Message-ID: <20180622152554.00006c3d@huawei.com>
Date:   Fri, 22 Jun 2018 15:26:24 +0100
From:   Jonathan Cameron <jonathan.cameron@...wei.com>
To:     Baolin Wang <baolin.wang@...aro.org>
CC:     Jonathan Cameron <jic23@...23.retrosnub.co.uk>,
        Jonathan Cameron <jic23@...nel.org>,
        Rob Herring <robh+dt@...nel.org>,
        Mark Rutland <mark.rutland@....com>,
        Hartmut Knaack <knaack.h@....de>,
        "Lars-Peter Clausen" <lars@...afoo.de>,
        Peter Meerwald-Stadler <pmeerw@...erw.net>,
        <freeman.liu@...eadtrum.com>, Mark Brown <broonie@...nel.org>,
        DTML <devicetree@...r.kernel.org>, <linux-iio@...r.kernel.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v2 2/2] iio: adc: Add Spreadtrum SC27XX PMICs ADC
 support

On Mon, 18 Jun 2018 18:47:18 +0800
Baolin Wang <baolin.wang@...aro.org> wrote:

> Hi Jonathan,
> 
> On 18 June 2018 at 18:20, Jonathan Cameron <jic23@...23.retrosnub.co.uk> wrote:
> >
> >
> > On 17 June 2018 09:03:04 BST, Baolin Wang <baolin.wang@...aro.org> wrote:  
> >>Hi Jonathan,
> >>
> >>On 17 June 2018 at 02:35, Jonathan Cameron <jic23@...nel.org> wrote:  
> >>> On Fri, 15 Jun 2018 15:03:36 +0800
> >>> Baolin Wang <baolin.wang@...aro.org> wrote:
> >>>  
> >>>> From: Freeman Liu <freeman.liu@...eadtrum.com>
> >>>>
> >>>> The Spreadtrum SC27XX PMICs ADC controller contains 32 channels,
> >>>> which is used to sample voltages with 12 bits conversion.
> >>>>
> >>>> Signed-off-by: Freeman Liu <freeman.liu@...eadtrum.com>
> >>>> Signed-off-by: Baolin Wang <baolin.wang@...aro.org>  
> >>>
> >>> Hi,
> >>>
> >>> There are some race conditions around the probe and remove.
> >>> More care is needed when we have a mixture of managed and unmanaged  
> >>cleanup  
> >>> like here.  
> >>
> >>Thanks to point the race issue.
> >>  
> >>>
> >>> I'm not understanding the way you have exposed a simple _raw and  
> >>_scale  
> >>> attributes with what looks to be different scaling to that applied
> >>> in _processed.   As I say below, we should not have both of those  
> >>interface  
> >>> options anyway.  The ABI is that (X_raw + X_offset)*X_scale =  
> >>X_processed.  
> >>> (with defaults of X_scale = 1 and X_offset = 0).  
> >>
> >>See below comments.
> >>  
> >>>
> >>> Please rename to avoid using wild cards in the name.  That's gone
> >>> wrong so many times in the past you wouldn't believe it!
> >>> Hmm Awkward though if the MFD is already upstream. Ah well, I guess
> >>> for consistency we should follow that and groan when it goes wrong.  
> >>
> >>Can I rename to be 'sprd-pmic-adc.c'? I can not rename it as
> >>'sc2731-adc', we have differnet PMICs (SC2730, SC2731, SC2720 etc.),
> >>but they are all integrated the same ADC controller.  
> >
> > You can rename as this is a common problem throughout drivers. There is no good solution.
> >
> > Given mfd naming, just leave it with wild cards as better than a name no one will recognise  
> 
> OK.
> 
> >>  
> >>>> ---
> >>>> Changes since v1:
> >>>>  - Add const for static structures definition.
> >>>>  - Change SC27XX_ADC_TO_VOLTAGE macro to be one function.
> >>>>  - Move channel scale accessing into mutex protection.
> >>>>  - Fix some typos.
> >>>> ---
> >>>>  drivers/iio/adc/Kconfig      |   10 +
> >>>>  drivers/iio/adc/Makefile     |    1 +
> >>>>  drivers/iio/adc/sc27xx_adc.c |  547  
> >>++++++++++++++++++++++++++++++++++++++++++  
> >>>>  3 files changed, 558 insertions(+)
> >>>>  create mode 100644 drivers/iio/adc/sc27xx_adc.c
> >>>>
> >>>> diff --git a/drivers/iio/adc/Kconfig b/drivers/iio/adc/Kconfig
> >>>> index 9da7907..985b73e 100644
> >>>> --- a/drivers/iio/adc/Kconfig
> >>>> +++ b/drivers/iio/adc/Kconfig
> >>>> @@ -621,6 +621,16 @@ config ROCKCHIP_SARADC
> >>>>         To compile this driver as a module, choose M here: the
> >>>>         module will be called rockchip_saradc.
> >>>>
> >>>> +config SC27XX_ADC
> >>>> +     tristate "Spreadtrum SC27xx series PMICs ADC"
> >>>> +     depends on MFD_SC27XX_PMIC || COMPILE_TEST
> >>>> +     help
