On 2019/06/15 10:10, Tetsuo Handa wrote: > I'm not sure this patch is correct/safe. Can you try memcg OOM torture > test (including memcg group OOM killing enabled) with this patch applied? Well, I guess this patch was wrong. The ordering of removing threads does not matter as long as we start traversing via signal_struct. The reason why crashed at for_each_thread() is unknown...
