12:54OftenTimeConsuming: Okay, I found a reproducible crash in speed-dreams-2
12:54OftenTimeConsuming: GALLIUM_HUD=".w550.h280.c70fps:70,.w550.h280.c80temperature:80" NOUVEAU_LIBDRM_DEBUG=2 speed-dreams-2
12:56OftenTimeConsuming: Road Tracks -> E-Track 4 -> Only player with Diamond R25 -> 2 laps -> noon time of day -> no clouds -> no rain
12:58OftenTimeConsuming: Setup should look like this; https://0x0.st/XYas.webp
13:00OftenTimeConsuming: You need to drive the first lap very fast, reaching 398 km/hr and then on the downhill of the second lap, as soon as you reach 397 km/hr, libgallium crashes. Unfortunately the error is masked with NOUVEAU_LIBDRM_DEBUG=2 and the game is unplayable with NOUVEAU_LIBDRM_DEBUG=1.
13:01OftenTimeConsuming: The game does have automatic players, but they suck at driving and keep crashing into the walls and thus fail to reach 397 km/hr.
13:02karolherbst[d]: is there no benchmark mode or anything? Because as I said earlier, if there are _any_ manual steps involved, it's less likely for anybody to look into it
13:02karolherbst[d]: though...
13:02karolherbst[d]: maybe a trace could trigger it as well
13:02karolherbst[d]: with apitrace or something else, not sure if we got a replacement for apitrace these days
13:03OftenTimeConsuming: I'll see if there's a player that doesn't suck at driving.
13:12OftenTimeConsuming: Valentino Fontana from Monoposto is fast enough to trigger the bug it seems.
13:13OftenTimeConsuming: I'll see if it trigger with; NOUVEAU_LIBDRM_DEBUG=1
13:13OftenTimeConsuming: *Monoposto 1
13:17OftenTimeConsuming: It does not trigger with 1, as the game runs too slow.
13:36OftenTimeConsuming: Got one with road Tracks -> Street 1 -> NOUVEAU_LIBDRM_DEBUG=1 Valentino Fontana
13:37karolherbst[d]: does it trigger with 2?
13:37karolherbst[d]: because that would be a bit more useful or maybe even 3 (which does both)
13:37OftenTimeConsuming: It triggers with 2 also.
13:39OftenTimeConsuming: But has no error output.
13:39OftenTimeConsuming: DEBUG=1; https://0x0.st/XYaL.txt.xz
13:40karolherbst[d]: OftenTimeConsuming: huh.. that's weird
13:40karolherbst[d]: did that broke...
13:42karolherbst[d]: OftenTimeConsuming: what's like the actual reason this one breaks?
13:43OftenTimeConsuming: No idea. It just crashes at runtime at seemingly random.
13:44karolherbst[d]: that's probably a different problem needing different debugging then
13:44OftenTimeConsuming: The crash is always in libgallium/
13:44karolherbst[d]: sure
13:44karolherbst[d]: but it always depends on the type of crash
13:45karolherbst[d]: like what's the stacktrace
13:46OftenTimeConsuming: https://termbin.com/0qxp
13:47karolherbst[d]: so the `INVALID_BITFIELD` thing should dump something with `=2`, but for the others it could be random stuff going on, so yeah, I'd need like the proper stacktraces
13:53OftenTimeConsuming: I'm hoping it'll crash with =3, but it seems to not want to anymore.
13:58karolherbst[d]: right, but if it's not always causing a GPU error, but just a crash, it's likely it's just some CPU code being wrong, in which case the dump won't really help that much, because it could just a memory corruption. Running with libasan _could_ also show something up, maybe it's a race, it's kinda hard to tell
14:18OftenTimeConsuming: Better compile it with asan then.
14:23OftenTimeConsuming: Of course it fails to compile, but LeakSanitizer found memory leak while compiling
14:23OftenTimeConsuming: *as
14:31karolherbst[d]: yeah.. that's kinda normal, don't worry about it
14:42OftenTimeConsuming: It's a memory error in libgallium - compiling speed-dreams-2 with asan stops the crash from occuring.
14:42OftenTimeConsuming: *occurring
14:47karolherbst[d]: oh no
14:47karolherbst[d]: I hate those
14:48karolherbst[d]: could be some data race though, libasan is quite at hiding those
19:18demonkingofsalvation[d]: I am back to once again request your help, reinstalled my os to ensure that I got rid of anything stopping nouveau from working but still got issues
19:19demonkingofsalvation[d]: my gpu is the NVIDIA GeForce RTX 3050 Ti Mobile, I asked earlier in the year and the gsp wasn't working or something
19:22demonkingofsalvation[d]: journaltctl:
19:22demonkingofsalvation[d]: https://pastebin.com/kHEL7kAg
19:23demonkingofsalvation[d]: dmesg:
19:23demonkingofsalvation[d]: https://pastebin.com/xPJ2PCgV
19:34karolherbst[d]: demonkingofsalvation[d]: does that happen with GSP disabled?
19:36demonkingofsalvation[d]: karolherbst[d]: this correct:
19:36demonkingofsalvation[d]: nouveau.config=NvGspRm=0
19:36karolherbst[d]: yeah
19:36demonkingofsalvation[d]: let me restart
19:40demonkingofsalvation[d]: nah seems to fix it, I'll send the pastebins
19:42demonkingofsalvation[d]: journalctl: https://pastebin.com/0M9yY2ni
19:43demonkingofsalvation[d]: karolherbst[d]: dmesg: https://pastebin.com/3HAusJ6a
19:44demonkingofsalvation[d]: I'm on a laptop if that matters
19:51karolherbst[d]: mhhhh
19:51karolherbst[d]: could be runtime pm, but like...
19:51karolherbst[d]: `nouveau.runpm=0` could fix it with GSP, worth a try
19:51karolherbst[d]: but then you'd probably want to run with GSP disabled instead, because runpm=0 will kill your battery runtime
19:52demonkingofsalvation[d]: I'll test it to see if it works, then decide afterwards
19:52karolherbst[d]: could also reboot depending on what you want to do 🙃
19:53demonkingofsalvation[d]: karolherbst[d]: as in it'll crash my system or rebuild grub?
19:54karolherbst[d]: oh no, I mean if you need perf you can enable gsp again and reboot
19:54demonkingofsalvation[d]: ah like that
19:54demonkingofsalvation[d]: well, back to restarting
19:58demonkingofsalvation[d]: seems like it worked, will send logs
19:59demonkingofsalvation[d]: journal: https://pastebin.com/SErqkG37
20:00karolherbst[d]: yeah, so I guess something goes wrong waking up the GPU
20:00karolherbst[d]: not sure if airlied[d] is aware of any patches still pending or not
20:00demonkingofsalvation[d]: dmesg:https://pastebin.com/mM2b6WJn
20:01demonkingofsalvation[d]: do I have to create an issue on the gitlab repo?
20:01karolherbst[d]: but yeah... guess you can use GSP once you need more performance, but normally you really should run with runpm=1 enabled.. though runpm=0 isn't too bad with GSP
20:01karolherbst[d]: I'd wait for airlied to reply
20:03demonkingofsalvation[d]: sure, thanks
20:03demonkingofsalvation[d]: how do I tell an application to use the my nvidia gpu, like normally I'd do prime-run or something (wanted to test with mangohud vkcube but just used my cpu)
20:04karolherbst[d]: `DRI_PRIME=1` for GL, not sure if it's the same for vulkan, because I think in vulkan it's up to the app to select a device
20:04karolherbst[d]: vkcube has a `--gpu_number` flag
20:04karolherbst[d]: but maybe you need to install nvk
20:04demonkingofsalvation[d]: already have nouveau-vulkan or whatever it's called (on arch)
20:06demonkingofsalvation[d]: or do I have to delete my intel graphics driver?
20:10karolherbst[d]: you shouldn't have to
20:10karolherbst[d]: just try passing different numbers to `vkcube --gpu_number`
20:11magic_rb[d]: My laptop also sometimes fails to resume from suspend if thats the topic here
20:11magic_rb[d]: Ive a 3060 mobile
20:12demonkingofsalvation[d]: karolherbst[d]: it only shows 1 gpu
20:12karolherbst[d]: mhhh
20:12demonkingofsalvation[d]: wait I'm being a dumbasss
20:12demonkingofsalvation[d]: it says nvidia rtx 3050ti
20:12demonkingofsalvation[d]: not sure why the other one isn't showing tho
20:15demonkingofsalvation[d]: weird when I try play left 4 dead 2, it compiles vulkan shaders but that makes all my cpu cores hit 100%
20:15karolherbst[d]: yeah, that's normal
20:15demonkingofsalvation[d]: isn't the gpu meant to do that or am I mistaken?
20:15karolherbst[d]: shaders are compiled on the CPU
20:16demonkingofsalvation[d]: thanks, was about to waste a couple hours
20:16demonkingofsalvation[d]: I assume this is since I'm not using the propitery drivers (mangohud vkcube): ] [error] [loader_nvml.cpp:42] Failed to open 64bit libnvidia-ml.so.1: libnvidia-ml.so.1: cannot open shared object file: No such file or directory
20:19huntercz122[d]: demonkingofsalvation[d]: mangohud uses nvml for gpu statistics on nvidia
20:19karolherbst[d]: yeah, you can ignore that one
20:21huntercz122[d]: huntercz122[d]: which obviously doesn't work on nouveau, mangohud always expects proprietary drivers when Nvidia gets detected
20:21huntercz122[d]: You can ignore it
20:32demonkingofsalvation[d]: is there a way to check if the gpu is being used since when I play left 4 dead 2, a single core goes to 100%?
20:32notthatclippy[d]: karolherbst[d]: ~~few more years and maybe nouveau and proprietary can have the same interface for this stuff~~
20:34notthatclippy[d]: (something something hwmon something GPL something)
21:02karolherbst[d]: demonkingofsalvation[d]: some games are simply cpu bottlenecked, but yeah... if it's decently fast, it's probably running on the gpu. on the cpu it would use all cires at max.
21:09demonkingofsalvation[d]: ah I see, thanks for all the help