Memory barrier is needed after setting LRU bit, but smp_mb() is too
strong. Some architectures, i.e. x86, imply memory barrier with atomic
operations, so replacing it with smp_mb__after_atomic() sounds better,
which is nop on strong ordered machines, and full memory barriers on
others. With this change the vm-calability cases would perform better
on x86, I saw total 6% improvement with this patch and previous inline
fix.
The test data (lru-file-readtwice throughput) against v5.6-rc4:
mainline w/ inline fix w/ both (adding this)
150MB 154MB 159MB
Fixes: 9c4e6b1a7027 ("mm, mlock, vmscan: no more skipping pagevecs")
Cc: Shakeel Butt <shakeelb@xxxxxxxxxx>
Cc: Vlastimil Babka <vbabka@xxxxxxx>
Signed-off-by: Yang Shi <yang.shi@xxxxxxxxxxxxxxxxx>
---
mm/swap.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/mm/swap.c b/mm/swap.c
index cf39d24..118bac4 100644
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -945,20 +945,20 @@ static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
* #0: __pagevec_lru_add_fn #1: clear_page_mlock
*
* SetPageLRU() TestClearPageMlocked()
- * smp_mb() // explicit ordering // above provides strict
+ * MB() // explicit ordering // above provides strict
* // ordering
* PageMlocked() PageLRU()
*
*
* if '#1' does not observe setting of PG_lru by '#0' and fails
* isolation, the explicit barrier will make sure that page_evictable
- * check will put the page in correct LRU. Without smp_mb(), SetPageLRU
+ * check will put the page in correct LRU. Without MB(), SetPageLRU
* can be reordered after PageMlocked check and can make '#1' to fail
* the isolation of the page whose Mlocked bit is cleared (#0 is also
* looking at the same page) and the evictable page will be stranded
* in an unevictable LRU.
*/
- smp_mb();
+ smp_mb__after_atomic();
if (page_evictable(page)) {
lru = page_lru(page);
--
1.8.3.1
