blob: d3d76136f53adea6293244f7cd9a9aa07f24b4d0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * 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 *
Jens Axboe0fe23472006-09-04 15:41:16 +02007 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/module.h>
Al Viro1cc9be62006-03-18 12:29:52 -050010#include <linux/blkdev.h>
11#include <linux/elevator.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/hash.h>
13#include <linux/rbtree.h>
Jens Axboe22e2c502005-06-27 10:55:12 +020014#include <linux/ioprio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015
16/*
17 * tunables
18 */
Arjan van de Ven64100092006-01-06 09:46:02 +010019static const int cfq_quantum = 4; /* max queue in one round of service */
Arjan van de Ven64100092006-01-06 09:46:02 +010020static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
21static const int cfq_back_max = 16 * 1024; /* maximum backwards seek, in KiB */
22static const int cfq_back_penalty = 2; /* penalty of a backwards seek */
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
Arjan van de Ven64100092006-01-06 09:46:02 +010024static const int cfq_slice_sync = HZ / 10;
Jens Axboe3b181522005-06-27 10:56:24 +020025static int cfq_slice_async = HZ / 25;
Arjan van de Ven64100092006-01-06 09:46:02 +010026static const int cfq_slice_async_rq = 2;
Jens Axboecaaa5f92006-06-16 11:23:00 +020027static int cfq_slice_idle = HZ / 125;
Jens Axboe22e2c502005-06-27 10:55:12 +020028
29#define CFQ_IDLE_GRACE (HZ / 10)
30#define CFQ_SLICE_SCALE (5)
31
32#define CFQ_KEY_ASYNC (0)
Jens Axboe22e2c502005-06-27 10:55:12 +020033
Linus Torvalds1da177e2005-04-16 15:20:36 -070034/*
35 * for the hash of cfqq inside the cfqd
36 */
37#define CFQ_QHASH_SHIFT 6
38#define CFQ_QHASH_ENTRIES (1 << CFQ_QHASH_SHIFT)
39#define list_entry_qhash(entry) hlist_entry((entry), struct cfq_queue, cfq_hash)
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#define list_entry_cfqq(ptr) list_entry((ptr), struct cfq_queue, cfq_list)
42
Jens Axboe5e705372006-07-13 12:39:25 +020043#define RQ_CIC(rq) ((struct cfq_io_context*)(rq)->elevator_private)
44#define RQ_CFQQ(rq) ((rq)->elevator_private2)
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046static kmem_cache_t *cfq_pool;
47static kmem_cache_t *cfq_ioc_pool;
48
Jens Axboe4050cf12006-07-19 05:07:12 +020049static DEFINE_PER_CPU(unsigned long, ioc_count);
Al Viro334e94d2006-03-18 15:05:53 -050050static struct completion *ioc_gone;
51
Jens Axboe22e2c502005-06-27 10:55:12 +020052#define CFQ_PRIO_LISTS IOPRIO_BE_NR
53#define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
Jens Axboe22e2c502005-06-27 10:55:12 +020054#define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
55
Jens Axboe3b181522005-06-27 10:56:24 +020056#define ASYNC (0)
57#define SYNC (1)
58
59#define cfq_cfqq_dispatched(cfqq) \
60 ((cfqq)->on_dispatch[ASYNC] + (cfqq)->on_dispatch[SYNC])
61
62#define cfq_cfqq_class_sync(cfqq) ((cfqq)->key != CFQ_KEY_ASYNC)
63
64#define cfq_cfqq_sync(cfqq) \
65 (cfq_cfqq_class_sync(cfqq) || (cfqq)->on_dispatch[SYNC])
Jens Axboe22e2c502005-06-27 10:55:12 +020066
Jens Axboe206dc692006-03-28 13:03:44 +020067#define sample_valid(samples) ((samples) > 80)
68
Jens Axboe22e2c502005-06-27 10:55:12 +020069/*
70 * Per block device queue structure
71 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070072struct cfq_data {
Jens Axboe22e2c502005-06-27 10:55:12 +020073 request_queue_t *queue;
74
75 /*
76 * rr list of queues with requests and the count of them
77 */
78 struct list_head rr_list[CFQ_PRIO_LISTS];
79 struct list_head busy_rr;
80 struct list_head cur_rr;
81 struct list_head idle_rr;
82 unsigned int busy_queues;
83
84 /*
Jens Axboe22e2c502005-06-27 10:55:12 +020085 * cfqq lookup hash
86 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 struct hlist_head *cfq_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Jens Axboe22e2c502005-06-27 10:55:12 +020089 int rq_in_driver;
Jens Axboe25776e32006-06-01 10:12:26 +020090 int hw_tag;
Jens Axboe22e2c502005-06-27 10:55:12 +020091
92 /*
Jens Axboe22e2c502005-06-27 10:55:12 +020093 * idle window management
94 */
95 struct timer_list idle_slice_timer;
96 struct work_struct unplug_work;
97
98 struct cfq_queue *active_queue;
99 struct cfq_io_context *active_cic;
100 int cur_prio, cur_end_prio;
101 unsigned int dispatch_slice;
102
103 struct timer_list idle_class_timer;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
105 sector_t last_sector;
Jens Axboe22e2c502005-06-27 10:55:12 +0200106 unsigned long last_end_request;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 /*
109 * tunables, see top of file
110 */
111 unsigned int cfq_quantum;
Jens Axboe22e2c502005-06-27 10:55:12 +0200112 unsigned int cfq_fifo_expire[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 unsigned int cfq_back_penalty;
114 unsigned int cfq_back_max;
Jens Axboe22e2c502005-06-27 10:55:12 +0200115 unsigned int cfq_slice[2];
116 unsigned int cfq_slice_async_rq;
117 unsigned int cfq_slice_idle;
Al Virod9ff4182006-03-18 13:51:22 -0500118
119 struct list_head cic_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120};
121
Jens Axboe22e2c502005-06-27 10:55:12 +0200122/*
123 * Per process-grouping structure
124 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125struct cfq_queue {
126 /* reference count */
127 atomic_t ref;
128 /* parent cfq_data */
129 struct cfq_data *cfqd;
Jens Axboe22e2c502005-06-27 10:55:12 +0200130 /* cfqq lookup hash */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 struct hlist_node cfq_hash;
132 /* hash key */
Jens Axboe22e2c502005-06-27 10:55:12 +0200133 unsigned int key;
Jens Axboe981a7972006-07-19 14:56:28 +0200134 /* member of the rr/busy/cur/idle cfqd list */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 struct list_head cfq_list;
136 /* sorted list of pending requests */
137 struct rb_root sort_list;
138 /* if fifo isn't expired, next request to serve */
Jens Axboe5e705372006-07-13 12:39:25 +0200139 struct request *next_rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 /* requests queued in sort_list */
141 int queued[2];
142 /* currently allocated requests */
143 int allocated[2];
Jens Axboe374f84a2006-07-23 01:42:19 +0200144 /* pending metadata requests */
145 int meta_pending;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 /* fifo list of requests in sort_list */
Jens Axboe22e2c502005-06-27 10:55:12 +0200147 struct list_head fifo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Jens Axboe22e2c502005-06-27 10:55:12 +0200149 unsigned long slice_start;
150 unsigned long slice_end;
151 unsigned long slice_left;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Jens Axboe3b181522005-06-27 10:56:24 +0200153 /* number of requests that are on the dispatch list */
154 int on_dispatch[2];
Jens Axboe22e2c502005-06-27 10:55:12 +0200155
156 /* io prio of this group */
157 unsigned short ioprio, org_ioprio;
158 unsigned short ioprio_class, org_ioprio_class;
159
Jens Axboe3b181522005-06-27 10:56:24 +0200160 /* various state flags, see below */
161 unsigned int flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162};
163
Jens Axboe3b181522005-06-27 10:56:24 +0200164enum cfqq_state_flags {
165 CFQ_CFQQ_FLAG_on_rr = 0,
166 CFQ_CFQQ_FLAG_wait_request,
167 CFQ_CFQQ_FLAG_must_alloc,
168 CFQ_CFQQ_FLAG_must_alloc_slice,
169 CFQ_CFQQ_FLAG_must_dispatch,
170 CFQ_CFQQ_FLAG_fifo_expire,
171 CFQ_CFQQ_FLAG_idle_window,
172 CFQ_CFQQ_FLAG_prio_changed,
Jens Axboe53b037442006-07-28 09:48:51 +0200173 CFQ_CFQQ_FLAG_queue_new,
Jens Axboe3b181522005-06-27 10:56:24 +0200174};
175
176#define CFQ_CFQQ_FNS(name) \
177static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \
178{ \
179 cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name); \
180} \
181static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \
182{ \
183 cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \
184} \
185static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \
186{ \
187 return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \
188}
189
190CFQ_CFQQ_FNS(on_rr);
191CFQ_CFQQ_FNS(wait_request);
192CFQ_CFQQ_FNS(must_alloc);
193CFQ_CFQQ_FNS(must_alloc_slice);
194CFQ_CFQQ_FNS(must_dispatch);
195CFQ_CFQQ_FNS(fifo_expire);
196CFQ_CFQQ_FNS(idle_window);
197CFQ_CFQQ_FNS(prio_changed);
Jens Axboe53b037442006-07-28 09:48:51 +0200198CFQ_CFQQ_FNS(queue_new);
Jens Axboe3b181522005-06-27 10:56:24 +0200199#undef CFQ_CFQQ_FNS
200
Jens Axboe3b181522005-06-27 10:56:24 +0200201static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short);
Jens Axboe5e705372006-07-13 12:39:25 +0200202static void cfq_dispatch_insert(request_queue_t *, struct request *);
Al Viro6f325a12006-03-18 14:58:37 -0500203static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, gfp_t gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205/*
Andrew Morton99f95e52005-06-27 20:14:05 -0700206 * scheduler run of queue, if there are requests pending and no one in the
207 * driver that will restart queueing
208 */
209static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
210{
Jens Axboe7b14e3b2006-02-28 09:35:11 +0100211 if (cfqd->busy_queues)
Andrew Morton99f95e52005-06-27 20:14:05 -0700212 kblockd_schedule_work(&cfqd->unplug_work);
213}
214
215static int cfq_queue_empty(request_queue_t *q)
216{
217 struct cfq_data *cfqd = q->elevator->elevator_data;
218
Jens Axboeb4878f22005-10-20 16:42:29 +0200219 return !cfqd->busy_queues;
Andrew Morton99f95e52005-06-27 20:14:05 -0700220}
221
Jens Axboe206dc692006-03-28 13:03:44 +0200222static inline pid_t cfq_queue_pid(struct task_struct *task, int rw)
223{
Jens Axboeb31dc662006-06-13 08:26:10 +0200224 if (rw == READ || rw == WRITE_SYNC)
Jens Axboe206dc692006-03-28 13:03:44 +0200225 return task->pid;
226
227 return CFQ_KEY_ASYNC;
228}
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230/*
Jens Axboe5e705372006-07-13 12:39:25 +0200231 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 * We choose the request that is closest to the head right now. Distance
Andreas Mohre8a99052006-03-28 08:59:49 +0200233 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 */
Jens Axboe5e705372006-07-13 12:39:25 +0200235static struct request *
236cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
238 sector_t last, s1, s2, d1 = 0, d2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 unsigned long back_max;
Andreas Mohre8a99052006-03-28 08:59:49 +0200240#define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */
241#define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */
242 unsigned wrap = 0; /* bit mask: requests behind the disk head? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Jens Axboe5e705372006-07-13 12:39:25 +0200244 if (rq1 == NULL || rq1 == rq2)
245 return rq2;
246 if (rq2 == NULL)
247 return rq1;
Jens Axboe9c2c38a2005-08-24 14:57:54 +0200248
Jens Axboe5e705372006-07-13 12:39:25 +0200249 if (rq_is_sync(rq1) && !rq_is_sync(rq2))
250 return rq1;
251 else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
252 return rq2;
Jens Axboe374f84a2006-07-23 01:42:19 +0200253 if (rq_is_meta(rq1) && !rq_is_meta(rq2))
254 return rq1;
255 else if (rq_is_meta(rq2) && !rq_is_meta(rq1))
256 return rq2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Jens Axboe5e705372006-07-13 12:39:25 +0200258 s1 = rq1->sector;
259 s2 = rq2->sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 last = cfqd->last_sector;
262
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 /*
264 * by definition, 1KiB is 2 sectors
265 */
266 back_max = cfqd->cfq_back_max * 2;
267
268 /*
269 * Strict one way elevator _except_ in the case where we allow
270 * short backward seeks which are biased as twice the cost of a
271 * similar forward seek.
272 */
273 if (s1 >= last)
274 d1 = s1 - last;
275 else if (s1 + back_max >= last)
276 d1 = (last - s1) * cfqd->cfq_back_penalty;
277 else
Andreas Mohre8a99052006-03-28 08:59:49 +0200278 wrap |= CFQ_RQ1_WRAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
280 if (s2 >= last)
281 d2 = s2 - last;
282 else if (s2 + back_max >= last)
283 d2 = (last - s2) * cfqd->cfq_back_penalty;
284 else
Andreas Mohre8a99052006-03-28 08:59:49 +0200285 wrap |= CFQ_RQ2_WRAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287 /* Found required data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
Andreas Mohre8a99052006-03-28 08:59:49 +0200289 /*
290 * By doing switch() on the bit mask "wrap" we avoid having to
291 * check two variables for all permutations: --> faster!
292 */
293 switch (wrap) {
Jens Axboe5e705372006-07-13 12:39:25 +0200294 case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
Andreas Mohre8a99052006-03-28 08:59:49 +0200295 if (d1 < d2)
Jens Axboe5e705372006-07-13 12:39:25 +0200296 return rq1;
Andreas Mohre8a99052006-03-28 08:59:49 +0200297 else if (d2 < d1)
Jens Axboe5e705372006-07-13 12:39:25 +0200298 return rq2;
Andreas Mohre8a99052006-03-28 08:59:49 +0200299 else {
300 if (s1 >= s2)
Jens Axboe5e705372006-07-13 12:39:25 +0200301 return rq1;
Andreas Mohre8a99052006-03-28 08:59:49 +0200302 else
Jens Axboe5e705372006-07-13 12:39:25 +0200303 return rq2;
Andreas Mohre8a99052006-03-28 08:59:49 +0200304 }
305
306 case CFQ_RQ2_WRAP:
Jens Axboe5e705372006-07-13 12:39:25 +0200307 return rq1;
Andreas Mohre8a99052006-03-28 08:59:49 +0200308 case CFQ_RQ1_WRAP:
Jens Axboe5e705372006-07-13 12:39:25 +0200309 return rq2;
310 case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
Andreas Mohre8a99052006-03-28 08:59:49 +0200311 default:
312 /*
313 * Since both rqs are wrapped,
314 * start with the one that's further behind head
315 * (--> only *one* back seek required),
316 * since back seek takes more time than forward.
317 */
318 if (s1 <= s2)
Jens Axboe5e705372006-07-13 12:39:25 +0200319 return rq1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 else
Jens Axboe5e705372006-07-13 12:39:25 +0200321 return rq2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 }
323}
324
325/*
326 * would be nice to take fifo expire time into account as well
327 */
Jens Axboe5e705372006-07-13 12:39:25 +0200328static struct request *
329cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
330 struct request *last)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
Jens Axboe21183b02006-07-13 12:33:14 +0200332 struct rb_node *rbnext = rb_next(&last->rb_node);
333 struct rb_node *rbprev = rb_prev(&last->rb_node);
Jens Axboe5e705372006-07-13 12:39:25 +0200334 struct request *next = NULL, *prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
Jens Axboe21183b02006-07-13 12:33:14 +0200336 BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
338 if (rbprev)
Jens Axboe5e705372006-07-13 12:39:25 +0200339 prev = rb_entry_rq(rbprev);
Jens Axboe21183b02006-07-13 12:33:14 +0200340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 if (rbnext)
Jens Axboe5e705372006-07-13 12:39:25 +0200342 next = rb_entry_rq(rbnext);
Jens Axboe21183b02006-07-13 12:33:14 +0200343 else {
344 rbnext = rb_first(&cfqq->sort_list);
345 if (rbnext && rbnext != &last->rb_node)
Jens Axboe5e705372006-07-13 12:39:25 +0200346 next = rb_entry_rq(rbnext);
Jens Axboe21183b02006-07-13 12:33:14 +0200347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
Jens Axboe21183b02006-07-13 12:33:14 +0200349 return cfq_choose_req(cfqd, next, prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350}
351
Jens Axboe22e2c502005-06-27 10:55:12 +0200352static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
Jens Axboe22e2c502005-06-27 10:55:12 +0200354 struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe53b037442006-07-28 09:48:51 +0200355 struct list_head *list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
Jens Axboe3b181522005-06-27 10:56:24 +0200357 BUG_ON(!cfq_cfqq_on_rr(cfqq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
359 list_del(&cfqq->cfq_list);
360
Jens Axboe22e2c502005-06-27 10:55:12 +0200361 if (cfq_class_rt(cfqq))
362 list = &cfqd->cur_rr;
363 else if (cfq_class_idle(cfqq))
364 list = &cfqd->idle_rr;
365 else {
366 /*
367 * if cfqq has requests in flight, don't allow it to be
368 * found in cfq_set_active_queue before it has finished them.
369 * this is done to increase fairness between a process that
370 * has lots of io pending vs one that only generates one
371 * sporadically or synchronously
372 */
Jens Axboe3b181522005-06-27 10:56:24 +0200373 if (cfq_cfqq_dispatched(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +0200374 list = &cfqd->busy_rr;
375 else
376 list = &cfqd->rr_list[cfqq->ioprio];
377 }
378
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 /*
Jens Axboe53b037442006-07-28 09:48:51 +0200380 * If this queue was preempted or is new (never been serviced), let
381 * it be added first for fairness but beind other new queues.
382 * Otherwise, just add to the back of the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 */
Jens Axboe53b037442006-07-28 09:48:51 +0200384 if (preempted || cfq_cfqq_queue_new(cfqq)) {
385 struct list_head *n = list;
386 struct cfq_queue *__cfqq;
Jens Axboeb52a8342006-06-01 18:53:43 +0200387
Jens Axboe53b037442006-07-28 09:48:51 +0200388 while (n->next != list) {
389 __cfqq = list_entry_cfqq(n->next);
390 if (!cfq_cfqq_queue_new(__cfqq))
391 break;
392
393 n = n->next;
394 }
395
396 list = n;
Jens Axboe22e2c502005-06-27 10:55:12 +0200397 }
398
Jens Axboe53b037442006-07-28 09:48:51 +0200399 list_add_tail(&cfqq->cfq_list, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400}
401
402/*
403 * add to busy list of queues for service, trying to be fair in ordering
Jens Axboe22e2c502005-06-27 10:55:12 +0200404 * the pending list according to last request service
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 */
406static inline void
Jens Axboeb4878f22005-10-20 16:42:29 +0200407cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408{
Jens Axboe3b181522005-06-27 10:56:24 +0200409 BUG_ON(cfq_cfqq_on_rr(cfqq));
410 cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 cfqd->busy_queues++;
412
Jens Axboeb4878f22005-10-20 16:42:29 +0200413 cfq_resort_rr_list(cfqq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414}
415
416static inline void
417cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
418{
Jens Axboe3b181522005-06-27 10:56:24 +0200419 BUG_ON(!cfq_cfqq_on_rr(cfqq));
420 cfq_clear_cfqq_on_rr(cfqq);
Jens Axboe981a7972006-07-19 14:56:28 +0200421 list_del_init(&cfqq->cfq_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
423 BUG_ON(!cfqd->busy_queues);
424 cfqd->busy_queues--;
425}
426
427/*
428 * rb tree support functions
429 */
Jens Axboe5e705372006-07-13 12:39:25 +0200430static inline void cfq_del_rq_rb(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
Jens Axboe5e705372006-07-13 12:39:25 +0200432 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboeb4878f22005-10-20 16:42:29 +0200433 struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe5e705372006-07-13 12:39:25 +0200434 const int sync = rq_is_sync(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Jens Axboeb4878f22005-10-20 16:42:29 +0200436 BUG_ON(!cfqq->queued[sync]);
437 cfqq->queued[sync]--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Jens Axboe5e705372006-07-13 12:39:25 +0200439 elv_rb_del(&cfqq->sort_list, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
Jens Axboedd67d052006-06-21 09:36:18 +0200441 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
Jens Axboeb4878f22005-10-20 16:42:29 +0200442 cfq_del_cfqq_rr(cfqd, cfqq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
Jens Axboe5e705372006-07-13 12:39:25 +0200445static void cfq_add_rq_rb(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446{
Jens Axboe5e705372006-07-13 12:39:25 +0200447 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe21183b02006-07-13 12:33:14 +0200449 struct request *__alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Jens Axboe5380a102006-07-13 12:37:56 +0200451 cfqq->queued[rq_is_sync(rq)]++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
453 /*
454 * looks a little odd, but the first insert might return an alias.
455 * if that happens, put the alias on the dispatch list
456 */
Jens Axboe21183b02006-07-13 12:33:14 +0200457 while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
Jens Axboe5e705372006-07-13 12:39:25 +0200458 cfq_dispatch_insert(cfqd->queue, __alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459}
460
461static inline void
Jens Axboe5e705372006-07-13 12:39:25 +0200462cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463{
Jens Axboe5380a102006-07-13 12:37:56 +0200464 elv_rb_del(&cfqq->sort_list, rq);
465 cfqq->queued[rq_is_sync(rq)]--;
Jens Axboe5e705372006-07-13 12:39:25 +0200466 cfq_add_rq_rb(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
Jens Axboe206dc692006-03-28 13:03:44 +0200469static struct request *
470cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
Jens Axboe206dc692006-03-28 13:03:44 +0200472 struct task_struct *tsk = current;
473 pid_t key = cfq_queue_pid(tsk, bio_data_dir(bio));
474 struct cfq_queue *cfqq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Jens Axboe206dc692006-03-28 13:03:44 +0200476 cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio);
Jens Axboe89850f72006-07-22 16:48:31 +0200477 if (cfqq) {
478 sector_t sector = bio->bi_sector + bio_sectors(bio);
479
Jens Axboe21183b02006-07-13 12:33:14 +0200480 return elv_rb_find(&cfqq->sort_list, sector);
Jens Axboe89850f72006-07-22 16:48:31 +0200481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 return NULL;
484}
485
Jens Axboeb4878f22005-10-20 16:42:29 +0200486static void cfq_activate_request(request_queue_t *q, struct request *rq)
487{
488 struct cfq_data *cfqd = q->elevator->elevator_data;
489
490 cfqd->rq_in_driver++;
Jens Axboe25776e32006-06-01 10:12:26 +0200491
492 /*
493 * If the depth is larger 1, it really could be queueing. But lets
494 * make the mark a little higher - idling could still be good for
495 * low queueing, and a low queueing number could also just indicate
496 * a SCSI mid layer like behaviour where limit+1 is often seen.
