Heh. You have hardware I haven't got access to, but I've seen the same
thing now that I have access to a SMP box.
The good news is that I _think_ I have found the problem, and I'm
running a personal test-kernel that has so far been up for almost an
hour under a load that usually crashed it within 15 minutes before my
fix.
Tha bad news is that my current fix is so horribly ugly that I don't
dare show it to anybody but Alan and David (and David hurled chunks when
he saw it ;)
Anyway, I hope to have this fixed in 2.1.28, with the really ugly fix
unless I can make it cleaner. This has been my #1 priority for the last
few days, and I finally feel like I have a good angle on it. Knock
wood.
Linus