[openib-general] Re: [PATCH] mad: add GID/class checking for matching received to sent MADs

[Jack Morgenstein]

On Wednesday 01 March 2006 19:21, Sean Hefty wrote:
> At this point, all we know is that the send and original request are from
> the same port.  We still don't know what GID was used in the send.  Don't
> we still need to check that?
>
We also know that the original TID has been assigned by this host -- in fact, 
by a specific process/agent. (the upper 32 bits are unique per assigned agent 
-- allocated sequentially per call to ib_register_mad_agent().

The application has the responsibility for assuring that the lower 32 bits are 
unique per transaction -- if the app does not do so, only its own MADs will 
suffer.

Thus, if we know that this host is the originator of the transaction, we 
should assume that the sending application has been debugged, and don't need 
to cover for sender/requester bugs -- the sending app should guarantee that 
the lower 32 bits are also unique on that agent.

In that case, TID alone is sufficient.

Jack