Re: [PATCH] net:stmmac:stmmac_platform:Add snps,xpcs devicetree parsing

[Serge Semin]

Hi Maxime, Jiangfeng

On Fri, Dec 08, 2023 at 09:14:08AM +0100, Maxime Chevallier wrote:
> Hello,
> 
> On Fri, 8 Dec 2023 07:02:19 +0000
> Jiangfeng Ma <Jiangfeng.Ma@xxxxxxxxxxxx> wrote:
> 
> > In order to setup xpcs, has_xpcs must be set to a non-zero value.
> > Add new optional devicetree properties representing this.
> > 
> > If has_xpcs is set to true, then xpcs_an_inband should preferably be
> > consistent with it, Otherwise, some errors may occur when starting
> > the network, For example, the phy interface that xpcs already supports,
> > but link up fails.
> 
> Can you elaborate on why you need this, and on which platform
> especially ? Usually drivers for the various stmmac variants know if
> they have XPCS or not, or can guess it based on some info such as the
> configured PHY mode (dwmac-intel).
> 
> Besides that, there are a few issues with your submission. If DT is the
> way to go (and I don't say it is), you would also need to update the
> bindings to document that property.
> 
> > The types of has_xpcs and xpcs_an_inband are unsigned int,
> > and generally used as flags. So it may be more reasonable to set them to
> > bool type. This can also be confirmed from the type of @ovr_an_inband.
> 
> And this part would go into a separate patch.

In addition to what Maxime already said having DT-bindings adjusted to
fit to the pattern implemented in the software part is a wrong way to
go. The best choice in this case is to add the DW XPCS DT-node to the
DW MAC MDIO/MI bus and then use the "pcs-handle" to inform the MAC
(mainly it's driver) of what PCS-device is actually attached to it.
The series I submitted on this week is exactly about that:
https://lore.kernel.org/netdev/20231205103559.9605-1-fancer.lancer@xxxxxxxxx/
I guess I'll need about a month or so to settle all the comments, but
the solution implemented there will be better than this one really.

-Serge(y)

> 
> Thanks,
> 
> Maxime
>