looks like you've done everything I did. I'm far from my working laptop today, so I can post my work on this issue only tomorrow so you can look through the code, compare and estimate solutions.
Regarding ARM build, I'm using xenalyze as both host tool (x86 build, e.g. for gnuplot scatterplot graphs) and target tool (ARM, for fast summary check); but since we're working with Xen 4.5, I'm building it outside Xen source tree with a custom Makefile.