Skip to contentSkip to conversation

mm: fix LRU balancing effect of new transparent huge pages

PATCH In review Shakeel Butt 2 participants 5 messages May 9, 2020 – May 11, 2020 Memory management

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm: fix LRU balancing effect of new transparent huge pages

From: Johannes Weiner <[email protected]>

Currently, THP are counted as single pages until they are split right before being swapped out. However, at that point the VM is already in the middle of reclaim, and adjusting the LRU balance then is useless.

Always account THP by the number of basepages, and remove the fixup from the splitting path.

---
Revived the patch from https://lore.kernel.org/patchwork/patch/685703/

 mm/swap.c | 23 +++++++++--------------
 1 file changed, 9 insertions(+), 14 deletions(-)
mm/swap.c +9 -14
index 4eb179ee0b72..b75c0ce90418 100644
@@ -262,14 +262,14 @@ void rotate_reclaimable_page(struct page *page)
 	}
 }
 
-static void update_page_reclaim_stat(struct lruvec *lruvec,
-				     int file, int rotated)
+static void update_page_reclaim_stat(struct lruvec *lruvec, int file,
+				     int rotated, int nr_pages)
 {
 	struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat;
 
-	reclaim_stat->recent_scanned[file]++;
+	reclaim_stat->recent_scanned[file] += nr_pages;
 	if (rotated)
-		reclaim_stat->recent_rotated[file]++;
+		reclaim_stat->recent_rotated[file] += nr_pages;
 }
 
 static void __activate_page(struct page *page, struct lruvec *lruvec,
@@ -288,7 +288,7 @@ static void __activate_page(struct page *page, struct lruvec *lruvec,
 
 		__count_vm_events(PGACTIVATE, nr_pages);
 		__count_memcg_events(lruvec_memcg(lruvec), PGACTIVATE, nr_pages);
-		update_page_reclaim_stat(lruvec, file, 1);
+		update_page_reclaim_stat(lruvec, file, 1, nr_pages);
 	}
 }
 
@@ -546,7 +546,7 @@ static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
 		__count_vm_events(PGDEACTIVATE, nr_pages);
 		__count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, nr_pages);
 	}
-	update_page_reclaim_stat(lruvec, file, 0);
+	update_page_reclaim_stat(lruvec, file, 0, nr_pages);
 }
 
 static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
@@ -564,7 +564,7 @@ static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec,
 
 		__count_vm_events(PGDEACTIVATE, nr_pages);
 		__count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE, nr_pages);
-		update_page_reclaim_stat(lruvec, file, 0);
+		update_page_reclaim_stat(lruvec, file, 0, nr_pages);
 	}
 }
 
@@ -590,7 +590,7 @@ static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec,
 
 		__count_vm_events(PGLAZYFREE, nr_pages);
 		__count_memcg_events(lruvec_memcg(lruvec), PGLAZYFREE, nr_pages);
-		update_page_reclaim_stat(lruvec, 1, 0);
+		update_page_reclaim_stat(lruvec, 1, 0, nr_pages);
 	}
 }
 
@@ -899,8 +899,6 @@ EXPORT_SYMBOL(__pagevec_release);
 void lru_add_page_tail(struct page *page, struct page *page_tail,
 		       struct lruvec *lruvec, struct list_head *list)
 {
-	const int file = 0;
-
 	VM_BUG_ON_PAGE(!PageHead(page), page);
 	VM_BUG_ON_PAGE(PageCompound(page_tail), page);
 	VM_BUG_ON_PAGE(PageLRU(page_tail), page);
@@ -926,9 +924,6 @@ void lru_add_page_tail(struct page *page, struct page *page_tail,
 		add_page_to_lru_list_tail(page_tail, lruvec,
 					  page_lru(page_tail));
 	}
-
-	if (!PageUnevictable(page))
-		update_page_reclaim_stat(lruvec, file, PageActive(page_tail));
 }
 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 
@@ -973,7 +968,7 @@ static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
 	if (page_evictable(page)) {
 		lru = page_lru(page);
 		update_page_reclaim_stat(lruvec, page_is_file_lru(page),
-					 PageActive(page));
+					 PageActive(page), nr_pages);
 		if (was_unevictable)
 			__count_vm_events(UNEVICTABLE_PGRESCUED, nr_pages);
 	} else {
Signature · 1 line
2.26.2.645.ge9eca65c58-goog
maintainer
#1 lore raw
Quoted 6 lines · On Sat, 9 May 2020 07:19:46 -0700 Shakeel Butt <[email protected]> wrote:

Currently, THP are counted as single pages until they are split right before being swapped out. However, at that point the VM is already in the middle of reclaim, and adjusting the LRU balance then is useless.

Always account THP by the number of basepages, and remove the fixup from the splitting path.

Confused. What kernel is this applicable to?

#2 lore raw
Quoted 11 lines · Andrew Morton
Quoted 6 lines · On Sat, 9 May 2020 07:19:46 -0700 Shakeel Butt <[email protected]> wrote:

Currently, THP are counted as single pages until they are split right before being swapped out. However, at that point the VM is already in the middle of reclaim, and adjusting the LRU balance then is useless.

Always account THP by the number of basepages, and remove the fixup from the splitting path.

Confused. What kernel is this applicable to?

It is still applicable to the latest Linux kernel. Basically lruvec->reclaim_stat->recent_[scanned|rotated] counters are used as heuristic in get_scan_count() to measure how much file and anon LRUs should be scanned by the current reclaim. Previously huge pages are treated as single page while updating the recent_[scanned|rotated] counters in swap.c while vmscan.c correctly updates them as huge pages.

maintainer
#3 lore raw
Quoted 8 lines · On Mon, 11 May 2020 14:38:23 -0700 Shakeel Butt <[email protected]> wrote:
Quoted 7 lines · Andrew Morton
Quoted 4 lines · On Sat, 9 May 2020 07:19:46 -0700 Shakeel Butt <[email protected]> wrote:

Currently, THP are counted as single pages until they are split right before being swapped out. However, at that point the VM is already in the middle of reclaim, and adjusting the LRU balance then is useless.

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

Always account THP by the number of basepages, and remove the fixup from the splitting path.

Confused. What kernel is this applicable to?

It is still applicable to the latest Linux kernel.

The patch has

mm/swap.c static void __activate_page(struct page *page, struct lruvec *lruvec, · line 293

Quoted 8 lines
mm/swap.c +1 -1
@@ -288,7 +288,7 @@ static void __activate_page(struct page *page, struct lruvec *lruvec,
 
 		__count_vm_events(PGACTIVATE, nr_pages);
 		__count_memcg_events(lruvec_memcg(lruvec), PGACTIVATE, nr_pages);
-		update_page_reclaim_stat(lruvec, file, 1);
+		update_page_reclaim_stat(lruvec, file, 1, nr_pages);
 	}
 }

but current mainline is quite different:

static void __activate_page(struct page *page, struct lruvec *lruvec,
void *arg)

{
	if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
		int file = page_is_file_lru(page);
		int lru = page_lru_base_type(page);

		del_page_from_lru_list(page, lruvec, lru);
		SetPageActive(page);
		lru += LRU_ACTIVE;
		add_page_to_lru_list(page, lruvec, lru);
		trace_mm_lru_activate(page);

		__count_vm_event(PGACTIVATE);
		update_page_reclaim_stat(lruvec, file, 1);
	}
}

q:/usr/src/linux-5.7-rc5> patch -p1 --dry-run < ~/x.txt checking file mm/swap.c
Hunk #2 FAILED at 288.
Hunk #3 FAILED at 546.
Hunk #4 FAILED at 564.
Hunk #5 FAILED at 590.
Hunk #6 succeeded at 890 (offset -9 lines). Hunk #7 succeeded at 915 (offset -9 lines). Hunk #8 succeeded at 958 with fuzz 2 (offset -10 lines). 4 out of 8 hunks FAILED

#4 lore raw
Quoted 49 lines · Andrew Morton
Quoted 8 lines · On Mon, 11 May 2020 14:38:23 -0700 Shakeel Butt <[email protected]> wrote:
Quoted 7 lines · Andrew Morton
Quoted 4 lines · On Sat, 9 May 2020 07:19:46 -0700 Shakeel Butt <[email protected]> wrote:

Currently, THP are counted as single pages until they are split right before being swapped out. However, at that point the VM is already in the middle of reclaim, and adjusting the LRU balance then is useless.

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

Always account THP by the number of basepages, and remove the fixup from the splitting path.

Confused. What kernel is this applicable to?

It is still applicable to the latest Linux kernel.

The patch has

mm/swap.c static void __activate_page(struct page *page, struct lruvec *lruvec, · line 293

Quoted 8 lines
mm/swap.c +1 -1
@@ -288,7 +288,7 @@ static void __activate_page(struct page *page, struct lruvec *lruvec,
 
              __count_vm_events(PGACTIVATE, nr_pages);
              __count_memcg_events(lruvec_memcg(lruvec), PGACTIVATE, nr_pages);
-             update_page_reclaim_stat(lruvec, file, 1);
+             update_page_reclaim_stat(lruvec, file, 1, nr_pages);
      }
 }

but current mainline is quite different:

static void __activate_page(struct page *page, struct lruvec *lruvec,
void *arg)

{
        if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) {
                int file = page_is_file_lru(page);
                int lru = page_lru_base_type(page);

                del_page_from_lru_list(page, lruvec, lru);
                SetPageActive(page);
                lru += LRU_ACTIVE;
                add_page_to_lru_list(page, lruvec, lru);
                trace_mm_lru_activate(page);

                __count_vm_event(PGACTIVATE);
                update_page_reclaim_stat(lruvec, file, 1);
        }
}
Quoted 11 lines

q:/usr/src/linux-5.7-rc5> patch -p1 --dry-run < ~/x.txt checking file mm/swap.c
Hunk #2 FAILED at 288.
Hunk #3 FAILED at 546.
Hunk #4 FAILED at 564.
Hunk #5 FAILED at 590.
Hunk #6 succeeded at 890 (offset -9 lines). Hunk #7 succeeded at 915 (offset -9 lines). Hunk #8 succeeded at 958 with fuzz 2 (offset -10 lines). 4 out of 8 hunks FAILED

Oh sorry my mistake. It is dependent on the first two patches at [1]. Basically I replaced the third patch of the series with this one. I should have re-send them all together.

[1] http://lkml.kernel.org/r/[email protected]