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; | |
f75edf2d | 120 | unsigned int allocated_slice; |
dae739eb VG |
121 | /* time when first request from queue completed and slice started. */ |
122 | unsigned long slice_start; | |
6118b70b JA |
123 | unsigned long slice_end; |
124 | long slice_resid; | |
125 | unsigned int slice_dispatch; | |
126 | ||
127 | /* pending metadata requests */ | |
128 | int meta_pending; | |
129 | /* number of requests that are on the dispatch list or inside driver */ | |
130 | int dispatched; | |
131 | ||
132 | /* io prio of this group */ | |
133 | unsigned short ioprio, org_ioprio; | |
134 | unsigned short ioprio_class, org_ioprio_class; | |
135 | ||
b2c18e1e JM |
136 | unsigned int seek_samples; |
137 | u64 seek_total; | |
138 | sector_t seek_mean; | |
139 | sector_t last_request_pos; | |
e6c5bc73 | 140 | unsigned long seeky_start; |
b2c18e1e | 141 | |
6118b70b | 142 | pid_t pid; |
df5fe3e8 | 143 | |
aa6f6a3d | 144 | struct cfq_rb_root *service_tree; |
df5fe3e8 | 145 | struct cfq_queue *new_cfqq; |
cdb16e8f | 146 | struct cfq_group *cfqg; |
22084190 VG |
147 | /* Sectors dispatched in current dispatch round */ |
148 | unsigned long nr_sectors; | |
6118b70b JA |
149 | }; |
150 | ||
c0324a02 | 151 | /* |
718eee05 | 152 | * First index in the service_trees. |
c0324a02 CZ |
153 | * IDLE is handled separately, so it has negative index |
154 | */ | |
155 | enum wl_prio_t { | |
c0324a02 | 156 | BE_WORKLOAD = 0, |
615f0259 VG |
157 | RT_WORKLOAD = 1, |
158 | IDLE_WORKLOAD = 2, | |
c0324a02 CZ |
159 | }; |
160 | ||
718eee05 CZ |
161 | /* |
162 | * Second index in the service_trees. | |
163 | */ | |
164 | enum wl_type_t { | |
165 | ASYNC_WORKLOAD = 0, | |
166 | SYNC_NOIDLE_WORKLOAD = 1, | |
167 | SYNC_WORKLOAD = 2 | |
168 | }; | |
169 | ||
cdb16e8f VG |
170 | /* This is per cgroup per device grouping structure */ |
171 | struct cfq_group { | |
1fa8f6d6 VG |
172 | /* group service_tree member */ |
173 | struct rb_node rb_node; | |
174 | ||
175 | /* group service_tree key */ | |
176 | u64 vdisktime; | |
25bc6b07 | 177 | unsigned int weight; |
1fa8f6d6 VG |
178 | bool on_st; |
179 | ||
180 | /* number of cfqq currently on this group */ | |
181 | int nr_cfqq; | |
182 | ||
58ff82f3 VG |
183 | /* Per group busy queus average. Useful for workload slice calc. */ |
184 | unsigned int busy_queues_avg[2]; | |
cdb16e8f VG |
185 | /* |
186 | * rr lists of queues with requests, onle rr for each priority class. | |
187 | * Counts are embedded in the cfq_rb_root | |
188 | */ | |
189 | struct cfq_rb_root service_trees[2][3]; | |
190 | struct cfq_rb_root service_tree_idle; | |
dae739eb VG |
191 | |
192 | unsigned long saved_workload_slice; | |
193 | enum wl_type_t saved_workload; | |
194 | enum wl_prio_t saved_serving_prio; | |
25fb5169 VG |
195 | struct blkio_group blkg; |
196 | #ifdef CONFIG_CFQ_GROUP_IOSCHED | |
197 | struct hlist_node cfqd_node; | |
b1c35769 | 198 | atomic_t ref; |
25fb5169 | 199 | #endif |
cdb16e8f | 200 | }; |
718eee05 | 201 | |
22e2c507 JA |
202 | /* |
203 | * Per block device queue structure | |
204 | */ | |
1da177e4 | 205 | struct cfq_data { |
165125e1 | 206 | struct request_queue *queue; |
1fa8f6d6 VG |
207 | /* Root service tree for cfq_groups */ |
208 | struct cfq_rb_root grp_service_tree; | |
cdb16e8f | 209 | struct cfq_group root_group; |
58ff82f3 VG |
210 | /* Number of active cfq groups on group service tree */ |
211 | int nr_groups; | |
22e2c507 | 212 | |
c0324a02 CZ |
213 | /* |
214 | * The priority currently being served | |
22e2c507 | 215 | */ |
c0324a02 | 216 | enum wl_prio_t serving_prio; |
718eee05 CZ |
217 | enum wl_type_t serving_type; |
218 | unsigned long workload_expires; | |
cdb16e8f | 219 | struct cfq_group *serving_group; |
8e550632 | 220 | bool noidle_tree_requires_idle; |
a36e71f9 JA |
221 | |
222 | /* | |
223 | * Each priority tree is sorted by next_request position. These | |
224 | * trees are used when determining if two or more queues are | |
225 | * interleaving requests (see cfq_close_cooperator). | |
226 | */ | |
227 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; | |
228 | ||
22e2c507 JA |
229 | unsigned int busy_queues; |
230 | ||
5ad531db | 231 | int rq_in_driver[2]; |
3ed9a296 | 232 | int sync_flight; |
45333d5a AC |
233 | |
234 | /* | |
235 | * queue-depth detection | |
236 | */ | |
237 | int rq_queued; | |
25776e35 | 238 | int hw_tag; |
e459dd08 CZ |
239 | /* |
240 | * hw_tag can be | |
241 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) | |
242 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) | |
243 | * 0 => no NCQ | |
244 | */ | |
245 | int hw_tag_est_depth; | |
246 | unsigned int hw_tag_samples; | |
1da177e4 | 247 | |
22e2c507 JA |
248 | /* |
249 | * idle window management | |
250 | */ | |
251 | struct timer_list idle_slice_timer; | |
23e018a1 | 252 | struct work_struct unplug_work; |
1da177e4 | 253 | |
22e2c507 JA |
254 | struct cfq_queue *active_queue; |
255 | struct cfq_io_context *active_cic; | |
22e2c507 | 256 | |
c2dea2d1 VT |
257 | /* |
258 | * async queue for each priority case | |
259 | */ | |
260 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; | |
261 | struct cfq_queue *async_idle_cfqq; | |
15c31be4 | 262 | |
6d048f53 | 263 | sector_t last_position; |
1da177e4 | 264 | |
1da177e4 LT |
265 | /* |
266 | * tunables, see top of file | |
267 | */ | |
268 | unsigned int cfq_quantum; | |
22e2c507 | 269 | unsigned int cfq_fifo_expire[2]; |
1da177e4 LT |
270 | unsigned int cfq_back_penalty; |
271 | unsigned int cfq_back_max; | |
22e2c507 JA |
272 | unsigned int cfq_slice[2]; |
273 | unsigned int cfq_slice_async_rq; | |
274 | unsigned int cfq_slice_idle; | |
963b72fc | 275 | unsigned int cfq_latency; |
d9ff4187 AV |
276 | |
277 | struct list_head cic_list; | |
1da177e4 | 278 | |
6118b70b JA |
279 | /* |
280 | * Fallback dummy cfqq for extreme OOM conditions | |
281 | */ | |
282 | struct cfq_queue oom_cfqq; | |
365722bb VG |
283 | |
284 | unsigned long last_end_sync_rq; | |
25fb5169 VG |
285 | |
286 | /* List of cfq groups being managed on this device*/ | |
287 | struct hlist_head cfqg_list; | |
1da177e4 LT |
288 | }; |
289 | ||
25fb5169 VG |
290 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
291 | ||
cdb16e8f VG |
292 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
293 | enum wl_prio_t prio, | |
718eee05 | 294 | enum wl_type_t type, |
c0324a02 CZ |
295 | struct cfq_data *cfqd) |
296 | { | |
1fa8f6d6 VG |
297 | if (!cfqg) |
298 | return NULL; | |
299 | ||
c0324a02 | 300 | if (prio == IDLE_WORKLOAD) |
cdb16e8f | 301 | return &cfqg->service_tree_idle; |
c0324a02 | 302 | |
cdb16e8f | 303 | return &cfqg->service_trees[prio][type]; |
c0324a02 CZ |
304 | } |
305 | ||
3b18152c | 306 | enum cfqq_state_flags { |
b0b8d749 JA |
307 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
308 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ | |
b029195d | 309 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
b0b8d749 | 310 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
b0b8d749 JA |
311 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
312 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ | |
313 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ | |
44f7c160 | 314 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
91fac317 | 315 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
b3b6d040 | 316 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
76280aff | 317 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
f75edf2d VG |
318 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
319 | CFQ_CFQQ_FLAG_wait_busy_done, /* Got new request. Expire the queue */ | |
3b18152c JA |
320 | }; |
321 | ||
322 | #define CFQ_CFQQ_FNS(name) \ | |
323 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ | |
324 | { \ | |
fe094d98 | 325 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
326 | } \ |
327 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ | |
328 | { \ | |
fe094d98 | 329 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
3b18152c JA |
330 | } \ |
331 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ | |
332 | { \ | |
fe094d98 | 333 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
3b18152c JA |
334 | } |
335 | ||
336 | CFQ_CFQQ_FNS(on_rr); | |
337 | CFQ_CFQQ_FNS(wait_request); | |
b029195d | 338 | CFQ_CFQQ_FNS(must_dispatch); |
3b18152c | 339 | CFQ_CFQQ_FNS(must_alloc_slice); |
3b18152c JA |
340 | CFQ_CFQQ_FNS(fifo_expire); |
341 | CFQ_CFQQ_FNS(idle_window); | |
342 | CFQ_CFQQ_FNS(prio_changed); | |
44f7c160 | 343 | CFQ_CFQQ_FNS(slice_new); |
91fac317 | 344 | CFQ_CFQQ_FNS(sync); |
a36e71f9 | 345 | CFQ_CFQQ_FNS(coop); |
76280aff | 346 | CFQ_CFQQ_FNS(deep); |
f75edf2d VG |
347 | CFQ_CFQQ_FNS(wait_busy); |
348 | CFQ_CFQQ_FNS(wait_busy_done); | |
3b18152c JA |
349 | #undef CFQ_CFQQ_FNS |
350 | ||
2868ef7b VG |
351 | #ifdef CONFIG_DEBUG_CFQ_IOSCHED |
352 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ | |
353 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ | |
354 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ | |
355 | blkg_path(&(cfqq)->cfqg->blkg), ##args); | |
356 | ||
357 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ | |
358 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ | |
359 | blkg_path(&(cfqg)->blkg), ##args); \ | |
360 | ||
361 | #else | |
7b679138 JA |
362 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
363 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) | |
2868ef7b VG |
364 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0); |
365 | #endif | |
7b679138 JA |
366 | #define cfq_log(cfqd, fmt, args...) \ |
367 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) | |
368 | ||
615f0259 VG |
369 | /* Traverses through cfq group service trees */ |
370 | #define for_each_cfqg_st(cfqg, i, j, st) \ | |
371 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ | |
372 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ | |
373 | : &cfqg->service_tree_idle; \ | |
374 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ | |
375 | (i == IDLE_WORKLOAD && j == 0); \ | |
376 | j++, st = i < IDLE_WORKLOAD ? \ | |
377 | &cfqg->service_trees[i][j]: NULL) \ | |
378 | ||
379 | ||
c0324a02 CZ |
380 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
381 | { | |
382 | if (cfq_class_idle(cfqq)) | |
383 | return IDLE_WORKLOAD; | |
384 | if (cfq_class_rt(cfqq)) | |
385 | return RT_WORKLOAD; | |
386 | return BE_WORKLOAD; | |
387 | } | |
388 | ||
718eee05 CZ |
389 | |
390 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) | |
391 | { | |
392 | if (!cfq_cfqq_sync(cfqq)) | |
393 | return ASYNC_WORKLOAD; | |
394 | if (!cfq_cfqq_idle_window(cfqq)) | |
395 | return SYNC_NOIDLE_WORKLOAD; | |
396 | return SYNC_WORKLOAD; | |
397 | } | |
398 | ||
58ff82f3 VG |
399 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
400 | struct cfq_data *cfqd, | |
401 | struct cfq_group *cfqg) | |
c0324a02 CZ |
402 | { |
403 | if (wl == IDLE_WORKLOAD) | |
cdb16e8f | 404 | return cfqg->service_tree_idle.count; |
c0324a02 | 405 | |
cdb16e8f VG |
406 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
407 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count | |
408 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; | |
c0324a02 CZ |
409 | } |
410 | ||
165125e1 | 411 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
a6151c3a | 412 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool, |
fd0928df | 413 | struct io_context *, gfp_t); |
4ac845a2 | 414 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
91fac317 VT |
415 | struct io_context *); |
416 | ||
5ad531db JA |
417 | static inline int rq_in_driver(struct cfq_data *cfqd) |
418 | { | |
419 | return cfqd->rq_in_driver[0] + cfqd->rq_in_driver[1]; | |
420 | } | |
421 | ||
91fac317 | 422 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, |
a6151c3a | 423 | bool is_sync) |
91fac317 | 424 | { |
a6151c3a | 425 | return cic->cfqq[is_sync]; |
91fac317 VT |
426 | } |
427 | ||
428 | static inline void cic_set_cfqq(struct cfq_io_context *cic, | |
a6151c3a | 429 | struct cfq_queue *cfqq, bool is_sync) |
91fac317 | 430 | { |
a6151c3a | 431 | cic->cfqq[is_sync] = cfqq; |
91fac317 VT |
432 | } |
433 | ||
434 | /* | |
435 | * We regard a request as SYNC, if it's either a read or has the SYNC bit | |
436 | * set (in which case it could also be direct WRITE). | |
437 | */ | |
a6151c3a | 438 | static inline bool cfq_bio_sync(struct bio *bio) |
91fac317 | 439 | { |
a6151c3a | 440 | return bio_data_dir(bio) == READ || bio_rw_flagged(bio, BIO_RW_SYNCIO); |
91fac317 | 441 | } |
1da177e4 | 442 | |
99f95e52 AM |
443 | /* |
444 | * scheduler run of queue, if there are requests pending and no one in the | |
445 | * driver that will restart queueing | |
446 | */ | |
23e018a1 | 447 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
99f95e52 | 448 | { |
7b679138 JA |
449 | if (cfqd->busy_queues) { |
450 | cfq_log(cfqd, "schedule dispatch"); | |
23e018a1 | 451 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
7b679138 | 452 | } |
99f95e52 AM |
453 | } |
454 | ||
165125e1 | 455 | static int cfq_queue_empty(struct request_queue *q) |
99f95e52 AM |
456 | { |
457 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
458 | ||
f04a6424 | 459 | return !cfqd->rq_queued; |
99f95e52 AM |
460 | } |
461 | ||
44f7c160 JA |
462 | /* |
463 | * Scale schedule slice based on io priority. Use the sync time slice only | |
464 | * if a queue is marked sync and has sync io queued. A sync queue with async | |
465 | * io only, should not get full sync slice length. | |
466 | */ | |
a6151c3a | 467 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
d9e7620e | 468 | unsigned short prio) |
44f7c160 | 469 | { |
d9e7620e | 470 | const int base_slice = cfqd->cfq_slice[sync]; |
44f7c160 | 471 | |
d9e7620e JA |
472 | WARN_ON(prio >= IOPRIO_BE_NR); |
473 | ||
474 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); | |
475 | } | |
44f7c160 | 476 | |
d9e7620e JA |
477 | static inline int |
478 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
479 | { | |
480 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); | |
44f7c160 JA |
481 | } |
482 | ||
25bc6b07 VG |
483 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
484 | { | |
485 | u64 d = delta << CFQ_SERVICE_SHIFT; | |
486 | ||
487 | d = d * BLKIO_WEIGHT_DEFAULT; | |
488 | do_div(d, cfqg->weight); | |
489 | return d; | |
490 | } | |
491 | ||
492 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
493 | { | |
494 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
495 | if (delta > 0) | |
496 | min_vdisktime = vdisktime; | |
497 | ||
498 | return min_vdisktime; | |
499 | } | |
500 | ||
501 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) | |
502 | { | |
503 | s64 delta = (s64)(vdisktime - min_vdisktime); | |
504 | if (delta < 0) | |
505 | min_vdisktime = vdisktime; | |
506 | ||
507 | return min_vdisktime; | |
508 | } | |
509 | ||
510 | static void update_min_vdisktime(struct cfq_rb_root *st) | |
511 | { | |
512 | u64 vdisktime = st->min_vdisktime; | |
513 | struct cfq_group *cfqg; | |
514 | ||
515 | if (st->active) { | |
516 | cfqg = rb_entry_cfqg(st->active); | |
517 | vdisktime = cfqg->vdisktime; | |
518 | } | |
519 | ||
520 | if (st->left) { | |
521 | cfqg = rb_entry_cfqg(st->left); | |
522 | vdisktime = min_vdisktime(vdisktime, cfqg->vdisktime); | |
523 | } | |
524 | ||
525 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, vdisktime); | |
526 | } | |
527 | ||
5db5d642 CZ |
528 | /* |
529 | * get averaged number of queues of RT/BE priority. | |
530 | * average is updated, with a formula that gives more weight to higher numbers, | |
531 | * to quickly follows sudden increases and decrease slowly | |
532 | */ | |
533 | ||
58ff82f3 VG |
534 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
535 | struct cfq_group *cfqg, bool rt) | |
5869619c | 536 | { |
5db5d642 CZ |
537 | unsigned min_q, max_q; |
538 | unsigned mult = cfq_hist_divisor - 1; | |
539 | unsigned round = cfq_hist_divisor / 2; | |
58ff82f3 | 540 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
5db5d642 | 541 | |
58ff82f3 VG |
542 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
543 | max_q = max(cfqg->busy_queues_avg[rt], busy); | |
544 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / | |
5db5d642 | 545 | cfq_hist_divisor; |
58ff82f3 VG |
546 | return cfqg->busy_queues_avg[rt]; |
547 | } | |
548 | ||
549 | static inline unsigned | |
550 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
551 | { | |
552 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
553 | ||
554 | return cfq_target_latency * cfqg->weight / st->total_weight; | |
5db5d642 CZ |
555 | } |
556 | ||
44f7c160 JA |
557 | static inline void |
558 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
559 | { | |
5db5d642 CZ |
560 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
561 | if (cfqd->cfq_latency) { | |
58ff82f3 VG |
562 | /* |
563 | * interested queues (we consider only the ones with the same | |
564 | * priority class in the cfq group) | |
565 | */ | |
566 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, | |
567 | cfq_class_rt(cfqq)); | |
5db5d642 CZ |
568 | unsigned sync_slice = cfqd->cfq_slice[1]; |
569 | unsigned expect_latency = sync_slice * iq; | |
58ff82f3 VG |
570 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
571 | ||
572 | if (expect_latency > group_slice) { | |
5db5d642 CZ |
573 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
574 | /* scale low_slice according to IO priority | |
575 | * and sync vs async */ | |
576 | unsigned low_slice = | |
577 | min(slice, base_low_slice * slice / sync_slice); | |
578 | /* the adapted slice value is scaled to fit all iqs | |
579 | * into the target latency */ | |
