Re: Random freezing failure with NFS and automount

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Monday, July 04, 2011, Vaidyanathan Srinivasan wrote:
> * Rafael J. Wysocki <rjw@sisk.pl> [2011-07-03 09:07:18]:
> 
> > Hi,
> > 
> > On Tuesday, June 28, 2011, Vaidyanathan Srinivasan wrote:
> > > Hi,
> > > 
> > > I have random freezing failures on my laptop running 2.6.39 kernel.
> > > The laptop has NFS client and automount.  Network could have been
> > > disconnected by the time suspend is attempted, hence nfs client should
> > > fail all operations, just freeze and allow laptop to suspend.
> > > 
> > > I need some help to drill deeper at this log and also suggestions on
> > > config options to try and get more information to help me root cause
> > > this issue.
> > > 
> > > This happens once in 4-5 suspend/resume cycles, does not succeed on
> > > retry, eventually I have to reboot.
> > 
> > This is a tasks freezer failure, ie. the freezing of tasks fails, because
> > one of them refuses to handle signals for 20 s.  This is probably related
> > to waiting on a VFS mutex in the TASK_UNINTERRUPTIBLE state.
> > 
> > We don't handle those cases nicely right now, sorry about that.
> 
> Hi Rafael,
> 
> Thanks for taking a look.  The NFS mount option in hard,intr so
> I would expect an interruptible sleep.  I will take this to file
> system folks and see if they can help.  I will also review my mount
> options to improve the situation.
> 
> When you said we are not handling the situation, what did you mean?

I meant that the freezing fails in those cases.

> We seem to cleanly unfreeze the tasks and return the system to working
> state (though suspend fails).  Maybe we should send some signals and
> try to prod the failing task to get to freeze?  What is needed here to
> improve our framework?

Probably there is a bug (or more bugs) in our error code paths.  That wouldn't
suprpise me too much, because those code paths are not tested very hard ...

Thanks,
Rafael