Minisforum HM90 / Ryzen 9 4900H – recurring kernel Oops, MCE errors and complete host freezes

[Surfer2010]

Hi,

I have been troubleshooting recurring instability on a Minisforum HM90 running Proxmox VE and would appreciate some help determining whether this looks like a hardware, kernel, firmware or AMD/Renoir issue.

Hardware​

• Minisforum / BESSTAR HM90
• AMD Ryzen 9 4900H (8C/16T)
• 32 GB RAM
• BIOS: 5.16, 10/13/2021
• Hardware ID: BESSTAR TECH LIMITED HM90/HM90

Proxmox​

Current system:

proxmox-ve: 9.2.0pve-manager: 9.2.4running kernel: 6.17.13-15-pveamd64-microcode: 3.20251202.1~bpo13+1pve-qemu-kvm: 11.0.2-1qemu-server: 9.1.18pve-container: 6.1.10zfsutils-linux: 2.4.3-pve1

Installed kernels include:

7.0.14-4-pve6.17.13-15-pve6.17.4-2-pve6.8.12-33-pve6.8.12-18-pve6.5.13-6-pve6.5.11-8-pve

I currently use 6.17.13-15-pve because I have also experienced severe problems with other kernels.

Problem​

The HM90 can run normally for several days and then becomes unresponsive.

Typical symptoms:

• SSH unavailable
• Proxmox GUI unavailable
• sometimes no network response
• kernel Oops/call traces on the local console
• hard reset eventually required

The problem is intermittent and not reliably reproducible.

Earlier MCE / CPU errors​

During previous incidents I found Machine Check / hardware errors associated with logical CPU 6, including messages referring to:

CPU: 6Instruction Fetch Unit Ext. ErrorL1 Cache

CPU 6 and CPU 7 are SMT siblings of the same physical core, so I started testing with both disabled.

Current state:

cat /sys/devices/system/cpu/online0-5,8-15
CPU 6: 0CPU 7: 0

A systemd service disables them during boot.

Kernel Oops despite CPU 6/7 being disabled​

Unfortunately this did not eliminate the problem.

On September 7 CPU 6 and CPU 7 were successfully disabled at 12:41:06:

Sep 07 12:41:06 kernel: smpboot: CPU 6 is now offlineSep 07 12:41:06 kernel: smpboot: CPU 7 is now offline

About 47 seconds later another kernel Oops occurred, this time on CPU 1:

Sep 07 12:41:53 kernel: BUG: unable to handle page fault for address: ffffffffffffff8eSep 07 12:41:53 kernel: #PF: supervisor write access in kernel modeSep 07 12:41:53 kernel: #PF: error_code(0x0002) - not-present pageSep 07 12:41:53 kernel: Oops: Oops: 0002 [#1] SMP NOPTI
CPU: 1UID: 100000PID: 1666Comm: (sd-executor)Tainted: P O6.17.13-15-pve
Hardware name: BESSTAR TECH LIMITED HM90/HM90BIOS 5.16 10/13/2021
RIP: mas_wr_store_type+0x210/0x670CR2: ffffffffffffff8e

Relevant call trace:

mas_storedup_mmapcopy_processkernel_clone__do_sys_clone__x64_sys_clonex64_sys_calldo_syscall_64

Immediately before this Oops, LXC 112 was being started.

Therefore the Oops happened after the suspected CPU core had already been taken offline.

Other crashes​

During another complete freeze, the local console showed KVM-related traces including:

kvm_arch_vcpu_ioctl_run
kvm_vcpu_ioctl
__apic_accept_irq

I initially suspected a VM as the trigger. One VM runs a relatively CPU-intensive local AI workload.

However, this does not explain all incidents. The September 7 Oops happened during LXC startup, and the AI VM is currently stopped.

Current status​

The host is currently running with CPU 6/7 disabled:

Kernel: 6.17.13-15-pveCPU online: 0-5,8-15CPU 6: offlineCPU 7: offline

At the last check:

Uptime: 2 days, 2:28Load average: 3.16, 3.16, 3.10
RAM:30 GiB total1.7 GiB used28 GiB free0 swap used
CPU Tctl: 48 CNVMe: 31.9 C

No failed systemd services are reported.

Questions​

At first I suspected a defective physical CPU core because of the MCE / L1 Instruction Fetch errors on CPU 6. However, the later page fault on CPU 1 occurred after CPU 6/7 were already offline.
Could these different errors have a common underlying cause such as RAM corruption, CPU/SoC memory controller problems or power instability?

In particular:

1. How would you interpret the mas_wr_store_type -> mas_store -> dup_mmap -> copy_process page fault?
2. Could this be hardware memory corruption rather than a kernel bug?
3. Are there known stability issues with Minisforum HM90 / Ryzen 9 4900H (Renoir) systems under Proxmox?
4. Are there known relevant BIOS, C-state, power-management or AMD-specific settings for the HM90?
5. Which of the installed kernels would be the most useful comparison test?
6. Would you recommend extensive RAM testing or other specific hardware tests?
7. What additional logs or diagnostics would be most useful after the next freeze?

My main goal is to determine whether this HM90 has an actual hardware defect or whether this is a kernel/firmware/software issue.

I have persistent logs and photos of the console from previous crashes and can provide the complete MCE/Oops output if needed.