| Message ID | 20230201120609.4151432-1-yukuai1@huaweicloud.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [-next] block, bfq: cleanup 'bfqg->online' | expand |
On Wed 01-02-23 20:06:09, Yu Kuai wrote: > From: Yu Kuai <yukuai3@huawei.com> > > After commit dfd6200a0954 ("blk-cgroup: support to track if policy is > online"), there is no need to do this again in bfq. > > Signed-off-by: Yu Kuai <yukuai3@huawei.com> So I agree this is nice to do but it isn't so simple. BFQ relies on the fact that 'online' is cleared under bfqd->lock so we cannot associate bio in bfq_bio_bfqg() with a bfqg that has already its bfq_pd_offline() function run. Maybe if you set 'online' to false before calling ->pd_offline() things would work fine for BFQ. Honza > --- > block/bfq-cgroup.c | 4 +--- > block/bfq-iosched.h | 2 -- > 2 files changed, 1 insertion(+), 5 deletions(-) > > diff --git a/block/bfq-cgroup.c b/block/bfq-cgroup.c > index b42956ab5550..a35136dae713 100644 > --- a/block/bfq-cgroup.c > +++ b/block/bfq-cgroup.c > @@ -551,7 +551,6 @@ static void bfq_pd_init(struct blkg_policy_data *pd) > bfqg->bfqd = bfqd; > bfqg->active_entities = 0; > bfqg->num_queues_with_pending_reqs = 0; > - bfqg->online = true; > bfqg->rq_pos_tree = RB_ROOT; > } > > @@ -614,7 +613,7 @@ struct bfq_group *bfq_bio_bfqg(struct bfq_data *bfqd, struct bio *bio) > continue; > } > bfqg = blkg_to_bfqg(blkg); > - if (bfqg->online) { > + if (bfqg->pd.online) { > bio_associate_blkg_from_css(bio, &blkg->blkcg->css); > return bfqg; > } > @@ -985,7 +984,6 @@ static void bfq_pd_offline(struct blkg_policy_data *pd) > > put_async_queues: > bfq_put_async_queues(bfqd, bfqg); > - bfqg->online = false; > > spin_unlock_irqrestore(&bfqd->lock, flags); > /* > diff --git a/block/bfq-iosched.h b/block/bfq-iosched.h > index 75cc6a324267..69aaee52285a 100644 > --- a/block/bfq-iosched.h > +++ b/block/bfq-iosched.h > @@ -1009,8 +1009,6 @@ struct bfq_group { > > /* reference counter (see comments in bfq_bic_update_cgroup) */ > refcount_t ref; > - /* Is bfq_group still online? */ > - bool online; > > struct bfq_entity entity; > struct bfq_sched_data sched_data; > -- > 2.31.1 >
Hi, Jan! 在 2023/02/01 21:10, Jan Kara 写道: > On Wed 01-02-23 20:06:09, Yu Kuai wrote: >> From: Yu Kuai <yukuai3@huawei.com> >> >> After commit dfd6200a0954 ("blk-cgroup: support to track if policy is >> online"), there is no need to do this again in bfq. >> >> Signed-off-by: Yu Kuai <yukuai3@huawei.com> > > So I agree this is nice to do but it isn't so simple. BFQ relies on the > fact that 'online' is cleared under bfqd->lock so we cannot associate bio > in bfq_bio_bfqg() with a bfqg that has already its bfq_pd_offline() > function run. > > Maybe if you set 'online' to false before calling ->pd_offline() things > would work fine for BFQ. Yes, you're right. Thanks for the explanation. Kuai
diff --git a/block/bfq-cgroup.c b/block/bfq-cgroup.c index b42956ab5550..a35136dae713 100644 --- a/block/bfq-cgroup.c +++ b/block/bfq-cgroup.c @@ -551,7 +551,6 @@ static void bfq_pd_init(struct blkg_policy_data *pd) bfqg->bfqd = bfqd; bfqg->active_entities = 0; bfqg->num_queues_with_pending_reqs = 0; - bfqg->online = true; bfqg->rq_pos_tree = RB_ROOT; } @@ -614,7 +613,7 @@ struct bfq_group *bfq_bio_bfqg(struct bfq_data *bfqd, struct bio *bio) continue; } bfqg = blkg_to_bfqg(blkg); - if (bfqg->online) { + if (bfqg->pd.online) { bio_associate_blkg_from_css(bio, &blkg->blkcg->css); return bfqg; } @@ -985,7 +984,6 @@ static void bfq_pd_offline(struct blkg_policy_data *pd) put_async_queues: bfq_put_async_queues(bfqd, bfqg); - bfqg->online = false; spin_unlock_irqrestore(&bfqd->lock, flags); /* diff --git a/block/bfq-iosched.h b/block/bfq-iosched.h index 75cc6a324267..69aaee52285a 100644 --- a/block/bfq-iosched.h +++ b/block/bfq-iosched.h @@ -1009,8 +1009,6 @@ struct bfq_group { /* reference counter (see comments in bfq_bic_update_cgroup) */ refcount_t ref; - /* Is bfq_group still online? */ - bool online; struct bfq_entity entity; struct bfq_sched_data sched_data;