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 | 19 | |
0381411e TH |
20 | static struct blkio_policy_type blkio_policy_cfq; |
21 | ||
1da177e4 LT |
22 | /* |
23 | * tunables | |
24 | */ | |
fe094d98 | 25 | /* max queue in one round of service */ |
abc3c744 | 26 | static const int cfq_quantum = 8; |
64100099 | 27 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
28 | /* maximum backwards seek, in KiB */ |
29 | static const int cfq_back_max = 16 * 1024; | |
30 | /* penalty of a backwards seek */ | |
31 | static const int cfq_back_penalty = 2; | |
64100099 | 32 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 33 | static int cfq_slice_async = HZ / 25; |
64100099 | 34 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 35 | static int cfq_slice_idle = HZ / 125; |
80bdf0c7 | 36 | static int cfq_group_idle = HZ / 125; |
5db5d642 CZ |
37 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
38 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 39 | |
d9e7620e | 40 | /* |
0871714e | 41 | * offset from end of service tree |
d9e7620e | 42 | */ |
0871714e | 43 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
44 | |
45 | /* | |
46 | * below this threshold, we consider thinktime immediate | |
47 | */ | |
48 | #define CFQ_MIN_TT (2) | |
49 | ||
22e2c507 | 50 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 51 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 52 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 53 | |
3dde36dd | 54 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 55 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 56 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 57 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 58 | |
a612fddf TH |
59 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
60 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
61 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 62 | |
e18b890b | 63 | static struct kmem_cache *cfq_pool; |
1da177e4 | 64 | |
22e2c507 JA |
65 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
66 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
67 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
68 | ||
206dc69b | 69 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 70 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 71 | |
c5869807 TH |
72 | struct cfq_ttime { |
73 | unsigned long last_end_request; | |
74 | ||
75 | unsigned long ttime_total; | |
76 | unsigned long ttime_samples; | |
77 | unsigned long ttime_mean; | |
78 | }; | |
79 | ||
cc09e299 JA |
80 | /* |
81 | * Most of our rbtree usage is for sorting with min extraction, so | |
82 | * if we cache the leftmost node we don't have to walk down the tree | |
83 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
84 | * move this into the elevator for the rq sorting as well. | |
85 | */ | |
86 | struct cfq_rb_root { | |
87 | struct rb_root rb; | |
88 | struct rb_node *left; | |
aa6f6a3d | 89 | unsigned count; |
73e9ffdd | 90 | unsigned total_weight; |
1fa8f6d6 | 91 | u64 min_vdisktime; |
f5f2b6ce | 92 | struct cfq_ttime ttime; |
cc09e299 | 93 | }; |
f5f2b6ce SL |
94 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
95 | .ttime = {.last_end_request = jiffies,},} | |
cc09e299 | 96 | |
6118b70b JA |
97 | /* |
98 | * Per process-grouping structure | |
99 | */ | |
100 | struct cfq_queue { | |
101 | /* reference count */ | |
30d7b944 | 102 | int ref; |
6118b70b JA |
103 | /* various state flags, see below */ |
104 | unsigned int flags; | |
105 | /* parent cfq_data */ | |
106 | struct cfq_data *cfqd; | |
107 | /* service_tree member */ | |
108 | struct rb_node rb_node; | |
109 | /* service_tree key */ | |
110 | unsigned long rb_key; | |
111 | /* prio tree member */ | |
112 | struct rb_node p_node; | |
113 | /* prio tree root we belong to, if any */ | |
114 | struct rb_root *p_root; | |
115 | /* sorted list of pending requests */ | |
116 | struct rb_root sort_list; | |
117 | /* if fifo isn't expired, next request to serve */ | |
118 | struct request *next_rq; | |
119 | /* requests queued in sort_list */ | |
120 | int queued[2]; | |
121 | /* currently allocated requests */ | |
122 | int allocated[2]; | |
123 | /* fifo list of requests in sort_list */ | |
124 | struct list_head fifo; | |
125 | ||
dae739eb VG |
126 | /* time when queue got scheduled in to dispatch first request. */ |
127 | unsigned long dispatch_start; | |
f75edf2d | 128 | unsigned int allocated_slice; |
c4081ba5 | 129 | unsigned int slice_dispatch; |
dae739eb VG |
130 | /* time when first request from queue completed and slice started. */ |
131 | unsigned long slice_start; | |
6118b70b JA |
132 | unsigned long slice_end; |
133 | long slice_resid; | |
6118b70b | 134 | |
65299a3b CH |
135 | /* pending priority requests */ |
136 | int prio_pending; | |
6118b70b JA |
137 | /* number of requests that are on the dispatch list or inside driver */ |
138 | int dispatched; | |
139 | ||
140 | /* io prio of this group */ | |
141 | unsigned short ioprio, org_ioprio; | |
4aede84b | 142 | unsigned short ioprio_class; |
6118b70b | 143 | |
c4081ba5 RK |
144 | pid_t pid; |
145 | ||
3dde36dd | 146 | u32 seek_history; |
b2c18e1e JM |
147 | sector_t last_request_pos; |
148 | ||
aa6f6a3d | 149 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 150 | struct cfq_queue *new_cfqq; |
cdb16e8f | 151 | struct cfq_group *cfqg; |
c4e7893e VG |
152 | /* Number of sectors dispatched from queue in single dispatch round */ |
153 | unsigned long nr_sectors; | |
6118b70b JA |
154 | }; |
155 | ||
c0324a02 | 156 | /* |
718eee05 | 157 | * First index in the service_trees. |
c0324a02 CZ |
158 | * IDLE is handled separately, so it has negative index |
159 | */ | |
160 | enum wl_prio_t { | |
c0324a02 | 161 | BE_WORKLOAD = 0, |
615f0259 VG |
162 | RT_WORKLOAD = 1, |
163 | IDLE_WORKLOAD = 2, | |
b4627321 | 164 | CFQ_PRIO_NR, |
c0324a02 CZ |
165 | }; |
166 | ||
718eee05 CZ |
167 | /* |
168 | * Second index in the service_trees. | |
169 | */ | |
170 | enum wl_type_t { | |
171 | ASYNC_WORKLOAD = 0, | |
172 | SYNC_NOIDLE_WORKLOAD = 1, | |
173 | SYNC_WORKLOAD = 2 | |
174 | }; | |
175 | ||
cdb16e8f VG |
176 | /* This is per cgroup per device grouping structure */ |
177 | struct cfq_group { | |
1fa8f6d6 VG |
178 | /* group service_tree member */ |
179 | struct rb_node rb_node; | |
180 | ||
181 | /* group service_tree key */ | |
182 | u64 vdisktime; | |
25bc6b07 | 183 | unsigned int weight; |
8184f93e JT |
184 | unsigned int new_weight; |
185 | bool needs_update; | |
1fa8f6d6 VG |
186 | |
187 | /* number of cfqq currently on this group */ | |
188 | int nr_cfqq; | |
189 | ||
cdb16e8f | 190 | /* |
4495a7d4 | 191 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
192 | * create the array for each prio class but at run time it is used |
193 | * only for RT and BE class and slot for IDLE class remains unused. | |
194 | * This is primarily done to avoid confusion and a gcc warning. | |
195 | */ | |
196 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
197 | /* | |
198 | * rr lists of queues with requests. We maintain service trees for | |
199 | * RT and BE classes. These trees are subdivided in subclasses | |
200 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
201 | * class there is no subclassification and all the cfq queues go on | |
202 | * a single tree service_tree_idle. | |
cdb16e8f VG |
203 | * Counts are embedded in the cfq_rb_root |
204 | */ | |
205 | struct cfq_rb_root service_trees[2][3]; | |
206 | struct cfq_rb_root service_tree_idle; | |
dae739eb VG |
207 | |
208 | unsigned long saved_workload_slice; | |
209 | enum wl_type_t saved_workload; | |
210 | enum wl_prio_t saved_serving_prio; | |
25fb5169 VG |
211 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
212 | struct hlist_node cfqd_node; | |
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; | |
f51b802c | 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 | ||
0381411e TH |
313 | static inline struct cfq_group *blkg_to_cfqg(struct blkio_group *blkg) |
314 | { | |
315 | return blkg_to_pdata(blkg, &blkio_policy_cfq); | |
316 | } | |
317 | ||
318 | static inline struct blkio_group *cfqg_to_blkg(struct cfq_group *cfqg) | |
319 | { | |
320 | return pdata_to_blkg(cfqg, &blkio_policy_cfq); | |
321 | } | |
322 | ||
25fb5169 VG |
323 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
324 | ||
cdb16e8f VG |
325 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
326 | enum wl_prio_t prio, | |
65b32a57 | 327 | enum wl_type_t type) |
c0324a02 | 328 | { |
1fa8f6d6 VG |
329 | if (!cfqg) |
330 | return NULL; | |
331 | ||
c0324a02 | 332 | if (prio == IDLE_WORKLOAD) |
cdb16e8f | 333 | return &cfqg->service_tree_idle; |
c0324a02 | 334 | |
cdb16e8f | 335 | return &cfqg->service_trees[prio][type]; |
c0324a02 CZ |
336 | } |
337 | ||
3b18152c | 338 | enum cfqq_state_flags { |
b0b8d749 JA |
339 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
340 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 341 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 342 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
343 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
344 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
345 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 346 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 347 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 348 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 349 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 350 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 351 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
352 | }; |
353 | ||
354 | #define CFQ_CFQQ_FNS(name) \ | |
355 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
356 | { \ | |
fe094d98 | 357 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
358 | } \ |
359 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
360 | { \ | |
fe094d98 | 361 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
362 | } \ |
363 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
364 | { \ | |
fe094d98 | 365 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
366 | } |
367 | ||
368 | CFQ_CFQQ_FNS(on_rr); | |
369 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 370 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 371 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
372 | CFQ_CFQQ_FNS(fifo_expire); |
373 | CFQ_CFQQ_FNS(idle_window); | |
374 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 375 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 376 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 377 | CFQ_CFQQ_FNS(coop); |
ae54abed | 378 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 379 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 380 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
381 | #undef CFQ_CFQQ_FNS |
382 | ||
afc24d49 | 383 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2868ef7b VG |
384 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
385 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ | |
386 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
0381411e | 387 | blkg_path(cfqg_to_blkg((cfqq)->cfqg)), ##args) |
2868ef7b VG |
388 | |
389 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ | |
390 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ | |
0381411e | 391 | blkg_path(cfqg_to_blkg((cfqg))), ##args) \ |
2868ef7b VG |
392 | |
393 | #else | |
7b679138 JA |
394 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
395 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
4495a7d4 | 396 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
2868ef7b | 397 | #endif |
7b679138 JA |
398 | #define cfq_log(cfqd, fmt, args...) \ |
399 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
400 | ||
615f0259 VG |
401 | /* Traverses through cfq group service trees */ |
402 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
403 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
404 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
405 | : &cfqg->service_tree_idle; \ | |
406 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
407 | (i == IDLE_WORKLOAD && j == 0); \ | |
408 | j++, st = i < IDLE_WORKLOAD ? \ | |
409 | &cfqg->service_trees[i][j]: NULL) \ | |
410 | ||
f5f2b6ce SL |
411 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
412 | struct cfq_ttime *ttime, bool group_idle) | |
413 | { | |
414 | unsigned long slice; | |
415 | if (!sample_valid(ttime->ttime_samples)) | |
416 | return false; | |
417 | if (group_idle) | |
418 | slice = cfqd->cfq_group_idle; | |
419 | else | |
420 | slice = cfqd->cfq_slice_idle; | |
421 | return ttime->ttime_mean > slice; | |
422 | } | |
615f0259 | 423 | |
02b35081 VG |
424 | static inline bool iops_mode(struct cfq_data *cfqd) |
425 | { | |
426 | /* | |
427 | * If we are not idling on queues and it is a NCQ drive, parallel | |
428 | * execution of requests is on and measuring time is not possible | |
429 | * in most of the cases until and unless we drive shallower queue | |
430 | * depths and that becomes a performance bottleneck. In such cases | |
431 | * switch to start providing fairness in terms of number of IOs. | |
432 | */ | |
433 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
434 | return true; | |
435 | else | |
436 | return false; | |
437 | } | |
438 | ||
c0324a02 CZ |
439 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
440 | { | |
441 | if (cfq_class_idle(cfqq)) | |
442 | return IDLE_WORKLOAD; | |
443 | if (cfq_class_rt(cfqq)) | |
444 | return RT_WORKLOAD; | |
445 | return BE_WORKLOAD; | |
446 | } | |
447 | ||
718eee05 CZ |
448 | |
449 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
450 | { | |
451 | if (!cfq_cfqq_sync(cfqq)) | |
452 | return ASYNC_WORKLOAD; | |
453 | if (!cfq_cfqq_idle_window(cfqq)) | |
454 | return SYNC_NOIDLE_WORKLOAD; | |
455 | return SYNC_WORKLOAD; | |
456 | } | |
457 | ||
58ff82f3 VG |
458 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
459 | struct cfq_data *cfqd, | |
460 | struct cfq_group *cfqg) | |
c0324a02 CZ |
461 | { |
462 | if (wl == IDLE_WORKLOAD) | |
cdb16e8f | 463 | return cfqg->service_tree_idle.count; |
c0324a02 | 464 | |
cdb16e8f VG |
465 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
466 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count | |
467 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
468 | } |
469 | ||
f26bd1f0 VG |
470 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
471 | struct cfq_group *cfqg) | |
472 | { | |
473 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count | |
474 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
475 | } | |
476 | ||
165125e1 | 477 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
a6151c3a | 478 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool, |
fd0928df | 479 | struct io_context *, gfp_t); |
91fac317 | 480 | |
c5869807 TH |
481 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
482 | { | |
483 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
484 | return container_of(icq, struct cfq_io_cq, icq); | |
485 | } | |
486 | ||
47fdd4ca TH |
487 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
488 | struct io_context *ioc) | |
489 | { | |
490 | if (ioc) | |
491 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
492 | return NULL; | |
493 | } | |
494 | ||
c5869807 | 495 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 496 | { |
a6151c3a | 497 | return cic->cfqq[is_sync]; |
91fac317 VT |
498 | } |
499 | ||
c5869807 TH |
500 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
501 | bool is_sync) | |
91fac317 | 502 | { |
a6151c3a | 503 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
504 | } |
505 | ||
c5869807 | 506 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 507 | { |
c5869807 | 508 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
509 | } |
510 | ||
91fac317 VT |
511 | /* |
512 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
513 | * set (in which case it could also be direct WRITE). | |
514 | */ | |
a6151c3a | 515 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 516 | { |
7b6d91da | 517 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 518 | } |
1da177e4 | 519 | |
99f95e52 AM |
520 | /* |
521 | * scheduler run of queue, if there are requests pending and no one in the | |
522 | * driver that will restart queueing | |
523 | */ | |
23e018a1 | 524 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 525 | { |
7b679138 JA |
526 | if (cfqd->busy_queues) { |
527 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 528 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 529 | } |
99f95e52 AM |
530 | } |
531 | ||
44f7c160 JA |
532 | /* |
533 | * Scale schedule slice based on io priority. Use the sync time slice only | |
534 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
535 | * io only, should not get full sync slice length. | |
536 | */ | |
a6151c3a | 537 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 538 | unsigned short prio) |
44f7c160 | 539 | { |
d9e7620e | 540 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 541 | |
d9e7620e JA |
542 | WARN_ON(prio >= IOPRIO_BE_NR); |
543 | ||
544 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
545 | } | |
44f7c160 | 546 | |
d9e7620e JA |
547 | static inline int |
548 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
549 | { | |
550 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
551 | } |
552 | ||
25bc6b07 VG |
553 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
554 | { | |
555 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
556 | ||
557 | d = d * BLKIO_WEIGHT_DEFAULT; | |
558 | do_div(d, cfqg->weight); | |
559 | return d; | |
560 | } | |
561 | ||
562 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
563 | { | |
564 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
565 | if (delta > 0) | |
566 | min_vdisktime = vdisktime; | |
567 | ||
568 | return min_vdisktime; | |
569 | } | |
570 | ||
571 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