497 */
498 if (!cfqd->hw_tag && cfqd->rq_in_driver > 4)
499 cfqd->hw_tag = 1;
Jens Axboeb4878f22005-10-20 16:42:29 +0200500}
501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502static void cfq_deactivate_request(request_queue_t *q, struct request *rq)
503{
Jens Axboe22e2c502005-06-27 10:55:12 +0200504 struct cfq_data *cfqd = q->elevator->elevator_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Jens Axboeb4878f22005-10-20 16:42:29 +0200506 WARN_ON(!cfqd->rq_in_driver);
507 cfqd->rq_in_driver--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508}
509
Jens Axboeb4878f22005-10-20 16:42:29 +0200510static void cfq_remove_request(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
Jens Axboe5e705372006-07-13 12:39:25 +0200512 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe21183b02006-07-13 12:33:14 +0200513
Jens Axboe5e705372006-07-13 12:39:25 +0200514 if (cfqq->next_rq == rq)
515 cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
Jens Axboeb4878f22005-10-20 16:42:29 +0200517 list_del_init(&rq->queuelist);
Jens Axboe5e705372006-07-13 12:39:25 +0200518 cfq_del_rq_rb(rq);
Jens Axboe374f84a2006-07-23 01:42:19 +0200519
520 if (rq_is_meta(rq)) {
521 WARN_ON(!cfqq->meta_pending);
522 cfqq->meta_pending--;
523 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524}
525
526static int
527cfq_merge(request_queue_t *q, struct request **req, struct bio *bio)
528{
529 struct cfq_data *cfqd = q->elevator->elevator_data;
530 struct request *__rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
Jens Axboe206dc692006-03-28 13:03:44 +0200532 __rq = cfq_find_rq_fmerge(cfqd, bio);
Jens Axboe22e2c502005-06-27 10:55:12 +0200533 if (__rq && elv_rq_merge_ok(__rq, bio)) {
Jens Axboe98170642006-07-28 09:23:08 +0200534 *req = __rq;
535 return ELEVATOR_FRONT_MERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 }
537
538 return ELEVATOR_NO_MERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539}
540
Jens Axboe21183b02006-07-13 12:33:14 +0200541static void cfq_merged_request(request_queue_t *q, struct request *req,
542 int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543{
Jens Axboe21183b02006-07-13 12:33:14 +0200544 if (type == ELEVATOR_FRONT_MERGE) {
Jens Axboe5e705372006-07-13 12:39:25 +0200545 struct cfq_queue *cfqq = RQ_CFQQ(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
Jens Axboe5e705372006-07-13 12:39:25 +0200547 cfq_reposition_rq_rb(cfqq, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549}
550
551static void
552cfq_merged_requests(request_queue_t *q, struct request *rq,
553 struct request *next)
554{
Jens Axboe22e2c502005-06-27 10:55:12 +0200555 /*
556 * reposition in fifo if next is older than rq
557 */
558 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
559 time_before(next->start_time, rq->start_time))
560 list_move(&rq->queuelist, &next->queuelist);
561
Jens Axboeb4878f22005-10-20 16:42:29 +0200562 cfq_remove_request(next);
Jens Axboe22e2c502005-06-27 10:55:12 +0200563}
564
565static inline void
566__cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
567{
568 if (cfqq) {
569 /*
570 * stop potential idle class queues waiting service
571 */
572 del_timer(&cfqd->idle_class_timer);
573
574 cfqq->slice_start = jiffies;
575 cfqq->slice_end = 0;
576 cfqq->slice_left = 0;
Jens Axboe3b181522005-06-27 10:56:24 +0200577 cfq_clear_cfqq_must_alloc_slice(cfqq);
578 cfq_clear_cfqq_fifo_expire(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +0200579 }
580
581 cfqd->active_queue = cfqq;
582}
583
584/*
Jens Axboe7b14e3b2006-02-28 09:35:11 +0100585 * current cfqq expired its slice (or was too idle), select new one
586 */
587static void
588__cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
589 int preempted)
590{
591 unsigned long now = jiffies;
592
593 if (cfq_cfqq_wait_request(cfqq))
594 del_timer(&cfqd->idle_slice_timer);
595
Jens Axboe53b037442006-07-28 09:48:51 +0200596 if (!preempted && !cfq_cfqq_dispatched(cfqq))
Jens Axboe7b14e3b2006-02-28 09:35:11 +0100597 cfq_schedule_dispatch(cfqd);
Jens Axboe7b14e3b2006-02-28 09:35:11 +0100598
599 cfq_clear_cfqq_must_dispatch(cfqq);
600 cfq_clear_cfqq_wait_request(cfqq);
Jens Axboe53b037442006-07-28 09:48:51 +0200601 cfq_clear_cfqq_queue_new(cfqq);
Jens Axboe7b14e3b2006-02-28 09:35:11 +0100602
603 /*
604 * store what was left of this slice, if the queue idled out
605 * or was preempted
606 */
607 if (time_after(cfqq->slice_end, now))
608 cfqq->slice_left = cfqq->slice_end - now;
609 else
610 cfqq->slice_left = 0;
611
612 if (cfq_cfqq_on_rr(cfqq))
613 cfq_resort_rr_list(cfqq, preempted);
614
615 if (cfqq == cfqd->active_queue)
616 cfqd->active_queue = NULL;
617
618 if (cfqd->active_cic) {
619 put_io_context(cfqd->active_cic->ioc);
620 cfqd->active_cic = NULL;
621 }
622
623 cfqd->dispatch_slice = 0;
624}
625
626static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted)
627{
628 struct cfq_queue *cfqq = cfqd->active_queue;
629
630 if (cfqq)
631 __cfq_slice_expired(cfqd, cfqq, preempted);
632}
633
634/*
Jens Axboe22e2c502005-06-27 10:55:12 +0200635 * 0
636 * 0,1
637 * 0,1,2
638 * 0,1,2,3
639 * 0,1,2,3,4
640 * 0,1,2,3,4,5
641 * 0,1,2,3,4,5,6
642 * 0,1,2,3,4,5,6,7
643 */
644static int cfq_get_next_prio_level(struct cfq_data *cfqd)
645{
646 int prio, wrap;
647
648 prio = -1;
649 wrap = 0;
650 do {
651 int p;
652
653 for (p = cfqd->cur_prio; p <= cfqd->cur_end_prio; p++) {
654 if (!list_empty(&cfqd->rr_list[p])) {
655 prio = p;
656 break;
657 }
658 }
659
660 if (prio != -1)
661 break;
662 cfqd->cur_prio = 0;
663 if (++cfqd->cur_end_prio == CFQ_PRIO_LISTS) {
664 cfqd->cur_end_prio = 0;
665 if (wrap)
666 break;
667 wrap = 1;
668 }
669 } while (1);
670
671 if (unlikely(prio == -1))
672 return -1;
673
674 BUG_ON(prio >= CFQ_PRIO_LISTS);
675
676 list_splice_init(&cfqd->rr_list[prio], &cfqd->cur_rr);
677
678 cfqd->cur_prio = prio + 1;
679 if (cfqd->cur_prio > cfqd->cur_end_prio) {
680 cfqd->cur_end_prio = cfqd->cur_prio;
681 cfqd->cur_prio = 0;
682 }
683 if (cfqd->cur_end_prio == CFQ_PRIO_LISTS) {
684 cfqd->cur_prio = 0;
685 cfqd->cur_end_prio = 0;
686 }
687
688 return prio;
689}
690
Jens Axboe3b181522005-06-27 10:56:24 +0200691static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd)
Jens Axboe22e2c502005-06-27 10:55:12 +0200692{
Jens Axboe7b14e3b2006-02-28 09:35:11 +0100693 struct cfq_queue *cfqq = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +0200694
Jens Axboe89850f72006-07-22 16:48:31 +0200695 if (!list_empty(&cfqd->cur_rr) || cfq_get_next_prio_level(cfqd) != -1) {
696 /*
697 * if current list is non-empty, grab first entry. if it is
698 * empty, get next prio level and grab first entry then if any
699 * are spliced
700 */
Jens Axboe22e2c502005-06-27 10:55:12 +0200701 cfqq = list_entry_cfqq(cfqd->cur_rr.next);
Jens Axboe89850f72006-07-22 16:48:31 +0200702 } else if (!list_empty(&cfqd->busy_rr)) {
703 /*
704 * If no new queues are available, check if the busy list has
705 * some before falling back to idle io.
706 */
Jens Axboee0de0202006-06-01 10:07:26 +0200707 cfqq = list_entry_cfqq(cfqd->busy_rr.next);
Jens Axboe89850f72006-07-22 16:48:31 +0200708 } else if (!list_empty(&cfqd->idle_rr)) {
709 /*
710 * if we have idle queues and no rt or be queues had pending
711 * requests, either allow immediate service if the grace period
712 * has passed or arm the idle grace timer
713 */
Jens Axboe22e2c502005-06-27 10:55:12 +0200714 unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE;
715
716 if (time_after_eq(jiffies, end))
717 cfqq = list_entry_cfqq(cfqd->idle_rr.next);
718 else
719 mod_timer(&cfqd->idle_class_timer, end);
720 }
721
722 __cfq_set_active_queue(cfqd, cfqq);
Jens Axboe3b181522005-06-27 10:56:24 +0200723 return cfqq;
Jens Axboe22e2c502005-06-27 10:55:12 +0200724}
725
Jens Axboecaaa5f92006-06-16 11:23:00 +0200726#define CIC_SEEKY(cic) ((cic)->seek_mean > (128 * 1024))
727
Jens Axboe22e2c502005-06-27 10:55:12 +0200728static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
729
730{
Jens Axboe206dc692006-03-28 13:03:44 +0200731 struct cfq_io_context *cic;
Jens Axboe7b14e3b2006-02-28 09:35:11 +0100732 unsigned long sl;
733
Jens Axboedd67d052006-06-21 09:36:18 +0200734 WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
Jens Axboe22e2c502005-06-27 10:55:12 +0200735 WARN_ON(cfqq != cfqd->active_queue);
736
737 /*
738 * idle is disabled, either manually or by past process history
739 */
740 if (!cfqd->cfq_slice_idle)
741 return 0;
Jens Axboe3b181522005-06-27 10:56:24 +0200742 if (!cfq_cfqq_idle_window(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +0200743 return 0;
744 /*
745 * task has exited, don't wait
746 */
Jens Axboe206dc692006-03-28 13:03:44 +0200747 cic = cfqd->active_cic;
748 if (!cic || !cic->ioc->task)
Jens Axboe22e2c502005-06-27 10:55:12 +0200749 return 0;
750
Jens Axboe3b181522005-06-27 10:56:24 +0200751 cfq_mark_cfqq_must_dispatch(cfqq);
752 cfq_mark_cfqq_wait_request(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +0200753
Jens Axboe7b14e3b2006-02-28 09:35:11 +0100754 sl = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle);
Jens Axboe206dc692006-03-28 13:03:44 +0200755
756 /*
757 * we don't want to idle for seeks, but we do want to allow
758 * fair distribution of slice time for a process doing back-to-back
759 * seeks. so allow a little bit of time for him to submit a new rq
760 */
Jens Axboecaaa5f92006-06-16 11:23:00 +0200761 if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
Jens Axboe44eb1232006-07-25 15:05:21 +0200762 sl = min(sl, msecs_to_jiffies(2));
Jens Axboe206dc692006-03-28 13:03:44 +0200763
Jens Axboe7b14e3b2006-02-28 09:35:11 +0100764 mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
Jens Axboe22e2c502005-06-27 10:55:12 +0200765 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766}
767
Jens Axboe5e705372006-07-13 12:39:25 +0200768static void cfq_dispatch_insert(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
770 struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe5e705372006-07-13 12:39:25 +0200771 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +0200772
Jens Axboe5380a102006-07-13 12:37:56 +0200773 cfq_remove_request(rq);
774 cfqq->on_dispatch[rq_is_sync(rq)]++;
775 elv_dispatch_sort(q, rq);
Jens Axboefd61af02006-06-16 15:35:39 +0200776
777 rq = list_entry(q->queue_head.prev, struct request, queuelist);
778 cfqd->last_sector = rq->sector + rq->nr_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779}
780
781/*
782 * return expired entry, or NULL to just start from scratch in rbtree
783 */
Jens Axboe5e705372006-07-13 12:39:25 +0200784static inline struct request *cfq_check_fifo(struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
786 struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe22e2c502005-06-27 10:55:12 +0200787 struct request *rq;
Jens Axboe89850f72006-07-22 16:48:31 +0200788 int fifo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
Jens Axboe3b181522005-06-27 10:56:24 +0200790 if (cfq_cfqq_fifo_expire(cfqq))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 return NULL;
Jens Axboe89850f72006-07-22 16:48:31 +0200792 if (list_empty(&cfqq->fifo))
793 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
Jens Axboe89850f72006-07-22 16:48:31 +0200795 fifo = cfq_cfqq_class_sync(cfqq);
796 rq = rq_entry_fifo(cfqq->fifo.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Jens Axboe89850f72006-07-22 16:48:31 +0200798 if (time_after(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) {
799 cfq_mark_cfqq_fifo_expire(cfqq);
800 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 }
802
803 return NULL;
804}
805
806/*
Jens Axboe3b181522005-06-27 10:56:24 +0200807 * Scale schedule slice based on io priority. Use the sync time slice only
808 * if a queue is marked sync and has sync io queued. A sync queue with async
809 * io only, should not get full sync slice length.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 */
Jens Axboe22e2c502005-06-27 10:55:12 +0200811static inline int
812cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
Jens Axboe22e2c502005-06-27 10:55:12 +0200814 const int base_slice = cfqd->cfq_slice[cfq_cfqq_sync(cfqq)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
Jens Axboe22e2c502005-06-27 10:55:12 +0200816 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
Jens Axboe22e2c502005-06-27 10:55:12 +0200818 return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - cfqq->ioprio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819}
820
Jens Axboe22e2c502005-06-27 10:55:12 +0200821static inline void
822cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
823{
824 cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
825}
826
827static inline int
828cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
829{
830 const int base_rq = cfqd->cfq_slice_async_rq;
831
832 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
833
834 return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
835}
836
837/*
838 * get next queue for service
839 */
Tejun Heo1b5ed5e12005-11-10 08:49:19 +0100840static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
Jens Axboe22e2c502005-06-27 10:55:12 +0200841{
842 unsigned long now = jiffies;
843 struct cfq_queue *cfqq;
844
845 cfqq = cfqd->active_queue;
846 if (!cfqq)
847 goto new_queue;
848
849 /*
850 * slice has expired
851 */
Jens Axboe3b181522005-06-27 10:56:24 +0200852 if (!cfq_cfqq_must_dispatch(cfqq) && time_after(now, cfqq->slice_end))
853 goto expire;
Jens Axboe22e2c502005-06-27 10:55:12 +0200854
855 /*
856 * if queue has requests, dispatch one. if not, check if
857 * enough slice is left to wait for one
858 */
Jens Axboedd67d052006-06-21 09:36:18 +0200859 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
Jens Axboe22e2c502005-06-27 10:55:12 +0200860 goto keep_queue;
Jens Axboecaaa5f92006-06-16 11:23:00 +0200861 else if (cfq_cfqq_dispatched(cfqq)) {
862 cfqq = NULL;
863 goto keep_queue;
864 } else if (cfq_cfqq_class_sync(cfqq)) {
Jens Axboe22e2c502005-06-27 10:55:12 +0200865 if (cfq_arm_slice_timer(cfqd, cfqq))
866 return NULL;
867 }
868
Jens Axboe3b181522005-06-27 10:56:24 +0200869expire:
Jens Axboe22e2c502005-06-27 10:55:12 +0200870 cfq_slice_expired(cfqd, 0);
Jens Axboe3b181522005-06-27 10:56:24 +0200871new_queue:
872 cfqq = cfq_set_active_queue(cfqd);
Jens Axboe22e2c502005-06-27 10:55:12 +0200873keep_queue:
Jens Axboe3b181522005-06-27 10:56:24 +0200874 return cfqq;
Jens Axboe22e2c502005-06-27 10:55:12 +0200875}
876
877static int
878__cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq,
879 int max_dispatch)
880{
881 int dispatched = 0;
882
Jens Axboedd67d052006-06-21 09:36:18 +0200883 BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
Jens Axboe22e2c502005-06-27 10:55:12 +0200884
885 do {
Jens Axboe5e705372006-07-13 12:39:25 +0200886 struct request *rq;
Jens Axboe22e2c502005-06-27 10:55:12 +0200887
888 /*
889 * follow expired path, else get first next available
890 */
Jens Axboe5e705372006-07-13 12:39:25 +0200891 if ((rq = cfq_check_fifo(cfqq)) == NULL)
892 rq = cfqq->next_rq;
Jens Axboe22e2c502005-06-27 10:55:12 +0200893
894 /*
895 * finally, insert request into driver dispatch list
896 */
Jens Axboe5e705372006-07-13 12:39:25 +0200897 cfq_dispatch_insert(cfqd->queue, rq);
Jens Axboe22e2c502005-06-27 10:55:12 +0200898
899 cfqd->dispatch_slice++;
900 dispatched++;
901
902 if (!cfqd->active_cic) {
Jens Axboe5e705372006-07-13 12:39:25 +0200903 atomic_inc(&RQ_CIC(rq)->ioc->refcount);
904 cfqd->active_cic = RQ_CIC(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +0200905 }
906
Jens Axboedd67d052006-06-21 09:36:18 +0200907 if (RB_EMPTY_ROOT(&cfqq->sort_list))
Jens Axboe22e2c502005-06-27 10:55:12 +0200908 break;
909
910 } while (dispatched < max_dispatch);
911
912 /*
Jens Axboecaaa5f92006-06-16 11:23:00 +0200913 * if slice end isn't set yet, set it.
Jens Axboe22e2c502005-06-27 10:55:12 +0200914 */
915 if (!cfqq->slice_end)
916 cfq_set_prio_slice(cfqd, cfqq);
917
918 /*
919 * expire an async queue immediately if it has used up its slice. idle
920 * queue always expire after 1 dispatch round.
921 */
922 if ((!cfq_cfqq_sync(cfqq) &&
923 cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
Jens Axboecaaa5f92006-06-16 11:23:00 +0200924 cfq_class_idle(cfqq) ||
925 !cfq_cfqq_idle_window(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +0200926 cfq_slice_expired(cfqd, 0);
927
928 return dispatched;
929}
930
931static int
Tejun Heo1b5ed5e12005-11-10 08:49:19 +0100932cfq_forced_dispatch_cfqqs(struct list_head *list)
933{
Tejun Heo1b5ed5e12005-11-10 08:49:19 +0100934 struct cfq_queue *cfqq, *next;
Jens Axboecaaa5f92006-06-16 11:23:00 +0200935 int dispatched;
Tejun Heo1b5ed5e12005-11-10 08:49:19 +0100936
Jens Axboecaaa5f92006-06-16 11:23:00 +0200937 dispatched = 0;
Tejun Heo1b5ed5e12005-11-10 08:49:19 +0100938 list_for_each_entry_safe(cfqq, next, list, cfq_list) {
Jens Axboe5e705372006-07-13 12:39:25 +0200939 while (cfqq->next_rq) {
940 cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq);
Tejun Heo1b5ed5e12005-11-10 08:49:19 +0100941 dispatched++;
942 }
943 BUG_ON(!list_empty(&cfqq->fifo));
944 }
Jens Axboecaaa5f92006-06-16 11:23:00 +0200945
Tejun Heo1b5ed5e12005-11-10 08:49:19 +0100946 return dispatched;
947}
948
949static int
950cfq_forced_dispatch(struct cfq_data *cfqd)
951{
952 int i, dispatched = 0;
953
954 for (i = 0; i < CFQ_PRIO_LISTS; i++)
955 dispatched += cfq_forced_dispatch_cfqqs(&cfqd->rr_list[i]);
956
957 dispatched += cfq_forced_dispatch_cfqqs(&cfqd->busy_rr);
958 dispatched += cfq_forced_dispatch_cfqqs(&cfqd->cur_rr);
959 dispatched += cfq_forced_dispatch_cfqqs(&cfqd->idle_rr);
960
961 cfq_slice_expired(cfqd, 0);
962
963 BUG_ON(cfqd->busy_queues);
964
965 return dispatched;
966}
967
968static int
Jens Axboeb4878f22005-10-20 16:42:29 +0200969cfq_dispatch_requests(request_queue_t *q, int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970{
971 struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboecaaa5f92006-06-16 11:23:00 +0200972 struct cfq_queue *cfqq, *prev_cfqq;
973 int dispatched;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
Jens Axboe22e2c502005-06-27 10:55:12 +0200975 if (!cfqd->busy_queues)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 return 0;
977
Tejun Heo1b5ed5e12005-11-10 08:49:19 +0100978 if (unlikely(force))
979 return cfq_forced_dispatch(cfqd);
980
Jens Axboecaaa5f92006-06-16 11:23:00 +0200981 dispatched = 0;
982 prev_cfqq = NULL;
983 while ((cfqq = cfq_select_queue(cfqd)) != NULL) {
Jens Axboeb4878f22005-10-20 16:42:29 +0200984 int max_dispatch;
985
Jens Axboecaaa5f92006-06-16 11:23:00 +0200986 /*
987 * Don't repeat dispatch from the previous queue.
988 */
989 if (prev_cfqq == cfqq)
990 break;
991
Jens Axboe3b181522005-06-27 10:56:24 +0200992 cfq_clear_cfqq_must_dispatch(cfqq);
993 cfq_clear_cfqq_wait_request(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +0200994 del_timer(&cfqd->idle_slice_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
Tejun Heo1b5ed5e12005-11-10 08:49:19 +0100996 max_dispatch = cfqd->cfq_quantum;
997 if (cfq_class_idle(cfqq))
998 max_dispatch = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
Jens Axboecaaa5f92006-06-16 11:23:00 +02001000 dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch);
1001
1002 /*
1003 * If the dispatch cfqq has idling enabled and is still
1004 * the active queue, break out.
1005 */
1006 if (cfq_cfqq_idle_window(cfqq) && cfqd->active_queue)
1007 break;
1008
1009 prev_cfqq = cfqq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 }
1011
Jens Axboecaaa5f92006-06-16 11:23:00 +02001012 return dispatched;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013}
1014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015/*
Jens Axboe5e705372006-07-13 12:39:25 +02001016 * task holds one reference to the queue, dropped when task exits. each rq
1017 * in-flight on this queue also holds a reference, dropped when rq is freed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 *
1019 * queue lock must be held here.
1020 */
1021static void cfq_put_queue(struct cfq_queue *cfqq)
1022{
Jens Axboe22e2c502005-06-27 10:55:12 +02001023 struct cfq_data *cfqd = cfqq->cfqd;
1024
1025 BUG_ON(atomic_read(&cfqq->ref) <= 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
1027 if (!atomic_dec_and_test(&cfqq->ref))
1028 return;
1029
1030 BUG_ON(rb_first(&cfqq->sort_list));
Jens Axboe22e2c502005-06-27 10:55:12 +02001031 BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
Jens Axboe3b181522005-06-27 10:56:24 +02001032 BUG_ON(cfq_cfqq_on_rr(cfqq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
Jens Axboe7b14e3b2006-02-28 09:35:11 +01001034 if (unlikely(cfqd->active_queue == cfqq))
Jens Axboe3b181522005-06-27 10:56:24 +02001035 __cfq_slice_expired(cfqd, cfqq, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02001036
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 /*
1038 * it's on the empty list and still hashed
1039 */
1040 list_del(&cfqq->cfq_list);
1041 hlist_del(&cfqq->cfq_hash);
1042 kmem_cache_free(cfq_pool, cfqq);
1043}
1044
Jens Axboe1ea25ecb2006-07-18 22:24:11 +02001045static struct cfq_queue *
Jens Axboe3b181522005-06-27 10:56:24 +02001046__cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio,
1047 const int hashval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048{
1049 struct hlist_head *hash_list = &cfqd->cfq_hash[hashval];
Jens Axboe206dc692006-03-28 13:03:44 +02001050 struct hlist_node *entry;
1051 struct cfq_queue *__cfqq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
Jens Axboe206dc692006-03-28 13:03:44 +02001053 hlist_for_each_entry(__cfqq, entry, hash_list, cfq_hash) {
Al Virob0a69162006-03-14 15:32:50 -05001054 const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
Jens Axboe206dc692006-03-28 13:03:44 +02001056 if (__cfqq->key == key && (__p == prio || !prio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 return __cfqq;
1058 }
1059
1060 return NULL;
1061}
1062
1063static struct cfq_queue *
Jens Axboe3b181522005-06-27 10:56:24 +02001064cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065{
Jens Axboe3b181522005-06-27 10:56:24 +02001066 return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067}
1068
Jens Axboee2d74ac2006-03-28 08:59:01 +02001069static void cfq_free_io_context(struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070{
Jens Axboe22e2c502005-06-27 10:55:12 +02001071 struct cfq_io_context *__cic;
Jens Axboee2d74ac2006-03-28 08:59:01 +02001072 struct rb_node *n;
1073 int freed = 0;
Jens Axboe22e2c502005-06-27 10:55:12 +02001074
Jens Axboee2d74ac2006-03-28 08:59:01 +02001075 while ((n = rb_first(&ioc->cic_root)) != NULL) {
1076 __cic = rb_entry(n, struct cfq_io_context, rb_node);
1077 rb_erase(&__cic->rb_node, &ioc->cic_root);
Jens Axboe22e2c502005-06-27 10:55:12 +02001078 kmem_cache_free(cfq_ioc_pool, __cic);
Al Viro334e94d2006-03-18 15:05:53 -05001079 freed++;
Jens Axboe22e2c502005-06-27 10:55:12 +02001080 }
1081
Jens Axboe4050cf12006-07-19 05:07:12 +02001082 elv_ioc_count_mod(ioc_count, -freed);
1083
1084 if (ioc_gone && !elv_ioc_count_read(ioc_count))
Al Viro334e94d2006-03-18 15:05:53 -05001085 complete(ioc_gone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086}
1087
Jens Axboe89850f72006-07-22 16:48:31 +02001088static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1089{
1090 if (unlikely(cfqq == cfqd->active_queue))
1091 __cfq_slice_expired(cfqd, cfqq, 0);
1092
1093 cfq_put_queue(cfqq);
1094}
1095
1096static void __cfq_exit_single_io_context(struct cfq_data *cfqd,
1097 struct cfq_io_context *cic)
1098{
Jens Axboefc463792006-08-29 09:05:44 +02001099 list_del_init(&cic->queue_list);
1100 smp_wmb();
1101 cic->key = NULL;
1102
Jens Axboe89850f72006-07-22 16:48:31 +02001103 if (cic->cfqq[ASYNC]) {
1104 cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]);
1105 cic->cfqq[ASYNC] = NULL;
1106 }
1107
1108 if (cic->cfqq[SYNC]) {
1109 cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]);
1110 cic->cfqq[SYNC] = NULL;
1111 }
Jens Axboe89850f72006-07-22 16:48:31 +02001112}
1113
1114
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115/*
Jens Axboe22e2c502005-06-27 10:55:12 +02001116 * Called with interrupts disabled
1117 */
1118static void cfq_exit_single_io_context(struct cfq_io_context *cic)
1119{
Al Viro478a82b2006-03-18 13:25:24 -05001120 struct cfq_data *cfqd = cic->key;
Jens Axboe22e2c502005-06-27 10:55:12 +02001121
Jens Axboe89850f72006-07-22 16:48:31 +02001122 if (cfqd) {
1123 request_queue_t *q = cfqd->queue;
Jens Axboe22e2c502005-06-27 10:55:12 +02001124
Jens Axboefc463792006-08-29 09:05:44 +02001125 spin_lock_irq(q->queue_lock);
Jens Axboe89850f72006-07-22 16:48:31 +02001126 __cfq_exit_single_io_context(cfqd, cic);
Jens Axboefc463792006-08-29 09:05:44 +02001127 spin_unlock_irq(q->queue_lock);
Al Viro12a05732006-03-18 13:38:01 -05001128 }
Jens Axboe22e2c502005-06-27 10:55:12 +02001129}
1130
Jens Axboee2d74ac2006-03-28 08:59:01 +02001131static void cfq_exit_io_context(struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132{
Jens Axboe22e2c502005-06-27 10:55:12 +02001133 struct cfq_io_context *__cic;
Jens Axboee2d74ac2006-03-28 08:59:01 +02001134 struct rb_node *n;
Jens Axboe22e2c502005-06-27 10:55:12 +02001135
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 /*
1137 * put the reference this task is holding to the various queues
1138 */
Jens Axboee2d74ac2006-03-28 08:59:01 +02001139
1140 n = rb_first(&ioc->cic_root);
1141 while (n != NULL) {
1142 __cic = rb_entry(n, struct cfq_io_context, rb_node);
1143
Jens Axboe22e2c502005-06-27 10:55:12 +02001144 cfq_exit_single_io_context(__cic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02001145 n = rb_next(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147}
1148
Jens Axboe22e2c502005-06-27 10:55:12 +02001149static struct cfq_io_context *
Al Viro8267e262005-10-21 03:20:53 -04001150cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
Jens Axboeb5deef92006-07-19 23:39:40 +02001152 struct cfq_io_context *cic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
Jens Axboeb5deef92006-07-19 23:39:40 +02001154 cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask, cfqd->queue->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 if (cic) {
Jens Axboe553698f2006-06-14 19:11:57 +02001156 memset(cic, 0, sizeof(*cic));
Jens Axboe22e2c502005-06-27 10:55:12 +02001157 cic->last_end_request = jiffies;
Jens Axboe553698f2006-06-14 19:11:57 +02001158 INIT_LIST_HEAD(&cic->queue_list);
Jens Axboe22e2c502005-06-27 10:55:12 +02001159 cic->dtor = cfq_free_io_context;
1160 cic->exit = cfq_exit_io_context;
Jens Axboe4050cf12006-07-19 05:07:12 +02001161 elv_ioc_count_inc(ioc_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 }
1163
1164 return cic;
1165}
1166
Jens Axboe22e2c502005-06-27 10:55:12 +02001167static void cfq_init_prio_data(struct cfq_queue *cfqq)
1168{
1169 struct task_struct *tsk = current;
1170 int ioprio_class;
1171
Jens Axboe3b181522005-06-27 10:56:24 +02001172 if (!cfq_cfqq_prio_changed(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +02001173 return;
1174
1175 ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio);
1176 switch (ioprio_class) {
1177 default:
1178 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
1179 case IOPRIO_CLASS_NONE:
1180 /*
1181 * no prio set, place us in the middle of the BE classes
1182 */
1183 cfqq->ioprio = task_nice_ioprio(tsk);
1184 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1185 break;
1186 case IOPRIO_CLASS_RT:
1187 cfqq->ioprio = task_ioprio(tsk);
1188 cfqq->ioprio_class = IOPRIO_CLASS_RT;
1189 break;
1190 case IOPRIO_CLASS_BE:
1191 cfqq->ioprio = task_ioprio(tsk);
1192 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1193 break;
1194 case IOPRIO_CLASS_IDLE:
1195 cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
1196 cfqq->ioprio = 7;
Jens Axboe3b181522005-06-27 10:56:24 +02001197 cfq_clear_cfqq_idle_window(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02001198 break;
1199 }
1200
1201 /*
1202 * keep track of original prio settings in case we have to temporarily
1203 * elevate the priority of this queue
1204 */
1205 cfqq->org_ioprio = cfqq->ioprio;
1206 cfqq->org_ioprio_class = cfqq->ioprio_class;
1207
Jens Axboe3b181522005-06-27 10:56:24 +02001208 if (cfq_cfqq_on_rr(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +02001209 cfq_resort_rr_list(cfqq, 0);
1210
Jens Axboe3b181522005-06-27 10:56:24 +02001211 cfq_clear_cfqq_prio_changed(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02001212}
1213
Al Viro478a82b2006-03-18 13:25:24 -05001214static inline void changed_ioprio(struct cfq_io_context *cic)
Jens Axboe22e2c502005-06-27 10:55:12 +02001215{
Al Viro478a82b2006-03-18 13:25:24 -05001216 struct cfq_data *cfqd = cic->key;
1217 struct cfq_queue *cfqq;
Jens Axboe35e60772006-06-14 09:10:45 +02001218
Jens Axboecaaa5f92006-06-16 11:23:00 +02001219 if (unlikely(!cfqd))
1220 return;
1221
1222 spin_lock(cfqd->queue->queue_lock);
1223
1224 cfqq = cic->cfqq[ASYNC];
1225 if (cfqq) {
1226 struct cfq_queue *new_cfqq;
1227 new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC, cic->ioc->task,
1228 GFP_ATOMIC);
1229 if (new_cfqq) {
1230 cic->cfqq[ASYNC] = new_cfqq;
1231 cfq_put_queue(cfqq);
1232 }
Jens Axboe22e2c502005-06-27 10:55:12 +02001233 }
Jens Axboecaaa5f92006-06-16 11:23:00 +02001234
1235 cfqq = cic->cfqq[SYNC];
1236 if (cfqq)
1237 cfq_mark_cfqq_prio_changed(cfqq);
1238
1239 spin_unlock(cfqd->queue->queue_lock);
Jens Axboe22e2c502005-06-27 10:55:12 +02001240}
1241
Jens Axboefc463792006-08-29 09:05:44 +02001242static void cfq_ioc_set_ioprio(struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
Al Viroa6a07632006-03-18 13:26:44 -05001244 struct cfq_io_context *cic;
Jens Axboee2d74ac2006-03-28 08:59:01 +02001245 struct rb_node *n;
Al Viroa6a07632006-03-18 13:26:44 -05001246
Jens Axboefc463792006-08-29 09:05:44 +02001247 ioc->ioprio_changed = 0;
Al Viroa6a07632006-03-18 13:26:44 -05001248
Jens Axboee2d74ac2006-03-28 08:59:01 +02001249 n = rb_first(&ioc->cic_root);
1250 while (n != NULL) {
1251 cic = rb_entry(n, struct cfq_io_context, rb_node);
Jens Axboe3793c652006-05-30 21:11:04 +02001252
Al Viro478a82b2006-03-18 13:25:24 -05001253 changed_ioprio(cic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02001254 n = rb_next(n);
1255 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256}
1257
1258static struct cfq_queue *
Al Viro6f325a12006-03-18 14:58:37 -05001259cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk,
Al Viro8267e262005-10-21 03:20:53 -04001260 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
1262 const int hashval = hash_long(key, CFQ_QHASH_SHIFT);
1263 struct cfq_queue *cfqq, *new_cfqq = NULL;
Al Viro6f325a12006-03-18 14:58:37 -05001264 unsigned short ioprio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
1266retry:
Al Viro6f325a12006-03-18 14:58:37 -05001267 ioprio = tsk->ioprio;
Jens Axboe3b181522005-06-27 10:56:24 +02001268 cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
1270 if (!cfqq) {
1271 if (new_cfqq) {
1272 cfqq = new_cfqq;
1273 new_cfqq = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02001274 } else if (gfp_mask & __GFP_WAIT) {
Jens Axboe89850f72006-07-22 16:48:31 +02001275 /*
1276 * Inform the allocator of the fact that we will
1277 * just repeat this allocation if it fails, to allow
1278 * the allocator to do whatever it needs to attempt to
1279 * free memory.
1280 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 spin_unlock_irq(cfqd->queue->queue_lock);
Jens Axboeb5deef92006-07-19 23:39:40 +02001282 new_cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask|__GFP_NOFAIL, cfqd->queue->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 spin_lock_irq(cfqd->queue->queue_lock);
1284 goto retry;
Jens Axboe22e2c502005-06-27 10:55:12 +02001285 } else {
Jens Axboeb5deef92006-07-19 23:39:40 +02001286 cfqq = kmem_cache_alloc_node(cfq_pool, gfp_mask, cfqd->queue->node);
Jens Axboe22e2c502005-06-27 10:55:12 +02001287 if (!cfqq)
1288 goto out;
Kiyoshi Ueda db3b5842005-06-17 16:15:10 +02001289 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 memset(cfqq, 0, sizeof(*cfqq));
1292
1293 INIT_HLIST_NODE(&cfqq->cfq_hash);
1294 INIT_LIST_HEAD(&cfqq->cfq_list);
Jens Axboe22e2c502005-06-27 10:55:12 +02001295 INIT_LIST_HEAD(&cfqq->fifo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297 cfqq->key = key;
1298 hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]);
1299 atomic_set(&cfqq->ref, 0);
1300 cfqq->cfqd = cfqd;
Jens Axboe22e2c502005-06-27 10:55:12 +02001301 /*
1302 * set ->slice_left to allow preemption for a new process
1303 */
1304 cfqq->slice_left = 2 * cfqd->cfq_slice_idle;
Jens Axboecaaa5f92006-06-16 11:23:00 +02001305 cfq_mark_cfqq_idle_window(cfqq);
Jens Axboe3b181522005-06-27 10:56:24 +02001306 cfq_mark_cfqq_prio_changed(cfqq);
Jens Axboe53b037442006-07-28 09:48:51 +02001307 cfq_mark_cfqq_queue_new(cfqq);
Jens Axboe3b181522005-06-27 10:56:24 +02001308 cfq_init_prio_data(cfqq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 }
1310
1311 if (new_cfqq)
1312 kmem_cache_free(cfq_pool, new_cfqq);
1313
1314 atomic_inc(&cfqq->ref);
1315out:
1316 WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq);
1317 return cfqq;
1318}
1319
OGAWA Hirofumidbecf3a2006-04-18 09:45:18 +02001320static void
1321cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic)
1322{
Jens Axboefc463792006-08-29 09:05:44 +02001323 WARN_ON(!list_empty(&cic->queue_list));
OGAWA Hirofumidbecf3a2006-04-18 09:45:18 +02001324 rb_erase(&cic->rb_node, &ioc->cic_root);
OGAWA Hirofumidbecf3a2006-04-18 09:45:18 +02001325 kmem_cache_free(cfq_ioc_pool, cic);
Jens Axboe4050cf12006-07-19 05:07:12 +02001326 elv_ioc_count_dec(ioc_count);
OGAWA Hirofumidbecf3a2006-04-18 09:45:18 +02001327}
1328
Jens Axboee2d74ac2006-03-28 08:59:01 +02001329static struct cfq_io_context *
1330cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc)
1331{
OGAWA Hirofumidbecf3a2006-04-18 09:45:18 +02001332 struct rb_node *n;
Jens Axboee2d74ac2006-03-28 08:59:01 +02001333 struct cfq_io_context *cic;
OGAWA Hirofumibe3b0752006-04-18 19:18:31 +02001334 void *k, *key = cfqd;
Jens Axboee2d74ac2006-03-28 08:59:01 +02001335
OGAWA Hirofumidbecf3a2006-04-18 09:45:18 +02001336restart:
1337 n = ioc->cic_root.rb_node;
Jens Axboee2d74ac2006-03-28 08:59:01 +02001338 while (n) {
1339 cic = rb_entry(n, struct cfq_io_context, rb_node);
OGAWA Hirofumibe3b0752006-04-18 19:18:31 +02001340 /* ->key must be copied to avoid race with cfq_exit_queue() */
1341 k = cic->key;
1342 if (unlikely(!k)) {
OGAWA Hirofumidbecf3a2006-04-18 09:45:18 +02001343 cfq_drop_dead_cic(ioc, cic);
1344 goto restart;
1345 }
Jens Axboee2d74ac2006-03-28 08:59:01 +02001346
OGAWA Hirofumibe3b0752006-04-18 19:18:31 +02001347 if (key < k)
Jens Axboee2d74ac2006-03-28 08:59:01 +02001348 n = n->rb_left;
OGAWA Hirofumibe3b0752006-04-18 19:18:31 +02001349 else if (key > k)
Jens Axboee2d74ac2006-03-28 08:59:01 +02001350 n = n->rb_right;
1351 else
1352 return cic;
1353 }
1354
1355 return NULL;
1356}
1357
1358static inline void
1359cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
1360 struct cfq_io_context *cic)
1361{
OGAWA Hirofumidbecf3a2006-04-18 09:45:18 +02001362 struct rb_node **p;
1363 struct rb_node *parent;
Jens Axboee2d74ac2006-03-28 08:59:01 +02001364 struct cfq_io_context *__cic;
OGAWA Hirofumibe3b0752006-04-18 19:18:31 +02001365 void *k;
Jens Axboee2d74ac2006-03-28 08:59:01 +02001366
Jens Axboee2d74ac2006-03-28 08:59:01 +02001367 cic->ioc = ioc;
1368 cic->key = cfqd;
1369
OGAWA Hirofumidbecf3a2006-04-18 09:45:18 +02001370restart:
1371 parent = NULL;
1372 p = &ioc->cic_root.rb_node;
Jens Axboee2d74ac2006-03-28 08:59:01 +02001373 while (*p) {
1374 parent = *p;
1375 __cic = rb_entry(parent, struct cfq_io_context, rb_node);
OGAWA Hirofumibe3b0752006-04-18 19:18:31 +02001376 /* ->key must be copied to avoid race with cfq_exit_queue() */
1377 k = __cic->key;
1378 if (unlikely(!k)) {
Oleg Nesterovbe33c3a2006-08-21 08:36:12 +02001379 cfq_drop_dead_cic(ioc, __cic);
OGAWA Hirofumidbecf3a2006-04-18 09:45:18 +02001380 goto restart;
1381 }
Jens Axboee2d74ac2006-03-28 08:59:01 +02001382
OGAWA Hirofumibe3b0752006-04-18 19:18:31 +02001383 if (cic->key < k)
Jens Axboee2d74ac2006-03-28 08:59:01 +02001384 p = &(*p)->rb_left;
OGAWA Hirofumibe3b0752006-04-18 19:18:31 +02001385 else if (cic->key > k)
Jens Axboee2d74ac2006-03-28 08:59:01 +02001386 p = &(*p)->rb_right;
1387 else
1388 BUG();
1389 }
1390
1391 rb_link_node(&cic->rb_node, parent, p);
1392 rb_insert_color(&cic->rb_node, &ioc->cic_root);
Jens Axboefc463792006-08-29 09:05:44 +02001393
1394 spin_lock_irq(cfqd->queue->queue_lock);
Jens Axboee2d74ac2006-03-28 08:59:01 +02001395 list_add(&cic->queue_list, &cfqd->cic_list);
Jens Axboefc463792006-08-29 09:05:44 +02001396 spin_unlock_irq(cfqd->queue->queue_lock);
Jens Axboee2d74ac2006-03-28 08:59:01 +02001397}
1398
Jens Axboe22e2c502005-06-27 10:55:12 +02001399/*
1400 * Setup general io context and cfq io context. There can be several cfq
1401 * io contexts per general io context, if this process is doing io to more
Jens Axboee2d74ac2006-03-28 08:59:01 +02001402 * than one device managed by cfq.
Jens Axboe22e2c502005-06-27 10:55:12 +02001403 */
1404static struct cfq_io_context *
Jens Axboee2d74ac2006-03-28 08:59:01 +02001405cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406{
Jens Axboe22e2c502005-06-27 10:55:12 +02001407 struct io_context *ioc = NULL;
1408 struct cfq_io_context *cic;
1409
1410 might_sleep_if(gfp_mask & __GFP_WAIT);
1411
Jens Axboeb5deef92006-07-19 23:39:40 +02001412 ioc = get_io_context(gfp_mask, cfqd->queue->node);
Jens Axboe22e2c502005-06-27 10:55:12 +02001413 if (!ioc)
1414 return NULL;
1415
Jens Axboee2d74ac2006-03-28 08:59:01 +02001416 cic = cfq_cic_rb_lookup(cfqd, ioc);
1417 if (cic)
1418 goto out;
Jens Axboe22e2c502005-06-27 10:55:12 +02001419
Jens Axboee2d74ac2006-03-28 08:59:01 +02001420 cic = cfq_alloc_io_context(cfqd, gfp_mask);
1421 if (cic == NULL)
1422 goto err;
Jens Axboe22e2c502005-06-27 10:55:12 +02001423
Jens Axboee2d74ac2006-03-28 08:59:01 +02001424 cfq_cic_link(cfqd, ioc, cic);
Jens Axboe22e2c502005-06-27 10:55:12 +02001425out:
Jens Axboefc463792006-08-29 09:05:44 +02001426 smp_read_barrier_depends();
1427 if (unlikely(ioc->ioprio_changed))
1428 cfq_ioc_set_ioprio(ioc);
1429
Jens Axboe22e2c502005-06-27 10:55:12 +02001430 return cic;
1431err:
1432 put_io_context(ioc);
1433 return NULL;
1434}
1435
1436static void
1437cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
1438{
1439 unsigned long elapsed, ttime;
1440
1441 /*
1442 * if this context already has stuff queued, thinktime is from
1443 * last queue not last end
1444 */
1445#if 0
1446 if (time_after(cic->last_end_request, cic->last_queue))
1447 elapsed = jiffies - cic->last_end_request;
1448 else
1449 elapsed = jiffies - cic->last_queue;
1450#else
1451 elapsed = jiffies - cic->last_end_request;
1452#endif
1453
1454 ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle);
1455
1456 cic->ttime_samples = (7*cic->ttime_samples + 256) / 8;
1457 cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8;
1458 cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples;
1459}
1460
Jens Axboe206dc692006-03-28 13:03:44 +02001461static void
1462cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic,
Jens Axboe5e705372006-07-13 12:39:25 +02001463 struct request *rq)
Jens Axboe206dc692006-03-28 13:03:44 +02001464{
1465 sector_t sdist;
1466 u64 total;
1467
Jens Axboe5e705372006-07-13 12:39:25 +02001468 if (cic->last_request_pos < rq->sector)
1469 sdist = rq->sector - cic->last_request_pos;
Jens Axboe206dc692006-03-28 13:03:44 +02001470 else
Jens Axboe5e705372006-07-13 12:39:25 +02001471 sdist = cic->last_request_pos - rq->sector;
Jens Axboe206dc692006-03-28 13:03:44 +02001472
1473 /*
1474 * Don't allow the seek distance to get too large from the
1475 * odd fragment, pagein, etc
1476 */
1477 if (cic->seek_samples <= 60) /* second&third seek */
1478 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024);
1479 else
1480 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64);
1481
1482 cic->seek_samples = (7*cic->seek_samples + 256) / 8;
1483 cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8;
1484 total = cic->seek_total + (cic->seek_samples/2);
1485 do_div(total, cic->seek_samples);
1486 cic->seek_mean = (sector_t)total;
1487}
Jens Axboe22e2c502005-06-27 10:55:12 +02001488
1489/*
1490 * Disable idle window if the process thinks too long or seeks so much that
1491 * it doesn't matter
1492 */
1493static void
1494cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1495 struct cfq_io_context *cic)
1496{
Jens Axboe3b181522005-06-27 10:56:24 +02001497 int enable_idle = cfq_cfqq_idle_window(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02001498
Jens Axboecaaa5f92006-06-16 11:23:00 +02001499 if (!cic->ioc->task || !cfqd->cfq_slice_idle ||
1500 (cfqd->hw_tag && CIC_SEEKY(cic)))
Jens Axboe22e2c502005-06-27 10:55:12 +02001501 enable_idle = 0;
1502 else if (sample_valid(cic->ttime_samples)) {
1503 if (cic->ttime_mean > cfqd->cfq_slice_idle)
1504 enable_idle = 0;
1505 else
1506 enable_idle = 1;
1507 }
1508
Jens Axboe3b181522005-06-27 10:56:24 +02001509 if (enable_idle)
1510 cfq_mark_cfqq_idle_window(cfqq);
1511 else
1512 cfq_clear_cfqq_idle_window(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02001513}
1514
1515
1516/*
1517 * Check if new_cfqq should preempt the currently active queue. Return 0 for
1518 * no or if we aren't sure, a 1 will cause a preempt.
1519 */
1520static int
1521cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
Jens Axboe5e705372006-07-13 12:39:25 +02001522 struct request *rq)
Jens Axboe22e2c502005-06-27 10:55:12 +02001523{
1524 struct cfq_queue *cfqq = cfqd->active_queue;
1525
1526 if (cfq_class_idle(new_cfqq))
1527 return 0;
1528
1529 if (!cfqq)
Jens Axboecaaa5f92006-06-16 11:23:00 +02001530 return 0;
Jens Axboe22e2c502005-06-27 10:55:12 +02001531
1532 if (cfq_class_idle(cfqq))
1533 return 1;
Jens Axboe3b181522005-06-27 10:56:24 +02001534 if (!cfq_cfqq_wait_request(new_cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +02001535 return 0;
1536 /*
1537 * if it doesn't have slice left, forget it
1538 */
1539 if (new_cfqq->slice_left < cfqd->cfq_slice_idle)
1540 return 0;
Jens Axboe374f84a2006-07-23 01:42:19 +02001541 /*
1542 * if the new request is sync, but the currently running queue is
1543 * not, let the sync request have priority.
1544 */
Jens Axboe5e705372006-07-13 12:39:25 +02001545 if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +02001546 return 1;
Jens Axboe374f84a2006-07-23 01:42:19 +02001547 /*
1548 * So both queues are sync. Let the new request get disk time if
1549 * it's a metadata request and the current queue is doing regular IO.
1550 */
1551 if (rq_is_meta(rq) && !cfqq->meta_pending)
1552 return 1;
Jens Axboe22e2c502005-06-27 10:55:12 +02001553
1554 return 0;
1555}
1556
1557/*
1558 * cfqq preempts the active queue. if we allowed preempt with no slice left,
1559 * let it have half of its nominal slice.
1560 */
1561static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1562{
Jens Axboebf572252006-07-19 20:29:12 +02001563 cfq_slice_expired(cfqd, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02001564
1565 if (!cfqq->slice_left)
1566 cfqq->slice_left = cfq_prio_to_slice(cfqd, cfqq) / 2;
1567
Jens Axboebf572252006-07-19 20:29:12 +02001568 /*
1569 * Put the new queue at the front of the of the current list,
1570 * so we know that it will be selected next.
1571 */
1572 BUG_ON(!cfq_cfqq_on_rr(cfqq));
1573 list_move(&cfqq->cfq_list, &cfqd->cur_rr);
1574
Jens Axboe22e2c502005-06-27 10:55:12 +02001575 cfqq->slice_end = cfqq->slice_left + jiffies;
Jens Axboe22e2c502005-06-27 10:55:12 +02001576}
1577
1578/*
Jens Axboe5e705372006-07-13 12:39:25 +02001579 * Called when a new fs request (rq) is added (to cfqq). Check if there's
Jens Axboe22e2c502005-06-27 10:55:12 +02001580 * something we should do about it
1581 */
1582static void
Jens Axboe5e705372006-07-13 12:39:25 +02001583cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1584 struct request *rq)
Jens Axboe22e2c502005-06-27 10:55:12 +02001585{
Jens Axboe5e705372006-07-13 12:39:25 +02001586 struct cfq_io_context *cic = RQ_CIC(rq);
Jens Axboe12e9fdd2006-06-01 10:09:56 +02001587
Jens Axboe374f84a2006-07-23 01:42:19 +02001588 if (rq_is_meta(rq))
1589 cfqq->meta_pending++;
1590
Jens Axboe9c2c38a2005-08-24 14:57:54 +02001591 /*
Jens Axboe5380a102006-07-13 12:37:56 +02001592 * check if this request is a better next-serve candidate)) {
Jens Axboe21183b02006-07-13 12:33:14 +02001593 */
Jens Axboe5e705372006-07-13 12:39:25 +02001594 cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq);
1595 BUG_ON(!cfqq->next_rq);
Jens Axboe21183b02006-07-13 12:33:14 +02001596
1597 /*
Jens Axboe9c2c38a2005-08-24 14:57:54 +02001598 * we never wait for an async request and we don't allow preemption
1599 * of an async request. so just return early
1600 */
Jens Axboe5e705372006-07-13 12:39:25 +02001601 if (!rq_is_sync(rq)) {
Jens Axboe12e9fdd2006-06-01 10:09:56 +02001602 /*
1603 * sync process issued an async request, if it's waiting
1604 * then expire it and kick rq handling.
1605 */
1606 if (cic == cfqd->active_cic &&
1607 del_timer(&cfqd->idle_slice_timer)) {
1608 cfq_slice_expired(cfqd, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +02001609 blk_start_queueing(cfqd->queue);
Jens Axboe12e9fdd2006-06-01 10:09:56 +02001610 }
Jens Axboe9c2c38a2005-08-24 14:57:54 +02001611 return;
Jens Axboe12e9fdd2006-06-01 10:09:56 +02001612 }
Jens Axboe22e2c502005-06-27 10:55:12 +02001613
Jens Axboe9c2c38a2005-08-24 14:57:54 +02001614 cfq_update_io_thinktime(cfqd, cic);
Jens Axboe5e705372006-07-13 12:39:25 +02001615 cfq_update_io_seektime(cfqd, cic, rq);
Jens Axboe9c2c38a2005-08-24 14:57:54 +02001616 cfq_update_idle_window(cfqd, cfqq, cic);
1617
1618 cic->last_queue = jiffies;
Jens Axboe5e705372006-07-13 12:39:25 +02001619 cic->last_request_pos = rq->sector + rq->nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02001620
1621 if (cfqq == cfqd->active_queue) {
1622 /*
1623 * if we are waiting for a request for this queue, let it rip
1624 * immediately and flag that we must not expire this queue
1625 * just now
1626 */
Jens Axboe3b181522005-06-27 10:56:24 +02001627 if (cfq_cfqq_wait_request(cfqq)) {
1628 cfq_mark_cfqq_must_dispatch(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02001629 del_timer(&cfqd->idle_slice_timer);
Jens Axboedc72ef42006-07-20 14:54:05 +02001630 blk_start_queueing(cfqd->queue);
Jens Axboe22e2c502005-06-27 10:55:12 +02001631 }
Jens Axboe5e705372006-07-13 12:39:25 +02001632 } else if (cfq_should_preempt(cfqd, cfqq, rq)) {
Jens Axboe22e2c502005-06-27 10:55:12 +02001633 /*
1634 * not the active queue - expire current slice if it is
1635 * idle and has expired it's mean thinktime or this new queue
1636 * has some old slice time left and is of higher priority
1637 */
1638 cfq_preempt_queue(cfqd, cfqq);
Jens Axboe3b181522005-06-27 10:56:24 +02001639 cfq_mark_cfqq_must_dispatch(cfqq);
Jens Axboedc72ef42006-07-20 14:54:05 +02001640 blk_start_queueing(cfqd->queue);
Jens Axboe22e2c502005-06-27 10:55:12 +02001641 }
1642}
1643
Jens Axboeb4878f22005-10-20 16:42:29 +02001644static void cfq_insert_request(request_queue_t *q, struct request *rq)
Jens Axboe22e2c502005-06-27 10:55:12 +02001645{
Jens Axboeb4878f22005-10-20 16:42:29 +02001646 struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe5e705372006-07-13 12:39:25 +02001647 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +02001648
1649 cfq_init_prio_data(cfqq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
Jens Axboe5e705372006-07-13 12:39:25 +02001651 cfq_add_rq_rb(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652
Jens Axboe21183b02006-07-13 12:33:14 +02001653 if (!cfq_cfqq_on_rr(cfqq))
1654 cfq_add_cfqq_rr(cfqd, cfqq);
1655
Jens Axboe22e2c502005-06-27 10:55:12 +02001656 list_add_tail(&rq->queuelist, &cfqq->fifo);
1657
Jens Axboe5e705372006-07-13 12:39:25 +02001658 cfq_rq_enqueued(cfqd, cfqq, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659}
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661static void cfq_completed_request(request_queue_t *q, struct request *rq)
1662{
Jens Axboe5e705372006-07-13 12:39:25 +02001663 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboeb4878f22005-10-20 16:42:29 +02001664 struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe5380a102006-07-13 12:37:56 +02001665 const int sync = rq_is_sync(rq);
Jens Axboeb4878f22005-10-20 16:42:29 +02001666 unsigned long now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
Jens Axboeb4878f22005-10-20 16:42:29 +02001668 now = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669
Jens Axboeb4878f22005-10-20 16:42:29 +02001670 WARN_ON(!cfqd->rq_in_driver);
1671 WARN_ON(!cfqq->on_dispatch[sync]);
1672 cfqd->rq_in_driver--;
1673 cfqq->on_dispatch[sync]--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Jens Axboeb4878f22005-10-20 16:42:29 +02001675 if (!cfq_class_idle(cfqq))
1676 cfqd->last_end_request = now;
Jens Axboe3b181522005-06-27 10:56:24 +02001677
Jens Axboe53b037442006-07-28 09:48:51 +02001678 if (!cfq_cfqq_dispatched(cfqq) && cfq_cfqq_on_rr(cfqq))
1679 cfq_resort_rr_list(cfqq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
Jens Axboecaaa5f92006-06-16 11:23:00 +02001681 if (sync)
Jens Axboe5e705372006-07-13 12:39:25 +02001682 RQ_CIC(rq)->last_end_request = now;
Jens Axboecaaa5f92006-06-16 11:23:00 +02001683
1684 /*
1685 * If this is the active queue, check if it needs to be expired,
1686 * or if we want to idle in case it has no pending requests.
1687 */
1688 if (cfqd->active_queue == cfqq) {
1689 if (time_after(now, cfqq->slice_end))
1690 cfq_slice_expired(cfqd, 0);
Jens Axboedd67d052006-06-21 09:36:18 +02001691 else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) {
Jens Axboecaaa5f92006-06-16 11:23:00 +02001692 if (!cfq_arm_slice_timer(cfqd, cfqq))
1693 cfq_schedule_dispatch(cfqd);
1694 }
1695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696}
1697
Jens Axboe22e2c502005-06-27 10:55:12 +02001698/*
1699 * we temporarily boost lower priority queues if they are holding fs exclusive
1700 * resources. they are boosted to normal prio (CLASS_BE/4)
1701 */
1702static void cfq_prio_boost(struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703{
Jens Axboe22e2c502005-06-27 10:55:12 +02001704 const int ioprio_class = cfqq->ioprio_class;
1705 const int ioprio = cfqq->ioprio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706
Jens Axboe22e2c502005-06-27 10:55:12 +02001707 if (has_fs_excl()) {
1708 /*
1709 * boost idle prio on transactions that would lock out other
1710 * users of the filesystem
1711 */
1712 if (cfq_class_idle(cfqq))
1713 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1714 if (cfqq->ioprio > IOPRIO_NORM)
1715 cfqq->ioprio = IOPRIO_NORM;
1716 } else {
1717 /*
1718 * check if we need to unboost the queue
1719 */
1720 if (cfqq->ioprio_class != cfqq->org_ioprio_class)
1721 cfqq->ioprio_class = cfqq->org_ioprio_class;
1722 if (cfqq->ioprio != cfqq->org_ioprio)
1723 cfqq->ioprio = cfqq->org_ioprio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 }
1725
Jens Axboe22e2c502005-06-27 10:55:12 +02001726 /*
1727 * refile between round-robin lists if we moved the priority class
1728 */
1729 if ((ioprio_class != cfqq->ioprio_class || ioprio != cfqq->ioprio) &&
Jens Axboe3b181522005-06-27 10:56:24 +02001730 cfq_cfqq_on_rr(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +02001731 cfq_resort_rr_list(cfqq, 0);
1732}
1733
Jens Axboe89850f72006-07-22 16:48:31 +02001734static inline int __cfq_may_queue(struct cfq_queue *cfqq)
Jens Axboe22e2c502005-06-27 10:55:12 +02001735{
Jens Axboe3b181522005-06-27 10:56:24 +02001736 if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) &&
Andrew Morton99f95e52005-06-27 20:14:05 -07001737 !cfq_cfqq_must_alloc_slice(cfqq)) {
Jens Axboe3b181522005-06-27 10:56:24 +02001738 cfq_mark_cfqq_must_alloc_slice(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02001739 return ELV_MQUEUE_MUST;
Jens Axboe3b181522005-06-27 10:56:24 +02001740 }
Jens Axboe22e2c502005-06-27 10:55:12 +02001741
1742 return ELV_MQUEUE_MAY;
Jens Axboe22e2c502005-06-27 10:55:12 +02001743}
1744
Jens Axboecb78b282006-07-28 09:32:57 +02001745static int cfq_may_queue(request_queue_t *q, int rw)
Jens Axboe22e2c502005-06-27 10:55:12 +02001746{
1747 struct cfq_data *cfqd = q->elevator->elevator_data;
1748 struct task_struct *tsk = current;
1749 struct cfq_queue *cfqq;
1750
1751 /*
1752 * don't force setup of a queue from here, as a call to may_queue
1753 * does not necessarily imply that a request actually will be queued.
1754 * so just lookup a possibly existing queue, or return 'may queue'
1755 * if that fails
1756 */
Jens Axboe3b181522005-06-27 10:56:24 +02001757 cfqq = cfq_find_cfq_hash(cfqd, cfq_queue_pid(tsk, rw), tsk->ioprio);
Jens Axboe22e2c502005-06-27 10:55:12 +02001758 if (cfqq) {
1759 cfq_init_prio_data(cfqq);
1760 cfq_prio_boost(cfqq);
1761
Jens Axboe89850f72006-07-22 16:48:31 +02001762 return __cfq_may_queue(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02001763 }
1764
1765 return ELV_MQUEUE_MAY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766}
1767
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768/*
1769 * queue lock held here
1770 */
1771static void cfq_put_request(request_queue_t *q, struct request *rq)
1772{
Jens Axboe5e705372006-07-13 12:39:25 +02001773 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
Jens Axboe5e705372006-07-13 12:39:25 +02001775 if (cfqq) {
Jens Axboe22e2c502005-06-27 10:55:12 +02001776 const int rw = rq_data_dir(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
Jens Axboe22e2c502005-06-27 10:55:12 +02001778 BUG_ON(!cfqq->allocated[rw]);
1779 cfqq->allocated[rw]--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
Jens Axboe5e705372006-07-13 12:39:25 +02001781 put_io_context(RQ_CIC(rq)->ioc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 rq->elevator_private = NULL;
Jens Axboe5e705372006-07-13 12:39:25 +02001784 rq->elevator_private2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 cfq_put_queue(cfqq);
1787 }
1788}
1789
1790/*
Jens Axboe22e2c502005-06-27 10:55:12 +02001791 * Allocate cfq data structures associated with this request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 */
Jens Axboe22e2c502005-06-27 10:55:12 +02001793static int
Jens Axboecb78b282006-07-28 09:32:57 +02001794cfq_set_request(request_queue_t *q, struct request *rq, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795{
1796 struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe3b181522005-06-27 10:56:24 +02001797 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 struct cfq_io_context *cic;
1799 const int rw = rq_data_dir(rq);
Jens Axboe3b181522005-06-27 10:56:24 +02001800 pid_t key = cfq_queue_pid(tsk, rw);
Jens Axboe22e2c502005-06-27 10:55:12 +02001801 struct cfq_queue *cfqq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 unsigned long flags;
Al Viro12a05732006-03-18 13:38:01 -05001803 int is_sync = key != CFQ_KEY_ASYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
1805 might_sleep_if(gfp_mask & __GFP_WAIT);
1806
Jens Axboee2d74ac2006-03-28 08:59:01 +02001807 cic = cfq_get_io_context(cfqd, gfp_mask);
Jens Axboe22e2c502005-06-27 10:55:12 +02001808
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 spin_lock_irqsave(q->queue_lock, flags);
1810
Jens Axboe22e2c502005-06-27 10:55:12 +02001811 if (!cic)
1812 goto queue_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
Al Viro12a05732006-03-18 13:38:01 -05001814 if (!cic->cfqq[is_sync]) {
Al Viro6f325a12006-03-18 14:58:37 -05001815 cfqq = cfq_get_queue(cfqd, key, tsk, gfp_mask);
Jens Axboe22e2c502005-06-27 10:55:12 +02001816 if (!cfqq)
1817 goto queue_fail;
1818
Al Viro12a05732006-03-18 13:38:01 -05001819 cic->cfqq[is_sync] = cfqq;
Jens Axboe22e2c502005-06-27 10:55:12 +02001820 } else
Al Viro12a05732006-03-18 13:38:01 -05001821 cfqq = cic->cfqq[is_sync];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
1823 cfqq->allocated[rw]++;
Jens Axboe3b181522005-06-27 10:56:24 +02001824 cfq_clear_cfqq_must_alloc(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02001825 atomic_inc(&cfqq->ref);
Jens Axboe5e705372006-07-13 12:39:25 +02001826
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 spin_unlock_irqrestore(q->queue_lock, flags);
1828
Jens Axboe5e705372006-07-13 12:39:25 +02001829 rq->elevator_private = cic;
1830 rq->elevator_private2 = cfqq;
1831 return 0;
Jens Axboe3b181522005-06-27 10:56:24 +02001832
Jens Axboe22e2c502005-06-27 10:55:12 +02001833queue_fail:
1834 if (cic)
1835 put_io_context(cic->ioc);
Jens Axboe89850f72006-07-22 16:48:31 +02001836
Jens Axboe3b181522005-06-27 10:56:24 +02001837 cfq_schedule_dispatch(cfqd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 spin_unlock_irqrestore(q->queue_lock, flags);
1839 return 1;
1840}
1841
Jens Axboe22e2c502005-06-27 10:55:12 +02001842static void cfq_kick_queue(void *data)
1843{
1844 request_queue_t *q = data;
Jens Axboe22e2c502005-06-27 10:55:12 +02001845 unsigned long flags;
1846
1847 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboedc72ef42006-07-20 14:54:05 +02001848 blk_start_queueing(q);
Jens Axboe22e2c502005-06-27 10:55:12 +02001849 spin_unlock_irqrestore(q->queue_lock, flags);
1850}
1851
1852/*
1853 * Timer running if the active_queue is currently idling inside its time slice
1854 */
1855static void cfq_idle_slice_timer(unsigned long data)
1856{
1857 struct cfq_data *cfqd = (struct cfq_data *) data;
1858 struct cfq_queue *cfqq;
1859 unsigned long flags;
1860
1861 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
1862
1863 if ((cfqq = cfqd->active_queue) != NULL) {
1864 unsigned long now = jiffies;
1865
1866 /*
1867 * expired
1868 */
1869 if (time_after(now, cfqq->slice_end))
1870 goto expire;
1871
1872 /*
1873 * only expire and reinvoke request handler, if there are
1874 * other queues with pending requests
1875 */
Jens Axboecaaa5f92006-06-16 11:23:00 +02001876 if (!cfqd->busy_queues)
Jens Axboe22e2c502005-06-27 10:55:12 +02001877 goto out_cont;
Jens Axboe22e2c502005-06-27 10:55:12 +02001878
1879 /*
1880 * not expired and it has a request pending, let it dispatch
1881 */
Jens Axboedd67d052006-06-21 09:36:18 +02001882 if (!RB_EMPTY_ROOT(&cfqq->sort_list)) {
Jens Axboe3b181522005-06-27 10:56:24 +02001883 cfq_mark_cfqq_must_dispatch(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02001884 goto out_kick;
1885 }
1886 }
1887expire:
1888 cfq_slice_expired(cfqd, 0);
1889out_kick:
Jens Axboe3b181522005-06-27 10:56:24 +02001890 cfq_schedule_dispatch(cfqd);
Jens Axboe22e2c502005-06-27 10:55:12 +02001891out_cont:
1892 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
1893}
1894
1895/*
1896 * Timer running if an idle class queue is waiting for service
1897 */
1898static void cfq_idle_class_timer(unsigned long data)
1899{
1900 struct cfq_data *cfqd = (struct cfq_data *) data;
1901 unsigned long flags, end;
1902
1903 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
1904
1905 /*
1906 * race with a non-idle queue, reset timer
1907 */
1908 end = cfqd->last_end_request + CFQ_IDLE_GRACE;
Jens Axboeae818a32006-06-01 10:13:43 +02001909 if (!time_after_eq(jiffies, end))
1910 mod_timer(&cfqd->idle_class_timer, end);
1911 else
Jens Axboe3b181522005-06-27 10:56:24 +02001912 cfq_schedule_dispatch(cfqd);
Jens Axboe22e2c502005-06-27 10:55:12 +02001913
1914 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
1915}
1916
Jens Axboe3b181522005-06-27 10:56:24 +02001917static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
1918{
1919 del_timer_sync(&cfqd->idle_slice_timer);
1920 del_timer_sync(&cfqd->idle_class_timer);
1921 blk_sync_queue(cfqd->queue);
1922}
Jens Axboe22e2c502005-06-27 10:55:12 +02001923
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924static void cfq_exit_queue(elevator_t *e)
1925{
Jens Axboe22e2c502005-06-27 10:55:12 +02001926 struct cfq_data *cfqd = e->elevator_data;
Al Virod9ff4182006-03-18 13:51:22 -05001927 request_queue_t *q = cfqd->queue;
Jens Axboe22e2c502005-06-27 10:55:12 +02001928
Jens Axboe3b181522005-06-27 10:56:24 +02001929 cfq_shutdown_timer_wq(cfqd);
Jens Axboee2d74ac2006-03-28 08:59:01 +02001930
Al Virod9ff4182006-03-18 13:51:22 -05001931 spin_lock_irq(q->queue_lock);
Jens Axboee2d74ac2006-03-28 08:59:01 +02001932
Al Virod9ff4182006-03-18 13:51:22 -05001933 if (cfqd->active_queue)
1934 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
Jens Axboee2d74ac2006-03-28 08:59:01 +02001935
1936 while (!list_empty(&cfqd->cic_list)) {
Al Virod9ff4182006-03-18 13:51:22 -05001937 struct cfq_io_context *cic = list_entry(cfqd->cic_list.next,
1938 struct cfq_io_context,
1939 queue_list);
Jens Axboe89850f72006-07-22 16:48:31 +02001940
1941 __cfq_exit_single_io_context(cfqd, cic);
Al Virod9ff4182006-03-18 13:51:22 -05001942 }
Jens Axboee2d74ac2006-03-28 08:59:01 +02001943
Al Virod9ff4182006-03-18 13:51:22 -05001944 spin_unlock_irq(q->queue_lock);
Al Viroa90d7422006-03-18 12:05:37 -05001945
1946 cfq_shutdown_timer_wq(cfqd);
1947
Al Viroa90d7422006-03-18 12:05:37 -05001948 kfree(cfqd->cfq_hash);
1949 kfree(cfqd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950}
1951
Jens Axboebc1c1162006-06-08 08:49:06 +02001952static void *cfq_init_queue(request_queue_t *q, elevator_t *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953{
1954 struct cfq_data *cfqd;
1955 int i;
1956
Jens Axboeb5deef92006-07-19 23:39:40 +02001957 cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 if (!cfqd)
Jens Axboebc1c1162006-06-08 08:49:06 +02001959 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960
1961 memset(cfqd, 0, sizeof(*cfqd));
Jens Axboe22e2c502005-06-27 10:55:12 +02001962
1963 for (i = 0; i < CFQ_PRIO_LISTS; i++)
1964 INIT_LIST_HEAD(&cfqd->rr_list[i]);
1965
1966 INIT_LIST_HEAD(&cfqd->busy_rr);
1967 INIT_LIST_HEAD(&cfqd->cur_rr);
1968 INIT_LIST_HEAD(&cfqd->idle_rr);
Al Virod9ff4182006-03-18 13:51:22 -05001969 INIT_LIST_HEAD(&cfqd->cic_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
Jens Axboeb5deef92006-07-19 23:39:40 +02001971 cfqd->cfq_hash = kmalloc_node(sizeof(struct hlist_head) * CFQ_QHASH_ENTRIES, GFP_KERNEL, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 if (!cfqd->cfq_hash)
Jens Axboe5e705372006-07-13 12:39:25 +02001973 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 for (i = 0; i < CFQ_QHASH_ENTRIES; i++)
1976 INIT_HLIST_HEAD(&cfqd->cfq_hash[i]);
1977
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 cfqd->queue = q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
Jens Axboe22e2c502005-06-27 10:55:12 +02001980 init_timer(&cfqd->idle_slice_timer);
1981 cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
1982 cfqd->idle_slice_timer.data = (unsigned long) cfqd;
1983
1984 init_timer(&cfqd->idle_class_timer);
1985 cfqd->idle_class_timer.function = cfq_idle_class_timer;
1986 cfqd->idle_class_timer.data = (unsigned long) cfqd;
1987
1988 INIT_WORK(&cfqd->unplug_work, cfq_kick_queue, q);
1989
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 cfqd->cfq_quantum = cfq_quantum;
Jens Axboe22e2c502005-06-27 10:55:12 +02001991 cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
1992 cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 cfqd->cfq_back_max = cfq_back_max;
1994 cfqd->cfq_back_penalty = cfq_back_penalty;
Jens Axboe22e2c502005-06-27 10:55:12 +02001995 cfqd->cfq_slice[0] = cfq_slice_async;
1996 cfqd->cfq_slice[1] = cfq_slice_sync;
1997 cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
1998 cfqd->cfq_slice_idle = cfq_slice_idle;
Jens Axboe3b181522005-06-27 10:56:24 +02001999
Jens Axboebc1c1162006-06-08 08:49:06 +02002000 return cfqd;
Jens Axboe5e705372006-07-13 12:39:25 +02002001out_free:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 kfree(cfqd);
Jens Axboebc1c1162006-06-08 08:49:06 +02002003 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004}
2005
2006static void cfq_slab_kill(void)
2007{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 if (cfq_pool)
2009 kmem_cache_destroy(cfq_pool);
2010 if (cfq_ioc_pool)
2011 kmem_cache_destroy(cfq_ioc_pool);
2012}
2013
2014static int __init cfq_slab_setup(void)
2015{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 cfq_pool = kmem_cache_create("cfq_pool", sizeof(struct cfq_queue), 0, 0,
2017 NULL, NULL);
2018 if (!cfq_pool)
2019 goto fail;
2020
2021 cfq_ioc_pool = kmem_cache_create("cfq_ioc_pool",
2022 sizeof(struct cfq_io_context), 0, 0, NULL, NULL);
2023 if (!cfq_ioc_pool)
2024 goto fail;
2025
2026 return 0;
2027fail:
2028 cfq_slab_kill();
2029 return -ENOMEM;
2030}
2031
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032/*
2033 * sysfs parts below -->
2034 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
2036static ssize_t
2037cfq_var_show(unsigned int var, char *page)
2038{
2039 return sprintf(page, "%d\n", var);
2040}
2041
2042static ssize_t
2043cfq_var_store(unsigned int *var, const char *page, size_t count)
2044{
2045 char *p = (char *) page;
2046
2047 *var = simple_strtoul(p, &p, 10);
2048 return count;
2049}
2050
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051#define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
Al Viro3d1ab402006-03-18 18:35:43 -05002052static ssize_t __FUNC(elevator_t *e, char *page) \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053{ \
Al Viro3d1ab402006-03-18 18:35:43 -05002054 struct cfq_data *cfqd = e->elevator_data; \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 unsigned int __data = __VAR; \
2056 if (__CONV) \
2057 __data = jiffies_to_msecs(__data); \
2058 return cfq_var_show(__data, (page)); \
2059}
2060SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02002061SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
2062SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
Al Viroe572ec72006-03-18 22:27:18 -05002063SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
2064SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02002065SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
2066SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
2067SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
2068SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069#undef SHOW_FUNCTION
2070
2071#define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
Al Viro3d1ab402006-03-18 18:35:43 -05002072static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073{ \
Al Viro3d1ab402006-03-18 18:35:43 -05002074 struct cfq_data *cfqd = e->elevator_data; \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 unsigned int __data; \
2076 int ret = cfq_var_store(&__data, (page), count); \
2077 if (__data < (MIN)) \
2078 __data = (MIN); \
2079 else if (__data > (MAX)) \
2080 __data = (MAX); \
2081 if (__CONV) \
2082 *(__PTR) = msecs_to_jiffies(__data); \
2083 else \
2084 *(__PTR) = __data; \
2085 return ret; \
2086}
2087STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02002088STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1);
2089STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1);
Al Viroe572ec72006-03-18 22:27:18 -05002090STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
2091STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02002092STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
2093STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
2094STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
2095STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096#undef STORE_FUNCTION
2097
Al Viroe572ec72006-03-18 22:27:18 -05002098#define CFQ_ATTR(name) \
2099 __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
Jens Axboe3b181522005-06-27 10:56:24 +02002100
Al Viroe572ec72006-03-18 22:27:18 -05002101static struct elv_fs_entry cfq_attrs[] = {
2102 CFQ_ATTR(quantum),
Al Viroe572ec72006-03-18 22:27:18 -05002103 CFQ_ATTR(fifo_expire_sync),
2104 CFQ_ATTR(fifo_expire_async),
2105 CFQ_ATTR(back_seek_max),
2106 CFQ_ATTR(back_seek_penalty),
2107 CFQ_ATTR(slice_sync),
2108 CFQ_ATTR(slice_async),
2109 CFQ_ATTR(slice_async_rq),
2110 CFQ_ATTR(slice_idle),
Al Viroe572ec72006-03-18 22:27:18 -05002111 __ATTR_NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112};
2113
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114static struct elevator_type iosched_cfq = {
2115 .ops = {
2116 .elevator_merge_fn = cfq_merge,
2117 .elevator_merged_fn = cfq_merged_request,
2118 .elevator_merge_req_fn = cfq_merged_requests,
Jens Axboeb4878f22005-10-20 16:42:29 +02002119 .elevator_dispatch_fn = cfq_dispatch_requests,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 .elevator_add_req_fn = cfq_insert_request,
Jens Axboeb4878f22005-10-20 16:42:29 +02002121 .elevator_activate_req_fn = cfq_activate_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 .elevator_deactivate_req_fn = cfq_deactivate_request,
2123 .elevator_queue_empty_fn = cfq_queue_empty,
2124 .elevator_completed_req_fn = cfq_completed_request,
Jens Axboe21183b02006-07-13 12:33:14 +02002125 .elevator_former_req_fn = elv_rb_former_request,
2126 .elevator_latter_req_fn = elv_rb_latter_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 .elevator_set_req_fn = cfq_set_request,
2128 .elevator_put_req_fn = cfq_put_request,
2129 .elevator_may_queue_fn = cfq_may_queue,
2130 .elevator_init_fn = cfq_init_queue,
2131 .elevator_exit_fn = cfq_exit_queue,
Jens Axboefc463792006-08-29 09:05:44 +02002132 .trim = cfq_free_io_context,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 },
Al Viro3d1ab402006-03-18 18:35:43 -05002134 .elevator_attrs = cfq_attrs,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 .elevator_name = "cfq",
2136 .elevator_owner = THIS_MODULE,
2137};
2138
2139static int __init cfq_init(void)
2140{
2141 int ret;
2142
Jens Axboe22e2c502005-06-27 10:55:12 +02002143 /*
2144 * could be 0 on HZ < 1000 setups
2145 */
2146 if (!cfq_slice_async)
2147 cfq_slice_async = 1;
2148 if (!cfq_slice_idle)
2149 cfq_slice_idle = 1;
2150
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 if (cfq_slab_setup())
2152 return -ENOMEM;
2153
2154 ret = elv_register(&iosched_cfq);
Jens Axboe22e2c502005-06-27 10:55:12 +02002155 if (ret)
2156 cfq_slab_kill();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 return ret;
2159}
2160
2161static void __exit cfq_exit(void)
2162{
Peter Zijlstra6e9a4732006-09-30 23:28:10 -07002163 DECLARE_COMPLETION_ONSTACK(all_gone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 elv_unregister(&iosched_cfq);
Al Viro334e94d2006-03-18 15:05:53 -05002165 ioc_gone = &all_gone;
OGAWA Hirofumifba82272006-04-18 09:44:06 +02002166 /* ioc_gone's update must be visible before reading ioc_count */
2167 smp_wmb();
Jens Axboe4050cf12006-07-19 05:07:12 +02002168 if (elv_ioc_count_read(ioc_count))
OGAWA Hirofumifba82272006-04-18 09:44:06 +02002169 wait_for_completion(ioc_gone);
Al Viro334e94d2006-03-18 15:05:53 -05002170 synchronize_rcu();
Christoph Hellwig83521d32005-10-30 15:01:39 -08002171 cfq_slab_kill();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172}
2173
2174module_init(cfq_init);
2175module_exit(cfq_exit);
2176
2177MODULE_AUTHOR("Jens Axboe");
2178MODULE_LICENSE("GPL");
2179MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");