diff --git a/src/mem/ruby/protocol/chi/CHI-cache-actions.sm b/src/mem/ruby/protocol/chi/CHI-cache-actions.sm
index 5d614dcc24..5fa87420d8 100644
--- a/src/mem/ruby/protocol/chi/CHI-cache-actions.sm
+++ b/src/mem/ruby/protocol/chi/CHI-cache-actions.sm
@@ -823,7 +823,8 @@ action(Initiate_AtomicReturn_I, desc="") {
     tbe.atomic_to_be_done := true;
   } else if ((policy_type == 1) || // UNIQUE NEAR
              (policy_type == 2)) { // PRESENT NEAR
-    tbe.actions.push(Event:SendAtomicReturn_NoWait);
+    tbe.actions.push(Event:SendAtomicReturn);
+    tbe.actions.push(Event:SendARData);
     tbe.dataToBeInvalid := true;
     tbe.doCacheFill := false;
     tbe.atomic_to_be_done := false;
@@ -860,7 +861,8 @@ action(Initiate_AtomicReturn_SD, desc="") {
     tbe.actions.push(Event:TagArrayWrite);
     tbe.atomic_to_be_done := true;
   } else if (policy_type == 1) { // UNIQUE NEAR
-    tbe.actions.push(Event:SendAtomicReturn_NoWait);
+    tbe.actions.push(Event:SendAtomicReturn);
+    tbe.actions.push(Event:SendARData);
     tbe.dataToBeInvalid := true;
     tbe.doCacheFill := false;
     tbe.atomic_to_be_done := false;
@@ -897,7 +899,8 @@ action(Initiate_AtomicReturn_SC, desc="") {
     tbe.actions.push(Event:TagArrayWrite);
     tbe.atomic_to_be_done := true;
   } else if (policy_type == 1) { // UNIQUE NEAR
-    tbe.actions.push(Event:SendAtomicReturn_NoWait);
+    tbe.actions.push(Event:SendAtomicReturn);
+    tbe.actions.push(Event:SendARData);
     tbe.dataToBeInvalid := true;
     tbe.doCacheFill := false;
     tbe.atomic_to_be_done := false;
@@ -1049,7 +1052,7 @@ action(Initiate_AtomicReturn_LocalWrite, desc="") {
   }
   tbe.actions.push(Event:SendDBIDResp_AR);
   tbe.actions.pushNB(Event:WriteFEPipe);
-  tbe.actions.pushNB(Event:SendCompData_AR);
+  tbe.actions.push(Event:SendCompData_AR);
   tbe.actions.push(Event:WriteFEPipe);
   tbe.actions.push(Event:CheckCacheFill);
   tbe.actions.push(Event:DelayAtomic);
@@ -1082,11 +1085,11 @@ action(Initiate_AtomicNoReturn_LocalWrite, desc="") {
 
 
 action(Initiate_AtomicReturn_Forward, desc="") {
-  if ((tbe.dir_sharers.count() > 0) &&
-     (tbe.dir_sharers.isElement(tbe.requestor))){
-    tbe.dir_sharers.remove(tbe.requestor);
-  }
+  tbe.actions.pushNB(Event:WriteFEPipe);
   tbe.actions.push(Event:SendAtomicReturn);
+  tbe.actions.push(Event:SendDBIDResp_AR);
+  tbe.actions.push(Event:SendARData);
+  tbe.actions.pushNB(Event:WriteFEPipe);
   tbe.actions.push(Event:SendCompData_AR);
   tbe.actions.pushNB(Event:TagArrayWrite);
 
@@ -1094,10 +1097,6 @@ action(Initiate_AtomicReturn_Forward, desc="") {
 }
 
 action(Initiate_AtomicNoReturn_Forward, desc="") {
-  if ((tbe.dir_sharers.count() > 0) &&
-     (tbe.dir_sharers.isElement(tbe.requestor))){
-    tbe.dir_sharers.remove(tbe.requestor);
-  }
   if (comp_anr) {
     tbe.actions.push(Event:SendAtomicNoReturn);
     tbe.actions.push(Event:SendDBIDResp_ANR);
@@ -1118,7 +1117,7 @@ action(Initiate_AtomicReturn_Miss, desc="") {
   tbe.actions.pushNB(Event:WriteFEPipe);
   tbe.actions.push(Event:SendDBIDResp_AR);
   tbe.actions.pushNB(Event:WriteFEPipe);
-  tbe.actions.pushNB(Event:SendCompData_AR);
+  tbe.actions.push(Event:SendCompData_AR);
   tbe.actions.push(Event:WriteFEPipe);
   tbe.actions.push(Event:CheckCacheFill);
   tbe.actions.push(Event:DelayAtomic);
@@ -1664,18 +1663,6 @@ action(Send_AtomicReturn, desc="") {
   tbe.expected_req_resp.addExpectedCount(1);
 }
 
-action(Send_AtomicReturn_NoWait, desc="") {
-  assert(is_valid(tbe));
-
-  enqueue(reqOutPort, CHIRequestMsg, request_latency) {
-    prepareRequestAtomic(tbe, CHIRequestType:AtomicReturn, out_msg);
-    out_msg.Destination.add(mapAddressToDownstreamMachine(tbe.addr));
-    allowRequestRetry(tbe, out_msg);
-  }
-
-  tbe.dataAMOValid := false;
-}
-
 action(Send_AtomicNoReturn, desc="") {
   assert(is_valid(tbe));
 
@@ -2337,6 +2324,10 @@ action(UpdateDataState_FromADataResp, desc="") {
     tbe.dataDirty := true;
 
     DPRINTF(RubySlicc, "Atomic after %s\n", tbe.dataBlk);
+  } else if ((tbe.expected_req_resp.hasReceivedData()) &&
+             ((tbe.reqType == CHIRequestType:AtomicReturn) ||
+             (tbe.reqType == CHIRequestType:AtomicNoReturn))){
+    tbe.dataMaybeDirtyUpstream := false;
   }
   printTBEState(tbe);
 }
@@ -2504,10 +2495,6 @@ action(Receive_ReqResp_WUNeedComp, desc="") {
   tbe.defer_expected_comp := true;
 }
 
-action(Receive_ReqResp_AR, desc="") {
-  tbe.actions.pushFrontNB(Event:SendARData);
-}
-
 action(Receive_ReqResp_WUComp, desc="") {
   if (tbe.defer_expected_comp) {
     tbe.defer_expected_comp := false;
@@ -2730,16 +2717,17 @@ action(Send_ANRData, desc="") {
 
 action(CheckARComp, desc="") {
   assert(is_valid(tbe));
+  clearExpectedReqResp(tbe);
   tbe.expected_req_resp.addExpectedDataType(CHIDataType:CompData_I);
   tbe.expected_req_resp.addExpectedRespType(CHIResponseType:RespSepData);
-  tbe.expected_req_resp.addExpectedCount(2);
+  tbe.expected_req_resp.setExpectedCount(2);
 }
 
 action(CheckANRComp, desc="") {
   assert(is_valid(tbe));
   if (tbe.defer_expected_comp) {
     tbe.defer_expected_comp := false;
-    tbe.expected_req_resp.addExpectedCount(1);
+    tbe.expected_req_resp.setExpectedCount(1);
     tbe.expected_req_resp.addExpectedRespType(CHIResponseType:Comp);
   }
 }
@@ -3108,10 +3096,6 @@ action(Send_CompData_AR, desc="") {
   assert(is_valid(tbe));
   assert(tbe.dataValid);
 
-  if (is_HN) {
-      tbe.oldDataBlk := tbe.dataBlk;
-  }
-
   tbe.snd_msgType := CHIDataType:CompData_I;
   tbe.dataMaybeDirtyUpstream := false;
   tbe.requestorToBeExclusiveOwner := false;
diff --git a/src/mem/ruby/protocol/chi/CHI-cache-transitions.sm b/src/mem/ruby/protocol/chi/CHI-cache-transitions.sm
index 9ced9d1826..1f6f906c0c 100644
--- a/src/mem/ruby/protocol/chi/CHI-cache-transitions.sm
+++ b/src/mem/ruby/protocol/chi/CHI-cache-transitions.sm
@@ -942,15 +942,6 @@ transition(BUSY_BLKD, SendWriteUnique, BUSY_INTR) {DestinationAvailable} {
 transition(BUSY_BLKD, SendAtomicReturn, BUSY_INTR) {DestinationAvailable} {
   Pop_TriggerQueue;
   Send_AtomicReturn;
-  CheckARComp;
-  Profile_OutgoingStart;
-  ProcessNextState_ClearPending;
-}
-
-transition(BUSY_BLKD, SendAtomicReturn_NoWait, BUSY_INTR) {
-  Pop_TriggerQueue;
-  Send_AtomicReturn_NoWait;
-  CheckARComp;
   Profile_OutgoingStart;
   ProcessNextState_ClearPending;
 }
@@ -1016,6 +1007,7 @@ transition(BUSY_BLKD, SendWUDataCB) {
 transition({BUSY_BLKD,BUSY_INTR}, SendARData) {
   Pop_TriggerQueue;
   Send_ARData;
+  CheckARComp;
   ProcessNextState_ClearPending;
 }
 
@@ -1492,7 +1484,6 @@ transition({BUSY_INTR,BUSY_BLKD}, DBIDResp, BUSY_BLKD) {
   Receive_ReqResp;
   Receive_ReqResp_CopyDBID;
   Receive_ReqResp_WUNeedComp;
-  Receive_ReqResp_AR;
   Pop_RespInQueue;
   ProcessNextState;
 }
diff --git a/src/mem/ruby/protocol/chi/CHI-cache.sm b/src/mem/ruby/protocol/chi/CHI-cache.sm
index a5c75b167b..487fe60026 100644
--- a/src/mem/ruby/protocol/chi/CHI-cache.sm
+++ b/src/mem/ruby/protocol/chi/CHI-cache.sm
@@ -135,7 +135,7 @@ machine(MachineType:Cache, "Cache coherency protocol") :
   // All Near executes all Atomics at L1 (variable set to 0; default)
   // Unique Near executes Atomics at HNF for states I, SC, SD (set to 1)
   // Present Near execites all Atomics at L1 except when state is I (set to 2)
-  int policy_type := 1;
+  int policy_type := 0;
 
 
   // Use separate Comp/DBIDResp responses for WriteUnique
@@ -487,7 +487,6 @@ machine(MachineType:Cache, "Cache coherency protocol") :
 
     // Send an atomic request downstream.
     SendAtomicReturn,          out_trans="yes", desc="Send atomic request with return";
-    SendAtomicReturn_NoWait,   out_trans="yes", desc="Send atomic request with return, but no DBID";
     SendAtomicNoReturn,        out_trans="yes", desc="Send atomic request without return";
     SendARData,                desc="Send atomic return request data";
     SendANRData,               desc="Send atomic no return request data";