> After going through the specification for bus_space_barrier()
   > for the fifth time, and finding it lacking, I'd like to propose
   > the following enhancement:
   
   I don't particularly object to this, but given the fact (noted again
   on the mailing lists recently) that all the bus_space_{read,write}
   operations have implicit barriers to deal with non-barrier-aware
   drivers, this seems like misplaced effort...


but surely this is a sitution we'd like to change.  you meantioned
a way we canto do this on a per-driver basis...and cleaning up the
bus_space_barrier() API is part of making this possible...


.mrg.