Re: [PATCH v2 2/5] dt-bindings: gpu: add a power_model optional properties for MALI

[Caesar Wang]

Rob,

å 2017å07æ18æ 04:07, Rob Herring åé:
> On Mon, Jul 17, 2017 at 04:14:28PM +0800, Caesar Wang wrote:
> > This patch adds the MALI's power-model to set the IPA model to be used
> > for power management.
> What's IPA? India Pale Ale or Intermediate Physical Address?

IPA is intelligent Power Allocator.  (As the ARM introduced on 
https://developer.arm.com/open-source/intelligent-power-allocation)

> > Signed-off-by: Caesar Wang <wxt@xxxxxxxxxxxxxx>
> > ---
> >
> > Changes in v2: None
> >
> >   Documentation/devicetree/bindings/gpu/arm,mali-midgard.txt | 12 ++++++++++++
> >   1 file changed, 12 insertions(+)
> >
> > diff --git a/Documentation/devicetree/bindings/gpu/arm,mali-midgard.txt b/Documentation/devicetree/bindings/gpu/arm,mali-midgard.txt
> > index a461e47..b616e6b 100644
> > --- a/Documentation/devicetree/bindings/gpu/arm,mali-midgard.txt
> > +++ b/Documentation/devicetree/bindings/gpu/arm,mali-midgard.txt
> > @@ -37,6 +37,18 @@ Optional properties:
> >   - operating-points-v2 : Refer to Documentation/devicetree/bindings/power/opp.txt
> >     for details.
> >   
> > +- power_model : Sets power model parameters. Note that this model was designed for the Juno
> > +	        platform, and may not be suitable for other platforms. A structure containing :
> > +	- compatible: Should be arm,mali-simple-power-model
> > +	- dynamic-coefficient: Coefficient, in pW/(Hz V^2), which is multiplied
> > +	  by v^2*f to calculate the dynamic power consumption.
> > +	- static-coefficient: Coefficient, in uW/V^3, which is multiplied by
> > +	  v^3 to calculate the static power consumption.
> > +	- ts: An array containing coefficients for the temperature scaling
> > +	  factor. This is used to scale the static power by a factor of
> > +	  tsf/1000000, where tsf = ts[3]*T^3 + ts[2]*T^2 + ts[1]*T + ts[0],
> > +	  and T = temperature in degrees.
> > +	- thermal-zone: A string identifying the thermal zone used for the GPU
> This can all easily be implied by the compatible string. I'm not
> inclined to accept something Mali specific here.

Isn't  arm,mali-midgard.txt document suit for Mali specific? :-)

> This looks *very* precise, but I'd be surprised if these values are any
> more than magic values (at least the dynamic coef) adjusted until the
> desired power/performance requirements are achieved. To put it another
> way, why don't we have similar values for CPUs?

These value was calculated by running full GPU process.

CPU had the similar value for dtsi.

Say: arch/arm64/boot/dts/rockchip/rk3399.dtsi
        cpu_b0: cpu@100 {
            ...
            dynamic-power-coefficient = <436>;
            ...
        };

-Caesar

> Rob