Syslinux - Mar 2009 - menu.c32: CPU Fault on console 0

Hi,

I hit a rather strange problem.  I'm PXE booting an IBM Blade.  It has
a BMC which provides serial over LAN via the BladeCenter management
module.  Its BIOS has a "Console Redirection" option, which dumps its
VGA console messages to a serial port.  "Redirection after Boot" is
disabled, but pxelinux runs "before boot", so everything arrives twice
on the serial line (which I can attach to after telnetting into the
management module) with the following config:

serial 1 19200
console 1
prompt 0
timeout 100
default menu.c32
[some labels]

In other words, everything works as expected.  But if I set

console 0

instead, loading menu.c32 kills off the processors of the blade.
Sometimes only one, sometimes both.  The management module logs:

I  System Reboot 
E  Internal Error CPU 1 Fault 
I  System Reboot 
I  Recovery Internal Error CPU 1 Fault 
I  SMI Hdlr:   00151350 Processor MachineCheck Data a  reporting bank:0000
reporting APICID:00FF CR4:00
I  SMI Hdlr:   00151351 Processor MachineCheck Data b  Address:00000000 00000000
Timestamp:00000036 3F5
I  SMI Hdlr:   00151352 Processor MachineCheck Data c  Status:A2000000 84010400
I  Blade Server Powered Down 
E  Internal Error CPU 2 Fault 
I  Power unit power off event received

By resetting this and that, it's possible to recover from here.
Before the processor halts, menu.c32 manages to output the top left
corner to the serial port:

$ od -c topleft.ctrl
0000000 033   [   0   m 033   [   2   0   h 033   e 033   %   @ 033   )
0000020   0 033   (   B 033   [   0   ;   3   7   ;   4   0   m 033   [
0000040   ?   2   5   l 033   [   2   J 033   [   1   ;   1   1   H 033
0000060   [   0   ;   3   0   ;   4   4   m 016   l
0000073

This happens with PXELINUX 3.74 pre1-12-gfc7c93a (and same menu.c32).
Flow control (0x13) doesn't make any difference.  What can be wrong?
The native serial support of pxelinux is more usable than the BIOS VGA
redirection, so I'd like to use the former if possible.
-- 
Thanks,
Feri.

Ferenc Wagner <wferi at niif.hu> writes:
> [...] In other words, everything works as expected.  But if I set
> console 0
> instead, loading menu.c32 kills off the processors [...]
Hi,

I understand this is a rather strange report, but can't anybody give
it a shot, whatever long?  I can successfully run dmitest.c32,
pcitest.c32 and vesainfo.c32 (these are the ones I tried), but if I
start menu.c32, the system freezes after clearing the screen and
printing the top left corner.  Actually, it reboots after a while
leaving the terminal emulator in a confused state, printing strange
characters instead of letters, just like after a cat /vmlinuz.

I'd be grateful for a tip, how to investigate this further...
-- 
Thanks,
Feri.