[PATCH bpf-next 0/3] Fix test_cgroup_iter_memcg issues found during back-porting

[PATCH bpf-next 0/3] Fix test_cgroup_iter_memcg issues found during back-porting

[Hui Zhu]

From: Hui Zhu <zhuhui@kylinos.cn>

While back-porting "mm: bpf kfuncs to access memcg data", I
encountered issues with test_cgroup_iter_memcg, specifically
in test_kmem.

The test_cgroup_iter_memcg test would falsely pass when
bpf_mem_cgroup_page_state() failed due to incompatible enum
values across kernel versions. Additionally, test_kmem would
fail on systems with cgroup.memory=nokmem enabled.

This series addresses these issues:
1. Add return value checks for bpf_mem_cgroup_page_state()
2. Return error when accessing kmem with nokmem enabled
3. Skip test_kmem when cgroup.memory=nokmem is set

Hui Zhu (3):
  selftests/bpf: Check bpf_mem_cgroup_page_state return value
  mm/memcontrol: Return error when accessing kmem with nokmem
  selftests/bpf: Skip test_kmem when cgroup.memory=nokmem

 mm/memcontrol.c                               |  3 +-
 .../bpf/prog_tests/cgroup_iter_memcg.c        | 40 +++++++++++++++++++
 2 files changed, 42 insertions(+), 1 deletion(-)

-- 
2.43.0

[PATCH bpf-next 1/3] selftests/bpf: Check bpf_mem_cgroup_page_state return value

[Hui Zhu]

From: Hui Zhu <zhuhui@kylinos.cn>

When back-porting test_progs to different kernel versions, I encountered
an issue where the test_cgroup_iter_memcg test would falsely pass even
when bpf_mem_cgroup_page_state() failed.

The problem occurs when test_progs compiled on one kernel version is
executed on another kernel with different enum values for memory
statistics (e.g., NR_ANON_MAPPED, NR_FILE_PAGES). In such cases,
bpf_mem_cgroup_page_state() returns -1 to indicate failure, but the test
didn't check for this error condition and incorrectly reported success.

This patch adds explicit checks to ensure bpf_mem_cgroup_page_state()
doesn't return -1 before validating the actual statistics values. This
prevents false positives when running test_progs in cross-kernel
environments.

Signed-off-by: Hui Zhu <zhuhui@kylinos.cn>
---
 .../selftests/bpf/prog_tests/cgroup_iter_memcg.c     | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c b/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
index a5afd16705f0..13b299512429 100644
--- a/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
+++ b/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
@@ -53,6 +53,8 @@ static void test_anon(struct bpf_link *link, struct memcg_query *memcg_query)
 	if (!ASSERT_OK(read_stats(link), "read stats"))
 		goto cleanup;
 
+	ASSERT_NEQ(memcg_query->nr_anon_mapped, (unsigned long)-1,
+		  "bpf_mem_cgroup_page_state NR_ANON_MAPPED");
 	ASSERT_GT(memcg_query->nr_anon_mapped, 0, "final anon mapped val");
 
 cleanup:
@@ -88,6 +90,10 @@ static void test_file(struct bpf_link *link, struct memcg_query *memcg_query)
 	if (!ASSERT_OK(read_stats(link), "read stats"))
 		goto cleanup_map;
 
+	ASSERT_NEQ(memcg_query->nr_file_pages, (unsigned long)-1,
+		  "bpf_mem_cgroup_page_state NR_FILE_PAGES");
+	ASSERT_NEQ(memcg_query->nr_file_mapped, (unsigned long)-1,
+		  "bpf_mem_cgroup_page_state NR_FILE_MAPPED");
 	ASSERT_GT(memcg_query->nr_file_pages, 0, "final file value");
 	ASSERT_GT(memcg_query->nr_file_mapped, 0, "final file mapped value");
 
@@ -119,6 +125,8 @@ static void test_shmem(struct bpf_link *link, struct memcg_query *memcg_query)
 	if (!ASSERT_OK(read_stats(link), "read stats"))
 		goto cleanup;
 
+	ASSERT_NEQ(memcg_query->nr_shmem, (unsigned long)-1,
+		  "bpf_mem_cgroup_page_state NR_SHMEM");
 	ASSERT_GT(memcg_query->nr_shmem, 0, "final shmem value");
 
 cleanup:
@@ -143,6 +151,8 @@ static void test_kmem(struct bpf_link *link, struct memcg_query *memcg_query)
 	if (!ASSERT_OK(read_stats(link), "read stats"))
 		goto cleanup;
 
+	ASSERT_NEQ(memcg_query->memcg_kmem, (unsigned long)-1,
+		  "bpf_mem_cgroup_page_state MEMCG_KMEM");
 	ASSERT_GT(memcg_query->memcg_kmem, 0, "kmem value");
 
 cleanup:
@@ -170,6 +180,8 @@ static void test_pgfault(struct bpf_link *link, struct memcg_query *memcg_query)
 	if (!ASSERT_OK(read_stats(link), "read stats"))
 		goto cleanup;
 
+	ASSERT_NEQ(memcg_query->pgfault, (unsigned long)-1,
+		  "bpf_mem_cgroup_page_state PGFAULT");
 	ASSERT_GT(memcg_query->pgfault, 0, "final pgfault val");
 
 cleanup:
-- 
2.43.0

[PATCH bpf-next 2/3] mm/memcontrol: Return error when accessing kmem with nokmem

[Hui Zhu]

From: Hui Zhu <zhuhui@kylinos.cn>

When running tests on hosts with cgroup.memory=nokmem enabled for
performance reasons, test_kmem always gets a value of 0 for kmem
statistics.

Since BPF programs cannot easily determine whether kmem is enabled,
add a check in memcg_stat_item_valid() to return an error when
attempting to access MEMCG_KMEM statistics while kmem accounting
is disabled via cgroup_memory_nokmem.

This prevents BPF programs from silently receiving zero values and
allows them to properly handle the case where kmem accounting is
unavailable.

Signed-off-by: Hui Zhu <zhuhui@kylinos.cn>
---
 mm/memcontrol.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 129eed3ff5bb..4d8419623d1c 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -667,7 +667,8 @@ unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
 
 bool memcg_stat_item_valid(int idx)
 {
-	if ((u32)idx >= MEMCG_NR_STAT)
+	if ((u32)idx >= MEMCG_NR_STAT ||
+	    (cgroup_memory_nokmem && (u32)idx == MEMCG_KMEM))
 		return false;
 
 	return !BAD_STAT_IDX(memcg_stats_index(idx));
-- 
2.43.0

[PATCH bpf-next 3/3] selftests/bpf: Skip test_kmem when cgroup.memory=nokmem

[Hui Zhu]

From: Hui Zhu <zhuhui@kylinos.cn>

When cgroup.memory=nokmem is set in kernel command line, kmem
accounting is disabled and the test_kmem subtest will fail.

Add a check to skip this test when the parameter is present.

Signed-off-by: Hui Zhu <zhuhui@kylinos.cn>
---
 .../bpf/prog_tests/cgroup_iter_memcg.c        | 28 +++++++++++++++++++
 1 file changed, 28 insertions(+)

diff --git a/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c b/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
index 13b299512429..203e6b091a21 100644
--- a/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
+++ b/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
@@ -134,11 +134,39 @@ static void test_shmem(struct bpf_link *link, struct memcg_query *memcg_query)
 	shm_unlink("/tmp_shmem");
 }
 
+static bool cmdline_has(const char *arg)
+{
+	char cmdline[4096];
+	int fd;
+	ssize_t len;
+	bool ret = false;
+
+	fd = open("/proc/cmdline", O_RDONLY);
+	if (fd < 0)
+		return false;
+
+	len = read(fd, cmdline, sizeof(cmdline) - 1);
+	close(fd);
+	if (len < 0)
+		return false;
+
+	cmdline[len] = '\0';
+	if (strstr(cmdline, arg))
+		ret = true;
+
+	return ret;
+}
+
 #define NR_PIPES 64
 static void test_kmem(struct bpf_link *link, struct memcg_query *memcg_query)
 {
 	int fds[NR_PIPES][2], i;
 
+	if (cmdline_has("cgroup.memory=nokmem")) {
+		test__skip();
+		return;
+	}
+
 	/*
 	 * Increase kmem value by creating pipes which will allocate some
 	 * kernel buffers.
-- 
2.43.0

Re: [PATCH bpf-next 1/3] selftests/bpf: Check bpf_mem_cgroup_page_state return value

[bot+bpf-ci]

[-- Attachment #1: Type: text/plain, Size: 1644 bytes --]

> diff --git a/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c b/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
> --- a/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
> +++ b/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c

[ ... ]

> @@ -170,6 +180,8 @@ static void test_pgfault(struct bpf_link *link, struct memcg_query *memcg_query)
>  	if (!ASSERT_OK(read_stats(link), "read stats"))
>  		goto cleanup;
>
> +	ASSERT_NEQ(memcg_query->pgfault, (unsigned long)-1,
> +		  "bpf_mem_cgroup_page_state PGFAULT");
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^

The assertion message references bpf_mem_cgroup_page_state(), but the BPF
program actually uses bpf_mem_cgroup_vm_events() to retrieve the pgfault
value:

tools/testing/selftests/bpf/progs/cgroup_iter_memcg.c:
    memcg_query.pgfault = bpf_mem_cgroup_vm_events(memcg, PGFAULT);

Should the message be "bpf_mem_cgroup_vm_events PGFAULT" instead?


---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/21939182103

AI-authorship-score: low
AI-authorship-explanation: The commit message uses first-person narrative describing a real back-porting scenario, and the code changes are simple repetitive assertions typical of human-written test improvements.
issues-found: 1
issue-severity-score: low
issue-severity-explanation: Minor documentation issue where an assertion message references the wrong BPF function name; does not affect test correctness or system behavior.

Re: [PATCH bpf-next 1/3] selftests/bpf: Check bpf_mem_cgroup_page_state return value

[hui.zhu]

2026年2月12日 16:50, bot+bpf-ci@kernel.org mailto:bot+bpf-ci@kernel.org  写到:


> 
> > 
> > diff --git a/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c b/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
> >  --- a/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
> >  +++ b/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
> > 
> [ ... ]
> 
> > 
> > @@ -170,6 +180,8 @@ static void test_pgfault(struct bpf_link *link, struct memcg_query *memcg_query)
> >  if (!ASSERT_OK(read_stats(link), "read stats"))
> >  goto cleanup;
> > 
> >  + ASSERT_NEQ(memcg_query->pgfault, (unsigned long)-1,
> >  + "bpf_mem_cgroup_page_state PGFAULT");
> > 
>  ^^^^^^^^^^^^^^^^^^^^^^^^^^^
> 
> The assertion message references bpf_mem_cgroup_page_state(), but the BPF
> program actually uses bpf_mem_cgroup_vm_events() to retrieve the pgfault
> value:
> 
> tools/testing/selftests/bpf/progs/cgroup_iter_memcg.c:
>  memcg_query.pgfault = bpf_mem_cgroup_vm_events(memcg, PGFAULT);
> 
> Should the message be "bpf_mem_cgroup_vm_events PGFAULT" instead?

Good catch!  I will fix it in the next version.

Best,
Hui

> 
> ---
> AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
> See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md
> 
> CI run summary: https://github.com/kernel-patches/bpf/actions/runs/21939182103
> 
> AI-authorship-score: low
> AI-authorship-explanation: The commit message uses first-person narrative describing a real back-porting scenario, and the code changes are simple repetitive assertions typical of human-written test improvements.
> issues-found: 1
> issue-severity-score: low
> issue-severity-explanation: Minor documentation issue where an assertion message references the wrong BPF function name; does not affect test correctness or system behavior.
>

Re: [PATCH bpf-next 1/3] selftests/bpf: Check bpf_mem_cgroup_page_state return value

[JP Kobryn]

On 2/12/26 12:23 AM, Hui Zhu wrote:
> From: Hui Zhu <zhuhui@kylinos.cn>
> 
> When back-porting test_progs to different kernel versions, I encountered
> an issue where the test_cgroup_iter_memcg test would falsely pass even
> when bpf_mem_cgroup_page_state() failed.
> 
> The problem occurs when test_progs compiled on one kernel version is
> executed on another kernel with different enum values for memory
> statistics (e.g., NR_ANON_MAPPED, NR_FILE_PAGES). [...]

This patch looks good but I think to fully solve this cross-kernel issue
we should use co-re in the bpf program. In your second revision, can you
add an additional patch to make use of bpf_core_enum_value()? This way
instead of relying on enum values in vmlinux.h at compile-time, we use
the btf info at load-time instead to get the proper value for the given
kernel.

Re: [PATCH bpf-next 2/3] mm/memcontrol: Return error when accessing kmem with nokmem

[JP Kobryn (Meta)]

On 2/12/26 12:23 AM, Hui Zhu wrote:
> From: Hui Zhu <zhuhui@kylinos.cn>
> 
> When running tests on hosts with cgroup.memory=nokmem enabled for
> performance reasons, test_kmem always gets a value of 0 for kmem
> statistics.
> 
> Since BPF programs cannot easily determine whether kmem is enabled,
> add a check in memcg_stat_item_valid() to return an error when
> attempting to access MEMCG_KMEM statistics while kmem accounting
> is disabled via cgroup_memory_nokmem.
> 
> This prevents BPF programs from silently receiving zero values and
> allows them to properly handle the case where kmem accounting is
> unavailable.
> 
> Signed-off-by: Hui Zhu <zhuhui@kylinos.cn>
> ---
>   mm/memcontrol.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index 129eed3ff5bb..4d8419623d1c 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -667,7 +667,8 @@ unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
>   
>   bool memcg_stat_item_valid(int idx)
>   {
> -	if ((u32)idx >= MEMCG_NR_STAT)
> +	if ((u32)idx >= MEMCG_NR_STAT ||
> +	    (cgroup_memory_nokmem && (u32)idx == MEMCG_KMEM))
>   		return false;

It's still a valid stat though, right? When it's disabled the value will
just remain zero. I don't think this is necessary.

Re: [PATCH bpf-next 3/3] selftests/bpf: Skip test_kmem when cgroup.memory=nokmem

[JP Kobryn (Meta)]

On 2/12/26 12:23 AM, Hui Zhu wrote:
> From: Hui Zhu <zhuhui@kylinos.cn>
> 
> When cgroup.memory=nokmem is set in kernel command line, kmem
> accounting is disabled and the test_kmem subtest will fail.
> 
> Add a check to skip this test when the parameter is present.
> 
> Signed-off-by: Hui Zhu <zhuhui@kylinos.cn>
> ---
>   .../bpf/prog_tests/cgroup_iter_memcg.c        | 28 +++++++++++++++++++
>   1 file changed, 28 insertions(+)
> 
> diff --git a/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c b/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
> index 13b299512429..203e6b091a21 100644
> --- a/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
> +++ b/tools/testing/selftests/bpf/prog_tests/cgroup_iter_memcg.c
> @@ -134,11 +134,39 @@ static void test_shmem(struct bpf_link *link, struct memcg_query *memcg_query)
>   	shm_unlink("/tmp_shmem");
>   }
>   
> +static bool cmdline_has(const char *arg)
> +{
> +	char cmdline[4096];
> +	int fd;
> +	ssize_t len;
> +	bool ret = false;
> +
> +	fd = open("/proc/cmdline", O_RDONLY);
> +	if (fd < 0)
> +		return false;
> +
> +	len = read(fd, cmdline, sizeof(cmdline) - 1);
> +	close(fd);
> +	if (len < 0)
> +		return false;
> +
> +	cmdline[len] = '\0';
> +	if (strstr(cmdline, arg))
> +		ret = true;
> +
> +	return ret;
> +}
> +
>   #define NR_PIPES 64
>   static void test_kmem(struct bpf_link *link, struct memcg_query *memcg_query)
>   {
>   	int fds[NR_PIPES][2], i;
>   
> +	if (cmdline_has("cgroup.memory=nokmem")) {
> +		test__skip();
> +		return;
> +	}

Instead of just skipping what if we proceed and then confirm we get a
zero value after the allocations?

Re: [PATCH bpf-next 1/3] selftests/bpf: Check bpf_mem_cgroup_page_state return value

[hui.zhu]

2026年2月13日 08:14, "JP Kobryn" <inwardvessel@gmail.com mailto:inwardvessel@gmail.com?to=%22JP%20Kobryn%22%20%3Cinwardvessel%40gmail.com%3E > 写到:


> 
> On 2/12/26 12:23 AM, Hui Zhu wrote:
> 
> > 
> > From: Hui Zhu <zhuhui@kylinos.cn>
> >  When back-porting test_progs to different kernel versions, I encountered
> >  an issue where the test_cgroup_iter_memcg test would falsely pass even
> >  when bpf_mem_cgroup_page_state() failed.
> >  The problem occurs when test_progs compiled on one kernel version is
> >  executed on another kernel with different enum values for memory
> >  statistics (e.g., NR_ANON_MAPPED, NR_FILE_PAGES). [...]
> > 
> This patch looks good but I think to fully solve this cross-kernel issue
> we should use co-re in the bpf program. In your second revision, can you
> add an additional patch to make use of bpf_core_enum_value()? This way
> instead of relying on enum values in vmlinux.h at compile-time, we use
> the btf info at load-time instead to get the proper value for the given
> kernel.
>

I post v2 according to your comments.
Thanks for your review.

Best,
Hui