[PATCH v2 08/12] arm64: dts: rockchip: rk3588s-nanopi: remove always-on from vdd_npu_s0 reg

[Joachim Eastwood]

On Sun, 12 Jul 2026 at 20:22, Jonas Karlman <jonas at kwiboo.se> wrote:
>
> Hi Joachim,
>
> On 7/12/2026 8:00 PM, Joachim Eastwood wrote:
> > Hi Jonas,
> >
> > On Sat, 11 Jul 2026 at 01:24, Jonas Karlman <jonas at kwiboo.xyz> wrote:
> >>
> >> Hi Joachim,
> >>
> >> On 7/11/2026 12:11 AM, Joachim Eastwood via B4 Relay wrote:
> >>> From: Joachim Eastwood <joachim.eastwood at gmail.com>
> >>>
> >>> Since the NPU is hooked up on these the always-on isn't needed anymore.
> >>>
> >>> Signed-off-by: Joachim Eastwood <joachim.eastwood at gmail.com>
> >>> ---
> >>>  arch/arm64/boot/dts/rockchip/rk3588s-nanopi.dtsi | 1 -
> >>>  1 file changed, 1 deletion(-)
> >>>
> >>> diff --git a/arch/arm64/boot/dts/rockchip/rk3588s-nanopi.dtsi b/arch/arm64/boot/dts/rockchip/rk3588s-nanopi.dtsi
> >>> index bcb2f5d63a61..463cfeddf270 100644
> >>> --- a/arch/arm64/boot/dts/rockchip/rk3588s-nanopi.dtsi
> >>> +++ b/arch/arm64/boot/dts/rockchip/rk3588s-nanopi.dtsi
> >>> @@ -258,7 +258,6 @@ vdd_npu_s0: regulator at 42 {
> >>>               regulator-max-microvolt = <950000>;
> >>>               regulator-ramp-delay = <2300>;
> >>>               regulator-boot-on;
> >>
> >> Why is boot-on kept?, NPU is likely not needed until OS has started.
> >
> > I am not really sure.
> > Looking at the other rk3588(s) boards, they also kept the boot-on parameter
> > when the always-on was removed.
>
> Ahh, those board should likely also drop the boot-on prop. There should
> really not be any need to force the PMIC to enable this regulator from
> firmware. If it was left intentionally for other boards, maybe there is
> a bug in software that need to be fixed.
>
> >
> >> With PD domain-supply the regulator should be enabled automatically, or
> >> is there some other issue that requires boot firmware to enable this
> >> regulator before OS is started?
> >
> > Maybe some Rockchip people can enlighten us?
> >
> > I haven't had the time to actually test the NPU. I'll try to test it next week.
> > So this patch is based on similar changes done to other RK3588 boards.
>
> Yes, please drop the boot-on prop and test :-)

boot-on doesn't seems to matter on my NanoPi M6. My short test code for the
NPU runs either way. Note that I am running edk2-rk3588 bootloader.

Setting up the enviroment for running stuff on the NPU was less painful
than I feared. Using Mesa teflon and python-tflite-runtime.

I will repost the entire patch set later today.
Thanks for all the feedback so far.


best regards,
Joachim Eastwood