mm: add tlb_remove_check_page_size_change to track page size change

With commit e77b0852b551 ("mm/mmu_gather: track page size with mmu
gather and force flush if page size change") we added the ability to
force a tlb flush when the page size change in a mmu_gather loop.  We
did that by checking for a page size change every time we added a page
to mmu_gather for lazy flush/remove.  We can improve that by moving the
page size change check early and not doing it every time we add a page.

This also helps us to do tlb flush when invalidating a range covering
dax mapping.  Wrt dax mapping we don't have a backing struct page and
hence we don't call tlb_remove_page, which earlier forced the tlb flush
on page size change.  Moving the page size change check earlier means we
will do the same even for dax mapping.

We also avoid doing this check on architecture other than powerpc.

In a later patch we will remove page size check from tlb_remove_page().

Link: http://lkml.kernel.org/r/20161026084839.27299-5-aneesh.kumar@linux.vnet.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

4 files changed, 16 insertions, 1 deletions

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 0103728ffa94..26fd1161ca85 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1323,6 +1323,8 @@ bool madvise_free_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
 	struct mm_struct *mm = tlb->mm;
 	bool ret = false;
 
+	tlb_remove_check_page_size_change(tlb, HPAGE_PMD_SIZE);
+
 	ptl = pmd_trans_huge_lock(pmd, vma);
 	if (!ptl)
 		goto out_unlocked;
@@ -1384,6 +1386,8 @@ int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
 	pmd_t orig_pmd;
 	spinlock_t *ptl;
 
+	tlb_remove_check_page_size_change(tlb, HPAGE_PMD_SIZE);
+
 	ptl = __pmd_trans_huge_lock(pmd, vma);
 	if (!ptl)
 		return 0;
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 8e519da7242d..3edb759c5c7d 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3286,6 +3286,11 @@ void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
 	BUG_ON(start & ~huge_page_mask(h));
 	BUG_ON(end & ~huge_page_mask(h));
 
+	/*
+	 * This is a hugetlb vma, all the pte entries should point
+	 * to huge page.
+	 */
+	tlb_remove_check_page_size_change(tlb, sz);
 	tlb_start_vma(tlb, vma);
 	mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
 	address = start;
diff --git a/mm/madvise.c b/mm/madvise.c
index 93fb63e88b5e..0e3828eae9f8 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -281,6 +281,7 @@ static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
 	if (pmd_trans_unstable(pmd))
 		return 0;
 
+	tlb_remove_check_page_size_change(tlb, PAGE_SIZE);
 	orig_pte = pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
 	arch_enter_lazy_mmu_mode();
 	for (; addr != end; pte++, addr += PAGE_SIZE) {
diff --git a/mm/memory.c b/mm/memory.c
index d86b7b4afd7d..eae20eb66bfc 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -528,7 +528,11 @@ void free_pgd_range(struct mmu_gather *tlb,
 		end -= PMD_SIZE;
 	if (addr > end - 1)
 		return;
-
+	/*
+	 * We add page table cache pages with PAGE_SIZE,
+	 * (see pte_free_tlb()), flush the tlb if we need
+	 */
+	tlb_remove_check_page_size_change(tlb, PAGE_SIZE);
 	pgd = pgd_offset(tlb->mm, addr);
 	do {
 		next = pgd_addr_end(addr, end);
@@ -1120,6 +1124,7 @@ static unsigned long zap_pte_range(struct mmu_gather *tlb,
 	swp_entry_t entry;
 	struct page *pending_page = NULL;
 
+	tlb_remove_check_page_size_change(tlb, PAGE_SIZE);
 again:
 	init_rss_vec(rss);
 	start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);