Re: [PATCH] pinctrl: freescale: avoid overwriting pin config when freeing GPIO

From: Stefan Agner
Date: Tue Sep 27 2016 - 15:40:26 EST

On 2016-09-27 11:17, Vladimir Zapolskiy wrote:
> Hi Stefan,
> On 09/27/2016 07:37 PM, Stefan Agner wrote:
>> On 2016-09-27 05:12, Vladimir Zapolskiy wrote:
>>> Hi Stefan,
>>> On 09/27/2016 03:26 AM, Stefan Agner wrote:
>>>> If a GPIO gets freed after selecting a new pinctrl configuration
>>>> the driver should not change pinctrl anymore. Otherwise this will
>>>> likely lead to a unusable pin configuration for > the newly selected
>>>> pinctrl.
>>>> Signed-off-by: Stefan Agner <stefan@xxxxxxxx>
>>>> ---
>>>> This turned out to be problematic when using the I2C GPIO bus recovery
>>>> functionality. After muxing back to I2C, the GPIO is being freed, which
>>>> cased I2C to stop working completely.
>>> IMHO this recent "i.MX I2C GPIO bus recovery" feature is kind of a hack,
>>> for example I believe it breaks I2C bus driver initialization on i.MX31
>>> boards, where today there is no pinctrl driver at all.
>> This has been addressed by Li Yang's patch, already in the next branch:
> Nice to know about it, thank you for the link.
>>> IMHO something like I've partially described in the recent "Requesting as
>>> a GPIO a pin already used through pinctrl" topic should be done here.
>>> Could you consider to add another pinctrl-1 group with alternative GPIO
>>> line mux/config settings to an i2c controller device node and apply it,
>>> when you need a bus recovery? You may find references how this kind of
>>> dynamic pinctrl management is done within mmc/sd subsystem.
>> I don't quite understand, that is already the case. This is what device
>> tree looks like to get the I2C recovery functionality:
>> &i2c1 {
>> pinctrl-names = "default", "gpio";
>> pinctrl-0 = <&pinctrl_i2c1>;
>> pinctrl-1 = <&pinctrl_i2c1_gpio>;
>> scl-gpios = <&gpio1 6 GPIO_ACTIVE_HIGH>;
>> sda-gpios = <&gpio1 7 GPIO_ACTIVE_HIGH>;
>> status = "okay";
>> };
> Great, then why do you experience a problem you've described?
>>>> After muxing back to I2C, the GPIO is being freed, which cased I2C
>>>> to stop working completely.
> Release GPIO firstly, then mux back to I2C, that's the correct sequence
> and I believe it obsoletes this change.

Added Viresh Kumar to the discussion, he implemented the I2C recovery

Yes, reordering the pinctrl/gpio_free calls would fix the problem too.

However, I guess there is no explicit rule to that ("request/free GPIOs
only when they are muxed as GPIO"), so I think of it that the issue is
actually in the pinctrl driver.

On top of that it is not entirely trivial to reorder the calls the way
i2c_generic_gpio_recovery and i2c_generic_recovery are set up right now.

>>> By the way did I miss a patch, which falls back to mux settings on
>>> .gpio_disable_free call for non-Vybrid platforms?
>> Currently only Vybrid makes use of the .gpio_request_enable... and so
>> should .gpio_disable_free then.
> So, I guess this is a change with a runtime difference for Vybrid only.

That is correct.

> I find that it was initially done wrong that a number of Vybrid specific
> hooks were added to the shared pinctrl-imx.c, in my opinion it is better
> to make needed abstractions and move all code around SHARE_MUX_CONF_REG
> to pinctrl-vf610.c:
> ./pinctrl-imx.c:216: if (info->flags & SHARE_MUX_CONF_REG) {
> ./pinctrl-imx.c:317: if (!(info->flags & SHARE_MUX_CONF_REG))
> ./pinctrl-imx.c:357: if (!(info->flags & SHARE_MUX_CONF_REG))
> ./pinctrl-imx.c:382: if (!(info->flags & SHARE_MUX_CONF_REG))
> ./pinctrl-imx.c:425: if (info->flags & SHARE_MUX_CONF_REG)
> ./pinctrl-imx.c:450: if (info->flags & SHARE_MUX_CONF_REG) {
> ./pinctrl-imx.c:534: if (info->flags & SHARE_MUX_CONF_REG)
> ./pinctrl-imx.c:575: if (info->flags & SHARE_MUX_CONF_REG) {
> Nevertheless this is not directly related to the change.

Yeah I was really on the fence there. Despite the same name, the IOMUXC
on Vybrid is quite a bit different than on i.MX. The problem is it is
not easily possible to factor out the SoC specific stuff into
pinctrl-vf610.c I would have basically had to add tons of callbacks or
re-implement lots of functions in pinctrl-vf610.c. Maybe it would have
probably been better to basically implement Vybrids IOMUXC as its own
pinctrl driver.