Re: [PATCH] ACPI / GPIO: Pass index to acpi_get_gpiod_by_index() when using properties

[Rafael J. Wysocki]

On Tuesday, November 04, 2014 03:42:38 PM Darren Hart wrote:
> 
> On 11/4/14 14:54, Rafael J. Wysocki wrote:
> 
> > From: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> > Subject: ACPI / property: Drop size_prop from acpi_dev_get_property_reference()
> > 
> > The size_prop argument of the recently added function
> > acpi_dev_get_property_reference() is not used by the only current
> > caller of that function and is very unlikely to be used at any time
> > going forward.
> > 
> > Namely, for a property whose value is a list of items each containing
> > a references to a device object possibly accompanied by some integers,
> > the number of items in the list can always be computed as the number
> > of elements of type ACPI_TYPE_LOCAL_REFERENCE in the property package.
> > Thus it should never be necessary to provide an additional "cells"
> > property with a value equal to the number of items in that list.
> 
> In this case, do we never expect a property to contain more than one
> ACPI_TYPE_LOCAL_REFERENCE?
> 
> Package () { "foobar",
>               Package () {
>                   "PCI0.FOO", "PCI0.BAR", 0, 1, 0,
>                   "PCI0.FOO", "PCI0.BAR2", 0, 1, 1
>               }
> }
> 
> This seems like it could be useful for connecting various types of
> devices together, but I confess not to have a specific exmaple in mind.
> It does concern me to limit the data format in this way.

We don't support this even with size_prop, so it doesn't seem to be relevant here.

Now, if we were to support this, I'd rather not use acpi_dev_get_property_reference()
for that, but add a new function specifically for it.  Moreover, I would extend the
format definition then so that we could do

Package () {
	"foobar", Package () {
			Package () {"PCI0.FOO", "PCI0.BAR", 0, 1, 0},
			Package () {"PCI0.FOO", "PCI0.BAR2", 0, 1, 1}
	}
}

in which case adding a special "size" property could be avoided.

That said, I have no idea why it might be necessary.  One reference in a property
value means that we're connecting the current node (the owner of the _DSD
containing that property) with some other node in the namespace.  Two references
in there would mean that the current node is to be connected with *two* other
nodes in the namespace at the same time.  That raises some questions that I'd
rather not consider in detail here, unless you insist. ;-)

> I suppose should such a case become necessary, we can deal with the
> issue then - and still avoid having a potential abuse point in the API
> from the start.

What we have today is sufficient for all of the cases we've considered so far.
If we find a case where it is not sufficient, we'll need to consider extending
the data format as well as the API.

Rafael

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/