if (n) { | |
1374 | __cfqg = rb_entry_cfqg(n); | |
1375 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
1376 | } else | |
1377 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
1378 | cfq_group_service_tree_add(st, cfqg); |
1379 | } | |
1fa8f6d6 | 1380 | |
8184f93e JT |
1381 | static void |
1382 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1383 | { | |
7918ffb5 TH |
1384 | struct cfq_group *pos = cfqg; |
1385 | bool propagate; | |
1386 | ||
1387 | /* | |
1388 | * Undo activation from cfq_group_service_tree_add(). Deactivate | |
1389 | * @cfqg and propagate deactivation upwards. | |
1390 | */ | |
1391 | propagate = !--pos->nr_active; | |
1392 | pos->children_weight -= pos->leaf_weight; | |
1393 | ||
1394 | while (propagate) { | |
d02f7aa8 | 1395 | struct cfq_group *parent = cfqg_parent(pos); |
7918ffb5 TH |
1396 | |
1397 | /* @pos has 0 nr_active at this point */ | |
1398 | WARN_ON_ONCE(pos->children_weight); | |
1d3650f7 | 1399 | pos->vfraction = 0; |
7918ffb5 TH |
1400 | |
1401 | if (!parent) | |
1402 | break; | |
1403 | ||
1404 | propagate = !--parent->nr_active; | |
1405 | parent->children_weight -= pos->weight; | |
1406 | pos = parent; | |
1407 | } | |
1408 | ||
1409 | /* remove from the service tree */ | |
8184f93e JT |
1410 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1411 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
1412 | } |
1413 | ||
1414 | static void | |
8184f93e | 1415 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
1416 | { |
1417 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1418 | ||
1419 | BUG_ON(cfqg->nr_cfqq < 1); | |
1420 | cfqg->nr_cfqq--; | |
25bc6b07 | 1421 | |
1fa8f6d6 VG |
1422 | /* If there are other cfq queues under this group, don't delete it */ |
1423 | if (cfqg->nr_cfqq) | |
1424 | return; | |
1425 | ||
2868ef7b | 1426 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 1427 | cfq_group_service_tree_del(st, cfqg); |
4d2ceea4 | 1428 | cfqg->saved_wl_slice = 0; |
155fead9 | 1429 | cfqg_stats_update_dequeue(cfqg); |
dae739eb VG |
1430 | } |
1431 | ||
9a7f38c4 JM |
1432 | static inline u64 cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
1433 | u64 *unaccounted_time) | |
dae739eb | 1434 | { |
9a7f38c4 JM |
1435 | u64 slice_used; |
1436 | u64 now = ktime_get_ns(); | |
dae739eb VG |
1437 | |
1438 | /* | |
1439 | * Queue got expired before even a single request completed or | |
1440 | * got expired immediately after first request completion. | |
1441 | */ | |
9a7f38c4 | 1442 | if (!cfqq->slice_start || cfqq->slice_start == now) { |
dae739eb VG |
1443 | /* |
1444 | * Also charge the seek time incurred to the group, otherwise | |
1445 | * if there are mutiple queues in the group, each can dispatch | |
1446 | * a single request on seeky media and cause lots of seek time | |
1447 | * and group will never know it. | |
1448 | */ | |
0b31c10c JK |
1449 | slice_used = max_t(u64, (now - cfqq->dispatch_start), |
1450 | jiffies_to_nsecs(1)); | |
dae739eb | 1451 | } else { |
9a7f38c4 | 1452 | slice_used = now - cfqq->slice_start; |
167400d3 JT |
1453 | if (slice_used > cfqq->allocated_slice) { |
1454 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 1455 | slice_used = cfqq->allocated_slice; |
167400d3 | 1456 | } |
9a7f38c4 | 1457 | if (cfqq->slice_start > cfqq->dispatch_start) |
167400d3 JT |
1458 | *unaccounted_time += cfqq->slice_start - |
1459 | cfqq->dispatch_start; | |
dae739eb VG |
1460 | } |
1461 | ||
dae739eb VG |
1462 | return slice_used; |
1463 | } | |
1464 | ||
1465 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1466 | struct cfq_queue *cfqq) |
dae739eb VG |
1467 | { |
1468 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
9a7f38c4 | 1469 | u64 used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1470 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1471 | - cfqg->service_tree_idle.count; | |
1d3650f7 | 1472 | unsigned int vfr; |
9a7f38c4 | 1473 | u64 now = ktime_get_ns(); |
f26bd1f0 VG |
1474 | |
1475 | BUG_ON(nr_sync < 0); | |
167400d3 | 1476 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1477 | |
02b35081 VG |
1478 | if (iops_mode(cfqd)) |
1479 | charge = cfqq->slice_dispatch; | |
1480 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1481 | charge = cfqq->allocated_slice; | |
dae739eb | 1482 | |
1d3650f7 TH |
1483 | /* |
1484 | * Can't update vdisktime while on service tree and cfqg->vfraction | |
1485 | * is valid only while on it. Cache vfr, leave the service tree, | |
1486 | * update vdisktime and go back on. The re-addition to the tree | |
1487 | * will also update the weights as necessary. | |
1488 | */ | |
1489 | vfr = cfqg->vfraction; | |
8184f93e | 1490 | cfq_group_service_tree_del(st, cfqg); |
1d3650f7 | 1491 | cfqg->vdisktime += cfqg_scale_charge(charge, vfr); |
8184f93e | 1492 | cfq_group_service_tree_add(st, cfqg); |
dae739eb VG |
1493 | |
1494 | /* This group is being expired. Save the context */ | |
9a7f38c4 JM |
1495 | if (cfqd->workload_expires > now) { |
1496 | cfqg->saved_wl_slice = cfqd->workload_expires - now; | |
4d2ceea4 VG |
1497 | cfqg->saved_wl_type = cfqd->serving_wl_type; |
1498 | cfqg->saved_wl_class = cfqd->serving_wl_class; | |
dae739eb | 1499 | } else |
4d2ceea4 | 1500 | cfqg->saved_wl_slice = 0; |
2868ef7b VG |
1501 | |
1502 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1503 | st->min_vdisktime); | |
fd16d263 | 1504 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
9a7f38c4 | 1505 | "sl_used=%llu disp=%llu charge=%llu iops=%u sect=%lu", |
fd16d263 JP |
1506 | used_sl, cfqq->slice_dispatch, charge, |
1507 | iops_mode(cfqd), cfqq->nr_sectors); | |
155fead9 TH |
1508 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
1509 | cfqg_stats_set_start_empty_time(cfqg); | |
1fa8f6d6 VG |
1510 | } |
1511 | ||
f51b802c TH |
1512 | /** |
1513 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1514 | * @cfqg: cfq_group to initialize | |
1515 | * | |
1516 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1517 | * is enabled or not. | |
1518 | */ | |
1519 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1520 | { | |
1521 | struct cfq_rb_root *st; | |
1522 | int i, j; | |
1523 | ||
1524 | for_each_cfqg_st(cfqg, i, j, st) | |
1525 | *st = CFQ_RB_ROOT; | |
1526 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1527 | ||
9a7f38c4 | 1528 | cfqg->ttime.last_end_request = ktime_get_ns(); |
f51b802c TH |
1529 | } |
1530 | ||
25fb5169 | 1531 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
69d7fde5 TH |
1532 | static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val, |
1533 | bool on_dfl, bool reset_dev, bool is_leaf_weight); | |
1534 | ||
24bdb8ef | 1535 | static void cfqg_stats_exit(struct cfqg_stats *stats) |
90d3839b | 1536 | { |
24bdb8ef TH |
1537 | blkg_rwstat_exit(&stats->merged); |
1538 | blkg_rwstat_exit(&stats->service_time); | |
1539 | blkg_rwstat_exit(&stats->wait_time); | |
1540 | blkg_rwstat_exit(&stats->queued); | |
24bdb8ef TH |
1541 | blkg_stat_exit(&stats->time); |
1542 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
1543 | blkg_stat_exit(&stats->unaccounted_time); | |
1544 | blkg_stat_exit(&stats->avg_queue_size_sum); | |
1545 | blkg_stat_exit(&stats->avg_queue_size_samples); | |
1546 | blkg_stat_exit(&stats->dequeue); | |
1547 | blkg_stat_exit(&stats->group_wait_time); | |
1548 | blkg_stat_exit(&stats->idle_time); | |
1549 | blkg_stat_exit(&stats->empty_time); | |
1550 | #endif | |
1551 | } | |
1552 | ||
1553 | static int cfqg_stats_init(struct cfqg_stats *stats, gfp_t gfp) | |
1554 | { | |
77ea7338 | 1555 | if (blkg_rwstat_init(&stats->merged, gfp) || |
24bdb8ef TH |
1556 | blkg_rwstat_init(&stats->service_time, gfp) || |
1557 | blkg_rwstat_init(&stats->wait_time, gfp) || | |
1558 | blkg_rwstat_init(&stats->queued, gfp) || | |
24bdb8ef TH |
1559 | blkg_stat_init(&stats->time, gfp)) |
1560 | goto err; | |
90d3839b PZ |
1561 | |
1562 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
24bdb8ef TH |
1563 | if (blkg_stat_init(&stats->unaccounted_time, gfp) || |
1564 | blkg_stat_init(&stats->avg_queue_size_sum, gfp) || | |
1565 | blkg_stat_init(&stats->avg_queue_size_samples, gfp) || | |
1566 | blkg_stat_init(&stats->dequeue, gfp) || | |
1567 | blkg_stat_init(&stats->group_wait_time, gfp) || | |
1568 | blkg_stat_init(&stats->idle_time, gfp) || | |
1569 | blkg_stat_init(&stats->empty_time, gfp)) | |
1570 | goto err; | |
90d3839b | 1571 | #endif |
24bdb8ef TH |
1572 | return 0; |
1573 | err: | |
1574 | cfqg_stats_exit(stats); | |
1575 | return -ENOMEM; | |
90d3839b PZ |
1576 | } |
1577 | ||
e4a9bde9 TH |
1578 | static struct blkcg_policy_data *cfq_cpd_alloc(gfp_t gfp) |
1579 | { | |
1580 | struct cfq_group_data *cgd; | |
1581 | ||
ebc4ff66 | 1582 | cgd = kzalloc(sizeof(*cgd), gfp); |
e4a9bde9 TH |
1583 | if (!cgd) |
1584 | return NULL; | |
1585 | return &cgd->cpd; | |
1586 | } | |
1587 | ||
81437648 | 1588 | static void cfq_cpd_init(struct blkcg_policy_data *cpd) |
e48453c3 | 1589 | { |
81437648 | 1590 | struct cfq_group_data *cgd = cpd_to_cfqgd(cpd); |
9e10a130 | 1591 | unsigned int weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ? |
69d7fde5 | 1592 | CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL; |
e48453c3 | 1593 | |
69d7fde5 TH |
1594 | if (cpd_to_blkcg(cpd) == &blkcg_root) |
1595 | weight *= 2; | |
1596 | ||
1597 | cgd->weight = weight; | |
1598 | cgd->leaf_weight = weight; | |
e48453c3 AA |
1599 | } |
1600 | ||
e4a9bde9 TH |
1601 | static void cfq_cpd_free(struct blkcg_policy_data *cpd) |
1602 | { | |
1603 | kfree(cpd_to_cfqgd(cpd)); | |
1604 | } | |
1605 | ||
69d7fde5 TH |
1606 | static void cfq_cpd_bind(struct blkcg_policy_data *cpd) |
1607 | { | |
1608 | struct blkcg *blkcg = cpd_to_blkcg(cpd); | |
9e10a130 | 1609 | bool on_dfl = cgroup_subsys_on_dfl(io_cgrp_subsys); |
69d7fde5 TH |
1610 | unsigned int weight = on_dfl ? CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL; |
1611 | ||
1612 | if (blkcg == &blkcg_root) | |
1613 | weight *= 2; | |
1614 | ||
1615 | WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, false)); | |
1616 | WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, true)); | |
1617 | } | |
1618 | ||
001bea73 TH |
1619 | static struct blkg_policy_data *cfq_pd_alloc(gfp_t gfp, int node) |
1620 | { | |
b2ce2643 TH |
1621 | struct cfq_group *cfqg; |
1622 | ||
1623 | cfqg = kzalloc_node(sizeof(*cfqg), gfp, node); | |
1624 | if (!cfqg) | |
1625 | return NULL; | |
1626 | ||
1627 | cfq_init_cfqg_base(cfqg); | |
24bdb8ef TH |
1628 | if (cfqg_stats_init(&cfqg->stats, gfp)) { |
1629 | kfree(cfqg); | |
1630 | return NULL; | |
1631 | } | |
b2ce2643 TH |
1632 | |
1633 | return &cfqg->pd; | |
001bea73 TH |
1634 | } |
1635 | ||
a9520cd6 | 1636 | static void cfq_pd_init(struct blkg_policy_data *pd) |
f469a7b4 | 1637 | { |
a9520cd6 TH |
1638 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
1639 | struct cfq_group_data *cgd = blkcg_to_cfqgd(pd->blkg->blkcg); | |
25fb5169 | 1640 | |
e48453c3 AA |
1641 | cfqg->weight = cgd->weight; |
1642 | cfqg->leaf_weight = cgd->leaf_weight; | |
25fb5169 VG |
1643 | } |
1644 | ||
a9520cd6 | 1645 | static void cfq_pd_offline(struct blkg_policy_data *pd) |
0b39920b | 1646 | { |
a9520cd6 | 1647 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
60a83707 TH |
1648 | int i; |
1649 | ||
1650 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
1651 | if (cfqg->async_cfqq[0][i]) | |
1652 | cfq_put_queue(cfqg->async_cfqq[0][i]); | |
1653 | if (cfqg->async_cfqq[1][i]) | |
1654 | cfq_put_queue(cfqg->async_cfqq[1][i]); | |
1655 | } | |
1656 | ||
1657 | if (cfqg->async_idle_cfqq) | |
1658 | cfq_put_queue(cfqg->async_idle_cfqq); | |
1659 | ||
0b39920b TH |
1660 | /* |
1661 | * @blkg is going offline and will be ignored by | |
1662 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
1663 | * that they don't get lost. If IOs complete after this point, the | |
1664 | * stats for them will be lost. Oh well... | |
1665 | */ | |
60a83707 | 1666 | cfqg_stats_xfer_dead(cfqg); |
0b39920b TH |
1667 | } |
1668 | ||
001bea73 TH |
1669 | static void cfq_pd_free(struct blkg_policy_data *pd) |
1670 | { | |
24bdb8ef TH |
1671 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
1672 | ||
1673 | cfqg_stats_exit(&cfqg->stats); | |
1674 | return kfree(cfqg); | |
001bea73 TH |
1675 | } |
1676 | ||
a9520cd6 | 1677 | static void cfq_pd_reset_stats(struct blkg_policy_data *pd) |
689665af | 1678 | { |
a9520cd6 | 1679 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
689665af TH |
1680 | |
1681 | cfqg_stats_reset(&cfqg->stats); | |
25fb5169 VG |
1682 | } |
1683 | ||
ae118896 TH |
1684 | static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd, |
1685 | struct blkcg *blkcg) | |
25fb5169 | 1686 | { |
ae118896 | 1687 | struct blkcg_gq *blkg; |
f469a7b4 | 1688 | |
ae118896 TH |
1689 | blkg = blkg_lookup(blkcg, cfqd->queue); |
1690 | if (likely(blkg)) | |
1691 | return blkg_to_cfqg(blkg); | |
1692 | return NULL; | |
25fb5169 VG |
1693 | } |
1694 | ||
1695 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1696 | { | |
25fb5169 | 1697 | cfqq->cfqg = cfqg; |
b1c35769 | 1698 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1699 | cfqg_get(cfqg); |
b1c35769 VG |
1700 | } |
1701 | ||
f95a04af TH |
1702 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, |
1703 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1704 | { |
f95a04af | 1705 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
3381cb8d TH |
1706 | |
1707 | if (!cfqg->dev_weight) | |
60c2bc2d | 1708 | return 0; |
f95a04af | 1709 | return __blkg_prfill_u64(sf, pd, cfqg->dev_weight); |
60c2bc2d TH |
1710 | } |
1711 | ||
2da8ca82 | 1712 | static int cfqg_print_weight_device(struct seq_file *sf, void *v) |
60c2bc2d | 1713 | { |
2da8ca82 TH |
1714 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1715 | cfqg_prfill_weight_device, &blkcg_policy_cfq, | |
1716 | 0, false); | |
60c2bc2d TH |
1717 | return 0; |
1718 | } | |
1719 | ||
e71357e1 TH |
1720 | static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf, |
1721 | struct blkg_policy_data *pd, int off) | |
1722 | { | |
1723 | struct cfq_group *cfqg = pd_to_cfqg(pd); | |
1724 | ||
1725 | if (!cfqg->dev_leaf_weight) | |
1726 | return 0; | |
1727 | return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight); | |
1728 | } | |
1729 | ||
2da8ca82 | 1730 | static int cfqg_print_leaf_weight_device(struct seq_file *sf, void *v) |
e71357e1 | 1731 | { |
2da8ca82 TH |
1732 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1733 | cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq, | |
1734 | 0, false); | |
e71357e1 TH |
1735 | return 0; |
1736 | } | |
1737 | ||
2da8ca82 | 1738 | static int cfq_print_weight(struct seq_file *sf, void *v) |
60c2bc2d | 1739 | { |
e48453c3 | 1740 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
9470e4a6 JA |
1741 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); |
1742 | unsigned int val = 0; | |
e48453c3 | 1743 | |
9470e4a6 JA |
1744 | if (cgd) |
1745 | val = cgd->weight; | |
1746 | ||
1747 | seq_printf(sf, "%u\n", val); | |
60c2bc2d TH |
1748 | return 0; |
1749 | } | |
1750 | ||
2da8ca82 | 1751 | static int cfq_print_leaf_weight(struct seq_file *sf, void *v) |
e71357e1 | 1752 | { |
e48453c3 | 1753 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
9470e4a6 JA |
1754 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); |
1755 | unsigned int val = 0; | |
1756 | ||
1757 | if (cgd) | |
1758 | val = cgd->leaf_weight; | |
e48453c3 | 1759 | |
9470e4a6 | 1760 | seq_printf(sf, "%u\n", val); |
e71357e1 TH |
1761 | return 0; |
1762 | } | |
1763 | ||
451af504 TH |
1764 | static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of, |
1765 | char *buf, size_t nbytes, loff_t off, | |
2ee867dc | 1766 | bool on_dfl, bool is_leaf_weight) |
60c2bc2d | 1767 | { |
69d7fde5 TH |
1768 | unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN; |
1769 | unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX; | |
451af504 | 1770 | struct blkcg *blkcg = css_to_blkcg(of_css(of)); |
60c2bc2d | 1771 | struct blkg_conf_ctx ctx; |
3381cb8d | 1772 | struct cfq_group *cfqg; |
e48453c3 | 1773 | struct cfq_group_data *cfqgd; |
60c2bc2d | 1774 | int ret; |
36aa9e5f | 1775 | u64 v; |
60c2bc2d | 1776 | |
3c798398 | 1777 | ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx); |
60c2bc2d TH |
1778 | if (ret) |
1779 | return ret; | |
1780 | ||
2ee867dc TH |
1781 | if (sscanf(ctx.body, "%llu", &v) == 1) { |
1782 | /* require "default" on dfl */ | |
1783 | ret = -ERANGE; | |
1784 | if (!v && on_dfl) | |
1785 | goto out_finish; | |
1786 | } else if (!strcmp(strim(ctx.body), "default")) { | |
1787 | v = 0; | |
1788 | } else { | |
1789 | ret = -EINVAL; | |
36aa9e5f | 1790 | goto out_finish; |
2ee867dc | 1791 | } |
36aa9e5f | 1792 | |
3381cb8d | 1793 | cfqg = blkg_to_cfqg(ctx.blkg); |
e48453c3 | 1794 | cfqgd = blkcg_to_cfqgd(blkcg); |
ae994ea9 | 1795 | |
20386ce0 | 1796 | ret = -ERANGE; |
69d7fde5 | 1797 | if (!v || (v >= min && v <= max)) { |
e71357e1 | 1798 | if (!is_leaf_weight) { |
36aa9e5f TH |
1799 | cfqg->dev_weight = v; |
1800 | cfqg->new_weight = v ?: cfqgd->weight; | |
e71357e1 | 1801 | } else { |
36aa9e5f TH |
1802 | cfqg->dev_leaf_weight = v; |
1803 | cfqg->new_leaf_weight = v ?: cfqgd->leaf_weight; | |
e71357e1 | 1804 | } |
60c2bc2d TH |
1805 | ret = 0; |
1806 | } | |
36aa9e5f | 1807 | out_finish: |
60c2bc2d | 1808 | blkg_conf_finish(&ctx); |
451af504 | 1809 | return ret ?: nbytes; |
60c2bc2d TH |
1810 | } |
1811 | ||
451af504 TH |
1812 | static ssize_t cfqg_set_weight_device(struct kernfs_open_file *of, |
1813 | char *buf, size_t nbytes, loff_t off) | |
e71357e1 | 1814 | { |
2ee867dc | 1815 | return __cfqg_set_weight_device(of, buf, nbytes, off, false, false); |
e71357e1 TH |
1816 | } |
1817 | ||
451af504 TH |
1818 | static ssize_t cfqg_set_leaf_weight_device(struct kernfs_open_file *of, |
1819 | char *buf, size_t nbytes, loff_t off) | |
e71357e1 | 1820 | { |
2ee867dc | 1821 | return __cfqg_set_weight_device(of, buf, nbytes, off, false, true); |
e71357e1 TH |
1822 | } |
1823 | ||
dd165eb3 | 1824 | static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val, |
69d7fde5 | 1825 | bool on_dfl, bool reset_dev, bool is_leaf_weight) |
60c2bc2d | 1826 | { |
69d7fde5 TH |
1827 | unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN; |
1828 | unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX; | |
182446d0 | 1829 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 1830 | struct blkcg_gq *blkg; |
e48453c3 | 1831 | struct cfq_group_data *cfqgd; |
ae994ea9 | 1832 | int ret = 0; |
60c2bc2d | 1833 | |
69d7fde5 TH |
1834 | if (val < min || val > max) |
1835 | return -ERANGE; | |
60c2bc2d TH |
1836 | |
1837 | spin_lock_irq(&blkcg->lock); | |
e48453c3 | 1838 | cfqgd = blkcg_to_cfqgd(blkcg); |
ae994ea9 JA |
1839 | if (!cfqgd) { |
1840 | ret = -EINVAL; | |
1841 | goto out; | |
1842 | } | |
e71357e1 TH |
1843 | |
1844 | if (!is_leaf_weight) | |
e48453c3 | 1845 | cfqgd->weight = val; |
e71357e1 | 1846 | else |
e48453c3 | 1847 | cfqgd->leaf_weight = val; |
60c2bc2d | 1848 | |
b67bfe0d | 1849 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
3381cb8d | 1850 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
60c2bc2d | 1851 | |
e71357e1 TH |
1852 | if (!cfqg) |
1853 | continue; | |
1854 | ||
1855 | if (!is_leaf_weight) { | |
69d7fde5 TH |
1856 | if (reset_dev) |
1857 | cfqg->dev_weight = 0; | |
e71357e1 | 1858 | if (!cfqg->dev_weight) |
e48453c3 | 1859 | cfqg->new_weight = cfqgd->weight; |
e71357e1 | 1860 | } else { |
69d7fde5 TH |
1861 | if (reset_dev) |
1862 | cfqg->dev_leaf_weight = 0; | |
e71357e1 | 1863 | if (!cfqg->dev_leaf_weight) |
e48453c3 | 1864 | cfqg->new_leaf_weight = cfqgd->leaf_weight; |
e71357e1 | 1865 | } |
60c2bc2d TH |
1866 | } |
1867 | ||
ae994ea9 | 1868 | out: |
60c2bc2d | 1869 | spin_unlock_irq(&blkcg->lock); |
ae994ea9 | 1870 | return ret; |
60c2bc2d TH |
1871 | } |
1872 | ||
182446d0 TH |
1873 | static int cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft, |
1874 | u64 val) | |
e71357e1 | 1875 | { |
69d7fde5 | 1876 | return __cfq_set_weight(css, val, false, false, false); |
e71357e1 TH |
1877 | } |
1878 | ||
182446d0 TH |
1879 | static int cfq_set_leaf_weight(struct cgroup_subsys_state *css, |
1880 | struct cftype *cft, u64 val) | |
e71357e1 | 1881 | { |
69d7fde5 | 1882 | return __cfq_set_weight(css, val, false, false, true); |
e71357e1 TH |
1883 | } |
1884 | ||
2da8ca82 | 1885 | static int cfqg_print_stat(struct seq_file *sf, void *v) |
5bc4afb1 | 1886 | { |
2da8ca82 TH |
1887 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat, |
1888 | &blkcg_policy_cfq, seq_cft(sf)->private, false); | |
5bc4afb1 TH |
1889 | return 0; |
1890 | } | |
1891 | ||
2da8ca82 | 1892 | static int cfqg_print_rwstat(struct seq_file *sf, void *v) |
5bc4afb1 | 1893 | { |
2da8ca82 TH |
1894 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat, |
1895 | &blkcg_policy_cfq, seq_cft(sf)->private, true); | |
5bc4afb1 TH |
1896 | return 0; |
1897 | } | |
1898 | ||
43114018 TH |
1899 | static u64 cfqg_prfill_stat_recursive(struct seq_file *sf, |
1900 | struct blkg_policy_data *pd, int off) | |
1901 | { | |
f12c74ca TH |
1902 | u64 sum = blkg_stat_recursive_sum(pd_to_blkg(pd), |
1903 | &blkcg_policy_cfq, off); | |
43114018 TH |
1904 | return __blkg_prfill_u64(sf, pd, sum); |
1905 | } | |
1906 | ||
1907 | static u64 cfqg_prfill_rwstat_recursive(struct seq_file *sf, | |
1908 | struct blkg_policy_data *pd, int off) | |
1909 | { | |
f12c74ca TH |
1910 | struct blkg_rwstat sum = blkg_rwstat_recursive_sum(pd_to_blkg(pd), |
1911 | &blkcg_policy_cfq, off); | |
43114018 TH |
1912 | return __blkg_prfill_rwstat(sf, pd, &sum); |
1913 | } | |
1914 | ||
2da8ca82 | 1915 | static int cfqg_print_stat_recursive(struct seq_file *sf, void *v) |
43114018 | 1916 | { |
2da8ca82 TH |
1917 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1918 | cfqg_prfill_stat_recursive, &blkcg_policy_cfq, | |
1919 | seq_cft(sf)->private, false); | |
43114018 TH |
1920 | return 0; |
1921 | } | |
1922 | ||
2da8ca82 | 1923 | static int cfqg_print_rwstat_recursive(struct seq_file *sf, void *v) |
43114018 | 1924 | { |
2da8ca82 TH |
1925 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1926 | cfqg_prfill_rwstat_recursive, &blkcg_policy_cfq, | |
1927 | seq_cft(sf)->private, true); | |
43114018 TH |
1928 | return 0; |
1929 | } | |
1930 | ||
702747ca TH |
1931 | static u64 cfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd, |
1932 | int off) | |
1933 | { | |
1934 | u64 sum = blkg_rwstat_total(&pd->blkg->stat_bytes); | |
1935 | ||
1936 | return __blkg_prfill_u64(sf, pd, sum >> 9); | |
1937 | } | |
1938 | ||
1939 | static int cfqg_print_stat_sectors(struct seq_file *sf, void *v) | |
1940 | { | |
1941 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1942 | cfqg_prfill_sectors, &blkcg_policy_cfq, 0, false); | |
1943 | return 0; | |
1944 | } | |
1945 | ||
1946 | static u64 cfqg_prfill_sectors_recursive(struct seq_file *sf, | |
1947 | struct blkg_policy_data *pd, int off) | |
1948 | { | |
1949 | struct blkg_rwstat tmp = blkg_rwstat_recursive_sum(pd->blkg, NULL, | |
1950 | offsetof(struct blkcg_gq, stat_bytes)); | |
1951 | u64 sum = atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) + | |
1952 | atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]); | |
1953 | ||
1954 | return __blkg_prfill_u64(sf, pd, sum >> 9); | |
1955 | } | |
1956 | ||
1957 | static int cfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v) | |
1958 | { | |
1959 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1960 | cfqg_prfill_sectors_recursive, &blkcg_policy_cfq, 0, | |
1961 | false); | |
1962 | return 0; | |
1963 | } | |
1964 | ||
60c2bc2d | 1965 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
f95a04af TH |
1966 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, |
1967 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1968 | { |
f95a04af | 1969 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
155fead9 | 1970 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
60c2bc2d TH |
1971 | u64 v = 0; |
1972 | ||
1973 | if (samples) { | |
155fead9 | 1974 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
f3cff25f | 1975 | v = div64_u64(v, samples); |
60c2bc2d | 1976 | } |
f95a04af | 1977 | __blkg_prfill_u64(sf, pd, v); |
60c2bc2d TH |
1978 | return 0; |
1979 | } | |
1980 | ||
1981 | /* print avg_queue_size */ | |
2da8ca82 | 1982 | static int cfqg_print_avg_queue_size(struct seq_file *sf, void *v) |
60c2bc2d | 1983 | { |
2da8ca82 TH |
1984 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1985 | cfqg_prfill_avg_queue_size, &blkcg_policy_cfq, | |
1986 | 0, false); | |
60c2bc2d TH |
1987 | return 0; |
1988 | } | |
1989 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1990 | ||
880f50e2 | 1991 | static struct cftype cfq_blkcg_legacy_files[] = { |
1d3650f7 | 1992 | /* on root, weight is mapped to leaf_weight */ |
60c2bc2d TH |
1993 | { |
1994 | .name = "weight_device", | |
1d3650f7 | 1995 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 1996 | .seq_show = cfqg_print_leaf_weight_device, |
451af504 | 1997 | .write = cfqg_set_leaf_weight_device, |
60c2bc2d TH |
1998 | }, |
1999 | { | |
2000 | .name = "weight", | |
1d3650f7 | 2001 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 2002 | .seq_show = cfq_print_leaf_weight, |
1d3650f7 | 2003 | .write_u64 = cfq_set_leaf_weight, |
60c2bc2d | 2004 | }, |
e71357e1 | 2005 | |
1d3650f7 | 2006 | /* no such mapping necessary for !roots */ |
60c2bc2d TH |
2007 | { |
2008 | .name = "weight_device", | |
1d3650f7 | 2009 | .flags = CFTYPE_NOT_ON_ROOT, |
2da8ca82 | 2010 | .seq_show = cfqg_print_weight_device, |
451af504 | 2011 | .write = cfqg_set_weight_device, |
60c2bc2d TH |
2012 | }, |
2013 | { | |
2014 | .name = "weight", | |
1d3650f7 | 2015 | .flags = CFTYPE_NOT_ON_ROOT, |
2da8ca82 | 2016 | .seq_show = cfq_print_weight, |
3381cb8d | 2017 | .write_u64 = cfq_set_weight, |
60c2bc2d | 2018 | }, |
e71357e1 | 2019 | |
e71357e1 TH |
2020 | { |
2021 | .name = "leaf_weight_device", | |
2da8ca82 | 2022 | .seq_show = cfqg_print_leaf_weight_device, |
451af504 | 2023 | .write = cfqg_set_leaf_weight_device, |
e71357e1 TH |
2024 | }, |
2025 | { | |
2026 | .name = "leaf_weight", | |
2da8ca82 | 2027 | .seq_show = cfq_print_leaf_weight, |
e71357e1 TH |
2028 | .write_u64 = cfq_set_leaf_weight, |
2029 | }, | |
2030 | ||
43114018 | 2031 | /* statistics, covers only the tasks in the cfqg */ |
60c2bc2d TH |
2032 | { |
2033 | .name = "time", | |
5bc4afb1 | 2034 | .private = offsetof(struct cfq_group, stats.time), |
2da8ca82 | 2035 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2036 | }, |
2037 | { | |
2038 | .name = "sectors", | |
702747ca | 2039 | .seq_show = cfqg_print_stat_sectors, |
60c2bc2d TH |
2040 | }, |
2041 | { | |
2042 | .name = "io_service_bytes", | |
77ea7338 TH |
2043 | .private = (unsigned long)&blkcg_policy_cfq, |
2044 | .seq_show = blkg_print_stat_bytes, | |
60c2bc2d TH |
2045 | }, |
2046 | { | |
2047 | .name = "io_serviced", | |
77ea7338 TH |
2048 | .private = (unsigned long)&blkcg_policy_cfq, |
2049 | .seq_show = blkg_print_stat_ios, | |
60c2bc2d TH |
2050 | }, |
2051 | { | |
2052 | .name = "io_service_time", | |
5bc4afb1 | 2053 | .private = offsetof(struct cfq_group, stats.service_time), |
2da8ca82 | 2054 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2055 | }, |
2056 | { | |
2057 | .name = "io_wait_time", | |
5bc4afb1 | 2058 | .private = offsetof(struct cfq_group, stats.wait_time), |
2da8ca82 | 2059 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2060 | }, |
2061 | { | |
2062 | .name = "io_merged", | |
5bc4afb1 | 2063 | .private = offsetof(struct cfq_group, stats.merged), |
2da8ca82 | 2064 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2065 | }, |
2066 | { | |
2067 | .name = "io_queued", | |
5bc4afb1 | 2068 | .private = offsetof(struct cfq_group, stats.queued), |
2da8ca82 | 2069 | .seq_show = cfqg_print_rwstat, |
60c2bc2d | 2070 | }, |
43114018 TH |
2071 | |
2072 | /* the same statictics which cover the cfqg and its descendants */ | |
2073 | { | |
2074 | .name = "time_recursive", | |
2075 | .private = offsetof(struct cfq_group, stats.time), | |
2da8ca82 | 2076 | .seq_show = cfqg_print_stat_recursive, |
43114018 TH |
2077 | }, |
2078 | { | |
2079 | .name = "sectors_recursive", | |
702747ca | 2080 | .seq_show = cfqg_print_stat_sectors_recursive, |
43114018 TH |
2081 | }, |
2082 | { | |
2083 | .name = "io_service_bytes_recursive", | |
77ea7338 TH |
2084 | .private = (unsigned long)&blkcg_policy_cfq, |
2085 | .seq_show = blkg_print_stat_bytes_recursive, | |
43114018 TH |
2086 | }, |
2087 | { | |
2088 | .name = "io_serviced_recursive", | |
77ea7338 TH |
2089 | .private = (unsigned long)&blkcg_policy_cfq, |
2090 | .seq_show = blkg_print_stat_ios_recursive, | |
43114018 TH |
2091 | }, |
2092 | { | |
2093 | .name = "io_service_time_recursive", | |
2094 | .private = offsetof(struct cfq_group, stats.service_time), | |
2da8ca82 | 2095 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2096 | }, |
2097 | { | |
2098 | .name = "io_wait_time_recursive", | |
2099 | .private = offsetof(struct cfq_group, stats.wait_time), | |
2da8ca82 | 2100 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2101 | }, |
2102 | { | |
2103 | .name = "io_merged_recursive", | |
2104 | .private = offsetof(struct cfq_group, stats.merged), | |
2da8ca82 | 2105 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2106 | }, |
2107 | { | |
2108 | .name = "io_queued_recursive", | |
2109 | .private = offsetof(struct cfq_group, stats.queued), | |
2da8ca82 | 2110 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 | 2111 | }, |
60c2bc2d TH |
2112 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
2113 | { | |
2114 | .name = "avg_queue_size", | |
2da8ca82 | 2115 | .seq_show = cfqg_print_avg_queue_size, |
60c2bc2d TH |
2116 | }, |
2117 | { | |
2118 | .name = "group_wait_time", | |
5bc4afb1 | 2119 | .private = offsetof(struct cfq_group, stats.group_wait_time), |
2da8ca82 | 2120 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2121 | }, |
2122 | { | |
2123 | .name = "idle_time", | |
5bc4afb1 | 2124 | .private = offsetof(struct cfq_group, stats.idle_time), |
2da8ca82 | 2125 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2126 | }, |
2127 | { | |
2128 | .name = "empty_time", | |
5bc4afb1 | 2129 | .private = offsetof(struct cfq_group, stats.empty_time), |
2da8ca82 | 2130 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2131 | }, |
2132 | { | |
2133 | .name = "dequeue", | |
5bc4afb1 | 2134 | .private = offsetof(struct cfq_group, stats.dequeue), |
2da8ca82 | 2135 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2136 | }, |
2137 | { | |
2138 | .name = "unaccounted_time", | |
5bc4afb1 | 2139 | .private = offsetof(struct cfq_group, stats.unaccounted_time), |
2da8ca82 | 2140 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2141 | }, |
2142 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
2143 | { } /* terminate */ | |
2144 | }; | |
2ee867dc TH |
2145 | |
2146 | static int cfq_print_weight_on_dfl(struct seq_file *sf, void *v) | |
2147 | { | |
2148 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
2149 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); | |
2150 | ||
2151 | seq_printf(sf, "default %u\n", cgd->weight); | |
2152 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_weight_device, | |
2153 | &blkcg_policy_cfq, 0, false); | |
2154 | return 0; | |
2155 | } | |
2156 | ||
2157 | static ssize_t cfq_set_weight_on_dfl(struct kernfs_open_file *of, | |
2158 | char *buf, size_t nbytes, loff_t off) | |
2159 | { | |
2160 | char *endp; | |
2161 | int ret; | |
2162 | u64 v; | |
2163 | ||
2164 | buf = strim(buf); | |
2165 | ||
2166 | /* "WEIGHT" or "default WEIGHT" sets the default weight */ | |
2167 | v = simple_strtoull(buf, &endp, 0); | |
2168 | if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) { | |
69d7fde5 | 2169 | ret = __cfq_set_weight(of_css(of), v, true, false, false); |
2ee867dc TH |
2170 | return ret ?: nbytes; |
2171 | } | |
2172 | ||
2173 | /* "MAJ:MIN WEIGHT" */ | |
2174 | return __cfqg_set_weight_device(of, buf, nbytes, off, true, false); | |
2175 | } | |
2176 | ||
2177 | static struct cftype cfq_blkcg_files[] = { | |
2178 | { | |
2179 | .name = "weight", | |
2180 | .flags = CFTYPE_NOT_ON_ROOT, | |
2181 | .seq_show = cfq_print_weight_on_dfl, | |
2182 | .write = cfq_set_weight_on_dfl, | |
2183 | }, | |
2184 | { } /* terminate */ | |
2185 | }; | |
2186 | ||
25fb5169 | 2187 | #else /* GROUP_IOSCHED */ |
ae118896 TH |
2188 | static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd, |
2189 | struct blkcg *blkcg) | |
25fb5169 | 2190 | { |
f51b802c | 2191 | return cfqd->root_group; |
25fb5169 | 2192 | } |
7f1dc8a2 | 2193 | |
25fb5169 VG |
2194 | static inline void |
2195 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
2196 | cfqq->cfqg = cfqg; | |
2197 | } | |
2198 | ||
2199 | #endif /* GROUP_IOSCHED */ | |
2200 | ||
498d3aa2 | 2201 | /* |
c0324a02 | 2202 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
2203 | * requests waiting to be processed. It is sorted in the order that |
2204 | * we will service the queues. | |
2205 | */ | |
a36e71f9 | 2206 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2207 | bool add_front) |
d9e7620e | 2208 | { |
0871714e JA |
2209 | struct rb_node **p, *parent; |
2210 | struct cfq_queue *__cfqq; | |
9a7f38c4 | 2211 | u64 rb_key; |
34b98d03 | 2212 | struct cfq_rb_root *st; |
498d3aa2 | 2213 | int left; |
dae739eb | 2214 | int new_cfqq = 1; |
9a7f38c4 | 2215 | u64 now = ktime_get_ns(); |
ae30c286 | 2216 | |
34b98d03 | 2217 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq)); |
0871714e JA |
2218 | if (cfq_class_idle(cfqq)) { |
2219 | rb_key = CFQ_IDLE_DELAY; | |
34b98d03 | 2220 | parent = rb_last(&st->rb); |
0871714e JA |
2221 | if (parent && parent != &cfqq->rb_node) { |
2222 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2223 | rb_key += __cfqq->rb_key; | |
2224 | } else | |
9a7f38c4 | 2225 | rb_key += now; |
0871714e | 2226 | } else if (!add_front) { |
b9c8946b JA |
2227 | /* |
2228 | * Get our rb key offset. Subtract any residual slice | |
2229 | * value carried from last service. A negative resid | |
2230 | * count indicates slice overrun, and this should position | |
2231 | * the next service time further away in the tree. | |
2232 | */ | |
9a7f38c4 | 2233 | rb_key = cfq_slice_offset(cfqd, cfqq) + now; |
b9c8946b | 2234 | rb_key -= cfqq->slice_resid; |
edd75ffd | 2235 | cfqq->slice_resid = 0; |
48e025e6 | 2236 | } else { |
9a7f38c4 | 2237 | rb_key = -NSEC_PER_SEC; |
34b98d03 | 2238 | __cfqq = cfq_rb_first(st); |
9a7f38c4 | 2239 | rb_key += __cfqq ? __cfqq->rb_key : now; |
48e025e6 | 2240 | } |
1da177e4 | 2241 | |
d9e7620e | 2242 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 2243 | new_cfqq = 0; |
99f9628a | 2244 | /* |
d9e7620e | 2245 | * same position, nothing more to do |
99f9628a | 2246 | */ |
