Merge ktlim@zizzer:/bk/newmem

into  zamp.eecs.umich.edu:/z/ktlim2/clean/o3-merge/newmem

src/cpu/memtest/memtest.cc:
src/cpu/memtest/memtest.hh:
src/cpu/simple/timing.hh:
tests/configs/o3-timing-mp.py:
    Hand merge.

--HG--
extra : convert_revision : a58cc439eb5e8f900d175ed8b5a85b6c8723e558

4 files changed, 126 insertions, 72 deletions

diff --git a/src/cpu/simple/atomic.cc b/src/cpu/simple/atomic.cc
index 88698bfee..490be20ae 100644
--- a/src/cpu/simple/atomic.cc
+++ b/src/cpu/simple/atomic.cc
@@ -28,6 +28,7 @@
  * Authors: Steve Reinhardt
  */
 
+#include "arch/locked_mem.hh"
 #include "arch/utility.hh"
 #include "cpu/exetrace.hh"
 #include "cpu/simple/atomic.hh"
@@ -93,7 +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;
 }
 
@@ -107,7 +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
@@ -133,20 +135,19 @@ AtomicSimpleCPU::AtomicSimpleCPU(Params *p)
 {
     _status = Idle;
 
-    // @todo fix me and get the real cpu id & thread number!!!
     ifetch_req = new Request();
-    ifetch_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE
+    ifetch_req->setThreadContext(p->cpu_id, 0); // Add thread ID if we add MT
     ifetch_pkt = new Packet(ifetch_req, Packet::ReadReq, Packet::Broadcast);
     ifetch_pkt->dataStatic(&inst);
 
     data_read_req = new Request();
-    data_read_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE
+    data_read_req->setThreadContext(p->cpu_id, 0); // Add thread ID here too
     data_read_pkt = new Packet(data_read_req, Packet::ReadReq,
                                Packet::Broadcast);
     data_read_pkt->dataStatic(&dataReg);
 
     data_write_req = new Request();
-    data_write_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE
+    data_write_req->setThreadContext(p->cpu_id, 0); // Add thread ID here too
     data_write_pkt = new Packet(data_write_req, Packet::WriteReq,
                                 Packet::Broadcast);
 }
@@ -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));
 }
@@ -253,29 +257,36 @@ template <class T>
 Fault
 AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags)
 {
-    data_read_req->setVirt(0, addr, sizeof(T), flags, thread->readPC());
+    // use the CPU's statically allocated read request and packet objects
+    Request *req = data_read_req;
+    Packet  *pkt = data_read_pkt;
+
+    req->setVirt(0, addr, sizeof(T), flags, thread->readPC());
 
     if (traceData) {
         traceData->setAddr(addr);
     }
 
     // translate to physical address
-    Fault fault = thread->translateDataReadReq(data_read_req);
+    Fault fault = thread->translateDataReadReq(req);
 
     // Now do the access.
     if (fault == NoFault) {
-        data_read_pkt->reinitFromRequest();
+        pkt->reinitFromRequest();
 
-        dcache_latency = dcachePort.sendAtomic(data_read_pkt);
+        dcache_latency = dcachePort.sendAtomic(pkt);
         dcache_access = true;
 
-        assert(data_read_pkt->result == Packet::Success);
-        data = data_read_pkt->get<T>();
+        assert(pkt->result == Packet::Success);
+        data = pkt->get<T>();
 
+        if (req->isLocked()) {
+            TheISA::handleLockedRead(thread, req);
+        }
     }
 
     // This will need a new way to tell if it has a dcache attached.
-    if (data_read_req->getFlags() & UNCACHEABLE)
+    if (req->isUncacheable())
         recordEvent("Uncached Read");
 
     return fault;
@@ -328,33 +339,52 @@ template <class T>
 Fault
 AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res)
 {
-    data_write_req->setVirt(0, addr, sizeof(T), flags, thread->readPC());
+    // use the CPU's statically allocated write request and packet objects
+    Request *req = data_write_req;
+    Packet  *pkt = data_write_pkt;
+
+    req->setVirt(0, addr, sizeof(T), flags, thread->readPC());
 
     if (traceData) {
         traceData->setAddr(addr);
     }
 
     // translate to physical address
-    Fault fault = thread->translateDataWriteReq(data_write_req);
+    Fault fault = thread->translateDataWriteReq(req);
 
