Skip to content

Framework Desktop reboots on every shutdown - sync flood #200

@melpomene

Description

@melpomene

Device Information

System Model or SKU

  • Framework Desktop (AMD Ryzen™ AI 300 PRO Series)

BIOS VERSION

03.04 (Insyde)

DIY Edition information

Memory: 128 GB LPDDR5 @ 8000 MT/s — Micron MT62F4G32D8DV-023 WT ×8 (on-package unified memory)

Storage:

  • WD_BLACK SN850X 4TB ×2 — Firmware 624361WD (internal)
  • WD_BLACK SN770 2TB — Firmware 731120WD (via usb)

Port/Peripheral information

Apple Studio Display connected via dual-cable (DisplayPort + USB) to the Framework Desktop.

Desktop expansion cards:

  1. SD card reader
  2. MicroSD card reader

Describe the bug

systemctl poweroff triggers an AMD data fabric sync flood (0x08000800), causing the machine to hard-reset and reboot instead of powering off. This is 100% reproducible on every poweroff attempt.

systemctl halt (which performs the same systemd teardown but does not issue the ACPI S5 command) shuts down cleanly every time — the machine stays off.

The difference between the two commands is a single step: poweroff sends the ACPI S5 command; halt does not. This isolates the root cause to the BIOS ACPI S5 implementation.


Steps To Reproduce

  1. Boot to a logged-in desktop session
  2. Run sudo systemctl poweroff
  3. Machine shuts down services normally, then hard-resets and reboots to LUKS decrypt prompt instead of powering off

For comparison: sudo systemctl halt → machine stays off cleanly every time.


Expected behavior

systemctl poweroff should power the machine off. It should not reboot.


Operating System

  • OS/Distribution: Arch Linux
  • Kernel: 6.19.11-arch1-1
  • Kernel options: cryptdevice=UUID=<UUID>:root root=/dev/mapper/root zswap.enabled=0 rw rootfstype=ext4 usbcore.autosuspend=-1 amdgpu.sg_display=0 thunderbolt.clx=false xhci_hcd.quirks=0x80 pcie_ports=native pcie_ecrc=on

Additional context

Reset reason logged on every subsequent boot:

x86/amd: Previous system reset reason [0x08000800]: an uncorrected error caused a data fabric sync flood event

Confirmed this is triggered by the ACPI S5 command, not by:

  • Display connection — reboots with Apple Studio Display disconnected
  • USB/Thunderbolt wakeup sources — disabled all XHC0/1/3/4 and NHI0/1 wakeup sources before poweroff, sync flood still occurs
  • NVMe firmware — SN770 updated from 731100WD to 731120WD, no change
  • reboot=efi kernel parameter — no change
  • PCIe AER errors — enabled pcie_ports=native pcie_ecrc=on to force Linux native AER handling; zero AER errors logged during shutdown; sync flood occurs after journald stops, below the level AER can observe
  • Power on AC is set to off in the bios

Key isolation test:

Command ACPI S5 issued Result
systemctl poweroff Yes Sync flood 0x08000800, reboots
systemctl halt No Clean shutdown, stays off

The ACPI S5 path in BIOS 03.04 appears to trigger an uncorrected error in the AMD data fabric during the S5 transition. This is not recoverable at the OS level and requires a BIOS fix as far as I can tell.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions