[Linux-stm32] [PATCH net-next 1/2] dt-bindings: net: Add Siflower SF19A2890 GMAC

[Andrew Lunn]

On Thu, Nov 05, 2026 at 03:57:45PM -0500, Joseph Telaak wrote:
> > Do you have access to another board which has a PHY connected, not a
> > switch?
> 
> No. The GL-SFT1200 is the only SF19A2890 board I have, and the only
> other one I know of, the vendor's evaluation board in OpenWrt, also has
> a fixed-link on this port.
> 
> > There are unexplained delays here. Is it the switch, this MAC, or the
> > PCB?
> >
> > We need to understand if tx-internal-delay 0 really means 0, or it
> > actually means 1065ps.
> 
> I can't tell that from this board. The sweeps measure the sum of the
> MAC's delay line, the switch and the PCB, and I have no documentation
> of the line's own delay at step 0.
> 
> What they do show, in the MAC's steps: the windows repeat every 8 ns,
> which puts the start of both eyes about 4 steps (0.2 ns) below 0. TX
> ends at step 51 and RX at step 28. So with every delay at 0 the link
> works, but right at the early edge of both eyes.
> 
> Until that is known, how would you like v2 to describe this board?

The delay value makes no sense to me. It probably works because
something is broken somewhere, either the MAC or the switch.

We have to be careful with what we accept, because we generally
consider DT blobs ABI, old blobs need to work with new kernels.

When we find out what is actually broken and need to fix it, we would
normally have to do that without break running boards. And that is
often very difficult and messy.

So i think we need a get out from backwards compatibility. We need to
be able to break running boards with old DT blobs, in the future when
we fix whatever is broken. Maybe the best way to do that it make the
MAC glue driver and the DSA driver depend on BROKEN.

    Andrew