Dear stable maintainers,

Please consider cherry-picking 2a23a4e1159c ("kvm: retrynx_huge_page_recovery_thread creation") to 6.12. It fixes a problem where some VMMs (crosvm, firecracker, others) may unnecessarily terminate a VM when -ENOMEM is returned to userspace for a non-fatal condition.

Report of the change's success on 6.12 and first request: https://lore.kernel.org/all/aBOPWGPTCgnUgtw-@CMGLRV3/

Example downstream bug report from libkrun: https://github.com/containers/libkrun/issues/314

We are seeing this same flaky behavior at scale on infra machines running 6.12.38 from Debian bpo.

Thanks!