Re: [PATCH 1/8] arm64: dts: mt8183: add thermal zone node

[Matthias Kaehlcke]

Hi,

On Fri, May 03, 2019 at 04:03:58PM +0800, Hsin-Yi Wang wrote:
> On Thu, May 2, 2019 at 10:43 AM michael.kao <michael.kao@xxxxxxxxxxxx> wrote:
> >
> > Add thermal zone node to Mediatek MT8183 dts file.
> >
> > Signed-off-by: Michael Kao <michael.kao@xxxxxxxxxxxx>
> > ---
> >  arch/arm64/boot/dts/mediatek/mt8183.dtsi | 64 ++++++++++++++++++++++++++++++++
> >  1 file changed, 64 insertions(+)
> >
> > diff --git a/arch/arm64/boot/dts/mediatek/mt8183.dtsi b/arch/arm64/boot/dts/mediatek/mt8183.dtsi
> > index 926df75..b92116f 100644
> > --- a/arch/arm64/boot/dts/mediatek/mt8183.dtsi
> > +++ b/arch/arm64/boot/dts/mediatek/mt8183.dtsi
> > @@ -334,6 +334,67 @@
> >                         status = "disabled";
> >                 };
> >
> > +               thermal: thermal@1100b000 {
> > +                       #thermal-sensor-cells = <1>;
> > +                       compatible = "mediatek,mt8183-thermal";
> > +                       reg = <0 0x1100b000 0 0x1000>;
> > +                       interrupts = <0 76 IRQ_TYPE_LEVEL_LOW>;
> > +                       clocks = <&infracfg CLK_INFRA_THERM>,
> > +                                <&infracfg CLK_INFRA_AUXADC>;
> > +                       clock-names = "therm", "auxadc";
> > +                       resets = <&infracfg  MT8183_INFRACFG_AO_THERM_SW_RST>;
> > +                       mediatek,auxadc = <&auxadc>;
> > +                       mediatek,apmixedsys = <&apmixedsys>;
> > +                       mediatek,hw-reset-temp = <117000>;
> > +                       nvmem-cells = <&thermal_calibration>;
> > +                       nvmem-cell-names = "calibration-data";
> > +               };
> > +
> > +               thermal-zones {
> > +                       cpu_thermal: cpu_thermal {
> > +                               polling-delay-passive = <1000>;
> > +                               polling-delay = <1000>;
> > +
> > +                               thermal-sensors = <&thermal 0>;
> > +                               sustainable-power = <1500>;
> > +                       };
> > +
> > +                       tzts1: tzts1 {
> > +                               polling-delay-passive = <1000>;
> > +                               polling-delay = <1000>;
> > +                               thermal-sensors = <&thermal 1>;
> Is sustainable-power required for tzts? Though it's an optional
> property, kernel would have warning:
> [    0.631556] thermal thermal_zone1: power_allocator:
> sustainable_power will be estimated
> [    0.639586] thermal thermal_zone2: power_allocator:
> sustainable_power will be estimated
> [    0.647611] thermal thermal_zone3: power_allocator:
> sustainable_power will be estimated
> [    0.655635] thermal thermal_zone4: power_allocator:
> sustainable_power will be estimated
> [    0.663658] thermal thermal_zone5: power_allocator:
> sustainable_power will be estimated
> if no sustainable-power assigned.

The property is indeed optional, if it isn't specified IPA will use
the sum of the minimum power of all 'power actors' of the zone as
estimate (see estimate_sustainable_power()). This may lead to overly
agressive throttling, since the nominal sustainable power will always
be <= the requested power.

In my understanding the sustainable power may varies between devices,
even for the same SoC. One could have all the hardware crammed into a
tiny plastic enclosure (e.g. ASUS Chromebit), another might have a
laptop form factor and a metal enclosure (e.g. ASUS C201). Both
examples are based on an Rockchip rk3288, but they have completely
different thermal behavior, and would likely have different values for
'sustainable-power'.

In this sense I tend to consider 'sustainable-power' more a device,
than a SoC property. You could specify a 'reasonable' value as a
starting point, but it will likely not be optimal for all or even most
devices. The warning might even be useful for device makers by
indicating them that there is room for tweaking.

I'm not an expert in the matter though, just happend to look into this
recently :)

Cheers

Matthias