Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * CFQ, or complete fairness queueing, disk scheduler. |
3 | * | |
4 | * Based on ideas from a previously unfinished io | |
5 | * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli. | |
6 | * | |
0fe23479 | 7 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
1da177e4 | 8 | */ |
1da177e4 | 9 | #include <linux/module.h> |
5a0e3ad6 | 10 | #include <linux/slab.h> |
1cc9be68 AV |
11 | #include <linux/blkdev.h> |
12 | #include <linux/elevator.h> | |
ad5ebd2f | 13 | #include <linux/jiffies.h> |
1da177e4 | 14 | #include <linux/rbtree.h> |
22e2c507 | 15 | #include <linux/ioprio.h> |
7b679138 | 16 | #include <linux/blktrace_api.h> |
6e736be7 | 17 | #include "blk.h" |
e98ef89b | 18 | #include "cfq.h" |
1da177e4 | 19 | |
0381411e TH |
20 | static struct blkio_policy_type blkio_policy_cfq; |
21 | ||
1da177e4 LT |
22 | /* |
23 | * tunables | |
24 | */ | |
fe094d98 | 25 | /* max queue in one round of service */ |
abc3c744 | 26 | static const int cfq_quantum = 8; |
64100099 | 27 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
28 | /* maximum backwards seek, in KiB */ |
29 | static const int cfq_back_max = 16 * 1024; | |
30 | /* penalty of a backwards seek */ | |
31 | static const int cfq_back_penalty = 2; | |
64100099 | 32 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 33 | static int cfq_slice_async = HZ / 25; |
64100099 | 34 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 35 | static int cfq_slice_idle = HZ / 125; |
80bdf0c7 | 36 | static int cfq_group_idle = HZ / 125; |
5db5d642 CZ |
37 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
38 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 39 | |
d9e7620e | 40 | /* |
0871714e | 41 | * offset from end of service tree |
d9e7620e | 42 | */ |
0871714e | 43 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
44 | |
45 | /* | |
46 | * below this threshold, we consider thinktime immediate | |
47 | */ | |
48 | #define CFQ_MIN_TT (2) | |
49 | ||
22e2c507 | 50 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 51 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 52 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 53 | |
3dde36dd | 54 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 55 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 56 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 57 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 58 | |
a612fddf TH |
59 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
60 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) | |
61 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) | |
1da177e4 | 62 | |
e18b890b | 63 | static struct kmem_cache *cfq_pool; |
1da177e4 | 64 | |
22e2c507 JA |
65 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
66 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
67 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
68 | ||
206dc69b | 69 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 70 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 71 | |
c5869807 TH |
72 | struct cfq_ttime { |
73 | unsigned long last_end_request; | |
74 | ||
75 | unsigned long ttime_total; | |
76 | unsigned long ttime_samples; | |
77 | unsigned long ttime_mean; | |
78 | }; | |
79 | ||
cc09e299 JA |
80 | /* |
81 | * Most of our rbtree usage is for sorting with min extraction, so | |
82 | * if we cache the leftmost node we don't have to walk down the tree | |
83 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
84 | * move this into the elevator for the rq sorting as well. | |
85 | */ | |
86 | struct cfq_rb_root { | |
87 | struct rb_root rb; | |
88 | struct rb_node *left; | |
aa6f6a3d | 89 | unsigned count; |
73e9ffdd | 90 | unsigned total_weight; |
1fa8f6d6 | 91 | u64 min_vdisktime; |
f5f2b6ce | 92 | struct cfq_ttime ttime; |
cc09e299 | 93 | }; |
f5f2b6ce SL |
94 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
95 | .ttime = {.last_end_request = jiffies,},} | |
cc09e299 | 96 | |
6118b70b JA |
97 | /* |
98 | * Per process-grouping structure | |
99 | */ | |
100 | struct cfq_queue { | |
101 | /* reference count */ | |
30d7b944 | 102 | int ref; |
6118b70b JA |
103 | /* various state flags, see below */ |
104 | unsigned int flags; | |
105 | /* parent cfq_data */ | |
106 | struct cfq_data *cfqd; | |
107 | /* service_tree member */ | |
108 | struct rb_node rb_node; | |
109 | /* service_tree key */ | |
110 | unsigned long rb_key; | |
111 | /* prio tree member */ | |
112 | struct rb_node p_node; | |
113 | /* prio tree root we belong to, if any */ | |
114 | struct rb_root *p_root; | |
115 | /* sorted list of pending requests */ | |
116 | struct rb_root sort_list; | |
117 | /* if fifo isn't expired, next request to serve */ | |
118 | struct request *next_rq; | |
119 | /* requests queued in sort_list */ | |
120 | int queued[2]; | |
121 | /* currently allocated requests */ | |
122 | int allocated[2]; | |
123 | /* fifo list of requests in sort_list */ | |
124 | struct list_head fifo; | |
125 | ||
dae739eb VG |
126 | /* time when queue got scheduled in to dispatch first request. */ |
127 | unsigned long dispatch_start; | |
f75edf2d | 128 | unsigned int allocated_slice; |
c4081ba5 | 129 | unsigned int slice_dispatch; |
dae739eb VG |
130 | /* time when first request from queue completed and slice started. */ |
131 | unsigned long slice_start; | |
6118b70b JA |
132 | unsigned long slice_end; |
133 | long slice_resid; | |
6118b70b | 134 | |
65299a3b CH |
135 | /* pending priority requests */ |
136 | int prio_pending; | |
6118b70b JA |
137 | /* number of requests that are on the dispatch list or inside driver */ |
138 | int dispatched; | |
139 | ||
140 | /* io prio of this group */ | |
141 | unsigned short ioprio, org_ioprio; | |
4aede84b | 142 | unsigned short ioprio_class; |
6118b70b | 143 | |
c4081ba5 RK |
144 | pid_t pid; |
145 | ||
3dde36dd | 146 | u32 seek_history; |
b2c18e1e JM |
147 | sector_t last_request_pos; |
148 | ||
aa6f6a3d | 149 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 150 | struct cfq_queue *new_cfqq; |
cdb16e8f | 151 | struct cfq_group *cfqg; |
c4e7893e VG |
152 | /* Number of sectors dispatched from queue in single dispatch round */ |
153 | unsigned long nr_sectors; | |
6118b70b JA |
154 | }; |
155 | ||
c0324a02 | 156 | /* |
718eee05 | 157 | * First index in the service_trees. |
c0324a02 CZ |
158 | * IDLE is handled separately, so it has negative index |
159 | */ | |
160 | enum wl_prio_t { | |
c0324a02 | 161 | BE_WORKLOAD = 0, |
615f0259 VG |
162 | RT_WORKLOAD = 1, |
163 | IDLE_WORKLOAD = 2, | |
b4627321 | 164 | CFQ_PRIO_NR, |
c0324a02 CZ |
165 | }; |
166 | ||
718eee05 CZ |
167 | /* |
168 | * Second index in the service_trees. | |
169 | */ | |
170 | enum wl_type_t { | |
171 | ASYNC_WORKLOAD = 0, | |
172 | SYNC_NOIDLE_WORKLOAD = 1, | |
173 | SYNC_WORKLOAD = 2 | |
174 | }; | |
175 | ||
cdb16e8f VG |
176 | /* This is per cgroup per device grouping structure */ |
177 | struct cfq_group { | |
1fa8f6d6 VG |
178 | /* group service_tree member */ |
179 | struct rb_node rb_node; | |
180 | ||
181 | /* group service_tree key */ | |
182 | u64 vdisktime; | |
25bc6b07 | 183 | unsigned int weight; |
8184f93e JT |
184 | unsigned int new_weight; |
185 | bool needs_update; | |
1fa8f6d6 VG |
186 | |
187 | /* number of cfqq currently on this group */ | |
188 | int nr_cfqq; | |
189 | ||
cdb16e8f | 190 | /* |
4495a7d4 | 191 | * Per group busy queues average. Useful for workload slice calc. We |
b4627321 VG |
192 | * create the array for each prio class but at run time it is used |
193 | * only for RT and BE class and slot for IDLE class remains unused. | |
194 | * This is primarily done to avoid confusion and a gcc warning. | |
195 | */ | |
196 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; | |
197 | /* | |
198 | * rr lists of queues with requests. We maintain service trees for | |
199 | * RT and BE classes. These trees are subdivided in subclasses | |
200 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE | |
201 | * class there is no subclassification and all the cfq queues go on | |
202 | * a single tree service_tree_idle. | |
cdb16e8f VG |
203 | * Counts are embedded in the cfq_rb_root |
204 | */ | |
205 | struct cfq_rb_root service_trees[2][3]; | |
206 | struct cfq_rb_root service_tree_idle; | |
dae739eb VG |
207 | |
208 | unsigned long saved_workload_slice; | |
209 | enum wl_type_t saved_workload; | |
210 | enum wl_prio_t saved_serving_prio; | |
4eef3049 | 211 | |
80bdf0c7 VG |
212 | /* number of requests that are on the dispatch list or inside driver */ |
213 | int dispatched; | |
7700fc4f | 214 | struct cfq_ttime ttime; |
cdb16e8f | 215 | }; |
718eee05 | 216 | |
c5869807 TH |
217 | struct cfq_io_cq { |
218 | struct io_cq icq; /* must be the first member */ | |
219 | struct cfq_queue *cfqq[2]; | |
220 | struct cfq_ttime ttime; | |
598971bf TH |
221 | int ioprio; /* the current ioprio */ |
222 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
223 | uint64_t blkcg_id; /* the current blkcg ID */ | |
224 | #endif | |
c5869807 TH |
225 | }; |
226 | ||
22e2c507 JA |
227 | /* |
228 | * Per block device queue structure | |
229 | */ | |
1da177e4 | 230 | struct cfq_data { |
165125e1 | 231 | struct request_queue *queue; |
1fa8f6d6 VG |
232 | /* Root service tree for cfq_groups */ |
233 | struct cfq_rb_root grp_service_tree; | |
f51b802c | 234 | struct cfq_group *root_group; |
22e2c507 | 235 | |
c0324a02 CZ |
236 | /* |
237 | * The priority currently being served | |
22e2c507 | 238 | */ |
c0324a02 | 239 | enum wl_prio_t serving_prio; |
718eee05 CZ |
240 | enum wl_type_t serving_type; |
241 | unsigned long workload_expires; | |
cdb16e8f | 242 | struct cfq_group *serving_group; |
a36e71f9 JA |
243 | |
244 | /* | |
245 | * Each priority tree is sorted by next_request position. These | |
246 | * trees are used when determining if two or more queues are | |
247 | * interleaving requests (see cfq_close_cooperator). | |
248 | */ | |
249 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
250 | ||
22e2c507 | 251 | unsigned int busy_queues; |
ef8a41df | 252 | unsigned int busy_sync_queues; |
22e2c507 | 253 | |
53c583d2 CZ |
254 | int rq_in_driver; |
255 | int rq_in_flight[2]; | |
45333d5a AC |
256 | |
257 | /* | |
258 | * queue-depth detection | |
259 | */ | |
260 | int rq_queued; | |
25776e35 | 261 | int hw_tag; |
e459dd08 CZ |
262 | /* |
263 | * hw_tag can be | |
264 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
265 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
266 | * 0 => no NCQ | |
267 | */ | |
268 | int hw_tag_est_depth; | |
269 | unsigned int hw_tag_samples; | |
1da177e4 | 270 | |
22e2c507 JA |
271 | /* |
272 | * idle window management | |
273 | */ | |
274 | struct timer_list idle_slice_timer; | |
23e018a1 | 275 | struct work_struct unplug_work; |
1da177e4 | 276 | |
22e2c507 | 277 | struct cfq_queue *active_queue; |
c5869807 | 278 | struct cfq_io_cq *active_cic; |
22e2c507 | 279 | |
c2dea2d1 VT |
280 | /* |
281 | * async queue for each priority case | |
282 | */ | |
283 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
284 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 285 | |
6d048f53 | 286 | sector_t last_position; |
1da177e4 | 287 | |
1da177e4 LT |
288 | /* |
289 | * tunables, see top of file | |
290 | */ | |
291 | unsigned int cfq_quantum; | |
22e2c507 | 292 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
293 | unsigned int cfq_back_penalty; |
294 | unsigned int cfq_back_max; | |
22e2c507 JA |
295 | unsigned int cfq_slice[2]; |
296 | unsigned int cfq_slice_async_rq; | |
297 | unsigned int cfq_slice_idle; | |
80bdf0c7 | 298 | unsigned int cfq_group_idle; |
963b72fc | 299 | unsigned int cfq_latency; |
d9ff4187 | 300 | |
6118b70b JA |
301 | /* |
302 | * Fallback dummy cfqq for extreme OOM conditions | |
303 | */ | |
304 | struct cfq_queue oom_cfqq; | |
365722bb | 305 | |
573412b2 | 306 | unsigned long last_delayed_sync; |
1da177e4 LT |
307 | }; |
308 | ||
0381411e TH |
309 | static inline struct cfq_group *blkg_to_cfqg(struct blkio_group *blkg) |
310 | { | |
311 | return blkg_to_pdata(blkg, &blkio_policy_cfq); | |
312 | } | |
313 | ||
314 | static inline struct blkio_group *cfqg_to_blkg(struct cfq_group *cfqg) | |
315 | { | |
316 | return pdata_to_blkg(cfqg, &blkio_policy_cfq); | |
317 | } | |
318 | ||
25fb5169 VG |
319 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
320 | ||
cdb16e8f VG |
321 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
322 | enum wl_prio_t prio, | |
65b32a57 | 323 | enum wl_type_t type) |
c0324a02 | 324 | { |
1fa8f6d6 VG |
325 | if (!cfqg) |
326 | return NULL; | |
327 | ||
c0324a02 | 328 | if (prio == IDLE_WORKLOAD) |
cdb16e8f | 329 | return &cfqg->service_tree_idle; |
c0324a02 | 330 | |
cdb16e8f | 331 | return &cfqg->service_trees[prio][type]; |
c0324a02 CZ |
332 | } |
333 | ||
3b18152c | 334 | enum cfqq_state_flags { |
b0b8d749 JA |
335 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
336 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 337 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 338 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
339 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
340 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
341 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 342 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 343 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 344 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 345 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 346 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 347 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
348 | }; |
349 | ||
350 | #define CFQ_CFQQ_FNS(name) \ | |
351 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
352 | { \ | |
fe094d98 | 353 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
354 | } \ |
355 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
356 | { \ | |
fe094d98 | 357 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
358 | } \ |
359 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
360 | { \ | |
fe094d98 | 361 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
362 | } |
363 | ||
364 | CFQ_CFQQ_FNS(on_rr); | |
365 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 366 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 367 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
368 | CFQ_CFQQ_FNS(fifo_expire); |
369 | CFQ_CFQQ_FNS(idle_window); | |
370 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 371 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 372 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 373 | CFQ_CFQQ_FNS(coop); |
ae54abed | 374 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 375 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 376 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
377 | #undef CFQ_CFQQ_FNS |
378 | ||
afc24d49 | 379 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2868ef7b VG |
380 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
381 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ | |
382 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
0381411e | 383 | blkg_path(cfqg_to_blkg((cfqq)->cfqg)), ##args) |
2868ef7b VG |
384 | |
385 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ | |
386 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ | |
0381411e | 387 | blkg_path(cfqg_to_blkg((cfqg))), ##args) \ |
2868ef7b VG |
388 | |
389 | #else | |
7b679138 JA |
390 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
391 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
4495a7d4 | 392 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
2868ef7b | 393 | #endif |
7b679138 JA |
394 | #define cfq_log(cfqd, fmt, args...) \ |
395 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
396 | ||
615f0259 VG |
397 | /* Traverses through cfq group service trees */ |
398 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
399 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
400 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
401 | : &cfqg->service_tree_idle; \ | |
402 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
403 | (i == IDLE_WORKLOAD && j == 0); \ | |
404 | j++, st = i < IDLE_WORKLOAD ? \ | |
405 | &cfqg->service_trees[i][j]: NULL) \ | |
406 | ||
f5f2b6ce SL |
407 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
408 | struct cfq_ttime *ttime, bool group_idle) | |
409 | { | |
410 | unsigned long slice; | |
411 | if (!sample_valid(ttime->ttime_samples)) | |
412 | return false; | |
413 | if (group_idle) | |
414 | slice = cfqd->cfq_group_idle; | |
415 | else | |
416 | slice = cfqd->cfq_slice_idle; | |
417 | return ttime->ttime_mean > slice; | |
418 | } | |
615f0259 | 419 | |
02b35081 VG |
420 | static inline bool iops_mode(struct cfq_data *cfqd) |
421 | { | |
422 | /* | |
423 | * If we are not idling on queues and it is a NCQ drive, parallel | |
424 | * execution of requests is on and measuring time is not possible | |
425 | * in most of the cases until and unless we drive shallower queue | |
426 | * depths and that becomes a performance bottleneck. In such cases | |
427 | * switch to start providing fairness in terms of number of IOs. | |
428 | */ | |
429 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
430 | return true; | |
431 | else | |
432 | return false; | |
433 | } | |
434 | ||
c0324a02 CZ |
435 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
436 | { | |
437 | if (cfq_class_idle(cfqq)) | |
438 | return IDLE_WORKLOAD; | |
439 | if (cfq_class_rt(cfqq)) | |
440 | return RT_WORKLOAD; | |
441 | return BE_WORKLOAD; | |
442 | } | |
443 | ||
718eee05 CZ |
444 | |
445 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
446 | { | |
447 | if (!cfq_cfqq_sync(cfqq)) | |
448 | return ASYNC_WORKLOAD; | |
449 | if (!cfq_cfqq_idle_window(cfqq)) | |
450 | return SYNC_NOIDLE_WORKLOAD; | |
451 | return SYNC_WORKLOAD; | |
452 | } | |
453 | ||
