Ah, the patch simply fixes the issue that if a PhysicalMemory object
receives a functional request in timing mode, it alters the request
which was causing things to crash. I don't remember the precise error
and I'm not sure if it happens in the unstable version, it is simply
an issue I encountered using the stable version a while back.
- Clint
On Apr 27, 2009, at 2:52 PM, nathan binkert wrote:
> Can you tell us more about this patch? Should this be in the tree?
>
> Nate
>
> On Mon, Apr 27, 2009 at 11:40 AM, Clint Smullen
> <[email protected]> wrote:
>> The only issue I have ever had with using functional memory requests
>> during timing mode is that some modules explicitly disable them.
>> There
>> is no intrinsic reason (other than that) for them to not work at any
>> time, since the actual memory store is the same for the functional
>> and
>> "physical" memory. Below is my patch for tweaking the behavior of the
>> PhysicalMemory to not create errors, not sure if this relates to the
>> issue you are seeing, however.
>>
>> - Clint
>>
>> diff --git a/src/mem/physical.cc b/src/mem/physical.cc
>> --- a/src/mem/physical.cc
>> +++ b/src/mem/physical.cc
>> @@ -325,12 +325,16 @@
>> if (pmemAddr)
>> memcpy(pkt->getPtr<uint8_t>(), hostAddr, pkt-
>> >getSize());
>> TRACE_PACKET("Read");
>> - pkt->makeAtomicResponse();
>> + if (pkt->needsResponse()) {
>> + pkt->makeAtomicResponse();
>> + }
>> } else if (pkt->isWrite()) {
>> if (pmemAddr)
>> memcpy(hostAddr, pkt->getPtr<uint8_t>(), pkt-
>> >getSize());
>> TRACE_PACKET("Write");
>> - pkt->makeAtomicResponse();
>> + if (pkt->needsResponse()) {
>> + pkt->makeAtomicResponse();
>> + }
>> } else if (pkt->isPrint()) {
>> Packet::PrintReqState *prs =
>> dynamic_cast<Packet::PrintReqState*>(pkt->senderState);
>>
>> On Apr 27, 2009, at 2:44 AM, Rick Strong wrote:
>>
>>> Hi all,
>>>
>>> I am writing a memory debugger that sits in between L1 and the
>>> cpu. It
>>> maintains a functional memory (like MemTest) that mimics the actions
>>> the
>>> memory actions and tests the values for problems. I would like to be
>>> able to plug this in right after restoring a checkpoint when the
>>> physical memory has values (from checkpoint) but the functional
>>> memory
>>> is blank. Is there an easy way to initialize a functional memory
>>> from a
>>> physical memory?
>>>
>>> Also, I see references to MemDebugger class in the cache
>>> implementations
>>> which makes me think someone else has gone down this road before.
>>> Any
>>> one know anything?
>>>
>>> Thanks,
>>> -Rick
>>>
>>> _______________________________________________
>>> 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
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
