Public bug reported:
# Recurring soft lockup / RCU stall on dentry (VFS) and mm spinlocks — 3
incidents across kernel 7.0.0-28/29/30-generic, HWE 24.04
## Summary
Since 2026-08-17 this system has hit a hard freeze requiring a forced
power-off three times, roughly once every 3-7 days. Each time,
`dmesg`/`journalctl` shows a `watchdog: BUG: soft lockup` on one or more
CPUs, spinning inside `native_queued_spin_lock_slowpath`, which
escalates into `rcu_preempt self-detected stall` and eventually a fully
unresponsive system (no console, no SSH, no clean shutdown). A forced
power cycle is the only recovery.
The offending userspace process is **different every time** (a
Rust/tokio-based app, a Bun JS runtime worker, and `gitk`), but the
kernel-side lock path is consistently the same family: **VFS dentry
lookup/lockref** (two of three incidents) or a closely related **mm
page-table lock** (one incident). This looks like a kernel-level lock
livelock/regression rather than an application bug.
## Environment
- Ubuntu 24.04.4 LTS (noble), HWE kernel stack (`linux-generic-hwe-24.04`)
- Kernel versions seen affected: `7.0.0-28-generic`, `7.0.0-29-generic`,
`7.0.0-30-generic` (currently running `7.0.0-30-generic`, latest available in
`noble-updates`/`noble-security` as of 2026-08-24 — no newer point release to
test yet)
- CPU: Intel Core Ultra 5 125H (Meteor Lake, 14C/18T)
- Discrete GPU: NVIDIA RTX 5060 Ti via an external OCuLink→PCIe dock
(proprietary driver 595.84, CUDA 13.2) — added 2026-08-16, one day before the
first incident. **Investigated and ruled out as the direct cause** — see "Why I
don't think this is GPU-related" below — but noting it since it's the one
hardware change that precedes the onset of this pattern.
- No custom kernel patches; stock Ubuntu HWE kernel packages only.
## Incident timeline
### Incident 1 — 2026-08-17, kernel 7.0.0-28-generic
Symptom order: mouse/keyboard alive, everything else hung (terminal froze
mid-command, external commands stopped working, network dead in-browser).
Forced power-off.
`journalctl -k -b -1` showed CPU stuck ~1061s spinning on an mm spinlock:
```
pte_offset_map_lock (via filemap_map_pages)
```
Offending thread: `tokio-rt-worker` (Rust/tokio async runtime worker thread;
parent process unidentified post-reboot).
### Incident 2 — 2026-08-21, kernel 7.0.0-29-generic
Symptom order: keyboard blocked first, mouse a few seconds later, then total
freeze.
`journalctl -b -1 -p warning` showed CPU stuck 1055s+ spinning in:
```
native_queued_spin_lock_slowpath
-> list_lru_del (VFS dentry LRU lock)
-> __dentry_kill
-> dput
-> filename_renameat2 (i.e. a rename() syscall)
```
Offending thread: `Bun Pool 0` (a Bun JS runtime worker thread — Bun is bundled
inside some other application, not a standalone install; exact parent
unrecoverable post-reboot). First lockup logged 13:21:15, escalated to
`rcu_preempt self-detected stall` at 13:39:47, fully unresponsive by ~13:40.
### Incident 3 — 2026-08-24, kernel 7.0.0-30-generic
First soft lockup logged 04:58:49 on one CPU, offending process `gitk`
(Tk-based git history browser). Spinning in:
```
native_queued_spin_lock_slowpath
-> _raw_spin_lock
-> lockref_get_not_dead
-> __legitimize_path
-> try_to_unlazy
-> complete_walk
-> do_open
-> path_openat
-> do_file_open
-> do_sys_openat2
-> __x64_sys_openat
```
— i.e. an `openat()` syscall contending the same dentry lockref family as
incident 2. Over the following ~82 minutes, 8 separate `gitk` kernel threads
(different PIDs) independently soft-locked, spreading across 8 different CPUs —
the lock contention cascaded system-wide rather than staying pinned to one
core. Multiple `rcu_preempt self-detected stall` messages followed from ~06:11
onward. Log (and presumably the forced power-off) ends at 06:21:13. No
OOM-killer activity. The `gitk` invocation was against a git repository with
~560 branches / 12 worktrees — plausibly a heavier-than-usual dentry/refs
workload, though this is circumstantial.
## Why I don't think this is GPU-related
I looked closely at whether the NVIDIA eGPU (added 2026-08-16, the day
before incident 1) is responsible, since all three incidents postdate
that change:
- None of the three call traces above touch NVIDIA driver, DRM, or
GPU-scheduler code — all three are pure VFS-dentry or mm/page-table lock paths,
in unrelated userspace programs that do no GPU work.
- No NVIDIA Xid errors, no PCIe AER errors attributable to the GPU or its PCIe
port, in any of the three affected boots. (There were a handful of unrelated,
self-corrected AER *correctable* errors from an NVMe controller on one boot,
hours before the freeze — not the GPU, and correctable errors are expected
background noise, not implicated.)
- A GPU health-monitoring timer sampling every 5 minutes shows the GPU
completely idle (0% util, idle power draw, PCIe link in its idle Gen1 state) at
the last sample before incident 3's freeze began, and the timer's own systemd
log confirms it ran cleanly on schedule right up until the freeze, then resumed
cleanly right after reboot — consistent with the whole system hanging (not an
I/O-specific or GPU-specific event), and with the GPU not doing anything
unusual when the lockup started.
So I don't believe the GPU or its driver triggers this. The more likely
(unconfirmed) explanation for the timing is that adding the eGPU also
meant running heavier, more sustained workloads (large ML models
resident in memory, more PCIe/interrupt complexity from the added dock)
— increasing overall system load enough to make a pre-existing rare
kernel locking race manifest more often, rather than the GPU path itself
being at fault.
## Question / ask
- Is this a known regression in the `7.0.0-2x/3x-generic` HWE kernel series
(dentry lockref / mm page-table lock livelock under contention)?
- Is there a fix already queued for a kernel point release beyond `-30`, or
should this be tracked as a new bug?
- Any diagnostic I should capture on the next occurrence beyond `journalctl -b
-1 -p warning` (e.g., should I enable `kdump` for a full vmcore, given apport
does not capture soft-lockup freezes)?
Happy to attach full `journalctl -b -1 -k` output for any of the three
incidents (large, so not inlined here) and `lspci -vvv`/`lscpu` output
on request.
ProblemType: Bug
DistroRelease: Ubuntu 24.04
Package: linux-image-7.0.0-30-generic 7.0.0-30.30~24.04.1
ProcVersionSignature: Ubuntu 7.0.0-30.30~24.04.1-generic 7.0.12
Uname: Linux 7.0.0-30-generic x86_64
ApportVersion: 2.28.3-0ubuntu0.1
Architecture: amd64
CasperMD5CheckResult: pass
CurrentDesktop: ubuntu:GNOME
Date: Mon Aug 24 06:40:58 2026
InstallationDate: Installed on 2025-12-03 (264 days ago)
InstallationMedia: Ubuntu 24.04.3 LTS "Noble Numbat" - Release amd64
(20250805.1)
MachineType: Micro Computer (HK) Tech Limited AI Series
ProcFB: 0 i915drmfb
ProcKernelCmdLine: BOOT_IMAGE=/boot/vmlinuz-7.0.0-30-generic
root=UUID=3c033fd3-1ebc-4d46-8b53-c3ed69fe253c ro quiet splash vt.handoff=7
PulseList: Error: command ['pacmd', 'list'] failed with exit code 1: No
PulseAudio daemon running, or not running as session daemon.
RelatedPackageVersions:
linux-restricted-modules-7.0.0-30-generic N/A
linux-backports-modules-7.0.0-30-generic N/A
linux-firmware 20240318.git3b128b60-0ubuntu2.29
SourcePackage: linux-hwe-7.0
UpgradeStatus: No upgrade log present (probably fresh install)
dmi.bios.date: 06/19/2025
dmi.bios.release: 5.32
dmi.bios.vendor: American Megatrends International, LLC.
dmi.bios.version: 1.01
dmi.board.asset.tag: Default string
dmi.board.name: ARBSC
dmi.board.vendor: Shenzhen Meigao Electronic Equipment Co.,Ltd
dmi.board.version: 1.0
dmi.chassis.asset.tag: Default string
dmi.chassis.type: 3
dmi.chassis.vendor: Default string
dmi.chassis.version: Default string
dmi.ec.firmware.release: 1.0
dmi.modalias:
dmi:bvnAmericanMegatrendsInternational,LLC.:bvr1.01:bd06/19/2025:br5.32:efr1.0:svnMicroComputer(HK)TechLimited:pnAISeries:pvr1.0:rvnShenzhenMeigaoElectronicEquipmentCo.,Ltd:rnARBSC:rvr1.0:cvnDefaultstring:ct3:cvrDefaultstring:skuMGARBSC:pfaAI:
dmi.product.family: AI
dmi.product.name: AI Series
dmi.product.sku: MGARBSC
dmi.product.version: 1.0
dmi.sys.vendor: Micro Computer (HK) Tech Limited
** Affects: linux-hwe-7.0 (Ubuntu)
Importance: Undecided
Status: New
** Tags: amd64 apport-bug noble
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2164907
Title:
Recurring soft lockup / RCU stall on dentry lockref (VFS), 3
incidents across kernel 7.0.0-28/29/30-generic
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-hwe-7.0/+bug/2164907/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
