TigerLake Yocto MR2 (5.4): GPU driver hang

[Gennady]

For some unknown reason GPU sometimes hangs, so Desktop environment stops working.

1) The problem can be reproduced by issuing a software reboot in loop.

After 5-20 cycles, the GPU hangs at boot when video server is initialized, and and the following error message appears in dmesg:

i915 0000:00:02.0: GPU HANG: ecode 12:1:0x00000000, stopped heartbeat on rcs0

i915 0000:00:02.0: Resetting rcs0 for stopped heartbeat on rcs0 [drm:gen8_reset_engines [i915]] ERROR rcs0 reset request timed out: {request: 00000001, RESET_CTL: 00000001}

As a result, nothing will be displayed on connected display until next reboot/power_cycle.

Logs are attached.

2) The problem can be easily reproduced both with X and Wayland.

3) There are a lot of similar issues with Intel Graphics all over Internet:

 - both for Tiger Lake:

https://gitlab.freedesktop.org/mesa/mesa/-/issues/4025

https://gitlab.freedesktop.org/drm/intel/-/issues/3109

https://mesa-ci.01.org/kusma/builds/19/results/8302360

http://198.175.88.100/lepton/builds/5/results/8502129

https://uonfu.com/q/intel/compute-runtime/395/755414054

 - and for other Intel chips:

https://gitlab.freedesktop.org/drm/intel/-/issues/161

https://gitlab.freedesktop.org/drm/intel/-/issues/487

https://gitlab.freedesktop.org/drm/intel/-/issues/673

Note: some of these threads for non-Tiger-Lake GPUs contain fixes for this issue. But when we checked the sources of our kernel, we discovered that it already contains all of them. And other advices from that discussions have not helped either.

4) Moreover there are a lot of open bugs, which are related to graphics on Tiger Lake on Intel bugzilla. Moreover some of them lead to GPU hanging:

https://gitlab.freedesktop.org/drm/intel/-/issues?label_name%5B%5D=platform%3A+TGL&scope=all&search=hang&state=all  

Configuration:

 - Intel Yocto BSP for Tiger Lake (MR2):

              x 5.4.106-intel-ese-standard-lts kernel;

              x xf86-video-intel-2.99.917+git0+f66d39544b-r0

              x libdrm2-2.4+git0+991e95fd13-r0

              x libdrm-intel1-2.4+git0+991e95fd13-r0

              x xserver-xorg-1.20.8-r0

              x wayland-1.18.0-r0

 - tried different Tiger Lake boards with the following CPUs:

              x i5-1145G7E

              x i7-1185G7E

[CarlosAM_INTEL]

Hello, @Gennady:

Thank you for contacting Intel Embedded Community.

We sent an email to the address associated with this account with suggestions that may help you.

Best regards,

@CarlosAM_INTEL.