Re: [PATCH] iio: adf4350: fix a possible divided-by-zero bug in adf4350_set_freq()

From: Jonathan Cameron
Date: Fri May 21 2021 - 13:24:13 EST


On Fri, 21 May 2021 14:19:53 +0800
tq17373059@xxxxxxxxxxx wrote:

> From: Qi Teng <tq17373059@xxxxxxxxxxx>
>
> The variable st->r1_mod is checked in:
> if (st->r0_fract && st->r1_mod)
>
> This indicates that st->r1_mod can be zero. Its value is the same as
> that in:
> st->r0_fract = do_div(tmp, st->r1_mod);
>
> However, st->r1_mod performs as a divisor in this statement, which
> implies a possible divided-by-zero bug.
>
> To fix this possible bug, st->r1_mod is checked before the division
> operation. If it is zero, st->r0_fract is set to zero instead of
> do_div(tmp, st->r1_mod).
>
> Reported-by: TOTE Robot <oslab@xxxxxxxxxxxxxxx>
> Signed-off-by: Qi Teng <tq17373059@xxxxxxxxxxx>

Patch seems to be inverse of the intended. You are removing a
divide by 0 protection that isn't present in the current driver.

Also, a fix like this needs a Fixes: tag.

The maths is sufficiently messy that I can't immediately tell if this
can actually be zero or whether the check you are highlighting is
paranoia. Given that, I agree that a fix here makes sense whether
or not we have a verified bug.

Thanks,

Jonathan

> ---
> drivers/iio/frequency/adf4350.c | 5 +----
> 1 file changed, 1 insertion(+), 4 deletions(-)
>
> diff --git a/drivers/iio/frequency/adf4350.c b/drivers/iio/frequency/adf4350.c
> index 99c6f260cc21..1462a6a5bc6d 100644
> --- a/drivers/iio/frequency/adf4350.c
> +++ b/drivers/iio/frequency/adf4350.c
> @@ -182,10 +182,7 @@ static int adf4350_set_freq(struct adf4350_state *st, unsigned long long freq)
>
> tmp = freq * (u64)st->r1_mod + (st->fpfd >> 1);
> do_div(tmp, st->fpfd); /* Div round closest (n + d/2)/d */
> - if (st->r1_mod)
> - st->r0_fract = do_div(tmp, st->r1_mod);
> - else
> - st->r0_fract = 0;
> + st->r0_fract = do_div(tmp, st->r1_mod);
> st->r0_int = tmp;
> } while (mdiv > st->r0_int);
>