58ff82f3 | 580 | slice = max(slice * group_slice / expect_latency, |
5db5d642 CZ |
581 | low_slice); |
582 | } | |
583 | } | |
dae739eb | 584 | cfqq->slice_start = jiffies; |
5db5d642 | 585 | cfqq->slice_end = jiffies + slice; |
f75edf2d | 586 | cfqq->allocated_slice = slice; |
7b679138 | 587 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
44f7c160 JA |
588 | } |
589 | ||
590 | /* | |
591 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end | |
592 | * isn't valid until the first request from the dispatch is activated | |
593 | * and the slice time set. | |
594 | */ | |
a6151c3a | 595 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
44f7c160 JA |
596 | { |
597 | if (cfq_cfqq_slice_new(cfqq)) | |
598 | return 0; | |
599 | if (time_before(jiffies, cfqq->slice_end)) | |
600 | return 0; | |
601 | ||
602 | return 1; | |
603 | } | |
604 | ||
1da177e4 | 605 | /* |
5e705374 | 606 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
1da177e4 | 607 | * We choose the request that is closest to the head right now. Distance |
e8a99053 | 608 | * behind the head is penalized and only allowed to a certain extent. |
1da177e4 | 609 | */ |
5e705374 | 610 | static struct request * |
cf7c25cf | 611 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
1da177e4 | 612 | { |
cf7c25cf | 613 | sector_t s1, s2, d1 = 0, d2 = 0; |
1da177e4 | 614 | unsigned long back_max; |
e8a99053 AM |
615 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
616 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
617 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ | |
1da177e4 | 618 | |
5e705374 JA |
619 | if (rq1 == NULL || rq1 == rq2) |
620 | return rq2; | |
621 | if (rq2 == NULL) | |
622 | return rq1; | |
9c2c38a1 | 623 | |
5e705374 JA |
624 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) |
625 | return rq1; | |
626 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
627 | return rq2; | |
374f84ac JA |
628 | if (rq_is_meta(rq1) && !rq_is_meta(rq2)) |
629 | return rq1; | |
630 | else if (rq_is_meta(rq2) && !rq_is_meta(rq1)) | |
631 | return rq2; | |
1da177e4 | 632 | |
83096ebf TH |
633 | s1 = blk_rq_pos(rq1); |
634 | s2 = blk_rq_pos(rq2); | |
1da177e4 | 635 | |
1da177e4 LT |
636 | /* |
637 | * by definition, 1KiB is 2 sectors | |
638 | */ | |
639 | back_max = cfqd->cfq_back_max * 2; | |
640 | ||
641 | /* | |
642 | * Strict one way elevator _except_ in the case where we allow | |
643 | * short backward seeks which are biased as twice the cost of a | |
644 | * similar forward seek. | |
645 | */ | |
646 | if (s1 >= last) | |
647 | d1 = s1 - last; | |
648 | else if (s1 + back_max >= last) | |
649 | d1 = (last - s1) * cfqd->cfq_back_penalty; | |
650 | else | |
e8a99053 | 651 | wrap |= CFQ_RQ1_WRAP; |
1da177e4 LT |
652 | |
653 | if (s2 >= last) | |
654 | d2 = s2 - last; | |
655 | else if (s2 + back_max >= last) | |
656 | d2 = (last - s2) * cfqd->cfq_back_penalty; | |
657 | else | |
e8a99053 | 658 | wrap |= CFQ_RQ2_WRAP; |
1da177e4 LT |
659 | |
660 | /* Found required data */ | |
e8a99053 AM |
661 | |
662 | /* | |
663 | * By doing switch() on the bit mask "wrap" we avoid having to | |
664 | * check two variables for all permutations: --> faster! | |
665 | */ | |
666 | switch (wrap) { | |
5e705374 | 667 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
e8a99053 | 668 | if (d1 < d2) |
5e705374 | 669 | return rq1; |
e8a99053 | 670 | else if (d2 < d1) |
5e705374 | 671 | return rq2; |
e8a99053 AM |
672 | else { |
673 | if (s1 >= s2) | |
5e705374 | 674 | return rq1; |
e8a99053 | 675 | else |
5e705374 | 676 | return rq2; |
e8a99053 | 677 | } |
1da177e4 | 678 | |
e8a99053 | 679 | case CFQ_RQ2_WRAP: |
5e705374 | 680 | return rq1; |
e8a99053 | 681 | case CFQ_RQ1_WRAP: |
5e705374 JA |
682 | return rq2; |
683 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ | |
e8a99053 AM |
684 | default: |
685 | /* | |
686 | * Since both rqs are wrapped, | |
687 | * start with the one that's further behind head | |
688 | * (--> only *one* back seek required), | |
689 | * since back seek takes more time than forward. | |
690 | */ | |
691 | if (s1 <= s2) | |
5e705374 | 692 | return rq1; |
1da177e4 | 693 | else |
5e705374 | 694 | return rq2; |
1da177e4 LT |
695 | } |
696 | } | |
697 | ||
498d3aa2 JA |
698 | /* |
699 | * The below is leftmost cache rbtree addon | |
700 | */ | |
0871714e | 701 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
cc09e299 | 702 | { |
615f0259 VG |
703 | /* Service tree is empty */ |
704 | if (!root->count) | |
705 | return NULL; | |
706 | ||
cc09e299 JA |
707 | if (!root->left) |
708 | root->left = rb_first(&root->rb); | |
709 | ||
0871714e JA |
710 | if (root->left) |
711 | return rb_entry(root->left, struct cfq_queue, rb_node); | |
712 | ||
713 | return NULL; | |
cc09e299 JA |
714 | } |
715 | ||
1fa8f6d6 VG |
716 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
717 | { | |
718 | if (!root->left) | |
719 | root->left = rb_first(&root->rb); | |
720 | ||
721 | if (root->left) | |
722 | return rb_entry_cfqg(root->left); | |
723 | ||
724 | return NULL; | |
725 | } | |
726 | ||
a36e71f9 JA |
727 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
728 | { | |
729 | rb_erase(n, root); | |
730 | RB_CLEAR_NODE(n); | |
731 | } | |
732 | ||
cc09e299 JA |
733 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
734 | { | |
735 | if (root->left == n) | |
736 | root->left = NULL; | |
a36e71f9 | 737 | rb_erase_init(n, &root->rb); |
aa6f6a3d | 738 | --root->count; |
cc09e299 JA |
739 | } |
740 | ||
1da177e4 LT |
741 | /* |
742 | * would be nice to take fifo expire time into account as well | |
743 | */ | |
5e705374 JA |
744 | static struct request * |
745 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
746 | struct request *last) | |
1da177e4 | 747 | { |
21183b07 JA |
748 | struct rb_node *rbnext = rb_next(&last->rb_node); |
749 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
5e705374 | 750 | struct request *next = NULL, *prev = NULL; |
1da177e4 | 751 | |
21183b07 | 752 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
1da177e4 LT |
753 | |
754 | if (rbprev) | |
5e705374 | 755 | prev = rb_entry_rq(rbprev); |
1da177e4 | 756 | |
21183b07 | 757 | if (rbnext) |
5e705374 | 758 | next = rb_entry_rq(rbnext); |
21183b07 JA |
759 | else { |
760 | rbnext = rb_first(&cfqq->sort_list); | |
761 | if (rbnext && rbnext != &last->rb_node) | |
5e705374 | 762 | next = rb_entry_rq(rbnext); |
21183b07 | 763 | } |
1da177e4 | 764 | |
cf7c25cf | 765 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
1da177e4 LT |
766 | } |
767 | ||
d9e7620e JA |
768 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
769 | struct cfq_queue *cfqq) | |
1da177e4 | 770 | { |
d9e7620e JA |
771 | /* |
772 | * just an approximation, should be ok. | |
773 | */ | |
cdb16e8f | 774 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
464191c6 | 775 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
d9e7620e JA |
776 | } |
777 | ||
1fa8f6d6 VG |
778 | static inline s64 |
779 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
780 | { | |
781 | return cfqg->vdisktime - st->min_vdisktime; | |
782 | } | |
783 | ||
784 | static void | |
785 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) | |
786 | { | |
787 | struct rb_node **node = &st->rb.rb_node; | |
788 | struct rb_node *parent = NULL; | |
789 | struct cfq_group *__cfqg; | |
790 | s64 key = cfqg_key(st, cfqg); | |
791 | int left = 1; | |
792 | ||
793 | while (*node != NULL) { | |
794 | parent = *node; | |
795 | __cfqg = rb_entry_cfqg(parent); | |
796 | ||
797 | if (key < cfqg_key(st, __cfqg)) | |
798 | node = &parent->rb_left; | |
799 | else { | |
800 | node = &parent->rb_right; | |
801 | left = 0; | |
802 | } | |
803 | } | |
804 | ||
805 | if (left) | |
806 | st->left = &cfqg->rb_node; | |
807 | ||
808 | rb_link_node(&cfqg->rb_node, parent, node); | |
809 | rb_insert_color(&cfqg->rb_node, &st->rb); | |
810 | } | |
811 | ||
812 | static void | |
813 | cfq_group_service_tree_add(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
814 | { | |
815 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
816 | struct cfq_group *__cfqg; | |
817 | struct rb_node *n; | |
818 | ||
819 | cfqg->nr_cfqq++; | |
820 | if (cfqg->on_st) | |
821 | return; | |
822 | ||
823 | /* | |
824 | * Currently put the group at the end. Later implement something | |
825 | * so that groups get lesser vtime based on their weights, so that | |
826 | * if group does not loose all if it was not continously backlogged. | |
827 | */ | |
828 | n = rb_last(&st->rb); | |
829 | if (n) { | |
830 | __cfqg = rb_entry_cfqg(n); | |
831 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; | |
832 | } else | |
833 | cfqg->vdisktime = st->min_vdisktime; | |
834 | ||
835 | __cfq_group_service_tree_add(st, cfqg); | |
836 | cfqg->on_st = true; | |
58ff82f3 VG |
837 | cfqd->nr_groups++; |
838 | st->total_weight += cfqg->weight; | |
1fa8f6d6 VG |
839 | } |
840 | ||
841 | static void | |
842 | cfq_group_service_tree_del(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
843 | { | |
844 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
845 | ||
25bc6b07 VG |
846 | if (st->active == &cfqg->rb_node) |
847 | st->active = NULL; | |
848 | ||
1fa8f6d6 VG |
849 | BUG_ON(cfqg->nr_cfqq < 1); |
850 | cfqg->nr_cfqq--; | |
25bc6b07 | 851 | |
1fa8f6d6 VG |
852 | /* If there are other cfq queues under this group, don't delete it */ |
853 | if (cfqg->nr_cfqq) | |
854 | return; | |
855 | ||
2868ef7b | 856 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
1fa8f6d6 | 857 | cfqg->on_st = false; |
58ff82f3 VG |
858 | cfqd->nr_groups--; |
859 | st->total_weight -= cfqg->weight; | |
1fa8f6d6 VG |
860 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
861 | cfq_rb_erase(&cfqg->rb_node, st); | |
dae739eb | 862 | cfqg->saved_workload_slice = 0; |
22084190 | 863 | blkiocg_update_blkio_group_dequeue_stats(&cfqg->blkg, 1); |
dae739eb VG |
864 | } |
865 | ||
866 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq) | |
867 | { | |
f75edf2d | 868 | unsigned int slice_used; |
dae739eb VG |
869 | |
870 | /* | |
871 | * Queue got expired before even a single request completed or | |
872 | * got expired immediately after first request completion. | |
873 | */ | |
874 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { | |
875 | /* | |
876 | * Also charge the seek time incurred to the group, otherwise | |
877 | * if there are mutiple queues in the group, each can dispatch | |
878 | * a single request on seeky media and cause lots of seek time | |
879 | * and group will never know it. | |
880 | */ | |
881 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), | |
882 | 1); | |
883 | } else { | |
884 | slice_used = jiffies - cfqq->slice_start; | |
f75edf2d VG |
885 | if (slice_used > cfqq->allocated_slice) |
886 | slice_used = cfqq->allocated_slice; | |
dae739eb VG |
887 | } |
888 | ||
22084190 VG |
889 | cfq_log_cfqq(cfqq->cfqd, cfqq, "sl_used=%u sect=%lu", slice_used, |
890 | cfqq->nr_sectors); | |
dae739eb VG |
891 | return slice_used; |
892 | } | |
893 | ||
894 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, | |
895 | struct cfq_queue *cfqq) | |
896 | { | |
897 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
898 | unsigned int used_sl; | |
899 | ||
900 | used_sl = cfq_cfqq_slice_usage(cfqq); | |
901 | ||
902 | /* Can't update vdisktime while group is on service tree */ | |
903 | cfq_rb_erase(&cfqg->rb_node, st); | |
904 | cfqg->vdisktime += cfq_scale_slice(used_sl, cfqg); | |
905 | __cfq_group_service_tree_add(st, cfqg); | |
906 | ||
907 | /* This group is being expired. Save the context */ | |
908 | if (time_after(cfqd->workload_expires, jiffies)) { | |
909 | cfqg->saved_workload_slice = cfqd->workload_expires | |
910 | - jiffies; | |
911 | cfqg->saved_workload = cfqd->serving_type; | |
912 | cfqg->saved_serving_prio = cfqd->serving_prio; | |
913 | } else | |
914 | cfqg->saved_workload_slice = 0; | |
2868ef7b VG |
915 | |
916 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, | |
917 | st->min_vdisktime); | |
22084190 VG |
918 | blkiocg_update_blkio_group_stats(&cfqg->blkg, used_sl, |
919 | cfqq->nr_sectors); | |
1fa8f6d6 VG |
920 | } |
921 | ||
25fb5169 VG |
922 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
923 | static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg) | |
924 | { | |
925 | if (blkg) | |
926 | return container_of(blkg, struct cfq_group, blkg); | |
927 | return NULL; | |
928 | } | |
929 | ||
f8d461d6 VG |
930 | void |
931 | cfq_update_blkio_group_weight(struct blkio_group *blkg, unsigned int weight) | |
932 | { | |
933 | cfqg_of_blkg(blkg)->weight = weight; | |
934 | } | |
935 | ||
25fb5169 VG |
936 | static struct cfq_group * |
937 | cfq_find_alloc_cfqg(struct cfq_data *cfqd, struct cgroup *cgroup, int create) | |
938 | { | |
939 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
940 | struct cfq_group *cfqg = NULL; | |
941 | void *key = cfqd; | |
942 | int i, j; | |
943 | struct cfq_rb_root *st; | |
22084190 VG |
944 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; |
945 | unsigned int major, minor; | |
25fb5169 VG |
946 | |
947 | /* Do we need to take this reference */ | |
948 | if (!css_tryget(&blkcg->css)) | |
949 | return NULL;; | |
950 | ||
951 | cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key)); | |
952 | if (cfqg || !create) | |
953 | goto done; | |
954 | ||
955 | cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node); | |
956 | if (!cfqg) | |
957 | goto done; | |
958 | ||
959 | cfqg->weight = blkcg->weight; | |
960 | for_each_cfqg_st(cfqg, i, j, st) | |
961 | *st = CFQ_RB_ROOT; | |
962 | RB_CLEAR_NODE(&cfqg->rb_node); | |
963 | ||
b1c35769 VG |
964 | /* |
965 | * Take the initial reference that will be released on destroy | |
966 | * This can be thought of a joint reference by cgroup and | |
967 | * elevator which will be dropped by either elevator exit | |
968 | * or cgroup deletion path depending on who is exiting first. | |
969 | */ | |
970 | atomic_set(&cfqg->ref, 1); | |
971 | ||
25fb5169 | 972 | /* Add group onto cgroup list */ |
22084190 VG |
973 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); |
974 | blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd, | |
975 | MKDEV(major, minor)); | |
25fb5169 VG |
976 | |
977 | /* Add group on cfqd list */ | |
978 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); | |
979 | ||
980 | done: | |
981 | css_put(&blkcg->css); | |
982 | return cfqg; | |
983 | } | |
984 | ||
985 | /* | |
986 | * Search for the cfq group current task belongs to. If create = 1, then also | |
987 | * create the cfq group if it does not exist. request_queue lock must be held. | |
988 | */ | |
989 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create) | |
990 | { | |
991 | struct cgroup *cgroup; | |
992 | struct cfq_group *cfqg = NULL; | |
993 | ||
994 | rcu_read_lock(); | |
995 | cgroup = task_cgroup(current, blkio_subsys_id); | |
996 | cfqg = cfq_find_alloc_cfqg(cfqd, cgroup, create); | |
997 | if (!cfqg && create) | |
998 | cfqg = &cfqd->root_group; | |
999 | rcu_read_unlock(); | |
1000 | return cfqg; | |
1001 | } | |
1002 | ||
1003 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) | |
1004 | { | |
1005 | /* Currently, all async queues are mapped to root group */ | |
1006 | if (!cfq_cfqq_sync(cfqq)) | |
1007 | cfqg = &cfqq->cfqd->root_group; | |
1008 | ||
1009 | cfqq->cfqg = cfqg; | |
b1c35769 VG |
1010 | /* cfqq reference on cfqg */ |
1011 | atomic_inc(&cfqq->cfqg->ref); | |
1012 | } | |
1013 | ||
1014 | static void cfq_put_cfqg(struct cfq_group *cfqg) | |
1015 | { | |
1016 | struct cfq_rb_root *st; | |
1017 | int i, j; | |
1018 | ||
1019 | BUG_ON(atomic_read(&cfqg->ref) <= 0); | |
1020 | if (!atomic_dec_and_test(&cfqg->ref)) | |
1021 | return; | |
1022 | for_each_cfqg_st(cfqg, i, j, st) | |
1023 | BUG_ON(!RB_EMPTY_ROOT(&st->rb) || st->active != NULL); | |
1024 | kfree(cfqg); | |
1025 | } | |
1026 | ||
1027 | static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg) | |
1028 | { | |
1029 | /* Something wrong if we are trying to remove same group twice */ | |
1030 | BUG_ON(hlist_unhashed(&cfqg->cfqd_node)); | |
1031 | ||
1032 | hlist_del_init(&cfqg->cfqd_node); | |
1033 | ||
1034 | /* | |
1035 | * Put the reference taken at the time of creation so that when all | |
1036 | * queues are gone, group can be destroyed. | |
1037 | */ | |
1038 | cfq_put_cfqg(cfqg); | |
1039 | } | |
1040 | ||
1041 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) | |
1042 | { | |
1043 | struct hlist_node *pos, *n; | |
1044 | struct cfq_group *cfqg; | |
1045 | ||
1046 | hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) { | |
1047 | /* | |
1048 | * If cgroup removal path got to blk_group first and removed | |
1049 | * it from cgroup list, then it will take care of destroying | |
1050 | * cfqg also. | |
1051 | */ | |
1052 | if (!blkiocg_del_blkio_group(&cfqg->blkg)) | |
1053 | cfq_destroy_cfqg(cfqd, cfqg); | |
1054 | } | |
25fb5169 | 1055 | } |
b1c35769 VG |
1056 | |
1057 | /* | |
1058 | * Blk cgroup controller notification saying that blkio_group object is being | |
1059 | * delinked as associated cgroup object is going away. That also means that | |
1060 | * no new IO will come in this group. So get rid of this group as soon as | |
1061 | * any pending IO in the group is finished. | |
1062 | * | |
1063 | * This function is called under rcu_read_lock(). key is the rcu protected | |
1064 | * pointer. That means "key" is a valid cfq_data pointer as long as we are rcu | |
1065 | * read lock. | |
1066 | * | |
1067 | * "key" was fetched from blkio_group under blkio_cgroup->lock. That means | |
1068 | * it should not be NULL as even if elevator was exiting, cgroup deltion | |
1069 | * path got to it first. | |
1070 | */ | |
1071 | void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg) | |
1072 | { | |
1073 | unsigned long flags; | |
1074 | struct cfq_data *cfqd = key; | |
1075 | ||
1076 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
1077 | cfq_destroy_cfqg(cfqd, cfqg_of_blkg(blkg)); | |
1078 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
1079 | } | |
1080 | ||
25fb5169 VG |
1081 | #else /* GROUP_IOSCHED */ |
1082 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create) | |
1083 | { | |
1084 | return &cfqd->root_group; | |
1085 | } | |
1086 | static inline void | |
1087 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { | |
1088 | cfqq->cfqg = cfqg; | |
1089 | } | |
1090 | ||
b1c35769 VG |
1091 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) {} |
1092 | static inline void cfq_put_cfqg(struct cfq_group *cfqg) {} | |
1093 | ||
25fb5169 VG |
1094 | #endif /* GROUP_IOSCHED */ |
1095 | ||
498d3aa2 | 1096 | /* |
c0324a02 | 1097 | * The cfqd->service_trees holds all pending cfq_queue's that have |
498d3aa2 JA |
1098 | * requests waiting to be processed. It is sorted in the order that |
1099 | * we will service the queues. | |
1100 | */ | |
a36e71f9 | 1101 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 1102 | bool add_front) |
d9e7620e | 1103 | { |
0871714e JA |
1104 | struct rb_node **p, *parent; |
1105 | struct cfq_queue *__cfqq; | |
d9e7620e | 1106 | unsigned long rb_key; |
c0324a02 | 1107 | struct cfq_rb_root *service_tree; |
498d3aa2 | 1108 | int left; |
dae739eb | 1109 | int new_cfqq = 1; |
d9e7620e | 1110 | |
cdb16e8f VG |
1111 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
1112 | cfqq_type(cfqq), cfqd); | |
0871714e JA |
1113 | if (cfq_class_idle(cfqq)) { |
1114 | rb_key = CFQ_IDLE_DELAY; | |
aa6f6a3d | 1115 | parent = rb_last(&service_tree->rb); |
0871714e JA |
1116 | if (parent && parent != &cfqq->rb_node) { |
1117 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1118 | rb_key += __cfqq->rb_key; | |
1119 | } else | |
1120 | rb_key += jiffies; | |
1121 | } else if (!add_front) { | |
b9c8946b JA |
1122 | /* |
1123 | * Get our rb key offset. Subtract any residual slice | |
1124 | * value carried from last service. A negative resid | |
1125 | * count indicates slice overrun, and this should position | |
1126 | * the next service time further away in the tree. | |
1127 | */ | |
edd75ffd | 1128 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
b9c8946b | 1129 | rb_key -= cfqq->slice_resid; |
edd75ffd | 1130 | cfqq->slice_resid = 0; |
48e025e6 CZ |
1131 | } else { |
1132 | rb_key = -HZ; | |
aa6f6a3d | 1133 | __cfqq = cfq_rb_first(service_tree); |
48e025e6 CZ |
1134 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
1135 | } | |
1da177e4 | 1136 | |
d9e7620e | 1137 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
dae739eb | 1138 | new_cfqq = 0; |
99f9628a | 1139 | /* |
d9e7620e | 1140 | * same position, nothing more to do |
99f9628a | 1141 | */ |
c0324a02 CZ |
1142 | if (rb_key == cfqq->rb_key && |
1143 | cfqq->service_tree == service_tree) | |
d9e7620e | 1144 | return; |
1da177e4 | 1145 | |
aa6f6a3d CZ |
1146 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
1147 | cfqq->service_tree = NULL; | |
1da177e4 | 1148 | } |
d9e7620e | 1149 | |
498d3aa2 | 1150 | left = 1; |
0871714e | 1151 | parent = NULL; |
aa6f6a3d CZ |
1152 | cfqq->service_tree = service_tree; |
1153 | p = &service_tree->rb.rb_node; | |
d9e7620e | 1154 | while (*p) { |
67060e37 | 1155 | struct rb_node **n; |
cc09e299 | 1156 | |
d9e7620e JA |
1157 | parent = *p; |
1158 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); | |
1159 | ||
0c534e0a | 1160 | /* |
c0324a02 | 1161 | * sort by key, that represents service time. |
0c534e0a | 1162 | */ |
c0324a02 | 1163 | if (time_before(rb_key, __cfqq->rb_key)) |
67060e37 | 1164 | n = &(*p)->rb_left; |
c0324a02 | 1165 | else { |
67060e37 | 1166 | n = &(*p)->rb_right; |
cc09e299 | 1167 | left = 0; |
c0324a02 | 1168 | } |
67060e37 JA |
1169 | |
1170 | p = n; | |
d9e7620e JA |
1171 | } |
1172 | ||
cc09e299 | 1173 | if (left) |
aa6f6a3d | 1174 | service_tree->left = &cfqq->rb_node; |
cc09e299 | 1175 | |
d9e7620e JA |
1176 | cfqq->rb_key = rb_key; |
1177 | rb_link_node(&cfqq->rb_node, parent, p); | |
aa6f6a3d CZ |
1178 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
1179 | service_tree->count++; | |
dae739eb VG |
1180 | if (add_front || !new_cfqq) |
1181 | return; | |
1fa8f6d6 | 1182 | cfq_group_service_tree_add(cfqd, cfqq->cfqg); |
1da177e4 LT |
1183 | } |
1184 | ||
a36e71f9 | 1185 | static struct cfq_queue * |
f2d1f0ae JA |
1186 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
1187 | sector_t sector, struct rb_node **ret_parent, | |
1188 | struct rb_node ***rb_link) | |
a36e71f9 | 1189 | { |
a36e71f9 JA |
1190 | struct rb_node **p, *parent; |
1191 | struct cfq_queue *cfqq = NULL; | |
1192 | ||
1193 | parent = NULL; | |
1194 | p = &root->rb_node; | |
1195 | while (*p) { | |
1196 | struct rb_node **n; | |
1197 | ||
1198 | parent = *p; | |
1199 | cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
1200 | ||
1201 | /* | |
1202 | * Sort strictly based on sector. Smallest to the left, | |
1203 | * largest to the right. | |
1204 | */ | |
2e46e8b2 | 1205 | if (sector > blk_rq_pos(cfqq->next_rq)) |
a36e71f9 | 1206 | n = &(*p)->rb_right; |
2e46e8b2 | 1207 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
a36e71f9 JA |
1208 | n = &(*p)->rb_left; |
1209 | else | |
1210 | break; | |
1211 | p = n; | |
3ac6c9f8 | 1212 | cfqq = NULL; |
a36e71f9 JA |
1213 | } |
1214 | ||
1215 | *ret_parent = parent; | |
1216 | if (rb_link) | |
1217 | *rb_link = p; | |
3ac6c9f8 | 1218 | return cfqq; |
a36e71f9 JA |
1219 | } |
1220 | ||
1221 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1222 | { | |
a36e71f9 JA |
1223 | struct rb_node **p, *parent; |
1224 | struct cfq_queue *__cfqq; | |
1225 | ||
f2d1f0ae JA |
1226 | if (cfqq->p_root) { |
1227 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1228 | cfqq->p_root = NULL; | |
1229 | } | |
a36e71f9 JA |
1230 | |
1231 | if (cfq_class_idle(cfqq)) | |
1232 | return; | |
1233 | if (!cfqq->next_rq) | |
1234 | return; | |
1235 | ||
f2d1f0ae | 1236 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
2e46e8b2 TH |
1237 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
1238 | blk_rq_pos(cfqq->next_rq), &parent, &p); | |
3ac6c9f8 JA |
1239 | if (!__cfqq) { |
1240 | rb_link_node(&cfqq->p_node, parent, p); | |
f2d1f0ae JA |
1241 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
1242 | } else | |
1243 | cfqq->p_root = NULL; | |
a36e71f9 JA |
1244 | } |
1245 | ||
498d3aa2 JA |
1246 | /* |
1247 | * Update cfqq's position in the service tree. | |
1248 | */ | |
edd75ffd | 1249 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
6d048f53 | 1250 | { |
6d048f53 JA |
1251 | /* |
1252 | * Resorting requires the cfqq to be on the RR list already. | |
1253 | */ | |
a36e71f9 | 1254 | if (cfq_cfqq_on_rr(cfqq)) { |
edd75ffd | 1255 | cfq_service_tree_add(cfqd, cfqq, 0); |
a36e71f9 JA |
1256 | cfq_prio_tree_add(cfqd, cfqq); |
1257 | } | |
6d048f53 JA |
1258 | } |
1259 | ||
1da177e4 LT |
1260 | /* |
1261 | * add to busy list of queues for service, trying to be fair in ordering | |
22e2c507 | 1262 | * the pending list according to last request service |
1da177e4 | 1263 | */ |
febffd61 | 1264 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1265 | { |
7b679138 | 1266 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
3b18152c JA |
1267 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1268 | cfq_mark_cfqq_on_rr(cfqq); | |
1da177e4 LT |
1269 | cfqd->busy_queues++; |
1270 | ||
edd75ffd | 1271 | cfq_resort_rr_list(cfqd, cfqq); |
1da177e4 LT |
1272 | } |
1273 | ||
498d3aa2 JA |
1274 | /* |
1275 | * Called when the cfqq no longer has requests pending, remove it from | |
1276 | * the service tree. | |
1277 | */ | |
febffd61 | 1278 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 1279 | { |
7b679138 | 1280 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
3b18152c JA |
1281 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
1282 | cfq_clear_cfqq_on_rr(cfqq); | |
1da177e4 | 1283 | |
aa6f6a3d CZ |
1284 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
1285 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); | |
1286 | cfqq->service_tree = NULL; | |
1287 | } | |
f2d1f0ae JA |
1288 | if (cfqq->p_root) { |
1289 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1290 | cfqq->p_root = NULL; | |
1291 | } | |
d9e7620e | 1292 | |
1fa8f6d6 | 1293 | cfq_group_service_tree_del(cfqd, cfqq->cfqg); |
1da177e4 LT |
1294 | BUG_ON(!cfqd->busy_queues); |
1295 | cfqd->busy_queues--; | |
1296 | } | |
1297 | ||
1298 | /* | |
1299 | * rb tree support functions | |
1300 | */ | |
febffd61 | 1301 | static void cfq_del_rq_rb(struct request *rq) |
1da177e4 | 1302 | { |
5e705374 | 1303 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
5e705374 | 1304 | const int sync = rq_is_sync(rq); |
1da177e4 | 1305 | |
b4878f24 JA |
1306 | BUG_ON(!cfqq->queued[sync]); |
1307 | cfqq->queued[sync]--; | |
1da177e4 | 1308 | |
5e705374 | 1309 | elv_rb_del(&cfqq->sort_list, rq); |
1da177e4 | 1310 | |
f04a6424 VG |
1311 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
1312 | /* | |
1313 | * Queue will be deleted from service tree when we actually | |
1314 | * expire it later. Right now just remove it from prio tree | |
1315 | * as it is empty. | |
1316 | */ | |
1317 | if (cfqq->p_root) { | |
1318 | rb_erase(&cfqq->p_node, cfqq->p_root); | |
1319 | cfqq->p_root = NULL; | |
1320 | } | |
1321 | } | |
1da177e4 LT |
1322 | } |
1323 | ||
5e705374 | 1324 | static void cfq_add_rq_rb(struct request *rq) |
1da177e4 | 1325 | { |
5e705374 | 1326 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 1327 | struct cfq_data *cfqd = cfqq->cfqd; |
a36e71f9 | 1328 | struct request *__alias, *prev; |
1da177e4 | 1329 | |
5380a101 | 1330 | cfqq->queued[rq_is_sync(rq)]++; |
1da177e4 LT |
1331 | |
1332 | /* | |
1333 | * looks a little odd, but the first insert might return an alias. | |
1334 | * if that happens, put the alias on the dispatch list | |
1335 | */ | |
21183b07 | 1336 | while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL) |
5e705374 | 1337 | cfq_dispatch_insert(cfqd->queue, __alias); |
5fccbf61 JA |
1338 | |
1339 | if (!cfq_cfqq_on_rr(cfqq)) | |
1340 | cfq_add_cfqq_rr(cfqd, cfqq); | |
5044eed4 JA |
1341 | |
1342 | /* | |
1343 | * check if this request is a better next-serve candidate | |
1344 | */ | |
a36e71f9 | 1345 | prev = cfqq->next_rq; |
cf7c25cf | 1346 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
a36e71f9 JA |
1347 | |
1348 | /* | |
1349 | * adjust priority tree position, if ->next_rq changes | |
1350 | */ | |
1351 | if (prev != cfqq->next_rq) | |
1352 | cfq_prio_tree_add(cfqd, cfqq); | |
1353 | ||
5044eed4 | 1354 | BUG_ON(!cfqq->next_rq); |
1da177e4 LT |
1355 | } |
1356 | ||
febffd61 | 1357 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
1da177e4 | 1358 | { |
5380a101 JA |
1359 | elv_rb_del(&cfqq->sort_list, rq); |
1360 | cfqq->queued[rq_is_sync(rq)]--; | |
5e705374 | 1361 | cfq_add_rq_rb(rq); |
1da177e4 LT |
1362 | } |
1363 | ||
206dc69b JA |
1364 | static struct request * |
1365 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) | |
1da177e4 | 1366 | { |
206dc69b | 1367 | struct task_struct *tsk = current; |
91fac317 | 1368 | struct cfq_io_context *cic; |
206dc69b | 1369 | struct cfq_queue *cfqq; |
1da177e4 | 1370 | |
4ac845a2 | 1371 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
1372 | if (!cic) |
1373 | return NULL; | |
1374 | ||
1375 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); | |
89850f7e JA |
1376 | if (cfqq) { |
1377 | sector_t sector = bio->bi_sector + bio_sectors(bio); | |
1378 | ||
21183b07 | 1379 | return elv_rb_find(&cfqq->sort_list, sector); |
89850f7e | 1380 | } |
1da177e4 | 1381 | |
1da177e4 LT |
1382 | return NULL; |
1383 | } | |
1384 | ||
165125e1 | 1385 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1386 | { |
22e2c507 | 1387 | struct cfq_data *cfqd = q->elevator->elevator_data; |
3b18152c | 1388 | |
5ad531db | 1389 | cfqd->rq_in_driver[rq_is_sync(rq)]++; |
7b679138 | 1390 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
5ad531db | 1391 | rq_in_driver(cfqd)); |
25776e35 | 1392 | |
5b93629b | 1393 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
1da177e4 LT |
1394 | } |
1395 | ||
165125e1 | 1396 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1397 | { |
b4878f24 | 1398 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5ad531db | 1399 | const int sync = rq_is_sync(rq); |
b4878f24 | 1400 | |
5ad531db JA |
1401 | WARN_ON(!cfqd->rq_in_driver[sync]); |
1402 | cfqd->rq_in_driver[sync]--; | |
7b679138 | 1403 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
5ad531db | 1404 | rq_in_driver(cfqd)); |
1da177e4 LT |
1405 | } |
1406 | ||
b4878f24 | 1407 | static void cfq_remove_request(struct request *rq) |
1da177e4 | 1408 | { |
5e705374 | 1409 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
21183b07 | 1410 | |
5e705374 JA |
1411 | if (cfqq->next_rq == rq) |
1412 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); | |
1da177e4 | 1413 | |
b4878f24 | 1414 | list_del_init(&rq->queuelist); |
5e705374 | 1415 | cfq_del_rq_rb(rq); |
374f84ac | 1416 | |
45333d5a | 1417 | cfqq->cfqd->rq_queued--; |
374f84ac JA |
1418 | if (rq_is_meta(rq)) { |
1419 | WARN_ON(!cfqq->meta_pending); | |
1420 | cfqq->meta_pending--; | |
1421 | } | |
1da177e4 LT |
1422 | } |
1423 | ||
165125e1 JA |
1424 | static int cfq_merge(struct request_queue *q, struct request **req, |
1425 | struct bio *bio) | |
1da177e4 LT |
1426 | { |
1427 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
1428 | struct request *__rq; | |
1da177e4 | 1429 | |
206dc69b | 1430 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
22e2c507 | 1431 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
9817064b JA |
1432 | *req = __rq; |
1433 | return ELEVATOR_FRONT_MERGE; | |
1da177e4 LT |
1434 | } |
1435 | ||
1436 | return ELEVATOR_NO_MERGE; | |
1da177e4 LT |
1437 | } |
1438 | ||
165125e1 | 1439 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
21183b07 | 1440 | int type) |
1da177e4 | 1441 | { |
21183b07 | 1442 | if (type == ELEVATOR_FRONT_MERGE) { |
5e705374 | 1443 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
1da177e4 | 1444 | |
5e705374 | 1445 | cfq_reposition_rq_rb(cfqq, req); |
1da177e4 | 1446 | } |
1da177e4 LT |
1447 | } |
1448 | ||
1449 | static void | |
165125e1 | 1450 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
1451 | struct request *next) |
1452 | { | |
cf7c25cf | 1453 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 JA |
1454 | /* |
1455 | * reposition in fifo if next is older than rq | |
1456 | */ | |
1457 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
30996f40 | 1458 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
22e2c507 | 1459 | list_move(&rq->queuelist, &next->queuelist); |
30996f40 JA |
1460 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
1461 | } | |
22e2c507 | 1462 | |
cf7c25cf CZ |
1463 | if (cfqq->next_rq == next) |
1464 | cfqq->next_rq = rq; | |
b4878f24 | 1465 | cfq_remove_request(next); |
22e2c507 JA |
1466 | } |
1467 | ||
165125e1 | 1468 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
da775265 JA |
1469 | struct bio *bio) |
1470 | { | |
1471 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
91fac317 | 1472 | struct cfq_io_context *cic; |
da775265 | 1473 | struct cfq_queue *cfqq; |
da775265 | 1474 | |
8682e1f1 VG |
1475 | /* Deny merge if bio and rq don't belong to same cfq group */ |
1476 | if ((RQ_CFQQ(rq))->cfqg != cfq_get_cfqg(cfqd, 0)) | |
1477 | return false; | |
da775265 | 1478 | /* |
ec8acb69 | 1479 | * Disallow merge of a sync bio into an async request. |
da775265 | 1480 | */ |
91fac317 | 1481 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
a6151c3a | 1482 | return false; |
da775265 JA |
1483 | |
1484 | /* | |
719d3402 JA |
1485 | * Lookup the cfqq that this bio will be queued with. Allow |
1486 | * merge only if rq is queued there. | |
da775265 | 1487 | */ |
4ac845a2 | 1488 | cic = cfq_cic_lookup(cfqd, current->io_context); |
91fac317 | 1489 | if (!cic) |
a6151c3a | 1490 | return false; |
719d3402 | 1491 | |
91fac317 | 1492 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
a6151c3a | 1493 | return cfqq == RQ_CFQQ(rq); |
da775265 JA |
1494 | } |
1495 | ||
febffd61 JA |
1496 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
1497 | struct cfq_queue *cfqq) | |
22e2c507 JA |
1498 | { |
1499 | if (cfqq) { | |
7b679138 | 1500 | cfq_log_cfqq(cfqd, cfqq, "set_active"); |
dae739eb VG |
1501 | cfqq->slice_start = 0; |
1502 | cfqq->dispatch_start = jiffies; | |
f75edf2d | 1503 | cfqq->allocated_slice = 0; |
22e2c507 | 1504 | cfqq->slice_end = 0; |
2f5cb738 | 1505 | cfqq->slice_dispatch = 0; |
22084190 | 1506 | cfqq->nr_sectors = 0; |
2f5cb738 | 1507 | |
2f5cb738 | 1508 | cfq_clear_cfqq_wait_request(cfqq); |
b029195d | 1509 | cfq_clear_cfqq_must_dispatch(cfqq); |
3b18152c JA |
1510 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
1511 | cfq_clear_cfqq_fifo_expire(cfqq); | |
44f7c160 | 1512 | cfq_mark_cfqq_slice_new(cfqq); |
2f5cb738 JA |
1513 | |
1514 | del_timer(&cfqd->idle_slice_timer); | |
22e2c507 JA |
1515 | } |
1516 | ||
1517 | cfqd->active_queue = cfqq; | |
1518 | } | |
1519 | ||
7b14e3b5 JA |
1520 | /* |
1521 | * current cfqq expired its slice (or was too idle), select new one | |
1522 | */ | |
1523 | static void | |
1524 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
a6151c3a | 1525 | bool timed_out) |
7b14e3b5 | 1526 | { |
7b679138 JA |
1527 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
1528 | ||
7b14e3b5 JA |
