On Tue, Feb 23, 2010 at 06:51:56PM -0600, David Young wrote: > On Wed, Feb 24, 2010 at 01:19:41AM +0100, Joerg Sonnenberger wrote: > > Hi, > > based on the PAE discussion there is the question of what to do for > > drives like twa(4). E.g. drivers that can use a more compact 32bit > > layout or support the full 64bit PCI address space. > > > > Would it be reasonable to add a new MI variable > > bool pci_use_64bit_addressing; > > that gets set by MD to true if drivers should use 64bit support? > > For x86, it would be enabled if physical memory above 4GB is found. > > Other architectures might have similiar rules or have a IOMMU that makes > > it irrelevant. The other reason for avoiding 64bit is the increased > > bus traffic after all. > > This seems like a property that the MD mainbus_attach() routine should > set on the pcibus_attach_args before it attaches the host bridge(s). It > may suffice simply to initialize pba_dmat64 to non-NULL.
...and the child checks for pa_dmat64 != NULL? Joerg
