All, I proposed making some KVM related packaging changes. Motivation: Certain aspects of KVM support have changed over time, and we should adjust to better align to the current state of affairs. Among these are: - the KVM modules have been included with the base kernel for quite some time - there is no longer a qemu / qemu-kvm code split upstream qemu handles KVM completely - on x86, since v3.7 the KVM kernel modules are auto loaded based on detection of the required hardware virtualization features Our current kvm package originally included the KVM kernel modules, as well as the QEMU binaries which supported KVM acceleration. As noted above conditions have changed to where a kvm package has little of its original purpose. The qemu package now has the basic support for using KVM, and about the only meaningful use of a kvm package would be for KVM related configuration or deliverables which would not be appropriate (for whatever reason) in the qemu package. The current content of the kvm package in OBS Virtualization includes: - dependency on qemu package - a recommends for virt-utils - the legacy qemu-kvm executable, which is simply a wrapper for the QEMU provided executable with the parameter added that enables KVM acceleration. I believe this wrapper still provides the useful feature of having an executable name, compatible with prior kvm enabled usage. - man page for qemu-kvm (which maps to qemu) - a udev rule to change permissions and group ownership of /dev/kvm to be compatible with current libvirt usage as well as non-libvirt based (ie QEMU command line) kvm group administration. - a qemu-ifup script which I find useful for testing and creating simple usage scenarios - a kvm_stat tool which allows for monitoring of the lower level KVM module activities of the running KVM guests - in the case of the SLES version of the kvm package, a place for additional modprobing, and support documentation, which in effect, rounds out the complete base support for KVM. I propose that we rename (essentially) the kvm package to be qemu-kvm. This is partly to reduce the number of maintained packages, as I envision this simply being a subpackage of qemu. This will also make versioning simpler to maintain. This package will maintain a dependency on qemu. I propose that the permissions and group ownership udev rule be moved from qemu-kvm to the base qemu package so that libvirt and etc. can function without the qemu-kvm package, providing for a slightly smaller package set and simpler configuration. I propose that the recommends dependency for virt-utils be removed. If the additional features and documentation still remaining in qemu-kvm (the qemu-kvm wrapper and man page, the sample qemu-ifup script, the kvm_stat debug utility, possible scripts for modprobing of additional kernel modules which enhance qemu-kvm usage (eg vhost-net), and the support documentation) are desired, then that package may be installed, but it is not considered essential for basic KVM support. The virt-utils package currently provides little value. It was envisioned as a place to collect generic utilities useful for either KVM or Xen virtualization, but to date very little has been done there. QEMU based utilities are now in qemu-tools, and the only utility in virt-utils is the unmaintained vm-snapshot-disk. I propose that we eliminate this package, but if there is objection, and someone relies on the vm-snapshot-disk feature, we can keep it, but not have it be part of the default set of rpms installed by KVM Host Server pattern. If there are dependencies on the kvm package name which will take time to remove, we could have that stripped down package simply require qemu-kvm in the interim. Comments on these proposed changes are of course welcome. Thanks, Bruce Rogers