On Feb 19, 2008 3:02 PM, Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> wrote:
quoted text
> Sure,
>
> Initial thoughts :
>
> I'd like to get the complete config causing this bug. I suspect either :

The .config I used in 2.6.25-rc2 is attached.

quoted text
> - A race between the lockless algo and an IRQ in a driver allocating
>   memory.
> - stat(c, ALLOC_FASTPATH); seems to be using a var++, therefore
>   indicating it is not reentrant if IRQs are disabled. Since those are
>   only stats, I guess it's ok, but still weird.
> - CPU hotplug problem.
>   http://bugzilla.kernel.org/attachment.cgi?id=14877&action=view shows
>   last sysfs file:
>   /sys/devices/system/cpu/cpu3/cache/index2/shared_cpu_map
>   -- is this linked to a cpu up/down event ?

# CONFIG_HOTPLUG_CPU is not set

I expect my system would get rather angry if I try to remove a cpu
while it is running. ;)

And as I don't use suspend, I wanted to stay away from any bugs that
might result from this.

quoted text
> Since this shows mostly with network card drivers, I think the most
> plausible cause would be an IRQ nesting over kmem_cache_alloc_node and
> calling it.

Thanks to yanmin crashing his machine with hackbench I now have a much
better testcase than recompiling KDE.
Starting hackbench will result in an nearly instant crash, if there is
a floodping running against the system.

This new testcase also showed that 2.6.24-rc2-mm1 can also be crashed
in this way.

I will try the patches from this tread against this new testcase and
report the results.

Torsten