00:36karolherbst: marex: I actually have a GT218...
00:37karolherbst: to my big surprise
00:37marex: karolherbst: is that similar to that quadro ?
00:38karolherbst: maybe
00:38marex: if you need PCI IDs or whatnot, the machine is still on
00:40karolherbst: well.. would'nt help if I don't have such a GPU.. I was just wondering if I should try it out myself and see if I can reproduce it on any of my cards
00:41marex: karolherbst: this MTHD thing happens during suspend/resume cycle reliably
00:42marex: karolherbst: the sway/wlroots problem happens with default sway config, just start sway and suspend ... it will take a while
00:42karolherbst: sure.. but it doesn't help if I can't reproduce it :D
00:42marex: that;s probably the reproducer ^
00:42marex: in case you want to attempt it
00:43karolherbst: I know, but sometimes issues are annoying and they depend on weirdo timing
00:44marex: that's a valid point
00:48karolherbst: anyway.. just updated this machine to 5.14 :D
00:49karolherbst: mhh.. saw weird artefacts on the console
00:49karolherbst: "bios: OOB 1 d7500086 d7500086" that's not good at least... oh well
00:49karolherbst: for later
00:49marex: heh
00:50karolherbst: yeah so.. well
00:50karolherbst: the machine suspended
00:50karolherbst: resuming on the other hand...
00:50karolherbst: "[ 74.018029] nouveau 0000:01:00.0: fb: trapped read at 0000026000 on channel 1 [1fbbe000 DRM] engine 0d [PCE0] client 13 [] subclient 01 [] reason 00000006 [NULL_DMAOBJ]"
00:50karolherbst: something seems busted :)
00:51marex: I might;ve seen that in the log too
00:54karolherbst: okay.. at least I am hitting one bug here :D
01:31karolherbst: marex: yeah.. seems like the error is always different
10:45marex: karolherbst: in my case the error is always the same, but I dont exercise the GPU too much, just kms console emulation and sway/wlroots
16:12karolherbst: marex: seems like it broke with 5.12
17:04marex: karolherbst: it seems it got a lot of testing since nobody noticed until now :)
17:16karolherbst: most people use distributions like debian or so and stay on old kernels for a long time
17:16karolherbst: but I think we have a few bug reports, but never got ti either reproduce it or whatever
17:17karolherbst: could also be there is a patch but it never got applied
18:23marex: karolherbst: I am using distributions like debian or so :)
18:24marex: karolherbst: unstable and experimental are fine, unreleased is a bit flaky
18:26marex: karolherbst: but I get the idea, dont worry
18:50karolherbst: uhhg.. bisecting is annoying if you hit compilation errors in between
19:27karolherbst: marex: uhhh.... I think I know what this is...
19:27karolherbst: skeggsb_: did you had time to look into this engine_id bug on nv50?
19:30karolherbst: marex: mind verifying that 64f7c698bea9 is broken and the commit before that works?
19:31KitsuWhooa: Nice :p
19:35karolherbst: Lyude: ping on that MR
19:37karolherbst: even have gitlab issue for that https://gitlab.freedesktop.org/drm/nouveau/-/issues/120
20:03marex: karolherbst: do you think ->engine_id() is called with unsupported engine ?
20:03marex: if so, then I would hit the WARN_ON(1) and get a splat
20:22Lyude: karolherbst: can you link to it? was just about to try to review it but I don't see it on my gitlab fdo todo list
20:22karolherbst: Lyude: https://gitlab.freedesktop.org/drm/nouveau/-/merge_requests/8
20:22Lyude: thanks
20:23karolherbst: marex: the patch is broken so we end up using the wrong id
20:23karolherbst: G84_FIFO_ENGN_GR is 1 but g84_fifo_chan_engine returned 0 for NVKM_ENGINE_GR before
21:10marex: ah
21:45marex: karolherbst: I'll try https://paste.debian.net/1214537/ since its easier to test right now
21:45marex: it should have the same effect
21:51karolherbst: maybe
22:15marex: karolherbst: so far so good
23:18marex: karolherbst: still good :)
23:28skeggsb: that patch will be fine, you don't need the SW check though, it'll return already above for that case
23:29skeggsb: re-send from earlier, because i don't believe my reply went through...
23:38karolherbst: skeggsb: ah okay, so this will be enough to fix it? cool
23:38karolherbst: I think I mentioned an idea like that a few weeks back
23:39karolherbst: marex: if you send the patch to the ML tomorrow with a "Fixes:...." that would be great