Re: [PATCH net-next v4 4/4] net: ftgmac100: Add RGMII delay support for AST2600

[Andrew Lunn <andrew@lunn.ch>]

> Hi Andrew,
> 
> I miss one condition is using fixed-link property.
> In ftgmac100, there are RGMII, NC-SI and fixed-link property.
> On RGMII, we have solution on dedicated PHY, but there is an issue on fixed-link
> property.
> 
> Example on dedicated PHY.
> The driver can pass the "rgmii-id" to tell PHY driver to enable the internal delay on
> PHY side. Therefore, we can force to disable RGMII delay on MAC side.
> But there is not any driver when using fixed-link property, which means
> no body can tell the outside device, like switch or MAC-to-MAC, to enable the internal
> delay on them. Also mean the phy-mode in fixed-link case is not used.
> 
> Therefore, could we ignore the RGMII delay on MAC side when the ftgmac100 driver gets
> the fixed-link property? Just keep the original delay value?

MAC to MAC is one of the edge cases where phy-mode is not great. What
is generally accepted it to use rx-internal-delay-ps and
tx-internal-delay-ps with 2ns and a comment explaining it is a MAC to
MAC link, with this MAC adding the delays.

I suggest you see how messy it is to implement this for 2600. If the
code looks horrible, keep the bootloader delays. For 2700 you have a
clean slate, you can implement all this correctly from the start.

	Andrew