[Intel-gfx] [PATCH] [RFC] kernel/cpu: Use lockref for online CPU reference counting

[Joonas Lahtinen]

On ke, 2016-02-17 at 17:37 +0100, Peter Zijlstra wrote:
> On Wed, Feb 17, 2016 at 05:33:51PM +0100, Daniel Vetter wrote:
> > On Wed, Feb 17, 2016 at 05:14:57PM +0100, Peter Zijlstra wrote:
> > > On Wed, Feb 17, 2016 at 05:13:21PM +0100, Daniel Vetter wrote:
> > > > And for context we're hitting this on CI in a bunch of our machines, which
> > > 
> > > What's CI ?
> > 
> > Continuous integration, aka our own farm of machines dedicated to running
> > i915.ko testcases. Kinda like 0day (it does pre-merge testing on the m-l
> > and also post-merge on our own little integration tree), but for just the
> > graphics team and our needs.
> 
> So what patch triggered this new issue? Did cpufreq change or what?

It appeared right after enabling lockdep debugging on the continuous
integration system. So we do not have a history of it not being there.

Taking an another look at my code, it could indeed end up in double-
wait-looping scenario if suspend and initialization was performed
simultaneously (it had a couple of other bugs too, fixed in v2).
Strange thing is, I think that should have been caught by cpuhp_lock_*
lockdep tracking.

So I'll move the discussion to linux-pm list to change the CPUfreq code. Thanks for the comments.

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation