mm/swap: convert lru_deactivate to a folio_batch
authorMatthew Wilcox (Oracle) <willy@infradead.org>
Fri, 17 Jun 2022 17:50:04 +0000 (18:50 +0100)
committerakpm <akpm@linux-foundation.org>
Mon, 4 Jul 2022 01:08:46 +0000 (18:08 -0700)
Using folios instead of pages shrinks deactivate_page() and
lru_deactivate_fn() by 778 bytes between them.

Link: https://lkml.kernel.org/r/20220617175020.717127-7-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/swap.c

index 75c72d2..f23a549 100644 (file)
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -63,7 +63,7 @@ struct lru_pvecs {
        local_lock_t lock;
        struct folio_batch lru_add;
        struct folio_batch lru_deactivate_file;
-       struct pagevec lru_deactivate;
+       struct folio_batch lru_deactivate;
        struct pagevec lru_lazyfree;
 #ifdef CONFIG_SMP
        struct pagevec activate_page;
@@ -623,15 +623,15 @@ static void lru_deactivate_file_fn(struct lruvec *lruvec, struct folio *folio)
        }
 }
 
-static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec)
+static void lru_deactivate_fn(struct lruvec *lruvec, struct folio *folio)
 {
-       if (PageActive(page) && !PageUnevictable(page)) {
-               int nr_pages = thp_nr_pages(page);
+       if (folio_test_active(folio) && !folio_test_unevictable(folio)) {
+               long nr_pages = folio_nr_pages(folio);
 
-               del_page_from_lru_list(page, lruvec);
-               ClearPageActive(page);
-               ClearPageReferenced(page);
-               add_page_to_lru_list(page, lruvec);
+               lruvec_del_folio(lruvec, folio);
+               folio_clear_active(folio);
+               folio_clear_referenced(folio);
+               lruvec_add_folio(lruvec, folio);
 
                __count_vm_events(PGDEACTIVATE, nr_pages);
                __count_memcg_events(lruvec_memcg(lruvec), PGDEACTIVATE,
@@ -690,9 +690,9 @@ void lru_add_drain_cpu(int cpu)
        if (folio_batch_count(fbatch))
                folio_batch_move_lru(fbatch, lru_deactivate_file_fn);
 
-       pvec = &per_cpu(lru_pvecs.lru_deactivate, cpu);
-       if (pagevec_count(pvec))
-               pagevec_lru_move_fn(pvec, lru_deactivate_fn);
+       fbatch = &per_cpu(lru_pvecs.lru_deactivate, cpu);
+       if (folio_batch_count(fbatch))
+               folio_batch_move_lru(fbatch, lru_deactivate_fn);
 
        pvec = &per_cpu(lru_pvecs.lru_lazyfree, cpu);
        if (pagevec_count(pvec))
@@ -736,14 +736,16 @@ void deactivate_file_folio(struct folio *folio)
  */
 void deactivate_page(struct page *page)
 {
-       if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) {
-               struct pagevec *pvec;
+       struct folio *folio = page_folio(page);
 
+       if (folio_test_lru(folio) && folio_test_active(folio) &&
+           !folio_test_unevictable(folio)) {
+               struct folio_batch *fbatch;
+
+               folio_get(folio);
                local_lock(&lru_pvecs.lock);
-               pvec = this_cpu_ptr(&lru_pvecs.lru_deactivate);
-               get_page(page);
-               if (pagevec_add_and_need_flush(pvec, page))
-                       pagevec_lru_move_fn(pvec, lru_deactivate_fn);
+               fbatch = this_cpu_ptr(&lru_pvecs.lru_deactivate);
+               folio_batch_add_and_move(fbatch, folio, lru_deactivate_fn);
                local_unlock(&lru_pvecs.lock);
        }
 }
@@ -896,7 +898,7 @@ static inline void __lru_add_drain_all(bool force_all_cpus)
                if (folio_batch_count(&per_cpu(lru_pvecs.lru_add, cpu)) ||
                    data_race(folio_batch_count(&per_cpu(lru_rotate.fbatch, cpu))) ||
                    folio_batch_count(&per_cpu(lru_pvecs.lru_deactivate_file, cpu)) ||
-                   pagevec_count(&per_cpu(lru_pvecs.lru_deactivate, cpu)) ||
+                   folio_batch_count(&per_cpu(lru_pvecs.lru_deactivate, cpu)) ||
                    pagevec_count(&per_cpu(lru_pvecs.lru_lazyfree, cpu)) ||
                    need_activate_page_drain(cpu) ||
                    need_mlock_page_drain(cpu) ||