Re: [PATCH 2/2] hwmon: scmi: Scale values to target desired HWMON units

[Florian Fainelli]

On 5/7/19 6:55 AM, Guenter Roeck wrote:
> Hi Florian,
> 
> On 5/6/19 3:41 PM, Florian Fainelli wrote:
>> If the SCMI firmware implementation is reporting values in a scale that
>> is different from the HWMON units, we need to scale up or down the value
>> according to how far appart they are.
>>
>> Signed-off-by: Florian Fainelli <f.fainelli@xxxxxxxxx>
>> ---
>> Â drivers/hwmon/scmi-hwmon.c | 55 +++++++++++++++++++++++++++++++-------
>> Â 1 file changed, 46 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/hwmon/scmi-hwmon.c b/drivers/hwmon/scmi-hwmon.c
>> index a80183a488c5..e9913509cb88 100644
>> --- a/drivers/hwmon/scmi-hwmon.c
>> +++ b/drivers/hwmon/scmi-hwmon.c
>> @@ -18,6 +18,51 @@ struct scmi_sensors {
>> ÂÂÂÂÂ const struct scmi_sensor_info **info[hwmon_max];
>> Â };
>> Â +static enum hwmon_sensor_types scmi_types[] = {
>> +ÂÂÂ [TEMPERATURE_C] = hwmon_temp,
>> +ÂÂÂ [VOLTAGE] = hwmon_in,
>> +ÂÂÂ [CURRENT] = hwmon_curr,
>> +ÂÂÂ [POWER] = hwmon_power,
>> +ÂÂÂ [ENERGY] = hwmon_energy,
>> +};
>> +
>> +static u64 scmi_hwmon_scale(const struct scmi_sensor_info *sensor,
>> u64 value)
>> +{
>> +ÂÂÂ u64 scaled_value = value;
> 
> I don't think that variable is necessary.
> 
>> +ÂÂÂ s8 desired_scale;
> 
> Just scale ? Also, you could assign scale here directly, and subtract
> the offset below. Then "n" would not be necessary.
> Such as
> ÂÂÂÂs8 scale = sensor->scale;ÂÂÂ // assuming scale is s8
> ÂÂÂÂ...
> ÂÂÂÂcase CURRENT:
> ÂÂÂÂÂÂÂ scale += 3;
> ÂÂÂÂ...
> 
> That would also be less confusing, since it would avoid the double
> negation.
> 
>> +ÂÂÂ int n, p;
> 
>> +
>> +ÂÂÂ switch (sensor->type) {
>> +ÂÂÂ case TEMPERATURE_C:
>> +ÂÂÂ case VOLTAGE:
>> +ÂÂÂ case CURRENT:
>> +ÂÂÂÂÂÂÂ /* fall through */
> Unnecessary comment

Is not removing the comment going to upset gcc when using
-Wimplicit-fallthrough?

> 
>> +ÂÂÂÂÂÂÂ desired_scale = -3;
>> +ÂÂÂÂÂÂÂ break;
>> +ÂÂÂ case POWER:
>> +ÂÂÂ case ENERGY:
>> +ÂÂÂÂÂÂÂ /* fall through */
> Unnecessary comment.
> 
>> +ÂÂÂÂÂÂÂ desired_scale = -6;
>> +ÂÂÂÂÂÂÂ break;
>> +ÂÂÂ default:
>> +ÂÂÂÂÂÂÂ return scaled_value;
> 
> Here we presumably want a scale of 0. However, if the scale passed
> from SCMI is, say, -5 or +5, we return the original (unadjusted)
> value. Seems to me we would still want to adjust the value to match
> hwmon expectations. Am I missing something ?

You raise a valid point, not that could happen today because if the
sensor type has a value we don't recognize, we have not registered it,
so we would not even try to read rom it, but let's be future proof.

> 
>> +ÂÂÂ }
>> +
>> +ÂÂÂ n = (s8)sensor->scale - desired_scale;
>> +ÂÂÂÂÂÂÂ if (n == 0)
> 
> Indentation seems off here.
> 
>> +ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ return scaled_value;
>> +
>> +ÂÂÂ for (p = 0; p < abs(n); p++) {
>> +ÂÂÂÂÂÂÂ /* Need to scale up from sensor to HWMON */
>> +ÂÂÂÂÂÂÂ if (n > 0)
>> +ÂÂÂÂÂÂÂÂÂÂÂ scaled_value *= 10;
>> +ÂÂÂÂÂÂÂ else
>> +ÂÂÂÂÂÂÂÂÂÂÂ do_div(scaled_value, 10);
>> +ÂÂÂ }
> 
> Something like
> 
> ÂÂÂÂfactor = pow10(abs(scale));
> ÂÂÂÂif (scale > 0)
> ÂÂÂÂÂÂÂ value *= factor;
> ÂÂÂÂelse
> ÂÂÂÂÂÂÂ do_div(value, factor);
> 
> would avoid the repeated abs() and do_div(). Unfortunately there is
> no pow10() helper, so you would have to write that. Still, I think
> that would be much more efficient.

Sounds reasonable. Thanks for your feedback!
-- 
Florian