Slackware64-current [07-MAR-2020] AMDGPU lock up with Firefox and Googles Maps

kingbeowulf · 03-07-2020, 04:27 PM

This may not be specific to Slackware, or even Firefox and Googles Maps. I've been experiencing occasional intermittent X lock ups when Firefox pan and zooms Google Maps, in Xfce. I'm going to swicth today to KDE4 to see if the same thing happens, but could take days to pop up. So far, Steam Games and other 2D/3D workloads don't seem to cause this so far. Only error that pops up is in /var/log/syslog

Code:

Mar  7 12:20:54 gandalf kernel: [136030.252665] [drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* Waiting for fences timed out!
Mar  7 12:21:00 gandalf kernel: [136030.252775] [drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* Waiting for fences timed out!
Mar  7 12:21:00 gandalf kernel: [136035.372628] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx_0.0.0 timeout, signaled seq=5156567, emitted seq=5156569
Mar  7 12:21:00 gandalf kernel: [136035.372739] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process Web Content pid 19578 thread firefox:cs0 pid 19979

The GPU is XFX Radeon RX 5700 XT THICC III Ultra, with kernel 5.4.24 amdgpu. This has been happening since around 5.4.19. I canb ssh in an pokek around but X server etc is AWOL and unresponsive - not a black screen, just static image of whatever was last displayed and no keyboard or mouse. Only solution so far is a reboot. Windows users around the 'net have been complaining agout black screen lockups with the AMD Windows driver with NAVI GPUs and speculating it has something to do with the AMD GPU BIOS/firmware.

Anyone have any ideas or similar experiences? So far more of an annoyance that anything else.

zuriel · 03-07-2020, 06:42 PM

I had some GPU hangs on the 5.4 kernels with my Vega 64. Running 5.5 kernels for the last few weeks, it's been rock solid.

kingbeowulf · 03-08-2020, 02:58 PM

Interesting. I just had a black screen X or KDE lockup on the Lenovo T510 laptop (all Intel) this morning, so this may not be just an amdgpu issue. Access via ssh and 'ps' and 'htop' showed nothing amiss that I could see; with a reboot. Hard to say though: absolutely nothing of note in the T510 system or X.org logs.

kingbeowulf · 10-20-2020, 12:09 AM

Seems this amdgpu bug is present for AMD NAVI GPUs with kernels below 5.5 utilizing Mesa and amdgpu. This does not apply to amdgpu-pro. I'm still getting X.org freezes, usually only when zooming google maps in chromium and firefox:
https://community.amd.com/thread/250909
https://gitlab.freedesktop.org/drm/amd/-/issues/892
Thus, on the "Requests for -current (14.2-->15.0)" thread, I recommended bumping Slackware Current to a higher kernel, even if that means moving to a stable kernel instead of LTS.

A possible workaround is the following, but I have no tested it yet. Create /etc/profile.d/amdgpu.sh (or whatever filename.sh):

Code:

#!/bin/sh

#export AMD_DEBUG="nongg,nodma"
export AMD_DEBUG="nongg"

Try "nongg" first and if that doesn't work, "nongg,nodma"