[PATCH net-next v2 8/8] dt-bindings: net: sparx5: document RGMII MAC delays
Daniel Machon
daniel.machon at microchip.com
Tue Nov 19 03:43:04 PST 2024
> > Hi Andrew,
> >
> > > > The lan969x switch device supports two RGMII port interfaces that can be
> > > > configured for MAC level rx and tx delays.
> > > >
> > > > Document two new properties {rx,tx}-internal-delay-ps. Make them
> > > > required properties, if the phy-mode is one of: rgmii, rgmii_id,
> > > > rgmii-rxid or rgmii-txid. Also specify accepted values.
> > >
> > > This is unusual if you look at other uses of {rt}x-internal-delay-ps.
> > > It is generally an optional parameter, and states it defaults to 0 if
> > > missing, and is ignored by the driver if phy-mode is not an rgmii
> > > variant.
> >
> > Is unusual bad? :-)
>
> Depends. Having a uniform usage is good, it causes less confusion. But
> strict enforcement also has its plus side.
>
> > I thought that requiring the properties would make
> > misconfigurations (mismatching phy-modes and MAC delays) more obvious,
> > as you were forced to specify exactly what combination you want in the
> > DT. Maybe not. I can change it, no problem.
>
> Do these ports only support RGMII? The general pattern is that ports
> supporting RGMII also support other modes, GMII, MII, rev-GMII,
> rev-MII etc. For these other modes RGMII delays are meaningless. The
> general pattern is that they are allowed in DT, but are just ignored.
RGMII and RMII.
>
> If the LAN969x ports only support RGMII, and you are enforcing the
> four RGMII modes in DT, you could also enforce the delays are present
> and only have all allowed values. But i would not have the enforcement
> any more strict than the other ports. Do you enforce the phy-modes for
> the ports with a PCS?
No, we do not enforce that in the DT. For the PCS ports, you can specify
whatever phy-mode in the DT, and if that phy-mode is not advertised in
the driver, it will just be rejected.
I decided to go ahead with v3 (which needs to be reposted when net-next
opens), where the properties are not required.
>
> Andrew
Thanks for your feedback.
/Daniel
More information about the linux-arm-kernel
mailing list