[Resolved] Early kernel 7.0.0 testing shows massive regression

Ask Fedora
, ,
1

Hi everyone, I’m looking for some else’s experience with kernel 7.0.0 on top of Fedora.

I’ve tested kernel-7.0.0-0.rc7.55.fc45 and kernel-7.0.0-62.fc45 on top of Fedora 44 KDE.

On rc.7 I had a complete system lock up, with logs showing drm issues, kernel: watchdog: CPU8: Watchdog detected hard LOCKUP on cpu 8 errors and so on.

On final release I’m not having lockups, but I have system stalls/freezes, mostly when working with 3D (games) and video playback, but also feels like networking also getting stuck, while browsing with Firefox, sites simple stop network activity mid load.

I’ve already opened a amdgpu drm issue here Making sure you're not a bot! , but feels like it’s not (only) drm issue.

My system is 9950X3D CPU with RX 7900 XTX GPU, there was a last minute Zen 3 fix, so I’m curious if something else was missed?

Anyone else have experience with new kernel? I’m wondering if that’s F45 kernel, my system specific or there is a missed issue.

2

Hello,

I compiled kernel 7.0.0 (gcc and rust) today - vanilla source ( Index of /pub/linux/kernel/ ). Everythink work fine - no regression. But it is on my HW - AMD ryzen 7 3700, mb AM4, NVIDIA gc etc. Yes on another HW it is not the same (?)…

3

I’ve also built kernel with v7.0 tag from Linus tree, but have not tried it yet. Gonna see if that helps.

4

Also note that you are testing a rawhide kernel and this forum has very little support for rawhide. Additionally, f44 has not even been released yet, so expecting stability with mixing f45 kernel on a Beta version of f44 seems a far stretch when expecting stability.
F45 packages are subject to frequent glitches/bugs/issues since it is purely development and changes occur extremely rapidly. Sometimes things are not even fully functional.

I am glad that you are testing the newer kernels, but I prefer a stable system and not a development system.

5

I’m perfectly aware of my setup and I’ve been doing exactly that testing (.0 build of kernel from rawhide on top of current (or beta) Fedora version) it for many years for exactly this reason, checking what’s coming early, and this is probably the first time I see huge issues right away.

Since there is a huge latency in kernel/drm fixes, I just wanted to raise awareness and gather feedback, not solution or support :slight_smile:

As far as “vanilla” kernel (with F44 configuration and missing new parameters set to default) goes, it’s also affected, but not to the same degree, the video playback is not affected in Firefox and 3D hiccups are much smaller (so probably some kernel tuning), so basically a micro-stutter rather then several seconds of hang.

Maybe @jforbes can chime in here?

I’m planning to retry “vanilla” kernel with rawhide config and then will try to do bisect, but it usually takes a lot of time, so not before weekend.

6

What is the best way for users who test rawhide to report back on what they find? Mailing list and/or Matrix and/or Bugzilla?

7

(no more IRC, Matrix now: #devel or #quality there—opened a pull request to update that)

8

For what it’s worth, I have had freezes yesterday and today with the log showing drm issues. But that was still with Fedora 43 and kernel 6.19.11-200.fc43.x86_64
I’ve updated to Fedora 44 Beta today to try to “outrun” the problem. Maybe you want to compare the journalctl messages:

$ journalctl -b -3 -r
Apr 15 15:16:54 nx1 kernel: amdgpu 0000:03:00.0: [drm] *ERROR* [CRTC:363:crtc-0] commit wait timed out
Apr 15 15:16:54 nx1 kernel: amdgpu 0000:03:00.0: [drm] *ERROR* flip_done timed out
Apr 15 15:16:43 nx1 kernel: amdgpu 0000:03:00.0: amdgpu: [drm] *ERROR* [CRTC:363:crtc-0] hw_done or flip_done timed out
Apr 15 15:16:35 nx1 kernel: amdgpu 0000:03:00.0: [drm] *ERROR* [CRTC:363:crtc-0] flip_done timed out
9

Yeah, there are already opened drm issues about it for example here: Making sure you're not a bot!, but like I commented on this bug, it was in 6.19.11 and 7.0.0-rc.7, but not in 6.19.12 or 7.0.0.

10

I don’t test rawhide, so am not sure of the best way to get back to the developers.

I do know that the developers seldom have time to monitor this forum which supports the fully released versions. They are sure to see bug reports, and probably the test mailing list. I am unsure about matrix.

12

Just to close the loop on this one, the issue got isolated to commit d6e152d905bdb1f32f9d99775e2f453350399a6a (reverting it fixes the problem), there is a proposed patch Making sure you're not a bot! that is reported to also fix the problem, but I’m still trying to test it myself due to code drift from the proposed patch to the current v7.0 tag, but, hopefully, final version will be picked up by 7.0.1

Adjusted patch for the v7.0 tag can be found here: Making sure you're not a bot!

13

You indicated in the title that this was resolved, but failed to mark the appropriate post as the solution. I suspect it was your last post just above.

Please mark the appropriate post as the solution.