Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * CFQ, or complete fairness queueing, disk scheduler. |
3 | * | |
4 | * Based on ideas from a previously unfinished io | |
5 | * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli. | |
6 | * | |
0fe23479 | 7 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
1da177e4 | 8 | */ |
1da177e4 | 9 | #include <linux/module.h> |
5a0e3ad6 | 10 | #include <linux/slab.h> |
e6017571 | 11 | #include <linux/sched/clock.h> |
1cc9be68 AV |
12 | #include <linux/blkdev.h> |
13 | #include <linux/elevator.h> | |
9a7f38c4 | 14 | #include <linux/ktime.h> |
1da177e4 | 15 | #include <linux/rbtree.h> |
22e2c507 | 16 | #include <linux/ioprio.h> |
7b679138 | 17 | #include <linux/blktrace_api.h> |
eea8f41c | 18 | #include <linux/blk-cgroup.h> |
6e736be7 | 19 | #include "blk.h" |
87760e5e | 20 | #include "blk-wbt.h" |
1da177e4 LT |
21 | |
22 | /* | |
23 | * tunables | |
24 | */ | |
fe094d98 | 25 | /* max queue in one round of service */ |
abc3c744 | 26 | static const int cfq_quantum = 8; |
9a7f38c4 | 27 | static const u64 cfq_fifo_expire[2] = { NSEC_PER_SEC / 4, NSEC_PER_SEC / 8 }; |
fe094d98 JA |
28 | /* maximum backwards seek, in KiB */ |
29 | static const int cfq_back_max = 16 * 1024; | |
30 | /* penalty of a backwards seek */ | |
31 | static const int cfq_back_penalty = 2; | |
9a7f38c4 JM |
32 | static const u64 cfq_slice_sync = NSEC_PER_SEC / 10; |
33 | static u64 cfq_slice_async = NSEC_PER_SEC / 25; | |
64100099 | 34 | static const int cfq_slice_async_rq = 2; |
9a7f38c4 JM |
35 | static u64 cfq_slice_idle = NSEC_PER_SEC / 125; |
36 | static u64 cfq_group_idle = NSEC_PER_SEC / 125; | |
37 | static const u64 cfq_target_latency = (u64)NSEC_PER_SEC * 3/10; /* 300 ms */ | |
5db5d642 | 38 | static const int cfq_hist_divisor = 4; |
22e2c507 | 39 | |
d9e7620e | 40 | /* |
5be6b756 | 41 | * offset from end of queue service tree for idle class |
d9e7620e | 42 | */ |
9a7f38c4 | 43 | #define CFQ_IDLE_DELAY (NSEC_PER_SEC / 5) |
5be6b756 HT |
44 | /* offset from end of group service tree under time slice mode */ |
45 | #define CFQ_SLICE_MODE_GROUP_DELAY (NSEC_PER_SEC / 5) | |
46 | /* offset from end of group service under IOPS mode */ | |
47 | #define CFQ_IOPS_MODE_GROUP_DELAY (HZ / 5) | |
d9e7620e JA |
48 | |
49 | /* | |
50 | * below this threshold, we consider thinktime immediate | |
51 | */ | |
9a7f38c4 | 52 | #define CFQ_MIN_TT (2 * NSEC_PER_SEC / HZ) |
d9e7620e | 53 | |
22e2c507 | 54 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 55 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 56 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 57 | |
3dde36dd | 58 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 59 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 60 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 61 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 62 | |
a612fddf TH |
63 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
64 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
65 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 66 | |
e18b890b | 67 | static struct kmem_cache *cfq_pool; |
1da177e4 | 68 | |
22e2c507 JA |
69 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
70 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
71 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
72 | ||
206dc69b | 73 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 74 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 75 | |
e48453c3 | 76 | /* blkio-related constants */ |
3ecca629 TH |
77 | #define CFQ_WEIGHT_LEGACY_MIN 10 |
78 | #define CFQ_WEIGHT_LEGACY_DFL 500 | |
79 | #define CFQ_WEIGHT_LEGACY_MAX 1000 | |
e48453c3 | 80 | |
c5869807 | 81 | struct cfq_ttime { |
9a7f38c4 | 82 | u64 last_end_request; |
c5869807 | 83 | |
9a7f38c4 JM |
84 | u64 ttime_total; |
85 | u64 ttime_mean; | |
c5869807 | 86 | unsigned long ttime_samples; |
c5869807 TH |
87 | }; |
88 | ||
cc09e299 JA |
89 | /* |
90 | * Most of our rbtree usage is for sorting with min extraction, so | |
91 | * if we cache the leftmost node we don't have to walk down the tree | |
92 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
93 | * move this into the elevator for the rq sorting as well. | |
94 | */ | |
95 | struct cfq_rb_root { | |
96 | struct rb_root rb; | |
97 | struct rb_node *left; | |
aa6f6a3d | 98 | unsigned count; |
1fa8f6d6 | 99 | u64 min_vdisktime; |
f5f2b6ce | 100 | struct cfq_ttime ttime; |
cc09e299 | 101 | }; |
f5f2b6ce | 102 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
9a7f38c4 | 103 | .ttime = {.last_end_request = ktime_get_ns(),},} |
cc09e299 | 104 | |
6118b70b JA |
105 | /* |
106 | * Per process-grouping structure | |
107 | */ | |
108 | struct cfq_queue { | |
109 | /* reference count */ | |
30d7b944 | 110 | int ref; |
6118b70b JA |
111 | /* various state flags, see below */ |
112 | unsigned int flags; | |
113 | /* parent cfq_data */ | |
114 | struct cfq_data *cfqd; | |
115 | /* service_tree member */ | |
116 | struct rb_node rb_node; | |
117 | /* service_tree key */ | |
9a7f38c4 | 118 | u64 rb_key; |
6118b70b JA |
119 | /* prio tree member */ |
120 | struct rb_node p_node; | |
121 | /* prio tree root we belong to, if any */ | |
122 | struct rb_root *p_root; | |
123 | /* sorted list of pending requests */ | |
124 | struct rb_root sort_list; | |
125 | /* if fifo isn't expired, next request to serve */ | |
126 | struct request *next_rq; | |
127 | /* requests queued in sort_list */ | |
128 | int queued[2]; | |
129 | /* currently allocated requests */ | |
130 | int allocated[2]; | |
131 | /* fifo list of requests in sort_list */ | |
132 | struct list_head fifo; | |
133 | ||
dae739eb | 134 | /* time when queue got scheduled in to dispatch first request. */ |
9a7f38c4 JM |
135 | u64 dispatch_start; |
136 | u64 allocated_slice; | |
137 | u64 slice_dispatch; | |
dae739eb | 138 | /* time when first request from queue completed and slice started. */ |
9a7f38c4 JM |
139 | u64 slice_start; |
140 | u64 slice_end; | |
93fdf147 | 141 | s64 slice_resid; |
6118b70b | 142 | |
65299a3b CH |
143 | /* pending priority requests */ |
144 | int prio_pending; | |
6118b70b JA |
145 | /* number of requests that are on the dispatch list or inside driver */ |
146 | int dispatched; | |
147 | ||
148 | /* io prio of this group */ | |
149 | unsigned short ioprio, org_ioprio; | |
b8269db4 | 150 | unsigned short ioprio_class, org_ioprio_class; |
6118b70b | 151 | |
c4081ba5 RK |
152 | pid_t pid; |
153 | ||
3dde36dd | 154 | u32 seek_history; |
b2c18e1e JM |
155 | sector_t last_request_pos; |
156 | ||
aa6f6a3d | 157 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 158 | struct cfq_queue *new_cfqq; |
cdb16e8f | 159 | struct cfq_group *cfqg; |
c4e7893e VG |
160 | /* Number of sectors dispatched from queue in single dispatch round */ |
161 | unsigned long nr_sectors; | |
6118b70b JA |
162 | }; |
163 | ||
c0324a02 | 164 | /* |
718eee05 | 165 | * First index in the service_trees. |
c0324a02 CZ |
166 | * IDLE is handled separately, so it has negative index |
167 | */ | |
3bf10fea | 168 | enum wl_class_t { |
c0324a02 | 169 | BE_WORKLOAD = 0, |
615f0259 VG |
170 | RT_WORKLOAD = 1, |
171 | IDLE_WORKLOAD = 2, | |
b4627321 | 172 | CFQ_PRIO_NR, |
c0324a02 CZ |
173 | }; |
174 | ||
718eee05 CZ |
175 | /* |
176 | * Second index in the service_trees. | |
177 | */ | |
178 | enum wl_type_t { | |
179 | ASYNC_WORKLOAD = 0, | |
180 | SYNC_NOIDLE_WORKLOAD = 1, | |
181 | SYNC_WORKLOAD = 2 | |
182 | }; | |
183 | ||
155fead9 TH |
184 | struct cfqg_stats { |
185 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
155fead9 TH |
186 | /* number of ios merged */ |
187 | struct blkg_rwstat merged; | |
188 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
189 | struct blkg_rwstat service_time; | |
190 | /* total time spent waiting in scheduler queue in ns */ | |
191 | struct blkg_rwstat wait_time; | |
192 | /* number of IOs queued up */ | |
193 | struct blkg_rwstat queued; | |
155fead9 TH |
194 | /* total disk time and nr sectors dispatched by this group */ |
195 | struct blkg_stat time; | |
196 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
197 | /* time not charged to this cgroup */ | |
198 | struct blkg_stat unaccounted_time; | |
199 | /* sum of number of ios queued across all samples */ | |
200 | struct blkg_stat avg_queue_size_sum; | |
201 | /* count of samples taken for average */ | |
202 | struct blkg_stat avg_queue_size_samples; | |
203 | /* how many times this group has been removed from service tree */ | |
204 | struct blkg_stat dequeue; | |
205 | /* total time spent waiting for it to be assigned a timeslice. */ | |
206 | struct blkg_stat group_wait_time; | |
3c798398 | 207 | /* time spent idling for this blkcg_gq */ |
155fead9 TH |
208 | struct blkg_stat idle_time; |
209 | /* total time with empty current active q with other requests queued */ | |
210 | struct blkg_stat empty_time; | |
211 | /* fields after this shouldn't be cleared on stat reset */ | |
212 | uint64_t start_group_wait_time; | |
213 | uint64_t start_idle_time; | |
214 | uint64_t start_empty_time; | |
215 | uint16_t flags; | |
216 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
217 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
218 | }; | |
219 | ||
e48453c3 AA |
220 | /* Per-cgroup data */ |
221 | struct cfq_group_data { | |
222 | /* must be the first member */ | |
81437648 | 223 | struct blkcg_policy_data cpd; |
e48453c3 AA |
224 | |
225 | unsigned int weight; | |
226 | unsigned int leaf_weight; | |
227 | }; | |
228 | ||
cdb16e8f VG |
229 | /* This is per cgroup per device grouping structure */ |
230 | struct cfq_group { | |
f95a04af TH |
231 | /* must be the first member */ |
232 | struct blkg_policy_data pd; | |
233 | ||
1fa8f6d6 VG |
234 | /* group service_tree member */ |
235 | struct rb_node rb_node; | |
236 | ||
237 | /* group service_tree key */ | |
238 | u64 vdisktime; | |
e71357e1 | 239 | |
7918ffb5 TH |
240 | /* |
241 | * The number of active cfqgs and sum of their weights under this | |
242 | * cfqg. This covers this cfqg's leaf_weight and all children's | |
243 | * weights, but does not cover weights of further descendants. | |
244 | * | |
245 | * If a cfqg is on the service tree, it's active. An active cfqg | |
246 | * also activates its parent and contributes to the children_weight | |
247 | * of the parent. | |
248 | */ | |
249 | int nr_active; | |
250 | unsigned int children_weight; | |
251 | ||
1d3650f7 TH |
252 | /* |
253 | * vfraction is the fraction of vdisktime that the tasks in this | |
254 | * cfqg are entitled to. This is determined by compounding the | |
255 | * ratios walking up from this cfqg to the root. | |
256 | * | |
257 | * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all | |
258 | * vfractions on a service tree is approximately 1. The sum may | |
259 | * deviate a bit due to rounding errors and fluctuations caused by | |
260 | * cfqgs entering and leaving the service tree. | |
261 | */ | |
262 | unsigned int vfraction; | |
263 | ||
e71357e1 TH |
264 | /* |
265 | * There are two weights - (internal) weight is the weight of this | |
266 | * cfqg against the sibling cfqgs. leaf_weight is the wight of | |
267 | * this cfqg against the child cfqgs. For the root cfqg, both | |
268 | * weights are kept in sync for backward compatibility. | |
269 | */ | |
25bc6b07 | 270 | unsigned int weight; |
8184f93e | 271 | unsigned int new_weight; |
3381cb8d | 272 | unsigned int dev_weight; |
1fa8f6d6 | 273 | |
e71357e1 TH |
274 | unsigned int leaf_weight; |
275 | unsigned int new_leaf_weight; | |
276 | unsigned int dev_leaf_weight; | |
277 | ||
1fa8f6d6 VG |
278 | /* number of cfqq currently on this group */ |
279 | int nr_cfqq; | |
280 | ||
cdb16e8f | 281 | /* |
4495a7d4 | 282 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
283 | * create the array for each prio class but at run time it is used |
284 | * only for RT and BE class and slot for IDLE class remains unused. | |
285 | * This is primarily done to avoid confusion and a gcc warning. | |
286 | */ | |
287 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
288 | /* | |
289 | * rr lists of queues with requests. We maintain service trees for | |
290 | * RT and BE classes. These trees are subdivided in subclasses | |
291 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
292 | * class there is no subclassification and all the cfq queues go on | |
293 | * a single tree service_tree_idle. | |
cdb16e8f VG |
294 | * Counts are embedded in the cfq_rb_root |
295 | */ | |
296 | struct cfq_rb_root service_trees[2][3]; | |
297 | struct cfq_rb_root service_tree_idle; | |
dae739eb | 298 | |
9a7f38c4 | 299 | u64 saved_wl_slice; |
4d2ceea4 VG |
300 | enum wl_type_t saved_wl_type; |
301 | enum wl_class_t saved_wl_class; | |
4eef3049 | 302 | |
80bdf0c7 VG |
303 | /* number of requests that are on the dispatch list or inside driver */ |
304 | int dispatched; | |
7700fc4f | 305 | struct cfq_ttime ttime; |
0b39920b | 306 | struct cfqg_stats stats; /* stats for this cfqg */ |
60a83707 TH |
307 | |
308 | /* async queue for each priority case */ | |
309 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
310 | struct cfq_queue *async_idle_cfqq; | |
311 | ||
cdb16e8f | 312 | }; |
718eee05 | 313 | |
c5869807 TH |
314 | struct cfq_io_cq { |
315 | struct io_cq icq; /* must be the first member */ | |
316 | struct cfq_queue *cfqq[2]; | |
317 | struct cfq_ttime ttime; | |
598971bf TH |
318 | int ioprio; /* the current ioprio */ |
319 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
f4da8072 | 320 | uint64_t blkcg_serial_nr; /* the current blkcg serial */ |
598971bf | 321 | #endif |
c5869807 TH |
322 | }; |
323 | ||
22e2c507 JA |
324 | /* |
325 | * Per block device queue structure | |
326 | */ | |
1da177e4 | 327 | struct cfq_data { |
165125e1 | 328 | struct request_queue *queue; |
1fa8f6d6 VG |
329 | /* Root service tree for cfq_groups */ |
330 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 331 | struct cfq_group *root_group; |
22e2c507 | 332 | |
c0324a02 CZ |
333 | /* |
334 | * The priority currently being served | |
22e2c507 | 335 | */ |
4d2ceea4 VG |
336 | enum wl_class_t serving_wl_class; |
337 | enum wl_type_t serving_wl_type; | |
9a7f38c4 | 338 | u64 workload_expires; |
cdb16e8f | 339 | struct cfq_group *serving_group; |
a36e71f9 JA |
340 | |
341 | /* | |
342 | * Each priority tree is sorted by next_request position. These | |
343 | * trees are used when determining if two or more queues are | |
344 | * interleaving requests (see cfq_close_cooperator). | |
345 | */ | |
346 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
347 | ||
22e2c507 | 348 | unsigned int busy_queues; |
ef8a41df | 349 | unsigned int busy_sync_queues; |
22e2c507 | 350 | |
53c583d2 CZ |
351 | int rq_in_driver; |
352 | int rq_in_flight[2]; | |
45333d5a AC |
353 | |
354 | /* | |
355 | * queue-depth detection | |
356 | */ | |
357 | int rq_queued; | |
25776e35 | 358 | int hw_tag; |
e459dd08 CZ |
359 | /* |
360 | * hw_tag can be | |
361 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
362 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
363 | * 0 => no NCQ | |
364 | */ | |
365 | int hw_tag_est_depth; | |
366 | unsigned int hw_tag_samples; | |
1da177e4 | 367 | |
22e2c507 JA |
368 | /* |
369 | * idle window management | |
370 | */ | |
91148325 | 371 | struct hrtimer idle_slice_timer; |
23e018a1 | 372 | struct work_struct unplug_work; |
1da177e4 | 373 | |
22e2c507 | 374 | struct cfq_queue *active_queue; |
c5869807 | 375 | struct cfq_io_cq *active_cic; |
22e2c507 | 376 | |
6d048f53 | 377 | sector_t last_position; |
1da177e4 | 378 | |
1da177e4 LT |
379 | /* |
380 | * tunables, see top of file | |
381 | */ | |
382 | unsigned int cfq_quantum; | |
1da177e4 LT |
383 | unsigned int cfq_back_penalty; |
384 | unsigned int cfq_back_max; | |
22e2c507 | 385 | unsigned int cfq_slice_async_rq; |
963b72fc | 386 | unsigned int cfq_latency; |
9a7f38c4 JM |
387 | u64 cfq_fifo_expire[2]; |
388 | u64 cfq_slice[2]; | |
389 | u64 cfq_slice_idle; | |
390 | u64 cfq_group_idle; | |
391 | u64 cfq_target_latency; | |
d9ff4187 | 392 | |
6118b70b JA |
393 | /* |
394 | * Fallback dummy cfqq for extreme OOM conditions | |
395 | */ | |
396 | struct cfq_queue oom_cfqq; | |
365722bb | 397 | |
9a7f38c4 | 398 | u64 last_delayed_sync; |
1da177e4 LT |
399 | }; |
400 | ||
25fb5169 | 401 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
60a83707 | 402 | static void cfq_put_queue(struct cfq_queue *cfqq); |
25fb5169 | 403 | |
34b98d03 | 404 | static struct cfq_rb_root *st_for(struct cfq_group *cfqg, |
3bf10fea | 405 | enum wl_class_t class, |
65b32a57 | 406 | enum wl_type_t type) |
c0324a02 | 407 | { |
1fa8f6d6 VG |
408 | if (!cfqg) |
409 | return NULL; | |
410 | ||
3bf10fea | 411 | if (class == IDLE_WORKLOAD) |
cdb16e8f | 412 | return &cfqg->service_tree_idle; |
c0324a02 | 413 | |
3bf10fea | 414 | return &cfqg->service_trees[class][type]; |
c0324a02 CZ |
415 | } |
416 | ||
3b18152c | 417 | enum cfqq_state_flags { |
b0b8d749 JA |
418 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
419 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 420 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 421 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
422 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
423 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
424 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 425 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 426 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 427 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 428 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 429 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 430 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
431 | }; |
432 | ||
433 | #define CFQ_CFQQ_FNS(name) \ | |
434 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
435 | { \ | |
fe094d98 | 436 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
437 | } \ |
438 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
439 | { \ | |
fe094d98 | 440 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
441 | } \ |
442 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
443 | { \ | |
fe094d98 | 444 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
445 | } |
446 | ||
447 | CFQ_CFQQ_FNS(on_rr); | |
448 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 449 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 450 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
451 | CFQ_CFQQ_FNS(fifo_expire); |
452 | CFQ_CFQQ_FNS(idle_window); | |
453 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 454 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 455 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 456 | CFQ_CFQQ_FNS(coop); |
ae54abed | 457 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 458 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 459 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
460 | #undef CFQ_CFQQ_FNS |
461 | ||
629ed0b1 | 462 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
2ce4d50f | 463 | |
155fead9 TH |
464 | /* cfqg stats flags */ |
465 | enum cfqg_stats_flags { | |
466 | CFQG_stats_waiting = 0, | |
467 | CFQG_stats_idling, | |
468 | CFQG_stats_empty, | |
629ed0b1 TH |
469 | }; |
470 | ||
155fead9 TH |
471 | #define CFQG_FLAG_FNS(name) \ |
472 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ | |
629ed0b1 | 473 | { \ |
155fead9 | 474 | stats->flags |= (1 << CFQG_stats_##name); \ |
629ed0b1 | 475 | } \ |
155fead9 | 476 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 477 | { \ |
155fead9 | 478 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
629ed0b1 | 479 | } \ |
155fead9 | 480 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 481 | { \ |
155fead9 | 482 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
629ed0b1 TH |
483 | } \ |
484 | ||
155fead9 TH |
485 | CFQG_FLAG_FNS(waiting) |
486 | CFQG_FLAG_FNS(idling) | |
487 | CFQG_FLAG_FNS(empty) | |
488 | #undef CFQG_FLAG_FNS | |
629ed0b1 TH |
489 | |
490 | /* This should be called with the queue_lock held. */ | |
155fead9 | 491 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
629ed0b1 TH |
492 | { |
493 | unsigned long long now; | |
494 | ||
155fead9 | 495 | if (!cfqg_stats_waiting(stats)) |
629ed0b1 TH |
496 | return; |
497 | ||
498 | now = sched_clock(); | |
499 | if (time_after64(now, stats->start_group_wait_time)) | |
500 | blkg_stat_add(&stats->group_wait_time, | |
501 | now - stats->start_group_wait_time); | |
155fead9 | 502 | cfqg_stats_clear_waiting(stats); |
629ed0b1 TH |
503 | } |
504 | ||
505 | /* This should be called with the queue_lock held. */ | |
155fead9 TH |
506 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
507 | struct cfq_group *curr_cfqg) | |
629ed0b1 | 508 | { |
155fead9 | 509 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 510 | |
155fead9 | 511 | if (cfqg_stats_waiting(stats)) |
629ed0b1 | 512 | return; |
155fead9 | 513 | if (cfqg == curr_cfqg) |
629ed0b1 | 514 | return; |
155fead9 TH |
515 | stats->start_group_wait_time = sched_clock(); |
516 | cfqg_stats_mark_waiting(stats); | |
629ed0b1 TH |
517 | } |
518 | ||
519 | /* This should be called with the queue_lock held. */ | |
155fead9 | 520 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
629ed0b1 TH |
521 | { |
522 | unsigned long long now; | |
523 | ||
155fead9 | 524 | if (!cfqg_stats_empty(stats)) |
629ed0b1 TH |
525 | return; |
526 | ||
527 | now = sched_clock(); | |
528 | if (time_after64(now, stats->start_empty_time)) | |
529 | blkg_stat_add(&stats->empty_time, | |
530 | now - stats->start_empty_time); | |
155fead9 | 531 | cfqg_stats_clear_empty(stats); |
629ed0b1 TH |
532 | } |
533 | ||
155fead9 | 534 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
629ed0b1 | 535 | { |
155fead9 | 536 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
629ed0b1 TH |
537 | } |
538 | ||
155fead9 | 539 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
629ed0b1 | 540 | { |
155fead9 | 541 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 542 | |
4d5e80a7 | 543 | if (blkg_rwstat_total(&stats->queued)) |
629ed0b1 TH |
544 | return; |
545 | ||
546 | /* | |
547 | * group is already marked empty. This can happen if cfqq got new | |
548 | * request in parent group and moved to this group while being added | |
549 | * to service tree. Just ignore the event and move on. | |
550 | */ | |
155fead9 | 551 | if (cfqg_stats_empty(stats)) |
629ed0b1 TH |
552 | return; |
553 | ||
554 | stats->start_empty_time = sched_clock(); | |
155fead9 | 555 | cfqg_stats_mark_empty(stats); |
629ed0b1 TH |
556 | } |
557 | ||
155fead9 | 558 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 559 | { |
155fead9 | 560 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 561 | |
155fead9 | 562 | if (cfqg_stats_idling(stats)) { |
629ed0b1 TH |
563 | unsigned long long now = sched_clock(); |
564 | ||
565 | if (time_after64(now, stats->start_idle_time)) | |
566 | blkg_stat_add(&stats->idle_time, | |
567 | now - stats->start_idle_time); | |
155fead9 | 568 | cfqg_stats_clear_idling(stats); |
629ed0b1 TH |
569 | } |
570 | } | |
571 | ||
155fead9 | 572 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 573 | { |
155fead9 | 574 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 575 | |
155fead9 | 576 | BUG_ON(cfqg_stats_idling(stats)); |
629ed0b1 TH |
577 | |
578 | stats->start_idle_time = sched_clock(); | |
155fead9 | 579 | cfqg_stats_mark_idling(stats); |
629ed0b1 TH |
580 | } |
581 | ||
155fead9 | 582 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
629ed0b1 | 583 | { |
155fead9 | 584 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
585 | |
586 | blkg_stat_add(&stats->avg_queue_size_sum, | |
4d5e80a7 | 587 | blkg_rwstat_total(&stats->queued)); |
629ed0b1 | 588 | blkg_stat_add(&stats->avg_queue_size_samples, 1); |
155fead9 | 589 | cfqg_stats_update_group_wait_time(stats); |
629ed0b1 TH |
590 | } |
591 | ||
592 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
593 | ||
f48ec1d7 TH |
594 | static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { } |
595 | static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } | |
596 | static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } | |
597 | static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } | |
598 | static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } | |
599 | static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } | |
600 | static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } | |
629ed0b1 TH |
601 | |
602 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
603 | ||
604 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
2ce4d50f | 605 | |
4ceab71b JA |
606 | static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd) |
607 | { | |
608 | return pd ? container_of(pd, struct cfq_group, pd) : NULL; | |
609 | } | |
610 | ||
611 | static struct cfq_group_data | |
612 | *cpd_to_cfqgd(struct blkcg_policy_data *cpd) | |
613 | { | |
81437648 | 614 | return cpd ? container_of(cpd, struct cfq_group_data, cpd) : NULL; |
4ceab71b JA |
615 | } |
616 | ||
617 | static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg) | |
618 | { | |
619 | return pd_to_blkg(&cfqg->pd); | |
620 | } | |
621 | ||
ffea73fc TH |
622 | static struct blkcg_policy blkcg_policy_cfq; |
623 | ||
624 | static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg) | |
625 | { | |
626 | return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq)); | |
627 | } | |
628 | ||
e48453c3 AA |
629 | static struct cfq_group_data *blkcg_to_cfqgd(struct blkcg *blkcg) |
630 | { | |
631 | return cpd_to_cfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_cfq)); | |
632 | } | |
633 | ||
d02f7aa8 | 634 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) |
7918ffb5 | 635 | { |
d02f7aa8 | 636 | struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent; |
7918ffb5 | 637 | |
d02f7aa8 | 638 | return pblkg ? blkg_to_cfqg(pblkg) : NULL; |
7918ffb5 TH |
639 | } |
640 | ||
3984aa55 JK |
641 | static inline bool cfqg_is_descendant(struct cfq_group *cfqg, |
642 | struct cfq_group *ancestor) | |
643 | { | |
644 | return cgroup_is_descendant(cfqg_to_blkg(cfqg)->blkcg->css.cgroup, | |
645 | cfqg_to_blkg(ancestor)->blkcg->css.cgroup); | |
646 | } | |
647 | ||
eb7d8c07 TH |
648 | static inline void cfqg_get(struct cfq_group *cfqg) |
649 | { | |
650 | return blkg_get(cfqg_to_blkg(cfqg)); | |
651 | } | |
652 | ||
653 | static inline void cfqg_put(struct cfq_group *cfqg) | |
654 | { | |
655 | return blkg_put(cfqg_to_blkg(cfqg)); | |
656 | } | |
657 | ||
54e7ed12 TH |
658 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \ |
659 | char __pbuf[128]; \ | |
660 | \ | |
661 | blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \ | |
b226e5c4 VG |
662 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \ |
663 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
664 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
54e7ed12 TH |
665 | __pbuf, ##args); \ |
666 | } while (0) | |
667 | ||
668 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \ | |
669 | char __pbuf[128]; \ | |
670 | \ | |
671 | blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \ | |
672 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \ | |
673 | } while (0) | |
2868ef7b | 674 | |
155fead9 | 675 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
ef295ecf CH |
676 | struct cfq_group *curr_cfqg, |
677 | unsigned int op) | |
2ce4d50f | 678 | { |
ef295ecf | 679 | blkg_rwstat_add(&cfqg->stats.queued, op, 1); |
155fead9 TH |
680 | cfqg_stats_end_empty_time(&cfqg->stats); |
681 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); | |
2ce4d50f TH |
682 | } |
683 | ||
155fead9 | 684 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
9a7f38c4 | 685 | uint64_t time, unsigned long unaccounted_time) |
2ce4d50f | 686 | { |
155fead9 | 687 | blkg_stat_add(&cfqg->stats.time, time); |
629ed0b1 | 688 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
155fead9 | 689 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
629ed0b1 | 690 | #endif |
2ce4d50f TH |
691 | } |
692 | ||
ef295ecf CH |
693 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, |
694 | unsigned int op) | |
2ce4d50f | 695 | { |
ef295ecf | 696 | blkg_rwstat_add(&cfqg->stats.queued, op, -1); |
2ce4d50f TH |
697 | } |
698 | ||
ef295ecf CH |
699 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, |
700 | unsigned int op) | |
2ce4d50f | 701 | { |
ef295ecf | 702 | blkg_rwstat_add(&cfqg->stats.merged, op, 1); |
2ce4d50f TH |
703 | } |
704 | ||
155fead9 | 705 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
ef295ecf CH |
706 | uint64_t start_time, uint64_t io_start_time, |
707 | unsigned int op) | |
2ce4d50f | 708 | { |
155fead9 | 709 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 710 | unsigned long long now = sched_clock(); |
629ed0b1 TH |
711 | |
712 | if (time_after64(now, io_start_time)) | |
ef295ecf | 713 | blkg_rwstat_add(&stats->service_time, op, now - io_start_time); |
629ed0b1 | 714 | if (time_after64(io_start_time, start_time)) |
ef295ecf | 715 | blkg_rwstat_add(&stats->wait_time, op, |
629ed0b1 | 716 | io_start_time - start_time); |
2ce4d50f TH |
717 | } |
718 | ||
689665af TH |
719 | /* @stats = 0 */ |
720 | static void cfqg_stats_reset(struct cfqg_stats *stats) | |
155fead9 | 721 | { |
155fead9 | 722 | /* queued stats shouldn't be cleared */ |
155fead9 TH |
723 | blkg_rwstat_reset(&stats->merged); |
724 | blkg_rwstat_reset(&stats->service_time); | |
725 | blkg_rwstat_reset(&stats->wait_time); | |
726 | blkg_stat_reset(&stats->time); | |
727 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
728 | blkg_stat_reset(&stats->unaccounted_time); | |
729 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
730 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
731 | blkg_stat_reset(&stats->dequeue); | |
732 | blkg_stat_reset(&stats->group_wait_time); | |
733 | blkg_stat_reset(&stats->idle_time); | |
734 | blkg_stat_reset(&stats->empty_time); | |
735 | #endif | |
736 | } | |
737 | ||
0b39920b | 738 | /* @to += @from */ |
e6269c44 | 739 | static void cfqg_stats_add_aux(struct cfqg_stats *to, struct cfqg_stats *from) |
0b39920b TH |
740 | { |
741 | /* queued stats shouldn't be cleared */ | |
e6269c44 TH |
742 | blkg_rwstat_add_aux(&to->merged, &from->merged); |
743 | blkg_rwstat_add_aux(&to->service_time, &from->service_time); | |
744 | blkg_rwstat_add_aux(&to->wait_time, &from->wait_time); | |
745 | blkg_stat_add_aux(&from->time, &from->time); | |
0b39920b | 746 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
e6269c44 TH |
747 | blkg_stat_add_aux(&to->unaccounted_time, &from->unaccounted_time); |
748 | blkg_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum); | |
749 | blkg_stat_add_aux(&to->avg_queue_size_samples, &from->avg_queue_size_samples); | |
750 | blkg_stat_add_aux(&to->dequeue, &from->dequeue); | |
751 | blkg_stat_add_aux(&to->group_wait_time, &from->group_wait_time); | |
752 | blkg_stat_add_aux(&to->idle_time, &from->idle_time); | |
753 | blkg_stat_add_aux(&to->empty_time, &from->empty_time); | |
0b39920b TH |
754 | #endif |
755 | } | |
756 | ||
757 | /* | |
e6269c44 | 758 | * Transfer @cfqg's stats to its parent's aux counts so that the ancestors' |
0b39920b TH |
759 | * recursive stats can still account for the amount used by this cfqg after |
760 | * it's gone. | |
761 | */ | |
762 | static void cfqg_stats_xfer_dead(struct cfq_group *cfqg) | |
763 | { | |
764 | struct cfq_group *parent = cfqg_parent(cfqg); | |
765 | ||
766 | lockdep_assert_held(cfqg_to_blkg(cfqg)->q->queue_lock); | |
767 | ||
768 | if (unlikely(!parent)) | |
769 | return; | |
770 | ||
e6269c44 | 771 | cfqg_stats_add_aux(&parent->stats, &cfqg->stats); |
0b39920b | 772 | cfqg_stats_reset(&cfqg->stats); |
0b39920b TH |
773 | } |
774 | ||
eb7d8c07 TH |
775 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
776 | ||
d02f7aa8 | 777 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) { return NULL; } |
3984aa55 JK |
778 | static inline bool cfqg_is_descendant(struct cfq_group *cfqg, |
779 | struct cfq_group *ancestor) | |
780 | { | |
781 | return true; | |
782 | } | |
eb7d8c07 TH |
783 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
784 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
785 | ||
7b679138 | 786 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
b226e5c4 VG |
787 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \ |
788 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
789 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
790 | ##args) | |
4495a7d4 | 791 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 | 792 | |
155fead9 | 793 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
ef295ecf | 794 | struct cfq_group *curr_cfqg, unsigned int op) { } |
155fead9 | 795 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
9a7f38c4 | 796 | uint64_t time, unsigned long unaccounted_time) { } |
ef295ecf CH |
797 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, |
798 | unsigned int op) { } | |
799 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, | |
800 | unsigned int op) { } | |
155fead9 | 801 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
ef295ecf CH |
802 | uint64_t start_time, uint64_t io_start_time, |
803 | unsigned int op) { } | |
2ce4d50f | 804 | |
eb7d8c07 TH |
805 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
806 | ||
7b679138 JA |
807 | #define cfq_log(cfqd, fmt, args...) \ |
808 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
809 | ||
615f0259 VG |
810 | /* Traverses through cfq group service trees */ |
811 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
812 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
813 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
814 | : &cfqg->service_tree_idle; \ | |
815 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
816 | (i == IDLE_WORKLOAD && j == 0); \ | |
817 | j++, st = i < IDLE_WORKLOAD ? \ | |
818 | &cfqg->service_trees[i][j]: NULL) \ | |
819 | ||
f5f2b6ce SL |
820 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
821 | struct cfq_ttime *ttime, bool group_idle) | |
822 | { | |
9a7f38c4 | 823 | u64 slice; |
f5f2b6ce SL |
824 | if (!sample_valid(ttime->ttime_samples)) |
825 | return false; | |
826 | if (group_idle) | |
827 | slice = cfqd->cfq_group_idle; | |
828 | else | |
829 | slice = cfqd->cfq_slice_idle; | |
830 | return ttime->ttime_mean > slice; | |
831 | } | |
615f0259 | 832 | |
02b35081 VG |
833 | static inline bool iops_mode(struct cfq_data *cfqd) |
834 | { | |
835 | /* | |
836 | * If we are not idling on queues and it is a NCQ drive, parallel | |
837 | * execution of requests is on and measuring time is not possible | |
838 | * in most of the cases until and unless we drive shallower queue | |
839 | * depths and that becomes a performance bottleneck. In such cases | |
840 | * switch to start providing fairness in terms of number of IOs. | |
841 | */ | |
842 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
843 | return true; | |
844 | else | |
845 | return false; | |
846 | } | |
847 | ||
3bf10fea | 848 | static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq) |
c0324a02 CZ |
849 | { |
850 | if (cfq_class_idle(cfqq)) | |
851 | return IDLE_WORKLOAD; | |
852 | if (cfq_class_rt(cfqq)) | |
853 | return RT_WORKLOAD; | |
854 | return BE_WORKLOAD; | |
855 | } | |
856 | ||
718eee05 CZ |
857 | |
858 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
859 | { | |
860 | if (!cfq_cfqq_sync(cfqq)) | |
861 | return ASYNC_WORKLOAD; | |
862 | if (!cfq_cfqq_idle_window(cfqq)) | |
863 | return SYNC_NOIDLE_WORKLOAD; | |
864 | return SYNC_WORKLOAD; | |
865 | } | |
866 | ||
3bf10fea | 867 | static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class, |
58ff82f3 VG |
868 | struct cfq_data *cfqd, |
869 | struct cfq_group *cfqg) | |
c0324a02 | 870 | { |
3bf10fea | 871 | if (wl_class == IDLE_WORKLOAD) |
cdb16e8f | 872 | return cfqg->service_tree_idle.count; |
c0324a02 | 873 | |
34b98d03 VG |
874 | return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count + |
875 | cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count + | |
876 | cfqg->service_trees[wl_class][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
877 | } |
878 | ||
f26bd1f0 VG |
879 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
880 | struct cfq_group *cfqg) | |
881 | { | |
34b98d03 VG |
882 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count + |
883 | cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
f26bd1f0 VG |
884 | } |
885 | ||
165125e1 | 886 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 887 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
2da8de0b | 888 | struct cfq_io_cq *cic, struct bio *bio); |
91fac317 | 889 | |
c5869807 TH |
890 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
891 | { | |
892 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
893 | return container_of(icq, struct cfq_io_cq, icq); | |
894 | } | |
895 | ||
47fdd4ca TH |
896 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
897 | struct io_context *ioc) | |
898 | { | |
899 | if (ioc) | |
900 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
901 | return NULL; | |
902 | } | |
903 | ||
c5869807 | 904 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 905 | { |
a6151c3a | 906 | return cic->cfqq[is_sync]; |
91fac317 VT |
907 | } |
908 | ||
c5869807 TH |
909 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
910 | bool is_sync) | |
91fac317 | 911 | { |
a6151c3a | 912 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
913 | } |
914 | ||
c5869807 | 915 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 916 | { |
c5869807 | 917 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
918 | } |
919 | ||
99f95e52 AM |
920 | /* |
921 | * scheduler run of queue, if there are requests pending and no one in the | |
922 | * driver that will restart queueing | |
923 | */ | |
23e018a1 | 924 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 925 | { |
7b679138 JA |
926 | if (cfqd->busy_queues) { |
927 | cfq_log(cfqd, "schedule dispatch"); | |
59c3d45e | 928 | kblockd_schedule_work(&cfqd->unplug_work); |
7b679138 | 929 | } |
99f95e52 AM |
930 | } |
931 | ||
44f7c160 JA |
932 | /* |
933 | * Scale schedule slice based on io priority. Use the sync time slice only | |
934 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
935 | * io only, should not get full sync slice length. | |
936 | */ | |
9a7f38c4 | 937 | static inline u64 cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 938 | unsigned short prio) |
44f7c160 | 939 | { |
9a7f38c4 JM |
940 | u64 base_slice = cfqd->cfq_slice[sync]; |
941 | u64 slice = div_u64(base_slice, CFQ_SLICE_SCALE); | |
44f7c160 | 942 | |
d9e7620e JA |
943 | WARN_ON(prio >= IOPRIO_BE_NR); |
944 | ||
9a7f38c4 | 945 | return base_slice + (slice * (4 - prio)); |
d9e7620e | 946 | } |
44f7c160 | 947 | |
9a7f38c4 | 948 | static inline u64 |
d9e7620e JA |
949 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
950 | { | |
951 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
952 | } |
953 | ||
1d3650f7 TH |
954 | /** |
955 | * cfqg_scale_charge - scale disk time charge according to cfqg weight | |
956 | * @charge: disk time being charged | |
957 | * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT | |
958 | * | |
959 | * Scale @charge according to @vfraction, which is in range (0, 1]. The | |
960 | * scaling is inversely proportional. | |
961 | * | |
962 | * scaled = charge / vfraction | |
963 | * | |
964 | * The result is also in fixed point w/ CFQ_SERVICE_SHIFT. | |
965 | */ | |
9a7f38c4 | 966 | static inline u64 cfqg_scale_charge(u64 charge, |
1d3650f7 | 967 | unsigned int vfraction) |
25bc6b07 | 968 | { |
1d3650f7 | 969 | u64 c = charge << CFQ_SERVICE_SHIFT; /* make it fixed point */ |
25bc6b07 | 970 | |
1d3650f7 TH |
971 | /* charge / vfraction */ |
972 | c <<= CFQ_SERVICE_SHIFT; | |
9a7f38c4 | 973 | return div_u64(c, vfraction); |
25bc6b07 VG |
974 | } |
975 | ||
976 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
977 | { | |
978 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
979 | if (delta > 0) | |
980 | min_vdisktime = vdisktime; | |
981 | ||
982 | return min_vdisktime; | |
983 | } | |
984 | ||
25bc6b07 VG |
985 | static void update_min_vdisktime(struct cfq_rb_root *st) |
986 | { | |
25bc6b07 VG |
987 | struct cfq_group *cfqg; |
988 | ||
25bc6b07 VG |
989 | if (st->left) { |
990 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
991 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
992 | cfqg->vdisktime); | |
25bc6b07 | 993 | } |
25bc6b07 VG |
994 | } |
995 | ||
5db5d642 CZ |
996 | /* |
997 | * get averaged number of queues of RT/BE priority. | |
998 | * average is updated, with a formula that gives more weight to higher numbers, | |
999 | * to quickly follows sudden increases and decrease slowly | |
1000 | */ | |
1001 | ||
58ff82f3 VG |
1002 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
1003 | struct cfq_group *cfqg, bool rt) | |
5869619c | 1004 | { |
5db5d642 CZ |
1005 | unsigned min_q, max_q; |
1006 | unsigned mult = cfq_hist_divisor - 1; | |
1007 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 1008 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 1009 | |
58ff82f3 VG |
1010 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
1011 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
1012 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 1013 | cfq_hist_divisor; |
58ff82f3 VG |
1014 | return cfqg->busy_queues_avg[rt]; |
1015 | } | |
1016 | ||
9a7f38c4 | 1017 | static inline u64 |
58ff82f3 VG |
1018 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1019 | { | |
41cad6ab | 1020 | return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT; |
5db5d642 CZ |
1021 | } |
1022 | ||
9a7f38c4 | 1023 | static inline u64 |
ba5bd520 | 1024 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 1025 | { |
9a7f38c4 | 1026 | u64 slice = cfq_prio_to_slice(cfqd, cfqq); |
5db5d642 | 1027 | if (cfqd->cfq_latency) { |
58ff82f3 VG |
1028 | /* |
1029 | * interested queues (we consider only the ones with the same | |
1030 | * priority class in the cfq group) | |
1031 | */ | |
1032 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
1033 | cfq_class_rt(cfqq)); | |
9a7f38c4 JM |
1034 | u64 sync_slice = cfqd->cfq_slice[1]; |
1035 | u64 expect_latency = sync_slice * iq; | |
1036 | u64 group_slice = cfq_group_slice(cfqd, cfqq->cfqg); | |
58ff82f3 VG |
1037 | |
1038 | if (expect_latency > group_slice) { | |
9a7f38c4 JM |
1039 | u64 base_low_slice = 2 * cfqd->cfq_slice_idle; |
1040 | u64 low_slice; | |
1041 | ||
5db5d642 CZ |
1042 | /* scale low_slice according to IO priority |
1043 | * and sync vs async */ | |
9a7f38c4 JM |
1044 | low_slice = div64_u64(base_low_slice*slice, sync_slice); |
1045 | low_slice = min(slice, low_slice); | |
5db5d642 CZ |
1046 | /* the adapted slice value is scaled to fit all iqs |
1047 | * into the target latency */ | |
9a7f38c4 JM |
1048 | slice = div64_u64(slice*group_slice, expect_latency); |
1049 | slice = max(slice, low_slice); | |
5db5d642 CZ |
1050 | } |
1051 | } | |
c553f8e3 SL |
1052 | return slice; |
1053 | } | |
1054 | ||
1055 | static inline void | |
1056 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1057 | { | |
9a7f38c4 JM |
1058 | u64 slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
1059 | u64 now = ktime_get_ns(); | |
c553f8e3 | 1060 | |
9a7f38c4 JM |
1061 | cfqq->slice_start = now; |
1062 | cfqq->slice_end = now + slice; | |
f75edf2d | 1063 | cfqq->allocated_slice = slice; |
9a7f38c4 | 1064 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%llu", cfqq->slice_end - now); |
44f7c160 JA |
1065 | } |
1066 | ||
1067 | /* | |
1068 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
1069 | * isn't valid until the first request from the dispatch is activated | |
1070 | * and the slice time set. | |
1071 | */ | |
a6151c3a | 1072 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
1073 | { |
1074 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 1075 | return false; |
9a7f38c4 | 1076 | if (ktime_get_ns() < cfqq->slice_end) |
c1e44756 | 1077 | return false; |
44f7c160 | 1078 | |
c1e44756 | 1079 | return true; |
44f7c160 JA |
1080 | } |
1081 | ||
1da177e4 | 1082 | /* |
5e705374 | 1083 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 1084 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 1085 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 1086 | */ |
5e705374 | 1087 | static struct request * |
cf7c25cf | 1088 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 1089 | { |
cf7c25cf | 1090 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 1091 | unsigned long back_max; |
e8a99053 AM |
1092 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
1093 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
1094 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 1095 | |
5e705374 JA |
1096 | if (rq1 == NULL || rq1 == rq2) |
1097 | return rq2; | |
1098 | if (rq2 == NULL) | |
1099 | return rq1; | |
9c2c38a1 | 1100 | |
229836bd NK |
1101 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
1102 | return rq_is_sync(rq1) ? rq1 : rq2; | |
1103 | ||
65299a3b CH |
1104 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
1105 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 1106 | |
83096ebf TH |
1107 | s1 = blk_rq_pos(rq1); |
1108 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 1109 | |
1da177e4 LT |
1110 | /* |
1111 | * by definition, 1KiB is 2 sectors | |
1112 | */ | |
1113 | back_max = cfqd->cfq_back_max * 2; | |
1114 | ||
1115 | /* | |
1116 | * Strict one way elevator _except_ in the case where we allow | |
1117 | * short backward seeks which are biased as twice the cost of a | |
1118 | * similar forward seek. | |
1119 | */ | |
1120 | if (s1 >= last) | |
1121 | d1 = s1 - last; | |
1122 | else if (s1 + back_max >= last) | |
1123 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
1124 | else | |
e8a99053 | 1125 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
1126 | |
1127 | if (s2 >= last) | |
1128 | d2 = s2 - last; | |
1129 | else if (s2 + back_max >= last) | |
1130 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
1131 | else | |
e8a99053 | 1132 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
1133 | |
1134 | /* Found required data */ | |
e8a99053 AM |
1135 | |
1136 | /* | |
1137 | * By doing switch() on the bit mask "wrap" we avoid having to | |
1138 | * check two variables for all permutations: --> faster! | |
1139 | */ | |
1140 | switch (wrap) { | |
5e705374 | 1141 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 1142 | if (d1 < d2) |
5e705374 | 1143 | return rq1; |
e8a99053 | 1144 | else if (d2 < d1) |
5e705374 | 1145 | return rq2; |
e8a99053 AM |
1146 | else { |
1147 | if (s1 >= s2) | |
5e705374 | 1148 | return rq1; |
e8a99053 | 1149 | else |
5e705374 | 1150 | return rq2; |
e8a99053 | 1151 | } |
1da177e4 | 1152 | |
e8a99053 | 1153 | case CFQ_RQ2_WRAP: |
5e705374 | 1154 | return rq1; |
e8a99053 | 1155 | case CFQ_RQ1_WRAP: |
5e705374 JA |
1156 | return rq2; |
1157 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
1158 | default: |
1159 | /* | |
1160 | * Since both rqs are wrapped, | |
1161 | * start with the one that's further behind head | |
1162 | * (--> only *one* back seek required), | |
1163 | * since back seek takes more time than forward. | |
1164 | */ | |
1165 | if (s1 <= s2) | |
5e705374 | 1166 | return rq1; |
1da177e4 | 1167 | else |
5e705374 | 1168 | return rq2; |
1da177e4 LT |
1169 | } |
1170 | } | |
1171 | ||
498d3aa2 JA |
1172 | /* |
1173 | * The below is leftmost cache rbtree addon | |
1174 | */ | |
0871714e | 1175 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 1176 | { |
615f0259 VG |
1177 | /* Service tree is empty */ |
1178 | if (!root->count) | |
1179 | return NULL; | |
1180 | ||
cc09e299 JA |
1181 | if (!root->left) |
1182 | root->left = rb_first(&root->rb); | |
1183 | ||
0871714e JA |
1184 | if (root->left) |
1185 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
1186 | ||
1187 | return NULL; | |
cc09e299 JA |
1188 | } |
1189 | ||
1fa8f6d6 VG |
1190 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
1191 | { | |
1192 | if (!root->left) | |
1193 | root->left = rb_first(&root->rb); | |
1194 | ||
1195 | if (root->left) | |
1196 | return rb_entry_cfqg(root->left); | |
1197 | ||
1198 | return NULL; | |
1199 | } | |
1200 | ||
a36e71f9 JA |
1201 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
1202 | { | |
1203 | rb_erase(n, root); | |
1204 | RB_CLEAR_NODE(n); | |
1205 | } | |
1206 | ||
cc09e299 JA |
1207 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
1208 | { | |
1209 | if (root->left == n) | |
1210 | root->left = NULL; | |
