Jiri:
the main issue is that the compilation is running on virtual machine, not actuall hardware. So now we need to get hands on some computer that I would just connect to the network. Im looking for some LGA1366 workstations, even dual socket maybe...quite cheap but still powerfull with 32nm Gluft based chips. Also usefull for git, Qt compilation...
Second, we don't need to rebuilt kernel that often, you are right. We are editing current drivers for different pinmuxes and making sure McASP interface can handle 8 channels and so on... Also the LCD drivers, they are directly in kernel itself and we needed to edit them.
Every other driver uses DTB, that means we are good.
Virtual machine shouldn't really matter as long as you're running HVM. These days, an outdated mainstream sandy bridge (or skylake, 1151) works just fine - as said, I'm getting 4 minutes as kernel compilation time with reasonably limited config - it's the drivers for all the peripheral devices that take most of the time during distro-config rebuilds, which take about 1 hour here.
As 44000khz mentioned, though, if you have the disk space, ccache works really well.
About LCD drivers - right, if you need to compile them in, you're right.
If "directly in kernel itself" was meant to be "in the upstream source tree", you could still copy the .c / .h out and compile out-of-tree, it's what I
used to do for undervolting a C2D CPU and found it more maintainable, maybe something to consider.
I would understand if you chose not to do that and maintain the patches in git, though.
For the rest - sure, makes sense, thanks for the explanation. Here was me hoping I'd be able to run FDAC on a vanilla kernel.