Testing an Open Firmware interrupt-map patch

[Andrew Turner]

On Tue, 15 Dec 2015 00:47:10 +0100
Marius Strobl <marius at alchemy.franken.de> wrote:

> On Mon, Dec 14, 2015 at 01:46:25PM +0000, Andrew Turner wrote:
> > Hello,
> > 
> > I'm looking for testers for a patch to update how we parse the
> > interrupt-map property to follow the ePAPR spec. This property is
> > commonly used with PCIe controllers.
> > 
> > The current code doesn't take the parent address size property. When
> > this is non-zero we need to also read these values. This is needed
> > on an arm64 board I have as the interrupt parent has memory-mapped
> > children so needs this to be set.
> > 
> > I have a patch at [1] to add this, however would like it if this
> > could be tested on other platforms using this code to check it
> > doesn't break these platforms before I commit it.  
> 
> +		    "#address-cells", &paddrsz, sizeof(paddrsz)) ==
> -1)
> +			paddrsz = 0;	/* default */
> 
> According to IEEE 1275-1994 (page 110) the default in case of a
> "#address-cells" property is 2 (see also ofw_bus_setup_iinfo()),
> with the latter also being known as the correct default value
> for sparc64 machines.

However both Linux and NetBSD have similar code, both assume a missing
interrupt-parent #address-size property the correct value is 0.

> In any case, this will have quite an impact on the offset
> calculation on sparc64 machines for e. g. EBusses beneath PCI
> busses (which use 3 address cells there) and, thus, most likely
> will cause havoc. Maybe the parent unit address should be only
> taken into account on platforms using interrupt parents?

It would seem there is a difference between the encoding on Sparc64 and
ePAPR. Using the the parent #address-size may lead to incorrect parsing
on Sparc64, as such I will disable it there.

Andrew