Re: [PATCH v1 1/5] pwm: mediatek: add a property "mediatek,num-pwms"

[Uwe Kleine-König]

Hello Ryder,

On Fri, Jan 18, 2019 at 05:42:54PM +0800, Ryder Lee wrote:
> On Fri, 2019-01-18 at 08:59 +0100, Uwe Kleine-König wrote:
> > Hello,
> > 
> > On Fri, Jan 18, 2019 at 11:24:41AM +0800, Ryder Lee wrote:
> > > This adds a property "mediatek,num-pwms" to avoid having an endless
> > > list of compatibles with no differences for the same driver.
> > > 
> > > Thus, the driver should have backwards compatibility to older DTs.
> > 
> > I still think Thierry should bless "num-pwms" without vendor prefix.
> 
> Okay.
> 
> > > Signed-off-by: Ryder Lee <ryder.lee@xxxxxxxxxxxx>
> > > ---
> > > Changes since v1: add some checks for backwards compatibility.
> > > ---
> > >  drivers/pwm/pwm-mediatek.c | 27 ++++++++++++++++++---------
> > >  1 file changed, 18 insertions(+), 9 deletions(-)
> > > 
> > > diff --git a/drivers/pwm/pwm-mediatek.c b/drivers/pwm/pwm-mediatek.c
> > > index eb6674c..81b7e5e 100644
> > > --- a/drivers/pwm/pwm-mediatek.c
> > > +++ b/drivers/pwm/pwm-mediatek.c
> > > @@ -55,7 +55,7 @@ enum {
> > >  };
> > >  
> > >  struct mtk_pwm_platform_data {
> > 
> > Unrelated to this patch: This name is bad. This struct is not used as
> > platform_data and so should better be named mtk_pwm_of_data. While at
> > criticizing existing stuff: I'd prefer pwm_mediatek as common prefix to
> > match the filename.
> 
> I think we can take care about that in another patch.

That's what I wanted to say, right. Do you follow up?

> > > -	unsigned int num_pwms;
> > > +	unsigned int num_pwms;	/* it should not be used in the future SoCs */
> > 
> > I'd drop this comment in favour of a runtime warning.
> 
> Sorry, I can't get you here.

I'd do a

	dev_warn(dev, "dt didn't specify number of PWMs, falling back to %d\n", pc->soc->num_pwms);

to make people aware that updating the dt would be nice.

> 
> > >  	bool pwm45_fixup;
> > >  	bool has_clks;
> > >  };
> > > @@ -226,27 +226,36 @@ static void mtk_pwm_disable(struct pwm_chip *chip, struct pwm_device *pwm)
> > >  
> > >  static int mtk_pwm_probe(struct platform_device *pdev)
> > >  {
> > > -	const struct mtk_pwm_platform_data *data;
> > > +	struct device_node *np = pdev->dev.of_node;
> > >  	struct mtk_pwm_chip *pc;
> > >  	struct resource *res;
> > > -	unsigned int i;
> > > +	unsigned int i, num_pwms;
> > >  	int ret;
> > >  
> > >  	pc = devm_kzalloc(&pdev->dev, sizeof(*pc), GFP_KERNEL);
> > >  	if (!pc)
> > >  		return -ENOMEM;
> > >  
> > > -	data = of_device_get_match_data(&pdev->dev);
> > > -	if (data == NULL)
> > > -		return -EINVAL;
> > > -	pc->soc = data;
> > > +	pc->soc = of_device_get_match_data(&pdev->dev);
> > 
> > This might return NULL which ...
> 
> The only way to call probe() is to match an entry in
> mtk_pwm_of_match[], so match cannot be NULL.

(<pedantic>Theoretically the driver can be probed by device name, but
that is not what I meant here.</pedantic>).

You're right, as long as all entries in mtk_pwm_of_match have a non-NULL
.data entry, you're fine. I somehow thought there might be some without
one. I wouldn't oppose to check for that anyhow as a defensive measure.

> > > [...]
> > > +	/* Check if we have set 'num-pwms' in DTs. */
> > > +	ret = of_property_read_u32(np, "mediatek,num-pwms", &num_pwms);
> > > +	if (ret < 0) {
> > > +		/* If no, fallback to use SoC data for backwards compatibility. */
> > > +		if (pc->soc->num_pwms) {
> > 
> > ... here then results in a NULL pointer dereference. I think you want
> 
> So we have no chance to get a NULL pointer, right?

Ack.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |