Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Barry Song <21cnbao@gmail.com>]

> >> I would appreciate if we leave the rmap out here.
> >>
> >> Can't we handle that when actually freeing the folio? folio_test_anon()
> >> is sticky until freed.
> >
> > To be clearer: we increment the counter when we set a folio anon, which
> > should indeed only happen in folio_add_new_anon_rmap(). We'll have to
> > ignore hugetlb here where we do it in hugetlb_add_new_anon_rmap().
> >
> > Then, when we free an anon folio we decrement the counter. (hugetlb
> > should clear the anon flag when an anon folio gets freed back to its
> > allocator -- likely that is already done).
> >
>
> Sorry that I am talking to myself: I'm wondering if we also have to
> adjust the counter when splitting a large folio to multiple
> smaller-but-still-large folios.

Hi David,

The conceptual code is shown below. Does this make more
sense to you? we have a line "mod_mthp_stat(new_order,
MTHP_STAT_NR_ANON, 1 << (order - new_order));"

@@ -3270,8 +3272,9 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 	struct deferred_split *ds_queue = get_deferred_split_queue(folio);
 	/* reset xarray order to new order after split */
 	XA_STATE_ORDER(xas, &folio->mapping->i_pages, folio->index, new_order);
-	struct anon_vma *anon_vma = NULL;
+	bool is_anon = folio_test_anon(folio);
 	struct address_space *mapping = NULL;
+	struct anon_vma *anon_vma = NULL;
 	int order = folio_order(folio);
 	int extra_pins, ret;
 	pgoff_t end;
@@ -3283,7 +3286,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 	if (new_order >= folio_order(folio))
 		return -EINVAL;
 
-	if (folio_test_anon(folio)) {
+	if (is_anon) {
 		/* order-1 is not supported for anonymous THP. */
 		if (new_order == 1) {
 			VM_WARN_ONCE(1, "Cannot split to order-1 folio");
@@ -3323,7 +3326,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 	if (folio_test_writeback(folio))
 		return -EBUSY;
 
-	if (folio_test_anon(folio)) {
+	if (is_anon) {
 		/*
 		 * The caller does not necessarily hold an mmap_lock that would
 		 * prevent the anon_vma disappearing so we first we take a
@@ -3437,6 +3440,10 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 			}
 		}
 
+		if (is_anon) {
+			mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
+			mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order));
+		}
 		__split_huge_page(page, list, end, new_order);
 		ret = 0;
 	} else {
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 408ef3d25cf5..c869d0601614 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -1039,6 +1039,7 @@ __always_inline bool free_pages_prepare(struct page *page,
 	bool skip_kasan_poison = should_skip_kasan_poison(page);
 	bool init = want_init_on_free();
 	bool compound = PageCompound(page);
+	bool anon = PageAnon(page);
 
 	VM_BUG_ON_PAGE(PageTail(page), page);
 
@@ -1130,6 +1131,9 @@ __always_inline bool free_pages_prepare(struct page *page,
 
 	debug_pagealloc_unmap_pages(page, 1 << order);
 
+	if (anon && compound)
+		mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
+
 	return true;
 }
 
diff --git a/mm/rmap.c b/mm/rmap.c
index 8d432051e970..982862cbf5ba 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma,
 	}
 
 	__folio_mod_stat(folio, nr, nr_pmdmapped);
+	mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
 }
 
 static __always_inline void __folio_add_file_rmap(struct folio *folio,