Dear Philippe De Muyter,
In message 20100220235945.GA15901@frolo.macqel you wrote:
Complain about this to the Linux architecture maintainers - there have been many and longf discussions about this before.
Do you mean lkml or some other mailing list or individual ?
I had especially the ARM list and RMK in mind.
Today we consider the device tree to be the Right Thing (TM) to pass such information to the kernel, and more and more architectures use this method.
AFAIK, device trees are powerpc-specific and my board is coldfire based.
Several other architecture use the device tree as well, and more are following. For example, an adaption for ARM is on the way right now.
As long as it's not available for your architecture, the most straightforward way is to pass an "ethaddr=..." argument on the kernel command line. Drivers can pick it up easily there. Just don't expect that such Linux driver code will be accepted for mainline.
I don't like that method because it can easily be forgotten by someone changing bootargs for some other reason. IMO bootargs should only be used for configuration choices.
Well, that's mostly a matter of taste. At least it's an architecutre independent, standardized way to pass information.
I'd rather go for an additional parameter given by u-boot to linux with the address of the used u-boot environment, passed the same way that the address of bd_info is given to linux (for coldfire's, that's on the stack)
i. e. you create yet another, non-standard solution - exactly what you complained about originally.
Best regards,
Wolfgang Denk