kiyohara@kk.iij4u.or.jp wrote:

> >   NE2000 variants in a config file.
> >   Is there no way to probe them properly?
> >   How FreeBSD/pc98 detects them?
> 
> I don't know and there might not be means. 
> 
>   http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/pc98/conf/GENERIC.hints?rev=1.26&content-type=text/x-cvsweb-markup
> 
>   e.g.
>     hint.ed.5.flags="0x600000"
>     hint.ed.1.flags="0x200000"
> 
> IMHO can't support on ne(4) currently, because some boards do the map
> distorted. It is necessary the handlers as each registers.  (like wdc(4))

Are these variants really NE2000 compatible? Or just using dp8390?
If they are not real NE2000, we should have different attachment
like sys/arch/mac68k/nubus/if_ae_nubus.c, so that you can handle
weird mappings by sc_reg_map[] in struct dp8390_softc.

Anyway, I'm afraid using flags in MD attachments might cause conflicts
against MI ne(4) driver.

> > - isn't it better to attach legacy devices (dcom, pic, wdc etc.)
> >   at cbus (which is a dumb bus anyway) rather than mainbus?
> >   I have a handmade two port dcom cbus board..
> 
> dcom_mainbus use the system-port.  However dcom at cbus not use system-
> port (PORT C). In addition, the mainbus of PC-98x1 does mapped a device
> different in the address of the even number and the odd number.  However,
> the map of cbus device is done perhaps linear. (not support yet. ;-)

Ah, I just remember the system-port of pc98.
I admit it's okay to have different attachment from cbus for such
devices, but how about having some pseudo bus like intio rather than
using mainbus directly?
---
Izumi Tsutsui