lmb_end_of_DRAM() returns the address of the end of RAM, not the starting
address of the last page.  We've been accessing mem_map[] out of bounds for
quite a while.  But, it's just a read, so it's probably never caused a real
problem.

But, during my port of CONFIG_NONLINEAR to ppc64, I have a check to make
sure that all __va() calls are given with valid physical addresses.  This
code tripped that check.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>