On Fri, Dec 16, 2016 at 11:06 AM, Christian Ehrhardt < [email protected]> wrote:
> I'll be throwing that against all architectures and get back to you if it > works for me. > (IIRC you already accepted the d/t/control dependency to qemu-system in my > former bug, so not listed) > Hi, what we had on Friday now for me creates different, but very similar issues, to what I described before. I made the note before that using kvm enabled gets the guest running only for a few seconds as it is mostly in a 2nd level virtualization environment. There i had after a few moments seen it dying on missing CPUid features (that was in the bug we closed by accepting most of my initial diff). Now switching to qemu-system without KVM the way as we discussed closes this "race" in a bad way. It fails right away now loading the vmlinuz. So in the new case (with the diff proposed before in this bug applied) it now fails the smoke-qemu-session with: + virsh start sqs error: Failed to start domain sqs error: internal error: qemu unexpectedly closed the monitor: warning: TCG doesn't support requested feature: CPUID.01H:ECX.vmx [bit 5] qemu: could not load kernel '/vmlinuz': Permission denied IMHO I'd tend to move the validate qemu step down just before the start/stop of the guests and skip that part if the virt-host-validate fails. I'm short of loosing overview, so for the next steps after this I'll rebase > to your latest commit on Monday first. > And then start over again. > I'm now rebasing to integrate on top of your last 5 commits of Friday and check again. But I assume I'll hit the same. Qemu is not yet merged to latest yet, so still at 2.6.1 there - which is old, but not terribly old. Tests take a while every time but I hope to come back with a solution later today. -- Christian Ehrhardt Software Engineer, Ubuntu Server Canonical Ltd
