On Sun, 26 Jan 2020 20:27:22 +0000
Jonathan Bakker <xc-racer2@xxxxxxx> wrote:
> Hi Linus,
>
> On 2020-01-26 7:16 a.m., Linus Walleij wrote:
> > On Fri, Jan 24, 2020 at 1:47 AM Jonathan Bakker <xc-racer2@xxxxxxx> wrote:
> >
> >> Thanks for the driver, I've tested it on a first-gen Galaxy S
> >> device with a GP2AP002A00F. I have a few comments that I've put inline
> >> based on my experiences.
> >
> > Thanks a lot!
> >
> >> Both shortly after probe (when runtime pm timeout occurs?) and after
> >> manually disabling the proximity event, this
> >> irq handler is called. Since the chip is in low power state, it obviously
> >> fails to read the proximity value and write to the OCON register below, eg
> >>
> >> [ 7.215875] gp2ap002 11-0044: error reading proximity
> >> [ 8.105878] gp2ap002 11-0044: error setting up VOUT control 1
> >>
> >> Can we do something like disable_irq() in the runtime pm function to prevent
> >> this?
> >
> > I added that in v6, I hope this solves your problem.
>
> Yep, it appears to. Thanks.
>
> >
> >> The gp2ap002s00f_regmap_i2c_read function works on the gp2ap002a00f as well,
> >> so this can be simplified/dropped.
> >
> > Fixed this too in v6.
> >
> >>> + if (ch_type != IIO_CURRENT) {
> >>> + dev_err(dev,
> >>> + "wrong type of IIO channel specified for ALSOUT\n");
> >>> + return -EINVAL;
> >>> + }
> >>
> >> This enforces a current ADC, while several devices besides mine (eg Galaxy Nexus)
> >> use a resistor and a voltage ADC. For this case, could we add a device property such as
> >> sharp,adc-adjustment-ratio to convert from the raw ADC values to a "current" that
> >> could be used in the lookup table? So the "current" would be the raw ADC divided
> >> by that special value.
> >>
> >> Instructions for converting the ADC back to the current can be found eg at
> >> https://android.googlesource.com/device/samsung/crespo/+/2e0ab7265e3039fee787c2216e0c98d92ea0b49e%5E%21/#F0
> >
> > I'd like to keep that as a future enhancement unless someone is very eager
> > to get it and has a device they can test it on. Otherwise it will be
> > just dry-coding
> > on my part.
>
> Well, I've got such a device and can test :)
>
> >
> > I think the property we would add in the device tree in that case should
> > be the resistance value. This is based on the following assumption
> > which is indeed a bit of speculation since there is no proper datasheet
> > for the light sensor part of the component:
> >
> > - The light sensor part is simply a photodiode
> > - This emits a nonlinear current in relation to how many
> > photons hit the photodiode in a time interval, the relationship
> > is described in the curren->lux table we have
> > - Some vendors do not have any current ADC, so they connect
> > this to a resistor, and measure the voltage over the
> > resistor because the have a voltage ADC
> >
> > Since current is linear to the voltage over the resistor, we should
> > include the resistance in the device tree, then using that the
> > corresponding current can be calculated and we use the same
> > look-up table to find the lux. Probably each system may need
> > to also subtract some bias voltage or so.
>
> Yes, this is my understanding of it as well (I also have no datasheet).
>
> Given V = actual voltage in V, Vref = reference voltage of ADC in V, ADC = value
> read from ADC, ADCmax = maximum possible value read from ADC, I = current in amps,
> R = resistor value in ohms, uA = current in microamps
>
> V / Vref = ADC / ADCmax
> V = (ADC / ADCmax) * Vref
>
> V = I * R
> I * R = (ADC / ADCmax) * Vref
> I = ADC * Vref / ADCmax / R
>
> However, because we want the current in uA for the table, (note, your comment says
> that the table is based on mA, but I believe that it should actually be uA)
>
> uA = ADC * Vref / ADCmax / R * 1000000
>
> Then, in order to avoid multiplying by a decimal, the uA is the ADC value divided
> by the inverse of
>
> (1000000 * Vref / ADCmax / R)
>
> For example, on the first gen Galaxy S series and the Nexus S, the Vref is 3.3V,
> the ADC is 12 bit (2^12 = 4096, so high value is 4095), the resistor is 47000ohms,
>
> inverse of (1000000 * 3.3 / 4095 / 47000) = 58
>
> so we need to divide the raw ADC reading by 58 in order to get the uA for the current->lux
> table.
>
> A quick patch that I used for testing (based off of v5) is
>
> diff --git a/drivers/iio/light/gp2ap002.c b/drivers/iio/light/gp2ap002.c
> index a5897959f70d..b98aec337f8b 100644
> --- a/drivers/iio/light/gp2ap002.c
> +++ b/drivers/iio/light/gp2ap002.c
> @@ -130,6 +130,7 @@
> * @vdd: regulator controlling VDD
> * @vio: regulator controlling VIO
> * @alsout: IIO ADC channel to convert the ALSOUT signal
> + * @adc_adj: conversion factor if voltage ADC used instead of current ADC
> * @is_gp2ap002s00f: this is the GP2AP002F variant of the chip
> * @enabled: we cannot read the status of the hardware so we need to
> * keep track of whether the event is enabled using this state variable
> @@ -143,6 +144,7 @@ struct gp2ap002 {
> enum iio_event_direction dir;
> u8 hys_far;
> u8 hys_close;
> + u8 adc_adj;
> bool is_gp2ap002s00f;
> bool enabled;
> };
> @@ -272,6 +274,9 @@ static int gp2ap002_get_lux(struct gp2ap002 *gp2ap002)
> if (ret < 0)
> return ret;
>
> + if (gp2ap002->adc_adj)
> + res /= gp2ap002->adc_adj;
> +
> dev_dbg(gp2ap002->dev, "read %d mA from ADC\n", res);
>
> ill1 = &gp2ap002_illuminance_table[0];
> @@ -588,7 +593,16 @@ static int gp2ap002_probe(struct i2c_client *client,
> ret = iio_get_channel_type(gp2ap002->alsout, &ch_type);
> if (ret < 0)
> return ret;
> - if (ch_type != IIO_CURRENT) {
> + if (ch_type == IIO_VOLTAGE) {
> + ret = device_property_read_u8(dev,
> + "sharp,adc-adjustment-ratio", &val);
> + if (ret) {
> + dev_err(dev,
> + "failed to obtain adc conversion\n");
> + return -EINVAL;
> + }
> + gp2ap002->adc_adj = val;
> + } else if (ch_type != IIO_CURRENT) {
> dev_err(dev,
> "wrong type of IIO channel specified for ALSOUT\n");
> return -EINVAL;
>
> Alternatively, you could collect the resistor value, the ADC precision (this doesn't
> appear to be queryable via the IIO layer), and the reference voltage level - but I'm
> not sure how you'd do the inverse calculation in the kernel.
An alternative to doing this is to represent the resistor circuit explicitly.
You end up with a really small ADC driver that is a consumer of a voltage
and provides a current channel. That has all the properties of the
circuit via DT.
We do some stuff a bit similar to this in the envelope detector driver.
In general I would prefer we handle this sort of conversion generically
rather than bolting it into a light sensor driver like you are doing here,
even if it comes at the cost of a bit more complexity.
Jonathan
>
> >
> > But we definately need something to test on to do this right>
> > Yours,
> > Linus Walleij
> >
>
> Thanks,
> Jonathan Bakker
