Re: target termination

[Joe Cowan <jpc@hpmckee.fc.hp.com>]

Responding to Michael Tresidder:

If I understand your proposal correctly, you would like to be able to
save the normally wasted cycle required by FRAME deassertion protocol
when a target disconnects.  This would be accomplished by the target
asserting STOP on the next-to-the-last data phase instead of doing so on
the last data phase.

While this might be considered for future I/O bus protocols, I think
it's highly risky for PCI, and would provide almost no benefit for most
applications.  Saving one cycle per transaction when a target
disconnects doesn't seem very significant, particularly when such
transactions usually transfer a cacheline or more of data.  If the
approach could be applied to the case of target retries, the benefit
might be a little higher for environments with PCI-to-PCI bridges.
However, I don't think that's possible.

Joe Cowan

*----------------------------------------------------------------------*
|                              Joe Cowan                               |
|                  ESG System Design and Performance                   |
| Email: jpc@fc.hp.com                   Hewlett-Packard Company       |
| Work: (+1) 970 898-2404 | (T)898-2404  3404 East Harmony Road; MS F2 |
| Fax:  (+1) 970 898-3197 | (T)898-3197  Fort Collins, CO 80528 USA    |
*----------------------------------------------------------------------*