Re: [PATCH] clk: mvebu: Prevent division by zero in clk_double_div_recalc_rate()

[Andrew Lunn]

On Mon, Sep 09, 2024 at 04:38:07PM +0300, Alexandra Diupina wrote:
> get_div() may return zero, so it is necessary to check
> before calling DIV_ROUND_UP_ULL().
> 
> Found by Linux Verification Center (linuxtesting.org) with SVACE.
> 
> Fixes: 8ca4746a78ab ("clk: mvebu: Add the peripheral clock driver for Armada 3700")
> Signed-off-by: Alexandra Diupina <adiupina@xxxxxxxxxxxxx>
> ---
>  drivers/clk/mvebu/armada-37xx-periph.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/clk/mvebu/armada-37xx-periph.c b/drivers/clk/mvebu/armada-37xx-periph.c
> index 8701a58a5804..d0e1d591e4f2 100644
> --- a/drivers/clk/mvebu/armada-37xx-periph.c
> +++ b/drivers/clk/mvebu/armada-37xx-periph.c
> @@ -343,7 +343,10 @@ static unsigned long clk_double_div_recalc_rate(struct clk_hw *hw,
>  	div = get_div(double_div->reg1, double_div->shift1);
>  	div *= get_div(double_div->reg2, double_div->shift2);
>  
> -	return DIV_ROUND_UP_ULL((u64)parent_rate, div);
> +	if (!div)
> +		return 0;

Looking at this code, it seems to me some fundamental assumption has
gone wrong here, if the dividers are 0. We want to know about this,
and a kernel taking a / 0 exception would be a good way to indicate
something is very wrong. Won't returning 0 just hide the problem, not
make it obvious?

Checking for a /0 on user input makes a lot of sense, but here, i
think you are just hiding bugs. Please consider this when making
similar changes in other parts of the kernel. Why has a /0 happened?

Tools like SVACE just point at possible problems. You then need to
look at them in detail, understand the context, and decide on the
proper fix, which might actually be, a /0 is good.

	Andrew