Johannes Weiner wrote: > On Wed, Mar 09, 2016 at 12:15:10AM +0900, Tetsuo Handa wrote: > > Since mem_cgroup_out_of_memory() is called by > > mem_cgroup_oom_synchronize(true) via pagefault_out_of_memory() via > > page fault, and possible allocations between setting PF_EXITING and > > calling exit_mm() are tty_audit_exit() and taskstats_exit() which will > > not trigger page fault, task_will_free_mem(current) in > > mem_cgroup_out_of_memory() is never true. > > > > Signed-off-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx> > > This opens us up to subtle bugs when somebody later changes the order > and adds new possible allocation sites between the sequence points you > describe above, or maybe adds other mem_cgroup_out_of_memory() callers. > > It looks like a simplification, but it actually complicates things. > If currently not needed, it should be removed. This is for a clarification. Also, what is the reason we do not need below change? I think there is a small race window because oom_killer_disabled needs to be checked after oom_killer_disable() held oom_lock. Is it because all userspace processes except current are frozen before oom_killer_disable() is called and not-yet frozen threads (i.e. kernel threads) never call mem_cgroup_out_of_memory() ? diff --git a/mm/memcontrol.c b/mm/memcontrol.c index ae8b81c..521cd33 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -1253,6 +1253,10 @@ static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, mutex_lock(&oom_lock); + /* Check if we raced with oom_killer_disable(). */ + if (oom_killer_disabled) + goto unlock; + /* * If current has a pending SIGKILL or is exiting, then automatically * select it. The goal is to allow it to allocate so that it may -- 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>