Re: [PATCH 4/5] arm64: dts: allwinner: a527: add EMAC0 to Radxa A5E board

[Andre Przywara <andre.przywara@arm.com>]

On Fri, 25 Apr 2025 04:01:30 +0200
Andrew Lunn <andrew@lunn.ch> wrote:

> > Ah, right, I dimly remembered there was some hardware setting, but your
> > mentioning of those strap resistors now tickled my memory!
> > 
> > So according to the Radxa board schematic, RGMII0-RXD0/RXDLY is pulled
> > up to VCCIO via 4.7K, while RGMII0-RXD1/TXDLY is pulled to GND (also via
> > 4K7). According to the Motorcom YT8531 datasheet this means that RX
> > delay is enabled, but TX delay is not.
> > The Avaota board uses the same setup, albeit with an RTL8211F-CG PHY,
> > but its datasheet confirms it uses the same logic.
> > 
> > So does this mean we should say rgmii-rxid, so that the MAC adds the TX
> > delay? Does the stmmac driver actually support this? I couldn't find
> > this part by quickly checking the code.  
> 
> No. It is what the PCB provides which matters. A very small number of
> PCB have extra long clock lines to add the 2ns delay. Those boards
> should use 'rgmii'. All other boards should use rgmii-id, meaning the
> delays need to be provided somewhere else. Typically it is the PHY
> which adds the delays.
> 
> The strapping should not matter, the PHY driver will override that. So
> 'rgmii-id' should result in the PHY doing the basis 2ns in both
> directions. The MAC DT properties then add additional delays, which i
> consider fine tuning. Most systems don't actually need fine tuning,
> but the YT8531 is funky, it often does need it for some reason.

Ah, many thanks for the explanation, that clears that up! I read something
about the MAC adding delays, which confused me, but what you say now makes
sense.

Thanks!
Andre.