Re: [BUG] linux-firmware-amdgpu 20260622 prevents clean poweroff on AMD RX 6800

[Harris Landgarten <harrisl@deeplearntech.com>]

It is unlikely that it is the firmware. I had the same issue and it started with 7.1-rc2. I am running gfx1150. The issue appears to have been fixed by 7.2-rc1
Sent from my iPhone

> On Jul 2, 2026, at 4:07 AM, Thorsten Leemhuis <regressions@leemhuis.info> wrote:
> 
> [adding a few people to the CC]
> 
> On 6/27/26 02:42, IT4roundtheW0rd wrote:
> >
> > I'd like to report a regression in linux-firmware-amdgpu introduced with
> > the 20260622 release.
> >
> > *System:*
> > - Distro: CachyOS
> > - Kernel: 7.1.1-2-cachyos
> > - GPU: AMD RX 6800-series
> > - Bootloader: GRUB
> >
> > *Problem:*
> > After updating linux-firmware-amdgpu from 1:20260519-1 to 1:20260622-1,
> > the system no longer shuts down completely when running `poweroff`.
> > Fans, keyboard LEDs, and power button LED remain active after the
> > shutdown sequence. The system hangs after "Sending SIGTERM to remaining
> > processes..." and never reaches full power-off (ACPI *S5).*
> 
> A few questions:
> 
> * Which firmware files does your GPU use exactly (a full dmesg log
> should answer this)
> * Does the problem happen with a vanilla kernel as well (it most likely
> will, but would be good to confirm, as heavily patched vendor kernels
> can lead to all sort of issues).
> 
> Ciao, Thorsten
> 
> > *Relevant log excerpt (journalctl -b -1):*
> >
> >     Jun 27 02:04:14 cachyos systemd-shutdown[1]: Sending SIGTERM to
> > remaining processes...
> >
> > (no further entries — system hangs here)
> >
> > *Workaround:*
> > Downgrading to linux-firmware-amdgpu 1:20260519-1 resolves the issue
> > completely.
> >
> > *Steps to reproduce:*
> > 1. Update linux-firmware-amdgpu to 1:20260622-1
> > 2. Run `poweroff`
> > 3. System hangs — power is not cut
> >
> > *Additional notes:*
> > The root cause appears to be that the new firmware blob prevents the
> > amdgpu driver from cleanly releasing the GPU before the ACPI S5 power
> > state transition. No other packages were updated in the same transaction
> > that could explain this behavior.
> >
> > This issue was researched and this report was drafted with the
> > assistance of Claude (Anthropic AI).
> >
> > Best regards,
> >
> > Benjamin Maas 
>