[PATCH] gpio: uclass: Introduce gpio-hog-optional property

[Nate Drude]

Hi Simon and Fabio,

On 9/12/22 3:16 PM, Simon Glass wrote:
> Hi,
> 
> On Mon, 12 Sept 2022 at 12:48, Fabio Estevam <festevam at gmail.com> wrote:
>>
>> Hi Nate and Simon,
>>
>> On Mon, Sep 12, 2022 at 2:55 PM Nate Drude <nate.d at variscite.com> wrote:
>>>
>>> gpio_hog_probe_all is invoked by init_sequence_r in board_r.c.
>>> If device_probe fails for any gpio-hog, boot hangs with the following error:
>>>
>>>> initcall sequence 00000000fffc8e18 failed at call 000000004023b320 (err=-121)
>>>> ### ERROR ### Please RESET the board ###
>>>
>>> gpio-hog-optional allows the boot sequence to continue if device_probe
>>> fails for optional gpio-hog(s).
>>>
>>> Signed-off-by: Nate Drude <nate.d at variscite.com>
>>> ---
>>>   doc/device-tree-bindings/gpio/gpio.txt | 1 +
>>>   drivers/gpio/gpio-uclass.c             | 4 +++-
>>>   2 files changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/doc/device-tree-bindings/gpio/gpio.txt b/doc/device-tree-bindings/gpio/gpio.txt
>>> index 1481ed607d..02d296316b 100644
>>> --- a/doc/device-tree-bindings/gpio/gpio.txt
>>> +++ b/doc/device-tree-bindings/gpio/gpio.txt
>>> @@ -189,6 +189,7 @@ gpio-controller's driver probe function.
>>>   Each GPIO hog definition is represented as a child node of the GPIO controller.
>>>   Required properties:
>>>   - gpio-hog:   A property specifying that this child node represents a GPIO hog.
>>> +- gpio-hog-optional: A property specifying to continue boot when device_probe fails in gpio_hog_probe_all
>>
>> gpio-hog-optional property does not exist in Linux.
>>
>> If this property is introduced then U-Boot and Linux devicetrees will
>> not be in sync.
>>
>> Can this be fixed differently?
> 
> Nate, can you send a patch to Linux with the binding update?
> 
> Regards,
> Simon

Thanks for your responses and feedback.

I don't think gpio-hog-optional is relevant to Linux.

The problem is if gpio_hog_probe_all returns an error, board_init_r will 
hang()

See:
- https://github.com/u-boot/u-boot/blob/v2022.07/common/board_r.c#L824-L825
- https://github.com/u-boot/u-boot/blob/v2022.07/common/board_r.c#L763
- 
https://github.com/u-boot/u-boot/blob/v2022.07/drivers/gpio/gpio-uclass.c#L330-L332

A practical example of how this may occur is when an i2c gpio expander 
(e.g. nxp,pca9534) uses gpio-hog, but the gpio expander is depopulated.

Arguably, the best solution is to use a different device tree when the 
gpio expander is not populated. This patch allows the gpio-hog to fail 
gracefully and continue booting if gpio-hog-optional is used.

Do you have any suggestions for a better approach? Does it make sense 
for gpio_hog_probe_all to cause a fatal error when the gpio hog probe 
fails (most devices, including the gpio-expander, will not cause a hang 
if they fail to probe)?

Regards,
Nate