[amdgpu] Severe freezes that begin with minor stutters

[Douglas Silva]

It starts with very subtle stuttering that you notice when moving windows around, then keeps getting worse and worse until it starts freezing the UI for many seconds. I could only reproduce it twice in a couple of days.

The applications open at the time are usually Atom, Firefox, Hexchat, GNOME Files and occasionally Eye of GNOME. I checked resource consumption at the time of the stutters. Memory consumption was about 50% of 8GB, while the swap partition was nearly full (after a long gaming session). Storage device is a SATA SSD. During the stutters and freezes, CPU usage is normal. Through the process list of System Monitor I cannot see any application overusing resources. Temperatures are also normal.

One potentially problematic hardware device is the GPU: AMD RX 560 4GB (open source AMDGPU). Its temperature and fan speed were also normal at the time.

I know my system is capable of handling this workload. Also, this is only reproducible on Ubuntu 19.10 Eoan with GNOME. The LTS version of Ubuntu (18.04.3) is unaffected. I have used it for months before switching to the latest non-LTS version.

What I've tried to stop the freezes:
- Closed all running applications and waited.

It didn't improve the situation, so I restarted my user session. The freezes continued, so I performed a system reboot, which finally helped normalize things.

I did not try to reproduce this on the Wayland session.

ProblemType: Bug
DistroRelease: Ubuntu 19.10
Package: gnome-shell 3.34.1+git20191024-1ubuntu1~19.10.1
ProcVersionSignature: Ubuntu 5.3.0-26.28-generic 5.3.13
Uname: Linux 5.3.0-26-generic x86_64
ApportVersion: 2.20.11-0ubuntu8.2
Architecture: amd64
CurrentDesktop: ubuntu:GNOME
Date: Fri Jan 24 11:13:56 2020
DisplayManager: gdm3
InstallationDate: Installed on 2019-12-27 (28 days ago)
InstallationMedia: Ubuntu 19.10 "Eoan Ermine" - Release amd64 (20191017)
RelatedPackageVersions: mutter-common 3.34.3-1ubuntu1~19.10.1
SourcePackage: gnome-shell
UpgradeStatus: No upgrade log present (probably fresh install)

[Daniel van Vugt (vanvugt)]

Please:

 1. Run:

    lspci -k > lspcik.txt

    and attach the file 'lspcik.txt' here.

 2. Reproduce the bug again and collect a full system log by running:

    journalctl -b0 > journal.txt

    and attach the file 'journal.txt' here.

 3. Try booting some newer or older kernels to see if any are able to avoid the bug:

    https://kernel.ubuntu.com/~kernel-ppa/mainline/

[Douglas Silva (o-alquimista)]

I have been trying to reproduce it in the last days, but I couldn't. I managed to get nearly 100% use of swap space again, while simultaneously using Nautilus and Eye of GNOME (which is what I was doing at the exact moment the freezes began), but it didn't happen again. Nothing changed about my workload. I have the same applications open, consuming the same usual amount of resources.

This seems to be a tricky one to reproduce, but I'll keep trying. By the way, I'm attaching the lspci command output.

[Douglas Silva (o-alquimista)]

I was able to reproduce it again. The system quickly became unresponsive, so I had to reboot with the REISUB combination. After the system started again I collected the logs. I wasn't using Nautilus of the image viewer this time.

[Douglas Silva (o-alquimista)]

The logs I attached previously only refer to the session AFTER the system restarted. So I've collected the logs from the previous boot this time, which will contain the relevant information.

I'm attaching the output of: `journalctl -b-1`

[Daniel van Vugt (vanvugt)]

Thanks. Unfortunately I can't yet see anything going wrong at the end there.

It's interesting you say that 18.04 is unaffected. I wonder are you able to try some other kernel versions to see if that's the important factor here?

https://kernel.ubuntu.com/~kernel-ppa/mainline/

[Douglas Silva (o-alquimista)]

You mean I should try an older version? Then I think I should try v5.0.x first. As you can see, it takes time to reproduce it again. I don't know how to intentionally reproduce it.

[Launchpad Janitor (janitor)]

[Expired for linux (Ubuntu) because there has been no activity for 60 days.]

[Launchpad Janitor (janitor)]

[Expired for xserver-xorg-video-amdgpu (Ubuntu) because there has been no activity for 60 days.]