FreeBSD 5.3-R-p5: frequent kernel panics

[Doug White]

On Mon, 28 Feb 2005, Piotr Gnyp wrote:

> On Fri, Feb 25, 2005 at 05:11:46PM -0800, Doug White wrote:
> > Please capture the panic messages and relevant kernel output along with
> > your traces. It helps to know what the system was doing when it tanked,
> > and the trap info decodes the frame in an easy-to-digest format for
> > humans. :)
>
> In info files:
> Good dump found on device /dev/da2s1b
>   Architecture: i386
>   Architecture version: 1
>   Dump length: 1073152000B (1023 MB)
>   Blocksize: 512
>   Dumptime: Wed Feb 23 15:31:58 2005
>   Versionstring: FreeBSD 5.3-RELEASE-p5 #1: Fri Feb 18 11:30:01 CET 2005
>   Panicstring: page fault
>   Bounds: 6

This is insufficient... use kgdb to load up the crashdump and post the
output of 'print msgbuf'. That will have the trap message in it.

> > Your traces are quite bizarre... somehow ttwakeup() jumps off into space.
> > Line 2370 of ttwakeup() is ttecho() which implies your system is suffering
> > from severe memory corruption, or your kernel binary is damaged.  I'd
> > suggest doing a fresh buildworld+kernel and running memtest86. Also the
> > mpt driver in 5.3 is known to react badly to SCSI errors but I don't think
> > it runs off and randomly corrupts memory.
>
> I phoned HP, they told me to reorganize RAM in banks, so now 512MB is in
> bank 0 and 256 in bank 1 and 2. I did also as you adviced - make buildworld,
> make kernel, make installworld.
>
> After changing MP from 1.4 to 1.1 system was stable for 3 days. However I
> switched again to 1.4 (according to HP support). Now I`ll wait if it will
> panic again.

This sounds like you have a bad CPU. FreeBSD has supported MPspec 1.4
since the beginning.  I've never encountered any machine that needs that
toggled. I'm sure if there are ones that need it, Compaq would be one :)

> > Also check the system event log for any errors logged from hardware, such
> > as ECC corrections, power problems, or temperature alarms.  The ProLiants
> > (this is a DL380 or something of that nature?) have sophisticated internal
> > monitoring and can log events even if the OS is damaged.
>
> None errors on console and log (server - HP Proliant ML110)
>
> > What were your kernel compile flags?
>
> make.conf:
> COPTFLAGS= -O -pipe
> WANT_FORCE_OPTIMIZATION_DOWNGRADE=1

I don't know what WANT_FORCE_OPTIMIZATION_DOWNGRADE=1 is, so I'd remove
it. COPTFLAGS would override any changes anyway.

-- 
Doug White                    |  FreeBSD: The Power to Serve
dwhite at gumbysoft.com          |  www.FreeBSD.org