Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * CFQ, or complete fairness queueing, disk scheduler. |
3 | * | |
4 | * Based on ideas from a previously unfinished io | |
5 | * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli. | |
6 | * | |
0fe23479 | 7 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
1da177e4 | 8 | */ |
1da177e4 | 9 | #include <linux/module.h> |
5a0e3ad6 | 10 | #include <linux/slab.h> |
1cc9be68 AV |
11 | #include <linux/blkdev.h> |
12 | #include <linux/elevator.h> | |
ad5ebd2f | 13 | #include <linux/jiffies.h> |
1da177e4 | 14 | #include <linux/rbtree.h> |
22e2c507 | 15 | #include <linux/ioprio.h> |
7b679138 | 16 | #include <linux/blktrace_api.h> |
6e736be7 | 17 | #include "blk.h" |
629ed0b1 | 18 | #include "blk-cgroup.h" |
1da177e4 LT |
19 | |
20 | /* | |
21 | * tunables | |
22 | */ | |
fe094d98 | 23 | /* max queue in one round of service */ |
abc3c744 | 24 | static const int cfq_quantum = 8; |
64100099 | 25 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
26 | /* maximum backwards seek, in KiB */ |
27 | static const int cfq_back_max = 16 * 1024; | |
28 | /* penalty of a backwards seek */ | |
29 | static const int cfq_back_penalty = 2; | |
64100099 | 30 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 31 | static int cfq_slice_async = HZ / 25; |
64100099 | 32 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 33 | static int cfq_slice_idle = HZ / 125; |
3c2a0909 | 34 | static int cfq_rt_idle_only = 1; |
80bdf0c7 | 35 | static int cfq_group_idle = HZ / 125; |
5db5d642 CZ |
36 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
37 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 38 | |
d9e7620e | 39 | /* |
0871714e | 40 | * offset from end of service tree |
d9e7620e | 41 | */ |
0871714e | 42 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
43 | |
44 | /* | |
45 | * below this threshold, we consider thinktime immediate | |
46 | */ | |
47 | #define CFQ_MIN_TT (2) | |
48 | ||
22e2c507 | 49 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 50 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 51 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 52 | |
3dde36dd | 53 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 54 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 55 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 56 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 57 | |
a612fddf TH |
58 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
59 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
60 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 61 | |
e18b890b | 62 | static struct kmem_cache *cfq_pool; |
1da177e4 | 63 | |
22e2c507 JA |
64 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
65 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
66 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
67 | ||
206dc69b | 68 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 69 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 70 | |
c5869807 TH |
71 | struct cfq_ttime { |
72 | unsigned long last_end_request; | |
73 | ||
74 | unsigned long ttime_total; | |
75 | unsigned long ttime_samples; | |
76 | unsigned long ttime_mean; | |
77 | }; | |
78 | ||
cc09e299 JA |
79 | /* |
80 | * Most of our rbtree usage is for sorting with min extraction, so | |
81 | * if we cache the leftmost node we don't have to walk down the tree | |
82 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
83 | * move this into the elevator for the rq sorting as well. | |
84 | */ | |
85 | struct cfq_rb_root { | |
86 | struct rb_root rb; | |
87 | struct rb_node *left; | |
aa6f6a3d | 88 | unsigned count; |
1fa8f6d6 | 89 | u64 min_vdisktime; |
f5f2b6ce | 90 | struct cfq_ttime ttime; |
cc09e299 | 91 | }; |
f5f2b6ce SL |
92 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
93 | .ttime = {.last_end_request = jiffies,},} | |
cc09e299 | 94 | |
6118b70b JA |
95 | /* |
96 | * Per process-grouping structure | |
97 | */ | |
98 | struct cfq_queue { | |
99 | /* reference count */ | |
30d7b944 | 100 | int ref; |
6118b70b JA |
101 | /* various state flags, see below */ |
102 | unsigned int flags; | |
103 | /* parent cfq_data */ | |
104 | struct cfq_data *cfqd; | |
105 | /* service_tree member */ | |
106 | struct rb_node rb_node; | |
107 | /* service_tree key */ | |
108 | unsigned long rb_key; | |
109 | /* prio tree member */ | |
110 | struct rb_node p_node; | |
111 | /* prio tree root we belong to, if any */ | |
112 | struct rb_root *p_root; | |
113 | /* sorted list of pending requests */ | |
114 | struct rb_root sort_list; | |
115 | /* if fifo isn't expired, next request to serve */ | |
116 | struct request *next_rq; | |
117 | /* requests queued in sort_list */ | |
118 | int queued[2]; | |
119 | /* currently allocated requests */ | |
120 | int allocated[2]; | |
121 | /* fifo list of requests in sort_list */ | |
122 | struct list_head fifo; | |
123 | ||
dae739eb VG |
124 | /* time when queue got scheduled in to dispatch first request. */ |
125 | unsigned long dispatch_start; | |
f75edf2d | 126 | unsigned int allocated_slice; |
c4081ba5 | 127 | unsigned int slice_dispatch; |
dae739eb VG |
128 | /* time when first request from queue completed and slice started. */ |
129 | unsigned long slice_start; | |
6118b70b JA |
130 | unsigned long slice_end; |
131 | long slice_resid; | |
6118b70b | 132 | |
65299a3b CH |
133 | /* pending priority requests */ |
134 | int prio_pending; | |
6118b70b JA |
135 | /* number of requests that are on the dispatch list or inside driver */ |
136 | int dispatched; | |
137 | ||
138 | /* io prio of this group */ | |
139 | unsigned short ioprio, org_ioprio; | |
4aede84b | 140 | unsigned short ioprio_class; |
6118b70b | 141 | |
c4081ba5 RK |
142 | pid_t pid; |
143 | ||
3dde36dd | 144 | u32 seek_history; |
b2c18e1e JM |
145 | sector_t last_request_pos; |
146 | ||
aa6f6a3d | 147 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 148 | struct cfq_queue *new_cfqq; |
cdb16e8f | 149 | struct cfq_group *cfqg; |
c4e7893e VG |
150 | /* Number of sectors dispatched from queue in single dispatch round */ |
151 | unsigned long nr_sectors; | |
6118b70b JA |
152 | }; |
153 | ||
c0324a02 | 154 | /* |
718eee05 | 155 | * First index in the service_trees. |
c0324a02 CZ |
156 | * IDLE is handled separately, so it has negative index |
157 | */ | |
3bf10fea | 158 | enum wl_class_t { |
c0324a02 | 159 | BE_WORKLOAD = 0, |
615f0259 VG |
160 | RT_WORKLOAD = 1, |
161 | IDLE_WORKLOAD = 2, | |
b4627321 | 162 | CFQ_PRIO_NR, |
c0324a02 CZ |
163 | }; |
164 | ||
718eee05 CZ |
165 | /* |
166 | * Second index in the service_trees. | |
167 | */ | |
168 | enum wl_type_t { | |
169 | ASYNC_WORKLOAD = 0, | |
170 | SYNC_NOIDLE_WORKLOAD = 1, | |
171 | SYNC_WORKLOAD = 2 | |
172 | }; | |
173 | ||
155fead9 TH |
174 | struct cfqg_stats { |
175 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
176 | /* total bytes transferred */ | |
177 | struct blkg_rwstat service_bytes; | |
178 | /* total IOs serviced, post merge */ | |
179 | struct blkg_rwstat serviced; | |
180 | /* number of ios merged */ | |
181 | struct blkg_rwstat merged; | |
182 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
183 | struct blkg_rwstat service_time; | |
184 | /* total time spent waiting in scheduler queue in ns */ | |
185 | struct blkg_rwstat wait_time; | |
186 | /* number of IOs queued up */ | |
187 | struct blkg_rwstat queued; | |
188 | /* total sectors transferred */ | |
189 | struct blkg_stat sectors; | |
190 | /* total disk time and nr sectors dispatched by this group */ | |
191 | struct blkg_stat time; | |
192 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
193 | /* time not charged to this cgroup */ | |
194 | struct blkg_stat unaccounted_time; | |
195 | /* sum of number of ios queued across all samples */ | |
196 | struct blkg_stat avg_queue_size_sum; | |
197 | /* count of samples taken for average */ | |
198 | struct blkg_stat avg_queue_size_samples; | |
199 | /* how many times this group has been removed from service tree */ | |
200 | struct blkg_stat dequeue; | |
201 | /* total time spent waiting for it to be assigned a timeslice. */ | |
202 | struct blkg_stat group_wait_time; | |
3c798398 | 203 | /* time spent idling for this blkcg_gq */ |
155fead9 TH |
204 | struct blkg_stat idle_time; |
205 | /* total time with empty current active q with other requests queued */ | |
206 | struct blkg_stat empty_time; | |
207 | /* fields after this shouldn't be cleared on stat reset */ | |
208 | uint64_t start_group_wait_time; | |
209 | uint64_t start_idle_time; | |
210 | uint64_t start_empty_time; | |
211 | uint16_t flags; | |
212 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
213 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
214 | }; | |
215 | ||
cdb16e8f VG |
216 | /* This is per cgroup per device grouping structure */ |
217 | struct cfq_group { | |
f95a04af TH |
218 | /* must be the first member */ |
219 | struct blkg_policy_data pd; | |
220 | ||
1fa8f6d6 VG |
221 | /* group service_tree member */ |
222 | struct rb_node rb_node; | |
223 | ||
224 | /* group service_tree key */ | |
225 | u64 vdisktime; | |
e71357e1 | 226 | |
7918ffb5 TH |
227 | /* |
228 | * The number of active cfqgs and sum of their weights under this | |
229 | * cfqg. This covers this cfqg's leaf_weight and all children's | |
230 | * weights, but does not cover weights of further descendants. | |
231 | * | |
232 | * If a cfqg is on the service tree, it's active. An active cfqg | |
233 | * also activates its parent and contributes to the children_weight | |
234 | * of the parent. | |
235 | */ | |
236 | int nr_active; | |
237 | unsigned int children_weight; | |
238 | ||
1d3650f7 TH |
239 | /* |
240 | * vfraction is the fraction of vdisktime that the tasks in this | |
241 | * cfqg are entitled to. This is determined by compounding the | |
242 | * ratios walking up from this cfqg to the root. | |
243 | * | |
244 | * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all | |
245 | * vfractions on a service tree is approximately 1. The sum may | |
246 | * deviate a bit due to rounding errors and fluctuations caused by | |
247 | * cfqgs entering and leaving the service tree. | |
248 | */ | |
249 | unsigned int vfraction; | |
250 | ||
e71357e1 TH |
251 | /* |
252 | * There are two weights - (internal) weight is the weight of this | |
253 | * cfqg against the sibling cfqgs. leaf_weight is the wight of | |
254 | * this cfqg against the child cfqgs. For the root cfqg, both | |
255 | * weights are kept in sync for backward compatibility. | |
256 | */ | |
25bc6b07 | 257 | unsigned int weight; |
8184f93e | 258 | unsigned int new_weight; |
3381cb8d | 259 | unsigned int dev_weight; |
1fa8f6d6 | 260 | |
e71357e1 TH |
261 | unsigned int leaf_weight; |
262 | unsigned int new_leaf_weight; | |
263 | unsigned int dev_leaf_weight; | |
264 | ||
1fa8f6d6 VG |
265 | /* number of cfqq currently on this group */ |
266 | int nr_cfqq; | |
267 | ||
cdb16e8f | 268 | /* |
4495a7d4 | 269 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
270 | * create the array for each prio class but at run time it is used |
271 | * only for RT and BE class and slot for IDLE class remains unused. | |
272 | * This is primarily done to avoid confusion and a gcc warning. | |
273 | */ | |
274 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
275 | /* | |
276 | * rr lists of queues with requests. We maintain service trees for | |
277 | * RT and BE classes. These trees are subdivided in subclasses | |
278 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
279 | * class there is no subclassification and all the cfq queues go on | |
280 | * a single tree service_tree_idle. | |
cdb16e8f VG |
281 | * Counts are embedded in the cfq_rb_root |
282 | */ | |
283 | struct cfq_rb_root service_trees[2][3]; | |
284 | struct cfq_rb_root service_tree_idle; | |
dae739eb | 285 | |
4d2ceea4 VG |
286 | unsigned long saved_wl_slice; |
287 | enum wl_type_t saved_wl_type; | |
288 | enum wl_class_t saved_wl_class; | |
4eef3049 | 289 | |
80bdf0c7 VG |
290 | /* number of requests that are on the dispatch list or inside driver */ |
291 | int dispatched; | |
7700fc4f | 292 | struct cfq_ttime ttime; |
0b39920b TH |
293 | struct cfqg_stats stats; /* stats for this cfqg */ |
294 | struct cfqg_stats dead_stats; /* stats pushed from dead children */ | |
cdb16e8f | 295 | }; |
718eee05 | 296 | |
c5869807 TH |
297 | struct cfq_io_cq { |
298 | struct io_cq icq; /* must be the first member */ | |
299 | struct cfq_queue *cfqq[2]; | |
300 | struct cfq_ttime ttime; | |
598971bf TH |
301 | int ioprio; /* the current ioprio */ |
302 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
303 | uint64_t blkcg_id; /* the current blkcg ID */ | |
304 | #endif | |
c5869807 TH |
305 | }; |
306 | ||
22e2c507 JA |
307 | /* |
308 | * Per block device queue structure | |
309 | */ | |
1da177e4 | 310 | struct cfq_data { |
165125e1 | 311 | struct request_queue *queue; |
1fa8f6d6 VG |
312 | /* Root service tree for cfq_groups */ |
313 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 314 | struct cfq_group *root_group; |
22e2c507 | 315 | |
c0324a02 CZ |
316 | /* |
317 | * The priority currently being served | |
22e2c507 | 318 | */ |
4d2ceea4 VG |
319 | enum wl_class_t serving_wl_class; |
320 | enum wl_type_t serving_wl_type; | |
718eee05 | 321 | unsigned long workload_expires; |
cdb16e8f | 322 | struct cfq_group *serving_group; |
a36e71f9 JA |
323 | |
324 | /* | |
325 | * Each priority tree is sorted by next_request position. These | |
326 | * trees are used when determining if two or more queues are | |
327 | * interleaving requests (see cfq_close_cooperator). | |
328 | */ | |
329 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
330 | ||
22e2c507 | 331 | unsigned int busy_queues; |
ef8a41df | 332 | unsigned int busy_sync_queues; |
22e2c507 | 333 | |
53c583d2 CZ |
334 | int rq_in_driver; |
335 | int rq_in_flight[2]; | |
45333d5a AC |
336 | |
337 | /* | |
338 | * queue-depth detection | |
339 | */ | |
340 | int rq_queued; | |
25776e35 | 341 | int hw_tag; |
e459dd08 CZ |
342 | /* |
343 | * hw_tag can be | |
344 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
345 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
346 | * 0 => no NCQ | |
347 | */ | |
348 | int hw_tag_est_depth; | |
349 | unsigned int hw_tag_samples; | |
1da177e4 | 350 | |
22e2c507 JA |
351 | /* |
352 | * idle window management | |
353 | */ | |
354 | struct timer_list idle_slice_timer; | |
23e018a1 | 355 | struct work_struct unplug_work; |
1da177e4 | 356 | |
22e2c507 | 357 | struct cfq_queue *active_queue; |
c5869807 | 358 | struct cfq_io_cq *active_cic; |
22e2c507 | 359 | |
c2dea2d1 VT |
360 | /* |
361 | * async queue for each priority case | |
362 | */ | |
363 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
364 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 365 | |
6d048f53 | 366 | sector_t last_position; |
1da177e4 | 367 | |
1da177e4 LT |
368 | /* |
369 | * tunables, see top of file | |
370 | */ | |
371 | unsigned int cfq_quantum; | |
22e2c507 | 372 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
373 | unsigned int cfq_back_penalty; |
374 | unsigned int cfq_back_max; | |
22e2c507 JA |
375 | unsigned int cfq_slice[2]; |
376 | unsigned int cfq_slice_async_rq; | |
377 | unsigned int cfq_slice_idle; | |
3c2a0909 | 378 | unsigned int cfq_rt_idle_only; |
80bdf0c7 | 379 | unsigned int cfq_group_idle; |
963b72fc | 380 | unsigned int cfq_latency; |
5bf14c07 | 381 | unsigned int cfq_target_latency; |
d9ff4187 | 382 | |
6118b70b JA |
383 | /* |
384 | * Fallback dummy cfqq for extreme OOM conditions | |
385 | */ | |
386 | struct cfq_queue oom_cfqq; | |
365722bb | 387 | |
573412b2 | 388 | unsigned long last_delayed_sync; |
1da177e4 LT |
389 | }; |
390 | ||
25fb5169 VG |
391 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
392 | ||
34b98d03 | 393 | static struct cfq_rb_root *st_for(struct cfq_group *cfqg, |
3bf10fea | 394 | enum wl_class_t class, |
65b32a57 | 395 | enum wl_type_t type) |
c0324a02 | 396 | { |
1fa8f6d6 VG |
397 | if (!cfqg) |
398 | return NULL; | |
399 | ||
3bf10fea | 400 | if (class == IDLE_WORKLOAD) |
cdb16e8f | 401 | return &cfqg->service_tree_idle; |
c0324a02 | 402 | |
3bf10fea | 403 | return &cfqg->service_trees[class][type]; |
c0324a02 CZ |
404 | } |
405 | ||
3b18152c | 406 | enum cfqq_state_flags { |
b0b8d749 JA |
407 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
408 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 409 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 410 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
411 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
412 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
413 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 414 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 415 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 416 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 417 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 418 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 419 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
420 | }; |
421 | ||
422 | #define CFQ_CFQQ_FNS(name) \ | |
423 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
424 | { \ | |
fe094d98 | 425 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
426 | } \ |
427 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
428 | { \ | |
fe094d98 | 429 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
430 | } \ |
431 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
432 | { \ | |
fe094d98 | 433 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
434 | } |
435 | ||
436 | CFQ_CFQQ_FNS(on_rr); | |
437 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 438 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 439 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
440 | CFQ_CFQQ_FNS(fifo_expire); |
441 | CFQ_CFQQ_FNS(idle_window); | |
442 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 443 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 444 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 445 | CFQ_CFQQ_FNS(coop); |
ae54abed | 446 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 447 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 448 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
449 | #undef CFQ_CFQQ_FNS |
450 | ||
f95a04af TH |
451 | static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd) |
452 | { | |
453 | return pd ? container_of(pd, struct cfq_group, pd) : NULL; | |
454 | } | |
455 | ||
f95a04af TH |
456 | static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg) |
457 | { | |
458 | return pd_to_blkg(&cfqg->pd); | |
459 | } | |
460 | ||
629ed0b1 | 461 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
2ce4d50f | 462 | |
155fead9 TH |
463 | /* cfqg stats flags */ |
464 | enum cfqg_stats_flags { | |
465 | CFQG_stats_waiting = 0, | |
466 | CFQG_stats_idling, | |
467 | CFQG_stats_empty, | |
629ed0b1 TH |
468 | }; |
469 | ||
155fead9 TH |
470 | #define CFQG_FLAG_FNS(name) \ |
471 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ | |
629ed0b1 | 472 | { \ |
155fead9 | 473 | stats->flags |= (1 << CFQG_stats_##name); \ |
629ed0b1 | 474 | } \ |
155fead9 | 475 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 476 | { \ |
155fead9 | 477 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
629ed0b1 | 478 | } \ |
155fead9 | 479 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
629ed0b1 | 480 | { \ |
155fead9 | 481 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
629ed0b1 TH |
482 | } \ |
483 | ||
155fead9 TH |
484 | CFQG_FLAG_FNS(waiting) |
485 | CFQG_FLAG_FNS(idling) | |
486 | CFQG_FLAG_FNS(empty) | |
487 | #undef CFQG_FLAG_FNS | |
629ed0b1 TH |
488 | |
489 | /* This should be called with the queue_lock held. */ | |
155fead9 | 490 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
629ed0b1 TH |
491 | { |
492 | unsigned long long now; | |
493 | ||
155fead9 | 494 | if (!cfqg_stats_waiting(stats)) |
629ed0b1 TH |
495 | return; |
496 | ||
497 | now = sched_clock(); | |
498 | if (time_after64(now, stats->start_group_wait_time)) | |
499 | blkg_stat_add(&stats->group_wait_time, | |
500 | now - stats->start_group_wait_time); | |
155fead9 | 501 | cfqg_stats_clear_waiting(stats); |
629ed0b1 TH |
502 | } |
503 | ||
504 | /* This should be called with the queue_lock held. */ | |
155fead9 TH |
505 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
506 | struct cfq_group *curr_cfqg) | |
629ed0b1 | 507 | { |
155fead9 | 508 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 509 | |
155fead9 | 510 | if (cfqg_stats_waiting(stats)) |
629ed0b1 | 511 | return; |
155fead9 | 512 | if (cfqg == curr_cfqg) |
629ed0b1 | 513 | return; |
155fead9 TH |
514 | stats->start_group_wait_time = sched_clock(); |
515 | cfqg_stats_mark_waiting(stats); | |
629ed0b1 TH |
516 | } |
517 | ||
518 | /* This should be called with the queue_lock held. */ | |
155fead9 | 519 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
629ed0b1 TH |
520 | { |
521 | unsigned long long now; | |
522 | ||
155fead9 | 523 | if (!cfqg_stats_empty(stats)) |
629ed0b1 TH |
524 | return; |
525 | ||
526 | now = sched_clock(); | |
527 | if (time_after64(now, stats->start_empty_time)) | |
528 | blkg_stat_add(&stats->empty_time, | |
529 | now - stats->start_empty_time); | |
155fead9 | 530 | cfqg_stats_clear_empty(stats); |
629ed0b1 TH |
531 | } |
532 | ||
155fead9 | 533 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
629ed0b1 | 534 | { |
155fead9 | 535 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
629ed0b1 TH |
536 | } |
537 | ||
155fead9 | 538 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
629ed0b1 | 539 | { |
155fead9 | 540 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 541 | |
4d5e80a7 | 542 | if (blkg_rwstat_total(&stats->queued)) |
629ed0b1 TH |
543 | return; |
544 | ||
545 | /* | |
546 | * group is already marked empty. This can happen if cfqq got new | |
547 | * request in parent group and moved to this group while being added | |
548 | * to service tree. Just ignore the event and move on. | |
549 | */ | |
155fead9 | 550 | if (cfqg_stats_empty(stats)) |
629ed0b1 TH |
551 | return; |
552 | ||
553 | stats->start_empty_time = sched_clock(); | |
155fead9 | 554 | cfqg_stats_mark_empty(stats); |
629ed0b1 TH |
555 | } |
556 | ||
155fead9 | 557 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 558 | { |
155fead9 | 559 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 560 | |
155fead9 | 561 | if (cfqg_stats_idling(stats)) { |
629ed0b1 TH |
562 | unsigned long long now = sched_clock(); |
563 | ||
564 | if (time_after64(now, stats->start_idle_time)) | |
565 | blkg_stat_add(&stats->idle_time, | |
566 | now - stats->start_idle_time); | |
155fead9 | 567 | cfqg_stats_clear_idling(stats); |
629ed0b1 TH |
568 | } |
569 | } | |
570 | ||
155fead9 | 571 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
629ed0b1 | 572 | { |
155fead9 | 573 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 574 | |
155fead9 | 575 | BUG_ON(cfqg_stats_idling(stats)); |
629ed0b1 TH |
576 | |
577 | stats->start_idle_time = sched_clock(); | |
155fead9 | 578 | cfqg_stats_mark_idling(stats); |
629ed0b1 TH |
579 | } |
580 | ||
155fead9 | 581 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
629ed0b1 | 582 | { |
155fead9 | 583 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 TH |
584 | |
585 | blkg_stat_add(&stats->avg_queue_size_sum, | |
4d5e80a7 | 586 | blkg_rwstat_total(&stats->queued)); |
629ed0b1 | 587 | blkg_stat_add(&stats->avg_queue_size_samples, 1); |
155fead9 | 588 | cfqg_stats_update_group_wait_time(stats); |
629ed0b1 TH |
589 | } |
590 | ||
591 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
592 | ||
f48ec1d7 TH |
593 | static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { } |
594 | static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } | |
595 | static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } | |
596 | static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } | |
597 | static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } | |
598 | static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } | |
599 | static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } | |
629ed0b1 TH |
600 | |
601 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ | |
602 | ||
603 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
2ce4d50f | 604 | |
ffea73fc TH |
605 | static struct blkcg_policy blkcg_policy_cfq; |
606 | ||
607 | static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg) | |
608 | { | |
609 | return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq)); | |
610 | } | |
611 | ||
d02f7aa8 | 612 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) |
7918ffb5 | 613 | { |
d02f7aa8 | 614 | struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent; |
7918ffb5 | 615 | |
d02f7aa8 | 616 | return pblkg ? blkg_to_cfqg(pblkg) : NULL; |
7918ffb5 TH |
617 | } |
618 | ||
eb7d8c07 TH |
619 | static inline void cfqg_get(struct cfq_group *cfqg) |
620 | { | |
621 | return blkg_get(cfqg_to_blkg(cfqg)); | |
622 | } | |
623 | ||
624 | static inline void cfqg_put(struct cfq_group *cfqg) | |
625 | { | |
626 | return blkg_put(cfqg_to_blkg(cfqg)); | |
627 | } | |
628 | ||
54e7ed12 TH |
629 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \ |
630 | char __pbuf[128]; \ | |
631 | \ | |
632 | blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \ | |
b226e5c4 VG |
633 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \ |
634 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
635 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
54e7ed12 TH |
636 | __pbuf, ##args); \ |
637 | } while (0) | |
638 | ||
639 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \ | |
640 | char __pbuf[128]; \ | |
641 | \ | |
642 | blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \ | |
643 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \ | |
644 | } while (0) | |
2868ef7b | 645 | |
155fead9 TH |
646 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
647 | struct cfq_group *curr_cfqg, int rw) | |
2ce4d50f | 648 | { |
155fead9 TH |
649 | blkg_rwstat_add(&cfqg->stats.queued, rw, 1); |
650 | cfqg_stats_end_empty_time(&cfqg->stats); | |
651 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); | |
2ce4d50f TH |
652 | } |
653 | ||
155fead9 TH |
654 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
655 | unsigned long time, unsigned long unaccounted_time) | |
2ce4d50f | 656 | { |
155fead9 | 657 | blkg_stat_add(&cfqg->stats.time, time); |
629ed0b1 | 658 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
155fead9 | 659 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
629ed0b1 | 660 | #endif |
2ce4d50f TH |
661 | } |
662 | ||
155fead9 | 663 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) |
2ce4d50f | 664 | { |
155fead9 | 665 | blkg_rwstat_add(&cfqg->stats.queued, rw, -1); |
2ce4d50f TH |
666 | } |
667 | ||
155fead9 | 668 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) |
2ce4d50f | 669 | { |
155fead9 | 670 | blkg_rwstat_add(&cfqg->stats.merged, rw, 1); |
2ce4d50f TH |
671 | } |
672 | ||
155fead9 TH |
673 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, |
674 | uint64_t bytes, int rw) | |
2ce4d50f | 675 | { |
155fead9 TH |
676 | blkg_stat_add(&cfqg->stats.sectors, bytes >> 9); |
677 | blkg_rwstat_add(&cfqg->stats.serviced, rw, 1); | |
678 | blkg_rwstat_add(&cfqg->stats.service_bytes, rw, bytes); | |
2ce4d50f TH |
679 | } |
680 | ||
155fead9 TH |
681 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
682 | uint64_t start_time, uint64_t io_start_time, int rw) | |
2ce4d50f | 683 | { |
155fead9 | 684 | struct cfqg_stats *stats = &cfqg->stats; |
629ed0b1 | 685 | unsigned long long now = sched_clock(); |
629ed0b1 TH |
686 | |
687 | if (time_after64(now, io_start_time)) | |
688 | blkg_rwstat_add(&stats->service_time, rw, now - io_start_time); | |
689 | if (time_after64(io_start_time, start_time)) | |
690 | blkg_rwstat_add(&stats->wait_time, rw, | |
691 | io_start_time - start_time); | |
2ce4d50f TH |
692 | } |
693 | ||
689665af TH |
694 | /* @stats = 0 */ |
695 | static void cfqg_stats_reset(struct cfqg_stats *stats) | |
155fead9 | 696 | { |
155fead9 TH |
697 | /* queued stats shouldn't be cleared */ |
698 | blkg_rwstat_reset(&stats->service_bytes); | |
699 | blkg_rwstat_reset(&stats->serviced); | |
700 | blkg_rwstat_reset(&stats->merged); | |
701 | blkg_rwstat_reset(&stats->service_time); | |
702 | blkg_rwstat_reset(&stats->wait_time); | |
703 | blkg_stat_reset(&stats->time); | |
704 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
705 | blkg_stat_reset(&stats->unaccounted_time); | |
706 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
707 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
708 | blkg_stat_reset(&stats->dequeue); | |
709 | blkg_stat_reset(&stats->group_wait_time); | |
710 | blkg_stat_reset(&stats->idle_time); | |
711 | blkg_stat_reset(&stats->empty_time); | |
712 | #endif | |
713 | } | |
714 | ||
0b39920b TH |
715 | /* @to += @from */ |
716 | static void cfqg_stats_merge(struct cfqg_stats *to, struct cfqg_stats *from) | |
717 | { | |
718 | /* queued stats shouldn't be cleared */ | |
719 | blkg_rwstat_merge(&to->service_bytes, &from->service_bytes); | |
720 | blkg_rwstat_merge(&to->serviced, &from->serviced); | |
721 | blkg_rwstat_merge(&to->merged, &from->merged); | |
722 | blkg_rwstat_merge(&to->service_time, &from->service_time); | |
723 | blkg_rwstat_merge(&to->wait_time, &from->wait_time); | |
724 | blkg_stat_merge(&from->time, &from->time); | |
725 | #ifdef CONFIG_DEBUG_BLK_CGROUP | |
726 | blkg_stat_merge(&to->unaccounted_time, &from->unaccounted_time); | |
727 | blkg_stat_merge(&to->avg_queue_size_sum, &from->avg_queue_size_sum); | |
728 | blkg_stat_merge(&to->avg_queue_size_samples, &from->avg_queue_size_samples); | |
729 | blkg_stat_merge(&to->dequeue, &from->dequeue); | |
730 | blkg_stat_merge(&to->group_wait_time, &from->group_wait_time); | |
731 | blkg_stat_merge(&to->idle_time, &from->idle_time); | |
732 | blkg_stat_merge(&to->empty_time, &from->empty_time); | |
733 | #endif | |
734 | } | |
735 | ||
736 | /* | |
737 | * Transfer @cfqg's stats to its parent's dead_stats so that the ancestors' | |
738 | * recursive stats can still account for the amount used by this cfqg after | |
739 | * it's gone. | |
740 | */ | |
741 | static void cfqg_stats_xfer_dead(struct cfq_group *cfqg) | |
742 | { | |
743 | struct cfq_group *parent = cfqg_parent(cfqg); | |
744 | ||
745 | lockdep_assert_held(cfqg_to_blkg(cfqg)->q->queue_lock); | |
746 | ||
747 | if (unlikely(!parent)) | |
748 | return; | |
749 | ||
750 | cfqg_stats_merge(&parent->dead_stats, &cfqg->stats); | |
751 | cfqg_stats_merge(&parent->dead_stats, &cfqg->dead_stats); | |
752 | cfqg_stats_reset(&cfqg->stats); | |
753 | cfqg_stats_reset(&cfqg->dead_stats); | |
754 | } | |
755 | ||
eb7d8c07 TH |
756 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
757 | ||
d02f7aa8 | 758 | static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) { return NULL; } |
eb7d8c07 TH |
759 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
760 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
761 | ||
7b679138 | 762 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
b226e5c4 VG |
763 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \ |
764 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
765 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ | |
766 | ##args) | |
4495a7d4 | 767 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 | 768 | |
155fead9 TH |
769 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
770 | struct cfq_group *curr_cfqg, int rw) { } | |
771 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, | |
772 | unsigned long time, unsigned long unaccounted_time) { } | |
773 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) { } | |
774 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) { } | |
775 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, | |
776 | uint64_t bytes, int rw) { } | |
777 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, | |
778 | uint64_t start_time, uint64_t io_start_time, int rw) { } | |
2ce4d50f | 779 | |
eb7d8c07 TH |
780 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
781 | ||
7b679138 JA |
782 | #define cfq_log(cfqd, fmt, args...) \ |
783 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
784 | ||
615f0259 VG |
785 | /* Traverses through cfq group service trees */ |
786 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
787 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
788 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
789 | : &cfqg->service_tree_idle; \ | |
790 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
791 | (i == IDLE_WORKLOAD && j == 0); \ | |
792 | j++, st = i < IDLE_WORKLOAD ? \ | |
793 | &cfqg->service_trees[i][j]: NULL) \ | |
794 | ||
f5f2b6ce SL |
795 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
796 | struct cfq_ttime *ttime, bool group_idle) | |
797 | { | |
798 | unsigned long slice; | |
799 | if (!sample_valid(ttime->ttime_samples)) | |
800 | return false; | |
801 | if (group_idle) | |
802 | slice = cfqd->cfq_group_idle; | |
803 | else | |
804 | slice = cfqd->cfq_slice_idle; | |
805 | return ttime->ttime_mean > slice; | |
806 | } | |
615f0259 | 807 | |
02b35081 VG |
808 | static inline bool iops_mode(struct cfq_data *cfqd) |
809 | { | |
810 | /* | |
811 | * If we are not idling on queues and it is a NCQ drive, parallel | |
812 | * execution of requests is on and measuring time is not possible | |
813 | * in most of the cases until and unless we drive shallower queue | |
814 | * depths and that becomes a performance bottleneck. In such cases | |
815 | * switch to start providing fairness in terms of number of IOs. | |
816 | */ | |
817 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
818 | return true; | |
819 | else | |
820 | return false; | |
821 | } | |
822 | ||
3bf10fea | 823 | static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq) |
c0324a02 CZ |
824 | { |
825 | if (cfq_class_idle(cfqq)) | |
826 | return IDLE_WORKLOAD; | |
827 | if (cfq_class_rt(cfqq)) | |
828 | return RT_WORKLOAD; | |
829 | return BE_WORKLOAD; | |
830 | } | |
831 | ||
718eee05 CZ |
832 | |
833 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
834 | { | |
835 | if (!cfq_cfqq_sync(cfqq)) | |
836 | return ASYNC_WORKLOAD; | |
837 | if (!cfq_cfqq_idle_window(cfqq)) | |
838 | return SYNC_NOIDLE_WORKLOAD; | |
839 | return SYNC_WORKLOAD; | |
840 | } | |
841 | ||
3bf10fea | 842 | static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class, |
58ff82f3 VG |
843 | struct cfq_data *cfqd, |
844 | struct cfq_group *cfqg) | |
c0324a02 | 845 | { |
3bf10fea | 846 | if (wl_class == IDLE_WORKLOAD) |
cdb16e8f | 847 | return cfqg->service_tree_idle.count; |
c0324a02 | 848 | |
34b98d03 VG |
849 | return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count + |
850 | cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count + | |
851 | cfqg->service_trees[wl_class][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
852 | } |
853 | ||
f26bd1f0 VG |
854 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
855 | struct cfq_group *cfqg) | |
856 | { | |
34b98d03 VG |
857 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count + |
858 | cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
f26bd1f0 VG |
859 | } |
860 | ||
165125e1 | 861 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 862 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
abede6da | 863 | struct cfq_io_cq *cic, struct bio *bio, |
4f85cb96 | 864 | gfp_t gfp_mask); |
91fac317 | 865 | |
c5869807 TH |
866 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
867 | { | |
868 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
869 | return container_of(icq, struct cfq_io_cq, icq); | |
870 | } | |
871 | ||
47fdd4ca TH |
872 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
873 | struct io_context *ioc) | |
874 | { | |
875 | if (ioc) | |
876 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
877 | return NULL; | |
878 | } | |
879 | ||
c5869807 | 880 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 881 | { |
a6151c3a | 882 | return cic->cfqq[is_sync]; |
91fac317 VT |
883 | } |
884 | ||
c5869807 TH |
885 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
886 | bool is_sync) | |
91fac317 | 887 | { |
a6151c3a | 888 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
889 | } |
890 | ||
c5869807 | 891 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 892 | { |
c5869807 | 893 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
894 | } |
895 | ||
91fac317 VT |
896 | /* |
897 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
898 | * set (in which case it could also be direct WRITE). | |
899 | */ | |
a6151c3a | 900 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 901 | { |
7b6d91da | 902 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 903 | } |
1da177e4 | 904 | |
99f95e52 AM |
905 | /* |
906 | * scheduler run of queue, if there are requests pending and no one in the | |
907 | * driver that will restart queueing | |
908 | */ | |
23e018a1 | 909 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 910 | { |
7b679138 JA |
911 | if (cfqd->busy_queues) { |
912 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 913 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 914 | } |
99f95e52 AM |
915 | } |
916 | ||
44f7c160 JA |
917 | /* |
918 | * Scale schedule slice based on io priority. Use the sync time slice only | |
919 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
920 | * io only, should not get full sync slice length. | |
921 | */ | |
a6151c3a | 922 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 923 | unsigned short prio) |
44f7c160 | 924 | { |
d9e7620e | 925 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 926 | |
d9e7620e JA |
927 | WARN_ON(prio >= IOPRIO_BE_NR); |
928 | ||
929 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
930 | } | |
44f7c160 | 931 | |
d9e7620e JA |
932 | static inline int |
933 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
934 | { | |
935 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
936 | } |
937 | ||
1d3650f7 TH |
938 | /** |
939 | * cfqg_scale_charge - scale disk time charge according to cfqg weight | |
940 | * @charge: disk time being charged | |
941 | * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT | |
942 | * | |
943 | * Scale @charge according to @vfraction, which is in range (0, 1]. The | |
944 | * scaling is inversely proportional. | |
945 | * | |
946 | * scaled = charge / vfraction | |
947 | * | |
948 | * The result is also in fixed point w/ CFQ_SERVICE_SHIFT. | |
949 | */ | |
950 | static inline u64 cfqg_scale_charge(unsigned long charge, | |
951 | unsigned int vfraction) | |
25bc6b07 | 952 | { |
1d3650f7 | 953 | u64 c = charge << CFQ_SERVICE_SHIFT; /* make it fixed point */ |
25bc6b07 | 954 | |
1d3650f7 TH |
955 | /* charge / vfraction */ |
956 | c <<= CFQ_SERVICE_SHIFT; | |
957 | do_div(c, vfraction); | |
958 | return c; | |
25bc6b07 VG |
959 | } |
960 | ||
961 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
962 | { | |
963 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
964 | if (delta > 0) | |
965 | min_vdisktime = vdisktime; | |
966 | ||
967 | return min_vdisktime; | |
968 | } | |
969 | ||
970 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
971 | { | |
972 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
973 | if (delta < 0) | |
974 | min_vdisktime = vdisktime; | |
975 | ||
976 | return min_vdisktime; | |
977 | } | |
978 | ||
979 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
980 | { | |
25bc6b07 VG |
981 | struct cfq_group *cfqg; |
982 | ||
25bc6b07 VG |
983 | if (st->left) { |
984 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
985 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
986 | cfqg->vdisktime); | |
25bc6b07 | 987 | } |
25bc6b07 VG |
988 | } |
989 | ||
5db5d642 CZ |
990 | /* |
991 | * get averaged number of queues of RT/BE priority. | |
992 | * average is updated, with a formula that gives more weight to higher numbers, | |
993 | * to quickly follows sudden increases and decrease slowly | |
994 | */ | |
995 | ||
58ff82f3 VG |
996 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
997 | struct cfq_group *cfqg, bool rt) | |
5869619c | 998 | { |
5db5d642 CZ |
999 | unsigned min_q, max_q; |
1000 | unsigned mult = cfq_hist_divisor - 1; | |
1001 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 1002 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 1003 | |
58ff82f3 VG |
1004 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
1005 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
1006 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 1007 | cfq_hist_divisor; |
58ff82f3 VG |
1008 | return cfqg->busy_queues_avg[rt]; |
1009 | } | |
1010 | ||
1011 | static inline unsigned | |
1012 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1013 | { | |
41cad6ab | 1014 | return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT; |
5db5d642 CZ |
1015 | } |
1016 | ||
c553f8e3 | 1017 | static inline unsigned |
ba5bd520 | 1018 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 1019 | { |
5db5d642 CZ |
1020 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
1021 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
1022 | /* |
1023 | * interested queues (we consider only the ones with the same | |
1024 | * priority class in the cfq group) | |
1025 | */ | |
1026 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
1027 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
1028 | unsigned sync_slice = cfqd->cfq_slice[1]; |
1029 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
1030 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
1031 | ||
1032 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
1033 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
1034 | /* scale low_slice according to IO priority | |
1035 | * and sync vs async */ | |
1036 | unsigned low_slice = | |
1037 | min(slice, base_low_slice * slice / sync_slice); | |
1038 | /* the adapted slice value is scaled to fit all iqs | |
1039 | * into the target latency */ | |
58ff82f3 | 1040 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
1041 | low_slice); |
1042 | } | |
1043 | } | |
c553f8e3 SL |
1044 | return slice; |
1045 | } | |
1046 | ||
1047 | static inline void | |
1048 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1049 | { | |
ba5bd520 | 1050 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 1051 | |
dae739eb | 1052 | cfqq->slice_start = jiffies; |
5db5d642 | 1053 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 1054 | cfqq->allocated_slice = slice; |
7b679138 | 1055 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
1056 | } |
1057 | ||
1058 | /* | |
1059 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
1060 | * isn't valid until the first request from the dispatch is activated | |
1061 | * and the slice time set. | |
1062 | */ | |
a6151c3a | 1063 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
1064 | { |
1065 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 1066 | return false; |
44f7c160 | 1067 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 1068 | return false; |
44f7c160 | 1069 | |
c1e44756 | 1070 | return true; |
44f7c160 JA |
1071 | } |
1072 | ||
1da177e4 | 1073 | /* |
5e705374 | 1074 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 1075 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 1076 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 1077 | */ |
5e705374 | 1078 | static struct request * |
cf7c25cf | 1079 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 1080 | { |
cf7c25cf | 1081 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 1082 | unsigned long back_max; |
e8a99053 AM |
1083 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
1084 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
1085 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 1086 | |
5e705374 JA |
1087 | if (rq1 == NULL || rq1 == rq2) |
1088 | return rq2; | |
1089 | if (rq2 == NULL) | |
1090 | return rq1; | |
9c2c38a1 | 1091 | |
229836bd NK |
1092 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
1093 | return rq_is_sync(rq1) ? rq1 : rq2; | |
1094 | ||
65299a3b CH |
1095 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
1096 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 1097 | |
83096ebf TH |
1098 | s1 = blk_rq_pos(rq1); |
1099 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 1100 | |
1da177e4 LT |
1101 | /* |
1102 | * by definition, 1KiB is 2 sectors | |
1103 | */ | |
1104 | back_max = cfqd->cfq_back_max * 2; | |
1105 | ||
1106 | /* | |
1107 | * Strict one way elevator _except_ in the case where we allow | |
1108 | * short backward seeks which are biased as twice the cost of a | |
1109 | * similar forward seek. | |
1110 | */ | |
1111 | if (s1 >= last) | |
1112 | d1 = s1 - last; | |
1113 | else if (s1 + back_max >= last) | |
1114 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
1115 | else | |
e8a99053 | 1116 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
1117 | |
1118 | if (s2 >= last) | |
1119 | d2 = s2 - last; | |
1120 | else if (s2 + back_max >= last) | |
1121 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
1122 | else | |
e8a99053 | 1123 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
1124 | |
1125 | /* Found required data */ | |
e8a99053 AM |
1126 | |
1127 | /* | |
1128 | * By doing switch() on the bit mask "wrap" we avoid having to | |
1129 | * check two variables for all permutations: --> faster! | |
1130 | */ | |
1131 | switch (wrap) { | |
5e705374 | 1132 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 1133 | if (d1 < d2) |
5e705374 | 1134 | return rq1; |
e8a99053 | 1135 | else if (d2 < d1) |
5e705374 | 1136 | return rq2; |
e8a99053 AM |
1137 | else { |
1138 | if (s1 >= s2) | |
5e705374 | 1139 | return rq1; |
e8a99053 | 1140 | else |
5e705374 | 1141 | return rq2; |
e8a99053 | 1142 | } |
1da177e4 | 1143 | |
e8a99053 | 1144 | case CFQ_RQ2_WRAP: |
5e705374 | 1145 | return rq1; |
e8a99053 | 1146 | case CFQ_RQ1_WRAP: |
5e705374 JA |
1147 | return rq2; |
1148 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
1149 | default: |
1150 | /* | |
1151 | * Since both rqs are wrapped, | |
1152 | * start with the one that's further behind head | |
1153 | * (--> only *one* back seek required), | |
1154 | * since back seek takes more time than forward. | |
1155 | */ | |
1156 | if (s1 <= s2) | |
5e705374 | 1157 | return rq1; |
1da177e4 | 1158 | else |
5e705374 | 1159 | return rq2; |
1da177e4 LT |
1160 | } |
1161 | } | |
1162 | ||
498d3aa2 JA |
1163 | /* |
1164 | * The below is leftmost cache rbtree addon | |
1165 | */ | |
0871714e | 1166 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 1167 | { |
615f0259 VG |
1168 | /* Service tree is empty */ |
1169 | if (!root->count) | |
1170 | return NULL; | |
1171 | ||
cc09e299 JA |
1172 | if (!root->left) |
1173 | root->left = rb_first(&root->rb); | |
1174 | ||
0871714e JA |
1175 | if (root->left) |
1176 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
1177 | ||
1178 | return NULL; | |
cc09e299 JA |
1179 | } |
1180 | ||
1fa8f6d6 VG |
1181 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
1182 | { | |
1183 | if (!root->left) | |
1184 | root->left = rb_first(&root->rb); | |
1185 | ||
1186 | if (root->left) | |
1187 | return rb_entry_cfqg(root->left); | |
1188 | ||
1189 | return NULL; | |
1190 | } | |
1191 | ||
a36e71f9 JA |
1192 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
1193 | { | |
1194 | rb_erase(n, root); | |
1195 | RB_CLEAR_NODE(n); | |
1196 | } | |
1197 | ||
cc09e299 JA |
1198 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
1199 | { | |
1200 | if (root->left == n) | |
1201 | root->left = NULL; | |
a36e71f9 | 1202 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 1203 | --root->count; |
cc09e299 JA |
1204 | } |
1205 | ||
1da177e4 LT |
1206 | /* |
1207 | * would be nice to take fifo expire time into account as well | |
1208 | */ | |
5e705374 JA |
1209 | static struct request * |
1210 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
1211 | struct request *last) | |
1da177e4 | 1212 | { |
21183b07 JA |
1213 | struct rb_node *rbnext = rb_next(&last->rb_node); |
1214 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 1215 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 1216 | |
21183b07 | 1217 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
1218 | |
1219 | if (rbprev) | |
5e705374 | 1220 | prev = rb_entry_rq(rbprev); |
1da177e4 | 1221 | |
21183b07 | 1222 | if (rbnext) |
5e705374 | 1223 | next = rb_entry_rq(rbnext); |
21183b07 JA |
1224 | else { |
1225 | rbnext = rb_first(&cfqq->sort_list); | |
1226 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 1227 | next = rb_entry_rq(rbnext); |
21183b07 | 1228 | } |
1da177e4 | 1229 | |
cf7c25cf | 1230 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
1231 | } |
1232 | ||
d9e7620e JA |
1233 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
1234 | struct cfq_queue *cfqq) | |
1da177e4 | 1235 | { |
d9e7620e JA |
1236 | /* |
1237 | * just an approximation, should be ok. | |
1238 | */ | |
cdb16e8f | 1239 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 1240 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
1241 | } |
1242 | ||
1fa8f6d6 VG |
1243 | static inline s64 |
1244 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1245 | { | |
1246 | return cfqg->vdisktime - st->min_vdisktime; | |
1247 | } | |
1248 | ||
1249 | static void | |
1250 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1251 | { | |
1252 | struct rb_node **node = &st->rb.rb_node; | |
1253 | struct rb_node *parent = NULL; | |
1254 | struct cfq_group *__cfqg; | |
1255 | s64 key = cfqg_key(st, cfqg); | |
1256 | int left = 1; | |
1257 | ||
1258 | while (*node != NULL) { | |
1259 | parent = *node; | |
1260 | __cfqg = rb_entry_cfqg(parent); | |
1261 | ||
1262 | if (key < cfqg_key(st, __cfqg)) | |
1263 | node = &parent->rb_left; | |
1264 | else { | |
1265 | node = &parent->rb_right; | |
1266 | left = 0; | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | if (left) | |
1271 | st->left = &cfqg->rb_node; | |
1272 | ||
1273 | rb_link_node(&cfqg->rb_node, parent, node); | |
1274 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
1275 | } | |
1276 | ||
1277 | static void | |
8184f93e JT |
1278 | cfq_update_group_weight(struct cfq_group *cfqg) |
1279 | { | |
3381cb8d | 1280 | if (cfqg->new_weight) { |
8184f93e | 1281 | cfqg->weight = cfqg->new_weight; |
3381cb8d | 1282 | cfqg->new_weight = 0; |
8184f93e | 1283 | } |
fe63ce51 TM |
1284 | } |
1285 | ||
1286 | static void | |
1287 | cfq_update_group_leaf_weight(struct cfq_group *cfqg) | |
1288 | { | |
1289 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
e71357e1 TH |
1290 | |
1291 | if (cfqg->new_leaf_weight) { | |
1292 | cfqg->leaf_weight = cfqg->new_leaf_weight; | |
1293 | cfqg->new_leaf_weight = 0; | |
1294 | } | |
8184f93e JT |
1295 | } |
1296 | ||
1297 | static void | |
1298 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1299 | { | |
1d3650f7 | 1300 | unsigned int vfr = 1 << CFQ_SERVICE_SHIFT; /* start with 1 */ |
7918ffb5 | 1301 | struct cfq_group *pos = cfqg; |
1d3650f7 | 1302 | struct cfq_group *parent; |
7918ffb5 TH |
1303 | bool propagate; |
1304 | ||
1305 | /* add to the service tree */ | |
8184f93e JT |
1306 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
1307 | ||
fe63ce51 | 1308 | cfq_update_group_leaf_weight(cfqg); |
8184f93e | 1309 | __cfq_group_service_tree_add(st, cfqg); |
7918ffb5 TH |
1310 | |
1311 | /* | |
1d3650f7 TH |
1312 | * Activate @cfqg and calculate the portion of vfraction @cfqg is |
1313 | * entitled to. vfraction is calculated by walking the tree | |
1314 | * towards the root calculating the fraction it has at each level. | |
1315 | * The compounded ratio is how much vfraction @cfqg owns. | |
1316 | * | |
1317 | * Start with the proportion tasks in this cfqg has against active | |
1318 | * children cfqgs - its leaf_weight against children_weight. | |
7918ffb5 TH |
1319 | */ |
1320 | propagate = !pos->nr_active++; | |
1321 | pos->children_weight += pos->leaf_weight; | |
1d3650f7 | 1322 | vfr = vfr * pos->leaf_weight / pos->children_weight; |
7918ffb5 | 1323 | |
1d3650f7 TH |
1324 | /* |
1325 | * Compound ->weight walking up the tree. Both activation and | |
1326 | * vfraction calculation are done in the same loop. Propagation | |
1327 | * stops once an already activated node is met. vfraction | |
1328 | * calculation should always continue to the root. | |
1329 | */ | |
d02f7aa8 | 1330 | while ((parent = cfqg_parent(pos))) { |
1d3650f7 | 1331 | if (propagate) { |
fe63ce51 | 1332 | cfq_update_group_weight(pos); |
1d3650f7 TH |
1333 | propagate = !parent->nr_active++; |
1334 | parent->children_weight += pos->weight; | |
1335 | } | |
1336 | vfr = vfr * pos->weight / parent->children_weight; | |
7918ffb5 TH |
1337 | pos = parent; |
1338 | } | |
1d3650f7 TH |
1339 | |
1340 | cfqg->vfraction = max_t(unsigned, vfr, 1); | |
8184f93e JT |
1341 | } |
1342 | ||
1343 | static void | |
1344 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
1345 | { |
1346 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1347 | struct cfq_group *__cfqg; | |
1348 | struct rb_node *n; | |
1349 | ||
1350 | cfqg->nr_cfqq++; | |
760701bf | 1351 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
1352 | return; |
1353 | ||
1354 | /* | |
1355 | * Currently put the group at the end. Later implement something | |
1356 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 1357 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
1358 | */ |
1359 | n = rb_last(&st->rb); | |
1360 | if (n) { | |
1361 | __cfqg = rb_entry_cfqg(n); | |
1362 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
1363 | } else | |
1364 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
1365 | cfq_group_service_tree_add(st, cfqg); |
1366 | } | |
1fa8f6d6 | 1367 | |
8184f93e JT |
1368 | static void |
1369 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
1370 | { | |
7918ffb5 TH |
1371 | struct cfq_group *pos = cfqg; |
1372 | bool propagate; | |
1373 | ||
1374 | /* | |
1375 | * Undo activation from cfq_group_service_tree_add(). Deactivate | |
1376 | * @cfqg and propagate deactivation upwards. | |
1377 | */ | |
1378 | propagate = !--pos->nr_active; | |
1379 | pos->children_weight -= pos->leaf_weight; | |
1380 | ||
1381 | while (propagate) { | |
d02f7aa8 | 1382 | struct cfq_group *parent = cfqg_parent(pos); |
7918ffb5 TH |
1383 | |
1384 | /* @pos has 0 nr_active at this point */ | |
1385 | WARN_ON_ONCE(pos->children_weight); | |
1d3650f7 | 1386 | pos->vfraction = 0; |
7918ffb5 TH |
1387 | |
1388 | if (!parent) | |
1389 | break; | |
1390 | ||
1391 | propagate = !--parent->nr_active; | |
1392 | parent->children_weight -= pos->weight; | |
1393 | pos = parent; | |
1394 | } | |
1395 | ||
1396 | /* remove from the service tree */ | |
8184f93e JT |
1397 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1398 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
1399 | } |
1400 | ||
1401 | static void | |
8184f93e | 1402 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
1403 | { |
1404 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
1405 | ||
1406 | BUG_ON(cfqg->nr_cfqq < 1); | |
1407 | cfqg->nr_cfqq--; | |
25bc6b07 | 1408 | |
1fa8f6d6 VG |
1409 | /* If there are other cfq queues under this group, don't delete it */ |
1410 | if (cfqg->nr_cfqq) | |
1411 | return; | |
1412 | ||
2868ef7b | 1413 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 1414 | cfq_group_service_tree_del(st, cfqg); |
4d2ceea4 | 1415 | cfqg->saved_wl_slice = 0; |
155fead9 | 1416 | cfqg_stats_update_dequeue(cfqg); |
dae739eb VG |
1417 | } |
1418 | ||
167400d3 JT |
1419 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
1420 | unsigned int *unaccounted_time) | |
dae739eb | 1421 | { |
f75edf2d | 1422 | unsigned int slice_used; |
dae739eb VG |
1423 | |
1424 | /* | |
1425 | * Queue got expired before even a single request completed or | |
1426 | * got expired immediately after first request completion. | |
1427 | */ | |
1428 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
1429 | /* | |
1430 | * Also charge the seek time incurred to the group, otherwise | |
1431 | * if there are mutiple queues in the group, each can dispatch | |
1432 | * a single request on seeky media and cause lots of seek time | |
1433 | * and group will never know it. | |
1434 | */ | |
1435 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
1436 | 1); | |
1437 | } else { | |
1438 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
1439 | if (slice_used > cfqq->allocated_slice) { |
1440 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 1441 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
1442 | } |
1443 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
1444 | *unaccounted_time += cfqq->slice_start - | |
1445 | cfqq->dispatch_start; | |
dae739eb VG |
1446 | } |
1447 | ||
dae739eb VG |
1448 | return slice_used; |
1449 | } | |
1450 | ||
1451 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1452 | struct cfq_queue *cfqq) |
dae739eb VG |
1453 | { |
1454 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 1455 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1456 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1457 | - cfqg->service_tree_idle.count; | |
1d3650f7 | 1458 | unsigned int vfr; |
f26bd1f0 VG |
1459 | |
1460 | BUG_ON(nr_sync < 0); | |
167400d3 | 1461 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1462 | |
02b35081 VG |
1463 | if (iops_mode(cfqd)) |
1464 | charge = cfqq->slice_dispatch; | |
1465 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1466 | charge = cfqq->allocated_slice; | |
dae739eb | 1467 | |
1d3650f7 TH |
1468 | /* |
1469 | * Can't update vdisktime while on service tree and cfqg->vfraction | |
1470 | * is valid only while on it. Cache vfr, leave the service tree, | |
1471 | * update vdisktime and go back on. The re-addition to the tree | |
1472 | * will also update the weights as necessary. | |
1473 | */ | |
1474 | vfr = cfqg->vfraction; | |
8184f93e | 1475 | cfq_group_service_tree_del(st, cfqg); |
1d3650f7 | 1476 | cfqg->vdisktime += cfqg_scale_charge(charge, vfr); |
8184f93e | 1477 | cfq_group_service_tree_add(st, cfqg); |
dae739eb VG |
1478 | |
1479 | /* This group is being expired. Save the context */ | |
1480 | if (time_after(cfqd->workload_expires, jiffies)) { | |
4d2ceea4 | 1481 | cfqg->saved_wl_slice = cfqd->workload_expires |
dae739eb | 1482 | - jiffies; |
4d2ceea4 VG |
1483 | cfqg->saved_wl_type = cfqd->serving_wl_type; |
1484 | cfqg->saved_wl_class = cfqd->serving_wl_class; | |
dae739eb | 1485 | } else |
4d2ceea4 | 1486 | cfqg->saved_wl_slice = 0; |
2868ef7b VG |
1487 | |
1488 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1489 | st->min_vdisktime); | |
fd16d263 JP |
1490 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1491 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1492 | used_sl, cfqq->slice_dispatch, charge, | |
1493 | iops_mode(cfqd), cfqq->nr_sectors); | |
155fead9 TH |
1494 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
1495 | cfqg_stats_set_start_empty_time(cfqg); | |
1fa8f6d6 VG |
1496 | } |
1497 | ||
f51b802c TH |
1498 | /** |
1499 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1500 | * @cfqg: cfq_group to initialize | |
1501 | * | |
1502 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1503 | * is enabled or not. | |
1504 | */ | |
1505 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1506 | { | |
1507 | struct cfq_rb_root *st; | |
1508 | int i, j; | |
1509 | ||
1510 | for_each_cfqg_st(cfqg, i, j, st) | |
1511 | *st = CFQ_RB_ROOT; | |
1512 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1513 | ||
1514 | cfqg->ttime.last_end_request = jiffies; | |
1515 | } | |
1516 | ||
25fb5169 | 1517 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 1518 | static void cfq_pd_init(struct blkcg_gq *blkg) |
f469a7b4 | 1519 | { |
0381411e | 1520 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
25fb5169 | 1521 | |
f51b802c | 1522 | cfq_init_cfqg_base(cfqg); |
3381cb8d | 1523 | cfqg->weight = blkg->blkcg->cfq_weight; |
e71357e1 | 1524 | cfqg->leaf_weight = blkg->blkcg->cfq_leaf_weight; |
25fb5169 VG |
1525 | } |
1526 | ||
0b39920b TH |
1527 | static void cfq_pd_offline(struct blkcg_gq *blkg) |
1528 | { | |
1529 | /* | |
1530 | * @blkg is going offline and will be ignored by | |
1531 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
1532 | * that they don't get lost. If IOs complete after this point, the | |
1533 | * stats for them will be lost. Oh well... | |
1534 | */ | |
1535 | cfqg_stats_xfer_dead(blkg_to_cfqg(blkg)); | |
1536 | } | |
1537 | ||
43114018 TH |
1538 | /* offset delta from cfqg->stats to cfqg->dead_stats */ |
1539 | static const int dead_stats_off_delta = offsetof(struct cfq_group, dead_stats) - | |
1540 | offsetof(struct cfq_group, stats); | |
1541 | ||
1542 | /* to be used by recursive prfill, sums live and dead stats recursively */ | |
1543 | static u64 cfqg_stat_pd_recursive_sum(struct blkg_policy_data *pd, int off) | |
1544 | { | |
1545 | u64 sum = 0; | |
1546 | ||
1547 | sum += blkg_stat_recursive_sum(pd, off); | |
1548 | sum += blkg_stat_recursive_sum(pd, off + dead_stats_off_delta); | |
1549 | return sum; | |
1550 | } | |
1551 | ||
1552 | /* to be used by recursive prfill, sums live and dead rwstats recursively */ | |
1553 | static struct blkg_rwstat cfqg_rwstat_pd_recursive_sum(struct blkg_policy_data *pd, | |
1554 | int off) | |
1555 | { | |
1556 | struct blkg_rwstat a, b; | |
1557 | ||
1558 | a = blkg_rwstat_recursive_sum(pd, off); | |
1559 | b = blkg_rwstat_recursive_sum(pd, off + dead_stats_off_delta); | |
1560 | blkg_rwstat_merge(&a, &b); | |
1561 | return a; | |
1562 | } | |
1563 | ||
689665af TH |
1564 | static void cfq_pd_reset_stats(struct blkcg_gq *blkg) |
1565 | { | |
1566 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); | |
1567 | ||
1568 | cfqg_stats_reset(&cfqg->stats); | |
0b39920b | 1569 | cfqg_stats_reset(&cfqg->dead_stats); |
25fb5169 VG |
1570 | } |
1571 | ||
1572 | /* | |
3e59cf9d VG |
1573 | * Search for the cfq group current task belongs to. request_queue lock must |
1574 | * be held. | |
25fb5169 | 1575 | */ |
cd1604fa | 1576 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
3c798398 | 1577 | struct blkcg *blkcg) |
25fb5169 | 1578 | { |
f469a7b4 | 1579 | struct request_queue *q = cfqd->queue; |
cd1604fa | 1580 | struct cfq_group *cfqg = NULL; |
25fb5169 | 1581 | |
3c798398 TH |
1582 | /* avoid lookup for the common case where there's no blkcg */ |
1583 | if (blkcg == &blkcg_root) { | |
cd1604fa TH |
1584 | cfqg = cfqd->root_group; |
1585 | } else { | |
3c798398 | 1586 | struct blkcg_gq *blkg; |
f469a7b4 | 1587 | |
3c96cb32 | 1588 | blkg = blkg_lookup_create(blkcg, q); |
cd1604fa | 1589 | if (!IS_ERR(blkg)) |
0381411e | 1590 | cfqg = blkg_to_cfqg(blkg); |
cd1604fa | 1591 | } |
f469a7b4 | 1592 | |
25fb5169 VG |
1593 | return cfqg; |
1594 | } | |
1595 | ||
1596 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1597 | { | |
1598 | /* Currently, all async queues are mapped to root group */ | |
1599 | if (!cfq_cfqq_sync(cfqq)) | |
f51b802c | 1600 | cfqg = cfqq->cfqd->root_group; |
25fb5169 VG |
1601 | |
1602 | cfqq->cfqg = cfqg; | |
b1c35769 | 1603 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1604 | cfqg_get(cfqg); |
b1c35769 VG |
1605 | } |
1606 | ||
f95a04af TH |
1607 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, |
1608 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1609 | { |
f95a04af | 1610 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
3381cb8d TH |
1611 | |
1612 | if (!cfqg->dev_weight) | |
60c2bc2d | 1613 | return 0; |
f95a04af | 1614 | return __blkg_prfill_u64(sf, pd, cfqg->dev_weight); |
60c2bc2d TH |
1615 | } |
1616 | ||
3381cb8d TH |
1617 | static int cfqg_print_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1618 | struct seq_file *sf) | |
60c2bc2d | 1619 | { |
3c798398 TH |
1620 | blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), |
1621 | cfqg_prfill_weight_device, &blkcg_policy_cfq, 0, | |
60c2bc2d TH |
1622 | false); |
1623 | return 0; | |
1624 | } | |
1625 | ||
e71357e1 TH |
1626 | static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf, |
1627 | struct blkg_policy_data *pd, int off) | |
1628 | { | |
1629 | struct cfq_group *cfqg = pd_to_cfqg(pd); | |
1630 | ||
1631 | if (!cfqg->dev_leaf_weight) | |
1632 | return 0; | |
1633 | return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight); | |
1634 | } | |
1635 | ||
1636 | static int cfqg_print_leaf_weight_device(struct cgroup *cgrp, | |
1637 | struct cftype *cft, | |
1638 | struct seq_file *sf) | |
1639 | { | |
1640 | blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), | |
1641 | cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq, 0, | |
1642 | false); | |
1643 | return 0; | |
1644 | } | |
1645 | ||
3381cb8d TH |
1646 | static int cfq_print_weight(struct cgroup *cgrp, struct cftype *cft, |
1647 | struct seq_file *sf) | |
60c2bc2d | 1648 | { |
3c798398 | 1649 | seq_printf(sf, "%u\n", cgroup_to_blkcg(cgrp)->cfq_weight); |
60c2bc2d TH |
1650 | return 0; |
1651 | } | |
1652 | ||
e71357e1 TH |
1653 | static int cfq_print_leaf_weight(struct cgroup *cgrp, struct cftype *cft, |
1654 | struct seq_file *sf) | |
1655 | { | |
1656 | seq_printf(sf, "%u\n", | |
1657 | cgroup_to_blkcg(cgrp)->cfq_leaf_weight); | |
1658 | return 0; | |
1659 | } | |
1660 | ||
1661 | static int __cfqg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, | |
1662 | const char *buf, bool is_leaf_weight) | |
60c2bc2d | 1663 | { |
3c798398 | 1664 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
60c2bc2d | 1665 | struct blkg_conf_ctx ctx; |
3381cb8d | 1666 | struct cfq_group *cfqg; |
60c2bc2d TH |
1667 | int ret; |
1668 | ||
3c798398 | 1669 | ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx); |
60c2bc2d TH |
1670 | if (ret) |
1671 | return ret; | |
1672 | ||
1673 | ret = -EINVAL; | |
3381cb8d | 1674 | cfqg = blkg_to_cfqg(ctx.blkg); |
a2b1693b | 1675 | if (!ctx.v || (ctx.v >= CFQ_WEIGHT_MIN && ctx.v <= CFQ_WEIGHT_MAX)) { |
e71357e1 TH |
1676 | if (!is_leaf_weight) { |
1677 | cfqg->dev_weight = ctx.v; | |
1678 | cfqg->new_weight = ctx.v ?: blkcg->cfq_weight; | |
1679 | } else { | |
1680 | cfqg->dev_leaf_weight = ctx.v; | |
1681 | cfqg->new_leaf_weight = ctx.v ?: blkcg->cfq_leaf_weight; | |
1682 | } | |
60c2bc2d TH |
1683 | ret = 0; |
1684 | } | |
1685 | ||
1686 | blkg_conf_finish(&ctx); | |
1687 | return ret; | |
1688 | } | |
1689 | ||
e71357e1 TH |
1690 | static int cfqg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, |
1691 | const char *buf) | |
1692 | { | |
1693 | return __cfqg_set_weight_device(cgrp, cft, buf, false); | |
1694 | } | |
1695 | ||
1696 | static int cfqg_set_leaf_weight_device(struct cgroup *cgrp, struct cftype *cft, | |
1697 | const char *buf) | |
1698 | { | |
1699 | return __cfqg_set_weight_device(cgrp, cft, buf, true); | |
1700 | } | |
1701 | ||
1702 | static int __cfq_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val, | |
1703 | bool is_leaf_weight) | |
60c2bc2d | 1704 | { |
3c798398 TH |
1705 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
1706 | struct blkcg_gq *blkg; | |
60c2bc2d | 1707 | |
3381cb8d | 1708 | if (val < CFQ_WEIGHT_MIN || val > CFQ_WEIGHT_MAX) |
60c2bc2d TH |
1709 | return -EINVAL; |
1710 | ||
1711 | spin_lock_irq(&blkcg->lock); | |
e71357e1 TH |
1712 | |
1713 | if (!is_leaf_weight) | |
1714 | blkcg->cfq_weight = val; | |
1715 | else | |
1716 | blkcg->cfq_leaf_weight = val; | |
60c2bc2d | 1717 | |
b67bfe0d | 1718 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
3381cb8d | 1719 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
60c2bc2d | 1720 | |
e71357e1 TH |
1721 | if (!cfqg) |
1722 | continue; | |
1723 | ||
1724 | if (!is_leaf_weight) { | |
1725 | if (!cfqg->dev_weight) | |
1726 | cfqg->new_weight = blkcg->cfq_weight; | |
