Subject: Re: IP20 memory suckage
To: Gerald heinig <>
From: Steve Rumble <>
List: port-sgimips
Date: 02/05/2004 17:51:48
On Thu, Feb 05, 2004 at 10:26:18PM +0100, Gerald heinig wrote:
> I've been trying in vain to get a kernel compiled on my Indigo R4000 and
> I've noticed the build keeps dying at slightly different places with an
> internal compiler error due to a segfault. I'm assuming this is due to
> flaky memory. Does anyone else have any better ideas?

I'd certainly suspect memory first. If it's not the ram, it'll
probably be a painful endeavour to find out what's wrong (flakey
cpu, mainboard, power supply, etc). Of course, it could still
potentially be a problem with netbsd, but since the R4k mips
code is widely used, and IP20 is very similar to its younger
siblings, I'm not inclined to suspect that. However, I don't
recall ever trying to natively compile a NetBSD kernel on IP20.

> I've discovered a memory checking utility under the FreeBSD ports
> system, memtest, which I haven't been able to find for NetBSD. I'll try
> and port it over, unless anyone has any better ideas for testing memory
> under NetBSD.

You could just try pulling out the cpu board and removing the
memory from one bank at a time until it works, assuming there is
a bank without a bad simm.

A cursory look suggests that memtest is fairly portable already.

> Assuming the memory is bad, does anyone have any suggestions for getting
> SGI Indigo memory, apart from, say, If it's in Europe,
> or even better, in Germany, that would be really cool.

The IP20 uses regular true-parity 72-pinn SIMMs. I wouldn't
think that they're all that difficult to track down. Perhaps