Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   "Doug Smythies" <dsmythies@telus.net>
To:     "'Chen Yu'" <yu.c.chen@intel.com>
Cc:     "'Len Brown'" <lenb@kernel.org>,
        "'Rafael J. Wysocki'" <rjw@rjwysocki.net>,
        <linux-kernel@vger.kernel.org>, <linux-pm@vger.kernel.org>
References: <cover.1586782089.git.yu.c.chen@intel.com> <db96fd31afd0ff65e4041665293b96c984e675bc.1586782089.git.yu.c.chen@intel.com>
In-Reply-To: <db96fd31afd0ff65e4041665293b96c984e675bc.1586782089.git.yu.c.chen@intel.com>
Subject: RE: [PATCH 2/3][v2] tools/power turbostat: Introduce functions to accumulate RAPL consumption
Date:   Wed, 15 Apr 2020 21:03:34 -0700
Message-ID: <001901d613a4$010e0a70$032a1f50$@net>
MIME-Version: 1.0
Content-Type: text/plain;
        charset="us-ascii"
Content-Transfer-Encoding: 7bit
Content-Language: en-ca
Thread-Index: AdYSXDnMXkJ5RB5zStyDB6tksB289gBRJlyA
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 2020.04.15 05:57 Chen Yu wrote:

...

> v2: According to Len's suggestion:
>    1. Enable the accumulated RAPL mechanism by default.

I am not a fan of this, but O.K.

>    2. Re-use the rapl_joule_counter_range to represent the
>       the timeout of periodical timer.

No, please no. It is too easy to still have an overflow.

...
> +	/*
> +	 * A wraparound time is calculated early.
> +	 */
> +	its.it_interval.tv_sec = rapl_joule_counter_range;

Would this be o.K.?

+	its.it_interval.tv_sec = rapl_joule_counter_range / 2;

> +	its.it_interval.tv_nsec = 0;

The way it was sent, this patch set does not work.
It still overflows.

Example, sample time calculated to ensure overflow:

Busy%   Bzy_MHz IRQ     PkgTmp  PkgWatt GFXWatt
100.00  3500    3592125 80      9.72    0.12
100.00  3500    3587391 79      9.77    0.12

Actual package watts was around 65.

However, if this additional patch is applied (I only fixed one of them):

doug@s18:~/temp-k-git/linux/tools/power/x86/turbostat$ git diff
diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c
index 29fc4069f467..4d72d9be5209 100644
--- a/tools/power/x86/turbostat/turbostat.c
+++ b/tools/power/x86/turbostat/turbostat.c
@@ -1350,7 +1350,8 @@ delta_package(struct pkg_data *new, struct pkg_data *old)

        old->gfx_mhz = new->gfx_mhz;

-       DELTA_WRAP32(new->energy_pkg, old->energy_pkg);
+/*     DELTA_WRAP32(new->energy_pkg, old->energy_pkg);  */
+       old->energy_pkg = new->energy_pkg - old->energy_pkg;
        DELTA_WRAP32(new->energy_cores, old->energy_cores);
        DELTA_WRAP32(new->energy_gfx, old->energy_gfx);
        DELTA_WRAP32(new->energy_dram, old->energy_dram);

Then it seems to work.

Example:

doug@s15:~/temp-turbostat$ sudo ./turbostat --Summary --show Busy%,Bzy_MHz,PkgTmp,PkgWatt,GFXWatt,IRQ --interval 1200
...
RAPL: 690 sec. Joule Counter Range, at 95 Watts
...
Busy%   Bzy_MHz IRQ     PkgTmp  PkgWatt GFXWatt
100.00  3500    3592328 80      64.32   0.12
100.00  3500    3595195 79      64.37   0.12

... Doug