Created attachment 307405 [details]
dmesg-from-journalctl

Hello,

situation here:

- Kernel: 6.12.6 (issue also reproducible with 6.12.3 and also random 6.11.x versions)
- Mainboard: MSI B550M-PRO-VDH-WIFI, replaced a few months ago
- CPU: AMD Ryzen 5700G
- Video: Radeon RX 6700G

The combo has been working fine with kernels 6.9.x, 6.10.x, including regular suspend/resume cycles (S3). I noticed first trouble when testing 6.11-rcX versions. It failed to resume but sometimes only on the second or third suspend cycle. The obvious symptoms were that system powered up but video never came back, neither did the onboard NIC, although USB keyboard was still responsive and I could sync and restart via Magic-SysRQ combos. No journalctl logs were shown, so I guess that userspace never has been resumed.

I have tried the CONFIG_PM_TRACE / CONFIG_PM_DEBUG method already. No usable data reported from there, i.e. from dmesg:

Dez 25 16:11:46 zombie kernel: PM: RTC time: 15:11:30, date: 2024-12-25
...
Dez 25 16:11:46 zombie kernel: PM:   Magic number: 4:125:185

$ cat /sys/power/pm_trace_dev_match
usb
acpi-wmi
acpi

I wanted to be helpful and have tried the bisecting already. That is leading to nowhere, I have given up after some hours. Because some builds are not showing the same behavior, i.e. the repro runs eventually end up in strange behavior with non-responsive USB keyboard (either not lighted at all or blinking leds on key strokes but not doing anything). And skipping those bisect steps means that it keeps jumping between 6.9.x and 6.11.x and 6.10.x branches, and I have no further ideas on how to continue.

On the other hand, 6.10.14 version is running perfectly right now, but IMHO that issue smells extremely and I am using regular&popular hardware parts, so it might affect lots of users with similar setups.