Merge zizzer.eecs.umich.edu:/bk/newmem

into  zeep.eecs.umich.edu:/home/gblack/m5/newmem_bus

--HG--
extra : convert_revision : 8267487b935eaf11665841ace3a5c664751b53b0

32 files changed, 397 insertions, 225 deletions

diff --git a/src/cpu/SConscript b/src/cpu/SConscript
index 2bb9a2399..5771a7904 100644
--- a/src/cpu/SConscript
+++ b/src/cpu/SConscript
@@ -158,6 +158,7 @@ if 'O3CPU' in env['CPU_MODELS']:
         o3/scoreboard.cc
         o3/store_set.cc
         ''')
+    sources += Split('memtest/memtest.cc')
     if env['USE_CHECKER']:
         sources += Split('o3/checker_builder.cc')
     else:
diff --git a/src/cpu/memtest/memtest.cc b/src/cpu/memtest/memtest.cc
index 7ea9eaefc..609a07a8e 100644
--- a/src/cpu/memtest/memtest.cc
+++ b/src/cpu/memtest/memtest.cc
@@ -38,39 +38,80 @@
 
 #include "base/misc.hh"
 #include "base/statistics.hh"
-#include "cpu/simple_thread.hh"
+//#include "cpu/simple_thread.hh"
 #include "cpu/memtest/memtest.hh"
-#include "mem/cache/base_cache.hh"
+//#include "mem/cache/base_cache.hh"
+//#include "mem/physical.hh"
 #include "sim/builder.hh"
 #include "sim/sim_events.hh"
 #include "sim/stats.hh"
+#include "mem/packet.hh"
+#include "mem/request.hh"
+#include "mem/port.hh"
+#include "mem/mem_object.hh"
 
 using namespace std;
-using namespace TheISA;
 
 int TESTER_ALLOCATOR=0;
 
+bool
+MemTest::CpuPort::recvTiming(Packet *pkt)
+{
+    memtest->completeRequest(pkt);
+    return true;
+}
+
+Tick
+MemTest::CpuPort::recvAtomic(Packet *pkt)
+{
+    panic("MemTest doesn't expect recvAtomic callback!");
+    return curTick;
+}
+
+void
+MemTest::CpuPort::recvFunctional(Packet *pkt)
+{
+    memtest->completeRequest(pkt);
+}
+
+void
+MemTest::CpuPort::recvStatusChange(Status status)
+{
+    if (status == RangeChange)
+        return;
+
+    panic("MemTest doesn't expect recvStatusChange callback!");
+}
+
+void
+MemTest::CpuPort::recvRetry()
+{
+    memtest->doRetry();
+}
+
 MemTest::MemTest(const string &name,
-                 MemInterface *_cache_interface,
-                 FunctionalMemory *main_mem,
-                 FunctionalMemory *check_mem,
+//		 MemInterface *_cache_interface,
+//		 PhysicalMemory *main_mem,
+//		 PhysicalMemory *check_mem,
                  unsigned _memorySize,
                  unsigned _percentReads,
-                 unsigned _percentCopies,
+//		 unsigned _percentCopies,
                  unsigned _percentUncacheable,
                  unsigned _progressInterval,
                  unsigned _percentSourceUnaligned,
                  unsigned _percentDestUnaligned,
                  Addr _traceAddr,
                  Counter _max_loads)
-    : SimObject(name),
+    : MemObject(name),
       tickEvent(this),
-      cacheInterface(_cache_interface),
-      mainMem(main_mem),
-      checkMem(check_mem),
+      cachePort("test", this),
+      funcPort("functional", this),
+      retryPkt(NULL),
+//      mainMem(main_mem),
+//      checkMem(check_mem),
       size(_memorySize),
       percentReads(_percentReads),
-      percentCopies(_percentCopies),
+//      percentCopies(_percentCopies),
       percentUncacheable(_percentUncacheable),
       progressInterval(_progressInterval),
       nextProgressMessage(_progressInterval),
@@ -81,43 +122,54 @@ MemTest::MemTest(const string &name,
     vector<string> cmd;
     cmd.push_back("/bin/ls");
     vector<string> null_vec;
-    thread = new SimpleThread(NULL, 0, mainMem, 0);
-
-    blockSize = cacheInterface->getBlockSize();
-    blockAddrMask = blockSize - 1;
-    traceBlockAddr = blockAddr(_traceAddr);
-
-    //setup data storage with interesting values
-    uint8_t *data1 = new uint8_t[size];
-    uint8_t *data2 = new uint8_t[size];
-    uint8_t *data3 = new uint8_t[size];
-    memset(data1, 1, size);
-    memset(data2, 2, size);
-    memset(data3, 3, size);
+    //  thread = new SimpleThread(NULL, 0, NULL, 0, mainMem);
     curTick = 0;
 
+    // Needs to be masked off once we know the block size.
+    traceBlockAddr = _traceAddr;
     baseAddr1 = 0x100000;
     baseAddr2 = 0x400000;
     uncacheAddr = 0x800000;
 
-    // set up intial memory contents here
-    mainMem->prot_write(baseAddr1, data1, size);
-    checkMem->prot_write(baseAddr1, data1, size);
-    mainMem->prot_write(baseAddr2, data2, size);
-    checkMem->prot_write(baseAddr2, data2, size);
-    mainMem->prot_write(uncacheAddr, data3, size);
-    checkMem->prot_write(uncacheAddr, data3, size);
-
-    delete [] data1;
-    delete [] data2;
-    delete [] data3;
-
     // set up counters
     noResponseCycles = 0;
     numReads = 0;
     tickEvent.schedule(0);
 
     id = TESTER_ALLOCATOR++;
+    if (TESTER_ALLOCATOR > 8)
+        panic("False sharing memtester only allows up to 8 testers");
+
+    accessRetry = false;
+}
+
+Port *
+MemTest::getPort(const std::string &if_name, int idx)
+{
+    if (if_name == "functional")
+        return &funcPort;
+    else if (if_name == "test")
+        return &cachePort;
+    else
+        panic("No Such Port\n");
+}
+
+void
+MemTest::init()
+{
+    // By the time init() is called, the ports should be hooked up.
+    blockSize = cachePort.peerBlockSize();
+    blockAddrMask = blockSize - 1;
+    traceBlockAddr = blockAddr(traceBlockAddr);
+
+    // set up intial memory contents here
+
+    cachePort.memsetBlob(baseAddr1, 1, size);
+    funcPort.memsetBlob(baseAddr1, 1, size);
+    cachePort.memsetBlob(baseAddr2, 2, size);
+    funcPort.memsetBlob(baseAddr2, 2, size);
+    cachePort.memsetBlob(uncacheAddr, 3, size);
+    funcPort.memsetBlob(uncacheAddr, 3, size);
 }
 
 static void
@@ -132,23 +184,31 @@ printData(ostream &os, uint8_t *data, int nbytes)
 }
 
 void
-MemTest::completeRequest(MemReqPtr &req, uint8_t *data)
+MemTest::completeRequest(Packet *pkt)
 {
+    MemTestSenderState *state =
+        dynamic_cast<MemTestSenderState *>(pkt->senderState);
+
+    uint8_t *data = state->data;
+    uint8_t *pkt_data = pkt->getPtr<uint8_t>();
+    Request *req = pkt->req;
+
     //Remove the address from the list of outstanding
-    std::set<unsigned>::iterator removeAddr = outstandingAddrs.find(req->paddr);
+    std::set<unsigned>::iterator removeAddr = outstandingAddrs.find(req->getPaddr());
     assert(removeAddr != outstandingAddrs.end());
     outstandingAddrs.erase(removeAddr);
 
