Implement Alpha LL/SC support for SimpleCPU (Atomic & Timing)

and PhysicalMemory.  *No* support for caches or O3CPU.
Note that properly setting cpu_id on all CPUs is now required
for correct operation.

src/arch/SConscript:
src/base/traceflags.py:
src/cpu/base.hh:
src/cpu/simple/atomic.cc:
src/cpu/simple/timing.cc:
src/cpu/simple/timing.hh:
src/mem/physical.cc:
src/mem/physical.hh:
src/mem/request.hh:
src/python/m5/objects/BaseCPU.py:
tests/configs/simple-atomic.py:
tests/configs/simple-timing.py:
tests/configs/tsunami-simple-atomic-dual.py:
tests/configs/tsunami-simple-atomic.py:
tests/configs/tsunami-simple-timing-dual.py:
tests/configs/tsunami-simple-timing.py:
    Implement Alpha LL/SC support for SimpleCPU (Atomic & Timing)
    and PhysicalMemory.  *No* support for caches or O3CPU.

--HG--
extra : convert_revision : 6ce982d44924cc477e049b9adf359818908e72be

3 files changed, 154 insertions, 9 deletions

diff --git a/src/mem/physical.cc b/src/mem/physical.cc
index 8fea733ec..23b1d5ffc 100644
--- a/src/mem/physical.cc
+++ b/src/mem/physical.cc
@@ -110,6 +110,88 @@ PhysicalMemory::calculateLatency(Packet *pkt)
     return lat;
 }
 
+
+
+// Add load-locked to tracking list.  Should only be called if the
+// operation is a load and the LOCKED flag is set.
+void
+PhysicalMemory::trackLoadLocked(Request *req)
+{
+    Addr paddr = LockedAddr::mask(req->getPaddr());
+
+    // first we check if we already have a locked addr for this
+    // xc.  Since each xc only gets one, we just update the
+    // existing record with the new address.
+    list<LockedAddr>::iterator i;
+
+    for (i = lockedAddrList.begin(); i != lockedAddrList.end(); ++i) {
+        if (i->matchesContext(req)) {
+            DPRINTF(LLSC, "Modifying lock record: cpu %d thread %d addr %#x\n",
+                    req->getCpuNum(), req->getThreadNum(), paddr);
+            i->addr = paddr;
+            return;
+        }
+    }
+
+    // no record for this xc: need to allocate a new one
+    DPRINTF(LLSC, "Adding lock record: cpu %d thread %d addr %#x\n",
+            req->getCpuNum(), req->getThreadNum(), paddr);
+    lockedAddrList.push_front(LockedAddr(req));
+}
+
+
+// Called on *writes* only... both regular stores and
+// store-conditional operations.  Check for conventional stores which
+// conflict with locked addresses, and for success/failure of store
+// conditionals.
+bool
+PhysicalMemory::checkLockedAddrList(Request *req)
+{
+    Addr paddr = LockedAddr::mask(req->getPaddr());
+    bool isLocked = req->isLocked();
+
+    // Initialize return value.  Non-conditional stores always
+    // succeed.  Assume conditional stores will fail until proven
+    // otherwise.
+    bool success = !isLocked;
+
+    // Iterate over list.  Note that there could be multiple matching
+    // records, as more than one context could have done a load locked
+    // to this location.
+    list<LockedAddr>::iterator i = lockedAddrList.begin();
+
+    while (i != lockedAddrList.end()) {
+
+        if (i->addr == paddr) {
+            // we have a matching address
+
+            if (isLocked && i->matchesContext(req)) {
+                // it's a store conditional, and as far as the memory
+                // system can tell, the requesting context's lock is
+                // still valid.
+                DPRINTF(LLSC, "StCond success: cpu %d thread %d addr %#x\n",
+                        req->getCpuNum(), req->getThreadNum(), paddr);
+                success = true;
+            }
+
+            // Get rid of our record of this lock and advance to next
+            DPRINTF(LLSC, "Erasing lock record: cpu %d thread %d addr %#x\n",
+                    i->cpuNum, i->threadNum, paddr);
+            i = lockedAddrList.erase(i);
+        }
+        else {
+            // no match: advance to next record
+            ++i;
+        }
+    }
+
+    if (isLocked) {
+        req->setScResult(success ? 1 : 0);
+    }
+
+    return success;
+}
+
 void
 PhysicalMemory::doFunctionalAccess(Packet *pkt)
 {
@@ -117,18 +199,17 @@ PhysicalMemory::doFunctionalAccess(Packet *pkt)
 
     switch (pkt->cmd) {
       case Packet::ReadReq:
+        if (pkt->req->isLocked()) {
+            trackLoadLocked(pkt->req);
+        }
         memcpy(pkt->getPtr<uint8_t>(),
                pmemAddr + pkt->getAddr() - params()->addrRange.start,
                pkt->getSize());
         break;
       case Packet::WriteReq:
-        memcpy(pmemAddr + pkt->getAddr() - params()->addrRange.start,
-               pkt->getPtr<uint8_t>(),
-               pkt->getSize());
-        // temporary hack: will need to add real LL/SC implementation
-        // for cacheless systems later.
-        if (pkt->req->getFlags() & LOCKED) {
-            pkt->req->setScResult(1);
+        if (writeOK(pkt->req)) {
+            memcpy(pmemAddr + pkt->getAddr() - params()->addrRange.start,
+                   pkt->getPtr<uint8_t>(), pkt->getSize());
         }
         break;
       default:
diff --git a/src/mem/physical.hh b/src/mem/physical.hh
index 02308b2ef..97bea2ec4 100644
--- a/src/mem/physical.hh
+++ b/src/mem/physical.hh
@@ -78,6 +78,68 @@ class PhysicalMemory : public MemObject
     const PhysicalMemory &operator=(const PhysicalMemory &specmem);
 
   protected:
+
+    class LockedAddr {
+      public:
+        // on alpha, minimum LL/SC granularity is 16 bytes, so lower
+        // bits need to masked off.
+        static const Addr Addr_Mask = 0xf;
+
+        static Addr mask(Addr paddr) { return (paddr & ~Addr_Mask); }
+
+        Addr addr; 	// locked address
+        int cpuNum;	// locking CPU
+        int threadNum;	// locking thread ID within CPU
+
+        // check for matching execution context
+        bool matchesContext(Request *req)
+        {
+            return (cpuNum == req->getCpuNum() &&
+                    threadNum == req->getThreadNum());
+        }
+
+        LockedAddr(Request *req)
+            : addr(mask(req->getPaddr())),
+              cpuNum(req->getCpuNum()),
+              threadNum(req->getThreadNum())
+        {
+        }
+    };
+
+    std::list<LockedAddr> lockedAddrList;
+
+    // helper function for checkLockedAddrs(): we really want to
+    // inline a quick check for an empty locked addr list (hopefully
+    // the common case), and do the full list search (if necessary) in
+    // this out-of-line function
+    bool checkLockedAddrList(Request *req);
+
+    // Record the address of a load-locked operation so that we can
+    // clear the execution context's lock flag if a matching store is
+    // performed
+    void trackLoadLocked(Request *req);
+
+    // Compare a store address with any locked addresses so we can
+    // clear the lock flag appropriately.  Return value set to 'false'
+    // if store operation should be suppressed (because it was a
+    // conditional store and the address was no longer locked by the
+    // requesting execution context), 'true' otherwise.  Note that
+    // this method must be called on *all* stores since even
+    // non-conditional stores must clear any matching lock addresses.
+    bool writeOK(Request *req) {
+        if (lockedAddrList.empty()) {
+            // no locked addrs: nothing to check, store_conditional fails
+            bool isLocked = req->isLocked();
+            if (isLocked) {
+                req->setScResult(0);
+            }
+            return !isLocked; // only do write if not an sc
+        } else {
+            // iterate over list...
+            return checkLockedAddrList(req);
+        }
+    }
+
     uint8_t *pmemAddr;
     MemoryPort *port;
     int pagePtr;
diff --git a/src/mem/request.hh b/src/mem/request.hh
index 6acd7526c..e54984fcd 100644
--- a/src/mem/request.hh
+++ b/src/mem/request.hh
@@ -232,9 +232,11 @@ class Request
     Addr getPC() { assert(validPC); return pc; }
 
     /** Accessor Function to Check Cacheability. */
-    bool isUncacheable() { return getFlags() & UNCACHEABLE; }
+    bool isUncacheable() { return (getFlags() & UNCACHEABLE) != 0; }
 
-    bool isInstRead() { return getFlags() & INST_READ; }
+    bool isInstRead() { return (getFlags() & INST_READ) != 0; }
+
+    bool isLocked() { return (getFlags() & LOCKED) != 0; }
 
     friend class Packet;
 };