linux-kernel - Re: [PATCH 0/3] perf/x86/rapl: Enable RAPL for AMD Fam17h

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <alpine.LNX.2.20.13.2005161908000.16957@monopod.intra.ispras.ru>
Date:   Sat, 16 May 2020 19:56:22 +0300 (MSK)
From:   Alexander Monakov <amonakov@...ras.ru>
To:     Stephane Eranian <eranian@...gle.com>
cc:     linux-kernel@...r.kernel.org, peterz@...radead.org, mingo@...e.hu,
        irogers@...gle.com, kim.phillips@....com, jolsa@...hat.com
Subject: Re: [PATCH 0/3] perf/x86/rapl: Enable RAPL for AMD Fam17h

On Fri, 15 May 2020, Stephane Eranian wrote:

> The series first moves the rapl.c file to common perf_events x86 and then
> adds the support.
> From the user's point of view, the interface is identical with
> /sys/devices/power. The energy-pkg event is the only one supported.

AMD also has per-core energy metering via MSR 0xc001029a, and I wonder
if you have plans to expose it to perf as well. I see it does not fit
so nicely with the existing code (as it's per-core instead of per-die).

The turbostat tool already exposes these per-core readings:

Core    CPU     Avg_MHz Busy%   Bzy_MHz TSC_MHz CorWatt PkgWatt
-       -       3951    100.00  3951    2373    54.92   30.04
0       0       3945    100.00  3945    2370    8.97    29.98
1       1       3945    100.00  3945    2370    9.11
2       2       3945    100.00  3945    2370    8.96
4       3       3946    100.00  3946    2370    9.32
5       4       3946    100.00  3946    2370    9.11
6       5       3946    100.00  3946    2370    9.39

turbostat sums the per-core energy figures to show the per-socket 54.92W
value. Though as you can see on this example, the figure is not in agreement
with the per-socket MSR you're using in your patch. Maybe the per-core
values are less reliable, but I believe I have a test that shows per-package
figure to be inaccurate as well. It would be nice if AMD clarified how the
counters work.

And, for what (little) it's worth, the series is:

Tested-by: Alexander Monakov <amonakov@...ras.ru>

Thank you.
Alexander