09:07tagr: karolherbst: one thing I remember from Tegra194 GPU work that I had done at some point is that newer generations of the hardware have gotten increasingly picky about values in certain registers
09:08tagr: one example is the aperture type that we always program as 0 (VRAM, I think) on Tegra as memory on Tegra was always considered VRAM, even though it was actually "shared" system memory
09:10tagr: newer GPUs, such as gv11b on Tegra194 now treat VRAM differently from system memory, so programming 0 into the aperture field will actually cause an error if it's not in fact referring to dedicated video memory
09:11tagr: so yeah, it's also entirely possible that some registers/methods use a default value that happens to work in some cases but then fail in others
09:12tagr: but I also recall that there are some instances where you just can't use block-linear buffers (I don't recall exactly what that was, but it might have had something to do with compression)
15:49Wally: `
17:46Newbyte: apologies if you got this question already, but in light of the Nouveau Vulkan driver, are you expecting to be able to figure out the firmware situation for newer NVIDIA GPUs?
17:47Newbyte: from what I understand they still won't provide a stable firmware "API" (I'm not sure that's the right term), right?
17:50karolherbst: Newbyte: we are allowed to use nvidias GSP firmware
17:50karolherbst: but yeah... if the API isn't stable it might be a bit of work to keep supporting it
17:50karolherbst: _but_
17:50karolherbst: we don't have to upgrade to every new firmware if there is no reason for it
17:50Newbyte: how will you use it? I assume it won't be going to linux-firmware
17:51karolherbst: that's to be determined
17:51Newbyte: all right, thanks :)
17:51karolherbst: well.. especially the linux-firmware situatio will be complicated, as gsp is massive
17:51karolherbst: and I doubt people will like it if we end up with 20 files or so :D
17:51karolherbst: dunno
17:51Newbyte: yeah, linux-firmware is already huge
17:52Newbyte: for example there was some commotion in Alpine Linux recently when they realised that like 90% storage use of a minimal Alpine Linux installation is linux-firmware alone
17:52Newbyte: but yeah, that's off-topic
17:53Newbyte: I was also wondering, will this Vulkan driver support older GPUs without the usable firmware? (ones without reclocking)
17:55karolherbst: if people find time to work on it, yes
17:56karolherbst: I hope we will get it to work on kepler and newer though
17:56karolherbst: older gens might be too painful to support
17:56karolherbst: atm we will start with turing, because of the firmware thing, but supporting at least maxwell shouldn't be a huge issue
17:56karolherbst: uhm.. maxwell and pascal
17:57Newbyte: makes sense
17:58karolherbst: _technically_ I think nv50 _could_ potentially be supported, but everything before kepler is probably a huge pita
18:00Newbyte: is there any chance that fan control could be possible on desktop Maxwell v2 GPUs?
18:00karolherbst: sure
18:00karolherbst: if nvidia releases their signing keys
18:00Newbyte: I would love to use Nouveau but my GPU's fans are super loud with it
18:00Newbyte: oh
18:00Newbyte: I had my hopes up for a moment
18:01karolherbst: welllll
18:01karolherbst: there is some blog post in the wild of some switch hackers being able to extract that key from the GPU
18:01karolherbst: and I've heard it's supposed to work on maxwell and pascal
18:01Newbyte: that's very interesting
18:01karolherbst: but that's a fuzzy situation legally
18:01karolherbst: you can't really make such a key public
18:02Newbyte: I assume you'd also have to write your own firmware then?
18:02karolherbst: that's not the problem
18:02karolherbst: the same reclocking code works for maxwell
18:02Newbyte: oh, I see
18:02karolherbst: we would just have to add fan control support to our own pmu fimrware
18:03karolherbst: we even use it for 1st gen maxwell
18:03karolherbst: there are a couple of GPUs out there
18:03karolherbst: GTX 750 I think?
18:03Newbyte: where is this firmware source code available by the way? I was trying to look for it a while ago but I wasn't sure where to find it
18:03Newbyte: yeah, 750 and the Ti variant right?
18:03karolherbst: somewhere inside nouveau... let me see
18:03Newbyte: thanks!
18:03karolherbst: drivers/gpu/drm/nouveau/nvkm/subdev/pmu/fuc/
18:04karolherbst: and a handful of mobile chips
18:04karolherbst: anyway.. on mobile we can just ignore this issue as the GPU isn't controling the fans anyway
18:04karolherbst: so it's really just about the 900 series
18:06Newbyte: I'm surprised how small this firmware is
18:06karolherbst: yeah.. it's not doing all that much
18:06karolherbst: it's essentially just parsing a script we upload
21:16TimurTabi: Lyude, karolherbst: are either of you going to pick up my nvbios_addr patch?
21:28Lyude: TimurTabi: if you sent it a while ago I may just not have noticed it, do you have a patchwork link?
21:28Lyude: (fwiw, there's probably still a good number of old patches that haven't gone through, but I've been trying to keep up to date with just about everything that's gone onto the list in the past few months
21:29Lyude: *that we haven't gone through
21:30TimurTabi: What's the patchwork for the nouveau list?
21:41TimurTabi: Here's the mailing list post: https://www.spinics.net/lists/nouveau/msg09598.html
22:35Lyude: TimurTabi: should be on patchwork.freedesktop.org
22:36Lyude: …weird, no idea why I didn't see that patch before.
22:36Lyude: TimurTabi: Reviewed-by: Lyude Paul <lyude@redhat.com>, I will push it in a little bit
22:41TimurTabi: Thanks,\