> 
> has anybody tried putting a PCI-PCI bridge in an LCA-based
> (e.g. Multia or AXPpci) system?  Does it work?
> 
>
We have a fair amount of experience now with PCI-to-PCI bridging with the
AXPpci.  It has some problems.  For one thing, although the PCI spec implies
that the firmware is supposed to configure PCI-to-PCI bridges and the devices
behind them, in our experience it does not.  I don't know if DEC has fixed
this in firmware, but I don't think so.  We have code (written by Trevor
Mendez, who no longer works here) that will configure the bridges and
devices behind them.  I believe that we shared that code but that it has
not been incorporated into the distribution because it's really a work-around
for hardware/firmware brokenness.  Beyond this, this code is not very robust
or flexible and hasn't been tested except in our admittedly weird configuration.
Also, another thing you should know if you plan to use PCI-to-PCI bridging
if you intend to have more than one level of bridging is that this
configuration seems to be prone to live-lock problems when the AXPpci wants
to master a device far downstream at the same time as some far downstream
bus master device wants to master the AXPpci's memory.  It looks like the
AXPpci acquires the "top" bus, the mastering peripheral acquires the
"bottom" bus, neither can really get the "middle" bus, and eventually they
back off and retry, slamming into each other again, ad infinitum, until
the Alpha's 2^24 maximum bus retry timer goes off and you get a machine
check.  We're using older bridge chips right now (21050, I think?) and there
are newer chips (21150?) that may rectify this problem.  We're currently
waiting on new bridge chips, they're heavily back-ordered at this time.
-ben