Re: [PATCH v6 08/13] iio: afe: rescale: reduce risk of integer overflow

[Peter Rosin]

On 2021-07-28 02:07, Liam Beguin wrote:
> On Fri Jul 23, 2021 at 5:17 PM EDT, Peter Rosin wrote:
>> On 2021-07-21 05:06, Liam Beguin wrote:
>>> From: Liam Beguin <lvb@xxxxxxxxxx>
>>>
>>> Reduce the risk of integer overflow by doing the scale calculation with
>>> 64bit integers and looking for a Greatest Common Divider for both parts
>>> of the fractional value when required.
>>>
>>> Signed-off-by: Liam Beguin <lvb@xxxxxxxxxx>
>>> ---
>>>  drivers/iio/afe/iio-rescale.c | 15 ++++++++++++---
>>>  1 file changed, 12 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/iio/afe/iio-rescale.c b/drivers/iio/afe/iio-rescale.c
>>> index 6f6a711ae3ae..35fa3b4e53e0 100644
>>> --- a/drivers/iio/afe/iio-rescale.c
>>> +++ b/drivers/iio/afe/iio-rescale.c
>>> @@ -21,12 +21,21 @@
>>>  int rescale_process_scale(struct rescale *rescale, int scale_type,
>>>  			  int *val, int *val2)
>>>  {
>>> -	unsigned long long tmp;
>>> +	s64 tmp, tmp2;
>>> +	u32 factor;
>>>  
>>>  	switch (scale_type) {
>>>  	case IIO_VAL_FRACTIONAL:
>>> -		*val *= rescale->numerator;
>>> -		*val2 *= rescale->denominator;
>>> +		if (check_mul_overflow(*val, rescale->numerator, (s32 *)&tmp) ||
>>> +		    check_mul_overflow(*val2, rescale->denominator, (s32 *)&tmp2)) {
>>> +			tmp = (s64)*val * rescale->numerator;
>>> +			tmp2 = (s64)*val2 * rescale->denominator;
>>> +			factor = gcd(tmp, tmp2);
> 
> Hi Peter,
> 
>>
>> Hi!
>>
>> Reiterating that gcd() only works for unsigned operands, so this is
>> broken for
>> negative values.
> 
> Apologies, I didn't mean to make it seem like I ignored your comments. I
> should've added a note. After you pointed out that gcd() only works for
> unsigned elements, I added test cases for negative values, and all tests
> passed. I'll look into it more.

Maybe I've misread the code and gcd is in fact working for negative
numbers? However, I imagine it might be arch specific, so testing on
a single arch feels insufficient and deeper analysis is required.

However, looking at lib/math/gcd.c it certainly still looks like
negative values will work very poorly, and there is no macro magic
in include/linux/gcd.h to handle it by wrapping the core C routine.

> rescale_voltage_divider_props() seems to also use gcd() with signed
> integers.

The type of the operands may be s32, but if you look at how those values
are populated, and with what they are populated, I think you will find that
only positive scale factors are sensible for a voltage divider. Using
resistors with so high resistance that s32 is not enough is simply not
supported.

Cheers,
Peter

> Thanks,
> Liam
> 
>>
>> Cheers,
>> Peter
>>
>>> +			tmp = div_s64(tmp, factor);
>>> +			tmp2 = div_s64(tmp2, factor);
>>> +		}
>>> +		*val = tmp;
>>> +		*val2 = tmp2;
>>>  		return scale_type;
>>>  	case IIO_VAL_INT:
>>>  		*val *= rescale->numerator;
>>>
>