In article <4415706A.6030103@netbsd.org>
shige@NetBSD.org wrote:

> >> - options IBM405GP (or something) is needed to specify ibm405gp.c
> >>   (or should it be renamed to obppci.c??)
> > PCI doesn't need to be 405gp specific.  It attaches to the PLB
> > (processor local bus), so plbpci.c would be better.
> 
> ibm4xx/pci/pchb.c + ibm4xx/dev/ibm405gp.c = plbpci.c ?

Maybe.

> And...
> Two functions in ibm4xx/pci/pci_machdep.c
> (pci_intr_map()/pci_conf_interrupt()) are machine-dependent.
> We should move them to evbppc/<board>/pci_machdep.c.
> I think that other functions to powerpc/dev/plbpci.c. Is it ok?

Hmm, it's okay to put these functions into MD files,
but I'm not sure if pci_machdep.c is appropriate or not.
Is there any other board-specific file like evbmips?

> OpenBlockS266 port use files.ibm405gp.
> but it is easy only to change `include' declaration
> in kernel config. Is it ok?

At least rtc.c will cause some conflicts (inittodr(9) etc.)
on walnut unless evbppc/walnut/dev/todclock.c is fixed.
Anyway rtc.c has no "device" specific function, so it should be
moved to evbppc/evbppc/todclock.c (or something) for now, IMO.
(eventually these todr(9) functions should be in sys/dev/clock_subr.c)

> There's still 405GP-specific part (opb_tag declaration) in
> sys/arch/powerpc/ibm4xx/dev/opb.c.
> This should be fixed.

Yeah, but for now you could proceed gpio stuff as interim fix ;-)
---
Izumi Tsutsui