From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751419AbdJDVAh (ORCPT ); Wed, 4 Oct 2017 17:00:37 -0400 Received: from gum.cmpxchg.org ([85.214.110.215]:53152 "EHLO gum.cmpxchg.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751120AbdJDVAg (ORCPT ); Wed, 4 Oct 2017 17:00:36 -0400 Date: Wed, 4 Oct 2017 17:00:27 -0400 From: Johannes Weiner To: Tetsuo Handa Cc: Andrew Morton , Alan Cox , Christoph Hellwig , Michal Hocko , linux-mm@kvack.org, linux-kernel@vger.kernel.org, kernel-team@fb.com Subject: Re: [PATCH 1/2] Revert "vmalloc: back off when the current task is killed" Message-ID: <20171004210027.GA2973@cmpxchg.org> References: <20171003225504.GA966@cmpxchg.org> <20171004185813.GA2136@cmpxchg.org> <20171004185906.GB2136@cmpxchg.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.8.3 (2017-05-23) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Oct 05, 2017 at 05:49:43AM +0900, Tetsuo Handa wrote: > On 2017/10/05 3:59, Johannes Weiner wrote: > > But the justification to make that vmalloc() call fail like this isn't > > convincing, either. The patch mentions an OOM victim exhausting the > > memory reserves and thus deadlocking the machine. But the OOM killer > > is only one, improbable source of fatal signals. It doesn't make sense > > to fail allocations preemptively with plenty of memory in most cases. > > By the time the current thread reaches do_exit(), fatal_signal_pending(current) > should become false. As far as I can guess, the source of fatal signal will be > tty_signal_session_leader(tty, exit_session) which is called just before > tty_ldisc_hangup(tty, cons_filp != NULL) rather than the OOM killer. I don't > know whether it is possible to make fatal_signal_pending(current) true inside > do_exit() though... It's definitely not the OOM killer, the memory situation looks fine when this happens. I didn't look closer where the signal comes from. That said, we trigger this issue fairly easily. We tested the revert over night on a couple thousand machines, and it fixed the issue (whereas the control group still saw the crashes).