572 | { | |
573 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
574 | if (delta < 0) | |
575 | min_vdisktime = vdisktime; | |
576 | ||
577 | return min_vdisktime; | |
578 | } | |
579 | ||
580 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
581 | { | |
25bc6b07 VG |
582 | struct cfq_group *cfqg; |
583 | ||
25bc6b07 VG |
584 | if (st->left) { |
585 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
586 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
587 | cfqg->vdisktime); | |
25bc6b07 | 588 | } |
25bc6b07 VG |
589 | } |
590 | ||
5db5d642 CZ |
591 | /* |
592 | * get averaged number of queues of RT/BE priority. | |
593 | * average is updated, with a formula that gives more weight to higher numbers, | |
594 | * to quickly follows sudden increases and decrease slowly | |
595 | */ | |
596 | ||
58ff82f3 VG |
597 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
598 | struct cfq_group *cfqg, bool rt) | |
5869619c | 599 | { |
5db5d642 CZ |
600 | unsigned min_q, max_q; |
601 | unsigned mult = cfq_hist_divisor - 1; | |
602 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 603 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 604 | |
58ff82f3 VG |
605 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
606 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
607 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 608 | cfq_hist_divisor; |
58ff82f3 VG |
609 | return cfqg->busy_queues_avg[rt]; |
610 | } | |
611 | ||
612 | static inline unsigned | |
613 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
614 | { | |
615 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
616 | ||
617 | return cfq_target_latency * cfqg->weight / st->total_weight; | |
5db5d642 CZ |
618 | } |
619 | ||
c553f8e3 | 620 | static inline unsigned |
ba5bd520 | 621 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 622 | { |
5db5d642 CZ |
623 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
624 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
625 | /* |
626 | * interested queues (we consider only the ones with the same | |
627 | * priority class in the cfq group) | |
628 | */ | |
629 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
630 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
631 | unsigned sync_slice = cfqd->cfq_slice[1]; |
632 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
633 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
634 | ||
635 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
636 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
637 | /* scale low_slice according to IO priority | |
638 | * and sync vs async */ | |
639 | unsigned low_slice = | |
640 | min(slice, base_low_slice * slice / sync_slice); | |
641 | /* the adapted slice value is scaled to fit all iqs | |
642 | * into the target latency */ | |
58ff82f3 | 643 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
644 | low_slice); |
645 | } | |
646 | } | |
c553f8e3 SL |
647 | return slice; |
648 | } | |
649 | ||
650 | static inline void | |
651 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
652 | { | |
ba5bd520 | 653 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 654 | |
dae739eb | 655 | cfqq->slice_start = jiffies; |
5db5d642 | 656 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 657 | cfqq->allocated_slice = slice; |
7b679138 | 658 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
659 | } |
660 | ||
661 | /* | |
662 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
663 | * isn't valid until the first request from the dispatch is activated | |
664 | * and the slice time set. | |
665 | */ | |
a6151c3a | 666 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
667 | { |
668 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 669 | return false; |
44f7c160 | 670 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 671 | return false; |
44f7c160 | 672 | |
c1e44756 | 673 | return true; |
44f7c160 JA |
674 | } |
675 | ||
1da177e4 | 676 | /* |
5e705374 | 677 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 678 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 679 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 680 | */ |
5e705374 | 681 | static struct request * |
cf7c25cf | 682 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 683 | { |
cf7c25cf | 684 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 685 | unsigned long back_max; |
e8a99053 AM |
686 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
687 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
688 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 689 | |
5e705374 JA |
690 | if (rq1 == NULL || rq1 == rq2) |
691 | return rq2; | |
692 | if (rq2 == NULL) | |
693 | return rq1; | |
9c2c38a1 | 694 | |
229836bd NK |
695 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
696 | return rq_is_sync(rq1) ? rq1 : rq2; | |
697 | ||
65299a3b CH |
698 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
699 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 700 | |
83096ebf TH |
701 | s1 = blk_rq_pos(rq1); |
702 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 703 | |
1da177e4 LT |
704 | /* |
705 | * by definition, 1KiB is 2 sectors | |
706 | */ | |
707 | back_max = cfqd->cfq_back_max * 2; | |
708 | ||
709 | /* | |
710 | * Strict one way elevator _except_ in the case where we allow | |
711 | * short backward seeks which are biased as twice the cost of a | |
712 | * similar forward seek. | |
713 | */ | |
714 | if (s1 >= last) | |
715 | d1 = s1 - last; | |
716 | else if (s1 + back_max >= last) | |
717 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
718 | else | |
e8a99053 | 719 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
720 | |
721 | if (s2 >= last) | |
722 | d2 = s2 - last; | |
723 | else if (s2 + back_max >= last) | |
724 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
725 | else | |
e8a99053 | 726 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
727 | |
728 | /* Found required data */ | |
e8a99053 AM |
729 | |
730 | /* | |
731 | * By doing switch() on the bit mask "wrap" we avoid having to | |
732 | * check two variables for all permutations: --> faster! | |
733 | */ | |
734 | switch (wrap) { | |
5e705374 | 735 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 736 | if (d1 < d2) |
5e705374 | 737 | return rq1; |
e8a99053 | 738 | else if (d2 < d1) |
5e705374 | 739 | return rq2; |
e8a99053 AM |
740 | else { |
741 | if (s1 >= s2) | |
5e705374 | 742 | return rq1; |
e8a99053 | 743 | else |
5e705374 | 744 | return rq2; |
e8a99053 | 745 | } |
1da177e4 | 746 | |
e8a99053 | 747 | case CFQ_RQ2_WRAP: |
5e705374 | 748 | return rq1; |
e8a99053 | 749 | case CFQ_RQ1_WRAP: |
5e705374 JA |
750 | return rq2; |
751 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
752 | default: |
753 | /* | |
754 | * Since both rqs are wrapped, | |
755 | * start with the one that's further behind head | |
756 | * (--> only *one* back seek required), | |
757 | * since back seek takes more time than forward. | |
758 | */ | |
759 | if (s1 <= s2) | |
5e705374 | 760 | return rq1; |
1da177e4 | 761 | else |
5e705374 | 762 | return rq2; |
1da177e4 LT |
763 | } |
764 | } | |
765 | ||
498d3aa2 JA |
766 | /* |
767 | * The below is leftmost cache rbtree addon | |
768 | */ | |
0871714e | 769 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 770 | { |
615f0259 VG |
771 | /* Service tree is empty */ |
772 | if (!root->count) | |
773 | return NULL; | |
774 | ||
cc09e299 JA |
775 | if (!root->left) |
776 | root->left = rb_first(&root->rb); | |
777 | ||
0871714e JA |
778 | if (root->left) |
779 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
780 | ||
781 | return NULL; | |
cc09e299 JA |
782 | } |
783 | ||
1fa8f6d6 VG |
784 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
785 | { | |
786 | if (!root->left) | |
787 | root->left = rb_first(&root->rb); | |
788 | ||
789 | if (root->left) | |
790 | return rb_entry_cfqg(root->left); | |
791 | ||
792 | return NULL; | |
793 | } | |
794 | ||
a36e71f9 JA |
795 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
796 | { | |
797 | rb_erase(n, root); | |
798 | RB_CLEAR_NODE(n); | |
799 | } | |
800 | ||
cc09e299 JA |
801 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
802 | { | |
803 | if (root->left == n) | |
804 | root->left = NULL; | |
a36e71f9 | 805 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 806 | --root->count; |
cc09e299 JA |
807 | } |
808 | ||
1da177e4 LT |
809 | /* |
810 | * would be nice to take fifo expire time into account as well | |
811 | */ | |
5e705374 JA |
812 | static struct request * |
813 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
814 | struct request *last) | |
1da177e4 | 815 | { |
21183b07 JA |
816 | struct rb_node *rbnext = rb_next(&last->rb_node); |
817 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 818 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 819 | |
21183b07 | 820 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
821 | |
822 | if (rbprev) | |
5e705374 | 823 | prev = rb_entry_rq(rbprev); |
1da177e4 | 824 | |
21183b07 | 825 | if (rbnext) |
5e705374 | 826 | next = rb_entry_rq(rbnext); |
21183b07 JA |
827 | else { |
828 | rbnext = rb_first(&cfqq->sort_list); | |
829 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 830 | next = rb_entry_rq(rbnext); |
21183b07 | 831 | } |
1da177e4 | 832 | |
cf7c25cf | 833 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
834 | } |
835 | ||
d9e7620e JA |
836 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
837 | struct cfq_queue *cfqq) | |
1da177e4 | 838 | { |
d9e7620e JA |
839 | /* |
840 | * just an approximation, should be ok. | |
841 | */ | |
cdb16e8f | 842 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 843 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
844 | } |
845 | ||
1fa8f6d6 VG |
846 | static inline s64 |
847 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
848 | { | |
849 | return cfqg->vdisktime - st->min_vdisktime; | |
850 | } | |
851 | ||
852 | static void | |
853 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
854 | { | |
855 | struct rb_node **node = &st->rb.rb_node; | |
856 | struct rb_node *parent = NULL; | |
857 | struct cfq_group *__cfqg; | |
858 | s64 key = cfqg_key(st, cfqg); | |
859 | int left = 1; | |
860 | ||
861 | while (*node != NULL) { | |
862 | parent = *node; | |
863 | __cfqg = rb_entry_cfqg(parent); | |
864 | ||
865 | if (key < cfqg_key(st, __cfqg)) | |
866 | node = &parent->rb_left; | |
867 | else { | |
868 | node = &parent->rb_right; | |
869 | left = 0; | |
870 | } | |
871 | } | |
872 | ||
873 | if (left) | |
874 | st->left = &cfqg->rb_node; | |
875 | ||
876 | rb_link_node(&cfqg->rb_node, parent, node); | |
877 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
878 | } | |
879 | ||
880 | static void | |
8184f93e JT |
881 | cfq_update_group_weight(struct cfq_group *cfqg) |
882 | { | |
883 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
884 | if (cfqg->needs_update) { | |
885 | cfqg->weight = cfqg->new_weight; | |
886 | cfqg->needs_update = false; | |
887 | } | |
888 | } | |
889 | ||
890 | static void | |
891 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
892 | { | |
893 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
894 | ||
895 | cfq_update_group_weight(cfqg); | |
896 | __cfq_group_service_tree_add(st, cfqg); | |
897 | st->total_weight += cfqg->weight; | |
898 | } | |
899 | ||
900 | static void | |
901 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
902 | { |
903 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
904 | struct cfq_group *__cfqg; | |
905 | struct rb_node *n; | |
906 | ||
907 | cfqg->nr_cfqq++; | |
760701bf | 908 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
909 | return; |
910 | ||
911 | /* | |
912 | * Currently put the group at the end. Later implement something | |
913 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 914 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
915 | */ |
916 | n = rb_last(&st->rb); | |
917 | if (n) { | |
918 | __cfqg = rb_entry_cfqg(n); | |
919 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
920 | } else | |
921 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
922 | cfq_group_service_tree_add(st, cfqg); |
923 | } | |
1fa8f6d6 | 924 | |
8184f93e JT |
925 | static void |
926 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
927 | { | |
928 | st->total_weight -= cfqg->weight; | |
929 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) | |
930 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
931 | } |
932 | ||
933 | static void | |
8184f93e | 934 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
935 | { |
936 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
937 | ||
938 | BUG_ON(cfqg->nr_cfqq < 1); | |
939 | cfqg->nr_cfqq--; | |
25bc6b07 | 940 | |
1fa8f6d6 VG |
941 | /* If there are other cfq queues under this group, don't delete it */ |
942 | if (cfqg->nr_cfqq) | |
943 | return; | |
944 | ||
2868ef7b | 945 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 946 | cfq_group_service_tree_del(st, cfqg); |
dae739eb | 947 | cfqg->saved_workload_slice = 0; |
c1768268 TH |
948 | cfq_blkiocg_update_dequeue_stats(cfqg_to_blkg(cfqg), |
949 | &blkio_policy_cfq, 1); | |
dae739eb VG |
950 | } |
951 | ||
167400d3 JT |
952 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
953 | unsigned int *unaccounted_time) | |
dae739eb | 954 | { |
f75edf2d | 955 | unsigned int slice_used; |
dae739eb VG |
956 | |
957 | /* | |
958 | * Queue got expired before even a single request completed or | |
959 | * got expired immediately after first request completion. | |
960 | */ | |
961 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
962 | /* | |
963 | * Also charge the seek time incurred to the group, otherwise | |
964 | * if there are mutiple queues in the group, each can dispatch | |
965 | * a single request on seeky media and cause lots of seek time | |
966 | * and group will never know it. | |
967 | */ | |
968 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
969 | 1); | |
970 | } else { | |
971 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
972 | if (slice_used > cfqq->allocated_slice) { |
973 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 974 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
975 | } |
976 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
977 | *unaccounted_time += cfqq->slice_start - | |
978 | cfqq->dispatch_start; | |
dae739eb VG |
979 | } |
980 | ||
dae739eb VG |
981 | return slice_used; |
982 | } | |
983 | ||
984 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 985 | struct cfq_queue *cfqq) |
dae739eb VG |
986 | { |
987 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 988 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
989 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
990 | - cfqg->service_tree_idle.count; | |
991 | ||
992 | BUG_ON(nr_sync < 0); | |
167400d3 | 993 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 994 | |
02b35081 VG |
995 | if (iops_mode(cfqd)) |
996 | charge = cfqq->slice_dispatch; | |
997 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
998 | charge = cfqq->allocated_slice; | |
dae739eb VG |
999 | |
1000 | /* Can't update vdisktime while group is on service tree */ | |
8184f93e | 1001 | cfq_group_service_tree_del(st, cfqg); |
02b35081 | 1002 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
8184f93e JT |
1003 | /* If a new weight was requested, update now, off tree */ |
1004 | cfq_group_service_tree_add(st, cfqg); | |
dae739eb VG |
1005 | |
1006 | /* This group is being expired. Save the context */ | |
1007 | if (time_after(cfqd->workload_expires, jiffies)) { | |
1008 | cfqg->saved_workload_slice = cfqd->workload_expires | |
1009 | - jiffies; | |
1010 | cfqg->saved_workload = cfqd->serving_type; | |
1011 | cfqg->saved_serving_prio = cfqd->serving_prio; | |
1012 | } else | |
1013 | cfqg->saved_workload_slice = 0; | |
2868ef7b VG |
1014 | |
1015 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1016 | st->min_vdisktime); | |
fd16d263 JP |
1017 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1018 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1019 | used_sl, cfqq->slice_dispatch, charge, | |
1020 | iops_mode(cfqd), cfqq->nr_sectors); | |
c1768268 TH |
1021 | cfq_blkiocg_update_timeslice_used(cfqg_to_blkg(cfqg), &blkio_policy_cfq, |
1022 | used_sl, unaccounted_sl); | |
1023 | cfq_blkiocg_set_start_empty_time(cfqg_to_blkg(cfqg), &blkio_policy_cfq); | |
1fa8f6d6 VG |
1024 | } |
1025 | ||
f51b802c TH |
1026 | /** |
1027 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1028 | * @cfqg: cfq_group to initialize | |
1029 | * | |
1030 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1031 | * is enabled or not. | |
1032 | */ | |
1033 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1034 | { | |
1035 | struct cfq_rb_root *st; | |
1036 | int i, j; | |
1037 | ||
1038 | for_each_cfqg_st(cfqg, i, j, st) | |
1039 | *st = CFQ_RB_ROOT; | |
1040 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1041 | ||
1042 | cfqg->ttime.last_end_request = jiffies; | |
1043 | } | |
1044 | ||
25fb5169 | 1045 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
ca32aefc TH |
1046 | static void cfq_update_blkio_group_weight(struct request_queue *q, |
1047 | struct blkio_group *blkg, | |
8aea4545 | 1048 | unsigned int weight) |
f8d461d6 | 1049 | { |
0381411e TH |
1050 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
1051 | ||
8184f93e JT |
1052 | cfqg->new_weight = weight; |
1053 | cfqg->needs_update = true; | |
f8d461d6 VG |
1054 | } |
1055 | ||
cd1604fa TH |
1056 | static void cfq_link_blkio_group(struct request_queue *q, |
1057 | struct blkio_group *blkg) | |
25fb5169 | 1058 | { |
cd1604fa | 1059 | struct cfq_data *cfqd = q->elevator->elevator_data; |
0381411e | 1060 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
f469a7b4 VG |
1061 | |
1062 | cfqd->nr_blkcg_linked_grps++; | |
f469a7b4 VG |
1063 | |
1064 | /* Add group on cfqd list */ | |
