[Etherboot-users] pxelinux.0: not a valid image

[David T. P. <phe...@un...>]

Hello,

    I have a cluster that I am booting via Etherboot. The nodes have no
CD-Rom or Floppy, so a network install is the only way to get an OS onto
the nodes. From what I have read, the basic operation of Etheboot and
PXE is that etherboot boots, gets a dhcp address, gets pxelinux.0 from
the tftp server, boots the pxelinux.0 file and then reads the contents
of ./pxelinux in the tftp root to get menu options. If this is not the
case, please correct me on how etherboot really works.

I have tested dhcp and tftp and know they work correctly. When I point
dhcp to serve pxelinux.0 to etherboot clients, the booting node gives a
line:

    Fliename pxelinux.0: ... not a valid image

At this point I was pulling out my hair because pxelinux.0 should be a
good image because it was installed with/by the OS and hasn't changed
from what was distributed with the OS (I'm using Rocks 4.1-Fuji). So, I
compiled memtest86 as an elf file and told dhcp to serve that instead.
Worked like a charm. Etherboot got a dhcp address, downloaded the
memtest.elf file and booted right into memtest. I then made an elf
kernel from vmlinuz and initrd.img which were in the tftproot and would
have been booted if pxe was working correctly. This booted the install
correctly. So, I am interested in why the elf works but the pxe doesn't.
I don't know what options Etherboot was compiled with because it is
burned into a LinuxBOIS rom on each node. The version line when it boots is:
    Etherboot 5.0.4eb4.1 (GPL) Tagged ELF (Multiboot) for [EEPRO100]
So, is there a way I can force pxe booting over elf through a dhcp
option or something?

Thanks for the help in advance,

-David Pheasant