22:04karolherbst: soo.. even more patches https://gitlab.freedesktop.org/drm/nouveau/-/merge_requests/8
22:04karolherbst: Lyude, skeggsb_ ^^
22:32marex: karolherbst: you dont happen to have an idea what's up with that quadro of mine above , do you ?
22:36karolherbst: marex: what was the issue?
22:37karolherbst: ohh
22:37karolherbst: seeing it now
22:37karolherbst: I think there was a fix for that
22:38karolherbst: let me check
22:40karolherbst: marex: do you think you'd be up for git bisecting the problem?
22:44marex: karolherbst: in where, kernel, mesa, (is libdrm still used?)
22:44karolherbst: kernel
22:45karolherbst: marex: I guess it's super relibly triggered? like it happens 100% of the time?
22:46marex: I think there are two problems now
22:46karolherbst: I have a vague idea what might be broken, but I think it also makes sense to figure out what broke it...
22:46karolherbst: yeah.. could be
22:46marex: one this is MTHD stuff, that's new
22:46karolherbst: I can try to reproduce it tomorrow maybe
22:46karolherbst: but changes are ... low for whatever reason
22:46marex: the other is this wlroots/sway suspend where some object is still referenced by the kernel/gpu/whatever and the system takes forever to suspend
22:47karolherbst: yeah...
22:47karolherbst: mhh
22:47marex: karolherbst: got any specific commit in mind ?
22:47Lyude: karolherbst will look tomorrow, feel free to poke if I forget
22:47karolherbst: Lyude: cool
22:47karolherbst: marex: nope
22:47karolherbst: at least now I am more motivated to work on those fixes, because I have a path to push them upstream relibly :D
22:48marex: I can start complaining about unresponsive maintainers anytime ...
22:49marex: hmmmm, so debian has either linux 5.10.y or 5.14.y prepackaged, that sucks
22:49marex: that's too large range
22:50marex: but then I should be able to build packages for the rest, which would at least reduce the bisect range ther
22:50marex: *there
22:50karolherbst: yeah.. I think you need to compile anyway :(
22:50karolherbst: well...
22:50karolherbst: you'd skip two rebuilds if you figure out the major version breaking it
22:51karolherbst: out of ~14...
22:52marex: but then I can look at git log --oneline --no-merges v5.1n..v5.1m -- drivers/gpu/drm/nouveau/ and try to guess which commit is relevant
22:52marex: all those gk10x and stuff I can likely filter out already
22:52karolherbst: marex: I stoped doing subdirectory bisect because they also break for weird reasons
22:52karolherbst: I just don't assume anything, so I never have to restart stuff ...
22:53marex: karolherbst: what does "gr:" and "ILLEGAL_MTHD" even mean anyway ?
22:53karolherbst: we submitted a command to the GPU which isn't known
22:53karolherbst: ILLEGAL_CLASS means we used an "object class" which doesn't exist
22:53karolherbst: it's like namespaces but in hw
22:53marex: and can I tell which command it was ?
22:54karolherbst: it happens inside nv50_fbcon_imageblit
22:54karolherbst: but...
22:54karolherbst: I don't think this helps
22:54karolherbst: all this stuff is quite complicated, so a random change could break it
22:57marex: oh, so this mthd 0x0860 is just opcode for pushing pixel data from CPU into the GPU ?
22:59karolherbst: yeah
22:59karolherbst: the thing is
22:59karolherbst: this method exists
22:59karolherbst: but the class is 0
22:59karolherbst: which is wrong :)
22:59karolherbst: so I think we teared down the object on the GPU too soon or something
23:13marex: karolherbst: all right, if you forget, I'll try to bisect this (i.e. I'll add it to the todo)