[RFC PATCH v2 2/3] mm: handle large folio when large folio in VM_LOCKED VMA range

[Yin Fengwei <fengwei.yin@intel.com>]

If large folio is in the range of VM_LOCKED VMA, it should be
mlocked to avoid being picked by page reclaim. Which may split
the large folio and then mlock each pages again.

Mlock this kind of large folio to prevent them being picked by
page reclaim.

For the large folio which cross the boundary of VM_LOCKED VMA,
we'd better not to mlock it. So if the system is under memory
pressure, this kind of large folio will be split and the pages
ouf of VM_LOCKED VMA can be reclaimed.

Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
---
 mm/internal.h | 11 ++++++++---
 mm/rmap.c     | 34 +++++++++++++++++++++++++++-------
 2 files changed, 35 insertions(+), 10 deletions(-)

diff --git a/mm/internal.h b/mm/internal.h
index c7dd15d8de3ef..776141de2797a 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -643,7 +643,8 @@ static inline void mlock_vma_folio(struct folio *folio,
 	 *    still be set while VM_SPECIAL bits are added: so ignore it then.
 	 */
 	if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED) &&
-	    (compound || !folio_test_large(folio)))
+	    (compound || !folio_test_large(folio) ||
+	    folio_in_range(folio, vma, vma->vm_start, vma->vm_end)))
 		mlock_folio(folio);
 }
 
@@ -651,8 +652,12 @@ void munlock_folio(struct folio *folio);
 static inline void munlock_vma_folio(struct folio *folio,
 			struct vm_area_struct *vma, bool compound)
 {
-	if (unlikely(vma->vm_flags & VM_LOCKED) &&
-	    (compound || !folio_test_large(folio)))
+	/*
+	 * To handle the case that a mlocked large folio is unmapped from VMA
+	 * piece by piece, allow munlock the large folio which is partially
+	 * mapped to VMA.
+	 */
+	if (unlikely(vma->vm_flags & VM_LOCKED))
 		munlock_folio(folio);
 }
 
diff --git a/mm/rmap.c b/mm/rmap.c
index 2668f5ea35342..455f415d8d9ca 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -803,6 +803,14 @@ struct folio_referenced_arg {
 	unsigned long vm_flags;
 	struct mem_cgroup *memcg;
 };
+
+static inline bool should_restore_mlock(struct folio *folio,
+		struct vm_area_struct *vma, bool pmd_mapped)
+{
+	return !folio_test_large(folio) ||
+			pmd_mapped || folio_within_vma(folio, vma);
+}
+
 /*
  * arg: folio_referenced_arg will be passed
  */
@@ -816,13 +824,25 @@ static bool folio_referenced_one(struct folio *folio,
 	while (page_vma_mapped_walk(&pvmw)) {
 		address = pvmw.address;
 
-		if ((vma->vm_flags & VM_LOCKED) &&
-		    (!folio_test_large(folio) || !pvmw.pte)) {
-			/* Restore the mlock which got missed */
-			mlock_vma_folio(folio, vma, !pvmw.pte);
-			page_vma_mapped_walk_done(&pvmw);
-			pra->vm_flags |= VM_LOCKED;
-			return false; /* To break the loop */
+		if (vma->vm_flags & VM_LOCKED) {
+			if (should_restore_mlock(folio, vma, !pvmw.pte)) {
+				/* Restore the mlock which got missed */
+				mlock_vma_folio(folio, vma, !pvmw.pte);
+				page_vma_mapped_walk_done(&pvmw);
+				pra->vm_flags |= VM_LOCKED;
+				return false; /* To break the loop */
+			} else {
+				/*
+				 * For large folio cross VMA boundaries, it's
+				 * expected to be picked  by page reclaim. But
+				 * should skip reference of pages which are in
+				 * the range of VM_LOCKED vma. As page reclaim
+				 * should just count the reference of pages out
+				 * the range of VM_LOCKED vma.
+				 */
+				pra->mapcount--;
+				continue;
+			}
 		}
 
 		if (pvmw.pte) {
-- 
2.39.2