Re: [pull-sys940x 3/4] genmac: Replace RANDOM_MAC in network/interfaces with a randomly generated MAC

[William Mills <wmills@ti.com>]

On 02/02/2012 12:19 PM, Darren Hart wrote:
> On 02/02/2012 06:09 AM, William Mills wrote:
>> So on a non-volatile r/w filesystem this will assign a new random mac to
>> each interface on the first boot and reuse that MAC on each subsquent boot.
> Correct.
>
>> On a volatile r/w filesystem this will use a new MAC addr for each
>> boot.  I guess thats better than nothing and not much to do on a
>> volatile system.
> Correct. The alternative would be to try and sort out some kind of hash
> based on unique data, maybe dmidecode data - but that's not very
> consistent across platforms. I think there is something like this
> available for the Beagleboard.
>
>
>> What does this do on a ro filesystem?  No network?  network with
>> 0:0:0:0:0:0 MAC addr? (horror! but I suspect the kernel won't allow
>> that).  Certainly this script failing should not cause the rest of boot
>> to fail.
> On a read only filesystem with genmac installed and RANDOM_MAC in the
> interfaces file, genmac will run and NOT update interfaces. This was
> written with the pch_gbe driver in mind. If no MAC is in the EEPROM, the
> pch_gbe driver leaves the MAC as 00:00:00:00:00:00 and leaves the
> interface down. It will fail any up command until the MAC has been set.
> So the subsequent networking script will fail to bring up the relevant
> interface. ifconfig -a will list the interface as down and with a MAC of
> 00:00:00:00:00:00.
>
> Other drivers will assign a random MAC within the kernel, making this
> all unnecessary, but David Miller refused this patch for this driver
> (despite the precedent).
>
> If you want to use a RO image on a platform with a pch_gbe nic and no
> EEPROM, you need to hardcode the generated MAC in the interfaces file.
>
> The alternative would be for me to call this pch_gbe_genmac and have it
> call ifconfig directly instead of delegating it to the interfaces file.
> But then I would still want to use a config file to store the
> first-boot-generated MAC so it was the same across boots. Once I have to
> generate a config file, I felt it made more sense to just use
> interfaces, and then it makes sense to make it generic (not pch_gbe
> specific), and that led me to this implementation.
>
> Seem reasonable?
>

Yes I think so.  Just wanted to make sure it failed gracefully.   I 
think your doing the best that can be expected.

Thanks for the detail.

[I'm interested as I have a board that might need this.  Not in OE yet, 
but someday ...]