Re: [PATCH v8] kernel/fork: beware of __put_task_struct calling context

[Wander Lairson Costa]

On Mon, May 15, 2023 at 3:09 PM Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
>
> On 05/15, Wander Lairson Costa wrote:
> > On Mon, May 15, 2023 at 1:43 PM Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
> > >
> > > Certainly I have missed something...
> > >
> > > but,
> > >
> > > On 05/15, Wander Lairson Costa wrote:
> > > >
> > > > -extern void __put_task_struct(struct task_struct *t);
> > > > +extern void ___put_task_struct(struct task_struct *t);
> > > > +extern void __put_task_struct_rcu_cb(struct rcu_head *rhp);
> > >
> > > I don't understand these renames, why can't you simply put this fix
> > > into put_task_struct() ?
> > >
> >
> > No particular reason, it was just a matter of style and keep the parts simple.
>
> Well, to me a single/simple change in put_task_struct() makes more
> sense, but I won't argue.
>

My initial thought was to break the code in smaller functions, but
maybe just changing put_task_struct() could be better.

>         static inline void put_task_struct(struct task_struct *t)
>         {
>                 if (!refcount_dec_and_test(...))
>                         return;
>
>                 if (IS_ENABLED(PREEMPT_RT) && ...)
>                         return call_rcu(...);
>
>                 ...
>                 __put_task_struct();
>                 ...
>         }
>
> > > > +static inline void __put_task_struct(struct task_struct *tsk)
> > > > +{
> > > ...
> > > > +     if (IS_ENABLED(CONFIG_PREEMPT_RT) && !preemptible())
> > > > +             call_rcu(&tsk->rcu, __put_task_struct_rcu_cb);
> > > > +     else
> > > > +             ___put_task_struct(tsk);
> > > > +}
> > >
> > > did you see the emails from Peter? In particular, this one:
> > >
> > >         https://lore.kernel.org/lkml/20230505133902.GC38236@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> > >
> >
> > I didn't notice the lock_acquire/lock_release part. However, I tested
> > the patch with CONFIG_PROVE_RAW_LOCK_NESTING and there was no warning.
>
> Hmm. I tend to trust the Sebastian's analysis in
>
>         https://lore.kernel.org/all/Y+zFNrCjBn53%2F+Q2@xxxxxxxxxxxxx/
>
> I'll try to look at it later, although I hope Sebastian or Peter
> can explain this before I try ;)
>

The inability to see and reproduce the potential issue is part of my
confusion about addressing this property.

> Oleg.
>