On Oct 21, 2008, at 9:46 AM, Wolfgang Denk wrote:
Dear Kumar Gala,
In message <1224598531-2698-1-git-send-email- galak@kernel.crashing.org> you wrote:
PCI bus is inherently 64-bit. We should treat all PCI related bus addresses as 64-bit quanities. This allows us to have the ability to support devices or memory on the PCI bus above the 32-bit boundary.
I don't think this is a good idea. There are pure 32 bit systems out there which will never use more than 32 bit for the PCI resources, so why load them with the additional memory size and execution time?
Should we not enable this only for such systems that actually need it?
We could, however it seems the impact is pretty small on those systems with pure 32-bit PCI. I think the different in testing etc may not be worth it:
text data bss dec hex filename 350588 44964 41909 437461 6acd5 u-boot (32-bit pci structs) 351100 44964 42349 438413 6b08d u-boot (64-bit pci structs) ---------------------------------------------- 512 0 440 952 3b8
If this is desired to be configurable should I introduce a pci_addr_t/ pci_size_t? Coupling to phys_addr_t/phys_size_t doesn't make sense to me because its perfectly reasonable to have a 64-bit PCI device work in a 32-bit system.
- k