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 | ||
25fb5169 VG |
309 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
310 | ||
cdb16e8f VG |
311 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
312 | enum wl_prio_t prio, | |
65b32a57 | 313 | enum wl_type_t type) |
c0324a02 | 314 | { |
1fa8f6d6 VG |
315 | if (!cfqg) |
316 | return NULL; | |
317 | ||
c0324a02 | 318 | if (prio == IDLE_WORKLOAD) |
cdb16e8f | 319 | return &cfqg->service_tree_idle; |
c0324a02 | 320 | |
cdb16e8f | 321 | return &cfqg->service_trees[prio][type]; |
c0324a02 CZ |
322 | } |
323 | ||
3b18152c | 324 | enum cfqq_state_flags { |
b0b8d749 JA |
325 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
326 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 327 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 328 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
329 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
330 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
331 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 332 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 333 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 334 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 335 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 336 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 337 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
338 | }; |
339 | ||
340 | #define CFQ_CFQQ_FNS(name) \ | |
341 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
342 | { \ | |
fe094d98 | 343 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
344 | } \ |
345 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
346 | { \ | |
fe094d98 | 347 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
348 | } \ |
349 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
350 | { \ | |
fe094d98 | 351 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
352 | } |
353 | ||
354 | CFQ_CFQQ_FNS(on_rr); | |
355 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 356 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 357 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
358 | CFQ_CFQQ_FNS(fifo_expire); |
359 | CFQ_CFQQ_FNS(idle_window); | |
360 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 361 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 362 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 363 | CFQ_CFQQ_FNS(coop); |
ae54abed | 364 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 365 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 366 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
367 | #undef CFQ_CFQQ_FNS |
368 | ||
afc24d49 | 369 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
eb7d8c07 TH |
370 | static inline struct cfq_group *blkg_to_cfqg(struct blkio_group *blkg) |
371 | { | |
372 | return blkg_to_pdata(blkg, &blkio_policy_cfq); | |
373 | } | |
374 | ||
375 | static inline struct blkio_group *cfqg_to_blkg(struct cfq_group *cfqg) | |
376 | { | |
aaec55a0 | 377 | return pdata_to_blkg(cfqg); |
eb7d8c07 TH |
378 | } |
379 | ||
380 | static inline void cfqg_get(struct cfq_group *cfqg) | |
381 | { | |
382 | return blkg_get(cfqg_to_blkg(cfqg)); | |
383 | } | |
384 | ||
385 | static inline void cfqg_put(struct cfq_group *cfqg) | |
386 | { | |
387 | return blkg_put(cfqg_to_blkg(cfqg)); | |
388 | } | |
389 | ||
2868ef7b VG |
390 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
391 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ | |
392 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
0381411e | 393 | blkg_path(cfqg_to_blkg((cfqq)->cfqg)), ##args) |
2868ef7b VG |
394 | |
395 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ | |
396 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ | |
0381411e | 397 | blkg_path(cfqg_to_blkg((cfqg))), ##args) \ |
2868ef7b | 398 | |
eb7d8c07 TH |
399 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
400 | ||
401 | static inline struct cfq_group *blkg_to_cfqg(struct blkio_group *blkg) { return NULL; } | |
402 | static inline struct blkio_group *cfqg_to_blkg(struct cfq_group *cfqg) { return NULL; } | |
403 | static inline void cfqg_get(struct cfq_group *cfqg) { } | |
404 | static inline void cfqg_put(struct cfq_group *cfqg) { } | |
405 | ||
7b679138 JA |
406 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
407 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
4495a7d4 | 408 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
eb7d8c07 TH |
409 | |
410 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
411 | ||
7b679138 JA |
412 | #define cfq_log(cfqd, fmt, args...) \ |
413 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
414 | ||
615f0259 VG |
415 | /* Traverses through cfq group service trees */ |
416 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
417 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
418 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
419 | : &cfqg->service_tree_idle; \ | |
420 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
421 | (i == IDLE_WORKLOAD && j == 0); \ | |
422 | j++, st = i < IDLE_WORKLOAD ? \ | |
423 | &cfqg->service_trees[i][j]: NULL) \ | |
424 | ||
f5f2b6ce SL |
425 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
426 | struct cfq_ttime *ttime, bool group_idle) | |
427 | { | |
428 | unsigned long slice; | |
429 | if (!sample_valid(ttime->ttime_samples)) | |
430 | return false; | |
431 | if (group_idle) | |
432 | slice = cfqd->cfq_group_idle; | |
433 | else | |
434 | slice = cfqd->cfq_slice_idle; | |
435 | return ttime->ttime_mean > slice; | |
436 | } | |
615f0259 | 437 | |
02b35081 VG |
438 | static inline bool iops_mode(struct cfq_data *cfqd) |
439 | { | |
440 | /* | |
441 | * If we are not idling on queues and it is a NCQ drive, parallel | |
442 | * execution of requests is on and measuring time is not possible | |
443 | * in most of the cases until and unless we drive shallower queue | |
444 | * depths and that becomes a performance bottleneck. In such cases | |
445 | * switch to start providing fairness in terms of number of IOs. | |
446 | */ | |
447 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
448 | return true; | |
449 | else | |
450 | return false; | |
451 | } | |
452 | ||
c0324a02 CZ |
453 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
454 | { | |
455 | if (cfq_class_idle(cfqq)) | |
456 | return IDLE_WORKLOAD; | |
457 | if (cfq_class_rt(cfqq)) | |
458 | return RT_WORKLOAD; | |
459 | return BE_WORKLOAD; | |
460 | } | |
461 | ||
718eee05 CZ |
462 | |
463 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
464 | { | |
465 | if (!cfq_cfqq_sync(cfqq)) | |
466 | return ASYNC_WORKLOAD; | |
467 | if (!cfq_cfqq_idle_window(cfqq)) | |
468 | return SYNC_NOIDLE_WORKLOAD; | |
469 | return SYNC_WORKLOAD; | |
470 | } | |
471 | ||
58ff82f3 VG |
472 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
473 | struct cfq_data *cfqd, | |
474 | struct cfq_group *cfqg) | |
c0324a02 CZ |
475 | { |
476 | if (wl == IDLE_WORKLOAD) | |
cdb16e8f | 477 | return cfqg->service_tree_idle.count; |
c0324a02 | 478 | |
cdb16e8f VG |
479 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
480 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count | |
481 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
482 | } |
483 | ||
f26bd1f0 VG |
484 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
485 | struct cfq_group *cfqg) | |
486 | { | |
487 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count | |
488 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
489 | } | |
490 | ||
165125e1 | 491 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
4f85cb96 | 492 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
abede6da | 493 | struct cfq_io_cq *cic, struct bio *bio, |
4f85cb96 | 494 | gfp_t gfp_mask); |
91fac317 | 495 | |
c5869807 TH |
496 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
497 | { | |
498 | /* cic->icq is the first member, %NULL will convert to %NULL */ | |
499 | return container_of(icq, struct cfq_io_cq, icq); | |
500 | } | |
501 | ||
47fdd4ca TH |
502 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
503 | struct io_context *ioc) | |
504 | { | |
505 | if (ioc) | |
506 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); | |
507 | return NULL; | |
508 | } | |
509 | ||
c5869807 | 510 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
91fac317 | 511 | { |
a6151c3a | 512 | return cic->cfqq[is_sync]; |
91fac317 VT |
513 | } |
514 | ||
c5869807 TH |
515 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
516 | bool is_sync) | |
91fac317 | 517 | { |
a6151c3a | 518 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
519 | } |
520 | ||
c5869807 | 521 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
bca4b914 | 522 | { |
c5869807 | 523 | return cic->icq.q->elevator->elevator_data; |
bca4b914 KK |
524 | } |
525 | ||
91fac317 VT |
526 | /* |
527 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
528 | * set (in which case it could also be direct WRITE). | |
529 | */ | |
a6151c3a | 530 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 531 | { |
7b6d91da | 532 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 533 | } |
1da177e4 | 534 | |
99f95e52 AM |
535 | /* |
536 | * scheduler run of queue, if there are requests pending and no one in the | |
537 | * driver that will restart queueing | |
538 | */ | |
23e018a1 | 539 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 540 | { |
7b679138 JA |
541 | if (cfqd->busy_queues) { |
542 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 543 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 544 | } |
99f95e52 AM |
545 | } |
546 | ||
44f7c160 JA |
547 | /* |
548 | * Scale schedule slice based on io priority. Use the sync time slice only | |
549 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
550 | * io only, should not get full sync slice length. | |
551 | */ | |
a6151c3a | 552 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 553 | unsigned short prio) |
44f7c160 | 554 | { |
d9e7620e | 555 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 556 | |
d9e7620e JA |
557 | WARN_ON(prio >= IOPRIO_BE_NR); |
558 | ||
559 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
560 | } | |
44f7c160 | 561 | |
d9e7620e JA |
562 | static inline int |
563 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
564 | { | |
565 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
566 | } |
567 | ||
25bc6b07 VG |
568 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
569 | { | |
570 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
571 | ||
572 | d = d * BLKIO_WEIGHT_DEFAULT; | |
573 | do_div(d, cfqg->weight); | |
574 | return d; | |
575 | } | |
576 | ||
577 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
578 | { | |
579 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
580 | if (delta > 0) | |
581 | min_vdisktime = vdisktime; | |
582 | ||
583 | return min_vdisktime; | |
584 | } | |
585 | ||
586 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
587 | { | |
588 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
589 | if (delta < 0) | |
590 | min_vdisktime = vdisktime; | |
591 | ||
592 | return min_vdisktime; | |
593 | } | |
594 | ||
595 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
596 | { | |
25bc6b07 VG |
597 | struct cfq_group *cfqg; |
598 | ||
25bc6b07 VG |
599 | if (st->left) { |
600 | cfqg = rb_entry_cfqg(st->left); | |
a6032710 GJ |
601 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
602 | cfqg->vdisktime); | |
25bc6b07 | 603 | } |
25bc6b07 VG |
604 | } |
605 | ||
5db5d642 CZ |
606 | /* |
607 | * get averaged number of queues of RT/BE priority. | |
608 | * average is updated, with a formula that gives more weight to higher numbers, | |
609 | * to quickly follows sudden increases and decrease slowly | |
610 | */ | |
611 | ||
58ff82f3 VG |
612 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
613 | struct cfq_group *cfqg, bool rt) | |
5869619c | 614 | { |
5db5d642 CZ |
615 | unsigned min_q, max_q; |
616 | unsigned mult = cfq_hist_divisor - 1; | |
617 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 618 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 619 | |
58ff82f3 VG |
620 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
621 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
622 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 623 | cfq_hist_divisor; |
58ff82f3 VG |
624 | return cfqg->busy_queues_avg[rt]; |
625 | } | |
626 | ||
627 | static inline unsigned | |
628 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
629 | { | |
630 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
631 | ||
632 | return cfq_target_latency * cfqg->weight / st->total_weight; | |
5db5d642 CZ |
633 | } |
634 | ||
c553f8e3 | 635 | static inline unsigned |
ba5bd520 | 636 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
44f7c160 | 637 | { |
5db5d642 CZ |
638 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
639 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
640 | /* |
641 | * interested queues (we consider only the ones with the same | |
642 | * priority class in the cfq group) | |
643 | */ | |
644 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
645 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
646 | unsigned sync_slice = cfqd->cfq_slice[1]; |
647 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
648 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
649 | ||
650 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
651 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
652 | /* scale low_slice according to IO priority | |
653 | * and sync vs async */ | |
654 | unsigned low_slice = | |
655 | min(slice, base_low_slice * slice / sync_slice); | |
656 | /* the adapted slice value is scaled to fit all iqs | |
657 | * into the target latency */ | |
58ff82f3 | 658 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
659 | low_slice); |
660 | } | |
661 | } | |
c553f8e3 SL |
662 | return slice; |
663 | } | |
664 | ||
665 | static inline void | |
666 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
667 | { | |
ba5bd520 | 668 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 | 669 | |
dae739eb | 670 | cfqq->slice_start = jiffies; |
5db5d642 | 671 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 672 | cfqq->allocated_slice = slice; |
7b679138 | 673 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
674 | } |
675 | ||
676 | /* | |
677 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
678 | * isn't valid until the first request from the dispatch is activated | |
679 | * and the slice time set. | |
680 | */ | |
a6151c3a | 681 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
682 | { |
683 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 684 | return false; |
44f7c160 | 685 | if (time_before(jiffies, cfqq->slice_end)) |
c1e44756 | 686 | return false; |
44f7c160 | 687 | |
c1e44756 | 688 | return true; |
44f7c160 JA |
689 | } |
690 | ||
1da177e4 | 691 | /* |
5e705374 | 692 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 693 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 694 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 695 | */ |
5e705374 | 696 | static struct request * |
cf7c25cf | 697 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 698 | { |
cf7c25cf | 699 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 700 | unsigned long back_max; |
e8a99053 AM |
701 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
702 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
703 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 704 | |
5e705374 JA |
705 | if (rq1 == NULL || rq1 == rq2) |
706 | return rq2; | |
707 | if (rq2 == NULL) | |
708 | return rq1; | |
9c2c38a1 | 709 | |
229836bd NK |
710 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
711 | return rq_is_sync(rq1) ? rq1 : rq2; | |
712 | ||
65299a3b CH |
713 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
714 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; | |
b53d1ed7 | 715 | |
83096ebf TH |
716 | s1 = blk_rq_pos(rq1); |
717 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 718 | |
1da177e4 LT |
719 | /* |
720 | * by definition, 1KiB is 2 sectors | |
721 | */ | |
722 | back_max = cfqd->cfq_back_max * 2; | |
723 | ||
724 | /* | |
725 | * Strict one way elevator _except_ in the case where we allow | |
726 | * short backward seeks which are biased as twice the cost of a | |
727 | * similar forward seek. | |
728 | */ | |
729 | if (s1 >= last) | |
730 | d1 = s1 - last; | |
731 | else if (s1 + back_max >= last) | |
732 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
733 | else | |
e8a99053 | 734 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
735 | |
736 | if (s2 >= last) | |
737 | d2 = s2 - last; | |
738 | else if (s2 + back_max >= last) | |
739 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
740 | else | |
e8a99053 | 741 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
742 | |
743 | /* Found required data */ | |
e8a99053 AM |
744 | |
745 | /* | |
746 | * By doing switch() on the bit mask "wrap" we avoid having to | |
747 | * check two variables for all permutations: --> faster! | |
748 | */ | |
749 | switch (wrap) { | |
5e705374 | 750 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 751 | if (d1 < d2) |
5e705374 | 752 | return rq1; |
e8a99053 | 753 | else if (d2 < d1) |
5e705374 | 754 | return rq2; |
e8a99053 AM |
755 | else { |
756 | if (s1 >= s2) | |
5e705374 | 757 | return rq1; |
e8a99053 | 758 | else |
5e705374 | 759 | return rq2; |
e8a99053 | 760 | } |
1da177e4 | 761 | |
e8a99053 | 762 | case CFQ_RQ2_WRAP: |
5e705374 | 763 | return rq1; |
e8a99053 | 764 | case CFQ_RQ1_WRAP: |
5e705374 JA |
765 | return rq2; |
766 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
767 | default: |
768 | /* | |
769 | * Since both rqs are wrapped, | |
770 | * start with the one that's further behind head | |
771 | * (--> only *one* back seek required), | |
772 | * since back seek takes more time than forward. | |
773 | */ | |
774 | if (s1 <= s2) | |
5e705374 | 775 | return rq1; |
1da177e4 | 776 | else |
5e705374 | 777 | return rq2; |
1da177e4 LT |
778 | } |
779 | } | |
780 | ||
498d3aa2 JA |
781 | /* |
782 | * The below is leftmost cache rbtree addon | |
783 | */ | |
0871714e | 784 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 785 | { |
615f0259 VG |
786 | /* Service tree is empty */ |
787 | if (!root->count) | |
788 | return NULL; | |
789 | ||
cc09e299 JA |
790 | if (!root->left) |
791 | root->left = rb_first(&root->rb); | |
792 | ||
0871714e JA |
793 | if (root->left) |
794 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
795 | ||
796 | return NULL; | |
cc09e299 JA |
797 | } |
798 | ||
1fa8f6d6 VG |
799 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
800 | { | |
801 | if (!root->left) | |
802 | root->left = rb_first(&root->rb); | |
803 | ||
804 | if (root->left) | |
805 | return rb_entry_cfqg(root->left); | |
806 | ||
807 | return NULL; | |
808 | } | |
809 | ||
a36e71f9 JA |
810 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
811 | { | |
812 | rb_erase(n, root); | |
813 | RB_CLEAR_NODE(n); | |
814 | } | |
815 | ||
cc09e299 JA |
816 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
817 | { | |
818 | if (root->left == n) | |
819 | root->left = NULL; | |
a36e71f9 | 820 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 821 | --root->count; |
cc09e299 JA |
822 | } |
823 | ||
1da177e4 LT |
824 | /* |
825 | * would be nice to take fifo expire time into account as well | |
826 | */ | |
5e705374 JA |
827 | static struct request * |
828 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
829 | struct request *last) | |
1da177e4 | 830 | { |
21183b07 JA |
831 | struct rb_node *rbnext = rb_next(&last->rb_node); |
832 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 833 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 834 | |
21183b07 | 835 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
836 | |
837 | if (rbprev) | |
5e705374 | 838 | prev = rb_entry_rq(rbprev); |
1da177e4 | 839 | |
21183b07 | 840 | if (rbnext) |
5e705374 | 841 | next = rb_entry_rq(rbnext); |
21183b07 JA |
842 | else { |
843 | rbnext = rb_first(&cfqq->sort_list); | |
844 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 845 | next = rb_entry_rq(rbnext); |
21183b07 | 846 | } |
1da177e4 | 847 | |
cf7c25cf | 848 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
849 | } |
850 | ||
d9e7620e JA |
851 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
852 | struct cfq_queue *cfqq) | |
1da177e4 | 853 | { |
d9e7620e JA |
854 | /* |
855 | * just an approximation, should be ok. | |
856 | */ | |
cdb16e8f | 857 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 858 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
859 | } |
860 | ||
1fa8f6d6 VG |
861 | static inline s64 |
862 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
863 | { | |
864 | return cfqg->vdisktime - st->min_vdisktime; | |
865 | } | |
866 | ||
867 | static void | |
868 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
869 | { | |
870 | struct rb_node **node = &st->rb.rb_node; | |
871 | struct rb_node *parent = NULL; | |
872 | struct cfq_group *__cfqg; | |
873 | s64 key = cfqg_key(st, cfqg); | |
874 | int left = 1; | |
875 | ||
876 | while (*node != NULL) { | |
877 | parent = *node; | |
878 | __cfqg = rb_entry_cfqg(parent); | |
879 | ||
880 | if (key < cfqg_key(st, __cfqg)) | |
881 | node = &parent->rb_left; | |
882 | else { | |
883 | node = &parent->rb_right; | |
884 | left = 0; | |
885 | } | |
886 | } | |
887 | ||
888 | if (left) | |
889 | st->left = &cfqg->rb_node; | |
890 | ||
891 | rb_link_node(&cfqg->rb_node, parent, node); | |
892 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
893 | } | |
894 | ||
895 | static void | |
8184f93e JT |
896 | cfq_update_group_weight(struct cfq_group *cfqg) |
897 | { | |
898 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
899 | if (cfqg->needs_update) { | |
900 | cfqg->weight = cfqg->new_weight; | |
901 | cfqg->needs_update = false; | |
902 | } | |
903 | } | |
904 | ||
905 | static void | |
906 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
907 | { | |
908 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); | |
909 | ||
910 | cfq_update_group_weight(cfqg); | |
911 | __cfq_group_service_tree_add(st, cfqg); | |
912 | st->total_weight += cfqg->weight; | |
913 | } | |
914 | ||
915 | static void | |
916 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1fa8f6d6 VG |
917 | { |
918 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
919 | struct cfq_group *__cfqg; | |
920 | struct rb_node *n; | |
921 | ||
922 | cfqg->nr_cfqq++; | |
760701bf | 923 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
1fa8f6d6 VG |
924 | return; |
925 | ||
926 | /* | |
927 | * Currently put the group at the end. Later implement something | |
928 | * so that groups get lesser vtime based on their weights, so that | |
25985edc | 929 | * if group does not loose all if it was not continuously backlogged. |
1fa8f6d6 VG |
930 | */ |
931 | n = rb_last(&st->rb); | |
932 | if (n) { | |
933 | __cfqg = rb_entry_cfqg(n); | |
934 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
935 | } else | |
936 | cfqg->vdisktime = st->min_vdisktime; | |
8184f93e JT |
937 | cfq_group_service_tree_add(st, cfqg); |
938 | } | |
1fa8f6d6 | 939 | |
8184f93e JT |
940 | static void |
941 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
942 | { | |
943 | st->total_weight -= cfqg->weight; | |
944 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) | |
945 | cfq_rb_erase(&cfqg->rb_node, st); | |
1fa8f6d6 VG |
946 | } |
947 | ||
948 | static void | |
8184f93e | 949 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
1fa8f6d6 VG |
950 | { |
951 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
952 | ||
953 | BUG_ON(cfqg->nr_cfqq < 1); | |
954 | cfqg->nr_cfqq--; | |
25bc6b07 | 955 | |
1fa8f6d6 VG |
956 | /* If there are other cfq queues under this group, don't delete it */ |
957 | if (cfqg->nr_cfqq) | |
958 | return; | |
959 | ||
2868ef7b | 960 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
8184f93e | 961 | cfq_group_service_tree_del(st, cfqg); |
dae739eb | 962 | cfqg->saved_workload_slice = 0; |
c1768268 TH |
963 | cfq_blkiocg_update_dequeue_stats(cfqg_to_blkg(cfqg), |
964 | &blkio_policy_cfq, 1); | |
dae739eb VG |
965 | } |
966 | ||
167400d3 JT |
967 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
968 | unsigned int *unaccounted_time) | |
dae739eb | 969 | { |
f75edf2d | 970 | unsigned int slice_used; |
dae739eb VG |
971 | |
972 | /* | |
973 | * Queue got expired before even a single request completed or | |
974 | * got expired immediately after first request completion. | |
975 | */ | |
976 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
977 | /* | |
978 | * Also charge the seek time incurred to the group, otherwise | |
979 | * if there are mutiple queues in the group, each can dispatch | |
980 | * a single request on seeky media and cause lots of seek time | |
981 | * and group will never know it. | |
982 | */ | |
983 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
984 | 1); | |
985 | } else { | |
986 | slice_used = jiffies - cfqq->slice_start; | |
167400d3 JT |
987 | if (slice_used > cfqq->allocated_slice) { |
988 | *unaccounted_time = slice_used - cfqq->allocated_slice; | |
f75edf2d | 989 | slice_used = cfqq->allocated_slice; |
167400d3 JT |
990 | } |
991 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) | |
992 | *unaccounted_time += cfqq->slice_start - | |
993 | cfqq->dispatch_start; | |
dae739eb VG |
994 | } |
995 | ||
dae739eb VG |
996 | return slice_used; |
997 | } | |
998 | ||
999 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 1000 | struct cfq_queue *cfqq) |
dae739eb VG |
1001 | { |
1002 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
167400d3 | 1003 | unsigned int used_sl, charge, unaccounted_sl = 0; |
f26bd1f0 VG |
1004 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
1005 | - cfqg->service_tree_idle.count; | |
1006 | ||
1007 | BUG_ON(nr_sync < 0); | |
167400d3 | 1008 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
dae739eb | 1009 | |
02b35081 VG |
1010 | if (iops_mode(cfqd)) |
1011 | charge = cfqq->slice_dispatch; | |
1012 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
1013 | charge = cfqq->allocated_slice; | |
dae739eb VG |
1014 | |
1015 | /* Can't update vdisktime while group is on service tree */ | |
8184f93e | 1016 | cfq_group_service_tree_del(st, cfqg); |
02b35081 | 1017 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
8184f93e JT |
1018 | /* If a new weight was requested, update now, off tree */ |
1019 | cfq_group_service_tree_add(st, cfqg); | |
dae739eb VG |
1020 | |
1021 | /* This group is being expired. Save the context */ | |
1022 | if (time_after(cfqd->workload_expires, jiffies)) { | |
1023 | cfqg->saved_workload_slice = cfqd->workload_expires | |
1024 | - jiffies; | |
1025 | cfqg->saved_workload = cfqd->serving_type; | |
1026 | cfqg->saved_serving_prio = cfqd->serving_prio; | |
1027 | } else | |
1028 | cfqg->saved_workload_slice = 0; | |
2868ef7b VG |
1029 | |
1030 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
1031 | st->min_vdisktime); | |
fd16d263 JP |
1032 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
1033 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", | |
1034 | used_sl, cfqq->slice_dispatch, charge, | |
1035 | iops_mode(cfqd), cfqq->nr_sectors); | |
c1768268 TH |
1036 | cfq_blkiocg_update_timeslice_used(cfqg_to_blkg(cfqg), &blkio_policy_cfq, |
1037 | used_sl, unaccounted_sl); | |
1038 | cfq_blkiocg_set_start_empty_time(cfqg_to_blkg(cfqg), &blkio_policy_cfq); | |
1fa8f6d6 VG |
1039 | } |
1040 | ||
f51b802c TH |
1041 | /** |
1042 | * cfq_init_cfqg_base - initialize base part of a cfq_group | |
1043 | * @cfqg: cfq_group to initialize | |
1044 | * | |
1045 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED | |
1046 | * is enabled or not. | |
1047 | */ | |
1048 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) | |
1049 | { | |
1050 | struct cfq_rb_root *st; | |
1051 | int i, j; | |
1052 | ||
1053 | for_each_cfqg_st(cfqg, i, j, st) | |
1054 | *st = CFQ_RB_ROOT; | |
1055 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1056 | ||
1057 | cfqg->ttime.last_end_request = jiffies; | |
1058 | } | |
1059 | ||
25fb5169 | 1060 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
ca32aefc TH |
1061 | static void cfq_update_blkio_group_weight(struct request_queue *q, |
1062 | struct blkio_group *blkg, | |
8aea4545 | 1063 | unsigned int weight) |
f8d461d6 | 1064 | { |
0381411e TH |
1065 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
1066 | ||
8184f93e JT |
1067 | cfqg->new_weight = weight; |
1068 | cfqg->needs_update = true; | |
f8d461d6 VG |
1069 | } |
1070 | ||
0381411e | 1071 | static void cfq_init_blkio_group(struct blkio_group *blkg) |
f469a7b4 | 1072 | { |
0381411e | 1073 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
25fb5169 | 1074 | |
f51b802c | 1075 | cfq_init_cfqg_base(cfqg); |
0381411e | 1076 | cfqg->weight = blkg->blkcg->weight; |
25fb5169 VG |
1077 | } |
1078 | ||
1079 | /* | |
3e59cf9d VG |
1080 | * Search for the cfq group current task belongs to. request_queue lock must |
1081 | * be held. | |
25fb5169 | 1082 | */ |
cd1604fa TH |
1083 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
1084 | struct blkio_cgroup *blkcg) | |
25fb5169 | 1085 | { |
f469a7b4 | 1086 | struct request_queue *q = cfqd->queue; |
cd1604fa | 1087 | struct cfq_group *cfqg = NULL; |
25fb5169 | 1088 | |
cd1604fa TH |
1089 | /* avoid lookup for the common case where there's no blkio cgroup */ |
1090 | if (blkcg == &blkio_root_cgroup) { | |
1091 | cfqg = cfqd->root_group; | |
1092 | } else { | |
1093 | struct blkio_group *blkg; | |
f469a7b4 | 1094 | |
aaec55a0 | 1095 | blkg = blkg_lookup_create(blkcg, q, false); |
cd1604fa | 1096 | if (!IS_ERR(blkg)) |
0381411e | 1097 | cfqg = blkg_to_cfqg(blkg); |
cd1604fa | 1098 | } |
f469a7b4 | 1099 | |
25fb5169 VG |
1100 | return cfqg; |
1101 | } | |
1102 | ||
1103 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1104 | { | |
1105 | /* Currently, all async queues are mapped to root group */ | |
1106 | if (!cfq_cfqq_sync(cfqq)) | |
f51b802c | 1107 | cfqg = cfqq->cfqd->root_group; |
25fb5169 VG |
1108 | |
1109 | cfqq->cfqg = cfqg; | |
b1c35769 | 1110 | /* cfqq reference on cfqg */ |
eb7d8c07 | 1111 | cfqg_get(cfqg); |
b1c35769 VG |
1112 | } |
1113 | ||
25fb5169 | 1114 | #else /* GROUP_IOSCHED */ |
cd1604fa TH |
1115 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
1116 | struct blkio_cgroup *blkcg) | |
25fb5169 | 1117 | { |
f51b802c | 1118 | return cfqd->root_group; |
25fb5169 | 1119 | } |
7f1dc8a2 | 1120 | |
25fb5169 VG |
1121 | static inline void |
1122 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1123 | cfqq->cfqg = cfqg; | |
1124 | } | |
1125 | ||
1126 | #endif /* GROUP_IOSCHED */ | |
1127 | ||
498d3aa2 | 1128 | /* |
c0324a02 | 1129 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1130 | * requests waiting to be processed. It is sorted in the order that |
1131 | * we will service the queues. | |
1132 | */ | |
a36e71f9 | 1133 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1134 | bool add_front) |
d9e7620e | 1135 | { |
0871714e JA |
1136 | struct rb_node **p, *parent; |
1137 | struct cfq_queue *__cfqq; | |
d9e7620e | 1138 | unsigned long rb_key; |
c0324a02 | 1139 | struct cfq_rb_root *service_tree; |
498d3aa2 | 1140 | int left; |
dae739eb | 1141 | int new_cfqq = 1; |
ae30c286 | 1142 | |
cdb16e8f | 1143 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
65b32a57 | 1144 | cfqq_type(cfqq)); |
0871714e JA |
1145 | if (cfq_class_idle(cfqq)) { |
1146 | rb_key = CFQ_IDLE_DELAY; | |
aa6f6a3d | 1147 | parent = rb_last(&service_tree->rb); |
0871714e JA |
1148 | if (parent && parent != &cfqq->rb_node) { |
1149 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1150 | rb_key += __cfqq->rb_key; | |
1151 | } else | |
1152 | rb_key += jiffies; | |
1153 | } else if (!add_front) { | |
b9c8946b JA |
1154 | /* |
1155 | * Get our rb key offset. Subtract any residual slice | |
1156 | * value carried from last service. A negative resid | |
1157 | * count indicates slice overrun, and this should position | |
1158 | * the next service time further away in the tree. | |
1159 | */ | |
edd75ffd | 1160 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1161 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1162 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1163 | } else { |
1164 | rb_key = -HZ; | |
aa6f6a3d | 1165 | __cfqq = cfq_rb_first(service_tree); |
48e025e6 CZ |
1166 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1167 | } | |
1da177e4 | 1168 | |
d9e7620e | 1169 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1170 | new_cfqq = 0; |
99f9628a | 1171 | /* |
d9e7620e | 1172 | * same position, nothing more to do |
99f9628a | 1173 | */ |
c0324a02 CZ |
1174 | if (rb_key == cfqq->rb_key && |
1175 | cfqq->service_tree == service_tree) | |
d9e7620e | 1176 | return; |
1da177e4 | 1177 | |
aa6f6a3d CZ |
1178 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1179 | cfqq->service_tree = NULL; | |
1da177e4 | 1180 | } |
d9e7620e | 1181 | |
498d3aa2 | 1182 | left = 1; |
0871714e | 1183 | parent = NULL; |
aa6f6a3d CZ |
1184 | cfqq->service_tree = service_tree; |
1185 | p = &service_tree->rb.rb_node; | |
d9e7620e | 1186 | while (*p) { |
67060e37 | 1187 | struct rb_node **n; |
cc09e299 | 1188 | |
d9e7620e JA |
1189 | parent = *p; |
1190 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1191 | ||
0c534e0a | 1192 | /* |
c0324a02 | 1193 | * sort by key, that represents service time. |
0c534e0a | 1194 | */ |
c0324a02 | 1195 | if (time_before(rb_key, __cfqq->rb_key)) |
67060e37 | 1196 | n = &(*p)->rb_left; |
c0324a02 | 1197 | else { |
67060e37 | 1198 | n = &(*p)->rb_right; |
cc09e299 | 1199 | left = 0; |
c0324a02 | 1200 | } |
67060e37 JA |
1201 | |
1202 | p = n; | |
d9e7620e JA |
1203 | } |
1204 | ||
cc09e299 | 1205 | if (left) |
aa6f6a3d | 1206 | service_tree->left = &cfqq->rb_node; |
cc09e299 | 1207 | |
d9e7620e JA |
1208 | cfqq->rb_key = rb_key; |
1209 | rb_link_node(&cfqq->rb_node, parent, p); | |
aa6f6a3d CZ |
1210 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
1211 | service_tree->count++; | |
20359f27 | 1212 | if (add_front || !new_cfqq) |
dae739eb | 1213 | return; |
8184f93e | 1214 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1215 | } |
1216 | ||
a36e71f9 | 1217 | static struct cfq_queue * |
f2d1f0ae JA |
1218 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1219 | sector_t sector, struct rb_node **ret_parent, | |
1220 | struct rb_node ***rb_link) | |
a36e71f9 | 1221 | { |
a36e71f9 JA |
1222 | struct rb_node **p, *parent; |
1223 | struct cfq_queue *cfqq = NULL; | |
1224 | ||
1225 | parent = NULL; | |
1226 | p = &root->rb_node; | |
1227 | while (*p) { | |
1228 | struct rb_node **n; | |
1229 | ||
1230 | parent = *p; | |
1231 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1232 | ||
1233 | /* | |
1234 | * Sort strictly based on sector. Smallest to the left, | |
1235 | * largest to the right. | |
1236 | */ | |
2e46e8b2 | 1237 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1238 | n = &(*p)->rb_right; |
2e46e8b2 | 1239 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1240 | n = &(*p)->rb_left; |
1241 | else | |
1242 | break; | |
1243 | p = n; | |
3ac6c9f8 | 1244 | cfqq = NULL; |
a36e71f9 JA |
1245 | } |
1246 | ||
1247 | *ret_parent = parent; | |
1248 | if (rb_link) | |
1249 | *rb_link = p; | |
3ac6c9f8 | 1250 | return cfqq; |
a36e71f9 JA |
1251 | } |
1252 | ||
1253 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1254 | { | |
a36e71f9 JA |
1255 | struct rb_node **p, *parent; |
1256 | struct cfq_queue *__cfqq; | |
1257 | ||
f2d1f0ae JA |
1258 | if (cfqq->p_root) { |
1259 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1260 | cfqq->p_root = NULL; | |
1261 | } | |
a36e71f9 JA |
1262 | |
1263 | if (cfq_class_idle(cfqq)) | |
1264 | return; | |
1265 | if (!cfqq->next_rq) | |
1266 | return; | |
1267 | ||
f2d1f0ae | 1268 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1269 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1270 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1271 | if (!__cfqq) { |
1272 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1273 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1274 | } else | |
1275 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1276 | } |
1277 | ||
498d3aa2 JA |
1278 | /* |
1279 | * Update cfqq's position in the service tree. | |
1280 | */ | |
edd75ffd | 1281 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1282 | { |
6d048f53 JA |
1283 | /* |
1284 | * Resorting requires the cfqq to be on the RR list already. | |
1285 | */ | |
a36e71f9 | 1286 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1287 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1288 | cfq_prio_tree_add(cfqd, cfqq); |
1289 | } | |
6d048f53 JA |
1290 | } |
1291 | ||
1da177e4 LT |
1292 | /* |
1293 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1294 | * the pending list according to last request service |
1da177e4 | 1295 | */ |
febffd61 | 1296 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1297 | { |
7b679138 | 1298 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1299 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1300 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 | 1301 | cfqd->busy_queues++; |
ef8a41df SL |
1302 | if (cfq_cfqq_sync(cfqq)) |
1303 | cfqd->busy_sync_queues++; | |
1da177e4 | 1304 | |
edd75ffd | 1305 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1306 | } |
1307 | ||
498d3aa2 JA |
1308 | /* |
1309 | * Called when the cfqq no longer has requests pending, remove it from | |
1310 | * the service tree. | |
1311 | */ | |
febffd61 | 1312 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1313 | { |
7b679138 | 1314 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1315 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1316 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1317 | |
aa6f6a3d CZ |
1318 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1319 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1320 | cfqq->service_tree = NULL; | |
1321 | } | |
f2d1f0ae JA |
1322 | if (cfqq->p_root) { |
1323 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1324 | cfqq->p_root = NULL; | |
1325 | } | |
d9e7620e | 1326 | |
8184f93e | 1327 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1328 | BUG_ON(!cfqd->busy_queues); |
1329 | cfqd->busy_queues--; | |
ef8a41df SL |
1330 | if (cfq_cfqq_sync(cfqq)) |
1331 | cfqd->busy_sync_queues--; | |
1da177e4 LT |
1332 | } |
1333 | ||
1334 | /* | |
1335 | * rb tree support functions | |
1336 | */ | |
febffd61 | 1337 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1338 | { |
5e705374 | 1339 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1340 | const int sync = rq_is_sync(rq); |
1da177e4 | 1341 | |
b4878f24 JA |
1342 | BUG_ON(!cfqq->queued[sync]); |
1343 | cfqq->queued[sync]--; | |
1da177e4 | 1344 | |
5e705374 | 1345 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1346 | |
f04a6424 VG |
1347 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1348 | /* | |
1349 | * Queue will be deleted from service tree when we actually | |
1350 | * expire it later. Right now just remove it from prio tree | |
1351 | * as it is empty. | |
1352 | */ | |
1353 | if (cfqq->p_root) { | |
1354 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1355 | cfqq->p_root = NULL; | |
1356 | } | |
1357 | } | |
1da177e4 LT |
1358 | } |
1359 | ||
5e705374 | 1360 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1361 | { |
5e705374 | 1362 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1363 | struct cfq_data *cfqd = cfqq->cfqd; |
796d5116 | 1364 | struct request *prev; |
1da177e4 | 1365 | |
5380a101 | 1366 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 | 1367 | |
796d5116 | 1368 | elv_rb_add(&cfqq->sort_list, rq); |
5fccbf61 JA |
1369 | |
1370 | if (!cfq_cfqq_on_rr(cfqq)) | |
1371 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1372 | |
1373 | /* | |
1374 | * check if this request is a better next-serve candidate | |
1375 | */ | |
a36e71f9 | 1376 | prev = cfqq->next_rq; |
cf7c25cf | 1377 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1378 | |
1379 | /* | |
1380 | * adjust priority tree position, if ->next_rq changes | |
1381 | */ | |
1382 | if (prev != cfqq->next_rq) | |
1383 | cfq_prio_tree_add(cfqd, cfqq); | |
1384 | ||
5044eed4 | 1385 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1386 | } |
1387 | ||
febffd61 | 1388 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1389 | { |
5380a101 JA |
1390 | elv_rb_del(&cfqq->sort_list, rq); |
1391 | cfqq->queued[rq_is_sync(rq)]--; | |
0381411e | 1392 | cfq_blkiocg_update_io_remove_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 TH |
1393 | &blkio_policy_cfq, rq_data_dir(rq), |
1394 | rq_is_sync(rq)); | |
5e705374 | 1395 | cfq_add_rq_rb(rq); |
0381411e | 1396 | cfq_blkiocg_update_io_add_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 | 1397 | &blkio_policy_cfq, |
0381411e TH |
1398 | cfqg_to_blkg(cfqq->cfqd->serving_group), |
1399 | rq_data_dir(rq), rq_is_sync(rq)); | |
1da177e4 LT |
1400 | } |
1401 | ||
206dc69b JA |
1402 | static struct request * |
1403 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1404 | { |
206dc69b | 1405 | struct task_struct *tsk = current; |
c5869807 | 1406 | struct cfq_io_cq *cic; |
206dc69b | 1407 | struct cfq_queue *cfqq; |
1da177e4 | 1408 | |
4ac845a2 | 1409 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1410 | if (!cic) |
1411 | return NULL; | |
1412 | ||
1413 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
1414 | if (cfqq) { |
1415 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
1416 | ||
21183b07 | 1417 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 1418 | } |
1da177e4 | 1419 | |
1da177e4 LT |
1420 | return NULL; |
1421 | } | |
1422 | ||
165125e1 | 1423 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1424 | { |
22e2c507 | 1425 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 1426 | |
53c583d2 | 1427 | cfqd->rq_in_driver++; |
7b679138 | 1428 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 1429 | cfqd->rq_in_driver); |
25776e35 | 1430 | |
5b93629b | 1431 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
1432 | } |
1433 | ||
165125e1 | 1434 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1435 | { |
b4878f24 JA |
1436 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1437 | ||
53c583d2 CZ |
1438 | WARN_ON(!cfqd->rq_in_driver); |
1439 | cfqd->rq_in_driver--; | |
7b679138 | 1440 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 1441 | cfqd->rq_in_driver); |
1da177e4 LT |
1442 | } |
1443 | ||
b4878f24 | 1444 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 1445 | { |
5e705374 | 1446 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 1447 | |
5e705374 JA |
1448 | if (cfqq->next_rq == rq) |
1449 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 1450 | |
b4878f24 | 1451 | list_del_init(&rq->queuelist); |
5e705374 | 1452 | cfq_del_rq_rb(rq); |
374f84ac | 1453 | |
45333d5a | 1454 | cfqq->cfqd->rq_queued--; |
0381411e | 1455 | cfq_blkiocg_update_io_remove_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 TH |
1456 | &blkio_policy_cfq, rq_data_dir(rq), |
1457 | rq_is_sync(rq)); | |
65299a3b CH |
1458 | if (rq->cmd_flags & REQ_PRIO) { |
1459 | WARN_ON(!cfqq->prio_pending); | |
1460 | cfqq->prio_pending--; | |
b53d1ed7 | 1461 | } |
1da177e4 LT |
1462 | } |
1463 | ||
165125e1 JA |
1464 | static int cfq_merge(struct request_queue *q, struct request **req, |
1465 | struct bio *bio) | |
1da177e4 LT |
1466 | { |
1467 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1468 | struct request *__rq; | |
1da177e4 | 1469 | |
206dc69b | 1470 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 1471 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
1472 | *req = __rq; |
1473 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
1474 | } |
1475 | ||
1476 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
1477 | } |
1478 | ||
165125e1 | 1479 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 1480 | int type) |
1da177e4 | 1481 | { |
21183b07 | 1482 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 1483 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 1484 | |
5e705374 | 1485 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 1486 | } |
1da177e4 LT |
1487 | } |
1488 | ||
812d4026 DS |
1489 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
1490 | struct bio *bio) | |
1491 | { | |
0381411e | 1492 | cfq_blkiocg_update_io_merged_stats(cfqg_to_blkg(RQ_CFQG(req)), |
c1768268 TH |
1493 | &blkio_policy_cfq, bio_data_dir(bio), |
1494 | cfq_bio_sync(bio)); | |
812d4026 DS |
1495 | } |
1496 | ||
1da177e4 | 1497 | static void |
165125e1 | 1498 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
1499 | struct request *next) |
1500 | { | |
cf7c25cf | 1501 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
4a0b75c7 SL |
1502 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1503 | ||
22e2c507 JA |
1504 | /* |
1505 | * reposition in fifo if next is older than rq | |
1506 | */ | |
1507 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
30996f40 | 1508 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
22e2c507 | 1509 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
1510 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1511 | } | |
22e2c507 | 1512 | |
cf7c25cf CZ |
1513 | if (cfqq->next_rq == next) |
1514 | cfqq->next_rq = rq; | |
b4878f24 | 1515 | cfq_remove_request(next); |
0381411e | 1516 | cfq_blkiocg_update_io_merged_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 TH |
1517 | &blkio_policy_cfq, rq_data_dir(next), |
1518 | rq_is_sync(next)); | |
4a0b75c7 SL |
1519 | |
1520 | cfqq = RQ_CFQQ(next); | |
1521 | /* | |
1522 | * all requests of this queue are merged to other queues, delete it | |
1523 | * from the service tree. If it's the active_queue, | |
1524 | * cfq_dispatch_requests() will choose to expire it or do idle | |
1525 | */ | |
1526 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && | |
1527 | cfqq != cfqd->active_queue) | |
1528 | cfq_del_cfqq_rr(cfqd, cfqq); | |
22e2c507 JA |
1529 | } |
1530 | ||
165125e1 | 1531 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
1532 | struct bio *bio) |
1533 | { | |
1534 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
c5869807 | 1535 | struct cfq_io_cq *cic; |
da775265 | 1536 | struct cfq_queue *cfqq; |
da775265 JA |
1537 | |
1538 | /* | |
ec8acb69 | 1539 | * Disallow merge of a sync bio into an async request. |
da775265 | 1540 | */ |
91fac317 | 1541 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 1542 | return false; |
da775265 JA |
1543 | |
1544 | /* | |
f1a4f4d3 | 1545 | * Lookup the cfqq that this bio will be queued with and allow |
07c2bd37 | 1546 | * merge only if rq is queued there. |
f1a4f4d3 | 1547 | */ |
07c2bd37 TH |
1548 | cic = cfq_cic_lookup(cfqd, current->io_context); |
1549 | if (!cic) | |
1550 | return false; | |
719d3402 | 1551 | |
91fac317 | 1552 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 1553 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
1554 | } |
1555 | ||
812df48d DS |
1556 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1557 | { | |
1558 | del_timer(&cfqd->idle_slice_timer); | |
c1768268 TH |
1559 | cfq_blkiocg_update_idle_time_stats(cfqg_to_blkg(cfqq->cfqg), |
1560 | &blkio_policy_cfq); | |
812df48d DS |
1561 | } |
1562 | ||
febffd61 JA |
1563 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
1564 | struct cfq_queue *cfqq) | |
22e2c507 JA |
1565 | { |
1566 | if (cfqq) { | |
b1ffe737 DS |
1567 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", |
1568 | cfqd->serving_prio, cfqd->serving_type); | |
c1768268 TH |
1569 | cfq_blkiocg_update_avg_queue_size_stats(cfqg_to_blkg(cfqq->cfqg), |
1570 | &blkio_policy_cfq); | |
62a37f6b JT |
1571 | cfqq->slice_start = 0; |
1572 | cfqq->dispatch_start = jiffies; | |
1573 | cfqq->allocated_slice = 0; | |
1574 | cfqq->slice_end = 0; | |
1575 | cfqq->slice_dispatch = 0; | |
1576 | cfqq->nr_sectors = 0; | |
1577 | ||
1578 | cfq_clear_cfqq_wait_request(cfqq); | |
1579 | cfq_clear_cfqq_must_dispatch(cfqq); | |
1580 | cfq_clear_cfqq_must_alloc_slice(cfqq); | |
1581 | cfq_clear_cfqq_fifo_expire(cfqq); | |
1582 | cfq_mark_cfqq_slice_new(cfqq); | |
1583 | ||
1584 | cfq_del_timer(cfqd, cfqq); | |
22e2c507 JA |
1585 | } |
1586 | ||
1587 | cfqd->active_queue = cfqq; | |
1588 | } | |
1589 | ||
7b14e3b5 JA |
1590 | /* |
1591 | * current cfqq expired its slice (or was too idle), select new one | |
1592 | */ | |
1593 | static void | |
1594 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 1595 | bool timed_out) |
7b14e3b5 | 1596 | { |
7b679138 JA |
1597 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
1598 | ||
7b14e3b5 | 1599 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 1600 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 1601 | |
7b14e3b5 | 1602 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 1603 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 1604 | |
ae54abed SL |
1605 | /* |
1606 | * If this cfqq is shared between multiple processes, check to | |
1607 | * make sure that those processes are still issuing I/Os within | |
1608 | * the mean seek distance. If not, it may be time to break the | |
1609 | * queues apart again. | |
1610 | */ | |
1611 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
1612 | cfq_mark_cfqq_split_coop(cfqq); | |
1613 | ||
7b14e3b5 | 1614 | /* |
6084cdda | 1615 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 1616 | */ |
c553f8e3 SL |
1617 | if (timed_out) { |
1618 | if (cfq_cfqq_slice_new(cfqq)) | |
ba5bd520 | 1619 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
c553f8e3 SL |
1620 | else |
1621 | cfqq->slice_resid = cfqq->slice_end - jiffies; | |
7b679138 JA |
1622 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
1623 | } | |
7b14e3b5 | 1624 | |
e5ff082e | 1625 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 1626 | |
f04a6424 VG |
1627 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1628 | cfq_del_cfqq_rr(cfqd, cfqq); | |
1629 | ||
edd75ffd | 1630 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
1631 | |
1632 | if (cfqq == cfqd->active_queue) | |
1633 | cfqd->active_queue = NULL; | |
1634 | ||
1635 | if (cfqd->active_cic) { | |
11a3122f | 1636 | put_io_context(cfqd->active_cic->icq.ioc); |
7b14e3b5 JA |
1637 | cfqd->active_cic = NULL; |
1638 | } | |
7b14e3b5 JA |
1639 | } |
1640 | ||
e5ff082e | 1641 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
1642 | { |
1643 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1644 | ||
1645 | if (cfqq) | |
e5ff082e | 1646 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
1647 | } |
1648 | ||
498d3aa2 JA |
1649 | /* |
1650 | * Get next queue for service. Unless we have a queue preemption, | |
1651 | * we'll simply select the first cfqq in the service tree. | |
1652 | */ | |
6d048f53 | 1653 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 1654 | { |
c0324a02 | 1655 | struct cfq_rb_root *service_tree = |
cdb16e8f | 1656 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
65b32a57 | 1657 | cfqd->serving_type); |
d9e7620e | 1658 | |
f04a6424 VG |
1659 | if (!cfqd->rq_queued) |
1660 | return NULL; | |
1661 | ||
1fa8f6d6 VG |
1662 | /* There is nothing to dispatch */ |
1663 | if (!service_tree) | |
1664 | return NULL; | |
c0324a02 CZ |
1665 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
1666 | return NULL; | |
1667 | return cfq_rb_first(service_tree); | |
6d048f53 JA |
1668 | } |
1669 | ||
f04a6424 VG |
1670 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
1671 | { | |
25fb5169 | 1672 | struct cfq_group *cfqg; |
f04a6424 VG |
1673 | struct cfq_queue *cfqq; |
1674 | int i, j; | |
1675 | struct cfq_rb_root *st; | |
1676 | ||
1677 | if (!cfqd->rq_queued) | |
1678 | return NULL; | |
1679 | ||
25fb5169 VG |
1680 | cfqg = cfq_get_next_cfqg(cfqd); |
1681 | if (!cfqg) | |
1682 | return NULL; | |
1683 | ||
f04a6424 VG |
1684 | for_each_cfqg_st(cfqg, i, j, st) |
1685 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
1686 | return cfqq; | |
1687 | return NULL; | |
1688 | } | |
1689 | ||
498d3aa2 JA |
1690 | /* |
1691 | * Get and set a new active queue for service. | |
1692 | */ | |
a36e71f9 JA |
1693 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
1694 | struct cfq_queue *cfqq) | |
6d048f53 | 1695 | { |
e00ef799 | 1696 | if (!cfqq) |
a36e71f9 | 1697 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 1698 | |
22e2c507 | 1699 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 1700 | return cfqq; |
22e2c507 JA |
1701 | } |
1702 | ||
d9e7620e JA |
1703 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
1704 | struct request *rq) | |
1705 | { | |
83096ebf TH |
1706 | if (blk_rq_pos(rq) >= cfqd->last_position) |
1707 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 1708 | else |
83096ebf | 1709 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
1710 | } |
1711 | ||
b2c18e1e | 1712 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 1713 | struct request *rq) |
6d048f53 | 1714 | { |
e9ce335d | 1715 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
1716 | } |
1717 | ||
a36e71f9 JA |
1718 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
1719 | struct cfq_queue *cur_cfqq) | |
1720 | { | |
f2d1f0ae | 1721 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
1722 | struct rb_node *parent, *node; |
1723 | struct cfq_queue *__cfqq; | |
1724 | sector_t sector = cfqd->last_position; | |
1725 | ||
1726 | if (RB_EMPTY_ROOT(root)) | |
1727 | return NULL; | |
1728 | ||
1729 | /* | |
1730 | * First, if we find a request starting at the end of the last | |
1731 | * request, choose it. | |
1732 | */ | |
f2d1f0ae | 1733 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
1734 | if (__cfqq) |
1735 | return __cfqq; | |
1736 | ||
1737 | /* | |
1738 | * If the exact sector wasn't found, the parent of the NULL leaf | |
1739 | * will contain the closest sector. | |
1740 | */ | |
1741 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 1742 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1743 | return __cfqq; |
1744 | ||
2e46e8b2 | 1745 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
1746 | node = rb_next(&__cfqq->p_node); |
1747 | else | |
1748 | node = rb_prev(&__cfqq->p_node); | |
1749 | if (!node) | |
1750 | return NULL; | |
1751 | ||
1752 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 1753 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1754 | return __cfqq; |
1755 | ||
1756 | return NULL; | |
1757 | } | |
1758 | ||
1759 | /* | |
1760 | * cfqd - obvious | |
1761 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
1762 | * closely cooperating with itself. | |
1763 | * | |
1764 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
1765 | * one request, and that cfqd->last_position reflects a position on the disk | |
1766 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
1767 | * assumption. | |
1768 | */ | |
1769 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 1770 | struct cfq_queue *cur_cfqq) |
6d048f53 | 1771 | { |
a36e71f9 JA |
1772 | struct cfq_queue *cfqq; |
1773 | ||
39c01b21 DS |
1774 | if (cfq_class_idle(cur_cfqq)) |
1775 | return NULL; | |
e6c5bc73 JM |
1776 | if (!cfq_cfqq_sync(cur_cfqq)) |
1777 | return NULL; | |
1778 | if (CFQQ_SEEKY(cur_cfqq)) | |
1779 | return NULL; | |
1780 | ||
b9d8f4c7 GJ |
1781 | /* |
1782 | * Don't search priority tree if it's the only queue in the group. | |
1783 | */ | |
1784 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
1785 | return NULL; | |
1786 | ||
6d048f53 | 1787 | /* |
d9e7620e JA |
1788 | * We should notice if some of the queues are cooperating, eg |
1789 | * working closely on the same area of the disk. In that case, | |
1790 | * we can group them together and don't waste time idling. | |
6d048f53 | 1791 | */ |
a36e71f9 JA |
1792 | cfqq = cfqq_close(cfqd, cur_cfqq); |
1793 | if (!cfqq) | |
1794 | return NULL; | |
1795 | ||
8682e1f1 VG |
1796 | /* If new queue belongs to different cfq_group, don't choose it */ |
1797 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
1798 | return NULL; | |
1799 | ||
df5fe3e8 JM |
1800 | /* |
1801 | * It only makes sense to merge sync queues. | |
1802 | */ | |
1803 | if (!cfq_cfqq_sync(cfqq)) | |
1804 | return NULL; | |
e6c5bc73 JM |
1805 | if (CFQQ_SEEKY(cfqq)) |
1806 | return NULL; | |
df5fe3e8 | 1807 | |
c0324a02 CZ |
1808 | /* |
1809 | * Do not merge queues of different priority classes | |
1810 | */ | |
1811 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
1812 | return NULL; | |
1813 | ||
a36e71f9 | 1814 | return cfqq; |
6d048f53 JA |
1815 | } |
1816 | ||
a6d44e98 CZ |
1817 | /* |
1818 | * Determine whether we should enforce idle window for this queue. | |
1819 | */ | |
1820 | ||
1821 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1822 | { | |
1823 | enum wl_prio_t prio = cfqq_prio(cfqq); | |
718eee05 | 1824 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
a6d44e98 | 1825 | |
f04a6424 VG |
1826 | BUG_ON(!service_tree); |
1827 | BUG_ON(!service_tree->count); | |
1828 | ||
b6508c16 VG |
1829 | if (!cfqd->cfq_slice_idle) |
1830 | return false; | |
1831 | ||
a6d44e98 CZ |
1832 | /* We never do for idle class queues. */ |
1833 | if (prio == IDLE_WORKLOAD) | |
1834 | return false; | |
1835 | ||
1836 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
1837 | if (cfq_cfqq_idle_window(cfqq) && |
1838 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
1839 | return true; |
1840 | ||
1841 | /* | |
1842 | * Otherwise, we do only if they are the last ones | |
1843 | * in their service tree. | |
1844 | */ | |
f5f2b6ce SL |
1845 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq) && |
1846 | !cfq_io_thinktime_big(cfqd, &service_tree->ttime, false)) | |
c1e44756 | 1847 | return true; |
b1ffe737 DS |
1848 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", |
1849 | service_tree->count); | |
c1e44756 | 1850 | return false; |
a6d44e98 CZ |
1851 | } |
1852 | ||
6d048f53 | 1853 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 1854 | { |
1792669c | 1855 | struct cfq_queue *cfqq = cfqd->active_queue; |
c5869807 | 1856 | struct cfq_io_cq *cic; |
80bdf0c7 | 1857 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 1858 | |
a68bbddb | 1859 | /* |
f7d7b7a7 JA |
1860 | * SSD device without seek penalty, disable idling. But only do so |
1861 | * for devices that support queuing, otherwise we still have a problem | |
1862 | * with sync vs async workloads. | |
a68bbddb | 1863 | */ |
f7d7b7a7 | 1864 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
1865 | return; |
1866 | ||
dd67d051 | 1867 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 1868 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
1869 | |
1870 | /* | |
1871 | * idle is disabled, either manually or by past process history | |
1872 | */ | |
80bdf0c7 VG |
1873 | if (!cfq_should_idle(cfqd, cfqq)) { |
1874 | /* no queue idling. Check for group idling */ | |
1875 | if (cfqd->cfq_group_idle) | |
1876 | group_idle = cfqd->cfq_group_idle; | |
1877 | else | |
1878 | return; | |
1879 | } | |
6d048f53 | 1880 | |
7b679138 | 1881 | /* |
8e550632 | 1882 | * still active requests from this queue, don't idle |
7b679138 | 1883 | */ |
8e550632 | 1884 | if (cfqq->dispatched) |
7b679138 JA |
1885 | return; |
1886 | ||
22e2c507 JA |
1887 | /* |
1888 | * task has exited, don't wait | |
1889 | */ | |
206dc69b | 1890 | cic = cfqd->active_cic; |
f6e8d01b | 1891 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
6d048f53 JA |
1892 | return; |
1893 | ||
355b659c CZ |
1894 | /* |
1895 | * If our average think time is larger than the remaining time | |
1896 | * slice, then don't idle. This avoids overrunning the allotted | |
1897 | * time slice. | |
1898 | */ | |
383cd721 SL |
1899 | if (sample_valid(cic->ttime.ttime_samples) && |
1900 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { | |
fd16d263 | 1901 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
383cd721 | 1902 | cic->ttime.ttime_mean); |
355b659c | 1903 | return; |
b1ffe737 | 1904 | } |
355b659c | 1905 | |
80bdf0c7 VG |
1906 | /* There are other queues in the group, don't do group idle */ |
1907 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
1908 | return; | |
1909 | ||
3b18152c | 1910 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 1911 | |
80bdf0c7 VG |
1912 | if (group_idle) |
1913 | sl = cfqd->cfq_group_idle; | |
1914 | else | |
1915 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 1916 | |
7b14e3b5 | 1917 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
c1768268 TH |
1918 | cfq_blkiocg_update_set_idle_time_stats(cfqg_to_blkg(cfqq->cfqg), |
1919 | &blkio_policy_cfq); | |
80bdf0c7 VG |
1920 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
1921 | group_idle ? 1 : 0); | |
1da177e4 LT |
1922 | } |
1923 | ||
498d3aa2 JA |
1924 | /* |
1925 | * Move request from internal lists to the request queue dispatch list. | |
1926 | */ | |
165125e1 | 1927 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 1928 | { |
3ed9a296 | 1929 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1930 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1931 | |
7b679138 JA |
1932 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
1933 | ||
06d21886 | 1934 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 1935 | cfq_remove_request(rq); |
6d048f53 | 1936 | cfqq->dispatched++; |
80bdf0c7 | 1937 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 1938 | elv_dispatch_sort(q, rq); |
3ed9a296 | 1939 | |
53c583d2 | 1940 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 1941 | cfqq->nr_sectors += blk_rq_sectors(rq); |
0381411e | 1942 | cfq_blkiocg_update_dispatch_stats(cfqg_to_blkg(cfqq->cfqg), |
c1768268 TH |
1943 | &blkio_policy_cfq, blk_rq_bytes(rq), |
1944 | rq_data_dir(rq), rq_is_sync(rq)); | |
1da177e4 LT |
1945 | } |
1946 | ||
1947 | /* | |
1948 | * return expired entry, or NULL to just start from scratch in rbtree | |
1949 | */ | |
febffd61 | 1950 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 1951 | { |
30996f40 | 1952 | struct request *rq = NULL; |
1da177e4 | 1953 | |
3b18152c | 1954 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 1955 | return NULL; |
cb887411 JA |
1956 | |
1957 | cfq_mark_cfqq_fifo_expire(cfqq); | |
1958 | ||
89850f7e JA |
1959 | if (list_empty(&cfqq->fifo)) |
1960 | return NULL; | |
1da177e4 | 1961 | |
89850f7e | 1962 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 1963 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 1964 | rq = NULL; |
1da177e4 | 1965 | |
30996f40 | 1966 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 1967 | return rq; |
1da177e4 LT |
1968 | } |
1969 | ||
22e2c507 JA |
1970 | static inline int |
1971 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1972 | { | |
1973 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 1974 | |
22e2c507 | 1975 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 1976 | |
b9f8ce05 | 1977 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
1da177e4 LT |
1978 | } |
1979 | ||
df5fe3e8 JM |
1980 | /* |
1981 | * Must be called with the queue_lock held. | |
1982 | */ | |
1983 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
1984 | { | |
1985 | int process_refs, io_refs; | |
1986 | ||
1987 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
30d7b944 | 1988 | process_refs = cfqq->ref - io_refs; |
df5fe3e8 JM |
1989 | BUG_ON(process_refs < 0); |
1990 | return process_refs; | |
1991 | } | |
1992 | ||
1993 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
1994 | { | |
e6c5bc73 | 1995 | int process_refs, new_process_refs; |
df5fe3e8 JM |
1996 | struct cfq_queue *__cfqq; |
1997 | ||
c10b61f0 JM |
1998 | /* |
1999 | * If there are no process references on the new_cfqq, then it is | |
2000 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2001 | * chain may have dropped their last reference (not just their | |
2002 | * last process reference). | |
2003 | */ | |
2004 | if (!cfqq_process_refs(new_cfqq)) | |
2005 | return; | |
2006 | ||
df5fe3e8 JM |
2007 | /* Avoid a circular list and skip interim queue merges */ |
2008 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2009 | if (__cfqq == cfqq) | |
2010 | return; | |
2011 | new_cfqq = __cfqq; | |
2012 | } | |
2013 | ||
2014 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2015 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2016 | /* |
2017 | * If the process for the cfqq has gone away, there is no | |
2018 | * sense in merging the queues. | |
2019 | */ | |
c10b61f0 | 2020 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2021 | return; |
2022 | ||
e6c5bc73 JM |
2023 | /* |
2024 | * Merge in the direction of the lesser amount of work. | |
2025 | */ | |
e6c5bc73 JM |
2026 | if (new_process_refs >= process_refs) { |
2027 | cfqq->new_cfqq = new_cfqq; | |
30d7b944 | 2028 | new_cfqq->ref += process_refs; |
e6c5bc73 JM |
2029 | } else { |
2030 | new_cfqq->new_cfqq = cfqq; | |
30d7b944 | 2031 | cfqq->ref += new_process_refs; |
e6c5bc73 | 2032 | } |
df5fe3e8 JM |
2033 | } |
2034 | ||
cdb16e8f | 2035 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
65b32a57 | 2036 | struct cfq_group *cfqg, enum wl_prio_t prio) |
718eee05 CZ |
2037 | { |
2038 | struct cfq_queue *queue; | |
2039 | int i; | |
2040 | bool key_valid = false; | |
2041 | unsigned long lowest_key = 0; | |
2042 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2043 | ||
65b32a57 VG |
2044 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2045 | /* select the one with lowest rb_key */ | |
2046 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); | |
718eee05 CZ |
2047 | if (queue && |
2048 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2049 | lowest_key = queue->rb_key; | |
2050 | cur_best = i; | |
2051 | key_valid = true; | |
2052 | } | |
2053 | } | |
2054 | ||
2055 | return cur_best; | |
2056 | } | |
2057 | ||
cdb16e8f | 2058 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
718eee05 | 2059 | { |
718eee05 CZ |
2060 | unsigned slice; |
2061 | unsigned count; | |
cdb16e8f | 2062 | struct cfq_rb_root *st; |
58ff82f3 | 2063 | unsigned group_slice; |
e4ea0c16 | 2064 | enum wl_prio_t original_prio = cfqd->serving_prio; |
1fa8f6d6 | 2065 | |
718eee05 | 2066 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2067 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
718eee05 | 2068 | cfqd->serving_prio = RT_WORKLOAD; |
58ff82f3 | 2069 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
718eee05 CZ |
2070 | cfqd->serving_prio = BE_WORKLOAD; |
2071 | else { | |
2072 | cfqd->serving_prio = IDLE_WORKLOAD; | |
2073 | cfqd->workload_expires = jiffies + 1; | |
2074 | return; | |
2075 | } | |
2076 | ||
e4ea0c16 SL |
2077 | if (original_prio != cfqd->serving_prio) |
2078 | goto new_workload; | |
2079 | ||
718eee05 CZ |
2080 | /* |
2081 | * For RT and BE, we have to choose also the type | |
2082 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2083 | * expiration time | |
2084 | */ | |
65b32a57 | 2085 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
cdb16e8f | 2086 | count = st->count; |
718eee05 CZ |
2087 | |
2088 | /* | |
65b32a57 | 2089 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2090 | */ |
65b32a57 | 2091 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2092 | return; |
2093 | ||
e4ea0c16 | 2094 | new_workload: |
718eee05 CZ |
2095 | /* otherwise select new workload type */ |
2096 | cfqd->serving_type = | |
65b32a57 VG |
2097 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); |
2098 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); | |
cdb16e8f | 2099 | count = st->count; |
718eee05 CZ |
2100 | |
2101 | /* | |
2102 | * the workload slice is computed as a fraction of target latency | |
2103 | * proportional to the number of queues in that workload, over | |
2104 | * all the queues in the same priority class | |
2105 | */ | |
58ff82f3 VG |
2106 | group_slice = cfq_group_slice(cfqd, cfqg); |
2107 | ||
2108 | slice = group_slice * count / | |
2109 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], | |
2110 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); | |
718eee05 | 2111 | |
f26bd1f0 VG |
2112 | if (cfqd->serving_type == ASYNC_WORKLOAD) { |
2113 | unsigned int tmp; | |
2114 | ||
2115 | /* | |
2116 | * Async queues are currently system wide. Just taking | |
2117 | * proportion of queues with-in same group will lead to higher | |
2118 | * async ratio system wide as generally root group is going | |
2119 | * to have higher weight. A more accurate thing would be to | |
2120 | * calculate system wide asnc/sync ratio. | |
2121 | */ | |
2122 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); | |
2123 | tmp = tmp/cfqd->busy_queues; | |
2124 | slice = min_t(unsigned, slice, tmp); | |
2125 | ||
718eee05 CZ |
2126 | /* async workload slice is scaled down according to |
2127 | * the sync/async slice ratio. */ | |
2128 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2129 | } else |
718eee05 CZ |
2130 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2131 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2132 | ||
2133 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2134 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 CZ |
2135 | cfqd->workload_expires = jiffies + slice; |
2136 | } | |
2137 | ||
1fa8f6d6 VG |
2138 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2139 | { | |
2140 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2141 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2142 | |
2143 | if (RB_EMPTY_ROOT(&st->rb)) | |
2144 | return NULL; | |
25bc6b07 | 2145 | cfqg = cfq_rb_first_group(st); |
25bc6b07 VG |
2146 | update_min_vdisktime(st); |
2147 | return cfqg; | |
1fa8f6d6 VG |
2148 | } |
2149 | ||
cdb16e8f VG |
2150 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2151 | { | |
1fa8f6d6 VG |
2152 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2153 | ||
2154 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2155 | |
2156 | /* Restore the workload type data */ | |
2157 | if (cfqg->saved_workload_slice) { | |
2158 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; | |
2159 | cfqd->serving_type = cfqg->saved_workload; | |
2160 | cfqd->serving_prio = cfqg->saved_serving_prio; | |
66ae2919 GJ |
2161 | } else |
2162 | cfqd->workload_expires = jiffies - 1; | |
2163 | ||
1fa8f6d6 | 2164 | choose_service_tree(cfqd, cfqg); |
cdb16e8f VG |
2165 | } |
2166 | ||
22e2c507 | 2167 | /* |
498d3aa2 JA |
2168 | * Select a queue for service. If we have a current active queue, |
2169 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2170 | */ |
1b5ed5e1 | 2171 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2172 | { |
a36e71f9 | 2173 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2174 | |
22e2c507 JA |
2175 | cfqq = cfqd->active_queue; |
2176 | if (!cfqq) | |
2177 | goto new_queue; | |
1da177e4 | 2178 | |
f04a6424 VG |
2179 | if (!cfqd->rq_queued) |
2180 | return NULL; | |
c244bb50 VG |
2181 | |
2182 | /* | |
2183 | * We were waiting for group to get backlogged. Expire the queue | |
2184 | */ | |
2185 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2186 | goto expire; | |
2187 | ||
22e2c507 | 2188 | /* |
6d048f53 | 2189 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2190 | */ |
7667aa06 VG |
2191 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2192 | /* | |
2193 | * If slice had not expired at the completion of last request | |
2194 | * we might not have turned on wait_busy flag. Don't expire | |
2195 | * the queue yet. Allow the group to get backlogged. | |
2196 | * | |
2197 | * The very fact that we have used the slice, that means we | |
2198 | * have been idling all along on this queue and it should be | |
2199 | * ok to wait for this request to complete. | |
2200 | */ | |
82bbbf28 VG |
2201 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2202 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2203 | cfqq = NULL; | |
7667aa06 | 2204 | goto keep_queue; |
82bbbf28 | 2205 | } else |
80bdf0c7 | 2206 | goto check_group_idle; |
7667aa06 | 2207 | } |
1da177e4 | 2208 | |
22e2c507 | 2209 | /* |
6d048f53 JA |
2210 | * The active queue has requests and isn't expired, allow it to |
2211 | * dispatch. | |
22e2c507 | 2212 | */ |
dd67d051 | 2213 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2214 | goto keep_queue; |
6d048f53 | 2215 | |
a36e71f9 JA |
2216 | /* |
2217 | * If another queue has a request waiting within our mean seek | |
2218 | * distance, let it run. The expire code will check for close | |
2219 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2220 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2221 | */ |
b3b6d040 | 2222 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2223 | if (new_cfqq) { |
2224 | if (!cfqq->new_cfqq) | |
2225 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2226 | goto expire; |
df5fe3e8 | 2227 | } |
a36e71f9 | 2228 | |
6d048f53 JA |
2229 | /* |
2230 | * No requests pending. If the active queue still has requests in | |
2231 | * flight or is idling for a new request, allow either of these | |
2232 | * conditions to happen (or time out) before selecting a new queue. | |
2233 | */ | |
80bdf0c7 VG |
2234 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2235 | cfqq = NULL; | |
2236 | goto keep_queue; | |
2237 | } | |
2238 | ||
8e1ac665 SL |
2239 | /* |
2240 | * This is a deep seek queue, but the device is much faster than | |
2241 | * the queue can deliver, don't idle | |
2242 | **/ | |
2243 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && | |
2244 | (cfq_cfqq_slice_new(cfqq) || | |
2245 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { | |
2246 | cfq_clear_cfqq_deep(cfqq); | |
2247 | cfq_clear_cfqq_idle_window(cfqq); | |
2248 | } | |
2249 | ||
80bdf0c7 VG |
2250 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
2251 | cfqq = NULL; | |
2252 | goto keep_queue; | |
2253 | } | |
2254 | ||
2255 | /* | |
2256 | * If group idle is enabled and there are requests dispatched from | |
2257 | * this group, wait for requests to complete. | |
2258 | */ | |
2259 | check_group_idle: | |
7700fc4f SL |
2260 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
2261 | cfqq->cfqg->dispatched && | |
2262 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { | |
caaa5f9f JA |
2263 | cfqq = NULL; |
2264 | goto keep_queue; | |
22e2c507 JA |
2265 | } |
2266 | ||
3b18152c | 2267 | expire: |
e5ff082e | 2268 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2269 | new_queue: |
718eee05 CZ |
2270 | /* |
2271 | * Current queue expired. Check if we have to switch to a new | |
2272 | * service tree | |
2273 | */ | |
2274 | if (!new_cfqq) | |
cdb16e8f | 2275 | cfq_choose_cfqg(cfqd); |
718eee05 | 2276 | |
a36e71f9 | 2277 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2278 | keep_queue: |
3b18152c | 2279 | return cfqq; |
22e2c507 JA |
2280 | } |
2281 | ||
febffd61 | 2282 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2283 | { |
2284 | int dispatched = 0; | |
2285 | ||
2286 | while (cfqq->next_rq) { | |
2287 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2288 | dispatched++; | |
2289 | } | |
2290 | ||
2291 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2292 | |
2293 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 2294 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
2295 | return dispatched; |
2296 | } | |
2297 | ||
498d3aa2 JA |
2298 | /* |
2299 | * Drain our current requests. Used for barriers and when switching | |
2300 | * io schedulers on-the-fly. | |
2301 | */ | |
d9e7620e | 2302 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2303 | { |
0871714e | 2304 | struct cfq_queue *cfqq; |
d9e7620e | 2305 | int dispatched = 0; |
cdb16e8f | 2306 | |
3440c49f | 2307 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 2308 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
2309 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2310 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 2311 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 2312 | } |
1b5ed5e1 | 2313 | |
1b5ed5e1 TH |
2314 | BUG_ON(cfqd->busy_queues); |
2315 | ||
6923715a | 2316 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2317 | return dispatched; |
2318 | } | |
2319 | ||
abc3c744 SL |
2320 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2321 | struct cfq_queue *cfqq) | |
2322 | { | |
2323 | /* the queue hasn't finished any request, can't estimate */ | |
2324 | if (cfq_cfqq_slice_new(cfqq)) | |
c1e44756 | 2325 | return true; |
abc3c744 SL |
2326 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
2327 | cfqq->slice_end)) | |
c1e44756 | 2328 | return true; |
abc3c744 | 2329 | |
c1e44756 | 2330 | return false; |
abc3c744 SL |
2331 | } |
2332 | ||
0b182d61 | 2333 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2334 | { |
2f5cb738 | 2335 | unsigned int max_dispatch; |
22e2c507 | 2336 | |
5ad531db JA |
2337 | /* |
2338 | * Drain async requests before we start sync IO | |
2339 | */ | |
53c583d2 | 2340 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 2341 | return false; |
5ad531db | 2342 | |
2f5cb738 JA |
2343 | /* |
2344 | * If this is an async queue and we have sync IO in flight, let it wait | |
2345 | */ | |
53c583d2 | 2346 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 2347 | return false; |
2f5cb738 | 2348 | |
abc3c744 | 2349 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
2350 | if (cfq_class_idle(cfqq)) |
2351 | max_dispatch = 1; | |
b4878f24 | 2352 | |
2f5cb738 JA |
2353 | /* |
2354 | * Does this cfqq already have too much IO in flight? | |
2355 | */ | |
2356 | if (cfqq->dispatched >= max_dispatch) { | |
ef8a41df | 2357 | bool promote_sync = false; |
2f5cb738 JA |
2358 | /* |
2359 | * idle queue must always only have a single IO in flight | |
2360 | */ | |
3ed9a296 | 2361 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2362 | return false; |
3ed9a296 | 2363 | |
ef8a41df | 2364 | /* |
c4ade94f LS |
2365 | * If there is only one sync queue |
2366 | * we can ignore async queue here and give the sync | |
ef8a41df SL |
2367 | * queue no dispatch limit. The reason is a sync queue can |
2368 | * preempt async queue, limiting the sync queue doesn't make | |
2369 | * sense. This is useful for aiostress test. | |
2370 | */ | |
c4ade94f LS |
2371 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
2372 | promote_sync = true; | |
ef8a41df | 2373 | |
2f5cb738 JA |
2374 | /* |
2375 | * We have other queues, don't allow more IO from this one | |
2376 | */ | |
ef8a41df SL |
2377 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
2378 | !promote_sync) | |
0b182d61 | 2379 | return false; |
9ede209e | 2380 | |
365722bb | 2381 | /* |
474b18cc | 2382 | * Sole queue user, no limit |
365722bb | 2383 | */ |
ef8a41df | 2384 | if (cfqd->busy_queues == 1 || promote_sync) |
abc3c744 SL |
2385 | max_dispatch = -1; |
2386 | else | |
2387 | /* | |
2388 | * Normally we start throttling cfqq when cfq_quantum/2 | |
2389 | * requests have been dispatched. But we can drive | |
2390 | * deeper queue depths at the beginning of slice | |
2391 | * subjected to upper limit of cfq_quantum. | |
2392 | * */ | |
2393 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
2394 | } |
2395 | ||
2396 | /* | |
2397 | * Async queues must wait a bit before being allowed dispatch. | |
2398 | * We also ramp up the dispatch depth gradually for async IO, | |
2399 | * based on the last sync IO we serviced | |
2400 | */ | |
963b72fc | 2401 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 2402 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 2403 | unsigned int depth; |
365722bb | 2404 | |
61f0c1dc | 2405 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2406 | if (!depth && !cfqq->dispatched) |
2407 | depth = 1; | |
8e296755 JA |
2408 | if (depth < max_dispatch) |
2409 | max_dispatch = depth; | |
2f5cb738 | 2410 | } |
3ed9a296 | 2411 | |
0b182d61 JA |
2412 | /* |
2413 | * If we're below the current max, allow a dispatch | |
2414 | */ | |
2415 | return cfqq->dispatched < max_dispatch; | |
2416 | } | |
2417 | ||
2418 | /* | |
2419 | * Dispatch a request from cfqq, moving them to the request queue | |
2420 | * dispatch list. | |
2421 | */ | |
2422 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2423 | { | |
2424 | struct request *rq; | |
2425 | ||
2426 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
2427 | ||
2428 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
2429 | return false; | |
2430 | ||
2431 | /* | |
2432 | * follow expired path, else get first next available | |
2433 | */ | |
2434 | rq = cfq_check_fifo(cfqq); | |
2435 | if (!rq) | |
2436 | rq = cfqq->next_rq; | |
2437 | ||
2438 | /* | |
2439 | * insert request into driver dispatch list | |
2440 | */ | |
2441 | cfq_dispatch_insert(cfqd->queue, rq); | |
2442 | ||
2443 | if (!cfqd->active_cic) { | |
c5869807 | 2444 | struct cfq_io_cq *cic = RQ_CIC(rq); |
0b182d61 | 2445 | |
c5869807 | 2446 | atomic_long_inc(&cic->icq.ioc->refcount); |
0b182d61 JA |
2447 | cfqd->active_cic = cic; |
2448 | } | |
2449 | ||
2450 | return true; | |
2451 | } | |
2452 | ||
2453 | /* | |
2454 | * Find the cfqq that we need to service and move a request from that to the | |
2455 | * dispatch list | |
2456 | */ | |
2457 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
2458 | { | |
2459 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2460 | struct cfq_queue *cfqq; | |
2461 | ||
2462 | if (!cfqd->busy_queues) | |
2463 | return 0; | |
2464 | ||
2465 | if (unlikely(force)) | |
2466 | return cfq_forced_dispatch(cfqd); | |
2467 | ||
2468 | cfqq = cfq_select_queue(cfqd); | |
2469 | if (!cfqq) | |
8e296755 JA |
2470 | return 0; |
2471 | ||
2f5cb738 | 2472 | /* |
0b182d61 | 2473 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 2474 | */ |
0b182d61 JA |
2475 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2476 | return 0; | |
2477 | ||
2f5cb738 | 2478 | cfqq->slice_dispatch++; |
b029195d | 2479 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 2480 | |
2f5cb738 JA |
2481 | /* |
2482 | * expire an async queue immediately if it has used up its slice. idle | |
2483 | * queue always expire after 1 dispatch round. | |
2484 | */ | |
2485 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
2486 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
2487 | cfq_class_idle(cfqq))) { | |
2488 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 2489 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
2490 | } |
2491 | ||
b217a903 | 2492 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 2493 | return 1; |
1da177e4 LT |
2494 | } |
2495 | ||
1da177e4 | 2496 | /* |
5e705374 JA |
2497 | * task holds one reference to the queue, dropped when task exits. each rq |
2498 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 2499 | * |
b1c35769 | 2500 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
2501 | * queue lock must be held here. |
2502 | */ | |
2503 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
2504 | { | |
22e2c507 | 2505 | struct cfq_data *cfqd = cfqq->cfqd; |
0bbfeb83 | 2506 | struct cfq_group *cfqg; |
22e2c507 | 2507 | |
30d7b944 | 2508 | BUG_ON(cfqq->ref <= 0); |
1da177e4 | 2509 | |
30d7b944 SL |
2510 | cfqq->ref--; |
2511 | if (cfqq->ref) | |
1da177e4 LT |
2512 | return; |
2513 | ||
7b679138 | 2514 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 2515 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 2516 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 2517 | cfqg = cfqq->cfqg; |
1da177e4 | 2518 | |
28f95cbc | 2519 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 2520 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2521 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2522 | } |
22e2c507 | 2523 | |
f04a6424 | 2524 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 2525 | kmem_cache_free(cfq_pool, cfqq); |
eb7d8c07 | 2526 | cfqg_put(cfqg); |
1da177e4 LT |
2527 | } |
2528 | ||
d02a2c07 | 2529 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 2530 | { |
df5fe3e8 JM |
2531 | struct cfq_queue *__cfqq, *next; |
2532 | ||
df5fe3e8 JM |
2533 | /* |
2534 | * If this queue was scheduled to merge with another queue, be | |
2535 | * sure to drop the reference taken on that queue (and others in | |
2536 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
2537 | */ | |
2538 | __cfqq = cfqq->new_cfqq; | |
2539 | while (__cfqq) { | |
2540 | if (__cfqq == cfqq) { | |
2541 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
2542 | break; | |
2543 | } | |
2544 | next = __cfqq->new_cfqq; | |
2545 | cfq_put_queue(__cfqq); | |
2546 | __cfqq = next; | |
2547 | } | |
d02a2c07 SL |
2548 | } |
2549 | ||
2550 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2551 | { | |
2552 | if (unlikely(cfqq == cfqd->active_queue)) { | |
2553 | __cfq_slice_expired(cfqd, cfqq, 0); | |
2554 | cfq_schedule_dispatch(cfqd); | |
2555 | } | |
2556 | ||
2557 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 2558 | |
89850f7e JA |
2559 | cfq_put_queue(cfqq); |
2560 | } | |
22e2c507 | 2561 | |
9b84cacd TH |
2562 | static void cfq_init_icq(struct io_cq *icq) |
2563 | { | |
2564 | struct cfq_io_cq *cic = icq_to_cic(icq); | |
2565 | ||
2566 | cic->ttime.last_end_request = jiffies; | |
2567 | } | |
2568 | ||
c5869807 | 2569 | static void cfq_exit_icq(struct io_cq *icq) |
89850f7e | 2570 | { |
c5869807 | 2571 | struct cfq_io_cq *cic = icq_to_cic(icq); |
283287a5 | 2572 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
4faa3c81 | 2573 | |
ff6657c6 JA |
2574 | if (cic->cfqq[BLK_RW_ASYNC]) { |
2575 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
2576 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
2577 | } |
2578 | ||
ff6657c6 JA |
2579 | if (cic->cfqq[BLK_RW_SYNC]) { |
2580 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
2581 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 2582 | } |
89850f7e JA |
2583 | } |
2584 | ||
abede6da | 2585 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
22e2c507 JA |
2586 | { |
2587 | struct task_struct *tsk = current; | |
2588 | int ioprio_class; | |
2589 | ||
3b18152c | 2590 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
2591 | return; |
2592 | ||
598971bf | 2593 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
22e2c507 | 2594 | switch (ioprio_class) { |
fe094d98 JA |
2595 | default: |
2596 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
2597 | case IOPRIO_CLASS_NONE: | |
2598 | /* | |
6d63c275 | 2599 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
2600 | */ |
2601 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 2602 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
2603 | break; |
2604 | case IOPRIO_CLASS_RT: | |
598971bf | 2605 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
2606 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
2607 | break; | |
2608 | case IOPRIO_CLASS_BE: | |
598971bf | 2609 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
fe094d98 JA |
2610 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
2611 | break; | |
2612 | case IOPRIO_CLASS_IDLE: | |
2613 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
2614 | cfqq->ioprio = 7; | |
2615 | cfq_clear_cfqq_idle_window(cfqq); | |
2616 | break; | |
22e2c507 JA |
2617 | } |
2618 | ||
2619 | /* | |
2620 | * keep track of original prio settings in case we have to temporarily | |
2621 | * elevate the priority of this queue | |
2622 | */ | |
2623 | cfqq->org_ioprio = cfqq->ioprio; | |
3b18152c | 2624 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
2625 | } |
2626 | ||
598971bf | 2627 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
22e2c507 | 2628 | { |
598971bf | 2629 | int ioprio = cic->icq.ioc->ioprio; |
bca4b914 | 2630 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 2631 | struct cfq_queue *cfqq; |
35e6077c | 2632 | |
598971bf TH |
2633 | /* |
2634 | * Check whether ioprio has changed. The condition may trigger | |
2635 | * spuriously on a newly created cic but there's no harm. | |
2636 | */ | |
2637 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) | |
caaa5f9f JA |
2638 | return; |
2639 | ||
ff6657c6 | 2640 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
2641 | if (cfqq) { |
2642 | struct cfq_queue *new_cfqq; | |
abede6da TH |
2643 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio, |
2644 | GFP_ATOMIC); | |
caaa5f9f | 2645 | if (new_cfqq) { |
ff6657c6 | 2646 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
2647 | cfq_put_queue(cfqq); |
2648 | } | |
22e2c507 | 2649 | } |
caaa5f9f | 2650 | |
ff6657c6 | 2651 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
2652 | if (cfqq) |
2653 | cfq_mark_cfqq_prio_changed(cfqq); | |
598971bf TH |
2654 | |
2655 | cic->ioprio = ioprio; | |
22e2c507 JA |
2656 | } |
2657 | ||
d5036d77 | 2658 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2659 | pid_t pid, bool is_sync) |
d5036d77 JA |
2660 | { |
2661 | RB_CLEAR_NODE(&cfqq->rb_node); | |
2662 | RB_CLEAR_NODE(&cfqq->p_node); | |
2663 | INIT_LIST_HEAD(&cfqq->fifo); | |
2664 | ||
30d7b944 | 2665 | cfqq->ref = 0; |
d5036d77 JA |
2666 | cfqq->cfqd = cfqd; |
2667 | ||
2668 | cfq_mark_cfqq_prio_changed(cfqq); | |
2669 | ||
2670 | if (is_sync) { | |
2671 | if (!cfq_class_idle(cfqq)) | |
2672 | cfq_mark_cfqq_idle_window(cfqq); | |
2673 | cfq_mark_cfqq_sync(cfqq); | |
2674 | } | |
2675 | cfqq->pid = pid; | |
2676 | } | |
2677 | ||
24610333 | 2678 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
598971bf | 2679 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
24610333 | 2680 | { |
bca4b914 | 2681 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
598971bf TH |
2682 | struct cfq_queue *sync_cfqq; |
2683 | uint64_t id; | |
24610333 | 2684 | |
598971bf TH |
2685 | rcu_read_lock(); |
2686 | id = bio_blkio_cgroup(bio)->id; | |
2687 | rcu_read_unlock(); | |
24610333 | 2688 | |
598971bf TH |
2689 | /* |
2690 | * Check whether blkcg has changed. The condition may trigger | |
2691 | * spuriously on a newly created cic but there's no harm. | |
2692 | */ | |
2693 | if (unlikely(!cfqd) || likely(cic->blkcg_id == id)) | |
2694 | return; | |
24610333 | 2695 | |
598971bf | 2696 | sync_cfqq = cic_to_cfqq(cic, 1); |
24610333 VG |
2697 | if (sync_cfqq) { |
2698 | /* | |
2699 | * Drop reference to sync queue. A new sync queue will be | |
2700 | * assigned in new group upon arrival of a fresh request. | |
2701 | */ | |
2702 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
2703 | cic_set_cfqq(cic, NULL, 1); | |
2704 | cfq_put_queue(sync_cfqq); | |
2705 | } | |
598971bf TH |
2706 | |
2707 | cic->blkcg_id = id; | |
24610333 | 2708 | } |
598971bf TH |
2709 | #else |
2710 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } | |
24610333 VG |
2711 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
2712 | ||
22e2c507 | 2713 | static struct cfq_queue * |
abede6da TH |
2714 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
2715 | struct bio *bio, gfp_t gfp_mask) | |
22e2c507 | 2716 | { |
0a5a7d0e | 2717 | struct blkio_cgroup *blkcg; |
22e2c507 | 2718 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
cdb16e8f | 2719 | struct cfq_group *cfqg; |
22e2c507 JA |
2720 | |
2721 | retry: | |
2a7f1244 TH |
2722 | rcu_read_lock(); |
2723 | ||
4f85cb96 | 2724 | blkcg = bio_blkio_cgroup(bio); |
cd1604fa | 2725 | cfqg = cfq_lookup_create_cfqg(cfqd, blkcg); |
91fac317 | 2726 | cfqq = cic_to_cfqq(cic, is_sync); |
22e2c507 | 2727 | |
6118b70b JA |
2728 | /* |
2729 | * Always try a new alloc if we fell back to the OOM cfqq | |
2730 | * originally, since it should just be a temporary situation. | |
2731 | */ | |
2732 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
2733 | cfqq = NULL; | |
22e2c507 JA |
2734 | if (new_cfqq) { |
2735 | cfqq = new_cfqq; | |
2736 | new_cfqq = NULL; | |
2737 | } else if (gfp_mask & __GFP_WAIT) { | |
2a7f1244 | 2738 | rcu_read_unlock(); |
22e2c507 | 2739 | spin_unlock_irq(cfqd->queue->queue_lock); |
94f6030c | 2740 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 2741 | gfp_mask | __GFP_ZERO, |
94f6030c | 2742 | cfqd->queue->node); |
22e2c507 | 2743 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
2744 | if (new_cfqq) |
2745 | goto retry; | |
22e2c507 | 2746 | } else { |
94f6030c CL |
2747 | cfqq = kmem_cache_alloc_node(cfq_pool, |
2748 | gfp_mask | __GFP_ZERO, | |
2749 | cfqd->queue->node); | |
22e2c507 JA |
2750 | } |
2751 | ||
6118b70b JA |
2752 | if (cfqq) { |
2753 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
abede6da | 2754 | cfq_init_prio_data(cfqq, cic); |
cdb16e8f | 2755 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
2756 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
2757 | } else | |
2758 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
2759 | } |
2760 | ||
2761 | if (new_cfqq) | |
2762 | kmem_cache_free(cfq_pool, new_cfqq); | |
2763 | ||
2a7f1244 | 2764 | rcu_read_unlock(); |
22e2c507 JA |
2765 | return cfqq; |
2766 | } | |
2767 | ||
c2dea2d1 VT |
2768 | static struct cfq_queue ** |
2769 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
2770 | { | |
fe094d98 | 2771 | switch (ioprio_class) { |
c2dea2d1 VT |
2772 | case IOPRIO_CLASS_RT: |
2773 | return &cfqd->async_cfqq[0][ioprio]; | |
598971bf TH |
2774 | case IOPRIO_CLASS_NONE: |
2775 | ioprio = IOPRIO_NORM; | |
2776 | /* fall through */ | |
c2dea2d1 VT |
2777 | case IOPRIO_CLASS_BE: |
2778 | return &cfqd->async_cfqq[1][ioprio]; | |
2779 | case IOPRIO_CLASS_IDLE: | |
2780 | return &cfqd->async_idle_cfqq; | |
2781 | default: | |
2782 | BUG(); | |
2783 | } | |
2784 | } | |
2785 | ||
15c31be4 | 2786 | static struct cfq_queue * |
abede6da | 2787 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
4f85cb96 | 2788 | struct bio *bio, gfp_t gfp_mask) |
15c31be4 | 2789 | { |
598971bf TH |
2790 | const int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
2791 | const int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); | |
c2dea2d1 | 2792 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
2793 | struct cfq_queue *cfqq = NULL; |
2794 | ||
c2dea2d1 VT |
2795 | if (!is_sync) { |
2796 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
2797 | cfqq = *async_cfqq; | |
2798 | } | |
2799 | ||
6118b70b | 2800 | if (!cfqq) |
abede6da | 2801 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, cic, bio, gfp_mask); |
15c31be4 JA |
2802 | |
2803 | /* | |
2804 | * pin the queue now that it's allocated, scheduler exit will prune it | |
2805 | */ | |
c2dea2d1 | 2806 | if (!is_sync && !(*async_cfqq)) { |
30d7b944 | 2807 | cfqq->ref++; |
c2dea2d1 | 2808 | *async_cfqq = cfqq; |
15c31be4 JA |
2809 | } |
2810 | ||
30d7b944 | 2811 | cfqq->ref++; |
15c31be4 JA |
2812 | return cfqq; |
2813 | } | |
2814 | ||
22e2c507 | 2815 | static void |
383cd721 | 2816 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
1da177e4 | 2817 | { |
383cd721 SL |
2818 | unsigned long elapsed = jiffies - ttime->last_end_request; |
2819 | elapsed = min(elapsed, 2UL * slice_idle); | |
db3b5848 | 2820 | |
383cd721 SL |
2821 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
2822 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; | |
2823 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; | |
2824 | } | |
2825 | ||
2826 | static void | |
2827 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 2828 | struct cfq_io_cq *cic) |
383cd721 | 2829 | { |
f5f2b6ce | 2830 | if (cfq_cfqq_sync(cfqq)) { |
383cd721 | 2831 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
f5f2b6ce SL |
2832 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
2833 | cfqd->cfq_slice_idle); | |
2834 | } | |
7700fc4f SL |
2835 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2836 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); | |
2837 | #endif | |
22e2c507 | 2838 | } |
1da177e4 | 2839 | |
206dc69b | 2840 | static void |
b2c18e1e | 2841 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 2842 | struct request *rq) |
206dc69b | 2843 | { |
3dde36dd | 2844 | sector_t sdist = 0; |
41647e7a | 2845 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
2846 | if (cfqq->last_request_pos) { |
2847 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
2848 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
2849 | else | |
2850 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
2851 | } | |
206dc69b | 2852 | |
3dde36dd | 2853 | cfqq->seek_history <<= 1; |
41647e7a CZ |
2854 | if (blk_queue_nonrot(cfqd->queue)) |
2855 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
2856 | else | |
2857 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 2858 | } |
1da177e4 | 2859 | |
22e2c507 JA |
2860 | /* |
2861 | * Disable idle window if the process thinks too long or seeks so much that | |
2862 | * it doesn't matter | |
2863 | */ | |
2864 | static void | |
2865 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
c5869807 | 2866 | struct cfq_io_cq *cic) |
22e2c507 | 2867 | { |
7b679138 | 2868 | int old_idle, enable_idle; |
1be92f2f | 2869 | |
0871714e JA |
2870 | /* |
2871 | * Don't idle for async or idle io prio class | |
2872 | */ | |
2873 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
2874 | return; |
2875 | ||
c265a7f4 | 2876 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 2877 | |
76280aff CZ |
2878 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
2879 | cfq_mark_cfqq_deep(cfqq); | |
2880 | ||
749ef9f8 CZ |
2881 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
2882 | enable_idle = 0; | |
f6e8d01b | 2883 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
c5869807 TH |
2884 | !cfqd->cfq_slice_idle || |
2885 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) | |
22e2c507 | 2886 | enable_idle = 0; |
383cd721 SL |
2887 | else if (sample_valid(cic->ttime.ttime_samples)) { |
2888 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) | |
22e2c507 JA |
2889 | enable_idle = 0; |
2890 | else | |
2891 | enable_idle = 1; | |
1da177e4 LT |
2892 | } |
2893 | ||
7b679138 JA |
2894 | if (old_idle != enable_idle) { |
2895 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
2896 | if (enable_idle) | |
2897 | cfq_mark_cfqq_idle_window(cfqq); | |
2898 | else | |
2899 | cfq_clear_cfqq_idle_window(cfqq); | |
2900 | } | |
22e2c507 | 2901 | } |
1da177e4 | 2902 | |
22e2c507 JA |
2903 | /* |
2904 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
2905 | * no or if we aren't sure, a 1 will cause a preempt. | |
2906 | */ | |
a6151c3a | 2907 | static bool |
22e2c507 | 2908 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 2909 | struct request *rq) |
22e2c507 | 2910 | { |
6d048f53 | 2911 | struct cfq_queue *cfqq; |
22e2c507 | 2912 | |
6d048f53 JA |
2913 | cfqq = cfqd->active_queue; |
2914 | if (!cfqq) | |
a6151c3a | 2915 | return false; |
22e2c507 | 2916 | |
6d048f53 | 2917 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 2918 | return false; |
22e2c507 JA |
2919 | |
2920 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 2921 | return true; |
1e3335de | 2922 | |
875feb63 DS |
2923 | /* |
2924 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
2925 | */ | |
2926 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
2927 | return false; | |
2928 | ||
374f84ac JA |
2929 | /* |
2930 | * if the new request is sync, but the currently running queue is | |
2931 | * not, let the sync request have priority. | |
2932 | */ | |
5e705374 | 2933 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 2934 | return true; |
1e3335de | 2935 | |
8682e1f1 VG |
2936 | if (new_cfqq->cfqg != cfqq->cfqg) |
2937 | return false; | |
2938 | ||
2939 | if (cfq_slice_used(cfqq)) | |
2940 | return true; | |
2941 | ||
2942 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
2943 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && | |
2944 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && | |
2945 | new_cfqq->service_tree->count == 2 && | |
2946 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2947 | return true; | |
2948 | ||
b53d1ed7 JA |
2949 | /* |
2950 | * So both queues are sync. Let the new request get disk time if | |
2951 | * it's a metadata request and the current queue is doing regular IO. | |
2952 | */ | |
65299a3b | 2953 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
b53d1ed7 JA |
2954 | return true; |
2955 | ||
3a9a3f6c DS |
2956 | /* |
2957 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
2958 | */ | |
2959 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 2960 | return true; |
3a9a3f6c | 2961 | |
d2d59e18 SL |
2962 | /* An idle queue should not be idle now for some reason */ |
2963 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) | |
2964 | return true; | |
2965 | ||
1e3335de | 2966 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 2967 | return false; |
1e3335de JA |
2968 | |
2969 | /* | |
2970 | * if this request is as-good as one we would expect from the | |
2971 | * current cfqq, let it preempt | |
2972 | */ | |
e9ce335d | 2973 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 2974 | return true; |
1e3335de | 2975 | |
a6151c3a | 2976 | return false; |
22e2c507 JA |
2977 | } |
2978 | ||
2979 | /* | |
2980 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
2981 | * let it have half of its nominal slice. | |
2982 | */ | |
2983 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2984 | { | |
df0793ab SL |
2985 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
2986 | ||
7b679138 | 2987 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
df0793ab | 2988 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 2989 | |
f8ae6e3e SL |
2990 | /* |
2991 | * workload type is changed, don't save slice, otherwise preempt | |
2992 | * doesn't happen | |
2993 | */ | |
df0793ab | 2994 | if (old_type != cfqq_type(cfqq)) |
f8ae6e3e SL |
2995 | cfqq->cfqg->saved_workload_slice = 0; |
2996 | ||
bf572256 JA |
2997 | /* |
2998 | * Put the new queue at the front of the of the current list, | |
2999 | * so we know that it will be selected next. | |
3000 | */ | |
3001 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3002 | |
3003 | cfq_service_tree_add(cfqd, cfqq, 1); | |
eda5e0c9 | 3004 | |
62a37f6b JT |
3005 | cfqq->slice_end = 0; |
3006 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3007 | } |
3008 | ||
22e2c507 | 3009 | /* |
5e705374 | 3010 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3011 | * something we should do about it |
3012 | */ | |
3013 | static void | |
5e705374 JA |
3014 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3015 | struct request *rq) | |
22e2c507 | 3016 | { |
c5869807 | 3017 | struct cfq_io_cq *cic = RQ_CIC(rq); |
12e9fddd | 3018 | |
45333d5a | 3019 | cfqd->rq_queued++; |
65299a3b CH |
3020 | if (rq->cmd_flags & REQ_PRIO) |
3021 | cfqq->prio_pending++; | |
374f84ac | 3022 | |
383cd721 | 3023 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
b2c18e1e | 3024 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3025 | cfq_update_idle_window(cfqd, cfqq, cic); |
3026 | ||
b2c18e1e | 3027 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3028 | |
3029 | if (cfqq == cfqd->active_queue) { | |
3030 | /* | |
b029195d JA |
3031 | * Remember that we saw a request from this process, but |
3032 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3033 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3034 | * and merging. If the request is already larger than a single |
3035 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3036 | * merging is already done. Ditto for a busy system that |
3037 | * has other work pending, don't risk delaying until the | |
3038 | * idle timer unplug to continue working. | |
22e2c507 | 3039 | */ |
d6ceb25e | 3040 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3041 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3042 | cfqd->busy_queues > 1) { | |
812df48d | 3043 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3044 | cfq_clear_cfqq_wait_request(cfqq); |
24ecfbe2 | 3045 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3046 | } else { |
e98ef89b | 3047 | cfq_blkiocg_update_idle_time_stats( |
c1768268 TH |
3048 | cfqg_to_blkg(cfqq->cfqg), |
3049 | &blkio_policy_cfq); | |
bf791937 | 3050 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3051 | } |
d6ceb25e | 3052 | } |
5e705374 | 3053 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3054 | /* |
3055 | * not the active queue - expire current slice if it is | |
3056 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3057 | * has some old slice time left and is of higher priority or |
3058 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3059 | */ |
3060 | cfq_preempt_queue(cfqd, cfqq); | |
24ecfbe2 | 3061 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3062 | } |
1da177e4 LT |
3063 | } |
3064 | ||
165125e1 | 3065 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3066 | { |
b4878f24 | 3067 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3068 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3069 | |
7b679138 | 3070 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
abede6da | 3071 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
1da177e4 | 3072 | |
30996f40 | 3073 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3074 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3075 | cfq_add_rq_rb(rq); |
0381411e | 3076 | cfq_blkiocg_update_io_add_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
c1768268 | 3077 | &blkio_policy_cfq, |
0381411e TH |
3078 | cfqg_to_blkg(cfqd->serving_group), |
3079 | rq_data_dir(rq), rq_is_sync(rq)); | |
5e705374 | 3080 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3081 | } |
3082 | ||
45333d5a AC |
3083 | /* |
3084 | * Update hw_tag based on peak queue depth over 50 samples under | |
3085 | * sufficient load. | |
3086 | */ | |
3087 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3088 | { | |
1a1238a7 SL |
3089 | struct cfq_queue *cfqq = cfqd->active_queue; |
3090 | ||
53c583d2 CZ |
3091 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3092 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3093 | |
3094 | if (cfqd->hw_tag == 1) | |
3095 | return; | |
45333d5a AC |
3096 | |
3097 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3098 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3099 | return; |
3100 | ||
1a1238a7 SL |
3101 | /* |
3102 | * If active queue hasn't enough requests and can idle, cfq might not | |
3103 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3104 | * case | |
3105 | */ | |
3106 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3107 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3108 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3109 | return; |
3110 | ||
45333d5a AC |
3111 | if (cfqd->hw_tag_samples++ < 50) |
3112 | return; | |
3113 | ||
e459dd08 | 3114 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3115 | cfqd->hw_tag = 1; |
3116 | else | |
3117 | cfqd->hw_tag = 0; | |
45333d5a AC |
3118 | } |
3119 | ||
7667aa06 VG |
3120 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3121 | { | |
c5869807 | 3122 | struct cfq_io_cq *cic = cfqd->active_cic; |
7667aa06 | 3123 | |
02a8f01b JT |
3124 | /* If the queue already has requests, don't wait */ |
3125 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3126 | return false; | |
3127 | ||
7667aa06 VG |
3128 | /* If there are other queues in the group, don't wait */ |
3129 | if (cfqq->cfqg->nr_cfqq > 1) | |
3130 | return false; | |
3131 | ||
7700fc4f SL |
3132 | /* the only queue in the group, but think time is big */ |
3133 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) | |
3134 | return false; | |
3135 | ||
7667aa06 VG |
3136 | if (cfq_slice_used(cfqq)) |
3137 | return true; | |
3138 | ||
3139 | /* if slice left is less than think time, wait busy */ | |
383cd721 SL |
3140 | if (cic && sample_valid(cic->ttime.ttime_samples) |
3141 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) | |
7667aa06 VG |
3142 | return true; |
3143 | ||
3144 | /* | |
3145 | * If think times is less than a jiffy than ttime_mean=0 and above | |
3146 | * will not be true. It might happen that slice has not expired yet | |
3147 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
3148 | * case where think time is less than a jiffy, mark the queue wait | |
3149 | * busy if only 1 jiffy is left in the slice. | |
3150 | */ | |
3151 | if (cfqq->slice_end - jiffies == 1) | |
3152 | return true; | |
3153 | ||
3154 | return false; | |
3155 | } | |
3156 | ||
165125e1 | 3157 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3158 | { |
5e705374 | 3159 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3160 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3161 | const int sync = rq_is_sync(rq); |
b4878f24 | 3162 | unsigned long now; |
1da177e4 | 3163 | |
b4878f24 | 3164 | now = jiffies; |
33659ebb CH |
3165 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3166 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 3167 | |
45333d5a AC |
3168 | cfq_update_hw_tag(cfqd); |
3169 | ||
53c583d2 | 3170 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 3171 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 3172 | cfqd->rq_in_driver--; |
6d048f53 | 3173 | cfqq->dispatched--; |
80bdf0c7 | 3174 | (RQ_CFQG(rq))->dispatched--; |
0381411e | 3175 | cfq_blkiocg_update_completion_stats(cfqg_to_blkg(cfqq->cfqg), |
c1768268 TH |
3176 | &blkio_policy_cfq, rq_start_time_ns(rq), |
3177 | rq_io_start_time_ns(rq), rq_data_dir(rq), | |
3178 | rq_is_sync(rq)); | |
1da177e4 | 3179 | |
53c583d2 | 3180 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 3181 | |
365722bb | 3182 | if (sync) { |
f5f2b6ce SL |
3183 | struct cfq_rb_root *service_tree; |
3184 | ||
383cd721 | 3185 | RQ_CIC(rq)->ttime.last_end_request = now; |
f5f2b6ce SL |
3186 | |
3187 | if (cfq_cfqq_on_rr(cfqq)) | |
3188 | service_tree = cfqq->service_tree; | |
3189 | else | |
3190 | service_tree = service_tree_for(cfqq->cfqg, | |
3191 | cfqq_prio(cfqq), cfqq_type(cfqq)); | |
3192 | service_tree->ttime.last_end_request = now; | |
573412b2 CZ |
3193 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3194 | cfqd->last_delayed_sync = now; | |
365722bb | 3195 | } |
caaa5f9f | 3196 | |
7700fc4f SL |
3197 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3198 | cfqq->cfqg->ttime.last_end_request = now; | |
3199 | #endif | |
3200 | ||
caaa5f9f JA |
3201 | /* |
3202 | * If this is the active queue, check if it needs to be expired, | |
3203 | * or if we want to idle in case it has no pending requests. | |
3204 | */ | |
3205 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3206 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3207 | ||
44f7c160 JA |
3208 | if (cfq_cfqq_slice_new(cfqq)) { |
3209 | cfq_set_prio_slice(cfqd, cfqq); | |
3210 | cfq_clear_cfqq_slice_new(cfqq); | |
3211 | } | |
f75edf2d VG |
3212 | |
3213 | /* | |
7667aa06 VG |
3214 | * Should we wait for next request to come in before we expire |
3215 | * the queue. | |
f75edf2d | 3216 | */ |
7667aa06 | 3217 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
3218 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3219 | if (!cfqd->cfq_slice_idle) | |
3220 | extend_sl = cfqd->cfq_group_idle; | |
3221 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 3222 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 3223 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
3224 | } |
3225 | ||
a36e71f9 | 3226 | /* |
8e550632 CZ |
3227 | * Idling is not enabled on: |
3228 | * - expired queues | |
3229 | * - idle-priority queues | |
3230 | * - async queues | |
3231 | * - queues with still some requests queued | |
3232 | * - when there is a close cooperator | |
a36e71f9 | 3233 | */ |
0871714e | 3234 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 3235 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3236 | else if (sync && cfqq_empty && |
3237 | !cfq_close_cooperator(cfqd, cfqq)) { | |
749ef9f8 | 3238 | cfq_arm_slice_timer(cfqd); |
8e550632 | 3239 | } |
caaa5f9f | 3240 | } |
6d048f53 | 3241 | |
53c583d2 | 3242 | if (!cfqd->rq_in_driver) |
23e018a1 | 3243 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3244 | } |
3245 | ||
89850f7e | 3246 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3247 | { |
1b379d8d | 3248 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3249 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3250 | return ELV_MQUEUE_MUST; |
3b18152c | 3251 | } |
1da177e4 | 3252 | |
22e2c507 | 3253 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3254 | } |
3255 | ||
165125e1 | 3256 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3257 | { |
3258 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3259 | struct task_struct *tsk = current; | |
c5869807 | 3260 | struct cfq_io_cq *cic; |
22e2c507 JA |
3261 | struct cfq_queue *cfqq; |
3262 | ||
3263 | /* | |
3264 | * don't force setup of a queue from here, as a call to may_queue | |
3265 | * does not necessarily imply that a request actually will be queued. | |
3266 | * so just lookup a possibly existing queue, or return 'may queue' | |
3267 | * if that fails | |
3268 | */ | |
4ac845a2 | 3269 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3270 | if (!cic) |
3271 | return ELV_MQUEUE_MAY; | |
3272 | ||
b0b78f81 | 3273 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3274 | if (cfqq) { |
abede6da | 3275 | cfq_init_prio_data(cfqq, cic); |
22e2c507 | 3276 | |
89850f7e | 3277 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3278 | } |
3279 | ||
3280 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3281 | } |
3282 | ||
1da177e4 LT |
3283 | /* |
3284 | * queue lock held here | |
3285 | */ | |
bb37b94c | 3286 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3287 | { |
5e705374 | 3288 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3289 | |
5e705374 | 3290 | if (cfqq) { |
22e2c507 | 3291 | const int rw = rq_data_dir(rq); |
1da177e4 | 3292 | |
22e2c507 JA |
3293 | BUG_ON(!cfqq->allocated[rw]); |
3294 | cfqq->allocated[rw]--; | |
1da177e4 | 3295 | |
7f1dc8a2 | 3296 | /* Put down rq reference on cfqg */ |
eb7d8c07 | 3297 | cfqg_put(RQ_CFQG(rq)); |
a612fddf TH |
3298 | rq->elv.priv[0] = NULL; |
3299 | rq->elv.priv[1] = NULL; | |
7f1dc8a2 | 3300 | |
1da177e4 LT |
3301 | cfq_put_queue(cfqq); |
3302 | } | |
3303 | } | |
3304 | ||
df5fe3e8 | 3305 | static struct cfq_queue * |
c5869807 | 3306 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
df5fe3e8 JM |
3307 | struct cfq_queue *cfqq) |
3308 | { | |
3309 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3310 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3311 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3312 | cfq_put_queue(cfqq); |
3313 | return cic_to_cfqq(cic, 1); | |
3314 | } | |
3315 | ||
e6c5bc73 JM |
3316 | /* |
3317 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3318 | * was the last process referring to said cfqq. | |
3319 | */ | |
3320 | static struct cfq_queue * | |
c5869807 | 3321 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
e6c5bc73 JM |
3322 | { |
3323 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
3324 | cfqq->pid = current->pid; |
3325 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 3326 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
3327 | return cfqq; |
3328 | } | |
3329 | ||
3330 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
3331 | |
3332 | cfq_put_cooperator(cfqq); | |
3333 | ||
e6c5bc73 JM |
3334 | cfq_put_queue(cfqq); |
3335 | return NULL; | |
3336 | } | |
1da177e4 | 3337 | /* |
22e2c507 | 3338 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3339 | */ |
22e2c507 | 3340 | static int |
852c788f TH |
3341 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
3342 | gfp_t gfp_mask) | |
1da177e4 LT |
3343 | { |
3344 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
f1f8cc94 | 3345 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
1da177e4 | 3346 | const int rw = rq_data_dir(rq); |
a6151c3a | 3347 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3348 | struct cfq_queue *cfqq; |
1da177e4 LT |
3349 | |
3350 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3351 | ||
216284c3 | 3352 | spin_lock_irq(q->queue_lock); |
f1f8cc94 | 3353 | |
598971bf TH |
3354 | check_ioprio_changed(cic, bio); |
3355 | check_blkcg_changed(cic, bio); | |
e6c5bc73 | 3356 | new_queue: |
91fac317 | 3357 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3358 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
abede6da | 3359 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio, gfp_mask); |
91fac317 | 3360 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3361 | } else { |
e6c5bc73 JM |
3362 | /* |
3363 | * If the queue was seeky for too long, break it apart. | |
3364 | */ | |
ae54abed | 3365 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
3366 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
3367 | cfqq = split_cfqq(cic, cfqq); | |
3368 | if (!cfqq) | |
3369 | goto new_queue; | |
3370 | } | |
3371 | ||
df5fe3e8 JM |
3372 | /* |
3373 | * Check to see if this queue is scheduled to merge with | |
3374 | * another, closely cooperating queue. The merging of | |
3375 | * queues happens here as it must be done in process context. | |
3376 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3377 | */ | |
3378 | if (cfqq->new_cfqq) | |
3379 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3380 | } |
1da177e4 LT |
3381 | |
3382 | cfqq->allocated[rw]++; | |
1da177e4 | 3383 | |
6fae9c25 | 3384 | cfqq->ref++; |
eb7d8c07 | 3385 | cfqg_get(cfqq->cfqg); |
a612fddf | 3386 | rq->elv.priv[0] = cfqq; |
1adaf3dd | 3387 | rq->elv.priv[1] = cfqq->cfqg; |
216284c3 | 3388 | spin_unlock_irq(q->queue_lock); |
5e705374 | 3389 | return 0; |
1da177e4 LT |
3390 | } |
3391 | ||
65f27f38 | 3392 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3393 | { |
65f27f38 | 3394 | struct cfq_data *cfqd = |
23e018a1 | 3395 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3396 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3397 | |
40bb54d1 | 3398 | spin_lock_irq(q->queue_lock); |
24ecfbe2 | 3399 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3400 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3401 | } |
3402 | ||
3403 | /* | |
3404 | * Timer running if the active_queue is currently idling inside its time slice | |
3405 | */ | |
3406 | static void cfq_idle_slice_timer(unsigned long data) | |
3407 | { | |
3408 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3409 | struct cfq_queue *cfqq; | |
3410 | unsigned long flags; | |
3c6bd2f8 | 3411 | int timed_out = 1; |
22e2c507 | 3412 | |
7b679138 JA |
3413 | cfq_log(cfqd, "idle timer fired"); |
3414 | ||
22e2c507 JA |
3415 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3416 | ||
fe094d98 JA |
3417 | cfqq = cfqd->active_queue; |
3418 | if (cfqq) { | |
3c6bd2f8 JA |
3419 | timed_out = 0; |
3420 | ||
b029195d JA |
3421 | /* |
3422 | * We saw a request before the queue expired, let it through | |
3423 | */ | |
3424 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3425 | goto out_kick; | |
3426 | ||
22e2c507 JA |
3427 | /* |
3428 | * expired | |
3429 | */ | |
44f7c160 | 3430 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
3431 | goto expire; |
3432 | ||
3433 | /* | |
3434 | * only expire and reinvoke request handler, if there are | |
3435 | * other queues with pending requests | |
3436 | */ | |
caaa5f9f | 3437 | if (!cfqd->busy_queues) |
22e2c507 | 3438 | goto out_cont; |
22e2c507 JA |
3439 | |
3440 | /* | |
3441 | * not expired and it has a request pending, let it dispatch | |
3442 | */ | |
75e50984 | 3443 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3444 | goto out_kick; |
76280aff CZ |
3445 | |
3446 | /* | |
3447 | * Queue depth flag is reset only when the idle didn't succeed | |
3448 | */ | |
3449 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
3450 | } |
3451 | expire: | |
e5ff082e | 3452 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 3453 | out_kick: |
23e018a1 | 3454 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
3455 | out_cont: |
3456 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3457 | } | |
3458 | ||
3b18152c JA |
3459 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3460 | { | |
3461 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 3462 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 3463 | } |
22e2c507 | 3464 | |
c2dea2d1 VT |
3465 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3466 | { | |
3467 | int i; | |
3468 | ||
3469 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
3470 | if (cfqd->async_cfqq[0][i]) | |
3471 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
3472 | if (cfqd->async_cfqq[1][i]) | |
3473 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 3474 | } |
2389d1ef ON |
3475 | |
3476 | if (cfqd->async_idle_cfqq) | |
3477 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
3478 | } |
3479 | ||
b374d18a | 3480 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 3481 | { |
22e2c507 | 3482 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 3483 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3484 | |
3b18152c | 3485 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 3486 | |
d9ff4187 | 3487 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3488 | |
d9ff4187 | 3489 | if (cfqd->active_queue) |
e5ff082e | 3490 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 | 3491 | |
c2dea2d1 | 3492 | cfq_put_async_queues(cfqd); |
03aa264a TH |
3493 | |
3494 | spin_unlock_irq(q->queue_lock); | |
3495 | ||
a90d742e AV |
3496 | cfq_shutdown_timer_wq(cfqd); |
3497 | ||
f51b802c TH |
3498 | #ifndef CONFIG_CFQ_GROUP_IOSCHED |
3499 | kfree(cfqd->root_group); | |
2abae55f | 3500 | #endif |
e8989fae | 3501 | update_root_blkg_pd(q, BLKIO_POLICY_PROP); |
56edf7d7 | 3502 | kfree(cfqd); |
1da177e4 LT |
3503 | } |
3504 | ||
b2fab5ac | 3505 | static int cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
3506 | { |
3507 | struct cfq_data *cfqd; | |
cd1604fa | 3508 | struct blkio_group *blkg __maybe_unused; |
f51b802c | 3509 | int i; |
1da177e4 | 3510 | |
94f6030c | 3511 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
a73f730d | 3512 | if (!cfqd) |
b2fab5ac | 3513 | return -ENOMEM; |
80b15c73 | 3514 | |
f51b802c TH |
3515 | cfqd->queue = q; |
3516 | q->elevator->elevator_data = cfqd; | |
3517 | ||
1fa8f6d6 VG |
3518 | /* Init root service tree */ |
3519 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
3520 | ||
f51b802c | 3521 | /* Init root group and prefer root group over other groups by default */ |
25fb5169 | 3522 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
f51b802c TH |
3523 | rcu_read_lock(); |
3524 | spin_lock_irq(q->queue_lock); | |
5624a4e4 | 3525 | |
aaec55a0 | 3526 | blkg = blkg_lookup_create(&blkio_root_cgroup, q, true); |
cd1604fa | 3527 | if (!IS_ERR(blkg)) |
0381411e | 3528 | cfqd->root_group = blkg_to_cfqg(blkg); |
f51b802c TH |
3529 | |
3530 | spin_unlock_irq(q->queue_lock); | |
3531 | rcu_read_unlock(); | |
3532 | #else | |
3533 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), | |
3534 | GFP_KERNEL, cfqd->queue->node); | |
3535 | if (cfqd->root_group) | |
3536 | cfq_init_cfqg_base(cfqd->root_group); | |
3537 | #endif | |
3538 | if (!cfqd->root_group) { | |
5624a4e4 | 3539 | kfree(cfqd); |
b2fab5ac | 3540 | return -ENOMEM; |
5624a4e4 VG |
3541 | } |
3542 | ||
f51b802c | 3543 | cfqd->root_group->weight = 2*BLKIO_WEIGHT_DEFAULT; |
5624a4e4 | 3544 | |
26a2ac00 JA |
3545 | /* |
3546 | * Not strictly needed (since RB_ROOT just clears the node and we | |
3547 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
3548 | * to add magic to the rb code | |
3549 | */ | |
3550 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
3551 | cfqd->prio_trees[i] = RB_ROOT; | |
3552 | ||
6118b70b JA |
3553 | /* |
3554 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
3555 | * Grab a permanent reference to it, so that the normal code flow | |
f51b802c TH |
3556 | * will not attempt to free it. oom_cfqq is linked to root_group |
3557 | * but shouldn't hold a reference as it'll never be unlinked. Lose | |
3558 | * the reference from linking right away. | |
6118b70b JA |
3559 | */ |
3560 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
30d7b944 | 3561 | cfqd->oom_cfqq.ref++; |
1adaf3dd TH |
3562 | |
3563 | spin_lock_irq(q->queue_lock); | |
f51b802c | 3564 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
eb7d8c07 | 3565 | cfqg_put(cfqd->root_group); |
1adaf3dd | 3566 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 3567 | |
22e2c507 JA |
3568 | init_timer(&cfqd->idle_slice_timer); |
3569 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
3570 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
3571 | ||
23e018a1 | 3572 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 3573 | |
1da177e4 | 3574 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
3575 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
3576 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
3577 | cfqd->cfq_back_max = cfq_back_max; |
3578 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
3579 | cfqd->cfq_slice[0] = cfq_slice_async; |
3580 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
3581 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
3582 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
80bdf0c7 | 3583 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 3584 | cfqd->cfq_latency = 1; |
e459dd08 | 3585 | cfqd->hw_tag = -1; |
edc71131 CZ |
3586 | /* |
3587 | * we optimistically start assuming sync ops weren't delayed in last | |
3588 | * second, in order to have larger depth for async operations. | |
3589 | */ | |
573412b2 | 3590 | cfqd->last_delayed_sync = jiffies - HZ; |
b2fab5ac | 3591 | return 0; |
1da177e4 LT |
3592 | } |
3593 | ||
1da177e4 LT |
3594 | /* |
3595 | * sysfs parts below --> | |
3596 | */ | |
1da177e4 LT |
3597 | static ssize_t |
3598 | cfq_var_show(unsigned int var, char *page) | |
3599 | { | |
3600 | return sprintf(page, "%d\n", var); | |
3601 | } | |
3602 | ||
3603 | static ssize_t | |
3604 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
3605 | { | |
3606 | char *p = (char *) page; | |
3607 | ||
3608 | *var = simple_strtoul(p, &p, 10); | |
3609 | return count; | |
3610 | } | |
3611 | ||
1da177e4 | 3612 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 3613 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 3614 | { \ |
3d1ab40f | 3615 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3616 | unsigned int __data = __VAR; \ |
3617 | if (__CONV) \ | |
3618 | __data = jiffies_to_msecs(__data); \ | |
3619 | return cfq_var_show(__data, (page)); \ | |
3620 | } | |
3621 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
3622 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
3623 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
3624 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
3625 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 3626 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 3627 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
3628 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
3629 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
3630 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 3631 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
1da177e4 LT |
3632 | #undef SHOW_FUNCTION |
3633 | ||
3634 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 3635 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 3636 | { \ |
3d1ab40f | 3637 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3638 | unsigned int __data; \ |
3639 | int ret = cfq_var_store(&__data, (page), count); \ | |
3640 | if (__data < (MIN)) \ | |
3641 | __data = (MIN); \ | |
3642 | else if (__data > (MAX)) \ | |
3643 | __data = (MAX); \ | |
3644 | if (__CONV) \ | |
3645 | *(__PTR) = msecs_to_jiffies(__data); \ | |
3646 | else \ | |
3647 | *(__PTR) = __data; \ | |
3648 | return ret; \ | |
3649 | } | |
3650 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
3651 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
3652 | UINT_MAX, 1); | |
3653 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
3654 | UINT_MAX, 1); | |
e572ec7e | 3655 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
3656 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
3657 | UINT_MAX, 0); | |
22e2c507 | 3658 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 3659 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
3660 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
3661 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
3662 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
3663 | UINT_MAX, 0); | |
963b72fc | 3664 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
1da177e4 LT |
3665 | #undef STORE_FUNCTION |
3666 | ||
e572ec7e AV |
3667 | #define CFQ_ATTR(name) \ |
3668 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
3669 | ||
3670 | static struct elv_fs_entry cfq_attrs[] = { | |
3671 | CFQ_ATTR(quantum), | |
e572ec7e AV |
3672 | CFQ_ATTR(fifo_expire_sync), |
3673 | CFQ_ATTR(fifo_expire_async), | |
3674 | CFQ_ATTR(back_seek_max), | |
3675 | CFQ_ATTR(back_seek_penalty), | |
3676 | CFQ_ATTR(slice_sync), | |
3677 | CFQ_ATTR(slice_async), | |
3678 | CFQ_ATTR(slice_async_rq), | |
3679 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 3680 | CFQ_ATTR(group_idle), |
963b72fc | 3681 | CFQ_ATTR(low_latency), |
e572ec7e | 3682 | __ATTR_NULL |
1da177e4 LT |
3683 | }; |
3684 | ||
1da177e4 LT |
3685 | static struct elevator_type iosched_cfq = { |
3686 | .ops = { | |
3687 | .elevator_merge_fn = cfq_merge, | |
3688 | .elevator_merged_fn = cfq_merged_request, | |
3689 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 3690 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 3691 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 3692 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 3693 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 3694 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 | 3695 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
1da177e4 | 3696 | .elevator_completed_req_fn = cfq_completed_request, |
21183b07 JA |
3697 | .elevator_former_req_fn = elv_rb_former_request, |
3698 | .elevator_latter_req_fn = elv_rb_latter_request, | |
9b84cacd | 3699 | .elevator_init_icq_fn = cfq_init_icq, |
7e5a8794 | 3700 | .elevator_exit_icq_fn = cfq_exit_icq, |
1da177e4 LT |
3701 | .elevator_set_req_fn = cfq_set_request, |
3702 | .elevator_put_req_fn = cfq_put_request, | |
3703 | .elevator_may_queue_fn = cfq_may_queue, | |
3704 | .elevator_init_fn = cfq_init_queue, | |
3705 | .elevator_exit_fn = cfq_exit_queue, | |
3706 | }, | |
3d3c2379 TH |
3707 | .icq_size = sizeof(struct cfq_io_cq), |
3708 | .icq_align = __alignof__(struct cfq_io_cq), | |
3d1ab40f | 3709 | .elevator_attrs = cfq_attrs, |
3d3c2379 | 3710 | .elevator_name = "cfq", |
1da177e4 LT |
3711 | .elevator_owner = THIS_MODULE, |
3712 | }; | |
3713 | ||
3e252066 VG |
3714 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3715 | static struct blkio_policy_type blkio_policy_cfq = { | |
3716 | .ops = { | |
0381411e | 3717 | .blkio_init_group_fn = cfq_init_blkio_group, |
3e252066 VG |
3718 | .blkio_update_group_weight_fn = cfq_update_blkio_group_weight, |
3719 | }, | |
062a644d | 3720 | .plid = BLKIO_POLICY_PROP, |
0381411e | 3721 | .pdata_size = sizeof(struct cfq_group), |
3e252066 | 3722 | }; |
3e252066 VG |
3723 | #endif |
3724 | ||
1da177e4 LT |
3725 | static int __init cfq_init(void) |
3726 | { | |
3d3c2379 TH |
3727 | int ret; |
3728 | ||
22e2c507 JA |
3729 | /* |
3730 | * could be 0 on HZ < 1000 setups | |
3731 | */ | |
3732 | if (!cfq_slice_async) | |
3733 | cfq_slice_async = 1; | |
3734 | if (!cfq_slice_idle) | |
3735 | cfq_slice_idle = 1; | |
3736 | ||
80bdf0c7 VG |
3737 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3738 | if (!cfq_group_idle) | |
3739 | cfq_group_idle = 1; | |
3740 | #else | |
3741 | cfq_group_idle = 0; | |
3742 | #endif | |
3d3c2379 TH |
3743 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
3744 | if (!cfq_pool) | |
1da177e4 LT |
3745 | return -ENOMEM; |
3746 | ||
3d3c2379 TH |
3747 | ret = elv_register(&iosched_cfq); |
3748 | if (ret) { | |
3749 | kmem_cache_destroy(cfq_pool); | |
3750 | return ret; | |
3751 | } | |
3d3c2379 | 3752 | |
b95ada55 | 3753 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3e252066 | 3754 | blkio_policy_register(&blkio_policy_cfq); |
b95ada55 | 3755 | #endif |
2fdd82bd | 3756 | return 0; |
1da177e4 LT |
3757 | } |
3758 | ||
3759 | static void __exit cfq_exit(void) | |
3760 | { | |
b95ada55 | 3761 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3e252066 | 3762 | blkio_policy_unregister(&blkio_policy_cfq); |
b95ada55 | 3763 | #endif |
1da177e4 | 3764 | elv_unregister(&iosched_cfq); |
3d3c2379 | 3765 | kmem_cache_destroy(cfq_pool); |
1da177e4 LT |
3766 | } |
3767 | ||
3768 | module_init(cfq_init); | |
3769 | module_exit(cfq_exit); | |
3770 | ||
3771 | MODULE_AUTHOR("Jens Axboe"); | |
3772 | MODULE_LICENSE("GPL"); | |
3773 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |