From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751575AbeFDSDb (ORCPT ); Mon, 4 Jun 2018 14:03:31 -0400 Received: from mx3-rdu2.redhat.com ([66.187.233.73]:56820 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751445AbeFDSD2 (ORCPT ); Mon, 4 Jun 2018 14:03:28 -0400 Date: Mon, 4 Jun 2018 13:03:25 -0500 From: Josh Poimboeuf To: Miroslav Benes Cc: jikos@kernel.org, jeyu@kernel.org, pmladek@suse.com, live-patching@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH 1/2] livepatch: Send a fake signal periodically Message-ID: <20180604180325.yeewbafxpjkt6gi5@treble> References: <20180604141636.11523-1-mbenes@suse.cz> <20180604141636.11523-2-mbenes@suse.cz> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline In-Reply-To: <20180604141636.11523-2-mbenes@suse.cz> User-Agent: NeoMutt/20180323 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, Jun 04, 2018 at 04:16:35PM +0200, Miroslav Benes wrote: > An administrator may send a fake signal to all remaining blocking tasks > of a running transition by writing to > /sys/kernel/livepatch//signal attribute. Let's do it > automatically after 10 seconds. The timeout is chosen deliberately. It > gives the tasks enough time to transition themselves. > > Theoretically, sending it once should be more than enough. Better be safe > than sorry, so send it periodically. This is the part I don't understand. Why do it periodically? Instead, might it make sense to just send the signals once, and if that doesn't work, reverse the transition? Then we could make patching a synchronous operation. But then, it might be remotely possible that the reverse operation also stalls (e.g., on a kthread). So, maybe it's best to just leave all these controls in the hands of the user. All that said, a few code review comments: - AFAICT, it does an 8 second delay instead of a 10 second delay, because a) try_complete_transition() is first called before there's any delay; b) the preincrement operator used on signals_cnt. - I think 15 seconds might be a better default. I've seen longer patching delays on a system with 100+ CPUs. - If a kthread or idle task is sleeping on a patched function, the pr_notice("signaling remaining tasks\n") will be repeated continously. - It might be cleaner to do it from the delayed work function (klp_transition_work_fn). -- Josh