Re: [PATCH v2] software_node: Add support for fwnode_graph*() family of functions

[Dan Scally <djrscally@gmail.com>]

On 18/09/2020 09:57, Heikki Krogerus wrote:
> On Fri, Sep 18, 2020 at 10:57:41AM +0300, Sakari Ailus wrote:
>> On Fri, Sep 18, 2020 at 08:46:52AM +0100, Dan Scally wrote:
>>> On 18/09/2020 08:34, Sakari Ailus wrote:
>>>> On Fri, Sep 18, 2020 at 07:49:31AM +0100, Dan Scally wrote:
>>>>> Good morning
>>>>>
>>>>> On 18/09/2020 07:22, Sakari Ailus wrote:
>>>>>> I wonder if this has something to do with replacing the device's fwnode
>>>>>> in the cio2-bridge patch.
>>>>>>
>>>>>> It's the device that needs to be enabled, and it's not a software node.
>>>>>>
>>>>> I think it is because of that yes, but I don't see a way around it at
>>>>> the moment - unless there's a way to attach the software_node port and
>>>>> endpoints that cio2-bridge creates to the device's existing firmware
>>>>> instead.
>>>> I thought this was how it was meant to be used?
>>>>
>>>> The secondary field is there for this purpose. But it may be not all fwnode
>>>> interface functions operate on fwnode->secondary?
>>> Let me test it; it might just require some changes to
>>> software_node_graph_get_port_parent() to check if the parent fwnode is a
>>> secondary, and if it is to return the primary instead.
>> Ah, indeed. I forgot this part. I wonder if it'd cause issues to return the
>> primary if you've got the secondary swnode.
>>
>> Heikki, any idea?
>>
>> Code elsewhere (e.g. V4L2 fwnode framework + drivers) assume a device is
>> identified by a single fwnode, not two --- currently the swnode graph
>> function returning port parent returns the secondary so there's no match
>> with the primary fwnode.
> Sorry I don't think I understand the scenario here, but never return
> the primary node when the software node is the secondary from the
> software node API! The software node functions deal and return
> software nodes, and nothing else, just like ACPI deals with ACPI nodes
> only and DT deals with OF nodes only. We must never jump between the
> fwnode types at this level. That also means that if you want to
> describe the device graph with software nodes, then every node in the
> graph, starting from the port parents, must be a software node.
> Whether or not the node is secondary is irrelevant. But I guess this
> is not a problem here (or is it?).
>
> Considering the secondary node will unfortunately need to be done by
> the callers of fwnode API when the fwnode API can't take care of that.
>
Alright, so if we want to attach software nodes as secondaries to a
devices existing fwnode we'd need to modify things like
fwnode_graph_get_next_endpoint_by_id() [1] to consider whether the
returned node was a software_node secondary when they try to get the
device's node to run *is_available()


I did sort of wonder whether this was the right approach before, but
there's other comments [2] in the source that reassured me, for example
device_add_properties():

>  * WARNING: The callers should not use this function if it is known that there
>  * is no real firmware node associated with @dev! In that case the callers
>  * should create a software node and assign it to @dev directly.


[1]
https://elixir.bootlin.com/linux/latest/source/drivers/base/property.c#L1126

[2]
https://elixir.bootlin.com/linux/latest/source/drivers/base/property.c#L541