memory.c - bad pmd - x86_64

[Peter J. Stieber]

PJS = Peter J. Stieber
PJS>> I have a system with a Tyan 2885 motherboard
PJS>> (S2885-ANRF) that uses dual Opteron 244 processors.
PJS>> Each processor has 1 GB of memory for a total of
PJS>> 2 GB. I am using a SATA HD. I am running the latest
PJS>> stock release of the SMP version of the FC3 kernel
PJS>> for x86_64. uname -a output follows:
PJS>>
PJS>> Linux maggie 2.6.11-1.14_FC3smp #1 SMP
PJS>> Thu Apr 7 19:36:23 EDT 2005
PJS>> x86_64 x86_64 x86_64 GNU/Linux
PJS>>
PJS>> The computer is worldly node for a small cluster of
PJS>> computers. It is resposible for building a code that
PJS>> is run on the cluster. A shell script is used to
PJS>> start the build process. Occasionally when the script
PJS>> is started it crashes and the following messages are
PJS>> place in /var/log/messages (sorry for the ugly line
PJS>> wrap):
PJS>>
PJS>> May 11 16:26:56 maggie kernel: mm/memory.c:97:
PJS>> bad pmd ffff81002f6a4000(0000000000000008).

DJ = Dave Jones
DJ> Please grab the latest test kernel from
DJ> http://people.redhat.com/davej/kernels/Fedora/FC3
DJ> and try to reproduce this. It contains debugging code
DJ> that hopefully will help nail this.

Thanks Dave.
I loaded the kernel:

Linux maggie 2.6.11-1.24_FC3smp #1
SMP Tue May 10 19:12:22 EDT 2005
x86_64 x86_64 x86_64 GNU/Linux

I'm trying to force the problem to occur, but as was reported on the 
linux-kenel list, it isn't obvious how to make the problem rear its ugly 
head.

Are you looking for /var/log/messages output when it happens?

Thanks again for the help. I'm very willing to serve as a debug test bed 
as my worldly node is a Tyan S2885 Thunder K8W motherboard running the 
SMP version of x86_64 FC3 and my compute nodes are Tyan S2850 Tomcat K8S 
motherboards running the non-SMP version of x86_64 FC3.

Will reply to this thread when the problem pops up,
Pete