[RFC PATCH v2 net-next 02/15] phy: introduce the PHY_MODE_ETHTOOL mode for phy_set_mode_ext()

[Florian Fainelli]

On 9/23/23 06:48, Vladimir Oltean wrote:
> In networking, we have 2 distinct data types:
> 
> - phy_interface_t describes the link between a MAC (or MAC-side PCS) and
>    an attached PHY or SFP cage
> 
> - enum ethtool_link_mode_bit_indices describes the link between a local
>    PHY and a remote PHY (for example, gigabit RJ45 twisted copper pairs)
> 
> Currently, phy_set_mode_ext(PHY_MODE_ETHERNET) takes arguments of the
> phy_interface_t type, and there is no way to pass an argument of the
> enum ethtool_link_mode_bit_indices type. The new PHY_MODE_ETHTOOL
> intends to address that.
> 
> It is true that there is currently some overlap between these data
> types, namely:
> 
>   phy_interface_t                enum ethtool_link_mode_bit_indices
>   -----------------------------------------------------------------
>   PHY_INTERFACE_MODE_10GKR       ETHTOOL_LINK_MODE_10000baseKR_Full_BIT
>   PHY_INTERFACE_MODE_1000BASEKX  ETHTOOL_LINK_MODE_1000baseKX_Full_BIT
> 
> but those overlaps were deemed to be mistakes, and PHY-to-PHY link modes
> should only be added to ethtool_link_mode_bit_indices going forward.
> Thus, I believe that the distinction is necessary, rather than hacking
> more improper PHY modes. Some of the PHY-to-PHY link modes which may be
> added in the future (to ethtool_link_mode_bit_indices and not to
> phy_interface_t) are:
> 
> 	ETHTOOL_LINK_MODE_100000baseKP4_Full_BIT
> 	ETHTOOL_LINK_MODE_100000baseCR10_Full_BIT
> 	ETHTOOL_LINK_MODE_25000baseKR_S_Full_BIT
> 	ETHTOOL_LINK_MODE_25000baseCR_S_Full_BIT
> 
> One user of PHY_MODE_ETHTOOL will be the MTIP backplane AN/LT + Lynx
> SerDes PHY combo, where the backplane autoneg protocol (IEEE 802.3
> clause 73) selects the operating PHY-to-PHY link mode.
> 
> There are electrical differences between the PHY-to-PHY backplane link
> modes (like ETHTOOL_LINK_MODE_10000baseKR_Full_BIT) and their
> non-backplane counterparts (like PHY_INTERFACE_MODE_10GBASER), namely
> the number of TX signal equalization taps and their configurability.
> This further justifies distinguishing between them in the generic PHY
> API.
> 
> Signed-off-by: Vladimir Oltean <vladimir.oltean at nxp.com>
> ---
> v1->v2: rename PHY_MODE_ETHERNET_PHY to PHY_MODE_ETHTOOL at Russell's
> suggestion
> 
>   include/linux/phy/phy.h | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/phy/phy.h b/include/linux/phy/phy.h
> index 6be348f1fa0e..72ef4afcda81 100644
> --- a/include/linux/phy/phy.h
> +++ b/include/linux/phy/phy.h
> @@ -39,6 +39,7 @@ enum phy_mode {
>   	PHY_MODE_UFS_HS_B,
>   	PHY_MODE_PCIE,
>   	PHY_MODE_ETHERNET,
> +	PHY_MODE_ETHTOOL,

Not feeling very comfortable with using ETHTOOL here because that is a 
Linux sub-subsystem name as opposed to the other enumeration values 
which are electrical modes of operation and/or industry standards names.

How about PHY_MODE_ETHERNET_EXPLICIT or PHY_MODE_ETHERNET_LINKMODE?
-- 
Florian