RFD: OF device tree vs. PHY flags.

[David Daney]

I am in the process of planning a conversion of Octeon SOC platform code 
to use the OF device tree in the Linux kernel.

One issue that I have encountered, is that for some boards, we need to 
pass a non-zero flags argument to the phy_attach_direct() method.  The 
value of the flags is board dependent, so it would make some sense to 
encode its value in the device tree itself.  The flags I am interested 
in control the configuration of clocking modes and status LED connections.

I would suggest the following:

o Add a new property to "ethernet-phy" dts bindings called 
"linux,flags".  It would contain a comma separated string of flag names. 
  Something like "led-mode1,clock-mode2".  The semantics of the flag 
names would be interpreted by the PHY driver...

o Add a new function pointer to struct phy_driver: u32 
(*of_parse_flags)(struct phy_device *phydev).  This would parse and 
return the flags value for the "linux,flags" property from the 
device_node associated with the particular PHY device in question.

o Modify of_phy_connect() to do something like the following:

.
.
.
	if (phy->driver && phy->driver->of_parse_flags)
		flags |= phy->driver->of_parse_flags(phy);
.
.
.

o Perhaps add some helper functions to of_mdio.c to assist in parsing 
the "linux,flags" properties string.

o Any extra code in the PHY drivers and struct phy_driver would be 
protected by #ifdef CONFIG_OF

How does this sound?

Any suggestions as for improvements, or better names for things?


Thanks,
David Daney