58ff82f3 VG |
454 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
455 | struct cfq_data *cfqd, | |
456 | struct cfq_group *cfqg) | |
c0324a02 CZ |
457 | { |
458 | if (wl == IDLE_WORKLOAD) | |
cdb16e8f | 459 | return cfqg->service_tree_idle.count; |
c0324a02 | 460 | |
cdb16e8f VG |
461 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
462 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count | |
463 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
464 | } |
465 | ||
f26bd1f0 VG |
466 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
467 | struct cfq_group *cfqg) | |
468 | { | |
469 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count | |
470 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
471 | } | |
472 | ||
165125e1 | 473 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 474 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
abede6da | 475 | struct cfq_io_cq *cic, struct bio *bio, |
4f85cb96 | 476 | gfp_t gfp_mask); |
91fac317 | 477 | |
c5869807 TH |
478 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
479 | { | |
480 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
481 | return container_of(icq, struct cfq_io_cq, icq); | |
482 | } | |
483 | ||
47fdd4ca TH |
484 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
485 | struct io_context *ioc) | |
486 | { | |
487 | if (ioc) | |
488 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
489 | return NULL; | |
490 | } | |
491 | ||
c5869807 | 492 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 493 | { |
a6151c3a | 494 | return cic->cfqq[is_sync]; |
91fac317 VT |
495 | } |
496 | ||
c5869807 TH |
497 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
498 | bool is_sync) | |
91fac317 | 499 | { |
a6151c3a | 500 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
501 | } |
502 | ||
c5869807 | 503 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 504 | { |
c5869807 | 505 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
506 | } |
507 | ||
91fac317 VT |
508 | /* |
509 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
510 | * set (in which case it could also be direct WRITE). | |
511 | */ | |
a6151c3a | 512 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 513 | { |
7b6d91da | 514 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 515 | } |
1da177e4 | 516 | |
99f95e52 AM |
517 | /* |
518 | * scheduler run of queue, if there are requests pending and no one in the | |
519 | * driver that will restart queueing | |
520 | */ | |
23e018a1 | 521 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 522 | { |
7b679138 JA |
523 | if (cfqd->busy_queues) { |
524 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 525 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 526 | } |
99f95e52 AM |
527 | } |
528 | ||
44f7c160 JA |
529 | /* |
530 | * Scale schedule slice based on io priority. Use the sync time slice only | |
531 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
532 | * io only, should not get full sync slice length. | |
533 | */ | |
a6151c3a | 534 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 535 | unsigned short prio) |
44f7c160 | 536 | { |
d9e7620e | 537 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 538 | |
d9e7620e JA |
539 | WARN_ON(prio >= IOPRIO_BE_NR); |
540 | ||
541 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
542 | } | |
44f7c160 | 543 | |
d9e7620e JA |
544 | static inline int |
545 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
546 | { | |
547 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
548 | } |
549 | ||
25bc6b07 VG |
550 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
551 | { | |
552 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
553 | ||
554 | d = d * BLKIO_WEIGHT_DEFAULT; | |
555 | do_div(d, cfqg->weight); | |
556 | return d; | |
557 | } | |
558 | ||
559 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
560 | { | |
561 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
562 | if (delta > 0) | |
563 | min_vdisktime = vdisktime; | |
564 | ||
565 | return min_vdisktime; | |
566 | } | |
567 | ||
568 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
569 | { | |
570 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
571 | if (delta < 0) | |
572 | min_vdisktime = vdisktime; | |
573 | ||
574 | return min_vdisktime; | |
575 | } | |
576 | ||
577 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
578 | { | |
25bc6b07 VG |
579 | struct cfq_group *cfqg; |
580 | ||
25bc6b07 VG |
581 | if (st->left) { |
582 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
583 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
584 | cfqg->vdisktime); | |
25bc6b07 | 585 | } |
25bc6b07 VG |
586 | } |
587 | ||
5db5d642 CZ |
588 | /* |
589 | * get averaged number of queues of RT/BE priority. | |
590 | * average is updated, with a formula that gives more weight to higher numbers, | |
591 | * to quickly follows sudden increases and decrease slowly | |
592 | */ | |
593 | ||
58ff82f3 VG |
594 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
595 | struct cfq_group *cfqg, bool rt) | |
5869619c | 596 | { |
5db5d642 CZ |
597 | unsigned min_q, max_q; |
598 | unsigned mult = cfq_hist_divisor - 1; | |
599 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 600 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 601 | |
58ff82f3 VG |
602 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
603 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
604 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 605 | cfq_hist_divisor; |
58ff82f3 VG |
606 | return cfqg->busy_queues_avg[rt]; |
607 | } | |
608 | ||
609 | static inline unsigned | |
610 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
611 | { | |
612 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
613 | ||
614 | return cfq_target_latency * cfqg->weight / st->total_weight; | |
5db5d642 CZ |
615 | } |
616 | ||
c553f8e3 | 617 | static inline unsigned |
ba5bd520 | 618 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 619 | { |
5db5d642 CZ |
620 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
621 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
622 | /* |
623 | * interested queues (we consider only the ones with the same | |
624 | * priority class in the cfq group) | |
625 | */ | |
626 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
627 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
628 | unsigned sync_slice = cfqd->cfq_slice[1]; |
629 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
630 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
631 | ||
632 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
633 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
634 | /* scale low_slice according to IO priority | |
635 | * and sync vs async */ | |
636 | unsigned low_slice = | |
637 | min(slice, base_low_slice * slice / sync_slice); | |
638 | /* the adapted slice value is scaled to fit all iqs | |
639 | * into the target latency */ | |
58ff82f3 | 640 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
641 | low_slice); |
642 | } | |
643 | } | |
c553f8e3 SL |
644 | return slice; |
645 | } | |
646 | ||
647 | static inline void | |
648 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
649 | { | |
ba5bd520 | 650 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 651 | |
dae739eb | 652 | cfqq->slice_start = jiffies; |
5db5d642 | 653 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 654 | cfqq->allocated_slice = slice; |
7b679138 | 655 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
656 | } |
657 | ||
658 | /* | |
659 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
660 | * isn't valid until the first request from the dispatch is activated | |
661 | * and the slice time set. | |
662 | */ | |
a6151c3a | 663 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
664 | { |
665 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 666 | return false; |
44f7c160 | 667 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 668 | return false; |
44f7c160 | 669 | |
c1e44756 | 670 | return true; |
44f7c160 JA |
671 | } |
672 | ||
1da177e4 | 673 | /* |
5e705374 | 674 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 675 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 676 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 677 | */ |
5e705374 | 678 | static struct request * |
cf7c25cf | 679 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 680 | { |
cf7c25cf | 681 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 682 | unsigned long back_max; |
e8a99053 AM |
683 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
684 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
685 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 686 | |
5e705374 JA |
687 | if (rq1 == NULL || rq1 == rq2) |
688 | return rq2; | |
689 | if (rq2 == NULL) | |
690 | return rq1; | |
9c2c38a1 | 691 | |
229836bd NK |
692 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
693 | return rq_is_sync(rq1) ? rq1 : rq2; | |
694 | ||
65299a3b CH |
695 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
696 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 697 | |
83096ebf TH |
698 | s1 = blk_rq_pos(rq1); |
699 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 700 | |
1da177e4 LT |
701 | /* |
702 | * by definition, 1KiB is 2 sectors | |
703 | */ | |
704 | back_max = cfqd->cfq_back_max * 2; | |
705 | ||
706 | /* | |
707 | * Strict one way elevator _except_ in the case where we allow | |
708 | * short backward seeks which are biased as twice the cost of a | |
709 | * similar forward seek. | |
710 | */ | |
711 | if (s1 >= last) | |
712 | d1 = s1 - last; | |
713 | else if (s1 + back_max >= last) | |
714 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
715 | else | |
e8a99053 | 716 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
717 | |
718 | if (s2 >= last) | |
719 | d2 = s2 - last; | |
720 | else if (s2 + back_max >= last) | |
721 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
722 | else | |
e8a99053 | 723 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
724 | |
725 | /* Found required data */ | |
e8a99053 AM |
726 | |
727 | /* | |
728 | * By doing switch() on the bit mask "wrap" we avoid having to | |
729 | * check two variables for all permutations: --> faster! | |
730 | */ | |
731 | switch (wrap) { | |
5e705374 | 732 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 733 | if (d1 < d2) |
5e705374 | 734 | return rq1; |
e8a99053 | 735 | else if (d2 < d1) |
5e705374 | 736 | return rq2; |
e8a99053 AM |
737 | else { |
738 | if (s1 >= s2) | |
5e705374 | 739 | return rq1; |
e8a99053 | 740 | else |
5e705374 | 741 | return rq2; |
e8a99053 | 742 | } |
1da177e4 | 743 | |
e8a99053 | 744 | case CFQ_RQ2_WRAP: |
5e705374 | 745 | return rq1; |
e8a99053 | 746 | case CFQ_RQ1_WRAP: |
5e705374 JA |
747 | return rq2; |
748 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
749 | default: |
750 | /* | |
751 | * Since both rqs are wrapped, | |
752 | * start with the one that's further behind head | |
753 | * (--> only *one* back seek required), | |
754 | * since back seek takes more time than forward. | |
755 | */ | |
756 | if (s1 <= s2) | |
5e705374 | 757 | return rq1; |
1da177e4 | 758 | else |
5e705374 | 759 | return rq2; |
1da177e4 LT |
760 | } |
761 | } | |
762 | ||
498d3aa2 JA |
763 | /* |
764 | * The below is leftmost cache rbtree addon | |
765 | */ | |
0871714e | 766 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 767 | { |
615f0259 VG |
768 | /* Service tree is empty */ |
769 | if (!root->count) | |
770 | return NULL; | |
771 | ||
cc09e299 JA |
772 | if (!root->left) |
773 | root->left = rb_first(&root->rb); | |
774 | ||
0871714e JA |
775 | if (root->left) |
776 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
777 | ||
778 | return NULL; | |
cc09e299 JA |
779 | } |
780 | ||
1fa8f6d6 VG |
781 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
782 | { | |
783 | if (!root->left) | |
784 | root->left = rb_first(&root->rb); | |
785 | ||
786 | if (root->left) | |
787 | return rb_entry_cfqg(root->left); | |
788 | ||
789 | return NULL; | |
790 | } | |
791 | ||
a36e71f9 JA |
792 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
793 | { | |
794 | rb_erase(n, root); | |
795 | RB_CLEAR_NODE(n); | |
796 | } | |
797 | ||
cc09e299 JA |
798 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
799 | { | |
800 | if (root->left == n) | |
801 | root->left = NULL; | |
a36e71f9 | 802 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 803 | --root->count; |
cc09e299 JA |
804 | } |
805 | ||
1da177e4 LT |
806 | /* |
807 | * would be nice to take fifo expire time into account as well | |
808 | */ | |
5e705374 JA |
809 | static struct request * |
810 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
811 | struct request *last) | |
1da177e4 | 812 | { |
21183b07 JA |
813 | struct rb_node *rbnext = rb_next(&last->rb_node); |
814 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 815 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 816 | |
21183b07 | 817 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
818 | |
819 | if (rbprev) | |
5e705374 | 820 | prev = rb_entry_rq(rbprev); |
1da177e4 | 821 | |
21183b07 | 822 | if (rbnext) |
5e705374 | 823 | next = rb_entry_rq(rbnext); |
21183b07 JA |
824 | else { |
825 | rbnext = rb_first(&cfqq->sort_list); | |
826 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 827 | next = rb_entry_rq(rbnext); |
21183b07 | 828 | } |
1da177e4 | 829 | |
cf7c25cf | 830 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
831 | } |
832 | ||
d9e7620e JA |
833 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
834 | struct cfq_queue *cfqq) | |
1da177e4 | 835 | { |
d9e7620e JA |
836 | /* |
837 | * just an approximation, should be ok. | |
838 | */ | |
cdb16e8f | 839 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 840 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
841 | } |
842 | ||
1fa8f6d6 VG |
843 | static inline s64 |
844 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
845 | { | |
846 | return cfqg->vdisktime - st->min_vdisktime; | |
847 | } | |
848 | ||
849 | static void | |
850 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
851 | { | |
852 | struct rb_node **node = &st->rb.rb_node; | |
853 | struct rb_node *parent = NULL; | |
854 | struct cfq_group *__cfqg; | |
855 | s64 key = cfqg_key(st, cfqg); | |
856 | int left = 1; | |
857 | ||
858 | while (*node != NULL) { | |
859 | parent = *node; | |
860 | __cfqg = rb_entry_cfqg(parent); | |
861 | ||
862 | if (key < cfqg_key(st, __cfqg)) | |
863 | node = &parent->rb_left; | |
864 | else { | |
865 | node = &parent->rb_right; | |
866 | left = 0; | |
867 | } | |
868 | } | |
869 | ||
870 | if (left) | |
871 | st->left = &cfqg->rb_node; | |
872 | ||
873 | rb_link_node(&cfqg->rb_node, parent, node); | |
874 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
875 | } | |
876 | ||
877 | static void | |
8184f93e JT |
878 | cfq_update_group_weight(struct cfq_group *cfqg) |
879 | { | |
880 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
881 | if (cfqg->needs_update) { | |
882 | cfqg->weight = cfqg->new_weight; | |
883 | cfqg->needs_update = false; | |
884 | } | |
885 | } | |
886 | ||
887 | static void | |
888 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
889 | { | |
890 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
891 | ||
892 | cfq_update_group_weight(cfqg); | |
893 | __cfq_group_service_tree_add(st, cfqg); | |
894 | st->total_weight += cfqg->weight; | |
895 | } | |
896 | ||
897 | static void | |
898 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
899 | { |
900 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
901 | struct cfq_group *__cfqg; | |
902 | struct rb_node *n; | |
903 | ||
904 | cfqg->nr_cfqq++; | |
760701bf | 905 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
906 | return; |
907 | ||
908 | /* | |
909 | * Currently put the group at the end. Later implement something | |
910 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 911 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
912 | */ |
913 | n = rb_last(&st->rb); | |
914 | if (n) { | |
915 | __cfqg = rb_entry_cfqg(n); | |
916 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
917 | } else | |
918 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
919 | cfq_group_service_tree_add(st, cfqg); |
920 | } | |
1fa8f6d6 | 921 | |
8184f93e JT |
922 | static void |
923 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
924 | { | |
925 | st->total_weight -= cfqg->weight; | |
926 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) | |
927 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
928 | } |
929 | ||
930 | static void | |
8184f93e | 931 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
932 | { |
933 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
934 | ||
935 | BUG_ON(cfqg->nr_cfqq < 1); | |
936 | cfqg->nr_cfqq--; | |
25bc6b07 | 937 | |
1fa8f6d6 VG |
938 | /* If there are other cfq queues under this group, don't delete it */ |
939 | if (cfqg->nr_cfqq) | |
940 | return; | |
941 | ||
2868ef7b | 942 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 943 | cfq_group_service_tree_del(st, cfqg); |
dae739eb | 944 | cfqg->saved_workload_slice = 0; |
c1768268 TH |
945 | cfq_blkiocg_update_dequeue_stats(cfqg_to_blkg(cfqg), |
946 | &blkio_policy_cfq, 1); | |
dae739eb VG |
947 | } |
948 | ||
167400d3 JT |
949 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
950 | unsigned int *unaccounted_time) | |
dae739eb | 951 | { |
f75edf2d | 952 | unsigned int slice_used; |
dae739eb VG |
953 | |
954 | /* | |
955 | * Queue got expired before even a single request completed or | |
956 | * got expired immediately after first request completion. | |
957 | */ | |
958 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
959 | /* | |
960 | * Also charge the seek time incurred to the group, otherwise | |
961 | * if there are mutiple queues in the group, each can dispatch | |
962 | * a single request on seeky media and cause lots of seek time | |
963 | * and group will never know it. | |
964 | */ | |
965 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
966 | 1); | |
967 | } else { | |
968 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
969 | if (slice_used > cfqq->allocated_slice) { |
970 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 971 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
972 | } |
973 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
974 | *unaccounted_time += cfqq->slice_start - | |
975 | cfqq->dispatch_start; | |
dae739eb VG |
976 | } |
977 | ||
dae739eb VG |
978 | return slice_used; |
979 | } | |
980 | ||
981 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 982 | struct cfq_queue *cfqq) |
dae739eb VG |
983 | { |
984 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 985 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
986 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
987 | - cfqg->service_tree_idle.count; | |
988 | ||
989 | BUG_ON(nr_sync < 0); | |
167400d3 | 990 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 991 | |
02b35081 VG |
992 | if (iops_mode(cfqd)) |
993 | charge = cfqq->slice_dispatch; | |
994 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
995 | charge = cfqq->allocated_slice; | |
dae739eb VG |
996 | |
997 | /* Can't update vdisktime while group is on service tree */ | |
8184f93e | 998 | cfq_group_service_tree_del(st, cfqg); |
02b35081 | 999 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
8184f93e JT |
1000 | /* If a new weight was requested, update now, off tree */ |
1001 | cfq_group_service_tree_add(st, cfqg); | |
dae739eb VG |
1002 | |
1003 | /* This group is being expired. Save the context */ | |
1004 | if (time_after(cfqd->workload_expires, jiffies)) { | |
1005 | cfqg->saved_workload_slice = cfqd->workload_expires | |
1006 | - jiffies; | |
1007 | cfqg->saved_workload = cfqd->serving_type; | |
1008 | cfqg->saved_serving_prio = cfqd->serving_prio; | |
1009 | } else | |
1010 | cfqg->saved_workload_slice = 0; | |
2868ef7b VG |
1011 | |
1012 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1013 | st->min_vdisktime); | |
fd16d263 JP |
1014 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1015 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1016 | used_sl, cfqq->slice_dispatch, charge, | |
1017 | iops_mode(cfqd), cfqq->nr_sectors); | |
c1768268 TH |
1018 | cfq_blkiocg_update_timeslice_used(cfqg_to_blkg(cfqg), &blkio_policy_cfq, |
1019 | used_sl, unaccounted_sl); | |
1020 | cfq_blkiocg_set_start_empty_time(cfqg_to_blkg(cfqg), &blkio_policy_cfq); | |
1fa8f6d6 VG |
1021 | } |
1022 | ||
f51b802c TH |
1023 | /** |
1024 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1025 | * @cfqg: cfq_group to initialize | |
1026 | * | |
1027 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1028 | * is enabled or not. | |
1029 | */ | |
1030 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1031 | { | |
1032 | struct cfq_rb_root *st; | |
1033 | int i, j; | |
1034 | ||
1035 | for_each_cfqg_st(cfqg, i, j, st) | |
1036 | *st = CFQ_RB_ROOT; | |
1037 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1038 | ||
1039 | cfqg->ttime.last_end_request = jiffies; | |
1040 | } | |
1041 | ||
25fb5169 | 1042 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
ca32aefc TH |
1043 | static void cfq_update_blkio_group_weight(struct request_queue *q, |
1044 | struct blkio_group *blkg, | |
8aea4545 | 1045 | unsigned int weight) |
f8d461d6 | 1046 | { |
0381411e TH |
1047 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
1048 | ||
8184f93e JT |
1049 | cfqg->new_weight = weight; |
1050 | cfqg->needs_update = true; | |
f8d461d6 VG |
1051 | } |
1052 | ||
0381411e | 1053 | static void cfq_init_blkio_group(struct blkio_group *blkg) |
f469a7b4 | 1054 | { |
0381411e | 1055 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
25fb5169 | 1056 | |
f51b802c | 1057 | cfq_init_cfqg_base(cfqg); |
0381411e | 1058 | cfqg->weight = blkg->blkcg->weight; |
25fb5169 VG |
1059 | } |
1060 | ||
1061 | /* | |
3e59cf9d VG |
1062 | * Search for the cfq group current task belongs to. request_queue lock must |
1063 | * be held. | |
25fb5169 | 1064 | */ |
cd1604fa TH |
1065 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
1066 | struct blkio_cgroup *blkcg) | |
25fb5169 | 1067 | { |
f469a7b4 | 1068 | struct request_queue *q = cfqd->queue; |
cd1604fa | 1069 | struct cfq_group *cfqg = NULL; |
25fb5169 | 1070 | |
cd1604fa TH |
1071 | /* avoid lookup for the common case where there's no blkio cgroup */ |
1072 | if (blkcg == &blkio_root_cgroup) { | |
1073 | cfqg = cfqd->root_group; | |
1074 | } else { | |
1075 | struct blkio_group *blkg; | |
f469a7b4 | 1076 | |
cd1604fa TH |
1077 | blkg = blkg_lookup_create(blkcg, q, BLKIO_POLICY_PROP, false); |
1078 | if (!IS_ERR(blkg)) | |
0381411e | 1079 | cfqg = blkg_to_cfqg(blkg); |
cd1604fa | 1080 | } |
f469a7b4 | 1081 | |
25fb5169 VG |
1082 | return cfqg; |
1083 | } | |
1084 | ||
1085 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1086 | { | |
1087 | /* Currently, all async queues are mapped to root group */ | |
1088 | if (!cfq_cfqq_sync(cfqq)) | |
f51b802c | 1089 | cfqg = cfqq->cfqd->root_group; |
25fb5169 VG |
1090 | |
1091 | cfqq->cfqg = cfqg; | |
b1c35769 | 1092 | /* cfqq reference on cfqg */ |
1adaf3dd | 1093 | blkg_get(cfqg_to_blkg(cfqg)); |
b1c35769 VG |
1094 | } |
1095 | ||
25fb5169 | 1096 | #else /* GROUP_IOSCHED */ |
cd1604fa TH |
1097 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
1098 | struct blkio_cgroup *blkcg) | |
25fb5169 | 1099 | { |
f51b802c | 1100 | return cfqd->root_group; |
25fb5169 | 1101 | } |
7f1dc8a2 | 1102 | |
25fb5169 VG |
1103 | static inline void |
1104 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1105 | cfqq->cfqg = cfqg; | |
1106 | } | |
1107 | ||
1108 | #endif /* GROUP_IOSCHED */ | |
1109 | ||
498d3aa2 | 1110 | /* |
c0324a02 | 1111 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1112 | * requests waiting to be processed. It is sorted in the order that |
1113 | * we will service the queues. | |
1114 | */ | |
a36e71f9 | 1115 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1116 | bool add_front) |
d9e7620e | 1117 | { |
0871714e JA |
1118 | struct rb_node **p, *parent; |
1119 | struct cfq_queue *__cfqq; | |
d9e7620e | 1120 | unsigned long rb_key; |
c0324a02 | 1121 | struct cfq_rb_root *service_tree; |
498d3aa2 | 1122 | int left; |
dae739eb | 1123 | int new_cfqq = 1; |
ae30c286 | 1124 | |
cdb16e8f | 1125 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
65b32a57 | 1126 | cfqq_type(cfqq)); |
0871714e JA |
1127 | if (cfq_class_idle(cfqq)) { |
1128 | rb_key = CFQ_IDLE_DELAY; | |
aa6f6a3d | 1129 | parent = rb_last(&service_tree->rb); |
0871714e JA |
1130 | if (parent && parent != &cfqq->rb_node) { |
1131 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1132 | rb_key += __cfqq->rb_key; | |
1133 | } else | |
1134 | rb_key += jiffies; | |
1135 | } else if (!add_front) { | |
b9c8946b JA |
1136 | /* |
1137 | * Get our rb key offset. Subtract any residual slice | |
1138 | * value carried from last service. A negative resid | |
1139 | * count indicates slice overrun, and this should position | |
1140 | * the next service time further away in the tree. | |
1141 | */ | |
edd75ffd | 1142 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1143 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1144 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1145 | } else { |
1146 | rb_key = -HZ; | |
aa6f6a3d | 1147 | __cfqq = cfq_rb_first(service_tree); |
48e025e6 CZ |
1148 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1149 | } | |
1da177e4 | 1150 | |
d9e7620e | 1151 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1152 | new_cfqq = 0; |
99f9628a | 1153 | /* |
d9e7620e | 1154 | * same position, nothing more to do |
99f9628a | 1155 | */ |
c0324a02 CZ |
1156 | if (rb_key == cfqq->rb_key && |
1157 | cfqq->service_tree == service_tree) | |
d9e7620e | 1158 | return; |
1da177e4 | 1159 | |
aa6f6a3d CZ |
1160 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1161 | cfqq->service_tree = NULL; | |
1da177e4 | 1162 | } |
d9e7620e | 1163 | |
498d3aa2 | 1164 | left = 1; |
0871714e | 1165 | parent = NULL; |
aa6f6a3d CZ |
1166 | cfqq->service_tree = service_tree; |
1167 | p = &service_tree->rb.rb_node; | |
d9e7620e | 1168 | while (*p) { |
67060e37 | 1169 | struct rb_node **n; |
cc09e299 | 1170 | |
d9e7620e JA |
1171 | parent = *p; |
1172 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1173 | ||
0c534e0a | 1174 | /* |
c0324a02 | 1175 | * sort by key, that represents service time. |
0c534e0a | 1176 | */ |
c0324a02 | 1177 | if (time_before(rb_key, __cfqq->rb_key)) |
67060e37 | 1178 | n = &(*p)->rb_left; |
c0324a02 | 1179 | else { |
67060e37 | 1180 | n = &(*p)->rb_right; |
cc09e299 | 1181 | left = 0; |
c0324a02 | 1182 | } |
67060e37 JA |
1183 | |
1184 | p = n; | |
d9e7620e JA |
1185 | } |
1186 | ||
cc09e299 | 1187 | if (left) |
aa6f6a3d | 1188 | service_tree->left = &cfqq->rb_node; |
cc09e299 | 1189 | |
d9e7620e JA |
1190 | cfqq->rb_key = rb_key; |
1191 | rb_link_node(&cfqq->rb_node, parent, p); | |
aa6f6a3d CZ |
1192 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
1193 | service_tree->count++; | |
20359f27 | 1194 | if (add_front || !new_cfqq) |
dae739eb | 1195 | return; |
8184f93e | 1196 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1197 | } |
1198 | ||
a36e71f9 | 1199 | static struct cfq_queue * |
f2d1f0ae JA |
1200 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1201 | sector_t sector, struct rb_node **ret_parent, | |
1202 | struct rb_node ***rb_link) | |
a36e71f9 | 1203 | { |
a36e71f9 JA |
1204 | struct rb_node **p, *parent; |
1205 | struct cfq_queue *cfqq = NULL; | |
1206 | ||
1207 | parent = NULL; | |
1208 | p = &root->rb_node; | |
1209 | while (*p) { | |
1210 | struct rb_node **n; | |
1211 | ||
1212 | parent = *p; | |
1213 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1214 | ||
1215 | /* | |
1216 | * Sort strictly based on sector. Smallest to the left, | |
1217 | * largest to the right. | |
1218 | */ | |
2e46e8b2 | 1219 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1220 | n = &(*p)->rb_right; |
2e46e8b2 | 1221 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1222 | n = &(*p)->rb_left; |
1223 | else | |
1224 | break; | |
1225 | p = n; | |
3ac6c9f8 | 1226 | cfqq = NULL; |
a36e71f9 JA |
1227 | } |
1228 | ||
1229 | *ret_parent = parent; | |
1230 | if (rb_link) | |
1231 | *rb_link = p; | |
3ac6c9f8 | 1232 | return cfqq; |
a36e71f9 JA |
1233 | } |
1234 | ||
1235 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1236 | { | |
a36e71f9 JA |
1237 | struct rb_node **p, *parent; |
1238 | struct cfq_queue *__cfqq; | |
1239 | ||
f2d1f0ae JA |
1240 | if (cfqq->p_root) { |
1241 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1242 | cfqq->p_root = NULL; | |
1243 | } | |
a36e71f9 JA |
1244 | |
1245 | if (cfq_class_idle(cfqq)) | |
1246 | return; | |
1247 | if (!cfqq->next_rq) | |
1248 | return; | |
1249 | ||
f2d1f0ae | 1250 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1251 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1252 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1253 | if (!__cfqq) { |
1254 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1255 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1256 | } else | |
1257 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1258 | } |
1259 | ||
498d3aa2 JA |
1260 | /* |
1261 | * Update cfqq's position in the service tree. | |
1262 | */ | |
edd75ffd | 1263 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1264 | { |
6d048f53 JA |
1265 | /* |
1266 | * Resorting requires the cfqq to be on the RR list already. | |
1267 | */ | |
a36e71f9 | 1268 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1269 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1270 | cfq_prio_tree_add(cfqd, cfqq); |
1271 | } | |
6d048f53 JA |
1272 | } |
1273 | ||
1da177e4 LT |
1274 | /* |
1275 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1276 | * the pending list according to last request service |
1da177e4 | 1277 | */ |
febffd61 | 1278 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1279 | { |
7b679138 | 1280 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1281 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1282 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 1283 | cfqd->busy_queues++; |
ef8a41df SL |
1284 | if (cfq_cfqq_sync(cfqq)) |
1285 | cfqd->busy_sync_queues++; | |
1da177e4 | 1286 | |
edd75ffd | 1287 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1288 | } |
1289 | ||
498d3aa2 JA |
1290 | /* |
1291 | * Called when the cfqq no longer has requests pending, remove it from | |
1292 | * the service tree. | |
1293 | */ | |
febffd61 | 1294 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1295 | { |
7b679138 | 1296 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1297 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1298 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1299 | |
aa6f6a3d CZ |
1300 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1301 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1302 | cfqq->service_tree = NULL; | |
1303 | } | |
f2d1f0ae JA |
1304 | if (cfqq->p_root) { |
1305 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1306 | cfqq->p_root = NULL; | |
1307 | } | |
d9e7620e | 1308 | |
8184f93e | 1309 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1310 | BUG_ON(!cfqd->busy_queues); |
1311 | cfqd->busy_queues--; | |
ef8a41df SL |
1312 | if (cfq_cfqq_sync(cfqq)) |
1313 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
1314 | } |
1315 | ||
1316 | /* | |
1317 | * rb tree support functions | |
1318 | */ | |
febffd61 | 1319 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1320 | { |
5e705374 | 1321 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1322 | const int sync = rq_is_sync(rq); |
1da177e4 | 1323 | |
b4878f24 JA |
1324 | BUG_ON(!cfqq->queued[sync]); |
1325 | cfqq->queued[sync]--; | |
1da177e4 | 1326 | |
5e705374 | 1327 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1328 | |
f04a6424 VG |
1329 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1330 | /* | |
1331 | * Queue will be deleted from service tree when we actually | |
1332 | * expire it later. Right now just remove it from prio tree | |
1333 | * as it is empty. | |
1334 | */ | |
1335 | if (cfqq->p_root) { | |
1336 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1337 | cfqq->p_root = NULL; | |
1338 | } | |
1339 | } | |
1da177e4 LT |
1340 | } |
1341 | ||
5e705374 | 1342 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1343 | { |
5e705374 | 1344 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1345 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 1346 | struct request *prev; |
1da177e4 | 1347 | |
5380a101 | 1348 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 1349 | |
796d5116 | 1350 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
1351 | |
1352 | if (!cfq_cfqq_on_rr(cfqq)) | |
1353 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1354 | |
1355 | /* | |
1356 | * check if this request is a better next-serve candidate | |
1357 | */ | |
a36e71f9 | 1358 | prev = cfqq->next_rq; |
cf7c25cf | 1359 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1360 | |
1361 | /* | |
1362 | * adjust priority tree position, if ->next_rq changes | |
1363 | */ | |
1364 | if (prev != cfqq->next_rq) | |
1365 | cfq_prio_tree_add(cfqd, cfqq); | |
1366 | ||
5044eed4 | 1367 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1368 | } |
1369 | ||
febffd61 | 1370 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1371 | { |
5380a101 JA |
1372 | elv_rb_del(&cfqq->sort_list, rq); |
1373 | cfqq->queued[rq_is_sync(rq)]--; | |
0381411e | 1374 | cfq_blkiocg_update_io_remove_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 TH |
1375 | &blkio_policy_cfq, rq_data_dir(rq), |
1376 | rq_is_sync(rq)); | |
5e705374 | 1377 | cfq_add_rq_rb(rq); |
0381411e | 1378 | cfq_blkiocg_update_io_add_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 | 1379 | &blkio_policy_cfq, |
0381411e TH |
1380 | cfqg_to_blkg(cfqq->cfqd->serving_group), |
1381 | rq_data_dir(rq), rq_is_sync(rq)); | |
1da177e4 LT |
1382 | } |
1383 | ||
206dc69b JA |
1384 | static struct request * |
1385 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1386 | { |
206dc69b | 1387 | struct task_struct *tsk = current; |
c5869807 | 1388 | struct cfq_io_cq *cic; |
206dc69b | 1389 | struct cfq_queue *cfqq; |
1da177e4 | 1390 | |
4ac845a2 | 1391 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1392 | if (!cic) |
1393 | return NULL; | |
1394 | ||
1395 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
1396 | if (cfqq) { |
1397 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
1398 | ||
21183b07 | 1399 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 1400 | } |
1da177e4 | 1401 | |
1da177e4 LT |
1402 | return NULL; |
1403 | } | |
1404 | ||
165125e1 | 1405 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1406 | { |
22e2c507 | 1407 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 1408 | |
53c583d2 | 1409 | cfqd->rq_in_driver++; |
7b679138 | 1410 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 1411 | cfqd->rq_in_driver); |
25776e35 | 1412 | |
5b93629b | 1413 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
1414 | } |
1415 | ||
165125e1 | 1416 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1417 | { |
b4878f24 JA |
1418 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1419 | ||
53c583d2 CZ |
1420 | WARN_ON(!cfqd->rq_in_driver); |
1421 | cfqd->rq_in_driver--; | |
7b679138 | 1422 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 1423 | cfqd->rq_in_driver); |
1da177e4 LT |
1424 | } |
1425 | ||
b4878f24 | 1426 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 1427 | { |
5e705374 | 1428 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 1429 | |
5e705374 JA |
1430 | if (cfqq->next_rq == rq) |
1431 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 1432 | |
b4878f24 | 1433 | list_del_init(&rq->queuelist); |
5e705374 | 1434 | cfq_del_rq_rb(rq); |
374f84ac | 1435 | |
45333d5a | 1436 | cfqq->cfqd->rq_queued--; |
0381411e | 1437 | cfq_blkiocg_update_io_remove_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 TH |
1438 | &blkio_policy_cfq, rq_data_dir(rq), |
1439 | rq_is_sync(rq)); | |
65299a3b CH |
1440 | if (rq->cmd_flags & REQ_PRIO) { |
1441 | WARN_ON(!cfqq->prio_pending); | |
1442 | cfqq->prio_pending--; | |
b53d1ed7 | 1443 | } |
1da177e4 LT |
1444 | } |
1445 | ||
165125e1 JA |
1446 | static int cfq_merge(struct request_queue *q, struct request **req, |
1447 | struct bio *bio) | |
1da177e4 LT |
1448 | { |
1449 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1450 | struct request *__rq; | |
1da177e4 | 1451 | |
206dc69b | 1452 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 1453 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
1454 | *req = __rq; |
1455 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
1456 | } |
1457 | ||
1458 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
1459 | } |
1460 | ||
165125e1 | 1461 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 1462 | int type) |
1da177e4 | 1463 | { |
21183b07 | 1464 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 1465 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 1466 | |
5e705374 | 1467 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 1468 | } |
1da177e4 LT |
1469 | } |
1470 | ||
812d4026 DS |
1471 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
1472 | struct bio *bio) | |
1473 | { | |
0381411e | 1474 | cfq_blkiocg_update_io_merged_stats(cfqg_to_blkg(RQ_CFQG(req)), |
c1768268 TH |
1475 | &blkio_policy_cfq, bio_data_dir(bio), |
1476 | cfq_bio_sync(bio)); | |
812d4026 DS |
1477 | } |
1478 | ||
1da177e4 | 1479 | static void |
165125e1 | 1480 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
1481 | struct request *next) |
1482 | { | |
cf7c25cf | 1483 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
1484 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1485 | ||
22e2c507 JA |
1486 | /* |
1487 | * reposition in fifo if next is older than rq | |
1488 | */ | |
1489 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
30996f40 | 1490 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
22e2c507 | 1491 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
1492 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1493 | } | |
22e2c507 | 1494 | |
cf7c25cf CZ |
1495 | if (cfqq->next_rq == next) |
1496 | cfqq->next_rq = rq; | |
b4878f24 | 1497 | cfq_remove_request(next); |
0381411e | 1498 | cfq_blkiocg_update_io_merged_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 TH |
1499 | &blkio_policy_cfq, rq_data_dir(next), |
1500 | rq_is_sync(next)); | |
4a0b75c7 SL |
1501 | |
1502 | cfqq = RQ_CFQQ(next); | |
1503 | /* | |
1504 | * all requests of this queue are merged to other queues, delete it | |
1505 | * from the service tree. If it's the active_queue, | |
1506 | * cfq_dispatch_requests() will choose to expire it or do idle | |
1507 | */ | |
1508 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
1509 | cfqq != cfqd->active_queue) | |
1510 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
1511 | } |
1512 | ||
165125e1 | 1513 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
1514 | struct bio *bio) |
1515 | { | |
1516 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 1517 | struct cfq_io_cq *cic; |
da775265 | 1518 | struct cfq_queue *cfqq; |
da775265 JA |
1519 | |
1520 | /* | |
ec8acb69 | 1521 | * Disallow merge of a sync bio into an async request. |
da775265 | 1522 | */ |
91fac317 | 1523 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 1524 | return false; |
da775265 JA |
1525 | |
1526 | /* | |
f1a4f4d3 | 1527 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 1528 | * merge only if rq is queued there. |
f1a4f4d3 | 1529 | */ |
07c2bd37 TH |
1530 | cic = cfq_cic_lookup(cfqd, current->io_context); |
1531 | if (!cic) | |
1532 | return false; | |
719d3402 | 1533 | |
91fac317 | 1534 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 1535 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
1536 | } |
1537 | ||
812df48d DS |
1538 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1539 | { | |
1540 | del_timer(&cfqd->idle_slice_timer); | |
c1768268 TH |
1541 | cfq_blkiocg_update_idle_time_stats(cfqg_to_blkg(cfqq->cfqg), |
1542 | &blkio_policy_cfq); | |
812df48d DS |
1543 | } |
1544 | ||
febffd61 JA |
1545 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
1546 | struct cfq_queue *cfqq) | |
22e2c507 JA |
1547 | { |
1548 | if (cfqq) { | |
b1ffe737 DS |
1549 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", |
1550 | cfqd->serving_prio, cfqd->serving_type); | |
c1768268 TH |
1551 | cfq_blkiocg_update_avg_queue_size_stats(cfqg_to_blkg(cfqq->cfqg), |
1552 | &blkio_policy_cfq); | |
62a37f6b JT |
1553 | cfqq->slice_start = 0; |
1554 | cfqq->dispatch_start = jiffies; | |
1555 | cfqq->allocated_slice = 0; | |
1556 | cfqq->slice_end = 0; | |
1557 | cfqq->slice_dispatch = 0; | |
1558 | cfqq->nr_sectors = 0; | |
1559 | ||
1560 | cfq_clear_cfqq_wait_request(cfqq); | |
1561 | cfq_clear_cfqq_must_dispatch(cfqq); | |
1562 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
1563 | cfq_clear_cfqq_fifo_expire(cfqq); | |
1564 | cfq_mark_cfqq_slice_new(cfqq); | |
1565 | ||
1566 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
1567 | } |
1568 | ||
1569 | cfqd->active_queue = cfqq; | |
1570 | } | |
1571 | ||
7b14e3b5 JA |
1572 | /* |
1573 | * current cfqq expired its slice (or was too idle), select new one | |
1574 | */ | |
1575 | static void | |
1576 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 1577 | bool timed_out) |
7b14e3b5 | 1578 | { |
7b679138 JA |
1579 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
1580 | ||
7b14e3b5 | 1581 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 1582 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 1583 | |
7b14e3b5 | 1584 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 1585 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 1586 | |
ae54abed SL |
1587 | /* |
1588 | * If this cfqq is shared between multiple processes, check to | |
1589 | * make sure that those processes are still issuing I/Os within | |
1590 | * the mean seek distance. If not, it may be time to break the | |
1591 | * queues apart again. | |
1592 | */ | |
1593 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
1594 | cfq_mark_cfqq_split_coop(cfqq); | |
1595 | ||
7b14e3b5 | 1596 | /* |
6084cdda | 1597 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 1598 | */ |
c553f8e3 SL |
1599 | if (timed_out) { |
1600 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 1601 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
1602 | else |
1603 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
1604 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
1605 | } | |
7b14e3b5 | 1606 | |
e5ff082e | 1607 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 1608 | |
f04a6424 VG |
1609 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1610 | cfq_del_cfqq_rr(cfqd, cfqq); | |
1611 | ||
edd75ffd | 1612 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
1613 | |
1614 | if (cfqq == cfqd->active_queue) | |
1615 | cfqd->active_queue = NULL; | |
1616 | ||
1617 | if (cfqd->active_cic) { | |
11a3122f | 1618 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
1619 | cfqd->active_cic = NULL; |
1620 | } | |
7b14e3b5 JA |
1621 | } |
1622 | ||
e5ff082e | 1623 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
1624 | { |
1625 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1626 | ||
1627 | if (cfqq) | |
e5ff082e | 1628 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
1629 | } |
1630 | ||
498d3aa2 JA |
1631 | /* |
1632 | * Get next queue for service. Unless we have a queue preemption, | |
1633 | * we'll simply select the first cfqq in the service tree. | |
1634 | */ | |
6d048f53 | 1635 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 1636 | { |
c0324a02 | 1637 | struct cfq_rb_root *service_tree = |
cdb16e8f | 1638 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
65b32a57 | 1639 | cfqd->serving_type); |
d9e7620e | 1640 | |
f04a6424 VG |
1641 | if (!cfqd->rq_queued) |
1642 | return NULL; | |
1643 | ||
1fa8f6d6 VG |
1644 | /* There is nothing to dispatch */ |
1645 | if (!service_tree) | |
1646 | return NULL; | |
c0324a02 CZ |
1647 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
1648 | return NULL; | |
1649 | return cfq_rb_first(service_tree); | |
6d048f53 JA |
1650 | } |
1651 | ||
f04a6424 VG |
1652 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
1653 | { | |
25fb5169 | 1654 | struct cfq_group *cfqg; |
f04a6424 VG |
1655 | struct cfq_queue *cfqq; |
1656 | int i, j; | |
1657 | struct cfq_rb_root *st; | |
1658 | ||
1659 | if (!cfqd->rq_queued) | |
1660 | return NULL; | |
1661 | ||
25fb5169 VG |
1662 | cfqg = cfq_get_next_cfqg(cfqd); |
1663 | if (!cfqg) | |
1664 | return NULL; | |
1665 | ||
f04a6424 VG |
1666 | for_each_cfqg_st(cfqg, i, j, st) |
1667 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
1668 | return cfqq; | |
1669 | return NULL; | |
1670 | } | |
1671 | ||
498d3aa2 JA |
1672 | /* |
1673 | * Get and set a new active queue for service. | |
1674 | */ | |
a36e71f9 JA |
1675 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
1676 | struct cfq_queue *cfqq) | |
6d048f53 | 1677 | { |
e00ef799 | 1678 | if (!cfqq) |
a36e71f9 | 1679 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 1680 | |
22e2c507 | 1681 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 1682 | return cfqq; |
22e2c507 JA |
1683 | } |
1684 | ||
d9e7620e JA |
1685 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
1686 | struct request *rq) | |
1687 | { | |
83096ebf TH |
1688 | if (blk_rq_pos(rq) >= cfqd->last_position) |
1689 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 1690 | else |
83096ebf | 1691 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
1692 | } |
1693 | ||
b2c18e1e | 1694 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 1695 | struct request *rq) |
6d048f53 | 1696 | { |
e9ce335d | 1697 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
1698 | } |
1699 | ||
a36e71f9 JA |
1700 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
1701 | struct cfq_queue *cur_cfqq) | |
1702 | { | |
f2d1f0ae | 1703 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
1704 | struct rb_node *parent, *node; |
1705 | struct cfq_queue *__cfqq; | |
1706 | sector_t sector = cfqd->last_position; | |
1707 | ||
1708 | if (RB_EMPTY_ROOT(root)) | |
1709 | return NULL; | |
1710 | ||
1711 | /* | |
1712 | * First, if we find a request starting at the end of the last | |
1713 | * request, choose it. | |
1714 | */ | |
f2d1f0ae | 1715 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
1716 | if (__cfqq) |
1717 | return __cfqq; | |
1718 | ||
1719 | /* | |
1720 | * If the exact sector wasn't found, the parent of the NULL leaf | |
1721 | * will contain the closest sector. | |
1722 | */ | |
1723 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 1724 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1725 | return __cfqq; |
1726 | ||
2e46e8b2 | 1727 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
1728 | node = rb_next(&__cfqq->p_node); |
1729 | else | |
1730 | node = rb_prev(&__cfqq->p_node); | |
1731 | if (!node) | |
1732 | return NULL; | |
1733 | ||
1734 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 1735 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1736 | return __cfqq; |
1737 | ||
1738 | return NULL; | |
1739 | } | |
1740 | ||
1741 | /* | |
1742 | * cfqd - obvious | |
1743 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
1744 | * closely cooperating with itself. | |
1745 | * | |
1746 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
1747 | * one request, and that cfqd->last_position reflects a position on the disk | |
1748 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
1749 | * assumption. | |
1750 | */ | |
1751 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 1752 | struct cfq_queue *cur_cfqq) |
6d048f53 | 1753 | { |
a36e71f9 JA |
1754 | struct cfq_queue *cfqq; |
1755 | ||
39c01b21 DS |
1756 | if (cfq_class_idle(cur_cfqq)) |
1757 | return NULL; | |
e6c5bc73 JM |
1758 | if (!cfq_cfqq_sync(cur_cfqq)) |
1759 | return NULL; | |
1760 | if (CFQQ_SEEKY(cur_cfqq)) | |
1761 | return NULL; | |
1762 | ||
b9d8f4c7 GJ |
1763 | /* |
1764 | * Don't search priority tree if it's the only queue in the group. | |
1765 | */ | |
1766 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
1767 | return NULL; | |
1768 | ||
6d048f53 | 1769 | /* |
d9e7620e JA |
1770 | * We should notice if some of the queues are cooperating, eg |
1771 | * working closely on the same area of the disk. In that case, | |
1772 | * we can group them together and don't waste time idling. | |
6d048f53 | 1773 | */ |
a36e71f9 JA |
1774 | cfqq = cfqq_close(cfqd, cur_cfqq); |
1775 | if (!cfqq) | |
1776 | return NULL; | |
1777 | ||
8682e1f1 VG |
1778 | /* If new queue belongs to different cfq_group, don't choose it */ |
1779 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
1780 | return NULL; | |
1781 | ||
df5fe3e8 JM |
1782 | /* |
1783 | * It only makes sense to merge sync queues. | |
1784 | */ | |
1785 | if (!cfq_cfqq_sync(cfqq)) | |
1786 | return NULL; | |
e6c5bc73 JM |
1787 | if (CFQQ_SEEKY(cfqq)) |
1788 | return NULL; | |
df5fe3e8 | 1789 | |
c0324a02 CZ |
1790 | /* |
1791 | * Do not merge queues of different priority classes | |
1792 | */ | |
1793 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
1794 | return NULL; | |
1795 | ||
a36e71f9 | 1796 | return cfqq; |
6d048f53 JA |
1797 | } |
1798 | ||
a6d44e98 CZ |
1799 | /* |
1800 | * Determine whether we should enforce idle window for this queue. | |
1801 | */ | |
1802 | ||
1803 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1804 | { | |
1805 | enum wl_prio_t prio = cfqq_prio(cfqq); | |
718eee05 | 1806 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
a6d44e98 | 1807 | |
f04a6424 VG |
1808 | BUG_ON(!service_tree); |
1809 | BUG_ON(!service_tree->count); | |
1810 | ||
b6508c16 VG |
1811 | if (!cfqd->cfq_slice_idle) |
1812 | return false; | |
1813 | ||
a6d44e98 CZ |
1814 | /* We never do for idle class queues. */ |
1815 | if (prio == IDLE_WORKLOAD) | |
1816 | return false; | |
1817 | ||
1818 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
1819 | if (cfq_cfqq_idle_window(cfqq) && |
1820 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
1821 | return true; |
1822 | ||
1823 | /* | |
1824 | * Otherwise, we do only if they are the last ones | |
1825 | * in their service tree. | |
1826 | */ | |
f5f2b6ce SL |
1827 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq) && |
1828 | !cfq_io_thinktime_big(cfqd, &service_tree->ttime, false)) | |
c1e44756 | 1829 | return true; |
b1ffe737 DS |
1830 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", |
1831 | service_tree->count); | |
c1e44756 | 1832 | return false; |
a6d44e98 CZ |
1833 | } |
1834 | ||
6d048f53 | 1835 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 1836 | { |
1792669c | 1837 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 1838 | struct cfq_io_cq *cic; |
80bdf0c7 | 1839 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 1840 | |
a68bbddb | 1841 | /* |
f7d7b7a7 JA |
1842 | * SSD device without seek penalty, disable idling. But only do so |
1843 | * for devices that support queuing, otherwise we still have a problem | |
1844 | * with sync vs async workloads. | |
a68bbddb | 1845 | */ |
f7d7b7a7 | 1846 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
1847 | return; |
1848 | ||
dd67d051 | 1849 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 1850 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
1851 | |
1852 | /* | |
1853 | * idle is disabled, either manually or by past process history | |
1854 | */ | |
80bdf0c7 VG |
1855 | if (!cfq_should_idle(cfqd, cfqq)) { |
1856 | /* no queue idling. Check for group idling */ | |
1857 | if (cfqd->cfq_group_idle) | |
1858 | group_idle = cfqd->cfq_group_idle; | |
1859 | else | |
1860 | return; | |
1861 | } | |
6d048f53 | 1862 | |
7b679138 | 1863 | /* |
8e550632 | 1864 | * still active requests from this queue, don't idle |
7b679138 | 1865 | */ |
8e550632 | 1866 | if (cfqq->dispatched) |
7b679138 JA |
1867 | return; |
1868 | ||
22e2c507 JA |
1869 | /* |
1870 | * task has exited, don't wait | |
1871 | */ | |
206dc69b | 1872 | cic = cfqd->active_cic; |
f6e8d01b | 1873 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
1874 | return; |
1875 | ||
355b659c CZ |
1876 | /* |
1877 | * If our average think time is larger than the remaining time | |
1878 | * slice, then don't idle. This avoids overrunning the allotted | |
1879 | * time slice. | |
1880 | */ | |
383cd721 SL |
1881 | if (sample_valid(cic->ttime.ttime_samples) && |
1882 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 1883 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 1884 | cic->ttime.ttime_mean); |
355b659c | 1885 | return; |
b1ffe737 | 1886 | } |
355b659c | 1887 | |
80bdf0c7 VG |
1888 | /* There are other queues in the group, don't do group idle */ |
1889 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
1890 | return; | |
1891 | ||
3b18152c | 1892 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 1893 | |
80bdf0c7 VG |
1894 | if (group_idle) |
1895 | sl = cfqd->cfq_group_idle; | |
1896 | else | |
1897 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 1898 | |
7b14e3b5 | 1899 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
c1768268 TH |
1900 | cfq_blkiocg_update_set_idle_time_stats(cfqg_to_blkg(cfqq->cfqg), |
1901 | &blkio_policy_cfq); | |
80bdf0c7 VG |
1902 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
1903 | group_idle ? 1 : 0); | |
1da177e4 LT |
1904 | } |
1905 | ||
498d3aa2 JA |
1906 | /* |
1907 | * Move request from internal lists to the request queue dispatch list. | |
1908 | */ | |
165125e1 | 1909 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 1910 | { |
3ed9a296 | 1911 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1912 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1913 | |
7b679138 JA |
1914 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
1915 | ||
06d21886 | 1916 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 1917 | cfq_remove_request(rq); |
6d048f53 | 1918 | cfqq->dispatched++; |
80bdf0c7 | 1919 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 1920 | elv_dispatch_sort(q, rq); |
3ed9a296 | 1921 | |
53c583d2 | 1922 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 1923 | cfqq->nr_sectors += blk_rq_sectors(rq); |
0381411e | 1924 | cfq_blkiocg_update_dispatch_stats(cfqg_to_blkg(cfqq->cfqg), |
c1768268 TH |
1925 | &blkio_policy_cfq, blk_rq_bytes(rq), |
1926 | rq_data_dir(rq), rq_is_sync(rq)); | |
1da177e4 LT |
1927 | } |
1928 | ||
1929 | /* | |
1930 | * return expired entry, or NULL to just start from scratch in rbtree | |
1931 | */ | |
febffd61 | 1932 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 1933 | { |
30996f40 | 1934 | struct request *rq = NULL; |
1da177e4 | 1935 | |
3b18152c | 1936 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 1937 | return NULL; |
cb887411 JA |
1938 | |
1939 | cfq_mark_cfqq_fifo_expire(cfqq); | |
1940 | ||
89850f7e JA |
1941 | if (list_empty(&cfqq->fifo)) |
1942 | return NULL; | |
1da177e4 | 1943 | |
89850f7e | 1944 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 1945 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 1946 | rq = NULL; |
1da177e4 | 1947 | |
30996f40 | 1948 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 1949 | return rq; |
1da177e4 LT |
1950 | } |
1951 | ||
22e2c507 JA |
1952 | static inline int |
1953 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1954 | { | |
1955 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 1956 | |
22e2c507 | 1957 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 1958 | |
b9f8ce05 | 1959 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
1960 | } |
1961 | ||
df5fe3e8 JM |
1962 | /* |
1963 | * Must be called with the queue_lock held. | |
1964 | */ | |
1965 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
1966 | { | |
1967 | int process_refs, io_refs; | |
1968 | ||
1969 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 1970 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
1971 | BUG_ON(process_refs < 0); |
1972 | return process_refs; | |
1973 | } | |
1974 | ||
1975 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
1976 | { | |
e6c5bc73 | 1977 | int process_refs, new_process_refs; |
df5fe3e8 JM |
1978 | struct cfq_queue *__cfqq; |
1979 | ||
c10b61f0 JM |
1980 | /* |
1981 | * If there are no process references on the new_cfqq, then it is | |
1982 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
1983 | * chain may have dropped their last reference (not just their | |
1984 | * last process reference). | |
1985 | */ | |
1986 | if (!cfqq_process_refs(new_cfqq)) | |
1987 | return; | |
1988 | ||
df5fe3e8 JM |
1989 | /* Avoid a circular list and skip interim queue merges */ |
1990 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
1991 | if (__cfqq == cfqq) | |
1992 | return; | |
1993 | new_cfqq = __cfqq; | |
1994 | } | |
1995 | ||
1996 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 1997 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
1998 | /* |
1999 | * If the process for the cfqq has gone away, there is no | |
2000 | * sense in merging the queues. | |
2001 | */ | |
c10b61f0 | 2002 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2003 | return; |
2004 | ||
e6c5bc73 JM |
2005 | /* |
2006 | * Merge in the direction of the lesser amount of work. | |
2007 | */ | |
e6c5bc73 JM |
2008 | if (new_process_refs >= process_refs) { |
2009 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2010 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2011 | } else { |
2012 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2013 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2014 | } |
df5fe3e8 JM |
2015 | } |
2016 | ||
cdb16e8f | 2017 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
65b32a57 | 2018 | struct cfq_group *cfqg, enum wl_prio_t prio) |
718eee05 CZ |
2019 | { |
2020 | struct cfq_queue *queue; | |
2021 | int i; | |
2022 | bool key_valid = false; | |
2023 | unsigned long lowest_key = 0; | |
2024 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2025 | ||
65b32a57 VG |
2026 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2027 | /* select the one with lowest rb_key */ | |
2028 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); | |
718eee05 CZ |
2029 | if (queue && |
2030 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2031 | lowest_key = queue->rb_key; | |
2032 | cur_best = i; | |
2033 | key_valid = true; | |
2034 | } | |
2035 | } | |
2036 | ||
2037 | return cur_best; | |
2038 | } | |
2039 | ||
cdb16e8f | 2040 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
718eee05 | 2041 | { |
718eee05 CZ |
2042 | unsigned slice; |
2043 | unsigned count; | |
cdb16e8f | 2044 | struct cfq_rb_root *st; |
58ff82f3 | 2045 | unsigned group_slice; |
e4ea0c16 | 2046 | enum wl_prio_t original_prio = cfqd->serving_prio; |
1fa8f6d6 | 2047 | |
718eee05 | 2048 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2049 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
718eee05 | 2050 | cfqd->serving_prio = RT_WORKLOAD; |
58ff82f3 | 2051 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
718eee05 CZ |
2052 | cfqd->serving_prio = BE_WORKLOAD; |
2053 | else { | |
2054 | cfqd->serving_prio = IDLE_WORKLOAD; | |
2055 | cfqd->workload_expires = jiffies + 1; | |
2056 | return; | |
2057 | } | |
2058 | ||
e4ea0c16 SL |
2059 | if (original_prio != cfqd->serving_prio) |
2060 | goto new_workload; | |
2061 | ||
718eee05 CZ |
2062 | /* |
2063 | * For RT and BE, we have to choose also the type | |
2064 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2065 | * expiration time | |
2066 | */ | |
65b32a57 | 2067 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
cdb16e8f | 2068 | count = st->count; |
718eee05 CZ |
2069 | |
2070 | /* | |
65b32a57 | 2071 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2072 | */ |
65b32a57 | 2073 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2074 | return; |
2075 | ||
e4ea0c16 | 2076 | new_workload: |
718eee05 CZ |
2077 | /* otherwise select new workload type */ |
2078 | cfqd->serving_type = | |
65b32a57 VG |
2079 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); |
2080 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); | |
cdb16e8f | 2081 | count = st->count; |
718eee05 CZ |
2082 | |
2083 | /* | |
2084 | * the workload slice is computed as a fraction of target latency | |
2085 | * proportional to the number of queues in that workload, over | |
2086 | * all the queues in the same priority class | |
2087 | */ | |
58ff82f3 VG |
2088 | group_slice = cfq_group_slice(cfqd, cfqg); |
2089 | ||
2090 | slice = group_slice * count / | |
2091 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], | |
2092 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); | |
718eee05 | 2093 | |
f26bd1f0 VG |
2094 | if (cfqd->serving_type == ASYNC_WORKLOAD) { |
2095 | unsigned int tmp; | |
2096 | ||
2097 | /* | |
2098 | * Async queues are currently system wide. Just taking | |
2099 | * proportion of queues with-in same group will lead to higher | |
2100 | * async ratio system wide as generally root group is going | |
2101 | * to have higher weight. A more accurate thing would be to | |
2102 | * calculate system wide asnc/sync ratio. | |
2103 | */ | |
2104 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); | |
2105 | tmp = tmp/cfqd->busy_queues; | |
2106 | slice = min_t(unsigned, slice, tmp); | |
2107 | ||
718eee05 CZ |
2108 | /* async workload slice is scaled down according to |
2109 | * the sync/async slice ratio. */ | |
2110 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2111 | } else |
718eee05 CZ |
2112 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2113 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2114 | ||
2115 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2116 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
2117 | cfqd->workload_expires = jiffies + slice; |
2118 | } | |
2119 | ||
1fa8f6d6 VG |
2120 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2121 | { | |
2122 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2123 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2124 | |
2125 | if (RB_EMPTY_ROOT(&st->rb)) | |
2126 | return NULL; | |
25bc6b07 | 2127 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
2128 | update_min_vdisktime(st); |
2129 | return cfqg; | |
1fa8f6d6 VG |
2130 | } |
2131 | ||
cdb16e8f VG |
2132 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2133 | { | |
1fa8f6d6 VG |
2134 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2135 | ||
2136 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2137 | |
2138 | /* Restore the workload type data */ | |
2139 | if (cfqg->saved_workload_slice) { | |
2140 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; | |
2141 | cfqd->serving_type = cfqg->saved_workload; | |
2142 | cfqd->serving_prio = cfqg->saved_serving_prio; | |
66ae2919 GJ |
2143 | } else |
2144 | cfqd->workload_expires = jiffies - 1; | |
2145 | ||
1fa8f6d6 | 2146 | choose_service_tree(cfqd, cfqg); |
cdb16e8f VG |
2147 | } |
2148 | ||
22e2c507 | 2149 | /* |
498d3aa2 JA |
2150 | * Select a queue for service. If we have a current active queue, |
2151 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2152 | */ |
1b5ed5e1 | 2153 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2154 | { |
a36e71f9 | 2155 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2156 | |
22e2c507 JA |
2157 | cfqq = cfqd->active_queue; |
2158 | if (!cfqq) | |
2159 | goto new_queue; | |
1da177e4 | 2160 | |
f04a6424 VG |
2161 | if (!cfqd->rq_queued) |
2162 | return NULL; | |
c244bb50 VG |
2163 | |
2164 | /* | |
2165 | * We were waiting for group to get backlogged. Expire the queue | |
2166 | */ | |
2167 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2168 | goto expire; | |
2169 | ||
22e2c507 | 2170 | /* |
6d048f53 | 2171 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2172 | */ |
7667aa06 VG |
2173 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2174 | /* | |
2175 | * If slice had not expired at the completion of last request | |
2176 | * we might not have turned on wait_busy flag. Don't expire | |
2177 | * the queue yet. Allow the group to get backlogged. | |
2178 | * | |
2179 | * The very fact that we have used the slice, that means we | |
2180 | * have been idling all along on this queue and it should be | |
2181 | * ok to wait for this request to complete. | |
2182 | */ | |
82bbbf28 VG |
2183 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2184 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2185 | cfqq = NULL; | |
7667aa06 | 2186 | goto keep_queue; |
82bbbf28 | 2187 | } else |
80bdf0c7 | 2188 | goto check_group_idle; |
7667aa06 | 2189 | } |
1da177e4 | 2190 | |
22e2c507 | 2191 | /* |
6d048f53 JA |
2192 | * The active queue has requests and isn't expired, allow it to |
2193 | * dispatch. | |
22e2c507 | 2194 | */ |
dd67d051 | 2195 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2196 | goto keep_queue; |
6d048f53 | 2197 | |
a36e71f9 JA |
2198 | /* |
2199 | * If another queue has a request waiting within our mean seek | |
2200 | * distance, let it run. The expire code will check for close | |
2201 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2202 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2203 | */ |
b3b6d040 | 2204 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2205 | if (new_cfqq) { |
2206 | if (!cfqq->new_cfqq) | |
2207 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2208 | goto expire; |
df5fe3e8 | 2209 | } |
a36e71f9 | 2210 | |
6d048f53 JA |
2211 | /* |
2212 | * No requests pending. If the active queue still has requests in | |
2213 | * flight or is idling for a new request, allow either of these | |
2214 | * conditions to happen (or time out) before selecting a new queue. | |
2215 | */ | |
80bdf0c7 VG |
2216 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2217 | cfqq = NULL; | |
2218 | goto keep_queue; | |
2219 | } | |
2220 | ||
8e1ac665 SL |
2221 | /* |
2222 | * This is a deep seek queue, but the device is much faster than | |
2223 | * the queue can deliver, don't idle | |
2224 | **/ | |
2225 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
2226 | (cfq_cfqq_slice_new(cfqq) || | |
2227 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
2228 | cfq_clear_cfqq_deep(cfqq); | |
2229 | cfq_clear_cfqq_idle_window(cfqq); | |
2230 | } | |
2231 | ||
80bdf0c7 VG |
2232 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2233 | cfqq = NULL; | |
2234 | goto keep_queue; | |
2235 | } | |
2236 | ||
2237 | /* | |
2238 | * If group idle is enabled and there are requests dispatched from | |
2239 | * this group, wait for requests to complete. | |
2240 | */ | |
2241 | check_group_idle: | |
7700fc4f SL |
2242 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
2243 | cfqq->cfqg->dispatched && | |
2244 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
2245 | cfqq = NULL; |
2246 | goto keep_queue; | |
22e2c507 JA |
2247 | } |
2248 | ||
3b18152c | 2249 | expire: |
e5ff082e | 2250 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2251 | new_queue: |
718eee05 CZ |
2252 | /* |
2253 | * Current queue expired. Check if we have to switch to a new | |
2254 | * service tree | |
2255 | */ | |
2256 | if (!new_cfqq) | |
cdb16e8f | 2257 | cfq_choose_cfqg(cfqd); |
718eee05 | 2258 | |
a36e71f9 | 2259 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2260 | keep_queue: |
3b18152c | 2261 | return cfqq; |
22e2c507 JA |
2262 | } |
2263 | ||
febffd61 | 2264 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2265 | { |
2266 | int dispatched = 0; | |
2267 | ||
2268 | while (cfqq->next_rq) { | |
2269 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2270 | dispatched++; | |
2271 | } | |
2272 | ||
2273 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2274 | |
2275 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 2276 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
2277 | return dispatched; |
2278 | } | |
2279 | ||
498d3aa2 JA |
2280 | /* |
2281 | * Drain our current requests. Used for barriers and when switching | |
2282 | * io schedulers on-the-fly. | |
2283 | */ | |
d9e7620e | 2284 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2285 | { |
0871714e | 2286 | struct cfq_queue *cfqq; |
d9e7620e | 2287 | int dispatched = 0; |
cdb16e8f | 2288 | |
3440c49f | 2289 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 2290 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
2291 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2292 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 2293 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 2294 | } |
1b5ed5e1 | 2295 | |
1b5ed5e1 TH |
2296 | BUG_ON(cfqd->busy_queues); |
2297 | ||
6923715a | 2298 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2299 | return dispatched; |
2300 | } | |
2301 | ||
abc3c744 SL |
2302 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2303 | struct cfq_queue *cfqq) | |
2304 | { | |
2305 | /* the queue hasn't finished any request, can't estimate */ | |
2306 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 2307 | return true; |
abc3c744 SL |
2308 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
2309 | cfqq->slice_end)) | |
c1e44756 | 2310 | return true; |
abc3c744 | 2311 | |
c1e44756 | 2312 | return false; |
abc3c744 SL |
2313 | } |
2314 | ||
0b182d61 | 2315 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2316 | { |
2f5cb738 | 2317 | unsigned int max_dispatch; |
22e2c507 | 2318 | |
5ad531db JA |
2319 | /* |
2320 | * Drain async requests before we start sync IO | |
2321 | */ | |
53c583d2 | 2322 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 2323 | return false; |
5ad531db | 2324 | |
2f5cb738 JA |
2325 | /* |
2326 | * If this is an async queue and we have sync IO in flight, let it wait | |
2327 | */ | |
53c583d2 | 2328 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 2329 | return false; |
2f5cb738 | 2330 | |
abc3c744 | 2331 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
2332 | if (cfq_class_idle(cfqq)) |
2333 | max_dispatch = 1; | |
b4878f24 | 2334 | |
2f5cb738 JA |
2335 | /* |
2336 | * Does this cfqq already have too much IO in flight? | |
2337 | */ | |
2338 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 2339 | bool promote_sync = false; |
2f5cb738 JA |
2340 | /* |
2341 | * idle queue must always only have a single IO in flight | |
2342 | */ | |
3ed9a296 | 2343 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2344 | return false; |
3ed9a296 | 2345 | |
ef8a41df | 2346 | /* |
c4ade94f LS |
2347 | * If there is only one sync queue |
2348 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
2349 | * queue no dispatch limit. The reason is a sync queue can |
2350 | * preempt async queue, limiting the sync queue doesn't make | |
2351 | * sense. This is useful for aiostress test. | |
2352 | */ | |
c4ade94f LS |
2353 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
2354 | promote_sync = true; | |
ef8a41df | 2355 | |
2f5cb738 JA |
2356 | /* |
2357 | * We have other queues, don't allow more IO from this one | |
2358 | */ | |
ef8a41df SL |
2359 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
2360 | !promote_sync) | |
0b182d61 | 2361 | return false; |
9ede209e | 2362 | |
365722bb | 2363 | /* |
474b18cc | 2364 | * Sole queue user, no limit |
365722bb | 2365 | */ |
ef8a41df | 2366 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
2367 | max_dispatch = -1; |
2368 | else | |
2369 | /* | |
2370 | * Normally we start throttling cfqq when cfq_quantum/2 | |
2371 | * requests have been dispatched. But we can drive | |
2372 | * deeper queue depths at the beginning of slice | |
2373 | * subjected to upper limit of cfq_quantum. | |
2374 | * */ | |
2375 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
2376 | } |
2377 | ||
2378 | /* | |
2379 | * Async queues must wait a bit before being allowed dispatch. | |
2380 | * We also ramp up the dispatch depth gradually for async IO, | |
2381 | * based on the last sync IO we serviced | |
2382 | */ | |
963b72fc | 2383 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 2384 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 2385 | unsigned int depth; |
365722bb | 2386 | |
61f0c1dc | 2387 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2388 | if (!depth && !cfqq->dispatched) |
2389 | depth = 1; | |
8e296755 JA |
2390 | if (depth < max_dispatch) |
2391 | max_dispatch = depth; | |
2f5cb738 | 2392 | } |
3ed9a296 | 2393 | |
0b182d61 JA |
2394 | /* |
2395 | * If we're below the current max, allow a dispatch | |
2396 | */ | |
2397 | return cfqq->dispatched < max_dispatch; | |
2398 | } | |
2399 | ||
2400 | /* | |
2401 | * Dispatch a request from cfqq, moving them to the request queue | |
2402 | * dispatch list. | |
2403 | */ | |
2404 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2405 | { | |
2406 | struct request *rq; | |
2407 | ||
2408 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
2409 | ||
2410 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
2411 | return false; | |
2412 | ||
2413 | /* | |
2414 | * follow expired path, else get first next available | |
2415 | */ | |
2416 | rq = cfq_check_fifo(cfqq); | |
2417 | if (!rq) | |
2418 | rq = cfqq->next_rq; | |
2419 | ||
2420 | /* | |
2421 | * insert request into driver dispatch list | |
2422 | */ | |
2423 | cfq_dispatch_insert(cfqd->queue, rq); | |
2424 | ||
2425 | if (!cfqd->active_cic) { | |
c5869807 | 2426 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 2427 | |
c5869807 | 2428 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
2429 | cfqd->active_cic = cic; |
2430 | } | |
2431 | ||
2432 | return true; | |
2433 | } | |
2434 | ||
2435 | /* | |
2436 | * Find the cfqq that we need to service and move a request from that to the | |
2437 | * dispatch list | |
2438 | */ | |
2439 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
2440 | { | |
2441 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2442 | struct cfq_queue *cfqq; | |
2443 | ||
2444 | if (!cfqd->busy_queues) | |
2445 | return 0; | |
2446 | ||
2447 | if (unlikely(force)) | |
2448 | return cfq_forced_dispatch(cfqd); | |
2449 | ||
2450 | cfqq = cfq_select_queue(cfqd); | |
2451 | if (!cfqq) | |
8e296755 JA |
2452 | return 0; |
2453 | ||
2f5cb738 | 2454 | /* |
0b182d61 | 2455 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 2456 | */ |
0b182d61 JA |
2457 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2458 | return 0; | |
2459 | ||
2f5cb738 | 2460 | cfqq->slice_dispatch++; |
b029195d | 2461 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 2462 | |
2f5cb738 JA |
2463 | /* |
2464 | * expire an async queue immediately if it has used up its slice. idle | |
2465 | * queue always expire after 1 dispatch round. | |
2466 | */ | |
2467 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
2468 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
2469 | cfq_class_idle(cfqq))) { | |
2470 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 2471 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
2472 | } |
2473 | ||
b217a903 | 2474 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 2475 | return 1; |
1da177e4 LT |
2476 | } |
2477 | ||
1da177e4 | 2478 | /* |
5e705374 JA |
2479 | * task holds one reference to the queue, dropped when task exits. each rq |
2480 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 2481 | * |
b1c35769 | 2482 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
2483 | * queue lock must be held here. |
2484 | */ | |
2485 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
2486 | { | |
22e2c507 | 2487 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 2488 | struct cfq_group *cfqg; |
22e2c507 | 2489 | |
30d7b944 | 2490 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 2491 | |
30d7b944 SL |
2492 | cfqq->ref--; |
2493 | if (cfqq->ref) | |
1da177e4 LT |
2494 | return; |
2495 | ||
7b679138 | 2496 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 2497 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 2498 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 2499 | cfqg = cfqq->cfqg; |
1da177e4 | 2500 | |
28f95cbc | 2501 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 2502 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2503 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2504 | } |
22e2c507 | 2505 | |
f04a6424 | 2506 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 2507 | kmem_cache_free(cfq_pool, cfqq); |
1adaf3dd | 2508 | blkg_put(cfqg_to_blkg(cfqg)); |
1da177e4 LT |
2509 | } |
2510 | ||
d02a2c07 | 2511 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 2512 | { |
df5fe3e8 JM |
2513 | struct cfq_queue *__cfqq, *next; |
2514 | ||
df5fe3e8 JM |
2515 | /* |
2516 | * If this queue was scheduled to merge with another queue, be | |
2517 | * sure to drop the reference taken on that queue (and others in | |
2518 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
2519 | */ | |
2520 | __cfqq = cfqq->new_cfqq; | |
2521 | while (__cfqq) { | |
2522 | if (__cfqq == cfqq) { | |
2523 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
2524 | break; | |
2525 | } | |
2526 | next = __cfqq->new_cfqq; | |
2527 | cfq_put_queue(__cfqq); | |
2528 | __cfqq = next; | |
2529 | } | |
d02a2c07 SL |
2530 | } |
2531 | ||
2532 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2533 | { | |
2534 | if (unlikely(cfqq == cfqd->active_queue)) { | |
2535 | __cfq_slice_expired(cfqd, cfqq, 0); | |
2536 | cfq_schedule_dispatch(cfqd); | |
2537 | } | |
2538 | ||
2539 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 2540 | |
89850f7e JA |
2541 | cfq_put_queue(cfqq); |
2542 | } | |
22e2c507 | 2543 | |
9b84cacd TH |
2544 | static void cfq_init_icq(struct io_cq *icq) |
2545 | { | |
2546 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
2547 | ||
2548 | cic->ttime.last_end_request = jiffies; | |
2549 | } | |
2550 | ||
c5869807 | 2551 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 2552 | { |
c5869807 | 2553 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 2554 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 2555 | |
ff6657c6 JA |
2556 | if (cic->cfqq[BLK_RW_ASYNC]) { |
2557 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
2558 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
2559 | } |
2560 | ||
ff6657c6 JA |
2561 | if (cic->cfqq[BLK_RW_SYNC]) { |
2562 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
2563 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 2564 | } |
89850f7e JA |
2565 | } |
2566 | ||
abede6da | 2567 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
2568 | { |
2569 | struct task_struct *tsk = current; | |
2570 | int ioprio_class; | |
2571 | ||
3b18152c | 2572 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
2573 | return; |
2574 | ||
598971bf | 2575 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 2576 | switch (ioprio_class) { |
fe094d98 JA |
2577 | default: |
2578 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
2579 | case IOPRIO_CLASS_NONE: | |
2580 | /* | |
6d63c275 | 2581 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
2582 | */ |
2583 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 2584 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
2585 | break; |
2586 | case IOPRIO_CLASS_RT: | |
598971bf | 2587 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
2588 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
2589 | break; | |
2590 | case IOPRIO_CLASS_BE: | |
598971bf | 2591 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
2592 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
2593 | break; | |
2594 | case IOPRIO_CLASS_IDLE: | |
2595 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
2596 | cfqq->ioprio = 7; | |
2597 | cfq_clear_cfqq_idle_window(cfqq); | |
2598 | break; | |
22e2c507 JA |
2599 | } |
2600 | ||
2601 | /* | |
2602 | * keep track of original prio settings in case we have to temporarily | |
2603 | * elevate the priority of this queue | |
2604 | */ | |
2605 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 2606 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
2607 | } |
2608 | ||
598971bf | 2609 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 2610 | { |
598971bf | 2611 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 2612 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 2613 | struct cfq_queue *cfqq; |
35e6077c | 2614 | |
598971bf TH |
2615 | /* |
2616 | * Check whether ioprio has changed. The condition may trigger | |
2617 | * spuriously on a newly created cic but there's no harm. | |
2618 | */ | |
2619 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
2620 | return; |
2621 | ||
ff6657c6 | 2622 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
2623 | if (cfqq) { |
2624 | struct cfq_queue *new_cfqq; | |
abede6da TH |
2625 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio, |
2626 | GFP_ATOMIC); | |
caaa5f9f | 2627 | if (new_cfqq) { |
ff6657c6 | 2628 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
2629 | cfq_put_queue(cfqq); |
2630 | } | |
22e2c507 | 2631 | } |
caaa5f9f | 2632 | |
ff6657c6 | 2633 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
2634 | if (cfqq) |
2635 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
2636 | |
2637 | cic->ioprio = ioprio; | |
22e2c507 JA |
2638 | } |
2639 | ||
d5036d77 | 2640 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2641 | pid_t pid, bool is_sync) |
d5036d77 JA |
2642 | { |
2643 | RB_CLEAR_NODE(&cfqq->rb_node); | |
2644 | RB_CLEAR_NODE(&cfqq->p_node); | |
2645 | INIT_LIST_HEAD(&cfqq->fifo); | |
2646 | ||
30d7b944 | 2647 | cfqq->ref = 0; |
d5036d77 JA |
2648 | cfqq->cfqd = cfqd; |
2649 | ||
2650 | cfq_mark_cfqq_prio_changed(cfqq); | |
2651 | ||
2652 | if (is_sync) { | |
2653 | if (!cfq_class_idle(cfqq)) | |
2654 | cfq_mark_cfqq_idle_window(cfqq); | |
2655 | cfq_mark_cfqq_sync(cfqq); | |
2656 | } | |
2657 | cfqq->pid = pid; | |
2658 | } | |
2659 | ||
24610333 | 2660 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
598971bf | 2661 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 2662 | { |
bca4b914 | 2663 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
598971bf TH |
2664 | struct cfq_queue *sync_cfqq; |
2665 | uint64_t id; | |
24610333 | 2666 | |
598971bf TH |
2667 | rcu_read_lock(); |
2668 | id = bio_blkio_cgroup(bio)->id; | |
2669 | rcu_read_unlock(); | |
24610333 | 2670 | |
598971bf TH |
2671 | /* |
2672 | * Check whether blkcg has changed. The condition may trigger | |
2673 | * spuriously on a newly created cic but there's no harm. | |
2674 | */ | |
2675 | if (unlikely(!cfqd) || likely(cic->blkcg_id == id)) | |
2676 | return; | |
24610333 | 2677 | |
598971bf | 2678 | sync_cfqq = cic_to_cfqq(cic, 1); |
24610333 VG |
2679 | if (sync_cfqq) { |
2680 | /* | |
2681 | * Drop reference to sync queue. A new sync queue will be | |
2682 | * assigned in new group upon arrival of a fresh request. | |
2683 | */ | |
2684 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
2685 | cic_set_cfqq(cic, NULL, 1); | |
2686 | cfq_put_queue(sync_cfqq); | |
2687 | } | |
598971bf TH |
2688 | |
2689 | cic->blkcg_id = id; | |
24610333 | 2690 | } |
598971bf TH |
2691 | #else |
2692 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } | |
24610333 VG |
2693 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
2694 | ||
22e2c507 | 2695 | static struct cfq_queue * |
abede6da TH |
2696 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
2697 | struct bio *bio, gfp_t gfp_mask) | |
22e2c507 | 2698 | { |
0a5a7d0e | 2699 | struct blkio_cgroup *blkcg; |
22e2c507 | 2700 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
cdb16e8f | 2701 | struct cfq_group *cfqg; |
22e2c507 JA |
2702 | |
2703 | retry: | |
2a7f1244 TH |
2704 | rcu_read_lock(); |
2705 | ||
4f85cb96 | 2706 | blkcg = bio_blkio_cgroup(bio); |
cd1604fa | 2707 | cfqg = cfq_lookup_create_cfqg(cfqd, blkcg); |
91fac317 | 2708 | cfqq = cic_to_cfqq(cic, is_sync); |
22e2c507 | 2709 | |
6118b70b JA |
2710 | /* |
2711 | * Always try a new alloc if we fell back to the OOM cfqq | |
2712 | * originally, since it should just be a temporary situation. | |
2713 | */ | |
2714 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
2715 | cfqq = NULL; | |
22e2c507 JA |
2716 | if (new_cfqq) { |
2717 | cfqq = new_cfqq; | |
2718 | new_cfqq = NULL; | |
2719 | } else if (gfp_mask & __GFP_WAIT) { | |
2a7f1244 | 2720 | rcu_read_unlock(); |
22e2c507 | 2721 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c | 2722 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 2723 | gfp_mask | __GFP_ZERO, |
94f6030c | 2724 | cfqd->queue->node); |
22e2c507 | 2725 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
2726 | if (new_cfqq) |
2727 | goto retry; | |
22e2c507 | 2728 | } else { |
94f6030c CL |
2729 | cfqq = kmem_cache_alloc_node(cfq_pool, |
2730 | gfp_mask | __GFP_ZERO, | |
2731 | cfqd->queue->node); | |
22e2c507 JA |
2732 | } |
2733 | ||
6118b70b JA |
2734 | if (cfqq) { |
2735 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
abede6da | 2736 | cfq_init_prio_data(cfqq, cic); |
cdb16e8f | 2737 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
2738 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
2739 | } else | |
2740 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
2741 | } |
2742 | ||
2743 | if (new_cfqq) | |
2744 | kmem_cache_free(cfq_pool, new_cfqq); | |
2745 | ||
2a7f1244 | 2746 | rcu_read_unlock(); |
22e2c507 JA |
2747 | return cfqq; |
2748 | } | |
2749 | ||
c2dea2d1 VT |
2750 | static struct cfq_queue ** |
2751 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
2752 | { | |
fe094d98 | 2753 | switch (ioprio_class) { |
c2dea2d1 VT |
2754 | case IOPRIO_CLASS_RT: |
2755 | return &cfqd->async_cfqq[0][ioprio]; | |
598971bf TH |
2756 | case IOPRIO_CLASS_NONE: |
2757 | ioprio = IOPRIO_NORM; | |
2758 | /* fall through */ | |
c2dea2d1 VT |
2759 | case IOPRIO_CLASS_BE: |
2760 | return &cfqd->async_cfqq[1][ioprio]; | |
2761 | case IOPRIO_CLASS_IDLE: | |
2762 | return &cfqd->async_idle_cfqq; | |
2763 | default: | |
2764 | BUG(); | |
2765 | } | |
2766 | } | |
2767 | ||
15c31be4 | 2768 | static struct cfq_queue * |
abede6da | 2769 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
4f85cb96 | 2770 | struct bio *bio, gfp_t gfp_mask) |
15c31be4 | 2771 | { |
598971bf TH |
2772 | const int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
2773 | const int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
c2dea2d1 | 2774 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
2775 | struct cfq_queue *cfqq = NULL; |
2776 | ||
c2dea2d1 VT |
2777 | if (!is_sync) { |
2778 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
2779 | cfqq = *async_cfqq; | |
2780 | } | |
2781 | ||
6118b70b | 2782 | if (!cfqq) |
abede6da | 2783 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, cic, bio, gfp_mask); |
15c31be4 JA |
2784 | |
2785 | /* | |
2786 | * pin the queue now that it's allocated, scheduler exit will prune it | |
2787 | */ | |
c2dea2d1 | 2788 | if (!is_sync && !(*async_cfqq)) { |
30d7b944 | 2789 | cfqq->ref++; |
c2dea2d1 | 2790 | *async_cfqq = cfqq; |
15c31be4 JA |
2791 | } |
2792 | ||
30d7b944 | 2793 | cfqq->ref++; |
15c31be4 JA |
2794 | return cfqq; |
2795 | } | |
2796 | ||
22e2c507 | 2797 | static void |
383cd721 | 2798 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 2799 | { |
383cd721 SL |
2800 | unsigned long elapsed = jiffies - ttime->last_end_request; |
2801 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 2802 | |
383cd721 SL |
2803 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
2804 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
2805 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
2806 | } | |
2807 | ||
2808 | static void | |
2809 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 2810 | struct cfq_io_cq *cic) |
383cd721 | 2811 | { |
f5f2b6ce | 2812 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 2813 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
2814 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
2815 | cfqd->cfq_slice_idle); | |
2816 | } | |
7700fc4f SL |
2817 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2818 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
2819 | #endif | |
22e2c507 | 2820 | } |
1da177e4 | 2821 | |
206dc69b | 2822 | static void |
b2c18e1e | 2823 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 2824 | struct request *rq) |
206dc69b | 2825 | { |
3dde36dd | 2826 | sector_t sdist = 0; |
41647e7a | 2827 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
2828 | if (cfqq->last_request_pos) { |
2829 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
2830 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
2831 | else | |
2832 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
2833 | } | |
206dc69b | 2834 | |
3dde36dd | 2835 | cfqq->seek_history <<= 1; |
41647e7a CZ |
2836 | if (blk_queue_nonrot(cfqd->queue)) |
2837 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
2838 | else | |
2839 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 2840 | } |
1da177e4 | 2841 | |
22e2c507 JA |
2842 | /* |
2843 | * Disable idle window if the process thinks too long or seeks so much that | |
2844 | * it doesn't matter | |
2845 | */ | |
2846 | static void | |
2847 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 2848 | struct cfq_io_cq *cic) |
22e2c507 | 2849 | { |
7b679138 | 2850 | int old_idle, enable_idle; |
1be92f2f | 2851 | |
0871714e JA |
2852 | /* |
2853 | * Don't idle for async or idle io prio class | |
2854 | */ | |
2855 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
2856 | return; |
2857 | ||
c265a7f4 | 2858 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 2859 | |
76280aff CZ |
2860 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
2861 | cfq_mark_cfqq_deep(cfqq); | |
2862 | ||
749ef9f8 CZ |
2863 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
2864 | enable_idle = 0; | |
f6e8d01b | 2865 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
2866 | !cfqd->cfq_slice_idle || |
2867 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 2868 | enable_idle = 0; |
383cd721 SL |
2869 | else if (sample_valid(cic->ttime.ttime_samples)) { |
2870 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
2871 | enable_idle = 0; |
2872 | else | |
2873 | enable_idle = 1; | |
1da177e4 LT |
2874 | } |
2875 | ||
7b679138 JA |
2876 | if (old_idle != enable_idle) { |
2877 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
2878 | if (enable_idle) | |
2879 | cfq_mark_cfqq_idle_window(cfqq); | |
2880 | else | |
2881 | cfq_clear_cfqq_idle_window(cfqq); | |
2882 | } | |
22e2c507 | 2883 | } |
1da177e4 | 2884 | |
22e2c507 JA |
2885 | /* |
2886 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
2887 | * no or if we aren't sure, a 1 will cause a preempt. | |
2888 | */ | |
a6151c3a | 2889 | static bool |
22e2c507 | 2890 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 2891 | struct request *rq) |
22e2c507 | 2892 | { |
6d048f53 | 2893 | struct cfq_queue *cfqq; |
22e2c507 | 2894 | |
6d048f53 JA |
2895 | cfqq = cfqd->active_queue; |
2896 | if (!cfqq) | |
a6151c3a | 2897 | return false; |
22e2c507 | 2898 | |
6d048f53 | 2899 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 2900 | return false; |
22e2c507 JA |
2901 | |
2902 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 2903 | return true; |
1e3335de | 2904 | |
875feb63 DS |
2905 | /* |
2906 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
2907 | */ | |
2908 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
2909 | return false; | |
2910 | ||
374f84ac JA |
2911 | /* |
2912 | * if the new request is sync, but the currently running queue is | |
2913 | * not, let the sync request have priority. | |
2914 | */ | |
5e705374 | 2915 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 2916 | return true; |
1e3335de | 2917 | |
8682e1f1 VG |
2918 | if (new_cfqq->cfqg != cfqq->cfqg) |
2919 | return false; | |
2920 | ||
2921 | if (cfq_slice_used(cfqq)) | |
2922 | return true; | |
2923 | ||
2924 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
2925 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && | |
2926 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && | |
2927 | new_cfqq->service_tree->count == 2 && | |
2928 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2929 | return true; | |
2930 | ||
b53d1ed7 JA |
2931 | /* |
2932 | * So both queues are sync. Let the new request get disk time if | |
2933 | * it's a metadata request and the current queue is doing regular IO. | |
2934 | */ | |
65299a3b | 2935 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
2936 | return true; |
2937 | ||
3a9a3f6c DS |
2938 | /* |
2939 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
2940 | */ | |
2941 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 2942 | return true; |
3a9a3f6c | 2943 | |
d2d59e18 SL |
2944 | /* An idle queue should not be idle now for some reason */ |
2945 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
2946 | return true; | |
2947 | ||
1e3335de | 2948 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 2949 | return false; |
1e3335de JA |
2950 | |
2951 | /* | |
2952 | * if this request is as-good as one we would expect from the | |
2953 | * current cfqq, let it preempt | |
2954 | */ | |
e9ce335d | 2955 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 2956 | return true; |
1e3335de | 2957 | |
a6151c3a | 2958 | return false; |
22e2c507 JA |
2959 | } |
2960 | ||
2961 | /* | |
2962 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
2963 | * let it have half of its nominal slice. | |
2964 | */ | |
2965 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2966 | { | |
df0793ab SL |
2967 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
2968 | ||
7b679138 | 2969 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 2970 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 2971 | |
f8ae6e3e SL |
2972 | /* |
2973 | * workload type is changed, don't save slice, otherwise preempt | |
2974 | * doesn't happen | |
2975 | */ | |
df0793ab | 2976 | if (old_type != cfqq_type(cfqq)) |
f8ae6e3e SL |
2977 | cfqq->cfqg->saved_workload_slice = 0; |
2978 | ||
bf572256 JA |
2979 | /* |
2980 | * Put the new queue at the front of the of the current list, | |
2981 | * so we know that it will be selected next. | |
2982 | */ | |
2983 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
2984 | |
2985 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 2986 | |
62a37f6b JT |
2987 | cfqq->slice_end = 0; |
2988 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
2989 | } |
2990 | ||
22e2c507 | 2991 | /* |
5e705374 | 2992 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
2993 | * something we should do about it |
2994 | */ | |
2995 | static void | |
5e705374 JA |
2996 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
2997 | struct request *rq) | |
22e2c507 | 2998 | { |
c5869807 | 2999 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3000 | |
45333d5a | 3001 | cfqd->rq_queued++; |
65299a3b CH |
3002 | if (rq->cmd_flags & REQ_PRIO) |
3003 | cfqq->prio_pending++; | |
374f84ac | 3004 | |
383cd721 | 3005 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3006 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3007 | cfq_update_idle_window(cfqd, cfqq, cic); |
3008 | ||
b2c18e1e | 3009 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3010 | |
3011 | if (cfqq == cfqd->active_queue) { | |
3012 | /* | |
b029195d JA |
3013 | * Remember that we saw a request from this process, but |
3014 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3015 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3016 | * and merging. If the request is already larger than a single |
3017 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3018 | * merging is already done. Ditto for a busy system that |
3019 | * has other work pending, don't risk delaying until the | |
3020 | * idle timer unplug to continue working. | |
22e2c507 | 3021 | */ |
d6ceb25e | 3022 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3023 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3024 | cfqd->busy_queues > 1) { | |
812df48d | 3025 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3026 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3027 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3028 | } else { |
e98ef89b | 3029 | cfq_blkiocg_update_idle_time_stats( |
c1768268 TH |
3030 | cfqg_to_blkg(cfqq->cfqg), |
3031 | &blkio_policy_cfq); | |
bf791937 | 3032 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3033 | } |
d6ceb25e | 3034 | } |
5e705374 | 3035 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3036 | /* |
3037 | * not the active queue - expire current slice if it is | |
3038 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3039 | * has some old slice time left and is of higher priority or |
3040 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3041 | */ |
3042 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3043 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3044 | } |
1da177e4 LT |
3045 | } |
3046 | ||
165125e1 | 3047 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3048 | { |
b4878f24 | 3049 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3050 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3051 | |
7b679138 | 3052 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 3053 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 3054 | |
30996f40 | 3055 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3056 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3057 | cfq_add_rq_rb(rq); |
0381411e | 3058 | cfq_blkiocg_update_io_add_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 | 3059 | &blkio_policy_cfq, |
0381411e TH |
3060 | cfqg_to_blkg(cfqd->serving_group), |
3061 | rq_data_dir(rq), rq_is_sync(rq)); | |
5e705374 | 3062 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3063 | } |
3064 | ||
45333d5a AC |
3065 | /* |
3066 | * Update hw_tag based on peak queue depth over 50 samples under | |
3067 | * sufficient load. | |
3068 | */ | |
3069 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3070 | { | |
1a1238a7 SL |
3071 | struct cfq_queue *cfqq = cfqd->active_queue; |
3072 | ||
53c583d2 CZ |
3073 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3074 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3075 | |
3076 | if (cfqd->hw_tag == 1) | |
3077 | return; | |
45333d5a AC |
3078 | |
3079 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3080 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3081 | return; |
3082 | ||
1a1238a7 SL |
3083 | /* |
3084 | * If active queue hasn't enough requests and can idle, cfq might not | |
3085 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3086 | * case | |
3087 | */ | |
3088 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3089 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3090 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3091 | return; |
3092 | ||
45333d5a AC |
3093 | if (cfqd->hw_tag_samples++ < 50) |
3094 | return; | |
3095 | ||
e459dd08 | 3096 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3097 | cfqd->hw_tag = 1; |
3098 | else | |
3099 | cfqd->hw_tag = 0; | |
45333d5a AC |
3100 | } |
3101 | ||
7667aa06 VG |
3102 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3103 | { | |
c5869807 | 3104 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 3105 | |
02a8f01b JT |
3106 | /* If the queue already has requests, don't wait */ |
3107 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3108 | return false; | |
3109 | ||
7667aa06 VG |
3110 | /* If there are other queues in the group, don't wait */ |
3111 | if (cfqq->cfqg->nr_cfqq > 1) | |
3112 | return false; | |
3113 | ||
7700fc4f SL |
3114 | /* the only queue in the group, but think time is big */ |
3115 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
3116 | return false; | |
3117 | ||
7667aa06 VG |
3118 | if (cfq_slice_used(cfqq)) |
3119 | return true; | |
3120 | ||
3121 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
3122 | if (cic && sample_valid(cic->ttime.ttime_samples) |
3123 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
3124 | return true; |
3125 | ||
3126 | /* | |
3127 | * If think times is less than a jiffy than ttime_mean=0 and above | |
3128 | * will not be true. It might happen that slice has not expired yet | |
3129 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
3130 | * case where think time is less than a jiffy, mark the queue wait | |
3131 | * busy if only 1 jiffy is left in the slice. | |
3132 | */ | |
3133 | if (cfqq->slice_end - jiffies == 1) | |
3134 | return true; | |
3135 | ||
3136 | return false; | |
3137 | } | |
3138 | ||
165125e1 | 3139 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3140 | { |
5e705374 | 3141 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3142 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3143 | const int sync = rq_is_sync(rq); |
b4878f24 | 3144 | unsigned long now; |
1da177e4 | 3145 | |
b4878f24 | 3146 | now = jiffies; |
33659ebb CH |
3147 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3148 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 3149 | |
45333d5a AC |
3150 | cfq_update_hw_tag(cfqd); |
3151 | ||
53c583d2 | 3152 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 3153 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 3154 | cfqd->rq_in_driver--; |
6d048f53 | 3155 | cfqq->dispatched--; |
80bdf0c7 | 3156 | (RQ_CFQG(rq))->dispatched--; |
0381411e | 3157 | cfq_blkiocg_update_completion_stats(cfqg_to_blkg(cfqq->cfqg), |
c1768268 TH |
3158 | &blkio_policy_cfq, rq_start_time_ns(rq), |
3159 | rq_io_start_time_ns(rq), rq_data_dir(rq), | |
3160 | rq_is_sync(rq)); | |
1da177e4 | 3161 | |
53c583d2 | 3162 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 3163 | |
365722bb | 3164 | if (sync) { |
f5f2b6ce SL |
3165 | struct cfq_rb_root *service_tree; |
3166 | ||
383cd721 | 3167 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
3168 | |
3169 | if (cfq_cfqq_on_rr(cfqq)) | |
3170 | service_tree = cfqq->service_tree; | |
3171 | else | |
3172 | service_tree = service_tree_for(cfqq->cfqg, | |
3173 | cfqq_prio(cfqq), cfqq_type(cfqq)); | |
3174 | service_tree->ttime.last_end_request = now; | |
573412b2 CZ |
3175 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3176 | cfqd->last_delayed_sync = now; | |
365722bb | 3177 | } |
caaa5f9f | 3178 | |
7700fc4f SL |
3179 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3180 | cfqq->cfqg->ttime.last_end_request = now; | |
3181 | #endif | |
3182 | ||
caaa5f9f JA |
3183 | /* |
3184 | * If this is the active queue, check if it needs to be expired, | |
3185 | * or if we want to idle in case it has no pending requests. | |
3186 | */ | |
3187 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3188 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3189 | ||
44f7c160 JA |
3190 | if (cfq_cfqq_slice_new(cfqq)) { |
3191 | cfq_set_prio_slice(cfqd, cfqq); | |
3192 | cfq_clear_cfqq_slice_new(cfqq); | |
3193 | } | |
f75edf2d VG |
3194 | |
3195 | /* | |
7667aa06 VG |
3196 | * Should we wait for next request to come in before we expire |
3197 | * the queue. | |
f75edf2d | 3198 | */ |
7667aa06 | 3199 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
3200 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3201 | if (!cfqd->cfq_slice_idle) | |
3202 | extend_sl = cfqd->cfq_group_idle; | |
3203 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 3204 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 3205 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
3206 | } |
3207 | ||
a36e71f9 | 3208 | /* |
8e550632 CZ |
3209 | * Idling is not enabled on: |
3210 | * - expired queues | |
3211 | * - idle-priority queues | |
3212 | * - async queues | |
3213 | * - queues with still some requests queued | |
3214 | * - when there is a close cooperator | |
a36e71f9 | 3215 | */ |
0871714e | 3216 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 3217 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3218 | else if (sync && cfqq_empty && |
3219 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 3220 | cfq_arm_slice_timer(cfqd); |
8e550632 | 3221 | } |
caaa5f9f | 3222 | } |
6d048f53 | 3223 | |
53c583d2 | 3224 | if (!cfqd->rq_in_driver) |
23e018a1 | 3225 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3226 | } |
3227 | ||
89850f7e | 3228 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3229 | { |
1b379d8d | 3230 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3231 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3232 | return ELV_MQUEUE_MUST; |
3b18152c | 3233 | } |
1da177e4 | 3234 | |
22e2c507 | 3235 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3236 | } |
3237 | ||
165125e1 | 3238 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3239 | { |
3240 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3241 | struct task_struct *tsk = current; | |
c5869807 | 3242 | struct cfq_io_cq *cic; |
22e2c507 JA |
3243 | struct cfq_queue *cfqq; |
3244 | ||
3245 | /* | |
3246 | * don't force setup of a queue from here, as a call to may_queue | |
3247 | * does not necessarily imply that a request actually will be queued. | |
3248 | * so just lookup a possibly existing queue, or return 'may queue' | |
3249 | * if that fails | |
3250 | */ | |
4ac845a2 | 3251 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3252 | if (!cic) |
3253 | return ELV_MQUEUE_MAY; | |
3254 | ||
b0b78f81 | 3255 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3256 | if (cfqq) { |
abede6da | 3257 | cfq_init_prio_data(cfqq, cic); |
22e2c507 | 3258 | |
89850f7e | 3259 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3260 | } |
3261 | ||
3262 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3263 | } |
3264 | ||
1da177e4 LT |
3265 | /* |
3266 | * queue lock held here | |
3267 | */ | |
bb37b94c | 3268 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3269 | { |
5e705374 | 3270 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3271 | |
5e705374 | 3272 | if (cfqq) { |
22e2c507 | 3273 | const int rw = rq_data_dir(rq); |
1da177e4 | 3274 | |
22e2c507 JA |
3275 | BUG_ON(!cfqq->allocated[rw]); |
3276 | cfqq->allocated[rw]--; | |
1da177e4 | 3277 | |
7f1dc8a2 | 3278 | /* Put down rq reference on cfqg */ |
1adaf3dd | 3279 | blkg_put(cfqg_to_blkg(RQ_CFQG(rq))); |
a612fddf TH |
3280 | rq->elv.priv[0] = NULL; |
3281 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 3282 | |
1da177e4 LT |
3283 | cfq_put_queue(cfqq); |
3284 | } | |
3285 | } | |
3286 | ||
df5fe3e8 | 3287 | static struct cfq_queue * |
c5869807 | 3288 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
3289 | struct cfq_queue *cfqq) |
3290 | { | |
3291 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3292 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3293 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3294 | cfq_put_queue(cfqq); |
3295 | return cic_to_cfqq(cic, 1); | |
3296 | } | |
3297 | ||
e6c5bc73 JM |
3298 | /* |
3299 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3300 | * was the last process referring to said cfqq. | |
3301 | */ | |
3302 | static struct cfq_queue * | |
c5869807 | 3303 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
3304 | { |
3305 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
3306 | cfqq->pid = current->pid; |
3307 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 3308 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
3309 | return cfqq; |
3310 | } | |
3311 | ||
3312 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
3313 | |
3314 | cfq_put_cooperator(cfqq); | |
3315 | ||
e6c5bc73 JM |
3316 | cfq_put_queue(cfqq); |
3317 | return NULL; | |
3318 | } | |
1da177e4 | 3319 | /* |
22e2c507 | 3320 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3321 | */ |
22e2c507 | 3322 | static int |
852c788f TH |
3323 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
3324 | gfp_t gfp_mask) | |
1da177e4 LT |
3325 | { |
3326 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 3327 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 3328 | const int rw = rq_data_dir(rq); |
a6151c3a | 3329 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3330 | struct cfq_queue *cfqq; |
1da177e4 LT |
3331 | |
3332 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3333 | ||
216284c3 | 3334 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 3335 | |
598971bf TH |
3336 | check_ioprio_changed(cic, bio); |
3337 | check_blkcg_changed(cic, bio); | |
e6c5bc73 | 3338 | new_queue: |
91fac317 | 3339 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3340 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
abede6da | 3341 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio, gfp_mask); |
91fac317 | 3342 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3343 | } else { |
e6c5bc73 JM |
3344 | /* |
3345 | * If the queue was seeky for too long, break it apart. | |
3346 | */ | |
ae54abed | 3347 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
3348 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
3349 | cfqq = split_cfqq(cic, cfqq); | |
3350 | if (!cfqq) | |
3351 | goto new_queue; | |
3352 | } | |
3353 | ||
df5fe3e8 JM |
3354 | /* |
3355 | * Check to see if this queue is scheduled to merge with | |
3356 | * another, closely cooperating queue. The merging of | |
3357 | * queues happens here as it must be done in process context. | |
3358 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3359 | */ | |
3360 | if (cfqq->new_cfqq) | |
3361 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3362 | } |
1da177e4 LT |
3363 | |
3364 | cfqq->allocated[rw]++; | |
1da177e4 | 3365 | |
6fae9c25 | 3366 | cfqq->ref++; |
1adaf3dd | 3367 | blkg_get(cfqg_to_blkg(cfqq->cfqg)); |
a612fddf | 3368 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 3369 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 3370 | spin_unlock_irq(q->queue_lock); |
5e705374 | 3371 | return 0; |
1da177e4 LT |
3372 | } |
3373 | ||
65f27f38 | 3374 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3375 | { |
65f27f38 | 3376 | struct cfq_data *cfqd = |
23e018a1 | 3377 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3378 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3379 | |
40bb54d1 | 3380 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 3381 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3382 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3383 | } |
3384 | ||
3385 | /* | |
3386 | * Timer running if the active_queue is currently idling inside its time slice | |
3387 | */ | |
3388 | static void cfq_idle_slice_timer(unsigned long data) | |
3389 | { | |
3390 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3391 | struct cfq_queue *cfqq; | |
3392 | unsigned long flags; | |
3c6bd2f8 | 3393 | int timed_out = 1; |
22e2c507 | 3394 | |
7b679138 JA |
3395 | cfq_log(cfqd, "idle timer fired"); |
3396 | ||
22e2c507 JA |
3397 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3398 | ||
fe094d98 JA |
3399 | cfqq = cfqd->active_queue; |
3400 | if (cfqq) { | |
3c6bd2f8 JA |
3401 | timed_out = 0; |
3402 | ||
b029195d JA |
3403 | /* |
3404 | * We saw a request before the queue expired, let it through | |
3405 | */ | |
3406 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3407 | goto out_kick; | |
3408 | ||
22e2c507 JA |
3409 | /* |
3410 | * expired | |
3411 | */ | |
44f7c160 | 3412 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
3413 | goto expire; |
3414 | ||
3415 | /* | |
3416 | * only expire and reinvoke request handler, if there are | |
3417 | * other queues with pending requests | |
3418 | */ | |
caaa5f9f | 3419 | if (!cfqd->busy_queues) |
22e2c507 | 3420 | goto out_cont; |
22e2c507 JA |
3421 | |
3422 | /* | |
3423 | * not expired and it has a request pending, let it dispatch | |
3424 | */ | |
75e50984 | 3425 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3426 | goto out_kick; |
76280aff CZ |
3427 | |
3428 | /* | |
3429 | * Queue depth flag is reset only when the idle didn't succeed | |
3430 | */ | |
3431 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
3432 | } |
3433 | expire: | |
e5ff082e | 3434 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 3435 | out_kick: |
23e018a1 | 3436 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
3437 | out_cont: |
3438 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3439 | } | |
3440 | ||
3b18152c JA |
3441 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3442 | { | |
3443 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 3444 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 3445 | } |
22e2c507 | 3446 | |
c2dea2d1 VT |
3447 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3448 | { | |
3449 | int i; | |
3450 | ||
3451 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
3452 | if (cfqd->async_cfqq[0][i]) | |
3453 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
3454 | if (cfqd->async_cfqq[1][i]) | |
3455 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 3456 | } |
2389d1ef ON |
3457 | |
3458 | if (cfqd->async_idle_cfqq) | |
3459 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
3460 | } |
3461 | ||
b374d18a | 3462 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 3463 | { |
22e2c507 | 3464 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 3465 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3466 | |
3b18152c | 3467 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 3468 | |
d9ff4187 | 3469 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3470 | |
d9ff4187 | 3471 | if (cfqd->active_queue) |
e5ff082e | 3472 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 3473 | |
c2dea2d1 | 3474 | cfq_put_async_queues(cfqd); |
03aa264a TH |
3475 | |
3476 | spin_unlock_irq(q->queue_lock); | |
3477 | ||
a90d742e AV |
3478 | cfq_shutdown_timer_wq(cfqd); |
3479 | ||
f51b802c TH |
3480 | #ifndef CONFIG_CFQ_GROUP_IOSCHED |
3481 | kfree(cfqd->root_group); | |
2abae55f | 3482 | #endif |
e8989fae | 3483 | update_root_blkg_pd(q, BLKIO_POLICY_PROP); |
56edf7d7 | 3484 | kfree(cfqd); |
1da177e4 LT |
3485 | } |
3486 | ||
b2fab5ac | 3487 | static int cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
3488 | { |
3489 | struct cfq_data *cfqd; | |
cd1604fa | 3490 | struct blkio_group *blkg __maybe_unused; |
f51b802c | 3491 | int i; |
1da177e4 | 3492 | |
94f6030c | 3493 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
a73f730d | 3494 | if (!cfqd) |
b2fab5ac | 3495 | return -ENOMEM; |
80b15c73 | 3496 | |
f51b802c TH |
3497 | cfqd->queue = q; |
3498 | q->elevator->elevator_data = cfqd; | |
3499 | ||
1fa8f6d6 VG |
3500 | /* Init root service tree */ |
3501 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
3502 | ||
f51b802c | 3503 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 3504 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
f51b802c TH |
3505 | rcu_read_lock(); |
3506 | spin_lock_irq(q->queue_lock); | |
5624a4e4 | 3507 | |
cd1604fa TH |
3508 | blkg = blkg_lookup_create(&blkio_root_cgroup, q, BLKIO_POLICY_PROP, |
3509 | true); | |
3510 | if (!IS_ERR(blkg)) | |
0381411e | 3511 | cfqd->root_group = blkg_to_cfqg(blkg); |
f51b802c TH |
3512 | |
3513 | spin_unlock_irq(q->queue_lock); | |
3514 | rcu_read_unlock(); | |
3515 | #else | |
3516 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), | |
3517 | GFP_KERNEL, cfqd->queue->node); | |
3518 | if (cfqd->root_group) | |
3519 | cfq_init_cfqg_base(cfqd->root_group); | |
3520 | #endif | |
3521 | if (!cfqd->root_group) { | |
5624a4e4 | 3522 | kfree(cfqd); |
b2fab5ac | 3523 | return -ENOMEM; |
5624a4e4 VG |
3524 | } |
3525 | ||
f51b802c | 3526 | cfqd->root_group->weight = 2*BLKIO_WEIGHT_DEFAULT; |
5624a4e4 | 3527 | |
26a2ac00 JA |
3528 | /* |
3529 | * Not strictly needed (since RB_ROOT just clears the node and we | |
3530 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
3531 | * to add magic to the rb code | |
3532 | */ | |
3533 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
3534 | cfqd->prio_trees[i] = RB_ROOT; | |
3535 | ||
6118b70b JA |
3536 | /* |
3537 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
3538 | * Grab a permanent reference to it, so that the normal code flow | |
f51b802c TH |
3539 | * will not attempt to free it. oom_cfqq is linked to root_group |
3540 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
3541 | * the reference from linking right away. | |
6118b70b JA |
3542 | */ |
3543 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 3544 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
3545 | |
3546 | spin_lock_irq(q->queue_lock); | |
f51b802c | 3547 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
1adaf3dd TH |
3548 | blkg_put(cfqg_to_blkg(cfqd->root_group)); |
3549 | spin_unlock_irq(q->queue_lock); | |
1da177e4 | 3550 | |
22e2c507 JA |
3551 | init_timer(&cfqd->idle_slice_timer); |
3552 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
3553 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
3554 | ||
23e018a1 | 3555 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 3556 | |
1da177e4 | 3557 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
3558 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
3559 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
3560 | cfqd->cfq_back_max = cfq_back_max; |
3561 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
3562 | cfqd->cfq_slice[0] = cfq_slice_async; |
3563 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
3564 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
3565 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
80bdf0c7 | 3566 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 3567 | cfqd->cfq_latency = 1; |
e459dd08 | 3568 | cfqd->hw_tag = -1; |
edc71131 CZ |
3569 | /* |
3570 | * we optimistically start assuming sync ops weren't delayed in last | |
3571 | * second, in order to have larger depth for async operations. | |
3572 | */ | |
573412b2 | 3573 | cfqd->last_delayed_sync = jiffies - HZ; |
b2fab5ac | 3574 | return 0; |
1da177e4 LT |
3575 | } |
3576 | ||
1da177e4 LT |
3577 | /* |
3578 | * sysfs parts below --> | |
3579 | */ | |
1da177e4 LT |
3580 | static ssize_t |
3581 | cfq_var_show(unsigned int var, char *page) | |
3582 | { | |
3583 | return sprintf(page, "%d\n", var); | |
3584 | } | |
3585 | ||
3586 | static ssize_t | |
3587 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
3588 | { | |
3589 | char *p = (char *) page; | |
3590 | ||
3591 | *var = simple_strtoul(p, &p, 10); | |
3592 | return count; | |
3593 | } | |
3594 | ||
1da177e4 | 3595 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 3596 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 3597 | { \ |
3d1ab40f | 3598 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3599 | unsigned int __data = __VAR; \ |
3600 | if (__CONV) \ | |
3601 | __data = jiffies_to_msecs(__data); \ | |
3602 | return cfq_var_show(__data, (page)); \ | |
3603 | } | |
3604 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
3605 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
3606 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
3607 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
3608 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 3609 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 3610 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
3611 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
3612 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
3613 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 3614 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
1da177e4 LT |
3615 | #undef SHOW_FUNCTION |
3616 | ||
3617 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 3618 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 3619 | { \ |
3d1ab40f | 3620 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3621 | unsigned int __data; \ |
3622 | int ret = cfq_var_store(&__data, (page), count); \ | |
3623 | if (__data < (MIN)) \ | |
3624 | __data = (MIN); \ | |
3625 | else if (__data > (MAX)) \ | |
3626 | __data = (MAX); \ | |
3627 | if (__CONV) \ | |
3628 | *(__PTR) = msecs_to_jiffies(__data); \ | |
3629 | else \ | |
3630 | *(__PTR) = __data; \ | |
3631 | return ret; \ | |
3632 | } | |
3633 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
3634 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
3635 | UINT_MAX, 1); | |
3636 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
3637 | UINT_MAX, 1); | |
e572ec7e | 3638 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
3639 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
3640 | UINT_MAX, 0); | |
22e2c507 | 3641 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 3642 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
3643 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
3644 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
3645 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
3646 | UINT_MAX, 0); | |
963b72fc | 3647 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
1da177e4 LT |
3648 | #undef STORE_FUNCTION |
3649 | ||
e572ec7e AV |
3650 | #define CFQ_ATTR(name) \ |
3651 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
3652 | ||
3653 | static struct elv_fs_entry cfq_attrs[] = { | |
3654 | CFQ_ATTR(quantum), | |
e572ec7e AV |
3655 | CFQ_ATTR(fifo_expire_sync), |
3656 | CFQ_ATTR(fifo_expire_async), | |
3657 | CFQ_ATTR(back_seek_max), | |
3658 | CFQ_ATTR(back_seek_penalty), | |
3659 | CFQ_ATTR(slice_sync), | |
3660 | CFQ_ATTR(slice_async), | |
3661 | CFQ_ATTR(slice_async_rq), | |
3662 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 3663 | CFQ_ATTR(group_idle), |
963b72fc | 3664 | CFQ_ATTR(low_latency), |
e572ec7e | 3665 | __ATTR_NULL |
1da177e4 LT |
3666 | }; |
3667 | ||
1da177e4 LT |
3668 | static struct elevator_type iosched_cfq = { |
3669 | .ops = { | |
3670 | .elevator_merge_fn = cfq_merge, | |
3671 | .elevator_merged_fn = cfq_merged_request, | |
3672 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 3673 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 3674 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 3675 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 3676 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 3677 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 3678 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 3679 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
3680 | .elevator_former_req_fn = elv_rb_former_request, |
3681 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 3682 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 3683 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
3684 | .elevator_set_req_fn = cfq_set_request, |
3685 | .elevator_put_req_fn = cfq_put_request, | |
3686 | .elevator_may_queue_fn = cfq_may_queue, | |
3687 | .elevator_init_fn = cfq_init_queue, | |
3688 | .elevator_exit_fn = cfq_exit_queue, | |
3689 | }, | |
3d3c2379 TH |
3690 | .icq_size = sizeof(struct cfq_io_cq), |
3691 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 3692 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 3693 | .elevator_name = "cfq", |
1da177e4 LT |
3694 | .elevator_owner = THIS_MODULE, |
3695 | }; | |
3696 | ||
3e252066 VG |
3697 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3698 | static struct blkio_policy_type blkio_policy_cfq = { | |
3699 | .ops = { | |
0381411e | 3700 | .blkio_init_group_fn = cfq_init_blkio_group, |
3e252066 VG |
3701 | .blkio_update_group_weight_fn = cfq_update_blkio_group_weight, |
3702 | }, | |
062a644d | 3703 | .plid = BLKIO_POLICY_PROP, |
0381411e | 3704 | .pdata_size = sizeof(struct cfq_group), |
3e252066 | 3705 | }; |
3e252066 VG |
3706 | #endif |
3707 | ||
1da177e4 LT |
3708 | static int __init cfq_init(void) |
3709 | { | |
3d3c2379 TH |
3710 | int ret; |
3711 | ||
22e2c507 JA |
3712 | /* |
3713 | * could be 0 on HZ < 1000 setups | |
3714 | */ | |
3715 | if (!cfq_slice_async) | |
3716 | cfq_slice_async = 1; | |
3717 | if (!cfq_slice_idle) | |
3718 | cfq_slice_idle = 1; | |
3719 | ||
80bdf0c7 VG |
3720 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3721 | if (!cfq_group_idle) | |
3722 | cfq_group_idle = 1; | |
3723 | #else | |
3724 | cfq_group_idle = 0; | |
3725 | #endif | |
3d3c2379 TH |
3726 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
3727 | if (!cfq_pool) | |
1da177e4 LT |
3728 | return -ENOMEM; |
3729 | ||
3d3c2379 TH |
3730 | ret = elv_register(&iosched_cfq); |
3731 | if (ret) { | |
3732 | kmem_cache_destroy(cfq_pool); | |
3733 | return ret; | |
3734 | } | |
3d3c2379 | 3735 | |
b95ada55 | 3736 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3e252066 | 3737 | blkio_policy_register(&blkio_policy_cfq); |
b95ada55 | 3738 | #endif |
2fdd82bd | 3739 | return 0; |
1da177e4 LT |
3740 | } |
3741 | ||
3742 | static void __exit cfq_exit(void) | |
3743 | { | |
b95ada55 | 3744 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3e252066 | 3745 | blkio_policy_unregister(&blkio_policy_cfq); |
b95ada55 | 3746 | #endif |
1da177e4 | 3747 | elv_unregister(&iosched_cfq); |
3d3c2379 | 3748 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
3749 | } |
3750 | ||
3751 | module_init(cfq_init); | |
3752 | module_exit(cfq_exit); | |
3753 | ||
3754 | MODULE_AUTHOR("Jens Axboe"); | |
3755 | MODULE_LICENSE("GPL"); | |
3756 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |