block: make sure local irq is disabled when calling __blkcg_rstat_flush
Subsystems (3)
PATCH 1/1 block: make sure local irq is disabled when calling __blkcg_rstat_flush
When __blkcg_rstat_flush() is called from cgroup_rstat_flush*() code path, interrupt is always disabled.
When we start to flush blkcg per-cpu stats list in __blkg_release() for avoiding to leak blkcg_gq's reference in commit 20cb1c2fb756 ("blk-cgroup: Flush stats before releasing blkcg_gq"), local irq isn't disabled yet, then lockdep warning may be triggered because the dependent cgroup locks may be acquired from irq(soft irq) handler.
Fix the issue by disabling local irq always.
--- block/blk-cgroup.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
block/blk-cgroup.c +3 -2
@@ -970,6 +970,7 @@ static void __blkcg_rstat_flush(struct blkcg *blkcg, int cpu)struct llist_head *lhead =per_cpu_ptr (blkcg->lhead, cpu);struct llist_node *lnode;struct blkg_iostat_set *bisc, *next_bisc; +unsigned long flags;rcu_read_lock ();
@@ -983,7 +984,7 @@ static void __blkcg_rstat_flush(struct blkcg *blkcg, int cpu)* When flushing from cgroup, cgroup_rstat_lock is always held, so * this lock won't cause contention most of time. */ -raw_spin_lock (&blkg_stat_lock); +raw_spin_lock_irqsave (&blkg_stat_lock, flags);/* * Iterate only the iostat_cpu's queued in the lockless list.
@@ -1009,7 +1010,7 @@ static void __blkcg_rstat_flush(struct blkcg *blkcg, int cpu)blkcg_iostat_update (parent, &blkg->iostat.cur, &blkg->iostat.last); } -raw_spin_unlock (&blkg_stat_lock); +raw_spin_unlock_irqrestore (&blkg_stat_lock, flags); out:rcu_read_unlock (); }
Signature · 1 line
2.40.1
Quoted 45 lines · Ming Lei
When
__blkcg_rstat_flush()is called from cgroup_rstat_flush*() code path, interrupt is always disabled.When we start to flush blkcg per-cpu stats list in
__blkg_release()for avoiding to leak blkcg_gq's reference in commit20cb1c2fb756("blk-cgroup: Flush stats before releasingblkcg_gq"), local irq isn't disabled yet, then lockdep warning may be triggered because the dependent cgroup locks may be acquired from irq(soft irq) handler.Fix the issue by disabling local irq always.
Fixes 20cb1c2fb756 ("blk-cgroup: Flush stats before releasing blkcg_gq")Reported-by Shinichiro Kawasaki <[email protected]>Closes https://lore.kernel.org/linux-block/pz2wzwnmn5tk3pwpskmjhli6g3qly7eoknilb26of376c7kwxy@qydzpvt6zpis/T/#u open hereCc Jay Shin <[email protected]>Cc Tejun Heo <[email protected]>Cc Waiman Long <[email protected]>Signed-off-by Ming Lei <[email protected]>--- block/blk-cgroup.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)block/blk-cgroup.c +2
-1@@ -970,6 +970,7 @@ static void __blkcg_rstat_flush(struct blkcg *blkcg, int cpu)struct llist_head *lhead =per_cpu_ptr (blkcg->lhead, cpu);struct llist_node *lnode;struct blkg_iostat_set *bisc, *next_bisc; +unsigned long flags;rcu_read_lock ();@@ -983,7 +984,7 @@ static void __blkcg_rstat_flush(struct blkcg *blkcg, int cpu)* When flushing from cgroup, cgroup_rstat_lock is always held, so * this lock won't cause contention most of time. */ -raw_spin_lock (&blkg_stat_lock); +raw_spin_lock_irqsave (&blkg_stat_lock, flags);/* * Iterate only the iostat_cpu's queued in the lockless list.
block/blk-cgroup.c static void __blkcg_rstat_flush(struct blkcg *blkcg, int cpu) · line 1016
Quoted 8 lines
block/blk-cgroup.c +1
-1@@ -1009,7 +1010,7 @@ static void __blkcg_rstat_flush(struct blkcg *blkcg, int cpu)blkcg_iostat_update (parent, &blkg->iostat.cur, &blkg->iostat.last); } -raw_spin_unlock (&blkg_stat_lock); +raw_spin_unlock_irqrestore (&blkg_stat_lock, flags); out:rcu_read_unlock (); }
Quoted 10 lines · Ming Lei
When
__blkcg_rstat_flush()is called from cgroup_rstat_flush*() code path, interrupt is always disabled.When we start to flush blkcg per-cpu stats list in
__blkg_release()for avoiding to leak blkcg_gq's reference in commit20cb1c2fb756("blk-cgroup: Flush stats before releasingblkcg_gq"), local irq isn't disabled yet, then lockdep warning may be triggered because the dependent cgroup locks may be acquired from irq(soft irq) handler.[...]
Applied, thanks!
[1/1] block: make sure local irq is disabled when calling __blkcg_rstat_flush
commit: 9c39b7a905d84b7da5f59d80f2e455853fea7217
Best regards,
Signature · 1 line
Jens Axboe
Quoted 11 lines · Ming Lei
When
__blkcg_rstat_flush()is called from cgroup_rstat_flush*() code path, interrupt is always disabled.When we start to flush blkcg per-cpu stats list in
__blkg_release()for avoiding to leak blkcg_gq's reference in commit20cb1c2fb756("blk-cgroup: Flush stats before releasingblkcg_gq"), local irq isn't disabled yet, then lockdep warning may be triggered because the dependent cgroup locks may be acquired from irq(soft irq) handler.Fix the issue by disabling local irq always.
Quoted 8 lines
Fixes 20cb1c2fb756 ("blk-cgroup: Flush stats before releasing blkcg_gq")Reported-by Shinichiro Kawasaki <[email protected]>Closes https://lore.kernel.org/linux-block/pz2wzwnmn5tk3pwpskmjhli6g3qly7eoknilb26of376c7kwxy@qydzpvt6zpis/T/#u open hereCc Jay Shin <[email protected]>Cc Tejun Heo <[email protected]>Cc Waiman Long <[email protected]>Signed-off-by Ming Lei <[email protected]>
Thanks Ming, I confirmed this avoids the WARN I reported. I also ran my test sets and observed no regression. Looks good.