ruby: support to stallAndWait the mandatory queue

By stalling and waiting the mandatory queue instead of recycling it, one can
ensure that no incoming messages are starved when the mandatory queue puts
signficant of pressure on the L1 cache controller (i.e. the ruby memtester).

--HG--
rename : src/mem/slicc/ast/WakeUpDependentsStatementAST.py => src/mem/slicc/ast/WakeUpAllDependentsStatementAST.py

2 files changed, 178 insertions, 106 deletions

diff --git a/src/mem/protocol/MOESI_CMP_token-L1cache.sm b/src/mem/protocol/MOESI_CMP_token-L1cache.sm
index 8cb45249e..00e9404c9 100644
--- a/src/mem/protocol/MOESI_CMP_token-L1cache.sm
+++ b/src/mem/protocol/MOESI_CMP_token-L1cache.sm
@@ -433,7 +433,7 @@ machine(L1Cache, "Token protocol")
   // ** IN_PORTS **
 
   // Use Timer
-  in_port(useTimerTable_in, Address, useTimerTable) {
+  in_port(useTimerTable_in, Address, useTimerTable, rank=5) {
     if (useTimerTable_in.isReady()) {
       TBE tbe := L1_TBEs[useTimerTable.readyAddress()];
 
@@ -459,7 +459,7 @@ machine(L1Cache, "Token protocol")
   }
 
   // Reissue Timer
-  in_port(reissueTimerTable_in, Address, reissueTimerTable) {
+  in_port(reissueTimerTable_in, Address, reissueTimerTable, rank=4) {
     if (reissueTimerTable_in.isReady()) {
       trigger(Event:Request_Timeout, reissueTimerTable.readyAddress(),
               getCacheEntry(reissueTimerTable.readyAddress()),
@@ -467,10 +467,8 @@ machine(L1Cache, "Token protocol")
     }
   }
 
-
-
   // Persistent Network
-  in_port(persistentNetwork_in, PersistentMsg, persistentToL1Cache) {
+  in_port(persistentNetwork_in, PersistentMsg, persistentToL1Cache, rank=3) {
     if (persistentNetwork_in.isReady()) {
       peek(persistentNetwork_in, PersistentMsg, block_on="Address") {
         assert(in_msg.Destination.isElement(machineID));
@@ -519,56 +517,8 @@ machine(L1Cache, "Token protocol")
     }
   }
 
-
-  // Request Network
-  in_port(requestNetwork_in, RequestMsg, requestToL1Cache) {
-    if (requestNetwork_in.isReady()) {
-      peek(requestNetwork_in, RequestMsg, block_on="Address") {
-        assert(in_msg.Destination.isElement(machineID));
-
-        Entry cache_entry := getCacheEntry(in_msg.Address);
-        TBE tbe := L1_TBEs[in_msg.Address];
-
-        if (in_msg.Type == CoherenceRequestType:GETX) {
-          if (in_msg.isLocal) {
-            trigger(Event:Transient_Local_GETX, in_msg.Address,
-                    cache_entry, tbe);
-          }
-          else {
-            trigger(Event:Transient_GETX, in_msg.Address,
-                    cache_entry, tbe);
-          }
-        } else if (in_msg.Type == CoherenceRequestType:GETS) {
-          if (getTokens(cache_entry) == 1 || 
-              getTokens(cache_entry) == (max_tokens() / 2) + 1) {
-            if (in_msg.isLocal) {
-              trigger(Event:Transient_Local_GETS_Last_Token, in_msg.Address,
-                      cache_entry, tbe);
-            }
-            else {
-              trigger(Event:Transient_GETS_Last_Token, in_msg.Address,
-                      cache_entry, tbe);
-            }
-          }
-          else {
-            if (in_msg.isLocal) {
-              trigger(Event:Transient_Local_GETS, in_msg.Address,
-                      cache_entry, tbe);
-            }
-            else {
-              trigger(Event:Transient_GETS, in_msg.Address,
-                      cache_entry, tbe);
-            }
-          }
-        } else {
-          error("Unexpected message");
-        }
-      }
-    }
-  }
-
   // Response Network
-  in_port(responseNetwork_in, ResponseMsg, responseToL1Cache) {
+  in_port(responseNetwork_in, ResponseMsg, responseToL1Cache, rank=2) {
     if (responseNetwork_in.isReady()) {
       peek(responseNetwork_in, ResponseMsg, block_on="Address") {
         assert(in_msg.Destination.isElement(machineID));
@@ -639,8 +589,55 @@ machine(L1Cache, "Token protocol")
     }
   }
 