1065 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); | |
1066 | } | |
1067 | ||
0381411e | 1068 | static void cfq_init_blkio_group(struct blkio_group *blkg) |
f469a7b4 | 1069 | { |
0381411e | 1070 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
25fb5169 | 1071 | |
f51b802c | 1072 | cfq_init_cfqg_base(cfqg); |
0381411e | 1073 | cfqg->weight = blkg->blkcg->weight; |
25fb5169 VG |
1074 | } |
1075 | ||
1076 | /* | |
3e59cf9d VG |
1077 | * Search for the cfq group current task belongs to. request_queue lock must |
1078 | * be held. | |
25fb5169 | 1079 | */ |
cd1604fa TH |
1080 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
1081 | struct blkio_cgroup *blkcg) | |
25fb5169 | 1082 | { |
f469a7b4 | 1083 | struct request_queue *q = cfqd->queue; |
cd1604fa | 1084 | struct cfq_group *cfqg = NULL; |
25fb5169 | 1085 | |
cd1604fa TH |
1086 | /* avoid lookup for the common case where there's no blkio cgroup */ |
1087 | if (blkcg == &blkio_root_cgroup) { | |
1088 | cfqg = cfqd->root_group; | |
1089 | } else { | |
1090 | struct blkio_group *blkg; | |
f469a7b4 | 1091 | |
cd1604fa TH |
1092 | blkg = blkg_lookup_create(blkcg, q, BLKIO_POLICY_PROP, false); |
1093 | if (!IS_ERR(blkg)) | |
0381411e | 1094 | cfqg = blkg_to_cfqg(blkg); |
cd1604fa | 1095 | } |
f469a7b4 | 1096 | |
25fb5169 VG |
1097 | return cfqg; |
1098 | } | |
1099 | ||
1100 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1101 | { | |
1102 | /* Currently, all async queues are mapped to root group */ | |
1103 | if (!cfq_cfqq_sync(cfqq)) | |
f51b802c | 1104 | cfqg = cfqq->cfqd->root_group; |
25fb5169 VG |
1105 | |
1106 | cfqq->cfqg = cfqg; | |
b1c35769 | 1107 | /* cfqq reference on cfqg */ |
1adaf3dd | 1108 | blkg_get(cfqg_to_blkg(cfqg)); |
b1c35769 VG |
1109 | } |
1110 | ||
1111 | static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1112 | { | |
1113 | /* Something wrong if we are trying to remove same group twice */ | |
1114 | BUG_ON(hlist_unhashed(&cfqg->cfqd_node)); | |
1115 | ||
1116 | hlist_del_init(&cfqg->cfqd_node); | |
1117 | ||
a5395b83 VG |
1118 | BUG_ON(cfqd->nr_blkcg_linked_grps <= 0); |
1119 | cfqd->nr_blkcg_linked_grps--; | |
1120 | ||
b1c35769 VG |
1121 | /* |
1122 | * Put the reference taken at the time of creation so that when all | |
1123 | * queues are gone, group can be destroyed. | |
1124 | */ | |
1adaf3dd | 1125 | blkg_put(cfqg_to_blkg(cfqg)); |
b1c35769 VG |
1126 | } |
1127 | ||
72e06c25 | 1128 | static bool cfq_release_cfq_groups(struct cfq_data *cfqd) |
b1c35769 VG |
1129 | { |
1130 | struct hlist_node *pos, *n; | |
1131 | struct cfq_group *cfqg; | |
72e06c25 | 1132 | bool empty = true; |
b1c35769 VG |
1133 | |
1134 | hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) { | |
1135 | /* | |
1136 | * If cgroup removal path got to blk_group first and removed | |
1137 | * it from cgroup list, then it will take care of destroying | |
1138 | * cfqg also. | |
1139 | */ | |
0381411e | 1140 | if (!cfq_blkiocg_del_blkio_group(cfqg_to_blkg(cfqg))) |
b1c35769 | 1141 | cfq_destroy_cfqg(cfqd, cfqg); |
72e06c25 TH |
1142 | else |
1143 | empty = false; | |
b1c35769 | 1144 | } |
72e06c25 | 1145 | return empty; |
25fb5169 | 1146 | } |
b1c35769 VG |
1147 | |
1148 | /* | |
1149 | * Blk cgroup controller notification saying that blkio_group object is being | |
1150 | * delinked as associated cgroup object is going away. That also means that | |
1151 | * no new IO will come in this group. So get rid of this group as soon as | |
1152 | * any pending IO in the group is finished. | |
1153 | * | |
1154 | * This function is called under rcu_read_lock(). key is the rcu protected | |
ca32aefc TH |
1155 | * pointer. That means @q is a valid request_queue pointer as long as we |
1156 | * are rcu read lock. | |
b1c35769 | 1157 | * |
ca32aefc | 1158 | * @q was fetched from blkio_group under blkio_cgroup->lock. That means |
b1c35769 VG |
1159 | * it should not be NULL as even if elevator was exiting, cgroup deltion |
1160 | * path got to it first. | |
1161 | */ | |
ca32aefc TH |
1162 | static void cfq_unlink_blkio_group(struct request_queue *q, |
1163 | struct blkio_group *blkg) | |
b1c35769 | 1164 | { |
ca32aefc TH |
1165 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1166 | unsigned long flags; | |
b1c35769 | 1167 | |
ca32aefc | 1168 | spin_lock_irqsave(q->queue_lock, flags); |
0381411e | 1169 | cfq_destroy_cfqg(cfqd, blkg_to_cfqg(blkg)); |
ca32aefc | 1170 | spin_unlock_irqrestore(q->queue_lock, flags); |
b1c35769 VG |
1171 | } |
1172 | ||
72e06c25 TH |
1173 | static struct elevator_type iosched_cfq; |
1174 | ||
1175 | static bool cfq_clear_queue(struct request_queue *q) | |
1176 | { | |
1177 | lockdep_assert_held(q->queue_lock); | |
1178 | ||
1179 | /* shoot down blkgs iff the current elevator is cfq */ | |
1180 | if (!q->elevator || q->elevator->type != &iosched_cfq) | |
1181 | return true; | |
1182 | ||
1183 | return cfq_release_cfq_groups(q->elevator->elevator_data); | |
1184 | } | |
1185 | ||
25fb5169 | 1186 | #else /* GROUP_IOSCHED */ |
cd1604fa TH |
1187 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
1188 | struct blkio_cgroup *blkcg) | |
25fb5169 | 1189 | { |
f51b802c | 1190 | return cfqd->root_group; |
25fb5169 | 1191 | } |
7f1dc8a2 | 1192 | |
25fb5169 VG |
1193 | static inline void |
1194 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1195 | cfqq->cfqg = cfqg; | |
1196 | } | |
1197 | ||
b1c35769 | 1198 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) {} |
b1c35769 | 1199 | |
25fb5169 VG |
1200 | #endif /* GROUP_IOSCHED */ |
1201 | ||
498d3aa2 | 1202 | /* |
c0324a02 | 1203 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1204 | * requests waiting to be processed. It is sorted in the order that |
1205 | * we will service the queues. | |
1206 | */ | |
a36e71f9 | 1207 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1208 | bool add_front) |
d9e7620e | 1209 | { |
0871714e JA |
1210 | struct rb_node **p, *parent; |
1211 | struct cfq_queue *__cfqq; | |
d9e7620e | 1212 | unsigned long rb_key; |
c0324a02 | 1213 | struct cfq_rb_root *service_tree; |
498d3aa2 | 1214 | int left; |
dae739eb | 1215 | int new_cfqq = 1; |
ae30c286 | 1216 | |
cdb16e8f | 1217 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
65b32a57 | 1218 | cfqq_type(cfqq)); |
0871714e JA |
1219 | if (cfq_class_idle(cfqq)) { |
1220 | rb_key = CFQ_IDLE_DELAY; | |
aa6f6a3d | 1221 | parent = rb_last(&service_tree->rb); |
0871714e JA |
1222 | if (parent && parent != &cfqq->rb_node) { |
1223 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1224 | rb_key += __cfqq->rb_key; | |
1225 | } else | |
1226 | rb_key += jiffies; | |
1227 | } else if (!add_front) { | |
b9c8946b JA |
1228 | /* |
1229 | * Get our rb key offset. Subtract any residual slice | |
1230 | * value carried from last service. A negative resid | |
1231 | * count indicates slice overrun, and this should position | |
1232 | * the next service time further away in the tree. | |
1233 | */ | |
edd75ffd | 1234 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1235 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1236 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1237 | } else { |
1238 | rb_key = -HZ; | |
aa6f6a3d | 1239 | __cfqq = cfq_rb_first(service_tree); |
48e025e6 CZ |
1240 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1241 | } | |
1da177e4 | 1242 | |
d9e7620e | 1243 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1244 | new_cfqq = 0; |
99f9628a | 1245 | /* |
d9e7620e | 1246 | * same position, nothing more to do |
99f9628a | 1247 | */ |
c0324a02 CZ |
1248 | if (rb_key == cfqq->rb_key && |
1249 | cfqq->service_tree == service_tree) | |
d9e7620e | 1250 | return; |
1da177e4 | 1251 | |
aa6f6a3d CZ |
1252 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1253 | cfqq->service_tree = NULL; | |
1da177e4 | 1254 | } |
d9e7620e | 1255 | |
498d3aa2 | 1256 | left = 1; |
0871714e | 1257 | parent = NULL; |
aa6f6a3d CZ |
1258 | cfqq->service_tree = service_tree; |
1259 | p = &service_tree->rb.rb_node; | |
d9e7620e | 1260 | while (*p) { |
67060e37 | 1261 | struct rb_node **n; |
cc09e299 | 1262 | |
d9e7620e JA |
1263 | parent = *p; |
1264 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1265 | ||
0c534e0a | 1266 | /* |
c0324a02 | 1267 | * sort by key, that represents service time. |
0c534e0a | 1268 | */ |
c0324a02 | 1269 | if (time_before(rb_key, __cfqq->rb_key)) |
67060e37 | 1270 | n = &(*p)->rb_left; |
c0324a02 | 1271 | else { |
67060e37 | 1272 | n = &(*p)->rb_right; |
cc09e299 | 1273 | left = 0; |
c0324a02 | 1274 | } |
67060e37 JA |
1275 | |
1276 | p = n; | |
d9e7620e JA |
1277 | } |
1278 | ||
cc09e299 | 1279 | if (left) |
aa6f6a3d | 1280 | service_tree->left = &cfqq->rb_node; |
cc09e299 | 1281 | |
d9e7620e JA |
1282 | cfqq->rb_key = rb_key; |
1283 | rb_link_node(&cfqq->rb_node, parent, p); | |
aa6f6a3d CZ |
1284 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
1285 | service_tree->count++; | |
20359f27 | 1286 | if (add_front || !new_cfqq) |
dae739eb | 1287 | return; |
8184f93e | 1288 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1289 | } |
1290 | ||
a36e71f9 | 1291 | static struct cfq_queue * |
f2d1f0ae JA |
1292 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1293 | sector_t sector, struct rb_node **ret_parent, | |
1294 | struct rb_node ***rb_link) | |
a36e71f9 | 1295 | { |
a36e71f9 JA |
1296 | struct rb_node **p, *parent; |
1297 | struct cfq_queue *cfqq = NULL; | |
1298 | ||
1299 | parent = NULL; | |
1300 | p = &root->rb_node; | |
1301 | while (*p) { | |
1302 | struct rb_node **n; | |
1303 | ||
1304 | parent = *p; | |
1305 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1306 | ||
1307 | /* | |
1308 | * Sort strictly based on sector. Smallest to the left, | |
1309 | * largest to the right. | |
1310 | */ | |
2e46e8b2 | 1311 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1312 | n = &(*p)->rb_right; |
2e46e8b2 | 1313 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1314 | n = &(*p)->rb_left; |
1315 | else | |
1316 | break; | |
1317 | p = n; | |
3ac6c9f8 | 1318 | cfqq = NULL; |
a36e71f9 JA |
1319 | } |
1320 | ||
1321 | *ret_parent = parent; | |
1322 | if (rb_link) | |
1323 | *rb_link = p; | |
3ac6c9f8 | 1324 | return cfqq; |
a36e71f9 JA |
1325 | } |
1326 | ||
1327 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1328 | { | |
a36e71f9 JA |
1329 | struct rb_node **p, *parent; |
1330 | struct cfq_queue *__cfqq; | |
1331 | ||
f2d1f0ae JA |
1332 | if (cfqq->p_root) { |
1333 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1334 | cfqq->p_root = NULL; | |
1335 | } | |
a36e71f9 JA |
1336 | |
1337 | if (cfq_class_idle(cfqq)) | |
1338 | return; | |
1339 | if (!cfqq->next_rq) | |
1340 | return; | |
1341 | ||
f2d1f0ae | 1342 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1343 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1344 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1345 | if (!__cfqq) { |
1346 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1347 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1348 | } else | |
1349 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1350 | } |
1351 | ||
498d3aa2 JA |
1352 | /* |
1353 | * Update cfqq's position in the service tree. | |
1354 | */ | |
edd75ffd | 1355 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1356 | { |
6d048f53 JA |
1357 | /* |
1358 | * Resorting requires the cfqq to be on the RR list already. | |
1359 | */ | |
a36e71f9 | 1360 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1361 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1362 | cfq_prio_tree_add(cfqd, cfqq); |
1363 | } | |
6d048f53 JA |
1364 | } |
1365 | ||
1da177e4 LT |
1366 | /* |
1367 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1368 | * the pending list according to last request service |
1da177e4 | 1369 | */ |
febffd61 | 1370 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1371 | { |
7b679138 | 1372 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1373 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1374 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 1375 | cfqd->busy_queues++; |
ef8a41df SL |
1376 | if (cfq_cfqq_sync(cfqq)) |
1377 | cfqd->busy_sync_queues++; | |
1da177e4 | 1378 | |
edd75ffd | 1379 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1380 | } |
1381 | ||
498d3aa2 JA |
1382 | /* |
1383 | * Called when the cfqq no longer has requests pending, remove it from | |
1384 | * the service tree. | |
1385 | */ | |
febffd61 | 1386 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1387 | { |
7b679138 | 1388 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1389 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1390 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1391 | |
aa6f6a3d CZ |
1392 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1393 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1394 | cfqq->service_tree = NULL; | |
1395 | } | |
f2d1f0ae JA |
1396 | if (cfqq->p_root) { |
1397 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1398 | cfqq->p_root = NULL; | |
1399 | } | |
d9e7620e | 1400 | |
8184f93e | 1401 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1402 | BUG_ON(!cfqd->busy_queues); |
1403 | cfqd->busy_queues--; | |
ef8a41df SL |
1404 | if (cfq_cfqq_sync(cfqq)) |
1405 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
1406 | } |
1407 | ||
1408 | /* | |
1409 | * rb tree support functions | |
1410 | */ | |
febffd61 | 1411 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1412 | { |
5e705374 | 1413 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1414 | const int sync = rq_is_sync(rq); |
1da177e4 | 1415 | |
b4878f24 JA |
1416 | BUG_ON(!cfqq->queued[sync]); |
1417 | cfqq->queued[sync]--; | |
1da177e4 | 1418 | |
5e705374 | 1419 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1420 | |
f04a6424 VG |
1421 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1422 | /* | |
1423 | * Queue will be deleted from service tree when we actually | |
1424 | * expire it later. Right now just remove it from prio tree | |
1425 | * as it is empty. | |
1426 | */ | |
1427 | if (cfqq->p_root) { | |
1428 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1429 | cfqq->p_root = NULL; | |
1430 | } | |
1431 | } | |
1da177e4 LT |
1432 | } |
1433 | ||
5e705374 | 1434 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1435 | { |
5e705374 | 1436 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1437 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 1438 | struct request *prev; |
1da177e4 | 1439 | |
5380a101 | 1440 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 1441 | |
796d5116 | 1442 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
1443 | |
1444 | if (!cfq_cfqq_on_rr(cfqq)) | |
1445 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1446 | |
1447 | /* | |
1448 | * check if this request is a better next-serve candidate | |
1449 | */ | |
a36e71f9 | 1450 | prev = cfqq->next_rq; |
cf7c25cf | 1451 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1452 | |
1453 | /* | |
1454 | * adjust priority tree position, if ->next_rq changes | |
1455 | */ | |
1456 | if (prev != cfqq->next_rq) | |
1457 | cfq_prio_tree_add(cfqd, cfqq); | |
1458 | ||
5044eed4 | 1459 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1460 | } |
1461 | ||
febffd61 | 1462 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1463 | { |
5380a101 JA |
1464 | elv_rb_del(&cfqq->sort_list, rq); |
1465 | cfqq->queued[rq_is_sync(rq)]--; | |
0381411e | 1466 | cfq_blkiocg_update_io_remove_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 TH |
1467 | &blkio_policy_cfq, rq_data_dir(rq), |
1468 | rq_is_sync(rq)); | |
5e705374 | 1469 | cfq_add_rq_rb(rq); |
0381411e | 1470 | cfq_blkiocg_update_io_add_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 | 1471 | &blkio_policy_cfq, |
0381411e TH |
1472 | cfqg_to_blkg(cfqq->cfqd->serving_group), |
1473 | rq_data_dir(rq), rq_is_sync(rq)); | |
1da177e4 LT |
1474 | } |
1475 | ||
206dc69b JA |
1476 | static struct request * |
1477 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1478 | { |
206dc69b | 1479 | struct task_struct *tsk = current; |
c5869807 | 1480 | struct cfq_io_cq *cic; |
206dc69b | 1481 | struct cfq_queue *cfqq; |
1da177e4 | 1482 | |
4ac845a2 | 1483 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1484 | if (!cic) |
1485 | return NULL; | |
1486 | ||
1487 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
1488 | if (cfqq) { |
1489 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
1490 | ||
21183b07 | 1491 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 1492 | } |
1da177e4 | 1493 | |
1da177e4 LT |
1494 | return NULL; |
1495 | } | |
1496 | ||
165125e1 | 1497 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1498 | { |
22e2c507 | 1499 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 1500 | |
53c583d2 | 1501 | cfqd->rq_in_driver++; |
7b679138 | 1502 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 1503 | cfqd->rq_in_driver); |
25776e35 | 1504 | |
5b93629b | 1505 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
1506 | } |
1507 | ||
165125e1 | 1508 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1509 | { |
b4878f24 JA |
1510 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1511 | ||
53c583d2 CZ |
1512 | WARN_ON(!cfqd->rq_in_driver); |
1513 | cfqd->rq_in_driver--; | |
7b679138 | 1514 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 1515 | cfqd->rq_in_driver); |
1da177e4 LT |
1516 | } |
1517 | ||
b4878f24 | 1518 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 1519 | { |
5e705374 | 1520 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 1521 | |
5e705374 JA |
1522 | if (cfqq->next_rq == rq) |
1523 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 1524 | |
b4878f24 | 1525 | list_del_init(&rq->queuelist); |
5e705374 | 1526 | cfq_del_rq_rb(rq); |
374f84ac | 1527 | |
45333d5a | 1528 | cfqq->cfqd->rq_queued--; |
0381411e | 1529 | cfq_blkiocg_update_io_remove_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 TH |
1530 | &blkio_policy_cfq, rq_data_dir(rq), |
1531 | rq_is_sync(rq)); | |
65299a3b CH |
1532 | if (rq->cmd_flags & REQ_PRIO) { |
1533 | WARN_ON(!cfqq->prio_pending); | |
1534 | cfqq->prio_pending--; | |
b53d1ed7 | 1535 | } |
1da177e4 LT |
1536 | } |
1537 | ||
165125e1 JA |
1538 | static int cfq_merge(struct request_queue *q, struct request **req, |
1539 | struct bio *bio) | |
1da177e4 LT |
1540 | { |
1541 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1542 | struct request *__rq; | |
1da177e4 | 1543 | |
206dc69b | 1544 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 1545 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
1546 | *req = __rq; |
1547 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
1548 | } |
1549 | ||
1550 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
1551 | } |
1552 | ||
165125e1 | 1553 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 1554 | int type) |
1da177e4 | 1555 | { |
21183b07 | 1556 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 1557 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 1558 | |
5e705374 | 1559 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 1560 | } |
1da177e4 LT |
1561 | } |
1562 | ||
812d4026 DS |
1563 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
1564 | struct bio *bio) | |
1565 | { | |
0381411e | 1566 | cfq_blkiocg_update_io_merged_stats(cfqg_to_blkg(RQ_CFQG(req)), |
c1768268 TH |
1567 | &blkio_policy_cfq, bio_data_dir(bio), |
1568 | cfq_bio_sync(bio)); | |
812d4026 DS |
1569 | } |
1570 | ||
1da177e4 | 1571 | static void |
165125e1 | 1572 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
1573 | struct request *next) |
1574 | { | |
cf7c25cf | 1575 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
1576 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1577 | ||
22e2c507 JA |
1578 | /* |
1579 | * reposition in fifo if next is older than rq | |
1580 | */ | |
1581 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
30996f40 | 1582 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
22e2c507 | 1583 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
1584 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1585 | } | |
22e2c507 | 1586 | |
cf7c25cf CZ |
1587 | if (cfqq->next_rq == next) |
1588 | cfqq->next_rq = rq; | |
b4878f24 | 1589 | cfq_remove_request(next); |
0381411e | 1590 | cfq_blkiocg_update_io_merged_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 TH |
1591 | &blkio_policy_cfq, rq_data_dir(next), |
1592 | rq_is_sync(next)); | |
4a0b75c7 SL |
1593 | |
1594 | cfqq = RQ_CFQQ(next); | |
1595 | /* | |
1596 | * all requests of this queue are merged to other queues, delete it | |
1597 | * from the service tree. If it's the active_queue, | |
1598 | * cfq_dispatch_requests() will choose to expire it or do idle | |
1599 | */ | |
1600 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
1601 | cfqq != cfqd->active_queue) | |
1602 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
1603 | } |
1604 | ||
165125e1 | 1605 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
1606 | struct bio *bio) |
1607 | { | |
1608 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 1609 | struct cfq_io_cq *cic; |
da775265 | 1610 | struct cfq_queue *cfqq; |
da775265 JA |
1611 | |
1612 | /* | |
ec8acb69 | 1613 | * Disallow merge of a sync bio into an async request. |
da775265 | 1614 | */ |
91fac317 | 1615 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 1616 | return false; |
da775265 JA |
1617 | |
1618 | /* | |
f1a4f4d3 | 1619 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 1620 | * merge only if rq is queued there. |
f1a4f4d3 | 1621 | */ |
07c2bd37 TH |
1622 | cic = cfq_cic_lookup(cfqd, current->io_context); |
1623 | if (!cic) | |
1624 | return false; | |
719d3402 | 1625 | |
91fac317 | 1626 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 1627 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
1628 | } |
1629 | ||
812df48d DS |
1630 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1631 | { | |
1632 | del_timer(&cfqd->idle_slice_timer); | |
c1768268 TH |
1633 | cfq_blkiocg_update_idle_time_stats(cfqg_to_blkg(cfqq->cfqg), |
1634 | &blkio_policy_cfq); | |
812df48d DS |
1635 | } |
1636 | ||
febffd61 JA |
1637 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
1638 | struct cfq_queue *cfqq) | |
22e2c507 JA |
1639 | { |
1640 | if (cfqq) { | |
b1ffe737 DS |
1641 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", |
1642 | cfqd->serving_prio, cfqd->serving_type); | |
c1768268 TH |
1643 | cfq_blkiocg_update_avg_queue_size_stats(cfqg_to_blkg(cfqq->cfqg), |
1644 | &blkio_policy_cfq); | |
62a37f6b JT |
1645 | cfqq->slice_start = 0; |
1646 | cfqq->dispatch_start = jiffies; | |
1647 | cfqq->allocated_slice = 0; | |
1648 | cfqq->slice_end = 0; | |
1649 | cfqq->slice_dispatch = 0; | |
1650 | cfqq->nr_sectors = 0; | |
1651 | ||
1652 | cfq_clear_cfqq_wait_request(cfqq); | |
1653 | cfq_clear_cfqq_must_dispatch(cfqq); | |
1654 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
1655 | cfq_clear_cfqq_fifo_expire(cfqq); | |
1656 | cfq_mark_cfqq_slice_new(cfqq); | |
1657 | ||
1658 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
1659 | } |
1660 | ||
1661 | cfqd->active_queue = cfqq; | |
1662 | } | |
1663 | ||
7b14e3b5 JA |
1664 | /* |
1665 | * current cfqq expired its slice (or was too idle), select new one | |
1666 | */ | |
1667 | static void | |
1668 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 1669 | bool timed_out) |
7b14e3b5 | 1670 | { |
7b679138 JA |
1671 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
1672 | ||
7b14e3b5 | 1673 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 1674 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 1675 | |
7b14e3b5 | 1676 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 1677 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 1678 | |
ae54abed SL |
1679 | /* |
1680 | * If this cfqq is shared between multiple processes, check to | |
1681 | * make sure that those processes are still issuing I/Os within | |
1682 | * the mean seek distance. If not, it may be time to break the | |
1683 | * queues apart again. | |
1684 | */ | |
1685 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
1686 | cfq_mark_cfqq_split_coop(cfqq); | |
1687 | ||
7b14e3b5 | 1688 | /* |
6084cdda | 1689 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 1690 | */ |
c553f8e3 SL |
1691 | if (timed_out) { |
1692 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 1693 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
1694 | else |
1695 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
1696 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
1697 | } | |
7b14e3b5 | 1698 | |
e5ff082e | 1699 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 1700 | |
f04a6424 VG |
1701 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1702 | cfq_del_cfqq_rr(cfqd, cfqq); | |
1703 | ||
edd75ffd | 1704 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
1705 | |
1706 | if (cfqq == cfqd->active_queue) | |
1707 | cfqd->active_queue = NULL; | |
1708 | ||
1709 | if (cfqd->active_cic) { | |
11a3122f | 1710 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
1711 | cfqd->active_cic = NULL; |
1712 | } | |
7b14e3b5 JA |
1713 | } |
1714 | ||
e5ff082e | 1715 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
1716 | { |
1717 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1718 | ||
1719 | if (cfqq) | |
e5ff082e | 1720 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
1721 | } |
1722 | ||
498d3aa2 JA |
1723 | /* |
1724 | * Get next queue for service. Unless we have a queue preemption, | |
1725 | * we'll simply select the first cfqq in the service tree. | |
1726 | */ | |
6d048f53 | 1727 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 1728 | { |
c0324a02 | 1729 | struct cfq_rb_root *service_tree = |
cdb16e8f | 1730 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
65b32a57 | 1731 | cfqd->serving_type); |
d9e7620e | 1732 | |
f04a6424 VG |
1733 | if (!cfqd->rq_queued) |
1734 | return NULL; | |
1735 | ||
1fa8f6d6 VG |
1736 | /* There is nothing to dispatch */ |
1737 | if (!service_tree) | |
1738 | return NULL; | |
c0324a02 CZ |
1739 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
1740 | return NULL; | |
1741 | return cfq_rb_first(service_tree); | |
6d048f53 JA |
1742 | } |
1743 | ||
f04a6424 VG |
1744 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
1745 | { | |
25fb5169 | 1746 | struct cfq_group *cfqg; |
f04a6424 VG |
1747 | struct cfq_queue *cfqq; |
1748 | int i, j; | |
1749 | struct cfq_rb_root *st; | |
1750 | ||
1751 | if (!cfqd->rq_queued) | |
1752 | return NULL; | |
1753 | ||
25fb5169 VG |
1754 | cfqg = cfq_get_next_cfqg(cfqd); |
1755 | if (!cfqg) | |
1756 | return NULL; | |
1757 | ||
f04a6424 VG |
1758 | for_each_cfqg_st(cfqg, i, j, st) |
1759 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
1760 | return cfqq; | |
1761 | return NULL; | |
1762 | } | |
1763 | ||
498d3aa2 JA |
1764 | /* |
1765 | * Get and set a new active queue for service. | |
1766 | */ | |
a36e71f9 JA |
1767 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
1768 | struct cfq_queue *cfqq) | |
6d048f53 | 1769 | { |
e00ef799 | 1770 | if (!cfqq) |
a36e71f9 | 1771 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 1772 | |
22e2c507 | 1773 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 1774 | return cfqq; |
22e2c507 JA |
1775 | } |
1776 | ||
d9e7620e JA |
1777 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
1778 | struct request *rq) | |
1779 | { | |
83096ebf TH |
1780 | if (blk_rq_pos(rq) >= cfqd->last_position) |
1781 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 1782 | else |
83096ebf | 1783 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
1784 | } |
1785 | ||
b2c18e1e | 1786 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 1787 | struct request *rq) |
6d048f53 | 1788 | { |
e9ce335d | 1789 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
1790 | } |
1791 | ||
a36e71f9 JA |
1792 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
1793 | struct cfq_queue *cur_cfqq) | |
1794 | { | |
f2d1f0ae | 1795 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
1796 | struct rb_node *parent, *node; |
1797 | struct cfq_queue *__cfqq; | |
1798 | sector_t sector = cfqd->last_position; | |
1799 | ||
1800 | if (RB_EMPTY_ROOT(root)) | |
1801 | return NULL; | |
1802 | ||
1803 | /* | |
1804 | * First, if we find a request starting at the end of the last | |
1805 | * request, choose it. | |
1806 | */ | |
f2d1f0ae | 1807 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
1808 | if (__cfqq) |
1809 | return __cfqq; | |
1810 | ||
1811 | /* | |
1812 | * If the exact sector wasn't found, the parent of the NULL leaf | |
1813 | * will contain the closest sector. | |
1814 | */ | |
1815 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 1816 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1817 | return __cfqq; |
1818 | ||
2e46e8b2 | 1819 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
1820 | node = rb_next(&__cfqq->p_node); |
1821 | else | |
1822 | node = rb_prev(&__cfqq->p_node); | |
1823 | if (!node) | |
1824 | return NULL; | |
1825 | ||
1826 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 1827 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1828 | return __cfqq; |
1829 | ||
1830 | return NULL; | |
1831 | } | |
1832 | ||
1833 | /* | |
1834 | * cfqd - obvious | |
1835 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
1836 | * closely cooperating with itself. | |
1837 | * | |
1838 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
1839 | * one request, and that cfqd->last_position reflects a position on the disk | |
1840 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
1841 | * assumption. | |
1842 | */ | |
1843 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 1844 | struct cfq_queue *cur_cfqq) |
6d048f53 | 1845 | { |
a36e71f9 JA |
1846 | struct cfq_queue *cfqq; |
1847 | ||
39c01b21 DS |
1848 | if (cfq_class_idle(cur_cfqq)) |
1849 | return NULL; | |
e6c5bc73 JM |
1850 | if (!cfq_cfqq_sync(cur_cfqq)) |
1851 | return NULL; | |
1852 | if (CFQQ_SEEKY(cur_cfqq)) | |
1853 | return NULL; | |
1854 | ||
b9d8f4c7 GJ |
1855 | /* |
1856 | * Don't search priority tree if it's the only queue in the group. | |
1857 | */ | |
1858 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
1859 | return NULL; | |
1860 | ||
6d048f53 | 1861 | /* |
d9e7620e JA |
1862 | * We should notice if some of the queues are cooperating, eg |
1863 | * working closely on the same area of the disk. In that case, | |
1864 | * we can group them together and don't waste time idling. | |
6d048f53 | 1865 | */ |
a36e71f9 JA |
1866 | cfqq = cfqq_close(cfqd, cur_cfqq); |
1867 | if (!cfqq) | |
1868 | return NULL; | |
1869 | ||
8682e1f1 VG |
1870 | /* If new queue belongs to different cfq_group, don't choose it */ |
1871 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
1872 | return NULL; | |
1873 | ||
df5fe3e8 JM |
1874 | /* |
1875 | * It only makes sense to merge sync queues. | |
1876 | */ | |
1877 | if (!cfq_cfqq_sync(cfqq)) | |
1878 | return NULL; | |
e6c5bc73 JM |
1879 | if (CFQQ_SEEKY(cfqq)) |
1880 | return NULL; | |
df5fe3e8 | 1881 | |
c0324a02 CZ |
1882 | /* |
1883 | * Do not merge queues of different priority classes | |
1884 | */ | |
1885 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
1886 | return NULL; | |
1887 | ||
a36e71f9 | 1888 | return cfqq; |
6d048f53 JA |
1889 | } |
1890 | ||
a6d44e98 CZ |
1891 | /* |
1892 | * Determine whether we should enforce idle window for this queue. | |
1893 | */ | |
1894 | ||
1895 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1896 | { | |
1897 | enum wl_prio_t prio = cfqq_prio(cfqq); | |
718eee05 | 1898 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
a6d44e98 | 1899 | |
f04a6424 VG |
1900 | BUG_ON(!service_tree); |
1901 | BUG_ON(!service_tree->count); | |
1902 | ||
b6508c16 VG |
1903 | if (!cfqd->cfq_slice_idle) |
1904 | return false; | |
1905 | ||
a6d44e98 CZ |
1906 | /* We never do for idle class queues. */ |
1907 | if (prio == IDLE_WORKLOAD) | |
1908 | return false; | |
1909 | ||
1910 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
1911 | if (cfq_cfqq_idle_window(cfqq) && |
1912 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
1913 | return true; |
1914 | ||
1915 | /* | |
1916 | * Otherwise, we do only if they are the last ones | |
1917 | * in their service tree. | |
1918 | */ | |
f5f2b6ce SL |
1919 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq) && |
1920 | !cfq_io_thinktime_big(cfqd, &service_tree->ttime, false)) | |
c1e44756 | 1921 | return true; |
b1ffe737 DS |
1922 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", |
1923 | service_tree->count); | |
c1e44756 | 1924 | return false; |
a6d44e98 CZ |
1925 | } |
1926 | ||
6d048f53 | 1927 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 1928 | { |
1792669c | 1929 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 1930 | struct cfq_io_cq *cic; |
80bdf0c7 | 1931 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 1932 | |
a68bbddb | 1933 | /* |
f7d7b7a7 JA |
1934 | * SSD device without seek penalty, disable idling. But only do so |
1935 | * for devices that support queuing, otherwise we still have a problem | |
1936 | * with sync vs async workloads. | |
a68bbddb | 1937 | */ |
f7d7b7a7 | 1938 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
1939 | return; |
1940 | ||
dd67d051 | 1941 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 1942 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
1943 | |
1944 | /* | |
1945 | * idle is disabled, either manually or by past process history | |
1946 | */ | |
80bdf0c7 VG |
1947 | if (!cfq_should_idle(cfqd, cfqq)) { |
1948 | /* no queue idling. Check for group idling */ | |
1949 | if (cfqd->cfq_group_idle) | |
1950 | group_idle = cfqd->cfq_group_idle; | |
1951 | else | |
1952 | return; | |
1953 | } | |
6d048f53 | 1954 | |
7b679138 | 1955 | /* |
8e550632 | 1956 | * still active requests from this queue, don't idle |
7b679138 | 1957 | */ |
8e550632 | 1958 | if (cfqq->dispatched) |
7b679138 JA |
1959 | return; |
1960 | ||
22e2c507 JA |
1961 | /* |
1962 | * task has exited, don't wait | |
1963 | */ | |
206dc69b | 1964 | cic = cfqd->active_cic; |
c5869807 | 1965 | if (!cic || !atomic_read(&cic->icq.ioc->nr_tasks)) |
6d048f53 JA |
1966 | return; |
1967 | ||
355b659c CZ |
1968 | /* |
1969 | * If our average think time is larger than the remaining time | |
1970 | * slice, then don't idle. This avoids overrunning the allotted | |
1971 | * time slice. | |
1972 | */ | |
383cd721 SL |
1973 | if (sample_valid(cic->ttime.ttime_samples) && |
1974 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 1975 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 1976 | cic->ttime.ttime_mean); |
355b659c | 1977 | return; |
b1ffe737 | 1978 | } |
355b659c | 1979 | |
80bdf0c7 VG |
1980 | /* There are other queues in the group, don't do group idle */ |
1981 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
1982 | return; | |
1983 | ||
3b18152c | 1984 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 1985 | |
80bdf0c7 VG |
1986 | if (group_idle) |
1987 | sl = cfqd->cfq_group_idle; | |
1988 | else | |
1989 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 1990 | |
7b14e3b5 | 1991 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
c1768268 TH |
1992 | cfq_blkiocg_update_set_idle_time_stats(cfqg_to_blkg(cfqq->cfqg), |
1993 | &blkio_policy_cfq); | |
80bdf0c7 VG |
1994 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
1995 | group_idle ? 1 : 0); | |
1da177e4 LT |
1996 | } |
1997 | ||
498d3aa2 JA |
1998 | /* |
1999 | * Move request from internal lists to the request queue dispatch list. | |
2000 | */ | |
165125e1 | 2001 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2002 | { |
3ed9a296 | 2003 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2004 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2005 | |
7b679138 JA |
2006 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2007 | ||
06d21886 | 2008 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2009 | cfq_remove_request(rq); |
6d048f53 | 2010 | cfqq->dispatched++; |
80bdf0c7 | 2011 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2012 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2013 | |
53c583d2 | 2014 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2015 | cfqq->nr_sectors += blk_rq_sectors(rq); |
0381411e | 2016 | cfq_blkiocg_update_dispatch_stats(cfqg_to_blkg(cfqq->cfqg), |
c1768268 TH |
2017 | &blkio_policy_cfq, blk_rq_bytes(rq), |
2018 | rq_data_dir(rq), rq_is_sync(rq)); | |
1da177e4 LT |
2019 | } |
2020 | ||
2021 | /* | |
2022 | * return expired entry, or NULL to just start from scratch in rbtree | |
2023 | */ | |
febffd61 | 2024 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 2025 | { |
30996f40 | 2026 | struct request *rq = NULL; |
1da177e4 | 2027 | |
3b18152c | 2028 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 2029 | return NULL; |
cb887411 JA |
2030 | |
2031 | cfq_mark_cfqq_fifo_expire(cfqq); | |
2032 | ||
89850f7e JA |
2033 | if (list_empty(&cfqq->fifo)) |
2034 | return NULL; | |
1da177e4 | 2035 | |
89850f7e | 2036 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 2037 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 2038 | rq = NULL; |
1da177e4 | 2039 | |
30996f40 | 2040 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 2041 | return rq; |
1da177e4 LT |
2042 | } |
2043 | ||
22e2c507 JA |
2044 | static inline int |
2045 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2046 | { | |
2047 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2048 | |
22e2c507 | 2049 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2050 | |
b9f8ce05 | 2051 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
2052 | } |
2053 | ||
df5fe3e8 JM |
2054 | /* |
2055 | * Must be called with the queue_lock held. | |
2056 | */ | |
2057 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2058 | { | |
2059 | int process_refs, io_refs; | |
2060 | ||
2061 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 2062 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
2063 | BUG_ON(process_refs < 0); |
2064 | return process_refs; | |
2065 | } | |
2066 | ||
2067 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2068 | { | |
e6c5bc73 | 2069 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2070 | struct cfq_queue *__cfqq; |
2071 | ||
c10b61f0 JM |
2072 | /* |
2073 | * If there are no process references on the new_cfqq, then it is | |
2074 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2075 | * chain may have dropped their last reference (not just their | |
2076 | * last process reference). | |
2077 | */ | |
2078 | if (!cfqq_process_refs(new_cfqq)) | |
2079 | return; | |
2080 | ||
df5fe3e8 JM |
2081 | /* Avoid a circular list and skip interim queue merges */ |
2082 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2083 | if (__cfqq == cfqq) | |
2084 | return; | |
2085 | new_cfqq = __cfqq; | |
2086 | } | |
2087 | ||
2088 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2089 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2090 | /* |
2091 | * If the process for the cfqq has gone away, there is no | |
2092 | * sense in merging the queues. | |
2093 | */ | |
c10b61f0 | 2094 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2095 | return; |
2096 | ||
e6c5bc73 JM |
2097 | /* |
2098 | * Merge in the direction of the lesser amount of work. | |
2099 | */ | |
e6c5bc73 JM |
2100 | if (new_process_refs >= process_refs) { |
2101 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2102 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2103 | } else { |
2104 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2105 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2106 | } |
df5fe3e8 JM |
2107 | } |
2108 | ||
cdb16e8f | 2109 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
65b32a57 | 2110 | struct cfq_group *cfqg, enum wl_prio_t prio) |
718eee05 CZ |
2111 | { |
2112 | struct cfq_queue *queue; | |
2113 | int i; | |
2114 | bool key_valid = false; | |
2115 | unsigned long lowest_key = 0; | |
2116 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2117 | ||
65b32a57 VG |
2118 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2119 | /* select the one with lowest rb_key */ | |
2120 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); | |
718eee05 CZ |
2121 | if (queue && |
2122 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2123 | lowest_key = queue->rb_key; | |
2124 | cur_best = i; | |
2125 | key_valid = true; | |
2126 | } | |
2127 | } | |
2128 | ||
2129 | return cur_best; | |
2130 | } | |
2131 | ||
cdb16e8f | 2132 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
718eee05 | 2133 | { |
718eee05 CZ |
2134 | unsigned slice; |
2135 | unsigned count; | |
cdb16e8f | 2136 | struct cfq_rb_root *st; |
58ff82f3 | 2137 | unsigned group_slice; |
e4ea0c16 | 2138 | enum wl_prio_t original_prio = cfqd->serving_prio; |
1fa8f6d6 | 2139 | |
718eee05 | 2140 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2141 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
718eee05 | 2142 | cfqd->serving_prio = RT_WORKLOAD; |
58ff82f3 | 2143 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
718eee05 CZ |
2144 | cfqd->serving_prio = BE_WORKLOAD; |
2145 | else { | |
2146 | cfqd->serving_prio = IDLE_WORKLOAD; | |
2147 | cfqd->workload_expires = jiffies + 1; | |
2148 | return; | |
2149 | } | |
2150 | ||
e4ea0c16 SL |
2151 | if (original_prio != cfqd->serving_prio) |
2152 | goto new_workload; | |
2153 | ||
718eee05 CZ |
2154 | /* |
2155 | * For RT and BE, we have to choose also the type | |
2156 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2157 | * expiration time | |
2158 | */ | |
65b32a57 | 2159 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
cdb16e8f | 2160 | count = st->count; |
718eee05 CZ |
2161 | |
2162 | /* | |
65b32a57 | 2163 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2164 | */ |
65b32a57 | 2165 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2166 | return; |
2167 | ||
e4ea0c16 | 2168 | new_workload: |
718eee05 CZ |
2169 | /* otherwise select new workload type */ |
2170 | cfqd->serving_type = | |
65b32a57 VG |
2171 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); |
2172 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); | |
cdb16e8f | 2173 | count = st->count; |
718eee05 CZ |
2174 | |
2175 | /* | |
2176 | * the workload slice is computed as a fraction of target latency | |
2177 | * proportional to the number of queues in that workload, over | |
2178 | * all the queues in the same priority class | |
2179 | */ | |
58ff82f3 VG |
2180 | group_slice = cfq_group_slice(cfqd, cfqg); |
2181 | ||
2182 | slice = group_slice * count / | |
2183 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], | |
2184 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); | |
718eee05 | 2185 | |
f26bd1f0 VG |
2186 | if (cfqd->serving_type == ASYNC_WORKLOAD) { |
2187 | unsigned int tmp; | |
2188 | ||
2189 | /* | |
2190 | * Async queues are currently system wide. Just taking | |
2191 | * proportion of queues with-in same group will lead to higher | |
2192 | * async ratio system wide as generally root group is going | |
2193 | * to have higher weight. A more accurate thing would be to | |
2194 | * calculate system wide asnc/sync ratio. | |
2195 | */ | |
2196 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); | |
2197 | tmp = tmp/cfqd->busy_queues; | |
2198 | slice = min_t(unsigned, slice, tmp); | |
2199 | ||
718eee05 CZ |
2200 | /* async workload slice is scaled down according to |
2201 | * the sync/async slice ratio. */ | |
2202 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2203 | } else |
718eee05 CZ |
2204 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2205 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2206 | ||
2207 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2208 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
2209 | cfqd->workload_expires = jiffies + slice; |
2210 | } | |
2211 | ||
1fa8f6d6 VG |
2212 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2213 | { | |
2214 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2215 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2216 | |
2217 | if (RB_EMPTY_ROOT(&st->rb)) | |
2218 | return NULL; | |
25bc6b07 | 2219 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
2220 | update_min_vdisktime(st); |
2221 | return cfqg; | |
1fa8f6d6 VG |
2222 | } |
2223 | ||
cdb16e8f VG |
2224 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2225 | { | |
1fa8f6d6 VG |
2226 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2227 | ||
2228 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2229 | |
2230 | /* Restore the workload type data */ | |
2231 | if (cfqg->saved_workload_slice) { | |
2232 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; | |
2233 | cfqd->serving_type = cfqg->saved_workload; | |
2234 | cfqd->serving_prio = cfqg->saved_serving_prio; | |
66ae2919 GJ |
2235 | } else |
2236 | cfqd->workload_expires = jiffies - 1; | |
2237 | ||
1fa8f6d6 | 2238 | choose_service_tree(cfqd, cfqg); |
cdb16e8f VG |
2239 | } |
2240 | ||
22e2c507 | 2241 | /* |
498d3aa2 JA |
2242 | * Select a queue for service. If we have a current active queue, |
2243 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2244 | */ |
1b5ed5e1 | 2245 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2246 | { |
a36e71f9 | 2247 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2248 | |
22e2c507 JA |
2249 | cfqq = cfqd->active_queue; |
2250 | if (!cfqq) | |
2251 | goto new_queue; | |
1da177e4 | 2252 | |
f04a6424 VG |
2253 | if (!cfqd->rq_queued) |
2254 | return NULL; | |
c244bb50 VG |
2255 | |
2256 | /* | |
2257 | * We were waiting for group to get backlogged. Expire the queue | |
2258 | */ | |
2259 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2260 | goto expire; | |
2261 | ||
22e2c507 | 2262 | /* |
6d048f53 | 2263 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2264 | */ |
7667aa06 VG |
2265 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2266 | /* | |
2267 | * If slice had not expired at the completion of last request | |
2268 | * we might not have turned on wait_busy flag. Don't expire | |
2269 | * the queue yet. Allow the group to get backlogged. | |
2270 | * | |
2271 | * The very fact that we have used the slice, that means we | |
2272 | * have been idling all along on this queue and it should be | |
2273 | * ok to wait for this request to complete. | |
2274 | */ | |
82bbbf28 VG |
2275 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2276 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2277 | cfqq = NULL; | |
7667aa06 | 2278 | goto keep_queue; |
82bbbf28 | 2279 | } else |
80bdf0c7 | 2280 | goto check_group_idle; |
7667aa06 | 2281 | } |
1da177e4 | 2282 | |
22e2c507 | 2283 | /* |
6d048f53 JA |
2284 | * The active queue has requests and isn't expired, allow it to |
2285 | * dispatch. | |
22e2c507 | 2286 | */ |
dd67d051 | 2287 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2288 | goto keep_queue; |
6d048f53 | 2289 | |
a36e71f9 JA |
2290 | /* |
2291 | * If another queue has a request waiting within our mean seek | |
2292 | * distance, let it run. The expire code will check for close | |
2293 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2294 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2295 | */ |
b3b6d040 | 2296 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2297 | if (new_cfqq) { |
2298 | if (!cfqq->new_cfqq) | |
2299 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2300 | goto expire; |
df5fe3e8 | 2301 | } |
a36e71f9 | 2302 | |
6d048f53 JA |
2303 | /* |
2304 | * No requests pending. If the active queue still has requests in | |
2305 | * flight or is idling for a new request, allow either of these | |
2306 | * conditions to happen (or time out) before selecting a new queue. | |
2307 | */ | |
80bdf0c7 VG |
2308 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2309 | cfqq = NULL; | |
2310 | goto keep_queue; | |
2311 | } | |
2312 | ||
8e1ac665 SL |
2313 | /* |
2314 | * This is a deep seek queue, but the device is much faster than | |
2315 | * the queue can deliver, don't idle | |
2316 | **/ | |
2317 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
2318 | (cfq_cfqq_slice_new(cfqq) || | |
2319 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
2320 | cfq_clear_cfqq_deep(cfqq); | |
2321 | cfq_clear_cfqq_idle_window(cfqq); | |
2322 | } | |
2323 | ||
80bdf0c7 VG |
2324 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2325 | cfqq = NULL; | |
2326 | goto keep_queue; | |
2327 | } | |
2328 | ||
2329 | /* | |
2330 | * If group idle is enabled and there are requests dispatched from | |
2331 | * this group, wait for requests to complete. | |
2332 | */ | |
2333 | check_group_idle: | |
7700fc4f SL |
2334 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
2335 | cfqq->cfqg->dispatched && | |
2336 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
2337 | cfqq = NULL; |
2338 | goto keep_queue; | |
22e2c507 JA |
2339 | } |
2340 | ||
3b18152c | 2341 | expire: |
e5ff082e | 2342 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2343 | new_queue: |
718eee05 CZ |
2344 | /* |
2345 | * Current queue expired. Check if we have to switch to a new | |
2346 | * service tree | |
2347 | */ | |
2348 | if (!new_cfqq) | |
cdb16e8f | 2349 | cfq_choose_cfqg(cfqd); |
718eee05 | 2350 | |
a36e71f9 | 2351 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2352 | keep_queue: |
3b18152c | 2353 | return cfqq; |
22e2c507 JA |
2354 | } |
2355 | ||
febffd61 | 2356 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2357 | { |
2358 | int dispatched = 0; | |
2359 | ||
2360 | while (cfqq->next_rq) { | |
2361 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2362 | dispatched++; | |
2363 | } | |
2364 | ||
2365 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2366 | |
2367 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 2368 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
2369 | return dispatched; |
2370 | } | |
2371 | ||
498d3aa2 JA |
2372 | /* |
2373 | * Drain our current requests. Used for barriers and when switching | |
2374 | * io schedulers on-the-fly. | |
2375 | */ | |
d9e7620e | 2376 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2377 | { |
0871714e | 2378 | struct cfq_queue *cfqq; |
d9e7620e | 2379 | int dispatched = 0; |
cdb16e8f | 2380 | |
3440c49f | 2381 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 2382 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
2383 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2384 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 2385 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 2386 | } |
1b5ed5e1 | 2387 | |
1b5ed5e1 TH |
2388 | BUG_ON(cfqd->busy_queues); |
2389 | ||
6923715a | 2390 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2391 | return dispatched; |
2392 | } | |
2393 | ||
abc3c744 SL |
2394 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2395 | struct cfq_queue *cfqq) | |
2396 | { | |
2397 | /* the queue hasn't finished any request, can't estimate */ | |
2398 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 2399 | return true; |
abc3c744 SL |
2400 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
2401 | cfqq->slice_end)) | |
c1e44756 | 2402 | return true; |
abc3c744 | 2403 | |
c1e44756 | 2404 | return false; |
abc3c744 SL |
2405 | } |
2406 | ||
0b182d61 | 2407 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2408 | { |
2f5cb738 | 2409 | unsigned int max_dispatch; |
22e2c507 | 2410 | |
5ad531db JA |
2411 | /* |
2412 | * Drain async requests before we start sync IO | |
2413 | */ | |
53c583d2 | 2414 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 2415 | return false; |
5ad531db | 2416 | |
2f5cb738 JA |
2417 | /* |
2418 | * If this is an async queue and we have sync IO in flight, let it wait | |
2419 | */ | |
53c583d2 | 2420 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 2421 | return false; |
2f5cb738 | 2422 | |
abc3c744 | 2423 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
2424 | if (cfq_class_idle(cfqq)) |
2425 | max_dispatch = 1; | |
b4878f24 | 2426 | |
2f5cb738 JA |
2427 | /* |
2428 | * Does this cfqq already have too much IO in flight? | |
2429 | */ | |
2430 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 2431 | bool promote_sync = false; |
2f5cb738 JA |
2432 | /* |
2433 | * idle queue must always only have a single IO in flight | |
2434 | */ | |
3ed9a296 | 2435 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2436 | return false; |
3ed9a296 | 2437 | |
ef8a41df | 2438 | /* |
c4ade94f LS |
2439 | * If there is only one sync queue |
2440 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
2441 | * queue no dispatch limit. The reason is a sync queue can |
2442 | * preempt async queue, limiting the sync queue doesn't make | |
2443 | * sense. This is useful for aiostress test. | |
2444 | */ | |
c4ade94f LS |
2445 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
2446 | promote_sync = true; | |
ef8a41df | 2447 | |
2f5cb738 JA |
2448 | /* |
2449 | * We have other queues, don't allow more IO from this one | |
2450 | */ | |
ef8a41df SL |
2451 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
2452 | !promote_sync) | |
0b182d61 | 2453 | return false; |
9ede209e | 2454 | |
365722bb | 2455 | /* |
474b18cc | 2456 | * Sole queue user, no limit |
365722bb | 2457 | */ |
ef8a41df | 2458 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
2459 | max_dispatch = -1; |
2460 | else | |
2461 | /* | |
2462 | * Normally we start throttling cfqq when cfq_quantum/2 | |
2463 | * requests have been dispatched. But we can drive | |
2464 | * deeper queue depths at the beginning of slice | |
2465 | * subjected to upper limit of cfq_quantum. | |
2466 | * */ | |
2467 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
2468 | } |
2469 | ||
2470 | /* | |
2471 | * Async queues must wait a bit before being allowed dispatch. | |
