Dear Kumar,

In message AB3960D1-288C-4A22-B9CF-9929C6B68CC7@kernel.crashing.org you wrote:

On Dec 4, 2008, at 10:58 AM, Kumar Gala wrote:

...

We have an issue with inbound windows overlapping outbound windows. This is fine, but it does mean that pci_hose_{bus_to_phys,phys_to_bus} can return the wrong thing.

I was wondering if there was any issue if PCI_REGION_MEMORY was set for a region if we skipped that region in the lookup functions:

pci_hose_phys_to_bus(...) { ... for (i = 0; i < hose->region_count; i++) { res = &hose->regions[i];

            if (((res->flags ^ flags) & PCI_REGION_TYPE) != 0)
                    continue;

if (reg->flags & PCI_REGION_MEMORY)
	continue;

...

any issues with this change? If not I'll work up a proper patch.

Wolfgang, any comments on this?

I'm not much of a PCI expert. If nobody else complains, feel free to move on.

Best regards,

Wolfgang Denk