Comment 31 for bug 1208455

As it seemed that starting a 64bit hypervisor (qemu-system-x86_64) from a 32bit user-space running on another 64bit hypervisor has different issues which seem to go back even further, I concentrated on bisecting the case of host running 64bit user-space on a 64bit CPU and the first level guest running 64bit user-space under a 64bit hypervisor. The second level guest is started with a 64bit hypervisor but a 32bit userspace. All host and guests are Saucy and started with the same kernel.

Then for bisecting I only replaced the host kernel with kernels between v3.9 and v3.10. Then started a guest and ran the prepare-testbed repeatedly. The first change that broke things was:

commit 5f3d5799974b89100268ba813cec8db7bd0693fb
KVM: nVMX: Rework event injection and recovery

However at that point starting a 2nd level guest is completely broken until:

commit 5a2892ce72e010e3cb96b438d7cdddce0c88e0e6
KVM: nVMX: Skip PF interception check when queuing during nested run

So potentially the rework patch broke things and the current issue is remaining after fixing the major issues. Though it could be introduced by one of the patches in between but it cannot be bisected.