00:19fdobridge: <J., Echo (she) 🇱🇹> This looks kinda weird 🐸
00:19fdobridge: <J., Echo (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1085356626950766674/Screenshot_20230315_021801.png
00:27fdobridge: <J., Echo (she) 🇱🇹> Here's Xonotic as well 🐸
00:27fdobridge: <J., Echo (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1085358613964529806/Screenshot_20230315_022630.png
00:45fdobridge: <J., Echo (she) 🇱🇹> One commit revert later and :ferris_happy:
00:45fdobridge: <J., Echo (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1085363018864537640/Screenshot_20230315_023819.png
01:16fdobridge: <karolherbst🐧🦀> heh
01:16fdobridge: <karolherbst🐧🦀> looks funky
01:16fdobridge: <karolherbst🐧🦀> yeah..
01:16fdobridge: <karolherbst🐧🦀> I'm convinced that the modifier stuff is still super broken in nouveau
01:16fdobridge: <karolherbst🐧🦀> one way or the other
10:07bylaws: Where is this channel bridged to?
10:11fdobridge: <Esdras Tarsis> bylaws: Discord
10:11bylaws: Do you have a link?
10:12fdobridge: <Esdras Tarsis> yes
10:12fdobridge: <Esdras Tarsis> https://discord.gg/D7xcHS4V
10:21fdobridge: <ByLaws> Thanks :)
11:24fdobridge: <J., Echo (she) 🇱🇹> Why does nouveau have higher CPU temperatures than NVIDIA driver? 🐸
11:47fdobridge: <J., Echo (she) 🇱🇹> And could GSP fix that?
13:01fdobridge: <karolherbst🐧🦀> because we are bad at power management, and yes
19:11fdobridge: <J., Echo (she) 🇱🇹> I'm typing this from non-GSP nouveau 😅
19:37fdobridge: <J., Echo (she) 🇱🇹> When will nouveau provide proper GPU statistics that MangoHUD can use for example? Also a program like radeontop/nvtop/intel_gpu_top would be nice 🐸
19:39fdobridge: <J., Echo (she) 🇱🇹> Also Discord app weirdly crashes the compositor on nouveau (only in Wayland mode though)
19:40fdobridge: <J., Echo (she) 🇱🇹> I should probably make an issue for this too because it's kinda a dealbreaker
19:42fdobridge: <Esdras Tarsis> I think nouveau needs to provide more information via hwmon, at the moment it only informs fan speed and temperature iirc
19:45fdobridge: <J., Echo (she) 🇱🇹> Also will there be any work done on HW video stuff for newer GPUs (VA-API) once nouveau becomes more attractive with NVK and GSP work currently being done?
19:49airlied: at some point nvk will get vulkan video
19:49fdobridge: <J., Echo (she) 🇱🇹> Basically nothing supports Vulkan video though (I think)
19:50airlied: by the time nvk is useful things will
19:52fdobridge: <Conan Kudo (ニール・ゴンパ)> if it would help, I could provide a copr with vkvideo enabled ffmpeg
19:53fdobridge: <Conan Kudo (ニール・ゴンパ)> I just went through the painful process of upgrading ffmpeg to v6 in f38+, so it's a thing we can do
19:54fdobridge: <Conan Kudo (ニール・ゴンパ)> since I think the patches apply fairly cleanly now
19:55fdobridge: <J., Echo (she) 🇱🇹> I wonder when the year of the nouveau desktop will come
19:56fdobridge: <karolherbst🐧🦀> when GSP support is ready obviously 😛
19:59fdobridge: <Conan Kudo (ニール・ゴンパ)> I'm surprised we haven't gotten the GSP into linux-firmware yet
20:00fdobridge: <Conan Kudo (ニール・ゴンパ)> people have been asking about it for a while, especially after seeing the work in linux 6.2...
20:02fdobridge: <karolherbst🐧🦀> it's..... huge
20:02fdobridge: <karolherbst🐧🦀> and every driver version releases a new binary
20:02fdobridge: <karolherbst🐧🦀> like 30MB per file
20:03fdobridge: <karolherbst🐧🦀> and we have two of those per driver release atm
20:03fdobridge: <Conan Kudo (ニール・ゴンパ)> what about at least getting the licensing changed so it can be shipped without the rest of the nv driver?
