Build the phys_to_machine_mapping array in Xen rather than

Build the phys_to_machine_mapping array in Xen rather than
reusing the 1:1 page table built by the builder, and it also establishes
the 1:1 direct mapping at runtime a la shadow page tables. Since the
builder constructs the full 1:1 direct mapping (but <4GB), the current
implementation wastes memory but cannot support guests with >=4GB
memory (even for x64-64).

This is also required for HVM support on the PAE host. That patch
should be sent soon.

For SVM, I think the svm code needs changes. Please look at the changes
to vmx.c and vmcs.c; should be straightforward.

Signed-off-by: Jun Nakajima <jun.nakajima@intel.com>
Signed-off-by: Xiaohui Xin <xiaohui.xin@intel.com>