]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
cgroup: Add test_cpucg_max_nested() testcase
authorDavid Vernet <void@manifault.com>
Sat, 23 Apr 2022 12:30:53 +0000 (05:30 -0700)
committerTejun Heo <tj@kernel.org>
Mon, 25 Apr 2022 17:27:31 +0000 (07:27 -1000)
The cgroup cpu controller selftests have a test_cpucg_max() testcase
that validates the behavior of the cpu.max knob. Let's also add a
testcase that verifies that the behavior works correctly when set on a
nested cgroup.

Signed-off-by: David Vernet <void@manifault.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
tools/testing/selftests/cgroup/test_cpu.c

index 715922c15c78f930f56e6aed0e94b8a94223da6b..24020a2c68dcdd9fff0e8d0f4af279d6840bbe77 100644 (file)
@@ -617,6 +617,68 @@ cleanup:
        return ret;
 }
 
+/*
+ * This test verifies that a process inside of a nested cgroup whose parent
+ * group has a cpu.max value set, is properly throttled.
+ */
+static int test_cpucg_max_nested(const char *root)
+{
+       int ret = KSFT_FAIL;
+       long usage_usec, user_usec;
+       long usage_seconds = 1;
+       long expected_usage_usec = usage_seconds * USEC_PER_SEC;
+       char *parent, *child;
+
+       parent = cg_name(root, "cpucg_parent");
+       child = cg_name(parent, "cpucg_child");
+       if (!parent || !child)
+               goto cleanup;
+
+       if (cg_create(parent))
+               goto cleanup;
+
+       if (cg_write(parent, "cgroup.subtree_control", "+cpu"))
+               goto cleanup;
+
+       if (cg_create(child))
+               goto cleanup;
+
+       if (cg_write(parent, "cpu.max", "1000"))
+               goto cleanup;
+
+       struct cpu_hog_func_param param = {
+               .nprocs = 1,
+               .ts = {
+                       .tv_sec = usage_seconds,
+                       .tv_nsec = 0,
+               },
+               .clock_type = CPU_HOG_CLOCK_WALL,
+       };
+       if (cg_run(child, hog_cpus_timed, (void *)&param))
+               goto cleanup;
+
+       usage_usec = cg_read_key_long(child, "cpu.stat", "usage_usec");
+       user_usec = cg_read_key_long(child, "cpu.stat", "user_usec");
+       if (user_usec <= 0)
+               goto cleanup;
+
+       if (user_usec >= expected_usage_usec)
+               goto cleanup;
+
+       if (values_close(usage_usec, expected_usage_usec, 95))
+               goto cleanup;
+
+       ret = KSFT_PASS;
+
+cleanup:
+       cg_destroy(child);
+       free(child);
+       cg_destroy(parent);
+       free(parent);
+
+       return ret;
+}
+
 #define T(x) { x, #x }
 struct cpucg_test {
        int (*fn)(const char *root);
@@ -629,6 +691,7 @@ struct cpucg_test {
        T(test_cpucg_nested_weight_overprovisioned),
        T(test_cpucg_nested_weight_underprovisioned),
        T(test_cpucg_max),
+       T(test_cpucg_max_nested),
 };
 #undef T