RHEL 4 has kernel-devel that does not define empty_zero_page.

update 200901:

  In fact, it's easier to kmalloc and use our own zero page for now.
  I'm disabling this compat test.

update 201208:

  There has been get_zeroed_page since the initial git commit with
  2.6.12, so I'm using that instead.  Allocating two consecutive
  pages using kcalloc was resulting in us manipulating the _count
  of the second page, which was a tail of a compound page.  That
  resulted in panics on 3.4.y (LXOSI-68).