Hey, we're running a VM-pool based on libvirt and QEMU/KVM. The host machines run debian 6 (squeeze). All VM's have a similar libvirt XML-definition. Since we've moved to the combination of Linux kernel 3.2.0-1 and libvirt 0.9.9 (installed from testing/wheezy repo's), we periodically notice poor performance of new(ly started) VM's and after inspection they turn out to be started without the qemu option -enable-kvm. In /var/log/libvirt/qemu/$VM-name.log a line like this is shown: 20529: error : qemuBuildCommandLine:3681 : unsupported configuration: the QEMU binary /usr/bin/kvm does not support kvm After a virsh destroy/start, using the same XML-definition, the VM is started WITH -enable-kvm. Looking at the libvirt source of 0.9.9, I can't imagine this being the result of a race condition or other strange circumstance: it simply parse 'qemu -help', which always give exactly the same output. I've tried reproducing the problem by continuously destroy, undefine, define and start the same VM and then grepping the kvm-commandline for the absence of -enable-kvm, without any result. The XML-definition includes <domain type='kvm'> and <emulator>/usr/bin/kvm</emulator> Does anyone recognize this behaviour of libvirt? Thanks, Reinier Schoof # virsh version Compiled against library: libvir 0.9.9 Using library: libvir 0.9.9 Using API: QEMU 0.9.9 Running hypervisor: QEMU 0.15.0 Linux 3.2.0-1-amd64 #1 SMP Fri Feb 17 05:17:36 UTC 2012 x86_64 GNU/Linux -- TransIP BV | https://www.transip.nl/