Re: 2.6.17-rc4-mm3-lockdep BUG: possible deadlock detected!

[Takashi Iwai <tiwai@suse.de>]

At Mon, 29 May 2006 11:47:49 -0700,
Andrew Morton wrote:
> 
> On Mon, 29 May 2006 20:32:47 +0200
> "Michal Piotrowski" <michal.k.k.piotrowski@gmail.com> wrote:
> 
> > I get this with Ingo's lockdep patch from
> > http://people.redhat.com/mingo/generic-irq-subsystem/
> > 
> > ====================================
> > [ BUG: possible deadlock detected! ]
> > ------------------------------------
> > modprobe/592 is trying to acquire lock:
> >  (&grp->list_mutex){----}, at: [<fd9ee555>]
> > snd_seq_port_connect+0xc0/0x337 [snd_seq]
> > 
> > but task is already holding lock:
> >  (&grp->list_mutex){----}, at: [<fd9ee4fb>]
> > snd_seq_port_connect+0x66/0x337 [snd_seq]
> > 
> > which could potentially lead to deadlocks!
> > 
> > other info that might help us debug this:
> > 1 locks held by modprobe/592:
> >  #0:  (&grp->list_mutex){----}, at: [<fd9ee4fb>]
> > snd_seq_port_connect+0x66/0x337 [snd_seq]
> 
> yes,
> 
>         down_write(&src->list_mutex);
>         down_write(&dest->list_mutex);
> 
> I wonder if there's anything which prevents another task from concurrently
> coming in and trying to perform the opposite connection and causing a
> deadlock.

I'm not 100% sure but I don't think the deadlock occurs because the
object has actually two locks, namely,

	port.c_src.list_mutex
	port.c_dst.list_mutex

so the above is equivalent with

	down(port_A.c_src.list_mutex);
	down(port_B.c_dst.list_mutex);

When the opposite connection is performed concurrently, different
locks are taken:

	down(port_B.c_src.list_mutex);
	down(port_A.c_dst.list_mutex);

In the code, there is another place that using down_write() (in
clear_subscriber_list()), but it's already after closing accesses to
that port, hence there can be no race there.


Takashi