Lags/freezes/hangs, Intel Raptor Lake/XE graphics?

[jmbreuer]

Certain workloads cause hangs of parts of my desktop, as best as I can describe:

First, I encountered this around xfreerdp: https://github.com/FreeRDP/FreeRDP/issues/12535

Where, when running more than one instance of xfreerdp, the new instance would hang with an all black window immediately on start, and the previous instance becomes unresponsive at that moment.

This is only the tip of the iceberg though. Any full-screen application running under wine/proton (ie typically, a game) will not start, but rather hang forever consuming little CPU. The konsole terminal I've started wine/proton/this game from will also become slow to respond (ie changing tabs), and even other konsole windows will appear hung. Shell in Yakuake or ssh in is unaffected. Just starting steam (which wants to do a pending update) will currently trigger the issue for me, after downloading some data the download window as well as the konsole I've started steam from will stall/only process one input event every 5-10 seconds, something like that.

The (firefox-bin) browser I'm typing this in is unaffected, while I have a hanging konsole/proton instance trying to run Only Cans (what I happened to have around that should be available for free from itch.io).

This behavior started with my most recent world update, including upgrading from gentoo kernel 6.18.8 to 6.18.18.
Starting the previous 6.18.8 kernel now does not change the behavior in any way, so it seems to be more of a userland issue.

When I set up this system at the beginning of the year, it did not seem particularly happy with the default i915 driver, so I added the kernel options to choose xe by default. At that time, the system then worked well, at least for a while.

I already ran into a similar issue after an earlier update (the FreeRDP multi instance thing); but that resolved itself while I was trying to follow up possible reverse dependencies / libraries built by a different version of the compiler / ... - no such luck this time, at least not so far.

Now, I get quite exactly the same behavior no matter whether I use the xe or i915 kernel driver, or KWin X11 (which is what I normally use) vs KWin Wayland (which I tried once to check).

Killing (possibly -9, especially wineserver) 'enough' of the zoo of processes created by wine/proton will (usually) un-stick the hung konsole windows etc again.

It feels to me like something is serialized/locking goes wrong when accessing certain 'more advanced' (3D-ish, fullscreen-like, accelerated, ...) graphics functions.
Once I've "managed" to lock up the (at the time fairly loaded) system by ((too) quickly) toggling the KWin compositor (Shift+Alt+F12) as the compositor was turning on again. Unfortunately, the system was so stuck then that I could not get any meaningful debug log out of it.

I've had a look at https://drm.pages.freedesktop.org/intel ... -bugs.html but I don't see any smoking guns at all in the places mentioned there (ie debug logging does not change noticeably when I start an application that then causes the hanging).

This is with the integrated graphics of an i7-1360P CPU (on a laptop, so no easy change); it gets detected as Raptor Lake-P [Iris Xe Graphics] [8086:A7A0].

Any ideas how to further debug this/places I could look/things I could try?

[jmbreuer]

I played around with different terminal emulators/shells etc, and it seems that exactly those processes/applications become sluggish/freeze periodically/only process input sporadically, which are parents of the process that initially causes/exhibits the hangs. ...

For example, starting uxterm from yakuake and then one of the triggering applications from uxterm will lag both uxterm and yakuake as soon as xfreerdp/wine/proton/steam triggers the issue.