-    switch (req->cmd) {
-      case Read:
-        if (memcmp(req->data, data, req->size) != 0) {
-            cerr << name() << ": on read of 0x" << hex << req->paddr
-                 << " (0x" << hex << blockAddr(req->paddr) << ")"
+    switch (pkt->cmd) {
+      case Packet::ReadResp:
+
+        if (memcmp(pkt_data, data, pkt->getSize()) != 0) {
+            cerr << name() << ": on read of 0x" << hex << req->getPaddr()
+                 << " (0x" << hex << blockAddr(req->getPaddr()) << ")"
                  << "@ cycle " << dec << curTick
                  << ", cache returns 0x";
-            printData(cerr, req->data, req->size);
+            printData(cerr, pkt_data, pkt->getSize());
             cerr << ", expected 0x";
-            printData(cerr, data, req->size);
+            printData(cerr, data, pkt->getSize());
             cerr << endl;
             fatal("");
         }
@@ -163,13 +223,13 @@ MemTest::completeRequest(MemReqPtr &req, uint8_t *data)
         }
 
         if (numReads >= maxLoads)
-            SimExit(curTick, "Maximum number of loads reached!");
+            exitSimLoop("Maximum number of loads reached!");
         break;
 
-      case Write:
+      case Packet::WriteResp:
         numWritesStat++;
         break;
-
+/*
       case Copy:
         //Also remove dest from outstanding list
         removeAddr = outstandingAddrs.find(req->dest);
@@ -177,36 +237,37 @@ MemTest::completeRequest(MemReqPtr &req, uint8_t *data)
         outstandingAddrs.erase(removeAddr);
         numCopiesStat++;
         break;
-
+*/
       default:
         panic("invalid command");
     }
 
-    if (blockAddr(req->paddr) == traceBlockAddr) {
+    if (blockAddr(req->getPaddr()) == traceBlockAddr) {
         cerr << name() << ": completed "
-             << (req->cmd.isWrite() ? "write" : "read")
+             << (pkt->isWrite() ? "write" : "read")
              << " access of "
-             << dec << req->size << " bytes at address 0x"
-             << hex << req->paddr
-             << " (0x" << hex << blockAddr(req->paddr) << ")"
+             << dec << pkt->getSize() << " bytes at address 0x"
+             << hex << req->getPaddr()
+             << " (0x" << hex << blockAddr(req->getPaddr()) << ")"
              << ", value = 0x";
-        printData(cerr, req->data, req->size);
+        printData(cerr, pkt_data, pkt->getSize());
         cerr << " @ cycle " << dec << curTick;
 
         cerr << endl;
     }
 
     noResponseCycles = 0;
+    delete state;
     delete [] data;
+    delete pkt->req;
+    delete pkt;
 }
 
-
 void
 MemTest::regStats()
 {
     using namespace Stats;
 
-
     numReadsStat
         .name(name() + ".num_reads")
         .desc("number of read accesses completed")
@@ -234,7 +295,7 @@ MemTest::tick()
         fatal("");
     }
 
-    if (cacheInterface->isBlocked()) {
+    if (accessRetry) {
         return;
     }
 
@@ -248,30 +309,30 @@ MemTest::tick()
 
     //If we aren't doing copies, use id as offset, and do a false sharing
     //mem tester
-    if (percentCopies == 0) {
-        //We can eliminate the lower bits of the offset, and then use the id
-        //to offset within the blks
-        offset &= ~63; //Not the low order bits
-        offset += id;
-        access_size = 0;
-    }
+    //We can eliminate the lower bits of the offset, and then use the id
+    //to offset within the blks
+    offset &= ~63; //Not the low order bits
+    offset += id;
+    access_size = 0;
 
-    MemReqPtr req = new MemReq();
+    Request *req = new Request();
+    uint32_t flags = 0;
+    Addr paddr;
 
     if (cacheable < percentUncacheable) {
-        req->flags |= UNCACHEABLE;
-        req->paddr = uncacheAddr + offset;
+        flags |= UNCACHEABLE;
+        paddr = uncacheAddr + offset;
     } else {
-        req->paddr = ((base) ? baseAddr1 : baseAddr2) + offset;
+        paddr = ((base) ? baseAddr1 : baseAddr2) + offset;
     }
     // bool probe = (random() % 2 == 1) && !req->isUncacheable();
     bool probe = false;
 
-    req->size = 1 << access_size;
-    req->data = new uint8_t[req->size];
-    req->paddr &= ~(req->size - 1);
-    req->time = curTick;
-    req->xc = thread->getProxy();
+    paddr &= ~((1 << access_size) - 1);
+    req->setPhys(paddr, 1 << access_size, flags);
+    req->setThreadContext(id,0);
+
+    uint8_t *result = new uint8_t[8];
 
     if (cmd < percentReads) {
         // read
@@ -279,60 +340,81 @@ MemTest::tick()
         //For now we only allow one outstanding request per addreess per tester
         //This means we assume CPU does write forwarding to reads that alias something
         //in the cpu store buffer.
-        if (outstandingAddrs.find(req->paddr) != outstandingAddrs.end()) return;
-        else outstandingAddrs.insert(req->paddr);
+        if (outstandingAddrs.find(paddr) != outstandingAddrs.end()) return;
+        else outstandingAddrs.insert(paddr);
 
-        req->cmd = Read;
-        uint8_t *result = new uint8_t[8];
-        checkMem->access(Read, req->paddr, result, req->size);
-        if (blockAddr(req->paddr) == traceBlockAddr) {
+        // ***** NOTE FOR RON: I'm not sure how to access checkMem. - Kevin
+        funcPort.readBlob(req->getPaddr(), result, req->getSize());
+
+        if (blockAddr(paddr) == traceBlockAddr) {
             cerr << name()
                  << ": initiating read "
                  << ((probe) ? "probe of " : "access of ")
-                 << dec << req->size << " bytes from addr 0x"
-                 << hex << req->paddr
-                 << " (0x" << hex << blockAddr(req->paddr) << ")"
+                 << dec << req->getSize() << " bytes from addr 0x"
+                 << hex << paddr
+                 << " (0x" << hex << blockAddr(paddr) << ")"
                  << " at cycle "
                  << dec << curTick << endl;
         }
+
+        Packet *pkt = new Packet(req, Packet::ReadReq, Packet::Broadcast);
+        pkt->dataDynamicArray(new uint8_t[req->getSize()]);
+        MemTestSenderState *state = new MemTestSenderState(result);
+        pkt->senderState = state;
+
         if (probe) {
-            cacheInterface->probeAndUpdate(req);
-            completeRequest(req, result);
+            cachePort.sendFunctional(pkt);
+//	    completeRequest(pkt, result);
         } else {
-            req->completionEvent = new MemCompleteEvent(req, result, this);
-            cacheInterface->access(req);
+//	    req->completionEvent = new MemCompleteEvent(req, result, this);
+            if (!cachePort.sendTiming(pkt)) {
+                accessRetry = true;
+                retryPkt = pkt;
+            }
         }
-    } else if (cmd < (100 - percentCopies)){
+    } else {
         // write
 
         //For now we only allow one outstanding request per addreess per tester
         //This means we assume CPU does write forwarding to reads that alias something
         //in the cpu store buffer.
-        if (outstandingAddrs.find(req->paddr) != outstandingAddrs.end()) return;
-        else outstandingAddrs.insert(req->paddr);
+        if (outstandingAddrs.find(paddr) != outstandingAddrs.end()) return;
+        else outstandingAddrs.insert(paddr);
 
-        req->cmd = Write;
-        memcpy(req->data, &data, req->size);
-        checkMem->access(Write, req->paddr, req->data, req->size);
-        if (blockAddr(req->paddr) == traceBlockAddr) {
+/*
+        if (blockAddr(req->getPaddr()) == traceBlockAddr) {
             cerr << name() << ": initiating write "
                  << ((probe)?"probe of ":"access of ")
-                 << dec << req->size << " bytes (value = 0x";
-            printData(cerr, req->data, req->size);
+                 << dec << req->getSize() << " bytes (value = 0x";
+            printData(cerr, data_pkt->getPtr(), req->getSize());
             cerr << ") to addr 0x"
-                 << hex << req->paddr
-                 << " (0x" << hex << blockAddr(req->paddr) << ")"
+                 << hex << req->getPaddr()
+                 << " (0x" << hex << blockAddr(req->getPaddr()) << ")"
                  << " at cycle "
                  << dec << curTick << endl;
         }
+*/
+        Packet *pkt = new Packet(req, Packet::WriteReq, Packet::Broadcast);
+        uint8_t *pkt_data = new uint8_t[req->getSize()];
+        pkt->dataDynamicArray(pkt_data);
+        memcpy(pkt_data, &data, req->getSize());
+        MemTestSenderState *state = new MemTestSenderState(result);
+        pkt->senderState = state;
+
+        funcPort.writeBlob(req->getPaddr(), pkt_data, req->getSize());
+
         if (probe) {
-            cacheInterface->probeAndUpdate(req);
-            completeRequest(req, NULL);
+            cachePort.sendFunctional(pkt);
+//	    completeRequest(req, NULL);
         } else {
-            req->completionEvent = new MemCompleteEvent(req, NULL, this);
-            cacheInterface->access(req);
+//	    req->completionEvent = new MemCompleteEvent(req, NULL, this);
+            if (!cachePort.sendTiming(pkt)) {
+                accessRetry = true;
+                retryPkt = pkt;
+            }
         }
-    } else {
+    }
+/*    else {
         // copy
         unsigned source_align = random() % 100;
         unsigned dest_align = random() % 100;
@@ -369,38 +451,32 @@ MemTest::tick()
                  << " (0x" << hex << blockAddr(dest) << ")"
                  << " at cycle "
                  << dec << curTick << endl;
-        }
+        }*
         cacheInterface->access(req);
         uint8_t result[blockSize];
         checkMem->access(Read, source, &result, blockSize);
         checkMem->access(Write, dest, &result, blockSize);
     }
+*/
 }
 
-
 void
-MemCompleteEvent::process()
-{
-    tester->completeRequest(req, data);
-    delete this;
-}
-
-
-const char *
-MemCompleteEvent::description()
+MemTest::doRetry()
 {
-    return "memory access completion";
+    if (cachePort.sendTiming(retryPkt)) {
+        accessRetry = false;
+        retryPkt = NULL;
+    }
 }
 
-
 BEGIN_DECLARE_SIM_OBJECT_PARAMS(MemTest)
 
-    SimObjectParam<BaseCache *> cache;
-    SimObjectParam<FunctionalMemory *> main_mem;
-    SimObjectParam<FunctionalMemory *> check_mem;
+//    SimObjectParam<BaseCache *> cache;
+//    SimObjectParam<PhysicalMemory *> main_mem;
+//    SimObjectParam<PhysicalMemory *> check_mem;
     Param<unsigned> memory_size;
     Param<unsigned> percent_reads;
-    Param<unsigned> percent_copies;
+//    Param<unsigned> percent_copies;
     Param<unsigned> percent_uncacheable;
     Param<unsigned> progress_interval;
     Param<unsigned> percent_source_unaligned;
@@ -413,12 +489,12 @@ END_DECLARE_SIM_OBJECT_PARAMS(MemTest)
 
 BEGIN_INIT_SIM_OBJECT_PARAMS(MemTest)
 
-    INIT_PARAM(cache, "L1 cache"),
-    INIT_PARAM(main_mem, "hierarchical memory"),
-    INIT_PARAM(check_mem, "check memory"),
+//    INIT_PARAM(cache, "L1 cache"),
+//    INIT_PARAM(main_mem, "hierarchical memory"),
+//    INIT_PARAM(check_mem, "check memory"),
     INIT_PARAM(memory_size, "memory size"),
     INIT_PARAM(percent_reads, "target read percentage"),
-    INIT_PARAM(percent_copies, "target copy percentage"),
+//    INIT_PARAM(percent_copies, "target copy percentage"),
     INIT_PARAM(percent_uncacheable, "target uncacheable percentage"),
     INIT_PARAM(progress_interval, "progress report interval (in accesses)"),
     INIT_PARAM(percent_source_unaligned,
@@ -433,8 +509,8 @@ END_INIT_SIM_OBJECT_PARAMS(MemTest)
 
 CREATE_SIM_OBJECT(MemTest)
 {
-    return new MemTest(getInstanceName(), cache->getInterface(), main_mem,
-                       check_mem, memory_size, percent_reads, percent_copies,
+    return new MemTest(getInstanceName(), /*cache->getInterface(),*/ /*main_mem,*/
+                       /*check_mem,*/ memory_size, percent_reads, /*percent_copies,*/
                        percent_uncacheable, progress_interval,
                        percent_source_unaligned, percent_dest_unaligned,
                        trace_addr, max_loads);
diff --git a/src/cpu/memtest/memtest.hh b/src/cpu/memtest/memtest.hh
index 42fb235db..278012eba 100644
--- a/src/cpu/memtest/memtest.hh
+++ b/src/cpu/memtest/memtest.hh
@@ -35,25 +35,27 @@
 #include <set>
 
 #include "base/statistics.hh"
-#include "mem/functional/functional.hh"
-#include "mem/mem_interface.hh"
+//#include "mem/functional/functional.hh"
+//#include "mem/mem_interface.hh"
 #include "sim/eventq.hh"
 #include "sim/sim_exit.hh"
 #include "sim/sim_object.hh"
 #include "sim/stats.hh"
+#include "mem/mem_object.hh"
+#include "mem/port.hh"
 
-class ThreadContext;
-class MemTest : public SimObject
+class Packet;
+class MemTest : public MemObject
 {
   public:
 
     MemTest(const std::string &name,
-            MemInterface *_cache_interface,
-            FunctionalMemory *main_mem,
-            FunctionalMemory *check_mem,
+//	    MemInterface *_cache_interface,
+//	    PhysicalMemory *main_mem,
+//	    PhysicalMemory *check_mem,
             unsigned _memorySize,
             unsigned _percentReads,
-            unsigned _percentCopies,
+//	    unsigned _percentCopies,
             unsigned _percentUncacheable,
             unsigned _progressInterval,
             unsigned _percentSourceUnaligned,
@@ -61,6 +63,8 @@ class MemTest : public SimObject
             Addr _traceAddr,
             Counter _max_loads);
 
+    virtual void init();
+
     // register statistics
     virtual void regStats();
 
@@ -69,6 +73,8 @@ class MemTest : public SimObject
     // main simulation loop (one cycle)
     void tick();
 
