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 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 4F2BBC433F5 for ; Mon, 15 Nov 2021 02:24:56 +0000 (UTC) Received: from kanga.kvack.org (kanga.kvack.org [205.233.56.17]) by mail.kernel.org (Postfix) with ESMTP id CE514610A6 for ; Mon, 15 Nov 2021 02:24:55 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.4.1 mail.kernel.org CE514610A6 Authentication-Results: mail.kernel.org; dmarc=fail (p=none dis=none) header.from=gmail.com Authentication-Results: mail.kernel.org; spf=pass smtp.mailfrom=kvack.org Received: by kanga.kvack.org (Postfix) id 655726B007B; Sun, 14 Nov 2021 21:24:55 -0500 (EST) Received: by kanga.kvack.org (Postfix, from userid 40) id 603616B007D; Sun, 14 Nov 2021 21:24:55 -0500 (EST) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id 4CB8C6B007E; Sun, 14 Nov 2021 21:24:55 -0500 (EST) X-Delivered-To: linux-mm@kvack.org Received: from forelay.hostedemail.com (smtprelay0114.hostedemail.com [216.40.44.114]) by kanga.kvack.org (Postfix) with ESMTP id 3F0546B007B for ; Sun, 14 Nov 2021 21:24:55 -0500 (EST) Received: from smtpin18.hostedemail.com (10.5.19.251.rfc1918.com [10.5.19.251]) by forelay05.hostedemail.com (Postfix) with ESMTP id D3C8A184818C4 for ; Mon, 15 Nov 2021 02:24:54 +0000 (UTC) X-FDA: 78809571708.18.F587BA5 Received: from mail-qv1-f47.google.com (mail-qv1-f47.google.com [209.85.219.47]) by imf25.hostedemail.com (Postfix) with ESMTP id 80422B000180 for ; Mon, 15 Nov 2021 02:24:41 +0000 (UTC) Received: by mail-qv1-f47.google.com with SMTP id m17so10336678qvx.8 for ; Sun, 14 Nov 2021 18:24:54 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20210112; h=mime-version:references:in-reply-to:from:date:message-id:subject:to :cc; bh=DKSWxpIr3zViMwXzWHpUhtajLyKeIiM7AaaIXOUY46E=; b=HAAHptK8yUr2RfCi7gFPbSimiuK/fL1EYdE7BD/IG3IEovGjABrTVPWul9pG6OQkuA veE6s+FHeIs90EVkBISga89t/jvFzWEIJKHwcpi111dUEagYLCsT32hjA6WqYjtCOzT5 mHX2sBSut3b95xd6KjS4AM8pevSsUbbCiBEXjBdtS93QgEPwiYiGC+lYG2vLbSnTQHkX m/dT3g9BJ0ZQFIy4VZjo902PtGh2EQP4E2M3bV+BuXK5oVOd35V5UWJjTW3Dt2oMelA9 cbgzoq5mScnLjGLe67GhP2VT1VsOtz7dBCwF05CbcgMKKnUfg34SQhB+Dd4wsbjuugr4 V04A== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc; bh=DKSWxpIr3zViMwXzWHpUhtajLyKeIiM7AaaIXOUY46E=; b=wOiyVHlJUNTJB2VOvWNrHHPYoEHcZggj46otyT6hVSUDgAd2SWnp4tepYKs1uaibHi 1n9Yo9rmqb5qkHjA46KwSQNv20ari+5DfIOJJRcppHGygs7LaTCYJS19JtTtNm7zVcnI jVOy/8lmmndpIzpnESH7O7tVltYaC9ioF9pb+MB9QFvqeVA/huHJe+ZsXZ7tvwWzre88 eF8eaPVZiBszTdRpLek6ophuJCeSYL+NTA/Y1GU42qSCwIe+Mrx5m6KR1m37qYe5dVxq sn7XfLOc9IPr8vZDtZdriGNMgt3DHckUZUZWTvMPqsPnidN5pL6bZfHdCeIpvUFETrnd /GhA== X-Gm-Message-State: AOAM533O0NcUo09oebM3YjU5F1w3RGvSReUk+9Efz9ajAEsi+t76rsIM xtm0vKCvp/vrk3nTYRWax+Ym1PkypsexebE+zsI= X-Google-Smtp-Source: ABdhPJzlQFx8l1W2SjZV/u/Gduhwanko3nQFvNcW70G59NFPIFI/K5+MaLKYAy4DFfyoVsbaPVVGJ+XfIQDOg5duPQU= X-Received: by 2002:a05:6214:d8e:: with SMTP id e14mr33545375qve.37.1636943093860; Sun, 14 Nov 2021 18:24:53 -0800 (PST) MIME-Version: 1.0 References: <1634278612-17055-1-git-send-email-huangzhaoyang@gmail.com> In-Reply-To: From: Zhaoyang Huang Date: Mon, 15 Nov 2021 10:24:32 +0800 Message-ID: Subject: Re: [Resend PATCH] psi : calc cfs task memstall time more precisely To: Johannes Weiner Cc: Peter Zijlstra , Andrew Morton , Michal Hocko , Vladimir Davydov , Zhaoyang Huang , "open list:MEMORY MANAGEMENT" , LKML , xuewen.yan@unisoc.com, Ke Wang Content-Type: text/plain; charset="UTF-8" Authentication-Results: imf25.hostedemail.com; dkim=pass header.d=gmail.com header.s=20210112 header.b=HAAHptK8; spf=pass (imf25.hostedemail.com: domain of huangzhaoyang@gmail.com designates 209.85.219.47 as permitted sender) smtp.mailfrom=huangzhaoyang@gmail.com; dmarc=pass (policy=none) header.from=gmail.com X-Rspamd-Server: rspam04 X-Rspamd-Queue-Id: 80422B000180 X-Stat-Signature: 8miefiam4twsfy6umdnjczoa1yfbk5ka X-HE-Tag: 1636943081-32227 X-Bogosity: Ham, tests=bogofilter, spamicity=0.000000, version=1.2.4 Sender: owner-linux-mm@kvack.org Precedence: bulk X-Loop: owner-majordomo@kvack.org List-ID: On Sat, Nov 13, 2021 at 12:36 AM Johannes Weiner wrote: > > On Tue, Nov 09, 2021 at 03:56:36PM +0100, Peter Zijlstra wrote: > > On Tue, Nov 02, 2021 at 03:47:33PM -0400, Johannes Weiner wrote: > > > CC peterz as well for rt and timekeeping magic > > > > > > On