Intel i915 GPU hang after kernel update to 5.8 (Pop OS 20.04) #29
Description
I am using Pop!_OS 20.04 LTS on a Haswell Intel(R) Core(TM) i7-4710HQ CPU.
After updating to kernel 5.8 a couple of days ago, the system constantly crashes on any heavier GPU load. As soon as a race starts in SuperTuxKart, the following happens 100% of the time:
[ 288.066977] i915 0000:00:02.0: [drm] GPU HANG: ecode 7:1:85ddfffd, in Xorg [1571]
[ 288.066978] GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.
[ 288.066979] Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.
[ 288.066979] Please see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details.
[ 288.066979] drm/i915 developers can then reassign to the right component if it's not a kernel issue.
[ 288.066980] The GPU crash dump is required to analyze GPU hangs, so please always attach it.
[ 288.066980] GPU crash dump saved to /sys/class/drm/card0/error
[ 288.067604] i915 0000:00:02.0: [drm] Resetting chip for stopped heartbeat on rcs0
[ 288.169353] i915 0000:00:02.0: [drm] Xorg[1571] context reset due to GPU hang
[ 291.138874] i915 0000:00:02.0: [drm] GPU HANG: ecode 7:1:85ddfffd, in Xorg [1571]
[ 291.138955] i915 0000:00:02.0: [drm] Resetting chip for stopped heartbeat on rcs0
[ 291.242212] i915 0000:00:02.0: [drm] Xorg[1571] context reset due to GPU hang
[ 297.030397] i915 0000:00:02.0: [drm] GPU HANG: ecode 7:1:85ddfffd, in Xorg [1571]
[ 297.030448] i915 0000:00:02.0: [drm] Resetting chip for stopped heartbeat on rcs0
[ 297.131629] i915 0000:00:02.0: [drm] Xorg[1571] context reset due to GPU hang
After this, the GPU is reset, all apps are closed and the login screen is started again. This crash can also happen randomly anywhere, but very rarely. However, full load always triggers it.
This problem is exactly the same as shown in this bug report:
https://gitlab.freedesktop.org/drm/intel/-/issues/2024
It seems that this problem was introduced by the newer kernel in Pop. There is a proposed fix (& 0) in the thread. It seems that more and more people will be affected as kernels of 5.7 and above are rolling out in various distributions.