On 30/09/15 16:39, Robin Murphy wrote:

Hi Marc,

On 22/09/15 18:52, Marc Zyngier wrote:

quoted

of_parse_phandle_with_args() is slightly inflexible as it doesn't
allow the (unusual) case where the #*-cells property is not defined.
In order to support this, introduce of_parse_phandle_with_opt_args()
which assumes that #*-cells is zero when it is not defined,
as required by the msi-parent binding

This is done by turning __of_parse_phandle_with_args into an even
bigger monster, which is a bit frightening.

A monster indeed; I can't quite figure out the exact effect this change 
has on of_count_phandle_with_args(), but I have a lingering doubt it may 
be something undesirable, since AFAICS that's now going to proceed from 
where it would have errored out before, with a count of -2.

I think it might be nicer to implement this by passing an extra "assume 
zero if #cells not found" boolean to __of_parse_phandle_with_args().

Alternatively, what's the actual likelihood of legacy bindings being 
mixed in with new ones? Could we not simply mandate that anyone adding 
an MSI controller with #msi-cells to a DT must ensure any existing nodes 
are also updated with #msi-cells = 0, and keep the legacy workaround 
self-contained in the MSI layer? e.g. paraphrasing from patch 2/2:

msi_np = of_parse_phandle(np, "msi-parent", 0);
if (!of_property_read_bool(msi_np, "#msi-cells"))
	return parse_this_thing(...);
else
	while (!of_parse_phandle_with_opt_args(np, "msi-parent"...
		if (parse_this_thing(...))
			return;

Having tried this, it doesn't look too bad. Specially turned into some
kind of library function. that can be shared between platform and PCI.

So I'll drop this patch altogether and repost an updated series.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...