[LTP] [PATCH v1] memcontrol03: Using clean page cache to avoid dependency on IO rate

[Cyril Hrubis]

Hi!
> Bad IO situation(storage bandwidth ~10MB/sec) will lead background
> writeback has uncertain progress for dirty page. So system can not
> reclaim enough memory for new process and finally lead a unexpected
> OOM.
> 
> memcontrol03.c:218: TPASS: Expect: (A/B/E memory.current=0) ~= 0
> memcontrol03.c:116: TPASS: Child 1918 killed by OOM
> memcontrol03.c:224: TPASS: Expect: (A/B memory.current=52588544) ~= 52428800
> memcontrol03.c:129: TFAIL: Expected child 1944 to exit(0), but instead killed by SIGKILL

So I've read the corresponding bug:

https://bugzilla.suse.com/show_bug.cgi?id=1218178

And now I understand what this is supposed to fix. Hover doing sync() is
a too big hammer, there is no need to sync the whole system, we just
need to make sure that the pages from the file we wrote into are written
to a pernament storage, so we likely just need:

diff --git a/testcases/kernel/controllers/memcg/memcontrol03.c b/testcases/kernel/controllers/memcg/memcontrol03.c
index e927dfd19..6d9c490b5 100644
--- a/testcases/kernel/controllers/memcg/memcontrol03.c
+++ b/testcases/kernel/controllers/memcg/memcontrol03.c
@@ -144,6 +144,7 @@ static void alloc_pagecache_in_child(const struct tst_cg_group *const cg,
        tst_res(TINFO, "Child %d in %s: Allocating pagecache: %"PRIdPTR,
                getpid(), tst_cg_group_name(cg), size);
        alloc_pagecache(fd, size);
+       SAFE_FSYNC(fd);

        TST_CHECKPOINT_WAKE(CHILD_IDLE);
        TST_CHECKPOINT_WAIT(TEST_DONE);

-- 
Cyril Hrubis
chrubis@suse.cz