On 2020/03/11 7:54, David Rientjes wrote: > The patch certainly prevents unnecessary oom kills when there is a pending > victim that uncharges its memory between invoking the oom killer and > finding MMF_OOM_SKIP in the list of eligible tasks and its much more > common on systems with limited cpu cores. I think that it is dump_header() which currently spends much time (due to synchronous printing) enough to make "the second memcg oom kill shows usage is >40MB below its limit of 100MB" happen. Shouldn't we call dump_header() and then do the last check and end with "but did not kill anybody" message?
