03:04fdobridge: <gfxstrand> I've pulled the change into origin/main
09:38fdobridge: <karolherbst🐧🦀> mhhh.. there are two new interesting bugs https://gitlab.freedesktop.org/drm/nouveau/-/issues/257 and https://gitlab.freedesktop.org/drm/nouveau/-/issues/256
09:39fdobridge: <karolherbst🐧🦀> both complain userspace running into `PAGE_NOT_PRESENT` errors after a kernel upgrade
09:39fdobridge: <karolherbst🐧🦀> ehh maybe not the first one
09:39fdobridge: <karolherbst🐧🦀> but I had another bug like somewhere else
09:39fdobridge: <karolherbst🐧🦀> would be funky if some of the memory bugs are indeed kernel regressions, would make it way easier to track down
09:43fdobridge: <karolherbst🐧🦀> anyway, I'll check if I can reproduce it reliably and if it's indeed something stable enough to run a `git bisect` on
13:10fdobridge: <karolherbst🐧🦀> ohhh nice
13:10fdobridge: <karolherbst🐧🦀> channel recovery seems to work way more reliable on 6.1 with fermi...
13:10fdobridge: <karolherbst🐧🦀> maybe I can figure this out and maybe this also affects other gens before turing
13:27fdobridge: <karolherbst🐧🦀> okay anyway, I'm convinced that channel recovery being unreliable is indeed a regression, and I have found an application which is reliably trigger crashes on the channel.. very nice
19:44fdobridge: <conan_kudo> have we figured out a plan for using GSP firmware for nouveau?
19:53fdobridge: <gfxstrand> What do you mean by "plan". There are patches and they're working on stabilizing it all.
20:08fdobridge: <conan_kudo> I mean a way to deal with not having to stuff it into initramfs
20:08fdobridge: <conan_kudo> I know about the patches and other things, but the big hangup I've heard is how to not keep bloating initramfs images
20:09fdobridge: <airlied> it's going into initramfs
20:10fdobridge: <airlied> the question is whether we can avoid carrying multiple of them in initramfs as we switch versions to support newer hw
20:10fdobridge: <conan_kudo> up to a point, probably
20:11fdobridge: <conan_kudo> when the NVIDIA driver drops older hardware, then probably not
20:11fdobridge: <conan_kudo> but GSP is GTX 16 / RTX 20 series+, so that may be far away
20:11fdobridge: <airlied> yes at that point we would have to address how many of them to ship or somehow use local installed hardware to pick it
20:12fdobridge: <conan_kudo> we could eventually plumb DNF to be aware to install needed firmware and autoremove unneeded firmware
20:13fdobridge: <conan_kudo> this is possible on openSUSE with Zypper, there was a proposal to port this to DNF a while ago: https://fedoraproject.org/wiki/Changes/Linux_Firmware_Minimization
20:20fdobridge: <airlied> there has also been discussions to move firmware to a separate initramfs so if we have X kernels installed we only have 1 copy of the fw
20:23fdobridge: <mohamexiety> given how each binary is tied to a particular driver release though, how could something like this be dealt with?
20:28fdobridge: <airlied> probably have to store some metadata for each installed kernel and take that into account when generating the updated fw initramfs
20:28fdobridge: <airlied> or just look at all the installed kernels and modules and work it out from scratch
20:29fdobridge: <mohamexiety> I see
20:35fdobridge: <conan_kudo> I keep wondering about this because the current state of affairs for nouveau is atrocious and I really would like for it to improve soon in Fedora :/
20:36fdobridge: <conan_kudo> and I know not everything has landed yet, but still...
20:36fdobridge: <conan_kudo> "so close and yet so far" is apropos here
21:28fdobridge: <karolherbst🐧🦀> pepole are discussing it afaik
21:29fdobridge: <karolherbst🐧🦀> there is a thread, it's not very public though
21:29fdobridge: <karolherbst🐧🦀> but anyway
21:29fdobridge: <conan_kudo> is nouveau with GSP firmware in as good or better state than nouveau with the older stuff yet?
21:29fdobridge: <karolherbst🐧🦀> people have it on the radar who are known to deal with those issues
21:29fdobridge: <karolherbst🐧🦀> uhm.. only time will tell
21:30fdobridge: <karolherbst🐧🦀> I won't make any statement on the quality unless users are starting to use it and we get some data
21:30fdobridge: <karolherbst🐧🦀> we can't really make any statements on that until it's out in the wild
21:31fdobridge: <karolherbst🐧🦀> the painful part with Nvidia is that you have to consider _every_ GPU to be different
21:31fdobridge: <karolherbst🐧🦀> and unless you test all GPUs you don't know what state anything is in
21:32fdobridge: <karolherbst🐧🦀> and with every I don't mean the model name or chipset name
21:32fdobridge: <karolherbst🐧🦀> I mean literally every product
21:33fdobridge: <karolherbst🐧🦀> well.. in theory even two GPUs from the same product line can be different, it's always hard to tell
21:35fdobridge: <karolherbst🐧🦀> but at least we won't have to ask nvidia 100 times to give us updated firmware because we think the one they gave us doesn't work on "newer" batches of GPUs
21:47fdobridge: <conan_kudo> should we try in Fedora then?
21:47fdobridge: <conan_kudo> should we try in Fedora then? maybe make some preview images for people to try with? (edited)
21:49fdobridge: <karolherbst🐧🦀> maybe? my biggest concern is mostly that if we enable it for everybody the sheer amount of bugs could overwhelm us. I was more thinking of doing something vendor specific. Like ask Lenovo if they are okay if we enable it only for all Lenovo laptops or something like that
21:49fdobridge: <karolherbst🐧🦀> and then go from there
21:50fdobridge: <karolherbst🐧🦀> or just enable it for everybody and deal with the bugs
21:56fdobridge: <airlied> No point rushing on this
21:57fdobridge: <airlied> When it goes upstream I think there will be a lot of issues to solve but with GSP and with older devices that regress due to refractors needed.to support GSP
21:57fdobridge: <karolherbst🐧🦀> yeah...
21:57fdobridge: <karolherbst🐧🦀> I think we should land everything in batches