>From thorpej@lestat.nas.nasa.gov Tue Sep 17 22:08:32 1996
>
>Also, it's worth noting that those chips don't live directly on the 
>KFTIA.  They're actually on a PCI bus, which sits on some sort of 
>bridge.  I'm not certain, but that particular PCI instance may be one of 
>the (multiple) primary PCIs on the 8200; the KFTIA holds 2 SCSI busses, 
>and 2 Tulips (21140s, I think), and one "hose" (which, on the particular 
>8200 you're working on, connects to a PCI bus, which in turn bridges to 
>EISA).

I misspoke myself. They're on a logical PCI slot. And the KFTIA has
4 SCSI busses- 3 Differential Wide + 1 SE Narrow (which is what it
boots off of...).

>
> > Has a driver for this been ported to NetBSD, or should I plan on
> > possibly porting the Linux version (or doing my own, or....)?
>
>If it were me, I'd say "write my own".  I just looked at the Linux 
>driver.  Besides being GPL'd, it's ... a mess.  It might be reasonable 
>documentation on how the chip works, but beyond that...

I already actually know the chip a bit. I was just leaving Sun when
they were working with Emulex in designing it.


>
> > It's also possible I can ignore these ISP SCSI busses for the moment,
> > shove an Adaptec 2940 or an NCR script engine PCI card into a PCI slot
> > and boot off of that (yes, I probably won't get very far past that anyway-
> > but that's what I'm starting to work on...)
>
>If you're going to use an alternate SCSI, use an NCR53c810 SIOP.  The 
>"ahc" driver is _far_ from being safe on 64-bit systems, and probably has 
>alignment problems on the alpha, too.  Probably your best bet is to just 
>load the kernel from DUh and nfs_mountroot ... talking to one of those 
>PCI slots requires figuring out how the bridge between them works; it'll 
>likely be a lot easier to just figure out the first PCI and then talk to 
>the Tulip, first :-)

This is a good point- Net might be a better way to go- although I believe
that the bridge issues may not obtain for the KFTIA- that's more of an issue
for the KFTHA (the full hose connections...)

>
>Gee, Matt ... you ought to pop by my office sometime (I'm in 222-1), and 
>I'll give you all the code I wrote for the 8200 O(1 year) ago... :-)


Ooops. Yer right. Silly me. I'll be down there tomorrow afternoon
and I'll drop on by....

-matt

p.s.: we should also talk about what's wrong with GPL and what 'mess'
means- the current linux qlogicisp.c is actually fairly clean for a
SCSI host adapter driver.