pmdp_get_lockless() should itself imply any ordering required.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 mm/hmm.c    |    1 -
 mm/vmscan.c |    3 ---
 2 files changed, 4 deletions(-)

--- a/mm/hmm.c
+++ b/mm/hmm.c
@@ -362,7 +362,6 @@ static int hmm_vma_walk_pmd(pmd_t *pmdp,
 		 * values.
 		 */
 		pmd = pmdp_get_lockless(pmdp);
-		barrier();
 		if (!pmd_devmap(pmd) && !pmd_trans_huge(pmd))
 			goto again;
 
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -4041,9 +4041,6 @@ static void walk_pmd_range(pud_t *pud, u
 	for (i = pmd_index(start), addr = start; addr != end; i++, addr = next) {
 		pmd_t val = pmdp_get_lockless(pmd + i);
 
-		/* for pmdp_get_lockless() */
-		barrier();
-
 		next = pmd_addr_end(addr, end);
 
 		if (!pmd_present(val) || is_huge_zero_pmd(val)) {

On Sat, Oct 22, 2022 at 5:48 AM Peter Zijlstra <peterz@infradead.org> wrote:
>
> pmdp_get_lockless() should itself imply any ordering required.

There are three remaining barriers that should be removed as well.