[RFC] Inconsistency in MDIO ioctl behaviour

[Ben Hutchings <bhutchings@solarflare.com>]

While comparing the various implementations of SIOC{G,S}MIIREG and
SIOCGMIIPHY, I found some differences in behaviour that could make some
applications unreliable across different drivers.

1. Implementations of SIOCGMIIPHY must set phy_id to the PHY's MDIO
address (PRTAD) or to a dummy address if there is no real MDIO bus.
Many implementations, including the generic ones (mii, phylib,
pci-skeleton and now mdio) then proceed as for SIOCGMIIPHY.  Others
return without accessing any registers.  Which behaviour is right?

2. Implementations of SIOC{G,S}MIIREG that do not support access to
arbitary MDIO addresses handle non-matching values of phy_id in at least
three different ways:
(a) ignore it and always use the expected PHY address
(b) ignore writes and return a dummy value for reads
(c) fail
I favour behaviour (c) but I'm not sure what the error code should be.
ENODEV, EINVAL or EIO?

Ben.

-- 
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.