Re: [LTP] memcg/memcontrol05: add cgroup v2 task migration charge accounting test

Re: [LTP] memcg/memcontrol05: add cgroup v2 task migration charge accounting test

[linuxtestproject.agent]

Hi Pavithra,

On Aug 4, 2026, Pavithra wrote:
> memcg/memcontrol05: add cgroup v2 task migration charge accounting test

> 	buf1 = SAFE_MMAP(NULL, ALLOC_SIZE, PROT_READ | PROT_WRITE,
> 			 MAP_PRIVATE | MAP_ANONYMOUS | MAP_POPULATE, -1, 0);
>
> 	TST_CHECKPOINT_WAKE(WORKER_ALLOC_DONE);
> [...]
> 	buf2 = SAFE_MMAP(NULL, ALLOC_SIZE2, PROT_READ | PROT_WRITE,
> 			 MAP_PRIVATE | MAP_ANONYMOUS | MAP_POPULATE, -1, 0);
>
> 	TST_CHECKPOINT_WAKE(WORKER_ALLOC2_DONE);

Could these mappings be populated by explicitly writing one byte per page,
and could the test declare an appropriate .min_mem_avail requirement?

mmap(2) states that mmap() does not fail when MAP_POPULATE cannot populate
the whole mapping. Linux 7.2 also discards mm_populate() errors. Under memory
pressure, the parent can therefore report TFAIL for an incomplete allocation
even though cgroup migration accounting is correct. The test needs 80 MB, but
currently has no memory prerequisite to turn that condition into TCONF.

> 	pid_t pid;
> 	long baseline_b, after_migrate, after_alloc2, current_a;
> 	worker_resumed = false;

Could a blank line be added after the declarations? check-memcontrol05
reports:

  memcontrol05.c:80: WARNING: Missing a blank line after declarations

>  * Two child cgroups (group_a, group_b) are created under the test cgroup.
>  * A worker process is placed in group_a and allocates a known amount of
> [...]
>  *   group_b/cgroup.procs. In cgroup v2 memory charges are never
> [...]
>  *   that allocated them. group_b/memory.current must not increase

Could identifiers and file names in this exported RST block use double
backticks as required for inline literals?

> 	.needs_root	= 1,

Could the description also state why root is required? Root-only tests must
document the reason; here it is needed to create cgroups and move the worker
through cgroup.procs.

Verdict - Needs revision

---
Note:

The agent can sometimes produce false positives although often its
findings are genuine. If you find issues with the review, please
comment this email or ignore the suggestions.

Regards,
LTP AI Reviewer

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

[LTP] [PATCH v2] memcg/memcontrol05: add cgroup v2 task migration charge accounting test

[Pavithra]

Verify that in cgroup v2, memory charges remain pinned to the cgroup
that allocated them when a task migrates, and that new allocations after
migration are charged to the destination cgroup.

Signed-off-by: Pavithra <pavrampu@linux.ibm.com>
---
v1 -> v2:
- Replace MAP_POPULATE with explicit page-touching via touch_pages()
- Add .min_mem_avail = 80 (MB) to skip test under low memory
- Add blank line after declarations in test_memcg_task_migration()
- Use double backticks for identifiers and paths in RST block
- Document why .needs_root = 1 is required in test description
Link to v1: https://lore.kernel.org/ltp/20260804095925.495813-1-pavrampu@linux.ibm.com/
---
 runtest/controllers                           |   1 +
 testcases/kernel/controllers/memcg/.gitignore |   1 +
 .../kernel/controllers/memcg/memcontrol05.c   | 175 ++++++++++++++++++
 3 files changed, 177 insertions(+)
 create mode 100644 testcases/kernel/controllers/memcg/memcontrol05.c

diff --git a/runtest/controllers b/runtest/controllers
index 93c52c439..a9050ef0e 100644
--- a/runtest/controllers
+++ b/runtest/controllers
@@ -24,6 +24,7 @@ memcontrol01 memcontrol01
 memcontrol02 memcontrol02
 memcontrol03 memcontrol03
 memcontrol04 memcontrol04
+memcontrol05 memcontrol05
 
 cgroup_fj_function_debug cgroup_fj_function.sh debug
 cgroup_fj_function_cpuset cgroup_fj_function.sh cpuset
diff --git a/testcases/kernel/controllers/memcg/.gitignore b/testcases/kernel/controllers/memcg/.gitignore
index 3883cede6..8b9f6005c 100644
--- a/testcases/kernel/controllers/memcg/.gitignore
+++ b/testcases/kernel/controllers/memcg/.gitignore
@@ -9,3 +9,4 @@ memcontrol01
 memcontrol02
 memcontrol03
 memcontrol04
+memcontrol05
diff --git a/testcases/kernel/controllers/memcg/memcontrol05.c b/testcases/kernel/controllers/memcg/memcontrol05.c
new file mode 100644
index 000000000..4cb727ffa
--- /dev/null
+++ b/testcases/kernel/controllers/memcg/memcontrol05.c
@@ -0,0 +1,175 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2026 IBM
+ * Author : Pavithra <pavrampu@linux.ibm.com>
+ */
+
+/*\
+ * Origin: ``testcases/kernel/controllers/memctl/memctl_testplan.txt``
+ * Verify memory controller charge accounting during task migration.
+ *
+ * Two child cgroups (``group_a``, ``group_b``) are created under the test cgroup.
+ * A worker process is placed in ``group_a`` and allocates a known amount of
+ * anonymous memory, touching every page to ensure physical allocation.
+ * Root is required to create cgroup subdirectories and to write a PID
+ * into ``cgroup.procs`` to migrate the worker between cgroups.
+ *
+ * Test 1 - No charge transfer on migration:
+ *   The worker is migrated to ``group_b`` by writing its PID to
+ *   ``group_b/cgroup.procs``. In cgroup v2 memory charges are never
+ *   transferred on task migration; they remain pinned to the cgroup
+ *   that allocated them. ``group_b/memory.current`` must not increase
+ *   beyond its baseline after the migration.
+ *
+ * Test 2 - New allocations charged to destination cgroup:
+ *   After migration the worker allocates an additional chunk of memory.
+ *   This new allocation must be charged to ``group_b``, so
+ *   ``group_b/memory.current`` must increase from the post-migration
+ *   baseline.
+ */
+
+#define _GNU_SOURCE
+
+#include <stdbool.h>
+#include <stdlib.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include "tst_test.h"
+#include "memcontrol_common.h"
+
+#define ALLOC_SIZE	MB(60)
+#define ALLOC_SIZE2	MB(20)
+#define MIN_MEM_AVAIL	80
+
+enum checkpoints {
+	WORKER_ALLOC_DONE,
+	WORKER_RESUME,
+	WORKER_ALLOC2_DONE,
+	WORKER_EXIT,
+};
+
+static struct tst_cg_group *group_a;
+static struct tst_cg_group *group_b;
+static bool worker_resumed;
+
+static void touch_pages(char *buf, size_t size)
+{
+	size_t i;
+
+	for (i = 0; i < size; i += getpagesize())
+		buf[i] = 1;
+}
+
+static void worker(void)
+{
+	char *buf1, *buf2;
+
+	SAFE_CG_PRINTF(group_a, "cgroup.procs", "%d", getpid());
+
+	buf1 = SAFE_MMAP(NULL, ALLOC_SIZE, PROT_READ | PROT_WRITE,
+			 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+	touch_pages(buf1, ALLOC_SIZE);
+
+	TST_CHECKPOINT_WAKE(WORKER_ALLOC_DONE);
+
+	TST_CHECKPOINT_WAIT(WORKER_RESUME);
+
+	buf2 = SAFE_MMAP(NULL, ALLOC_SIZE2, PROT_READ | PROT_WRITE,
+			 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+	touch_pages(buf2, ALLOC_SIZE2);
+
+	TST_CHECKPOINT_WAKE(WORKER_ALLOC2_DONE);
+
+	TST_CHECKPOINT_WAIT(WORKER_EXIT);
+
+	SAFE_MUNMAP(buf1, ALLOC_SIZE);
+	SAFE_MUNMAP(buf2, ALLOC_SIZE2);
+}
+
+static void test_memcg_task_migration(void)
+{
+	pid_t pid;
+	long baseline_b, after_migrate, after_alloc2, current_a;
+
+	worker_resumed = false;
+
+	group_a = tst_cg_group_mk(tst_cg, "group_a");
+	group_b = tst_cg_group_mk(tst_cg, "group_b");
+
+	SAFE_CG_SCANF(group_b, "memory.current", "%ld", &baseline_b);
+	tst_res(TINFO, "group_b baseline memory.current=%ld", baseline_b);
+
+	pid = SAFE_FORK();
+	if (!pid) {
+		worker();
+		exit(0);
+	}
+
+	TST_CHECKPOINT_WAIT(WORKER_ALLOC_DONE);
+
+	SAFE_CG_SCANF(group_a, "memory.current", "%ld", &current_a);
+	tst_res(TINFO, "group_a memory.current=%ld after alloc", current_a);
+	if (current_a < (long)ALLOC_SIZE) {
+		tst_res(TFAIL,
+			"group_a memory.current (%ld) < ALLOC_SIZE (%ld)",
+			current_a, (long)ALLOC_SIZE);
+		goto done;
+	}
+	tst_res(TPASS,
+		"group_a memory.current (%ld) >= ALLOC_SIZE (%ld)",
+		current_a, (long)ALLOC_SIZE);
+
+	SAFE_CG_PRINTF(group_b, "cgroup.procs", "%d", pid);
+	tst_res(TINFO, "Migrated worker PID %d to group_b", pid);
+
+	SAFE_CG_SCANF(group_b, "memory.current", "%ld", &after_migrate);
+	tst_res(TINFO,
+		"group_b memory.current=%ld after migration (baseline=%ld)",
+		after_migrate, baseline_b);
+
+	TST_EXP_EXPR(after_migrate <= baseline_b + (long)MB(4),
+		     "group_b memory.current (%ld) not increased after migration (baseline=%ld)",
+		     after_migrate, baseline_b);
+
+	TST_CHECKPOINT_WAKE(WORKER_RESUME);
+	worker_resumed = true;
+	TST_CHECKPOINT_WAIT(WORKER_ALLOC2_DONE);
+
+	SAFE_CG_SCANF(group_b, "memory.current", "%ld", &after_alloc2);
+	tst_res(TINFO, "group_b memory.current=%ld after second alloc (baseline=%ld)",
+		after_alloc2, after_migrate);
+
+	TST_EXP_EXPR(after_alloc2 >= after_migrate + (long)ALLOC_SIZE2,
+		     "group_b memory.current (%ld) increased by >= ALLOC_SIZE2 (%ld)",
+		     after_alloc2, (long)ALLOC_SIZE2);
+
+done:
+	if (!worker_resumed) {
+		TST_CHECKPOINT_WAKE(WORKER_RESUME);
+		TST_CHECKPOINT_WAIT(WORKER_ALLOC2_DONE);
+	}
+	TST_CHECKPOINT_WAKE(WORKER_EXIT);
+	tst_reap_children();
+
+	group_a = tst_cg_group_rm(group_a);
+	group_b = tst_cg_group_rm(group_b);
+}
+
+static void cleanup(void)
+{
+	if (group_a)
+		group_a = tst_cg_group_rm(group_a);
+	if (group_b)
+		group_b = tst_cg_group_rm(group_b);
+}
+
+static struct tst_test test = {
+	.test_all	= test_memcg_task_migration,
+	.cleanup	= cleanup,
+	.forks_child	= 1,
+	.needs_root	= 1,
+	.needs_checkpoints = 1,
+	.needs_cgroup_ver  = TST_CG_V2,
+	.needs_cgroup_ctrls = (const char *const []){ "memory", NULL },
+	.min_mem_avail	= MIN_MEM_AVAIL,
+};
-- 
2.55.0


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

Re: [LTP] memcg/memcontrol05: add cgroup v2 task migration charge accounting test

[linuxtestproject.agent]

Hi Pavithra,

On Aug 12, 2026, Pavithra wrote:
> memcg/memcontrol05: add cgroup v2 task migration charge accounting test

> +#define ALLOC_SIZE	MB(60)
> +#define ALLOC_SIZE2	MB(20)
> +#define MIN_MEM_AVAIL	80

[Medium] Could MIN_MEM_AVAIL include headroom above the two mappings? The
worker keeps both mappings resident simultaneously, so a host with exactly
80 MiB available passes the framework check but still needs memory for page
tables and other process and kernel allocations. The worker can be OOM-killed
while touching the mappings, producing a timeout or TBROK instead of testing
charge accounting.

> +	SAFE_CG_SCANF(group_b, "memory.current", "%ld", &after_migrate);
> +	TST_EXP_EXPR(after_migrate <= baseline_b + (long)MB(4),
> +		     "group_b memory.current (%ld) not increased after migration (baseline=%ld)",
> +		     after_migrate, baseline_b);

[Medium] Could this also read and verify group_a/memory.current after the
migration? The stated invariant is that existing charges remain pinned to
group_a, but this checks only that group_b did not gain them. A regression
that drops those charges during migration leaves group_b within the allowed
range and reports TPASS despite violating the invariant.

Verdict - Needs revision

---
Note:

The agent can sometimes produce false positives although often its
findings are genuine. If you find issues with the review, please
comment this email or ignore the suggestions.

Regards,
LTP AI Reviewer

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp