On Thu 15-06-17 14:26:26, David Rientjes wrote: > On Thu, 15 Jun 2017, Michal Hocko wrote: > > > > If mm->mm_users is not incremented because it is already zero by the oom > > > reaper, meaning the final refcount has been dropped, do not set > > > MMF_OOM_SKIP prematurely. > > > > > > __mmput() may not have had a chance to do exit_mmap() yet, so memory from > > > a previous oom victim is still mapped. > > > > true and do we have a _guarantee_ it will do it? E.g. can somebody block > > exit_aio from completing? Or can somebody hold mmap_sem and thus block > > ksm_exit resp. khugepaged_exit from completing? The reason why I was > > conservative and set such a mm as MMF_OOM_SKIP was because I couldn't > > give a definitive answer to those questions. And we really _want_ to > > have a guarantee of a forward progress here. Killing an additional > > proecess is a price to pay and if that doesn't trigger normall it sounds > > like a reasonable compromise to me. > > > > I have not seen any issues where __mmput() stalls and exit_mmap() fails to > free its mapped memory once mm->mm_users has dropped to 0. > > > > __mput() naturally requires no > > > references on mm->mm_users to do exit_mmap(). > > > > > > Without this, several processes can be oom killed unnecessarily and the > > > oom log can show an abundance of memory available if exit_mmap() is in > > > progress at the time the process is skipped. > > > > Have you seen this happening in the real life? > > > > Yes, quite a bit in testing. > > One oom kill shows the system to be oom: > > [22999.488705] Node 0 Normal free:90484kB min:90500kB ... > [22999.488711] Node 1 Normal free:91536kB min:91948kB ... > > followed up by one or more unnecessary oom kills showing the oom killer > racing with memory freeing of the victim: > > [22999.510329] Node 0 Normal free:229588kB min:90500kB ... > [22999.510334] Node 1 Normal free:600036kB min:91948kB ... > > The patch is absolutely required for us to prevent continuous oom killing > of processes after a single process has been oom killed and its memory is > in the process of being freed. OK, could you play with the patch/idea suggested in http://lkml.kernel.org/r/20170615122031.GL1486@xxxxxxxxxxxxxx? -- Michal Hocko SUSE Labs -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
