Re: [PATCH 2/2] arm64: dts: allwinner: h6: Add OrangePi 3 LTS DTS

["Jernej Škrabec" <jernej.skrabec@gmail.com>]

Dne torek, 29. april 2025 ob 17:24:18 Srednjeevropski poletni čas je Jernej Škrabec napisal(a):
> Dne torek, 29. april 2025 ob 17:09:22 Srednjeevropski poletni čas je Andrew Lunn napisal(a):
> > On Tue, Apr 29, 2025 at 04:51:59PM +0200, Jernej Škrabec wrote:
> > > Dne ponedeljek, 28. april 2025 ob 14:37:48 Srednjeevropski poletni čas je Andrew Lunn napisal(a):
> > > > On Sat, Apr 26, 2025 at 08:00:49PM +0200, Jernej Škrabec wrote:
> > > > > Dne petek, 25. april 2025 ob 17:34:14 Srednjeevropski poletni čas je Andrew Lunn napisal(a):
> > > > > > > > +&emac {
> > > > > > > > +	pinctrl-names = "default";
> > > > > > > > +	pinctrl-0 = <&ext_rgmii_pins>;
> > > > > > > > +	phy-mode = "rgmii-rxid";
> > > > > > > 
> > > > > > > So relating to what Andrew said earlier today, should this read rgmii-id
> > > > > > > instead? Since the strap resistors just set some boot-up value, but we
> > > > > > > want the PHY driver to enable both RX and TX delay programmatically?
> > > > > > 
> > > > > > Yes.
> > > > > > 
> > > > > > There is a checkpatch.pl patch working its way through the system
> > > > > > which will add warning about any rgmii value other than rgmii-id. Such
> > > > > > values need a comment that the PCB has extra long clock
> > > > > > lines. Hopefully that will make people actually stop and think about
> > > > > > this, rather than just copy broken vendor code.
> > > > > 
> > > > > I spent quite some time working on ethernet support for this board. Once
> > > > > I've found PHY datasheet, I confirmed that there is added delay. So this
> > > > > particular board needs "rgmii-rxid" mode.
> > > > 
> > > > There have been numerous discussions about what these rgmii modes
> > > > mean, because DT developers frequently get them wrong.
> > > > 
> > > > Does the PCB have an extra long clock line in the TX direction? That
> > > > is what rgmii-rxid means, the PCB is providing the TX delay, the
> > > > MAC/PHY pair needs to add the RX delay.
> > > 
> > > While schematic is accessible, AFAIK PCB/gerbers are not, so I can't really
> > > tell how long it is. But without this extra delay, ethernet doesn't work.
> > 
> > You are not adding an extra delay, you are subtracting a
> > delay. 'rgmii-rxid' says the TX delay is implemented by the PCB, hence
> > the PHY does not need to add the delay.
> > 
> > What is normal is that the PCB adds no delays, and the PHY adds the
> > delay for both the RX and the TX. And you represent this with
> > 'rgmii-id'.
> 
> ok, thanks for explanation.
> 
> > 
> > So what you need to find out is, where does the TX delay come from?
> 
> How to do that? Strapping show me this way and testing confirmed it. Not
> sure what more I can do? As a hobbyist, I don't have access to anything more
> than schematic.

I just to be clear, I tested various combinations, including rgmii-id, and it
didn't work, except rgmii-rxid, which matches strapping. Of course Motorcomm
PHY driver took that into account and set registers accordingly.

Best regards,
Jernej