System completely locking down "randomly?" when using an eGPU

Brief summary of the problem:

When using my eGPU (laptop screen turned off), KDE Plasma (kwin) runs alright but I had to set KWIN_DRM_DEVICES in /etc/environment to make it use the eGPU by default so I can get the full 144fps in kwin. But, at random times (have not found any pattern), the system completely locks down :

 ° the external screen is on, but completely frozen

 ° NO input is taken (not even the `REISUB` method works)

 ° even `SSH` doesn't connect

I have to force shut down my laptop by long pressing the power button.

Hardware description:

• CPU: i5-1335U
• GPU 1: 0000:00:02.0 VGA compatible controller [0300]: Intel Corporation Raptor Lake-P [Iris Xe Graphics] [8086:a7a1] (rev 04)
• GPU 2 (eGPU) : 0000:03:00.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc. [AMD/ATI] Ellesmere [Radeon RX 470/480/570/570X/580/580X/590] [1002:67df] (rev e7)
• System Memory: 16GB DDR5
• Display(s): 1920x1200@60hz internal display ; 1920x1200@144hz external connected to the eGPU
• Type of Display Connection: the internal display is connected via eDP ; the external one is connected via HDMI to the eGPU

System information:

• Distro name and Version: Fedora 40 KDE
• Kernel version: Linux inspiron7430fedora 6.8.8-300.fc40.x86_64 #1 SMP PREEMPT_DYNAMIC Sat Apr 27 17:53:31 UTC 2024 x86_64 GNU/Linux
• Custom kernel: N/A
• AMD official driver version: N/A
• DE : KDE Plasma 6.0.4 on WAYLAND

How to reproduce the issue:

1. Use KDE Plasma 6 Wayland
2. Have an eGPU connected (for me, it's an AMD RX580 connected via ThunderBolt3)
3. Wait, and eventually, after an undetermined and possibly variable time, the whole system will completely freeze.
4. It seems to happen even faster if I'm using XWayland apps such as OnlyOffice, or if I am playing videos. Not tested on GNOME, working in Windows 10 (it's not hardware malfunction)

Attached files:

Some (maybe useful) logs captured via sudo journalctl -b-1 | tail -n50 after a crash. journalctl-logs.txt

Screenshots/video files

N/A

Log files (for system lockups / game freezes / crashes)

See attachment, if you have any other method to acquire more useful logs, please do tell me and I'll share them.

added Polaris label

Can you attach your dmesg output?

Here's sudo journalctl -k -b-1 | tail -n100 (I had a crash on my last boot as well)

journaltcl-k-log.txt

Possibly the same issue but I managed to get a core dump.

Hardware description:

• CPU: Intel i5-13600K
• GPU: Advanced Micro Devices, Inc. [AMD/ATI] Navi 22 [Radeon RX 6700/6700 XT/6750 XT / 6800M/6850M XT] [1002:73df] (rev c0)
• System Memory: 32GB of DDR5 RAM @ 6400MHz
• Display(s): DELL 4K
• Type of Display Connection: DP

System information:

• Distro name and Version: EndeavourOS (Arch) latest on 05/10/2024.
• Kernel version: Linux fractal 6.8.9-arch1-2 #1 (closed) SMP PREEMPT_DYNAMIC Tue, 07 May 2024 21:35:54 +0000 x86_64 GNU/Linux
• Custom kernel: NA
• AMD official driver version: Mesa 24.0.6-arch1.2
• KDE Plasma: 6.0.4-1
• Wayland: 1.22.0-1

How to reproduce the issue:

I was mainly running Guild Wars 2 launcher using Steam Proton compatibility.

Attached files:

core.Xwayland.1000.243e4cbb4d8e40199b47830295891d7a.1154.1715315871000000.zst

dmesg.log

journalctl.log

Some more Information: I noticed that it seems to happen faster if I don't kwin --replace after hotplugging my eGPU (before doing so, kwin is using my iGPU and is passing the frames to the eGPU, restarting it makes it use eGPU exclusively). Hot-unplugging the eGPU leads immediately to the lockdown.

Although it seems there has been not much going on since the issue was opened, I just want to add that when running KDE Plasma at 120+ Hz on my monitor, the Memory Clock of my RX580 is stuck at 2000 MHz. This seems to be a known issue #1709 (closed) #1912 etc..

I'll try newer kernels (Fedora 40 is on 6.8.11-300) on Arch / OpenSUSE TW to see if the regression's gone..

I'm experiencing the same lockdown, after eGPU hot-unplug:

relevant logs.txt last lines before forced shutdown

Hardware description:

• CPU: AMD Ryzen 7 7840U w/ Radeon 780M Graphics
• GPU: iGPU: AMD Radeon 780M, eGPU: AMD RX 6700 XT 12G

66:00.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc. [AMD/ATI] Navi 22 [Radeon RX 6700/6700 XT/6750 XT / 6800M/6850M XT] [1002:73df] (rev c1)
c1:00.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc. [AMD/ATI] Phoenix1 [1002:15bf] (rev c4)

• System Memory: 32Gb
• Display(s): MSI 2K
• Type of Display Connection: DP

System information:

• Distro name and Version: NixOS unstable: 25.05.20241207.22c3f2c (Warbler)
• Kernel version: Linux 6.12.3 #1 (closed)-NixOS SMP PREEMPT_DYNAMIC x86_64 GNU/Linux
• Custom kernel: N/A
• AMD official driver version: Mesa 24.2.6

If other information can be helpful, I would gladly provide it

Okay I have done some testing and I think I got a reproducable way to trigger it:

• open a browser on the external monitor rendered by the egpu
• open youtube tabs, a lot of them, in different windows etc (to simulate a certain load)
• and then disconnect the egpu. If the load is low enough, the system seems to recover after a 1~5 seconds or so, otherwise it locks up indefinitely

multiple plug/unplugs, only the last one does not recover properly: dmesg.txt

mentioned in issue #3877