Re: [PATCH] drm/i915: Check that the plane points to the pipes framebuffer before enabling

From: Chris Wilson <chris@chris-wilson.co.uk>
To: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: intel-gfx@lists.freedesktop.org
Subject: Re: [PATCH] drm/i915: Check that the plane points to the pipes framebuffer before enabling
Date: Tue, 19 Apr 2011 06:55:57 +0100	[thread overview]
Message-ID: <0d30dc$lta0jo@orsmga001.jf.intel.com> (raw)
In-Reply-To: <20110418085444.2d020b6c@jbarnes-desktop>

On Mon, 18 Apr 2011 08:54:44 -0700, Jesse Barnes <jbarnes@virtuousgeek.org> wrote:
> I like it (though now the comment talks about DSPADDR and leaves
> DSPSURF out in the cold).

D'oh. That'll teach me to try and write a comment to explain a function
first!

> Sounds like our dpms code may be causing trouble for this x11perf run
> perhaps?  What else would cause us to change base addrs during a bench
> run like that?

Right. The early observation is that we don't assign a framebuffer to
the CRTC borrowed for VGA/TV load detection. And I can postulate that we
contrived to access a PTE during a hotplug poll whilst rewriting the GTT
(or something similar to the object following the ringbuffer).

I think we can just attach the fbcon for the purposes of load detection.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre