Public bug reported:
# Probable AMDGPU-related reboot regression on Ubuntu HWE 7.0
Prepared locally on 2026-09-13. Tested on Linux Mint 22.3 using Ubuntu-
provided kernels. This has not yet been reproduced on an Ubuntu
installation.
## Summary
On Linux Mint 22.3 (Ubuntu 24.04 base), restarting with Ubuntu HWE
kernel 7.0 can leave the screen off and fans running/ramping, requiring
a power-button reset. Kernel 6.17.0-40 has restarted successfully with
graphics enabled and is the current workaround. Normal power-off and
restart directly from BIOS have also worked.
The strongest comparison is between two text-mode boots of the same
7.0.0-31 kernel: reboot hung with Intel graphics blocked and AMDGPU
initialized; reboot succeeded after additionally blocking AMDGPU. This
strongly implicates the AMDGPU initialization/shutdown path or its
interaction with other kernel components. It does not establish the
exact defective function or commit. A malformed earlier blacklist test
reportedly restarted with Intel still loaded, so intermittency remains
possible.
## Environment
- Linux Mint 22.3, Ubuntu noble base, x86_64.
- Working installed kernel: 6.17.0-40-generic, package 6.17.0-40.40~24.04.1.
- Failing kernel: 7.0.0-31-generic, package 7.0.0-31.31~24.04.1, upstream
version reported as 7.0.14.
- Also reported failing: 7.0.0-28 and 7.0.0-30.
- Discrete GPU: AMD Navi 31, PCI ID 1002:744c, ASRock subsystem 1849:5308. PCI
naming covers RX 7900 XT/XTX variants; exact retail model not established here.
- Integrated GPU: Intel Alder Lake-S GT1, PCI ID 8086:4680, i915 driver under
normal operation.
- Motherboard: MSI PRO Z690-A WIFI (MS-7D25).
- BIOS: A.D0, dated 2023-07-10.
## Tests and evidence
| Test | Boot configuration | Reported result |
| --- | --- | --- |
| Installed 6.17.0-40 | Normal graphical boot | Reboot worked |
| Mint live USB, 6.14.0-37 | Live desktop | Reboot worked |
| 7.0.0-31 | Normal desktop | Reboot hung |
| 7.0.0-31 | `systemd.unit=multi-user.target` | Reboot hung; LightDM did not
start |
| 7.0.0-31 | `systemd.unit=multi-user.target nomodeset` | Reboot worked |
| Test B, 7.0.0-31 | `systemd.unit=multi-user.target module_blacklist=i915,xe`
| Latest test hung |
| Test C, 7.0.0-31 | `systemd.unit=multi-user.target
module_blacklist=i915,xe,amdgpu` | Reboot worked |
Test B on September 13 at 10:26 CDT recorded:
```text
Module i915 is blacklisted
Module xe is blacklisted
[drm] Initialized amdgpu 3.64.0 for 0000:03:00.0 on minor 1
```
Test C on September 13 at 10:33 CDT recorded:
```text
Module i915 is blacklisted
Module xe is blacklisted
Module amdgpu is blacklisted
```
The preserved journal reaches the final systemd shutdown stage and stops
when journald terminates. No preserved stack trace identifies the hang.
Reaching reboot.target or seeing Journal stopped alone does not prove a
reboot succeeded; success/failure outcomes above come from the user's
physical observations.
Earlier 7.0 boots logged AMD display REG_WAIT timeouts, but those
messages alone do not establish the cause. A temporary EFI reset-method
test also hung. Current measurements taken during normal operation did
not show overheating; temperatures during the actual hang were not
captured.
## Current workaround
GRUB defaults to 6.17.0-40-generic with normal graphics drivers enabled.
The kernel image and module packages are marked manually installed.
Kernel updates remain enabled, but newly installed kernels will not
automatically replace this explicit default.
Local configuration:
- `/etc/default/grub.d/99-working-kernel.cfg`: working default and five-second
menu.
- `/etc/grub.d/42_reboot_diagnostics`: optional tests A, B, C.
Do not use the all-GPU blacklist or nomodeset as the normal daily boot
configuration if GPU acceleration is required.
## Suggested maintainer follow-up
Review as an AMDGPU-related regression between 6.17 and 7.0. Identify a
relevant fixed distribution kernel or request targeted late-shutdown
traces. Repeat the B/C comparison if needed because the failure may be
intermittent. Further isolation should be based on maintainer guidance
rather than changing unrelated BIOS settings or disabling broad power-
management features.
Related reports found during research involve other AMD hardware and do
not establish that the same defect or fix applies here. No specific
published fix was verified for this machine.
** Affects: linux (Ubuntu)
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2167190
Title:
[amdgpu][Navi 31] Reboot hangs on HWE 7.0; 6.17 works; blocking amdgpu
avoids hang
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2167190/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
