> It's not the pulldown that usually gets you in a mixed 5V/3V PCI design. It's the
> ringback associated with the pullup, where there are no clamps to attenuate
> the overshoot.
That's certainly true if your pullups comply with the 5v drive curve;
the PCI pullup spec for 5v has a *minimum* short-circuit Ioh of 44mA!
Pumping that much current out of a 3.3v pullup without 3.3v clamping
is fer-sher begging for disaster. OTOH, staying within the 12-24mA
range for 3.3v (no, PCI doesn't have a max but 2:1 is achievable and
weak is better) keeps ringback within reason.
We've had a lot of experience here with 5v-tolerant PCI design and
have found that drivers that just barely meet the minimum pullup spec
for 3.3v work beautifully in every application we've seen. Basically,
Intel snarched it when they blew off the tolerant condition with a
one-liner about using 5v drive levels.
-- D. C. Sessions email@example.com