[parisc-linux] Re: pa-risc V0.9 on Model 720 (+ others)
Michael S.Zick
mszick@pflash.com
Tue, 5 Jun 2001 14:01:31 -0500
To All -
Reference my last on this subject - (attached)
Using xconfig as a model - on finds:
xconfig main menu
=09network device support
=09=09ethernet (10 or 100 mbit)
=09=09=09EISA, VLB, PCI and on-board controllers
Selecting that choice would lead one to believe that the proper code=20
for the "on-board controllers" would be generated.
Which should cover the built-in Ethernet on the older boxes (including
the standard, model 720) and when also combined with the proper selection=
of
card - a model 720 with an optional EISA BA having a Ethernet card plugge=
d
into it.
Perhaps similar configuration logic could be included for the built-in (5=
3c73xx)
SCSI controller - with the additional option of selecting an additional c=
ard
living in the optional EISA BA card slot.
Note: The prior owners of my machine had just such a configuration of on-=
board
and EISA bus SCSI controllers - running under HP-UX.
Mike
On Tue, 05 Jun 2001, Matthew Wilcox wrote:
> On Tue, Jun 05, 2001 at 11:36:37AM -0500, Michael S. Zick wrote:
> > Without either BA's defined - in/out Byte, Word, Long does not have
> > a prototype in mem.c (amoung other places) - the compiler guesses
> > (incorrectly) at what inb() and outb() mean.
>=20
> That's a known-bad configuration. You have to have some kind of PCI
> or EISA controller configured in.
Only if you intend to never support non-PCI, non-EISA older boxes.
Regardless of your intent - proper combinations of CONFIG_xxxxx
are supposed to be handled by the (x)config step of the build process.
After all - that is what "configure" means (in English).
To do a clean job of this - a selection of "Box Model" (similar to "Proce=
ssor Type")
may have to be added to the configuration scripts. With corresponding #i=
fdef/#ifndef
statements in the relavent code.
At the moment - I have tunnel vision - I am focused on getting a personal=
copy
of parisc-linux to boot on my model 720 box.
The comments and findings I post to the mailing lists may be freely disre=
garded
by anyone not interested them. Still, they will exist in the mail list a=
rchives so that
any developer working on a public version of this kernel can use them for=
hints, leads
and other ideas.
Mike
>=20
> --=20
> Revolutions do not require corporate support.