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" |
e98ef89b | 18 | #include "cfq.h" |
1da177e4 LT |
19 | |
20 | /* | |
21 | * tunables | |
22 | */ | |
fe094d98 | 23 | /* max queue in one round of service */ |
abc3c744 | 24 | static const int cfq_quantum = 8; |
64100099 | 25 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
26 | /* maximum backwards seek, in KiB */ |
27 | static const int cfq_back_max = 16 * 1024; | |
28 | /* penalty of a backwards seek */ | |
29 | static const int cfq_back_penalty = 2; | |
64100099 | 30 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 31 | static int cfq_slice_async = HZ / 25; |
64100099 | 32 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 33 | static int cfq_slice_idle = HZ / 125; |
80bdf0c7 | 34 | static int cfq_group_idle = HZ / 125; |
5db5d642 CZ |
35 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
36 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 37 | |
d9e7620e | 38 | /* |
0871714e | 39 | * offset from end of service tree |
d9e7620e | 40 | */ |
0871714e | 41 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
42 | |
43 | /* | |
44 | * below this threshold, we consider thinktime immediate | |
45 | */ | |
46 | #define CFQ_MIN_TT (2) | |
47 | ||
22e2c507 | 48 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 49 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 50 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 51 | |
3dde36dd | 52 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 53 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 54 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 55 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 56 | |
a612fddf TH |
57 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
58 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
59 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 60 | |
e18b890b | 61 | static struct kmem_cache *cfq_pool; |
1da177e4 | 62 | |
22e2c507 JA |
63 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
64 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
65 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
66 | ||
206dc69b | 67 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 68 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 69 | |
c5869807 TH |
70 | struct cfq_ttime { |
71 | unsigned long last_end_request; | |
72 | ||
73 | unsigned long ttime_total; | |
74 | unsigned long ttime_samples; | |
75 | unsigned long ttime_mean; | |
76 | }; | |
77 | ||
cc09e299 JA |
78 | /* |
79 | * Most of our rbtree usage is for sorting with min extraction, so | |
80 | * if we cache the leftmost node we don't have to walk down the tree | |
81 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
82 | * move this into the elevator for the rq sorting as well. | |
83 | */ | |
84 | struct cfq_rb_root { | |
85 | struct rb_root rb; | |
86 | struct rb_node *left; | |
aa6f6a3d | 87 | unsigned count; |
73e9ffdd | 88 | unsigned total_weight; |
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 | */ | |
158 | enum wl_prio_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 | ||
cdb16e8f VG |
174 | /* This is per cgroup per device grouping structure */ |
175 | struct cfq_group { | |
1fa8f6d6 VG |
176 | /* group service_tree member */ |
177 | struct rb_node rb_node; | |
178 | ||
179 | /* group service_tree key */ | |
180 | u64 vdisktime; | |
25bc6b07 | 181 | unsigned int weight; |
8184f93e JT |
182 | unsigned int new_weight; |
183 | bool needs_update; | |
1fa8f6d6 VG |
184 | |
185 | /* number of cfqq currently on this group */ | |
186 | int nr_cfqq; | |
187 | ||
cdb16e8f | 188 | /* |
4495a7d4 | 189 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
190 | * create the array for each prio class but at run time it is used |
191 | * only for RT and BE class and slot for IDLE class remains unused. | |
192 | * This is primarily done to avoid confusion and a gcc warning. | |
193 | */ | |
194 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
195 | /* | |
196 | * rr lists of queues with requests. We maintain service trees for | |
197 | * RT and BE classes. These trees are subdivided in subclasses | |
198 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
199 | * class there is no subclassification and all the cfq queues go on | |
200 | * a single tree service_tree_idle. | |
cdb16e8f VG |
201 | * Counts are embedded in the cfq_rb_root |
202 | */ | |
203 | struct cfq_rb_root service_trees[2][3]; | |
204 | struct cfq_rb_root service_tree_idle; | |
dae739eb VG |
205 | |
206 | unsigned long saved_workload_slice; | |
207 | enum wl_type_t saved_workload; | |
208 | enum wl_prio_t saved_serving_prio; | |
25fb5169 VG |
209 | struct blkio_group blkg; |
210 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
211 | struct hlist_node cfqd_node; | |
329a6781 | 212 | int ref; |
25fb5169 | 213 | #endif |
80bdf0c7 VG |
214 | /* number of requests that are on the dispatch list or inside driver */ |
215 | int dispatched; | |
7700fc4f | 216 | struct cfq_ttime ttime; |
cdb16e8f | 217 | }; |
718eee05 | 218 | |
c5869807 TH |
219 | struct cfq_io_cq { |
220 | struct io_cq icq; /* must be the first member */ | |
221 | struct cfq_queue *cfqq[2]; | |
222 | struct cfq_ttime ttime; | |
223 | }; | |
224 | ||
22e2c507 JA |
225 | /* |
226 | * Per block device queue structure | |
227 | */ | |
1da177e4 | 228 | struct cfq_data { |
165125e1 | 229 | struct request_queue *queue; |
1fa8f6d6 VG |
230 | /* Root service tree for cfq_groups */ |
231 | struct cfq_rb_root grp_service_tree; | |
cdb16e8f | 232 | struct cfq_group root_group; |
22e2c507 | 233 | |
c0324a02 CZ |
234 | /* |
235 | * The priority currently being served | |
22e2c507 | 236 | */ |
c0324a02 | 237 | enum wl_prio_t serving_prio; |
718eee05 CZ |
238 | enum wl_type_t serving_type; |
239 | unsigned long workload_expires; | |
cdb16e8f | 240 | struct cfq_group *serving_group; |
a36e71f9 JA |
241 | |
242 | /* | |
243 | * Each priority tree is sorted by next_request position. These | |
244 | * trees are used when determining if two or more queues are | |
245 | * interleaving requests (see cfq_close_cooperator). | |
246 | */ | |
247 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
248 | ||
22e2c507 | 249 | unsigned int busy_queues; |
ef8a41df | 250 | unsigned int busy_sync_queues; |
22e2c507 | 251 | |
53c583d2 CZ |
252 | int rq_in_driver; |
253 | int rq_in_flight[2]; | |
45333d5a AC |
254 | |
255 | /* | |
256 | * queue-depth detection | |
257 | */ | |
258 | int rq_queued; | |
25776e35 | 259 | int hw_tag; |
e459dd08 CZ |
260 | /* |
261 | * hw_tag can be | |
262 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
263 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
264 | * 0 => no NCQ | |
265 | */ | |
266 | int hw_tag_est_depth; | |
267 | unsigned int hw_tag_samples; | |
1da177e4 | 268 | |
22e2c507 JA |
269 | /* |
270 | * idle window management | |
271 | */ | |
272 | struct timer_list idle_slice_timer; | |
23e018a1 | 273 | struct work_struct unplug_work; |
1da177e4 | 274 | |
22e2c507 | 275 | struct cfq_queue *active_queue; |
c5869807 | 276 | struct cfq_io_cq *active_cic; |
22e2c507 | 277 | |
c2dea2d1 VT |
278 | /* |
279 | * async queue for each priority case | |
280 | */ | |
281 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
282 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 283 | |
6d048f53 | 284 | sector_t last_position; |
1da177e4 | 285 | |
1da177e4 LT |
286 | /* |
287 | * tunables, see top of file | |
288 | */ | |
289 | unsigned int cfq_quantum; | |
22e2c507 | 290 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
291 | unsigned int cfq_back_penalty; |
292 | unsigned int cfq_back_max; | |
22e2c507 JA |
293 | unsigned int cfq_slice[2]; |
294 | unsigned int cfq_slice_async_rq; | |
295 | unsigned int cfq_slice_idle; | |
80bdf0c7 | 296 | unsigned int cfq_group_idle; |
963b72fc | 297 | unsigned int cfq_latency; |
d9ff4187 | 298 | |
6118b70b JA |
299 | /* |
300 | * Fallback dummy cfqq for extreme OOM conditions | |
301 | */ | |
302 | struct cfq_queue oom_cfqq; | |
365722bb | 303 | |
573412b2 | 304 | unsigned long last_delayed_sync; |
25fb5169 VG |
305 | |
306 | /* List of cfq groups being managed on this device*/ | |
307 | struct hlist_head cfqg_list; | |
56edf7d7 VG |
308 | |
309 | /* Number of groups which are on blkcg->blkg_list */ | |
310 | unsigned int nr_blkcg_linked_grps; | |
1da177e4 LT |
311 | }; |
312 | ||
25fb5169 VG |
313 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
314 | ||
cdb16e8f VG |
315 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
316 | enum wl_prio_t prio, | |
65b32a57 | 317 | enum wl_type_t type) |
c0324a02 | 318 | { |
1fa8f6d6 VG |
319 | if (!cfqg) |
320 | return NULL; | |
321 | ||
c0324a02 | 322 | if (prio == IDLE_WORKLOAD) |
cdb16e8f | 323 | return &cfqg->service_tree_idle; |
c0324a02 | 324 | |
cdb16e8f | 325 | return &cfqg->service_trees[prio][type]; |
c0324a02 CZ |
326 | } |
327 | ||
3b18152c | 328 | enum cfqq_state_flags { |
b0b8d749 JA |
329 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
330 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 331 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 332 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
333 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
334 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
335 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 336 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 337 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 338 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 339 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 340 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 341 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
342 | }; |
343 | ||
344 | #define CFQ_CFQQ_FNS(name) \ | |
345 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
346 | { \ | |
fe094d98 | 347 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
348 | } \ |
349 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
350 | { \ | |
fe094d98 | 351 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
352 | } \ |
353 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
354 | { \ | |
fe094d98 | 355 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
356 | } |
357 | ||
358 | CFQ_CFQQ_FNS(on_rr); | |
359 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 360 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 361 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
362 | CFQ_CFQQ_FNS(fifo_expire); |
363 | CFQ_CFQQ_FNS(idle_window); | |
364 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 365 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 366 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 367 | CFQ_CFQQ_FNS(coop); |
ae54abed | 368 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 369 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 370 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
371 | #undef CFQ_CFQQ_FNS |
372 | ||
afc24d49 | 373 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2868ef7b VG |
374 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
375 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ | |
376 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
4495a7d4 | 377 | blkg_path(&(cfqq)->cfqg->blkg), ##args) |
2868ef7b VG |
378 | |
379 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ | |
380 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ | |
4495a7d4 | 381 | blkg_path(&(cfqg)->blkg), ##args) \ |
2868ef7b VG |
382 | |
383 | #else | |
7b679138 JA |
384 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
385 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
4495a7d4 | 386 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
2868ef7b | 387 | #endif |
7b679138 JA |
388 | #define cfq_log(cfqd, fmt, args...) \ |
389 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
390 | ||
615f0259 VG |
391 | /* Traverses through cfq group service trees */ |
392 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
393 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
394 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
395 | : &cfqg->service_tree_idle; \ | |
396 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
397 | (i == IDLE_WORKLOAD && j == 0); \ | |
398 | j++, st = i < IDLE_WORKLOAD ? \ | |
399 | &cfqg->service_trees[i][j]: NULL) \ | |
400 | ||
f5f2b6ce SL |
401 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
402 | struct cfq_ttime *ttime, bool group_idle) | |
403 | { | |
404 | unsigned long slice; | |
405 | if (!sample_valid(ttime->ttime_samples)) | |
406 | return false; | |
407 | if (group_idle) | |
408 | slice = cfqd->cfq_group_idle; | |
409 | else | |
410 | slice = cfqd->cfq_slice_idle; | |
411 | return ttime->ttime_mean > slice; | |
412 | } | |
615f0259 | 413 | |
02b35081 VG |
414 | static inline bool iops_mode(struct cfq_data *cfqd) |
415 | { | |
416 | /* | |
417 | * If we are not idling on queues and it is a NCQ drive, parallel | |
418 | * execution of requests is on and measuring time is not possible | |
419 | * in most of the cases until and unless we drive shallower queue | |
420 | * depths and that becomes a performance bottleneck. In such cases | |
421 | * switch to start providing fairness in terms of number of IOs. | |
422 | */ | |
423 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
424 | return true; | |
425 | else | |
426 | return false; | |
427 | } | |
428 | ||
c0324a02 CZ |
429 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
430 | { | |
431 | if (cfq_class_idle(cfqq)) | |
432 | return IDLE_WORKLOAD; | |
433 | if (cfq_class_rt(cfqq)) | |
434 | return RT_WORKLOAD; | |
435 | return BE_WORKLOAD; | |
436 | } | |
437 | ||
718eee05 CZ |
438 | |
439 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
440 | { | |
441 | if (!cfq_cfqq_sync(cfqq)) | |
442 | return ASYNC_WORKLOAD; | |
443 | if (!cfq_cfqq_idle_window(cfqq)) | |
444 | return SYNC_NOIDLE_WORKLOAD; | |
445 | return SYNC_WORKLOAD; | |
446 | } | |
447 | ||
58ff82f3 VG |
448 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
449 | struct cfq_data *cfqd, | |
450 | struct cfq_group *cfqg) | |
c0324a02 CZ |
451 | { |
452 | if (wl == IDLE_WORKLOAD) | |
cdb16e8f | 453 | return cfqg->service_tree_idle.count; |
c0324a02 | 454 | |
cdb16e8f VG |
455 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
456 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count | |
457 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
458 | } |
459 | ||
f26bd1f0 VG |
460 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
461 | struct cfq_group *cfqg) | |
462 | { | |
463 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count | |
464 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
465 | } | |
466 | ||
165125e1 | 467 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
a6151c3a | 468 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool, |
fd0928df | 469 | struct io_context *, gfp_t); |
91fac317 | 470 | |
c5869807 TH |
471 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
472 | { | |
473 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
474 | return container_of(icq, struct cfq_io_cq, icq); | |
475 | } | |
476 | ||
47fdd4ca TH |
477 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
478 | struct io_context *ioc) | |
479 | { | |
480 | if (ioc) | |
481 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
482 | return NULL; | |
483 | } | |
484 | ||
c5869807 | 485 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 486 | { |
a6151c3a | 487 | return cic->cfqq[is_sync]; |
91fac317 VT |
488 | } |
489 | ||
c5869807 TH |
490 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
491 | bool is_sync) | |
91fac317 | 492 | { |
a6151c3a | 493 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
494 | } |
495 | ||
c5869807 | 496 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 497 | { |
c5869807 | 498 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
499 | } |
500 | ||
91fac317 VT |
501 | /* |
502 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
503 | * set (in which case it could also be direct WRITE). | |
504 | */ | |
a6151c3a | 505 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 506 | { |
7b6d91da | 507 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 508 | } |
1da177e4 | 509 | |
99f95e52 AM |
510 | /* |
511 | * scheduler run of queue, if there are requests pending and no one in the | |
512 | * driver that will restart queueing | |
513 | */ | |
23e018a1 | 514 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 515 | { |
7b679138 JA |
516 | if (cfqd->busy_queues) { |
517 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 518 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 519 | } |
99f95e52 AM |
520 | } |
521 | ||
44f7c160 JA |
522 | /* |
523 | * Scale schedule slice based on io priority. Use the sync time slice only | |
524 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
525 | * io only, should not get full sync slice length. | |
526 | */ | |
a6151c3a | 527 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 528 | unsigned short prio) |
44f7c160 | 529 | { |
d9e7620e | 530 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 531 | |
d9e7620e JA |
532 | WARN_ON(prio >= IOPRIO_BE_NR); |
533 | ||
534 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
535 | } | |
44f7c160 | 536 | |
d9e7620e JA |
537 | static inline int |
538 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
539 | { | |
540 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
541 | } |
542 | ||
25bc6b07 VG |
543 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
544 | { | |
545 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
546 | ||
547 | d = d * BLKIO_WEIGHT_DEFAULT; | |
548 | do_div(d, cfqg->weight); | |
549 | return d; | |
550 | } | |
551 | ||
552 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
553 | { | |
554 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
555 | if (delta > 0) | |
556 | min_vdisktime = vdisktime; | |
557 | ||
558 | return min_vdisktime; | |
559 | } | |
560 | ||
561 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
562 | { | |
563 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
564 | if (delta < 0) | |
565 | min_vdisktime = vdisktime; | |
566 | ||
567 | return min_vdisktime; | |
568 | } | |
569 | ||
570 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
571 | { | |
25bc6b07 VG |
572 | struct cfq_group *cfqg; |
573 | ||
25bc6b07 VG |
574 | if (st->left) { |
575 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
576 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
577 | cfqg->vdisktime); | |
25bc6b07 | 578 | } |
25bc6b07 VG |
579 | } |
580 | ||
5db5d642 CZ |
581 | /* |
582 | * get averaged number of queues of RT/BE priority. | |
583 | * average is updated, with a formula that gives more weight to higher numbers, | |
