From: Hui Zhu <zhuhui@kylinos.cn>

When cgroup.memory=nokmem is set in the kernel command line, kmem
accounting is disabled. This causes the test_kmem subtest in
cgroup_iter_memcg to fail because it expects non-zero kmem values.

Fix this by checking /proc/cmdline for the nokmem parameter. If
found, verify that kmem value is zero and return early, skipping
the pipe creation test that would otherwise fail.

Signed-off-by: Hui Zhu <zhuhui@kylinos.cn>
---
 .../bpf/prog_tests/cgroup_iter_memcg.c        | 30 +++++++++++++++++++
 1 file changed, 30 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 897b17b58df3..2b9c148cebf0 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 +134,41 @@ 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")) {
+		if (!ASSERT_OK(read_stats(link), "read stats"))
+			return;
+		ASSERT_EQ(memcg_query->memcg_kmem, 0, "kmem value");
+		return;
+	}
+
 	/*
 	 * Increase kmem value by creating pipes which will allocate some
 	 * kernel buffers.
-- 
2.43.0

> 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

[ ... ]

> +static bool cmdline_has(const char *arg)
> +{

[ ... ]

> +	cmdline[len] = '\0';
> +	if (strstr(cmdline, arg))
> +		ret = true;
> +
> +	return ret;
> +}

[ ... ]

> +	if (cmdline_has("cgroup.memory=nokmem")) {

The kernel's cgroup.memory= parameter accepts a comma-separated list
of options.  The parser in mm/memcontrol.c cgroup_memory() splits the
value on commas with strsep():

mm/memcontrol.c:cgroup_memory() {
    while ((token = strsep(&s, ",")) != NULL) {
        ...
        if (!strcmp(token, "nokmem"))
            cgroup_memory_nokmem = true;
        ...
    }
}

This means valid command lines include configurations like
"cgroup.memory=nosocket,nokmem" or "cgroup.memory=nobpf,nokmem".

Since cmdline_has() uses strstr() to match the full literal string
"cgroup.memory=nokmem", it only matches when nokmem is the first (or
only) option.  For "cgroup.memory=nosocket,nokmem", strstr() won't
find "cgroup.memory=nokmem" as a substring, so the check returns
false even though nokmem is active.

Would it be more robust to search for just "nokmem" within the
cgroup.memory= value, or to split on commas similar to how the kernel
parses it?

> +		if (!ASSERT_OK(read_stats(link), "read stats"))
> +			return;
> +		ASSERT_EQ(memcg_query->memcg_kmem, 0, "kmem value");
> +		return;
> +	}


---
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/21978734425

AI-authorship-score: low
AI-authorship-explanation: The patch addresses a specific real-world back-porting issue, shows iterative development across versions in response to human reviewer feedback, and uses standard kernel C patterns with no hallmarks of AI generation.
issues-found: 1
issue-severity-score: low
issue-severity-explanation: The strstr-based cmdline check misses nokmem when it is not the first option in a comma-separated cgroup.memory= value, causing the test to fail on those configurations instead of being skipped.