Re: [PATCH 1/3] mm, lru_gen: batch update counters on againg

[Yu Zhao]

On Fri, Dec 22, 2023 at 3:24 AM Kairui Song <ryncsn@xxxxxxxxx> wrote:
>
> From: Kairui Song <kasong@xxxxxxxxxxx>
>
> When lru_gen is aging, it will update mm counters page by page,
> which causes a higher overhead if age happens frequently or there
> are a lot of pages in one generation getting moved.
> Optimize this by doing the counter update in batch.
>
> Although most __mod_*_state has its own caches the overhead
> is still observable.
>
> Tested in a 4G memcg on a EPYC 7K62 with:
>
>   memcached -u nobody -m 16384 -s /tmp/memcached.socket \
>     -a 0766 -t 16 -B binary &
>
>   memtier_benchmark -S /tmp/memcached.socket \
>     -P memcache_binary -n allkeys \
>     --key-minimum=1 --key-maximum=16000000 -d 1024 \
>     --ratio=1:0 --key-pattern=P:P -c 2 -t 16 --pipeline 8 -x 6
>
> Average result of 18 test runs:
>
> Before: 44017.78 Ops/sec
> After:  44687.08 Ops/sec (+1.5%)
>
> Signed-off-by: Kairui Song <kasong@xxxxxxxxxxx>
> ---
>  mm/vmscan.c | 64 +++++++++++++++++++++++++++++++++++++++++++++--------
>  1 file changed, 55 insertions(+), 9 deletions(-)

Usually most reclaim activity happens in kswapd, e.g., from the
MongoDB benchmark (--duration=900):
pgscan_kswapd 11294317
pgscan_direct 128
And kswapd always has current->reclaim_state->mm_walk. So the
following should bring the vast majority of the improvement (assuming
it's not noise) with far less code change:

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 9dd8977de5a2..c06e00635d2b 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -3095,6 +3095,8 @@ static int folio_update_gen(struct folio *folio, int gen)
 static int folio_inc_gen(struct lruvec *lruvec, struct folio *folio,
bool reclaiming)
 {
        int type = folio_is_file_lru(folio);
+       struct lru_gen_mm_walk *walk = current->reclaim_state->mm_walk;
+
        struct lru_gen_folio *lrugen = &lruvec->lrugen;
        int new_gen, old_gen = lru_gen_from_seq(lrugen->min_seq[type]);
        unsigned long new_flags, old_flags = READ_ONCE(folio->flags);
@@ -3116,7 +3118,10 @@ static int folio_inc_gen(struct lruvec *lruvec,
struct folio *folio, bool reclai
                        new_flags |= BIT(PG_reclaim);
        } while (!try_cmpxchg(&folio->flags, &old_flags, new_flags));

-       lru_gen_update_size(lruvec, folio, old_gen, new_gen);
+       if (walk)
+               update_batch_size(walk, folio, old_gen, new_gen);
+       else
+               lru_gen_update_size(lruvec, folio, old_gen, new_gen);

        return new_gen;
 }
@@ -3739,6 +3744,8 @@ static void inc_max_seq(struct lruvec *lruvec,
bool can_swap, bool force_scan)
        int prev, next;
        int type, zone;
        struct lru_gen_folio *lrugen = &lruvec->lrugen;
+       struct lru_gen_mm_walk *walk = current->reclaim_state->mm_walk;
+
 restart:
        spin_lock_irq(&lruvec->lru_lock);

@@ -3758,6 +3765,9 @@ static void inc_max_seq(struct lruvec *lruvec,
bool can_swap, bool force_scan)
                goto restart;
        }

+       if (walk && walk->batched)
+               reset_batch_size(lruvec, walk);
+
        /*
         * Update the active/inactive LRU sizes for compatibility. Both sides of
         * the current max_seq need to be covered, since max_seq+1 can overlap