rvb@sicily.odyssey.cs.cmu.edu wrote:
> Point of clarification: The reason we round to BIOS partitions is to
> live compatibly with other OS's on the disk.

A fair bit of code out there assumes that BIOS partitions end at the
end of a cylinder.  Some of it is even in the NetBSD kernel;
arch/i386/i386/disksubr.c resets part of the geometry in the fake
disklabel if it finds a NetBSD partition, with the assumption that it
ends at the end of a cylinder.  (Granted, I'd like to see this code
excised, since it just causes trouble, but the i386 portmaster(s)
didn't comment on my PR.)