Host bridge latency on ALPHA Server

[fmj@pe.dk (Finn Martin Johansen)]

I've designed a PCI add-in card that runs under NT on Power PC and Pentium
platforms.
The card is a master on the PCI bus during DMA reads.
We've experienced that the DMA accesses are quite different on a Digital
ALPHA Server 2000 5/250 (two processors) than on Power PC and Pentium
platforms. We know off course that it has a Digital host bridge controller
but can it be true that when we have initiated a master access (FRAME#  and
IRDY# active) it takes approx. 1us before the target respons with a TRDY#
(This is a PCI spec. violation!!!!!!!!!!). But what is the reason for this???
The hardwired value MIN_GNT in PCI config is set to 8, i.e. 2us, but the
system arbiter initialize the latency register with a value of 20h, thus
with the abovementioned host bridge latency we only get to make to data
transfers in a burst. 
We don't see the desired burst length (16 transfers setup by our DMA state
machine) on a DMA read eventhough CBE# has the value of MRM (I can't
remember the bit value).
We the tried to reconfigure the latency register with the NT driver. By
increasing the value to 40h we experienced that the data transfers in a
burst increased to seven ,and then did a bus master arbitration for new REQ#
and GNT# to reach eight.
Has anyone seen the same kind of things on ALPHA Server systems???
Please give me anything to go on from here. Now I'm in a kind of deadlock!!!!
If you need more specifics on the problem please e-mail me

 /Finn M. Johansen
*******************************
Finn M. Johansen, M.Sc.E.E.
Hardware Design Engineer, R&D

PURUP PREPRESS A/S
5 Soenderskovvej
DK-8520 Lystrup
Denmark
Phone: +45 87 434 343
Fax: +45 87 434 445
E-mail: fmj@pe.dk                           
******************************
�
��