Re: [PATCH net-next v3 07/11] net: phy: introduce phy_mdiobus_read_mmd()

[Michael Walle <mwalle@kernel.org>]

Am 2023-07-19 01:54, schrieb Andrew Lunn:
>> +static int __phy_mdiobus_read_mmd(struct mii_bus *bus, int phy_addr,
>> +				  enum phy_access_mode access_mode,
>> +				  int devad, u32 regnum)
>> +{
>> +	switch (access_mode) {
>> +	case PHY_ACCESS_C45:
>> +		return __mdiobus_c45_read(bus, phy_addr, devad, regnum);
>> +	case PHY_ACCESS_C22:
>> +		/* ignore return value for legacy reasons */
>> +		mmd_phy_indirect(bus, phy_addr, devad, regnum, false);
>> +
>> +		/* Read the content of the MMD's selected register */
>> +		return __mdiobus_read(bus, phy_addr, MII_MMD_DATA);
>> +	default:
>> +		return -EOPNOTSUPP;
>> +	}
> 
> So this is reading a C45 register space register, otherwise it would
> not be called _mmd and have a devad. So access_mode should really be
> transfer mode. Until now, only transfer mode C45 can be used to access
> C45 register space. The point of this patchset is to add a new
> C45_OVER_C22 transfer mode.

So you suggest to rename access_mode to transfer_mode, right?

> And C22 would should give -EINVAL, since you cannot use plain C22 bus
> transactions to access C45 register space.

We had indirect mmd access before with c22 PHYs before, we could
theoretically fold that into c45-over-c22, but the old indirect
mmd access wasn't checking for error codes and according to Russell
we cannot change that. Honestly, I'd just duplicate the code and
leave the old non-error checking code in __phy_read_mmd() while
__phy_mdiobus_read_mmd() will do error checking and return -EINVAL
with PHY_ACCESS_C22. I had that in one of the first versions but
you suggested to not copy the code :), then this ugly check_rc
thing came. Or __phy_mdiobus_(read,write)_mmd() will have a
check_rc parameter.

-michael