Emulex CNAs would not iPXE boot in UEFI mode.

[Om Kumar]

While using a HPE ProLiant 490c G9 in UEFI mode, during provisioning phase, the server got stuck in iPXE script after downloading kernel and ramdisk.

:deploy
imgfree
kernel {% if pxe_options.ipxe_timeout > 0 %}--timeout {{ pxe_options.ipxe_timeout }} {% endif %}{{ pxe_options.deployment_aki_path }} selinux=0 disk={{ pxe_options.disk }} iscsi_target_iqn={{ pxe_options.iscsi_target_iqn }} deployment_id={{ pxe_options.deployment_id }} deployment_key={{ pxe_options.deployment_key }} ironic_api_url={{ pxe_options.ironic_api_url }} troubleshoot=0 text {{ pxe_options.pxe_append_params|default("", true) }} boot_option={{ pxe_options.boot_option }} ip=${ip}:${next-server}:${gateway}:${netmask} BOOTIF=${mac} {% if pxe_options.root_device %}root_device={{ pxe_options.root_device }}{% endif %} ipa-api-url={{ pxe_options['ipa-api-url'] }} ipa-driver-name={{ pxe_options['ipa-driver-name'] }} boot_mode={{ pxe_options['boot_mode'] }} initrd=deploy_ramdisk coreos.configdrive=0 || goto deploy

initrd {% if pxe_options.ipxe_timeout > 0 %}--timeout {{ pxe_options.ipxe_timeout }} {% endif %}{{ pxe_options.deployment_ari_path }} || goto deploy
boot

When a sleep of 10s is added before boot, as follows, it works.

:deploy
imgfree
kernel {% if pxe_options.ipxe_timeout > 0 %}--timeout {{ pxe_options.ipxe_timeout }} {% endif %}{{ pxe_options.deployment_aki_path }} selinux=0 disk={{ pxe_options.disk }} iscsi_target_iqn={{ pxe_options.iscsi_target_iqn }} deployment_id={{ pxe_options.deployment_id }} deployment_key={{ pxe_options.deployment_key }} ironic_api_url={{ pxe_options.ironic_api_url }} troubleshoot=0 text {{ pxe_options.pxe_append_params|default("", true) }} boot_option={{ pxe_options.boot_option }} ip=${ip}:${next-server}:${gateway}:${netmask} BOOTIF=${mac} {% if pxe_options.root_device %}root_device={{ pxe_options.root_device }}{% endif %} ipa-api-url={{ pxe_options['ipa-api-url'] }} ipa-driver-name={{ pxe_options['ipa-driver-name'] }} boot_mode={{ pxe_options['boot_mode'] }} initrd=deploy_ramdisk coreos.configdrive=0 || goto deploy

initrd {% if pxe_options.ipxe_timeout > 0 %}--timeout {{ pxe_options.ipxe_timeout }} {% endif %}{{ pxe_options.deployment_ari_path }} || goto deploy
sleep 10
boot

iPXE used from http://boot.ipxe.org/ipxe.efi

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/373194

[Julia Kreger (juliaashleykreger)]

Out of curiosity, Did the CNA download the iPXE binary and use that? or did it use an onboard iPXE image?

[Miles Gould (mgould)]

Can you please post the generated inspector.ipxe?

[Miles Gould (mgould)]

Sorry, I'm mixing up inspection and deployment. Please ignore my request above.

[Miles Gould (mgould)]

OTOH, did you try introspection with this setup? And if so, did it work, or did you encounter the same issue?

[Om Kumar (om-kumar)]

@juliaashleykreger - CNA did download ipxe and used it instead of onboard iPXE. We confirmed this by looking at server console logs.

@mgould - We tried this with a simple setup of iPXE with a isc-dhcp-server and apache web-server. We also tried using different IPA kernel and ramdisk based on Fedora, ubuntu and Debian and kernel/ramdisk from running RHEL OS. But they all resulted in the same error. However, using sleep helped resolve it in all case.

To me, it looks like the issue is due to buggy Emulex Firmware. We tried the same iPXE server with Intel Network Adapters in UEFI mode and they worked without issues.

[Dmitry Tantsur (divius)]

Hi! I highly recommend you raise this to the iPXE developers. It would be much better if they provided a work around in their code, this way the whole world would benefit from it.

[OpenStack Infra (hudson-openstack)]

Change abandoned by Dmitry Tantsur (<email address hidden>) on branch: master
Review: https://review.openstack.org/373194
Reason: Hi! This patch has been without updates for 10 months, so I have to abandon it.

[Dmitry Tantsur (divius)]

Hi! All patches are abandoned, so I unassign this bug.

[Vladyslav Drok (vdrok)]

So is there anything that can be done on ironic side? adding sleep is obviously not an option upstream :)

[Dmitry Tantsur (divius)]

Yeah, I'll mark it as Opinion until somebody comes up with an upstreamable way to fix it.