We got bitten by silent "put cpu failures" multiple times, the most recent one is RHEL-7558. We need to fix this for good at some point, or we can keep getting silent CPU failures in guest and they're hard to diagnose.

Here the problem is after migration, QEMU will setup destination vCPUs with source vCPU states, however if anything wrong happens during this procedure, instead of properly failing the migration, QEMU currently will ignore that failure.

One example is FPU, consider after migration applying FPU state failed but QEMU ignores it, it will cause guest to start without knowing that it has a corrupted FPU state. It can cause all kinds of guest OS failures due to wrong vCPU state randomly.

For more information, see [1,2].

Use this bug to track that work. We need to upstream the patchset in some form [2], then backport to all relevant downstream branches.

[1] https://issues.redhat.com/browse/RHEL-7558?focusedId=23698892&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-23698892
[2] https://lore.kernel.org/all/20220617144857.34189-1-peterx@redhat.com/