On Fri 06-09-19 22:41:39, Tetsuo Handa wrote: > On 2019/09/06 22:08, Qian Cai wrote: > > Yes, mlocked is troublesome. I have other incidents where crond and systemd- > > udevd were killed by mistake, > > Yes. How to mitigate this regression is a controversial topic. > Michal thinks that we should make mlocked pages reclaimable, but > we haven't reached there. I think that we can monitor whether > counters decay over time (with timeout), but Michal refuses any > timeout based approach. We are deadlocked there. There is really nothing controversial here. It just needs somebody to do the actual work. Besides that I haven't seen any real workload to mlock so much memory that there won't be anything reapable. LTP oom test simply doesn't represent any real world workload. So while not optimal I do not really think this is something to lose sleep over. -- Michal Hocko SUSE Labs
