[Etherboot-discuss] gPXE and IBM TPMfOSd with NetXen Phantom Issue

[Miller, S. <Sha...@yr...>]

Good day to All,

I've got a NetXen VEN 0x4040 DEV 0x0100 4-port card here.  Some new HP
server computers ship with such, along with a version of gPXE 0.9.6+ on
the NIC's ROM.

I'm having an issue with IBM's Tivoli Provisioning Manager for OS
Deployment.  I've seen the same issue with gPXE's legacy drivers, too.

TPMfOSd shuts down the UNDI stack and restarts it as part of its loading
procedure.  It then does repeatedly performs ARP requests for the
TPMfOSd server, but appears to be deaf.  Packet captures reveal that the
server is most certainly responding immediately after the ARP request,
but this doesn't seem t help the TPMfOSd client, who uses gPXE-provided
UNDI.

make
DEBUG=arp:15,pxe_call:15,pxe_errors:15,pxe_loader:15,pxe_preboot:15,pxe_
undi:15,phantom:15 bin/40400100.lkrn

revealed quite a bit of information.  I do not see any ARP hit/miss
messages during the troubled stage, though I do see them earlier on
during the successful gPXE DHCP negotiation and file download.  I do see
UNDI_TRANSMITs at TPMfOSd-chosen retry intervals.  I see the Phantom
driver's phantom_post_cds() debug messages with each ARP request, but do
not see _rds() messages; the NIC appears to be deaf.

Quite unfortunately, the serial output slowed down the TFTP transfer
enough so that I could not both obtain a log and download the Rembo-ia32
file both successfully.  Also unfortunately, the nature of the TPMfOSd
product appears to generate a per-client and per-request unique file, so
building gPXE with an embedded Rembo-ia32 also does not help.

Also interestingly, when booting the embedded file, gPXE went through so
many lines of 'Phantom 0xXXXX does not have option ""' messages, that I
had to leave and come back a long time later to see that the embedded
Rembo-ia32 was not going to work as a proper downloaded one would.
These many lines strike me as odd, since the same gPXE behaviour does
not show when downloading Rembo-ia32 from the TPMfOSd server and
attempting to boot it.  This little bit will require more investigation.

Does anybody have any suggestions for what could be causing a deaf gPXE
driver, or any tests or trials?  The gPXE driver works just fine for
everything except how TPMfOSd seems to use it.  Could it be related to
the UNDI shutdown and restart?

I do see messages from the Phantom driver where it adds the Ethernet
broadcast MAC and the MAC address for the correct port.

Thanks, anyone.

- Shao Miller