cgroup: Allow 5% of extra pages when RSS only limit

Moving pages to swap is a background task and can take a while. In the
mean-time the proces might alocate new pages breaking the limit
temporarily. Let's allow 5% overcommit in such case.

Note: this should never happen with mem+swap limit, because we know in
advance how much swap and mem we have left.
Signed-off-by: NLukáš Doktor <ldoktor@redhat.com>

qemu/tests/cgroup.py

@@ -1941,9 +1941,10 @@ def run(test, params, env):
                     logging.info(out)
             else:   # only RSS limit
                 exit_nr = session.cmd_output("echo $?")[:-1]
-                if max_rss > mem_limit:
-                    err = ("The limit was broken: max_rss=%s, limit=%s" %
-                           (max_rss, mem_limit))
+                if max_rss > mem_limit * 1.05:
+                    # Allow 5% pages to be in-progress of swapping out
+                    err = ("The limit was broken: max_rss=%s, limit=%s (+5%%)"
+                           % (max_rss, mem_limit))
                 elif exit_nr != '0':
                     err = ("dd command failed(%s) output: %s" % (exit_nr, out))
                 elif (max_rssswap) < mem_limit: