Re: [RFC PATCH 2/2] PCI: rockchip-host: drop wait on PERST# toggle

[Geraldo Nascimento <geraldogabriel@gmail.com>]

On Mon, Nov 10, 2025 at 12:51:49AM +0100, Dragan Simic wrote:
> Hello Geraldo,
> 
> On Wednesday, November 05, 2025 04:55 CET, Geraldo Nascimento <geraldogabriel@gmail.com> wrote:
> > I did some more testing, intrigued by why would a delay of more than
> > 5 ms after the enablement of the power rails trigger failure in
> > initial link-training.
> > 
> > Something in my intuition kept telling me this was PERST# related,
> > and so I followed that rabbit-hole.
> > 
> > It seems the following change will allow the SSD to work with the
> > Rockchip-IP PCIe core without any other changes. So it is purely
> > a DT change and we are able to keep the mandatory 100ms delay
> > after driving PERST# low, as well as the always-on/boot-on
> > properties of the 3v3 power regulator.
> > 
> > This time everything is within the PCIe spec AFAICT, PERST# indeed
> > is an Open Drain signal, and indeed it does requires pull-up resistor
> > to maintain the drive after driving it high.
> > 
> > I'm still testing the overall stability of this, let's hope for the
> > best!
> > 
> > diff --git a/arch/arm64/boot/dts/rockchip/rk3399pro-vmarc-som.dtsi b/arch/arm64/boot/dts/rockchip/rk3399pro-vmarc-som.dtsi
> > index aa70776e898a..1c5afc0413bc 100644
> > --- a/arch/arm64/boot/dts/rockchip/rk3399pro-vmarc-som.dtsi
> > +++ b/arch/arm64/boot/dts/rockchip/rk3399pro-vmarc-som.dtsi
> > @@ -383,13 +383,14 @@ &pcie_phy {
> >  };
> >  
> >  &pcie0 {
> > -	ep-gpios = <&gpio0 RK_PB4 GPIO_ACTIVE_HIGH>;
> > +	ep-gpios = <&gpio0 RK_PB4 (GPIO_ACTIVE_HIGH | GPIO_OPEN_DRAIN)>;
> >  	num-lanes = <4>;
> > -	pinctrl-0 = <&pcie_clkreqnb_cpm>;
> > +	pinctrl-0 = <&pcie_clkreqnb_cpm>, <&pcie_perst>;
> >  	pinctrl-names = "default";
> >  	vpcie0v9-supply = <&vcca_0v9>;	/* VCC_0V9_S0 */
> >  	vpcie1v8-supply = <&vcca_1v8>;	/* VCC_1V8_S0 */
> >  	vpcie3v3-supply = <&vcc3v3_pcie>;
> > +	max-link-speed = <2>;
> 
> FWIW, we shouldn't be enabling PCIe Gen2 here, because it's been
> already disabled for the RK3399 due to unknown errata in the commit
> 712fa1777207 ("arm64: dts: rockchip: add max-link-speed for rk3399",
> 2016-12-16).  It's perfectly reasonable to assume the same for the
> RK3399Pro, which is basically RK3399 packaged together with RK1808,
> AFAIK with no on-package interconnects.

Hi Dragan!

Thanks for the catch, you are correct. But in this case it was just
for my tests and it crept in in the git diff. I wasn't really proposing
to make that change.

Thanks,
Geraldo Nascimento