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