Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * CFQ, or complete fairness queueing, disk scheduler. |
3 | * | |
4 | * Based on ideas from a previously unfinished io | |
5 | * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli. | |
6 | * | |
0fe23479 | 7 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
1da177e4 | 8 | */ |
1da177e4 | 9 | #include <linux/module.h> |
5a0e3ad6 | 10 | #include <linux/slab.h> |
1cc9be68 AV |
11 | #include <linux/blkdev.h> |
12 | #include <linux/elevator.h> | |
ad5ebd2f | 13 | #include <linux/jiffies.h> |
1da177e4 | 14 | #include <linux/rbtree.h> |
22e2c507 | 15 | #include <linux/ioprio.h> |
7b679138 | 16 | #include <linux/blktrace_api.h> |
6e736be7 | 17 | #include "blk.h" |
e98ef89b | 18 | #include "cfq.h" |
1da177e4 LT |
19 | |
20 | /* | |
21 | * tunables | |
22 | */ | |
fe094d98 | 23 | /* max queue in one round of service */ |
abc3c744 | 24 | static const int cfq_quantum = 8; |
64100099 | 25 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
26 | /* maximum backwards seek, in KiB */ |
27 | static const int cfq_back_max = 16 * 1024; | |
28 | /* penalty of a backwards seek */ | |
29 | static const int cfq_back_penalty = 2; | |
64100099 | 30 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 31 | static int cfq_slice_async = HZ / 25; |
64100099 | 32 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 33 | static int cfq_slice_idle = HZ / 125; |
80bdf0c7 | 34 | static int cfq_group_idle = HZ / 125; |
5db5d642 CZ |
35 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
36 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 37 | |
d9e7620e | 38 | /* |
0871714e | 39 | * offset from end of service tree |
d9e7620e | 40 | */ |
0871714e | 41 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
42 | |
43 | /* | |
44 | * below this threshold, we consider thinktime immediate | |
45 | */ | |
46 | #define CFQ_MIN_TT (2) | |
47 | ||
22e2c507 | 48 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 49 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 50 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 51 | |
3dde36dd | 52 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 53 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 54 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 55 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 56 | |
fe094d98 | 57 | #define RQ_CIC(rq) \ |
c186794d MS |
58 | ((struct cfq_io_context *) (rq)->elevator_private[0]) |
59 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private[1]) | |
60 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elevator_private[2]) | |
1da177e4 | 61 | |
e18b890b CL |
62 | static struct kmem_cache *cfq_pool; |
63 | static struct kmem_cache *cfq_ioc_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 | |
cc09e299 JA |
72 | /* |
73 | * Most of our rbtree usage is for sorting with min extraction, so | |
74 | * if we cache the leftmost node we don't have to walk down the tree | |
75 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
76 | * move this into the elevator for the rq sorting as well. | |
77 | */ | |
78 | struct cfq_rb_root { | |
79 | struct rb_root rb; | |
80 | struct rb_node *left; | |
aa6f6a3d | 81 | unsigned count; |
73e9ffdd | 82 | unsigned total_weight; |
1fa8f6d6 | 83 | u64 min_vdisktime; |
f5f2b6ce | 84 | struct cfq_ttime ttime; |
cc09e299 | 85 | }; |
f5f2b6ce SL |
86 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
87 | .ttime = {.last_end_request = jiffies,},} | |
cc09e299 | 88 | |
6118b70b JA |
89 | /* |
90 | * Per process-grouping structure | |
91 | */ | |
92 | struct cfq_queue { | |
93 | /* reference count */ | |
30d7b944 | 94 | int ref; |
6118b70b JA |
95 | /* various state flags, see below */ |
96 | unsigned int flags; | |
97 | /* parent cfq_data */ | |
98 | struct cfq_data *cfqd; | |
99 | /* service_tree member */ | |
100 | struct rb_node rb_node; | |
101 | /* service_tree key */ | |
102 | unsigned long rb_key; | |
103 | /* prio tree member */ | |
104 | struct rb_node p_node; | |
105 | /* prio tree root we belong to, if any */ | |
106 | struct rb_root *p_root; | |
107 | /* sorted list of pending requests */ | |
108 | struct rb_root sort_list; | |
109 | /* if fifo isn't expired, next request to serve */ | |
110 | struct request *next_rq; | |
111 | /* requests queued in sort_list */ | |
112 | int queued[2]; | |
113 | /* currently allocated requests */ | |
114 | int allocated[2]; | |
115 | /* fifo list of requests in sort_list */ | |
116 | struct list_head fifo; | |
117 | ||
dae739eb VG |
118 | /* time when queue got scheduled in to dispatch first request. */ |
119 | unsigned long dispatch_start; | |
f75edf2d | 120 | unsigned int allocated_slice; |
c4081ba5 | 121 | unsigned int slice_dispatch; |
dae739eb VG |
122 | /* time when first request from queue completed and slice started. */ |
123 | unsigned long slice_start; | |
6118b70b JA |
124 | unsigned long slice_end; |
125 | long slice_resid; | |
6118b70b | 126 | |
65299a3b CH |
127 | /* pending priority requests */ |
128 | int prio_pending; | |
6118b70b JA |
129 | /* number of requests that are on the dispatch list or inside driver */ |
130 | int dispatched; | |
131 | ||
132 | /* io prio of this group */ | |
133 | unsigned short ioprio, org_ioprio; | |
4aede84b | 134 | unsigned short ioprio_class; |
6118b70b | 135 | |
c4081ba5 RK |
136 | pid_t pid; |
137 | ||
3dde36dd | 138 | u32 seek_history; |
b2c18e1e JM |
139 | sector_t last_request_pos; |
140 | ||
aa6f6a3d | 141 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 142 | struct cfq_queue *new_cfqq; |
cdb16e8f | 143 | struct cfq_group *cfqg; |
c4e7893e VG |
144 | /* Number of sectors dispatched from queue in single dispatch round */ |
145 | unsigned long nr_sectors; | |
6118b70b JA |
146 | }; |
147 | ||
c0324a02 | 148 | /* |
718eee05 | 149 | * First index in the service_trees. |
c0324a02 CZ |
150 | * IDLE is handled separately, so it has negative index |
151 | */ | |
152 | enum wl_prio_t { | |
c0324a02 | 153 | BE_WORKLOAD = 0, |
615f0259 VG |
154 | RT_WORKLOAD = 1, |
155 | IDLE_WORKLOAD = 2, | |
b4627321 | 156 | CFQ_PRIO_NR, |
c0324a02 CZ |
157 | }; |
158 | ||
718eee05 CZ |
159 | /* |
160 | * Second index in the service_trees. | |
161 | */ | |
162 | enum wl_type_t { | |
163 | ASYNC_WORKLOAD = 0, | |
164 | SYNC_NOIDLE_WORKLOAD = 1, | |
165 | SYNC_WORKLOAD = 2 | |
166 | }; | |
167 | ||
cdb16e8f VG |
168 | /* This is per cgroup per device grouping structure */ |
169 | struct cfq_group { | |
1fa8f6d6 VG |
170 | /* group service_tree member */ |
171 | struct rb_node rb_node; | |
172 | ||
173 | /* group service_tree key */ | |
174 | u64 vdisktime; | |
25bc6b07 | 175 | unsigned int weight; |
8184f93e JT |
176 | unsigned int new_weight; |
177 | bool needs_update; | |
1fa8f6d6 VG |
178 | |
179 | /* number of cfqq currently on this group */ | |
180 | int nr_cfqq; | |
181 | ||
cdb16e8f | 182 | /* |
4495a7d4 | 183 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
184 | * create the array for each prio class but at run time it is used |
185 | * only for RT and BE class and slot for IDLE class remains unused. | |
186 | * This is primarily done to avoid confusion and a gcc warning. | |
187 | */ | |
188 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
189 | /* | |
190 | * rr lists of queues with requests. We maintain service trees for | |
191 | * RT and BE classes. These trees are subdivided in subclasses | |
192 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
193 | * class there is no subclassification and all the cfq queues go on | |
194 | * a single tree service_tree_idle. | |
cdb16e8f VG |
195 | * Counts are embedded in the cfq_rb_root |
196 | */ | |
197 | struct cfq_rb_root service_trees[2][3]; | |
198 | struct cfq_rb_root service_tree_idle; | |
dae739eb VG |
199 | |
200 | unsigned long saved_workload_slice; | |
201 | enum wl_type_t saved_workload; | |
202 | enum wl_prio_t saved_serving_prio; | |
25fb5169 VG |
203 | struct blkio_group blkg; |
204 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
205 | struct hlist_node cfqd_node; | |
329a6781 | 206 | int ref; |
25fb5169 | 207 | #endif |
80bdf0c7 VG |
208 | /* number of requests that are on the dispatch list or inside driver */ |
209 | int dispatched; | |
7700fc4f | 210 | struct cfq_ttime ttime; |
cdb16e8f | 211 | }; |
718eee05 | 212 | |
22e2c507 JA |
213 | /* |
214 | * Per block device queue structure | |
215 | */ | |
1da177e4 | 216 | struct cfq_data { |
165125e1 | 217 | struct request_queue *queue; |
1fa8f6d6 VG |
218 | /* Root service tree for cfq_groups */ |
219 | struct cfq_rb_root grp_service_tree; | |
cdb16e8f | 220 | struct cfq_group root_group; |
22e2c507 | 221 | |
c0324a02 CZ |
222 | /* |
223 | * The priority currently being served | |
22e2c507 | 224 | */ |
c0324a02 | 225 | enum wl_prio_t serving_prio; |
718eee05 CZ |
226 | enum wl_type_t serving_type; |
227 | unsigned long workload_expires; | |
cdb16e8f | 228 | struct cfq_group *serving_group; |
a36e71f9 JA |
229 | |
230 | /* | |
231 | * Each priority tree is sorted by next_request position. These | |
232 | * trees are used when determining if two or more queues are | |
233 | * interleaving requests (see cfq_close_cooperator). | |
234 | */ | |
235 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
236 | ||
22e2c507 | 237 | unsigned int busy_queues; |
ef8a41df | 238 | unsigned int busy_sync_queues; |
22e2c507 | 239 | |
53c583d2 CZ |
240 | int rq_in_driver; |
241 | int rq_in_flight[2]; | |
45333d5a AC |
242 | |
243 | /* | |
244 | * queue-depth detection | |
245 | */ | |
246 | int rq_queued; | |
25776e35 | 247 | int hw_tag; |
e459dd08 CZ |
248 | /* |
249 | * hw_tag can be | |
250 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
251 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
252 | * 0 => no NCQ | |
253 | */ | |
254 | int hw_tag_est_depth; | |
255 | unsigned int hw_tag_samples; | |
1da177e4 | 256 | |
22e2c507 JA |
257 | /* |
258 | * idle window management | |
259 | */ | |
260 | struct timer_list idle_slice_timer; | |
23e018a1 | 261 | struct work_struct unplug_work; |
1da177e4 | 262 | |
22e2c507 JA |
263 | struct cfq_queue *active_queue; |
264 | struct cfq_io_context *active_cic; | |
22e2c507 | 265 | |
c2dea2d1 VT |
266 | /* |
267 | * async queue for each priority case | |
268 | */ | |
269 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
270 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 271 | |
6d048f53 | 272 | sector_t last_position; |
1da177e4 | 273 | |
1da177e4 LT |
274 | /* |
275 | * tunables, see top of file | |
276 | */ | |
277 | unsigned int cfq_quantum; | |
22e2c507 | 278 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
279 | unsigned int cfq_back_penalty; |
280 | unsigned int cfq_back_max; | |
22e2c507 JA |
281 | unsigned int cfq_slice[2]; |
282 | unsigned int cfq_slice_async_rq; | |
283 | unsigned int cfq_slice_idle; | |
80bdf0c7 | 284 | unsigned int cfq_group_idle; |
963b72fc | 285 | unsigned int cfq_latency; |
d9ff4187 AV |
286 | |
287 | struct list_head cic_list; | |
1da177e4 | 288 | |
6118b70b JA |
289 | /* |
290 | * Fallback dummy cfqq for extreme OOM conditions | |
291 | */ | |
292 | struct cfq_queue oom_cfqq; | |
365722bb | 293 | |
573412b2 | 294 | unsigned long last_delayed_sync; |
25fb5169 VG |
295 | |
296 | /* List of cfq groups being managed on this device*/ | |
297 | struct hlist_head cfqg_list; | |
56edf7d7 VG |
298 | |
299 | /* Number of groups which are on blkcg->blkg_list */ | |
300 | unsigned int nr_blkcg_linked_grps; | |
1da177e4 LT |
301 | }; |
302 | ||
25fb5169 VG |
303 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
304 | ||
cdb16e8f VG |
305 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
306 | enum wl_prio_t prio, | |
65b32a57 | 307 | enum wl_type_t type) |
c0324a02 | 308 | { |
1fa8f6d6 VG |
309 | if (!cfqg) |
310 | return NULL; | |
311 | ||
c0324a02 | 312 | if (prio == IDLE_WORKLOAD) |
cdb16e8f | 313 | return &cfqg->service_tree_idle; |
c0324a02 | 314 | |
cdb16e8f | 315 | return &cfqg->service_trees[prio][type]; |
c0324a02 CZ |
316 | } |
317 | ||
3b18152c | 318 | enum cfqq_state_flags { |
b0b8d749 JA |
319 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
320 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 321 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 322 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
323 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
324 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
325 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 326 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 327 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 328 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 329 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 330 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 331 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
332 | }; |
333 | ||
334 | #define CFQ_CFQQ_FNS(name) \ | |
335 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
336 | { \ | |
fe094d98 | 337 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
338 | } \ |
339 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
340 | { \ | |
fe094d98 | 341 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
342 | } \ |
343 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
344 | { \ | |
fe094d98 | 345 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
346 | } |
347 | ||
348 | CFQ_CFQQ_FNS(on_rr); | |
349 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 350 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 351 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
352 | CFQ_CFQQ_FNS(fifo_expire); |
353 | CFQ_CFQQ_FNS(idle_window); | |
354 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 355 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 356 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 357 | CFQ_CFQQ_FNS(coop); |
ae54abed | 358 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 359 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 360 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
361 | #undef CFQ_CFQQ_FNS |
362 | ||
afc24d49 | 363 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2868ef7b VG |
364 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
365 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ | |
366 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
4495a7d4 | 367 | blkg_path(&(cfqq)->cfqg->blkg), ##args) |
2868ef7b VG |
368 | |
369 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ | |
370 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ | |
4495a7d4 | 371 | blkg_path(&(cfqg)->blkg), ##args) \ |
2868ef7b VG |
372 | |
373 | #else | |
7b679138 JA |
374 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
375 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
4495a7d4 | 376 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
2868ef7b | 377 | #endif |
7b679138 JA |
378 | #define cfq_log(cfqd, fmt, args...) \ |
379 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
380 | ||
615f0259 VG |
381 | /* Traverses through cfq group service trees */ |
382 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
383 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
384 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
385 | : &cfqg->service_tree_idle; \ | |
386 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
387 | (i == IDLE_WORKLOAD && j == 0); \ | |
388 | j++, st = i < IDLE_WORKLOAD ? \ | |
389 | &cfqg->service_trees[i][j]: NULL) \ | |
390 | ||
f5f2b6ce SL |
391 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
392 | struct cfq_ttime *ttime, bool group_idle) | |
393 | { | |
394 | unsigned long slice; | |
395 | if (!sample_valid(ttime->ttime_samples)) | |
396 | return false; | |
397 | if (group_idle) | |
398 | slice = cfqd->cfq_group_idle; | |
399 | else | |
400 | slice = cfqd->cfq_slice_idle; | |
401 | return ttime->ttime_mean > slice; | |
402 | } | |
615f0259 | 403 | |
02b35081 VG |
404 | static inline bool iops_mode(struct cfq_data *cfqd) |
405 | { | |
406 | /* | |
407 | * If we are not idling on queues and it is a NCQ drive, parallel | |
408 | * execution of requests is on and measuring time is not possible | |
409 | * in most of the cases until and unless we drive shallower queue | |
410 | * depths and that becomes a performance bottleneck. In such cases | |
411 | * switch to start providing fairness in terms of number of IOs. | |
412 | */ | |
413 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
414 | return true; | |
415 | else | |
416 | return false; | |
417 | } | |
418 | ||
c0324a02 CZ |
419 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
420 | { | |
421 | if (cfq_class_idle(cfqq)) | |
422 | return IDLE_WORKLOAD; | |
423 | if (cfq_class_rt(cfqq)) | |
424 | return RT_WORKLOAD; | |
425 | return BE_WORKLOAD; | |
426 | } | |
427 | ||
718eee05 CZ |
428 | |
429 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
430 | { | |
431 | if (!cfq_cfqq_sync(cfqq)) | |
432 | return ASYNC_WORKLOAD; | |
433 | if (!cfq_cfqq_idle_window(cfqq)) | |
434 | return SYNC_NOIDLE_WORKLOAD; | |
435 | return SYNC_WORKLOAD; | |
436 | } | |
437 | ||
58ff82f3 VG |
438 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
439 | struct cfq_data *cfqd, | |
440 | struct cfq_group *cfqg) | |
c0324a02 CZ |
441 | { |
442 | if (wl == IDLE_WORKLOAD) | |
cdb16e8f | 443 | return cfqg->service_tree_idle.count; |
c0324a02 | 444 | |
cdb16e8f VG |
445 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
446 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count | |
447 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
448 | } |
449 | ||
f26bd1f0 VG |
450 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
451 | struct cfq_group *cfqg) | |
452 | { | |
453 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count | |
454 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
455 | } | |
456 | ||
165125e1 | 457 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
a6151c3a | 458 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool, |
fd0928df | 459 | struct io_context *, gfp_t); |
4ac845a2 | 460 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
91fac317 VT |
461 | struct io_context *); |
462 | ||
463 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, | |
a6151c3a | 464 | bool is_sync) |
91fac317 | 465 | { |
a6151c3a | 466 | return cic->cfqq[is_sync]; |
91fac317 VT |
467 | } |
468 | ||
469 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
a6151c3a | 470 | struct cfq_queue *cfqq, bool is_sync) |
91fac317 | 471 | { |
a6151c3a | 472 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
473 | } |
474 | ||
bca4b914 KK |
475 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_context *cic) |
476 | { | |
1238033c | 477 | return cic->q->elevator->elevator_data; |
bca4b914 KK |
478 | } |
479 | ||
91fac317 VT |
480 | /* |
481 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
482 | * set (in which case it could also be direct WRITE). | |
483 | */ | |
a6151c3a | 484 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 485 | { |
7b6d91da | 486 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 487 | } |
1da177e4 | 488 | |
99f95e52 AM |
489 | /* |
490 | * scheduler run of queue, if there are requests pending and no one in the | |
491 | * driver that will restart queueing | |
492 | */ | |
23e018a1 | 493 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 494 | { |
7b679138 JA |
495 | if (cfqd->busy_queues) { |
496 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 497 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 498 | } |
99f95e52 AM |
499 | } |
500 | ||
44f7c160 JA |
501 | /* |
502 | * Scale schedule slice based on io priority. Use the sync time slice only | |
503 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
504 | * io only, should not get full sync slice length. | |
505 | */ | |
a6151c3a | 506 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 507 | unsigned short prio) |
44f7c160 | 508 | { |
d9e7620e | 509 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 510 | |
d9e7620e JA |
511 | WARN_ON(prio >= IOPRIO_BE_NR); |
512 | ||
513 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
514 | } | |
44f7c160 | 515 | |
d9e7620e JA |
516 | static inline int |
517 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
518 | { | |
519 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
520 | } |
521 | ||
25bc6b07 VG |
522 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
523 | { | |
524 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
525 | ||
526 | d = d * BLKIO_WEIGHT_DEFAULT; | |
527 | do_div(d, cfqg->weight); | |
528 | return d; | |
529 | } | |
530 | ||
531 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
532 | { | |
533 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
534 | if (delta > 0) | |
535 | min_vdisktime = vdisktime; | |
536 | ||
537 | return min_vdisktime; | |
538 | } | |
539 | ||
540 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
541 | { | |
542 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
543 | if (delta < 0) | |
544 | min_vdisktime = vdisktime; | |
545 | ||
546 | return min_vdisktime; | |
547 | } | |
548 | ||
549 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
550 | { | |
25bc6b07 VG |
551 | struct cfq_group *cfqg; |
552 | ||
25bc6b07 VG |
553 | if (st->left) { |
554 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
555 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
556 | cfqg->vdisktime); | |
25bc6b07 | 557 | } |
25bc6b07 VG |
558 | } |
559 | ||
5db5d642 CZ |
560 | /* |
561 | * get averaged number of queues of RT/BE priority. | |
562 | * average is updated, with a formula that gives more weight to higher numbers, | |
563 | * to quickly follows sudden increases and decrease slowly | |
564 | */ | |
565 | ||
58ff82f3 VG |
566 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
567 | struct cfq_group *cfqg, bool rt) | |
5869619c | 568 | { |
5db5d642 CZ |
569 | unsigned min_q, max_q; |
570 | unsigned mult = cfq_hist_divisor - 1; | |
571 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 572 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 573 | |
58ff82f3 VG |
574 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
575 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
576 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 577 | cfq_hist_divisor; |
58ff82f3 VG |
578 | return cfqg->busy_queues_avg[rt]; |
579 | } | |
580 | ||
581 | static inline unsigned | |
582 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
583 | { | |
584 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
585 | ||
586 | return cfq_target_latency * cfqg->weight / st->total_weight; | |
5db5d642 CZ |
587 | } |
588 | ||
c553f8e3 | 589 | static inline unsigned |
ba5bd520 | 590 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 591 | { |
5db5d642 CZ |
592 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
593 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
594 | /* |
595 | * interested queues (we consider only the ones with the same | |
596 | * priority class in the cfq group) | |
597 | */ | |
598 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
599 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
600 | unsigned sync_slice = cfqd->cfq_slice[1]; |
601 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
602 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
603 | ||
604 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
605 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
606 | /* scale low_slice according to IO priority | |
607 | * and sync vs async */ | |
608 | unsigned low_slice = | |
609 | min(slice, base_low_slice * slice / sync_slice); | |
610 | /* the adapted slice value is scaled to fit all iqs | |
611 | * into the target latency */ | |
58ff82f3 | 612 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
613 | low_slice); |
614 | } | |
615 | } | |
c553f8e3 SL |
616 | return slice; |
617 | } | |
618 | ||
619 | static inline void | |
620 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
621 | { | |
ba5bd520 | 622 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 623 | |
dae739eb | 624 | cfqq->slice_start = jiffies; |
5db5d642 | 625 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 626 | cfqq->allocated_slice = slice; |
7b679138 | 627 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
628 | } |
629 | ||
630 | /* | |
631 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
632 | * isn't valid until the first request from the dispatch is activated | |
633 | * and the slice time set. | |
634 | */ | |
a6151c3a | 635 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
636 | { |
637 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 638 | return false; |
44f7c160 | 639 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 640 | return false; |
44f7c160 | 641 | |
c1e44756 | 642 | return true; |
44f7c160 JA |
643 | } |
644 | ||
1da177e4 | 645 | /* |
5e705374 | 646 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 647 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 648 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 649 | */ |
5e705374 | 650 | static struct request * |
cf7c25cf | 651 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 652 | { |
cf7c25cf | 653 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 654 | unsigned long back_max; |
e8a99053 AM |
655 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
656 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
657 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 658 | |
5e705374 JA |
659 | if (rq1 == NULL || rq1 == rq2) |
660 | return rq2; | |
661 | if (rq2 == NULL) | |
662 | return rq1; | |
9c2c38a1 | 663 | |
229836bd NK |
664 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
665 | return rq_is_sync(rq1) ? rq1 : rq2; | |
666 | ||
65299a3b CH |
667 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
668 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 669 | |
83096ebf TH |
670 | s1 = blk_rq_pos(rq1); |
671 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 672 | |
1da177e4 LT |
673 | /* |
674 | * by definition, 1KiB is 2 sectors | |
675 | */ | |
676 | back_max = cfqd->cfq_back_max * 2; | |
677 | ||
678 | /* | |
679 | * Strict one way elevator _except_ in the case where we allow | |
680 | * short backward seeks which are biased as twice the cost of a | |
681 | * similar forward seek. | |
682 | */ | |
683 | if (s1 >= last) | |
684 | d1 = s1 - last; | |
685 | else if (s1 + back_max >= last) | |
686 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
687 | else | |
e8a99053 | 688 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
689 | |
690 | if (s2 >= last) | |
691 | d2 = s2 - last; | |
692 | else if (s2 + back_max >= last) | |
693 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
694 | else | |
e8a99053 | 695 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
696 | |
697 | /* Found required data */ | |
e8a99053 AM |
698 | |
699 | /* | |
700 | * By doing switch() on the bit mask "wrap" we avoid having to | |
701 | * check two variables for all permutations: --> faster! | |
702 | */ | |
703 | switch (wrap) { | |
5e705374 | 704 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 705 | if (d1 < d2) |
5e705374 | 706 | return rq1; |
e8a99053 | 707 | else if (d2 < d1) |
5e705374 | 708 | return rq2; |
e8a99053 AM |
709 | else { |
710 | if (s1 >= s2) | |
5e705374 | 711 | return rq1; |
e8a99053 | 712 | else |
5e705374 | 713 | return rq2; |
e8a99053 | 714 | } |
1da177e4 | 715 | |
e8a99053 | 716 | case CFQ_RQ2_WRAP: |
5e705374 | 717 | return rq1; |
e8a99053 | 718 | case CFQ_RQ1_WRAP: |
5e705374 JA |
719 | return rq2; |
720 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
721 | default: |
722 | /* | |
723 | * Since both rqs are wrapped, | |
724 | * start with the one that's further behind head | |
725 | * (--> only *one* back seek required), | |
726 | * since back seek takes more time than forward. | |
727 | */ | |
728 | if (s1 <= s2) | |
5e705374 | 729 | return rq1; |
1da177e4 | 730 | else |
5e705374 | 731 | return rq2; |
1da177e4 LT |
732 | } |
733 | } | |
734 | ||
498d3aa2 JA |
735 | /* |
736 | * The below is leftmost cache rbtree addon | |
737 | */ | |
0871714e | 738 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 739 | { |
615f0259 VG |
740 | /* Service tree is empty */ |
741 | if (!root->count) | |
742 | return NULL; | |
743 | ||
cc09e299 JA |
744 | if (!root->left) |
745 | root->left = rb_first(&root->rb); | |
746 | ||
0871714e JA |
747 | if (root->left) |
748 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
749 | ||
750 | return NULL; | |
cc09e299 JA |
751 | } |
752 | ||
1fa8f6d6 VG |
753 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
754 | { | |
755 | if (!root->left) | |
756 | root->left = rb_first(&root->rb); | |
757 | ||
758 | if (root->left) | |
759 | return rb_entry_cfqg(root->left); | |
760 | ||
761 | return NULL; | |
762 | } | |
763 | ||
a36e71f9 JA |
764 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
765 | { | |
766 | rb_erase(n, root); | |
767 | RB_CLEAR_NODE(n); | |
768 | } | |
769 | ||
cc09e299 JA |
770 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
771 | { | |
772 | if (root->left == n) | |
773 | root->left = NULL; | |
a36e71f9 | 774 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 775 | --root->count; |
cc09e299 JA |
776 | } |
777 | ||
1da177e4 LT |
778 | /* |
779 | * would be nice to take fifo expire time into account as well | |
780 | */ | |
5e705374 JA |
781 | static struct request * |
782 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
783 | struct request *last) | |
1da177e4 | 784 | { |
21183b07 JA |
785 | struct rb_node *rbnext = rb_next(&last->rb_node); |
786 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 787 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 788 | |
21183b07 | 789 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
790 | |
791 | if (rbprev) | |
5e705374 | 792 | prev = rb_entry_rq(rbprev); |
1da177e4 | 793 | |
21183b07 | 794 | if (rbnext) |
5e705374 | 795 | next = rb_entry_rq(rbnext); |
21183b07 JA |
796 | else { |
797 | rbnext = rb_first(&cfqq->sort_list); | |
798 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 799 | next = rb_entry_rq(rbnext); |
21183b07 | 800 | } |
1da177e4 | 801 | |
cf7c25cf | 802 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
803 | } |
804 | ||
d9e7620e JA |
805 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
806 | struct cfq_queue *cfqq) | |
1da177e4 | 807 | { |
d9e7620e JA |
808 | /* |
809 | * just an approximation, should be ok. | |
810 | */ | |
cdb16e8f | 811 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 812 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
813 | } |
814 | ||
1fa8f6d6 VG |
815 | static inline s64 |
816 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
817 | { | |
818 | return cfqg->vdisktime - st->min_vdisktime; | |
819 | } | |
820 | ||
821 | static void | |
822 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
823 | { | |
824 | struct rb_node **node = &st->rb.rb_node; | |
825 | struct rb_node *parent = NULL; | |
826 | struct cfq_group *__cfqg; | |
827 | s64 key = cfqg_key(st, cfqg); | |
828 | int left = 1; | |
829 | ||
830 | while (*node != NULL) { | |
831 | parent = *node; | |
832 | __cfqg = rb_entry_cfqg(parent); | |
833 | ||
834 | if (key < cfqg_key(st, __cfqg)) | |
835 | node = &parent->rb_left; | |
836 | else { | |
837 | node = &parent->rb_right; | |
838 | left = 0; | |
839 | } | |
840 | } | |
841 | ||
842 | if (left) | |
843 | st->left = &cfqg->rb_node; | |
844 | ||
845 | rb_link_node(&cfqg->rb_node, parent, node); | |
846 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
847 | } | |
848 | ||
849 | static void | |
8184f93e JT |
850 | cfq_update_group_weight(struct cfq_group *cfqg) |
851 | { | |
852 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
853 | if (cfqg->needs_update) { | |
854 | cfqg->weight = cfqg->new_weight; | |
855 | cfqg->needs_update = false; | |
856 | } | |
857 | } | |
858 | ||
859 | static void | |
860 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
861 | { | |
862 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
863 | ||
864 | cfq_update_group_weight(cfqg); | |
865 | __cfq_group_service_tree_add(st, cfqg); | |
866 | st->total_weight += cfqg->weight; | |
867 | } | |
868 | ||
869 | static void | |
870 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
871 | { |
872 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
873 | struct cfq_group *__cfqg; | |
874 | struct rb_node *n; | |
875 | ||
876 | cfqg->nr_cfqq++; | |
760701bf | 877 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
878 | return; |
879 | ||
880 | /* | |
881 | * Currently put the group at the end. Later implement something | |
882 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 883 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
884 | */ |
885 | n = rb_last(&st->rb); | |
886 | if (n) { | |
887 | __cfqg = rb_entry_cfqg(n); | |
888 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
889 | } else | |
890 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
891 | cfq_group_service_tree_add(st, cfqg); |
892 | } | |
1fa8f6d6 | 893 | |
8184f93e JT |
894 | static void |
895 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
896 | { | |
897 | st->total_weight -= cfqg->weight; | |
898 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) | |
899 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
900 | } |
901 | ||
902 | static void | |
8184f93e | 903 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
904 | { |
905 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
906 | ||
907 | BUG_ON(cfqg->nr_cfqq < 1); | |
908 | cfqg->nr_cfqq--; | |
25bc6b07 | 909 | |
1fa8f6d6 VG |
910 | /* If there are other cfq queues under this group, don't delete it */ |
911 | if (cfqg->nr_cfqq) | |
912 | return; | |
913 | ||
2868ef7b | 914 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 915 | cfq_group_service_tree_del(st, cfqg); |
dae739eb | 916 | cfqg->saved_workload_slice = 0; |
e98ef89b | 917 | cfq_blkiocg_update_dequeue_stats(&cfqg->blkg, 1); |
dae739eb VG |
918 | } |
919 | ||
167400d3 JT |
920 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
921 | unsigned int *unaccounted_time) | |
dae739eb | 922 | { |
f75edf2d | 923 | unsigned int slice_used; |
dae739eb VG |
924 | |
925 | /* | |
926 | * Queue got expired before even a single request completed or | |
927 | * got expired immediately after first request completion. | |
928 | */ | |
929 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
930 | /* | |
931 | * Also charge the seek time incurred to the group, otherwise | |
932 | * if there are mutiple queues in the group, each can dispatch | |
933 | * a single request on seeky media and cause lots of seek time | |
934 | * and group will never know it. | |
935 | */ | |
936 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
937 | 1); | |
938 | } else { | |
939 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
940 | if (slice_used > cfqq->allocated_slice) { |
941 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 942 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
943 | } |
944 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
945 | *unaccounted_time += cfqq->slice_start - | |
946 | cfqq->dispatch_start; | |
dae739eb VG |
947 | } |
948 | ||
dae739eb VG |
949 | return slice_used; |
950 | } | |
951 | ||
952 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 953 | struct cfq_queue *cfqq) |
dae739eb VG |
954 | { |
955 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 956 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
957 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
958 | - cfqg->service_tree_idle.count; | |
959 | ||
960 | BUG_ON(nr_sync < 0); | |
167400d3 | 961 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 962 | |
02b35081 VG |
963 | if (iops_mode(cfqd)) |
964 | charge = cfqq->slice_dispatch; | |
965 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
966 | charge = cfqq->allocated_slice; | |
dae739eb VG |
967 | |
968 | /* Can't update vdisktime while group is on service tree */ | |
8184f93e | 969 | cfq_group_service_tree_del(st, cfqg); |
02b35081 | 970 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
8184f93e JT |
971 | /* If a new weight was requested, update now, off tree */ |
972 | cfq_group_service_tree_add(st, cfqg); | |
dae739eb VG |
973 | |
974 | /* This group is being expired. Save the context */ | |
975 | if (time_after(cfqd->workload_expires, jiffies)) { | |
976 | cfqg->saved_workload_slice = cfqd->workload_expires | |
977 | - jiffies; | |
978 | cfqg->saved_workload = cfqd->serving_type; | |
979 | cfqg->saved_serving_prio = cfqd->serving_prio; | |
980 | } else | |
981 | cfqg->saved_workload_slice = 0; | |
2868ef7b VG |
982 | |
983 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
984 | st->min_vdisktime); | |
fd16d263 JP |
985 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
986 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
987 | used_sl, cfqq->slice_dispatch, charge, | |
988 | iops_mode(cfqd), cfqq->nr_sectors); | |
167400d3 JT |
989 | cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl, |
990 | unaccounted_sl); | |
e98ef89b | 991 | cfq_blkiocg_set_start_empty_time(&cfqg->blkg); |
1fa8f6d6 VG |
992 | } |
993 | ||
25fb5169 VG |
994 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
995 | static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg) | |
996 | { | |
997 | if (blkg) | |
998 | return container_of(blkg, struct cfq_group, blkg); | |
999 | return NULL; | |
1000 | } | |
1001 | ||
8aea4545 PB |
1002 | static void cfq_update_blkio_group_weight(void *key, struct blkio_group *blkg, |
1003 | unsigned int weight) | |
f8d461d6 | 1004 | { |
8184f93e JT |
1005 | struct cfq_group *cfqg = cfqg_of_blkg(blkg); |
1006 | cfqg->new_weight = weight; | |
1007 | cfqg->needs_update = true; | |
f8d461d6 VG |
1008 | } |
1009 | ||
f469a7b4 VG |
1010 | static void cfq_init_add_cfqg_lists(struct cfq_data *cfqd, |
1011 | struct cfq_group *cfqg, struct blkio_cgroup *blkcg) | |
25fb5169 | 1012 | { |
22084190 VG |
1013 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; |
1014 | unsigned int major, minor; | |
25fb5169 | 1015 | |
f469a7b4 VG |
1016 | /* |
1017 | * Add group onto cgroup list. It might happen that bdi->dev is | |
1018 | * not initialized yet. Initialize this new group without major | |
1019 | * and minor info and this info will be filled in once a new thread | |
1020 | * comes for IO. | |
1021 | */ | |
1022 | if (bdi->dev) { | |
a74b2ada | 1023 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); |
f469a7b4 VG |
1024 | cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, |
1025 | (void *)cfqd, MKDEV(major, minor)); | |
1026 | } else | |
1027 | cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, | |
1028 | (void *)cfqd, 0); | |
1029 | ||
1030 | cfqd->nr_blkcg_linked_grps++; | |
1031 | cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev); | |
1032 | ||
1033 | /* Add group on cfqd list */ | |
1034 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); | |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * Should be called from sleepable context. No request queue lock as per | |
1039 | * cpu stats are allocated dynamically and alloc_percpu needs to be called | |
1040 | * from sleepable context. | |
1041 | */ | |
1042 | static struct cfq_group * cfq_alloc_cfqg(struct cfq_data *cfqd) | |
1043 | { | |
1044 | struct cfq_group *cfqg = NULL; | |
5624a4e4 | 1045 | int i, j, ret; |
f469a7b4 | 1046 | struct cfq_rb_root *st; |
25fb5169 VG |
1047 | |
1048 | cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node); | |
1049 | if (!cfqg) | |
f469a7b4 | 1050 | return NULL; |
25fb5169 | 1051 | |
25fb5169 VG |
1052 | for_each_cfqg_st(cfqg, i, j, st) |
1053 | *st = CFQ_RB_ROOT; | |
1054 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1055 | ||
7700fc4f SL |
1056 | cfqg->ttime.last_end_request = jiffies; |
1057 | ||
b1c35769 VG |
1058 | /* |
1059 | * Take the initial reference that will be released on destroy | |
1060 | * This can be thought of a joint reference by cgroup and | |
1061 | * elevator which will be dropped by either elevator exit | |
1062 | * or cgroup deletion path depending on who is exiting first. | |
1063 | */ | |
329a6781 | 1064 | cfqg->ref = 1; |
5624a4e4 VG |
1065 | |
1066 | ret = blkio_alloc_blkg_stats(&cfqg->blkg); | |
1067 | if (ret) { | |
1068 | kfree(cfqg); | |
1069 | return NULL; | |
1070 | } | |
1071 | ||
f469a7b4 VG |
1072 | return cfqg; |
1073 | } | |
1074 | ||
1075 | static struct cfq_group * | |
1076 | cfq_find_cfqg(struct cfq_data *cfqd, struct blkio_cgroup *blkcg) | |
1077 | { | |
1078 | struct cfq_group *cfqg = NULL; | |
1079 | void *key = cfqd; | |
1080 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; | |
1081 | unsigned int major, minor; | |
b1c35769 | 1082 | |
180be2a0 | 1083 | /* |
f469a7b4 VG |
1084 | * This is the common case when there are no blkio cgroups. |
1085 | * Avoid lookup in this case | |
180be2a0 | 1086 | */ |
f469a7b4 VG |
1087 | if (blkcg == &blkio_root_cgroup) |
1088 | cfqg = &cfqd->root_group; | |
1089 | else | |
1090 | cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key)); | |
25fb5169 | 1091 | |
f469a7b4 VG |
1092 | if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) { |
1093 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); | |
1094 | cfqg->blkg.dev = MKDEV(major, minor); | |
1095 | } | |
25fb5169 | 1096 | |
25fb5169 VG |
1097 | return cfqg; |
1098 | } | |
1099 | ||
1100 | /* | |
3e59cf9d VG |
1101 | * Search for the cfq group current task belongs to. request_queue lock must |
1102 | * be held. | |
25fb5169 | 1103 | */ |
3e59cf9d | 1104 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd) |
25fb5169 | 1105 | { |
70087dc3 | 1106 | struct blkio_cgroup *blkcg; |
f469a7b4 VG |
1107 | struct cfq_group *cfqg = NULL, *__cfqg = NULL; |
1108 | struct request_queue *q = cfqd->queue; | |
25fb5169 VG |
1109 | |
1110 | rcu_read_lock(); | |
70087dc3 | 1111 | blkcg = task_blkio_cgroup(current); |
f469a7b4 VG |
1112 | cfqg = cfq_find_cfqg(cfqd, blkcg); |
1113 | if (cfqg) { | |
1114 | rcu_read_unlock(); | |
1115 | return cfqg; | |
1116 | } | |
1117 | ||
1118 | /* | |
1119 | * Need to allocate a group. Allocation of group also needs allocation | |
1120 | * of per cpu stats which in-turn takes a mutex() and can block. Hence | |
1121 | * we need to drop rcu lock and queue_lock before we call alloc. | |
1122 | * | |
1123 | * Not taking any queue reference here and assuming that queue is | |
1124 | * around by the time we return. CFQ queue allocation code does | |
1125 | * the same. It might be racy though. | |
1126 | */ | |
1127 | ||
1128 | rcu_read_unlock(); | |
1129 | spin_unlock_irq(q->queue_lock); | |
1130 | ||
1131 | cfqg = cfq_alloc_cfqg(cfqd); | |
1132 | ||
1133 | spin_lock_irq(q->queue_lock); | |
1134 | ||
1135 | rcu_read_lock(); | |
1136 | blkcg = task_blkio_cgroup(current); | |
1137 | ||
1138 | /* | |
1139 | * If some other thread already allocated the group while we were | |
1140 | * not holding queue lock, free up the group | |
1141 | */ | |
1142 | __cfqg = cfq_find_cfqg(cfqd, blkcg); | |
1143 | ||
1144 | if (__cfqg) { | |
1145 | kfree(cfqg); | |
1146 | rcu_read_unlock(); | |
1147 | return __cfqg; | |
1148 | } | |
1149 | ||
3e59cf9d | 1150 | if (!cfqg) |
25fb5169 | 1151 | cfqg = &cfqd->root_group; |
f469a7b4 VG |
1152 | |
1153 | cfq_init_add_cfqg_lists(cfqd, cfqg, blkcg); | |
25fb5169 VG |
1154 | rcu_read_unlock(); |
1155 | return cfqg; | |
1156 | } | |
1157 | ||
7f1dc8a2 VG |
1158 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) |
1159 | { | |
329a6781 | 1160 | cfqg->ref++; |
7f1dc8a2 VG |
1161 | return cfqg; |
1162 | } | |
1163 | ||
25fb5169 VG |
1164 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) |
1165 | { | |
1166 | /* Currently, all async queues are mapped to root group */ | |
1167 | if (!cfq_cfqq_sync(cfqq)) | |
1168 | cfqg = &cfqq->cfqd->root_group; | |
1169 | ||
1170 | cfqq->cfqg = cfqg; | |
b1c35769 | 1171 | /* cfqq reference on cfqg */ |
329a6781 | 1172 | cfqq->cfqg->ref++; |
b1c35769 VG |
1173 | } |
1174 | ||
1175 | static void cfq_put_cfqg(struct cfq_group *cfqg) | |
1176 | { | |
1177 | struct cfq_rb_root *st; | |
1178 | int i, j; | |
1179 | ||
329a6781 SL |
1180 | BUG_ON(cfqg->ref <= 0); |
1181 | cfqg->ref--; | |
1182 | if (cfqg->ref) | |
b1c35769 VG |
1183 | return; |
1184 | for_each_cfqg_st(cfqg, i, j, st) | |
b54ce60e | 1185 | BUG_ON(!RB_EMPTY_ROOT(&st->rb)); |
5624a4e4 | 1186 | free_percpu(cfqg->blkg.stats_cpu); |
b1c35769 VG |
1187 | kfree(cfqg); |
1188 | } | |
1189 | ||
1190 | static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1191 | { | |
1192 | /* Something wrong if we are trying to remove same group twice */ | |
1193 | BUG_ON(hlist_unhashed(&cfqg->cfqd_node)); | |
1194 | ||
1195 | hlist_del_init(&cfqg->cfqd_node); | |
1196 | ||
a5395b83 VG |
1197 | BUG_ON(cfqd->nr_blkcg_linked_grps <= 0); |
1198 | cfqd->nr_blkcg_linked_grps--; | |
1199 | ||
b1c35769 VG |
1200 | /* |
1201 | * Put the reference taken at the time of creation so that when all | |
1202 | * queues are gone, group can be destroyed. | |
1203 | */ | |
1204 | cfq_put_cfqg(cfqg); | |
1205 | } | |
1206 | ||
1207 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) | |
1208 | { | |
1209 | struct hlist_node *pos, *n; | |
1210 | struct cfq_group *cfqg; | |
1211 | ||
1212 | hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) { | |
1213 | /* | |
1214 | * If cgroup removal path got to blk_group first and removed | |
1215 | * it from cgroup list, then it will take care of destroying | |
1216 | * cfqg also. | |
1217 | */ | |
e98ef89b | 1218 | if (!cfq_blkiocg_del_blkio_group(&cfqg->blkg)) |
b1c35769 VG |
1219 | cfq_destroy_cfqg(cfqd, cfqg); |
1220 | } | |
25fb5169 | 1221 | } |
b1c35769 VG |
1222 | |
1223 | /* | |
1224 | * Blk cgroup controller notification saying that blkio_group object is being | |
1225 | * delinked as associated cgroup object is going away. That also means that | |
1226 | * no new IO will come in this group. So get rid of this group as soon as | |
1227 | * any pending IO in the group is finished. | |
1228 | * | |
1229 | * This function is called under rcu_read_lock(). key is the rcu protected | |
1230 | * pointer. That means "key" is a valid cfq_data pointer as long as we are rcu | |
1231 | * read lock. | |
1232 | * | |
1233 | * "key" was fetched from blkio_group under blkio_cgroup->lock. That means | |
1234 | * it should not be NULL as even if elevator was exiting, cgroup deltion | |
1235 | * path got to it first. | |
1236 | */ | |
8aea4545 | 1237 | static void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg) |
b1c35769 VG |
1238 | { |
1239 | unsigned long flags; | |
1240 | struct cfq_data *cfqd = key; | |
1241 | ||
1242 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1243 | cfq_destroy_cfqg(cfqd, cfqg_of_blkg(blkg)); | |
1244 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1245 | } | |
1246 | ||
25fb5169 | 1247 | #else /* GROUP_IOSCHED */ |
3e59cf9d | 1248 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd) |
25fb5169 VG |
1249 | { |
1250 | return &cfqd->root_group; | |
1251 | } | |
7f1dc8a2 VG |
1252 | |
1253 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) | |
1254 | { | |
50eaeb32 | 1255 | return cfqg; |
7f1dc8a2 VG |
1256 | } |
1257 | ||
25fb5169 VG |
1258 | static inline void |
1259 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1260 | cfqq->cfqg = cfqg; | |
1261 | } | |
1262 | ||
b1c35769 VG |
1263 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) {} |
1264 | static inline void cfq_put_cfqg(struct cfq_group *cfqg) {} | |
1265 | ||
25fb5169 VG |
1266 | #endif /* GROUP_IOSCHED */ |
1267 | ||
498d3aa2 | 1268 | /* |
c0324a02 | 1269 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1270 | * requests waiting to be processed. It is sorted in the order that |
1271 | * we will service the queues. | |
1272 | */ | |
a36e71f9 | 1273 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1274 | bool add_front) |
d9e7620e | 1275 | { |
0871714e JA |
1276 | struct rb_node **p, *parent; |
1277 | struct cfq_queue *__cfqq; | |
d9e7620e | 1278 | unsigned long rb_key; |
c0324a02 | 1279 | struct cfq_rb_root *service_tree; |
498d3aa2 | 1280 | int left; |
dae739eb | 1281 | int new_cfqq = 1; |
ae30c286 | 1282 | |
cdb16e8f | 1283 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
65b32a57 | 1284 | cfqq_type(cfqq)); |
0871714e JA |
1285 | if (cfq_class_idle(cfqq)) { |
1286 | rb_key = CFQ_IDLE_DELAY; | |
aa6f6a3d | 1287 | parent = rb_last(&service_tree->rb); |
0871714e JA |
1288 | if (parent && parent != &cfqq->rb_node) { |
1289 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1290 | rb_key += __cfqq->rb_key; | |
1291 | } else | |
1292 | rb_key += jiffies; | |
1293 | } else if (!add_front) { | |
b9c8946b JA |
1294 | /* |
1295 | * Get our rb key offset. Subtract any residual slice | |
1296 | * value carried from last service. A negative resid | |
1297 | * count indicates slice overrun, and this should position | |
1298 | * the next service time further away in the tree. | |
1299 | */ | |
edd75ffd | 1300 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1301 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1302 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1303 | } else { |
1304 | rb_key = -HZ; | |
aa6f6a3d | 1305 | __cfqq = cfq_rb_first(service_tree); |
48e025e6 CZ |
1306 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1307 | } | |
1da177e4 | 1308 | |
d9e7620e | 1309 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1310 | new_cfqq = 0; |
99f9628a | 1311 | /* |
d9e7620e | 1312 | * same position, nothing more to do |
99f9628a | 1313 | */ |
c0324a02 CZ |
1314 | if (rb_key == cfqq->rb_key && |
1315 | cfqq->service_tree == service_tree) | |
d9e7620e | 1316 | return; |
1da177e4 | 1317 | |
aa6f6a3d CZ |
1318 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1319 | cfqq->service_tree = NULL; | |
1da177e4 | 1320 | } |
d9e7620e | 1321 | |
498d3aa2 | 1322 | left = 1; |
0871714e | 1323 | parent = NULL; |
aa6f6a3d CZ |
1324 | cfqq->service_tree = service_tree; |
1325 | p = &service_tree->rb.rb_node; | |
d9e7620e | 1326 | while (*p) { |
67060e37 | 1327 | struct rb_node **n; |
cc09e299 | 1328 | |
d9e7620e JA |
1329 | parent = *p; |
1330 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1331 | ||
0c534e0a | 1332 | /* |
c0324a02 | 1333 | * sort by key, that represents service time. |
0c534e0a | 1334 | */ |
c0324a02 | 1335 | if (time_before(rb_key, __cfqq->rb_key)) |
67060e37 | 1336 | n = &(*p)->rb_left; |
c0324a02 | 1337 | else { |
67060e37 | 1338 | n = &(*p)->rb_right; |
cc09e299 | 1339 | left = 0; |
c0324a02 | 1340 | } |
67060e37 JA |
1341 | |
1342 | p = n; | |
d9e7620e JA |
1343 | } |
1344 | ||
cc09e299 | 1345 | if (left) |
aa6f6a3d | 1346 | service_tree->left = &cfqq->rb_node; |
cc09e299 | 1347 | |
d9e7620e JA |
1348 | cfqq->rb_key = rb_key; |
1349 | rb_link_node(&cfqq->rb_node, parent, p); | |
aa6f6a3d CZ |
1350 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
1351 | service_tree->count++; | |
20359f27 | 1352 | if (add_front || !new_cfqq) |
dae739eb | 1353 | return; |
8184f93e | 1354 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1355 | } |
1356 | ||
a36e71f9 | 1357 | static struct cfq_queue * |
f2d1f0ae JA |
1358 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1359 | sector_t sector, struct rb_node **ret_parent, | |
1360 | struct rb_node ***rb_link) | |
a36e71f9 | 1361 | { |
a36e71f9 JA |
1362 | struct rb_node **p, *parent; |
1363 | struct cfq_queue *cfqq = NULL; | |
1364 | ||
1365 | parent = NULL; | |
1366 | p = &root->rb_node; | |
1367 | while (*p) { | |
1368 | struct rb_node **n; | |
1369 | ||
1370 | parent = *p; | |
1371 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1372 | ||
1373 | /* | |
1374 | * Sort strictly based on sector. Smallest to the left, | |
1375 | * largest to the right. | |
1376 | */ | |
2e46e8b2 | 1377 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1378 | n = &(*p)->rb_right; |
2e46e8b2 | 1379 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1380 | n = &(*p)->rb_left; |
1381 | else | |
1382 | break; | |
1383 | p = n; | |
3ac6c9f8 | 1384 | cfqq = NULL; |
a36e71f9 JA |
1385 | } |
1386 | ||
1387 | *ret_parent = parent; | |
1388 | if (rb_link) | |
1389 | *rb_link = p; | |
3ac6c9f8 | 1390 | return cfqq; |
a36e71f9 JA |
1391 | } |
1392 | ||
1393 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1394 | { | |
a36e71f9 JA |
1395 | struct rb_node **p, *parent; |
1396 | struct cfq_queue *__cfqq; | |
1397 | ||
f2d1f0ae JA |
1398 | if (cfqq->p_root) { |
1399 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1400 | cfqq->p_root = NULL; | |
1401 | } | |
a36e71f9 JA |
1402 | |
1403 | if (cfq_class_idle(cfqq)) | |
1404 | return; | |
1405 | if (!cfqq->next_rq) | |
1406 | return; | |
1407 | ||
f2d1f0ae | 1408 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1409 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1410 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1411 | if (!__cfqq) { |
1412 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1413 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1414 | } else | |
1415 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1416 | } |
1417 | ||
498d3aa2 JA |
1418 | /* |
1419 | * Update cfqq's position in the service tree. | |
1420 | */ | |
edd75ffd | 1421 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1422 | { |
6d048f53 JA |
1423 | /* |
1424 | * Resorting requires the cfqq to be on the RR list already. | |
1425 | */ | |
a36e71f9 | 1426 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1427 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1428 | cfq_prio_tree_add(cfqd, cfqq); |
1429 | } | |
6d048f53 JA |
1430 | } |
1431 | ||
1da177e4 LT |
1432 | /* |
1433 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1434 | * the pending list according to last request service |
1da177e4 | 1435 | */ |
febffd61 | 1436 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1437 | { |
7b679138 | 1438 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1439 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1440 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 1441 | cfqd->busy_queues++; |
ef8a41df SL |
1442 | if (cfq_cfqq_sync(cfqq)) |
1443 | cfqd->busy_sync_queues++; | |
1da177e4 | 1444 | |
edd75ffd | 1445 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1446 | } |
1447 | ||
498d3aa2 JA |
1448 | /* |
1449 | * Called when the cfqq no longer has requests pending, remove it from | |
1450 | * the service tree. | |
1451 | */ | |
febffd61 | 1452 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1453 | { |
7b679138 | 1454 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1455 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1456 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1457 | |
aa6f6a3d CZ |
1458 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1459 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1460 | cfqq->service_tree = NULL; | |
1461 | } | |
f2d1f0ae JA |
1462 | if (cfqq->p_root) { |
1463 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1464 | cfqq->p_root = NULL; | |
1465 | } | |
d9e7620e | 1466 | |
8184f93e | 1467 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1468 | BUG_ON(!cfqd->busy_queues); |
1469 | cfqd->busy_queues--; | |
ef8a41df SL |
1470 | if (cfq_cfqq_sync(cfqq)) |
1471 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
1472 | } |
1473 | ||
1474 | /* | |
1475 | * rb tree support functions | |
1476 | */ | |
febffd61 | 1477 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1478 | { |
5e705374 | 1479 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1480 | const int sync = rq_is_sync(rq); |
1da177e4 | 1481 | |
b4878f24 JA |
1482 | BUG_ON(!cfqq->queued[sync]); |
1483 | cfqq->queued[sync]--; | |
1da177e4 | 1484 | |
5e705374 | 1485 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1486 | |
f04a6424 VG |
1487 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1488 | /* | |
1489 | * Queue will be deleted from service tree when we actually | |
1490 | * expire it later. Right now just remove it from prio tree | |
1491 | * as it is empty. | |
1492 | */ | |
1493 | if (cfqq->p_root) { | |
1494 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1495 | cfqq->p_root = NULL; | |
1496 | } | |
1497 | } | |
1da177e4 LT |
1498 | } |
1499 | ||
5e705374 | 1500 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1501 | { |
5e705374 | 1502 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1503 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 1504 | struct request *prev; |
1da177e4 | 1505 | |
5380a101 | 1506 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 1507 | |
796d5116 | 1508 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
1509 | |
1510 | if (!cfq_cfqq_on_rr(cfqq)) | |
1511 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1512 | |
1513 | /* | |
1514 | * check if this request is a better next-serve candidate | |
1515 | */ | |
a36e71f9 | 1516 | prev = cfqq->next_rq; |
cf7c25cf | 1517 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1518 | |
1519 | /* | |
1520 | * adjust priority tree position, if ->next_rq changes | |
1521 | */ | |
1522 | if (prev != cfqq->next_rq) | |
1523 | cfq_prio_tree_add(cfqd, cfqq); | |
1524 | ||
5044eed4 | 1525 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1526 | } |
1527 | ||
febffd61 | 1528 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1529 | { |
5380a101 JA |
1530 | elv_rb_del(&cfqq->sort_list, rq); |
1531 | cfqq->queued[rq_is_sync(rq)]--; | |
e98ef89b VG |
1532 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, |
1533 | rq_data_dir(rq), rq_is_sync(rq)); | |
5e705374 | 1534 | cfq_add_rq_rb(rq); |
e98ef89b | 1535 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, |
7f1dc8a2 VG |
1536 | &cfqq->cfqd->serving_group->blkg, rq_data_dir(rq), |
1537 | rq_is_sync(rq)); | |
1da177e4 LT |
1538 | } |
1539 | ||
206dc69b JA |
1540 | static struct request * |
1541 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1542 | { |
206dc69b | 1543 | struct task_struct *tsk = current; |
91fac317 | 1544 | struct cfq_io_context *cic; |
206dc69b | 1545 | struct cfq_queue *cfqq; |
1da177e4 | 1546 | |
4ac845a2 | 1547 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1548 | if (!cic) |
1549 | return NULL; | |
1550 | ||
1551 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
1552 | if (cfqq) { |
1553 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
1554 | ||
21183b07 | 1555 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 1556 | } |
1da177e4 | 1557 | |
1da177e4 LT |
1558 | return NULL; |
1559 | } | |
1560 | ||
165125e1 | 1561 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1562 | { |
22e2c507 | 1563 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 1564 | |
53c583d2 | 1565 | cfqd->rq_in_driver++; |
7b679138 | 1566 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 1567 | cfqd->rq_in_driver); |
25776e35 | 1568 | |
