On Thu, 19 May 2022 22:13:47 -1000 Tejun Heo wrote: > On Thu, May 19, 2022 at 04:26:51PM -0700, Tadeusz Struk wrote: > > On 5/19/22 04:23, Hillf Danton wrote: > > > On Wed, 18 May 2022 09:48:21 -0700 Tadeusz Struk wrote: > > > > On 4/22/22 04:05, Michal Koutny wrote: > > > > > On Thu, Apr 21, 2022 at 02:00:56PM -1000, Tejun Heo<tj@xxxxxxxxxx> wrote: > > > > > > If this is the case, we need to hold an extra reference to be put by the > > > > > > css_killed_work_fn(), right? > > > > > > That put could trigger INIT_WORK in css_release() and warning [1] > > > on init active (active state 0) object OTOH as the same > > > css->destroy_work is used in both kill and release pathes. > > Hmm... wouldn't the extra reference keep release from happening? Hm...Hm... have difficulty following up given the risk of memleak without release. > > > Will this help if there would be two WQs, one for the css_release path > > and one for the rcu_work? Unlikely as adding one more queue does not help INIT_WORK. Hillf > > > > diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c > > index adb820e98f24..a4873b33e488 100644 > > --- a/kernel/cgroup/cgroup.c > > +++ b/kernel/cgroup/cgroup.c > > @@ -124,6 +124,7 @@ DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem); > > * which may lead to deadlock. > > */ > > static struct workqueue_struct *cgroup_destroy_wq; > > +static struct workqueue_struct *cgroup_destroy_rcu_wq; > > I don't understand why this would help. Care to elaborate? > > Thanks. > > -- > tejun