20:03fdobridge: <karolherbst🐧🦀> I think that's already allowed
20:04fdobridge: <karolherbst🐧🦀> @airlied brought that topic up with some kernel folks I think
20:04fdobridge: <karolherbst🐧🦀> but we might have to ask nvidia to throw it into a special git repo
20:04fdobridge: <Conan Kudo (ニール・ゴンパ)> Not yet? https://github.com/NVIDIA/open-gpu-kernel-modules/issues/62#issuecomment-1124575459
20:04fdobridge: <karolherbst🐧🦀> well, it does say it's allowed?
20:04fdobridge: <Conan Kudo (ニール・ゴンパ)> if it's split out or whatever, then we can ship it and people can optionally switch it on
20:05fdobridge: <karolherbst🐧🦀> ohh `No Separation of Components. ` ehh
20:05fdobridge: <karolherbst🐧🦀> Let me bring it up with Nvidia then 😄
20:06fdobridge: <karolherbst🐧🦀> I have more effective ways than a github issue to do this
20:07fdobridge: <Conan Kudo (ニール・ゴンパ)> Thank you!
20:07fdobridge: <Conan Kudo (ニール・ゴンパ)> I would like to have this in Fedora, and we basically can't right now 😢
20:08fdobridge: <karolherbst🐧🦀> yeah.. but we have more serious issues
20:08fdobridge: <airlied> we won't be throwing in Fedora anytime soon
20:08fdobridge: <karolherbst🐧🦀> people will fight this if it means +150MB isos
20:08fdobridge:<Conan Kudo (ニール・ゴンパ)> did his own packaging of the driver for RH/Fedora ecosystem to play with it with his one computer with a compatible NV GPU
20:08fdobridge: <Esdras Tarsis> we need stable ABI support
20:08fdobridge: <airlied> getting linux-firmware right is going to be hard enough
20:08fdobridge: <airlied> I expect we will pick a version and try to stick to it for long periods of time
20:09fdobridge: <karolherbst🐧🦀> I'm sure we won't get it
20:09fdobridge: <karolherbst🐧🦀> or at least not any time soon
20:09fdobridge: <Conan Kudo (ニール・ゴンパ)> sure, it's a long road, but fixing legal issues up front would be good first 🙂
20:09fdobridge: <karolherbst🐧🦀> yeah.. I have experience with Nvidia lawyers...
20:09fdobridge: <Conan Kudo (ニール・ゴンパ)> because they tend to be the slow path
20:09fdobridge: <karolherbst🐧🦀> and I leave it at that
20:09fdobridge: <Conan Kudo (ニール・ゴンパ)> lol
20:09fdobridge: <Conan Kudo (ニール・ゴンパ)> I'm just excited to see a revitalization of nouveau
20:09fdobridge: <airlied> I think they've already cleared it for redistribution, just nobody fixed up the docs
20:10fdobridge: <karolherbst🐧🦀> yeah, I think everybody already thinks it's redistributable 😄
20:10fdobridge: <airlied> but figuring out what version to redistribute is going to be the hard problem
20:10fdobridge: <karolherbst🐧🦀> well.. we have to version it anyway
20:10fdobridge: <airlied> certainly no point in considering it too much before we get feature parity with non-gsp
20:11fdobridge: <karolherbst🐧🦀> initramfs size will also blow
20:11fdobridge: <airlied> I don't think any of the current ones are worth redistributing yet
20:11fdobridge: <karolherbst🐧🦀> so what should nouveau do? declare it supports the past 10 versions?
20:11fdobridge: <karolherbst🐧🦀> bye bye `/boot`
20:11fdobridge: <karolherbst🐧🦀> only support one and regress on systems only having old ones?
20:11fdobridge: <karolherbst🐧🦀> that will give us kernel regression bugs
20:11fdobridge: <airlied> I'm considering we don't put it in the initramfs and only load nouveau late
20:11fdobridge: <karolherbst🐧🦀> no
20:11fdobridge: <karolherbst🐧🦀> we have to
20:12fdobridge: <airlied> why?
20:12fdobridge: <karolherbst🐧🦀> dracut e.g.
20:12fdobridge: <airlied> why?
20:12fdobridge: <karolherbst🐧🦀> we need kms
20:12fdobridge: <Conan Kudo (ニール・ゴンパ)> wouldn't KMS be utterly broken if we didn't?
20:12fdobridge: <airlied> we have simplekms
20:12fdobridge: <karolherbst🐧🦀> es need a proper driver
20:12fdobridge: <airlied> es?
20:12fdobridge: <karolherbst🐧🦀> typo
20:12fdobridge: <karolherbst🐧🦀> full disc encryption relies on it
20:12fdobridge: <airlied> no it doesn't
20:12fdobridge: <karolherbst🐧🦀> think about a laptop, which is closed + external display
20:12fdobridge: <karolherbst🐧🦀> it does
20:12fdobridge: <karolherbst🐧🦀> you don't want to have to open your laptop to boot it
20:12fdobridge: <karolherbst🐧🦀> that's just stupid
20:12fdobridge: <airlied> if your laptop is closed + external display, will you see anything?
20:12fdobridge: <karolherbst🐧🦀> point is: we _require_ kms
20:13fdobridge: <karolherbst🐧🦀> not without kms
20:13fdobridge: <karolherbst🐧🦀> yes with kms
20:13fdobridge: <Conan Kudo (ニール・ゴンパ)> you're supposed to
20:13fdobridge: <karolherbst🐧🦀> full disc encryption prompt is mirrored on every display
20:13fdobridge: <karolherbst🐧🦀> anyway. I'll NAK it either way
20:13fdobridge: <airlied> don't closed laptops generally boot external displays?
20:13fdobridge: <Conan Kudo (ニール・ゴンパ)> nope
20:13fdobridge: <karolherbst🐧🦀> we need kms, I'm not debating this
20:13fdobridge: <Conan Kudo (ニール・ゴンパ)> because the external display port is wired into the NV GPU
20:14fdobridge: <Conan Kudo (ニール・ゴンパ)> whereas the internal display is not
20:14fdobridge: <airlied> ah you mean for dual-gpu ah yes
20:14fdobridge: <Conan Kudo (ニール・ゴンパ)> there are no single-GPU NV laptops on the market
20:14fdobridge: <karolherbst🐧🦀> also not all single ones mirror on the firmware level
20:14fdobridge: <karolherbst🐧🦀> it's all very silly
20:14fdobridge: <karolherbst🐧🦀> and then there are random other systems
20:14fdobridge: <karolherbst🐧🦀> and random other reasons we might need kms in dracut
20:14fdobridge: <Conan Kudo (ニール・ゴンパ)> my new work P1 has that problem with a simple Intel iGPU
20:15fdobridge: <Conan Kudo (ニール・ゴンパ)> it makes me hate FDE so much
20:15fdobridge: <karolherbst🐧🦀> yeah..
20:15fdobridge: <karolherbst🐧🦀> kms is the only sane and reliable option here
20:15fdobridge: <airlied> so it's broken now even with a kms driver?
20:15fdobridge: <Conan Kudo (ニール・ゴンパ)> yes
20:15fdobridge: <Conan Kudo (ニール・ゴンパ)> randomly broken
20:15fdobridge: <karolherbst🐧🦀> well.. it's supposed to work at least with kms
20:15fdobridge: <Conan Kudo (ニール・ゴンパ)> I accidentally bricked my laptop yesterday doing it
20:16fdobridge: <Conan Kudo (ニール・ゴンパ)> * doing software updates
20:16fdobridge: <Conan Kudo (ニール・ゴンパ)> (nearly)
20:16fdobridge: <karolherbst🐧🦀> on my dell laptop the external display (because you know, thunderbolt) only lights up once `i915` is loaded
20:16fdobridge: <Conan Kudo (ニール・ゴンパ)> same!
20:16fdobridge: <Conan Kudo (ニール・ゴンパ)> also Thunderbolt is evil
20:16fdobridge: <airlied> okay then someone needs to work out how to make the fw loadable 😛
20:16fdobridge: <karolherbst🐧🦀> yeah.. it's a bit scuffed
20:16fdobridge: <karolherbst🐧🦀> yeah...
20:16fdobridge: <karolherbst🐧🦀> it's a big mess
20:16fdobridge: <airlied> I had considered making a secondary initramfs
20:16fdobridge: <Conan Kudo (ニール・ゴンパ)> there was a bit of work done for that for specifically apple hardware
20:17fdobridge: <airlied> so at least we'd only have one copy of it instead of 3-10
20:17fdobridge: <karolherbst🐧🦀> we could also make dracut smarter
20:17fdobridge: <Conan Kudo (ニール・ゴンパ)> we do this for initializing Apple Silicon hardware
20:17fdobridge: <karolherbst🐧🦀> and only copy in the latest one
20:17fdobridge: <karolherbst🐧🦀> do we have a way of declaring hierarchy of files in kernel modules?
20:17fdobridge: <Conan Kudo (ニール・ゴンパ)> the `asahi` dracut module does some simple stuff for doing this
20:17fdobridge: <karolherbst🐧🦀> I think other drivers might benefit from it as well
20:17fdobridge: <Conan Kudo (ニール・ゴンパ)> though it probably isn't exactly what you want
20:18fdobridge: <karolherbst🐧🦀> yeah.. I'm more thinking kernel level fixing
20:18fdobridge: <Conan Kudo (ニール・ゴンパ)> there's a hierarchy of firmware directories
20:18fdobridge: <Conan Kudo (ニール・ゴンパ)> I don't know about firmware files
20:18fdobridge: <karolherbst🐧🦀> like to be able to declare "those files are updates of this other group of files, copy only the 'best' one"
20:18fdobridge: <airlied> we'd need a better way of stating firmware deps in that I only need one of those
20:18fdobridge: <airlied> not all of them like we have now
20:18fdobridge: <karolherbst🐧🦀> yeah
20:19fdobridge: <karolherbst🐧🦀> which might allow us to shrink initramfs size today as well
20:19fdobridge: <karolherbst🐧🦀> (just to make it bigger with gsp 🙃 )
20:19fdobridge: <karolherbst🐧🦀> at least it makes the situation less bad
20:19fdobridge: <Conan Kudo (ニール・ゴンパ)> one of the patches for kernel-asahi is to add more firmware load paths: https://gitlab.com/fedora-asahi/kernel-asahi/-/commit/f04e2f542586846ff4e0a135f0a1707e21b772f9
20:20fdobridge: <karolherbst🐧🦀> mhh
20:20fdobridge: <Conan Kudo (ニール・ゴンパ)> there's just no intelligence beyond that yet, though
20:20fdobridge: <karolherbst🐧🦀> ohhh
20:20fdobridge: <karolherbst🐧🦀> @airlied I have an idea... we could add a version field, and dracut would pick up the latest?
20:20fdobridge: <karolherbst🐧🦀> that might not be terrible
20:21fdobridge: <airlied> I think we'd need to have some sort of way to group them in the module info
20:21fdobridge: <airlied> so that userspace could work it out
20:21fdobridge: <karolherbst🐧🦀> like a "firmware_type/name/whtvr" + "version" field rather
20:21fdobridge: <karolherbst🐧🦀> and then per type it picks the newest
20:21fdobridge: <airlied> yeah
20:21fdobridge: <airlied> just not sure how to encode that
20:21fdobridge: <airlied> and be kinda backwards compatible
20:21fdobridge: <karolherbst🐧🦀> maybe bring it up with kernel maintainers? I wouldn't be surprised if somebody already has a plan, but was like "nah.. not going to do this to save a few kb"
20:23fdobridge: <airlied> yeah I might kick off a thread and see what crap greg comes back with to say it's a bad idea 😛
20:23fdobridge: <karolherbst🐧🦀> cool
20:23fdobridge: <karolherbst🐧🦀> could be a cool GSoC project, but could also be a bad one, because it means dealing with kernel maintainers 🙃
20:24fdobridge: <airlied> not sure if we have the runway for doing it via GSoC if it's a requirement for us to land gsp
20:24fdobridge: <karolherbst🐧🦀> yeah..
