Re: Bad Illegal instruction traps on dual-Xeon (p4) Linux Dell box

["James Washer" <washer@us.ibm.com>]

[-- Attachment #1: Type: text/plain, Size: 1965 bytes --]


Just to clarify things.

Lots of processes die from illegal op traps.. gcc, bash, make, etc... but
the instruction is ALWAYS  opcode 0x55 and is part of a subroutine preamble
in every case..  You are correct... 0x55 should not generate a trap.

Bad cpu? Hmmm, Tom has 6 different CPU's ( all p4 xeons ), on three
systems, that have this EXACT same problem.

and why does this require the system to be running smp?


 - jim

Kurt Garloff <kurt@garloff.de>@vger.kernel.org on 03/20/2002 03:26:10 PM

Sent by:    linux-kernel-owner@vger.kernel.org


To:    Tom Epperly <tepperly@llnl.gov>
cc:    Linux kernel list <linux-kernel@vger.kernel.org>
Subject:    Re: Bad Illegal instruction traps on dual-Xeon (p4) Linux Dell
       box



On Wed, Mar 20, 2002 at 01:35:30PM -0800, Tom Epperly wrote:
> The kernel log showed me that various standard programs such as
> /bin/sh are generating bogus illegal instruction traps on a legal
> opcode (0x55) as part of a standard function preamble. After receiving
> an illegal instruction trap on opcode (0x55), the modified kernel does
> a wbinvd() to flush the cache and a __flush_tlb() to flush the TLB
> and then retries the "illegal" opcode. The retry produces a second
> illegal instruction trap on the same legal opcode (0x55). Information
> from /var/log/messages is shown below.

The CPU is what triggers the exception.
So this sounds like a defect (or overheated) CPU to me.

OTOH, the kernel logs "invalid operand". Could you run ksymoops to get a
disassembly?
AFAICS, its a push %ebp instruction, which should not be illegal. So either
your stack is overflowing or my suspicion with the defect CPU is
applicable.

Regards,
--
Kurt Garloff                   <kurt@garloff.de>         [Eindhoven, NL]
Physics: Plasma simulations  <K.Garloff@Phys.TUE.NL>  [TU Eindhoven, NL]
Linux: SCSI, Security          <garloff@suse.de>    [SuSE Nuernberg, DE]
 (See mail header or public key servers for PGP2 and GPG public keys.)


[-- Attachment #2: C.DTF --]
[-- Type: application/octet-stream, Size: 242 bytes --]

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: For info see http://www.gnupg.org

iD8DBQE8mRqSxmLh6hyYd04RAjseAJ9D9WFchN4IdWbh/rUcJ9C55RT6ngCgl4p9
HA1QFzDVq2UdL939jr2bu7U=
=jGsY
-----END PGP SIGNATURE-----