[PATCH] selftests/mm: pkeys: wait for the child in become_child()

[PATCH] selftests/mm: pkeys: wait for the child in become_child()

[Carlos López]

become_child() forks during test_pkey_alloc_exhaust() and has the parent
exit, so that the remaining tests run in the child. However, the test
runner only waits for the original process, so it sees the test finish
early with a success status while the remaining tests keep running in
the background. Failures after the first fork are thus not reported in
the exit status, and the remaining tests overlap with the next selftest.

Fix this by waiting for the child in the parent and exiting with its
status. A previous fix avoided waiting, arguing that the parents would
keep holding their resources [1]. Particularly, the parent keeps the
hugetlb reservation of its test memory, so the child runs out of hugetlb
pages after a few iterations. Avoid this by freeing the test memory
before waiting.

Link: https://lore.kernel.org/all/20240419115027.3848958-1-usama.anjum@collabora.com [1]
Fixes: e1812933b17b ("x86/selftests/pkeys: Fork() to check for state being preserved")
Signed-off-by: Carlos López <clopez@suse.de>
---
 tools/testing/selftests/mm/protection_keys.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/tools/testing/selftests/mm/protection_keys.c b/tools/testing/selftests/mm/protection_keys.c
index b7882ab97683..5dece144ab95 100644
--- a/tools/testing/selftests/mm/protection_keys.c
+++ b/tools/testing/selftests/mm/protection_keys.c
@@ -1009,9 +1009,10 @@ static void test_pkey_syscalls_bad_args(int *ptr, u16 pkey)
 	pkey_assert(err);
 }
 
-static void become_child(void)
+static void become_child(int *ptr)
 {
 	pid_t forkret;
+	int status;
 
 	forkret = fork();
 	pkey_assert(forkret >= 0);
@@ -1021,7 +1022,16 @@ static void become_child(void)
 		/* in the child */
 		return;
 	}
-	_exit(0);
+
+	/*
+	 * Wait for the child to finish the tests and pass on its result.
+	 * Free the test memory first, as it may reserve pages from the
+	 * small hugetlb pool that the child needs.
+	 */
+	free_pkey_malloc(ptr);
+	if (waitpid(forkret, &status, 0) != forkret || !WIFEXITED(status))
+		_exit(KSFT_FAIL);
+	_exit(WEXITSTATUS(status));
 }
 
 /* Assumes that all pkeys other than 'pkey' are unallocated */
@@ -1060,7 +1070,7 @@ static void test_pkey_alloc_exhaust(int *ptr, u16 pkey)
 		 * preserved across fork().
 		 */
 		if (i == NR_PKEYS*2)
-			become_child();
+			become_child(ptr);
 	}
 
 	dprintf3("%s()::%d\n", __func__, __LINE__);

base-commit: 843c58022865a7b0f9743416c2b48bc2632da23c
-- 
2.51.0

Re: [PATCH] selftests/mm: pkeys: wait for the child in become_child()

[Andrew Morton]

On Fri,  6 Nov 2026 17:46:39 +0100 Carlos López <clopez@suse.de> wrote:

> become_child() forks during test_pkey_alloc_exhaust() and has the parent
> exit, so that the remaining tests run in the child. However, the test
> runner only waits for the original process, so it sees the test finish
> early with a success status while the remaining tests keep running in
> the background. Failures after the first fork are thus not reported in
> the exit status, and the remaining tests overlap with the next selftest.
> 
> Fix this by waiting for the child in the parent and exiting with its
> status. A previous fix avoided waiting, arguing that the parents would
> keep holding their resources [1]. Particularly, the parent keeps the
> hugetlb reservation of its test memory, so the child runs out of hugetlb
> pages after a few iterations. Avoid this by freeing the test memory
> before waiting.

Thanks.

>  tools/testing/selftests/mm/protection_keys.c | 16 +++++++++++++---

get_maintainer coverage of selftests is poor, alas.  So it's a good
idea to poke around in the git-blame output to figure out who best to
cc.  I added four new lucky duckies to this thread.

> --- a/tools/testing/selftests/mm/protection_keys.c
> +++ b/tools/testing/selftests/mm/protection_keys.c
> @@ -1009,9 +1009,10 @@ static void test_pkey_syscalls_bad_args(int *ptr, u16 pkey)
>  	pkey_assert(err);
>  }
>  
> -static void become_child(void)
> +static void become_child(int *ptr)
>  {
>  	pid_t forkret;
> +	int status;
>  
>  	forkret = fork();
>  	pkey_assert(forkret >= 0);
> @@ -1021,7 +1022,16 @@ static void become_child(void)
>  		/* in the child */
>  		return;
>  	}
> -	_exit(0);
> +
> +	/*
> +	 * Wait for the child to finish the tests and pass on its result.
> +	 * Free the test memory first, as it may reserve pages from the
> +	 * small hugetlb pool that the child needs.
> +	 */
> +	free_pkey_malloc(ptr);
> +	if (waitpid(forkret, &status, 0) != forkret || !WIFEXITED(status))
> +		_exit(KSFT_FAIL);
> +	_exit(WEXITSTATUS(status));
>  }
>  
>  /* Assumes that all pkeys other than 'pkey' are unallocated */
> @@ -1060,7 +1070,7 @@ static void test_pkey_alloc_exhaust(int *ptr, u16 pkey)
>  		 * preserved across fork().
>  		 */
>  		if (i == NR_PKEYS*2)
> -			become_child();
> +			become_child(ptr);
>  	}
>  
>  	dprintf3("%s()::%d\n", __func__, __LINE__);
> 

Looks good, checks out, I'll queue it up.