]>
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> |
e98ef89b | 17 | #include "cfq.h" |
1da177e4 LT |
18 | |
19 | /* | |
20 | * tunables | |
21 | */ | |
fe094d98 | 22 | /* max queue in one round of service */ |
abc3c744 | 23 | static const int cfq_quantum = 8; |
64100099 | 24 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
25 | /* maximum backwards seek, in KiB */ |
26 | static const int cfq_back_max = 16 * 1024; | |
27 | /* penalty of a backwards seek */ | |
28 | static const int cfq_back_penalty = 2; | |
64100099 | 29 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 30 | static int cfq_slice_async = HZ / 25; |
64100099 | 31 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 32 | static int cfq_slice_idle = HZ / 125; |
80bdf0c7 | 33 | static int cfq_group_idle = HZ / 125; |
5db5d642 CZ |
34 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
35 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 36 | |
d9e7620e | 37 | /* |
0871714e | 38 | * offset from end of service tree |
d9e7620e | 39 | */ |
0871714e | 40 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
41 | |
42 | /* | |
43 | * below this threshold, we consider thinktime immediate | |
44 | */ | |
45 | #define CFQ_MIN_TT (2) | |
46 | ||
22e2c507 | 47 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 48 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 49 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 50 | |
3dde36dd | 51 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
e9ce335d | 52 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
41647e7a | 53 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
3dde36dd | 54 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
ae54abed | 55 | |
fe094d98 JA |
56 | #define RQ_CIC(rq) \ |
57 | ((struct cfq_io_context *) (rq)->elevator_private) | |
7b679138 | 58 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private2) |
7f1dc8a2 | 59 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elevator_private3) |
1da177e4 | 60 | |
e18b890b CL |
61 | static struct kmem_cache *cfq_pool; |
62 | static struct kmem_cache *cfq_ioc_pool; | |
1da177e4 | 63 | |
245b2e70 | 64 | static DEFINE_PER_CPU(unsigned long, cfq_ioc_count); |
334e94de | 65 | static struct completion *ioc_gone; |
9a11b4ed | 66 | static DEFINE_SPINLOCK(ioc_gone_lock); |
334e94de | 67 | |
80b15c73 KK |
68 | static DEFINE_SPINLOCK(cic_index_lock); |
69 | static DEFINE_IDA(cic_index_ida); | |
70 | ||
22e2c507 JA |
71 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
72 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
73 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
74 | ||
206dc69b | 75 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 76 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 77 | |
cc09e299 JA |
78 | /* |
79 | * Most of our rbtree usage is for sorting with min extraction, so | |
80 | * if we cache the leftmost node we don't have to walk down the tree | |
81 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
82 | * move this into the elevator for the rq sorting as well. | |
83 | */ | |
84 | struct cfq_rb_root { | |
85 | struct rb_root rb; | |
86 | struct rb_node *left; | |
aa6f6a3d | 87 | unsigned count; |
73e9ffdd | 88 | unsigned total_weight; |
1fa8f6d6 | 89 | u64 min_vdisktime; |
25bc6b07 | 90 | struct rb_node *active; |
cc09e299 | 91 | }; |
73e9ffdd RK |
92 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, .left = NULL, \ |
93 | .count = 0, .min_vdisktime = 0, } | |
cc09e299 | 94 | |
6118b70b JA |
95 | /* |
96 | * Per process-grouping structure | |
97 | */ | |
98 | struct cfq_queue { | |
99 | /* reference count */ | |
100 | atomic_t ref; | |
101 | /* various state flags, see below */ | |
102 | unsigned int flags; | |
103 | /* parent cfq_data */ | |
104 | struct cfq_data *cfqd; | |
105 | /* service_tree member */ | |
106 | struct rb_node rb_node; | |
107 | /* service_tree key */ | |
108 | unsigned long rb_key; | |
109 | /* prio tree member */ | |
110 | struct rb_node p_node; | |
111 | /* prio tree root we belong to, if any */ | |
112 | struct rb_root *p_root; | |
113 | /* sorted list of pending requests */ | |
114 | struct rb_root sort_list; | |
115 | /* if fifo isn't expired, next request to serve */ | |
116 | struct request *next_rq; | |
117 | /* requests queued in sort_list */ | |
118 | int queued[2]; | |
119 | /* currently allocated requests */ | |
120 | int allocated[2]; | |
121 | /* fifo list of requests in sort_list */ | |
122 | struct list_head fifo; | |
123 | ||
dae739eb VG |
124 | /* time when queue got scheduled in to dispatch first request. */ |
125 | unsigned long dispatch_start; | |
f75edf2d | 126 | unsigned int allocated_slice; |
c4081ba5 | 127 | unsigned int slice_dispatch; |
dae739eb VG |
128 | /* time when first request from queue completed and slice started. */ |
129 | unsigned long slice_start; | |
6118b70b JA |
130 | unsigned long slice_end; |
131 | long slice_resid; | |
6118b70b JA |
132 | |
133 | /* pending metadata requests */ | |
134 | int meta_pending; | |
135 | /* number of requests that are on the dispatch list or inside driver */ | |
136 | int dispatched; | |
137 | ||
138 | /* io prio of this group */ | |
139 | unsigned short ioprio, org_ioprio; | |
140 | unsigned short ioprio_class, org_ioprio_class; | |
141 | ||
c4081ba5 RK |
142 | pid_t pid; |
143 | ||
3dde36dd | 144 | u32 seek_history; |
b2c18e1e JM |
145 | sector_t last_request_pos; |
146 | ||
aa6f6a3d | 147 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 148 | struct cfq_queue *new_cfqq; |
cdb16e8f | 149 | struct cfq_group *cfqg; |
ae30c286 | 150 | struct cfq_group *orig_cfqg; |
c4e7893e VG |
151 | /* Number of sectors dispatched from queue in single dispatch round */ |
152 | unsigned long nr_sectors; | |
6118b70b JA |
153 | }; |
154 | ||
c0324a02 | 155 | /* |
718eee05 | 156 | * First index in the service_trees. |
c0324a02 CZ |
157 | * IDLE is handled separately, so it has negative index |
158 | */ | |
159 | enum wl_prio_t { | |
c0324a02 | 160 | BE_WORKLOAD = 0, |
615f0259 VG |
161 | RT_WORKLOAD = 1, |
162 | IDLE_WORKLOAD = 2, | |
c0324a02 CZ |
163 | }; |
164 | ||
718eee05 CZ |
165 | /* |
166 | * Second index in the service_trees. | |
167 | */ | |
168 | enum wl_type_t { | |
169 | ASYNC_WORKLOAD = 0, | |
170 | SYNC_NOIDLE_WORKLOAD = 1, | |
171 | SYNC_WORKLOAD = 2 | |
172 | }; | |
173 | ||
cdb16e8f VG |
174 | /* This is per cgroup per device grouping structure */ |
175 | struct cfq_group { | |
1fa8f6d6 VG |
176 | /* group service_tree member */ |
177 | struct rb_node rb_node; | |
178 | ||
179 | /* group service_tree key */ | |
180 | u64 vdisktime; | |
25bc6b07 | 181 | unsigned int weight; |
1fa8f6d6 VG |
182 | bool on_st; |
183 | ||
184 | /* number of cfqq currently on this group */ | |
185 | int nr_cfqq; | |
186 | ||
58ff82f3 VG |
187 | /* Per group busy queus average. Useful for workload slice calc. */ |
188 | unsigned int busy_queues_avg[2]; | |
cdb16e8f VG |
189 | /* |
190 | * rr lists of queues with requests, onle rr for each priority class. | |
191 | * Counts are embedded in the cfq_rb_root | |
192 | */ | |
193 | struct cfq_rb_root service_trees[2][3]; | |
194 | struct cfq_rb_root service_tree_idle; | |
dae739eb VG |
195 | |
196 | unsigned long saved_workload_slice; | |
197 | enum wl_type_t saved_workload; | |
198 | enum wl_prio_t saved_serving_prio; | |
25fb5169 VG |
199 | struct blkio_group blkg; |
200 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
201 | struct hlist_node cfqd_node; | |
b1c35769 | 202 | atomic_t ref; |
25fb5169 | 203 | #endif |
80bdf0c7 VG |
204 | /* number of requests that are on the dispatch list or inside driver */ |
205 | int dispatched; | |
cdb16e8f | 206 | }; |
718eee05 | 207 | |
22e2c507 JA |
208 | /* |
209 | * Per block device queue structure | |
210 | */ | |
1da177e4 | 211 | struct cfq_data { |
165125e1 | 212 | struct request_queue *queue; |
1fa8f6d6 VG |
213 | /* Root service tree for cfq_groups */ |
214 | struct cfq_rb_root grp_service_tree; | |
cdb16e8f | 215 | struct cfq_group root_group; |
22e2c507 | 216 | |
c0324a02 CZ |
217 | /* |
218 | * The priority currently being served | |
22e2c507 | 219 | */ |
c0324a02 | 220 | enum wl_prio_t serving_prio; |
718eee05 CZ |
221 | enum wl_type_t serving_type; |
222 | unsigned long workload_expires; | |
cdb16e8f | 223 | struct cfq_group *serving_group; |
8e550632 | 224 | bool noidle_tree_requires_idle; |
a36e71f9 JA |
225 | |
226 | /* | |
227 | * Each priority tree is sorted by next_request position. These | |
228 | * trees are used when determining if two or more queues are | |
229 | * interleaving requests (see cfq_close_cooperator). | |
230 | */ | |
231 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
232 | ||
22e2c507 JA |
233 | unsigned int busy_queues; |
234 | ||
53c583d2 CZ |
235 | int rq_in_driver; |
236 | int rq_in_flight[2]; | |
45333d5a AC |
237 | |
238 | /* | |
239 | * queue-depth detection | |
240 | */ | |
241 | int rq_queued; | |
25776e35 | 242 | int hw_tag; |
e459dd08 CZ |
243 | /* |
244 | * hw_tag can be | |
245 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
246 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
247 | * 0 => no NCQ | |
248 | */ | |
249 | int hw_tag_est_depth; | |
250 | unsigned int hw_tag_samples; | |
1da177e4 | 251 | |
22e2c507 JA |
252 | /* |
253 | * idle window management | |
254 | */ | |
255 | struct timer_list idle_slice_timer; | |
23e018a1 | 256 | struct work_struct unplug_work; |
1da177e4 | 257 | |
22e2c507 JA |
258 | struct cfq_queue *active_queue; |
259 | struct cfq_io_context *active_cic; | |
22e2c507 | 260 | |
c2dea2d1 VT |
261 | /* |
262 | * async queue for each priority case | |
263 | */ | |
264 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
265 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 266 | |
6d048f53 | 267 | sector_t last_position; |
1da177e4 | 268 | |
1da177e4 LT |
269 | /* |
270 | * tunables, see top of file | |
271 | */ | |
272 | unsigned int cfq_quantum; | |
22e2c507 | 273 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
274 | unsigned int cfq_back_penalty; |
275 | unsigned int cfq_back_max; | |
22e2c507 JA |
276 | unsigned int cfq_slice[2]; |
277 | unsigned int cfq_slice_async_rq; | |
278 | unsigned int cfq_slice_idle; | |
80bdf0c7 | 279 | unsigned int cfq_group_idle; |
963b72fc | 280 | unsigned int cfq_latency; |
ae30c286 | 281 | unsigned int cfq_group_isolation; |
d9ff4187 | 282 | |
80b15c73 | 283 | unsigned int cic_index; |
d9ff4187 | 284 | struct list_head cic_list; |
1da177e4 | 285 | |
6118b70b JA |
286 | /* |
287 | * Fallback dummy cfqq for extreme OOM conditions | |
288 | */ | |
289 | struct cfq_queue oom_cfqq; | |
365722bb | 290 | |
573412b2 | 291 | unsigned long last_delayed_sync; |
25fb5169 VG |
292 | |
293 | /* List of cfq groups being managed on this device*/ | |
294 | struct hlist_head cfqg_list; | |
bb729bc9 | 295 | struct rcu_head rcu; |
1da177e4 LT |
296 | }; |
297 | ||
25fb5169 VG |
298 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
299 | ||
cdb16e8f VG |
300 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
301 | enum wl_prio_t prio, | |
65b32a57 | 302 | enum wl_type_t type) |
c0324a02 | 303 | { |
1fa8f6d6 VG |
304 | if (!cfqg) |
305 | return NULL; | |
306 | ||
c0324a02 | 307 | if (prio == IDLE_WORKLOAD) |
cdb16e8f | 308 | return &cfqg->service_tree_idle; |
c0324a02 | 309 | |
cdb16e8f | 310 | return &cfqg->service_trees[prio][type]; |
c0324a02 CZ |
311 | } |
312 | ||
3b18152c | 313 | enum cfqq_state_flags { |
b0b8d749 JA |
314 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
315 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 316 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 317 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
318 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
319 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
320 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 321 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 322 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 323 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
ae54abed | 324 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
76280aff | 325 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d | 326 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
3b18152c JA |
327 | }; |
328 | ||
329 | #define CFQ_CFQQ_FNS(name) \ | |
330 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
331 | { \ | |
fe094d98 | 332 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
333 | } \ |
334 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
335 | { \ | |
fe094d98 | 336 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
337 | } \ |
338 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
339 | { \ | |
fe094d98 | 340 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
341 | } |
342 | ||
343 | CFQ_CFQQ_FNS(on_rr); | |
344 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 345 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 346 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
347 | CFQ_CFQQ_FNS(fifo_expire); |
348 | CFQ_CFQQ_FNS(idle_window); | |
349 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 350 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 351 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 352 | CFQ_CFQQ_FNS(coop); |
ae54abed | 353 | CFQ_CFQQ_FNS(split_coop); |
76280aff | 354 | CFQ_CFQQ_FNS(deep); |
f75edf2d | 355 | CFQ_CFQQ_FNS(wait_busy); |
3b18152c JA |
356 | #undef CFQ_CFQQ_FNS |
357 | ||
afc24d49 | 358 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2868ef7b VG |
359 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
360 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ | |
361 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
362 | blkg_path(&(cfqq)->cfqg->blkg), ##args); | |
363 | ||
364 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ | |
365 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ | |
366 | blkg_path(&(cfqg)->blkg), ##args); \ | |
367 | ||
368 | #else | |
7b679138 JA |
369 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
370 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
2868ef7b VG |
371 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0); |
372 | #endif | |
7b679138 JA |
373 | #define cfq_log(cfqd, fmt, args...) \ |
374 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
375 | ||
615f0259 VG |
376 | /* Traverses through cfq group service trees */ |
377 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
378 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
379 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
380 | : &cfqg->service_tree_idle; \ | |
381 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
382 | (i == IDLE_WORKLOAD && j == 0); \ | |
383 | j++, st = i < IDLE_WORKLOAD ? \ | |
384 | &cfqg->service_trees[i][j]: NULL) \ | |
385 | ||
386 | ||
02b35081 VG |
387 | static inline bool iops_mode(struct cfq_data *cfqd) |
388 | { | |
389 | /* | |
390 | * If we are not idling on queues and it is a NCQ drive, parallel | |
391 | * execution of requests is on and measuring time is not possible | |
392 | * in most of the cases until and unless we drive shallower queue | |
393 | * depths and that becomes a performance bottleneck. In such cases | |
394 | * switch to start providing fairness in terms of number of IOs. | |
395 | */ | |
396 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) | |
397 | return true; | |
398 | else | |
399 | return false; | |
400 | } | |
401 | ||
c0324a02 CZ |
402 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
403 | { | |
404 | if (cfq_class_idle(cfqq)) | |
405 | return IDLE_WORKLOAD; | |
406 | if (cfq_class_rt(cfqq)) | |
407 | return RT_WORKLOAD; | |
408 | return BE_WORKLOAD; | |
409 | } | |
410 | ||
718eee05 CZ |
411 | |
412 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
413 | { | |
414 | if (!cfq_cfqq_sync(cfqq)) | |
415 | return ASYNC_WORKLOAD; | |
416 | if (!cfq_cfqq_idle_window(cfqq)) | |
417 | return SYNC_NOIDLE_WORKLOAD; | |
418 | return SYNC_WORKLOAD; | |
419 | } | |
420 | ||
58ff82f3 VG |
421 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
422 | struct cfq_data *cfqd, | |
423 | struct cfq_group *cfqg) | |
c0324a02 CZ |
424 | { |
425 | if (wl == IDLE_WORKLOAD) | |
cdb16e8f | 426 | return cfqg->service_tree_idle.count; |
c0324a02 | 427 | |
cdb16e8f VG |
428 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
429 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count | |
430 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
431 | } |
432 | ||
f26bd1f0 VG |
433 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
434 | struct cfq_group *cfqg) | |
435 | { | |
436 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count | |
437 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; | |
438 | } | |
439 | ||
165125e1 | 440 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
a6151c3a | 441 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool, |
fd0928df | 442 | struct io_context *, gfp_t); |
4ac845a2 | 443 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
91fac317 VT |
444 | struct io_context *); |
445 | ||
446 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, | |
a6151c3a | 447 | bool is_sync) |
91fac317 | 448 | { |
a6151c3a | 449 | return cic->cfqq[is_sync]; |
91fac317 VT |
450 | } |
451 | ||
452 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
a6151c3a | 453 | struct cfq_queue *cfqq, bool is_sync) |
91fac317 | 454 | { |
a6151c3a | 455 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
456 | } |
457 | ||
bca4b914 | 458 | #define CIC_DEAD_KEY 1ul |
80b15c73 | 459 | #define CIC_DEAD_INDEX_SHIFT 1 |
bca4b914 KK |
460 | |
461 | static inline void *cfqd_dead_key(struct cfq_data *cfqd) | |
462 | { | |
80b15c73 | 463 | return (void *)(cfqd->cic_index << CIC_DEAD_INDEX_SHIFT | CIC_DEAD_KEY); |
bca4b914 KK |
464 | } |
465 | ||
466 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_context *cic) | |
467 | { | |
468 | struct cfq_data *cfqd = cic->key; | |
469 | ||
470 | if (unlikely((unsigned long) cfqd & CIC_DEAD_KEY)) | |
471 | return NULL; | |
472 | ||
473 | return cfqd; | |
474 | } | |
475 | ||
91fac317 VT |
476 | /* |
477 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
478 | * set (in which case it could also be direct WRITE). | |
479 | */ | |
a6151c3a | 480 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 481 | { |
7b6d91da | 482 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
91fac317 | 483 | } |
1da177e4 | 484 | |
99f95e52 AM |
485 | /* |
486 | * scheduler run of queue, if there are requests pending and no one in the | |
487 | * driver that will restart queueing | |
488 | */ | |
23e018a1 | 489 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 490 | { |
7b679138 JA |
491 | if (cfqd->busy_queues) { |
492 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 493 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 494 | } |
99f95e52 AM |
495 | } |
496 | ||
165125e1 | 497 | static int cfq_queue_empty(struct request_queue *q) |
99f95e52 AM |
498 | { |
499 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
500 | ||
f04a6424 | 501 | return !cfqd->rq_queued; |
99f95e52 AM |
502 | } |
503 | ||
44f7c160 JA |
504 | /* |
505 | * Scale schedule slice based on io priority. Use the sync time slice only | |
506 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
507 | * io only, should not get full sync slice length. | |
508 | */ | |
a6151c3a | 509 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 510 | unsigned short prio) |
44f7c160 | 511 | { |
d9e7620e | 512 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 513 | |
d9e7620e JA |
514 | WARN_ON(prio >= IOPRIO_BE_NR); |
515 | ||
516 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
517 | } | |
44f7c160 | 518 | |
d9e7620e JA |
519 | static inline int |
520 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
521 | { | |
522 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
523 | } |
524 | ||
25bc6b07 VG |
525 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
526 | { | |
527 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
528 | ||
529 | d = d * BLKIO_WEIGHT_DEFAULT; | |
530 | do_div(d, cfqg->weight); | |
531 | return d; | |
532 | } | |
533 | ||
534 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
535 | { | |
536 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
537 | if (delta > 0) | |
538 | min_vdisktime = vdisktime; | |
539 | ||
540 | return min_vdisktime; | |
541 | } | |
542 | ||
543 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
544 | { | |
545 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
546 | if (delta < 0) | |
547 | min_vdisktime = vdisktime; | |
548 | ||
549 | return min_vdisktime; | |
550 | } | |
551 | ||
552 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
553 | { | |
554 | u64 vdisktime = st->min_vdisktime; | |
555 | struct cfq_group *cfqg; | |
556 | ||
557 | if (st->active) { | |
558 | cfqg = rb_entry_cfqg(st->active); | |
559 | vdisktime = cfqg->vdisktime; | |
560 | } | |
561 | ||
562 | if (st->left) { | |
563 | cfqg = rb_entry_cfqg(st->left); | |
564 | vdisktime = min_vdisktime(vdisktime, cfqg->vdisktime); | |
565 | } | |
566 | ||
567 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, vdisktime); | |
568 | } | |
569 | ||
5db5d642 CZ |
570 | /* |
571 | * get averaged number of queues of RT/BE priority. | |
572 | * average is updated, with a formula that gives more weight to higher numbers, | |
573 | * to quickly follows sudden increases and decrease slowly | |
574 | */ | |
575 | ||
58ff82f3 VG |
576 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
577 | struct cfq_group *cfqg, bool rt) | |
5869619c | 578 | { |
5db5d642 CZ |
579 | unsigned min_q, max_q; |
580 | unsigned mult = cfq_hist_divisor - 1; | |
581 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 582 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 583 | |
58ff82f3 VG |
584 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
585 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
586 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 587 | cfq_hist_divisor; |
58ff82f3 VG |
588 | return cfqg->busy_queues_avg[rt]; |
589 | } | |
590 | ||
591 | static inline unsigned | |
592 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
593 | { | |
594 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
595 | ||
596 | return cfq_target_latency * cfqg->weight / st->total_weight; | |
5db5d642 CZ |
597 | } |
598 | ||
44f7c160 JA |
599 | static inline void |
600 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
601 | { | |
5db5d642 CZ |
602 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
603 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
604 | /* |
605 | * interested queues (we consider only the ones with the same | |
606 | * priority class in the cfq group) | |
607 | */ | |
608 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
609 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
610 | unsigned sync_slice = cfqd->cfq_slice[1]; |
611 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
612 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
613 | ||
614 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
615 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
616 | /* scale low_slice according to IO priority | |
617 | * and sync vs async */ | |
618 | unsigned low_slice = | |
619 | min(slice, base_low_slice * slice / sync_slice); | |
620 | /* the adapted slice value is scaled to fit all iqs | |
621 | * into the target latency */ | |
58ff82f3 | 622 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
623 | low_slice); |
624 | } | |
625 | } | |
dae739eb | 626 | cfqq->slice_start = jiffies; |
5db5d642 | 627 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 628 | cfqq->allocated_slice = slice; |
7b679138 | 629 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
630 | } |
631 | ||
632 | /* | |
633 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
634 | * isn't valid until the first request from the dispatch is activated | |
635 | * and the slice time set. | |
636 | */ | |
a6151c3a | 637 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
638 | { |
639 | if (cfq_cfqq_slice_new(cfqq)) | |
640 | return 0; | |
641 | if (time_before(jiffies, cfqq->slice_end)) | |
642 | return 0; | |
643 | ||
644 | return 1; | |
645 | } | |
646 | ||
1da177e4 | 647 | /* |
5e705374 | 648 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 649 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 650 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 651 | */ |
5e705374 | 652 | static struct request * |
cf7c25cf | 653 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 654 | { |
cf7c25cf | 655 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 656 | unsigned long back_max; |
e8a99053 AM |
657 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
658 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
659 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 660 | |
5e705374 JA |
661 | if (rq1 == NULL || rq1 == rq2) |
662 | return rq2; | |
663 | if (rq2 == NULL) | |
664 | return rq1; | |
9c2c38a1 | 665 | |
5e705374 JA |
666 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
667 | return rq1; | |
668 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
669 | return rq2; | |
7b6d91da | 670 | if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META)) |
374f84ac | 671 | return rq1; |
7b6d91da CH |
672 | else if ((rq2->cmd_flags & REQ_META) && |
673 | !(rq1->cmd_flags & REQ_META)) | |
374f84ac | 674 | return rq2; |
1da177e4 | 675 | |
83096ebf TH |
676 | s1 = blk_rq_pos(rq1); |
677 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 678 | |
1da177e4 LT |
679 | /* |
680 | * by definition, 1KiB is 2 sectors | |
681 | */ | |
682 | back_max = cfqd->cfq_back_max * 2; | |
683 | ||
684 | /* | |
685 | * Strict one way elevator _except_ in the case where we allow | |
686 | * short backward seeks which are biased as twice the cost of a | |
687 | * similar forward seek. | |
688 | */ | |
689 | if (s1 >= last) | |
690 | d1 = s1 - last; | |
691 | else if (s1 + back_max >= last) | |
692 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
693 | else | |
e8a99053 | 694 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
695 | |
696 | if (s2 >= last) | |
697 | d2 = s2 - last; | |
698 | else if (s2 + back_max >= last) | |
699 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
700 | else | |
e8a99053 | 701 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
702 | |
703 | /* Found required data */ | |
e8a99053 AM |
704 | |
705 | /* | |
706 | * By doing switch() on the bit mask "wrap" we avoid having to | |
707 | * check two variables for all permutations: --> faster! | |
708 | */ | |
709 | switch (wrap) { | |
5e705374 | 710 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 711 | if (d1 < d2) |
5e705374 | 712 | return rq1; |
e8a99053 | 713 | else if (d2 < d1) |
5e705374 | 714 | return rq2; |
e8a99053 AM |
715 | else { |
716 | if (s1 >= s2) | |
5e705374 | 717 | return rq1; |
e8a99053 | 718 | else |
5e705374 | 719 | return rq2; |
e8a99053 | 720 | } |
1da177e4 | 721 | |
e8a99053 | 722 | case CFQ_RQ2_WRAP: |
5e705374 | 723 | return rq1; |
e8a99053 | 724 | case CFQ_RQ1_WRAP: |
5e705374 JA |
725 | return rq2; |
726 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
727 | default: |
728 | /* | |
729 | * Since both rqs are wrapped, | |
730 | * start with the one that's further behind head | |
731 | * (--> only *one* back seek required), | |
732 | * since back seek takes more time than forward. | |
733 | */ | |
734 | if (s1 <= s2) | |
5e705374 | 735 | return rq1; |
1da177e4 | 736 | else |
5e705374 | 737 | return rq2; |
1da177e4 LT |
738 | } |
739 | } | |
740 | ||
498d3aa2 JA |
741 | /* |
742 | * The below is leftmost cache rbtree addon | |
743 | */ | |
0871714e | 744 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 745 | { |
615f0259 VG |
746 | /* Service tree is empty */ |
747 | if (!root->count) | |
748 | return NULL; | |
749 | ||
cc09e299 JA |
750 | if (!root->left) |
751 | root->left = rb_first(&root->rb); | |
752 | ||
0871714e JA |
753 | if (root->left) |
754 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
755 | ||
756 | return NULL; | |
cc09e299 JA |
757 | } |
758 | ||
1fa8f6d6 VG |
759 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
760 | { | |
761 | if (!root->left) | |
762 | root->left = rb_first(&root->rb); | |
763 | ||
764 | if (root->left) | |
765 | return rb_entry_cfqg(root->left); | |
766 | ||
767 | return NULL; | |
768 | } | |
769 | ||
a36e71f9 JA |
770 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
771 | { | |
772 | rb_erase(n, root); | |
773 | RB_CLEAR_NODE(n); | |
774 | } | |
775 | ||
cc09e299 JA |
776 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
777 | { | |
778 | if (root->left == n) | |
779 | root->left = NULL; | |
a36e71f9 | 780 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 781 | --root->count; |
cc09e299 JA |
782 | } |
783 | ||
1da177e4 LT |
784 | /* |
785 | * would be nice to take fifo expire time into account as well | |
786 | */ | |
5e705374 JA |
787 | static struct request * |
788 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
789 | struct request *last) | |
1da177e4 | 790 | { |
21183b07 JA |
791 | struct rb_node *rbnext = rb_next(&last->rb_node); |
792 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 793 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 794 | |
21183b07 | 795 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
796 | |
797 | if (rbprev) | |
5e705374 | 798 | prev = rb_entry_rq(rbprev); |
1da177e4 | 799 | |
21183b07 | 800 | if (rbnext) |
5e705374 | 801 | next = rb_entry_rq(rbnext); |
21183b07 JA |
802 | else { |
803 | rbnext = rb_first(&cfqq->sort_list); | |
804 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 805 | next = rb_entry_rq(rbnext); |
21183b07 | 806 | } |
1da177e4 | 807 | |
cf7c25cf | 808 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
809 | } |
810 | ||
d9e7620e JA |
811 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
812 | struct cfq_queue *cfqq) | |
1da177e4 | 813 | { |
d9e7620e JA |
814 | /* |
815 | * just an approximation, should be ok. | |
816 | */ | |
cdb16e8f | 817 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 818 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
819 | } |
820 | ||
1fa8f6d6 VG |
821 | static inline s64 |
822 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
823 | { | |
824 | return cfqg->vdisktime - st->min_vdisktime; | |
825 | } | |
826 | ||
827 | static void | |
828 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
829 | { | |
830 | struct rb_node **node = &st->rb.rb_node; | |
831 | struct rb_node *parent = NULL; | |
832 | struct cfq_group *__cfqg; | |
833 | s64 key = cfqg_key(st, cfqg); | |
834 | int left = 1; | |
835 | ||
836 | while (*node != NULL) { | |
837 | parent = *node; | |
838 | __cfqg = rb_entry_cfqg(parent); | |
839 | ||
840 | if (key < cfqg_key(st, __cfqg)) | |
841 | node = &parent->rb_left; | |
842 | else { | |
843 | node = &parent->rb_right; | |
844 | left = 0; | |
845 | } | |
846 | } | |
847 | ||
848 | if (left) | |
849 | st->left = &cfqg->rb_node; | |
850 | ||
851 | rb_link_node(&cfqg->rb_node, parent, node); | |
852 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
853 | } | |
854 | ||
855 | static void | |
856 | cfq_group_service_tree_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
857 | { | |
858 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
859 | struct cfq_group *__cfqg; | |
860 | struct rb_node *n; | |
861 | ||
862 | cfqg->nr_cfqq++; | |
863 | if (cfqg->on_st) | |
864 | return; | |
865 | ||
866 | /* | |
867 | * Currently put the group at the end. Later implement something | |
868 | * so that groups get lesser vtime based on their weights, so that | |
869 | * if group does not loose all if it was not continously backlogged. | |
870 | */ | |
871 | n = rb_last(&st->rb); | |
872 | if (n) { | |
873 | __cfqg = rb_entry_cfqg(n); | |
874 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
875 | } else | |
876 | cfqg->vdisktime = st->min_vdisktime; | |
877 | ||
878 | __cfq_group_service_tree_add(st, cfqg); | |
879 | cfqg->on_st = true; | |
58ff82f3 | 880 | st->total_weight += cfqg->weight; |
1fa8f6d6 VG |
881 | } |
882 | ||
883 | static void | |
884 | cfq_group_service_tree_del(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
885 | { | |
886 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
887 | ||
25bc6b07 VG |
888 | if (st->active == &cfqg->rb_node) |
889 | st->active = NULL; | |
890 | ||
1fa8f6d6 VG |
891 | BUG_ON(cfqg->nr_cfqq < 1); |
892 | cfqg->nr_cfqq--; | |
25bc6b07 | 893 | |
1fa8f6d6 VG |
894 | /* If there are other cfq queues under this group, don't delete it */ |
895 | if (cfqg->nr_cfqq) | |
896 | return; | |
897 | ||
2868ef7b | 898 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
1fa8f6d6 | 899 | cfqg->on_st = false; |
58ff82f3 | 900 | st->total_weight -= cfqg->weight; |
1fa8f6d6 VG |
901 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
902 | cfq_rb_erase(&cfqg->rb_node, st); | |
dae739eb | 903 | cfqg->saved_workload_slice = 0; |
e98ef89b | 904 | cfq_blkiocg_update_dequeue_stats(&cfqg->blkg, 1); |
dae739eb VG |
905 | } |
906 | ||
907 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq) | |
908 | { | |
f75edf2d | 909 | unsigned int slice_used; |
dae739eb VG |
910 | |
911 | /* | |
912 | * Queue got expired before even a single request completed or | |
913 | * got expired immediately after first request completion. | |
914 | */ | |
915 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
916 | /* | |
917 | * Also charge the seek time incurred to the group, otherwise | |
918 | * if there are mutiple queues in the group, each can dispatch | |
919 | * a single request on seeky media and cause lots of seek time | |
920 | * and group will never know it. | |
921 | */ | |
922 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
923 | 1); | |
924 | } else { | |
925 | slice_used = jiffies - cfqq->slice_start; | |
f75edf2d VG |
926 | if (slice_used > cfqq->allocated_slice) |
927 | slice_used = cfqq->allocated_slice; | |
dae739eb VG |
928 | } |
929 | ||
dae739eb VG |
930 | return slice_used; |
931 | } | |
932 | ||
933 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
e5ff082e | 934 | struct cfq_queue *cfqq) |
dae739eb VG |
935 | { |
936 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
02b35081 | 937 | unsigned int used_sl, charge; |
f26bd1f0 VG |
938 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
939 | - cfqg->service_tree_idle.count; | |
940 | ||
941 | BUG_ON(nr_sync < 0); | |
02b35081 | 942 | used_sl = charge = cfq_cfqq_slice_usage(cfqq); |
dae739eb | 943 | |
02b35081 VG |
944 | if (iops_mode(cfqd)) |
945 | charge = cfqq->slice_dispatch; | |
946 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) | |
947 | charge = cfqq->allocated_slice; | |
dae739eb VG |
948 | |
949 | /* Can't update vdisktime while group is on service tree */ | |
950 | cfq_rb_erase(&cfqg->rb_node, st); | |
02b35081 | 951 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
dae739eb VG |
952 | __cfq_group_service_tree_add(st, cfqg); |
953 | ||
954 | /* This group is being expired. Save the context */ | |
955 | if (time_after(cfqd->workload_expires, jiffies)) { | |
956 | cfqg->saved_workload_slice = cfqd->workload_expires | |
957 | - jiffies; | |
958 | cfqg->saved_workload = cfqd->serving_type; | |
959 | cfqg->saved_serving_prio = cfqd->serving_prio; | |
960 | } else | |
961 | cfqg->saved_workload_slice = 0; | |
2868ef7b VG |
962 | |
963 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
964 | st->min_vdisktime); | |
c4e7893e VG |
965 | cfq_log_cfqq(cfqq->cfqd, cfqq, "sl_used=%u disp=%u charge=%u iops=%u" |
966 | " sect=%u", used_sl, cfqq->slice_dispatch, charge, | |
967 | iops_mode(cfqd), cfqq->nr_sectors); | |
e98ef89b VG |
968 | cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl); |
969 | cfq_blkiocg_set_start_empty_time(&cfqg->blkg); | |
1fa8f6d6 VG |
970 | } |
971 | ||
25fb5169 VG |
972 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
973 | static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg) | |
974 | { | |
975 | if (blkg) | |
976 | return container_of(blkg, struct cfq_group, blkg); | |
977 | return NULL; | |
978 | } | |
979 | ||
f8d461d6 VG |
980 | void |
981 | cfq_update_blkio_group_weight(struct blkio_group *blkg, unsigned int weight) | |
982 | { | |
983 | cfqg_of_blkg(blkg)->weight = weight; | |
984 | } | |
985 | ||
25fb5169 VG |
986 | static struct cfq_group * |
987 | cfq_find_alloc_cfqg(struct cfq_data *cfqd, struct cgroup *cgroup, int create) | |
988 | { | |
989 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
990 | struct cfq_group *cfqg = NULL; | |
991 | void *key = cfqd; | |
992 | int i, j; | |
993 | struct cfq_rb_root *st; | |
22084190 VG |
994 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; |
995 | unsigned int major, minor; | |
25fb5169 | 996 | |
25fb5169 | 997 | cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key)); |
a74b2ada RB |
998 | if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) { |
999 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); | |
1000 | cfqg->blkg.dev = MKDEV(major, minor); | |
1001 | goto done; | |
1002 | } | |
25fb5169 VG |
1003 | if (cfqg || !create) |
1004 | goto done; | |
1005 | ||
1006 | cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node); | |
1007 | if (!cfqg) | |
1008 | goto done; | |
1009 | ||
25fb5169 VG |
1010 | for_each_cfqg_st(cfqg, i, j, st) |
1011 | *st = CFQ_RB_ROOT; | |
1012 | RB_CLEAR_NODE(&cfqg->rb_node); | |
1013 | ||
b1c35769 VG |
1014 | /* |
1015 | * Take the initial reference that will be released on destroy | |
1016 | * This can be thought of a joint reference by cgroup and | |
1017 | * elevator which will be dropped by either elevator exit | |
1018 | * or cgroup deletion path depending on who is exiting first. | |
1019 | */ | |
1020 | atomic_set(&cfqg->ref, 1); | |
1021 | ||
25fb5169 | 1022 | /* Add group onto cgroup list */ |
22084190 | 1023 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); |
e98ef89b | 1024 | cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd, |
22084190 | 1025 | MKDEV(major, minor)); |
34d0f179 | 1026 | cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev); |
25fb5169 VG |
1027 | |
1028 | /* Add group on cfqd list */ | |
1029 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); | |
1030 | ||
1031 | done: | |
25fb5169 VG |
1032 | return cfqg; |
1033 | } | |
1034 | ||
1035 | /* | |
1036 | * Search for the cfq group current task belongs to. If create = 1, then also | |
1037 | * create the cfq group if it does not exist. request_queue lock must be held. | |
1038 | */ | |
1039 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create) | |
1040 | { | |
1041 | struct cgroup *cgroup; | |
1042 | struct cfq_group *cfqg = NULL; | |
1043 | ||
1044 | rcu_read_lock(); | |
1045 | cgroup = task_cgroup(current, blkio_subsys_id); | |
1046 | cfqg = cfq_find_alloc_cfqg(cfqd, cgroup, create); | |
1047 | if (!cfqg && create) | |
1048 | cfqg = &cfqd->root_group; | |
1049 | rcu_read_unlock(); | |
1050 | return cfqg; | |
1051 | } | |
1052 | ||
7f1dc8a2 VG |
1053 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) |
1054 | { | |
1055 | atomic_inc(&cfqg->ref); | |
1056 | return cfqg; | |
1057 | } | |
1058 | ||
25fb5169 VG |
1059 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) |
1060 | { | |
1061 | /* Currently, all async queues are mapped to root group */ | |
1062 | if (!cfq_cfqq_sync(cfqq)) | |
1063 | cfqg = &cfqq->cfqd->root_group; | |
1064 | ||
1065 | cfqq->cfqg = cfqg; | |
b1c35769 VG |
1066 | /* cfqq reference on cfqg */ |
1067 | atomic_inc(&cfqq->cfqg->ref); | |
1068 | } | |
1069 | ||
1070 | static void cfq_put_cfqg(struct cfq_group *cfqg) | |
1071 | { | |
1072 | struct cfq_rb_root *st; | |
1073 | int i, j; | |
1074 | ||
1075 | BUG_ON(atomic_read(&cfqg->ref) <= 0); | |
1076 | if (!atomic_dec_and_test(&cfqg->ref)) | |
1077 | return; | |
1078 | for_each_cfqg_st(cfqg, i, j, st) | |
1079 | BUG_ON(!RB_EMPTY_ROOT(&st->rb) || st->active != NULL); | |
1080 | kfree(cfqg); | |
1081 | } | |
1082 | ||
1083 | static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1084 | { | |
1085 | /* Something wrong if we are trying to remove same group twice */ | |
1086 | BUG_ON(hlist_unhashed(&cfqg->cfqd_node)); | |
1087 | ||
1088 | hlist_del_init(&cfqg->cfqd_node); | |
1089 | ||
1090 | /* | |
1091 | * Put the reference taken at the time of creation so that when all | |
1092 | * queues are gone, group can be destroyed. | |
1093 | */ | |
1094 | cfq_put_cfqg(cfqg); | |
1095 | } | |
1096 | ||
1097 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) | |
1098 | { | |
1099 | struct hlist_node *pos, *n; | |
1100 | struct cfq_group *cfqg; | |
1101 | ||
1102 | hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) { | |
1103 | /* | |
1104 | * If cgroup removal path got to blk_group first and removed | |
1105 | * it from cgroup list, then it will take care of destroying | |
1106 | * cfqg also. | |
1107 | */ | |
e98ef89b | 1108 | if (!cfq_blkiocg_del_blkio_group(&cfqg->blkg)) |
b1c35769 VG |
1109 | cfq_destroy_cfqg(cfqd, cfqg); |
1110 | } | |
25fb5169 | 1111 | } |
b1c35769 VG |
1112 | |
1113 | /* | |
1114 | * Blk cgroup controller notification saying that blkio_group object is being | |
1115 | * delinked as associated cgroup object is going away. That also means that | |
1116 | * no new IO will come in this group. So get rid of this group as soon as | |
1117 | * any pending IO in the group is finished. | |
1118 | * | |
1119 | * This function is called under rcu_read_lock(). key is the rcu protected | |
1120 | * pointer. That means "key" is a valid cfq_data pointer as long as we are rcu | |
1121 | * read lock. | |
1122 | * | |
1123 | * "key" was fetched from blkio_group under blkio_cgroup->lock. That means | |
1124 | * it should not be NULL as even if elevator was exiting, cgroup deltion | |
1125 | * path got to it first. | |
1126 | */ | |
1127 | void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg) | |
1128 | { | |
1129 | unsigned long flags; | |
1130 | struct cfq_data *cfqd = key; | |
1131 | ||
1132 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1133 | cfq_destroy_cfqg(cfqd, cfqg_of_blkg(blkg)); | |
1134 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1135 | } | |
1136 | ||
25fb5169 VG |
1137 | #else /* GROUP_IOSCHED */ |
1138 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create) | |
1139 | { | |
1140 | return &cfqd->root_group; | |
1141 | } | |
7f1dc8a2 VG |
1142 | |
1143 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) | |
1144 | { | |
50eaeb32 | 1145 | return cfqg; |
7f1dc8a2 VG |
1146 | } |
1147 | ||
25fb5169 VG |
1148 | static inline void |
1149 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1150 | cfqq->cfqg = cfqg; | |
1151 | } | |
1152 | ||
b1c35769 VG |
1153 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) {} |
1154 | static inline void cfq_put_cfqg(struct cfq_group *cfqg) {} | |
1155 | ||
25fb5169 VG |
1156 | #endif /* GROUP_IOSCHED */ |
1157 | ||
498d3aa2 | 1158 | /* |
c0324a02 | 1159 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1160 | * requests waiting to be processed. It is sorted in the order that |
1161 | * we will service the queues. | |
1162 | */ | |
a36e71f9 | 1163 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1164 | bool add_front) |
d9e7620e | 1165 | { |
0871714e JA |
1166 | struct rb_node **p, *parent; |
1167 | struct cfq_queue *__cfqq; | |
d9e7620e | 1168 | unsigned long rb_key; |
c0324a02 | 1169 | struct cfq_rb_root *service_tree; |
498d3aa2 | 1170 | int left; |
dae739eb | 1171 | int new_cfqq = 1; |
ae30c286 VG |
1172 | int group_changed = 0; |
1173 | ||
1174 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
1175 | if (!cfqd->cfq_group_isolation | |
1176 | && cfqq_type(cfqq) == SYNC_NOIDLE_WORKLOAD | |
1177 | && cfqq->cfqg && cfqq->cfqg != &cfqd->root_group) { | |
1178 | /* Move this cfq to root group */ | |
1179 | cfq_log_cfqq(cfqd, cfqq, "moving to root group"); | |
1180 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) | |
1181 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); | |
1182 | cfqq->orig_cfqg = cfqq->cfqg; | |
1183 | cfqq->cfqg = &cfqd->root_group; | |
1184 | atomic_inc(&cfqd->root_group.ref); | |
1185 | group_changed = 1; | |
1186 | } else if (!cfqd->cfq_group_isolation | |
1187 | && cfqq_type(cfqq) == SYNC_WORKLOAD && cfqq->orig_cfqg) { | |
1188 | /* cfqq is sequential now needs to go to its original group */ | |
1189 | BUG_ON(cfqq->cfqg != &cfqd->root_group); | |
1190 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) | |
1191 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); | |
1192 | cfq_put_cfqg(cfqq->cfqg); | |
1193 | cfqq->cfqg = cfqq->orig_cfqg; | |
1194 | cfqq->orig_cfqg = NULL; | |
1195 | group_changed = 1; | |
1196 | cfq_log_cfqq(cfqd, cfqq, "moved to origin group"); | |
1197 | } | |
1198 | #endif | |
d9e7620e | 1199 | |
cdb16e8f | 1200 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
65b32a57 | 1201 | cfqq_type(cfqq)); |
0871714e JA |
1202 | if (cfq_class_idle(cfqq)) { |
1203 | rb_key = CFQ_IDLE_DELAY; | |
aa6f6a3d | 1204 | parent = rb_last(&service_tree->rb); |
0871714e JA |
1205 | if (parent && parent != &cfqq->rb_node) { |
1206 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1207 | rb_key += __cfqq->rb_key; | |
1208 | } else | |
1209 | rb_key += jiffies; | |
1210 | } else if (!add_front) { | |
b9c8946b JA |
1211 | /* |
1212 | * Get our rb key offset. Subtract any residual slice | |
1213 | * value carried from last service. A negative resid | |
1214 | * count indicates slice overrun, and this should position | |
1215 | * the next service time further away in the tree. | |
1216 | */ | |
edd75ffd | 1217 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1218 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1219 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1220 | } else { |
1221 | rb_key = -HZ; | |
aa6f6a3d | 1222 | __cfqq = cfq_rb_first(service_tree); |
48e025e6 CZ |
1223 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1224 | } | |
1da177e4 | 1225 | |
d9e7620e | 1226 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1227 | new_cfqq = 0; |
99f9628a | 1228 | /* |
d9e7620e | 1229 | * same position, nothing more to do |
99f9628a | 1230 | */ |
c0324a02 CZ |
1231 | if (rb_key == cfqq->rb_key && |
1232 | cfqq->service_tree == service_tree) | |
d9e7620e | 1233 | return; |
1da177e4 | 1234 | |
aa6f6a3d CZ |
1235 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1236 | cfqq->service_tree = NULL; | |
1da177e4 | 1237 | } |
d9e7620e | 1238 | |
498d3aa2 | 1239 | left = 1; |
0871714e | 1240 | parent = NULL; |
aa6f6a3d CZ |
1241 | cfqq->service_tree = service_tree; |
1242 | p = &service_tree->rb.rb_node; | |
d9e7620e | 1243 | while (*p) { |
67060e37 | 1244 | struct rb_node **n; |
cc09e299 | 1245 | |
d9e7620e JA |
1246 | parent = *p; |
1247 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1248 | ||
0c534e0a | 1249 | /* |
c0324a02 | 1250 | * sort by key, that represents service time. |
0c534e0a | 1251 | */ |
c0324a02 | 1252 | if (time_before(rb_key, __cfqq->rb_key)) |
67060e37 | 1253 | n = &(*p)->rb_left; |
c0324a02 | 1254 | else { |
67060e37 | 1255 | n = &(*p)->rb_right; |
cc09e299 | 1256 | left = 0; |
c0324a02 | 1257 | } |
67060e37 JA |
1258 | |
1259 | p = n; | |
d9e7620e JA |
1260 | } |
1261 | ||
cc09e299 | 1262 | if (left) |
aa6f6a3d | 1263 | service_tree->left = &cfqq->rb_node; |
cc09e299 | 1264 | |
d9e7620e JA |
1265 | cfqq->rb_key = rb_key; |
1266 | rb_link_node(&cfqq->rb_node, parent, p); | |
aa6f6a3d CZ |
1267 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
1268 | service_tree->count++; | |
ae30c286 | 1269 | if ((add_front || !new_cfqq) && !group_changed) |
dae739eb | 1270 | return; |
1fa8f6d6 | 1271 | cfq_group_service_tree_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1272 | } |
1273 | ||
a36e71f9 | 1274 | static struct cfq_queue * |
f2d1f0ae JA |
1275 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1276 | sector_t sector, struct rb_node **ret_parent, | |
1277 | struct rb_node ***rb_link) | |
a36e71f9 | 1278 | { |
a36e71f9 JA |
1279 | struct rb_node **p, *parent; |
1280 | struct cfq_queue *cfqq = NULL; | |
1281 | ||
1282 | parent = NULL; | |
1283 | p = &root->rb_node; | |
1284 | while (*p) { | |
1285 | struct rb_node **n; | |
1286 | ||
1287 | parent = *p; | |
1288 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1289 | ||
1290 | /* | |
1291 | * Sort strictly based on sector. Smallest to the left, | |
1292 | * largest to the right. | |
1293 | */ | |
2e46e8b2 | 1294 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1295 | n = &(*p)->rb_right; |
2e46e8b2 | 1296 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1297 | n = &(*p)->rb_left; |
1298 | else | |
1299 | break; | |
1300 | p = n; | |
3ac6c9f8 | 1301 | cfqq = NULL; |
a36e71f9 JA |
1302 | } |
1303 | ||
1304 | *ret_parent = parent; | |
1305 | if (rb_link) | |
1306 | *rb_link = p; | |
3ac6c9f8 | 1307 | return cfqq; |
a36e71f9 JA |
1308 | } |
1309 | ||
1310 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1311 | { | |
a36e71f9 JA |
1312 | struct rb_node **p, *parent; |
1313 | struct cfq_queue *__cfqq; | |
1314 | ||
f2d1f0ae JA |
1315 | if (cfqq->p_root) { |
1316 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1317 | cfqq->p_root = NULL; | |
1318 | } | |
a36e71f9 JA |
1319 | |
1320 | if (cfq_class_idle(cfqq)) | |
1321 | return; | |
1322 | if (!cfqq->next_rq) | |
1323 | return; | |
1324 | ||
f2d1f0ae | 1325 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1326 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1327 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1328 | if (!__cfqq) { |
1329 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1330 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1331 | } else | |
1332 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1333 | } |
1334 | ||
498d3aa2 JA |
1335 | /* |
1336 | * Update cfqq's position in the service tree. | |
1337 | */ | |
edd75ffd | 1338 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1339 | { |
6d048f53 JA |
1340 | /* |
1341 | * Resorting requires the cfqq to be on the RR list already. | |
1342 | */ | |
a36e71f9 | 1343 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1344 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1345 | cfq_prio_tree_add(cfqd, cfqq); |
1346 | } | |
6d048f53 JA |
1347 | } |
1348 | ||
1da177e4 LT |
1349 | /* |
1350 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1351 | * the pending list according to last request service |
1da177e4 | 1352 | */ |
febffd61 | 1353 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1354 | { |
7b679138 | 1355 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1356 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1357 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
1358 | cfqd->busy_queues++; |
1359 | ||
edd75ffd | 1360 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1361 | } |
1362 | ||
498d3aa2 JA |
1363 | /* |
1364 | * Called when the cfqq no longer has requests pending, remove it from | |
1365 | * the service tree. | |
1366 | */ | |
febffd61 | 1367 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1368 | { |
7b679138 | 1369 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1370 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1371 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1372 | |
aa6f6a3d CZ |
1373 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1374 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1375 | cfqq->service_tree = NULL; | |
1376 | } | |
f2d1f0ae JA |
1377 | if (cfqq->p_root) { |
1378 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1379 | cfqq->p_root = NULL; | |
1380 | } | |
d9e7620e | 1381 | |
1fa8f6d6 | 1382 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1383 | BUG_ON(!cfqd->busy_queues); |
1384 | cfqd->busy_queues--; | |
1385 | } | |
1386 | ||
1387 | /* | |
1388 | * rb tree support functions | |
1389 | */ | |
febffd61 | 1390 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1391 | { |
5e705374 | 1392 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1393 | const int sync = rq_is_sync(rq); |
1da177e4 | 1394 | |
b4878f24 JA |
1395 | BUG_ON(!cfqq->queued[sync]); |
1396 | cfqq->queued[sync]--; | |
1da177e4 | 1397 | |
5e705374 | 1398 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1399 | |
f04a6424 VG |
1400 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1401 | /* | |
1402 | * Queue will be deleted from service tree when we actually | |
1403 | * expire it later. Right now just remove it from prio tree | |
1404 | * as it is empty. | |
1405 | */ | |
1406 | if (cfqq->p_root) { | |
1407 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1408 | cfqq->p_root = NULL; | |
1409 | } | |
1410 | } | |
1da177e4 LT |
1411 | } |
1412 | ||
5e705374 | 1413 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1414 | { |
5e705374 | 1415 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1416 | struct cfq_data *cfqd = cfqq->cfqd; |
a36e71f9 | 1417 | struct request *__alias, *prev; |
1da177e4 | 1418 | |
5380a101 | 1419 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
1420 | |
1421 | /* | |
1422 | * looks a little odd, but the first insert might return an alias. | |
1423 | * if that happens, put the alias on the dispatch list | |
1424 | */ | |
21183b07 | 1425 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 1426 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
1427 | |
1428 | if (!cfq_cfqq_on_rr(cfqq)) | |
1429 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1430 | |
1431 | /* | |
1432 | * check if this request is a better next-serve candidate | |
1433 | */ | |
a36e71f9 | 1434 | prev = cfqq->next_rq; |
cf7c25cf | 1435 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1436 | |
1437 | /* | |
1438 | * adjust priority tree position, if ->next_rq changes | |
1439 | */ | |
1440 | if (prev != cfqq->next_rq) | |
1441 | cfq_prio_tree_add(cfqd, cfqq); | |
1442 | ||
5044eed4 | 1443 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1444 | } |
1445 | ||
febffd61 | 1446 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1447 | { |
5380a101 JA |
1448 | elv_rb_del(&cfqq->sort_list, rq); |
1449 | cfqq->queued[rq_is_sync(rq)]--; | |
e98ef89b VG |
1450 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, |
1451 | rq_data_dir(rq), rq_is_sync(rq)); | |
5e705374 | 1452 | cfq_add_rq_rb(rq); |
e98ef89b | 1453 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, |
7f1dc8a2 VG |
1454 | &cfqq->cfqd->serving_group->blkg, rq_data_dir(rq), |
1455 | rq_is_sync(rq)); | |
1da177e4 LT |
1456 | } |
1457 | ||
206dc69b JA |
1458 | static struct request * |
1459 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1460 | { |
206dc69b | 1461 | struct task_struct *tsk = current; |
91fac317 | 1462 | struct cfq_io_context *cic; |
206dc69b | 1463 | struct cfq_queue *cfqq; |
1da177e4 | 1464 | |
4ac845a2 | 1465 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1466 | if (!cic) |
1467 | return NULL; | |
1468 | ||
1469 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
1470 | if (cfqq) { |
1471 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
1472 | ||
21183b07 | 1473 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 1474 | } |
1da177e4 | 1475 | |
1da177e4 LT |
1476 | return NULL; |
1477 | } | |
1478 | ||
165125e1 | 1479 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1480 | { |
22e2c507 | 1481 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 1482 | |
53c583d2 | 1483 | cfqd->rq_in_driver++; |
7b679138 | 1484 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
53c583d2 | 1485 | cfqd->rq_in_driver); |
25776e35 | 1486 | |
5b93629b | 1487 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
1488 | } |
1489 | ||
165125e1 | 1490 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1491 | { |
b4878f24 JA |
1492 | struct cfq_data *cfqd = q->elevator->elevator_data; |
1493 | ||
53c583d2 CZ |
1494 | WARN_ON(!cfqd->rq_in_driver); |
1495 | cfqd->rq_in_driver--; | |
7b679138 | 1496 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
53c583d2 | 1497 | cfqd->rq_in_driver); |
1da177e4 LT |
1498 | } |
1499 | ||
b4878f24 | 1500 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 1501 | { |
5e705374 | 1502 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 1503 | |
5e705374 JA |
1504 | if (cfqq->next_rq == rq) |
1505 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 1506 | |
b4878f24 | 1507 | list_del_init(&rq->queuelist); |
5e705374 | 1508 | cfq_del_rq_rb(rq); |
374f84ac | 1509 | |
45333d5a | 1510 | cfqq->cfqd->rq_queued--; |
e98ef89b VG |
1511 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, |
1512 | rq_data_dir(rq), rq_is_sync(rq)); | |
7b6d91da | 1513 | if (rq->cmd_flags & REQ_META) { |
374f84ac JA |
1514 | WARN_ON(!cfqq->meta_pending); |
1515 | cfqq->meta_pending--; | |
1516 | } | |
1da177e4 LT |
1517 | } |
1518 | ||
165125e1 JA |
1519 | static int cfq_merge(struct request_queue *q, struct request **req, |
1520 | struct bio *bio) | |
1da177e4 LT |
1521 | { |
1522 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1523 | struct request *__rq; | |
1da177e4 | 1524 | |
206dc69b | 1525 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 1526 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
1527 | *req = __rq; |
1528 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
1529 | } |
1530 | ||
1531 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
1532 | } |
1533 | ||
165125e1 | 1534 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 1535 | int type) |
1da177e4 | 1536 | { |
21183b07 | 1537 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 1538 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 1539 | |
5e705374 | 1540 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 1541 | } |
1da177e4 LT |
1542 | } |
1543 | ||
812d4026 DS |
1544 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
1545 | struct bio *bio) | |
1546 | { | |
e98ef89b VG |
1547 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(req))->blkg, |
1548 | bio_data_dir(bio), cfq_bio_sync(bio)); | |
812d4026 DS |
1549 | } |
1550 | ||
1da177e4 | 1551 | static void |
165125e1 | 1552 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
1553 | struct request *next) |
1554 | { | |
cf7c25cf | 1555 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1556 | /* |
1557 | * reposition in fifo if next is older than rq | |
1558 | */ | |
1559 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
30996f40 | 1560 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
22e2c507 | 1561 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
1562 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1563 | } | |
22e2c507 | 1564 | |
cf7c25cf CZ |
1565 | if (cfqq->next_rq == next) |
1566 | cfqq->next_rq = rq; | |
b4878f24 | 1567 | cfq_remove_request(next); |
e98ef89b VG |
1568 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(rq))->blkg, |
1569 | rq_data_dir(next), rq_is_sync(next)); | |
22e2c507 JA |
1570 | } |
1571 | ||
165125e1 | 1572 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
1573 | struct bio *bio) |
1574 | { | |
1575 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
91fac317 | 1576 | struct cfq_io_context *cic; |
da775265 | 1577 | struct cfq_queue *cfqq; |
da775265 JA |
1578 | |
1579 | /* | |
ec8acb69 | 1580 | * Disallow merge of a sync bio into an async request. |
da775265 | 1581 | */ |
91fac317 | 1582 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 1583 | return false; |
da775265 JA |
1584 | |
1585 | /* | |
719d3402 JA |
1586 | * Lookup the cfqq that this bio will be queued with. Allow |
1587 | * merge only if rq is queued there. | |
da775265 | 1588 | */ |
4ac845a2 | 1589 | cic = cfq_cic_lookup(cfqd, current->io_context); |
91fac317 | 1590 | if (!cic) |
a6151c3a | 1591 | return false; |
719d3402 | 1592 | |
91fac317 | 1593 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 1594 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
1595 | } |
1596 | ||
812df48d DS |
1597 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1598 | { | |
1599 | del_timer(&cfqd->idle_slice_timer); | |
e98ef89b | 1600 | cfq_blkiocg_update_idle_time_stats(&cfqq->cfqg->blkg); |
812df48d DS |
1601 | } |
1602 | ||
febffd61 JA |
1603 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
1604 | struct cfq_queue *cfqq) | |
22e2c507 JA |
1605 | { |
1606 | if (cfqq) { | |
b1ffe737 DS |
1607 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", |
1608 | cfqd->serving_prio, cfqd->serving_type); | |
e98ef89b | 1609 | cfq_blkiocg_update_avg_queue_size_stats(&cfqq->cfqg->blkg); |
dae739eb VG |
1610 | cfqq->slice_start = 0; |
1611 | cfqq->dispatch_start = jiffies; | |
f75edf2d | 1612 | cfqq->allocated_slice = 0; |
22e2c507 | 1613 | cfqq->slice_end = 0; |
2f5cb738 | 1614 | cfqq->slice_dispatch = 0; |
c4e7893e | 1615 | cfqq->nr_sectors = 0; |
2f5cb738 | 1616 | |
2f5cb738 | 1617 | cfq_clear_cfqq_wait_request(cfqq); |
b029195d | 1618 | cfq_clear_cfqq_must_dispatch(cfqq); |
3b18152c JA |
1619 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
1620 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 1621 | cfq_mark_cfqq_slice_new(cfqq); |
2f5cb738 | 1622 | |
812df48d | 1623 | cfq_del_timer(cfqd, cfqq); |
22e2c507 JA |
1624 | } |
1625 | ||
1626 | cfqd->active_queue = cfqq; | |
1627 | } | |
1628 | ||
7b14e3b5 JA |
1629 | /* |
1630 | * current cfqq expired its slice (or was too idle), select new one | |
1631 | */ | |
1632 | static void | |
1633 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
e5ff082e | 1634 | bool timed_out) |
7b14e3b5 | 1635 | { |
7b679138 JA |
1636 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
1637 | ||
7b14e3b5 | 1638 | if (cfq_cfqq_wait_request(cfqq)) |
812df48d | 1639 | cfq_del_timer(cfqd, cfqq); |
7b14e3b5 | 1640 | |
7b14e3b5 | 1641 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d | 1642 | cfq_clear_cfqq_wait_busy(cfqq); |
7b14e3b5 | 1643 | |
ae54abed SL |
1644 | /* |
1645 | * If this cfqq is shared between multiple processes, check to | |
1646 | * make sure that those processes are still issuing I/Os within | |
1647 | * the mean seek distance. If not, it may be time to break the | |
1648 | * queues apart again. | |
1649 | */ | |
1650 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) | |
1651 | cfq_mark_cfqq_split_coop(cfqq); | |
1652 | ||
7b14e3b5 | 1653 | /* |
6084cdda | 1654 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 1655 | */ |
7b679138 | 1656 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) { |
c5b680f3 | 1657 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b679138 JA |
1658 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
1659 | } | |
7b14e3b5 | 1660 | |
e5ff082e | 1661 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
dae739eb | 1662 | |
f04a6424 VG |
1663 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1664 | cfq_del_cfqq_rr(cfqd, cfqq); | |
1665 | ||
edd75ffd | 1666 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
1667 | |
1668 | if (cfqq == cfqd->active_queue) | |
1669 | cfqd->active_queue = NULL; | |
1670 | ||
dae739eb VG |
1671 | if (&cfqq->cfqg->rb_node == cfqd->grp_service_tree.active) |
1672 | cfqd->grp_service_tree.active = NULL; | |
1673 | ||
7b14e3b5 JA |
1674 | if (cfqd->active_cic) { |
1675 | put_io_context(cfqd->active_cic->ioc); | |
1676 | cfqd->active_cic = NULL; | |
1677 | } | |
7b14e3b5 JA |
1678 | } |
1679 | ||
e5ff082e | 1680 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
1681 | { |
1682 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1683 | ||
1684 | if (cfqq) | |
e5ff082e | 1685 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
1686 | } |
1687 | ||
498d3aa2 JA |
1688 | /* |
1689 | * Get next queue for service. Unless we have a queue preemption, | |
1690 | * we'll simply select the first cfqq in the service tree. | |
1691 | */ | |
6d048f53 | 1692 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 1693 | { |
c0324a02 | 1694 | struct cfq_rb_root *service_tree = |
cdb16e8f | 1695 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
65b32a57 | 1696 | cfqd->serving_type); |
d9e7620e | 1697 | |
f04a6424 VG |
1698 | if (!cfqd->rq_queued) |
1699 | return NULL; | |
1700 | ||
1fa8f6d6 VG |
1701 | /* There is nothing to dispatch */ |
1702 | if (!service_tree) | |
1703 | return NULL; | |
c0324a02 CZ |
1704 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
1705 | return NULL; | |
1706 | return cfq_rb_first(service_tree); | |
6d048f53 JA |
1707 | } |
1708 | ||
f04a6424 VG |
1709 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
1710 | { | |
25fb5169 | 1711 | struct cfq_group *cfqg; |
f04a6424 VG |
1712 | struct cfq_queue *cfqq; |
1713 | int i, j; | |
1714 | struct cfq_rb_root *st; | |
1715 | ||
1716 | if (!cfqd->rq_queued) | |
1717 | return NULL; | |
1718 | ||
25fb5169 VG |
1719 | cfqg = cfq_get_next_cfqg(cfqd); |
1720 | if (!cfqg) | |
1721 | return NULL; | |
1722 | ||
f04a6424 VG |
1723 | for_each_cfqg_st(cfqg, i, j, st) |
1724 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
1725 | return cfqq; | |
1726 | return NULL; | |
1727 | } | |
1728 | ||
498d3aa2 JA |
1729 | /* |
1730 | * Get and set a new active queue for service. | |
1731 | */ | |
a36e71f9 JA |
1732 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
1733 | struct cfq_queue *cfqq) | |
6d048f53 | 1734 | { |
e00ef799 | 1735 | if (!cfqq) |
a36e71f9 | 1736 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 1737 | |
22e2c507 | 1738 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 1739 | return cfqq; |
22e2c507 JA |
1740 | } |
1741 | ||
d9e7620e JA |
1742 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
1743 | struct request *rq) | |
1744 | { | |
83096ebf TH |
1745 | if (blk_rq_pos(rq) >= cfqd->last_position) |
1746 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 1747 | else |
83096ebf | 1748 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
1749 | } |
1750 | ||
b2c18e1e | 1751 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
e9ce335d | 1752 | struct request *rq) |
6d048f53 | 1753 | { |
e9ce335d | 1754 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
6d048f53 JA |
1755 | } |
1756 | ||
a36e71f9 JA |
1757 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
1758 | struct cfq_queue *cur_cfqq) | |
1759 | { | |
f2d1f0ae | 1760 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
1761 | struct rb_node *parent, *node; |
1762 | struct cfq_queue *__cfqq; | |
1763 | sector_t sector = cfqd->last_position; | |
1764 | ||
1765 | if (RB_EMPTY_ROOT(root)) | |
1766 | return NULL; | |
1767 | ||
1768 | /* | |
1769 | * First, if we find a request starting at the end of the last | |
1770 | * request, choose it. | |
1771 | */ | |
f2d1f0ae | 1772 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
1773 | if (__cfqq) |
1774 | return __cfqq; | |
1775 | ||
1776 | /* | |
1777 | * If the exact sector wasn't found, the parent of the NULL leaf | |
1778 | * will contain the closest sector. | |
1779 | */ | |
1780 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
e9ce335d | 1781 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1782 | return __cfqq; |
1783 | ||
2e46e8b2 | 1784 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
1785 | node = rb_next(&__cfqq->p_node); |
1786 | else | |
1787 | node = rb_prev(&__cfqq->p_node); | |
1788 | if (!node) | |
1789 | return NULL; | |
1790 | ||
1791 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
e9ce335d | 1792 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1793 | return __cfqq; |
1794 | ||
1795 | return NULL; | |
1796 | } | |
1797 | ||
1798 | /* | |
1799 | * cfqd - obvious | |
1800 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
1801 | * closely cooperating with itself. | |
1802 | * | |
1803 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
1804 | * one request, and that cfqd->last_position reflects a position on the disk | |
1805 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
1806 | * assumption. | |
1807 | */ | |
1808 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 1809 | struct cfq_queue *cur_cfqq) |
6d048f53 | 1810 | { |
a36e71f9 JA |
1811 | struct cfq_queue *cfqq; |
1812 | ||
39c01b21 DS |
1813 | if (cfq_class_idle(cur_cfqq)) |
1814 | return NULL; | |
e6c5bc73 JM |
1815 | if (!cfq_cfqq_sync(cur_cfqq)) |
1816 | return NULL; | |
1817 | if (CFQQ_SEEKY(cur_cfqq)) | |
1818 | return NULL; | |
1819 | ||
b9d8f4c7 GJ |
1820 | /* |
1821 | * Don't search priority tree if it's the only queue in the group. | |
1822 | */ | |
1823 | if (cur_cfqq->cfqg->nr_cfqq == 1) | |
1824 | return NULL; | |
1825 | ||
6d048f53 | 1826 | /* |
d9e7620e JA |
1827 | * We should notice if some of the queues are cooperating, eg |
1828 | * working closely on the same area of the disk. In that case, | |
1829 | * we can group them together and don't waste time idling. | |
6d048f53 | 1830 | */ |
a36e71f9 JA |
1831 | cfqq = cfqq_close(cfqd, cur_cfqq); |
1832 | if (!cfqq) | |
1833 | return NULL; | |
1834 | ||
8682e1f1 VG |
1835 | /* If new queue belongs to different cfq_group, don't choose it */ |
1836 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
1837 | return NULL; | |
1838 | ||
df5fe3e8 JM |
1839 | /* |
1840 | * It only makes sense to merge sync queues. | |
1841 | */ | |
1842 | if (!cfq_cfqq_sync(cfqq)) | |
1843 | return NULL; | |
e6c5bc73 JM |
1844 | if (CFQQ_SEEKY(cfqq)) |
1845 | return NULL; | |
df5fe3e8 | 1846 | |
c0324a02 CZ |
1847 | /* |
1848 | * Do not merge queues of different priority classes | |
1849 | */ | |
1850 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
1851 | return NULL; | |
1852 | ||
a36e71f9 | 1853 | return cfqq; |
6d048f53 JA |
1854 | } |
1855 | ||
a6d44e98 CZ |
1856 | /* |
1857 | * Determine whether we should enforce idle window for this queue. | |
1858 | */ | |
1859 | ||
1860 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1861 | { | |
1862 | enum wl_prio_t prio = cfqq_prio(cfqq); | |
718eee05 | 1863 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
a6d44e98 | 1864 | |
f04a6424 VG |
1865 | BUG_ON(!service_tree); |
1866 | BUG_ON(!service_tree->count); | |
1867 | ||
b6508c16 VG |
1868 | if (!cfqd->cfq_slice_idle) |
1869 | return false; | |
1870 | ||
a6d44e98 CZ |
1871 | /* We never do for idle class queues. */ |
1872 | if (prio == IDLE_WORKLOAD) | |
1873 | return false; | |
1874 | ||
1875 | /* We do for queues that were marked with idle window flag. */ | |
3c764b7a SL |
1876 | if (cfq_cfqq_idle_window(cfqq) && |
1877 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) | |
a6d44e98 CZ |
1878 | return true; |
1879 | ||
1880 | /* | |
1881 | * Otherwise, we do only if they are the last ones | |
1882 | * in their service tree. | |
1883 | */ | |
b1ffe737 DS |
1884 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq)) |
1885 | return 1; | |
1886 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", | |
1887 | service_tree->count); | |
1888 | return 0; | |
a6d44e98 CZ |
1889 | } |
1890 | ||
6d048f53 | 1891 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 1892 | { |
1792669c | 1893 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 1894 | struct cfq_io_context *cic; |
80bdf0c7 | 1895 | unsigned long sl, group_idle = 0; |
7b14e3b5 | 1896 | |
a68bbddb | 1897 | /* |
f7d7b7a7 JA |
1898 | * SSD device without seek penalty, disable idling. But only do so |
1899 | * for devices that support queuing, otherwise we still have a problem | |
1900 | * with sync vs async workloads. | |
a68bbddb | 1901 | */ |
f7d7b7a7 | 1902 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
1903 | return; |
1904 | ||
dd67d051 | 1905 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 1906 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
1907 | |
1908 | /* | |
1909 | * idle is disabled, either manually or by past process history | |
1910 | */ | |
80bdf0c7 VG |
1911 | if (!cfq_should_idle(cfqd, cfqq)) { |
1912 | /* no queue idling. Check for group idling */ | |
1913 | if (cfqd->cfq_group_idle) | |
1914 | group_idle = cfqd->cfq_group_idle; | |
1915 | else | |
1916 | return; | |
1917 | } | |
6d048f53 | 1918 | |
7b679138 | 1919 | /* |
8e550632 | 1920 | * still active requests from this queue, don't idle |
7b679138 | 1921 | */ |
8e550632 | 1922 | if (cfqq->dispatched) |
7b679138 JA |
1923 | return; |
1924 | ||
22e2c507 JA |
1925 | /* |
1926 | * task has exited, don't wait | |
1927 | */ | |
206dc69b | 1928 | cic = cfqd->active_cic; |
66dac98e | 1929 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
6d048f53 JA |
1930 | return; |
1931 | ||
355b659c CZ |
1932 | /* |
1933 | * If our average think time is larger than the remaining time | |
1934 | * slice, then don't idle. This avoids overrunning the allotted | |
1935 | * time slice. | |
1936 | */ | |
1937 | if (sample_valid(cic->ttime_samples) && | |
b1ffe737 DS |
1938 | (cfqq->slice_end - jiffies < cic->ttime_mean)) { |
1939 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%d", | |
1940 | cic->ttime_mean); | |
355b659c | 1941 | return; |
b1ffe737 | 1942 | } |
355b659c | 1943 | |
80bdf0c7 VG |
1944 | /* There are other queues in the group, don't do group idle */ |
1945 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) | |
1946 | return; | |
1947 | ||
3b18152c | 1948 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 1949 | |
80bdf0c7 VG |
1950 | if (group_idle) |
1951 | sl = cfqd->cfq_group_idle; | |
1952 | else | |
1953 | sl = cfqd->cfq_slice_idle; | |
206dc69b | 1954 | |
7b14e3b5 | 1955 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
e98ef89b | 1956 | cfq_blkiocg_update_set_idle_time_stats(&cfqq->cfqg->blkg); |
80bdf0c7 VG |
1957 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
1958 | group_idle ? 1 : 0); | |
1da177e4 LT |
1959 | } |
1960 | ||
498d3aa2 JA |
1961 | /* |
1962 | * Move request from internal lists to the request queue dispatch list. | |
1963 | */ | |
165125e1 | 1964 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 1965 | { |
3ed9a296 | 1966 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1967 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1968 | |
7b679138 JA |
1969 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
1970 | ||
06d21886 | 1971 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 1972 | cfq_remove_request(rq); |
6d048f53 | 1973 | cfqq->dispatched++; |
80bdf0c7 | 1974 | (RQ_CFQG(rq))->dispatched++; |
5380a101 | 1975 | elv_dispatch_sort(q, rq); |
3ed9a296 | 1976 | |
53c583d2 | 1977 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
c4e7893e | 1978 | cfqq->nr_sectors += blk_rq_sectors(rq); |
e98ef89b | 1979 | cfq_blkiocg_update_dispatch_stats(&cfqq->cfqg->blkg, blk_rq_bytes(rq), |
84c124da | 1980 | rq_data_dir(rq), rq_is_sync(rq)); |
1da177e4 LT |
1981 | } |
1982 | ||
1983 | /* | |
1984 | * return expired entry, or NULL to just start from scratch in rbtree | |
1985 | */ | |
febffd61 | 1986 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 1987 | { |
30996f40 | 1988 | struct request *rq = NULL; |
1da177e4 | 1989 | |
3b18152c | 1990 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 1991 | return NULL; |
cb887411 JA |
1992 | |
1993 | cfq_mark_cfqq_fifo_expire(cfqq); | |
1994 | ||
89850f7e JA |
1995 | if (list_empty(&cfqq->fifo)) |
1996 | return NULL; | |
1da177e4 | 1997 | |
89850f7e | 1998 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 1999 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 2000 | rq = NULL; |
1da177e4 | 2001 | |
30996f40 | 2002 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 2003 | return rq; |
1da177e4 LT |
2004 | } |
2005 | ||
22e2c507 JA |
2006 | static inline int |
2007 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2008 | { | |
2009 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 2010 | |
22e2c507 | 2011 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 2012 | |
22e2c507 | 2013 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
2014 | } |
2015 | ||
df5fe3e8 JM |
2016 | /* |
2017 | * Must be called with the queue_lock held. | |
2018 | */ | |
2019 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
2020 | { | |
2021 | int process_refs, io_refs; | |
2022 | ||
2023 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
2024 | process_refs = atomic_read(&cfqq->ref) - io_refs; | |
2025 | BUG_ON(process_refs < 0); | |
2026 | return process_refs; | |
2027 | } | |
2028 | ||
2029 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
2030 | { | |
e6c5bc73 | 2031 | int process_refs, new_process_refs; |
df5fe3e8 JM |
2032 | struct cfq_queue *__cfqq; |
2033 | ||
c10b61f0 JM |
2034 | /* |
2035 | * If there are no process references on the new_cfqq, then it is | |
2036 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the | |
2037 | * chain may have dropped their last reference (not just their | |
2038 | * last process reference). | |
2039 | */ | |
2040 | if (!cfqq_process_refs(new_cfqq)) | |
2041 | return; | |
2042 | ||
df5fe3e8 JM |
2043 | /* Avoid a circular list and skip interim queue merges */ |
2044 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
2045 | if (__cfqq == cfqq) | |
2046 | return; | |
2047 | new_cfqq = __cfqq; | |
2048 | } | |
2049 | ||
2050 | process_refs = cfqq_process_refs(cfqq); | |
c10b61f0 | 2051 | new_process_refs = cfqq_process_refs(new_cfqq); |
df5fe3e8 JM |
2052 | /* |
2053 | * If the process for the cfqq has gone away, there is no | |
2054 | * sense in merging the queues. | |
2055 | */ | |
c10b61f0 | 2056 | if (process_refs == 0 || new_process_refs == 0) |
df5fe3e8 JM |
2057 | return; |
2058 | ||
e6c5bc73 JM |
2059 | /* |
2060 | * Merge in the direction of the lesser amount of work. | |
2061 | */ | |
e6c5bc73 JM |
2062 | if (new_process_refs >= process_refs) { |
2063 | cfqq->new_cfqq = new_cfqq; | |
2064 | atomic_add(process_refs, &new_cfqq->ref); | |
2065 | } else { | |
2066 | new_cfqq->new_cfqq = cfqq; | |
2067 | atomic_add(new_process_refs, &cfqq->ref); | |
2068 | } | |
df5fe3e8 JM |
2069 | } |
2070 | ||
cdb16e8f | 2071 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
65b32a57 | 2072 | struct cfq_group *cfqg, enum wl_prio_t prio) |
718eee05 CZ |
2073 | { |
2074 | struct cfq_queue *queue; | |
2075 | int i; | |
2076 | bool key_valid = false; | |
2077 | unsigned long lowest_key = 0; | |
2078 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
2079 | ||
65b32a57 VG |
2080 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
2081 | /* select the one with lowest rb_key */ | |
2082 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); | |
718eee05 CZ |
2083 | if (queue && |
2084 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
2085 | lowest_key = queue->rb_key; | |
2086 | cur_best = i; | |
2087 | key_valid = true; | |
2088 | } | |
2089 | } | |
2090 | ||
2091 | return cur_best; | |
2092 | } | |
2093 | ||
cdb16e8f | 2094 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
718eee05 | 2095 | { |
718eee05 CZ |
2096 | unsigned slice; |
2097 | unsigned count; | |
cdb16e8f | 2098 | struct cfq_rb_root *st; |
58ff82f3 | 2099 | unsigned group_slice; |
718eee05 | 2100 | |
1fa8f6d6 VG |
2101 | if (!cfqg) { |
2102 | cfqd->serving_prio = IDLE_WORKLOAD; | |
2103 | cfqd->workload_expires = jiffies + 1; | |
2104 | return; | |
2105 | } | |
2106 | ||
718eee05 | 2107 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 2108 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
718eee05 | 2109 | cfqd->serving_prio = RT_WORKLOAD; |
58ff82f3 | 2110 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
718eee05 CZ |
2111 | cfqd->serving_prio = BE_WORKLOAD; |
2112 | else { | |
2113 | cfqd->serving_prio = IDLE_WORKLOAD; | |
2114 | cfqd->workload_expires = jiffies + 1; | |
2115 | return; | |
2116 | } | |
2117 | ||
2118 | /* | |
2119 | * For RT and BE, we have to choose also the type | |
2120 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
2121 | * expiration time | |
2122 | */ | |
65b32a57 | 2123 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
cdb16e8f | 2124 | count = st->count; |
718eee05 CZ |
2125 | |
2126 | /* | |
65b32a57 | 2127 | * check workload expiration, and that we still have other queues ready |
718eee05 | 2128 | */ |
65b32a57 | 2129 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
718eee05 CZ |
2130 | return; |
2131 | ||
2132 | /* otherwise select new workload type */ | |
2133 | cfqd->serving_type = | |
65b32a57 VG |
2134 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); |
2135 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); | |
cdb16e8f | 2136 | count = st->count; |
718eee05 CZ |
2137 | |
2138 | /* | |
2139 | * the workload slice is computed as a fraction of target latency | |
2140 | * proportional to the number of queues in that workload, over | |
2141 | * all the queues in the same priority class | |
2142 | */ | |
58ff82f3 VG |
2143 | group_slice = cfq_group_slice(cfqd, cfqg); |
2144 | ||
2145 | slice = group_slice * count / | |
2146 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], | |
2147 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); | |
718eee05 | 2148 | |
f26bd1f0 VG |
2149 | if (cfqd->serving_type == ASYNC_WORKLOAD) { |
2150 | unsigned int tmp; | |
2151 | ||
2152 | /* | |
2153 | * Async queues are currently system wide. Just taking | |
2154 | * proportion of queues with-in same group will lead to higher | |
2155 | * async ratio system wide as generally root group is going | |
2156 | * to have higher weight. A more accurate thing would be to | |
2157 | * calculate system wide asnc/sync ratio. | |
2158 | */ | |
2159 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); | |
2160 | tmp = tmp/cfqd->busy_queues; | |
2161 | slice = min_t(unsigned, slice, tmp); | |
2162 | ||
718eee05 CZ |
2163 | /* async workload slice is scaled down according to |
2164 | * the sync/async slice ratio. */ | |
2165 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
f26bd1f0 | 2166 | } else |
718eee05 CZ |
2167 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
2168 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2169 | ||
2170 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
b1ffe737 | 2171 | cfq_log(cfqd, "workload slice:%d", slice); |
718eee05 | 2172 | cfqd->workload_expires = jiffies + slice; |
8e550632 | 2173 | cfqd->noidle_tree_requires_idle = false; |
718eee05 CZ |
2174 | } |
2175 | ||
1fa8f6d6 VG |
2176 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2177 | { | |
2178 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2179 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2180 | |
2181 | if (RB_EMPTY_ROOT(&st->rb)) | |
2182 | return NULL; | |
25bc6b07 VG |
2183 | cfqg = cfq_rb_first_group(st); |
2184 | st->active = &cfqg->rb_node; | |
2185 | update_min_vdisktime(st); | |
2186 | return cfqg; | |
1fa8f6d6 VG |
2187 | } |
2188 | ||
cdb16e8f VG |
2189 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2190 | { | |
1fa8f6d6 VG |
2191 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2192 | ||
2193 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2194 | |
2195 | /* Restore the workload type data */ | |
2196 | if (cfqg->saved_workload_slice) { | |
2197 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; | |
2198 | cfqd->serving_type = cfqg->saved_workload; | |
2199 | cfqd->serving_prio = cfqg->saved_serving_prio; | |
66ae2919 GJ |
2200 | } else |
2201 | cfqd->workload_expires = jiffies - 1; | |
2202 | ||
1fa8f6d6 | 2203 | choose_service_tree(cfqd, cfqg); |
cdb16e8f VG |
2204 | } |
2205 | ||
22e2c507 | 2206 | /* |
498d3aa2 JA |
2207 | * Select a queue for service. If we have a current active queue, |
2208 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2209 | */ |
1b5ed5e1 | 2210 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2211 | { |
a36e71f9 | 2212 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2213 | |
22e2c507 JA |
2214 | cfqq = cfqd->active_queue; |
2215 | if (!cfqq) | |
2216 | goto new_queue; | |
1da177e4 | 2217 | |
f04a6424 VG |
2218 | if (!cfqd->rq_queued) |
2219 | return NULL; | |
c244bb50 VG |
2220 | |
2221 | /* | |
2222 | * We were waiting for group to get backlogged. Expire the queue | |
2223 | */ | |
2224 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) | |
2225 | goto expire; | |
2226 | ||
22e2c507 | 2227 | /* |
6d048f53 | 2228 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2229 | */ |
7667aa06 VG |
2230 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
2231 | /* | |
2232 | * If slice had not expired at the completion of last request | |
2233 | * we might not have turned on wait_busy flag. Don't expire | |
2234 | * the queue yet. Allow the group to get backlogged. | |
2235 | * | |
2236 | * The very fact that we have used the slice, that means we | |
2237 | * have been idling all along on this queue and it should be | |
2238 | * ok to wait for this request to complete. | |
2239 | */ | |
82bbbf28 VG |
2240 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
2241 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2242 | cfqq = NULL; | |
7667aa06 | 2243 | goto keep_queue; |
82bbbf28 | 2244 | } else |
80bdf0c7 | 2245 | goto check_group_idle; |
7667aa06 | 2246 | } |
1da177e4 | 2247 | |
22e2c507 | 2248 | /* |
6d048f53 JA |
2249 | * The active queue has requests and isn't expired, allow it to |
2250 | * dispatch. | |
22e2c507 | 2251 | */ |
dd67d051 | 2252 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2253 | goto keep_queue; |
6d048f53 | 2254 | |
a36e71f9 JA |
2255 | /* |
2256 | * If another queue has a request waiting within our mean seek | |
2257 | * distance, let it run. The expire code will check for close | |
2258 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2259 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2260 | */ |
b3b6d040 | 2261 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2262 | if (new_cfqq) { |
2263 | if (!cfqq->new_cfqq) | |
2264 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2265 | goto expire; |
df5fe3e8 | 2266 | } |
a36e71f9 | 2267 | |
6d048f53 JA |
2268 | /* |
2269 | * No requests pending. If the active queue still has requests in | |
2270 | * flight or is idling for a new request, allow either of these | |
2271 | * conditions to happen (or time out) before selecting a new queue. | |
2272 | */ | |
80bdf0c7 VG |
2273 | if (timer_pending(&cfqd->idle_slice_timer)) { |
2274 | cfqq = NULL; | |
2275 | goto keep_queue; | |
2276 | } | |
2277 | ||
2278 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { | |
2279 | cfqq = NULL; | |
2280 | goto keep_queue; | |
2281 | } | |
2282 | ||
2283 | /* | |
2284 | * If group idle is enabled and there are requests dispatched from | |
2285 | * this group, wait for requests to complete. | |
2286 | */ | |
2287 | check_group_idle: | |
2288 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 | |
2289 | && cfqq->cfqg->dispatched) { | |
caaa5f9f JA |
2290 | cfqq = NULL; |
2291 | goto keep_queue; | |
22e2c507 JA |
2292 | } |
2293 | ||
3b18152c | 2294 | expire: |
e5ff082e | 2295 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2296 | new_queue: |
718eee05 CZ |
2297 | /* |
2298 | * Current queue expired. Check if we have to switch to a new | |
2299 | * service tree | |
2300 | */ | |
2301 | if (!new_cfqq) | |
cdb16e8f | 2302 | cfq_choose_cfqg(cfqd); |
718eee05 | 2303 | |
a36e71f9 | 2304 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2305 | keep_queue: |
3b18152c | 2306 | return cfqq; |
22e2c507 JA |
2307 | } |
2308 | ||
febffd61 | 2309 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2310 | { |
2311 | int dispatched = 0; | |
2312 | ||
2313 | while (cfqq->next_rq) { | |
2314 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2315 | dispatched++; | |
2316 | } | |
2317 | ||
2318 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2319 | |
2320 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
e5ff082e | 2321 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
d9e7620e JA |
2322 | return dispatched; |
2323 | } | |
2324 | ||
498d3aa2 JA |
2325 | /* |
2326 | * Drain our current requests. Used for barriers and when switching | |
2327 | * io schedulers on-the-fly. | |
2328 | */ | |
d9e7620e | 2329 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2330 | { |
0871714e | 2331 | struct cfq_queue *cfqq; |
d9e7620e | 2332 | int dispatched = 0; |
cdb16e8f | 2333 | |
3440c49f | 2334 | /* Expire the timeslice of the current active queue first */ |
e5ff082e | 2335 | cfq_slice_expired(cfqd, 0); |
3440c49f DS |
2336 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
2337 | __cfq_set_active_queue(cfqd, cfqq); | |
f04a6424 | 2338 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
3440c49f | 2339 | } |
1b5ed5e1 | 2340 | |
1b5ed5e1 TH |
2341 | BUG_ON(cfqd->busy_queues); |
2342 | ||
6923715a | 2343 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2344 | return dispatched; |
2345 | } | |
2346 | ||
abc3c744 SL |
2347 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
2348 | struct cfq_queue *cfqq) | |
2349 | { | |
2350 | /* the queue hasn't finished any request, can't estimate */ | |
2351 | if (cfq_cfqq_slice_new(cfqq)) | |
2352 | return 1; | |
2353 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, | |
2354 | cfqq->slice_end)) | |
2355 | return 1; | |
2356 | ||
2357 | return 0; | |
2358 | } | |
2359 | ||
0b182d61 | 2360 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2361 | { |
2f5cb738 | 2362 | unsigned int max_dispatch; |
22e2c507 | 2363 | |
5ad531db JA |
2364 | /* |
2365 | * Drain async requests before we start sync IO | |
2366 | */ | |
53c583d2 | 2367 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
0b182d61 | 2368 | return false; |
5ad531db | 2369 | |
2f5cb738 JA |
2370 | /* |
2371 | * If this is an async queue and we have sync IO in flight, let it wait | |
2372 | */ | |
53c583d2 | 2373 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
0b182d61 | 2374 | return false; |
2f5cb738 | 2375 | |
abc3c744 | 2376 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
2f5cb738 JA |
2377 | if (cfq_class_idle(cfqq)) |
2378 | max_dispatch = 1; | |
b4878f24 | 2379 | |
2f5cb738 JA |
2380 | /* |
2381 | * Does this cfqq already have too much IO in flight? | |
2382 | */ | |
2383 | if (cfqq->dispatched >= max_dispatch) { | |
2384 | /* | |
2385 | * idle queue must always only have a single IO in flight | |
2386 | */ | |
3ed9a296 | 2387 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2388 | return false; |
3ed9a296 | 2389 | |
2f5cb738 JA |
2390 | /* |
2391 | * We have other queues, don't allow more IO from this one | |
2392 | */ | |
abc3c744 | 2393 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq)) |
0b182d61 | 2394 | return false; |
9ede209e | 2395 | |
365722bb | 2396 | /* |
474b18cc | 2397 | * Sole queue user, no limit |
365722bb | 2398 | */ |
abc3c744 SL |
2399 | if (cfqd->busy_queues == 1) |
2400 | max_dispatch = -1; | |
2401 | else | |
2402 | /* | |
2403 | * Normally we start throttling cfqq when cfq_quantum/2 | |
2404 | * requests have been dispatched. But we can drive | |
2405 | * deeper queue depths at the beginning of slice | |
2406 | * subjected to upper limit of cfq_quantum. | |
2407 | * */ | |
2408 | max_dispatch = cfqd->cfq_quantum; | |
8e296755 JA |
2409 | } |
2410 | ||
2411 | /* | |
2412 | * Async queues must wait a bit before being allowed dispatch. | |
2413 | * We also ramp up the dispatch depth gradually for async IO, | |
2414 | * based on the last sync IO we serviced | |
2415 | */ | |
963b72fc | 2416 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
573412b2 | 2417 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
8e296755 | 2418 | unsigned int depth; |
365722bb | 2419 | |
61f0c1dc | 2420 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2421 | if (!depth && !cfqq->dispatched) |
2422 | depth = 1; | |
8e296755 JA |
2423 | if (depth < max_dispatch) |
2424 | max_dispatch = depth; | |
2f5cb738 | 2425 | } |
3ed9a296 | 2426 | |
0b182d61 JA |
2427 | /* |
2428 | * If we're below the current max, allow a dispatch | |
2429 | */ | |
2430 | return cfqq->dispatched < max_dispatch; | |
2431 | } | |
2432 | ||
2433 | /* | |
2434 | * Dispatch a request from cfqq, moving them to the request queue | |
2435 | * dispatch list. | |
2436 | */ | |
2437 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2438 | { | |
2439 | struct request *rq; | |
2440 | ||
2441 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
2442 | ||
2443 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
2444 | return false; | |
2445 | ||
2446 | /* | |
2447 | * follow expired path, else get first next available | |
2448 | */ | |
2449 | rq = cfq_check_fifo(cfqq); | |
2450 | if (!rq) | |
2451 | rq = cfqq->next_rq; | |
2452 | ||
2453 | /* | |
2454 | * insert request into driver dispatch list | |
2455 | */ | |
2456 | cfq_dispatch_insert(cfqd->queue, rq); | |
2457 | ||
2458 | if (!cfqd->active_cic) { | |
2459 | struct cfq_io_context *cic = RQ_CIC(rq); | |
2460 | ||
2461 | atomic_long_inc(&cic->ioc->refcount); | |
2462 | cfqd->active_cic = cic; | |
2463 | } | |
2464 | ||
2465 | return true; | |
2466 | } | |
2467 | ||
2468 | /* | |
2469 | * Find the cfqq that we need to service and move a request from that to the | |
2470 | * dispatch list | |
2471 | */ | |
2472 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
2473 | { | |
2474 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2475 | struct cfq_queue *cfqq; | |
2476 | ||
2477 | if (!cfqd->busy_queues) | |
2478 | return 0; | |
2479 | ||
2480 | if (unlikely(force)) | |
2481 | return cfq_forced_dispatch(cfqd); | |
2482 | ||
2483 | cfqq = cfq_select_queue(cfqd); | |
2484 | if (!cfqq) | |
8e296755 JA |
2485 | return 0; |
2486 | ||
2f5cb738 | 2487 | /* |
0b182d61 | 2488 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 2489 | */ |
0b182d61 JA |
2490 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2491 | return 0; | |
2492 | ||
2f5cb738 | 2493 | cfqq->slice_dispatch++; |
b029195d | 2494 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 2495 | |
2f5cb738 JA |
2496 | /* |
2497 | * expire an async queue immediately if it has used up its slice. idle | |
2498 | * queue always expire after 1 dispatch round. | |
2499 | */ | |
2500 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
2501 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
2502 | cfq_class_idle(cfqq))) { | |
2503 | cfqq->slice_end = jiffies + 1; | |
e5ff082e | 2504 | cfq_slice_expired(cfqd, 0); |
1da177e4 LT |
2505 | } |
2506 | ||
b217a903 | 2507 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 2508 | return 1; |
1da177e4 LT |
2509 | } |
2510 | ||
1da177e4 | 2511 | /* |
5e705374 JA |
2512 | * task holds one reference to the queue, dropped when task exits. each rq |
2513 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 2514 | * |
b1c35769 | 2515 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
2516 | * queue lock must be held here. |
2517 | */ | |
2518 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
2519 | { | |
22e2c507 | 2520 | struct cfq_data *cfqd = cfqq->cfqd; |
878eaddd | 2521 | struct cfq_group *cfqg, *orig_cfqg; |
22e2c507 JA |
2522 | |
2523 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
2524 | |
2525 | if (!atomic_dec_and_test(&cfqq->ref)) | |
2526 | return; | |
2527 | ||
7b679138 | 2528 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 2529 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 2530 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 2531 | cfqg = cfqq->cfqg; |
878eaddd | 2532 | orig_cfqg = cfqq->orig_cfqg; |
1da177e4 | 2533 | |
28f95cbc | 2534 | if (unlikely(cfqd->active_queue == cfqq)) { |
e5ff082e | 2535 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2536 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2537 | } |
22e2c507 | 2538 | |
f04a6424 | 2539 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 2540 | kmem_cache_free(cfq_pool, cfqq); |
b1c35769 | 2541 | cfq_put_cfqg(cfqg); |
878eaddd VG |
2542 | if (orig_cfqg) |
2543 | cfq_put_cfqg(orig_cfqg); | |
1da177e4 LT |
2544 | } |
2545 | ||
d6de8be7 JA |
2546 | /* |
2547 | * Must always be called with the rcu_read_lock() held | |
2548 | */ | |
07416d29 JA |
2549 | static void |
2550 | __call_for_each_cic(struct io_context *ioc, | |
2551 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
2552 | { | |
2553 | struct cfq_io_context *cic; | |
2554 | struct hlist_node *n; | |
2555 | ||
2556 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) | |
2557 | func(ioc, cic); | |
2558 | } | |
2559 | ||
4ac845a2 | 2560 | /* |
34e6bbf2 | 2561 | * Call func for each cic attached to this ioc. |
4ac845a2 | 2562 | */ |
34e6bbf2 | 2563 | static void |
4ac845a2 JA |
2564 | call_for_each_cic(struct io_context *ioc, |
2565 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1da177e4 | 2566 | { |
4ac845a2 | 2567 | rcu_read_lock(); |
07416d29 | 2568 | __call_for_each_cic(ioc, func); |
4ac845a2 | 2569 | rcu_read_unlock(); |
34e6bbf2 FC |
2570 | } |
2571 | ||
2572 | static void cfq_cic_free_rcu(struct rcu_head *head) | |
2573 | { | |
2574 | struct cfq_io_context *cic; | |
2575 | ||
2576 | cic = container_of(head, struct cfq_io_context, rcu_head); | |
2577 | ||
2578 | kmem_cache_free(cfq_ioc_pool, cic); | |
245b2e70 | 2579 | elv_ioc_count_dec(cfq_ioc_count); |
34e6bbf2 | 2580 | |
9a11b4ed JA |
2581 | if (ioc_gone) { |
2582 | /* | |
2583 | * CFQ scheduler is exiting, grab exit lock and check | |
2584 | * the pending io context count. If it hits zero, | |
2585 | * complete ioc_gone and set it back to NULL | |
2586 | */ | |
2587 | spin_lock(&ioc_gone_lock); | |
245b2e70 | 2588 | if (ioc_gone && !elv_ioc_count_read(cfq_ioc_count)) { |
9a11b4ed JA |
2589 | complete(ioc_gone); |
2590 | ioc_gone = NULL; | |
2591 | } | |
2592 | spin_unlock(&ioc_gone_lock); | |
2593 | } | |
34e6bbf2 | 2594 | } |
4ac845a2 | 2595 | |
34e6bbf2 FC |
2596 | static void cfq_cic_free(struct cfq_io_context *cic) |
2597 | { | |
2598 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | |
4ac845a2 JA |
2599 | } |
2600 | ||
2601 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) | |
2602 | { | |
2603 | unsigned long flags; | |
bca4b914 | 2604 | unsigned long dead_key = (unsigned long) cic->key; |
4ac845a2 | 2605 | |
bca4b914 | 2606 | BUG_ON(!(dead_key & CIC_DEAD_KEY)); |
4ac845a2 JA |
2607 | |
2608 | spin_lock_irqsave(&ioc->lock, flags); | |
80b15c73 | 2609 | radix_tree_delete(&ioc->radix_root, dead_key >> CIC_DEAD_INDEX_SHIFT); |
ffc4e759 | 2610 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
2611 | spin_unlock_irqrestore(&ioc->lock, flags); |
2612 | ||
34e6bbf2 | 2613 | cfq_cic_free(cic); |
4ac845a2 JA |
2614 | } |
2615 | ||
d6de8be7 JA |
2616 | /* |
2617 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. | |
2618 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), | |
2619 | * and ->trim() which is called with the task lock held | |
2620 | */ | |
4ac845a2 JA |
2621 | static void cfq_free_io_context(struct io_context *ioc) |
2622 | { | |
4ac845a2 | 2623 | /* |
34e6bbf2 FC |
2624 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
2625 | * so no more cic's are allowed to be linked into this ioc. So it | |
2626 | * should be ok to iterate over the known list, we will see all cic's | |
2627 | * since no new ones are added. | |
4ac845a2 | 2628 | */ |
07416d29 | 2629 | __call_for_each_cic(ioc, cic_free_func); |
1da177e4 LT |
2630 | } |
2631 | ||
d02a2c07 | 2632 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
1da177e4 | 2633 | { |
df5fe3e8 JM |
2634 | struct cfq_queue *__cfqq, *next; |
2635 | ||
df5fe3e8 JM |
2636 | /* |
2637 | * If this queue was scheduled to merge with another queue, be | |
2638 | * sure to drop the reference taken on that queue (and others in | |
2639 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
2640 | */ | |
2641 | __cfqq = cfqq->new_cfqq; | |
2642 | while (__cfqq) { | |
2643 | if (__cfqq == cfqq) { | |
2644 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
2645 | break; | |
2646 | } | |
2647 | next = __cfqq->new_cfqq; | |
2648 | cfq_put_queue(__cfqq); | |
2649 | __cfqq = next; | |
2650 | } | |
d02a2c07 SL |
2651 | } |
2652 | ||
2653 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2654 | { | |
2655 | if (unlikely(cfqq == cfqd->active_queue)) { | |
2656 | __cfq_slice_expired(cfqd, cfqq, 0); | |
2657 | cfq_schedule_dispatch(cfqd); | |
2658 | } | |
2659 | ||
2660 | cfq_put_cooperator(cfqq); | |
df5fe3e8 | 2661 | |
89850f7e JA |
2662 | cfq_put_queue(cfqq); |
2663 | } | |
22e2c507 | 2664 | |
89850f7e JA |
2665 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
2666 | struct cfq_io_context *cic) | |
2667 | { | |
4faa3c81 FC |
2668 | struct io_context *ioc = cic->ioc; |
2669 | ||
fc46379d | 2670 | list_del_init(&cic->queue_list); |
4ac845a2 JA |
2671 | |
2672 | /* | |
bca4b914 | 2673 | * Make sure dead mark is seen for dead queues |
4ac845a2 | 2674 | */ |
fc46379d | 2675 | smp_wmb(); |
bca4b914 | 2676 | cic->key = cfqd_dead_key(cfqd); |
fc46379d | 2677 | |
4faa3c81 FC |
2678 | if (ioc->ioc_data == cic) |
2679 | rcu_assign_pointer(ioc->ioc_data, NULL); | |
2680 | ||
ff6657c6 JA |
2681 | if (cic->cfqq[BLK_RW_ASYNC]) { |
2682 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
2683 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
2684 | } |
2685 | ||
ff6657c6 JA |
2686 | if (cic->cfqq[BLK_RW_SYNC]) { |
2687 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
2688 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 2689 | } |
89850f7e JA |
2690 | } |
2691 | ||
4ac845a2 JA |
2692 | static void cfq_exit_single_io_context(struct io_context *ioc, |
2693 | struct cfq_io_context *cic) | |
89850f7e | 2694 | { |
bca4b914 | 2695 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
89850f7e | 2696 | |
89850f7e | 2697 | if (cfqd) { |
165125e1 | 2698 | struct request_queue *q = cfqd->queue; |
4ac845a2 | 2699 | unsigned long flags; |
89850f7e | 2700 | |
4ac845a2 | 2701 | spin_lock_irqsave(q->queue_lock, flags); |
62c1fe9d JA |
2702 | |
2703 | /* | |
2704 | * Ensure we get a fresh copy of the ->key to prevent | |
2705 | * race between exiting task and queue | |
2706 | */ | |
2707 | smp_read_barrier_depends(); | |
bca4b914 | 2708 | if (cic->key == cfqd) |
62c1fe9d JA |
2709 | __cfq_exit_single_io_context(cfqd, cic); |
2710 | ||
4ac845a2 | 2711 | spin_unlock_irqrestore(q->queue_lock, flags); |
89850f7e | 2712 | } |
1da177e4 LT |
2713 | } |
2714 | ||
498d3aa2 JA |
2715 | /* |
2716 | * The process that ioc belongs to has exited, we need to clean up | |
2717 | * and put the internal structures we have that belongs to that process. | |
2718 | */ | |
e2d74ac0 | 2719 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 2720 | { |
4ac845a2 | 2721 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
1da177e4 LT |
2722 | } |
2723 | ||
22e2c507 | 2724 | static struct cfq_io_context * |
8267e268 | 2725 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 2726 | { |
b5deef90 | 2727 | struct cfq_io_context *cic; |
1da177e4 | 2728 | |
94f6030c CL |
2729 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
2730 | cfqd->queue->node); | |
1da177e4 | 2731 | if (cic) { |
22e2c507 | 2732 | cic->last_end_request = jiffies; |
553698f9 | 2733 | INIT_LIST_HEAD(&cic->queue_list); |
ffc4e759 | 2734 | INIT_HLIST_NODE(&cic->cic_list); |
22e2c507 JA |
2735 | cic->dtor = cfq_free_io_context; |
2736 | cic->exit = cfq_exit_io_context; | |
245b2e70 | 2737 | elv_ioc_count_inc(cfq_ioc_count); |
1da177e4 LT |
2738 | } |
2739 | ||
2740 | return cic; | |
2741 | } | |
2742 | ||
fd0928df | 2743 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
2744 | { |
2745 | struct task_struct *tsk = current; | |
2746 | int ioprio_class; | |
2747 | ||
3b18152c | 2748 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
2749 | return; |
2750 | ||
fd0928df | 2751 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 2752 | switch (ioprio_class) { |
fe094d98 JA |
2753 | default: |
2754 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
2755 | case IOPRIO_CLASS_NONE: | |
2756 | /* | |
6d63c275 | 2757 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
2758 | */ |
2759 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 2760 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
2761 | break; |
2762 | case IOPRIO_CLASS_RT: | |
2763 | cfqq->ioprio = task_ioprio(ioc); | |
2764 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
2765 | break; | |
2766 | case IOPRIO_CLASS_BE: | |
2767 | cfqq->ioprio = task_ioprio(ioc); | |
2768 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
2769 | break; | |
2770 | case IOPRIO_CLASS_IDLE: | |
2771 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
2772 | cfqq->ioprio = 7; | |
2773 | cfq_clear_cfqq_idle_window(cfqq); | |
2774 | break; | |
22e2c507 JA |
2775 | } |
2776 | ||
2777 | /* | |
2778 | * keep track of original prio settings in case we have to temporarily | |
2779 | * elevate the priority of this queue | |
2780 | */ | |
2781 | cfqq->org_ioprio = cfqq->ioprio; | |
2782 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 2783 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
2784 | } |
2785 | ||
febffd61 | 2786 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
22e2c507 | 2787 | { |
bca4b914 | 2788 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
478a82b0 | 2789 | struct cfq_queue *cfqq; |
c1b707d2 | 2790 | unsigned long flags; |
35e6077c | 2791 | |
caaa5f9f JA |
2792 | if (unlikely(!cfqd)) |
2793 | return; | |
2794 | ||
c1b707d2 | 2795 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f | 2796 | |
ff6657c6 | 2797 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
2798 | if (cfqq) { |
2799 | struct cfq_queue *new_cfqq; | |
ff6657c6 JA |
2800 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc, |
2801 | GFP_ATOMIC); | |
caaa5f9f | 2802 | if (new_cfqq) { |
ff6657c6 | 2803 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
2804 | cfq_put_queue(cfqq); |
2805 | } | |
22e2c507 | 2806 | } |
caaa5f9f | 2807 | |
ff6657c6 | 2808 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
2809 | if (cfqq) |
2810 | cfq_mark_cfqq_prio_changed(cfqq); | |
2811 | ||
c1b707d2 | 2812 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
2813 | } |
2814 | ||
fc46379d | 2815 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 2816 | { |
4ac845a2 | 2817 | call_for_each_cic(ioc, changed_ioprio); |
fc46379d | 2818 | ioc->ioprio_changed = 0; |
22e2c507 JA |
2819 | } |
2820 | ||
d5036d77 | 2821 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2822 | pid_t pid, bool is_sync) |
d5036d77 JA |
2823 | { |
2824 | RB_CLEAR_NODE(&cfqq->rb_node); | |
2825 | RB_CLEAR_NODE(&cfqq->p_node); | |
2826 | INIT_LIST_HEAD(&cfqq->fifo); | |
2827 | ||
2828 | atomic_set(&cfqq->ref, 0); | |
2829 | cfqq->cfqd = cfqd; | |
2830 | ||
2831 | cfq_mark_cfqq_prio_changed(cfqq); | |
2832 | ||
2833 | if (is_sync) { | |
2834 | if (!cfq_class_idle(cfqq)) | |
2835 | cfq_mark_cfqq_idle_window(cfqq); | |
2836 | cfq_mark_cfqq_sync(cfqq); | |
2837 | } | |
2838 | cfqq->pid = pid; | |
2839 | } | |
2840 | ||
24610333 VG |
2841 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2842 | static void changed_cgroup(struct io_context *ioc, struct cfq_io_context *cic) | |
2843 | { | |
2844 | struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1); | |
bca4b914 | 2845 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
24610333 VG |
2846 | unsigned long flags; |
2847 | struct request_queue *q; | |
2848 | ||
2849 | if (unlikely(!cfqd)) | |
2850 | return; | |
2851 | ||
2852 | q = cfqd->queue; | |
2853 | ||
2854 | spin_lock_irqsave(q->queue_lock, flags); | |
2855 | ||
2856 | if (sync_cfqq) { | |
2857 | /* | |
2858 | * Drop reference to sync queue. A new sync queue will be | |
2859 | * assigned in new group upon arrival of a fresh request. | |
2860 | */ | |
2861 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
2862 | cic_set_cfqq(cic, NULL, 1); | |
2863 | cfq_put_queue(sync_cfqq); | |
2864 | } | |
2865 | ||
2866 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2867 | } | |
2868 | ||
2869 | static void cfq_ioc_set_cgroup(struct io_context *ioc) | |
2870 | { | |
2871 | call_for_each_cic(ioc, changed_cgroup); | |
2872 | ioc->cgroup_changed = 0; | |
2873 | } | |
2874 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
2875 | ||
22e2c507 | 2876 | static struct cfq_queue * |
a6151c3a | 2877 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, |
fd0928df | 2878 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 2879 | { |
22e2c507 | 2880 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 2881 | struct cfq_io_context *cic; |
cdb16e8f | 2882 | struct cfq_group *cfqg; |
22e2c507 JA |
2883 | |
2884 | retry: | |
cdb16e8f | 2885 | cfqg = cfq_get_cfqg(cfqd, 1); |
4ac845a2 | 2886 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
2887 | /* cic always exists here */ |
2888 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 | 2889 | |
6118b70b JA |
2890 | /* |
2891 | * Always try a new alloc if we fell back to the OOM cfqq | |
2892 | * originally, since it should just be a temporary situation. | |
2893 | */ | |
2894 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
2895 | cfqq = NULL; | |
22e2c507 JA |
2896 | if (new_cfqq) { |
2897 | cfqq = new_cfqq; | |
2898 | new_cfqq = NULL; | |
2899 | } else if (gfp_mask & __GFP_WAIT) { | |
2900 | spin_unlock_irq(cfqd->queue->queue_lock); | |
94f6030c | 2901 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 2902 | gfp_mask | __GFP_ZERO, |
94f6030c | 2903 | cfqd->queue->node); |
22e2c507 | 2904 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
2905 | if (new_cfqq) |
2906 | goto retry; | |
22e2c507 | 2907 | } else { |
94f6030c CL |
2908 | cfqq = kmem_cache_alloc_node(cfq_pool, |
2909 | gfp_mask | __GFP_ZERO, | |
2910 | cfqd->queue->node); | |
22e2c507 JA |
2911 | } |
2912 | ||
6118b70b JA |
2913 | if (cfqq) { |
2914 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
2915 | cfq_init_prio_data(cfqq, ioc); | |
cdb16e8f | 2916 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
2917 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
2918 | } else | |
2919 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
2920 | } |
2921 | ||
2922 | if (new_cfqq) | |
2923 | kmem_cache_free(cfq_pool, new_cfqq); | |
2924 | ||
22e2c507 JA |
2925 | return cfqq; |
2926 | } | |
2927 | ||
c2dea2d1 VT |
2928 | static struct cfq_queue ** |
2929 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
2930 | { | |
fe094d98 | 2931 | switch (ioprio_class) { |
c2dea2d1 VT |
2932 | case IOPRIO_CLASS_RT: |
2933 | return &cfqd->async_cfqq[0][ioprio]; | |
2934 | case IOPRIO_CLASS_BE: | |
2935 | return &cfqd->async_cfqq[1][ioprio]; | |
2936 | case IOPRIO_CLASS_IDLE: | |
2937 | return &cfqd->async_idle_cfqq; | |
2938 | default: | |
2939 | BUG(); | |
2940 | } | |
2941 | } | |
2942 | ||
15c31be4 | 2943 | static struct cfq_queue * |
a6151c3a | 2944 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc, |
15c31be4 JA |
2945 | gfp_t gfp_mask) |
2946 | { | |
fd0928df JA |
2947 | const int ioprio = task_ioprio(ioc); |
2948 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 2949 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
2950 | struct cfq_queue *cfqq = NULL; |
2951 | ||
c2dea2d1 VT |
2952 | if (!is_sync) { |
2953 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
2954 | cfqq = *async_cfqq; | |
2955 | } | |
2956 | ||
6118b70b | 2957 | if (!cfqq) |
fd0928df | 2958 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
15c31be4 JA |
2959 | |
2960 | /* | |
2961 | * pin the queue now that it's allocated, scheduler exit will prune it | |
2962 | */ | |
c2dea2d1 | 2963 | if (!is_sync && !(*async_cfqq)) { |
15c31be4 | 2964 | atomic_inc(&cfqq->ref); |
c2dea2d1 | 2965 | *async_cfqq = cfqq; |
15c31be4 JA |
2966 | } |
2967 | ||
2968 | atomic_inc(&cfqq->ref); | |
2969 | return cfqq; | |
2970 | } | |
2971 | ||
498d3aa2 JA |
2972 | /* |
2973 | * We drop cfq io contexts lazily, so we may find a dead one. | |
2974 | */ | |
dbecf3ab | 2975 | static void |
4ac845a2 JA |
2976 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
2977 | struct cfq_io_context *cic) | |
dbecf3ab | 2978 | { |
4ac845a2 JA |
2979 | unsigned long flags; |
2980 | ||
fc46379d | 2981 | WARN_ON(!list_empty(&cic->queue_list)); |
bca4b914 | 2982 | BUG_ON(cic->key != cfqd_dead_key(cfqd)); |
597bc485 | 2983 | |
4ac845a2 JA |
2984 | spin_lock_irqsave(&ioc->lock, flags); |
2985 | ||
4faa3c81 | 2986 | BUG_ON(ioc->ioc_data == cic); |
597bc485 | 2987 | |
80b15c73 | 2988 | radix_tree_delete(&ioc->radix_root, cfqd->cic_index); |
ffc4e759 | 2989 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
2990 | spin_unlock_irqrestore(&ioc->lock, flags); |
2991 | ||
2992 | cfq_cic_free(cic); | |
dbecf3ab OH |
2993 | } |
2994 | ||
e2d74ac0 | 2995 | static struct cfq_io_context * |
4ac845a2 | 2996 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
e2d74ac0 | 2997 | { |
e2d74ac0 | 2998 | struct cfq_io_context *cic; |
d6de8be7 | 2999 | unsigned long flags; |
e2d74ac0 | 3000 | |
91fac317 VT |
3001 | if (unlikely(!ioc)) |
3002 | return NULL; | |
3003 | ||
d6de8be7 JA |
3004 | rcu_read_lock(); |
3005 | ||
597bc485 JA |
3006 | /* |
3007 | * we maintain a last-hit cache, to avoid browsing over the tree | |
3008 | */ | |
4ac845a2 | 3009 | cic = rcu_dereference(ioc->ioc_data); |
d6de8be7 JA |
3010 | if (cic && cic->key == cfqd) { |
3011 | rcu_read_unlock(); | |
597bc485 | 3012 | return cic; |
d6de8be7 | 3013 | } |
597bc485 | 3014 | |
4ac845a2 | 3015 | do { |
80b15c73 | 3016 | cic = radix_tree_lookup(&ioc->radix_root, cfqd->cic_index); |
4ac845a2 JA |
3017 | rcu_read_unlock(); |
3018 | if (!cic) | |
3019 | break; | |
bca4b914 | 3020 | if (unlikely(cic->key != cfqd)) { |
4ac845a2 | 3021 | cfq_drop_dead_cic(cfqd, ioc, cic); |
d6de8be7 | 3022 | rcu_read_lock(); |
4ac845a2 | 3023 | continue; |
dbecf3ab | 3024 | } |
e2d74ac0 | 3025 | |
d6de8be7 | 3026 | spin_lock_irqsave(&ioc->lock, flags); |
4ac845a2 | 3027 | rcu_assign_pointer(ioc->ioc_data, cic); |
d6de8be7 | 3028 | spin_unlock_irqrestore(&ioc->lock, flags); |
4ac845a2 JA |
3029 | break; |
3030 | } while (1); | |
e2d74ac0 | 3031 | |
4ac845a2 | 3032 | return cic; |
e2d74ac0 JA |
3033 | } |
3034 | ||
4ac845a2 JA |
3035 | /* |
3036 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup | |
3037 | * the process specific cfq io context when entered from the block layer. | |
3038 | * Also adds the cic to a per-cfqd list, used when this queue is removed. | |
3039 | */ | |
febffd61 JA |
3040 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
3041 | struct cfq_io_context *cic, gfp_t gfp_mask) | |
e2d74ac0 | 3042 | { |
0261d688 | 3043 | unsigned long flags; |
4ac845a2 | 3044 | int ret; |
e2d74ac0 | 3045 | |
4ac845a2 JA |
3046 | ret = radix_tree_preload(gfp_mask); |
3047 | if (!ret) { | |
3048 | cic->ioc = ioc; | |
3049 | cic->key = cfqd; | |
e2d74ac0 | 3050 | |
4ac845a2 JA |
3051 | spin_lock_irqsave(&ioc->lock, flags); |
3052 | ret = radix_tree_insert(&ioc->radix_root, | |
80b15c73 | 3053 | cfqd->cic_index, cic); |
ffc4e759 JA |
3054 | if (!ret) |
3055 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); | |
4ac845a2 | 3056 | spin_unlock_irqrestore(&ioc->lock, flags); |
e2d74ac0 | 3057 | |
4ac845a2 JA |
3058 | radix_tree_preload_end(); |
3059 | ||
3060 | if (!ret) { | |
3061 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
3062 | list_add(&cic->queue_list, &cfqd->cic_list); | |
3063 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3064 | } | |
e2d74ac0 JA |
3065 | } |
3066 | ||
4ac845a2 JA |
3067 | if (ret) |
3068 | printk(KERN_ERR "cfq: cic link failed!\n"); | |
fc46379d | 3069 | |
4ac845a2 | 3070 | return ret; |
e2d74ac0 JA |
3071 | } |
3072 | ||
1da177e4 LT |
3073 | /* |
3074 | * Setup general io context and cfq io context. There can be several cfq | |
3075 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 3076 | * than one device managed by cfq. |
1da177e4 LT |
3077 | */ |
3078 | static struct cfq_io_context * | |
e2d74ac0 | 3079 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 3080 | { |
22e2c507 | 3081 | struct io_context *ioc = NULL; |
1da177e4 | 3082 | struct cfq_io_context *cic; |
1da177e4 | 3083 | |
22e2c507 | 3084 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 3085 | |
b5deef90 | 3086 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
3087 | if (!ioc) |
3088 | return NULL; | |
3089 | ||
4ac845a2 | 3090 | cic = cfq_cic_lookup(cfqd, ioc); |
e2d74ac0 JA |
3091 | if (cic) |
3092 | goto out; | |
1da177e4 | 3093 | |
e2d74ac0 JA |
3094 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
3095 | if (cic == NULL) | |
3096 | goto err; | |
1da177e4 | 3097 | |
4ac845a2 JA |
3098 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) |
3099 | goto err_free; | |
3100 | ||
1da177e4 | 3101 | out: |
fc46379d JA |
3102 | smp_read_barrier_depends(); |
3103 | if (unlikely(ioc->ioprio_changed)) | |
3104 | cfq_ioc_set_ioprio(ioc); | |
3105 | ||
24610333 VG |
3106 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
3107 | if (unlikely(ioc->cgroup_changed)) | |
3108 | cfq_ioc_set_cgroup(ioc); | |
3109 | #endif | |
1da177e4 | 3110 | return cic; |
4ac845a2 JA |
3111 | err_free: |
3112 | cfq_cic_free(cic); | |
1da177e4 LT |
3113 | err: |
3114 | put_io_context(ioc); | |
3115 | return NULL; | |
3116 | } | |
3117 | ||
22e2c507 JA |
3118 | static void |
3119 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 3120 | { |
aaf1228d JA |
3121 | unsigned long elapsed = jiffies - cic->last_end_request; |
3122 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 3123 | |
22e2c507 JA |
3124 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
3125 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
3126 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
3127 | } | |
1da177e4 | 3128 | |
206dc69b | 3129 | static void |
b2c18e1e | 3130 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 3131 | struct request *rq) |
206dc69b | 3132 | { |
3dde36dd | 3133 | sector_t sdist = 0; |
41647e7a | 3134 | sector_t n_sec = blk_rq_sectors(rq); |
3dde36dd CZ |
3135 | if (cfqq->last_request_pos) { |
3136 | if (cfqq->last_request_pos < blk_rq_pos(rq)) | |
3137 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
3138 | else | |
3139 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); | |
3140 | } | |
206dc69b | 3141 | |
3dde36dd | 3142 | cfqq->seek_history <<= 1; |
41647e7a CZ |
3143 | if (blk_queue_nonrot(cfqd->queue)) |
3144 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); | |
3145 | else | |
3146 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); | |
206dc69b | 3147 | } |
1da177e4 | 3148 | |
22e2c507 JA |
3149 | /* |
3150 | * Disable idle window if the process thinks too long or seeks so much that | |
3151 | * it doesn't matter | |
3152 | */ | |
3153 | static void | |
3154 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
3155 | struct cfq_io_context *cic) | |
3156 | { | |
7b679138 | 3157 | int old_idle, enable_idle; |
1be92f2f | 3158 | |
0871714e JA |
3159 | /* |
3160 | * Don't idle for async or idle io prio class | |
3161 | */ | |
3162 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
3163 | return; |
3164 | ||
c265a7f4 | 3165 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 3166 | |
76280aff CZ |
3167 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
3168 | cfq_mark_cfqq_deep(cfqq); | |
3169 | ||
66dac98e | 3170 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
3dde36dd | 3171 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) |
22e2c507 JA |
3172 | enable_idle = 0; |
3173 | else if (sample_valid(cic->ttime_samples)) { | |
718eee05 | 3174 | if (cic->ttime_mean > cfqd->cfq_slice_idle) |
22e2c507 JA |
3175 | enable_idle = 0; |
3176 | else | |
3177 | enable_idle = 1; | |
1da177e4 LT |
3178 | } |
3179 | ||
7b679138 JA |
3180 | if (old_idle != enable_idle) { |
3181 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
3182 | if (enable_idle) | |
3183 | cfq_mark_cfqq_idle_window(cfqq); | |
3184 | else | |
3185 | cfq_clear_cfqq_idle_window(cfqq); | |
3186 | } | |
22e2c507 | 3187 | } |
1da177e4 | 3188 | |
22e2c507 JA |
3189 | /* |
3190 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3191 | * no or if we aren't sure, a 1 will cause a preempt. | |
3192 | */ | |
a6151c3a | 3193 | static bool |
22e2c507 | 3194 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3195 | struct request *rq) |
22e2c507 | 3196 | { |
6d048f53 | 3197 | struct cfq_queue *cfqq; |
22e2c507 | 3198 | |
6d048f53 JA |
3199 | cfqq = cfqd->active_queue; |
3200 | if (!cfqq) | |
a6151c3a | 3201 | return false; |
22e2c507 | 3202 | |
6d048f53 | 3203 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3204 | return false; |
22e2c507 JA |
3205 | |
3206 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3207 | return true; |
1e3335de | 3208 | |
875feb63 DS |
3209 | /* |
3210 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. | |
3211 | */ | |
3212 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) | |
3213 | return false; | |
3214 | ||
374f84ac JA |
3215 | /* |
3216 | * if the new request is sync, but the currently running queue is | |
3217 | * not, let the sync request have priority. | |
3218 | */ | |
5e705374 | 3219 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3220 | return true; |
1e3335de | 3221 | |
8682e1f1 VG |
3222 | if (new_cfqq->cfqg != cfqq->cfqg) |
3223 | return false; | |
3224 | ||
3225 | if (cfq_slice_used(cfqq)) | |
3226 | return true; | |
3227 | ||
3228 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
3229 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && | |
3230 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && | |
3231 | new_cfqq->service_tree->count == 2 && | |
3232 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3233 | return true; | |
3234 | ||
374f84ac JA |
3235 | /* |
3236 | * So both queues are sync. Let the new request get disk time if | |
3237 | * it's a metadata request and the current queue is doing regular IO. | |
3238 | */ | |
7b6d91da | 3239 | if ((rq->cmd_flags & REQ_META) && !cfqq->meta_pending) |
e6ec4fe2 | 3240 | return true; |
22e2c507 | 3241 | |
3a9a3f6c DS |
3242 | /* |
3243 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3244 | */ | |
3245 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3246 | return true; |
3a9a3f6c | 3247 | |
1e3335de | 3248 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3249 | return false; |
1e3335de JA |
3250 | |
3251 | /* | |
3252 | * if this request is as-good as one we would expect from the | |
3253 | * current cfqq, let it preempt | |
3254 | */ | |
e9ce335d | 3255 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3256 | return true; |
1e3335de | 3257 | |
a6151c3a | 3258 | return false; |
22e2c507 JA |
3259 | } |
3260 | ||
3261 | /* | |
3262 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3263 | * let it have half of its nominal slice. | |
3264 | */ | |
3265 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3266 | { | |
7b679138 | 3267 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
e5ff082e | 3268 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3269 | |
bf572256 JA |
3270 | /* |
3271 | * Put the new queue at the front of the of the current list, | |
3272 | * so we know that it will be selected next. | |
3273 | */ | |
3274 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3275 | |
3276 | cfq_service_tree_add(cfqd, cfqq, 1); | |
bf572256 | 3277 | |
44f7c160 JA |
3278 | cfqq->slice_end = 0; |
3279 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3280 | } |
3281 | ||
22e2c507 | 3282 | /* |
5e705374 | 3283 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3284 | * something we should do about it |
3285 | */ | |
3286 | static void | |
5e705374 JA |
3287 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3288 | struct request *rq) | |
22e2c507 | 3289 | { |
5e705374 | 3290 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 3291 | |
45333d5a | 3292 | cfqd->rq_queued++; |
7b6d91da | 3293 | if (rq->cmd_flags & REQ_META) |
374f84ac JA |
3294 | cfqq->meta_pending++; |
3295 | ||
9c2c38a1 | 3296 | cfq_update_io_thinktime(cfqd, cic); |
b2c18e1e | 3297 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3298 | cfq_update_idle_window(cfqd, cfqq, cic); |
3299 | ||
b2c18e1e | 3300 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3301 | |
3302 | if (cfqq == cfqd->active_queue) { | |
3303 | /* | |
b029195d JA |
3304 | * Remember that we saw a request from this process, but |
3305 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3306 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3307 | * and merging. If the request is already larger than a single |
3308 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3309 | * merging is already done. Ditto for a busy system that |
3310 | * has other work pending, don't risk delaying until the | |
3311 | * idle timer unplug to continue working. | |
22e2c507 | 3312 | */ |
d6ceb25e | 3313 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3314 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3315 | cfqd->busy_queues > 1) { | |
812df48d | 3316 | cfq_del_timer(cfqd, cfqq); |
554554f6 | 3317 | cfq_clear_cfqq_wait_request(cfqq); |
bf791937 | 3318 | __blk_run_queue(cfqd->queue); |
a11cdaa7 | 3319 | } else { |
e98ef89b | 3320 | cfq_blkiocg_update_idle_time_stats( |
a11cdaa7 | 3321 | &cfqq->cfqg->blkg); |
bf791937 | 3322 | cfq_mark_cfqq_must_dispatch(cfqq); |
a11cdaa7 | 3323 | } |
d6ceb25e | 3324 | } |
5e705374 | 3325 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3326 | /* |
3327 | * not the active queue - expire current slice if it is | |
3328 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3329 | * has some old slice time left and is of higher priority or |
3330 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3331 | */ |
3332 | cfq_preempt_queue(cfqd, cfqq); | |
a7f55792 | 3333 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3334 | } |
1da177e4 LT |
3335 | } |
3336 | ||
165125e1 | 3337 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3338 | { |
b4878f24 | 3339 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3340 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3341 | |
7b679138 | 3342 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
fd0928df | 3343 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
1da177e4 | 3344 | |
30996f40 | 3345 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3346 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3347 | cfq_add_rq_rb(rq); |
e98ef89b | 3348 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, |
cdc1184c DS |
3349 | &cfqd->serving_group->blkg, rq_data_dir(rq), |
3350 | rq_is_sync(rq)); | |
5e705374 | 3351 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3352 | } |
3353 | ||
45333d5a AC |
3354 | /* |
3355 | * Update hw_tag based on peak queue depth over 50 samples under | |
3356 | * sufficient load. | |
3357 | */ | |
3358 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3359 | { | |
1a1238a7 SL |
3360 | struct cfq_queue *cfqq = cfqd->active_queue; |
3361 | ||
53c583d2 CZ |
3362 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
3363 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; | |
e459dd08 CZ |
3364 | |
3365 | if (cfqd->hw_tag == 1) | |
3366 | return; | |
45333d5a AC |
3367 | |
3368 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
53c583d2 | 3369 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3370 | return; |
3371 | ||
1a1238a7 SL |
3372 | /* |
3373 | * If active queue hasn't enough requests and can idle, cfq might not | |
3374 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3375 | * case | |
3376 | */ | |
3377 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3378 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
53c583d2 | 3379 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
1a1238a7 SL |
3380 | return; |
3381 | ||
45333d5a AC |
3382 | if (cfqd->hw_tag_samples++ < 50) |
3383 | return; | |
3384 | ||
e459dd08 | 3385 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3386 | cfqd->hw_tag = 1; |
3387 | else | |
3388 | cfqd->hw_tag = 0; | |
45333d5a AC |
3389 | } |
3390 | ||
7667aa06 VG |
3391 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
3392 | { | |
3393 | struct cfq_io_context *cic = cfqd->active_cic; | |
3394 | ||
3395 | /* If there are other queues in the group, don't wait */ | |
3396 | if (cfqq->cfqg->nr_cfqq > 1) | |
3397 | return false; | |
3398 | ||
3399 | if (cfq_slice_used(cfqq)) | |
3400 | return true; | |
3401 | ||
3402 | /* if slice left is less than think time, wait busy */ | |
3403 | if (cic && sample_valid(cic->ttime_samples) | |
3404 | && (cfqq->slice_end - jiffies < cic->ttime_mean)) | |
3405 | return true; | |
3406 | ||
3407 | /* | |
3408 | * If think times is less than a jiffy than ttime_mean=0 and above | |
3409 | * will not be true. It might happen that slice has not expired yet | |
3410 | * but will expire soon (4-5 ns) during select_queue(). To cover the | |
3411 | * case where think time is less than a jiffy, mark the queue wait | |
3412 | * busy if only 1 jiffy is left in the slice. | |
3413 | */ | |
3414 | if (cfqq->slice_end - jiffies == 1) | |
3415 | return true; | |
3416 | ||
3417 | return false; | |
3418 | } | |
3419 | ||
165125e1 | 3420 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3421 | { |
5e705374 | 3422 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3423 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3424 | const int sync = rq_is_sync(rq); |
b4878f24 | 3425 | unsigned long now; |
1da177e4 | 3426 | |
b4878f24 | 3427 | now = jiffies; |
33659ebb CH |
3428 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
3429 | !!(rq->cmd_flags & REQ_NOIDLE)); | |
1da177e4 | 3430 | |
45333d5a AC |
3431 | cfq_update_hw_tag(cfqd); |
3432 | ||
53c583d2 | 3433 | WARN_ON(!cfqd->rq_in_driver); |
6d048f53 | 3434 | WARN_ON(!cfqq->dispatched); |
53c583d2 | 3435 | cfqd->rq_in_driver--; |
6d048f53 | 3436 | cfqq->dispatched--; |
80bdf0c7 | 3437 | (RQ_CFQG(rq))->dispatched--; |
e98ef89b VG |
3438 | cfq_blkiocg_update_completion_stats(&cfqq->cfqg->blkg, |
3439 | rq_start_time_ns(rq), rq_io_start_time_ns(rq), | |
3440 | rq_data_dir(rq), rq_is_sync(rq)); | |
1da177e4 | 3441 | |
53c583d2 | 3442 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
3ed9a296 | 3443 | |
365722bb | 3444 | if (sync) { |
5e705374 | 3445 | RQ_CIC(rq)->last_end_request = now; |
573412b2 CZ |
3446 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
3447 | cfqd->last_delayed_sync = now; | |
365722bb | 3448 | } |
caaa5f9f JA |
3449 | |
3450 | /* | |
3451 | * If this is the active queue, check if it needs to be expired, | |
3452 | * or if we want to idle in case it has no pending requests. | |
3453 | */ | |
3454 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3455 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3456 | ||
44f7c160 JA |
3457 | if (cfq_cfqq_slice_new(cfqq)) { |
3458 | cfq_set_prio_slice(cfqd, cfqq); | |
3459 | cfq_clear_cfqq_slice_new(cfqq); | |
3460 | } | |
f75edf2d VG |
3461 | |
3462 | /* | |
7667aa06 VG |
3463 | * Should we wait for next request to come in before we expire |
3464 | * the queue. | |
f75edf2d | 3465 | */ |
7667aa06 | 3466 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
80bdf0c7 VG |
3467 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
3468 | if (!cfqd->cfq_slice_idle) | |
3469 | extend_sl = cfqd->cfq_group_idle; | |
3470 | cfqq->slice_end = jiffies + extend_sl; | |
f75edf2d | 3471 | cfq_mark_cfqq_wait_busy(cfqq); |
b1ffe737 | 3472 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
f75edf2d VG |
3473 | } |
3474 | ||
a36e71f9 | 3475 | /* |
8e550632 CZ |
3476 | * Idling is not enabled on: |
3477 | * - expired queues | |
3478 | * - idle-priority queues | |
3479 | * - async queues | |
3480 | * - queues with still some requests queued | |
3481 | * - when there is a close cooperator | |
a36e71f9 | 3482 | */ |
0871714e | 3483 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
e5ff082e | 3484 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3485 | else if (sync && cfqq_empty && |
3486 | !cfq_close_cooperator(cfqd, cfqq)) { | |
33659ebb CH |
3487 | cfqd->noidle_tree_requires_idle |= |
3488 | !(rq->cmd_flags & REQ_NOIDLE); | |
8e550632 CZ |
3489 | /* |
3490 | * Idling is enabled for SYNC_WORKLOAD. | |
3491 | * SYNC_NOIDLE_WORKLOAD idles at the end of the tree | |
33659ebb | 3492 | * only if we processed at least one !REQ_NOIDLE request |
8e550632 CZ |
3493 | */ |
3494 | if (cfqd->serving_type == SYNC_WORKLOAD | |
c04645e5 VG |
3495 | || cfqd->noidle_tree_requires_idle |
3496 | || cfqq->cfqg->nr_cfqq == 1) | |
8e550632 CZ |
3497 | cfq_arm_slice_timer(cfqd); |
3498 | } | |
caaa5f9f | 3499 | } |
6d048f53 | 3500 | |
53c583d2 | 3501 | if (!cfqd->rq_in_driver) |
23e018a1 | 3502 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3503 | } |
3504 | ||
22e2c507 JA |
3505 | /* |
3506 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
3507 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
3508 | */ | |
3509 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 3510 | { |
22e2c507 JA |
3511 | if (has_fs_excl()) { |
3512 | /* | |
3513 | * boost idle prio on transactions that would lock out other | |
3514 | * users of the filesystem | |
3515 | */ | |
3516 | if (cfq_class_idle(cfqq)) | |
3517 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
3518 | if (cfqq->ioprio > IOPRIO_NORM) | |
3519 | cfqq->ioprio = IOPRIO_NORM; | |
3520 | } else { | |
3521 | /* | |
dddb7451 | 3522 | * unboost the queue (if needed) |
22e2c507 | 3523 | */ |
dddb7451 CZ |
3524 | cfqq->ioprio_class = cfqq->org_ioprio_class; |
3525 | cfqq->ioprio = cfqq->org_ioprio; | |
22e2c507 | 3526 | } |
22e2c507 | 3527 | } |
1da177e4 | 3528 | |
89850f7e | 3529 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3530 | { |
1b379d8d | 3531 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3532 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3533 | return ELV_MQUEUE_MUST; |
3b18152c | 3534 | } |
1da177e4 | 3535 | |
22e2c507 | 3536 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3537 | } |
3538 | ||
165125e1 | 3539 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3540 | { |
3541 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3542 | struct task_struct *tsk = current; | |
91fac317 | 3543 | struct cfq_io_context *cic; |
22e2c507 JA |
3544 | struct cfq_queue *cfqq; |
3545 | ||
3546 | /* | |
3547 | * don't force setup of a queue from here, as a call to may_queue | |
3548 | * does not necessarily imply that a request actually will be queued. | |
3549 | * so just lookup a possibly existing queue, or return 'may queue' | |
3550 | * if that fails | |
3551 | */ | |
4ac845a2 | 3552 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3553 | if (!cic) |
3554 | return ELV_MQUEUE_MAY; | |
3555 | ||
b0b78f81 | 3556 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3557 | if (cfqq) { |
fd0928df | 3558 | cfq_init_prio_data(cfqq, cic->ioc); |
22e2c507 JA |
3559 | cfq_prio_boost(cfqq); |
3560 | ||
89850f7e | 3561 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3562 | } |
3563 | ||
3564 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3565 | } |
3566 | ||
1da177e4 LT |
3567 | /* |
3568 | * queue lock held here | |
3569 | */ | |
bb37b94c | 3570 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3571 | { |
5e705374 | 3572 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3573 | |
5e705374 | 3574 | if (cfqq) { |
22e2c507 | 3575 | const int rw = rq_data_dir(rq); |
1da177e4 | 3576 | |
22e2c507 JA |
3577 | BUG_ON(!cfqq->allocated[rw]); |
3578 | cfqq->allocated[rw]--; | |
1da177e4 | 3579 | |
5e705374 | 3580 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 3581 | |
1da177e4 | 3582 | rq->elevator_private = NULL; |
5e705374 | 3583 | rq->elevator_private2 = NULL; |
1da177e4 | 3584 | |
7f1dc8a2 VG |
3585 | /* Put down rq reference on cfqg */ |
3586 | cfq_put_cfqg(RQ_CFQG(rq)); | |
3587 | rq->elevator_private3 = NULL; | |
3588 | ||
1da177e4 LT |
3589 | cfq_put_queue(cfqq); |
3590 | } | |
3591 | } | |
3592 | ||
df5fe3e8 JM |
3593 | static struct cfq_queue * |
3594 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_context *cic, | |
3595 | struct cfq_queue *cfqq) | |
3596 | { | |
3597 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3598 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3599 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3600 | cfq_put_queue(cfqq); |
3601 | return cic_to_cfqq(cic, 1); | |
3602 | } | |
3603 | ||
e6c5bc73 JM |
3604 | /* |
3605 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3606 | * was the last process referring to said cfqq. | |
3607 | */ | |
3608 | static struct cfq_queue * | |
3609 | split_cfqq(struct cfq_io_context *cic, struct cfq_queue *cfqq) | |
3610 | { | |
3611 | if (cfqq_process_refs(cfqq) == 1) { | |
e6c5bc73 JM |
3612 | cfqq->pid = current->pid; |
3613 | cfq_clear_cfqq_coop(cfqq); | |
ae54abed | 3614 | cfq_clear_cfqq_split_coop(cfqq); |
e6c5bc73 JM |
3615 | return cfqq; |
3616 | } | |
3617 | ||
3618 | cic_set_cfqq(cic, NULL, 1); | |
d02a2c07 SL |
3619 | |
3620 | cfq_put_cooperator(cfqq); | |
3621 | ||
e6c5bc73 JM |
3622 | cfq_put_queue(cfqq); |
3623 | return NULL; | |
3624 | } | |
1da177e4 | 3625 | /* |
22e2c507 | 3626 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3627 | */ |
22e2c507 | 3628 | static int |
165125e1 | 3629 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
3630 | { |
3631 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3632 | struct cfq_io_context *cic; | |
3633 | const int rw = rq_data_dir(rq); | |
a6151c3a | 3634 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3635 | struct cfq_queue *cfqq; |
1da177e4 LT |
3636 | unsigned long flags; |
3637 | ||
3638 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3639 | ||
e2d74ac0 | 3640 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 3641 | |
1da177e4 LT |
3642 | spin_lock_irqsave(q->queue_lock, flags); |
3643 | ||
22e2c507 JA |
3644 | if (!cic) |
3645 | goto queue_fail; | |
3646 | ||
e6c5bc73 | 3647 | new_queue: |
91fac317 | 3648 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3649 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
fd0928df | 3650 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
91fac317 | 3651 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3652 | } else { |
e6c5bc73 JM |
3653 | /* |
3654 | * If the queue was seeky for too long, break it apart. | |
3655 | */ | |
ae54abed | 3656 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
e6c5bc73 JM |
3657 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
3658 | cfqq = split_cfqq(cic, cfqq); | |
3659 | if (!cfqq) | |
3660 | goto new_queue; | |
3661 | } | |
3662 | ||
df5fe3e8 JM |
3663 | /* |
3664 | * Check to see if this queue is scheduled to merge with | |
3665 | * another, closely cooperating queue. The merging of | |
3666 | * queues happens here as it must be done in process context. | |
3667 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3668 | */ | |
3669 | if (cfqq->new_cfqq) | |
3670 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3671 | } |
1da177e4 LT |
3672 | |
3673 | cfqq->allocated[rw]++; | |
22e2c507 | 3674 | atomic_inc(&cfqq->ref); |
1da177e4 | 3675 | |
5e705374 | 3676 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 3677 | |
5e705374 JA |
3678 | rq->elevator_private = cic; |
3679 | rq->elevator_private2 = cfqq; | |
7f1dc8a2 | 3680 | rq->elevator_private3 = cfq_ref_get_cfqg(cfqq->cfqg); |
5e705374 | 3681 | return 0; |
1da177e4 | 3682 | |
22e2c507 JA |
3683 | queue_fail: |
3684 | if (cic) | |
3685 | put_io_context(cic->ioc); | |
89850f7e | 3686 | |
23e018a1 | 3687 | cfq_schedule_dispatch(cfqd); |
1da177e4 | 3688 | spin_unlock_irqrestore(q->queue_lock, flags); |
7b679138 | 3689 | cfq_log(cfqd, "set_request fail"); |
1da177e4 LT |
3690 | return 1; |
3691 | } | |
3692 | ||
65f27f38 | 3693 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3694 | { |
65f27f38 | 3695 | struct cfq_data *cfqd = |
23e018a1 | 3696 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3697 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3698 | |
40bb54d1 | 3699 | spin_lock_irq(q->queue_lock); |
a7f55792 | 3700 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3701 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3702 | } |
3703 | ||
3704 | /* | |
3705 | * Timer running if the active_queue is currently idling inside its time slice | |
3706 | */ | |
3707 | static void cfq_idle_slice_timer(unsigned long data) | |
3708 | { | |
3709 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3710 | struct cfq_queue *cfqq; | |
3711 | unsigned long flags; | |
3c6bd2f8 | 3712 | int timed_out = 1; |
22e2c507 | 3713 | |
7b679138 JA |
3714 | cfq_log(cfqd, "idle timer fired"); |
3715 | ||
22e2c507 JA |
3716 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3717 | ||
fe094d98 JA |
3718 | cfqq = cfqd->active_queue; |
3719 | if (cfqq) { | |
3c6bd2f8 JA |
3720 | timed_out = 0; |
3721 | ||
b029195d JA |
3722 | /* |
3723 | * We saw a request before the queue expired, let it through | |
3724 | */ | |
3725 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3726 | goto out_kick; | |
3727 | ||
22e2c507 JA |
3728 | /* |
3729 | * expired | |
3730 | */ | |
44f7c160 | 3731 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
3732 | goto expire; |
3733 | ||
3734 | /* | |
3735 | * only expire and reinvoke request handler, if there are | |
3736 | * other queues with pending requests | |
3737 | */ | |
caaa5f9f | 3738 | if (!cfqd->busy_queues) |
22e2c507 | 3739 | goto out_cont; |
22e2c507 JA |
3740 | |
3741 | /* | |
3742 | * not expired and it has a request pending, let it dispatch | |
3743 | */ | |
75e50984 | 3744 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3745 | goto out_kick; |
76280aff CZ |
3746 | |
3747 | /* | |
3748 | * Queue depth flag is reset only when the idle didn't succeed | |
3749 | */ | |
3750 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
3751 | } |
3752 | expire: | |
e5ff082e | 3753 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 3754 | out_kick: |
23e018a1 | 3755 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
3756 | out_cont: |
3757 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3758 | } | |
3759 | ||
3b18152c JA |
3760 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3761 | { | |
3762 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 3763 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 3764 | } |
22e2c507 | 3765 | |
c2dea2d1 VT |
3766 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3767 | { | |
3768 | int i; | |
3769 | ||
3770 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
3771 | if (cfqd->async_cfqq[0][i]) | |
3772 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
3773 | if (cfqd->async_cfqq[1][i]) | |
3774 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 3775 | } |
2389d1ef ON |
3776 | |
3777 | if (cfqd->async_idle_cfqq) | |
3778 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
3779 | } |
3780 | ||
bb729bc9 JA |
3781 | static void cfq_cfqd_free(struct rcu_head *head) |
3782 | { | |
3783 | kfree(container_of(head, struct cfq_data, rcu)); | |
3784 | } | |
3785 | ||
b374d18a | 3786 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 3787 | { |
22e2c507 | 3788 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 3789 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3790 | |
3b18152c | 3791 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 3792 | |
d9ff4187 | 3793 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3794 | |
d9ff4187 | 3795 | if (cfqd->active_queue) |
e5ff082e | 3796 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
3797 | |
3798 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
3799 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
3800 | struct cfq_io_context, | |
3801 | queue_list); | |
89850f7e JA |
3802 | |
3803 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 3804 | } |
e2d74ac0 | 3805 | |
c2dea2d1 | 3806 | cfq_put_async_queues(cfqd); |
b1c35769 | 3807 | cfq_release_cfq_groups(cfqd); |
e98ef89b | 3808 | cfq_blkiocg_del_blkio_group(&cfqd->root_group.blkg); |
15c31be4 | 3809 | |
d9ff4187 | 3810 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
3811 | |
3812 | cfq_shutdown_timer_wq(cfqd); | |
3813 | ||
80b15c73 KK |
3814 | spin_lock(&cic_index_lock); |
3815 | ida_remove(&cic_index_ida, cfqd->cic_index); | |
3816 | spin_unlock(&cic_index_lock); | |
3817 | ||
b1c35769 | 3818 | /* Wait for cfqg->blkg->key accessors to exit their grace periods. */ |
bb729bc9 | 3819 | call_rcu(&cfqd->rcu, cfq_cfqd_free); |
1da177e4 LT |
3820 | } |
3821 | ||
80b15c73 KK |
3822 | static int cfq_alloc_cic_index(void) |
3823 | { | |
3824 | int index, error; | |
3825 | ||
3826 | do { | |
3827 | if (!ida_pre_get(&cic_index_ida, GFP_KERNEL)) | |
3828 | return -ENOMEM; | |
3829 | ||
3830 | spin_lock(&cic_index_lock); | |
3831 | error = ida_get_new(&cic_index_ida, &index); | |
3832 | spin_unlock(&cic_index_lock); | |
3833 | if (error && error != -EAGAIN) | |
3834 | return error; | |
3835 | } while (error); | |
3836 | ||
3837 | return index; | |
3838 | } | |
3839 | ||
165125e1 | 3840 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
3841 | { |
3842 | struct cfq_data *cfqd; | |
718eee05 | 3843 | int i, j; |
cdb16e8f | 3844 | struct cfq_group *cfqg; |
615f0259 | 3845 | struct cfq_rb_root *st; |
1da177e4 | 3846 | |
80b15c73 KK |
3847 | i = cfq_alloc_cic_index(); |
3848 | if (i < 0) | |
3849 | return NULL; | |
3850 | ||
94f6030c | 3851 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
1da177e4 | 3852 | if (!cfqd) |
bc1c1169 | 3853 | return NULL; |
1da177e4 | 3854 | |
80b15c73 KK |
3855 | cfqd->cic_index = i; |
3856 | ||
1fa8f6d6 VG |
3857 | /* Init root service tree */ |
3858 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
3859 | ||
cdb16e8f VG |
3860 | /* Init root group */ |
3861 | cfqg = &cfqd->root_group; | |
615f0259 VG |
3862 | for_each_cfqg_st(cfqg, i, j, st) |
3863 | *st = CFQ_RB_ROOT; | |
1fa8f6d6 | 3864 | RB_CLEAR_NODE(&cfqg->rb_node); |
26a2ac00 | 3865 | |
25bc6b07 VG |
3866 | /* Give preference to root group over other groups */ |
3867 | cfqg->weight = 2*BLKIO_WEIGHT_DEFAULT; | |
3868 | ||
25fb5169 | 3869 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
b1c35769 VG |
3870 | /* |
3871 | * Take a reference to root group which we never drop. This is just | |
3872 | * to make sure that cfq_put_cfqg() does not try to kfree root group | |
3873 | */ | |
3874 | atomic_set(&cfqg->ref, 1); | |
dcf097b2 | 3875 | rcu_read_lock(); |
e98ef89b VG |
3876 | cfq_blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg, |
3877 | (void *)cfqd, 0); | |
dcf097b2 | 3878 | rcu_read_unlock(); |
25fb5169 | 3879 | #endif |
26a2ac00 JA |
3880 | /* |
3881 | * Not strictly needed (since RB_ROOT just clears the node and we | |
3882 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
3883 | * to add magic to the rb code | |
3884 | */ | |
3885 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
3886 | cfqd->prio_trees[i] = RB_ROOT; | |
3887 | ||
6118b70b JA |
3888 | /* |
3889 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
3890 | * Grab a permanent reference to it, so that the normal code flow | |
3891 | * will not attempt to free it. | |
3892 | */ | |
3893 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
3894 | atomic_inc(&cfqd->oom_cfqq.ref); | |
cdb16e8f | 3895 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group); |
6118b70b | 3896 | |
d9ff4187 | 3897 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 3898 | |
1da177e4 | 3899 | cfqd->queue = q; |
1da177e4 | 3900 | |
22e2c507 JA |
3901 | init_timer(&cfqd->idle_slice_timer); |
3902 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
3903 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
3904 | ||
23e018a1 | 3905 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 3906 | |
1da177e4 | 3907 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
3908 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
3909 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
3910 | cfqd->cfq_back_max = cfq_back_max; |
3911 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
3912 | cfqd->cfq_slice[0] = cfq_slice_async; |
3913 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
3914 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
3915 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
80bdf0c7 | 3916 | cfqd->cfq_group_idle = cfq_group_idle; |
963b72fc | 3917 | cfqd->cfq_latency = 1; |
ae30c286 | 3918 | cfqd->cfq_group_isolation = 0; |
e459dd08 | 3919 | cfqd->hw_tag = -1; |
edc71131 CZ |
3920 | /* |
3921 | * we optimistically start assuming sync ops weren't delayed in last | |
3922 | * second, in order to have larger depth for async operations. | |
3923 | */ | |
573412b2 | 3924 | cfqd->last_delayed_sync = jiffies - HZ; |
bc1c1169 | 3925 | return cfqd; |
1da177e4 LT |
3926 | } |
3927 | ||
3928 | static void cfq_slab_kill(void) | |
3929 | { | |
d6de8be7 JA |
3930 | /* |
3931 | * Caller already ensured that pending RCU callbacks are completed, | |
3932 | * so we should have no busy allocations at this point. | |
3933 | */ | |
1da177e4 LT |
3934 | if (cfq_pool) |
3935 | kmem_cache_destroy(cfq_pool); | |
3936 | if (cfq_ioc_pool) | |
3937 | kmem_cache_destroy(cfq_ioc_pool); | |
3938 | } | |
3939 | ||
3940 | static int __init cfq_slab_setup(void) | |
3941 | { | |
0a31bd5f | 3942 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
1da177e4 LT |
3943 | if (!cfq_pool) |
3944 | goto fail; | |
3945 | ||
34e6bbf2 | 3946 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
1da177e4 LT |
3947 | if (!cfq_ioc_pool) |
3948 | goto fail; | |
3949 | ||
3950 | return 0; | |
3951 | fail: | |
3952 | cfq_slab_kill(); | |
3953 | return -ENOMEM; | |
3954 | } | |
3955 | ||
1da177e4 LT |
3956 | /* |
3957 | * sysfs parts below --> | |
3958 | */ | |
1da177e4 LT |
3959 | static ssize_t |
3960 | cfq_var_show(unsigned int var, char *page) | |
3961 | { | |
3962 | return sprintf(page, "%d\n", var); | |
3963 | } | |
3964 | ||
3965 | static ssize_t | |
3966 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
3967 | { | |
3968 | char *p = (char *) page; | |
3969 | ||
3970 | *var = simple_strtoul(p, &p, 10); | |
3971 | return count; | |
3972 | } | |
3973 | ||
1da177e4 | 3974 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 3975 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 3976 | { \ |
3d1ab40f | 3977 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3978 | unsigned int __data = __VAR; \ |
3979 | if (__CONV) \ | |
3980 | __data = jiffies_to_msecs(__data); \ | |
3981 | return cfq_var_show(__data, (page)); \ | |
3982 | } | |
3983 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
3984 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
3985 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
3986 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
3987 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 | 3988 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
80bdf0c7 | 3989 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
22e2c507 JA |
3990 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
3991 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
3992 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 3993 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
ae30c286 | 3994 | SHOW_FUNCTION(cfq_group_isolation_show, cfqd->cfq_group_isolation, 0); |
1da177e4 LT |
3995 | #undef SHOW_FUNCTION |
3996 | ||
3997 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 3998 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 3999 | { \ |
3d1ab40f | 4000 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
4001 | unsigned int __data; \ |
4002 | int ret = cfq_var_store(&__data, (page), count); \ | |
4003 | if (__data < (MIN)) \ | |
4004 | __data = (MIN); \ | |
4005 | else if (__data > (MAX)) \ | |
4006 | __data = (MAX); \ | |
4007 | if (__CONV) \ | |
4008 | *(__PTR) = msecs_to_jiffies(__data); \ | |
4009 | else \ | |
4010 | *(__PTR) = __data; \ | |
4011 | return ret; \ | |
4012 | } | |
4013 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
4014 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
4015 | UINT_MAX, 1); | |
4016 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
4017 | UINT_MAX, 1); | |
e572ec7e | 4018 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
4019 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
4020 | UINT_MAX, 0); | |
22e2c507 | 4021 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
80bdf0c7 | 4022 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
22e2c507 JA |
4023 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
4024 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
4025 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
4026 | UINT_MAX, 0); | |
963b72fc | 4027 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
ae30c286 | 4028 | STORE_FUNCTION(cfq_group_isolation_store, &cfqd->cfq_group_isolation, 0, 1, 0); |
1da177e4 LT |
4029 | #undef STORE_FUNCTION |
4030 | ||
e572ec7e AV |
4031 | #define CFQ_ATTR(name) \ |
4032 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
4033 | ||
4034 | static struct elv_fs_entry cfq_attrs[] = { | |
4035 | CFQ_ATTR(quantum), | |
e572ec7e AV |
4036 | CFQ_ATTR(fifo_expire_sync), |
4037 | CFQ_ATTR(fifo_expire_async), | |
4038 | CFQ_ATTR(back_seek_max), | |
4039 | CFQ_ATTR(back_seek_penalty), | |
4040 | CFQ_ATTR(slice_sync), | |
4041 | CFQ_ATTR(slice_async), | |
4042 | CFQ_ATTR(slice_async_rq), | |
4043 | CFQ_ATTR(slice_idle), | |
80bdf0c7 | 4044 | CFQ_ATTR(group_idle), |
963b72fc | 4045 | CFQ_ATTR(low_latency), |
ae30c286 | 4046 | CFQ_ATTR(group_isolation), |
e572ec7e | 4047 | __ATTR_NULL |
1da177e4 LT |
4048 | }; |
4049 | ||
1da177e4 LT |
4050 | static struct elevator_type iosched_cfq = { |
4051 | .ops = { | |
4052 | .elevator_merge_fn = cfq_merge, | |
4053 | .elevator_merged_fn = cfq_merged_request, | |
4054 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 4055 | .elevator_allow_merge_fn = cfq_allow_merge, |
812d4026 | 4056 | .elevator_bio_merged_fn = cfq_bio_merged, |
b4878f24 | 4057 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 4058 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 4059 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
4060 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
4061 | .elevator_queue_empty_fn = cfq_queue_empty, | |
4062 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
4063 | .elevator_former_req_fn = elv_rb_former_request, |
4064 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
4065 | .elevator_set_req_fn = cfq_set_request, |
4066 | .elevator_put_req_fn = cfq_put_request, | |
4067 | .elevator_may_queue_fn = cfq_may_queue, | |
4068 | .elevator_init_fn = cfq_init_queue, | |
4069 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 4070 | .trim = cfq_free_io_context, |
1da177e4 | 4071 | }, |
3d1ab40f | 4072 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
4073 | .elevator_name = "cfq", |
4074 | .elevator_owner = THIS_MODULE, | |
4075 | }; | |
4076 | ||
3e252066 VG |
4077 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4078 | static struct blkio_policy_type blkio_policy_cfq = { | |
4079 | .ops = { | |
4080 | .blkio_unlink_group_fn = cfq_unlink_blkio_group, | |
4081 | .blkio_update_group_weight_fn = cfq_update_blkio_group_weight, | |
4082 | }, | |
4083 | }; | |
4084 | #else | |
4085 | static struct blkio_policy_type blkio_policy_cfq; | |
4086 | #endif | |
4087 | ||
1da177e4 LT |
4088 | static int __init cfq_init(void) |
4089 | { | |
22e2c507 JA |
4090 | /* |
4091 | * could be 0 on HZ < 1000 setups | |
4092 | */ | |
4093 | if (!cfq_slice_async) | |
4094 | cfq_slice_async = 1; | |
4095 | if (!cfq_slice_idle) | |
4096 | cfq_slice_idle = 1; | |
4097 | ||
80bdf0c7 VG |
4098 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
4099 | if (!cfq_group_idle) | |
4100 | cfq_group_idle = 1; | |
4101 | #else | |
4102 | cfq_group_idle = 0; | |
4103 | #endif | |
1da177e4 LT |
4104 | if (cfq_slab_setup()) |
4105 | return -ENOMEM; | |
4106 | ||
2fdd82bd | 4107 | elv_register(&iosched_cfq); |
3e252066 | 4108 | blkio_policy_register(&blkio_policy_cfq); |
1da177e4 | 4109 | |
2fdd82bd | 4110 | return 0; |
1da177e4 LT |
4111 | } |
4112 | ||
4113 | static void __exit cfq_exit(void) | |
4114 | { | |
6e9a4738 | 4115 | DECLARE_COMPLETION_ONSTACK(all_gone); |
3e252066 | 4116 | blkio_policy_unregister(&blkio_policy_cfq); |
1da177e4 | 4117 | elv_unregister(&iosched_cfq); |
334e94de | 4118 | ioc_gone = &all_gone; |
fba82272 OH |
4119 | /* ioc_gone's update must be visible before reading ioc_count */ |
4120 | smp_wmb(); | |
d6de8be7 JA |
4121 | |
4122 | /* | |
4123 | * this also protects us from entering cfq_slab_kill() with | |
4124 | * pending RCU callbacks | |
4125 | */ | |
245b2e70 | 4126 | if (elv_ioc_count_read(cfq_ioc_count)) |
9a11b4ed | 4127 | wait_for_completion(&all_gone); |
80b15c73 | 4128 | ida_destroy(&cic_index_ida); |
83521d3e | 4129 | cfq_slab_kill(); |
1da177e4 LT |
4130 | } |
4131 | ||
4132 | module_init(cfq_init); | |
4133 | module_exit(cfq_exit); | |
4134 | ||
4135 | MODULE_AUTHOR("Jens Axboe"); | |
4136 | MODULE_LICENSE("GPL"); | |
4137 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |