Re: [PATCH v4 net-next 2/8] net: phylink: introduce generic method to query PHY in-band autoneg capability

[Vladimir Oltean <vladimir.oltean@nxp.com>]

On Tue, Nov 22, 2022 at 09:21:36AM +0000, Russell King (Oracle) wrote:
> > +enum phy_an_inband {
> > +	PHY_AN_INBAND_UNKNOWN		= BIT(0),
> > +	PHY_AN_INBAND_OFF		= BIT(1),
> > +	PHY_AN_INBAND_ON		= BIT(2),
> > +};
> 
> There is another option here:
> 
> - unknown (basically, PHY driver doesn't implement the function or
>   can't report)
> - off (PHY driver knows for certain that in-band isn't used)
> - on (PHY driver knows that in-band is required and must be
>   acknowledged)
> - on-but-not-required (PHY driver knows that in-band can be used, but
>   the PHY has hardware support for timing out waiting for the in-band
>   acknowledgement - Marvell PHYs support this.)
> 
> Maybe the fourth state can be indicated by setting both _OFF and _ON ?
> 
> Given that there's four states, does it make sense for this to be a
> bitfield?

Setting both _OFF and _ON in the capability report would already have
the meaning that it's configurable via a subsequent call to
phy_config_an_inband(). It's really configurable in VSC8514. I suppose
introducing PHY_AN_INBAND_ON_TIMEOUT = BIT(3) could make sense as
another option for the capability, orthogonal to the other 2.

Maybe it would be useful in itself if the MAC cannot support MLO_AN_INBAND,
like the Lynx PCS in 2500base-x, and the PHY only reports PHY_AN_INBAND_ON |
PHY_AN_INBAND_ON_TIMEOUT (hypothetical example). Phylink would pick
PHY_AN_INBAND_ON_TIMEOUT.

Given that I don't have a use case for this, should I add PHY_AN_INBAND_ON_TIMEOUT
to this patch set or should that be done by someone for whom it makes a difference?

The relevant implication for this patch set is the function prototype of
phy_config_an_enabled(struct phy_device *phydev, bool enabled). It
shouldn't take a bool enabled, but an enum phy_an_inband for future
extensibility (and reject/ignore PHY_AN_INBAND_UNKNOWN).