[U-Boot] [PATCH 1/3] core: Add function to get device for ofnode

[Simon Glass]

Hi Mario,

On 18 April 2018 at 02:20, Mario Six <mario.six at gdsys.cc> wrote:
> Hi Simon,
>
> On Thu, Apr 12, 2018 at 6:42 PM, Simon Glass <sjg at chromium.org> wrote:
>> Hi Mario,
>>
>> On 10 April 2018 at 05:34, Mario Six <mario.six at gdsys.cc> wrote:
>>> Hi Simon,
>>>
>>> On Fri, Mar 30, 2018 at 12:43 AM, Simon Glass <sjg at chromium.org> wrote:
>>>> Hi Mario,
>>>>
>>>> On 28 March 2018 at 20:37, Mario Six <mario.six at gdsys.cc> wrote:
>>>>> It's sometimes useful to get the device associated with a given
ofnode.
>>>>> Implement a function to implement this lookup operation.
>>>>
>>>> Where would you use this? Can you not use phandles to find the device?
>>>> Or uclass_get_device_by_ofnode() ?
>>>>
>>>
>>> The function is used with the dev_{enable,disable}_by_path in the next
patch:
>>> If I used any of the uclass_* functions or similar, the device would be
probed,
>>> which is not what I want, since the device may not actually be
physically
>>> present.
>>
>> So how about using uclass_find_device_by_ofnode() ?
>>
>
> That would work for the disabling, true, but not for the enabling (which
is
> what is used in the upcoming board): Since the node is declared as
disabled in
> the DT, the device is not even bound (so uclass_find_device_by_ofnode)
won't
> return it.
>
> A more elegant solution would be to have device_probe check again if the
> underlying ofnode is disabled, and stop the probing if that's the case.
In this
> scenario the disabled devices would still be displayed in the tree, but
never
> probed, which is probably OK (I don't know if there would be any side
effects
> with iterating through devices, for example). But changing the behavior
of such
> elementary API functions is probably not a good idea.

That seems to be a different topic.

Fundamentally I don't see the difference between
uclass_find_device_by_ofnode() and your ofnode_dev().

If you want to enable something after probing you will have to call
device_bind() or similar. If that is your intent, I think you need a
different function from ofnode_dev(), since it also relies on the device
already being bound.

Regards,
Simon