a36e71f9 | 1211 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 1212 | --root->count; |
cc09e299 JA |
1213 | } |
1214 | ||
1da177e4 LT |
1215 | /* |
1216 | * would be nice to take fifo expire time into account as well | |
1217 | */ | |
5e705374 JA |
1218 | static struct request * |
1219 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1220 | struct request *last) | |
1da177e4 | 1221 | { |
21183b07 JA |
1222 | struct rb_node *rbnext = rb_next(&last->rb_node); |
1223 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 1224 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 1225 | |
21183b07 | 1226 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
1227 | |
1228 | if (rbprev) | |
5e705374 | 1229 | prev = rb_entry_rq(rbprev); |
1da177e4 | 1230 | |
21183b07 | 1231 | if (rbnext) |
5e705374 | 1232 | next = rb_entry_rq(rbnext); |
21183b07 JA |
1233 | else { |
1234 | rbnext = rb_first(&cfqq->sort_list); | |
1235 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 1236 | next = rb_entry_rq(rbnext); |
21183b07 | 1237 | } |
1da177e4 | 1238 | |
cf7c25cf | 1239 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
1240 | } |
1241 | ||
9a7f38c4 JM |
1242 | static u64 cfq_slice_offset(struct cfq_data *cfqd, |
1243 | struct cfq_queue *cfqq) | |
1da177e4 | 1244 | { |
d9e7620e JA |
1245 | /* |
1246 | * just an approximation, should be ok. | |
1247 | */ | |
cdb16e8f | 1248 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 1249 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
1250 | } |
1251 | ||
1fa8f6d6 VG |
1252 | static inline s64 |
1253 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1254 | { | |
1255 | return cfqg->vdisktime - st->min_vdisktime; | |
1256 | } | |
1257 | ||
1258 | static void | |
1259 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1260 | { | |
1261 | struct rb_node **node = &st->rb.rb_node; | |
1262 | struct rb_node *parent = NULL; | |
1263 | struct cfq_group *__cfqg; | |
1264 | s64 key = cfqg_key(st, cfqg); | |
1265 | int left = 1; | |
1266 | ||
1267 | while (*node != NULL) { | |
1268 | parent = *node; | |
1269 | __cfqg = rb_entry_cfqg(parent); | |
1270 | ||
1271 | if (key < cfqg_key(st, __cfqg)) | |
1272 | node = &parent->rb_left; | |
1273 | else { | |
1274 | node = &parent->rb_right; | |
1275 | left = 0; | |
1276 | } | |
1277 | } | |
1278 | ||
1279 | if (left) | |
1280 | st->left = &cfqg->rb_node; | |
1281 | ||
1282 | rb_link_node(&cfqg->rb_node, parent, node); | |
1283 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
1284 | } | |
1285 | ||
7b5af5cf TM |
1286 | /* |
1287 | * This has to be called only on activation of cfqg | |
1288 | */ | |
1fa8f6d6 | 1289 | static void |
8184f93e JT |
1290 | cfq_update_group_weight(struct cfq_group *cfqg) |
1291 | { | |
3381cb8d | 1292 | if (cfqg->new_weight) { |
8184f93e | 1293 | cfqg->weight = cfqg->new_weight; |
3381cb8d | 1294 | cfqg->new_weight = 0; |
8184f93e | 1295 | } |
e15693ef TM |
1296 | } |
1297 | ||
1298 | static void | |
1299 | cfq_update_group_leaf_weight(struct cfq_group *cfqg) | |
1300 | { | |
1301 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
e71357e1 TH |
1302 | |
1303 | if (cfqg->new_leaf_weight) { | |
1304 | cfqg->leaf_weight = cfqg->new_leaf_weight; | |
1305 | cfqg->new_leaf_weight = 0; | |
1306 | } | |
8184f93e JT |
1307 | } |
1308 | ||
1309 | static void | |
1310 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1311 | { | |
1d3650f7 | 1312 | unsigned int vfr = 1 << CFQ_SERVICE_SHIFT; /* start with 1 */ |
7918ffb5 | 1313 | struct cfq_group *pos = cfqg; |
1d3650f7 | 1314 | struct cfq_group *parent; |
7918ffb5 TH |
1315 | bool propagate; |
1316 | ||
1317 | /* add to the service tree */ | |
8184f93e JT |
1318 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
1319 | ||
7b5af5cf TM |
1320 | /* |
1321 | * Update leaf_weight. We cannot update weight at this point | |
1322 | * because cfqg might already have been activated and is | |
1323 | * contributing its current weight to the parent's child_weight. | |
1324 | */ | |
e15693ef | 1325 | cfq_update_group_leaf_weight(cfqg); |
8184f93e | 1326 | __cfq_group_service_tree_add(st, cfqg); |
7918ffb5 TH |
1327 | |
1328 | /* | |
1d3650f7 TH |
1329 | * Activate @cfqg and calculate the portion of vfraction @cfqg is |
1330 | * entitled to. vfraction is calculated by walking the tree | |
1331 | * towards the root calculating the fraction it has at each level. | |
1332 | * The compounded ratio is how much vfraction @cfqg owns. | |
1333 | * | |
1334 | * Start with the proportion tasks in this cfqg has against active | |
1335 | * children cfqgs - its leaf_weight against children_weight. | |
7918ffb5 TH |
1336 | */ |
1337 | propagate = !pos->nr_active++; | |
1338 | pos->children_weight += pos->leaf_weight; | |
1d3650f7 | 1339 | vfr = vfr * pos->leaf_weight / pos->children_weight; |
7918ffb5 | 1340 | |
1d3650f7 TH |
1341 | /* |
1342 | * Compound ->weight walking up the tree. Both activation and | |
1343 | * vfraction calculation are done in the same loop. Propagation | |
1344 | * stops once an already activated node is met. vfraction | |
1345 | * calculation should always continue to the root. | |
1346 | */ | |
d02f7aa8 | 1347 | while ((parent = cfqg_parent(pos))) { |
1d3650f7 | 1348 | if (propagate) { |
e15693ef | 1349 | cfq_update_group_weight(pos); |
1d3650f7 TH |
1350 | propagate = !parent->nr_active++; |
1351 | parent->children_weight += pos->weight; | |
1352 | } | |
1353 | vfr = vfr * pos->weight / parent->children_weight; | |
7918ffb5 TH |
1354 | pos = parent; |
1355 | } | |
1d3650f7 TH |
1356 | |
1357 | cfqg->vfraction = max_t(unsigned, vfr, 1); | |
8184f93e JT |
1358 | } |
1359 | ||
5be6b756 HT |
1360 | static inline u64 cfq_get_cfqg_vdisktime_delay(struct cfq_data *cfqd) |
1361 | { | |
1362 | if (!iops_mode(cfqd)) | |
1363 | return CFQ_SLICE_MODE_GROUP_DELAY; | |
1364 | else | |
1365 | return CFQ_IOPS_MODE_GROUP_DELAY; | |
1366 | } | |
1367 | ||
8184f93e JT |
1368 | static void |
1369 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
1370 | { |
1371 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1372 | struct cfq_group *__cfqg; | |
1373 | struct rb_node *n; | |
1374 | ||
1375 | cfqg->nr_cfqq++; | |
760701bf | 1376 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
1377 | return; |
1378 | ||
1379 | /* | |
1380 | * Currently put the group at the end. Later implement something | |
1381 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 1382 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
1383 | */ |
1384 | n = rb_last(&st->rb); | |
1385 | if (n) { | |
1386 | __cfqg = rb_entry_cfqg(n); | |
5be6b756 HT |
1387 | cfqg->vdisktime = __cfqg->vdisktime + |
1388 | cfq_get_cfqg_vdisktime_delay(cfqd); | |
1fa8f6d6 VG |
1389 | } else |
1390 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
1391 | cfq_group_service_tree_add(st, cfqg); |
1392 | } | |
1fa8f6d6 | 1393 | |
8184f93e JT |
1394 | static void |
1395 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1396 | { | |
7918ffb5 TH |
1397 | struct cfq_group *pos = cfqg; |
1398 | bool propagate; | |
1399 | ||
1400 | /* | |
1401 | * Undo activation from cfq_group_service_tree_add(). Deactivate | |
1402 | * @cfqg and propagate deactivation upwards. | |
1403 | */ | |
1404 | propagate = !--pos->nr_active; | |
1405 | pos->children_weight -= pos->leaf_weight; | |
1406 | ||
1407 | while (propagate) { | |
d02f7aa8 | 1408 | struct cfq_group *parent = cfqg_parent(pos); |
7918ffb5 TH |
1409 | |
1410 | /* @pos has 0 nr_active at this point */ | |
1411 | WARN_ON_ONCE(pos->children_weight); | |
1d3650f7 | 1412 | pos->vfraction = 0; |
7918ffb5 TH |
1413 | |
1414 | if (!parent) | |
1415 | break; | |
1416 | ||
1417 | propagate = !--parent->nr_active; | |
1418 | parent->children_weight -= pos->weight; | |
1419 | pos = parent; | |
1420 | } | |
1421 | ||
1422 | /* remove from the service tree */ | |
8184f93e JT |
1423 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1424 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
1425 | } |
1426 | ||
1427 | static void | |
8184f93e | 1428 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
1429 | { |
1430 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1431 | ||
1432 | BUG_ON(cfqg->nr_cfqq < 1); | |
1433 | cfqg->nr_cfqq--; | |
25bc6b07 | 1434 | |
1fa8f6d6 VG |
1435 | /* If there are other cfq queues under this group, don't delete it */ |
1436 | if (cfqg->nr_cfqq) | |
1437 | return; | |
1438 | ||
2868ef7b | 1439 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 1440 | cfq_group_service_tree_del(st, cfqg); |
4d2ceea4 | 1441 | cfqg->saved_wl_slice = 0; |
155fead9 | 1442 | cfqg_stats_update_dequeue(cfqg); |
dae739eb VG |
1443 | } |
1444 | ||
9a7f38c4 JM |
1445 | static inline u64 cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
1446 | u64 *unaccounted_time) | |
dae739eb | 1447 | { |
9a7f38c4 JM |
1448 | u64 slice_used; |
1449 | u64 now = ktime_get_ns(); | |
dae739eb VG |
1450 | |
1451 | /* | |
1452 | * Queue got expired before even a single request completed or | |
1453 | * got expired immediately after first request completion. | |
1454 | */ | |
9a7f38c4 | 1455 | if (!cfqq->slice_start || cfqq->slice_start == now) { |
dae739eb VG |
1456 | /* |
1457 | * Also charge the seek time incurred to the group, otherwise | |
1458 | * if there are mutiple queues in the group, each can dispatch | |
1459 | * a single request on seeky media and cause lots of seek time | |
1460 | * and group will never know it. | |
1461 | */ | |
0b31c10c JK |
1462 | slice_used = max_t(u64, (now - cfqq->dispatch_start), |
1463 | jiffies_to_nsecs(1)); | |
dae739eb | 1464 | } else { |
9a7f38c4 | 1465 | slice_used = now - cfqq->slice_start; |
167400d3 JT |
1466 | if (slice_used > cfqq->allocated_slice) { |
1467 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 1468 | slice_used = cfqq->allocated_slice; |
167400d3 | 1469 | } |
9a7f38c4 | 1470 | if (cfqq->slice_start > cfqq->dispatch_start) |
167400d3 JT |
1471 | *unaccounted_time += cfqq->slice_start - |
1472 | cfqq->dispatch_start; | |
dae739eb VG |
1473 | } |
1474 | ||
dae739eb VG |
1475 | return slice_used; |
1476 | } | |
1477 | ||
1478 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1479 | struct cfq_queue *cfqq) |
dae739eb VG |
1480 | { |
1481 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
9a7f38c4 | 1482 | u64 used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1483 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1484 | - cfqg->service_tree_idle.count; | |
1d3650f7 | 1485 | unsigned int vfr; |
9a7f38c4 | 1486 | u64 now = ktime_get_ns(); |
f26bd1f0 VG |
1487 | |
1488 | BUG_ON(nr_sync < 0); | |
167400d3 | 1489 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1490 | |
02b35081 VG |
1491 | if (iops_mode(cfqd)) |
1492 | charge = cfqq->slice_dispatch; | |
1493 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1494 | charge = cfqq->allocated_slice; | |
dae739eb | 1495 | |
1d3650f7 TH |
1496 | /* |
1497 | * Can't update vdisktime while on service tree and cfqg->vfraction | |
1498 | * is valid only while on it. Cache vfr, leave the service tree, | |
1499 | * update vdisktime and go back on. The re-addition to the tree | |
1500 | * will also update the weights as necessary. | |
1501 | */ | |
1502 | vfr = cfqg->vfraction; | |
8184f93e | 1503 | cfq_group_service_tree_del(st, cfqg); |
1d3650f7 | 1504 | cfqg->vdisktime += cfqg_scale_charge(charge, vfr); |
8184f93e | 1505 | cfq_group_service_tree_add(st, cfqg); |
dae739eb VG |
1506 | |
1507 | /* This group is being expired. Save the context */ | |
9a7f38c4 JM |
1508 | if (cfqd->workload_expires > now) { |
1509 | cfqg->saved_wl_slice = cfqd->workload_expires - now; | |
4d2ceea4 VG |
1510 | cfqg->saved_wl_type = cfqd->serving_wl_type; |
1511 | cfqg->saved_wl_class = cfqd->serving_wl_class; | |
dae739eb | 1512 | } else |
4d2ceea4 | 1513 | cfqg->saved_wl_slice = 0; |
2868ef7b VG |
1514 | |
1515 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1516 | st->min_vdisktime); | |
fd16d263 | 1517 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
9a7f38c4 | 1518 | "sl_used=%llu disp=%llu charge=%llu iops=%u sect=%lu", |
fd16d263 JP |
1519 | used_sl, cfqq->slice_dispatch, charge, |
1520 | iops_mode(cfqd), cfqq->nr_sectors); | |
155fead9 TH |
1521 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
1522 | cfqg_stats_set_start_empty_time(cfqg); | |
1fa8f6d6 VG |
1523 | } |
1524 | ||
f51b802c TH |
1525 | /** |
1526 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1527 | * @cfqg: cfq_group to initialize | |
1528 | * | |
1529 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1530 | * is enabled or not. | |
1531 | */ | |
1532 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1533 | { | |
1534 | struct cfq_rb_root *st; | |
1535 | int i, j; | |
1536 | ||
1537 | for_each_cfqg_st(cfqg, i, j, st) | |
1538 | *st = CFQ_RB_ROOT; | |
1539 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1540 | ||
9a7f38c4 | 1541 | cfqg->ttime.last_end_request = ktime_get_ns(); |
f51b802c TH |
1542 | } |
1543 | ||
25fb5169 | 1544 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
69d7fde5 TH |
1545 | static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val, |
1546 | bool on_dfl, bool reset_dev, bool is_leaf_weight); | |
1547 | ||
24bdb8ef | 1548 | static void cfqg_stats_exit(struct cfqg_stats *stats) |
90d3839b | 1549 | { |
24bdb8ef TH |
1550 | blkg_rwstat_exit(&stats->merged); |
1551 | blkg_rwstat_exit(&stats->service_time); | |
1552 | blkg_rwstat_exit(&stats->wait_time); | |
1553 | blkg_rwstat_exit(&stats->queued); | |
24bdb8ef TH |
1554 | blkg_stat_exit(&stats->time); |
1555 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
1556 | blkg_stat_exit(&stats->unaccounted_time); | |
1557 | blkg_stat_exit(&stats->avg_queue_size_sum); | |
1558 | blkg_stat_exit(&stats->avg_queue_size_samples); | |
1559 | blkg_stat_exit(&stats->dequeue); | |
1560 | blkg_stat_exit(&stats->group_wait_time); | |
1561 | blkg_stat_exit(&stats->idle_time); | |
1562 | blkg_stat_exit(&stats->empty_time); | |
1563 | #endif | |
1564 | } | |
1565 | ||
1566 | static int cfqg_stats_init(struct cfqg_stats *stats, gfp_t gfp) | |
1567 | { | |
77ea7338 | 1568 | if (blkg_rwstat_init(&stats->merged, gfp) || |
24bdb8ef TH |
1569 | blkg_rwstat_init(&stats->service_time, gfp) || |
1570 | blkg_rwstat_init(&stats->wait_time, gfp) || | |
1571 | blkg_rwstat_init(&stats->queued, gfp) || | |
24bdb8ef TH |
1572 | blkg_stat_init(&stats->time, gfp)) |
1573 | goto err; | |
90d3839b PZ |
1574 | |
1575 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
24bdb8ef TH |
1576 | if (blkg_stat_init(&stats->unaccounted_time, gfp) || |
1577 | blkg_stat_init(&stats->avg_queue_size_sum, gfp) || | |
1578 | blkg_stat_init(&stats->avg_queue_size_samples, gfp) || | |
1579 | blkg_stat_init(&stats->dequeue, gfp) || | |
1580 | blkg_stat_init(&stats->group_wait_time, gfp) || | |
1581 | blkg_stat_init(&stats->idle_time, gfp) || | |
1582 | blkg_stat_init(&stats->empty_time, gfp)) | |
1583 | goto err; | |
90d3839b | 1584 | #endif |
24bdb8ef TH |
1585 | return 0; |
1586 | err: | |
1587 | cfqg_stats_exit(stats); | |
1588 | return -ENOMEM; | |
90d3839b PZ |
1589 | } |
1590 | ||
e4a9bde9 TH |
1591 | static struct blkcg_policy_data *cfq_cpd_alloc(gfp_t gfp) |
1592 | { | |
1593 | struct cfq_group_data *cgd; | |
1594 | ||
ebc4ff66 | 1595 | cgd = kzalloc(sizeof(*cgd), gfp); |
e4a9bde9 TH |
1596 | if (!cgd) |
1597 | return NULL; | |
1598 | return &cgd->cpd; | |
1599 | } | |
1600 | ||
81437648 | 1601 | static void cfq_cpd_init(struct blkcg_policy_data *cpd) |
e48453c3 | 1602 | { |
81437648 | 1603 | struct cfq_group_data *cgd = cpd_to_cfqgd(cpd); |
9e10a130 | 1604 | unsigned int weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ? |
69d7fde5 | 1605 | CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL; |
e48453c3 | 1606 | |
69d7fde5 TH |
1607 | if (cpd_to_blkcg(cpd) == &blkcg_root) |
1608 | weight *= 2; | |
1609 | ||
1610 | cgd->weight = weight; | |
1611 | cgd->leaf_weight = weight; | |
e48453c3 AA |
1612 | } |
1613 | ||
e4a9bde9 TH |
1614 | static void cfq_cpd_free(struct blkcg_policy_data *cpd) |
1615 | { | |
1616 | kfree(cpd_to_cfqgd(cpd)); | |
1617 | } | |
1618 | ||
69d7fde5 TH |
1619 | static void cfq_cpd_bind(struct blkcg_policy_data *cpd) |
1620 | { | |
1621 | struct blkcg *blkcg = cpd_to_blkcg(cpd); | |
9e10a130 | 1622 | bool on_dfl = cgroup_subsys_on_dfl(io_cgrp_subsys); |
69d7fde5 TH |
1623 | unsigned int weight = on_dfl ? CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL; |
1624 | ||
1625 | if (blkcg == &blkcg_root) | |
1626 | weight *= 2; | |
1627 | ||
1628 | WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, false)); | |
1629 | WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, true)); | |
1630 | } | |
1631 | ||
001bea73 TH |
1632 | static struct blkg_policy_data *cfq_pd_alloc(gfp_t gfp, int node) |
1633 | { | |
b2ce2643 TH |
1634 | struct cfq_group *cfqg; |
1635 | ||
1636 | cfqg = kzalloc_node(sizeof(*cfqg), gfp, node); | |
1637 | if (!cfqg) | |
1638 | return NULL; | |
1639 | ||
1640 | cfq_init_cfqg_base(cfqg); | |
24bdb8ef TH |
1641 | if (cfqg_stats_init(&cfqg->stats, gfp)) { |
1642 | kfree(cfqg); | |
1643 | return NULL; | |
1644 | } | |
b2ce2643 TH |
1645 | |
1646 | return &cfqg->pd; | |
001bea73 TH |
1647 | } |
1648 | ||
a9520cd6 | 1649 | static void cfq_pd_init(struct blkg_policy_data *pd) |
f469a7b4 | 1650 | { |
a9520cd6 TH |
1651 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
1652 | struct cfq_group_data *cgd = blkcg_to_cfqgd(pd->blkg->blkcg); | |
25fb5169 | 1653 | |
e48453c3 AA |
1654 | cfqg->weight = cgd->weight; |
1655 | cfqg->leaf_weight = cgd->leaf_weight; | |
25fb5169 VG |
1656 | } |
1657 | ||
a9520cd6 | 1658 | static void cfq_pd_offline(struct blkg_policy_data *pd) |
0b39920b | 1659 | { |
a9520cd6 | 1660 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
60a83707 TH |
1661 | int i; |
1662 | ||
1663 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
1664 | if (cfqg->async_cfqq[0][i]) | |
1665 | cfq_put_queue(cfqg->async_cfqq[0][i]); | |
1666 | if (cfqg->async_cfqq[1][i]) | |
1667 | cfq_put_queue(cfqg->async_cfqq[1][i]); | |
1668 | } | |
1669 | ||
1670 | if (cfqg->async_idle_cfqq) | |
1671 | cfq_put_queue(cfqg->async_idle_cfqq); | |
1672 | ||
0b39920b TH |
1673 | /* |
1674 | * @blkg is going offline and will be ignored by | |
1675 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
1676 | * that they don't get lost. If IOs complete after this point, the | |
1677 | * stats for them will be lost. Oh well... | |
1678 | */ | |
60a83707 | 1679 | cfqg_stats_xfer_dead(cfqg); |
0b39920b TH |
1680 | } |
1681 | ||
001bea73 TH |
1682 | static void cfq_pd_free(struct blkg_policy_data *pd) |
1683 | { | |
24bdb8ef TH |
1684 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
1685 | ||
1686 | cfqg_stats_exit(&cfqg->stats); | |
1687 | return kfree(cfqg); | |
001bea73 TH |
1688 | } |
1689 | ||
a9520cd6 | 1690 | static void cfq_pd_reset_stats(struct blkg_policy_data *pd) |
689665af | 1691 | { |
a9520cd6 | 1692 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
689665af TH |
1693 | |
1694 | cfqg_stats_reset(&cfqg->stats); | |
25fb5169 VG |
1695 | } |
1696 | ||
ae118896 TH |
1697 | static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd, |
1698 | struct blkcg *blkcg) | |
25fb5169 | 1699 | { |
ae118896 | 1700 | struct blkcg_gq *blkg; |
f469a7b4 | 1701 | |
ae118896 TH |
1702 | blkg = blkg_lookup(blkcg, cfqd->queue); |
1703 | if (likely(blkg)) | |
1704 | return blkg_to_cfqg(blkg); | |
1705 | return NULL; | |
25fb5169 VG |
1706 | } |
1707 | ||
1708 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1709 | { | |
25fb5169 | 1710 | cfqq->cfqg = cfqg; |
b1c35769 | 1711 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1712 | cfqg_get(cfqg); |
b1c35769 VG |
1713 | } |
1714 | ||
f95a04af TH |
1715 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, |
1716 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1717 | { |
f95a04af | 1718 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
3381cb8d TH |
1719 | |
1720 | if (!cfqg->dev_weight) | |
60c2bc2d | 1721 | return 0; |
f95a04af | 1722 | return __blkg_prfill_u64(sf, pd, cfqg->dev_weight); |
60c2bc2d TH |
1723 | } |
1724 | ||
2da8ca82 | 1725 | static int cfqg_print_weight_device(struct seq_file *sf, void *v) |
60c2bc2d | 1726 | { |
2da8ca82 TH |
1727 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1728 | cfqg_prfill_weight_device, &blkcg_policy_cfq, | |
1729 | 0, false); | |
60c2bc2d TH |
1730 | return 0; |
1731 | } | |
1732 | ||
e71357e1 TH |
1733 | static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf, |
1734 | struct blkg_policy_data *pd, int off) | |
1735 | { | |
1736 | struct cfq_group *cfqg = pd_to_cfqg(pd); | |
1737 | ||
1738 | if (!cfqg->dev_leaf_weight) | |
1739 | return 0; | |
1740 | return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight); | |
1741 | } | |
1742 | ||
2da8ca82 | 1743 | static int cfqg_print_leaf_weight_device(struct seq_file *sf, void *v) |
e71357e1 | 1744 | { |
2da8ca82 TH |
1745 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1746 | cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq, | |
1747 | 0, false); | |
e71357e1 TH |
1748 | return 0; |
1749 | } | |
1750 | ||
2da8ca82 | 1751 | static int cfq_print_weight(struct seq_file *sf, void *v) |
60c2bc2d | 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; | |
e48453c3 | 1756 | |
9470e4a6 JA |
1757 | if (cgd) |
1758 | val = cgd->weight; | |
1759 | ||
1760 | seq_printf(sf, "%u\n", val); | |
60c2bc2d TH |
1761 | return 0; |
1762 | } | |
1763 | ||
2da8ca82 | 1764 | static int cfq_print_leaf_weight(struct seq_file *sf, void *v) |
e71357e1 | 1765 | { |
e48453c3 | 1766 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
9470e4a6 JA |
1767 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); |
1768 | unsigned int val = 0; | |
1769 | ||
1770 | if (cgd) | |
1771 | val = cgd->leaf_weight; | |
e48453c3 | 1772 | |
9470e4a6 | 1773 | seq_printf(sf, "%u\n", val); |
e71357e1 TH |
1774 | return 0; |
1775 | } | |
1776 | ||
451af504 TH |
1777 | static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of, |
1778 | char *buf, size_t nbytes, loff_t off, | |
2ee867dc | 1779 | bool on_dfl, bool is_leaf_weight) |
60c2bc2d | 1780 | { |
69d7fde5 TH |
1781 | unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN; |
1782 | unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX; | |
451af504 | 1783 | struct blkcg *blkcg = css_to_blkcg(of_css(of)); |
60c2bc2d | 1784 | struct blkg_conf_ctx ctx; |
3381cb8d | 1785 | struct cfq_group *cfqg; |
e48453c3 | 1786 | struct cfq_group_data *cfqgd; |
60c2bc2d | 1787 | int ret; |
36aa9e5f | 1788 | u64 v; |
60c2bc2d | 1789 | |
3c798398 | 1790 | ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx); |
60c2bc2d TH |
1791 | if (ret) |
1792 | return ret; | |
1793 | ||
2ee867dc TH |
1794 | if (sscanf(ctx.body, "%llu", &v) == 1) { |
1795 | /* require "default" on dfl */ | |
1796 | ret = -ERANGE; | |
1797 | if (!v && on_dfl) | |
1798 | goto out_finish; | |
1799 | } else if (!strcmp(strim(ctx.body), "default")) { | |
1800 | v = 0; | |
1801 | } else { | |
1802 | ret = -EINVAL; | |
36aa9e5f | 1803 | goto out_finish; |
2ee867dc | 1804 | } |
36aa9e5f | 1805 | |
3381cb8d | 1806 | cfqg = blkg_to_cfqg(ctx.blkg); |
e48453c3 | 1807 | cfqgd = blkcg_to_cfqgd(blkcg); |
ae994ea9 | 1808 | |
20386ce0 | 1809 | ret = -ERANGE; |
69d7fde5 | 1810 | if (!v || (v >= min && v <= max)) { |
e71357e1 | 1811 | if (!is_leaf_weight) { |
36aa9e5f TH |
1812 | cfqg->dev_weight = v; |
1813 | cfqg->new_weight = v ?: cfqgd->weight; | |
e71357e1 | 1814 | } else { |
36aa9e5f TH |
1815 | cfqg->dev_leaf_weight = v; |
1816 | cfqg->new_leaf_weight = v ?: cfqgd->leaf_weight; | |
e71357e1 | 1817 | } |
60c2bc2d TH |
1818 | ret = 0; |
1819 | } | |
36aa9e5f | 1820 | out_finish: |
60c2bc2d | 1821 | blkg_conf_finish(&ctx); |
451af504 | 1822 | return ret ?: nbytes; |
60c2bc2d TH |
1823 | } |
1824 | ||
451af504 TH |
1825 | static ssize_t cfqg_set_weight_device(struct kernfs_open_file *of, |
1826 | char *buf, size_t nbytes, loff_t off) | |
e71357e1 | 1827 | { |
2ee867dc | 1828 | return __cfqg_set_weight_device(of, buf, nbytes, off, false, false); |
e71357e1 TH |
1829 | } |
1830 | ||
451af504 TH |
1831 | static ssize_t cfqg_set_leaf_weight_device(struct kernfs_open_file *of, |
1832 | char *buf, size_t nbytes, loff_t off) | |
