On 12/19/24 6:58 PM, Conor Dooley wrote:
> On Thu, Dec 19, 2024 at 03:42:11PM +0000, Tudor Ambarus wrote:
>> Hi, Conor,
>>
>> On 12/19/24 2:11 PM, Conor Dooley wrote:
>>>> There are mailbox clients that can discover the mailbox channel ID at
>>>> run-time. For such cases passing the channel identifier via DT is
>>>> redundant. Add support for referencing controllers solely by node.
>>> I don't really get your implementation, why not just allow #mbox-cells = 0?
>>> That's what's done for things like fixed frequency clocks that only have
>>> a single output.
>>
>> Ah, indeed!
>>
>> instead of:
>> of_parse_phandle(dev->of_node, "mbox", 0);
>> I can do a:
>> of_parse_phandle_with_args(dev->of_node, "mboxes",
>> "#mbox-cells", 0, &of_args)
>> where #mbox-cells = 0;
>>
>> Or ... can I pass NULL for cells_name and make the #mbox-cells property
>> optional and still keeping its requirement of being at least 1?
>
> I think the mbox-cells = 0 approach is preferred, that property is what
> marks it as a mailbox controller after all. Perhaps Rob or Krzysztof can
> comment?
I think using mbox-cells = 0 is better indeed. In my proposal I
considered the list to always have a single phandle, thus a reference to
a single mailbox controller, whereas it may be possible that clients to
reference multiple mailbox controllers. If so, #mbox-cells needs to be
defined in all the controllers, for consistency reasons, similar to what
happens with fixed clocks, as you already mentioned.
Thus I'll change the method to:
struct mbox_chan *mbox_request_channel_by_args(struct mbox_client *cl,
int index, const struct mbox_xlate_args *spec);
and use of_parse_phandle_with_args() in it.
Thanks, Conor!
ta