584 | * to quickly follows sudden increases and decrease slowly | |
585 | */ | |
586 | ||
58ff82f3 VG |
587 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
588 | struct cfq_group *cfqg, bool rt) | |
5869619c | 589 | { |
5db5d642 CZ |
590 | unsigned min_q, max_q; |
591 | unsigned mult = cfq_hist_divisor - 1; | |
592 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 593 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 594 | |
58ff82f3 VG |
595 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
596 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
597 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 598 | cfq_hist_divisor; |
58ff82f3 VG |
599 | return cfqg->busy_queues_avg[rt]; |
600 | } | |
601 | ||
602 | static inline unsigned | |
603 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
604 | { | |
605 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
606 | ||
607 | return cfq_target_latency * cfqg->weight / st->total_weight; | |
5db5d642 CZ |
608 | } |
609 | ||
c553f8e3 | 610 | static inline unsigned |
ba5bd520 | 611 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 612 | { |
5db5d642 CZ |
613 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
614 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
615 | /* |
616 | * interested queues (we consider only the ones with the same | |
617 | * priority class in the cfq group) | |
618 | */ | |
619 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
620 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
621 | unsigned sync_slice = cfqd->cfq_slice[1]; |
622 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
623 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
624 | ||
625 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
626 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
627 | /* scale low_slice according to IO priority | |
628 | * and sync vs async */ | |
629 | unsigned low_slice = | |
630 | min(slice, base_low_slice * slice / sync_slice); | |
631 | /* the adapted slice value is scaled to fit all iqs | |
632 | * into the target latency */ | |
58ff82f3 | 633 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
634 | low_slice); |
635 | } | |
636 | } | |
c553f8e3 SL |
637 | return slice; |
638 | } | |
639 | ||
640 | static inline void | |
641 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
642 | { | |
ba5bd520 | 643 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 644 | |
dae739eb | 645 | cfqq->slice_start = jiffies; |
5db5d642 | 646 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 647 | cfqq->allocated_slice = slice; |
7b679138 | 648 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
649 | } |
650 | ||
651 | /* | |
652 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
653 | * isn't valid until the first request from the dispatch is activated | |
654 | * and the slice time set. | |
655 | */ | |
a6151c3a | 656 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
657 | { |
658 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 659 | return false; |
44f7c160 | 660 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 661 | return false; |
44f7c160 | 662 | |
c1e44756 | 663 | return true; |
44f7c160 JA |
664 | } |
665 | ||
1da177e4 | 666 | /* |
5e705374 | 667 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 668 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 669 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 670 | */ |
5e705374 | 671 | static struct request * |
cf7c25cf | 672 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 673 | { |
cf7c25cf | 674 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 675 | unsigned long back_max; |
e8a99053 AM |
676 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
677 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
678 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 679 | |
5e705374 JA |
680 | if (rq1 == NULL || rq1 == rq2) |
681 | return rq2; | |
682 | if (rq2 == NULL) | |
683 | return rq1; | |
9c2c38a1 | 684 | |
229836bd NK |
685 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
686 | return rq_is_sync(rq1) ? rq1 : rq2; | |
687 | ||
65299a3b CH |
688 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
689 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 690 | |
83096ebf TH |
691 | s1 = blk_rq_pos(rq1); |
692 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 693 | |
1da177e4 LT |
694 | /* |
695 | * by definition, 1KiB is 2 sectors | |
696 | */ | |
697 | back_max = cfqd->cfq_back_max * 2; | |
698 | ||
699 | /* | |
700 | * Strict one way elevator _except_ in the case where we allow | |
701 | * short backward seeks which are biased as twice the cost of a | |
702 | * similar forward seek. | |
703 | */ | |
704 | if (s1 >= last) | |
705 | d1 = s1 - last; | |
706 | else if (s1 + back_max >= last) | |
707 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
708 | else | |
e8a99053 | 709 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
710 | |
711 | if (s2 >= last) | |
712 | d2 = s2 - last; | |
713 | else if (s2 + back_max >= last) | |
714 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
715 | else | |
e8a99053 | 716 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
717 | |
718 | /* Found required data */ | |
e8a99053 AM |
719 | |
720 | /* | |
721 | * By doing switch() on the bit mask "wrap" we avoid having to | |
722 | * check two variables for all permutations: --> faster! | |
723 | */ | |
724 | switch (wrap) { | |
5e705374 | 725 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 726 | if (d1 < d2) |
5e705374 | 727 | return rq1; |
e8a99053 | 728 | else if (d2 < d1) |
5e705374 | 729 | return rq2; |
e8a99053 AM |
730 | else { |
731 | if (s1 >= s2) | |
5e705374 | 732 | return rq1; |
e8a99053 | 733 | else |
5e705374 | 734 | return rq2; |
e8a99053 | 735 | } |
1da177e4 | 736 | |
e8a99053 | 737 | case CFQ_RQ2_WRAP: |
5e705374 | 738 | return rq1; |
e8a99053 | 739 | case CFQ_RQ1_WRAP: |
5e705374 JA |
740 | return rq2; |
741 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
742 | default: |
743 | /* | |
744 | * Since both rqs are wrapped, | |
745 | * start with the one that's further behind head | |
746 | * (--> only *one* back seek required), | |
747 | * since back seek takes more time than forward. | |
748 | */ | |
749 | if (s1 <= s2) | |
5e705374 | 750 | return rq1; |
1da177e4 | 751 | else |
5e705374 | 752 | return rq2; |
1da177e4 LT |
753 | } |
754 | } | |
755 | ||
498d3aa2 JA |
756 | /* |
757 | * The below is leftmost cache rbtree addon | |
758 | */ | |
0871714e | 759 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 760 | { |
615f0259 VG |
761 | /* Service tree is empty */ |
762 | if (!root->count) | |
763 | return NULL; | |
764 | ||
cc09e299 JA |
765 | if (!root->left) |
766 | root->left = rb_first(&root->rb); | |
767 | ||
0871714e JA |
768 | if (root->left) |
769 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
770 | ||
771 | return NULL; | |
cc09e299 JA |
772 | } |
773 | ||
1fa8f6d6 VG |
774 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
775 | { | |
776 | if (!root->left) | |
777 | root->left = rb_first(&root->rb); | |
778 | ||
779 | if (root->left) | |
780 | return rb_entry_cfqg(root->left); | |
781 | ||
782 | return NULL; | |
783 | } | |
784 | ||
a36e71f9 JA |
785 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
786 | { | |
787 | rb_erase(n, root); | |
788 | RB_CLEAR_NODE(n); | |
789 | } | |
790 | ||
cc09e299 JA |
791 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
792 | { | |
793 | if (root->left == n) | |
794 | root->left = NULL; | |
a36e71f9 | 795 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 796 | --root->count; |
cc09e299 JA |
797 | } |
798 | ||
1da177e4 LT |
799 | /* |
800 | * would be nice to take fifo expire time into account as well | |
801 | */ | |
5e705374 JA |
802 | static struct request * |
803 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
804 | struct request *last) | |
1da177e4 | 805 | { |
21183b07 JA |
806 | struct rb_node *rbnext = rb_next(&last->rb_node); |
807 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 808 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 809 | |
21183b07 | 810 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
811 | |
812 | if (rbprev) | |
5e705374 | 813 | prev = rb_entry_rq(rbprev); |
1da177e4 | 814 | |
21183b07 | 815 | if (rbnext) |
5e705374 | 816 | next = rb_entry_rq(rbnext); |
21183b07 JA |
817 | else { |
818 | rbnext = rb_first(&cfqq->sort_list); | |
819 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 820 | next = rb_entry_rq(rbnext); |
21183b07 | 821 | } |
1da177e4 | 822 | |
cf7c25cf | 823 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
824 | } |
825 | ||
d9e7620e JA |
826 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
827 | struct cfq_queue *cfqq) | |
1da177e4 | 828 | { |
d9e7620e JA |
829 | /* |
830 | * just an approximation, should be ok. | |
831 | */ | |
cdb16e8f | 832 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 833 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
834 | } |
835 | ||
1fa8f6d6 VG |
836 | static inline s64 |
837 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
838 | { | |
839 | return cfqg->vdisktime - st->min_vdisktime; | |
840 | } | |
841 | ||
842 | static void | |
843 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
844 | { | |
845 | struct rb_node **node = &st->rb.rb_node; | |
846 | struct rb_node *parent = NULL; | |
847 | struct cfq_group *__cfqg; | |
848 | s64 key = cfqg_key(st, cfqg); | |
849 | int left = 1; | |
850 | ||
851 | while (*node != NULL) { | |
852 | parent = *node; | |
853 | __cfqg = rb_entry_cfqg(parent); | |
854 | ||
855 | if (key < cfqg_key(st, __cfqg)) | |
856 | node = &parent->rb_left; | |
857 | else { | |
858 | node = &parent->rb_right; | |
859 | left = 0; | |
860 | } | |
861 | } | |
862 | ||
863 | if (left) | |
864 | st->left = &cfqg->rb_node; | |
865 | ||
866 | rb_link_node(&cfqg->rb_node, parent, node); | |
867 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
868 | } | |
869 | ||
870 | static void | |
8184f93e JT |
871 | cfq_update_group_weight(struct cfq_group *cfqg) |
872 | { | |
873 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
874 | if (cfqg->needs_update) { | |
875 | cfqg->weight = cfqg->new_weight; | |
876 | cfqg->needs_update = false; | |
877 | } | |
878 | } | |
879 | ||
880 | static void | |
881 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
882 | { | |
883 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
884 | ||
885 | cfq_update_group_weight(cfqg); | |
886 | __cfq_group_service_tree_add(st, cfqg); | |
887 | st->total_weight += cfqg->weight; | |
888 | } | |
889 | ||
890 | static void | |
891 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
892 | { |
893 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
894 | struct cfq_group *__cfqg; | |
895 | struct rb_node *n; | |
896 | ||
897 | cfqg->nr_cfqq++; | |
760701bf | 898 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
899 | return; |
900 | ||
901 | /* | |
902 | * Currently put the group at the end. Later implement something | |
903 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 904 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
905 | */ |
906 | n = rb_last(&st->rb); | |
907 | if (n) { | |
908 | __cfqg = rb_entry_cfqg(n); | |
909 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
910 | } else | |
911 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
912 | cfq_group_service_tree_add(st, cfqg); |
913 | } | |
1fa8f6d6 | 914 | |
8184f93e JT |
915 | static void |
916 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
917 | { | |
918 | st->total_weight -= cfqg->weight; | |
919 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) | |
920 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
921 | } |
922 | ||
923 | static void | |
8184f93e | 924 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
925 | { |
926 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
927 | ||
928 | BUG_ON(cfqg->nr_cfqq < 1); | |
929 | cfqg->nr_cfqq--; | |
25bc6b07 | 930 | |
1fa8f6d6 VG |
931 | /* If there are other cfq queues under this group, don't delete it */ |
932 | if (cfqg->nr_cfqq) | |
933 | return; | |
934 | ||
2868ef7b | 935 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 936 | cfq_group_service_tree_del(st, cfqg); |
dae739eb | 937 | cfqg->saved_workload_slice = 0; |
e98ef89b | 938 | cfq_blkiocg_update_dequeue_stats(&cfqg->blkg, 1); |
dae739eb VG |
939 | } |
940 | ||
167400d3 JT |
941 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
942 | unsigned int *unaccounted_time) | |
dae739eb | 943 | { |
f75edf2d | 944 | unsigned int slice_used; |
dae739eb VG |
945 | |
946 | /* | |
947 | * Queue got expired before even a single request completed or | |
948 | * got expired immediately after first request completion. | |
949 | */ | |
950 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
951 | /* | |
952 | * Also charge the seek time incurred to the group, otherwise | |
953 | * if there are mutiple queues in the group, each can dispatch | |
954 | * a single request on seeky media and cause lots of seek time | |
955 | * and group will never know it. | |
956 | */ | |
957 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
958 | 1); | |
959 | } else { | |
960 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
961 | if (slice_used > cfqq->allocated_slice) { |
962 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 963 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
964 | } |
965 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
966 | *unaccounted_time += cfqq->slice_start - | |
967 | cfqq->dispatch_start; | |
dae739eb VG |
968 | } |
969 | ||
dae739eb VG |
970 | return slice_used; |
971 | } | |
972 | ||
973 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 974 | struct cfq_queue *cfqq) |
dae739eb VG |
975 | { |
976 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 977 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
978 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
979 | - cfqg->service_tree_idle.count; | |
980 | ||
981 | BUG_ON(nr_sync < 0); | |
167400d3 | 982 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 983 | |
02b35081 VG |
984 | if (iops_mode(cfqd)) |
985 | charge = cfqq->slice_dispatch; | |
986 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
987 | charge = cfqq->allocated_slice; | |
dae739eb VG |
988 | |
989 | /* Can't update vdisktime while group is on service tree */ | |
8184f93e | 990 | cfq_group_service_tree_del(st, cfqg); |
02b35081 | 991 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
8184f93e JT |
992 | /* If a new weight was requested, update now, off tree */ |
993 | cfq_group_service_tree_add(st, cfqg); | |
dae739eb VG |
994 | |
995 | /* This group is being expired. Save the context */ | |
996 | if (time_after(cfqd->workload_expires, jiffies)) { | |
997 | cfqg->saved_workload_slice = cfqd->workload_expires | |
998 | - jiffies; | |
999 | cfqg->saved_workload = cfqd->serving_type; | |
1000 | cfqg->saved_serving_prio = cfqd->serving_prio; | |
1001 | } else | |
1002 | cfqg->saved_workload_slice = 0; | |
2868ef7b VG |
1003 | |
1004 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1005 | st->min_vdisktime); | |
fd16d263 JP |
1006 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1007 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1008 | used_sl, cfqq->slice_dispatch, charge, | |
1009 | iops_mode(cfqd), cfqq->nr_sectors); | |
167400d3 JT |
1010 | cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl, |
1011 | unaccounted_sl); | |
e98ef89b | 1012 | cfq_blkiocg_set_start_empty_time(&cfqg->blkg); |
1fa8f6d6 VG |
1013 | } |
1014 | ||
25fb5169 VG |
1015 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
1016 | static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg) | |
1017 | { | |
1018 | if (blkg) | |
1019 | return container_of(blkg, struct cfq_group, blkg); | |
1020 | return NULL; | |
1021 | } | |
1022 | ||
8aea4545 PB |
1023 | static void cfq_update_blkio_group_weight(void *key, struct blkio_group *blkg, |
1024 | unsigned int weight) | |
f8d461d6 | 1025 | { |
8184f93e JT |
1026 | struct cfq_group *cfqg = cfqg_of_blkg(blkg); |
1027 | cfqg->new_weight = weight; | |
1028 | cfqg->needs_update = true; | |
f8d461d6 VG |
1029 | } |
1030 | ||
f469a7b4 VG |
1031 | static void cfq_init_add_cfqg_lists(struct cfq_data *cfqd, |
1032 | struct cfq_group *cfqg, struct blkio_cgroup *blkcg) | |
25fb5169 | 1033 | { |
22084190 VG |
1034 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; |
1035 | unsigned int major, minor; | |
25fb5169 | 1036 | |
f469a7b4 VG |
1037 | /* |
1038 | * Add group onto cgroup list. It might happen that bdi->dev is | |
1039 | * not initialized yet. Initialize this new group without major | |
1040 | * and minor info and this info will be filled in once a new thread | |
1041 | * comes for IO. | |
1042 | */ | |
1043 | if (bdi->dev) { | |
a74b2ada | 1044 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); |
f469a7b4 VG |
1045 | cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, |
1046 | (void *)cfqd, MKDEV(major, minor)); | |
1047 | } else | |
1048 | cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, | |
1049 | (void *)cfqd, 0); | |
1050 | ||
1051 | cfqd->nr_blkcg_linked_grps++; | |
1052 | cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev); | |
1053 | ||
1054 | /* Add group on cfqd list */ | |
1055 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * Should be called from sleepable context. No request queue lock as per | |
1060 | * cpu stats are allocated dynamically and alloc_percpu needs to be called | |
1061 | * from sleepable context. | |
1062 | */ | |
1063 | static struct cfq_group * cfq_alloc_cfqg(struct cfq_data *cfqd) | |
1064 | { | |
1065 | struct cfq_group *cfqg = NULL; | |
5624a4e4 | 1066 | int i, j, ret; |
f469a7b4 | 1067 | struct cfq_rb_root *st; |
25fb5169 VG |
1068 | |
1069 | cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node); | |
1070 | if (!cfqg) | |
f469a7b4 | 1071 | return NULL; |
25fb5169 | 1072 | |
25fb5169 VG |
1073 | for_each_cfqg_st(cfqg, i, j, st) |
1074 | *st = CFQ_RB_ROOT; | |
1075 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1076 | ||
7700fc4f SL |
1077 | cfqg->ttime.last_end_request = jiffies; |
1078 | ||
b1c35769 VG |
1079 | /* |
1080 | * Take the initial reference that will be released on destroy | |
1081 | * This can be thought of a joint reference by cgroup and | |
1082 | * elevator which will be dropped by either elevator exit | |
1083 | * or cgroup deletion path depending on who is exiting first. | |
1084 | */ | |
329a6781 | 1085 | cfqg->ref = 1; |
5624a4e4 VG |
1086 | |
1087 | ret = blkio_alloc_blkg_stats(&cfqg->blkg); | |
1088 | if (ret) { | |
1089 | kfree(cfqg); | |
1090 | return NULL; | |
1091 | } | |
1092 | ||
f469a7b4 VG |
1093 | return cfqg; |
1094 | } | |
1095 | ||
1096 | static struct cfq_group * | |
1097 | cfq_find_cfqg(struct cfq_data *cfqd, struct blkio_cgroup *blkcg) | |
1098 | { | |
1099 | struct cfq_group *cfqg = NULL; | |
1100 | void *key = cfqd; | |
1101 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; | |
1102 | unsigned int major, minor; | |
b1c35769 | 1103 | |
180be2a0 | 1104 | /* |
f469a7b4 VG |
1105 | * This is the common case when there are no blkio cgroups. |
1106 | * Avoid lookup in this case | |
180be2a0 | 1107 | */ |
f469a7b4 VG |
1108 | if (blkcg == &blkio_root_cgroup) |
1109 | cfqg = &cfqd->root_group; | |
1110 | else | |
1111 | cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key)); | |
25fb5169 | 1112 | |
f469a7b4 VG |
1113 | if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) { |
1114 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); | |
1115 | cfqg->blkg.dev = MKDEV(major, minor); | |
1116 | } | |
25fb5169 | 1117 | |
25fb5169 VG |
1118 | return cfqg; |
1119 | } | |
1120 | ||
1121 | /* | |
3e59cf9d VG |
1122 | * Search for the cfq group current task belongs to. request_queue lock must |
1123 | * be held. | |
25fb5169 | 1124 | */ |
3e59cf9d | 1125 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd) |
25fb5169 | 1126 | { |
70087dc3 | 1127 | struct blkio_cgroup *blkcg; |
f469a7b4 VG |
1128 | struct cfq_group *cfqg = NULL, *__cfqg = NULL; |
1129 | struct request_queue *q = cfqd->queue; | |
25fb5169 VG |
1130 | |
1131 | rcu_read_lock(); | |
70087dc3 | 1132 | blkcg = task_blkio_cgroup(current); |
f469a7b4 VG |
1133 | cfqg = cfq_find_cfqg(cfqd, blkcg); |
1134 | if (cfqg) { | |
1135 | rcu_read_unlock(); | |
1136 | return cfqg; | |
1137 | } | |
1138 | ||
1139 | /* | |
1140 | * Need to allocate a group. Allocation of group also needs allocation | |
1141 | * of per cpu stats which in-turn takes a mutex() and can block. Hence | |
1142 | * we need to drop rcu lock and queue_lock before we call alloc. | |
1143 | * | |
1144 | * Not taking any queue reference here and assuming that queue is | |
1145 | * around by the time we return. CFQ queue allocation code does | |
1146 | * the same. It might be racy though. | |
1147 | */ | |
1148 | ||
1149 | rcu_read_unlock(); | |
1150 | spin_unlock_irq(q->queue_lock); | |
1151 | ||
1152 | cfqg = cfq_alloc_cfqg(cfqd); | |
1153 | ||
1154 | spin_lock_irq(q->queue_lock); | |
1155 | ||
1156 | rcu_read_lock(); | |
1157 | blkcg = task_blkio_cgroup(current); | |
1158 | ||
1159 | /* | |
1160 | * If some other thread already allocated the group while we were | |
1161 | * not holding queue lock, free up the group | |
1162 | */ | |
1163 | __cfqg = cfq_find_cfqg(cfqd, blkcg); | |
1164 | ||
1165 | if (__cfqg) { | |
1166 | kfree(cfqg); | |
1167 | rcu_read_unlock(); | |
1168 | return __cfqg; | |
1169 | } | |
1170 | ||
3e59cf9d | 1171 | if (!cfqg) |
25fb5169 | 1172 | cfqg = &cfqd->root_group; |
f469a7b4 VG |
1173 | |
1174 | cfq_init_add_cfqg_lists(cfqd, cfqg, blkcg); | |
25fb5169 VG |
1175 | rcu_read_unlock(); |
1176 | return cfqg; | |
1177 | } | |
1178 | ||
7f1dc8a2 VG |
1179 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) |
1180 | { | |
329a6781 | 1181 | cfqg->ref++; |
7f1dc8a2 VG |
1182 | return cfqg; |
1183 | } | |
1184 | ||
25fb5169 VG |
1185 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) |
1186 | { | |
1187 | /* Currently, all async queues are mapped to root group */ | |
1188 | if (!cfq_cfqq_sync(cfqq)) | |
1189 | cfqg = &cfqq->cfqd->root_group; | |
1190 | ||
1191 | cfqq->cfqg = cfqg; | |
b1c35769 | 1192 | /* cfqq reference on cfqg */ |
329a6781 | 1193 | cfqq->cfqg->ref++; |
b1c35769 VG |
1194 | } |
1195 | ||
1196 | static void cfq_put_cfqg(struct cfq_group *cfqg) | |
1197 | { | |
1198 | struct cfq_rb_root *st; | |
1199 | int i, j; | |
1200 | ||
329a6781 SL |
1201 | BUG_ON(cfqg->ref <= 0); |
1202 | cfqg->ref--; | |
1203 | if (cfqg->ref) | |
b1c35769 VG |
1204 | return; |
1205 | for_each_cfqg_st(cfqg, i, j, st) | |
b54ce60e | 1206 | BUG_ON(!RB_EMPTY_ROOT(&st->rb)); |
5624a4e4 | 1207 | free_percpu(cfqg->blkg.stats_cpu); |
b1c35769 VG |
1208 | kfree(cfqg); |
1209 | } | |
1210 | ||
1211 | static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1212 | { | |
1213 | /* Something wrong if we are trying to remove same group twice */ | |
1214 | BUG_ON(hlist_unhashed(&cfqg->cfqd_node)); | |
1215 | ||
1216 | hlist_del_init(&cfqg->cfqd_node); | |
1217 | ||
a5395b83 VG |
1218 | BUG_ON(cfqd->nr_blkcg_linked_grps <= 0); |
1219 | cfqd->nr_blkcg_linked_grps--; | |
1220 | ||
b1c35769 VG |
1221 | /* |
1222 | * Put the reference taken at the time of creation so that when all | |
1223 | * queues are gone, group can be destroyed. | |
1224 | */ | |
1225 | cfq_put_cfqg(cfqg); | |
1226 | } | |
1227 | ||
1228 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) | |
1229 | { | |
1230 | struct hlist_node *pos, *n; | |
1231 | struct cfq_group *cfqg; | |
1232 | ||
1233 | hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) { | |
1234 | /* | |
1235 | * If cgroup removal path got to blk_group first and removed | |
1236 | * it from cgroup list, then it will take care of destroying | |
1237 | * cfqg also. | |
1238 | */ | |
e98ef89b | 1239 | if (!cfq_blkiocg_del_blkio_group(&cfqg->blkg)) |
b1c35769 VG |
1240 | cfq_destroy_cfqg(cfqd, cfqg); |
1241 | } | |
25fb5169 | 1242 | } |
b1c35769 VG |
1243 | |
1244 | /* | |
1245 | * Blk cgroup controller notification saying that blkio_group object is being | |
1246 | * delinked as associated cgroup object is going away. That also means that | |
1247 | * no new IO will come in this group. So get rid of this group as soon as | |
1248 | * any pending IO in the group is finished. | |
1249 | * | |
1250 | * This function is called under rcu_read_lock(). key is the rcu protected | |
1251 | * pointer. That means "key" is a valid cfq_data pointer as long as we are rcu | |
1252 | * read lock. | |
1253 | * | |
1254 | * "key" was fetched from blkio_group under blkio_cgroup->lock. That means | |
1255 | * it should not be NULL as even if elevator was exiting, cgroup deltion | |
1256 | * path got to it first. | |
1257 | */ | |
8aea4545 | 1258 | static void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg) |
b1c35769 VG |
1259 | { |
1260 | unsigned long flags; | |
1261 | struct cfq_data *cfqd = key; | |
1262 | ||
1263 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1264 | cfq_destroy_cfqg(cfqd, cfqg_of_blkg(blkg)); | |
1265 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1266 | } | |
1267 | ||
25fb5169 | 1268 | #else /* GROUP_IOSCHED */ |
3e59cf9d | 1269 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd) |
25fb5169 VG |
1270 | { |
1271 | return &cfqd->root_group; | |
1272 | } | |
7f1dc8a2 VG |
1273 | |
1274 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) | |
1275 | { | |
50eaeb32 | 1276 | return cfqg; |
7f1dc8a2 VG |
1277 | } |
1278 | ||
25fb5169 VG |
1279 | static inline void |
1280 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1281 | cfqq->cfqg = cfqg; | |
1282 | } | |
1283 | ||
b1c35769 VG |
1284 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) {} |
1285 | static inline void cfq_put_cfqg(struct cfq_group *cfqg) {} | |
1286 | ||
25fb5169 VG |
1287 | #endif /* GROUP_IOSCHED */ |
1288 | ||
498d3aa2 | 1289 | /* |
c0324a02 | 1290 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1291 | * requests waiting to be processed. It is sorted in the order that |
1292 | * we will service the queues. | |
1293 | */ | |
a36e71f9 | 1294 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1295 | bool add_front) |
d9e7620e | 1296 | { |
0871714e JA |
1297 | struct rb_node **p, *parent; |
1298 | struct cfq_queue *__cfqq; | |
d9e7620e | 1299 | unsigned long rb_key; |
c0324a02 | 1300 | struct cfq_rb_root *service_tree; |
498d3aa2 | 1301 | int left; |
dae739eb | 1302 | int new_cfqq = 1; |
ae30c286 | 1303 | |
cdb16e8f | 1304 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
65b32a57 | 1305 | cfqq_type(cfqq)); |
0871714e JA |
1306 | if (cfq_class_idle(cfqq)) { |
1307 | rb_key = CFQ_IDLE_DELAY; | |
aa6f6a3d | 1308 | parent = rb_last(&service_tree->rb); |
0871714e JA |
1309 | if (parent && parent != &cfqq->rb_node) { |
1310 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1311 | rb_key += __cfqq->rb_key; | |
1312 | } else | |
1313 | rb_key += jiffies; | |
1314 | } else if (!add_front) { | |
b9c8946b JA |
1315 | /* |
1316 | * Get our rb key offset. Subtract any residual slice | |
1317 | * value carried from last service. A negative resid | |
1318 | * count indicates slice overrun, and this should position | |
1319 | * the next service time further away in the tree. | |
1320 | */ | |
edd75ffd | 1321 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1322 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1323 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1324 | } else { |
1325 | rb_key = -HZ; | |
aa6f6a3d | 1326 | __cfqq = cfq_rb_first(service_tree); |
48e025e6 CZ |
1327 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1328 | } | |
1da177e4 | 1329 | |
d9e7620e | 1330 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1331 | new_cfqq = 0; |
99f9628a | 1332 | /* |
d9e7620e | 1333 | * same position, nothing more to do |
99f9628a | 1334 | */ |
c0324a02 CZ |
1335 | if (rb_key == cfqq->rb_key && |
1336 | cfqq->service_tree == service_tree) | |
d9e7620e | 1337 | return; |
1da177e4 | 1338 | |
aa6f6a3d CZ |
1339 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1340 | cfqq->service_tree = NULL; | |
1da177e4 | 1341 | } |
d9e7620e | 1342 | |
498d3aa2 | 1343 | left = 1; |
0871714e | 1344 | parent = NULL; |
aa6f6a3d CZ |
1345 | cfqq->service_tree = service_tree; |
1346 | p = &service_tree->rb.rb_node; | |
d9e7620e | 1347 | while (*p) { |
67060e37 | 1348 | struct rb_node **n; |
cc09e299 | 1349 | |
d9e7620e JA |
1350 | parent = *p; |
1351 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1352 | ||
0c534e0a | 1353 | /* |
c0324a02 | 1354 | * sort by key, that represents service time. |
0c534e0a | 1355 | */ |
c0324a02 | 1356 | if (time_before(rb_key, __cfqq->rb_key)) |
67060e37 | 1357 | n = &(*p)->rb_left; |
c0324a02 | 1358 | else { |
67060e37 | 1359 | n = &(*p)->rb_right; |
cc09e299 | 1360 | left = 0; |
c0324a02 | 1361 | } |
67060e37 JA |
1362 | |
1363 | p = n; | |
d9e7620e JA |
1364 | } |
1365 | ||
cc09e299 | 1366 | if (left) |
aa6f6a3d | 1367 | service_tree->left = &cfqq->rb_node; |
cc09e299 | 1368 | |
d9e7620e JA |
1369 | cfqq->rb_key = rb_key; |
1370 | rb_link_node(&cfqq->rb_node, parent, p); | |
aa6f6a3d CZ |
1371 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
1372 | service_tree->count++; | |
20359f27 | 1373 | if (add_front || !new_cfqq) |
dae739eb | 1374 | return; |
8184f93e | 1375 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1376 | } |
1377 | ||
a36e71f9 | 1378 | static struct cfq_queue * |
f2d1f0ae JA |
1379 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1380 | sector_t sector, struct rb_node **ret_parent, | |
1381 | struct rb_node ***rb_link) | |
a36e71f9 | 1382 | { |
a36e71f9 JA |
1383 | struct rb_node **p, *parent; |
1384 | struct cfq_queue *cfqq = NULL; | |
1385 | ||
1386 | parent = NULL; | |
1387 | p = &root->rb_node; | |
1388 | while (*p) { | |
1389 | struct rb_node **n; | |
1390 | ||
1391 | parent = *p; | |
1392 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1393 | ||
1394 | /* | |
1395 | * Sort strictly based on sector. Smallest to the left, | |
1396 | * largest to the right. | |
1397 | */ | |
2e46e8b2 | 1398 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1399 | n = &(*p)->rb_right; |
2e46e8b2 | 1400 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1401 | n = &(*p)->rb_left; |
1402 | else | |
1403 | break; | |
1404 | p = n; | |
3ac6c9f8 | 1405 | cfqq = NULL; |
a36e71f9 JA |
1406 | } |
1407 | ||
1408 | *ret_parent = parent; | |
1409 | if (rb_link) | |
1410 | *rb_link = p; | |
3ac6c9f8 | 1411 | return cfqq; |
a36e71f9 JA |
1412 | } |
1413 | ||
1414 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1415 | { | |
a36e71f9 JA |
1416 | struct rb_node **p, *parent; |
1417 | struct cfq_queue *__cfqq; | |
1418 | ||
f2d1f0ae JA |
1419 | if (cfqq->p_root) { |
1420 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1421 | cfqq->p_root = NULL; | |
1422 | } | |
a36e71f9 JA |
1423 | |
1424 | if (cfq_class_idle(cfqq)) | |
1425 | return; | |
1426 | if (!cfqq->next_rq) | |
1427 | return; | |
1428 | ||
f2d1f0ae | 1429 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1430 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1431 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1432 | if (!__cfqq) { |
1433 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1434 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1435 | } else | |
1436 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1437 | } |
1438 | ||
498d3aa2 JA |
1439 | /* |
1440 | * Update cfqq's position in the service tree. | |
1441 | */ | |
edd75ffd | 1442 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1443 | { |
6d048f53 JA |
1444 | /* |
1445 | * Resorting requires the cfqq to be on the RR list already. | |
1446 | */ | |
a36e71f9 | 1447 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1448 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1449 | cfq_prio_tree_add(cfqd, cfqq); |
1450 | } | |
6d048f53 JA |
1451 | } |
1452 | ||
1da177e4 LT |
1453 | /* |
1454 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1455 | * the pending list according to last request service |
1da177e4 | 1456 | */ |
febffd61 | 1457 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1458 | { |
7b679138 | 1459 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1460 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1461 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 1462 | cfqd->busy_queues++; |
ef8a41df SL |
1463 | if (cfq_cfqq_sync(cfqq)) |
1464 | cfqd->busy_sync_queues++; | |
1da177e4 | 1465 | |
edd75ffd | 1466 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1467 | } |
1468 | ||
498d3aa2 JA |
1469 | /* |
1470 | * Called when the cfqq no longer has requests pending, remove it from | |
1471 | * the service tree. | |
1472 | */ | |
febffd61 | 1473 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1474 | { |
7b679138 | 1475 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1476 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1477 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1478 | |
aa6f6a3d CZ |
1479 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1480 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1481 | cfqq->service_tree = NULL; | |
1482 | } | |
f2d1f0ae JA |
1483 | if (cfqq->p_root) { |
1484 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1485 | cfqq->p_root = NULL; | |
1486 | } | |
d9e7620e | 1487 | |
8184f93e | 1488 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1489 | BUG_ON(!cfqd->busy_queues); |
1490 | cfqd->busy_queues--; | |
ef8a41df SL |
1491 | if (cfq_cfqq_sync(cfqq)) |
1492 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
1493 | } |
1494 | ||
1495 | /* | |
1496 | * rb tree support functions | |
1497 | */ | |
febffd61 | 1498 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1499 | { |
5e705374 | 1500 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1501 | const int sync = rq_is_sync(rq); |
1da177e4 | 1502 | |
b4878f24 JA |
1503 | BUG_ON(!cfqq->queued[sync]); |
1504 | cfqq->queued[sync]--; | |
1da177e4 | 1505 | |
5e705374 | 1506 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1507 | |
f04a6424 VG |
1508 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1509 | /* | |
1510 | * Queue will be deleted from service tree when we actually | |
1511 | * expire it later. Right now just remove it from prio tree | |
1512 | * as it is empty. | |
1513 | */ | |
1514 | if (cfqq->p_root) { | |
1515 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1516 | cfqq->p_root = NULL; | |
1517 | } | |
1518 | } | |
1da177e4 LT |
1519 | } |
1520 | ||
5e705374 | 1521 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1522 | { |
5e705374 | 1523 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1524 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 1525 | struct request *prev; |
1da177e4 | 1526 | |
5380a101 | 1527 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 1528 | |
796d5116 | 1529 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
1530 | |
1531 | if (!cfq_cfqq_on_rr(cfqq)) | |
1532 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1533 | |
1534 | /* | |
1535 | * check if this request is a better next-serve candidate | |
1536 | */ | |
a36e71f9 | 1537 | prev = cfqq->next_rq; |
cf7c25cf | 1538 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1539 | |
1540 | /* | |
1541 | * adjust priority tree position, if ->next_rq changes | |
1542 | */ | |
1543 | if (prev != cfqq->next_rq) | |
1544 | cfq_prio_tree_add(cfqd, cfqq); | |
1545 | ||
5044eed4 | 1546 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1547 | } |
1548 | ||
febffd61 | 1549 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1550 | { |
5380a101 JA |
1551 | elv_rb_del(&cfqq->sort_list, rq); |
1552 | cfqq->queued[rq_is_sync(rq)]--; | |
e98ef89b VG |
1553 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, |
1554 | rq_data_dir(rq), rq_is_sync(rq)); | |
5e705374 | 1555 | cfq_add_rq_rb(rq); |
e98ef89b | 1556 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, |
7f1dc8a2 VG |
1557 | &cfqq->cfqd->serving_group->blkg, rq_data_dir(rq), |
1558 | rq_is_sync(rq)); | |
1da177e4 LT |
1559 | } |
1560 | ||
206dc69b JA |
1561 | static struct request * |
1562 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1563 | { |
206dc69b | 1564 | struct task_struct *tsk = current; |
c5869807 | 1565 | struct cfq_io_cq *cic; |
206dc69b | 1566 | struct cfq_queue *cfqq; |
1da177e4 | 1567 | |
4ac845a2 | 1568 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1569 | if (!cic) |
1570 | return NULL; | |
1571 | ||
1572 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
1573 | if (cfqq) { |
1574 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
1575 | ||
21183b07 | 1576 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 1577 | } |
1da177e4 | 1578 | |
1da177e4 LT |
1579 | return NULL; |
1580 | } | |
1581 | ||
165125e1 | 1582 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1583 | { |
22e2c507 | 1584 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 1585 | |
53c583d2 | 1586 | cfqd->rq_in_driver++; |
7b679138 | 1587 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 1588 | cfqd->rq_in_driver); |
25776e35 | 1589 | |
5b93629b | 1590 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
1591 | } |
1592 | ||
165125e1 | 1593 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1594 | { |
b4878f24 JA |
1595 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1596 | ||
53c583d2 CZ |
1597 | WARN_ON(!cfqd->rq_in_driver); |
1598 | cfqd->rq_in_driver--; | |
7b679138 | 1599 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 1600 | cfqd->rq_in_driver); |
1da177e4 LT |
1601 | } |
1602 | ||
b4878f24 | 1603 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 1604 | { |
5e705374 | 1605 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 1606 | |
5e705374 JA |
1607 | if (cfqq->next_rq == rq) |
1608 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 1609 | |
b4878f24 | 1610 | list_del_init(&rq->queuelist); |
5e705374 | 1611 | cfq_del_rq_rb(rq); |
374f84ac | 1612 | |
45333d5a | 1613 | cfqq->cfqd->rq_queued--; |
e98ef89b VG |
1614 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, |
1615 | rq_data_dir(rq), rq_is_sync(rq)); | |
65299a3b CH |
1616 | if (rq->cmd_flags & REQ_PRIO) { |
1617 | WARN_ON(!cfqq->prio_pending); | |
1618 | cfqq->prio_pending--; | |
b53d1ed7 | 1619 | } |
1da177e4 LT |
1620 | } |
1621 | ||
165125e1 JA |
1622 | static int cfq_merge(struct request_queue *q, struct request **req, |
1623 | struct bio *bio) | |
1da177e4 LT |
1624 | { |
1625 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1626 | struct request *__rq; | |
1da177e4 | 1627 | |
206dc69b | 1628 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 1629 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
1630 | *req = __rq; |
1631 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
1632 | } |
1633 | ||
1634 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
1635 | } |
1636 | ||
165125e1 | 1637 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 1638 | int type) |
1da177e4 | 1639 | { |
21183b07 | 1640 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 1641 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 1642 | |
5e705374 | 1643 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 1644 | } |
1da177e4 LT |
1645 | } |
1646 | ||
812d4026 DS |
1647 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
1648 | struct bio *bio) | |
1649 | { | |
e98ef89b VG |
1650 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(req))->blkg, |
1651 | bio_data_dir(bio), cfq_bio_sync(bio)); | |
812d4026 DS |
1652 | } |
1653 | ||
1da177e4 | 1654 | static void |
165125e1 | 1655 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
1656 | struct request *next) |
1657 | { | |
cf7c25cf | 1658 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1659 | /* |
1660 | * reposition in fifo if next is older than rq | |
1661 | */ | |
1662 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
30996f40 | 1663 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
22e2c507 | 1664 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
1665 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1666 | } | |
22e2c507 | 1667 | |
cf7c25cf CZ |
1668 | if (cfqq->next_rq == next) |
1669 | cfqq->next_rq = rq; | |
b4878f24 | 1670 | cfq_remove_request(next); |
e98ef89b VG |
1671 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(rq))->blkg, |
1672 | rq_data_dir(next), rq_is_sync(next)); | |
22e2c507 JA |
1673 | } |
1674 | ||
165125e1 | 1675 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
1676 | struct bio *bio) |
1677 | { | |
1678 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 1679 | struct cfq_io_cq *cic; |
da775265 | 1680 | struct cfq_queue *cfqq; |
da775265 JA |
1681 | |
1682 | /* | |
ec8acb69 | 1683 | * Disallow merge of a sync bio into an async request. |
da775265 | 1684 | */ |
91fac317 | 1685 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 1686 | return false; |
da775265 JA |
1687 | |
1688 | /* | |
f1a4f4d3 TH |
1689 | * Lookup the cfqq that this bio will be queued with and allow |
1690 | * merge only if rq is queued there. This function can be called | |
1691 | * from plug merge without queue_lock. In such cases, ioc of @rq | |
1692 | * and %current are guaranteed to be equal. Avoid lookup which | |
1693 | * requires queue_lock by using @rq's cic. | |
1694 | */ | |
c5869807 | 1695 | if (current->io_context == RQ_CIC(rq)->icq.ioc) { |
f1a4f4d3 TH |
1696 | cic = RQ_CIC(rq); |
1697 | } else { | |
1698 | cic = cfq_cic_lookup(cfqd, current->io_context); | |
1699 | if (!cic) | |
1700 | return false; | |
1701 | } | |
719d3402 | 1702 | |
91fac317 | 1703 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 1704 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
1705 | } |
1706 | ||
812df48d DS |
1707 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1708 | { | |
1709 | del_timer(&cfqd->idle_slice_timer); | |
e98ef89b | 1710 | cfq_blkiocg_update_idle_time_stats(&cfqq->cfqg->blkg); |
812df48d DS |
1711 | } |
1712 | ||
febffd61 JA |
1713 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
1714 | struct cfq_queue *cfqq) | |
22e2c507 JA |
1715 | { |
1716 | if (cfqq) { | |
b1ffe737 DS |
1717 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", |
1718 | cfqd->serving_prio, cfqd->serving_type); | |
62a37f6b JT |
1719 | cfq_blkiocg_update_avg_queue_size_stats(&cfqq->cfqg->blkg); |
1720 | cfqq->slice_start = 0; | |
1721 | cfqq->dispatch_start = jiffies; | |
1722 | cfqq->allocated_slice = 0; | |
1723 | cfqq->slice_end = 0; | |
1724 | cfqq->slice_dispatch = 0; | |
1725 | cfqq->nr_sectors = 0; | |
1726 | ||
1727 | cfq_clear_cfqq_wait_request(cfqq); | |
1728 | cfq_clear_cfqq_must_dispatch(cfqq); | |
1729 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
1730 | cfq_clear_cfqq_fifo_expire(cfqq); | |
1731 | cfq_mark_cfqq_slice_new(cfqq); | |
1732 | ||
1733 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
1734 | } |
1735 | ||
1736 | cfqd->active_queue = cfqq; | |
1737 | } | |
1738 | ||
7b14e3b5 JA |
1739 | /* |
1740 | * current cfqq expired its slice (or was too idle), select new one | |
1741 | */ | |
1742 | static void | |
1743 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 1744 | bool timed_out) |
7b14e3b5 | 1745 | { |
7b679138 JA |
1746 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
1747 | ||
7b14e3b5 | 1748 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 1749 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 1750 | |
7b14e3b5 | 1751 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 1752 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 1753 | |
ae54abed SL |
1754 | /* |
1755 | * If this cfqq is shared between multiple processes, check to | |
1756 | * make sure that those processes are still issuing I/Os within | |
1757 | * the mean seek distance. If not, it may be time to break the | |
1758 | * queues apart again. | |
1759 | */ | |
1760 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
1761 | cfq_mark_cfqq_split_coop(cfqq); | |
1762 | ||
7b14e3b5 | 1763 | /* |
6084cdda | 1764 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 1765 | */ |
c553f8e3 SL |
1766 | if (timed_out) { |
1767 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 1768 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
1769 | else |
1770 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
1771 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
1772 | } | |
7b14e3b5 | 1773 | |
e5ff082e | 1774 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 1775 | |
f04a6424 VG |
1776 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1777 | cfq_del_cfqq_rr(cfqd, cfqq); | |
1778 | ||
edd75ffd | 1779 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
1780 | |
1781 | if (cfqq == cfqd->active_queue) | |
1782 | cfqd->active_queue = NULL; | |
1783 | ||
1784 | if (cfqd->active_cic) { | |
c5869807 | 1785 | put_io_context(cfqd->active_cic->icq.ioc, cfqd->queue); |
7b14e3b5 JA |
1786 | cfqd->active_cic = NULL; |
1787 | } | |
7b14e3b5 JA |
1788 | } |
1789 | ||
e5ff082e | 1790 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
1791 | { |
1792 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1793 | ||
1794 | if (cfqq) | |
e5ff082e | 1795 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
1796 | } |
1797 | ||
498d3aa2 JA |
1798 | /* |
1799 | * Get next queue for service. Unless we have a queue preemption, | |
1800 | * we'll simply select the first cfqq in the service tree. | |
1801 | */ | |
6d048f53 | 1802 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 1803 | { |
c0324a02 | 1804 | struct cfq_rb_root *service_tree = |
cdb16e8f | 1805 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
65b32a57 | 1806 | cfqd->serving_type); |
d9e7620e | 1807 | |
f04a6424 VG |
1808 | if (!cfqd->rq_queued) |
1809 | return NULL; | |
1810 | ||
1fa8f6d6 VG |
1811 | /* There is nothing to dispatch */ |
1812 | if (!service_tree) | |
1813 | return NULL; | |
c0324a02 CZ |
1814 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
1815 | return NULL; | |
1816 | return cfq_rb_first(service_tree); | |
6d048f53 JA |
1817 | } |
1818 | ||
f04a6424 VG |
1819 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
1820 | { | |
25fb5169 | 1821 | struct cfq_group *cfqg; |
f04a6424 VG |
1822 | struct cfq_queue *cfqq; |
1823 | int i, j; | |
1824 | struct cfq_rb_root *st; | |
1825 | ||
1826 | if (!cfqd->rq_queued) | |
1827 | return NULL; | |
1828 | ||
25fb5169 VG |
1829 | cfqg = cfq_get_next_cfqg(cfqd); |
1830 | if (!cfqg) | |
1831 | return NULL; | |
1832 | ||
f04a6424 VG |
1833 | for_each_cfqg_st(cfqg, i, j, st) |
1834 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
1835 | return cfqq; | |
1836 | return NULL; | |
1837 | } | |
1838 | ||
498d3aa2 JA |
1839 | /* |
1840 | * Get and set a new active queue for service. | |
1841 | */ | |
a36e71f9 JA |
1842 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
1843 | struct cfq_queue *cfqq) | |
6d048f53 | 1844 | { |
e00ef799 | 1845 | if (!cfqq) |
a36e71f9 | 1846 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 1847 | |
22e2c507 | 1848 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 1849 | return cfqq; |
22e2c507 JA |
1850 | } |
1851 | ||
d9e7620e JA |
1852 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
1853 | struct request *rq) | |
1854 | { | |
83096ebf TH |
1855 | if (blk_rq_pos(rq) >= cfqd->last_position) |
1856 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 1857 | else |
83096ebf | 1858 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
1859 | } |
1860 | ||
b2c18e1e | 1861 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 1862 | struct request *rq) |
6d048f53 | 1863 | { |
e9ce335d | 1864 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
1865 | } |
1866 | ||
a36e71f9 JA |
1867 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
1868 | struct cfq_queue *cur_cfqq) | |
1869 | { | |
f2d1f0ae | 1870 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
1871 | struct rb_node *parent, *node; |
1872 | struct cfq_queue *__cfqq; | |
1873 | sector_t sector = cfqd->last_position; | |
1874 | ||
1875 | if (RB_EMPTY_ROOT(root)) | |
1876 | return NULL; | |
1877 | ||
1878 | /* | |
1879 | * First, if we find a request starting at the end of the last | |
1880 | * request, choose it. | |
1881 | */ | |
f2d1f0ae | 1882 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
1883 | if (__cfqq) |
1884 | return __cfqq; | |
1885 | ||
1886 | /* | |
1887 | * If the exact sector wasn't found, the parent of the NULL leaf | |
1888 | * will contain the closest sector. | |
1889 | */ | |
1890 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 1891 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1892 | return __cfqq; |
1893 | ||
2e46e8b2 | 1894 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
1895 | node = rb_next(&__cfqq->p_node); |
1896 | else | |
1897 | node = rb_prev(&__cfqq->p_node); | |
1898 | if (!node) | |
1899 | return NULL; | |
1900 | ||
1901 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 1902 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1903 | return __cfqq; |
1904 | ||
1905 | return NULL; | |
1906 | } | |
1907 | ||
1908 | /* | |
1909 | * cfqd - obvious | |
1910 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
1911 | * closely cooperating with itself. | |
1912 | * | |
1913 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
1914 | * one request, and that cfqd->last_position reflects a position on the disk | |
1915 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
1916 | * assumption. | |
1917 | */ | |
1918 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 1919 | struct cfq_queue *cur_cfqq) |
6d048f53 | 1920 | { |
a36e71f9 JA |
1921 | struct cfq_queue *cfqq; |
1922 | ||
39c01b21 DS |
1923 | if (cfq_class_idle(cur_cfqq)) |
1924 | return NULL; | |
e6c5bc73 JM |
1925 | if (!cfq_cfqq_sync(cur_cfqq)) |
1926 | return NULL; | |
1927 | if (CFQQ_SEEKY(cur_cfqq)) | |
1928 | return NULL; | |
1929 | ||
b9d8f4c7 GJ |
1930 | /* |
1931 | * Don't search priority tree if it's the only queue in the group. | |
1932 | */ | |
1933 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
1934 | return NULL; | |
1935 | ||
6d048f53 | 1936 | /* |
d9e7620e JA |
1937 | * We should notice if some of the queues are cooperating, eg |
1938 | * working closely on the same area of the disk. In that case, | |
1939 | * we can group them together and don't waste time idling. | |
6d048f53 | 1940 | */ |
a36e71f9 JA |
1941 | cfqq = cfqq_close(cfqd, cur_cfqq); |
1942 | if (!cfqq) | |
1943 | return NULL; | |
1944 | ||
8682e1f1 VG |
1945 | /* If new queue belongs to different cfq_group, don't choose it */ |
1946 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
1947 | return NULL; | |
1948 | ||
df5fe3e8 JM |
1949 | /* |
1950 | * It only makes sense to merge sync queues. | |
1951 | */ | |
1952 | if (!cfq_cfqq_sync(cfqq)) | |
1953 | return NULL; | |
e6c5bc73 JM |
1954 | if (CFQQ_SEEKY(cfqq)) |
1955 | return NULL; | |
df5fe3e8 | 1956 | |
c0324a02 CZ |
1957 | /* |
1958 | * Do not merge queues of different priority classes | |
1959 | */ | |
1960 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
1961 | return NULL; | |
1962 | ||
a36e71f9 | 1963 | return cfqq; |
6d048f53 JA |
1964 | } |
1965 | ||
a6d44e98 CZ |
1966 | /* |
1967 | * Determine whether we should enforce idle window for this queue. | |
1968 | */ | |
1969 | ||
1970 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1971 | { | |
1972 | enum wl_prio_t prio = cfqq_prio(cfqq); | |
718eee05 | 1973 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
a6d44e98 | 1974 | |
f04a6424 VG |
1975 | BUG_ON(!service_tree); |
1976 | BUG_ON(!service_tree->count); | |
1977 | ||
b6508c16 VG |
1978 | if (!cfqd->cfq_slice_idle) |
1979 | return false; | |
1980 | ||
a6d44e98 CZ |
1981 | /* We never do for idle class queues. */ |
1982 | if (prio == IDLE_WORKLOAD) | |
1983 | return false; | |
1984 | ||
1985 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
1986 | if (cfq_cfqq_idle_window(cfqq) && |
1987 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
1988 | return true; |
1989 | ||
1990 | /* | |
1991 | * Otherwise, we do only if they are the last ones | |
1992 | * in their service tree. | |
1993 | */ | |
f5f2b6ce SL |
1994 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq) && |
1995 | !cfq_io_thinktime_big(cfqd, &service_tree->ttime, false)) | |
c1e44756 | 1996 | return true; |
b1ffe737 DS |
1997 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", |
1998 | service_tree->count); | |
c1e44756 | 1999 | return false; |
a6d44e98 CZ |
2000 | } |
2001 | ||
6d048f53 | 2002 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 2003 | { |
1792669c | 2004 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 2005 | struct cfq_io_cq *cic; |
80bdf0c7 | 2006 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 2007 | |
a68bbddb | 2008 | /* |
f7d7b7a7 JA |
2009 | * SSD device without seek penalty, disable idling. But only do so |
2010 | * for devices that support queuing, otherwise we still have a problem | |
2011 | * with sync vs async workloads. | |
a68bbddb | 2012 | */ |
f7d7b7a7 | 2013 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
2014 | return; |
2015 | ||
dd67d051 | 2016 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 2017 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
2018 | |
2019 | /* | |
2020 | * idle is disabled, either manually or by past process history | |
2021 | */ | |
80bdf0c7 VG |
2022 | if (!cfq_should_idle(cfqd, cfqq)) { |
2023 | /* no queue idling. Check for group idling */ | |
2024 | if (cfqd->cfq_group_idle) | |
2025 | group_idle = cfqd->cfq_group_idle; | |
2026 | else | |
2027 | return; | |
2028 | } | |
6d048f53 | 2029 | |
7b679138 | 2030 | /* |
8e550632 | 2031 | * still active requests from this queue, don't idle |
7b679138 | 2032 | */ |
8e550632 | 2033 | if (cfqq->dispatched) |
7b679138 JA |
2034 | return; |
2035 | ||
22e2c507 JA |
2036 | /* |
2037 | * task has exited, don't wait | |
2038 | */ | |
206dc69b | 2039 | cic = cfqd->active_cic; |
c5869807 | 2040 | if (!cic || !atomic_read(&cic->icq.ioc->nr_tasks)) |
6d048f53 JA |
2041 | return; |
2042 | ||
355b659c CZ |
2043 | /* |
2044 | * If our average think time is larger than the remaining time | |
2045 | * slice, then don't idle. This avoids overrunning the allotted | |
2046 | * time slice. | |
2047 | */ | |
383cd721 SL |
2048 | if (sample_valid(cic->ttime.ttime_samples) && |
2049 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 2050 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 2051 | cic->ttime.ttime_mean); |
355b659c | 2052 | return; |
b1ffe737 | 2053 | } |
355b659c | 2054 | |
80bdf0c7 VG |
2055 | /* There are other queues in the group, don't do group idle */ |
2056 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
2057 | return; | |
2058 | ||
3b18152c | 2059 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2060 | |
80bdf0c7 VG |
2061 | if (group_idle) |
2062 | sl = cfqd->cfq_group_idle; | |
2063 | else | |
2064 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2065 | |
7b14e3b5 | 2066 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
e98ef89b | 2067 | cfq_blkiocg_update_set_idle_time_stats(&cfqq->cfqg->blkg); |
80bdf0c7 VG |
2068 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
2069 | group_idle ? 1 : 0); | |
1da177e4 LT |
2070 | } |
2071 | ||
498d3aa2 JA |
2072 | /* |
2073 | * Move request from internal lists to the request queue dispatch list. | |
2074 | */ | |
165125e1 | 2075 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2076 | { |
3ed9a296 | 2077 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2078 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2079 | |
7b679138 JA |
2080 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2081 | ||
06d21886 | 2082 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2083 | cfq_remove_request(rq); |
6d048f53 | 2084 | cfqq->dispatched++; |
80bdf0c7 | 2085 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2086 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2087 | |
53c583d2 | 2088 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2089 | cfqq->nr_sectors += blk_rq_sectors(rq); |
e98ef89b | 2090 | cfq_blkiocg_update_dispatch_stats(&cfqq->cfqg->blkg, blk_rq_bytes(rq), |
