Re: SoCFPGA ethernet broken

[Andrew Lunn <andrew@lunn.ch>]

On Fri, Oct 16, 2015 at 09:38:37AM -0500, Dinh Nguyen wrote:
> On Fri, 16 Oct 2015, Andrew Lunn wrote:
> 
> > > Another debugging point, the SoCFPGA board has a Micrel ksz9021 PHY attached
> > > to the ethernet port. What I'm seeing is that with 8b63ec1837fa patch, when
> > > the call to ksz9021_config_init() is made both of_node and dev->parent->of_node
> > > are NULL, without the patch the dev->parent->of_node is a valid pointer. Thus
> > > the skew values get programmed to the phy.
> > 
> > Ah!
> > 
> > You have the phy device tree parameters in the wrong place. These are
> > phy paramters, so should really be in the phy node. But
> > socfpga_cyclone5_socdk.dts has them in the MAC node.
> >
> 
> Alright, let me see if I can rework the DTS.

Well, we are not supposed to break device tree bindings. So we should
try to make this work again.
  
> > There is nothing in Documentation/devicetree/bindings/net/micrel.txt
> > which says you are allowed to place them in the MAC node. Obviously
> > the code did allow this, which is what has now broken.
> 
> I was following Documentation/devicetree/bindings/net/micrel-ksz90x1.txt and
> in this document I was following the autodetected PHY example. Did I mis-interpret
> the example?

I was looking at the wrong binding documentation. So yes, it is
documented you can do this.

But i still think it is wrong. These are phy properties, implemented
by the phy, so should be in the phy node. So moving them would be
good. But as i said, we should fix backwards compatibility if
possible.

	Andrew