Public bug reported:
## Problem / impact
The **linux-hwe-6.17** kernel (6.17.0-35 / 6.17.0-40) **causes intermittent
idle hard-hangs** on an **Intel N150** (Alder Lake-N / "Twin Lake") mini-PC
(GMKtec NucBox G3 Plus) running Ubuntu 24.04.1 LTS. After hours to days of
otherwise-normal uptime, the box becomes fully unresponsive (no network, no
SSH, no console) and **does NOT self-reboot** — a **manual power-cycle is
required** to recover. It is an always-on home server/router, so each hang is a
total outage. On the 6.8 GA kernel the same hardware was stable for months
(107-day and 41-day uptimes; zero hangs/BERT events in 90 days of journal).
## Regression — yes
- Working: `linux-image-6.8.0-134-generic` (GA).
- Failing: `linux-image-6.17.0-35-generic` and `linux-image-6.17.0-40-generic`
(HWE). Two idle hangs in one day (2026-07-15).
- Prior 90 days of journal: **zero BERT hardware-error events and zero
unclean-shutdown markers** — the hangs began only with the 6.17 kernel.
## Reproducibility
Intermittent, **not on-demand**. Observed twice in one day after ~16 days and
~5.6 hours of uptime respectively; other boots/short runs on 6.17 did not hang.
The probabilistic, idle-only nature points to a deep C-state trigger (see
below).
## Hardware
- GMKtec NucBox G3 Plus
- CPU: Intel N150 (family 6, model 0xbe; Alder Lake-N / Twin Lake), 4 cores, no
HT
- RAM: 16 GB DDR4-3200, non-ECC
- BIOS: AMI (EDK2) version 5.27, dated 2024-11-06
- NIC: Intel I22x (igc); GPU: Alder Lake-N iGPU (i915); cpuidle driver
`intel_idle`
## Symptom detail (2026-07-15)
- Hang #1 on 6.17.0-35 after ~16 days uptime; Hang #2 on 6.17.0-40 after ~5.6
hours.
- Both at **near-idle** (only per-minute cron jobs running) — no load spike, no
OOM, no thermal event.
- **No oops/panic/soft-lockup/MCE** is logged before the hang — the journal
just stops mid-normal-operation; the box stays dead until manually power-cycled.
- After each recovery the kernel logs:
BERT: [Hardware Error]: Skipped 1 error records
BERT: Total records found: 1
The BERT boot-error region (3300 bytes, read via `/dev/mem`) is one
**FATAL-severity CPER record** with an OEM section GUID
(`81212A96-09ED-4996-9471-8D729C8E69ED`, AMI EDK2). Hexdump attached.
## Ruled out
- Memory: EDAC igen6 `ce_count=0` / `ue_count=0` (non-ECC; rasdaemon now
installed).
- Disk: SMART PASSED on both drives, 0 reallocated sectors.
- Thermal: CPU 73–76 °C (tjmax 105), NVMe 52–64 °C — normal.
- Capacity: 5/15 GB RAM used, swap untouched; rootfs 44 %.
- Hardware was clean for 90 days prior — only the 6.17 kernel triggers it.
## Likely cause (deep C-state / idle path)
The 6.17 kernel exposes deep package C-states on this N150 part and the CPU
idles in C8/C10 **millions of times** (~6.6M C8 / ~7.4M C10 entries across
cores in the first ~1h45m of uptime — see cpuidle-states.txt; counters are
cumulative since boot). The 6.8 GA kernel predates Twin Lake enablement and did
not expose these deep states. Entering/exiting C8/C10 appears to trip a latent
fault in the N150 SoC / GMKtec firmware → fatal hardware error → dead hang
requiring power-cycle. This matches the well-known Alder Lake-N / Twin Lake
deep-C-state instability class on newer kernels.
## Workaround
`intel_idle.max_cstate=2` on the kernel cmdline (cap deep C-states, keep C6) —
applied today, both persistently (GRUB) and live (sysfs `disable=1` on C8/C10).
Hangs appear to stop; stability under observation. This is both the workaround
and confirmation of the deep-idle trigger.
## Request
1. Confirm whether **linux-hwe-6.17 is known to be unstable on Intel N150 /
Alder Lake-N**, and whether a newer kernel already fixes it.
2. Consider a **default C-state quirk** for Alder Lake-N / Twin Lake
(N100/N150/N97) on affected OEM platforms — the failure is total (hard hang of
an always-on box) and this hardware is common.
3. Happy to provide the full BERT CPER hexdump, journal excerpts, or run an
`intel_idle.max_cstate=1` soak / PCIe ASPM tests on request.
** Affects: ubuntu
Importance: Undecided
Status: New
** Tags: intel-n150 kernel-power regression
** Attachment added: "system-info.txt"
https://bugs.launchpad.net/bugs/2160711/+attachment/5982037/+files/system-info.txt
** Project changed: launchpad => ubuntu
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2160711
Title:
linux-hwe-6.17 causes intermittent idle hard-hangs on Intel N150
(Alder Lake-N); needs power-cycle (FATAL BERT); regression from 6.8 GA
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+bug/2160711/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
