[Intel-gfx] GMA 500 Support

[David A. Greene]

On Saturday 26 September 2009 17:07:39 Ben Gamari wrote:

> On Sat, Sep 26, 2009 at 03:39:53PM -0500, David A. Greene wrote:
> > I am interested in getting the GMA 500 (Pouslbo) chipset working in
> > Linux. I've done some small investigating into the (unmaintained) Ubuntu
> > driver. I believe this should get integrated with the intel-gfx effort
> > and be an officially supported driver.
> 
> This doesn't seem like this would make too much sense. The driver at least
> doesn't belong in xv86-video-intel. For one, it seems that something
> went horribly wrong in the writing of the psb driver. There is horrible
> code duplication in the driver and no one has been keeping it up-to-date
> with the recent changes in the stack.

Yes, there is a lot of duplication.  It is in pretty sorry shape.  It also 
uses the TTM interface instead of GEM.  Is TTM even supported
anymore in modern kernels/X drivers?  This to me is the biggest
problem.  Is GEM documented somewhere?  A google search
didn't turn up anything.

> Moreover, the GMA500 hardware is a completely different design from the
> other GMA products.  The core itself was licensed from a third-party IP
> core company (a derivative of the PowerVR SGX design from Imagination
> Technologies). It doesn't look like Imagination is going to release
> documentation or even a reasonable driver unless someone with power
> (e.g. Intel or Dell) demands it of them.

But I found this message on this very mailing list:

http://lists.freedesktop.org/archives/intel-gfx/2009-May/002427.html

So apparently someone knows something about this chip and thinks
it shares at least a small something with i915.

Can anyone clarify?

The open source bits of the existing poulsbo driver expose at least some
register information and other bits.  What's needed for a small bit of
2d acceleration?

> Honestly, I doubt that this will happen. I'm certain that there are
> endless IP issues with the PowerVR core and thus it is unlikely that
> they will ever be able to release adequate documentation.  The best
> situation I am hoping for is to realize they made a mistake in selecting
> PowerVR and use their own design in future platform iterations. Linux
> has a very good chance in the netbook market; it would be a horrible
> shame to see this opportunity slip by because of a lack of usable
> embedded GPU.

Is the PowerVR core for 2d as well?  I'm confused given the above message.

I definitely understand the problems with that core.  The Ubuntu driver
includes a firmware blob.  I'm entirely new to graphics drivers.  Is that
firmware likely dependent on TTM or is it independent of memory management?

Any resources to come up to speed on video driver development?  I'm a
quick study (my background is compilers, probably at least as convoluted
as video drivers.  :) ).

> I hope that I am wrong about the above. From what I can see, the SGX
> > Any documentation on the GMA 500 I can look at?  I don't care about
> > 3D yet.  Getting decent 2D performance is the first step.
> 
> I'm almost certain that there (unfortunately) is no documentation
> available for this chip. If there were, I'm sure we would be farther
> along in supporting it. Even the folks in the Intel Linux graphics group
> do not have access to the documentation.
> 
> Anyways, let's all hope this mess is resolved before long. Please
> correct me if I got anything wrong above; I try to stay informed but
> sometimes I miss things.

Given that Dell is using this platform in some newer Minis, maybe
something can be worked out.  It's a shame that Imagination Tech
is not being cooperative.  Since Intel essentially forces some
netbooks to use this chip (due to restrictions on N-series Atom
marketing[1]), one would hope they are exerting some muscle here.

                             -Dave

[1] http://www.netbookmarket.net/intel-atom-n-series-vs-z-series/
    (speculation, but netbook specs seem to back it up)