Re: Two Arbiters in a bus

[Dave New <den@aisinc.com>]

Young-Moo Lee wrote:
> 
> Hi there,
> 
> Glad to write to you.
> I am Young Lee who has an experience of developing PCI bus target device.
> Now I am going to develop a board which has several pci masters.
> 

Greetings Mr. Lee,

I believe that from this you are putting together a board that has a local
PCI bus, with your PCI masters (more correctly called initiators) connected
to that bus.

> Especially
> I want to connect both pentium-pro(i686) and intel's i960rp to the same
> pci bus. In this case, there are two bus arbiter in a bus. I can't find
> what happens in the case of more than one arbiters in a pci bus.
> 
I believe that from this you are using the i960RP as an intelligent 'bridge'
between the local PCI bus on your board, and the motherboard PCI bus, which
contains the P6.  If the P6 is on the local PCI bus, this complicates
the problem considerably (see below).

I believe that you cannot have two arbiters on a single PCI bus.

On the other hand, the i960RP can be told to not arbitrate both of the busses
it is connected to (or so I have been told by our hardware engineering folks).

So, it should only arbitrate the local PCI bus, and slave to the motherboard
arbiter.  This does not mean that the i960 cannot initiate transactions on the
motherboard (P6) PCI bus.  It just means that it must request arbitration
from the motherboard PCI bus, just like any other PCI bus initiator.

One additional wrinkle that I've been made aware of, is that you must run the local
PCI bus clock synchronous to the motherboard clock.  Although the i960RP was
designed to be able to handle asynchronous clocks on its primary and
secondary PCI bus interfaces, there is an Intel errata that states that this
doesn't work.  Further, I've been told that Intel will not fix this problem,
which means that using the i960RP could be non-compliant for certain cases.

For instance, if you wish to place a processor on the local PCI bus (a Pentium,
for instance), the clock requirements of that processor (and the fact that
the i960RP will need to drive the local bus clock synchronous to the motherboard
PCI bus clock) will generally not allow you to support 'green' systems or so-called
'spread-spectrum' motherboard clocks.  Most processors I've looked at require
that *they* drive the local bus clock, using a companion clock generator.

> Could you possibly tell me about this case?
> Thank you in advance.
> 
> Young Lee
> ymlee@vlsi.kaist.ac.kr

--

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
-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
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
ÉìÜ