Re: [PATCH] drm/i915: fix regression after clock gating init split

[Jesse Barnes <jbarnes@virtuousgeek.org>]

On Sat, 28 May 2011 14:26:28 +1000
Jason Stubbs <jasonbstubbs@gmail.com> wrote:

> On Sat, 28 May 2011 05:44:11 Keith Packard wrote:
> > On Fri, 27 May 2011 23:44:19 +1000, Jason Stubbs <jasonbstubbs@gmail.com> 
> wrote:
> > > From: Jason Stubbs <jasonbstubbs@gmail.com>
> > > 
> > > However, there are some code paths, notably IS_PINEVIEW(dev), where 
> > > init_clock_gating is not set and not needed.
> > 
> > Looks like Pineview should be using the gen3_init_clock_gating function.
> 
> Yep, you are right. Adding printk()s to the original implementation confirmed 
> that - at least on my hardware. I should have known to check that...
> 
> There are two other code paths where an init_clock_gating function isn't 
> specified though. The specific code paths are:
> 
> HAS_PCH_SPLIT() && !IS_GEN5() && !IS_GEN6() && !IS_IVYBRIDGE()
> 
> Looking at the original intel_enable_clock_gating() function, there would
> have been some intialization done for this case.
> 
> IS_GEN4() && !IS_CRESTLINE() && !IS_BROADWATER()
> 
> It looks like this would have gone into the final "no gating match" if/else 
> branch.
> 
> If both of the above don't happen in the real world, then I guess it's fine as 
> is. If either are possible, then they need to be fixed too.
> 
> I'll just (re)send a patch for the IS_PINEVIEW() case and leave the above to
> somebody who knows better than I...

Yeah, the above combos aren't valid, so I don't think we need to handle
them.

Thanks for the Pineview fix!

-- 
Jesse Barnes, Intel Open Source Technology Center