> 
> > > For systems that have two SCSI busses (e.g., Power Macintosh 8100s), 
> > > the two SCSI ID maps are logically merged into one. Devices sharing 
> > > the same ID, but sitting on separate busses, are not supported. Only 
> > > the device on the external bus will be accessible in such a case. 
> 
> As David implies, this is solving a naming problem.  It unfortunately
> doesn't solve the "how to access the bus" problem which is what "we're"
> up against.

Right. The question is where is that second SCSI controller in IO space.

> BSD grabs SCSI drives in order and starts assigning them ids.  First one
> bus (internal), then another bus (2nd bus/Dock/nubus/etc.).  You are limited
> by the number of sd devices that you have made in /dev and by the number
> representable in the device node...  If you're really curious, I'll dig
> it up for you.  Before you start running up against that barrier, I
> suspect that you'll be wanting to glob some of the disks together into
> ccds, anyway...  ;-)

With 8 partitions per disk, and minor device #'s being 8 bits, we
can have 32 disks (that is 32 sd's; st's, ss's, and cd's don't count).

Supporting multiple, sparse-ly filled SCSI busses is one reason for
dynamically numbering disks.

Take care,

Bill