08:29 ellagay[d]: ./
10:40 pac85[d]: ellagay[d]: `./: ./: Is a directory`
12:29 karolherbst[d]: I got reminded today, that we have a few users with GPUs where the initial clock config (voltage, clock, etc..) is invalid... and leads to random crashes 🥲
12:29 karolherbst[d]: which can be fixed by reclocking the GPU, which has bugs, so...
20:55 _lyude[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1460376619519840550/image.png?ex=6966b136&is=69655fb6&hm=94cf53a01a69fe9f4b46e0f41924cb16a7adc6a9dd7d14ab630400d30733b787&
20:55 _lyude[d]: is the artifacting on the icons here the same as the other checkerboard-looking graphical bug I ran into with nouveau the other day? https://gitlab.freedesktop.org/mesa/mesa/-/issues/13909 this one specifically
21:05 mhenning[d]: Does it still happen on main? The bug you linked has been fixed.
21:06 _lyude[d]: mh - would be easy to test if I knew how to override graphics drivers in flatpak :V
21:07 mhenning[d]: oh, right that's a headache
21:08 _lyude[d]: i have a lot of applications on here under flatpak that are having graphics issues unfortunately
21:11 mangodev[d]: _lyude[d]: interesting
21:11 mangodev[d]: i get similar-ish artifacts in firefox when `backdrop-filter: blur` is used on a page
21:12 mangodev[d]: although the checkerboard extends to the entire page, not just the element
21:12 mhenning[d]: From the image alone it's hard to tell if it's the same bug or not
21:12 _lyude[d]: it seems like I really need to focus on just getting it so I can run flatpak stuff with abritrary driver versions
21:12 mangodev[d]: yeah, but it gives me similar vibes
21:12 mangodev[d]: i think the checkerboard direction is even the same
21:12 _lyude[d]: because out of the stuff I run: I think I only have two applications (firefox and nvim-gtk) that aren't running through flatpak
21:13 mangodev[d]: i have gotten weird graphical glitches on the server list gradient on servers with a banner image
21:14 airlied[d]: firefox with large youtube vids gives me the odd black box flicker, but I need to retest
21:16 mhenning[d]: _lyude[d]: I managed to get that working using https://gitlab.com/freedesktop-sdk/mesa-git-extension in the past although I forget exactly what I did
21:19 mangodev[d]: airlied[d]: yes
21:19 mangodev[d]: especially bad in theater mode
21:19 mangodev[d]: although YouTube generally flickers for me no matter where i am because their newer site layout has a blurred top bar and buttons
21:19 mangodev[d]: i do need to try more browsers, but i'm concerned that most won't work
21:20 mangodev[d]: tried Vivaldi, hard crashes when a single image is loaded (same width fault that discord intermittently runs into)
21:20 mhenning[d]: I haven't managed to reproduce that one with youtube, but the background video on https://www.qt.io/ triggers it for me
21:20 _lyude[d]: ooooo neat. I've got full screen blinking now
21:20 mangodev[d]: google chrome works mostly fine though?
21:21 mangodev[d]: mhenning[d]: weird
21:21 mangodev[d]: sometimes flickers black, sometimes flickers white?
21:22 mangodev[d]: oooh hover over "Development and Design"
21:22 mhenning[d]: mangodev[d]: Does that one reproduce easily? I've never managed to reproduce your discord issues
21:22 airlied[d]: I assume this is all zink on nvk
21:22 mangodev[d]: mangodev[d]: the Get Started button flickers in and out, mostly out
21:23 mangodev[d]: mhenning[d]: has happened any time i load a site with images
21:23 mangodev[d]: YouTube instantly crashes
21:23 mhenning[d]: airlied[d]: I believe so yes
21:23 mangodev[d]: airlied[d]: i've had width faults happen even with vulkan mode forced on
21:23 _lyude[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1460383724305776722/PXL_20260112_212248597.mp4?ex=6966b7d4&is=69656654&hm=195d322f9ccab00596442f5d1dcf385cada8b0d6b13ba97c92bc1cd136296385&
21:23 _lyude[d]: New bug dropped
21:24 mangodev[d]: _lyude[d]: i'm lucky my setup's not that bad 😭
21:24 _lyude[d]: to b honest I WAIT it's showing me the framebuffer for the right-most monitor
21:24 _lyude[d]: oh this is weird
21:24 _lyude[d]: mangodev[d]: I have no idea what on earth triggers this, this is the first time I actually had a camera ready
21:25 _lyude[d]: oh it moved to the right monitor
21:25 mangodev[d]: i will say that the only bug i've had with chrome is that it can randomly crash on the second time i sign into it on a session
21:25 mangodev[d]: specifically the second time
21:25 mangodev[d]: although i haven't been able to replicate it enough to check logs and see why
21:25 mangodev[d]: it just
21:25 mangodev[d]: *closes*
21:26 mhenning[d]: mangodev[d]: nothing in dmesg?
21:26 mangodev[d]: mangodev[d]: .
21:26 _lyude[d]: airlied[d]: any chance you've got any experience with testing driver bugs through flatpak? I found https://blog.sebastianwick.net/posts/flatpak-graphics-drivers/ but this is, a lot more involved then I was hoping 🙁
21:29 airlied[d]: no you normally have to hack the flatpak somehow, it's always painful
21:29 _lyude[d]: ouch
21:30 _lyude[d]: Hm. I see they're able to make runtime extensions for specific nvidia versions…
21:30 mangodev[d]: _lyude[d]: and stuff like this is why i still like nvk better than prop…
21:34 mhenning[d]: _lyude[d]: again, https://gitlab.com/freedesktop-sdk/mesa-git-extension is probably the way to do it for nvk
21:34 _lyude[d]: oh! sorry mhenning[d] I didn't see your message the first time
21:35 _lyude[d]: that's awesome, should let me actually figure out what's going on here finally
21:36 jannau: _lyude[d]: for asahi we used https://pagure.io/fedora-asahi/mesa-asahi-flatpak/tree/24.08 and https://pagure.io/fedora-asahi/fdo-sdk-flatpak/tree/24.08 to build a host extension with the non-upstream driver (mesa-git is probably easier)
21:43 _lyude[d]: btrw if people are curious: I didn't end up managing to figure out the suspend/resume issue on my RTX6000, it looks like it's not any obvious difference with the firmware layout (after looking closer I realized we're almost identical in terms of how we're laying stuff out for the fw, and pretty much all the codepaths I was finding in openrm don't actually seem to apply for this card).
21:44 airlied[d]: so just sometimes resume fails?
21:44 _lyude[d]: though it's hard to tell if older versions of the driver that match the firmware version might do something, since I haven't come up with a nice way of either being able to run nvidia's driver headless (so I don't need the userspace drivers for the same nvidia driver version installed) or figuring out a way to setup an arbitrary nvidia driver version that doesn't pollute my system
21:45 _lyude[d]: airlied[d]: No it always fails, it almost comes back with r535 but seems to hit a different looking there
21:46 _lyude[d]: R570 seems to get stuck on the initial suspend step, which we don't notice in nouveau - supposedly due to some non-WPR region not having enough space. Then as a result resume doesn't actually wok and the GPU dies
21:47 _lyude[d]: (the WPR-region bit coming from what notthatclippy[d] was able to decode from the GSP logs)
21:47 airlied: I think you can just build openrm and load the drm module with modeset=1 fbdev=1 orsomething
21:47 airlied[d]: without having the userspace
21:48 _lyude[d]: oh! that would totally solve the problem then, so long as it actually loads the firmware and such
21:48 airlied[d]: oh you need to put the firmware somewhere alright
21:48 airlied[d]: I've extracted the run file locally and copied the fw into the right place in the past
21:48 airlied[d]: so I don't get all the userspace bits
21:49 _lyude[d]: yeah - i've just been installing the rpm, blacklisting nvidia at boot, loading into multi-user and then inserting the driver here
21:50 _lyude[d]: airlied[d]: you said you had a RTX6000 packed away somewhere right?
21:51 airlied[d]: I'll see if I can get into a box this week
21:55 _lyude[d]: yeah - would be good to know if it behaves any differently. especially because after reading some of the gsp loading code in openrm I'm slightly suspicious it's actually possible there are regions of memory being reserved by the vbios itself that might differ.
21:55 _lyude[d]: but i'm mostly curious what it's going to just look like with r570 now that I have a way of testing that 🙂