On Fri, 10 Mar 2000, Bill Studenmund wrote:

> True. But the main thing is that we can quite readily support nubus and
> pci powermacs from the same kernel config. :-) We don't need two NetBSD
> architectures to do it.

:-)


> If "bart" or whatever we call it attaches, we are a nubus machine. If
> bandit attaches, we're a non-G3 generation powermac. If grackle or
> uninorth attatch, we're some sort of G3 or G4 generation. :-) 
> 
> > > (AMIC or HMC would be other possible names). Then just have the bart bus
> > 
> > AMIC is the DMA engine, which should also be at mainbus?
> 
> Maybe. We don't have a specific DMA driver, though that's probably because
> everything has done dbdma until now. Probably the DMA controller should
> attach either to the hba (host-bus addapter, bandit, grackle, uninorth, or
> "bart"), or to the bus (pci or nubus). Not really sure. :-)

Hmm.  Well, the AMIC DMA tends to behave pretty differently from dbdma,
and isn't consistent from device to device.  Even with AMIC info for SCSI
and serial, it still took several months to figure out what was wrong with
AMIC for floppy support.  In the case of AMIC, I hate to say it, but I
think the OSF people probably did the most efficient thing possible when
they wrote separate sets of routines for each device.  Nasty....

However, it might be logical to have some sort of DMA type variable, and
in the case of AMIC, use a different set of routines.  It wouldn't really
be "attaching" to anything, but one could call it that....  ;-)


David