[OpenBIOS] Back at it again( PCI Passthrough )

[Segher Boessenkool]

On Fri, May 11, 2018 at 09:11:40AM +0100, Mark Cave-Ayland wrote:
> On 11/05/18 08:11, Jd Lyons via OpenBIOS wrote:
> 
> >Seems I could add config-1@ as a colon definition, however I’m not sure 
> >how too deal with r1@?
> >
> >: rl at -le  rl@ lbflip ;
> >: >config  f1000000 + ;
> >: config-l@  >config cr ." config-l@ " dup . rl at -le space dup . ;
> >
> >Just can’t figure how r1@ is implemented in SLOF?

rl@, not r1 at .  It is defined in hvcall.code or nativeio.code, depending
on what platform you are building for.  It does a "raw" cache-inhibited
32-bit read.

> That's definitely the long way around.

> My reading of the IEEE-1275 PCI bindings is that you take the first 
> 32-bit word of "reg" for the PCI configuration space address, so as you 
> can see we return the device-id/vendor-id for the NE2000 NIC as above.

But the AGP and PCIe buses on a powermac are on separate PCI domains
from the other PCI buses, and the PHBs have a different programming
interface, too.

I don't know how things work in QEMU, but if it emulates the hardware
at all, you will have to find the proper config-* to use via walking
parents in the device tree, etc., not shortcut to one implementation
for all buses.


Segher