2004-01-21 22:08:06

by Brian Ristuccia

[permalink] [raw]
Subject: 2.6.0 CONFIG_NUMA=y leads to random segfaults

While building a kernel for a Dual Athlon MP 2800 system, I accidentally
enabled CONFIG_NUMA, "Numa Memory Allocation Support". This option should
have been a no-op on the system I was using it on, since on this motherboard
all system memory is equally close to each of the two CPU's. Instead, it
caused userspace processes to randomly segfault.

Booting a 2.6.0 rebuilt without CONFIG_NUMA made the problem go away.

--
Brian Ristuccia
[email protected]


2004-01-22 19:33:50

by Martin J. Bligh

[permalink] [raw]
Subject: Re: 2.6.0 CONFIG_NUMA=y leads to random segfaults

> While building a kernel for a Dual Athlon MP 2800 system, I accidentally
> enabled CONFIG_NUMA, "Numa Memory Allocation Support". This option should
> have been a no-op on the system I was using it on, since on this motherboard
> all system memory is equally close to each of the two CPU's. Instead, it
> caused userspace processes to randomly segfault.
>
> Booting a 2.6.0 rebuilt without CONFIG_NUMA made the problem go away.

Can you try 2.6.1 with Andi's latest patchset from here:
ftp://ftp.x86-64.org/pub/linux/v2.6/x86_64-2.6.1-2.bz2
There's been quite a few updates to AMD oustanding since before 2.6.0
that Andi said were critical.

If it's still bust, could you fish around in /var/log/messages or consoles
for any stack traces or anything you can see?

M.