Re: [5/5] ARM: tegra: jetson-tk1: enable GK20A GPU

[Thierry Reding]

On Thu, Sep 25, 2014 at 12:07:01PM -0600, Stephen Warren wrote:
> On 09/25/2014 10:41 AM, Thierry Reding wrote:
> >On Thu, Sep 25, 2014 at 09:48:01AM -0600, Stephen Warren wrote:
> >>On 09/25/2014 07:27 AM, Sjoerd Simons wrote:
> >>>Playing a bit with todays linux-next on my jetson, it seems this patch is
> >>>still required for enabling the GPU. Is there anything blocking it (firmware
> >>>not available yet in liux-firmware?)
> >>
> >>I think initially I was waiting for the DRM patch "drm/nouvea: support for
> >>probing platform devices" to be applied, but it looks like that's been
> >>applied already, so only patches 4 and 5 in this series are still
> >>outstanding.
> >>
> >>Alex, wasn't there also some issue where the VPR register had to be
> >>programmed, and if it wasn't there'd be a hang when the GPU registers were
> >>touched? If we've added code to Nouveau/tegradrm to detect that and avoid
> >>the problem, then I guess we can commit these last two patches for 3.19. A
> >>resend after the 3.18 merge window might help.
> >
> >A patch that programs VPR was merged into U-Boot (though I don't think
> >it's made it into master yet). I'm not sure we can reasonably check for
> >that in Nouveau, given that the register is somewhere completely
> >unrelated. In fact I think the U-Boot patch was triggered by some
> >discussion about how to solve this and it was decided that it shouldn't
> >be done in the kernel, but U-Boot should set it up.
> >
> >That said, perhaps one solution would be to make U-Boot enable the gk20a
> >device if it's set up the VPR and disable it otherwise?
> 
> For that to work, we'd need the DT to say status="disabled" by default for
> the GPU, and for the fixed U-Boot (and indeed every other bootloader...) to
> enable the GPU node. This would allow people with old versions of U-Boot (or
> other bootloaders) to continue to boot. This means bootloaders would only
> have to set status="okay", but never have to set status="disabled", which at
> least simplifies them a tiny bit.

Sounds like a reasonable requirement on bootloaders to me. If it's clear
that the device will not work without an initialized VPR region, part of
the "boot protocol" should be for the bootloader to tell the kernel when
it's safe to enable the GPU.

Thierry

Attachment: pgpUfhkuhr0ni.pgp
Description: PGP signature