05:38fdobridge: <airlied> Running dEQP on 16 threads in 484-test groups
05:38fdobridge: <airlied> still no luck
05:50fdobridge: <airlied> @karolherbst is this one of those netconsole type situations? I'm am occasionally seeing a dead laptop
06:53fdobridge: <airlied> also what is latest kernel you have reproduced the kasan on? 6.3.1?
08:46fdobridge: <karolherbst🐧🦀> master/drm-misc-next/random commits since the one breaking it.
08:47fdobridge: <karolherbst🐧🦀> yeah, the machine just dies
08:47fdobridge: <karolherbst🐧🦀> it crashes within kmalloc, soo
08:47fdobridge: <karolherbst🐧🦀> I'm using a serial console attached to my desktop
08:50fdobridge: <karolherbst🐧🦀> sometimes i had to wait 2 hours to trigger it. It did take me a week to bisect for a reason
08:53fdobridge: <karolherbst🐧🦀> I'm sure I also tried on stable/linux-6.3.y to trgger it
09:08fdobridge: <airlied> Okay I'll see if my netconsole efforts pay off
09:13fdobridge: <karolherbst🐧🦀> hopefully 🙂
09:14fdobridge: <karolherbst🐧🦀> not sure how kmalloc free netconsole is
09:14fdobridge: <karolherbst🐧🦀> but if your machine dies you probably hit the bug as other issues shouldn't be near as fatal as this one
19:50fdobridge: <gfxstrand> Well, NAK now builds with stable rust again. \o/
19:50fdobridge: <gfxstrand> (Rust 1.17.0 stabilized `std::sync::OnceLock`)
19:51fdobridge: <karolherbst🐧🦀> funky 😛
19:54fdobridge: <gfxstrand> I'm using it for parsing debug env vars. Probably not strictly necessary but it's useful.
19:54fdobridge: <karolherbst🐧🦀> same
19:54fdobridge: <karolherbst🐧🦀> but I have a dirty workaround
19:55fdobridge: <gfxstrand> Yeah, I thought about hand-rolling it but it was about to go stable so why bother
19:55fdobridge: <karolherbst🐧🦀> https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/gallium/frontends/rusticl/core/platform.rs#L51-114
19:55fdobridge: <Samantas5855> I tend to forget nak is rust
19:55fdobridge: <karolherbst🐧🦀> yeah, but it's also not much code
19:56fdobridge: <karolherbst🐧🦀> at least I'm trying to play a little nice to distributions 😄
19:57fdobridge: <karolherbst🐧🦀> I'm sure if we go newest rust all the time, we'll get our inbox spamed with people complaining
19:57fdobridge: <karolherbst🐧🦀> so my rough goal is to follow whatever firefox esr and/or linux require
20:03fdobridge: <lingm> btw, which distros ship firefox esr but not normal firefox? other than debian and its close derivatives
20:04fdobridge: <karolherbst🐧🦀> debian
20:04fdobridge: <karolherbst🐧🦀> and uhm...
20:04fdobridge: <karolherbst🐧🦀> RHEL
20:04fdobridge: <karolherbst🐧🦀> I think?
20:04fdobridge: <karolherbst🐧🦀> anything enterprise probably
20:06fdobridge: <lingm> and these enterprise distros are going to ship a modern mesa?
20:06fdobridge: <karolherbst🐧🦀> yes
20:06fdobridge: <karolherbst🐧🦀> RHEL updates mesa constantly
20:07fdobridge: <karolherbst🐧🦀> because hardware enablement
20:10fdobridge: <karolherbst🐧🦀> but generally it's a distributions problem
20:10fdobridge: <gfxstrand> I already know what lengths RHEL goes to in order to update Mesa. Grabbing a fresh rustc isn't that big of a deal, probably.
20:10fdobridge: <gfxstrand> That too
20:10fdobridge: <karolherbst🐧🦀> yeah.. I'm not too worried about it, I just also don't want to require the newest rustc on every release
20:10fdobridge: <karolherbst🐧🦀> whatever firefox esr needs kinda feels like the proper middleground between annoying distributions and not being held down too much
20:11fdobridge: <karolherbst🐧🦀> the next ESR release will use 1.66.0
20:12fdobridge: <karolherbst🐧🦀> and the current one is 1.59
20:12fdobridge: <karolherbst🐧🦀> and mesa is already at 1.60 anyway 🙃
20:26fdobridge: <lingm> @karolherbst there's RHEL's rust policy. anyone can find out for sure if the mesa policy is the same?
20:26fdobridge: <lingm> https://cdn.discordapp.com/attachments/1034184951790305330/1118637564870541353/Screenshot_from_2023-06-14_22-24-14.png
20:37fdobridge: <karolherbst🐧🦀> sounds like RHEL uses whatever was newest on branch point
20:38fdobridge: <lingm> yeah, 1.66 came out ~5 months before RHEL 9.2
20:38fdobridge: <karolherbst🐧🦀> which is roughly when 9.2 gets branched out 😄
20:38fdobridge: <karolherbst🐧🦀> which is roughly when 9.2 got branched out 😄 (edited)
20:40fdobridge: <karolherbst🐧🦀> but anyway
20:40fdobridge: <karolherbst🐧🦀> that sounds like "next FireFox ESR" as well 😛
20:40fdobridge: <karolherbst🐧🦀> I could actually figure out what the precise rule is
20:40fdobridge: <lingm> precise rule for what?
20:40fdobridge: <karolherbst🐧🦀> updating rustc
20:41fdobridge: <karolherbst🐧🦀> but anyway
20:41fdobridge: <karolherbst🐧🦀> older versions don't get new hardware support in RHEL
20:41fdobridge: <karolherbst🐧🦀> it's all on y stream only
20:42fdobridge: <karolherbst🐧🦀> rhel 9.3 will target 23.2? probably, but anyway, I don't think RHEL will be an issue
20:42fdobridge: <karolherbst🐧🦀> just all the debian based distros will shout at us
20:43fdobridge: <karolherbst🐧🦀> it's not really the distros itself, just people backporting newest mesa to old bases
20:55fdobridge: <lingm> `rustup update` 🙃 (anything ubuntu based ships normal firefox btw. i'm only aware of debian and raspbian shipping only esr. although i'm not well versed in the debian universe)
20:58fdobridge: <karolherbst🐧🦀> yeah..... I mean we'll probably update faster than ESR in some cases if it makes sense, but I also assume we'll settle, because some OS are just pure pain
20:59fdobridge: <karolherbst🐧🦀> @lingm btw, what's the way to enforce using unsafe blocks in unsafe functions?
21:01fdobridge: <lingm> putting `#[warn(unsafe_op_in_unsafe_fn)]` in the crate root should show the warning. you can turn it into a `deny` once the existing cases are fixed
21:01fdobridge: <karolherbst🐧🦀> ahh good idea, will clean that up later I guess
21:01fdobridge: <lingm> you can probably pass it on the command line somehow too
21:02fdobridge: <karolherbst🐧🦀> yeah
21:03fdobridge: <lingm> most people put it in the code though. just don't do `#![deny(warnings)]` or you'll get pain on compiler updates
21:03fdobridge: <lingm> putting `#![warn(unsafe_op_in_unsafe_fn)]` in the crate root should show the warning. you can turn it into a `deny` once the existing cases are fixed (edited)
21:15fdobridge: <karolherbst🐧🦀> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23660
21:25fdobridge: <marysaka> soon `undocumented_unsafe_blocks` enabled? :aki_peek:
21:26fdobridge: <karolherbst🐧🦀> uhhhh..... too much work for now 😄 at least I want to require it for changed code... I should add more docs there as well
21:28fdobridge: <marysaka> yeah makes sense, maybe I could look a bit and help on that side when I'm done with my falcon stuffs 😄
21:29fdobridge: <karolherbst🐧🦀> @lingm I actually don't know what tag to add for your rb 😄 maybe I just leave it out, dunno It's pointless anyway 🙃
21:29fdobridge: <karolherbst🐧🦀> still busy with falcon stuff? 😄
21:30fdobridge: <Samantas5855> is falcon the riscv chip?
21:31fdobridge: <karolherbst🐧🦀> no, that's pre riscv
21:31fdobridge: <Samantas5855> so its the one pre gsp?
21:31fdobridge: <karolherbst🐧🦀> yeah
21:31fdobridge: <karolherbst🐧🦀> well.. "one"
21:31fdobridge: <Samantas5855> on kepler, maxwell.pascal, volta
21:31fdobridge: <karolherbst🐧🦀> I think they have 7 versions of it?
21:31fdobridge: <marysaka> haven't had much time to poke at it in the past weeks as I was travelling a bit (and busy wrapping stuffs at work before leaving)
21:31fdobridge: <marysaka> But yeah I kind of got an half baked nouveau with only DEVINIT in place, will need to see if it actually work with my testing
21:31fdobridge: <Samantas5855> 7?
21:31fdobridge: <karolherbst🐧🦀> yeah
21:31fdobridge: <karolherbst🐧🦀> or 6
21:31fdobridge: <marysaka> I think it's 6 no?
21:31fdobridge: <karolherbst🐧🦀> I am sure there are at least 5
21:31fdobridge: <marysaka> oh you are accounting for v0
21:32fdobridge: <karolherbst🐧🦀> I don't know.. there are many 🙃
21:32fdobridge: <Samantas5855> kepler,maxwell a, b, pascal, volta
21:32fdobridge: <Samantas5855> which one is the sixth
21:32fdobridge: <marysaka> no we are talking about Falcon versions not GPU arch
21:32fdobridge: <karolherbst🐧🦀> falcon is older than that
21:33fdobridge: <Samantas5855> I thought every gpu arch had its own falcon
21:33fdobridge: <karolherbst🐧🦀> I don't know which arch came wiht falcon first 😄
21:33fdobridge: <Samantas5855> cause it restricted us more with each gen
21:33fdobridge: <airlied> gsp is both a falcon and a riscv
21:33fdobridge: <airlied> because why not
21:33fdobridge: <karolherbst🐧🦀> uhhh
21:34fdobridge: <karolherbst🐧🦀> so falcon boots the riscv?
21:34fdobridge: <karolherbst🐧🦀> or does it depend on the subsystem?
21:34fdobridge: <airlied> there's a mode switch
21:34fdobridge: <marysaka> ... does it still have the old DMA engine?
21:34fdobridge: <karolherbst🐧🦀> ahh yeah...
21:34fdobridge: <karolherbst🐧🦀> falcons ISA is cursed
21:34fdobridge: <marysaka> yeah it's quite a mess...
21:35fdobridge: <Samantas5855> if maxwell b and pascal become usable on nouveau it would be a big w
21:36fdobridge: <Samantas5855> im guessing you guys are trying to exploit the falcon chip on these gens?
21:36fdobridge: <karolherbst🐧🦀> and who is finding the time to write all the reclocking code?
21:37fdobridge: <marysaka> Probably not me :vReiAgony:
21:37fdobridge: <Samantas5855> Vitaliy Kuzmin in 12 years probably 🐸
21:37fdobridge: <karolherbst🐧🦀> probably
21:38fdobridge: <karolherbst🐧🦀> it's a massive amount of work sadly
21:38fdobridge: <karolherbst🐧🦀> I wished more would be willing to actually invest into nouveau to make it not suck so much
21:38fdobridge: <Samantas5855> I wonder why does the nvidia os driver not support non gsp gpus
21:38fdobridge: <Samantas5855> while their prop driver still supports them
21:38fdobridge: <karolherbst🐧🦀> because they want to hide stuff in gsp
21:39fdobridge: <karolherbst🐧🦀> gsp contains all the code they didn't want to open source
21:39fdobridge: <karolherbst🐧🦀> it's really that simple
21:39fdobridge: <marysaka> Not to mention that would leak HDCP interfaces with the falcon for example
21:39fdobridge: <Samantas5855> they could still offer the binaries you guys need for pmu tho right?
21:39fdobridge: <karolherbst🐧🦀> well they could but won't
21:39fdobridge: <Samantas5855> yeah that's what I'm asking
21:40fdobridge: <marysaka> tbh I wish they would do that at least at EOL of their devices :painpeko:
21:40fdobridge: <Samantas5855> isnt it possible to dump them from the driver package?
21:41fdobridge: <karolherbst🐧🦀> sure, but how would you distribute them?
21:41fdobridge: <lingm> yeah, those tags are nonsense busywork imo. leave it out.
21:41fdobridge: <Samantas5855> you don't
21:41fdobridge: <karolherbst🐧🦀> yeah, then it's useless
21:42fdobridge: <Samantas5855> ask the user to provide them or make the distro extract them at install time
21:42fdobridge: <Samantas5855> android x86 used to do that with houdini iirc
21:42fdobridge: <marysaka> That's not viable
21:43fdobridge: <karolherbst🐧🦀> in theory it _could_ be done, but then we also have to reverse engineer all the interfaces
21:43fdobridge: <Samantas5855> doesnt that already happen if you want vaapi with nouveau?
21:43fdobridge: <karolherbst🐧🦀> wouldn't be all that bad for EOL hardware where we just pick the newest and call it a day
21:43fdobridge: <karolherbst🐧🦀> yeah, and almost nobody uses it
21:43fdobridge: <Samantas5855> I did 🐸
21:43fdobridge: <karolherbst🐧🦀> did it actually work?
21:43fdobridge: <Samantas5855> it had an aur package
21:44fdobridge: <Samantas5855> it did with mpv but the video was a bit broken iirc
21:44fdobridge: <karolherbst🐧🦀> yeah.. figures
21:44fdobridge: <karolherbst🐧🦀> seeing how many bugs we get on that stuff (like 1 in a year) I can safely say that almost nobody actually uses it 😄
21:45fdobridge: <Samantas5855> do people actually use nouveau as a daily driver
21:45fdobridge: <Samantas5855> I use it for a gt710
21:45fdobridge: <Samantas5855> cause I only have that pc for browsing
21:45fdobridge: <karolherbst🐧🦀> yeah, a lot actually do
21:45fdobridge: <Samantas5855> but thata a very niche usecase of an external gpu
21:45fdobridge: <Samantas5855> but thats a very niche usecase of an external gpu (edited)
21:45fdobridge: <karolherbst🐧🦀> it's good enough as long as you don't run anything demanding
21:46fdobridge: <Samantas5855> as long as it can run wayland
21:46fdobridge: <karolherbst🐧🦀> it gets better with each generation actually...
21:46fdobridge: <karolherbst🐧🦀> the older it gets the more broken it is
21:46fdobridge: <Samantas5855> thats a good thing
21:46fdobridge: <karolherbst🐧🦀> but anyway
21:46fdobridge: <Samantas5855> for the future
21:46fdobridge: <karolherbst🐧🦀> we don't really have the people to actually fix shit
21:47fdobridge: <Samantas5855> but the gt710 is fine
21:48fdobridge: <Samantas5855> I thought more people started working on nouveau after the open modules dropped
21:49fdobridge: <karolherbst🐧🦀> ehhh no
21:50fdobridge: <karolherbst🐧🦀> nvk is the only part where more people showed up, but that's more because reclocking comes through gsp
21:50fdobridge: <karolherbst🐧🦀> but that still doesn't fix the situation with the kernel driver being an unstable mess on a lot of GPUs if you just wiggle it hard enough
21:50fdobridge: <Samantas5855> would it be a better idea to create a new one
21:50fdobridge: <Samantas5855> by forking the open kernel modules
21:52fdobridge: <karolherbst🐧🦀> open kernel modules is nowhere near getting upstreamed and it requires GSP
21:52fdobridge: <karolherbst🐧🦀> doesn't do anything for the older GPUs we support
21:52fdobridge: <karolherbst🐧🦀> and the code quality is crap
21:53fdobridge: <karolherbst🐧🦀> it would take less time to just write a new driver
21:53fdobridge: <Samantas5855> I didnt know about the code quality
21:53fdobridge: <Samantas5855> haven't delved into the code
21:56fdobridge: <karolherbst🐧🦀> anyway, be it as it may, it won't address the actual problem: not having enough people to work on it
21:58fdobridge: <airlied> yeah someone should write a new gsp only driver in rust 😛
21:59fdobridge: <karolherbst🐧🦀> mhhh
21:59fdobridge: <airlied> and make it cuda compatible 😛
22:00fdobridge: <karolherbst🐧🦀> I know for sure that I won't be the person being in the mood for doing so 🙃
22:01fdobridge: <airlied> but you love 🦀
22:02fdobridge: <karolherbst🐧🦀> sure, but I also know what level of pain I would get into here and I know nobody cares enough to do actual kernel level funding for nouveau
22:02fdobridge: <karolherbst🐧🦀> or proper funding
22:03fdobridge: <karolherbst🐧🦀> so opting in to do that, would also just means I burn me out on purpose
22:07fdobridge: <karolherbst🐧🦀> if somebody finds us 5 more people to do it, then I might reconsider
23:05fdobridge: <esdrastarsis> Is there a good getting started documentation for nouveau-drm for newbies?
23:05fdobridge: <karolherbst🐧🦀> no
23:06fdobridge: <karolherbst🐧🦀> I advice fixing bugs you run into