mcelog-error in mail from hourly cron job using Unbreakable Kernel

elygre

After installing OEL 5.5 + Unbreakable Kernel, I get the following error message sent from the hourly cron:

/etc/cron.hourly/mcelog.cron:

mcelog: warning: record length longer than expected. Consider update.

Some relevant info:

[[email protected] ~]# uname -a
Linux XXXXX 2.6.32-100.20.1.el5 #1 SMP Fri Oct 1 16:37:36 EDT 2010 x86_64 x86_64 x86_64 GNU/Linux
[[email protected] ~]# rpm -q mcelog
mcelog-0.9pre-1.30.el5
[[email protected] ~]# up2date --show-channels
el5_x86_64_latest
ol5_x86_64_latest
el5_x86_64_oracle
[[email protected] ~]# up2date -u -f
...
All packages are currently up to date

A similar bug report was filed against some Fedora-version last year, with updates through this year. This might be relevant and/or helpful, or not: https://bugzilla.redhat.com/show_bug.cgi?id=507026

Eirik

Dude!

I get the same error message by just typing mcelog at the command shell, but I'm running in Virtualbox.

MCE is nothing but feature of AMD / Intel 64 bit systems which is used to detect an unrecoverable hardware problem. Program such mcelog decodes machine check events (hardware errors) on x86-64 machines running a 64-bit Linux kernel. It should be run regularly as a cron job on any x86-64 Linux system. This is useful for predicting server hardware failure before actual server crash.

On my OL 5.5 systems mcelog is version: mcelog-0.9pre-1.30.el5.x86_64

Sergio-Oracle

Eirik,

Looks like you have network support at a minimum. If you have Basic or Premier Linux support, please report this issue via My Oracle Support. If you don't, I recommend you file a bug via bugzilla.oracle.com

Sergio

elygre

We got basic linux support, I think. even though our ULN CSI expired while I was goofing around on a sailboat in the pacific :-) That's being worked on, though.

Either way, I'll go with bugzilla, which I'd wager is faster, and probably closer to home for whoever needs to actually fix this: http://bugzilla.oracle.com/bugzilla/show_bug.cgi?id=11323.

Dude!

I'm not sure if this is the real fix, but....

Contents of mcelog.cron:
# /usr/sbin/mcelog ignorenodev filter >> /var/log/mcelog
mcelog: warning: record length longer than expected. Consider update.

Perhaps the mcelog utility is not in sync with the current kernel headers. After downloading the source and recompiling the error disappears:

# ./mcelog ignorenodev filter >> /var/log/mcelog
no error


Here is what I did:

# wget http://public-yum.oracle.com/repo/OracleLinux/OL5/5/base/x86_64/mcelog-0.9pre-1.30.el5.src.rpm

# mkdir -p /usr/src/redhat/SOURCES
# rpm -Uvh mcelog-0.9pre-1.30.el5.src.rpm

# cd /usr/src/redhat/SOURCES/
# tar zxvf mcelog-0.9pre.tar.gz

# cd /usr/src/redhat/SOURCES/mcelog-0.9pre
# make

# ./mcelog

# mv /usr/sbin/mcelog /usr/sbin/mcelog.orig
# ln -sf /usr/src/redhat/SOURCES/mcelog-0.9pre/mcelog /usr/sbin/mcelog
or if you trust it:
# cp /usr/src/redhat/SOURCES/mcelog-0.9pre/mcelog /usr/sbin/mcelog

elygre

That sounds like a reasonable fix to me :-)

For myself, I'll leave it as is, though. I don't want to compile OS binaries, and will just wait until a fixed binary comes along.

849715

This is a known bug in the underlying kernel. Upstream support for mcelog has dried up for years and a new method may be coming out. It appears the "immediate" fix in January 10 was never implemented--

https://bugzilla.redhat.com/show_bug.cgi?id=507026

Thanks,
Jim