+  // Request Network
+  in_port(requestNetwork_in, RequestMsg, requestToL1Cache) {
+    if (requestNetwork_in.isReady()) {
+      peek(requestNetwork_in, RequestMsg, block_on="Address") {
+        assert(in_msg.Destination.isElement(machineID));
+
+        Entry cache_entry := getCacheEntry(in_msg.Address);
+        TBE tbe := L1_TBEs[in_msg.Address];
+
+        if (in_msg.Type == CoherenceRequestType:GETX) {
+          if (in_msg.isLocal) {
+            trigger(Event:Transient_Local_GETX, in_msg.Address,
+                    cache_entry, tbe);
+          }
+          else {
+            trigger(Event:Transient_GETX, in_msg.Address,
+                    cache_entry, tbe);
+          }
+        } else if (in_msg.Type == CoherenceRequestType:GETS) {
+          if (getTokens(cache_entry) == 1 || 
+              getTokens(cache_entry) == (max_tokens() / 2) + 1) {
+            if (in_msg.isLocal) {
+              trigger(Event:Transient_Local_GETS_Last_Token, in_msg.Address,
+                      cache_entry, tbe);
+            }
+            else {
+              trigger(Event:Transient_GETS_Last_Token, in_msg.Address,
+                      cache_entry, tbe);
+            }
+          }
+          else {
+            if (in_msg.isLocal) {
+              trigger(Event:Transient_Local_GETS, in_msg.Address,
+                      cache_entry, tbe);
+            }
+            else {
+              trigger(Event:Transient_GETS, in_msg.Address,
+                      cache_entry, tbe);
+            }
+          }
+        } else {
+          error("Unexpected message");
+        }
+      }
+    }
+  }
+
   // Mandatory Queue
-  in_port(mandatoryQueue_in, CacheMsg, mandatoryQueue, desc="...") {
+  in_port(mandatoryQueue_in, CacheMsg, mandatoryQueue, desc="...", rank=0) {
     if (mandatoryQueue_in.isReady()) {
       peek(mandatoryQueue_in, CacheMsg, block_on="LineAddress") {
         // Check for data access to blocks in I-cache and ifetchs to blocks in D-cache
@@ -1339,6 +1336,11 @@ machine(L1Cache, "Token protocol")
     tbe.IssueTime := get_time();
   }
 
+  action(ta_traceStalledAddress, "ta", desc="Trace Stalled Address") {
+    peek(mandatoryQueue_in, CacheMsg) {
+      APPEND_TRANSITION_COMMENT(in_msg.LineAddress);
+    }
+  }
 
   action(j_unsetReissueTimer, "j", desc="Unset reissue timer.") {
     if (reissueTimerTable.isSet(address)) {
@@ -1527,8 +1529,19 @@ machine(L1Cache, "Token protocol")
     }
   }
 
-  action(zz_recycleMandatoryQueue, "\z", desc="Send the head of the mandatory queue to the back of the queue.") {
-    mandatoryQueue_in.recycle();
+  action(zz_stallAndWaitMandatoryQueue, "\z", desc="Send the head of the mandatory queue to the back of the queue.") {
+    peek(mandatoryQueue_in, CacheMsg) {
+      APPEND_TRANSITION_COMMENT(in_msg.LineAddress);
+    } 
+    stall_and_wait(mandatoryQueue_in, address);    
+  }
+
+  action(kd_wakeUpDependents, "kd", desc="wake-up dependents") {
+    wake_up_dependents(address);
+  }
+
+  action(ka_wakeUpAllDependents, "ka", desc="wake-up all dependents") {
+    wake_up_all_dependents();
   }
 
