From: "Rafael J. Wysocki" <rafael@kernel.org>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Giovanni Gherdovich <ggherdovich@suse.cz>,
Borislav Petkov <bp@alien8.de>, Ingo Molnar <mingo@redhat.com>,
"Rafael J . Wysocki" <rjw@rjwysocki.net>,
Viresh Kumar <viresh.kumar@linaro.org>,
Jon Grimm <Jon.Grimm@amd.com>,
Nathan Fontenot <Nathan.Fontenot@amd.com>,
Yazen Ghannam <Yazen.Ghannam@amd.com>,
Thomas Lendacky <Thomas.Lendacky@amd.com>,
Suthikulpanit Suravee <Suravee.Suthikulpanit@amd.com>,
Mel Gorman <mgorman@techsingularity.net>, Pu Wen <puwen@hygon.cn>,
Juri Lelli <juri.lelli@redhat.com>,
Vincent Guittot <vincent.guittot@linaro.org>,
Dietmar Eggemann <dietmar.eggemann@arm.com>,
Michael Larabel <Michael@phoronix.com>,
"the arch/x86 maintainers" <x86@kernel.org>,
Linux PM <linux-pm@vger.kernel.org>,
Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
ACPI Devel Maling List <linux-acpi@vger.kernel.org>
Subject: Re: [PATCH v2 1/1] x86,sched: On AMD EPYC set freq_max = max_boost in schedutil invariant formula
Date: Tue, 2 Feb 2021 19:40:26 +0100 [thread overview]
Message-ID: <CAJZ5v0iu9eqhwM58NNfkG5k=d4mQsA9SiRAnajpd7e0+Ub7k1A@mail.gmail.com> (raw)
In-Reply-To: <YA6XmO2nuivdpE8M@hirez.programming.kicks-ass.net>
On Mon, Jan 25, 2021 at 11:11 AM Peter Zijlstra <peterz@infradead.org> wrote:
>
> On Fri, Jan 22, 2021 at 09:40:38PM +0100, Giovanni Gherdovich wrote:
> > This workload is constant in time, so instead of using the PELT sum we can
> > pretend that scale invariance is obtained with
> >
> > util_inv = util_raw * freq_curr / freq_max1 [formula-1]
> >
> > where util_raw is the PELT util from v5.10 (which is to say, not invariant),
> > and util_inv is the PELT util from v5.11-rc4. freq_max1 comes from
> > commit 976df7e5730e ("x86, sched: Use midpoint of max_boost and max_P for
> > frequency invariance on AMD EPYC") and is (P0+max_boost)/2 = (2.25+3.4)/2 =
> > 2.825 GHz. Then we have the schedutil formula
> >
> > freq_next = 1.25 * freq_max2 * util_inv [formula-2]
> >
> > Here v5.11-rc4 uses freq_max2 = P0 = 2.25 GHz (and this patch changes it to
> > 3.4 GHz).
> >
> > Since all cores are busy, there is no boost available. Let's be generous and say
> > the tasks initially get P0, i.e. freq_curr = 2.25 GHz. Combining the formulas
> > above and taking util_raw = 825/1024 = 0.8, freq_next is:
> >
> > freq_next = 1.25 * 2.25 * 0.8 * 2.25 / 2.825 = 1.79 GHz
>
> Right, so here's a 'problem' between schedutil and cpufreq, they don't
> use the same f_max at all times.
>
> And this is also an inconsistency between acpi_cpufreq and intel_pstate
> (passive). IIRC the intel_pstate cpufreq drivers uses 4C/1C/P0 resp,
> while ACPI seems to stick to P0 f_max.
The only place where 4C is used is the scale invariance code AFAICS.
intel_pstate uses P0 as the f_max unless turbo is disabled.
The difference between intel_pstate and acpi_cpufreq is that (a) the
latter uses a frequency table and the former doesn't and (b) the
latter uses the P0 entry of the frequency table to represent the
entire turbo range,
> Rafael; should ACPI change that behaviour rather than adding yet another
> magic variable?
I'm not sure. That may change the behavior from what is expected by some users.
next prev parent reply other threads:[~2021-02-02 18:44 UTC|newest]
Thread overview: 26+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-01-22 20:40 [PATCH v2 0/1] AMD EPYC: fix schedutil perf regression (freq-invariance) Giovanni Gherdovich
2021-01-22 20:40 ` [PATCH v2 1/1] x86,sched: On AMD EPYC set freq_max = max_boost in schedutil invariant formula Giovanni Gherdovich
2021-01-25 10:04 ` Peter Zijlstra
2021-01-26 9:28 ` Giovanni Gherdovich
2021-01-26 10:02 ` Peter Zijlstra
2021-02-02 18:45 ` Rafael J. Wysocki
2021-02-02 19:11 ` Rafael J. Wysocki
2021-02-03 9:56 ` Giovanni Gherdovich
2021-02-02 18:40 ` Rafael J. Wysocki [this message]
2021-01-25 10:06 ` Peter Zijlstra
2021-01-26 9:09 ` Giovanni Gherdovich
2021-01-26 9:31 ` Mel Gorman
2021-01-26 10:05 ` Peter Zijlstra
[not found] ` <1611933781.15858.48.camel@suse.cz>
2021-02-02 14:17 ` Giovanni Gherdovich
2021-02-02 18:21 ` Peter Zijlstra
2021-02-02 18:29 ` Rafael J. Wysocki
2021-02-02 19:00 ` Rafael J. Wysocki
2021-02-02 18:59 ` Rafael J. Wysocki
2021-02-02 19:26 ` Rafael J. Wysocki
2021-02-03 8:39 ` Giovanni Gherdovich
2021-02-03 13:40 ` Rafael J. Wysocki
2021-02-03 9:12 ` Giovanni Gherdovich
2021-02-03 6:04 ` Viresh Kumar
2021-01-24 22:30 ` [PATCH v2 0/1] AMD EPYC: fix schedutil perf regression (freq-invariance) Michael Larabel
2021-01-25 8:34 ` Peter Zijlstra
2021-01-26 9:01 ` Giovanni Gherdovich
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAJZ5v0iu9eqhwM58NNfkG5k=d4mQsA9SiRAnajpd7e0+Ub7k1A@mail.gmail.com' \
--to=rafael@kernel.org \
--cc=Jon.Grimm@amd.com \
--cc=Michael@phoronix.com \
--cc=Nathan.Fontenot@amd.com \
--cc=Suravee.Suthikulpanit@amd.com \
--cc=Thomas.Lendacky@amd.com \
--cc=Yazen.Ghannam@amd.com \
--cc=bp@alien8.de \
--cc=dietmar.eggemann@arm.com \
--cc=ggherdovich@suse.cz \
--cc=juri.lelli@redhat.com \
--cc=linux-acpi@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-pm@vger.kernel.org \
--cc=mgorman@techsingularity.net \
--cc=mingo@redhat.com \
--cc=peterz@infradead.org \
--cc=puwen@hygon.cn \
--cc=rjw@rjwysocki.net \
--cc=vincent.guittot@linaro.org \
--cc=viresh.kumar@linaro.org \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).