20:35:55 <steev> robclark: is mesa 21.3 rc1 known to have corruption on a680?
20:36:41 <robclark> not really.. although tbh we haven't been using release branches for CrOS
20:36:51 <robclark> and I do basically all my dev/test on main
20:38:20 <steev> https://usercontent.irccloud-cdn.com/file/CYgQOFm8/1635298694.JPG

1635298694.JPG112.44KB • image/jpeg
20:38:34 <steev> Sorry for picture, I can’t get an actual screenshot to happen
20:39:22 <robclark> maybe ubwc settings aren't right (either on gpu or dpu side) for 8cx?
20:39:42 <robclark> dump /sys/kernel/debug/dri/0/state ?
20:39:44 <steev> that's possible, i'm not sure.  i'm using bamse's newest -next stuff for the flex5g
20:40:25 <robclark> if you can start gnome-shell from cmdline (or otherwise inject env vars), FD_MESA_DEBUG=noubwc ?
20:40:29 <steev> https://paste.debian.net/1217013/
20:40:53 <robclark> ok, modifier=0x500000000000001 .. it is using ubwc
20:41:19 <steev> yeah that works
20:42:38 <robclark> ok.. yeah, then *some* sort of UBWC disagreement between gpu and dpu, I guess.. if this worked before, I suppose you could bisect kernel?
20:43:05 <steev> it worked fine on 21.2.2
20:43:28 <steev> but 21.3-rc1 is the corruption, but definitely could be kernel side
20:45:34 <steev> ubwc is widebus?
20:46:08 <robclark> no, bandwidth compressed format
20:46:53 <robclark> you could try reverting 0e15d5af81bba7c1e798cf10fc838988c86daafd .. if that wasn't in 21.2.x.. that changed a bit the logic about what modifier we pick, so possible the issue wasn't seen before because we were picking linear instead of ubwc?
20:52:19 <steev> i'll take a look
20:54:26 <robclark> easy check is dump $debugfs/dri/0/state in the working case.. modifier=0x0 is linear, modifier=0x500000000000001 is ubwc
21:02:28 <steev> yeah working it shows 0x0
21:03:15 <robclark> ok.. sounds like there is probably some incorrect ubwc configuration on kernel side on dpu and/or gpu side of things
21:03:38 <robclark> I guess the mesa commit I mentioned switched us from linear to ubwc