From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-2.4 required=3.0 tests=DKIM_SIGNED,DKIM_VALID, DKIM_VALID_AU,HEADER_FROM_DIFFERENT_DOMAINS,MAILING_LIST_MULTI,SPF_HELO_NONE, SPF_PASS,USER_AGENT_SANE_1 autolearn=no autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 89317C742A2 for ; Thu, 11 Jul 2019 19:58:45 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 50E6B208E4 for ; Thu, 11 Jul 2019 19:58:45 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=pass (1024-bit key) header.d=joelfernandes.org header.i=@joelfernandes.org header.b="q+taS8xQ" Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729005AbfGKT6n (ORCPT ); Thu, 11 Jul 2019 15:58:43 -0400 Received: from mail-pl1-f195.google.com ([209.85.214.195]:45535 "EHLO mail-pl1-f195.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1728931AbfGKT6n (ORCPT ); Thu, 11 Jul 2019 15:58:43 -0400 Received: by mail-pl1-f195.google.com with SMTP id y8so3558919plr.12 for ; Thu, 11 Jul 2019 12:58:42 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=joelfernandes.org; s=google; h=date:from:to:cc:subject:message-id:references:mime-version :content-disposition:content-transfer-encoding:in-reply-to :user-agent; bh=ht9JIkdKnsND0qRfdxvtr9JOrmvYFxMRa0bAr2nWrT8=; b=q+taS8xQftTPcsu718h0VpXS4k0nrAc3iWEcqUf6v5fSr4dJ2XXxIsWOSkx3b8ISS4 UR+u0ZqbrLa6iN+GGEqoRwP/jIexCCBuYa9mZAe0NN4UuwiqJct5pSFqdU/72RZM//M4 PDD9mnKb7UVUZWH5yvsRk0b6AwObX0/b8yxV8= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:date:from:to:cc:subject:message-id:references :mime-version:content-disposition:content-transfer-encoding :in-reply-to:user-agent; bh=ht9JIkdKnsND0qRfdxvtr9JOrmvYFxMRa0bAr2nWrT8=; b=X63tmWLMDG5vj1uBqrOyqn0Bt3saK03u+fSnH1X0w+uSHMyOJNBwD1OP3p7V2bTXf5 ZjJ9MpaVKWBgZorX6wh6TzJAqpmDHdtveK5daPW0ESqwu7TKh/eHexX7BYAiyzx+JbKQ Hpt0CN5IFuOVVnU3mRDonLLrbWjEYN06A6UmnxLESEpODglqSs+nbqdjtk8Rx3WxxQx8 +JloNhyR/oR46B9YTmKs4+ohaHPKTDjwPbInb9pgW3gtAfKs8GSy5tlTMmUK7xIaJFD+ /vuZ2NcRSrEJCflI3XfX2U2TdaQ2O3E7w3a9gieQwkOitxCjt5yAARjj0g4hdFXrRaRZ Wlew== X-Gm-Message-State: APjAAAW3GxAv76IVzgBr2onJ6HCeZNlvLRyGWXRNn4sn5KR0Y9aNv+r7 Ju4tqDUJmvslze7miDtEpzg= X-Google-Smtp-Source: APXvYqwGHjCIXofIQD1MYuS6DMYCoGyPvFzy/BnOMEeKNsXK91RwKc4AOWUpMANIiyqd0XHBR+arvw== X-Received: by 2002:a17:902:44f:: with SMTP id 73mr6700269ple.192.1562875122087; Thu, 11 Jul 2019 12:58:42 -0700 (PDT) Received: from localhost ([2620:15c:6:12:9c46:e0da:efbf:69cc]) by smtp.gmail.com with ESMTPSA id q144sm6337161pfc.103.2019.07.11.12.58.40 (version=TLS1_3 cipher=AEAD-AES256-GCM-SHA384 bits=256/256); Thu, 11 Jul 2019 12:58:41 -0700 (PDT) Date: Thu, 11 Jul 2019 15:58:39 -0400 From: Joel Fernandes To: "Paul E. McKenney" Cc: Byungchul Park , josh@joshtriplett.org, rostedt@goodmis.org, mathieu.desnoyers@efficios.com, jiangshanlai@gmail.com, rcu@vger.kernel.org, linux-kernel@vger.kernel.org, kernel-team@lge.com Subject: Re: [PATCH] rcu: Make jiffies_till_sched_qs writable Message-ID: <20190711195839.GA163275@google.com> References: <1562565609-12482-1-git-send-email-byungchul.park@lge.com> <20190708125013.GG26519@linux.ibm.com> <20190708130359.GA42888@google.com> <20190709055815.GA19459@X58A-UD3R> <20190709124102.GR26519@linux.ibm.com> <20190710012025.GA20711@X58A-UD3R> <20190711123052.GI26519@linux.ibm.com> <20190711130849.GA212044@google.com> <20190711150215.GK26519@linux.ibm.com> <20190711164818.GA260447@google.com> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: <20190711164818.GA260447@google.com> User-Agent: Mutt/1.10.1 (2018-07-13) Sender: linux-kernel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Jul 11, 2019 at 12:48:18PM -0400, Joel Fernandes wrote: > On Thu, Jul 11, 2019 at 08:02:15AM -0700, Paul E. McKenney wrote: > > On Thu, Jul 11, 2019 at 09:08:49AM -0400, Joel Fernandes wrote: > > > On Thu, Jul 11, 2019 at 05:30:52AM -0700, Paul E. McKenney wrote: > > > > On Wed, Jul 10, 2019 at 10:20:25AM +0900, Byungchul Park wrote: > > > > > On Tue, Jul 09, 2019 at 05:41:02AM -0700, Paul E. McKenney wrote: > > > > > > > Hi Paul, > > > > > > > > > > > > > > IMHO, as much as we want to tune the time for fqs to be initiated, we > > > > > > > can also want to tune the time for the help from scheduler to start. > > > > > > > I thought only difference between them is a level of urgency. I might be > > > > > > > wrong. It would be appreciated if you let me know if I miss something. > > > > > > > > > > > > Hello, Byungchul, > > > > > > > > > > > > I understand that one hypothetically might want to tune this at runtime, > > > > > > but have you had need to tune this at runtime on a real production > > > > > > workload? If so, what problem was happening that caused you to want to > > > > > > do this tuning? > > > > > > > > > > Not actually. > > > > > > > > > > > > And it's ok even if the patch is turned down based on your criteria. :) > > > > > > > > > > > > If there is a real need, something needs to be provided to meet that > > > > > > need. But in the absence of a real need, past experience has shown > > > > > > that speculative tuning knobs usually do more harm than good. ;-) > > > > > > > > > > It makes sense, "A speculative tuning knobs do more harm than good". > > > > > > > > > > Then, it would be better to leave jiffies_till_{first,next}_fqs tunnable > > > > > but jiffies_till_sched_qs until we need it. > > > > > > > > > > However, > > > > > > > > > > (1) In case that jiffies_till_sched_qs is tunnable: > > > > > > > > > > We might need all of jiffies_till_{first,next}_qs, > > > > > jiffies_till_sched_qs and jiffies_to_sched_qs because > > > > > jiffies_to_sched_qs can be affected by any of them. So we > > > > > should be able to read each value at any time. > > > > > > > > > > (2) In case that jiffies_till_sched_qs is not tunnable: > > > > > > > > > > I think we don't have to keep the jiffies_till_sched_qs any > > > > > longer since that's only for setting jiffies_to_sched_qs at > > > > > *booting time*, which can be done with jiffies_to_sched_qs too. > > > > > It's meaningless to keep all of tree variables. > > > > > > > > > > The simpler and less knobs that we really need we have, the better. > > > > > > > > > > what do you think about it? > > > > > > > > > > In the following patch, I (1) removed jiffies_till_sched_qs and then > > > > > (2) renamed jiffies_*to*_sched_qs to jiffies_*till*_sched_qs because I > > > > > think jiffies_till_sched_qs is a much better name for the purpose. I > > > > > will resend it with a commit msg after knowing your opinion on it. > > > > > > > > I will give you a definite "maybe". > > > > > > > > Here are the two reasons for changing RCU's embarrassingly large array > > > > of tuning parameters: > > > > > > > > 1. They are causing a problem in production. This would represent a > > > > bug that clearly must be fixed. As you say, this change is not > > > > in this category. > > > > > > > > 2. The change simplifies either RCU's code or the process of tuning > > > > RCU, but without degrading RCU's ability to run everywhere and > > > > without removing debugging tools. > > > > > > > > The change below clearly simplifies things by removing a few lines of > > > > code, and it does not change RCU's default self-configuration. But are > > > > we sure about the debugging aspect? (Please keep in mind that many more > > > > sites are willing to change boot parameters than are willing to patch > > > > their kernels.) > > > > > > > > What do you think? > > > > > > Just to add that independent of whether the runtime tunable make sense or > > > not, may be it is still worth correcting the 0444 to be 0644 to be a separate > > > patch? > > > > You lost me on this one. Doesn't changing from 0444 to 0644 make it be > > a runtime tunable? > > I was going by our earlier discussion that the parameter is still writable at > boot time. You mentioned something like the following: > --- > In Byungchul's defense, the current module_param() permissions are > 0444, which really is read-only. Although I do agree that they can > be written at boot, one could use this same line of reasoning to argue > that const variables can be written at compile time (or, for on-stack > const variables, at function-invocation time). But we still call them > "const". > --- > > Sorry if I got confused. You are right that we could leave it as read-only. > > > > > Finally, I urge you to join with Joel Fernandes and go through these > > > > grace-period-duration tuning parameters.  Once you guys get your heads > > > > completely around all of them and how they interact across the different > > > > possible RCU configurations, I bet that the two of you will have excellent > > > > ideas for improvement. > > > > > > Yes, I am quite happy to join forces. Byungchul, let me know what about this > > > or other things you had in mind. I have some other RCU topics too I am trying > > > to get my head around and planning to work on more patches. > > > > > > Paul, in case you had any other specific tunables or experiments in mind, let > > > me know. I am quite happy to try out new experiments and learn something > > > based on tuning something. > > > > These would be the tunables controlling how quickly RCU takes its > > various actions to encourage the current grace period to end quickly. > > I would be happy to give you the exact list if you wish, but most of > > them have appeared in this thread. > > > > The experiments should be designed to work out whether the current > > default settings have configurations where they act badly.  This might > > also come up with advice for people attempting hand-tuning, or proposed > > parameter-checking code to avoid bad combinations. > > > > For one example, setting the RCU CPU stall timeout too low will definitely > > cause some unwanted splats.  (Yes, one could argue that other things in > > the kernel should change to allow this value to decrease, but things > > like latency tracer and friends are probably more useful and important.) > > Ok, thank you for the hints. Hmm, speaking of grace period durations, it seems to me the maximum grace period ever is recorded in rcu_state.gp_max. However it is not read from anywhere. Any idea why it was added but not used? I am interested in dumping this value just for fun, and seeing what I get. I wonder also it is useful to dump it in rcutorture/rcuperf to find any issues, or even expose it in sys/proc fs to see what worst case grace periods look like. - Joel