On Thu, 2024-10-03 at 10:22 -0400, Chinmaya Mahesh wrote: > Hello, > I was reading the KVM documentation that mentions live-migrating an L1 > guest with a running L2 guest on AMD systems results in undefined > behavior: https://www.kernel.org/doc/html/latest/virt/kvm/x86/running-nested-guests.html#live-migration-with-nested-kvm. > However, we noticed that this documentation hasn't been updated in a > while (last edit of that section was May 6 2020 according to the > blame), and notably there have been some AMD nested migration > improvements in June 2020: > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=039aeb9deb9291f3b19c375a8bc6fa7f768996cc. > > We did some stress testing of live migrating systems with nested VMs > on AMD and noticed that they seem to be running fine with no crashes > so far. Do we know if the docs are stale on this? If we have tested > this and it seems to work fine, are we taking on a lot of risk by live > migrating VMs with L2 vms running inside of them on AMD? Are there > specific workloads that are known to result in undefined behavior more > frequently? > > We have tested 2 scenarios: KVM running in KVM, and Hyper-V in KVM, > both on AMD EPYC CPUs. Both seem to do fine with repeated live > migrations although we are yet to try this long-term. > > Thanks! > Hi, While nested migration, and nested virtualization as a whole, is not supported by RedHat AFAIK, I few years ago did an extensive round of testing and bugfixing in the area of AMD nested virtualization and especially nested migration on my Zen2 machine - this way my way of studying KVM internals, and having some fun. So I happy to hear that it works, and yes the documentation must be updated, I'll send a patch for that soon. Best regards, Maxim Levitsky
