Elsewhere, NR_SHMEM is updated at the same time as shmem NR_FILE_PAGES;
but shmem_replace_page() was forgetting to do that - so NR_SHMEM stats
could grow too big or too small, in those unusual cases when it's used.
Signed-off-by: Hugh Dickins <[email protected]>
---
This is not terribly important, and will clash with one of Matthew's
59 for 5.21; I don't mind if this gets delayed, and we just do it again
on top of his series later, or he fold the equivalent into his series;
but thought I'd better send it in as another fix to shmem_replace_page()
while that function is on our minds.
mm/shmem.c | 2 ++
1 file changed, 2 insertions(+)
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1659,7 +1659,9 @@ static int shmem_replace_page(struct page **pagep, gfp_t gfp,
new = page_folio(newpage);
mem_cgroup_migrate(old, new);
__inc_lruvec_page_state(newpage, NR_FILE_PAGES);
+ __inc_lruvec_page_state(newpage, NR_SHMEM);
__dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
+ __dec_lruvec_page_state(oldpage, NR_SHMEM);
}
xa_unlock_irq(&swap_mapping->i_pages);
On Wed, Aug 10, 2022 at 10:06:33PM -0700, Hugh Dickins wrote:
> Elsewhere, NR_SHMEM is updated at the same time as shmem NR_FILE_PAGES;
> but shmem_replace_page() was forgetting to do that - so NR_SHMEM stats
> could grow too big or too small, in those unusual cases when it's used.
>
> Signed-off-by: Hugh Dickins <[email protected]>
Reviewed-by: Matthew Wilcox (Oracle) <[email protected]>
> ---
> This is not terribly important, and will clash with one of Matthew's
> 59 for 5.21; I don't mind if this gets delayed, and we just do it again
> on top of his series later, or he fold the equivalent into his series;
> but thought I'd better send it in as another fix to shmem_replace_page()
> while that function is on our minds.
Let's get this into 6.0 since it's a bugfix, and I'll rebase my patches
for 6.1 on top of it.