Re: [PATCH v5 05/10] iio: afe: rescale: add INT_PLUS_{MICRO,NANO} support

[Peter Rosin <peda@axentia.se>]

On 2021-07-19 01:44, Liam Beguin wrote:
> On Sat Jul 17, 2021 at 12:55 PM EDT, Jonathan Cameron wrote:
>> On Fri, 16 Jul 2021 15:18:33 -0400
>> "Liam Beguin" <liambeguin@gmail.com> wrote:
>>
>>> On Thu Jul 15, 2021 at 5:48 AM EDT, Peter Rosin wrote:
>>>>
>>>> On 2021-07-15 05:12, Liam Beguin wrote:  
>>>>> From: Liam Beguin <lvb@xiphos.com>
>>>>>
>>>>> Some ADCs use IIO_VAL_INT_PLUS_{NANO,MICRO} scale types.
>>>>> Add support for these to allow using the iio-rescaler with them.
>>>>>
>>>>> Signed-off-by: Liam Beguin <lvb@xiphos.com>
>>>>> ---
>>>>>  drivers/iio/afe/iio-rescale.c | 15 +++++++++++++++
>>>>>  1 file changed, 15 insertions(+)
>>>>>
>>>>> diff --git a/drivers/iio/afe/iio-rescale.c b/drivers/iio/afe/iio-rescale.c
>>>>> index 4c3cfd4d5181..a2b220b5ba86 100644
>>>>> --- a/drivers/iio/afe/iio-rescale.c
>>>>> +++ b/drivers/iio/afe/iio-rescale.c
>>>>> @@ -92,7 +92,22 @@ static int rescale_read_raw(struct iio_dev *indio_dev,
>>>>>  			do_div(tmp, 1000000000LL);
>>>>>  			*val = tmp;
>>>>>  			return ret;
>>>>> +		case IIO_VAL_INT_PLUS_NANO:
>>>>> +			tmp = ((s64)*val * 1000000000LL + *val2) * rescale->numerator;
>>>>> +			do_div(tmp, rescale->denominator);
>>>>> +
>>>>> +			*val = div_s64(tmp, 1000000000LL);
>>>>> +			*val2 = tmp - *val * 1000000000LL;
>>>>> +			return ret;  
>>>>
>>>> This is too simplistic and prone to overflow. We need something like
>>>> this
>>>> (untested)
>>>>
>>>> tmp = (s64)*val * rescale->numerator;
>>>> rem = do_div(tmp, rescale->denominator);
>>>> *val = tmp;
>>>> tmp = ((s64)rem * 1000000000LL + (s64)*val2) * rescale->numerator;
>>>> do_div(tmp, rescale->denominator);
>>>> *val2 = tmp;
>>>>
>>>> Still not very safe with numerator and denominator both "large", but
>>>> much
>>>> better. And then we need normalizing the fraction part after the above,
>>>> of
>>>> course.
>>>>  
>>>
>>> Understood, I'll test that.
>>>
>>>> And, of course, I'm not sure what *val == -1 and *val2 == 500000000
>>>> really
>>>> means. Is that -1.5 or -0.5? The above may very well need adjusting for
>>>> negative values...
>>>>  
>>>
>>> I would've assumed the correct answer is -1 + 500000000e-9 = -0.5
>>> but adding a test case to iio-test-format.c seems to return -1.5...
>>
> 
> Hi Jonathan,
> 
>> No. -1.5 is as intended, though the IIO_VAL_PLUS_MICRO is rather
>> confusing
>> naming :( We should perhaps add more documentation for that. Signs were
>> always a bit of a pain with this two integer scheme for fixed point.
>>
>> The intent is to have moderately readable look up tables with the
>> problem that
>> we don't have a signed 0 available. Meh, maybe this decision a long time
>> back wasn't a the right one, but it may be a pain to change now as too
>> many
>> drivers to check!
>>
>> 1, 0000000 == 1
>> 0, 5000000 == 0.5
>> 0, 0000000 == 0
>> 0, -5000000 == -0.5
>> -1, 5000000 == -1.5
>>
> 
> Understood, thanks for clearing that out.

I just realized that do_div assumes unsigned operands...

:-(

Cheers,
Peter