[rescue] USB port card for sBus?

[Jochen Kunz]

On Sun, 24 Aug 2008 21:23:43 -0400
Joshua Boyd <jdboyd at jdboyd.net> wrote:

> Actually, there is a PCI->wishbone bridge already, so if I
> understand correctly, only a SBUS->wishbone bridge would need to be
> written.
Wishbone?
[google]
Wishbone = system on a chip bus.
But then you have to build a SBus to Wishbone bus bridge.

Maybe worthwhile as it would ease the attachment of other PCI stuff,
e.g. FireWire, SATA, ...

> Actually, if the OS of choice is NetBSD (or OpenBSD, or FreeBSD) and
> he used the SBUS->FPGA->PCI route, wouldn't he only need a SBUS->PCI
> driver for the FPGA?
Yes. But in practice this can be complicated. As der Mouse already
noted there are endianes issues. There are also alignment issues. Some
silly PCI devices need 8 or 16 bit read / write access for proper
function... (Look at the sparse and dense register mapping areas in the
PCI host bridges of Alphas...) There is also an IOMMU in the SBus host
bridge, so you need proper pre- and post DMA operations in all PCI
device drivers to setup and tear down DMA mappings. IIRC a SBus card
has only a small memory window into the CPU address space. ...

No real problems, but many litle things to observe that can fail in
subtile ways. I got this knowledge the hard way when I worked on the
HP9000 GSC to PCI bus bridge bus_dma(9) code in NetBSD. GSC and PCI are
quite diferent in concept. PCI on the old GSC based HPPA machines is an
alien. An alien like a SBus to PCI bridge. ;-)
--


tsch|_,
       Jochen

Homepage: http://www.unixag-kl.fh-kl.de/~jkunz/