[alsa-devel] [PATCH v1] ASoC: tlv320aic32x4: handle regmap_read error gracefully

[Peter Seiderer]

Hello Mark,

On Thu, 9 Jan 2020 20:38:19 +0000, Mark Brown <broonie at kernel.org> wrote:

> On Fri, Dec 27, 2019 at 04:20:56PM +0100, Peter Seiderer wrote:
> > @@ -338,7 +338,8 @@ static unsigned long clk_aic32x4_div_recalc_rate(struct clk_hw *hw,
>
> >  	unsigned int val;
>
> > -	regmap_read(div->regmap, div->reg, &val);
> > +	if (regmap_read(div->regmap, div->reg, &val))

Unrelated to your question, but I would change this line (on next patch
iteration) to (as all other return value checked places for regmap_read
in the same file):

	if (regmap_read(div->regmap, div->reg, &val) < 0)

> > +		return 0;
>
> Is this the best fix - shouldn't we be returning an error here?  We
> don't know what the value programmed into the device actually is so zero
> might be wrong, and we still have the risk that the value we read from
> the device may be zero if the device is misprogrammed.

clk_aic32x4_div_recalc_rate() is used for clk_ops aic32x4_div_ops.recalc_rate,
did not check/or see on first sight if there is a error handling on the usage
of recalc_rate, but did take a look at some other places where the error
handling seems to be to return zero, e.g. sound/soc/codecs/da7219.c
sound/soc/intel/skylake/skl-ssp-clk.c, etc.

The error occurred with linux-5.3.18, with the earlier versions on regmap_read
failure val was (by chance) near 2^31 and evaluated with (val & AIC32X4_DIV_MASK)
to 96 (or similar)...but with 5.3.18 evaluated to 0 and the line

	return DIV_ROUND_UP(parent_rate, val & AIC32X4_DIV_MASK);

produced the 'Division by zero'...

Regards,
Peter