Re: [PATCH] powerpc: add of_get_mac_address() and update fsl_soc.c to use it

[Timur Tabi <timur@freescale.com>]

Segher Boessenkool wrote:
>>> +    return get_property(np, "address", NULL);
>>
>> Since there is so much room for confusion with "address", please at
>> least do some basic sanity checking, like check that it's size is 6
>> bytes and maybe that the multicast bit isn't set.
> 
> Preferably, don't use "address" at all -- there shouldn't

Unfortunately, I really don't have that luxury.  I *must* support 'address'.

> be any trees in the wild that use it.  If there actually
> are some, the should use a quirk to change it to "local-
> mac-address" instead.

The whole point behind checking for 'address' is to support older device trees 
that have only that property.  We have a problem where some installations use 
'address', and if we don't support it here in the kernel, then the kernel won't 
boot.

> This holds in general: workarounds for gross errors in
> device trees should be applied much more selectively
> than is done now, it just doesn't scale this way.

There are trees in the kernel today that have 'address' in them.  Before I can 
fix the trees, though, I need to update U-Boot and the kernel to support 
local-mac-address *without* breaking support for older trees.  That's why this 
patch is written the way it is.

I have patches for U-Boot already out there, although currently none of the 
U-Boot maintainers have indicated a willingness to apply them.

This whole local-mac-address mess needs to be fixed in an iterative process. 
One day, we can remove the check for 'address' from of_get_mac_address(), but 
not yet.

I've implemented all the other suggestion so far.  I *really* need this patch to 
go in as is.

-- 
Timur Tabi
Linux Kernel Developer @ Freescale