00:07karolherbst: anybody up for a quick review or shall I just merge https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17969 ?
02:09Wally: Did anyone answer my gem offset question? (My X server hanged and crashed a couople times)
02:09airlied: Wally: nope, but I think the offset is the vm address for the bo
02:10Wally: The mapped address?
02:12airlied: not sure what you mean by mapped there
02:12airlied: but it's the GPU address for the bo
02:12Wally: So not in CPU ram...
02:13Wally: The virtual memory of the GPU
02:13airlied: yes the vm address on the gpu side
02:31Wally: I doubt that the offset is the sum of the the sizes of all the buffers that come before them but I also doubt that there isnt some pattern to it...ogk doesnt seem to have some way of getting
02:31Wally: it...will investigate tommorow
02:42airlied: nope it's allocated from a gpu va address space allocator
02:42airlied: it's like ogk doesn't have one of those kernel side, and it needs to be done in the compat laye
02:43airlied: plans for nouveau to start doing it in mesa at some point
02:44Wally: Where would I start on creating one of those compat layers?
02:46Wally: (also are you objecting to it being the sum of all the sizes or objecting to there being a pattern)
02:47airlied: bit of both
02:47airlied: there is definitely alignment requirements
02:47airlied: but you can try doing something simple to start I suppose
02:54Wally: I guess I could calculate it if I was doing block linear uncompressed allocations though
02:54airlied: tiling also matters I think
02:56Wally: wouldnt block linear imply contiguos(no idea how to spell it) tiling?
07:29casejuarez: karolherbst: I can confirm now that the issue must be related to the distribution (qubes). I did a fresh install of Arch and no problems there, thanks a lot for the help
17:49ilgaz: firefox vaapi causes entire OS to freeze with nvidia 9400 (yes that one) https://bugzilla.mozilla.org/show_bug.cgi?id=1780962
17:49ilgaz: running in usermode of course
17:50ilgaz: Good thing is it comes disabled by default
17:54ilgaz:verified vaapi acceleration works perfectly. e.g. mpv
18:22karolherbst: ilgaz: probably multithreading at work here... I have an MR which might or might not fix it in case you want to try it out: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10752
18:22karolherbst: didn't do a lot of vaapi/vdpau testing though
18:24ilgaz: hello karolherbst. I will sure test tomorrow. mpv etc are all fine. It is just firefox it seems
18:25karolherbst: firefox also is doing OpenGL
18:25karolherbst: doing multiple GPU things in the same process is causing issues
18:34ilgaz: thank you BBL
20:43karolherbst: glennk: I assume with 12154 gnome will render fine? I think I will try that out on my nv4x GPUs tomorrow then if I don't forget. Is there anything else left to fix or is that essentially it?
20:43karolherbst: though my nv4x GPUs tend to be super unstable for whatever reason
20:43karolherbst: but maybe that's also just the scissor state
20:45glennk: oh, totally forgot about that one, haven't touched nouveau in a while...
20:46glennk: kernels after the big nouveau refactor are super unstable with nv4x
20:46glennk: try with something older like 4.15 or so
20:47karolherbst: I don't have any networking with such old kernels
20:47karolherbst: guess why I never figured out if there was a regression or not :(
20:47karolherbst: glennk: do you think you might be able to find some time git bisecting the kernel to figure out what broke it?
20:47karolherbst: might give me an idea on how to fix it then
20:49glennk: karolherbst, i've mentioned this before in the channel, its just a big overwrite merge with 20k+ lines where before = stable, after = random lockups
20:50karolherbst: .... crap
20:50karolherbst: I suspect memory management is totally busted, but...
20:55karolherbst: glennk: anyway.. I don't think I have the commit in my logs or anything :(
21:00glennk: its a big merge commit with ttm and nouveau refactorings
21:01glennk: can probably just look for the biggest nouveau commit in git log
21:01glennk: post 5.4 something
21:02karolherbst: maybe it's easier to try the git bisect myself then...
21:02glennk: the lockups aren't hard to reproduce at least
21:02karolherbst: my system dies after I launch glxgears a few times
21:05glennk: glxgears tended to work on mine, but anything complicated would just lock up after a second or so
21:05karolherbst: for me gnome worked fine... well besides the rendering issues
21:05glennk: gnome is not complicated rendering wise
21:06karolherbst: yeah.. probably
21:06glennk: the only thing is it _really_ likes to use glScissor
21:07karolherbst: probably for a reason
21:08glennk: as always, better options available, but more implementation effort