1727 | } else { | |
1728 | if (!cfqg->dev_leaf_weight) | |
1729 | cfqg->new_leaf_weight = blkcg->cfq_leaf_weight; | |
1730 | } | |
60c2bc2d TH |
1731 | } |
1732 | ||
1733 | spin_unlock_irq(&blkcg->lock); | |
1734 | return 0; | |
1735 | } | |
1736 | ||
e71357e1 TH |
1737 | static int cfq_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) |
1738 | { | |
1739 | return __cfq_set_weight(cgrp, cft, val, false); | |
1740 | } | |
1741 | ||
1742 | static int cfq_set_leaf_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) | |
1743 | { | |
1744 | return __cfq_set_weight(cgrp, cft, val, true); | |
1745 | } | |
1746 | ||
5bc4afb1 TH |
1747 | static int cfqg_print_stat(struct cgroup *cgrp, struct cftype *cft, |
1748 | struct seq_file *sf) | |
1749 | { | |
3c798398 | 1750 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
5bc4afb1 | 1751 | |
3c798398 | 1752 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_stat, &blkcg_policy_cfq, |
5bc4afb1 TH |
1753 | cft->private, false); |
1754 | return 0; | |
1755 | } | |
1756 | ||
1757 | static int cfqg_print_rwstat(struct cgroup *cgrp, struct cftype *cft, | |
1758 | struct seq_file *sf) | |
1759 | { | |
3c798398 | 1760 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
5bc4afb1 | 1761 | |
3c798398 | 1762 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_rwstat, &blkcg_policy_cfq, |
5bc4afb1 TH |
1763 | cft->private, true); |
1764 | return 0; | |
1765 | } | |
1766 | ||
43114018 TH |
1767 | static u64 cfqg_prfill_stat_recursive(struct seq_file *sf, |
1768 | struct blkg_policy_data *pd, int off) | |
1769 | { | |
1770 | u64 sum = cfqg_stat_pd_recursive_sum(pd, off); | |
1771 | ||
1772 | return __blkg_prfill_u64(sf, pd, sum); | |
1773 | } | |
1774 | ||
1775 | static u64 cfqg_prfill_rwstat_recursive(struct seq_file *sf, | |
1776 | struct blkg_policy_data *pd, int off) | |
1777 | { | |
1778 | struct blkg_rwstat sum = cfqg_rwstat_pd_recursive_sum(pd, off); | |
1779 | ||
1780 | return __blkg_prfill_rwstat(sf, pd, &sum); | |
1781 | } | |
1782 | ||
1783 | static int cfqg_print_stat_recursive(struct cgroup *cgrp, struct cftype *cft, | |
1784 | struct seq_file *sf) | |
1785 | { | |
1786 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); | |
1787 | ||
1788 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_stat_recursive, | |
1789 | &blkcg_policy_cfq, cft->private, false); | |
1790 | return 0; | |
1791 | } | |
1792 | ||
1793 | static int cfqg_print_rwstat_recursive(struct cgroup *cgrp, struct cftype *cft, | |
1794 | struct seq_file *sf) | |
1795 | { | |
1796 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); | |
1797 | ||
1798 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_rwstat_recursive, | |
1799 | &blkcg_policy_cfq, cft->private, true); | |
1800 | return 0; | |
1801 | } | |
1802 | ||
60c2bc2d | 1803 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
f95a04af TH |
1804 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, |
1805 | struct blkg_policy_data *pd, int off) | |
60c2bc2d | 1806 | { |
f95a04af | 1807 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
155fead9 | 1808 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
60c2bc2d TH |
1809 | u64 v = 0; |
1810 | ||
1811 | if (samples) { | |
155fead9 | 1812 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
85f58908 | 1813 | v = div64_u64(v, samples); |
60c2bc2d | 1814 | } |
f95a04af | 1815 | __blkg_prfill_u64(sf, pd, v); |
60c2bc2d TH |
1816 | return 0; |
1817 | } | |
1818 | ||
1819 | /* print avg_queue_size */ | |
155fead9 TH |
1820 | static int cfqg_print_avg_queue_size(struct cgroup *cgrp, struct cftype *cft, |
1821 | struct seq_file *sf) | |
60c2bc2d | 1822 | { |
3c798398 | 1823 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
60c2bc2d | 1824 | |
155fead9 | 1825 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_avg_queue_size, |
3c798398 | 1826 | &blkcg_policy_cfq, 0, false); |
60c2bc2d TH |
1827 | return 0; |
1828 | } | |
1829 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1830 | ||
1831 | static struct cftype cfq_blkcg_files[] = { | |
1d3650f7 | 1832 | /* on root, weight is mapped to leaf_weight */ |
60c2bc2d TH |
1833 | { |
1834 | .name = "weight_device", | |
1d3650f7 TH |
1835 | .flags = CFTYPE_ONLY_ON_ROOT, |
1836 | .read_seq_string = cfqg_print_leaf_weight_device, | |
1837 | .write_string = cfqg_set_leaf_weight_device, | |
60c2bc2d TH |
1838 | .max_write_len = 256, |
1839 | }, | |
1840 | { | |
1841 | .name = "weight", | |
1d3650f7 TH |
1842 | .flags = CFTYPE_ONLY_ON_ROOT, |
1843 | .read_seq_string = cfq_print_leaf_weight, | |
1844 | .write_u64 = cfq_set_leaf_weight, | |
60c2bc2d | 1845 | }, |
e71357e1 | 1846 | |
1d3650f7 | 1847 | /* no such mapping necessary for !roots */ |
60c2bc2d TH |
1848 | { |
1849 | .name = "weight_device", | |
1d3650f7 | 1850 | .flags = CFTYPE_NOT_ON_ROOT, |
3381cb8d TH |
1851 | .read_seq_string = cfqg_print_weight_device, |
1852 | .write_string = cfqg_set_weight_device, | |
60c2bc2d TH |
1853 | .max_write_len = 256, |
1854 | }, | |
1855 | { | |
1856 | .name = "weight", | |
1d3650f7 | 1857 | .flags = CFTYPE_NOT_ON_ROOT, |
3381cb8d TH |
1858 | .read_seq_string = cfq_print_weight, |
1859 | .write_u64 = cfq_set_weight, | |
60c2bc2d | 1860 | }, |
e71357e1 | 1861 | |
e71357e1 TH |
1862 | { |
1863 | .name = "leaf_weight_device", | |
e71357e1 TH |
1864 | .read_seq_string = cfqg_print_leaf_weight_device, |
1865 | .write_string = cfqg_set_leaf_weight_device, | |
1866 | .max_write_len = 256, | |
1867 | }, | |
1868 | { | |
1869 | .name = "leaf_weight", | |
e71357e1 TH |
1870 | .read_seq_string = cfq_print_leaf_weight, |
1871 | .write_u64 = cfq_set_leaf_weight, | |
1872 | }, | |
1873 | ||
43114018 | 1874 | /* statistics, covers only the tasks in the cfqg */ |
60c2bc2d TH |
1875 | { |
1876 | .name = "time", | |
5bc4afb1 TH |
1877 | .private = offsetof(struct cfq_group, stats.time), |
1878 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1879 | }, |
1880 | { | |
1881 | .name = "sectors", | |
5bc4afb1 TH |
1882 | .private = offsetof(struct cfq_group, stats.sectors), |
1883 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1884 | }, |
1885 | { | |
1886 | .name = "io_service_bytes", | |
5bc4afb1 TH |
1887 | .private = offsetof(struct cfq_group, stats.service_bytes), |
1888 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1889 | }, |
1890 | { | |
1891 | .name = "io_serviced", | |
5bc4afb1 TH |
1892 | .private = offsetof(struct cfq_group, stats.serviced), |
1893 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1894 | }, |
1895 | { | |
1896 | .name = "io_service_time", | |
5bc4afb1 TH |
1897 | .private = offsetof(struct cfq_group, stats.service_time), |
1898 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1899 | }, |
1900 | { | |
1901 | .name = "io_wait_time", | |
5bc4afb1 TH |
1902 | .private = offsetof(struct cfq_group, stats.wait_time), |
1903 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1904 | }, |
1905 | { | |
1906 | .name = "io_merged", | |
5bc4afb1 TH |
1907 | .private = offsetof(struct cfq_group, stats.merged), |
1908 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d TH |
1909 | }, |
1910 | { | |
1911 | .name = "io_queued", | |
5bc4afb1 TH |
1912 | .private = offsetof(struct cfq_group, stats.queued), |
1913 | .read_seq_string = cfqg_print_rwstat, | |
60c2bc2d | 1914 | }, |
43114018 TH |
1915 | |
1916 | /* the same statictics which cover the cfqg and its descendants */ | |
1917 | { | |
1918 | .name = "time_recursive", | |
1919 | .private = offsetof(struct cfq_group, stats.time), | |
1920 | .read_seq_string = cfqg_print_stat_recursive, | |
1921 | }, | |
1922 | { | |
1923 | .name = "sectors_recursive", | |
1924 | .private = offsetof(struct cfq_group, stats.sectors), | |
1925 | .read_seq_string = cfqg_print_stat_recursive, | |
1926 | }, | |
1927 | { | |
1928 | .name = "io_service_bytes_recursive", | |
1929 | .private = offsetof(struct cfq_group, stats.service_bytes), | |
1930 | .read_seq_string = cfqg_print_rwstat_recursive, | |
1931 | }, | |
1932 | { | |
1933 | .name = "io_serviced_recursive", | |
1934 | .private = offsetof(struct cfq_group, stats.serviced), | |
1935 | .read_seq_string = cfqg_print_rwstat_recursive, | |
1936 | }, | |
1937 | { | |
1938 | .name = "io_service_time_recursive", | |
1939 | .private = offsetof(struct cfq_group, stats.service_time), | |
1940 | .read_seq_string = cfqg_print_rwstat_recursive, | |
1941 | }, | |
1942 | { | |
1943 | .name = "io_wait_time_recursive", | |
1944 | .private = offsetof(struct cfq_group, stats.wait_time), | |
1945 | .read_seq_string = cfqg_print_rwstat_recursive, | |
1946 | }, | |
1947 | { | |
1948 | .name = "io_merged_recursive", | |
1949 | .private = offsetof(struct cfq_group, stats.merged), | |
1950 | .read_seq_string = cfqg_print_rwstat_recursive, | |
1951 | }, | |
1952 | { | |
1953 | .name = "io_queued_recursive", | |
1954 | .private = offsetof(struct cfq_group, stats.queued), | |
1955 | .read_seq_string = cfqg_print_rwstat_recursive, | |
1956 | }, | |
60c2bc2d TH |
1957 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
1958 | { | |
1959 | .name = "avg_queue_size", | |
155fead9 | 1960 | .read_seq_string = cfqg_print_avg_queue_size, |
60c2bc2d TH |
1961 | }, |
1962 | { | |
1963 | .name = "group_wait_time", | |
5bc4afb1 TH |
1964 | .private = offsetof(struct cfq_group, stats.group_wait_time), |
1965 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1966 | }, |
1967 | { | |
1968 | .name = "idle_time", | |
5bc4afb1 TH |
1969 | .private = offsetof(struct cfq_group, stats.idle_time), |
1970 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1971 | }, |
1972 | { | |
1973 | .name = "empty_time", | |
5bc4afb1 TH |
1974 | .private = offsetof(struct cfq_group, stats.empty_time), |
1975 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1976 | }, |
1977 | { | |
1978 | .name = "dequeue", | |
5bc4afb1 TH |
1979 | .private = offsetof(struct cfq_group, stats.dequeue), |
1980 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1981 | }, |
1982 | { | |
1983 | .name = "unaccounted_time", | |
5bc4afb1 TH |
1984 | .private = offsetof(struct cfq_group, stats.unaccounted_time), |
1985 | .read_seq_string = cfqg_print_stat, | |
60c2bc2d TH |
1986 | }, |
1987 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ | |
1988 | { } /* terminate */ | |
1989 | }; | |
25fb5169 | 1990 | #else /* GROUP_IOSCHED */ |
cd1604fa | 1991 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
3c798398 | 1992 | struct blkcg *blkcg) |
25fb5169 | 1993 | { |
f51b802c | 1994 | return cfqd->root_group; |
25fb5169 | 1995 | } |
7f1dc8a2 | 1996 | |
25fb5169 VG |
1997 | static inline void |
1998 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1999 | cfqq->cfqg = cfqg; | |
2000 | } | |
2001 | ||
2002 | #endif /* GROUP_IOSCHED */ | |
2003 | ||
498d3aa2 | 2004 | /* |
c0324a02 | 2005 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
2006 | * requests waiting to be processed. It is sorted in the order that |
2007 | * we will service the queues. | |
2008 | */ | |
a36e71f9 | 2009 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2010 | bool add_front) |
d9e7620e | 2011 | { |
0871714e JA |
2012 | struct rb_node **p, *parent; |
2013 | struct cfq_queue *__cfqq; | |
d9e7620e | 2014 | unsigned long rb_key; |
34b98d03 | 2015 | struct cfq_rb_root *st; |
498d3aa2 | 2016 | int left; |
dae739eb | 2017 | int new_cfqq = 1; |
ae30c286 | 2018 | |
34b98d03 | 2019 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq)); |
0871714e JA |
2020 | if (cfq_class_idle(cfqq)) { |
2021 | rb_key = CFQ_IDLE_DELAY; | |
34b98d03 | 2022 | parent = rb_last(&st->rb); |
0871714e JA |
2023 | if (parent && parent != &cfqq->rb_node) { |
2024 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2025 | rb_key += __cfqq->rb_key; | |
2026 | } else | |
2027 | rb_key += jiffies; | |
2028 | } else if (!add_front) { | |
b9c8946b JA |
2029 | /* |
2030 | * Get our rb key offset. Subtract any residual slice | |
2031 | * value carried from last service. A negative resid | |
2032 | * count indicates slice overrun, and this should position | |
2033 | * the next service time further away in the tree. | |
2034 | */ | |
edd75ffd | 2035 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 2036 | rb_key -= cfqq->slice_resid; |
edd75ffd | 2037 | cfqq->slice_resid = 0; |
48e025e6 CZ |
2038 | } else { |
2039 | rb_key = -HZ; | |
34b98d03 | 2040 | __cfqq = cfq_rb_first(st); |
48e025e6 CZ |
2041 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
2042 | } | |
1da177e4 | 2043 | |
d9e7620e | 2044 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 2045 | new_cfqq = 0; |
99f9628a | 2046 | /* |
d9e7620e | 2047 | * same position, nothing more to do |
99f9628a | 2048 | */ |
34b98d03 | 2049 | if (rb_key == cfqq->rb_key && cfqq->service_tree == st) |
d9e7620e | 2050 | return; |
1da177e4 | 2051 | |
aa6f6a3d CZ |
2052 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
2053 | cfqq->service_tree = NULL; | |
1da177e4 | 2054 | } |
d9e7620e | 2055 | |
498d3aa2 | 2056 | left = 1; |
0871714e | 2057 | parent = NULL; |
34b98d03 VG |
2058 | cfqq->service_tree = st; |
2059 | p = &st->rb.rb_node; | |
d9e7620e JA |
2060 | while (*p) { |
2061 | parent = *p; | |
2062 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
2063 | ||
0c534e0a | 2064 | /* |
c0324a02 | 2065 | * sort by key, that represents service time. |
0c534e0a | 2066 | */ |
c0324a02 | 2067 | if (time_before(rb_key, __cfqq->rb_key)) |
1f23f121 | 2068 | p = &parent->rb_left; |
c0324a02 | 2069 | else { |
1f23f121 | 2070 | p = &parent->rb_right; |
cc09e299 | 2071 | left = 0; |
c0324a02 | 2072 | } |
d9e7620e JA |
2073 | } |
2074 | ||
cc09e299 | 2075 | if (left) |
34b98d03 | 2076 | st->left = &cfqq->rb_node; |
cc09e299 | 2077 | |
d9e7620e JA |
2078 | cfqq->rb_key = rb_key; |
2079 | rb_link_node(&cfqq->rb_node, parent, p); | |
34b98d03 VG |
2080 | rb_insert_color(&cfqq->rb_node, &st->rb); |
2081 | st->count++; | |
20359f27 | 2082 | if (add_front || !new_cfqq) |
dae739eb | 2083 | return; |
8184f93e | 2084 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
2085 | } |
2086 | ||
a36e71f9 | 2087 | static struct cfq_queue * |
f2d1f0ae JA |
2088 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
2089 | sector_t sector, struct rb_node **ret_parent, | |
2090 | struct rb_node ***rb_link) | |
a36e71f9 | 2091 | { |
a36e71f9 JA |
2092 | struct rb_node **p, *parent; |
2093 | struct cfq_queue *cfqq = NULL; | |
2094 | ||
2095 | parent = NULL; | |
2096 | p = &root->rb_node; | |
2097 | while (*p) { | |
2098 | struct rb_node **n; | |
2099 | ||
2100 | parent = *p; | |
2101 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
2102 | ||
2103 | /* | |
2104 | * Sort strictly based on sector. Smallest to the left, | |
2105 | * largest to the right. | |
2106 | */ | |
2e46e8b2 | 2107 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 2108 | n = &(*p)->rb_right; |
2e46e8b2 | 2109 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
2110 | n = &(*p)->rb_left; |
2111 | else | |
2112 | break; | |
2113 | p = n; | |
3ac6c9f8 | 2114 | cfqq = NULL; |
a36e71f9 JA |
2115 | } |
2116 | ||
2117 | *ret_parent = parent; | |
2118 | if (rb_link) | |
2119 | *rb_link = p; | |
3ac6c9f8 | 2120 | return cfqq; |
a36e71f9 JA |
2121 | } |
2122 | ||
2123 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2124 | { | |
a36e71f9 JA |
2125 | struct rb_node **p, *parent; |
2126 | struct cfq_queue *__cfqq; | |
2127 | ||
f2d1f0ae JA |
2128 | if (cfqq->p_root) { |
2129 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2130 | cfqq->p_root = NULL; | |
2131 | } | |
a36e71f9 JA |
2132 | |
2133 | if (cfq_class_idle(cfqq)) | |
2134 | return; | |
2135 | if (!cfqq->next_rq) | |
2136 | return; | |
2137 | ||
f2d1f0ae | 2138 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
2139 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
2140 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
2141 | if (!__cfqq) { |
2142 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
2143 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
2144 | } else | |
2145 | cfqq->p_root = NULL; | |
a36e71f9 JA |
2146 | } |
2147 | ||
498d3aa2 JA |
2148 | /* |
2149 | * Update cfqq's position in the service tree. | |
2150 | */ | |
edd75ffd | 2151 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 2152 | { |
6d048f53 JA |
2153 | /* |
2154 | * Resorting requires the cfqq to be on the RR list already. | |
2155 | */ | |
a36e71f9 | 2156 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 2157 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
2158 | cfq_prio_tree_add(cfqd, cfqq); |
2159 | } | |
6d048f53 JA |
2160 | } |
2161 | ||
1da177e4 LT |
2162 | /* |
2163 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 2164 | * the pending list according to last request service |
1da177e4 | 2165 | */ |
febffd61 | 2166 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2167 | { |
7b679138 | 2168 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
2169 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
2170 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 2171 | cfqd->busy_queues++; |
ef8a41df SL |
2172 | if (cfq_cfqq_sync(cfqq)) |
2173 | cfqd->busy_sync_queues++; | |
1da177e4 | 2174 | |
edd75ffd | 2175 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
2176 | } |
2177 | ||
498d3aa2 JA |
2178 | /* |
2179 | * Called when the cfqq no longer has requests pending, remove it from | |
2180 | * the service tree. | |
2181 | */ | |
febffd61 | 2182 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2183 | { |
7b679138 | 2184 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
2185 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
2186 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 2187 | |
aa6f6a3d CZ |
2188 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
2189 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
2190 | cfqq->service_tree = NULL; | |
2191 | } | |
f2d1f0ae JA |
2192 | if (cfqq->p_root) { |
2193 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2194 | cfqq->p_root = NULL; | |
2195 | } | |
d9e7620e | 2196 | |
8184f93e | 2197 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
2198 | BUG_ON(!cfqd->busy_queues); |
2199 | cfqd->busy_queues--; | |
ef8a41df SL |
2200 | if (cfq_cfqq_sync(cfqq)) |
2201 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
2202 | } |
2203 | ||
2204 | /* | |
2205 | * rb tree support functions | |
2206 | */ | |
febffd61 | 2207 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 2208 | { |
5e705374 | 2209 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 2210 | const int sync = rq_is_sync(rq); |
1da177e4 | 2211 | |
b4878f24 JA |
2212 | BUG_ON(!cfqq->queued[sync]); |
2213 | cfqq->queued[sync]--; | |
1da177e4 | 2214 | |
5e705374 | 2215 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 2216 | |
f04a6424 VG |
2217 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
2218 | /* | |
2219 | * Queue will be deleted from service tree when we actually | |
2220 | * expire it later. Right now just remove it from prio tree | |
2221 | * as it is empty. | |
2222 | */ | |
2223 | if (cfqq->p_root) { | |
2224 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
2225 | cfqq->p_root = NULL; | |
2226 | } | |
2227 | } | |
1da177e4 LT |
2228 | } |
2229 | ||
5e705374 | 2230 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 2231 | { |
5e705374 | 2232 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 2233 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 2234 | struct request *prev; |
1da177e4 | 2235 | |
5380a101 | 2236 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 2237 | |
796d5116 | 2238 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
2239 | |
2240 | if (!cfq_cfqq_on_rr(cfqq)) | |
2241 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
2242 | |
2243 | /* | |
2244 | * check if this request is a better next-serve candidate | |
2245 | */ | |
a36e71f9 | 2246 | prev = cfqq->next_rq; |
cf7c25cf | 2247 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
2248 | |
2249 | /* | |
2250 | * adjust priority tree position, if ->next_rq changes | |
2251 | */ | |
2252 | if (prev != cfqq->next_rq) | |
2253 | cfq_prio_tree_add(cfqd, cfqq); | |
2254 | ||
5044eed4 | 2255 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
2256 | } |
2257 | ||
febffd61 | 2258 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 2259 | { |
5380a101 JA |
2260 | elv_rb_del(&cfqq->sort_list, rq); |
2261 | cfqq->queued[rq_is_sync(rq)]--; | |
155fead9 | 2262 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
5e705374 | 2263 | cfq_add_rq_rb(rq); |
155fead9 TH |
2264 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
2265 | rq->cmd_flags); | |
1da177e4 LT |
2266 | } |
2267 | ||
206dc69b JA |
2268 | static struct request * |
2269 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 2270 | { |
206dc69b | 2271 | struct task_struct *tsk = current; |
c5869807 | 2272 | struct cfq_io_cq *cic; |
206dc69b | 2273 | struct cfq_queue *cfqq; |
1da177e4 | 2274 | |
4ac845a2 | 2275 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
2276 | if (!cic) |
2277 | return NULL; | |
2278 | ||
2279 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
f73a1c7d KO |
2280 | if (cfqq) |
2281 | return elv_rb_find(&cfqq->sort_list, bio_end_sector(bio)); | |
1da177e4 | 2282 | |
1da177e4 LT |
2283 | return NULL; |
2284 | } | |
2285 | ||
165125e1 | 2286 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2287 | { |
22e2c507 | 2288 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 2289 | |
53c583d2 | 2290 | cfqd->rq_in_driver++; |
7b679138 | 2291 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 2292 | cfqd->rq_in_driver); |
25776e35 | 2293 | |
5b93629b | 2294 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
2295 | } |
2296 | ||
165125e1 | 2297 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2298 | { |
b4878f24 JA |
2299 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2300 | ||
53c583d2 CZ |
2301 | WARN_ON(!cfqd->rq_in_driver); |
2302 | cfqd->rq_in_driver--; | |
7b679138 | 2303 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 2304 | cfqd->rq_in_driver); |
1da177e4 LT |
2305 | } |
2306 | ||
b4878f24 | 2307 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 2308 | { |
5e705374 | 2309 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 2310 | |
5e705374 JA |
2311 | if (cfqq->next_rq == rq) |
2312 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 2313 | |
b4878f24 | 2314 | list_del_init(&rq->queuelist); |
5e705374 | 2315 | cfq_del_rq_rb(rq); |
374f84ac | 2316 | |
45333d5a | 2317 | cfqq->cfqd->rq_queued--; |
155fead9 | 2318 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
65299a3b CH |
2319 | if (rq->cmd_flags & REQ_PRIO) { |
2320 | WARN_ON(!cfqq->prio_pending); | |
