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

Reply via email to