2 * Interface for controlling IO bandwidth on a request queue
4 * Copyright (C) 2010 Vivek Goyal <vgoyal@redhat.com>
7 #include <linux/module.h>
8 #include <linux/slab.h>
9 #include <linux/blkdev.h>
10 #include <linux/bio.h>
11 #include <linux/blktrace_api.h>
12 #include "blk-cgroup.h"
14 /* Max dispatch from a group in 1 round */
15 static int throtl_grp_quantum = 8;
17 /* Total max dispatch from all groups in one round */
18 static int throtl_quantum = 32;
20 /* Throttling is performed over 100ms slice and after that slice is renewed */
21 static unsigned long throtl_slice = HZ/10; /* 100 ms */
23 /* A workqueue to queue throttle related work */
24 static struct workqueue_struct *kthrotld_workqueue;
25 static void throtl_schedule_delayed_work(struct throtl_data *td,
28 struct throtl_rb_root {
32 unsigned long min_disptime;
35 #define THROTL_RB_ROOT (struct throtl_rb_root) { .rb = RB_ROOT, .left = NULL, \
36 .count = 0, .min_disptime = 0}
38 #define rb_entry_tg(node) rb_entry((node), struct throtl_grp, rb_node)
41 /* List of throtl groups on the request queue*/
42 struct hlist_node tg_node;
44 /* active throtl group service_tree member */
45 struct rb_node rb_node;
48 * Dispatch time in jiffies. This is the estimated time when group
49 * will unthrottle and is ready to dispatch more bio. It is used as
50 * key to sort active groups in service tree.
52 unsigned long disptime;
54 struct blkio_group blkg;
58 /* Two lists for READ and WRITE */
59 struct bio_list bio_lists[2];
61 /* Number of queued bios on READ and WRITE lists */
62 unsigned int nr_queued[2];
64 /* bytes per second rate limits */
70 /* Number of bytes disptached in current slice */
71 uint64_t bytes_disp[2];
72 /* Number of bio's dispatched in current slice */
73 unsigned int io_disp[2];
75 /* When did we start a new slice */
76 unsigned long slice_start[2];
77 unsigned long slice_end[2];
79 /* Some throttle limits got updated for the group */
85 /* List of throtl groups */
86 struct hlist_head tg_list;
88 /* service tree for active throtl groups */
89 struct throtl_rb_root tg_service_tree;
91 struct throtl_grp root_tg;
92 struct request_queue *queue;
94 /* Total Number of queued bios on READ and WRITE lists */
95 unsigned int nr_queued[2];
98 * number of total undestroyed groups
100 unsigned int nr_undestroyed_grps;
102 /* Work for dispatching throttled bios */
103 struct delayed_work throtl_work;
108 enum tg_state_flags {
109 THROTL_TG_FLAG_on_rr = 0, /* on round-robin busy list */
112 #define THROTL_TG_FNS(name) \
113 static inline void throtl_mark_tg_##name(struct throtl_grp *tg) \
115 (tg)->flags |= (1 << THROTL_TG_FLAG_##name); \
117 static inline void throtl_clear_tg_##name(struct throtl_grp *tg) \
119 (tg)->flags &= ~(1 << THROTL_TG_FLAG_##name); \
121 static inline int throtl_tg_##name(const struct throtl_grp *tg) \
123 return ((tg)->flags & (1 << THROTL_TG_FLAG_##name)) != 0; \
126 THROTL_TG_FNS(on_rr);
128 #define throtl_log_tg(td, tg, fmt, args...) \
129 blk_add_trace_msg((td)->queue, "throtl %s " fmt, \
130 blkg_path(&(tg)->blkg), ##args); \
132 #define throtl_log(td, fmt, args...) \
133 blk_add_trace_msg((td)->queue, "throtl " fmt, ##args)
135 static inline struct throtl_grp *tg_of_blkg(struct blkio_group *blkg)
138 return container_of(blkg, struct throtl_grp, blkg);
143 static inline int total_nr_queued(struct throtl_data *td)
145 return (td->nr_queued[0] + td->nr_queued[1]);
148 static inline struct throtl_grp *throtl_ref_get_tg(struct throtl_grp *tg)
150 atomic_inc(&tg->ref);
154 static void throtl_put_tg(struct throtl_grp *tg)
156 BUG_ON(atomic_read(&tg->ref) <= 0);
157 if (!atomic_dec_and_test(&tg->ref))
162 static void throtl_init_group(struct throtl_grp *tg)
164 INIT_HLIST_NODE(&tg->tg_node);
165 RB_CLEAR_NODE(&tg->rb_node);
166 bio_list_init(&tg->bio_lists[0]);
167 bio_list_init(&tg->bio_lists[1]);
168 tg->limits_changed = false;
170 /* Practically unlimited BW */
171 tg->bps[0] = tg->bps[1] = -1;
172 tg->iops[0] = tg->iops[1] = -1;
175 * Take the initial reference that will be released on destroy
176 * This can be thought of a joint reference by cgroup and
177 * request queue which will be dropped by either request queue
178 * exit or cgroup deletion path depending on who is exiting first.
