03:21voxadam: I have a 550 Ti. A few months ago I was experincing "random" lockups when using the Nouveau driver. Eventually I decided that for the sake of getting work done I'd use the integrated Intel GPU instead.
03:22voxadam: Today I reinstalled my 550 Ti in hopes that things had changed but within an hour I experienced a hard lock.
03:22voxadam: Well, not a hard lock, I was still able to ssh in.
03:23Samsai: voxadam, just randomly or did you do something that could have triggered it?
03:24voxadam: I've never been able to pin down an action that would trigger it.
03:24voxadam: I can say that KDE would trigger it faster than when I logged in to Gnome.
03:25voxadam: That's the dmesg from my lockup earlier.
03:25voxadam: IIRC it's not what I used to see.
03:25Samsai: i messed around with a 550 ti yesterday, not very long though and i don't use gnome or kde
03:27voxadam: Here's an old one. https://gist.github.com/voxadam/6bbf98e63e60f697c75d
03:27voxadam: Scroll to 01:08:21
03:28voxadam: BTW, I'm currently connected using the machine in question so I might appear to disappear for a bit in the event of a lock/crash.
04:17mogorva: Hi! I compiled & installed libdrm and xf86-video-nouveau from git with the help of this guide: http://nouveau.freedesktop.org/wiki/InstallNouveau/ Can you tell me why certain apps crash on start if Mesa was compiled with DRI3 support?
04:17mogorva: One of my games dies with 'nv50/nv50_miptree.c:132: nv50_mt_choose_storage_type: Assertion `ms == 0' failed.' stacktrace -> http://pastebin.com/TTTsqsms
04:25mogorva: what does this commit exactly do (introduced the problem for me): http://cgit.freedesktop.org/nouveau/xf86-video-nouveau/commit/?id=241e7289f25a342a457952b9b0e539c2f0b81d99
04:56xexaxo: mogorva: dri3 is relatively recent addition. so you're just hitting bug :\
04:57mogorva: used to work until that commit though
04:58xexaxo: (lazy explanation) - dri3 requires that both the ddx (xf86-video-nouveau) and the GL implementation support it.
04:59xexaxo: until that commit the ddx did not support it (on your setup/hardware) so it was unused.
05:00mogorva: so it's best to leave it disabled when compiling mesa
05:00xexaxo: you can toggle it at runtime as well, let me fish out the env. variable
05:01mogorva: do you mean LIBGL_DRI3_DISABLE ?
05:01xexaxo: the same.
05:01xexaxo: sooner or later we ought to fix that bug though ;)
05:02mogorva: okay, thanks for the explanation :)
05:10xexaxo: yw. although I do wonder if we should really disable assertions on release mesa builds.
09:50imirkin: mlankhorst: see mogorva's backtrace above... looks like it's trying to allocate a texture with ms > 0 and PIPE_BIND_SCANOUT
09:50imirkin: mlankhorst: happens with dri3 + exa
09:55mlankhorst: good times :)
09:59imirkin: since you added those dri3 paths i was hoping you might have some thoughts
10:06imirkin: mogorva: make sure to file a bug + trace btw. dri3 isn't particularly well-tested. could be that any application that uses a ms visual will just crash.
10:08mogorva: imirkin: where to report bugs in xf86-video-nouveau?
10:08imirkin: xorg -> Driver/nouveau
10:08mogorva: ah ,ok
10:09imirkin: but it all just goes to the same mailing list... you could report it against mesa as well
10:09imirkin: it's unclear to me who's at fault
10:10mogorva: a coredump is also generated but I don't know what to do with that
10:10imirkin: ignore that
10:10imirkin: as long as i can generate the issue with the trace, that should be enough
10:14imirkin: mogorva: btw what X version are you using?
10:15imirkin: hm ok. there was a patch to only enable it for X > 1.16.3
10:15imirkin: er, >=
10:15mogorva: the trace file is only 3 kB: https://drive.google.com/open?id=0B-tTbLKBl-tOVWpMMmlaakgyNkU
10:17mogorva: this is the terminal output: http://hastebin.com/honacazazu.hs
10:18imirkin: does replaying the trace cause the same crash?
10:21mogorva: imirkin: yes it does
10:21imirkin: excellent :) file a bug then please. you can attach the trace into the bug instead of the google drive thing
10:21mogorva: i mean i get this when replaying with glretrace: http://hastebin.com/awigegowuc.hs
10:23imirkin: weird that you don't see the assert
10:23mogorva: i don't see the 'nv50/nv50_miptree.c:132: nv50_mt_choose_storage_type: Assertion `ms == 0' failed' message
10:24mogorva: in dmesg I have this: http://hastebin.com/foyapiquwa.css
10:25imirkin: hm ok
10:29mogorva: imirkin, re: bug 91171: what does it mean if the trace plays correctly with the '-sb' (single buffer) switch?
10:30imirkin: mogorva: it means that the game is missing some flushes. or that nouveau is.
10:31imirkin: since you're also seeing some issues with swrast, i'll say that either something in core mesa is missing a flush, or the game is doing something wrong
10:32imirkin: the nvidia blob driver has some odd behaviour wrt flushes iirc
10:32imirkin: which might be why it happens to work
10:32imirkin: alternatively perhaps the game expects a single-buffered visual while wine does double-buffering. dunno.
10:33imirkin: unfortunately i have no experience with that sort of thing... i barely know what visuals are ;)
10:38imirkin: and i have no clue what's going on with the zoos thing
10:38imirkin: i've found a bunch of very mildly bogus stuff in the process, but none of it has had any effect
10:41mogorva: if you have the doubt that the problem might be in Wine as well just let me know and I open a bug report @winehq bugtracker
10:43imirkin: mogorva: the zoos thing is def a nouveau issue. i suspect that the --sb thing may be a wine issue. at least i'd like them to take a look at it and see if they think it's at all odd.
21:57imirkin: hmmm... GTX 750 for $65. almost tempting.