00:19 asch: warning: ralloc.c: source file checksum mismatch between line table (293ff95abf3dae8cee43d4f18cc5a2d5) and file on disk () on frame 3 above
00:37 airlied[d]: mohamexiety[d]: no 3d hw " GA100 does not support DirectX. For GPU compute applications, OpenCL version 3.0 and CUDA 8.0 can be used. "
01:02 asch: Could it be happening because I lack DRM_NOUVEAU_SVM on the kernel?
01:04 mhenning[d]: No, it's probably not a kernel issue. It looks like an odd userspace bug
01:04 mhenning[d]: Maybe you could try in valgrind and see if it reports anything interesting?
01:06 karolherbst: asch: any funky compiler flags you pass in to compile mesa?
01:08 karolherbst: anyway.. seems like a NULL pointer access somewhere and nobody tests mesa with musl
01:08 asch: CFLAGS="-O0 -march=native -gfull -gdwarf -fno-omit-frame-pointer -pipe"
01:08 karolherbst: I don't even think we support musl at all for various reasons
01:08 asch: CXXFLAGS="-O0 -march=native -gfull -gdwarf -fno-omit-frame-pointer -pipe"
01:08 karolherbst: like most games won't even run with musl
01:08 karolherbst: probably
01:08 asch: On a global level I only use -O2 -pipe -march=native
01:09 asch: Well, supertuxkart would run fine with the nouveau renderer up until December
01:09 asch: Wesnoth still runs fine
01:09 asch: =P
01:09 asch: vulkan renderer I mean
01:09 asch: It still runs with opengl
01:09 karolherbst: yeah.. but wouldn't be surprising if something broke related to musl
01:09 karolherbst: no support doesn't mean it might or might not work, just that nobody tests anything with musl really
01:10 karolherbst: anyway.. either it's a memory corruption or a bug in musl
01:12 karolherbst: also....
01:12 karolherbst: what is this nvkcl.cpp file?
01:12 karolherbst: ohh wait, it's generated.. nvm
01:13 karolherbst: mhhhhhhhhhhh
01:13 asch: valgrind vulkaninfo: https://pastebin.com/1m1g9iL3
01:13 asch: Well, I just ran memtest86+ for 25 hours the other day. No errors found.
01:14 karolherbst: I _think_ this is some module load order thing going weird
01:14 asch: Ditto for Windows Memory Diagnostic Tool
01:14 karolherbst: all of this happens before main is even hit
01:15 karolherbst: could be that nvk gets to initialize stuff before musl is done with malloc things?
01:17 karolherbst: asch: do you have mallocng included?
01:19 karolherbst: though maybe it's the only option these days...
01:24 asch: I don't really know. I don't have any packages with malloc on its name installed and apropos malloc only returns GC_malloc'
01:28 karolherbst: soooo
01:28 karolherbst: there is one thing that is weird
01:28 karolherbst: most of the time it uses libc functions from /usr/lib/libc.so
01:29 karolherbst: but the crashes inside nvk are inside /usr/lib/libvulkan_nouveau.so
01:29 karolherbst: I wonder if there is some weird linker thing going on and nvk just ends up calling into libc functions that aren't in the actual libc.so
01:29 karolherbst: and weird things happen
01:30 karolherbst: having line numbers would kinda help
01:31 asch: LDFLAGS="-Wl,-O1 -Wl,--as-needed -Wl,-z,pack-relative-relocs -Wl,--as-needed" LD="ld.lld"
01:31 asch: Would gdb provide those line numbers?
01:32 asch: I never touched any LDFLAGS anywhere. Those are default values
01:33 karolherbst: yeah.. no idea
01:33 karolherbst: I only know that weird things are happening
01:52 mhenning[d]: Does setting `MESA_PROCESS_NAME=something` change what happens?
01:55 mhenning[d]: Also found https://github.com/rui314/mold/issues/1071 which could be relevant if you're building musl with an affected version of mold
02:51 asch: Hello. Thank you for your attention and time. You are awesome ^^ Wayland just crashed here. It never happened before like this and here is the dmesg output: https://pastebin.com/LwgCcxj2
02:52 asch: Maybe some of you might find it relevant. I noticed that xdg-portal start complaining whenever I begin testing with vulkan/vulkaninfo, so this issue might be affecting some other things, or maybe it is a result of some other thing affecting mesa
02:52 asch: anyways, thank you a lot once again and see ya :)
05:36 CompanionCube: so last year i filed https://gitlab.freedesktop.org/drm/nouveau/-/issues/427 and now i've finally bothred to actually look into it more. This turned out to be two differnt problems: the error in in dmesg and the not-working OpenGL in Xorg. The first problem seems to be the result of
05:36 CompanionCube: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/gpu/drm/nouveau/nouveau_drm.c?id=04e0481526e30ab8c7e7580033d2f88b7ef2da3f, The second is the result of the nouveau DDX only enabling DRI2 by default, even though Mesa has dropped DRI2. Enabling DRI3 makes OpenGL at least seem to work fine (compositor, webrender, glslsandbox) , even on a 'bad' kernel.
05:36 CompanionCube: Going to update the issue now.
10:18 mohamexiety[d]: airlied[d]: Hm I can’t find anything that directly states it had it beyond the hopper white paper pointing out that only 2 TPCs are 3d capable while the GA100 one doesn’t so now I am not sure
10:19 airlied[d]: I did try bringing up nouveau on gh100 but it stopped being fun 😛
15:21 karolherbst[d]: mhenning[d]: were you looking into the prepass problem we've talked about recently? I could also maybe look into it, but was wondering if you came up with a solution. I was also wondering if this could resolve some of the perf regressions I was seeing with the prepass opt enabled.
15:33 mhenning[d]: Yeah, I have an idea I want to try out and I might get to it this upcoming week.
15:34 karolherbst[d]: okay, cool