Memory modified panic -- was: Re: cvs commit: src/sys/kern kern_malloc.c

(cc'd to current)

On Wed, Feb 01, 2006 at 12:54:56PM -0500, Andrew Gallatin wrote:
> Steve Kargl [sgk_at_troutmask.apl.washington.edu] wrote:
>> On Wed, Feb 01, 2006 at 11:44:27AM -0500, Andrew Gallatin wrote:
>>> Steve Kargl writes:
>>>> On Wed, Feb 01, 2006 at 10:47:52AM -0500, Andrew Gallatin wrote:
>>>>> WARNING: WITNESS option enabled, expect reduced performance.
>>>>> Memory modified after free 0xffffff0000006d00(248) val=5 _at_
>>>>> 0xffffff0000006dd0
>>>>> kernel trap 9 with interrupts disabled
>>>> 
>>>> You can trigger this panic without the red zone stuff.
>>>> See my string of post from yesterday.  Something went
>>> 
>>> Your panic looks like the same panic as mine, but with memguard rather
>>> than redzone.  These systems do similar things, I suppose it is only
>>> natural that they'd be hit by the same bug.
>> 
>> You get it without memguard, too.  Kris suggested that I try memguard
>> to capture the problem.  Unfortunately, memguard actually made
>> matters worse in that I did not even make it to single user mode
>> before a panic.
> 
> Odd.  I do not get it without using either memguard or redzone.
> I have a (mostly generic) kernel with WITNESS & INVARIANTS
> that boots fine.
> 
> Maybe we should move this to -current or -amd64, as it does not
> seem to be directly related to this commit..

The system boots without memguard or redzone here as well.
However, I can panic the system with any significant load
and filesystem activity.  Currently, a "make buildworld"
with a "gmake check-gfortran" in a GCC build tree will
kill my system.

-- 
Steve