Steve, I will send you a quickly-understandable model of system I am implementing in a few days so you can see where this is happening and how changing the order of declaration makes my system work perfectly properly.
---- Original message ---- >Date: Mon, 27 Oct 2008 07:36:26 -0700 >From: "Steve Reinhardt" <[EMAIL PROTECTED]> >Subject: Re: [m5-users] Non-Deterministic Bus Port Id Assignments >To: "M5 users mailing list" <[email protected]> > >Bus port IDs are assigned as the connections to the bus are processed, >which happens in a fairly arbitrary order... so yes, I can believe >that seemingly minor changes to your configuration would affect the >bus ID assignments. > >I don't see how this could cause a problem though... your whole >discussion of src and dest fields being the same is confusing to me. >If the src and dest fields are the same then a device is sending a >message to itself. This only happens in one scenario that I know of, >which is when a cache broadcasts a request up the hierarchy to force >an invalidation (e.g., an L2 broadcasts a request from the L2-L3 bus >up to the L1-L2 bus)... since it is a request, the lower level cache >is the implicit destination, even though it's also the source. I >think there's some explicit code in the bus that deals with this. > >Other than that, I don't believe you should ever see packets where src >== dest (hence the assertion). If you're talking about IDs across >different buses, then that's a bug; the src & dest fields on a packet >should be using IDs that are relative to the bus the packet is being >sent on. If a packet is transferred from one bus to another then the >src field should get rewritten to reflect the new sender on the second >bus. (Note that this happens implicitly in the bus port code.) > >Steve > >On Sun, Oct 26, 2008 at 10:46 PM, Shoaib Akram <[EMAIL PROTECTED]> wrote: >> when creating a complex interconnection network where request and response >> paths are very different, sometimes there is a problem of desta and src have >> same port id, and thus the failure of dest!=src assertion. However, if the >> order of declaration of buses is changed, the port ids somehow change, and >> the assertion failure no more and thus messages are received by nodes who >> earlier sent them, because src and dest are different. Now, one should be >> careful about final destination port being same as beginning (src) port, but >> hte order of declaration changing the bus port-ids look suspicious? >> _______________________________________________ >> m5-users mailing list >> [email protected] >> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users >> >_______________________________________________ >m5-users mailing list >[email protected] >http://m5sim.org/cgi-bin/mailman/listinfo/m5-users _______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