34b98d03 | 2247 | if (rb_key == cfqq->rb_key && cfqq->service_tree == st) |
d9e7620e | 2248 | return; |
1da177e4 | 2249 | |
aa6f6a3d CZ |
2250 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
2251 | cfqq->service_tree = NULL; | |
1da177e4 | 2252 | } |
d9e7620e | 2253 | |
498d3aa2 | 2254 | left = 1; |
0871714e | 2255 | parent = NULL; |
34b98d03 VG |
2256 | cfqq->service_tree = st; |
2257 | p = &st->rb.rb_node; | |
d9e7620e JA |
2258 | while (*p) { |
2259 | parent = *p; | |
2260 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2261 | ||
0c534e0a | 2262 | /* |
c0324a02 | 2263 | * sort by key, that represents service time. |
0c534e0a | 2264 | */ |
9a7f38c4 | 2265 | if (rb_key < __cfqq->rb_key) |
1f23f121 | 2266 | p = &parent->rb_left; |
c0324a02 | 2267 | else { |
1f23f121 | 2268 | p = &parent->rb_right; |
cc09e299 | 2269 | left = 0; |
c0324a02 | 2270 | } |
d9e7620e JA |
2271 | } |
2272 | ||
cc09e299 | 2273 | if (left) |
34b98d03 | 2274 | st->left = &cfqq->rb_node; |
cc09e299 | 2275 | |
d9e7620e JA |
2276 | cfqq->rb_key = rb_key; |
2277 | rb_link_node(&cfqq->rb_node, parent, p); | |
34b98d03 VG |
2278 | rb_insert_color(&cfqq->rb_node, &st->rb); |
2279 | st->count++; | |
20359f27 | 2280 | if (add_front || !new_cfqq) |
dae739eb | 2281 | return; |
8184f93e | 2282 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
2283 | } |
2284 | ||
a36e71f9 | 2285 | static struct cfq_queue * |
f2d1f0ae JA |
2286 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
2287 | sector_t sector, struct rb_node **ret_parent, | |
2288 | struct rb_node ***rb_link) | |
a36e71f9 | 2289 | { |
a36e71f9 JA |
2290 | struct rb_node **p, *parent; |
2291 | struct cfq_queue *cfqq = NULL; | |
2292 | ||
2293 | parent = NULL; | |
2294 | p = &root->rb_node; | |
2295 | while (*p) { | |
2296 | struct rb_node **n; | |
2297 | ||
2298 | parent = *p; | |
2299 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
2300 | ||
2301 | /* | |
2302 | * Sort strictly based on sector. Smallest to the left, | |
2303 | * largest to the right. | |
2304 | */ | |
2e46e8b2 | 2305 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 2306 | n = &(*p)->rb_right; |
2e46e8b2 | 2307 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
2308 | n = &(*p)->rb_left; |
2309 | else | |
2310 | break; | |
2311 | p = n; | |
3ac6c9f8 | 2312 | cfqq = NULL; |
a36e71f9 JA |
2313 | } |
2314 | ||
2315 | *ret_parent = parent; | |
2316 | if (rb_link) | |
2317 | *rb_link = p; | |
3ac6c9f8 | 2318 | return cfqq; |
a36e71f9 JA |
2319 | } |
2320 | ||
2321 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2322 | { | |
a36e71f9 JA |
2323 | struct rb_node **p, *parent; |
2324 | struct cfq_queue *__cfqq; | |
2325 | ||
f2d1f0ae JA |
2326 | if (cfqq->p_root) { |
2327 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2328 | cfqq->p_root = NULL; | |
2329 | } | |
a36e71f9 JA |
2330 | |
2331 | if (cfq_class_idle(cfqq)) | |
2332 | return; | |
2333 | if (!cfqq->next_rq) | |
2334 | return; | |
2335 | ||
f2d1f0ae | 2336 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
2337 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
2338 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
2339 | if (!__cfqq) { |
2340 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
2341 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
2342 | } else | |
2343 | cfqq->p_root = NULL; | |
a36e71f9 JA |
2344 | } |
2345 | ||
498d3aa2 JA |
2346 | /* |
2347 | * Update cfqq's position in the service tree. | |
2348 | */ | |
edd75ffd | 2349 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 2350 | { |
6d048f53 JA |
2351 | /* |
2352 | * Resorting requires the cfqq to be on the RR list already. | |
2353 | */ | |
a36e71f9 | 2354 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 2355 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
2356 | cfq_prio_tree_add(cfqd, cfqq); |
2357 | } | |
6d048f53 JA |
2358 | } |
2359 | ||
1da177e4 LT |
2360 | /* |
2361 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 2362 | * the pending list according to last request service |
1da177e4 | 2363 | */ |
febffd61 | 2364 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2365 | { |
7b679138 | 2366 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
2367 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
2368 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 2369 | cfqd->busy_queues++; |
ef8a41df SL |
2370 | if (cfq_cfqq_sync(cfqq)) |
2371 | cfqd->busy_sync_queues++; | |
1da177e4 | 2372 | |
edd75ffd | 2373 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
2374 | } |
2375 | ||
498d3aa2 JA |
2376 | /* |
2377 | * Called when the cfqq no longer has requests pending, remove it from | |
2378 | * the service tree. | |
2379 | */ | |
febffd61 | 2380 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2381 | { |
7b679138 | 2382 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
2383 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
2384 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 2385 | |
aa6f6a3d CZ |
2386 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
2387 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
2388 | cfqq->service_tree = NULL; | |
2389 | } | |
f2d1f0ae JA |
2390 | if (cfqq->p_root) { |
2391 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2392 | cfqq->p_root = NULL; | |
2393 | } | |
d9e7620e | 2394 | |
8184f93e | 2395 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
2396 | BUG_ON(!cfqd->busy_queues); |
2397 | cfqd->busy_queues--; | |
ef8a41df SL |
2398 | if (cfq_cfqq_sync(cfqq)) |
2399 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
2400 | } |
2401 | ||
2402 | /* | |
2403 | * rb tree support functions | |
2404 | */ | |
febffd61 | 2405 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 2406 | { |
5e705374 | 2407 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 2408 | const int sync = rq_is_sync(rq); |
1da177e4 | 2409 | |
b4878f24 JA |
2410 | BUG_ON(!cfqq->queued[sync]); |
2411 | cfqq->queued[sync]--; | |
1da177e4 | 2412 | |
5e705374 | 2413 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 2414 | |
f04a6424 VG |
2415 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
2416 | /* | |
2417 | * Queue will be deleted from service tree when we actually | |
2418 | * expire it later. Right now just remove it from prio tree | |
2419 | * as it is empty. | |
2420 | */ | |
2421 | if (cfqq->p_root) { | |
2422 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2423 | cfqq->p_root = NULL; | |
2424 | } | |
2425 | } | |
1da177e4 LT |
2426 | } |
2427 | ||
5e705374 | 2428 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 2429 | { |
5e705374 | 2430 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2431 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 2432 | struct request *prev; |
1da177e4 | 2433 | |
5380a101 | 2434 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 2435 | |
796d5116 | 2436 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
2437 | |
2438 | if (!cfq_cfqq_on_rr(cfqq)) | |
2439 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
2440 | |
2441 | /* | |
2442 | * check if this request is a better next-serve candidate | |
2443 | */ | |
a36e71f9 | 2444 | prev = cfqq->next_rq; |
cf7c25cf | 2445 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
2446 | |
2447 | /* | |
2448 | * adjust priority tree position, if ->next_rq changes | |
2449 | */ | |
2450 | if (prev != cfqq->next_rq) | |
2451 | cfq_prio_tree_add(cfqd, cfqq); | |
2452 | ||
5044eed4 | 2453 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
2454 | } |
2455 | ||
febffd61 | 2456 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 2457 | { |
5380a101 JA |
2458 | elv_rb_del(&cfqq->sort_list, rq); |
2459 | cfqq->queued[rq_is_sync(rq)]--; | |
ef295ecf | 2460 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
5e705374 | 2461 | cfq_add_rq_rb(rq); |
155fead9 | 2462 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
ef295ecf | 2463 | rq->cmd_flags); |
1da177e4 LT |
2464 | } |
2465 | ||
206dc69b JA |
2466 | static struct request * |
2467 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 2468 | { |
206dc69b | 2469 | struct task_struct *tsk = current; |
c5869807 | 2470 | struct cfq_io_cq *cic; |
206dc69b | 2471 | struct cfq_queue *cfqq; |
1da177e4 | 2472 | |
4ac845a2 | 2473 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2474 | if (!cic) |
2475 | return NULL; | |
2476 | ||
aa39ebd4 | 2477 | cfqq = cic_to_cfqq(cic, op_is_sync(bio->bi_opf)); |
f73a1c7d KO |
2478 | if (cfqq) |
2479 | return elv_rb_find(&cfqq->sort_list, bio_end_sector(bio)); | |
1da177e4 | 2480 | |
1da177e4 LT |
2481 | return NULL; |
2482 | } | |
2483 | ||
165125e1 | 2484 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2485 | { |
22e2c507 | 2486 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 2487 | |
53c583d2 | 2488 | cfqd->rq_in_driver++; |
7b679138 | 2489 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 2490 | cfqd->rq_in_driver); |
25776e35 | 2491 | |
5b93629b | 2492 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
2493 | } |
2494 | ||
165125e1 | 2495 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2496 | { |
b4878f24 JA |
2497 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2498 | ||
53c583d2 CZ |
2499 | WARN_ON(!cfqd->rq_in_driver); |
2500 | cfqd->rq_in_driver--; | |
7b679138 | 2501 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 2502 | cfqd->rq_in_driver); |
1da177e4 LT |
2503 | } |
2504 | ||
b4878f24 | 2505 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 2506 | { |
5e705374 | 2507 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 2508 | |
5e705374 JA |
2509 | if (cfqq->next_rq == rq) |
2510 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 2511 | |
b4878f24 | 2512 | list_del_init(&rq->queuelist); |
5e705374 | 2513 | cfq_del_rq_rb(rq); |
374f84ac | 2514 | |
45333d5a | 2515 | cfqq->cfqd->rq_queued--; |
ef295ecf | 2516 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
65299a3b CH |
2517 | if (rq->cmd_flags & REQ_PRIO) { |
2518 | WARN_ON(!cfqq->prio_pending); | |
2519 | cfqq->prio_pending--; | |
b53d1ed7 | 2520 | } |
1da177e4 LT |
2521 | } |
2522 | ||
34fe7c05 | 2523 | static enum elv_merge cfq_merge(struct request_queue *q, struct request **req, |
165125e1 | 2524 | struct bio *bio) |
1da177e4 LT |
2525 | { |
2526 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2527 | struct request *__rq; | |
1da177e4 | 2528 | |
206dc69b | 2529 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
72ef799b | 2530 | if (__rq && elv_bio_merge_ok(__rq, bio)) { |
9817064b JA |
2531 | *req = __rq; |
2532 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
2533 | } |
2534 | ||
2535 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
2536 | } |
2537 | ||
165125e1 | 2538 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
34fe7c05 | 2539 | enum elv_merge type) |
1da177e4 | 2540 | { |
21183b07 | 2541 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 2542 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 2543 | |
5e705374 | 2544 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 2545 | } |
1da177e4 LT |
2546 | } |
2547 | ||
812d4026 DS |
2548 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
2549 | struct bio *bio) | |
2550 | { | |
ef295ecf | 2551 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_opf); |
812d4026 DS |
2552 | } |
2553 | ||
1da177e4 | 2554 | static void |
165125e1 | 2555 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
2556 | struct request *next) |
2557 | { | |
cf7c25cf | 2558 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
2559 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2560 | ||
22e2c507 JA |
2561 | /* |
2562 | * reposition in fifo if next is older than rq | |
2563 | */ | |
2564 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
9a7f38c4 | 2565 | next->fifo_time < rq->fifo_time && |
3d106fba | 2566 | cfqq == RQ_CFQQ(next)) { |
22e2c507 | 2567 | list_move(&rq->queuelist, &next->queuelist); |
8b4922d3 | 2568 | rq->fifo_time = next->fifo_time; |
30996f40 | 2569 | } |
22e2c507 | 2570 | |
cf7c25cf CZ |
2571 | if (cfqq->next_rq == next) |
2572 | cfqq->next_rq = rq; | |
b4878f24 | 2573 | cfq_remove_request(next); |
ef295ecf | 2574 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
4a0b75c7 SL |
2575 | |
2576 | cfqq = RQ_CFQQ(next); | |
2577 | /* | |
2578 | * all requests of this queue are merged to other queues, delete it | |
2579 | * from the service tree. If it's the active_queue, | |
2580 | * cfq_dispatch_requests() will choose to expire it or do idle | |
2581 | */ | |
2582 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
2583 | cfqq != cfqd->active_queue) | |
2584 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
2585 | } |
2586 | ||
72ef799b TE |
2587 | static int cfq_allow_bio_merge(struct request_queue *q, struct request *rq, |
2588 | struct bio *bio) | |
da775265 JA |
2589 | { |
2590 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
aa39ebd4 | 2591 | bool is_sync = op_is_sync(bio->bi_opf); |
c5869807 | 2592 | struct cfq_io_cq *cic; |
da775265 | 2593 | struct cfq_queue *cfqq; |
da775265 JA |
2594 | |
2595 | /* | |
ec8acb69 | 2596 | * Disallow merge of a sync bio into an async request. |
da775265 | 2597 | */ |
aa39ebd4 | 2598 | if (is_sync && !rq_is_sync(rq)) |
a6151c3a | 2599 | return false; |
da775265 JA |
2600 | |
2601 | /* | |
f1a4f4d3 | 2602 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 2603 | * merge only if rq is queued there. |
f1a4f4d3 | 2604 | */ |
07c2bd37 TH |
2605 | cic = cfq_cic_lookup(cfqd, current->io_context); |
2606 | if (!cic) | |
2607 | return false; | |
719d3402 | 2608 | |
aa39ebd4 | 2609 | cfqq = cic_to_cfqq(cic, is_sync); |
a6151c3a | 2610 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
2611 | } |
2612 | ||
72ef799b TE |
2613 | static int cfq_allow_rq_merge(struct request_queue *q, struct request *rq, |
2614 | struct request *next) | |
2615 | { | |
2616 | return RQ_CFQQ(rq) == RQ_CFQQ(next); | |
2617 | } | |
2618 | ||
812df48d DS |
2619 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2620 | { | |
91148325 | 2621 | hrtimer_try_to_cancel(&cfqd->idle_slice_timer); |
155fead9 | 2622 | cfqg_stats_update_idle_time(cfqq->cfqg); |
812df48d DS |
2623 | } |
2624 | ||
febffd61 JA |
2625 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
2626 | struct cfq_queue *cfqq) | |
22e2c507 JA |
2627 | { |
2628 | if (cfqq) { | |
3bf10fea | 2629 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d", |
4d2ceea4 | 2630 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
155fead9 | 2631 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
62a37f6b | 2632 | cfqq->slice_start = 0; |
9a7f38c4 | 2633 | cfqq->dispatch_start = ktime_get_ns(); |
62a37f6b JT |
2634 | cfqq->allocated_slice = 0; |
2635 | cfqq->slice_end = 0; | |
2636 | cfqq->slice_dispatch = 0; | |
2637 | cfqq->nr_sectors = 0; | |
2638 | ||
2639 | cfq_clear_cfqq_wait_request(cfqq); | |
2640 | cfq_clear_cfqq_must_dispatch(cfqq); | |
2641 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
2642 | cfq_clear_cfqq_fifo_expire(cfqq); | |
2643 | cfq_mark_cfqq_slice_new(cfqq); | |
2644 | ||
2645 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
2646 | } |
2647 | ||
2648 | cfqd->active_queue = cfqq; | |
2649 | } | |
2650 | ||
7b14e3b5 JA |
2651 | /* |
2652 | * current cfqq expired its slice (or was too idle), select new one | |
2653 | */ | |
2654 | static void | |
2655 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 2656 | bool timed_out) |
7b14e3b5 | 2657 | { |
7b679138 JA |
2658 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
2659 | ||
7b14e3b5 | 2660 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 2661 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 2662 | |
7b14e3b5 | 2663 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 2664 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 2665 | |
ae54abed SL |
2666 | /* |
2667 | * If this cfqq is shared between multiple processes, check to | |
2668 | * make sure that those processes are still issuing I/Os within | |
2669 | * the mean seek distance. If not, it may be time to break the | |
2670 | * queues apart again. | |
2671 | */ | |
2672 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
2673 | cfq_mark_cfqq_split_coop(cfqq); | |
2674 | ||
7b14e3b5 | 2675 | /* |
6084cdda | 2676 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 2677 | */ |
c553f8e3 SL |
2678 | if (timed_out) { |
2679 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 2680 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 2681 | else |
9a7f38c4 | 2682 | cfqq->slice_resid = cfqq->slice_end - ktime_get_ns(); |
93fdf147 | 2683 | cfq_log_cfqq(cfqd, cfqq, "resid=%lld", cfqq->slice_resid); |
7b679138 | 2684 | } |
7b14e3b5 | 2685 | |
e5ff082e | 2686 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 2687 | |
f04a6424 VG |
2688 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
2689 | cfq_del_cfqq_rr(cfqd, cfqq); | |
2690 | ||
edd75ffd | 2691 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
2692 | |
2693 | if (cfqq == cfqd->active_queue) | |
2694 | cfqd->active_queue = NULL; | |
2695 | ||
2696 | if (cfqd->active_cic) { | |
11a3122f | 2697 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
2698 | cfqd->active_cic = NULL; |
2699 | } | |
7b14e3b5 JA |
2700 | } |
2701 | ||
e5ff082e | 2702 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
2703 | { |
2704 | struct cfq_queue *cfqq = cfqd->active_queue; | |
2705 | ||
2706 | if (cfqq) | |
e5ff082e | 2707 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
2708 | } |
2709 | ||
498d3aa2 JA |
2710 | /* |
2711 | * Get next queue for service. Unless we have a queue preemption, | |
2712 | * we'll simply select the first cfqq in the service tree. | |
2713 | */ | |
6d048f53 | 2714 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 2715 | { |
34b98d03 VG |
2716 | struct cfq_rb_root *st = st_for(cfqd->serving_group, |
2717 | cfqd->serving_wl_class, cfqd->serving_wl_type); | |
d9e7620e | 2718 | |
f04a6424 VG |
2719 | if (!cfqd->rq_queued) |
2720 | return NULL; | |
2721 | ||
1fa8f6d6 | 2722 | /* There is nothing to dispatch */ |
34b98d03 | 2723 | if (!st) |
1fa8f6d6 | 2724 | return NULL; |
34b98d03 | 2725 | if (RB_EMPTY_ROOT(&st->rb)) |
c0324a02 | 2726 | return NULL; |
34b98d03 | 2727 | return cfq_rb_first(st); |
6d048f53 JA |
2728 | } |
2729 | ||
f04a6424 VG |
2730 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
2731 | { | |
25fb5169 | 2732 | struct cfq_group *cfqg; |
f04a6424 VG |
2733 | struct cfq_queue *cfqq; |
2734 | int i, j; | |
2735 | struct cfq_rb_root *st; | |
2736 | ||
2737 | if (!cfqd->rq_queued) | |
2738 | return NULL; | |
2739 | ||
25fb5169 VG |
2740 | cfqg = cfq_get_next_cfqg(cfqd); |
2741 | if (!cfqg) | |
2742 | return NULL; | |
2743 | ||
1cf41753 ME |
2744 | for_each_cfqg_st(cfqg, i, j, st) { |
2745 | cfqq = cfq_rb_first(st); | |
2746 | if (cfqq) | |
f04a6424 | 2747 | return cfqq; |
1cf41753 | 2748 | } |
f04a6424 VG |
2749 | return NULL; |
2750 | } | |
2751 | ||
498d3aa2 JA |
2752 | /* |
2753 | * Get and set a new active queue for service. | |
2754 | */ | |
a36e71f9 JA |
2755 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
2756 | struct cfq_queue *cfqq) | |
6d048f53 | 2757 | { |
e00ef799 | 2758 | if (!cfqq) |
a36e71f9 | 2759 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 2760 | |
22e2c507 | 2761 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 2762 | return cfqq; |
22e2c507 JA |
2763 | } |
2764 | ||
d9e7620e JA |
2765 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
2766 | struct request *rq) | |
2767 | { | |
83096ebf TH |
2768 | if (blk_rq_pos(rq) >= cfqd->last_position) |
2769 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 2770 | else |
83096ebf | 2771 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
2772 | } |
2773 | ||
b2c18e1e | 2774 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 2775 | struct request *rq) |
6d048f53 | 2776 | { |
e9ce335d | 2777 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
2778 | } |
2779 | ||
a36e71f9 JA |
2780 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
2781 | struct cfq_queue *cur_cfqq) | |
2782 | { | |
f2d1f0ae | 2783 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
2784 | struct rb_node *parent, *node; |
2785 | struct cfq_queue *__cfqq; | |
2786 | sector_t sector = cfqd->last_position; | |
2787 | ||
2788 | if (RB_EMPTY_ROOT(root)) | |
2789 | return NULL; | |
2790 | ||
2791 | /* | |
2792 | * First, if we find a request starting at the end of the last | |
2793 | * request, choose it. | |
2794 | */ | |
f2d1f0ae | 2795 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
2796 | if (__cfqq) |
2797 | return __cfqq; | |
2798 | ||
2799 | /* | |
2800 | * If the exact sector wasn't found, the parent of the NULL leaf | |
2801 | * will contain the closest sector. | |
2802 | */ | |
2803 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 2804 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2805 | return __cfqq; |
2806 | ||
2e46e8b2 | 2807 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
2808 | node = rb_next(&__cfqq->p_node); |
2809 | else | |
2810 | node = rb_prev(&__cfqq->p_node); | |
2811 | if (!node) | |
2812 | return NULL; | |
2813 | ||
2814 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 2815 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2816 | return __cfqq; |
2817 | ||
2818 | return NULL; | |
2819 | } | |
2820 | ||
2821 | /* | |
2822 | * cfqd - obvious | |
2823 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
2824 | * closely cooperating with itself. | |
2825 | * | |
2826 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
2827 | * one request, and that cfqd->last_position reflects a position on the disk | |
2828 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
2829 | * assumption. | |
2830 | */ | |
2831 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 2832 | struct cfq_queue *cur_cfqq) |
6d048f53 | 2833 | { |
a36e71f9 JA |
2834 | struct cfq_queue *cfqq; |
2835 | ||
39c01b21 DS |
2836 | if (cfq_class_idle(cur_cfqq)) |
2837 | return NULL; | |
e6c5bc73 JM |
2838 | if (!cfq_cfqq_sync(cur_cfqq)) |
2839 | return NULL; | |
2840 | if (CFQQ_SEEKY(cur_cfqq)) | |
2841 | return NULL; | |
2842 | ||
b9d8f4c7 GJ |
2843 | /* |
2844 | * Don't search priority tree if it's the only queue in the group. | |
2845 | */ | |
2846 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
2847 | return NULL; | |
2848 | ||
6d048f53 | 2849 | /* |
d9e7620e JA |
2850 | * We should notice if some of the queues are cooperating, eg |
2851 | * working closely on the same area of the disk. In that case, | |
2852 | * we can group them together and don't waste time idling. | |
6d048f53 | 2853 | */ |
a36e71f9 JA |
2854 | cfqq = cfqq_close(cfqd, cur_cfqq); |
2855 | if (!cfqq) | |
2856 | return NULL; | |
2857 | ||
8682e1f1 VG |
2858 | /* If new queue belongs to different cfq_group, don't choose it */ |
2859 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
2860 | return NULL; | |
2861 | ||
df5fe3e8 JM |
2862 | /* |
2863 | * It only makes sense to merge sync queues. | |
2864 | */ | |
2865 | if (!cfq_cfqq_sync(cfqq)) | |
2866 | return NULL; | |
e6c5bc73 JM |
2867 | if (CFQQ_SEEKY(cfqq)) |
2868 | return NULL; | |
df5fe3e8 | 2869 | |
c0324a02 CZ |
2870 | /* |
2871 | * Do not merge queues of different priority classes | |
2872 | */ | |
2873 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
2874 | return NULL; | |
2875 | ||
a36e71f9 | 2876 | return cfqq; |
6d048f53 JA |
2877 | } |
2878 | ||
a6d44e98 CZ |
2879 | /* |
2880 | * Determine whether we should enforce idle window for this queue. | |
2881 | */ | |
2882 | ||
2883 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2884 | { | |
3bf10fea | 2885 | enum wl_class_t wl_class = cfqq_class(cfqq); |
34b98d03 | 2886 | struct cfq_rb_root *st = cfqq->service_tree; |
a6d44e98 | 2887 | |
34b98d03 VG |
2888 | BUG_ON(!st); |
2889 | BUG_ON(!st->count); | |
f04a6424 | 2890 | |
b6508c16 VG |
2891 | if (!cfqd->cfq_slice_idle) |
2892 | return false; | |
2893 | ||
a6d44e98 | 2894 | /* We never do for idle class queues. */ |
3bf10fea | 2895 | if (wl_class == IDLE_WORKLOAD) |
a6d44e98 CZ |
2896 | return false; |
2897 | ||
2898 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
2899 | if (cfq_cfqq_idle_window(cfqq) && |
2900 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
2901 | return true; |
2902 | ||
2903 | /* | |
2904 | * Otherwise, we do only if they are the last ones | |
2905 | * in their service tree. | |
2906 | */ | |
34b98d03 VG |
2907 | if (st->count == 1 && cfq_cfqq_sync(cfqq) && |
2908 | !cfq_io_thinktime_big(cfqd, &st->ttime, false)) | |
c1e44756 | 2909 | return true; |
34b98d03 | 2910 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count); |
c1e44756 | 2911 | return false; |
a6d44e98 CZ |
2912 | } |
2913 | ||
6d048f53 | 2914 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2915 | { |
1792669c | 2916 | struct cfq_queue *cfqq = cfqd->active_queue; |
e795421e | 2917 | struct cfq_rb_root *st = cfqq->service_tree; |
c5869807 | 2918 | struct cfq_io_cq *cic; |
9a7f38c4 JM |
2919 | u64 sl, group_idle = 0; |
2920 | u64 now = ktime_get_ns(); | |
7b14e3b5 | 2921 | |
a68bbddb | 2922 | /* |
f7d7b7a7 JA |
2923 | * SSD device without seek penalty, disable idling. But only do so |
2924 | * for devices that support queuing, otherwise we still have a problem | |
2925 | * with sync vs async workloads. | |
a68bbddb | 2926 | */ |
f7d7b7a7 | 2927 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
2928 | return; |
2929 | ||
dd67d051 | 2930 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2931 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2932 | |
2933 | /* | |
2934 | * idle is disabled, either manually or by past process history | |
2935 | */ | |
80bdf0c7 VG |
2936 | if (!cfq_should_idle(cfqd, cfqq)) { |
2937 | /* no queue idling. Check for group idling */ | |
2938 | if (cfqd->cfq_group_idle) | |
2939 | group_idle = cfqd->cfq_group_idle; | |
2940 | else | |
2941 | return; | |
2942 | } | |
6d048f53 | 2943 | |
7b679138 | 2944 | /* |
8e550632 | 2945 | * still active requests from this queue, don't idle |
7b679138 | 2946 | */ |
8e550632 | 2947 | if (cfqq->dispatched) |
7b679138 JA |
2948 | return; |
2949 | ||
22e2c507 JA |
2950 | /* |
2951 | * task has exited, don't wait | |
2952 | */ | |
206dc69b | 2953 | cic = cfqd->active_cic; |
f6e8d01b | 2954 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
2955 | return; |
2956 | ||
355b659c CZ |
2957 | /* |
2958 | * If our average think time is larger than the remaining time | |
2959 | * slice, then don't idle. This avoids overrunning the allotted | |
2960 | * time slice. | |
2961 | */ | |
383cd721 | 2962 | if (sample_valid(cic->ttime.ttime_samples) && |
9a7f38c4 JM |
2963 | (cfqq->slice_end - now < cic->ttime.ttime_mean)) { |
2964 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%llu", | |
383cd721 | 2965 | cic->ttime.ttime_mean); |
355b659c | 2966 | return; |
b1ffe737 | 2967 | } |
355b659c | 2968 | |
e795421e JK |
2969 | /* |
2970 | * There are other queues in the group or this is the only group and | |
2971 | * it has too big thinktime, don't do group idle. | |
2972 | */ | |
2973 | if (group_idle && | |
2974 | (cfqq->cfqg->nr_cfqq > 1 || | |
2975 | cfq_io_thinktime_big(cfqd, &st->ttime, true))) | |
80bdf0c7 VG |
2976 | return; |
2977 | ||
3b18152c | 2978 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2979 | |
80bdf0c7 VG |
2980 | if (group_idle) |
2981 | sl = cfqd->cfq_group_idle; | |
2982 | else | |
2983 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2984 | |
91148325 JK |
2985 | hrtimer_start(&cfqd->idle_slice_timer, ns_to_ktime(sl), |
2986 | HRTIMER_MODE_REL); | |
155fead9 | 2987 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
9a7f38c4 | 2988 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %llu group_idle: %d", sl, |
80bdf0c7 | 2989 | group_idle ? 1 : 0); |
1da177e4 LT |
2990 | } |
2991 | ||
498d3aa2 JA |
2992 | /* |
2993 | * Move request from internal lists to the request queue dispatch list. | |
2994 | */ | |
165125e1 | 2995 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2996 | { |
3ed9a296 | 2997 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2998 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2999 | |
7b679138 JA |
3000 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
3001 | ||
06d21886 | 3002 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 3003 | cfq_remove_request(rq); |
6d048f53 | 3004 | cfqq->dispatched++; |
80bdf0c7 | 3005 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 3006 | elv_dispatch_sort(q, rq); |
3ed9a296 | 3007 | |
53c583d2 | 3008 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 3009 | cfqq->nr_sectors += blk_rq_sectors(rq); |
1da177e4 LT |
3010 | } |
3011 | ||
3012 | /* | |
3013 | * return expired entry, or NULL to just start from scratch in rbtree | |
3014 | */ | |
febffd61 | 3015 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 3016 | { |
30996f40 | 3017 | struct request *rq = NULL; |
1da177e4 | 3018 | |
3b18152c | 3019 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 3020 | return NULL; |
cb887411 JA |
3021 | |
3022 | cfq_mark_cfqq_fifo_expire(cfqq); | |
3023 | ||
89850f7e JA |
3024 | if (list_empty(&cfqq->fifo)) |
3025 | return NULL; | |
1da177e4 | 3026 | |
89850f7e | 3027 | rq = rq_entry_fifo(cfqq->fifo.next); |
9a7f38c4 | 3028 | if (ktime_get_ns() < rq->fifo_time) |
7b679138 | 3029 | rq = NULL; |
1da177e4 | 3030 | |
6d048f53 | 3031 | return rq; |
1da177e4 LT |
3032 | } |
3033 | ||
22e2c507 JA |
3034 | static inline int |
3035 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3036 | { | |
3037 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 3038 | |
22e2c507 | 3039 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 3040 | |
b9f8ce05 | 3041 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
3042 | } |
3043 | ||
df5fe3e8 JM |
3044 | /* |
3045 | * Must be called with the queue_lock held. | |
3046 | */ | |
3047 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
3048 | { | |
3049 | int process_refs, io_refs; | |
3050 | ||
3051 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 3052 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
3053 | BUG_ON(process_refs < 0); |
3054 | return process_refs; | |
3055 | } | |
3056 | ||
3057 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
3058 | { | |
e6c5bc73 | 3059 | int process_refs, new_process_refs; |
df5fe3e8 JM |
3060 | struct cfq_queue *__cfqq; |
3061 | ||
c10b61f0 JM |
3062 | /* |
3063 | * If there are no process references on the new_cfqq, then it is | |
3064 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
3065 | * chain may have dropped their last reference (not just their | |
3066 | * last process reference). | |
3067 | */ | |
3068 | if (!cfqq_process_refs(new_cfqq)) | |
3069 | return; | |
3070 | ||
df5fe3e8 JM |
3071 | /* Avoid a circular list and skip interim queue merges */ |
3072 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
3073 | if (__cfqq == cfqq) | |
3074 | return; | |
3075 | new_cfqq = __cfqq; | |
3076 | } | |
3077 | ||
3078 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 3079 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
3080 | /* |
3081 | * If the process for the cfqq has gone away, there is no | |
3082 | * sense in merging the queues. | |
3083 | */ | |
c10b61f0 | 3084 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
3085 | return; |
3086 | ||
e6c5bc73 JM |
3087 | /* |
3088 | * Merge in the direction of the lesser amount of work. | |
3089 | */ | |
e6c5bc73 JM |
3090 | if (new_process_refs >= process_refs) { |
3091 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 3092 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
3093 | } else { |
3094 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 3095 | cfqq->ref += new_process_refs; |
e6c5bc73 | 3096 | } |
df5fe3e8 JM |
3097 | } |
3098 | ||
6d816ec7 | 3099 | static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd, |
3bf10fea | 3100 | struct cfq_group *cfqg, enum wl_class_t wl_class) |
718eee05 CZ |
3101 | { |
3102 | struct cfq_queue *queue; | |
3103 | int i; | |
3104 | bool key_valid = false; | |
9a7f38c4 | 3105 | u64 lowest_key = 0; |
718eee05 CZ |
3106 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; |
3107 | ||
65b32a57 VG |
3108 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
3109 | /* select the one with lowest rb_key */ | |
34b98d03 | 3110 | queue = cfq_rb_first(st_for(cfqg, wl_class, i)); |
718eee05 | 3111 | if (queue && |
9a7f38c4 | 3112 | (!key_valid || queue->rb_key < lowest_key)) { |
718eee05 CZ |
3113 | lowest_key = queue->rb_key; |
3114 | cur_best = i; | |
3115 | key_valid = true; | |
3116 | } | |
3117 | } | |
3118 | ||
3119 | return cur_best; | |
3120 | } | |
3121 | ||
6d816ec7 VG |
3122 | static void |
3123 | choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
718eee05 | 3124 | { |
9a7f38c4 | 3125 | u64 slice; |
718eee05 | 3126 | unsigned count; |
cdb16e8f | 3127 | struct cfq_rb_root *st; |
9a7f38c4 | 3128 | u64 group_slice; |
4d2ceea4 | 3129 | enum wl_class_t original_class = cfqd->serving_wl_class; |
9a7f38c4 | 3130 | u64 now = ktime_get_ns(); |
1fa8f6d6 | 3131 | |
718eee05 | 3132 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 3133 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 3134 | cfqd->serving_wl_class = RT_WORKLOAD; |
58ff82f3 | 3135 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 3136 | cfqd->serving_wl_class = BE_WORKLOAD; |
718eee05 | 3137 | else { |
4d2ceea4 | 3138 | cfqd->serving_wl_class = IDLE_WORKLOAD; |
9a7f38c4 | 3139 | cfqd->workload_expires = now + jiffies_to_nsecs(1); |
718eee05 CZ |
3140 | return; |
3141 | } | |
3142 | ||
4d2ceea4 | 3143 | if (original_class != cfqd->serving_wl_class) |
e4ea0c16 SL |
3144 | goto new_workload; |
3145 | ||
718eee05 CZ |
3146 | /* |
3147 | * For RT and BE, we have to choose also the type | |
3148 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
3149 | * expiration time | |
3150 | */ | |
34b98d03 | 3151 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 3152 | count = st->count; |
718eee05 CZ |
3153 | |
3154 | /* | |
65b32a57 | 3155 | * check workload expiration, and that we still have other queues ready |
718eee05 | 3156 | */ |
9a7f38c4 | 3157 | if (count && !(now > cfqd->workload_expires)) |
718eee05 CZ |
3158 | return; |
3159 | ||
e4ea0c16 | 3160 | new_workload: |
718eee05 | 3161 | /* otherwise select new workload type */ |
6d816ec7 | 3162 | cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg, |
4d2ceea4 | 3163 | cfqd->serving_wl_class); |
34b98d03 | 3164 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 3165 | count = st->count; |
718eee05 CZ |
3166 | |
3167 | /* | |
3168 | * the workload slice is computed as a fraction of target latency | |
3169 | * proportional to the number of queues in that workload, over | |
3170 | * all the queues in the same priority class | |
3171 | */ | |
58ff82f3 VG |
3172 | group_slice = cfq_group_slice(cfqd, cfqg); |
3173 | ||
9a7f38c4 | 3174 | slice = div_u64(group_slice * count, |
4d2ceea4 VG |
3175 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class], |
3176 | cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd, | |
9a7f38c4 | 3177 | cfqg))); |
718eee05 | 3178 | |
4d2ceea4 | 3179 | if (cfqd->serving_wl_type == ASYNC_WORKLOAD) { |
9a7f38c4 | 3180 | u64 tmp; |
f26bd1f0 VG |
3181 | |
3182 | /* | |
3183 | * Async queues are currently system wide. Just taking | |
3184 | * proportion of queues with-in same group will lead to higher | |
3185 | * async ratio system wide as generally root group is going | |
3186 | * to have higher weight. A more accurate thing would be to | |
3187 | * calculate system wide asnc/sync ratio. | |
3188 | */ | |
5bf14c07 TM |
3189 | tmp = cfqd->cfq_target_latency * |
3190 | cfqg_busy_async_queues(cfqd, cfqg); | |
9a7f38c4 JM |
3191 | tmp = div_u64(tmp, cfqd->busy_queues); |
3192 | slice = min_t(u64, slice, tmp); | |
f26bd1f0 | 3193 | |
718eee05 CZ |
3194 | /* async workload slice is scaled down according to |
3195 | * the sync/async slice ratio. */ | |
9a7f38c4 | 3196 | slice = div64_u64(slice*cfqd->cfq_slice[0], cfqd->cfq_slice[1]); |
f26bd1f0 | 3197 | } else |
718eee05 CZ |
3198 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
3199 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
3200 | ||
9a7f38c4 JM |
3201 | slice = max_t(u64, slice, CFQ_MIN_TT); |
3202 | cfq_log(cfqd, "workload slice:%llu", slice); | |
3203 | cfqd->workload_expires = now + slice; | |
718eee05 CZ |
3204 | } |
3205 | ||
1fa8f6d6 VG |
3206 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
3207 | { | |
3208 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 3209 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
3210 | |
3211 | if (RB_EMPTY_ROOT(&st->rb)) | |
3212 | return NULL; | |
25bc6b07 | 3213 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
3214 | update_min_vdisktime(st); |
3215 | return cfqg; | |
1fa8f6d6 VG |
3216 | } |
3217 | ||
cdb16e8f VG |
3218 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
3219 | { | |
1fa8f6d6 | 3220 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
9a7f38c4 | 3221 | u64 now = ktime_get_ns(); |
1fa8f6d6 VG |
3222 | |
3223 | cfqd->serving_group = cfqg; | |
dae739eb VG |
3224 | |
3225 | /* Restore the workload type data */ | |
4d2ceea4 | 3226 | if (cfqg->saved_wl_slice) { |
9a7f38c4 | 3227 | cfqd->workload_expires = now + cfqg->saved_wl_slice; |
4d2ceea4 VG |
3228 | cfqd->serving_wl_type = cfqg->saved_wl_type; |
3229 | cfqd->serving_wl_class = cfqg->saved_wl_class; | |
66ae2919 | 3230 | } else |
9a7f38c4 | 3231 | cfqd->workload_expires = now - 1; |
66ae2919 | 3232 | |
6d816ec7 | 3233 | choose_wl_class_and_type(cfqd, cfqg); |
cdb16e8f VG |
3234 | } |
3235 | ||
22e2c507 | 3236 | /* |
498d3aa2 JA |
3237 | * Select a queue for service. If we have a current active queue, |
3238 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 3239 | */ |
1b5ed5e1 | 3240 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 3241 | { |
a36e71f9 | 3242 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
9a7f38c4 | 3243 | u64 now = ktime_get_ns(); |
1da177e4 | 3244 | |
22e2c507 JA |
3245 | cfqq = cfqd->active_queue; |
3246 | if (!cfqq) | |
3247 | goto new_queue; | |
1da177e4 | 3248 | |
f04a6424 VG |
3249 | if (!cfqd->rq_queued) |
3250 | return NULL; | |
c244bb50 VG |
3251 | |
3252 | /* | |
3253 | * We were waiting for group to get backlogged. Expire the queue | |
3254 | */ | |
3255 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3256 | goto expire; | |
3257 | ||
22e2c507 | 3258 | /* |
6d048f53 | 3259 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 3260 | */ |
7667aa06 VG |
3261 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
3262 | /* | |
3263 | * If slice had not expired at the completion of last request | |
3264 | * we might not have turned on wait_busy flag. Don't expire | |
3265 | * the queue yet. Allow the group to get backlogged. | |
3266 | * | |
3267 | * The very fact that we have used the slice, that means we | |
3268 | * have been idling all along on this queue and it should be | |
3269 | * ok to wait for this request to complete. | |
3270 | */ | |
82bbbf28 VG |
3271 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
3272 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
3273 | cfqq = NULL; | |
7667aa06 | 3274 | goto keep_queue; |
82bbbf28 | 3275 | } else |
80bdf0c7 | 3276 | goto check_group_idle; |
7667aa06 | 3277 | } |
1da177e4 | 3278 | |
22e2c507 | 3279 | /* |
6d048f53 JA |
3280 | * The active queue has requests and isn't expired, allow it to |
3281 | * dispatch. | |
22e2c507 | 3282 | */ |
dd67d051 | 3283 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3284 | goto keep_queue; |
6d048f53 | 3285 | |
a36e71f9 JA |
3286 | /* |
3287 | * If another queue has a request waiting within our mean seek | |
3288 | * distance, let it run. The expire code will check for close | |
3289 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 3290 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 3291 | */ |
b3b6d040 | 3292 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
3293 | if (new_cfqq) { |
3294 | if (!cfqq->new_cfqq) | |
3295 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 3296 | goto expire; |
df5fe3e8 | 3297 | } |
a36e71f9 | 3298 | |
6d048f53 JA |
3299 | /* |
3300 | * No requests pending. If the active queue still has requests in | |
3301 | * flight or is idling for a new request, allow either of these | |
3302 | * conditions to happen (or time out) before selecting a new queue. | |
3303 | */ | |
91148325 | 3304 | if (hrtimer_active(&cfqd->idle_slice_timer)) { |
80bdf0c7 VG |
3305 | cfqq = NULL; |
3306 | goto keep_queue; | |
3307 | } | |
3308 | ||
8e1ac665 SL |
3309 | /* |
3310 | * This is a deep seek queue, but the device is much faster than | |
3311 | * the queue can deliver, don't idle | |
3312 | **/ | |
3313 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
3314 | (cfq_cfqq_slice_new(cfqq) || | |
9a7f38c4 | 3315 | (cfqq->slice_end - now > now - cfqq->slice_start))) { |
8e1ac665 SL |
3316 | cfq_clear_cfqq_deep(cfqq); |
3317 | cfq_clear_cfqq_idle_window(cfqq); | |
3318 | } | |
3319 | ||
80bdf0c7 VG |
3320 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
3321 | cfqq = NULL; | |
3322 | goto keep_queue; | |
3323 | } | |
3324 | ||
3325 | /* | |
3326 | * If group idle is enabled and there are requests dispatched from | |
3327 | * this group, wait for requests to complete. | |
3328 | */ | |
3329 | check_group_idle: | |
7700fc4f SL |
3330 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
3331 | cfqq->cfqg->dispatched && | |
3332 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
3333 | cfqq = NULL; |
3334 | goto keep_queue; | |
22e2c507 JA |
3335 | } |
3336 | ||
3b18152c | 3337 | expire: |
e5ff082e | 3338 | cfq_slice_expired(cfqd, 0); |
3b18152c | 3339 | new_queue: |
718eee05 CZ |
3340 | /* |
3341 | * Current queue expired. Check if we have to switch to a new | |
3342 | * service tree | |
3343 | */ | |
3344 | if (!new_cfqq) | |
cdb16e8f | 3345 | cfq_choose_cfqg(cfqd); |
718eee05 | 3346 | |
a36e71f9 | 3347 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 3348 | keep_queue: |
3b18152c | 3349 | return cfqq; |
22e2c507 JA |
3350 | } |
3351 | ||
febffd61 | 3352 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
3353 | { |
3354 | int dispatched = 0; | |
3355 | ||
3356 | while (cfqq->next_rq) { | |
3357 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
3358 | dispatched++; | |
3359 | } | |
3360 | ||
3361 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
3362 | |
3363 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 3364 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
3365 | return dispatched; |
3366 | } | |
3367 | ||
498d3aa2 JA |
3368 | /* |
3369 | * Drain our current requests. Used for barriers and when switching | |
3370 | * io schedulers on-the-fly. | |
3371 | */ | |
d9e7620e | 3372 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 3373 | { |
0871714e | 3374 | struct cfq_queue *cfqq; |
d9e7620e | 3375 | int dispatched = 0; |
cdb16e8f | 3376 | |
3440c49f | 3377 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 3378 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
3379 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
3380 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 3381 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 3382 | } |
1b5ed5e1 | 3383 | |
1b5ed5e1 TH |
3384 | BUG_ON(cfqd->busy_queues); |
3385 | ||
6923715a | 3386 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
3387 | return dispatched; |
3388 | } | |
3389 | ||
abc3c744 SL |
3390 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
3391 | struct cfq_queue *cfqq) | |
3392 | { | |
9a7f38c4 JM |
3393 | u64 now = ktime_get_ns(); |
3394 | ||
abc3c744 SL |
3395 | /* the queue hasn't finished any request, can't estimate */ |
3396 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 3397 | return true; |
9a7f38c4 | 3398 | if (now + cfqd->cfq_slice_idle * cfqq->dispatched > cfqq->slice_end) |
c1e44756 | 3399 | return true; |
abc3c744 | 3400 | |
c1e44756 | 3401 | return false; |
abc3c744 SL |
3402 | } |
3403 | ||
0b182d61 | 3404 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 3405 | { |
2f5cb738 | 3406 | unsigned int max_dispatch; |
22e2c507 | 3407 | |
3932a86b GC |
3408 | if (cfq_cfqq_must_dispatch(cfqq)) |
3409 | return true; | |
3410 | ||
5ad531db JA |
3411 | /* |
3412 | * Drain async requests before we start sync IO | |
3413 | */ | |
53c583d2 | 3414 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 3415 | return false; |
5ad531db | 3416 | |
2f5cb738 JA |
3417 | /* |
3418 | * If this is an async queue and we have sync IO in flight, let it wait | |
3419 | */ | |
53c583d2 | 3420 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 3421 | return false; |
2f5cb738 | 3422 | |
abc3c744 | 3423 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
3424 | if (cfq_class_idle(cfqq)) |
3425 | max_dispatch = 1; | |
b4878f24 | 3426 | |
2f5cb738 JA |
3427 | /* |
3428 | * Does this cfqq already have too much IO in flight? | |
3429 | */ | |
3430 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 3431 | bool promote_sync = false; |
2f5cb738 JA |
3432 | /* |
3433 | * idle queue must always only have a single IO in flight | |
3434 | */ | |
3ed9a296 | 3435 | if (cfq_class_idle(cfqq)) |
0b182d61 | 3436 | return false; |
3ed9a296 | 3437 | |
ef8a41df | 3438 | /* |
c4ade94f LS |
3439 | * If there is only one sync queue |
3440 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
3441 | * queue no dispatch limit. The reason is a sync queue can |
3442 | * preempt async queue, limiting the sync queue doesn't make | |
3443 | * sense. This is useful for aiostress test. | |
3444 | */ | |
c4ade94f LS |
3445 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
3446 | promote_sync = true; | |
ef8a41df | 3447 | |
2f5cb738 JA |
3448 | /* |
3449 | * We have other queues, don't allow more IO from this one | |
3450 | */ | |
ef8a41df SL |
3451 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
3452 | !promote_sync) | |
0b182d61 | 3453 | return false; |
9ede209e | 3454 | |
365722bb | 3455 | /* |
474b18cc | 3456 | * Sole queue user, no limit |
365722bb | 3457 | */ |
ef8a41df | 3458 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
3459 | max_dispatch = -1; |
3460 | else | |
3461 | /* | |
3462 | * Normally we start throttling cfqq when cfq_quantum/2 | |
3463 | * requests have been dispatched. But we can drive | |
3464 | * deeper queue depths at the beginning of slice | |
3465 | * subjected to upper limit of cfq_quantum. | |
3466 | * */ | |
3467 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
3468 | } |
3469 | ||
3470 | /* | |
3471 | * Async queues must wait a bit before being allowed dispatch. | |
3472 | * We also ramp up the dispatch depth gradually for async IO, | |
3473 | * based on the last sync IO we serviced | |
3474 | */ | |
963b72fc | 3475 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
9a7f38c4 | 3476 | u64 last_sync = ktime_get_ns() - cfqd->last_delayed_sync; |
8e296755 | 3477 | unsigned int depth; |
365722bb | 3478 | |
9a7f38c4 | 3479 | depth = div64_u64(last_sync, cfqd->cfq_slice[1]); |
e00c54c3 JA |
3480 | if (!depth && !cfqq->dispatched) |
3481 | depth = 1; | |
8e296755 JA |
3482 | if (depth < max_dispatch) |
3483 | max_dispatch = depth; | |
2f5cb738 | 3484 | } |
3ed9a296 | 3485 | |
0b182d61 JA |
3486 | /* |
3487 | * If we're below the current max, allow a dispatch | |
3488 | */ | |
3489 | return cfqq->dispatched < max_dispatch; | |
3490 | } | |
3491 | ||
3492 | /* | |
3493 | * Dispatch a request from cfqq, moving them to the request queue | |
3494 | * dispatch list. | |
3495 | */ | |
3496 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3497 | { | |
3498 | struct request *rq; | |
3499 | ||
3500 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
3501 | ||
3932a86b GC |
3502 | rq = cfq_check_fifo(cfqq); |
3503 | if (rq) | |
3504 | cfq_mark_cfqq_must_dispatch(cfqq); | |
3505 | ||
0b182d61 JA |
3506 | if (!cfq_may_dispatch(cfqd, cfqq)) |
3507 | return false; | |
3508 | ||
3509 | /* | |
3510 | * follow expired path, else get first next available | |
3511 | */ | |
0b182d61 JA |
3512 | if (!rq) |
3513 | rq = cfqq->next_rq; | |
3932a86b GC |
3514 | else |
3515 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); | |
0b182d61 JA |
3516 | |
3517 | /* | |
3518 | * insert request into driver dispatch list | |
3519 | */ | |
3520 | cfq_dispatch_insert(cfqd->queue, rq); | |
3521 | ||
3522 | if (!cfqd->active_cic) { | |
c5869807 | 3523 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 3524 | |
c5869807 | 3525 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
3526 | cfqd->active_cic = cic; |
3527 | } | |
3528 | ||
3529 | return true; | |
3530 | } | |
3531 | ||
3532 | /* | |
3533 | * Find the cfqq that we need to service and move a request from that to the | |
3534 | * dispatch list | |
3535 | */ | |
3536 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
3537 | { | |
3538 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3539 | struct cfq_queue *cfqq; | |
3540 | ||
3541 | if (!cfqd->busy_queues) | |
3542 | return 0; | |
3543 | ||
3544 | if (unlikely(force)) | |
3545 | return cfq_forced_dispatch(cfqd); | |
3546 | ||
3547 | cfqq = cfq_select_queue(cfqd); | |
3548 | if (!cfqq) | |
8e296755 JA |
3549 | return 0; |
3550 | ||
2f5cb738 | 3551 | /* |
0b182d61 | 3552 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 3553 | */ |
0b182d61 JA |
3554 | if (!cfq_dispatch_request(cfqd, cfqq)) |
3555 | return 0; | |
3556 | ||
2f5cb738 | 3557 | cfqq->slice_dispatch++; |
b029195d | 3558 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 3559 | |
2f5cb738 JA |
3560 | /* |
3561 | * expire an async queue immediately if it has used up its slice. idle | |
3562 | * queue always expire after 1 dispatch round. | |
3563 | */ | |
3564 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
3565 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
3566 | cfq_class_idle(cfqq))) { | |
9a7f38c4 | 3567 | cfqq->slice_end = ktime_get_ns() + 1; |
e5ff082e | 3568 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
3569 | } |
3570 | ||
b217a903 | 3571 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 3572 | return 1; |
1da177e4 LT |
3573 | } |
3574 | ||
1da177e4 | 3575 | /* |
5e705374 JA |
3576 | * task holds one reference to the queue, dropped when task exits. each rq |
3577 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 3578 | * |
b1c35769 | 3579 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
3580 | * queue lock must be held here. |
3581 | */ | |
3582 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
3583 | { | |
22e2c507 | 3584 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 3585 | struct cfq_group *cfqg; |
22e2c507 | 3586 | |
30d7b944 | 3587 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 3588 | |
30d7b944 SL |
3589 | cfqq->ref--; |
3590 | if (cfqq->ref) | |
1da177e4 LT |
3591 | return; |
3592 | ||
7b679138 | 3593 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 3594 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 3595 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 3596 | cfqg = cfqq->cfqg; |
1da177e4 | 3597 | |
28f95cbc | 3598 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 3599 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 3600 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 3601 | } |
22e2c507 | 3602 | |
f04a6424 | 3603 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 3604 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 3605 | cfqg_put(cfqg); |
1da177e4 LT |
3606 | } |
3607 | ||
d02a2c07 | 3608 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 3609 | { |
df5fe3e8 JM |
3610 | struct cfq_queue *__cfqq, *next; |
3611 | ||
df5fe3e8 JM |
3612 | /* |
3613 | * If this queue was scheduled to merge with another queue, be | |
3614 | * sure to drop the reference taken on that queue (and others in | |
3615 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
3616 | */ | |
3617 | __cfqq = cfqq->new_cfqq; | |
3618 | while (__cfqq) { | |
3619 | if (__cfqq == cfqq) { | |
3620 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
3621 | break; | |
3622 | } | |
3623 | next = __cfqq->new_cfqq; | |
3624 | cfq_put_queue(__cfqq); | |
3625 | __cfqq = next; | |
3626 | } | |
d02a2c07 SL |
3627 | } |
3628 | ||
3629 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3630 | { | |
3631 | if (unlikely(cfqq == cfqd->active_queue)) { | |
3632 | __cfq_slice_expired(cfqd, cfqq, 0); | |
3633 | cfq_schedule_dispatch(cfqd); | |
3634 | } | |
3635 | ||
3636 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 3637 | |
89850f7e JA |
3638 | cfq_put_queue(cfqq); |
3639 | } | |
22e2c507 | 3640 | |
9b84cacd TH |
3641 | static void cfq_init_icq(struct io_cq *icq) |
3642 | { | |
3643 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
3644 | ||
9a7f38c4 | 3645 | cic->ttime.last_end_request = ktime_get_ns(); |
9b84cacd TH |
3646 | } |
3647 | ||
c5869807 | 3648 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 3649 | { |
c5869807 | 3650 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 3651 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 3652 | |
563180a4 TH |
3653 | if (cic_to_cfqq(cic, false)) { |
3654 | cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, false)); | |
3655 | cic_set_cfqq(cic, NULL, false); | |
12a05732 AV |
3656 | } |
3657 | ||
563180a4 TH |
3658 | if (cic_to_cfqq(cic, true)) { |
3659 | cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, true)); | |
3660 | cic_set_cfqq(cic, NULL, true); | |
12a05732 | 3661 | } |
89850f7e JA |
3662 | } |
3663 | ||
abede6da | 3664 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
3665 | { |
3666 | struct task_struct *tsk = current; | |
3667 | int ioprio_class; | |
3668 | ||
3b18152c | 3669 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
3670 | return; |
3671 | ||
598971bf | 3672 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 3673 | switch (ioprio_class) { |
fe094d98 JA |
3674 | default: |
3675 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
3676 | case IOPRIO_CLASS_NONE: | |
3677 | /* | |
6d63c275 | 3678 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
3679 | */ |
3680 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 3681 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
3682 | break; |
3683 | case IOPRIO_CLASS_RT: | |
598971bf | 3684 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3685 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
3686 | break; | |
3687 | case IOPRIO_CLASS_BE: | |
598971bf | 3688 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3689 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3690 | break; | |
3691 | case IOPRIO_CLASS_IDLE: | |
3692 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
3693 | cfqq->ioprio = 7; | |
3694 | cfq_clear_cfqq_idle_window(cfqq); | |
3695 | break; | |
22e2c507 JA |
3696 | } |
3697 | ||
3698 | /* | |
3699 | * keep track of original prio settings in case we have to temporarily | |
3700 | * elevate the priority of this queue | |
3701 | */ | |
3702 | cfqq->org_ioprio = cfqq->ioprio; | |
b8269db4 | 3703 | cfqq->org_ioprio_class = cfqq->ioprio_class; |
3b18152c | 3704 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
3705 | } |
3706 | ||
598971bf | 3707 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 3708 | { |
598971bf | 3709 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 3710 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 3711 | struct cfq_queue *cfqq; |
35e6077c | 3712 | |
598971bf TH |
3713 | /* |
3714 | * Check whether ioprio has changed. The condition may trigger | |
3715 | * spuriously on a newly created cic but there's no harm. | |
3716 | */ | |
3717 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
3718 | return; |
3719 | ||
563180a4 | 3720 | cfqq = cic_to_cfqq(cic, false); |
caaa5f9f | 3721 | if (cfqq) { |
563180a4 | 3722 | cfq_put_queue(cfqq); |
2da8de0b | 3723 | cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio); |
563180a4 | 3724 | cic_set_cfqq(cic, cfqq, false); |
22e2c507 | 3725 | } |
caaa5f9f | 3726 | |
563180a4 | 3727 | cfqq = cic_to_cfqq(cic, true); |
caaa5f9f JA |
3728 | if (cfqq) |
3729 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
3730 | |
3731 | cic->ioprio = ioprio; | |
22e2c507 JA |
3732 | } |
3733 | ||
d5036d77 | 3734 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 3735 | pid_t pid, bool is_sync) |
d5036d77 JA |
3736 | { |
3737 | RB_CLEAR_NODE(&cfqq->rb_node); | |
3738 | RB_CLEAR_NODE(&cfqq->p_node); | |
3739 | INIT_LIST_HEAD(&cfqq->fifo); | |
3740 | ||
30d7b944 | 3741 | cfqq->ref = 0; |
d5036d77 JA |
3742 | cfqq->cfqd = cfqd; |
3743 | ||
3744 | cfq_mark_cfqq_prio_changed(cfqq); | |
3745 | ||
3746 | if (is_sync) { | |
3747 | if (!cfq_class_idle(cfqq)) | |
3748 | cfq_mark_cfqq_idle_window(cfqq); | |
3749 | cfq_mark_cfqq_sync(cfqq); | |
3750 | } | |
3751 | cfqq->pid = pid; | |
3752 | } | |
3753 | ||
24610333 | 3754 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
142bbdfc | 3755 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 3756 | { |
bca4b914 | 3757 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
60a83707 | 3758 | struct cfq_queue *cfqq; |
f4da8072 | 3759 | uint64_t serial_nr; |
24610333 | 3760 | |
598971bf | 3761 | rcu_read_lock(); |
f4da8072 | 3762 | serial_nr = bio_blkcg(bio)->css.serial_nr; |
598971bf | 3763 | rcu_read_unlock(); |
24610333 | 3764 | |
598971bf TH |
3765 | /* |
3766 | * Check whether blkcg has changed. The condition may trigger | |
3767 | * spuriously on a newly created cic but there's no harm. | |
3768 | */ | |
f4da8072 | 3769 | if (unlikely(!cfqd) || likely(cic->blkcg_serial_nr == serial_nr)) |
142bbdfc | 3770 | return; |
87760e5e | 3771 | |
60a83707 TH |
3772 | /* |
3773 | * Drop reference to queues. New queues will be assigned in new | |
3774 | * group upon arrival of fresh requests. | |
3775 | */ | |
3776 | cfqq = cic_to_cfqq(cic, false); | |
3777 | if (cfqq) { | |
3778 | cfq_log_cfqq(cfqd, cfqq, "changed cgroup"); | |
3779 | cic_set_cfqq(cic, NULL, false); | |
3780 | cfq_put_queue(cfqq); | |
3781 | } | |
3782 | ||
3783 | cfqq = cic_to_cfqq(cic, true); | |
3784 | if (cfqq) { | |
3785 | cfq_log_cfqq(cfqd, cfqq, "changed cgroup"); | |
3786 | cic_set_cfqq(cic, NULL, true); | |
3787 | cfq_put_queue(cfqq); | |
24610333 | 3788 | } |
598971bf | 3789 | |
f4da8072 | 3790 | cic->blkcg_serial_nr = serial_nr; |
24610333 | 3791 | } |
598971bf | 3792 | #else |
142bbdfc | 3793 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
5d7f5ce1 | 3794 | { |
5d7f5ce1 | 3795 | } |
24610333 VG |
3796 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
3797 | ||
c2dea2d1 | 3798 | static struct cfq_queue ** |
60a83707 | 3799 | cfq_async_queue_prio(struct cfq_group *cfqg, int ioprio_class, int ioprio) |
c2dea2d1 | 3800 | { |
fe094d98 | 3801 | switch (ioprio_class) { |
c2dea2d1 | 3802 | case IOPRIO_CLASS_RT: |
60a83707 | 3803 | return &cfqg->async_cfqq[0][ioprio]; |
598971bf TH |
3804 | case IOPRIO_CLASS_NONE: |
3805 | ioprio = IOPRIO_NORM; | |
3806 | /* fall through */ | |
c2dea2d1 | 3807 | case IOPRIO_CLASS_BE: |
60a83707 | 3808 | return &cfqg->async_cfqq[1][ioprio]; |
c2dea2d1 | 3809 | case IOPRIO_CLASS_IDLE: |
60a83707 | 3810 | return &cfqg->async_idle_cfqq; |
c2dea2d1 VT |
3811 | default: |
3812 | BUG(); | |
3813 | } | |
3814 | } | |
3815 | ||
15c31be4 | 3816 | static struct cfq_queue * |
abede6da | 3817 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
2da8de0b | 3818 | struct bio *bio) |
15c31be4 | 3819 | { |
c6ce1943 JM |
3820 | int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
3821 | int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
d4aad7ff | 3822 | struct cfq_queue **async_cfqq = NULL; |
4ebc1c61 | 3823 | struct cfq_queue *cfqq; |
322731ed TH |
3824 | struct cfq_group *cfqg; |
3825 | ||
3826 | rcu_read_lock(); | |
ae118896 | 3827 | cfqg = cfq_lookup_cfqg(cfqd, bio_blkcg(bio)); |
322731ed TH |
3828 | if (!cfqg) { |
3829 | cfqq = &cfqd->oom_cfqq; | |
3830 | goto out; | |
3831 | } | |
15c31be4 | 3832 | |
c2dea2d1 | 3833 | if (!is_sync) { |
c6ce1943 JM |
3834 | if (!ioprio_valid(cic->ioprio)) { |
3835 | struct task_struct *tsk = current; | |
3836 | ioprio = task_nice_ioprio(tsk); | |
3837 | ioprio_class = task_nice_ioclass(tsk); | |
3838 | } | |
60a83707 | 3839 | async_cfqq = cfq_async_queue_prio(cfqg, ioprio_class, ioprio); |
c2dea2d1 | 3840 | cfqq = *async_cfqq; |
4ebc1c61 TH |
3841 | if (cfqq) |
3842 | goto out; | |
c2dea2d1 VT |
3843 | } |
3844 | ||
e00f4f4d TH |
3845 | cfqq = kmem_cache_alloc_node(cfq_pool, |
3846 | GFP_NOWAIT | __GFP_ZERO | __GFP_NOWARN, | |
d4aad7ff TH |
3847 | cfqd->queue->node); |
3848 | if (!cfqq) { | |
3849 | cfqq = &cfqd->oom_cfqq; | |
3850 | goto out; | |
3851 | } | |
3852 | ||
4d608baa AP |
3853 | /* cfq_init_cfqq() assumes cfqq->ioprio_class is initialized. */ |
3854 | cfqq->ioprio_class = IOPRIO_CLASS_NONE; | |
d4aad7ff TH |
3855 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); |
3856 | cfq_init_prio_data(cfqq, cic); | |
3857 | cfq_link_cfqq_cfqg(cfqq, cfqg); | |
3858 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | |
15c31be4 | 3859 | |
d4aad7ff TH |
3860 | if (async_cfqq) { |
3861 | /* a new async queue is created, pin and remember */ | |
30d7b944 | 3862 | cfqq->ref++; |
c2dea2d1 | 3863 | *async_cfqq = cfqq; |
15c31be4 | 3864 | } |
4ebc1c61 | 3865 | out: |
30d7b944 | 3866 | cfqq->ref++; |
322731ed | 3867 | rcu_read_unlock(); |
15c31be4 JA |
3868 | return cfqq; |
3869 | } | |
3870 | ||
22e2c507 | 3871 | static void |
9a7f38c4 | 3872 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, u64 slice_idle) |
1da177e4 | 3873 | { |
9a7f38c4 | 3874 | u64 elapsed = ktime_get_ns() - ttime->last_end_request; |
383cd721 | 3875 | elapsed = min(elapsed, 2UL * slice_idle); |
db3b5848 | 3876 | |
383cd721 | 3877 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
9a7f38c4 JM |
3878 | ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed, 8); |
3879 | ttime->ttime_mean = div64_ul(ttime->ttime_total + 128, | |
3880 | ttime->ttime_samples); | |
383cd721 SL |
3881 | } |
3882 | ||
3883 | static void | |
3884 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3885 | struct cfq_io_cq *cic) |
383cd721 | 3886 | { |
f5f2b6ce | 3887 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3888 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3889 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3890 | cfqd->cfq_slice_idle); | |
3891 | } | |
7700fc4f SL |
3892 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3893 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3894 | #endif | |
22e2c507 | 3895 | } |
1da177e4 | 3896 | |
206dc69b | 3897 | static void |
b2c18e1e | 3898 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3899 | struct request *rq) |
206dc69b | 3900 | { |
3dde36dd | 3901 | sector_t sdist = 0; |
41647e7a | 3902 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3903 | if (cfqq->last_request_pos) { |
3904 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3905 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3906 | else | |
3907 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3908 | } | |
206dc69b | 3909 | |
3dde36dd | 3910 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3911 | if (blk_queue_nonrot(cfqd->queue)) |
3912 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3913 | else | |
3914 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3915 | } |
1da177e4 | 3916 | |
a2b80967 CH |
3917 | static inline bool req_noidle(struct request *req) |
3918 | { | |
3919 | return req_op(req) == REQ_OP_WRITE && | |
3920 | (req->cmd_flags & (REQ_SYNC | REQ_IDLE)) == REQ_SYNC; | |
3921 | } | |
3922 | ||
22e2c507 JA |
3923 | /* |
3924 | * Disable idle window if the process thinks too long or seeks so much that | |
3925 | * it doesn't matter | |
3926 | */ | |
3927 | static void | |
3928 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3929 | struct cfq_io_cq *cic) |
22e2c507 | 3930 | { |
7b679138 | 3931 | int old_idle, enable_idle; |
1be92f2f | 3932 | |
0871714e JA |
3933 | /* |
3934 | * Don't idle for async or idle io prio class | |
3935 | */ | |
3936 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3937 | return; |
3938 | ||
c265a7f4 | 3939 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3940 | |
76280aff CZ |
3941 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3942 | cfq_mark_cfqq_deep(cfqq); | |
3943 | ||
a2b80967 | 3944 | if (cfqq->next_rq && req_noidle(cfqq->next_rq)) |
749ef9f8 | 3945 | enable_idle = 0; |
f6e8d01b | 3946 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
3947 | !cfqd->cfq_slice_idle || |
3948 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3949 | enable_idle = 0; |
383cd721 SL |
3950 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3951 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3952 | enable_idle = 0; |
3953 | else | |
3954 | enable_idle = 1; | |
1da177e4 LT |
3955 | } |
3956 | ||
7b679138 JA |
3957 | if (old_idle != enable_idle) { |
3958 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3959 | if (enable_idle) | |
3960 | cfq_mark_cfqq_idle_window(cfqq); | |
3961 | else | |
3962 | cfq_clear_cfqq_idle_window(cfqq); | |
3963 | } | |
22e2c507 | 3964 | } |
1da177e4 | 3965 | |
22e2c507 JA |
3966 | /* |
3967 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3968 | * no or if we aren't sure, a 1 will cause a preempt. | |
3969 | */ | |
a6151c3a | 3970 | static bool |
22e2c507 | 3971 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3972 | struct request *rq) |
22e2c507 | 3973 | { |
6d048f53 | 3974 | struct cfq_queue *cfqq; |
22e2c507 | 3975 | |
6d048f53 JA |
3976 | cfqq = cfqd->active_queue; |
3977 | if (!cfqq) | |
a6151c3a | 3978 | return false; |
22e2c507 | 3979 | |
6d048f53 | 3980 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3981 | return false; |
22e2c507 JA |
3982 | |
3983 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3984 | return true; |
1e3335de | 3985 | |
875feb63 DS |
3986 | /* |
3987 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3988 | */ | |
3989 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3990 | return false; | |
3991 | ||
374f84ac JA |
3992 | /* |
3993 | * if the new request is sync, but the currently running queue is | |
3994 | * not, let the sync request have priority. | |
3995 | */ | |
3932a86b | 3996 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) |
a6151c3a | 3997 | return true; |
1e3335de | 3998 | |
3984aa55 JK |
3999 | /* |
4000 | * Treat ancestors of current cgroup the same way as current cgroup. | |
4001 | * For anybody else we disallow preemption to guarantee service | |
4002 | * fairness among cgroups. | |
4003 | */ | |
4004 | if (!cfqg_is_descendant(cfqq->cfqg, new_cfqq->cfqg)) | |
8682e1f1 VG |
4005 | return false; |
4006 | ||
4007 | if (cfq_slice_used(cfqq)) | |
4008 | return true; | |
4009 | ||
6c80731c JK |
4010 | /* |
4011 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
4012 | */ | |
4013 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
4014 | return true; | |
4015 | ||
4016 | WARN_ON_ONCE(cfqq->ioprio_class != new_cfqq->ioprio_class); | |
8682e1f1 | 4017 | /* Allow preemption only if we are idling on sync-noidle tree */ |
4d2ceea4 | 4018 | if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 | 4019 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 VG |
4020 | RB_EMPTY_ROOT(&cfqq->sort_list)) |
4021 | return true; | |
4022 | ||
b53d1ed7 JA |
4023 | /* |
4024 | * So both queues are sync. Let the new request get disk time if | |
4025 | * it's a metadata request and the current queue is doing regular IO. | |
4026 | */ | |
65299a3b | 4027 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
4028 | return true; |
4029 | ||
d2d59e18 SL |
4030 | /* An idle queue should not be idle now for some reason */ |
4031 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
4032 | return true; | |
4033 | ||
1e3335de | 4034 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 4035 | return false; |
1e3335de JA |
4036 | |
4037 | /* | |
4038 | * if this request is as-good as one we would expect from the | |
4039 | * current cfqq, let it preempt | |
4040 | */ | |
e9ce335d | 4041 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 4042 | return true; |
1e3335de | 4043 | |
a6151c3a | 4044 | return false; |
22e2c507 JA |
4045 | } |
4046 | ||
4047 | /* | |
4048 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
4049 | * let it have half of its nominal slice. | |
4050 | */ | |
4051 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
4052 | { | |
df0793ab SL |
4053 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
4054 | ||
7b679138 | 4055 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 4056 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 4057 | |
f8ae6e3e SL |
4058 | /* |
4059 | * workload type is changed, don't save slice, otherwise preempt | |
4060 | * doesn't happen | |
4061 | */ | |
df0793ab | 4062 | if (old_type != cfqq_type(cfqq)) |
4d2ceea4 | 4063 | cfqq->cfqg->saved_wl_slice = 0; |
f8ae6e3e | 4064 | |
bf572256 JA |
4065 | /* |
4066 | * Put the new queue at the front of the of the current list, | |
4067 | * so we know that it will be selected next. | |
4068 | */ | |
4069 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
4070 | |
4071 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 4072 | |
62a37f6b JT |
4073 | cfqq->slice_end = 0; |
4074 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
4075 | } |
4076 | ||
22e2c507 | 4077 | /* |
5e705374 | 4078 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
4079 | * something we should do about it |
4080 | */ | |
4081 | static void | |
5e705374 JA |
4082 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
4083 | struct request *rq) | |
22e2c507 | 4084 | { |
c5869807 | 4085 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 4086 | |
45333d5a | 4087 | cfqd->rq_queued++; |
65299a3b CH |
4088 | if (rq->cmd_flags & REQ_PRIO) |
4089 | cfqq->prio_pending++; | |
374f84ac | 4090 | |
383cd721 | 4091 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 4092 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
4093 | cfq_update_idle_window(cfqd, cfqq, cic); |
4094 | ||
b2c18e1e | 4095 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
4096 | |
4097 | if (cfqq == cfqd->active_queue) { | |
4098 | /* | |
b029195d JA |
4099 | * Remember that we saw a request from this process, but |
4100 | * don't start queuing just yet. Otherwise we risk seeing lots | |
4101 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
4102 | * and merging. If the request is already larger than a single |
4103 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
4104 | * merging is already done. Ditto for a busy system that |
4105 | * has other work pending, don't risk delaying until the | |
4106 | * idle timer unplug to continue working. | |
22e2c507 | 4107 | */ |
d6ceb25e | 4108 | if (cfq_cfqq_wait_request(cfqq)) { |
09cbfeaf | 4109 | if (blk_rq_bytes(rq) > PAGE_SIZE || |
2d870722 | 4110 | cfqd->busy_queues > 1) { |
812df48d | 4111 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 4112 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 4113 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 4114 | } else { |
155fead9 | 4115 | cfqg_stats_update_idle_time(cfqq->cfqg); |
bf791937 | 4116 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 4117 | } |
d6ceb25e | 4118 | } |
5e705374 | 4119 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
4120 | /* |
4121 | * not the active queue - expire current slice if it is | |
4122 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
4123 | * has some old slice time left and is of higher priority or |
4124 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
4125 | */ |
4126 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 4127 | __blk_run_queue(cfqd->queue); |
22e2c507 | 4128 | } |
1da177e4 LT |
4129 | } |
4130 | ||
165125e1 | 4131 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 4132 | { |
b4878f24 | 4133 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 4134 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 4135 | |
7b679138 | 4136 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 4137 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 4138 | |
9a7f38c4 | 4139 | rq->fifo_time = ktime_get_ns() + cfqd->cfq_fifo_expire[rq_is_sync(rq)]; |
22e2c507 | 4140 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 4141 | cfq_add_rq_rb(rq); |
ef295ecf | 4142 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
155fead9 | 4143 | rq->cmd_flags); |
5e705374 | 4144 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
4145 | } |
4146 | ||
45333d5a AC |
4147 | /* |
4148 | * Update hw_tag based on peak queue depth over 50 samples under | |
4149 | * sufficient load. | |
4150 | */ | |
4151 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
4152 | { | |
1a1238a7 SL |
4153 | struct cfq_queue *cfqq = cfqd->active_queue; |
4154 | ||
53c583d2 CZ |
4155 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
4156 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
4157 | |
4158 | if (cfqd->hw_tag == 1) | |
4159 | return; | |
45333d5a AC |
4160 | |
4161 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 4162 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
4163 | return; |
4164 | ||
1a1238a7 SL |
4165 | /* |
4166 | * If active queue hasn't enough requests and can idle, cfq might not | |
4167 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
4168 | * case | |
4169 | */ | |
4170 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
4171 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 4172 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
4173 | return; |
4174 | ||
45333d5a AC |
4175 | if (cfqd->hw_tag_samples++ < 50) |
4176 | return; | |
4177 | ||
e459dd08 | 4178 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
4179 | cfqd->hw_tag = 1; |
4180 | else | |
4181 | cfqd->hw_tag = 0; | |
45333d5a AC |
4182 | } |
4183 | ||
7667aa06 VG |
4184 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
4185 | { | |
c5869807 | 4186 | struct cfq_io_cq *cic = cfqd->active_cic; |
9a7f38c4 | 4187 | u64 now = ktime_get_ns(); |
7667aa06 | 4188 | |
02a8f01b JT |
4189 | /* If the queue already has requests, don't wait */ |
4190 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
4191 | return false; | |
4192 | ||
7667aa06 VG |
4193 | /* If there are other queues in the group, don't wait */ |
4194 | if (cfqq->cfqg->nr_cfqq > 1) | |
4195 | return false; | |
4196 | ||
7700fc4f SL |
4197 | /* the only queue in the group, but think time is big */ |
4198 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
4199 | return false; | |
4200 | ||
7667aa06 VG |
4201 | if (cfq_slice_used(cfqq)) |
4202 | return true; | |
4203 | ||
4204 | /* if slice left is less than think time, wait busy */ | |
383cd721 | 4205 | if (cic && sample_valid(cic->ttime.ttime_samples) |
9a7f38c4 | 4206 | && (cfqq->slice_end - now < cic->ttime.ttime_mean)) |
7667aa06 VG |
4207 | return true; |
4208 | ||
4209 | /* | |
4210 | * If think times is less than a jiffy than ttime_mean=0 and above | |
4211 | * will not be true. It might happen that slice has not expired yet | |
4212 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
4213 | * case where think time is less than a jiffy, mark the queue wait | |
4214 | * busy if only 1 jiffy is left in the slice. | |
4215 | */ | |
9a7f38c4 | 4216 | if (cfqq->slice_end - now <= jiffies_to_nsecs(1)) |
7667aa06 VG |
4217 | return true; |
4218 | ||
4219 | return false; | |
4220 | } | |
4221 | ||
165125e1 | 4222 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 4223 | { |
5e705374 | 4224 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 4225 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 4226 | const int sync = rq_is_sync(rq); |
9a7f38c4 | 4227 | u64 now = ktime_get_ns(); |
1da177e4 | 4228 | |
a2b80967 | 4229 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", req_noidle(rq)); |
1da177e4 | 4230 | |
45333d5a AC |
4231 | cfq_update_hw_tag(cfqd); |
4232 | ||
53c583d2 | 4233 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 4234 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 4235 | cfqd->rq_in_driver--; |
6d048f53 | 4236 | cfqq->dispatched--; |
80bdf0c7 | 4237 | (RQ_CFQG(rq))->dispatched--; |
155fead9 | 4238 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
ef295ecf | 4239 | rq_io_start_time_ns(rq), rq->cmd_flags); |
1da177e4 | 4240 | |
53c583d2 | 4241 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 4242 | |
365722bb | 4243 | if (sync) { |
34b98d03 | 4244 | struct cfq_rb_root *st; |
f5f2b6ce | 4245 | |
383cd721 | 4246 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
4247 | |
4248 | if (cfq_cfqq_on_rr(cfqq)) | |
34b98d03 | 4249 | st = cfqq->service_tree; |
f5f2b6ce | 4250 | else |
34b98d03 VG |
4251 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), |
4252 | cfqq_type(cfqq)); | |
4253 | ||
4254 | st->ttime.last_end_request = now; | |
149321a6 JK |
4255 | /* |
4256 | * We have to do this check in jiffies since start_time is in | |
4257 | * jiffies and it is not trivial to convert to ns. If | |
4258 | * cfq_fifo_expire[1] ever comes close to 1 jiffie, this test | |
4259 | * will become problematic but so far we are fine (the default | |
4260 | * is 128 ms). | |
4261 | */ | |
4262 | if (!time_after(rq->start_time + | |
4263 | nsecs_to_jiffies(cfqd->cfq_fifo_expire[1]), | |
4264 | jiffies)) | |
573412b2 | 4265 | cfqd->last_delayed_sync = now; |
365722bb | 4266 | } |
caaa5f9f | 4267 | |
7700fc4f SL |
4268 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4269 | cfqq->cfqg->ttime.last_end_request = now; | |
4270 | #endif | |
4271 | ||
caaa5f9f JA |
4272 | /* |
4273 | * If this is the active queue, check if it needs to be expired, | |
4274 | * or if we want to idle in case it has no pending requests. | |
4275 | */ | |
4276 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
4277 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
4278 | ||
44f7c160 JA |
4279 | if (cfq_cfqq_slice_new(cfqq)) { |
4280 | cfq_set_prio_slice(cfqd, cfqq); | |
4281 | cfq_clear_cfqq_slice_new(cfqq); | |
4282 | } | |
f75edf2d VG |
4283 | |
4284 | /* | |
7667aa06 VG |
4285 | * Should we wait for next request to come in before we expire |
4286 | * the queue. | |
f75edf2d | 4287 | */ |
7667aa06 | 4288 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
9a7f38c4 | 4289 | u64 extend_sl = cfqd->cfq_slice_idle; |
80bdf0c7 VG |
4290 | if (!cfqd->cfq_slice_idle) |
4291 | extend_sl = cfqd->cfq_group_idle; | |
9a7f38c4 | 4292 | cfqq->slice_end = now + extend_sl; |
f75edf2d | 4293 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 4294 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
4295 | } |
4296 | ||
a36e71f9 | 4297 | /* |
8e550632 CZ |
4298 | * Idling is not enabled on: |
4299 | * - expired queues | |
4300 | * - idle-priority queues | |
4301 | * - async queues | |
4302 | * - queues with still some requests queued | |
4303 | * - when there is a close cooperator | |
a36e71f9 | 4304 | */ |
0871714e | 4305 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 4306 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
4307 | else if (sync && cfqq_empty && |
4308 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 4309 | cfq_arm_slice_timer(cfqd); |
8e550632 | 4310 | } |
caaa5f9f | 4311 | } |
6d048f53 | 4312 | |
53c583d2 | 4313 | if (!cfqd->rq_in_driver) |
23e018a1 | 4314 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
4315 | } |
4316 | ||
ef295ecf | 4317 | static void cfqq_boost_on_prio(struct cfq_queue *cfqq, unsigned int op) |
b8269db4 JA |
4318 | { |
4319 | /* | |
4320 | * If REQ_PRIO is set, boost class and prio level, if it's below | |
4321 | * BE/NORM. If prio is not set, restore the potentially boosted | |
4322 | * class/prio level. | |
4323 | */ | |
ef295ecf | 4324 | if (!(op & REQ_PRIO)) { |
b8269db4 JA |
4325 | cfqq->ioprio_class = cfqq->org_ioprio_class; |
4326 | cfqq->ioprio = cfqq->org_ioprio; | |
4327 | } else { | |
4328 | if (cfq_class_idle(cfqq)) | |
4329 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
4330 | if (cfqq->ioprio > IOPRIO_NORM) | |
4331 | cfqq->ioprio = IOPRIO_NORM; | |
4332 | } | |
4333 | } | |
4334 | ||
89850f7e | 4335 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 4336 | { |
1b379d8d | 4337 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 4338 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 4339 | return ELV_MQUEUE_MUST; |
3b18152c | 4340 | } |
1da177e4 | 4341 | |
22e2c507 | 4342 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
4343 | } |
4344 | ||
ef295ecf | 4345 | static int cfq_may_queue(struct request_queue *q, unsigned int op) |
22e2c507 JA |
4346 | { |
4347 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
4348 | struct task_struct *tsk = current; | |
c5869807 | 4349 | struct cfq_io_cq *cic; |
22e2c507 JA |
4350 | struct cfq_queue *cfqq; |
4351 | ||
4352 | /* | |
4353 | * don't force setup of a queue from here, as a call to may_queue | |
4354 | * does not necessarily imply that a request actually will be queued. | |
4355 | * so just lookup a possibly existing queue, or return 'may queue' | |
4356 | * if that fails | |
4357 | */ | |
4ac845a2 | 4358 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
4359 | if (!cic) |
4360 | return ELV_MQUEUE_MAY; | |
4361 | ||
ef295ecf | 4362 | cfqq = cic_to_cfqq(cic, op_is_sync(op)); |
22e2c507 | 4363 | if (cfqq) { |
abede6da | 4364 | cfq_init_prio_data(cfqq, cic); |
ef295ecf | 4365 | cfqq_boost_on_prio(cfqq, op); |
22e2c507 | 4366 | |
89850f7e | 4367 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
4368 | } |
4369 | ||
4370 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
4371 | } |
4372 | ||
1da177e4 LT |
4373 | /* |
4374 | * queue lock held here | |
4375 | */ | |
bb37b94c | 4376 | static void cfq_put_request(struct request *rq) |
1da177e4 | 4377 | { |
5e705374 | 4378 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 4379 | |
5e705374 | 4380 | if (cfqq) { |
22e2c507 | 4381 | const int rw = rq_data_dir(rq); |
1da177e4 | 4382 | |
22e2c507 JA |
4383 | BUG_ON(!cfqq->allocated[rw]); |
4384 | cfqq->allocated[rw]--; | |
1da177e4 | 4385 | |
7f1dc8a2 | 4386 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 4387 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
4388 | rq->elv.priv[0] = NULL; |
4389 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 4390 | |
1da177e4 LT |
4391 | cfq_put_queue(cfqq); |
4392 | } | |
4393 | } | |
4394 | ||
df5fe3e8 | 4395 | static struct cfq_queue * |
c5869807 | 4396 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
4397 | struct cfq_queue *cfqq) |
4398 | { | |
4399 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
4400 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 4401 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
4402 | cfq_put_queue(cfqq); |
4403 | return cic_to_cfqq(cic, 1); | |
4404 | } | |
4405 | ||
e6c5bc73 JM |
4406 | /* |
4407 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
4408 | * was the last process referring to said cfqq. | |
4409 | */ | |
4410 | static struct cfq_queue * | |
c5869807 | 4411 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
4412 | { |
4413 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
4414 | cfqq->pid = current->pid; |
4415 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 4416 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
4417 | return cfqq; |
4418 | } | |
4419 | ||
4420 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
4421 | |
4422 | cfq_put_cooperator(cfqq); | |
4423 | ||
e6c5bc73 JM |
4424 | cfq_put_queue(cfqq); |
4425 | return NULL; | |
4426 | } | |
1da177e4 | 4427 | /* |
22e2c507 | 4428 | * Allocate cfq data structures associated with this request. |
1da177e4 | 4429 | */ |
22e2c507 | 4430 | static int |
852c788f TH |
4431 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
4432 | gfp_t gfp_mask) | |
1da177e4 LT |
4433 | { |
4434 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 4435 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 4436 | const int rw = rq_data_dir(rq); |
a6151c3a | 4437 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 4438 | struct cfq_queue *cfqq; |
1da177e4 | 4439 | |
216284c3 | 4440 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 4441 | |
598971bf | 4442 | check_ioprio_changed(cic, bio); |
142bbdfc | 4443 | check_blkcg_changed(cic, bio); |
e6c5bc73 | 4444 | new_queue: |
91fac317 | 4445 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 4446 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
bce6133b TH |
4447 | if (cfqq) |
4448 | cfq_put_queue(cfqq); | |
2da8de0b | 4449 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio); |
91fac317 | 4450 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 4451 | } else { |
e6c5bc73 JM |
4452 | /* |
4453 | * If the queue was seeky for too long, break it apart. | |
4454 | */ | |
ae54abed | 4455 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
4456 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
4457 | cfqq = split_cfqq(cic, cfqq); | |
4458 | if (!cfqq) | |
4459 | goto new_queue; | |
4460 | } | |
4461 | ||
df5fe3e8 JM |
4462 | /* |
4463 | * Check to see if this queue is scheduled to merge with | |
4464 | * another, closely cooperating queue. The merging of | |
4465 | * queues happens here as it must be done in process context. | |
4466 | * The reference on new_cfqq was taken in merge_cfqqs. | |
4467 | */ | |
4468 | if (cfqq->new_cfqq) | |
4469 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 4470 | } |
1da177e4 LT |
4471 | |
4472 | cfqq->allocated[rw]++; | |
1da177e4 | 4473 | |
6fae9c25 | 4474 | cfqq->ref++; |
eb7d8c07 | 4475 | cfqg_get(cfqq->cfqg); |
a612fddf | 4476 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 4477 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 4478 | spin_unlock_irq(q->queue_lock); |
5d7f5ce1 | 4479 | |
5e705374 | 4480 | return 0; |
1da177e4 LT |
4481 | } |
4482 | ||
65f27f38 | 4483 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 4484 | { |
65f27f38 | 4485 | struct cfq_data *cfqd = |
23e018a1 | 4486 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 4487 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4488 | |
40bb54d1 | 4489 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 4490 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 4491 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
4492 | } |
4493 | ||
4494 | /* | |
4495 | * Timer running if the active_queue is currently idling inside its time slice | |
4496 | */ | |
91148325 | 4497 | static enum hrtimer_restart cfq_idle_slice_timer(struct hrtimer *timer) |
22e2c507 | 4498 | { |
91148325 JK |
4499 | struct cfq_data *cfqd = container_of(timer, struct cfq_data, |
4500 | idle_slice_timer); | |
22e2c507 JA |
4501 | struct cfq_queue *cfqq; |
4502 | unsigned long flags; | |
3c6bd2f8 | 4503 | int timed_out = 1; |
22e2c507 | 4504 | |
7b679138 JA |
4505 | cfq_log(cfqd, "idle timer fired"); |
4506 | ||
22e2c507 JA |
4507 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
4508 | ||
fe094d98 JA |
4509 | cfqq = cfqd->active_queue; |
4510 | if (cfqq) { | |
3c6bd2f8 JA |
4511 | timed_out = 0; |
4512 | ||
b029195d JA |
4513 | /* |
4514 | * We saw a request before the queue expired, let it through | |
4515 | */ | |
4516 | if (cfq_cfqq_must_dispatch(cfqq)) | |
4517 | goto out_kick; | |
4518 | ||
22e2c507 JA |
4519 | /* |
4520 | * expired | |
4521 | */ | |
44f7c160 | 4522 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
4523 | goto expire; |
4524 | ||
4525 | /* | |
4526 | * only expire and reinvoke request handler, if there are | |
4527 | * other queues with pending requests | |
4528 | */ | |
caaa5f9f | 4529 | if (!cfqd->busy_queues) |
22e2c507 | 4530 | goto out_cont; |
22e2c507 JA |
4531 | |
4532 | /* | |
4533 | * not expired and it has a request pending, let it dispatch | |
4534 | */ | |
75e50984 | 4535 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 4536 | goto out_kick; |
76280aff CZ |
4537 | |
4538 | /* | |
4539 | * Queue depth flag is reset only when the idle didn't succeed | |
4540 | */ | |
4541 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
4542 | } |
4543 | expire: | |
e5ff082e | 4544 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 4545 | out_kick: |
23e018a1 | 4546 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
4547 | out_cont: |
4548 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
91148325 | 4549 | return HRTIMER_NORESTART; |
22e2c507 JA |
4550 | } |
4551 | ||
3b18152c JA |
4552 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
4553 | { | |
91148325 | 4554 | hrtimer_cancel(&cfqd->idle_slice_timer); |
23e018a1 | 4555 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 4556 | } |
22e2c507 | 4557 | |
b374d18a | 4558 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 4559 | { |
22e2c507 | 4560 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 4561 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4562 | |
3b18152c | 4563 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 4564 | |
d9ff4187 | 4565 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 4566 | |
d9ff4187 | 4567 | if (cfqd->active_queue) |
e5ff082e | 4568 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 4569 | |
03aa264a TH |
4570 | spin_unlock_irq(q->queue_lock); |
4571 | ||
a90d742e AV |
4572 | cfq_shutdown_timer_wq(cfqd); |
4573 | ||
ffea73fc TH |
4574 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4575 | blkcg_deactivate_policy(q, &blkcg_policy_cfq); | |
4576 | #else | |
f51b802c | 4577 | kfree(cfqd->root_group); |
2abae55f | 4578 | #endif |
56edf7d7 | 4579 | kfree(cfqd); |
1da177e4 LT |
4580 | } |
4581 | ||
d50235b7 | 4582 | static int cfq_init_queue(struct request_queue *q, struct elevator_type *e) |
1da177e4 LT |
4583 | { |
4584 | struct cfq_data *cfqd; | |
3c798398 | 4585 | struct blkcg_gq *blkg __maybe_unused; |
a2b1693b | 4586 | int i, ret; |
d50235b7 JM |
4587 | struct elevator_queue *eq; |
4588 | ||
4589 | eq = elevator_alloc(q, e); | |
4590 | if (!eq) | |
4591 | return -ENOMEM; | |
1da177e4 | 4592 | |
c1b511eb | 4593 | cfqd = kzalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node); |
d50235b7 JM |
4594 | if (!cfqd) { |
4595 | kobject_put(&eq->kobj); | |
b2fab5ac | 4596 | return -ENOMEM; |
d50235b7 JM |
4597 | } |
4598 | eq->elevator_data = cfqd; | |
80b15c73 | 4599 | |
f51b802c | 4600 | cfqd->queue = q; |
d50235b7 JM |
4601 | spin_lock_irq(q->queue_lock); |
4602 | q->elevator = eq; | |
4603 | spin_unlock_irq(q->queue_lock); | |
f51b802c | 4604 | |
1fa8f6d6 VG |
4605 | /* Init root service tree */ |
4606 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
4607 | ||
f51b802c | 4608 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 4609 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4610 | ret = blkcg_activate_policy(q, &blkcg_policy_cfq); |
a2b1693b TH |
4611 | if (ret) |
4612 | goto out_free; | |
f51b802c | 4613 | |
a2b1693b | 4614 | cfqd->root_group = blkg_to_cfqg(q->root_blkg); |
f51b802c | 4615 | #else |
a2b1693b | 4616 | ret = -ENOMEM; |
f51b802c TH |
4617 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
4618 | GFP_KERNEL, cfqd->queue->node); | |
a2b1693b TH |
4619 | if (!cfqd->root_group) |
4620 | goto out_free; | |
5624a4e4 | 4621 | |
a2b1693b | 4622 | cfq_init_cfqg_base(cfqd->root_group); |
3ecca629 TH |
4623 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_LEGACY_DFL; |
4624 | cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_LEGACY_DFL; | |
69d7fde5 | 4625 | #endif |
5624a4e4 | 4626 | |
26a2ac00 JA |
4627 | /* |
4628 | * Not strictly needed (since RB_ROOT just clears the node and we | |
4629 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
4630 | * to add magic to the rb code | |
4631 | */ | |
4632 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
4633 | cfqd->prio_trees[i] = RB_ROOT; | |
4634 | ||
6118b70b | 4635 | /* |
d4aad7ff | 4636 | * Our fallback cfqq if cfq_get_queue() runs into OOM issues. |
6118b70b | 4637 | * Grab a permanent reference to it, so that the normal code flow |
f51b802c TH |
4638 | * will not attempt to free it. oom_cfqq is linked to root_group |
4639 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
4640 | * the reference from linking right away. | |
6118b70b JA |
4641 | */ |
4642 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 4643 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
4644 | |
4645 | spin_lock_irq(q->queue_lock); | |
f51b802c | 4646 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 4647 | cfqg_put(cfqd->root_group); |
1adaf3dd | 4648 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 4649 | |
91148325 JK |
4650 | hrtimer_init(&cfqd->idle_slice_timer, CLOCK_MONOTONIC, |
4651 | HRTIMER_MODE_REL); | |
22e2c507 | 4652 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; |
22e2c507 | 4653 | |
23e018a1 | 4654 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 4655 | |
1da177e4 | 4656 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
4657 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
4658 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
4659 | cfqd->cfq_back_max = cfq_back_max; |
4660 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
4661 | cfqd->cfq_slice[0] = cfq_slice_async; |
4662 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
5bf14c07 | 4663 | cfqd->cfq_target_latency = cfq_target_latency; |
22e2c507 | 4664 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
0bb97947 | 4665 | cfqd->cfq_slice_idle = cfq_slice_idle; |
80bdf0c7 | 4666 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 4667 | cfqd->cfq_latency = 1; |
e459dd08 | 4668 | cfqd->hw_tag = -1; |
edc71131 CZ |
4669 | /* |
4670 | * we optimistically start assuming sync ops weren't delayed in last | |
4671 | * second, in order to have larger depth for async operations. | |
4672 | */ | |
9a7f38c4 | 4673 | cfqd->last_delayed_sync = ktime_get_ns() - NSEC_PER_SEC; |
b2fab5ac | 4674 | return 0; |
a2b1693b TH |
4675 | |
4676 | out_free: | |
4677 | kfree(cfqd); | |
d50235b7 | 4678 | kobject_put(&eq->kobj); |
a2b1693b | 4679 | return ret; |
1da177e4 LT |
4680 | } |
4681 | ||
0bb97947 JA |
4682 | static void cfq_registered_queue(struct request_queue *q) |
4683 | { | |
4684 | struct elevator_queue *e = q->elevator; | |
4685 | struct cfq_data *cfqd = e->elevator_data; | |
4686 | ||
4687 | /* | |
4688 | * Default to IOPS mode with no idling for SSDs | |
4689 | */ | |
4690 | if (blk_queue_nonrot(q)) | |
4691 | cfqd->cfq_slice_idle = 0; | |
142bbdfc | 4692 | wbt_disable_default(q); |
0bb97947 JA |
4693 | } |
4694 | ||
1da177e4 LT |
4695 | /* |
4696 | * sysfs parts below --> | |
4697 | */ | |
1da177e4 LT |
4698 | static ssize_t |
4699 | cfq_var_show(unsigned int var, char *page) | |
4700 | { | |
176167ad | 4701 | return sprintf(page, "%u\n", var); |
1da177e4 LT |
4702 | } |
4703 | ||
4704 | static ssize_t | |
4705 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
4706 | { | |
4707 | char *p = (char *) page; | |
4708 | ||
4709 | *var = simple_strtoul(p, &p, 10); | |
4710 | return count; | |
4711 | } | |
4712 | ||
1da177e4 | 4713 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 4714 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 4715 | { \ |
3d1ab40f | 4716 | struct cfq_data *cfqd = e->elevator_data; \ |
9a7f38c4 | 4717 | u64 __data = __VAR; \ |
1da177e4 | 4718 | if (__CONV) \ |
9a7f38c4 | 4719 | __data = div_u64(__data, NSEC_PER_MSEC); \ |
1da177e4 LT |
4720 | return cfq_var_show(__data, (page)); \ |
4721 | } | |
4722 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
4723 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
4724 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
4725 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
4726 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 4727 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 4728 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
4729 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4730 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4731 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4732 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
5bf14c07 | 4733 | SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1); |
1da177e4 LT |
4734 | #undef SHOW_FUNCTION |
4735 | ||
d2d481d0 JM |
4736 | #define USEC_SHOW_FUNCTION(__FUNC, __VAR) \ |
4737 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ | |
4738 | { \ | |
4739 | struct cfq_data *cfqd = e->elevator_data; \ | |
4740 | u64 __data = __VAR; \ | |
4741 | __data = div_u64(__data, NSEC_PER_USEC); \ | |
4742 | return cfq_var_show(__data, (page)); \ | |
4743 | } | |
4744 | USEC_SHOW_FUNCTION(cfq_slice_idle_us_show, cfqd->cfq_slice_idle); | |
4745 | USEC_SHOW_FUNCTION(cfq_group_idle_us_show, cfqd->cfq_group_idle); | |
4746 | USEC_SHOW_FUNCTION(cfq_slice_sync_us_show, cfqd->cfq_slice[1]); | |
4747 | USEC_SHOW_FUNCTION(cfq_slice_async_us_show, cfqd->cfq_slice[0]); | |
4748 | USEC_SHOW_FUNCTION(cfq_target_latency_us_show, cfqd->cfq_target_latency); | |
4749 | #undef USEC_SHOW_FUNCTION | |
4750 | ||
1da177e4 | 4751 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ |
b374d18a | 4752 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4753 | { \ |
3d1ab40f | 4754 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4755 | unsigned int __data; \ |
4756 | int ret = cfq_var_store(&__data, (page), count); \ | |
4757 | if (__data < (MIN)) \ | |
4758 | __data = (MIN); \ | |
4759 | else if (__data > (MAX)) \ | |
4760 | __data = (MAX); \ | |
4761 | if (__CONV) \ | |
9a7f38c4 | 4762 | *(__PTR) = (u64)__data * NSEC_PER_MSEC; \ |
1da177e4 LT |
4763 | else \ |
4764 | *(__PTR) = __data; \ | |
4765 | return ret; \ | |
4766 | } | |
4767 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4768 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4769 | UINT_MAX, 1); | |
4770 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4771 | UINT_MAX, 1); | |
e572ec7e | 4772 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4773 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4774 | UINT_MAX, 0); | |
22e2c507 | 4775 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4776 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4777 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4778 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4779 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4780 | UINT_MAX, 0); | |
963b72fc | 4781 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
5bf14c07 | 4782 | STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1); |
1da177e4 LT |
4783 | #undef STORE_FUNCTION |
4784 | ||
d2d481d0 JM |
4785 | #define USEC_STORE_FUNCTION(__FUNC, __PTR, MIN, MAX) \ |
4786 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ | |
4787 | { \ | |
4788 | struct cfq_data *cfqd = e->elevator_data; \ | |
4789 | unsigned int __data; \ | |
4790 | int ret = cfq_var_store(&__data, (page), count); \ | |
4791 | if (__data < (MIN)) \ | |
4792 | __data = (MIN); \ | |
4793 | else if (__data > (MAX)) \ | |
4794 | __data = (MAX); \ | |
4795 | *(__PTR) = (u64)__data * NSEC_PER_USEC; \ | |
4796 | return ret; \ | |
4797 | } | |
4798 | USEC_STORE_FUNCTION(cfq_slice_idle_us_store, &cfqd->cfq_slice_idle, 0, UINT_MAX); | |
4799 | USEC_STORE_FUNCTION(cfq_group_idle_us_store, &cfqd->cfq_group_idle, 0, UINT_MAX); | |
4800 | USEC_STORE_FUNCTION(cfq_slice_sync_us_store, &cfqd->cfq_slice[1], 1, UINT_MAX); | |
4801 | USEC_STORE_FUNCTION(cfq_slice_async_us_store, &cfqd->cfq_slice[0], 1, UINT_MAX); | |
4802 | USEC_STORE_FUNCTION(cfq_target_latency_us_store, &cfqd->cfq_target_latency, 1, UINT_MAX); | |
4803 | #undef USEC_STORE_FUNCTION | |
4804 | ||
e572ec7e AV |
4805 | #define CFQ_ATTR(name) \ |
4806 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4807 | ||
4808 | static struct elv_fs_entry cfq_attrs[] = { | |
4809 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4810 | CFQ_ATTR(fifo_expire_sync), |
4811 | CFQ_ATTR(fifo_expire_async), | |
4812 | CFQ_ATTR(back_seek_max), | |
4813 | CFQ_ATTR(back_seek_penalty), | |
4814 | CFQ_ATTR(slice_sync), | |
d2d481d0 | 4815 | CFQ_ATTR(slice_sync_us), |
e572ec7e | 4816 | CFQ_ATTR(slice_async), |
d2d481d0 | 4817 | CFQ_ATTR(slice_async_us), |
e572ec7e AV |
4818 | CFQ_ATTR(slice_async_rq), |
4819 | CFQ_ATTR(slice_idle), | |
d2d481d0 | 4820 | CFQ_ATTR(slice_idle_us), |
80bdf0c7 | 4821 | CFQ_ATTR(group_idle), |
d2d481d0 | 4822 | CFQ_ATTR(group_idle_us), |
963b72fc | 4823 | CFQ_ATTR(low_latency), |
5bf14c07 | 4824 | CFQ_ATTR(target_latency), |
d2d481d0 | 4825 | CFQ_ATTR(target_latency_us), |
e572ec7e | 4826 | __ATTR_NULL |
1da177e4 LT |
4827 | }; |
4828 | ||
1da177e4 | 4829 | static struct elevator_type iosched_cfq = { |
c51ca6cf | 4830 | .ops.sq = { |
1da177e4 LT |
4831 | .elevator_merge_fn = cfq_merge, |
4832 | .elevator_merged_fn = cfq_merged_request, | |
4833 | .elevator_merge_req_fn = cfq_merged_requests, | |
72ef799b TE |
4834 | .elevator_allow_bio_merge_fn = cfq_allow_bio_merge, |
4835 | .elevator_allow_rq_merge_fn = cfq_allow_rq_merge, | |
812d4026 | 4836 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4837 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4838 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4839 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4840 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4841 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4842 | .elevator_former_req_fn = elv_rb_former_request, |
4843 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 4844 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 4845 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
4846 | .elevator_set_req_fn = cfq_set_request, |
4847 | .elevator_put_req_fn = cfq_put_request, | |
4848 | .elevator_may_queue_fn = cfq_may_queue, | |
4849 | .elevator_init_fn = cfq_init_queue, | |
4850 | .elevator_exit_fn = cfq_exit_queue, | |
0bb97947 | 4851 | .elevator_registered_fn = cfq_registered_queue, |
1da177e4 | 4852 | }, |
3d3c2379 TH |
4853 | .icq_size = sizeof(struct cfq_io_cq), |
4854 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 4855 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 4856 | .elevator_name = "cfq", |
1da177e4 LT |
4857 | .elevator_owner = THIS_MODULE, |
4858 | }; | |
4859 | ||
3e252066 | 4860 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4861 | static struct blkcg_policy blkcg_policy_cfq = { |
2ee867dc | 4862 | .dfl_cftypes = cfq_blkcg_files, |
880f50e2 | 4863 | .legacy_cftypes = cfq_blkcg_legacy_files, |
f9fcc2d3 | 4864 | |
e4a9bde9 | 4865 | .cpd_alloc_fn = cfq_cpd_alloc, |
e48453c3 | 4866 | .cpd_init_fn = cfq_cpd_init, |
e4a9bde9 | 4867 | .cpd_free_fn = cfq_cpd_free, |
69d7fde5 | 4868 | .cpd_bind_fn = cfq_cpd_bind, |
e4a9bde9 | 4869 | |
001bea73 | 4870 | .pd_alloc_fn = cfq_pd_alloc, |
f9fcc2d3 | 4871 | .pd_init_fn = cfq_pd_init, |
0b39920b | 4872 | .pd_offline_fn = cfq_pd_offline, |
001bea73 | 4873 | .pd_free_fn = cfq_pd_free, |
f9fcc2d3 | 4874 | .pd_reset_stats_fn = cfq_pd_reset_stats, |
3e252066 | 4875 | }; |
3e252066 VG |
4876 | #endif |
4877 | ||
1da177e4 LT |
4878 | static int __init cfq_init(void) |
4879 | { | |
3d3c2379 TH |
4880 | int ret; |
4881 | ||
80bdf0c7 | 4882 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4883 | ret = blkcg_policy_register(&blkcg_policy_cfq); |
8bd435b3 TH |
4884 | if (ret) |
4885 | return ret; | |
ffea73fc TH |
4886 | #else |
4887 | cfq_group_idle = 0; | |
4888 | #endif | |
8bd435b3 | 4889 | |
fd794956 | 4890 | ret = -ENOMEM; |
3d3c2379 TH |
4891 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4892 | if (!cfq_pool) | |
8bd435b3 | 4893 | goto err_pol_unreg; |
1da177e4 | 4894 | |
3d3c2379 | 4895 | ret = elv_register(&iosched_cfq); |
8bd435b3 TH |
4896 | if (ret) |
4897 | goto err_free_pool; | |
3d3c2379 | 4898 | |
2fdd82bd | 4899 | return 0; |
8bd435b3 TH |
4900 | |
4901 | err_free_pool: | |
4902 | kmem_cache_destroy(cfq_pool); | |
4903 | err_pol_unreg: | |
ffea73fc | 4904 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4905 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4906 | #endif |
8bd435b3 | 4907 | return ret; |
1da177e4 LT |
4908 | } |
4909 | ||
4910 | static void __exit cfq_exit(void) | |
4911 | { | |
ffea73fc | 4912 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4913 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4914 | #endif |
1da177e4 | 4915 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4916 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4917 | } |
4918 | ||
4919 | module_init(cfq_init); | |
4920 | module_exit(cfq_exit); | |
4921 | ||
4922 | MODULE_AUTHOR("Jens Axboe"); | |
4923 | MODULE_LICENSE("GPL"); | |
4924 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |