Re: [BUG] TASK_DEAD task is able to be woken up in special condition

[Ingo Molnar <mingo@elte.hu>]

* Oleg Nesterov <oleg@redhat.com> wrote:

> On 01/17, Ingo Molnar wrote:
> >
> > * Yasunori Goto <y-goto@jp.fujitsu.com> wrote:
> >
> > > --- linux-3.2.orig/kernel/exit.c
> > > +++ linux-3.2/kernel/exit.c
> > > @@ -1038,6 +1038,22 @@ NORET_TYPE void do_exit(long code)
> > >
> > >  	preempt_disable();
> > >  	exit_rcu();
> > > +
> > > +	/*
> > > +	 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
> > > +	 * when the following two conditions become true.
> > > +	 *   - There is race condition of mmap_sem (It is acquired by
> > > +	 *     exit_mm()), and
> > > +	 *   - SMI occurs before setting TASK_RUNINNG.
> > > +	 *     (or hypervisor of virtual machine switches to other guest)
> > > +	 *  As a result, we may become TASK_RUNNING after becoming TASK_DEAD
> > > +	 *
> > > +	 * To avoid it, we have to wait for releasing tsk->pi_lock which
> > > +	 * is held by try_to_wake_up()
> > > +	 */
> > > +	smp_mb();
> > > +	raw_spin_unlock_wait(&tsk->pi_lock);
> >
> > Hm, unlock_wait() is really nasty. Wouldnt the adoption of 
> > the -rt kernel's delayed task put logic solve most of these 
> > races?
> 
> How? The problem is that the exiting task can do the last 
> schedule() in TASK_RUNNING state, this breaks the TASK_DEAD 
> logic in finish_task_switch().

Well, but does the -rt kernel suffer from the same race? It can 
generate delays at the exact same place, and can generate much 
longer delays than an SMI, if a high-prio RT task comes along.

So if there's something in the -rt kernel that fixes this race 
we'd like to have that. If the bug is present in the -rt kernel 
then why didn't it ever get triggered? We caught much more 
narrow races in -rt, and very early on in the project.

Thanks,

	Ingo