linux-kernel - Re: _cpu_down deadlock [was Re: 2.6.19-rc1-mm1]

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 11 Oct 2006 09:39:20 -0700
From:	Andrew Morton <akpm@...l.org>
To:	Neil Brown <neilb@...e.de>
Cc:	Arjan van de Ven <arjan@...radead.org>,
	Michal Piotrowski <michal.k.k.piotrowski@...il.com>,
	Pavel Machek <pavel@....cz>, "Rafael J. Wysocki" <rjw@...k.pl>,
	linux-kernel@...r.kernel.org, Ingo Molnar <mingo@...e.hu>,
	rusty@...tcorp.com.au
Subject: Re: _cpu_down deadlock [was Re: 2.6.19-rc1-mm1]

On Wed, 11 Oct 2006 23:08:21 +1000
Neil Brown <neilb@...e.de> wrote:

> On Wednesday October 11, arjan@...radead.org wrote:
> > 
> > > > blocking_notifier_call_chain is
> > > >         down_read(&nh->rwsem);
> > > >         ret = notifier_call_chain(&nh->head, val, v);
> > > >         up_read(&nh->rwsem);
> > > >
> > > > and so holds ->rwsem while calling the callback.
> > > > So the locking sequence ends up as:
> > > >
> > > >  down_read(&cpu_chain.rwsem);
> > > >  mutex_lock(&workqueue_mutex);
> > > >  up_read(&cpu_chain.rwsem);
> > > >
> > > >  down_read(&cpu_chain.rwsem);
> > > >  mutex_unlock(&workqueue_mutex);
> > > >  up_read(&workqueue_mutex);
> > > >
> > > > and lockdep doesn't seem to like this.  It sees workqueue_mutex
> > > > claimed while cpu_chain.rwsem is held. and then it sees
> > > > cpu_chain.rwsem claimed while workqueue_mutex is held, which looks a
> > > > bit like a class ABBA deadlock.
> > > > Of course because it is a 'down_read' rather than a 'down', it isn't
> > > > really a dead lock.
> > 
> > ok can you explain to me why "down_read" doesn't make this a deadlock
> > while "down" would make it a deadlock? I have trouble following your
> > reasoning.....
> > 
> > (remember that rwsems are strictly fair)
> 
> I see your point.
> 
> While thread A holds just workqueue_mutex,
> thread B takes cpu_chain.rwsem for read then tries to take
> workqueue_mutex and blocks.
> Now thread C tries to get a write lock on cpu_chain.rwsem and blocks
> as well.
> Finally thread A moves on to try to get a read lock on cpu_chain.rwsem
> and this blocks because thread C is waiting for a write lock.
> 
> So A waits on B and C, C waits on B, B waits on A.
> Deadlock.

Except the entire operation is serialised by the the two top-level callers
(cpu_up() and cpu_down()) taking mutex_lock(&cpu_add_remove_lock).  Can
lockdep be taught about that?

> Who do we blame this on?  Are you still the cpu-hot-plug guy Rusty?

It's fun blaming Rusty for stuff, but he can dodge this one with
more-than-usual ease, I'm afraid.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/