     // Now do the access.
     if (fault == NoFault) {
-        data = htog(data);
-        data_write_pkt->reinitFromRequest();
-        data_write_pkt->dataStatic(&data);
+        bool do_access = true;  // flag to suppress cache access
 
-        dcache_latency = dcachePort.sendAtomic(data_write_pkt);
-        dcache_access = true;
+        if (req->isLocked()) {
+            do_access = TheISA::handleLockedWrite(thread, req);
+        }
+
+        if (do_access) {
+            data = htog(data);
+            pkt->reinitFromRequest();
+            pkt->dataStatic(&data);
 
-        assert(data_write_pkt->result == Packet::Success);
+            dcache_latency = dcachePort.sendAtomic(pkt);
+            dcache_access = true;
 
-        if (res && data_write_req->getFlags() & LOCKED) {
-            *res = data_write_req->getScResult();
+            assert(pkt->result == Packet::Success);
+        }
+
+        if (req->isLocked()) {
+            uint64_t scResult = req->getScResult();
+            if (scResult != 0) {
+                // clear failure counter
+                thread->setStCondFailures(0);
+            }
+            if (res) {
+                *res = req->getScResult();
+            }
         }
     }
 
     // This will need a new way to tell if it's hooked up to a cache or not.
-    if (data_write_req->getFlags() & UNCACHEABLE)
+    if (req->isUncacheable())
         recordEvent("Uncached Write");
 
     // If the write needs to have a fault on the access, consider calling
@@ -467,11 +497,11 @@ BEGIN_DECLARE_SIM_OBJECT_PARAMS(AtomicSimpleCPU)
     Param<Tick> progress_interval;
     SimObjectParam<MemObject *> mem;
     SimObjectParam<System *> system;
+    Param<int> cpu_id;
 
 #if FULL_SYSTEM
     SimObjectParam<AlphaITB *> itb;
     SimObjectParam<AlphaDTB *> dtb;
-    Param<int> cpu_id;
     Param<Tick> profile;
 #else
     SimObjectParam<Process *> workload;
@@ -500,11 +530,11 @@ BEGIN_INIT_SIM_OBJECT_PARAMS(AtomicSimpleCPU)
     INIT_PARAM(progress_interval, "Progress interval"),
     INIT_PARAM(mem, "memory"),
     INIT_PARAM(system, "system object"),
+    INIT_PARAM(cpu_id, "processor ID"),
 
 #if FULL_SYSTEM
     INIT_PARAM(itb, "Instruction TLB"),
     INIT_PARAM(dtb, "Data TLB"),
-    INIT_PARAM(cpu_id, "processor ID"),
     INIT_PARAM(profile, ""),
 #else
     INIT_PARAM(workload, "processes to run"),
@@ -538,11 +568,11 @@ CREATE_SIM_OBJECT(AtomicSimpleCPU)
     params->simulate_stalls = simulate_stalls;
     params->mem = mem;
     params->system = system;
+    params->cpu_id = cpu_id;
 
 #if FULL_SYSTEM
     params->itb = itb;
     params->dtb = dtb;
-    params->cpu_id = cpu_id;
     params->profile = profile;
 #else
     params->process = workload;
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 9bed5dab1..33f673cbc 100644
--- a/src/cpu/simple/timing.cc
+++ b/src/cpu/simple/timing.cc
@@ -28,6 +28,7 @@
  * Authors: Steve Reinhardt
  */
 
+#include "arch/locked_mem.hh"
 #include "arch/utility.hh"
 #include "cpu/exetrace.hh"
 #include "cpu/simple/timing.hh"
@@ -73,7 +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
@@ -94,7 +96,8 @@ TimingSimpleCPU::CpuPort::TickEvent::schedule(Packet *_pkt, Tick t)
 }
 
 TimingSimpleCPU::TimingSimpleCPU(Params *p)
-    : BaseSimpleCPU(p), icachePort(this, p->clock), dcachePort(this, p->clock)
+    : BaseSimpleCPU(p), icachePort(this, p->clock), dcachePort(this, p->clock),
+      cpu_id(p->cpu_id)
 {
     _status = Idle;
     ifetch_pkt = dcache_pkt = NULL;
@@ -251,35 +254,35 @@ template <class T>
 Fault
 TimingSimpleCPU::read(Addr addr, T &data, unsigned flags)
 {
-    // need to fill in CPU & thread IDs here
-    Request *data_read_req = new Request();
-    data_read_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE
-    data_read_req->setVirt(0, addr, sizeof(T), flags, thread->readPC());
+    Request *req =
+        new Request(/* asid */ 0, addr, sizeof(T), flags, thread->readPC(),
+                    cpu_id, /* thread ID */ 0);
 
     if (traceData) {
-        traceData->setAddr(data_read_req->getVaddr());
+        traceData->setAddr(req->getVaddr());
     }
 
