Skip to contentSkip to conversation

memcg_control_test: Disable swapping in test cgroup

PATCH In review T.J. Mercier via ltp 2 participants 3 messages Mar 26, 2025 – Mar 27, 2025 1 message not in the archive

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 memcg_control_test: Disable swapping in test cgroup

memcg_control_test_1 sets both a memory limit and a memsw/swap limit to a single page, and then asks mem_process to allocate 2 pages in an attempt to cause a memcg OOM kill. However, if swapping is enabled the first page can be swapped before the second is allocated causing neither limit to be exceeded, and no OOM kill to occur which results in a test failure. Fix this by disabling swap in the test cgroup if it is enabled, causing both pages to always be accounted under the same counter such that memory.max (v2) / memory.limit_in_bytes (v1) causes an OOM kill.

Fixes c0b815736 ("New testcase added by Naufal.")
---
 .../kernel/controllers/memcg/control/memcg_control_test.sh   | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
testcases/kernel/controllers/memcg/control/memcg_control_test.sh +2 -3
index 68287a70c..eb8db3e30 100644
@@ -12,7 +12,6 @@ TST_NEEDS_TMPDIR=1
 
 PAGE_SIZE=$(tst_getconf PAGESIZE)
 
-TOT_MEM_LIMIT=$PAGE_SIZE
 ACTIVE_MEM_LIMIT=$PAGE_SIZE
 PROC_MEM=$((PAGE_SIZE * 2))
 
@@ -50,13 +49,13 @@ test1()
 
 	# If the kernel is built without swap, the $memsw_memory_limit file is missing
 	if [ -e "$test_dir/$memsw_memory_limit" ]; then
-		ROD echo "$TOT_MEM_LIMIT" \> "$test_dir/$memsw_memory_limit"
+		ROD echo 0 \> "$test_dir/$memsw_memory_limit"
 	fi
 
 	KILLED_CNT=0
 	test_proc_kill
 
-	if [ $PROC_MEM -gt $TOT_MEM_LIMIT ] && [ $KILLED_CNT -eq 0 ]; then
+	if [ $KILLED_CNT -eq 0 ]; then
 		tst_res TFAIL "Test #1: failed"
 	else
 		tst_res TPASS "Test #1: passed"
Signature · 5 lines
2.49.0.395.g12beb8f557-goog


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
Reviewed-by on v1 1/1
#1 lore raw

Hi,

Quoted 8 lines

memcg_control_test_1 sets both a memory limit and a memsw/swap limit to a single page, and then asks mem_process to allocate 2 pages in an attempt to cause a memcg OOM kill. However, if swapping is enabled the first page can be swapped before the second is allocated causing neither limit to be exceeded, and no OOM kill to occur which results in a test failure. Fix this by disabling swap in the test cgroup if it is enabled, causing both pages to always be accounted under the same counter such that memory.max (v2) / memory.limit_in_bytes (v1) causes an OOM kill.

LGTM.

I suppose the old shell tests which cgroupv1 memory.memsw.limit_in_bytes are written to test memory limits, thus only this one is suitable to modify.

The new C based tests (cgroupv2) should not be affected either.

Kind regards,
Petr

Quoted 34 lines
Fixes c0b815736 ("New testcase added by Naufal.")
---
 .../kernel/controllers/memcg/control/memcg_control_test.sh   | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
testcases/kernel/controllers/memcg/control/memcg_control_test.sh +2 -3
index 68287a70c..eb8db3e30 100644
@@ -12,7 +12,6 @@ TST_NEEDS_TMPDIR=1
 
 PAGE_SIZE=$(tst_getconf PAGESIZE)
 
-TOT_MEM_LIMIT=$PAGE_SIZE
 ACTIVE_MEM_LIMIT=$PAGE_SIZE
 PROC_MEM=$((PAGE_SIZE * 2))
 
@@ -50,13 +49,13 @@ test1()
 
 	# If the kernel is built without swap, the $memsw_memory_limit file is missing
 	if [ -e "$test_dir/$memsw_memory_limit" ]; then
-		ROD echo "$TOT_MEM_LIMIT" \> "$test_dir/$memsw_memory_limit"
+		ROD echo 0 \> "$test_dir/$memsw_memory_limit"
 	fi
 
 	KILLED_CNT=0
 	test_proc_kill
 
-	if [ $PROC_MEM -gt $TOT_MEM_LIMIT ] && [ $KILLED_CNT -eq 0 ]; then
+	if [ $KILLED_CNT -eq 0 ]; then
 		tst_res TFAIL "Test #1: failed"
 	else
 		tst_res TPASS "Test #1: passed"
Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp

A message that is not in the archive (sent off-list or not yet received): lo4sjpn6uvhyhgi7welsp4dsvi6jxrl4h5gwacuvqmgfvfqcq7@ayoafjktxzyc

#3 lore raw
Quoted 13 lines · Michal Koutný

Hello.

Quoted 8 lines · T.J. Mercier

memcg_control_test_1 sets both a memory limit and a memsw/swap limit to a single page, and then asks mem_process to allocate 2 pages in an attempt to cause a memcg OOM kill. However, if swapping is enabled the first page can be swapped before the second is allocated causing neither limit to be exceeded, and no OOM kill to occur which results in a test failure. Fix this by disabling swap in the test cgroup if it is enabled, causing both pages to always be accounted under the same counter such that memory.max (v2) / memory.limit_in_bytes (v1) causes an OOM kill.

Quoted 7 lines

I see how this fixes the test for v2 but for v1 the write would fail with -EINVAL because memory.memsw.limit_in_bytes < memory.limit_in_bytes and swapout would still be possible (which it wasn't _on v1_ before this patch IIUC).

IOW the testcase should not treat memory.swap.max (v2) and memory.memsw.limit_in_bytes (v1) as equivalent quantities.

Thanks Michal. I'm about to send a v2 where swap is disabled only in memcg v2 but the memsw limit is still applied in memcg v1. Tested on both versions of memcg this time.

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp