00:03 <bamse> not interesting enough, it was just better at hiding

00:36 <steev> ah dang

16:33 <steev> i do have a very odd issue with the flex 5g that i have absolutely no idea how to track down - the first time logging in, i'm always shoved into Xorg... but if i log out and back in, i'm properly on wayland

17:17 <steev> bamse: there's definitely something up with wifi on the flex though, doing something like running "apt update" where it downloads ~70MB of files and doing something as simple as moving the cursor becomes a bit of a struggle

17:20 <bamse> i see that sometimes with usb ethernet as well...so i suspect it could be storage related

17:20 <bamse> steev:

17:21 <steev> i wasn't sure if it was storage or not, it could be there too

17:32 <bamse> perhaps worth doing some io benchmarking and see if that shows the problem as well?

17:35 <robclark> there has been some recent work around smmu unmap optimizations on sc7180 (which seem to benefit wlan, emmc, etc)..

17:58 <bamse> robclark: don't see the same thing on c630...so i don't think it's related to that...this time

17:58 <bamse> robclark: but that work certainly seems interesting

17:59 <robclark> there were a couple different fixes.. one related to compute dsp (which might have been a coreboot change) might be sc7180 specific.. the wait-for-safe dpu change probably isn't

18:00 <robclark> but I think they are all somehow related to the hw wait-for-safe thing

18:24 <bamse> ahh, forgot about that thing

18:25 <bamse> on 845 the wait for safe issue caused a 1000 fold slowdown in ufs performance, making it impossible to use the board

18:28 <robclark> it looks like `drm/msm/disp/dpu1: add safe lut config in dpu driver` is a thing that could also be helping 845..

18:40 <steev> i do have that pulled in on my c630 kernels

18:56 <steev> bamse: fwiw, the reason i think/thought wifi over IO is that i can do things like build mesa/libdrm and don't see the issues at all

20:53 <bamse> steev: ahh, you're right

20:53 <bamse> same here