570.124.04 prime offload / runtime D3 power management broken on electron app launch on wayland

Hello, I have problems with prime offload and runtime D3 power management. I have semi-recently switched to Hyprland on my Lenovo Legion 5 15IMH05H which has RTX 2060 and Intel GPU. Running Arch Linux using wayland and hyprland, tested on kernels 6.13.5, 6.13.5-zen and 6.12.8-lts, latest nvidia-dkms driver version 570.124.04.

Here is bug report before it happened:
nvidia-bug-report_working.log.gz (1.0 MB)

Here is bug report after it happened:
nvidia-bug-report_not_working.log.gz (1.1 MB)

Both reports are from the same session.

I have found that it’s electron causing this issue. The gpu works as expected, going in and out of d3cold upon prime-run or nvidia-smi.
If I start a electron application using prime-run, it wakes the gpu and it never returns to d3cold, crashing. prime-run glxgears and vkcube show black window and for vkcube the error NVRM: Xid (PCI:0000:01:00): 109, pid=18558, name=vkcube, Ch 00000008, errorString CTX SWITCH TIMEOUT, Info 0x14001 shows up in dmesg. nvidia-smi still works.

I have tested Discord, Obsidian and PulseAudio Graph Control electron apps.

Discord queries the gpu right on start, so it crashes it even without prime-run. Tried passing --enable-features=UseOzonePlatform --ozone-platform=wayland but behavior is the same, xwayland doesn’t seem to have an effect.

Obsidian crashes the gpu without prime-run, but with the wayland enable options it does touch the gpu and crashes it even without prime-run.

PulseAudio Graph Control with prime-run crashes on start with following error, but doesn’t crash the gpu.

[10743:0312/161240.837537:FATAL:gpu_data_manager_impl_private.cc(445)] GPU process isn't usable. Goodbye.
/usr/bin/prime-run: line 2: 10743 Trace/breakpoint trap   (core dumped) __NV_PRIME_RENDER_OFFLOAD=1 __VK_LAYER_NV_optimus=NVIDIA_only __GLX_VENDOR_LIBRARY_NAME=nvidia "$@"

I have set udev rules and nvidia options according to docs. If I omit the two which are supposed to work on kernel 5.5+, I get i2c timeout error and xHC error with root hub lost power following, but that’s probably not related.