1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * cgroups support for the BFQ I/O scheduler.
5 #include <linux/module.h>
6 #include <linux/slab.h>
7 #include <linux/blkdev.h>
8 #include <linux/cgroup.h>
9 #include <linux/ktime.h>
10 #include <linux/rbtree.h>
11 #include <linux/ioprio.h>
12 #include <linux/sbitmap.h>
13 #include <linux/delay.h>
16 #include "bfq-iosched.h"
18 #ifdef CONFIG_BFQ_CGROUP_DEBUG
19 static int bfq_stat_init(struct bfq_stat *stat, gfp_t gfp)
23 ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp);
27 atomic64_set(&stat->aux_cnt, 0);
31 static void bfq_stat_exit(struct bfq_stat *stat)
33 percpu_counter_destroy(&stat->cpu_cnt);
37 * bfq_stat_add - add a value to a bfq_stat
38 * @stat: target bfq_stat
41 * Add @val to @stat. The caller must ensure that IRQ on the same CPU
42 * don't re-enter this function for the same counter.
44 static inline void bfq_stat_add(struct bfq_stat *stat, uint64_t val)
46 percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH);
50 * bfq_stat_read - read the current value of a bfq_stat
51 * @stat: bfq_stat to read
53 static inline uint64_t bfq_stat_read(struct bfq_stat *stat)
55 return percpu_counter_sum_positive(&stat->cpu_cnt);
59 * bfq_stat_reset - reset a bfq_stat
60 * @stat: bfq_stat to reset
62 static inline void bfq_stat_reset(struct bfq_stat *stat)
64 percpu_counter_set(&stat->cpu_cnt, 0);
65 atomic64_set(&stat->aux_cnt, 0);
69 * bfq_stat_add_aux - add a bfq_stat into another's aux count
70 * @to: the destination bfq_stat
73 * Add @from's count including the aux one to @to's aux count.
75 static inline void bfq_stat_add_aux(struct bfq_stat *to,
76 struct bfq_stat *from)
78 atomic64_add(bfq_stat_read(from) + atomic64_read(&from->aux_cnt),
83 * blkg_prfill_stat - prfill callback for bfq_stat
84 * @sf: seq_file to print to
85 * @pd: policy private data of interest
86 * @off: offset to the bfq_stat in @pd
88 * prfill callback for printing a bfq_stat.
90 static u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd,
93 return __blkg_prfill_u64(sf, pd, bfq_stat_read((void *)pd + off));
96 /* bfqg stats flags */
97 enum bfqg_stats_flags {
98 BFQG_stats_waiting = 0,
103 #define BFQG_FLAG_FNS(name) \
104 static void bfqg_stats_mark_##name(struct bfqg_stats *stats) \
106 stats->flags |= (1 << BFQG_stats_##name); \
108 static void bfqg_stats_clear_##name(struct bfqg_stats *stats) \
110 stats->flags &= ~(1 << BFQG_stats_##name); \
112 static int bfqg_stats_##name(struct bfqg_stats *stats) \
114 return (stats->flags & (1 << BFQG_stats_##name)) != 0; \
117 BFQG_FLAG_FNS(waiting)
118 BFQG_FLAG_FNS(idling)
122 /* This should be called with the scheduler lock held. */
123 static void bfqg_stats_update_group_wait_time(struct bfqg_stats *stats)
127 if (!bfqg_stats_waiting(stats))
130 now = ktime_get_ns();
131 if (now > stats->start_group_wait_time)
132 bfq_stat_add(&stats->group_wait_time,
133 now - stats->start_group_wait_time);
134 bfqg_stats_clear_waiting(stats);
137 /* This should be called with the scheduler lock held. */
138 static void bfqg_stats_set_start_group_wait_time(struct bfq_group *bfqg,
139 struct bfq_group *curr_bfqg)
141 struct bfqg_stats *stats = &bfqg->stats;
143 if (bfqg_stats_waiting(stats))
145 if (bfqg == curr_bfqg)
147 stats->start_group_wait_time = ktime_get_ns();
148 bfqg_stats_mark_waiting(stats);
151 /* This should be called with the scheduler lock held. */
152 static void bfqg_stats_end_empty_time(struct bfqg_stats *stats)
156 if (!bfqg_stats_empty(stats))
159 now = ktime_get_ns();
160 if (now > stats->start_empty_time)
161 bfq_stat_add(&stats->empty_time,
162 now - stats->start_empty_time);
163 bfqg_stats_clear_empty(stats);
166 void bfqg_stats_update_dequeue(struct bfq_group *bfqg)
168 bfq_stat_add(&bfqg->stats.dequeue, 1);
171 void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg)
173 struct bfqg_stats *stats = &bfqg->stats;
175 if (blkg_rwstat_total(&stats->queued))
179 * group is already marked empty. This can happen if bfqq got new
180 * request in parent group and moved to this group while being added
181 * to service tree. Just ignore the event and move on.
183 if (bfqg_stats_empty(stats))
186 stats->start_empty_time = ktime_get_ns();
187 bfqg_stats_mark_empty(stats);
190 void bfqg_stats_update_idle_time(struct bfq_group *bfqg)
192 struct bfqg_stats *stats = &bfqg->stats;
194 if (bfqg_stats_idling(stats)) {
195 u64 now = ktime_get_ns();
197 if (now > stats->start_idle_time)
198 bfq_stat_add(&stats->idle_time,
199 now - stats->start_idle_time);
200 bfqg_stats_clear_idling(stats);
204 void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg)
206 struct bfqg_stats *stats = &bfqg->stats;
208 stats->start_idle_time = ktime_get_ns();
209 bfqg_stats_mark_idling(stats);
212 void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg)
214 struct bfqg_stats *stats = &bfqg->stats;
216 bfq_stat_add(&stats->avg_queue_size_sum,
217 blkg_rwstat_total(&stats->queued));
218 bfq_stat_add(&stats->avg_queue_size_samples, 1);
219 bfqg_stats_update_group_wait_time(stats);
222 void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq,
225 blkg_rwstat_add(&bfqg->stats.queued, opf, 1);
226 bfqg_stats_end_empty_time(&bfqg->stats);
227 if (!(bfqq == bfqg->bfqd->in_service_queue))
228 bfqg_stats_set_start_group_wait_time(bfqg, bfqq_group(bfqq));
231 void bfqg_stats_update_io_remove(struct bfq_group *bfqg, blk_opf_t opf)
233 blkg_rwstat_add(&bfqg->stats.queued, opf, -1);
236 void bfqg_stats_update_io_merged(struct bfq_group *bfqg, blk_opf_t opf)
238 blkg_rwstat_add(&bfqg->stats.merged, opf, 1);
241 void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
242 u64 io_start_time_ns, blk_opf_t opf)
244 struct bfqg_stats *stats = &bfqg->stats;
245 u64 now = ktime_get_ns();
247 if (now > io_start_time_ns)
248 blkg_rwstat_add(&stats->service_time, opf,
249 now - io_start_time_ns);
250 if (io_start_time_ns > start_time_ns)
251 blkg_rwstat_add(&stats->wait_time, opf,
252 io_start_time_ns - start_time_ns);
255 #else /* CONFIG_BFQ_CGROUP_DEBUG */
257 void bfqg_stats_update_io_remove(struct bfq_group *bfqg, blk_opf_t opf) { }
258 void bfqg_stats_update_io_merged(struct bfq_group *bfqg, blk_opf_t opf) { }
259 void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
260 u64 io_start_time_ns, blk_opf_t opf) { }
261 void bfqg_stats_update_dequeue(struct bfq_group *bfqg) { }
262 void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg) { }
264 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
266 #ifdef CONFIG_BFQ_GROUP_IOSCHED
269 * blk-cgroup policy-related handlers
270 * The following functions help in converting between blk-cgroup
271 * internal structures and BFQ-specific structures.
274 static struct bfq_group *pd_to_bfqg(struct blkg_policy_data *pd)
276 return pd ? container_of(pd, struct bfq_group, pd) : NULL;
279 struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg)
281 return pd_to_blkg(&bfqg->pd);
284 static struct bfq_group *blkg_to_bfqg(struct blkcg_gq *blkg)
286 return pd_to_bfqg(blkg_to_pd(blkg, &blkcg_policy_bfq));
291 * The following functions help in navigating the bfq_group hierarchy
292 * by allowing to find the parent of a bfq_group or the bfq_group
293 * associated to a bfq_queue.
296 static struct bfq_group *bfqg_parent(struct bfq_group *bfqg)
298 struct blkcg_gq *pblkg = bfqg_to_blkg(bfqg)->parent;
300 return pblkg ? blkg_to_bfqg(pblkg) : NULL;
303 struct bfq_group *bfqq_group(struct bfq_queue *bfqq)
305 struct bfq_entity *group_entity = bfqq->entity.parent;
307 return group_entity ? container_of(group_entity, struct bfq_group,
309 bfqq->bfqd->root_group;
313 * The following two functions handle get and put of a bfq_group by
314 * wrapping the related blk-cgroup hooks.
317 static void bfqg_get(struct bfq_group *bfqg)
319 refcount_inc(&bfqg->ref);
322 static void bfqg_put(struct bfq_group *bfqg)
324 if (refcount_dec_and_test(&bfqg->ref))
328 static void bfqg_and_blkg_get(struct bfq_group *bfqg)
330 /* see comments in bfq_bic_update_cgroup for why refcounting bfqg */
333 blkg_get(bfqg_to_blkg(bfqg));
336 void bfqg_and_blkg_put(struct bfq_group *bfqg)
338 blkg_put(bfqg_to_blkg(bfqg));
343 void bfqg_stats_update_legacy_io(struct request_queue *q, struct request *rq)
345 struct bfq_group *bfqg = blkg_to_bfqg(rq->bio->bi_blkg);
350 blkg_rwstat_add(&bfqg->stats.bytes, rq->cmd_flags, blk_rq_bytes(rq));
351 blkg_rwstat_add(&bfqg->stats.ios, rq->cmd_flags, 1);
355 static void bfqg_stats_reset(struct bfqg_stats *stats)
357 #ifdef CONFIG_BFQ_CGROUP_DEBUG
358 /* queued stats shouldn't be cleared */
359 blkg_rwstat_reset(&stats->merged);
360 blkg_rwstat_reset(&stats->service_time);
361 blkg_rwstat_reset(&stats->wait_time);
362 bfq_stat_reset(&stats->time);
363 bfq_stat_reset(&stats->avg_queue_size_sum);
364 bfq_stat_reset(&stats->avg_queue_size_samples);
365 bfq_stat_reset(&stats->dequeue);
366 bfq_stat_reset(&stats->group_wait_time);
367 bfq_stat_reset(&stats->idle_time);
368 bfq_stat_reset(&stats->empty_time);
373 static void bfqg_stats_add_aux(struct bfqg_stats *to, struct bfqg_stats *from)
378 #ifdef CONFIG_BFQ_CGROUP_DEBUG
379 /* queued stats shouldn't be cleared */
380 blkg_rwstat_add_aux(&to->merged, &from->merged);
381 blkg_rwstat_add_aux(&to->service_time, &from->service_time);
382 blkg_rwstat_add_aux(&to->wait_time, &from->wait_time);
383 bfq_stat_add_aux(&from->time, &from->time);
384 bfq_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum);
385 bfq_stat_add_aux(&to->avg_queue_size_samples,
386 &from->avg_queue_size_samples);
387 bfq_stat_add_aux(&to->dequeue, &from->dequeue);
388 bfq_stat_add_aux(&to->group_wait_time, &from->group_wait_time);
389 bfq_stat_add_aux(&to->idle_time, &from->idle_time);
390 bfq_stat_add_aux(&to->empty_time, &from->empty_time);
395 * Transfer @bfqg's stats to its parent's aux counts so that the ancestors'
396 * recursive stats can still account for the amount used by this bfqg after
399 static void bfqg_stats_xfer_dead(struct bfq_group *bfqg)
401 struct bfq_group *parent;
403 if (!bfqg) /* root_group */
406 parent = bfqg_parent(bfqg);
408 lockdep_assert_held(&bfqg_to_blkg(bfqg)->q->queue_lock);
410 if (unlikely(!parent))
413 bfqg_stats_add_aux(&parent->stats, &bfqg->stats);
414 bfqg_stats_reset(&bfqg->stats);
417 void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg)
419 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
421 entity->weight = entity->new_weight;
422 entity->orig_weight = entity->new_weight;
424 bfqq->ioprio = bfqq->new_ioprio;
425 bfqq->ioprio_class = bfqq->new_ioprio_class;
427 * Make sure that bfqg and its associated blkg do not
428 * disappear before entity.
430 bfqg_and_blkg_get(bfqg);
432 entity->parent = bfqg->my_entity; /* NULL for root group */
433 entity->sched_data = &bfqg->sched_data;
436 static void bfqg_stats_exit(struct bfqg_stats *stats)
438 blkg_rwstat_exit(&stats->bytes);
439 blkg_rwstat_exit(&stats->ios);
440 #ifdef CONFIG_BFQ_CGROUP_DEBUG
441 blkg_rwstat_exit(&stats->merged);
442 blkg_rwstat_exit(&stats->service_time);
443 blkg_rwstat_exit(&stats->wait_time);
444 blkg_rwstat_exit(&stats->queued);
445 bfq_stat_exit(&stats->time);
446 bfq_stat_exit(&stats->avg_queue_size_sum);
447 bfq_stat_exit(&stats->avg_queue_size_samples);
448 bfq_stat_exit(&stats->dequeue);
449 bfq_stat_exit(&stats->group_wait_time);
450 bfq_stat_exit(&stats->idle_time);
451 bfq_stat_exit(&stats->empty_time);
455 static int bfqg_stats_init(struct bfqg_stats *stats, gfp_t gfp)
457 if (blkg_rwstat_init(&stats->bytes, gfp) ||
458 blkg_rwstat_init(&stats->ios, gfp))
461 #ifdef CONFIG_BFQ_CGROUP_DEBUG
462 if (blkg_rwstat_init(&stats->merged, gfp) ||
463 blkg_rwstat_init(&stats->service_time, gfp) ||
464 blkg_rwstat_init(&stats->wait_time, gfp) ||
465 blkg_rwstat_init(&stats->queued, gfp) ||
466 bfq_stat_init(&stats->time, gfp) ||
467 bfq_stat_init(&stats->avg_queue_size_sum, gfp) ||
468 bfq_stat_init(&stats->avg_queue_size_samples, gfp) ||
469 bfq_stat_init(&stats->dequeue, gfp) ||
470 bfq_stat_init(&stats->group_wait_time, gfp) ||
471 bfq_stat_init(&stats->idle_time, gfp) ||
472 bfq_stat_init(&stats->empty_time, gfp))
479 bfqg_stats_exit(stats);
483 static struct bfq_group_data *cpd_to_bfqgd(struct blkcg_policy_data *cpd)
485 return cpd ? container_of(cpd, struct bfq_group_data, pd) : NULL;
488 static struct bfq_group_data *blkcg_to_bfqgd(struct blkcg *blkcg)
490 return cpd_to_bfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_bfq));
493 static struct blkcg_policy_data *bfq_cpd_alloc(gfp_t gfp)
495 struct bfq_group_data *bgd;
497 bgd = kzalloc(sizeof(*bgd), gfp);
503 static void bfq_cpd_init(struct blkcg_policy_data *cpd)
505 struct bfq_group_data *d = cpd_to_bfqgd(cpd);
507 d->weight = CGROUP_WEIGHT_DFL;
510 static void bfq_cpd_free(struct blkcg_policy_data *cpd)
512 kfree(cpd_to_bfqgd(cpd));
515 static struct blkg_policy_data *bfq_pd_alloc(struct gendisk *disk,
516 struct blkcg *blkcg, gfp_t gfp)
518 struct bfq_group *bfqg;
520 bfqg = kzalloc_node(sizeof(*bfqg), gfp, disk->node_id);
524 if (bfqg_stats_init(&bfqg->stats, gfp)) {
529 /* see comments in bfq_bic_update_cgroup for why refcounting */
530 refcount_set(&bfqg->ref, 1);
534 static void bfq_pd_init(struct blkg_policy_data *pd)
536 struct blkcg_gq *blkg = pd_to_blkg(pd);
537 struct bfq_group *bfqg = blkg_to_bfqg(blkg);
538 struct bfq_data *bfqd = blkg->q->elevator->elevator_data;
539 struct bfq_entity *entity = &bfqg->entity;
540 struct bfq_group_data *d = blkcg_to_bfqgd(blkg->blkcg);
542 entity->orig_weight = entity->weight = entity->new_weight = d->weight;
543 entity->my_sched_data = &bfqg->sched_data;
544 entity->last_bfqq_created = NULL;
546 bfqg->my_entity = entity; /*
547 * the root_group's will be set to NULL
548 * in bfq_init_queue()
551 bfqg->active_entities = 0;
552 bfqg->num_queues_with_pending_reqs = 0;
553 bfqg->rq_pos_tree = RB_ROOT;
556 static void bfq_pd_free(struct blkg_policy_data *pd)
558 struct bfq_group *bfqg = pd_to_bfqg(pd);
560 bfqg_stats_exit(&bfqg->stats);
564 static void bfq_pd_reset_stats(struct blkg_policy_data *pd)
566 struct bfq_group *bfqg = pd_to_bfqg(pd);
568 bfqg_stats_reset(&bfqg->stats);
571 static void bfq_group_set_parent(struct bfq_group *bfqg,
572 struct bfq_group *parent)
574 struct bfq_entity *entity;
576 entity = &bfqg->entity;
577 entity->parent = parent->my_entity;
578 entity->sched_data = &parent->sched_data;
581 static void bfq_link_bfqg(struct bfq_data *bfqd, struct bfq_group *bfqg)
583 struct bfq_group *parent;
584 struct bfq_entity *entity;
587 * Update chain of bfq_groups as we might be handling a leaf group
588 * which, along with some of its relatives, has not been hooked yet
589 * to the private hierarchy of BFQ.
591 entity = &bfqg->entity;
592 for_each_entity(entity) {
593 struct bfq_group *curr_bfqg = container_of(entity,
594 struct bfq_group, entity);
595 if (curr_bfqg != bfqd->root_group) {
596 parent = bfqg_parent(curr_bfqg);
598 parent = bfqd->root_group;
599 bfq_group_set_parent(curr_bfqg, parent);
604 struct bfq_group *bfq_bio_bfqg(struct bfq_data *bfqd, struct bio *bio)
606 struct blkcg_gq *blkg = bio->bi_blkg;
607 struct bfq_group *bfqg;
614 bfqg = blkg_to_bfqg(blkg);
615 if (bfqg->pd.online) {
616 bio_associate_blkg_from_css(bio, &blkg->blkcg->css);
621 bio_associate_blkg_from_css(bio,
622 &bfqg_to_blkg(bfqd->root_group)->blkcg->css);
623 return bfqd->root_group;
627 * bfq_bfqq_move - migrate @bfqq to @bfqg.
628 * @bfqd: queue descriptor.
629 * @bfqq: the queue to move.
630 * @bfqg: the group to move to.
632 * Move @bfqq to @bfqg, deactivating it from its old group and reactivating
633 * it on the new one. Avoid putting the entity on the old group idle tree.
635 * Must be called under the scheduler lock, to make sure that the blkg
636 * owning @bfqg does not disappear (see comments in
637 * bfq_bic_update_cgroup on guaranteeing the consistency of blkg
640 void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
641 struct bfq_group *bfqg)
643 struct bfq_entity *entity = &bfqq->entity;
644 struct bfq_group *old_parent = bfqq_group(bfqq);
645 bool has_pending_reqs = false;
648 * No point to move bfqq to the same group, which can happen when
649 * root group is offlined
651 if (old_parent == bfqg)
655 * oom_bfqq is not allowed to move, oom_bfqq will hold ref to root_group
656 * until elevator exit.
658 if (bfqq == &bfqd->oom_bfqq)
661 * Get extra reference to prevent bfqq from being freed in
662 * next possible expire or deactivate.
666 if (entity->in_groups_with_pending_reqs) {
667 has_pending_reqs = true;
668 bfq_del_bfqq_in_groups_with_pending_reqs(bfqq);
671 /* If bfqq is empty, then bfq_bfqq_expire also invokes
672 * bfq_del_bfqq_busy, thereby removing bfqq and its entity
673 * from data structures related to current group. Otherwise we
674 * need to remove bfqq explicitly with bfq_deactivate_bfqq, as
677 if (bfqq == bfqd->in_service_queue)
678 bfq_bfqq_expire(bfqd, bfqd->in_service_queue,
679 false, BFQQE_PREEMPTED);
681 if (bfq_bfqq_busy(bfqq))
682 bfq_deactivate_bfqq(bfqd, bfqq, false, false);
683 else if (entity->on_st_or_in_serv)
684 bfq_put_idle_entity(bfq_entity_service_tree(entity), entity);
685 bfqg_and_blkg_put(old_parent);
687 if (entity->parent &&
688 entity->parent->last_bfqq_created == bfqq)
689 entity->parent->last_bfqq_created = NULL;
690 else if (bfqd->last_bfqq_created == bfqq)
691 bfqd->last_bfqq_created = NULL;
693 entity->parent = bfqg->my_entity;
694 entity->sched_data = &bfqg->sched_data;
695 /* pin down bfqg and its associated blkg */
696 bfqg_and_blkg_get(bfqg);
698 if (has_pending_reqs)
699 bfq_add_bfqq_in_groups_with_pending_reqs(bfqq);
701 if (bfq_bfqq_busy(bfqq)) {
702 if (unlikely(!bfqd->nonrot_with_queueing))
703 bfq_pos_tree_add_move(bfqd, bfqq);
704 bfq_activate_bfqq(bfqd, bfqq);
707 if (!bfqd->in_service_queue && !bfqd->tot_rq_in_driver)
708 bfq_schedule_dispatch(bfqd);
709 /* release extra ref taken above, bfqq may happen to be freed now */
713 static void bfq_sync_bfqq_move(struct bfq_data *bfqd,
714 struct bfq_queue *sync_bfqq,
715 struct bfq_io_cq *bic,
716 struct bfq_group *bfqg,
717 unsigned int act_idx)
719 struct bfq_queue *bfqq;
721 if (!sync_bfqq->new_bfqq && !bfq_bfqq_coop(sync_bfqq)) {
722 /* We are the only user of this bfqq, just move it */
723 if (sync_bfqq->entity.sched_data != &bfqg->sched_data)
724 bfq_bfqq_move(bfqd, sync_bfqq, bfqg);
729 * The queue was merged to a different queue. Check
730 * that the merge chain still belongs to the same
733 for (bfqq = sync_bfqq; bfqq; bfqq = bfqq->new_bfqq)
734 if (bfqq->entity.sched_data != &bfqg->sched_data)
738 * Some queue changed cgroup so the merge is not valid
739 * anymore. We cannot easily just cancel the merge (by
740 * clearing new_bfqq) as there may be other processes
741 * using this queue and holding refs to all queues
742 * below sync_bfqq->new_bfqq. Similarly if the merge
743 * already happened, we need to detach from bfqq now
744 * so that we cannot merge bio to a request from the
747 bfq_put_cooperator(sync_bfqq);
748 bic_set_bfqq(bic, NULL, true, act_idx);
749 bfq_release_process_ref(bfqd, sync_bfqq);
754 * __bfq_bic_change_cgroup - move @bic to @bfqg.
755 * @bfqd: the queue descriptor.
756 * @bic: the bic to move.
757 * @bfqg: the group to move to.
759 * Move bic to blkcg, assuming that bfqd->lock is held; which makes
760 * sure that the reference to cgroup is valid across the call (see
761 * comments in bfq_bic_update_cgroup on this issue)
763 static void __bfq_bic_change_cgroup(struct bfq_data *bfqd,
764 struct bfq_io_cq *bic,
765 struct bfq_group *bfqg)
767 unsigned int act_idx;
769 for (act_idx = 0; act_idx < bfqd->num_actuators; act_idx++) {
770 struct bfq_queue *async_bfqq = bic_to_bfqq(bic, false, act_idx);
771 struct bfq_queue *sync_bfqq = bic_to_bfqq(bic, true, act_idx);
774 async_bfqq->entity.sched_data != &bfqg->sched_data) {
775 bic_set_bfqq(bic, NULL, false, act_idx);
776 bfq_release_process_ref(bfqd, async_bfqq);
780 bfq_sync_bfqq_move(bfqd, sync_bfqq, bic, bfqg, act_idx);
784 void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio)
786 struct bfq_data *bfqd = bic_to_bfqd(bic);
787 struct bfq_group *bfqg = bfq_bio_bfqg(bfqd, bio);
790 serial_nr = bfqg_to_blkg(bfqg)->blkcg->css.serial_nr;
793 * Check whether blkcg has changed. The condition may trigger
794 * spuriously on a newly created cic but there's no harm.
796 if (unlikely(!bfqd) || likely(bic->blkcg_serial_nr == serial_nr))
800 * New cgroup for this process. Make sure it is linked to bfq internal
803 bfq_link_bfqg(bfqd, bfqg);
804 __bfq_bic_change_cgroup(bfqd, bic, bfqg);
806 * Update blkg_path for bfq_log_* functions. We cache this
807 * path, and update it here, for the following
808 * reasons. Operations on blkg objects in blk-cgroup are
809 * protected with the request_queue lock, and not with the
810 * lock that protects the instances of this scheduler
811 * (bfqd->lock). This exposes BFQ to the following sort of
814 * The blkg_lookup performed in bfq_get_queue, protected
815 * through rcu, may happen to return the address of a copy of
816 * the original blkg. If this is the case, then the
817 * bfqg_and_blkg_get performed in bfq_get_queue, to pin down
818 * the blkg, is useless: it does not prevent blk-cgroup code
819 * from destroying both the original blkg and all objects
820 * directly or indirectly referred by the copy of the
823 * On the bright side, destroy operations on a blkg invoke, as
824 * a first step, hooks of the scheduler associated with the
825 * blkg. And these hooks are executed with bfqd->lock held for
826 * BFQ. As a consequence, for any blkg associated with the
827 * request queue this instance of the scheduler is attached
828 * to, we are guaranteed that such a blkg is not destroyed, and
829 * that all the pointers it contains are consistent, while we
830 * are holding bfqd->lock. A blkg_lookup performed with
831 * bfqd->lock held then returns a fully consistent blkg, which
832 * remains consistent until this lock is held.
834 * Thanks to the last fact, and to the fact that: (1) bfqg has
835 * been obtained through a blkg_lookup in the above
836 * assignment, and (2) bfqd->lock is being held, here we can
837 * safely use the policy data for the involved blkg (i.e., the
838 * field bfqg->pd) to get to the blkg associated with bfqg,
839 * and then we can safely use any field of blkg. After we
840 * release bfqd->lock, even just getting blkg through this
841 * bfqg may cause dangling references to be traversed, as
842 * bfqg->pd may not exist any more.
844 * In view of the above facts, here we cache, in the bfqg, any
845 * blkg data we may need for this bic, and for its associated
846 * bfq_queue. As of now, we need to cache only the path of the
847 * blkg, which is used in the bfq_log_* functions.
849 * Finally, note that bfqg itself needs to be protected from
850 * destruction on the blkg_free of the original blkg (which
851 * invokes bfq_pd_free). We use an additional private
852 * refcounter for bfqg, to let it disappear only after no
853 * bfq_queue refers to it any longer.
855 blkg_path(bfqg_to_blkg(bfqg), bfqg->blkg_path, sizeof(bfqg->blkg_path));
856 bic->blkcg_serial_nr = serial_nr;
860 * bfq_flush_idle_tree - deactivate any entity on the idle tree of @st.
861 * @st: the service tree being flushed.
863 static void bfq_flush_idle_tree(struct bfq_service_tree *st)
865 struct bfq_entity *entity = st->first_idle;
867 for (; entity ; entity = st->first_idle)
868 __bfq_deactivate_entity(entity, false);
872 * bfq_reparent_leaf_entity - move leaf entity to the root_group.
873 * @bfqd: the device data structure with the root group.
874 * @entity: the entity to move, if entity is a leaf; or the parent entity
875 * of an active leaf entity to move, if entity is not a leaf.
876 * @ioprio_class: I/O priority class to reparent.
878 static void bfq_reparent_leaf_entity(struct bfq_data *bfqd,
879 struct bfq_entity *entity,
882 struct bfq_queue *bfqq;
883 struct bfq_entity *child_entity = entity;
885 while (child_entity->my_sched_data) { /* leaf not reached yet */
886 struct bfq_sched_data *child_sd = child_entity->my_sched_data;
887 struct bfq_service_tree *child_st = child_sd->service_tree +
889 struct rb_root *child_active = &child_st->active;
891 child_entity = bfq_entity_of(rb_first(child_active));
894 child_entity = child_sd->in_service_entity;
897 bfqq = bfq_entity_to_bfqq(child_entity);
898 bfq_bfqq_move(bfqd, bfqq, bfqd->root_group);
902 * bfq_reparent_active_queues - move to the root group all active queues.
903 * @bfqd: the device data structure with the root group.
904 * @bfqg: the group to move from.
905 * @st: the service tree to start the search from.
906 * @ioprio_class: I/O priority class to reparent.
908 static void bfq_reparent_active_queues(struct bfq_data *bfqd,
909 struct bfq_group *bfqg,
910 struct bfq_service_tree *st,
913 struct rb_root *active = &st->active;
914 struct bfq_entity *entity;
916 while ((entity = bfq_entity_of(rb_first(active))))
917 bfq_reparent_leaf_entity(bfqd, entity, ioprio_class);
919 if (bfqg->sched_data.in_service_entity)
920 bfq_reparent_leaf_entity(bfqd,
921 bfqg->sched_data.in_service_entity,
926 * bfq_pd_offline - deactivate the entity associated with @pd,
927 * and reparent its children entities.
928 * @pd: descriptor of the policy going offline.
930 * blkio already grabs the queue_lock for us, so no need to use
933 static void bfq_pd_offline(struct blkg_policy_data *pd)
935 struct bfq_service_tree *st;
936 struct bfq_group *bfqg = pd_to_bfqg(pd);
937 struct bfq_data *bfqd = bfqg->bfqd;
938 struct bfq_entity *entity = bfqg->my_entity;
942 spin_lock_irqsave(&bfqd->lock, flags);
944 if (!entity) /* root group */
945 goto put_async_queues;
948 * Empty all service_trees belonging to this group before
949 * deactivating the group itself.
951 for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) {
952 st = bfqg->sched_data.service_tree + i;
955 * It may happen that some queues are still active
956 * (busy) upon group destruction (if the corresponding
957 * processes have been forced to terminate). We move
958 * all the leaf entities corresponding to these queues
960 * Also, it may happen that the group has an entity
961 * in service, which is disconnected from the active
962 * tree: it must be moved, too.
963 * There is no need to put the sync queues, as the
964 * scheduler has taken no reference.
966 bfq_reparent_active_queues(bfqd, bfqg, st, i);
969 * The idle tree may still contain bfq_queues
970 * belonging to exited task because they never
971 * migrated to a different cgroup from the one being
972 * destroyed now. In addition, even
973 * bfq_reparent_active_queues() may happen to add some
974 * entities to the idle tree. It happens if, in some
975 * of the calls to bfq_bfqq_move() performed by
976 * bfq_reparent_active_queues(), the queue to move is
977 * empty and gets expired.
979 bfq_flush_idle_tree(st);
982 __bfq_deactivate_entity(entity, false);
985 bfq_put_async_queues(bfqd, bfqg);
987 spin_unlock_irqrestore(&bfqd->lock, flags);
989 * @blkg is going offline and will be ignored by
990 * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so
991 * that they don't get lost. If IOs complete after this point, the
992 * stats for them will be lost. Oh well...
994 bfqg_stats_xfer_dead(bfqg);
997 void bfq_end_wr_async(struct bfq_data *bfqd)
999 struct blkcg_gq *blkg;
1001 list_for_each_entry(blkg, &bfqd->queue->blkg_list, q_node) {
1002 struct bfq_group *bfqg = blkg_to_bfqg(blkg);
1004 bfq_end_wr_async_queues(bfqd, bfqg);
1006 bfq_end_wr_async_queues(bfqd, bfqd->root_group);
1009 static int bfq_io_show_weight_legacy(struct seq_file *sf, void *v)
1011 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
1012 struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg);
1013 unsigned int val = 0;
1016 val = bfqgd->weight;
1018 seq_printf(sf, "%u\n", val);
1023 static u64 bfqg_prfill_weight_device(struct seq_file *sf,
1024 struct blkg_policy_data *pd, int off)
1026 struct bfq_group *bfqg = pd_to_bfqg(pd);
1028 if (!bfqg->entity.dev_weight)
1030 return __blkg_prfill_u64(sf, pd, bfqg->entity.dev_weight);
1033 static int bfq_io_show_weight(struct seq_file *sf, void *v)
1035 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
1036 struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg);
1038 seq_printf(sf, "default %u\n", bfqgd->weight);
1039 blkcg_print_blkgs(sf, blkcg, bfqg_prfill_weight_device,
1040 &blkcg_policy_bfq, 0, false);
1044 static void bfq_group_set_weight(struct bfq_group *bfqg, u64 weight, u64 dev_weight)
1046 weight = dev_weight ?: weight;
1048 bfqg->entity.dev_weight = dev_weight;
1050 * Setting the prio_changed flag of the entity
1051 * to 1 with new_weight == weight would re-set
1052 * the value of the weight to its ioprio mapping.
1053 * Set the flag only if necessary.
1055 if ((unsigned short)weight != bfqg->entity.new_weight) {
1056 bfqg->entity.new_weight = (unsigned short)weight;
1058 * Make sure that the above new value has been
1059 * stored in bfqg->entity.new_weight before
1060 * setting the prio_changed flag. In fact,
1061 * this flag may be read asynchronously (in
1062 * critical sections protected by a different
1063 * lock than that held here), and finding this
1064 * flag set may cause the execution of the code
1065 * for updating parameters whose value may
1066 * depend also on bfqg->entity.new_weight (in
1067 * __bfq_entity_update_weight_prio).
1068 * This barrier makes sure that the new value
1069 * of bfqg->entity.new_weight is correctly
1070 * seen in that code.
1073 bfqg->entity.prio_changed = 1;
1077 static int bfq_io_set_weight_legacy(struct cgroup_subsys_state *css,
1078 struct cftype *cftype,
1081 struct blkcg *blkcg = css_to_blkcg(css);
1082 struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg);
1083 struct blkcg_gq *blkg;
1086 if (val < BFQ_MIN_WEIGHT || val > BFQ_MAX_WEIGHT)
1090 spin_lock_irq(&blkcg->lock);
1091 bfqgd->weight = (unsigned short)val;
1092 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
1093 struct bfq_group *bfqg = blkg_to_bfqg(blkg);
1096 bfq_group_set_weight(bfqg, val, 0);
1098 spin_unlock_irq(&blkcg->lock);
1103 static ssize_t bfq_io_set_device_weight(struct kernfs_open_file *of,
1104 char *buf, size_t nbytes,
1108 struct blkg_conf_ctx ctx;
1109 struct blkcg *blkcg = css_to_blkcg(of_css(of));
1110 struct bfq_group *bfqg;
1113 ret = blkg_conf_prep(blkcg, &blkcg_policy_bfq, buf, &ctx);
1117 if (sscanf(ctx.body, "%llu", &v) == 1) {
1118 /* require "default" on dfl */
1122 } else if (!strcmp(strim(ctx.body), "default")) {
1129 bfqg = blkg_to_bfqg(ctx.blkg);
1132 if (!v || (v >= BFQ_MIN_WEIGHT && v <= BFQ_MAX_WEIGHT)) {
1133 bfq_group_set_weight(bfqg, bfqg->entity.weight, v);
1137 blkg_conf_finish(&ctx);
1138 return ret ?: nbytes;
1141 static ssize_t bfq_io_set_weight(struct kernfs_open_file *of,
1142 char *buf, size_t nbytes,
1151 /* "WEIGHT" or "default WEIGHT" sets the default weight */
1152 v = simple_strtoull(buf, &endp, 0);
1153 if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) {
1154 ret = bfq_io_set_weight_legacy(of_css(of), NULL, v);
1155 return ret ?: nbytes;
1158 return bfq_io_set_device_weight(of, buf, nbytes, off);
1161 static int bfqg_print_rwstat(struct seq_file *sf, void *v)
1163 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat,
1164 &blkcg_policy_bfq, seq_cft(sf)->private, true);
1168 static u64 bfqg_prfill_rwstat_recursive(struct seq_file *sf,
1169 struct blkg_policy_data *pd, int off)
1171 struct blkg_rwstat_sample sum;
1173 blkg_rwstat_recursive_sum(pd_to_blkg(pd), &blkcg_policy_bfq, off, &sum);
1174 return __blkg_prfill_rwstat(sf, pd, &sum);
1177 static int bfqg_print_rwstat_recursive(struct seq_file *sf, void *v)
1179 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1180 bfqg_prfill_rwstat_recursive, &blkcg_policy_bfq,
1181 seq_cft(sf)->private, true);
1185 #ifdef CONFIG_BFQ_CGROUP_DEBUG
1186 static int bfqg_print_stat(struct seq_file *sf, void *v)
1188 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat,
1189 &blkcg_policy_bfq, seq_cft(sf)->private, false);
1193 static u64 bfqg_prfill_stat_recursive(struct seq_file *sf,
1194 struct blkg_policy_data *pd, int off)
1196 struct blkcg_gq *blkg = pd_to_blkg(pd);
1197 struct blkcg_gq *pos_blkg;
1198 struct cgroup_subsys_state *pos_css;
1201 lockdep_assert_held(&blkg->q->queue_lock);
1204 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
1205 struct bfq_stat *stat;
1207 if (!pos_blkg->online)
1210 stat = (void *)blkg_to_pd(pos_blkg, &blkcg_policy_bfq) + off;
1211 sum += bfq_stat_read(stat) + atomic64_read(&stat->aux_cnt);
1215 return __blkg_prfill_u64(sf, pd, sum);
1218 static int bfqg_print_stat_recursive(struct seq_file *sf, void *v)
1220 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1221 bfqg_prfill_stat_recursive, &blkcg_policy_bfq,
1222 seq_cft(sf)->private, false);
1226 static u64 bfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd,
1229 struct bfq_group *bfqg = blkg_to_bfqg(pd->blkg);
1230 u64 sum = blkg_rwstat_total(&bfqg->stats.bytes);
1232 return __blkg_prfill_u64(sf, pd, sum >> 9);
1235 static int bfqg_print_stat_sectors(struct seq_file *sf, void *v)
1237 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1238 bfqg_prfill_sectors, &blkcg_policy_bfq, 0, false);
1242 static u64 bfqg_prfill_sectors_recursive(struct seq_file *sf,
1243 struct blkg_policy_data *pd, int off)
1245 struct blkg_rwstat_sample tmp;
1247 blkg_rwstat_recursive_sum(pd->blkg, &blkcg_policy_bfq,
1248 offsetof(struct bfq_group, stats.bytes), &tmp);
1250 return __blkg_prfill_u64(sf, pd,
1251 (tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE]) >> 9);
1254 static int bfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v)
1256 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1257 bfqg_prfill_sectors_recursive, &blkcg_policy_bfq, 0,
1262 static u64 bfqg_prfill_avg_queue_size(struct seq_file *sf,
1263 struct blkg_policy_data *pd, int off)
1265 struct bfq_group *bfqg = pd_to_bfqg(pd);
1266 u64 samples = bfq_stat_read(&bfqg->stats.avg_queue_size_samples);
1270 v = bfq_stat_read(&bfqg->stats.avg_queue_size_sum);
1271 v = div64_u64(v, samples);
1273 __blkg_prfill_u64(sf, pd, v);
1277 /* print avg_queue_size */
1278 static int bfqg_print_avg_queue_size(struct seq_file *sf, void *v)
1280 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1281 bfqg_prfill_avg_queue_size, &blkcg_policy_bfq,
1285 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
1287 struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node)
1291 ret = blkcg_activate_policy(bfqd->queue->disk, &blkcg_policy_bfq);
1295 return blkg_to_bfqg(bfqd->queue->root_blkg);
1298 struct blkcg_policy blkcg_policy_bfq = {
1299 .dfl_cftypes = bfq_blkg_files,
1300 .legacy_cftypes = bfq_blkcg_legacy_files,
1302 .cpd_alloc_fn = bfq_cpd_alloc,
1303 .cpd_init_fn = bfq_cpd_init,
1304 .cpd_free_fn = bfq_cpd_free,
1306 .pd_alloc_fn = bfq_pd_alloc,
1307 .pd_init_fn = bfq_pd_init,
1308 .pd_offline_fn = bfq_pd_offline,
1309 .pd_free_fn = bfq_pd_free,
1310 .pd_reset_stats_fn = bfq_pd_reset_stats,
1313 struct cftype bfq_blkcg_legacy_files[] = {
1315 .name = "bfq.weight",
1316 .flags = CFTYPE_NOT_ON_ROOT,
1317 .seq_show = bfq_io_show_weight_legacy,
1318 .write_u64 = bfq_io_set_weight_legacy,
1321 .name = "bfq.weight_device",
1322 .flags = CFTYPE_NOT_ON_ROOT,
1323 .seq_show = bfq_io_show_weight,
1324 .write = bfq_io_set_weight,
1327 /* statistics, covers only the tasks in the bfqg */
1329 .name = "bfq.io_service_bytes",
1330 .private = offsetof(struct bfq_group, stats.bytes),
1331 .seq_show = bfqg_print_rwstat,
1334 .name = "bfq.io_serviced",
1335 .private = offsetof(struct bfq_group, stats.ios),
1336 .seq_show = bfqg_print_rwstat,
1338 #ifdef CONFIG_BFQ_CGROUP_DEBUG
1341 .private = offsetof(struct bfq_group, stats.time),
1342 .seq_show = bfqg_print_stat,
1345 .name = "bfq.sectors",
1346 .seq_show = bfqg_print_stat_sectors,
1349 .name = "bfq.io_service_time",
1350 .private = offsetof(struct bfq_group, stats.service_time),
1351 .seq_show = bfqg_print_rwstat,
1354 .name = "bfq.io_wait_time",
1355 .private = offsetof(struct bfq_group, stats.wait_time),
1356 .seq_show = bfqg_print_rwstat,
1359 .name = "bfq.io_merged",
1360 .private = offsetof(struct bfq_group, stats.merged),
1361 .seq_show = bfqg_print_rwstat,
1364 .name = "bfq.io_queued",
1365 .private = offsetof(struct bfq_group, stats.queued),
1366 .seq_show = bfqg_print_rwstat,
1368 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
1370 /* the same statistics which cover the bfqg and its descendants */
1372 .name = "bfq.io_service_bytes_recursive",
1373 .private = offsetof(struct bfq_group, stats.bytes),
1374 .seq_show = bfqg_print_rwstat_recursive,
1377 .name = "bfq.io_serviced_recursive",
1378 .private = offsetof(struct bfq_group, stats.ios),
1379 .seq_show = bfqg_print_rwstat_recursive,
1381 #ifdef CONFIG_BFQ_CGROUP_DEBUG
1383 .name = "bfq.time_recursive",
1384 .private = offsetof(struct bfq_group, stats.time),
1385 .seq_show = bfqg_print_stat_recursive,
1388 .name = "bfq.sectors_recursive",
1389 .seq_show = bfqg_print_stat_sectors_recursive,
1392 .name = "bfq.io_service_time_recursive",
1393 .private = offsetof(struct bfq_group, stats.service_time),
1394 .seq_show = bfqg_print_rwstat_recursive,
1397 .name = "bfq.io_wait_time_recursive",
1398 .private = offsetof(struct bfq_group, stats.wait_time),
1399 .seq_show = bfqg_print_rwstat_recursive,
1402 .name = "bfq.io_merged_recursive",
1403 .private = offsetof(struct bfq_group, stats.merged),
1404 .seq_show = bfqg_print_rwstat_recursive,
1407 .name = "bfq.io_queued_recursive",
1408 .private = offsetof(struct bfq_group, stats.queued),
1409 .seq_show = bfqg_print_rwstat_recursive,
1412 .name = "bfq.avg_queue_size",
1413 .seq_show = bfqg_print_avg_queue_size,
1416 .name = "bfq.group_wait_time",
1417 .private = offsetof(struct bfq_group, stats.group_wait_time),
1418 .seq_show = bfqg_print_stat,
1421 .name = "bfq.idle_time",
1422 .private = offsetof(struct bfq_group, stats.idle_time),
1423 .seq_show = bfqg_print_stat,
1426 .name = "bfq.empty_time",
1427 .private = offsetof(struct bfq_group, stats.empty_time),
1428 .seq_show = bfqg_print_stat,
1431 .name = "bfq.dequeue",
1432 .private = offsetof(struct bfq_group, stats.dequeue),
1433 .seq_show = bfqg_print_stat,
1435 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
1439 struct cftype bfq_blkg_files[] = {
1441 .name = "bfq.weight",
1442 .flags = CFTYPE_NOT_ON_ROOT,
1443 .seq_show = bfq_io_show_weight,
1444 .write = bfq_io_set_weight,
1449 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1451 void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1452 struct bfq_group *bfqg) {}
1454 void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg)
1456 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
1458 entity->weight = entity->new_weight;
1459 entity->orig_weight = entity->new_weight;
1461 bfqq->ioprio = bfqq->new_ioprio;
1462 bfqq->ioprio_class = bfqq->new_ioprio_class;
1464 entity->sched_data = &bfqg->sched_data;
1467 void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) {}
1469 void bfq_end_wr_async(struct bfq_data *bfqd)
1471 bfq_end_wr_async_queues(bfqd, bfqd->root_group);
1474 struct bfq_group *bfq_bio_bfqg(struct bfq_data *bfqd, struct bio *bio)
1476 return bfqd->root_group;
1479 struct bfq_group *bfqq_group(struct bfq_queue *bfqq)
1481 return bfqq->bfqd->root_group;
1484 void bfqg_and_blkg_put(struct bfq_group *bfqg) {}
1486 struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node)
1488 struct bfq_group *bfqg;
1491 bfqg = kmalloc_node(sizeof(*bfqg), GFP_KERNEL | __GFP_ZERO, node);
1495 for (i = 0; i < BFQ_IOPRIO_CLASSES; i++)
1496 bfqg->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT;
1500 #endif /* CONFIG_BFQ_GROUP_IOSCHED */