rmap: introduce rmap_walk_locked()

This patchset rewrites freeze_page() and unfreeze_page() using
try_to_unmap() and remove_migration_ptes().  Result is much simpler, but
somewhat slower.

Migration 8GiB worth of PMD-mapped THP:

  Baseline	20.21 +/- 0.393
  Patched	20.73 +/- 0.082
  Slowdown	1.03x

It's 3% slower, comparing to 14% in v1.  I don't it should be a stopper.

Splitting of PTE-mapped pages slowed more.  But this is not a common
case.

Migration 8GiB worth of PMD-mapped THP:

  Baseline	20.39 +/- 0.225
  Patched	22.43 +/- 0.496
  Slowdown	1.10x

rmap_walk_locked() is the same as rmap_walk(), but the caller takes care
of the relevant rmap lock.

This is preparation for switching THP splitting from custom rmap walk in
freeze_page()/unfreeze_page() to the generic one.

There is no support for KSM pages for now: not clear which lock is
implied.

Signed-off-by: Kirill A. Shutemov <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

include/linux/rmap.h

@@ -266,6 +266,7 @@ struct rmap_walk_control {
 };
 
 int rmap_walk(struct page *page, struct rmap_walk_control *rwc);
+int rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc);
 
 #else	/* !CONFIG_MMU */
 

mm/rmap.c

@@ -1715,14 +1715,21 @@ static struct anon_vma *rmap_walk_anon_lock(struct page *page,
  * vm_flags for that VMA.  That should be OK, because that vma shouldn't be
  * LOCKED.
  */
-static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
+static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc,
+		bool locked)
 {
 	struct anon_vma *anon_vma;
 	pgoff_t pgoff;
 	struct anon_vma_chain *avc;
 	int ret = SWAP_AGAIN;
 
-	anon_vma = rmap_walk_anon_lock(page, rwc);
+	if (locked) {
+		anon_vma = page_anon_vma(page);
+		/* anon_vma disappear under us? */
+		VM_BUG_ON_PAGE(!anon_vma, page);
+	} else {
+		anon_vma = rmap_walk_anon_lock(page, rwc);
+	}
 	if (!anon_vma)
 		return ret;
 
@@ -1742,7 +1749,9 @@ static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
 		if (rwc->done && rwc->done(page))
 			break;
 	}
-	anon_vma_unlock_read(anon_vma);
+
+	if (!locked)
+		anon_vma_unlock_read(anon_vma);
 	return ret;
 }
 
@@ -1759,9 +1768,10 @@ static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
  * vm_flags for that VMA.  That should be OK, because that vma shouldn't be
  * LOCKED.
  */
-static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
+static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc,
+		bool locked)
 {
-	struct address_space *mapping = page->mapping;
+	struct address_space *mapping = page_mapping(page);
 	pgoff_t pgoff;
 	struct vm_area_struct *vma;
 	int ret = SWAP_AGAIN;
@@ -1778,7 +1788,8 @@ static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
 		return ret;
 
 	pgoff = page_to_pgoff(page);
-	i_mmap_lock_read(mapping);
+	if (!locked)
+		i_mmap_lock_read(mapping);
 	vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
 		unsigned long address = vma_address(page, vma);
 
@@ -1795,7 +1806,8 @@ static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
 	}
 
 done:
-	i_mmap_unlock_read(mapping);
+	if (!locked)
+		i_mmap_unlock_read(mapping);
 	return ret;
 }
 
@@ -1804,9 +1816,20 @@ int rmap_walk(struct page *page, struct rmap_walk_control *rwc)
 	if (unlikely(PageKsm(page)))
 		return rmap_walk_ksm(page, rwc);
 	else if (PageAnon(page))
-		return rmap_walk_anon(page, rwc);
+		return rmap_walk_anon(page, rwc, false);
+	else
+		return rmap_walk_file(page, rwc, false);
+}
+
+/* Like rmap_walk, but caller holds relevant rmap lock */
+int rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc)
+{
+	/* no ksm support for now */
+	VM_BUG_ON_PAGE(PageKsm(page), page);
+	if (PageAnon(page))
+		return rmap_walk_anon(page, rwc, true);
 	else
-		return rmap_walk_file(page, rwc);
+		return rmap_walk_file(page, rwc, true);
 }
 
 #ifdef CONFIG_HUGETLB_PAGE