2 * CFQ, or complete fairness queueing, disk scheduler.
4 * Based on ideas from a previously unfinished io
5 * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli.
7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
9 #include <linux/module.h>
10 #include <linux/blkdev.h>
11 #include <linux/elevator.h>
12 #include <linux/rbtree.h>
13 #include <linux/ioprio.h>
14 #include <linux/blktrace_api.h>
19 /* max queue in one round of service */
20 static const int cfq_quantum = 4;
21 static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
22 /* maximum backwards seek, in KiB */
23 static const int cfq_back_max = 16 * 1024;
24 /* penalty of a backwards seek */
25 static const int cfq_back_penalty = 2;
26 static const int cfq_slice_sync = HZ / 10;
27 static int cfq_slice_async = HZ / 25;
28 static const int cfq_slice_async_rq = 2;
29 static int cfq_slice_idle = HZ / 125;
32 * offset from end of service tree
34 #define CFQ_IDLE_DELAY (HZ / 5)
37 * below this threshold, we consider thinktime immediate
39 #define CFQ_MIN_TT (2)
41 #define CFQ_SLICE_SCALE (5)
42 #define CFQ_HW_QUEUE_MIN (5)
45 ((struct cfq_io_context *) (rq)->elevator_private)
46 #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private2)
48 static struct kmem_cache *cfq_pool;
49 static struct kmem_cache *cfq_ioc_pool;
51 static DEFINE_PER_CPU(unsigned long, ioc_count);
52 static struct completion *ioc_gone;
53 static DEFINE_SPINLOCK(ioc_gone_lock);
55 #define CFQ_PRIO_LISTS IOPRIO_BE_NR
56 #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
57 #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
62 #define sample_valid(samples) ((samples) > 80)
65 * Most of our rbtree usage is for sorting with min extraction, so
66 * if we cache the leftmost node we don't have to walk down the tree
67 * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should
68 * move this into the elevator for the rq sorting as well.
74 #define CFQ_RB_ROOT (struct cfq_rb_root) { RB_ROOT, NULL, }
77 * Per block device queue structure
80 struct request_queue *queue;
83 * rr list of queues with requests and the count of them
85 struct cfq_rb_root service_tree;
86 unsigned int busy_queues;
88 * Used to track any pending rt requests so we can pre-empt current
89 * non-RT cfqq in service when this value is non-zero.
91 unsigned int busy_rt_queues;
97 * queue-depth detection
102 int rq_in_driver_peak;
105 * idle window management
107 struct timer_list idle_slice_timer;
108 struct work_struct unplug_work;
110 struct cfq_queue *active_queue;
111 struct cfq_io_context *active_cic;
114 * async queue for each priority case
116 struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
117 struct cfq_queue *async_idle_cfqq;
119 sector_t last_position;
120 unsigned long last_end_request;
123 * tunables, see top of file
125 unsigned int cfq_quantum;
126 unsigned int cfq_fifo_expire[2];
127 unsigned int cfq_back_penalty;
128 unsigned int cfq_back_max;
129 unsigned int cfq_slice[2];
130 unsigned int cfq_slice_async_rq;
131 unsigned int cfq_slice_idle;
133 struct list_head cic_list;
137 * Per process-grouping structure
140 /* reference count */
142 /* various state flags, see below */
144 /* parent cfq_data */
145 struct cfq_data *cfqd;
146 /* service_tree member */
147 struct rb_node rb_node;
148 /* service_tree key */
149 unsigned long rb_key;
150 /* sorted list of pending requests */
151 struct rb_root sort_list;
152 /* if fifo isn't expired, next request to serve */
153 struct request *next_rq;
154 /* requests queued in sort_list */
156 /* currently allocated requests */
158 /* fifo list of requests in sort_list */
159 struct list_head fifo;
161 unsigned long slice_end;
164 /* pending metadata requests */
166 /* number of requests that are on the dispatch list or inside driver */
169 /* io prio of this group */
170 unsigned short ioprio, org_ioprio;
171 unsigned short ioprio_class, org_ioprio_class;
176 enum cfqq_state_flags {
177 CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */
178 CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */
179 CFQ_CFQQ_FLAG_must_alloc, /* must be allowed rq alloc */
180 CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */
181 CFQ_CFQQ_FLAG_must_dispatch, /* must dispatch, even if expired */
182 CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */
183 CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */
184 CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */
185 CFQ_CFQQ_FLAG_queue_new, /* queue never been serviced */
186 CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */
187 CFQ_CFQQ_FLAG_sync, /* synchronous queue */
190 #define CFQ_CFQQ_FNS(name) \
191 static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \
193 (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \
195 static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \
197 (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \
199 static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \
201 return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \
205 CFQ_CFQQ_FNS(wait_request);
206 CFQ_CFQQ_FNS(must_alloc);
207 CFQ_CFQQ_FNS(must_alloc_slice);
208 CFQ_CFQQ_FNS(must_dispatch);
209 CFQ_CFQQ_FNS(fifo_expire);
210 CFQ_CFQQ_FNS(idle_window);
211 CFQ_CFQQ_FNS(prio_changed);
212 CFQ_CFQQ_FNS(queue_new);
213 CFQ_CFQQ_FNS(slice_new);
217 #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \
218 blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
219 #define cfq_log(cfqd, fmt, args...) \
220 blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)
222 static void cfq_dispatch_insert(struct request_queue *, struct request *);
223 static struct cfq_queue *cfq_get_queue(struct cfq_data *, int,
224 struct io_context *, gfp_t);
225 static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
226 struct io_context *);
228 static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
231 return cic->cfqq[!!is_sync];
234 static inline void cic_set_cfqq(struct cfq_io_context *cic,
235 struct cfq_queue *cfqq, int is_sync)
237 cic->cfqq[!!is_sync] = cfqq;
241 * We regard a request as SYNC, if it's either a read or has the SYNC bit
242 * set (in which case it could also be direct WRITE).
244 static inline int cfq_bio_sync(struct bio *bio)
246 if (bio_data_dir(bio) == READ || bio_sync(bio))
253 * scheduler run of queue, if there are requests pending and no one in the
254 * driver that will restart queueing
256 static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
258 if (cfqd->busy_queues) {
259 cfq_log(cfqd, "schedule dispatch");
260 kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
264 static int cfq_queue_empty(struct request_queue *q)
266 struct cfq_data *cfqd = q->elevator->elevator_data;
268 return !cfqd->busy_queues;
272 * Scale schedule slice based on io priority. Use the sync time slice only
273 * if a queue is marked sync and has sync io queued. A sync queue with async
274 * io only, should not get full sync slice length.
276 static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync,
279 const int base_slice = cfqd->cfq_slice[sync];
281 WARN_ON(prio >= IOPRIO_BE_NR);
283 return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
287 cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
289 return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
293 cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
295 cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
296 cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
300 * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end
301 * isn't valid until the first request from the dispatch is activated
302 * and the slice time set.
304 static inline int cfq_slice_used(struct cfq_queue *cfqq)
306 if (cfq_cfqq_slice_new(cfqq))
308 if (time_before(jiffies, cfqq->slice_end))
315 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
316 * We choose the request that is closest to the head right now. Distance
317 * behind the head is penalized and only allowed to a certain extent.