180 atomic_set(&tg->ref, 1);
183 /* Should be called with rcu read lock held (needed for blkcg) */
185 throtl_add_group_to_td_list(struct throtl_data *td, struct throtl_grp *tg)
187 hlist_add_head(&tg->tg_node, &td->tg_list);
188 td->nr_undestroyed_grps++;
191 static struct throtl_grp * throtl_find_alloc_tg(struct throtl_data *td,
192 struct blkio_cgroup *blkcg)
194 struct throtl_grp *tg = NULL;
196 struct backing_dev_info *bdi = &td->queue->backing_dev_info;
197 unsigned int major, minor;
200 * TODO: Speed up blkiocg_lookup_group() by maintaining a radix
201 * tree of blkg (instead of traversing through hash list all
206 * This is the common case when there are no blkio cgroups.
207 * Avoid lookup in this case
209 if (blkcg == &blkio_root_cgroup)
212 tg = tg_of_blkg(blkiocg_lookup_group(blkcg, key));
214 /* Fill in device details for root group */
215 if (tg && !tg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
216 sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
217 tg->blkg.dev = MKDEV(major, minor);
224 tg = kzalloc_node(sizeof(*tg), GFP_ATOMIC, td->queue->node);
228 throtl_init_group(tg);
230 /* Add group onto cgroup list */
231 sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
232 blkiocg_add_blkio_group(blkcg, &tg->blkg, (void *)td,
233 MKDEV(major, minor), BLKIO_POLICY_THROTL);
235 tg->bps[READ] = blkcg_get_read_bps(blkcg, tg->blkg.dev);
236 tg->bps[WRITE] = blkcg_get_write_bps(blkcg, tg->blkg.dev);
237 tg->iops[READ] = blkcg_get_read_iops(blkcg, tg->blkg.dev);
238 tg->iops[WRITE] = blkcg_get_write_iops(blkcg, tg->blkg.dev);
240 throtl_add_group_to_td_list(td, tg);
245 static struct throtl_grp * throtl_get_tg(struct throtl_data *td)
247 struct throtl_grp *tg = NULL;
248 struct blkio_cgroup *blkcg;
251 blkcg = task_blkio_cgroup(current);
252 tg = throtl_find_alloc_tg(td, blkcg);
259 static struct throtl_grp *throtl_rb_first(struct throtl_rb_root *root)
261 /* Service tree is empty */
266 root->left = rb_first(&root->rb);
269 return rb_entry_tg(root->left);
274 static void rb_erase_init(struct rb_node *n, struct rb_root *root)
280 static void throtl_rb_erase(struct rb_node *n, struct throtl_rb_root *root)
284 rb_erase_init(n, &root->rb);
288 static void update_min_dispatch_time(struct throtl_rb_root *st)
290 struct throtl_grp *tg;
292 tg = throtl_rb_first(st);
296 st->min_disptime = tg->disptime;
300 tg_service_tree_add(struct throtl_rb_root *st, struct throtl_grp *tg)
302 struct rb_node **node = &st->rb.rb_node;
303 struct rb_node *parent = NULL;
304 struct throtl_grp *__tg;
305 unsigned long key = tg->disptime;
308 while (*node != NULL) {
310 __tg = rb_entry_tg(parent);
312 if (time_before(key, __tg->disptime))
313 node = &parent->rb_left;
315 node = &parent->rb_right;
321 st->left = &tg->rb_node;
323 rb_link_node(&tg->rb_node, parent, node);
324 rb_insert_color(&tg->rb_node, &st->rb);
327 static void __throtl_enqueue_tg(struct throtl_data *td, struct throtl_grp *tg)
