GPU HANG: ecode 8:0:00000000

Description

frequent GPU HANG errors when waking up from idle ( maybe ) using h264 vaapi de & encode
requires a reboot to recover
might be a regression since all 5.8.x & 5.11.x kernels worked fine so far

[Thu Dec 16 08:01:54 2021] i915 0000:00:02.0: [drm] GPU HANG: ecode 8:0:00000000
[Thu Dec 16 08:01:54 2021] i915 0000:00:02.0: [drm] Resetting rcs0 for stopped heartbeat on rcs0
[Thu Dec 16 08:01:57 2021] i915 0000:00:02.0: [drm] GPU HANG: ecode 8:0:00000000
[Thu Dec 16 08:01:57 2021] i915 0000:00:02.0: [drm] Resetting rcs0 for stopped heartbeat on rcs0
[Thu Dec 16 08:02:00 2021] i915 0000:00:02.0: [drm] GPU HANG: ecode 8:0:00000000
[Thu Dec 16 08:02:00 2021] i915 0000:00:02.0: [drm] Resetting rcs0 for stopped heartbeat on rcs0
[Thu Dec 16 08:02:00 2021] Fence expiration time out i915-0000:00:02.0:queue56:src[1237]:aee504!
[Thu Dec 16 08:02:00 2021] Fence expiration time out i915-0000:00:02.0:queue56:src[1237]:aee502!

unfortunately no GPU crash report

Hardware/OS

Ubuntu 21.04
5.13.0-20-generic
Intel(R) Xeon(R) CPU E3-1285L v4 @ 3.40GHz
00:02.0 Display controller: Intel Corporation Iris Pro Graphics P6300 (rev 0a)

added platform: BDW label

Have you tried with latest drm-tip. See also what information is needed: https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs

i can't since this is a production system

actually adding the i915.mitigations=off kernel parameter seems to have fixed it

at least i haven't had a GPU hang since then

#4858 just leaving this here.

Happened again unfortunately while waking up from idle ( starting vaapi de/encode )

/sys/class/drm/card1/error still without any error state

[Thu Feb 17 08:30:50 2022] i915 0000:00:02.0: [drm] Resetting rcs0 for stopped heartbeat on rcs0
[Thu Feb 17 08:30:53 2022] i915 0000:00:02.0: [drm] GPU HANG: ecode 8:0:00000000

Kernel version: stock Ubuntu 5.13.0-28-generic

there is a working solution for #4858, mentioned in the first post. could be a different situation, but worth a try.

I have found a way to reproduce: #8687 (comment 2007347).

When I migrated from XFCE/Xorg on Debian Buster to Gnome/Wayland on Debian Bookworm, I started experiencing this issue at random about 4 times per hour when watching YouTube videos in Firefox. I am using an Intel i7-5775C CPU, which has the Iris Pro 6200 integrated GPU. The solution reported in #4858 reduced the hang frequency to about once per day for me.

Instead of running intel_gpu_frequency -m on every boot, I made the mitigation permanent by creating the following udev rule in /etc/udev/rules.d/10-card0.rules after finding the appropriate frequency with intel_gpu_frequency:

ACTION=="add|change", KERNEL=="card0", SUBSYSTEM=="drm", ATTR{gt_min_freq_mhz}="1150"