[syslinux] PXElinux boot sequence with multiple ethernets

[jon.milan at stativa.com]

   Hello,

   I am attempting a PXE boot between two systems, each with multiple
   network cards.   While there are a total of 8 ports on each computer,
   only two (each) are connected as follows:

   Boot Server

        eth0 - 10GbE fiber channel (private to the set of computers being
   managed) (Qlogic)

        eth4 - 1Gb ethernet (public and out of my sphere of management)
   (NetExtreme II)

   Boot Client

        eth4 - 10GbE fiber channel (same private as above)

        eth0 - 1Gb ethernet (same public as above)

   These are both IBM systems.

   The firmware in the boot client is configured to PXE boot using eth4.
   DHCPDISCOVERY, OFFER, REQUEST and ACK appear as I expect with a
   pxelinux.0 file specified.  Vendor class identifier is
   "PXEClient:Arch:00007:UNDI:003016"

   After a negotiation (option ack fails on first request cause of tsize,
   and I'm using tftpd-hpa 5.0), TFTP transfers the pxelinux.0 file (26
   packets).

   Then the problem starts:

   1)  The whole DHCPDISCOVERY / OFFER / REQ / ACK / TFTP (pxelinux.0)
   sequence repeats on the same eth4 port.

   2)  The client's eth0 port attempts DHCPDISCOVERY which times out (no
   DHCP for this "public" subnet).  This appears to be a PXE extended
   discovery as it also includes a Vendor class identifier "PXEClient"
   (different values).

   3)  The client falls through to the next boot option specified.

   I have attempted using option classes in the DHCP configuration, but
   have seen no better results.

   This is my first foray into PXE, though I am well versed in the UN*X
   world (starting 1976).  Perhaps the symptom is known to someone already
   -- might anyone have a suggestion?

   As a data point, syslinux boot from a thumb drive launches me in to the
   menu properly.   We are looking for a hands-free option, however, so I
   am continuing to pursue PXE.

   Thank you in advance,

   Jon