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/slab.h>
11 #include <linux/blkdev.h>
12 #include <linux/elevator.h>
13 #include <linux/jiffies.h>
14 #include <linux/rbtree.h>
15 #include <linux/ioprio.h>
16 #include <linux/blktrace_api.h>
22 /* max queue in one round of service */
23 static const int cfq_quantum = 8;
24 static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
25 /* maximum backwards seek, in KiB */
26 static const int cfq_back_max = 16 * 1024;
27 /* penalty of a backwards seek */
28 static const int cfq_back_penalty = 2;
29 static const int cfq_slice_sync = HZ / 10;
30 static int cfq_slice_async = HZ / 25;
31 static const int cfq_slice_async_rq = 2;
32 static int cfq_slice_idle = HZ / 125;
33 static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
34 static const int cfq_hist_divisor = 4;
37 * offset from end of service tree
39 #define CFQ_IDLE_DELAY (HZ / 5)
42 * below this threshold, we consider thinktime immediate
44 #define CFQ_MIN_TT (2)
46 #define CFQ_SLICE_SCALE (5)
47 #define CFQ_HW_QUEUE_MIN (5)
48 #define CFQ_SERVICE_SHIFT 12
50 #define CFQQ_SEEK_THR (sector_t)(8 * 100)
51 #define CFQQ_CLOSE_THR (sector_t)(8 * 1024)
52 #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32)
53 #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8)
56 ((struct cfq_io_context *) (rq)->elevator_private)
57 #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private2)
58 #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elevator_private3)
60 static struct kmem_cache *cfq_pool;
61 static struct kmem_cache *cfq_ioc_pool;
63 static DEFINE_PER_CPU(unsigned long, cfq_ioc_count);
64 static struct completion *ioc_gone;
65 static DEFINE_SPINLOCK(ioc_gone_lock);
67 static DEFINE_SPINLOCK(cic_index_lock);
68 static DEFINE_IDA(cic_index_ida);
70 #define CFQ_PRIO_LISTS IOPRIO_BE_NR
71 #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
72 #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
74 #define sample_valid(samples) ((samples) > 80)
75 #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node)
78 * Most of our rbtree usage is for sorting with min extraction, so
79 * if we cache the leftmost node we don't have to walk down the tree
80 * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should
81 * move this into the elevator for the rq sorting as well.
87 unsigned total_weight;
89 struct rb_node *active;
91 #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, .left = NULL, \
92 .count = 0, .min_vdisktime = 0, }
95 * Per process-grouping structure
100 /* various state flags, see below */
102 /* parent cfq_data */
103 struct cfq_data *cfqd;
104 /* service_tree member */
105 struct rb_node rb_node;
106 /* service_tree key */
107 unsigned long rb_key;
108 /* prio tree member */
109 struct rb_node p_node;
110 /* prio tree root we belong to, if any */
111 struct rb_root *p_root;
112 /* sorted list of pending requests */
113 struct rb_root sort_list;
114 /* if fifo isn't expired, next request to serve */
115 struct request *next_rq;
116 /* requests queued in sort_list */
118 /* currently allocated requests */
120 /* fifo list of requests in sort_list */
121 struct list_head fifo;
123 /* time when queue got scheduled in to dispatch first request. */
124 unsigned long dispatch_start;
125 unsigned int allocated_slice;
126 unsigned int slice_dispatch;
127 /* time when first request from queue completed and slice started. */
128 unsigned long slice_start;
129 unsigned long slice_end;
132 /* pending metadata requests */
134 /* number of requests that are on the dispatch list or inside driver */
137 /* io prio of this group */
138 unsigned short ioprio, org_ioprio;
139 unsigned short ioprio_class, org_ioprio_class;
144 sector_t last_request_pos;
146 struct cfq_rb_root *service_tree;
147 struct cfq_queue *new_cfqq;
148 struct cfq_group *cfqg;
149 struct cfq_group *orig_cfqg;
153 * First index in the service_trees.
154 * IDLE is handled separately, so it has negative index
163 * Second index in the service_trees.
167 SYNC_NOIDLE_WORKLOAD = 1,
171 /* This is per cgroup per device grouping structure */
173 /* group service_tree member */
174 struct rb_node rb_node;
176 /* group service_tree key */
181 /* number of cfqq currently on this group */
184 /* Per group busy queus average. Useful for workload slice calc. */
185 unsigned int busy_queues_avg[2];
187 * rr lists of queues with requests, onle rr for each priority class.
188 * Counts are embedded in the cfq_rb_root
190 struct cfq_rb_root service_trees[2][3];
191 struct cfq_rb_root service_tree_idle;
193 unsigned long saved_workload_slice;
194 enum wl_type_t saved_workload;
195 enum wl_prio_t saved_serving_prio;
196 struct blkio_group blkg;
197 #ifdef CONFIG_CFQ_GROUP_IOSCHED
198 struct hlist_node cfqd_node;
204 * Per block device queue structure
207 struct request_queue *queue;
208 /* Root service tree for cfq_groups */
209 struct cfq_rb_root grp_service_tree;
210 struct cfq_group root_group;
213 * The priority currently being served
215 enum wl_prio_t serving_prio;
216 enum wl_type_t serving_type;
217 unsigned long workload_expires;
218 struct cfq_group *serving_group;
219 bool noidle_tree_requires_idle;
222 * Each priority tree is sorted by next_request position. These
223 * trees are used when determining if two or more queues are
224 * interleaving requests (see cfq_close_cooperator).
226 struct rb_root prio_trees[CFQ_PRIO_LISTS];
228 unsigned int busy_queues;
234 * queue-depth detection
240 * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection)
241 * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth)
244 int hw_tag_est_depth;
245 unsigned int hw_tag_samples;
248 * idle window management
250 struct timer_list idle_slice_timer;
251 struct work_struct unplug_work;
253 struct cfq_queue *active_queue;
254 struct cfq_io_context *active_cic;
257 * async queue for each priority case
259 struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
260 struct cfq_queue *async_idle_cfqq;
262 sector_t last_position;
265 * tunables, see top of file
267 unsigned int cfq_quantum;
268 unsigned int cfq_fifo_expire[2];
269 unsigned int cfq_back_penalty;
270 unsigned int cfq_back_max;
271 unsigned int cfq_slice[2];
272 unsigned int cfq_slice_async_rq;
273 unsigned int cfq_slice_idle;
274 unsigned int cfq_latency;
275 unsigned int cfq_group_isolation;
277 unsigned int cic_index;
278 struct list_head cic_list;
281 * Fallback dummy cfqq for extreme OOM conditions
283 struct cfq_queue oom_cfqq;
285 unsigned long last_delayed_sync;
287 /* List of cfq groups being managed on this device*/
288 struct hlist_head cfqg_list;
292 static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);
294 static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg,
301 if (prio == IDLE_WORKLOAD)
302 return &cfqg->service_tree_idle;
304 return &cfqg->service_trees[prio][type];
307 enum cfqq_state_flags {
308 CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */
309 CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */
310 CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */
311 CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */
312 CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */
313 CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */
314 CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */
315 CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */
316 CFQ_CFQQ_FLAG_sync, /* synchronous queue */
317 CFQ_CFQQ_FLAG_coop, /* cfqq is shared */
318 CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */
319 CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */
320 CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */
323 #define CFQ_CFQQ_FNS(name) \
324 static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \
326 (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \
328 static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \
330 (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \
332 static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \
334 return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \
338 CFQ_CFQQ_FNS(wait_request);
339 CFQ_CFQQ_FNS(must_dispatch);
340 CFQ_CFQQ_FNS(must_alloc_slice);
341 CFQ_CFQQ_FNS(fifo_expire);
342 CFQ_CFQQ_FNS(idle_window);
343 CFQ_CFQQ_FNS(prio_changed);
344 CFQ_CFQQ_FNS(slice_new);
347 CFQ_CFQQ_FNS(split_coop);
349 CFQ_CFQQ_FNS(wait_busy);
352 #ifdef CONFIG_CFQ_GROUP_IOSCHED
353 #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \
354 blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \
355 cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
356 blkg_path(&(cfqq)->cfqg->blkg), ##args);
358 #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \
359 blk_add_trace_msg((cfqd)->queue, "%s " fmt, \
360 blkg_path(&(cfqg)->blkg), ##args); \
363 #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \
364 blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
365 #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0);
367 #define cfq_log(cfqd, fmt, args...) \
368 blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)
370 /* Traverses through cfq group service trees */
371 #define for_each_cfqg_st(cfqg, i, j, st) \
372 for (i = 0; i <= IDLE_WORKLOAD; i++) \
373 for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\
374 : &cfqg->service_tree_idle; \
375 (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \
376 (i == IDLE_WORKLOAD && j == 0); \
377 j++, st = i < IDLE_WORKLOAD ? \
378 &cfqg->service_trees[i][j]: NULL) \
381 static inline bool iops_mode(struct cfq_data *cfqd)
384 * If we are not idling on queues and it is a NCQ drive, parallel
385 * execution of requests is on and measuring time is not possible
386 * in most of the cases until and unless we drive shallower queue
387 * depths and that becomes a performance bottleneck. In such cases
388 * switch to start providing fairness in terms of number of IOs.
390 if (!cfqd->cfq_slice_idle && cfqd->hw_tag)
396 static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq)
398 if (cfq_class_idle(cfqq))
399 return IDLE_WORKLOAD;
400 if (cfq_class_rt(cfqq))
406 static enum wl_type_t cfqq_type(struct cfq_queue *cfqq)
408 if (!cfq_cfqq_sync(cfqq))
409 return ASYNC_WORKLOAD;
410 if (!cfq_cfqq_idle_window(cfqq))
411 return SYNC_NOIDLE_WORKLOAD;
412 return SYNC_WORKLOAD;
415 static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl,
416 struct cfq_data *cfqd,
417 struct cfq_group *cfqg)
419 if (wl == IDLE_WORKLOAD)
420 return cfqg->service_tree_idle.count;
422 return cfqg->service_trees[wl][ASYNC_WORKLOAD].count
423 + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
424 + cfqg->service_trees[wl][SYNC_WORKLOAD].count;
427 static inline int cfqg_busy_async_queues(struct cfq_data *cfqd,
428 struct cfq_group *cfqg)
430 return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count
431 + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count;
434 static void cfq_dispatch_insert(struct request_queue *, struct request *);
435 static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
436 struct io_context *, gfp_t);
437 static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
438 struct io_context *);
440 static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
443 return cic->cfqq[is_sync];
446 static inline void cic_set_cfqq(struct cfq_io_context *cic,
447 struct cfq_queue *cfqq, bool is_sync)
449 cic->cfqq[is_sync] = cfqq;
452 #define CIC_DEAD_KEY 1ul
453 #define CIC_DEAD_INDEX_SHIFT 1
455 static inline void *cfqd_dead_key(struct cfq_data *cfqd)
457 return (void *)(cfqd->cic_index << CIC_DEAD_INDEX_SHIFT | CIC_DEAD_KEY);
460 static inline struct cfq_data *cic_to_cfqd(struct cfq_io_context *cic)
462 struct cfq_data *cfqd = cic->key;
464 if (unlikely((unsigned long) cfqd & CIC_DEAD_KEY))
471 * We regard a request as SYNC, if it's either a read or has the SYNC bit
472 * set (in which case it could also be direct WRITE).
474 static inline bool cfq_bio_sync(struct bio *bio)
476 return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC);
480 * scheduler run of queue, if there are requests pending and no one in the
481 * driver that will restart queueing
483 static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
485 if (cfqd->busy_queues) {
486 cfq_log(cfqd, "schedule dispatch");
487 kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
491 static int cfq_queue_empty(struct request_queue *q)
493 struct cfq_data *cfqd = q->elevator->elevator_data;
495 return !cfqd->rq_queued;
499 * Scale schedule slice based on io priority. Use the sync time slice only
500 * if a queue is marked sync and has sync io queued. A sync queue with async
501 * io only, should not get full sync slice length.
503 static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
506 const int base_slice = cfqd->cfq_slice[sync];
508 WARN_ON(prio >= IOPRIO_BE_NR);
510 return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
514 cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
516 return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
519 static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg)
521 u64 d = delta << CFQ_SERVICE_SHIFT;
523 d = d * BLKIO_WEIGHT_DEFAULT;
524 do_div(d, cfqg->weight);
528 static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime)
530 s64 delta = (s64)(vdisktime - min_vdisktime);
532 min_vdisktime = vdisktime;
534 return min_vdisktime;
537 static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime)
539 s64 delta = (s64)(vdisktime - min_vdisktime);
541 min_vdisktime = vdisktime;
543 return min_vdisktime;
546 static void update_min_vdisktime(struct cfq_rb_root *st)
548 u64 vdisktime = st->min_vdisktime;
549 struct cfq_group *cfqg;
552 cfqg = rb_entry_cfqg(st->active);
553 vdisktime = cfqg->vdisktime;
557 cfqg = rb_entry_cfqg(st->left);
558 vdisktime = min_vdisktime(vdisktime, cfqg->vdisktime);
561 st->min_vdisktime = max_vdisktime(st->min_vdisktime, vdisktime);
565 * get averaged number of queues of RT/BE priority.
