In article <445CF17B.5020905@netbsd.org>
shige@NetBSD.org wrote:

> Jason Thorpe wrote:
> > There is no need to have the extra layer of indirection.  You can
> > determine which plb devices are present by simply checking the cpu type
> > in the plb driver.
> 
> The opinion of `key' members of evbppc port developers
> is that there is no problem in the current ibm4xx device structure.

Well, I guess Jason just said not to attach plb at cpu.
Anyway, we should remember that Jason is an important key developer
of whole NetBSD ;-)

> I understood it. I will forget a pseudo mainbus. ;-)

I have a feeling that introducing a pseudo mainbus and attaching
cpus and local busses at the mainbus looks a bit more consistent
than current situation:
- EV64260 (with 750) has mainbus for root, cpu* at mainbus0 and
  gt0 (system controller) at mainbus0.
- EXPLORA451 (with 403GCX, no internal peripheral?) has elb
  (explora local bus?) for root, cpu0 at elb? and com0 at elb? etc.
- WALNUT (with 405GP?) has plb as root, cpu0 at plb? and opb* at plb0 etc.
But I know I'm not a key person of this area, just my two yen ;-)

> But I don't agree that we will add new declarations (such as (*1)) to
> plb and opb driver and update plb.c and opb.c when supporting for new
> ibm4xx cpu.

IMO, it's better to summarize how many possible 4xx CPU variants
should be handled in that layer, and then design structure for
each of them before implementing it. At least there is another
port for PPC405D5Xn core:
http://mail-index.netbsd.org/port-powerpc/2006/04/26/0000.html
---
Izumi Tsutsui