Re: [PATCH 5/6] arm64: tegra: Add node for the GB10B GPU on Tegra264
From: Thierry Reding
Date: Wed Nov 18 2026 - 07:49:51 EST
On Wed, Nov 18, 2026 at 09:30:59AM +0900, Mikko Perttunen wrote:
> On Tuesday, November 17, 2026 9:17 PM Thierry Reding wrote:
> > On Thu, Oct 29, 2026 at 12:32:07PM +0900, Mikko Perttunen wrote:
> > > While the integrated GB10B GPU on Tegra264 is on the PCI bus, it
> > > requires a BPMP power domain to be enabled or it won't enumerate on the
> > > bus. Therefore, describe it and the power domain in the device tree to
> > > allow the power domain to be enabled prior to enumeration.
> > >
> > > Signed-off-by: Mikko Perttunen <mperttunen@xxxxxxxxxx>
> > > ---
> > > arch/arm64/boot/dts/nvidia/tegra264.dtsi | 6 ++++++
> > > 1 file changed, 6 insertions(+)
> > >
> > > diff --git a/arch/arm64/boot/dts/nvidia/tegra264.dtsi b/arch/arm64/boot/dts/nvidia/tegra264.dtsi
> > > index 5ab34cd09a3d..4c8b320ecc40 100644
> > > --- a/arch/arm64/boot/dts/nvidia/tegra264.dtsi
> > > +++ b/arch/arm64/boot/dts/nvidia/tegra264.dtsi
> > > @@ -3597,6 +3597,12 @@ pcie@0 {
> > > #address-cells = <3>;
> > > #size-cells = <2>;
> > > ranges;
> > > +
> > > + gpu@0 {
> > > + compatible = "pci10de,2b00";
> > > + reg = <0x00010000 0x0 0x0 0x0 0x0>;
> > > + power-domains = <&bpmp TEGRA264_POWER_DOMAIN_GPU>;
> > > + };
> >
> > My understanding is that the GPU is the only device on this PCI bus, so
> > wouldn't it be simpler to add the power-domains property to the bus' DT
> > node? I guess that's just another interpretation of how this hardware
> > works, and I like how this power control for PCI devices allows
> > individual devices to be turned on before the bus probes (there's a case
> > like this on Jetson Nano where Ethernet needs a GPIO pin to be pulled up
> > in order to start enumerating on the PCI bus, though I don't remember
> > how we ended up solving that, sources don't have any indication of that
> > GPIO being permanently set high or something).
> >
> > Thierry
>
> The bus itself appears (at least to the point of showing up on lspci and
> kernel logs without any errors) to work fine without the power domain,
> so I thought this the correct way to model the hardware. It seems pretty
> straightforward like this anyway.
Agreed, this looks like a nice extension of the PCI enablement code.
I'll wait for the DT maintainers to review the bindings change before
picking this up.
Thierry
Attachment:
signature.asc
Description: PGP signature