1529 | if (cfq_cfqq_wait_request(cfqq)) |
1530 | del_timer(&cfqd->idle_slice_timer); | |
1531 | ||
7b14e3b5 | 1532 | cfq_clear_cfqq_wait_request(cfqq); |
f75edf2d VG |
1533 | cfq_clear_cfqq_wait_busy(cfqq); |
1534 | cfq_clear_cfqq_wait_busy_done(cfqq); | |
7b14e3b5 JA |
1535 | |
1536 | /* | |
6084cdda | 1537 | * store what was left of this slice, if the queue idled/timed out |
7b14e3b5 | 1538 | */ |
7b679138 | 1539 | if (timed_out && !cfq_cfqq_slice_new(cfqq)) { |
c5b680f3 | 1540 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
7b679138 JA |
1541 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
1542 | } | |
7b14e3b5 | 1543 | |
dae739eb VG |
1544 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
1545 | ||
f04a6424 VG |
1546 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
1547 | cfq_del_cfqq_rr(cfqd, cfqq); | |
1548 | ||
edd75ffd | 1549 | cfq_resort_rr_list(cfqd, cfqq); |
7b14e3b5 JA |
1550 | |
1551 | if (cfqq == cfqd->active_queue) | |
1552 | cfqd->active_queue = NULL; | |
1553 | ||
dae739eb VG |
1554 | if (&cfqq->cfqg->rb_node == cfqd->grp_service_tree.active) |
1555 | cfqd->grp_service_tree.active = NULL; | |
1556 | ||
7b14e3b5 JA |
1557 | if (cfqd->active_cic) { |
1558 | put_io_context(cfqd->active_cic->ioc); | |
1559 | cfqd->active_cic = NULL; | |
1560 | } | |
7b14e3b5 JA |
1561 | } |
1562 | ||
a6151c3a | 1563 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
7b14e3b5 JA |
1564 | { |
1565 | struct cfq_queue *cfqq = cfqd->active_queue; | |
1566 | ||
1567 | if (cfqq) | |
6084cdda | 1568 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
7b14e3b5 JA |
1569 | } |
1570 | ||
498d3aa2 JA |
1571 | /* |
1572 | * Get next queue for service. Unless we have a queue preemption, | |
1573 | * we'll simply select the first cfqq in the service tree. | |
1574 | */ | |
6d048f53 | 1575 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
22e2c507 | 1576 | { |
c0324a02 | 1577 | struct cfq_rb_root *service_tree = |
cdb16e8f VG |
1578 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
1579 | cfqd->serving_type, cfqd); | |
d9e7620e | 1580 | |
f04a6424 VG |
1581 | if (!cfqd->rq_queued) |
1582 | return NULL; | |
1583 | ||
1fa8f6d6 VG |
1584 | /* There is nothing to dispatch */ |
1585 | if (!service_tree) | |
1586 | return NULL; | |
c0324a02 CZ |
1587 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
1588 | return NULL; | |
1589 | return cfq_rb_first(service_tree); | |
6d048f53 JA |
1590 | } |
1591 | ||
f04a6424 VG |
1592 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
1593 | { | |
25fb5169 | 1594 | struct cfq_group *cfqg; |
f04a6424 VG |
1595 | struct cfq_queue *cfqq; |
1596 | int i, j; | |
1597 | struct cfq_rb_root *st; | |
1598 | ||
1599 | if (!cfqd->rq_queued) | |
1600 | return NULL; | |
1601 | ||
25fb5169 VG |
1602 | cfqg = cfq_get_next_cfqg(cfqd); |
1603 | if (!cfqg) | |
1604 | return NULL; | |
1605 | ||
f04a6424 VG |
1606 | for_each_cfqg_st(cfqg, i, j, st) |
1607 | if ((cfqq = cfq_rb_first(st)) != NULL) | |
1608 | return cfqq; | |
1609 | return NULL; | |
1610 | } | |
1611 | ||
498d3aa2 JA |
1612 | /* |
1613 | * Get and set a new active queue for service. | |
1614 | */ | |
a36e71f9 JA |
1615 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
1616 | struct cfq_queue *cfqq) | |
6d048f53 | 1617 | { |
e00ef799 | 1618 | if (!cfqq) |
a36e71f9 | 1619 | cfqq = cfq_get_next_queue(cfqd); |
6d048f53 | 1620 | |
22e2c507 | 1621 | __cfq_set_active_queue(cfqd, cfqq); |
3b18152c | 1622 | return cfqq; |
22e2c507 JA |
1623 | } |
1624 | ||
d9e7620e JA |
1625 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
1626 | struct request *rq) | |
1627 | { | |
83096ebf TH |
1628 | if (blk_rq_pos(rq) >= cfqd->last_position) |
1629 | return blk_rq_pos(rq) - cfqd->last_position; | |
d9e7620e | 1630 | else |
83096ebf | 1631 | return cfqd->last_position - blk_rq_pos(rq); |
d9e7620e JA |
1632 | } |
1633 | ||
b2c18e1e JM |
1634 | #define CFQQ_SEEK_THR 8 * 1024 |
1635 | #define CFQQ_SEEKY(cfqq) ((cfqq)->seek_mean > CFQQ_SEEK_THR) | |
04dc6e71 | 1636 | |
b2c18e1e JM |
1637 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
1638 | struct request *rq) | |
6d048f53 | 1639 | { |
b2c18e1e | 1640 | sector_t sdist = cfqq->seek_mean; |
6d048f53 | 1641 | |
b2c18e1e JM |
1642 | if (!sample_valid(cfqq->seek_samples)) |
1643 | sdist = CFQQ_SEEK_THR; | |
6d048f53 | 1644 | |
04dc6e71 | 1645 | return cfq_dist_from_last(cfqd, rq) <= sdist; |
6d048f53 JA |
1646 | } |
1647 | ||
a36e71f9 JA |
1648 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
1649 | struct cfq_queue *cur_cfqq) | |
1650 | { | |
f2d1f0ae | 1651 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
a36e71f9 JA |
1652 | struct rb_node *parent, *node; |
1653 | struct cfq_queue *__cfqq; | |
1654 | sector_t sector = cfqd->last_position; | |
1655 | ||
1656 | if (RB_EMPTY_ROOT(root)) | |
1657 | return NULL; | |
1658 | ||
1659 | /* | |
1660 | * First, if we find a request starting at the end of the last | |
1661 | * request, choose it. | |
1662 | */ | |
f2d1f0ae | 1663 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
a36e71f9 JA |
1664 | if (__cfqq) |
1665 | return __cfqq; | |
1666 | ||
1667 | /* | |
1668 | * If the exact sector wasn't found, the parent of the NULL leaf | |
1669 | * will contain the closest sector. | |
1670 | */ | |
1671 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); | |
b2c18e1e | 1672 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1673 | return __cfqq; |
1674 | ||
2e46e8b2 | 1675 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
a36e71f9 JA |
1676 | node = rb_next(&__cfqq->p_node); |
1677 | else | |
1678 | node = rb_prev(&__cfqq->p_node); | |
1679 | if (!node) | |
1680 | return NULL; | |
1681 | ||
1682 | __cfqq = rb_entry(node, struct cfq_queue, p_node); | |
b2c18e1e | 1683 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
a36e71f9 JA |
1684 | return __cfqq; |
1685 | ||
1686 | return NULL; | |
1687 | } | |
1688 | ||
1689 | /* | |
1690 | * cfqd - obvious | |
1691 | * cur_cfqq - passed in so that we don't decide that the current queue is | |
1692 | * closely cooperating with itself. | |
1693 | * | |
1694 | * So, basically we're assuming that that cur_cfqq has dispatched at least | |
1695 | * one request, and that cfqd->last_position reflects a position on the disk | |
1696 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid | |
1697 | * assumption. | |
1698 | */ | |
1699 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, | |
b3b6d040 | 1700 | struct cfq_queue *cur_cfqq) |
6d048f53 | 1701 | { |
a36e71f9 JA |
1702 | struct cfq_queue *cfqq; |
1703 | ||
e6c5bc73 JM |
1704 | if (!cfq_cfqq_sync(cur_cfqq)) |
1705 | return NULL; | |
1706 | if (CFQQ_SEEKY(cur_cfqq)) | |
1707 | return NULL; | |
1708 | ||
6d048f53 | 1709 | /* |
d9e7620e JA |
1710 | * We should notice if some of the queues are cooperating, eg |
1711 | * working closely on the same area of the disk. In that case, | |
1712 | * we can group them together and don't waste time idling. | |
6d048f53 | 1713 | */ |
a36e71f9 JA |
1714 | cfqq = cfqq_close(cfqd, cur_cfqq); |
1715 | if (!cfqq) | |
1716 | return NULL; | |
1717 | ||
8682e1f1 VG |
1718 | /* If new queue belongs to different cfq_group, don't choose it */ |
1719 | if (cur_cfqq->cfqg != cfqq->cfqg) | |
1720 | return NULL; | |
1721 | ||
df5fe3e8 JM |
1722 | /* |
1723 | * It only makes sense to merge sync queues. | |
1724 | */ | |
1725 | if (!cfq_cfqq_sync(cfqq)) | |
1726 | return NULL; | |
e6c5bc73 JM |
1727 | if (CFQQ_SEEKY(cfqq)) |
1728 | return NULL; | |
df5fe3e8 | 1729 | |
c0324a02 CZ |
1730 | /* |
1731 | * Do not merge queues of different priority classes | |
1732 | */ | |
1733 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) | |
1734 | return NULL; | |
1735 | ||
a36e71f9 | 1736 | return cfqq; |
6d048f53 JA |
1737 | } |
1738 | ||
a6d44e98 CZ |
1739 | /* |
1740 | * Determine whether we should enforce idle window for this queue. | |
1741 | */ | |
1742 | ||
1743 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1744 | { | |
1745 | enum wl_prio_t prio = cfqq_prio(cfqq); | |
718eee05 | 1746 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
a6d44e98 | 1747 | |
f04a6424 VG |
1748 | BUG_ON(!service_tree); |
1749 | BUG_ON(!service_tree->count); | |
1750 | ||
a6d44e98 CZ |
1751 | /* We never do for idle class queues. */ |
1752 | if (prio == IDLE_WORKLOAD) | |
1753 | return false; | |
1754 | ||
1755 | /* We do for queues that were marked with idle window flag. */ | |
1756 | if (cfq_cfqq_idle_window(cfqq)) | |
1757 | return true; | |
1758 | ||
1759 | /* | |
1760 | * Otherwise, we do only if they are the last ones | |
1761 | * in their service tree. | |
1762 | */ | |
f04a6424 | 1763 | return service_tree->count == 1; |
a6d44e98 CZ |
1764 | } |
1765 | ||
6d048f53 | 1766 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
22e2c507 | 1767 | { |
1792669c | 1768 | struct cfq_queue *cfqq = cfqd->active_queue; |
206dc69b | 1769 | struct cfq_io_context *cic; |
7b14e3b5 JA |
1770 | unsigned long sl; |
1771 | ||
a68bbddb | 1772 | /* |
f7d7b7a7 JA |
1773 | * SSD device without seek penalty, disable idling. But only do so |
1774 | * for devices that support queuing, otherwise we still have a problem | |
1775 | * with sync vs async workloads. | |
a68bbddb | 1776 | */ |
f7d7b7a7 | 1777 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
a68bbddb JA |
1778 | return; |
1779 | ||
dd67d051 | 1780 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
6d048f53 | 1781 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
22e2c507 JA |
1782 | |
1783 | /* | |
1784 | * idle is disabled, either manually or by past process history | |
1785 | */ | |
a6d44e98 | 1786 | if (!cfqd->cfq_slice_idle || !cfq_should_idle(cfqd, cfqq)) |
6d048f53 JA |
1787 | return; |
1788 | ||
7b679138 | 1789 | /* |
8e550632 | 1790 | * still active requests from this queue, don't idle |
7b679138 | 1791 | */ |
8e550632 | 1792 | if (cfqq->dispatched) |
7b679138 JA |
1793 | return; |
1794 | ||
22e2c507 JA |
1795 | /* |
1796 | * task has exited, don't wait | |
1797 | */ | |
206dc69b | 1798 | cic = cfqd->active_cic; |
66dac98e | 1799 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
6d048f53 JA |
1800 | return; |
1801 | ||
355b659c CZ |
1802 | /* |
1803 | * If our average think time is larger than the remaining time | |
1804 | * slice, then don't idle. This avoids overrunning the allotted | |
1805 | * time slice. | |
1806 | */ | |
1807 | if (sample_valid(cic->ttime_samples) && | |
1808 | (cfqq->slice_end - jiffies < cic->ttime_mean)) | |
1809 | return; | |
1810 | ||
3b18152c | 1811 | cfq_mark_cfqq_wait_request(cfqq); |
22e2c507 | 1812 | |
6d048f53 | 1813 | sl = cfqd->cfq_slice_idle; |
206dc69b | 1814 | |
7b14e3b5 | 1815 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
9481ffdc | 1816 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu", sl); |
1da177e4 LT |
1817 | } |
1818 | ||
498d3aa2 JA |
1819 | /* |
1820 | * Move request from internal lists to the request queue dispatch list. | |
1821 | */ | |
165125e1 | 1822 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
1da177e4 | 1823 | { |
3ed9a296 | 1824 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 1825 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 1826 | |
7b679138 JA |
1827 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
1828 | ||
06d21886 | 1829 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
5380a101 | 1830 | cfq_remove_request(rq); |
6d048f53 | 1831 | cfqq->dispatched++; |
5380a101 | 1832 | elv_dispatch_sort(q, rq); |
3ed9a296 JA |
1833 | |
1834 | if (cfq_cfqq_sync(cfqq)) | |
1835 | cfqd->sync_flight++; | |
22084190 | 1836 | cfqq->nr_sectors += blk_rq_sectors(rq); |
1da177e4 LT |
1837 | } |
1838 | ||
1839 | /* | |
1840 | * return expired entry, or NULL to just start from scratch in rbtree | |
1841 | */ | |
febffd61 | 1842 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
1da177e4 | 1843 | { |
30996f40 | 1844 | struct request *rq = NULL; |
1da177e4 | 1845 | |
3b18152c | 1846 | if (cfq_cfqq_fifo_expire(cfqq)) |
1da177e4 | 1847 | return NULL; |
cb887411 JA |
1848 | |
1849 | cfq_mark_cfqq_fifo_expire(cfqq); | |
1850 | ||
89850f7e JA |
1851 | if (list_empty(&cfqq->fifo)) |
1852 | return NULL; | |
1da177e4 | 1853 | |
89850f7e | 1854 | rq = rq_entry_fifo(cfqq->fifo.next); |
30996f40 | 1855 | if (time_before(jiffies, rq_fifo_time(rq))) |
7b679138 | 1856 | rq = NULL; |
1da177e4 | 1857 | |
30996f40 | 1858 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
6d048f53 | 1859 | return rq; |
1da177e4 LT |
1860 | } |
1861 | ||
22e2c507 JA |
1862 | static inline int |
1863 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
1864 | { | |
1865 | const int base_rq = cfqd->cfq_slice_async_rq; | |
1da177e4 | 1866 | |
22e2c507 | 1867 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
1da177e4 | 1868 | |
22e2c507 | 1869 | return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); |
1da177e4 LT |
1870 | } |
1871 | ||
df5fe3e8 JM |
1872 | /* |
1873 | * Must be called with the queue_lock held. | |
1874 | */ | |
1875 | static int cfqq_process_refs(struct cfq_queue *cfqq) | |
1876 | { | |
1877 | int process_refs, io_refs; | |
1878 | ||
1879 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; | |
1880 | process_refs = atomic_read(&cfqq->ref) - io_refs; | |
1881 | BUG_ON(process_refs < 0); | |
1882 | return process_refs; | |
1883 | } | |
1884 | ||
1885 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) | |
1886 | { | |
e6c5bc73 | 1887 | int process_refs, new_process_refs; |
df5fe3e8 JM |
1888 | struct cfq_queue *__cfqq; |
1889 | ||
1890 | /* Avoid a circular list and skip interim queue merges */ | |
1891 | while ((__cfqq = new_cfqq->new_cfqq)) { | |
1892 | if (__cfqq == cfqq) | |
1893 | return; | |
1894 | new_cfqq = __cfqq; | |
1895 | } | |
1896 | ||
1897 | process_refs = cfqq_process_refs(cfqq); | |
1898 | /* | |
1899 | * If the process for the cfqq has gone away, there is no | |
1900 | * sense in merging the queues. | |
1901 | */ | |
1902 | if (process_refs == 0) | |
1903 | return; | |
1904 | ||
e6c5bc73 JM |
1905 | /* |
1906 | * Merge in the direction of the lesser amount of work. | |
1907 | */ | |
1908 | new_process_refs = cfqq_process_refs(new_cfqq); | |
1909 | if (new_process_refs >= process_refs) { | |
1910 | cfqq->new_cfqq = new_cfqq; | |
1911 | atomic_add(process_refs, &new_cfqq->ref); | |
1912 | } else { | |
1913 | new_cfqq->new_cfqq = cfqq; | |
1914 | atomic_add(new_process_refs, &cfqq->ref); | |
1915 | } | |
df5fe3e8 JM |
1916 | } |
1917 | ||
cdb16e8f VG |
1918 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
1919 | struct cfq_group *cfqg, enum wl_prio_t prio, | |
1920 | bool prio_changed) | |
718eee05 CZ |
1921 | { |
1922 | struct cfq_queue *queue; | |
1923 | int i; | |
1924 | bool key_valid = false; | |
1925 | unsigned long lowest_key = 0; | |
1926 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; | |
1927 | ||
1928 | if (prio_changed) { | |
1929 | /* | |
1930 | * When priorities switched, we prefer starting | |
1931 | * from SYNC_NOIDLE (first choice), or just SYNC | |
1932 | * over ASYNC | |
1933 | */ | |
cdb16e8f | 1934 | if (service_tree_for(cfqg, prio, cur_best, cfqd)->count) |
718eee05 CZ |
1935 | return cur_best; |
1936 | cur_best = SYNC_WORKLOAD; | |
cdb16e8f | 1937 | if (service_tree_for(cfqg, prio, cur_best, cfqd)->count) |
718eee05 CZ |
1938 | return cur_best; |
1939 | ||
1940 | return ASYNC_WORKLOAD; | |
1941 | } | |
1942 | ||
1943 | for (i = 0; i < 3; ++i) { | |
1944 | /* otherwise, select the one with lowest rb_key */ | |
cdb16e8f | 1945 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i, cfqd)); |
718eee05 CZ |
1946 | if (queue && |
1947 | (!key_valid || time_before(queue->rb_key, lowest_key))) { | |
1948 | lowest_key = queue->rb_key; | |
1949 | cur_best = i; | |
1950 | key_valid = true; | |
1951 | } | |
1952 | } | |
1953 | ||
1954 | return cur_best; | |
1955 | } | |
1956 | ||
cdb16e8f | 1957 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
718eee05 CZ |
1958 | { |
1959 | enum wl_prio_t previous_prio = cfqd->serving_prio; | |
1960 | bool prio_changed; | |
1961 | unsigned slice; | |
1962 | unsigned count; | |
cdb16e8f | 1963 | struct cfq_rb_root *st; |
58ff82f3 | 1964 | unsigned group_slice; |
718eee05 | 1965 | |
1fa8f6d6 VG |
1966 | if (!cfqg) { |
1967 | cfqd->serving_prio = IDLE_WORKLOAD; | |
1968 | cfqd->workload_expires = jiffies + 1; | |
1969 | return; | |
1970 | } | |
1971 | ||
718eee05 | 1972 | /* Choose next priority. RT > BE > IDLE */ |
58ff82f3 | 1973 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
718eee05 | 1974 | cfqd->serving_prio = RT_WORKLOAD; |
58ff82f3 | 1975 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
718eee05 CZ |
1976 | cfqd->serving_prio = BE_WORKLOAD; |
1977 | else { | |
1978 | cfqd->serving_prio = IDLE_WORKLOAD; | |
1979 | cfqd->workload_expires = jiffies + 1; | |
1980 | return; | |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * For RT and BE, we have to choose also the type | |
1985 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload | |
1986 | * expiration time | |
1987 | */ | |
1988 | prio_changed = (cfqd->serving_prio != previous_prio); | |
cdb16e8f VG |
1989 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type, |
1990 | cfqd); | |
1991 | count = st->count; | |
718eee05 CZ |
1992 | |
1993 | /* | |
1994 | * If priority didn't change, check workload expiration, | |
1995 | * and that we still have other queues ready | |
1996 | */ | |
1997 | if (!prio_changed && count && | |
1998 | !time_after(jiffies, cfqd->workload_expires)) | |
1999 | return; | |
2000 | ||
2001 | /* otherwise select new workload type */ | |
2002 | cfqd->serving_type = | |
cdb16e8f VG |
2003 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio, prio_changed); |
2004 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type, | |
2005 | cfqd); | |
2006 | count = st->count; | |
718eee05 CZ |
2007 | |
2008 | /* | |
2009 | * the workload slice is computed as a fraction of target latency | |
2010 | * proportional to the number of queues in that workload, over | |
2011 | * all the queues in the same priority class | |
2012 | */ | |
58ff82f3 VG |
2013 | group_slice = cfq_group_slice(cfqd, cfqg); |
2014 | ||
2015 | slice = group_slice * count / | |
2016 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], | |
2017 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); | |
718eee05 CZ |
2018 | |
2019 | if (cfqd->serving_type == ASYNC_WORKLOAD) | |
2020 | /* async workload slice is scaled down according to | |
2021 | * the sync/async slice ratio. */ | |
2022 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; | |
2023 | else | |
2024 | /* sync workload slice is at least 2 * cfq_slice_idle */ | |
2025 | slice = max(slice, 2 * cfqd->cfq_slice_idle); | |
2026 | ||
2027 | slice = max_t(unsigned, slice, CFQ_MIN_TT); | |
2028 | cfqd->workload_expires = jiffies + slice; | |
8e550632 | 2029 | cfqd->noidle_tree_requires_idle = false; |
718eee05 CZ |
2030 | } |
2031 | ||
1fa8f6d6 VG |
2032 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
2033 | { | |
2034 | struct cfq_rb_root *st = &cfqd->grp_service_tree; | |
25bc6b07 | 2035 | struct cfq_group *cfqg; |
1fa8f6d6 VG |
2036 | |
2037 | if (RB_EMPTY_ROOT(&st->rb)) | |
2038 | return NULL; | |
25bc6b07 VG |
2039 | cfqg = cfq_rb_first_group(st); |
2040 | st->active = &cfqg->rb_node; | |
2041 | update_min_vdisktime(st); | |
2042 | return cfqg; | |
1fa8f6d6 VG |
2043 | } |
2044 | ||
cdb16e8f VG |
2045 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
2046 | { | |
1fa8f6d6 VG |
2047 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
2048 | ||
2049 | cfqd->serving_group = cfqg; | |
dae739eb VG |
2050 | |
2051 | /* Restore the workload type data */ | |
2052 | if (cfqg->saved_workload_slice) { | |
2053 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; | |
2054 | cfqd->serving_type = cfqg->saved_workload; | |
2055 | cfqd->serving_prio = cfqg->saved_serving_prio; | |
2056 | } | |
1fa8f6d6 | 2057 | choose_service_tree(cfqd, cfqg); |
cdb16e8f VG |
2058 | } |
2059 | ||
22e2c507 | 2060 | /* |
498d3aa2 JA |
2061 | * Select a queue for service. If we have a current active queue, |
2062 | * check whether to continue servicing it, or retrieve and set a new one. | |
22e2c507 | 2063 | */ |
1b5ed5e1 | 2064 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
1da177e4 | 2065 | { |
a36e71f9 | 2066 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
1da177e4 | 2067 | |
22e2c507 JA |
2068 | cfqq = cfqd->active_queue; |
2069 | if (!cfqq) | |
2070 | goto new_queue; | |
1da177e4 | 2071 | |
f04a6424 VG |
2072 | if (!cfqd->rq_queued) |
2073 | return NULL; | |
22e2c507 | 2074 | /* |
6d048f53 | 2075 | * The active queue has run out of time, expire it and select new. |
22e2c507 | 2076 | */ |
f75edf2d VG |
2077 | if ((cfq_slice_used(cfqq) || cfq_cfqq_wait_busy_done(cfqq)) |
2078 | && !cfq_cfqq_must_dispatch(cfqq)) | |
3b18152c | 2079 | goto expire; |
1da177e4 | 2080 | |
22e2c507 | 2081 | /* |
6d048f53 JA |
2082 | * The active queue has requests and isn't expired, allow it to |
2083 | * dispatch. | |
22e2c507 | 2084 | */ |
dd67d051 | 2085 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 2086 | goto keep_queue; |
6d048f53 | 2087 | |
a36e71f9 JA |
2088 | /* |
2089 | * If another queue has a request waiting within our mean seek | |
2090 | * distance, let it run. The expire code will check for close | |
2091 | * cooperators and put the close queue at the front of the service | |
df5fe3e8 | 2092 | * tree. If possible, merge the expiring queue with the new cfqq. |
a36e71f9 | 2093 | */ |
b3b6d040 | 2094 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
df5fe3e8 JM |
2095 | if (new_cfqq) { |
2096 | if (!cfqq->new_cfqq) | |
2097 | cfq_setup_merge(cfqq, new_cfqq); | |
a36e71f9 | 2098 | goto expire; |
df5fe3e8 | 2099 | } |
a36e71f9 | 2100 | |
6d048f53 JA |
2101 | /* |
2102 | * No requests pending. If the active queue still has requests in | |
2103 | * flight or is idling for a new request, allow either of these | |
2104 | * conditions to happen (or time out) before selecting a new queue. | |
2105 | */ | |
cc197479 | 2106 | if (timer_pending(&cfqd->idle_slice_timer) || |
a6d44e98 | 2107 | (cfqq->dispatched && cfq_should_idle(cfqd, cfqq))) { |
caaa5f9f JA |
2108 | cfqq = NULL; |
2109 | goto keep_queue; | |
22e2c507 JA |
2110 | } |
2111 | ||
3b18152c | 2112 | expire: |
6084cdda | 2113 | cfq_slice_expired(cfqd, 0); |
3b18152c | 2114 | new_queue: |
718eee05 CZ |
2115 | /* |
2116 | * Current queue expired. Check if we have to switch to a new | |
2117 | * service tree | |
2118 | */ | |
2119 | if (!new_cfqq) | |
cdb16e8f | 2120 | cfq_choose_cfqg(cfqd); |
718eee05 | 2121 | |
a36e71f9 | 2122 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
22e2c507 | 2123 | keep_queue: |
3b18152c | 2124 | return cfqq; |
22e2c507 JA |
2125 | } |
2126 | ||
febffd61 | 2127 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
d9e7620e JA |
2128 | { |
2129 | int dispatched = 0; | |
2130 | ||
2131 | while (cfqq->next_rq) { | |
2132 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); | |
2133 | dispatched++; | |
2134 | } | |
2135 | ||
2136 | BUG_ON(!list_empty(&cfqq->fifo)); | |
f04a6424 VG |
2137 | |
2138 | /* By default cfqq is not expired if it is empty. Do it explicitly */ | |
2139 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); | |
d9e7620e JA |
2140 | return dispatched; |
2141 | } | |
2142 | ||
498d3aa2 JA |
2143 | /* |
2144 | * Drain our current requests. Used for barriers and when switching | |
2145 | * io schedulers on-the-fly. | |
2146 | */ | |
d9e7620e | 2147 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
1b5ed5e1 | 2148 | { |
0871714e | 2149 | struct cfq_queue *cfqq; |
d9e7620e | 2150 | int dispatched = 0; |
cdb16e8f | 2151 | |
f04a6424 VG |
2152 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) |
2153 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); | |
1b5ed5e1 | 2154 | |
6084cdda | 2155 | cfq_slice_expired(cfqd, 0); |
1b5ed5e1 TH |
2156 | BUG_ON(cfqd->busy_queues); |
2157 | ||
6923715a | 2158 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
1b5ed5e1 TH |
2159 | return dispatched; |
2160 | } | |
2161 | ||
0b182d61 | 2162 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
2f5cb738 | 2163 | { |
2f5cb738 | 2164 | unsigned int max_dispatch; |
22e2c507 | 2165 | |
5ad531db JA |
2166 | /* |
2167 | * Drain async requests before we start sync IO | |
2168 | */ | |
a6d44e98 | 2169 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_driver[BLK_RW_ASYNC]) |
0b182d61 | 2170 | return false; |
5ad531db | 2171 | |
2f5cb738 JA |
2172 | /* |
2173 | * If this is an async queue and we have sync IO in flight, let it wait | |
2174 | */ | |
2175 | if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq)) | |
0b182d61 | 2176 | return false; |
2f5cb738 JA |
2177 | |
2178 | max_dispatch = cfqd->cfq_quantum; | |
2179 | if (cfq_class_idle(cfqq)) | |
2180 | max_dispatch = 1; | |
b4878f24 | 2181 | |
2f5cb738 JA |
2182 | /* |
2183 | * Does this cfqq already have too much IO in flight? | |
2184 | */ | |
2185 | if (cfqq->dispatched >= max_dispatch) { | |
2186 | /* | |
2187 | * idle queue must always only have a single IO in flight | |
2188 | */ | |
3ed9a296 | 2189 | if (cfq_class_idle(cfqq)) |
0b182d61 | 2190 | return false; |
3ed9a296 | 2191 | |
2f5cb738 JA |
2192 | /* |
2193 | * We have other queues, don't allow more IO from this one | |
2194 | */ | |
2195 | if (cfqd->busy_queues > 1) | |
0b182d61 | 2196 | return false; |
9ede209e | 2197 | |
365722bb | 2198 | /* |
474b18cc | 2199 | * Sole queue user, no limit |
365722bb | 2200 | */ |
474b18cc | 2201 | max_dispatch = -1; |
8e296755 JA |
2202 | } |
2203 | ||
2204 | /* | |
2205 | * Async queues must wait a bit before being allowed dispatch. | |
2206 | * We also ramp up the dispatch depth gradually for async IO, | |
2207 | * based on the last sync IO we serviced | |
2208 | */ | |
963b72fc | 2209 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
8e296755 JA |
2210 | unsigned long last_sync = jiffies - cfqd->last_end_sync_rq; |
2211 | unsigned int depth; | |
365722bb | 2212 | |
61f0c1dc | 2213 | depth = last_sync / cfqd->cfq_slice[1]; |
e00c54c3 JA |
2214 | if (!depth && !cfqq->dispatched) |
2215 | depth = 1; | |
8e296755 JA |
2216 | if (depth < max_dispatch) |
2217 | max_dispatch = depth; | |
2f5cb738 | 2218 | } |
3ed9a296 | 2219 | |
0b182d61 JA |
2220 | /* |
2221 | * If we're below the current max, allow a dispatch | |
2222 | */ | |
2223 | return cfqq->dispatched < max_dispatch; | |
2224 | } | |
2225 | ||
2226 | /* | |
2227 | * Dispatch a request from cfqq, moving them to the request queue | |
2228 | * dispatch list. | |
2229 | */ | |
2230 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
2231 | { | |
2232 | struct request *rq; | |
2233 | ||
2234 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); | |
2235 | ||
2236 | if (!cfq_may_dispatch(cfqd, cfqq)) | |
2237 | return false; | |
2238 | ||
2239 | /* | |
2240 | * follow expired path, else get first next available | |
2241 | */ | |
2242 | rq = cfq_check_fifo(cfqq); | |
2243 | if (!rq) | |
2244 | rq = cfqq->next_rq; | |
2245 | ||
2246 | /* | |
2247 | * insert request into driver dispatch list | |
2248 | */ | |
2249 | cfq_dispatch_insert(cfqd->queue, rq); | |
2250 | ||
2251 | if (!cfqd->active_cic) { | |
2252 | struct cfq_io_context *cic = RQ_CIC(rq); | |
2253 | ||
2254 | atomic_long_inc(&cic->ioc->refcount); | |
2255 | cfqd->active_cic = cic; | |
2256 | } | |
2257 | ||
2258 | return true; | |
2259 | } | |
2260 | ||
2261 | /* | |
2262 | * Find the cfqq that we need to service and move a request from that to the | |
2263 | * dispatch list | |
2264 | */ | |
2265 | static int cfq_dispatch_requests(struct request_queue *q, int force) | |
2266 | { | |
2267 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
2268 | struct cfq_queue *cfqq; | |
2269 | ||
2270 | if (!cfqd->busy_queues) | |
2271 | return 0; | |
2272 | ||
2273 | if (unlikely(force)) | |
2274 | return cfq_forced_dispatch(cfqd); | |
2275 | ||
2276 | cfqq = cfq_select_queue(cfqd); | |
2277 | if (!cfqq) | |
8e296755 JA |
2278 | return 0; |
2279 | ||
2f5cb738 | 2280 | /* |
0b182d61 | 2281 | * Dispatch a request from this cfqq, if it is allowed |
2f5cb738 | 2282 | */ |
0b182d61 JA |
2283 | if (!cfq_dispatch_request(cfqd, cfqq)) |
2284 | return 0; | |
2285 | ||
2f5cb738 | 2286 | cfqq->slice_dispatch++; |
b029195d | 2287 | cfq_clear_cfqq_must_dispatch(cfqq); |
22e2c507 | 2288 | |
2f5cb738 JA |
2289 | /* |
2290 | * expire an async queue immediately if it has used up its slice. idle | |
2291 | * queue always expire after 1 dispatch round. | |
2292 | */ | |
2293 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && | |
2294 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || | |
2295 | cfq_class_idle(cfqq))) { | |
2296 | cfqq->slice_end = jiffies + 1; | |
2297 | cfq_slice_expired(cfqd, 0); | |
1da177e4 LT |
2298 | } |
2299 | ||
b217a903 | 2300 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
2f5cb738 | 2301 | return 1; |
1da177e4 LT |
2302 | } |
2303 | ||
1da177e4 | 2304 | /* |
5e705374 JA |
2305 | * task holds one reference to the queue, dropped when task exits. each rq |
2306 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
1da177e4 | 2307 | * |
b1c35769 | 2308 | * Each cfq queue took a reference on the parent group. Drop it now. |
1da177e4 LT |
2309 | * queue lock must be held here. |
2310 | */ | |
2311 | static void cfq_put_queue(struct cfq_queue *cfqq) | |
2312 | { | |
22e2c507 | 2313 | struct cfq_data *cfqd = cfqq->cfqd; |
b1c35769 | 2314 | struct cfq_group *cfqg; |
22e2c507 JA |
2315 | |
2316 | BUG_ON(atomic_read(&cfqq->ref) <= 0); | |
1da177e4 LT |
2317 | |
2318 | if (!atomic_dec_and_test(&cfqq->ref)) | |
2319 | return; | |
2320 | ||
7b679138 | 2321 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
1da177e4 | 2322 | BUG_ON(rb_first(&cfqq->sort_list)); |
22e2c507 | 2323 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
b1c35769 | 2324 | cfqg = cfqq->cfqg; |
1da177e4 | 2325 | |
28f95cbc | 2326 | if (unlikely(cfqd->active_queue == cfqq)) { |
6084cdda | 2327 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2328 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2329 | } |
22e2c507 | 2330 | |
f04a6424 | 2331 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
1da177e4 | 2332 | kmem_cache_free(cfq_pool, cfqq); |
b1c35769 | 2333 | cfq_put_cfqg(cfqg); |
1da177e4 LT |
2334 | } |
2335 | ||
d6de8be7 JA |
2336 | /* |
2337 | * Must always be called with the rcu_read_lock() held | |
2338 | */ | |
07416d29 JA |
2339 | static void |
2340 | __call_for_each_cic(struct io_context *ioc, | |
2341 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
2342 | { | |
2343 | struct cfq_io_context *cic; | |
2344 | struct hlist_node *n; | |
2345 | ||
2346 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) | |
2347 | func(ioc, cic); | |
2348 | } | |
2349 | ||
4ac845a2 | 2350 | /* |
34e6bbf2 | 2351 | * Call func for each cic attached to this ioc. |
4ac845a2 | 2352 | */ |
34e6bbf2 | 2353 | static void |
4ac845a2 JA |
2354 | call_for_each_cic(struct io_context *ioc, |
2355 | void (*func)(struct io_context *, struct cfq_io_context *)) | |
1da177e4 | 2356 | { |
4ac845a2 | 2357 | rcu_read_lock(); |
07416d29 | 2358 | __call_for_each_cic(ioc, func); |
4ac845a2 | 2359 | rcu_read_unlock(); |
34e6bbf2 FC |
2360 | } |
2361 | ||
2362 | static void cfq_cic_free_rcu(struct rcu_head *head) | |
2363 | { | |
2364 | struct cfq_io_context *cic; | |
2365 | ||
2366 | cic = container_of(head, struct cfq_io_context, rcu_head); | |
2367 | ||
2368 | kmem_cache_free(cfq_ioc_pool, cic); | |
245b2e70 | 2369 | elv_ioc_count_dec(cfq_ioc_count); |
34e6bbf2 | 2370 | |
9a11b4ed JA |
2371 | if (ioc_gone) { |
2372 | /* | |
2373 | * CFQ scheduler is exiting, grab exit lock and check | |
2374 | * the pending io context count. If it hits zero, | |
2375 | * complete ioc_gone and set it back to NULL | |
2376 | */ | |
2377 | spin_lock(&ioc_gone_lock); | |
245b2e70 | 2378 | if (ioc_gone && !elv_ioc_count_read(cfq_ioc_count)) { |
9a11b4ed JA |
2379 | complete(ioc_gone); |
2380 | ioc_gone = NULL; | |
2381 | } | |
2382 | spin_unlock(&ioc_gone_lock); | |
2383 | } | |
34e6bbf2 | 2384 | } |
4ac845a2 | 2385 | |
34e6bbf2 FC |
2386 | static void cfq_cic_free(struct cfq_io_context *cic) |
2387 | { | |
2388 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); | |
4ac845a2 JA |
2389 | } |
2390 | ||
2391 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) | |
2392 | { | |
2393 | unsigned long flags; | |
2394 | ||
2395 | BUG_ON(!cic->dead_key); | |
2396 | ||
2397 | spin_lock_irqsave(&ioc->lock, flags); | |
2398 | radix_tree_delete(&ioc->radix_root, cic->dead_key); | |
ffc4e759 | 2399 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
2400 | spin_unlock_irqrestore(&ioc->lock, flags); |
2401 | ||
34e6bbf2 | 2402 | cfq_cic_free(cic); |
4ac845a2 JA |
2403 | } |
2404 | ||
d6de8be7 JA |
2405 | /* |
2406 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. | |
2407 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), | |
2408 | * and ->trim() which is called with the task lock held | |
2409 | */ | |
4ac845a2 JA |
2410 | static void cfq_free_io_context(struct io_context *ioc) |
2411 | { | |
4ac845a2 | 2412 | /* |
34e6bbf2 FC |
2413 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
2414 | * so no more cic's are allowed to be linked into this ioc. So it | |
2415 | * should be ok to iterate over the known list, we will see all cic's | |
2416 | * since no new ones are added. | |
4ac845a2 | 2417 | */ |
07416d29 | 2418 | __call_for_each_cic(ioc, cic_free_func); |
1da177e4 LT |
2419 | } |
2420 | ||
89850f7e | 2421 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
1da177e4 | 2422 | { |
df5fe3e8 JM |
2423 | struct cfq_queue *__cfqq, *next; |
2424 | ||
28f95cbc | 2425 | if (unlikely(cfqq == cfqd->active_queue)) { |
6084cdda | 2426 | __cfq_slice_expired(cfqd, cfqq, 0); |
23e018a1 | 2427 | cfq_schedule_dispatch(cfqd); |
28f95cbc | 2428 | } |
22e2c507 | 2429 | |
df5fe3e8 JM |
2430 | /* |
2431 | * If this queue was scheduled to merge with another queue, be | |
2432 | * sure to drop the reference taken on that queue (and others in | |
2433 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. | |
2434 | */ | |
2435 | __cfqq = cfqq->new_cfqq; | |
2436 | while (__cfqq) { | |
2437 | if (__cfqq == cfqq) { | |
2438 | WARN(1, "cfqq->new_cfqq loop detected\n"); | |
2439 | break; | |
2440 | } | |
2441 | next = __cfqq->new_cfqq; | |
2442 | cfq_put_queue(__cfqq); | |
2443 | __cfqq = next; | |
2444 | } | |
2445 | ||
89850f7e JA |
2446 | cfq_put_queue(cfqq); |
2447 | } | |
22e2c507 | 2448 | |
89850f7e JA |
2449 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
2450 | struct cfq_io_context *cic) | |
2451 | { | |
4faa3c81 FC |
2452 | struct io_context *ioc = cic->ioc; |
2453 | ||
fc46379d | 2454 | list_del_init(&cic->queue_list); |
4ac845a2 JA |
2455 | |
2456 | /* | |
2457 | * Make sure key == NULL is seen for dead queues | |
2458 | */ | |
fc46379d | 2459 | smp_wmb(); |
4ac845a2 | 2460 | cic->dead_key = (unsigned long) cic->key; |
fc46379d JA |
2461 | cic->key = NULL; |
2462 | ||
4faa3c81 FC |
2463 | if (ioc->ioc_data == cic) |
2464 | rcu_assign_pointer(ioc->ioc_data, NULL); | |
2465 | ||
ff6657c6 JA |
2466 | if (cic->cfqq[BLK_RW_ASYNC]) { |
2467 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); | |
2468 | cic->cfqq[BLK_RW_ASYNC] = NULL; | |
12a05732 AV |
2469 | } |
2470 | ||
ff6657c6 JA |
2471 | if (cic->cfqq[BLK_RW_SYNC]) { |
2472 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); | |
2473 | cic->cfqq[BLK_RW_SYNC] = NULL; | |
12a05732 | 2474 | } |
89850f7e JA |
2475 | } |
2476 | ||
4ac845a2 JA |
2477 | static void cfq_exit_single_io_context(struct io_context *ioc, |
2478 | struct cfq_io_context *cic) | |
89850f7e JA |
2479 | { |
2480 | struct cfq_data *cfqd = cic->key; | |
2481 | ||
89850f7e | 2482 | if (cfqd) { |
165125e1 | 2483 | struct request_queue *q = cfqd->queue; |
4ac845a2 | 2484 | unsigned long flags; |
89850f7e | 2485 | |
4ac845a2 | 2486 | spin_lock_irqsave(q->queue_lock, flags); |
62c1fe9d JA |
2487 | |
2488 | /* | |
2489 | * Ensure we get a fresh copy of the ->key to prevent | |
2490 | * race between exiting task and queue | |
2491 | */ | |
2492 | smp_read_barrier_depends(); | |
2493 | if (cic->key) | |
2494 | __cfq_exit_single_io_context(cfqd, cic); | |
2495 | ||
4ac845a2 | 2496 | spin_unlock_irqrestore(q->queue_lock, flags); |
89850f7e | 2497 | } |
1da177e4 LT |
2498 | } |
2499 | ||
498d3aa2 JA |
2500 | /* |
2501 | * The process that ioc belongs to has exited, we need to clean up | |
2502 | * and put the internal structures we have that belongs to that process. | |
2503 | */ | |
e2d74ac0 | 2504 | static void cfq_exit_io_context(struct io_context *ioc) |
1da177e4 | 2505 | { |
4ac845a2 | 2506 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
1da177e4 LT |
2507 | } |
2508 | ||
22e2c507 | 2509 | static struct cfq_io_context * |
8267e268 | 2510 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 2511 | { |
b5deef90 | 2512 | struct cfq_io_context *cic; |
1da177e4 | 2513 | |
94f6030c CL |
2514 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
2515 | cfqd->queue->node); | |
1da177e4 | 2516 | if (cic) { |
22e2c507 | 2517 | cic->last_end_request = jiffies; |
553698f9 | 2518 | INIT_LIST_HEAD(&cic->queue_list); |
ffc4e759 | 2519 | INIT_HLIST_NODE(&cic->cic_list); |
22e2c507 JA |
2520 | cic->dtor = cfq_free_io_context; |
2521 | cic->exit = cfq_exit_io_context; | |
245b2e70 | 2522 | elv_ioc_count_inc(cfq_ioc_count); |
1da177e4 LT |
2523 | } |
2524 | ||
2525 | return cic; | |
2526 | } | |
2527 | ||
fd0928df | 2528 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
22e2c507 JA |
2529 | { |
2530 | struct task_struct *tsk = current; | |
2531 | int ioprio_class; | |
2532 | ||
3b18152c | 2533 | if (!cfq_cfqq_prio_changed(cfqq)) |
22e2c507 JA |
2534 | return; |
2535 | ||
fd0928df | 2536 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
22e2c507 | 2537 | switch (ioprio_class) { |
fe094d98 JA |
2538 | default: |
2539 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); | |
2540 | case IOPRIO_CLASS_NONE: | |
2541 | /* | |
6d63c275 | 2542 | * no prio set, inherit CPU scheduling settings |
fe094d98 JA |
2543 | */ |
2544 | cfqq->ioprio = task_nice_ioprio(tsk); | |
6d63c275 | 2545 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
fe094d98 JA |
2546 | break; |
2547 | case IOPRIO_CLASS_RT: | |
2548 | cfqq->ioprio = task_ioprio(ioc); | |
2549 | cfqq->ioprio_class = IOPRIO_CLASS_RT; | |
2550 | break; | |
2551 | case IOPRIO_CLASS_BE: | |
2552 | cfqq->ioprio = task_ioprio(ioc); | |
2553 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
2554 | break; | |
2555 | case IOPRIO_CLASS_IDLE: | |
2556 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; | |
2557 | cfqq->ioprio = 7; | |
2558 | cfq_clear_cfqq_idle_window(cfqq); | |
2559 | break; | |
22e2c507 JA |
2560 | } |
2561 | ||
2562 | /* | |
2563 | * keep track of original prio settings in case we have to temporarily | |
2564 | * elevate the priority of this queue | |
2565 | */ | |
2566 | cfqq->org_ioprio = cfqq->ioprio; | |
2567 | cfqq->org_ioprio_class = cfqq->ioprio_class; | |
3b18152c | 2568 | cfq_clear_cfqq_prio_changed(cfqq); |
22e2c507 JA |
2569 | } |
2570 | ||
febffd61 | 2571 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
22e2c507 | 2572 | { |
478a82b0 AV |
2573 | struct cfq_data *cfqd = cic->key; |
2574 | struct cfq_queue *cfqq; | |
c1b707d2 | 2575 | unsigned long flags; |
35e6077c | 2576 | |
caaa5f9f JA |
2577 | if (unlikely(!cfqd)) |
2578 | return; | |
2579 | ||
c1b707d2 | 2580 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
caaa5f9f | 2581 | |
ff6657c6 | 2582 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
caaa5f9f JA |
2583 | if (cfqq) { |
2584 | struct cfq_queue *new_cfqq; | |
ff6657c6 JA |
2585 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc, |
2586 | GFP_ATOMIC); | |
caaa5f9f | 2587 | if (new_cfqq) { |
ff6657c6 | 2588 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
caaa5f9f JA |
2589 | cfq_put_queue(cfqq); |
2590 | } | |
22e2c507 | 2591 | } |
caaa5f9f | 2592 | |
ff6657c6 | 2593 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
caaa5f9f JA |
2594 | if (cfqq) |
2595 | cfq_mark_cfqq_prio_changed(cfqq); | |
2596 | ||
c1b707d2 | 2597 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
22e2c507 JA |
2598 | } |
2599 | ||
fc46379d | 2600 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
22e2c507 | 2601 | { |
4ac845a2 | 2602 | call_for_each_cic(ioc, changed_ioprio); |
fc46379d | 2603 | ioc->ioprio_changed = 0; |
22e2c507 JA |
2604 | } |
2605 | ||
d5036d77 | 2606 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
a6151c3a | 2607 | pid_t pid, bool is_sync) |
d5036d77 JA |
2608 | { |
2609 | RB_CLEAR_NODE(&cfqq->rb_node); | |
2610 | RB_CLEAR_NODE(&cfqq->p_node); | |
2611 | INIT_LIST_HEAD(&cfqq->fifo); | |
2612 | ||
2613 | atomic_set(&cfqq->ref, 0); | |
2614 | cfqq->cfqd = cfqd; | |
2615 | ||
2616 | cfq_mark_cfqq_prio_changed(cfqq); | |
2617 | ||
2618 | if (is_sync) { | |
2619 | if (!cfq_class_idle(cfqq)) | |
2620 | cfq_mark_cfqq_idle_window(cfqq); | |
2621 | cfq_mark_cfqq_sync(cfqq); | |
2622 | } | |
2623 | cfqq->pid = pid; | |
2624 | } | |
2625 | ||
24610333 VG |
2626 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2627 | static void changed_cgroup(struct io_context *ioc, struct cfq_io_context *cic) | |
2628 | { | |
2629 | struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1); | |
2630 | struct cfq_data *cfqd = cic->key; | |
2631 | unsigned long flags; | |
2632 | struct request_queue *q; | |
2633 | ||
2634 | if (unlikely(!cfqd)) | |
2635 | return; | |
2636 | ||
2637 | q = cfqd->queue; | |
2638 | ||
2639 | spin_lock_irqsave(q->queue_lock, flags); | |
2640 | ||
2641 | if (sync_cfqq) { | |
2642 | /* | |
2643 | * Drop reference to sync queue. A new sync queue will be | |
2644 | * assigned in new group upon arrival of a fresh request. | |
2645 | */ | |
2646 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); | |
2647 | cic_set_cfqq(cic, NULL, 1); | |
2648 | cfq_put_queue(sync_cfqq); | |
2649 | } | |
2650 | ||
2651 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2652 | } | |
2653 | ||
2654 | static void cfq_ioc_set_cgroup(struct io_context *ioc) | |
2655 | { | |
2656 | call_for_each_cic(ioc, changed_cgroup); | |
2657 | ioc->cgroup_changed = 0; | |
2658 | } | |
2659 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ | |
2660 | ||
22e2c507 | 2661 | static struct cfq_queue * |
a6151c3a | 2662 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, |
fd0928df | 2663 | struct io_context *ioc, gfp_t gfp_mask) |
22e2c507 | 2664 | { |
22e2c507 | 2665 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
91fac317 | 2666 | struct cfq_io_context *cic; |
cdb16e8f | 2667 | struct cfq_group *cfqg; |
22e2c507 JA |
2668 | |
2669 | retry: | |
cdb16e8f | 2670 | cfqg = cfq_get_cfqg(cfqd, 1); |
4ac845a2 | 2671 | cic = cfq_cic_lookup(cfqd, ioc); |
91fac317 VT |
2672 | /* cic always exists here */ |
2673 | cfqq = cic_to_cfqq(cic, is_sync); | |
22e2c507 | 2674 | |
6118b70b JA |
2675 | /* |
2676 | * Always try a new alloc if we fell back to the OOM cfqq | |
2677 | * originally, since it should just be a temporary situation. | |
2678 | */ | |
2679 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { | |
2680 | cfqq = NULL; | |
22e2c507 JA |
2681 | if (new_cfqq) { |
2682 | cfqq = new_cfqq; | |
2683 | new_cfqq = NULL; | |
2684 | } else if (gfp_mask & __GFP_WAIT) { | |
2685 | spin_unlock_irq(cfqd->queue->queue_lock); | |
94f6030c | 2686 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
6118b70b | 2687 | gfp_mask | __GFP_ZERO, |
94f6030c | 2688 | cfqd->queue->node); |
22e2c507 | 2689 | spin_lock_irq(cfqd->queue->queue_lock); |
6118b70b JA |
2690 | if (new_cfqq) |
2691 | goto retry; | |
22e2c507 | 2692 | } else { |
94f6030c CL |
2693 | cfqq = kmem_cache_alloc_node(cfq_pool, |
2694 | gfp_mask | __GFP_ZERO, | |
2695 | cfqd->queue->node); | |
22e2c507 JA |
2696 | } |
2697 | ||
6118b70b JA |
2698 | if (cfqq) { |
2699 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); | |
2700 | cfq_init_prio_data(cfqq, ioc); | |
cdb16e8f | 2701 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
6118b70b JA |
2702 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
2703 | } else | |
2704 | cfqq = &cfqd->oom_cfqq; | |
22e2c507 JA |
2705 | } |
2706 | ||
2707 | if (new_cfqq) | |
2708 | kmem_cache_free(cfq_pool, new_cfqq); | |
2709 | ||
22e2c507 JA |
2710 | return cfqq; |
2711 | } | |
2712 | ||
c2dea2d1 VT |
2713 | static struct cfq_queue ** |
2714 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) | |
2715 | { | |
fe094d98 | 2716 | switch (ioprio_class) { |
c2dea2d1 VT |
2717 | case IOPRIO_CLASS_RT: |
2718 | return &cfqd->async_cfqq[0][ioprio]; | |
2719 | case IOPRIO_CLASS_BE: | |
2720 | return &cfqd->async_cfqq[1][ioprio]; | |
2721 | case IOPRIO_CLASS_IDLE: | |
2722 | return &cfqd->async_idle_cfqq; | |
2723 | default: | |
2724 | BUG(); | |
2725 | } | |
2726 | } | |
2727 | ||
15c31be4 | 2728 | static struct cfq_queue * |
a6151c3a | 2729 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc, |
15c31be4 JA |
2730 | gfp_t gfp_mask) |
2731 | { | |
fd0928df JA |
2732 | const int ioprio = task_ioprio(ioc); |
2733 | const int ioprio_class = task_ioprio_class(ioc); | |
c2dea2d1 | 2734 | struct cfq_queue **async_cfqq = NULL; |
15c31be4 JA |
2735 | struct cfq_queue *cfqq = NULL; |
2736 | ||
c2dea2d1 VT |
2737 | if (!is_sync) { |
2738 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); | |
2739 | cfqq = *async_cfqq; | |
2740 | } | |
2741 | ||
6118b70b | 2742 | if (!cfqq) |
fd0928df | 2743 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
15c31be4 JA |
2744 | |
2745 | /* | |
2746 | * pin the queue now that it's allocated, scheduler exit will prune it | |
2747 | */ | |
c2dea2d1 | 2748 | if (!is_sync && !(*async_cfqq)) { |
15c31be4 | 2749 | atomic_inc(&cfqq->ref); |
c2dea2d1 | 2750 | *async_cfqq = cfqq; |
15c31be4 JA |
2751 | } |
2752 | ||
2753 | atomic_inc(&cfqq->ref); | |
2754 | return cfqq; | |
2755 | } | |
2756 | ||
498d3aa2 JA |
2757 | /* |
2758 | * We drop cfq io contexts lazily, so we may find a dead one. | |
2759 | */ | |
dbecf3ab | 2760 | static void |
4ac845a2 JA |
2761 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
2762 | struct cfq_io_context *cic) | |
dbecf3ab | 2763 | { |
4ac845a2 JA |
2764 | unsigned long flags; |
2765 | ||
fc46379d | 2766 | WARN_ON(!list_empty(&cic->queue_list)); |
597bc485 | 2767 | |
4ac845a2 JA |
2768 | spin_lock_irqsave(&ioc->lock, flags); |
2769 | ||
4faa3c81 | 2770 | BUG_ON(ioc->ioc_data == cic); |
597bc485 | 2771 | |
4ac845a2 | 2772 | radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd); |
ffc4e759 | 2773 | hlist_del_rcu(&cic->cic_list); |
4ac845a2 JA |
2774 | spin_unlock_irqrestore(&ioc->lock, flags); |
2775 | ||
2776 | cfq_cic_free(cic); | |
dbecf3ab OH |
2777 | } |
2778 | ||
e2d74ac0 | 2779 | static struct cfq_io_context * |
4ac845a2 | 2780 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
e2d74ac0 | 2781 | { |
e2d74ac0 | 2782 | struct cfq_io_context *cic; |
d6de8be7 | 2783 | unsigned long flags; |
4ac845a2 | 2784 | void *k; |
e2d74ac0 | 2785 | |
91fac317 VT |
2786 | if (unlikely(!ioc)) |
2787 | return NULL; | |
2788 | ||
d6de8be7 JA |
2789 | rcu_read_lock(); |
2790 | ||
597bc485 JA |
2791 | /* |
2792 | * we maintain a last-hit cache, to avoid browsing over the tree | |
2793 | */ | |
4ac845a2 | 2794 | cic = rcu_dereference(ioc->ioc_data); |
d6de8be7 JA |
2795 | if (cic && cic->key == cfqd) { |
2796 | rcu_read_unlock(); | |
597bc485 | 2797 | return cic; |
d6de8be7 | 2798 | } |
597bc485 | 2799 | |
4ac845a2 | 2800 | do { |
4ac845a2 JA |
2801 | cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd); |
2802 | rcu_read_unlock(); | |
2803 | if (!cic) | |
2804 | break; | |
be3b0753 OH |
2805 | /* ->key must be copied to avoid race with cfq_exit_queue() */ |
2806 | k = cic->key; | |
2807 | if (unlikely(!k)) { | |
4ac845a2 | 2808 | cfq_drop_dead_cic(cfqd, ioc, cic); |
d6de8be7 | 2809 | rcu_read_lock(); |
4ac845a2 | 2810 | continue; |
dbecf3ab | 2811 | } |
e2d74ac0 | 2812 | |
d6de8be7 | 2813 | spin_lock_irqsave(&ioc->lock, flags); |
4ac845a2 | 2814 | rcu_assign_pointer(ioc->ioc_data, cic); |
d6de8be7 | 2815 | spin_unlock_irqrestore(&ioc->lock, flags); |
4ac845a2 JA |
2816 | break; |
2817 | } while (1); | |
e2d74ac0 | 2818 | |
4ac845a2 | 2819 | return cic; |
e2d74ac0 JA |
2820 | } |
2821 | ||
4ac845a2 JA |
2822 | /* |
2823 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup | |
2824 | * the process specific cfq io context when entered from the block layer. | |
2825 | * Also adds the cic to a per-cfqd list, used when this queue is removed. | |
2826 | */ | |
febffd61 JA |
2827 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
2828 | struct cfq_io_context *cic, gfp_t gfp_mask) | |
e2d74ac0 | 2829 | { |
0261d688 | 2830 | unsigned long flags; |
4ac845a2 | 2831 | int ret; |
e2d74ac0 | 2832 | |
4ac845a2 JA |
2833 | ret = radix_tree_preload(gfp_mask); |
2834 | if (!ret) { | |
2835 | cic->ioc = ioc; | |
2836 | cic->key = cfqd; | |
e2d74ac0 | 2837 | |
4ac845a2 JA |
2838 | spin_lock_irqsave(&ioc->lock, flags); |
2839 | ret = radix_tree_insert(&ioc->radix_root, | |
2840 | (unsigned long) cfqd, cic); | |
ffc4e759 JA |
2841 | if (!ret) |
2842 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); | |
4ac845a2 | 2843 | spin_unlock_irqrestore(&ioc->lock, flags); |
e2d74ac0 | 2844 | |
4ac845a2 JA |
2845 | radix_tree_preload_end(); |
2846 | ||
2847 | if (!ret) { | |
2848 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); | |
2849 | list_add(&cic->queue_list, &cfqd->cic_list); | |
2850 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
2851 | } | |
e2d74ac0 JA |
2852 | } |
2853 | ||
4ac845a2 JA |
2854 | if (ret) |
2855 | printk(KERN_ERR "cfq: cic link failed!\n"); | |
fc46379d | 2856 | |
4ac845a2 | 2857 | return ret; |
e2d74ac0 JA |
2858 | } |
2859 | ||
1da177e4 LT |
2860 | /* |
2861 | * Setup general io context and cfq io context. There can be several cfq | |
2862 | * io contexts per general io context, if this process is doing io to more | |
e2d74ac0 | 2863 | * than one device managed by cfq. |
1da177e4 LT |
2864 | */ |
2865 | static struct cfq_io_context * | |
e2d74ac0 | 2866 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
1da177e4 | 2867 | { |
22e2c507 | 2868 | struct io_context *ioc = NULL; |
1da177e4 | 2869 | struct cfq_io_context *cic; |
1da177e4 | 2870 | |
22e2c507 | 2871 | might_sleep_if(gfp_mask & __GFP_WAIT); |
1da177e4 | 2872 | |
b5deef90 | 2873 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
1da177e4 LT |
2874 | if (!ioc) |
2875 | return NULL; | |
2876 | ||
4ac845a2 | 2877 | cic = cfq_cic_lookup(cfqd, ioc); |
e2d74ac0 JA |
2878 | if (cic) |
2879 | goto out; | |
1da177e4 | 2880 | |
e2d74ac0 JA |
2881 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
2882 | if (cic == NULL) | |
2883 | goto err; | |
1da177e4 | 2884 | |
4ac845a2 JA |
2885 | if (cfq_cic_link(cfqd, ioc, cic, gfp_mask)) |
2886 | goto err_free; | |
2887 | ||
1da177e4 | 2888 | out: |
fc46379d JA |
2889 | smp_read_barrier_depends(); |
2890 | if (unlikely(ioc->ioprio_changed)) | |
2891 | cfq_ioc_set_ioprio(ioc); | |
2892 | ||
24610333 VG |
2893 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
2894 | if (unlikely(ioc->cgroup_changed)) | |
2895 | cfq_ioc_set_cgroup(ioc); | |
2896 | #endif | |
1da177e4 | 2897 | return cic; |
4ac845a2 JA |
2898 | err_free: |
2899 | cfq_cic_free(cic); | |
1da177e4 LT |
2900 | err: |
2901 | put_io_context(ioc); | |
2902 | return NULL; | |
2903 | } | |
2904 | ||
22e2c507 JA |
2905 | static void |
2906 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) | |
1da177e4 | 2907 | { |
aaf1228d JA |
2908 | unsigned long elapsed = jiffies - cic->last_end_request; |
2909 | unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); | |
db3b5848 | 2910 | |
22e2c507 JA |
2911 | cic->ttime_samples = (7*cic->ttime_samples + 256) / 8; |
2912 | cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8; | |
2913 | cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples; | |
2914 | } | |
1da177e4 | 2915 | |
206dc69b | 2916 | static void |
b2c18e1e | 2917 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
6d048f53 | 2918 | struct request *rq) |
206dc69b JA |
2919 | { |
2920 | sector_t sdist; | |
2921 | u64 total; | |
2922 | ||
b2c18e1e | 2923 | if (!cfqq->last_request_pos) |
4d00aa47 | 2924 | sdist = 0; |
b2c18e1e JM |
2925 | else if (cfqq->last_request_pos < blk_rq_pos(rq)) |
2926 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; | |
206dc69b | 2927 | else |
b2c18e1e | 2928 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); |
206dc69b JA |
2929 | |
2930 | /* | |
2931 | * Don't allow the seek distance to get too large from the | |
2932 | * odd fragment, pagein, etc | |
2933 | */ | |
b2c18e1e JM |
2934 | if (cfqq->seek_samples <= 60) /* second&third seek */ |
2935 | sdist = min(sdist, (cfqq->seek_mean * 4) + 2*1024*1024); | |
206dc69b | 2936 | else |
b2c18e1e | 2937 | sdist = min(sdist, (cfqq->seek_mean * 4) + 2*1024*64); |
206dc69b | 2938 | |
b2c18e1e JM |
2939 | cfqq->seek_samples = (7*cfqq->seek_samples + 256) / 8; |
2940 | cfqq->seek_total = (7*cfqq->seek_total + (u64)256*sdist) / 8; | |
2941 | total = cfqq->seek_total + (cfqq->seek_samples/2); | |
2942 | do_div(total, cfqq->seek_samples); | |
2943 | cfqq->seek_mean = (sector_t)total; | |
e6c5bc73 JM |
2944 | |
2945 | /* | |
2946 | * If this cfqq is shared between multiple processes, check to | |
2947 | * make sure that those processes are still issuing I/Os within | |
2948 | * the mean seek distance. If not, it may be time to break the | |
2949 | * queues apart again. | |
2950 | */ | |
2951 | if (cfq_cfqq_coop(cfqq)) { | |
2952 | if (CFQQ_SEEKY(cfqq) && !cfqq->seeky_start) | |
2953 | cfqq->seeky_start = jiffies; | |
2954 | else if (!CFQQ_SEEKY(cfqq)) | |
2955 | cfqq->seeky_start = 0; | |
2956 | } | |
206dc69b | 2957 | } |
1da177e4 | 2958 | |
22e2c507 JA |
2959 | /* |
2960 | * Disable idle window if the process thinks too long or seeks so much that | |
2961 | * it doesn't matter | |
2962 | */ | |
2963 | static void | |
2964 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, | |
2965 | struct cfq_io_context *cic) | |
2966 | { | |
7b679138 | 2967 | int old_idle, enable_idle; |
1be92f2f | 2968 | |
0871714e JA |
2969 | /* |
2970 | * Don't idle for async or idle io prio class | |
2971 | */ | |
2972 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) | |
1be92f2f JA |
2973 | return; |
2974 | ||
c265a7f4 | 2975 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
1da177e4 | 2976 | |
76280aff CZ |
2977 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
2978 | cfq_mark_cfqq_deep(cfqq); | |
2979 | ||
66dac98e | 2980 | if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
76280aff CZ |
2981 | (!cfq_cfqq_deep(cfqq) && sample_valid(cfqq->seek_samples) |
2982 | && CFQQ_SEEKY(cfqq))) | |
22e2c507 JA |
2983 | enable_idle = 0; |
2984 | else if (sample_valid(cic->ttime_samples)) { | |
718eee05 | 2985 | if (cic->ttime_mean > cfqd->cfq_slice_idle) |
22e2c507 JA |
2986 | enable_idle = 0; |
2987 | else | |
2988 | enable_idle = 1; | |
1da177e4 LT |
2989 | } |
2990 | ||
7b679138 JA |
2991 | if (old_idle != enable_idle) { |
2992 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); | |
2993 | if (enable_idle) | |
2994 | cfq_mark_cfqq_idle_window(cfqq); | |
2995 | else | |
2996 | cfq_clear_cfqq_idle_window(cfqq); | |
2997 | } | |
22e2c507 | 2998 | } |
1da177e4 | 2999 | |
22e2c507 JA |
3000 | /* |
3001 | * Check if new_cfqq should preempt the currently active queue. Return 0 for | |
3002 | * no or if we aren't sure, a 1 will cause a preempt. | |
3003 | */ | |
a6151c3a | 3004 | static bool |
22e2c507 | 3005 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
5e705374 | 3006 | struct request *rq) |
22e2c507 | 3007 | { |
6d048f53 | 3008 | struct cfq_queue *cfqq; |
22e2c507 | 3009 | |
6d048f53 JA |
3010 | cfqq = cfqd->active_queue; |
3011 | if (!cfqq) | |
a6151c3a | 3012 | return false; |
22e2c507 | 3013 | |
6d048f53 | 3014 | if (cfq_class_idle(new_cfqq)) |
a6151c3a | 3015 | return false; |
22e2c507 JA |
3016 | |
3017 | if (cfq_class_idle(cfqq)) | |
a6151c3a | 3018 | return true; |
1e3335de | 3019 | |
374f84ac JA |
3020 | /* |
3021 | * if the new request is sync, but the currently running queue is | |
3022 | * not, let the sync request have priority. | |
3023 | */ | |
5e705374 | 3024 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
a6151c3a | 3025 | return true; |
1e3335de | 3026 | |
8682e1f1 VG |
3027 | if (new_cfqq->cfqg != cfqq->cfqg) |
3028 | return false; | |
3029 | ||
3030 | if (cfq_slice_used(cfqq)) | |
3031 | return true; | |
3032 | ||
3033 | /* Allow preemption only if we are idling on sync-noidle tree */ | |
3034 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && | |
3035 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && | |
3036 | new_cfqq->service_tree->count == 2 && | |
3037 | RB_EMPTY_ROOT(&cfqq->sort_list)) | |
3038 | return true; | |
3039 | ||
374f84ac JA |
3040 | /* |
3041 | * So both queues are sync. Let the new request get disk time if | |
3042 | * it's a metadata request and the current queue is doing regular IO. | |
3043 | */ | |
3044 | if (rq_is_meta(rq) && !cfqq->meta_pending) | |
e6ec4fe2 | 3045 | return true; |
22e2c507 | 3046 | |
3a9a3f6c DS |
3047 | /* |
3048 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. | |
3049 | */ | |
3050 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) | |
a6151c3a | 3051 | return true; |
3a9a3f6c | 3052 | |
1e3335de | 3053 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
a6151c3a | 3054 | return false; |
1e3335de JA |
3055 | |
3056 | /* | |
3057 | * if this request is as-good as one we would expect from the | |
3058 | * current cfqq, let it preempt | |
3059 | */ | |
e00ef799 | 3060 | if (cfq_rq_close(cfqd, cfqq, rq)) |
a6151c3a | 3061 | return true; |
1e3335de | 3062 | |
a6151c3a | 3063 | return false; |
22e2c507 JA |
3064 | } |
3065 | ||
3066 | /* | |
3067 | * cfqq preempts the active queue. if we allowed preempt with no slice left, | |
3068 | * let it have half of its nominal slice. | |
3069 | */ | |
3070 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) | |
3071 | { | |
7b679138 | 3072 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
6084cdda | 3073 | cfq_slice_expired(cfqd, 1); |
22e2c507 | 3074 | |
bf572256 JA |
3075 | /* |
3076 | * Put the new queue at the front of the of the current list, | |
3077 | * so we know that it will be selected next. | |
3078 | */ | |
3079 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); | |
edd75ffd JA |
3080 | |
3081 | cfq_service_tree_add(cfqd, cfqq, 1); | |
bf572256 | 3082 | |
44f7c160 JA |
3083 | cfqq->slice_end = 0; |
3084 | cfq_mark_cfqq_slice_new(cfqq); | |
22e2c507 JA |
3085 | } |
3086 | ||
22e2c507 | 3087 | /* |
5e705374 | 3088 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
22e2c507 JA |
3089 | * something we should do about it |
3090 | */ | |
3091 | static void | |
5e705374 JA |
3092 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
3093 | struct request *rq) | |
22e2c507 | 3094 | { |
5e705374 | 3095 | struct cfq_io_context *cic = RQ_CIC(rq); |
12e9fddd | 3096 | |
45333d5a | 3097 | cfqd->rq_queued++; |
374f84ac JA |
3098 | if (rq_is_meta(rq)) |
3099 | cfqq->meta_pending++; | |
3100 | ||
9c2c38a1 | 3101 | cfq_update_io_thinktime(cfqd, cic); |
b2c18e1e | 3102 | cfq_update_io_seektime(cfqd, cfqq, rq); |
9c2c38a1 JA |
3103 | cfq_update_idle_window(cfqd, cfqq, cic); |
3104 | ||
b2c18e1e | 3105 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
22e2c507 JA |
3106 | |
3107 | if (cfqq == cfqd->active_queue) { | |
f75edf2d VG |
3108 | if (cfq_cfqq_wait_busy(cfqq)) { |
3109 | cfq_clear_cfqq_wait_busy(cfqq); | |
3110 | cfq_mark_cfqq_wait_busy_done(cfqq); | |
3111 | } | |
22e2c507 | 3112 | /* |
b029195d JA |
3113 | * Remember that we saw a request from this process, but |
3114 | * don't start queuing just yet. Otherwise we risk seeing lots | |
3115 | * of tiny requests, because we disrupt the normal plugging | |
d6ceb25e JA |
3116 | * and merging. If the request is already larger than a single |
3117 | * page, let it rip immediately. For that case we assume that | |
2d870722 JA |
3118 | * merging is already done. Ditto for a busy system that |
3119 | * has other work pending, don't risk delaying until the | |
3120 | * idle timer unplug to continue working. | |
22e2c507 | 3121 | */ |
d6ceb25e | 3122 | if (cfq_cfqq_wait_request(cfqq)) { |
2d870722 JA |
3123 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
3124 | cfqd->busy_queues > 1) { | |
d6ceb25e | 3125 | del_timer(&cfqd->idle_slice_timer); |
bf791937 VG |
3126 | __blk_run_queue(cfqd->queue); |
3127 | } else | |
3128 | cfq_mark_cfqq_must_dispatch(cfqq); | |
d6ceb25e | 3129 | } |
5e705374 | 3130 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
22e2c507 JA |
3131 | /* |
3132 | * not the active queue - expire current slice if it is | |
3133 | * idle and has expired it's mean thinktime or this new queue | |
3a9a3f6c DS |
3134 | * has some old slice time left and is of higher priority or |
3135 | * this new queue is RT and the current one is BE | |
22e2c507 JA |
3136 | */ |
3137 | cfq_preempt_queue(cfqd, cfqq); | |
a7f55792 | 3138 | __blk_run_queue(cfqd->queue); |
22e2c507 | 3139 | } |
1da177e4 LT |
3140 | } |
3141 | ||
165125e1 | 3142 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3143 | { |
b4878f24 | 3144 | struct cfq_data *cfqd = q->elevator->elevator_data; |
5e705374 | 3145 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
22e2c507 | 3146 | |
7b679138 | 3147 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
fd0928df | 3148 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
1da177e4 | 3149 | |
30996f40 | 3150 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
22e2c507 | 3151 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
aa6f6a3d | 3152 | cfq_add_rq_rb(rq); |
22e2c507 | 3153 | |
5e705374 | 3154 | cfq_rq_enqueued(cfqd, cfqq, rq); |
1da177e4 LT |
3155 | } |
3156 | ||
45333d5a AC |
3157 | /* |
3158 | * Update hw_tag based on peak queue depth over 50 samples under | |
3159 | * sufficient load. | |
3160 | */ | |
3161 | static void cfq_update_hw_tag(struct cfq_data *cfqd) | |
3162 | { | |
1a1238a7 SL |
3163 | struct cfq_queue *cfqq = cfqd->active_queue; |
3164 | ||
e459dd08 CZ |
3165 | if (rq_in_driver(cfqd) > cfqd->hw_tag_est_depth) |
3166 | cfqd->hw_tag_est_depth = rq_in_driver(cfqd); | |
3167 | ||
3168 | if (cfqd->hw_tag == 1) | |
3169 | return; | |
45333d5a AC |
3170 | |
3171 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && | |
5ad531db | 3172 | rq_in_driver(cfqd) <= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3173 | return; |
3174 | ||
1a1238a7 SL |
3175 | /* |
3176 | * If active queue hasn't enough requests and can idle, cfq might not | |
3177 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this | |
3178 | * case | |
3179 | */ | |
3180 | if (cfqq && cfq_cfqq_idle_window(cfqq) && | |
3181 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < | |
3182 | CFQ_HW_QUEUE_MIN && rq_in_driver(cfqd) < CFQ_HW_QUEUE_MIN) | |
3183 | return; | |
3184 | ||
45333d5a AC |
3185 | if (cfqd->hw_tag_samples++ < 50) |
3186 | return; | |
3187 | ||
e459dd08 | 3188 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
45333d5a AC |
3189 | cfqd->hw_tag = 1; |
3190 | else | |
3191 | cfqd->hw_tag = 0; | |
45333d5a AC |
3192 | } |
3193 | ||
165125e1 | 3194 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 | 3195 | { |
5e705374 | 3196 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
b4878f24 | 3197 | struct cfq_data *cfqd = cfqq->cfqd; |
5380a101 | 3198 | const int sync = rq_is_sync(rq); |
b4878f24 | 3199 | unsigned long now; |
1da177e4 | 3200 | |
b4878f24 | 3201 | now = jiffies; |
2868ef7b | 3202 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", !!rq_noidle(rq)); |
1da177e4 | 3203 | |
45333d5a AC |
3204 | cfq_update_hw_tag(cfqd); |
3205 | ||
5ad531db | 3206 | WARN_ON(!cfqd->rq_in_driver[sync]); |
6d048f53 | 3207 | WARN_ON(!cfqq->dispatched); |
5ad531db | 3208 | cfqd->rq_in_driver[sync]--; |
6d048f53 | 3209 | cfqq->dispatched--; |
1da177e4 | 3210 | |
3ed9a296 JA |
3211 | if (cfq_cfqq_sync(cfqq)) |
3212 | cfqd->sync_flight--; | |
3213 | ||
365722bb | 3214 | if (sync) { |
5e705374 | 3215 | RQ_CIC(rq)->last_end_request = now; |
365722bb VG |
3216 | cfqd->last_end_sync_rq = now; |
3217 | } | |
caaa5f9f JA |
3218 | |
3219 | /* | |
3220 | * If this is the active queue, check if it needs to be expired, | |
3221 | * or if we want to idle in case it has no pending requests. | |
3222 | */ | |
3223 | if (cfqd->active_queue == cfqq) { | |
a36e71f9 JA |
3224 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
3225 | ||
44f7c160 JA |
3226 | if (cfq_cfqq_slice_new(cfqq)) { |
3227 | cfq_set_prio_slice(cfqd, cfqq); | |
3228 | cfq_clear_cfqq_slice_new(cfqq); | |
3229 | } | |
f75edf2d VG |
3230 | |
3231 | /* | |
3232 | * If this queue consumed its slice and this is last queue | |
3233 | * in the group, wait for next request before we expire | |
3234 | * the queue | |
3235 | */ | |
3236 | if (cfq_slice_used(cfqq) && cfqq->cfqg->nr_cfqq == 1) { | |
3237 | cfqq->slice_end = jiffies + cfqd->cfq_slice_idle; | |
3238 | cfq_mark_cfqq_wait_busy(cfqq); | |
3239 | } | |
3240 | ||
a36e71f9 | 3241 | /* |
8e550632 CZ |
3242 | * Idling is not enabled on: |
3243 | * - expired queues | |
3244 | * - idle-priority queues | |
3245 | * - async queues | |
3246 | * - queues with still some requests queued | |
3247 | * - when there is a close cooperator | |
a36e71f9 | 3248 | */ |
0871714e | 3249 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
6084cdda | 3250 | cfq_slice_expired(cfqd, 1); |
8e550632 CZ |
3251 | else if (sync && cfqq_empty && |
3252 | !cfq_close_cooperator(cfqd, cfqq)) { | |
3253 | cfqd->noidle_tree_requires_idle |= !rq_noidle(rq); | |
3254 | /* | |
3255 | * Idling is enabled for SYNC_WORKLOAD. | |
3256 | * SYNC_NOIDLE_WORKLOAD idles at the end of the tree | |
3257 | * only if we processed at least one !rq_noidle request | |
3258 | */ | |
3259 | if (cfqd->serving_type == SYNC_WORKLOAD | |
3260 | || cfqd->noidle_tree_requires_idle) | |
3261 | cfq_arm_slice_timer(cfqd); | |
3262 | } | |
caaa5f9f | 3263 | } |
6d048f53 | 3264 | |
5ad531db | 3265 | if (!rq_in_driver(cfqd)) |
23e018a1 | 3266 | cfq_schedule_dispatch(cfqd); |
1da177e4 LT |
3267 | } |
3268 | ||
22e2c507 JA |
3269 | /* |
3270 | * we temporarily boost lower priority queues if they are holding fs exclusive | |
3271 | * resources. they are boosted to normal prio (CLASS_BE/4) | |
3272 | */ | |
3273 | static void cfq_prio_boost(struct cfq_queue *cfqq) | |
1da177e4 | 3274 | { |
22e2c507 JA |
3275 | if (has_fs_excl()) { |
3276 | /* | |
3277 | * boost idle prio on transactions that would lock out other | |
3278 | * users of the filesystem | |
3279 | */ | |
3280 | if (cfq_class_idle(cfqq)) | |
3281 | cfqq->ioprio_class = IOPRIO_CLASS_BE; | |
3282 | if (cfqq->ioprio > IOPRIO_NORM) | |
3283 | cfqq->ioprio = IOPRIO_NORM; | |
3284 | } else { | |
3285 | /* | |
dddb7451 | 3286 | * unboost the queue (if needed) |
22e2c507 | 3287 | */ |
dddb7451 CZ |
3288 | cfqq->ioprio_class = cfqq->org_ioprio_class; |
3289 | cfqq->ioprio = cfqq->org_ioprio; | |
22e2c507 | 3290 | } |
22e2c507 | 3291 | } |
1da177e4 | 3292 | |
89850f7e | 3293 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
22e2c507 | 3294 | { |
1b379d8d | 3295 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
3b18152c | 3296 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
22e2c507 | 3297 | return ELV_MQUEUE_MUST; |
3b18152c | 3298 | } |
1da177e4 | 3299 | |
22e2c507 | 3300 | return ELV_MQUEUE_MAY; |
22e2c507 JA |
3301 | } |
3302 | ||
165125e1 | 3303 | static int cfq_may_queue(struct request_queue *q, int rw) |
22e2c507 JA |
3304 | { |
3305 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3306 | struct task_struct *tsk = current; | |
91fac317 | 3307 | struct cfq_io_context *cic; |
22e2c507 JA |
3308 | struct cfq_queue *cfqq; |
3309 | ||
3310 | /* | |
3311 | * don't force setup of a queue from here, as a call to may_queue | |
3312 | * does not necessarily imply that a request actually will be queued. | |
3313 | * so just lookup a possibly existing queue, or return 'may queue' | |
3314 | * if that fails | |
3315 | */ | |
4ac845a2 | 3316 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
91fac317 VT |
3317 | if (!cic) |
3318 | return ELV_MQUEUE_MAY; | |
3319 | ||
b0b78f81 | 3320 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
22e2c507 | 3321 | if (cfqq) { |
fd0928df | 3322 | cfq_init_prio_data(cfqq, cic->ioc); |
22e2c507 JA |
3323 | cfq_prio_boost(cfqq); |
3324 | ||
89850f7e | 3325 | return __cfq_may_queue(cfqq); |
22e2c507 JA |
3326 | } |
3327 | ||
3328 | return ELV_MQUEUE_MAY; | |
1da177e4 LT |
3329 | } |
3330 | ||
1da177e4 LT |
3331 | /* |
3332 | * queue lock held here | |
3333 | */ | |
bb37b94c | 3334 | static void cfq_put_request(struct request *rq) |
1da177e4 | 3335 | { |
5e705374 | 3336 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
1da177e4 | 3337 | |
5e705374 | 3338 | if (cfqq) { |
22e2c507 | 3339 | const int rw = rq_data_dir(rq); |
1da177e4 | 3340 | |
22e2c507 JA |
3341 | BUG_ON(!cfqq->allocated[rw]); |
3342 | cfqq->allocated[rw]--; | |
1da177e4 | 3343 | |
5e705374 | 3344 | put_io_context(RQ_CIC(rq)->ioc); |
1da177e4 | 3345 | |
1da177e4 | 3346 | rq->elevator_private = NULL; |
5e705374 | 3347 | rq->elevator_private2 = NULL; |
1da177e4 | 3348 | |
1da177e4 LT |
3349 | cfq_put_queue(cfqq); |
3350 | } | |
3351 | } | |
3352 | ||
df5fe3e8 JM |
3353 | static struct cfq_queue * |
3354 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_context *cic, | |
3355 | struct cfq_queue *cfqq) | |
3356 | { | |
3357 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); | |
3358 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); | |
b3b6d040 | 3359 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
df5fe3e8 JM |
3360 | cfq_put_queue(cfqq); |
3361 | return cic_to_cfqq(cic, 1); | |
3362 | } | |
3363 | ||
e6c5bc73 JM |
3364 | static int should_split_cfqq(struct cfq_queue *cfqq) |
3365 | { | |
3366 | if (cfqq->seeky_start && | |
3367 | time_after(jiffies, cfqq->seeky_start + CFQQ_COOP_TOUT)) | |
3368 | return 1; | |
3369 | return 0; | |
3370 | } | |
3371 | ||
3372 | /* | |
3373 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this | |
3374 | * was the last process referring to said cfqq. | |
3375 | */ | |
3376 | static struct cfq_queue * | |
3377 | split_cfqq(struct cfq_io_context *cic, struct cfq_queue *cfqq) | |
3378 | { | |
3379 | if (cfqq_process_refs(cfqq) == 1) { | |
3380 | cfqq->seeky_start = 0; | |
3381 | cfqq->pid = current->pid; | |
3382 | cfq_clear_cfqq_coop(cfqq); | |
3383 | return cfqq; | |
3384 | } | |
3385 | ||
3386 | cic_set_cfqq(cic, NULL, 1); | |
3387 | cfq_put_queue(cfqq); | |
3388 | return NULL; | |
3389 | } | |
1da177e4 | 3390 | /* |
22e2c507 | 3391 | * Allocate cfq data structures associated with this request. |
1da177e4 | 3392 | */ |
22e2c507 | 3393 | static int |
165125e1 | 3394 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
3395 | { |
3396 | struct cfq_data *cfqd = q->elevator->elevator_data; | |
3397 | struct cfq_io_context *cic; | |
3398 | const int rw = rq_data_dir(rq); | |
a6151c3a | 3399 | const bool is_sync = rq_is_sync(rq); |
22e2c507 | 3400 | struct cfq_queue *cfqq; |
1da177e4 LT |
3401 | unsigned long flags; |
3402 | ||
3403 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
3404 | ||
e2d74ac0 | 3405 | cic = cfq_get_io_context(cfqd, gfp_mask); |
22e2c507 | 3406 | |
1da177e4 LT |
3407 | spin_lock_irqsave(q->queue_lock, flags); |
3408 | ||
22e2c507 JA |
3409 | if (!cic) |
3410 | goto queue_fail; | |
3411 | ||
e6c5bc73 | 3412 | new_queue: |
91fac317 | 3413 | cfqq = cic_to_cfqq(cic, is_sync); |
32f2e807 | 3414 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
fd0928df | 3415 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
91fac317 | 3416 | cic_set_cfqq(cic, cfqq, is_sync); |
df5fe3e8 | 3417 | } else { |
e6c5bc73 JM |
3418 | /* |
3419 | * If the queue was seeky for too long, break it apart. | |
3420 | */ | |
3421 | if (cfq_cfqq_coop(cfqq) && should_split_cfqq(cfqq)) { | |
3422 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); | |
3423 | cfqq = split_cfqq(cic, cfqq); | |
3424 | if (!cfqq) | |
3425 | goto new_queue; | |
3426 | } | |
3427 | ||
df5fe3e8 JM |
3428 | /* |
3429 | * Check to see if this queue is scheduled to merge with | |
3430 | * another, closely cooperating queue. The merging of | |
3431 | * queues happens here as it must be done in process context. | |
3432 | * The reference on new_cfqq was taken in merge_cfqqs. | |
3433 | */ | |
3434 | if (cfqq->new_cfqq) | |
3435 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); | |
91fac317 | 3436 | } |
1da177e4 LT |
3437 | |
3438 | cfqq->allocated[rw]++; | |
22e2c507 | 3439 | atomic_inc(&cfqq->ref); |
1da177e4 | 3440 | |
5e705374 | 3441 | spin_unlock_irqrestore(q->queue_lock, flags); |
3b18152c | 3442 | |
5e705374 JA |
3443 | rq->elevator_private = cic; |
3444 | rq->elevator_private2 = cfqq; | |
3445 | return 0; | |
1da177e4 | 3446 | |
22e2c507 JA |
3447 | queue_fail: |
3448 | if (cic) | |
3449 | put_io_context(cic->ioc); | |
89850f7e | 3450 | |
23e018a1 | 3451 | cfq_schedule_dispatch(cfqd); |
1da177e4 | 3452 | spin_unlock_irqrestore(q->queue_lock, flags); |
7b679138 | 3453 | cfq_log(cfqd, "set_request fail"); |
1da177e4 LT |
3454 | return 1; |
3455 | } | |
3456 | ||
65f27f38 | 3457 | static void cfq_kick_queue(struct work_struct *work) |
22e2c507 | 3458 | { |
65f27f38 | 3459 | struct cfq_data *cfqd = |
23e018a1 | 3460 | container_of(work, struct cfq_data, unplug_work); |
165125e1 | 3461 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3462 | |
40bb54d1 | 3463 | spin_lock_irq(q->queue_lock); |
a7f55792 | 3464 | __blk_run_queue(cfqd->queue); |
40bb54d1 | 3465 | spin_unlock_irq(q->queue_lock); |
22e2c507 JA |
3466 | } |
3467 | ||
3468 | /* | |
3469 | * Timer running if the active_queue is currently idling inside its time slice | |
3470 | */ | |
3471 | static void cfq_idle_slice_timer(unsigned long data) | |
3472 | { | |
3473 | struct cfq_data *cfqd = (struct cfq_data *) data; | |
3474 | struct cfq_queue *cfqq; | |
3475 | unsigned long flags; | |
3c6bd2f8 | 3476 | int timed_out = 1; |
22e2c507 | 3477 | |
7b679138 JA |
3478 | cfq_log(cfqd, "idle timer fired"); |
3479 | ||
22e2c507 JA |
3480 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
3481 | ||
fe094d98 JA |
3482 | cfqq = cfqd->active_queue; |
3483 | if (cfqq) { | |
3c6bd2f8 JA |
3484 | timed_out = 0; |
3485 | ||
b029195d JA |
3486 | /* |
3487 | * We saw a request before the queue expired, let it through | |
3488 | */ | |
3489 | if (cfq_cfqq_must_dispatch(cfqq)) | |
3490 | goto out_kick; | |
3491 | ||
22e2c507 JA |
3492 | /* |
3493 | * expired | |
3494 | */ | |
44f7c160 | 3495 | if (cfq_slice_used(cfqq)) |
22e2c507 JA |
3496 | goto expire; |
3497 | ||
3498 | /* | |
3499 | * only expire and reinvoke request handler, if there are | |
3500 | * other queues with pending requests | |
3501 | */ | |
caaa5f9f | 3502 | if (!cfqd->busy_queues) |
22e2c507 | 3503 | goto out_cont; |
22e2c507 JA |
3504 | |
3505 | /* | |
3506 | * not expired and it has a request pending, let it dispatch | |
3507 | */ | |
75e50984 | 3508 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
22e2c507 | 3509 | goto out_kick; |
76280aff CZ |
3510 | |
3511 | /* | |
3512 | * Queue depth flag is reset only when the idle didn't succeed | |
3513 | */ | |
3514 | cfq_clear_cfqq_deep(cfqq); | |
22e2c507 JA |
3515 | } |
3516 | expire: | |
6084cdda | 3517 | cfq_slice_expired(cfqd, timed_out); |
22e2c507 | 3518 | out_kick: |
23e018a1 | 3519 | cfq_schedule_dispatch(cfqd); |
22e2c507 JA |
3520 | out_cont: |
3521 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); | |
3522 | } | |
3523 | ||
3b18152c JA |
3524 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
3525 | { | |
3526 | del_timer_sync(&cfqd->idle_slice_timer); | |
23e018a1 | 3527 | cancel_work_sync(&cfqd->unplug_work); |
3b18152c | 3528 | } |
22e2c507 | 3529 | |
c2dea2d1 VT |
3530 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
3531 | { | |
3532 | int i; | |
3533 | ||
3534 | for (i = 0; i < IOPRIO_BE_NR; i++) { | |
3535 | if (cfqd->async_cfqq[0][i]) | |
3536 | cfq_put_queue(cfqd->async_cfqq[0][i]); | |
3537 | if (cfqd->async_cfqq[1][i]) | |
3538 | cfq_put_queue(cfqd->async_cfqq[1][i]); | |
c2dea2d1 | 3539 | } |
2389d1ef ON |
3540 | |
3541 | if (cfqd->async_idle_cfqq) | |
3542 | cfq_put_queue(cfqd->async_idle_cfqq); | |
c2dea2d1 VT |
3543 | } |
3544 | ||
b374d18a | 3545 | static void cfq_exit_queue(struct elevator_queue *e) |
1da177e4 | 3546 | { |
22e2c507 | 3547 | struct cfq_data *cfqd = e->elevator_data; |
165125e1 | 3548 | struct request_queue *q = cfqd->queue; |
22e2c507 | 3549 | |
3b18152c | 3550 | cfq_shutdown_timer_wq(cfqd); |
e2d74ac0 | 3551 | |
d9ff4187 | 3552 | spin_lock_irq(q->queue_lock); |
e2d74ac0 | 3553 | |
d9ff4187 | 3554 | if (cfqd->active_queue) |
6084cdda | 3555 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
e2d74ac0 JA |
3556 | |
3557 | while (!list_empty(&cfqd->cic_list)) { | |
d9ff4187 AV |
3558 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
3559 | struct cfq_io_context, | |
3560 | queue_list); | |
89850f7e JA |
3561 | |
3562 | __cfq_exit_single_io_context(cfqd, cic); | |
d9ff4187 | 3563 | } |
e2d74ac0 | 3564 | |
c2dea2d1 | 3565 | cfq_put_async_queues(cfqd); |
b1c35769 VG |
3566 | cfq_release_cfq_groups(cfqd); |
3567 | blkiocg_del_blkio_group(&cfqd->root_group.blkg); | |
15c31be4 | 3568 | |
d9ff4187 | 3569 | spin_unlock_irq(q->queue_lock); |
a90d742e AV |
3570 | |
3571 | cfq_shutdown_timer_wq(cfqd); | |
3572 | ||
b1c35769 VG |
3573 | /* Wait for cfqg->blkg->key accessors to exit their grace periods. */ |
3574 | synchronize_rcu(); | |
a90d742e | 3575 | kfree(cfqd); |
1da177e4 LT |
3576 | } |
3577 | ||
165125e1 | 3578 | static void *cfq_init_queue(struct request_queue *q) |
1da177e4 LT |
3579 | { |
3580 | struct cfq_data *cfqd; | |
718eee05 | 3581 | int i, j; |
cdb16e8f | 3582 | struct cfq_group *cfqg; |
615f0259 | 3583 | struct cfq_rb_root *st; |
1da177e4 | 3584 | |
94f6030c | 3585 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
1da177e4 | 3586 | if (!cfqd) |
bc1c1169 | 3587 | return NULL; |
1da177e4 | 3588 | |
1fa8f6d6 VG |
3589 | /* Init root service tree */ |
3590 | cfqd->grp_service_tree = CFQ_RB_ROOT; | |
3591 | ||
cdb16e8f VG |
3592 | /* Init root group */ |
3593 | cfqg = &cfqd->root_group; | |
615f0259 VG |
3594 | for_each_cfqg_st(cfqg, i, j, st) |
3595 | *st = CFQ_RB_ROOT; | |
1fa8f6d6 | 3596 | RB_CLEAR_NODE(&cfqg->rb_node); |
26a2ac00 | 3597 | |
25bc6b07 VG |
3598 | /* Give preference to root group over other groups */ |
3599 | cfqg->weight = 2*BLKIO_WEIGHT_DEFAULT; | |
3600 | ||
25fb5169 | 3601 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
b1c35769 VG |
3602 | /* |
3603 | * Take a reference to root group which we never drop. This is just | |
3604 | * to make sure that cfq_put_cfqg() does not try to kfree root group | |
3605 | */ | |
3606 | atomic_set(&cfqg->ref, 1); | |
22084190 VG |
3607 | blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg, (void *)cfqd, |
3608 | 0); | |
25fb5169 | 3609 | #endif |
26a2ac00 JA |
3610 | /* |
3611 | * Not strictly needed (since RB_ROOT just clears the node and we | |
3612 | * zeroed cfqd on alloc), but better be safe in case someone decides | |
3613 | * to add magic to the rb code | |
3614 | */ | |
3615 | for (i = 0; i < CFQ_PRIO_LISTS; i++) | |
3616 | cfqd->prio_trees[i] = RB_ROOT; | |
3617 | ||
6118b70b JA |
3618 | /* |
3619 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. | |
3620 | * Grab a permanent reference to it, so that the normal code flow | |
3621 | * will not attempt to free it. | |
3622 | */ | |
3623 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); | |
3624 | atomic_inc(&cfqd->oom_cfqq.ref); | |
cdb16e8f | 3625 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group); |
6118b70b | 3626 | |
d9ff4187 | 3627 | INIT_LIST_HEAD(&cfqd->cic_list); |
1da177e4 | 3628 | |
1da177e4 | 3629 | cfqd->queue = q; |
1da177e4 | 3630 | |
22e2c507 JA |
3631 | init_timer(&cfqd->idle_slice_timer); |
3632 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; | |
3633 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; | |
3634 | ||
23e018a1 | 3635 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
22e2c507 | 3636 | |
1da177e4 | 3637 | cfqd->cfq_quantum = cfq_quantum; |
22e2c507 JA |
3638 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
3639 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; | |
1da177e4 LT |
3640 | cfqd->cfq_back_max = cfq_back_max; |
3641 | cfqd->cfq_back_penalty = cfq_back_penalty; | |
22e2c507 JA |
3642 | cfqd->cfq_slice[0] = cfq_slice_async; |
3643 | cfqd->cfq_slice[1] = cfq_slice_sync; | |
3644 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; | |
3645 | cfqd->cfq_slice_idle = cfq_slice_idle; | |
963b72fc | 3646 | cfqd->cfq_latency = 1; |
e459dd08 | 3647 | cfqd->hw_tag = -1; |
365722bb | 3648 | cfqd->last_end_sync_rq = jiffies; |
bc1c1169 | 3649 | return cfqd; |
1da177e4 LT |
3650 | } |
3651 | ||
3652 | static void cfq_slab_kill(void) | |
3653 | { | |
d6de8be7 JA |
3654 | /* |
3655 | * Caller already ensured that pending RCU callbacks are completed, | |
3656 | * so we should have no busy allocations at this point. | |
3657 | */ | |
1da177e4 LT |
3658 | if (cfq_pool) |
3659 | kmem_cache_destroy(cfq_pool); | |
3660 | if (cfq_ioc_pool) | |
3661 | kmem_cache_destroy(cfq_ioc_pool); | |
3662 | } | |
3663 | ||
3664 | static int __init cfq_slab_setup(void) | |
3665 | { | |
0a31bd5f | 3666 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
1da177e4 LT |
3667 | if (!cfq_pool) |
3668 | goto fail; | |
3669 | ||
34e6bbf2 | 3670 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
1da177e4 LT |
3671 | if (!cfq_ioc_pool) |
3672 | goto fail; | |
3673 | ||
3674 | return 0; | |
3675 | fail: | |
3676 | cfq_slab_kill(); | |
3677 | return -ENOMEM; | |
3678 | } | |
3679 | ||
1da177e4 LT |
3680 | /* |
3681 | * sysfs parts below --> | |
3682 | */ | |
1da177e4 LT |
3683 | static ssize_t |
3684 | cfq_var_show(unsigned int var, char *page) | |
3685 | { | |
3686 | return sprintf(page, "%d\n", var); | |
3687 | } | |
3688 | ||
3689 | static ssize_t | |
3690 | cfq_var_store(unsigned int *var, const char *page, size_t count) | |
3691 | { | |
3692 | char *p = (char *) page; | |
3693 | ||
3694 | *var = simple_strtoul(p, &p, 10); | |
3695 | return count; | |
3696 | } | |
3697 | ||
1da177e4 | 3698 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
b374d18a | 3699 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
1da177e4 | 3700 | { \ |
3d1ab40f | 3701 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3702 | unsigned int __data = __VAR; \ |
3703 | if (__CONV) \ | |
3704 | __data = jiffies_to_msecs(__data); \ | |
3705 | return cfq_var_show(__data, (page)); \ | |
3706 | } | |
3707 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); | |
22e2c507 JA |
3708 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
3709 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); | |
e572ec7e AV |
3710 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
3711 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); | |
22e2c507 JA |
3712 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
3713 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); | |
3714 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); | |
3715 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); | |
963b72fc | 3716 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
1da177e4 LT |
3717 | #undef SHOW_FUNCTION |
3718 | ||
3719 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
b374d18a | 3720 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
1da177e4 | 3721 | { \ |
3d1ab40f | 3722 | struct cfq_data *cfqd = e->elevator_data; \ |
1da177e4 LT |
3723 | unsigned int __data; \ |
3724 | int ret = cfq_var_store(&__data, (page), count); \ | |
3725 | if (__data < (MIN)) \ | |
3726 | __data = (MIN); \ | |
3727 | else if (__data > (MAX)) \ | |
3728 | __data = (MAX); \ | |
3729 | if (__CONV) \ | |
3730 | *(__PTR) = msecs_to_jiffies(__data); \ | |
3731 | else \ | |
3732 | *(__PTR) = __data; \ | |
3733 | return ret; \ | |
3734 | } | |
3735 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); | |
fe094d98 JA |
3736 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
3737 | UINT_MAX, 1); | |
3738 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, | |
3739 | UINT_MAX, 1); | |
e572ec7e | 3740 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
fe094d98 JA |
3741 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
3742 | UINT_MAX, 0); | |
22e2c507 JA |
3743 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
3744 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); | |
3745 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); | |
fe094d98 JA |
3746 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
3747 | UINT_MAX, 0); | |
963b72fc | 3748 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
1da177e4 LT |
3749 | #undef STORE_FUNCTION |
3750 | ||
e572ec7e AV |
3751 | #define CFQ_ATTR(name) \ |
3752 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) | |
3753 | ||
3754 | static struct elv_fs_entry cfq_attrs[] = { | |
3755 | CFQ_ATTR(quantum), | |
e572ec7e AV |
3756 | CFQ_ATTR(fifo_expire_sync), |
3757 | CFQ_ATTR(fifo_expire_async), | |
3758 | CFQ_ATTR(back_seek_max), | |
3759 | CFQ_ATTR(back_seek_penalty), | |
3760 | CFQ_ATTR(slice_sync), | |
3761 | CFQ_ATTR(slice_async), | |
3762 | CFQ_ATTR(slice_async_rq), | |
3763 | CFQ_ATTR(slice_idle), | |
963b72fc | 3764 | CFQ_ATTR(low_latency), |
e572ec7e | 3765 | __ATTR_NULL |
1da177e4 LT |
3766 | }; |
3767 | ||
1da177e4 LT |
3768 | static struct elevator_type iosched_cfq = { |
3769 | .ops = { | |
3770 | .elevator_merge_fn = cfq_merge, | |
3771 | .elevator_merged_fn = cfq_merged_request, | |
3772 | .elevator_merge_req_fn = cfq_merged_requests, | |
da775265 | 3773 | .elevator_allow_merge_fn = cfq_allow_merge, |
b4878f24 | 3774 | .elevator_dispatch_fn = cfq_dispatch_requests, |
1da177e4 | 3775 | .elevator_add_req_fn = cfq_insert_request, |
b4878f24 | 3776 | .elevator_activate_req_fn = cfq_activate_request, |
1da177e4 LT |
3777 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
3778 | .elevator_queue_empty_fn = cfq_queue_empty, | |
3779 | .elevator_completed_req_fn = cfq_completed_request, | |
21183b07 JA |
3780 | .elevator_former_req_fn = elv_rb_former_request, |
3781 | .elevator_latter_req_fn = elv_rb_latter_request, | |
1da177e4 LT |
3782 | .elevator_set_req_fn = cfq_set_request, |
3783 | .elevator_put_req_fn = cfq_put_request, | |
3784 | .elevator_may_queue_fn = cfq_may_queue, | |
3785 | .elevator_init_fn = cfq_init_queue, | |
3786 | .elevator_exit_fn = cfq_exit_queue, | |
fc46379d | 3787 | .trim = cfq_free_io_context, |
1da177e4 | 3788 | }, |
3d1ab40f | 3789 | .elevator_attrs = cfq_attrs, |
1da177e4 LT |
3790 | .elevator_name = "cfq", |
3791 | .elevator_owner = THIS_MODULE, | |
3792 | }; | |
3793 | ||
3794 | static int __init cfq_init(void) | |
3795 | { | |
22e2c507 JA |
3796 | /* |
3797 | * could be 0 on HZ < 1000 setups | |
3798 | */ | |
3799 | if (!cfq_slice_async) | |
3800 | cfq_slice_async = 1; | |
3801 | if (!cfq_slice_idle) | |
3802 | cfq_slice_idle = 1; | |
3803 | ||
1da177e4 LT |
3804 | if (cfq_slab_setup()) |
3805 | return -ENOMEM; | |
3806 | ||
2fdd82bd | 3807 | elv_register(&iosched_cfq); |
1da177e4 | 3808 | |
2fdd82bd | 3809 | return 0; |
1da177e4 LT |
3810 | } |
3811 | ||
3812 | static void __exit cfq_exit(void) | |
3813 | { | |
6e9a4738 | 3814 | DECLARE_COMPLETION_ONSTACK(all_gone); |
1da177e4 | 3815 | elv_unregister(&iosched_cfq); |
334e94de | 3816 | ioc_gone = &all_gone; |
fba82272 OH |
3817 | /* ioc_gone's update must be visible before reading ioc_count */ |
3818 | smp_wmb(); | |
d6de8be7 JA |
3819 | |
3820 | /* | |
3821 | * this also protects us from entering cfq_slab_kill() with | |
3822 | * pending RCU callbacks | |
3823 | */ | |
245b2e70 | 3824 | if (elv_ioc_count_read(cfq_ioc_count)) |
9a11b4ed | 3825 | wait_for_completion(&all_gone); |
83521d3e | 3826 | cfq_slab_kill(); |
1da177e4 LT |
3827 | } |
3828 | ||
3829 | module_init(cfq_init); | |
3830 | module_exit(cfq_exit); | |
3831 | ||
3832 | MODULE_AUTHOR("Jens Axboe"); | |
3833 | MODULE_LICENSE("GPL"); | |
3834 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |