Re: NMI errors in 2.0.30??u

Rogier Wolff (
Mon, 28 Apr 1997 15:23:44 +0200 (MET DST)

Alan Cox wrote:
> > This test detects the simplest type of errors, and not much more.
> > This test does not find all coupling errors that you describe.
> >
> > A more sofisticated test like marchg finds all errors, upto all the
> > single coupling errors that may exist.
> Only if you blindly assume everything else is within spec. You should
> do continual heavy DMA during all your testing. Lots of memory errors
> don't show up except during DMA accesses where the cache is much
> more stressed.

Yes, the DMA and "power supply interference" suggestions provide an
explanation of why a normal memory test doesn't find most memory

Indeed the "model" for the memory errors determines what the test will
look like and what it is going to find. The question that always
remains is: "Does the model conform to reality". In this case the
answer is "no".

Although I know that a memory test is not going to find 90% of the
memory errors, I might still want to write a kernel-option that does a
memory test just after boot. Alan, (or anybody else,) would you know
of a generic way to have DMA occurring concurrently with the memory
test to trigger that type of errors? I'm afraid that this would be
impossible right?