> Yea, Ali's got it right... the way to think about address-based > routing on the bus is that the packet logically does get broadcast to > determine the destination; all our mucking with address ranges is > really just some combination of optimization and sanity check. I buy that, but I'd argue that we shouldn't use the term Broadcast shouldn't be set by the sender since the sender shouldn't care unless the sender really wants a broadcast. Can I create a new term Automatic and make everyone use it (i.e. s/Broadcast/Automatic/)? I'll then re-add broadcast to really mean broadcast in that the message, regardless of destination or whatever is really destined for all targets.
> As far as the setSrc() calls in memtest, I vaguely recall that there > were problems with cache responses failing because there was no valid > source ID on the request packet. Basically there's a hole in our port > interoperability story (where you're not supposed to care what's on > the other side of a port) in that a packet that comes straight into a > cache with no intervening bus might not have a valid source ID. I > don't remember why memtest needs (needed?) this but normal CPUs don't. It seems that the src should be set by the peer of the sender, not the sender, so I don't see why it would matter. I'll dig deeper. My goal is to improve things for non Bus interconnects. Nate _______________________________________________ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