2321 | cfqq->prio_pending--; | |
b53d1ed7 | 2322 | } |
1da177e4 LT |
2323 | } |
2324 | ||
165125e1 JA |
2325 | static int cfq_merge(struct request_queue *q, struct request **req, |
2326 | struct bio *bio) | |
1da177e4 LT |
2327 | { |
2328 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2329 | struct request *__rq; | |
1da177e4 | 2330 | |
206dc69b | 2331 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 2332 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
2333 | *req = __rq; |
2334 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
2335 | } |
2336 | ||
2337 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
2338 | } |
2339 | ||
165125e1 | 2340 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 2341 | int type) |
1da177e4 | 2342 | { |
21183b07 | 2343 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 2344 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 2345 | |
5e705374 | 2346 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 2347 | } |
1da177e4 LT |
2348 | } |
2349 | ||
812d4026 DS |
2350 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
2351 | struct bio *bio) | |
2352 | { | |
155fead9 | 2353 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_rw); |
812d4026 DS |
2354 | } |
2355 | ||
1da177e4 | 2356 | static void |
165125e1 | 2357 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
2358 | struct request *next) |
2359 | { | |
cf7c25cf | 2360 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
2361 | struct cfq_data *cfqd = q->elevator->elevator_data; |
2362 | ||
22e2c507 JA |
2363 | /* |
2364 | * reposition in fifo if next is older than rq | |
2365 | */ | |
2366 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
3d106fba SL |
2367 | time_before(rq_fifo_time(next), rq_fifo_time(rq)) && |
2368 | cfqq == RQ_CFQQ(next)) { | |
22e2c507 | 2369 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
2370 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
2371 | } | |
22e2c507 | 2372 | |
cf7c25cf CZ |
2373 | if (cfqq->next_rq == next) |
2374 | cfqq->next_rq = rq; | |
b4878f24 | 2375 | cfq_remove_request(next); |
155fead9 | 2376 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
4a0b75c7 SL |
2377 | |
2378 | cfqq = RQ_CFQQ(next); | |
2379 | /* | |
2380 | * all requests of this queue are merged to other queues, delete it | |
2381 | * from the service tree. If it's the active_queue, | |
2382 | * cfq_dispatch_requests() will choose to expire it or do idle | |
2383 | */ | |
2384 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
2385 | cfqq != cfqd->active_queue) | |
2386 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
2387 | } |
2388 | ||
165125e1 | 2389 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
2390 | struct bio *bio) |
2391 | { | |
2392 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 2393 | struct cfq_io_cq *cic; |
da775265 | 2394 | struct cfq_queue *cfqq; |
da775265 JA |
2395 | |
2396 | /* | |
ec8acb69 | 2397 | * Disallow merge of a sync bio into an async request. |
da775265 | 2398 | */ |
91fac317 | 2399 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 2400 | return false; |
da775265 JA |
2401 | |
2402 | /* | |
f1a4f4d3 | 2403 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 2404 | * merge only if rq is queued there. |
f1a4f4d3 | 2405 | */ |
07c2bd37 TH |
2406 | cic = cfq_cic_lookup(cfqd, current->io_context); |
2407 | if (!cic) | |
2408 | return false; | |
719d3402 | 2409 | |
91fac317 | 2410 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 2411 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
2412 | } |
2413 | ||
812df48d DS |
2414 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2415 | { | |
2416 | del_timer(&cfqd->idle_slice_timer); | |
155fead9 | 2417 | cfqg_stats_update_idle_time(cfqq->cfqg); |
812df48d DS |
2418 | } |
2419 | ||
febffd61 JA |
2420 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
2421 | struct cfq_queue *cfqq) | |
22e2c507 JA |
2422 | { |
2423 | if (cfqq) { | |
3bf10fea | 2424 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d", |
4d2ceea4 | 2425 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
155fead9 | 2426 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
62a37f6b JT |
2427 | cfqq->slice_start = 0; |
2428 | cfqq->dispatch_start = jiffies; | |
2429 | cfqq->allocated_slice = 0; | |
2430 | cfqq->slice_end = 0; | |
2431 | cfqq->slice_dispatch = 0; | |
2432 | cfqq->nr_sectors = 0; | |
2433 | ||
2434 | cfq_clear_cfqq_wait_request(cfqq); | |
2435 | cfq_clear_cfqq_must_dispatch(cfqq); | |
2436 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
2437 | cfq_clear_cfqq_fifo_expire(cfqq); | |
2438 | cfq_mark_cfqq_slice_new(cfqq); | |
2439 | ||
2440 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
2441 | } |
2442 | ||
2443 | cfqd->active_queue = cfqq; | |
2444 | } | |
2445 | ||
7b14e3b5 JA |
2446 | /* |
2447 | * current cfqq expired its slice (or was too idle), select new one | |
2448 | */ | |
2449 | static void | |
2450 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 2451 | bool timed_out) |
7b14e3b5 | 2452 | { |
7b679138 JA |
2453 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
2454 | ||
7b14e3b5 | 2455 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 2456 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 2457 | |
7b14e3b5 | 2458 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 2459 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 2460 | |
ae54abed SL |
2461 | /* |
2462 | * If this cfqq is shared between multiple processes, check to | |
2463 | * make sure that those processes are still issuing I/Os within | |
2464 | * the mean seek distance. If not, it may be time to break the | |
2465 | * queues apart again. | |
2466 | */ | |
2467 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
2468 | cfq_mark_cfqq_split_coop(cfqq); | |
2469 | ||
7b14e3b5 | 2470 | /* |
6084cdda | 2471 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 2472 | */ |
c553f8e3 SL |
2473 | if (timed_out) { |
2474 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 2475 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
2476 | else |
2477 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
2478 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
2479 | } | |
7b14e3b5 | 2480 | |
e5ff082e | 2481 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 2482 | |
f04a6424 VG |
2483 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
2484 | cfq_del_cfqq_rr(cfqd, cfqq); | |
2485 | ||
edd75ffd | 2486 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
2487 | |
2488 | if (cfqq == cfqd->active_queue) | |
2489 | cfqd->active_queue = NULL; | |
2490 | ||
2491 | if (cfqd->active_cic) { | |
11a3122f | 2492 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
2493 | cfqd->active_cic = NULL; |
2494 | } | |
7b14e3b5 JA |
2495 | } |
2496 | ||
e5ff082e | 2497 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
2498 | { |
2499 | struct cfq_queue *cfqq = cfqd->active_queue; | |
2500 | ||
2501 | if (cfqq) | |
e5ff082e | 2502 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
2503 | } |
2504 | ||
498d3aa2 JA |
2505 | /* |
2506 | * Get next queue for service. Unless we have a queue preemption, | |
2507 | * we'll simply select the first cfqq in the service tree. | |
2508 | */ | |
6d048f53 | 2509 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 2510 | { |
34b98d03 VG |
2511 | struct cfq_rb_root *st = st_for(cfqd->serving_group, |
2512 | cfqd->serving_wl_class, cfqd->serving_wl_type); | |
d9e7620e | 2513 | |
f04a6424 VG |
2514 | if (!cfqd->rq_queued) |
2515 | return NULL; | |
2516 | ||
1fa8f6d6 | 2517 | /* There is nothing to dispatch */ |
34b98d03 | 2518 | if (!st) |
1fa8f6d6 | 2519 | return NULL; |
34b98d03 | 2520 | if (RB_EMPTY_ROOT(&st->rb)) |
c0324a02 | 2521 | return NULL; |
34b98d03 | 2522 | return cfq_rb_first(st); |
6d048f53 JA |
2523 | } |
2524 | ||
f04a6424 VG |
2525 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
2526 | { | |
25fb5169 | 2527 | struct cfq_group *cfqg; |
f04a6424 VG |
2528 | struct cfq_queue *cfqq; |
2529 | int i, j; | |
2530 | struct cfq_rb_root *st; | |
2531 | ||
2532 | if (!cfqd->rq_queued) | |
2533 | return NULL; | |
2534 | ||
25fb5169 VG |
2535 | cfqg = cfq_get_next_cfqg(cfqd); |
2536 | if (!cfqg) | |
2537 | return NULL; | |
2538 | ||
f04a6424 VG |
2539 | for_each_cfqg_st(cfqg, i, j, st) |
2540 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
2541 | return cfqq; | |
2542 | return NULL; | |
2543 | } | |
2544 | ||
498d3aa2 JA |
2545 | /* |
2546 | * Get and set a new active queue for service. | |
2547 | */ | |
a36e71f9 JA |
2548 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
2549 | struct cfq_queue *cfqq) | |
6d048f53 | 2550 | { |
e00ef799 | 2551 | if (!cfqq) |
a36e71f9 | 2552 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 2553 | |
22e2c507 | 2554 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 2555 | return cfqq; |
22e2c507 JA |
2556 | } |
2557 | ||
d9e7620e JA |
2558 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
2559 | struct request *rq) | |
2560 | { | |
83096ebf TH |
2561 | if (blk_rq_pos(rq) >= cfqd->last_position) |
2562 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 2563 | else |
83096ebf | 2564 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
2565 | } |
2566 | ||
b2c18e1e | 2567 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 2568 | struct request *rq) |
6d048f53 | 2569 | { |
e9ce335d | 2570 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
2571 | } |
2572 | ||
a36e71f9 JA |
2573 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
2574 | struct cfq_queue *cur_cfqq) | |
2575 | { | |
f2d1f0ae | 2576 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
2577 | struct rb_node *parent, *node; |
2578 | struct cfq_queue *__cfqq; | |
2579 | sector_t sector = cfqd->last_position; | |
2580 | ||
2581 | if (RB_EMPTY_ROOT(root)) | |
2582 | return NULL; | |
2583 | ||
2584 | /* | |
2585 | * First, if we find a request starting at the end of the last | |
2586 | * request, choose it. | |
2587 | */ | |
f2d1f0ae | 2588 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
2589 | if (__cfqq) |
2590 | return __cfqq; | |
2591 | ||
2592 | /* | |
2593 | * If the exact sector wasn't found, the parent of the NULL leaf | |
2594 | * will contain the closest sector. | |
2595 | */ | |
2596 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 2597 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2598 | return __cfqq; |
2599 | ||
2e46e8b2 | 2600 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
2601 | node = rb_next(&__cfqq->p_node); |
2602 | else | |
2603 | node = rb_prev(&__cfqq->p_node); | |
2604 | if (!node) | |
2605 | return NULL; | |
2606 | ||
2607 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 2608 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
2609 | return __cfqq; |
2610 | ||
2611 | return NULL; | |
2612 | } | |
2613 | ||
2614 | /* | |
2615 | * cfqd - obvious | |
2616 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
2617 | * closely cooperating with itself. | |
2618 | * | |
2619 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
2620 | * one request, and that cfqd->last_position reflects a position on the disk | |
2621 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
2622 | * assumption. | |
2623 | */ | |
2624 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 2625 | struct cfq_queue *cur_cfqq) |
6d048f53 | 2626 | { |
a36e71f9 JA |
2627 | struct cfq_queue *cfqq; |
2628 | ||
39c01b21 DS |
2629 | if (cfq_class_idle(cur_cfqq)) |
2630 | return NULL; | |
e6c5bc73 JM |
2631 | if (!cfq_cfqq_sync(cur_cfqq)) |
2632 | return NULL; | |
2633 | if (CFQQ_SEEKY(cur_cfqq)) | |
2634 | return NULL; | |
2635 | ||
b9d8f4c7 GJ |
2636 | /* |
2637 | * Don't search priority tree if it's the only queue in the group. | |
2638 | */ | |
2639 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
2640 | return NULL; | |
2641 | ||
6d048f53 | 2642 | /* |
d9e7620e JA |
2643 | * We should notice if some of the queues are cooperating, eg |
2644 | * working closely on the same area of the disk. In that case, | |
2645 | * we can group them together and don't waste time idling. | |
6d048f53 | 2646 | */ |
a36e71f9 JA |
2647 | cfqq = cfqq_close(cfqd, cur_cfqq); |
2648 | if (!cfqq) | |
2649 | return NULL; | |
2650 | ||
8682e1f1 VG |
2651 | /* If new queue belongs to different cfq_group, don't choose it */ |
2652 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
2653 | return NULL; | |
2654 | ||
df5fe3e8 JM |
2655 | /* |
2656 | * It only makes sense to merge sync queues. | |
2657 | */ | |
2658 | if (!cfq_cfqq_sync(cfqq)) | |
2659 | return NULL; | |
e6c5bc73 JM |
2660 | if (CFQQ_SEEKY(cfqq)) |
2661 | return NULL; | |
df5fe3e8 | 2662 | |
c0324a02 CZ |
2663 | /* |
2664 | * Do not merge queues of different priority classes | |
2665 | */ | |
2666 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
2667 | return NULL; | |
2668 | ||
a36e71f9 | 2669 | return cfqq; |
6d048f53 JA |
2670 | } |
2671 | ||
a6d44e98 CZ |
2672 | /* |
2673 | * Determine whether we should enforce idle window for this queue. | |
2674 | */ | |
2675 | ||
2676 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2677 | { | |
3bf10fea | 2678 | enum wl_class_t wl_class = cfqq_class(cfqq); |
34b98d03 | 2679 | struct cfq_rb_root *st = cfqq->service_tree; |
a6d44e98 | 2680 | |
34b98d03 VG |
2681 | BUG_ON(!st); |
2682 | BUG_ON(!st->count); | |
f04a6424 | 2683 | |
b6508c16 VG |
2684 | if (!cfqd->cfq_slice_idle) |
2685 | return false; | |
2686 | ||
a6d44e98 | 2687 | /* We never do for idle class queues. */ |
3bf10fea | 2688 | if (wl_class == IDLE_WORKLOAD) |
a6d44e98 CZ |
2689 | return false; |
2690 | ||
3c2a0909 S |
2691 | if (cfqd->cfq_rt_idle_only && wl_class != RT_WORKLOAD) |
2692 | return false; | |
2693 | ||
a6d44e98 | 2694 | /* We do for queues that were marked with idle window flag. */ |
3c764b7a SL |
2695 | if (cfq_cfqq_idle_window(cfqq) && |
2696 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
2697 | return true; |
2698 | ||
2699 | /* | |
2700 | * Otherwise, we do only if they are the last ones | |
2701 | * in their service tree. | |
2702 | */ | |
34b98d03 VG |
2703 | if (st->count == 1 && cfq_cfqq_sync(cfqq) && |
2704 | !cfq_io_thinktime_big(cfqd, &st->ttime, false)) | |
c1e44756 | 2705 | return true; |
34b98d03 | 2706 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count); |
c1e44756 | 2707 | return false; |
a6d44e98 CZ |
2708 | } |
2709 | ||
6d048f53 | 2710 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2711 | { |
1792669c | 2712 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 2713 | struct cfq_io_cq *cic; |
80bdf0c7 | 2714 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 2715 | |
a68bbddb | 2716 | /* |
f7d7b7a7 JA |
2717 | * SSD device without seek penalty, disable idling. But only do so |
2718 | * for devices that support queuing, otherwise we still have a problem | |
2719 | * with sync vs async workloads. | |
a68bbddb | 2720 | */ |
f7d7b7a7 | 2721 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
2722 | return; |
2723 | ||
dd67d051 | 2724 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2725 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2726 | |
2727 | /* | |
2728 | * idle is disabled, either manually or by past process history | |
2729 | */ | |
80bdf0c7 VG |
2730 | if (!cfq_should_idle(cfqd, cfqq)) { |
2731 | /* no queue idling. Check for group idling */ | |
2732 | if (cfqd->cfq_group_idle) | |
2733 | group_idle = cfqd->cfq_group_idle; | |
2734 | else | |
2735 | return; | |
2736 | } | |
6d048f53 | 2737 | |
7b679138 | 2738 | /* |
8e550632 | 2739 | * still active requests from this queue, don't idle |
7b679138 | 2740 | */ |
8e550632 | 2741 | if (cfqq->dispatched) |
7b679138 JA |
2742 | return; |
2743 | ||
22e2c507 JA |
2744 | /* |
2745 | * task has exited, don't wait | |
2746 | */ | |
206dc69b | 2747 | cic = cfqd->active_cic; |
f6e8d01b | 2748 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
2749 | return; |
2750 | ||
355b659c CZ |
2751 | /* |
2752 | * If our average think time is larger than the remaining time | |
2753 | * slice, then don't idle. This avoids overrunning the allotted | |
2754 | * time slice. | |
2755 | */ | |
383cd721 SL |
2756 | if (sample_valid(cic->ttime.ttime_samples) && |
2757 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 2758 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 2759 | cic->ttime.ttime_mean); |
355b659c | 2760 | return; |
b1ffe737 | 2761 | } |
355b659c | 2762 | |
80bdf0c7 VG |
2763 | /* There are other queues in the group, don't do group idle */ |
2764 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
2765 | return; | |
2766 | ||
3b18152c | 2767 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2768 | |
80bdf0c7 VG |
2769 | if (group_idle) |
2770 | sl = cfqd->cfq_group_idle; | |
2771 | else | |
2772 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2773 | |
7b14e3b5 | 2774 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
155fead9 | 2775 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
80bdf0c7 VG |
2776 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
2777 | group_idle ? 1 : 0); | |
1da177e4 LT |
2778 | } |
2779 | ||
498d3aa2 JA |
2780 | /* |
2781 | * Move request from internal lists to the request queue dispatch list. | |
2782 | */ | |
165125e1 | 2783 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2784 | { |
3ed9a296 | 2785 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2786 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2787 | |
7b679138 JA |
2788 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2789 | ||
06d21886 | 2790 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2791 | cfq_remove_request(rq); |
6d048f53 | 2792 | cfqq->dispatched++; |
80bdf0c7 | 2793 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2794 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2795 | |
53c583d2 | 2796 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2797 | cfqq->nr_sectors += blk_rq_sectors(rq); |
155fead9 | 2798 | cfqg_stats_update_dispatch(cfqq->cfqg, blk_rq_bytes(rq), rq->cmd_flags); |
1da177e4 LT |
2799 | } |
2800 | ||
2801 | /* | |
2802 | * return expired entry, or NULL to just start from scratch in rbtree | |
2803 | */ | |
febffd61 | 2804 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 2805 | { |
30996f40 | 2806 | struct request *rq = NULL; |
1da177e4 | 2807 | |
3b18152c | 2808 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 2809 | return NULL; |
cb887411 JA |
2810 | |
2811 | cfq_mark_cfqq_fifo_expire(cfqq); | |
2812 | ||
89850f7e JA |
2813 | if (list_empty(&cfqq->fifo)) |
2814 | return NULL; | |
1da177e4 | 2815 | |
89850f7e | 2816 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 2817 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 2818 | rq = NULL; |
1da177e4 | 2819 | |
6d048f53 | 2820 | return rq; |
1da177e4 LT |
2821 | } |
2822 | ||
22e2c507 JA |
2823 | static inline int |
2824 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2825 | { | |
2826 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2827 | |
22e2c507 | 2828 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2829 | |
b9f8ce05 | 2830 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
2831 | } |
2832 | ||
df5fe3e8 JM |
2833 | /* |
2834 | * Must be called with the queue_lock held. | |
2835 | */ | |
2836 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2837 | { | |
2838 | int process_refs, io_refs; | |
2839 | ||
2840 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 2841 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
2842 | BUG_ON(process_refs < 0); |
2843 | return process_refs; | |
2844 | } | |
2845 | ||
2846 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2847 | { | |
e6c5bc73 | 2848 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2849 | struct cfq_queue *__cfqq; |
2850 | ||
c10b61f0 JM |
2851 | /* |
2852 | * If there are no process references on the new_cfqq, then it is | |
2853 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2854 | * chain may have dropped their last reference (not just their | |
2855 | * last process reference). | |
2856 | */ | |
2857 | if (!cfqq_process_refs(new_cfqq)) | |
2858 | return; | |
2859 | ||
df5fe3e8 JM |
2860 | /* Avoid a circular list and skip interim queue merges */ |
2861 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2862 | if (__cfqq == cfqq) | |
2863 | return; | |
2864 | new_cfqq = __cfqq; | |
2865 | } | |
2866 | ||
2867 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2868 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2869 | /* |
2870 | * If the process for the cfqq has gone away, there is no | |
2871 | * sense in merging the queues. | |
2872 | */ | |
c10b61f0 | 2873 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2874 | return; |
2875 | ||
e6c5bc73 JM |
2876 | /* |
2877 | * Merge in the direction of the lesser amount of work. | |
2878 | */ | |
e6c5bc73 JM |
2879 | if (new_process_refs >= process_refs) { |
2880 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2881 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2882 | } else { |
2883 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2884 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2885 | } |
df5fe3e8 JM |
2886 | } |
2887 | ||
6d816ec7 | 2888 | static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd, |
3bf10fea | 2889 | struct cfq_group *cfqg, enum wl_class_t wl_class) |
718eee05 CZ |
2890 | { |
2891 | struct cfq_queue *queue; | |
2892 | int i; | |
2893 | bool key_valid = false; | |
2894 | unsigned long lowest_key = 0; | |
2895 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2896 | ||
65b32a57 VG |
2897 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2898 | /* select the one with lowest rb_key */ | |
34b98d03 | 2899 | queue = cfq_rb_first(st_for(cfqg, wl_class, i)); |
718eee05 CZ |
2900 | if (queue && |
2901 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2902 | lowest_key = queue->rb_key; | |
2903 | cur_best = i; | |
2904 | key_valid = true; | |
2905 | } | |
2906 | } | |
2907 | ||
2908 | return cur_best; | |
2909 | } | |
2910 | ||
6d816ec7 VG |
2911 | static void |
2912 | choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
718eee05 | 2913 | { |
718eee05 CZ |
2914 | unsigned slice; |
2915 | unsigned count; | |
cdb16e8f | 2916 | struct cfq_rb_root *st; |
58ff82f3 | 2917 | unsigned group_slice; |
4d2ceea4 | 2918 | enum wl_class_t original_class = cfqd->serving_wl_class; |
1fa8f6d6 | 2919 | |
718eee05 | 2920 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2921 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 2922 | cfqd->serving_wl_class = RT_WORKLOAD; |
58ff82f3 | 2923 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
4d2ceea4 | 2924 | cfqd->serving_wl_class = BE_WORKLOAD; |
718eee05 | 2925 | else { |
4d2ceea4 | 2926 | cfqd->serving_wl_class = IDLE_WORKLOAD; |
718eee05 CZ |
2927 | cfqd->workload_expires = jiffies + 1; |
2928 | return; | |
2929 | } | |
2930 | ||
4d2ceea4 | 2931 | if (original_class != cfqd->serving_wl_class) |
e4ea0c16 SL |
2932 | goto new_workload; |
2933 | ||
718eee05 CZ |
2934 | /* |
2935 | * For RT and BE, we have to choose also the type | |
2936 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2937 | * expiration time | |
2938 | */ | |
34b98d03 | 2939 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 2940 | count = st->count; |
718eee05 CZ |
2941 | |
2942 | /* | |
65b32a57 | 2943 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2944 | */ |
65b32a57 | 2945 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2946 | return; |
2947 | ||
e4ea0c16 | 2948 | new_workload: |
718eee05 | 2949 | /* otherwise select new workload type */ |
6d816ec7 | 2950 | cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg, |
4d2ceea4 | 2951 | cfqd->serving_wl_class); |
34b98d03 | 2952 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
cdb16e8f | 2953 | count = st->count; |
718eee05 CZ |
2954 | |
2955 | /* | |
2956 | * the workload slice is computed as a fraction of target latency | |
2957 | * proportional to the number of queues in that workload, over | |
2958 | * all the queues in the same priority class | |
2959 | */ | |
58ff82f3 VG |
2960 | group_slice = cfq_group_slice(cfqd, cfqg); |
2961 | ||
2962 | slice = group_slice * count / | |
4d2ceea4 VG |
2963 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class], |
2964 | cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd, | |
3bf10fea | 2965 | cfqg)); |
718eee05 | 2966 | |
4d2ceea4 | 2967 | if (cfqd->serving_wl_type == ASYNC_WORKLOAD) { |
f26bd1f0 VG |
2968 | unsigned int tmp; |
2969 | ||
2970 | /* | |
2971 | * Async queues are currently system wide. Just taking | |
2972 | * proportion of queues with-in same group will lead to higher | |
2973 | * async ratio system wide as generally root group is going | |
2974 | * to have higher weight. A more accurate thing would be to | |
2975 | * calculate system wide asnc/sync ratio. | |
2976 | */ | |
5bf14c07 TM |
2977 | tmp = cfqd->cfq_target_latency * |
2978 | cfqg_busy_async_queues(cfqd, cfqg); | |
f26bd1f0 VG |
2979 | tmp = tmp/cfqd->busy_queues; |
2980 | slice = min_t(unsigned, slice, tmp); | |
2981 | ||
718eee05 CZ |
2982 | /* async workload slice is scaled down according to |
2983 | * the sync/async slice ratio. */ | |
2984 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2985 | } else |
718eee05 CZ |
2986 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2987 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2988 | ||
2989 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2990 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
2991 | cfqd->workload_expires = jiffies + slice; |
2992 | } | |
2993 | ||
1fa8f6d6 VG |
2994 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2995 | { | |
2996 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2997 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2998 | |
2999 | if (RB_EMPTY_ROOT(&st->rb)) | |
3000 | return NULL; | |
25bc6b07 | 3001 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
3002 | update_min_vdisktime(st); |
3003 | return cfqg; | |
1fa8f6d6 VG |
3004 | } |
3005 | ||
cdb16e8f VG |
3006 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
3007 | { | |
1fa8f6d6 VG |
3008 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
3009 | ||
3010 | cfqd->serving_group = cfqg; | |
dae739eb VG |
3011 | |
3012 | /* Restore the workload type data */ | |
4d2ceea4 VG |
3013 | if (cfqg->saved_wl_slice) { |
3014 | cfqd->workload_expires = jiffies + cfqg->saved_wl_slice; | |
3015 | cfqd->serving_wl_type = cfqg->saved_wl_type; | |
3016 | cfqd->serving_wl_class = cfqg->saved_wl_class; | |
66ae2919 GJ |
3017 | } else |
3018 | cfqd->workload_expires = jiffies - 1; | |
3019 | ||
6d816ec7 | 3020 | choose_wl_class_and_type(cfqd, cfqg); |
cdb16e8f VG |
3021 | } |
3022 | ||
22e2c507 | 3023 | /* |
498d3aa2 JA |
3024 | * Select a queue for service. If we have a current active queue, |
3025 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 3026 | */ |
1b5ed5e1 | 3027 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 3028 | { |
a36e71f9 | 3029 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 3030 | |
22e2c507 JA |
3031 | cfqq = cfqd->active_queue; |
3032 | if (!cfqq) | |
3033 | goto new_queue; | |
1da177e4 | 3034 | |
f04a6424 VG |
3035 | if (!cfqd->rq_queued) |
3036 | return NULL; | |
c244bb50 VG |
3037 | |
3038 | /* | |
3039 | * We were waiting for group to get backlogged. Expire the queue | |
3040 | */ | |
3041 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3042 | goto expire; | |
3043 | ||
22e2c507 | 3044 | /* |
6d048f53 | 3045 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 3046 | */ |
7667aa06 VG |
3047 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
3048 | /* | |
3049 | * If slice had not expired at the completion of last request | |
3050 | * we might not have turned on wait_busy flag. Don't expire | |
3051 | * the queue yet. Allow the group to get backlogged. | |
3052 | * | |
3053 | * The very fact that we have used the slice, that means we | |
3054 | * have been idling all along on this queue and it should be | |
3055 | * ok to wait for this request to complete. | |
3056 | */ | |
82bbbf28 VG |
3057 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
3058 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
3059 | cfqq = NULL; | |
7667aa06 | 3060 | goto keep_queue; |
82bbbf28 | 3061 | } else |
80bdf0c7 | 3062 | goto check_group_idle; |
7667aa06 | 3063 | } |
1da177e4 | 3064 | |
22e2c507 | 3065 | /* |
6d048f53 JA |
3066 | * The active queue has requests and isn't expired, allow it to |
3067 | * dispatch. | |
22e2c507 | 3068 | */ |
dd67d051 | 3069 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3070 | goto keep_queue; |
6d048f53 | 3071 | |
a36e71f9 JA |
3072 | /* |
3073 | * If another queue has a request waiting within our mean seek | |
3074 | * distance, let it run. The expire code will check for close | |
3075 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 3076 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 3077 | */ |
b3b6d040 | 3078 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
3079 | if (new_cfqq) { |
3080 | if (!cfqq->new_cfqq) | |
3081 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 3082 | goto expire; |
df5fe3e8 | 3083 | } |
a36e71f9 | 3084 | |
6d048f53 JA |
3085 | /* |
3086 | * No requests pending. If the active queue still has requests in | |
3087 | * flight or is idling for a new request, allow either of these | |
3088 | * conditions to happen (or time out) before selecting a new queue. | |
3089 | */ | |
80bdf0c7 VG |
3090 | if (timer_pending(&cfqd->idle_slice_timer)) { |
3091 | cfqq = NULL; | |
3092 | goto keep_queue; | |
3093 | } | |
3094 | ||
8e1ac665 SL |
3095 | /* |
3096 | * This is a deep seek queue, but the device is much faster than | |
3097 | * the queue can deliver, don't idle | |
3098 | **/ | |
3099 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
3100 | (cfq_cfqq_slice_new(cfqq) || | |
3101 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
3102 | cfq_clear_cfqq_deep(cfqq); | |
3103 | cfq_clear_cfqq_idle_window(cfqq); | |
3104 | } | |
3105 | ||
80bdf0c7 VG |
3106 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
3107 | cfqq = NULL; | |
3108 | goto keep_queue; | |
3109 | } | |
3110 | ||
3111 | /* | |
3112 | * If group idle is enabled and there are requests dispatched from | |
3113 | * this group, wait for requests to complete. | |
3114 | */ | |
3115 | check_group_idle: | |
7700fc4f SL |
3116 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
3117 | cfqq->cfqg->dispatched && | |
3118 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
3119 | cfqq = NULL; |
3120 | goto keep_queue; | |
22e2c507 JA |
3121 | } |
3122 | ||
3b18152c | 3123 | expire: |
e5ff082e | 3124 | cfq_slice_expired(cfqd, 0); |
3b18152c | 3125 | new_queue: |
718eee05 CZ |
3126 | /* |
3127 | * Current queue expired. Check if we have to switch to a new | |
3128 | * service tree | |
3129 | */ | |
3130 | if (!new_cfqq) | |
cdb16e8f | 3131 | cfq_choose_cfqg(cfqd); |
718eee05 | 3132 | |
a36e71f9 | 3133 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 3134 | keep_queue: |
3b18152c | 3135 | return cfqq; |
22e2c507 JA |
3136 | } |
3137 | ||
febffd61 | 3138 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
3139 | { |
3140 | int dispatched = 0; | |
3141 | ||
3142 | while (cfqq->next_rq) { | |
3143 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
3144 | dispatched++; | |
3145 | } | |
3146 | ||
3147 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
3148 | |
3149 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 3150 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
3151 | return dispatched; |
3152 | } | |
3153 | ||
498d3aa2 JA |
3154 | /* |
3155 | * Drain our current requests. Used for barriers and when switching | |
3156 | * io schedulers on-the-fly. | |
3157 | */ | |
d9e7620e | 3158 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 3159 | { |
0871714e | 3160 | struct cfq_queue *cfqq; |
d9e7620e | 3161 | int dispatched = 0; |
cdb16e8f | 3162 | |
3440c49f | 3163 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 3164 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
3165 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
3166 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 3167 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 3168 | } |
1b5ed5e1 | 3169 | |
1b5ed5e1 TH |
3170 | BUG_ON(cfqd->busy_queues); |
3171 | ||
6923715a | 3172 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
3173 | return dispatched; |
3174 | } | |
3175 | ||
abc3c744 SL |
3176 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
3177 | struct cfq_queue *cfqq) | |
3178 | { | |
3179 | /* the queue hasn't finished any request, can't estimate */ | |
3180 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 3181 | return true; |
abc3c744 SL |
3182 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
3183 | cfqq->slice_end)) | |
c1e44756 | 3184 | return true; |
abc3c744 | 3185 | |
c1e44756 | 3186 | return false; |
abc3c744 SL |
3187 | } |
3188 | ||
0b182d61 | 3189 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 3190 | { |
2f5cb738 | 3191 | unsigned int max_dispatch; |
22e2c507 | 3192 | |
3c2a0909 S |
3193 | if (cfq_cfqq_must_dispatch(cfqq)) |
3194 | return true; | |
3195 | ||
5ad531db JA |
3196 | /* |
3197 | * Drain async requests before we start sync IO | |
3198 | */ | |
53c583d2 | 3199 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 3200 | return false; |
5ad531db | 3201 | |
2f5cb738 JA |
3202 | /* |
3203 | * If this is an async queue and we have sync IO in flight, let it wait | |
3204 | */ | |
53c583d2 | 3205 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 3206 | return false; |
2f5cb738 | 3207 | |
abc3c744 | 3208 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
3209 | if (cfq_class_idle(cfqq)) |
3210 | max_dispatch = 1; | |
b4878f24 | 3211 | |
2f5cb738 JA |
3212 | /* |
3213 | * Does this cfqq already have too much IO in flight? | |
3214 | */ | |
3215 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 3216 | bool promote_sync = false; |
2f5cb738 JA |
3217 | /* |
3218 | * idle queue must always only have a single IO in flight | |
3219 | */ | |
3ed9a296 | 3220 | if (cfq_class_idle(cfqq)) |
0b182d61 | 3221 | return false; |
3ed9a296 | 3222 | |
ef8a41df | 3223 | /* |
c4ade94f LS |
3224 | * If there is only one sync queue |
3225 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
3226 | * queue no dispatch limit. The reason is a sync queue can |
3227 | * preempt async queue, limiting the sync queue doesn't make | |
3228 | * sense. This is useful for aiostress test. | |
3229 | */ | |
c4ade94f LS |
3230 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
3231 | promote_sync = true; | |
ef8a41df | 3232 | |
2f5cb738 JA |
3233 | /* |
3234 | * We have other queues, don't allow more IO from this one | |
3235 | */ | |
ef8a41df SL |
3236 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
3237 | !promote_sync) | |
0b182d61 | 3238 | return false; |
9ede209e | 3239 | |
365722bb | 3240 | /* |
474b18cc | 3241 | * Sole queue user, no limit |
365722bb | 3242 | */ |
ef8a41df | 3243 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
3244 | max_dispatch = -1; |
3245 | else | |
3246 | /* | |
3247 | * Normally we start throttling cfqq when cfq_quantum/2 | |
3248 | * requests have been dispatched. But we can drive | |
3249 | * deeper queue depths at the beginning of slice | |
3250 | * subjected to upper limit of cfq_quantum. | |
3251 | * */ | |
3252 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
3253 | } |
3254 | ||
3255 | /* | |
3256 | * Async queues must wait a bit before being allowed dispatch. | |
3257 | * We also ramp up the dispatch depth gradually for async IO, | |
3258 | * based on the last sync IO we serviced | |
3259 | */ | |
963b72fc | 3260 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 3261 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 3262 | unsigned int depth; |
365722bb | 3263 | |
61f0c1dc | 3264 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
3265 | if (!depth && !cfqq->dispatched) |
3266 | depth = 1; | |
8e296755 JA |
3267 | if (depth < max_dispatch) |
3268 | max_dispatch = depth; | |
2f5cb738 | 3269 | } |
3ed9a296 | 3270 | |
0b182d61 JA |
3271 | /* |
3272 | * If we're below the current max, allow a dispatch | |
3273 | */ | |
3274 | return cfqq->dispatched < max_dispatch; | |
3275 | } | |
3276 | ||
3277 | /* | |
3278 | * Dispatch a request from cfqq, moving them to the request queue | |
3279 | * dispatch list. | |
3280 | */ | |
3281 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3282 | { | |
3283 | struct request *rq; | |
3284 | ||
3285 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
3286 | ||
3c2a0909 S |
3287 | rq = cfq_check_fifo(cfqq); |
3288 | if (rq) | |
3289 | cfq_mark_cfqq_must_dispatch(cfqq); | |
3290 | ||
0b182d61 JA |
3291 | if (!cfq_may_dispatch(cfqd, cfqq)) |
3292 | return false; | |
3293 | ||
3294 | /* | |
3295 | * follow expired path, else get first next available | |
3296 | */ | |
0b182d61 JA |
3297 | if (!rq) |
3298 | rq = cfqq->next_rq; | |
3c2a0909 S |
3299 | else |
3300 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); | |
0b182d61 JA |
3301 | |
3302 | /* | |
3303 | * insert request into driver dispatch list | |
3304 | */ | |
3305 | cfq_dispatch_insert(cfqd->queue, rq); | |
3306 | ||
3307 | if (!cfqd->active_cic) { | |
c5869807 | 3308 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 3309 | |
c5869807 | 3310 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
3311 | cfqd->active_cic = cic; |
3312 | } | |
3313 | ||
3314 | return true; | |
3315 | } | |
3316 | ||
3317 | /* | |
3318 | * Find the cfqq that we need to service and move a request from that to the | |
3319 | * dispatch list | |
3320 | */ | |
3321 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
3322 | { | |
3323 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3324 | struct cfq_queue *cfqq; | |
3325 | ||
3326 | if (!cfqd->busy_queues) | |
3327 | return 0; | |
3328 | ||
3329 | if (unlikely(force)) | |
3330 | return cfq_forced_dispatch(cfqd); | |
3331 | ||
3332 | cfqq = cfq_select_queue(cfqd); | |
3333 | if (!cfqq) | |
8e296755 JA |
3334 | return 0; |
3335 | ||
2f5cb738 | 3336 | /* |
0b182d61 | 3337 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 3338 | */ |
0b182d61 JA |
3339 | if (!cfq_dispatch_request(cfqd, cfqq)) |
3340 | return 0; | |
3341 | ||
2f5cb738 | 3342 | cfqq->slice_dispatch++; |
b029195d | 3343 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 3344 | |
2f5cb738 JA |
3345 | /* |
3346 | * expire an async queue immediately if it has used up its slice. idle | |
3347 | * queue always expire after 1 dispatch round. | |
3348 | */ | |
3349 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
3350 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
3351 | cfq_class_idle(cfqq))) { | |
3352 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 3353 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
3354 | } |
3355 | ||
b217a903 | 3356 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 3357 | return 1; |
1da177e4 LT |
3358 | } |
3359 | ||
1da177e4 | 3360 | /* |
5e705374 JA |
3361 | * task holds one reference to the queue, dropped when task exits. each rq |
3362 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 3363 | * |
b1c35769 | 3364 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
3365 | * queue lock must be held here. |
3366 | */ | |
3367 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
3368 | { | |
22e2c507 | 3369 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 3370 | struct cfq_group *cfqg; |
22e2c507 | 3371 | |
30d7b944 | 3372 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 3373 | |
30d7b944 SL |
3374 | cfqq->ref--; |
3375 | if (cfqq->ref) | |
1da177e4 LT |
3376 | return; |
3377 | ||
7b679138 | 3378 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 3379 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 3380 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 3381 | cfqg = cfqq->cfqg; |
1da177e4 | 3382 | |
28f95cbc | 3383 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 3384 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 3385 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 3386 | } |
22e2c507 | 3387 | |
f04a6424 | 3388 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 3389 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 3390 | cfqg_put(cfqg); |
1da177e4 LT |
3391 | } |
3392 | ||
d02a2c07 | 3393 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 3394 | { |
df5fe3e8 JM |
3395 | struct cfq_queue *__cfqq, *next; |
3396 | ||
df5fe3e8 JM |
3397 | /* |
3398 | * If this queue was scheduled to merge with another queue, be | |
3399 | * sure to drop the reference taken on that queue (and others in | |
3400 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
3401 | */ | |
3402 | __cfqq = cfqq->new_cfqq; | |
3403 | while (__cfqq) { | |
3404 | if (__cfqq == cfqq) { | |
3405 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
3406 | break; | |
3407 | } | |
3408 | next = __cfqq->new_cfqq; | |
3409 | cfq_put_queue(__cfqq); | |
3410 | __cfqq = next; | |
3411 | } | |
d02a2c07 SL |
3412 | } |
3413 | ||
3414 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3415 | { | |
3416 | if (unlikely(cfqq == cfqd->active_queue)) { | |
3417 | __cfq_slice_expired(cfqd, cfqq, 0); | |
3418 | cfq_schedule_dispatch(cfqd); | |
3419 | } | |
3420 | ||
3421 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 3422 | |
89850f7e JA |
3423 | cfq_put_queue(cfqq); |
3424 | } | |
22e2c507 | 3425 | |
9b84cacd TH |
3426 | static void cfq_init_icq(struct io_cq *icq) |
3427 | { | |
3428 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
3429 | ||
3430 | cic->ttime.last_end_request = jiffies; | |
3431 | } | |
3432 | ||
c5869807 | 3433 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 3434 | { |
c5869807 | 3435 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 3436 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 3437 | |
ff6657c6 JA |
3438 | if (cic->cfqq[BLK_RW_ASYNC]) { |
3439 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
3440 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
3441 | } |
3442 | ||
ff6657c6 JA |
3443 | if (cic->cfqq[BLK_RW_SYNC]) { |
3444 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
3445 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 3446 | } |
89850f7e JA |
3447 | } |
3448 | ||
abede6da | 3449 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
3450 | { |
3451 | struct task_struct *tsk = current; | |
3452 | int ioprio_class; | |
3453 | ||
3b18152c | 3454 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
3455 | return; |
3456 | ||
598971bf | 3457 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 3458 | switch (ioprio_class) { |
fe094d98 JA |
3459 | default: |
3460 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
3461 | case IOPRIO_CLASS_NONE: | |
3462 | /* | |
6d63c275 | 3463 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
3464 | */ |
3465 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 3466 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
3467 | break; |
3468 | case IOPRIO_CLASS_RT: | |
598971bf | 3469 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3470 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
3471 | break; | |
3472 | case IOPRIO_CLASS_BE: | |
598971bf | 3473 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
3474 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
3475 | break; | |
3476 | case IOPRIO_CLASS_IDLE: | |
3477 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
3478 | cfqq->ioprio = 7; | |
3479 | cfq_clear_cfqq_idle_window(cfqq); | |
3480 | break; | |
22e2c507 JA |
3481 | } |
3482 | ||
3483 | /* | |
3484 | * keep track of original prio settings in case we have to temporarily | |
3485 | * elevate the priority of this queue | |
3486 | */ | |
3487 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 3488 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
3489 | } |
3490 | ||
598971bf | 3491 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 3492 | { |
598971bf | 3493 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 3494 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 3495 | struct cfq_queue *cfqq; |
35e6077c | 3496 | |
598971bf TH |
3497 | /* |
3498 | * Check whether ioprio has changed. The condition may trigger | |
3499 | * spuriously on a newly created cic but there's no harm. | |
3500 | */ | |
3501 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
3502 | return; |
3503 | ||
ff6657c6 | 3504 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
3505 | if (cfqq) { |
3506 | struct cfq_queue *new_cfqq; | |
abede6da TH |
3507 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio, |
3508 | GFP_ATOMIC); | |
caaa5f9f | 3509 | if (new_cfqq) { |
ff6657c6 | 3510 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
3511 | cfq_put_queue(cfqq); |
3512 | } | |
22e2c507 | 3513 | } |
caaa5f9f | 3514 | |
ff6657c6 | 3515 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
3516 | if (cfqq) |
3517 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
3518 | |
3519 | cic->ioprio = ioprio; | |
22e2c507 JA |
3520 | } |
3521 | ||
d5036d77 | 3522 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 3523 | pid_t pid, bool is_sync) |
d5036d77 JA |
3524 | { |
3525 | RB_CLEAR_NODE(&cfqq->rb_node); | |
3526 | RB_CLEAR_NODE(&cfqq->p_node); | |
3527 | INIT_LIST_HEAD(&cfqq->fifo); | |
3528 | ||
30d7b944 | 3529 | cfqq->ref = 0; |
d5036d77 JA |
3530 | cfqq->cfqd = cfqd; |
3531 | ||
3532 | cfq_mark_cfqq_prio_changed(cfqq); | |
3533 | ||
3534 | if (is_sync) { | |
3c2a0909 | 3535 | if (!cfq_class_idle(cfqq) && (!cfqd->cfq_rt_idle_only || cfq_class_rt(cfqq))) |
d5036d77 JA |
3536 | cfq_mark_cfqq_idle_window(cfqq); |
3537 | cfq_mark_cfqq_sync(cfqq); | |
3538 | } | |
3539 | cfqq->pid = pid; | |
3540 | } | |
3541 | ||
24610333 | 3542 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
598971bf | 3543 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 3544 | { |
bca4b914 | 3545 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
598971bf TH |
3546 | struct cfq_queue *sync_cfqq; |
3547 | uint64_t id; | |
24610333 | 3548 | |
598971bf | 3549 | rcu_read_lock(); |
3c798398 | 3550 | id = bio_blkcg(bio)->id; |
598971bf | 3551 | rcu_read_unlock(); |
24610333 | 3552 | |
598971bf TH |
3553 | /* |
3554 | * Check whether blkcg has changed. The condition may trigger | |
3555 | * spuriously on a newly created cic but there's no harm. | |
3556 | */ | |
3557 | if (unlikely(!cfqd) || likely(cic->blkcg_id == id)) | |
3558 | return; | |
24610333 | 3559 | |
598971bf | 3560 | sync_cfqq = cic_to_cfqq(cic, 1); |
24610333 VG |
3561 | if (sync_cfqq) { |
3562 | /* | |
3563 | * Drop reference to sync queue. A new sync queue will be | |
3564 | * assigned in new group upon arrival of a fresh request. | |
3565 | */ | |
3566 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
3567 | cic_set_cfqq(cic, NULL, 1); | |
3568 | cfq_put_queue(sync_cfqq); | |
3569 | } | |
598971bf TH |
3570 | |
3571 | cic->blkcg_id = id; | |
24610333 | 3572 | } |
598971bf TH |
3573 | #else |
3574 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } | |
24610333 VG |
3575 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
3576 | ||
22e2c507 | 3577 | static struct cfq_queue * |
abede6da TH |
3578 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
3579 | struct bio *bio, gfp_t gfp_mask) | |
22e2c507 | 3580 | { |
3c798398 | 3581 | struct blkcg *blkcg; |
22e2c507 | 3582 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
cdb16e8f | 3583 | struct cfq_group *cfqg; |
22e2c507 JA |
3584 | |
3585 | retry: | |
2a7f1244 TH |
3586 | rcu_read_lock(); |
3587 | ||
3c798398 | 3588 | blkcg = bio_blkcg(bio); |
cd1604fa | 3589 | cfqg = cfq_lookup_create_cfqg(cfqd, blkcg); |
e1d6c219 KK |
3590 | if (!cfqg) { |
3591 | cfqq = &cfqd->oom_cfqq; | |
3592 | goto out; | |
3593 | } | |
3594 | ||
91fac317 | 3595 | cfqq = cic_to_cfqq(cic, is_sync); |
22e2c507 | 3596 | |
6118b70b JA |
3597 | /* |
3598 | * Always try a new alloc if we fell back to the OOM cfqq | |
3599 | * originally, since it should just be a temporary situation. | |
3600 | */ | |
3601 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
3602 | cfqq = NULL; | |
22e2c507 JA |
3603 | if (new_cfqq) { |
3604 | cfqq = new_cfqq; | |
3605 | new_cfqq = NULL; | |
3606 | } else if (gfp_mask & __GFP_WAIT) { | |
2a7f1244 | 3607 | rcu_read_unlock(); |
22e2c507 | 3608 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c | 3609 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 3610 | gfp_mask | __GFP_ZERO, |
94f6030c | 3611 | cfqd->queue->node); |
22e2c507 | 3612 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
3613 | if (new_cfqq) |
3614 | goto retry; | |
a3cc86c2 GC |
3615 | else |
3616 | return &cfqd->oom_cfqq; | |
22e2c507 | 3617 | } else { |
94f6030c CL |
3618 | cfqq = kmem_cache_alloc_node(cfq_pool, |
3619 | gfp_mask | __GFP_ZERO, | |
3620 | cfqd->queue->node); | |
22e2c507 JA |
3621 | } |
3622 | ||
6118b70b JA |
3623 | if (cfqq) { |
3624 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
abede6da | 3625 | cfq_init_prio_data(cfqq, cic); |
cdb16e8f | 3626 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
3627 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
3628 | } else | |
3629 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 | 3630 | } |
e1d6c219 | 3631 | out: |
22e2c507 JA |
3632 | if (new_cfqq) |
3633 | kmem_cache_free(cfq_pool, new_cfqq); | |
3634 | ||
2a7f1244 | 3635 | rcu_read_unlock(); |
22e2c507 JA |
3636 | return cfqq; |
3637 | } | |
3638 | ||
c2dea2d1 VT |
3639 | static struct cfq_queue ** |
3640 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
3641 | { | |
fe094d98 | 3642 | switch (ioprio_class) { |
c2dea2d1 VT |
3643 | case IOPRIO_CLASS_RT: |
3644 | return &cfqd->async_cfqq[0][ioprio]; | |
598971bf TH |
3645 | case IOPRIO_CLASS_NONE: |
3646 | ioprio = IOPRIO_NORM; | |
3647 | /* fall through */ | |
c2dea2d1 VT |
3648 | case IOPRIO_CLASS_BE: |
3649 | return &cfqd->async_cfqq[1][ioprio]; | |
3650 | case IOPRIO_CLASS_IDLE: | |
3651 | return &cfqd->async_idle_cfqq; | |
3652 | default: | |
3653 | BUG(); | |
3654 | } | |
3655 | } | |
3656 | ||
15c31be4 | 3657 | static struct cfq_queue * |
abede6da | 3658 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
4f85cb96 | 3659 | struct bio *bio, gfp_t gfp_mask) |
15c31be4 | 3660 | { |
faad78c6 JM |
3661 | int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
3662 | int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
c2dea2d1 | 3663 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
3664 | struct cfq_queue *cfqq = NULL; |
3665 | ||
c2dea2d1 | 3666 | if (!is_sync) { |
faad78c6 JM |
3667 | if (!ioprio_valid(cic->ioprio)) { |
3668 | struct task_struct *tsk = current; | |
3669 | ioprio = task_nice_ioprio(tsk); | |
3670 | ioprio_class = task_nice_ioclass(tsk); | |
3671 | } | |
c2dea2d1 VT |
3672 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); |
3673 | cfqq = *async_cfqq; | |
3674 | } | |
3675 | ||
6118b70b | 3676 | if (!cfqq) |
abede6da | 3677 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, cic, bio, gfp_mask); |
15c31be4 JA |
3678 | |
3679 | /* | |
3680 | * pin the queue now that it's allocated, scheduler exit will prune it | |
3681 | */ | |
c2dea2d1 | 3682 | if (!is_sync && !(*async_cfqq)) { |
30d7b944 | 3683 | cfqq->ref++; |
c2dea2d1 | 3684 | *async_cfqq = cfqq; |
15c31be4 JA |
3685 | } |
3686 | ||
30d7b944 | 3687 | cfqq->ref++; |
15c31be4 JA |
3688 | return cfqq; |
3689 | } | |
3690 | ||
22e2c507 | 3691 | static void |
383cd721 | 3692 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 3693 | { |
383cd721 SL |
3694 | unsigned long elapsed = jiffies - ttime->last_end_request; |
3695 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 3696 | |
383cd721 SL |
3697 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
3698 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
3699 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
3700 | } | |
3701 | ||
3702 | static void | |
3703 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3704 | struct cfq_io_cq *cic) |
383cd721 | 3705 | { |
f5f2b6ce | 3706 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3707 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3708 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3709 | cfqd->cfq_slice_idle); | |
3710 | } | |
7700fc4f SL |
3711 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3712 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3713 | #endif | |
22e2c507 | 3714 | } |
1da177e4 | 3715 | |
206dc69b | 3716 | static void |
b2c18e1e | 3717 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3718 | struct request *rq) |
206dc69b | 3719 | { |
3dde36dd | 3720 | sector_t sdist = 0; |
41647e7a | 3721 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3722 | if (cfqq->last_request_pos) { |
3723 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3724 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3725 | else | |
3726 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3727 | } | |
206dc69b | 3728 | |
3dde36dd | 3729 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3730 | if (blk_queue_nonrot(cfqd->queue)) |
3731 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3732 | else | |
3733 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3734 | } |
1da177e4 | 3735 | |
22e2c507 JA |
3736 | /* |
3737 | * Disable idle window if the process thinks too long or seeks so much that | |
3738 | * it doesn't matter | |
3739 | */ | |
3740 | static void | |
3741 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3742 | struct cfq_io_cq *cic) |
22e2c507 | 3743 | { |
7b679138 | 3744 | int old_idle, enable_idle; |
1be92f2f | 3745 | |
0871714e JA |
3746 | /* |
3747 | * Don't idle for async or idle io prio class | |
3748 | */ | |
3749 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3750 | return; |
3751 | ||
3c2a0909 S |
3752 | if (cfqd->cfq_rt_idle_only && !cfq_class_rt(cfqq)) |
3753 | return; | |
3754 | ||
c265a7f4 | 3755 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3756 | |
76280aff CZ |
3757 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3758 | cfq_mark_cfqq_deep(cfqq); | |
3759 | ||
749ef9f8 CZ |
3760 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
3761 | enable_idle = 0; | |
f6e8d01b | 3762 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
3763 | !cfqd->cfq_slice_idle || |
3764 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3765 | enable_idle = 0; |
383cd721 SL |
3766 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3767 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3768 | enable_idle = 0; |
3769 | else | |
3770 | enable_idle = 1; | |
1da177e4 LT |
3771 | } |
3772 | ||
7b679138 JA |
3773 | if (old_idle != enable_idle) { |
3774 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3775 | if (enable_idle) | |
3776 | cfq_mark_cfqq_idle_window(cfqq); | |
3777 | else | |
3778 | cfq_clear_cfqq_idle_window(cfqq); | |
3779 | } | |
22e2c507 | 3780 | } |
1da177e4 | 3781 | |
22e2c507 JA |
3782 | /* |
3783 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3784 | * no or if we aren't sure, a 1 will cause a preempt. | |
3785 | */ | |
a6151c3a | 3786 | static bool |
22e2c507 | 3787 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3788 | struct request *rq) |
22e2c507 | 3789 | { |
6d048f53 | 3790 | struct cfq_queue *cfqq; |
22e2c507 | 3791 | |
6d048f53 JA |
3792 | cfqq = cfqd->active_queue; |
3793 | if (!cfqq) | |
a6151c3a | 3794 | return false; |
22e2c507 | 3795 | |
6d048f53 | 3796 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3797 | return false; |
22e2c507 JA |
3798 | |
3799 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3800 | return true; |
1e3335de | 3801 | |
875feb63 DS |
3802 | /* |
3803 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3804 | */ | |
3805 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3806 | return false; | |
3807 | ||
374f84ac JA |
3808 | /* |
3809 | * if the new request is sync, but the currently running queue is | |
3810 | * not, let the sync request have priority. | |
3811 | */ | |
3c2a0909 | 3812 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) |
a6151c3a | 3813 | return true; |
1e3335de | 3814 | |
8682e1f1 VG |
3815 | if (new_cfqq->cfqg != cfqq->cfqg) |
3816 | return false; | |
3817 | ||
3818 | if (cfq_slice_used(cfqq)) | |
3819 | return true; | |
3820 | ||
3821 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
4d2ceea4 | 3822 | if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD && |
8682e1f1 VG |
3823 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
3824 | new_cfqq->service_tree->count == 2 && | |
3825 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3826 | return true; | |
3827 | ||
b53d1ed7 JA |
3828 | /* |
3829 | * So both queues are sync. Let the new request get disk time if | |
3830 | * it's a metadata request and the current queue is doing regular IO. | |
3831 | */ | |
65299a3b | 3832 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
3833 | return true; |
3834 | ||
3a9a3f6c DS |
3835 | /* |
3836 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3837 | */ | |
3838 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3839 | return true; |
3a9a3f6c | 3840 | |
d2d59e18 SL |
3841 | /* An idle queue should not be idle now for some reason */ |
3842 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
3843 | return true; | |
3844 | ||
1e3335de | 3845 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3846 | return false; |
1e3335de JA |
3847 | |
3848 | /* | |
3849 | * if this request is as-good as one we would expect from the | |
3850 | * current cfqq, let it preempt | |
3851 | */ | |
e9ce335d | 3852 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3853 | return true; |
1e3335de | 3854 | |
a6151c3a | 3855 | return false; |
22e2c507 JA |
3856 | } |
3857 | ||
3858 | /* | |
3859 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3860 | * let it have half of its nominal slice. | |
3861 | */ | |
3862 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3863 | { | |
df0793ab SL |
3864 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
3865 | ||
7b679138 | 3866 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 3867 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3868 | |
f8ae6e3e SL |
3869 | /* |
3870 | * workload type is changed, don't save slice, otherwise preempt | |
3871 | * doesn't happen | |
3872 | */ | |
df0793ab | 3873 | if (old_type != cfqq_type(cfqq)) |
4d2ceea4 | 3874 | cfqq->cfqg->saved_wl_slice = 0; |
f8ae6e3e | 3875 | |
bf572256 JA |
3876 | /* |
3877 | * Put the new queue at the front of the of the current list, | |
3878 | * so we know that it will be selected next. | |
3879 | */ | |
3880 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3881 | |
3882 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3883 | |
62a37f6b JT |
3884 | cfqq->slice_end = 0; |
3885 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3886 | } |
3887 | ||
22e2c507 | 3888 | /* |
5e705374 | 3889 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3890 | * something we should do about it |
3891 | */ | |
3892 | static void | |
5e705374 JA |
3893 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3894 | struct request *rq) | |
22e2c507 | 3895 | { |
c5869807 | 3896 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3897 | |
45333d5a | 3898 | cfqd->rq_queued++; |
65299a3b CH |
3899 | if (rq->cmd_flags & REQ_PRIO) |
3900 | cfqq->prio_pending++; | |
374f84ac | 3901 | |
383cd721 | 3902 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3903 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3904 | cfq_update_idle_window(cfqd, cfqq, cic); |
3905 | ||
b2c18e1e | 3906 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3907 | |
3908 | if (cfqq == cfqd->active_queue) { | |
3909 | /* | |
b029195d JA |
3910 | * Remember that we saw a request from this process, but |
3911 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3912 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3913 | * and merging. If the request is already larger than a single |
3914 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3915 | * merging is already done. Ditto for a busy system that |
3916 | * has other work pending, don't risk delaying until the | |
3917 | * idle timer unplug to continue working. | |
22e2c507 | 3918 | */ |
d6ceb25e | 3919 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3920 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3921 | cfqd->busy_queues > 1) { | |
812df48d | 3922 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3923 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3924 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3925 | } else { |
155fead9 | 3926 | cfqg_stats_update_idle_time(cfqq->cfqg); |
bf791937 | 3927 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3928 | } |
d6ceb25e | 3929 | } |
5e705374 | 3930 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3931 | /* |
3932 | * not the active queue - expire current slice if it is | |
3933 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3934 | * has some old slice time left and is of higher priority or |
3935 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3936 | */ |
3937 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3938 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3939 | } |
1da177e4 LT |
3940 | } |
3941 | ||
165125e1 | 3942 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3943 | { |
b4878f24 | 3944 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3945 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3946 | |
7b679138 | 3947 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 3948 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 3949 | |
30996f40 | 3950 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3951 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3952 | cfq_add_rq_rb(rq); |
155fead9 TH |
3953 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
3954 | rq->cmd_flags); | |
5e705374 | 3955 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3956 | } |
3957 | ||
45333d5a AC |
3958 | /* |
3959 | * Update hw_tag based on peak queue depth over 50 samples under | |
3960 | * sufficient load. | |
3961 | */ | |
3962 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3963 | { | |
1a1238a7 SL |
3964 | struct cfq_queue *cfqq = cfqd->active_queue; |
3965 | ||
53c583d2 CZ |
3966 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3967 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3968 | |
3969 | if (cfqd->hw_tag == 1) | |
3970 | return; | |
45333d5a AC |
3971 | |
3972 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3973 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3974 | return; |
3975 | ||
1a1238a7 SL |
3976 | /* |
3977 | * If active queue hasn't enough requests and can idle, cfq might not | |
3978 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3979 | * case | |
3980 | */ | |
3981 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3982 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3983 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3984 | return; |
3985 | ||
45333d5a AC |
3986 | if (cfqd->hw_tag_samples++ < 50) |
3987 | return; | |
3988 | ||
e459dd08 | 3989 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3990 | cfqd->hw_tag = 1; |
3991 | else | |
3992 | cfqd->hw_tag = 0; | |
45333d5a AC |
3993 | } |
3994 | ||
7667aa06 VG |
3995 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3996 | { | |
c5869807 | 3997 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 3998 | |
02a8f01b JT |
3999 | /* If the queue already has requests, don't wait */ |
4000 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
4001 | return false; | |
4002 | ||
7667aa06 VG |
4003 | /* If there are other queues in the group, don't wait */ |
4004 | if (cfqq->cfqg->nr_cfqq > 1) | |
4005 | return false; | |
4006 | ||
7700fc4f SL |
4007 | /* the only queue in the group, but think time is big */ |
4008 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
4009 | return false; | |
4010 | ||
7667aa06 VG |
4011 | if (cfq_slice_used(cfqq)) |
4012 | return true; | |
4013 | ||
4014 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
4015 | if (cic && sample_valid(cic->ttime.ttime_samples) |
4016 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
4017 | return true; |
4018 | ||
4019 | /* | |
4020 | * If think times is less than a jiffy than ttime_mean=0 and above | |
4021 | * will not be true. It might happen that slice has not expired yet | |
4022 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
4023 | * case where think time is less than a jiffy, mark the queue wait | |
4024 | * busy if only 1 jiffy is left in the slice. | |
4025 | */ | |
4026 | if (cfqq->slice_end - jiffies == 1) | |
4027 | return true; | |
4028 | ||
4029 | return false; | |
4030 | } | |
4031 | ||
165125e1 | 4032 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 4033 | { |
5e705374 | 4034 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 4035 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 4036 | const int sync = rq_is_sync(rq); |
b4878f24 | 4037 | unsigned long now; |
1da177e4 | 4038 | |
b4878f24 | 4039 | now = jiffies; |
33659ebb CH |
4040 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
4041 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 4042 | |
45333d5a AC |
4043 | cfq_update_hw_tag(cfqd); |
4044 | ||
53c583d2 | 4045 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 4046 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 4047 | cfqd->rq_in_driver--; |
6d048f53 | 4048 | cfqq->dispatched--; |
80bdf0c7 | 4049 | (RQ_CFQG(rq))->dispatched--; |
155fead9 TH |
4050 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
4051 | rq_io_start_time_ns(rq), rq->cmd_flags); | |
1da177e4 | 4052 | |
53c583d2 | 4053 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 4054 | |
365722bb | 4055 | if (sync) { |
34b98d03 | 4056 | struct cfq_rb_root *st; |
f5f2b6ce | 4057 | |
383cd721 | 4058 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
4059 | |
4060 | if (cfq_cfqq_on_rr(cfqq)) | |
34b98d03 | 4061 | st = cfqq->service_tree; |
f5f2b6ce | 4062 | else |
34b98d03 VG |
4063 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), |
4064 | cfqq_type(cfqq)); | |
4065 | ||
4066 | st->ttime.last_end_request = now; | |
573412b2 CZ |
4067 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
4068 | cfqd->last_delayed_sync = now; | |
365722bb | 4069 | } |
caaa5f9f | 4070 | |
7700fc4f SL |
4071 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4072 | cfqq->cfqg->ttime.last_end_request = now; | |
4073 | #endif | |
4074 | ||
caaa5f9f JA |
4075 | /* |
4076 | * If this is the active queue, check if it needs to be expired, | |
4077 | * or if we want to idle in case it has no pending requests. | |
4078 | */ | |
4079 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
4080 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
4081 | ||
44f7c160 JA |
4082 | if (cfq_cfqq_slice_new(cfqq)) { |
4083 | cfq_set_prio_slice(cfqd, cfqq); | |
4084 | cfq_clear_cfqq_slice_new(cfqq); | |
4085 | } | |
f75edf2d VG |
4086 | |
4087 | /* | |
7667aa06 VG |
4088 | * Should we wait for next request to come in before we expire |
4089 | * the queue. | |
f75edf2d | 4090 | */ |
7667aa06 | 4091 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
4092 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
4093 | if (!cfqd->cfq_slice_idle) | |
4094 | extend_sl = cfqd->cfq_group_idle; | |
4095 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 4096 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 4097 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
4098 | } |
4099 | ||
a36e71f9 | 4100 | /* |
8e550632 CZ |
4101 | * Idling is not enabled on: |
4102 | * - expired queues | |
4103 | * - idle-priority queues | |
4104 | * - async queues | |
4105 | * - queues with still some requests queued | |
4106 | * - when there is a close cooperator | |
a36e71f9 | 4107 | */ |
0871714e | 4108 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 4109 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
4110 | else if (sync && cfqq_empty && |
4111 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 4112 | cfq_arm_slice_timer(cfqd); |
8e550632 | 4113 | } |
caaa5f9f | 4114 | } |
6d048f53 | 4115 | |
53c583d2 | 4116 | if (!cfqd->rq_in_driver) |
23e018a1 | 4117 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
4118 | } |
4119 | ||
89850f7e | 4120 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 4121 | { |
1b379d8d | 4122 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 4123 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 4124 | return ELV_MQUEUE_MUST; |
3b18152c | 4125 | } |
1da177e4 | 4126 | |
22e2c507 | 4127 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
4128 | } |
4129 | ||
165125e1 | 4130 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
4131 | { |
4132 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
4133 | struct task_struct *tsk = current; | |
c5869807 | 4134 | struct cfq_io_cq *cic; |
22e2c507 JA |
4135 | struct cfq_queue *cfqq; |
4136 | ||
4137 | /* | |
4138 | * don't force setup of a queue from here, as a call to may_queue | |
4139 | * does not necessarily imply that a request actually will be queued. | |
4140 | * so just lookup a possibly existing queue, or return 'may queue' | |
4141 | * if that fails | |
4142 | */ | |
4ac845a2 | 4143 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
4144 | if (!cic) |
4145 | return ELV_MQUEUE_MAY; | |
4146 | ||
b0b78f81 | 4147 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 4148 | if (cfqq) { |
abede6da | 4149 | cfq_init_prio_data(cfqq, cic); |
22e2c507 | 4150 | |
89850f7e | 4151 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
4152 | } |
4153 | ||
4154 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
4155 | } |
4156 | ||
1da177e4 LT |
4157 | /* |
4158 | * queue lock held here | |
4159 | */ | |
bb37b94c | 4160 | static void cfq_put_request(struct request *rq) |
1da177e4 | 4161 | { |
5e705374 | 4162 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 4163 | |
5e705374 | 4164 | if (cfqq) { |
22e2c507 | 4165 | const int rw = rq_data_dir(rq); |
1da177e4 | 4166 | |
22e2c507 JA |
4167 | BUG_ON(!cfqq->allocated[rw]); |
4168 | cfqq->allocated[rw]--; | |
1da177e4 | 4169 | |
7f1dc8a2 | 4170 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 4171 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
4172 | rq->elv.priv[0] = NULL; |
4173 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 4174 | |
1da177e4 LT |
4175 | cfq_put_queue(cfqq); |
4176 | } | |
4177 | } | |
4178 | ||
df5fe3e8 | 4179 | static struct cfq_queue * |
c5869807 | 4180 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
4181 | struct cfq_queue *cfqq) |
4182 | { | |
4183 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
4184 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 4185 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
4186 | cfq_put_queue(cfqq); |
4187 | return cic_to_cfqq(cic, 1); | |
4188 | } | |
4189 | ||
e6c5bc73 JM |
4190 | /* |
4191 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
4192 | * was the last process referring to said cfqq. | |
4193 | */ | |
4194 | static struct cfq_queue * | |
c5869807 | 4195 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
4196 | { |
4197 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
4198 | cfqq->pid = current->pid; |
4199 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 4200 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
4201 | return cfqq; |
4202 | } | |
4203 | ||
4204 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
4205 | |
4206 | cfq_put_cooperator(cfqq); | |
4207 | ||
e6c5bc73 JM |
4208 | cfq_put_queue(cfqq); |
4209 | return NULL; | |
4210 | } | |
1da177e4 | 4211 | /* |
22e2c507 | 4212 | * Allocate cfq data structures associated with this request. |
1da177e4 | 4213 | */ |
22e2c507 | 4214 | static int |
852c788f TH |
4215 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
4216 | gfp_t gfp_mask) | |
1da177e4 LT |
4217 | { |
4218 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 4219 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 4220 | const int rw = rq_data_dir(rq); |
a6151c3a | 4221 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 4222 | struct cfq_queue *cfqq; |
1da177e4 LT |
4223 | |
4224 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
4225 | ||
216284c3 | 4226 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 4227 | |
598971bf TH |
4228 | check_ioprio_changed(cic, bio); |
4229 | check_blkcg_changed(cic, bio); | |
e6c5bc73 | 4230 | new_queue: |
91fac317 | 4231 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 4232 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
abede6da | 4233 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio, gfp_mask); |
91fac317 | 4234 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 4235 | } else { |
e6c5bc73 JM |
4236 | /* |
4237 | * If the queue was seeky for too long, break it apart. | |
4238 | */ | |
ae54abed | 4239 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
4240 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
4241 | cfqq = split_cfqq(cic, cfqq); | |
4242 | if (!cfqq) | |
4243 | goto new_queue; | |
4244 | } | |
4245 | ||
df5fe3e8 JM |
4246 | /* |
4247 | * Check to see if this queue is scheduled to merge with | |
4248 | * another, closely cooperating queue. The merging of | |
4249 | * queues happens here as it must be done in process context. | |
4250 | * The reference on new_cfqq was taken in merge_cfqqs. | |
4251 | */ | |
4252 | if (cfqq->new_cfqq) | |
4253 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 4254 | } |
1da177e4 LT |
4255 | |
4256 | cfqq->allocated[rw]++; | |
1da177e4 | 4257 | |
6fae9c25 | 4258 | cfqq->ref++; |
eb7d8c07 | 4259 | cfqg_get(cfqq->cfqg); |
a612fddf | 4260 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 4261 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 4262 | spin_unlock_irq(q->queue_lock); |
5e705374 | 4263 | return 0; |
1da177e4 LT |
4264 | } |
4265 | ||
65f27f38 | 4266 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 4267 | { |
65f27f38 | 4268 | struct cfq_data *cfqd = |
23e018a1 | 4269 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 4270 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4271 | |
40bb54d1 | 4272 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 4273 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 4274 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
4275 | } |
4276 | ||
4277 | /* | |
4278 | * Timer running if the active_queue is currently idling inside its time slice | |
4279 | */ | |
4280 | static void cfq_idle_slice_timer(unsigned long data) | |
4281 | { | |
4282 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
4283 | struct cfq_queue *cfqq; | |
4284 | unsigned long flags; | |
3c6bd2f8 | 4285 | int timed_out = 1; |
22e2c507 | 4286 | |
7b679138 JA |
4287 | cfq_log(cfqd, "idle timer fired"); |
4288 | ||
22e2c507 JA |
4289 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
4290 | ||
fe094d98 JA |
4291 | cfqq = cfqd->active_queue; |
4292 | if (cfqq) { | |
3c6bd2f8 JA |
4293 | timed_out = 0; |
4294 | ||
b029195d JA |
4295 | /* |
4296 | * We saw a request before the queue expired, let it through | |
4297 | */ | |
4298 | if (cfq_cfqq_must_dispatch(cfqq)) | |
4299 | goto out_kick; | |
4300 | ||
22e2c507 JA |
4301 | /* |
4302 | * expired | |
4303 | */ | |
44f7c160 | 4304 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
4305 | goto expire; |
4306 | ||
4307 | /* | |
4308 | * only expire and reinvoke request handler, if there are | |
4309 | * other queues with pending requests | |
4310 | */ | |
caaa5f9f | 4311 | if (!cfqd->busy_queues) |
22e2c507 | 4312 | goto out_cont; |
22e2c507 JA |
4313 | |
4314 | /* | |
4315 | * not expired and it has a request pending, let it dispatch | |
4316 | */ | |
75e50984 | 4317 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 4318 | goto out_kick; |
76280aff CZ |
4319 | |
4320 | /* | |
4321 | * Queue depth flag is reset only when the idle didn't succeed | |
4322 | */ | |
4323 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
4324 | } |
4325 | expire: | |
e5ff082e | 4326 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 4327 | out_kick: |
23e018a1 | 4328 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
4329 | out_cont: |
4330 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
4331 | } | |
4332 | ||
3b18152c JA |
4333 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
4334 | { | |
4335 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 4336 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 4337 | } |
22e2c507 | 4338 | |
c2dea2d1 VT |
4339 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
4340 | { | |
4341 | int i; | |
4342 | ||
4343 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
4344 | if (cfqd->async_cfqq[0][i]) | |
4345 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
4346 | if (cfqd->async_cfqq[1][i]) | |
4347 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 4348 | } |
2389d1ef ON |
4349 | |
4350 | if (cfqd->async_idle_cfqq) | |
4351 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
4352 | } |
4353 | ||
b374d18a | 4354 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 4355 | { |
22e2c507 | 4356 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 4357 | struct request_queue *q = cfqd->queue; |
22e2c507 | 4358 | |
3b18152c | 4359 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 4360 | |
d9ff4187 | 4361 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 4362 | |
d9ff4187 | 4363 | if (cfqd->active_queue) |
e5ff082e | 4364 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 4365 | |
c2dea2d1 | 4366 | cfq_put_async_queues(cfqd); |
03aa264a TH |
4367 | |
4368 | spin_unlock_irq(q->queue_lock); | |
4369 | ||
a90d742e AV |
4370 | cfq_shutdown_timer_wq(cfqd); |
4371 | ||
ffea73fc TH |
4372 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4373 | blkcg_deactivate_policy(q, &blkcg_policy_cfq); | |
4374 | #else | |
f51b802c | 4375 | kfree(cfqd->root_group); |
2abae55f | 4376 | #endif |
56edf7d7 | 4377 | kfree(cfqd); |
1da177e4 LT |
4378 | } |
4379 | ||
a6ad83fc | 4380 | static int cfq_init_queue(struct request_queue *q, struct elevator_type *e) |
1da177e4 LT |
4381 | { |
4382 | struct cfq_data *cfqd; | |
3c798398 | 4383 | struct blkcg_gq *blkg __maybe_unused; |
a2b1693b | 4384 | int i, ret; |
a6ad83fc JM |
4385 | struct elevator_queue *eq; |
4386 | ||
4387 | eq = elevator_alloc(q, e); | |
4388 | if (!eq) | |
4389 | return -ENOMEM; | |
1da177e4 | 4390 | |
94f6030c | 4391 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
a6ad83fc JM |
4392 | if (!cfqd) { |
4393 | kobject_put(&eq->kobj); | |
b2fab5ac | 4394 | return -ENOMEM; |
a6ad83fc JM |
4395 | } |
4396 | eq->elevator_data = cfqd; | |
80b15c73 | 4397 | |
f51b802c | 4398 | cfqd->queue = q; |
a6ad83fc JM |
4399 | spin_lock_irq(q->queue_lock); |
4400 | q->elevator = eq; | |
4401 | spin_unlock_irq(q->queue_lock); | |
f51b802c | 4402 | |
1fa8f6d6 VG |
4403 | /* Init root service tree */ |
4404 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
4405 | ||
f51b802c | 4406 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 4407 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4408 | ret = blkcg_activate_policy(q, &blkcg_policy_cfq); |
a2b1693b TH |
4409 | if (ret) |
4410 | goto out_free; | |
f51b802c | 4411 | |
a2b1693b | 4412 | cfqd->root_group = blkg_to_cfqg(q->root_blkg); |
f51b802c | 4413 | #else |
a2b1693b | 4414 | ret = -ENOMEM; |
f51b802c TH |
4415 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
4416 | GFP_KERNEL, cfqd->queue->node); | |
a2b1693b TH |
4417 | if (!cfqd->root_group) |
4418 | goto out_free; | |
5624a4e4 | 4419 | |
a2b1693b TH |
4420 | cfq_init_cfqg_base(cfqd->root_group); |
4421 | #endif | |
3381cb8d | 4422 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_DEFAULT; |
e71357e1 | 4423 | cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_DEFAULT; |
5624a4e4 | 4424 | |
26a2ac00 JA |
4425 | /* |
4426 | * Not strictly needed (since RB_ROOT just clears the node and we | |
4427 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
4428 | * to add magic to the rb code | |
4429 | */ | |
4430 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
4431 | cfqd->prio_trees[i] = RB_ROOT; | |
4432 | ||
6118b70b JA |
4433 | /* |
4434 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
4435 | * Grab a permanent reference to it, so that the normal code flow | |
f51b802c TH |
4436 | * will not attempt to free it. oom_cfqq is linked to root_group |
4437 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
4438 | * the reference from linking right away. | |
6118b70b JA |
4439 | */ |
4440 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 4441 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
4442 | |
4443 | spin_lock_irq(q->queue_lock); | |
f51b802c | 4444 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 4445 | cfqg_put(cfqd->root_group); |
1adaf3dd | 4446 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 4447 | |
22e2c507 JA |
4448 | init_timer(&cfqd->idle_slice_timer); |
4449 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
4450 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
4451 | ||
23e018a1 | 4452 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 4453 | |
1da177e4 | 4454 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
4455 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
4456 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
4457 | cfqd->cfq_back_max = cfq_back_max; |
4458 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
4459 | cfqd->cfq_slice[0] = cfq_slice_async; |
4460 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
5bf14c07 | 4461 | cfqd->cfq_target_latency = cfq_target_latency; |
22e2c507 JA |
4462 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
4463 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
3c2a0909 | 4464 | cfqd->cfq_rt_idle_only = cfq_rt_idle_only; |
80bdf0c7 | 4465 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 4466 | cfqd->cfq_latency = 1; |
e459dd08 | 4467 | cfqd->hw_tag = -1; |
edc71131 CZ |
4468 | /* |
4469 | * we optimistically start assuming sync ops weren't delayed in last | |
4470 | * second, in order to have larger depth for async operations. | |
4471 | */ | |
573412b2 | 4472 | cfqd->last_delayed_sync = jiffies - HZ; |
b2fab5ac | 4473 | return 0; |
a2b1693b TH |
4474 | |
4475 | out_free: | |
4476 | kfree(cfqd); | |
a6ad83fc | 4477 | kobject_put(&eq->kobj); |
a2b1693b | 4478 | return ret; |
1da177e4 LT |
4479 | } |
4480 | ||
1da177e4 LT |
4481 | /* |
4482 | * sysfs parts below --> | |
4483 | */ | |
1da177e4 LT |
4484 | static ssize_t |
4485 | cfq_var_show(unsigned int var, char *page) | |
4486 | { | |
4487 | return sprintf(page, "%d\n", var); | |
4488 | } | |
4489 | ||
4490 | static ssize_t | |
4491 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
4492 | { | |
4493 | char *p = (char *) page; | |
4494 | ||
4495 | *var = simple_strtoul(p, &p, 10); | |
4496 | return count; | |
4497 | } | |
4498 | ||
1da177e4 | 4499 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 4500 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 4501 | { \ |
3d1ab40f | 4502 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4503 | unsigned int __data = __VAR; \ |
4504 | if (__CONV) \ | |
4505 | __data = jiffies_to_msecs(__data); \ | |
4506 | return cfq_var_show(__data, (page)); \ | |
4507 | } | |
4508 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
4509 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
4510 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
4511 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
4512 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 4513 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
3c2a0909 | 4514 | SHOW_FUNCTION(cfq_rt_idle_only_show, cfqd->cfq_rt_idle_only, 0); |
80bdf0c7 | 4515 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
4516 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4517 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4518 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4519 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
5bf14c07 | 4520 | SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1); |
1da177e4 LT |
4521 | #undef SHOW_FUNCTION |
4522 | ||
4523 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 4524 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4525 | { \ |
3d1ab40f | 4526 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4527 | unsigned int __data; \ |
4528 | int ret = cfq_var_store(&__data, (page), count); \ | |
4529 | if (__data < (MIN)) \ | |
4530 | __data = (MIN); \ | |
4531 | else if (__data > (MAX)) \ | |
4532 | __data = (MAX); \ | |
4533 | if (__CONV) \ | |
4534 | *(__PTR) = msecs_to_jiffies(__data); \ | |
4535 | else \ | |
4536 | *(__PTR) = __data; \ | |
4537 | return ret; \ | |
4538 | } | |
4539 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4540 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4541 | UINT_MAX, 1); | |
4542 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4543 | UINT_MAX, 1); | |
e572ec7e | 4544 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4545 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4546 | UINT_MAX, 0); | |
22e2c507 | 4547 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
3c2a0909 | 4548 | STORE_FUNCTION(cfq_rt_idle_only_store, &cfqd->cfq_rt_idle_only, 0, 1, 0); |
80bdf0c7 | 4549 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4550 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4551 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4552 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4553 | UINT_MAX, 0); | |
963b72fc | 4554 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
5bf14c07 | 4555 | STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1); |
1da177e4 LT |
4556 | #undef STORE_FUNCTION |
4557 | ||
e572ec7e AV |
4558 | #define CFQ_ATTR(name) \ |
4559 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4560 | ||
4561 | static struct elv_fs_entry cfq_attrs[] = { | |
4562 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4563 | CFQ_ATTR(fifo_expire_sync), |
4564 | CFQ_ATTR(fifo_expire_async), | |
4565 | CFQ_ATTR(back_seek_max), | |
4566 | CFQ_ATTR(back_seek_penalty), | |
4567 | CFQ_ATTR(slice_sync), | |
4568 | CFQ_ATTR(slice_async), | |
4569 | CFQ_ATTR(slice_async_rq), | |
4570 | CFQ_ATTR(slice_idle), | |
3c2a0909 | 4571 | CFQ_ATTR(rt_idle_only), |
80bdf0c7 | 4572 | CFQ_ATTR(group_idle), |
963b72fc | 4573 | CFQ_ATTR(low_latency), |
5bf14c07 | 4574 | CFQ_ATTR(target_latency), |
e572ec7e | 4575 | __ATTR_NULL |
1da177e4 LT |
4576 | }; |
4577 | ||
1da177e4 LT |
4578 | static struct elevator_type iosched_cfq = { |
4579 | .ops = { | |
4580 | .elevator_merge_fn = cfq_merge, | |
4581 | .elevator_merged_fn = cfq_merged_request, | |
4582 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 4583 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 4584 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4585 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4586 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4587 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4588 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4589 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4590 | .elevator_former_req_fn = elv_rb_former_request, |
4591 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 4592 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 4593 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
4594 | .elevator_set_req_fn = cfq_set_request, |
4595 | .elevator_put_req_fn = cfq_put_request, | |
4596 | .elevator_may_queue_fn = cfq_may_queue, | |
4597 | .elevator_init_fn = cfq_init_queue, | |
4598 | .elevator_exit_fn = cfq_exit_queue, | |
4599 | }, | |
3d3c2379 TH |
4600 | .icq_size = sizeof(struct cfq_io_cq), |
4601 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 4602 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 4603 | .elevator_name = "cfq", |
1da177e4 LT |
4604 | .elevator_owner = THIS_MODULE, |
4605 | }; | |
4606 | ||
3e252066 | 4607 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4608 | static struct blkcg_policy blkcg_policy_cfq = { |
f9fcc2d3 TH |
4609 | .pd_size = sizeof(struct cfq_group), |
4610 | .cftypes = cfq_blkcg_files, | |
4611 | ||
4612 | .pd_init_fn = cfq_pd_init, | |
0b39920b | 4613 | .pd_offline_fn = cfq_pd_offline, |
f9fcc2d3 | 4614 | .pd_reset_stats_fn = cfq_pd_reset_stats, |
3e252066 | 4615 | }; |
3e252066 VG |
4616 | #endif |
4617 | ||
1da177e4 LT |
4618 | static int __init cfq_init(void) |
4619 | { | |
3d3c2379 TH |
4620 | int ret; |
4621 | ||
22e2c507 JA |
4622 | /* |
4623 | * could be 0 on HZ < 1000 setups | |
4624 | */ | |
4625 | if (!cfq_slice_async) | |
4626 | cfq_slice_async = 1; | |
4627 | if (!cfq_slice_idle) | |
4628 | cfq_slice_idle = 1; | |
4629 | ||
80bdf0c7 VG |
4630 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4631 | if (!cfq_group_idle) | |
4632 | cfq_group_idle = 1; | |
8bd435b3 | 4633 | |
3c798398 | 4634 | ret = blkcg_policy_register(&blkcg_policy_cfq); |
8bd435b3 TH |
4635 | if (ret) |
4636 | return ret; | |
ffea73fc TH |
4637 | #else |
4638 | cfq_group_idle = 0; | |
4639 | #endif | |
8bd435b3 | 4640 | |
fd794956 | 4641 | ret = -ENOMEM; |
3d3c2379 TH |
4642 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4643 | if (!cfq_pool) | |
8bd435b3 | 4644 | goto err_pol_unreg; |
1da177e4 | 4645 | |
3d3c2379 | 4646 | ret = elv_register(&iosched_cfq); |
8bd435b3 TH |
4647 | if (ret) |
4648 | goto err_free_pool; | |
3d3c2379 | 4649 | |
2fdd82bd | 4650 | return 0; |
8bd435b3 TH |
4651 | |
4652 | err_free_pool: | |
4653 | kmem_cache_destroy(cfq_pool); | |
4654 | err_pol_unreg: | |
ffea73fc | 4655 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4656 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4657 | #endif |
8bd435b3 | 4658 | return ret; |
1da177e4 LT |
4659 | } |
4660 | ||
4661 | static void __exit cfq_exit(void) | |
4662 | { | |
ffea73fc | 4663 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3c798398 | 4664 | blkcg_policy_unregister(&blkcg_policy_cfq); |
ffea73fc | 4665 | #endif |
1da177e4 | 4666 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4667 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4668 | } |
4669 | ||
4670 | module_init(cfq_init); | |
4671 | module_exit(cfq_exit); | |
4672 | ||
4673 | MODULE_AUTHOR("Jens Axboe"); | |
4674 | MODULE_LICENSE("GPL"); | |
4675 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |