2023-06-19 10:42:35

by Jan Glauber

Subject: [PATCH v2] mm: Fix shmem THP counters on migration

The per node numa_stat values for shmem don't change on
page migration for THP:

grep shmem /sys/fs/cgroup/machine.slice/.../memory.numa_stat:

shmem N0=1092616192 N1=10485760
shmem_thp N0=1092616192 N1=10485760

migratepages 9181 0 1:

shmem N0=0 N1=1103101952
shmem_thp N0=1092616192 N1=10485760

Fix that by updating shmem_thp counters likewise to shmem counters
on page migration.

Signed-off-by: Jan Glauber <[email protected]>
---
mm/migrate.c | 5 +++++
1 file changed, 5 insertions(+)

diff --git a/mm/migrate.c b/mm/migrate.c
index 01cac26a3127..d2ba786ea105 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -492,6 +492,11 @@ int folio_migrate_mapping(struct address_space *mapping,
if (folio_test_swapbacked(folio) && !folio_test_swapcache(folio)) {
__mod_lruvec_state(old_lruvec, NR_SHMEM, -nr);
__mod_lruvec_state(new_lruvec, NR_SHMEM, nr);
+
+ if (folio_test_transhuge(folio)) {
+ __mod_lruvec_state(old_lruvec, NR_SHMEM_THPS, -nr);
+ __mod_lruvec_state(new_lruvec, NR_SHMEM_THPS, nr);
+ }
}
#ifdef CONFIG_SWAP
if (folio_test_swapcache(folio)) {
--
2.25.1

2023-06-21 07:29:33

by Baolin Wang

Subject: Re: [PATCH v2] mm: Fix shmem THP counters on migration

Cc: Huang Ying

On 6/19/2023 6:33 PM, Jan Glauber wrote:
> The per node numa_stat values for shmem don't change on
> page migration for THP:
>
> grep shmem /sys/fs/cgroup/machine.slice/.../memory.numa_stat:
>
> shmem N0=1092616192 N1=10485760
> shmem_thp N0=1092616192 N1=10485760
>
> migratepages 9181 0 1:
>
> shmem N0=0 N1=1103101952
> shmem_thp N0=1092616192 N1=10485760
>
> Fix that by updating shmem_thp counters likewise to shmem counters
> on page migration.
>
> Signed-off-by: Jan Glauber <[email protected]>
> ---

Please add your change history.

> mm/migrate.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/mm/migrate.c b/mm/migrate.c
> index 01cac26a3127..d2ba786ea105 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -492,6 +492,11 @@ int folio_migrate_mapping(struct address_space *mapping,
> if (folio_test_swapbacked(folio) && !folio_test_swapcache(folio)) {
> __mod_lruvec_state(old_lruvec, NR_SHMEM, -nr);
> __mod_lruvec_state(new_lruvec, NR_SHMEM, nr);
> +
> + if (folio_test_transhuge(folio)) {

I am afraid this validation is fragile, IIUC the file backed folio can
contain various numbers of pages in future.

So seems using folio_test_pmd_mappable() seems more suitable for THP.

> + __mod_lruvec_state(old_lruvec, NR_SHMEM_THPS, -nr);
> + __mod_lruvec_state(new_lruvec, NR_SHMEM_THPS, nr);
> + }
> }
> #ifdef CONFIG_SWAP
> if (folio_test_swapcache(folio)) {
> --
> 2.25.1
>