Re: [RFC PATCH] mm: vmscan: avoid anon scanning for GFP_NOIO with low swapcache in MGLRU

[Barry Song]

On Sat, Oct 10, 2026 at 10:11 PM Xiang Liu <liuxiang.277@xxxxxxxxxxxxx> wrote:
>
> Bo Zhang's GFP_NOIO patch [1] addressed anonymous scanning with low
> swapcache in the traditional LRU and left MGLRU as a TODO. This patch
> addresses that remaining case.
>
> Under GFP_NOIO, anonymous folios requiring swap IO cannot be reclaimed.
> When very few anonymous folios are already in swapcache, scanning anon
> can spend substantial time isolating and processing folios with little
> reclaim progress.
>
> Apply the existing can_reclaim_anon_pages() check to MGLRU.
> lruvec_is_sizable() excludes unavailable anon from its size estimate.
> In try_to_shrink_lruvec(), file-only scan_swappiness is passed to
> get_nr_to_scan(), should_run_aging() and evict_folios().
>
> The tradeoff is synchronous aging. With four anon generations and two
> file generations, advancing max_seq can require moving the oldest anon
> folios to the next generation. This patch accepts that cost to avoid
> low-yield anonymous scanning. It keeps the original swappiness in
> try_to_inc_max_seq(), avoiding additional use of the inc_min_seq()
> single-type shortcut that advances min_seq without processing the
> oldest folios and can cause cold/hot inversions.
>
> Tested against the mm-unstable base-commit below using dm-verity
> hash-block reads through dm-bufio, with unchanged NOIO allocation flags.
>
> Allocation time spans alloc_buffer() entry to return with valid buffer
> and data. Reclaim time is the time spent in try_to_free_pages() within
> that allocation. Eviction and aging are the time spent in evict_folios()
> and try_to_inc_max_seq() within that reclaim.
>
> 1. NOIO allocations under memory pressure
>
> The workload triggered NOIO reclaim during dm-verity reads, with
> generations allowed to evolve naturally. The table reports median
> times for successful allocations that entered direct reclaim.
>
> Time (ms, median)            Baseline  Patched
> ===========================  ========  =======
> Successful alloc_buffer()     181.196    0.465
> Direct reclaim                181.185    0.131
> Eviction within reclaim       179.147    0.091
>
> 2. Synchronous-aging cost
>
> Prepared four anon generations and two file generations, with zero
> swapcache and pages in the oldest file generation. The tests varied
> the oldest anon population to compare baseline eviction with patched
> synchronous aging.
>
> Each row reports a successful alloc_buffer() call and its reclaim
> costs. Oldest anon is the measured population of the oldest generation.
>
> Kernel    Oldest anon  Allocation   Reclaim  Eviction    Aging
>                 (MiB)        (ms)      (ms)      (ms)     (ms)
> ========  ===========  ==========  ========  ========  =======
> Baseline     1755.266     131.704   131.699   127.985    1.672
> Patched      1840.422      38.316    38.311     0.307   37.785
> Baseline     4835.418     681.796   681.766   675.477    4.382
> Patched      4813.180      91.161    91.140     0.450   90.407
> Baseline     8941.582    1474.245  1474.202  1469.084    0.488
> Patched      9012.660     143.748   143.709     1.026  142.341
> Baseline    13104.676    1098.372  1098.354  1087.976    2.307
> Patched     13046.621     139.450   139.390     1.083  137.954
> Baseline    17187.043    2694.909  2694.902  2686.470    0.784
> Patched     17124.902     209.240   209.233     0.638  208.270
>
> Across these tested sizes, synchronous aging was substantially cheaper
> than baseline eviction, and successful allocation time decreased.
>
> [1]
> https://patchew.org/linux/20260908062649.1045883-1-zhangbo56@xxxxxxxxxx/
>
> Signed-off-by: Xiang Liu <liuxiang.277@xxxxxxxxxxxxx>
> ---
>  mm/vmscan.c | 26 ++++++++++++++++----------
>  1 file changed, 16 insertions(+), 10 deletions(-)
>
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 29b31446fc11..550c16cad9f9 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -379,13 +379,6 @@ static inline bool reclaimable_anon_is_low(struct mem_cgroup *memcg,
>         if (!sc || (sc->gfp_mask & __GFP_IO))
>                 return false;
>
> -       /*
> -        * FIXME: MGLRU doesn't fully respect can_reclaim_anon_pages() for the
> -        * scanning type, so only apply this to the traditional LRU for now.
> -        */
> -       if (lru_gen_enabled())
> -               return false;
> -
>         if (memcg) {
>                 struct lruvec *lruvec = mem_cgroup_lruvec(memcg, pgdat);
>
> @@ -4357,6 +4350,13 @@ static bool lruvec_is_sizable(struct lruvec *lruvec, struct scan_control *sc)
>         unsigned long total;
>         int swappiness = get_swappiness(lruvec, sc);
>         struct mem_cgroup *memcg = lruvec_memcg(lruvec);
> +       int nid = lruvec_pgdat(lruvec)->node_id;
> +
> +       if (swappiness && !can_reclaim_anon_pages(memcg, nid, sc)) {
> +               if (swappiness == SWAPPINESS_ANON_ONLY)
> +                       return false;
> +               swappiness = MIN_SWAPPINESS;
> +       }

Hi Xiang,

Thanks for your patch.

I'm not a fan of this fix, as setting
scan_swappiness = MIN_SWAPPINESS will force a cold/hot
inversion for anonymous folios.

static bool inc_min_seq(struct lruvec *lruvec, int type, int
swappiness)
{
        ...

        /* For file type, skip the check if swappiness is anon only */
        if (type && (swappiness == SWAPPINESS_ANON_ONLY))
                goto done;

        /* For anon type, skip the check if swappiness is zero (file
only) */
        if (!type && !swappiness)
                goto done;
        ...
}

We should try to avoid this.

Best Regards
Barry