On Wed, 11 May 2011 11:40:56 -0700, Guenter Roeck wrote: > On Wed, 2011-05-11 at 12:57 -0400, Jean Delvare wrote: > > What if we create attribute curr[1-*]_resistor, which would hold the > > value of the resistor used? Unit to be defined (milli-Ohm?) It would > > either be initialized to 1 milli-Ohm as you've been using so far, or > > left uninitialized (0) and the current value would only be returned by > > the driver after user-space sets the value (through a set statement in > > the configuration file.) Either way, this would let the user specify > > the resistor value, as we do with the ones used for voltage input > > scaling. > > Just as good or bad as defining a calibration factor, in my opinion. Calibration factor? You mean, the compute statement is 0.001 Ohm isn't OK? > Both convert uV readings to mA. If we go along that route, we might as > well do the same for voltages etc. Err, but we don't _need_ it for voltages, so why would we make things needlessly complex? > Milli-Ohm would be insufficient, since there are smaller resistors > available. Mouser offers a sense resistor with 0.000125 Ohms, so you > would have to use micro-Ohm. Sure, fine with me. This will still allow values up to 4.2 kOhm on 32-bit systems, I presume this is sufficient. > > This would solve your concern (drivers can keep exporting curr[1-*] > > attributes for current sensors) and mind (components external to the > > monitoring device are specified by user-space). > > Not really ... current sense resistors are similar to voltage divider > resistors, and external to the chip. Assumption for voltage divider > resistors is that the driver doesn't care and conversion is handled in > userland. Unfortunately, that does not work for currents, since the > driver has to make _some_ assumption how to convert uV as reported by > the ADC to mA. Unless we go for suboption b, i.e. the resistor value starts at 0 and the driver doesn't report any current value until user-space has set the value. Then the driver doesn't have to assume anything. I agree that voltage and current cases aren't fundamentally different, except for the fact that unscaled voltage values do exist, while every current sensor will need a resistor. What I would like is that we come up with a standard and documented way to implement and use current sensors. It can be my proposal above, or it can be your arbitrary 0.001 Ohm resistor. But I don't want to leave the implementation up to every driver author. This is how lm-sensors 2 worked, and we definitely do NOT want anything like this ever again. -- Jean Delvare _______________________________________________ lm-sensors mailing list lm-sensors@xxxxxxxxxxxxxx http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