e71357e1 | 1833 | { |
2ee867dc | 1834 | return __cfqg_set_weight_device(of, buf, nbytes, off, false, true); |
e71357e1 TH |
1835 | } |
1836 | ||
dd165eb3 | 1837 | static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val, |
69d7fde5 | 1838 | bool on_dfl, bool reset_dev, bool is_leaf_weight) |
60c2bc2d | 1839 | { |
69d7fde5 TH |
1840 | unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN; |
1841 | unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX; | |
182446d0 | 1842 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 1843 | struct blkcg_gq *blkg; |
e48453c3 | 1844 | struct cfq_group_data *cfqgd; |
ae994ea9 | 1845 | int ret = 0; |
60c2bc2d | 1846 | |
69d7fde5 TH |
1847 | if (val < min || val > max) |
1848 | return -ERANGE; | |
60c2bc2d TH |
1849 | |
1850 | spin_lock_irq(&blkcg->lock); | |
e48453c3 | 1851 | cfqgd = blkcg_to_cfqgd(blkcg); |
ae994ea9 JA |
1852 | if (!cfqgd) { |
1853 | ret = -EINVAL; | |
1854 | goto out; | |
1855 | } | |
e71357e1 TH |
1856 | |
1857 | if (!is_leaf_weight) | |
e48453c3 | 1858 | cfqgd->weight = val; |
e71357e1 | 1859 | else |
e48453c3 | 1860 | cfqgd->leaf_weight = val; |
60c2bc2d | 1861 | |
b67bfe0d | 1862 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
3381cb8d | 1863 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
60c2bc2d | 1864 | |
e71357e1 TH |
1865 | if (!cfqg) |
1866 | continue; | |
1867 | ||
1868 | if (!is_leaf_weight) { | |
69d7fde5 TH |
1869 | if (reset_dev) |
1870 | cfqg->dev_weight = 0; | |
e71357e1 | 1871 | if (!cfqg->dev_weight) |
e48453c3 | 1872 | cfqg->new_weight = cfqgd->weight; |
e71357e1 | 1873 | } else { |
69d7fde5 TH |
1874 | if (reset_dev) |
1875 | cfqg->dev_leaf_weight = 0; | |
e71357e1 | 1876 | if (!cfqg->dev_leaf_weight) |
e48453c3 | 1877 | cfqg->new_leaf_weight = cfqgd->leaf_weight; |
e71357e1 | 1878 | } |
60c2bc2d TH |
1879 | } |
1880 | ||
ae994ea9 | 1881 | out: |
60c2bc2d | 1882 | spin_unlock_irq(&blkcg->lock); |
ae994ea9 | 1883 | return ret; |
60c2bc2d TH |
1884 | } |
1885 | ||
182446d0 TH |
1886 | static int cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft, |
1887 | u64 val) | |
e71357e1 | 1888 | { |
69d7fde5 | 1889 | return __cfq_set_weight(css, val, false, false, false); |
e71357e1 TH |
1890 | } |
1891 | ||
182446d0 TH |
1892 | static int cfq_set_leaf_weight(struct cgroup_subsys_state *css, |
1893 | struct cftype *cft, u64 val) | |
e71357e1 | 1894 | { |
69d7fde5 | 1895 | return __cfq_set_weight(css, val, false, false, true); |
e71357e1 TH |
1896 | } |
1897 | ||
2da8ca82 | 1898 | static int cfqg_print_stat(struct seq_file *sf, void *v) |
5bc4afb1 | 1899 | { |
2da8ca82 TH |
1900 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat, |
1901 | &blkcg_policy_cfq, seq_cft(sf)->private, false); | |
5bc4afb1 TH |
1902 | return 0; |
1903 | } | |
1904 | ||
2da8ca82 | 1905 | static int cfqg_print_rwstat(struct seq_file *sf, void *v) |
5bc4afb1 | 1906 | { |
2da8ca82 TH |
1907 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat, |
1908 | &blkcg_policy_cfq, seq_cft(sf)->private, true); | |
5bc4afb1 TH |
1909 | return 0; |
1910 | } | |
1911 | ||
43114018 TH |
1912 | static u64 cfqg_prfill_stat_recursive(struct seq_file *sf, |
1913 | struct blkg_policy_data *pd, int off) | |
1914 | { | |
f12c74ca TH |
1915 | u64 sum = blkg_stat_recursive_sum(pd_to_blkg(pd), |
1916 | &blkcg_policy_cfq, off); | |
43114018 TH |
1917 | return __blkg_prfill_u64(sf, pd, sum); |
1918 | } | |
1919 | ||
1920 | static u64 cfqg_prfill_rwstat_recursive(struct seq_file *sf, | |
1921 | struct blkg_policy_data *pd, int off) | |
1922 | { | |
f12c74ca TH |
1923 | struct blkg_rwstat sum = blkg_rwstat_recursive_sum(pd_to_blkg(pd), |
1924 | &blkcg_policy_cfq, off); | |
43114018 TH |
1925 | return __blkg_prfill_rwstat(sf, pd, &sum); |
1926 | } | |
1927 | ||
2da8ca82 | 1928 | static int cfqg_print_stat_recursive(struct seq_file *sf, void *v) |
43114018 | 1929 | { |
2da8ca82 TH |
1930 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1931 | cfqg_prfill_stat_recursive, &blkcg_policy_cfq, | |
1932 | seq_cft(sf)->private, false); | |
43114018 TH |
1933 | return 0; |
1934 | } | |
1935 | ||
2da8ca82 | 1936 | static int cfqg_print_rwstat_recursive(struct seq_file *sf, void *v) |
43114018 | 1937 | { |
2da8ca82 TH |
1938 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1939 | cfqg_prfill_rwstat_recursive, &blkcg_policy_cfq, | |
1940 | seq_cft(sf)->private, true); | |
43114018 TH |
1941 | return 0; |
1942 | } | |
1943 | ||
702747ca TH |
1944 | static u64 cfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd, |
1945 | int off) | |
1946 | { | |
1947 | u64 sum = blkg_rwstat_total(&pd->blkg->stat_bytes); | |
1948 | ||
1949 | return __blkg_prfill_u64(sf, pd, sum >> 9); | |
1950 | } | |
1951 | ||
1952 | static int cfqg_print_stat_sectors(struct seq_file *sf, void *v) | |
1953 | { | |
1954 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1955 | cfqg_prfill_sectors, &blkcg_policy_cfq, 0, false); | |
1956 | return 0; | |
1957 | } | |
1958 | ||
1959 | static u64 cfqg_prfill_sectors_recursive(struct seq_file *sf, | |
1960 | struct blkg_policy_data *pd, int off) | |
1961 | { | |
1962 | struct blkg_rwstat tmp = blkg_rwstat_recursive_sum(pd->blkg, NULL, | |
1963 | offsetof(struct blkcg_gq, stat_bytes)); | |
1964 | u64 sum = atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) + | |
1965 | atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]); | |
1966 | ||
1967 | return __blkg_prfill_u64(sf, pd, sum >> 9); | |
1968 | } | |
1969 | ||
1970 | static int cfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v) | |
1971 | { | |
1972 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1973 | cfqg_prfill_sectors_recursive, &blkcg_policy_cfq, 0, | |
1974 | false); | |
1975 | return 0; | |
1976 | } | |
1977 | ||
60c2bc2d | 1978 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
f95a04af TH |
1979 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, |
1980 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1981 | { |
f95a04af | 1982 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
155fead9 | 1983 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
60c2bc2d TH |
1984 | u64 v = 0; |
1985 | ||
1986 | if (samples) { | |
155fead9 | 1987 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
f3cff25f | 1988 | v = div64_u64(v, samples); |
60c2bc2d | 1989 | } |
f95a04af | 1990 | __blkg_prfill_u64(sf, pd, v); |
60c2bc2d TH |
1991 | return 0; |
1992 | } | |
1993 | ||
1994 | /* print avg_queue_size */ | |
2da8ca82 | 1995 | static int cfqg_print_avg_queue_size(struct seq_file *sf, void *v) |
60c2bc2d | 1996 | { |
2da8ca82 TH |
1997 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
1998 | cfqg_prfill_avg_queue_size, &blkcg_policy_cfq, | |
1999 | 0, false); | |
60c2bc2d TH |
2000 | return 0; |
2001 | } | |
2002 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
2003 | ||
880f50e2 | 2004 | static struct cftype cfq_blkcg_legacy_files[] = { |
1d3650f7 | 2005 | /* on root, weight is mapped to leaf_weight */ |
60c2bc2d TH |
2006 | { |
2007 | .name = "weight_device", | |
1d3650f7 | 2008 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 2009 | .seq_show = cfqg_print_leaf_weight_device, |
451af504 | 2010 | .write = cfqg_set_leaf_weight_device, |
60c2bc2d TH |
2011 | }, |
2012 | { | |
2013 | .name = "weight", | |
1d3650f7 | 2014 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 2015 | .seq_show = cfq_print_leaf_weight, |
1d3650f7 | 2016 | .write_u64 = cfq_set_leaf_weight, |
60c2bc2d | 2017 | }, |
e71357e1 | 2018 | |
1d3650f7 | 2019 | /* no such mapping necessary for !roots */ |
60c2bc2d TH |
2020 | { |
2021 | .name = "weight_device", | |
1d3650f7 | 2022 | .flags = CFTYPE_NOT_ON_ROOT, |
2da8ca82 | 2023 | .seq_show = cfqg_print_weight_device, |
451af504 | 2024 | .write = cfqg_set_weight_device, |
60c2bc2d TH |
2025 | }, |
2026 | { | |
2027 | .name = "weight", | |
1d3650f7 | 2028 | .flags = CFTYPE_NOT_ON_ROOT, |
2da8ca82 | 2029 | .seq_show = cfq_print_weight, |
3381cb8d | 2030 | .write_u64 = cfq_set_weight, |
60c2bc2d | 2031 | }, |
e71357e1 | 2032 | |
e71357e1 TH |
2033 | { |
2034 | .name = "leaf_weight_device", | |
2da8ca82 | 2035 | .seq_show = cfqg_print_leaf_weight_device, |
451af504 | 2036 | .write = cfqg_set_leaf_weight_device, |
e71357e1 TH |
2037 | }, |
2038 | { | |
2039 | .name = "leaf_weight", | |
2da8ca82 | 2040 | .seq_show = cfq_print_leaf_weight, |
e71357e1 TH |
2041 | .write_u64 = cfq_set_leaf_weight, |
2042 | }, | |
2043 | ||
43114018 | 2044 | /* statistics, covers only the tasks in the cfqg */ |
60c2bc2d TH |
2045 | { |
2046 | .name = "time", | |
5bc4afb1 | 2047 | .private = offsetof(struct cfq_group, stats.time), |
2da8ca82 | 2048 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2049 | }, |
2050 | { | |
2051 | .name = "sectors", | |
702747ca | 2052 | .seq_show = cfqg_print_stat_sectors, |
60c2bc2d TH |
2053 | }, |
2054 | { | |
2055 | .name = "io_service_bytes", | |
77ea7338 TH |
2056 | .private = (unsigned long)&blkcg_policy_cfq, |
2057 | .seq_show = blkg_print_stat_bytes, | |
60c2bc2d TH |
2058 | }, |
2059 | { | |
2060 | .name = "io_serviced", | |
77ea7338 TH |
2061 | .private = (unsigned long)&blkcg_policy_cfq, |
2062 | .seq_show = blkg_print_stat_ios, | |
60c2bc2d TH |
2063 | }, |
2064 | { | |
2065 | .name = "io_service_time", | |
5bc4afb1 | 2066 | .private = offsetof(struct cfq_group, stats.service_time), |
2da8ca82 | 2067 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2068 | }, |
2069 | { | |
2070 | .name = "io_wait_time", | |
5bc4afb1 | 2071 | .private = offsetof(struct cfq_group, stats.wait_time), |
2da8ca82 | 2072 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2073 | }, |
2074 | { | |
2075 | .name = "io_merged", | |
5bc4afb1 | 2076 | .private = offsetof(struct cfq_group, stats.merged), |
2da8ca82 | 2077 | .seq_show = cfqg_print_rwstat, |
60c2bc2d TH |
2078 | }, |
2079 | { | |
2080 | .name = "io_queued", | |
5bc4afb1 | 2081 | .private = offsetof(struct cfq_group, stats.queued), |
2da8ca82 | 2082 | .seq_show = cfqg_print_rwstat, |
60c2bc2d | 2083 | }, |
43114018 TH |
2084 | |
2085 | /* the same statictics which cover the cfqg and its descendants */ | |
2086 | { | |
2087 | .name = "time_recursive", | |
2088 | .private = offsetof(struct cfq_group, stats.time), | |
2da8ca82 | 2089 | .seq_show = cfqg_print_stat_recursive, |
43114018 TH |
2090 | }, |
2091 | { | |
2092 | .name = "sectors_recursive", | |
702747ca | 2093 | .seq_show = cfqg_print_stat_sectors_recursive, |
43114018 TH |
2094 | }, |
2095 | { | |
2096 | .name = "io_service_bytes_recursive", | |
77ea7338 TH |
2097 | .private = (unsigned long)&blkcg_policy_cfq, |
2098 | .seq_show = blkg_print_stat_bytes_recursive, | |
43114018 TH |
2099 | }, |
2100 | { | |
2101 | .name = "io_serviced_recursive", | |
77ea7338 TH |
2102 | .private = (unsigned long)&blkcg_policy_cfq, |
2103 | .seq_show = blkg_print_stat_ios_recursive, | |
43114018 TH |
2104 | }, |
2105 | { | |
2106 | .name = "io_service_time_recursive", | |
2107 | .private = offsetof(struct cfq_group, stats.service_time), | |
2da8ca82 | 2108 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2109 | }, |
2110 | { | |
2111 | .name = "io_wait_time_recursive", | |
2112 | .private = offsetof(struct cfq_group, stats.wait_time), | |
2da8ca82 | 2113 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2114 | }, |
2115 | { | |
2116 | .name = "io_merged_recursive", | |
2117 | .private = offsetof(struct cfq_group, stats.merged), | |
2da8ca82 | 2118 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 TH |
2119 | }, |
2120 | { | |
2121 | .name = "io_queued_recursive", | |
2122 | .private = offsetof(struct cfq_group, stats.queued), | |
2da8ca82 | 2123 | .seq_show = cfqg_print_rwstat_recursive, |
43114018 | 2124 | }, |
60c2bc2d TH |
2125 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
2126 | { | |
2127 | .name = "avg_queue_size", | |
2da8ca82 | 2128 | .seq_show = cfqg_print_avg_queue_size, |
60c2bc2d TH |
2129 | }, |
2130 | { | |
2131 | .name = "group_wait_time", | |
5bc4afb1 | 2132 | .private = offsetof(struct cfq_group, stats.group_wait_time), |
2da8ca82 | 2133 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2134 | }, |
2135 | { | |
2136 | .name = "idle_time", | |
5bc4afb1 | 2137 | .private = offsetof(struct cfq_group, stats.idle_time), |
2da8ca82 | 2138 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2139 | }, |
2140 | { | |
2141 | .name = "empty_time", | |
5bc4afb1 | 2142 | .private = offsetof(struct cfq_group, stats.empty_time), |
2da8ca82 | 2143 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2144 | }, |
2145 | { | |
2146 | .name = "dequeue", | |
5bc4afb1 | 2147 | .private = offsetof(struct cfq_group, stats.dequeue), |
2da8ca82 | 2148 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2149 | }, |
2150 | { | |
2151 | .name = "unaccounted_time", | |
5bc4afb1 | 2152 | .private = offsetof(struct cfq_group, stats.unaccounted_time), |
2da8ca82 | 2153 | .seq_show = cfqg_print_stat, |
60c2bc2d TH |
2154 | }, |
2155 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
2156 | { } /* terminate */ | |
2157 | }; | |
2ee867dc TH |
2158 | |
2159 | static int cfq_print_weight_on_dfl(struct seq_file *sf, void *v) | |
2160 | { | |
2161 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
2162 | struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg); | |
2163 | ||
2164 | seq_printf(sf, "default %u\n", cgd->weight); | |
2165 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_weight_device, | |
2166 | &blkcg_policy_cfq, 0, false); | |
2167 | return 0; | |
2168 | } | |
2169 | ||
2170 | static ssize_t cfq_set_weight_on_dfl(struct kernfs_open_file *of, | |
2171 | char *buf, size_t nbytes, loff_t off) | |
2172 | { | |
2173 | char *endp; | |
2174 | int ret; | |
2175 | u64 v; | |
2176 | ||
2177 | buf = strim(buf); | |
2178 | ||
2179 | /* "WEIGHT" or "default WEIGHT" sets the default weight */ | |
2180 | v = simple_strtoull(buf, &endp, 0); | |
2181 | if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) { | |
69d7fde5 | 2182 | ret = __cfq_set_weight(of_css(of), v, true, false, false); |
2ee867dc TH |
2183 | return ret ?: nbytes; |
2184 | } | |
2185 | ||
2186 | /* "MAJ:MIN WEIGHT" */ | |
2187 | return __cfqg_set_weight_device(of, buf, nbytes, off, true, false); | |
2188 | } | |
2189 | ||
2190 | static struct cftype cfq_blkcg_files[] = { | |
2191 | { | |
2192 | .name = "weight", | |
2193 | .flags = CFTYPE_NOT_ON_ROOT, | |
2194 | .seq_show = cfq_print_weight_on_dfl, | |
2195 | .write = cfq_set_weight_on_dfl, | |
2196 | }, | |
2197 | { } /* terminate */ | |
2198 | }; | |
2199 | ||
25fb5169 | 2200 | #else /* GROUP_IOSCHED */ |
ae118896 TH |
2201 | static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd, |
2202 | struct blkcg *blkcg) | |
25fb5169 | 2203 | { |
f51b802c | 2204 | return cfqd->root_group; |
25fb5169 | 2205 | } |
7f1dc8a2 | 2206 | |
25fb5169 VG |
2207 | static inline void |
2208 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
2209 | cfqq->cfqg = cfqg; | |
2210 | } | |
2211 | ||
2212 | #endif /* GROUP_IOSCHED */ | |
2213 | ||
498d3aa2 | 2214 | /* |
c0324a02 | 2215 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
2216 | * requests waiting to be processed. It is sorted in the order that |
2217 | * we will service the queues. | |
2218 | */ | |
a36e71f9 | 2219 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2220 | bool add_front) |
d9e7620e | 2221 | { |
0871714e JA |
2222 | struct rb_node **p, *parent; |
2223 | struct cfq_queue *__cfqq; | |
9a7f38c4 | 2224 | u64 rb_key; |
34b98d03 | 2225 | struct cfq_rb_root *st; |
498d3aa2 | 2226 | int left; |
dae739eb | 2227 | int new_cfqq = 1; |
9a7f38c4 | 2228 | u64 now = ktime_get_ns(); |
ae30c286 | 2229 | |
34b98d03 | 2230 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq)); |
0871714e JA |
2231 | if (cfq_class_idle(cfqq)) { |
2232 | rb_key = CFQ_IDLE_DELAY; | |
34b98d03 | 2233 | parent = rb_last(&st->rb); |
0871714e JA |
2234 | if (parent && parent != &cfqq->rb_node) { |
2235 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2236 | rb_key += __cfqq->rb_key; | |
2237 | } else | |
9a7f38c4 | 2238 | rb_key += now; |
0871714e | 2239 | } else if (!add_front) { |
b9c8946b JA |
2240 | /* |
2241 | * Get our rb key offset. Subtract any residual slice | |
2242 | * value carried from last service. A negative resid | |
2243 | * count indicates slice overrun, and this should position | |
2244 | * the next service time further away in the tree. | |
2245 | */ | |
9a7f38c4 | 2246 | rb_key = cfq_slice_offset(cfqd, cfqq) + now; |
b9c8946b | 2247 | rb_key -= cfqq->slice_resid; |
edd75ffd | 2248 | cfqq->slice_resid = 0; |
48e025e6 | 2249 | } else { |
9a7f38c4 | 2250 | rb_key = -NSEC_PER_SEC; |
34b98d03 | 2251 | __cfqq = cfq_rb_first(st); |
9a7f38c4 | 2252 | rb_key += __cfqq ? __cfqq->rb_key : now; |
48e025e6 | 2253 | } |
1da177e4 | 2254 | |
d9e7620e | 2255 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 2256 | new_cfqq = 0; |
99f9628a | 2257 | /* |
d9e7620e | 2258 | * same position, nothing more to do |
99f9628a | 2259 | */ |
34b98d03 | 2260 | if (rb_key == cfqq->rb_key && cfqq->service_tree == st) |
d9e7620e | 2261 | return; |
1da177e4 | 2262 | |
aa6f6a3d CZ |
2263 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
2264 | cfqq->service_tree = NULL; | |
1da177e4 | 2265 | } |
d9e7620e | 2266 | |
498d3aa2 | 2267 | left = 1; |
0871714e | 2268 | parent = NULL; |
34b98d03 VG |
2269 | cfqq->service_tree = st; |
2270 | p = &st->rb.rb_node; | |
d9e7620e JA |
2271 | while (*p) { |
2272 | parent = *p; | |
2273 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2274 | ||
0c534e0a | 2275 | /* |
c0324a02 | 2276 | * sort by key, that represents service time. |
0c534e0a | 2277 | */ |
9a7f38c4 | 2278 | if (rb_key < __cfqq->rb_key) |
1f23f121 | 2279 | p = &parent->rb_left; |
c0324a02 | 2280 | else { |
1f23f121 | 2281 | p = &parent->rb_right; |
cc09e299 | 2282 | left = 0; |
c0324a02 | 2283 | } |
d9e7620e JA |
2284 | } |
2285 | ||
cc09e299 | 2286 | if (left) |
34b98d03 | 2287 | st->left = &cfqq->rb_node; |
cc09e299 | 2288 | |
d9e7620e JA |
2289 | cfqq->rb_key = rb_key; |
2290 | rb_link_node(&cfqq->rb_node, parent, p); | |
34b98d03 VG |
2291 | rb_insert_color(&cfqq->rb_node, &st->rb); |
2292 | st->count++; | |
20359f27 | 2293 | if (add_front || !new_cfqq) |
dae739eb | 2294 | return; |
8184f93e | 2295 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
2296 | } |
2297 | ||
a36e71f9 | 2298 | static struct cfq_queue * |
f2d1f0ae JA |
2299 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
2300 | sector_t sector, struct rb_node **ret_parent, | |
2301 | struct rb_node ***rb_link) | |
a36e71f9 | 2302 | { |
a36e71f9 JA |
2303 | struct rb_node **p, *parent; |
2304 | struct cfq_queue *cfqq = NULL; | |
2305 | ||
2306 | parent = NULL; | |
2307 | p = &root->rb_node; | |
2308 | while (*p) { | |
2309 | struct rb_node **n; | |
2310 | ||
2311 | parent = *p; | |
2312 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
2313 | ||
2314 | /* | |
2315 | * Sort strictly based on sector. Smallest to the left, | |
2316 | * largest to the right. | |
2317 | */ | |
2e46e8b2 | 2318 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 2319 | n = &(*p)->rb_right; |
2e46e8b2 | 2320 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
2321 | n = &(*p)->rb_left; |
2322 | else | |
2323 | break; | |
2324 | p = n; | |
3ac6c9f8 | 2325 | cfqq = NULL; |
a36e71f9 JA |
2326 | } |
2327 | ||
2328 | *ret_parent = parent; | |
2329 | if (rb_link) | |
2330 | *rb_link = p; | |
3ac6c9f8 | 2331 | return cfqq; |
a36e71f9 JA |
2332 | } |
2333 | ||
2334 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2335 | { | |
a36e71f9 JA |
2336 | struct rb_node **p, *parent; |
2337 | struct cfq_queue *__cfqq; | |
2338 | ||
f2d1f0ae JA |
2339 | if (cfqq->p_root) { |
2340 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2341 | cfqq->p_root = NULL; | |
2342 | } | |
a36e71f9 JA |
2343 | |
2344 | if (cfq_class_idle(cfqq)) | |
2345 | return; | |
2346 | if (!cfqq->next_rq) | |
2347 | return; | |
2348 | ||
f2d1f0ae | 2349 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
2350 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
2351 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
2352 | if (!__cfqq) { |
2353 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
2354 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
2355 | } else | |
2356 | cfqq->p_root = NULL; | |
a36e71f9 JA |
2357 | } |
2358 | ||
498d3aa2 JA |
2359 | /* |
2360 | * Update cfqq's position in the service tree. | |
2361 | */ | |
edd75ffd | 2362 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 2363 | { |
6d048f53 JA |
2364 | /* |
2365 | * Resorting requires the cfqq to be on the RR list already. | |
2366 | */ | |
a36e71f9 | 2367 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 2368 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
2369 | cfq_prio_tree_add(cfqd, cfqq); |
2370 | } | |
6d048f53 JA |
2371 | } |
2372 | ||
1da177e4 LT |
2373 | /* |
2374 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 2375 | * the pending list according to last request service |
1da177e4 | 2376 | */ |
febffd61 | 2377 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2378 | { |
7b679138 | 2379 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
2380 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
2381 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 2382 | cfqd->busy_queues++; |
ef8a41df SL |
2383 | if (cfq_cfqq_sync(cfqq)) |
2384 | cfqd->busy_sync_queues++; | |
1da177e4 | 2385 | |
edd75ffd | 2386 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
2387 | } |
2388 | ||
498d3aa2 JA |
2389 | /* |
2390 | * Called when the cfqq no longer has requests pending, remove it from | |
2391 | * the service tree. | |
2392 | */ | |
febffd61 | 2393 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2394 | { |
7b679138 | 2395 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
2396 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
2397 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 2398 | |
aa6f6a3d CZ |
2399 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
2400 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
2401 | cfqq->service_tree = NULL; | |
2402 | } | |
f2d1f0ae JA |
2403 | if (cfqq->p_root) { |
2404 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2405 | cfqq->p_root = NULL; | |
2406 | } | |
d9e7620e | 2407 | |
8184f93e | 2408 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
2409 | BUG_ON(!cfqd->busy_queues); |
2410 | cfqd->busy_queues--; | |
ef8a41df SL |
2411 | if (cfq_cfqq_sync(cfqq)) |
2412 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
2413 | } |
2414 | ||
2415 | /* | |
2416 | * rb tree support functions | |
2417 | */ | |
febffd61 | 2418 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 2419 | { |
5e705374 | 2420 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 2421 | const int sync = rq_is_sync(rq); |
1da177e4 | 2422 | |
b4878f24 JA |
2423 | BUG_ON(!cfqq->queued[sync]); |
2424 | cfqq->queued[sync]--; | |
1da177e4 | 2425 | |
5e705374 | 2426 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 2427 | |
f04a6424 VG |
2428 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
2429 | /* | |
2430 | * Queue will be deleted from service tree when we actually | |
2431 | * expire it later. Right now just remove it from prio tree | |
2432 | * as it is empty. | |
2433 | */ | |
2434 | if (cfqq->p_root) { | |
2435 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2436 | cfqq->p_root = NULL; | |
2437 | } | |
2438 | } | |
1da177e4 LT |
2439 | } |
2440 | ||
5e705374 | 2441 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 2442 | { |
5e705374 | 2443 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2444 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 2445 | struct request *prev; |
1da177e4 | 2446 | |
5380a101 | 2447 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 2448 | |
796d5116 | 2449 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
2450 | |
2451 | if (!cfq_cfqq_on_rr(cfqq)) | |
2452 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
2453 | |
2454 | /* | |
2455 | * check if this request is a better next-serve candidate | |
2456 | */ | |
a36e71f9 | 2457 | prev = cfqq->next_rq; |
cf7c25cf | 2458 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
2459 | |
2460 | /* | |
2461 | * adjust priority tree position, if ->next_rq changes | |
2462 | */ | |
2463 | if (prev != cfqq->next_rq) | |
2464 | cfq_prio_tree_add(cfqd, cfqq); | |
2465 | ||
5044eed4 | 2466 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
2467 | } |
2468 | ||
febffd61 | 2469 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 2470 | { |
5380a101 JA |
2471 | elv_rb_del(&cfqq->sort_list, rq); |
2472 | cfqq->queued[rq_is_sync(rq)]--; | |
ef295ecf | 2473 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
5e705374 | 2474 | cfq_add_rq_rb(rq); |
155fead9 | 2475 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
ef295ecf | 2476 | rq->cmd_flags); |
1da177e4 LT |
2477 | } |
2478 | ||
206dc69b JA |
2479 | static struct request * |
2480 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 2481 | { |
206dc69b | 2482 | struct task_struct *tsk = current; |
c5869807 | 2483 | struct cfq_io_cq *cic; |
206dc69b | 2484 | struct cfq_queue *cfqq; |
1da177e4 | 2485 | |
4ac845a2 | 2486 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2487 | if (!cic) |
2488 | return NULL; | |
2489 | ||
aa39ebd4 | 2490 | cfqq = cic_to_cfqq(cic, op_is_sync(bio->bi_opf)); |
f73a1c7d KO |
2491 | if (cfqq) |
2492 | return elv_rb_find(&cfqq->sort_list, bio_end_sector(bio)); | |
1da177e4 | 2493 | |
1da177e4 LT |
2494 | return NULL; |
2495 | } | |
2496 | ||
165125e1 | 2497 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2498 | { |
22e2c507 | 2499 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 2500 | |
53c583d2 | 2501 | cfqd->rq_in_driver++; |
7b679138 | 2502 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 2503 | cfqd->rq_in_driver); |
25776e35 | 2504 | |
5b93629b | 2505 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
2506 | } |
2507 | ||
165125e1 | 2508 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2509 | { |
b4878f24 JA |
2510 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2511 | ||
53c583d2 CZ |
2512 | WARN_ON(!cfqd->rq_in_driver); |
2513 | cfqd->rq_in_driver--; | |
7b679138 | 2514 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 2515 | cfqd->rq_in_driver); |
1da177e4 LT |
2516 | } |
2517 | ||
b4878f24 | 2518 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 2519 | { |
5e705374 | 2520 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 2521 | |
5e705374 JA |
2522 | if (cfqq->next_rq == rq) |
2523 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 2524 | |
b4878f24 | 2525 | list_del_init(&rq->queuelist); |
5e705374 | 2526 | cfq_del_rq_rb(rq); |
374f84ac | 2527 | |
45333d5a | 2528 | cfqq->cfqd->rq_queued--; |
ef295ecf | 2529 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
65299a3b CH |
2530 | if (rq->cmd_flags & REQ_PRIO) { |
2531 | WARN_ON(!cfqq->prio_pending); | |
2532 | cfqq->prio_pending--; | |
b53d1ed7 | 2533 | } |
1da177e4 LT |
2534 | } |
2535 | ||
34fe7c05 | 2536 | static enum elv_merge cfq_merge(struct request_queue *q, struct request **req, |
165125e1 | 2537 | struct bio *bio) |
1da177e4 LT |
2538 | { |
2539 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2540 | struct request *__rq; | |
1da177e4 | 2541 | |
206dc69b | 2542 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
72ef799b | 2543 | if (__rq && elv_bio_merge_ok(__rq, bio)) { |
9817064b JA |
2544 | *req = __rq; |
2545 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
2546 | } |
2547 | ||
2548 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
2549 | } |
2550 | ||
165125e1 | 2551 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
34fe7c05 | 2552 | enum elv_merge type) |
1da177e4 | 2553 | { |
21183b07 | 2554 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 2555 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 2556 | |
5e705374 | 2557 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 2558 | } |
1da177e4 LT |
2559 | } |
2560 | ||
812d4026 DS |
2561 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
2562 | struct bio *bio) | |
2563 | { | |
ef295ecf | 2564 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_opf); |
812d4026 DS |
2565 | } |
2566 | ||
1da177e4 | 2567 | static void |
165125e1 | 2568 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
2569 | struct request *next) |
2570 | { | |
cf7c25cf | 2571 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
2572 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2573 | ||
22e2c507 JA |
2574 | /* |
2575 | * reposition in fifo if next is older than rq | |
2576 | */ | |
2577 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
9a7f38c4 | 2578 | next->fifo_time < rq->fifo_time && |
3d106fba | 2579 | cfqq == RQ_CFQQ(next)) { |
22e2c507 | 2580 | list_move(&rq->queuelist, &next->queuelist); |
8b4922d3 | 2581 | rq->fifo_time = next->fifo_time; |
30996f40 | 2582 | } |
22e2c507 | 2583 | |
cf7c25cf CZ |
2584 | if (cfqq->next_rq == next) |
2585 | cfqq->next_rq = rq; | |
b4878f24 | 2586 | cfq_remove_request(next); |
ef295ecf | 2587 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
4a0b75c7 SL |
2588 | |
2589 | cfqq = RQ_CFQQ(next); | |
2590 | /* | |
2591 | * all requests of this queue are merged to other queues, delete it | |
2592 | * from the service tree. If it's the active_queue, | |
2593 | * cfq_dispatch_requests() will choose to expire it or do idle | |
2594 | */ | |
2595 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
2596 | cfqq != cfqd->active_queue) | |
2597 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
2598 | } |
2599 | ||
72ef799b TE |
2600 | static int cfq_allow_bio_merge(struct request_queue *q, struct request *rq, |
2601 | struct bio *bio) | |
da775265 JA |
2602 | { |
2603 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
aa39ebd4 | 2604 | bool is_sync = op_is_sync(bio->bi_opf); |
c5869807 | 2605 | struct cfq_io_cq *cic; |
da775265 | 2606 | struct cfq_queue *cfqq; |
da775265 JA |
2607 | |
2608 | /* | |
ec8acb69 | 2609 | * Disallow merge of a sync bio into an async request. |
da775265 | 2610 | */ |
aa39ebd4 | 2611 | if (is_sync && !rq_is_sync(rq)) |
a6151c3a | 2612 | return false; |
da775265 JA |
2613 | |
2614 | /* | |
f1a4f4d3 | 2615 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 2616 | * merge only if rq is queued there. |
f1a4f4d3 | 2617 | */ |
07c2bd37 TH |
2618 | cic = cfq_cic_lookup(cfqd, current->io_context); |
2619 | if (!cic) | |
2620 | return false; | |
719d3402 | 2621 | |
aa39ebd4 | 2622 | cfqq = cic_to_cfqq(cic, is_sync); |
a6151c3a | 2623 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
2624 | } |
2625 | ||
72ef799b TE |
2626 | static int cfq_allow_rq_merge(struct request_queue *q, struct request *rq, |
2627 | struct request *next) | |
2628 | { | |
2629 | return RQ_CFQQ(rq) == RQ_CFQQ(next); | |
2630 | } | |
2631 | ||
812df48d DS |
2632 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2633 | { | |
91148325 | 2634 | hrtimer_try_to_cancel(&cfqd->idle_slice_timer); |
155fead9 | 2635 | cfqg_stats_update_idle_time(cfqq->cfqg); |
812df48d DS |
2636 | } |
2637 | ||
febffd61 JA |
2638 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
2639 | struct cfq_queue *cfqq) | |
22e2c507 JA |
2640 | { |
2641 | if (cfqq) { | |
3bf10fea | 2642 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d", |
4d2ceea4 | 2643 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
155fead9 | 2644 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
62a37f6b | 2645 | cfqq->slice_start = 0; |
9a7f38c4 | 2646 | cfqq->dispatch_start = ktime_get_ns(); |
62a37f6b JT |
2647 | cfqq->allocated_slice = 0; |
2648 | cfqq->slice_end = 0; | |
2649 | cfqq->slice_dispatch = 0; | |
2650 | cfqq->nr_sectors = 0; | |
2651 | ||
2652 | cfq_clear_cfqq_wait_request(cfqq); | |
2653 | cfq_clear_cfqq_must_dispatch(cfqq); | |
2654 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
2655 | cfq_clear_cfqq_fifo_expire(cfqq); | |
2656 | cfq_mark_cfqq_slice_new(cfqq); | |
2657 | ||
2658 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
2659 | } |
2660 | ||
2661 | cfqd->active_queue = cfqq; | |
2662 | } | |
2663 | ||
7b14e3b5 JA |
2664 | /* |
2665 | * current cfqq expired its slice (or was too idle), select new one | |
2666 | */ | |
2667 | static void | |
2668 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 2669 | bool timed_out) |
7b14e3b5 | 2670 | { |
7b679138 JA |
2671 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
2672 | ||
7b14e3b5 | 2673 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 2674 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 2675 | |
7b14e3b5 | 2676 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 2677 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 2678 | |
ae54abed SL |
2679 | /* |
2680 | * If this cfqq is shared between multiple processes, check to | |
2681 | * make sure that those processes are still issuing I/Os within | |
2682 | * the mean seek distance. If not, it may be time to break the | |
2683 | * queues apart again. | |
2684 | */ | |
2685 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
2686 | cfq_mark_cfqq_split_coop(cfqq); | |
2687 | ||
7b14e3b5 | 2688 | /* |
6084cdda | 2689 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 2690 | */ |
c553f8e3 SL |
2691 | if (timed_out) { |
2692 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 2693 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 2694 | else |
9a7f38c4 | 2695 | cfqq->slice_resid = cfqq->slice_end - ktime_get_ns(); |
93fdf147 | 2696 | cfq_log_cfqq(cfqd, cfqq, "resid=%lld", cfqq->slice_resid); |
7b679138 | 2697 | } |
7b14e3b5 | 2698 | |
e5ff082e | 2699 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 2700 | |
f04a6424 VG |
2701 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
2702 | cfq_del_cfqq_rr(cfqd, cfqq); | |
2703 | ||
edd75ffd | 2704 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
2705 | |
2706 | if (cfqq == cfqd->active_queue) | |
2707 | cfqd->active_queue = NULL; | |
2708 | ||
2709 | if (cfqd->active_cic) { | |
11a3122f | 2710 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
2711 | cfqd->active_cic = NULL; |
2712 | } | |
7b14e3b5 JA |
2713 | } |
2714 | ||
e5ff082e | 2715 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
2716 | { |
2717 | struct cfq_queue *cfqq = cfqd->active_queue; | |
2718 | ||
2719 | if (cfqq) | |
e5ff082e | 2720 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
2721 | } |
2722 | ||
498d3aa2 JA |
2723 | /* |
2724 | * Get next queue for service. Unless we have a queue preemption, | |
2725 | * we'll simply select the first cfqq in the service tree. | |
2726 | */ | |
6d048f53 | 2727 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 2728 | { |
34b98d03 VG |
2729 | struct cfq_rb_root *st = st_for(cfqd->serving_group, |
2730 | cfqd->serving_wl_class, cfqd->serving_wl_type); | |
d9e7620e | 2731 | |
f04a6424 VG |
2732 | if (!cfqd->rq_queued) |
2733 | return NULL; | |
2734 | ||
1fa8f6d6 | 2735 | /* There is nothing to dispatch */ |
34b98d03 | 2736 | if (!st) |
1fa8f6d6 | 2737 | return NULL; |
34b98d03 | 2738 | if (RB_EMPTY_ROOT(&st->rb)) |
c0324a02 | 2739 | return NULL; |
34b98d03 | 2740 | return cfq_rb_first(st); |
6d048f53 JA |
2741 | } |
2742 | ||
f04a6424 VG |
2743 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
2744 | { | |
25fb5169 | 2745 | struct cfq_group *cfqg; |
f04a6424 VG |
2746 | struct cfq_queue *cfqq; |
2747 | int i, j; | |
2748 | struct cfq_rb_root *st; | |
2749 | ||
2750 | if (!cfqd->rq_queued) | |
2751 | return NULL; | |
2752 | ||
25fb5169 VG |
2753 | cfqg = cfq_get_next_cfqg(cfqd); |
2754 | if (!cfqg) | |
2755 | return NULL; | |
2756 | ||
1cf41753 ME |
2757 | for_each_cfqg_st(cfqg, i, j, st) { |
2758 | cfqq = cfq_rb_first(st); | |
2759 | if (cfqq) | |
f04a6424 | 2760 | return cfqq; |
1cf41753 | 2761 | } |
f04a6424 VG |
2762 | return NULL; |
2763 | } | |
2764 | ||
498d3aa2 JA |
2765 | /* |
2766 | * Get and set a new active queue for service. | |
2767 | */ | |
a36e71f9 JA |
2768 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
2769 | struct cfq_queue *cfqq) | |
6d048f53 | 2770 | { |
e00ef799 | 2771 | if (!cfqq) |
a36e71f9 | 2772 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 2773 | |
22e2c507 | 2774 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 2775 | return cfqq; |
22e2c507 JA |
2776 | } |
2777 | ||
d9e7620e JA |
2778 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
2779 | struct request *rq) | |
2780 | { | |
83096ebf TH |
2781 | if (blk_rq_pos(rq) >= cfqd->last_position) |
2782 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 2783 | else |
83096ebf | 2784 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
2785 | } |
2786 | ||
b2c18e1e | 2787 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 2788 | struct request *rq) |
6d048f53 | 2789 | { |
e9ce335d | 2790 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
2791 | } |
2792 | ||
a36e71f9 JA |
2793 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
2794 | struct cfq_queue *cur_cfqq) | |
2795 | { | |
f2d1f0ae | 2796 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
2797 | struct rb_node *parent, *node; |
2798 | struct cfq_queue *__cfqq; | |
2799 | sector_t sector = cfqd->last_position; | |
2800 | ||
2801 | if (RB_EMPTY_ROOT(root)) | |
2802 | return NULL; | |
2803 | ||
2804 | /* | |
2805 | * First, if we find a request starting at the end of the last | |
2806 | * request, choose it. | |
2807 | */ | |
f2d1f0ae | 2808 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
2809 | if (__cfqq) |
2810 | return __cfqq; | |
2811 | ||
2812 | /* | |
2813 | * If the exact sector wasn't found, the parent of the NULL leaf | |
2814 | * will contain the closest sector. | |
2815 | */ | |
2816 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 2817 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2818 | return __cfqq; |
2819 | ||
2e46e8b2 | 2820 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
2821 | node = rb_next(&__cfqq->p_node); |
2822 | else | |
2823 | node = rb_prev(&__cfqq->p_node); | |
2824 | if (!node) | |
2825 | return NULL; | |
2826 | ||
2827 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 2828 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2829 | return __cfqq; |
2830 | ||
2831 | return NULL; | |
2832 | } | |
2833 | ||
2834 | /* | |
2835 | * cfqd - obvious | |
2836 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
2837 | * closely cooperating with itself. | |
2838 | * | |
2839 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
2840 | * one request, and that cfqd->last_position reflects a position on the disk | |
2841 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
2842 | * assumption. | |
2843 | */ | |
2844 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 2845 | struct cfq_queue *cur_cfqq) |
6d048f53 | 2846 | { |
a36e71f9 JA |
2847 | struct cfq_queue *cfqq; |
2848 | ||
39c01b21 DS |
2849 | if (cfq_class_idle(cur_cfqq)) |
2850 | return NULL; | |
e6c5bc73 JM |
2851 | if (!cfq_cfqq_sync(cur_cfqq)) |
2852 | return NULL; | |
2853 | if (CFQQ_SEEKY(cur_cfqq)) | |
2854 | return NULL; | |
2855 | ||
b9d8f4c7 GJ |
2856 | /* |
2857 | * Don't search priority tree if it's the only queue in the group. | |
2858 | */ | |
2859 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
2860 | return NULL; | |
2861 | ||
6d048f53 | 2862 | /* |
d9e7620e JA |
2863 | * We should notice if some of the queues are cooperating, eg |
2864 | * working closely on the same area of the disk. In that case, | |
2865 | * we can group them together and don't waste time idling. | |
6d048f53 | 2866 | */ |
a36e71f9 JA |
2867 | cfqq = cfqq_close(cfqd, cur_cfqq); |
2868 | if (!cfqq) | |
2869 | return NULL; | |
2870 | ||
8682e1f1 VG |
2871 | /* If new queue belongs to different cfq_group, don't choose it */ |
2872 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
2873 | return NULL; | |
2874 | ||
df5fe3e8 JM |
2875 | /* |
2876 | * It only makes sense to merge sync queues. | |
2877 | */ | |
2878 | if (!cfq_cfqq_sync(cfqq)) | |
2879 | return NULL; | |
e6c5bc73 JM |
2880 | if (CFQQ_SEEKY(cfqq)) |
2881 | return NULL; | |
df5fe3e8 | 2882 | |
c0324a02 CZ |
2883 | /* |
2884 | * Do not merge queues of different priority classes | |
2885 | */ | |
2886 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
2887 | return NULL; | |
2888 | ||
a36e71f9 | 2889 | return cfqq; |
6d048f53 JA |
2890 | } |
2891 | ||
a6d44e98 CZ |
2892 | /* |
2893 | * Determine whether we should enforce idle window for this queue. | |
2894 | */ | |
2895 | ||
2896 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2897 | { | |
3bf10fea | 2898 | enum wl_class_t wl_class = cfqq_class(cfqq); |
34b98d03 | 2899 | struct cfq_rb_root *st = cfqq->service_tree; |
a6d44e98 | 2900 | |
34b98d03 VG |
2901 | BUG_ON(!st); |
2902 | BUG_ON(!st->count); | |
f04a6424 | 2903 | |
b6508c16 VG |
2904 | if (!cfqd->cfq_slice_idle) |
2905 | return false; | |
2906 | ||
a6d44e98 | 2907 | /* We never do for idle class queues. */ |
3bf10fea | 2908 | if (wl_class == IDLE_WORKLOAD) |
a6d44e98 CZ |
2909 | return false; |
2910 | ||
2911 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
2912 | if (cfq_cfqq_idle_window(cfqq) && |
2913 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
2914 | return true; |
2915 | ||
2916 | /* | |
2917 | * Otherwise, we do only if they are the last ones | |
2918 | * in their service tree. | |
2919 | */ | |
34b98d03 VG |
2920 | if (st->count == 1 && cfq_cfqq_sync(cfqq) && |
2921 | !cfq_io_thinktime_big(cfqd, &st->ttime, false)) | |
c1e44756 | 2922 | return true; |
34b98d03 | 2923 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count); |
c1e44756 | 2924 | return false; |
a6d44e98 CZ |
2925 | } |
2926 | ||
6d048f53 | 2927 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2928 | { |
1792669c | 2929 | struct cfq_queue *cfqq = cfqd->active_queue; |
e795421e | 2930 | struct cfq_rb_root *st = cfqq->service_tree; |
c5869807 | 2931 | struct cfq_io_cq *cic; |
9a7f38c4 JM |
2932 | u64 sl, group_idle = 0; |
2933 | u64 now = ktime_get_ns(); | |
7b14e3b5 | 2934 | |
a68bbddb | 2935 | /* |
f7d7b7a7 JA |
2936 | * SSD device without seek penalty, disable idling. But only do so |
2937 | * for devices that support queuing, otherwise we still have a problem | |
2938 | * with sync vs async workloads. | |
a68bbddb | 2939 | */ |
f7d7b7a7 | 2940 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
2941 | return; |
2942 | ||
dd67d051 | 2943 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2944 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2945 | |
2946 | /* | |
2947 | * idle is disabled, either manually or by past process history | |
2948 | */ | |
80bdf0c7 VG |
2949 | if (!cfq_should_idle(cfqd, cfqq)) { |
2950 | /* no queue idling. Check for group idling */ | |
2951 | if (cfqd->cfq_group_idle) | |
2952 | group_idle = cfqd->cfq_group_idle; | |
2953 | else | |
2954 | return; | |
2955 | } | |
6d048f53 | 2956 | |
7b679138 | 2957 | /* |
8e550632 | 2958 | * still active requests from this queue, don't idle |
7b679138 | 2959 | */ |
8e550632 | 2960 | if (cfqq->dispatched) |
7b679138 JA |
2961 | return; |
2962 | ||
22e2c507 JA |
2963 | /* |
2964 | * task has exited, don't wait | |
2965 | */ | |
206dc69b | 2966 | cic = cfqd->active_cic; |
f6e8d01b | 2967 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
2968 | return; |
2969 | ||
355b659c CZ |
2970 | /* |
2971 | * If our average think time is larger than the remaining time | |
2972 | * slice, then don't idle. This avoids overrunning the allotted | |
2973 | * time slice. | |
2974 | */ | |
383cd721 | 2975 | if (sample_valid(cic->ttime.ttime_samples) && |
9a7f38c4 JM |
2976 | (cfqq->slice_end - now < cic->ttime.ttime_mean)) { |
2977 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%llu", | |
383cd721 | 2978 | cic->ttime.ttime_mean); |
355b659c | 2979 | return; |
b1ffe737 | 2980 | } |
355b659c | 2981 | |
e795421e JK |
2982 | /* |
2983 | * There are other queues in the group or this is the only group and | |
2984 | * it has too big thinktime, don't do group idle. | |
2985 | */ | |
2986 | if (group_idle && | |
2987 | (cfqq->cfqg->nr_cfqq > 1 || | |
2988 | cfq_io_thinktime_big(cfqd, &st->ttime, true))) | |
80bdf0c7 VG |
2989 | return; |
2990 | ||
3b18152c | 2991 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2992 | |
80bdf0c7 VG |
2993 | if (group_idle) |
2994 | sl = cfqd->cfq_group_idle; | |
2995 | else | |
2996 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2997 | |
91148325 JK |
2998 | hrtimer_start(&cfqd->idle_slice_timer, ns_to_ktime(sl), |
2999 | HRTIMER_MODE_REL); | |
155fead9 | 3000 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
9a7f38c4 | 3001 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %llu group_idle: %d", sl, |
80bdf0c7 | 3002 | group_idle ? 1 : 0); |
1da177e4 LT |
3003 | } |
3004 | ||
498d3aa2 JA |
3005 | /* |
3006 | * Move request from internal lists to the request queue dispatch list. | |
3007 | */ | |
165125e1 | 3008 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 3009 | { |
3ed9a296 | 3010 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3011 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3012 | |
7b679138 JA |
3013 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
3014 | ||
06d21886 | 3015 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 3016 | cfq_remove_request(rq); |
6d048f53 | 3017 | cfqq->dispatched++; |
80bdf0c7 | 3018 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 3019 | elv_dispatch_sort(q, rq); |
3ed9a296 | 3020 | |
53c583d2 | 3021 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 3022 | cfqq->nr_sectors += blk_rq_sectors(rq); |
1da177e4 LT |
3023 | } |
3024 | ||
3025 | /* | |
3026 | * return expired entry, or NULL to just start from scratch in rbtree | |
3027 | */ | |
febffd61 | 3028 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 3029 | { |
30996f40 | 3030 | struct request *rq = NULL; |
1da177e4 | 3031 | |
3b18152c | 3032 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 3033 | return NULL; |
cb887411 JA |
3034 | |
3035 | cfq_mark_cfqq_fifo_expire(cfqq); | |
3036 | ||
89850f7e JA |
3037 | if (list_empty(&cfqq->fifo)) |
3038 | return NULL; | |
1da177e4 | 3039 | |
89850f7e | 3040 | rq = rq_entry_fifo(cfqq->fifo.next); |
9a7f38c4 | 3041 | if (ktime_get_ns() < rq->fifo_time) |
7b679138 | 3042 | rq = NULL; |
1da177e4 | 3043 | |
6d048f53 | 3044 | return rq; |
1da177e4 LT |
3045 | } |
3046 | ||
22e2c507 JA |
3047 | static inline int |
3048 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3049 | { | |
3050 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 3051 | |
22e2c507 | 3052 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 3053 | |
b9f8ce05 | 3054 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
3055 | } |
3056 | ||
df5fe3e8 JM |
3057 | /* |
3058 | * Must be called with the queue_lock held. | |
3059 | */ | |
3060 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
3061 | { | |
3062 | int process_refs, io_refs; | |
3063 | ||
3064 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 3065 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
3066 | BUG_ON(process_refs < 0); |
3067 | return process_refs; | |
3068 | } | |
3069 | ||
3070 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
3071 | { | |
e6c5bc73 | 3072 | int process_refs, new_process_refs; |
df5fe3e8 JM |
3073 | struct cfq_queue *__cfqq; |
3074 | ||
c10b61f0 JM |
3075 | /* |
3076 | * If there are no process references on the new_cfqq, then it is | |
3077 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
3078 | * chain may have dropped their last reference (not just their | |
3079 | * last process reference). | |
3080 | */ | |
3081 | if (!cfqq_process_refs(new_cfqq)) | |
3082 | return; | |
3083 | ||
df5fe3e8 JM |
3084 | /* Avoid a circular list and skip interim queue merges */ |
3085 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
3086 | if (__cfqq == cfqq) | |
3087 | return; | |
3088 | new_cfqq = __cfqq; | |
3089 | } | |
3090 | ||
3091 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 3092 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
3093 | /* |
3094 | * If the process for the cfqq has gone away, there is no | |
3095 | * sense in merging the queues. | |
3096 | */ | |
c10b61f0 | 3097 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
3098 | return; |
3099 | ||
e6c5bc73 JM |
3100 | /* |
3101 | * Merge in the direction of the lesser amount of work. | |
3102 | */ | |
e6c5bc73 JM |
3103 | if (new_process_refs >= process_refs) { |
3104 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 3105 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
3106 | } else { |
3107 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 3108 | cfqq->ref += new_process_refs; |
e6c5bc73 | 3109 | } |
df5fe3e8 JM |
3110 | } |
3111 | ||
6d816ec7 | 3112 | static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd, |
3bf10fea | 3113 | struct cfq_group *cfqg, enum wl_class_t wl_class) |
718eee05 CZ |
3114 | { |
3115 | struct cfq_queue *queue; | |
3116 | int i; | |
3117 | bool key_valid = false; | |
9a7f38c4 | 3118 | u64 lowest_key = 0; |
718eee05 CZ |
3119 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; |
3120 | ||
65b32a57 VG |
3121 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
3122 | /* select the one with lowest rb_key */ | |
34b98d03 | 3123 | queue = cfq_rb_first(st_for(cfqg, wl_class, i)); |
718eee05 | 3124 | if (queue && |
9a7f38c4 | 3125 | (!key_valid || queue->rb_key < lowest_key)) { |
718eee05 CZ |
3126 | lowest_key = queue->rb_key; |
3127 | cur_best = i; | |
3128 | key_valid = true; | |
3129 | } | |
3130 | } | |
3131 | ||
3132 | return cur_best; | |
3133 | } | |
3134 | ||
6d816ec7 VG |
3135 | static void |
3136 | choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
718eee05 | 3137 | { |
9a7f38c4 | 3138 | u64 slice; |
718eee05 | 3139 | unsigned count; |
cdb16e8f | 3140 | struct cfq_rb_root *st; |
9a7f38c4 | 3141 | u64 group_slice; |
4d2ceea4 | 3142 | enum wl_class_t original_class = cfqd->serving_wl_class; |
9a7f38c4 | 3143 | u64 now = ktime_get_ns(); |
1fa8f6d6 | 3144 | |
718eee05 | 3145 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 3146 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 3147 | cfqd->serving_wl_class = RT_WORKLOAD; |
58ff82f3 | 3148 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 3149 | cfqd->serving_wl_class = BE_WORKLOAD; |
718eee05 | 3150 | else { |
4d2ceea4 | 3151 | cfqd->serving_wl_class = IDLE_WORKLOAD; |
9a7f38c4 | 3152 | cfqd->workload_expires = now + jiffies_to_nsecs(1); |
718eee05 CZ |
3153 | return; |
3154 | } | |
3155 | ||
4d2ceea4 | 3156 | if (original_class != cfqd->serving_wl_class) |
e4ea0c16 SL |
3157 | goto new_workload; |
3158 | ||
718eee05 CZ |
3159 | /* |
3160 | * For RT and BE, we have to choose also the type | |
3161 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
3162 | * expiration time | |
3163 | */ | |
34b98d03 | 3164 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 3165 | count = st->count; |
718eee05 CZ |
3166 | |
3167 | /* | |
65b32a57 | 3168 | * check workload expiration, and that we still have other queues ready |
718eee05 | 3169 | */ |
9a7f38c4 | 3170 | if (count && !(now > cfqd->workload_expires)) |
718eee05 CZ |
3171 | return; |
3172 | ||
e4ea0c16 | 3173 | new_workload: |
718eee05 | 3174 | /* otherwise select new workload type */ |
6d816ec7 | 3175 | cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg, |
4d2ceea4 | 3176 | cfqd->serving_wl_class); |
34b98d03 | 3177 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 3178 | count = st->count; |
718eee05 CZ |
3179 | |
3180 | /* | |
3181 | * the workload slice is computed as a fraction of target latency | |
3182 | * proportional to the number of queues in that workload, over | |
3183 | * all the queues in the same priority class | |
3184 | */ | |
58ff82f3 VG |
3185 | group_slice = cfq_group_slice(cfqd, cfqg); |
3186 | ||
9a7f38c4 | 3187 | slice = div_u64(group_slice * count, |
4d2ceea4 VG |
3188 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class], |
3189 | cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd, | |
9a7f38c4 | 3190 | cfqg))); |
718eee05 | 3191 | |
4d2ceea4 | 3192 | if (cfqd->serving_wl_type == ASYNC_WORKLOAD) { |
9a7f38c4 | 3193 | u64 tmp; |
f26bd1f0 VG |
3194 | |
3195 | /* | |
3196 | * Async queues are currently system wide. Just taking | |
3197 | * proportion of queues with-in same group will lead to higher | |
3198 | * async ratio system wide as generally root group is going | |
3199 | * to have higher weight. A more accurate thing would be to | |
3200 | * calculate system wide asnc/sync ratio. | |
3201 | */ | |
5bf14c07 TM |
3202 | tmp = cfqd->cfq_target_latency * |
3203 | cfqg_busy_async_queues(cfqd, cfqg); | |
9a7f38c4 JM |
3204 | tmp = div_u64(tmp, cfqd->busy_queues); |
3205 | slice = min_t(u64, slice, tmp); | |
f26bd1f0 | 3206 | |
718eee05 CZ |
3207 | /* async workload slice is scaled down according to |
3208 | * the sync/async slice ratio. */ | |
9a7f38c4 | 3209 | slice = div64_u64(slice*cfqd->cfq_slice[0], cfqd->cfq_slice[1]); |
f26bd1f0 | 3210 | } else |
718eee05 CZ |
3211 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
3212 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
3213 | ||
9a7f38c4 JM |
3214 | slice = max_t(u64, slice, CFQ_MIN_TT); |
3215 | cfq_log(cfqd, "workload slice:%llu", slice); | |
3216 | cfqd->workload_expires = now + slice; | |
718eee05 CZ |
3217 | } |
3218 | ||
1fa8f6d6 VG |
3219 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
3220 | { | |
3221 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 3222 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
3223 | |
3224 | if (RB_EMPTY_ROOT(&st->rb)) | |
3225 | return NULL; | |
25bc6b07 | 3226 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
3227 | update_min_vdisktime(st); |
3228 | return cfqg; | |
1fa8f6d6 VG |
3229 | } |
3230 | ||
cdb16e8f VG |
3231 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
3232 | { | |
1fa8f6d6 | 3233 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
9a7f38c4 | 3234 | u64 now = ktime_get_ns(); |
1fa8f6d6 VG |
3235 | |
3236 | cfqd->serving_group = cfqg; | |
dae739eb VG |
3237 | |
3238 | /* Restore the workload type data */ | |
4d2ceea4 | 3239 | if (cfqg->saved_wl_slice) { |
9a7f38c4 | 3240 | cfqd->workload_expires = now + cfqg->saved_wl_slice; |
4d2ceea4 VG |
3241 | cfqd->serving_wl_type = cfqg->saved_wl_type; |
3242 | cfqd->serving_wl_class = cfqg->saved_wl_class; | |
66ae2919 | 3243 | } else |
9a7f38c4 | 3244 | cfqd->workload_expires = now - 1; |
66ae2919 | 3245 | |
6d816ec7 | 3246 | choose_wl_class_and_type(cfqd, cfqg); |
cdb16e8f VG |
3247 | } |
3248 | ||
22e2c507 | 3249 | /* |
498d3aa2 JA |
3250 | * Select a queue for service. If we have a current active queue, |
3251 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 3252 | */ |
1b5ed5e1 | 3253 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 3254 | { |
a36e71f9 | 3255 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
9a7f38c4 | 3256 | u64 now = ktime_get_ns(); |
1da177e4 | 3257 | |
22e2c507 JA |
3258 | cfqq = cfqd->active_queue; |
3259 | if (!cfqq) | |
3260 | goto new_queue; | |
1da177e4 | 3261 | |
f04a6424 VG |
3262 | if (!cfqd->rq_queued) |
3263 | return NULL; | |
c244bb50 VG |
3264 | |
3265 | /* | |
3266 | * We were waiting for group to get backlogged. Expire the queue | |
3267 | */ | |
3268 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3269 | goto expire; | |
3270 | ||
22e2c507 | 3271 | /* |
6d048f53 | 3272 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 3273 | */ |
7667aa06 VG |
3274 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
3275 | /* | |
3276 | * If slice had not expired at the completion of last request | |
3277 | * we might not have turned on wait_busy flag. Don't expire | |
3278 | * the queue yet. Allow the group to get backlogged. | |
3279 | * | |
3280 | * The very fact that we have used the slice, that means we | |
3281 | * have been idling all along on this queue and it should be | |
3282 | * ok to wait for this request to complete. | |
3283 | */ | |
82bbbf28 VG |
3284 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
3285 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
3286 | cfqq = NULL; | |
7667aa06 | 3287 | goto keep_queue; |
82bbbf28 | 3288 | } else |
80bdf0c7 | 3289 | goto check_group_idle; |
7667aa06 | 3290 | } |
1da177e4 | 3291 | |
22e2c507 | 3292 | /* |
6d048f53 JA |
3293 | * The active queue has requests and isn't expired, allow it to |
3294 | * dispatch. | |
22e2c507 | 3295 | */ |
dd67d051 | 3296 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3297 | goto keep_queue; |
6d048f53 | 3298 | |
a36e71f9 JA |
3299 | /* |
3300 | * If another queue has a request waiting within our mean seek | |
3301 | * distance, let it run. The expire code will check for close | |
3302 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 3303 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 3304 | */ |
b3b6d040 | 3305 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
3306 | if (new_cfqq) { |
3307 | if (!cfqq->new_cfqq) | |
3308 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 3309 | goto expire; |
df5fe3e8 | 3310 | } |
a36e71f9 | 3311 | |
6d048f53 JA |
3312 | /* |
3313 | * No requests pending. If the active queue still has requests in | |
3314 | * flight or is idling for a new request, allow either of these | |
3315 | * conditions to happen (or time out) before selecting a new queue. | |
3316 | */ | |
91148325 | 3317 | if (hrtimer_active(&cfqd->idle_slice_timer)) { |
80bdf0c7 VG |
3318 | cfqq = NULL; |
3319 | goto keep_queue; | |
3320 | } | |
3321 | ||
8e1ac665 SL |
3322 | /* |
3323 | * This is a deep seek queue, but the device is much faster than | |
3324 | * the queue can deliver, don't idle | |
3325 | **/ | |
3326 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
3327 | (cfq_cfqq_slice_new(cfqq) || | |
9a7f38c4 | 3328 | (cfqq->slice_end - now > now - cfqq->slice_start))) { |
8e1ac665 SL |
3329 | cfq_clear_cfqq_deep(cfqq); |
3330 | cfq_clear_cfqq_idle_window(cfqq); | |
3331 | } | |
3332 | ||
80bdf0c7 VG |
3333 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
3334 | cfqq = NULL; | |
3335 | goto keep_queue; | |
3336 | } | |
3337 | ||
3338 | /* | |
3339 | * If group idle is enabled and there are requests dispatched from | |
3340 | * this group, wait for requests to complete. | |
3341 | */ | |
3342 | check_group_idle: | |
7700fc4f SL |
3343 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
3344 | cfqq->cfqg->dispatched && | |
3345 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
3346 | cfqq = NULL; |
3347 | goto keep_queue; | |
22e2c507 JA |
3348 | } |
3349 | ||
3b18152c | 3350 | expire: |
e5ff082e | 3351 | cfq_slice_expired(cfqd, 0); |
3b18152c | 3352 | new_queue: |
718eee05 CZ |
3353 | /* |
3354 | * Current queue expired. Check if we have to switch to a new | |
3355 | * service tree | |
3356 | */ | |
3357 | if (!new_cfqq) | |
cdb16e8f | 3358 | cfq_choose_cfqg(cfqd); |
718eee05 | 3359 | |
a36e71f9 | 3360 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 3361 | keep_queue: |
3b18152c | 3362 | return cfqq; |
22e2c507 JA |
3363 | } |
3364 | ||
febffd61 | 3365 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
3366 | { |
3367 | int dispatched = 0; | |
3368 | ||
3369 | while (cfqq->next_rq) { | |
3370 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
3371 | dispatched++; | |
3372 | } | |
3373 | ||
3374 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
3375 | |
3376 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 3377 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
3378 | return dispatched; |
3379 | } | |
3380 | ||
498d3aa2 JA |
3381 | /* |
3382 | * Drain our current requests. Used for barriers and when switching | |
3383 | * io schedulers on-the-fly. | |
3384 | */ | |
d9e7620e | 3385 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 3386 | { |
0871714e | 3387 | struct cfq_queue *cfqq; |
d9e7620e | 3388 | int dispatched = 0; |
cdb16e8f | 3389 | |
3440c49f | 3390 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 3391 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
3392 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
3393 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 3394 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 3395 | } |
1b5ed5e1 | 3396 | |
1b5ed5e1 TH |
3397 | BUG_ON(cfqd->busy_queues); |
3398 | ||
6923715a | 3399 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
3400 | return dispatched; |
3401 | } | |
3402 | ||
abc3c744 SL |
3403 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
3404 | struct cfq_queue *cfqq) | |
3405 | { | |
9a7f38c4 JM |
3406 | u64 now = ktime_get_ns(); |
3407 | ||
abc3c744 SL |
3408 | /* the queue hasn't finished any request, can't estimate */ |
3409 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 3410 | return true; |
9a7f38c4 | 3411 | if (now + cfqd->cfq_slice_idle * cfqq->dispatched > cfqq->slice_end) |
c1e44756 | 3412 | return true; |
abc3c744 | 3413 | |
c1e44756 | 3414 | return false; |
abc3c744 SL |
3415 | } |
3416 | ||
0b182d61 | 3417 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 3418 | { |
2f5cb738 | 3419 | unsigned int max_dispatch; |
22e2c507 | 3420 | |
3932a86b GC |
3421 | if (cfq_cfqq_must_dispatch(cfqq)) |
3422 | return true; | |
3423 | ||
5ad531db JA |
3424 | /* |
3425 | * Drain async requests before we start sync IO | |
3426 | */ | |
53c583d2 | 3427 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 3428 | return false; |
5ad531db | 3429 | |
2f5cb738 JA |
3430 | /* |
3431 | * If this is an async queue and we have sync IO in flight, let it wait | |
3432 | */ | |
53c583d2 | 3433 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 3434 | return false; |
2f5cb738 | 3435 | |
abc3c744 | 3436 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
3437 | if (cfq_class_idle(cfqq)) |
3438 | max_dispatch = 1; | |
b4878f24 | 3439 | |
2f5cb738 JA |
3440 | /* |
3441 | * Does this cfqq already have too much IO in flight? | |
3442 | */ | |
3443 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 3444 | bool promote_sync = false; |
2f5cb738 JA |
3445 | /* |
3446 | * idle queue must always only have a single IO in flight | |
3447 | */ | |
3ed9a296 | 3448 | if (cfq_class_idle(cfqq)) |
0b182d61 | 3449 | return false; |
3ed9a296 | 3450 | |
ef8a41df | 3451 | /* |
c4ade94f LS |
3452 | * If there is only one sync queue |
3453 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
3454 | * queue no dispatch limit. The reason is a sync queue can |
3455 | * preempt async queue, limiting the sync queue doesn't make | |
3456 | * sense. This is useful for aiostress test. | |
3457 | */ | |
c4ade94f LS |
3458 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
3459 | promote_sync = true; | |
ef8a41df | 3460 | |
2f5cb738 JA |
3461 | /* |
3462 | * We have other queues, don't allow more IO from this one | |
3463 | */ | |
ef8a41df SL |
3464 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
3465 | !promote_sync) | |
0b182d61 | 3466 | return false; |
9ede209e | 3467 | |
365722bb | 3468 | /* |
474b18cc | 3469 | * Sole queue user, no limit |
365722bb | 3470 | */ |
ef8a41df | 3471 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
3472 | max_dispatch = -1; |
3473 | else | |
3474 | /* | |
3475 | * Normally we start throttling cfqq when cfq_quantum/2 | |
3476 | * requests have been dispatched. But we can drive | |
3477 | * deeper queue depths at the beginning of slice | |
3478 | * subjected to upper limit of cfq_quantum. | |
3479 | * */ | |
3480 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
3481 | } |
3482 | ||
3483 | /* | |
3484 | * Async queues must wait a bit before being allowed dispatch. | |
3485 | * We also ramp up the dispatch depth gradually for async IO, | |
3486 | * based on the last sync IO we serviced | |
3487 | */ | |
963b72fc | 3488 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
9a7f38c4 | 3489 | u64 last_sync = ktime_get_ns() - cfqd->last_delayed_sync; |
8e296755 | 3490 | unsigned int depth; |
365722bb | 3491 | |
9a7f38c4 | 3492 | depth = div64_u64(last_sync, cfqd->cfq_slice[1]); |
e00c54c3 JA |
3493 | if (!depth && !cfqq->dispatched) |
3494 | depth = 1; | |
8e296755 JA |
3495 | if (depth < max_dispatch) |
3496 | max_dispatch = depth; | |
2f5cb738 | 3497 | } |
3ed9a296 | 3498 | |
0b182d61 JA |
3499 | /* |
3500 | * If we're below the current max, allow a dispatch | |
3501 | */ | |
3502 | return cfqq->dispatched < max_dispatch; | |
3503 | } | |
3504 | ||
3505 | /* | |
3506 | * Dispatch a request from cfqq, moving them to the request queue | |
3507 | * dispatch list. | |
3508 | */ | |
3509 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3510 | { | |
3511 | struct request *rq; | |
3512 | ||
3513 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
3514 | ||
3932a86b GC |
3515 | rq = cfq_check_fifo(cfqq); |
3516 | if (rq) | |
3517 | cfq_mark_cfqq_must_dispatch(cfqq); | |
3518 | ||
0b182d61 JA |
3519 | if (!cfq_may_dispatch(cfqd, cfqq)) |
3520 | return false; | |
3521 | ||
3522 | /* | |
3523 | * follow expired path, else get first next available | |
3524 | */ | |
0b182d61 JA |
3525 | if (!rq) |
3526 | rq = cfqq->next_rq; | |
3932a86b GC |
3527 | else |
3528 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); | |
0b182d61 JA |
3529 | |
3530 | /* | |
3531 | * insert request into driver dispatch list | |
3532 | */ | |
3533 | cfq_dispatch_insert(cfqd->queue, rq); | |
3534 | ||
3535 | if (!cfqd->active_cic) { | |
c5869807 | 3536 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 3537 | |
c5869807 | 3538 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
3539 | cfqd->active_cic = cic; |
3540 | } | |
3541 | ||
3542 | return true; | |
3543 | } | |
3544 | ||
3545 | /* | |
3546 | * Find the cfqq that we need to service and move a request from that to the | |
3547 | * dispatch list | |
3548 | */ | |
3549 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
3550 | { | |
3551 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3552 | struct cfq_queue *cfqq; | |
3553 | ||
3554 | if (!cfqd->busy_queues) | |
3555 | return 0; | |
3556 | ||
3557 | if (unlikely(force)) | |
3558 | return cfq_forced_dispatch(cfqd); | |
3559 | ||
3560 | cfqq = cfq_select_queue(cfqd); | |
3561 | if (!cfqq) | |
8e296755 JA |
3562 | return 0; |
3563 | ||
2f5cb738 | 3564 | /* |
0b182d61 | 3565 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 3566 | */ |
0b182d61 JA |
3567 | if (!cfq_dispatch_request(cfqd, cfqq)) |
3568 | return 0; | |
3569 | ||
2f5cb738 | 3570 | cfqq->slice_dispatch++; |
b029195d | 3571 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 3572 | |
2f5cb738 JA |
3573 | /* |
3574 | * expire an async queue immediately if it has used up its slice. idle | |
3575 | * queue always expire after 1 dispatch round. | |
3576 | */ | |
3577 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
3578 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
3579 | cfq_class_idle(cfqq))) { | |
9a7f38c4 | 3580 | cfqq->slice_end = ktime_get_ns() + 1; |
e5ff082e | 3581 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
3582 | } |
3583 | ||
b217a903 | 3584 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 3585 | return 1; |
1da177e4 LT |
3586 | } |
3587 | ||
1da177e4 | 3588 | /* |
5e705374 JA |
3589 | * task holds one reference to the queue, dropped when task exits. each rq |
3590 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 3591 | * |
b1c35769 | 3592 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
3593 | * queue lock must be held here. |
3594 | */ | |
3595 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
3596 | { | |
22e2c507 | 3597 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 3598 | struct cfq_group *cfqg; |
22e2c507 | 3599 | |
30d7b944 | 3600 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 3601 | |
30d7b944 SL |
3602 | cfqq->ref--; |
3603 | if (cfqq->ref) | |
1da177e4 LT |
3604 | return; |
3605 | ||
7b679138 | 3606 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 3607 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 3608 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 3609 | cfqg = cfqq->cfqg; |
1da177e4 | 3610 | |
28f95cbc | 3611 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 3612 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 3613 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 3614 | } |
22e2c507 | 3615 | |
f04a6424 | 3616 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 3617 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 3618 | cfqg_put(cfqg); |
1da177e4 LT |
3619 | } |
3620 | ||
d02a2c07 | 3621 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 3622 | { |
df5fe3e8 JM |
3623 | struct cfq_queue *__cfqq, *next; |
3624 | ||
df5fe3e8 JM |
3625 | /* |
3626 | * If this queue was scheduled to merge with another queue, be | |
3627 | * sure to drop the reference taken on that queue (and others in | |
3628 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
3629 | */ | |
3630 | __cfqq = cfqq->new_cfqq; | |
3631 | while (__cfqq) { | |
3632 | if (__cfqq == cfqq) { | |
3633 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
3634 | break; | |
3635 | } | |
3636 | next = __cfqq->new_cfqq; | |
3637 | cfq_put_queue(__cfqq); | |
3638 | __cfqq = next; | |
3639 | } | |
d02a2c07 SL |
3640 | } |
3641 | ||
3642 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3643 | { | |
3644 | if (unlikely(cfqq == cfqd->active_queue)) { | |
3645 | __cfq_slice_expired(cfqd, cfqq, 0); | |
3646 | cfq_schedule_dispatch(cfqd); | |
3647 | } | |
3648 | ||
3649 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 3650 | |
89850f7e JA |
3651 | cfq_put_queue(cfqq); |
3652 | } | |
22e2c507 | 3653 | |
9b84cacd TH |
3654 | static void cfq_init_icq(struct io_cq *icq) |
3655 | { | |
3656 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
3657 | ||
9a7f38c4 | 3658 | cic->ttime.last_end_request = ktime_get_ns(); |
9b84cacd TH |
3659 | } |
3660 | ||
c5869807 | 3661 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 3662 | { |
c5869807 | 3663 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 3664 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 3665 | |
563180a4 TH |
3666 | if (cic_to_cfqq(cic, false)) { |
3667 | cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, false)); | |
3668 | cic_set_cfqq(cic, NULL, false); | |
12a05732 AV |
3669 | } |
3670 | ||
563180a4 TH |
3671 | if (cic_to_cfqq(cic, true)) { |
3672 | cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, true)); | |
3673 | cic_set_cfqq(cic, NULL, true); | |
12a05732 | 3674 | } |
89850f7e JA |
3675 | } |
3676 | ||
abede6da | 3677 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
3678 | { |
3679 | struct task_struct *tsk = current; | |
3680 | int ioprio_class; | |
3681 | ||
3b18152c | 3682 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
3683 | return; |
3684 | ||
598971bf | 3685 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 3686 | switch (ioprio_class) { |
fe094d98 JA |
3687 | default: |
3688 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
3689 | case IOPRIO_CLASS_NONE: | |
3690 | /* | |
6d63c275 | 3691 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
3692 | */ |
3693 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 3694 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
3695 | break; |
3696 | case IOPRIO_CLASS_RT: | |
598971bf | 3697 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3698 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
3699 | break; | |
3700 | case IOPRIO_CLASS_BE: | |
598971bf | 3701 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3702 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3703 | break; | |
3704 | case IOPRIO_CLASS_IDLE: | |
3705 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
3706 | cfqq->ioprio = 7; | |
3707 | cfq_clear_cfqq_idle_window(cfqq); | |
3708 | break; | |
22e2c507 JA |
3709 | } |
3710 | ||
3711 | /* | |
3712 | * keep track of original prio settings in case we have to temporarily | |
3713 | * elevate the priority of this queue | |
3714 | */ | |
3715 | cfqq->org_ioprio = cfqq->ioprio; | |
b8269db4 | 3716 | cfqq->org_ioprio_class = cfqq->ioprio_class; |
3b18152c | 3717 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
3718 | } |
3719 | ||
598971bf | 3720 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 3721 | { |
598971bf | 3722 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 3723 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 3724 | struct cfq_queue *cfqq; |
35e6077c | 3725 | |
598971bf TH |
3726 | /* |
3727 | * Check whether ioprio has changed. The condition may trigger | |
3728 | * spuriously on a newly created cic but there's no harm. | |
3729 | */ | |
3730 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
3731 | return; |
3732 | ||
563180a4 | 3733 | cfqq = cic_to_cfqq(cic, false); |
caaa5f9f | 3734 | if (cfqq) { |
563180a4 | 3735 | cfq_put_queue(cfqq); |
2da8de0b | 3736 | cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio); |
563180a4 | 3737 | cic_set_cfqq(cic, cfqq, false); |
22e2c507 | 3738 | } |
caaa5f9f | 3739 | |
563180a4 | 3740 | cfqq = cic_to_cfqq(cic, true); |
caaa5f9f JA |
3741 | if (cfqq) |
3742 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
3743 | |
3744 | cic->ioprio = ioprio; | |
22e2c507 JA |
3745 | } |
3746 | ||
d5036d77 | 3747 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 3748 | pid_t pid, bool is_sync) |
d5036d77 JA |
3749 | { |
3750 | RB_CLEAR_NODE(&cfqq->rb_node); | |
3751 | RB_CLEAR_NODE(&cfqq->p_node); | |
3752 | INIT_LIST_HEAD(&cfqq->fifo); | |
3753 | ||
30d7b944 | 3754 | cfqq->ref = 0; |
d5036d77 JA |
3755 | cfqq->cfqd = cfqd; |
3756 | ||
3757 | cfq_mark_cfqq_prio_changed(cfqq); | |
3758 | ||
3759 | if (is_sync) { | |
3760 | if (!cfq_class_idle(cfqq)) | |
3761 | cfq_mark_cfqq_idle_window(cfqq); | |
3762 | cfq_mark_cfqq_sync(cfqq); | |
3763 | } | |
3764 | cfqq->pid = pid; | |
3765 | } | |
3766 | ||
24610333 | 3767 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
142bbdfc | 3768 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 3769 | { |
bca4b914 | 3770 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
60a83707 | 3771 | struct cfq_queue *cfqq; |
f4da8072 | 3772 | uint64_t serial_nr; |
24610333 | 3773 | |
598971bf | 3774 | rcu_read_lock(); |
f4da8072 | 3775 | serial_nr = bio_blkcg(bio)->css.serial_nr; |
598971bf | 3776 | rcu_read_unlock(); |
24610333 | 3777 | |
598971bf TH |
3778 | /* |
3779 | * Check whether blkcg has changed. The condition may trigger | |
3780 | * spuriously on a newly created cic but there's no harm. | |
3781 | */ | |
f4da8072 | 3782 | if (unlikely(!cfqd) || likely(cic->blkcg_serial_nr == serial_nr)) |
142bbdfc | 3783 | return; |
87760e5e | 3784 | |
60a83707 TH |
3785 | /* |
3786 | * Drop reference to queues. New queues will be assigned in new | |
3787 | * group upon arrival of fresh requests. | |
3788 | */ | |
3789 | cfqq = cic_to_cfqq(cic, false); | |
3790 | if (cfqq) { | |
3791 | cfq_log_cfqq(cfqd, cfqq, "changed cgroup"); | |
3792 | cic_set_cfqq(cic, NULL, false); | |
3793 | cfq_put_queue(cfqq); | |
3794 | } | |
3795 | ||
3796 | cfqq = cic_to_cfqq(cic, true); | |
3797 | if (cfqq) { | |
3798 | cfq_log_cfqq(cfqd, cfqq, "changed cgroup"); | |
3799 | cic_set_cfqq(cic, NULL, true); | |
3800 | cfq_put_queue(cfqq); | |
24610333 | 3801 | } |
598971bf | 3802 | |
f4da8072 | 3803 | cic->blkcg_serial_nr = serial_nr; |
24610333 | 3804 | } |
598971bf | 3805 | #else |
142bbdfc | 3806 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
5d7f5ce1 | 3807 | { |
5d7f5ce1 | 3808 | } |
24610333 VG |
3809 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
3810 | ||
c2dea2d1 | 3811 | static struct cfq_queue ** |
60a83707 | 3812 | cfq_async_queue_prio(struct cfq_group *cfqg, int ioprio_class, int ioprio) |
c2dea2d1 | 3813 | { |
fe094d98 | 3814 | switch (ioprio_class) { |
c2dea2d1 | 3815 | case IOPRIO_CLASS_RT: |
60a83707 | 3816 | return &cfqg->async_cfqq[0][ioprio]; |
598971bf TH |
3817 | case IOPRIO_CLASS_NONE: |
3818 | ioprio = IOPRIO_NORM; | |
3819 | /* fall through */ | |
c2dea2d1 | 3820 | case IOPRIO_CLASS_BE: |
60a83707 | 3821 | return &cfqg->async_cfqq[1][ioprio]; |
c2dea2d1 | 3822 | case IOPRIO_CLASS_IDLE: |
60a83707 | 3823 | return &cfqg->async_idle_cfqq; |
c2dea2d1 VT |
3824 | default: |
3825 | BUG(); | |
3826 | } | |
3827 | } | |
3828 | ||
15c31be4 | 3829 | static struct cfq_queue * |
abede6da | 3830 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
2da8de0b | 3831 | struct bio *bio) |
15c31be4 | 3832 | { |
c6ce1943 JM |
3833 | int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
3834 | int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
d4aad7ff | 3835 | struct cfq_queue **async_cfqq = NULL; |
4ebc1c61 | 3836 | struct cfq_queue *cfqq; |
322731ed TH |
3837 | struct cfq_group *cfqg; |
3838 | ||
3839 | rcu_read_lock(); | |
ae118896 | 3840 | cfqg = cfq_lookup_cfqg(cfqd, bio_blkcg(bio)); |
322731ed TH |
3841 | if (!cfqg) { |
3842 | cfqq = &cfqd->oom_cfqq; | |
3843 | goto out; | |
3844 | } | |
15c31be4 | 3845 | |
c2dea2d1 | 3846 | if (!is_sync) { |
c6ce1943 JM |
3847 | if (!ioprio_valid(cic->ioprio)) { |
3848 | struct task_struct *tsk = current; | |
3849 | ioprio = task_nice_ioprio(tsk); | |
3850 | ioprio_class = task_nice_ioclass(tsk); | |
3851 | } | |
60a83707 | 3852 | async_cfqq = cfq_async_queue_prio(cfqg, ioprio_class, ioprio); |
c2dea2d1 | 3853 | cfqq = *async_cfqq; |
4ebc1c61 TH |
3854 | if (cfqq) |
3855 | goto out; | |
c2dea2d1 VT |
3856 | } |
3857 | ||
e00f4f4d TH |
3858 | cfqq = kmem_cache_alloc_node(cfq_pool, |
3859 | GFP_NOWAIT | __GFP_ZERO | __GFP_NOWARN, | |
d4aad7ff TH |
3860 | cfqd->queue->node); |
3861 | if (!cfqq) { | |
3862 | cfqq = &cfqd->oom_cfqq; | |
3863 | goto out; | |
3864 | } | |
3865 | ||
4d608baa AP |
3866 | /* cfq_init_cfqq() assumes cfqq->ioprio_class is initialized. */ |
3867 | cfqq->ioprio_class = IOPRIO_CLASS_NONE; | |
d4aad7ff TH |
3868 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); |
3869 | cfq_init_prio_data(cfqq, cic); | |
3870 | cfq_link_cfqq_cfqg(cfqq, cfqg); | |
3871 | cfq_log_cfqq(cfqd, cfqq, "alloced"); | |
15c31be4 | 3872 | |
d4aad7ff TH |
3873 | if (async_cfqq) { |
3874 | /* a new async queue is created, pin and remember */ | |
30d7b944 | 3875 | cfqq->ref++; |
c2dea2d1 | 3876 | *async_cfqq = cfqq; |
15c31be4 | 3877 | } |
4ebc1c61 | 3878 | out: |
30d7b944 | 3879 | cfqq->ref++; |
322731ed | 3880 | rcu_read_unlock(); |
15c31be4 JA |
3881 | return cfqq; |
3882 | } | |
3883 | ||
22e2c507 | 3884 | static void |
9a7f38c4 | 3885 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, u64 slice_idle) |
1da177e4 | 3886 | { |
9a7f38c4 | 3887 | u64 elapsed = ktime_get_ns() - ttime->last_end_request; |
383cd721 | 3888 | elapsed = min(elapsed, 2UL * slice_idle); |
db3b5848 | 3889 | |
383cd721 | 3890 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
9a7f38c4 JM |
3891 | ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed, 8); |
3892 | ttime->ttime_mean = div64_ul(ttime->ttime_total + 128, | |
3893 | ttime->ttime_samples); | |
383cd721 SL |
3894 | } |
3895 | ||
3896 | static void | |
3897 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3898 | struct cfq_io_cq *cic) |
383cd721 | 3899 | { |
f5f2b6ce | 3900 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3901 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3902 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3903 | cfqd->cfq_slice_idle); | |
3904 | } | |
7700fc4f SL |
3905 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3906 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3907 | #endif | |
22e2c507 | 3908 | } |
1da177e4 | 3909 | |
206dc69b | 3910 | static void |
b2c18e1e | 3911 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3912 | struct request *rq) |
206dc69b | 3913 | { |
3dde36dd | 3914 | sector_t sdist = 0; |
41647e7a | 3915 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3916 | if (cfqq->last_request_pos) { |
3917 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3918 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3919 | else | |
3920 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3921 | } | |
206dc69b | 3922 | |
3dde36dd | 3923 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3924 | if (blk_queue_nonrot(cfqd->queue)) |
3925 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3926 | else | |
3927 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3928 | } |
1da177e4 | 3929 | |
a2b80967 CH |
3930 | static inline bool req_noidle(struct request *req) |
3931 | { | |
3932 | return req_op(req) == REQ_OP_WRITE && | |
3933 | (req->cmd_flags & (REQ_SYNC | REQ_IDLE)) == REQ_SYNC; | |
3934 | } | |
3935 | ||
22e2c507 JA |
3936 | /* |
3937 | * Disable idle window if the process thinks too long or seeks so much that | |
3938 | * it doesn't matter | |
3939 | */ | |
3940 | static void | |
3941 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3942 | struct cfq_io_cq *cic) |
22e2c507 | 3943 | { |
7b679138 | 3944 | int old_idle, enable_idle; |
1be92f2f | 3945 | |
0871714e JA |
3946 | /* |
3947 | * Don't idle for async or idle io prio class | |
3948 | */ | |
3949 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3950 | return; |
3951 | ||
c265a7f4 | 3952 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3953 | |
76280aff CZ |
3954 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3955 | cfq_mark_cfqq_deep(cfqq); | |
3956 | ||
a2b80967 | 3957 | if (cfqq->next_rq && req_noidle(cfqq->next_rq)) |
749ef9f8 | 3958 | enable_idle = 0; |
f6e8d01b | 3959 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
3960 | !cfqd->cfq_slice_idle || |
3961 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3962 | enable_idle = 0; |
383cd721 SL |
3963 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3964 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3965 | enable_idle = 0; |
3966 | else | |
3967 | enable_idle = 1; | |
1da177e4 LT |
3968 | } |
3969 | ||
7b679138 JA |
3970 | if (old_idle != enable_idle) { |
3971 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3972 | if (enable_idle) | |
3973 | cfq_mark_cfqq_idle_window(cfqq); | |
3974 | else | |
3975 | cfq_clear_cfqq_idle_window(cfqq); | |
3976 | } | |
22e2c507 | 3977 | } |
1da177e4 | 3978 | |
22e2c507 JA |
3979 | /* |
3980 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3981 | * no or if we aren't sure, a 1 will cause a preempt. | |
3982 | */ | |
a6151c3a | 3983 | static bool |
22e2c507 | 3984 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3985 | struct request *rq) |
22e2c507 | 3986 | { |
6d048f53 | 3987 | struct cfq_queue *cfqq; |
22e2c507 | 3988 | |
6d048f53 JA |
3989 | cfqq = cfqd->active_queue; |
3990 | if (!cfqq) | |
a6151c3a | 3991 | return false; |
22e2c507 | 3992 | |
6d048f53 | 3993 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3994 | return false; |
22e2c507 JA |
3995 | |
3996 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3997 | return true; |
1e3335de | 3998 | |
875feb63 DS |
3999 | /* |
4000 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
4001 | */ | |
4002 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
4003 | return false; | |
4004 | ||
374f84ac JA |
4005 | /* |
4006 | * if the new request is sync, but the currently running queue is | |
4007 | * not, let the sync request have priority. | |
4008 | */ | |
3932a86b | 4009 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) |
a6151c3a | 4010 | return true; |
1e3335de | 4011 | |
3984aa55 JK |
4012 | /* |
4013 | * Treat ancestors of current cgroup the same way as current cgroup. | |
4014 | * For anybody else we disallow preemption to guarantee service | |
4015 | * fairness among cgroups. | |
4016 | */ | |
4017 | if (!cfqg_is_descendant(cfqq->cfqg, new_cfqq->cfqg)) | |
8682e1f1 VG |
4018 | return false; |
4019 | ||
4020 | if (cfq_slice_used(cfqq)) | |
4021 | return true; | |
4022 | ||
6c80731c JK |
4023 | /* |
4024 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
4025 | */ | |
4026 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
4027 | return true; | |
4028 | ||
4029 | WARN_ON_ONCE(cfqq->ioprio_class != new_cfqq->ioprio_class); | |
8682e1f1 | 4030 | /* Allow preemption only if we are idling on sync-noidle tree */ |
4d2ceea4 | 4031 | if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 | 4032 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 VG |
4033 | RB_EMPTY_ROOT(&cfqq->sort_list)) |
4034 | return true; | |
4035 | ||
b53d1ed7 JA |
4036 | /* |
4037 | * So both queues are sync. Let the new request get disk time if | |
4038 | * it's a metadata request and the current queue is doing regular IO. | |
4039 | */ | |
65299a3b | 4040 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
4041 | return true; |
4042 | ||
d2d59e18 SL |
4043 | /* An idle queue should not be idle now for some reason */ |
4044 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
4045 | return true; | |
4046 | ||
1e3335de | 4047 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 4048 | return false; |
1e3335de JA |
4049 | |
4050 | /* | |
4051 | * if this request is as-good as one we would expect from the | |
4052 | * current cfqq, let it preempt | |
4053 | */ | |
e9ce335d | 4054 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 4055 | return true; |
1e3335de | 4056 | |
a6151c3a | 4057 | return false; |
22e2c507 JA |
4058 | } |
4059 | ||
4060 | /* | |
4061 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
4062 | * let it have half of its nominal slice. | |
4063 | */ | |
4064 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
4065 | { | |
df0793ab SL |
4066 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
4067 | ||
7b679138 | 4068 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 4069 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 4070 | |
f8ae6e3e SL |
4071 | /* |
4072 | * workload type is changed, don't save slice, otherwise preempt | |
4073 | * doesn't happen | |
4074 | */ | |
df0793ab | 4075 | if (old_type != cfqq_type(cfqq)) |
4d2ceea4 | 4076 | cfqq->cfqg->saved_wl_slice = 0; |
f8ae6e3e | 4077 | |
bf572256 JA |
4078 | /* |
4079 | * Put the new queue at the front of the of the current list, | |
4080 | * so we know that it will be selected next. | |
4081 | */ | |
4082 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
4083 | |
4084 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 4085 | |
62a37f6b JT |
4086 | cfqq->slice_end = 0; |
4087 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
4088 | } |
4089 | ||
22e2c507 | 4090 | /* |
5e705374 | 4091 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
4092 | * something we should do about it |
4093 | */ | |
4094 | static void | |
5e705374 JA |
4095 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
4096 | struct request *rq) | |
22e2c507 | 4097 | { |
c5869807 | 4098 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 4099 | |
45333d5a | 4100 | cfqd->rq_queued++; |
65299a3b CH |
4101 | if (rq->cmd_flags & REQ_PRIO) |
4102 | cfqq->prio_pending++; | |
374f84ac | 4103 | |
383cd721 | 4104 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 4105 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
4106 | cfq_update_idle_window(cfqd, cfqq, cic); |
4107 | ||
b2c18e1e | 4108 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
4109 | |
4110 | if (cfqq == cfqd->active_queue) { | |
4111 | /* | |
b029195d JA |
4112 | * Remember that we saw a request from this process, but |
4113 | * don't start queuing just yet. Otherwise we risk seeing lots | |
4114 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
4115 | * and merging. If the request is already larger than a single |
4116 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
4117 | * merging is already done. Ditto for a busy system that |
4118 | * has other work pending, don't risk delaying until the | |
4119 | * idle timer unplug to continue working. | |
22e2c507 | 4120 | */ |
d6ceb25e | 4121 | if (cfq_cfqq_wait_request(cfqq)) { |
09cbfeaf | 4122 | if (blk_rq_bytes(rq) > PAGE_SIZE || |
2d870722 | 4123 | cfqd->busy_queues > 1) { |
812df48d | 4124 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 4125 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 4126 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 4127 | } else { |
155fead9 | 4128 | cfqg_stats_update_idle_time(cfqq->cfqg); |
bf791937 | 4129 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 4130 | } |
d6ceb25e | 4131 | } |
5e705374 | 4132 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
4133 | /* |
4134 | * not the active queue - expire current slice if it is | |
4135 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
4136 | * has some old slice time left and is of higher priority or |
4137 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
4138 | */ |
4139 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 4140 | __blk_run_queue(cfqd->queue); |
22e2c507 | 4141 | } |
1da177e4 LT |
4142 | } |
4143 | ||
165125e1 | 4144 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 4145 | { |
b4878f24 | 4146 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 4147 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 4148 | |
7b679138 | 4149 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 4150 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 4151 | |
9a7f38c4 | 4152 | rq->fifo_time = ktime_get_ns() + cfqd->cfq_fifo_expire[rq_is_sync(rq)]; |
22e2c507 | 4153 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 4154 | cfq_add_rq_rb(rq); |
ef295ecf | 4155 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
155fead9 | 4156 | rq->cmd_flags); |
5e705374 | 4157 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
4158 | } |
4159 | ||
45333d5a AC |
4160 | /* |
4161 | * Update hw_tag based on peak queue depth over 50 samples under | |
4162 | * sufficient load. | |
4163 | */ | |
4164 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
4165 | { | |
1a1238a7 SL |
4166 | struct cfq_queue *cfqq = cfqd->active_queue; |
4167 | ||
53c583d2 CZ |
4168 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
4169 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
4170 | |
4171 | if (cfqd->hw_tag == 1) | |
4172 | return; | |
45333d5a AC |
4173 | |
4174 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 4175 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
4176 | return; |
4177 | ||
1a1238a7 SL |
4178 | /* |
4179 | * If active queue hasn't enough requests and can idle, cfq might not | |
4180 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
4181 | * case | |
4182 | */ | |
4183 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
4184 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 4185 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
4186 | return; |
4187 | ||
45333d5a AC |
4188 | if (cfqd->hw_tag_samples++ < 50) |
4189 | return; | |
4190 | ||
e459dd08 | 4191 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
4192 | cfqd->hw_tag = 1; |
4193 | else | |
4194 | cfqd->hw_tag = 0; | |
45333d5a AC |
4195 | } |
4196 | ||
7667aa06 VG |
4197 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
4198 | { | |
c5869807 | 4199 | struct cfq_io_cq *cic = cfqd->active_cic; |
9a7f38c4 | 4200 | u64 now = ktime_get_ns(); |
7667aa06 | 4201 | |
02a8f01b JT |
4202 | /* If the queue already has requests, don't wait */ |
4203 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
4204 | return false; | |
4205 | ||
7667aa06 VG |
4206 | /* If there are other queues in the group, don't wait */ |
4207 | if (cfqq->cfqg->nr_cfqq > 1) | |
4208 | return false; | |
4209 | ||
7700fc4f SL |
4210 | /* the only queue in the group, but think time is big */ |
4211 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
4212 | return false; | |
4213 | ||
7667aa06 VG |
4214 | if (cfq_slice_used(cfqq)) |
4215 | return true; | |
4216 | ||
4217 | /* if slice left is less than think time, wait busy */ | |
383cd721 | 4218 | if (cic && sample_valid(cic->ttime.ttime_samples) |
9a7f38c4 | 4219 | && (cfqq->slice_end - now < cic->ttime.ttime_mean)) |
7667aa06 VG |
4220 | return true; |
4221 | ||
4222 | /* | |
4223 | * If think times is less than a jiffy than ttime_mean=0 and above | |
4224 | * will not be true. It might happen that slice has not expired yet | |
4225 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
4226 | * case where think time is less than a jiffy, mark the queue wait | |
4227 | * busy if only 1 jiffy is left in the slice. | |
4228 | */ | |
9a7f38c4 | 4229 | if (cfqq->slice_end - now <= jiffies_to_nsecs(1)) |
7667aa06 VG |
4230 | return true; |
4231 | ||
4232 | return false; | |
4233 | } | |
4234 | ||
165125e1 | 4235 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 4236 | { |
5e705374 | 4237 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 4238 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 4239 | const int sync = rq_is_sync(rq); |
9a7f38c4 | 4240 | u64 now = ktime_get_ns(); |
1da177e4 | 4241 | |
a2b80967 | 4242 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", req_noidle(rq)); |
1da177e4 | 4243 | |
45333d5a AC |
4244 | cfq_update_hw_tag(cfqd); |
4245 | ||
53c583d2 | 4246 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 4247 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 4248 | cfqd->rq_in_driver--; |
6d048f53 | 4249 | cfqq->dispatched--; |
80bdf0c7 | 4250 | (RQ_CFQG(rq))->dispatched--; |
155fead9 | 4251 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
ef295ecf | 4252 | rq_io_start_time_ns(rq), rq->cmd_flags); |
1da177e4 | 4253 | |
53c583d2 | 4254 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 4255 | |
365722bb | 4256 | if (sync) { |
34b98d03 | 4257 | struct cfq_rb_root *st; |
f5f2b6ce | 4258 | |
383cd721 | 4259 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
4260 | |
4261 | if (cfq_cfqq_on_rr(cfqq)) | |
34b98d03 | 4262 | st = cfqq->service_tree; |
f5f2b6ce | 4263 | else |
34b98d03 VG |
4264 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), |
4265 | cfqq_type(cfqq)); | |
4266 | ||
4267 | st->ttime.last_end_request = now; | |
149321a6 JK |
4268 | /* |
4269 | * We have to do this check in jiffies since start_time is in | |
4270 | * jiffies and it is not trivial to convert to ns. If | |
4271 | * cfq_fifo_expire[1] ever comes close to 1 jiffie, this test | |
4272 | * will become problematic but so far we are fine (the default | |
4273 | * is 128 ms). | |
4274 | */ | |
4275 | if (!time_after(rq->start_time + | |
4276 | nsecs_to_jiffies(cfqd->cfq_fifo_expire[1]), | |
4277 | jiffies)) | |
573412b2 | 4278 | cfqd->last_delayed_sync = now; |
365722bb | 4279 | } |
caaa5f9f | 4280 | |
7700fc4f SL |
4281 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4282 | cfqq->cfqg->ttime.last_end_request = now; | |
4283 | #endif | |
4284 | ||
caaa5f9f JA |
4285 | /* |
4286 | * If this is the active queue, check if it needs to be expired, | |
4287 | * or if we want to idle in case it has no pending requests. | |
4288 | */ | |
4289 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
4290 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
4291 | ||
44f7c160 JA |
4292 | if (cfq_cfqq_slice_new(cfqq)) { |
4293 | cfq_set_prio_slice(cfqd, cfqq); | |
4294 | cfq_clear_cfqq_slice_new(cfqq); | |
4295 | } | |
f75edf2d VG |
4296 | |
4297 | /* | |
7667aa06 VG |
4298 | * Should we wait for next request to come in before we expire |
4299 | * the queue. | |
f75edf2d | 4300 | */ |
7667aa06 | 4301 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
9a7f38c4 | 4302 | u64 extend_sl = cfqd->cfq_slice_idle; |
80bdf0c7 VG |
4303 | if (!cfqd->cfq_slice_idle) |
4304 | extend_sl = cfqd->cfq_group_idle; | |
9a7f38c4 | 4305 | cfqq->slice_end = now + extend_sl; |
f75edf2d | 4306 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 4307 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
4308 | } |
4309 | ||
a36e71f9 | 4310 | /* |
8e550632 CZ |
4311 | * Idling is not enabled on: |
4312 | * - expired queues | |
4313 | * - idle-priority queues | |
4314 | * - async queues | |
4315 | * - queues with still some requests queued | |
4316 | * - when there is a close cooperator | |
a36e71f9 | 4317 | */ |
0871714e | 4318 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 4319 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
4320 | else if (sync && cfqq_empty && |
4321 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 4322 | cfq_arm_slice_timer(cfqd); |
8e550632 | 4323 | } |
caaa5f9f | 4324 | } |
6d048f53 | 4325 | |
53c583d2 | 4326 | if (!cfqd->rq_in_driver) |
23e018a1 | 4327 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
4328 | } |
4329 | ||
ef295ecf | 4330 | static void cfqq_boost_on_prio(struct cfq_queue *cfqq, unsigned int op) |
b8269db4 JA |
4331 | { |
4332 | /* | |
4333 | * If REQ_PRIO is set, boost class and prio level, if it's below | |
4334 | * BE/NORM. If prio is not set, restore the potentially boosted | |
4335 | * class/prio level. | |
4336 | */ | |
ef295ecf | 4337 | if (!(op & REQ_PRIO)) { |
b8269db4 JA |
4338 | cfqq->ioprio_class = cfqq->org_ioprio_class; |
4339 | cfqq->ioprio = cfqq->org_ioprio; | |
4340 | } else { | |
4341 | if (cfq_class_idle(cfqq)) | |
4342 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
4343 | if (cfqq->ioprio > IOPRIO_NORM) | |
4344 | cfqq->ioprio = IOPRIO_NORM; | |
4345 | } | |
4346 | } | |
4347 | ||
89850f7e | 4348 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 4349 | { |
1b379d8d | 4350 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 4351 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 4352 | return ELV_MQUEUE_MUST; |
3b18152c | 4353 | } |
1da177e4 | 4354 | |
22e2c507 | 4355 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
4356 | } |
4357 | ||
ef295ecf | 4358 | static int cfq_may_queue(struct request_queue *q, unsigned int op) |
22e2c507 JA |
4359 | { |
4360 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
4361 | struct task_struct *tsk = current; | |
c5869807 | 4362 | struct cfq_io_cq *cic; |
22e2c507 JA |
4363 | struct cfq_queue *cfqq; |
4364 | ||
4365 | /* | |
4366 | * don't force setup of a queue from here, as a call to may_queue | |
4367 | * does not necessarily imply that a request actually will be queued. | |
4368 | * so just lookup a possibly existing queue, or return 'may queue' | |
4369 | * if that fails | |
4370 | */ | |
4ac845a2 | 4371 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
4372 | if (!cic) |
4373 | return ELV_MQUEUE_MAY; | |
4374 | ||
ef295ecf | 4375 | cfqq = cic_to_cfqq(cic, op_is_sync(op)); |
22e2c507 | 4376 | if (cfqq) { |
abede6da | 4377 | cfq_init_prio_data(cfqq, cic); |
ef295ecf | 4378 | cfqq_boost_on_prio(cfqq, op); |
22e2c507 | 4379 | |
89850f7e | 4380 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
4381 | } |
4382 | ||
4383 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
4384 | } |
4385 | ||
1da177e4 LT |
4386 | /* |
4387 | * queue lock held here | |
4388 | */ | |
bb37b94c | 4389 | static void cfq_put_request(struct request *rq) |
1da177e4 | 4390 | { |
5e705374 | 4391 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 4392 | |
5e705374 | 4393 | if (cfqq) { |
22e2c507 | 4394 | const int rw = rq_data_dir(rq); |
1da177e4 | 4395 | |
22e2c507 JA |
4396 | BUG_ON(!cfqq->allocated[rw]); |
4397 | cfqq->allocated[rw]--; | |
1da177e4 | 4398 | |
7f1dc8a2 | 4399 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 4400 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
4401 | rq->elv.priv[0] = NULL; |
4402 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 4403 | |
1da177e4 LT |
4404 | cfq_put_queue(cfqq); |
4405 | } | |
4406 | } | |
4407 | ||
df5fe3e8 | 4408 | static struct cfq_queue * |
c5869807 | 4409 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
4410 | struct cfq_queue *cfqq) |
4411 | { | |
4412 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
4413 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 4414 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
4415 | cfq_put_queue(cfqq); |
4416 | return cic_to_cfqq(cic, 1); | |
4417 | } | |
4418 | ||
e6c5bc73 JM |
4419 | /* |
4420 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
4421 | * was the last process referring to said cfqq. | |
4422 | */ | |
4423 | static struct cfq_queue * | |
c5869807 | 4424 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
4425 | { |
4426 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
4427 | cfqq->pid = current->pid; |
4428 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 4429 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
4430 | return cfqq; |
4431 | } | |
4432 | ||
4433 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
4434 | |
4435 | cfq_put_cooperator(cfqq); | |
4436 | ||
e6c5bc73 JM |
4437 | cfq_put_queue(cfqq); |
4438 | return NULL; | |
4439 | } | |
1da177e4 | 4440 | /* |
22e2c507 | 4441 | * Allocate cfq data structures associated with this request. |
1da177e4 | 4442 | */ |
22e2c507 | 4443 | static int |
852c788f TH |
4444 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
4445 | gfp_t gfp_mask) | |
1da177e4 LT |
4446 | { |
4447 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 4448 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 4449 | const int rw = rq_data_dir(rq); |
a6151c3a | 4450 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 4451 | struct cfq_queue *cfqq; |
1da177e4 | 4452 | |
216284c3 | 4453 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 4454 | |
598971bf | 4455 | check_ioprio_changed(cic, bio); |
142bbdfc | 4456 | check_blkcg_changed(cic, bio); |
e6c5bc73 | 4457 | new_queue: |
91fac317 | 4458 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 4459 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
bce6133b TH |
4460 | if (cfqq) |
4461 | cfq_put_queue(cfqq); | |
2da8de0b | 4462 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio); |
91fac317 | 4463 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 4464 | } else { |
e6c5bc73 JM |
4465 | /* |
4466 | * If the queue was seeky for too long, break it apart. | |
4467 | */ | |
ae54abed | 4468 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
4469 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
4470 | cfqq = split_cfqq(cic, cfqq); | |
4471 | if (!cfqq) | |
4472 | goto new_queue; | |
4473 | } | |
4474 | ||
df5fe3e8 JM |
4475 | /* |
4476 | * Check to see if this queue is scheduled to merge with | |
4477 | * another, closely cooperating queue. The merging of | |
4478 | * queues happens here as it must be done in process context. | |
4479 | * The reference on new_cfqq was taken in merge_cfqqs. | |
4480 | */ | |
4481 | if (cfqq->new_cfqq) | |
4482 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 4483 | } |
1da177e4 LT |
4484 | |
4485 | cfqq->allocated[rw]++; | |
1da177e4 | 4486 | |
6fae9c25 | 4487 | cfqq->ref++; |
eb7d8c07 | 4488 | cfqg_get(cfqq->cfqg); |
a612fddf | 4489 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 4490 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 4491 | spin_unlock_irq(q->queue_lock); |
5d7f5ce1 | 4492 | |
5e705374 | 4493 | return 0; |
1da177e4 LT |
4494 | } |
4495 | ||
65f27f38 | 4496 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 4497 | { |
65f27f38 | 4498 | struct cfq_data *cfqd = |
23e018a1 | 4499 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 4500 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4501 | |
40bb54d1 | 4502 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 4503 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 4504 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
4505 | } |
4506 | ||
4507 | /* | |
4508 | * Timer running if the active_queue is currently idling inside its time slice | |
4509 | */ | |
91148325 | 4510 | static enum hrtimer_restart cfq_idle_slice_timer(struct hrtimer *timer) |
22e2c507 | 4511 | { |
91148325 JK |
4512 | struct cfq_data *cfqd = container_of(timer, struct cfq_data, |
4513 | idle_slice_timer); | |
22e2c507 JA |
4514 | struct cfq_queue *cfqq; |
4515 | unsigned long flags; | |
3c6bd2f8 | 4516 | int timed_out = 1; |
22e2c507 | 4517 | |
7b679138 JA |
4518 | cfq_log(cfqd, "idle timer fired"); |
4519 | ||
22e2c507 JA |
4520 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
4521 | ||
fe094d98 JA |
4522 | cfqq = cfqd->active_queue; |
4523 | if (cfqq) { | |
3c6bd2f8 JA |
4524 | timed_out = 0; |
4525 | ||
b029195d JA |
4526 | /* |
4527 | * We saw a request before the queue expired, let it through | |
4528 | */ | |
4529 | if (cfq_cfqq_must_dispatch(cfqq)) | |
4530 | goto out_kick; | |
4531 | ||
22e2c507 JA |
4532 | /* |
4533 | * expired | |
4534 | */ | |
44f7c160 | 4535 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
4536 | goto expire; |
4537 | ||
4538 | /* | |
4539 | * only expire and reinvoke request handler, if there are | |
4540 | * other queues with pending requests | |
4541 | */ | |
caaa5f9f | 4542 | if (!cfqd->busy_queues) |
22e2c507 | 4543 | goto out_cont; |
22e2c507 JA |
4544 | |
4545 | /* | |
4546 | * not expired and it has a request pending, let it dispatch | |
4547 | */ | |
75e50984 | 4548 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 4549 | goto out_kick; |
76280aff CZ |
4550 | |
4551 | /* | |
4552 | * Queue depth flag is reset only when the idle didn't succeed | |
4553 | */ | |
4554 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
4555 | } |
4556 | expire: | |
e5ff082e | 4557 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 4558 | out_kick: |
23e018a1 | 4559 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
4560 | out_cont: |
4561 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
91148325 | 4562 | return HRTIMER_NORESTART; |
22e2c507 JA |
4563 | } |
4564 | ||
3b18152c JA |
4565 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
4566 | { | |
91148325 | 4567 | hrtimer_cancel(&cfqd->idle_slice_timer); |
23e018a1 | 4568 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 4569 | } |
22e2c507 | 4570 | |
b374d18a | 4571 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 4572 | { |
22e2c507 | 4573 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 4574 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4575 | |
3b18152c | 4576 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 4577 | |
d9ff4187 | 4578 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 4579 | |
d9ff4187 | 4580 | if (cfqd->active_queue) |
e5ff082e | 4581 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 4582 | |
03aa264a TH |
4583 | spin_unlock_irq(q->queue_lock); |
4584 | ||
a90d742e AV |
4585 | cfq_shutdown_timer_wq(cfqd); |
4586 | ||
ffea73fc TH |
4587 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4588 | blkcg_deactivate_policy(q, &blkcg_policy_cfq); | |
4589 | #else | |
f51b802c | 4590 | kfree(cfqd->root_group); |
2abae55f | 4591 | #endif |
56edf7d7 | 4592 | kfree(cfqd); |
1da177e4 LT |
4593 | } |
4594 | ||
d50235b7 | 4595 | static int cfq_init_queue(struct request_queue *q, struct elevator_type *e) |
1da177e4 LT |
4596 | { |
4597 | struct cfq_data *cfqd; | |
3c798398 | 4598 | struct blkcg_gq *blkg __maybe_unused; |
a2b1693b | 4599 | int i, ret; |
d50235b7 JM |
4600 | struct elevator_queue *eq; |
4601 | ||
4602 | eq = elevator_alloc(q, e); | |
4603 | if (!eq) | |
4604 | return -ENOMEM; | |
1da177e4 | 4605 | |
c1b511eb | 4606 | cfqd = kzalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node); |
d50235b7 JM |
4607 | if (!cfqd) { |
4608 | kobject_put(&eq->kobj); | |
b2fab5ac | 4609 | return -ENOMEM; |
d50235b7 JM |
4610 | } |
4611 | eq->elevator_data = cfqd; | |
80b15c73 | 4612 | |
f51b802c | 4613 | cfqd->queue = q; |
d50235b7 JM |
4614 | spin_lock_irq(q->queue_lock); |
4615 | q->elevator = eq; | |
4616 | spin_unlock_irq(q->queue_lock); | |
f51b802c | 4617 | |
1fa8f6d6 VG |
4618 | /* Init root service tree */ |
4619 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
4620 | ||
f51b802c | 4621 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 4622 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4623 | ret = blkcg_activate_policy(q, &blkcg_policy_cfq); |
a2b1693b TH |
4624 | if (ret) |
4625 | goto out_free; | |
f51b802c | 4626 | |
a2b1693b | 4627 | cfqd->root_group = blkg_to_cfqg(q->root_blkg); |
f51b802c | 4628 | #else |
a2b1693b | 4629 | ret = -ENOMEM; |
f51b802c TH |
4630 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
4631 | GFP_KERNEL, cfqd->queue->node); | |
a2b1693b TH |
4632 | if (!cfqd->root_group) |
4633 | goto out_free; | |
5624a4e4 | 4634 | |
a2b1693b | 4635 | cfq_init_cfqg_base(cfqd->root_group); |
3ecca629 TH |
4636 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_LEGACY_DFL; |
4637 | cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_LEGACY_DFL; | |
69d7fde5 | 4638 | #endif |
5624a4e4 | 4639 | |
26a2ac00 JA |
4640 | /* |
4641 | * Not strictly needed (since RB_ROOT just clears the node and we | |
4642 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
4643 | * to add magic to the rb code | |
4644 | */ | |
4645 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
4646 | cfqd->prio_trees[i] = RB_ROOT; | |
4647 | ||
6118b70b | 4648 | /* |
d4aad7ff | 4649 | * Our fallback cfqq if cfq_get_queue() runs into OOM issues. |
6118b70b | 4650 | * Grab a permanent reference to it, so that the normal code flow |
f51b802c TH |
4651 | * will not attempt to free it. oom_cfqq is linked to root_group |
4652 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
4653 | * the reference from linking right away. | |
6118b70b JA |
4654 | */ |
4655 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 4656 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
4657 | |
4658 | spin_lock_irq(q->queue_lock); | |
f51b802c | 4659 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 4660 | cfqg_put(cfqd->root_group); |
1adaf3dd | 4661 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 4662 | |
91148325 JK |
4663 | hrtimer_init(&cfqd->idle_slice_timer, CLOCK_MONOTONIC, |
4664 | HRTIMER_MODE_REL); | |
22e2c507 | 4665 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; |
22e2c507 | 4666 | |
23e018a1 | 4667 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 4668 | |
1da177e4 | 4669 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
4670 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
4671 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
4672 | cfqd->cfq_back_max = cfq_back_max; |
4673 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
4674 | cfqd->cfq_slice[0] = cfq_slice_async; |
4675 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
5bf14c07 | 4676 | cfqd->cfq_target_latency = cfq_target_latency; |
22e2c507 | 4677 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
0bb97947 | 4678 | cfqd->cfq_slice_idle = cfq_slice_idle; |
80bdf0c7 | 4679 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 4680 | cfqd->cfq_latency = 1; |
e459dd08 | 4681 | cfqd->hw_tag = -1; |
edc71131 CZ |
4682 | /* |
4683 | * we optimistically start assuming sync ops weren't delayed in last | |
4684 | * second, in order to have larger depth for async operations. | |
4685 | */ | |
9a7f38c4 | 4686 | cfqd->last_delayed_sync = ktime_get_ns() - NSEC_PER_SEC; |
b2fab5ac | 4687 | return 0; |
a2b1693b TH |
4688 | |
4689 | out_free: | |
4690 | kfree(cfqd); | |
d50235b7 | 4691 | kobject_put(&eq->kobj); |
a2b1693b | 4692 | return ret; |
1da177e4 LT |
4693 | } |
4694 | ||
0bb97947 JA |
4695 | static void cfq_registered_queue(struct request_queue *q) |
4696 | { | |
4697 | struct elevator_queue *e = q->elevator; | |
4698 | struct cfq_data *cfqd = e->elevator_data; | |
4699 | ||
4700 | /* | |
4701 | * Default to IOPS mode with no idling for SSDs | |
4702 | */ | |
4703 | if (blk_queue_nonrot(q)) | |
4704 | cfqd->cfq_slice_idle = 0; | |
142bbdfc | 4705 | wbt_disable_default(q); |
0bb97947 JA |
4706 | } |
4707 | ||
1da177e4 LT |
4708 | /* |
4709 | * sysfs parts below --> | |
4710 | */ | |
1da177e4 LT |
4711 | static ssize_t |
4712 | cfq_var_show(unsigned int var, char *page) | |
4713 | { | |
176167ad | 4714 | return sprintf(page, "%u\n", var); |
1da177e4 LT |
4715 | } |
4716 | ||
4717 | static ssize_t | |
4718 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
4719 | { | |
4720 | char *p = (char *) page; | |
4721 | ||
4722 | *var = simple_strtoul(p, &p, 10); | |
4723 | return count; | |
4724 | } | |
4725 | ||
1da177e4 | 4726 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 4727 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 4728 | { \ |
3d1ab40f | 4729 | struct cfq_data *cfqd = e->elevator_data; \ |
9a7f38c4 | 4730 | u64 __data = __VAR; \ |
1da177e4 | 4731 | if (__CONV) \ |
9a7f38c4 | 4732 | __data = div_u64(__data, NSEC_PER_MSEC); \ |
1da177e4 LT |
4733 | return cfq_var_show(__data, (page)); \ |
4734 | } | |
4735 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
4736 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
4737 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
4738 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
4739 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 4740 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 4741 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
4742 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4743 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4744 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4745 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
5bf14c07 | 4746 | SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1); |
1da177e4 LT |
4747 | #undef SHOW_FUNCTION |
4748 | ||
d2d481d0 JM |
4749 | #define USEC_SHOW_FUNCTION(__FUNC, __VAR) \ |
4750 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ | |
4751 | { \ | |
4752 | struct cfq_data *cfqd = e->elevator_data; \ | |
4753 | u64 __data = __VAR; \ | |
4754 | __data = div_u64(__data, NSEC_PER_USEC); \ | |
4755 | return cfq_var_show(__data, (page)); \ | |
4756 | } | |
4757 | USEC_SHOW_FUNCTION(cfq_slice_idle_us_show, cfqd->cfq_slice_idle); | |
4758 | USEC_SHOW_FUNCTION(cfq_group_idle_us_show, cfqd->cfq_group_idle); | |
4759 | USEC_SHOW_FUNCTION(cfq_slice_sync_us_show, cfqd->cfq_slice[1]); | |
4760 | USEC_SHOW_FUNCTION(cfq_slice_async_us_show, cfqd->cfq_slice[0]); | |
4761 | USEC_SHOW_FUNCTION(cfq_target_latency_us_show, cfqd->cfq_target_latency); | |
4762 | #undef USEC_SHOW_FUNCTION | |
4763 | ||
1da177e4 | 4764 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ |
b374d18a | 4765 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4766 | { \ |
3d1ab40f | 4767 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4768 | unsigned int __data; \ |
4769 | int ret = cfq_var_store(&__data, (page), count); \ | |
4770 | if (__data < (MIN)) \ | |
4771 | __data = (MIN); \ | |
4772 | else if (__data > (MAX)) \ | |
4773 | __data = (MAX); \ | |
4774 | if (__CONV) \ | |
9a7f38c4 | 4775 | *(__PTR) = (u64)__data * NSEC_PER_MSEC; \ |
1da177e4 LT |
4776 | else \ |
4777 | *(__PTR) = __data; \ | |
4778 | return ret; \ | |
4779 | } | |
4780 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4781 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4782 | UINT_MAX, 1); | |
4783 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4784 | UINT_MAX, 1); | |
e572ec7e | 4785 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4786 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4787 | UINT_MAX, 0); | |
22e2c507 | 4788 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4789 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4790 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4791 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4792 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4793 | UINT_MAX, 0); | |
963b72fc | 4794 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
5bf14c07 | 4795 | STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1); |
1da177e4 LT |
4796 | #undef STORE_FUNCTION |
4797 | ||
d2d481d0 JM |
4798 | #define USEC_STORE_FUNCTION(__FUNC, __PTR, MIN, MAX) \ |
4799 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ | |
4800 | { \ | |
4801 | struct cfq_data *cfqd = e->elevator_data; \ | |
4802 | unsigned int __data; \ | |
4803 | int ret = cfq_var_store(&__data, (page), count); \ | |
4804 | if (__data < (MIN)) \ | |
4805 | __data = (MIN); \ | |
4806 | else if (__data > (MAX)) \ | |
4807 | __data = (MAX); \ | |
4808 | *(__PTR) = (u64)__data * NSEC_PER_USEC; \ | |
4809 | return ret; \ | |
4810 | } | |
4811 | USEC_STORE_FUNCTION(cfq_slice_idle_us_store, &cfqd->cfq_slice_idle, 0, UINT_MAX); | |
4812 | USEC_STORE_FUNCTION(cfq_group_idle_us_store, &cfqd->cfq_group_idle, 0, UINT_MAX); | |
4813 | USEC_STORE_FUNCTION(cfq_slice_sync_us_store, &cfqd->cfq_slice[1], 1, UINT_MAX); | |
4814 | USEC_STORE_FUNCTION(cfq_slice_async_us_store, &cfqd->cfq_slice[0], 1, UINT_MAX); | |
4815 | USEC_STORE_FUNCTION(cfq_target_latency_us_store, &cfqd->cfq_target_latency, 1, UINT_MAX); | |
4816 | #undef USEC_STORE_FUNCTION | |
4817 | ||
e572ec7e AV |
4818 | #define CFQ_ATTR(name) \ |
4819 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4820 | ||
4821 | static struct elv_fs_entry cfq_attrs[] = { | |
4822 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4823 | CFQ_ATTR(fifo_expire_sync), |
4824 | CFQ_ATTR(fifo_expire_async), | |
4825 | CFQ_ATTR(back_seek_max), | |
4826 | CFQ_ATTR(back_seek_penalty), | |
4827 | CFQ_ATTR(slice_sync), | |
d2d481d0 | 4828 | CFQ_ATTR(slice_sync_us), |
e572ec7e | 4829 | CFQ_ATTR(slice_async), |
d2d481d0 | 4830 | CFQ_ATTR(slice_async_us), |
e572ec7e AV |
4831 | CFQ_ATTR(slice_async_rq), |
4832 | CFQ_ATTR(slice_idle), | |
d2d481d0 | 4833 | CFQ_ATTR(slice_idle_us), |
80bdf0c7 | 4834 | CFQ_ATTR(group_idle), |
d2d481d0 | 4835 | CFQ_ATTR(group_idle_us), |
963b72fc | 4836 | CFQ_ATTR(low_latency), |
5bf14c07 | 4837 | CFQ_ATTR(target_latency), |
d2d481d0 | 4838 | CFQ_ATTR(target_latency_us), |
e572ec7e | 4839 | __ATTR_NULL |
1da177e4 LT |
4840 | }; |
4841 | ||
1da177e4 | 4842 | static struct elevator_type iosched_cfq = { |
c51ca6cf | 4843 | .ops.sq = { |
1da177e4 LT |
4844 | .elevator_merge_fn = cfq_merge, |
4845 | .elevator_merged_fn = cfq_merged_request, | |
4846 | .elevator_merge_req_fn = cfq_merged_requests, | |
72ef799b TE |
4847 | .elevator_allow_bio_merge_fn = cfq_allow_bio_merge, |
4848 | .elevator_allow_rq_merge_fn = cfq_allow_rq_merge, | |
812d4026 | 4849 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4850 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4851 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4852 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4853 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4854 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4855 | .elevator_former_req_fn = elv_rb_former_request, |
4856 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 4857 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 4858 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
4859 | .elevator_set_req_fn = cfq_set_request, |
4860 | .elevator_put_req_fn = cfq_put_request, | |
4861 | .elevator_may_queue_fn = cfq_may_queue, | |
4862 | .elevator_init_fn = cfq_init_queue, | |
4863 | .elevator_exit_fn = cfq_exit_queue, | |
0bb97947 | 4864 | .elevator_registered_fn = cfq_registered_queue, |
1da177e4 | 4865 | }, |
3d3c2379 TH |
4866 | .icq_size = sizeof(struct cfq_io_cq), |
4867 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 4868 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 4869 | .elevator_name = "cfq", |
1da177e4 LT |
4870 | .elevator_owner = THIS_MODULE, |
4871 | }; | |
4872 | ||
3e252066 | 4873 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4874 | static struct blkcg_policy blkcg_policy_cfq = { |
2ee867dc | 4875 | .dfl_cftypes = cfq_blkcg_files, |
880f50e2 | 4876 | .legacy_cftypes = cfq_blkcg_legacy_files, |
f9fcc2d3 | 4877 | |
e4a9bde9 | 4878 | .cpd_alloc_fn = cfq_cpd_alloc, |
e48453c3 | 4879 | .cpd_init_fn = cfq_cpd_init, |
e4a9bde9 | 4880 | .cpd_free_fn = cfq_cpd_free, |
69d7fde5 | 4881 | .cpd_bind_fn = cfq_cpd_bind, |
e4a9bde9 | 4882 | |
001bea73 | 4883 | .pd_alloc_fn = cfq_pd_alloc, |
f9fcc2d3 | 4884 | .pd_init_fn = cfq_pd_init, |
0b39920b | 4885 | .pd_offline_fn = cfq_pd_offline, |
001bea73 | 4886 | .pd_free_fn = cfq_pd_free, |
f9fcc2d3 | 4887 | .pd_reset_stats_fn = cfq_pd_reset_stats, |
3e252066 | 4888 | }; |
3e252066 VG |
4889 | #endif |
4890 | ||
1da177e4 LT |
4891 | static int __init cfq_init(void) |
4892 | { | |
3d3c2379 TH |
4893 | int ret; |
4894 | ||
80bdf0c7 | 4895 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4896 | ret = blkcg_policy_register(&blkcg_policy_cfq); |
8bd435b3 TH |
4897 | if (ret) |
4898 | return ret; | |
ffea73fc TH |
4899 | #else |
4900 | cfq_group_idle = 0; | |
4901 | #endif | |
8bd435b3 | 4902 | |
fd794956 | 4903 | ret = -ENOMEM; |
3d3c2379 TH |
4904 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4905 | if (!cfq_pool) | |
8bd435b3 | 4906 | goto err_pol_unreg; |
1da177e4 | 4907 | |
3d3c2379 | 4908 | ret = elv_register(&iosched_cfq); |
8bd435b3 TH |
4909 | if (ret) |
4910 | goto err_free_pool; | |
3d3c2379 | 4911 | |
2fdd82bd | 4912 | return 0; |
8bd435b3 TH |
4913 | |
4914 | err_free_pool: | |
4915 | kmem_cache_destroy(cfq_pool); | |
4916 | err_pol_unreg: | |
ffea73fc | 4917 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4918 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4919 | #endif |
8bd435b3 | 4920 | return ret; |
1da177e4 LT |
4921 | } |
4922 | ||
4923 | static void __exit cfq_exit(void) | |
4924 | { | |
ffea73fc | 4925 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4926 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4927 | #endif |
1da177e4 | 4928 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4929 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4930 | } |
4931 | ||
4932 | module_init(cfq_init); | |
4933 | module_exit(cfq_exit); | |
4934 | ||
4935 | MODULE_AUTHOR("Jens Axboe"); | |
4936 | MODULE_LICENSE("GPL"); | |
4937 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |