Re: [PATCH] mm/vmscan: fix delayed flusher wakeup in MGLRU

[Kairui Song]

On Thu, Apr 30, 2026 at 2:55 AM Vineet Agarwal
<agarwal.vineet2006@xxxxxxxxx> wrote:
>
> MGLRU currently decides whether to wake flusher threads in
> try_to_shrink_lruvec() using cumulative reclaim counters:
>
> sc->nr.unqueued_dirty == sc->nr.file_taken
>
> However, these counters are accumulated across multiple evict_folios()
> passes before the check is performed.
>
> This can delay or suppress flusher wakeup when an earlier reclaim batch
> isolates only dirty file folios, but a later batch isolates clean file
> folios before try_to_shrink_lruvec() performs the final comparison.
>
> For example:
>
> batch 1: file_taken = 100, unqueued_dirty = 100
> batch 2: file_taken += 60, unqueued_dirty += 0
>
> Final check becomes 100 != 160 and flusher wakeup is skipped, even
> though reclaim was already blocked by dirty file folios in batch 1.
>
> Classic reclaim avoids this by using per-batch values:
>
> stat.nr_unqueued_dirty == nr_taken
>
> and waking flushers immediately when the condition is met.
>
> Make MGLRU use the same per-batch flusher wakeup behavior as classic
> reclaim by moving the flusher wakeup into evict_folios(), using
> batch-local isolation results from scan_folios() instead of the
> cumulative counters checked later in try_to_shrink_lruvec().
>
> This avoids missed flusher wakeups and makes dirty folio reclaim
> behavior consistent with classic reclaim.
>
> Fixes: 1bc542c6a0d14 ("mm/vmscan: wake up flushers conditionally to avoid cgroup OOM")
> Signed-off-by: Vineet Agarwal <agarwal.vineet2006@xxxxxxxxx>
> ---
>  mm/vmscan.c | 46 ++++++++++++++++++++--------------------------
>  1 file changed, 20 insertions(+), 26 deletions(-)
>
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index bd1b1aa12581..f9b6cc146a3d 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -4680,7 +4680,8 @@ static bool isolate_folio(struct lruvec *lruvec, struct folio *folio, struct sca
>
>  static int scan_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>                        struct scan_control *sc, int type, int tier,
> -                      struct list_head *list)
> +                      struct list_head *list,
> +                      unsigned long *file_taken)
>  {
>         int i;
>         int gen;
> @@ -4749,7 +4750,7 @@ static int scan_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>                                 scanned, skipped, isolated,
>                                 type ? LRU_INACTIVE_FILE : LRU_INACTIVE_ANON);
>         if (type == LRU_GEN_FILE)
> -               sc->nr.file_taken += isolated;
> +               *file_taken += isolated;
>         /*
>          * There might not be eligible folios due to reclaim_idx. Check the
>          * remaining to prevent livelock if it's not making progress.
> @@ -4798,7 +4799,8 @@ static int get_type_to_scan(struct lruvec *lruvec, int swappiness)
>
>  static int isolate_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>                           struct scan_control *sc, int swappiness,
> -                         int *type_scanned, struct list_head *list)
> +                         int *type_scanned, struct list_head *list,
> +                         unsigned long *file_taken)
>  {
>         int i;
>         int type = get_type_to_scan(lruvec, swappiness);
> @@ -4809,7 +4811,8 @@ static int isolate_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>
>                 *type_scanned = type;
>
> -               scanned = scan_folios(nr_to_scan, lruvec, sc, type, tier, list);
> +               scanned = scan_folios(nr_to_scan, lruvec, sc, type, tier,
> +                                     list, file_taken);
>                 if (scanned)
>                         return scanned;
>
> @@ -4825,6 +4828,7 @@ static int evict_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>         int type;
>         int scanned;
>         int reclaimed;
> +       unsigned long file_taken = 0;
>         LIST_HEAD(list);
>         LIST_HEAD(clean);
>         struct folio *folio;
> @@ -4839,8 +4843,8 @@ static int evict_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>
>         lruvec_lock_irq(lruvec);
>
> -       scanned = isolate_folios(nr_to_scan, lruvec, sc, swappiness, &type, &list);
> -
> +       scanned = isolate_folios(nr_to_scan, lruvec, sc, swappiness,
> +                                &type, &list, &file_taken);
>         scanned += try_to_inc_min_seq(lruvec, swappiness);
>
>         if (evictable_min_seq(lrugen->min_seq, swappiness) + MIN_NR_GENS > lrugen->max_seq)
> @@ -4852,6 +4856,14 @@ static int evict_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>                 return scanned;
>  retry:
>         reclaimed = shrink_folio_list(&list, pgdat, sc, &stat, false, memcg);
> +
> +       if (stat.nr_unqueued_dirty && stat.nr_unqueued_dirty == file_taken) {
> +               wakeup_flusher_threads(WB_REASON_VMSCAN);
> +
> +               if (!writeback_throttling_sane(sc))
> +                       reclaim_throttle(pgdat, VMSCAN_THROTTLE_WRITEBACK);
> +       }
> +       sc->nr.file_taken += file_taken;
>         sc->nr.unqueued_dirty += stat.nr_unqueued_dirty;
>         sc->nr_reclaimed += reclaimed;
>         trace_mm_vmscan_lru_shrink_inactive(pgdat->node_id,
> @@ -5021,27 +5033,9 @@ static bool try_to_shrink_lruvec(struct lruvec *lruvec, struct scan_control *sc)
>         }
>
>         /*
> -        * If too many file cache in the coldest generation can't be evicted
> -        * due to being dirty, wake up the flusher.
> +        * Flusher wakeup and writeback throttling are handled in
> +        * evict_folios() based on per-batch reclaim results.
>          */
> -       if (sc->nr.unqueued_dirty && sc->nr.unqueued_dirty == sc->nr.file_taken) {
> -               struct pglist_data *pgdat = lruvec_pgdat(lruvec);
> -
> -               wakeup_flusher_threads(WB_REASON_VMSCAN);
> -
> -               /*
> -                * For cgroupv1 dirty throttling is achieved by waking up
> -                * the kernel flusher here and later waiting on folios
> -                * which are in writeback to finish (see shrink_folio_list()).
> -                *
> -                * Flusher may not be able to issue writeback quickly
> -                * enough for cgroupv1 writeback throttling to work
> -                * on a large system.
> -                */
> -               if (!writeback_throttling_sane(sc))
> -                       reclaim_throttle(pgdat, VMSCAN_THROTTLE_WRITEBACK);
> -       }
> -
>         /* whether this lruvec should be rotated */
>         return nr_to_scan < 0;
>  }
> --
> 2.54.0
>
>

Hi, Vineet,

Thanks for the patch, I got what you mean, but even with this patch,
the wakeup and flush logic of MGLRU is still very different from
classical LRU and doesn't perform well. No throttling, the batch is
still over large, and it could over aggressively wake up the flusher
as the folios are stuck at the tail.

There are a few more issues related, so there is already a ongoing
work to fix them properly, and make them more unified:
https://lore.kernel.org/linux-mm/20260428-mglru-reclaim-v7-0-02fabb92dc43@xxxxxxxxxxx/

Global throttling or isolation throttling are still missing after
that, I think maybe we can focus on that part after that series. How
do you think?