20:25fdobridge: <airlied> much more likely some Red Hat engineer would be asked to look at it 😛
20:25fdobridge: <karolherbst🐧🦀> I mean.. in the end I'd think it's a bed idea, because.. $upstream 😛
20:25fdobridge: <karolherbst🐧🦀> heh
20:31fdobridge: <airlied> like I think there will be the push back of don't support it without a stable ABI
20:31fdobridge: <airlied> which assumes we are in a better bargaining position than we are 😛
20:34fdobridge: <Mohamexiety> kinda unrelated, sorry, but speaking of this.. I will be picking up another GPU soon for NVK work. do you know if having a GSP+non-GSP GPU connected would work (and NVK would work properly)? @airlied
20:34fdobridge: <airlied> yes two of them should be fine
20:34fdobridge: <Mohamexiety> it'll most likely either be a GTX 750(Ti) (GM107) or GT 1030 (GP108)
20:34fdobridge: <Mohamexiety> alright then, thanks!
20:35fdobridge: <airlied> @karolherbst🐧 okay I kicked off a thread
20:48fdobridge: <karolherbst🐧🦀> nice
20:48fdobridge: <karolherbst🐧🦀> which topic?
20:49fdobridge: <karolherbst🐧🦀> or rather.. which ML 😄
21:12fdobridge: <airlied> modules on dri-devel/modules
21:33fdobridge: <J., Echo (she) 🇱🇹> What could cause this?: `kernel: nouveau 0000:01:00.0: gr: TRAP ch 2 [00ffe46000 kwin_wayland[13400]]`
21:36fdobridge: <karolherbst🐧🦀> anything in a shader
21:36fdobridge: <karolherbst🐧🦀> is this reliable btw?
21:38fdobridge: <J., Echo (she) 🇱🇹> I encountered this exact error in multiple scenarios consistently (opening Discord in Wayland mode, running Firefox with DRI_PRIME=1, resizing the SuperTuxKart window with DRI_PRIME=1)
21:46anholt: I don't suppose any of the recent work on nvk or drm would have come up with a reason that SSBOs are tremendously unstable on tegra, would it?
21:47karolherbst: mhhhh
21:47karolherbst: soooo
21:47karolherbst: we have a lurking bug in nouveau since forever
21:47karolherbst: fp helper invocations can't load from memory
21:48karolherbst: or well.. I think it's kepler+ actually
21:48karolherbst: but that's since forever, but maybe a fix for that helps the instability?
21:48karolherbst: dunno
21:48anholt: kepler and maxwell is what I've tested.
21:48anholt: how bad's the fix?
21:49anholt: also, it's unstable in tess and compute, so helper invocations seem unlikely
21:50karolherbst: depends on how proper the fix is
21:51karolherbst: anholt: try this: https://gitlab.freedesktop.org/drm/nouveau/-/commit/99401da29004b777d6999bf78206f989423e3985
21:51karolherbst: the "proper" fix is, to flip that from userspace on a per context base, and there is some magic gr firmware stuf fhappening for this
21:51karolherbst: still need to verify it and write down the code
21:51karolherbst: so if that helps with stuff in CI, I'd focus on it even more
21:51karolherbst: but for nvk and the CTS it was causing infinite loops and stuff
21:52karolherbst: (not having that)
21:57fdobridge: <J., Echo (she) 🇱🇹> nouveau has some weird glitch when switching between cursors on KWin (having a display connected to nouveau driver is enough to reproduce this)
23:26fdobridge: <J., Echo (she) 🇱🇹> i can't find NAK source code for some reason
23:42fdobridge: <Esdras Tarsis> https://gitlab.freedesktop.org/gfxstrand/mesa/-/tree/nak/main
23:43fdobridge: <Esdras Tarsis> src/nouveau/compiler
23:50fdobridge: <J., Echo (she) 🇱🇹> I don't see it in nouveau/mesa though