+    virtual Port *getPort(const std::string &if_name, int idx = -1);
+
   protected:
     class TickEvent : public Event
     {
@@ -82,16 +88,62 @@ class MemTest : public SimObject
     };
 
     TickEvent tickEvent;
+    class CpuPort : public Port
+    {
+
+        MemTest *memtest;
+
+      public:
+
+        CpuPort(const std::string &_name, MemTest *_memtest)
+            : Port(_name), memtest(_memtest)
+        { }
+
+      protected:
+
+        virtual bool recvTiming(Packet *pkt);
+
+        virtual Tick recvAtomic(Packet *pkt);
+
+        virtual void recvFunctional(Packet *pkt);
+
+        virtual void recvStatusChange(Status status);
+
+        virtual void recvRetry();
+
+        virtual void getDeviceAddressRanges(AddrRangeList &resp,
+            AddrRangeList &snoop)
+        { resp.clear(); snoop.clear(); }
+    };
+
+    CpuPort cachePort;
+    CpuPort funcPort;
+
+    class MemTestSenderState : public Packet::SenderState
+    {
+      public:
+        /** Constructor. */
+        MemTestSenderState(uint8_t *_data)
+            : data(_data)
+        { }
+
+        // Hold onto data pointer
+        uint8_t *data;
+    };
+
+//    Request *dataReq;
+    Packet  *retryPkt;
+//    MemInterface *cacheInterface;
+//    PhysicalMemory *mainMem;
+//    PhysicalMemory *checkMem;
+//    SimpleThread *thread;
 
-    MemInterface *cacheInterface;
-    FunctionalMemory *mainMem;
-    FunctionalMemory *checkMem;
-    SimpleThread *thread;
+    bool accessRetry;
 
     unsigned size;		// size of testing memory region
 
     unsigned percentReads;	// target percentage of read accesses
-    unsigned percentCopies;	// target percentage of copy accesses
+//    unsigned percentCopies;	// target percentage of copy accesses
     unsigned percentUncacheable;
 
     int id;
@@ -128,29 +180,11 @@ class MemTest : public SimObject
     Stats::Scalar<> numCopiesStat;
 
     // called by MemCompleteEvent::process()
-    void completeRequest(MemReqPtr &req, uint8_t *data);
+    void completeRequest(Packet *pkt);
 
-    friend class MemCompleteEvent;
-};
+    void doRetry();
 
-
-class MemCompleteEvent : public Event
-{
-    MemReqPtr req;
-    uint8_t *data;
-    MemTest *tester;
-
-  public:
-
-    MemCompleteEvent(MemReqPtr &_req, uint8_t *_data, MemTest *_tester)
-        : Event(&mainEventQueue),
-          req(_req), data(_data), tester(_tester)
-    {
-    }
-
-    void process();
-
-    virtual const char *description();
+    friend class MemCompleteEvent;
 };
 
 #endif // __CPU_MEMTEST_MEMTEST_HH__
diff --git a/src/cpu/o3/fetch.hh b/src/cpu/o3/fetch.hh
index 1a2ca32a4..280bf0e71 100644
--- a/src/cpu/o3/fetch.hh
+++ b/src/cpu/o3/fetch.hh
@@ -96,7 +96,7 @@ class DefaultFetch
         /** Returns the address ranges of this device. */
         virtual void getDeviceAddressRanges(AddrRangeList &resp,
                                             AddrRangeList &snoop)
-        { resp.clear(); snoop.clear(); }
+        { resp.clear(); snoop.clear(); snoop.push_back(RangeSize(0,-1)); }
 
         /** Timing version of receive.  Handles setting fetch to the
          * proper status to start fetching. */
diff --git a/src/cpu/o3/fetch_impl.hh b/src/cpu/o3/fetch_impl.hh
index 497179576..b3c3caaad 100644
--- a/src/cpu/o3/fetch_impl.hh
+++ b/src/cpu/o3/fetch_impl.hh
@@ -63,7 +63,7 @@ template<class Impl>
 void
 DefaultFetch<Impl>::IcachePort::recvFunctional(PacketPtr pkt)
 {
-    panic("DefaultFetch doesn't expect recvFunctional callback!");
+    warn("Default fetch doesn't update it's state from a functional call.");
 }
 
 template<class Impl>
diff --git a/src/cpu/o3/lsq.hh b/src/cpu/o3/lsq.hh
index 190734dc2..6b12d75b4 100644
--- a/src/cpu/o3/lsq.hh
+++ b/src/cpu/o3/lsq.hh
@@ -311,7 +311,7 @@ class LSQ {
         /** Returns the address ranges of this device. */
         virtual void getDeviceAddressRanges(AddrRangeList &resp,
                                             AddrRangeList &snoop)
-        { resp.clear(); snoop.clear(); }
+        { resp.clear(); snoop.clear(); snoop.push_back(RangeSize(0,-1)); }
 
         /** Timing version of receive.  Handles writing back and
          * completing the load or store that has returned from
diff --git a/src/cpu/o3/lsq_impl.hh b/src/cpu/o3/lsq_impl.hh
index 2bbab71f0..7b7d1eb8e 100644
--- a/src/cpu/o3/lsq_impl.hh
+++ b/src/cpu/o3/lsq_impl.hh
@@ -46,7 +46,7 @@ template <class Impl>
 void
 LSQ<Impl>::DcachePort::recvFunctional(PacketPtr pkt)
 {
-    panic("O3CPU doesn't expect recvFunctional callback!");
+    warn("O3CPU doesn't update things on a recvFunctional.");
 }
 
 template <class Impl>
diff --git a/src/cpu/ozone/front_end.hh b/src/cpu/ozone/front_end.hh
index 5ffd3666e..59cf9785c 100644
--- a/src/cpu/ozone/front_end.hh
+++ b/src/cpu/ozone/front_end.hh
@@ -92,7 +92,7 @@ class FrontEnd
         /** Returns the address ranges of this device. */
         virtual void getDeviceAddressRanges(AddrRangeList &resp,
                                             AddrRangeList &snoop)
-        { resp.clear(); snoop.clear(); }
+        { resp.clear(); snoop.clear(); snoop.push_back(RangeSize(0,-1)); }
 
         /** Timing version of receive.  Handles setting fetch to the
          * proper status to start fetching. */
diff --git a/src/cpu/ozone/front_end_impl.hh b/src/cpu/ozone/front_end_impl.hh
index 5956c5cba..c814ff9c7 100644
--- a/src/cpu/ozone/front_end_impl.hh
+++ b/src/cpu/ozone/front_end_impl.hh
@@ -59,7 +59,7 @@ template<class Impl>
 void
 FrontEnd<Impl>::IcachePort::recvFunctional(PacketPtr pkt)
 {
-    panic("FrontEnd doesn't expect recvFunctional callback!");
+    warn("FrontEnd doesn't update state from functional calls");
 }
 
 template<class Impl>
diff --git a/src/cpu/ozone/lw_lsq.hh b/src/cpu/ozone/lw_lsq.hh
index 347f4569b..9b93ce74f 100644
--- a/src/cpu/ozone/lw_lsq.hh
+++ b/src/cpu/ozone/lw_lsq.hh
@@ -260,7 +260,7 @@ class OzoneLWLSQ {
 
         virtual void getDeviceAddressRanges(AddrRangeList &resp,
                                             AddrRangeList &snoop)
-        { resp.clear(); snoop.clear(); }
+        { resp.clear(); snoop.clear(); snoop.push_back(RangeSize(0,-1); }
 
