Date:	Fri, 15 Nov 2013 09:07:36 +0100
From:	Jean Delvare <khali@...ux-fr.org>
To:	Eduardo Valentin <eduardo.valentin@...com>
Cc:	<swarren@...dotorg.org>, <pawel.moll@....com>,
	<mark.rutland@....com>, <ian.campbell@...rix.com>,
	<rob.herring@...xeda.com>, <linux@...ck-us.net>,
	<rui.zhang@...el.com>, devicetree@...r.kernel.org, wni@...dia.com,
	linux-pm@...r.kernel.org, linux-kernel@...r.kernel.org,
	lm-sensors@...sensors.org, grant.likely@...aro.org
Subject: Re: [lm-sensors] [PATCHv9 02/20] thermal: introduce device tree
 parser

Hi Eduardo,

On Tue, 12 Nov 2013 15:46:04 -0400, Eduardo Valentin wrote:
> This patch introduces a device tree bindings for
> describing the hardware thermal behavior and limits.
> Also a parser to read and interpret the data and feed
> it in the thermal framework is presented.
> 
> This patch introduces a thermal data parser for device
> tree. The parsed data is used to build thermal zones
> and thermal binding parameters. The output data
> can then be used to deploy thermal policies.
> 
> This patch adds also documentation regarding this
> API and how to define tree nodes to use
> this infrastructure.
> 
> Note that, in order to be able to have control
> on the sensor registration on the DT thermal zone,
> it was required to allow changing the thermal zone
> .get_temp callback. For this reason, this patch
> also removes the 'const' modifier from the .ops
> field of thermal zone devices.
> 
> Cc: Zhang Rui <rui.zhang@...el.com>
> Cc: linux-pm@...r.kernel.org
> Cc: linux-kernel@...r.kernel.org
> Cc: Mark Rutland <mark.rutland@....com>
> Signed-off-by: Eduardo Valentin <eduardo.valentin@...com>
> ---
> (...)
> +static int of_thermal_get_trend(struct thermal_zone_device *tz, int trip,
> +				enum thermal_trend *trend)
> +{
> +	struct __thermal_zone *data = tz->devdata;
> +	long dev_trend;
> +	int r;
> +
> +	if (!data->get_trend)
> +		return -EINVAL;
> +
> +	r = data->get_trend(data->sensor_data, &dev_trend);
> +	if (r)
> +		return r;
> +
> +	/* TODO: These intervals might have some thresholds, but in core code */
> +	if (dev_trend > 0)
> +		*trend = THERMAL_TREND_RAISING;
> +	else if (dev_trend < 0)
> +		*trend = THERMAL_TREND_DROPPING;
> +	else
> +		*trend = THERMAL_TREND_STABLE;
> +
> +	return 0;
> +}

I don't like the whole trend thing.

I've been writing hwmon drivers for the past decade and I've never seen
a thermal sensor device being able to report trends. And as a rule of
thumb, if the hardware can't do it then the (hardware-specific) drivers
should not report it.

Hwmon devices (and drivers) report temperature values, and sometimes
historical min/max. They can do that because these are facts that need
no interpretation.

Defining a trend, however, requires care and, more importantly,
decisions. For example, consider a thermal sensor which reports 50°C at
time t, then 47°C at time t+3s, then 48°C at time t+6s. At t+7s someone
asks for the trend, what should the driver reply? If you consider only
the last 5 seconds, temperature is raising. If you consider the last 10
seconds instead, temperature is dropping. How would the driver know
what time frame the caller is interested in?

Another example: "small" temperature variations. If temperatures varies
by 1°C in my kitchen's oven, I'd call it stable. If my body temperature
varies by 1°C, I'd call it non-stable, and my doctor for an appointment
also ;-)

My point is that only the caller, and not the driver, knows how to
convert a series of measurements into a trend. So I don't think drivers
should implement anything like get_trend(). Whatever piece of code
needs to establish a trend should call get_temp() repeatedly, store the
results and do its own analysis of the data.

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

Powered by blists - more mailing lists