I noticed that in the v0.21 image, you’ve compiled the lima Mali400 drivers into the kernel:
# dmesg | grep -i lima
[ 0.882332] lima 1c40000.gpu: bus rate = 200000000
[ 0.887121] lima 1c40000.gpu: mod rate = 384000000
[ 0.892318] lima 1c40000.gpu: gp - mali400 version major 1 minor 1
[ 0.898555] lima 1c40000.gpu: pp0 - mali400 version major 1 minor 1
[ 0.904850] lima 1c40000.gpu: pp1 - mali400 version major 1 minor 1
[ 0.911155] lima 1c40000.gpu: l2 cache 64K, 4-way, 64byte cache line, 64bit external bus
[ 0.919735] [drm] Initialized lima 1.0.0 20170325 for 1c40000.gpu on minor 0
With this, and with compiling the lima-mesa fork, I was able to get an accelerated kmscube running, and am working on compiling glmark2 right now. Unfortunately I wasn’t able to get retroarch or other apps to leverage the DRI layer (yet).
Before I spend any more time on this, will one of you say what the current state is of the dev team working on Mali400 acceleration? I don’t want to waste my entire holiday vacation working on this if you’re already ahead of me. Obviously, the whole community is looking forward to this 
Similarly, if I can help you at all in any way, please let me know.
Finally, what steps did you take to compile lima into the kernel? Your other post here doesn’t include any patches for lima: CPI Kernel Source