Thanks Steve! The hints help me through the source code a lot!
Jiayuan
> On Dec 2, 2007 8:24 AM, Jiayuan Meng <[EMAIL PROTECTED]> wrote:
>>
>>
>> Hi all,
>>
>> We have a mesh interconnection model and I'm thinking about refine it and
>> integrate it into M5. But I got several questions:
>>
>> 1. it seems the bus is tightly coupled with address mapping
>> (findPort(addr)). I don't understand why and how to implement the same
>> functionality into the mesh. I'm planning to just identify ports with their
>> IDs (plus a reserved ID for broadcasting). Is this okay?
>
> For a given memory address, you need to know where the "home node" is
> (i.e. the memory module where that address resides, that normally
> responds to requests for that location). In a real bus this is done
> by broadcasting the request to every device and seeing which one
> accepts it. Our bus model uses the info from getDeviceAddressRanges()
> to avoid that unnecessary broadcast.
>
>>
>> 2. I'm not sure what the snoop ports are for. In the splash2 SE script, the
>> bus doesn't seem to have snoop ports.
>
> I'm confused... are you talking about the snoopPorts list in the Bus
> object? Not sure how that relates to the splash2 SE script. Anyway,
> it's just a list of normal ports that have indicated they want to
> snoop on coherence traffic (i.e., caches). Again, a real bus would
> broadcast every transaction to every device, but there's no point in
> sending coherence traffic to devices that don't care about snooping.
>
>>
>> 3. I assume the functionality of the bus/interconnection is to forward
>> packets from src to dst while modeling some sort of contention. Is there any
>> other roles this module plays? Does the bus in M5 carry any responsibilities
>> in coherence?
>
> Mostly the role it plays in coherence is just in making sure that all
> the devices that want to snoop get a chance to do so. There's also
> the "express snoop" functionality that was added for the new
> multi-level coherence scheme. Other than that it's agnostic about the
> coherence protocol.
>
> Steve
>
>>
>> Thanks!
>>
>> Jiayuan
>> _______________________________________________
>> m5-users mailing list
>> m5-users@m5sim.org
>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>
> _______________________________________________
> m5-users mailing list
> m5-users@m5sim.org
> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>
_______________________________________________
m5-users mailing list
m5-users@m5sim.org
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
