VRR randomly causes a full system freeze

Can you bisect?

Yes, I'll probably get to it on Sunday.

By the way, you may be able to capture crash or freeze info with netconsole. Or, I guess, a serial console.

@kode54 Thanks for the suggestion! Unfortunately, I don't have access to another linux machine and can't setup a netconsole. I've also tried using kdumpst, but although this tool manages to reboot my system when the freeze occurs, it fails to log anything.

I've started bisecting but had a few starting issues and couldn't finish it today. So far, I've built the kernels v6.10.0-rc5-1-git and v6.6.10-arch1-1. The git kernel caused a bit of trouble since it prevented the system from properly shutting down. However, this kernel did not freeze when I tried the reproduction steps. Instead, there was a subtle black screen flicker on the desktop, but that's a different issue.

As expected, the freeze did not occur with the v6.6.10-arch1-1 kernel, so I'll continue bisecting from there tomorrow.

Unfortunately, I ran into a problem while trying to bisect the issue: I'm able to compile the kernel after setting up the bisect, but the new kernel then fails to boot and gets stuck at "a start job is running" for UUID="d79a0c42-b1e3-41e2-8f45-88b81213a036" which is strange because that is the correct boot drive (/dev/nvme0n1p2).

I've followed the Arch wiki Bisecting bugs with Git and Kernel/Arch build system. The exact steps I've taken are as follows:

yay -S linux-git
cd /home/user/.cache/yay/linux-git/
makepkg -od
cd src/linux-torvalds
make clean
git bisect start, git bisect good v6.6, git bisect bad v6.7
cd ../..
makepkg -efsi

This seems to reset the kernel config (Restart config...) which is then followed by a lot of prompts for which I entered default values. The kernel builds successfully and gets installed, but fails to boot. The label of the installed kernel is always 6.10.0-rc6-1-git-00069-g795c58e4c7fc instead of 6.6.r9152.deefd5024f07. I don't know if that's how it's supposed to be or if I have to change a config somewhere.

I started over 4 times, thinking I might have missed something and also tried skipping to the next commit, but the result was the same.

Any help or hint would be appreciated.

did you skip with git bisect skip?, I know I had to run it a couple times when I was bisecting an issue a while back

Yes, but I only skipped once and then stopped because it didn't change the result. I'll try skipping a few more commits tomorrow.

I've tried skipping and rebuilding the kernel 5 times now and the result was always the same: the kernel gets stuck at "a start job is running" when trying to boot

maybe try doing it without make clean? I don't remember running that when bisecting the kernel because I remember reading it would wipe the config, otherwise my steps were about the same.

you can run make localmodconfig in the directory to make a config with only the modules you are currently using on your system, it makes building it faster. there maybe other steps I have forgotten, I don't have access to my machine right to check exactly what I did.

Thanks for the help! I'll try that.

@Jelgnum I managed to get it to work by copying the prepare logic in PKGBUILD to the build step.

@agd5f Here are my bisect results: bisect.log first-bad-commit.log

There were a few commits tagged as v6.5 that I couldn't properly test because the graphics driver didn't work with those kernels, causing it to fall back to llvmpipe. Since those commits came before v6.6 I tagged them as good.

The bad commit seems pretty big, so I'm not sure how useful this information is. Is there a way to narrow it down further?

Btw: I've also now tested the kernel 6.10.arch1-2 and can still reproduce the crash.

mentioned in issue #3464

I'm experiencing the exact same issue, my system is also similar to yours.

Hardware description:

Mainboard: MSI MAG B650 TOMAHAWK WIFI
CPU: Ryzen 7 7700x
GPU: AMD Navi 31 Radeon RX 7900 XTX
System Memory: DDR5 32 GiB (2x16GB, 6000mhz)
Display: Samsung LC27HG70QQUXEN 27" Curved Gaming Monitor - 144Hz, WQHD, Freesync
Type of Display Connection: DP 1.2 (HBR2 21.6 Gbit/s)

System information:

Distro name and Version: Arch Linux
Kernel version: 6.9.7-arch1-1
mesa 1:24.1.2-1
vulkan-radeon 1:24.1.2-1
DE: Plasma 6.1.1 (Wayland)
WM: kwin 6.1.1.2-1 (Wayland)

I kept triggering the freeze by setting VRR to "Automatic" on KDE Plasma and watching fullscreen content, mostly Youtube on Chromium and MPV.

I am also unable to reproduce the freeze with LTS Kernel 6.6.36-1.

X11 with VRR disabled doesn't trigger the freeze, i might try enabling VRR to see if something changes.

Yesterday i was able to set up kdumpst and got a kernel panic stacktrace, hopefully it helps dmesg-pstore.202407020018-0

If you guys have any suggestions for testing or ways to get more logs, i will try to help.

Thanks, it's good to know that this issue also occurs on other systems Are you able to trigger the freeze by taking screenshots or does it freeze all by itself as soon as you go fullscreen?

I still need to finish bisecting - haven't had a chance yet to continue.

Yes, i was also able to trigger it at low FPS (10-30) using VRRTest and Spectacle, i kept the "PrintScreen" key pressed on my keyboard to keep taking screenshots but without saving them to disk.

Before seeing your issue posted here, i always triggered the crash watching fullscreen video content and occasionally moving the mouse and switching back and forth from fullscreen to windowed. Games didn't seem to have triggered the crash for me yet, maybe because we always need low FPS in order to trigger it.

In my case, i observed that the crash never triggers if the system hasn't been on for at least half an hour.

Just now, i was able to obtain a new stacktrace triggered with VRRTest and Spectacle (the old one i posted looks to contain corrupted data): dmesg-pstore.202407041959-0

I experienced full system freezes with VRR occurring completely at random once or twice a week on my Lenovo Legion 7 Gen 7 AMD laptop with an RX 6700M dGPU (and 680M iGPU). After I disabled VRR, the freezes stopped happening. Not sure how to diagnose it since there's nothing in the journal after a hard reboot, and I can't SSH into a frozen system.

mentioned in issue #3421 (closed)

changed the description

The same issue happened to me several times during games, but I can't trigger it with VRRTest and Spectacle.

Have encountered this when pausing or seeking in mpv while fullscreen with VRR set to Automatic in Plasma 6.

Mainboard: MSI MAG B650 TOMAHAWK WIFI
CPU: AMD Ryzen 7900X
GPU: AMD Navi 31 Radeon RX 7900 XTX [1002:744c] (rev c8)
System Memory: 32 GiB (2x16GB)
Display: ASUS XG27AQ
Type of Display Connection: DP 1.2
Distro name and Version: Fedora Linux
Kernel version: 6.9.12-200.fc40.x86_64
mesa 24.1.5-2
DE: Plasma 6.1.3 (Wayland)
WM: kwin 6.1.3-2 (Wayland)

Stacktrace is very similar to the previous one: vrr-bt.txt

changed the description

mentioned in issue #3581 (closed)

This, #3581 (closed) and #3142 (closed) are the exact same issue. It'd be helpful to track them all in one place

As the other reports all cover Plasma. Just for completeness sake: I can reproduce the issue on Gnome 46 (experimental VRR) and kernel 6.9.12/6.10.6

And I could reproduce #3142 (closed) with labwc/wlroots and kernel 6.10.x, but apply the linked patch there on all kernels going forward.

Instead, I end up with: #3487

mentioned in issue #3142 (closed)

mentioned in issue mesa/mesa#11854

This fits the symptoms I've experienced though I hadn't associated it with VRR (I started to use KDE Plasma's automatic VRR but I have window rules that disable it for everything other than games).

I've been playing Hunt Showdown and getting completely random hangs/freezes with nothing in the systemd log files. I tried to SSH into the system when it happened a few minutes ago and could not connect to the system, so it really seems to be more than just a "user space/graphics" crash, the entire system becomes unresponsive even to network traffic.

Operating System: Fedora Linux 40
KDE Plasma Version: 6.1.5
KDE Frameworks Version: 6.6.0
Qt Version: 6.7.2
Kernel Version: 6.10.10-200.fc40.x86_64 (64-bit)
Graphics Platform: Wayland
Processors: 32 × AMD Ryzen 9 7950X 16-Core Processor
Memory: 61.9 GiB of RAM
Graphics Processor: AMD Radeon RX 7900 XTX
Manufacturer: ASUS

A fix has been merged in 6.11.0 and 6.10.11, please give it a try. See the root cause: #3142 (closed)

That's great news! I'll wait a few days as I assume 6.10.11 will be hitting Fedora soon.

Closing since this has been fixed in kernel 6.10.11. See #3460 (comment 2578529).

closed

VRR randomly causes a full system freeze

Brief summary of the problem:

Hardware description:

System information:

How to reproduce the issue:

Attached files:

Screenshots

Log files (for system lockups / game freezes / crashes)

Designs

Child items 0

Activity

Admin message

Admin message

VRR randomly causes a full system freeze

Brief summary of the problem:

Hardware description:

System information:

How to reproduce the issue:

Attached files:

Screenshots

Log files (for system lockups / game freezes / crashes)

Activity