.. and thereby losing the point of doing it the old way.
The old way caused more warnigns, but more importantly, it caused warnings
on machines that didn't actually _have_ highmem pages. Which is actually
the big majority of them.
That was the whole (and only) point of the debugging! If you only test the
page address, you lose all the coverage!
It would be better if we actually passed in the gfp_flags, and then we
could test the __GFP_HIGH bit rather than the page address. But for now,
the rule is that GFP_ATOMIC and __GFP_ZERO do not work together, because
this sanity test currently cannot work for that case.
Linus
--
