On Sun 03-07-16 11:35:56, Tetsuo Handa wrote: > This is my alternative proposal compared to what Michal posted at > http://lkml.kernel.org/r/1467365190-24640-1-git-send-email-mhocko@xxxxxxxxxx . > > The series is based on top of linux-next-20160701 + > http://lkml.kernel.org/r/1467201562-6709-1-git-send-email-mhocko@xxxxxxxxxx . > > The key point of the series is [PATCH 3/8]. I have only checked the diff between the whole patchset applied with what I have posted as an RFC last week, so I cannot comment on specific patches. Let me summarize the differences between the two approaches though. My proposal adds a stable reference of the killed mm struct into the signal struct and most oom decisions can refer to this mm and its flags because signal struct life time exceeds its visible task_struct. We still need signal->oom_victims counter to catch different threads lifetime. Yours enqueues the mm to a linked list and has a similar effect with an advantage that signal->oom_victims is no longer needed because you have pulled the OOM_SCAN_ABORT out of select_bad_process to earlier {mem_cgroup_}out_of_memory and check existence of a compatible mm for the oom domain. This means that mm struct has to remember all the information that might be gone by the time we look at the enqueued mm again. This means a slightly larger memory foot print (nothing earth shattering though). That being said, I believe both approaches are sound. So let's discuss {ad,dis}vantages of those approaches. You are introducing more code but to be fair I guess the mm rather than task queuing is better long term. Copying the state for later use is unfortunate but it might turn out better to have all the oom specific stuff inside the mm rather than spread around in other structures. As I've said I haven't looked very deeply into details but at least memcg handling would need more work, I will respond to the specific patch. I guess the mm visibility is basically same with both approaches. Even though you hide the mm from __mmput while mine has it alive until signal struct goes away this is basically the equivalent because mine is hiding the mm with MMF_OOM_REAPED from the oom reaper and oom_reaper is just a weaker form of __mmput. I am not tight to my approach but could you name main arguments why you think yours is better? -- 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>