[RFC PATCH 6/6] WORKAROUND: Don't split large folios on madvise

[Yin Fengwei]

On 3/17/23 18:58, Ryan Roberts wrote:
> Signed-off-by: Ryan Roberts <ryan.roberts at arm.com>
> ---
>   mm/madvise.c | 8 ++++++++
>   1 file changed, 8 insertions(+)
> 
> diff --git a/mm/madvise.c b/mm/madvise.c
> index 340125d08c03..8fb84da744e1 100644
> --- a/mm/madvise.c
> +++ b/mm/madvise.c
> @@ -447,6 +447,7 @@ static int madvise_cold_or_pageout_pte_range(pmd_t *pmd,
>   		 * are sure it's worth. Split it if we are only owner.
>   		 */
>   		if (folio_test_large(folio)) {
> +#if 0
>   			if (folio_mapcount(folio) != 1)
>   				break;
>   			if (pageout_anon_only_filter && !folio_test_anon(folio))
> @@ -469,6 +470,9 @@ static int madvise_cold_or_pageout_pte_range(pmd_t *pmd,
>   			pte--;
>   			addr -= PAGE_SIZE;
>   			continue;
> +#else
> +			break;
> +#endif
>   		}
> 
>   		/*
> @@ -664,6 +668,7 @@ static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
>   		 * deactivate all pages.
>   		 */
>   		if (folio_test_large(folio)) {
> +#if 0
>   			if (folio_mapcount(folio) != 1)
>   				goto out;
>   			folio_get(folio);
> @@ -684,6 +689,9 @@ static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
>   			pte--;
>   			addr -= PAGE_SIZE;
>   			continue;
> +#else
> +			goto out;
> +#endif
>   		}
 From this workaround change, you hit an case that large folio has
1 as folio_mapcount()? Can you share the kernel crash log to me? Thanks.


Regards
Yin, Fengwei

> 
>   		if (folio_test_swapcache(folio) || folio_test_dirty(folio)) {
> --
> 2.25.1
>