On Wed, Sep 13, 2023 at 09:00:03PM -0400, Tyler Stachecki wrote:
qemu then both ceases to put the remaining (non-XSAVE) x86 architectural state into KVM and makes the fateful mistake of resuming the guest anyways. This usually results in immediate guest corruption, silent or not.
I just want to highlight that although this is probably more of a bug with respect to how qemu is handling things, the original patches from Leo are starting to appear in many distro stable kernels and are really putting a spanner in the works for maintaining VMs that are long-lived in nature.
At present, if you take the fix for PKRU migration issues (or if you are just in need a more recent kernel), you are dealt with a situation where live- migrating VMs to a kernel patched for the PKRU issue from one that is not potentially crashes or corrupts skads of VMs.
There is no fix for qemu that I am aware of yet. Although, I am willing to look into one if that is more palatable, I filed this patch on the premise of "don't break userspace"...