Re: Linux 2.6.30-rc1 tg3 endian issues with MAC addresses on BCM5701. Bisected.

[David Miller <davem@davemloft.net>]

From: Robin Holt <holt@sgi.com>
Date: Mon, 13 Apr 2009 15:15:58 -0500

> On Mon, Apr 13, 2009 at 10:57:24AM -0700, Matt Carlson wrote:
>> Thank you for doing this bisection Robin.  This, and the data you
>> provided in your previous email is really helpful.  Unfortunately, they
>> raise more questions than answers.
>> 
>> I reviewed this patch again, and it still looks correct to me.  It
>> should be one big behavioral no-op.  Obviously something is wrong, but
>> I'm not seeing the root cause at the moment.  I need to think about this
>> some more.
> 
> If you want me to build test kernels with debug printk's, etc.  Let me
> know.  I will be working that issue on the side, so my responses may be
> slightly delayed.  I have access to many tests machines that exhibit
> this problem.
> 
> The problem seems to affect BCM5701 based systems only.  BCM5704 have
> the correct MAC addresses.  Nearly all of our SGI Altix 3700/4700
> machines have the BCM5701 adapters on their base I/O board.

I think it's only going to hit chips that access those particular
NVRAM chip read/write paths.

Matt, realize that on big-endian there are implicit endian
conversions going on.  The read*/write* macros are converting
little-to-big endian on big-endian systems.

Actually, is that the case, that we only see this corrupt MAC
address bug on big-endian systems?  Or are we seeing this on
little-endian too?