   //*****************************************************
@@ -1537,15 +1550,16 @@ machine(L1Cache, "Token protocol")
 
   // Transitions for Load/Store/L2_Replacement from transient states
   transition({IM, SM, OM, IS, IM_L, IS_L, I_L, S_L, SM_L, M_W, MM_W}, L1_Replacement) {
-    zz_recycleMandatoryQueue;
+    ta_traceStalledAddress;
+    zz_stallAndWaitMandatoryQueue;
   }
 
   transition({IM, SM, OM, IS, IM_L, IS_L, SM_L}, {Store, Atomic}) {
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
   }
 
   transition({IM, IS, IM_L, IS_L}, {Load, Ifetch}) {
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
   }
 
 
@@ -1615,8 +1629,10 @@ machine(L1Cache, "Token protocol")
   }
 
   transition(I, L1_Replacement) {
+    ta_traceStalledAddress;
     tr_tokenReplacement;
     gg_deallocateL1CacheBlock;
+    ka_wakeUpAllDependents;
   }
 
   transition(I, {Transient_GETX, Transient_Local_GETX}) {
@@ -1674,8 +1690,10 @@ machine(L1Cache, "Token protocol")
   }
 
   transition(S, L1_Replacement, I) {
+    ta_traceStalledAddress;
     cc_sharedReplacement; // Only needed in some cases
     gg_deallocateL1CacheBlock;
+    ka_wakeUpAllDependents;
   }
 
   transition(S, {Transient_GETX, Transient_Local_GETX}, I) {
@@ -1750,8 +1768,10 @@ machine(L1Cache, "Token protocol")
   }
 
   transition(O, L1_Replacement, I) {
+    ta_traceStalledAddress;
     c_ownedReplacement;
     gg_deallocateL1CacheBlock;
+    ka_wakeUpAllDependents;
   }
 
   transition(O, {Transient_GETX, Transient_Local_GETX}, I) {
@@ -1835,8 +1855,10 @@ machine(L1Cache, "Token protocol")
   }
 
   transition(MM, L1_Replacement, I) {
+    ta_traceStalledAddress;
     c_ownedReplacement;
     gg_deallocateL1CacheBlock;
+    ka_wakeUpAllDependents;
   }
 
   transition(MM, {Transient_GETX, Transient_Local_GETX, Transient_GETS, Transient_Local_GETS}, I) {
@@ -1864,11 +1886,13 @@ machine(L1Cache, "Token protocol")
   transition(MM_W, Use_TimeoutNoStarvers, MM) {
     s_deallocateTBE;
     jj_unsetUseTimer;
+    kd_wakeUpDependents;
   }
 
   transition(MM_W, Use_TimeoutNoStarvers_NoMig, M) {
     s_deallocateTBE;
     jj_unsetUseTimer;
+    kd_wakeUpDependents;
   }
 
   // Transitions from Dirty Exclusive
@@ -1898,8 +1922,10 @@ machine(L1Cache, "Token protocol")
   }
 
   transition(M, L1_Replacement, I) {
+    ta_traceStalledAddress;
     c_ownedReplacement;
     gg_deallocateL1CacheBlock;
+    ka_wakeUpAllDependents;
   }
 
   transition(M, {Transient_GETX, Transient_Local_GETX}, I) {
@@ -1948,6 +1974,7 @@ machine(L1Cache, "Token protocol")
   transition(M_W, {Use_TimeoutNoStarvers, Use_TimeoutNoStarvers_NoMig}, M) {
     s_deallocateTBE;
     jj_unsetUseTimer;
+    kd_wakeUpDependents;
   }
 
   transition(M_W, Use_TimeoutStarverX, I_L) {
@@ -2056,6 +2083,7 @@ machine(L1Cache, "Token protocol")
     o_scheduleUseTimeout;
     j_unsetReissueTimer;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   transition(SM, Data_Shared) {
@@ -2077,6 +2105,7 @@ machine(L1Cache, "Token protocol")
     o_scheduleUseTimeout;
     j_unsetReissueTimer;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   transition({IM, SM}, {Transient_GETX, Transient_Local_GETX}, IM) { // We don't have the data yet, but we might have collected some tokens.  We give them up here to avoid livelock
@@ -2106,6 +2135,7 @@ machine(L1Cache, "Token protocol")
     o_scheduleUseTimeout;
     j_unsetReissueTimer;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   transition(OM, Data_Shared) {
@@ -2121,6 +2151,7 @@ machine(L1Cache, "Token protocol")
     o_scheduleUseTimeout;
     j_unsetReissueTimer;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   transition(OM, Request_Timeout) {
@@ -2142,6 +2173,7 @@ machine(L1Cache, "Token protocol")
     s_deallocateTBE;
     j_unsetReissueTimer;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   transition(IS, Data_Owner, O) {
@@ -2151,6 +2183,7 @@ machine(L1Cache, "Token protocol")
     s_deallocateTBE;
     j_unsetReissueTimer;
     n_popResponseQueue;
+    kd_wakeUpDependents;    
   }
 
   transition(IS, Data_All_Tokens, M_W) {
@@ -2160,6 +2193,7 @@ machine(L1Cache, "Token protocol")
     o_scheduleUseTimeout;
     j_unsetReissueTimer;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   transition(IS, Request_Timeout) {
@@ -2247,6 +2281,7 @@ machine(L1Cache, "Token protocol")
     j_unsetReissueTimer;
     o_scheduleUseTimeout;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   transition(SM_L, Data_All_Tokens, S_L) {
@@ -2288,6 +2323,7 @@ machine(L1Cache, "Token protocol")
     j_unsetReissueTimer;
     o_scheduleUseTimeout;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
 
@@ -2295,22 +2331,27 @@ machine(L1Cache, "Token protocol")
 
   transition(I_L, Own_Lock_or_Unlock, I) {
     l_popPersistentQueue;
+    kd_wakeUpDependents;
   }
 
   transition(S_L, Own_Lock_or_Unlock, S) {
     l_popPersistentQueue;
+    kd_wakeUpDependents;
   }
 
   transition(IM_L, Own_Lock_or_Unlock, IM) {
     l_popPersistentQueue;
+    kd_wakeUpDependents;
   }
 
   transition(IS_L, Own_Lock_or_Unlock, IS) {
     l_popPersistentQueue;
+    kd_wakeUpDependents;
   }
 
   transition(SM_L, Own_Lock_or_Unlock, SM) {
     l_popPersistentQueue;
+    kd_wakeUpDependents;
   }
 }
 
diff --git a/src/mem/protocol/MOESI_hammer-cache.sm b/src/mem/protocol/MOESI_hammer-cache.sm
index 1f14db4f7..78bc9e3e7 100644
--- a/src/mem/protocol/MOESI_hammer-cache.sm
+++ b/src/mem/protocol/MOESI_hammer-cache.sm
@@ -278,7 +278,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
   // ** IN_PORTS **
 
   // Trigger Queue
-  in_port(triggerQueue_in, TriggerMsg, triggerQueue) {
+  in_port(triggerQueue_in, TriggerMsg, triggerQueue, rank=3) {
     if (triggerQueue_in.isReady()) {
       peek(triggerQueue_in, TriggerMsg) {
 
@@ -298,10 +298,35 @@ machine(L1Cache, "AMD Hammer-like protocol")
     }
   }
 
-  // Nothing from the request network
+  // Nothing from the unblock network
+
+  // Response Network
+  in_port(responseToCache_in, ResponseMsg, responseToCache, rank=2) {
+    if (responseToCache_in.isReady()) {
+      peek(responseToCache_in, ResponseMsg, block_on="Address") {
+
+        Entry cache_entry := getCacheEntry(in_msg.Address);
+        TBE tbe := TBEs[in_msg.Address];
+
+        if (in_msg.Type == CoherenceResponseType:ACK) {
+          trigger(Event:Ack, in_msg.Address, cache_entry, tbe);
+        } else if (in_msg.Type == CoherenceResponseType:ACK_SHARED) {
+          trigger(Event:Shared_Ack, in_msg.Address, cache_entry, tbe);
+        } else if (in_msg.Type == CoherenceResponseType:DATA) {
+          trigger(Event:Data, in_msg.Address, cache_entry, tbe);
+        } else if (in_msg.Type == CoherenceResponseType:DATA_SHARED) {
+          trigger(Event:Shared_Data, in_msg.Address, cache_entry, tbe);
+        } else if (in_msg.Type == CoherenceResponseType:DATA_EXCLUSIVE) {
+          trigger(Event:Exclusive_Data, in_msg.Address, cache_entry, tbe);
+        } else {
+          error("Unexpected message");
+        }
+      }
+    }
+  }
 
   // Forward Network
-  in_port(forwardToCache_in, RequestMsg, forwardToCache) {
+  in_port(forwardToCache_in, RequestMsg, forwardToCache, rank=1) {
     if (forwardToCache_in.isReady()) {
       peek(forwardToCache_in, RequestMsg, block_on="Address") {
 
@@ -339,35 +364,10 @@ machine(L1Cache, "AMD Hammer-like protocol")
     }
   }
 
-  // Response Network
-  in_port(responseToCache_in, ResponseMsg, responseToCache) {
-    if (responseToCache_in.isReady()) {
-      peek(responseToCache_in, ResponseMsg, block_on="Address") {
-
-        Entry cache_entry := getCacheEntry(in_msg.Address);
-        TBE tbe := TBEs[in_msg.Address];
-
-        if (in_msg.Type == CoherenceResponseType:ACK) {
-          trigger(Event:Ack, in_msg.Address, cache_entry, tbe);
-        } else if (in_msg.Type == CoherenceResponseType:ACK_SHARED) {
-          trigger(Event:Shared_Ack, in_msg.Address, cache_entry, tbe);
-        } else if (in_msg.Type == CoherenceResponseType:DATA) {
-          trigger(Event:Data, in_msg.Address, cache_entry, tbe);
-        } else if (in_msg.Type == CoherenceResponseType:DATA_SHARED) {
-          trigger(Event:Shared_Data, in_msg.Address, cache_entry, tbe);
-        } else if (in_msg.Type == CoherenceResponseType:DATA_EXCLUSIVE) {
-          trigger(Event:Exclusive_Data, in_msg.Address, cache_entry, tbe);
-        } else {
-          error("Unexpected message");
-        }
-      }
-    }
-  }
-
-  // Nothing from the unblock network
+  // Nothing from the request network
 
   // Mandatory Queue
-  in_port(mandatoryQueue_in, CacheMsg, mandatoryQueue, desc="...") {
+  in_port(mandatoryQueue_in, CacheMsg, mandatoryQueue, desc="...", rank=0) {
     if (mandatoryQueue_in.isReady()) {
       peek(mandatoryQueue_in, CacheMsg, block_on="LineAddress") {
 
@@ -1017,8 +1017,16 @@ machine(L1Cache, "AMD Hammer-like protocol")
     }
   }
 
-  action(zz_recycleMandatoryQueue, "\z", desc="Send the head of the mandatory queue to the back of the queue.") {
-    mandatoryQueue_in.recycle();
+  action(zz_stallAndWaitMandatoryQueue, "\z", desc="Send the head of the mandatory queue to the back of the queue.") {
+    stall_and_wait(mandatoryQueue_in, address);    
+  }
+
+  action(kd_wakeUpDependents, "kd", desc="wake-up dependents") {
+    wake_up_dependents(address);
+  }
+
+  action(ka_wakeUpAllDependents, "ka", desc="wake-up all dependents") {
+    wake_up_all_dependents();
   }
 
   //*****************************************************
@@ -1027,19 +1035,19 @@ machine(L1Cache, "AMD Hammer-like protocol")
 
   // Transitions for Load/Store/L2_Replacement from transient states
   transition({IM, SM, ISM, OM, IS, SS, OI, MI, II, IT, ST, OT, MT, MMT}, {Store, L2_Replacement}) {
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
   }
 
   transition({M_W, MM_W}, {L2_Replacement}) {
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
   }
 
   transition({IM, IS, OI, MI, II, IT, ST, OT, MT, MMT}, {Load, Ifetch}) {
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
   }
 
   transition({IM, SM, ISM, OM, IS, SS, MM_W, M_W, OI, MI, II, IT, ST, OT, MT, MMT}, L1_to_L2) {
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
   }
 
   transition({IT, ST, OT, MT, MMT}, {Other_GETX, NC_DMA_GETS, Other_GETS, Merged_GETS, Other_GETS_No_Mig, Invalidate}) {
@@ -1053,6 +1061,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     vv_allocateL2CacheBlock;
     hp_copyFromTBEToL2;
     s_deallocateTBE;
+    ka_wakeUpAllDependents;
   }
 
   transition(I, Trigger_L2_to_L1D, IT) {
@@ -1062,7 +1071,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     nb_copyFromTBEToL1; // Not really needed for state I
     s_deallocateTBE;
     uu_profileMiss;
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
     ll_L2toL1Transfer;
   }
 
@@ -1073,7 +1082,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     nb_copyFromTBEToL1;
     s_deallocateTBE;
     uu_profileMiss;
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
     ll_L2toL1Transfer;
   }
 
@@ -1084,7 +1093,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     nb_copyFromTBEToL1;
     s_deallocateTBE;
     uu_profileMiss;
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
     ll_L2toL1Transfer;
   }
 
@@ -1095,7 +1104,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     nb_copyFromTBEToL1;
     s_deallocateTBE;
     uu_profileMiss;
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
     ll_L2toL1Transfer;
   }
 
@@ -1106,7 +1115,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     nb_copyFromTBEToL1;
     s_deallocateTBE;
     uu_profileMiss;
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
     ll_L2toL1Transfer;
   }
 
@@ -1117,7 +1126,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     nb_copyFromTBEToL1;
     s_deallocateTBE;
     uu_profileMiss;
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
     ll_L2toL1Transfer;
   }
 
@@ -1128,7 +1137,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     nb_copyFromTBEToL1;
     s_deallocateTBE;
     uu_profileMiss;
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
     ll_L2toL1Transfer;
   }
 
@@ -1139,7 +1148,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     nb_copyFromTBEToL1;
     s_deallocateTBE;
     uu_profileMiss;
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
     ll_L2toL1Transfer;
   }
 
@@ -1150,7 +1159,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     nb_copyFromTBEToL1;
     s_deallocateTBE;
     uu_profileMiss;
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
     ll_L2toL1Transfer;
   }
 
@@ -1161,28 +1170,33 @@ machine(L1Cache, "AMD Hammer-like protocol")
     nb_copyFromTBEToL1;
     s_deallocateTBE;
     uu_profileMiss;
-    zz_recycleMandatoryQueue;
+    zz_stallAndWaitMandatoryQueue;
     ll_L2toL1Transfer;
   }
 
   transition(IT, Complete_L2_to_L1, I) {
     j_popTriggerQueue;
+    kd_wakeUpDependents;
   }
 
   transition(ST, Complete_L2_to_L1, S) {
     j_popTriggerQueue;
+    kd_wakeUpDependents;
   }
 
   transition(OT, Complete_L2_to_L1, O) {
     j_popTriggerQueue;
+    kd_wakeUpDependents;
   }
 
   transition(MT, Complete_L2_to_L1, M) {
     j_popTriggerQueue;
+    kd_wakeUpDependents;
   }
 
   transition(MMT, Complete_L2_to_L1, MM) {
     j_popTriggerQueue;
+    kd_wakeUpDependents;
   }
 
   // Transitions from Idle
@@ -1212,6 +1226,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
 
   transition(I, L2_Replacement) {
     rr_deallocateL2CacheBlock;
+    ka_wakeUpAllDependents;
   }
 
   transition(I, {Other_GETX, NC_DMA_GETS, Other_GETS, Other_GETS_No_Mig, Invalidate}) {
@@ -1234,6 +1249,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
 
   transition(S, L2_Replacement, I) {
     rr_deallocateL2CacheBlock;
+    ka_wakeUpAllDependents;
   }
 
   transition(S, {Other_GETX, Invalidate}, I) {
@@ -1264,6 +1280,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     i_allocateTBE;
     d_issuePUT;
     rr_deallocateL2CacheBlock;
+    ka_wakeUpAllDependents;
   }
 
   transition(O, {Other_GETX, Invalidate}, I) {
@@ -1296,6 +1313,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     i_allocateTBE;
     d_issuePUT;
     rr_deallocateL2CacheBlock;
+    ka_wakeUpAllDependents;
   }
 
   transition(MM, {Other_GETX, Invalidate}, I) {
@@ -1338,6 +1356,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     i_allocateTBE;
     d_issuePUT;
     rr_deallocateL2CacheBlock;
+    ka_wakeUpAllDependents;
   }
 
   transition(M, {Other_GETX, Invalidate}, I) {
@@ -1386,6 +1405,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     o_checkForCompletion;
     sx_external_store_hit;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   // Transitions from SM
@@ -1424,6 +1444,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     gm_sendUnblockM;
     s_deallocateTBE;
     j_popTriggerQueue;
+    kd_wakeUpDependents;
   }
 
   // Transitions from OM
@@ -1455,6 +1476,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     gm_sendUnblockM;
     s_deallocateTBE;
     j_popTriggerQueue;
+    kd_wakeUpDependents;
   }
 
   // Transitions from IS
@@ -1484,6 +1506,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     hx_external_load_hit;
     uo_updateCurrentOwner;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   transition(IS, Exclusive_Data, M_W) {
@@ -1492,6 +1515,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     o_checkForCompletion;
     hx_external_load_hit;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   transition(IS, Shared_Data, SS) {
@@ -1502,6 +1526,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     hx_external_load_hit;
     uo_updateCurrentOwner;
     n_popResponseQueue;
+    kd_wakeUpDependents;
   }
 
   // Transitions from SS
@@ -1549,6 +1574,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     gm_sendUnblockM;
     s_deallocateTBE;
     j_popTriggerQueue;
+    kd_wakeUpDependents;
   }
 
   // Transitions from M_W
@@ -1568,6 +1594,7 @@ machine(L1Cache, "AMD Hammer-like protocol")
     gm_sendUnblockM;
     s_deallocateTBE;
     j_popTriggerQueue;
+    kd_wakeUpDependents;
   }
 
   // Transitions from OI/MI
@@ -1591,12 +1618,14 @@ machine(L1Cache, "AMD Hammer-like protocol")
     t_sendExclusiveDataFromTBEToMemory;
     s_deallocateTBE;
     l_popForwardQueue;
+    kd_wakeUpDependents;
   }
 
   transition(OI, Writeback_Ack, I) {
     qq_sendDataFromTBEToMemory;
     s_deallocateTBE;
     l_popForwardQueue;
+    kd_wakeUpDependents;
   }
 
   // Transitions from II
@@ -1609,10 +1638,12 @@ machine(L1Cache, "AMD Hammer-like protocol")
     g_sendUnblock;
     s_deallocateTBE;
     l_popForwardQueue;
+    kd_wakeUpDependents;
   }
 
   transition(II, Writeback_Nack, I) {
     s_deallocateTBE;
     l_popForwardQueue;
+    kd_wakeUpDependents;
   }
 }