Date: Mon, 31 Aug 2015 21:11:07 +0800
From: Huang Rui <ray.huang@amd.com>
To: Guenter Roeck <linux@roeck-us.net>
CC: Borislav Petkov <bp@suse.de>, Jean Delvare <jdelvare@suse.de>,
        "Andy Lutomirski" <luto@amacapital.net>,
        Andreas Herrmann <herrmann.der.user@gmail.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        "Peter Zijlstra" <peterz@infradead.org>,
        Ingo Molnar <mingo@kernel.org>,
        "Rafael J. Wysocki" <rjw@rjwysocki.net>, Len Brown <lenb@kernel.org>,
        John Stultz <john.stultz@linaro.org>,
        =?iso-8859-1?Q?Fr=E9d=E9ric?= Weisbecker <fweisbec@gmail.com>,
        <lm-sensors@lm-sensors.org>, <linux-kernel@vger.kernel.org>,
        <x86@kernel.org>, Andreas Herrmann <herrmann.der.user@googlemail.com>,
        Aravind Gopalakrishnan <Aravind.Gopalakrishnan@amd.com>,
        Borislav Petkov <bp@alien8.de>,
        "Fengguang Wu" <fengguang.wu@intel.com>, Aaron Lu <aaron.lu@intel.com>,
        Tony Li <tony.li@amd.com>
Subject: Re: [PATCH 12/15] hwmon, fam15h_power: introduce a cpu accumulated
 power reporting algorithm
Message-ID: <20150831131107.GB15543@hr-slim.amd.com>
References: <1440662866-28716-1-git-send-email-ray.huang@amd.com>
 <1440662866-28716-13-git-send-email-ray.huang@amd.com>
 <20150827173043.GB27452@roeck-us.net>
 <20150828104525.GD4191@hr-slim.amd.com>
 <55E06A99.7070800@roeck-us.net>
 <20150831041607.GA15543@hr-slim.amd.com>
 <55E3D864.6090500@roeck-us.net>
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Disposition: inline
In-Reply-To: <55E3D864.6090500@roeck-us.net>
User-Agent: Mutt/1.5.21 (2010-09-15)
Content-Transfer-Encoding: 8BIT
SpamDiagnosticOutput: 1:23
SpamDiagnosticMetadata: NSPM
X-MS-Exchange-CrossTenant-OriginalArrivalTime: 31 Aug 2015 13:10:36.3437
 (UTC)
X-MS-Exchange-CrossTenant-Id: 3dd8961f-e488-4e60-8e11-a82d994e183d
X-MS-Exchange-CrossTenant-OriginalAttributedTenantConnectingIp: TenantId=3dd8961f-e488-4e60-8e11-a82d994e183d;Ip=[165.204.84.222];Helo=[atltwp02.amd.com]
X-MS-Exchange-CrossTenant-FromEntityHeader: HybridOnPrem
X-MS-Exchange-Transport-CrossTenantHeadersStamped: SN1PR12MB0719
X-Microsoft-Exchange-Diagnostics: 1;SN1PR12MB0766;2:UIGSjB18VksjXYxfXc6+ptvxAh8X4KlKPFDVYEXavRnBs4zVofsJDwLUp/HpbBa+wQogtiyfDYWGwgc+W8Sxbj66YnJrVxXa3LYt67DhiaEq2l0T3aRRmvr+tG88G/pYYF72P9a442cLS+DzXV/3st7tuWWrPEXn2rhhNTriI4o=;3:C8nBsceIZseFIYA3Av7Y0a6m5me2Dy7EinJqb7mwqT+qddJa59kibRDwvZiZzCeCYyJH+hwq/6lvdKUCWOQBWGXbPVotaBBumtmVya1/jVb3AezhCPObfmgt1diUfE6Zz2rP/xGeg0fGZM1gdCxH/2x+PyJoMsIn5NwH8aXoQzHp/ou2AztyL0tRW9TL3VogESS8+zUi6iR1pKfYylzmyFp+BAwoVWmKRcsRAqnHH9q2jhG8Yri4rRsa3o/7npvA;25:fFymer+dD/E+SGOIyEePVyguI9jZHMRMeJprM0hQEsiqpSCNRFE2A55UdZgpGcclj5sXHvKTce9CyOek8tqqGyRNV8Cw9uO88pQYVZSrDd51VPBngs60b0RUxZrlFhCX+RwmM+49vGGDGFyDtGe3JUSsgC6cF5XgGyGbhDu3szLfEGYN4sTPS3S7II3/8BYlsLJm1jo+CVvN02PAZ1Ghn45RqjzivzmmuQv1R+63V4Tfibd1ZLysL3ELbCmUJaQarJlxNI7kVXgYJN9PeVuJTQ==;23:zDh/kbOMZ7l+DMyBT0p1ZG1RcuPNjmotjcpTibyXZ3ltPvnbTg2trWV3rFPLxwuVKpnQ1eW46lkoZrOOs9DR+xyp4LknkHvIbRTOGYcrK99WJkKQvGa0bMbKxqv1lkV9D7VLtDMqSXqI9kQ37BLutgVHgHNIWHExgAhSyd8ZvCtEJiLKpv0ojVC/bUMmfAvW
X-OriginatorOrg: amd.com
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org
Content-Length: 5575
Lines: 135

On Sun, Aug 30, 2015 at 09:30:28PM -0700, Guenter Roeck wrote:
> On 08/30/2015 09:16 PM, Huang Rui wrote:
> >On Fri, Aug 28, 2015 at 07:05:13AM -0700, Guenter Roeck wrote:
> >>On 08/28/2015 03:45 AM, Huang Rui wrote:
> >>>On Thu, Aug 27, 2015 at 10:30:43AM -0700, Guenter Roeck wrote:
> >>>>On Thu, Aug 27, 2015 at 04:07:43PM +0800, Huang Rui wrote:
> >>>>>This patch introduces an algorithm that computes the average power by
> >>>>>reading a delta value of “core power accumulator” register during
> >>>>>measurement interval, and then dividing delta value by the length of
> >>>>>the time interval.
> >>>>>
> >>>>>User is able to use power1_acc entry to measure the processor power
> >>>>>consumption and power1_acc just needs to be read twice with an needed
> >>>>>interval in-between.
> >>>>>
> >>>>>A simple example:
> >>>>>
> >>>>>$ cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc
> >>>>>$ sleep 10000s
> >>>>>$ cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc
> >>>>>
> >>>>>The result is current average processor power consumption in 10000
> >>>>>seconds. The unit of the result is uWatt.
> >>>>>
> >>>>>Signed-off-by: Huang Rui <ray.huang@amd.com>
> >>>>>---
> >>>>>  drivers/hwmon/fam15h_power.c | 62 ++++++++++++++++++++++++++++++++++++++++++++
> >>>>>  1 file changed, 62 insertions(+)
> >>>>>
> >>>>>diff --git a/drivers/hwmon/fam15h_power.c b/drivers/hwmon/fam15h_power.c
> >>>>>index d529e4b..3bab797 100644
> >>>>>--- a/drivers/hwmon/fam15h_power.c
> >>>>>+++ b/drivers/hwmon/fam15h_power.c
> >>>>>@@ -60,6 +60,7 @@ struct fam15h_power_data {
> >>>>>  	u64 cu_acc_power[MAX_CUS];
> >>>>>  	/* performance timestamp counter */
> >>>>>  	u64 cpu_sw_pwr_ptsc[MAX_CUS];
> >>>>>+	struct mutex acc_pwr_mutex;
> >>>>>  };
> >>>>>
> >>>>>  static ssize_t show_power(struct device *dev,
> >>>>>@@ -121,17 +122,74 @@ static DEVICE_ATTR(power1_crit, S_IRUGO, show_power_crit, NULL);
> >>>>>  static struct attribute_group fam15h_power_group;
> >>>>>  __ATTRIBUTE_GROUPS(fam15h_power);
> >>>>>
> >>>>>+static ssize_t show_power_acc(struct device *dev,
> >>>>>+			      struct device_attribute *attr, char *buf)
> >>>>>+{
> >>>>>+	int cpu, cu, cu_num, cores_per_cu;
> >>>>>+	u64 curr_cu_acc_power[MAX_CUS],
> >>>>>+	    curr_ptsc[MAX_CUS], jdelta[MAX_CUS];
> >>>>>+	u64 tdelta, avg_acc;
> >>>>>+	struct fam15h_power_data *data = dev_get_drvdata(dev);
> >>>>>+
> >>>>>+	cores_per_cu = amd_get_cores_per_cu();
> >>>>>+	cu_num = boot_cpu_data.x86_max_cores / cores_per_cu;
> >>>>>+
> >>>>>+	for (cpu = 0, avg_acc = 0; cpu < cu_num * cores_per_cu; cpu += cores_per_cu) {
> >>>>>+		cu = cpu / cores_per_cu;
> >>>>>+		if (rdmsrl_safe_on_cpu(cpu, MSR_F15H_PTSC, &curr_ptsc[cu])) {
> >>>>>+			pr_err("Failed to read PTSC counter MSR on core%d\n",
> >>>>>+			       cpu);
> >>>>>+			return 0;
> >>>>>+		}
> >>>>>+
> >>>>>+		if (rdmsrl_safe_on_cpu(cpu, MSR_F15H_CU_PWR_ACCUMULATOR,
> >>>>>+				       &curr_cu_acc_power[cu])) {
> >>>>>+			pr_err("Failed to read compute unit power accumulator MSR on core%d\n",
> >>>>>+			       cpu);
> >>>>>+			return 0;
> >>>>>+		}
> >>>>>+
> >>>>>+		if (curr_cu_acc_power[cu] < data->cu_acc_power[cu]) {
> >>>>>+			jdelta[cu] = data->max_cu_acc_power + curr_cu_acc_power[cu];
> >>>>>+			jdelta[cu] -= data->cu_acc_power[cu];
> >>>>>+		} else {
> >>>>>+			jdelta[cu] = curr_cu_acc_power[cu] - data->cu_acc_power[cu];
> >>>>>+		}
> >>>>>+		tdelta = curr_ptsc[cu] - data->cpu_sw_pwr_ptsc[cu];
> >>>>>+		jdelta[cu] *= data->cpu_pwr_sample_ratio * 1000;
> >>>>>+		do_div(jdelta[cu], tdelta);
> >>>>>+
> >>>>>+		mutex_lock(&data->acc_pwr_mutex);
> >>>>>+		data->cu_acc_power[cu] = curr_cu_acc_power[cu];
> >>>>>+		data->cpu_sw_pwr_ptsc[cu] = curr_ptsc[cu];
> >>>>>+		mutex_unlock(&data->acc_pwr_mutex);
> >>>>>+
> >>>>>+		/* the unit is microWatt */
> >>>>>+		avg_acc += jdelta[cu];
> >>>>>+	}
> >>>>>+
> >>>>>+	return sprintf(buf, "%u\n", (unsigned int) avg_acc);
> >>>>>+}
> >>>>>+static DEVICE_ATTR(power1_acc, S_IRUGO, show_power_acc, NULL);
> >>>>
> >>>>I am not really a friend of introducing a non-standard attribute.
> >>>>Does the energy attribute not work here ?
> >>>>
> >>>
> >>>You're right. Non-standard attribute might not be good. Could you
> >>>please give me some hints if I use "energy" instead?
> >>>
> >>1 Joule = 1 Watt-second.
> >>
> >>Something else, though - did you make sure that your code doesn't overflow ?
> >>Even though you calculate the average in an u64, you display it as unsigned.
> >>
> >
> >Thanks to your reminder. It should not be overflow. The maximum power
> >consumption of processor (AMD CZ and future 15h) is about 15 Watts =
> >15,000,000 uWatts = 0xE4E1C0 uWatts, the size is 24 < 32 < 64 bits.
> >
> >Actually, the unit of jdelta is not Joule. Because the tdelta is the
> >loops (cycles) that PTSC counter (the freqency is about 100 MHz)
> >counts not seconds.
> >
> >So avg_acc is the average power consumption not the accumulated energy.
> >
> 
> Would power1_average then be better suitable for the attribute ?
> There is also power1_average_interval which could be used to make
> the interval configurable.
> 

power1_average and power1_average_interval looks better except we
might use a conversion that make interval from milliseconds to PTSC
cycles instead of "cat-ting" the sysfs file twice, right? 

Thanks,
Rui
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/