5b93629b | 1569 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
1570 | } |
1571 | ||
165125e1 | 1572 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1573 | { |
b4878f24 JA |
1574 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1575 | ||
53c583d2 CZ |
1576 | WARN_ON(!cfqd->rq_in_driver); |
1577 | cfqd->rq_in_driver--; | |
7b679138 | 1578 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 1579 | cfqd->rq_in_driver); |
1da177e4 LT |
1580 | } |
1581 | ||
b4878f24 | 1582 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 1583 | { |
5e705374 | 1584 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 1585 | |
5e705374 JA |
1586 | if (cfqq->next_rq == rq) |
1587 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 1588 | |
b4878f24 | 1589 | list_del_init(&rq->queuelist); |
5e705374 | 1590 | cfq_del_rq_rb(rq); |
374f84ac | 1591 | |
45333d5a | 1592 | cfqq->cfqd->rq_queued--; |
e98ef89b VG |
1593 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, |
1594 | rq_data_dir(rq), rq_is_sync(rq)); | |
65299a3b CH |
1595 | if (rq->cmd_flags & REQ_PRIO) { |
1596 | WARN_ON(!cfqq->prio_pending); | |
1597 | cfqq->prio_pending--; | |
b53d1ed7 | 1598 | } |
1da177e4 LT |
1599 | } |
1600 | ||
165125e1 JA |
1601 | static int cfq_merge(struct request_queue *q, struct request **req, |
1602 | struct bio *bio) | |
1da177e4 LT |
1603 | { |
1604 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1605 | struct request *__rq; | |
1da177e4 | 1606 | |
206dc69b | 1607 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 1608 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
1609 | *req = __rq; |
1610 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
1611 | } |
1612 | ||
1613 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
1614 | } |
1615 | ||
165125e1 | 1616 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 1617 | int type) |
1da177e4 | 1618 | { |
21183b07 | 1619 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 1620 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 1621 | |
5e705374 | 1622 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 1623 | } |
1da177e4 LT |
1624 | } |
1625 | ||
812d4026 DS |
1626 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
1627 | struct bio *bio) | |
1628 | { | |
e98ef89b VG |
1629 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(req))->blkg, |
1630 | bio_data_dir(bio), cfq_bio_sync(bio)); | |
812d4026 DS |
1631 | } |
1632 | ||
1da177e4 | 1633 | static void |
165125e1 | 1634 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
1635 | struct request *next) |
1636 | { | |
cf7c25cf | 1637 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1638 | /* |
1639 | * reposition in fifo if next is older than rq | |
1640 | */ | |
1641 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
30996f40 | 1642 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
22e2c507 | 1643 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
1644 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1645 | } | |
22e2c507 | 1646 | |
cf7c25cf CZ |
1647 | if (cfqq->next_rq == next) |
1648 | cfqq->next_rq = rq; | |
b4878f24 | 1649 | cfq_remove_request(next); |
e98ef89b VG |
1650 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(rq))->blkg, |
1651 | rq_data_dir(next), rq_is_sync(next)); | |
22e2c507 JA |
1652 | } |
1653 | ||
165125e1 | 1654 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
1655 | struct bio *bio) |
1656 | { | |
1657 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
91fac317 | 1658 | struct cfq_io_context *cic; |
da775265 | 1659 | struct cfq_queue *cfqq; |
da775265 JA |
1660 | |
1661 | /* | |
ec8acb69 | 1662 | * Disallow merge of a sync bio into an async request. |
da775265 | 1663 | */ |
91fac317 | 1664 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 1665 | return false; |
da775265 JA |
1666 | |
1667 | /* | |
f1a4f4d3 TH |
1668 | * Lookup the cfqq that this bio will be queued with and allow |
1669 | * merge only if rq is queued there. This function can be called | |
1670 | * from plug merge without queue_lock. In such cases, ioc of @rq | |
1671 | * and %current are guaranteed to be equal. Avoid lookup which | |
1672 | * requires queue_lock by using @rq's cic. | |
1673 | */ | |
1674 | if (current->io_context == RQ_CIC(rq)->ioc) { | |
1675 | cic = RQ_CIC(rq); | |
1676 | } else { | |
1677 | cic = cfq_cic_lookup(cfqd, current->io_context); | |
1678 | if (!cic) | |
1679 | return false; | |
1680 | } | |
719d3402 | 1681 | |
91fac317 | 1682 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 1683 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
1684 | } |
1685 | ||
812df48d DS |
1686 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1687 | { | |
1688 | del_timer(&cfqd->idle_slice_timer); | |
e98ef89b | 1689 | cfq_blkiocg_update_idle_time_stats(&cfqq->cfqg->blkg); |
812df48d DS |
1690 | } |
1691 | ||
febffd61 JA |
1692 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
1693 | struct cfq_queue *cfqq) | |
22e2c507 JA |
1694 | { |
1695 | if (cfqq) { | |
b1ffe737 DS |
1696 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", |
1697 | cfqd->serving_prio, cfqd->serving_type); | |
62a37f6b JT |
1698 | cfq_blkiocg_update_avg_queue_size_stats(&cfqq->cfqg->blkg); |
1699 | cfqq->slice_start = 0; | |
1700 | cfqq->dispatch_start = jiffies; | |
1701 | cfqq->allocated_slice = 0; | |
1702 | cfqq->slice_end = 0; | |
1703 | cfqq->slice_dispatch = 0; | |
1704 | cfqq->nr_sectors = 0; | |
1705 | ||
1706 | cfq_clear_cfqq_wait_request(cfqq); | |
1707 | cfq_clear_cfqq_must_dispatch(cfqq); | |
1708 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
1709 | cfq_clear_cfqq_fifo_expire(cfqq); | |
1710 | cfq_mark_cfqq_slice_new(cfqq); | |
1711 | ||
1712 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
1713 | } |
1714 | ||
1715 | cfqd->active_queue = cfqq; | |
1716 | } | |
1717 | ||
7b14e3b5 JA |
1718 | /* |
1719 | * current cfqq expired its slice (or was too idle), select new one | |
1720 | */ | |
1721 | static void | |
1722 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 1723 | bool timed_out) |
7b14e3b5 | 1724 | { |
7b679138 JA |
1725 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
1726 | ||
7b14e3b5 | 1727 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 1728 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 1729 | |
7b14e3b5 | 1730 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 1731 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 1732 | |
ae54abed SL |
1733 | /* |
1734 | * If this cfqq is shared between multiple processes, check to | |
1735 | * make sure that those processes are still issuing I/Os within | |
1736 | * the mean seek distance. If not, it may be time to break the | |
1737 | * queues apart again. | |
1738 | */ | |
1739 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
1740 | cfq_mark_cfqq_split_coop(cfqq); | |
1741 | ||
7b14e3b5 | 1742 | /* |
6084cdda | 1743 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 1744 | */ |
c553f8e3 SL |
1745 | if (timed_out) { |
1746 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 1747 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
1748 | else |
1749 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
1750 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
1751 | } | |
7b14e3b5 | 1752 | |
e5ff082e | 1753 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 1754 | |
f04a6424 VG |
1755 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1756 | cfq_del_cfqq_rr(cfqd, cfqq); | |
1757 | ||
edd75ffd | 1758 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
1759 | |
1760 | if (cfqq == cfqd->active_queue) | |
1761 | cfqd->active_queue = NULL; | |
1762 | ||
1763 | if (cfqd->active_cic) { | |
b2efa052 | 1764 | put_io_context(cfqd->active_cic->ioc, cfqd->queue); |
7b14e3b5 JA |
1765 | cfqd->active_cic = NULL; |
1766 | } | |
7b14e3b5 JA |
1767 | } |
1768 | ||
e5ff082e | 1769 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
1770 | { |
1771 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1772 | ||
1773 | if (cfqq) | |
e5ff082e | 1774 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
1775 | } |
1776 | ||
498d3aa2 JA |
1777 | /* |
1778 | * Get next queue for service. Unless we have a queue preemption, | |
1779 | * we'll simply select the first cfqq in the service tree. | |
1780 | */ | |
6d048f53 | 1781 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 1782 | { |
c0324a02 | 1783 | struct cfq_rb_root *service_tree = |
cdb16e8f | 1784 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
65b32a57 | 1785 | cfqd->serving_type); |
d9e7620e | 1786 | |
f04a6424 VG |
1787 | if (!cfqd->rq_queued) |
1788 | return NULL; | |
1789 | ||
1fa8f6d6 VG |
1790 | /* There is nothing to dispatch */ |
1791 | if (!service_tree) | |
1792 | return NULL; | |
c0324a02 CZ |
1793 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
1794 | return NULL; | |
1795 | return cfq_rb_first(service_tree); | |
6d048f53 JA |
1796 | } |
1797 | ||
f04a6424 VG |
1798 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
1799 | { | |
25fb5169 | 1800 | struct cfq_group *cfqg; |
f04a6424 VG |
1801 | struct cfq_queue *cfqq; |
1802 | int i, j; | |
1803 | struct cfq_rb_root *st; | |
1804 | ||
1805 | if (!cfqd->rq_queued) | |
1806 | return NULL; | |
1807 | ||
25fb5169 VG |
1808 | cfqg = cfq_get_next_cfqg(cfqd); |
1809 | if (!cfqg) | |
1810 | return NULL; | |
1811 | ||
f04a6424 VG |
1812 | for_each_cfqg_st(cfqg, i, j, st) |
1813 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
1814 | return cfqq; | |
1815 | return NULL; | |
1816 | } | |
1817 | ||
498d3aa2 JA |
1818 | /* |
1819 | * Get and set a new active queue for service. | |
1820 | */ | |
a36e71f9 JA |
1821 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
1822 | struct cfq_queue *cfqq) | |
6d048f53 | 1823 | { |
e00ef799 | 1824 | if (!cfqq) |
a36e71f9 | 1825 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 1826 | |
22e2c507 | 1827 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 1828 | return cfqq; |
22e2c507 JA |
1829 | } |
1830 | ||
d9e7620e JA |
1831 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
1832 | struct request *rq) | |
1833 | { | |
83096ebf TH |
1834 | if (blk_rq_pos(rq) >= cfqd->last_position) |
1835 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 1836 | else |
83096ebf | 1837 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
1838 | } |
1839 | ||
b2c18e1e | 1840 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 1841 | struct request *rq) |
6d048f53 | 1842 | { |
e9ce335d | 1843 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
1844 | } |
1845 | ||
a36e71f9 JA |
1846 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
1847 | struct cfq_queue *cur_cfqq) | |
1848 | { | |
f2d1f0ae | 1849 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
1850 | struct rb_node *parent, *node; |
1851 | struct cfq_queue *__cfqq; | |
1852 | sector_t sector = cfqd->last_position; | |
1853 | ||
1854 | if (RB_EMPTY_ROOT(root)) | |
1855 | return NULL; | |
1856 | ||
1857 | /* | |
1858 | * First, if we find a request starting at the end of the last | |
1859 | * request, choose it. | |
1860 | */ | |
f2d1f0ae | 1861 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
1862 | if (__cfqq) |
1863 | return __cfqq; | |
1864 | ||
1865 | /* | |
1866 | * If the exact sector wasn't found, the parent of the NULL leaf | |
1867 | * will contain the closest sector. | |
1868 | */ | |
1869 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 1870 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1871 | return __cfqq; |
1872 | ||
2e46e8b2 | 1873 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
1874 | node = rb_next(&__cfqq->p_node); |
1875 | else | |
1876 | node = rb_prev(&__cfqq->p_node); | |
1877 | if (!node) | |
1878 | return NULL; | |
1879 | ||
1880 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 1881 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1882 | return __cfqq; |
1883 | ||
1884 | return NULL; | |
1885 | } | |
1886 | ||
1887 | /* | |
1888 | * cfqd - obvious | |
1889 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
1890 | * closely cooperating with itself. | |
1891 | * | |
1892 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
1893 | * one request, and that cfqd->last_position reflects a position on the disk | |
1894 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
1895 | * assumption. | |
1896 | */ | |
1897 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 1898 | struct cfq_queue *cur_cfqq) |
6d048f53 | 1899 | { |
a36e71f9 JA |
1900 | struct cfq_queue *cfqq; |
1901 | ||
39c01b21 DS |
1902 | if (cfq_class_idle(cur_cfqq)) |
1903 | return NULL; | |
e6c5bc73 JM |
1904 | if (!cfq_cfqq_sync(cur_cfqq)) |
1905 | return NULL; | |
1906 | if (CFQQ_SEEKY(cur_cfqq)) | |
1907 | return NULL; | |
1908 | ||
b9d8f4c7 GJ |
1909 | /* |
1910 | * Don't search priority tree if it's the only queue in the group. | |
1911 | */ | |
1912 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
1913 | return NULL; | |
1914 | ||
6d048f53 | 1915 | /* |
d9e7620e JA |
1916 | * We should notice if some of the queues are cooperating, eg |
1917 | * working closely on the same area of the disk. In that case, | |
1918 | * we can group them together and don't waste time idling. | |
6d048f53 | 1919 | */ |
a36e71f9 JA |
1920 | cfqq = cfqq_close(cfqd, cur_cfqq); |
1921 | if (!cfqq) | |
1922 | return NULL; | |
1923 | ||
8682e1f1 VG |
1924 | /* If new queue belongs to different cfq_group, don't choose it */ |
1925 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
1926 | return NULL; | |
1927 | ||
df5fe3e8 JM |
1928 | /* |
1929 | * It only makes sense to merge sync queues. | |
1930 | */ | |
1931 | if (!cfq_cfqq_sync(cfqq)) | |
1932 | return NULL; | |
e6c5bc73 JM |
1933 | if (CFQQ_SEEKY(cfqq)) |
1934 | return NULL; | |
df5fe3e8 | 1935 | |
c0324a02 CZ |
1936 | /* |
1937 | * Do not merge queues of different priority classes | |
1938 | */ | |
1939 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
1940 | return NULL; | |
1941 | ||
a36e71f9 | 1942 | return cfqq; |
6d048f53 JA |
1943 | } |
1944 | ||
a6d44e98 CZ |
1945 | /* |
1946 | * Determine whether we should enforce idle window for this queue. | |
1947 | */ | |
1948 | ||
1949 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1950 | { | |
1951 | enum wl_prio_t prio = cfqq_prio(cfqq); | |
718eee05 | 1952 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
a6d44e98 | 1953 | |
f04a6424 VG |
1954 | BUG_ON(!service_tree); |
1955 | BUG_ON(!service_tree->count); | |
1956 | ||
b6508c16 VG |
1957 | if (!cfqd->cfq_slice_idle) |
1958 | return false; | |
1959 | ||
a6d44e98 CZ |
1960 | /* We never do for idle class queues. */ |
1961 | if (prio == IDLE_WORKLOAD) | |
1962 | return false; | |
1963 | ||
1964 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
1965 | if (cfq_cfqq_idle_window(cfqq) && |
1966 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
1967 | return true; |
1968 | ||
1969 | /* | |
1970 | * Otherwise, we do only if they are the last ones | |
1971 | * in their service tree. | |
1972 | */ | |
f5f2b6ce SL |
1973 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq) && |
1974 | !cfq_io_thinktime_big(cfqd, &service_tree->ttime, false)) | |
c1e44756 | 1975 | return true; |
b1ffe737 DS |
1976 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", |
1977 | service_tree->count); | |
c1e44756 | 1978 | return false; |
a6d44e98 CZ |
1979 | } |
1980 | ||
6d048f53 | 1981 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 1982 | { |
1792669c | 1983 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 1984 | struct cfq_io_context *cic; |
80bdf0c7 | 1985 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 1986 | |
a68bbddb | 1987 | /* |
f7d7b7a7 JA |
1988 | * SSD device without seek penalty, disable idling. But only do so |
1989 | * for devices that support queuing, otherwise we still have a problem | |
1990 | * with sync vs async workloads. | |
a68bbddb | 1991 | */ |
f7d7b7a7 | 1992 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
1993 | return; |
1994 | ||
dd67d051 | 1995 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 1996 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
1997 | |
1998 | /* | |
1999 | * idle is disabled, either manually or by past process history | |
2000 | */ | |
80bdf0c7 VG |
2001 | if (!cfq_should_idle(cfqd, cfqq)) { |
2002 | /* no queue idling. Check for group idling */ | |
2003 | if (cfqd->cfq_group_idle) | |
2004 | group_idle = cfqd->cfq_group_idle; | |
2005 | else | |
2006 | return; | |
2007 | } | |
6d048f53 | 2008 | |
7b679138 | 2009 | /* |
8e550632 | 2010 | * still active requests from this queue, don't idle |
7b679138 | 2011 | */ |
8e550632 | 2012 | if (cfqq->dispatched) |
7b679138 JA |
2013 | return; |
2014 | ||
22e2c507 JA |
2015 | /* |
2016 | * task has exited, don't wait | |
2017 | */ | |
206dc69b | 2018 | cic = cfqd->active_cic; |
66dac98e | 2019 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
6d048f53 JA |
2020 | return; |
2021 | ||
355b659c CZ |
2022 | /* |
2023 | * If our average think time is larger than the remaining time | |
2024 | * slice, then don't idle. This avoids overrunning the allotted | |
2025 | * time slice. | |
2026 | */ | |
383cd721 SL |
2027 | if (sample_valid(cic->ttime.ttime_samples) && |
2028 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 2029 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 2030 | cic->ttime.ttime_mean); |
355b659c | 2031 | return; |
b1ffe737 | 2032 | } |
355b659c | 2033 | |
80bdf0c7 VG |
2034 | /* There are other queues in the group, don't do group idle */ |
2035 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
2036 | return; | |
2037 | ||
3b18152c | 2038 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 2039 | |
80bdf0c7 VG |
2040 | if (group_idle) |
2041 | sl = cfqd->cfq_group_idle; | |
2042 | else | |
2043 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 2044 | |
7b14e3b5 | 2045 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
e98ef89b | 2046 | cfq_blkiocg_update_set_idle_time_stats(&cfqq->cfqg->blkg); |
80bdf0c7 VG |
2047 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
2048 | group_idle ? 1 : 0); | |
1da177e4 LT |
2049 | } |
2050 | ||
498d3aa2 JA |
2051 | /* |
2052 | * Move request from internal lists to the request queue dispatch list. | |
2053 | */ | |
165125e1 | 2054 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 2055 | { |
3ed9a296 | 2056 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 2057 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 2058 | |
7b679138 JA |
2059 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
2060 | ||
06d21886 | 2061 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 2062 | cfq_remove_request(rq); |
6d048f53 | 2063 | cfqq->dispatched++; |
80bdf0c7 | 2064 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 2065 | elv_dispatch_sort(q, rq); |
3ed9a296 | 2066 | |
53c583d2 | 2067 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 2068 | cfqq->nr_sectors += blk_rq_sectors(rq); |
e98ef89b | 2069 | cfq_blkiocg_update_dispatch_stats(&cfqq->cfqg->blkg, blk_rq_bytes(rq), |
84c124da | 2070 | rq_data_dir(rq), rq_is_sync(rq)); |
1da177e4 LT |
2071 | } |
2072 | ||
2073 | /* | |
2074 | * return expired entry, or NULL to just start from scratch in rbtree | |
2075 | */ | |
febffd61 | 2076 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 2077 | { |
30996f40 | 2078 | struct request *rq = NULL; |
1da177e4 | 2079 | |
3b18152c | 2080 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 2081 | return NULL; |
cb887411 JA |
2082 | |
2083 | cfq_mark_cfqq_fifo_expire(cfqq); | |
2084 | ||
89850f7e JA |
2085 | if (list_empty(&cfqq->fifo)) |
2086 | return NULL; | |
1da177e4 | 2087 | |
89850f7e | 2088 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 2089 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 2090 | rq = NULL; |
1da177e4 | 2091 | |
30996f40 | 2092 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 2093 | return rq; |
1da177e4 LT |
2094 | } |
2095 | ||
22e2c507 JA |
2096 | static inline int |
2097 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2098 | { | |
2099 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2100 | |
22e2c507 | 2101 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2102 | |
b9f8ce05 | 2103 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
2104 | } |
2105 | ||
df5fe3e8 JM |
2106 | /* |
2107 | * Must be called with the queue_lock held. | |
2108 | */ | |
2109 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2110 | { | |
2111 | int process_refs, io_refs; | |
2112 | ||
2113 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 2114 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
2115 | BUG_ON(process_refs < 0); |
2116 | return process_refs; | |
2117 | } | |
2118 | ||
2119 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2120 | { | |
e6c5bc73 | 2121 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2122 | struct cfq_queue *__cfqq; |
2123 | ||
c10b61f0 JM |
2124 | /* |
2125 | * If there are no process references on the new_cfqq, then it is | |
2126 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2127 | * chain may have dropped their last reference (not just their | |
2128 | * last process reference). | |
2129 | */ | |
2130 | if (!cfqq_process_refs(new_cfqq)) | |
2131 | return; | |
2132 | ||
df5fe3e8 JM |
2133 | /* Avoid a circular list and skip interim queue merges */ |
2134 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2135 | if (__cfqq == cfqq) | |
2136 | return; | |
2137 | new_cfqq = __cfqq; | |
2138 | } | |
2139 | ||
2140 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2141 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2142 | /* |
2143 | * If the process for the cfqq has gone away, there is no | |
2144 | * sense in merging the queues. | |
2145 | */ | |
c10b61f0 | 2146 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2147 | return; |
2148 | ||
e6c5bc73 JM |
2149 | /* |
2150 | * Merge in the direction of the lesser amount of work. | |
2151 | */ | |
e6c5bc73 JM |
2152 | if (new_process_refs >= process_refs) { |
2153 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2154 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2155 | } else { |
2156 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2157 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2158 | } |
df5fe3e8 JM |
2159 | } |
2160 | ||
cdb16e8f | 2161 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
65b32a57 | 2162 | struct cfq_group *cfqg, enum wl_prio_t prio) |
718eee05 CZ |
2163 | { |
2164 | struct cfq_queue *queue; | |
2165 | int i; | |
2166 | bool key_valid = false; | |
2167 | unsigned long lowest_key = 0; | |
2168 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2169 | ||
65b32a57 VG |
2170 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2171 | /* select the one with lowest rb_key */ | |
2172 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); | |
718eee05 CZ |
2173 | if (queue && |
2174 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2175 | lowest_key = queue->rb_key; | |
2176 | cur_best = i; | |
2177 | key_valid = true; | |
2178 | } | |
2179 | } | |
2180 | ||
2181 | return cur_best; | |
2182 | } | |
2183 | ||
cdb16e8f | 2184 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
718eee05 | 2185 | { |
718eee05 CZ |
2186 | unsigned slice; |
2187 | unsigned count; | |
cdb16e8f | 2188 | struct cfq_rb_root *st; |
58ff82f3 | 2189 | unsigned group_slice; |
e4ea0c16 | 2190 | enum wl_prio_t original_prio = cfqd->serving_prio; |
1fa8f6d6 | 2191 | |
718eee05 | 2192 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2193 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
718eee05 | 2194 | cfqd->serving_prio = RT_WORKLOAD; |
58ff82f3 | 2195 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
718eee05 CZ |
2196 | cfqd->serving_prio = BE_WORKLOAD; |
2197 | else { | |
2198 | cfqd->serving_prio = IDLE_WORKLOAD; | |
2199 | cfqd->workload_expires = jiffies + 1; | |
2200 | return; | |
2201 | } | |
2202 | ||
e4ea0c16 SL |
2203 | if (original_prio != cfqd->serving_prio) |
2204 | goto new_workload; | |
2205 | ||
718eee05 CZ |
2206 | /* |
2207 | * For RT and BE, we have to choose also the type | |
2208 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2209 | * expiration time | |
2210 | */ | |
65b32a57 | 2211 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
cdb16e8f | 2212 | count = st->count; |
718eee05 CZ |
2213 | |
2214 | /* | |
65b32a57 | 2215 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2216 | */ |
65b32a57 | 2217 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2218 | return; |
2219 | ||
e4ea0c16 | 2220 | new_workload: |
718eee05 CZ |
2221 | /* otherwise select new workload type */ |
2222 | cfqd->serving_type = | |
65b32a57 VG |
2223 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); |
2224 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); | |
cdb16e8f | 2225 | count = st->count; |
718eee05 CZ |
2226 | |
2227 | /* | |
2228 | * the workload slice is computed as a fraction of target latency | |
2229 | * proportional to the number of queues in that workload, over | |
2230 | * all the queues in the same priority class | |
2231 | */ | |
58ff82f3 VG |
2232 | group_slice = cfq_group_slice(cfqd, cfqg); |
2233 | ||
2234 | slice = group_slice * count / | |
2235 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], | |
2236 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); | |
718eee05 | 2237 | |
f26bd1f0 VG |
2238 | if (cfqd->serving_type == ASYNC_WORKLOAD) { |
2239 | unsigned int tmp; | |
2240 | ||
2241 | /* | |
2242 | * Async queues are currently system wide. Just taking | |
2243 | * proportion of queues with-in same group will lead to higher | |
2244 | * async ratio system wide as generally root group is going | |
2245 | * to have higher weight. A more accurate thing would be to | |
2246 | * calculate system wide asnc/sync ratio. | |
2247 | */ | |
2248 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); | |
2249 | tmp = tmp/cfqd->busy_queues; | |
2250 | slice = min_t(unsigned, slice, tmp); | |
2251 | ||
718eee05 CZ |
2252 | /* async workload slice is scaled down according to |
2253 | * the sync/async slice ratio. */ | |
2254 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2255 | } else |
718eee05 CZ |
2256 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2257 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2258 | ||
2259 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2260 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
2261 | cfqd->workload_expires = jiffies + slice; |
2262 | } | |
2263 | ||
1fa8f6d6 VG |
2264 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2265 | { | |
2266 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2267 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2268 | |
2269 | if (RB_EMPTY_ROOT(&st->rb)) | |
2270 | return NULL; | |
25bc6b07 | 2271 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
2272 | update_min_vdisktime(st); |
2273 | return cfqg; | |
1fa8f6d6 VG |
2274 | } |
2275 | ||
cdb16e8f VG |
2276 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2277 | { | |
1fa8f6d6 VG |
2278 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2279 | ||
2280 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2281 | |
2282 | /* Restore the workload type data */ | |
2283 | if (cfqg->saved_workload_slice) { | |
2284 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; | |
2285 | cfqd->serving_type = cfqg->saved_workload; | |
2286 | cfqd->serving_prio = cfqg->saved_serving_prio; | |
66ae2919 GJ |
2287 | } else |
2288 | cfqd->workload_expires = jiffies - 1; | |
2289 | ||
1fa8f6d6 | 2290 | choose_service_tree(cfqd, cfqg); |
cdb16e8f VG |
2291 | } |
2292 | ||
22e2c507 | 2293 | /* |
498d3aa2 JA |
2294 | * Select a queue for service. If we have a current active queue, |
2295 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2296 | */ |
1b5ed5e1 | 2297 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2298 | { |
a36e71f9 | 2299 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2300 | |
22e2c507 JA |
2301 | cfqq = cfqd->active_queue; |
2302 | if (!cfqq) | |
2303 | goto new_queue; | |
1da177e4 | 2304 | |
f04a6424 VG |
2305 | if (!cfqd->rq_queued) |
2306 | return NULL; | |
c244bb50 VG |
2307 | |
2308 | /* | |
2309 | * We were waiting for group to get backlogged. Expire the queue | |
2310 | */ | |
2311 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2312 | goto expire; | |
2313 | ||
22e2c507 | 2314 | /* |
6d048f53 | 2315 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2316 | */ |
7667aa06 VG |
2317 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2318 | /* | |
2319 | * If slice had not expired at the completion of last request | |
2320 | * we might not have turned on wait_busy flag. Don't expire | |
2321 | * the queue yet. Allow the group to get backlogged. | |
2322 | * | |
2323 | * The very fact that we have used the slice, that means we | |
2324 | * have been idling all along on this queue and it should be | |
2325 | * ok to wait for this request to complete. | |
2326 | */ | |
82bbbf28 VG |
2327 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2328 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2329 | cfqq = NULL; | |
7667aa06 | 2330 | goto keep_queue; |
82bbbf28 | 2331 | } else |
80bdf0c7 | 2332 | goto check_group_idle; |
7667aa06 | 2333 | } |
1da177e4 | 2334 | |
22e2c507 | 2335 | /* |
6d048f53 JA |
2336 | * The active queue has requests and isn't expired, allow it to |
2337 | * dispatch. | |
22e2c507 | 2338 | */ |
dd67d051 | 2339 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2340 | goto keep_queue; |
6d048f53 | 2341 | |
a36e71f9 JA |
2342 | /* |
2343 | * If another queue has a request waiting within our mean seek | |
2344 | * distance, let it run. The expire code will check for close | |
2345 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2346 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2347 | */ |
b3b6d040 | 2348 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2349 | if (new_cfqq) { |
2350 | if (!cfqq->new_cfqq) | |
2351 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2352 | goto expire; |
df5fe3e8 | 2353 | } |
a36e71f9 | 2354 | |
6d048f53 JA |
2355 | /* |
2356 | * No requests pending. If the active queue still has requests in | |
2357 | * flight or is idling for a new request, allow either of these | |
2358 | * conditions to happen (or time out) before selecting a new queue. | |
2359 | */ | |
80bdf0c7 VG |
2360 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2361 | cfqq = NULL; | |
2362 | goto keep_queue; | |
2363 | } | |
2364 | ||
8e1ac665 SL |
2365 | /* |
2366 | * This is a deep seek queue, but the device is much faster than | |
2367 | * the queue can deliver, don't idle | |
2368 | **/ | |
2369 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
2370 | (cfq_cfqq_slice_new(cfqq) || | |
2371 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
2372 | cfq_clear_cfqq_deep(cfqq); | |
2373 | cfq_clear_cfqq_idle_window(cfqq); | |
2374 | } | |
2375 | ||
80bdf0c7 VG |
2376 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2377 | cfqq = NULL; | |
2378 | goto keep_queue; | |
2379 | } | |
2380 | ||
2381 | /* | |
2382 | * If group idle is enabled and there are requests dispatched from | |
2383 | * this group, wait for requests to complete. | |
2384 | */ | |
2385 | check_group_idle: | |
7700fc4f SL |
2386 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
2387 | cfqq->cfqg->dispatched && | |
2388 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
2389 | cfqq = NULL; |
2390 | goto keep_queue; | |
22e2c507 JA |
2391 | } |
2392 | ||
3b18152c | 2393 | expire: |
e5ff082e | 2394 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2395 | new_queue: |
718eee05 CZ |
2396 | /* |
2397 | * Current queue expired. Check if we have to switch to a new | |
2398 | * service tree | |
2399 | */ | |
2400 | if (!new_cfqq) | |
cdb16e8f | 2401 | cfq_choose_cfqg(cfqd); |
718eee05 | 2402 | |
a36e71f9 | 2403 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2404 | keep_queue: |
3b18152c | 2405 | return cfqq; |
22e2c507 JA |
2406 | } |
2407 | ||
febffd61 | 2408 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2409 | { |
2410 | int dispatched = 0; | |
2411 | ||
2412 | while (cfqq->next_rq) { | |
2413 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2414 | dispatched++; | |
2415 | } | |
2416 | ||
2417 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2418 | |
2419 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 2420 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
2421 | return dispatched; |
2422 | } | |
2423 | ||
498d3aa2 JA |
2424 | /* |
2425 | * Drain our current requests. Used for barriers and when switching | |
2426 | * io schedulers on-the-fly. | |
2427 | */ | |
d9e7620e | 2428 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2429 | { |
0871714e | 2430 | struct cfq_queue *cfqq; |
d9e7620e | 2431 | int dispatched = 0; |
cdb16e8f | 2432 | |
3440c49f | 2433 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 2434 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
2435 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2436 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 2437 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 2438 | } |
1b5ed5e1 | 2439 | |
1b5ed5e1 TH |
2440 | BUG_ON(cfqd->busy_queues); |
2441 | ||
6923715a | 2442 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2443 | return dispatched; |
2444 | } | |
2445 | ||
abc3c744 SL |
2446 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2447 | struct cfq_queue *cfqq) | |
2448 | { | |
2449 | /* the queue hasn't finished any request, can't estimate */ | |
2450 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 2451 | return true; |
abc3c744 SL |
2452 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
2453 | cfqq->slice_end)) | |
c1e44756 | 2454 | return true; |
abc3c744 | 2455 | |
c1e44756 | 2456 | return false; |
abc3c744 SL |
2457 | } |
2458 | ||
0b182d61 | 2459 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2460 | { |
2f5cb738 | 2461 | unsigned int max_dispatch; |
22e2c507 | 2462 | |
5ad531db JA |
2463 | /* |
2464 | * Drain async requests before we start sync IO | |
2465 | */ | |
53c583d2 | 2466 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 2467 | return false; |
5ad531db | 2468 | |
2f5cb738 JA |
2469 | /* |
2470 | * If this is an async queue and we have sync IO in flight, let it wait | |
2471 | */ | |
53c583d2 | 2472 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 2473 | return false; |
2f5cb738 | 2474 | |
abc3c744 | 2475 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
2476 | if (cfq_class_idle(cfqq)) |
2477 | max_dispatch = 1; | |
b4878f24 | 2478 | |
2f5cb738 JA |
2479 | /* |
2480 | * Does this cfqq already have too much IO in flight? | |
2481 | */ | |
2482 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 2483 | bool promote_sync = false; |
2f5cb738 JA |
2484 | /* |
2485 | * idle queue must always only have a single IO in flight | |
2486 | */ | |
3ed9a296 | 2487 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2488 | return false; |
3ed9a296 | 2489 | |
ef8a41df | 2490 | /* |
c4ade94f LS |
2491 | * If there is only one sync queue |
2492 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
2493 | * queue no dispatch limit. The reason is a sync queue can |
2494 | * preempt async queue, limiting the sync queue doesn't make | |
2495 | * sense. This is useful for aiostress test. | |
2496 | */ | |
c4ade94f LS |
2497 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
2498 | promote_sync = true; | |
ef8a41df | 2499 | |
2f5cb738 JA |
2500 | /* |
2501 | * We have other queues, don't allow more IO from this one | |
2502 | */ | |
ef8a41df SL |
2503 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
2504 | !promote_sync) | |
0b182d61 | 2505 | return false; |
9ede209e | 2506 | |
365722bb | 2507 | /* |
474b18cc | 2508 | * Sole queue user, no limit |
365722bb | 2509 | */ |
ef8a41df | 2510 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
2511 | max_dispatch = -1; |
2512 | else | |
2513 | /* | |
2514 | * Normally we start throttling cfqq when cfq_quantum/2 | |
2515 | * requests have been dispatched. But we can drive | |
2516 | * deeper queue depths at the beginning of slice | |
2517 | * subjected to upper limit of cfq_quantum. | |
2518 | * */ | |
2519 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
2520 | } |
2521 | ||
2522 | /* | |
2523 | * Async queues must wait a bit before being allowed dispatch. | |
2524 | * We also ramp up the dispatch depth gradually for async IO, | |
2525 | * based on the last sync IO we serviced | |
2526 | */ | |
963b72fc | 2527 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 2528 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 2529 | unsigned int depth; |
365722bb | 2530 | |
61f0c1dc | 2531 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2532 | if (!depth && !cfqq->dispatched) |
2533 | depth = 1; | |
8e296755 JA |
2534 | if (depth < max_dispatch) |
2535 | max_dispatch = depth; | |
2f5cb738 | 2536 | } |
3ed9a296 | 2537 | |
0b182d61 JA |
2538 | /* |
2539 | * If we're below the current max, allow a dispatch | |
2540 | */ | |
2541 | return cfqq->dispatched < max_dispatch; | |
2542 | } | |
2543 | ||
2544 | /* | |
2545 | * Dispatch a request from cfqq, moving them to the request queue | |
2546 | * dispatch list. | |
2547 | */ | |
2548 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2549 | { | |
2550 | struct request *rq; | |
2551 | ||
2552 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
2553 | ||
2554 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
2555 | return false; | |
2556 | ||
2557 | /* | |
2558 | * follow expired path, else get first next available | |
2559 | */ | |
2560 | rq = cfq_check_fifo(cfqq); | |
2561 | if (!rq) | |
2562 | rq = cfqq->next_rq; | |
2563 | ||
2564 | /* | |
2565 | * insert request into driver dispatch list | |
2566 | */ | |
2567 | cfq_dispatch_insert(cfqd->queue, rq); | |
2568 | ||
2569 | if (!cfqd->active_cic) { | |
2570 | struct cfq_io_context *cic = RQ_CIC(rq); | |
2571 | ||
2572 | atomic_long_inc(&cic->ioc->refcount); | |
2573 | cfqd->active_cic = cic; | |
2574 | } | |
2575 | ||
2576 | return true; | |
2577 | } | |
2578 | ||
2579 | /* | |
2580 | * Find the cfqq that we need to service and move a request from that to the | |
2581 | * dispatch list | |
2582 | */ | |
2583 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
2584 | { | |
2585 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2586 | struct cfq_queue *cfqq; | |
2587 | ||
2588 | if (!cfqd->busy_queues) | |
2589 | return 0; | |
2590 | ||
2591 | if (unlikely(force)) | |
2592 | return cfq_forced_dispatch(cfqd); | |
2593 | ||
2594 | cfqq = cfq_select_queue(cfqd); | |
2595 | if (!cfqq) | |
8e296755 JA |
2596 | return 0; |
2597 | ||
2f5cb738 | 2598 | /* |
0b182d61 | 2599 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 2600 | */ |
0b182d61 JA |
2601 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2602 | return 0; | |
2603 | ||
2f5cb738 | 2604 | cfqq->slice_dispatch++; |
b029195d | 2605 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 2606 | |
2f5cb738 JA |
2607 | /* |
2608 | * expire an async queue immediately if it has used up its slice. idle | |
2609 | * queue always expire after 1 dispatch round. | |
2610 | */ | |
