]>
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> |
1cc9be68 AV |
10 | #include <linux/blkdev.h> |
11 | #include <linux/elevator.h> | |
ad5ebd2f | 12 | #include <linux/jiffies.h> |
1da177e4 | 13 | #include <linux/rbtree.h> |
22e2c507 | 14 | #include <linux/ioprio.h> |
7b679138 | 15 | #include <linux/blktrace_api.h> |
25bc6b07 | 16 | #include "blk-cgroup.h" |
1da177e4 LT |
17 | |
18 | /* | |
19 | * tunables | |
20 | */ | |
fe094d98 JA |
21 | /* max queue in one round of service */ |
22 | static const int cfq_quantum = 4; | |
64100099 | 23 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
fe094d98 JA |
24 | /* maximum backwards seek, in KiB */ |
25 | static const int cfq_back_max = 16 * 1024; | |
26 | /* penalty of a backwards seek */ | |
27 | static const int cfq_back_penalty = 2; | |
64100099 | 28 | static const int cfq_slice_sync = HZ / 10; |
3b18152c | 29 | static int cfq_slice_async = HZ / 25; |
64100099 | 30 | static const int cfq_slice_async_rq = 2; |
caaa5f9f | 31 | static int cfq_slice_idle = HZ / 125; |
5db5d642 CZ |
32 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
33 | static const int cfq_hist_divisor = 4; | |
22e2c507 | 34 | |
d9e7620e | 35 | /* |
0871714e | 36 | * offset from end of service tree |
d9e7620e | 37 | */ |
0871714e | 38 | #define CFQ_IDLE_DELAY (HZ / 5) |
d9e7620e JA |
39 | |
40 | /* | |
41 | * below this threshold, we consider thinktime immediate | |
42 | */ | |
43 | #define CFQ_MIN_TT (2) | |
44 | ||
e6c5bc73 JM |
45 | /* |
46 | * Allow merged cfqqs to perform this amount of seeky I/O before | |
47 | * deciding to break the queues up again. | |
48 | */ | |
49 | #define CFQQ_COOP_TOUT (HZ) | |
50 | ||
22e2c507 | 51 | #define CFQ_SLICE_SCALE (5) |
45333d5a | 52 | #define CFQ_HW_QUEUE_MIN (5) |
25bc6b07 | 53 | #define CFQ_SERVICE_SHIFT 12 |
22e2c507 | 54 | |
fe094d98 JA |
55 | #define RQ_CIC(rq) \ |
56 | ((struct cfq_io_context *) (rq)->elevator_private) | |
7b679138 | 57 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private2) |
1da177e4 | 58 | |
e18b890b CL |
59 | static struct kmem_cache *cfq_pool; |
60 | static struct kmem_cache *cfq_ioc_pool; | |
1da177e4 | 61 | |
245b2e70 | 62 | static DEFINE_PER_CPU(unsigned long, cfq_ioc_count); |
334e94de | 63 | static struct completion *ioc_gone; |
9a11b4ed | 64 | static DEFINE_SPINLOCK(ioc_gone_lock); |
334e94de | 65 | |
22e2c507 JA |
66 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
67 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
22e2c507 JA |
68 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
69 | ||
206dc69b | 70 | #define sample_valid(samples) ((samples) > 80) |
1fa8f6d6 | 71 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
206dc69b | 72 | |
cc09e299 JA |
73 | /* |
74 | * Most of our rbtree usage is for sorting with min extraction, so | |
75 | * if we cache the leftmost node we don't have to walk down the tree | |
76 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should | |
77 | * move this into the elevator for the rq sorting as well. | |
78 | */ | |
79 | struct cfq_rb_root { | |
80 | struct rb_root rb; | |
81 | struct rb_node *left; | |
aa6f6a3d | 82 | unsigned count; |
1fa8f6d6 | 83 | u64 min_vdisktime; |
25bc6b07 | 84 | struct rb_node *active; |
58ff82f3 | 85 | unsigned total_weight; |
cc09e299 | 86 | }; |
1fa8f6d6 | 87 | #define CFQ_RB_ROOT (struct cfq_rb_root) { RB_ROOT, NULL, 0, 0, } |
cc09e299 | 88 | |
6118b70b JA |
89 | /* |
90 | * Per process-grouping structure | |
91 | */ | |
92 | struct cfq_queue { | |
93 | /* reference count */ | |
94 | atomic_t ref; | |
95 | /* various state flags, see below */ | |
96 | unsigned int flags; | |
97 | /* parent cfq_data */ | |
98 | struct cfq_data *cfqd; | |
99 | /* service_tree member */ | |
100 | struct rb_node rb_node; | |
101 | /* service_tree key */ | |
102 | unsigned long rb_key; | |
103 | /* prio tree member */ | |
104 | struct rb_node p_node; | |
105 | /* prio tree root we belong to, if any */ | |
106 | struct rb_root *p_root; | |
107 | /* sorted list of pending requests */ | |
108 | struct rb_root sort_list; | |
109 | /* if fifo isn't expired, next request to serve */ | |
110 | struct request *next_rq; | |
111 | /* requests queued in sort_list */ | |
112 | int queued[2]; | |
113 | /* currently allocated requests */ | |
114 | int allocated[2]; | |
115 | /* fifo list of requests in sort_list */ | |
116 | struct list_head fifo; | |
117 | ||
dae739eb VG |
118 | /* time when queue got scheduled in to dispatch first request. */ |
119 | unsigned long dispatch_start; | |
120 | /* time when first request from queue completed and slice started. */ | |
121 | unsigned long slice_start; | |
6118b70b JA |
122 | unsigned long slice_end; |
123 | long slice_resid; | |
124 | unsigned int slice_dispatch; | |
125 | ||
126 | /* pending metadata requests */ | |
127 | int meta_pending; | |
128 | /* number of requests that are on the dispatch list or inside driver */ | |
129 | int dispatched; | |
130 | ||
131 | /* io prio of this group */ | |
132 | unsigned short ioprio, org_ioprio; | |
133 | unsigned short ioprio_class, org_ioprio_class; | |
134 | ||
b2c18e1e JM |
135 | unsigned int seek_samples; |
136 | u64 seek_total; | |
137 | sector_t seek_mean; | |
138 | sector_t last_request_pos; | |
e6c5bc73 | 139 | unsigned long seeky_start; |
b2c18e1e | 140 | |
6118b70b | 141 | pid_t pid; |
df5fe3e8 | 142 | |
aa6f6a3d | 143 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 144 | struct cfq_queue *new_cfqq; |
cdb16e8f | 145 | struct cfq_group *cfqg; |
22084190 VG |
146 | /* Sectors dispatched in current dispatch round */ |
147 | unsigned long nr_sectors; | |
6118b70b JA |
148 | }; |
149 | ||
c0324a02 | 150 | /* |
718eee05 | 151 | * First index in the service_trees. |
c0324a02 CZ |
152 | * IDLE is handled separately, so it has negative index |
153 | */ | |
154 | enum wl_prio_t { | |
c0324a02 | 155 | BE_WORKLOAD = 0, |
615f0259 VG |
156 | RT_WORKLOAD = 1, |
157 | IDLE_WORKLOAD = 2, | |
c0324a02 CZ |
158 | }; |
159 | ||
718eee05 CZ |
160 | /* |
161 | * Second index in the service_trees. | |
162 | */ | |
163 | enum wl_type_t { | |
164 | ASYNC_WORKLOAD = 0, | |
165 | SYNC_NOIDLE_WORKLOAD = 1, | |
166 | SYNC_WORKLOAD = 2 | |
167 | }; | |
168 | ||
cdb16e8f VG |
169 | /* This is per cgroup per device grouping structure */ |
170 | struct cfq_group { | |
1fa8f6d6 VG |
171 | /* group service_tree member */ |
172 | struct rb_node rb_node; | |
173 | ||
174 | /* group service_tree key */ | |
175 | u64 vdisktime; | |
25bc6b07 | 176 | unsigned int weight; |
1fa8f6d6 VG |
177 | bool on_st; |
178 | ||
179 | /* number of cfqq currently on this group */ | |
180 | int nr_cfqq; | |
181 | ||
58ff82f3 VG |
182 | /* Per group busy queus average. Useful for workload slice calc. */ |
183 | unsigned int busy_queues_avg[2]; | |
cdb16e8f VG |
184 | /* |
185 | * rr lists of queues with requests, onle rr for each priority class. | |
186 | * Counts are embedded in the cfq_rb_root | |
187 | */ | |
188 | struct cfq_rb_root service_trees[2][3]; | |
189 | struct cfq_rb_root service_tree_idle; | |
dae739eb VG |
190 | |
191 | unsigned long saved_workload_slice; | |
192 | enum wl_type_t saved_workload; | |
193 | enum wl_prio_t saved_serving_prio; | |
25fb5169 VG |
194 | struct blkio_group blkg; |
195 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
196 | struct hlist_node cfqd_node; | |
b1c35769 | 197 | atomic_t ref; |
25fb5169 | 198 | #endif |
cdb16e8f | 199 | }; |
718eee05 | 200 | |
22e2c507 JA |
201 | /* |
202 | * Per block device queue structure | |
203 | */ | |
1da177e4 | 204 | struct cfq_data { |
165125e1 | 205 | struct request_queue *queue; |
1fa8f6d6 VG |
206 | /* Root service tree for cfq_groups */ |
207 | struct cfq_rb_root grp_service_tree; | |
cdb16e8f | 208 | struct cfq_group root_group; |
58ff82f3 VG |
209 | /* Number of active cfq groups on group service tree */ |
210 | int nr_groups; | |
22e2c507 | 211 | |
c0324a02 CZ |
212 | /* |
213 | * The priority currently being served | |
22e2c507 | 214 | */ |
c0324a02 | 215 | enum wl_prio_t serving_prio; |
718eee05 CZ |
216 | enum wl_type_t serving_type; |
217 | unsigned long workload_expires; | |
cdb16e8f | 218 | struct cfq_group *serving_group; |
8e550632 | 219 | bool noidle_tree_requires_idle; |
a36e71f9 JA |
220 | |
221 | /* | |
222 | * Each priority tree is sorted by next_request position. These | |
223 | * trees are used when determining if two or more queues are | |
224 | * interleaving requests (see cfq_close_cooperator). | |
225 | */ | |
226 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
227 | ||
22e2c507 JA |
228 | unsigned int busy_queues; |
229 | ||
5ad531db | 230 | int rq_in_driver[2]; |
3ed9a296 | 231 | int sync_flight; |
45333d5a AC |
232 | |
233 | /* | |
234 | * queue-depth detection | |
235 | */ | |
236 | int rq_queued; | |
25776e35 | 237 | int hw_tag; |
e459dd08 CZ |
238 | /* |
239 | * hw_tag can be | |
240 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
241 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
242 | * 0 => no NCQ | |
243 | */ | |
244 | int hw_tag_est_depth; | |
245 | unsigned int hw_tag_samples; | |
1da177e4 | 246 | |
22e2c507 JA |
247 | /* |
248 | * idle window management | |
249 | */ | |
250 | struct timer_list idle_slice_timer; | |
23e018a1 | 251 | struct work_struct unplug_work; |
1da177e4 | 252 | |
22e2c507 JA |
253 | struct cfq_queue *active_queue; |
254 | struct cfq_io_context *active_cic; | |
22e2c507 | 255 | |
c2dea2d1 VT |
256 | /* |
257 | * async queue for each priority case | |
258 | */ | |
259 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
260 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 261 | |
6d048f53 | 262 | sector_t last_position; |
1da177e4 | 263 | |
1da177e4 LT |
264 | /* |
265 | * tunables, see top of file | |
266 | */ | |
267 | unsigned int cfq_quantum; | |
22e2c507 | 268 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
269 | unsigned int cfq_back_penalty; |
270 | unsigned int cfq_back_max; | |
22e2c507 JA |
271 | unsigned int cfq_slice[2]; |
272 | unsigned int cfq_slice_async_rq; | |
273 | unsigned int cfq_slice_idle; | |
963b72fc | 274 | unsigned int cfq_latency; |
d9ff4187 AV |
275 | |
276 | struct list_head cic_list; | |
1da177e4 | 277 | |
6118b70b JA |
278 | /* |
279 | * Fallback dummy cfqq for extreme OOM conditions | |
280 | */ | |
281 | struct cfq_queue oom_cfqq; | |
365722bb VG |
282 | |
283 | unsigned long last_end_sync_rq; | |
25fb5169 VG |
284 | |
285 | /* List of cfq groups being managed on this device*/ | |
286 | struct hlist_head cfqg_list; | |
1da177e4 LT |
287 | }; |
288 | ||
25fb5169 VG |
289 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
290 | ||
cdb16e8f VG |
291 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
292 | enum wl_prio_t prio, | |
718eee05 | 293 | enum wl_type_t type, |
c0324a02 CZ |
294 | struct cfq_data *cfqd) |
295 | { | |
1fa8f6d6 VG |
296 | if (!cfqg) |
297 | return NULL; | |
298 | ||
c0324a02 | 299 | if (prio == IDLE_WORKLOAD) |
cdb16e8f | 300 | return &cfqg->service_tree_idle; |
c0324a02 | 301 | |
cdb16e8f | 302 | return &cfqg->service_trees[prio][type]; |
c0324a02 CZ |
303 | } |
304 | ||
3b18152c | 305 | enum cfqq_state_flags { |
b0b8d749 JA |
306 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
307 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 308 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 309 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
310 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
311 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
312 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 313 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 314 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 315 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
76280aff | 316 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
3b18152c JA |
317 | }; |
318 | ||
319 | #define CFQ_CFQQ_FNS(name) \ | |
320 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
321 | { \ | |
fe094d98 | 322 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
323 | } \ |
324 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
325 | { \ | |
fe094d98 | 326 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
327 | } \ |
328 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
329 | { \ | |
fe094d98 | 330 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
331 | } |
332 | ||
333 | CFQ_CFQQ_FNS(on_rr); | |
334 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 335 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 336 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
337 | CFQ_CFQQ_FNS(fifo_expire); |
338 | CFQ_CFQQ_FNS(idle_window); | |
339 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 340 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 341 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 342 | CFQ_CFQQ_FNS(coop); |
76280aff | 343 | CFQ_CFQQ_FNS(deep); |
3b18152c JA |
344 | #undef CFQ_CFQQ_FNS |
345 | ||
2868ef7b VG |
346 | #ifdef CONFIG_DEBUG_CFQ_IOSCHED |
347 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ | |
348 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ | |
349 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
350 | blkg_path(&(cfqq)->cfqg->blkg), ##args); | |
351 | ||
352 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ | |
353 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ | |
354 | blkg_path(&(cfqg)->blkg), ##args); \ | |
355 | ||
356 | #else | |
7b679138 JA |
357 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
358 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
2868ef7b VG |
359 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0); |
360 | #endif | |
7b679138 JA |
361 | #define cfq_log(cfqd, fmt, args...) \ |
362 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
363 | ||
615f0259 VG |
364 | /* Traverses through cfq group service trees */ |
365 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
366 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
367 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
368 | : &cfqg->service_tree_idle; \ | |
369 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
370 | (i == IDLE_WORKLOAD && j == 0); \ | |
371 | j++, st = i < IDLE_WORKLOAD ? \ | |
372 | &cfqg->service_trees[i][j]: NULL) \ | |
373 | ||
374 | ||
c0324a02 CZ |
375 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
376 | { | |
377 | if (cfq_class_idle(cfqq)) | |
378 | return IDLE_WORKLOAD; | |
379 | if (cfq_class_rt(cfqq)) | |
380 | return RT_WORKLOAD; | |
381 | return BE_WORKLOAD; | |
382 | } | |
383 | ||
718eee05 CZ |
384 | |
385 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
386 | { | |
387 | if (!cfq_cfqq_sync(cfqq)) | |
388 | return ASYNC_WORKLOAD; | |
389 | if (!cfq_cfqq_idle_window(cfqq)) | |
390 | return SYNC_NOIDLE_WORKLOAD; | |
391 | return SYNC_WORKLOAD; | |
392 | } | |
393 | ||
58ff82f3 VG |
394 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
395 | struct cfq_data *cfqd, | |
396 | struct cfq_group *cfqg) | |
c0324a02 CZ |
397 | { |
398 | if (wl == IDLE_WORKLOAD) | |
cdb16e8f | 399 | return cfqg->service_tree_idle.count; |
c0324a02 | 400 | |
cdb16e8f VG |
401 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
402 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count | |
403 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
404 | } |
405 | ||
165125e1 | 406 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
a6151c3a | 407 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool, |
fd0928df | 408 | struct io_context *, gfp_t); |
4ac845a2 | 409 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
91fac317 VT |
410 | struct io_context *); |
411 | ||
5ad531db JA |
412 | static inline int rq_in_driver(struct cfq_data *cfqd) |
413 | { | |
414 | return cfqd->rq_in_driver[0] + cfqd->rq_in_driver[1]; | |
415 | } | |
416 | ||
91fac317 | 417 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, |
a6151c3a | 418 | bool is_sync) |
91fac317 | 419 | { |
a6151c3a | 420 | return cic->cfqq[is_sync]; |
91fac317 VT |
421 | } |
422 | ||
423 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
a6151c3a | 424 | struct cfq_queue *cfqq, bool is_sync) |
91fac317 | 425 | { |
a6151c3a | 426 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
427 | } |
428 | ||
429 | /* | |
430 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
431 | * set (in which case it could also be direct WRITE). | |
432 | */ | |
a6151c3a | 433 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 434 | { |
a6151c3a | 435 | return bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO); |
91fac317 | 436 | } |
1da177e4 | 437 | |
99f95e52 AM |
438 | /* |
439 | * scheduler run of queue, if there are requests pending and no one in the | |
440 | * driver that will restart queueing | |
441 | */ | |
23e018a1 | 442 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 443 | { |
7b679138 JA |
444 | if (cfqd->busy_queues) { |
445 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 446 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 447 | } |
99f95e52 AM |
448 | } |
449 | ||
165125e1 | 450 | static int cfq_queue_empty(struct request_queue *q) |
99f95e52 AM |
451 | { |
452 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
453 | ||
f04a6424 | 454 | return !cfqd->rq_queued; |
99f95e52 AM |
455 | } |
456 | ||
44f7c160 JA |
457 | /* |
458 | * Scale schedule slice based on io priority. Use the sync time slice only | |
459 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
460 | * io only, should not get full sync slice length. | |
461 | */ | |
a6151c3a | 462 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 463 | unsigned short prio) |
44f7c160 | 464 | { |
d9e7620e | 465 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 466 | |
d9e7620e JA |
467 | WARN_ON(prio >= IOPRIO_BE_NR); |
468 | ||
469 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
470 | } | |
44f7c160 | 471 | |
d9e7620e JA |
472 | static inline int |
473 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
474 | { | |
475 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
476 | } |
477 | ||
25bc6b07 VG |
478 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
479 | { | |
480 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
481 | ||
482 | d = d * BLKIO_WEIGHT_DEFAULT; | |
483 | do_div(d, cfqg->weight); | |
484 | return d; | |
485 | } | |
486 | ||
487 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
488 | { | |
489 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
490 | if (delta > 0) | |
491 | min_vdisktime = vdisktime; | |
492 | ||
493 | return min_vdisktime; | |
494 | } | |
495 | ||
496 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
497 | { | |
498 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
499 | if (delta < 0) | |
500 | min_vdisktime = vdisktime; | |
501 | ||
502 | return min_vdisktime; | |
503 | } | |
504 | ||
505 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
506 | { | |
507 | u64 vdisktime = st->min_vdisktime; | |
508 | struct cfq_group *cfqg; | |
509 | ||
510 | if (st->active) { | |
511 | cfqg = rb_entry_cfqg(st->active); | |
512 | vdisktime = cfqg->vdisktime; | |
513 | } | |
514 | ||
515 | if (st->left) { | |
516 | cfqg = rb_entry_cfqg(st->left); | |
517 | vdisktime = min_vdisktime(vdisktime, cfqg->vdisktime); | |
518 | } | |
519 | ||
520 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, vdisktime); | |
521 | } | |
522 | ||
5db5d642 CZ |
523 | /* |
524 | * get averaged number of queues of RT/BE priority. | |
525 | * average is updated, with a formula that gives more weight to higher numbers, | |
526 | * to quickly follows sudden increases and decrease slowly | |
527 | */ | |
528 | ||
58ff82f3 VG |
529 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
530 | struct cfq_group *cfqg, bool rt) | |
5869619c | 531 | { |
5db5d642 CZ |
532 | unsigned min_q, max_q; |
533 | unsigned mult = cfq_hist_divisor - 1; | |
534 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 535 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 536 | |
58ff82f3 VG |
537 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
538 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
539 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 540 | cfq_hist_divisor; |
58ff82f3 VG |
541 | return cfqg->busy_queues_avg[rt]; |
542 | } | |
543 | ||
544 | static inline unsigned | |
545 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
546 | { | |
547 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
548 | ||
549 | return cfq_target_latency * cfqg->weight / st->total_weight; | |
5db5d642 CZ |
550 | } |
551 | ||
44f7c160 JA |
552 | static inline void |
553 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
554 | { | |
5db5d642 CZ |
555 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
556 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
557 | /* |
558 | * interested queues (we consider only the ones with the same | |
559 | * priority class in the cfq group) | |
560 | */ | |
561 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
562 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
563 | unsigned sync_slice = cfqd->cfq_slice[1]; |
564 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
565 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
566 | ||
567 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
568 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
569 | /* scale low_slice according to IO priority | |
570 | * and sync vs async */ | |
571 | unsigned low_slice = | |
572 | min(slice, base_low_slice * slice / sync_slice); | |
573 | /* the adapted slice value is scaled to fit all iqs | |
574 | * into the target latency */ | |
58ff82f3 | 575 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
576 | low_slice); |
577 | } | |
578 | } | |
dae739eb | 579 | cfqq->slice_start = jiffies; |
5db5d642 | 580 | cfqq->slice_end = jiffies + slice; |
7b679138 | 581 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
582 | } |
583 | ||
584 | /* | |
585 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
586 | * isn't valid until the first request from the dispatch is activated | |
587 | * and the slice time set. | |
588 | */ | |
a6151c3a | 589 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
590 | { |
591 | if (cfq_cfqq_slice_new(cfqq)) | |
592 | return 0; | |
593 | if (time_before(jiffies, cfqq->slice_end)) | |
594 | return 0; | |
595 | ||
596 | return 1; | |
597 | } | |
598 | ||
1da177e4 | 599 | /* |
5e705374 | 600 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 601 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 602 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 603 | */ |
5e705374 | 604 | static struct request * |
cf7c25cf | 605 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 606 | { |
cf7c25cf | 607 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 608 | unsigned long back_max; |
e8a99053 AM |
609 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
610 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
611 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 612 | |
5e705374 JA |
613 | if (rq1 == NULL || rq1 == rq2) |
614 | return rq2; | |
615 | if (rq2 == NULL) | |
616 | return rq1; | |
9c2c38a1 | 617 | |
5e705374 JA |
618 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
619 | return rq1; | |
620 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
621 | return rq2; | |
374f84ac JA |
622 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
623 | return rq1; | |
624 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
625 | return rq2; | |
1da177e4 | 626 | |
83096ebf TH |
627 | s1 = blk_rq_pos(rq1); |
628 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 629 | |
1da177e4 LT |
630 | /* |
631 | * by definition, 1KiB is 2 sectors | |
632 | */ | |
633 | back_max = cfqd->cfq_back_max * 2; | |
634 | ||
635 | /* | |
636 | * Strict one way elevator _except_ in the case where we allow | |
637 | * short backward seeks which are biased as twice the cost of a | |
638 | * similar forward seek. | |
639 | */ | |
640 | if (s1 >= last) | |
641 | d1 = s1 - last; | |
642 | else if (s1 + back_max >= last) | |
643 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
644 | else | |
e8a99053 | 645 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
646 | |
647 | if (s2 >= last) | |
648 | d2 = s2 - last; | |
649 | else if (s2 + back_max >= last) | |
650 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
651 | else | |
e8a99053 | 652 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
653 | |
654 | /* Found required data */ | |
e8a99053 AM |
655 | |
656 | /* | |
657 | * By doing switch() on the bit mask "wrap" we avoid having to | |
658 | * check two variables for all permutations: --> faster! | |
659 | */ | |
660 | switch (wrap) { | |
5e705374 | 661 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 662 | if (d1 < d2) |
5e705374 | 663 | return rq1; |
e8a99053 | 664 | else if (d2 < d1) |
5e705374 | 665 | return rq2; |
e8a99053 AM |
666 | else { |
667 | if (s1 >= s2) | |
5e705374 | 668 | return rq1; |
e8a99053 | 669 | else |
5e705374 | 670 | return rq2; |
e8a99053 | 671 | } |
1da177e4 | 672 | |
e8a99053 | 673 | case CFQ_RQ2_WRAP: |
5e705374 | 674 | return rq1; |
e8a99053 | 675 | case CFQ_RQ1_WRAP: |
5e705374 JA |
676 | return rq2; |
677 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
678 | default: |
679 | /* | |
680 | * Since both rqs are wrapped, | |
681 | * start with the one that's further behind head | |
682 | * (--> only *one* back seek required), | |
683 | * since back seek takes more time than forward. | |
684 | */ | |
685 | if (s1 <= s2) | |
5e705374 | 686 | return rq1; |
1da177e4 | 687 | else |
5e705374 | 688 | return rq2; |
1da177e4 LT |
689 | } |
690 | } | |
691 | ||
498d3aa2 JA |
692 | /* |
693 | * The below is leftmost cache rbtree addon | |
694 | */ | |
0871714e | 695 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 696 | { |
615f0259 VG |
697 | /* Service tree is empty */ |
698 | if (!root->count) | |
699 | return NULL; | |
700 | ||
cc09e299 JA |
701 | if (!root->left) |
702 | root->left = rb_first(&root->rb); | |
703 | ||
0871714e JA |
704 | if (root->left) |
705 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
706 | ||
707 | return NULL; | |
cc09e299 JA |
708 | } |
709 | ||
1fa8f6d6 VG |
710 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
711 | { | |
712 | if (!root->left) | |
713 | root->left = rb_first(&root->rb); | |
714 | ||
715 | if (root->left) | |
716 | return rb_entry_cfqg(root->left); | |
717 | ||
718 | return NULL; | |
719 | } | |
720 | ||
a36e71f9 JA |
721 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
722 | { | |
723 | rb_erase(n, root); | |
724 | RB_CLEAR_NODE(n); | |
725 | } | |
726 | ||
cc09e299 JA |
727 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
728 | { | |
729 | if (root->left == n) | |
730 | root->left = NULL; | |
a36e71f9 | 731 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 732 | --root->count; |
cc09e299 JA |
733 | } |
734 | ||
1da177e4 LT |
735 | /* |
736 | * would be nice to take fifo expire time into account as well | |
737 | */ | |
5e705374 JA |
738 | static struct request * |
739 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
740 | struct request *last) | |
1da177e4 | 741 | { |
21183b07 JA |
742 | struct rb_node *rbnext = rb_next(&last->rb_node); |
743 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 744 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 745 | |
21183b07 | 746 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
747 | |
748 | if (rbprev) | |
5e705374 | 749 | prev = rb_entry_rq(rbprev); |
1da177e4 | 750 | |
21183b07 | 751 | if (rbnext) |
5e705374 | 752 | next = rb_entry_rq(rbnext); |
21183b07 JA |
753 | else { |
754 | rbnext = rb_first(&cfqq->sort_list); | |
755 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 756 | next = rb_entry_rq(rbnext); |
21183b07 | 757 | } |
1da177e4 | 758 | |
cf7c25cf | 759 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
760 | } |
761 | ||
d9e7620e JA |
762 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
763 | struct cfq_queue *cfqq) | |
1da177e4 | 764 | { |
d9e7620e JA |
765 | /* |
766 | * just an approximation, should be ok. | |
767 | */ | |
cdb16e8f | 768 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 769 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
770 | } |
771 | ||
1fa8f6d6 VG |
772 | static inline s64 |
773 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
774 | { | |
775 | return cfqg->vdisktime - st->min_vdisktime; | |
776 | } | |
777 | ||
778 | static void | |
779 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
780 | { | |
781 | struct rb_node **node = &st->rb.rb_node; | |
782 | struct rb_node *parent = NULL; | |
783 | struct cfq_group *__cfqg; | |
784 | s64 key = cfqg_key(st, cfqg); | |
785 | int left = 1; | |
786 | ||
787 | while (*node != NULL) { | |
788 | parent = *node; | |
789 | __cfqg = rb_entry_cfqg(parent); | |
790 | ||
791 | if (key < cfqg_key(st, __cfqg)) | |
792 | node = &parent->rb_left; | |
793 | else { | |
794 | node = &parent->rb_right; | |
795 | left = 0; | |
796 | } | |
797 | } | |
798 | ||
799 | if (left) | |
800 | st->left = &cfqg->rb_node; | |
801 | ||
802 | rb_link_node(&cfqg->rb_node, parent, node); | |
803 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
804 | } | |
805 | ||
806 | static void | |
807 | cfq_group_service_tree_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
808 | { | |
809 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
810 | struct cfq_group *__cfqg; | |
811 | struct rb_node *n; | |
812 | ||
813 | cfqg->nr_cfqq++; | |
814 | if (cfqg->on_st) | |
815 | return; | |
816 | ||
817 | /* | |
818 | * Currently put the group at the end. Later implement something | |
819 | * so that groups get lesser vtime based on their weights, so that | |
820 | * if group does not loose all if it was not continously backlogged. | |
821 | */ | |
822 | n = rb_last(&st->rb); | |
823 | if (n) { | |
824 | __cfqg = rb_entry_cfqg(n); | |
825 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
826 | } else | |
827 | cfqg->vdisktime = st->min_vdisktime; | |
828 | ||
829 | __cfq_group_service_tree_add(st, cfqg); | |
830 | cfqg->on_st = true; | |
58ff82f3 VG |
831 | cfqd->nr_groups++; |
832 | st->total_weight += cfqg->weight; | |
1fa8f6d6 VG |
833 | } |
834 | ||
835 | static void | |
836 | cfq_group_service_tree_del(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
837 | { | |
838 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
839 | ||
25bc6b07 VG |
840 | if (st->active == &cfqg->rb_node) |
841 | st->active = NULL; | |
842 | ||
1fa8f6d6 VG |
843 | BUG_ON(cfqg->nr_cfqq < 1); |
844 | cfqg->nr_cfqq--; | |
25bc6b07 | 845 | |
1fa8f6d6 VG |
846 | /* If there are other cfq queues under this group, don't delete it */ |
847 | if (cfqg->nr_cfqq) | |
848 | return; | |
849 | ||
2868ef7b | 850 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
1fa8f6d6 | 851 | cfqg->on_st = false; |
58ff82f3 VG |
852 | cfqd->nr_groups--; |
853 | st->total_weight -= cfqg->weight; | |
1fa8f6d6 VG |
854 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
855 | cfq_rb_erase(&cfqg->rb_node, st); | |
dae739eb | 856 | cfqg->saved_workload_slice = 0; |
22084190 | 857 | blkiocg_update_blkio_group_dequeue_stats(&cfqg->blkg, 1); |
dae739eb VG |
858 | } |
859 | ||
860 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq) | |
861 | { | |
862 | unsigned int slice_used, allocated_slice; | |
863 | ||
864 | /* | |
865 | * Queue got expired before even a single request completed or | |
866 | * got expired immediately after first request completion. | |
867 | */ | |
868 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
869 | /* | |
870 | * Also charge the seek time incurred to the group, otherwise | |
871 | * if there are mutiple queues in the group, each can dispatch | |
872 | * a single request on seeky media and cause lots of seek time | |
873 | * and group will never know it. | |
874 | */ | |
875 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
876 | 1); | |
877 | } else { | |
878 | slice_used = jiffies - cfqq->slice_start; | |
879 | allocated_slice = cfqq->slice_end - cfqq->slice_start; | |
880 | if (slice_used > allocated_slice) | |
881 | slice_used = allocated_slice; | |
882 | } | |
883 | ||
22084190 VG |
884 | cfq_log_cfqq(cfqq->cfqd, cfqq, "sl_used=%u sect=%lu", slice_used, |
885 | cfqq->nr_sectors); | |
dae739eb VG |
886 | return slice_used; |
887 | } | |
888 | ||
889 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
890 | struct cfq_queue *cfqq) | |
891 | { | |
892 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
893 | unsigned int used_sl; | |
894 | ||
895 | used_sl = cfq_cfqq_slice_usage(cfqq); | |
896 | ||
897 | /* Can't update vdisktime while group is on service tree */ | |
898 | cfq_rb_erase(&cfqg->rb_node, st); | |
899 | cfqg->vdisktime += cfq_scale_slice(used_sl, cfqg); | |
900 | __cfq_group_service_tree_add(st, cfqg); | |
901 | ||
902 | /* This group is being expired. Save the context */ | |
903 | if (time_after(cfqd->workload_expires, jiffies)) { | |
904 | cfqg->saved_workload_slice = cfqd->workload_expires | |
905 | - jiffies; | |
906 | cfqg->saved_workload = cfqd->serving_type; | |
907 | cfqg->saved_serving_prio = cfqd->serving_prio; | |
908 | } else | |
909 | cfqg->saved_workload_slice = 0; | |
2868ef7b VG |
910 | |
911 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
912 | st->min_vdisktime); | |
22084190 VG |
913 | blkiocg_update_blkio_group_stats(&cfqg->blkg, used_sl, |
914 | cfqq->nr_sectors); | |
1fa8f6d6 VG |
915 | } |
916 | ||
25fb5169 VG |
917 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
918 | static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg) | |
919 | { | |
920 | if (blkg) | |
921 | return container_of(blkg, struct cfq_group, blkg); | |
922 | return NULL; | |
923 | } | |
924 | ||
925 | static struct cfq_group * | |
926 | cfq_find_alloc_cfqg(struct cfq_data *cfqd, struct cgroup *cgroup, int create) | |
927 | { | |
928 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
929 | struct cfq_group *cfqg = NULL; | |
930 | void *key = cfqd; | |
931 | int i, j; | |
932 | struct cfq_rb_root *st; | |
22084190 VG |
933 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; |
934 | unsigned int major, minor; | |
25fb5169 VG |
935 | |
936 | /* Do we need to take this reference */ | |
937 | if (!css_tryget(&blkcg->css)) | |
938 | return NULL;; | |
939 | ||
940 | cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key)); | |
941 | if (cfqg || !create) | |
942 | goto done; | |
943 | ||
944 | cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node); | |
945 | if (!cfqg) | |
946 | goto done; | |
947 | ||
948 | cfqg->weight = blkcg->weight; | |
949 | for_each_cfqg_st(cfqg, i, j, st) | |
950 | *st = CFQ_RB_ROOT; | |
951 | RB_CLEAR_NODE(&cfqg->rb_node); | |
952 | ||
b1c35769 VG |
953 | /* |
954 | * Take the initial reference that will be released on destroy | |
955 | * This can be thought of a joint reference by cgroup and | |
956 | * elevator which will be dropped by either elevator exit | |
957 | * or cgroup deletion path depending on who is exiting first. | |
958 | */ | |
959 | atomic_set(&cfqg->ref, 1); | |
960 | ||
25fb5169 | 961 | /* Add group onto cgroup list */ |
22084190 VG |
962 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); |
963 | blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd, | |
964 | MKDEV(major, minor)); | |
25fb5169 VG |
965 | |
966 | /* Add group on cfqd list */ | |
967 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); | |
968 | ||
969 | done: | |
970 | css_put(&blkcg->css); | |
971 | return cfqg; | |
972 | } | |
973 | ||
974 | /* | |
975 | * Search for the cfq group current task belongs to. If create = 1, then also | |
976 | * create the cfq group if it does not exist. request_queue lock must be held. | |
977 | */ | |
978 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create) | |
979 | { | |
980 | struct cgroup *cgroup; | |
981 | struct cfq_group *cfqg = NULL; | |
982 | ||
983 | rcu_read_lock(); | |
984 | cgroup = task_cgroup(current, blkio_subsys_id); | |
985 | cfqg = cfq_find_alloc_cfqg(cfqd, cgroup, create); | |
986 | if (!cfqg && create) | |
987 | cfqg = &cfqd->root_group; | |
988 | rcu_read_unlock(); | |
989 | return cfqg; | |
990 | } | |
991 | ||
992 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
993 | { | |
994 | /* Currently, all async queues are mapped to root group */ | |
995 | if (!cfq_cfqq_sync(cfqq)) | |
996 | cfqg = &cfqq->cfqd->root_group; | |
997 | ||
998 | cfqq->cfqg = cfqg; | |
b1c35769 VG |
999 | /* cfqq reference on cfqg */ |
1000 | atomic_inc(&cfqq->cfqg->ref); | |
1001 | } | |
1002 | ||
1003 | static void cfq_put_cfqg(struct cfq_group *cfqg) | |
1004 | { | |
1005 | struct cfq_rb_root *st; | |
1006 | int i, j; | |
1007 | ||
1008 | BUG_ON(atomic_read(&cfqg->ref) <= 0); | |
1009 | if (!atomic_dec_and_test(&cfqg->ref)) | |
1010 | return; | |
1011 | for_each_cfqg_st(cfqg, i, j, st) | |
1012 | BUG_ON(!RB_EMPTY_ROOT(&st->rb) || st->active != NULL); | |
1013 | kfree(cfqg); | |
1014 | } | |
1015 | ||
1016 | static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1017 | { | |
1018 | /* Something wrong if we are trying to remove same group twice */ | |
1019 | BUG_ON(hlist_unhashed(&cfqg->cfqd_node)); | |
1020 | ||
1021 | hlist_del_init(&cfqg->cfqd_node); | |
1022 | ||
1023 | /* | |
1024 | * Put the reference taken at the time of creation so that when all | |
1025 | * queues are gone, group can be destroyed. | |
1026 | */ | |
1027 | cfq_put_cfqg(cfqg); | |
1028 | } | |
1029 | ||
1030 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) | |
1031 | { | |
1032 | struct hlist_node *pos, *n; | |
1033 | struct cfq_group *cfqg; | |
1034 | ||
1035 | hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) { | |
1036 | /* | |
1037 | * If cgroup removal path got to blk_group first and removed | |
1038 | * it from cgroup list, then it will take care of destroying | |
1039 | * cfqg also. | |
1040 | */ | |
1041 | if (!blkiocg_del_blkio_group(&cfqg->blkg)) | |
1042 | cfq_destroy_cfqg(cfqd, cfqg); | |
1043 | } | |
25fb5169 | 1044 | } |
b1c35769 VG |
1045 | |
1046 | /* | |
1047 | * Blk cgroup controller notification saying that blkio_group object is being | |
1048 | * delinked as associated cgroup object is going away. That also means that | |
1049 | * no new IO will come in this group. So get rid of this group as soon as | |
1050 | * any pending IO in the group is finished. | |
1051 | * | |
1052 | * This function is called under rcu_read_lock(). key is the rcu protected | |
1053 | * pointer. That means "key" is a valid cfq_data pointer as long as we are rcu | |
1054 | * read lock. | |
1055 | * | |
1056 | * "key" was fetched from blkio_group under blkio_cgroup->lock. That means | |
1057 | * it should not be NULL as even if elevator was exiting, cgroup deltion | |
1058 | * path got to it first. | |
1059 | */ | |
1060 | void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg) | |
1061 | { | |
1062 | unsigned long flags; | |
1063 | struct cfq_data *cfqd = key; | |
1064 | ||
1065 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1066 | cfq_destroy_cfqg(cfqd, cfqg_of_blkg(blkg)); | |
1067 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1068 | } | |
1069 | ||
25fb5169 VG |
1070 | #else /* GROUP_IOSCHED */ |
1071 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create) | |
1072 | { | |
1073 | return &cfqd->root_group; | |
1074 | } | |
1075 | static inline void | |
1076 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1077 | cfqq->cfqg = cfqg; | |
1078 | } | |
1079 | ||
b1c35769 VG |
1080 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) {} |
1081 | static inline void cfq_put_cfqg(struct cfq_group *cfqg) {} | |
1082 | ||
25fb5169 VG |
1083 | #endif /* GROUP_IOSCHED */ |
1084 | ||
498d3aa2 | 1085 | /* |
c0324a02 | 1086 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1087 | * requests waiting to be processed. It is sorted in the order that |
1088 | * we will service the queues. | |
1089 | */ | |
a36e71f9 | 1090 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1091 | bool add_front) |
d9e7620e | 1092 | { |
0871714e JA |
1093 | struct rb_node **p, *parent; |
1094 | struct cfq_queue *__cfqq; | |
d9e7620e | 1095 | unsigned long rb_key; |
c0324a02 | 1096 | struct cfq_rb_root *service_tree; |
498d3aa2 | 1097 | int left; |
dae739eb | 1098 | int new_cfqq = 1; |
d9e7620e | 1099 | |
cdb16e8f VG |
1100 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
1101 | cfqq_type(cfqq), cfqd); | |
0871714e JA |
1102 | if (cfq_class_idle(cfqq)) { |
1103 | rb_key = CFQ_IDLE_DELAY; | |
aa6f6a3d | 1104 | parent = rb_last(&service_tree->rb); |
0871714e JA |
1105 | if (parent && parent != &cfqq->rb_node) { |
1106 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1107 | rb_key += __cfqq->rb_key; | |
1108 | } else | |
1109 | rb_key += jiffies; | |
1110 | } else if (!add_front) { | |
b9c8946b JA |
1111 | /* |
1112 | * Get our rb key offset. Subtract any residual slice | |
1113 | * value carried from last service. A negative resid | |
1114 | * count indicates slice overrun, and this should position | |
1115 | * the next service time further away in the tree. | |
1116 | */ | |
edd75ffd | 1117 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1118 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1119 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1120 | } else { |
1121 | rb_key = -HZ; | |
aa6f6a3d | 1122 | __cfqq = cfq_rb_first(service_tree); |
48e025e6 CZ |
1123 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1124 | } | |
1da177e4 | 1125 | |
d9e7620e | 1126 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1127 | new_cfqq = 0; |
99f9628a | 1128 | /* |
d9e7620e | 1129 | * same position, nothing more to do |
99f9628a | 1130 | */ |
c0324a02 CZ |
1131 | if (rb_key == cfqq->rb_key && |
1132 | cfqq->service_tree == service_tree) | |
d9e7620e | 1133 | return; |
1da177e4 | 1134 | |
aa6f6a3d CZ |
1135 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1136 | cfqq->service_tree = NULL; | |
1da177e4 | 1137 | } |
d9e7620e | 1138 | |
498d3aa2 | 1139 | left = 1; |
0871714e | 1140 | parent = NULL; |
aa6f6a3d CZ |
1141 | cfqq->service_tree = service_tree; |
1142 | p = &service_tree->rb.rb_node; | |
d9e7620e | 1143 | while (*p) { |
67060e37 | 1144 | struct rb_node **n; |
cc09e299 | 1145 | |
d9e7620e JA |
1146 | parent = *p; |
1147 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1148 | ||
0c534e0a | 1149 | /* |
c0324a02 | 1150 | * sort by key, that represents service time. |
0c534e0a | 1151 | */ |
c0324a02 | 1152 | if (time_before(rb_key, __cfqq->rb_key)) |
67060e37 | 1153 | n = &(*p)->rb_left; |
c0324a02 | 1154 | else { |
67060e37 | 1155 | n = &(*p)->rb_right; |
cc09e299 | 1156 | left = 0; |
c0324a02 | 1157 | } |
67060e37 JA |
1158 | |
1159 | p = n; | |
d9e7620e JA |
1160 | } |
1161 | ||
cc09e299 | 1162 | if (left) |
aa6f6a3d | 1163 | service_tree->left = &cfqq->rb_node; |
cc09e299 | 1164 | |
d9e7620e JA |
1165 | cfqq->rb_key = rb_key; |
1166 | rb_link_node(&cfqq->rb_node, parent, p); | |
aa6f6a3d CZ |
1167 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
1168 | service_tree->count++; | |
dae739eb VG |
1169 | if (add_front || !new_cfqq) |
1170 | return; | |
1fa8f6d6 | 1171 | cfq_group_service_tree_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1172 | } |
1173 | ||
a36e71f9 | 1174 | static struct cfq_queue * |
f2d1f0ae JA |
1175 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1176 | sector_t sector, struct rb_node **ret_parent, | |
1177 | struct rb_node ***rb_link) | |
a36e71f9 | 1178 | { |
a36e71f9 JA |
1179 | struct rb_node **p, *parent; |
1180 | struct cfq_queue *cfqq = NULL; | |
1181 | ||
1182 | parent = NULL; | |
1183 | p = &root->rb_node; | |
1184 | while (*p) { | |
1185 | struct rb_node **n; | |
1186 | ||
1187 | parent = *p; | |
1188 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1189 | ||
1190 | /* | |
1191 | * Sort strictly based on sector. Smallest to the left, | |
1192 | * largest to the right. | |
1193 | */ | |
2e46e8b2 | 1194 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1195 | n = &(*p)->rb_right; |
2e46e8b2 | 1196 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1197 | n = &(*p)->rb_left; |
1198 | else | |
1199 | break; | |
1200 | p = n; | |
3ac6c9f8 | 1201 | cfqq = NULL; |
a36e71f9 JA |
1202 | } |
1203 | ||
1204 | *ret_parent = parent; | |
1205 | if (rb_link) | |
1206 | *rb_link = p; | |
3ac6c9f8 | 1207 | return cfqq; |
a36e71f9 JA |
1208 | } |
1209 | ||
1210 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1211 | { | |
a36e71f9 JA |
1212 | struct rb_node **p, *parent; |
1213 | struct cfq_queue *__cfqq; | |
1214 | ||
f2d1f0ae JA |
1215 | if (cfqq->p_root) { |
1216 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1217 | cfqq->p_root = NULL; | |
1218 | } | |
a36e71f9 JA |
1219 | |
1220 | if (cfq_class_idle(cfqq)) | |
1221 | return; | |
1222 | if (!cfqq->next_rq) | |
1223 | return; | |
1224 | ||
f2d1f0ae | 1225 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1226 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1227 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1228 | if (!__cfqq) { |
1229 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1230 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1231 | } else | |
1232 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1233 | } |
1234 | ||
498d3aa2 JA |
1235 | /* |
1236 | * Update cfqq's position in the service tree. | |
1237 | */ | |
edd75ffd | 1238 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1239 | { |
6d048f53 JA |
1240 | /* |
1241 | * Resorting requires the cfqq to be on the RR list already. | |
1242 | */ | |
a36e71f9 | 1243 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1244 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1245 | cfq_prio_tree_add(cfqd, cfqq); |
1246 | } | |
6d048f53 JA |
1247 | } |
1248 | ||
1da177e4 LT |
1249 | /* |
1250 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1251 | * the pending list according to last request service |
1da177e4 | 1252 | */ |
febffd61 | 1253 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1254 | { |
7b679138 | 1255 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1256 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1257 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
1258 | cfqd->busy_queues++; |
1259 | ||
edd75ffd | 1260 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1261 | } |
1262 | ||
498d3aa2 JA |
1263 | /* |
1264 | * Called when the cfqq no longer has requests pending, remove it from | |
1265 | * the service tree. | |
1266 | */ | |
febffd61 | 1267 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1268 | { |
7b679138 | 1269 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1270 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1271 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1272 | |
aa6f6a3d CZ |
1273 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1274 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1275 | cfqq->service_tree = NULL; | |
1276 | } | |
f2d1f0ae JA |
1277 | if (cfqq->p_root) { |
1278 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1279 | cfqq->p_root = NULL; | |
1280 | } | |
d9e7620e | 1281 | |
1fa8f6d6 | 1282 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1283 | BUG_ON(!cfqd->busy_queues); |
1284 | cfqd->busy_queues--; | |
1285 | } | |
1286 | ||
1287 | /* | |
1288 | * rb tree support functions | |
1289 | */ | |
febffd61 | 1290 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1291 | { |
5e705374 | 1292 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1293 | const int sync = rq_is_sync(rq); |
1da177e4 | 1294 | |
b4878f24 JA |
1295 | BUG_ON(!cfqq->queued[sync]); |
1296 | cfqq->queued[sync]--; | |
1da177e4 | 1297 | |
5e705374 | 1298 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1299 | |
f04a6424 VG |
1300 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1301 | /* | |
1302 | * Queue will be deleted from service tree when we actually | |
1303 | * expire it later. Right now just remove it from prio tree | |
1304 | * as it is empty. | |
1305 | */ | |
1306 | if (cfqq->p_root) { | |
1307 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1308 | cfqq->p_root = NULL; | |
1309 | } | |
1310 | } | |
1da177e4 LT |
1311 | } |
1312 | ||
5e705374 | 1313 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1314 | { |
5e705374 | 1315 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1316 | struct cfq_data *cfqd = cfqq->cfqd; |
a36e71f9 | 1317 | struct request *__alias, *prev; |
1da177e4 | 1318 | |
5380a101 | 1319 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
1320 | |
1321 | /* | |
1322 | * looks a little odd, but the first insert might return an alias. | |
1323 | * if that happens, put the alias on the dispatch list | |
1324 | */ | |
21183b07 | 1325 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 1326 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
1327 | |
1328 | if (!cfq_cfqq_on_rr(cfqq)) | |
1329 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1330 | |
1331 | /* | |
1332 | * check if this request is a better next-serve candidate | |
1333 | */ | |
a36e71f9 | 1334 | prev = cfqq->next_rq; |
cf7c25cf | 1335 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1336 | |
1337 | /* | |
1338 | * adjust priority tree position, if ->next_rq changes | |
1339 | */ | |
1340 | if (prev != cfqq->next_rq) | |
1341 | cfq_prio_tree_add(cfqd, cfqq); | |
1342 | ||
5044eed4 | 1343 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1344 | } |
1345 | ||
febffd61 | 1346 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1347 | { |
5380a101 JA |
1348 | elv_rb_del(&cfqq->sort_list, rq); |
1349 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 1350 | cfq_add_rq_rb(rq); |
1da177e4 LT |
1351 | } |
1352 | ||
206dc69b JA |
1353 | static struct request * |
1354 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1355 | { |
206dc69b | 1356 | struct task_struct *tsk = current; |
91fac317 | 1357 | struct cfq_io_context *cic; |
206dc69b | 1358 | struct cfq_queue *cfqq; |
1da177e4 | 1359 | |
4ac845a2 | 1360 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1361 | if (!cic) |
1362 | return NULL; | |
1363 | ||
1364 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
1365 | if (cfqq) { |
1366 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
1367 | ||
21183b07 | 1368 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 1369 | } |
1da177e4 | 1370 | |
1da177e4 LT |
1371 | return NULL; |
1372 | } | |
1373 | ||
165125e1 | 1374 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1375 | { |
22e2c507 | 1376 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 1377 | |
5ad531db | 1378 | cfqd->rq_in_driver[rq_is_sync(rq)]++; |
7b679138 | 1379 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
5ad531db | 1380 | rq_in_driver(cfqd)); |
25776e35 | 1381 | |
5b93629b | 1382 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
1383 | } |
1384 | ||
165125e1 | 1385 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1386 | { |
b4878f24 | 1387 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5ad531db | 1388 | const int sync = rq_is_sync(rq); |
b4878f24 | 1389 | |
5ad531db JA |
1390 | WARN_ON(!cfqd->rq_in_driver[sync]); |
1391 | cfqd->rq_in_driver[sync]--; | |
7b679138 | 1392 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
5ad531db | 1393 | rq_in_driver(cfqd)); |
1da177e4 LT |
1394 | } |
1395 | ||
b4878f24 | 1396 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 1397 | { |
5e705374 | 1398 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 1399 | |
5e705374 JA |
1400 | if (cfqq->next_rq == rq) |
1401 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 1402 | |
b4878f24 | 1403 | list_del_init(&rq->queuelist); |
5e705374 | 1404 | cfq_del_rq_rb(rq); |
374f84ac | 1405 | |
45333d5a | 1406 | cfqq->cfqd->rq_queued--; |
374f84ac JA |
1407 | if (rq_is_meta(rq)) { |
1408 | WARN_ON(!cfqq->meta_pending); | |
1409 | cfqq->meta_pending--; | |
1410 | } | |
1da177e4 LT |
1411 | } |
1412 | ||
165125e1 JA |
1413 | static int cfq_merge(struct request_queue *q, struct request **req, |
1414 | struct bio *bio) | |
1da177e4 LT |
1415 | { |
1416 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1417 | struct request *__rq; | |
1da177e4 | 1418 | |
206dc69b | 1419 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 1420 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
1421 | *req = __rq; |
1422 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
1423 | } |
1424 | ||
1425 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
1426 | } |
1427 | ||
165125e1 | 1428 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 1429 | int type) |
1da177e4 | 1430 | { |
21183b07 | 1431 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 1432 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 1433 | |
5e705374 | 1434 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 1435 | } |
1da177e4 LT |
1436 | } |
1437 | ||
1438 | static void | |
165125e1 | 1439 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
1440 | struct request *next) |
1441 | { | |
cf7c25cf | 1442 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1443 | /* |
1444 | * reposition in fifo if next is older than rq | |
1445 | */ | |
1446 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
30996f40 | 1447 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
22e2c507 | 1448 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
1449 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1450 | } | |
22e2c507 | 1451 | |
cf7c25cf CZ |
1452 | if (cfqq->next_rq == next) |
1453 | cfqq->next_rq = rq; | |
b4878f24 | 1454 | cfq_remove_request(next); |
22e2c507 JA |
1455 | } |
1456 | ||
165125e1 | 1457 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
1458 | struct bio *bio) |
1459 | { | |
1460 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
91fac317 | 1461 | struct cfq_io_context *cic; |
da775265 | 1462 | struct cfq_queue *cfqq; |
da775265 | 1463 | |
8682e1f1 VG |
1464 | /* Deny merge if bio and rq don't belong to same cfq group */ |
1465 | if ((RQ_CFQQ(rq))->cfqg != cfq_get_cfqg(cfqd, 0)) | |
1466 | return false; | |
da775265 | 1467 | /* |
ec8acb69 | 1468 | * Disallow merge of a sync bio into an async request. |
da775265 | 1469 | */ |
91fac317 | 1470 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 1471 | return false; |
da775265 JA |
1472 | |
1473 | /* | |
719d3402 JA |
1474 | * Lookup the cfqq that this bio will be queued with. Allow |
1475 | * merge only if rq is queued there. | |
da775265 | 1476 | */ |
4ac845a2 | 1477 | cic = cfq_cic_lookup(cfqd, current->io_context); |
91fac317 | 1478 | if (!cic) |
a6151c3a | 1479 | return false; |
719d3402 | 1480 | |
91fac317 | 1481 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 1482 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
1483 | } |
1484 | ||
febffd61 JA |
1485 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
1486 | struct cfq_queue *cfqq) | |
22e2c507 JA |
1487 | { |
1488 | if (cfqq) { | |
7b679138 | 1489 | cfq_log_cfqq(cfqd, cfqq, "set_active"); |
dae739eb VG |
1490 | cfqq->slice_start = 0; |
1491 | cfqq->dispatch_start = jiffies; | |
22e2c507 | 1492 | cfqq->slice_end = 0; |
2f5cb738 | 1493 | cfqq->slice_dispatch = 0; |
22084190 | 1494 | cfqq->nr_sectors = 0; |
2f5cb738 | 1495 | |
2f5cb738 | 1496 | cfq_clear_cfqq_wait_request(cfqq); |
b029195d | 1497 | cfq_clear_cfqq_must_dispatch(cfqq); |
3b18152c JA |
1498 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
1499 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 1500 | cfq_mark_cfqq_slice_new(cfqq); |
2f5cb738 JA |
1501 | |
1502 | del_timer(&cfqd->idle_slice_timer); | |
22e2c507 JA |
1503 | } |
1504 | ||
1505 | cfqd->active_queue = cfqq; | |
1506 | } | |
1507 | ||
7b14e3b5 JA |
1508 | /* |
1509 | * current cfqq expired its slice (or was too idle), select new one | |
1510 | */ | |
1511 | static void | |
1512 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
a6151c3a | 1513 | bool timed_out) |
7b14e3b5 | 1514 | { |
7b679138 JA |
1515 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
1516 | ||
7b14e3b5 JA |
1517 | if (cfq_cfqq_wait_request(cfqq)) |
1518 | del_timer(&cfqd->idle_slice_timer); | |
1519 | ||
7b14e3b5 JA |
1520 | cfq_clear_cfqq_wait_request(cfqq); |
1521 | ||
1522 | /* | |
6084cdda | 1523 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 1524 | */ |
7b679138 | 1525 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) { |
c5b680f3 | 1526 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b679138 JA |
1527 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
1528 | } | |
7b14e3b5 | 1529 | |
dae739eb VG |
1530 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
1531 | ||
f04a6424 VG |
1532 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1533 | cfq_del_cfqq_rr(cfqd, cfqq); | |
1534 | ||
edd75ffd | 1535 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
1536 | |
1537 | if (cfqq == cfqd->active_queue) | |
1538 | cfqd->active_queue = NULL; | |
1539 | ||
dae739eb VG |
1540 | if (&cfqq->cfqg->rb_node == cfqd->grp_service_tree.active) |
1541 | cfqd->grp_service_tree.active = NULL; | |
1542 | ||
7b14e3b5 JA |
1543 | if (cfqd->active_cic) { |
1544 | put_io_context(cfqd->active_cic->ioc); | |
1545 | cfqd->active_cic = NULL; | |
1546 | } | |
7b14e3b5 JA |
1547 | } |
1548 | ||
a6151c3a | 1549 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
1550 | { |
1551 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1552 | ||
1553 | if (cfqq) | |
6084cdda | 1554 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
1555 | } |
1556 | ||
498d3aa2 JA |
1557 | /* |
1558 | * Get next queue for service. Unless we have a queue preemption, | |
1559 | * we'll simply select the first cfqq in the service tree. | |
1560 | */ | |
6d048f53 | 1561 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 1562 | { |
c0324a02 | 1563 | struct cfq_rb_root *service_tree = |
cdb16e8f VG |
1564 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
1565 | cfqd->serving_type, cfqd); | |
d9e7620e | 1566 | |
f04a6424 VG |
1567 | if (!cfqd->rq_queued) |
1568 | return NULL; | |
1569 | ||
1fa8f6d6 VG |
1570 | /* There is nothing to dispatch */ |
1571 | if (!service_tree) | |
1572 | return NULL; | |
c0324a02 CZ |
1573 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
1574 | return NULL; | |
1575 | return cfq_rb_first(service_tree); | |
6d048f53 JA |
1576 | } |
1577 | ||
f04a6424 VG |
1578 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
1579 | { | |
25fb5169 | 1580 | struct cfq_group *cfqg; |
f04a6424 VG |
1581 | struct cfq_queue *cfqq; |
1582 | int i, j; | |
1583 | struct cfq_rb_root *st; | |
1584 | ||
1585 | if (!cfqd->rq_queued) | |
1586 | return NULL; | |
1587 | ||
25fb5169 VG |
1588 | cfqg = cfq_get_next_cfqg(cfqd); |
1589 | if (!cfqg) | |
1590 | return NULL; | |
1591 | ||
f04a6424 VG |
1592 | for_each_cfqg_st(cfqg, i, j, st) |
1593 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
1594 | return cfqq; | |
1595 | return NULL; | |
1596 | } | |
1597 | ||
498d3aa2 JA |
1598 | /* |
1599 | * Get and set a new active queue for service. | |
1600 | */ | |
a36e71f9 JA |
1601 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
1602 | struct cfq_queue *cfqq) | |
6d048f53 | 1603 | { |
e00ef799 | 1604 | if (!cfqq) |
a36e71f9 | 1605 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 1606 | |
22e2c507 | 1607 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 1608 | return cfqq; |
22e2c507 JA |
1609 | } |
1610 | ||
d9e7620e JA |
1611 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
1612 | struct request *rq) | |
1613 | { | |
83096ebf TH |
1614 | if (blk_rq_pos(rq) >= cfqd->last_position) |
1615 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 1616 | else |
83096ebf | 1617 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
1618 | } |
1619 | ||
b2c18e1e JM |
1620 | #define CFQQ_SEEK_THR 8 * 1024 |
1621 | #define CFQQ_SEEKY(cfqq) ((cfqq)->seek_mean > CFQQ_SEEK_THR) | |
04dc6e71 | 1622 | |
b2c18e1e JM |
1623 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1624 | struct request *rq) | |
6d048f53 | 1625 | { |
b2c18e1e | 1626 | sector_t sdist = cfqq->seek_mean; |
6d048f53 | 1627 | |
b2c18e1e JM |
1628 | if (!sample_valid(cfqq->seek_samples)) |
1629 | sdist = CFQQ_SEEK_THR; | |
6d048f53 | 1630 | |
04dc6e71 | 1631 | return cfq_dist_from_last(cfqd, rq) <= sdist; |
6d048f53 JA |
1632 | } |
1633 | ||
a36e71f9 JA |
1634 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
1635 | struct cfq_queue *cur_cfqq) | |
1636 | { | |
f2d1f0ae | 1637 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
1638 | struct rb_node *parent, *node; |
1639 | struct cfq_queue *__cfqq; | |
1640 | sector_t sector = cfqd->last_position; | |
1641 | ||
1642 | if (RB_EMPTY_ROOT(root)) | |
1643 | return NULL; | |
1644 | ||
1645 | /* | |
1646 | * First, if we find a request starting at the end of the last | |
1647 | * request, choose it. | |
1648 | */ | |
f2d1f0ae | 1649 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
1650 | if (__cfqq) |
1651 | return __cfqq; | |
1652 | ||
1653 | /* | |
1654 | * If the exact sector wasn't found, the parent of the NULL leaf | |
1655 | * will contain the closest sector. | |
1656 | */ | |
1657 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
b2c18e1e | 1658 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1659 | return __cfqq; |
1660 | ||
2e46e8b2 | 1661 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
1662 | node = rb_next(&__cfqq->p_node); |
1663 | else | |
1664 | node = rb_prev(&__cfqq->p_node); | |
1665 | if (!node) | |
1666 | return NULL; | |
1667 | ||
1668 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
b2c18e1e | 1669 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1670 | return __cfqq; |
1671 | ||
1672 | return NULL; | |
1673 | } | |
1674 | ||
1675 | /* | |
1676 | * cfqd - obvious | |
1677 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
1678 | * closely cooperating with itself. | |
1679 | * | |
1680 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
1681 | * one request, and that cfqd->last_position reflects a position on the disk | |
1682 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
1683 | * assumption. | |
1684 | */ | |
1685 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 1686 | struct cfq_queue *cur_cfqq) |
6d048f53 | 1687 | { |
a36e71f9 JA |
1688 | struct cfq_queue *cfqq; |
1689 | ||
e6c5bc73 JM |
1690 | if (!cfq_cfqq_sync(cur_cfqq)) |
1691 | return NULL; | |
1692 | if (CFQQ_SEEKY(cur_cfqq)) | |
1693 | return NULL; | |
1694 | ||
6d048f53 | 1695 | /* |
d9e7620e JA |
1696 | * We should notice if some of the queues are cooperating, eg |
1697 | * working closely on the same area of the disk. In that case, | |
1698 | * we can group them together and don't waste time idling. | |
6d048f53 | 1699 | */ |
a36e71f9 JA |
1700 | cfqq = cfqq_close(cfqd, cur_cfqq); |
1701 | if (!cfqq) | |
1702 | return NULL; | |
1703 | ||
8682e1f1 VG |
1704 | /* If new queue belongs to different cfq_group, don't choose it */ |
1705 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
1706 | return NULL; | |
1707 | ||
df5fe3e8 JM |
1708 | /* |
1709 | * It only makes sense to merge sync queues. | |
1710 | */ | |
1711 | if (!cfq_cfqq_sync(cfqq)) | |
1712 | return NULL; | |
e6c5bc73 JM |
1713 | if (CFQQ_SEEKY(cfqq)) |
1714 | return NULL; | |
df5fe3e8 | 1715 | |
c0324a02 CZ |
1716 | /* |
1717 | * Do not merge queues of different priority classes | |
1718 | */ | |
1719 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
1720 | return NULL; | |
1721 | ||
a36e71f9 | 1722 | return cfqq; |
6d048f53 JA |
1723 | } |
1724 | ||
a6d44e98 CZ |
1725 | /* |
1726 | * Determine whether we should enforce idle window for this queue. | |
1727 | */ | |
1728 | ||
1729 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1730 | { | |
1731 | enum wl_prio_t prio = cfqq_prio(cfqq); | |
718eee05 | 1732 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
a6d44e98 | 1733 | |
f04a6424 VG |
1734 | BUG_ON(!service_tree); |
1735 | BUG_ON(!service_tree->count); | |
1736 | ||
a6d44e98 CZ |
1737 | /* We never do for idle class queues. */ |
1738 | if (prio == IDLE_WORKLOAD) | |
1739 | return false; | |
1740 | ||
1741 | /* We do for queues that were marked with idle window flag. */ | |
1742 | if (cfq_cfqq_idle_window(cfqq)) | |
1743 | return true; | |
1744 | ||
1745 | /* | |
1746 | * Otherwise, we do only if they are the last ones | |
1747 | * in their service tree. | |
1748 | */ | |
f04a6424 | 1749 | return service_tree->count == 1; |
a6d44e98 CZ |
1750 | } |
1751 | ||
6d048f53 | 1752 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 1753 | { |
1792669c | 1754 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 1755 | struct cfq_io_context *cic; |
7b14e3b5 JA |
1756 | unsigned long sl; |
1757 | ||
a68bbddb | 1758 | /* |
f7d7b7a7 JA |
1759 | * SSD device without seek penalty, disable idling. But only do so |
1760 | * for devices that support queuing, otherwise we still have a problem | |
1761 | * with sync vs async workloads. | |
a68bbddb | 1762 | */ |
f7d7b7a7 | 1763 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
1764 | return; |
1765 | ||
dd67d051 | 1766 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 1767 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
1768 | |
1769 | /* | |
1770 | * idle is disabled, either manually or by past process history | |
1771 | */ | |
a6d44e98 | 1772 | if (!cfqd->cfq_slice_idle || !cfq_should_idle(cfqd, cfqq)) |
6d048f53 JA |
1773 | return; |
1774 | ||
7b679138 | 1775 | /* |
8e550632 | 1776 | * still active requests from this queue, don't idle |
7b679138 | 1777 | */ |
8e550632 | 1778 | if (cfqq->dispatched) |
7b679138 JA |
1779 | return; |
1780 | ||
22e2c507 JA |
1781 | /* |
1782 | * task has exited, don't wait | |
1783 | */ | |
206dc69b | 1784 | cic = cfqd->active_cic; |
66dac98e | 1785 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
6d048f53 JA |
1786 | return; |
1787 | ||
355b659c CZ |
1788 | /* |
1789 | * If our average think time is larger than the remaining time | |
1790 | * slice, then don't idle. This avoids overrunning the allotted | |
1791 | * time slice. | |
1792 | */ | |
1793 | if (sample_valid(cic->ttime_samples) && | |
1794 | (cfqq->slice_end - jiffies < cic->ttime_mean)) | |
1795 | return; | |
1796 | ||
3b18152c | 1797 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 1798 | |
6d048f53 | 1799 | sl = cfqd->cfq_slice_idle; |
206dc69b | 1800 | |
7b14e3b5 | 1801 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
9481ffdc | 1802 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu", sl); |
1da177e4 LT |
1803 | } |
1804 | ||
498d3aa2 JA |
1805 | /* |
1806 | * Move request from internal lists to the request queue dispatch list. | |
1807 | */ | |
165125e1 | 1808 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 1809 | { |
3ed9a296 | 1810 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1811 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1812 | |
7b679138 JA |
1813 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
1814 | ||
06d21886 | 1815 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 1816 | cfq_remove_request(rq); |
6d048f53 | 1817 | cfqq->dispatched++; |
5380a101 | 1818 | elv_dispatch_sort(q, rq); |
3ed9a296 JA |
1819 | |
1820 | if (cfq_cfqq_sync(cfqq)) | |
1821 | cfqd->sync_flight++; | |
22084190 | 1822 | cfqq->nr_sectors += blk_rq_sectors(rq); |
1da177e4 LT |
1823 | } |
1824 | ||
1825 | /* | |
1826 | * return expired entry, or NULL to just start from scratch in rbtree | |
1827 | */ | |
febffd61 | 1828 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 1829 | { |
30996f40 | 1830 | struct request *rq = NULL; |
1da177e4 | 1831 | |
3b18152c | 1832 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 1833 | return NULL; |
cb887411 JA |
1834 | |
1835 | cfq_mark_cfqq_fifo_expire(cfqq); | |
1836 | ||
89850f7e JA |
1837 | if (list_empty(&cfqq->fifo)) |
1838 | return NULL; | |
1da177e4 | 1839 | |
89850f7e | 1840 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 1841 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 1842 | rq = NULL; |
1da177e4 | 1843 | |
30996f40 | 1844 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 1845 | return rq; |
1da177e4 LT |
1846 | } |
1847 | ||
22e2c507 JA |
1848 | static inline int |
1849 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1850 | { | |
1851 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 1852 | |
22e2c507 | 1853 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 1854 | |
22e2c507 | 1855 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
1856 | } |
1857 | ||
df5fe3e8 JM |
1858 | /* |
1859 | * Must be called with the queue_lock held. | |
1860 | */ | |
1861 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
1862 | { | |
1863 | int process_refs, io_refs; | |
1864 | ||
1865 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
1866 | process_refs = atomic_read(&cfqq->ref) - io_refs; | |
1867 | BUG_ON(process_refs < 0); | |
1868 | return process_refs; | |
1869 | } | |
1870 | ||
1871 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
1872 | { | |
e6c5bc73 | 1873 | int process_refs, new_process_refs; |
df5fe3e8 JM |
1874 | struct cfq_queue *__cfqq; |
1875 | ||
1876 | /* Avoid a circular list and skip interim queue merges */ | |
1877 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
1878 | if (__cfqq == cfqq) | |
1879 | return; | |
1880 | new_cfqq = __cfqq; | |
1881 | } | |
1882 | ||
1883 | process_refs = cfqq_process_refs(cfqq); | |
1884 | /* | |
1885 | * If the process for the cfqq has gone away, there is no | |
1886 | * sense in merging the queues. | |
1887 | */ | |
1888 | if (process_refs == 0) | |
1889 | return; | |
1890 | ||
e6c5bc73 JM |
1891 | /* |
1892 | * Merge in the direction of the lesser amount of work. | |
1893 | */ | |
1894 | new_process_refs = cfqq_process_refs(new_cfqq); | |
1895 | if (new_process_refs >= process_refs) { | |
1896 | cfqq->new_cfqq = new_cfqq; | |
1897 | atomic_add(process_refs, &new_cfqq->ref); | |
1898 | } else { | |
1899 | new_cfqq->new_cfqq = cfqq; | |
1900 | atomic_add(new_process_refs, &cfqq->ref); | |
1901 | } | |
df5fe3e8 JM |
1902 | } |
1903 | ||
cdb16e8f VG |
1904 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
1905 | struct cfq_group *cfqg, enum wl_prio_t prio, | |
1906 | bool prio_changed) | |
718eee05 CZ |
1907 | { |
1908 | struct cfq_queue *queue; | |
1909 | int i; | |
1910 | bool key_valid = false; | |
1911 | unsigned long lowest_key = 0; | |
1912 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
1913 | ||
1914 | if (prio_changed) { | |
1915 | /* | |
1916 | * When priorities switched, we prefer starting | |
1917 | * from SYNC_NOIDLE (first choice), or just SYNC | |
1918 | * over ASYNC | |
1919 | */ | |
cdb16e8f | 1920 | if (service_tree_for(cfqg, prio, cur_best, cfqd)->count) |
718eee05 CZ |
1921 | return cur_best; |
1922 | cur_best = SYNC_WORKLOAD; | |
cdb16e8f | 1923 | if (service_tree_for(cfqg, prio, cur_best, cfqd)->count) |
718eee05 CZ |
1924 | return cur_best; |
1925 | ||
1926 | return ASYNC_WORKLOAD; | |
1927 | } | |
1928 | ||
1929 | for (i = 0; i < 3; ++i) { | |
1930 | /* otherwise, select the one with lowest rb_key */ | |
cdb16e8f | 1931 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i, cfqd)); |
718eee05 CZ |
1932 | if (queue && |
1933 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
1934 | lowest_key = queue->rb_key; | |
1935 | cur_best = i; | |
1936 | key_valid = true; | |
1937 | } | |
1938 | } | |
1939 | ||
1940 | return cur_best; | |
1941 | } | |
1942 | ||
cdb16e8f | 1943 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
718eee05 CZ |
1944 | { |
1945 | enum wl_prio_t previous_prio = cfqd->serving_prio; | |
1946 | bool prio_changed; | |
1947 | unsigned slice; | |
1948 | unsigned count; | |
cdb16e8f | 1949 | struct cfq_rb_root *st; |
58ff82f3 | 1950 | unsigned group_slice; |
718eee05 | 1951 | |
1fa8f6d6 VG |
1952 | if (!cfqg) { |
1953 | cfqd->serving_prio = IDLE_WORKLOAD; | |
1954 | cfqd->workload_expires = jiffies + 1; | |
1955 | return; | |
1956 | } | |
1957 | ||
718eee05 | 1958 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 1959 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
718eee05 | 1960 | cfqd->serving_prio = RT_WORKLOAD; |
58ff82f3 | 1961 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
718eee05 CZ |
1962 | cfqd->serving_prio = BE_WORKLOAD; |
1963 | else { | |
1964 | cfqd->serving_prio = IDLE_WORKLOAD; | |
1965 | cfqd->workload_expires = jiffies + 1; | |
1966 | return; | |
1967 | } | |
1968 | ||
1969 | /* | |
1970 | * For RT and BE, we have to choose also the type | |
1971 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
1972 | * expiration time | |
1973 | */ | |
1974 | prio_changed = (cfqd->serving_prio != previous_prio); | |
cdb16e8f VG |
1975 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type, |
1976 | cfqd); | |
1977 | count = st->count; | |
718eee05 CZ |
1978 | |
1979 | /* | |
1980 | * If priority didn't change, check workload expiration, | |
1981 | * and that we still have other queues ready | |
1982 | */ | |
1983 | if (!prio_changed && count && | |
1984 | !time_after(jiffies, cfqd->workload_expires)) | |
1985 | return; | |
1986 | ||
1987 | /* otherwise select new workload type */ | |
1988 | cfqd->serving_type = | |
cdb16e8f VG |
1989 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio, prio_changed); |
1990 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type, | |
1991 | cfqd); | |
1992 | count = st->count; | |
718eee05 CZ |
1993 | |
1994 | /* | |
1995 | * the workload slice is computed as a fraction of target latency | |
1996 | * proportional to the number of queues in that workload, over | |
1997 | * all the queues in the same priority class | |
1998 | */ | |
58ff82f3 VG |
1999 | group_slice = cfq_group_slice(cfqd, cfqg); |
2000 | ||
2001 | slice = group_slice * count / | |
2002 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], | |
2003 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); | |
718eee05 CZ |
2004 | |
2005 | if (cfqd->serving_type == ASYNC_WORKLOAD) | |
2006 | /* async workload slice is scaled down according to | |
2007 | * the sync/async slice ratio. */ | |
2008 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
2009 | else | |
2010 | /* sync workload slice is at least 2 * cfq_slice_idle */ | |
2011 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2012 | ||
2013 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
2014 | cfqd->workload_expires = jiffies + slice; | |
8e550632 | 2015 | cfqd->noidle_tree_requires_idle = false; |
718eee05 CZ |
2016 | } |
2017 | ||
1fa8f6d6 VG |
2018 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2019 | { | |
2020 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2021 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2022 | |
2023 | if (RB_EMPTY_ROOT(&st->rb)) | |
2024 | return NULL; | |
25bc6b07 VG |
2025 | cfqg = cfq_rb_first_group(st); |
2026 | st->active = &cfqg->rb_node; | |
2027 | update_min_vdisktime(st); | |
2028 | return cfqg; | |
1fa8f6d6 VG |
2029 | } |
2030 | ||
cdb16e8f VG |
2031 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2032 | { | |
1fa8f6d6 VG |
2033 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2034 | ||
2035 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2036 | |
2037 | /* Restore the workload type data */ | |
2038 | if (cfqg->saved_workload_slice) { | |
2039 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; | |
2040 | cfqd->serving_type = cfqg->saved_workload; | |
2041 | cfqd->serving_prio = cfqg->saved_serving_prio; | |
2042 | } | |
1fa8f6d6 | 2043 | choose_service_tree(cfqd, cfqg); |
cdb16e8f VG |
2044 | } |
2045 | ||
22e2c507 | 2046 | /* |
498d3aa2 JA |
2047 | * Select a queue for service. If we have a current active queue, |
2048 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2049 | */ |
1b5ed5e1 | 2050 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2051 | { |
a36e71f9 | 2052 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2053 | |
22e2c507 JA |
2054 | cfqq = cfqd->active_queue; |
2055 | if (!cfqq) | |
2056 | goto new_queue; | |
1da177e4 | 2057 | |
f04a6424 VG |
2058 | if (!cfqd->rq_queued) |
2059 | return NULL; | |
22e2c507 | 2060 | /* |
6d048f53 | 2061 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2062 | */ |
b029195d | 2063 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) |
3b18152c | 2064 | goto expire; |
1da177e4 | 2065 | |
22e2c507 | 2066 | /* |
6d048f53 JA |
2067 | * The active queue has requests and isn't expired, allow it to |
2068 | * dispatch. | |
22e2c507 | 2069 | */ |
dd67d051 | 2070 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2071 | goto keep_queue; |
6d048f53 | 2072 | |
a36e71f9 JA |
2073 | /* |
2074 | * If another queue has a request waiting within our mean seek | |
2075 | * distance, let it run. The expire code will check for close | |
2076 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2077 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2078 | */ |
b3b6d040 | 2079 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2080 | if (new_cfqq) { |
2081 | if (!cfqq->new_cfqq) | |
2082 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2083 | goto expire; |
df5fe3e8 | 2084 | } |
a36e71f9 | 2085 | |
6d048f53 JA |
2086 | /* |
2087 | * No requests pending. If the active queue still has requests in | |
2088 | * flight or is idling for a new request, allow either of these | |
2089 | * conditions to happen (or time out) before selecting a new queue. | |
2090 | */ | |
cc197479 | 2091 | if (timer_pending(&cfqd->idle_slice_timer) || |
a6d44e98 | 2092 | (cfqq->dispatched && cfq_should_idle(cfqd, cfqq))) { |
caaa5f9f JA |
2093 | cfqq = NULL; |
2094 | goto keep_queue; | |
22e2c507 JA |
2095 | } |
2096 | ||
3b18152c | 2097 | expire: |
6084cdda | 2098 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2099 | new_queue: |
718eee05 CZ |
2100 | /* |
2101 | * Current queue expired. Check if we have to switch to a new | |
2102 | * service tree | |
2103 | */ | |
2104 | if (!new_cfqq) | |
cdb16e8f | 2105 | cfq_choose_cfqg(cfqd); |
718eee05 | 2106 | |
a36e71f9 | 2107 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2108 | keep_queue: |
3b18152c | 2109 | return cfqq; |
22e2c507 JA |
2110 | } |
2111 | ||
febffd61 | 2112 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2113 | { |
2114 | int dispatched = 0; | |
2115 | ||
2116 | while (cfqq->next_rq) { | |
2117 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2118 | dispatched++; | |
2119 | } | |
2120 | ||
2121 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2122 | |
2123 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
2124 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); | |
d9e7620e JA |
2125 | return dispatched; |
2126 | } | |
2127 | ||
498d3aa2 JA |
2128 | /* |
2129 | * Drain our current requests. Used for barriers and when switching | |
2130 | * io schedulers on-the-fly. | |
2131 | */ | |
d9e7620e | 2132 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2133 | { |
0871714e | 2134 | struct cfq_queue *cfqq; |
d9e7620e | 2135 | int dispatched = 0; |
cdb16e8f | 2136 | |
f04a6424 VG |
2137 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) |
2138 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); | |
1b5ed5e1 | 2139 | |
6084cdda | 2140 | cfq_slice_expired(cfqd, 0); |
1b5ed5e1 TH |
2141 | BUG_ON(cfqd->busy_queues); |
2142 | ||
6923715a | 2143 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2144 | return dispatched; |
2145 | } | |
2146 | ||
0b182d61 | 2147 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2148 | { |
2f5cb738 | 2149 | unsigned int max_dispatch; |
22e2c507 | 2150 | |
5ad531db JA |
2151 | /* |
2152 | * Drain async requests before we start sync IO | |
2153 | */ | |
a6d44e98 | 2154 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_driver[BLK_RW_ASYNC]) |
0b182d61 | 2155 | return false; |
5ad531db | 2156 | |
2f5cb738 JA |
2157 | /* |
2158 | * If this is an async queue and we have sync IO in flight, let it wait | |
2159 | */ | |
2160 | if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq)) | |
0b182d61 | 2161 | return false; |
2f5cb738 JA |
2162 | |
2163 | max_dispatch = cfqd->cfq_quantum; | |
2164 | if (cfq_class_idle(cfqq)) | |
2165 | max_dispatch = 1; | |
b4878f24 | 2166 | |
2f5cb738 JA |
2167 | /* |
2168 | * Does this cfqq already have too much IO in flight? | |
2169 | */ | |
2170 | if (cfqq->dispatched >= max_dispatch) { | |
2171 | /* | |
2172 | * idle queue must always only have a single IO in flight | |
2173 | */ | |
3ed9a296 | 2174 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2175 | return false; |
3ed9a296 | 2176 | |
2f5cb738 JA |
2177 | /* |
2178 | * We have other queues, don't allow more IO from this one | |
2179 | */ | |
2180 | if (cfqd->busy_queues > 1) | |
0b182d61 | 2181 | return false; |
9ede209e | 2182 | |
365722bb | 2183 | /* |
474b18cc | 2184 | * Sole queue user, no limit |
365722bb | 2185 | */ |
474b18cc | 2186 | max_dispatch = -1; |
8e296755 JA |
2187 | } |
2188 | ||
2189 | /* | |
2190 | * Async queues must wait a bit before being allowed dispatch. | |
2191 | * We also ramp up the dispatch depth gradually for async IO, | |
2192 | * based on the last sync IO we serviced | |
2193 | */ | |
963b72fc | 2194 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
8e296755 JA |
2195 | unsigned long last_sync = jiffies - cfqd->last_end_sync_rq; |
2196 | unsigned int depth; | |
365722bb | 2197 | |
61f0c1dc | 2198 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2199 | if (!depth && !cfqq->dispatched) |
2200 | depth = 1; | |
8e296755 JA |
2201 | if (depth < max_dispatch) |
2202 | max_dispatch = depth; | |
2f5cb738 | 2203 | } |
3ed9a296 | 2204 | |
0b182d61 JA |
2205 | /* |
2206 | * If we're below the current max, allow a dispatch | |
2207 | */ | |
2208 | return cfqq->dispatched < max_dispatch; | |
2209 | } | |
2210 | ||
2211 | /* | |
2212 | * Dispatch a request from cfqq, moving them to the request queue | |
2213 | * dispatch list. | |
2214 | */ | |
2215 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2216 | { | |
2217 | struct request *rq; | |
2218 | ||
2219 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
2220 | ||
2221 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
2222 | return false; | |
2223 | ||
2224 | /* | |
2225 | * follow expired path, else get first next available | |
2226 | */ | |
2227 | rq = cfq_check_fifo(cfqq); | |
2228 | if (!rq) | |
2229 | rq = cfqq->next_rq; | |
2230 | ||
2231 | /* | |
2232 | * insert request into driver dispatch list | |
2233 | */ | |
2234 | cfq_dispatch_insert(cfqd->queue, rq); | |
2235 | ||
2236 | if (!cfqd->active_cic) { | |
2237 | struct cfq_io_context *cic = RQ_CIC(rq); | |
2238 | ||
2239 | atomic_long_inc(&cic->ioc->refcount); | |
2240 | cfqd->active_cic = cic; | |
2241 | } | |
2242 | ||
2243 | return true; | |
2244 | } | |
2245 | ||
2246 | /* | |
2247 | * Find the cfqq that we need to service and move a request from that to the | |
2248 | * dispatch list | |
2249 | */ | |
2250 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
2251 | { | |
2252 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2253 | struct cfq_queue *cfqq; | |
2254 | ||
2255 | if (!cfqd->busy_queues) | |
2256 | return 0; | |
2257 | ||
2258 | if (unlikely(force)) | |
2259 | return cfq_forced_dispatch(cfqd); | |
2260 | ||
2261 | cfqq = cfq_select_queue(cfqd); | |
2262 | if (!cfqq) | |
8e296755 JA |
2263 | return 0; |
2264 | ||
2f5cb738 | 2265 | /* |
0b182d61 | 2266 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 2267 | */ |
0b182d61 JA |
2268 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2269 | return 0; | |
2270 | ||
2f5cb738 | 2271 | cfqq->slice_dispatch++; |
b029195d | 2272 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 2273 | |
2f5cb738 JA |
2274 | /* |
2275 | * expire an async queue immediately if it has used up its slice. idle | |
2276 | * queue always expire after 1 dispatch round. | |
2277 | */ | |
2278 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
2279 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
2280 | cfq_class_idle(cfqq))) { | |
2281 | cfqq->slice_end = jiffies + 1; | |
2282 | cfq_slice_expired(cfqd, 0); | |
1da177e4 LT |
2283 | } |
2284 | ||
b217a903 | 2285 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 2286 | return 1; |
1da177e4 LT |
2287 | } |
2288 | ||
1da177e4 | 2289 | /* |
5e705374 JA |
2290 | * task holds one reference to the queue, dropped when task exits. each rq |
2291 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 2292 | * |
b1c35769 | 2293 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
2294 | * queue lock must be held here. |
2295 | */ | |
2296 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
2297 | { | |
22e2c507 | 2298 | struct cfq_data *cfqd = cfqq->cfqd; |
b1c35769 | 2299 | struct cfq_group *cfqg; |
22e2c507 JA |
2300 | |
2301 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
2302 | |
2303 | if (!atomic_dec_and_test(&cfqq->ref)) | |
2304 | return; | |
2305 | ||
7b679138 | 2306 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 2307 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 2308 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 2309 | cfqg = cfqq->cfqg; |
1da177e4 | 2310 | |
28f95cbc | 2311 | if (unlikely(cfqd->active_queue == cfqq)) { |
6084cdda | 2312 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2313 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2314 | } |
22e2c507 | 2315 | |
f04a6424 | 2316 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 2317 | kmem_cache_free(cfq_pool, cfqq); |
b1c35769 | 2318 | cfq_put_cfqg(cfqg); |
1da177e4 LT |
2319 | } |
2320 | ||
d6de8be7 JA |
2321 | /* |
2322 | * Must always be called with the rcu_read_lock() held | |
2323 | */ | |
07416d29 JA |
2324 | static void |
2325 | __call_for_each_cic(struct io_context *ioc, | |
2326 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
2327 | { | |
2328 | struct cfq_io_context *cic; | |
2329 | struct hlist_node *n; | |
2330 | ||
2331 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) | |
2332 | func(ioc, cic); | |
2333 | } | |
2334 | ||
4ac845a2 | 2335 | /* |
34e6bbf2 | 2336 | * Call func for each cic attached to this ioc. |
4ac845a2 | 2337 | */ |
34e6bbf2 | 2338 | static void |
4ac845a2 JA |
2339 | call_for_each_cic(struct io_context *ioc, |
2340 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1da177e4 | 2341 | { |
4ac845a2 | 2342 | rcu_read_lock(); |
07416d29 | 2343 | __call_for_each_cic(ioc, func); |
4ac845a2 | 2344 | rcu_read_unlock(); |
34e6bbf2 FC |
2345 | } |
2346 | ||
2347 | static void cfq_cic_free_rcu(struct rcu_head *head) | |
2348 | { | |
2349 | struct cfq_io_context *cic; | |
2350 | ||
2351 | cic = container_of(head, struct cfq_io_context, rcu_head); | |
2352 | ||
2353 | kmem_cache_free(cfq_ioc_pool, cic); | |
245b2e70 | 2354 | elv_ioc_count_dec(cfq_ioc_count); |
34e6bbf2 | 2355 | |
9a11b4ed JA |
2356 | if (ioc_gone) { |
2357 | /* | |
2358 | * CFQ scheduler is exiting, grab exit lock and check | |
2359 | * the pending io context count. If it hits zero, | |
2360 | * complete ioc_gone and set it back to NULL | |
2361 | */ | |
2362 | spin_lock(&ioc_gone_lock); | |
245b2e70 | 2363 | if (ioc_gone && !elv_ioc_count_read(cfq_ioc_count)) { |
9a11b4ed JA |
2364 | complete(ioc_gone); |
2365 | ioc_gone = NULL; | |
2366 | } | |
2367 | spin_unlock(&ioc_gone_lock); | |
2368 | } | |
34e6bbf2 | 2369 | } |
4ac845a2 | 2370 | |
34e6bbf2 FC |
2371 | static void cfq_cic_free(struct cfq_io_context *cic) |
2372 | { | |
2373 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | |
4ac845a2 JA |
2374 | } |
2375 | ||
2376 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) | |
2377 | { | |
2378 | unsigned long flags; | |
2379 | ||
2380 | BUG_ON(!cic->dead_key); | |
2381 | ||
2382 | spin_lock_irqsave(&ioc->lock, flags); | |
2383 | radix_tree_delete(&ioc->radix_root, cic->dead_key); | |
ffc4e759 | 2384 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
2385 | spin_unlock_irqrestore(&ioc->lock, flags); |
2386 | ||
34e6bbf2 | 2387 | cfq_cic_free(cic); |
4ac845a2 JA |
2388 | } |
2389 | ||
d6de8be7 JA |
2390 | /* |
2391 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. | |
2392 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), | |
2393 | * and ->trim() which is called with the task lock held | |
2394 | */ | |
4ac845a2 JA |
2395 | static void cfq_free_io_context(struct io_context *ioc) |
2396 | { | |
4ac845a2 | 2397 | /* |
34e6bbf2 FC |
2398 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
2399 | * so no more cic's are allowed to be linked into this ioc. So it | |
2400 | * should be ok to iterate over the known list, we will see all cic's | |
2401 | * since no new ones are added. | |
4ac845a2 | 2402 | */ |
07416d29 | 2403 | __call_for_each_cic(ioc, cic_free_func); |
1da177e4 LT |
2404 | } |
2405 | ||
89850f7e | 2406 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2407 | { |
df5fe3e8 JM |
2408 | struct cfq_queue *__cfqq, *next; |
2409 | ||
28f95cbc | 2410 | if (unlikely(cfqq == cfqd->active_queue)) { |
6084cdda | 2411 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2412 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2413 | } |
22e2c507 | 2414 | |
df5fe3e8 JM |
2415 | /* |
2416 | * If this queue was scheduled to merge with another queue, be | |
2417 | * sure to drop the reference taken on that queue (and others in | |
2418 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
2419 | */ | |
2420 | __cfqq = cfqq->new_cfqq; | |
2421 | while (__cfqq) { | |
2422 | if (__cfqq == cfqq) { | |
2423 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
2424 | break; | |
2425 | } | |
2426 | next = __cfqq->new_cfqq; | |
2427 | cfq_put_queue(__cfqq); | |
2428 | __cfqq = next; | |
2429 | } | |
2430 | ||
89850f7e JA |
2431 | cfq_put_queue(cfqq); |
2432 | } | |
22e2c507 | 2433 | |
89850f7e JA |
2434 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
2435 | struct cfq_io_context *cic) | |
2436 | { | |
4faa3c81 FC |
2437 | struct io_context *ioc = cic->ioc; |
2438 | ||
fc46379d | 2439 | list_del_init(&cic->queue_list); |
4ac845a2 JA |
2440 | |
2441 | /* | |
2442 | * Make sure key == NULL is seen for dead queues | |
2443 | */ | |
fc46379d | 2444 | smp_wmb(); |
4ac845a2 | 2445 | cic->dead_key = (unsigned long) cic->key; |
fc46379d JA |
2446 | cic->key = NULL; |
2447 | ||
4faa3c81 FC |
2448 | if (ioc->ioc_data == cic) |
2449 | rcu_assign_pointer(ioc->ioc_data, NULL); | |
2450 | ||
ff6657c6 JA |
2451 | if (cic->cfqq[BLK_RW_ASYNC]) { |
2452 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
2453 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
2454 | } |
2455 | ||
ff6657c6 JA |
2456 | if (cic->cfqq[BLK_RW_SYNC]) { |
2457 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
2458 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 2459 | } |
89850f7e JA |
2460 | } |
2461 | ||
4ac845a2 JA |
2462 | static void cfq_exit_single_io_context(struct io_context *ioc, |
2463 | struct cfq_io_context *cic) | |
89850f7e JA |
2464 | { |
2465 | struct cfq_data *cfqd = cic->key; | |
2466 | ||
89850f7e | 2467 | if (cfqd) { |
165125e1 | 2468 | struct request_queue *q = cfqd->queue; |
4ac845a2 | 2469 | unsigned long flags; |
89850f7e | 2470 | |
4ac845a2 | 2471 | spin_lock_irqsave(q->queue_lock, flags); |
62c1fe9d JA |
2472 | |
2473 | /* | |
2474 | * Ensure we get a fresh copy of the ->key to prevent | |
2475 | * race between exiting task and queue | |
2476 | */ | |
2477 | smp_read_barrier_depends(); | |
2478 | if (cic->key) | |
2479 | __cfq_exit_single_io_context(cfqd, cic); | |
2480 | ||
4ac845a2 | 2481 | spin_unlock_irqrestore(q->queue_lock, flags); |
89850f7e | 2482 | } |
1da177e4 LT |
2483 | } |
2484 | ||
498d3aa2 JA |
2485 | /* |
2486 | * The process that ioc belongs to has exited, we need to clean up | |
2487 | * and put the internal structures we have that belongs to that process. | |
2488 | */ | |
e2d74ac0 | 2489 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 2490 | { |
4ac845a2 | 2491 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
1da177e4 LT |
2492 | } |
2493 | ||
22e2c507 | 2494 | static struct cfq_io_context * |
8267e268 | 2495 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 2496 | { |
b5deef90 | 2497 | struct cfq_io_context *cic; |
1da177e4 | 2498 | |
94f6030c CL |
2499 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
2500 | cfqd->queue->node); | |
1da177e4 | 2501 | if (cic) { |
22e2c507 | 2502 | cic->last_end_request = jiffies; |
553698f9 | 2503 | INIT_LIST_HEAD(&cic->queue_list); |
ffc4e759 | 2504 | INIT_HLIST_NODE(&cic->cic_list); |
22e2c507 JA |
2505 | cic->dtor = cfq_free_io_context; |
2506 | cic->exit = cfq_exit_io_context; | |
245b2e70 | 2507 | elv_ioc_count_inc(cfq_ioc_count); |
1da177e4 LT |
2508 | } |
2509 | ||
2510 | return cic; | |
2511 | } | |
2512 | ||
fd0928df | 2513 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
2514 | { |
2515 | struct task_struct *tsk = current; | |
2516 | int ioprio_class; | |
2517 | ||
3b18152c | 2518 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
2519 | return; |
2520 | ||
fd0928df | 2521 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 2522 | switch (ioprio_class) { |
fe094d98 JA |
2523 | default: |
2524 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
2525 | case IOPRIO_CLASS_NONE: | |
2526 | /* | |
6d63c275 | 2527 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
2528 | */ |
2529 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 2530 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
2531 | break; |
2532 | case IOPRIO_CLASS_RT: | |
2533 | cfqq->ioprio = task_ioprio(ioc); | |
2534 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
2535 | break; | |
2536 | case IOPRIO_CLASS_BE: | |
2537 | cfqq->ioprio = task_ioprio(ioc); | |
2538 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
2539 | break; | |
2540 | case IOPRIO_CLASS_IDLE: | |
2541 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
2542 | cfqq->ioprio = 7; | |
2543 | cfq_clear_cfqq_idle_window(cfqq); | |
2544 | break; | |
22e2c507 JA |
2545 | } |
2546 | ||
2547 | /* | |
2548 | * keep track of original prio settings in case we have to temporarily | |
2549 | * elevate the priority of this queue | |
2550 | */ | |
2551 | cfqq->org_ioprio = cfqq->ioprio; | |
2552 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 2553 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
2554 | } |
2555 | ||
febffd61 | 2556 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
22e2c507 | 2557 | { |
478a82b0 AV |
2558 | struct cfq_data *cfqd = cic->key; |
2559 | struct cfq_queue *cfqq; | |
c1b707d2 | 2560 | unsigned long flags; |
35e6077c | 2561 | |
caaa5f9f JA |
2562 | if (unlikely(!cfqd)) |
2563 | return; | |
2564 | ||
c1b707d2 | 2565 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f | 2566 | |
ff6657c6 | 2567 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
2568 | if (cfqq) { |
2569 | struct cfq_queue *new_cfqq; | |
ff6657c6 JA |
2570 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc, |
2571 | GFP_ATOMIC); | |
caaa5f9f | 2572 | if (new_cfqq) { |
ff6657c6 | 2573 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
2574 | cfq_put_queue(cfqq); |
2575 | } | |
22e2c507 | 2576 | } |
caaa5f9f | 2577 | |
ff6657c6 | 2578 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
2579 | if (cfqq) |
2580 | cfq_mark_cfqq_prio_changed(cfqq); | |
2581 | ||
c1b707d2 | 2582 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
2583 | } |
2584 | ||
fc46379d | 2585 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 2586 | { |
4ac845a2 | 2587 | call_for_each_cic(ioc, changed_ioprio); |
fc46379d | 2588 | ioc->ioprio_changed = 0; |
22e2c507 JA |
2589 | } |
2590 | ||
d5036d77 | 2591 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2592 | pid_t pid, bool is_sync) |
d5036d77 JA |
2593 | { |
2594 | RB_CLEAR_NODE(&cfqq->rb_node); | |
2595 | RB_CLEAR_NODE(&cfqq->p_node); | |
2596 | INIT_LIST_HEAD(&cfqq->fifo); | |
2597 | ||
2598 | atomic_set(&cfqq->ref, 0); | |
2599 | cfqq->cfqd = cfqd; | |
2600 | ||
2601 | cfq_mark_cfqq_prio_changed(cfqq); | |
2602 | ||
2603 | if (is_sync) { | |
2604 | if (!cfq_class_idle(cfqq)) | |
2605 | cfq_mark_cfqq_idle_window(cfqq); | |
2606 | cfq_mark_cfqq_sync(cfqq); | |
2607 | } | |
2608 | cfqq->pid = pid; | |
2609 | } | |
2610 | ||
24610333 VG |
2611 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2612 | static void changed_cgroup(struct io_context *ioc, struct cfq_io_context *cic) | |
2613 | { | |
2614 | struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1); | |
2615 | struct cfq_data *cfqd = cic->key; | |
2616 | unsigned long flags; | |
2617 | struct request_queue *q; | |
2618 | ||
2619 | if (unlikely(!cfqd)) | |
2620 | return; | |
2621 | ||
2622 | q = cfqd->queue; | |
2623 | ||
2624 | spin_lock_irqsave(q->queue_lock, flags); | |
2625 | ||
2626 | if (sync_cfqq) { | |
2627 | /* | |
2628 | * Drop reference to sync queue. A new sync queue will be | |
2629 | * assigned in new group upon arrival of a fresh request. | |
2630 | */ | |
2631 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
2632 | cic_set_cfqq(cic, NULL, 1); | |
2633 | cfq_put_queue(sync_cfqq); | |
2634 | } | |
2635 | ||
2636 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2637 | } | |
2638 | ||
2639 | static void cfq_ioc_set_cgroup(struct io_context *ioc) | |
2640 | { | |
2641 | call_for_each_cic(ioc, changed_cgroup); | |
2642 | ioc->cgroup_changed = 0; | |
2643 | } | |
2644 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
2645 | ||
22e2c507 | 2646 | static struct cfq_queue * |
a6151c3a | 2647 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, |
fd0928df | 2648 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 2649 | { |
22e2c507 | 2650 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 2651 | struct cfq_io_context *cic; |
cdb16e8f | 2652 | struct cfq_group *cfqg; |
22e2c507 JA |
2653 | |
2654 | retry: | |
cdb16e8f | 2655 | cfqg = cfq_get_cfqg(cfqd, 1); |
4ac845a2 | 2656 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
2657 | /* cic always exists here */ |
2658 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 | 2659 | |
6118b70b JA |
2660 | /* |
2661 | * Always try a new alloc if we fell back to the OOM cfqq | |
2662 | * originally, since it should just be a temporary situation. | |
2663 | */ | |
2664 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
2665 | cfqq = NULL; | |
22e2c507 JA |
2666 | if (new_cfqq) { |
2667 | cfqq = new_cfqq; | |
2668 | new_cfqq = NULL; | |
2669 | } else if (gfp_mask & __GFP_WAIT) { | |
2670 | spin_unlock_irq(cfqd->queue->queue_lock); | |
94f6030c | 2671 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 2672 | gfp_mask | __GFP_ZERO, |
94f6030c | 2673 | cfqd->queue->node); |
22e2c507 | 2674 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
2675 | if (new_cfqq) |
2676 | goto retry; | |
22e2c507 | 2677 | } else { |
94f6030c CL |
2678 | cfqq = kmem_cache_alloc_node(cfq_pool, |
2679 | gfp_mask | __GFP_ZERO, | |
2680 | cfqd->queue->node); | |
22e2c507 JA |
2681 | } |
2682 | ||
6118b70b JA |
2683 | if (cfqq) { |
2684 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
2685 | cfq_init_prio_data(cfqq, ioc); | |
cdb16e8f | 2686 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
2687 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
2688 | } else | |
2689 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
2690 | } |
2691 | ||
2692 | if (new_cfqq) | |
2693 | kmem_cache_free(cfq_pool, new_cfqq); | |
2694 | ||
22e2c507 JA |
2695 | return cfqq; |
2696 | } | |
2697 | ||
c2dea2d1 VT |
2698 | static struct cfq_queue ** |
2699 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
2700 | { | |
fe094d98 | 2701 | switch (ioprio_class) { |
c2dea2d1 VT |
2702 | case IOPRIO_CLASS_RT: |
2703 | return &cfqd->async_cfqq[0][ioprio]; | |
2704 | case IOPRIO_CLASS_BE: | |
2705 | return &cfqd->async_cfqq[1][ioprio]; | |
2706 | case IOPRIO_CLASS_IDLE: | |
2707 | return &cfqd->async_idle_cfqq; | |
2708 | default: | |
2709 | BUG(); | |
2710 | } | |
2711 | } | |
2712 | ||
15c31be4 | 2713 | static struct cfq_queue * |
a6151c3a | 2714 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc, |
15c31be4 JA |
2715 | gfp_t gfp_mask) |
2716 | { | |
fd0928df JA |
2717 | const int ioprio = task_ioprio(ioc); |
2718 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 2719 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
2720 | struct cfq_queue *cfqq = NULL; |
2721 | ||
c2dea2d1 VT |
2722 | if (!is_sync) { |
2723 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
2724 | cfqq = *async_cfqq; | |
2725 | } | |
2726 | ||
6118b70b | 2727 | if (!cfqq) |
fd0928df | 2728 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
15c31be4 JA |
2729 | |
2730 | /* | |
2731 | * pin the queue now that it's allocated, scheduler exit will prune it | |
2732 | */ | |
c2dea2d1 | 2733 | if (!is_sync && !(*async_cfqq)) { |
15c31be4 | 2734 | atomic_inc(&cfqq->ref); |
c2dea2d1 | 2735 | *async_cfqq = cfqq; |
15c31be4 JA |
2736 | } |
2737 | ||
2738 | atomic_inc(&cfqq->ref); | |
2739 | return cfqq; | |
2740 | } | |
2741 | ||
498d3aa2 JA |
2742 | /* |
2743 | * We drop cfq io contexts lazily, so we may find a dead one. | |
2744 | */ | |
dbecf3ab | 2745 | static void |
4ac845a2 JA |
2746 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
2747 | struct cfq_io_context *cic) | |
dbecf3ab | 2748 | { |
4ac845a2 JA |
2749 | unsigned long flags; |
2750 | ||
fc46379d | 2751 | WARN_ON(!list_empty(&cic->queue_list)); |
597bc485 | 2752 | |
4ac845a2 JA |
2753 | spin_lock_irqsave(&ioc->lock, flags); |
2754 | ||
4faa3c81 | 2755 | BUG_ON(ioc->ioc_data == cic); |
597bc485 | 2756 | |
4ac845a2 | 2757 | radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd); |
ffc4e759 | 2758 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
2759 | spin_unlock_irqrestore(&ioc->lock, flags); |
2760 | ||
2761 | cfq_cic_free(cic); | |
dbecf3ab OH |
2762 | } |
2763 | ||
e2d74ac0 | 2764 | static struct cfq_io_context * |
4ac845a2 | 2765 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
e2d74ac0 | 2766 | { |
e2d74ac0 | 2767 | struct cfq_io_context *cic; |
d6de8be7 | 2768 | unsigned long flags; |
4ac845a2 | 2769 | void *k; |
e2d74ac0 | 2770 | |
91fac317 VT |
2771 | if (unlikely(!ioc)) |
2772 | return NULL; | |
2773 | ||
d6de8be7 JA |
2774 | rcu_read_lock(); |
2775 | ||
597bc485 JA |
2776 | /* |
2777 | * we maintain a last-hit cache, to avoid browsing over the tree | |
2778 | */ | |
4ac845a2 | 2779 | cic = rcu_dereference(ioc->ioc_data); |
d6de8be7 JA |
2780 | if (cic && cic->key == cfqd) { |
2781 | rcu_read_unlock(); | |
597bc485 | 2782 | return cic; |
d6de8be7 | 2783 | } |
597bc485 | 2784 | |
4ac845a2 | 2785 | do { |
4ac845a2 JA |
2786 | cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd); |
2787 | rcu_read_unlock(); | |
2788 | if (!cic) | |
2789 | break; | |
be3b0753 OH |
2790 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
2791 | k = cic->key; | |
2792 | if (unlikely(!k)) { | |
4ac845a2 | 2793 | cfq_drop_dead_cic(cfqd, ioc, cic); |
d6de8be7 | 2794 | rcu_read_lock(); |
4ac845a2 | 2795 | continue; |
dbecf3ab | 2796 | } |
e2d74ac0 | 2797 | |
d6de8be7 | 2798 | spin_lock_irqsave(&ioc->lock, flags); |
4ac845a2 | 2799 | rcu_assign_pointer(ioc->ioc_data, cic); |
d6de8be7 | 2800 | spin_unlock_irqrestore(&ioc->lock, flags); |
4ac845a2 JA |
2801 | break; |
2802 | } while (1); | |
e2d74ac0 | 2803 | |
4ac845a2 | 2804 | return cic; |
e2d74ac0 JA |
2805 | } |
2806 | ||
4ac845a2 JA |
2807 | /* |
2808 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup | |
2809 | * the process specific cfq io context when entered from the block layer. | |
2810 | * Also adds the cic to a per-cfqd list, used when this queue is removed. | |
2811 | */ | |
febffd61 JA |
2812 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
2813 | struct cfq_io_context *cic, gfp_t gfp_mask) | |
e2d74ac0 | 2814 | { |
0261d688 | 2815 | unsigned long flags; |
4ac845a2 | 2816 | int ret; |
e2d74ac0 | 2817 | |
4ac845a2 JA |
2818 | ret = radix_tree_preload(gfp_mask); |
2819 | if (!ret) { | |
2820 | cic->ioc = ioc; | |
2821 | cic->key = cfqd; | |
e2d74ac0 | 2822 | |
4ac845a2 JA |
2823 | spin_lock_irqsave(&ioc->lock, flags); |
2824 | ret = radix_tree_insert(&ioc->radix_root, | |
2825 | (unsigned long) cfqd, cic); | |
ffc4e759 JA |
2826 | if (!ret) |
2827 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); | |
4ac845a2 | 2828 | spin_unlock_irqrestore(&ioc->lock, flags); |
e2d74ac0 | 2829 | |
4ac845a2 JA |
2830 | radix_tree_preload_end(); |
2831 | ||
2832 | if (!ret) { | |
2833 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
2834 | list_add(&cic->queue_list, &cfqd->cic_list); | |
2835 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
2836 | } | |
e2d74ac0 JA |
2837 | } |
2838 | ||
4ac845a2 JA |
2839 | if (ret) |
2840 | printk(KERN_ERR "cfq: cic link failed!\n"); | |
fc46379d | 2841 | |
4ac845a2 | 2842 | return ret; |
e2d74ac0 JA |
2843 | } |
2844 | ||
1da177e4 LT |
2845 | /* |
2846 | * Setup general io context and cfq io context. There can be several cfq | |
2847 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 2848 | * than one device managed by cfq. |
1da177e4 LT |
2849 | */ |
2850 | static struct cfq_io_context * | |
e2d74ac0 | 2851 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 2852 | { |
22e2c507 | 2853 | struct io_context *ioc = NULL; |
1da177e4 | 2854 | struct cfq_io_context *cic; |
1da177e4 | 2855 | |
22e2c507 | 2856 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 2857 | |
b5deef90 | 2858 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
2859 | if (!ioc) |
2860 | return NULL; | |
2861 | ||
4ac845a2 | 2862 | cic = cfq_cic_lookup(cfqd, ioc); |
e2d74ac0 JA |
2863 | if (cic) |
2864 | goto out; | |
1da177e4 | 2865 | |
e2d74ac0 JA |
2866 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
2867 | if (cic == NULL) | |
2868 | goto err; | |
1da177e4 | 2869 | |
4ac845a2 JA |
2870 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) |
2871 | goto err_free; | |
2872 | ||
1da177e4 | 2873 | out: |
fc46379d JA |
2874 | smp_read_barrier_depends(); |
2875 | if (unlikely(ioc->ioprio_changed)) | |
2876 | cfq_ioc_set_ioprio(ioc); | |
2877 | ||
24610333 VG |
2878 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2879 | if (unlikely(ioc->cgroup_changed)) | |
2880 | cfq_ioc_set_cgroup(ioc); | |
2881 | #endif | |
1da177e4 | 2882 | return cic; |
4ac845a2 JA |
2883 | err_free: |
2884 | cfq_cic_free(cic); | |
1da177e4 LT |
2885 | err: |
2886 | put_io_context(ioc); | |
2887 | return NULL; | |
2888 | } | |
2889 | ||
22e2c507 JA |
2890 | static void |
2891 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 2892 | { |
aaf1228d JA |
2893 | unsigned long elapsed = jiffies - cic->last_end_request; |
2894 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 2895 | |
22e2c507 JA |
2896 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
2897 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
2898 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
2899 | } | |
1da177e4 | 2900 | |
206dc69b | 2901 | static void |
b2c18e1e | 2902 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 2903 | struct request *rq) |
206dc69b JA |
2904 | { |
2905 | sector_t sdist; | |
2906 | u64 total; | |
2907 | ||
b2c18e1e | 2908 | if (!cfqq->last_request_pos) |
4d00aa47 | 2909 | sdist = 0; |
b2c18e1e JM |
2910 | else if (cfqq->last_request_pos < blk_rq_pos(rq)) |
2911 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
206dc69b | 2912 | else |
b2c18e1e | 2913 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); |
206dc69b JA |
2914 | |
2915 | /* | |
2916 | * Don't allow the seek distance to get too large from the | |
2917 | * odd fragment, pagein, etc | |
2918 | */ | |
b2c18e1e JM |
2919 | if (cfqq->seek_samples <= 60) /* second&third seek */ |
2920 | sdist = min(sdist, (cfqq->seek_mean * 4) + 2*1024*1024); | |
206dc69b | 2921 | else |
b2c18e1e | 2922 | sdist = min(sdist, (cfqq->seek_mean * 4) + 2*1024*64); |
206dc69b | 2923 | |
b2c18e1e JM |
2924 | cfqq->seek_samples = (7*cfqq->seek_samples + 256) / 8; |
2925 | cfqq->seek_total = (7*cfqq->seek_total + (u64)256*sdist) / 8; | |
2926 | total = cfqq->seek_total + (cfqq->seek_samples/2); | |
2927 | do_div(total, cfqq->seek_samples); | |
2928 | cfqq->seek_mean = (sector_t)total; | |
e6c5bc73 JM |
2929 | |
2930 | /* | |
2931 | * If this cfqq is shared between multiple processes, check to | |
2932 | * make sure that those processes are still issuing I/Os within | |
2933 | * the mean seek distance. If not, it may be time to break the | |
2934 | * queues apart again. | |
2935 | */ | |
2936 | if (cfq_cfqq_coop(cfqq)) { | |
2937 | if (CFQQ_SEEKY(cfqq) && !cfqq->seeky_start) | |
2938 | cfqq->seeky_start = jiffies; | |
2939 | else if (!CFQQ_SEEKY(cfqq)) | |
2940 | cfqq->seeky_start = 0; | |
2941 | } | |
206dc69b | 2942 | } |
1da177e4 | 2943 | |
22e2c507 JA |
2944 | /* |
2945 | * Disable idle window if the process thinks too long or seeks so much that | |
2946 | * it doesn't matter | |
2947 | */ | |
2948 | static void | |
2949 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
2950 | struct cfq_io_context *cic) | |
2951 | { | |
7b679138 | 2952 | int old_idle, enable_idle; |
1be92f2f | 2953 | |
0871714e JA |
2954 | /* |
2955 | * Don't idle for async or idle io prio class | |
2956 | */ | |
2957 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
2958 | return; |
2959 | ||
c265a7f4 | 2960 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 2961 | |
76280aff CZ |
2962 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
2963 | cfq_mark_cfqq_deep(cfqq); | |
2964 | ||
66dac98e | 2965 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
76280aff CZ |
2966 | (!cfq_cfqq_deep(cfqq) && sample_valid(cfqq->seek_samples) |
2967 | && CFQQ_SEEKY(cfqq))) | |
22e2c507 JA |
2968 | enable_idle = 0; |
2969 | else if (sample_valid(cic->ttime_samples)) { | |
718eee05 | 2970 | if (cic->ttime_mean > cfqd->cfq_slice_idle) |
22e2c507 JA |
2971 | enable_idle = 0; |
2972 | else | |
2973 | enable_idle = 1; | |
1da177e4 LT |
2974 | } |
2975 | ||
7b679138 JA |
2976 | if (old_idle != enable_idle) { |
2977 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
2978 | if (enable_idle) | |
2979 | cfq_mark_cfqq_idle_window(cfqq); | |
2980 | else | |
2981 | cfq_clear_cfqq_idle_window(cfqq); | |
2982 | } | |
22e2c507 | 2983 | } |
1da177e4 | 2984 | |
22e2c507 JA |
2985 | /* |
2986 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
2987 | * no or if we aren't sure, a 1 will cause a preempt. | |
2988 | */ | |
a6151c3a | 2989 | static bool |
22e2c507 | 2990 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 2991 | struct request *rq) |
22e2c507 | 2992 | { |
6d048f53 | 2993 | struct cfq_queue *cfqq; |
22e2c507 | 2994 | |
6d048f53 JA |
2995 | cfqq = cfqd->active_queue; |
2996 | if (!cfqq) | |
a6151c3a | 2997 | return false; |
22e2c507 | 2998 | |
6d048f53 | 2999 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3000 | return false; |
22e2c507 JA |
3001 | |
3002 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3003 | return true; |
1e3335de | 3004 | |
374f84ac JA |
3005 | /* |
3006 | * if the new request is sync, but the currently running queue is | |
3007 | * not, let the sync request have priority. | |
3008 | */ | |
5e705374 | 3009 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3010 | return true; |
1e3335de | 3011 | |
8682e1f1 VG |
3012 | if (new_cfqq->cfqg != cfqq->cfqg) |
3013 | return false; | |
3014 | ||
3015 | if (cfq_slice_used(cfqq)) | |
3016 | return true; | |
3017 | ||
3018 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
3019 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && | |
3020 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && | |
3021 | new_cfqq->service_tree->count == 2 && | |
3022 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3023 | return true; | |
3024 | ||
374f84ac JA |
3025 | /* |
3026 | * So both queues are sync. Let the new request get disk time if | |
3027 | * it's a metadata request and the current queue is doing regular IO. | |
3028 | */ | |
3029 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
e6ec4fe2 | 3030 | return true; |
22e2c507 | 3031 | |
3a9a3f6c DS |
3032 | /* |
3033 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3034 | */ | |
3035 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3036 | return true; |
3a9a3f6c | 3037 | |
1e3335de | 3038 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3039 | return false; |
1e3335de JA |
3040 | |
3041 | /* | |
3042 | * if this request is as-good as one we would expect from the | |
3043 | * current cfqq, let it preempt | |
3044 | */ | |
e00ef799 | 3045 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3046 | return true; |
1e3335de | 3047 | |
a6151c3a | 3048 | return false; |
22e2c507 JA |
3049 | } |
3050 | ||
3051 | /* | |
3052 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3053 | * let it have half of its nominal slice. | |
3054 | */ | |
3055 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3056 | { | |
7b679138 | 3057 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
6084cdda | 3058 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3059 | |
bf572256 JA |
3060 | /* |
3061 | * Put the new queue at the front of the of the current list, | |
3062 | * so we know that it will be selected next. | |
3063 | */ | |
3064 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3065 | |
3066 | cfq_service_tree_add(cfqd, cfqq, 1); | |
bf572256 | 3067 | |
44f7c160 JA |
3068 | cfqq->slice_end = 0; |
3069 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3070 | } |
3071 | ||
22e2c507 | 3072 | /* |
5e705374 | 3073 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3074 | * something we should do about it |
3075 | */ | |
3076 | static void | |
5e705374 JA |
3077 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3078 | struct request *rq) | |
22e2c507 | 3079 | { |
5e705374 | 3080 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 3081 | |
45333d5a | 3082 | cfqd->rq_queued++; |
374f84ac JA |
3083 | if (rq_is_meta(rq)) |
3084 | cfqq->meta_pending++; | |
3085 | ||
9c2c38a1 | 3086 | cfq_update_io_thinktime(cfqd, cic); |
b2c18e1e | 3087 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3088 | cfq_update_idle_window(cfqd, cfqq, cic); |
3089 | ||
b2c18e1e | 3090 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3091 | |
3092 | if (cfqq == cfqd->active_queue) { | |
3093 | /* | |
b029195d JA |
3094 | * Remember that we saw a request from this process, but |
3095 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3096 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3097 | * and merging. If the request is already larger than a single |
3098 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3099 | * merging is already done. Ditto for a busy system that |
3100 | * has other work pending, don't risk delaying until the | |
3101 | * idle timer unplug to continue working. | |
22e2c507 | 3102 | */ |
d6ceb25e | 3103 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3104 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3105 | cfqd->busy_queues > 1) { | |
d6ceb25e | 3106 | del_timer(&cfqd->idle_slice_timer); |
bf791937 VG |
3107 | __blk_run_queue(cfqd->queue); |
3108 | } else | |
3109 | cfq_mark_cfqq_must_dispatch(cfqq); | |
d6ceb25e | 3110 | } |
5e705374 | 3111 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3112 | /* |
3113 | * not the active queue - expire current slice if it is | |
3114 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3115 | * has some old slice time left and is of higher priority or |
3116 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3117 | */ |
3118 | cfq_preempt_queue(cfqd, cfqq); | |
a7f55792 | 3119 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3120 | } |
1da177e4 LT |
3121 | } |
3122 | ||
165125e1 | 3123 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3124 | { |
b4878f24 | 3125 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3126 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3127 | |
7b679138 | 3128 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
fd0928df | 3129 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
1da177e4 | 3130 | |
30996f40 | 3131 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3132 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3133 | cfq_add_rq_rb(rq); |
22e2c507 | 3134 | |
5e705374 | 3135 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3136 | } |
3137 | ||
45333d5a AC |
3138 | /* |
3139 | * Update hw_tag based on peak queue depth over 50 samples under | |
3140 | * sufficient load. | |
3141 | */ | |
3142 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3143 | { | |
1a1238a7 SL |
3144 | struct cfq_queue *cfqq = cfqd->active_queue; |
3145 | ||
e459dd08 CZ |
3146 | if (rq_in_driver(cfqd) > cfqd->hw_tag_est_depth) |
3147 | cfqd->hw_tag_est_depth = rq_in_driver(cfqd); | |
3148 | ||
3149 | if (cfqd->hw_tag == 1) | |
3150 | return; | |
45333d5a AC |
3151 | |
3152 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
5ad531db | 3153 | rq_in_driver(cfqd) <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3154 | return; |
3155 | ||
1a1238a7 SL |
3156 | /* |
3157 | * If active queue hasn't enough requests and can idle, cfq might not | |
3158 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3159 | * case | |
3160 | */ | |
3161 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3162 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
3163 | CFQ_HW_QUEUE_MIN && rq_in_driver(cfqd) < CFQ_HW_QUEUE_MIN) | |
3164 | return; | |
3165 | ||
45333d5a AC |
3166 | if (cfqd->hw_tag_samples++ < 50) |
3167 | return; | |
3168 | ||
e459dd08 | 3169 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3170 | cfqd->hw_tag = 1; |
3171 | else | |
3172 | cfqd->hw_tag = 0; | |
45333d5a AC |
3173 | } |
3174 | ||
165125e1 | 3175 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3176 | { |
5e705374 | 3177 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3178 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3179 | const int sync = rq_is_sync(rq); |
b4878f24 | 3180 | unsigned long now; |
1da177e4 | 3181 | |
b4878f24 | 3182 | now = jiffies; |
2868ef7b | 3183 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", !!rq_noidle(rq)); |
1da177e4 | 3184 | |
45333d5a AC |
3185 | cfq_update_hw_tag(cfqd); |
3186 | ||
5ad531db | 3187 | WARN_ON(!cfqd->rq_in_driver[sync]); |
6d048f53 | 3188 | WARN_ON(!cfqq->dispatched); |
5ad531db | 3189 | cfqd->rq_in_driver[sync]--; |
6d048f53 | 3190 | cfqq->dispatched--; |
1da177e4 | 3191 | |
3ed9a296 JA |
3192 | if (cfq_cfqq_sync(cfqq)) |
3193 | cfqd->sync_flight--; | |
3194 | ||
365722bb | 3195 | if (sync) { |
5e705374 | 3196 | RQ_CIC(rq)->last_end_request = now; |
365722bb VG |
3197 | cfqd->last_end_sync_rq = now; |
3198 | } | |
caaa5f9f JA |
3199 | |
3200 | /* | |
3201 | * If this is the active queue, check if it needs to be expired, | |
3202 | * or if we want to idle in case it has no pending requests. | |
3203 | */ | |
3204 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3205 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3206 | ||
44f7c160 JA |
3207 | if (cfq_cfqq_slice_new(cfqq)) { |
3208 | cfq_set_prio_slice(cfqd, cfqq); | |
3209 | cfq_clear_cfqq_slice_new(cfqq); | |
3210 | } | |
a36e71f9 | 3211 | /* |
8e550632 CZ |
3212 | * Idling is not enabled on: |
3213 | * - expired queues | |
3214 | * - idle-priority queues | |
3215 | * - async queues | |
3216 | * - queues with still some requests queued | |
3217 | * - when there is a close cooperator | |
a36e71f9 | 3218 | */ |
0871714e | 3219 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
6084cdda | 3220 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3221 | else if (sync && cfqq_empty && |
3222 | !cfq_close_cooperator(cfqd, cfqq)) { | |
3223 | cfqd->noidle_tree_requires_idle |= !rq_noidle(rq); | |
3224 | /* | |
3225 | * Idling is enabled for SYNC_WORKLOAD. | |
3226 | * SYNC_NOIDLE_WORKLOAD idles at the end of the tree | |
3227 | * only if we processed at least one !rq_noidle request | |
3228 | */ | |
3229 | if (cfqd->serving_type == SYNC_WORKLOAD | |
3230 | || cfqd->noidle_tree_requires_idle) | |
3231 | cfq_arm_slice_timer(cfqd); | |
3232 | } | |
caaa5f9f | 3233 | } |
6d048f53 | 3234 | |
5ad531db | 3235 | if (!rq_in_driver(cfqd)) |
23e018a1 | 3236 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3237 | } |
3238 | ||
22e2c507 JA |
3239 | /* |
3240 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
3241 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
3242 | */ | |
3243 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 3244 | { |
22e2c507 JA |
3245 | if (has_fs_excl()) { |
3246 | /* | |
3247 | * boost idle prio on transactions that would lock out other | |
3248 | * users of the filesystem | |
3249 | */ | |
3250 | if (cfq_class_idle(cfqq)) | |
3251 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
3252 | if (cfqq->ioprio > IOPRIO_NORM) | |
3253 | cfqq->ioprio = IOPRIO_NORM; | |
3254 | } else { | |
3255 | /* | |
dddb7451 | 3256 | * unboost the queue (if needed) |
22e2c507 | 3257 | */ |
dddb7451 CZ |
3258 | cfqq->ioprio_class = cfqq->org_ioprio_class; |
3259 | cfqq->ioprio = cfqq->org_ioprio; | |
22e2c507 | 3260 | } |
22e2c507 | 3261 | } |
1da177e4 | 3262 | |
89850f7e | 3263 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3264 | { |
1b379d8d | 3265 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3266 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3267 | return ELV_MQUEUE_MUST; |
3b18152c | 3268 | } |
1da177e4 | 3269 | |
22e2c507 | 3270 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3271 | } |
3272 | ||
165125e1 | 3273 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3274 | { |
3275 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3276 | struct task_struct *tsk = current; | |
91fac317 | 3277 | struct cfq_io_context *cic; |
22e2c507 JA |
3278 | struct cfq_queue *cfqq; |
3279 | ||
3280 | /* | |
3281 | * don't force setup of a queue from here, as a call to may_queue | |
3282 | * does not necessarily imply that a request actually will be queued. | |
3283 | * so just lookup a possibly existing queue, or return 'may queue' | |
3284 | * if that fails | |
3285 | */ | |
4ac845a2 | 3286 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3287 | if (!cic) |
3288 | return ELV_MQUEUE_MAY; | |
3289 | ||
b0b78f81 | 3290 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3291 | if (cfqq) { |
fd0928df | 3292 | cfq_init_prio_data(cfqq, cic->ioc); |
22e2c507 JA |
3293 | cfq_prio_boost(cfqq); |
3294 | ||
89850f7e | 3295 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3296 | } |
3297 | ||
3298 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3299 | } |
3300 | ||
1da177e4 LT |
3301 | /* |
3302 | * queue lock held here | |
3303 | */ | |
bb37b94c | 3304 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3305 | { |
5e705374 | 3306 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3307 | |
5e705374 | 3308 | if (cfqq) { |
22e2c507 | 3309 | const int rw = rq_data_dir(rq); |
1da177e4 | 3310 | |
22e2c507 JA |
3311 | BUG_ON(!cfqq->allocated[rw]); |
3312 | cfqq->allocated[rw]--; | |
1da177e4 | 3313 | |
5e705374 | 3314 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 3315 | |
1da177e4 | 3316 | rq->elevator_private = NULL; |
5e705374 | 3317 | rq->elevator_private2 = NULL; |
1da177e4 | 3318 | |
1da177e4 LT |
3319 | cfq_put_queue(cfqq); |
3320 | } | |
3321 | } | |
3322 | ||
df5fe3e8 JM |
3323 | static struct cfq_queue * |
3324 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_context *cic, | |
3325 | struct cfq_queue *cfqq) | |
3326 | { | |
3327 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3328 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3329 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3330 | cfq_put_queue(cfqq); |
3331 | return cic_to_cfqq(cic, 1); | |
3332 | } | |
3333 | ||
e6c5bc73 JM |
3334 | static int should_split_cfqq(struct cfq_queue *cfqq) |
3335 | { | |
3336 | if (cfqq->seeky_start && | |
3337 | time_after(jiffies, cfqq->seeky_start + CFQQ_COOP_TOUT)) | |
3338 | return 1; | |
3339 | return 0; | |
3340 | } | |
3341 | ||
3342 | /* | |
3343 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3344 | * was the last process referring to said cfqq. | |
3345 | */ | |
3346 | static struct cfq_queue * | |
3347 | split_cfqq(struct cfq_io_context *cic, struct cfq_queue *cfqq) | |
3348 | { | |
3349 | if (cfqq_process_refs(cfqq) == 1) { | |
3350 | cfqq->seeky_start = 0; | |
3351 | cfqq->pid = current->pid; | |
3352 | cfq_clear_cfqq_coop(cfqq); | |
3353 | return cfqq; | |
3354 | } | |
3355 | ||
3356 | cic_set_cfqq(cic, NULL, 1); | |
3357 | cfq_put_queue(cfqq); | |
3358 | return NULL; | |
3359 | } | |
1da177e4 | 3360 | /* |
22e2c507 | 3361 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3362 | */ |
22e2c507 | 3363 | static int |
165125e1 | 3364 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
3365 | { |
3366 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3367 | struct cfq_io_context *cic; | |
3368 | const int rw = rq_data_dir(rq); | |
a6151c3a | 3369 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3370 | struct cfq_queue *cfqq; |
1da177e4 LT |
3371 | unsigned long flags; |
3372 | ||
3373 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3374 | ||
e2d74ac0 | 3375 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 3376 | |
1da177e4 LT |
3377 | spin_lock_irqsave(q->queue_lock, flags); |
3378 | ||
22e2c507 JA |
3379 | if (!cic) |
3380 | goto queue_fail; | |
3381 | ||
e6c5bc73 | 3382 | new_queue: |
91fac317 | 3383 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3384 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
fd0928df | 3385 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
91fac317 | 3386 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3387 | } else { |
e6c5bc73 JM |
3388 | /* |
3389 | * If the queue was seeky for too long, break it apart. | |
3390 | */ | |
3391 | if (cfq_cfqq_coop(cfqq) && should_split_cfqq(cfqq)) { | |
3392 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); | |
3393 | cfqq = split_cfqq(cic, cfqq); | |
3394 | if (!cfqq) | |
3395 | goto new_queue; | |
3396 | } | |
3397 | ||
df5fe3e8 JM |
3398 | /* |
3399 | * Check to see if this queue is scheduled to merge with | |
3400 | * another, closely cooperating queue. The merging of | |
3401 | * queues happens here as it must be done in process context. | |
3402 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3403 | */ | |
3404 | if (cfqq->new_cfqq) | |
3405 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3406 | } |
1da177e4 LT |
3407 | |
3408 | cfqq->allocated[rw]++; | |
22e2c507 | 3409 | atomic_inc(&cfqq->ref); |
1da177e4 | 3410 | |
5e705374 | 3411 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 3412 | |
5e705374 JA |
3413 | rq->elevator_private = cic; |
3414 | rq->elevator_private2 = cfqq; | |
3415 | return 0; | |
1da177e4 | 3416 | |
22e2c507 JA |
3417 | queue_fail: |
3418 | if (cic) | |
3419 | put_io_context(cic->ioc); | |
89850f7e | 3420 | |
23e018a1 | 3421 | cfq_schedule_dispatch(cfqd); |
1da177e4 | 3422 | spin_unlock_irqrestore(q->queue_lock, flags); |
7b679138 | 3423 | cfq_log(cfqd, "set_request fail"); |
1da177e4 LT |
3424 | return 1; |
3425 | } | |
3426 | ||
65f27f38 | 3427 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3428 | { |
65f27f38 | 3429 | struct cfq_data *cfqd = |
23e018a1 | 3430 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3431 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3432 | |
40bb54d1 | 3433 | spin_lock_irq(q->queue_lock); |
a7f55792 | 3434 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3435 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3436 | } |
3437 | ||
3438 | /* | |
3439 | * Timer running if the active_queue is currently idling inside its time slice | |
3440 | */ | |
3441 | static void cfq_idle_slice_timer(unsigned long data) | |
3442 | { | |
3443 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3444 | struct cfq_queue *cfqq; | |
3445 | unsigned long flags; | |
3c6bd2f8 | 3446 | int timed_out = 1; |
22e2c507 | 3447 | |
7b679138 JA |
3448 | cfq_log(cfqd, "idle timer fired"); |
3449 | ||
22e2c507 JA |
3450 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3451 | ||
fe094d98 JA |
3452 | cfqq = cfqd->active_queue; |
3453 | if (cfqq) { | |
3c6bd2f8 JA |
3454 | timed_out = 0; |
3455 | ||
b029195d JA |
3456 | /* |
3457 | * We saw a request before the queue expired, let it through | |
3458 | */ | |
3459 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3460 | goto out_kick; | |
3461 | ||
22e2c507 JA |
3462 | /* |
3463 | * expired | |
3464 | */ | |
44f7c160 | 3465 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
3466 | goto expire; |
3467 | ||
3468 | /* | |
3469 | * only expire and reinvoke request handler, if there are | |
3470 | * other queues with pending requests | |
3471 | */ | |
caaa5f9f | 3472 | if (!cfqd->busy_queues) |
22e2c507 | 3473 | goto out_cont; |
22e2c507 JA |
3474 | |
3475 | /* | |
3476 | * not expired and it has a request pending, let it dispatch | |
3477 | */ | |
75e50984 | 3478 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3479 | goto out_kick; |
76280aff CZ |
3480 | |
3481 | /* | |
3482 | * Queue depth flag is reset only when the idle didn't succeed | |
3483 | */ | |
3484 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
3485 | } |
3486 | expire: | |
6084cdda | 3487 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 3488 | out_kick: |
23e018a1 | 3489 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
3490 | out_cont: |
3491 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3492 | } | |
3493 | ||
3b18152c JA |
3494 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3495 | { | |
3496 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 3497 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 3498 | } |
22e2c507 | 3499 | |
c2dea2d1 VT |
3500 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3501 | { | |
3502 | int i; | |
3503 | ||
3504 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
3505 | if (cfqd->async_cfqq[0][i]) | |
3506 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
3507 | if (cfqd->async_cfqq[1][i]) | |
3508 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 3509 | } |
2389d1ef ON |
3510 | |
3511 | if (cfqd->async_idle_cfqq) | |
3512 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
3513 | } |
3514 | ||
b374d18a | 3515 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 3516 | { |
22e2c507 | 3517 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 3518 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3519 | |
3b18152c | 3520 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 3521 | |
d9ff4187 | 3522 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3523 | |
d9ff4187 | 3524 | if (cfqd->active_queue) |
6084cdda | 3525 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
3526 | |
3527 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
3528 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
3529 | struct cfq_io_context, | |
3530 | queue_list); | |
89850f7e JA |
3531 | |
3532 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 3533 | } |
e2d74ac0 | 3534 | |
c2dea2d1 | 3535 | cfq_put_async_queues(cfqd); |
b1c35769 VG |
3536 | cfq_release_cfq_groups(cfqd); |
3537 | blkiocg_del_blkio_group(&cfqd->root_group.blkg); | |
15c31be4 | 3538 | |
d9ff4187 | 3539 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
3540 | |
3541 | cfq_shutdown_timer_wq(cfqd); | |
3542 | ||
b1c35769 VG |
3543 | /* Wait for cfqg->blkg->key accessors to exit their grace periods. */ |
3544 | synchronize_rcu(); | |
a90d742e | 3545 | kfree(cfqd); |
1da177e4 LT |
3546 | } |
3547 | ||
165125e1 | 3548 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
3549 | { |
3550 | struct cfq_data *cfqd; | |
718eee05 | 3551 | int i, j; |
cdb16e8f | 3552 | struct cfq_group *cfqg; |
615f0259 | 3553 | struct cfq_rb_root *st; |
1da177e4 | 3554 | |
94f6030c | 3555 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
1da177e4 | 3556 | if (!cfqd) |
bc1c1169 | 3557 | return NULL; |
1da177e4 | 3558 | |
1fa8f6d6 VG |
3559 | /* Init root service tree */ |
3560 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
3561 | ||
cdb16e8f VG |
3562 | /* Init root group */ |
3563 | cfqg = &cfqd->root_group; | |
615f0259 VG |
3564 | for_each_cfqg_st(cfqg, i, j, st) |
3565 | *st = CFQ_RB_ROOT; | |
1fa8f6d6 | 3566 | RB_CLEAR_NODE(&cfqg->rb_node); |
26a2ac00 | 3567 | |
25bc6b07 VG |
3568 | /* Give preference to root group over other groups */ |
3569 | cfqg->weight = 2*BLKIO_WEIGHT_DEFAULT; | |
3570 | ||
25fb5169 | 3571 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
b1c35769 VG |
3572 | /* |
3573 | * Take a reference to root group which we never drop. This is just | |
3574 | * to make sure that cfq_put_cfqg() does not try to kfree root group | |
3575 | */ | |
3576 | atomic_set(&cfqg->ref, 1); | |
22084190 VG |
3577 | blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg, (void *)cfqd, |
3578 | 0); | |
25fb5169 | 3579 | #endif |
26a2ac00 JA |
3580 | /* |
3581 | * Not strictly needed (since RB_ROOT just clears the node and we | |
3582 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
3583 | * to add magic to the rb code | |
3584 | */ | |
3585 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
3586 | cfqd->prio_trees[i] = RB_ROOT; | |
3587 | ||
6118b70b JA |
3588 | /* |
3589 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
3590 | * Grab a permanent reference to it, so that the normal code flow | |
3591 | * will not attempt to free it. | |
3592 | */ | |
3593 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
3594 | atomic_inc(&cfqd->oom_cfqq.ref); | |
cdb16e8f | 3595 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group); |
6118b70b | 3596 | |
d9ff4187 | 3597 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 3598 | |
1da177e4 | 3599 | cfqd->queue = q; |
1da177e4 | 3600 | |
22e2c507 JA |
3601 | init_timer(&cfqd->idle_slice_timer); |
3602 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
3603 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
3604 | ||
23e018a1 | 3605 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 3606 | |
1da177e4 | 3607 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
3608 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
3609 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
3610 | cfqd->cfq_back_max = cfq_back_max; |
3611 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
3612 | cfqd->cfq_slice[0] = cfq_slice_async; |
3613 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
3614 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
3615 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
963b72fc | 3616 | cfqd->cfq_latency = 1; |
e459dd08 | 3617 | cfqd->hw_tag = -1; |
365722bb | 3618 | cfqd->last_end_sync_rq = jiffies; |
bc1c1169 | 3619 | return cfqd; |
1da177e4 LT |
3620 | } |
3621 | ||
3622 | static void cfq_slab_kill(void) | |
3623 | { | |
d6de8be7 JA |
3624 | /* |
3625 | * Caller already ensured that pending RCU callbacks are completed, | |
3626 | * so we should have no busy allocations at this point. | |
3627 | */ | |
1da177e4 LT |
3628 | if (cfq_pool) |
3629 | kmem_cache_destroy(cfq_pool); | |
3630 | if (cfq_ioc_pool) | |
3631 | kmem_cache_destroy(cfq_ioc_pool); | |
3632 | } | |
3633 | ||
3634 | static int __init cfq_slab_setup(void) | |
3635 | { | |
0a31bd5f | 3636 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
1da177e4 LT |
3637 | if (!cfq_pool) |
3638 | goto fail; | |
3639 | ||
34e6bbf2 | 3640 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
1da177e4 LT |
3641 | if (!cfq_ioc_pool) |
3642 | goto fail; | |
3643 | ||
3644 | return 0; | |
3645 | fail: | |
3646 | cfq_slab_kill(); | |
3647 | return -ENOMEM; | |
3648 | } | |
3649 | ||
1da177e4 LT |
3650 | /* |
3651 | * sysfs parts below --> | |
3652 | */ | |
1da177e4 LT |
3653 | static ssize_t |
3654 | cfq_var_show(unsigned int var, char *page) | |
3655 | { | |
3656 | return sprintf(page, "%d\n", var); | |
3657 | } | |
3658 | ||
3659 | static ssize_t | |
3660 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
3661 | { | |
3662 | char *p = (char *) page; | |
3663 | ||
3664 | *var = simple_strtoul(p, &p, 10); | |
3665 | return count; | |
3666 | } | |
3667 | ||
1da177e4 | 3668 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 3669 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 3670 | { \ |
3d1ab40f | 3671 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3672 | unsigned int __data = __VAR; \ |
3673 | if (__CONV) \ | |
3674 | __data = jiffies_to_msecs(__data); \ | |
3675 | return cfq_var_show(__data, (page)); \ | |
3676 | } | |
3677 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
3678 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
3679 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
3680 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
3681 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
3682 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
3683 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
3684 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
3685 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 3686 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
1da177e4 LT |
3687 | #undef SHOW_FUNCTION |
3688 | ||
3689 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 3690 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 3691 | { \ |
3d1ab40f | 3692 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3693 | unsigned int __data; \ |
3694 | int ret = cfq_var_store(&__data, (page), count); \ | |
3695 | if (__data < (MIN)) \ | |
3696 | __data = (MIN); \ | |
3697 | else if (__data > (MAX)) \ | |
3698 | __data = (MAX); \ | |
3699 | if (__CONV) \ | |
3700 | *(__PTR) = msecs_to_jiffies(__data); \ | |
3701 | else \ | |
3702 | *(__PTR) = __data; \ | |
3703 | return ret; \ | |
3704 | } | |
3705 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
3706 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
3707 | UINT_MAX, 1); | |
3708 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
3709 | UINT_MAX, 1); | |
e572ec7e | 3710 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
3711 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
3712 | UINT_MAX, 0); | |
22e2c507 JA |
3713 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
3714 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
3715 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
3716 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
3717 | UINT_MAX, 0); | |
963b72fc | 3718 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
1da177e4 LT |
3719 | #undef STORE_FUNCTION |
3720 | ||
e572ec7e AV |
3721 | #define CFQ_ATTR(name) \ |
3722 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
3723 | ||
3724 | static struct elv_fs_entry cfq_attrs[] = { | |
3725 | CFQ_ATTR(quantum), | |
e572ec7e AV |
3726 | CFQ_ATTR(fifo_expire_sync), |
3727 | CFQ_ATTR(fifo_expire_async), | |
3728 | CFQ_ATTR(back_seek_max), | |
3729 | CFQ_ATTR(back_seek_penalty), | |
3730 | CFQ_ATTR(slice_sync), | |
3731 | CFQ_ATTR(slice_async), | |
3732 | CFQ_ATTR(slice_async_rq), | |
3733 | CFQ_ATTR(slice_idle), | |
963b72fc | 3734 | CFQ_ATTR(low_latency), |
e572ec7e | 3735 | __ATTR_NULL |
1da177e4 LT |
3736 | }; |
3737 | ||
1da177e4 LT |
3738 | static struct elevator_type iosched_cfq = { |
3739 | .ops = { | |
3740 | .elevator_merge_fn = cfq_merge, | |
3741 | .elevator_merged_fn = cfq_merged_request, | |
3742 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 3743 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 3744 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 3745 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 3746 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
3747 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
3748 | .elevator_queue_empty_fn = cfq_queue_empty, | |
3749 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
3750 | .elevator_former_req_fn = elv_rb_former_request, |
3751 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
3752 | .elevator_set_req_fn = cfq_set_request, |
3753 | .elevator_put_req_fn = cfq_put_request, | |
3754 | .elevator_may_queue_fn = cfq_may_queue, | |
3755 | .elevator_init_fn = cfq_init_queue, | |
3756 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 3757 | .trim = cfq_free_io_context, |
1da177e4 | 3758 | }, |
3d1ab40f | 3759 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
3760 | .elevator_name = "cfq", |
3761 | .elevator_owner = THIS_MODULE, | |
3762 | }; | |
3763 | ||
3764 | static int __init cfq_init(void) | |
3765 | { | |
22e2c507 JA |
3766 | /* |
3767 | * could be 0 on HZ < 1000 setups | |
3768 | */ | |
3769 | if (!cfq_slice_async) | |
3770 | cfq_slice_async = 1; | |
3771 | if (!cfq_slice_idle) | |
3772 | cfq_slice_idle = 1; | |
3773 | ||
1da177e4 LT |
3774 | if (cfq_slab_setup()) |
3775 | return -ENOMEM; | |
3776 | ||
2fdd82bd | 3777 | elv_register(&iosched_cfq); |
1da177e4 | 3778 | |
2fdd82bd | 3779 | return 0; |
1da177e4 LT |
3780 | } |
3781 | ||
3782 | static void __exit cfq_exit(void) | |
3783 | { | |
6e9a4738 | 3784 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 3785 | elv_unregister(&iosched_cfq); |
334e94de | 3786 | ioc_gone = &all_gone; |
fba82272 OH |
3787 | /* ioc_gone's update must be visible before reading ioc_count */ |
3788 | smp_wmb(); | |
d6de8be7 JA |
3789 | |
3790 | /* | |
3791 | * this also protects us from entering cfq_slab_kill() with | |
3792 | * pending RCU callbacks | |
3793 | */ | |
245b2e70 | 3794 | if (elv_ioc_count_read(cfq_ioc_count)) |
9a11b4ed | 3795 | wait_for_completion(&all_gone); |
83521d3e | 3796 | cfq_slab_kill(); |
1da177e4 LT |
3797 | } |
3798 | ||
3799 | module_init(cfq_init); | |
3800 | module_exit(cfq_exit); | |
3801 | ||
3802 | MODULE_AUTHOR("Jens Axboe"); | |
3803 | MODULE_LICENSE("GPL"); | |
3804 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |