Re: [PATCH RFC 01/10] dt-bindings: gpu: mali-valhall-csf: add mediatek,mt8196-mali variant

[Nicolas Frattaroli <nicolas.frattaroli@collabora.com>]

Hi Rob,

On Saturday, 6 September 2025 01:26:57 Central European Summer Time Rob Herring wrote:
> On Fri, Sep 05, 2025 at 12:22:57PM +0200, Nicolas Frattaroli wrote:
> > The Mali-based GPU on the MediaTek MT8196 SoC is shackled to its concept
> > of "MFlexGraphics", which in this iteration includes an embedded MCU
> > that needs to be poked to power on the GPU, and is in charge of
> > controlling all the clocks and regulators.
> > 
> > In return, it lets us omit the OPP tables from the device tree, as those
> > can now be enumerated at runtime from the MCU.
> > 
> > Add the mediatek,mt8196-mali compatible, and a performance-controller
> > property which points to a node representing such setups. It's required
> > on mt8196 devices.
> > 
> > Signed-off-by: Nicolas Frattaroli <nicolas.frattaroli@collabora.com>
> > ---
> >  .../bindings/gpu/arm,mali-valhall-csf.yaml         | 36 +++++++++++++++++++++-
> >  1 file changed, 35 insertions(+), 1 deletion(-)
> > 
> > diff --git a/Documentation/devicetree/bindings/gpu/arm,mali-valhall-csf.yaml b/Documentation/devicetree/bindings/gpu/arm,mali-valhall-csf.yaml
> > index a5b4e00217587c5d1f889094e2fff7b76e6148eb..6df802e900b744d226395c29f8d87fb6d3282d26 100644
> > --- a/Documentation/devicetree/bindings/gpu/arm,mali-valhall-csf.yaml
> > +++ b/Documentation/devicetree/bindings/gpu/arm,mali-valhall-csf.yaml
> > @@ -19,6 +19,7 @@ properties:
> >        - items:
> >            - enum:
> >                - rockchip,rk3588-mali
> > +              - mediatek,mt8196-mali
> >            - const: arm,mali-valhall-csf   # Mali Valhall GPU model/revision is fully discoverable
> >  
> >    reg:
> > @@ -53,6 +54,13 @@ properties:
> >    opp-table:
> >      type: object
> >  
> > +  performance-controller:
> > +    $ref: /schemas/types.yaml#/definitions/phandle
> > +    description:
> > +      A phandle of a device that controls this GPU's power and frequency,
> > +      if any. If present, this is usually in the form of some specialised
> > +      embedded MCU.
> 
> We already abuse power-domains binding with both power and performance. 
> There's a performance-domain binding too, but only used on one platform 
> for CPUs (Mediatek too IIRC). Or perhaps you could just point to an 
> empty OPP table. I don't think you have anything new here, so don't 
> invent something new.

Oops, yeah, I forgot about performance-domain already existing. I agree
that it looks like a good fit; iirc I initially disregarded it because
I thought it was an actual heterogenous core cpufreq-y thing I'd be
overloading with new meaning, but I see now that this is not so, and
aside from mediatek, Apple appears to be the only user.

Thanks for the fast review.

Kind regards,
Nicolas Frattaroli

> 
> Rob
>