Re: [RFC PATCH v2 12/17] ACPI: property: Add support for nargs_prop in acpi_fwnode_get_reference_args()

[Andy Shevchenko]

On Mon, Feb 03, 2025 at 07:21:50PM +0530, Sunil V L wrote:
> On Mon, Feb 03, 2025 at 02:36:58PM +0200, Mika Westerberg wrote:
> > On Mon, Feb 03, 2025 at 05:54:18PM +0530, Sunil V L wrote:
> > > On Mon, Feb 03, 2025 at 12:58:40PM +0200, Mika Westerberg wrote:
> > > > On Mon, Feb 03, 2025 at 11:43:26AM +0200, Andy Shevchenko wrote:
> > > > > On Mon, Feb 03, 2025 at 02:19:01PM +0530, Anup Patel wrote:
> > > > > > From: Sunil V L <sunilvl@xxxxxxxxxxxxxxxx>
> > > > > > 
> > > > > > fwnode_get_reference_args() which is common for both DT and ACPI passes
> > > > > > a property name like #mbox-cells which needs to be fetched from the
> > > > > > reference node to determine the number of arguments needed for the
> > > > > > property. However, the ACPI version of this function doesn't support
> > > > > > this and simply ignores the parameter passed from the wrapper function.
> > > > > > Add support for dynamically finding number of arguments by reading the
> > > > > > nargs property value. Update the callers to pass extra parameter.
> > > > > 
> > > > > I don't like this (implementation).
> > > > 
> > > > Agree.
> > > > 
> > > > > It seems that we basically have two parameters which values are duplicating
> > > > > each other. This is error prone API and confusing in the cases when both are
> > > > > defined. If you want property, add a new API that takes const char *nargs
> > > > > and relies on the property be present.
> > > > 
> > > > Also this is not really needed for ACPI case because it has types so it can
> > > > distinguish references from integer. Having separate property for this just
> > > > makes things more complex than they need to be IMHO.
> > > 
> > > Thanks! Andy and Mika for your kind feedback. I agree that having both
> > > property name and nargs is confusing and also ACPI would not need
> > > nargs_prop. In fact, I think ACPI doesn't need even nargs integer value
> > > as well from the caller since all integers after the reference are
> > > counted as arguments.  However, the issue is acpi_get_ref_args() assumes
> > > that caller passes valid num_args. But typically the common
> > > fwnode_property_get_reference_args() doesn't usually pass both valid
> > > values. So, should fwnode_property_get_reference_args() pass both
> > > nargs_prop (for DT) and nargs (for ACPI). Or do you mean it is better to
> > > remove the check for num_args in the loop inside acpi_get_ref_args()
> > > function?
> > 
> > Can you show an example of a case you are trying to solve with this? So far
> > we have been able to go with the current implementation so why this is
> > needed now?
> 
> Basically one can call fwnode_property_get_reference_args()
> irrespective of DT/ACPI. The case we are trying is like below.
> 
> if (fwnode_property_get_reference_args(dev->fwnode, "mboxes",
> 					       "#mbox-cells", 0, index, &fwspec)) {
> ...
> }
> 
> As you can see this works for DT since OF interface handles
> "#mbox-cells". But since nargs is passed as 0, it won't work for ACPI
> due to the reason I mentioned earlier.
> 
> Mandating to pass both "#mbox-cell" and valid nargs count looks
> redundant to me.

Ah, interesting. The original change that introduces this 3e3119d3088f ("device
property: Introduce fwnode_property_get_reference_args") hadn't been reviewed
by Mika or me, that's probably why we are not familiar with.

Since interface is already established, I would recommend to fix
this as proposed, i.e. with a new API. This is the way to match
how OF seems to be doing.

-- 
With Best Regards,
Andy Shevchenko