Fri, Oct 15, 2021 at 02:16:52PM +0800, Huangzhaoyang wrote: > > > > From: Zhaoyang Huang > > > > > > > > In an EAS enabled system, there are two scenarios discordant to current design, > > > > > > > > 1. workload used to be heavy uneven among cores for sake of scheduler policy. > > > > RT task usually preempts CFS task in little core. > > > > 2. CFS task's memstall time is counted as simple as exit - entry so far, which > > > > ignore the preempted time by RT, DL and Irqs. > > > > It ignores preemption full-stop. I don't see why RT/IRQ should be > > special cased here. > > > > > > With these two constraints, the percpu nonidle time would be mainly consumed by > > > > none CFS tasks and couldn't be averaged. Eliminating them by calc the time growth > > > > via the proportion of cfs_rq's utilization on the whole rq. > > > > > > > > +static unsigned long psi_memtime_fixup(u32 growth) > > > > +{ > > > > + struct rq *rq = task_rq(current); > > > > + unsigned long growth_fixed = (unsigned long)growth; > > > > + > > > > + if (!(current->policy == SCHED_NORMAL || current->policy == SCHED_BATCH)) > > > > + return growth_fixed; > > > > + > > > > + if (current->in_memstall) > > > > + growth_fixed = div64_ul((1024 - rq->avg_rt.util_avg - rq->avg_dl.util_avg > > > > + - rq->avg_irq.util_avg + 1) * growth, 1024); > > > > + > > > > + return growth_fixed; > > > > +} > > > > + > > > > static void init_triggers(struct psi_group *group, u64 now) > > > > { > > > > struct psi_trigger *t; > > > > @@ -658,6 +675,7 @@ static void record_times(struct psi_group_cpu *groupc, u64 now) > > > > } > > > > > > > > if (groupc->state_mask & (1 << PSI_MEM_SOME)) { > > > > + delta = psi_memtime_fixup(delta); > > > > > > Ok, so we want to deduct IRQ and RT preemption time from the memstall > > > period of an active reclaimer, since it's technically not stalled on > > > memory during this time but on CPU. > > > > > > However, we do NOT want to deduct IRQ and RT time from memstalls that > > > are sleeping on refaults swapins, since they are not affected by what > > > is going on on the CPU. > > > > I think that focus on RT/IRQ is mis-guided here, and the implementation > > is horrendous. > > > > So the fundamental question seems to be; and I think Johannes is the one > > to answer that: What time-base do these metrics want to use? > > > > Do some of these states want to account in task-time instead of > > wall-time perhaps? I can't quite remember, but vague memories are > > telling me most of the PSI accounting was about blocked tasks, not > > running tasks, which makes all this rather more complicated. > > > > Randomly scaling time as proposed seems almost certainly wrong. What > > would that make the stats mean? > > It *could* be argued that IRQs and RT preemptions are CPU stalls. > > I'm less convinced we should subtract preemptions from memory stalls. > > Yes, when you're reclaiming and you get preempted for whatever reason, > you're technically stalled on CPU in this moment. However, reclaim > itself consumes CPU and walltime, and it could be what is causing > those preemptions to begin with! For example, reclaim could eat up 90% > of your scheduling timeslice and then cause a preemption when the > thread is back in userspace and trying to be productive. By consuming > time, it also drags out the overall timeline for userspace to finish > its work, and a longer timeline will have more disruptions from > independent events like IRQs and RT thread wakeups. > > So if you *were* to discount CPU contention from memory stalls, it > would also mean that you'd have to count *memory stalls* when > userspace experiences CPU contention caused by preceding reclaims. I > don't think it makes sense to try to go down that road... > > They're dependent resources. Just like faster CPUs and faster IO > devices mean less memory pressure for the same amount of reclaim and > paging activity, it seems logical that contention of those underlying > resources will result in longer memory stalls and higher pressure. Imagine that two triggers created on CPU and MEMORY with one RT non-memstall process consume 90% of the rq's util while a memstall CFS process get the rest of 10%. The problem is we will be misguided as both of the resources are busy under current mechanisms.