Re: [linux-pm] pm-hibernate : possible circular locking dependencydetected

[Peter Zijlstra]

On Mon, 2009-04-06 at 21:58 +0200, Rafael J. Wysocki wrote:
> On Monday 06 April 2009, Alan Stern wrote:
> > On Mon, 6 Apr 2009, Gautham R Shenoy wrote:
> > 
> > > > If I understand correctly it isn't really a deadlock scenario, but it
> > > > is a lockdep violation.  The violation is:
> > > > 
> > > > 	The pci_device_probe() path 2) proves that dpm_list_mtx [4] can
> > > > 	be acquired while cpu_hotplug.lock [3] is held;
> > > > 
> > > > 	The hibernate() path 3) proves that cpu_hotplug.lock [3] can be
> > > > 	acquired while dpm_list_mtx [4] is held.
> > > > 
> > > > The two pathways cannot run simultaneously (and hence cannot deadlock) 
> > > > because the prepare() stage of hibernation is supposed to stop all 
> > > > device probing.  But lockdep will still report a problem.
> > > 
> > > Thanks for clarifying this Alan. I guess it boils down to teaching
> > > lockdep about this false-positive.
> > 
> > Or else changing the code somehow to avoid the violation completely.  
> > But I have no idea how...  And AFAIK, teaching lockdep about special 
> > cases like this is not so easy to do.
> 
> Yeah, I've just wanted to ask about that.  Peter, how can we do it?

I think it would come down to modeling that blocking of probes as a lock
or something -- because that's basically what it is.

So on the regular probe path, take a read lock of this lock, and on the
suspend path take it as write or something.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/