Re: [PATCH net-next v3 7/8] net: dsa: lantiq_gswip: store switch API version in priv

From: Daniel Golle
Date: Thu Aug 21 2025 - 06:51:13 EST


On Thu, Aug 21, 2025 at 04:58:23AM +0200, Andrew Lunn wrote:
> > > + priv->version = le16_to_cpu((__le16 __force)version);
> >
> > I've researched this a bit more and came to the conclusion that while the
> > above works fine because all Lantiq SoCs with built-in switch are
> > big-endian machines it is still wrong.
> > I base this conclusion on the fact that when dealing with more recent
> > MDIO-connected switches (MaxLinear GSW1xx series) the host endian doesn't
> > play a role in the driver -- when dealing with 16-bit values on the MDIO
> > bus, the bus abstraction takes care of converting from/to host endianess.
>
> I agree that all MDIO bus registers are host endian, 16 bit. The shift
> register in the hardware is responsible for putting the bits on the
> wire in the correct order for MDIO.
>
> > Hence I believe this should simply be a swab16() which will always result
> > in the version being in the right byte order to use comparative operators
> > in a meaningful way.
>
> How is this described in the datasheet? And is version special, or do
> all registers need swapping?

The (anyway public) datasheets I have access to don't describe the VERSION
register at all. In the existing precompiler macros, however, you can see
that most-significant and least-significant byte are swapped. REV is
more significant than MOD:

#define GSWIP_VERSION 0x013
#define GSWIP_VERSION_REV_SHIFT 0
#define GSWIP_VERSION_REV_MASK GENMASK(7, 0)
#define GSWIP_VERSION_MOD_SHIFT 8
#define GSWIP_VERSION_MOD_MASK GENMASK(15, 8)
#define GSWIP_VERSION_2_0 0x100
#define GSWIP_VERSION_2_1 0x021
#define GSWIP_VERSION_2_2 0x122
#define GSWIP_VERSION_2_2_ETC 0x022

Now I'd like to add
#define GSWIP_VERSION_2_3 0x023

and then have a simple way to make features available starting from a
GSWIP_VERSION. Now in order for GSWIP_VERSION_2_3 to be greater than
GSWIP_VERSION_2_2, and GSWIP_VERSION_2_1 to be greater than
GSWIP_VERSION_2_0, the bytes need to be swapped.

I don't think the vendor even considered any specific order of the two
bytes but just defines them as separate 8-bit fields, considering it a
16-bit unsigned integer is my interpretation which came from the need for
comparability.