2611 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
2612 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
2613 | cfq_class_idle(cfqq))) { | |
2614 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 2615 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
2616 | } |
2617 | ||
b217a903 | 2618 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 2619 | return 1; |
1da177e4 LT |
2620 | } |
2621 | ||
1da177e4 | 2622 | /* |
5e705374 JA |
2623 | * task holds one reference to the queue, dropped when task exits. each rq |
2624 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 2625 | * |
b1c35769 | 2626 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
2627 | * queue lock must be held here. |
2628 | */ | |
2629 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
2630 | { | |
22e2c507 | 2631 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 2632 | struct cfq_group *cfqg; |
22e2c507 | 2633 | |
30d7b944 | 2634 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 2635 | |
30d7b944 SL |
2636 | cfqq->ref--; |
2637 | if (cfqq->ref) | |
1da177e4 LT |
2638 | return; |
2639 | ||
7b679138 | 2640 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 2641 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 2642 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 2643 | cfqg = cfqq->cfqg; |
1da177e4 | 2644 | |
28f95cbc | 2645 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 2646 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2647 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2648 | } |
22e2c507 | 2649 | |
f04a6424 | 2650 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 2651 | kmem_cache_free(cfq_pool, cfqq); |
b1c35769 | 2652 | cfq_put_cfqg(cfqg); |
1da177e4 LT |
2653 | } |
2654 | ||
34e6bbf2 FC |
2655 | static void cfq_cic_free_rcu(struct rcu_head *head) |
2656 | { | |
b50b636b TH |
2657 | kmem_cache_free(cfq_ioc_pool, |
2658 | container_of(head, struct cfq_io_context, rcu_head)); | |
34e6bbf2 | 2659 | } |
4ac845a2 | 2660 | |
34e6bbf2 FC |
2661 | static void cfq_cic_free(struct cfq_io_context *cic) |
2662 | { | |
2663 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | |
4ac845a2 JA |
2664 | } |
2665 | ||
283287a5 | 2666 | static void cfq_release_cic(struct cfq_io_context *cic) |
4ac845a2 | 2667 | { |
283287a5 | 2668 | struct io_context *ioc = cic->ioc; |
4ac845a2 | 2669 | |
1238033c | 2670 | radix_tree_delete(&ioc->radix_root, cic->q->id); |
b50b636b | 2671 | hlist_del(&cic->cic_list); |
34e6bbf2 | 2672 | cfq_cic_free(cic); |
4ac845a2 JA |
2673 | } |
2674 | ||
d02a2c07 | 2675 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 2676 | { |
df5fe3e8 JM |
2677 | struct cfq_queue *__cfqq, *next; |
2678 | ||
df5fe3e8 JM |
2679 | /* |
2680 | * If this queue was scheduled to merge with another queue, be | |
2681 | * sure to drop the reference taken on that queue (and others in | |
2682 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
2683 | */ | |
2684 | __cfqq = cfqq->new_cfqq; | |
2685 | while (__cfqq) { | |
2686 | if (__cfqq == cfqq) { | |
2687 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
2688 | break; | |
2689 | } | |
2690 | next = __cfqq->new_cfqq; | |
2691 | cfq_put_queue(__cfqq); | |
2692 | __cfqq = next; | |
2693 | } | |
d02a2c07 SL |
2694 | } |
2695 | ||
2696 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2697 | { | |
2698 | if (unlikely(cfqq == cfqd->active_queue)) { | |
2699 | __cfq_slice_expired(cfqd, cfqq, 0); | |
2700 | cfq_schedule_dispatch(cfqd); | |
2701 | } | |
2702 | ||
2703 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 2704 | |
89850f7e JA |
2705 | cfq_put_queue(cfqq); |
2706 | } | |
22e2c507 | 2707 | |
283287a5 | 2708 | static void cfq_exit_cic(struct cfq_io_context *cic) |
89850f7e | 2709 | { |
283287a5 | 2710 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 FC |
2711 | struct io_context *ioc = cic->ioc; |
2712 | ||
fc46379d | 2713 | list_del_init(&cic->queue_list); |
4ac845a2 JA |
2714 | |
2715 | /* | |
f1a4f4d3 TH |
2716 | * Both setting lookup hint to and clearing it from @cic are done |
2717 | * under queue_lock. If it's not pointing to @cic now, it never | |
2718 | * will. Hint assignment itself can race safely. | |
4ac845a2 | 2719 | */ |
f1a4f4d3 | 2720 | if (rcu_dereference_raw(ioc->ioc_data) == cic) |
4faa3c81 FC |
2721 | rcu_assign_pointer(ioc->ioc_data, NULL); |
2722 | ||
ff6657c6 JA |
2723 | if (cic->cfqq[BLK_RW_ASYNC]) { |
2724 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
2725 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
2726 | } |
2727 | ||
ff6657c6 JA |
2728 | if (cic->cfqq[BLK_RW_SYNC]) { |
2729 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
2730 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 2731 | } |
89850f7e JA |
2732 | } |
2733 | ||
22e2c507 | 2734 | static struct cfq_io_context * |
8267e268 | 2735 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 2736 | { |
b5deef90 | 2737 | struct cfq_io_context *cic; |
1da177e4 | 2738 | |
94f6030c CL |
2739 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
2740 | cfqd->queue->node); | |
1da177e4 | 2741 | if (cic) { |
383cd721 | 2742 | cic->ttime.last_end_request = jiffies; |
553698f9 | 2743 | INIT_LIST_HEAD(&cic->queue_list); |
ffc4e759 | 2744 | INIT_HLIST_NODE(&cic->cic_list); |
b2efa052 TH |
2745 | cic->exit = cfq_exit_cic; |
2746 | cic->release = cfq_release_cic; | |
1da177e4 LT |
2747 | } |
2748 | ||
2749 | return cic; | |
2750 | } | |
2751 | ||
fd0928df | 2752 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
2753 | { |
2754 | struct task_struct *tsk = current; | |
2755 | int ioprio_class; | |
2756 | ||
3b18152c | 2757 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
2758 | return; |
2759 | ||
fd0928df | 2760 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 2761 | switch (ioprio_class) { |
fe094d98 JA |
2762 | default: |
2763 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
2764 | case IOPRIO_CLASS_NONE: | |
2765 | /* | |
6d63c275 | 2766 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
2767 | */ |
2768 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 2769 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
2770 | break; |
2771 | case IOPRIO_CLASS_RT: | |
2772 | cfqq->ioprio = task_ioprio(ioc); | |
2773 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
2774 | break; | |
2775 | case IOPRIO_CLASS_BE: | |
2776 | cfqq->ioprio = task_ioprio(ioc); | |
2777 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
2778 | break; | |
2779 | case IOPRIO_CLASS_IDLE: | |
2780 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
2781 | cfqq->ioprio = 7; | |
2782 | cfq_clear_cfqq_idle_window(cfqq); | |
2783 | break; | |
22e2c507 JA |
2784 | } |
2785 | ||
2786 | /* | |
2787 | * keep track of original prio settings in case we have to temporarily | |
2788 | * elevate the priority of this queue | |
2789 | */ | |
2790 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 2791 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
2792 | } |
2793 | ||
dc86900e | 2794 | static void changed_ioprio(struct cfq_io_context *cic) |
22e2c507 | 2795 | { |
bca4b914 | 2796 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 2797 | struct cfq_queue *cfqq; |
35e6077c | 2798 | |
caaa5f9f JA |
2799 | if (unlikely(!cfqd)) |
2800 | return; | |
2801 | ||
ff6657c6 | 2802 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
2803 | if (cfqq) { |
2804 | struct cfq_queue *new_cfqq; | |
ff6657c6 JA |
2805 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc, |
2806 | GFP_ATOMIC); | |
caaa5f9f | 2807 | if (new_cfqq) { |
ff6657c6 | 2808 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
2809 | cfq_put_queue(cfqq); |
2810 | } | |
22e2c507 | 2811 | } |
caaa5f9f | 2812 | |
ff6657c6 | 2813 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
2814 | if (cfqq) |
2815 | cfq_mark_cfqq_prio_changed(cfqq); | |
22e2c507 JA |
2816 | } |
2817 | ||
d5036d77 | 2818 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2819 | pid_t pid, bool is_sync) |
d5036d77 JA |
2820 | { |
2821 | RB_CLEAR_NODE(&cfqq->rb_node); | |
2822 | RB_CLEAR_NODE(&cfqq->p_node); | |
2823 | INIT_LIST_HEAD(&cfqq->fifo); | |
2824 | ||
30d7b944 | 2825 | cfqq->ref = 0; |
d5036d77 JA |
2826 | cfqq->cfqd = cfqd; |
2827 | ||
2828 | cfq_mark_cfqq_prio_changed(cfqq); | |
2829 | ||
2830 | if (is_sync) { | |
2831 | if (!cfq_class_idle(cfqq)) | |
2832 | cfq_mark_cfqq_idle_window(cfqq); | |
2833 | cfq_mark_cfqq_sync(cfqq); | |
2834 | } | |
2835 | cfqq->pid = pid; | |
2836 | } | |
2837 | ||
24610333 | 2838 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
dc86900e | 2839 | static void changed_cgroup(struct cfq_io_context *cic) |
24610333 VG |
2840 | { |
2841 | struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1); | |
bca4b914 | 2842 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
24610333 VG |
2843 | struct request_queue *q; |
2844 | ||
2845 | if (unlikely(!cfqd)) | |
2846 | return; | |
2847 | ||
2848 | q = cfqd->queue; | |
2849 | ||
24610333 VG |
2850 | if (sync_cfqq) { |
2851 | /* | |
2852 | * Drop reference to sync queue. A new sync queue will be | |
2853 | * assigned in new group upon arrival of a fresh request. | |
2854 | */ | |
2855 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
2856 | cic_set_cfqq(cic, NULL, 1); | |
2857 | cfq_put_queue(sync_cfqq); | |
2858 | } | |
24610333 | 2859 | } |
24610333 VG |
2860 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
2861 | ||
22e2c507 | 2862 | static struct cfq_queue * |
a6151c3a | 2863 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, |
fd0928df | 2864 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 2865 | { |
22e2c507 | 2866 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 2867 | struct cfq_io_context *cic; |
cdb16e8f | 2868 | struct cfq_group *cfqg; |
22e2c507 JA |
2869 | |
2870 | retry: | |
3e59cf9d | 2871 | cfqg = cfq_get_cfqg(cfqd); |
4ac845a2 | 2872 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
2873 | /* cic always exists here */ |
2874 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 | 2875 | |
6118b70b JA |
2876 | /* |
2877 | * Always try a new alloc if we fell back to the OOM cfqq | |
2878 | * originally, since it should just be a temporary situation. | |
2879 | */ | |
2880 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
2881 | cfqq = NULL; | |
22e2c507 JA |
2882 | if (new_cfqq) { |
2883 | cfqq = new_cfqq; | |
2884 | new_cfqq = NULL; | |
2885 | } else if (gfp_mask & __GFP_WAIT) { | |
2886 | spin_unlock_irq(cfqd->queue->queue_lock); | |
94f6030c | 2887 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 2888 | gfp_mask | __GFP_ZERO, |
94f6030c | 2889 | cfqd->queue->node); |
22e2c507 | 2890 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
2891 | if (new_cfqq) |
2892 | goto retry; | |
22e2c507 | 2893 | } else { |
94f6030c CL |
2894 | cfqq = kmem_cache_alloc_node(cfq_pool, |
2895 | gfp_mask | __GFP_ZERO, | |
2896 | cfqd->queue->node); | |
22e2c507 JA |
2897 | } |
2898 | ||
6118b70b JA |
2899 | if (cfqq) { |
2900 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
2901 | cfq_init_prio_data(cfqq, ioc); | |
cdb16e8f | 2902 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
2903 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
2904 | } else | |
2905 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
2906 | } |
2907 | ||
2908 | if (new_cfqq) | |
2909 | kmem_cache_free(cfq_pool, new_cfqq); | |
2910 | ||
22e2c507 JA |
2911 | return cfqq; |
2912 | } | |
2913 | ||
c2dea2d1 VT |
2914 | static struct cfq_queue ** |
2915 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
2916 | { | |
fe094d98 | 2917 | switch (ioprio_class) { |
c2dea2d1 VT |
2918 | case IOPRIO_CLASS_RT: |
2919 | return &cfqd->async_cfqq[0][ioprio]; | |
2920 | case IOPRIO_CLASS_BE: | |
2921 | return &cfqd->async_cfqq[1][ioprio]; | |
2922 | case IOPRIO_CLASS_IDLE: | |
2923 | return &cfqd->async_idle_cfqq; | |
2924 | default: | |
2925 | BUG(); | |
2926 | } | |
2927 | } | |
2928 | ||
15c31be4 | 2929 | static struct cfq_queue * |
a6151c3a | 2930 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc, |
15c31be4 JA |
2931 | gfp_t gfp_mask) |
2932 | { | |
fd0928df JA |
2933 | const int ioprio = task_ioprio(ioc); |
2934 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 2935 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
2936 | struct cfq_queue *cfqq = NULL; |
2937 | ||
c2dea2d1 VT |
2938 | if (!is_sync) { |
2939 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
2940 | cfqq = *async_cfqq; | |
2941 | } | |
2942 | ||
6118b70b | 2943 | if (!cfqq) |
fd0928df | 2944 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
15c31be4 JA |
2945 | |
2946 | /* | |
2947 | * pin the queue now that it's allocated, scheduler exit will prune it | |
2948 | */ | |
c2dea2d1 | 2949 | if (!is_sync && !(*async_cfqq)) { |
30d7b944 | 2950 | cfqq->ref++; |
c2dea2d1 | 2951 | *async_cfqq = cfqq; |
15c31be4 JA |
2952 | } |
2953 | ||
30d7b944 | 2954 | cfqq->ref++; |
15c31be4 JA |
2955 | return cfqq; |
2956 | } | |
2957 | ||
f1a4f4d3 TH |
2958 | /** |
2959 | * cfq_cic_lookup - lookup cfq_io_context | |
2960 | * @cfqd: the associated cfq_data | |
2961 | * @ioc: the associated io_context | |
2962 | * | |
2963 | * Look up cfq_io_context associated with @cfqd - @ioc pair. Must be | |
2964 | * called with queue_lock held. | |
2965 | */ | |
e2d74ac0 | 2966 | static struct cfq_io_context * |
4ac845a2 | 2967 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
e2d74ac0 | 2968 | { |
1238033c | 2969 | struct request_queue *q = cfqd->queue; |
e2d74ac0 | 2970 | struct cfq_io_context *cic; |
e2d74ac0 | 2971 | |
f1a4f4d3 | 2972 | lockdep_assert_held(cfqd->queue->queue_lock); |
91fac317 VT |
2973 | if (unlikely(!ioc)) |
2974 | return NULL; | |
2975 | ||
597bc485 | 2976 | /* |
b9a19208 TH |
2977 | * cic's are indexed from @ioc using radix tree and hint pointer, |
2978 | * both of which are protected with RCU. All removals are done | |
2979 | * holding both q and ioc locks, and we're holding q lock - if we | |
2980 | * find a cic which points to us, it's guaranteed to be valid. | |
597bc485 | 2981 | */ |
b9a19208 | 2982 | rcu_read_lock(); |
4ac845a2 | 2983 | cic = rcu_dereference(ioc->ioc_data); |
1238033c | 2984 | if (cic && cic->q == q) |
f1a4f4d3 | 2985 | goto out; |
597bc485 | 2986 | |
b9a19208 | 2987 | cic = radix_tree_lookup(&ioc->radix_root, cfqd->queue->id); |
1238033c | 2988 | if (cic && cic->q == q) |
b9a19208 TH |
2989 | rcu_assign_pointer(ioc->ioc_data, cic); /* allowed to race */ |
2990 | else | |
2991 | cic = NULL; | |
f1a4f4d3 TH |
2992 | out: |
2993 | rcu_read_unlock(); | |
4ac845a2 | 2994 | return cic; |
e2d74ac0 JA |
2995 | } |
2996 | ||
216284c3 TH |
2997 | /** |
2998 | * cfq_create_cic - create and link a cfq_io_context | |
2999 | * @cfqd: cfqd of interest | |
3000 | * @gfp_mask: allocation mask | |
3001 | * | |
3002 | * Make sure cfq_io_context linking %current->io_context and @cfqd exists. | |
3003 | * If ioc and/or cic doesn't exist, they will be created using @gfp_mask. | |
4ac845a2 | 3004 | */ |
216284c3 | 3005 | static int cfq_create_cic(struct cfq_data *cfqd, gfp_t gfp_mask) |
e2d74ac0 | 3006 | { |
216284c3 TH |
3007 | struct request_queue *q = cfqd->queue; |
3008 | struct cfq_io_context *cic = NULL; | |
3009 | struct io_context *ioc; | |
3010 | int ret = -ENOMEM; | |
3011 | ||
3012 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3013 | ||
3014 | /* allocate stuff */ | |
f2dbd76a | 3015 | ioc = create_io_context(current, gfp_mask, q->node); |
216284c3 TH |
3016 | if (!ioc) |
3017 | goto out; | |
3018 | ||
3019 | cic = cfq_alloc_io_context(cfqd, gfp_mask); | |
3020 | if (!cic) | |
3021 | goto out; | |
e2d74ac0 | 3022 | |
4ac845a2 | 3023 | ret = radix_tree_preload(gfp_mask); |
283287a5 TH |
3024 | if (ret) |
3025 | goto out; | |
e2d74ac0 | 3026 | |
283287a5 | 3027 | cic->ioc = ioc; |
283287a5 TH |
3028 | cic->q = cfqd->queue; |
3029 | ||
216284c3 TH |
3030 | /* lock both q and ioc and try to link @cic */ |
3031 | spin_lock_irq(q->queue_lock); | |
3032 | spin_lock(&ioc->lock); | |
4ac845a2 | 3033 | |
216284c3 TH |
3034 | ret = radix_tree_insert(&ioc->radix_root, q->id, cic); |
3035 | if (likely(!ret)) { | |
b50b636b | 3036 | hlist_add_head(&cic->cic_list, &ioc->cic_list); |
283287a5 | 3037 | list_add(&cic->queue_list, &cfqd->cic_list); |
216284c3 TH |
3038 | cic = NULL; |
3039 | } else if (ret == -EEXIST) { | |
3040 | /* someone else already did it */ | |
3041 | ret = 0; | |
e2d74ac0 | 3042 | } |
216284c3 TH |
3043 | |
3044 | spin_unlock(&ioc->lock); | |
3045 | spin_unlock_irq(q->queue_lock); | |
3046 | ||
3047 | radix_tree_preload_end(); | |
283287a5 | 3048 | out: |
4ac845a2 JA |
3049 | if (ret) |
3050 | printk(KERN_ERR "cfq: cic link failed!\n"); | |
216284c3 TH |
3051 | if (cic) |
3052 | cfq_cic_free(cic); | |
4ac845a2 | 3053 | return ret; |
e2d74ac0 JA |
3054 | } |
3055 | ||
216284c3 TH |
3056 | /** |
3057 | * cfq_get_io_context - acquire cfq_io_context and bump refcnt on io_context | |
3058 | * @cfqd: cfqd to setup cic for | |
3059 | * @gfp_mask: allocation mask | |
3060 | * | |
3061 | * Return cfq_io_context associating @cfqd and %current->io_context and | |
3062 | * bump refcnt on io_context. If ioc or cic doesn't exist, they're created | |
3063 | * using @gfp_mask. | |
3064 | * | |
3065 | * Must be called under queue_lock which may be released and re-acquired. | |
3066 | * This function also may sleep depending on @gfp_mask. | |
1da177e4 LT |
3067 | */ |
3068 | static struct cfq_io_context * | |
e2d74ac0 | 3069 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 3070 | { |
216284c3 | 3071 | struct request_queue *q = cfqd->queue; |
6e736be7 | 3072 | struct cfq_io_context *cic = NULL; |
216284c3 TH |
3073 | struct io_context *ioc; |
3074 | int err; | |
3075 | ||
3076 | lockdep_assert_held(q->queue_lock); | |
3077 | ||
3078 | while (true) { | |
3079 | /* fast path */ | |
3080 | ioc = current->io_context; | |
3081 | if (likely(ioc)) { | |
3082 | cic = cfq_cic_lookup(cfqd, ioc); | |
3083 | if (likely(cic)) | |
3084 | break; | |
3085 | } | |
1da177e4 | 3086 | |
216284c3 TH |
3087 | /* slow path - unlock, create missing ones and retry */ |
3088 | spin_unlock_irq(q->queue_lock); | |
3089 | err = cfq_create_cic(cfqd, gfp_mask); | |
3090 | spin_lock_irq(q->queue_lock); | |
3091 | if (err) | |
3092 | return NULL; | |
3093 | } | |
1da177e4 | 3094 | |
216284c3 | 3095 | /* bump @ioc's refcnt and handle changed notifications */ |
6e736be7 TH |
3096 | get_io_context(ioc); |
3097 | ||
dc86900e TH |
3098 | if (unlikely(cic->changed)) { |
3099 | if (test_and_clear_bit(CIC_IOPRIO_CHANGED, &cic->changed)) | |
3100 | changed_ioprio(cic); | |
24610333 | 3101 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
dc86900e TH |
3102 | if (test_and_clear_bit(CIC_CGROUP_CHANGED, &cic->changed)) |
3103 | changed_cgroup(cic); | |
24610333 | 3104 | #endif |
dc86900e TH |
3105 | } |
3106 | ||
1da177e4 | 3107 | return cic; |
1da177e4 LT |
3108 | } |
3109 | ||
22e2c507 | 3110 | static void |
383cd721 | 3111 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 3112 | { |
383cd721 SL |
3113 | unsigned long elapsed = jiffies - ttime->last_end_request; |
3114 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 3115 | |
383cd721 SL |
3116 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
3117 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
3118 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
3119 | } | |
3120 | ||
3121 | static void | |
3122 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
3123 | struct cfq_io_context *cic) | |
3124 | { | |
f5f2b6ce | 3125 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 3126 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
3127 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
3128 | cfqd->cfq_slice_idle); | |
3129 | } | |
7700fc4f SL |
3130 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3131 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
3132 | #endif | |
22e2c507 | 3133 | } |
1da177e4 | 3134 | |
206dc69b | 3135 | static void |
b2c18e1e | 3136 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3137 | struct request *rq) |
206dc69b | 3138 | { |
3dde36dd | 3139 | sector_t sdist = 0; |
41647e7a | 3140 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3141 | if (cfqq->last_request_pos) { |
3142 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3143 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3144 | else | |
3145 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3146 | } | |
206dc69b | 3147 | |
3dde36dd | 3148 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3149 | if (blk_queue_nonrot(cfqd->queue)) |
3150 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3151 | else | |
3152 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3153 | } |
1da177e4 | 3154 | |
22e2c507 JA |
3155 | /* |
3156 | * Disable idle window if the process thinks too long or seeks so much that | |
3157 | * it doesn't matter | |
3158 | */ | |
3159 | static void | |
3160 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
3161 | struct cfq_io_context *cic) | |
3162 | { | |
7b679138 | 3163 | int old_idle, enable_idle; |
1be92f2f | 3164 | |
0871714e JA |
3165 | /* |
3166 | * Don't idle for async or idle io prio class | |
3167 | */ | |
3168 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3169 | return; |
3170 | ||
c265a7f4 | 3171 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3172 | |
76280aff CZ |
3173 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3174 | cfq_mark_cfqq_deep(cfqq); | |
3175 | ||
749ef9f8 CZ |
3176 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
3177 | enable_idle = 0; | |
3178 | else if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || | |
3dde36dd | 3179 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) |
22e2c507 | 3180 | enable_idle = 0; |
383cd721 SL |
3181 | else if (sample_valid(cic->ttime.ttime_samples)) { |
3182 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
3183 | enable_idle = 0; |
3184 | else | |
3185 | enable_idle = 1; | |
1da177e4 LT |
3186 | } |
3187 | ||
7b679138 JA |
3188 | if (old_idle != enable_idle) { |
3189 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3190 | if (enable_idle) | |
3191 | cfq_mark_cfqq_idle_window(cfqq); | |
3192 | else | |
3193 | cfq_clear_cfqq_idle_window(cfqq); | |
3194 | } | |
22e2c507 | 3195 | } |
1da177e4 | 3196 | |
22e2c507 JA |
3197 | /* |
3198 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3199 | * no or if we aren't sure, a 1 will cause a preempt. | |
3200 | */ | |
a6151c3a | 3201 | static bool |
22e2c507 | 3202 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3203 | struct request *rq) |
22e2c507 | 3204 | { |
6d048f53 | 3205 | struct cfq_queue *cfqq; |
22e2c507 | 3206 | |
6d048f53 JA |
3207 | cfqq = cfqd->active_queue; |
3208 | if (!cfqq) | |
a6151c3a | 3209 | return false; |
22e2c507 | 3210 | |
6d048f53 | 3211 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3212 | return false; |
22e2c507 JA |
3213 | |
3214 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3215 | return true; |
1e3335de | 3216 | |
875feb63 DS |
3217 | /* |
3218 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3219 | */ | |
3220 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3221 | return false; | |
3222 | ||
374f84ac JA |
3223 | /* |
3224 | * if the new request is sync, but the currently running queue is | |
3225 | * not, let the sync request have priority. | |
3226 | */ | |
5e705374 | 3227 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3228 | return true; |
1e3335de | 3229 | |
8682e1f1 VG |
3230 | if (new_cfqq->cfqg != cfqq->cfqg) |
3231 | return false; | |
3232 | ||
3233 | if (cfq_slice_used(cfqq)) | |
3234 | return true; | |
3235 | ||
3236 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
3237 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && | |
3238 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && | |
3239 | new_cfqq->service_tree->count == 2 && | |
3240 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3241 | return true; | |
3242 | ||
b53d1ed7 JA |
3243 | /* |
3244 | * So both queues are sync. Let the new request get disk time if | |
3245 | * it's a metadata request and the current queue is doing regular IO. | |
3246 | */ | |
65299a3b | 3247 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
3248 | return true; |
3249 | ||
3a9a3f6c DS |
3250 | /* |
3251 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3252 | */ | |
3253 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3254 | return true; |
3a9a3f6c | 3255 | |
d2d59e18 SL |
3256 | /* An idle queue should not be idle now for some reason */ |
3257 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
3258 | return true; | |
3259 | ||
1e3335de | 3260 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3261 | return false; |
1e3335de JA |
3262 | |
3263 | /* | |
3264 | * if this request is as-good as one we would expect from the | |
3265 | * current cfqq, let it preempt | |
3266 | */ | |
e9ce335d | 3267 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3268 | return true; |
1e3335de | 3269 | |
a6151c3a | 3270 | return false; |
22e2c507 JA |
3271 | } |
3272 | ||
3273 | /* | |
3274 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3275 | * let it have half of its nominal slice. | |
3276 | */ | |
3277 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3278 | { | |
f8ae6e3e SL |
3279 | struct cfq_queue *old_cfqq = cfqd->active_queue; |
3280 | ||
7b679138 | 3281 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
e5ff082e | 3282 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3283 | |
f8ae6e3e SL |
3284 | /* |
3285 | * workload type is changed, don't save slice, otherwise preempt | |
3286 | * doesn't happen | |
3287 | */ | |
3288 | if (cfqq_type(old_cfqq) != cfqq_type(cfqq)) | |
3289 | cfqq->cfqg->saved_workload_slice = 0; | |
3290 | ||
bf572256 JA |
3291 | /* |
3292 | * Put the new queue at the front of the of the current list, | |
3293 | * so we know that it will be selected next. | |
3294 | */ | |
3295 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3296 | |
3297 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3298 | |
62a37f6b JT |
3299 | cfqq->slice_end = 0; |
3300 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3301 | } |
3302 | ||
22e2c507 | 3303 | /* |
5e705374 | 3304 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3305 | * something we should do about it |
3306 | */ | |
3307 | static void | |
5e705374 JA |
3308 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3309 | struct request *rq) | |
22e2c507 | 3310 | { |
5e705374 | 3311 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 3312 | |
45333d5a | 3313 | cfqd->rq_queued++; |
65299a3b CH |
3314 | if (rq->cmd_flags & REQ_PRIO) |
3315 | cfqq->prio_pending++; | |
374f84ac | 3316 | |
383cd721 | 3317 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3318 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3319 | cfq_update_idle_window(cfqd, cfqq, cic); |
3320 | ||
b2c18e1e | 3321 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3322 | |
3323 | if (cfqq == cfqd->active_queue) { | |
3324 | /* | |
b029195d JA |
3325 | * Remember that we saw a request from this process, but |
3326 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3327 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3328 | * and merging. If the request is already larger than a single |
3329 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3330 | * merging is already done. Ditto for a busy system that |
3331 | * has other work pending, don't risk delaying until the | |
3332 | * idle timer unplug to continue working. | |
22e2c507 | 3333 | */ |
d6ceb25e | 3334 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3335 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3336 | cfqd->busy_queues > 1) { | |
812df48d | 3337 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3338 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3339 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3340 | } else { |
e98ef89b | 3341 | cfq_blkiocg_update_idle_time_stats( |
a11cdaa7 | 3342 | &cfqq->cfqg->blkg); |
bf791937 | 3343 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3344 | } |
d6ceb25e | 3345 | } |
5e705374 | 3346 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3347 | /* |
3348 | * not the active queue - expire current slice if it is | |
3349 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3350 | * has some old slice time left and is of higher priority or |
3351 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3352 | */ |
3353 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3354 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3355 | } |
1da177e4 LT |
3356 | } |
3357 | ||
165125e1 | 3358 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3359 | { |
b4878f24 | 3360 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3361 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3362 | |
7b679138 | 3363 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
fd0928df | 3364 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
1da177e4 | 3365 | |
30996f40 | 3366 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3367 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3368 | cfq_add_rq_rb(rq); |
e98ef89b | 3369 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, |
cdc1184c DS |
3370 | &cfqd->serving_group->blkg, rq_data_dir(rq), |
3371 | rq_is_sync(rq)); | |
5e705374 | 3372 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3373 | } |
3374 | ||
45333d5a AC |
3375 | /* |
3376 | * Update hw_tag based on peak queue depth over 50 samples under | |
3377 | * sufficient load. | |
3378 | */ | |
3379 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3380 | { | |
1a1238a7 SL |
3381 | struct cfq_queue *cfqq = cfqd->active_queue; |
3382 | ||
53c583d2 CZ |
3383 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3384 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3385 | |
3386 | if (cfqd->hw_tag == 1) | |
3387 | return; | |
45333d5a AC |
3388 | |
3389 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3390 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3391 | return; |
3392 | ||
1a1238a7 SL |
3393 | /* |
3394 | * If active queue hasn't enough requests and can idle, cfq might not | |
3395 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3396 | * case | |
3397 | */ | |
3398 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3399 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3400 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3401 | return; |
3402 | ||
45333d5a AC |
3403 | if (cfqd->hw_tag_samples++ < 50) |
3404 | return; | |
3405 | ||
e459dd08 | 3406 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3407 | cfqd->hw_tag = 1; |
3408 | else | |
3409 | cfqd->hw_tag = 0; | |
45333d5a AC |
3410 | } |
3411 | ||
7667aa06 VG |
3412 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3413 | { | |
3414 | struct cfq_io_context *cic = cfqd->active_cic; | |
3415 | ||
02a8f01b JT |
3416 | /* If the queue already has requests, don't wait */ |
3417 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3418 | return false; | |
3419 | ||
7667aa06 VG |
3420 | /* If there are other queues in the group, don't wait */ |
3421 | if (cfqq->cfqg->nr_cfqq > 1) | |
3422 | return false; | |
3423 | ||
7700fc4f SL |
3424 | /* the only queue in the group, but think time is big */ |
3425 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
3426 | return false; | |
3427 | ||
7667aa06 VG |
3428 | if (cfq_slice_used(cfqq)) |
3429 | return true; | |
3430 | ||
3431 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
3432 | if (cic && sample_valid(cic->ttime.ttime_samples) |
3433 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
3434 | return true; |
3435 | ||
3436 | /* | |
3437 | * If think times is less than a jiffy than ttime_mean=0 and above | |
3438 | * will not be true. It might happen that slice has not expired yet | |
3439 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
3440 | * case where think time is less than a jiffy, mark the queue wait | |
3441 | * busy if only 1 jiffy is left in the slice. | |
3442 | */ | |
3443 | if (cfqq->slice_end - jiffies == 1) | |
3444 | return true; | |
3445 | ||
3446 | return false; | |
3447 | } | |
3448 | ||
165125e1 | 3449 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3450 | { |
5e705374 | 3451 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3452 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3453 | const int sync = rq_is_sync(rq); |
b4878f24 | 3454 | unsigned long now; |
1da177e4 | 3455 | |
b4878f24 | 3456 | now = jiffies; |
33659ebb CH |
3457 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3458 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 3459 | |
45333d5a AC |
3460 | cfq_update_hw_tag(cfqd); |
3461 | ||
53c583d2 | 3462 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 3463 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 3464 | cfqd->rq_in_driver--; |
6d048f53 | 3465 | cfqq->dispatched--; |
80bdf0c7 | 3466 | (RQ_CFQG(rq))->dispatched--; |
e98ef89b VG |
3467 | cfq_blkiocg_update_completion_stats(&cfqq->cfqg->blkg, |
3468 | rq_start_time_ns(rq), rq_io_start_time_ns(rq), | |
3469 | rq_data_dir(rq), rq_is_sync(rq)); | |
1da177e4 | 3470 | |
53c583d2 | 3471 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 3472 | |
365722bb | 3473 | if (sync) { |
f5f2b6ce SL |
3474 | struct cfq_rb_root *service_tree; |
3475 | ||
383cd721 | 3476 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
3477 | |
3478 | if (cfq_cfqq_on_rr(cfqq)) | |
3479 | service_tree = cfqq->service_tree; | |
3480 | else | |
3481 | service_tree = service_tree_for(cfqq->cfqg, | |
3482 | cfqq_prio(cfqq), cfqq_type(cfqq)); | |
3483 | service_tree->ttime.last_end_request = now; | |
573412b2 CZ |
3484 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3485 | cfqd->last_delayed_sync = now; | |
365722bb | 3486 | } |
caaa5f9f | 3487 | |
7700fc4f SL |
3488 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3489 | cfqq->cfqg->ttime.last_end_request = now; | |
3490 | #endif | |
3491 | ||
caaa5f9f JA |
3492 | /* |
3493 | * If this is the active queue, check if it needs to be expired, | |
3494 | * or if we want to idle in case it has no pending requests. | |
3495 | */ | |
3496 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3497 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3498 | ||
44f7c160 JA |
3499 | if (cfq_cfqq_slice_new(cfqq)) { |
3500 | cfq_set_prio_slice(cfqd, cfqq); | |
3501 | cfq_clear_cfqq_slice_new(cfqq); | |
3502 | } | |
f75edf2d VG |
3503 | |
3504 | /* | |
7667aa06 VG |
3505 | * Should we wait for next request to come in before we expire |
3506 | * the queue. | |
f75edf2d | 3507 | */ |
7667aa06 | 3508 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
3509 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3510 | if (!cfqd->cfq_slice_idle) | |
3511 | extend_sl = cfqd->cfq_group_idle; | |
3512 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 3513 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 3514 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
3515 | } |
3516 | ||
a36e71f9 | 3517 | /* |
8e550632 CZ |
3518 | * Idling is not enabled on: |
3519 | * - expired queues | |
3520 | * - idle-priority queues | |
3521 | * - async queues | |
3522 | * - queues with still some requests queued | |
3523 | * - when there is a close cooperator | |
a36e71f9 | 3524 | */ |
0871714e | 3525 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 3526 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3527 | else if (sync && cfqq_empty && |
3528 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 3529 | cfq_arm_slice_timer(cfqd); |
8e550632 | 3530 | } |
caaa5f9f | 3531 | } |
6d048f53 | 3532 | |
53c583d2 | 3533 | if (!cfqd->rq_in_driver) |
23e018a1 | 3534 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3535 | } |
3536 | ||
89850f7e | 3537 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3538 | { |
1b379d8d | 3539 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3540 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3541 | return ELV_MQUEUE_MUST; |
3b18152c | 3542 | } |
1da177e4 | 3543 | |
22e2c507 | 3544 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3545 | } |
3546 | ||
165125e1 | 3547 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3548 | { |
3549 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3550 | struct task_struct *tsk = current; | |
91fac317 | 3551 | struct cfq_io_context *cic; |
22e2c507 JA |
3552 | struct cfq_queue *cfqq; |
3553 | ||
3554 | /* | |
3555 | * don't force setup of a queue from here, as a call to may_queue | |
3556 | * does not necessarily imply that a request actually will be queued. | |
3557 | * so just lookup a possibly existing queue, or return 'may queue' | |
3558 | * if that fails | |
3559 | */ | |
4ac845a2 | 3560 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3561 | if (!cic) |
3562 | return ELV_MQUEUE_MAY; | |
3563 | ||
b0b78f81 | 3564 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3565 | if (cfqq) { |
fd0928df | 3566 | cfq_init_prio_data(cfqq, cic->ioc); |
22e2c507 | 3567 | |
89850f7e | 3568 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3569 | } |
3570 | ||
3571 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3572 | } |
3573 | ||
1da177e4 LT |
3574 | /* |
3575 | * queue lock held here | |
3576 | */ | |
bb37b94c | 3577 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3578 | { |
5e705374 | 3579 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3580 | |
5e705374 | 3581 | if (cfqq) { |
22e2c507 | 3582 | const int rw = rq_data_dir(rq); |
1da177e4 | 3583 | |
22e2c507 JA |
3584 | BUG_ON(!cfqq->allocated[rw]); |
3585 | cfqq->allocated[rw]--; | |
1da177e4 | 3586 | |
b2efa052 | 3587 | put_io_context(RQ_CIC(rq)->ioc, cfqq->cfqd->queue); |
1da177e4 | 3588 | |
c186794d MS |
3589 | rq->elevator_private[0] = NULL; |
3590 | rq->elevator_private[1] = NULL; | |
1da177e4 | 3591 | |
7f1dc8a2 VG |
3592 | /* Put down rq reference on cfqg */ |
3593 | cfq_put_cfqg(RQ_CFQG(rq)); | |
c186794d | 3594 | rq->elevator_private[2] = NULL; |
7f1dc8a2 | 3595 | |
1da177e4 LT |
3596 | cfq_put_queue(cfqq); |
3597 | } | |
3598 | } | |
3599 | ||
df5fe3e8 JM |
3600 | static struct cfq_queue * |
3601 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_context *cic, | |
3602 | struct cfq_queue *cfqq) | |
3603 | { | |
3604 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3605 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3606 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3607 | cfq_put_queue(cfqq); |
3608 | return cic_to_cfqq(cic, 1); | |
3609 | } | |
3610 | ||
e6c5bc73 JM |
3611 | /* |
3612 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3613 | * was the last process referring to said cfqq. | |
3614 | */ | |
3615 | static struct cfq_queue * | |
3616 | split_cfqq(struct cfq_io_context *cic, struct cfq_queue *cfqq) | |
3617 | { | |
3618 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
3619 | cfqq->pid = current->pid; |
3620 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 3621 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
3622 | return cfqq; |
3623 | } | |
3624 | ||
3625 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
3626 | |
3627 | cfq_put_cooperator(cfqq); | |
3628 | ||
e6c5bc73 JM |
3629 | cfq_put_queue(cfqq); |
3630 | return NULL; | |
3631 | } | |
1da177e4 | 3632 | /* |
22e2c507 | 3633 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3634 | */ |
22e2c507 | 3635 | static int |
165125e1 | 3636 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
3637 | { |
3638 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3639 | struct cfq_io_context *cic; | |
3640 | const int rw = rq_data_dir(rq); | |
a6151c3a | 3641 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3642 | struct cfq_queue *cfqq; |
1da177e4 LT |
3643 | |
3644 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3645 | ||
216284c3 | 3646 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3647 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 JA |
3648 | if (!cic) |
3649 | goto queue_fail; | |
3650 | ||
e6c5bc73 | 3651 | new_queue: |
91fac317 | 3652 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3653 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
fd0928df | 3654 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
91fac317 | 3655 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3656 | } else { |
e6c5bc73 JM |
3657 | /* |
3658 | * If the queue was seeky for too long, break it apart. | |
3659 | */ | |
ae54abed | 3660 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
3661 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
3662 | cfqq = split_cfqq(cic, cfqq); | |
3663 | if (!cfqq) | |
3664 | goto new_queue; | |
3665 | } | |
3666 | ||
df5fe3e8 JM |
3667 | /* |
3668 | * Check to see if this queue is scheduled to merge with | |
3669 | * another, closely cooperating queue. The merging of | |
3670 | * queues happens here as it must be done in process context. | |
3671 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3672 | */ | |
3673 | if (cfqq->new_cfqq) | |
3674 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3675 | } |
1da177e4 LT |
3676 | |
3677 | cfqq->allocated[rw]++; | |
1da177e4 | 3678 | |
6fae9c25 | 3679 | cfqq->ref++; |
c186794d MS |
3680 | rq->elevator_private[0] = cic; |
3681 | rq->elevator_private[1] = cfqq; | |
3682 | rq->elevator_private[2] = cfq_ref_get_cfqg(cfqq->cfqg); | |
216284c3 | 3683 | spin_unlock_irq(q->queue_lock); |
5e705374 | 3684 | return 0; |
1da177e4 | 3685 | |
22e2c507 | 3686 | queue_fail: |
23e018a1 | 3687 | cfq_schedule_dispatch(cfqd); |
216284c3 | 3688 | spin_unlock_irq(q->queue_lock); |
7b679138 | 3689 | cfq_log(cfqd, "set_request fail"); |
1da177e4 LT |
3690 | return 1; |
3691 | } | |
3692 | ||
65f27f38 | 3693 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3694 | { |
65f27f38 | 3695 | struct cfq_data *cfqd = |
23e018a1 | 3696 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3697 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3698 | |
40bb54d1 | 3699 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 3700 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3701 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3702 | } |
3703 | ||
3704 | /* | |
3705 | * Timer running if the active_queue is currently idling inside its time slice | |
3706 | */ | |
3707 | static void cfq_idle_slice_timer(unsigned long data) | |
3708 | { | |
3709 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3710 | struct cfq_queue *cfqq; | |
3711 | unsigned long flags; | |
3c6bd2f8 | 3712 | int timed_out = 1; |
22e2c507 | 3713 | |
7b679138 JA |
3714 | cfq_log(cfqd, "idle timer fired"); |
3715 | ||
22e2c507 JA |
3716 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3717 | ||
fe094d98 JA |
3718 | cfqq = cfqd->active_queue; |
3719 | if (cfqq) { | |
3c6bd2f8 JA |
3720 | timed_out = 0; |
3721 | ||
b029195d JA |
3722 | /* |
3723 | * We saw a request before the queue expired, let it through | |
3724 | */ | |
3725 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3726 | goto out_kick; | |
3727 | ||
22e2c507 JA |
3728 | /* |
3729 | * expired | |
3730 | */ | |
44f7c160 | 3731 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
3732 | goto expire; |
3733 | ||
3734 | /* | |
3735 | * only expire and reinvoke request handler, if there are | |
3736 | * other queues with pending requests | |
3737 | */ | |
caaa5f9f | 3738 | if (!cfqd->busy_queues) |
22e2c507 | 3739 | goto out_cont; |
22e2c507 JA |
3740 | |
3741 | /* | |
3742 | * not expired and it has a request pending, let it dispatch | |
3743 | */ | |
75e50984 | 3744 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3745 | goto out_kick; |
76280aff CZ |
3746 | |
3747 | /* | |
3748 | * Queue depth flag is reset only when the idle didn't succeed | |
3749 | */ | |
3750 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
3751 | } |
3752 | expire: | |
e5ff082e | 3753 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 3754 | out_kick: |
23e018a1 | 3755 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
3756 | out_cont: |
3757 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3758 | } | |
3759 | ||
3b18152c JA |
3760 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3761 | { | |
3762 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 3763 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 3764 | } |
22e2c507 | 3765 | |
c2dea2d1 VT |
3766 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3767 | { | |
3768 | int i; | |
3769 | ||
3770 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
3771 | if (cfqd->async_cfqq[0][i]) | |
3772 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
3773 | if (cfqd->async_cfqq[1][i]) | |
3774 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 3775 | } |
2389d1ef ON |
3776 | |
3777 | if (cfqd->async_idle_cfqq) | |
3778 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
3779 | } |
3780 | ||
b374d18a | 3781 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 3782 | { |
22e2c507 | 3783 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 3784 | struct request_queue *q = cfqd->queue; |
56edf7d7 | 3785 | bool wait = false; |
22e2c507 | 3786 | |
3b18152c | 3787 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 3788 | |
d9ff4187 | 3789 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3790 | |
d9ff4187 | 3791 | if (cfqd->active_queue) |
e5ff082e | 3792 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
3793 | |
3794 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
3795 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
3796 | struct cfq_io_context, | |
3797 | queue_list); | |
b2efa052 | 3798 | struct io_context *ioc = cic->ioc; |
89850f7e | 3799 | |
b2efa052 | 3800 | spin_lock(&ioc->lock); |
283287a5 | 3801 | cfq_exit_cic(cic); |
b2efa052 TH |
3802 | cfq_release_cic(cic); |
3803 | spin_unlock(&ioc->lock); | |
d9ff4187 | 3804 | } |
e2d74ac0 | 3805 | |
c2dea2d1 | 3806 | cfq_put_async_queues(cfqd); |
b1c35769 | 3807 | cfq_release_cfq_groups(cfqd); |
56edf7d7 VG |
3808 | |
3809 | /* | |
3810 | * If there are groups which we could not unlink from blkcg list, | |
3811 | * wait for a rcu period for them to be freed. | |
3812 | */ | |
3813 | if (cfqd->nr_blkcg_linked_grps) | |
3814 | wait = true; | |
15c31be4 | 3815 | |
d9ff4187 | 3816 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
3817 | |
3818 | cfq_shutdown_timer_wq(cfqd); | |
3819 | ||
56edf7d7 VG |
3820 | /* |
3821 | * Wait for cfqg->blkg->key accessors to exit their grace periods. | |
3822 | * Do this wait only if there are other unlinked groups out | |
3823 | * there. This can happen if cgroup deletion path claimed the | |
3824 | * responsibility of cleaning up a group before queue cleanup code | |
3825 | * get to the group. | |
3826 | * | |
3827 | * Do not call synchronize_rcu() unconditionally as there are drivers | |
3828 | * which create/delete request queue hundreds of times during scan/boot | |
3829 | * and synchronize_rcu() can take significant time and slow down boot. | |
3830 | */ | |
3831 | if (wait) | |
3832 | synchronize_rcu(); | |
2abae55f VG |
3833 | |
3834 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
3835 | /* Free up per cpu stats for root group */ | |
3836 | free_percpu(cfqd->root_group.blkg.stats_cpu); | |
3837 | #endif | |
56edf7d7 | 3838 | kfree(cfqd); |
1da177e4 LT |
3839 | } |
3840 | ||
165125e1 | 3841 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
3842 | { |
3843 | struct cfq_data *cfqd; | |
718eee05 | 3844 | int i, j; |
cdb16e8f | 3845 | struct cfq_group *cfqg; |
615f0259 | 3846 | struct cfq_rb_root *st; |
1da177e4 | 3847 | |
94f6030c | 3848 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
a73f730d | 3849 | if (!cfqd) |
bc1c1169 | 3850 | return NULL; |
80b15c73 | 3851 | |
1fa8f6d6 VG |
3852 | /* Init root service tree */ |
3853 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
3854 | ||
cdb16e8f VG |
3855 | /* Init root group */ |
3856 | cfqg = &cfqd->root_group; | |
615f0259 VG |
3857 | for_each_cfqg_st(cfqg, i, j, st) |
3858 | *st = CFQ_RB_ROOT; | |
1fa8f6d6 | 3859 | RB_CLEAR_NODE(&cfqg->rb_node); |
26a2ac00 | 3860 | |
25bc6b07 VG |
3861 | /* Give preference to root group over other groups */ |
3862 | cfqg->weight = 2*BLKIO_WEIGHT_DEFAULT; | |
3863 | ||
25fb5169 | 3864 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
b1c35769 | 3865 | /* |
56edf7d7 VG |
3866 | * Set root group reference to 2. One reference will be dropped when |
3867 | * all groups on cfqd->cfqg_list are being deleted during queue exit. | |
3868 | * Other reference will remain there as we don't want to delete this | |
3869 | * group as it is statically allocated and gets destroyed when | |
3870 | * throtl_data goes away. | |
b1c35769 | 3871 | */ |
56edf7d7 | 3872 | cfqg->ref = 2; |
5624a4e4 VG |
3873 | |
3874 | if (blkio_alloc_blkg_stats(&cfqg->blkg)) { | |
3875 | kfree(cfqg); | |
3876 | kfree(cfqd); | |
3877 | return NULL; | |
3878 | } | |
3879 | ||
dcf097b2 | 3880 | rcu_read_lock(); |
5624a4e4 | 3881 | |
e98ef89b VG |
3882 | cfq_blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg, |
3883 | (void *)cfqd, 0); | |
dcf097b2 | 3884 | rcu_read_unlock(); |
56edf7d7 VG |
3885 | cfqd->nr_blkcg_linked_grps++; |
3886 | ||
3887 | /* Add group on cfqd->cfqg_list */ | |
3888 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); | |
25fb5169 | 3889 | #endif |
26a2ac00 JA |
3890 | /* |
3891 | * Not strictly needed (since RB_ROOT just clears the node and we | |
3892 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
3893 | * to add magic to the rb code | |
3894 | */ | |
3895 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
3896 | cfqd->prio_trees[i] = RB_ROOT; | |
3897 | ||
6118b70b JA |
3898 | /* |
3899 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
3900 | * Grab a permanent reference to it, so that the normal code flow | |
3901 | * will not attempt to free it. | |
3902 | */ | |
3903 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 3904 | cfqd->oom_cfqq.ref++; |
cdb16e8f | 3905 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group); |
6118b70b | 3906 | |
d9ff4187 | 3907 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 3908 | |
1da177e4 | 3909 | cfqd->queue = q; |
1da177e4 | 3910 | |
22e2c507 JA |
3911 | init_timer(&cfqd->idle_slice_timer); |
3912 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
3913 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
3914 | ||
23e018a1 | 3915 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 3916 | |
1da177e4 | 3917 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
3918 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
3919 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
3920 | cfqd->cfq_back_max = cfq_back_max; |
3921 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
3922 | cfqd->cfq_slice[0] = cfq_slice_async; |
3923 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
3924 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
3925 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
80bdf0c7 | 3926 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 3927 | cfqd->cfq_latency = 1; |
e459dd08 | 3928 | cfqd->hw_tag = -1; |
edc71131 CZ |
3929 | /* |
3930 | * we optimistically start assuming sync ops weren't delayed in last | |
3931 | * second, in order to have larger depth for async operations. | |
3932 | */ | |
573412b2 | 3933 | cfqd->last_delayed_sync = jiffies - HZ; |
bc1c1169 | 3934 | return cfqd; |
1da177e4 LT |
3935 | } |
3936 | ||
3937 | static void cfq_slab_kill(void) | |
3938 | { | |
d6de8be7 JA |
3939 | /* |
3940 | * Caller already ensured that pending RCU callbacks are completed, | |
3941 | * so we should have no busy allocations at this point. | |
3942 | */ | |
1da177e4 LT |
3943 | if (cfq_pool) |
3944 | kmem_cache_destroy(cfq_pool); | |
3945 | if (cfq_ioc_pool) | |
3946 | kmem_cache_destroy(cfq_ioc_pool); | |
3947 | } | |
3948 | ||
3949 | static int __init cfq_slab_setup(void) | |
3950 | { | |
0a31bd5f | 3951 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
1da177e4 LT |
3952 | if (!cfq_pool) |
3953 | goto fail; | |
3954 | ||
34e6bbf2 | 3955 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
1da177e4 LT |
3956 | if (!cfq_ioc_pool) |
3957 | goto fail; | |
3958 | ||
3959 | return 0; | |
3960 | fail: | |
3961 | cfq_slab_kill(); | |
3962 | return -ENOMEM; | |
3963 | } | |
3964 | ||
1da177e4 LT |
3965 | /* |
3966 | * sysfs parts below --> | |
3967 | */ | |
1da177e4 LT |
3968 | static ssize_t |
3969 | cfq_var_show(unsigned int var, char *page) | |
3970 | { | |
3971 | return sprintf(page, "%d\n", var); | |
3972 | } | |
3973 | ||
3974 | static ssize_t | |
3975 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
3976 | { | |
3977 | char *p = (char *) page; | |
3978 | ||
3979 | *var = simple_strtoul(p, &p, 10); | |
3980 | return count; | |
3981 | } | |
3982 | ||
1da177e4 | 3983 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 3984 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 3985 | { \ |
3d1ab40f | 3986 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3987 | unsigned int __data = __VAR; \ |
3988 | if (__CONV) \ | |
3989 | __data = jiffies_to_msecs(__data); \ | |
3990 | return cfq_var_show(__data, (page)); \ | |
3991 | } | |
3992 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
3993 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
3994 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
3995 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
3996 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 3997 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 3998 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
3999 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
4000 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
4001 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 4002 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
1da177e4 LT |
4003 | #undef SHOW_FUNCTION |
4004 | ||
4005 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 4006 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 4007 | { \ |
3d1ab40f | 4008 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4009 | unsigned int __data; \ |
4010 | int ret = cfq_var_store(&__data, (page), count); \ | |
4011 | if (__data < (MIN)) \ | |
4012 | __data = (MIN); \ | |
4013 | else if (__data > (MAX)) \ | |
4014 | __data = (MAX); \ | |
4015 | if (__CONV) \ | |
4016 | *(__PTR) = msecs_to_jiffies(__data); \ | |
4017 | else \ | |
4018 | *(__PTR) = __data; \ | |
4019 | return ret; \ | |
4020 | } | |
4021 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4022 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4023 | UINT_MAX, 1); | |
4024 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4025 | UINT_MAX, 1); | |
e572ec7e | 4026 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4027 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4028 | UINT_MAX, 0); | |
22e2c507 | 4029 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4030 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4031 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4032 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4033 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4034 | UINT_MAX, 0); | |
963b72fc | 4035 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
1da177e4 LT |
4036 | #undef STORE_FUNCTION |
4037 | ||
e572ec7e AV |
4038 | #define CFQ_ATTR(name) \ |
4039 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4040 | ||
4041 | static struct elv_fs_entry cfq_attrs[] = { | |
4042 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4043 | CFQ_ATTR(fifo_expire_sync), |
4044 | CFQ_ATTR(fifo_expire_async), | |
4045 | CFQ_ATTR(back_seek_max), | |
4046 | CFQ_ATTR(back_seek_penalty), | |
4047 | CFQ_ATTR(slice_sync), | |
4048 | CFQ_ATTR(slice_async), | |
4049 | CFQ_ATTR(slice_async_rq), | |
4050 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 4051 | CFQ_ATTR(group_idle), |
963b72fc | 4052 | CFQ_ATTR(low_latency), |
e572ec7e | 4053 | __ATTR_NULL |
1da177e4 LT |
4054 | }; |
4055 | ||
1da177e4 LT |
4056 | static struct elevator_type iosched_cfq = { |
4057 | .ops = { | |
4058 | .elevator_merge_fn = cfq_merge, | |
4059 | .elevator_merged_fn = cfq_merged_request, | |
4060 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 4061 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 4062 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4063 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4064 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4065 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 4066 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 4067 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
4068 | .elevator_former_req_fn = elv_rb_former_request, |
4069 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
4070 | .elevator_set_req_fn = cfq_set_request, |
4071 | .elevator_put_req_fn = cfq_put_request, | |
4072 | .elevator_may_queue_fn = cfq_may_queue, | |
4073 | .elevator_init_fn = cfq_init_queue, | |
4074 | .elevator_exit_fn = cfq_exit_queue, | |
4075 | }, | |
3d1ab40f | 4076 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
4077 | .elevator_name = "cfq", |
4078 | .elevator_owner = THIS_MODULE, | |
4079 | }; | |
4080 | ||
3e252066 VG |
4081 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4082 | static struct blkio_policy_type blkio_policy_cfq = { | |
4083 | .ops = { | |
4084 | .blkio_unlink_group_fn = cfq_unlink_blkio_group, | |
4085 | .blkio_update_group_weight_fn = cfq_update_blkio_group_weight, | |
4086 | }, | |
062a644d | 4087 | .plid = BLKIO_POLICY_PROP, |
3e252066 VG |
4088 | }; |
4089 | #else | |
4090 | static struct blkio_policy_type blkio_policy_cfq; | |
4091 | #endif | |
4092 | ||
1da177e4 LT |
4093 | static int __init cfq_init(void) |
4094 | { | |
22e2c507 JA |
4095 | /* |
4096 | * could be 0 on HZ < 1000 setups | |
4097 | */ | |
4098 | if (!cfq_slice_async) | |
4099 | cfq_slice_async = 1; | |
4100 | if (!cfq_slice_idle) | |
4101 | cfq_slice_idle = 1; | |
4102 | ||
80bdf0c7 VG |
4103 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4104 | if (!cfq_group_idle) | |
4105 | cfq_group_idle = 1; | |
4106 | #else | |
4107 | cfq_group_idle = 0; | |
4108 | #endif | |
1da177e4 LT |
4109 | if (cfq_slab_setup()) |
4110 | return -ENOMEM; | |
4111 | ||
2fdd82bd | 4112 | elv_register(&iosched_cfq); |
3e252066 | 4113 | blkio_policy_register(&blkio_policy_cfq); |
1da177e4 | 4114 | |
2fdd82bd | 4115 | return 0; |
1da177e4 LT |
4116 | } |
4117 | ||
4118 | static void __exit cfq_exit(void) | |
4119 | { | |
3e252066 | 4120 | blkio_policy_unregister(&blkio_policy_cfq); |
1da177e4 | 4121 | elv_unregister(&iosched_cfq); |
b50b636b | 4122 | rcu_barrier(); /* make sure all cic RCU frees are complete */ |
83521d3e | 4123 | cfq_slab_kill(); |
1da177e4 LT |
4124 | } |
4125 | ||
4126 | module_init(cfq_init); | |
4127 | module_exit(cfq_exit); | |
4128 | ||
4129 | MODULE_AUTHOR("Jens Axboe"); | |
4130 | MODULE_LICENSE("GPL"); | |
4131 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |