KVM test failure but see guest kernel output

[Ike Panhc]

Running `sudo certify-18.04` on arm64 platform and KVM test fails.
But in the log we can see guest kernel output

http://kernel.ubuntu.com/~ikepanhc/checkbox/d06-uose1806/submission_2018-06-18T03.08.02.153262.html#13-1-log

[dann frazier (dannf)]

I looked into this a bit and found it to be reproducible on this platform. I disabled the cloud-init service so that the guest OS would complete booting. I then manually ran "cloud-init init" to reproduce the hang, then attached strace to see where it might be hung.

root@ubuntu:/var/log# strace -p 1255
strace: Process 1255 attached
getrandom(^Cstrace: Process 1255 detached
 <detached ...>

Ah.. maybe this system just lacks entropy. Indeed, there is no rng device in the guest:

root@(none):/# cat /sys/devices/virtual/misc/hw_random/rng_available

root@(none):/# cat /sys/devices/virtual/misc/hw_random/rng_current
none

So, I attached the virtio-rng device to the guest by adding "-device virtio-rng-pci" to the qemu command line. Unfortunately, the driver for this device isn't in the initramfs, so I manually added it:

echo "virtio-rng" >> /etc/initramfs-tools/modules
mkinitramfs -o /boot/initrd.img-4.15.0-24-generic 4.15.0-24-generic

I also installed haveged on the host to try and generate more entropy - but I don't know if that made any real difference. I wouldn't recommend doing that as part of cert unless we know it is necessary.

Perhaps its possible to simply avoid the need for entropy w/ a cloud-init config - that might be a simpler way to go about this.

[Jeff Lane  (bladernr)]

Moved to P-P-C and re-opened so we can dig into this. Not sure it's recreatable or not outside of arm but we have two things here:

1 we can try to figure this out and figure out a fix for it in the virtualization script
2 there's a plan to redo the virtualization script KVM bits to be far more simplified.

I think we need to do 1 first and can keep this in mind for 2 later to avoid a regression.

[Maksim Beliaev (beliaev-maksim)]

Bug was migrated to GitHub: https://github.com/canonical/checkbox/issues/164.
Bug is no more monitored here.