Haswell GPU Hang with kernel 5.10

I've been having GPU hangs and picture corruption issues recently.

When the GPU hang happens, Sway will freeze and I am no longer able to interact or update my desktop. I have to switch to another TTY to pkill sway and then run until the next hang. The issues don't seem to happen consistently with opening any particular program or after a certain period of time.

System architecture: x86_64

Kernel version: 5.10.11-arch1-1

Linux distribution: Arch Linux

Display connector: Integrated laptop screen + HDMI

CPU: Intel(R) Core(TM) i7-4710MQ

Desktop environment: sway, wlroots, XWayland

dmesg.log.zst

journalctl.log.zst

hang-error.zst

Previous issue on Sway repo before being told to report here: https://github.com/swaywm/sway/issues/5988

changed the description

changed the description

Try

echo off | sudo tee /sys/module/i915/parameters/mitigations

#3038 indicates that it is currently causing render corruption on hsw-gt2, so a hang is also possible.

I've run that command. Will report back with whether I still get hangs.

It still hangs for me.

added kernel:5.10 label

Similar setup with the same issue. Happens at random times, but I notice it is mostly while using Google Meet over the day to pair with people. Here the systemd output:

Feb 03 18:05:01 kernel: i915 0000:00:02.0: [drm] Resetting rcs0 for invalid CSB event
Feb 03 18:05:01 kernel: i915 0000:00:02.0: [drm] sway[1484] context reset due to GPU hang
Feb 03 18:05:01 kernel: i915 0000:00:02.0: [drm] GPU HANG: ecode 11:1:85d7fffb, in sway [1484]
Feb 03 18:05:01 kernel: GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.
Feb 03 18:05:01 kernel: Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.
Feb 03 18:05:01 kernel: Please see https://gitlab.freedesktop.org/drm/intel/-/wikis/How-to-file-i915-bugs for details.
Feb 03 18:05:01 kernel: drm/i915 developers can then reassign to the right component if it's not a kernel issue.
Feb 03 18:05:01 kernel: The GPU crash dump is required to analyze GPU hangs, so please always attach it.
Feb 03 18:05:01 kernel: GPU crash dump saved to /sys/class/drm/card0/error
Feb 03 18:05:15 kernel: i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
Feb 03 18:05:15 kernel: i915 0000:00:02.0: [drm] sway[1484] context reset due to GPU hang
Feb 03 18:05:15 kernel: i915 0000:00:02.0: [drm] GPU HANG: ecode 11:1:85dfbfff, in sway [1484]

This is a different architecture, Icelake (gen11). The other crashes in this bug report are specific to Haswell gt2 (gen7). That indicates you have a different problem that requires a fresh bug report and include the error state.

@ickle Turning off mitigations seems to have stopped my DE crashing, thanks for that.

Am I right in thinking that this would just be a temporary solution?
Is there any information I can give to help getting this fixed for when mitigations are on?

mitigations=off also fixed triangles and gpu hangs for me.

There's a good chance that

commit 29d35b73ead4e41aa0d1a954c9bfbdce659ec5d6
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Mon Jan 25 12:50:33 2021 +0000

    drm/i915/gt: Always try to reserve GGTT address 0x0
    
    commit 489140b5ba2e7cc4b853c29e0591895ddb462a82 upstream.

in v5.10.14 accidentally prevents the hangs and glitches?

Running without mitigations=off on 5.10.14 today to test that.

I think I have the cause of the scratch overwrite (and it may end up overwriting anything afaict): https://patchwork.freedesktop.org/series/86947/

That will be in drm-tip soon, but may be a couple of weeks before it propagates to stable.

I've not had any graphical glitches or hangs today while running kernel 5.10.14 with the default mitigations value of auto.

Had lots of visual glitches on my Haswell GPU, too (Fedora 33).

Downgraded my kernel from 5.10.11 to 5.10.8 and the issue is solved.

More people are talking about this issue here: https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/3642

Same problem on thinkpad T450 Intel i5 5300U with iGPU HD-5500 with kernel 5.10.x and long term kernel 5.4.99 Fedora 33 x86_64 Gnome 3.38.3 wayland and video_2021-02-26_18-25-36

I have the same effects on debian/buster desktop machine on a Asus Z87M-PLUS with Intel(R) Core(TM) i5-4440 CPU uname -m: x86_64 uname -r: 5.10.0-3-amd64. Unfortunately, I've added my logs to another thread: Issue 3123

Problem still there on Linux 5.11.2 (Arch Linux on T440P)

After a couple of suspend/resume cycles on kernel 5.11.6, problem seems to be not present anymore.

Arch Linux stock kernel on T440P.

with kernel 5.11.7 the problem remain on i5 5300U intel HD 5500 (thikpad t450) on gnome 40

i915 0000:00:02.0: [drm] ERROR CPU pipe A FIFO underrun

Arch Linux stock 5.11.7-arch1-1 still working on t440p. No regression here.

5.11.11 problem remain

i915 0000:00:02.0: [drm] ERROR CPU pipe A FIFO underrun

added platform: HSW label