Re: [PATCH 2/2] selftests/mm: add tests for splitting pmd THPs to all lower orders.

[Zi Yan <ziy@nvidia.com>]

On 10 Jan 2025, at 18:50, Zi Yan wrote:

> Kernel already supports splitting a folio to any lower order. Test it.
>
> Signed-off-by: Zi Yan <ziy@nvidia.com>
> ---
>  tools/testing/selftests/mm/split_huge_page_test.c | 13 ++++++++-----
>  1 file changed, 8 insertions(+), 5 deletions(-)
>
> diff --git a/tools/testing/selftests/mm/split_huge_page_test.c b/tools/testing/selftests/mm/split_huge_page_test.c
> index 128004308233..3d3bc40a268b 100644
> --- a/tools/testing/selftests/mm/split_huge_page_test.c
> +++ b/tools/testing/selftests/mm/split_huge_page_test.c
> @@ -144,7 +144,7 @@ void split_pmd_zero_pages(void)
>  	free(one_page);
>  }
>
> -void split_pmd_thp(void)
> +void split_pmd_thp_to_order(int order)
>  {
>  	char *one_page;
>  	size_t len = 4 * pmd_pagesize;
> @@ -164,7 +164,7 @@ void split_pmd_thp(void)
>
>  	/* split all THPs */
>  	write_debugfs(PID_FMT, getpid(), (uint64_t)one_page,
> -		(uint64_t)one_page + len, 0);
> +		(uint64_t)one_page + len, order);
>
>  	for (i = 0; i < len; i++)
>  		if (one_page[i] != (char)i)
> @@ -174,7 +174,7 @@ void split_pmd_thp(void)
>  	if (!check_huge_anon(one_page, 0, pmd_pagesize))
>  		ksft_exit_fail_msg("Still AnonHugePages not split\n");
>
> -	ksft_test_result_pass("Split huge pages successful\n");
> +	ksft_test_result_pass("Split huge pages to order %d successful\n", order);
>  	free(one_page);
>  }
>
> @@ -481,7 +481,7 @@ int main(int argc, char **argv)
>  	if (argc > 1)
>  		optional_xfs_path = argv[1];
>
> -	ksft_set_plan(4+9);
> +	ksft_set_plan(1+9+2+9);
>
>  	pagesize = getpagesize();
>  	pageshift = ffs(pagesize) - 1;
> @@ -492,7 +492,10 @@ int main(int argc, char **argv)
>  	fd_size = 2 * pmd_pagesize;
>
>  	split_pmd_zero_pages();
> -	split_pmd_thp();
> +
> +	for (i = 0; i < 9; i++)
> +		split_pmd_thp_to_order(i);
> +
>  	split_pte_mapped_thp();
>  	split_file_backed_thp();
>

Hi Andrew,

Can you fold the patch below into this one? Anonymous memory does not
support order-1 folio, so there is no need to test splitting to order-1.
Kernel will WARN_ONCE for split_pmd_thp_to_order(1) test and the whole
test will fail at that point.

Thanks.

diff --git a/tools/testing/selftests/mm/split_huge_page_test.c b/tools/testing/selftests/mm/split_huge_page_test.c
index 3d3bc40a268b..4b47ab544a0c 100644
--- a/tools/testing/selftests/mm/split_huge_page_test.c
+++ b/tools/testing/selftests/mm/split_huge_page_test.c
@@ -494,7 +494,8 @@ int main(int argc, char **argv)
 	split_pmd_zero_pages();

 	for (i = 0; i < 9; i++)
-		split_pmd_thp_to_order(i);
+		if (i != 1)
+			split_pmd_thp_to_order(i);

 	split_pte_mapped_thp();
 	split_file_backed_thp();


--
Best Regards,
Yan, Zi