         virtual bool recvTiming(PacketPtr pkt);
 
diff --git a/src/cpu/ozone/lw_lsq_impl.hh b/src/cpu/ozone/lw_lsq_impl.hh
index 9d17b027f..e523712da 100644
--- a/src/cpu/ozone/lw_lsq_impl.hh
+++ b/src/cpu/ozone/lw_lsq_impl.hh
@@ -72,7 +72,7 @@ template <class Impl>
 void
 OzoneLWLSQ<Impl>::DcachePort::recvFunctional(PacketPtr pkt)
 {
-    panic("O3CPU doesn't expect recvFunctional callback!");
+    warn("O3CPU doesn't update things on a recvFunctional");
 }
 
 template <class Impl>
diff --git a/src/cpu/simple/atomic.cc b/src/cpu/simple/atomic.cc
index 42b0e9783..490be20ae 100644
--- a/src/cpu/simple/atomic.cc
+++ b/src/cpu/simple/atomic.cc
@@ -94,7 +94,7 @@ AtomicSimpleCPU::init()
 bool
 AtomicSimpleCPU::CpuPort::recvTiming(Packet *pkt)
 {
-    panic("AtomicSimpleCPU doesn't expect recvAtomic callback!");
+    panic("AtomicSimpleCPU doesn't expect recvTiming callback!");
     return true;
 }
 
@@ -108,7 +108,8 @@ AtomicSimpleCPU::CpuPort::recvAtomic(Packet *pkt)
 void
 AtomicSimpleCPU::CpuPort::recvFunctional(Packet *pkt)
 {
-    panic("AtomicSimpleCPU doesn't expect recvFunctional callback!");
+    //No internal storage to update, just return
+    return;
 }
 
 void
@@ -161,6 +162,8 @@ AtomicSimpleCPU::serialize(ostream &os)
 {
     SimObject::State so_state = SimObject::getState();
     SERIALIZE_ENUM(so_state);
+    Status _status = status();
+    SERIALIZE_ENUM(_status);
     BaseSimpleCPU::serialize(os);
     nameOut(os, csprintf("%s.tickEvent", name()));
     tickEvent.serialize(os);
@@ -171,6 +174,7 @@ AtomicSimpleCPU::unserialize(Checkpoint *cp, const string &section)
 {
     SimObject::State so_state;
     UNSERIALIZE_ENUM(so_state);
+    UNSERIALIZE_ENUM(_status);
     BaseSimpleCPU::unserialize(cp, section);
     tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
 }
diff --git a/src/cpu/simple/atomic.hh b/src/cpu/simple/atomic.hh
index b602af558..52afd76ef 100644
--- a/src/cpu/simple/atomic.hh
+++ b/src/cpu/simple/atomic.hh
@@ -104,9 +104,9 @@ class AtomicSimpleCPU : public BaseSimpleCPU
 
         virtual void getDeviceAddressRanges(AddrRangeList &resp,
             AddrRangeList &snoop)
-        { resp.clear(); snoop.clear(); }
-    };
+        { resp.clear(); snoop.clear(); snoop.push_back(RangeSize(0,-1)); }
 
+    };
     CpuPort icachePort;
     CpuPort dcachePort;
 
diff --git a/src/cpu/simple/timing.cc b/src/cpu/simple/timing.cc
index a394468b9..48362c42a 100644
--- a/src/cpu/simple/timing.cc
+++ b/src/cpu/simple/timing.cc
@@ -74,7 +74,8 @@ TimingSimpleCPU::CpuPort::recvAtomic(Packet *pkt)
 void
 TimingSimpleCPU::CpuPort::recvFunctional(Packet *pkt)
 {
-    panic("TimingSimpleCPU doesn't expect recvFunctional callback!");
+    //No internal storage to update, jusst return
+    return;
 }
 
 void
diff --git a/src/cpu/simple/timing.hh b/src/cpu/simple/timing.hh
index b65eebd99..18e13aeb2 100644
--- a/src/cpu/simple/timing.hh
+++ b/src/cpu/simple/timing.hh
@@ -92,7 +92,7 @@ class TimingSimpleCPU : public BaseSimpleCPU
 
         virtual void getDeviceAddressRanges(AddrRangeList &resp,
             AddrRangeList &snoop)
-        { resp.clear(); snoop.clear(); }
+        { resp.clear(); snoop.clear(); snoop.push_back(RangeSize(0,-1)); }
 
         struct TickEvent : public Event
         {
diff --git a/src/dev/ide_ctrl.cc b/src/dev/ide_ctrl.cc
index e8d7f4817..8007fda5e 100644
--- a/src/dev/ide_ctrl.cc
+++ b/src/dev/ide_ctrl.cc
@@ -742,7 +742,6 @@ IdeController::unserialize(Checkpoint *cp, const std::string &section)
     UNSERIALIZE_SCALAR(bm_enabled);
     UNSERIALIZE_ARRAY(cmd_in_progress,
                       sizeof(cmd_in_progress) / sizeof(cmd_in_progress[0]));
-    pioPort->sendStatusChange(Port::RangeChange);
 }
 
 #ifndef DOXYGEN_SHOULD_SKIP_THIS
diff --git a/src/dev/pcidev.cc b/src/dev/pcidev.cc
index c3b83f448..b16ddb31a 100644
--- a/src/dev/pcidev.cc
+++ b/src/dev/pcidev.cc
@@ -302,6 +302,8 @@ PciDev::unserialize(Checkpoint *cp, const std::string &section)
     UNSERIALIZE_ARRAY(BARAddrs, sizeof(BARAddrs) / sizeof(BARAddrs[0]));
     UNSERIALIZE_ARRAY(config.data,
                       sizeof(config.data) / sizeof(config.data[0]));
+    pioPort->sendStatusChange(Port::RangeChange);
+
 }
 
 #ifndef DOXYGEN_SHOULD_SKIP_THIS
diff --git a/src/mem/bus.cc b/src/mem/bus.cc
index c288e34c0..7584ffffd 100644
--- a/src/mem/bus.cc
+++ b/src/mem/bus.cc
@@ -270,6 +270,18 @@ Bus::atomicSnoop(Packet *pkt)
     }
 }
 
+void
+Bus::functionalSnoop(Packet *pkt)
+{
+    std::vector<int> ports = findSnoopPorts(pkt->getAddr(), pkt->getSrc());
+
+    while (!ports.empty())
+    {
+        interfaces[ports.back()]->sendFunctional(pkt);
+        ports.pop_back();
+    }
+}
+
 bool
 Bus::timingSnoop(Packet *pkt)
 {
@@ -306,7 +318,7 @@ Bus::recvFunctional(Packet *pkt)
     DPRINTF(Bus, "recvFunctional: packet src %d dest %d addr 0x%x cmd %s\n",
             pkt->getSrc(), pkt->getDest(), pkt->getAddr(), pkt->cmdString());
     assert(pkt->getDest() == Packet::Broadcast);
-    atomicSnoop(pkt);
+    functionalSnoop(pkt);
     findPort(pkt->getAddr(), pkt->getSrc())->sendFunctional(pkt);
 }
 
diff --git a/src/mem/bus.hh b/src/mem/bus.hh
index c9b0a76a0..a89738775 100644
--- a/src/mem/bus.hh
+++ b/src/mem/bus.hh
@@ -109,6 +109,9 @@ class Bus : public MemObject
     /** Snoop all relevant ports atomicly. */
     void atomicSnoop(Packet *pkt);
 
+    /** Snoop all relevant ports functionally. */
+    void functionalSnoop(Packet *pkt);
+
     /** Call snoop on caches, be sure to set SNOOP_COMMIT bit if you want
      * the snoop to happen
      * @return True if succeds.
diff --git a/src/mem/cache/base_cache.cc b/src/mem/cache/base_cache.cc
index 02a46b444..b2c3a0439 100644
--- a/src/mem/cache/base_cache.cc
+++ b/src/mem/cache/base_cache.cc
@@ -71,7 +71,7 @@ BaseCache::CachePort::deviceBlockSize()
 bool
 BaseCache::CachePort::recvTiming(Packet *pkt)
 {
-    if (blocked)
+    if (pkt->isRequest() && blocked)
     {
         DPRINTF(Cache,"Scheduling a retry while blocked\n");
         mustSendRetry = true;
@@ -108,10 +108,11 @@ BaseCache::CachePort::recvRetry()
     else if (!isCpuSide)
     {
         pkt = cache->getPacket();
+        MSHR* mshr = (MSHR*)pkt->senderState;
         bool success = sendTiming(pkt);
         DPRINTF(Cache, "Address %x was %s in sending the timing request\n",
                 pkt->getAddr(), success ? "succesful" : "unsuccesful");
-        cache->sendResult(pkt, success);
+        cache->sendResult(pkt, mshr, success);
         if (success && cache->doMasterRequest())
         {
             //Still more to issue, rerequest in 1 cycle
@@ -122,7 +123,9 @@ BaseCache::CachePort::recvRetry()
     }
     else
     {
-        pkt = cache->getCoherencePacket();
+        //pkt = cache->getCoherencePacket();
+        //We save the packet, no reordering on CSHRS
+        pkt = cshrRetry;
         bool success = sendTiming(pkt);
         if (success && cache->doSlaveRequest())
         {
@@ -181,10 +184,11 @@ BaseCache::CacheEvent::process()
         {
             //MSHR
             pkt = cachePort->cache->getPacket();
+            MSHR* mshr = (MSHR*) pkt->senderState;
             bool success = cachePort->sendTiming(pkt);
             DPRINTF(Cache, "Address %x was %s in sending the timing request\n",
                     pkt->getAddr(), success ? "succesful" : "unsuccesful");
-            cachePort->cache->sendResult(pkt, success);
+            cachePort->cache->sendResult(pkt, mshr, success);
             if (success && cachePort->cache->doMasterRequest())
             {
                 //Still more to issue, rerequest in 1 cycle
@@ -197,7 +201,11 @@ BaseCache::CacheEvent::process()
             //CSHR
             pkt = cachePort->cache->getCoherencePacket();
             bool success = cachePort->sendTiming(pkt);
-            if (success && cachePort->cache->doSlaveRequest())
+            if (!success) {
+                //Need to send on a retry
+                cachePort->cshrRetry = pkt;
+            }
+            else if (cachePort->cache->doSlaveRequest())
             {
                 //Still more to issue, rerequest in 1 cycle
                 pkt = NULL;
diff --git a/src/mem/cache/base_cache.hh b/src/mem/cache/base_cache.hh
index 2e92e7730..c45f3b71b 100644
--- a/src/mem/cache/base_cache.hh
+++ b/src/mem/cache/base_cache.hh
@@ -72,6 +72,7 @@ enum RequestCause{
     Request_PF
 };
 
+class MSHR;
 /**
  * A basic cache interface. Implements some common functions for speed.
  */
@@ -112,6 +113,8 @@ class BaseCache : public MemObject
         bool isCpuSide;
 
         std::list<Packet *> drainList;
+
+        Packet *cshrRetry;
     };
 
     struct CacheEvent : public Event
@@ -156,7 +159,7 @@ class BaseCache : public MemObject
         if (status == Port::RangeChange){
             if (!isCpuSide) {
                 cpuSidePort->sendStatusChange(Port::RangeChange);
-                if (topLevelCache && !snoopRangesSent) {
+                if (!snoopRangesSent) {
                     snoopRangesSent = true;
                     memSidePort->sendStatusChange(Port::RangeChange);
                 }
@@ -177,7 +180,7 @@ class BaseCache : public MemObject
         fatal("No implementation");
     }
 
-    virtual void sendResult(Packet* &pkt, bool success)
+    virtual void sendResult(Packet* &pkt, MSHR* mshr, bool success)
     {
 
         fatal("No implementation");
@@ -568,14 +571,14 @@ class BaseCache : public MemObject
         {
             //This is where snoops get updated
             AddrRangeList dummy;
-            if (!topLevelCache)
-            {
+//            if (!topLevelCache)
+//            {
                 cpuSidePort->getPeerAddressRanges(dummy, snoop);
-            }
-            else
-            {
-                snoop.push_back(RangeSize(0,-1));
-            }
+//            }
+//            else
+//            {
+//                snoop.push_back(RangeSize(0,-1));
+//            }
 
             return;
         }
diff --git a/src/mem/cache/cache.hh b/src/mem/cache/cache.hh
index 4b8870c95..923bf8255 100644
--- a/src/mem/cache/cache.hh
+++ b/src/mem/cache/cache.hh
@@ -175,7 +175,7 @@ class Cache : public BaseCache
      * @param pkt The request.
      * @param success True if the request was sent successfully.
      */
-    virtual void sendResult(Packet * &pkt, bool success);
+    virtual void sendResult(Packet * &pkt, MSHR* mshr, bool success);
 
     /**
      * Handles a response (cache line fill/write ack) from the bus.
diff --git a/src/mem/cache/cache_impl.hh b/src/mem/cache/cache_impl.hh
index 1f03065b6..bde7ac04b 100644
--- a/src/mem/cache/cache_impl.hh
+++ b/src/mem/cache/cache_impl.hh
@@ -73,7 +73,7 @@ doTimingAccess(Packet *pkt, CachePort *cachePort, bool isCpuSide)
             handleResponse(pkt);
         else {
             //Check if we should do the snoop
-            if (pkt->flags && SNOOP_COMMIT)
+            if (pkt->flags & SNOOP_COMMIT)
                 snoop(pkt);
         }
     }
@@ -236,9 +236,9 @@ Cache<TagStore,Buffering,Coherence>::access(PacketPtr &pkt)
         missQueue->doWriteback(writebacks.front());
         writebacks.pop_front();
     }
-    DPRINTF(Cache, "%s %x %s blk_addr: %x pc %x\n", pkt->cmdString(),
+    DPRINTF(Cache, "%s %x %s blk_addr: %x\n", pkt->cmdString(),
             pkt->getAddr() & (((ULL(1))<<48)-1), (blk) ? "hit" : "miss",
-            pkt->getAddr() & ~((Addr)blkSize - 1), pkt->req->getPC());
+            pkt->getAddr() & ~((Addr)blkSize - 1));
     if (blk) {
         // Hit
         hits[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
@@ -287,10 +287,10 @@ Cache<TagStore,Buffering,Coherence>::getPacket()
 
 template<class TagStore, class Buffering, class Coherence>
 void
-Cache<TagStore,Buffering,Coherence>::sendResult(PacketPtr &pkt, bool success)
+Cache<TagStore,Buffering,Coherence>::sendResult(PacketPtr &pkt, MSHR* mshr, bool success)
 {
     if (success) {
-        missQueue->markInService(pkt);
+              missQueue->markInService(pkt, mshr);
           //Temp Hack for UPGRADES
           if (pkt->cmd == Packet::UpgradeReq) {
               handleResponse(pkt);
@@ -314,9 +314,11 @@ Cache<TagStore,Buffering,Coherence>::handleResponse(Packet * &pkt)
             blk = tags->findBlock(pkt);
             CacheBlk::State old_state = (blk) ? blk->status : 0;
             PacketList writebacks;
+            CacheBlk::State new_state = coherence->getNewState(pkt,old_state);
+            DPRINTF(Cache, "Block for blk addr %x moving from state %i to %i\n",
+                    pkt->getAddr() & (((ULL(1))<<48)-1), old_state, new_state);
             blk = tags->handleFill(blk, (MSHR*)pkt->senderState,
-                                   coherence->getNewState(pkt,old_state),
-                                   writebacks, pkt);
+                                   new_state, writebacks, pkt);
             while (!writebacks.empty()) {
                     missQueue->doWriteback(writebacks.front());
                     writebacks.pop_front();
@@ -387,6 +389,7 @@ Cache<TagStore,Buffering,Coherence>::snoop(Packet * &pkt)
                     //If the outstanding request was an invalidate (upgrade,readex,..)
                     //Then we need to ACK the request until we get the data
                     //Also NACK if the outstanding request is not a cachefill (writeback)
+                    assert(!(pkt->flags & SATISFIED));
                     pkt->flags |= SATISFIED;
                     pkt->flags |= NACKED_LINE;
                     assert("Don't detect these on the other side yet\n");
@@ -426,6 +429,7 @@ Cache<TagStore,Buffering,Coherence>::snoop(Packet * &pkt)
                     if (pkt->isRead()) {
                         //Only Upgrades don't get here
                         //Supply the data
+                        assert(!(pkt->flags & SATISFIED));
                         pkt->flags |= SATISFIED;
 
                         //If we are in an exclusive protocol, make it ask again
@@ -444,7 +448,7 @@ Cache<TagStore,Buffering,Coherence>::snoop(Packet * &pkt)
 
                     if (pkt->isInvalidate()) {
                         //This must be an upgrade or other cache will take ownership
-                        missQueue->markInService(mshr->pkt);
+                        missQueue->markInService(mshr->pkt, mshr);
                     }
                     return;
                 }
@@ -454,10 +458,16 @@ Cache<TagStore,Buffering,Coherence>::snoop(Packet * &pkt)
     CacheBlk::State new_state;
     bool satisfy = coherence->handleBusRequest(pkt,blk,mshr, new_state);
     if (satisfy) {
+        DPRINTF(Cache, "Cache snooped a %s request and now supplying data,"
+                "new state is %i\n",
+                pkt->cmdString(), new_state);
+
         tags->handleSnoop(blk, new_state, pkt);
         respondToSnoop(pkt, curTick + hitLatency);
         return;
     }
+    if (blk) DPRINTF(Cache, "Cache snooped a %s request, new state is %i\n",
+                     pkt->cmdString(), new_state);
     tags->handleSnoop(blk, new_state);
 }
 
@@ -603,7 +613,7 @@ Cache<TagStore,Buffering,Coherence>::probe(Packet * &pkt, bool update, CachePort
             // update the cache state and statistics
             if (mshr || !writes.empty()){
                 // Can't handle it, return pktuest unsatisfied.
-                return 0;
+                panic("Atomic access ran into outstanding MSHR's or WB's!");
             }
             if (!pkt->req->isUncacheable()) {
                 // Fetch the cache block to fill
@@ -620,7 +630,9 @@ Cache<TagStore,Buffering,Coherence>::probe(Packet * &pkt, bool update, CachePort
                 lat = memSidePort->sendAtomic(busPkt);
 
                 //Be sure to flip the response to a request for coherence
-                busPkt->makeAtomicResponse();
+                if (busPkt->needsResponse()) {
+                    busPkt->makeAtomicResponse();
+                }
 
 /*		if (!(busPkt->flags & SATISFIED)) {
                     // blocked at a higher level, just return
@@ -655,7 +667,7 @@ Cache<TagStore,Buffering,Coherence>::probe(Packet * &pkt, bool update, CachePort
             hits[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
         } else if (pkt->isWrite()) {
             // Still need to change data in all locations.
-            return otherSidePort->sendAtomic(pkt);
+            otherSidePort->sendFunctional(pkt);
         }
         return curTick + lat;
     }
@@ -673,9 +685,15 @@ Cache<TagStore,Buffering,Coherence>::snoopProbe(PacketPtr &pkt)
         CacheBlk::State new_state = 0;
         bool satisfy = coherence->handleBusRequest(pkt,blk,mshr, new_state);
         if (satisfy) {
+            DPRINTF(Cache, "Cache snooped a %s request and now supplying data,"
+                    "new state is %i\n",
+                    pkt->cmdString(), new_state);
+
             tags->handleSnoop(blk, new_state, pkt);
             return hitLatency;
         }
+        if (blk) DPRINTF(Cache, "Cache snooped a %s request, new state is %i\n",
+                     pkt->cmdString(), new_state);
         tags->handleSnoop(blk, new_state);
         return 0;
 }
diff --git a/src/mem/cache/miss/blocking_buffer.cc b/src/mem/cache/miss/blocking_buffer.cc
index 7a6ea9133..f7aacff89 100644
--- a/src/mem/cache/miss/blocking_buffer.cc
+++ b/src/mem/cache/miss/blocking_buffer.cc
@@ -123,12 +123,12 @@ BlockingBuffer::restoreOrigCmd(Packet * &pkt)
 }
 
 void
-BlockingBuffer::markInService(Packet * &pkt)
+BlockingBuffer::markInService(Packet * &pkt, MSHR* mshr)
 {
     if (!pkt->isCacheFill() && pkt->isWrite()) {
         // Forwarding a write/ writeback, don't need to change
         // the command
-        assert((MSHR*)pkt->senderState == &wb);
+        assert(mshr == &wb);
         cache->clearMasterRequest(Request_WB);
         if (!pkt->needsResponse()) {
             assert(wb.getNumTargets() == 0);
@@ -138,7 +138,7 @@ BlockingBuffer::markInService(Packet * &pkt)
             wb.inService = true;
         }
     } else {
-        assert((MSHR*)pkt->senderState == &miss);
+        assert(mshr == &miss);
         cache->clearMasterRequest(Request_MSHR);
         if (!pkt->needsResponse()) {
             assert(miss.getNumTargets() == 0);
diff --git a/src/mem/cache/miss/blocking_buffer.hh b/src/mem/cache/miss/blocking_buffer.hh
index 641d5a798..f7069696c 100644
--- a/src/mem/cache/miss/blocking_buffer.hh
+++ b/src/mem/cache/miss/blocking_buffer.hh
@@ -152,7 +152,7 @@ public:
      * are successfully sent.
      * @param pkt The request that was sent on the bus.
      */
-    void markInService(Packet * &pkt);
+    void markInService(Packet * &pkt, MSHR* mshr);
 
     /**
      * Frees the resources of the pktuest and unblock the cache.
diff --git a/src/mem/cache/miss/miss_queue.cc b/src/mem/cache/miss/miss_queue.cc
index 273b6587f..bdb7a39c8 100644
--- a/src/mem/cache/miss/miss_queue.cc
+++ b/src/mem/cache/miss/miss_queue.cc
@@ -372,7 +372,7 @@ MissQueue::allocateMiss(Packet * &pkt, int size, Tick time)
 MSHR*
 MissQueue::allocateWrite(Packet * &pkt, int size, Tick time)
 {
-    MSHR* mshr = wb.allocate(pkt,blkSize);
+    MSHR* mshr = wb.allocate(pkt,size);
     mshr->order = order++;
 
 //REMOVING COMPRESSION FOR NOW
@@ -446,7 +446,7 @@ MissQueue::handleMiss(Packet * &pkt, int blkSize, Tick time)
         /**
          * @todo Add write merging here.
          */
-        mshr = allocateWrite(pkt, blkSize, time);
+        mshr = allocateWrite(pkt, pkt->getSize(), time);
         return;
     }
 
@@ -526,9 +526,8 @@ MissQueue::restoreOrigCmd(Packet * &pkt)
 }
 
 void
-MissQueue::markInService(Packet * &pkt)
+MissQueue::markInService(Packet * &pkt, MSHR* mshr)
 {
-    assert(pkt->senderState != 0);
     bool unblock = false;
     BlockedCause cause = NUM_BLOCKED_CAUSES;
 
@@ -540,7 +539,7 @@ MissQueue::markInService(Packet * &pkt)
         // Forwarding a write/ writeback, don't need to change
         // the command
         unblock = wb.isFull();
-        wb.markInService((MSHR*)pkt->senderState);
+        wb.markInService(mshr);
         if (!wb.havePending()){
             cache->clearMasterRequest(Request_WB);
         }
@@ -551,11 +550,11 @@ MissQueue::markInService(Packet * &pkt)
         }
     } else {
         unblock = mq.isFull();
-        mq.markInService((MSHR*)pkt->senderState);
+        mq.markInService(mshr);
         if (!mq.havePending()){
             cache->clearMasterRequest(Request_MSHR);
         }
-        if (((MSHR*)(pkt->senderState))->originalCmd == Packet::HardPFReq) {
+        if (mshr->originalCmd == Packet::HardPFReq) {
             DPRINTF(HWPrefetch, "%s:Marking a HW_PF in service\n",
                     cache->name());
             //Also clear pending if need be
diff --git a/src/mem/cache/miss/miss_queue.hh b/src/mem/cache/miss/miss_queue.hh
index 505d1f90c..179638d2b 100644
--- a/src/mem/cache/miss/miss_queue.hh
+++ b/src/mem/cache/miss/miss_queue.hh
@@ -256,7 +256,7 @@ class MissQueue
      * are successfully sent.
      * @param pkt The request that was sent on the bus.
      */
-    void markInService(Packet * &pkt);
+    void markInService(Packet * &pkt, MSHR* mshr);
 
     /**
      * Collect statistics and free resources of a satisfied pktuest.
diff --git a/src/mem/physical.cc b/src/mem/physical.cc
index 070693442..96d78bd99 100644
--- a/src/mem/physical.cc
+++ b/src/mem/physical.cc
@@ -231,7 +231,7 @@ PhysicalMemory::getPort(const std::string &if_name, int idx)
         port = new MemoryPort(name() + "-port", this);
         return port;
     } else if (if_name == "functional") {
-        /* special port for functional writes at startup. */
+        /* special port for functional writes at startup. And for memtester */
         return new MemoryPort(name() + "-funcport", this);
     } else {
         panic("PhysicalMemory::getPort: unknown port %s requested", if_name);
diff --git a/src/python/m5/objects/MemTest.py b/src/python/m5/objects/MemTest.py
index 97600768f..18aff03f4 100644
--- a/src/python/m5/objects/MemTest.py
+++ b/src/python/m5/objects/MemTest.py
@@ -1,13 +1,12 @@
 from m5.SimObject import SimObject
 from m5.params import *
+from m5.proxy import *
+from m5 import build_env
+
 class MemTest(SimObject):
     type = 'MemTest'
-    cache = Param.BaseCache("L1 cache")
-    check_mem = Param.FunctionalMemory("check memory")
-    main_mem = Param.FunctionalMemory("hierarchical memory")
     max_loads = Param.Counter("number of loads to execute")
     memory_size = Param.Int(65536, "memory size")
-    percent_copies = Param.Percent(0, "target copy percentage")
     percent_dest_unaligned = Param.Percent(50,
         "percent of copy dest address that are unaligned")
     percent_reads = Param.Percent(65, "target read percentage")
@@ -18,3 +17,6 @@ class MemTest(SimObject):
     progress_interval = Param.Counter(1000000,
         "progress report interval (in accesses)")
     trace_addr = Param.Addr(0, "address to trace")
+
+    test = Port("Port to the memory system to test")
+    functional = Port("Port to the functional memory used for verification")
diff --git a/src/python/m5/objects/PhysicalMemory.py b/src/python/m5/objects/PhysicalMemory.py
index dd3ffd651..4e097543d 100644
--- a/src/python/m5/objects/PhysicalMemory.py
+++ b/src/python/m5/objects/PhysicalMemory.py
@@ -5,6 +5,7 @@ from MemObject import *
 class PhysicalMemory(MemObject):
     type = 'PhysicalMemory'
     port = Port("the access port")
+    functional = Port("Functional Access Port")
     range = Param.AddrRange(AddrRange('128MB'), "Device Address")
     file = Param.String('', "memory mapped file")
     latency = Param.Latency(Parent.clock, "latency of an access")
diff --git a/src/python/m5/params.py b/src/python/m5/params.py
index cbbd23004..93d784181 100644
--- a/src/python/m5/params.py
+++ b/src/python/m5/params.py
@@ -804,7 +804,7 @@ class PortRef(object):
         newRef.simobj = simobj
         assert(isSimObject(newRef.simobj))
         if self.peer and not proxy.isproxy(self.peer):
-            peerObj = memo[self.peer.simobj]
+            peerObj = self.peer.simobj(_memo=memo)
             newRef.peer = self.peer.clone(peerObj, memo)
             assert(not isinstance(newRef.peer, VectorPortRef))
         return newRef
diff --git a/src/python/m5/proxy.py b/src/python/m5/proxy.py
index 7ebc0ae19..e539f14ee 100644
--- a/src/python/m5/proxy.py
+++ b/src/python/m5/proxy.py
@@ -33,6 +33,8 @@
 #
 #####################################################################
 
+import copy
+
 class BaseProxy(object):
     def __init__(self, search_self, search_up):
         self._search_self = search_self
@@ -129,15 +131,22 @@ class AttrProxy(BaseProxy):
             return super(AttrProxy, self).__getattr__(self, attr)
         if hasattr(self, '_pdesc'):
             raise AttributeError, "Attribute reference on bound proxy"
-        self._modifiers.append(attr)
-        return self
+        # Return a copy of self rather than modifying self in place
+        # since self could be an indirect reference via a variable or
+        # parameter
+        new_self = copy.deepcopy(self)
+        new_self._modifiers.append(attr)
+        return new_self
 
     # support indexing on proxies (e.g., Self.cpu[0])
     def __getitem__(self, key):
         if not isinstance(key, int):
             raise TypeError, "Proxy object requires integer index"
-        self._modifiers.append(key)
-        return self
+        if hasattr(self, '_pdesc'):
+            raise AttributeError, "Index operation on bound proxy"
+        new_self = copy.deepcopy(self)
+        new_self._modifiers.append(key)
+        return new_self
 
     def find(self, obj):
         try: