Hi Krystian, Important update - **srv06 (mainline v6.16.0) has now panicked** after running stable for approximately 25+ days.
However, the crash signature is **different** from the original bug: ``` [1374450.499466] ---[ end Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: start_hv_timer+0x171/0x180 [kvm] ]--- ``` This is a stack-protector triggered panic in KVM's `start_hv_timer()`, not the original `multi_cpu_stop`/`cpu_stopper_thread`/"Attempted to kill the idle task" signature we've been tracking. This raises a few possibilities I'm not sure how to weigh: 1. This could be a related but distinct bug also affecting KVM on this AMD Opteron 6300 / multi-node NUMA hardware, exposed by a much longer uptime window than the original panics (1-24h) 2. It could mean the original bug's fix (if commit 9709eb0f845b is indeed relevant) only reduces the frequency/timing of the failure rather than eliminating it outright, and this is a delayed manifestation of the same underlying instability 3. It could be entirely unrelated hardware-adjacent flakiness on this specific host (though we haven't seen ECC/MCE errors logged) For context, at time of the panic all other test hosts were still stable: - srv05 (HWE 7.0.0-29): running since 2026-08-18, no panic - srv07 (mainline 6.15.11): running since 2026-08-20, no panic - srv04 (mainline 6.8.12): running since ~3 days ago, no panic - srv08 (mainline 6.12.0, then reverted to Ubuntu 6.8.0-137 for a follow-up test suggested by another contributor - more below): stable throughout its 6.12.0 run except for one transient hung-task (XFS I/O) event that self-recovered Separately, a Launchpad contributor (Viktor Pashaiev) suggested a theory involving `membarrier(MEMBARRIER_CMD_PRIVATE_EXPEDITED)` relying on cpu_stopper threads, racing with KVM vCPU threads entering idle/mwait during `multi_cpu_stop` synchronization on this AMD multi-node NUMA hardware, and proposed testing with `processor.max_cstate=1 idle=nomwait` or `kvm halt_poll_ns=0` boot parameters on the original affected kernel (6.8.0-137) to isolate it. I've reinstalled 6.8.0-137 on srv08 and am currently running it *without* those parameters first as a baseline, before testing with them added, to keep the comparison clean. I'll reboot srv06 and let it continue running on 6.16.0 to see if this recurs, and will keep testing srv08 as planned. Let me know if you'd like me to prioritize differently given this new panic. -- 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
