Re: PCI - x86 coprocessor board

[Dave New <den@aisinc.com>]

Eric Rehm wrote:
> 
> I'm looking for a x86 (486, Pentium, or Pentium Pro) busmastering PCI coprocessor board.
> 
> I'd like to plug it in to a PCI slot a x86 or Alpha Windows NT system.
> The coprocessor board should be able to boot MS-DOS independent of the
> Windows NT system.  (A second keyboard and monitor is OK for
> development purposes.)  The coprocessor board should be able to:
> 
> 1) be a target of PCI transactions (memory or I/O space) from
> the NT host, AND
> 
> 2) be capable of acting as a PCI bus master to DMA data from
> coprocessor board memory to the NT host memory.
> 
> Suggestions?
> 
> Thanks,
> 
> /eric rehm
> Equator Technologies, Inc.
> Seattle, WA
> eric@equator.com
> (206) 328-6544 x213

Hi Eric,

When you find one, let us all know <grin>.

You would think that this was a simple request, but so far, all the
available off-the-shelf x86 host PCI chipsets won't support this.

In particular, they most all insist on mastering the PCI bus,
and mapping the conflicting memory translations is a mess, to say
nothing of how to handle the configuration spaces that each
processor would like to see, on the primary and secondary bus.  If you
treat the interfacing chipset as a bridge, who's driving the
configuration cycles when?  And who owns/drives the primary
and secondary bus clocks?

Add to that the idea that the embedded Pentium won't tolerate having
its clock slaved to the main PCI bus clock, and you have a non-PCI
compliant clock problem (no spread-sprectrum, here!)

The most hopeful thing I've come across is a V3 part (V96DPC) that puts
two of their bridges back-to-back in a single package.  They
internally hook the 'local' i960 buses together, and allow each primary
side (which is facing 'out') to be 'driven' by the PCI bus they
are facing.  Then the embedded Pentium would still need a host-to-
PCI chipset to drive the 'local' PCI bus, etc.

What a mess.  Is there anyone out there trying to put together
a host-to-PCI chipset that will truly allow slaving the processor?
Is it doable, considering Pentium's clock requirements?

And what about the software?  Anyone doing an I2O driver layer for
NT?  To talk to something besides the non-clock-compliant i960RP?

-- 

Cheers,

DaveN

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Dave New, den@aisinc.com    | Machine vision?
Applied Intelligent Systems |      I'm glad *they* can see the future...
3980 Ranchero Drive         | 
Ann Arbor, MI  48108        |        Opinions expressed are mine.    | PGP 2.6
(313) 332-7010              | 08 12 9F AF 5B 3E B2 9B  6F DC 66 5A 41 0B AB 29
(313) 332-7077 FAX
´äÓ