566 * average is updated, with a formula that gives more weight to higher numbers,
567 * to quickly follows sudden increases and decrease slowly
570 static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
571 struct cfq_group *cfqg, bool rt)
573 unsigned min_q, max_q;
574 unsigned mult = cfq_hist_divisor - 1;
575 unsigned round = cfq_hist_divisor / 2;
576 unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
578 min_q = min(cfqg->busy_queues_avg[rt], busy);
579 max_q = max(cfqg->busy_queues_avg[rt], busy);
580 cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) /
582 return cfqg->busy_queues_avg[rt];
585 static inline unsigned
586 cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg)
588 struct cfq_rb_root *st = &cfqd->grp_service_tree;
590 return cfq_target_latency * cfqg->weight / st->total_weight;
594 cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
596 unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
597 if (cfqd->cfq_latency) {
599 * interested queues (we consider only the ones with the same
600 * priority class in the cfq group)
602 unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg,
604 unsigned sync_slice = cfqd->cfq_slice[1];
605 unsigned expect_latency = sync_slice * iq;
606 unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg);
608 if (expect_latency > group_slice) {
609 unsigned base_low_slice = 2 * cfqd->cfq_slice_idle;
610 /* scale low_slice according to IO priority
611 * and sync vs async */
613 min(slice, base_low_slice * slice / sync_slice);
614 /* the adapted slice value is scaled to fit all iqs
615 * into the target latency */
616 slice = max(slice * group_slice / expect_latency,
620 cfqq->slice_start = jiffies;
621 cfqq->slice_end = jiffies + slice;
622 cfqq->allocated_slice = slice;
623 cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
627 * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end
628 * isn't valid until the first request from the dispatch is activated
629 * and the slice time set.
631 static inline bool cfq_slice_used(struct cfq_queue *cfqq)
633 if (cfq_cfqq_slice_new(cfqq))
635 if (time_before(jiffies, cfqq->slice_end))
642 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
643 * We choose the request that is closest to the head right now. Distance
644 * behind the head is penalized and only allowed to a certain extent.
646 static struct request *
647 cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
649 sector_t s1, s2, d1 = 0, d2 = 0;
650 unsigned long back_max;
651 #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */
652 #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */
653 unsigned wrap = 0; /* bit mask: requests behind the disk head? */
655 if (rq1 == NULL || rq1 == rq2)
660 if (rq_is_sync(rq1) && !rq_is_sync(rq2))
662 else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
664 if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))
666 else if ((rq2->cmd_flags & REQ_META) &&
667 !(rq1->cmd_flags & REQ_META))
670 s1 = blk_rq_pos(rq1);
671 s2 = blk_rq_pos(rq2);
674 * by definition, 1KiB is 2 sectors
676 back_max = cfqd->cfq_back_max * 2;
679 * Strict one way elevator _except_ in the case where we allow
680 * short backward seeks which are biased as twice the cost of a
681 * similar forward seek.
685 else if (s1 + back_max >= last)
686 d1 = (last - s1) * cfqd->cfq_back_penalty;
688 wrap |= CFQ_RQ1_WRAP;
692 else if (s2 + back_max >= last)
693 d2 = (last - s2) * cfqd->cfq_back_penalty;
695 wrap |= CFQ_RQ2_WRAP;
697 /* Found required data */
700 * By doing switch() on the bit mask "wrap" we avoid having to
701 * check two variables for all permutations: --> faster!
704 case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
720 case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
723 * Since both rqs are wrapped,
724 * start with the one that's further behind head
725 * (--> only *one* back seek required),
726 * since back seek takes more time than forward.
736 * The below is leftmost cache rbtree addon
738 static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
740 /* Service tree is empty */
745 root->left = rb_first(&root->rb);
748 return rb_entry(root->left, struct cfq_queue, rb_node);
753 static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root)
756 root->left = rb_first(&root->rb);
759 return rb_entry_cfqg(root->left);
764 static void rb_erase_init(struct rb_node *n, struct rb_root *root)
770 static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
774 rb_erase_init(n, &root->rb);
779 * would be nice to take fifo expire time into account as well
781 static struct request *
782 cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
783 struct request *last)
785 struct rb_node *rbnext = rb_next(&last->rb_node);
786 struct rb_node *rbprev = rb_prev(&last->rb_node);
787 struct request *next = NULL, *prev = NULL;
789 BUG_ON(RB_EMPTY_NODE(&last->rb_node));
792 prev = rb_entry_rq(rbprev);
795 next = rb_entry_rq(rbnext);
797 rbnext = rb_first(&cfqq->sort_list);
798 if (rbnext && rbnext != &last->rb_node)
799 next = rb_entry_rq(rbnext);
802 return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
805 static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
806 struct cfq_queue *cfqq)
809 * just an approximation, should be ok.
811 return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
812 cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
816 cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg)
818 return cfqg->vdisktime - st->min_vdisktime;
822 __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg)
824 struct rb_node **node = &st->rb.rb_node;
825 struct rb_node *parent = NULL;
826 struct cfq_group *__cfqg;
827 s64 key = cfqg_key(st, cfqg);
830 while (*node != NULL) {
832 __cfqg = rb_entry_cfqg(parent);
834 if (key < cfqg_key(st, __cfqg))
835 node = &parent->rb_left;
837 node = &parent->rb_right;
843 st->left = &cfqg->rb_node;
845 rb_link_node(&cfqg->rb_node, parent, node);
846 rb_insert_color(&cfqg->rb_node, &st->rb);
850 cfq_group_service_tree_add(struct cfq_data *cfqd, struct cfq_group *cfqg)
852 struct cfq_rb_root *st = &cfqd->grp_service_tree;
853 struct cfq_group *__cfqg;
861 * Currently put the group at the end. Later implement something
862 * so that groups get lesser vtime based on their weights, so that
863 * if group does not loose all if it was not continously backlogged.
865 n = rb_last(&st->rb);
867 __cfqg = rb_entry_cfqg(n);
868 cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY;
870 cfqg->vdisktime = st->min_vdisktime;
872 __cfq_group_service_tree_add(st, cfqg);
874 st->total_weight += cfqg->weight;
878 cfq_group_service_tree_del(struct cfq_data *cfqd, struct cfq_group *cfqg)
880 struct cfq_rb_root *st = &cfqd->grp_service_tree;
882 if (st->active == &cfqg->rb_node)
885 BUG_ON(cfqg->nr_cfqq < 1);
888 /* If there are other cfq queues under this group, don't delete it */
892 cfq_log_cfqg(cfqd, cfqg, "del_from_rr group");
894 st->total_weight -= cfqg->weight;
895 if (!RB_EMPTY_NODE(&cfqg->rb_node))
896 cfq_rb_erase(&cfqg->rb_node, st);
897 cfqg->saved_workload_slice = 0;
898 cfq_blkiocg_update_dequeue_stats(&cfqg->blkg, 1);
901 static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq)
903 unsigned int slice_used;
906 * Queue got expired before even a single request completed or
907 * got expired immediately after first request completion.
909 if (!cfqq->slice_start || cfqq->slice_start == jiffies) {
911 * Also charge the seek time incurred to the group, otherwise
912 * if there are mutiple queues in the group, each can dispatch
913 * a single request on seeky media and cause lots of seek time
914 * and group will never know it.
916 slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start),
919 slice_used = jiffies - cfqq->slice_start;
920 if (slice_used > cfqq->allocated_slice)
921 slice_used = cfqq->allocated_slice;
927 static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
928 struct cfq_queue *cfqq)
930 struct cfq_rb_root *st = &cfqd->grp_service_tree;
931 unsigned int used_sl, charge;
932 int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg)
933 - cfqg->service_tree_idle.count;
936 used_sl = charge = cfq_cfqq_slice_usage(cfqq);
939 charge = cfqq->slice_dispatch;
940 else if (!cfq_cfqq_sync(cfqq) && !nr_sync)
941 charge = cfqq->allocated_slice;
943 /* Can't update vdisktime while group is on service tree */
944 cfq_rb_erase(&cfqg->rb_node, st);
945 cfqg->vdisktime += cfq_scale_slice(charge, cfqg);
946 __cfq_group_service_tree_add(st, cfqg);
948 /* This group is being expired. Save the context */
949 if (time_after(cfqd->workload_expires, jiffies)) {
950 cfqg->saved_workload_slice = cfqd->workload_expires
952 cfqg->saved_workload = cfqd->serving_type;
953 cfqg->saved_serving_prio = cfqd->serving_prio;
955 cfqg->saved_workload_slice = 0;
957 cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
959 cfq_log_cfqq(cfqq->cfqd, cfqq, "sl_used=%u disp=%u charge=%u iops=%u",
960 used_sl, cfqq->slice_dispatch, charge, iops_mode(cfqd));
961 cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl);
962 cfq_blkiocg_set_start_empty_time(&cfqg->blkg);
965 #ifdef CONFIG_CFQ_GROUP_IOSCHED
966 static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg)
969 return container_of(blkg, struct cfq_group, blkg);
974 cfq_update_blkio_group_weight(struct blkio_group *blkg, unsigned int weight)
976 cfqg_of_blkg(blkg)->weight = weight;
979 static struct cfq_group *
980 cfq_find_alloc_cfqg(struct cfq_data *cfqd, struct cgroup *cgroup, int create)
982 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
983 struct cfq_group *cfqg = NULL;
986 struct cfq_rb_root *st;
987 struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
988 unsigned int major, minor;
990 cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key));
991 if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
992 sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
993 cfqg->blkg.dev = MKDEV(major, minor);
999 cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node);
1003 for_each_cfqg_st(cfqg, i, j, st)
1005 RB_CLEAR_NODE(&cfqg->rb_node);
1008 * Take the initial reference that will be released on destroy
1009 * This can be thought of a joint reference by cgroup and
1010 * elevator which will be dropped by either elevator exit
1011 * or cgroup deletion path depending on who is exiting first.
1013 atomic_set(&cfqg->ref, 1);
1015 /* Add group onto cgroup list */
1016 sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
1017 cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd,
1018 MKDEV(major, minor));
1019 cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev);
1021 /* Add group on cfqd list */
1022 hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list);
1029 * Search for the cfq group current task belongs to. If create = 1, then also
1030 * create the cfq group if it does not exist. request_queue lock must be held.
1032 static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create)
1034 struct cgroup *cgroup;
1035 struct cfq_group *cfqg = NULL;
1038 cgroup = task_cgroup(current, blkio_subsys_id);
1039 cfqg = cfq_find_alloc_cfqg(cfqd, cgroup, create);
1040 if (!cfqg && create)
1041 cfqg = &cfqd->root_group;
1046 static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg)
1048 atomic_inc(&cfqg->ref);
1052 static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg)
1054 /* Currently, all async queues are mapped to root group */
1055 if (!cfq_cfqq_sync(cfqq))
1056 cfqg = &cfqq->cfqd->root_group;
1059 /* cfqq reference on cfqg */
1060 atomic_inc(&cfqq->cfqg->ref);
1063 static void cfq_put_cfqg(struct cfq_group *cfqg)
1065 struct cfq_rb_root *st;
1068 BUG_ON(atomic_read(&cfqg->ref) <= 0);
1069 if (!atomic_dec_and_test(&cfqg->ref))
1071 for_each_cfqg_st(cfqg, i, j, st)
1072 BUG_ON(!RB_EMPTY_ROOT(&st->rb) || st->active != NULL);
1076 static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg)
1078 /* Something wrong if we are trying to remove same group twice */
1079 BUG_ON(hlist_unhashed(&cfqg->cfqd_node));
1081 hlist_del_init(&cfqg->cfqd_node);
1084 * Put the reference taken at the time of creation so that when all
1085 * queues are gone, group can be destroyed.
1090 static void cfq_release_cfq_groups(struct cfq_data *cfqd)
1092 struct hlist_node *pos, *n;
1093 struct cfq_group *cfqg;
1095 hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) {
1097 * If cgroup removal path got to blk_group first and removed
1098 * it from cgroup list, then it will take care of destroying
1101 if (!cfq_blkiocg_del_blkio_group(&cfqg->blkg))
1102 cfq_destroy_cfqg(cfqd, cfqg);
1107 * Blk cgroup controller notification saying that blkio_group object is being
1108 * delinked as associated cgroup object is going away. That also means that
1109 * no new IO will come in this group. So get rid of this group as soon as
1110 * any pending IO in the group is finished.
1112 * This function is called under rcu_read_lock(). key is the rcu protected
1113 * pointer. That means "key" is a valid cfq_data pointer as long as we are rcu
1116 * "key" was fetched from blkio_group under blkio_cgroup->lock. That means
1117 * it should not be NULL as even if elevator was exiting, cgroup deltion
1118 * path got to it first.
1120 void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg)
1122 unsigned long flags;
1123 struct cfq_data *cfqd = key;
1125 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
1126 cfq_destroy_cfqg(cfqd, cfqg_of_blkg(blkg));
1127 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
1130 #else /* GROUP_IOSCHED */
1131 static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create)
1133 return &cfqd->root_group;
1136 static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg)
1142 cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) {
1146 static void cfq_release_cfq_groups(struct cfq_data *cfqd) {}
1147 static inline void cfq_put_cfqg(struct cfq_group *cfqg) {}
1149 #endif /* GROUP_IOSCHED */
1152 * The cfqd->service_trees holds all pending cfq_queue's that have
1153 * requests waiting to be processed. It is sorted in the order that
1154 * we will service the queues.
1156 static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1159 struct rb_node **p, *parent;
1160 struct cfq_queue *__cfqq;
1161 unsigned long rb_key;
1162 struct cfq_rb_root *service_tree;
1165 int group_changed = 0;
1167 #ifdef CONFIG_CFQ_GROUP_IOSCHED
1168 if (!cfqd->cfq_group_isolation
1169 && cfqq_type(cfqq) == SYNC_NOIDLE_WORKLOAD
1170 && cfqq->cfqg && cfqq->cfqg != &cfqd->root_group) {
1171 /* Move this cfq to root group */
1172 cfq_log_cfqq(cfqd, cfqq, "moving to root group");
1173 if (!RB_EMPTY_NODE(&cfqq->rb_node))
1174 cfq_group_service_tree_del(cfqd, cfqq->cfqg);
1175 cfqq->orig_cfqg = cfqq->cfqg;
1176 cfqq->cfqg = &cfqd->root_group;
1177 atomic_inc(&cfqd->root_group.ref);
1179 } else if (!cfqd->cfq_group_isolation
1180 && cfqq_type(cfqq) == SYNC_WORKLOAD && cfqq->orig_cfqg) {
1181 /* cfqq is sequential now needs to go to its original group */
1182 BUG_ON(cfqq->cfqg != &cfqd->root_group);
1183 if (!RB_EMPTY_NODE(&cfqq->rb_node))
1184 cfq_group_service_tree_del(cfqd, cfqq->cfqg);
1185 cfq_put_cfqg(cfqq->cfqg);
1186 cfqq->cfqg = cfqq->orig_cfqg;
1187 cfqq->orig_cfqg = NULL;
1189 cfq_log_cfqq(cfqd, cfqq, "moved to origin group");
1193 service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq),
1195 if (cfq_class_idle(cfqq)) {
1196 rb_key = CFQ_IDLE_DELAY;
1197 parent = rb_last(&service_tree->rb);
1198 if (parent && parent != &cfqq->rb_node) {
1199 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
1200 rb_key += __cfqq->rb_key;
1203 } else if (!add_front) {
1205 * Get our rb key offset. Subtract any residual slice
1206 * value carried from last service. A negative resid
1207 * count indicates slice overrun, and this should position
1208 * the next service time further away in the tree.
1210 rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
1211 rb_key -= cfqq->slice_resid;
1212 cfqq->slice_resid = 0;
1215 __cfqq = cfq_rb_first(service_tree);
1216 rb_key += __cfqq ? __cfqq->rb_key : jiffies;
1219 if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
1222 * same position, nothing more to do
1224 if (rb_key == cfqq->rb_key &&
1225 cfqq->service_tree == service_tree)
1228 cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
1229 cfqq->service_tree = NULL;
1234 cfqq->service_tree = service_tree;
1235 p = &service_tree->rb.rb_node;
1240 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
1243 * sort by key, that represents service time.
1245 if (time_before(rb_key, __cfqq->rb_key))
1248 n = &(*p)->rb_right;
1256 service_tree->left = &cfqq->rb_node;
1258 cfqq->rb_key = rb_key;
1259 rb_link_node(&cfqq->rb_node, parent, p);
1260 rb_insert_color(&cfqq->rb_node, &service_tree->rb);
1261 service_tree->count++;
1262 if ((add_front || !new_cfqq) && !group_changed)
1264 cfq_group_service_tree_add(cfqd, cfqq->cfqg);
1267 static struct cfq_queue *
1268 cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root,
1269 sector_t sector, struct rb_node **ret_parent,
1270 struct rb_node ***rb_link)
1272 struct rb_node **p, *parent;
1273 struct cfq_queue *cfqq = NULL;
1281 cfqq = rb_entry(parent, struct cfq_queue, p_node);
1284 * Sort strictly based on sector. Smallest to the left,
1285 * largest to the right.
1287 if (sector > blk_rq_pos(cfqq->next_rq))
1288 n = &(*p)->rb_right;
1289 else if (sector < blk_rq_pos(cfqq->next_rq))
1297 *ret_parent = parent;
1303 static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1305 struct rb_node **p, *parent;
1306 struct cfq_queue *__cfqq;
1309 rb_erase(&cfqq->p_node, cfqq->p_root);
1310 cfqq->p_root = NULL;
1313 if (cfq_class_idle(cfqq))
1318 cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
1319 __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
1320 blk_rq_pos(cfqq->next_rq), &parent, &p);
1322 rb_link_node(&cfqq->p_node, parent, p);
1323 rb_insert_color(&cfqq->p_node, cfqq->p_root);
1325 cfqq->p_root = NULL;
1329 * Update cfqq's position in the service tree.
1331 static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1334 * Resorting requires the cfqq to be on the RR list already.
1336 if (cfq_cfqq_on_rr(cfqq)) {
1337 cfq_service_tree_add(cfqd, cfqq, 0);
1338 cfq_prio_tree_add(cfqd, cfqq);
1343 * add to busy list of queues for service, trying to be fair in ordering
1344 * the pending list according to last request service
1346 static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1348 cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
1349 BUG_ON(cfq_cfqq_on_rr(cfqq));
1350 cfq_mark_cfqq_on_rr(cfqq);
1351 cfqd->busy_queues++;
1353 cfq_resort_rr_list(cfqd, cfqq);
1357 * Called when the cfqq no longer has requests pending, remove it from
1360 static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1362 cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
1363 BUG_ON(!cfq_cfqq_on_rr(cfqq));
1364 cfq_clear_cfqq_on_rr(cfqq);
1366 if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
1367 cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
1368 cfqq->service_tree = NULL;
1371 rb_erase(&cfqq->p_node, cfqq->p_root);
1372 cfqq->p_root = NULL;
1375 cfq_group_service_tree_del(cfqd, cfqq->cfqg);
1376 BUG_ON(!cfqd->busy_queues);
1377 cfqd->busy_queues--;
1381 * rb tree support functions
1383 static void cfq_del_rq_rb(struct request *rq)
1385 struct cfq_queue *cfqq = RQ_CFQQ(rq);
1386 const int sync = rq_is_sync(rq);
1388 BUG_ON(!cfqq->queued[sync]);
1389 cfqq->queued[sync]--;
1391 elv_rb_del(&cfqq->sort_list, rq);
1393 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) {
1395 * Queue will be deleted from service tree when we actually
1396 * expire it later. Right now just remove it from prio tree
1400 rb_erase(&cfqq->p_node, cfqq->p_root);
1401 cfqq->p_root = NULL;
1406 static void cfq_add_rq_rb(struct request *rq)
1408 struct cfq_queue *cfqq = RQ_CFQQ(rq);
1409 struct cfq_data *cfqd = cfqq->cfqd;
1410 struct request *__alias, *prev;
1412 cfqq->queued[rq_is_sync(rq)]++;
1415 * looks a little odd, but the first insert might return an alias.
1416 * if that happens, put the alias on the dispatch list
1418 while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
1419 cfq_dispatch_insert(cfqd->queue, __alias);
1421 if (!cfq_cfqq_on_rr(cfqq))
1422 cfq_add_cfqq_rr(cfqd, cfqq);
1425 * check if this request is a better next-serve candidate
1427 prev = cfqq->next_rq;
1428 cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position);
1431 * adjust priority tree position, if ->next_rq changes
1433 if (prev != cfqq->next_rq)
1434 cfq_prio_tree_add(cfqd, cfqq);
1436 BUG_ON(!cfqq->next_rq);
1439 static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
1441 elv_rb_del(&cfqq->sort_list, rq);
1442 cfqq->queued[rq_is_sync(rq)]--;
1443 cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg,
1444 rq_data_dir(rq), rq_is_sync(rq));
1446 cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg,
1447 &cfqq->cfqd->serving_group->blkg, rq_data_dir(rq),
1451 static struct request *
1452 cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
1454 struct task_struct *tsk = current;
1455 struct cfq_io_context *cic;
1456 struct cfq_queue *cfqq;
1458 cic = cfq_cic_lookup(cfqd, tsk->io_context);
1462 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
1464 sector_t sector = bio->bi_sector + bio_sectors(bio);
1466 return elv_rb_find(&cfqq->sort_list, sector);
1472 static void cfq_activate_request(struct request_queue *q, struct request *rq)
1474 struct cfq_data *cfqd = q->elevator->elevator_data;
1476 cfqd->rq_in_driver++;
1477 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
1478 cfqd->rq_in_driver);
1480 cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
1483 static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
1485 struct cfq_data *cfqd = q->elevator->elevator_data;
1487 WARN_ON(!cfqd->rq_in_driver);
1488 cfqd->rq_in_driver--;
1489 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
1490 cfqd->rq_in_driver);
1493 static void cfq_remove_request(struct request *rq)
1495 struct cfq_queue *cfqq = RQ_CFQQ(rq);
1497 if (cfqq->next_rq == rq)
1498 cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
1500 list_del_init(&rq->queuelist);
1503 cfqq->cfqd->rq_queued--;
1504 cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg,
1505 rq_data_dir(rq), rq_is_sync(rq));
1506 if (rq->cmd_flags & REQ_META) {
1507 WARN_ON(!cfqq->meta_pending);
1508 cfqq->meta_pending--;
1512 static int cfq_merge(struct request_queue *q, struct request **req,
1515 struct cfq_data *cfqd = q->elevator->elevator_data;
1516 struct request *__rq;
1518 __rq = cfq_find_rq_fmerge(cfqd, bio);
1519 if (__rq && elv_rq_merge_ok(__rq, bio)) {
1521 return ELEVATOR_FRONT_MERGE;
1524 return ELEVATOR_NO_MERGE;
1527 static void cfq_merged_request(struct request_queue *q, struct request *req,
1530 if (type == ELEVATOR_FRONT_MERGE) {
1531 struct cfq_queue *cfqq = RQ_CFQQ(req);
1533 cfq_reposition_rq_rb(cfqq, req);
1537 static void cfq_bio_merged(struct request_queue *q, struct request *req,
1540 cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(req))->blkg,
1541 bio_data_dir(bio), cfq_bio_sync(bio));
1545 cfq_merged_requests(struct request_queue *q, struct request *rq,
1546 struct request *next)
1548 struct cfq_queue *cfqq = RQ_CFQQ(rq);
1550 * reposition in fifo if next is older than rq
1552 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
1553 time_before(rq_fifo_time(next), rq_fifo_time(rq))) {
1554 list_move(&rq->queuelist, &next->queuelist);
1555 rq_set_fifo_time(rq, rq_fifo_time(next));
1558 if (cfqq->next_rq == next)
1560 cfq_remove_request(next);
1561 cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(rq))->blkg,
1562 rq_data_dir(next), rq_is_sync(next));
1565 static int cfq_allow_merge(struct request_queue *q, struct request *rq,
1568 struct cfq_data *cfqd = q->elevator->elevator_data;
1569 struct cfq_io_context *cic;
1570 struct cfq_queue *cfqq;
1573 * Disallow merge of a sync bio into an async request.
1575 if (cfq_bio_sync(bio) && !rq_is_sync(rq))
1579 * Lookup the cfqq that this bio will be queued with. Allow
1580 * merge only if rq is queued there.
1582 cic = cfq_cic_lookup(cfqd, current->io_context);
1586 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
1587 return cfqq == RQ_CFQQ(rq);
1590 static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1592 del_timer(&cfqd->idle_slice_timer);
1593 cfq_blkiocg_update_idle_time_stats(&cfqq->cfqg->blkg);
1596 static void __cfq_set_active_queue(struct cfq_data *cfqd,
1597 struct cfq_queue *cfqq)
1600 cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d",
1601 cfqd->serving_prio, cfqd->serving_type);
1602 cfq_blkiocg_update_avg_queue_size_stats(&cfqq->cfqg->blkg);
1603 cfqq->slice_start = 0;
1604 cfqq->dispatch_start = jiffies;
1605 cfqq->allocated_slice = 0;
1606 cfqq->slice_end = 0;
1607 cfqq->slice_dispatch = 0;
1609 cfq_clear_cfqq_wait_request(cfqq);
1610 cfq_clear_cfqq_must_dispatch(cfqq);
1611 cfq_clear_cfqq_must_alloc_slice(cfqq);
1612 cfq_clear_cfqq_fifo_expire(cfqq);
1613 cfq_mark_cfqq_slice_new(cfqq);
1615 cfq_del_timer(cfqd, cfqq);
1618 cfqd->active_queue = cfqq;
1622 * current cfqq expired its slice (or was too idle), select new one
1625 __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1628 cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);
1630 if (cfq_cfqq_wait_request(cfqq))
1631 cfq_del_timer(cfqd, cfqq);
1633 cfq_clear_cfqq_wait_request(cfqq);
1634 cfq_clear_cfqq_wait_busy(cfqq);
1637 * If this cfqq is shared between multiple processes, check to
1638 * make sure that those processes are still issuing I/Os within
1639 * the mean seek distance. If not, it may be time to break the
1640 * queues apart again.
1642 if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq))
1643 cfq_mark_cfqq_split_coop(cfqq);
1646 * store what was left of this slice, if the queue idled/timed out
1648 if (timed_out && !cfq_cfqq_slice_new(cfqq)) {
1649 cfqq->slice_resid = cfqq->slice_end - jiffies;
1650 cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid);
1653 cfq_group_served(cfqd, cfqq->cfqg, cfqq);
1655 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
1656 cfq_del_cfqq_rr(cfqd, cfqq);
1658 cfq_resort_rr_list(cfqd, cfqq);
1660 if (cfqq == cfqd->active_queue)
1661 cfqd->active_queue = NULL;
1663 if (&cfqq->cfqg->rb_node == cfqd->grp_service_tree.active)
1664 cfqd->grp_service_tree.active = NULL;
1666 if (cfqd->active_cic) {
1667 put_io_context(cfqd->active_cic->ioc);
1668 cfqd->active_cic = NULL;
1672 static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out)
1674 struct cfq_queue *cfqq = cfqd->active_queue;
1677 __cfq_slice_expired(cfqd, cfqq, timed_out);
1681 * Get next queue for service. Unless we have a queue preemption,
1682 * we'll simply select the first cfqq in the service tree.
1684 static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
1686 struct cfq_rb_root *service_tree =
1687 service_tree_for(cfqd->serving_group, cfqd->serving_prio,
1688 cfqd->serving_type);
1690 if (!cfqd->rq_queued)
1693 /* There is nothing to dispatch */
1696 if (RB_EMPTY_ROOT(&service_tree->rb))
1698 return cfq_rb_first(service_tree);
1701 static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd)
1703 struct cfq_group *cfqg;
1704 struct cfq_queue *cfqq;
1706 struct cfq_rb_root *st;
1708 if (!cfqd->rq_queued)
1711 cfqg = cfq_get_next_cfqg(cfqd);
1715 for_each_cfqg_st(cfqg, i, j, st)
1716 if ((cfqq = cfq_rb_first(st)) != NULL)
1722 * Get and set a new active queue for service.
1724 static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd,
1725 struct cfq_queue *cfqq)
1728 cfqq = cfq_get_next_queue(cfqd);
1730 __cfq_set_active_queue(cfqd, cfqq);
1734 static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
1737 if (blk_rq_pos(rq) >= cfqd->last_position)
1738 return blk_rq_pos(rq) - cfqd->last_position;
1740 return cfqd->last_position - blk_rq_pos(rq);
1743 static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1746 return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR;
1749 static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
1750 struct cfq_queue *cur_cfqq)
1752 struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio];
1753 struct rb_node *parent, *node;
1754 struct cfq_queue *__cfqq;
1755 sector_t sector = cfqd->last_position;
1757 if (RB_EMPTY_ROOT(root))
1761 * First, if we find a request starting at the end of the last
1762 * request, choose it.
1764 __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL);
1769 * If the exact sector wasn't found, the parent of the NULL leaf
1770 * will contain the closest sector.
1772 __cfqq = rb_entry(parent, struct cfq_queue, p_node);
1773 if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
1776 if (blk_rq_pos(__cfqq->next_rq) < sector)
1777 node = rb_next(&__cfqq->p_node);
1779 node = rb_prev(&__cfqq->p_node);
1783 __cfqq = rb_entry(node, struct cfq_queue, p_node);
1784 if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
1792 * cur_cfqq - passed in so that we don't decide that the current queue is
1793 * closely cooperating with itself.
1795 * So, basically we're assuming that that cur_cfqq has dispatched at least
1796 * one request, and that cfqd->last_position reflects a position on the disk
1797 * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid
1800 static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd,
1801 struct cfq_queue *cur_cfqq)
1803 struct cfq_queue *cfqq;
1805 if (cfq_class_idle(cur_cfqq))
1807 if (!cfq_cfqq_sync(cur_cfqq))
1809 if (CFQQ_SEEKY(cur_cfqq))
1813 * Don't search priority tree if it's the only queue in the group.
1815 if (cur_cfqq->cfqg->nr_cfqq == 1)
1819 * We should notice if some of the queues are cooperating, eg
1820 * working closely on the same area of the disk. In that case,
1821 * we can group them together and don't waste time idling.
1823 cfqq = cfqq_close(cfqd, cur_cfqq);
1827 /* If new queue belongs to different cfq_group, don't choose it */
1828 if (cur_cfqq->cfqg != cfqq->cfqg)
1832 * It only makes sense to merge sync queues.
1834 if (!cfq_cfqq_sync(cfqq))
1836 if (CFQQ_SEEKY(cfqq))
1840 * Do not merge queues of different priority classes
1842 if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq))
1849 * Determine whether we should enforce idle window for this queue.
1852 static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1854 enum wl_prio_t prio = cfqq_prio(cfqq);
1855 struct cfq_rb_root *service_tree = cfqq->service_tree;
1857 BUG_ON(!service_tree);
1858 BUG_ON(!service_tree->count);
1860 if (!cfqd->cfq_slice_idle)
1863 /* We never do for idle class queues. */
1864 if (prio == IDLE_WORKLOAD)
1867 /* We do for queues that were marked with idle window flag. */
1868 if (cfq_cfqq_idle_window(cfqq) &&
1869 !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag))
1873 * Otherwise, we do only if they are the last ones
1874 * in their service tree.
1876 if (service_tree->count == 1 && cfq_cfqq_sync(cfqq))
1878 cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d",
1879 service_tree->count);
1883 static void cfq_arm_slice_timer(struct cfq_data *cfqd)
1885 struct cfq_queue *cfqq = cfqd->active_queue;
1886 struct cfq_io_context *cic;
1890 * SSD device without seek penalty, disable idling. But only do so
1891 * for devices that support queuing, otherwise we still have a problem
1892 * with sync vs async workloads.
1894 if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)
1897 WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
1898 WARN_ON(cfq_cfqq_slice_new(cfqq));
1901 * idle is disabled, either manually or by past process history
1903 if (!cfq_should_idle(cfqd, cfqq))
1907 * still active requests from this queue, don't idle
1909 if (cfqq->dispatched)
1913 * task has exited, don't wait
1915 cic = cfqd->active_cic;
1916 if (!cic || !atomic_read(&cic->ioc->nr_tasks))
1920 * If our average think time is larger than the remaining time
1921 * slice, then don't idle. This avoids overrunning the allotted
1924 if (sample_valid(cic->ttime_samples) &&
1925 (cfqq->slice_end - jiffies < cic->ttime_mean)) {
1926 cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%d",
1931 cfq_mark_cfqq_wait_request(cfqq);
1933 sl = cfqd->cfq_slice_idle;
1935 mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
1936 cfq_blkiocg_update_set_idle_time_stats(&cfqq->cfqg->blkg);
1937 cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu", sl);
1941 * Move request from internal lists to the request queue dispatch list.
1943 static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
1945 struct cfq_data *cfqd = q->elevator->elevator_data;
1946 struct cfq_queue *cfqq = RQ_CFQQ(rq);
1948 cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");
1950 cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq);
1951 cfq_remove_request(rq);
1953 elv_dispatch_sort(q, rq);
1955 cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++;
1956 cfq_blkiocg_update_dispatch_stats(&cfqq->cfqg->blkg, blk_rq_bytes(rq),
1957 rq_data_dir(rq), rq_is_sync(rq));
1961 * return expired entry, or NULL to just start from scratch in rbtree
1963 static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
1965 struct request *rq = NULL;
1967 if (cfq_cfqq_fifo_expire(cfqq))
1970 cfq_mark_cfqq_fifo_expire(cfqq);
1972 if (list_empty(&cfqq->fifo))
1975 rq = rq_entry_fifo(cfqq->fifo.next);
1976 if (time_before(jiffies, rq_fifo_time(rq)))
1979 cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq);
1984 cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1986 const int base_rq = cfqd->cfq_slice_async_rq;
1988 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
1990 return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
1994 * Must be called with the queue_lock held.
1996 static int cfqq_process_refs(struct cfq_queue *cfqq)
1998 int process_refs, io_refs;
2000 io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE];
2001 process_refs = atomic_read(&cfqq->ref) - io_refs;
2002 BUG_ON(process_refs < 0);
2003 return process_refs;
2006 static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq)
2008 int process_refs, new_process_refs;
2009 struct cfq_queue *__cfqq;
2012 * If there are no process references on the new_cfqq, then it is
2013 * unsafe to follow the ->new_cfqq chain as other cfqq's in the
2014 * chain may have dropped their last reference (not just their
2015 * last process reference).
2017 if (!cfqq_process_refs(new_cfqq))
2020 /* Avoid a circular list and skip interim queue merges */
2021 while ((__cfqq = new_cfqq->new_cfqq)) {
2027 process_refs = cfqq_process_refs(cfqq);
2028 new_process_refs = cfqq_process_refs(new_cfqq);
2030 * If the process for the cfqq has gone away, there is no
2031 * sense in merging the queues.
2033 if (process_refs == 0 || new_process_refs == 0)
2037 * Merge in the direction of the lesser amount of work.
2039 if (new_process_refs >= process_refs) {
2040 cfqq->new_cfqq = new_cfqq;
2041 atomic_add(process_refs, &new_cfqq->ref);
2043 new_cfqq->new_cfqq = cfqq;
2044 atomic_add(new_process_refs, &cfqq->ref);
2048 static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd,
2049 struct cfq_group *cfqg, enum wl_prio_t prio)
2051 struct cfq_queue *queue;
2053 bool key_valid = false;
2054 unsigned long lowest_key = 0;
2055 enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD;
2057 for (i = 0; i <= SYNC_WORKLOAD; ++i) {
2058 /* select the one with lowest rb_key */
2059 queue = cfq_rb_first(service_tree_for(cfqg, prio, i));
2061 (!key_valid || time_before(queue->rb_key, lowest_key))) {
2062 lowest_key = queue->rb_key;
2071 static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg)
2075 struct cfq_rb_root *st;
2076 unsigned group_slice;
2079 cfqd->serving_prio = IDLE_WORKLOAD;
2080 cfqd->workload_expires = jiffies + 1;
2084 /* Choose next priority. RT > BE > IDLE */
2085 if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg))
2086 cfqd->serving_prio = RT_WORKLOAD;
2087 else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg))
2088 cfqd->serving_prio = BE_WORKLOAD;
2090 cfqd->serving_prio = IDLE_WORKLOAD;
2091 cfqd->workload_expires = jiffies + 1;
2096 * For RT and BE, we have to choose also the type
2097 * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload
2100 st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type);
2104 * check workload expiration, and that we still have other queues ready
2106 if (count && !time_after(jiffies, cfqd->workload_expires))
2109 /* otherwise select new workload type */
2110 cfqd->serving_type =
2111 cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio);
2112 st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type);
2116 * the workload slice is computed as a fraction of target latency
2117 * proportional to the number of queues in that workload, over
2118 * all the queues in the same priority class
2120 group_slice = cfq_group_slice(cfqd, cfqg);
2122 slice = group_slice * count /
2123 max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio],
2124 cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg));
2126 if (cfqd->serving_type == ASYNC_WORKLOAD) {
2130 * Async queues are currently system wide. Just taking
2131 * proportion of queues with-in same group will lead to higher
2132 * async ratio system wide as generally root group is going
2133 * to have higher weight. A more accurate thing would be to
2134 * calculate system wide asnc/sync ratio.
2136 tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg);
2137 tmp = tmp/cfqd->busy_queues;
2138 slice = min_t(unsigned, slice, tmp);
2140 /* async workload slice is scaled down according to
2141 * the sync/async slice ratio. */
2142 slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1];
2144 /* sync workload slice is at least 2 * cfq_slice_idle */
2145 slice = max(slice, 2 * cfqd->cfq_slice_idle);
2147 slice = max_t(unsigned, slice, CFQ_MIN_TT);
2148 cfq_log(cfqd, "workload slice:%d", slice);
2149 cfqd->workload_expires = jiffies + slice;
2150 cfqd->noidle_tree_requires_idle = false;
2153 static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd)
2155 struct cfq_rb_root *st = &cfqd->grp_service_tree;
2156 struct cfq_group *cfqg;
2158 if (RB_EMPTY_ROOT(&st->rb))
2160 cfqg = cfq_rb_first_group(st);
2161 st->active = &cfqg->rb_node;
2162 update_min_vdisktime(st);
2166 static void cfq_choose_cfqg(struct cfq_data *cfqd)
2168 struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd);
2170 cfqd->serving_group = cfqg;
2172 /* Restore the workload type data */
2173 if (cfqg->saved_workload_slice) {
2174 cfqd->workload_expires = jiffies + cfqg->saved_workload_slice;
2175 cfqd->serving_type = cfqg->saved_workload;
2176 cfqd->serving_prio = cfqg->saved_serving_prio;
2178 cfqd->workload_expires = jiffies - 1;
2180 choose_service_tree(cfqd, cfqg);
2184 * Select a queue for service. If we have a current active queue,
2185 * check whether to continue servicing it, or retrieve and set a new one.
2187 static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
2189 struct cfq_queue *cfqq, *new_cfqq = NULL;
2191 cfqq = cfqd->active_queue;
2195 if (!cfqd->rq_queued)
2199 * We were waiting for group to get backlogged. Expire the queue
2201 if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list))
2205 * The active queue has run out of time, expire it and select new.
2207 if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) {
2209 * If slice had not expired at the completion of last request
2210 * we might not have turned on wait_busy flag. Don't expire
2211 * the queue yet. Allow the group to get backlogged.
2213 * The very fact that we have used the slice, that means we
2214 * have been idling all along on this queue and it should be
2215 * ok to wait for this request to complete.
2217 if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list)
2218 && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
2226 * The active queue has requests and isn't expired, allow it to
2229 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
2233 * If another queue has a request waiting within our mean seek
2234 * distance, let it run. The expire code will check for close
2235 * cooperators and put the close queue at the front of the service
2236 * tree. If possible, merge the expiring queue with the new cfqq.
2238 new_cfqq = cfq_close_cooperator(cfqd, cfqq);
2240 if (!cfqq->new_cfqq)
2241 cfq_setup_merge(cfqq, new_cfqq);
2246 * No requests pending. If the active queue still has requests in
2247 * flight or is idling for a new request, allow either of these
2248 * conditions to happen (or time out) before selecting a new queue.
2250 if (timer_pending(&cfqd->idle_slice_timer) ||
2251 (cfqq->dispatched && cfq_should_idle(cfqd, cfqq))) {
2257 cfq_slice_expired(cfqd, 0);
2260 * Current queue expired. Check if we have to switch to a new
2264 cfq_choose_cfqg(cfqd);
2266 cfqq = cfq_set_active_queue(cfqd, new_cfqq);
2271 static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
2275 while (cfqq->next_rq) {
2276 cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq);
2280 BUG_ON(!list_empty(&cfqq->fifo));
2282 /* By default cfqq is not expired if it is empty. Do it explicitly */
2283 __cfq_slice_expired(cfqq->cfqd, cfqq, 0);
2288 * Drain our current requests. Used for barriers and when switching
2289 * io schedulers on-the-fly.
2291 static int cfq_forced_dispatch(struct cfq_data *cfqd)
2293 struct cfq_queue *cfqq;
2296 /* Expire the timeslice of the current active queue first */
2297 cfq_slice_expired(cfqd, 0);
2298 while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) {
2299 __cfq_set_active_queue(cfqd, cfqq);
2300 dispatched += __cfq_forced_dispatch_cfqq(cfqq);
2303 BUG_ON(cfqd->busy_queues);
2305 cfq_log(cfqd, "forced_dispatch=%d", dispatched);
2309 static inline bool cfq_slice_used_soon(struct cfq_data *cfqd,
2310 struct cfq_queue *cfqq)
2312 /* the queue hasn't finished any request, can't estimate */
2313 if (cfq_cfqq_slice_new(cfqq))
2315 if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched,
2322 static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2324 unsigned int max_dispatch;
2327 * Drain async requests before we start sync IO
2329 if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC])
2333 * If this is an async queue and we have sync IO in flight, let it wait
2335 if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq))
2338 max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1);
2339 if (cfq_class_idle(cfqq))
2343 * Does this cfqq already have too much IO in flight?
2345 if (cfqq->dispatched >= max_dispatch) {
2347 * idle queue must always only have a single IO in flight
2349 if (cfq_class_idle(cfqq))
2353 * We have other queues, don't allow more IO from this one
2355 if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq))
2359 * Sole queue user, no limit
2361 if (cfqd->busy_queues == 1)
2365 * Normally we start throttling cfqq when cfq_quantum/2
2366 * requests have been dispatched. But we can drive
2367 * deeper queue depths at the beginning of slice
2368 * subjected to upper limit of cfq_quantum.
2370 max_dispatch = cfqd->cfq_quantum;
2374 * Async queues must wait a bit before being allowed dispatch.
2375 * We also ramp up the dispatch depth gradually for async IO,
2376 * based on the last sync IO we serviced
2378 if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) {
2379 unsigned long last_sync = jiffies - cfqd->last_delayed_sync;
2382 depth = last_sync / cfqd->cfq_slice[1];
2383 if (!depth && !cfqq->dispatched)
2385 if (depth < max_dispatch)
2386 max_dispatch = depth;
2390 * If we're below the current max, allow a dispatch
2392 return cfqq->dispatched < max_dispatch;
2396 * Dispatch a request from cfqq, moving them to the request queue
2399 static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2403 BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
2405 if (!cfq_may_dispatch(cfqd, cfqq))
2409 * follow expired path, else get first next available
2411 rq = cfq_check_fifo(cfqq);
2416 * insert request into driver dispatch list
2418 cfq_dispatch_insert(cfqd->queue, rq);
2420 if (!cfqd->active_cic) {
2421 struct cfq_io_context *cic = RQ_CIC(rq);
2423 atomic_long_inc(&cic->ioc->refcount);
2424 cfqd->active_cic = cic;
2431 * Find the cfqq that we need to service and move a request from that to the
2434 static int cfq_dispatch_requests(struct request_queue *q, int force)
2436 struct cfq_data *cfqd = q->elevator->elevator_data;
2437 struct cfq_queue *cfqq;
2439 if (!cfqd->busy_queues)
2442 if (unlikely(force))
2443 return cfq_forced_dispatch(cfqd);
2445 cfqq = cfq_select_queue(cfqd);
2450 * Dispatch a request from this cfqq, if it is allowed
2452 if (!cfq_dispatch_request(cfqd, cfqq))
2455 cfqq->slice_dispatch++;
2456 cfq_clear_cfqq_must_dispatch(cfqq);
2459 * expire an async queue immediately if it has used up its slice. idle
2460 * queue always expire after 1 dispatch round.
2462 if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) &&
2463 cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
2464 cfq_class_idle(cfqq))) {
2465 cfqq->slice_end = jiffies + 1;
2466 cfq_slice_expired(cfqd, 0);
2469 cfq_log_cfqq(cfqd, cfqq, "dispatched a request");
2474 * task holds one reference to the queue, dropped when task exits. each rq
2475 * in-flight on this queue also holds a reference, dropped when rq is freed.
2477 * Each cfq queue took a reference on the parent group. Drop it now.
2478 * queue lock must be held here.
2480 static void cfq_put_queue(struct cfq_queue *cfqq)
2482 struct cfq_data *cfqd = cfqq->cfqd;
2483 struct cfq_group *cfqg, *orig_cfqg;
2485 BUG_ON(atomic_read(&cfqq->ref) <= 0);
2487 if (!atomic_dec_and_test(&cfqq->ref))
2490 cfq_log_cfqq(cfqd, cfqq, "put_queue");
2491 BUG_ON(rb_first(&cfqq->sort_list));
2492 BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
2494 orig_cfqg = cfqq->orig_cfqg;
2496 if (unlikely(cfqd->active_queue == cfqq)) {
2497 __cfq_slice_expired(cfqd, cfqq, 0);
2498 cfq_schedule_dispatch(cfqd);
2501 BUG_ON(cfq_cfqq_on_rr(cfqq));
2502 kmem_cache_free(cfq_pool, cfqq);
2505 cfq_put_cfqg(orig_cfqg);
2509 * Must always be called with the rcu_read_lock() held
2512 __call_for_each_cic(struct io_context *ioc,
2513 void (*func)(struct io_context *, struct cfq_io_context *))
2515 struct cfq_io_context *cic;
2516 struct hlist_node *n;
2518 hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list)
2523 * Call func for each cic attached to this ioc.
2526 call_for_each_cic(struct io_context *ioc,
2527 void (*func)(struct io_context *, struct cfq_io_context *))
2530 __call_for_each_cic(ioc, func);
2534 static void cfq_cic_free_rcu(struct rcu_head *head)
2536 struct cfq_io_context *cic;
2538 cic = container_of(head, struct cfq_io_context, rcu_head);
2540 kmem_cache_free(cfq_ioc_pool, cic);
2541 elv_ioc_count_dec(cfq_ioc_count);
2545 * CFQ scheduler is exiting, grab exit lock and check
2546 * the pending io context count. If it hits zero,
2547 * complete ioc_gone and set it back to NULL
2549 spin_lock(&ioc_gone_lock);
2550 if (ioc_gone && !elv_ioc_count_read(cfq_ioc_count)) {
2554 spin_unlock(&ioc_gone_lock);
2558 static void cfq_cic_free(struct cfq_io_context *cic)
2560 call_rcu(&cic->rcu_head, cfq_cic_free_rcu);
2563 static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic)
2565 unsigned long flags;
2566 unsigned long dead_key = (unsigned long) cic->key;
2568 BUG_ON(!(dead_key & CIC_DEAD_KEY));
2570 spin_lock_irqsave(&ioc->lock, flags);
2571 radix_tree_delete(&ioc->radix_root, dead_key >> CIC_DEAD_INDEX_SHIFT);
2572 hlist_del_rcu(&cic->cic_list);
2573 spin_unlock_irqrestore(&ioc->lock, flags);
2579 * Must be called with rcu_read_lock() held or preemption otherwise disabled.
2580 * Only two callers of this - ->dtor() which is called with the rcu_read_lock(),
2581 * and ->trim() which is called with the task lock held
2583 static void cfq_free_io_context(struct io_context *ioc)
2586 * ioc->refcount is zero here, or we are called from elv_unregister(),
2587 * so no more cic's are allowed to be linked into this ioc. So it
2588 * should be ok to iterate over the known list, we will see all cic's
2589 * since no new ones are added.
2591 __call_for_each_cic(ioc, cic_free_func);
2594 static void cfq_put_cooperator(struct cfq_queue *cfqq)
2596 struct cfq_queue *__cfqq, *next;
2599 * If this queue was scheduled to merge with another queue, be
2600 * sure to drop the reference taken on that queue (and others in
2601 * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs.
2603 __cfqq = cfqq->new_cfqq;
2605 if (__cfqq == cfqq) {
2606 WARN(1, "cfqq->new_cfqq loop detected\n");
2609 next = __cfqq->new_cfqq;
2610 cfq_put_queue(__cfqq);
2615 static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2617 if (unlikely(cfqq == cfqd->active_queue)) {
2618 __cfq_slice_expired(cfqd, cfqq, 0);
2619 cfq_schedule_dispatch(cfqd);
2622 cfq_put_cooperator(cfqq);
2624 cfq_put_queue(cfqq);
2627 static void __cfq_exit_single_io_context(struct cfq_data *cfqd,
2628 struct cfq_io_context *cic)
2630 struct io_context *ioc = cic->ioc;
2632 list_del_init(&cic->queue_list);
2635 * Make sure dead mark is seen for dead queues
2638 cic->key = cfqd_dead_key(cfqd);
2640 if (ioc->ioc_data == cic)
2641 rcu_assign_pointer(ioc->ioc_data, NULL);
2643 if (cic->cfqq[BLK_RW_ASYNC]) {
2644 cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]);
2645 cic->cfqq[BLK_RW_ASYNC] = NULL;
2648 if (cic->cfqq[BLK_RW_SYNC]) {
2649 cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]);
2650 cic->cfqq[BLK_RW_SYNC] = NULL;
2654 static void cfq_exit_single_io_context(struct io_context *ioc,
2655 struct cfq_io_context *cic)
2657 struct cfq_data *cfqd = cic_to_cfqd(cic);
2660 struct request_queue *q = cfqd->queue;
2661 unsigned long flags;
2663 spin_lock_irqsave(q->queue_lock, flags);
2666 * Ensure we get a fresh copy of the ->key to prevent
2667 * race between exiting task and queue
2669 smp_read_barrier_depends();
2670 if (cic->key == cfqd)
2671 __cfq_exit_single_io_context(cfqd, cic);
2673 spin_unlock_irqrestore(q->queue_lock, flags);
2678 * The process that ioc belongs to has exited, we need to clean up
2679 * and put the internal structures we have that belongs to that process.
2681 static void cfq_exit_io_context(struct io_context *ioc)
2683 call_for_each_cic(ioc, cfq_exit_single_io_context);
2686 static struct cfq_io_context *
2687 cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
2689 struct cfq_io_context *cic;
2691 cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO,
2694 cic->last_end_request = jiffies;
2695 INIT_LIST_HEAD(&cic->queue_list);
2696 INIT_HLIST_NODE(&cic->cic_list);
2697 cic->dtor = cfq_free_io_context;
2698 cic->exit = cfq_exit_io_context;
2699 elv_ioc_count_inc(cfq_ioc_count);
2705 static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc)
2707 struct task_struct *tsk = current;
2710 if (!cfq_cfqq_prio_changed(cfqq))
2713 ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio);
2714 switch (ioprio_class) {
2716 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
2717 case IOPRIO_CLASS_NONE:
2719 * no prio set, inherit CPU scheduling settings
2721 cfqq->ioprio = task_nice_ioprio(tsk);
2722 cfqq->ioprio_class = task_nice_ioclass(tsk);
2724 case IOPRIO_CLASS_RT:
2725 cfqq->ioprio = task_ioprio(ioc);
2726 cfqq->ioprio_class = IOPRIO_CLASS_RT;
2728 case IOPRIO_CLASS_BE:
2729 cfqq->ioprio = task_ioprio(ioc);
2730 cfqq->ioprio_class = IOPRIO_CLASS_BE;
2732 case IOPRIO_CLASS_IDLE:
2733 cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
2735 cfq_clear_cfqq_idle_window(cfqq);
2740 * keep track of original prio settings in case we have to temporarily
2741 * elevate the priority of this queue
2743 cfqq->org_ioprio = cfqq->ioprio;
2744 cfqq->org_ioprio_class = cfqq->ioprio_class;
2745 cfq_clear_cfqq_prio_changed(cfqq);
2748 static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic)
2750 struct cfq_data *cfqd = cic_to_cfqd(cic);
2751 struct cfq_queue *cfqq;
2752 unsigned long flags;
2754 if (unlikely(!cfqd))
2757 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2759 cfqq = cic->cfqq[BLK_RW_ASYNC];
2761 struct cfq_queue *new_cfqq;
2762 new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc,
2765 cic->cfqq[BLK_RW_ASYNC] = new_cfqq;
2766 cfq_put_queue(cfqq);
2770 cfqq = cic->cfqq[BLK_RW_SYNC];
2772 cfq_mark_cfqq_prio_changed(cfqq);
2774 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2777 static void cfq_ioc_set_ioprio(struct io_context *ioc)
2779 call_for_each_cic(ioc, changed_ioprio);
2780 ioc->ioprio_changed = 0;
2783 static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2784 pid_t pid, bool is_sync)
2786 RB_CLEAR_NODE(&cfqq->rb_node);
2787 RB_CLEAR_NODE(&cfqq->p_node);
2788 INIT_LIST_HEAD(&cfqq->fifo);
2790 atomic_set(&cfqq->ref, 0);
2793 cfq_mark_cfqq_prio_changed(cfqq);
2796 if (!cfq_class_idle(cfqq))
2797 cfq_mark_cfqq_idle_window(cfqq);
2798 cfq_mark_cfqq_sync(cfqq);
2803 #ifdef CONFIG_CFQ_GROUP_IOSCHED
2804 static void changed_cgroup(struct io_context *ioc, struct cfq_io_context *cic)
2806 struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1);
2807 struct cfq_data *cfqd = cic_to_cfqd(cic);
2808 unsigned long flags;
2809 struct request_queue *q;
2811 if (unlikely(!cfqd))
2816 spin_lock_irqsave(q->queue_lock, flags);
2820 * Drop reference to sync queue. A new sync queue will be
2821 * assigned in new group upon arrival of a fresh request.
2823 cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup");
2824 cic_set_cfqq(cic, NULL, 1);
2825 cfq_put_queue(sync_cfqq);
2828 spin_unlock_irqrestore(q->queue_lock, flags);
2831 static void cfq_ioc_set_cgroup(struct io_context *ioc)
2833 call_for_each_cic(ioc, changed_cgroup);
2834 ioc->cgroup_changed = 0;
2836 #endif /* CONFIG_CFQ_GROUP_IOSCHED */
2838 static struct cfq_queue *
2839 cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync,
2840 struct io_context *ioc, gfp_t gfp_mask)
2842 struct cfq_queue *cfqq, *new_cfqq = NULL;
2843 struct cfq_io_context *cic;
2844 struct cfq_group *cfqg;
2847 cfqg = cfq_get_cfqg(cfqd, 1);
2848 cic = cfq_cic_lookup(cfqd, ioc);
2849 /* cic always exists here */
2850 cfqq = cic_to_cfqq(cic, is_sync);
2853 * Always try a new alloc if we fell back to the OOM cfqq
2854 * originally, since it should just be a temporary situation.
2856 if (!cfqq || cfqq == &cfqd->oom_cfqq) {
2861 } else if (gfp_mask & __GFP_WAIT) {
2862 spin_unlock_irq(cfqd->queue->queue_lock);
2863 new_cfqq = kmem_cache_alloc_node(cfq_pool,
2864 gfp_mask | __GFP_ZERO,
2866 spin_lock_irq(cfqd->queue->queue_lock);
2870 cfqq = kmem_cache_alloc_node(cfq_pool,
2871 gfp_mask | __GFP_ZERO,
2876 cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync);
2877 cfq_init_prio_data(cfqq, ioc);
2878 cfq_link_cfqq_cfqg(cfqq, cfqg);
2879 cfq_log_cfqq(cfqd, cfqq, "alloced");
2881 cfqq = &cfqd->oom_cfqq;
2885 kmem_cache_free(cfq_pool, new_cfqq);
2890 static struct cfq_queue **
2891 cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio)
2893 switch (ioprio_class) {
2894 case IOPRIO_CLASS_RT:
2895 return &cfqd->async_cfqq[0][ioprio];
2896 case IOPRIO_CLASS_BE:
2897 return &cfqd->async_cfqq[1][ioprio];
2898 case IOPRIO_CLASS_IDLE:
2899 return &cfqd->async_idle_cfqq;
2905 static struct cfq_queue *
2906 cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc,
2909 const int ioprio = task_ioprio(ioc);
2910 const int ioprio_class = task_ioprio_class(ioc);
2911 struct cfq_queue **async_cfqq = NULL;
2912 struct cfq_queue *cfqq = NULL;
2915 async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio);
2920 cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask);
2923 * pin the queue now that it's allocated, scheduler exit will prune it
2925 if (!is_sync && !(*async_cfqq)) {
2926 atomic_inc(&cfqq->ref);
2930 atomic_inc(&cfqq->ref);
2935 * We drop cfq io contexts lazily, so we may find a dead one.
2938 cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc,
2939 struct cfq_io_context *cic)
2941 unsigned long flags;
2943 WARN_ON(!list_empty(&cic->queue_list));
2944 BUG_ON(cic->key != cfqd_dead_key(cfqd));
2946 spin_lock_irqsave(&ioc->lock, flags);
2948 BUG_ON(ioc->ioc_data == cic);
2950 radix_tree_delete(&ioc->radix_root, cfqd->cic_index);
2951 hlist_del_rcu(&cic->cic_list);
2952 spin_unlock_irqrestore(&ioc->lock, flags);
2957 static struct cfq_io_context *
2958 cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc)
2960 struct cfq_io_context *cic;
2961 unsigned long flags;
2969 * we maintain a last-hit cache, to avoid browsing over the tree
2971 cic = rcu_dereference(ioc->ioc_data);
2972 if (cic && cic->key == cfqd) {
2978 cic = radix_tree_lookup(&ioc->radix_root, cfqd->cic_index);
2982 if (unlikely(cic->key != cfqd)) {
2983 cfq_drop_dead_cic(cfqd, ioc, cic);
2988 spin_lock_irqsave(&ioc->lock, flags);
2989 rcu_assign_pointer(ioc->ioc_data, cic);
2990 spin_unlock_irqrestore(&ioc->lock, flags);
2998 * Add cic into ioc, using cfqd as the search key. This enables us to lookup
2999 * the process specific cfq io context when entered from the block layer.
3000 * Also adds the cic to a per-cfqd list, used when this queue is removed.
3002 static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
3003 struct cfq_io_context *cic, gfp_t gfp_mask)
3005 unsigned long flags;
3008 ret = radix_tree_preload(gfp_mask);
3013 spin_lock_irqsave(&ioc->lock, flags);
3014 ret = radix_tree_insert(&ioc->radix_root,
3015 cfqd->cic_index, cic);
3017 hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list);
3018 spin_unlock_irqrestore(&ioc->lock, flags);
3020 radix_tree_preload_end();
3023 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
3024 list_add(&cic->queue_list, &cfqd->cic_list);
3025 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
3030 printk(KERN_ERR "cfq: cic link failed!\n");
3036 * Setup general io context and cfq io context. There can be several cfq
3037 * io contexts per general io context, if this process is doing io to more
3038 * than one device managed by cfq.
3040 static struct cfq_io_context *
3041 cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
3043 struct io_context *ioc = NULL;
3044 struct cfq_io_context *cic;
3046 might_sleep_if(gfp_mask & __GFP_WAIT);
3048 ioc = get_io_context(gfp_mask, cfqd->queue->node);
3052 cic = cfq_cic_lookup(cfqd, ioc);
3056 cic = cfq_alloc_io_context(cfqd, gfp_mask);
3060 if (cfq_cic_link(cfqd, ioc, cic, gfp_mask))
3064 smp_read_barrier_depends();
3065 if (unlikely(ioc->ioprio_changed))
3066 cfq_ioc_set_ioprio(ioc);
3068 #ifdef CONFIG_CFQ_GROUP_IOSCHED
3069 if (unlikely(ioc->cgroup_changed))
3070 cfq_ioc_set_cgroup(ioc);
3076 put_io_context(ioc);
3081 cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
3083 unsigned long elapsed = jiffies - cic->last_end_request;
3084 unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle);
3086 cic->ttime_samples = (7*cic->ttime_samples + 256) / 8;
3087 cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8;
3088 cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples;
3092 cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
3096 sector_t n_sec = blk_rq_sectors(rq);
3097 if (cfqq->last_request_pos) {
3098 if (cfqq->last_request_pos < blk_rq_pos(rq))
3099 sdist = blk_rq_pos(rq) - cfqq->last_request_pos;
3101 sdist = cfqq->last_request_pos - blk_rq_pos(rq);
3104 cfqq->seek_history <<= 1;
3105 if (blk_queue_nonrot(cfqd->queue))
3106 cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT);
3108 cfqq->seek_history |= (sdist > CFQQ_SEEK_THR);
3112 * Disable idle window if the process thinks too long or seeks so much that
3116 cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
3117 struct cfq_io_context *cic)
3119 int old_idle, enable_idle;
3122 * Don't idle for async or idle io prio class
3124 if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
3127 enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
3129 if (cfqq->queued[0] + cfqq->queued[1] >= 4)
3130 cfq_mark_cfqq_deep(cfqq);
3132 if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle ||
3133 (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq)))
3135 else if (sample_valid(cic->ttime_samples)) {
3136 if (cic->ttime_mean > cfqd->cfq_slice_idle)
3142 if (old_idle != enable_idle) {
3143 cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle);
3145 cfq_mark_cfqq_idle_window(cfqq);
3147 cfq_clear_cfqq_idle_window(cfqq);
3152 * Check if new_cfqq should preempt the currently active queue. Return 0 for
3153 * no or if we aren't sure, a 1 will cause a preempt.
3156 cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
3159 struct cfq_queue *cfqq;
3161 cfqq = cfqd->active_queue;
3165 if (cfq_class_idle(new_cfqq))
3168 if (cfq_class_idle(cfqq))
3172 * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice.
3174 if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq))
3178 * if the new request is sync, but the currently running queue is
3179 * not, let the sync request have priority.
3181 if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq))
3184 if (new_cfqq->cfqg != cfqq->cfqg)
3187 if (cfq_slice_used(cfqq))
3190 /* Allow preemption only if we are idling on sync-noidle tree */
3191 if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD &&
3192 cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD &&
3193 new_cfqq->service_tree->count == 2 &&
3194 RB_EMPTY_ROOT(&cfqq->sort_list))
3198 * So both queues are sync. Let the new request get disk time if
3199 * it's a metadata request and the current queue is doing regular IO.
3201 if ((rq->cmd_flags & REQ_META) && !cfqq->meta_pending)
3205 * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice.
3207 if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq))
3210 if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq))
3214 * if this request is as-good as one we would expect from the
3215 * current cfqq, let it preempt
3217 if (cfq_rq_close(cfqd, cfqq, rq))
3224 * cfqq preempts the active queue. if we allowed preempt with no slice left,
3225 * let it have half of its nominal slice.
3227 static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3229 cfq_log_cfqq(cfqd, cfqq, "preempt");
3230 cfq_slice_expired(cfqd, 1);
3233 * Put the new queue at the front of the of the current list,
3234 * so we know that it will be selected next.
3236 BUG_ON(!cfq_cfqq_on_rr(cfqq));
3238 cfq_service_tree_add(cfqd, cfqq, 1);
3240 cfqq->slice_end = 0;
3241 cfq_mark_cfqq_slice_new(cfqq);
3245 * Called when a new fs request (rq) is added (to cfqq). Check if there's
3246 * something we should do about it
3249 cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
3252 struct cfq_io_context *cic = RQ_CIC(rq);
3255 if (rq->cmd_flags & REQ_META)
3256 cfqq->meta_pending++;
3258 cfq_update_io_thinktime(cfqd, cic);
3259 cfq_update_io_seektime(cfqd, cfqq, rq);
3260 cfq_update_idle_window(cfqd, cfqq, cic);
3262 cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
3264 if (cfqq == cfqd->active_queue) {
3266 * Remember that we saw a request from this process, but
3267 * don't start queuing just yet. Otherwise we risk seeing lots
3268 * of tiny requests, because we disrupt the normal plugging
3269 * and merging. If the request is already larger than a single
3270 * page, let it rip immediately. For that case we assume that
3271 * merging is already done. Ditto for a busy system that
3272 * has other work pending, don't risk delaying until the
3273 * idle timer unplug to continue working.
3275 if (cfq_cfqq_wait_request(cfqq)) {
3276 if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE ||
3277 cfqd->busy_queues > 1) {
3278 cfq_del_timer(cfqd, cfqq);
3279 cfq_clear_cfqq_wait_request(cfqq);
3280 __blk_run_queue(cfqd->queue);
3282 cfq_blkiocg_update_idle_time_stats(
3284 cfq_mark_cfqq_must_dispatch(cfqq);
3287 } else if (cfq_should_preempt(cfqd, cfqq, rq)) {
3289 * not the active queue - expire current slice if it is
3290 * idle and has expired it's mean thinktime or this new queue
3291 * has some old slice time left and is of higher priority or
3292 * this new queue is RT and the current one is BE
3294 cfq_preempt_queue(cfqd, cfqq);
3295 __blk_run_queue(cfqd->queue);
3299 static void cfq_insert_request(struct request_queue *q, struct request *rq)
3301 struct cfq_data *cfqd = q->elevator->elevator_data;
3302 struct cfq_queue *cfqq = RQ_CFQQ(rq);
3304 cfq_log_cfqq(cfqd, cfqq, "insert_request");
3305 cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc);
3307 rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]);
3308 list_add_tail(&rq->queuelist, &cfqq->fifo);
3310 cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg,
3311 &cfqd->serving_group->blkg, rq_data_dir(rq),
3313 cfq_rq_enqueued(cfqd, cfqq, rq);
3317 * Update hw_tag based on peak queue depth over 50 samples under
3320 static void cfq_update_hw_tag(struct cfq_data *cfqd)
3322 struct cfq_queue *cfqq = cfqd->active_queue;
3324 if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth)
3325 cfqd->hw_tag_est_depth = cfqd->rq_in_driver;
3327 if (cfqd->hw_tag == 1)
3330 if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN &&
3331 cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN)
3335 * If active queue hasn't enough requests and can idle, cfq might not
3336 * dispatch sufficient requests to hardware. Don't zero hw_tag in this
3339 if (cfqq && cfq_cfqq_idle_window(cfqq) &&
3340 cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] <
3341 CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN)
3344 if (cfqd->hw_tag_samples++ < 50)
3347 if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN)
3353 static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3355 struct cfq_io_context *cic = cfqd->active_cic;
3357 /* If there are other queues in the group, don't wait */
3358 if (cfqq->cfqg->nr_cfqq > 1)
3361 if (cfq_slice_used(cfqq))
3364 /* if slice left is less than think time, wait busy */
3365 if (cic && sample_valid(cic->ttime_samples)
3366 && (cfqq->slice_end - jiffies < cic->ttime_mean))
3370 * If think times is less than a jiffy than ttime_mean=0 and above
3371 * will not be true. It might happen that slice has not expired yet
3372 * but will expire soon (4-5 ns) during select_queue(). To cover the
3373 * case where think time is less than a jiffy, mark the queue wait
3374 * busy if only 1 jiffy is left in the slice.
3376 if (cfqq->slice_end - jiffies == 1)
3382 static void cfq_completed_request(struct request_queue *q, struct request *rq)
3384 struct cfq_queue *cfqq = RQ_CFQQ(rq);
3385 struct cfq_data *cfqd = cfqq->cfqd;
3386 const int sync = rq_is_sync(rq);
3390 cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d",
3391 !!(rq->cmd_flags & REQ_NOIDLE));
3393 cfq_update_hw_tag(cfqd);
3395 WARN_ON(!cfqd->rq_in_driver);
3396 WARN_ON(!cfqq->dispatched);
3397 cfqd->rq_in_driver--;
3399 cfq_blkiocg_update_completion_stats(&cfqq->cfqg->blkg,
3400 rq_start_time_ns(rq), rq_io_start_time_ns(rq),
3401 rq_data_dir(rq), rq_is_sync(rq));
3403 cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--;
3406 RQ_CIC(rq)->last_end_request = now;
3407 if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now))
3408 cfqd->last_delayed_sync = now;
3412 * If this is the active queue, check if it needs to be expired,
3413 * or if we want to idle in case it has no pending requests.
3415 if (cfqd->active_queue == cfqq) {
3416 const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list);
3418 if (cfq_cfqq_slice_new(cfqq)) {
3419 cfq_set_prio_slice(cfqd, cfqq);
3420 cfq_clear_cfqq_slice_new(cfqq);
3424 * Should we wait for next request to come in before we expire
3427 if (cfq_should_wait_busy(cfqd, cfqq)) {
3428 cfqq->slice_end = jiffies + cfqd->cfq_slice_idle;
3429 cfq_mark_cfqq_wait_busy(cfqq);
3430 cfq_log_cfqq(cfqd, cfqq, "will busy wait");
3434 * Idling is not enabled on:
3436 * - idle-priority queues
3438 * - queues with still some requests queued
3439 * - when there is a close cooperator
3441 if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
3442 cfq_slice_expired(cfqd, 1);
3443 else if (sync && cfqq_empty &&
3444 !cfq_close_cooperator(cfqd, cfqq)) {
3445 cfqd->noidle_tree_requires_idle |=
3446 !(rq->cmd_flags & REQ_NOIDLE);
3448 * Idling is enabled for SYNC_WORKLOAD.
3449 * SYNC_NOIDLE_WORKLOAD idles at the end of the tree
3450 * only if we processed at least one !REQ_NOIDLE request
3452 if (cfqd->serving_type == SYNC_WORKLOAD
3453 || cfqd->noidle_tree_requires_idle
3454 || cfqq->cfqg->nr_cfqq == 1)
3455 cfq_arm_slice_timer(cfqd);
3459 if (!cfqd->rq_in_driver)
3460 cfq_schedule_dispatch(cfqd);
3464 * we temporarily boost lower priority queues if they are holding fs exclusive
3465 * resources. they are boosted to normal prio (CLASS_BE/4)
3467 static void cfq_prio_boost(struct cfq_queue *cfqq)
3469 if (has_fs_excl()) {
3471 * boost idle prio on transactions that would lock out other
3472 * users of the filesystem
3474 if (cfq_class_idle(cfqq))
3475 cfqq->ioprio_class = IOPRIO_CLASS_BE;
3476 if (cfqq->ioprio > IOPRIO_NORM)
3477 cfqq->ioprio = IOPRIO_NORM;
3480 * unboost the queue (if needed)
3482 cfqq->ioprio_class = cfqq->org_ioprio_class;
3483 cfqq->ioprio = cfqq->org_ioprio;
3487 static inline int __cfq_may_queue(struct cfq_queue *cfqq)
3489 if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) {
3490 cfq_mark_cfqq_must_alloc_slice(cfqq);
3491 return ELV_MQUEUE_MUST;
3494 return ELV_MQUEUE_MAY;
3497 static int cfq_may_queue(struct request_queue *q, int rw)
3499 struct cfq_data *cfqd = q->elevator->elevator_data;
3500 struct task_struct *tsk = current;
3501 struct cfq_io_context *cic;
3502 struct cfq_queue *cfqq;
3505 * don't force setup of a queue from here, as a call to may_queue
3506 * does not necessarily imply that a request actually will be queued.
3507 * so just lookup a possibly existing queue, or return 'may queue'
3510 cic = cfq_cic_lookup(cfqd, tsk->io_context);
3512 return ELV_MQUEUE_MAY;
3514 cfqq = cic_to_cfqq(cic, rw_is_sync(rw));
3516 cfq_init_prio_data(cfqq, cic->ioc);
3517 cfq_prio_boost(cfqq);
3519 return __cfq_may_queue(cfqq);
3522 return ELV_MQUEUE_MAY;
3526 * queue lock held here
3528 static void cfq_put_request(struct request *rq)
3530 struct cfq_queue *cfqq = RQ_CFQQ(rq);
3533 const int rw = rq_data_dir(rq);
3535 BUG_ON(!cfqq->allocated[rw]);
3536 cfqq->allocated[rw]--;
3538 put_io_context(RQ_CIC(rq)->ioc);
3540 rq->elevator_private = NULL;
3541 rq->elevator_private2 = NULL;
3543 /* Put down rq reference on cfqg */
3544 cfq_put_cfqg(RQ_CFQG(rq));
3545 rq->elevator_private3 = NULL;
3547 cfq_put_queue(cfqq);
3551 static struct cfq_queue *
3552 cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_context *cic,
3553 struct cfq_queue *cfqq)
3555 cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq);
3556 cic_set_cfqq(cic, cfqq->new_cfqq, 1);
3557 cfq_mark_cfqq_coop(cfqq->new_cfqq);
3558 cfq_put_queue(cfqq);
3559 return cic_to_cfqq(cic, 1);
3563 * Returns NULL if a new cfqq should be allocated, or the old cfqq if this
3564 * was the last process referring to said cfqq.
3566 static struct cfq_queue *
3567 split_cfqq(struct cfq_io_context *cic, struct cfq_queue *cfqq)
3569 if (cfqq_process_refs(cfqq) == 1) {
3570 cfqq->pid = current->pid;
3571 cfq_clear_cfqq_coop(cfqq);
3572 cfq_clear_cfqq_split_coop(cfqq);
3576 cic_set_cfqq(cic, NULL, 1);
3578 cfq_put_cooperator(cfqq);
3580 cfq_put_queue(cfqq);
3584 * Allocate cfq data structures associated with this request.
3587 cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
3589 struct cfq_data *cfqd = q->elevator->elevator_data;
3590 struct cfq_io_context *cic;
3591 const int rw = rq_data_dir(rq);
3592 const bool is_sync = rq_is_sync(rq);
3593 struct cfq_queue *cfqq;
3594 unsigned long flags;
3596 might_sleep_if(gfp_mask & __GFP_WAIT);
3598 cic = cfq_get_io_context(cfqd, gfp_mask);
3600 spin_lock_irqsave(q->queue_lock, flags);
3606 cfqq = cic_to_cfqq(cic, is_sync);
3607 if (!cfqq || cfqq == &cfqd->oom_cfqq) {
3608 cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask);
3609 cic_set_cfqq(cic, cfqq, is_sync);
3612 * If the queue was seeky for too long, break it apart.
3614 if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) {
3615 cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq");
3616 cfqq = split_cfqq(cic, cfqq);
3622 * Check to see if this queue is scheduled to merge with
3623 * another, closely cooperating queue. The merging of
3624 * queues happens here as it must be done in process context.
3625 * The reference on new_cfqq was taken in merge_cfqqs.
3628 cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq);
3631 cfqq->allocated[rw]++;
3632 atomic_inc(&cfqq->ref);
3634 spin_unlock_irqrestore(q->queue_lock, flags);
3636 rq->elevator_private = cic;
3637 rq->elevator_private2 = cfqq;
3638 rq->elevator_private3 = cfq_ref_get_cfqg(cfqq->cfqg);
3643 put_io_context(cic->ioc);
3645 cfq_schedule_dispatch(cfqd);
3646 spin_unlock_irqrestore(q->queue_lock, flags);
3647 cfq_log(cfqd, "set_request fail");
3651 static void cfq_kick_queue(struct work_struct *work)
3653 struct cfq_data *cfqd =
3654 container_of(work, struct cfq_data, unplug_work);
3655 struct request_queue *q = cfqd->queue;
3657 spin_lock_irq(q->queue_lock);
3658 __blk_run_queue(cfqd->queue);
3659 spin_unlock_irq(q->queue_lock);
3663 * Timer running if the active_queue is currently idling inside its time slice
3665 static void cfq_idle_slice_timer(unsigned long data)
3667 struct cfq_data *cfqd = (struct cfq_data *) data;
3668 struct cfq_queue *cfqq;
3669 unsigned long flags;
3672 cfq_log(cfqd, "idle timer fired");
3674 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
3676 cfqq = cfqd->active_queue;
3681 * We saw a request before the queue expired, let it through
3683 if (cfq_cfqq_must_dispatch(cfqq))
3689 if (cfq_slice_used(cfqq))
3693 * only expire and reinvoke request handler, if there are
3694 * other queues with pending requests
3696 if (!cfqd->busy_queues)
3700 * not expired and it has a request pending, let it dispatch
3702 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
3706 * Queue depth flag is reset only when the idle didn't succeed
3708 cfq_clear_cfqq_deep(cfqq);
3711 cfq_slice_expired(cfqd, timed_out);
3713 cfq_schedule_dispatch(cfqd);
3715 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
3718 static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
3720 del_timer_sync(&cfqd->idle_slice_timer);
3721 cancel_work_sync(&cfqd->unplug_work);
3724 static void cfq_put_async_queues(struct cfq_data *cfqd)
3728 for (i = 0; i < IOPRIO_BE_NR; i++) {
3729 if (cfqd->async_cfqq[0][i])
3730 cfq_put_queue(cfqd->async_cfqq[0][i]);
3731 if (cfqd->async_cfqq[1][i])
3732 cfq_put_queue(cfqd->async_cfqq[1][i]);
3735 if (cfqd->async_idle_cfqq)
3736 cfq_put_queue(cfqd->async_idle_cfqq);
3739 static void cfq_cfqd_free(struct rcu_head *head)
3741 kfree(container_of(head, struct cfq_data, rcu));
3744 static void cfq_exit_queue(struct elevator_queue *e)
3746 struct cfq_data *cfqd = e->elevator_data;
3747 struct request_queue *q = cfqd->queue;
3749 cfq_shutdown_timer_wq(cfqd);
3751 spin_lock_irq(q->queue_lock);
3753 if (cfqd->active_queue)
3754 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
3756 while (!list_empty(&cfqd->cic_list)) {
3757 struct cfq_io_context *cic = list_entry(cfqd->cic_list.next,
3758 struct cfq_io_context,
3761 __cfq_exit_single_io_context(cfqd, cic);
3764 cfq_put_async_queues(cfqd);
3765 cfq_release_cfq_groups(cfqd);
3766 cfq_blkiocg_del_blkio_group(&cfqd->root_group.blkg);
3768 spin_unlock_irq(q->queue_lock);
3770 cfq_shutdown_timer_wq(cfqd);
3772 spin_lock(&cic_index_lock);
3773 ida_remove(&cic_index_ida, cfqd->cic_index);
3774 spin_unlock(&cic_index_lock);
3776 /* Wait for cfqg->blkg->key accessors to exit their grace periods. */
3777 call_rcu(&cfqd->rcu, cfq_cfqd_free);
3780 static int cfq_alloc_cic_index(void)
3785 if (!ida_pre_get(&cic_index_ida, GFP_KERNEL))
3788 spin_lock(&cic_index_lock);
3789 error = ida_get_new(&cic_index_ida, &index);
3790 spin_unlock(&cic_index_lock);
3791 if (error && error != -EAGAIN)
3798 static void *cfq_init_queue(struct request_queue *q)
3800 struct cfq_data *cfqd;
3802 struct cfq_group *cfqg;
3803 struct cfq_rb_root *st;
3805 i = cfq_alloc_cic_index();
3809 cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node);
3813 cfqd->cic_index = i;
3815 /* Init root service tree */
3816 cfqd->grp_service_tree = CFQ_RB_ROOT;
3818 /* Init root group */
3819 cfqg = &cfqd->root_group;
3820 for_each_cfqg_st(cfqg, i, j, st)
3822 RB_CLEAR_NODE(&cfqg->rb_node);
3824 /* Give preference to root group over other groups */
3825 cfqg->weight = 2*BLKIO_WEIGHT_DEFAULT;
3827 #ifdef CONFIG_CFQ_GROUP_IOSCHED
3829 * Take a reference to root group which we never drop. This is just
3830 * to make sure that cfq_put_cfqg() does not try to kfree root group
3832 atomic_set(&cfqg->ref, 1);
3834 cfq_blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg,
3839 * Not strictly needed (since RB_ROOT just clears the node and we
3840 * zeroed cfqd on alloc), but better be safe in case someone decides
3841 * to add magic to the rb code
3843 for (i = 0; i < CFQ_PRIO_LISTS; i++)
3844 cfqd->prio_trees[i] = RB_ROOT;
3847 * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues.
3848 * Grab a permanent reference to it, so that the normal code flow
3849 * will not attempt to free it.
3851 cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0);
3852 atomic_inc(&cfqd->oom_cfqq.ref);
3853 cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group);
3855 INIT_LIST_HEAD(&cfqd->cic_list);
3859 init_timer(&cfqd->idle_slice_timer);
3860 cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
3861 cfqd->idle_slice_timer.data = (unsigned long) cfqd;
3863 INIT_WORK(&cfqd->unplug_work, cfq_kick_queue);
3865 cfqd->cfq_quantum = cfq_quantum;
3866 cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
3867 cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
3868 cfqd->cfq_back_max = cfq_back_max;
3869 cfqd->cfq_back_penalty = cfq_back_penalty;
3870 cfqd->cfq_slice[0] = cfq_slice_async;
3871 cfqd->cfq_slice[1] = cfq_slice_sync;
3872 cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
3873 cfqd->cfq_slice_idle = cfq_slice_idle;
3874 cfqd->cfq_latency = 1;
3875 cfqd->cfq_group_isolation = 0;
3878 * we optimistically start assuming sync ops weren't delayed in last
3879 * second, in order to have larger depth for async operations.
3881 cfqd->last_delayed_sync = jiffies - HZ;
3885 static void cfq_slab_kill(void)
3888 * Caller already ensured that pending RCU callbacks are completed,
3889 * so we should have no busy allocations at this point.
3892 kmem_cache_destroy(cfq_pool);
3894 kmem_cache_destroy(cfq_ioc_pool);
3897 static int __init cfq_slab_setup(void)
3899 cfq_pool = KMEM_CACHE(cfq_queue, 0);
3903 cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0);
3914 * sysfs parts below -->
3917 cfq_var_show(unsigned int var, char *page)
3919 return sprintf(page, "%d\n", var);
3923 cfq_var_store(unsigned int *var, const char *page, size_t count)
3925 char *p = (char *) page;
3927 *var = simple_strtoul(p, &p, 10);
3931 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
3932 static ssize_t __FUNC(struct elevator_queue *e, char *page) \
3934 struct cfq_data *cfqd = e->elevator_data; \
3935 unsigned int __data = __VAR; \
3937 __data = jiffies_to_msecs(__data); \
3938 return cfq_var_show(__data, (page)); \
3940 SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
3941 SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
3942 SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
3943 SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
3944 SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
3945 SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
3946 SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
3947 SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
3948 SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
3949 SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0);
3950 SHOW_FUNCTION(cfq_group_isolation_show, cfqd->cfq_group_isolation, 0);
3951 #undef SHOW_FUNCTION
3953 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
3954 static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
3956 struct cfq_data *cfqd = e->elevator_data; \
3957 unsigned int __data; \
3958 int ret = cfq_var_store(&__data, (page), count); \
3959 if (__data < (MIN)) \
3961 else if (__data > (MAX)) \
3964 *(__PTR) = msecs_to_jiffies(__data); \
3966 *(__PTR) = __data; \
3969 STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
3970 STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1,
3972 STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1,
3974 STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
3975 STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1,
3977 STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
3978 STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
3979 STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
3980 STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1,
3982 STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0);
3983 STORE_FUNCTION(cfq_group_isolation_store, &cfqd->cfq_group_isolation, 0, 1, 0);
3984 #undef STORE_FUNCTION
3986 #define CFQ_ATTR(name) \
3987 __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
3989 static struct elv_fs_entry cfq_attrs[] = {
3991 CFQ_ATTR(fifo_expire_sync),
3992 CFQ_ATTR(fifo_expire_async),
3993 CFQ_ATTR(back_seek_max),
3994 CFQ_ATTR(back_seek_penalty),
3995 CFQ_ATTR(slice_sync),
3996 CFQ_ATTR(slice_async),
3997 CFQ_ATTR(slice_async_rq),
3998 CFQ_ATTR(slice_idle),
3999 CFQ_ATTR(low_latency),
4000 CFQ_ATTR(group_isolation),
4004 static struct elevator_type iosched_cfq = {
4006 .elevator_merge_fn = cfq_merge,
4007 .elevator_merged_fn = cfq_merged_request,
4008 .elevator_merge_req_fn = cfq_merged_requests,
4009 .elevator_allow_merge_fn = cfq_allow_merge,
4010 .elevator_bio_merged_fn = cfq_bio_merged,
4011 .elevator_dispatch_fn = cfq_dispatch_requests,
4012 .elevator_add_req_fn = cfq_insert_request,
4013 .elevator_activate_req_fn = cfq_activate_request,
4014 .elevator_deactivate_req_fn = cfq_deactivate_request,
4015 .elevator_queue_empty_fn = cfq_queue_empty,
4016 .elevator_completed_req_fn = cfq_completed_request,
4017 .elevator_former_req_fn = elv_rb_former_request,
4018 .elevator_latter_req_fn = elv_rb_latter_request,
4019 .elevator_set_req_fn = cfq_set_request,
4020 .elevator_put_req_fn = cfq_put_request,
4021 .elevator_may_queue_fn = cfq_may_queue,
4022 .elevator_init_fn = cfq_init_queue,
4023 .elevator_exit_fn = cfq_exit_queue,
4024 .trim = cfq_free_io_context,
4026 .elevator_attrs = cfq_attrs,
4027 .elevator_name = "cfq",
4028 .elevator_owner = THIS_MODULE,
4031 #ifdef CONFIG_CFQ_GROUP_IOSCHED
4032 static struct blkio_policy_type blkio_policy_cfq = {
4034 .blkio_unlink_group_fn = cfq_unlink_blkio_group,
4035 .blkio_update_group_weight_fn = cfq_update_blkio_group_weight,
4039 static struct blkio_policy_type blkio_policy_cfq;
4042 static int __init cfq_init(void)
4045 * could be 0 on HZ < 1000 setups
4047 if (!cfq_slice_async)
4048 cfq_slice_async = 1;
4049 if (!cfq_slice_idle)
4052 if (cfq_slab_setup())
4055 elv_register(&iosched_cfq);
4056 blkio_policy_register(&blkio_policy_cfq);
4061 static void __exit cfq_exit(void)
4063 DECLARE_COMPLETION_ONSTACK(all_gone);
4064 blkio_policy_unregister(&blkio_policy_cfq);
4065 elv_unregister(&iosched_cfq);
4066 ioc_gone = &all_gone;
4067 /* ioc_gone's update must be visible before reading ioc_count */
4071 * this also protects us from entering cfq_slab_kill() with
4072 * pending RCU callbacks
4074 if (elv_ioc_count_read(cfq_ioc_count))
4075 wait_for_completion(&all_gone);
4076 ida_destroy(&cic_index_ida);
4080 module_init(cfq_init);
4081 module_exit(cfq_exit);
4083 MODULE_AUTHOR("Jens Axboe");
4084 MODULE_LICENSE("GPL");
4085 MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");