Tracking down a fixed bug: 10h38m problem

[Steven N. Severinghaus]

Andy Whitcroft wrote:
> On Fri, Nov 06, 2009 at 10:53:02AM -0600, Steven N. Severinghaus wrote:
>> Hi, all. I'm trying to track down something that is causing problems in
>> OpenWRT[1] that seems to be fixed with an Ubuntu kernel. In short, our Via
>> C3 motherboards hang after about 10 hours and 38 minutes (~32870 seconds,
>> plus or minus a few seconds), and they wake up as if no time has elapsed if
>> you hit "enter" on the serial console.
[...]
> I don't recognise the symptoms specifically no.  That kernel would have
> been 2.6.28 based and we did take a lot of stable updates for that.
> My first suggestion would be to try 2.6.28.10 which was the last full
> stable we took, and perhaps the latest 2.6.28.x.

Thanks for the suggestion, Andy. My latest test was to check out the latest 
Ubuntu kernel sources from the Karmic branch and compare that to the latest 
server-install CD. As I understand, when I git clone 
git://k.u.com/ubuntu/ubuntu-karmic.git, that's already got all of the 
Ubuntu-specific patches applied, right?

So having built that with the OpenWRT toolchain, using i386_defconfig, the 
problem persists, whereas with the same filesystem image, but with the 
server-install kernel swapped in, the problem is gone. I think this leaves 
me two possibilities, and I'd appreciate any thoughts on them or any 
alternatives.

1) There is a difference between ubuntu-karmic.git and whatever ends up on 
the server-install CD. How is that kernel built? Is it patched in some way 
beyond what's in git? What kernel config is used? Anything special about 
the toolchain that builds it?

2) There is a problem with the OpenWRT toolchain. This possibility worries 
me because it sounds like it would be very hard to track down. If this is 
the problem, then it has persisted across OpenWRT releases and on several 
different build platforms.

Any thoughts?

Cheers,
-Steve