[v3,2/2] clk:mmp: clk-mix.c fix divide-by-zero

Commit Message

Xiaoming Ni March 30, 2019, 1:55 p.m. UTC

The _get_div() function has a branch with a return value of 0
Add a check on the return value of _get_div() to avoid divide-by-zero

Signed-off-by: nixiaoming <[e-mail]>
Reviewed-by: Mukesh Ojha <[e-mail]>
---
 drivers/clk/mmp/clk-mix.c | 6 ++++++
 1 file changed, 6 insertions(+)

Comments

Stephen Boyd April 23, 2019, 11 p.m. UTC | #1

Quoting nixiaoming (2019-03-30 06:55:42)
> The _get_div() function has a branch with a return value of 0
> Add a check on the return value of _get_div() to avoid divide-by-zero
> 

Are you seeing this in practice? Or just trying to avoid a div-by-zero
case that you've found from inspection? 

> Signed-off-by: nixiaoming <[e-mail]>
> Reviewed-by: Mukesh Ojha <[e-mail]>

Xiaoming Ni April 24, 2019, 3:34 p.m. UTC | #2

On Wed, Apr 24, 2019 at 7:00 AM Stephen Boyd <[e-mail]> wrote:
>Quoting nixiaoming (2019-03-30 06:55:42)
>> The _get_div() function has a branch with a return value of 0
>> Add a check on the return value of _get_div() to avoid divide-by-zero
>> 
>
>Are you seeing this in practice? Or just trying to avoid a div-by-zero
>case that you've found from inspection? 
>
This potential bug is found by code inspection.
_get_div() is defined as a static function which is only refered twice 
in drivers/clk/mmp/clk-mix.c. In both cases the return value of _get_div()
 is used as divider without any check.
 If _get_div() never returns 0, then the branch returning 0 is dead code,
or the return value should be check to avoid dividing by zero error. 

thanks

>> Signed-off-by: nixiaoming <[e-mail]>
>> Reviewed-by: Mukesh Ojha <[e-mail]>
>

Stephen Boyd April 24, 2019, 8:55 p.m. UTC | #3

Quoting Nixiaoming (2019-04-24 08:34:54)
> On Wed, Apr 24, 2019 at 7:00 AM Stephen Boyd <[e-mail]> wrote:
> >Quoting nixiaoming (2019-03-30 06:55:42)
> >> The _get_div() function has a branch with a return value of 0
> >> Add a check on the return value of _get_div() to avoid divide-by-zero
> >> 
> >
> >Are you seeing this in practice? Or just trying to avoid a div-by-zero
> >case that you've found from inspection? 
> >
> This potential bug is found by code inspection.
> _get_div() is defined as a static function which is only refered twice 
> in drivers/clk/mmp/clk-mix.c. In both cases the return value of _get_div()
>  is used as divider without any check.
>  If _get_div() never returns 0, then the branch returning 0 is dead code,
> or the return value should be check to avoid dividing by zero error. 
> 

Ok. If it's found by code inspection then I'd rather wait until someone
can confirm what the behavior should be when _get_div() returns 0. Is it
an error case that never happens or does the hardware support 0 as a
divider value meaning "bypass"? I'd like to understand that before
patching up a potential div-by-zero case and not knowing what it means.

Patch

diff --git a/drivers/clk/mmp/clk-mix.c b/drivers/clk/mmp/clk-mix.c
index 90814b2..6ed5ad7 100644
--- a/drivers/clk/mmp/clk-mix.c
+++ b/drivers/clk/mmp/clk-mix.c
@@ -245,6 +245,9 @@  static int mmp_clk_mix_determine_rate(struct clk_hw *hw,
 			div_val_max = _get_maxdiv(mix);
 			for (j = 0; j < div_val_max; j++) {
 				div = _get_div(mix, j);
+				/* avoid divide-by-zero */
+				if (!div)
+					continue;
 				mix_rate = parent_rate / div;
 				gap = abs(mix_rate - req->rate);
 				if (!parent_best || gap < gap_best) {
@@ -341,6 +344,9 @@  static unsigned long mmp_clk_mix_recalc_rate(struct clk_hw *hw,
 	shift = mix->reg_info.shift_div;
 
 	div = _get_div(mix, MMP_CLK_BITS_GET_VAL(mux_div, width, shift));
+	/* avoid divide-by-zero */
+	if (!div)
+		return -EINVAL;
 
 	return parent_rate / div;
 }