Message-ID: <20170717200457.00000487@huawei.com>
Date:   Mon, 17 Jul 2017 20:04:57 +0800
From:   Jonathan Cameron <Jonathan.Cameron@...wei.com>
To:     Stefan Bruens <stefan.bruens@...h-aachen.de>
CC:     Jonathan Cameron <jic23@...nel.org>, <linux-iio@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>, Hartmut Knaack <knaack.h@....de>,
        "Lars-Peter Clausen" <lars@...afoo.de>,
        Peter Meerwald-Stadler <pmeerw@...erw.net>,
        "Marc Titinger" <mtitinger@...libre.com>, <pmeerw@...erw.net>,
        <manivannanece23@...il.com>
Subject: Re: [PATCH 2/2] iio: adc: Allow setting Shunt Voltage PGA gain and
 Bus Voltage range

On Mon, 17 Jul 2017 00:08:32 +0200
Stefan Bruens <stefan.bruens@...h-aachen.de> wrote:

> On Sonntag, 30. April 2017 18:19:39 CEST Jonathan Cameron wrote:
> > On 29/04/17 21:37, Stefan Bruens wrote:  
> > > On Mittwoch, 26. April 2017 08:59:47 CEST Jonathan Cameron wrote:  
> > >> On 26/04/17 07:19, Jonathan Cameron wrote:  
> > >>> On 17/04/17 23:08, Stefan Bruens wrote:  
> > >>>> On Freitag, 14. April 2017 17:12:03 CEST Jonathan Cameron wrote:  
> > > [...]
> > >   
> > >>>> 4. Any user of the gain settings had to be made aware of the
> > >>>> possibility
> > >>>> to
> > >>>> change it, no matter how it is exposed. Making it part of the scale,
> > >>>> and
> > >>>> thus changing the meaning of the raw values, would be breaking the
> > >>>> existing ABI.>  
> > >>> 
> > >>> The raw values should indeed not change.  That was a missunderstanding
> > >>> on
> > >>> my part.  Usually when a device has a PGA it is not compensated for in
> > >>> the output. So normally it's up to the driver to 'apply' the effective
> > >>> gain to the incoming reading.  When that isn't the case, it can be
> > >>> considered some sort of internal trim - hence the use of calibscale for
> > >>> this case.  
> > >> 
> > >> Mulling this over, calibscale might not work either in this case. The
> > >> datasheet helpfully sometimes uses ranges and sometimes uses scale
> > >> factors.
> > >> There is also obviously the calibration register kicking around which
> > >> would
> > >> also be handled with calibscale if exposed to userspace (currently it
> > >> isn't)
> > >> 
> > >> I'm out of time tonight so will think it bit more about this and get back
> > >> to you in the next few days...  
> > > 
> > > hardwaregain may be a viable option. For the shunt voltage, available
> > > values would be [0.125, 0.25, 0.5, 1.0], for the bus range we would have
> > > either [0.5, 1.0] or [1.0, 2.0] for bus ranges [32V, 16V].
> > > 
> > > Does hardwaregain have the right semantics for shunt voltage gain and/or
> > > bus range?  
> > 
> > Description we currently have in
> > Documentation/ABI/testing/sysfs-bus-iio is:
> > 		Hardware applied gain factor. If shared across all channels,
> > 		<type>_hardwaregain is used.
> > 
> > Just thinking about the use cases, it is mostly used for cases where the
> > gain is not of the measurement being acquired, but rather of something
> > related (like the gain on time of flight sensors or pulse counters).
> > 
> > It also gets used for output devices and amplifiers though so kind of
> > similar as in those cases we felt calibrationscale was a bit of a stretch!
> > 
> > So yes, I can see that working.  Whether it is a better choice than
> > simply allowing the range attributes (documented for this narrow
> > case to say they should only be used when the range is independent of
> > the scale) is an open question.  Given we have always preferred scales
> > to ranges if you think you can make hardwaregain fit well then lets
> > go with that, perhaps updating the docs to make this usecase explicit.
> > 
> > Looking back at the original emails we were actually thinking of
> > transistioning calibscale to hardwaregain in general as it covered
> > describing both uses, but it never happened...  
> 
> Hi John,
> 
> as all other patches for INA2xx went into or on their way into mainline, its 
> time to revisit the INA219/220 bus range and shunt voltage gain again.
> 
> TLDR: Using HARDWAREGAIN fits existing uses/semantics.
> 
> I had a look at current users of IIO_CHAN_INFO_HARDWAREGAIN:
> 
> amplifiers/ad8366.c: Variable gain amplifier without ADC or DAC, so no SCALE 
> attribute
> 
> light/vl6180.c: ToF sensor with ambient light sensor. The ALS sensor has two 
> settings affecting the RAW sensor readout, HARDWAREGAIN and INTegration_TIME. 
> Baseline settings are gain=1 and integration time=0.1(seconds), with a 
> corresponding raw reading of 1 ^= 0.32 lux.
> The SCALE value is correct for the baseline setting, but although modifying 
> HARDWAREGAIN and/or INT_TIME affects the RAW readout, this is not reflected in 
> the SCALE attribute, i.e. to get the correct physical value, one has to use:
> Light[lux] = raw_value * SCALE * (0.1s/INT_TIME) / HARDWAREGAIN
This isn't right.  User space should be able to just apply the single scale
value to get the correct real world value, not this complex interaction.

So I'd count this driver as buggy unfortunately. 

> 
> light/adjd_s311.c: HARDWAREGAIN affects the RAW readout, but as there is no 
> given fixed relationship between RAW values and irradiance, there is no SCALE 
> attribute.

> 
> adc/stx104.c: The ADC has a software controllable HARDWAREGAIN and a hardware 
> controlled (jumper) offset and single ended/differential setting with software 
> readback. HARDWAREGAIN and offset/differential are reflected in the SCALE and 
> OFFSET attributes, i.e. the physical value can be determined by:
> U[V] = (raw_value * SCALE) + OFFSET
> 
> So we have two users of HARDWAREGAIN with contradicting behaviour regarding 
> SCALE. IMHO, the stx104 behaviour is the correct one.
I go the other way, simply because we don't want to complicate the userspace
interface if we don't have to.
> 
> For the INA2xx, neither INT_TIME nor AVERAGE affect the RAW <-> physical value 
> relationship, i.e. the SCALE is fixed. The same is true for the INA219/220 bus 
> range/shunt voltage gain. So using HARDWAREGAIN for both shunt voltage gain 
> and bus voltage range does match existing semantics.

I'm uncomfortable with applying a second scaling within all user space code.
That should be handled in the kernel rather than pushing on the burden.
It's not a fast path so doesn't matter if we have to some nasty maths to
work out the right value.

Jonathan
> 
> Kind regards,
> 
> Stefan
> 

Powered by blists - more mailing lists