On Sun, 28 Oct 2012 14:10:14 +0100
Ingo Molnar [off-list ref] wrote:

There's a related problem on s390: other THP implementations 
have pmd_mknotpresent() while s390 not, resulting in:

  mm/huge_memory.c:1543:2: error: implicit declaration of function 'pmd_mknotpresent'

The (untested!) patch below adds the s390 version of this 
method.

Gerald, Martin, did I get the S390 details right?

The upstream thp patches for s390 fixed that by adding pmdp_invalidate() and
replacing the pmd_mknotpresent() in mm/huge_memory.c, see git commit 46dcde73.
The pmdp_invalidate() is already included linux-next, so it should be ok to
just change mm/huge_memory.c similar to upstream:

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index f1c2679..842b6df 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c

@@ -1537,8 +1537,7 @@ static int __split_huge_page_map(struct page *page,
 	 * complete. The ptl also protects against concurrent faults due to
 	 * making the pmd not-present.
 	 */
-	set_pmd_at(mm, address, pmd, pmd_mknotpresent(*pmd));
-	flush_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
+	pmdp_invalidate(vma, address, pmd);
 	pmd_populate(mm, pmd, pgtable);
 	ret = 1;

With this, s390 does not need a pmd_mknotpresent(), and using
set_pmd_at(..., pmd_mknotpresent()) would be wrong on s390 anyway because
we need a flushing operation to change a valid pmd.