ia64/linux-2.6.18-xen.hg

view block/cfq-iosched.c @ 890:2e94884f0e8d

Dom0 PCI: fix a regression introduced by the SR-IOV change

The device class may be changed during the early fixup. So need to
re-read the device class from pci_dev after the fixup.

The patch "PCI: centralize device setup code" (c/s 825) wrongly
cleaned up the device class re-read. This patch reverts that change.

Signed-off-by: Yu Zhao <yu.zhao@intel.com>
author Keir Fraser <keir.fraser@citrix.com>
date Wed Jun 03 11:21:52 2009 +0100 (2009-06-03)
parents 831230e53067
children
line source
1 /*
2 * CFQ, or complete fairness queueing, disk scheduler.
3 *
4 * Based on ideas from a previously unfinished io
5 * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli.
6 *
7 * Copyright (C) 2003 Jens Axboe <axboe@suse.de>
8 */
9 #include <linux/module.h>
10 #include <linux/blkdev.h>
11 #include <linux/elevator.h>
12 #include <linux/hash.h>
13 #include <linux/rbtree.h>
14 #include <linux/ioprio.h>
16 /*
17 * tunables
18 */
19 static const int cfq_quantum = 4; /* max queue in one round of service */
20 static const int cfq_queued = 8; /* minimum rq allocate limit per-queue*/
21 static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
22 static const int cfq_back_max = 16 * 1024; /* maximum backwards seek, in KiB */
23 static const int cfq_back_penalty = 2; /* penalty of a backwards seek */
25 static const int cfq_slice_sync = HZ / 10;
26 static int cfq_slice_async = HZ / 25;
27 static const int cfq_slice_async_rq = 2;
28 static int cfq_slice_idle = HZ / 125;
30 #define CFQ_IDLE_GRACE (HZ / 10)
31 #define CFQ_SLICE_SCALE (5)
33 #define CFQ_KEY_ASYNC (0)
35 static DEFINE_SPINLOCK(cfq_exit_lock);
37 /*
38 * for the hash of cfqq inside the cfqd
39 */
40 #define CFQ_QHASH_SHIFT 6
41 #define CFQ_QHASH_ENTRIES (1 << CFQ_QHASH_SHIFT)
42 #define list_entry_qhash(entry) hlist_entry((entry), struct cfq_queue, cfq_hash)
44 /*
45 * for the hash of crq inside the cfqq
46 */
47 #define CFQ_MHASH_SHIFT 6
48 #define CFQ_MHASH_BLOCK(sec) ((sec) >> 3)
49 #define CFQ_MHASH_ENTRIES (1 << CFQ_MHASH_SHIFT)
50 #define CFQ_MHASH_FN(sec) hash_long(CFQ_MHASH_BLOCK(sec), CFQ_MHASH_SHIFT)
51 #define rq_hash_key(rq) ((rq)->sector + (rq)->nr_sectors)
52 #define list_entry_hash(ptr) hlist_entry((ptr), struct cfq_rq, hash)
54 #define list_entry_cfqq(ptr) list_entry((ptr), struct cfq_queue, cfq_list)
55 #define list_entry_fifo(ptr) list_entry((ptr), struct request, queuelist)
57 #define RQ_DATA(rq) (rq)->elevator_private
59 /*
60 * rb-tree defines
61 */
62 #define rb_entry_crq(node) rb_entry((node), struct cfq_rq, rb_node)
63 #define rq_rb_key(rq) (rq)->sector
65 static kmem_cache_t *crq_pool;
66 static kmem_cache_t *cfq_pool;
67 static kmem_cache_t *cfq_ioc_pool;
69 static atomic_t ioc_count = ATOMIC_INIT(0);
70 static struct completion *ioc_gone;
72 #define CFQ_PRIO_LISTS IOPRIO_BE_NR
73 #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
74 #define cfq_class_be(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_BE)
75 #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
77 #define ASYNC (0)
78 #define SYNC (1)
80 #define cfq_cfqq_dispatched(cfqq) \
81 ((cfqq)->on_dispatch[ASYNC] + (cfqq)->on_dispatch[SYNC])
83 #define cfq_cfqq_class_sync(cfqq) ((cfqq)->key != CFQ_KEY_ASYNC)
85 #define cfq_cfqq_sync(cfqq) \
86 (cfq_cfqq_class_sync(cfqq) || (cfqq)->on_dispatch[SYNC])
88 #define sample_valid(samples) ((samples) > 80)
90 /*
91 * Per block device queue structure
92 */
93 struct cfq_data {
94 request_queue_t *queue;
96 /*
97 * rr list of queues with requests and the count of them
98 */
99 struct list_head rr_list[CFQ_PRIO_LISTS];
100 struct list_head busy_rr;
101 struct list_head cur_rr;
102 struct list_head idle_rr;
103 unsigned int busy_queues;
105 /*
106 * non-ordered list of empty cfqq's
107 */
108 struct list_head empty_list;
110 /*
111 * cfqq lookup hash
112 */
113 struct hlist_head *cfq_hash;
115 /*
116 * global crq hash for all queues
117 */
118 struct hlist_head *crq_hash;
120 mempool_t *crq_pool;
122 int rq_in_driver;
123 int hw_tag;
125 /*
126 * schedule slice state info
127 */
128 /*
129 * idle window management
130 */
131 struct timer_list idle_slice_timer;
132 struct work_struct unplug_work;
134 struct cfq_queue *active_queue;
135 struct cfq_io_context *active_cic;
136 int cur_prio, cur_end_prio;
137 unsigned int dispatch_slice;
139 struct timer_list idle_class_timer;
141 sector_t last_sector;
142 unsigned long last_end_request;
144 unsigned int rq_starved;
146 /*
147 * tunables, see top of file
148 */
149 unsigned int cfq_quantum;
150 unsigned int cfq_queued;
151 unsigned int cfq_fifo_expire[2];
152 unsigned int cfq_back_penalty;
153 unsigned int cfq_back_max;
154 unsigned int cfq_slice[2];
155 unsigned int cfq_slice_async_rq;
156 unsigned int cfq_slice_idle;
158 struct list_head cic_list;
159 };
161 /*
162 * Per process-grouping structure
163 */
164 struct cfq_queue {
165 /* reference count */
166 atomic_t ref;
167 /* parent cfq_data */
168 struct cfq_data *cfqd;
169 /* cfqq lookup hash */
170 struct hlist_node cfq_hash;
171 /* hash key */
172 unsigned int key;
173 /* on either rr or empty list of cfqd */
174 struct list_head cfq_list;
175 /* sorted list of pending requests */
176 struct rb_root sort_list;
177 /* if fifo isn't expired, next request to serve */
178 struct cfq_rq *next_crq;
179 /* requests queued in sort_list */
180 int queued[2];
181 /* currently allocated requests */
182 int allocated[2];
183 /* fifo list of requests in sort_list */
184 struct list_head fifo;
186 unsigned long slice_start;
187 unsigned long slice_end;
188 unsigned long slice_left;
189 unsigned long service_last;
191 /* number of requests that are on the dispatch list */
192 int on_dispatch[2];
194 /* io prio of this group */
195 unsigned short ioprio, org_ioprio;
196 unsigned short ioprio_class, org_ioprio_class;
198 /* various state flags, see below */
199 unsigned int flags;
200 };
202 struct cfq_rq {
203 struct rb_node rb_node;
204 sector_t rb_key;
205 struct request *request;
206 struct hlist_node hash;
208 struct cfq_queue *cfq_queue;
209 struct cfq_io_context *io_context;
211 unsigned int crq_flags;
212 };
214 enum cfqq_state_flags {
215 CFQ_CFQQ_FLAG_on_rr = 0,
216 CFQ_CFQQ_FLAG_wait_request,
217 CFQ_CFQQ_FLAG_must_alloc,
218 CFQ_CFQQ_FLAG_must_alloc_slice,
219 CFQ_CFQQ_FLAG_must_dispatch,
220 CFQ_CFQQ_FLAG_fifo_expire,
221 CFQ_CFQQ_FLAG_idle_window,
222 CFQ_CFQQ_FLAG_prio_changed,
223 };
225 #define CFQ_CFQQ_FNS(name) \
226 static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \
227 { \
228 cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name); \
229 } \
230 static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \
231 { \
232 cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \
233 } \
234 static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \
235 { \
236 return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \
237 }
239 CFQ_CFQQ_FNS(on_rr);
240 CFQ_CFQQ_FNS(wait_request);
241 CFQ_CFQQ_FNS(must_alloc);
242 CFQ_CFQQ_FNS(must_alloc_slice);
243 CFQ_CFQQ_FNS(must_dispatch);
244 CFQ_CFQQ_FNS(fifo_expire);
245 CFQ_CFQQ_FNS(idle_window);
246 CFQ_CFQQ_FNS(prio_changed);
247 #undef CFQ_CFQQ_FNS
249 enum cfq_rq_state_flags {
250 CFQ_CRQ_FLAG_is_sync = 0,
251 };
253 #define CFQ_CRQ_FNS(name) \
254 static inline void cfq_mark_crq_##name(struct cfq_rq *crq) \
255 { \
256 crq->crq_flags |= (1 << CFQ_CRQ_FLAG_##name); \
257 } \
258 static inline void cfq_clear_crq_##name(struct cfq_rq *crq) \
259 { \
260 crq->crq_flags &= ~(1 << CFQ_CRQ_FLAG_##name); \
261 } \
262 static inline int cfq_crq_##name(const struct cfq_rq *crq) \
263 { \
264 return (crq->crq_flags & (1 << CFQ_CRQ_FLAG_##name)) != 0; \
265 }
267 CFQ_CRQ_FNS(is_sync);
268 #undef CFQ_CRQ_FNS
270 static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short);
271 static void cfq_dispatch_insert(request_queue_t *, struct cfq_rq *);
272 static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, gfp_t gfp_mask);
274 /*
275 * lots of deadline iosched dupes, can be abstracted later...
276 */
277 static inline void cfq_del_crq_hash(struct cfq_rq *crq)
278 {
279 hlist_del_init(&crq->hash);
280 }
282 static inline void cfq_add_crq_hash(struct cfq_data *cfqd, struct cfq_rq *crq)
283 {
284 const int hash_idx = CFQ_MHASH_FN(rq_hash_key(crq->request));
286 hlist_add_head(&crq->hash, &cfqd->crq_hash[hash_idx]);
287 }
289 static struct request *cfq_find_rq_hash(struct cfq_data *cfqd, sector_t offset)
290 {
291 struct hlist_head *hash_list = &cfqd->crq_hash[CFQ_MHASH_FN(offset)];
292 struct hlist_node *entry, *next;
294 hlist_for_each_safe(entry, next, hash_list) {
295 struct cfq_rq *crq = list_entry_hash(entry);
296 struct request *__rq = crq->request;
298 if (!rq_mergeable(__rq)) {
299 cfq_del_crq_hash(crq);
300 continue;
301 }
303 if (rq_hash_key(__rq) == offset)
304 return __rq;
305 }
307 return NULL;
308 }
310 /*
311 * scheduler run of queue, if there are requests pending and no one in the
312 * driver that will restart queueing
313 */
314 static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
315 {
316 if (cfqd->busy_queues)
317 kblockd_schedule_work(&cfqd->unplug_work);
318 }
320 static int cfq_queue_empty(request_queue_t *q)
321 {
322 struct cfq_data *cfqd = q->elevator->elevator_data;
324 return !cfqd->busy_queues;
325 }
327 static inline pid_t cfq_queue_pid(struct task_struct *task, int rw)
328 {
329 if (rw == READ || rw == WRITE_SYNC)
330 return task->pid;
332 return CFQ_KEY_ASYNC;
333 }
335 /*
336 * Lifted from AS - choose which of crq1 and crq2 that is best served now.
337 * We choose the request that is closest to the head right now. Distance
338 * behind the head is penalized and only allowed to a certain extent.
339 */
340 static struct cfq_rq *
341 cfq_choose_req(struct cfq_data *cfqd, struct cfq_rq *crq1, struct cfq_rq *crq2)
342 {
343 sector_t last, s1, s2, d1 = 0, d2 = 0;
344 unsigned long back_max;
345 #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */
346 #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */
347 unsigned wrap = 0; /* bit mask: requests behind the disk head? */
349 if (crq1 == NULL || crq1 == crq2)
350 return crq2;
351 if (crq2 == NULL)
352 return crq1;
354 if (cfq_crq_is_sync(crq1) && !cfq_crq_is_sync(crq2))
355 return crq1;
356 else if (cfq_crq_is_sync(crq2) && !cfq_crq_is_sync(crq1))
357 return crq2;
359 s1 = crq1->request->sector;
360 s2 = crq2->request->sector;
362 last = cfqd->last_sector;
364 /*
365 * by definition, 1KiB is 2 sectors
366 */
367 back_max = cfqd->cfq_back_max * 2;
369 /*
370 * Strict one way elevator _except_ in the case where we allow
371 * short backward seeks which are biased as twice the cost of a
372 * similar forward seek.
373 */
374 if (s1 >= last)
375 d1 = s1 - last;
376 else if (s1 + back_max >= last)
377 d1 = (last - s1) * cfqd->cfq_back_penalty;
378 else
379 wrap |= CFQ_RQ1_WRAP;
381 if (s2 >= last)
382 d2 = s2 - last;
383 else if (s2 + back_max >= last)
384 d2 = (last - s2) * cfqd->cfq_back_penalty;
385 else
386 wrap |= CFQ_RQ2_WRAP;
388 /* Found required data */
390 /*
391 * By doing switch() on the bit mask "wrap" we avoid having to
392 * check two variables for all permutations: --> faster!
393 */
394 switch (wrap) {
395 case 0: /* common case for CFQ: crq1 and crq2 not wrapped */
396 if (d1 < d2)
397 return crq1;
398 else if (d2 < d1)
399 return crq2;
400 else {
401 if (s1 >= s2)
402 return crq1;
403 else
404 return crq2;
405 }
407 case CFQ_RQ2_WRAP:
408 return crq1;
409 case CFQ_RQ1_WRAP:
410 return crq2;
411 case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both crqs wrapped */
412 default:
413 /*
414 * Since both rqs are wrapped,
415 * start with the one that's further behind head
416 * (--> only *one* back seek required),
417 * since back seek takes more time than forward.
418 */
419 if (s1 <= s2)
420 return crq1;
421 else
422 return crq2;
423 }
424 }
426 /*
427 * would be nice to take fifo expire time into account as well
428 */
429 static struct cfq_rq *
430 cfq_find_next_crq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
431 struct cfq_rq *last)
432 {
433 struct cfq_rq *crq_next = NULL, *crq_prev = NULL;
434 struct rb_node *rbnext, *rbprev;
436 if (!(rbnext = rb_next(&last->rb_node))) {
437 rbnext = rb_first(&cfqq->sort_list);
438 if (rbnext == &last->rb_node)
439 rbnext = NULL;
440 }
442 rbprev = rb_prev(&last->rb_node);
444 if (rbprev)
445 crq_prev = rb_entry_crq(rbprev);
446 if (rbnext)
447 crq_next = rb_entry_crq(rbnext);
449 return cfq_choose_req(cfqd, crq_next, crq_prev);
450 }
452 static void cfq_update_next_crq(struct cfq_rq *crq)
453 {
454 struct cfq_queue *cfqq = crq->cfq_queue;
456 if (cfqq->next_crq == crq)
457 cfqq->next_crq = cfq_find_next_crq(cfqq->cfqd, cfqq, crq);
458 }
460 static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted)
461 {
462 struct cfq_data *cfqd = cfqq->cfqd;
463 struct list_head *list, *entry;
465 BUG_ON(!cfq_cfqq_on_rr(cfqq));
467 list_del(&cfqq->cfq_list);
469 if (cfq_class_rt(cfqq))
470 list = &cfqd->cur_rr;
471 else if (cfq_class_idle(cfqq))
472 list = &cfqd->idle_rr;
473 else {
474 /*
475 * if cfqq has requests in flight, don't allow it to be
476 * found in cfq_set_active_queue before it has finished them.
477 * this is done to increase fairness between a process that
478 * has lots of io pending vs one that only generates one
479 * sporadically or synchronously
480 */
481 if (cfq_cfqq_dispatched(cfqq))
482 list = &cfqd->busy_rr;
483 else
484 list = &cfqd->rr_list[cfqq->ioprio];
485 }
487 /*
488 * if queue was preempted, just add to front to be fair. busy_rr
489 * isn't sorted, but insert at the back for fairness.
490 */
491 if (preempted || list == &cfqd->busy_rr) {
492 if (preempted)
493 list = list->prev;
495 list_add_tail(&cfqq->cfq_list, list);
496 return;
497 }
499 /*
500 * sort by when queue was last serviced
501 */
502 entry = list;
503 while ((entry = entry->prev) != list) {
504 struct cfq_queue *__cfqq = list_entry_cfqq(entry);
506 if (!__cfqq->service_last)
507 break;
508 if (time_before(__cfqq->service_last, cfqq->service_last))
509 break;
510 }
512 list_add(&cfqq->cfq_list, entry);
513 }
515 /*
516 * add to busy list of queues for service, trying to be fair in ordering
517 * the pending list according to last request service
518 */
519 static inline void
520 cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
521 {
522 BUG_ON(cfq_cfqq_on_rr(cfqq));
523 cfq_mark_cfqq_on_rr(cfqq);
524 cfqd->busy_queues++;
526 cfq_resort_rr_list(cfqq, 0);
527 }
529 static inline void
530 cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
531 {
532 BUG_ON(!cfq_cfqq_on_rr(cfqq));
533 cfq_clear_cfqq_on_rr(cfqq);
534 list_move(&cfqq->cfq_list, &cfqd->empty_list);
536 BUG_ON(!cfqd->busy_queues);
537 cfqd->busy_queues--;
538 }
540 /*
541 * rb tree support functions
542 */
543 static inline void cfq_del_crq_rb(struct cfq_rq *crq)
544 {
545 struct cfq_queue *cfqq = crq->cfq_queue;
546 struct cfq_data *cfqd = cfqq->cfqd;
547 const int sync = cfq_crq_is_sync(crq);
549 BUG_ON(!cfqq->queued[sync]);
550 cfqq->queued[sync]--;
552 cfq_update_next_crq(crq);
554 rb_erase(&crq->rb_node, &cfqq->sort_list);
556 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
557 cfq_del_cfqq_rr(cfqd, cfqq);
558 }
560 static struct cfq_rq *
561 __cfq_add_crq_rb(struct cfq_rq *crq)
562 {
563 struct rb_node **p = &crq->cfq_queue->sort_list.rb_node;
564 struct rb_node *parent = NULL;
565 struct cfq_rq *__crq;
567 while (*p) {
568 parent = *p;
569 __crq = rb_entry_crq(parent);
571 if (crq->rb_key < __crq->rb_key)
572 p = &(*p)->rb_left;
573 else if (crq->rb_key > __crq->rb_key)
574 p = &(*p)->rb_right;
575 else
576 return __crq;
577 }
579 rb_link_node(&crq->rb_node, parent, p);
580 return NULL;
581 }
583 static void cfq_add_crq_rb(struct cfq_rq *crq)
584 {
585 struct cfq_queue *cfqq = crq->cfq_queue;
586 struct cfq_data *cfqd = cfqq->cfqd;
587 struct request *rq = crq->request;
588 struct cfq_rq *__alias;
590 crq->rb_key = rq_rb_key(rq);
591 cfqq->queued[cfq_crq_is_sync(crq)]++;
593 /*
594 * looks a little odd, but the first insert might return an alias.