329 struct throtl_rb_root *st = &td->tg_service_tree;
331 tg_service_tree_add(st, tg);
332 throtl_mark_tg_on_rr(tg);
336 static void throtl_enqueue_tg(struct throtl_data *td, struct throtl_grp *tg)
338 if (!throtl_tg_on_rr(tg))
339 __throtl_enqueue_tg(td, tg);
342 static void __throtl_dequeue_tg(struct throtl_data *td, struct throtl_grp *tg)
344 throtl_rb_erase(&tg->rb_node, &td->tg_service_tree);
345 throtl_clear_tg_on_rr(tg);
348 static void throtl_dequeue_tg(struct throtl_data *td, struct throtl_grp *tg)
350 if (throtl_tg_on_rr(tg))
351 __throtl_dequeue_tg(td, tg);
354 static void throtl_schedule_next_dispatch(struct throtl_data *td)
356 struct throtl_rb_root *st = &td->tg_service_tree;
359 * If there are more bios pending, schedule more work.
361 if (!total_nr_queued(td))
366 update_min_dispatch_time(st);
368 if (time_before_eq(st->min_disptime, jiffies))
369 throtl_schedule_delayed_work(td, 0);
371 throtl_schedule_delayed_work(td, (st->min_disptime - jiffies));
375 throtl_start_new_slice(struct throtl_data *td, struct throtl_grp *tg, bool rw)
377 tg->bytes_disp[rw] = 0;
379 tg->slice_start[rw] = jiffies;
380 tg->slice_end[rw] = jiffies + throtl_slice;
381 throtl_log_tg(td, tg, "[%c] new slice start=%lu end=%lu jiffies=%lu",
382 rw == READ ? 'R' : 'W', tg->slice_start[rw],
383 tg->slice_end[rw], jiffies);
386 static inline void throtl_set_slice_end(struct throtl_data *td,
387 struct throtl_grp *tg, bool rw, unsigned long jiffy_end)
389 tg->slice_end[rw] = roundup(jiffy_end, throtl_slice);
392 static inline void throtl_extend_slice(struct throtl_data *td,
393 struct throtl_grp *tg, bool rw, unsigned long jiffy_end)
395 tg->slice_end[rw] = roundup(jiffy_end, throtl_slice);
396 throtl_log_tg(td, tg, "[%c] extend slice start=%lu end=%lu jiffies=%lu",
397 rw == READ ? 'R' : 'W', tg->slice_start[rw],
398 tg->slice_end[rw], jiffies);
401 /* Determine if previously allocated or extended slice is complete or not */
403 throtl_slice_used(struct throtl_data *td, struct throtl_grp *tg, bool rw)
405 if (time_in_range(jiffies, tg->slice_start[rw], tg->slice_end[rw]))
411 /* Trim the used slices and adjust slice start accordingly */
413 throtl_trim_slice(struct throtl_data *td, struct throtl_grp *tg, bool rw)
415 unsigned long nr_slices, time_elapsed, io_trim;
418 BUG_ON(time_before(tg->slice_end[rw], tg->slice_start[rw]));
421 * If bps are unlimited (-1), then time slice don't get
422 * renewed. Don't try to trim the slice if slice is used. A new
423 * slice will start when appropriate.
425 if (throtl_slice_used(td, tg, rw))
429 * A bio has been dispatched. Also adjust slice_end. It might happen
430 * that initially cgroup limit was very low resulting in high
431 * slice_end, but later limit was bumped up and bio was dispached
432 * sooner, then we need to reduce slice_end. A high bogus slice_end
433 * is bad because it does not allow new slice to start.
436 throtl_set_slice_end(td, tg, rw, jiffies + throtl_slice);
438 time_elapsed = jiffies - tg->slice_start[rw];
440 nr_slices = time_elapsed / throtl_slice;
444 tmp = tg->bps[rw] * throtl_slice * nr_slices;
448 io_trim = (tg->iops[rw] * throtl_slice * nr_slices)/HZ;
450 if (!bytes_trim && !io_trim)
453 if (tg->bytes_disp[rw] >= bytes_trim)
454 tg->bytes_disp[rw] -= bytes_trim;
456 tg->bytes_disp[rw] = 0;
458 if (tg->io_disp[rw] >= io_trim)
459 tg->io_disp[rw] -= io_trim;
463 tg->slice_start[rw] += nr_slices * throtl_slice;
465 throtl_log_tg(td, tg, "[%c] trim slice nr=%lu bytes=%llu io=%lu"
466 " start=%lu end=%lu jiffies=%lu",
467 rw == READ ? 'R' : 'W', nr_slices, bytes_trim, io_trim,
468 tg->slice_start[rw], tg->slice_end[rw], jiffies);
471 static bool tg_with_in_iops_limit(struct throtl_data *td, struct throtl_grp *tg,
472 struct bio *bio, unsigned long *wait)
474 bool rw = bio_data_dir(bio);
475 unsigned int io_allowed;
476 unsigned long jiffy_elapsed, jiffy_wait, jiffy_elapsed_rnd;
479 jiffy_elapsed = jiffy_elapsed_rnd = jiffies - tg->slice_start[rw];
481 /* Slice has just started. Consider one slice interval */
483 jiffy_elapsed_rnd = throtl_slice;
485 jiffy_elapsed_rnd = roundup(jiffy_elapsed_rnd, throtl_slice);
488 * jiffy_elapsed_rnd should not be a big value as minimum iops can be
489 * 1 then at max jiffy elapsed should be equivalent of 1 second as we
490 * will allow dispatch after 1 second and after that slice should
494 tmp = (u64)tg->iops[rw] * jiffy_elapsed_rnd;
498 io_allowed = UINT_MAX;
502 if (tg->io_disp[rw] + 1 <= io_allowed) {
508 /* Calc approx time to dispatch */
509 jiffy_wait = ((tg->io_disp[rw] + 1) * HZ)/tg->iops[rw] + 1;
511 if (jiffy_wait > jiffy_elapsed)
512 jiffy_wait = jiffy_wait - jiffy_elapsed;
521 static bool tg_with_in_bps_limit(struct throtl_data *td, struct throtl_grp *tg,
522 struct bio *bio, unsigned long *wait)
524 bool rw = bio_data_dir(bio);
525 u64 bytes_allowed, extra_bytes, tmp;
526 unsigned long jiffy_elapsed, jiffy_wait, jiffy_elapsed_rnd;
528 jiffy_elapsed = jiffy_elapsed_rnd = jiffies - tg->slice_start[rw];
530 /* Slice has just started. Consider one slice interval */
532 jiffy_elapsed_rnd = throtl_slice;
534 jiffy_elapsed_rnd = roundup(jiffy_elapsed_rnd, throtl_slice);
536 tmp = tg->bps[rw] * jiffy_elapsed_rnd;
540 if (tg->bytes_disp[rw] + bio->bi_size <= bytes_allowed) {
546 /* Calc approx time to dispatch */
547 extra_bytes = tg->bytes_disp[rw] + bio->bi_size - bytes_allowed;
548 jiffy_wait = div64_u64(extra_bytes * HZ, tg->bps[rw]);
554 * This wait time is without taking into consideration the rounding
555 * up we did. Add that time also.
557 jiffy_wait = jiffy_wait + (jiffy_elapsed_rnd - jiffy_elapsed);
564 * Returns whether one can dispatch a bio or not. Also returns approx number
565 * of jiffies to wait before this bio is with-in IO rate and can be dispatched
567 static bool tg_may_dispatch(struct throtl_data *td, struct throtl_grp *tg,
568 struct bio *bio, unsigned long *wait)
570 bool rw = bio_data_dir(bio);
571 unsigned long bps_wait = 0, iops_wait = 0, max_wait = 0;
574 * Currently whole state machine of group depends on first bio
575 * queued in the group bio list. So one should not be calling
576 * this function with a different bio if there are other bios
579 BUG_ON(tg->nr_queued[rw] && bio != bio_list_peek(&tg->bio_lists[rw]));
581 /* If tg->bps = -1, then BW is unlimited */
582 if (tg->bps[rw] == -1 && tg->iops[rw] == -1) {
589 * If previous slice expired, start a new one otherwise renew/extend
590 * existing slice to make sure it is at least throtl_slice interval
593 if (throtl_slice_used(td, tg, rw))
594 throtl_start_new_slice(td, tg, rw);
596 if (time_before(tg->slice_end[rw], jiffies + throtl_slice))
597 throtl_extend_slice(td, tg, rw, jiffies + throtl_slice);
600 if (tg_with_in_bps_limit(td, tg, bio, &bps_wait)
601 && tg_with_in_iops_limit(td, tg, bio, &iops_wait)) {
607 max_wait = max(bps_wait, iops_wait);
612 if (time_before(tg->slice_end[rw], jiffies + max_wait))
613 throtl_extend_slice(td, tg, rw, jiffies + max_wait);
618 static void throtl_charge_bio(struct throtl_grp *tg, struct bio *bio)
620 bool rw = bio_data_dir(bio);
621 bool sync = bio->bi_rw & REQ_SYNC;
623 /* Charge the bio to the group */
624 tg->bytes_disp[rw] += bio->bi_size;
628 * TODO: This will take blkg->stats_lock. Figure out a way
629 * to avoid this cost.
631 blkiocg_update_dispatch_stats(&tg->blkg, bio->bi_size, rw, sync);
634 static void throtl_add_bio_tg(struct throtl_data *td, struct throtl_grp *tg,
637 bool rw = bio_data_dir(bio);
639 bio_list_add(&tg->bio_lists[rw], bio);
640 /* Take a bio reference on tg */
641 throtl_ref_get_tg(tg);
644 throtl_enqueue_tg(td, tg);
647 static void tg_update_disptime(struct throtl_data *td, struct throtl_grp *tg)
649 unsigned long read_wait = -1, write_wait = -1, min_wait = -1, disptime;
652 if ((bio = bio_list_peek(&tg->bio_lists[READ])))
653 tg_may_dispatch(td, tg, bio, &read_wait);
655 if ((bio = bio_list_peek(&tg->bio_lists[WRITE])))
656 tg_may_dispatch(td, tg, bio, &write_wait);
658 min_wait = min(read_wait, write_wait);
659 disptime = jiffies + min_wait;
661 /* Update dispatch time */
662 throtl_dequeue_tg(td, tg);
663 tg->disptime = disptime;
664 throtl_enqueue_tg(td, tg);
667 static void tg_dispatch_one_bio(struct throtl_data *td, struct throtl_grp *tg,
668 bool rw, struct bio_list *bl)
672 bio = bio_list_pop(&tg->bio_lists[rw]);
674 /* Drop bio reference on tg */
677 BUG_ON(td->nr_queued[rw] <= 0);
680 throtl_charge_bio(tg, bio);
681 bio_list_add(bl, bio);
682 bio->bi_rw |= REQ_THROTTLED;
684 throtl_trim_slice(td, tg, rw);
687 static int throtl_dispatch_tg(struct throtl_data *td, struct throtl_grp *tg,
690 unsigned int nr_reads = 0, nr_writes = 0;
691 unsigned int max_nr_reads = throtl_grp_quantum*3/4;
692 unsigned int max_nr_writes = throtl_grp_quantum - max_nr_reads;
695 /* Try to dispatch 75% READS and 25% WRITES */
697 while ((bio = bio_list_peek(&tg->bio_lists[READ]))
698 && tg_may_dispatch(td, tg, bio, NULL)) {
700 tg_dispatch_one_bio(td, tg, bio_data_dir(bio), bl);
703 if (nr_reads >= max_nr_reads)
707 while ((bio = bio_list_peek(&tg->bio_lists[WRITE]))
708 && tg_may_dispatch(td, tg, bio, NULL)) {
710 tg_dispatch_one_bio(td, tg, bio_data_dir(bio), bl);
713 if (nr_writes >= max_nr_writes)
717 return nr_reads + nr_writes;
720 static int throtl_select_dispatch(struct throtl_data *td, struct bio_list *bl)
722 unsigned int nr_disp = 0;
723 struct throtl_grp *tg;
724 struct throtl_rb_root *st = &td->tg_service_tree;
727 tg = throtl_rb_first(st);
732 if (time_before(jiffies, tg->disptime))
735 throtl_dequeue_tg(td, tg);
737 nr_disp += throtl_dispatch_tg(td, tg, bl);
739 if (tg->nr_queued[0] || tg->nr_queued[1]) {
740 tg_update_disptime(td, tg);
741 throtl_enqueue_tg(td, tg);
744 if (nr_disp >= throtl_quantum)
751 static void throtl_process_limit_change(struct throtl_data *td)
753 struct throtl_grp *tg;
754 struct hlist_node *pos, *n;
756 if (!td->limits_changed)
759 xchg(&td->limits_changed, false);
761 throtl_log(td, "limits changed");
763 hlist_for_each_entry_safe(tg, pos, n, &td->tg_list, tg_node) {
764 if (!tg->limits_changed)
767 if (!xchg(&tg->limits_changed, false))
770 throtl_log_tg(td, tg, "limit change rbps=%llu wbps=%llu"
771 " riops=%u wiops=%u", tg->bps[READ], tg->bps[WRITE],
772 tg->iops[READ], tg->iops[WRITE]);
775 * Restart the slices for both READ and WRITES. It
776 * might happen that a group's limit are dropped
777 * suddenly and we don't want to account recently
778 * dispatched IO with new low rate
780 throtl_start_new_slice(td, tg, 0);
781 throtl_start_new_slice(td, tg, 1);
783 if (throtl_tg_on_rr(tg))
784 tg_update_disptime(td, tg);
788 /* Dispatch throttled bios. Should be called without queue lock held. */
789 static int throtl_dispatch(struct request_queue *q)
791 struct throtl_data *td = q->td;
792 unsigned int nr_disp = 0;
793 struct bio_list bio_list_on_stack;
795 struct blk_plug plug;
797 spin_lock_irq(q->queue_lock);
799 throtl_process_limit_change(td);
801 if (!total_nr_queued(td))
804 bio_list_init(&bio_list_on_stack);
806 throtl_log(td, "dispatch nr_queued=%lu read=%u write=%u",
807 total_nr_queued(td), td->nr_queued[READ],
808 td->nr_queued[WRITE]);
810 nr_disp = throtl_select_dispatch(td, &bio_list_on_stack);
813 throtl_log(td, "bios disp=%u", nr_disp);
815 throtl_schedule_next_dispatch(td);
817 spin_unlock_irq(q->queue_lock);
820 * If we dispatched some requests, unplug the queue to make sure
824 blk_start_plug(&plug);
825 while((bio = bio_list_pop(&bio_list_on_stack)))
826 generic_make_request(bio);
827 blk_finish_plug(&plug);
832 void blk_throtl_work(struct work_struct *work)
834 struct throtl_data *td = container_of(work, struct throtl_data,
836 struct request_queue *q = td->queue;
841 /* Call with queue lock held */
843 throtl_schedule_delayed_work(struct throtl_data *td, unsigned long delay)
846 struct delayed_work *dwork = &td->throtl_work;
848 /* schedule work if limits changed even if no bio is queued */
849 if (total_nr_queued(td) > 0 || td->limits_changed) {
851 * We might have a work scheduled to be executed in future.
852 * Cancel that and schedule a new one.
854 __cancel_delayed_work(dwork);
855 queue_delayed_work(kthrotld_workqueue, dwork, delay);
856 throtl_log(td, "schedule work. delay=%lu jiffies=%lu",
862 throtl_destroy_tg(struct throtl_data *td, struct throtl_grp *tg)
864 /* Something wrong if we are trying to remove same group twice */
865 BUG_ON(hlist_unhashed(&tg->tg_node));
867 hlist_del_init(&tg->tg_node);
870 * Put the reference taken at the time of creation so that when all
871 * queues are gone, group can be destroyed.
