Re: Suspicious debounce handling code in pintctrl-baytrail

[Jean Delvare <jdelvare@suse.de>]

Hi Mika, hello Andy,

On Thu, 26 Jan 2017 15:43:41 +0200, Mika Westerberg wrote:
> On Thu, Jan 26, 2017 at 02:20:20PM +0100, Jean Delvare wrote:
> > Hi Cristina,
> > 
> > In this commit:
> > 
> > commit 658b476c742fe379e7020309fd590a27b457a4c1
> > Date:   Fri Apr 1 14:00:07 2016 +0300
> > 
> >     pinctrl: baytrail: Add debounce configuration
> > 
> > you added support for getting and setting debounce configuration for
> > the Baytrail pins. Now gcc complains about the following:
> > 
> >   CC [M]  drivers/pinctrl/intel/pinctrl-baytrail.o
> > drivers/pinctrl/intel/pinctrl-baytrail.c: In function ‘byt_pin_config_set’:
> > drivers/pinctrl/intel/pinctrl-baytrail.c:1181:17: warning: variable ‘debounce’ set but not used [-Wunused-but-set-variable]
> >   u32 conf, val, debounce;
> >                  ^
> > 
> > I looked at the code, and it indeed looks wrong. You are reading the
> > BYT_DEBOUNCE_REG register, clearing the debounce time bits from it, and
> > then writing the debounce time bits to the *BYT_CONF0_REG* register.
> > This is certainly corrupting the chip configuration, as the
> > configuration register bits 2-0 have a completely different meaning
> > (mux configuration.)
> 
> Yes, you are right.
> 
> This is fixed by following commit from Andy:
> 
> https://git.kernel.org/cgit/linux/kernel/git/linusw/linux-pinctrl.git/commit/?h=fixes&id=04ff5a095d662e0879f0eb04b9247e092210aeff

Thanks for the pointer, I wasn't aware of that patch. Looks overall
good, but:

+			case 0:
+				conf &= BYT_DEBOUNCE_EN;
+				break;

looks suspicious to me. I think ~BYT_DEBOUNCE_EN was intended?
Furthermore I would expect conf |= BYT_DEBOUNCE_EN in the other cases,
otherwise disabling debounce once will prevent you from enabling it
forever.

Also Andy's patch doesn't seem to address my second point:

> Additionally, I suspect that in the "get" path, the following is wrong:
> 
> 	case PIN_CONFIG_INPUT_DEBOUNCE:
> 		if (!(conf & BYT_DEBOUNCE_EN))
> 			return -EINVAL;
> 
> I'm not familiar with the pinctrl interface but my understanding of
> the PIN_CONFIG_INPUT_DEBOUNCE documentation is that the driver must
> return 0 if debouncing is disabled.

?

Thanks,
-- 
Jean Delvare
SUSE L3 Support