21:45 karolherbst: ahhh, I know now of 3 games which have quite often GPU hangs or causing dead channels :/
21:51 imirkin_: you want one that kills the entire comp?
21:51 imirkin_: F1 2016 (or 17 ... the last one that isn't vk-only)
21:52 imirkin_: once you try to go into a race ... ka-boom
21:52 imirkin_: iirc hakzsam debugged it to a shader that goes into an infinite loop based on some ssbo data
21:54 karolherbst: mhh
21:54 karolherbst: "kills the entire comp" you mean freeze X?
21:54 imirkin_: hangs the pci bus, i think
21:54 imirkin_: which in turn makes the box unhappy
22:00 karolherbst: uff
22:00 karolherbst: that's heavy
22:00 karolherbst: but I would be happy debugging the other fails as well :)
22:00 imirkin_: probably because the frag shader is in an infinite loop, and is accessing system memory somehow
22:00 imirkin_: and it can loop a lot faster than it can access memory
22:01 karolherbst: mhhhh
22:01 karolherbst: I really wished we knew how to setup our trap handler and debug running shaders :/
22:01 imirkin_: hakzsam wrote some patches to diff ssbo's relative to a "working" run on a trace
22:02 imirkin_: karolherbst: https://people.freedesktop.org/~imirkin/traces/F12015.trace.xz
22:02 imirkin_: good chance this hangs the gpu
22:02 imirkin_: so ... use with caution
22:04 karolherbst: shouldn't be too bad
22:05 karolherbst: ohh, wasn't fdo dead or something?
22:05 imirkin_: ?
22:06 karolherbst: can't connect
22:06 karolherbst: ahhh, now it works
22:06 imirkin_: it took me a whiel too
22:06 imirkin_: i dunno what's up
22:11 karolherbst: imirkin_: segfault
22:11 karolherbst: ohh could be because I used the frameretracer glretrace
22:11 karolherbst: but... kernel rejected a pushbuffer
22:12 imirkin_: well, we haven't looked at it in a LONG time
22:12 imirkin_: for all i know, some recent fix has improved things
22:15 karolherbst: mhh
22:15 karolherbst: now it didn't crash
22:16 karolherbst: still got that rejected pushbuffer
22:16 karolherbst: but I have some application which just cause nouveau to kill the channel
22:16 karolherbst: well those famous CTSW_TIMEOUT issues
22:16 karolherbst: would be cool to be able to debug those
22:26 imirkin_: oh well
22:26 imirkin_: something must have changed
22:26 imirkin_: used to be pretty reliable
22:29 karolherbst: it causes a couple of shader evicitons though
22:30 imirkin_: but doesn't use multiple contexts like that
22:31 karolherbst: ohh, it does use multiple contexts?
22:31 karolherbst: ahh I still had my 1 << 14 hack
22:43 HdkR: Oh snap, F1 201x uses infinite looping shaders waiting on SSBO results? That's lovely
22:44 imirkin_: HdkR: no, the ssbo value we have just causes an infinite loop
22:44 imirkin_: on another impl, the ssbo value had something which caused it to not infinite-loop
22:44 HdkR: interesting
22:46 imirkin_: basically like a while (ssbo != foo) { do things with foo }
22:46 imirkin_: but our value in the ssbo could never equal foo
22:46 imirkin_: or something
22:46 imirkin_: it was a long time ago