Bug 5665

Summary: CPU lockup when hitting OOM on amd64/NUMA
Product: Memory Management Reporter: Michael Renner (robe)
Component: Page AllocatorAssignee: Andrew Morton (akpm)
Status: REJECTED UNREPRODUCIBLE    
Severity: high CC: clameter, mrechberger, protasnb
Priority: P2    
Hardware: i386   
OS: Linux   
Kernel Version: 2.6.15-rc2 Subsystem:
Regression: --- Bisected commit-id:

Description Michael Renner 2005-11-26 22:43:36 UTC
Distribution: debian testing
Hardware Environment: iwill h8501, 8x dualcore opteron, 64gb ram
Problem Description: triggering OOM events under certain workloads causes lockups.

Steps to reproduce: I was able to reproducible trigger this bug by having
multiple processess (4-16) with large memory footprints grow until they
eventually reach the systems memory limit.

Here's a "screenshot" of a trace in 2.6.15-rc2 w/o swapspace:
http://666kb.com/i/10yom358azw8w.jpg
And here's one with a swapfile: http://666kb.com/i/10z1l2c61a8ld.jpg
Used .config: http://phpfi.com/88965
Comment 1 Natalie Protasevich 2007-10-29 10:48:44 UTC
Michael, is this still a problem with recent kernels?
Thanks.
Comment 2 Michael Renner 2007-11-04 17:28:19 UTC
I can't tell because I don't work with the systems in question anymore. The BUG() output looked pretty conclusive, but feel free to close this bug.
Comment 3 Natalie Protasevich 2008-03-10 23:52:27 UTC
I am pretty sure that the kernel is too old. It's been so many updates to page allocation/reclamation/throttling mechanism, the problem really needs to be confirmed. But I better let developers to decide.