Hi Radovan, Moving the AMD hosts to noble HWE 7.0 (7.0.0-38) is definitely the right call for production stability.
Your latest log provides the missing link: Notice the path during the lockup: `zap_page_range_single <- madvise (qemu worker, kvm_mmu_notifier_invalidate_range_end in the path)`. In 6.8, KVM's TDP MMU had a known race where `mmu_notifier` unmap callbacks did not synchronously zap all invalid roots before returning. As a result, KVM retained stale references to freed backing pages after QEMU unmapped them via madvise. When core kernel structures (CFS runqueues in `dequeue_entity`, memslot lists, or stack frames) were later allocated in those freed pages, concurrent vCPU access caused memory corruption. This explains why the crashes appeared days apart in completely unrelated subsystems (`dequeue_entity`, `kvm_replace_memslot`, `start_hv_timer`, and the `multi_cpu_stop` deadlock). In 6.9 through 6.13 and 7.0, the KVM TDP MMU invalidation sequencing was heavily reworked to guarantee all roots are synchronously zapped before mmu_notifier finishes. Because this fix spans an extensive series of KVM MMU lifecycle commits rather than a single standalone patch, backporting it to 6.8 is high-risk. Sticking with the official noble HWE 7.0 kernel is the cleanest and safest resolution. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2163642 Title: Kernel panic "Attempted to kill the idle task" on AMD Opteron multi- node NUMA under KVM (6.8.0-137) To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2163642/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
