Comment 43 for bug 1492037

I looked in the user journal, and found first this:

aug. 08 01:40:07 neruval kernel: [drm:gen8_irq_handler [i915]] *ERROR* The master control interrupt lied (SDE)!

Then it seems like kscreen (the display handler for Plasma) tried to launch another instance of itself (which fails, because it is already running), and then receives some weird notifications about the displays changing:

aug. 08 01:40:07 neruval kscreen_backend_launcher[28810]: kscreen.xrandr: Connected output 67 to CRTC 63
aug. 08 01:40:07 neruval kscreen_backend_launcher[28810]: kscreen.xcb.helper: Detected XRandR 1.4
aug. 08 01:40:07 neruval kscreen_backend_launcher[28810]: kscreen.xcb.helper: Event Base: 89
aug. 08 01:40:07 neruval kscreen_backend_launcher[28810]: kscreen.xcb.helper: Event Error: 147
aug. 08 01:40:07 neruval kscreen_backend_launcher[28810]: kscreen.backendLauncher: Loading "XRandR" backend
aug. 08 01:40:07 neruval kscreen_backend_launcher[28810]: kscreen.backendLauncher: Failed to register as DBus service: another launcher already running?
aug. 08 01:40:07 neruval kscreen_backend_launcher[28810]: kscreen.backendLauncher: ""
aug. 08 01:40:07 neruval kscreen_backend_launcher[10198]: kscreen: Primary output changed from KScreen::Output(Id: 67 , Name: "eDP1" ) ( "eDP1" ) to KScreen::Output(Id: 67 , Name: "eDP1" ) ( "eDP1" )
aug. 08 01:40:07 neruval kscreen_backend_launcher[10198]: kscreen: Primary output changed from KScreen::Output(Id: 67 , Name: "eDP1" ) ( "eDP1" ) to KScreen::Output(Id: 67 , Name: "eDP1" ) ( "eDP1" )
aug. 08 01:40:07 neruval kscreen_backend_launcher[10198]: kscreen: Primary output changed from KScreen::Output(Id: 67 , Name: "eDP1" ) ( "eDP1" ) to KScreen::Output(Id: 67 , Name: "eDP1" ) ( "eDP1" )
aug. 08 01:40:07 neruval kscreen_backend_launcher[10198]: kscreen: Primary output changed from KScreen::Output(Id: 67 , Name: "eDP1" ) ( "eDP1" ) to KScreen::Output(Id: 67 , Name: "eDP1" ) ( "eDP1" )

And a couple of seconds after that, Plasma crashes:

aug. 08 01:40:09 neruval drkonqi[28805]: Sending SIGSTOP to process

My theory from these logs was that because of these weird display changes Plasma resizes itself to some really weird sizes, which leads to mesa trying to allocate some really weird BOs, which the kernel fails to allocate, and then things crash.

I added some debug output to libdrm, to see what errors the kernel was actually returning:

================= DRMIOCTL FAILED 22: Invalid argument
================= DRMIOCTL FAILED 1: Operation not permitted
================= DRMIOCTL FAILED 22: Invalid argument

The invalid argument error I think might support my theory. I forgot to log the actual request, but I'll do that now, hopefully it will help to track this further. If it is the error from the IRQ handler that causes this, I'm even more out of my depth, though.