On Wed, 24 Feb 2016, Martin Schwidefsky wrote:

On Tue, 23 Feb 2016 22:33:45 +0300
"Kirill A. Shutemov" [off-list ref] wrote:

quoted

On Tue, Feb 23, 2016 at 07:19:07PM +0100, Gerald Schaefer wrote:

quoted

I'll check with Martin, maybe it is actually trivial, then we can
do a quick test it to rule that one out.

Oh. I found a bug in __split_huge_pmd_locked(). Although, not sure if it's
_the_ bug.

pmdp_invalidate() is called for the wrong address :-/
I guess that can be destructive on the architecture, right?

Could you check this?

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 1c317b85ea7d..4246bc70e55a 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c

@@ -2865,7 +2865,7 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
 	pgtable = pgtable_trans_huge_withdraw(mm, pmd);
 	pmd_populate(mm, &_pmd, pgtable);

-	for (i = 0; i < HPAGE_PMD_NR; i++, haddr += PAGE_SIZE) {
+	for (i = 0; i < HPAGE_PMD_NR; i++) {
 		pte_t entry, *pte;
 		/*
 		 * Note that NUMA hinting access restrictions are not

@@ -2886,9 +2886,9 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
 		}
 		if (dirty)
 			SetPageDirty(page + i);
-		pte = pte_offset_map(&_pmd, haddr);
+		pte = pte_offset_map(&_pmd, haddr + i * PAGE_SIZE);
 		BUG_ON(!pte_none(*pte));
-		set_pte_at(mm, haddr, pte, entry);
+		set_pte_at(mm, haddr + i * PAGE_SIZE, pte, entry);
 		atomic_inc(&page[i]._mapcount);
 		pte_unmap(pte);
 	}

@@ -2938,7 +2938,7 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
 	pmd_populate(mm, pmd, pgtable);

 	if (freeze) {
-		for (i = 0; i < HPAGE_PMD_NR; i++, haddr += PAGE_SIZE) {
+		for (i = 0; i < HPAGE_PMD_NR; i++) {
 			page_remove_rmap(page + i, false);
 			put_page(page + i);
 		}

Test is running and it looks good so far. For the final assessment I defer
to Gerald and Sebastian.

Yes, that one worked. My testsystem is doing make -j10 && make clean
in a loop since 4 hours now. Thanks!

Sebastian