06:11fililip: sorry, I still haven't figured out the matrix bridge so I use an IRC client for sending but I see everything just fine
06:12fililip: my question is whether there already is a userspace solution for reading IP dumps so that users can submit them alongside kernel / journal logs when they submit bug reports
07:26kode54: would IP dumps be useful for any of my RT bug reports?
14:09Venemo: kode54: if by RT you mean ray tracing, I think it's best to ask in the bug report
14:10Venemo: fililip: I am not aware of that other than with umr
14:38agd5f: fililip, when a GPU reset happens, a devcoredump event is generated and you can read the device core dump from /sys/class/devcoredump/
14:49fililip: oh, that's very handy, thanks!
14:49fililip: it seems to work even on 6.10 however, so is the information more limited in my case than it would be compared to 6.12?
14:55agd5f: yeah, 6.10 didn't expose as much data in the devcoredump
14:56fililip: also, that data file is very large, does the last part of it contain something like full VRAM contents?
15:14agd5f: register and ring dumps
15:14agd5f: it's human readable
15:15fililip: yes, by 'last part' I meant the very long list of address-value mappings
15:18fililip: thanks again for pointing me to where it's stored, that should make it much easier for Mesa devs to figure out GPU hangs, right?
15:27Venemo: agd5f: what does this device core dump contain?
15:28agd5f: Venemo, a bunch of driver state and register dumps
15:28Venemo: wasn't he asking for an IB dump? or is that also there?
15:29agd5f: just the rings at the moment
15:43Venemo: it would be very useful to see what commands were in flight when the hang happened
15:51tonyk: fililip: this was my last attempt on getting more reset information on amdgpu: https://lore.kernel.org/lkml/20230501185747.33519-1-andrealmeid@igalia.com/
15:51tonyk: fililip: are you looking for some info in particular?
15:58fililip: I saw your talk from around a year ago and was wondering if the situation with userspace reporting is better now than it was before
16:17tonyk: fililip: well we have some progress, devcoredump has more information, I think the user experience after a reset had improved too from my tests. But no groundbreaking advancements in managing to link the kernel dump with the actual reset cause
16:52fililip: ah, so even with IP block dumps coming in 6.12 it's not as good as it was intended to be?
17:13tonyk: it will do some help, but not sure in which level
18:04fililip: i think i figured the bridge thing out, does it work?
18:17Remco: fililip: Seems so
18:17fililip: thanks!