Re: [PATCH net-next v1 3/3] net: phy: micrel: Fix forced link mode for KSZ886X switches

[Russell King (Oracle)]

On Wed, Oct 11, 2023 at 03:29:49PM +0200, Alexander Stein wrote:
> Hi Oleksij,
> 
> Am Mittwoch, 11. Oktober 2023, 14:38:56 CEST schrieb Oleksij Rempel:
> > +	if (phydev->autoneg != AUTONEG_ENABLE) {
> > +		ret = genphy_setup_forced(phydev);
> > +		if (ret)
> > +			return ret;
> > +
...
> > +		ret = phy_set_bits(phydev, MII_KSZPHY_CTRL,
> > +				   KSZ886X_CTRL_FORCE_LINK);
> > +		if (ret)
> > +			return ret;
> > +	} else {
...
> > +		ret = phy_clear_bits(phydev, MII_KSZPHY_CTRL,
> > +				     KSZ886X_CTRL_FORCE_LINK);
> > +		if (ret)
> > +			return ret;
> 
> Isn't this call to phy_clear_bits() a fix for autonegotiation mode? This 
> should be a separate patch then.

No, I don't think that is the case. Compare the two paths above, noting
that patch 1 introduces the definition for KSZ886X_CTRL_FORCE_LINK.

If autoneg is disabled, then this bit is then set, which forces the
link. Clearly, if autoneg is then re-enabled, this bit has to be
cleared to allow the effects of the autoneg-disabled path to be undone.

So both of these, the phy_set_bits() and the phy_clear_bits() belong in
the same patch. Splitting them up, so we introduce phy_set_bits() first
will create a regression - which we don't want.

These two belong logically together.

What does concern me, however, is that the autoneg-disabled path avoids
calling genphy_setup_master_slave(), and since establishing which end
of the link is part of the fundamentals of a 1000base-T link, I wonder
whether both paths should still call genphy_config_aneg().

Apart from that, I think the patch is otherwise fine.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!