m400 cartridge (mcdivitt) unable to deploy Xenial

[Newell Jensen]

The m400 , (mcdivitt cartridge) is unable to boot Xenial cloud images. Primarily hosts under maas control. It would appear that flash-kernel is not being installed and run. flash-kernel runs mkimage which creates the boot.scr file.

** File not found /boot.scr **

I have included the full boot logs here:

https://pastebin.canonical.com/149343/

Booting PXE
Requesting DHCP address via NIC1
BOOTP broadcast 1
DHCP client bound to address 10.229.65.140
Retrieving file: pxelinux.cfg/602C7A67-95BE-53FD-960B-29C735B4A11F
Using NIC1 device
TFTP from server 10.229.32.21; our IP address is 10.229.65.140
Filename 'pxelinux.cfg/602C7A67-95BE-53FD-960B-29C735B4A11F'.
Load address: 0x4000800000
Loading: *
TFTP error: 'File not found' (1)
Not retrying...
Retrieving file: pxelinux.cfg/01-fc-15-b4-15-e9-91
Using NIC1 device
TFTP from server 10.229.32.21; our IP address is 10.229.65.140
Filename 'pxelinux.cfg/01-fc-15-b4-15-e9-91'.
Load address: 0x4000800000
Loading: #
  0 Bytes/s
done
Bytes transferred = 41 (29 hex)
Config file found
1: local
PXE: executing localboot
** File not found /boot.scr **
## Executing script at 4004000000
Wrong image format for "source" command
Booting M.2
** File not found boot.scr **
## Executing script at 4004000000
Wrong image format for "source" command

[Scott Moser (smoser)]

I'm not sure actually how this used to work. I dont know of any changes in the maas images taht would have caused it.

curtin (curtin/commands/curthooks.py) has a comment that says it calls update-initramfs if on arm and not uefi to get flash-kernel called.

It woudl be good if you could collect a install log with config that includes the following (change maas server's /etc/maas/preseeds/curtin_userdata)
 verbosity: 2

[Scott Moser (smoser)]

I added maas and curtin tasks.
Ultimately the bug is probably one of those.

things to note here:
a.) xenial maas images do not have a kernel installed in them.
    versions of curtin < 351 (the latest upload) will not choose to install a kernel so boot will fail.

b.) per dannf: one ugliness about m400 is that fw is hardcoded to read the boot.scr from the first partition
  maas 1.9 is probably not aware of this and probably not partitioning correctly, where as in maas 1.8 curtin would handle partitioning and previously do the right thing.

[Brad Marshall (brad-marshall)]

We're seeing the same thing with our m400 cartridges when deploying trusty with maas 1.9.0+bzr4533-0ubuntu1 (trusty1). I have a boot log with verbosity 2 at https://pastebin.canonical.com/151314/.

Please let us know if you need any further information, or if there's any debugging we can do to assist.

[Scott Moser (smoser)]

Brad and I got this functional for him (trusty deployment at least).
As dannf said above in comment 2, the boot.scr has to be on / of the first partition, which ends up being /boot for the installed system.

Thus, the system really needs /boot as the first partition. I suspect other layouts after that would work (such as lvm elsewhere).

Brad provided the cli command to make this work is:

maas <maas> node set-storage-layout node-<id> storage_layout=flat boot_size=2147483648

I'm deleting the curtin and maas-images task, but leaving the maas task. If this is "just work", maas would have to recognize the hardware and do the right thing.

[Newell Jensen (newell-jensen)]

Setting to Critical as this is blocking IS