On Wed, Dec 12, 2018 at 09:35:27AM -0500, [email protected] wrote:
> (totally willing to embarrass myself further, this time with respect to
> inconsistency in basic terminology in the kernel files.)
>
> from include/linux/mod_devicetable.h:
>
> /**
> * struct mdio_device_id - identifies PHY devices on an MDIO/MII bus
> * @phy_id: The result of
> * (mdio_read(&MII_PHYSID1) << 16 | mdio_read(&PHYSID2)) & @phy_id_mask
> * for this PHY type
> * @phy_id_mask: Defines the significant bits of @phy_id. A value of 0
> * is used to terminate an array of struct mdio_device_id.
> */
> struct mdio_device_id {
> __u32 phy_id;
> __u32 phy_id_mask;
> };
>
> stop right there ... ignoring the obvious typo in "&PHYSID2" (should
> of course be "&MII_PHYSID2"),
Yes, patch welcome.
the definition of "phy_id" is puzzling, as
> it's defined as the full 32-bit value with the mask applied to it.
> is that right? is the technical definition of phy_id the value *after*
> the phy_id_mask is applied to the register contents? because that
> certainly seems inconsistent with most explanations i've seen.
What might be the source of this confusion is that the lower nibble
mostly means the PHY silicon revision. So:
#define PHY_ID_ASIX_AX88796B 0x003b1841
This is the second revision of the silicon, but the mask
.phy_id_mask = 0xfffffff0,
indicates this driver is happy to work with any revision of the
silicon. The driver writer probably had the second revision of the
silicon, which is why they set the #define to that value.
But some vendors don't use the lower nibble as silicon revision,
e.g. ACME Corp.
Andrew
