[PATCH v2 1/2] thermal: mediatek: Add cpu power cooling model.

Punit Agrawal punit.agrawal at arm.com
Mon Oct 12 10:26:51 PDT 2015


Mark Rutland <mark.rutland at arm.com> writes:

> On Wed, Oct 07, 2015 at 08:22:40PM +0800, Dawei Chien wrote:
>> From: "Dawei.Chien" <dawei.chien at mediatek.com>
>> 
>> This power model is base on Intelligent Power Allocation (IPA) technical,
>> requires that the operating-points of the CPUs are registered using the
>> kernel's opp library and the `cpufreq_frequency_table` is assigned to the
>> `struct device` of the cpu MT8173.
>> 
>> Signed-off-by: Dawei.Chien <dawei.chien at mediatek.com>
>> ---
>> This patch is base on
>> https://patchwork.kernel.org/patch/7034601/
>> ---
>>  drivers/cpufreq/mt8173-cpufreq.c |   97 +++++++++++++++++++++++++++++++++-----
>>  1 file changed, 86 insertions(+), 11 deletions(-)
>> 
>> diff --git a/drivers/cpufreq/mt8173-cpufreq.c b/drivers/cpufreq/mt8173-cpufreq.c
>> index 49caed2..9233ec5 100644
>> --- a/drivers/cpufreq/mt8173-cpufreq.c
>> +++ b/drivers/cpufreq/mt8173-cpufreq.c
>> @@ -28,7 +28,8 @@
>>  #define MAX_VOLT_SHIFT		(200000)
>>  #define MAX_VOLT_LIMIT		(1150000)
>>  #define VOLT_TOL		(10000)
>> -
>> +#define CAPACITANCE_CA53	(263)
>> +#define CAPACITANCE_CA57	(530)
>>  /*
>>   * The struct mtk_cpu_dvfs_info holds necessary information for doing CPU DVFS
>>   * on each CPU power/clock domain of Mediatek SoCs. Each CPU cluster in
>> @@ -51,6 +52,72 @@ struct mtk_cpu_dvfs_info {
>>  	bool need_voltage_tracking;
>>  };
>>  
>> +struct mtk_cpu_static_power {
>> +	unsigned long voltage;
>> +	unsigned int power;
>> +};
>> +
>> +/* measured by WA program. */
>> +static const struct mtk_cpu_static_power mtk_ca53_static_power[] = {
>> +	{859000, 43},
>> +	{908000, 52},
>> +	{983000, 86},
>> +	{1009000, 123},
>> +	{1028000, 138},
>> +	{1083000, 172},
>> +	{1109000, 180},
>> +	{1125000, 192},
>> +};
>> +
>> +/* measured by WA program. */
>> +static const struct mtk_cpu_static_power mtk_ca57_static_power[] = {
>> +	{828000, 72},
>> +	{867000, 90},
>> +	{927000, 156},
>> +	{968000, 181},
>> +	{1007000, 298},
>> +	{1049000, 435},
>> +	{1089000, 533},
>> +	{1125000, 533},
>> +};
>
> This should be described in the DT, rather than necessitating tonnes of
> these tables in the kernel.

The above table is a simplification that is tying the static power of a
device to a voltage (indirectly via frequency for ease of indexing
perhaps).

We should definitely look at describing the static power model in the
device tree - but it is not entirely clear what is the best model to
use as basis for the device tree bindings.

Static power consumption depends on a few different parameters (silicon
process, voltage, temperature, etc.). The relationship between them
maybe non-linear and not all of these maybe significant for a given
platform. e.g., for Juno, we are using a regression based on voltage and
temperature which was deemed close enough (very subjective tbh).

On the other hand, dynamic power consumption is somewhat simpler and I
have patches[0][1][2] enabling it's description in device tree. Your comments
there would be very much appreciated.

As for this patch, I hope it can be moved to using the device tree when
the bindings for static power are agreed upon. Although sub-optimal, I
can't see any other way forward for now.

[0] http://thread.gmane.org/gmane.linux.kernel/2011466/focus=130373
[1] http://thread.gmane.org/gmane.linux.kernel/2011466/focus=2011465
[2] http://article.gmane.org/gmane.linux.drivers.sensors/37859

>
> Mark.
> --
> To unsubscribe from this list: send the line "unsubscribe linux-pm" in
> the body of a message to majordomo at vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html



More information about the linux-arm-kernel mailing list