09:21asdqueerfromeu[d]: NVK is in a real danger to losing to Turnip in terms of extension support (there are still some inactive MRs for extensions) 📉
09:38HdkR: If you plug an NVIDIA GPU in to a snapdragon then you get the cumulative total of both extensions :P
15:47clee_: HdkR: wait, that's illegal
17:40imachine: hello
17:40imachine: is there a way to force nouveau not change power state?
17:40imachine: it seems to hang for me (G86M [Quadro NVS 140M])
17:41imachine: recently upgraded to new ubuntu and it's an issue w/ this driver it hangs on 2d and performance in 3d is poorer than w/ nvidia-340
17:41imachine: but nvidia-340 support was dropped in ubuntu 22.04 which i upgraded to
17:41imachine: so either I downgrade to 20.04 or make this nouveau driver work and perform w/out hanging and ok working in 3d
17:42imachine: I play occasional savage-sxr and quake2; quake2 ok perf; savage - soso
17:42karolherbst: imachine: nouveau doesn't support changing power states on its own at all, so you get whatever the GPU is at when it boots
17:42imachine: but 2d performance issue most, it hangs just using firefox, performance slow on scrolling pages, artifacts come up w/ chromium rendering issues in 2d
17:42imachine: karolherbst, ok is there a way to change the power state ?
17:43imachine: because it seems the issue is related to a power state suddenly dropping performance.. I would assume, since once the issue happens the system is rendered inoperable from graphical point of view
17:43imachine: so can't even read logs or so
17:44karolherbst: there is a `/sys/kernel/debug/dri/0/pstate` file you can write the perf state you want it to use, read the file to see what's available, and the last line is the current state
17:44imachine: you can move the cursor a bit but it just hangs like does not respond to clicks or changing to text console ctrl-alt-f2
17:44karolherbst: (might be 1 isntead of 0 if you have multiple GPUs)
17:44karolherbst: yeah... there seems to be issues with very slow GPUs where the GPU times out on the job...
17:44imachine: says error writing function not implemented
17:44karolherbst: mhh, might not be enabled on the G86 then
17:45imachine: like I just echo "22" > /sys/kernel/debug/dri/0/pstate
17:45imachine: which is top mode
17:45karolherbst: yeah.. it seems we only support it for a bit newer GPUs sadly
17:46karolherbst: well "support", nothing in that area is really supported anyway. In regards to the issue you are seeing, you could share your kernel log from a boot running into the issue, maybe it's something fixable
17:53imachine: yeah that would mean it's old boot?
17:54imachine: because when it happens it's frozen to the point where can't really show kernel...
17:54imachine: I could just keep showing it "live" and run into issue then take picture of the screen
17:55karolherbst: yeah.. you might be able to fetch it via "journalctl --boot -1 --dmesg" (or use a different number than -1 depending if it was the previous boot or some older one)
17:55karolherbst: should probably see errors at the bottom, might also not see anything if the machine was rebooted without the kernel being able to write out the log to disc
17:55imachine: Direct firmware load for nouveau/nv84_xuc00f failed with error -2
17:55imachine: found this
17:55karolherbst: you can ignore that one
17:55imachine: vp: unable to load firmware nouveau/nv84_xuc00f
17:56karolherbst: really, it's easier to show the full log, because most/some of it can easily be ignored
17:58imachine: Aug 1 19:57:42 t40p kernel: [ 1563.018163] nouveau 0000:01:00.0: Xorg[1332]: validating bo list
17:58imachine: this is what I see
17:58imachine: Aug 1 19:57:42 t40p kernel: [ 1563.018152] nouveau 0000:01:00.0: Xorg[1332]: fail ttm_validate
17:58karolherbst: okay...
17:58imachine: Aug 1 19:57:42 t40p kernel: [ 1562.945282] nouveau 0000:01:00.0: Xorg[1332]: validate: -12
17:58karolherbst: yeah, that looks like an issue
17:58imachine: it happens w/ using firefox
17:58imachine: so I assume that's what's causing the hangs
17:59imachine: https://bbs.archlinux.org/viewtopic.php?id=270521
17:59karolherbst: it could be that you are simply running out of VRAM here and nouveau was always kinda bad at handling the situation
17:59imachine: looks like someone run into it
18:00imachine: and it looks like issue from 2016 :)
18:01imachine: yes it seems to be memory issue like you write
18:01karolherbst: right.. though I think I have to say that this issue would require a huge amount of work to be fixed, because handling the out of VRAM situation isn't necessarily an quick one to add.
18:01imachine: well so downgrade to ubuntu 20.04 it is then
18:01imachine: and back to nvidia-340
18:02imachine: on archlinux I saw someone do on unsupported aur patches for nvidia-340
18:02imachine: to work w/ kernel 5.15
18:02imachine: but I have not found yet way to use ppa or something seemless w/ ubuntu 24.04 to work w/ -340 and 5.15 kernel
18:04karolherbst: yeah.. I wished I could be of any help with your issue here, but I don't want to give you any false hopes here
18:04karolherbst: but yeah.. there are people around writing custom patches to support newer kernels with the nvidia drivers
18:05imachine: yeah that seems to work/ it would be possibly middle ground so to say - keep updated system (w/ arch) but also old drivers support (via patches/aur)
22:45karolherbst[d]: ahuillet[d]: any luck on this? And also, do you know what nvidia is doing for SoCs (e.g. Tegra)? Atm what I've came up with uses the PCI device id + bus address, but... tegras don't have a device id anyway.