Subject: Re: -current MP panics
To: None <port-alpha@netbsd.org>
From: Christos Zoulas <christos@zoulas.com>
List: port-alpha
Date: 08/09/2004 04:03:49
In article <200408090353.i793rwt7028858@egsner.cirr.com>,
Stephen M. Jones <smj@cirr.com> wrote:
>I've got two CS20s running the daily snapshots (latest is 08/02, I'll
>update to whatever releng has tonight).
>
>I'm running MP kernels with DIAGNOSTICS/LOCKDEBUG and -g makeoptions,
>but can't seem to get much out of them when they panic or hard hang.
>panics won't successful dump due to an fpsave 'ipi didn't'
>
>Anyway, here are the messages I've gotten. If you'd like, I'd be
>happy to run some other commands. Also, is it possible for an MP
>system to be more stable without all the debugging switched on? I've
>had an earlier build run for 12 days doing a looped 'build.sh build'
>successfully (takes about 4 hours to a complete each time). Now that
>I have debugging turned on and using the 0802 build, I'm not able to
>keep the systems up for more than 4 hours.
DIAGNOSTIC/LOCKDEBUG will catch problems sooner, and will add more
sanity checks that will cause panics. Some of the problems that
DIAGNOSTIC catches will not cause the system to crash... So yes,
adding DIAGNOSTIC will make the system crash more. In addition, gcc
seems to be best tested with -O, and just -g might produce buggy
code. YMMV.
christos