Re: [PATCH RFC net-next 09/23] net: dsa: lantiq_gswip: add support for SWAPI version 2.3

[Daniel Golle]

On Sun, Aug 17, 2025 at 05:36:13PM +0200, Andrew Lunn wrote:
> On Sat, Aug 16, 2025 at 08:52:42PM +0100, Daniel Golle wrote:
> > Add definition for switch API version 2.3 and a macro to make comparing
> > the version more conveniant.
> > 
> > Signed-off-by: Daniel Golle <daniel@xxxxxxxxxxxxxx>
> > ---
> >  drivers/net/dsa/lantiq_gswip.h | 3 +++
> >  1 file changed, 3 insertions(+)
> > 
> > diff --git a/drivers/net/dsa/lantiq_gswip.h b/drivers/net/dsa/lantiq_gswip.h
> > index 433b65b047dd..fd0c01edb914 100644
> > --- a/drivers/net/dsa/lantiq_gswip.h
> > +++ b/drivers/net/dsa/lantiq_gswip.h
> > @@ -7,6 +7,7 @@
> >  #include <linux/platform_device.h>
> >  #include <linux/regmap.h>
> >  #include <linux/reset.h>
> > +#include <linux/swab.h>
> >  #include <net/dsa.h>
> >  
> >  /* GSWIP MDIO Registers */
> > @@ -93,6 +94,8 @@
> >  #define   GSWIP_VERSION_2_1		0x021
> >  #define   GSWIP_VERSION_2_2		0x122
> >  #define   GSWIP_VERSION_2_2_ETC		0x022
> > +#define   GSWIP_VERSION_2_3		0x023
> > +#define GSWIP_VERSION_GE(priv, ver)	(swab16(priv->version) >= swab16(ver))
> 
> Don't this depend on the endiannes of the CPU?
> 
> It seems like it would be better to make your new version member cpu
> endian, and when writing to it, do le16_to_cpu().

Yes, that does make sense.

> 
> Also, if i remember correctly, you made version a u32. Should it
> really be a u16?

True, it should probably be u16. It would complicate things a bit though
as the (existing, currently supported) MMIO switches built-into the Lantiq
SoCs use 32-bit memory operations to access the 16-bit switch registers,
where the upper 16-bit are always all zero... Hence I got to use
.val_bits = 32 to not end up with bus errors, and also .reg_shift = -2
is needed as each 16-bit register address needs to be multiplied by 4 (ie.
addr =<< 2, which is what .reg_shift = -2 is doing).
So up to now, eg. when passed as a function paramter in existing code paths,
the version has been a u32 type. When using a different type for the version
stored in the priv struct, that would create quite a bit of confusion imho:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/dsa/lantiq_gswip.c?h=v6.16#n2018

Let me know what you think.