2472 | * We also ramp up the dispatch depth gradually for async IO, | |
2473 | * based on the last sync IO we serviced | |
2474 | */ | |
963b72fc | 2475 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 2476 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 2477 | unsigned int depth; |
365722bb | 2478 | |
61f0c1dc | 2479 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2480 | if (!depth && !cfqq->dispatched) |
2481 | depth = 1; | |
8e296755 JA |
2482 | if (depth < max_dispatch) |
2483 | max_dispatch = depth; | |
2f5cb738 | 2484 | } |
3ed9a296 | 2485 | |
0b182d61 JA |
2486 | /* |
2487 | * If we're below the current max, allow a dispatch | |
2488 | */ | |
2489 | return cfqq->dispatched < max_dispatch; | |
2490 | } | |
2491 | ||
2492 | /* | |
2493 | * Dispatch a request from cfqq, moving them to the request queue | |
2494 | * dispatch list. | |
2495 | */ | |
2496 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2497 | { | |
2498 | struct request *rq; | |
2499 | ||
2500 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
2501 | ||
2502 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
2503 | return false; | |
2504 | ||
2505 | /* | |
2506 | * follow expired path, else get first next available | |
2507 | */ | |
2508 | rq = cfq_check_fifo(cfqq); | |
2509 | if (!rq) | |
2510 | rq = cfqq->next_rq; | |
2511 | ||
2512 | /* | |
2513 | * insert request into driver dispatch list | |
2514 | */ | |
2515 | cfq_dispatch_insert(cfqd->queue, rq); | |
2516 | ||
2517 | if (!cfqd->active_cic) { | |
c5869807 | 2518 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 2519 | |
c5869807 | 2520 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
2521 | cfqd->active_cic = cic; |
2522 | } | |
2523 | ||
2524 | return true; | |
2525 | } | |
2526 | ||
2527 | /* | |
2528 | * Find the cfqq that we need to service and move a request from that to the | |
2529 | * dispatch list | |
2530 | */ | |
2531 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
2532 | { | |
2533 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2534 | struct cfq_queue *cfqq; | |
2535 | ||
2536 | if (!cfqd->busy_queues) | |
2537 | return 0; | |
2538 | ||
2539 | if (unlikely(force)) | |
2540 | return cfq_forced_dispatch(cfqd); | |
2541 | ||
2542 | cfqq = cfq_select_queue(cfqd); | |
2543 | if (!cfqq) | |
8e296755 JA |
2544 | return 0; |
2545 | ||
2f5cb738 | 2546 | /* |
0b182d61 | 2547 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 2548 | */ |
0b182d61 JA |
2549 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2550 | return 0; | |
2551 | ||
2f5cb738 | 2552 | cfqq->slice_dispatch++; |
b029195d | 2553 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 2554 | |
2f5cb738 JA |
2555 | /* |
2556 | * expire an async queue immediately if it has used up its slice. idle | |
2557 | * queue always expire after 1 dispatch round. | |
2558 | */ | |
2559 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
2560 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
2561 | cfq_class_idle(cfqq))) { | |
2562 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 2563 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
2564 | } |
2565 | ||
b217a903 | 2566 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 2567 | return 1; |
1da177e4 LT |
2568 | } |
2569 | ||
1da177e4 | 2570 | /* |
5e705374 JA |
2571 | * task holds one reference to the queue, dropped when task exits. each rq |
2572 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 2573 | * |
b1c35769 | 2574 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
2575 | * queue lock must be held here. |
2576 | */ | |
2577 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
2578 | { | |
22e2c507 | 2579 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 2580 | struct cfq_group *cfqg; |
22e2c507 | 2581 | |
30d7b944 | 2582 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 2583 | |
30d7b944 SL |
2584 | cfqq->ref--; |
2585 | if (cfqq->ref) | |
1da177e4 LT |
2586 | return; |
2587 | ||
7b679138 | 2588 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 2589 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 2590 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 2591 | cfqg = cfqq->cfqg; |
1da177e4 | 2592 | |
28f95cbc | 2593 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 2594 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2595 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2596 | } |
22e2c507 | 2597 | |
f04a6424 | 2598 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 2599 | kmem_cache_free(cfq_pool, cfqq); |
1adaf3dd | 2600 | blkg_put(cfqg_to_blkg(cfqg)); |
1da177e4 LT |
2601 | } |
2602 | ||
d02a2c07 | 2603 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 2604 | { |
df5fe3e8 JM |
2605 | struct cfq_queue *__cfqq, *next; |
2606 | ||
df5fe3e8 JM |
2607 | /* |
2608 | * If this queue was scheduled to merge with another queue, be | |
2609 | * sure to drop the reference taken on that queue (and others in | |
2610 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
2611 | */ | |
2612 | __cfqq = cfqq->new_cfqq; | |
2613 | while (__cfqq) { | |
2614 | if (__cfqq == cfqq) { | |
2615 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
2616 | break; | |
2617 | } | |
2618 | next = __cfqq->new_cfqq; | |
2619 | cfq_put_queue(__cfqq); | |
2620 | __cfqq = next; | |
2621 | } | |
d02a2c07 SL |
2622 | } |
2623 | ||
2624 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2625 | { | |
2626 | if (unlikely(cfqq == cfqd->active_queue)) { | |
2627 | __cfq_slice_expired(cfqd, cfqq, 0); | |
2628 | cfq_schedule_dispatch(cfqd); | |
2629 | } | |
2630 | ||
2631 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 2632 | |
89850f7e JA |
2633 | cfq_put_queue(cfqq); |
2634 | } | |
22e2c507 | 2635 | |
9b84cacd TH |
2636 | static void cfq_init_icq(struct io_cq *icq) |
2637 | { | |
2638 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
2639 | ||
2640 | cic->ttime.last_end_request = jiffies; | |
2641 | } | |
2642 | ||
c5869807 | 2643 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 2644 | { |
c5869807 | 2645 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 2646 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 2647 | |
ff6657c6 JA |
2648 | if (cic->cfqq[BLK_RW_ASYNC]) { |
2649 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
2650 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
2651 | } |
2652 | ||
ff6657c6 JA |
2653 | if (cic->cfqq[BLK_RW_SYNC]) { |
2654 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
2655 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 2656 | } |
89850f7e JA |
2657 | } |
2658 | ||
fd0928df | 2659 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
2660 | { |
2661 | struct task_struct *tsk = current; | |
2662 | int ioprio_class; | |
2663 | ||
3b18152c | 2664 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
2665 | return; |
2666 | ||
fd0928df | 2667 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 2668 | switch (ioprio_class) { |
fe094d98 JA |
2669 | default: |
2670 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
2671 | case IOPRIO_CLASS_NONE: | |
2672 | /* | |
6d63c275 | 2673 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
2674 | */ |
2675 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 2676 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
2677 | break; |
2678 | case IOPRIO_CLASS_RT: | |
2679 | cfqq->ioprio = task_ioprio(ioc); | |
2680 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
2681 | break; | |
2682 | case IOPRIO_CLASS_BE: | |
2683 | cfqq->ioprio = task_ioprio(ioc); | |
2684 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
2685 | break; | |
2686 | case IOPRIO_CLASS_IDLE: | |
2687 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
2688 | cfqq->ioprio = 7; | |
2689 | cfq_clear_cfqq_idle_window(cfqq); | |
2690 | break; | |
22e2c507 JA |
2691 | } |
2692 | ||
2693 | /* | |
2694 | * keep track of original prio settings in case we have to temporarily | |
2695 | * elevate the priority of this queue | |
2696 | */ | |
2697 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 2698 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
2699 | } |
2700 | ||
c5869807 | 2701 | static void changed_ioprio(struct cfq_io_cq *cic) |
22e2c507 | 2702 | { |
bca4b914 | 2703 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 2704 | struct cfq_queue *cfqq; |
35e6077c | 2705 | |
caaa5f9f JA |
2706 | if (unlikely(!cfqd)) |
2707 | return; | |
2708 | ||
ff6657c6 | 2709 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
2710 | if (cfqq) { |
2711 | struct cfq_queue *new_cfqq; | |
c5869807 | 2712 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->icq.ioc, |
ff6657c6 | 2713 | GFP_ATOMIC); |
caaa5f9f | 2714 | if (new_cfqq) { |
ff6657c6 | 2715 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
2716 | cfq_put_queue(cfqq); |
2717 | } | |
22e2c507 | 2718 | } |
caaa5f9f | 2719 | |
ff6657c6 | 2720 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
2721 | if (cfqq) |
2722 | cfq_mark_cfqq_prio_changed(cfqq); | |
22e2c507 JA |
2723 | } |
2724 | ||
d5036d77 | 2725 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2726 | pid_t pid, bool is_sync) |
d5036d77 JA |
2727 | { |
2728 | RB_CLEAR_NODE(&cfqq->rb_node); | |
2729 | RB_CLEAR_NODE(&cfqq->p_node); | |
2730 | INIT_LIST_HEAD(&cfqq->fifo); | |
2731 | ||
30d7b944 | 2732 | cfqq->ref = 0; |
d5036d77 JA |
2733 | cfqq->cfqd = cfqd; |
2734 | ||
2735 | cfq_mark_cfqq_prio_changed(cfqq); | |
2736 | ||
2737 | if (is_sync) { | |
2738 | if (!cfq_class_idle(cfqq)) | |
2739 | cfq_mark_cfqq_idle_window(cfqq); | |
2740 | cfq_mark_cfqq_sync(cfqq); | |
2741 | } | |
2742 | cfqq->pid = pid; | |
2743 | } | |
2744 | ||
24610333 | 2745 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
c5869807 | 2746 | static void changed_cgroup(struct cfq_io_cq *cic) |
24610333 VG |
2747 | { |
2748 | struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1); | |
bca4b914 | 2749 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
24610333 VG |
2750 | struct request_queue *q; |
2751 | ||
2752 | if (unlikely(!cfqd)) | |
2753 | return; | |
2754 | ||
2755 | q = cfqd->queue; | |
2756 | ||
24610333 VG |
2757 | if (sync_cfqq) { |
2758 | /* | |
2759 | * Drop reference to sync queue. A new sync queue will be | |
2760 | * assigned in new group upon arrival of a fresh request. | |
2761 | */ | |
2762 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
2763 | cic_set_cfqq(cic, NULL, 1); | |
2764 | cfq_put_queue(sync_cfqq); | |
2765 | } | |
24610333 | 2766 | } |
24610333 VG |
2767 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
2768 | ||
22e2c507 | 2769 | static struct cfq_queue * |
a6151c3a | 2770 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, |
fd0928df | 2771 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 2772 | { |
0a5a7d0e | 2773 | struct blkio_cgroup *blkcg; |
22e2c507 | 2774 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
c5869807 | 2775 | struct cfq_io_cq *cic; |
cdb16e8f | 2776 | struct cfq_group *cfqg; |
22e2c507 JA |
2777 | |
2778 | retry: | |
2a7f1244 TH |
2779 | rcu_read_lock(); |
2780 | ||
0a5a7d0e TH |
2781 | blkcg = task_blkio_cgroup(current); |
2782 | ||
cd1604fa TH |
2783 | cfqg = cfq_lookup_create_cfqg(cfqd, blkcg); |
2784 | ||
4ac845a2 | 2785 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
2786 | /* cic always exists here */ |
2787 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 | 2788 | |
6118b70b JA |
2789 | /* |
2790 | * Always try a new alloc if we fell back to the OOM cfqq | |
2791 | * originally, since it should just be a temporary situation. | |
2792 | */ | |
2793 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
2794 | cfqq = NULL; | |
22e2c507 JA |
2795 | if (new_cfqq) { |
2796 | cfqq = new_cfqq; | |
2797 | new_cfqq = NULL; | |
2798 | } else if (gfp_mask & __GFP_WAIT) { | |
2a7f1244 | 2799 | rcu_read_unlock(); |
22e2c507 | 2800 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c | 2801 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 2802 | gfp_mask | __GFP_ZERO, |
94f6030c | 2803 | cfqd->queue->node); |
22e2c507 | 2804 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
2805 | if (new_cfqq) |
2806 | goto retry; | |
22e2c507 | 2807 | } else { |
94f6030c CL |
2808 | cfqq = kmem_cache_alloc_node(cfq_pool, |
2809 | gfp_mask | __GFP_ZERO, | |
2810 | cfqd->queue->node); | |
22e2c507 JA |
2811 | } |
2812 | ||
6118b70b JA |
2813 | if (cfqq) { |
2814 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
2815 | cfq_init_prio_data(cfqq, ioc); | |
cdb16e8f | 2816 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
2817 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
2818 | } else | |
2819 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
2820 | } |
2821 | ||
2822 | if (new_cfqq) | |
2823 | kmem_cache_free(cfq_pool, new_cfqq); | |
2824 | ||
2a7f1244 | 2825 | rcu_read_unlock(); |
22e2c507 JA |
2826 | return cfqq; |
2827 | } | |
2828 | ||
c2dea2d1 VT |
2829 | static struct cfq_queue ** |
2830 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
2831 | { | |
fe094d98 | 2832 | switch (ioprio_class) { |
c2dea2d1 VT |
2833 | case IOPRIO_CLASS_RT: |
2834 | return &cfqd->async_cfqq[0][ioprio]; | |
2835 | case IOPRIO_CLASS_BE: | |
2836 | return &cfqd->async_cfqq[1][ioprio]; | |
2837 | case IOPRIO_CLASS_IDLE: | |
2838 | return &cfqd->async_idle_cfqq; | |
2839 | default: | |
2840 | BUG(); | |
2841 | } | |
2842 | } | |
2843 | ||
15c31be4 | 2844 | static struct cfq_queue * |
a6151c3a | 2845 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc, |
15c31be4 JA |
2846 | gfp_t gfp_mask) |
2847 | { | |
fd0928df JA |
2848 | const int ioprio = task_ioprio(ioc); |
2849 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 2850 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
2851 | struct cfq_queue *cfqq = NULL; |
2852 | ||
c2dea2d1 VT |
2853 | if (!is_sync) { |
2854 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
2855 | cfqq = *async_cfqq; | |
2856 | } | |
2857 | ||
6118b70b | 2858 | if (!cfqq) |
fd0928df | 2859 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
15c31be4 JA |
2860 | |
2861 | /* | |
2862 | * pin the queue now that it's allocated, scheduler exit will prune it | |
2863 | */ | |
c2dea2d1 | 2864 | if (!is_sync && !(*async_cfqq)) { |
30d7b944 | 2865 | cfqq->ref++; |
c2dea2d1 | 2866 | *async_cfqq = cfqq; |
15c31be4 JA |
2867 | } |
2868 | ||
30d7b944 | 2869 | cfqq->ref++; |
15c31be4 JA |
2870 | return cfqq; |
2871 | } | |
2872 | ||
22e2c507 | 2873 | static void |
383cd721 | 2874 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 2875 | { |
383cd721 SL |
2876 | unsigned long elapsed = jiffies - ttime->last_end_request; |
2877 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 2878 | |
383cd721 SL |
2879 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
2880 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
2881 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
2882 | } | |
2883 | ||
2884 | static void | |
2885 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 2886 | struct cfq_io_cq *cic) |
383cd721 | 2887 | { |
f5f2b6ce | 2888 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 2889 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
2890 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
2891 | cfqd->cfq_slice_idle); | |
2892 | } | |
7700fc4f SL |
2893 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2894 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
2895 | #endif | |
22e2c507 | 2896 | } |
1da177e4 | 2897 | |
206dc69b | 2898 | static void |
b2c18e1e | 2899 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 2900 | struct request *rq) |
206dc69b | 2901 | { |
3dde36dd | 2902 | sector_t sdist = 0; |
41647e7a | 2903 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
2904 | if (cfqq->last_request_pos) { |
2905 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
2906 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
2907 | else | |
2908 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
2909 | } | |
206dc69b | 2910 | |
3dde36dd | 2911 | cfqq->seek_history <<= 1; |
41647e7a CZ |
2912 | if (blk_queue_nonrot(cfqd->queue)) |
2913 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
2914 | else | |
2915 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 2916 | } |
1da177e4 | 2917 | |
22e2c507 JA |
2918 | /* |
2919 | * Disable idle window if the process thinks too long or seeks so much that | |
2920 | * it doesn't matter | |
2921 | */ | |
2922 | static void | |
2923 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 2924 | struct cfq_io_cq *cic) |
22e2c507 | 2925 | { |
7b679138 | 2926 | int old_idle, enable_idle; |
1be92f2f | 2927 | |
0871714e JA |
2928 | /* |
2929 | * Don't idle for async or idle io prio class | |
2930 | */ | |
2931 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
2932 | return; |
2933 | ||
c265a7f4 | 2934 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 2935 | |
76280aff CZ |
2936 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
2937 | cfq_mark_cfqq_deep(cfqq); | |
2938 | ||
749ef9f8 CZ |
2939 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
2940 | enable_idle = 0; | |
c5869807 TH |
2941 | else if (!atomic_read(&cic->icq.ioc->nr_tasks) || |
2942 | !cfqd->cfq_slice_idle || | |
2943 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 2944 | enable_idle = 0; |
383cd721 SL |
2945 | else if (sample_valid(cic->ttime.ttime_samples)) { |
2946 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
2947 | enable_idle = 0; |
2948 | else | |
2949 | enable_idle = 1; | |
1da177e4 LT |
2950 | } |
2951 | ||
7b679138 JA |
2952 | if (old_idle != enable_idle) { |
2953 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
2954 | if (enable_idle) | |
2955 | cfq_mark_cfqq_idle_window(cfqq); | |
2956 | else | |
2957 | cfq_clear_cfqq_idle_window(cfqq); | |
2958 | } | |
22e2c507 | 2959 | } |
1da177e4 | 2960 | |
22e2c507 JA |
2961 | /* |
2962 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
2963 | * no or if we aren't sure, a 1 will cause a preempt. | |
2964 | */ | |
a6151c3a | 2965 | static bool |
22e2c507 | 2966 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 2967 | struct request *rq) |
22e2c507 | 2968 | { |
6d048f53 | 2969 | struct cfq_queue *cfqq; |
22e2c507 | 2970 | |
6d048f53 JA |
2971 | cfqq = cfqd->active_queue; |
2972 | if (!cfqq) | |
a6151c3a | 2973 | return false; |
22e2c507 | 2974 | |
6d048f53 | 2975 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 2976 | return false; |
22e2c507 JA |
2977 | |
2978 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 2979 | return true; |
1e3335de | 2980 | |
875feb63 DS |
2981 | /* |
2982 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
2983 | */ | |
2984 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
2985 | return false; | |
2986 | ||
374f84ac JA |
2987 | /* |
2988 | * if the new request is sync, but the currently running queue is | |
2989 | * not, let the sync request have priority. | |
2990 | */ | |
5e705374 | 2991 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 2992 | return true; |
1e3335de | 2993 | |
8682e1f1 VG |
2994 | if (new_cfqq->cfqg != cfqq->cfqg) |
2995 | return false; | |
2996 | ||
2997 | if (cfq_slice_used(cfqq)) | |
2998 | return true; | |
2999 | ||
3000 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
3001 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && | |
3002 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && | |
3003 | new_cfqq->service_tree->count == 2 && | |
3004 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3005 | return true; | |
3006 | ||
b53d1ed7 JA |
3007 | /* |
3008 | * So both queues are sync. Let the new request get disk time if | |
3009 | * it's a metadata request and the current queue is doing regular IO. | |
3010 | */ | |
65299a3b | 3011 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
3012 | return true; |
3013 | ||
3a9a3f6c DS |
3014 | /* |
3015 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3016 | */ | |
3017 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3018 | return true; |
3a9a3f6c | 3019 | |
d2d59e18 SL |
3020 | /* An idle queue should not be idle now for some reason */ |
3021 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
3022 | return true; | |
3023 | ||
1e3335de | 3024 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3025 | return false; |
1e3335de JA |
3026 | |
3027 | /* | |
3028 | * if this request is as-good as one we would expect from the | |
3029 | * current cfqq, let it preempt | |
3030 | */ | |
e9ce335d | 3031 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3032 | return true; |
1e3335de | 3033 | |
a6151c3a | 3034 | return false; |
22e2c507 JA |
3035 | } |
3036 | ||
3037 | /* | |
3038 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3039 | * let it have half of its nominal slice. | |
3040 | */ | |
3041 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3042 | { | |
df0793ab SL |
3043 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
3044 | ||
7b679138 | 3045 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 3046 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3047 | |
f8ae6e3e SL |
3048 | /* |
3049 | * workload type is changed, don't save slice, otherwise preempt | |
3050 | * doesn't happen | |
3051 | */ | |
df0793ab | 3052 | if (old_type != cfqq_type(cfqq)) |
f8ae6e3e SL |
3053 | cfqq->cfqg->saved_workload_slice = 0; |
3054 | ||
bf572256 JA |
3055 | /* |
3056 | * Put the new queue at the front of the of the current list, | |
3057 | * so we know that it will be selected next. | |
3058 | */ | |
3059 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3060 | |
3061 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3062 | |
62a37f6b JT |
3063 | cfqq->slice_end = 0; |
3064 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3065 | } |
3066 | ||
22e2c507 | 3067 | /* |
5e705374 | 3068 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3069 | * something we should do about it |
3070 | */ | |
3071 | static void | |
5e705374 JA |
3072 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3073 | struct request *rq) | |
22e2c507 | 3074 | { |
c5869807 | 3075 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3076 | |
45333d5a | 3077 | cfqd->rq_queued++; |
65299a3b CH |
3078 | if (rq->cmd_flags & REQ_PRIO) |
3079 | cfqq->prio_pending++; | |
374f84ac | 3080 | |
383cd721 | 3081 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3082 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3083 | cfq_update_idle_window(cfqd, cfqq, cic); |
3084 | ||
b2c18e1e | 3085 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3086 | |
3087 | if (cfqq == cfqd->active_queue) { | |
3088 | /* | |
b029195d JA |
3089 | * Remember that we saw a request from this process, but |
3090 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3091 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3092 | * and merging. If the request is already larger than a single |
3093 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3094 | * merging is already done. Ditto for a busy system that |
3095 | * has other work pending, don't risk delaying until the | |
3096 | * idle timer unplug to continue working. | |
22e2c507 | 3097 | */ |
d6ceb25e | 3098 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3099 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3100 | cfqd->busy_queues > 1) { | |
812df48d | 3101 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3102 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3103 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3104 | } else { |
e98ef89b | 3105 | cfq_blkiocg_update_idle_time_stats( |
c1768268 TH |
3106 | cfqg_to_blkg(cfqq->cfqg), |
3107 | &blkio_policy_cfq); | |
bf791937 | 3108 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3109 | } |
d6ceb25e | 3110 | } |
5e705374 | 3111 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3112 | /* |
3113 | * not the active queue - expire current slice if it is | |
3114 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3115 | * has some old slice time left and is of higher priority or |
3116 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3117 | */ |
3118 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3119 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3120 | } |
1da177e4 LT |
3121 | } |
3122 | ||
165125e1 | 3123 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3124 | { |
b4878f24 | 3125 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3126 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3127 | |
7b679138 | 3128 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
c5869807 | 3129 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->icq.ioc); |
1da177e4 | 3130 | |
30996f40 | 3131 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3132 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3133 | cfq_add_rq_rb(rq); |
0381411e | 3134 | cfq_blkiocg_update_io_add_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 | 3135 | &blkio_policy_cfq, |
0381411e TH |
3136 | cfqg_to_blkg(cfqd->serving_group), |
3137 | rq_data_dir(rq), rq_is_sync(rq)); | |
5e705374 | 3138 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3139 | } |
3140 | ||
45333d5a AC |
3141 | /* |
3142 | * Update hw_tag based on peak queue depth over 50 samples under | |
3143 | * sufficient load. | |
3144 | */ | |
3145 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3146 | { | |
1a1238a7 SL |
3147 | struct cfq_queue *cfqq = cfqd->active_queue; |
3148 | ||
53c583d2 CZ |
3149 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3150 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3151 | |
3152 | if (cfqd->hw_tag == 1) | |
3153 | return; | |
45333d5a AC |
3154 | |
3155 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3156 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3157 | return; |
3158 | ||
1a1238a7 SL |
3159 | /* |
3160 | * If active queue hasn't enough requests and can idle, cfq might not | |
3161 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3162 | * case | |
3163 | */ | |
3164 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3165 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3166 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3167 | return; |
3168 | ||
45333d5a AC |
3169 | if (cfqd->hw_tag_samples++ < 50) |
3170 | return; | |
3171 | ||
e459dd08 | 3172 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3173 | cfqd->hw_tag = 1; |
3174 | else | |
3175 | cfqd->hw_tag = 0; | |
45333d5a AC |
3176 | } |
3177 | ||
7667aa06 VG |
3178 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3179 | { | |
c5869807 | 3180 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 3181 | |
02a8f01b JT |
3182 | /* If the queue already has requests, don't wait */ |
3183 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3184 | return false; | |
3185 | ||
7667aa06 VG |
3186 | /* If there are other queues in the group, don't wait */ |
3187 | if (cfqq->cfqg->nr_cfqq > 1) | |
3188 | return false; | |
3189 | ||
7700fc4f SL |
3190 | /* the only queue in the group, but think time is big */ |
3191 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
3192 | return false; | |
3193 | ||
7667aa06 VG |
3194 | if (cfq_slice_used(cfqq)) |
3195 | return true; | |
3196 | ||
3197 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
3198 | if (cic && sample_valid(cic->ttime.ttime_samples) |
3199 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
3200 | return true; |
3201 | ||
3202 | /* | |
3203 | * If think times is less than a jiffy than ttime_mean=0 and above | |
3204 | * will not be true. It might happen that slice has not expired yet | |
3205 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
3206 | * case where think time is less than a jiffy, mark the queue wait | |
3207 | * busy if only 1 jiffy is left in the slice. | |
3208 | */ | |
3209 | if (cfqq->slice_end - jiffies == 1) | |
3210 | return true; | |
3211 | ||
3212 | return false; | |
3213 | } | |
3214 | ||
165125e1 | 3215 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3216 | { |
5e705374 | 3217 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3218 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3219 | const int sync = rq_is_sync(rq); |
b4878f24 | 3220 | unsigned long now; |
1da177e4 | 3221 | |
b4878f24 | 3222 | now = jiffies; |
33659ebb CH |
3223 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3224 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 3225 | |
45333d5a AC |
3226 | cfq_update_hw_tag(cfqd); |
3227 | ||
53c583d2 | 3228 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 3229 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 3230 | cfqd->rq_in_driver--; |
6d048f53 | 3231 | cfqq->dispatched--; |
80bdf0c7 | 3232 | (RQ_CFQG(rq))->dispatched--; |
0381411e | 3233 | cfq_blkiocg_update_completion_stats(cfqg_to_blkg(cfqq->cfqg), |
c1768268 TH |
3234 | &blkio_policy_cfq, rq_start_time_ns(rq), |
3235 | rq_io_start_time_ns(rq), rq_data_dir(rq), | |
3236 | rq_is_sync(rq)); | |
1da177e4 | 3237 | |
53c583d2 | 3238 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 3239 | |
365722bb | 3240 | if (sync) { |
f5f2b6ce SL |
3241 | struct cfq_rb_root *service_tree; |
3242 | ||
383cd721 | 3243 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
3244 | |
3245 | if (cfq_cfqq_on_rr(cfqq)) | |
3246 | service_tree = cfqq->service_tree; | |
3247 | else | |
3248 | service_tree = service_tree_for(cfqq->cfqg, | |
3249 | cfqq_prio(cfqq), cfqq_type(cfqq)); | |
3250 | service_tree->ttime.last_end_request = now; | |
573412b2 CZ |
3251 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3252 | cfqd->last_delayed_sync = now; | |
365722bb | 3253 | } |
caaa5f9f | 3254 | |
7700fc4f SL |
3255 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3256 | cfqq->cfqg->ttime.last_end_request = now; | |
3257 | #endif | |
3258 | ||
caaa5f9f JA |
3259 | /* |
3260 | * If this is the active queue, check if it needs to be expired, | |
3261 | * or if we want to idle in case it has no pending requests. | |
3262 | */ | |
3263 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3264 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3265 | ||
44f7c160 JA |
3266 | if (cfq_cfqq_slice_new(cfqq)) { |
3267 | cfq_set_prio_slice(cfqd, cfqq); | |
3268 | cfq_clear_cfqq_slice_new(cfqq); | |
3269 | } | |
f75edf2d VG |
3270 | |
3271 | /* | |
7667aa06 VG |
3272 | * Should we wait for next request to come in before we expire |
3273 | * the queue. | |
f75edf2d | 3274 | */ |
7667aa06 | 3275 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
3276 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3277 | if (!cfqd->cfq_slice_idle) | |
3278 | extend_sl = cfqd->cfq_group_idle; | |
3279 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 3280 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 3281 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
3282 | } |
3283 | ||
a36e71f9 | 3284 | /* |
8e550632 CZ |
3285 | * Idling is not enabled on: |
3286 | * - expired queues | |
3287 | * - idle-priority queues | |
3288 | * - async queues | |
3289 | * - queues with still some requests queued | |
3290 | * - when there is a close cooperator | |
a36e71f9 | 3291 | */ |
0871714e | 3292 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 3293 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3294 | else if (sync && cfqq_empty && |
3295 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 3296 | cfq_arm_slice_timer(cfqd); |
8e550632 | 3297 | } |
caaa5f9f | 3298 | } |
6d048f53 | 3299 | |
53c583d2 | 3300 | if (!cfqd->rq_in_driver) |
23e018a1 | 3301 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3302 | } |
3303 | ||
89850f7e | 3304 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3305 | { |
1b379d8d | 3306 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3307 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3308 | return ELV_MQUEUE_MUST; |
3b18152c | 3309 | } |
1da177e4 | 3310 | |
22e2c507 | 3311 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3312 | } |
3313 | ||
165125e1 | 3314 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3315 | { |
3316 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3317 | struct task_struct *tsk = current; | |
c5869807 | 3318 | struct cfq_io_cq *cic; |
22e2c507 JA |
3319 | struct cfq_queue *cfqq; |
3320 | ||
3321 | /* | |
3322 | * don't force setup of a queue from here, as a call to may_queue | |
3323 | * does not necessarily imply that a request actually will be queued. | |
3324 | * so just lookup a possibly existing queue, or return 'may queue' | |
3325 | * if that fails | |
3326 | */ | |
4ac845a2 | 3327 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3328 | if (!cic) |
3329 | return ELV_MQUEUE_MAY; | |
3330 | ||
b0b78f81 | 3331 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3332 | if (cfqq) { |
c5869807 | 3333 | cfq_init_prio_data(cfqq, cic->icq.ioc); |
22e2c507 | 3334 | |
89850f7e | 3335 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3336 | } |
3337 | ||
3338 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3339 | } |
3340 | ||
1da177e4 LT |
3341 | /* |
3342 | * queue lock held here | |
3343 | */ | |
bb37b94c | 3344 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3345 | { |
5e705374 | 3346 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3347 | |
5e705374 | 3348 | if (cfqq) { |
22e2c507 | 3349 | const int rw = rq_data_dir(rq); |
1da177e4 | 3350 | |
22e2c507 JA |
3351 | BUG_ON(!cfqq->allocated[rw]); |
3352 | cfqq->allocated[rw]--; | |
1da177e4 | 3353 | |
7f1dc8a2 | 3354 | /* Put down rq reference on cfqg */ |
1adaf3dd | 3355 | blkg_put(cfqg_to_blkg(RQ_CFQG(rq))); |
a612fddf TH |
3356 | rq->elv.priv[0] = NULL; |
3357 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 3358 | |
1da177e4 LT |
3359 | cfq_put_queue(cfqq); |
3360 | } | |
3361 | } | |
3362 | ||
df5fe3e8 | 3363 | static struct cfq_queue * |
c5869807 | 3364 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
3365 | struct cfq_queue *cfqq) |
3366 | { | |
3367 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3368 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3369 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3370 | cfq_put_queue(cfqq); |
3371 | return cic_to_cfqq(cic, 1); | |
3372 | } | |
3373 | ||
e6c5bc73 JM |
3374 | /* |
3375 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3376 | * was the last process referring to said cfqq. | |
3377 | */ | |
3378 | static struct cfq_queue * | |
c5869807 | 3379 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
3380 | { |
3381 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
3382 | cfqq->pid = current->pid; |
3383 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 3384 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
3385 | return cfqq; |
3386 | } | |
3387 | ||
3388 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
3389 | |
3390 | cfq_put_cooperator(cfqq); | |
3391 | ||
e6c5bc73 JM |
3392 | cfq_put_queue(cfqq); |
3393 | return NULL; | |
3394 | } | |
1da177e4 | 3395 | /* |
22e2c507 | 3396 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3397 | */ |
22e2c507 | 3398 | static int |
165125e1 | 3399 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
3400 | { |
3401 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 3402 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 3403 | const int rw = rq_data_dir(rq); |
a6151c3a | 3404 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3405 | struct cfq_queue *cfqq; |
d705ae6b | 3406 | unsigned int changed; |
1da177e4 LT |
3407 | |
3408 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3409 | ||
216284c3 | 3410 | spin_lock_irq(q->queue_lock); |
f1f8cc94 TH |
3411 | |
3412 | /* handle changed notifications */ | |
d705ae6b TH |
3413 | changed = icq_get_changed(&cic->icq); |
3414 | if (unlikely(changed & ICQ_IOPRIO_CHANGED)) | |
3415 | changed_ioprio(cic); | |
f1f8cc94 | 3416 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
d705ae6b TH |
3417 | if (unlikely(changed & ICQ_CGROUP_CHANGED)) |
3418 | changed_cgroup(cic); | |
f1f8cc94 | 3419 | #endif |
22e2c507 | 3420 | |
e6c5bc73 | 3421 | new_queue: |
91fac317 | 3422 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3423 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
c5869807 | 3424 | cfqq = cfq_get_queue(cfqd, is_sync, cic->icq.ioc, gfp_mask); |
91fac317 | 3425 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3426 | } else { |
e6c5bc73 JM |
3427 | /* |
3428 | * If the queue was seeky for too long, break it apart. | |
3429 | */ | |
ae54abed | 3430 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
3431 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
3432 | cfqq = split_cfqq(cic, cfqq); | |
3433 | if (!cfqq) | |
3434 | goto new_queue; | |
3435 | } | |
3436 | ||
df5fe3e8 JM |
3437 | /* |
3438 | * Check to see if this queue is scheduled to merge with | |
3439 | * another, closely cooperating queue. The merging of | |
3440 | * queues happens here as it must be done in process context. | |
3441 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3442 | */ | |
3443 | if (cfqq->new_cfqq) | |
3444 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3445 | } |
1da177e4 LT |
3446 | |
3447 | cfqq->allocated[rw]++; | |
1da177e4 | 3448 | |
6fae9c25 | 3449 | cfqq->ref++; |
1adaf3dd | 3450 | blkg_get(cfqg_to_blkg(cfqq->cfqg)); |
a612fddf | 3451 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 3452 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 3453 | spin_unlock_irq(q->queue_lock); |
5e705374 | 3454 | return 0; |
1da177e4 LT |
3455 | } |
3456 | ||
65f27f38 | 3457 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3458 | { |
65f27f38 | 3459 | struct cfq_data *cfqd = |
23e018a1 | 3460 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3461 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3462 | |
40bb54d1 | 3463 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 3464 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3465 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3466 | } |
3467 | ||
3468 | /* | |
3469 | * Timer running if the active_queue is currently idling inside its time slice | |
3470 | */ | |
3471 | static void cfq_idle_slice_timer(unsigned long data) | |
3472 | { | |
3473 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3474 | struct cfq_queue *cfqq; | |
3475 | unsigned long flags; | |
3c6bd2f8 | 3476 | int timed_out = 1; |
22e2c507 | 3477 | |
7b679138 JA |
3478 | cfq_log(cfqd, "idle timer fired"); |
3479 | ||
22e2c507 JA |
3480 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3481 | ||
fe094d98 JA |
3482 | cfqq = cfqd->active_queue; |
3483 | if (cfqq) { | |
3c6bd2f8 JA |
3484 | timed_out = 0; |
3485 | ||
b029195d JA |
3486 | /* |
3487 | * We saw a request before the queue expired, let it through | |
3488 | */ | |
3489 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3490 | goto out_kick; | |
3491 | ||
22e2c507 JA |
3492 | /* |
3493 | * expired | |
3494 | */ | |
44f7c160 | 3495 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
3496 | goto expire; |
3497 | ||
3498 | /* | |
3499 | * only expire and reinvoke request handler, if there are | |
3500 | * other queues with pending requests | |
3501 | */ | |
caaa5f9f | 3502 | if (!cfqd->busy_queues) |
22e2c507 | 3503 | goto out_cont; |
22e2c507 JA |
3504 | |
3505 | /* | |
3506 | * not expired and it has a request pending, let it dispatch | |
3507 | */ | |
75e50984 | 3508 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3509 | goto out_kick; |
76280aff CZ |
3510 | |
3511 | /* | |
3512 | * Queue depth flag is reset only when the idle didn't succeed | |
3513 | */ | |
3514 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
3515 | } |
3516 | expire: | |
e5ff082e | 3517 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 3518 | out_kick: |
23e018a1 | 3519 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
3520 | out_cont: |
3521 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3522 | } | |
3523 | ||
3b18152c JA |
3524 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3525 | { | |
3526 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 3527 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 3528 | } |
22e2c507 | 3529 | |
c2dea2d1 VT |
3530 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3531 | { | |
3532 | int i; | |
3533 | ||
3534 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
3535 | if (cfqd->async_cfqq[0][i]) | |
3536 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
3537 | if (cfqd->async_cfqq[1][i]) | |
3538 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 3539 | } |
2389d1ef ON |
3540 | |
3541 | if (cfqd->async_idle_cfqq) | |
3542 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
3543 | } |
3544 | ||
b374d18a | 3545 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 3546 | { |
22e2c507 | 3547 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 3548 | struct request_queue *q = cfqd->queue; |
56edf7d7 | 3549 | bool wait = false; |
22e2c507 | 3550 | |
3b18152c | 3551 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 3552 | |
d9ff4187 | 3553 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3554 | |
d9ff4187 | 3555 | if (cfqd->active_queue) |
e5ff082e | 3556 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 3557 | |
c2dea2d1 | 3558 | cfq_put_async_queues(cfqd); |
b1c35769 | 3559 | cfq_release_cfq_groups(cfqd); |
56edf7d7 VG |
3560 | |
3561 | /* | |
3562 | * If there are groups which we could not unlink from blkcg list, | |
3563 | * wait for a rcu period for them to be freed. | |
3564 | */ | |
3565 | if (cfqd->nr_blkcg_linked_grps) | |
3566 | wait = true; | |
15c31be4 | 3567 | |
d9ff4187 | 3568 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
3569 | |
3570 | cfq_shutdown_timer_wq(cfqd); | |
3571 | ||
56edf7d7 VG |
3572 | /* |
3573 | * Wait for cfqg->blkg->key accessors to exit their grace periods. | |
3574 | * Do this wait only if there are other unlinked groups out | |
3575 | * there. This can happen if cgroup deletion path claimed the | |
3576 | * responsibility of cleaning up a group before queue cleanup code | |
3577 | * get to the group. | |
3578 | * | |
3579 | * Do not call synchronize_rcu() unconditionally as there are drivers | |
3580 | * which create/delete request queue hundreds of times during scan/boot | |
3581 | * and synchronize_rcu() can take significant time and slow down boot. | |
3582 | */ | |
3583 | if (wait) | |
3584 | synchronize_rcu(); | |
2abae55f | 3585 | |
f51b802c TH |
3586 | #ifndef CONFIG_CFQ_GROUP_IOSCHED |
3587 | kfree(cfqd->root_group); | |
2abae55f | 3588 | #endif |
56edf7d7 | 3589 | kfree(cfqd); |
1da177e4 LT |
3590 | } |
3591 | ||
b2fab5ac | 3592 | static int cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
3593 | { |
3594 | struct cfq_data *cfqd; | |
cd1604fa | 3595 | struct blkio_group *blkg __maybe_unused; |
f51b802c | 3596 | int i; |
1da177e4 | 3597 | |
94f6030c | 3598 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
a73f730d | 3599 | if (!cfqd) |
b2fab5ac | 3600 | return -ENOMEM; |
80b15c73 | 3601 | |
f51b802c TH |
3602 | cfqd->queue = q; |
3603 | q->elevator->elevator_data = cfqd; | |
3604 | ||
1fa8f6d6 VG |
3605 | /* Init root service tree */ |
3606 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
3607 | ||
f51b802c | 3608 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 3609 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
f51b802c TH |
3610 | rcu_read_lock(); |
3611 | spin_lock_irq(q->queue_lock); | |
5624a4e4 | 3612 | |
cd1604fa TH |
3613 | blkg = blkg_lookup_create(&blkio_root_cgroup, q, BLKIO_POLICY_PROP, |
3614 | true); | |
3615 | if (!IS_ERR(blkg)) | |
0381411e | 3616 | cfqd->root_group = blkg_to_cfqg(blkg); |
f51b802c TH |
3617 | |
3618 | spin_unlock_irq(q->queue_lock); | |
3619 | rcu_read_unlock(); | |
3620 | #else | |
3621 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), | |
3622 | GFP_KERNEL, cfqd->queue->node); | |
3623 | if (cfqd->root_group) | |
3624 | cfq_init_cfqg_base(cfqd->root_group); | |
3625 | #endif | |
3626 | if (!cfqd->root_group) { | |
5624a4e4 | 3627 | kfree(cfqd); |
b2fab5ac | 3628 | return -ENOMEM; |
5624a4e4 VG |
3629 | } |
3630 | ||
f51b802c | 3631 | cfqd->root_group->weight = 2*BLKIO_WEIGHT_DEFAULT; |
5624a4e4 | 3632 | |
26a2ac00 JA |
3633 | /* |
3634 | * Not strictly needed (since RB_ROOT just clears the node and we | |
3635 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
3636 | * to add magic to the rb code | |
3637 | */ | |
3638 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
3639 | cfqd->prio_trees[i] = RB_ROOT; | |
3640 | ||
6118b70b JA |
3641 | /* |
3642 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
3643 | * Grab a permanent reference to it, so that the normal code flow | |
f51b802c TH |
3644 | * will not attempt to free it. oom_cfqq is linked to root_group |
3645 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
3646 | * the reference from linking right away. | |
6118b70b JA |
3647 | */ |
3648 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 3649 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
3650 | |
3651 | spin_lock_irq(q->queue_lock); | |
f51b802c | 3652 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
1adaf3dd TH |
3653 | blkg_put(cfqg_to_blkg(cfqd->root_group)); |
3654 | spin_unlock_irq(q->queue_lock); | |
1da177e4 | 3655 | |
22e2c507 JA |
3656 | init_timer(&cfqd->idle_slice_timer); |
3657 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
3658 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
3659 | ||
23e018a1 | 3660 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 3661 | |
1da177e4 | 3662 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
3663 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
3664 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
3665 | cfqd->cfq_back_max = cfq_back_max; |
3666 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
3667 | cfqd->cfq_slice[0] = cfq_slice_async; |
3668 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
3669 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
3670 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
80bdf0c7 | 3671 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 3672 | cfqd->cfq_latency = 1; |
e459dd08 | 3673 | cfqd->hw_tag = -1; |
edc71131 CZ |
3674 | /* |
3675 | * we optimistically start assuming sync ops weren't delayed in last | |
3676 | * second, in order to have larger depth for async operations. | |
3677 | */ | |
573412b2 | 3678 | cfqd->last_delayed_sync = jiffies - HZ; |
b2fab5ac | 3679 | return 0; |
1da177e4 LT |
3680 | } |
3681 | ||
1da177e4 LT |
3682 | /* |
3683 | * sysfs parts below --> | |
3684 | */ | |
1da177e4 LT |
3685 | static ssize_t |
3686 | cfq_var_show(unsigned int var, char *page) | |
3687 | { | |
3688 | return sprintf(page, "%d\n", var); | |
3689 | } | |
3690 | ||
3691 | static ssize_t | |
3692 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
3693 | { | |
3694 | char *p = (char *) page; | |
3695 | ||
3696 | *var = simple_strtoul(p, &p, 10); | |
3697 | return count; | |
3698 | } | |
3699 | ||
1da177e4 | 3700 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 3701 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 3702 | { \ |
3d1ab40f | 3703 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3704 | unsigned int __data = __VAR; \ |
3705 | if (__CONV) \ | |
3706 | __data = jiffies_to_msecs(__data); \ | |
3707 | return cfq_var_show(__data, (page)); \ | |
3708 | } | |
3709 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
3710 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
3711 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
3712 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
3713 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 3714 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 3715 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
3716 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
3717 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
3718 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 3719 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
1da177e4 LT |
3720 | #undef SHOW_FUNCTION |
3721 | ||
3722 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 3723 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 3724 | { \ |
3d1ab40f | 3725 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3726 | unsigned int __data; \ |
3727 | int ret = cfq_var_store(&__data, (page), count); \ | |
3728 | if (__data < (MIN)) \ | |
3729 | __data = (MIN); \ | |
3730 | else if (__data > (MAX)) \ | |
3731 | __data = (MAX); \ | |
3732 | if (__CONV) \ | |
3733 | *(__PTR) = msecs_to_jiffies(__data); \ | |
3734 | else \ | |
3735 | *(__PTR) = __data; \ | |
3736 | return ret; \ | |
3737 | } | |
3738 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
3739 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
3740 | UINT_MAX, 1); | |
3741 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
3742 | UINT_MAX, 1); | |
e572ec7e | 3743 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
3744 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
3745 | UINT_MAX, 0); | |
22e2c507 | 3746 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 3747 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
3748 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
3749 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
3750 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
3751 | UINT_MAX, 0); | |
963b72fc | 3752 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
1da177e4 LT |
3753 | #undef STORE_FUNCTION |
3754 | ||
e572ec7e AV |
3755 | #define CFQ_ATTR(name) \ |
3756 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
3757 | ||
3758 | static struct elv_fs_entry cfq_attrs[] = { | |
3759 | CFQ_ATTR(quantum), | |
e572ec7e AV |
3760 | CFQ_ATTR(fifo_expire_sync), |
3761 | CFQ_ATTR(fifo_expire_async), | |
3762 | CFQ_ATTR(back_seek_max), | |
3763 | CFQ_ATTR(back_seek_penalty), | |
3764 | CFQ_ATTR(slice_sync), | |
3765 | CFQ_ATTR(slice_async), | |
3766 | CFQ_ATTR(slice_async_rq), | |
3767 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 3768 | CFQ_ATTR(group_idle), |
963b72fc | 3769 | CFQ_ATTR(low_latency), |
e572ec7e | 3770 | __ATTR_NULL |
1da177e4 LT |
3771 | }; |
3772 | ||
1da177e4 LT |
3773 | static struct elevator_type iosched_cfq = { |
3774 | .ops = { | |
3775 | .elevator_merge_fn = cfq_merge, | |
3776 | .elevator_merged_fn = cfq_merged_request, | |
3777 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 3778 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 3779 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 3780 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 3781 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 3782 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 3783 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 3784 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
3785 | .elevator_former_req_fn = elv_rb_former_request, |
3786 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 3787 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 3788 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
3789 | .elevator_set_req_fn = cfq_set_request, |
3790 | .elevator_put_req_fn = cfq_put_request, | |
3791 | .elevator_may_queue_fn = cfq_may_queue, | |
3792 | .elevator_init_fn = cfq_init_queue, | |
3793 | .elevator_exit_fn = cfq_exit_queue, | |
3794 | }, | |
3d3c2379 TH |
3795 | .icq_size = sizeof(struct cfq_io_cq), |
3796 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 3797 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 3798 | .elevator_name = "cfq", |
1da177e4 LT |
3799 | .elevator_owner = THIS_MODULE, |
3800 | }; | |
3801 | ||
3e252066 VG |
3802 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3803 | static struct blkio_policy_type blkio_policy_cfq = { | |
3804 | .ops = { | |
0381411e | 3805 | .blkio_init_group_fn = cfq_init_blkio_group, |
cd1604fa | 3806 | .blkio_link_group_fn = cfq_link_blkio_group, |
3e252066 | 3807 | .blkio_unlink_group_fn = cfq_unlink_blkio_group, |
72e06c25 | 3808 | .blkio_clear_queue_fn = cfq_clear_queue, |
3e252066 VG |
3809 | .blkio_update_group_weight_fn = cfq_update_blkio_group_weight, |
3810 | }, | |
062a644d | 3811 | .plid = BLKIO_POLICY_PROP, |
0381411e | 3812 | .pdata_size = sizeof(struct cfq_group), |
3e252066 | 3813 | }; |
3e252066 VG |
3814 | #endif |
3815 | ||
1da177e4 LT |
3816 | static int __init cfq_init(void) |
3817 | { | |
3d3c2379 TH |
3818 | int ret; |
3819 | ||
22e2c507 JA |
3820 | /* |
3821 | * could be 0 on HZ < 1000 setups | |
3822 | */ | |
3823 | if (!cfq_slice_async) | |
3824 | cfq_slice_async = 1; | |
3825 | if (!cfq_slice_idle) | |
3826 | cfq_slice_idle = 1; | |
3827 | ||
80bdf0c7 VG |
3828 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3829 | if (!cfq_group_idle) | |
3830 | cfq_group_idle = 1; | |
3831 | #else | |
3832 | cfq_group_idle = 0; | |
3833 | #endif | |
3d3c2379 TH |
3834 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
3835 | if (!cfq_pool) | |
1da177e4 LT |
3836 | return -ENOMEM; |
3837 | ||
3d3c2379 TH |
3838 | ret = elv_register(&iosched_cfq); |
3839 | if (ret) { | |
3840 | kmem_cache_destroy(cfq_pool); | |
3841 | return ret; | |
3842 | } | |
3d3c2379 | 3843 | |
b95ada55 | 3844 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3e252066 | 3845 | blkio_policy_register(&blkio_policy_cfq); |
b95ada55 | 3846 | #endif |
2fdd82bd | 3847 | return 0; |
1da177e4 LT |
3848 | } |
3849 | ||
3850 | static void __exit cfq_exit(void) | |
3851 | { | |
b95ada55 | 3852 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3e252066 | 3853 | blkio_policy_unregister(&blkio_policy_cfq); |
b95ada55 | 3854 | #endif |
1da177e4 | 3855 | elv_unregister(&iosched_cfq); |
3d3c2379 | 3856 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
3857 | } |
3858 | ||
3859 | module_init(cfq_init); | |
3860 | module_exit(cfq_exit); | |
3861 | ||
3862 | MODULE_AUTHOR("Jens Axboe"); | |
3863 | MODULE_LICENSE("GPL"); | |
3864 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |