4

System

Linux Mint 19.3 Cinnamon 64-bit, based on Ubuntu 18.04 LTS.


Related Hardware


Could anyone tell me if the following means anything special like that my Nvidia card is faulting? Could it be just a software error on Google Chrome (stable) side, or in nvidia-435 driver? How do I find out?

I just know my computer froze for a second or two, and this:

dmesg trail

[Thu Jan 16 16:01:38 2020] show_signal_msg: 23 callbacks suppressed
[Thu Jan 16 16:01:38 2020] GpuWatchdog[18858]: segfault at 0 ip 000055a9a5a6077d sp 00007f033f76c6c0 error 6 in chrome[55a9a1b25000+7170000]
[Thu Jan 16 16:01:38 2020] Code: 48 c1 c9 03 48 81 f9 af 00 00 00 0f 87 c9 00 00 00 48 8d 15 19 61 9c fb f6 04 11 20 0f 84 b8 00 00 00 be 01 00 00 00 ff 50 30 <c7> 04 25 00 00 00 00 37 13 00 00 c6 05 f1 6b a4 03 01 80 7d 8f 00

What I was doing at that time

I was playing an HTML5 game (Forge of Empires).

1 Answer 1

7

No, it's not your Nvidia card that is at fault. Neither really is Chrome, either.

What happens first is that the Nvidia software crashes, stopping the render pipeline. Then, after a few seconds, chrome detects the GPU not rendering any more, tries to handle that, fails, and throws the segfault.

When the machine is in that crashed state, and you ssh into it and run "top", you'll see two processes irq/75 nvidia and nv_queue alternately running at 100% cpu (the interrupt number may be different on your system).

Also, a few seconds before the GpuWatchdog, your syslog probably contains some messages from the nvidia driver:

Feb 10 17:00:24 natascha kernel: [157260.734117] NVRM: GPU at PCI:0000:08:00: GPU-f622f482-2ad1-4992-4d8a-9d62b465e084
Feb 10 17:00:24 natascha kernel: [157260.734120] NVRM: GPU Board Serial Number: 
Feb 10 17:00:24 natascha kernel: [157260.734124] NVRM: Xid (PCI:0000:08:00): 61, pid=1391, 0cde(308c) 00000000 00000000

Reports of the problem are all over the internet; I didn't find any fixes yet. I had the same problem on my new PC, not running chrome didn't prevent the crash but prevented the syslog message; reverting to 430 drivers from 435 made the problem go away (so far).


Update: The crash happens with 430 drivers as well. The 440 drivers, not part of Ubuntu, seem to fix this though. At least I didn't have the problem any more, and the post by amrits on https://devtalk.nvidia.com/default/topic/1060783/linux/random-xid-61-and-xorg-lock-up/7 confirms this.

As the 440 drivers are not part of the Ubuntu distribution, here's what I did - I got this info from https://linuxconfig.org/how-to-install-the-nvidia-drivers-on-ubuntu-19-10-eoan-ermine-linux which is about Ubuntu 19.10, but works on 18.04 as well:

sudo -i
add-apt-repository ppa:graphics-drivers/ppa
apt update

At this point, ubuntu-drivers devices should output, among other things,

# ubuntu-drivers devices
== /sys/devices/pci0000:00/0000:00:03.1/0000:08:00.0 ==
modalias : pci:v000010DEd00001F02sv000010DEsd00001F02bc03sc00i00
vendor   : NVIDIA Corporation
driver   : nvidia-driver-440 - third-party free recommended

then you can install the driver

apt install nvidia-driver-440

and as you need to reboot anyway to make the new driver active, I recommend updating the rest of your software as well:

apt upgrade
apt autoremove
reboot

Update Jun 15 - There is still no driver fix according to nvidia, they weren't able to reproduce the problem. See the thread at their forum. However, it seems like the issue happens on some mainboard/GPU combinations when the GPU goes from power save mode to a mode where it uses more power. Forcing the GPU to a higher frequency seems to prevent this, and some users report the following to work as a workaround:

nvidia-smi -pm ENABLED
sudo nvidia-smi -lgc 1000,1815

(This must be repeated at each reboot)

This sets a permanent (until reboot) higher frequency for the card, resulting in more power consumption and possibly less lifetime, but seems to work around the crash, so may be preferable to many users.

8
  • Very nice and full of detail. Good job. Commented Feb 10, 2020 at 16:53
  • updated to 440 but issue still occurs randomly. Could you please confirm you kernel version as well? Thank you
    – hbt
    Commented Apr 4, 2020 at 20:01
  • I still have a few crashes as well, but instead of one per day it's one per week with 440.64. Commented Apr 5, 2020 at 10:08
  • I started having this issue once a week using 440.64 as well
    – bksunday
    Commented May 9, 2020 at 12:49
  • 1
    This is most certainly caused by the garbage nvidia drivers on Linux and it is still happening with the 470 drivers. Wasn't a problem with the open AMD drivers. At least they fixed the monitor goes to sleep and you loose all but one display problem. If anyone is wondering when the year of the Linux desktop will be, the answer is never and this is why. Commented Jul 10, 2021 at 3:59

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .