Re: [PATCH net 2/2] net: pcs: xpcs: actively unset DW_VR_MII_DIG_CTRL1_2G5_EN for 1G SGMII

[Vladimir Oltean]

On Wed, Jan 15, 2025 at 02:44:52PM +0000, Russell King (Oracle) wrote:
> On Tue, Jan 14, 2025 at 06:47:21PM +0200, Vladimir Oltean wrote:
> > xpcs_config_2500basex() sets DW_VR_MII_DIG_CTRL1_2G5_EN, but
> > xpcs_config_aneg_c37_sgmii() never unsets it. So, on a protocol change
> > from 2500base-x to sgmii, the DW_VR_MII_DIG_CTRL1_2G5_EN bit will remain
> > set.
> > 
> > Fixes: f27abde3042a ("net: pcs: add 2500BASEX support for Intel mGbE controller")
> > Signed-off-by: Vladimir Oltean <vladimir.oltean@xxxxxxx>
> 
> Reviewed-by: Russell King (Oracle) <rmk+kernel@xxxxxxxxxxxxxxx>
> 
> Thanks!
> 
> I wonder whether, now that we have in-band capabilities, and thus
> phylink knows whether AN should be enabled or not, whether we can
> simplify all these different config functions and rely on the
> neg_mode from phylink to configure in-band appropriately.

I don't understand, many sub-functions of xpcs_do_config() use neg_mode
already.

If you're talking about replacing compat->an_mode with something derived
partially from the neg_mode and partially from state->interface, then in
principle yes, sure, but we will need new neg_modes for clause 73
auto-negotiation (to replace DW_AN_C73), plus appropriate handling in phylink.