Public bug reported:

**Title:** Kernel panic "Attempted to kill the idle task" on AMD Opteron
6300 series (multi-node NUMA) under KVM load - linux 6.8.0-137-generic
(24.04)

**Summary:**

On 5 out of 5 identical HP ProLiant DL385p Gen8 servers (AMD Opteron
6386 SE, 4 NUMA nodes each, running as KVM/libvirt hypervisors with
multiple guest VMs), kernel panics with "Attempted to kill the idle
task!" occurred within roughly 1 to 24 hours after upgrading from Ubuntu
22.04 (kernel 5.15.0-187-generic, stable in production for years) to
Ubuntu 24.04 (kernel 6.8.0-137-generic) via do-release-upgrade. Pinning
the boot kernel back to 5.15.0-187-generic (via GRUB_DEFAULT) while
keeping the 24.04 userspace fully resolved the instability on all
affected hosts, with no further crashes since.

**Hardware:**
- HP ProLiant DL385p Gen8, BIOS A28 (03/14/2018)
- AMD Opteron 6386 SE, microcode 0x6000852
- 4 NUMA nodes per host
- KVM/libvirt hypervisor, 9-20 guest VMs per host under production load (mixed 
Kubernetes control-plane/worker nodes, haproxy, monitoring, storage services)

**Steps to reproduce:**
1. Fresh `do-release-upgrade` from 22.04 (5.15.0-187-generic) to 24.04 
(6.8.0-137-generic) on the described hardware
2. Boot into 6.8.0-137-generic with multiple KVM guests running under 
moderate-to-high, mixed CPU/memory load
3. Panic occurs non-deterministically, anywhere from ~1 hour to ~24 hours after 
boot; not tied to any single reproducible user action

**Observed panic signature (consistent across all 5 affected hosts):**
```
Kernel panic - not syncing: Attempted to kill the idle task!
Call Trace:
 <TASK>
 ? multi_cpu_stop+0xa0/0x120
 ? __pfx_multi_cpu_stop+0x10/0x10
 cpu_stopper_thread+0x9c/0x170
 ? __pfx_smpboot_thread_fn+0x10/0x10
 smpboot_thread_fn+0xe3/0x1e0
 kthread+0xf2/0x120
 ? __pfx_kthread+0x10/0x10
 ret_from_fork+0x47/0x70
 ? __pfx_kthread+0x10/0x10
 ret_from_fork_asm+0x1b/0x30
 </TASK>
```

One instance on a related host (same hardware) additionally showed
`membarrier_private_expedited+0x1cf/0x2d0` in RIP immediately preceding
the panic.

**Possible related upstream discussion:**

A 2018 LKML thread by Matt Fleming (SUSE), "cpu stopper threads and load
balancing leads to deadlock"
(https://lkml.iu.edu/hypermail/linux/kernel/1804.2/01642.html),
describes a closely matching mechanism: a CPU stopper thread entering
idle/active load balancing while, concurrently, NUMA balancing on
another CPU attempts to migrate a task via `stop_two_cpus()`, resulting
in a deadlock over the stopper lock. Given that our affected hosts are
multi-node NUMA systems (4 nodes) and the panic call trace centers on
`cpu_stopper_thread`/`multi_cpu_stop`, this looks like a strong
candidate mechanism, though we cannot confirm whether it is the same
unresolved issue, a regression of a past fix, or a new but related bug
in 6.8.

**Workaround applied:**

GRUB kernel pinning to boot `5.15.0-187-generic` while keeping the 24.04
userspace intact (`GRUB_DEFAULT` set to the jammy kernel's advanced boot
entry ID, `update-grub`, reboot). All 5 hosts stable since applying this
workaround.

**Additional diagnostics performed:**
- Hardware (BIOS version, CPU model, microcode) verified identical across all 5 
affected hosts
- Memtester (4GB sample, full test cycle) passed clean on the two hosts tested
- HP Integrated Management Log (IML) reviewed on two hosts: one shows a 
documented Uncorrectable CPU Machine Check Exception from 2020 (Bank 0x04, both 
physical processors) - possibly a contributing factor for that specific host, 
but does not explain the same panic signature occurring on hosts without such 
MCE history
- Network/bonding layers (LACP, active-backup) verified healthy and unrelated 
to the panics
- `dmesg`/`journalctl` EDAC output shows no correctable/uncorrectable memory 
errors logged around panic times on the hosts checked

**Please let me know what additional information (full dmesg, kernel
config, /proc/cpuinfo, NUMA topology via `numactl --hardware`, etc.)
would help diagnose this further.**

** Affects: linux (Ubuntu)
     Importance: Undecided
         Status: New

** Attachment added: "version_signature and lspci output from affected host"
   https://bugs.launchpad.net/bugs/2163642/+attachment/5992749/+files/logs.zip

-- 
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