Re: [PATCH] usb: onboard_dev: add the Rockchip rk1808

On Fri, Aug 28, 2026 at 02:14:22PM +0200, Heiko Stübner wrote:

Hi Rudi,

Am Freitag, 28. August 2026, 12:55:56 Mitteleuropäische Sommerzeit schrieb Rudi Heitbaum:

quoted

On Thu, Aug 27, 2026 at 03:25:23PM +0200, Heiko Stuebner wrote:

quoted

Am Montag, 24. August 2026, 04:52:02 Mitteleuropäische Sommerzeit schrieb Rudi Heitbaum:

quoted

The rk1808 is the npu die in an rk3399pro package. It is wired to one
of the usb3 controllers rather than to an external port, and it needs
its reference clock running before it will boot, which is exactly what
this driver exists to do for onboard devices.

It takes no supply from the driver: the rails are sequenced elsewhere.
Only the clock matters here, and that is already optional and handled
generically.

the "elsewhere" might need a bit more explanation to verify it's the right
implementation. Even if the regulators are always-on or something, we
still want the supplies defined - for future uses.

As follow up on the bringup of the Rock-Pi-N10, the proposed change to the dtsi
for the rk3399pro-vmarc-som is here, but needs the provision of the usb first.
https://github.com/heitbaum/linux/commit/6b114b6fff4022c330003d7f70ce09915ffd774f

I might have formulated that strangly, what I meant was, who is enabling
the power supplies.

The "NPU Power" part of any rk3399pro schematics (should) list 5 power
supplies for the NPU block itself:

NPU_LOGIC_VDD
    --> regulator likely named NPU_VDD_LOG_S0
        --> enabled by NPU_PWREN_2 gpio fixed regulator

NPU_EFUSE_VQPS
    --> regulator likely named NPU_VCC1V8_EFUSE
        --> enabled by NPU_EFUSE_EN_H gpio fixed regulator

NPU_CORE_VDD
    --> regulator likely named NPU_VDD_S0
        --> TCS4525 regulator with voltage being set via the i2c controller
             inside the NPU - but enabled by the NPU_PWREN_6 gpio from
             core RK3399.

NPU_CPU_VDD
    --> regulator likely named NPU_VDD_CPU_S0
        --> enabled by NPU_PWREN_4 gpio fixed regulator

NPU_ADC_AVDD_1V8
    --> regulator likely named NPU_VCC_1V8_S3
        --> enabled by NPU_PWREN_3 gpio fixed regulator

Naming might vary, I found just some EVB schematics.

Someone needs to enable these, so the question was, who is doing that?

Hi Heiko,

I hadn't looked in to these from the mainline viewpoint. I had been
using an adapted vendor script based on the radxa npu_powerctl binary
which reimplements the npu_reset sequence over GPIO.

This has been the sequence:

gpio0 line 4   NPU_PWREN_1   -> 1, 2 ms
gpio0 line 10  NPU_PWREN_2   -> 1, 2 ms
gpio0 line 11  NPU_PWREN_3   -> 1, 2 ms
gpio1 line 22                -> 1, 2 ms
gpio1 line 23                -> 1, 2 ms
gpio1 line 24  NPU_PWREN_6   -> 1, 25 ms
gpio1 line 0                 -> 1  (reset release)

I have the schematics and the working N10 here, so let me look into
describing these in the .dts and update the onboard_usb_dev.h.

Would you prefer that once I have a working and tested dts description of
the rk1808 - submit the onboard_usb_dev.h and rk3399pro-vmarc-som.dtsi
changes as a patchset - or just update the onboard_usb_dev.h as a v2?

Regards
Rudi

Thanks
Heiko