[PATCH] i2c: imx: make bus recovery through pinctrl optional

[Li Yang]

On Thu, Nov 19, 2015 at 2:08 AM, Uwe Kleine-König
<u.kleine-koenig at pengutronix.de> wrote:
> Hello,
>
> Cc += Linus Walleij, linux-gpio
>
> On Wed, Nov 18, 2015 at 04:59:11PM -0600, Li Yang wrote:
>> On Wed, Nov 18, 2015 at 1:24 AM, Uwe Kleine-König
>> <u.kleine-koenig at pengutronix.de> wrote:
>> > On Tue, Nov 17, 2015 at 06:02:59PM -0600, Li Yang wrote:
>> >> @@ -1125,7 +1119,12 @@ static int i2c_imx_probe(struct platform_device *pdev)
>> >>               goto clk_disable;
>> >>       }
>> >>
>> >> -     i2c_imx_init_recovery_info(i2c_imx, pdev);
>> >> +     /* optional bus recovery feature through pinctrl */
>> >> +     i2c_imx->pinctrl = devm_pinctrl_get(&pdev->dev);
>> >> +     if (IS_ERR_OR_NULL(i2c_imx->pinctrl))
>> >> +             dev_info(&pdev->dev, "can't get pinctrl, bus recovery feature disabled\n");
>> >> +     else
>> >> +             i2c_imx_init_recovery_info(i2c_imx, pdev);
>> >
>> > I'm pretty sure this is wrong. If pinctrl isn't available
>> > devm_pinctrl_get returns NULL? But AFAIK you must not ignore an error,
>> > so the better thing to do is:
>>
>> If CONFIG_PINCTRL is not enabled, the devm_pinctrl_get() will return
>> NULL directly as defined in the include/linux/pinctrl/consumer.h.
>>
>> If CONFIG_PINCTRL is enabled because we are using a multi-platform
>> image but the actual hardware used doesn't have a pinctrl driver or
>> pinctrl device tree nodes.  It is expected that the devm_pinctrl_get()
>> will return error.  But as the pinctrl is only used for bus recovery
>> which is just an optional function of this driver.  We shouldn't
>> bailout the probe but keep the driver working without the bus recovery
>> function.  As for generic errors like (!dev) or out-of-memory, the
>> probe will fail elsewhere anyway.
>
> IMHO you shouldn't ignore -ENOMEM from devm_pinctrl_get but just the
> case for "there was no pinctrl specified in dt". Also consider the i2c
> driver is loaded before the pinctrl driver and devm_pinctrl_get returns
> -EPROBE_DEFER (I assume it does, didn't check), this should be handled.

Make sense.   -EPROBE_DEFER should be honored.  Maybe it's good to
check the -ENOMEM too.  I will send a new version.

Thanks,
Leo