00:13 imirkin: i think you're in the wrong channel
00:13 imirkin: but ... i'd be thoroughly surprised if amdgpu didn't support HDMI audio
00:13 imirkin: and slightly less but still very surprised if it didn't support DP audio
00:16 mangix: it's a 2021 GPU so who knows
00:30 HdkR: Who knows, could even be 2022 because there have only been leaks about it, not releases :P
00:36 RSpliet: imirkin: ah, the joke went over heads. I was simply amused by the codename. I mean sure, you'll get great graphics out of him, but he _is_(...was) missing an ear
00:37 imirkin: RSpliet: yeah, didn't make the connection. sad.
00:41 HdkR: oop
01:27 mangix: I find it interesting that GPUs all the way back to 200 series can be made to work with UEFI
01:28 mangix: year 2008, wow
01:29 raket: karolherbst: tried your threading branch now with nv120 (b64861ef334f7d367c6a3c5cb9fd1f6a0b40038f) higher fps in cs-go(!) but rendering hangs after a few seconds! this is what syslog reports kernel: nouveau 0000:01:00.0: gr: TRAP ch 9 [017d79c000 Xorg[1518]] -> kernel: nouveau 0000:01:00.0: gr: SHADER a2040a0e, sph: 0x040a0e, stage: 0x22 <- what is that anyway? GPU hang?
05:45 DrNefario: is there a way I can force an EDID after loading nouveau?
05:46 DrNefario: I have an EDID in a .bin file and I need to apply it to output DP-1
06:14 DrNefario: something new happened when I started nouveau this time, it didn't immediately fail to get the EDID. https://termbin.com/12nw
06:14 DrNefario: the display still turned off, but I've never seen "disp: outp 00:0006:0f82: link rate unsupported by sink" before
06:18 DrNefario: is it possible that nouveau is setting DP-1's bandwidth too low to support my display?
06:19 DrNefario: is there a way I can check/force the connection link rate?
06:28 DrNefario: this log might have a little more info: https://termbin.com/8nww
07:57 DrNefario: I think nouveau steals the EDID from my second monitor and applies it to the first
07:58 DrNefario: when the second monitor isn't connected, nouveau fails outright and basically turns the screen off
07:59 DrNefario: when it is connected, nouveau almost sort of works but in a very wrong way. it looks like the framebuffer is interlaced with itself.
08:08 DrNefario: https://imgur.com/a/7rbd57T
08:25 RSpliet: DrNefario: these are good questions. I have relatively few insights, but... I do think that nouveau's standard EDID read/write paths are quite robust given the large number of working multi-monitor set-ups.
08:26 RSpliet: Most EDID-related issues are caused by either a bad monitor (less common these days I feel, but others might disagree) or a bad cable/converter/KVM
08:27 DrNefario: my monitors work fine on Windows and Linux with the proprietary drivers, so I think this must be some rare case that nouveau fails
08:27 RSpliet: Lyude might have some insights on the "link rate unsupported", but... from the back of my head she's in the US, so is probably still catching z's
08:28 RSpliet: DrNefario: ah thanks, yes that points in a different direction then :-) Nouveau's mode setting code isn't perfect esp for newer HW. If the link rate issue is related to the EDID reading issue then I guess that explains some things.
08:30 RSpliet: I suspect that focussing on replacing the EDID might barking up the wrong tree then. Better fix that DP link set-up
08:38 DrNefario: is there anything I can do to help?
08:38 DrNefario: so far I've just opened an issue https://gitlab.freedesktop.org/drm/nouveau/-/issues/106
08:38 DrNefario: if link speed is the underlying cause then the title may be misleading
08:48 DrNefario: I just managed to get a new error: "nouveau 0000:26:00.0: disp: outp 04:0006:0f44: training failed"
08:52 RSpliet: Plop the new error onto the issue tracker. Also, do please try a 5.13 kernel if you haven't already. Just to rule out the potential case that the bug has already been fixed prior
08:53 RSpliet: nouveau.modeset=1 should be the default option btw, there shouldn't be a need to add that to the kernel params. Not sure about .config=NvI2C=1
08:54 DrNefario: those are the kernel parameters the arch ISO uses by default
08:54 DrNefario: so far I've only been doing this on a live USB, but at this point I think I have to do a proper install
08:54 RSpliet: No worries about the param, they don't harm either :-)
09:46 karolherbst: raket: mhhh, interesting... I'd have to take a look on what's going wrong there
11:21 karolherbst: raket: do I have to be ingame or ist the menu enough or something?
12:18 DrNefario: can someone help me make sense of this log? https://termbin.com/ju6v
12:19 karolherbst: DrNefario: that essentially means we configure the GPU wrongly or so for displaying
12:19 DrNefario: ah, does it say what went wrong?
12:19 karolherbst: not sure
12:20 karolherbst: I mean.. it does print the error the GPU reports and it dumps the state
12:20 karolherbst: but it's not easy to parse this info
12:20 karolherbst: that's something for skeggsb_ or Lyude to look into
12:21 karolherbst: DrNefario: but I guess it's just a simple 1680x1050 display, right?
12:21 karolherbst: did you had the nvidia driver loaded before or something?
12:21 DrNefario: no, I have two displays connected right now
12:21 karolherbst: ahh
12:21 karolherbst: does it work with one?
12:22 DrNefario: it works when only the secondary display is connected
12:22 DrNefario: the primary display can only work if nouveau is disabled with nomodeset or something similar
12:23 DrNefario: the moment I load nouveau my monitor craps itself
12:23 karolherbst: mhhh
12:24 karolherbst: oh well... csgo just crashes without my patches.. nice
12:26 karolherbst: ahh "multicore-rendering: enabled" :)
12:37 DrNefario: I have to go now. I'll read the logs later in case anyone figures something out
13:59 raket: karolherbst: start steam -> start csgo -> start a map with bots -> wait a few seconds -> gpu hang
14:05 karolherbst: raket: I also managed to crash it once without my patches, but sadly only once :/
14:14 karolherbst: raket: can you say how often that happens?
14:14 karolherbst: I see a potential reason, but I can also see that it usually doesn't happen
14:15 karolherbst: or maybe it's special to maxwell
14:15 karolherbst: nv120 did you say?
14:15 raket: karolherbst: i tested 3 times yesterday, the error was 3 times, and when i was about to test the last time i joined a competetive game to play and had to revert to mesa-21.1.5 to just finnish the game :D
14:15 raket: yaeh nv120
14:15 karolherbst: mhhh
14:16 karolherbst: okay... for me it just works
14:16 karolherbst: raket: but on 21.1.5 if you open the overlay and open the browser you probably get an instant crash, no?
14:17 raket: there's nothing wrong in mesa-21.1.5 but with b64861ef334f7d367c6a3c5cb9fd1f6a0b40038f it hangs
14:17 karolherbst: well.. at least with my turing GPU it worked fine
14:18 karolherbst: raket: well.. I am sure you can get it to crash with the browser inside the overlay.. maybe you need to enable browser acceleration before...
14:24 karolherbst: raket: settings -> interface -> "enable GPU acceelerated rendering in web views" with this enabled, 21.1.5 should crash quite quickly :D
14:25 karolherbst: I guess you have that disabled?
14:25 raket: sec
14:27 karolherbst: oh no... wrong GPU :D
14:33 raket: karolherbst: no crash with that
14:33 karolherbst: okay
14:34 karolherbst: sadly those things are way too random :D
17:01 karolherbst: raket: okay.. on my gm206 I am able to hit your issue :)
17:01 raket: ok!
17:01 raket: progress :D
17:01 karolherbst: yeah... but I think I know what's up
17:02 karolherbst: mhhh
17:08 Lyude: looks like that something with the display settings is programmed incorrectly
17:08 Lyude: I can translate tat code there in a little bit
17:08 Lyude: *that
17:17 Lyude: ...that's bizarre
17:17 Lyude: if I'm reading this correctly it appears the GPU is telling us that it's being programmed in clone mode but does not support clone mode
17:19 Lyude: that's for the core channel though, window seems to be reporting a mismatch between the produced and output color formats
17:19 karolherbst: huh :O
17:19 Lyude: unfortunately they're gone
17:19 karolherbst: yeah...
17:20 karolherbst: but still weird
17:20 Lyude: but my assumption is we're somehow very much programming the or settings incorrectly
17:20 karolherbst: yep
17:20 Lyude: and hitting one of the "nice to have but not implemented" nvidia display features on accident
17:20 karolherbst: ahh
17:20 karolherbst: maybe it is implemented on few chips?
17:20 karolherbst: dunno
17:21 karolherbst: bit they also have this interlaced on DP thing :D
17:21 karolherbst: for who knows what reasons
17:21 Lyude: karolherbst: I don't believe it is these days, this was something only supported with PIORs which no longer even exist with turing. but they might add it back at some point
17:21 karolherbst: ahh
17:21 Lyude: anyway-would definitely need to see the full atomic state so I can figure out how it comes to the conclusion this is the mode to set
17:22 karolherbst: do we have a tool to parse those dumps?
17:22 karolherbst: we really need better tooling around our debugging
17:22 Lyude: karolherbst: you mean the one I just looked at? unfortunately no, I'm not sure we can share all of the details from this but I've definitely mentioned this in the past that I'd like us to be able to get slightly more descriptive error messages…
17:23 karolherbst: ohh sure, but I am not directly talking about printing all the stuff in dmesg, more like having a tool to throw dmesg stuff against and it just parses it
17:23 karolherbst: like I did for push buffers
17:23 karolherbst: so if you have access to more stuff you could feed it into the tool
17:24 karolherbst: atm I just use the rnndb we have
17:24 Lyude: ahh - honestly I think we should implement something like that within nouveau itself, since for display we actually have nearly complete headers for most things
17:24 karolherbst: but I could use nvidia headers once we get the 3D stuff
17:24 karolherbst: ahh
17:24 karolherbst: cool
17:24 Lyude: very fancy headers, too
17:24 karolherbst: true
17:24 karolherbst: I really would like to wire up nvidias headers for rnn though
17:24 karolherbst: without 3D....
17:25 Lyude: karolherbst: also - I have a WIP of implementing the atomic state printing callbacks for nouveau somewhere
17:25 karolherbst: Lyude: btw.. if you enver want to look at kernel sides pushbuffers I could even add support for that format we see in dmesg or so
17:25 Lyude: so we can get debugging output that looks more similar to what i915 has
17:25 karolherbst: sounds like a plan
17:26 Lyude: yeah - most of the time you're really just trying to make it so it prints out a state you can look at and go "this seems funny"
17:26 karolherbst: yep
23:38 DrNefario: how can I upload the full atomic state of nouveau? I'd like to help in any way I can
23:49 DrNefario: could the issue have anything to do with my monitor supporting 3d vision?
23:50 karolherbst: DrNefario: uhhh.. hopefully not?