02:54armada: I'm trying to get nouveau running on my rtx3090, but as soon as linux boots the display is immediately disconnected and doesn't come back. Even booting the usb live environment of any distro (ubuntu, arch) is impossible without the 'nomodeset' kernel parameter.
02:56armada: Debian bookworm does seem to be able to get at least a lowres mode to work, though it doesn't manage to actually get far enough to start the desktop environment and if I update to the latest kernel in the unstable branch it shows the same issues as all other distros
02:59armada: I haven't found any obvious errors in dmesg, it seems to detect the display, its detects proper display modes and nouveau seems to initialize fine. I also tried connecting a different 1080p monitor without success.
03:01armada: Adding nouveau.config=NvGspRm=1 or nouveau.config=NvMSI=0 to the kernel parameters didn't work either
03:08armada: Here's the output of journalctl | grep nouveau: https://pastebin.com/8fPgPPcy
04:11armada: also my kernel version is 6.9, I'll try 6.10 as soon as it's available on arch
04:15armada: actually I can just enable core-testing and try it out now
04:23armada: no change when using the 6.10 kernel
04:24armada: also the same GPU works fine under Windows
04:25tiredchiku[d]: armada, what happens if you use `nomodeset` kernel cmdline option
04:27armada: that works fine, I can use a console like normal
04:30tiredchiku[d]: this sounds like NyB's issue, hmm
04:31clangcat[d]: We could hire a bear
04:31clangcat[d]: to steal the Nvidia GPUs
04:32tiredchiku[d]: Jul 21 04:10:09 archlinux kernel: nouveau 0000:0c:00.0: DRM: [DRM/0000000b:kmsOutp] [DP_AUX_XFER type:5 size:16 addr:00050] 0 size:16 (ret: 0)
04:32clangcat[d]: then people wouldn't have issues with them.
04:32tiredchiku[d]: huh
04:32clangcat[d]: Think about
04:32clangcat[d]: I'm right
04:32tiredchiku[d]: could you try switching the monitor to HDMI?
04:32tiredchiku[d]: armada ^
04:33armada: sure, one moment please
04:34tiredchiku[d]: of course ;)
04:34tiredchiku[d]: :)*
04:34tiredchiku[d]: (typing on phone sucks)
04:38armada: tiredchiku[d]: that actually worked :D
04:38armada: I'm not sure why I didn't think to try the HDMI port
04:41armada: thank you for the tip and if I can help out with debugging this DisplayPort issue I'd be happy to
04:43armada: clangcat[d]: if a bear steals my GPU then I would still have the "disconnected monitor" issue ;)
04:43clangcat[d]: armada: Not if the bear also robs your monitor
04:44clangcat[d]: :Foxy_Knife:
04:47armada: and then I can finally be free and live my life in nature while the bear is trying to figure out why the monitor won't working when connected over displayport
04:47tiredchiku[d]: am glad it worked, though it's more of a workaround and not a real solution...
04:48tiredchiku[d]: would you mind filing an issue over at https://https://gitlab.freedesktop.org/drm/nouveau
04:48tiredchiku[d]: could get dmesg from one of the previous boots with journalctl -b -1 -k
04:49tiredchiku[d]: replace -1 with howmanyever power cycles ago
04:49armada: you mean the one that successfully connected over HDMI?
04:49tiredchiku[d]: no, the DisplayPort that failed
04:53armada: sure, will file the issue tomorrow, in the mean time I'll stick around on IRC, I was planning to do some work on NVK anyway
04:56clangcat[d]: armada: Exactly and if the bear works it out you steal it back.
04:57clangcat[d]: Exploit the bear's working class
04:57clangcat[d]: Anyway thank god pride flags are simple patterns
15:36NyB: tiredchiku[d]: my issue was different; it's a complete freeze. I'm 98% sure it's a software issue, although I suppose there could be something wrong that makes the hardware pee outside the sandbox when modeset is enabled...
15:37NyB: *is different -> I still had last time I checked
15:39tiredchiku[d]: hm, yeah
15:39tiredchiku[d]: I still wonder if your modeset issues are related to DP jank though
15:44NyB: dunno. And with how nvk and the new nvidia OSS kmod thing seem to be grabbing all the attention (not without reason), I see myself running the thing on pure VBE framebuffer only before long :p
18:57armada: tiredchiku[d]: should I collect the logs with nouveau.debug=debug on the kernel params?
18:58armada: also I get a feeling this is related to some high referesh rate display modes being exposed over DisplayPort that nouveau can't handle
18:59armada: because if I connect it over HDMI 2.0 rather than a shitty HDMI 1.4 cable I get the same problems as DisplayPort
19:04tiredchiku[d]: nouveau.debug=trace, maybe
19:04tiredchiku[d]: not too familiar with nouveau's debug options, sorry
19:05armada: I'll post my logs as-is and we'll see if we need more
19:11tiredchiku[d]: sounds good to me
19:19armada: created issue #376
19:26armada: seems likely it's related to #356
19:28tiredchiku[d]: awesome! I'm no expert but I'll try to take a look in the morning
19:28tiredchiku[d]: currently 1am for me
19:28armada: I appreciate it, have a good night
22:01skeggsb9778[d]: does DP work if you use nouveau.config=NvGspRm=0?
22:05clangcat[d]: skeggsb9778[d]: If you mean like display port connectors they should work and I don't believe they have anything to do with GSP. I think.
22:05clangcat[d]: Though of course there may be other reasons it may not be working.
22:06skeggsb9778[d]: sorry, I was replying to armada - i'm interested to know if it works in the non-GSP paths
23:14armada: skeggsb9778[d]: No change when using NvGspRm=0
23:21skeggsb9778[d]: ok, thanks for checking! there's a couple of DP bugs with GSP-RM i haven't tracked down yet, so i was wondering if your issue could be related - but doesn't appear so
23:24clangcat[d]: skeggsb9778[d]: Interesting. Does GSP actually handle Display port stuff?
23:24skeggsb9778[d]: It handles a large portion of what happens during a modeset
23:27clangcat[d]: skeggsb9778[d]: Interesting guess it was largely true what I heard GSP just moves a lot of stuff into firmware.
23:30skeggsb9778[d]: NVIDIA HW is a bit different to others when it comes to modesetting. It has channels and push buffers, just like graphics. The KMS driver fills the push buffer with methods to describe the modeset, and says "go"
23:31skeggsb9778[d]: HW then goes off and does its own thing, but interrupts the driver 3 times at various stages to do whatever it needs to
23:31skeggsb9778[d]: GSP-RM handles all the "driver" side of that now
23:31clangcat[d]: Yea yea just interesting.
23:31clangcat[d]: and yea a little surprising how much functionality is in GSP
23:32armada: soon we will be communicating to Nvidia GPUs using Vulkan over PCI-e
23:34clangcat[d]: armada: I mean technically you can put as much stuff as you want in firmware it's all just code at the end of the day XD.
23:35armada: I mean there's already an Nvidia GPU that has an m2 SSD slot, someday your distro will just run on the GSP and you won't need the rest of the PC anymore