319 static struct request *
320 cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2)
322 sector_t last, s1, s2, d1 = 0, d2 = 0;
323 unsigned long back_max;
324 #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */
325 #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */
326 unsigned wrap = 0; /* bit mask: requests behind the disk head? */
328 if (rq1 == NULL || rq1 == rq2)
333 if (rq_is_sync(rq1) && !rq_is_sync(rq2))
335 else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
337 if (rq_is_meta(rq1) && !rq_is_meta(rq2))
339 else if (rq_is_meta(rq2) && !rq_is_meta(rq1))
345 last = cfqd->last_position;
348 * by definition, 1KiB is 2 sectors
350 back_max = cfqd->cfq_back_max * 2;
353 * Strict one way elevator _except_ in the case where we allow
354 * short backward seeks which are biased as twice the cost of a
355 * similar forward seek.
359 else if (s1 + back_max >= last)
360 d1 = (last - s1) * cfqd->cfq_back_penalty;
362 wrap |= CFQ_RQ1_WRAP;
366 else if (s2 + back_max >= last)
367 d2 = (last - s2) * cfqd->cfq_back_penalty;
369 wrap |= CFQ_RQ2_WRAP;
371 /* Found required data */
374 * By doing switch() on the bit mask "wrap" we avoid having to
375 * check two variables for all permutations: --> faster!
378 case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
394 case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
397 * Since both rqs are wrapped,
398 * start with the one that's further behind head
399 * (--> only *one* back seek required),
400 * since back seek takes more time than forward.
410 * The below is leftmost cache rbtree addon
412 static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
415 root->left = rb_first(&root->rb);
418 return rb_entry(root->left, struct cfq_queue, rb_node);
423 static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
428 rb_erase(n, &root->rb);
433 * would be nice to take fifo expire time into account as well
435 static struct request *
436 cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
437 struct request *last)
439 struct rb_node *rbnext = rb_next(&last->rb_node);
440 struct rb_node *rbprev = rb_prev(&last->rb_node);
441 struct request *next = NULL, *prev = NULL;
443 BUG_ON(RB_EMPTY_NODE(&last->rb_node));
446 prev = rb_entry_rq(rbprev);
449 next = rb_entry_rq(rbnext);
451 rbnext = rb_first(&cfqq->sort_list);
452 if (rbnext && rbnext != &last->rb_node)
453 next = rb_entry_rq(rbnext);
456 return cfq_choose_req(cfqd, next, prev);
459 static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
460 struct cfq_queue *cfqq)
463 * just an approximation, should be ok.
465 return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) -
466 cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
470 * The cfqd->service_tree holds all pending cfq_queue's that have
471 * requests waiting to be processed. It is sorted in the order that
472 * we will service the queues.
474 static void cfq_service_tree_add(struct cfq_data *cfqd,
475 struct cfq_queue *cfqq, int add_front)
477 struct rb_node **p, *parent;
478 struct cfq_queue *__cfqq;
479 unsigned long rb_key;
482 if (cfq_class_idle(cfqq)) {
483 rb_key = CFQ_IDLE_DELAY;
484 parent = rb_last(&cfqd->service_tree.rb);
485 if (parent && parent != &cfqq->rb_node) {
486 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
487 rb_key += __cfqq->rb_key;
490 } else if (!add_front) {
491 rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
492 rb_key += cfqq->slice_resid;
493 cfqq->slice_resid = 0;
497 if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
499 * same position, nothing more to do
501 if (rb_key == cfqq->rb_key)
504 cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
509 p = &cfqd->service_tree.rb.rb_node;
514 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
517 * sort RT queues first, we always want to give
518 * preference to them. IDLE queues goes to the back.
519 * after that, sort on the next service time.
521 if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq))
523 else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq))
525 else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq))
527 else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq))
529 else if (rb_key < __cfqq->rb_key)
534 if (n == &(*p)->rb_right)
541 cfqd->service_tree.left = &cfqq->rb_node;
543 cfqq->rb_key = rb_key;
544 rb_link_node(&cfqq->rb_node, parent, p);
545 rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb);
549 * Update cfqq's position in the service tree.
551 static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
554 * Resorting requires the cfqq to be on the RR list already.
556 if (cfq_cfqq_on_rr(cfqq))
557 cfq_service_tree_add(cfqd, cfqq, 0);
561 * add to busy list of queues for service, trying to be fair in ordering
562 * the pending list according to last request service
564 static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
566 cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
567 BUG_ON(cfq_cfqq_on_rr(cfqq));
568 cfq_mark_cfqq_on_rr(cfqq);
570 if (cfq_class_rt(cfqq))
571 cfqd->busy_rt_queues++;
573 cfq_resort_rr_list(cfqd, cfqq);
577 * Called when the cfqq no longer has requests pending, remove it from
580 static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
582 cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
583 BUG_ON(!cfq_cfqq_on_rr(cfqq));
584 cfq_clear_cfqq_on_rr(cfqq);
586 if (!RB_EMPTY_NODE(&cfqq->rb_node))
587 cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
589 BUG_ON(!cfqd->busy_queues);
591 if (cfq_class_rt(cfqq))
592 cfqd->busy_rt_queues--;
596 * rb tree support functions
598 static void cfq_del_rq_rb(struct request *rq)
600 struct cfq_queue *cfqq = RQ_CFQQ(rq);
601 struct cfq_data *cfqd = cfqq->cfqd;
602 const int sync = rq_is_sync(rq);
604 BUG_ON(!cfqq->queued[sync]);
605 cfqq->queued[sync]--;
607 elv_rb_del(&cfqq->sort_list, rq);
609 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
610 cfq_del_cfqq_rr(cfqd, cfqq);
613 static void cfq_add_rq_rb(struct request *rq)
615 struct cfq_queue *cfqq = RQ_CFQQ(rq);
616 struct cfq_data *cfqd = cfqq->cfqd;
617 struct request *__alias;
619 cfqq->queued[rq_is_sync(rq)]++;
622 * looks a little odd, but the first insert might return an alias.
