[PATCH v5] i2c: imx: make bus recovery through pinctrl optional

[Leo Li]

On Tue, Sep 6, 2016 at 4:51 PM, Stefan Agner <stefan at agner.ch> wrote:
> On 2016-09-06 13:06, Leo Li wrote:
>> On Tue, Sep 6, 2016 at 1:58 PM, Uwe Kleine-König
>> <u.kleine-koenig at pengutronix.de> wrote:
>>> On Fri, Aug 19, 2016 at 05:05:22PM -0500, Li Yang wrote:
>>>> Since commit 1c4b6c3bcf30 ("i2c: imx: implement bus recovery") the
>>>> driver starts to use gpio/pinctrl to do i2c bus recovery.  But pinctrl
>>>> is not always available for platforms with this controller such as ls1021a
>>>> and ls1043a, and the device tree binding also mentioned this gpio based
>>>> recovery mechanism as optional.  The patch makes it really optional that
>>>> the probe function won't bailout but just disable the bus recovery function
>>>> when pinctrl is not available.
>>>>
>>>> Signed-off-by: Li Yang <leoyang.li at nxp.com>
>>>> Cc: Gao Pan <pandy.gao at nxp.com>
>>>> Cc: Uwe Kleine-König <u.kleine-koenig at pengutronix.de>
>>>> Acked-by: Linus Walleij <linus.walleij at linaro.org>
>>>> ---
>>>> v5:
>>>> Revert the last minute change of recovery info initialization timing, it
>>>> will cause problem if initialized after i2c_add_numbered_adapter()
>>>>
>>>> v4:
>>>> Remove the use of IS_ERR_OR_NULL
>>>> Move the condition judgement to i2c_imx_init_recovery_info()
>>>> Change the timing of recovery initialization to be after bus registration
>>>>
>>>> v3:
>>>> Rebased to Wolfram's for-next branch
>>>> Added acked-by from Linus Walleij
>>>> Update to use new nxp email addresses due to company merge
>>>>
>>>>  drivers/i2c/busses/i2c-imx.c | 15 ++++++++++++++-
>>>>  1 file changed, 14 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/i2c/busses/i2c-imx.c b/drivers/i2c/busses/i2c-imx.c
>>>> index 1844bc9..7ae7992 100644
>>>> --- a/drivers/i2c/busses/i2c-imx.c
>>>> +++ b/drivers/i2c/busses/i2c-imx.c
>>>> @@ -989,6 +989,15 @@ static void i2c_imx_init_recovery_info(struct imx_i2c_struct *i2c_imx,
>>>>  {
>>>>       struct i2c_bus_recovery_info *rinfo = &i2c_imx->rinfo;
>>>>
>>>> +     /* if pinctrl is not supported on the system */
>>>> +     if (IS_ERR(i2c_imx->pinctrl))
>>>> +             i2c_imx->pinctrl = NULL;
>>>> +
>>>> +     if (!i2c_imx->pinctrl) {
>>>> +             dev_info(&pdev->dev, "can't get pinctrl, bus recovery not supported\n");
>>>> +             return;
>>>> +     }
>>>> +
>>>>       i2c_imx->pinctrl_pins_default = pinctrl_lookup_state(i2c_imx->pinctrl,
>>>>                       PINCTRL_STATE_DEFAULT);
>>>>       i2c_imx->pinctrl_pins_gpio = pinctrl_lookup_state(i2c_imx->pinctrl,
>>>> @@ -1081,8 +1090,11 @@ static int i2c_imx_probe(struct platform_device *pdev)
>>>>               return ret;
>>>>       }
>>>>
>>>> +     /* optional bus recovery feature through pinctrl */
>>>>       i2c_imx->pinctrl = devm_pinctrl_get(&pdev->dev);
>>>> -     if (IS_ERR(i2c_imx->pinctrl)) {
>>>> +     /* bailout on -ENOMEM or -EPROBE_DEFER, continue for other errors */
>>>> +     if (PTR_ERR(i2c_imx->pinctrl) == -ENOMEM ||
>>>> +                     PTR_ERR(i2c_imx->pinctrl) == -EPROBE_DEFER) {
>>>>               ret = PTR_ERR(i2c_imx->pinctrl);
>>>>               goto clk_disable;
>>>>       }
>>>
>>> devm_pinctrl_get might return the following error-valued pointers:
>>>  - -EINVAL
>>>  - -ENOMEM
>>>  - -ENODEV
>>>  - -EPROBE_DEFER
>>>
>>> There are several error paths returning -EINVAL, one is when an invalid
>>> phandle is used. Do you really want to ignore that?
>>>
>>> IMO error handling is better done with inverse logic, that is continue
>>> on some explicit error, bail out on all unknown stuff. This tends to be
>>> more robust. Also the comment should be improved to not explain that for
>>> -ENOMEM and -EPROBE_DEFER we bail out (which should be obvious for
>>> anyone who can read C) but to explain why.
>>
>> What you said is true for normal error handling, but in this scenario
>> it is intentional to ignore all pinctrl related errors except critical
>> ones because failing to have pinctrl for an optional feature shouldn't
>> impact the function of normal i2c.  We choose to catch -ENOMEM because
>> the error could also cause problem for i2c probe, and -EPROBE_DEFER
>> because it's possible that the pinctrl will be ready later and we want
>> to give it a chance.  The i2c driver really don't care why the pinctrl
>> was not usable.  I thought I added comment before the
>
> I don't agree. E.g. -EINVAL would appear if you pass devm_pinctrl_get an
> invalid device. Currently you would silently ignore that, which is not
> what you want.

It is not silently ignored, there will be a message printed out saying
pinctrl is not available and bus recovery is not supported.  On the
contrary, without this change the entire i2c driver fails to work
silently if pinctrl is somehow not working.  And if the system is so
broken that the pointer to the i2c device is NULL, the probe of i2c
would have already failed before this point.  We shouldn't count on an
optional function of the driver to catch fundamental issues like this.

>
> You want to get the pinctrl in any case expect there isn't one. And that
> is how you should formulate your if statement.
>
> /*
>  * It is ok if no pinctrl device is available. We'll not be able to use
> the
>  * bus recovery feature, but otherwise the driver works fine...
>  */
> if (PTR_ERR(i2c_imx->pinctrl) != -ENODEV)

I agree that there could be other possibilities that the pinctrl
failed to work beside the reason I described in the commit
message(platform doesn't support pinctrl at all).  But I don't think
any of them other than the -ENOMEM and -EPROBE_DEFER deserves a bail
out for the entire i2c driver.

Regards,
Leo