Re: bidirectional REQ#/GNT#

[Andy Ingraham <ingraham@wrksys.ENET.dec.com>]

>>Does anyone know why the PCI 2.1 specification defines the REQ# and
>>GNT# signals as bidirectional tri-state (section 2.2.4, page 11)?  Is
>>this just an artifact of the signal type definitions in section 2.1 on
>>page 8 or is there an obvious reason that I am missing?  Since these
>>signals are defined as point-to-point, there is no obvious reason for
>>any device other than a central resource to ever drive GNT#.

I think the answer was that, although REQ# and GNT# aren't
bi-directional, they must be tri-stateable to handle Reset.  This
means they don't use "ordinary" output drivers.

You are correct that GNT# is always an input and REQ# an output,
except at the central arbiter where it's reversed.

The issue was raised back when rev. 2.1 of the spec was being
prepared, but evidently the spec writers didn't feel the need to
change it, or were just stubborn.  The "PCI System Design Guide",
another document from the PCI SIG (don't know if it's still available)
did define REQ# and GNT# as "out" and "in" respectively.

Regards,
Andy Ingraham
��