    // translate to physical address
-    Fault fault = thread->translateDataReadReq(data_read_req);
+    Fault fault = thread->translateDataReadReq(req);
 
     // Now do the access.
     if (fault == NoFault) {
-        Packet *data_read_pkt =
-            new Packet(data_read_req, Packet::ReadReq, Packet::Broadcast);
-        data_read_pkt->dataDynamic<T>(new T);
+        Packet *pkt =
+            new Packet(req, Packet::ReadReq, Packet::Broadcast);
+        pkt->dataDynamic<T>(new T);
 
-        if (!dcachePort.sendTiming(data_read_pkt)) {
+        if (!dcachePort.sendTiming(pkt)) {
             _status = DcacheRetry;
-            dcache_pkt = data_read_pkt;
+            dcache_pkt = pkt;
         } else {
             _status = DcacheWaitResponse;
+            // memory system takes ownership of packet
             dcache_pkt = NULL;
         }
     }
 
     // This will need a new way to tell if it has a dcache attached.
-    if (data_read_req->getFlags() & UNCACHEABLE)
+    if (req->isUncacheable())
         recordEvent("Uncached Read");
 
     return fault;
@@ -332,31 +335,39 @@ template <class T>
 Fault
 TimingSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res)
 {
-    // need to fill in CPU & thread IDs here
-    Request *data_write_req = new Request();
-    data_write_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE
-    data_write_req->setVirt(0, addr, sizeof(T), flags, thread->readPC());
+    Request *req =
+        new Request(/* asid */ 0, addr, sizeof(T), flags, thread->readPC(),
+                    cpu_id, /* thread ID */ 0);
 
     // translate to physical address
-    Fault fault = thread->translateDataWriteReq(data_write_req);
+    Fault fault = thread->translateDataWriteReq(req);
+
     // Now do the access.
     if (fault == NoFault) {
-        Packet *data_write_pkt =
-            new Packet(data_write_req, Packet::WriteReq, Packet::Broadcast);
-        data_write_pkt->allocate();
-        data_write_pkt->set(data);
+        assert(dcache_pkt == NULL);
+        dcache_pkt = new Packet(req, Packet::WriteReq, Packet::Broadcast);
+        dcache_pkt->allocate();
+        dcache_pkt->set(data);
 
-        if (!dcachePort.sendTiming(data_write_pkt)) {
-            _status = DcacheRetry;
-            dcache_pkt = data_write_pkt;
-        } else {
-            _status = DcacheWaitResponse;
-            dcache_pkt = NULL;
+        bool do_access = true;  // flag to suppress cache access
+
+        if (req->isLocked()) {
+            do_access = TheISA::handleLockedWrite(thread, req);
+        }
+
+        if (do_access) {
+            if (!dcachePort.sendTiming(dcache_pkt)) {
+                _status = DcacheRetry;
+            } else {
+                _status = DcacheWaitResponse;
+                // memory system takes ownership of packet
+                dcache_pkt = NULL;
+            }
         }
     }
 
     // This will need a new way to tell if it's hooked up to a cache or not.
-    if (data_write_req->getFlags() & UNCACHEABLE)
+    if (req->isUncacheable())
         recordEvent("Uncached Write");
 
     // If the write needs to have a fault on the access, consider calling
@@ -416,9 +427,8 @@ TimingSimpleCPU::fetch()
 {
     checkForInterrupts();
 
-    // need to fill in CPU & thread IDs here
     Request *ifetch_req = new Request();
-    ifetch_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE
+    ifetch_req->setThreadContext(cpu_id, /* thread ID */ 0);
     Fault fault = setupFetchRequest(ifetch_req);
 
     ifetch_pkt = new Packet(ifetch_req, Packet::ReadReq, Packet::Broadcast);
@@ -483,12 +493,20 @@ TimingSimpleCPU::completeIfetch(Packet *pkt)
     if (curStaticInst->isMemRef() && !curStaticInst->isDataPrefetch()) {
         // load or store: just send to dcache
         Fault fault = curStaticInst->initiateAcc(this, traceData);
-        if (fault == NoFault) {
-            // successfully initiated access: instruction will
-            // complete in dcache response callback
-            assert(_status == DcacheWaitResponse);
+        if (_status != Running) {
+            // instruction will complete in dcache response callback
+            assert(_status == DcacheWaitResponse || _status == DcacheRetry);
+            assert(fault == NoFault);
         } else {
-            // fault: complete now to invoke fault handler
+            if (fault == NoFault) {
+                // early fail on store conditional: complete now
+                assert(dcache_pkt != NULL);
+                fault = curStaticInst->completeAcc(dcache_pkt, this,
+                                                   traceData);
+                delete dcache_pkt->req;
+                delete dcache_pkt;
+                dcache_pkt = NULL;
+            }
             postExecute();
             advanceInst(fault);
         }
@@ -509,8 +527,7 @@ TimingSimpleCPU::IcachePort::ITickEvent::process()
 bool
 TimingSimpleCPU::IcachePort::recvTiming(Packet *pkt)
 {
-    // These next few lines could be replaced with something faster
-    // who knows what though
+    // delay processing of returned data until next CPU clock edge
     Tick time = pkt->req->getTime();
     while (time < curTick)
         time += lat;
@@ -551,6 +568,10 @@ TimingSimpleCPU::completeDataAccess(Packet *pkt)
 
     Fault fault = curStaticInst->completeAcc(pkt, this, traceData);
 
+    if (pkt->isRead() && pkt->req->isLocked()) {
+        TheISA::handleLockedRead(thread, pkt->req);
+    }
+
     delete pkt->req;
     delete pkt;
 
@@ -578,6 +599,7 @@ TimingSimpleCPU::completeDrain()
 bool
 TimingSimpleCPU::DcachePort::recvTiming(Packet *pkt)
 {
+    // delay processing of returned data until next CPU clock edge
     Tick time = pkt->req->getTime();
     while (time < curTick)
         time += lat;
@@ -606,6 +628,7 @@ TimingSimpleCPU::DcachePort::recvRetry()
     Packet *tmp = cpu->dcache_pkt;
     if (sendTiming(tmp)) {
         cpu->_status = DcacheWaitResponse;
+        // memory system takes ownership of packet
         cpu->dcache_pkt = NULL;
     }
 }
@@ -624,11 +647,11 @@ BEGIN_DECLARE_SIM_OBJECT_PARAMS(TimingSimpleCPU)
     Param<Tick> progress_interval;
     SimObjectParam<MemObject *> mem;
     SimObjectParam<System *> system;
+    Param<int> cpu_id;
 
 #if FULL_SYSTEM
     SimObjectParam<AlphaITB *> itb;
     SimObjectParam<AlphaDTB *> dtb;
-    Param<int> cpu_id;
     Param<Tick> profile;
 #else
     SimObjectParam<Process *> workload;
@@ -657,11 +680,11 @@ BEGIN_INIT_SIM_OBJECT_PARAMS(TimingSimpleCPU)
     INIT_PARAM(progress_interval, "Progress interval"),
     INIT_PARAM(mem, "memory"),
     INIT_PARAM(system, "system object"),
+    INIT_PARAM(cpu_id, "processor ID"),
 
 #if FULL_SYSTEM
     INIT_PARAM(itb, "Instruction TLB"),
     INIT_PARAM(dtb, "Data TLB"),
-    INIT_PARAM(cpu_id, "processor ID"),
     INIT_PARAM(profile, ""),
 #else
     INIT_PARAM(workload, "processes to run"),
@@ -693,11 +716,11 @@ CREATE_SIM_OBJECT(TimingSimpleCPU)
     params->functionTraceStart = function_trace_start;
     params->mem = mem;
     params->system = system;
+    params->cpu_id = cpu_id;
 
 #if FULL_SYSTEM
     params->itb = itb;
     params->dtb = dtb;
-    params->cpu_id = cpu_id;
     params->profile = profile;
 #else
     params->process = workload;
diff --git a/src/cpu/simple/timing.hh b/src/cpu/simple/timing.hh
index 8a20d1cfe..988ddeded 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
         {
@@ -166,6 +166,7 @@ class TimingSimpleCPU : public BaseSimpleCPU
     Packet *ifetch_pkt;
     Packet *dcache_pkt;
 
+    int cpu_id;
     Tick previousTick;
 
   public: