Re: /proc dcache deadlock in do_exit

[Oleg Nesterov <oleg@tv-sign.ru>]

On 11/27, Eric W. Biederman wrote:
>
> Andrew Morton <akpm@linux-foundation.org> writes:
> 
> > On Tue, 27 Nov 2007 14:20:22 +0100
> > Andrea Arcangeli <andrea@suse.de> wrote:
> >
> >> do_exit->release_task->mark_inode_dirty_sync->schedule() (will never
> >> come back to run journal_stop)
> >
> > I don't see why the schedule() will not return?  Because the task has
> > PF_EXITING set?  Doesn't TASK_DEAD do that?
> 
> Yes, why do we not come back from schedule?
> 
> If we are not allowed to schedule after setting PF_EXITING before
> we set TASK_DEAD that entire code path sounds brittle and
> error prone.

Yes, it is fine to schedule after release_task(). As Eric pointed out, we
don't race with call_rcu(delayed_put_task_struct), scheduler has another
reference

	dup_task_struct:
		/* One for us, one for whoever does the "release_task()" (usually parent) */
		atomic_set(&tsk->usage,2);



However, with CONFIG_HOTPLUG_CPU we do have the problem here, but this is
off-topic. Preemption is fine, but deactivate_task() is not. We can't migrate
the deactivated released task from the dead CPU.

	migrate_live_tasks() can't find the task after __unhash_process()

	migrate_dead_tasks() doesn't see it after deactivate_task().

And afaics try_to_wake_up() doesn't necessary change task_cpu() if it is
offline.

No? But again, this is offtopic even if I am right.

Oleg.