data corruption with current (maybe sis chipset related?)

[Don Lewis]

On 10 May, Vallo Kallaste wrote:
> On Fri, May 09, 2003 at 01:59:37PM -0700, Don Lewis
> <truckman at FreeBSD.org> wrote:
> 
>> I've seen it on my Athlon XP box with 1GB of ECC RAM.  If I ran "make
>> buildworld" a number of times in a row, it would barf on a corrupted
>> file in /usr/src.  A 16 byte chunk of a source file would be changed to
>> some sort of random garbage.  It was just the copy of the file that was
>> cached in RAM that was corrupted.  Rebooting the system would show that
>> the file was undamaged.  Adding the kernel options mentioned earlier in
>> this thread makes the machine run flawlessly.
> 
> How did you built the world for testing? With missing /etc/make.conf
> or with some specific options, with -j or without? I have SMP Athlon
> box with 1GB ECC memory and would like to reproduce it. I'm able to
> reproduce it at will on P4 Celeron with 128MB of memory (gcc dies in
> parallel buildworld).

The only thing in /etc/make.conf was KENRCONF.  I was testing without
-j.  It took about five or six buildworlds for it to show up.