Michal Hocko wrote: > I do not think the placement in find_lock_task_mm is desirable nor > correct. This function is used in multiple contexts outside of the oom > proper. It only returns a locked task_struct for a thread that belongs > to the process. OK. Andrew, please drop from -mm tree for now. > What you are seeing is clearly undesirable of course but I believe we > should handle it at oom_kill_process layer. Blindly selecting a child > process even when it doesn't sit on some memory or when it has already > been killed is wrong. The heuristic is clearly too naive and so we > should touch it rather than compensating it somewhere else. What about > the following simple approach? It does two things and I will split it > up if this looks like a desirable approach. Please note I haven't tested > it because it is more of an idea than a finished thing. What do you think? I think we need to filter at select_bad_process() and oom_kill_process(). When P has no children, P is chosen and TIF_MEMDIE is set on P. But P can be chosen forever due to P->signal->oom_score_adj == OOM_SCORE_ADJ_MAX even if the OOM reaper reclaimed P's mm. We need to ensure that oom_kill_process() is not called with P if P already has TIF_MEMDIE. (By the way, we are already assuming the OOM reaper kernel thread is available. Changing to BUG_ON(IS_ERR(oom_reaper_th)) should be OK. ;-) ) -- 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>