Re: [PATCH/RFC 2/2] gpio: of: Add DT overlay support for GPIO hogs

[Frank Rowand <frowand.list@gmail.com>]

On 1/7/20 2:11 AM, Geert Uytterhoeven wrote:
> Hi Frank,
> 
> On Tue, Jan 7, 2020 at 8:10 AM Frank Rowand <frowand.list@gmail.com> wrote:
>> On 1/6/20 5:34 PM, Frank Rowand wrote:
>>> On 12/30/19 7:38 AM, Geert Uytterhoeven wrote:
>>>> As GPIO hogs are configured at GPIO controller initialization time,
>>>> adding/removing GPIO hogs in DT overlays does not work.
>>>>
>>>> Add support for GPIO hogs described in DT overlays by registering an OF
>>>> reconfiguration notifier, to handle the addition and removal of GPIO hog
>>>> subnodes to/from a GPIO controller device node.
>>>>
>>>> Note that when a GPIO hog device node is being removed, its "gpios"
>>>> properties is no longer available, so we have to keep track of which
>>>> node a hog belongs to, which is done by adding a pointer to the hog's
>>>> device node to struct gpio_desc.
>>>
>>> If I have read the patches and the existing overlay source correctly,
>>> then some observations:
>>>
>>> - A gpio hog node added in an overlay will be properly processed.
>>>
>>> - A gpio hog node already existing in the live devicetree, but with a
>>>   non-active status will be properly processed if the status of the
>>>   gpio hog node is changed to "ok" in the overlay.

Verified by testing.


>>> - If a gpio hog node already exists in the live devicetree with an
>>>   active status, then any updated or added properties in that gpio
>>>   hog node in the overlay will have no effect.

Should be documented.


>>>   There is a scenario where the updated property would have an effect:
>>>   apply a second overlay that sets the status to inactive, then apply
>>>   a third overlay that sets the status back to active.  This is a
>>>   rather contrived example and I think it should be documented as
>>>   not supported and the result undefined.

I was wrong in this case.

of_reconfig_get_state_change() does not simply report whether a node
is added or removed, which confused me because it returns
OF_RECONFIG_CHANGE_ADD and OF_RECONFIG_CHANGE_REMOVE (as well as
no change), which I was incorrectly translating to node added or
node removed.   OF_RECONFIG_CHANGE_ADD and OF_RECONFIG_CHANGE_REMOVE
properly report a node becoming available or available due to changes
in the "status" property, as well as accounting for a node being
added or removed.

So the case that I was worried about is handled correctly.


>> I went back and double checked the related code.  For gpio hog nodes
>> that are in a non-overlay, the status property is checked because
>> of_gpiochip_scan_gpios() uses for_each_available_child_of_node()
>> to search for gpio hog nodes, and for_each_available_child_of_node()
>> checks the status property.  But in the case of a gpio hog node
>> added by an overlay, of_gpio_notify() does not check the status
>> property in the gpio hog node.  The check for the status property
>> should be added to of_gpio_notify().
> 
> Right.  of_device_is_available() should be called to check this.
> Note that of_i2c_notify() and of_spi_notify() also lack such a check.
> of_platform_notify() calls of_platform_device_create_pdata(), which does
> have the check.

And thus I was wrong about this also, so of_gpio_notify() does not need to
check the status property, since of_reconfig_get_state_change() already
implicitly incorporates this check.

> 
> Gr{oetje,eeting}s,
> 
>                         Geert
>