Now for the good stuff! ;)

This one has had more changes...  I've changed the syscall from 
sys_membind to sys_mbind.  I liked Paolo's suggestion of aligning the 
naming.  I've fixed up the way the bitmaps are passed.  I pulled out all 
the ZONE_* code, and now just have it use all the zones on the node.  I 
made sure that the binding pointers are not compiled in for non-NUMA 
kernels.  All that is added for non-NUMA kernels is the cond_syscall and 
a small change in the page_cache_alloc callpath.  Now page_cache_alloc 
calls __page_cache_alloc, which is just the old page_cache_alloc for 
non-NUMA.  For NUMA, it's obviously a different function.  I also 
cleaned up the bitmask size issue, by just making sure userspace doesn't 
pass in a bitmask that's way too large.

I guess that's it...  As always, I'm looking forward to any comments!

Cheers!

-Matt