java 1.4.2 segfaults on new PE2850
rorem at ai.uic.edu
Thu Mar 3 00:07:01 CST 2005
Matt Domsch wrote:
>On Wed, Mar 02, 2005 at 04:36:40PM -0600, Doug Rorem wrote:
>>I'm wondering if anyone has seen this (and has a fix?)
>>We have a PE2850, dual 3.2Ghz Xeons, 4gig memory
>>10gig swap (details at end), three 73gig HD's in RAID 5
>>configuration. We're running RedHat EL 3 Update 3.
>>I've been trying to install Oracle 10g w/o success.
>>Finally realized (via strace) that the java executable
>>in their JRE (v 1.4.2) is seg faulting. This behavior is
>>repeatable with the J2SE 1.4.2_07 RPM from Sun (by just
>>running /usr/java/j2sdk1.4.2_07/bin/java -version).
>>I tried turning off hyperthreading (I saw a post on
>>the net that suggested trying this) but that didn't
>>seem to make a difference.
>>Oddly, Sun's J2SE 5.0 RPM doesn't exhibit this behavior,
>>but that isn't much help because Oracle's installer
>>uses Java v 1.4.2
>>We another PE2850 (a few months old) with the same
>>exact configuration that *doesn't* exhibit this
>>behavior. The only difference (that I know of is this)
>>new PE2850 - dual Xeon 3.2Ghz model 4 stepping 1
>>old PE2850 - dual Xeon 3.2Ghz model 3 stepping 4
>>Has anyone else experienced this?
>Yes. I believe it has to do with NX (No Execute, aka Execute
>Disabled) being enabled. If you check the BIOS versions on your
>systems, are they different?
>Sun Java JREs < 1.5.0 don't work with NX enabled. You have two
>1) upgrade to JRE >= 1.5.0
>2) disable NX with the kernel command line option "noexec=off".
> If you're running an ia32e (EM64T) kernel, also add "noexec32=off"
Thanks - adding the 'noexec=off' in grub seems to have
done the trick. Wow - that was a bit obscure and difficult
to track down, I'm surprised there isn't more mention about
it on the web (unless I was searching the wrong things).
Unfortunately, we don't have the liberty of insisting that
the JVM > 1.5 because:
Oracle 10g installer uses 1.4.2_03 JVM
We use Sun's ChiliSoft ASP, which uses 1.4 JRE and the
Sun Java System Web Server (which apparently with v6.1
out feb-05 *does* support Java 1.5).
Outside of losing the supposed protection that the XD (Intel)
(NX - AMD) bit provides (with the noexec=off switch)... are
there any other implications you're aware of? (specifically
More information about the Linux-PowerEdge