Re: TX1 mainline GPU/wifi support

[Thierry Reding <thierry.reding-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 1699 bytes --]

Closing the loop with what was discussed on IRC...

On Tue, Aug 02, 2016 at 01:32:31PM +0100, Jack Mitchell wrote:
> Hi,
> 
> I am currently investigating the opportunity to move our TX1 boards over to
> mainline linux as we require some recent kernel features. I have
> successfully booted linux-next and have USB/ethernet working along with
> Docker and a recent systemd. I see that GPU support is somewhat implemented
> but I'm not sure how to integrate the current nvidia userspace with the
> driver. We use CUDA on our boards, and as such I assume we need the nvidia
> kernel blob to be driving the GPU, right?

That's correct. The Nouveau userspace (and kernel driver for that
matter) don't support CUDA at this point in time. I think some work had
been done on that as part of the A44 project, but I don't know exactly
what the status is.

Alex, do you happen to know?

In order to run the proprietary userspace you'd need a copy of the nvgpu
driver (and possibly other dependencies) from downstream. Porting that
to an upstream kernel is, as far as I understand, non-trivial. So the
bottom line is that you're between a rock and a hard place.

Perhaps one alternative is to connect a discrete GPU via PCIe and use
the proprietary driver for 32-bit ARM. I think people have had some
success using that. That's of course not a very nice solution since it
doesn't make use of the Tegra GPU and requires a lot of extra power.

> As a second question, is there any roadmap for the wifi support on the TX1
> devkit?

I have not looked at wifi at all. I think it's connected via SDIO, so it
might be possible to get it to work with very little effort.

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]