cpu: Add support for instructions that zero cache lines.

src/cpu/base_dyn_inst_impl.hh

@@ -90,6 +90,7 @@ BaseDynInst<Impl>::initVars()
     effAddr = 0;
     physEffAddr = 0;
     readyRegs = 0;
+    memReqFlags = 0;
 
     status.reset();
 

src/cpu/o3/lsq_unit.hh

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2012 ARM Limited
+ * Copyright (c) 2012-2013 ARM Limited
  * All rights reserved
  *
  * The license below extends only to copyright in the software and shall
@@ -361,7 +361,7 @@ class LSQUnit {
         /** Constructs a store queue entry for a given instruction. */
         SQEntry(DynInstPtr &_inst)
             : inst(_inst), req(NULL), sreqLow(NULL), sreqHigh(NULL), size(0),
-              isSplit(0), canWB(0), committed(0), completed(0)
+              isSplit(0), canWB(0), committed(0), completed(0), isAllZeros(0)
         {
             std::memset(data, 0, sizeof(data));
         }
@@ -384,6 +384,11 @@ class LSQUnit {
         bool committed;
         /** Whether or not the store is completed. */
         bool completed;
+        /** Does this request write all zeros and thus doesn't
+         * have any data attached to it. Used for cache block zero
+         * style instructs (ARM DC ZVA; ALPHA WH64)
+         */
+        bool isAllZeros;
     };
 
   private:
@@ -691,13 +696,18 @@ LSQUnit<Impl>::read(Request *req, Request *sreqLow, Request *sreqHigh,
             // Get shift amount for offset into the store's data.
             int shift_amt = req->getVaddr() - storeQueue[store_idx].inst->effAddr;
 
-            memcpy(data, storeQueue[store_idx].data + shift_amt,
+            if (storeQueue[store_idx].isAllZeros)
+                memset(data, 0, req->getSize());
+            else
+                memcpy(data, storeQueue[store_idx].data + shift_amt,
                    req->getSize());
 
             assert(!load_inst->memData);
-            load_inst->memData = new uint8_t[64];
-
-            memcpy(load_inst->memData,
+            load_inst->memData = new uint8_t[req->getSize()];
+            if (storeQueue[store_idx].isAllZeros)
+                memset(load_inst->memData, 0, req->getSize());
+            else
+                memcpy(load_inst->memData,
                     storeQueue[store_idx].data + shift_amt, req->getSize());
 
             DPRINTF(LSQUnit, "Forwarding from store idx %i to load to "
@@ -777,7 +787,7 @@ LSQUnit<Impl>::read(Request *req, Request *sreqLow, Request *sreqHigh,
             load_inst->seqNum, load_inst->pcState());
 
     assert(!load_inst->memData);
-    load_inst->memData = new uint8_t[64];
+    load_inst->memData = new uint8_t[req->getSize()];
 
     ++usedPorts;
 
@@ -916,7 +926,9 @@ LSQUnit<Impl>::write(Request *req, Request *sreqLow, Request *sreqHigh,
     storeQueue[store_idx].sreqHigh = sreqHigh;
     unsigned size = req->getSize();
     storeQueue[store_idx].size = size;
-    assert(size <= sizeof(storeQueue[store_idx].data));
+    storeQueue[store_idx].isAllZeros = req->getFlags() & Request::CACHE_BLOCK_ZERO;
+    assert(size <= sizeof(storeQueue[store_idx].data) ||
+            (req->getFlags() & Request::CACHE_BLOCK_ZERO));
 
     // Split stores can only occur in ISAs with unaligned memory accesses.  If
     // a store request has been split, sreqLow and sreqHigh will be non-null.
@@ -924,7 +936,8 @@ LSQUnit<Impl>::write(Request *req, Request *sreqLow, Request *sreqHigh,
         storeQueue[store_idx].isSplit = true;
     }
 
-    memcpy(storeQueue[store_idx].data, data, size);
+    if (!(req->getFlags() & Request::CACHE_BLOCK_ZERO))
+        memcpy(storeQueue[store_idx].data, data, size);
 
     // This function only writes the data to the store queue, so no fault
     // can happen here.

src/cpu/o3/lsq_unit_impl.hh

@@ -816,9 +816,12 @@ LSQUnit<Impl>::writebackStores()
         storeQueue[storeWBIdx].committed = true;
 
         assert(!inst->memData);
-        inst->memData = new uint8_t[64];
+        inst->memData = new uint8_t[req->getSize()];
 
-        memcpy(inst->memData, storeQueue[storeWBIdx].data, req->getSize());
+        if (storeQueue[storeWBIdx].isAllZeros)
+            memset(inst->memData, 0, req->getSize());
+        else
+            memcpy(inst->memData, storeQueue[storeWBIdx].data, req->getSize());
 
         MemCmd command =
             req->isSwap() ? MemCmd::SwapReq :

src/cpu/simple/atomic.cc

@@ -399,6 +399,16 @@ Fault
 AtomicSimpleCPU::writeMem(uint8_t *data, unsigned size,
                           Addr addr, unsigned flags, uint64_t *res)
 {
+
+    static uint8_t zero_array[64] = {};
+
+    if (data == NULL) {
+        assert(size <= 64);
+        assert(flags & Request::CACHE_BLOCK_ZERO);
+        // This must be a cache block cleaning request
+        data = zero_array;
+    }
+
     // use the CPU's statically allocated write request and packet objects
     Request *req = &data_write_req;
 

src/cpu/simple/timing.cc

@@ -472,14 +472,20 @@ TimingSimpleCPU::writeMem(uint8_t *data, unsigned size,
                           Addr addr, unsigned flags, uint64_t *res)
 {
     uint8_t *newData = new uint8_t[size];
-    memcpy(newData, data, size);
-
     const int asid = 0;
     const ThreadID tid = 0;
     const Addr pc = thread->instAddr();
     unsigned block_size = cacheLineSize();
     BaseTLB::Mode mode = BaseTLB::Write;
 
+    if (data == NULL) {
+        assert(flags & Request::CACHE_BLOCK_ZERO);
+        // This must be a cache block cleaning request
+        memset(newData, 0, size);
+    } else {
+        memcpy(newData, data, size);
+    }
+
     if (traceData) {
         traceData->setAddr(addr);
     }

src/mem/request.hh

@@ -114,6 +114,11 @@ class Request
     /** This request is made in privileged mode. */
     static const FlagsType PRIVILEGED                   = 0x00008000;
 
+    /** This is a write that is targeted and zeroing an entire cache block.
+     * There is no need for a read/modify/write
+     */
+    static const FlagsType CACHE_BLOCK_ZERO            = 0x00010000;
+
     /** The request should not cause a memory access. */
     static const FlagsType NO_ACCESS                   = 0x00080000;
     /** This request will lock or unlock the accessed memory. When used with