623 * if that happens, put the alias on the dispatch list
625 while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
626 cfq_dispatch_insert(cfqd->queue, __alias);
628 if (!cfq_cfqq_on_rr(cfqq))
629 cfq_add_cfqq_rr(cfqd, cfqq);
632 * check if this request is a better next-serve candidate
634 cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq);
635 BUG_ON(!cfqq->next_rq);
638 static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
640 elv_rb_del(&cfqq->sort_list, rq);
641 cfqq->queued[rq_is_sync(rq)]--;
645 static struct request *
646 cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
648 struct task_struct *tsk = current;
649 struct cfq_io_context *cic;
650 struct cfq_queue *cfqq;
652 cic = cfq_cic_lookup(cfqd, tsk->io_context);
656 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
658 sector_t sector = bio->bi_sector + bio_sectors(bio);
660 return elv_rb_find(&cfqq->sort_list, sector);
666 static void cfq_activate_request(struct request_queue *q, struct request *rq)
668 struct cfq_data *cfqd = q->elevator->elevator_data;
670 cfqd->rq_in_driver++;
671 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
674 cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors;
677 static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
679 struct cfq_data *cfqd = q->elevator->elevator_data;
681 WARN_ON(!cfqd->rq_in_driver);
682 cfqd->rq_in_driver--;
683 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
687 static void cfq_remove_request(struct request *rq)
689 struct cfq_queue *cfqq = RQ_CFQQ(rq);
691 if (cfqq->next_rq == rq)
692 cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
694 list_del_init(&rq->queuelist);
697 cfqq->cfqd->rq_queued--;
698 if (rq_is_meta(rq)) {
699 WARN_ON(!cfqq->meta_pending);
700 cfqq->meta_pending--;
704 static int cfq_merge(struct request_queue *q, struct request **req,
707 struct cfq_data *cfqd = q->elevator->elevator_data;
708 struct request *__rq;
710 __rq = cfq_find_rq_fmerge(cfqd, bio);
711 if (__rq && elv_rq_merge_ok(__rq, bio)) {
713 return ELEVATOR_FRONT_MERGE;
716 return ELEVATOR_NO_MERGE;
719 static void cfq_merged_request(struct request_queue *q, struct request *req,
722 if (type == ELEVATOR_FRONT_MERGE) {
723 struct cfq_queue *cfqq = RQ_CFQQ(req);
725 cfq_reposition_rq_rb(cfqq, req);
730 cfq_merged_requests(struct request_queue *q, struct request *rq,
731 struct request *next)
734 * reposition in fifo if next is older than rq
736 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
737 time_before(next->start_time, rq->start_time))
738 list_move(&rq->queuelist, &next->queuelist);
740 cfq_remove_request(next);
743 static int cfq_allow_merge(struct request_queue *q, struct request *rq,
746 struct cfq_data *cfqd = q->elevator->elevator_data;
747 struct cfq_io_context *cic;
748 struct cfq_queue *cfqq;
751 * Disallow merge of a sync bio into an async request.
753 if (cfq_bio_sync(bio) && !rq_is_sync(rq))
757 * Lookup the cfqq that this bio will be queued with. Allow
758 * merge only if rq is queued there.
760 cic = cfq_cic_lookup(cfqd, current->io_context);
764 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
765 if (cfqq == RQ_CFQQ(rq))
771 static void __cfq_set_active_queue(struct cfq_data *cfqd,
772 struct cfq_queue *cfqq)
775 cfq_log_cfqq(cfqd, cfqq, "set_active");
777 cfq_clear_cfqq_must_alloc_slice(cfqq);
778 cfq_clear_cfqq_fifo_expire(cfqq);
779 cfq_mark_cfqq_slice_new(cfqq);
780 cfq_clear_cfqq_queue_new(cfqq);
783 cfqd->active_queue = cfqq;
787 * current cfqq expired its slice (or was too idle), select new one
790 __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
793 cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);
795 if (cfq_cfqq_wait_request(cfqq))
796 del_timer(&cfqd->idle_slice_timer);
798 cfq_clear_cfqq_must_dispatch(cfqq);
799 cfq_clear_cfqq_wait_request(cfqq);
802 * store what was left of this slice, if the queue idled/timed out
804 if (timed_out && !cfq_cfqq_slice_new(cfqq)) {
805 cfqq->slice_resid = cfqq->slice_end - jiffies;
806 cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid);
809 cfq_resort_rr_list(cfqd, cfqq);
811 if (cfqq == cfqd->active_queue)
812 cfqd->active_queue = NULL;
814 if (cfqd->active_cic) {
815 put_io_context(cfqd->active_cic->ioc);
816 cfqd->active_cic = NULL;
820 static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out)
822 struct cfq_queue *cfqq = cfqd->active_queue;
825 __cfq_slice_expired(cfqd, cfqq, timed_out);
829 * Get next queue for service. Unless we have a queue preemption,
830 * we'll simply select the first cfqq in the service tree.
832 static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
834 if (RB_EMPTY_ROOT(&cfqd->service_tree.rb))
837 return cfq_rb_first(&cfqd->service_tree);
841 * Get and set a new active queue for service.
843 static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd)
845 struct cfq_queue *cfqq;
847 cfqq = cfq_get_next_queue(cfqd);
848 __cfq_set_active_queue(cfqd, cfqq);
852 static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
855 if (rq->sector >= cfqd->last_position)
856 return rq->sector - cfqd->last_position;
858 return cfqd->last_position - rq->sector;
861 static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq)
863 struct cfq_io_context *cic = cfqd->active_cic;
865 if (!sample_valid(cic->seek_samples))
868 return cfq_dist_from_last(cfqd, rq) <= cic->seek_mean;
871 static int cfq_close_cooperator(struct cfq_data *cfq_data,
872 struct cfq_queue *cfqq)
875 * We should notice if some of the queues are cooperating, eg
876 * working closely on the same area of the disk. In that case,
877 * we can group them together and don't waste time idling.
882 #define CIC_SEEKY(cic) ((cic)->seek_mean > (8 * 1024))
884 static void cfq_arm_slice_timer(struct cfq_data *cfqd)
886 struct cfq_queue *cfqq = cfqd->active_queue;
887 struct cfq_io_context *cic;
891 * SSD device without seek penalty, disable idling. But only do so
892 * for devices that support queuing, otherwise we still have a problem
893 * with sync vs async workloads.
895 if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)
898 WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
899 WARN_ON(cfq_cfqq_slice_new(cfqq));
902 * idle is disabled, either manually or by past process history
904 if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq))
908 * still requests with the driver, don't idle
910 if (cfqd->rq_in_driver)
914 * task has exited, don't wait
916 cic = cfqd->active_cic;
917 if (!cic || !atomic_read(&cic->ioc->nr_tasks))
921 * See if this prio level has a good candidate
923 if (cfq_close_cooperator(cfqd, cfqq) &&
924 (sample_valid(cic->ttime_samples) && cic->ttime_mean > 2))
927 cfq_mark_cfqq_must_dispatch(cfqq);
928 cfq_mark_cfqq_wait_request(cfqq);
931 * we don't want to idle for seeks, but we do want to allow
932 * fair distribution of slice time for a process doing back-to-back
933 * seeks. so allow a little bit of time for him to submit a new rq
935 sl = cfqd->cfq_slice_idle;
936 if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
937 sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT));
939 mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
940 cfq_log(cfqd, "arm_idle: %lu", sl);
944 * Move request from internal lists to the request queue dispatch list.
946 static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
948 struct cfq_data *cfqd = q->elevator->elevator_data;
949 struct cfq_queue *cfqq = RQ_CFQQ(rq);
951 cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");
953 cfq_remove_request(rq);
955 elv_dispatch_sort(q, rq);
957 if (cfq_cfqq_sync(cfqq))
962 * return expired entry, or NULL to just start from scratch in rbtree
964 static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
966 struct cfq_data *cfqd = cfqq->cfqd;
970 if (cfq_cfqq_fifo_expire(cfqq))
973 cfq_mark_cfqq_fifo_expire(cfqq);
975 if (list_empty(&cfqq->fifo))
978 fifo = cfq_cfqq_sync(cfqq);
979 rq = rq_entry_fifo(cfqq->fifo.next);
981 if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo]))
984 cfq_log_cfqq(cfqd, cfqq, "fifo=%p", rq);
989 cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
991 const int base_rq = cfqd->cfq_slice_async_rq;
993 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
995 return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
999 * Select a queue for service. If we have a current active queue,
1000 * check whether to continue servicing it, or retrieve and set a new one.
1002 static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
1004 struct cfq_queue *cfqq;
1006 cfqq = cfqd->active_queue;
1011 * The active queue has run out of time, expire it and select new.
1013 if (cfq_slice_used(cfqq))
1017 * If we have a RT cfqq waiting, then we pre-empt the current non-rt
1020 if (!cfq_class_rt(cfqq) && cfqd->busy_rt_queues) {
1022 * We simulate this as cfqq timed out so that it gets to bank
1023 * the remaining of its time slice.
1025 cfq_log_cfqq(cfqd, cfqq, "preempt");
1026 cfq_slice_expired(cfqd, 1);
1031 * The active queue has requests and isn't expired, allow it to
1034 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
1038 * No requests pending. If the active queue still has requests in
1039 * flight or is idling for a new request, allow either of these
1040 * conditions to happen (or time out) before selecting a new queue.
1042 if (timer_pending(&cfqd->idle_slice_timer) ||
1043 (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) {
1049 cfq_slice_expired(cfqd, 0);
1051 cfqq = cfq_set_active_queue(cfqd);
1057 * Dispatch some requests from cfqq, moving them to the request queue
1061 __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1066 BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
1072 * follow expired path, else get first next available
1074 rq = cfq_check_fifo(cfqq);
1079 * finally, insert request into driver dispatch list
1081 cfq_dispatch_insert(cfqd->queue, rq);
1085 if (!cfqd->active_cic) {
1086 atomic_inc(&RQ_CIC(rq)->ioc->refcount);
1087 cfqd->active_cic = RQ_CIC(rq);
1090 if (RB_EMPTY_ROOT(&cfqq->sort_list))
1094 * If there is a non-empty RT cfqq waiting for current
1095 * cfqq's timeslice to complete, pre-empt this cfqq
1097 if (!cfq_class_rt(cfqq) && cfqd->busy_rt_queues)
1100 } while (dispatched < max_dispatch);
1103 * expire an async queue immediately if it has used up its slice. idle
1104 * queue always expire after 1 dispatch round.
1106 if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) &&
1107 dispatched >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
1108 cfq_class_idle(cfqq))) {
1109 cfqq->slice_end = jiffies + 1;
1110 cfq_slice_expired(cfqd, 0);
1116 static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
1120 while (cfqq->next_rq) {
1121 cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq);
1125 BUG_ON(!list_empty(&cfqq->fifo));
1130 * Drain our current requests. Used for barriers and when switching
1131 * io schedulers on-the-fly.
1133 static int cfq_forced_dispatch(struct cfq_data *cfqd)
1135 struct cfq_queue *cfqq;
1138 while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL)
1139 dispatched += __cfq_forced_dispatch_cfqq(cfqq);
1141 cfq_slice_expired(cfqd, 0);
1143 BUG_ON(cfqd->busy_queues);
1145 cfq_log(cfqd, "forced_dispatch=%d\n", dispatched);
1149 static int cfq_dispatch_requests(struct request_queue *q, int force)
1151 struct cfq_data *cfqd = q->elevator->elevator_data;
1152 struct cfq_queue *cfqq;
1155 if (!cfqd->busy_queues)
1158 if (unlikely(force))
1159 return cfq_forced_dispatch(cfqd);
1162 while ((cfqq = cfq_select_queue(cfqd)) != NULL) {
1165 max_dispatch = cfqd->cfq_quantum;
1166 if (cfq_class_idle(cfqq))
1169 if (cfqq->dispatched >= max_dispatch && cfqd->busy_queues > 1)
1172 if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq))
1175 cfq_clear_cfqq_must_dispatch(cfqq);
1176 cfq_clear_cfqq_wait_request(cfqq);
1177 del_timer(&cfqd->idle_slice_timer);
1179 dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch);
1182 cfq_log(cfqd, "dispatched=%d", dispatched);
1187 * task holds one reference to the queue, dropped when task exits. each rq
1188 * in-flight on this queue also holds a reference, dropped when rq is freed.
1190 * queue lock must be held here.
1192 static void cfq_put_queue(struct cfq_queue *cfqq)
1194 struct cfq_data *cfqd = cfqq->cfqd;
1196 BUG_ON(atomic_read(&cfqq->ref) <= 0);
1198 if (!atomic_dec_and_test(&cfqq->ref))
1201 cfq_log_cfqq(cfqd, cfqq, "put_queue");
1202 BUG_ON(rb_first(&cfqq->sort_list));
1203 BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
1204 BUG_ON(cfq_cfqq_on_rr(cfqq));
1206 if (unlikely(cfqd->active_queue == cfqq)) {
1207 __cfq_slice_expired(cfqd, cfqq, 0);
1208 cfq_schedule_dispatch(cfqd);
1211 kmem_cache_free(cfq_pool, cfqq);
1215 * Must always be called with the rcu_read_lock() held
1218 __call_for_each_cic(struct io_context *ioc,
1219 void (*func)(struct io_context *, struct cfq_io_context *))
1221 struct cfq_io_context *cic;
1222 struct hlist_node *n;
1224 hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list)
1229 * Call func for each cic attached to this ioc.
1232 call_for_each_cic(struct io_context *ioc,
1233 void (*func)(struct io_context *, struct cfq_io_context *))
1236 __call_for_each_cic(ioc, func);
1240 static void cfq_cic_free_rcu(struct rcu_head *head)
1242 struct cfq_io_context *cic;
1244 cic = container_of(head, struct cfq_io_context, rcu_head);
1246 kmem_cache_free(cfq_ioc_pool, cic);
1247 elv_ioc_count_dec(ioc_count);
1251 * CFQ scheduler is exiting, grab exit lock and check
1252 * the pending io context count. If it hits zero,
1253 * complete ioc_gone and set it back to NULL
1255 spin_lock(&ioc_gone_lock);
1256 if (ioc_gone && !elv_ioc_count_read(ioc_count)) {
1260 spin_unlock(&ioc_gone_lock);
1264 static void cfq_cic_free(struct cfq_io_context *cic)
1266 call_rcu(&cic->rcu_head, cfq_cic_free_rcu);
1269 static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic)
1271 unsigned long flags;
1273 BUG_ON(!cic->dead_key);
1275 spin_lock_irqsave(&ioc->lock, flags);
1276 radix_tree_delete(&ioc->radix_root, cic->dead_key);
1277 hlist_del_rcu(&cic->cic_list);
1278 spin_unlock_irqrestore(&ioc->lock, flags);
1284 * Must be called with rcu_read_lock() held or preemption otherwise disabled.
1285 * Only two callers of this - ->dtor() which is called with the rcu_read_lock(),
1286 * and ->trim() which is called with the task lock held
1288 static void cfq_free_io_context(struct io_context *ioc)
1291 * ioc->refcount is zero here, or we are called from elv_unregister(),
1292 * so no more cic's are allowed to be linked into this ioc. So it
1293 * should be ok to iterate over the known list, we will see all cic's
1294 * since no new ones are added.
1296 __call_for_each_cic(ioc, cic_free_func);
1299 static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1301 if (unlikely(cfqq == cfqd->active_queue)) {
1302 __cfq_slice_expired(cfqd, cfqq, 0);
1303 cfq_schedule_dispatch(cfqd);
1306 cfq_put_queue(cfqq);
1309 static void __cfq_exit_single_io_context(struct cfq_data *cfqd,
1310 struct cfq_io_context *cic)
1312 struct io_context *ioc = cic->ioc;
1314 list_del_init(&cic->queue_list);
1317 * Make sure key == NULL is seen for dead queues
1320 cic->dead_key = (unsigned long) cic->key;
1323 if (ioc->ioc_data == cic)
1324 rcu_assign_pointer(ioc->ioc_data, NULL);
1326 if (cic->cfqq[ASYNC]) {
1327 cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]);
1328 cic->cfqq[ASYNC] = NULL;
1331 if (cic->cfqq[SYNC]) {
1332 cfq_exit_cfqq(cfqd, cic->cfqq[SYNC]);
1333 cic->cfqq[SYNC] = NULL;
1337 static void cfq_exit_single_io_context(struct io_context *ioc,
1338 struct cfq_io_context *cic)
1340 struct cfq_data *cfqd = cic->key;
1343 struct request_queue *q = cfqd->queue;
1344 unsigned long flags;
1346 spin_lock_irqsave(q->queue_lock, flags);
1349 * Ensure we get a fresh copy of the ->key to prevent
1350 * race between exiting task and queue
1352 smp_read_barrier_depends();
1354 __cfq_exit_single_io_context(cfqd, cic);
1356 spin_unlock_irqrestore(q->queue_lock, flags);
1361 * The process that ioc belongs to has exited, we need to clean up
1362 * and put the internal structures we have that belongs to that process.
1364 static void cfq_exit_io_context(struct io_context *ioc)
1366 call_for_each_cic(ioc, cfq_exit_single_io_context);
1369 static struct cfq_io_context *
1370 cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1372 struct cfq_io_context *cic;
1374 cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO,
1377 cic->last_end_request = jiffies;
1378 INIT_LIST_HEAD(&cic->queue_list);
1379 INIT_HLIST_NODE(&cic->cic_list);
1380 cic->dtor = cfq_free_io_context;
1381 cic->exit = cfq_exit_io_context;
1382 elv_ioc_count_inc(ioc_count);
1388 static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc)
1390 struct task_struct *tsk = current;
1393 if (!cfq_cfqq_prio_changed(cfqq))
1396 ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio);
1397 switch (ioprio_class) {
1399 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
1400 case IOPRIO_CLASS_NONE:
1402 * no prio set, inherit CPU scheduling settings
1404 cfqq->ioprio = task_nice_ioprio(tsk);
1405 cfqq->ioprio_class = task_nice_ioclass(tsk);
1407 case IOPRIO_CLASS_RT:
1408 cfqq->ioprio = task_ioprio(ioc);
1409 cfqq->ioprio_class = IOPRIO_CLASS_RT;
1411 case IOPRIO_CLASS_BE:
1412 cfqq->ioprio = task_ioprio(ioc);
1413 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1415 case IOPRIO_CLASS_IDLE:
1416 cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
1418 cfq_clear_cfqq_idle_window(cfqq);
1423 * keep track of original prio settings in case we have to temporarily
1424 * elevate the priority of this queue
1426 cfqq->org_ioprio = cfqq->ioprio;
1427 cfqq->org_ioprio_class = cfqq->ioprio_class;
1428 cfq_clear_cfqq_prio_changed(cfqq);
1431 static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic)
1433 struct cfq_data *cfqd = cic->key;
1434 struct cfq_queue *cfqq;
1435 unsigned long flags;
1437 if (unlikely(!cfqd))
1440 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
1442 cfqq = cic->cfqq[ASYNC];
1444 struct cfq_queue *new_cfqq;
1445 new_cfqq = cfq_get_queue(cfqd, ASYNC, cic->ioc, GFP_ATOMIC);
1447 cic->cfqq[ASYNC] = new_cfqq;
1448 cfq_put_queue(cfqq);
1452 cfqq = cic->cfqq[SYNC];
1454 cfq_mark_cfqq_prio_changed(cfqq);
1456 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
1459 static void cfq_ioc_set_ioprio(struct io_context *ioc)
1461 call_for_each_cic(ioc, changed_ioprio);
1462 ioc->ioprio_changed = 0;
1465 static struct cfq_queue *
1466 cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync,
1467 struct io_context *ioc, gfp_t gfp_mask)
1469 struct cfq_queue *cfqq, *new_cfqq = NULL;
1470 struct cfq_io_context *cic;
1473 cic = cfq_cic_lookup(cfqd, ioc);
1474 /* cic always exists here */
1475 cfqq = cic_to_cfqq(cic, is_sync);
1481 } else if (gfp_mask & __GFP_WAIT) {
1483 * Inform the allocator of the fact that we will
1484 * just repeat this allocation if it fails, to allow
1485 * the allocator to do whatever it needs to attempt to
1488 spin_unlock_irq(cfqd->queue->queue_lock);
1489 new_cfqq = kmem_cache_alloc_node(cfq_pool,
1490 gfp_mask | __GFP_NOFAIL | __GFP_ZERO,
1492 spin_lock_irq(cfqd->queue->queue_lock);
1495 cfqq = kmem_cache_alloc_node(cfq_pool,
1496 gfp_mask | __GFP_ZERO,
1502 RB_CLEAR_NODE(&cfqq->rb_node);
1503 INIT_LIST_HEAD(&cfqq->fifo);
1505 atomic_set(&cfqq->ref, 0);
1508 cfq_mark_cfqq_prio_changed(cfqq);
1509 cfq_mark_cfqq_queue_new(cfqq);
1511 cfq_init_prio_data(cfqq, ioc);
1514 if (!cfq_class_idle(cfqq))
1515 cfq_mark_cfqq_idle_window(cfqq);
1516 cfq_mark_cfqq_sync(cfqq);
1518 cfqq->pid = current->pid;
1519 cfq_log_cfqq(cfqd, cfqq, "alloced");
1523 kmem_cache_free(cfq_pool, new_cfqq);
1526 WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq);
1530 static struct cfq_queue **
1531 cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio)
1533 switch (ioprio_class) {
1534 case IOPRIO_CLASS_RT:
1535 return &cfqd->async_cfqq[0][ioprio];
1536 case IOPRIO_CLASS_BE:
1537 return &cfqd->async_cfqq[1][ioprio];
1538 case IOPRIO_CLASS_IDLE:
1539 return &cfqd->async_idle_cfqq;
1545 static struct cfq_queue *
1546 cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc,
1549 const int ioprio = task_ioprio(ioc);
1550 const int ioprio_class = task_ioprio_class(ioc);
1551 struct cfq_queue **async_cfqq = NULL;
1552 struct cfq_queue *cfqq = NULL;
1555 async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio);
1560 cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask);
1566 * pin the queue now that it's allocated, scheduler exit will prune it
1568 if (!is_sync && !(*async_cfqq)) {
1569 atomic_inc(&cfqq->ref);
1573 atomic_inc(&cfqq->ref);
1578 * We drop cfq io contexts lazily, so we may find a dead one.
1581 cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc,
1582 struct cfq_io_context *cic)
1584 unsigned long flags;
1586 WARN_ON(!list_empty(&cic->queue_list));
1588 spin_lock_irqsave(&ioc->lock, flags);
1590 BUG_ON(ioc->ioc_data == cic);
1592 radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd);
1593 hlist_del_rcu(&cic->cic_list);
1594 spin_unlock_irqrestore(&ioc->lock, flags);
1599 static struct cfq_io_context *
1600 cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc)
1602 struct cfq_io_context *cic;
1603 unsigned long flags;
1612 * we maintain a last-hit cache, to avoid browsing over the tree
1614 cic = rcu_dereference(ioc->ioc_data);
1615 if (cic && cic->key == cfqd) {
1621 cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd);
1625 /* ->key must be copied to avoid race with cfq_exit_queue() */
1628 cfq_drop_dead_cic(cfqd, ioc, cic);
1633 spin_lock_irqsave(&ioc->lock, flags);
1634 rcu_assign_pointer(ioc->ioc_data, cic);
1635 spin_unlock_irqrestore(&ioc->lock, flags);
1643 * Add cic into ioc, using cfqd as the search key. This enables us to lookup
1644 * the process specific cfq io context when entered from the block layer.
1645 * Also adds the cic to a per-cfqd list, used when this queue is removed.
1647 static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
1648 struct cfq_io_context *cic, gfp_t gfp_mask)
1650 unsigned long flags;
1653 ret = radix_tree_preload(gfp_mask);
1658 spin_lock_irqsave(&ioc->lock, flags);
1659 ret = radix_tree_insert(&ioc->radix_root,
1660 (unsigned long) cfqd, cic);
1662 hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list);
1663 spin_unlock_irqrestore(&ioc->lock, flags);
1665 radix_tree_preload_end();
1668 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
1669 list_add(&cic->queue_list, &cfqd->cic_list);
1670 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
1675 printk(KERN_ERR "cfq: cic link failed!\n");
1681 * Setup general io context and cfq io context. There can be several cfq
1682 * io contexts per general io context, if this process is doing io to more
1683 * than one device managed by cfq.
1685 static struct cfq_io_context *
1686 cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1688 struct io_context *ioc = NULL;
1689 struct cfq_io_context *cic;
1691 might_sleep_if(gfp_mask & __GFP_WAIT);
1693 ioc = get_io_context(gfp_mask, cfqd->queue->node);
1697 cic = cfq_cic_lookup(cfqd, ioc);
1701 cic = cfq_alloc_io_context(cfqd, gfp_mask);
1705 if (cfq_cic_link(cfqd, ioc, cic, gfp_mask))
1709 smp_read_barrier_depends();
1710 if (unlikely(ioc->ioprio_changed))
1711 cfq_ioc_set_ioprio(ioc);
1717 put_io_context(ioc);
1722 cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
1724 unsigned long elapsed = jiffies - cic->last_end_request;
1725 unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle);
1727 cic->ttime_samples = (7*cic->ttime_samples + 256) / 8;
1728 cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8;
1729 cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples;
1733 cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic,
1739 if (cic->last_request_pos < rq->sector)
1740 sdist = rq->sector - cic->last_request_pos;
1742 sdist = cic->last_request_pos - rq->sector;
1745 * Don't allow the seek distance to get too large from the
1746 * odd fragment, pagein, etc
1748 if (cic->seek_samples <= 60) /* second&third seek */
1749 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024);
1751 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64);
1753 cic->seek_samples = (7*cic->seek_samples + 256) / 8;
1754 cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8;
1755 total = cic->seek_total + (cic->seek_samples/2);
1756 do_div(total, cic->seek_samples);
1757 cic->seek_mean = (sector_t)total;
1761 * Disable idle window if the process thinks too long or seeks so much that
1765 cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1766 struct cfq_io_context *cic)
1768 int old_idle, enable_idle;
1771 * Don't idle for async or idle io prio class
1773 if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
1776 enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
1778 if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle ||
1779 (cfqd->hw_tag && CIC_SEEKY(cic)))
1781 else if (sample_valid(cic->ttime_samples)) {
1782 if (cic->ttime_mean > cfqd->cfq_slice_idle)
1788 if (old_idle != enable_idle) {
1789 cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle);
1791 cfq_mark_cfqq_idle_window(cfqq);
1793 cfq_clear_cfqq_idle_window(cfqq);
1798 * Check if new_cfqq should preempt the currently active queue. Return 0 for
1799 * no or if we aren't sure, a 1 will cause a preempt.
1802 cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
1805 struct cfq_queue *cfqq;
1807 cfqq = cfqd->active_queue;
1811 if (cfq_slice_used(cfqq))
1814 if (cfq_class_idle(new_cfqq))
1817 if (cfq_class_idle(cfqq))
1821 * if the new request is sync, but the currently running queue is
1822 * not, let the sync request have priority.
1824 if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq))
1828 * So both queues are sync. Let the new request get disk time if
1829 * it's a metadata request and the current queue is doing regular IO.
1831 if (rq_is_meta(rq) && !cfqq->meta_pending)
1835 * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice.
1837 if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq))
1840 if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq))
1844 * if this request is as-good as one we would expect from the
1845 * current cfqq, let it preempt
1847 if (cfq_rq_close(cfqd, rq))
1854 * cfqq preempts the active queue. if we allowed preempt with no slice left,
1855 * let it have half of its nominal slice.
1857 static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1859 cfq_log_cfqq(cfqd, cfqq, "preempt");
1860 cfq_slice_expired(cfqd, 1);
1863 * Put the new queue at the front of the of the current list,
1864 * so we know that it will be selected next.
1866 BUG_ON(!cfq_cfqq_on_rr(cfqq));
1868 cfq_service_tree_add(cfqd, cfqq, 1);
1870 cfqq->slice_end = 0;
1871 cfq_mark_cfqq_slice_new(cfqq);
1875 * Called when a new fs request (rq) is added (to cfqq). Check if there's
1876 * something we should do about it
1879 cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1882 struct cfq_io_context *cic = RQ_CIC(rq);
1886 cfqq->meta_pending++;
1888 cfq_update_io_thinktime(cfqd, cic);
1889 cfq_update_io_seektime(cfqd, cic, rq);
1890 cfq_update_idle_window(cfqd, cfqq, cic);
1892 cic->last_request_pos = rq->sector + rq->nr_sectors;
1894 if (cfqq == cfqd->active_queue) {
1896 * if we are waiting for a request for this queue, let it rip
1897 * immediately and flag that we must not expire this queue
1900 if (cfq_cfqq_wait_request(cfqq)) {
1901 cfq_mark_cfqq_must_dispatch(cfqq);
1902 del_timer(&cfqd->idle_slice_timer);
1903 blk_start_queueing(cfqd->queue);
1905 } else if (cfq_should_preempt(cfqd, cfqq, rq)) {
1907 * not the active queue - expire current slice if it is
1908 * idle and has expired it's mean thinktime or this new queue
1909 * has some old slice time left and is of higher priority or
1910 * this new queue is RT and the current one is BE
1912 cfq_preempt_queue(cfqd, cfqq);
1913 cfq_mark_cfqq_must_dispatch(cfqq);
1914 blk_start_queueing(cfqd->queue);
1918 static void cfq_insert_request(struct request_queue *q, struct request *rq)
1920 struct cfq_data *cfqd = q->elevator->elevator_data;
1921 struct cfq_queue *cfqq = RQ_CFQQ(rq);
1923 cfq_log_cfqq(cfqd, cfqq, "insert_request");
1924 cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc);
1928 list_add_tail(&rq->queuelist, &cfqq->fifo);
1930 cfq_rq_enqueued(cfqd, cfqq, rq);
1934 * Update hw_tag based on peak queue depth over 50 samples under
1937 static void cfq_update_hw_tag(struct cfq_data *cfqd)
1939 if (cfqd->rq_in_driver > cfqd->rq_in_driver_peak)
1940 cfqd->rq_in_driver_peak = cfqd->rq_in_driver;
1942 if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN &&
1943 cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN)
1946 if (cfqd->hw_tag_samples++ < 50)
1949 if (cfqd->rq_in_driver_peak >= CFQ_HW_QUEUE_MIN)
1954 cfqd->hw_tag_samples = 0;
1955 cfqd->rq_in_driver_peak = 0;
1958 static void cfq_completed_request(struct request_queue *q, struct request *rq)
1960 struct cfq_queue *cfqq = RQ_CFQQ(rq);
1961 struct cfq_data *cfqd = cfqq->cfqd;
1962 const int sync = rq_is_sync(rq);
1966 cfq_log_cfqq(cfqd, cfqq, "complete");
1968 cfq_update_hw_tag(cfqd);
1970 WARN_ON(!cfqd->rq_in_driver);
1971 WARN_ON(!cfqq->dispatched);
1972 cfqd->rq_in_driver--;
1975 if (cfq_cfqq_sync(cfqq))
1976 cfqd->sync_flight--;
1978 if (!cfq_class_idle(cfqq))
1979 cfqd->last_end_request = now;
1982 RQ_CIC(rq)->last_end_request = now;
1985 * If this is the active queue, check if it needs to be expired,
1986 * or if we want to idle in case it has no pending requests.
1988 if (cfqd->active_queue == cfqq) {
1989 if (cfq_cfqq_slice_new(cfqq)) {
1990 cfq_set_prio_slice(cfqd, cfqq);
1991 cfq_clear_cfqq_slice_new(cfqq);
1993 if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
1994 cfq_slice_expired(cfqd, 1);
1995 else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list))
1996 cfq_arm_slice_timer(cfqd);
1999 if (!cfqd->rq_in_driver)
2000 cfq_schedule_dispatch(cfqd);
2004 * we temporarily boost lower priority queues if they are holding fs exclusive
2005 * resources. they are boosted to normal prio (CLASS_BE/4)
2007 static void cfq_prio_boost(struct cfq_queue *cfqq)
2009 if (has_fs_excl()) {
2011 * boost idle prio on transactions that would lock out other
2012 * users of the filesystem
2014 if (cfq_class_idle(cfqq))
2015 cfqq->ioprio_class = IOPRIO_CLASS_BE;
2016 if (cfqq->ioprio > IOPRIO_NORM)
2017 cfqq->ioprio = IOPRIO_NORM;
2020 * check if we need to unboost the queue
2022 if (cfqq->ioprio_class != cfqq->org_ioprio_class)
2023 cfqq->ioprio_class = cfqq->org_ioprio_class;
2024 if (cfqq->ioprio != cfqq->org_ioprio)
2025 cfqq->ioprio = cfqq->org_ioprio;
2029 static inline int __cfq_may_queue(struct cfq_queue *cfqq)
2031 if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) &&
2032 !cfq_cfqq_must_alloc_slice(cfqq)) {
2033 cfq_mark_cfqq_must_alloc_slice(cfqq);
2034 return ELV_MQUEUE_MUST;
2037 return ELV_MQUEUE_MAY;
2040 static int cfq_may_queue(struct request_queue *q, int rw)
2042 struct cfq_data *cfqd = q->elevator->elevator_data;
2043 struct task_struct *tsk = current;
2044 struct cfq_io_context *cic;
2045 struct cfq_queue *cfqq;
2048 * don't force setup of a queue from here, as a call to may_queue
2049 * does not necessarily imply that a request actually will be queued.
2050 * so just lookup a possibly existing queue, or return 'may queue'
2053 cic = cfq_cic_lookup(cfqd, tsk->io_context);
2055 return ELV_MQUEUE_MAY;
2057 cfqq = cic_to_cfqq(cic, rw & REQ_RW_SYNC);
2059 cfq_init_prio_data(cfqq, cic->ioc);
2060 cfq_prio_boost(cfqq);
2062 return __cfq_may_queue(cfqq);
2065 return ELV_MQUEUE_MAY;
2069 * queue lock held here
2071 static void cfq_put_request(struct request *rq)
2073 struct cfq_queue *cfqq = RQ_CFQQ(rq);
2076 const int rw = rq_data_dir(rq);
2078 BUG_ON(!cfqq->allocated[rw]);
2079 cfqq->allocated[rw]--;
2081 put_io_context(RQ_CIC(rq)->ioc);
2083 rq->elevator_private = NULL;
2084 rq->elevator_private2 = NULL;
2086 cfq_put_queue(cfqq);
2091 * Allocate cfq data structures associated with this request.
2094 cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
2096 struct cfq_data *cfqd = q->elevator->elevator_data;
2097 struct cfq_io_context *cic;
2098 const int rw = rq_data_dir(rq);
2099 const int is_sync = rq_is_sync(rq);
2100 struct cfq_queue *cfqq;
2101 unsigned long flags;
2103 might_sleep_if(gfp_mask & __GFP_WAIT);
2105 cic = cfq_get_io_context(cfqd, gfp_mask);
2107 spin_lock_irqsave(q->queue_lock, flags);
2112 cfqq = cic_to_cfqq(cic, is_sync);
2114 cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask);
2119 cic_set_cfqq(cic, cfqq, is_sync);
2122 cfqq->allocated[rw]++;
2123 cfq_clear_cfqq_must_alloc(cfqq);
2124 atomic_inc(&cfqq->ref);
2126 spin_unlock_irqrestore(q->queue_lock, flags);
2128 rq->elevator_private = cic;
2129 rq->elevator_private2 = cfqq;
2134 put_io_context(cic->ioc);
2136 cfq_schedule_dispatch(cfqd);
2137 spin_unlock_irqrestore(q->queue_lock, flags);
2138 cfq_log(cfqd, "set_request fail");
2142 static void cfq_kick_queue(struct work_struct *work)
2144 struct cfq_data *cfqd =
2145 container_of(work, struct cfq_data, unplug_work);
2146 struct request_queue *q = cfqd->queue;
2147 unsigned long flags;
2149 spin_lock_irqsave(q->queue_lock, flags);
2150 blk_start_queueing(q);
2151 spin_unlock_irqrestore(q->queue_lock, flags);
2155 * Timer running if the active_queue is currently idling inside its time slice
2157 static void cfq_idle_slice_timer(unsigned long data)
2159 struct cfq_data *cfqd = (struct cfq_data *) data;
2160 struct cfq_queue *cfqq;
2161 unsigned long flags;
2164 cfq_log(cfqd, "idle timer fired");
2166 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2168 cfqq = cfqd->active_queue;
2175 if (cfq_slice_used(cfqq))
2179 * only expire and reinvoke request handler, if there are
2180 * other queues with pending requests
2182 if (!cfqd->busy_queues)
2186 * not expired and it has a request pending, let it dispatch
2188 if (!RB_EMPTY_ROOT(&cfqq->sort_list)) {
2189 cfq_mark_cfqq_must_dispatch(cfqq);
2194 cfq_slice_expired(cfqd, timed_out);
2196 cfq_schedule_dispatch(cfqd);
2198 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2201 static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
2203 del_timer_sync(&cfqd->idle_slice_timer);
2204 cancel_work_sync(&cfqd->unplug_work);
2207 static void cfq_put_async_queues(struct cfq_data *cfqd)
2211 for (i = 0; i < IOPRIO_BE_NR; i++) {
2212 if (cfqd->async_cfqq[0][i])
2213 cfq_put_queue(cfqd->async_cfqq[0][i]);
2214 if (cfqd->async_cfqq[1][i])
2215 cfq_put_queue(cfqd->async_cfqq[1][i]);
2218 if (cfqd->async_idle_cfqq)
2219 cfq_put_queue(cfqd->async_idle_cfqq);
2222 static void cfq_exit_queue(struct elevator_queue *e)
2224 struct cfq_data *cfqd = e->elevator_data;
2225 struct request_queue *q = cfqd->queue;
2227 cfq_shutdown_timer_wq(cfqd);
2229 spin_lock_irq(q->queue_lock);
2231 if (cfqd->active_queue)
2232 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
2234 while (!list_empty(&cfqd->cic_list)) {
2235 struct cfq_io_context *cic = list_entry(cfqd->cic_list.next,
2236 struct cfq_io_context,
2239 __cfq_exit_single_io_context(cfqd, cic);
2242 cfq_put_async_queues(cfqd);
2244 spin_unlock_irq(q->queue_lock);
2246 cfq_shutdown_timer_wq(cfqd);
2251 static void *cfq_init_queue(struct request_queue *q)
2253 struct cfq_data *cfqd;
2255 cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node);
2259 cfqd->service_tree = CFQ_RB_ROOT;
2260 INIT_LIST_HEAD(&cfqd->cic_list);
2264 init_timer(&cfqd->idle_slice_timer);
2265 cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
2266 cfqd->idle_slice_timer.data = (unsigned long) cfqd;
2268 INIT_WORK(&cfqd->unplug_work, cfq_kick_queue);
2270 cfqd->last_end_request = jiffies;
2271 cfqd->cfq_quantum = cfq_quantum;
2272 cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
2273 cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
2274 cfqd->cfq_back_max = cfq_back_max;
2275 cfqd->cfq_back_penalty = cfq_back_penalty;
2276 cfqd->cfq_slice[0] = cfq_slice_async;
2277 cfqd->cfq_slice[1] = cfq_slice_sync;
2278 cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
2279 cfqd->cfq_slice_idle = cfq_slice_idle;
2285 static void cfq_slab_kill(void)
2288 * Caller already ensured that pending RCU callbacks are completed,
2289 * so we should have no busy allocations at this point.
2292 kmem_cache_destroy(cfq_pool);
2294 kmem_cache_destroy(cfq_ioc_pool);
2297 static int __init cfq_slab_setup(void)
2299 cfq_pool = KMEM_CACHE(cfq_queue, 0);
2303 cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0);
2314 * sysfs parts below -->
2317 cfq_var_show(unsigned int var, char *page)
2319 return sprintf(page, "%d\n", var);
2323 cfq_var_store(unsigned int *var, const char *page, size_t count)
2325 char *p = (char *) page;
2327 *var = simple_strtoul(p, &p, 10);
2331 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
2332 static ssize_t __FUNC(struct elevator_queue *e, char *page) \
2334 struct cfq_data *cfqd = e->elevator_data; \
2335 unsigned int __data = __VAR; \
2337 __data = jiffies_to_msecs(__data); \
2338 return cfq_var_show(__data, (page)); \
2340 SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
2341 SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
2342 SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
2343 SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
2344 SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
2345 SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
2346 SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
2347 SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
2348 SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
2349 #undef SHOW_FUNCTION
2351 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
2352 static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
2354 struct cfq_data *cfqd = e->elevator_data; \
2355 unsigned int __data; \
2356 int ret = cfq_var_store(&__data, (page), count); \
2357 if (__data < (MIN)) \
2359 else if (__data > (MAX)) \
2362 *(__PTR) = msecs_to_jiffies(__data); \
2364 *(__PTR) = __data; \
2367 STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
2368 STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1,
2370 STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1,
2372 STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
2373 STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1,
2375 STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
2376 STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
2377 STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
2378 STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1,
2380 #undef STORE_FUNCTION
2382 #define CFQ_ATTR(name) \
2383 __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
2385 static struct elv_fs_entry cfq_attrs[] = {
2387 CFQ_ATTR(fifo_expire_sync),
2388 CFQ_ATTR(fifo_expire_async),
2389 CFQ_ATTR(back_seek_max),
2390 CFQ_ATTR(back_seek_penalty),
2391 CFQ_ATTR(slice_sync),
2392 CFQ_ATTR(slice_async),
2393 CFQ_ATTR(slice_async_rq),
2394 CFQ_ATTR(slice_idle),
2398 static struct elevator_type iosched_cfq = {
2400 .elevator_merge_fn = cfq_merge,
2401 .elevator_merged_fn = cfq_merged_request,
2402 .elevator_merge_req_fn = cfq_merged_requests,
2403 .elevator_allow_merge_fn = cfq_allow_merge,
2404 .elevator_dispatch_fn = cfq_dispatch_requests,
2405 .elevator_add_req_fn = cfq_insert_request,
2406 .elevator_activate_req_fn = cfq_activate_request,
2407 .elevator_deactivate_req_fn = cfq_deactivate_request,
2408 .elevator_queue_empty_fn = cfq_queue_empty,
2409 .elevator_completed_req_fn = cfq_completed_request,
2410 .elevator_former_req_fn = elv_rb_former_request,
2411 .elevator_latter_req_fn = elv_rb_latter_request,
2412 .elevator_set_req_fn = cfq_set_request,
2413 .elevator_put_req_fn = cfq_put_request,
2414 .elevator_may_queue_fn = cfq_may_queue,
2415 .elevator_init_fn = cfq_init_queue,
2416 .elevator_exit_fn = cfq_exit_queue,
2417 .trim = cfq_free_io_context,
2419 .elevator_attrs = cfq_attrs,
2420 .elevator_name = "cfq",
2421 .elevator_owner = THIS_MODULE,
2424 static int __init cfq_init(void)
2427 * could be 0 on HZ < 1000 setups
2429 if (!cfq_slice_async)
2430 cfq_slice_async = 1;
2431 if (!cfq_slice_idle)
2434 if (cfq_slab_setup())
2437 elv_register(&iosched_cfq);
2442 static void __exit cfq_exit(void)
2444 DECLARE_COMPLETION_ONSTACK(all_gone);
2445 elv_unregister(&iosched_cfq);
2446 ioc_gone = &all_gone;
2447 /* ioc_gone's update must be visible before reading ioc_count */
2451 * this also protects us from entering cfq_slab_kill() with
2452 * pending RCU callbacks
2454 if (elv_ioc_count_read(ioc_count))
2455 wait_for_completion(&all_gone);
2459 module_init(cfq_init);
2460 module_exit(cfq_exit);
2462 MODULE_AUTHOR("Jens Axboe");
2463 MODULE_LICENSE("GPL");
2464 MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");