595 * if that happens, put the alias on the dispatch list
596 */
597 while ((__alias = __cfq_add_crq_rb(crq)) != NULL)
598 cfq_dispatch_insert(cfqd->queue, __alias);
600 rb_insert_color(&crq->rb_node, &cfqq->sort_list);
602 if (!cfq_cfqq_on_rr(cfqq))
603 cfq_add_cfqq_rr(cfqd, cfqq);
605 /*
606 * check if this request is a better next-serve candidate
607 */
608 cfqq->next_crq = cfq_choose_req(cfqd, cfqq->next_crq, crq);
609 }
611 static inline void
612 cfq_reposition_crq_rb(struct cfq_queue *cfqq, struct cfq_rq *crq)
613 {
614 rb_erase(&crq->rb_node, &cfqq->sort_list);
615 cfqq->queued[cfq_crq_is_sync(crq)]--;
617 cfq_add_crq_rb(crq);
618 }
620 static struct request *
621 cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
622 {
623 struct task_struct *tsk = current;
624 pid_t key = cfq_queue_pid(tsk, bio_data_dir(bio));
625 struct cfq_queue *cfqq;
626 struct rb_node *n;
627 sector_t sector;
629 cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio);
630 if (!cfqq)
631 goto out;
633 sector = bio->bi_sector + bio_sectors(bio);
634 n = cfqq->sort_list.rb_node;
635 while (n) {
636 struct cfq_rq *crq = rb_entry_crq(n);
638 if (sector < crq->rb_key)
639 n = n->rb_left;
640 else if (sector > crq->rb_key)
641 n = n->rb_right;
642 else
643 return crq->request;
644 }
646 out:
647 return NULL;
648 }
650 static void cfq_activate_request(request_queue_t *q, struct request *rq)
651 {
652 struct cfq_data *cfqd = q->elevator->elevator_data;
654 cfqd->rq_in_driver++;
656 /*
657 * If the depth is larger 1, it really could be queueing. But lets
658 * make the mark a little higher - idling could still be good for
659 * low queueing, and a low queueing number could also just indicate
660 * a SCSI mid layer like behaviour where limit+1 is often seen.
661 */
662 if (!cfqd->hw_tag && cfqd->rq_in_driver > 4)
663 cfqd->hw_tag = 1;
664 }
666 static void cfq_deactivate_request(request_queue_t *q, struct request *rq)
667 {
668 struct cfq_data *cfqd = q->elevator->elevator_data;
670 WARN_ON(!cfqd->rq_in_driver);
671 cfqd->rq_in_driver--;
672 }
674 static void cfq_remove_request(struct request *rq)
675 {
676 struct cfq_rq *crq = RQ_DATA(rq);
678 list_del_init(&rq->queuelist);
679 cfq_del_crq_rb(crq);
680 cfq_del_crq_hash(crq);
681 }
683 static int
684 cfq_merge(request_queue_t *q, struct request **req, struct bio *bio)
685 {
686 struct cfq_data *cfqd = q->elevator->elevator_data;
687 struct request *__rq;
688 int ret;
690 __rq = cfq_find_rq_hash(cfqd, bio->bi_sector);
691 if (__rq && elv_rq_merge_ok(__rq, bio)) {
692 ret = ELEVATOR_BACK_MERGE;
693 goto out;
694 }
696 __rq = cfq_find_rq_fmerge(cfqd, bio);
697 if (__rq && elv_rq_merge_ok(__rq, bio)) {
698 ret = ELEVATOR_FRONT_MERGE;
699 goto out;
700 }
702 return ELEVATOR_NO_MERGE;
703 out:
704 *req = __rq;
705 return ret;
706 }
708 static void cfq_merged_request(request_queue_t *q, struct request *req)
709 {
710 struct cfq_data *cfqd = q->elevator->elevator_data;
711 struct cfq_rq *crq = RQ_DATA(req);
713 cfq_del_crq_hash(crq);
714 cfq_add_crq_hash(cfqd, crq);
716 if (rq_rb_key(req) != crq->rb_key) {
717 struct cfq_queue *cfqq = crq->cfq_queue;
719 cfq_update_next_crq(crq);
720 cfq_reposition_crq_rb(cfqq, crq);
721 }
722 }
724 static void
725 cfq_merged_requests(request_queue_t *q, struct request *rq,
726 struct request *next)
727 {
728 cfq_merged_request(q, rq);
730 /*
731 * reposition in fifo if next is older than rq
732 */
733 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
734 time_before(next->start_time, rq->start_time))
735 list_move(&rq->queuelist, &next->queuelist);
737 cfq_remove_request(next);
738 }
740 static inline void
741 __cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
742 {
743 if (cfqq) {
744 /*
745 * stop potential idle class queues waiting service
746 */
747 del_timer(&cfqd->idle_class_timer);
749 cfqq->slice_start = jiffies;
750 cfqq->slice_end = 0;
751 cfqq->slice_left = 0;
752 cfq_clear_cfqq_must_alloc_slice(cfqq);
753 cfq_clear_cfqq_fifo_expire(cfqq);
754 }
756 cfqd->active_queue = cfqq;
757 }
759 /*
760 * current cfqq expired its slice (or was too idle), select new one
761 */
762 static void
763 __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
764 int preempted)
765 {
766 unsigned long now = jiffies;
768 if (cfq_cfqq_wait_request(cfqq))
769 del_timer(&cfqd->idle_slice_timer);
771 if (!preempted && !cfq_cfqq_dispatched(cfqq)) {
772 cfqq->service_last = now;
773 cfq_schedule_dispatch(cfqd);
774 }
776 cfq_clear_cfqq_must_dispatch(cfqq);
777 cfq_clear_cfqq_wait_request(cfqq);
779 /*
780 * store what was left of this slice, if the queue idled out
781 * or was preempted
782 */
783 if (time_after(cfqq->slice_end, now))
784 cfqq->slice_left = cfqq->slice_end - now;
785 else
786 cfqq->slice_left = 0;
788 if (cfq_cfqq_on_rr(cfqq))
789 cfq_resort_rr_list(cfqq, preempted);
791 if (cfqq == cfqd->active_queue)
792 cfqd->active_queue = NULL;
794 if (cfqd->active_cic) {
795 put_io_context(cfqd->active_cic->ioc);
796 cfqd->active_cic = NULL;
797 }
799 cfqd->dispatch_slice = 0;
800 }
802 static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted)
803 {
804 struct cfq_queue *cfqq = cfqd->active_queue;
806 if (cfqq)
807 __cfq_slice_expired(cfqd, cfqq, preempted);
808 }
810 /*
811 * 0
812 * 0,1
813 * 0,1,2
814 * 0,1,2,3
815 * 0,1,2,3,4
816 * 0,1,2,3,4,5
817 * 0,1,2,3,4,5,6
818 * 0,1,2,3,4,5,6,7
819 */
820 static int cfq_get_next_prio_level(struct cfq_data *cfqd)
821 {
822 int prio, wrap;
824 prio = -1;
825 wrap = 0;
826 do {
827 int p;
829 for (p = cfqd->cur_prio; p <= cfqd->cur_end_prio; p++) {
830 if (!list_empty(&cfqd->rr_list[p])) {
831 prio = p;
832 break;
833 }
834 }
836 if (prio != -1)
837 break;
838 cfqd->cur_prio = 0;
839 if (++cfqd->cur_end_prio == CFQ_PRIO_LISTS) {
840 cfqd->cur_end_prio = 0;
841 if (wrap)
842 break;
843 wrap = 1;
844 }
845 } while (1);
847 if (unlikely(prio == -1))
848 return -1;
850 BUG_ON(prio >= CFQ_PRIO_LISTS);
852 list_splice_init(&cfqd->rr_list[prio], &cfqd->cur_rr);
854 cfqd->cur_prio = prio + 1;
855 if (cfqd->cur_prio > cfqd->cur_end_prio) {
856 cfqd->cur_end_prio = cfqd->cur_prio;
857 cfqd->cur_prio = 0;
858 }
859 if (cfqd->cur_end_prio == CFQ_PRIO_LISTS) {
860 cfqd->cur_prio = 0;
861 cfqd->cur_end_prio = 0;
862 }
864 return prio;
865 }
867 static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd)
868 {
869 struct cfq_queue *cfqq = NULL;
871 /*
872 * if current list is non-empty, grab first entry. if it is empty,
873 * get next prio level and grab first entry then if any are spliced
874 */
875 if (!list_empty(&cfqd->cur_rr) || cfq_get_next_prio_level(cfqd) != -1)
876 cfqq = list_entry_cfqq(cfqd->cur_rr.next);
878 /*
879 * If no new queues are available, check if the busy list has some
880 * before falling back to idle io.
881 */
882 if (!cfqq && !list_empty(&cfqd->busy_rr))
883 cfqq = list_entry_cfqq(cfqd->busy_rr.next);
885 /*
886 * if we have idle queues and no rt or be queues had pending
887 * requests, either allow immediate service if the grace period
888 * has passed or arm the idle grace timer
889 */
890 if (!cfqq && !list_empty(&cfqd->idle_rr)) {
891 unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE;
893 if (time_after_eq(jiffies, end))
894 cfqq = list_entry_cfqq(cfqd->idle_rr.next);
895 else
896 mod_timer(&cfqd->idle_class_timer, end);
897 }
899 __cfq_set_active_queue(cfqd, cfqq);
900 return cfqq;
901 }
903 #define CIC_SEEKY(cic) ((cic)->seek_mean > (128 * 1024))
905 static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
907 {
908 struct cfq_io_context *cic;
909 unsigned long sl;
911 WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
912 WARN_ON(cfqq != cfqd->active_queue);
914 /*
915 * idle is disabled, either manually or by past process history
916 */
917 if (!cfqd->cfq_slice_idle)
918 return 0;
919 if (!cfq_cfqq_idle_window(cfqq))
920 return 0;
921 /*
922 * task has exited, don't wait
923 */
924 cic = cfqd->active_cic;
925 if (!cic || !cic->ioc->task)
926 return 0;
928 cfq_mark_cfqq_must_dispatch(cfqq);
929 cfq_mark_cfqq_wait_request(cfqq);
931 sl = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle);
933 /*
934 * we don't want to idle for seeks, but we do want to allow
935 * fair distribution of slice time for a process doing back-to-back
936 * seeks. so allow a little bit of time for him to submit a new rq
937 */
938 if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
939 sl = min(sl, msecs_to_jiffies(2));
941 mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
942 return 1;
943 }
945 static void cfq_dispatch_insert(request_queue_t *q, struct cfq_rq *crq)
946 {
947 struct cfq_data *cfqd = q->elevator->elevator_data;
948 struct cfq_queue *cfqq = crq->cfq_queue;
949 struct request *rq;
951 cfqq->next_crq = cfq_find_next_crq(cfqd, cfqq, crq);
952 cfq_remove_request(crq->request);
953 cfqq->on_dispatch[cfq_crq_is_sync(crq)]++;
954 elv_dispatch_sort(q, crq->request);
956 rq = list_entry(q->queue_head.prev, struct request, queuelist);
957 cfqd->last_sector = rq->sector + rq->nr_sectors;
958 }
960 /*
961 * return expired entry, or NULL to just start from scratch in rbtree
962 */
963 static inline struct cfq_rq *cfq_check_fifo(struct cfq_queue *cfqq)
964 {
965 struct cfq_data *cfqd = cfqq->cfqd;
966 struct request *rq;
967 struct cfq_rq *crq;
969 if (cfq_cfqq_fifo_expire(cfqq))
970 return NULL;
972 if (!list_empty(&cfqq->fifo)) {
973 int fifo = cfq_cfqq_class_sync(cfqq);
975 crq = RQ_DATA(list_entry_fifo(cfqq->fifo.next));
976 rq = crq->request;
977 if (time_after(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) {
978 cfq_mark_cfqq_fifo_expire(cfqq);
979 return crq;
980 }
981 }
983 return NULL;
984 }
986 /*
987 * Scale schedule slice based on io priority. Use the sync time slice only
988 * if a queue is marked sync and has sync io queued. A sync queue with async
989 * io only, should not get full sync slice length.
990 */
991 static inline int
992 cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
993 {
994 const int base_slice = cfqd->cfq_slice[cfq_cfqq_sync(cfqq)];
996 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
998 return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - cfqq->ioprio));
999 }
1001 static inline void
1002 cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1004 cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
1007 static inline int
1008 cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1010 const int base_rq = cfqd->cfq_slice_async_rq;
1012 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
1014 return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
1017 /*
1018 * get next queue for service
1019 */
1020 static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
1022 unsigned long now = jiffies;
1023 struct cfq_queue *cfqq;
1025 cfqq = cfqd->active_queue;
1026 if (!cfqq)
1027 goto new_queue;
1029 /*
1030 * slice has expired
1031 */
1032 if (!cfq_cfqq_must_dispatch(cfqq) && time_after(now, cfqq->slice_end))
1033 goto expire;
1035 /*
1036 * if queue has requests, dispatch one. if not, check if
1037 * enough slice is left to wait for one
1038 */
1039 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
1040 goto keep_queue;
1041 else if (cfq_cfqq_dispatched(cfqq)) {
1042 cfqq = NULL;
1043 goto keep_queue;
1044 } else if (cfq_cfqq_class_sync(cfqq)) {
1045 if (cfq_arm_slice_timer(cfqd, cfqq))
1046 return NULL;
1049 expire:
1050 cfq_slice_expired(cfqd, 0);
1051 new_queue:
1052 cfqq = cfq_set_active_queue(cfqd);
1053 keep_queue:
1054 return cfqq;
1057 static int
1058 __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1059 int max_dispatch)
1061 int dispatched = 0;
1063 BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
1065 do {
1066 struct cfq_rq *crq;
1068 /*
1069 * follow expired path, else get first next available
1070 */
1071 if ((crq = cfq_check_fifo(cfqq)) == NULL)
1072 crq = cfqq->next_crq;
1074 /*
1075 * finally, insert request into driver dispatch list
1076 */
1077 cfq_dispatch_insert(cfqd->queue, crq);
1079 cfqd->dispatch_slice++;
1080 dispatched++;
1082 if (!cfqd->active_cic) {
1083 atomic_inc(&crq->io_context->ioc->refcount);
1084 cfqd->active_cic = crq->io_context;
1087 if (RB_EMPTY_ROOT(&cfqq->sort_list))
1088 break;
1090 } while (dispatched < max_dispatch);
1092 /*
1093 * if slice end isn't set yet, set it.
1094 */
1095 if (!cfqq->slice_end)
1096 cfq_set_prio_slice(cfqd, cfqq);
1098 /*
1099 * expire an async queue immediately if it has used up its slice. idle
1100 * queue always expire after 1 dispatch round.
1101 */
1102 if ((!cfq_cfqq_sync(cfqq) &&
1103 cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
1104 cfq_class_idle(cfqq) ||
1105 !cfq_cfqq_idle_window(cfqq))
1106 cfq_slice_expired(cfqd, 0);
1108 return dispatched;
1111 static int
1112 cfq_forced_dispatch_cfqqs(struct list_head *list)
1114 struct cfq_queue *cfqq, *next;
1115 struct cfq_rq *crq;
1116 int dispatched;
1118 dispatched = 0;
1119 list_for_each_entry_safe(cfqq, next, list, cfq_list) {
1120 while ((crq = cfqq->next_crq)) {
1121 cfq_dispatch_insert(cfqq->cfqd->queue, crq);
1122 dispatched++;
1124 BUG_ON(!list_empty(&cfqq->fifo));
1127 return dispatched;
1130 static int
1131 cfq_forced_dispatch(struct cfq_data *cfqd)
1133 int i, dispatched = 0;
1135 for (i = 0; i < CFQ_PRIO_LISTS; i++)
1136 dispatched += cfq_forced_dispatch_cfqqs(&cfqd->rr_list[i]);
1138 dispatched += cfq_forced_dispatch_cfqqs(&cfqd->busy_rr);
1139 dispatched += cfq_forced_dispatch_cfqqs(&cfqd->cur_rr);
1140 dispatched += cfq_forced_dispatch_cfqqs(&cfqd->idle_rr);
1142 cfq_slice_expired(cfqd, 0);
1144 BUG_ON(cfqd->busy_queues);
1146 return dispatched;
1149 static int
1150 cfq_dispatch_requests(request_queue_t *q, int force)
1152 struct cfq_data *cfqd = q->elevator->elevator_data;
1153 struct cfq_queue *cfqq, *prev_cfqq;
1154 int dispatched;
1156 if (!cfqd->busy_queues)
1157 return 0;
1159 if (unlikely(force))
1160 return cfq_forced_dispatch(cfqd);
1162 dispatched = 0;
1163 prev_cfqq = NULL;
1164 while ((cfqq = cfq_select_queue(cfqd)) != NULL) {
1165 int max_dispatch;
1167 /*
1168 * Don't repeat dispatch from the previous queue.
1169 */
1170 if (prev_cfqq == cfqq)
1171 break;
1173 cfq_clear_cfqq_must_dispatch(cfqq);
1174 cfq_clear_cfqq_wait_request(cfqq);
1175 del_timer(&cfqd->idle_slice_timer);
1177 max_dispatch = cfqd->cfq_quantum;
1178 if (cfq_class_idle(cfqq))
1179 max_dispatch = 1;
1181 dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch);
1183 /*
1184 * If the dispatch cfqq has idling enabled and is still
1185 * the active queue, break out.
1186 */
1187 if (cfq_cfqq_idle_window(cfqq) && cfqd->active_queue)
1188 break;
1190 prev_cfqq = cfqq;
1193 return dispatched;
1196 /*
1197 * task holds one reference to the queue, dropped when task exits. each crq
1198 * in-flight on this queue also holds a reference, dropped when crq is freed.
1200 * queue lock must be held here.
1201 */
1202 static void cfq_put_queue(struct cfq_queue *cfqq)
1204 struct cfq_data *cfqd = cfqq->cfqd;
1206 BUG_ON(atomic_read(&cfqq->ref) <= 0);
1208 if (!atomic_dec_and_test(&cfqq->ref))
1209 return;
1211 BUG_ON(rb_first(&cfqq->sort_list));
1212 BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
1213 BUG_ON(cfq_cfqq_on_rr(cfqq));
1215 if (unlikely(cfqd->active_queue == cfqq))
1216 __cfq_slice_expired(cfqd, cfqq, 0);
1218 /*
1219 * it's on the empty list and still hashed
1220 */
1221 list_del(&cfqq->cfq_list);
1222 hlist_del(&cfqq->cfq_hash);
1223 kmem_cache_free(cfq_pool, cfqq);
1226 static inline struct cfq_queue *
1227 __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio,
1228 const int hashval)
1230 struct hlist_head *hash_list = &cfqd->cfq_hash[hashval];
1231 struct hlist_node *entry;
1232 struct cfq_queue *__cfqq;
1234 hlist_for_each_entry(__cfqq, entry, hash_list, cfq_hash) {
1235 const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio);
1237 if (__cfqq->key == key && (__p == prio || !prio))
1238 return __cfqq;
1241 return NULL;
1244 static struct cfq_queue *
1245 cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio)
1247 return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT));
1250 static void cfq_free_io_context(struct io_context *ioc)
1252 struct cfq_io_context *__cic;
1253 struct rb_node *n;
1254 int freed = 0;
1256 while ((n = rb_first(&ioc->cic_root)) != NULL) {
1257 __cic = rb_entry(n, struct cfq_io_context, rb_node);
1258 rb_erase(&__cic->rb_node, &ioc->cic_root);
1259 kmem_cache_free(cfq_ioc_pool, __cic);
1260 freed++;
1263 if (atomic_sub_and_test(freed, &ioc_count) && ioc_gone)
1264 complete(ioc_gone);
1267 static void cfq_trim(struct io_context *ioc)
1269 ioc->set_ioprio = NULL;
1270 cfq_free_io_context(ioc);
1273 /*
1274 * Called with interrupts disabled
1275 */
1276 static void cfq_exit_single_io_context(struct cfq_io_context *cic)
1278 struct cfq_data *cfqd = cic->key;
1279 request_queue_t *q;
1281 if (!cfqd)
1282 return;
1284 q = cfqd->queue;
1286 WARN_ON(!irqs_disabled());
1288 spin_lock(q->queue_lock);
1290 if (cic->cfqq[ASYNC]) {
1291 if (unlikely(cic->cfqq[ASYNC] == cfqd->active_queue))
1292 __cfq_slice_expired(cfqd, cic->cfqq[ASYNC], 0);
1293 cfq_put_queue(cic->cfqq[ASYNC]);
1294 cic->cfqq[ASYNC] = NULL;
1297 if (cic->cfqq[SYNC]) {
1298 if (unlikely(cic->cfqq[SYNC] == cfqd->active_queue))
1299 __cfq_slice_expired(cfqd, cic->cfqq[SYNC], 0);
1300 cfq_put_queue(cic->cfqq[SYNC]);
1301 cic->cfqq[SYNC] = NULL;
1304 cic->key = NULL;
1305 list_del_init(&cic->queue_list);
1306 spin_unlock(q->queue_lock);
1309 static void cfq_exit_io_context(struct io_context *ioc)
1311 struct cfq_io_context *__cic;
1312 unsigned long flags;
1313 struct rb_node *n;
1315 /*
1316 * put the reference this task is holding to the various queues
1317 */
1318 spin_lock_irqsave(&cfq_exit_lock, flags);
1320 n = rb_first(&ioc->cic_root);
1321 while (n != NULL) {
1322 __cic = rb_entry(n, struct cfq_io_context, rb_node);
1324 cfq_exit_single_io_context(__cic);
1325 n = rb_next(n);
1328 spin_unlock_irqrestore(&cfq_exit_lock, flags);
1331 static struct cfq_io_context *
1332 cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1334 struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_mask);
1336 if (cic) {
1337 memset(cic, 0, sizeof(*cic));
1338 cic->last_end_request = jiffies;
1339 INIT_LIST_HEAD(&cic->queue_list);
1340 cic->dtor = cfq_free_io_context;
1341 cic->exit = cfq_exit_io_context;
1342 atomic_inc(&ioc_count);
1345 return cic;
1348 static void cfq_init_prio_data(struct cfq_queue *cfqq)
1350 struct task_struct *tsk = current;
1351 int ioprio_class;
1353 if (!cfq_cfqq_prio_changed(cfqq))
1354 return;
1356 ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio);
1357 switch (ioprio_class) {
1358 default:
1359 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
1360 case IOPRIO_CLASS_NONE:
1361 /*
1362 * no prio set, place us in the middle of the BE classes
1363 */
1364 cfqq->ioprio = task_nice_ioprio(tsk);
1365 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1366 break;
1367 case IOPRIO_CLASS_RT:
1368 cfqq->ioprio = task_ioprio(tsk);
1369 cfqq->ioprio_class = IOPRIO_CLASS_RT;
1370 break;
1371 case IOPRIO_CLASS_BE:
1372 cfqq->ioprio = task_ioprio(tsk);
1373 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1374 break;
1375 case IOPRIO_CLASS_IDLE:
1376 cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
1377 cfqq->ioprio = 7;
1378 cfq_clear_cfqq_idle_window(cfqq);
1379 break;
1382 /*
1383 * keep track of original prio settings in case we have to temporarily
1384 * elevate the priority of this queue
1385 */
1386 cfqq->org_ioprio = cfqq->ioprio;
1387 cfqq->org_ioprio_class = cfqq->ioprio_class;
1389 if (cfq_cfqq_on_rr(cfqq))
1390 cfq_resort_rr_list(cfqq, 0);
1392 cfq_clear_cfqq_prio_changed(cfqq);
1395 static inline void changed_ioprio(struct cfq_io_context *cic)
1397 struct cfq_data *cfqd = cic->key;
1398 struct cfq_queue *cfqq;
1400 if (unlikely(!cfqd))
1401 return;
1403 spin_lock(cfqd->queue->queue_lock);
1405 cfqq = cic->cfqq[ASYNC];
1406 if (cfqq) {
1407 struct cfq_queue *new_cfqq;
1408 new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC, cic->ioc->task,
1409 GFP_ATOMIC);
1410 if (new_cfqq) {
1411 cic->cfqq[ASYNC] = new_cfqq;
1412 cfq_put_queue(cfqq);
1416 cfqq = cic->cfqq[SYNC];
1417 if (cfqq)
1418 cfq_mark_cfqq_prio_changed(cfqq);
1420 spin_unlock(cfqd->queue->queue_lock);
1423 /*
1424 * callback from sys_ioprio_set, irqs are disabled
1425 */
1426 static int cfq_ioc_set_ioprio(struct io_context *ioc, unsigned int ioprio)
1428 struct cfq_io_context *cic;
1429 struct rb_node *n;
1431 spin_lock(&cfq_exit_lock);
1433 n = rb_first(&ioc->cic_root);
1434 while (n != NULL) {
1435 cic = rb_entry(n, struct cfq_io_context, rb_node);
1437 changed_ioprio(cic);
1438 n = rb_next(n);
1441 spin_unlock(&cfq_exit_lock);
1443 return 0;
1446 static struct cfq_queue *
1447 cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk,
1448 gfp_t gfp_mask)
1450 const int hashval = hash_long(key, CFQ_QHASH_SHIFT);
1451 struct cfq_queue *cfqq, *new_cfqq = NULL;
1452 unsigned short ioprio;
1454 retry:
1455 ioprio = tsk->ioprio;
1456 cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval);
1458 if (!cfqq) {
1459 if (new_cfqq) {
1460 cfqq = new_cfqq;
1461 new_cfqq = NULL;
1462 } else if (gfp_mask & __GFP_WAIT) {
1463 spin_unlock_irq(cfqd->queue->queue_lock);
1464 new_cfqq = kmem_cache_alloc(cfq_pool, gfp_mask);
1465 spin_lock_irq(cfqd->queue->queue_lock);
1466 goto retry;
1467 } else {
1468 cfqq = kmem_cache_alloc(cfq_pool, gfp_mask);
1469 if (!cfqq)
1470 goto out;
1473 memset(cfqq, 0, sizeof(*cfqq));
1475 INIT_HLIST_NODE(&cfqq->cfq_hash);
1476 INIT_LIST_HEAD(&cfqq->cfq_list);
1477 INIT_LIST_HEAD(&cfqq->fifo);
1479 cfqq->key = key;
1480 hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]);
1481 atomic_set(&cfqq->ref, 0);
1482 cfqq->cfqd = cfqd;
1483 cfqq->service_last = 0;
1484 /*
1485 * set ->slice_left to allow preemption for a new process
1486 */
1487 cfqq->slice_left = 2 * cfqd->cfq_slice_idle;
1488 cfq_mark_cfqq_idle_window(cfqq);
1489 cfq_mark_cfqq_prio_changed(cfqq);
1490 cfq_init_prio_data(cfqq);
1493 if (new_cfqq)
1494 kmem_cache_free(cfq_pool, new_cfqq);
1496 atomic_inc(&cfqq->ref);
1497 out:
1498 WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq);
1499 return cfqq;
1502 static void
1503 cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic)
1505 spin_lock(&cfq_exit_lock);
1506 rb_erase(&cic->rb_node, &ioc->cic_root);
1507 list_del_init(&cic->queue_list);
1508 spin_unlock(&cfq_exit_lock);
1509 kmem_cache_free(cfq_ioc_pool, cic);
1510 atomic_dec(&ioc_count);
1513 static struct cfq_io_context *
1514 cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc)
1516 struct rb_node *n;
1517 struct cfq_io_context *cic;
1518 void *k, *key = cfqd;
1520 restart:
1521 n = ioc->cic_root.rb_node;
1522 while (n) {
1523 cic = rb_entry(n, struct cfq_io_context, rb_node);
1524 /* ->key must be copied to avoid race with cfq_exit_queue() */
1525 k = cic->key;
1526 if (unlikely(!k)) {
1527 cfq_drop_dead_cic(ioc, cic);
1528 goto restart;
1531 if (key < k)
1532 n = n->rb_left;
1533 else if (key > k)
1534 n = n->rb_right;
1535 else
1536 return cic;
1539 return NULL;
1542 static inline void
1543 cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
1544 struct cfq_io_context *cic)
1546 struct rb_node **p;
1547 struct rb_node *parent;
1548 struct cfq_io_context *__cic;
1549 void *k;
1551 cic->ioc = ioc;
1552 cic->key = cfqd;
1554 ioc->set_ioprio = cfq_ioc_set_ioprio;
1555 restart:
1556 parent = NULL;
1557 p = &ioc->cic_root.rb_node;
1558 while (*p) {
1559 parent = *p;
1560 __cic = rb_entry(parent, struct cfq_io_context, rb_node);
1561 /* ->key must be copied to avoid race with cfq_exit_queue() */
1562 k = __cic->key;
1563 if (unlikely(!k)) {
1564 cfq_drop_dead_cic(ioc, __cic);
1565 goto restart;
1568 if (cic->key < k)
1569 p = &(*p)->rb_left;
1570 else if (cic->key > k)
1571 p = &(*p)->rb_right;
1572 else
1573 BUG();
1576 spin_lock(&cfq_exit_lock);
1577 rb_link_node(&cic->rb_node, parent, p);
1578 rb_insert_color(&cic->rb_node, &ioc->cic_root);
1579 list_add(&cic->queue_list, &cfqd->cic_list);
1580 spin_unlock(&cfq_exit_lock);
1583 /*
1584 * Setup general io context and cfq io context. There can be several cfq
1585 * io contexts per general io context, if this process is doing io to more
1586 * than one device managed by cfq.
1587 */
1588 static struct cfq_io_context *
1589 cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1591 struct io_context *ioc = NULL;
1592 struct cfq_io_context *cic;
1594 might_sleep_if(gfp_mask & __GFP_WAIT);
1596 ioc = get_io_context(gfp_mask);
1597 if (!ioc)
1598 return NULL;
1600 cic = cfq_cic_rb_lookup(cfqd, ioc);
1601 if (cic)
1602 goto out;
1604 cic = cfq_alloc_io_context(cfqd, gfp_mask);
1605 if (cic == NULL)
1606 goto err;
1608 cfq_cic_link(cfqd, ioc, cic);
1609 out:
1610 return cic;
1611 err:
1612 put_io_context(ioc);
1613 return NULL;
1616 static void
1617 cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
1619 unsigned long elapsed, ttime;
1621 /*
1622 * if this context already has stuff queued, thinktime is from
1623 * last queue not last end
1624 */
1625 #if 0
1626 if (time_after(cic->last_end_request, cic->last_queue))
1627 elapsed = jiffies - cic->last_end_request;
1628 else
1629 elapsed = jiffies - cic->last_queue;
1630 #else
1631 elapsed = jiffies - cic->last_end_request;
1632 #endif
1634 ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle);
1636 cic->ttime_samples = (7*cic->ttime_samples + 256) / 8;
1637 cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8;
1638 cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples;
1641 static void
1642 cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic,
1643 struct cfq_rq *crq)
1645 sector_t sdist;
1646 u64 total;
1648 if (cic->last_request_pos < crq->request->sector)
1649 sdist = crq->request->sector - cic->last_request_pos;
1650 else
1651 sdist = cic->last_request_pos - crq->request->sector;
1653 /*
1654 * Don't allow the seek distance to get too large from the
1655 * odd fragment, pagein, etc
1656 */
1657 if (cic->seek_samples <= 60) /* second&third seek */
1658 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024);
1659 else
1660 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64);
1662 cic->seek_samples = (7*cic->seek_samples + 256) / 8;
1663 cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8;
1664 total = cic->seek_total + (cic->seek_samples/2);
1665 do_div(total, cic->seek_samples);
1666 cic->seek_mean = (sector_t)total;
1669 /*
1670 * Disable idle window if the process thinks too long or seeks so much that
1671 * it doesn't matter
1672 */
1673 static void
1674 cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1675 struct cfq_io_context *cic)
1677 int enable_idle = cfq_cfqq_idle_window(cfqq);
1679 if (!cic->ioc->task || !cfqd->cfq_slice_idle ||
1680 (cfqd->hw_tag && CIC_SEEKY(cic)))
1681 enable_idle = 0;
1682 else if (sample_valid(cic->ttime_samples)) {
1683 if (cic->ttime_mean > cfqd->cfq_slice_idle)
1684 enable_idle = 0;
1685 else
1686 enable_idle = 1;
1689 if (enable_idle)
1690 cfq_mark_cfqq_idle_window(cfqq);
1691 else
1692 cfq_clear_cfqq_idle_window(cfqq);
1696 /*
1697 * Check if new_cfqq should preempt the currently active queue. Return 0 for
1698 * no or if we aren't sure, a 1 will cause a preempt.
1699 */
1700 static int
1701 cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
1702 struct cfq_rq *crq)
1704 struct cfq_queue *cfqq = cfqd->active_queue;
1706 if (cfq_class_idle(new_cfqq))
1707 return 0;
1709 if (!cfqq)
1710 return 0;
1712 if (cfq_class_idle(cfqq))
1713 return 1;
1714 if (!cfq_cfqq_wait_request(new_cfqq))
1715 return 0;
1716 /*
1717 * if it doesn't have slice left, forget it
1718 */
1719 if (new_cfqq->slice_left < cfqd->cfq_slice_idle)
1720 return 0;
1721 if (cfq_crq_is_sync(crq) && !cfq_cfqq_sync(cfqq))
1722 return 1;
1724 return 0;
1727 /*
1728 * cfqq preempts the active queue. if we allowed preempt with no slice left,
1729 * let it have half of its nominal slice.
1730 */
1731 static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1733 struct cfq_queue *__cfqq, *next;
1735 list_for_each_entry_safe(__cfqq, next, &cfqd->cur_rr, cfq_list)
1736 cfq_resort_rr_list(__cfqq, 1);
1738 if (!cfqq->slice_left)
1739 cfqq->slice_left = cfq_prio_to_slice(cfqd, cfqq) / 2;
1741 cfqq->slice_end = cfqq->slice_left + jiffies;
1742 cfq_slice_expired(cfqd, 1);
1743 __cfq_set_active_queue(cfqd, cfqq);
1746 /*
1747 * should really be a ll_rw_blk.c helper
1748 */
1749 static void cfq_start_queueing(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1751 request_queue_t *q = cfqd->queue;
1753 if (!blk_queue_plugged(q))
1754 q->request_fn(q);
1755 else
1756 __generic_unplug_device(q);
1759 /*
1760 * Called when a new fs request (crq) is added (to cfqq). Check if there's
1761 * something we should do about it
1762 */
1763 static void
1764 cfq_crq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1765 struct cfq_rq *crq)
1767 struct cfq_io_context *cic = crq->io_context;
1769 /*
1770 * we never wait for an async request and we don't allow preemption
1771 * of an async request. so just return early
1772 */
1773 if (!cfq_crq_is_sync(crq)) {
1774 /*
1775 * sync process issued an async request, if it's waiting
1776 * then expire it and kick rq handling.
1777 */
1778 if (cic == cfqd->active_cic &&
1779 del_timer(&cfqd->idle_slice_timer)) {
1780 cfq_slice_expired(cfqd, 0);
1781 cfq_start_queueing(cfqd, cfqq);
1783 return;
1786 cfq_update_io_thinktime(cfqd, cic);
1787 cfq_update_io_seektime(cfqd, cic, crq);
1788 cfq_update_idle_window(cfqd, cfqq, cic);
1790 cic->last_queue = jiffies;
1791 cic->last_request_pos = crq->request->sector + crq->request->nr_sectors;
1793 if (cfqq == cfqd->active_queue) {
1794 /*
1795 * if we are waiting for a request for this queue, let it rip
1796 * immediately and flag that we must not expire this queue
1797 * just now
1798 */
1799 if (cfq_cfqq_wait_request(cfqq)) {
1800 cfq_mark_cfqq_must_dispatch(cfqq);
1801 del_timer(&cfqd->idle_slice_timer);
1802 cfq_start_queueing(cfqd, cfqq);
1804 } else if (cfq_should_preempt(cfqd, cfqq, crq)) {
1805 /*
1806 * not the active queue - expire current slice if it is
1807 * idle and has expired it's mean thinktime or this new queue
1808 * has some old slice time left and is of higher priority
1809 */
1810 cfq_preempt_queue(cfqd, cfqq);
1811 cfq_mark_cfqq_must_dispatch(cfqq);
1812 cfq_start_queueing(cfqd, cfqq);
1816 static void cfq_insert_request(request_queue_t *q, struct request *rq)
1818 struct cfq_data *cfqd = q->elevator->elevator_data;
1819 struct cfq_rq *crq = RQ_DATA(rq);
1820 struct cfq_queue *cfqq = crq->cfq_queue;
1822 cfq_init_prio_data(cfqq);
1824 cfq_add_crq_rb(crq);
1826 list_add_tail(&rq->queuelist, &cfqq->fifo);
1828 if (rq_mergeable(rq))
1829 cfq_add_crq_hash(cfqd, crq);
1831 cfq_crq_enqueued(cfqd, cfqq, crq);
1834 static void cfq_completed_request(request_queue_t *q, struct request *rq)
1836 struct cfq_rq *crq = RQ_DATA(rq);
1837 struct cfq_queue *cfqq = crq->cfq_queue;
1838 struct cfq_data *cfqd = cfqq->cfqd;
1839 const int sync = cfq_crq_is_sync(crq);
1840 unsigned long now;
1842 now = jiffies;
1844 WARN_ON(!cfqd->rq_in_driver);
1845 WARN_ON(!cfqq->on_dispatch[sync]);
1846 cfqd->rq_in_driver--;
1847 cfqq->on_dispatch[sync]--;
1849 if (!cfq_class_idle(cfqq))
1850 cfqd->last_end_request = now;
1852 if (!cfq_cfqq_dispatched(cfqq)) {
1853 if (cfq_cfqq_on_rr(cfqq)) {
1854 cfqq->service_last = now;
1855 cfq_resort_rr_list(cfqq, 0);
1859 if (sync)
1860 crq->io_context->last_end_request = now;
1862 /*
1863 * If this is the active queue, check if it needs to be expired,
1864 * or if we want to idle in case it has no pending requests.
1865 */
1866 if (cfqd->active_queue == cfqq) {
1867 if (time_after(now, cfqq->slice_end))
1868 cfq_slice_expired(cfqd, 0);
1869 else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) {
1870 if (!cfq_arm_slice_timer(cfqd, cfqq))
1871 cfq_schedule_dispatch(cfqd);
1876 static struct request *
1877 cfq_former_request(request_queue_t *q, struct request *rq)
1879 struct cfq_rq *crq = RQ_DATA(rq);
1880 struct rb_node *rbprev = rb_prev(&crq->rb_node);
1882 if (rbprev)
1883 return rb_entry_crq(rbprev)->request;
1885 return NULL;
1888 static struct request *
1889 cfq_latter_request(request_queue_t *q, struct request *rq)
1891 struct cfq_rq *crq = RQ_DATA(rq);
1892 struct rb_node *rbnext = rb_next(&crq->rb_node);
1894 if (rbnext)
1895 return rb_entry_crq(rbnext)->request;
1897 return NULL;
1900 /*
1901 * we temporarily boost lower priority queues if they are holding fs exclusive
1902 * resources. they are boosted to normal prio (CLASS_BE/4)
1903 */
1904 static void cfq_prio_boost(struct cfq_queue *cfqq)
1906 const int ioprio_class = cfqq->ioprio_class;
1907 const int ioprio = cfqq->ioprio;
1909 if (has_fs_excl()) {
1910 /*
1911 * boost idle prio on transactions that would lock out other
1912 * users of the filesystem
1913 */
1914 if (cfq_class_idle(cfqq))
1915 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1916 if (cfqq->ioprio > IOPRIO_NORM)
1917 cfqq->ioprio = IOPRIO_NORM;
1918 } else {
1919 /*
1920 * check if we need to unboost the queue
1921 */
1922 if (cfqq->ioprio_class != cfqq->org_ioprio_class)
1923 cfqq->ioprio_class = cfqq->org_ioprio_class;
1924 if (cfqq->ioprio != cfqq->org_ioprio)
1925 cfqq->ioprio = cfqq->org_ioprio;
1928 /*
1929 * refile between round-robin lists if we moved the priority class
1930 */
1931 if ((ioprio_class != cfqq->ioprio_class || ioprio != cfqq->ioprio) &&
1932 cfq_cfqq_on_rr(cfqq))
1933 cfq_resort_rr_list(cfqq, 0);
1936 static inline int
1937 __cfq_may_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1938 struct task_struct *task, int rw)
1940 if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) &&
1941 !cfq_cfqq_must_alloc_slice(cfqq)) {
1942 cfq_mark_cfqq_must_alloc_slice(cfqq);
1943 return ELV_MQUEUE_MUST;
1946 return ELV_MQUEUE_MAY;
1949 static int cfq_may_queue(request_queue_t *q, int rw, struct bio *bio)
1951 struct cfq_data *cfqd = q->elevator->elevator_data;
1952 struct task_struct *tsk = current;
1953 struct cfq_queue *cfqq;
1955 /*
1956 * don't force setup of a queue from here, as a call to may_queue
1957 * does not necessarily imply that a request actually will be queued.
1958 * so just lookup a possibly existing queue, or return 'may queue'
1959 * if that fails
1960 */
1961 cfqq = cfq_find_cfq_hash(cfqd, cfq_queue_pid(tsk, rw), tsk->ioprio);
1962 if (cfqq) {
1963 cfq_init_prio_data(cfqq);
1964 cfq_prio_boost(cfqq);
1966 return __cfq_may_queue(cfqd, cfqq, tsk, rw);
1969 return ELV_MQUEUE_MAY;
1972 static void cfq_check_waiters(request_queue_t *q, struct cfq_queue *cfqq)
1974 struct cfq_data *cfqd = q->elevator->elevator_data;
1976 if (unlikely(cfqd->rq_starved)) {
1977 struct request_list *rl = &q->rq;
1979 smp_mb();
1980 if (waitqueue_active(&rl->wait[READ]))
1981 wake_up(&rl->wait[READ]);
1982 if (waitqueue_active(&rl->wait[WRITE]))
1983 wake_up(&rl->wait[WRITE]);
1987 /*
1988 * queue lock held here
1989 */
1990 static void cfq_put_request(request_queue_t *q, struct request *rq)
1992 struct cfq_data *cfqd = q->elevator->elevator_data;
1993 struct cfq_rq *crq = RQ_DATA(rq);
1995 if (crq) {
1996 struct cfq_queue *cfqq = crq->cfq_queue;
1997 const int rw = rq_data_dir(rq);
1999 BUG_ON(!cfqq->allocated[rw]);
2000 cfqq->allocated[rw]--;
2002 put_io_context(crq->io_context->ioc);
2004 mempool_free(crq, cfqd->crq_pool);
2005 rq->elevator_private = NULL;
2007 cfq_check_waiters(q, cfqq);
2008 cfq_put_queue(cfqq);
2012 /*
2013 * Allocate cfq data structures associated with this request.
2014 */
2015 static int
2016 cfq_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
2017 gfp_t gfp_mask)
2019 struct cfq_data *cfqd = q->elevator->elevator_data;
2020 struct task_struct *tsk = current;
2021 struct cfq_io_context *cic;
2022 const int rw = rq_data_dir(rq);
2023 pid_t key = cfq_queue_pid(tsk, rw);
2024 struct cfq_queue *cfqq;
2025 struct cfq_rq *crq;
2026 unsigned long flags;
2027 int is_sync = key != CFQ_KEY_ASYNC;
2029 might_sleep_if(gfp_mask & __GFP_WAIT);
2031 cic = cfq_get_io_context(cfqd, gfp_mask);
2033 spin_lock_irqsave(q->queue_lock, flags);
2035 if (!cic)
2036 goto queue_fail;
2038 if (!cic->cfqq[is_sync]) {
2039 cfqq = cfq_get_queue(cfqd, key, tsk, gfp_mask);
2040 if (!cfqq)
2041 goto queue_fail;
2043 cic->cfqq[is_sync] = cfqq;
2044 } else
2045 cfqq = cic->cfqq[is_sync];
2047 cfqq->allocated[rw]++;
2048 cfq_clear_cfqq_must_alloc(cfqq);
2049 cfqd->rq_starved = 0;
2050 atomic_inc(&cfqq->ref);
2051 spin_unlock_irqrestore(q->queue_lock, flags);
2053 crq = mempool_alloc(cfqd->crq_pool, gfp_mask);
2054 if (crq) {
2055 RB_CLEAR_NODE(&crq->rb_node);
2056 crq->rb_key = 0;
2057 crq->request = rq;
2058 INIT_HLIST_NODE(&crq->hash);
2059 crq->cfq_queue = cfqq;
2060 crq->io_context = cic;
2062 if (is_sync)
2063 cfq_mark_crq_is_sync(crq);
2064 else
2065 cfq_clear_crq_is_sync(crq);
2067 rq->elevator_private = crq;
2068 return 0;
2071 spin_lock_irqsave(q->queue_lock, flags);
2072 cfqq->allocated[rw]--;
2073 if (!(cfqq->allocated[0] + cfqq->allocated[1]))
2074 cfq_mark_cfqq_must_alloc(cfqq);
2075 cfq_put_queue(cfqq);
2076 queue_fail:
2077 if (cic)
2078 put_io_context(cic->ioc);
2079 /*
2080 * mark us rq allocation starved. we need to kickstart the process
2081 * ourselves if there are no pending requests that can do it for us.
2082 * that would be an extremely rare OOM situation
2083 */
2084 cfqd->rq_starved = 1;
2085 cfq_schedule_dispatch(cfqd);
2086 spin_unlock_irqrestore(q->queue_lock, flags);
2087 return 1;
2090 static void cfq_kick_queue(void *data)
2092 request_queue_t *q = data;
2093 struct cfq_data *cfqd = q->elevator->elevator_data;
2094 unsigned long flags;
2096 spin_lock_irqsave(q->queue_lock, flags);
2098 if (cfqd->rq_starved) {
2099 struct request_list *rl = &q->rq;
2101 /*
2102 * we aren't guaranteed to get a request after this, but we
2103 * have to be opportunistic
2104 */
2105 smp_mb();
2106 if (waitqueue_active(&rl->wait[READ]))
2107 wake_up(&rl->wait[READ]);
2108 if (waitqueue_active(&rl->wait[WRITE]))
2109 wake_up(&rl->wait[WRITE]);
2112 blk_remove_plug(q);
2113 q->request_fn(q);
2114 spin_unlock_irqrestore(q->queue_lock, flags);
2117 /*
2118 * Timer running if the active_queue is currently idling inside its time slice
2119 */
2120 static void cfq_idle_slice_timer(unsigned long data)
2122 struct cfq_data *cfqd = (struct cfq_data *) data;
2123 struct cfq_queue *cfqq;
2124 unsigned long flags;
2126 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2128 if ((cfqq = cfqd->active_queue) != NULL) {
2129 unsigned long now = jiffies;
2131 /*
2132 * expired
2133 */
2134 if (time_after(now, cfqq->slice_end))
2135 goto expire;
2137 /*
2138 * only expire and reinvoke request handler, if there are
2139 * other queues with pending requests
2140 */
2141 if (!cfqd->busy_queues)
2142 goto out_cont;
2144 /*
2145 * not expired and it has a request pending, let it dispatch
2146 */
2147 if (!RB_EMPTY_ROOT(&cfqq->sort_list)) {
2148 cfq_mark_cfqq_must_dispatch(cfqq);
2149 goto out_kick;
2152 expire:
2153 cfq_slice_expired(cfqd, 0);
2154 out_kick:
2155 cfq_schedule_dispatch(cfqd);
2156 out_cont:
2157 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2160 /*
2161 * Timer running if an idle class queue is waiting for service
2162 */
2163 static void cfq_idle_class_timer(unsigned long data)
2165 struct cfq_data *cfqd = (struct cfq_data *) data;
2166 unsigned long flags, end;
2168 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2170 /*
2171 * race with a non-idle queue, reset timer
2172 */
2173 end = cfqd->last_end_request + CFQ_IDLE_GRACE;
2174 if (!time_after_eq(jiffies, end))
2175 mod_timer(&cfqd->idle_class_timer, end);
2176 else
2177 cfq_schedule_dispatch(cfqd);
2179 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2182 static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
2184 del_timer_sync(&cfqd->idle_slice_timer);
2185 del_timer_sync(&cfqd->idle_class_timer);
2186 blk_sync_queue(cfqd->queue);
2189 static void cfq_exit_queue(elevator_t *e)
2191 struct cfq_data *cfqd = e->elevator_data;
2192 request_queue_t *q = cfqd->queue;
2194 cfq_shutdown_timer_wq(cfqd);
2196 spin_lock(&cfq_exit_lock);
2197 spin_lock_irq(q->queue_lock);
2199 if (cfqd->active_queue)
2200 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
2202 while (!list_empty(&cfqd->cic_list)) {
2203 struct cfq_io_context *cic = list_entry(cfqd->cic_list.next,
2204 struct cfq_io_context,
2205 queue_list);
2206 if (cic->cfqq[ASYNC]) {
2207 cfq_put_queue(cic->cfqq[ASYNC]);
2208 cic->cfqq[ASYNC] = NULL;
2210 if (cic->cfqq[SYNC]) {
2211 cfq_put_queue(cic->cfqq[SYNC]);
2212 cic->cfqq[SYNC] = NULL;
2214 cic->key = NULL;
2215 list_del_init(&cic->queue_list);
2218 spin_unlock_irq(q->queue_lock);
2219 spin_unlock(&cfq_exit_lock);
2221 cfq_shutdown_timer_wq(cfqd);
2223 mempool_destroy(cfqd->crq_pool);
2224 kfree(cfqd->crq_hash);
2225 kfree(cfqd->cfq_hash);
2226 kfree(cfqd);
2229 static void *cfq_init_queue(request_queue_t *q, elevator_t *e)
2231 struct cfq_data *cfqd;
2232 int i;
2234 cfqd = kmalloc(sizeof(*cfqd), GFP_KERNEL);
2235 if (!cfqd)
2236 return NULL;
2238 memset(cfqd, 0, sizeof(*cfqd));
2240 for (i = 0; i < CFQ_PRIO_LISTS; i++)
2241 INIT_LIST_HEAD(&cfqd->rr_list[i]);
2243 INIT_LIST_HEAD(&cfqd->busy_rr);
2244 INIT_LIST_HEAD(&cfqd->cur_rr);
2245 INIT_LIST_HEAD(&cfqd->idle_rr);
2246 INIT_LIST_HEAD(&cfqd->empty_list);
2247 INIT_LIST_HEAD(&cfqd->cic_list);
2249 cfqd->crq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_MHASH_ENTRIES, GFP_KERNEL);
2250 if (!cfqd->crq_hash)
2251 goto out_crqhash;
2253 cfqd->cfq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_QHASH_ENTRIES, GFP_KERNEL);
2254 if (!cfqd->cfq_hash)
2255 goto out_cfqhash;
2257 cfqd->crq_pool = mempool_create_slab_pool(BLKDEV_MIN_RQ, crq_pool);
2258 if (!cfqd->crq_pool)
2259 goto out_crqpool;
2261 for (i = 0; i < CFQ_MHASH_ENTRIES; i++)
2262 INIT_HLIST_HEAD(&cfqd->crq_hash[i]);
2263 for (i = 0; i < CFQ_QHASH_ENTRIES; i++)
2264 INIT_HLIST_HEAD(&cfqd->cfq_hash[i]);
2266 cfqd->queue = q;
2268 init_timer(&cfqd->idle_slice_timer);
2269 cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
2270 cfqd->idle_slice_timer.data = (unsigned long) cfqd;
2272 init_timer(&cfqd->idle_class_timer);
2273 cfqd->idle_class_timer.function = cfq_idle_class_timer;
2274 cfqd->idle_class_timer.data = (unsigned long) cfqd;
2276 INIT_WORK(&cfqd->unplug_work, cfq_kick_queue, q);
2278 cfqd->cfq_queued = cfq_queued;
2279 cfqd->cfq_quantum = cfq_quantum;
2280 cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
2281 cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
2282 cfqd->cfq_back_max = cfq_back_max;
2283 cfqd->cfq_back_penalty = cfq_back_penalty;
2284 cfqd->cfq_slice[0] = cfq_slice_async;
2285 cfqd->cfq_slice[1] = cfq_slice_sync;
2286 cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
2287 cfqd->cfq_slice_idle = cfq_slice_idle;
2289 return cfqd;
2290 out_crqpool:
2291 kfree(cfqd->cfq_hash);
2292 out_cfqhash:
2293 kfree(cfqd->crq_hash);
2294 out_crqhash:
2295 kfree(cfqd);
2296 return NULL;
2299 static void cfq_slab_kill(void)
2301 if (crq_pool)
2302 kmem_cache_destroy(crq_pool);
2303 if (cfq_pool)
2304 kmem_cache_destroy(cfq_pool);
2305 if (cfq_ioc_pool)
2306 kmem_cache_destroy(cfq_ioc_pool);
2309 static int __init cfq_slab_setup(void)
2311 crq_pool = kmem_cache_create("crq_pool", sizeof(struct cfq_rq), 0, 0,
2312 NULL, NULL);
2313 if (!crq_pool)
2314 goto fail;
2316 cfq_pool = kmem_cache_create("cfq_pool", sizeof(struct cfq_queue), 0, 0,
2317 NULL, NULL);
2318 if (!cfq_pool)
2319 goto fail;
2321 cfq_ioc_pool = kmem_cache_create("cfq_ioc_pool",
2322 sizeof(struct cfq_io_context), 0, 0, NULL, NULL);
2323 if (!cfq_ioc_pool)
2324 goto fail;
2326 return 0;
2327 fail:
2328 cfq_slab_kill();
2329 return -ENOMEM;
2332 /*
2333 * sysfs parts below -->
2334 */
2336 static ssize_t
2337 cfq_var_show(unsigned int var, char *page)
2339 return sprintf(page, "%d\n", var);
2342 static ssize_t
2343 cfq_var_store(unsigned int *var, const char *page, size_t count)
2345 char *p = (char *) page;
2347 *var = simple_strtoul(p, &p, 10);
2348 return count;
2351 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
2352 static ssize_t __FUNC(elevator_t *e, char *page) \
2353 { \
2354 struct cfq_data *cfqd = e->elevator_data; \
2355 unsigned int __data = __VAR; \
2356 if (__CONV) \
2357 __data = jiffies_to_msecs(__data); \
2358 return cfq_var_show(__data, (page)); \
2360 SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
2361 SHOW_FUNCTION(cfq_queued_show, cfqd->cfq_queued, 0);
2362 SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
2363 SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
2364 SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
2365 SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
2366 SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
2367 SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
2368 SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
2369 SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
2370 #undef SHOW_FUNCTION
2372 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
2373 static ssize_t __FUNC(elevator_t *e, const char *page, size_t count) \
2374 { \
2375 struct cfq_data *cfqd = e->elevator_data; \
2376 unsigned int __data; \
2377 int ret = cfq_var_store(&__data, (page), count); \
2378 if (__data < (MIN)) \
2379 __data = (MIN); \
2380 else if (__data > (MAX)) \
2381 __data = (MAX); \
2382 if (__CONV) \
2383 *(__PTR) = msecs_to_jiffies(__data); \
2384 else \
2385 *(__PTR) = __data; \
2386 return ret; \
2388 STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
2389 STORE_FUNCTION(cfq_queued_store, &cfqd->cfq_queued, 1, UINT_MAX, 0);
2390 STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1);
2391 STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1);
2392 STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
2393 STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0);
2394 STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
2395 STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
2396 STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
2397 STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0);
2398 #undef STORE_FUNCTION
2400 #define CFQ_ATTR(name) \
2401 __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
2403 static struct elv_fs_entry cfq_attrs[] = {
2404 CFQ_ATTR(quantum),
2405 CFQ_ATTR(queued),
2406 CFQ_ATTR(fifo_expire_sync),
2407 CFQ_ATTR(fifo_expire_async),
2408 CFQ_ATTR(back_seek_max),
2409 CFQ_ATTR(back_seek_penalty),
2410 CFQ_ATTR(slice_sync),
2411 CFQ_ATTR(slice_async),
2412 CFQ_ATTR(slice_async_rq),
2413 CFQ_ATTR(slice_idle),
2414 __ATTR_NULL
2415 };
2417 static struct elevator_type iosched_cfq = {
2418 .ops = {
2419 .elevator_merge_fn = cfq_merge,
2420 .elevator_merged_fn = cfq_merged_request,
2421 .elevator_merge_req_fn = cfq_merged_requests,
2422 .elevator_dispatch_fn = cfq_dispatch_requests,
2423 .elevator_add_req_fn = cfq_insert_request,
2424 .elevator_activate_req_fn = cfq_activate_request,
2425 .elevator_deactivate_req_fn = cfq_deactivate_request,
2426 .elevator_queue_empty_fn = cfq_queue_empty,
2427 .elevator_completed_req_fn = cfq_completed_request,
2428 .elevator_former_req_fn = cfq_former_request,
2429 .elevator_latter_req_fn = cfq_latter_request,
2430 .elevator_set_req_fn = cfq_set_request,
2431 .elevator_put_req_fn = cfq_put_request,
2432 .elevator_may_queue_fn = cfq_may_queue,
2433 .elevator_init_fn = cfq_init_queue,
2434 .elevator_exit_fn = cfq_exit_queue,
2435 .trim = cfq_trim,
2436 },
2437 .elevator_attrs = cfq_attrs,
2438 .elevator_name = "cfq",
2439 .elevator_owner = THIS_MODULE,
2440 };
2442 static int __init cfq_init(void)
2444 int ret;
2446 /*
2447 * could be 0 on HZ < 1000 setups
2448 */
2449 if (!cfq_slice_async)
2450 cfq_slice_async = 1;
2451 if (!cfq_slice_idle)
2452 cfq_slice_idle = 1;
2454 if (cfq_slab_setup())
2455 return -ENOMEM;
2457 ret = elv_register(&iosched_cfq);
2458 if (ret)
2459 cfq_slab_kill();
2461 return ret;
2464 static void __exit cfq_exit(void)
2466 DECLARE_COMPLETION(all_gone);
2467 elv_unregister(&iosched_cfq);
2468 ioc_gone = &all_gone;
2469 /* ioc_gone's update must be visible before reading ioc_count */
2470 smp_wmb();
2471 if (atomic_read(&ioc_count))
2472 wait_for_completion(ioc_gone);
2473 synchronize_rcu();
2474 cfq_slab_kill();
2477 module_init(cfq_init);
2478 module_exit(cfq_exit);
2480 MODULE_AUTHOR("Jens Axboe");
2481 MODULE_LICENSE("GPL");
2482 MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");