Re: [PATCH v3 1/3] iio: light: isl29018: fix overflow and precision in isl29018_read_lux()

[Andy Shevchenko]

On Thu, Jun 04, 2026 at 12:06:15PM +0200, Herman van Hazendonk wrote:
> The intermediate calculations in isl29018_read_lux() use 32-bit
> arithmetic throughout, which overflows in two distinct ways:
> 
> 1. lux_data * chip->scale.uscale — at 16-bit integration time and the
>    1000 fc range, scale.uscale is 976562.  A full-scale 16-bit reading
>    (65535) gives 65535 * 976562 ≈ 64 billion, far beyond UINT_MAX.
>    The value wraps before the /1000000 division can save it, producing
>    a wildly wrong data_x_range.
> 
> 2. data_x_range * chip->calibscale — even after a correct data_x_range,
>    multiplying by a calibscale of a few hundred (reasonable for a deeply
>    tinted cover glass) pushes the product past INT_MAX, causing *lux to
>    wrap negative.
> 
> Additionally, dividing lux_data * scale.uscale by 1000000 before
> applying calibscale discards the fractional-lux remainder.  For low
> ranges where scale.scale is zero, any reading below 1000000/scale.uscale
> counts truncates to a data_x_range of zero, so the calibscale
> multiplication cannot rescue it.  This creates a dead-band at low light
> levels that is especially visible when a large cover-glass compensation
> gain is in use.
> 
> Fix the overflows by widening the intermediate variables to u64 and
> using div_u64() for the divisions (plain 64-bit division emits
> __aeabi_uldivmod on ARM32, which is not available in kernel builds).
> Preserve the uscale remainder across the first division so that the
> calibscale multiplication captures the sub-lux contribution.  Clamp
> the final result to INT_MAX before storing it in the signed int *lux
> out parameter.

...

> -	unsigned int data_x_range;
> +	u32 uscale_rem;
> +	u64 uscale_term, data_x_range, result;

Please, try to keep reversed xmas tree order.

...

> +	*lux = (int)min_t(u64, result, INT_MAX);

Too many castings. What you most likely want here is clamp(), try never use
min_t().

-- 
With Best Regards,
Andy Shevchenko