11:52pabs3: I just had a GPU hang with 4.19.28: http://paste.debian.net/1082073/
11:53pabs3: 01:00.0 VGA compatible controller : NVIDIA Corporation GK107 [GeForce GT 740] [10de:0fc8] (rev a1)
11:57RSpliet: pabs3: STACK_ERROR sounds like something that can be analysed. It's been a while since I dived into nouveau, but it'd be nice if we could capture the offending shader
11:57RSpliet: karolherbst: any idea whether this is possible?
11:57RSpliet: pabs3: When did this happen? Straight after boot?
11:57pabs3: after several days of uptime
11:58RSpliet: Ah... were you doing something with the machine?
11:58pabs3: I had GNOME shell and totem running watching a movie
11:58pabs3: the display froze so I had to reboot
11:58RSpliet: Hmm... ok, so nothing that you can reproduce.
11:59pabs3: nope. if I ever have it again, is there anything I can check over ssh?
11:59RSpliet: Nah, once it's reported in dmesg it's too late
11:59RSpliet: pabs3: what version of mesa you running?
12:00pabs3: OpenGL core profile version string: 4.3 (Core Profile) Mesa 18.3.4
12:00pabs3: (Debian buster)
12:01RSpliet: Slightly outdated, but not sure whether any control flow changes have been made since.
12:02RSpliet: Of course the second problem is the hang. It nicely kills the channel, but ideally you'd have a fresh wayland or x.org popping up asking you to log in rather than a system freeze
12:37karolherbst: pabs3: ufff
12:37karolherbst: is that reproducible?
12:38karolherbst: I guess something messed up the hardware context and killing the offending process would unfreeze X
12:38karolherbst: that's kind of a silly issue
12:38pabs3: I haven't had a GPU hang in ages, its a one-off
14:40sigod: could audio/video sync problems be caused by not having non-free firmware (eg nouveau-fw) installed?
14:42RSpliet: sigod: without nouveau-fw, no attempt is made to decode video using VDPAU or VA-API.
14:44sigod: RSpliet, so would that lead to sync issues?
14:45RSpliet: Rather, software codecs are used for video decoding. Sync problems then generally point to bugs in the software decoder, or the audio stack, or the 2D graphics stack. Not very specific I know. Installing these firmwares could replace the software decoder with a hardware decoder - and because the frames live on the GPU already, the 2D operations needed for display compositing are different.
14:45RSpliet: If sync issues are the result of the audio stack, it wouldn't make much of a difference. If the software decoder isn't keeping up, using the HW decoder could solve issues (... but are prone to introduce different ones).
14:46RSpliet: So in short: it's worth a shot?
14:47sigod: RSpliet, i am using parabola so the kernel won't load non-free firmware...
14:48sigod: and my gtx660 has the context switching crash so hardware acceleration is disabled too
14:56sigod: it's weird because html5 videos are in sync but videos i play in vlc and smplayer are out of sync
15:03RSpliet: sigod: I see. However, if you disable hardware acceleration, I don't see how this could be a nouveau problem.
15:04RSpliet: As nouveau and your NVIDIA graphics card are just funnelling pixels to your screen, nothing else.
15:05sigod: RSpliet, oh ok
15:05RSpliet: Problems more likely lie with the used codecs or the output methods chosen in your video player. VLC supports like 8 different output methods, try choosing a few different ones and see if they make a difference.
15:06sigod: it's just strange that both vlc and smplayer do it
15:06sigod: i have tried them
15:08karolherbst: sigod: sync as in audio/video sync?
15:08karolherbst: sometimes the movie file is just out of sync
15:11RSpliet: sigod: did you try all of these: http://www.attabot.org/wp-content/uploads/2015/02/vlc-output.jpg ?
15:11sigod: but on windows it is ok in some instances
15:11RSpliet: (well, likely some of them aren't available on your machine, but... ;-))
15:11sigod: RSpliet, thanks
15:14sigod: RSpliet, maybe opengl will work better
15:15RSpliet: Unlikely, you said you disabled acceleration so I'd expect that to add a lot of unnecessary overhead... but worth a try.
15:22karolherbst: well, audio/video synchronisation isn't exactly a trivial issue
15:23karolherbst: could be that some timestamping is off?
19:03Kubast2: Hey I have a very specific question, but it might be more fit for freedesktop/x11/wayland/mesa channels
19:04Kubast2: What happens when a program that uses a lot of vram gets sigstopped
19:04Kubast2: Another vram heavy app gets started
19:04Kubast2: It then gets stopped and original program getd sigcont
19:04Kubast2: Is there some page out like thing for gpus in linux?
19:07Kubast2: Would a gpu just request the geometry textures again, the program would had to implement it or what
19:07karolherbst: Kubast2: there is some paging out, but at draw time everything one single draw requires has to be paged in
19:08karolherbst: which kind of works for some stuff, but I think it needs a more careful look on how to improve things here and kind of being able to tell the driver that something should be inside VRAM, but it wouldn't be bad if it has to be pushed into system mem
19:14Lyude: karolherbst: just noticed: v2 is missing the cc to stable. Also, I've got someone who's going to help out with testing this on their Xaomi laptop running 5.1, so if greg pokes you later down the road for a 5.1 backport just let me know and I can send that to them
19:16karolherbst: I am still kind of waiting on what skeggsb is thinking about that fix. If it helps with at least some laptops then it's definietly a step into the right direction, but I would also understand if he complains that we don't even know the actual issue here :/
19:17Lyude: karolherbst: I understand
19:18karolherbst: I was hoping we could get a faster answer from nvidia, but... well, that didn't work out so well
19:18karolherbst: Lyude: Maybe we could ask them if at some point they had to workaround silly intel PCIe controllers for fixing the on windows
19:19karolherbst: this gives me an idea
19:20karolherbst: mhh yeah well... searching for windows bug in this direction doesn't really work out so well