[rescue] Time to replace the IPX
paul at doorslam.net
Sat Dec 20 18:08:24 CST 2003
Jochen Kunz [jkunz at unixag-kl.fh-kl.de] wrote:
> On 2003.11.26 18:04 Paul Weissmann wrote:
> > If so, I think I still have some Phobos EISA card for HP 9000s I
> > could try.
> Phobos EISA card? Tel me more!
It was a card specifically sold for HP-UX systems, complete with a
HP-UX 10 driver on a DAT.
I think it is a different card than those rebadgeded 3Com from
Phobos, but I'm not sure.
It has an Intel chip on it, can't remember it now, would have to look
again, it's Fast-Ethernet for sure.
> > You misunderstood me - systems with a CPU newer than PA-8200 have a
> > very different I/O-subsystem than the older ones.
> Sure. The IO is (mostly) PCI based?
> It is much simpler writing a new bus attachment with the needed
> bus_space(9) / bus_dma(9) glue then adding a whole new architecture. A
> new architecture requires a lot of nasty low level work including
> assembler, interrupt routing, pmap(9) interface to the VM system...
It is a completely new architecture, I don't think these class of
machines (B2000) have much in common with earlier 32-bit (capable)
> > The B2000 in turn is based on the Astro system controller together
> > with Elroy PCI bridges and has zero GSC-devices left.
> That's no problem. If the IO is build with generic PCI devices all you
> need is a bus attachment and everything works automagicly.
They have totally different memory and bus-controllers.
They use so called "I/O-Ropes" from the Astro-system controller to
which then the PCI-bridges attach. On these bridges you then find
generic PCI-devices. HP used again a lot of custom ASICs to implement
some parts of the I/O-subsystem (IDE, USB, RS232, ...), but these
parts are supposedly kind of buggy, esp. on DMA transfers.
Also, the PDC routines are supposedly a lot different on PA2.0
systems than on 1.1 ones.
All IIRC, I did not dig that much into PA2.0 for now.
zwei jaeger treffen sich. beide tot.
More information about the rescue