> >>>> +       Say yes here to build support for the integrated ADC inside  
> >>the  
> >>>> +       Spreadtrum SC27xx series PMICs.
> >>>> +
> >>>> +       This driver can also be built as a module. If so, the module
> >>>> +       will be called sc27xx_adc.
> >>>> +
> >>>>  config SPEAR_ADC
> >>>>       tristate "ST SPEAr ADC"
> >>>>       depends on PLAT_SPEAR || COMPILE_TEST
> >>>> diff --git a/drivers/iio/adc/Makefile b/drivers/iio/adc/Makefile
> >>>> index 28a9423..03db7b5 100644
> >>>> --- a/drivers/iio/adc/Makefile
> >>>> +++ b/drivers/iio/adc/Makefile
> >>>> @@ -59,6 +59,7 @@ obj-$(CONFIG_QCOM_SPMI_VADC) += qcom-spmi-vadc.o
> >>>>  obj-$(CONFIG_QCOM_PM8XXX_XOADC) += qcom-pm8xxx-xoadc.o
> >>>>  obj-$(CONFIG_RCAR_GYRO_ADC) += rcar-gyroadc.o
> >>>>  obj-$(CONFIG_ROCKCHIP_SARADC) += rockchip_saradc.o
> >>>> +obj-$(CONFIG_SC27XX_ADC) += sc27xx_adc.o
> >>>>  obj-$(CONFIG_SPEAR_ADC) += spear_adc.o
> >>>>  obj-$(CONFIG_STX104) += stx104.o
> >>>>  obj-$(CONFIG_SUN4I_GPADC) += sun4i-gpadc-iio.o
> >>>> diff --git a/drivers/iio/adc/sc27xx_adc.c  
> >>b/drivers/iio/adc/sc27xx_adc.c  
> >>>> new file mode 100644
> >>>> index 0000000..52e5b74
> >>>> --- /dev/null
> >>>> +++ b/drivers/iio/adc/sc27xx_adc.c  
> >>>
> >>> In general (i.e. when we notice in time) we don't allow wild cards in  
> >>names.  
> >>> Far too many times we did this in the past and ended up with later  
> >>parts  
> >>> that fitted the name, but could not be supported by the driver.
> >>>
> >>> The convention is to name everything after the first part supported.
> >>> So here, sc2731. (I relaxed my thoughts on this later having seen the  
> >>mfd  
> >>> has this naming - so there are no ideal options left..)  
> >>
> >>Like I explained above, maybe change to 'sprd_pmic_adc.c', is this OK
> >>for you?  
> > Goes wrong just as quickly as wild cards...  
> 
> OK.
> 
> >>>> +
> >>>> +static int sc27xx_adc_convert_volt(struct sc27xx_adc_data *data,  
> >>int channel,  
> >>>> +                                int scale, int raw_adc)
> >>>> +{
> >>>> +     u32 numerator, denominator;
> >>>> +     u32 volt;
> >>>> +
> >>>> +     /*
> >>>> +      * Convert ADC values to voltage values according to the  
> >>linear graph,  
> >>>> +      * and channel 5 and channel 1 has been calibrated, so we can  
> >>just  
> >>>> +      * return the voltage values calculated by the linear graph.  
> >>But other  
> >>>> +      * channels need be calculated to the real voltage values with  
> >>the  
> >>>> +      * voltage ratio.
> >>>> +      */
> >>>> +     switch (channel) {
> >>>> +     case 5:
> >>>> +             return sc27xx_adc_to_volt(&big_scale_graph, raw_adc);
> >>>> +
> >>>> +     case 1:
> >>>> +             return sc27xx_adc_to_volt(&small_scale_graph,  
> >>raw_adc);  
> >>>> +
> >>>> +     default:
> >>>> +             volt = sc27xx_adc_to_volt(&small_scale_graph,  
> >>raw_adc);  
> >>>> +             break;
> >>>> +     }  
> >>>
> >>> This looks a lot more complex than simple scaling that is indicated  
> >>by the  
> >>> raw and scale attributes? They can't both be right..  
> >>
> >>Since this is special for our ADC controller, we have 2 channels that
> >>has been calibrated in hardware, but for other
> >>none-calibrated-channels, we should care about the channel voltage
> >>ratio when converting to a real voltage values, that is because some
> >>channel's voltage is larger so we need one voltage ratio to sample the
> >>ADC values.  
> >
> > It's still a question of one or the other. Channels should not do processed and raw without a very good reason.  
> 
> I think I have explained why we need our special processed approach as below.
> 
> >
> > Issue with processed is that you can't easily do buffered chrdev streaming in future.
> >
> >  
> >>  
> >>>> +
> >>>> +     sc27xx_adc_volt_ratio(data, channel, scale, &numerator,  
> >>&denominator);  
> >>>> +
> >>>> +     return (volt * denominator + numerator / 2) / numerator;
> >>>> +}
> >>>> +
> >>>> +static int sc27xx_adc_read_processed(struct sc27xx_adc_data *data,
> >>>> +                                  int channel, int scale, int *val)
> >>>> +{
> >>>> +     int ret, raw_adc;
> >>>> +
> >>>> +     ret = sc27xx_adc_read(data, channel, scale, &raw_adc);
> >>>> +     if (ret)
> >>>> +             return ret;
> >>>> +
> >>>> +     *val = sc27xx_adc_convert_volt(data, channel, scale, raw_adc);
> >>>> +     return 0;
> >>>> +}
> >>>> +
> >>>> +static int sc27xx_adc_read_raw(struct iio_dev *indio_dev,
> >>>> +                            struct iio_chan_spec const *chan,
> >>>> +                            int *val, int *val2, long mask)
> >>>> +{
> >>>> +     struct sc27xx_adc_data *data = iio_priv(indio_dev);
> >>>> +     int scale, ret, tmp;
> >>>> +
> >>>> +     switch (mask) {
> >>>> +     case IIO_CHAN_INFO_RAW:
> >>>> +     case IIO_CHAN_INFO_AVERAGE_RAW:
> >>>> +             mutex_lock(&indio_dev->mlock);
> >>>> +             scale = data->channel_scale[chan->channel];
> >>>> +             ret = sc27xx_adc_read(data, chan->channel, scale,  
> >>&tmp);  
> >>>> +             mutex_unlock(&indio_dev->mlock);
> >>>> +
> >>>> +             if (ret)
> >>>> +                     return ret;
> >>>> +
> >>>> +             *val = tmp;
> >>>> +             return IIO_VAL_INT;
> >>>> +
> >>>> +     case IIO_CHAN_INFO_PROCESSED:
> >>>> +             mutex_lock(&indio_dev->mlock);
> >>>> +             scale = data->channel_scale[chan->channel];
> >>>> +             ret = sc27xx_adc_read_processed(data, chan->channel,  
> >>scale,  
> >>>> +                                             &tmp);  
> >>>
> >>> To keep to the rule of 'one way to read a value' we don't tend to  
> >>support  
> >>> both raw and processed.  The only exception is made for devices where  
> >>we got  
> >>> this wrong in the first place and so have to support both to avoid a  
> >>potential  
> >>> regression due to ABI changes.
> >>>
> >>> If it is a simple linear scaling (like here I think) then the  
> >>preferred option  
> >>> is to not supply the processed version.  Just do raw.  
> >>
> >>Unfortunately, we can not use the formula ( (X_raw + X_offset)*X_scale
> >>= X_processed) for our ADC controller to get a processed value.
> >>Firstly, the ADC hardware will do the sampling with the scale value.  
> >
> > Hmm fair enough, scale is fine for that. But don't provide raw unless real voltage is scale *raw
> >  
> >>Secondly we should convert a raw value to a voltage value by the
> >>linear graph table, for some channels, we should also use the channel
> >>voltage ratio to get a real voltage value. So I think we should keep
> >>our special processed approach for consumers.  
> >
> > That's fine but drop the raw access or you are not obeying the abi.  
> 
> Sorry, I think I did not get your points. Could you elaborate on why
> we can not provide raw and processed?
> I saw many drivers not only provide the raw access but also the
> processed access. Especially for our special processed approach, I
> think the raw access and the processed access are both needed for
> consumers. Thanks.
> 

That's not true.  There are a few 'very special cases' where this happens.
We had cases where the driver was already putting out RAW values when
it turned out there was no sensible way of converting it to processed values
(non linear as in your case).  As the RAW interface was existing ABI we had
to maintain it even though in some senses this was a bug.

The other cases that look a bit like this are when multiple input sensors
are fused to produce a computed output value.  This happens with light
sensors.  However, the raw and processed channels are different channels
(as there is no one to one mapping).

It is not uncommon to provide a 'mixture' of raw and processed but they
are not for the same channels.

So in general we simply do not allow both to be provided for a single channel.
There is no useful purpose in doing so and it complicates the exposed ABI.

Jonathan

Powered by blists - more mailing lists