From: Yijia Wang <wangyijia.yeah@bytedance.com>

check_file_mmap() faults a page in the middle of a file mapping and
expects the mmap read-around path to make neighbouring pages resident.
The test currently counts only pages after the faulted page.

That misses valid read-around on systems with large base page sizes. On
arm64 with 64K pages and the default 128K readahead setting, the
read-around window is two pages wide and centred on the faulting page.
Faulting page 32 makes pages 31 and 32 resident, so the forward-only
scan from page 33 reports ra_pages == 0 even though a neighbouring page
was brought in.

Keep the existing readahead assertion, but count resident neighbouring
pages on both sides of the faulted page. This fixes the 64K-page false
failure without teaching the selftest to compute the expected readahead
window from sysfs or other implementation details.

Signed-off-by: Yijia Wang <wangyijia.yeah@bytedance.com>
---
Changes in v4:
- Drop the sysfs read_ahead_kb helper and skip logic from v3.
- Keep the existing ra_pages assertion and count resident readahead
  pages on both sides of the faulted page.

 tools/testing/selftests/mincore/mincore_selftest.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/tools/testing/selftests/mincore/mincore_selftest.c b/tools/testing/selftests/mincore/mincore_selftest.c
index 3182043..ebf2b06 100644
--- a/tools/testing/selftests/mincore/mincore_selftest.c
+++ b/tools/testing/selftests/mincore/mincore_selftest.c
@@ -243,8 +243,10 @@ TEST(check_file_mmap)
 	}
 
 	/*
-	 * Touch a page in the middle of the mapping. We expect the next
-	 * few pages (the readahead window) to be populated too.
+	 * Touch a page in the middle of the mapping. We expect some
+	 * surrounding pages (the readahead window) to be populated too.
+	 * Depending on the page size and readahead setting, the pages may
+	 * land before the faulted page rather than after it.
 	 */
 	addr[FILE_SIZE / 2] = 1;
 	retval = mincore(addr, FILE_SIZE, vec);
@@ -253,6 +255,12 @@ TEST(check_file_mmap)
 		TH_LOG("Page not found in memory after use");
 	}
 
+	i = FILE_SIZE / 2 / page_size - 1;
+	while (i >= 0 && vec[i]) {
+		ra_pages++;
+		i--;
+	}
+
 	i = FILE_SIZE / 2 / page_size + 1;
 	while (i < vec_size && vec[i]) {
 		ra_pages++;
-- 
2.43.0

On Mon, 13 Jul 2026 17:43:19 +0800 王翊嘉 <wangyijia.yeah@bytedance.com> wrote:

> From: Yijia Wang <wangyijia.yeah@bytedance.com>
> 
> check_file_mmap() faults a page in the middle of a file mapping and
> expects the mmap read-around path to make neighbouring pages resident.
> The test currently counts only pages after the faulted page.
> 
> That misses valid read-around on systems with large base page sizes. On
> arm64 with 64K pages and the default 128K readahead setting, the
> read-around window is two pages wide and centred on the faulting page.
> Faulting page 32 makes pages 31 and 32 resident, so the forward-only
> scan from page 33 reports ra_pages == 0 even though a neighbouring page
> was brought in.
> 
> Keep the existing readahead assertion, but count resident neighbouring
> pages on both sides of the faulted page. This fixes the 64K-page false
> failure without teaching the selftest to compute the expected readahead
> window from sysfs or other implementation details.
> 

Thanks.

I continue to have bad feelings about this sefltest - checking on
internal and undocumented(?) behaviors of kernel-of-the-day. 
Pretending that these behaviors are ABI.

But oh well, if it causes problems we'll fix them.  And if we're to
test this at all, we may as well be more complete about it.

> --- a/tools/testing/selftests/mincore/mincore_selftest.c
> +++ b/tools/testing/selftests/mincore/mincore_selftest.c
> @@ -243,8 +243,10 @@ TEST(check_file_mmap)
>  	}
>  
>  	/*
> -	 * Touch a page in the middle of the mapping. We expect the next
> -	 * few pages (the readahead window) to be populated too.
> +	 * Touch a page in the middle of the mapping. We expect some
> +	 * surrounding pages (the readahead window) to be populated too.
> +	 * Depending on the page size and readahead setting, the pages may
> +	 * land before the faulted page rather than after it.
>  	 */
>  	addr[FILE_SIZE / 2] = 1;
>  	retval = mincore(addr, FILE_SIZE, vec);
> @@ -253,6 +255,12 @@ TEST(check_file_mmap)
>  		TH_LOG("Page not found in memory after use");
>  	}
>  
> +	i = FILE_SIZE / 2 / page_size - 1;
> +	while (i >= 0 && vec[i]) {
> +		ra_pages++;
> +		i--;
> +	}
> +
>  	i = FILE_SIZE / 2 / page_size + 1;
>  	while (i < vec_size && vec[i]) {
>  		ra_pages++;
> -- 
> 2.43.0