2024-06-12 05:10:54

by Hugh Dickins

Subject: [PATCH unstable] mm: rmap: abstract updating per-node and per-memcg stats fix

/proc/meminfo is showing ridiculously large numbers on some lines:
__folio_remove_rmap()'s __folio_mod_stat() should be subtracting!

Signed-off-by: Hugh Dickins <[email protected]>
---
A fix for folding into mm-unstable, not needed for 6.10-rc.

mm/rmap.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1567,7 +1567,7 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,
list_empty(&folio->_deferred_list))
deferred_split_folio(folio);
}
- __folio_mod_stat(folio, nr, nr_pmdmapped);
+ __folio_mod_stat(folio, -nr, -nr_pmdmapped);

/*
* It would be tidy to reset folio_test_anon mapping when fully
--
2.35.3

2024-06-12 06:53:38

by Yosry Ahmed

Subject: Re: [PATCH unstable] mm: rmap: abstract updating per-node and per-memcg stats fix

On Tue, Jun 11, 2024 at 10:10 PM Hugh Dickins <[email protected]> wrote:
>
> /proc/meminfo is showing ridiculously large numbers on some lines:
> __folio_remove_rmap()'s __folio_mod_stat() should be subtracting!
>
> Signed-off-by: Hugh Dickins <[email protected]>

Reviewed-by: Yosry Ahmed <[email protected]>

Thanks a lot for fixing this! I was just looking at a test failure
report by the kernel robot caused by this [1].

Just to document my own stupidity here:

1. In [2], I sent a fix to use __mod_node_page_state() instead of
__lruvec_stat_mod_folio() in __folio_remove_rmap(). I made the same
mistake of replacing subtraction with addition.

2. In [3], I sent a v2 of that fix that kept the subtraction in
__folio_remove_rmap() correctly.

3. In [4], I sent a cleanup on top of the fix, and that cleanup
replaced the subtraction in __folio_remove_rmap() with an addition,
again.

Apparently, I just suck at subtraction :)

[1]https://lore.kernel.org/linux-mm/[email protected]/
[2]https://lore.kernel.org/lkml/[email protected]/
[3]https://lore.kernel.org/lkml/[email protected]/
[4]https://lore.kernel.org/lkml/[email protected]/

> ---
> A fix for folding into mm-unstable, not needed for 6.10-rc.
>
> mm/rmap.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1567,7 +1567,7 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,
> list_empty(&folio->_deferred_list))
> deferred_split_folio(folio);
> }
> - __folio_mod_stat(folio, nr, nr_pmdmapped);
> + __folio_mod_stat(folio, -nr, -nr_pmdmapped);
>
> /*
> * It would be tidy to reset folio_test_anon mapping when fully
> --
> 2.35.3
>

2024-06-12 07:26:17

by David Hildenbrand

Subject: Re: [PATCH unstable] mm: rmap: abstract updating per-node and per-memcg stats fix

On 12.06.24 07:10, Hugh Dickins wrote:
> /proc/meminfo is showing ridiculously large numbers on some lines:
> __folio_remove_rmap()'s __folio_mod_stat() should be subtracting!
>
> Signed-off-by: Hugh Dickins <[email protected]>
> ---
> A fix for folding into mm-unstable, not needed for 6.10-rc.
>
> mm/rmap.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1567,7 +1567,7 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,
> list_empty(&folio->_deferred_list))
> deferred_split_folio(folio);
> }
> - __folio_mod_stat(folio, nr, nr_pmdmapped);
> + __folio_mod_stat(folio, -nr, -nr_pmdmapped);
>
> /*
> * It would be tidy to reset folio_test_anon mapping when fully

Missed that detail, thanks!

Acked-by: David Hildenbrand <[email protected]>

--
Cheers,

David / dhildenb