It turns out that the WARN_ON_ONCE() call in css_release_work_fn
introduced by commit ab0312526867 ("cgroup: Show # of subsystem CSSes
in cgroup.stat") is incorrect. Although css->nr_descendants must be
0 when a css is released and ready to be freed, the corresponding
cgrp->nr_dying_subsys[ss->id] may not be 0 if a subsystem is activated
and deactivated multiple times with one or more of its previous
activation leaving behind dying csses.

Fix the incorrect warning by removing the cgrp->nr_dying_subsys check.

Fixes: ab0312526867 ("cgroup: Show # of subsystem CSSes in cgroup.stat")
Closes: https://lore.kernel.org/cgroups/6f301773-2fce-4602-a391-8af7ef00b2fb@redhat.com/T/#t
Signed-off-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cgroup.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
index 601600afdd20..244ec600b4d8 100644
--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c
@@ -5465,7 +5465,14 @@ static void css_release_work_fn(struct work_struct *work)
 			ss->css_released(css);
 
 		cgrp->nr_dying_subsys[ss->id]--;
-		WARN_ON_ONCE(css->nr_descendants || cgrp->nr_dying_subsys[ss->id]);
+		/*
+		 * When a css is released and ready to be freed, its
+		 * nr_descendants must be zero. However, the corresponding
+		 * cgrp->nr_dying_subsys[ss->id] may not be 0 if a subsystem
+		 * is activated and deactivated multiple times with one or
+		 * more of its previous activation leaving behind dying csses.
+		 */
+		WARN_ON_ONCE(css->nr_descendants);
 		parent_cgrp = cgroup_parent(cgrp);
 		while (parent_cgrp) {
 			parent_cgrp->nr_dying_subsys[ss->id]--;
-- 
2.43.5

On Thu, Aug 08, 2024 at 11:22:59PM -0400, Waiman Long wrote:
> It turns out that the WARN_ON_ONCE() call in css_release_work_fn
> introduced by commit ab0312526867 ("cgroup: Show # of subsystem CSSes
> in cgroup.stat") is incorrect. Although css->nr_descendants must be
> 0 when a css is released and ready to be freed, the corresponding
> cgrp->nr_dying_subsys[ss->id] may not be 0 if a subsystem is activated
> and deactivated multiple times with one or more of its previous
> activation leaving behind dying csses.
> 
> Fix the incorrect warning by removing the cgrp->nr_dying_subsys check.
> 
> Fixes: ab0312526867 ("cgroup: Show # of subsystem CSSes in cgroup.stat")
> Closes: https://lore.kernel.org/cgroups/6f301773-2fce-4602-a391-8af7ef00b2fb@redhat.com/T/#t
> Signed-off-by: Waiman Long <longman@redhat.com>

Applied to cgroup/for-6.12.

Thanks.

-- 
tejun