09:20daxe: hi I'm new to IRC, I was wondering if this is the right place to ask troubleshooting/support questions about nouveau/mesa?
09:21RSpliet: If you're running an upstream/new enough kernel and userspace then yes. Otherwise the first advice is going to be "try something newer"
09:21RSpliet: daxe: ^
09:21daxe: I'm on gentoo with 6.4.1
09:21daxe: kernel that is
09:22RSpliet: That counts as new enough. Same ofc goes for mesa. But otherwise don't ask to ask, just ask!
09:22RSpliet: (and someone else might know the answer :D)
09:24DodoGTA: daxe: What problem do you have?
09:24daxe: I'm trying to run a game client (RuneLite) but when I turn on the GPU acceleration it causes a crash and I get a kernel warning in dmesg.
09:24daxe: I just realised it's a bad time to ask because the game is down for an hour
09:27DodoGTA: daxe: What kernel warning do you get?
09:27daxe: I'll upload to pastebin or similar one sec
09:28daxe: stdout from game client: https://pastebin.com/kTBLhL7v. dmesg kernel warning: https://pastebin.com/JZVhhpkB
09:30DodoGTA: daxe: Does the same happen with glxgears?
09:30daxe: no that runs fine
09:31daxe: not sure if it makes a difference but I'm using wayland (sway)
09:32DodoGTA: Turing should have working 3D acceleration so that's bizzare (did you install the required NVIDIA firmware?)
09:33daxe: I installed the nvidia-firmware package through the package manager
09:33daxe: is there a way to check if that's installed correctly?
09:35daxe: I've got the /lib/firmware/nouveau direcotry with files in it which I think is what the package installs
09:36DodoGTA: daxe: That's the wrong firmware (I meant the nvidia stuff in linux-firmware)
09:37daxe: gotcha, I've got /lib/firmware/nvidia/tu102/ is that the one I need?
09:37DodoGTA: daxe: That should be the right folder
09:38DodoGTA: Do you get anything in dmesg before the dmesg kernel warning?
09:40daxe: https://pastebin.com/pMvWK3a5
10:46karolherbst: looks like you simply run out of VRAM
10:47karolherbst: which is odd...
10:48karolherbst: daxe: I think we wired up the memory consumption reporting. Something with GALLIUM_HUD or other tools displaying current VRAM usage could help to see if it's really that or something buggy
10:49HdkR: karolherbst: Going to wire up nouveau memory consumption to fdinfo so that nvtop can use it?
10:49HdkR: :)
10:54DodoGTA: karolherbst: In what kernel version will it be available?
10:55karolherbst: it's a userspace thing
10:56DodoGTA: karolherbst: So it's a part of Mesa or libdrm then?
10:56karolherbst: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19405
10:56karolherbst: ehh wait, wrong one
10:56karolherbst: huh.. I thought someobdy wired it up though...
10:57karolherbst: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17931
10:57karolherbst: this one!
11:00daxe: you want me to build and test with that branch?
11:00daxe: I can try
11:03DodoGTA: karolherbst: I think this only exposes the information through a GL extension (sysfs node or fdinfo would be nice)
11:07karolherbst: daxe: it doesn't fix anything, it just allows you to monitor VRAM usage
11:08daxe: ok, will see if I can get that working
11:22daxe: if I'm using mesa 23.1.3 that change should already be merged in? I can't seem to get tools to display VRAM usage
11:22daxe: I tried GALLIUM_HUD="VRAM-usage" glxgears and got "gallium_hud: unknown driver query 'VRAM-usage'"
11:23daxe: sorry if I misunderstood the instructions
11:24karolherbst: daxe: but it does show up in `glxinfo` as "Total available memory", right?
11:25daxe: yes that shows the total amount but not how much is in use
11:25daxe: oh just under that it shows it
11:25daxe: Currently available dedicated video memory: 8969 MB
11:26karolherbst: mhh.. let me see if something is wrong and why it doens't show in gallium hud...
11:27karolherbst: oh, looks like there is no query for it
11:28daxe: I am just using this for now: watch -n 0.1 "glxinfo | grep 'Currently available'"
11:28daxe: tried to run the program but it caused my whole display to freeze, I'll try it again but run the watch command over ssh
11:29karolherbst: how is this HUD called everybody uses?
11:30karolherbst: mangohud
11:31daxe: I tried that one it didn't show VRAM usage unfortunately
11:31daxe: I don't think it will help because my display freezes completely so I can't even see the HUD
11:31daxe: display freezes until the application crashes
11:32karolherbst: heh.. it indeed doesn't show VRAM...
11:32karolherbst: curious on what's missing
11:32karolherbst: ohhh...
11:32karolherbst: mangohud only supports this on the amdgpu driver...
11:32karolherbst: that's a mangohud bug then :)
11:34karolherbst: yeah.. it's implemented in the weirdest way
11:35karolherbst: daxe: ohh, it's more interesting if you get close to 0 or not
11:37DodoGTA: karolherbst: I already have temperature support hooked up for nouveau in MangoHUD
11:38karolherbst: DodoGTA: Sure, but what about using standard APIs for memory consumption/info?
11:38karolherbst: the GL/Vk extensions exist for that
11:40daxe: it doesn't seem like the memory consumption increases when the crash happens
11:40karolherbst: but for the temperature I _hope_ MangoHUD just uses the hwmon interfaces
11:40karolherbst: daxe: okay... it might be something wrong with the check on the kernel side then
11:40karolherbst: maybe the check only works up to 32 bit values or something silly :D
11:40karolherbst: let me check..
11:41daxe: I have to step away for a bit but I'll be back later
11:45karolherbst: daxe: it might actually be that the GPU is either too slow or is indeed stuck processing something mhh
11:45karolherbst: but can also be related to the freeze actually
11:47karolherbst: mhhh
11:47karolherbst: "fifo: fault 00 [VIRT_READ] at 0000000009da3000 engine 40 [gr] client 09 [GPC2/PE_1] reason 02 [PTE] on channel 9 [02bf35f000 Discord[9017]]"
11:49karolherbst: but yeah..I suspect the GPU acceleration might just trip some weirdo bug
11:50DodoGTA: karolherbst: MangoHUD does use hwmon on AMD GPUs
11:53DodoGTA: And NVML on the proprietary NVIDIA driver (because the temperature isn't probably exposed through hwmon there)
11:54karolherbst: yeah, for nvidia it makes total sense. I'm just curious on why not GL/Vk are used to retrieve the vram infos?
11:58DodoGTA: karolherbst: Maybe because a graphics API-independent solution is more reliable? 🤷♀️️
12:16DodoGTA: karolherbst: nouveau driver is missing pipe queries for memory usage (which are needed by Gallium HUD)
12:26daxe: I figured the discord log was unrelated
12:27DodoGTA: daxe: We do actually have a secret Discord server
12:36karolherbst: DodoGTA: which queries? I couldn't find anything off hand
12:44DodoGTA: karolherbst: These are driver-dependent, but here's radeonsi: https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/gallium/drivers/radeonsi/si_query.c#L1752
12:52DodoGTA: Also the memory usage returned by glxinfo seems broken