Video/audio hesitation in 5.15.11 kernel?

tld · Veteran Joined: 09 Dec 2003 Posts: 1852

I recently updated all my machines from gentoo-sources-5.10.76-r1 to gentoo-sources-5.15.11. I'm noticing some hesitation and similar issues that I don't have with the 5.1.76-r1 kernel, for example when jumping forward or back. On my MythTV fromend I just switched back to the old kernel for that reason, as it was noticeable enough to be annoying. That machine is an x86 machine using nVidia with the nvidia-drivers-390.147 driver (as required by my GT 430 card). I have the same nVidia driver for both kernels. I'm using vdpau. To be clear, I have no idea if the issues are specifically related to the video or drivers at all.

Here's some of what I noticed on that machine: When playing OTA TV recordings and jumping forward or back, it takes a moment for the audio to kick after a jump. On things like h264 videos, sometimes after a jump there's a temporary slowdown in the video motion. None of this happens in 5.10. Also note that all of that is getting video over my LAN (using a wired connection).

Has anyone seem or heard of anything like this? I'm actually not sure I've ever had a notable regression upgrading to a stable kernel that I can recall until this.

Thanks in advance.
Tom

tld · Veteran Joined: 09 Dec 2003 Posts: 1852

I've tried a few more things here. First of all I've tried the newest stable 5.15.16 kernel, as well as the newest stable 5.10.93 kernel. As was the case before, I get that odd hesitation only in the 5.15 kernel.

So far it appears that this is only happening for my in MythTV, and things like xine or mpv don't seem to be doing this. Interestingly, this happens in MythTV regardless of whether I'm watching TV recordings that use the mythtv protocol, but also when I play videos directly from the disk on the frontend itself...so it doesn't appear to be network related. That all complicates things a bit, because I'm still running MythTV 29.1 using an ebuild of my own in my local overlay. I'll just have to stick with 5.10 and try new kernels as they come out to see if there's any change. Sort of sucks.

Tom

Hu · Administrator Joined: 06 Mar 2007 Posts: 23082

Can you narrow down in which kernel version this regressed? The number of commits between 5.10 and 5.15 is quite large.

Can you check whether this happens with an untainted 5.15 kernel?

tld · Veteran Joined: 09 Dec 2003 Posts: 1852

Wow...testing all those interim kernel versions would be a massive undertaking. When you say "untainted", do you mean without the nVidia driver? If so then that's pretty much impossible for me with MythTV. That's all heavily dependent on vdpau, and no way would that old x86 machine even function trying to play content with something like nouveau for example.

I think I'm just going to have to see if this eventually gets resolved in some future kernel. This one's really tough as the issue could be related to video and/or audio/alsa etc...very hard to know. The audio I'm using is via the nVidia HDMI too. Really ugly one this is. I've posted to the MythTV users mailing list as well in case someone's seen anything like this.

EDIT: I just realized that there are no kernels in portage between the 5.10 and 5.15 kernels I have installed, so that test becomes basically impossible.

Thanks!
Tom

Hu · Administrator Joined: 06 Mar 2007 Posts: 23082

There are 78218 commits between v5.10 and v5.15.11. An optimal git bisect would require log2(78218) = ~17 tests. This is probably overly optimistic, and 20-25 would be more likely.

Yes, the nVidia driver taints the kernel. You're unlikely to get much upstream interest as long as the kernel is tainted by a proprietary blob.

You wouldn't use the Gentoo Portage ebuilds for this. It does not offer you the precision required to search non-tagged kernel releases.

tld · Veteran Joined: 09 Dec 2003 Posts: 1852

Thanks for the suggestions! I have serious doubts I'd get anywhere with this. I suspect this is just a matter of that legacy nVidia driver, combined with the fact that this is all 32-bit. nVidia no longer supports 32-bit, and I believe that support for that driver ends after 2022. This is almost surely buggy behavior in nVidia's support for the 5.15 kernel.

If I don't want to put up with those video hesitations (which again, only happen when jumping forward or back) I'll probably just have to stick with 5.10. Thanks again!

Tom