On 26/01/16 17:06, Andrew Lunn wrote: > On Tue, Jan 26, 2016 at 04:33:11PM -0800, Florian Fainelli wrote: >> On 26/01/16 16:11, Andrew Lunn wrote: >>> PHY devices may only list clause 22, 45, and their PHY identifier >>> values as compatible values. No other compatible strings are allowed. >>> Make this clear in the documentation, and remove examples where >>> make/model compatible strings are listed. >> >> Humm, should not we rather require Ethernet PHY Device Tree nodes to >> have *at least* a "ethernet-phy-ieee802.3-c22" or >> "ethernet-phy-ieee802.3-c45", and any other compatible string which >> further specifies the hardware is also welcome? > > At the moment, "ethernet-phy-ieee802.3-c45" is used, we look for it > and act upon it. "ethernet-phy-ieee802.3-c22" is not used anywhere, > other than my new of_mdiobus_child_is_phy(). Also, for backwards > compatibility with older blobs, we can never assume one or the other > will be present. > > So you are suggesting we change around 200 ethphy nodes to add in > "ethernet-phy-ieee802.3-c22", yet we don't actually do anything with > it?
Well, we do now, since that is one of the results used by of_mdiobus_child_is_phy(), but you are right, this does not scale. What I would prefer seeing though is not removing nodes that have at least two compatible strings, including one that is "ethernet-phy-ieee802.3-c22", but those which have only one, like the marvell ones that you patch, should have either an additional "ethernet-phy-ieee802.3-c22", or none. Does that make sense? -- Florian
