some GPU machines fail to pxe-boot after update to iPXE 2.0.0

[mwennrich]

Description

Some GPU machines end up in a boot loop trying to boot from a non existing device:

iPXE 2.0.0 (g12798) -- Open Source Network Boot Firmware -- https://ipxe.org
Features: DNS HTTP HTTPS iSCSI TFTP VLAN SRP AoE EFI Menu
metal-stack pixie
Manufacturer: Supermicro
Waiting for link-up on net0... ok
Configuring (net0 be:3a:f2:b6:05:9f)...... ok
No ProxyDHCP response, retrying (attempt 0/10)
Configuring (net0 be:3a:f2:b6:05:9f)...... ok
No ProxyDHCP response, retrying (attempt 1/10)
(...)
No ProxyDHCP response, retrying (attempt 9/10)
Configuring (net0 be:3a:f2:b6:05:9f)...... ok
No ProxyDHCP response, retrying (attempt 10/10)
Failed to get a ProxyDHCP response after 10 attempts

(reboot)

These are the existing devices:

|  (B153/D0/F0) UEFI PXE IPv4 Intel(R) Ethernet Controller         |  E810-XXV for SFP(MAC:3CECEFE668FA)
|  (B153/D0/F1) UEFI PXE IPv4 Intel(R) Ethernet Controller         |  E810-XXV for SFP(MAC:3CECEFE668FB)
|  (B61/D0/F0) UEFI PXE IPv4 Intel(R) Ethernet Controller          |  X550(MAC:7CC255A864FA) 
|  (B61/D0/F1) UEFI PXE IPv4 Intel(R) Ethernet Controller          |  X550(MAC:7CC255A864FB) 

net0 with be:3a:f2:b6:05:9f does not exist, and seems to be a virtual device.

[mwennrich]

One option is to use snponly.efi (#48). But this has failed on dell-machines? (#42)

Another option is to change the boot.ipxe, to not give up on the first device, like this (AI generated, not tested):

#!ipxe

set attempts:int32 10
set try:int32 0
set max_if:int32 32

set user-class pixiecore

echo metal-stack pixie
echo Manufacturer: ${manufacturer}

set console ttyS1
set sp:hex 20 && set sp ${sp:string}
iseq ${manufacturer} Dell${sp}Inc. && set console ttyS0 ||
iseq ${manufacturer} Giga${sp}Computing && set console ttyS0 ||
iseq ${manufacturer} QEMU && set console ttyS0 ||
iseq ${manufacturer} FUJITSU && set console ttyS0 ||

:outer
echo ProxyDHCP scan (attempt ${try}/${attempts})
set i:int32 0
goto scan_if

:scan_if
iseq ${i} ${max_if} && goto retry ||

set iface net${i}
ifopen ${iface} || goto next_if

clear filename
dhcp ${iface} || goto next_if
isset ${filename} && goto boot

:next_if
inc i
goto scan_if

:retry
iseq ${try} ${attempts} && goto fail ||
inc try
goto outer

:boot
chain ${filename} || goto fail

:fail
echo Failed to get a ProxyDHCP response after ${attempts} attempts
echo Rebooting in 5 seconds...
sleep 5
reboot

[mwennrich]

One option is to use snponly.efi (#48). But this has failed on dell-machines? (#42)

• tested with our Dell machines, seems to work just fine