[RFC PATCH 0/2] mm/gup: move pinned pages to unevictable list to avoid meaningless scan

[RFC PATCH 0/2] mm/gup: move pinned pages to unevictable list to avoid meaningless scan

[Jinjiang Tu]

The pages pinned by GUP interfaces are unable to be swapped out. Currently,
these pages are put into active/inactive list, leading to meaningless scan
when reclaiming memory.

Even worse, when we try to pin memory more than free memory, the system
fails to response for long time, all tasks are reclaiming memory. However,
the process takes long time to scan too much pinned pages, but only reclaim
few pages. Because each scan takes too much time, during this time, some
pages are swapped out again, some pages can be reclaimed again during next
scan. As a result, the OOM condition is not easy to meet.

I reproduce this problem by gup_test utility (CONFIG_GUP_TEST is needed).
The system has 3.3G free memory, and I pins 4G memory by command
'./gup_test -b -m 4096'. The system will fail to response for 10+ minutes
before triggering OOM.

To solve this problem, this patchset moves pinned pages to unevictable list
to avoid meaningless scanning during memory reclaim.

Jinjiang Tu (2):
  mm/gup: move pinned pages to unevictable list to avoid meaningless
    scan
  mm/gup: remove folio_maybe_dma_pinned() calls in shrink_page_list()

 mm/gup.c      | 13 +++++++++++++
 mm/internal.h |  2 +-
 mm/vmscan.c   | 12 ------------
 3 files changed, 14 insertions(+), 13 deletions(-)

-- 
2.25.1

[RFC PATCH 1/2] mm/gup: move pinned pages to unevictable list to avoid meaningless scan

[Jinjiang Tu]

To avoid meaningless scan during memory reclaim, this patch moves pinned
pages to unevictable list to avoid meaningless scan during memory reclaim.
Update folio_evictable() to check if the folio is pinned. If the folio is
pinned, the function returns false. The vmscan code will put the folio to
inevictable list when scanning the folio, so each pinned page will be
scanned at most once. When the folio is unpinned, check if the folio can
be put back to inactive/active list by calling folio_evictable().

Signed-off-by: Jinjiang Tu <tujinjiang@huawei.com>
---
 mm/gup.c      | 13 +++++++++++++
 mm/internal.h |  2 +-
 2 files changed, 14 insertions(+), 1 deletion(-)

diff --git a/mm/gup.c b/mm/gup.c
index 6e2f9e9d6537..ca82e1bc6826 100644
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -200,6 +200,19 @@ static void gup_put_folio(struct folio *folio, int refs, unsigned int flags)
 
 	if (!put_devmap_managed_page_refs(&folio->page, refs))
 		folio_put_refs(folio, refs);
+
+	if (folio_test_unevictable(folio) && folio_evictable(folio)) {
+		struct lruvec *lruvec = folio_lruvec_lock_irq(folio);
+
+		lruvec_del_folio(lruvec, folio);
+		folio_clear_unevictable(folio);
+		lruvec_add_folio(lruvec, folio);
+		folio_set_lru(folio);
+		__count_vm_events(UNEVICTABLE_PGRESCUED,
+			folio_nr_pages(folio));
+
+		unlock_page_lruvec_irq(lruvec);
+	}
 }
 
 /**
diff --git a/mm/internal.h b/mm/internal.h
index 8ed127c1c808..9ce158de40cf 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -154,7 +154,7 @@ static inline bool folio_evictable(struct folio *folio)
 	/* Prevent address_space of inode and swap cache from being freed */
 	rcu_read_lock();
 	ret = !mapping_unevictable(folio_mapping(folio)) &&
-			!folio_test_mlocked(folio);
+			!folio_test_mlocked(folio) && !folio_maybe_dma_pinned(folio);
 	rcu_read_unlock();
 	return ret;
 }
-- 
2.25.1

[RFC PATCH 2/2] mm/gup: remove folio_maybe_dma_pinned() calls in shrink_page_list()

[Jinjiang Tu]

Since the pinned pages have been put into inevictable list,
folio_maybe_dma_pinned() calls in shrink_page_list() could
be removed.

Signed-off-by: Jinjiang Tu <tujinjiang@huawei.com>
---
 mm/vmscan.c | 12 ------------
 1 file changed, 12 deletions(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 2fe4a11d63f4..590833c1fdbe 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -1891,8 +1891,6 @@ static unsigned int shrink_folio_list(struct list_head *folio_list,
 			if (!folio_test_swapcache(folio)) {
 				if (!(sc->gfp_mask & __GFP_IO))
 					goto keep_locked;
-				if (folio_maybe_dma_pinned(folio))
-					goto keep_locked;
 				if (folio_test_large(folio)) {
 					/* cannot split folio, skip it */
 					if (!can_split_folio(folio, NULL))
@@ -1959,16 +1957,6 @@ static unsigned int shrink_folio_list(struct list_head *folio_list,
 			}
 		}
 
-		/*
-		 * Folio is unmapped now so it cannot be newly pinned anymore.
-		 * No point in trying to reclaim folio if it is pinned.
-		 * Furthermore we don't want to reclaim underlying fs metadata
-		 * if the folio is pinned and thus potentially modified by the
-		 * pinning process as that may upset the filesystem.
-		 */
-		if (folio_maybe_dma_pinned(folio))
-			goto activate_locked;
-
 		mapping = folio_mapping(folio);
 		if (folio_test_dirty(folio)) {
 			/*
-- 
2.25.1