Re: [PATCH] selftests/mm: pagemap_scan ioctl: add PFN ZERO test cases

[David Hildenbrand]

On 30.06.25 12:24, Muhammad Usama Anjum wrote:
> Add test cases to test the correctness of PFN ZERO flag of pagemap_scan
> ioctl. Test with normal pages backed memory and huge pages backed
> memory.

Just to verify: would this trigger on kernels before my fix?

> Cc: David Hildenbrand <david@xxxxxxxxxx>
> Signed-off-by: Muhammad Usama Anjum <usama.anjum@xxxxxxxxxxxxx>
> ---
> The bug has been fixed [1].
>
> [1] https://lore.kernel.org/all/20250617143532.2375383-1-david@xxxxxxxxxx
> ---
>   tools/testing/selftests/mm/pagemap_ioctl.c | 57 +++++++++++++++++++++-
>   1 file changed, 56 insertions(+), 1 deletion(-)
>
> diff --git a/tools/testing/selftests/mm/pagemap_ioctl.c b/tools/testing/selftests/mm/pagemap_ioctl.c
> index 57b4bba2b45f3..6138de0087edf 100644
> --- a/tools/testing/selftests/mm/pagemap_ioctl.c
> +++ b/tools/testing/selftests/mm/pagemap_ioctl.c
> @@ -1,4 +1,5 @@
>   // SPDX-License-Identifier: GPL-2.0
> +
>   #define _GNU_SOURCE
>   #include <stdio.h>
>   #include <fcntl.h>
> @@ -1480,6 +1481,57 @@ static void transact_test(int page_size)
>   			      extra_thread_faults);
>   }
>   
> +void zeropfn_tests(void)
> +{
> +	unsigned long long mem_size;
> +	struct page_region vec;
> +	int i, ret;
> +	char *mem;
> +
> +	/* Test with page backed memory */

What is "page backed memory" ? :)

> +	mem_size = 10 * page_size;
> +	mem = mmap(NULL, mem_size, PROT_READ, MAP_PRIVATE | MAP_ANON, -1, 0);
> +	if (mem == MAP_FAILED)
> +		ksft_exit_fail_msg("error nomem\n");
> +
> +	/* Touch each page to ensure it's mapped */
> +	for (i = 0; i < mem_size; i += page_size)
> +		(void)((volatile char *)mem)[i];
> +
> +	ret = pagemap_ioctl(mem, mem_size, &vec, 1, 0,
> +			    (mem_size / page_size), PAGE_IS_PFNZERO, 0, 0, PAGE_IS_PFNZERO);
> +	if (ret < 0)
> +		ksft_exit_fail_msg("error %d %d %s\n", ret, errno, strerror(errno));
> +
> +	ksft_test_result(ret == 1 && LEN(vec) == (mem_size / page_size),
> +			 "%s all pages must have PFNZERO set\n", __func__);
> +
> +	munmap(mem, mem_size);
> +
> +	/* Test with huge page */
> +	mem_size = 10 * hpage_size;
> +	mem = memalign(hpage_size, mem_size);
> +	if (!mem)
> +		ksft_exit_fail_msg("error nomem\n");
> +
> +	ret = madvise(mem, mem_size, MADV_HUGEPAGE);
> +	if (ret)
> +		ksft_exit_fail_msg("madvise failed %d %s\n", errno, strerror(errno));

Might fail on older kernels, so we usually treat this as a skip.

> +
> +	for (i = 0; i < mem_size; i += hpage_size)
> +		(void)((volatile char *)mem)[i];
> +
> +	ret = pagemap_ioctl(mem, mem_size, &vec, 1, 0,
> +			    (mem_size / page_size), PAGE_IS_PFNZERO, 0, 0, PAGE_IS_PFNZERO);
> +	if (ret < 0)
> +		ksft_exit_fail_msg("error %d %d %s\n", ret, errno, strerror(errno));
> +
> +	ksft_test_result(ret == 1 && LEN(vec) == (mem_size / page_size),
> +			 "%s all huge pages must have PFNZERO set\n", __func__);

Wouldn't this be able to fail if 
/sys/kernel/mm/transparent_hugepage/use_zero_page is set to false, or if 
mmap() gave us a suboptimally-aligned range?

You'd have to read each and every page to get the ordinary shared 
zeropage in these configs instead without making the test too complicated.

> +
> +	free(mem);


Shouldn't this be an munmap() ?

> +}
> +
>   int main(int __attribute__((unused)) argc, char *argv[])
>   {
>   	int shmid, buf_size, fd, i, ret;
> @@ -1494,7 +1546,7 @@ int main(int __attribute__((unused)) argc, char *argv[])
>   	if (init_uffd())
>   		ksft_exit_pass();
>   
> -	ksft_set_plan(115);
> +	ksft_set_plan(117);

We should probably look into converting this test to kselftest_harness.

--
Cheers,

David / dhildenb