Huge interrupt overhead reported after RAM added to Atom-based system

[Brett Glass]

Have been doing more experimentation regarding this problem. It 
doesn't occur with a 1 GB memory module in the machine, nor with a 
2 GB module -- only a 4 GB module.

This makes me wonder if there's some sort of memory bank switching 
or extended addressing mechanism here (like PAE). Perhaps it's 
causing some sort of interrupt every time a certain area of memory 
(perhaps the topmost portion) is accessed?

I'd like to experiment with having FreeBSD try to use less than the 
full 4 GB (e.g., to make it act as if memory ended at, say, 3 GB) 
but I'm not sure how to tell the kernel to do that.

--Brett Glass