874 td->nr_undestroyed_grps--;
877 static void throtl_release_tgs(struct throtl_data *td)
879 struct hlist_node *pos, *n;
880 struct throtl_grp *tg;
882 hlist_for_each_entry_safe(tg, pos, n, &td->tg_list, tg_node) {
884 * If cgroup removal path got to blk_group first and removed
885 * it from cgroup list, then it will take care of destroying
888 if (!blkiocg_del_blkio_group(&tg->blkg))
889 throtl_destroy_tg(td, tg);
893 static void throtl_td_free(struct throtl_data *td)
899 * Blk cgroup controller notification saying that blkio_group object is being
900 * delinked as associated cgroup object is going away. That also means that
901 * no new IO will come in this group. So get rid of this group as soon as
902 * any pending IO in the group is finished.
904 * This function is called under rcu_read_lock(). key is the rcu protected
905 * pointer. That means "key" is a valid throtl_data pointer as long as we are
908 * "key" was fetched from blkio_group under blkio_cgroup->lock. That means
909 * it should not be NULL as even if queue was going away, cgroup deltion
910 * path got to it first.
912 void throtl_unlink_blkio_group(void *key, struct blkio_group *blkg)
915 struct throtl_data *td = key;
917 spin_lock_irqsave(td->queue->queue_lock, flags);
918 throtl_destroy_tg(td, tg_of_blkg(blkg));
919 spin_unlock_irqrestore(td->queue->queue_lock, flags);
922 static void throtl_update_blkio_group_common(struct throtl_data *td,
923 struct throtl_grp *tg)
925 xchg(&tg->limits_changed, true);
926 xchg(&td->limits_changed, true);
927 /* Schedule a work now to process the limit change */
928 throtl_schedule_delayed_work(td, 0);
932 * For all update functions, key should be a valid pointer because these
933 * update functions are called under blkcg_lock, that means, blkg is
934 * valid and in turn key is valid. queue exit path can not race because
937 * Can not take queue lock in update functions as queue lock under blkcg_lock
938 * is not allowed. Under other paths we take blkcg_lock under queue_lock.
940 static void throtl_update_blkio_group_read_bps(void *key,
941 struct blkio_group *blkg, u64 read_bps)
943 struct throtl_data *td = key;
944 struct throtl_grp *tg = tg_of_blkg(blkg);
946 tg->bps[READ] = read_bps;
947 throtl_update_blkio_group_common(td, tg);
950 static void throtl_update_blkio_group_write_bps(void *key,
951 struct blkio_group *blkg, u64 write_bps)
953 struct throtl_data *td = key;
954 struct throtl_grp *tg = tg_of_blkg(blkg);
956 tg->bps[WRITE] = write_bps;
957 throtl_update_blkio_group_common(td, tg);
960 static void throtl_update_blkio_group_read_iops(void *key,
961 struct blkio_group *blkg, unsigned int read_iops)
963 struct throtl_data *td = key;
964 struct throtl_grp *tg = tg_of_blkg(blkg);
966 tg->iops[READ] = read_iops;
967 throtl_update_blkio_group_common(td, tg);
970 static void throtl_update_blkio_group_write_iops(void *key,
971 struct blkio_group *blkg, unsigned int write_iops)
973 struct throtl_data *td = key;
974 struct throtl_grp *tg = tg_of_blkg(blkg);
976 tg->iops[WRITE] = write_iops;
977 throtl_update_blkio_group_common(td, tg);
980 static void throtl_shutdown_wq(struct request_queue *q)
982 struct throtl_data *td = q->td;
984 cancel_delayed_work_sync(&td->throtl_work);
987 static struct blkio_policy_type blkio_policy_throtl = {
989 .blkio_unlink_group_fn = throtl_unlink_blkio_group,
990 .blkio_update_group_read_bps_fn =
991 throtl_update_blkio_group_read_bps,
992 .blkio_update_group_write_bps_fn =
993 throtl_update_blkio_group_write_bps,
994 .blkio_update_group_read_iops_fn =
995 throtl_update_blkio_group_read_iops,
996 .blkio_update_group_write_iops_fn =
997 throtl_update_blkio_group_write_iops,
999 .plid = BLKIO_POLICY_THROTL,
1002 int blk_throtl_bio(struct request_queue *q, struct bio **biop)
1004 struct throtl_data *td = q->td;
1005 struct throtl_grp *tg;
1006 struct bio *bio = *biop;
1007 bool rw = bio_data_dir(bio), update_disptime = true;
1009 if (bio->bi_rw & REQ_THROTTLED) {
1010 bio->bi_rw &= ~REQ_THROTTLED;
1014 spin_lock_irq(q->queue_lock);
1015 tg = throtl_get_tg(td);
1017 if (tg->nr_queued[rw]) {
1019 * There is already another bio queued in same dir. No
1020 * need to update dispatch time.
1022 update_disptime = false;
1027 /* Bio is with-in rate limit of group */
1028 if (tg_may_dispatch(td, tg, bio, NULL)) {
1029 throtl_charge_bio(tg, bio);
1032 * We need to trim slice even when bios are not being queued
1033 * otherwise it might happen that a bio is not queued for
1034 * a long time and slice keeps on extending and trim is not
1035 * called for a long time. Now if limits are reduced suddenly
1036 * we take into account all the IO dispatched so far at new
1037 * low rate and * newly queued IO gets a really long dispatch
1040 * So keep on trimming slice even if bio is not queued.
1042 throtl_trim_slice(td, tg, rw);
1047 throtl_log_tg(td, tg, "[%c] bio. bdisp=%u sz=%u bps=%llu"
1048 " iodisp=%u iops=%u queued=%d/%d",
1049 rw == READ ? 'R' : 'W',
1050 tg->bytes_disp[rw], bio->bi_size, tg->bps[rw],
1051 tg->io_disp[rw], tg->iops[rw],
1052 tg->nr_queued[READ], tg->nr_queued[WRITE]);
1054 throtl_add_bio_tg(q->td, tg, bio);
1057 if (update_disptime) {
1058 tg_update_disptime(td, tg);
1059 throtl_schedule_next_dispatch(td);
1063 spin_unlock_irq(q->queue_lock);
1067 int blk_throtl_init(struct request_queue *q)
1069 struct throtl_data *td;
1070 struct throtl_grp *tg;
1072 td = kzalloc_node(sizeof(*td), GFP_KERNEL, q->node);
1076 INIT_HLIST_HEAD(&td->tg_list);
1077 td->tg_service_tree = THROTL_RB_ROOT;
1078 td->limits_changed = false;
1079 INIT_DELAYED_WORK(&td->throtl_work, blk_throtl_work);
1081 /* Init root group */
1083 throtl_init_group(tg);
1086 * Set root group reference to 2. One reference will be dropped when
1087 * all groups on tg_list are being deleted during queue exit. Other
1088 * reference will remain there as we don't want to delete this group
1089 * as it is statically allocated and gets destroyed when throtl_data
1092 atomic_inc(&tg->ref);
1095 blkiocg_add_blkio_group(&blkio_root_cgroup, &tg->blkg, (void *)td,
1096 0, BLKIO_POLICY_THROTL);
1098 throtl_add_group_to_td_list(td, tg);
1100 /* Attach throtl data to request queue */
1106 void blk_throtl_exit(struct request_queue *q)
1108 struct throtl_data *td = q->td;
1113 throtl_shutdown_wq(q);
1115 spin_lock_irq(q->queue_lock);
1116 throtl_release_tgs(td);
1118 /* If there are other groups */
1119 if (td->nr_undestroyed_grps > 0)
1122 spin_unlock_irq(q->queue_lock);
1125 * Wait for tg->blkg->key accessors to exit their grace periods.
1126 * Do this wait only if there are other undestroyed groups out
1127 * there (other than root group). This can happen if cgroup deletion
1128 * path claimed the responsibility of cleaning up a group before
1129 * queue cleanup code get to the group.
1131 * Do not call synchronize_rcu() unconditionally as there are drivers
1132 * which create/delete request queue hundreds of times during scan/boot
1133 * and synchronize_rcu() can take significant time and slow down boot.
1139 * Just being safe to make sure after previous flush if some body did
1140 * update limits through cgroup and another work got queued, cancel
1143 throtl_shutdown_wq(q);
1147 static int __init throtl_init(void)
1149 kthrotld_workqueue = alloc_workqueue("kthrotld", WQ_MEM_RECLAIM, 0);
1150 if (!kthrotld_workqueue)
1151 panic("Failed to create kthrotld\n");
1153 blkio_policy_register(&blkio_policy_throtl);
1157 module_init(throtl_init);