On Wed, May 26, 2010 at 12:47 AM, Andreas Färber andreas.faerber@web.de wrote:
Hey,
I'm still a bit puzzled about the cell vs. int concept here... Some of the patches seem to account for cell being 64-bit on sparc64. On ppc(64) however ucell is 32-bit, so wouldn't we need to modify at least our new ofmem_arch_get_translation_entry_size and ofmem_arch_create_translation_entry as well to account for 64-bit physical and virtual addresses?
Yes, those must be changed as well. Current implementation is OK at the moment though since hardcoded encode method yields the same result as address-cells-aware one. F.e. on sparc64 we will encode 6 u32 values instead of 3 u64 we do now. In my wish list: - ofmem_common: switch to 32bit properties - pci_set_AAPL_address: same