84c124da | 2091 | rq_data_dir(rq), rq_is_sync(rq)); |
1da177e4 LT |
2092 | } |
2093 | ||
2094 | /* | |
2095 | * return expired entry, or NULL to just start from scratch in rbtree | |
2096 | */ | |
febffd61 | 2097 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 2098 | { |
30996f40 | 2099 | struct request *rq = NULL; |
1da177e4 | 2100 | |
3b18152c | 2101 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 2102 | return NULL; |
cb887411 JA |
2103 | |
2104 | cfq_mark_cfqq_fifo_expire(cfqq); | |
2105 | ||
89850f7e JA |
2106 | if (list_empty(&cfqq->fifo)) |
2107 | return NULL; | |
1da177e4 | 2108 | |
89850f7e | 2109 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 2110 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 2111 | rq = NULL; |
1da177e4 | 2112 | |
30996f40 | 2113 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 2114 | return rq; |
1da177e4 LT |
2115 | } |
2116 | ||
22e2c507 JA |
2117 | static inline int |
2118 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2119 | { | |
2120 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2121 | |
22e2c507 | 2122 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2123 | |
b9f8ce05 | 2124 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
2125 | } |
2126 | ||
df5fe3e8 JM |
2127 | /* |
2128 | * Must be called with the queue_lock held. | |
2129 | */ | |
2130 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2131 | { | |
2132 | int process_refs, io_refs; | |
2133 | ||
2134 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 2135 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
2136 | BUG_ON(process_refs < 0); |
2137 | return process_refs; | |
2138 | } | |
2139 | ||
2140 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2141 | { | |
e6c5bc73 | 2142 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2143 | struct cfq_queue *__cfqq; |
2144 | ||
c10b61f0 JM |
2145 | /* |
2146 | * If there are no process references on the new_cfqq, then it is | |
2147 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2148 | * chain may have dropped their last reference (not just their | |
2149 | * last process reference). | |
2150 | */ | |
2151 | if (!cfqq_process_refs(new_cfqq)) | |
2152 | return; | |
2153 | ||
df5fe3e8 JM |
2154 | /* Avoid a circular list and skip interim queue merges */ |
2155 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2156 | if (__cfqq == cfqq) | |
2157 | return; | |
2158 | new_cfqq = __cfqq; | |
2159 | } | |
2160 | ||
2161 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2162 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2163 | /* |
2164 | * If the process for the cfqq has gone away, there is no | |
2165 | * sense in merging the queues. | |
2166 | */ | |
c10b61f0 | 2167 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2168 | return; |
2169 | ||
e6c5bc73 JM |
2170 | /* |
2171 | * Merge in the direction of the lesser amount of work. | |
2172 | */ | |
e6c5bc73 JM |
2173 | if (new_process_refs >= process_refs) { |
2174 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2175 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2176 | } else { |
2177 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2178 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2179 | } |
df5fe3e8 JM |
2180 | } |
2181 | ||
cdb16e8f | 2182 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
65b32a57 | 2183 | struct cfq_group *cfqg, enum wl_prio_t prio) |
718eee05 CZ |
2184 | { |
2185 | struct cfq_queue *queue; | |
2186 | int i; | |
2187 | bool key_valid = false; | |
2188 | unsigned long lowest_key = 0; | |
2189 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2190 | ||
65b32a57 VG |
2191 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2192 | /* select the one with lowest rb_key */ | |
2193 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); | |
718eee05 CZ |
2194 | if (queue && |
2195 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2196 | lowest_key = queue->rb_key; | |
2197 | cur_best = i; | |
2198 | key_valid = true; | |
2199 | } | |
2200 | } | |
2201 | ||
2202 | return cur_best; | |
2203 | } | |
2204 | ||
cdb16e8f | 2205 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
718eee05 | 2206 | { |
718eee05 CZ |
2207 | unsigned slice; |
2208 | unsigned count; | |
cdb16e8f | 2209 | struct cfq_rb_root *st; |
58ff82f3 | 2210 | unsigned group_slice; |
e4ea0c16 | 2211 | enum wl_prio_t original_prio = cfqd->serving_prio; |
1fa8f6d6 | 2212 | |
718eee05 | 2213 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2214 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
718eee05 | 2215 | cfqd->serving_prio = RT_WORKLOAD; |
58ff82f3 | 2216 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
718eee05 CZ |
2217 | cfqd->serving_prio = BE_WORKLOAD; |
2218 | else { | |
2219 | cfqd->serving_prio = IDLE_WORKLOAD; | |
2220 | cfqd->workload_expires = jiffies + 1; | |
2221 | return; | |
2222 | } | |
2223 | ||
e4ea0c16 SL |
2224 | if (original_prio != cfqd->serving_prio) |
2225 | goto new_workload; | |
2226 | ||
718eee05 CZ |
2227 | /* |
2228 | * For RT and BE, we have to choose also the type | |
2229 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2230 | * expiration time | |
2231 | */ | |
65b32a57 | 2232 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
cdb16e8f | 2233 | count = st->count; |
718eee05 CZ |
2234 | |
2235 | /* | |
65b32a57 | 2236 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2237 | */ |
65b32a57 | 2238 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2239 | return; |
2240 | ||
e4ea0c16 | 2241 | new_workload: |
718eee05 CZ |
2242 | /* otherwise select new workload type */ |
2243 | cfqd->serving_type = | |
65b32a57 VG |
2244 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); |
2245 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); | |
cdb16e8f | 2246 | count = st->count; |
718eee05 CZ |
2247 | |
2248 | /* | |
2249 | * the workload slice is computed as a fraction of target latency | |
2250 | * proportional to the number of queues in that workload, over | |
2251 | * all the queues in the same priority class | |
2252 | */ | |
58ff82f3 VG |
2253 | group_slice = cfq_group_slice(cfqd, cfqg); |
2254 | ||
2255 | slice = group_slice * count / | |
2256 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], | |
2257 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); | |
718eee05 | 2258 | |
f26bd1f0 VG |
2259 | if (cfqd->serving_type == ASYNC_WORKLOAD) { |
2260 | unsigned int tmp; | |
2261 | ||
2262 | /* | |
2263 | * Async queues are currently system wide. Just taking | |
2264 | * proportion of queues with-in same group will lead to higher | |
2265 | * async ratio system wide as generally root group is going | |
2266 | * to have higher weight. A more accurate thing would be to | |
2267 | * calculate system wide asnc/sync ratio. | |
2268 | */ | |
2269 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); | |
2270 | tmp = tmp/cfqd->busy_queues; | |
2271 | slice = min_t(unsigned, slice, tmp); | |
2272 | ||
718eee05 CZ |
2273 | /* async workload slice is scaled down according to |
2274 | * the sync/async slice ratio. */ | |
2275 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2276 | } else |
718eee05 CZ |
2277 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2278 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2279 | ||
2280 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2281 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
2282 | cfqd->workload_expires = jiffies + slice; |
2283 | } | |
2284 | ||
1fa8f6d6 VG |
2285 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2286 | { | |
2287 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2288 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2289 | |
2290 | if (RB_EMPTY_ROOT(&st->rb)) | |
2291 | return NULL; | |
25bc6b07 | 2292 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
2293 | update_min_vdisktime(st); |
2294 | return cfqg; | |
1fa8f6d6 VG |
2295 | } |
2296 | ||
cdb16e8f VG |
2297 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2298 | { | |
1fa8f6d6 VG |
2299 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2300 | ||
2301 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2302 | |
2303 | /* Restore the workload type data */ | |
2304 | if (cfqg->saved_workload_slice) { | |
2305 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; | |
2306 | cfqd->serving_type = cfqg->saved_workload; | |
2307 | cfqd->serving_prio = cfqg->saved_serving_prio; | |
66ae2919 GJ |
2308 | } else |
2309 | cfqd->workload_expires = jiffies - 1; | |
2310 | ||
1fa8f6d6 | 2311 | choose_service_tree(cfqd, cfqg); |
cdb16e8f VG |
2312 | } |
2313 | ||
22e2c507 | 2314 | /* |
498d3aa2 JA |
2315 | * Select a queue for service. If we have a current active queue, |
2316 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2317 | */ |
1b5ed5e1 | 2318 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2319 | { |
a36e71f9 | 2320 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2321 | |
22e2c507 JA |
2322 | cfqq = cfqd->active_queue; |
2323 | if (!cfqq) | |
2324 | goto new_queue; | |
1da177e4 | 2325 | |
f04a6424 VG |
2326 | if (!cfqd->rq_queued) |
2327 | return NULL; | |
c244bb50 VG |
2328 | |
2329 | /* | |
2330 | * We were waiting for group to get backlogged. Expire the queue | |
2331 | */ | |
2332 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2333 | goto expire; | |
2334 | ||
22e2c507 | 2335 | /* |
6d048f53 | 2336 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2337 | */ |
7667aa06 VG |
2338 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2339 | /* | |
2340 | * If slice had not expired at the completion of last request | |
2341 | * we might not have turned on wait_busy flag. Don't expire | |
2342 | * the queue yet. Allow the group to get backlogged. | |
2343 | * | |
2344 | * The very fact that we have used the slice, that means we | |
2345 | * have been idling all along on this queue and it should be | |
2346 | * ok to wait for this request to complete. | |
2347 | */ | |
82bbbf28 VG |
2348 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2349 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2350 | cfqq = NULL; | |
7667aa06 | 2351 | goto keep_queue; |
82bbbf28 | 2352 | } else |
80bdf0c7 | 2353 | goto check_group_idle; |
7667aa06 | 2354 | } |
1da177e4 | 2355 | |
22e2c507 | 2356 | /* |
6d048f53 JA |
2357 | * The active queue has requests and isn't expired, allow it to |
2358 | * dispatch. | |
22e2c507 | 2359 | */ |
dd67d051 | 2360 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2361 | goto keep_queue; |
6d048f53 | 2362 | |
a36e71f9 JA |
2363 | /* |
2364 | * If another queue has a request waiting within our mean seek | |
2365 | * distance, let it run. The expire code will check for close | |
2366 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2367 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2368 | */ |
b3b6d040 | 2369 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2370 | if (new_cfqq) { |
2371 | if (!cfqq->new_cfqq) | |
2372 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2373 | goto expire; |
df5fe3e8 | 2374 | } |
a36e71f9 | 2375 | |
6d048f53 JA |
2376 | /* |
2377 | * No requests pending. If the active queue still has requests in | |
2378 | * flight or is idling for a new request, allow either of these | |
2379 | * conditions to happen (or time out) before selecting a new queue. | |
2380 | */ | |
80bdf0c7 VG |
2381 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2382 | cfqq = NULL; | |
2383 | goto keep_queue; | |
2384 | } | |
2385 | ||
8e1ac665 SL |
2386 | /* |
2387 | * This is a deep seek queue, but the device is much faster than | |
2388 | * the queue can deliver, don't idle | |
2389 | **/ | |
2390 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
2391 | (cfq_cfqq_slice_new(cfqq) || | |
2392 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
2393 | cfq_clear_cfqq_deep(cfqq); | |
2394 | cfq_clear_cfqq_idle_window(cfqq); | |
2395 | } | |
2396 | ||
80bdf0c7 VG |
2397 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2398 | cfqq = NULL; | |
2399 | goto keep_queue; | |
2400 | } | |
2401 | ||
2402 | /* | |
2403 | * If group idle is enabled and there are requests dispatched from | |
2404 | * this group, wait for requests to complete. | |
2405 | */ | |
2406 | check_group_idle: | |
7700fc4f SL |
2407 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
2408 | cfqq->cfqg->dispatched && | |
2409 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
2410 | cfqq = NULL; |
2411 | goto keep_queue; | |
22e2c507 JA |
2412 | } |
2413 | ||
3b18152c | 2414 | expire: |
e5ff082e | 2415 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2416 | new_queue: |
718eee05 CZ |
2417 | /* |
2418 | * Current queue expired. Check if we have to switch to a new | |
2419 | * service tree | |
2420 | */ | |
2421 | if (!new_cfqq) | |
cdb16e8f | 2422 | cfq_choose_cfqg(cfqd); |
718eee05 | 2423 | |
a36e71f9 | 2424 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2425 | keep_queue: |
3b18152c | 2426 | return cfqq; |
22e2c507 JA |
2427 | } |
2428 | ||
febffd61 | 2429 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2430 | { |
2431 | int dispatched = 0; | |
2432 | ||
2433 | while (cfqq->next_rq) { | |
2434 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2435 | dispatched++; | |
2436 | } | |
2437 | ||
2438 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2439 | |
2440 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 2441 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
2442 | return dispatched; |
2443 | } | |
2444 | ||
498d3aa2 JA |
2445 | /* |
2446 | * Drain our current requests. Used for barriers and when switching | |
2447 | * io schedulers on-the-fly. | |
2448 | */ | |
d9e7620e | 2449 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2450 | { |
0871714e | 2451 | struct cfq_queue *cfqq; |
d9e7620e | 2452 | int dispatched = 0; |
cdb16e8f | 2453 | |
3440c49f | 2454 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 2455 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
2456 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2457 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 2458 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 2459 | } |
1b5ed5e1 | 2460 | |
1b5ed5e1 TH |
2461 | BUG_ON(cfqd->busy_queues); |
2462 | ||
6923715a | 2463 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2464 | return dispatched; |
2465 | } | |
2466 | ||
abc3c744 SL |
2467 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2468 | struct cfq_queue *cfqq) | |
2469 | { | |
2470 | /* the queue hasn't finished any request, can't estimate */ | |
2471 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 2472 | return true; |
abc3c744 SL |
2473 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
2474 | cfqq->slice_end)) | |
c1e44756 | 2475 | return true; |
abc3c744 | 2476 | |
c1e44756 | 2477 | return false; |
abc3c744 SL |
2478 | } |
2479 | ||
0b182d61 | 2480 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2481 | { |
2f5cb738 | 2482 | unsigned int max_dispatch; |
22e2c507 | 2483 | |
5ad531db JA |
2484 | /* |
2485 | * Drain async requests before we start sync IO | |
2486 | */ | |
53c583d2 | 2487 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 2488 | return false; |
5ad531db | 2489 | |
2f5cb738 JA |
2490 | /* |
2491 | * If this is an async queue and we have sync IO in flight, let it wait | |
2492 | */ | |
53c583d2 | 2493 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 2494 | return false; |
2f5cb738 | 2495 | |
abc3c744 | 2496 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
2497 | if (cfq_class_idle(cfqq)) |
2498 | max_dispatch = 1; | |
b4878f24 | 2499 | |
2f5cb738 JA |
2500 | /* |
2501 | * Does this cfqq already have too much IO in flight? | |
2502 | */ | |
2503 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 2504 | bool promote_sync = false; |
2f5cb738 JA |
2505 | /* |
2506 | * idle queue must always only have a single IO in flight | |
2507 | */ | |
3ed9a296 | 2508 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2509 | return false; |
3ed9a296 | 2510 | |
ef8a41df | 2511 | /* |
c4ade94f LS |
2512 | * If there is only one sync queue |
2513 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
2514 | * queue no dispatch limit. The reason is a sync queue can |
2515 | * preempt async queue, limiting the sync queue doesn't make | |
2516 | * sense. This is useful for aiostress test. | |
2517 | */ | |
c4ade94f LS |
2518 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
2519 | promote_sync = true; | |
ef8a41df | 2520 | |
2f5cb738 JA |
2521 | /* |
2522 | * We have other queues, don't allow more IO from this one | |
2523 | */ | |
ef8a41df SL |
2524 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
2525 | !promote_sync) | |
0b182d61 | 2526 | return false; |
9ede209e | 2527 | |
365722bb | 2528 | /* |
474b18cc | 2529 | * Sole queue user, no limit |
365722bb | 2530 | */ |
ef8a41df | 2531 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
2532 | max_dispatch = -1; |
2533 | else | |
2534 | /* | |
2535 | * Normally we start throttling cfqq when cfq_quantum/2 | |
2536 | * requests have been dispatched. But we can drive | |
2537 | * deeper queue depths at the beginning of slice | |
2538 | * subjected to upper limit of cfq_quantum. | |
2539 | * */ | |
2540 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
2541 | } |
2542 | ||
2543 | /* | |
2544 | * Async queues must wait a bit before being allowed dispatch. | |
2545 | * We also ramp up the dispatch depth gradually for async IO, | |
2546 | * based on the last sync IO we serviced | |
2547 | */ | |
963b72fc | 2548 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 2549 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 2550 | unsigned int depth; |
365722bb | 2551 | |
61f0c1dc | 2552 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2553 | if (!depth && !cfqq->dispatched) |
2554 | depth = 1; | |
8e296755 JA |
2555 | if (depth < max_dispatch) |
2556 | max_dispatch = depth; | |
2f5cb738 | 2557 | } |
3ed9a296 | 2558 | |
0b182d61 JA |
2559 | /* |
2560 | * If we're below the current max, allow a dispatch | |
2561 | */ | |
2562 | return cfqq->dispatched < max_dispatch; | |
2563 | } | |
2564 | ||
2565 | /* | |
2566 | * Dispatch a request from cfqq, moving them to the request queue | |
2567 | * dispatch list. | |
2568 | */ | |
2569 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2570 | { | |
2571 | struct request *rq; | |
2572 | ||
2573 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
2574 | ||
2575 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
2576 | return false; | |
2577 | ||
2578 | /* | |
2579 | * follow expired path, else get first next available | |
2580 | */ | |
2581 | rq = cfq_check_fifo(cfqq); | |
2582 | if (!rq) | |
2583 | rq = cfqq->next_rq; | |
2584 | ||
2585 | /* | |
2586 | * insert request into driver dispatch list | |
2587 | */ | |
2588 | cfq_dispatch_insert(cfqd->queue, rq); | |
2589 | ||
2590 | if (!cfqd->active_cic) { | |
c5869807 | 2591 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 2592 | |
c5869807 | 2593 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
2594 | cfqd->active_cic = cic; |
2595 | } | |
2596 | ||
2597 | return true; | |
2598 | } | |
2599 | ||
2600 | /* | |
2601 | * Find the cfqq that we need to service and move a request from that to the | |
2602 | * dispatch list | |
2603 | */ | |
2604 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
2605 | { | |
2606 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2607 | struct cfq_queue *cfqq; | |
2608 | ||
2609 | if (!cfqd->busy_queues) | |
2610 | return 0; | |
2611 | ||
2612 | if (unlikely(force)) | |
2613 | return cfq_forced_dispatch(cfqd); | |
2614 | ||
2615 | cfqq = cfq_select_queue(cfqd); | |
2616 | if (!cfqq) | |
8e296755 JA |
2617 | return 0; |
2618 | ||
2f5cb738 | 2619 | /* |
0b182d61 | 2620 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 2621 | */ |
0b182d61 JA |
2622 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2623 | return 0; | |
2624 | ||
2f5cb738 | 2625 | cfqq->slice_dispatch++; |
b029195d | 2626 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 2627 | |
2f5cb738 JA |
2628 | /* |
2629 | * expire an async queue immediately if it has used up its slice. idle | |
2630 | * queue always expire after 1 dispatch round. | |
2631 | */ | |
2632 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
2633 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
2634 | cfq_class_idle(cfqq))) { | |
2635 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 2636 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
2637 | } |
2638 | ||
b217a903 | 2639 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 2640 | return 1; |
1da177e4 LT |
2641 | } |
2642 | ||
1da177e4 | 2643 | /* |
5e705374 JA |
2644 | * task holds one reference to the queue, dropped when task exits. each rq |
2645 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 2646 | * |
b1c35769 | 2647 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
2648 | * queue lock must be held here. |
2649 | */ | |
2650 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
2651 | { | |
22e2c507 | 2652 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 2653 | struct cfq_group *cfqg; |
22e2c507 | 2654 | |
30d7b944 | 2655 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 2656 | |
30d7b944 SL |
2657 | cfqq->ref--; |
2658 | if (cfqq->ref) | |
1da177e4 LT |
2659 | return; |
2660 | ||
7b679138 | 2661 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 2662 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 2663 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 2664 | cfqg = cfqq->cfqg; |
1da177e4 | 2665 | |
28f95cbc | 2666 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 2667 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2668 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2669 | } |
22e2c507 | 2670 | |
f04a6424 | 2671 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 2672 | kmem_cache_free(cfq_pool, cfqq); |
b1c35769 | 2673 | cfq_put_cfqg(cfqg); |
1da177e4 LT |
2674 | } |
2675 | ||
d02a2c07 | 2676 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 2677 | { |
df5fe3e8 JM |
2678 | struct cfq_queue *__cfqq, *next; |
2679 | ||
df5fe3e8 JM |
2680 | /* |
2681 | * If this queue was scheduled to merge with another queue, be | |
2682 | * sure to drop the reference taken on that queue (and others in | |
2683 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
2684 | */ | |
2685 | __cfqq = cfqq->new_cfqq; | |
2686 | while (__cfqq) { | |
2687 | if (__cfqq == cfqq) { | |
2688 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
2689 | break; | |
2690 | } | |
2691 | next = __cfqq->new_cfqq; | |
2692 | cfq_put_queue(__cfqq); | |
2693 | __cfqq = next; | |
2694 | } | |
d02a2c07 SL |
2695 | } |
2696 | ||
2697 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2698 | { | |
2699 | if (unlikely(cfqq == cfqd->active_queue)) { | |
2700 | __cfq_slice_expired(cfqd, cfqq, 0); | |
2701 | cfq_schedule_dispatch(cfqd); | |
2702 | } | |
2703 | ||
2704 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 2705 | |
89850f7e JA |
2706 | cfq_put_queue(cfqq); |
2707 | } | |
22e2c507 | 2708 | |
9b84cacd TH |
2709 | static void cfq_init_icq(struct io_cq *icq) |
2710 | { | |
2711 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
2712 | ||
2713 | cic->ttime.last_end_request = jiffies; | |
2714 | } | |
2715 | ||
c5869807 | 2716 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 2717 | { |
c5869807 | 2718 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 2719 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 2720 | |
ff6657c6 JA |
2721 | if (cic->cfqq[BLK_RW_ASYNC]) { |
2722 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
2723 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
2724 | } |
2725 | ||
ff6657c6 JA |
2726 | if (cic->cfqq[BLK_RW_SYNC]) { |
2727 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
2728 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 2729 | } |
89850f7e JA |
2730 | } |
2731 | ||
fd0928df | 2732 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
2733 | { |
2734 | struct task_struct *tsk = current; | |
2735 | int ioprio_class; | |
2736 | ||
3b18152c | 2737 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
2738 | return; |
2739 | ||
fd0928df | 2740 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 2741 | switch (ioprio_class) { |
fe094d98 JA |
2742 | default: |
2743 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
2744 | case IOPRIO_CLASS_NONE: | |
2745 | /* | |
6d63c275 | 2746 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
2747 | */ |
2748 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 2749 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
2750 | break; |
2751 | case IOPRIO_CLASS_RT: | |
2752 | cfqq->ioprio = task_ioprio(ioc); | |
2753 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
2754 | break; | |
2755 | case IOPRIO_CLASS_BE: | |
2756 | cfqq->ioprio = task_ioprio(ioc); | |
2757 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
2758 | break; | |
2759 | case IOPRIO_CLASS_IDLE: | |
2760 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
2761 | cfqq->ioprio = 7; | |
2762 | cfq_clear_cfqq_idle_window(cfqq); | |
2763 | break; | |
22e2c507 JA |
2764 | } |
2765 | ||
2766 | /* | |
2767 | * keep track of original prio settings in case we have to temporarily | |
2768 | * elevate the priority of this queue | |
2769 | */ | |
2770 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 2771 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
2772 | } |
2773 | ||
c5869807 | 2774 | static void changed_ioprio(struct cfq_io_cq *cic) |
22e2c507 | 2775 | { |
bca4b914 | 2776 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 2777 | struct cfq_queue *cfqq; |
35e6077c | 2778 | |
caaa5f9f JA |
2779 | if (unlikely(!cfqd)) |
2780 | return; | |
2781 | ||
ff6657c6 | 2782 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
2783 | if (cfqq) { |
2784 | struct cfq_queue *new_cfqq; | |
c5869807 | 2785 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->icq.ioc, |
ff6657c6 | 2786 | GFP_ATOMIC); |
caaa5f9f | 2787 | if (new_cfqq) { |
ff6657c6 | 2788 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
2789 | cfq_put_queue(cfqq); |
2790 | } | |
22e2c507 | 2791 | } |
caaa5f9f | 2792 | |
ff6657c6 | 2793 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
2794 | if (cfqq) |
2795 | cfq_mark_cfqq_prio_changed(cfqq); | |
22e2c507 JA |
2796 | } |
2797 | ||
d5036d77 | 2798 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2799 | pid_t pid, bool is_sync) |
d5036d77 JA |
2800 | { |
2801 | RB_CLEAR_NODE(&cfqq->rb_node); | |
2802 | RB_CLEAR_NODE(&cfqq->p_node); | |
2803 | INIT_LIST_HEAD(&cfqq->fifo); | |
2804 | ||
30d7b944 | 2805 | cfqq->ref = 0; |
d5036d77 JA |
2806 | cfqq->cfqd = cfqd; |
2807 | ||
2808 | cfq_mark_cfqq_prio_changed(cfqq); | |
2809 | ||
2810 | if (is_sync) { | |
2811 | if (!cfq_class_idle(cfqq)) | |
2812 | cfq_mark_cfqq_idle_window(cfqq); | |
2813 | cfq_mark_cfqq_sync(cfqq); | |
2814 | } | |
2815 | cfqq->pid = pid; | |
2816 | } | |
2817 | ||
24610333 | 2818 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
c5869807 | 2819 | static void changed_cgroup(struct cfq_io_cq *cic) |
24610333 VG |
2820 | { |
2821 | struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1); | |
bca4b914 | 2822 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
24610333 VG |
2823 | struct request_queue *q; |
2824 | ||
2825 | if (unlikely(!cfqd)) | |
2826 | return; | |
2827 | ||
2828 | q = cfqd->queue; | |
2829 | ||
24610333 VG |
2830 | if (sync_cfqq) { |
2831 | /* | |
2832 | * Drop reference to sync queue. A new sync queue will be | |
2833 | * assigned in new group upon arrival of a fresh request. | |
2834 | */ | |
2835 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
2836 | cic_set_cfqq(cic, NULL, 1); | |
2837 | cfq_put_queue(sync_cfqq); | |
2838 | } | |
24610333 | 2839 | } |
24610333 VG |
2840 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
2841 | ||
22e2c507 | 2842 | static struct cfq_queue * |
a6151c3a | 2843 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, |
fd0928df | 2844 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 2845 | { |
22e2c507 | 2846 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
c5869807 | 2847 | struct cfq_io_cq *cic; |
cdb16e8f | 2848 | struct cfq_group *cfqg; |
22e2c507 JA |
2849 | |
2850 | retry: | |
3e59cf9d | 2851 | cfqg = cfq_get_cfqg(cfqd); |
4ac845a2 | 2852 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
2853 | /* cic always exists here */ |
2854 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 | 2855 | |
6118b70b JA |
2856 | /* |
2857 | * Always try a new alloc if we fell back to the OOM cfqq | |
2858 | * originally, since it should just be a temporary situation. | |
2859 | */ | |
2860 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
2861 | cfqq = NULL; | |
22e2c507 JA |
2862 | if (new_cfqq) { |
2863 | cfqq = new_cfqq; | |
2864 | new_cfqq = NULL; | |
2865 | } else if (gfp_mask & __GFP_WAIT) { | |
2866 | spin_unlock_irq(cfqd->queue->queue_lock); | |
94f6030c | 2867 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 2868 | gfp_mask | __GFP_ZERO, |
94f6030c | 2869 | cfqd->queue->node); |
22e2c507 | 2870 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
2871 | if (new_cfqq) |
2872 | goto retry; | |
22e2c507 | 2873 | } else { |
94f6030c CL |
2874 | cfqq = kmem_cache_alloc_node(cfq_pool, |
2875 | gfp_mask | __GFP_ZERO, | |
2876 | cfqd->queue->node); | |
22e2c507 JA |
2877 | } |
2878 | ||
6118b70b JA |
2879 | if (cfqq) { |
2880 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
2881 | cfq_init_prio_data(cfqq, ioc); | |
cdb16e8f | 2882 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
2883 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
2884 | } else | |
2885 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
2886 | } |
2887 | ||
2888 | if (new_cfqq) | |
2889 | kmem_cache_free(cfq_pool, new_cfqq); | |
2890 | ||
22e2c507 JA |
2891 | return cfqq; |
2892 | } | |
2893 | ||
c2dea2d1 VT |
2894 | static struct cfq_queue ** |
2895 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
2896 | { | |
fe094d98 | 2897 | switch (ioprio_class) { |
c2dea2d1 VT |
2898 | case IOPRIO_CLASS_RT: |
2899 | return &cfqd->async_cfqq[0][ioprio]; | |
2900 | case IOPRIO_CLASS_BE: | |
2901 | return &cfqd->async_cfqq[1][ioprio]; | |
2902 | case IOPRIO_CLASS_IDLE: | |
2903 | return &cfqd->async_idle_cfqq; | |
2904 | default: | |
2905 | BUG(); | |
2906 | } | |
2907 | } | |
2908 | ||
15c31be4 | 2909 | static struct cfq_queue * |
a6151c3a | 2910 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc, |
15c31be4 JA |
2911 | gfp_t gfp_mask) |
2912 | { | |
fd0928df JA |
2913 | const int ioprio = task_ioprio(ioc); |
2914 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 2915 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
2916 | struct cfq_queue *cfqq = NULL; |
2917 | ||
c2dea2d1 VT |
2918 | if (!is_sync) { |
2919 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
2920 | cfqq = *async_cfqq; | |
2921 | } | |
2922 | ||
6118b70b | 2923 | if (!cfqq) |
fd0928df | 2924 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
15c31be4 JA |
2925 | |
2926 | /* | |
2927 | * pin the queue now that it's allocated, scheduler exit will prune it | |
2928 | */ | |
c2dea2d1 | 2929 | if (!is_sync && !(*async_cfqq)) { |
30d7b944 | 2930 | cfqq->ref++; |
c2dea2d1 | 2931 | *async_cfqq = cfqq; |
15c31be4 JA |
2932 | } |
2933 | ||
30d7b944 | 2934 | cfqq->ref++; |
15c31be4 JA |
2935 | return cfqq; |
2936 | } | |
2937 | ||
216284c3 | 2938 | /** |
9b84cacd TH |
2939 | * ioc_create_icq - create and link io_cq |
2940 | * @q: request_queue of interest | |
216284c3 TH |
2941 | * @gfp_mask: allocation mask |
2942 | * | |
9b84cacd TH |
2943 | * Make sure io_cq linking %current->io_context and @q exists. If either |
2944 | * io_context and/or icq don't exist, they will be created using @gfp_mask. | |
2945 | * | |
2946 | * The caller is responsible for ensuring @ioc won't go away and @q is | |
2947 | * alive and will stay alive until this function returns. | |
4ac845a2 | 2948 | */ |
9b84cacd | 2949 | static struct io_cq *ioc_create_icq(struct request_queue *q, gfp_t gfp_mask) |
e2d74ac0 | 2950 | { |
9b84cacd | 2951 | struct elevator_type *et = q->elevator->type; |
216284c3 | 2952 | struct io_context *ioc; |
9b84cacd | 2953 | struct io_cq *icq; |
216284c3 TH |
2954 | |
2955 | /* allocate stuff */ | |
f2dbd76a | 2956 | ioc = create_io_context(current, gfp_mask, q->node); |
216284c3 | 2957 | if (!ioc) |
9b84cacd | 2958 | return NULL; |
216284c3 | 2959 | |
9b84cacd TH |
2960 | icq = kmem_cache_alloc_node(et->icq_cache, gfp_mask | __GFP_ZERO, |
2961 | q->node); | |
2962 | if (!icq) | |
2963 | return NULL; | |
e2d74ac0 | 2964 | |
9b84cacd TH |
2965 | if (radix_tree_preload(gfp_mask) < 0) { |
2966 | kmem_cache_free(et->icq_cache, icq); | |
2967 | return NULL; | |
2968 | } | |
e2d74ac0 | 2969 | |
c5869807 | 2970 | icq->ioc = ioc; |
9b84cacd TH |
2971 | icq->q = q; |
2972 | INIT_LIST_HEAD(&icq->q_node); | |
2973 | INIT_HLIST_NODE(&icq->ioc_node); | |
283287a5 | 2974 | |
c5869807 | 2975 | /* lock both q and ioc and try to link @icq */ |
216284c3 TH |
2976 | spin_lock_irq(q->queue_lock); |
2977 | spin_lock(&ioc->lock); | |
4ac845a2 | 2978 | |
9b84cacd | 2979 | if (likely(!radix_tree_insert(&ioc->icq_tree, q->id, icq))) { |
c5869807 | 2980 | hlist_add_head(&icq->ioc_node, &ioc->icq_list); |
a612fddf | 2981 | list_add(&icq->q_node, &q->icq_list); |
9b84cacd TH |
2982 | if (et->ops.elevator_init_icq_fn) |
2983 | et->ops.elevator_init_icq_fn(icq); | |
2984 | } else { | |
2985 | kmem_cache_free(et->icq_cache, icq); | |
2986 | icq = ioc_lookup_icq(ioc, q); | |
2987 | if (!icq) | |
2988 | printk(KERN_ERR "cfq: icq link failed!\n"); | |
e2d74ac0 | 2989 | } |
216284c3 TH |
2990 | |
2991 | spin_unlock(&ioc->lock); | |
2992 | spin_unlock_irq(q->queue_lock); | |
216284c3 | 2993 | radix_tree_preload_end(); |
9b84cacd | 2994 | return icq; |
e2d74ac0 JA |
2995 | } |
2996 | ||
216284c3 | 2997 | /** |
c5869807 | 2998 | * cfq_get_cic - acquire cfq_io_cq and bump refcnt on io_context |
216284c3 TH |
2999 | * @cfqd: cfqd to setup cic for |
3000 | * @gfp_mask: allocation mask | |
3001 | * | |
c5869807 | 3002 | * Return cfq_io_cq associating @cfqd and %current->io_context and |
216284c3 TH |
3003 | * bump refcnt on io_context. If ioc or cic doesn't exist, they're created |
3004 | * using @gfp_mask. | |
3005 | * | |
3006 | * Must be called under queue_lock which may be released and re-acquired. | |
3007 | * This function also may sleep depending on @gfp_mask. | |
1da177e4 | 3008 | */ |
c5869807 | 3009 | static struct cfq_io_cq *cfq_get_cic(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 3010 | { |
216284c3 | 3011 | struct request_queue *q = cfqd->queue; |
c5869807 | 3012 | struct cfq_io_cq *cic = NULL; |
216284c3 | 3013 | struct io_context *ioc; |
216284c3 TH |
3014 | |
3015 | lockdep_assert_held(q->queue_lock); | |
3016 | ||
3017 | while (true) { | |
3018 | /* fast path */ | |
3019 | ioc = current->io_context; | |
3020 | if (likely(ioc)) { | |
3021 | cic = cfq_cic_lookup(cfqd, ioc); | |
3022 | if (likely(cic)) | |
3023 | break; | |
3024 | } | |
1da177e4 | 3025 | |
216284c3 TH |
3026 | /* slow path - unlock, create missing ones and retry */ |
3027 | spin_unlock_irq(q->queue_lock); | |
9b84cacd | 3028 | cic = icq_to_cic(ioc_create_icq(q, gfp_mask)); |
216284c3 | 3029 | spin_lock_irq(q->queue_lock); |
9b84cacd | 3030 | if (!cic) |
216284c3 TH |
3031 | return NULL; |
3032 | } | |
1da177e4 | 3033 | |
216284c3 | 3034 | /* bump @ioc's refcnt and handle changed notifications */ |
6e736be7 TH |
3035 | get_io_context(ioc); |
3036 | ||
c5869807 TH |
3037 | if (unlikely(cic->icq.changed)) { |
3038 | if (test_and_clear_bit(ICQ_IOPRIO_CHANGED, &cic->icq.changed)) | |
dc86900e | 3039 | changed_ioprio(cic); |
24610333 | 3040 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
c5869807 | 3041 | if (test_and_clear_bit(ICQ_CGROUP_CHANGED, &cic->icq.changed)) |
dc86900e | 3042 | changed_cgroup(cic); |
24610333 | 3043 | #endif |
dc86900e TH |
3044 | } |
3045 | ||
1da177e4 | 3046 | return cic; |
1da177e4 LT |
3047 | } |
3048 | ||
22e2c507 | 3049 | static void |
383cd721 | 3050 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 3051 | { |
383cd721 SL |
3052 | unsigned long elapsed = jiffies - ttime->last_end_request; |
3053 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 3054 | |
383cd721 SL |
3055 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
3056 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
3057 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
3058 | } | |
3059 | ||
3060 | static void | |
3061 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3062 | struct cfq_io_cq *cic) |
383cd721 | 3063 | { |
f5f2b6ce | 3064 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3065 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3066 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3067 | cfqd->cfq_slice_idle); | |
3068 | } | |
7700fc4f SL |
3069 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3070 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3071 | #endif | |
22e2c507 | 3072 | } |
1da177e4 | 3073 | |
206dc69b | 3074 | static void |
b2c18e1e | 3075 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3076 | struct request *rq) |
206dc69b | 3077 | { |
3dde36dd | 3078 | sector_t sdist = 0; |
41647e7a | 3079 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3080 | if (cfqq->last_request_pos) { |
3081 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3082 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3083 | else | |
