09:11pmoreau_: skeggsb, Lyude: Was it one of you who was interested in CRC readback? https://github.com/NVIDIA/open-gpu-doc/commit/36db1ace0c3658ad43b5d94ccbba9be15f14dee5
10:24cosurgi: in about 15 minutes I will have to terminate this live gdb session attached to crashed xserver
10:24cosurgi: pmoreau: you seem to be around? I have sent gdb backtrace to skeggsb and imirkin. Do you want to see it too?
10:25cosurgi: it's from nouveau compiled with -O0 -g, so has plenty of info.
10:47pmoreau: cosurgi: I am kinda around, but at work and busy with something that I need to finish within the next 30 minutes, so I don’t have time to look at it, sorry.
10:47cosurgi: pmoreau: ok. No worries, thanks :)
10:54cosurgi: eventually we will find this bug :)
11:00cosurgi: anyway, here's the gdb backtrace: https://paste.ubuntu.com/p/dSYBKzq5mW/
11:01cosurgi: with p *bo; i args; disas $pc,+16
11:02cosurgi: HdkR: though `info regs` apparentnly is not a gdb command. Not an info menu item either (or maybe I have some -doc package not installed).
11:03cosurgi: fortunately with the habit to change VT to the "debug" xserver before going to sleep preserves my work sessions in the other "work/science" xservers (usually).
12:50Lyude: pmoreau: it was me and skeggsb yeah
12:51Lyude: pmoreau: !!!!!!!
12:51Lyude: igt on nouveau, here we come
13:09pmoreau: Have fun, Lyude!! Hopefully they provided enough information for your needs!
13:12Lyude: skeggsb: I'm going to play around a bit with this
14:19Lyude: skeggsb: so-it looks like the way to allocate a dma buffer that's accessible by the GPU/CPU would be to create an nvif_object for it?
14:22imirkin: Lyude: have a look at how clut's are done
14:23imirkin: (search for "lut" in dispnv50)
14:23imirkin: they also have to allocate a buffer
14:26Lyude: imirkin: ahhh, I see now. thanks for the tip!
14:27imirkin: tbh i don't remember what that way is, i just know it's done and it works :)
14:27imirkin: gtg, good luck with crc's -- great to see that got released.
14:28imirkin: cosurgi: given the SIGBUS, and that we checked that the memory is in-range, i don't think there's any need to run with gdb anymore
14:28imirkin: cosurgi: i think at this point, the fault lies in the kernel
14:29imirkin: (get it? fault? ha.)
14:51cosurgi: imirkin: thanks for the hint! I got used to attaching gdb to running xserver. Uh. so we need to nail the bug inside kernel?
14:52cosurgi:already has two lines of ode in the kernel. Resuming from hibernation wasn't working when more than 4 CPUs were in the system.
14:52cosurgi: but the kernel hibernation dev did all the hard work. I was only a tester in fact.
14:53cosurgi: imirkin: can you tell me which files in the kernel are affectin my problem?
14:53cosurgi: or even better, which functions :)
18:34scientes: NVidia GeForce MX130?
23:42imirkin: mwk: know of a way to tell on tesla whether the current invocation is a helper invocation or not?
23:42imirkin: i.e. thread_kill on fermi+
23:42mwk: I can think of a workaround
23:43mwk: there's this "live pixels only" attribute on tex instruction
23:43imirkin: a little ... too clever.
23:43mwk: not sure if there's a proper way
23:44imirkin: i think it's just not going to be supported on nv50 :)