Re: [libvirt] [PATCH 2/5] Introduce SMM feature

On 01.08.2016 21:10, Laszlo Ersek wrote: > On 08/01/16 19:05, Gerd Hoffmann wrote: >> Hi, >> >>> If it does, then the host kernel should be 4.4 or later. (If we want to >>> capture version numbers, then for each minor release, the latest patch >>> level should be required: 4.4.16, 4.5.7, 4.6.5, 4.7.0, because there >>> have been a number of KVM tweaks and stable backports for SMM support.) >> >> Better check /dev/kvm for the KVM_CAP_X86_SMM capability, otherwise the >> rhel-7 kernel is out even though smm is backported. >> >> Probably libvirt already has checks for other caps so it should be easy. > > The problem is that KVM_CAP_X86_SMM does not guarantee that all the > tricky bugfixes are present in the host kernel. > > Moreover, libvirt does not validate nor check guest kernel. And the only /dev/kvm/ caps we check are: KVM_CAP_IOMMU - to see if legacy PCI passthrough works KVM_CAP_PPC_SMT - KVM_CAP_MAX_VCPUS - KVM_CAP_NR_VCPUS - for some vCPUs counts I don't think libvirt should check for KVM_CAP_x86_SMM. I mean, if we go down the rabbit hole, libvirt would have to check everything. The buck has to stop somewhere.