3084 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3085 | } | |
206dc69b | 3086 | |
3dde36dd | 3087 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3088 | if (blk_queue_nonrot(cfqd->queue)) |
3089 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3090 | else | |
3091 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3092 | } |
1da177e4 | 3093 | |
22e2c507 JA |
3094 | /* |
3095 | * Disable idle window if the process thinks too long or seeks so much that | |
3096 | * it doesn't matter | |
3097 | */ | |
3098 | static void | |
3099 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 3100 | struct cfq_io_cq *cic) |
22e2c507 | 3101 | { |
7b679138 | 3102 | int old_idle, enable_idle; |
1be92f2f | 3103 | |
0871714e JA |
3104 | /* |
3105 | * Don't idle for async or idle io prio class | |
3106 | */ | |
3107 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3108 | return; |
3109 | ||
c265a7f4 | 3110 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3111 | |
76280aff CZ |
3112 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3113 | cfq_mark_cfqq_deep(cfqq); | |
3114 | ||
749ef9f8 CZ |
3115 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
3116 | enable_idle = 0; | |
c5869807 TH |
3117 | else if (!atomic_read(&cic->icq.ioc->nr_tasks) || |
3118 | !cfqd->cfq_slice_idle || | |
3119 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 3120 | enable_idle = 0; |
383cd721 SL |
3121 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3122 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3123 | enable_idle = 0; |
3124 | else | |
3125 | enable_idle = 1; | |
1da177e4 LT |
3126 | } |
3127 | ||
7b679138 JA |
3128 | if (old_idle != enable_idle) { |
3129 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3130 | if (enable_idle) | |
3131 | cfq_mark_cfqq_idle_window(cfqq); | |
3132 | else | |
3133 | cfq_clear_cfqq_idle_window(cfqq); | |
3134 | } | |
22e2c507 | 3135 | } |
1da177e4 | 3136 | |
22e2c507 JA |
3137 | /* |
3138 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3139 | * no or if we aren't sure, a 1 will cause a preempt. | |
3140 | */ | |
a6151c3a | 3141 | static bool |
22e2c507 | 3142 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3143 | struct request *rq) |
22e2c507 | 3144 | { |
6d048f53 | 3145 | struct cfq_queue *cfqq; |
22e2c507 | 3146 | |
6d048f53 JA |
3147 | cfqq = cfqd->active_queue; |
3148 | if (!cfqq) | |
a6151c3a | 3149 | return false; |
22e2c507 | 3150 | |
6d048f53 | 3151 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3152 | return false; |
22e2c507 JA |
3153 | |
3154 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3155 | return true; |
1e3335de | 3156 | |
875feb63 DS |
3157 | /* |
3158 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3159 | */ | |
3160 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3161 | return false; | |
3162 | ||
374f84ac JA |
3163 | /* |
3164 | * if the new request is sync, but the currently running queue is | |
3165 | * not, let the sync request have priority. | |
3166 | */ | |
5e705374 | 3167 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3168 | return true; |
1e3335de | 3169 | |
8682e1f1 VG |
3170 | if (new_cfqq->cfqg != cfqq->cfqg) |
3171 | return false; | |
3172 | ||
3173 | if (cfq_slice_used(cfqq)) | |
3174 | return true; | |
3175 | ||
3176 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
3177 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && | |
3178 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && | |
3179 | new_cfqq->service_tree->count == 2 && | |
3180 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3181 | return true; | |
3182 | ||
b53d1ed7 JA |
3183 | /* |
3184 | * So both queues are sync. Let the new request get disk time if | |
3185 | * it's a metadata request and the current queue is doing regular IO. | |
3186 | */ | |
65299a3b | 3187 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
3188 | return true; |
3189 | ||
3a9a3f6c DS |
3190 | /* |
3191 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3192 | */ | |
3193 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3194 | return true; |
3a9a3f6c | 3195 | |
d2d59e18 SL |
3196 | /* An idle queue should not be idle now for some reason */ |
3197 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
3198 | return true; | |
3199 | ||
1e3335de | 3200 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3201 | return false; |
1e3335de JA |
3202 | |
3203 | /* | |
3204 | * if this request is as-good as one we would expect from the | |
3205 | * current cfqq, let it preempt | |
3206 | */ | |
e9ce335d | 3207 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3208 | return true; |
1e3335de | 3209 | |
a6151c3a | 3210 | return false; |
22e2c507 JA |
3211 | } |
3212 | ||
3213 | /* | |
3214 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3215 | * let it have half of its nominal slice. | |
3216 | */ | |
3217 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3218 | { | |
f8ae6e3e SL |
3219 | struct cfq_queue *old_cfqq = cfqd->active_queue; |
3220 | ||
7b679138 | 3221 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
e5ff082e | 3222 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3223 | |
f8ae6e3e SL |
3224 | /* |
3225 | * workload type is changed, don't save slice, otherwise preempt | |
3226 | * doesn't happen | |
3227 | */ | |
3228 | if (cfqq_type(old_cfqq) != cfqq_type(cfqq)) | |
3229 | cfqq->cfqg->saved_workload_slice = 0; | |
3230 | ||
bf572256 JA |
3231 | /* |
3232 | * Put the new queue at the front of the of the current list, | |
3233 | * so we know that it will be selected next. | |
3234 | */ | |
3235 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3236 | |
3237 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3238 | |
62a37f6b JT |
3239 | cfqq->slice_end = 0; |
3240 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3241 | } |
3242 | ||
22e2c507 | 3243 | /* |
5e705374 | 3244 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3245 | * something we should do about it |
3246 | */ | |
3247 | static void | |
5e705374 JA |
3248 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3249 | struct request *rq) | |
22e2c507 | 3250 | { |
c5869807 | 3251 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3252 | |
45333d5a | 3253 | cfqd->rq_queued++; |
65299a3b CH |
3254 | if (rq->cmd_flags & REQ_PRIO) |
3255 | cfqq->prio_pending++; | |
374f84ac | 3256 | |
383cd721 | 3257 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3258 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3259 | cfq_update_idle_window(cfqd, cfqq, cic); |
3260 | ||
b2c18e1e | 3261 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3262 | |
3263 | if (cfqq == cfqd->active_queue) { | |
3264 | /* | |
b029195d JA |
3265 | * Remember that we saw a request from this process, but |
3266 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3267 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3268 | * and merging. If the request is already larger than a single |
3269 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3270 | * merging is already done. Ditto for a busy system that |
3271 | * has other work pending, don't risk delaying until the | |
3272 | * idle timer unplug to continue working. | |
22e2c507 | 3273 | */ |
d6ceb25e | 3274 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3275 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3276 | cfqd->busy_queues > 1) { | |
812df48d | 3277 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3278 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3279 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3280 | } else { |
e98ef89b | 3281 | cfq_blkiocg_update_idle_time_stats( |
a11cdaa7 | 3282 | &cfqq->cfqg->blkg); |
bf791937 | 3283 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3284 | } |
d6ceb25e | 3285 | } |
5e705374 | 3286 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3287 | /* |
3288 | * not the active queue - expire current slice if it is | |
3289 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3290 | * has some old slice time left and is of higher priority or |
3291 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3292 | */ |
3293 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3294 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3295 | } |
1da177e4 LT |
3296 | } |
3297 | ||
165125e1 | 3298 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3299 | { |
b4878f24 | 3300 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3301 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3302 | |
7b679138 | 3303 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
c5869807 | 3304 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->icq.ioc); |
1da177e4 | 3305 | |
30996f40 | 3306 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3307 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3308 | cfq_add_rq_rb(rq); |
e98ef89b | 3309 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, |
cdc1184c DS |
3310 | &cfqd->serving_group->blkg, rq_data_dir(rq), |
3311 | rq_is_sync(rq)); | |
5e705374 | 3312 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3313 | } |
3314 | ||
45333d5a AC |
3315 | /* |
3316 | * Update hw_tag based on peak queue depth over 50 samples under | |
3317 | * sufficient load. | |
3318 | */ | |
3319 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3320 | { | |
1a1238a7 SL |
3321 | struct cfq_queue *cfqq = cfqd->active_queue; |
3322 | ||
53c583d2 CZ |
3323 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3324 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3325 | |
3326 | if (cfqd->hw_tag == 1) | |
3327 | return; | |
45333d5a AC |
3328 | |
3329 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3330 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3331 | return; |
3332 | ||
1a1238a7 SL |
3333 | /* |
3334 | * If active queue hasn't enough requests and can idle, cfq might not | |
3335 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3336 | * case | |
3337 | */ | |
3338 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3339 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3340 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3341 | return; |
3342 | ||
45333d5a AC |
3343 | if (cfqd->hw_tag_samples++ < 50) |
3344 | return; | |
3345 | ||
e459dd08 | 3346 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3347 | cfqd->hw_tag = 1; |
3348 | else | |
3349 | cfqd->hw_tag = 0; | |
45333d5a AC |
3350 | } |
3351 | ||
7667aa06 VG |
3352 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3353 | { | |
c5869807 | 3354 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 3355 | |
02a8f01b JT |
3356 | /* If the queue already has requests, don't wait */ |
3357 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3358 | return false; | |
3359 | ||
7667aa06 VG |
3360 | /* If there are other queues in the group, don't wait */ |
3361 | if (cfqq->cfqg->nr_cfqq > 1) | |
3362 | return false; | |
3363 | ||
7700fc4f SL |
3364 | /* the only queue in the group, but think time is big */ |
3365 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
3366 | return false; | |
3367 | ||
7667aa06 VG |
3368 | if (cfq_slice_used(cfqq)) |
3369 | return true; | |
3370 | ||
3371 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
3372 | if (cic && sample_valid(cic->ttime.ttime_samples) |
3373 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
3374 | return true; |
3375 | ||
3376 | /* | |
3377 | * If think times is less than a jiffy than ttime_mean=0 and above | |
3378 | * will not be true. It might happen that slice has not expired yet | |
3379 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
3380 | * case where think time is less than a jiffy, mark the queue wait | |
3381 | * busy if only 1 jiffy is left in the slice. | |
3382 | */ | |
3383 | if (cfqq->slice_end - jiffies == 1) | |
3384 | return true; | |
3385 | ||
3386 | return false; | |
3387 | } | |
3388 | ||
165125e1 | 3389 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3390 | { |
5e705374 | 3391 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3392 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3393 | const int sync = rq_is_sync(rq); |
b4878f24 | 3394 | unsigned long now; |
1da177e4 | 3395 | |
b4878f24 | 3396 | now = jiffies; |
33659ebb CH |
3397 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3398 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 3399 | |
45333d5a AC |
3400 | cfq_update_hw_tag(cfqd); |
3401 | ||
53c583d2 | 3402 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 3403 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 3404 | cfqd->rq_in_driver--; |
6d048f53 | 3405 | cfqq->dispatched--; |
80bdf0c7 | 3406 | (RQ_CFQG(rq))->dispatched--; |
e98ef89b VG |
3407 | cfq_blkiocg_update_completion_stats(&cfqq->cfqg->blkg, |
3408 | rq_start_time_ns(rq), rq_io_start_time_ns(rq), | |
3409 | rq_data_dir(rq), rq_is_sync(rq)); | |
1da177e4 | 3410 | |
53c583d2 | 3411 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 3412 | |
365722bb | 3413 | if (sync) { |
f5f2b6ce SL |
3414 | struct cfq_rb_root *service_tree; |
3415 | ||
383cd721 | 3416 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
3417 | |
3418 | if (cfq_cfqq_on_rr(cfqq)) | |
3419 | service_tree = cfqq->service_tree; | |
3420 | else | |
3421 | service_tree = service_tree_for(cfqq->cfqg, | |
3422 | cfqq_prio(cfqq), cfqq_type(cfqq)); | |
3423 | service_tree->ttime.last_end_request = now; | |
573412b2 CZ |
3424 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3425 | cfqd->last_delayed_sync = now; | |
365722bb | 3426 | } |
caaa5f9f | 3427 | |
7700fc4f SL |
3428 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3429 | cfqq->cfqg->ttime.last_end_request = now; | |
3430 | #endif | |
3431 | ||
caaa5f9f JA |
3432 | /* |
3433 | * If this is the active queue, check if it needs to be expired, | |
3434 | * or if we want to idle in case it has no pending requests. | |
3435 | */ | |
3436 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3437 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3438 | ||
44f7c160 JA |
3439 | if (cfq_cfqq_slice_new(cfqq)) { |
3440 | cfq_set_prio_slice(cfqd, cfqq); | |
3441 | cfq_clear_cfqq_slice_new(cfqq); | |
3442 | } | |
f75edf2d VG |
3443 | |
3444 | /* | |
7667aa06 VG |
3445 | * Should we wait for next request to come in before we expire |
3446 | * the queue. | |
f75edf2d | 3447 | */ |
7667aa06 | 3448 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
3449 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3450 | if (!cfqd->cfq_slice_idle) | |
3451 | extend_sl = cfqd->cfq_group_idle; | |
3452 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 3453 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 3454 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
3455 | } |
3456 | ||
a36e71f9 | 3457 | /* |
8e550632 CZ |
3458 | * Idling is not enabled on: |
3459 | * - expired queues | |
3460 | * - idle-priority queues | |
3461 | * - async queues | |
3462 | * - queues with still some requests queued | |
3463 | * - when there is a close cooperator | |
a36e71f9 | 3464 | */ |
0871714e | 3465 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 3466 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3467 | else if (sync && cfqq_empty && |
3468 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 3469 | cfq_arm_slice_timer(cfqd); |
8e550632 | 3470 | } |
caaa5f9f | 3471 | } |
6d048f53 | 3472 | |
53c583d2 | 3473 | if (!cfqd->rq_in_driver) |
23e018a1 | 3474 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3475 | } |
3476 | ||
89850f7e | 3477 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3478 | { |
1b379d8d | 3479 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3480 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3481 | return ELV_MQUEUE_MUST; |
3b18152c | 3482 | } |
1da177e4 | 3483 | |
22e2c507 | 3484 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3485 | } |
3486 | ||
165125e1 | 3487 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3488 | { |
3489 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3490 | struct task_struct *tsk = current; | |
c5869807 | 3491 | struct cfq_io_cq *cic; |
22e2c507 JA |
3492 | struct cfq_queue *cfqq; |
3493 | ||
3494 | /* | |
3495 | * don't force setup of a queue from here, as a call to may_queue | |
3496 | * does not necessarily imply that a request actually will be queued. | |
3497 | * so just lookup a possibly existing queue, or return 'may queue' | |
3498 | * if that fails | |
3499 | */ | |
4ac845a2 | 3500 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3501 | if (!cic) |
3502 | return ELV_MQUEUE_MAY; | |
3503 | ||
b0b78f81 | 3504 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3505 | if (cfqq) { |
c5869807 | 3506 | cfq_init_prio_data(cfqq, cic->icq.ioc); |
22e2c507 | 3507 | |
89850f7e | 3508 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3509 | } |
3510 | ||
3511 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3512 | } |
3513 | ||
1da177e4 LT |
3514 | /* |
3515 | * queue lock held here | |
3516 | */ | |
bb37b94c | 3517 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3518 | { |
5e705374 | 3519 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3520 | |
5e705374 | 3521 | if (cfqq) { |
22e2c507 | 3522 | const int rw = rq_data_dir(rq); |
1da177e4 | 3523 | |
22e2c507 JA |
3524 | BUG_ON(!cfqq->allocated[rw]); |
3525 | cfqq->allocated[rw]--; | |
1da177e4 | 3526 | |
c5869807 | 3527 | put_io_context(RQ_CIC(rq)->icq.ioc, cfqq->cfqd->queue); |
1da177e4 | 3528 | |
7f1dc8a2 VG |
3529 | /* Put down rq reference on cfqg */ |
3530 | cfq_put_cfqg(RQ_CFQG(rq)); | |
a612fddf TH |
3531 | rq->elv.priv[0] = NULL; |
3532 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 3533 | |
1da177e4 LT |
3534 | cfq_put_queue(cfqq); |
3535 | } | |
3536 | } | |
3537 | ||
df5fe3e8 | 3538 | static struct cfq_queue * |
c5869807 | 3539 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
3540 | struct cfq_queue *cfqq) |
3541 | { | |
3542 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3543 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3544 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3545 | cfq_put_queue(cfqq); |
3546 | return cic_to_cfqq(cic, 1); | |
3547 | } | |
3548 | ||
e6c5bc73 JM |
3549 | /* |
3550 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3551 | * was the last process referring to said cfqq. | |
3552 | */ | |
3553 | static struct cfq_queue * | |
c5869807 | 3554 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
3555 | { |
3556 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
3557 | cfqq->pid = current->pid; |
3558 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 3559 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
3560 | return cfqq; |
3561 | } | |
3562 | ||
3563 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
3564 | |
3565 | cfq_put_cooperator(cfqq); | |
3566 | ||
e6c5bc73 JM |
3567 | cfq_put_queue(cfqq); |
3568 | return NULL; | |
3569 | } | |
1da177e4 | 3570 | /* |
22e2c507 | 3571 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3572 | */ |
22e2c507 | 3573 | static int |
165125e1 | 3574 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
3575 | { |
3576 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 3577 | struct cfq_io_cq *cic; |
1da177e4 | 3578 | const int rw = rq_data_dir(rq); |
a6151c3a | 3579 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3580 | struct cfq_queue *cfqq; |
1da177e4 LT |
3581 | |
3582 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3583 | ||
216284c3 | 3584 | spin_lock_irq(q->queue_lock); |
c5869807 | 3585 | cic = cfq_get_cic(cfqd, gfp_mask); |
22e2c507 JA |
3586 | if (!cic) |
3587 | goto queue_fail; | |
3588 | ||
e6c5bc73 | 3589 | new_queue: |
91fac317 | 3590 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3591 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
c5869807 | 3592 | cfqq = cfq_get_queue(cfqd, is_sync, cic->icq.ioc, gfp_mask); |
91fac317 | 3593 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3594 | } else { |
e6c5bc73 JM |
3595 | /* |
3596 | * If the queue was seeky for too long, break it apart. | |
3597 | */ | |
ae54abed | 3598 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
3599 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
3600 | cfqq = split_cfqq(cic, cfqq); | |
3601 | if (!cfqq) | |
3602 | goto new_queue; | |
3603 | } | |
3604 | ||
df5fe3e8 JM |
3605 | /* |
3606 | * Check to see if this queue is scheduled to merge with | |
3607 | * another, closely cooperating queue. The merging of | |
3608 | * queues happens here as it must be done in process context. | |
3609 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3610 | */ | |
3611 | if (cfqq->new_cfqq) | |
3612 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3613 | } |
1da177e4 LT |
3614 | |
3615 | cfqq->allocated[rw]++; | |
1da177e4 | 3616 | |
6fae9c25 | 3617 | cfqq->ref++; |
a612fddf TH |
3618 | rq->elv.icq = &cic->icq; |
3619 | rq->elv.priv[0] = cfqq; | |
3620 | rq->elv.priv[1] = cfq_ref_get_cfqg(cfqq->cfqg); | |
216284c3 | 3621 | spin_unlock_irq(q->queue_lock); |
5e705374 | 3622 | return 0; |
1da177e4 | 3623 | |
22e2c507 | 3624 | queue_fail: |
23e018a1 | 3625 | cfq_schedule_dispatch(cfqd); |
216284c3 | 3626 | spin_unlock_irq(q->queue_lock); |
7b679138 | 3627 | cfq_log(cfqd, "set_request fail"); |
1da177e4 LT |
3628 | return 1; |
3629 | } | |
3630 | ||
65f27f38 | 3631 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3632 | { |
65f27f38 | 3633 | struct cfq_data *cfqd = |
23e018a1 | 3634 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3635 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3636 | |
40bb54d1 | 3637 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 3638 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3639 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3640 | } |
3641 | ||
3642 | /* | |
3643 | * Timer running if the active_queue is currently idling inside its time slice | |
3644 | */ | |
3645 | static void cfq_idle_slice_timer(unsigned long data) | |
3646 | { | |
3647 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3648 | struct cfq_queue *cfqq; | |
3649 | unsigned long flags; | |
3c6bd2f8 | 3650 | int timed_out = 1; |
22e2c507 | 3651 | |
7b679138 JA |
3652 | cfq_log(cfqd, "idle timer fired"); |
3653 | ||
22e2c507 JA |
3654 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3655 | ||
fe094d98 JA |
3656 | cfqq = cfqd->active_queue; |
3657 | if (cfqq) { | |
3c6bd2f8 JA |
3658 | timed_out = 0; |
3659 | ||
b029195d JA |
3660 | /* |
3661 | * We saw a request before the queue expired, let it through | |
3662 | */ | |
3663 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3664 | goto out_kick; | |
3665 | ||
22e2c507 JA |
3666 | /* |
3667 | * expired | |
3668 | */ | |
44f7c160 | 3669 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
3670 | goto expire; |
3671 | ||
3672 | /* | |
3673 | * only expire and reinvoke request handler, if there are | |
3674 | * other queues with pending requests | |
3675 | */ | |
caaa5f9f | 3676 | if (!cfqd->busy_queues) |
22e2c507 | 3677 | goto out_cont; |
22e2c507 JA |
3678 | |
3679 | /* | |
3680 | * not expired and it has a request pending, let it dispatch | |
3681 | */ | |
75e50984 | 3682 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3683 | goto out_kick; |
76280aff CZ |
3684 | |
3685 | /* | |
3686 | * Queue depth flag is reset only when the idle didn't succeed | |
3687 | */ | |
3688 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
3689 | } |
3690 | expire: | |
e5ff082e | 3691 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 3692 | out_kick: |
23e018a1 | 3693 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
3694 | out_cont: |
3695 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3696 | } | |
3697 | ||
3b18152c JA |
3698 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3699 | { | |
3700 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 3701 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 3702 | } |
22e2c507 | 3703 | |
c2dea2d1 VT |
3704 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3705 | { | |
3706 | int i; | |
3707 | ||
3708 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
3709 | if (cfqd->async_cfqq[0][i]) | |
3710 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
3711 | if (cfqd->async_cfqq[1][i]) | |
3712 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 3713 | } |
2389d1ef ON |
3714 | |
3715 | if (cfqd->async_idle_cfqq) | |
3716 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
3717 | } |
3718 | ||
b374d18a | 3719 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 3720 | { |
22e2c507 | 3721 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 3722 | struct request_queue *q = cfqd->queue; |
56edf7d7 | 3723 | bool wait = false; |
22e2c507 | 3724 | |
3b18152c | 3725 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 3726 | |
d9ff4187 | 3727 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3728 | |
d9ff4187 | 3729 | if (cfqd->active_queue) |
e5ff082e | 3730 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 3731 | |
c2dea2d1 | 3732 | cfq_put_async_queues(cfqd); |
b1c35769 | 3733 | cfq_release_cfq_groups(cfqd); |
56edf7d7 VG |
3734 | |
3735 | /* | |
3736 | * If there are groups which we could not unlink from blkcg list, | |
3737 | * wait for a rcu period for them to be freed. | |
3738 | */ | |
3739 | if (cfqd->nr_blkcg_linked_grps) | |
3740 | wait = true; | |
15c31be4 | 3741 | |
d9ff4187 | 3742 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
3743 | |
3744 | cfq_shutdown_timer_wq(cfqd); | |
3745 | ||
56edf7d7 VG |
3746 | /* |
3747 | * Wait for cfqg->blkg->key accessors to exit their grace periods. | |
3748 | * Do this wait only if there are other unlinked groups out | |
3749 | * there. This can happen if cgroup deletion path claimed the | |
3750 | * responsibility of cleaning up a group before queue cleanup code | |
3751 | * get to the group. | |
3752 | * | |
3753 | * Do not call synchronize_rcu() unconditionally as there are drivers | |
3754 | * which create/delete request queue hundreds of times during scan/boot | |
3755 | * and synchronize_rcu() can take significant time and slow down boot. | |
3756 | */ | |
3757 | if (wait) | |
3758 | synchronize_rcu(); | |
2abae55f VG |
3759 | |
3760 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
3761 | /* Free up per cpu stats for root group */ | |
3762 | free_percpu(cfqd->root_group.blkg.stats_cpu); | |
3763 | #endif | |
56edf7d7 | 3764 | kfree(cfqd); |
1da177e4 LT |
3765 | } |
3766 | ||
165125e1 | 3767 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
3768 | { |
3769 | struct cfq_data *cfqd; | |
718eee05 | 3770 | int i, j; |
cdb16e8f | 3771 | struct cfq_group *cfqg; |
615f0259 | 3772 | struct cfq_rb_root *st; |
1da177e4 | 3773 | |
94f6030c | 3774 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
a73f730d | 3775 | if (!cfqd) |
bc1c1169 | 3776 | return NULL; |
80b15c73 | 3777 | |
1fa8f6d6 VG |
3778 | /* Init root service tree */ |
3779 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
3780 | ||
cdb16e8f VG |
3781 | /* Init root group */ |
3782 | cfqg = &cfqd->root_group; | |
615f0259 VG |
3783 | for_each_cfqg_st(cfqg, i, j, st) |
3784 | *st = CFQ_RB_ROOT; | |
1fa8f6d6 | 3785 | RB_CLEAR_NODE(&cfqg->rb_node); |
26a2ac00 | 3786 | |
25bc6b07 VG |
3787 | /* Give preference to root group over other groups */ |
3788 | cfqg->weight = 2*BLKIO_WEIGHT_DEFAULT; | |
3789 | ||
25fb5169 | 3790 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
b1c35769 | 3791 | /* |
56edf7d7 VG |
3792 | * Set root group reference to 2. One reference will be dropped when |
3793 | * all groups on cfqd->cfqg_list are being deleted during queue exit. | |
3794 | * Other reference will remain there as we don't want to delete this | |
3795 | * group as it is statically allocated and gets destroyed when | |
3796 | * throtl_data goes away. | |
b1c35769 | 3797 | */ |
56edf7d7 | 3798 | cfqg->ref = 2; |
5624a4e4 VG |
3799 | |
3800 | if (blkio_alloc_blkg_stats(&cfqg->blkg)) { | |
3801 | kfree(cfqg); | |
3802 | kfree(cfqd); | |
3803 | return NULL; | |
3804 | } | |
3805 | ||
dcf097b2 | 3806 | rcu_read_lock(); |
5624a4e4 | 3807 | |
e98ef89b VG |
3808 | cfq_blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg, |
3809 | (void *)cfqd, 0); | |
dcf097b2 | 3810 | rcu_read_unlock(); |
56edf7d7 VG |
3811 | cfqd->nr_blkcg_linked_grps++; |
3812 | ||
3813 | /* Add group on cfqd->cfqg_list */ | |
3814 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); | |
25fb5169 | 3815 | #endif |
26a2ac00 JA |
3816 | /* |
3817 | * Not strictly needed (since RB_ROOT just clears the node and we | |
3818 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
3819 | * to add magic to the rb code | |
3820 | */ | |
3821 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
3822 | cfqd->prio_trees[i] = RB_ROOT; | |
3823 | ||
6118b70b JA |
3824 | /* |
3825 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
3826 | * Grab a permanent reference to it, so that the normal code flow | |
3827 | * will not attempt to free it. | |
3828 | */ | |
3829 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 3830 | cfqd->oom_cfqq.ref++; |
cdb16e8f | 3831 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group); |
6118b70b | 3832 | |
1da177e4 | 3833 | cfqd->queue = q; |
1da177e4 | 3834 | |
22e2c507 JA |
3835 | init_timer(&cfqd->idle_slice_timer); |
3836 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
3837 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
3838 | ||
23e018a1 | 3839 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 3840 | |
1da177e4 | 3841 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
3842 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
3843 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
3844 | cfqd->cfq_back_max = cfq_back_max; |
3845 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
3846 | cfqd->cfq_slice[0] = cfq_slice_async; |
3847 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
3848 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
3849 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
80bdf0c7 | 3850 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 3851 | cfqd->cfq_latency = 1; |
e459dd08 | 3852 | cfqd->hw_tag = -1; |
edc71131 CZ |
3853 | /* |
3854 | * we optimistically start assuming sync ops weren't delayed in last | |
3855 | * second, in order to have larger depth for async operations. | |
3856 | */ | |
573412b2 | 3857 | cfqd->last_delayed_sync = jiffies - HZ; |
bc1c1169 | 3858 | return cfqd; |
1da177e4 LT |
3859 | } |
3860 | ||
1da177e4 LT |
3861 | /* |
3862 | * sysfs parts below --> | |
3863 | */ | |
1da177e4 LT |
3864 | static ssize_t |
3865 | cfq_var_show(unsigned int var, char *page) | |
3866 | { | |
3867 | return sprintf(page, "%d\n", var); | |
3868 | } | |
3869 | ||
3870 | static ssize_t | |
3871 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
3872 | { | |
3873 | char *p = (char *) page; | |
3874 | ||
3875 | *var = simple_strtoul(p, &p, 10); | |
3876 | return count; | |
3877 | } | |
3878 | ||
1da177e4 | 3879 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 3880 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 3881 | { \ |
3d1ab40f | 3882 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3883 | unsigned int __data = __VAR; \ |
3884 | if (__CONV) \ | |
3885 | __data = jiffies_to_msecs(__data); \ | |
3886 | return cfq_var_show(__data, (page)); \ | |
3887 | } | |
3888 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
3889 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
3890 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
3891 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
3892 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 3893 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 3894 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
3895 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
3896 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
3897 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 3898 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
1da177e4 LT |
3899 | #undef SHOW_FUNCTION |
3900 | ||
3901 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 3902 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 3903 | { \ |
3d1ab40f | 3904 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3905 | unsigned int __data; \ |
3906 | int ret = cfq_var_store(&__data, (page), count); \ | |
3907 | if (__data < (MIN)) \ | |
3908 | __data = (MIN); \ | |
3909 | else if (__data > (MAX)) \ | |
3910 | __data = (MAX); \ | |
3911 | if (__CONV) \ | |
3912 | *(__PTR) = msecs_to_jiffies(__data); \ | |
3913 | else \ | |
3914 | *(__PTR) = __data; \ | |
3915 | return ret; \ | |
3916 | } | |
3917 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
3918 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
3919 | UINT_MAX, 1); | |
3920 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
3921 | UINT_MAX, 1); | |
e572ec7e | 3922 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
3923 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
3924 | UINT_MAX, 0); | |
22e2c507 | 3925 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 3926 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
3927 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
3928 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
3929 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
3930 | UINT_MAX, 0); | |
963b72fc | 3931 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
1da177e4 LT |
3932 | #undef STORE_FUNCTION |
3933 | ||
e572ec7e AV |
3934 | #define CFQ_ATTR(name) \ |
3935 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
3936 | ||
3937 | static struct elv_fs_entry cfq_attrs[] = { | |
3938 | CFQ_ATTR(quantum), | |
e572ec7e AV |
3939 | CFQ_ATTR(fifo_expire_sync), |
3940 | CFQ_ATTR(fifo_expire_async), | |
3941 | CFQ_ATTR(back_seek_max), | |
3942 | CFQ_ATTR(back_seek_penalty), | |
3943 | CFQ_ATTR(slice_sync), | |
3944 | CFQ_ATTR(slice_async), | |
3945 | CFQ_ATTR(slice_async_rq), | |
3946 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 3947 | CFQ_ATTR(group_idle), |
963b72fc | 3948 | CFQ_ATTR(low_latency), |
e572ec7e | 3949 | __ATTR_NULL |
1da177e4 LT |
3950 | }; |
3951 | ||
1da177e4 LT |
3952 | static struct elevator_type iosched_cfq = { |
3953 | .ops = { | |
3954 | .elevator_merge_fn = cfq_merge, | |
3955 | .elevator_merged_fn = cfq_merged_request, | |
3956 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 3957 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 3958 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 3959 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 3960 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 3961 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 3962 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 3963 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
3964 | .elevator_former_req_fn = elv_rb_former_request, |
3965 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 3966 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 3967 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
3968 | .elevator_set_req_fn = cfq_set_request, |
3969 | .elevator_put_req_fn = cfq_put_request, | |
3970 | .elevator_may_queue_fn = cfq_may_queue, | |
3971 | .elevator_init_fn = cfq_init_queue, | |
3972 | .elevator_exit_fn = cfq_exit_queue, | |
3973 | }, | |
3d3c2379 TH |
3974 | .icq_size = sizeof(struct cfq_io_cq), |
3975 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 3976 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 3977 | .elevator_name = "cfq", |
1da177e4 LT |
3978 | .elevator_owner = THIS_MODULE, |
3979 | }; | |
3980 | ||
3e252066 VG |
3981 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3982 | static struct blkio_policy_type blkio_policy_cfq = { | |
3983 | .ops = { | |
3984 | .blkio_unlink_group_fn = cfq_unlink_blkio_group, | |
3985 | .blkio_update_group_weight_fn = cfq_update_blkio_group_weight, | |
3986 | }, | |
062a644d | 3987 | .plid = BLKIO_POLICY_PROP, |
3e252066 VG |
3988 | }; |
3989 | #else | |
3990 | static struct blkio_policy_type blkio_policy_cfq; | |
3991 | #endif | |
3992 | ||
1da177e4 LT |
3993 | static int __init cfq_init(void) |
3994 | { | |
3d3c2379 TH |
3995 | int ret; |
3996 | ||
22e2c507 JA |
3997 | /* |
3998 | * could be 0 on HZ < 1000 setups | |
3999 | */ | |
4000 | if (!cfq_slice_async) | |
4001 | cfq_slice_async = 1; | |
4002 | if (!cfq_slice_idle) | |
4003 | cfq_slice_idle = 1; | |
4004 | ||
80bdf0c7 VG |
4005 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4006 | if (!cfq_group_idle) | |
4007 | cfq_group_idle = 1; | |
4008 | #else | |
4009 | cfq_group_idle = 0; | |
4010 | #endif | |
3d3c2379 TH |
4011 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
4012 | if (!cfq_pool) | |
1da177e4 LT |
4013 | return -ENOMEM; |
4014 | ||
3d3c2379 TH |
4015 | ret = elv_register(&iosched_cfq); |
4016 | if (ret) { | |
4017 | kmem_cache_destroy(cfq_pool); | |
4018 | return ret; | |
4019 | } | |
3d3c2379 | 4020 | |
3e252066 | 4021 | blkio_policy_register(&blkio_policy_cfq); |
1da177e4 | 4022 | |
2fdd82bd | 4023 | return 0; |
1da177e4 LT |
4024 | } |
4025 | ||
4026 | static void __exit cfq_exit(void) | |
4027 | { | |
3e252066 | 4028 | blkio_policy_unregister(&blkio_policy_cfq); |
1da177e4 | 4029 | elv_unregister(&iosched_cfq); |
3d3c2379 | 4030 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
4031 | } |
4032 | ||
4033 | module_init(cfq_init); | |
4034 | module_exit(cfq_exit); | |
4035 | ||
4036 | MODULE_AUTHOR("Jens Axboe"); | |
4037 | MODULE_LICENSE("GPL"); | |
4038 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |