diff options
Diffstat (limited to 'src/mem/protocol/MESI_CMP_filter_directory-L2cache.sm')
-rw-r--r-- | src/mem/protocol/MESI_CMP_filter_directory-L2cache.sm | 2123 |
1 files changed, 2123 insertions, 0 deletions
diff --git a/src/mem/protocol/MESI_CMP_filter_directory-L2cache.sm b/src/mem/protocol/MESI_CMP_filter_directory-L2cache.sm new file mode 100644 index 000000000..9085ae33f --- /dev/null +++ b/src/mem/protocol/MESI_CMP_filter_directory-L2cache.sm @@ -0,0 +1,2123 @@ + +/* + * Copyright (c) 1999-2005 Mark D. Hill and David A. Wood + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer; + * redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution; + * neither the name of the copyright holders nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +/* + * $Id: MSI_MOSI_CMP_directory-L2cache.sm 1.12 05/01/19 15:55:40-06:00 beckmann@s0-28.cs.wisc.edu $ + * + */ + +machine(L2Cache, "MESI Directory L2 Cache CMP") { + + // L2 BANK QUEUES + // From local bank of L2 cache TO the network + MessageBuffer DirRequestFromL2Cache, network="To", virtual_network="2", ordered="false"; // this L2 bank -> Memory + MessageBuffer L1RequestFromL2Cache, network="To", virtual_network="1", ordered="false"; // this L2 bank -> a local L1 + MessageBuffer responseFromL2Cache, network="To", virtual_network="3", ordered="false"; // this L2 bank -> a local L1 || Memory + + // FROM the network to this local bank of L2 cache + MessageBuffer L1RequestToL2Cache, network="From", virtual_network="0", ordered="false"; // a local L1 -> this L2 bank + MessageBuffer responseToL2Cache, network="From", virtual_network="3", ordered="false"; // a local L1 || Memory -> this L2 bank + MessageBuffer unblockToL2Cache, network="From", virtual_network="4", ordered="false"; // a local L1 || Memory -> this L2 bank + + // STATES + enumeration(State, desc="L2 Cache states", default="L2Cache_State_NP") { + // Base states + NP, desc="Not present in either cache"; + SS, desc="L2 cache entry Shared, also present in one or more L1s"; + M, desc="L2 cache entry Modified, not present in any L1s", format="!b"; + MT, desc="L2 cache entry Modified in a local L1, assume L2 copy stale", format="!b"; + + // L2 replacement + M_I, desc="L2 cache replacing, have all acks, sent dirty data to memory, waiting for ACK from memory"; + MT_I, desc="L2 cache replacing, getting data from exclusive"; + MCT_I, desc="L2 cache replacing, clean in L2, getting data or ack from exclusive"; + I_I, desc="L2 replacing clean data, need to inv sharers and then drop data"; + S_I, desc="L2 replacing dirty data, collecting acks from L1s"; + + // Transient States for fetching data from memory + ISS, desc="L2 idle, got single L1_GETS, issued memory fetch, have not seen response yet"; + IS, desc="L2 idle, got L1_GET_INSTR or multiple L1_GETS, issued memory fetch, have not seen response yet"; + IM, desc="L2 idle, got L1_GETX, issued memory fetch, have not seen response(s) yet"; + + // Blocking states + SS_MB, desc="Blocked for L1_GETX from SS"; + SS_SSB, desc="Blocked for L1_GETS from SS"; + MT_MB, desc="Blocked for L1_GETX from MT"; + M_MB, desc="Blocked for L1_GETX from M"; + ISS_MB, desc="Blocked for L1_GETS or L1_GETX from NP, received Mem Data"; + IS_SSB, desc="Blocked for L1_GET_INSTR from NP, received Mem Data"; + M_SSB, desc="Blocked for L1_GET_INSTR from M"; + + MT_IIB, desc="Blocked for L1_GETS from MT, waiting for unblock and data"; + MT_IB, desc="Blocked for L1_GETS from MT, got unblock, waiting for data"; + MT_SB, desc="Blocked for L1_GETS from MT, got data, waiting for unblock"; + + // for resolving PUTX/PUTS races + PB_MT, desc="Going to MT, got data and unblock, waiting for PUT"; + PB_SS, desc="Going to SS, got unblock, waiting for PUT"; + PB_MT_IB, desc="Blocked from MT, got unblock, waiting for data and PUT"; + + } + + // EVENTS + enumeration(Event, desc="L2 Cache events") { + // L2 events + + // events initiated by the local L1s + L1_GET_INSTR, desc="a L1I GET INSTR request for a block maped to us"; + L1_GET_INSTR_ESCAPE, desc="a L1I GET INSTR in an escape action request for a block mapped to us"; + L1_GETS, desc="a L1D GETS request for a block maped to us"; + L1_GETS_ESCAPE, desc="a L1D GETS in an escape action request for a block mapped to us"; + L1_GETX, desc="a L1D GETX request for a block maped to us"; + L1_GETX_ESCAPE, desc="a L1D GETX in an escape action request for a block mapped to us"; + L1_UPGRADE, desc="a L1D GETX request for a block maped to us"; + + L1_PUTX, desc="L1 replacing data"; + L1_PUTX_old, desc="L1 replacing data, but no longer sharer"; + L1_PUTS, desc="L1 replacing clean data"; + L1_PUTS_old, desc="L1 replacing clean data, but no longer sharer"; + L1_PUT_PENDING, desc="L1 PUT msg pending (recycled)"; + + Fwd_L1_GETX, desc="L1 did not have data, so we supply"; + Fwd_L1_GETS, desc="L1 did not have data, so we supply"; + Fwd_L1_GET_INSTR, desc="L1 did not have data, so we supply"; + + // events initiated by this L2 + L2_Replacement, desc="L2 Replacement", format="!r"; + L2_Replacement_XACT, desc="L2 Replacement of trans. data", format="!r"; + L2_Replacement_clean, desc="L2 Replacement, but data is clean", format="!r"; + L2_Replacement_clean_XACT, desc="L2 Replacement of trans. data, but data is clean", format="!r"; + + // events from memory controller + Mem_Data, desc="data from memory", format="!r"; + Mem_Ack, desc="ack from memory", format="!r"; + + // M->S data writeback + WB_Data, desc="data from L1"; + WB_Data_clean, desc="clean data from L1"; + Ack, desc="writeback ack"; + Ack_all, desc="writeback ack"; + // For transactional memory + Nack, desc="filter indicates conflict"; + Nack_all, desc="all filters have responded, at least one conflict"; + + Unblock, desc="Unblock from L1 requestor"; + Unblock_Cancel, desc="Unblock from L1 requestor (FOR XACT MEMORY)"; + Exclusive_Unblock, desc="Unblock from L1 requestor"; + + Unblock_WaitPUTold, desc="Unblock from L1 requestor, last requestor was replacing so wait for PUT msg"; + Exclusive_Unblock_WaitPUTold, desc="Unblock from L1 requestor, last requestor was replacing so wait for PUT msg"; + + } + + // TYPES + + // CacheEntry + structure(Entry, desc="...", interface="AbstractCacheEntry") { + State CacheState, desc="cache state"; + NetDest Sharers, desc="tracks the L1 shares on-chip"; + MachineID Exclusive, desc="Exclusive holder of block"; + DataBlock DataBlk, desc="data for the block"; + bool Dirty, default="false", desc="data is dirty"; + + bool Trans, desc="dummy bit for debugging"; + bool Read, desc="LogTM R bit"; + bool Write, desc="LogTM W bit"; + bool L2Miss, desc="Was this block sourced from memory"; + int L1PutsPending, default="0", desc="how many PUT_ are pending for this entry (being recyled)"; + } + + // TBE fields + structure(TBE, desc="...") { + Address Address, desc="Line address for this TBE"; + Address PhysicalAddress, desc="Physical address for this TBE"; + State TBEState, desc="Transient state"; + DataBlock DataBlk, desc="Buffer for the data block"; + bool Dirty, default="false", desc="Data is Dirty"; + + NetDest L1_GetS_IDs, desc="Set of the internal processors that want the block in shared state"; + MachineID L1_GetX_ID, desc="ID of the L1 cache to forward the block to once we get a response"; + bool isPrefetch, desc="Set if this was caused by a prefetch"; + + int pendingAcks, desc="number of pending acks for invalidates during writeback"; + bool nack, default="false", desc="has this request been NACKed?"; + } + + external_type(CacheMemory) { + bool cacheAvail(Address); + Address cacheProbe(Address); + void allocate(Address); + void deallocate(Address); + Entry lookup(Address); + void changePermission(Address, AccessPermission); + bool isTagPresent(Address); + void setMRU(Address); + } + + external_type(TBETable) { + TBE lookup(Address); + void allocate(Address); + void deallocate(Address); + bool isPresent(Address); + } + + TBETable L2_TBEs, template_hack="<L2Cache_TBE>"; + + CacheMemory L2cacheMemory, template_hack="<L2Cache_Entry>", constructor_hack='L2_CACHE_NUM_SETS_BITS,L2_CACHE_ASSOC,MachineType_L2Cache,int_to_string(i)'; + + // inclusive cache, returns L2 entries only + Entry getL2CacheEntry(Address addr), return_by_ref="yes" { + return L2cacheMemory[addr]; + } + + void changeL2Permission(Address addr, AccessPermission permission) { + if (L2cacheMemory.isTagPresent(addr)) { + return L2cacheMemory.changePermission(addr, permission); + } + } + + string getCoherenceRequestTypeStr(CoherenceRequestType type) { + return CoherenceRequestType_to_string(type); + } + + bool isL2CacheTagPresent(Address addr) { + return (L2cacheMemory.isTagPresent(addr)); + } + + bool isOneSharerLeft(Address addr, MachineID requestor) { + assert(L2cacheMemory[addr].Sharers.isElement(requestor)); + return (L2cacheMemory[addr].Sharers.count() == 1); + } + + bool isSharer(Address addr, MachineID requestor) { + if (L2cacheMemory.isTagPresent(addr)) { + return L2cacheMemory[addr].Sharers.isElement(requestor); + } else { + return false; + } + } + + void addSharer(Address addr, MachineID requestor) { + DEBUG_EXPR(machineID); + DEBUG_EXPR(requestor); + DEBUG_EXPR(addr); + assert(map_L1CacheMachId_to_L2Cache(addr, requestor) == machineID); + L2cacheMemory[addr].Sharers.add(requestor); + } + + State getState(Address addr) { + if(L2_TBEs.isPresent(addr)) { + return L2_TBEs[addr].TBEState; + } else if (isL2CacheTagPresent(addr)) { + return getL2CacheEntry(addr).CacheState; + } + return State:NP; + } + + string getStateStr(Address addr) { + return L2Cache_State_to_string(getState(addr)); + } + + // when is this called + void setState(Address addr, State state) { + + // MUST CHANGE + if (L2_TBEs.isPresent(addr)) { + L2_TBEs[addr].TBEState := state; + } + + if (isL2CacheTagPresent(addr)) { + getL2CacheEntry(addr).CacheState := state; + + // Set permission + if (state == State:SS ) { + changeL2Permission(addr, AccessPermission:Read_Only); + } else if (state == State:M) { + changeL2Permission(addr, AccessPermission:Read_Write); + } else if (state == State:MT) { + changeL2Permission(addr, AccessPermission:Stale); + } else { + changeL2Permission(addr, AccessPermission:Busy); + } + } + } + + Event L1Cache_request_type_to_event(CoherenceRequestType type, Address addr, MachineID requestor) { + if (L2cacheMemory.isTagPresent(addr)){ /* Present */ + if(getL2CacheEntry(addr).L1PutsPending > 0 && /* At least one PUT pending */ + (getL2CacheEntry(addr).CacheState == State:SS || getL2CacheEntry(addr).CacheState == State:MT || getL2CacheEntry(addr).CacheState == State:M )) { /* Base state */ + + /* Only allow PUTX/PUTS to go on */ + if (type != CoherenceRequestType:PUTX && + type != CoherenceRequestType:PUTS) { + return Event:L1_PUT_PENDING; // Don't serve any req until the wb is serviced + } + } + } + if(type == CoherenceRequestType:GETS) { + return Event:L1_GETS; + } else if(type == CoherenceRequestType:GETS_ESCAPE) { + return Event:L1_GETS_ESCAPE; + } else if(type == CoherenceRequestType:GET_INSTR) { + return Event:L1_GET_INSTR; + } else if(type == CoherenceRequestType:GET_INSTR_ESCAPE) { + return Event:L1_GET_INSTR_ESCAPE; + } else if (type == CoherenceRequestType:GETX) { + return Event:L1_GETX; + } else if(type == CoherenceRequestType:GETX_ESCAPE) { + return Event:L1_GETX_ESCAPE; + } else if (type == CoherenceRequestType:UPGRADE) { + if ( isL2CacheTagPresent(addr) && getL2CacheEntry(addr).Sharers.isElement(requestor) ) { + return Event:L1_UPGRADE; + } else { + return Event:L1_GETX; + } + } else if (type == CoherenceRequestType:PUTX) { + if ( isL2CacheTagPresent(addr) && getL2CacheEntry(addr).L1PutsPending > 0) { + getL2CacheEntry(addr).L1PutsPending := getL2CacheEntry(addr).L1PutsPending - 1; + DEBUG_EXPR("PUTX PutSPending "); + DEBUG_EXPR(getL2CacheEntry(addr).L1PutsPending); + } + if (isSharer(addr, requestor)) { + return Event:L1_PUTX; + } else { + return Event:L1_PUTX_old; + } + } else if (type == CoherenceRequestType:PUTS) { + if ( isL2CacheTagPresent(addr) && getL2CacheEntry(addr).L1PutsPending > 0) { + getL2CacheEntry(addr).L1PutsPending := getL2CacheEntry(addr).L1PutsPending - 1; + DEBUG_EXPR("PUTS PutSPending "); + DEBUG_EXPR(getL2CacheEntry(addr).L1PutsPending); + } + if (isSharer(addr, requestor)) { + return Event:L1_PUTS; + } else { + return Event:L1_PUTS_old; + } + } else { + DEBUG_EXPR(addr); + DEBUG_EXPR(type); + error("Invalid L1 forwarded request type"); + } + } + + // ** OUT_PORTS ** + + out_port(L1RequestIntraChipL2Network_out, RequestMsg, L1RequestFromL2Cache); + out_port(DirRequestIntraChipL2Network_out, RequestMsg, DirRequestFromL2Cache); + out_port(responseIntraChipL2Network_out, ResponseMsg, responseFromL2Cache); + + + // Response IntraChip L2 Network - response msg to this particular L2 bank + in_port(responseIntraChipL2Network_in, ResponseMsg, responseToL2Cache) { + if (responseIntraChipL2Network_in.isReady()) { + peek(responseIntraChipL2Network_in, ResponseMsg) { + // test wether it's from a local L1 or an off chip source + assert(in_msg.Destination.isElement(machineID)); + if(machineIDToMachineType(in_msg.Sender) == MachineType:L1Cache) { + if(in_msg.Type == CoherenceResponseType:DATA) { + if (in_msg.Dirty) { + trigger(Event:WB_Data, in_msg.Address); + } else { + trigger(Event:WB_Data_clean, in_msg.Address); + } + } else if (in_msg.Type == CoherenceResponseType:ACK) { + if ((L2_TBEs[in_msg.Address].pendingAcks - in_msg.AckCount) == 0) { + // check whether any previous responses have been NACKs + if(L2_TBEs[in_msg.Address].nack == false) { + trigger(Event:Ack_all, in_msg.Address); + } + else { + // at least one nack received + trigger(Event:Nack_all, in_msg.Address); + } + } else { + trigger(Event:Ack, in_msg.Address); + } + // for NACKs + } else if (in_msg.Type == CoherenceResponseType:NACK) { + if ((L2_TBEs[in_msg.Address].pendingAcks - in_msg.AckCount) == 0) { + trigger(Event:Nack_all, in_msg.Address); + } else { + trigger(Event:Nack, in_msg.Address); + } + } else { + error("unknown message type"); + } + + } else { // external message + if(in_msg.Type == CoherenceResponseType:MEMORY_DATA) { + trigger(Event:Mem_Data, in_msg.Address); // L2 now has data and all off-chip acks + } else if(in_msg.Type == CoherenceResponseType:MEMORY_ACK) { + trigger(Event:Mem_Ack, in_msg.Address); // L2 now has data and all off-chip acks + } else { + error("unknown message type"); + } + } + } + } // if not ready, do nothing + } + + // L1 Request + in_port(L1RequestIntraChipL2Network_in, RequestMsg, L1RequestToL2Cache) { + if(L1RequestIntraChipL2Network_in.isReady()) { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + /* + DEBUG_EXPR(in_msg.Address); + DEBUG_EXPR(id); + DEBUG_EXPR(getState(in_msg.Address)); + DEBUG_EXPR(in_msg.Requestor); + DEBUG_EXPR(in_msg.Type); + DEBUG_EXPR(in_msg.Destination); + */ + assert(machineIDToMachineType(in_msg.Requestor) == MachineType:L1Cache); + assert(in_msg.Destination.isElement(machineID)); + if (L2cacheMemory.isTagPresent(in_msg.Address)) { + // The L2 contains the block, so proceeded with handling the request + trigger(L1Cache_request_type_to_event(in_msg.Type, in_msg.Address, in_msg.Requestor), in_msg.Address); + } else { + if (L2cacheMemory.cacheAvail(in_msg.Address)) { + // L2 does't have the line, but we have space for it in the L2 + trigger(L1Cache_request_type_to_event(in_msg.Type, in_msg.Address, in_msg.Requestor), in_msg.Address); + } else { + // No room in the L2, so we need to make room before handling the request + if (L2cacheMemory[ L2cacheMemory.cacheProbe(in_msg.Address) ].Dirty ) { + // check whether block is transactional + if(L2cacheMemory[ L2cacheMemory.cacheProbe(in_msg.Address) ].Trans == true){ + trigger(Event:L2_Replacement_XACT, L2cacheMemory.cacheProbe(in_msg.Address)); + } + else{ + trigger(Event:L2_Replacement, L2cacheMemory.cacheProbe(in_msg.Address)); + } + } else { + // check whether block is transactional + if(L2cacheMemory[ L2cacheMemory.cacheProbe(in_msg.Address) ].Trans == true){ + trigger(Event:L2_Replacement_clean_XACT, L2cacheMemory.cacheProbe(in_msg.Address)); + } + else{ + trigger(Event:L2_Replacement_clean, L2cacheMemory.cacheProbe(in_msg.Address)); + } + } + } + } + } + } + } + + in_port(L1unblockNetwork_in, ResponseMsg, unblockToL2Cache) { + if(L1unblockNetwork_in.isReady()) { + peek(L1unblockNetwork_in, ResponseMsg) { + assert(in_msg.Destination.isElement(machineID)); + if (in_msg.Type == CoherenceResponseType:EXCLUSIVE_UNBLOCK) { + if (in_msg.RemoveLastOwnerFromDir == true) { + if (isSharer(in_msg.Address,in_msg.LastOwnerID)) { + trigger(Event:Exclusive_Unblock_WaitPUTold, in_msg.Address); + } + else { // PUT arrived, requestor already removed from dir + trigger(Event:Exclusive_Unblock, in_msg.Address); + } + } + else { + trigger(Event:Exclusive_Unblock, in_msg.Address); + } + } else if (in_msg.Type == CoherenceResponseType:UNBLOCK) { + if (in_msg.RemoveLastOwnerFromDir == true) { + if (isSharer(in_msg.Address,in_msg.LastOwnerID)) { + trigger(Event:Unblock_WaitPUTold, in_msg.Address); + } + else { // PUT arrived, requestor already removed from dir + trigger(Event:Unblock, in_msg.Address); + } + } + else { + trigger(Event:Unblock, in_msg.Address); + } + } else if (in_msg.Type == CoherenceResponseType:UNBLOCK_CANCEL) { + trigger(Event:Unblock_Cancel, in_msg.Address); + } else { + error("unknown unblock message"); + } + } + } + } + + // ACTIONS + + action(a_issueFetchToMemory, "a", desc="fetch data from memory") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(DirRequestIntraChipL2Network_out, RequestMsg, latency="L2_REQUEST_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceRequestType:GETS; + out_msg.Requestor := machineID; + out_msg.Destination.add(map_Address_to_Directory(address)); + out_msg.MessageSize := MessageSizeType:Control; + } + } + } + + action(b_forwardRequestToExclusive, "b", desc="Forward request to the exclusive L1") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := in_msg.Type; + out_msg.Requestor := in_msg.Requestor; + out_msg.Destination.add(L2cacheMemory[address].Exclusive); + out_msg.MessageSize := MessageSizeType:Request_Control; + // also pass along timestamp + out_msg.Timestamp := in_msg.Timestamp; + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + } + } + } + + action(c_exclusiveReplacement, "c", desc="Send data to memory") { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_RESPONSE_LATENCY") { + out_msg.Address := address; + out_msg.Type := CoherenceResponseType:MEMORY_DATA; + out_msg.Sender := machineID; + out_msg.Destination.add(map_Address_to_Directory(address)); + out_msg.DataBlk := getL2CacheEntry(address).DataBlk; + out_msg.Dirty := getL2CacheEntry(address).Dirty; + out_msg.MessageSize := MessageSizeType:Response_Data; + } + } + + action(ct_exclusiveReplacementFromTBE, "ct", desc="Send data to memory") { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_RESPONSE_LATENCY") { + out_msg.Address := address; + out_msg.Type := CoherenceResponseType:MEMORY_DATA; + out_msg.Sender := machineID; + out_msg.Destination.add(map_Address_to_Directory(address)); + out_msg.DataBlk := L2_TBEs[address].DataBlk; + out_msg.Dirty := L2_TBEs[address].Dirty; + out_msg.MessageSize := MessageSizeType:Response_Data; + } + } + + + //************Transactional memory actions ************** + //broadcast a write filter lookup request to all L1s except for the requestor + action(a_checkL1WriteFiltersExceptRequestor, "wr", desc="Broadcast a Write Filter lookup request"){ + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceRequestType:CHECK_WRITE_FILTER; + // make L1s forward responses to requestor + out_msg.Requestor := in_msg.Requestor; + out_msg.Destination := getLocalL1IDs(machineID); + // don't broadcast to requestor + out_msg.Destination.remove(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Response_Control; + // also pass along timestamp of requestor + out_msg.Timestamp := in_msg.Timestamp; + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + } + } + } + + //broadcast a read + write filter lookup request to all L1s except for the requestor + action(a_checkL1ReadWriteFiltersExceptRequestor, "rwr", desc="Broadcast a Read + Write Filter lookup request"){ + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceRequestType:CHECK_READ_WRITE_FILTER; + // make L1 forward responses to requestor + out_msg.Requestor := in_msg.Requestor; + out_msg.Destination := getLocalL1IDs(machineID); + // don't broadcast to requestor + out_msg.Destination.remove(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Response_Control; + // also pass along timestamp of requestor + out_msg.Timestamp := in_msg.Timestamp; + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + } + } + } + + // These are to send out filter checks to those NACKers in our sharers or exclusive ptr list + action(a_checkNackerL1WriteFiltersExceptRequestor, "wrn", desc="Broadcast a Write Filter lookup request"){ + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + // check if the L2 miss bit is set - if it is, send check filter requests to those in our Sharers list only + if(getL2CacheEntry(address).L2Miss == true){ + // check whether we are the only sharer on the list. If so, no need to broadcast. + if(isSharer(address, in_msg.Requestor) == true && isOneSharerLeft(address, in_msg.Requestor) == true){ + // no filter check needed + APPEND_TRANSITION_COMMENT("L2 Miss: No need to check L1 write filter "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + } + else{ + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceRequestType:CHECK_WRITE_FILTER; + // make L1s forward responses to requestor + out_msg.Requestor := in_msg.Requestor; + assert(getL2CacheEntry(address).Sharers.count() > 0); + out_msg.Destination := getL2CacheEntry(address).Sharers; + // don't broadcast to requestor + out_msg.Destination.remove(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Response_Control; + // also pass along timestamp of requestor + out_msg.Timestamp := in_msg.Timestamp; + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" dest: "); + APPEND_TRANSITION_COMMENT(out_msg.Destination); + } + } + } + else{ + // This is a read request, so check whether we have a writer + if(getL2CacheEntry(address).Sharers.count() == 0 && getL2CacheEntry(address).Exclusive != in_msg.Requestor){ + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="L2_TAG_LATENCY") { + // we have a writer, and it is not us + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceRequestType:CHECK_WRITE_FILTER; + // make L1s forward responses to requestor + out_msg.Requestor := in_msg.Requestor; + out_msg.Destination.add(getL2CacheEntry(address).Exclusive); + out_msg.MessageSize := MessageSizeType:Response_Control; + // also pass along timestamp of requestor + out_msg.Timestamp := in_msg.Timestamp; + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" dest: "); + APPEND_TRANSITION_COMMENT(out_msg.Destination); + } + } + else{ + APPEND_TRANSITION_COMMENT("L1 replacement: No need to check L1 write filter"); + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" exclusive: "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Exclusive); + // we should not have any sharers + assert( getL2CacheEntry(address).Sharers.count() == 0 ); + } + } + } + } + + action(a_checkNackerL1ReadWriteFiltersExceptRequestor, "wrrn", desc="Broadcast a Read + Write Filter lookup request"){ + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + // check if the L2 miss bit is set - if it is, send check filter requests to those in our Sharers list only + if(getL2CacheEntry(address).L2Miss == true){ + // check whether we are the only sharer on the list. If so, no need to broadcast. + if(isSharer(address, in_msg.Requestor) == true && isOneSharerLeft(address, in_msg.Requestor) == true){ + // no filter check needed + APPEND_TRANSITION_COMMENT("L2 Miss: No need to check L1 read/write filter"); + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + } + else{ + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceRequestType:CHECK_READ_WRITE_FILTER; + // make L1s forward responses to requestor + out_msg.Requestor := in_msg.Requestor; + assert(getL2CacheEntry(address).Sharers.count() > 0); + out_msg.Destination := getL2CacheEntry(address).Sharers; + // don't broadcast to requestor + out_msg.Destination.remove(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Response_Control; + // also pass along timestamp of requestor + out_msg.Timestamp := in_msg.Timestamp; + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" dest: "); + APPEND_TRANSITION_COMMENT(out_msg.Destination); + } + } + } + else{ + // This is a write request, so check whether we have readers not including us or a writer that is not us + if(getL2CacheEntry(address).Sharers.count() == 0 && getL2CacheEntry(address).Exclusive != in_msg.Requestor){ + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="L2_TAG_LATENCY") { + // we have a writer, and it is not us + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceRequestType:CHECK_READ_WRITE_FILTER; + // make L1s forward responses to requestor + out_msg.Requestor := in_msg.Requestor; + out_msg.Destination.add(getL2CacheEntry(address).Exclusive); + out_msg.MessageSize := MessageSizeType:Response_Control; + // also pass along timestamp of requestor + out_msg.Timestamp := in_msg.Timestamp; + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" dest: "); + APPEND_TRANSITION_COMMENT(out_msg.Destination); + } + } + else if(getL2CacheEntry(address).Sharers.count() > 0){ + // this should never happen - since we allow silent S replacements but we always track exclusive L1 + assert(false); + if(isSharer(address, in_msg.Requestor) == true && isOneSharerLeft(address, in_msg.Requestor) == true){ + // no filter check needed + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" L1 replacement: No need to check L1 read/write filter - we are only reader"); + } + else{ + // reader(s) exist that is not us + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceRequestType:CHECK_READ_WRITE_FILTER; + // make L1s forward responses to requestor + out_msg.Requestor := in_msg.Requestor; + out_msg.Destination := getL2CacheEntry(address).Sharers; + // don't check our own filter + out_msg.Destination.remove(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Response_Control; + // also pass along timestamp of requestor + out_msg.Timestamp := in_msg.Timestamp; + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" dest: "); + APPEND_TRANSITION_COMMENT(out_msg.Destination); + } + } + } + else{ + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Sharers); + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Exclusive); + APPEND_TRANSITION_COMMENT(" L1 replacement: No need to check L1 read/write filter"); + } + } + } + } + + // send data but force L1 requestor to wait for filter responses + action(f_sendDataToGetSRequestor, "f", desc="Send data from cache to reqeustor") { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_RESPONSE_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := L2_TBEs[address].PhysicalAddress; + out_msg.Type := CoherenceResponseType:L2_DATA; + out_msg.Sender := machineID; + out_msg.Destination := L2_TBEs[address].L1_GetS_IDs; // internal nodes + out_msg.DataBlk := getL2CacheEntry(address).DataBlk; + out_msg.Dirty := getL2CacheEntry(address).Dirty; + out_msg.MessageSize := MessageSizeType:Response_Data; + + // wait for the filter responses from other L1s + out_msg.AckCount := 0 - (numberOfL1CachePerChip() - 1); + } + } + + // send exclusive data + action(f_sendExclusiveDataToGetSRequestor, "fx", desc="Send data from cache to reqeustor") { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_RESPONSE_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := L2_TBEs[address].PhysicalAddress; + out_msg.Type := CoherenceResponseType:L2_DATA_EXCLUSIVE; + out_msg.Sender := machineID; + out_msg.Destination := L2_TBEs[address].L1_GetS_IDs; // internal nodes + out_msg.DataBlk := getL2CacheEntry(address).DataBlk; + out_msg.Dirty := getL2CacheEntry(address).Dirty; + out_msg.MessageSize := MessageSizeType:Response_Data; + + // wait for the filter responses from other L1s + out_msg.AckCount := 0 - (numberOfL1CachePerChip() - 1); + } + } + + action(f_sendDataToGetXRequestor, "fxx", desc="Send data from cache to reqeustor") { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_RESPONSE_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := L2_TBEs[address].PhysicalAddress; + out_msg.Type := CoherenceResponseType:L2_DATA; + out_msg.Sender := machineID; + out_msg.Destination.add(L2_TBEs[address].L1_GetX_ID); // internal nodes + out_msg.DataBlk := getL2CacheEntry(address).DataBlk; + out_msg.Dirty := getL2CacheEntry(address).Dirty; + out_msg.MessageSize := MessageSizeType:Response_Data; + + // wait for the filter responses from other L1s + out_msg.AckCount := 0 - (numberOfL1CachePerChip() - 1); + } + } + + action(f_sendDataToRequestor, "fd", desc="Send data from cache to reqeustor") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_RESPONSE_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceResponseType:L2_DATA; + out_msg.Sender := machineID; + out_msg.Destination.add(in_msg.Requestor); + out_msg.DataBlk := getL2CacheEntry(address).DataBlk; + out_msg.Dirty := getL2CacheEntry(address).Dirty; + out_msg.MessageSize := MessageSizeType:Response_Data; + + // different ack counts for different situations + if(in_msg.Type == CoherenceRequestType:GET_INSTR_ESCAPE || in_msg.Type == CoherenceRequestType:GETX_ESCAPE){ + // no acks needed + out_msg.AckCount := 0; + } + else{ + + // ORIGINAL + if( false ) { + out_msg.AckCount := 0 - (numberOfL1CachePerChip() - 1); + } + + else{ + // NEW*** + // differentiate btw read and write requests + if(in_msg.Type == CoherenceRequestType:GET_INSTR){ + if(getL2CacheEntry(address).L2Miss == true){ + // check whether we are the only sharer on the list. If so, no need to broadcast. + if(isSharer(address, in_msg.Requestor) == true && isOneSharerLeft(address, in_msg.Requestor) == true){ + // no filter check needed + APPEND_TRANSITION_COMMENT("We are only sharer"); + out_msg.AckCount := 0; + } + else{ + // wait for ACKs from the other NACKers + out_msg.AckCount := 0 - getL2CacheEntry(address).Sharers.count(); + if(isSharer(address, in_msg.Requestor)){ + // don't include us + out_msg.AckCount := out_msg.AckCount + 1; + } + APPEND_TRANSITION_COMMENT("Nackers exist"); + } + } + else{ + // This is a read request, so check whether we have a writer + if(getL2CacheEntry(address).Sharers.count() == 0 && getL2CacheEntry(address).Exclusive != in_msg.Requestor){ + // we have a writer and it is not us + out_msg.AckCount := 0 - 1; + + APPEND_TRANSITION_COMMENT(" Writer exists "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Exclusive); + } + else{ + // we should have no sharers! + assert(getL2CacheEntry(address).Sharers.count() == 0); + assert(getL2CacheEntry(address).Exclusive == in_msg.Requestor); + + APPEND_TRANSITION_COMMENT(" Sharers or we are writer exist, ok to read "); + APPEND_TRANSITION_COMMENT(" sharers: "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Sharers); + APPEND_TRANSITION_COMMENT(" exclusive: "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Exclusive); + out_msg.AckCount := 0; + } + } + } + else if(in_msg.Type == CoherenceRequestType:GETX){ + if(getL2CacheEntry(address).L2Miss == true){ + // check whether we are the only sharer on the list. If so, no need to broadcast. + if(isSharer(address, in_msg.Requestor) == true && isOneSharerLeft(address, in_msg.Requestor) == true){ + // no filter check needed + APPEND_TRANSITION_COMMENT(" L2Miss and we are only sharer "); + out_msg.AckCount := 0; + } + else{ + // nackers exist + out_msg.AckCount := 0 - getL2CacheEntry(address).Sharers.count(); + if(isSharer(address, in_msg.Requestor)){ + // don't include us + out_msg.AckCount := out_msg.AckCount + 1; + } + APPEND_TRANSITION_COMMENT("Nackers exist"); + } + } + else{ + // This is a write request, so check whether we have readers not including us or a writer that is not us + if(getL2CacheEntry(address).Sharers.count() == 0 && getL2CacheEntry(address).Exclusive != in_msg.Requestor){ + // we have a writer and it is not us + out_msg.AckCount := 0 - 1; + + APPEND_TRANSITION_COMMENT(" Writer exists "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Exclusive); + + } + else if(getL2CacheEntry(address).Sharers.count() > 0){ + // this shouldn't be possible - we always track exclusive owner, but allow silent S replacements + assert(false); + + if(isSharer(address, in_msg.Requestor) == true && isOneSharerLeft(address, in_msg.Requestor) == true){ + // no filter check needed + APPEND_TRANSITION_COMMENT(" L1 replacement: No need to check L1 read/write filter - we are only reader"); + out_msg.AckCount := 0; + } + else{ + // reader(s) exist that is not us + out_msg.AckCount := 0 - getL2CacheEntry(address).Sharers.count(); + if(isSharer(address, in_msg.Requestor)){ + // don't include us + out_msg.AckCount := out_msg.AckCount + 1; + } + APPEND_TRANSITION_COMMENT(" Readers exist "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Sharers); + } + } + else{ + // we should always have no sharers! + assert(getL2CacheEntry(address).Sharers.count() == 0); + assert(getL2CacheEntry(address).Exclusive == in_msg.Requestor); + + out_msg.AckCount := 0; + + APPEND_TRANSITION_COMMENT(" sharers: "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Sharers); + APPEND_TRANSITION_COMMENT(" exclusive: "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Exclusive); + APPEND_TRANSITION_COMMENT(" L1 replacement: No need to check L1 read/write filter"); + } + } + } // for GETX + else{ + // unknown request type + APPEND_TRANSITION_COMMENT(in_msg.Type); + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + assert(false); + } + } + } // for original vs new code + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" AckCount: "); + APPEND_TRANSITION_COMMENT(out_msg.AckCount); + } + } + } + + action(f_sendExclusiveDataToRequestor, "fdx", desc="Send data from cache to reqeustor") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_RESPONSE_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceResponseType:L2_DATA_EXCLUSIVE; + out_msg.Sender := machineID; + out_msg.Destination.add(in_msg.Requestor); + out_msg.DataBlk := getL2CacheEntry(address).DataBlk; + out_msg.Dirty := getL2CacheEntry(address).Dirty; + out_msg.MessageSize := MessageSizeType:Response_Data; + + // different ack counts depending on situation + // IMPORTANT: assuming data sent exclusively for GETS request + if(in_msg.Type == CoherenceRequestType:GETS_ESCAPE){ + // no acks needed + out_msg.AckCount := 0; + } + else{ + + // ORIGINAL : + if( false ){ + // request filter checks from all L1s + out_msg.AckCount := 0 - (numberOfL1CachePerChip() - 1); + } + else{ + // NEW*** + if(getL2CacheEntry(address).L2Miss == true){ + // check whether we are the only sharer on the list. If so, no need to broadcast. + if(isSharer(address, in_msg.Requestor) == true && isOneSharerLeft(address, in_msg.Requestor) == true){ + // no filter check needed + APPEND_TRANSITION_COMMENT("We are only sharer"); + out_msg.AckCount := 0; + } + else{ + // wait for ACKs from the other NACKers + out_msg.AckCount := 0 - getL2CacheEntry(address).Sharers.count(); + if(isSharer(address, in_msg.Requestor)){ + // don't include us + out_msg.AckCount := out_msg.AckCount + 1; + } + APPEND_TRANSITION_COMMENT("Nackers exist"); + } + } + else{ + // This is a read request, so check whether we have a writer + if(getL2CacheEntry(address).Sharers.count() == 0 && getL2CacheEntry(address).Exclusive != in_msg.Requestor){ + // we have a writer and it is not us + out_msg.AckCount := 0 - 1; + + APPEND_TRANSITION_COMMENT(" Writer exists "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Exclusive); + } + else{ + // we should always have no sharers! + APPEND_TRANSITION_COMMENT(address); + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" sharers: "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Sharers); + + DEBUG_EXPR(address); + DEBUG_EXPR(" requestor: "); + DEBUG_EXPR(in_msg.Requestor); + DEBUG_EXPR(" sharers: "); + DEBUG_EXPR(getL2CacheEntry(address).Sharers); + + assert(getL2CacheEntry(address).Sharers.count() == 0); + assert(getL2CacheEntry(address).Exclusive == in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" Sharers exist or we are writer, ok to read "); + out_msg.AckCount := 0; + } + } + } // for orginal vs new code + } + + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" AckCount: "); + APPEND_TRANSITION_COMMENT(out_msg.AckCount); + } + } + } + + // send an accumulated ACK to requestor when we don't care about checking filters (for escape actions) + action(f_sendAccumulatedAckToRequestor, "faa", desc="Send ACKs to requestor") { + // special case: don't send ACK if uniprocessor, since we don't need it (just send data) + if((numberOfL1CachePerChip() - 1) > 0){ + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="1") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceResponseType:ACK; + out_msg.Sender := machineID; + out_msg.Destination.add(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Response_Control; + // count all L1s except requestor + out_msg.AckCount := numberOfL1CachePerChip() - 1; + APPEND_TRANSITION_COMMENT(" Total L1s: "); + APPEND_TRANSITION_COMMENT(numberOfL1CachePerChip()); + APPEND_TRANSITION_COMMENT(" Total ACKS: "); + APPEND_TRANSITION_COMMENT(out_msg.AckCount); + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + } + } + } + } + + // special INV used when we receive an escape action request. Sharers cannot NACK this invalidate. + action(fwm_sendFwdInvEscapeToSharersMinusRequestor, "fwme", desc="invalidate sharers for request, requestor is sharer") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="1") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceRequestType:INV_ESCAPE; + out_msg.Requestor := in_msg.Requestor; + out_msg.Destination := L2cacheMemory[address].Sharers; + out_msg.Destination.remove(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Request_Control; + //also pass along timestamp + out_msg.Timestamp := in_msg.Timestamp; + } + } + } + + action(f_profileRequestor, "prq", desc="Profiles the requestor") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + APPEND_TRANSITION_COMMENT(" requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + } + } + + // marks the L2 block as transactional if request was transactional + action(f_markBlockTransIfTrans, "\mbt", desc="Mark an L2 block as transactional") { + peek(L1unblockNetwork_in, ResponseMsg) { + if(in_msg.Transactional == true){ + L2cacheMemory[address].Trans := true; + } + } + } + + action(q_profileOverflow, "po", desc="profile the overflowed block"){ + profileOverflow(address, machineID); + } + + action(p_profileRequest, "pcc", desc="Profile request msg") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + APPEND_TRANSITION_COMMENT(" request: Timestamp: "); + APPEND_TRANSITION_COMMENT(in_msg.Timestamp); + APPEND_TRANSITION_COMMENT(" Requestor: "); + APPEND_TRANSITION_COMMENT(in_msg.Requestor); + APPEND_TRANSITION_COMMENT(" Dest: "); + APPEND_TRANSITION_COMMENT(in_msg.Destination); + APPEND_TRANSITION_COMMENT(" PA: "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + APPEND_TRANSITION_COMMENT(" Type: "); + APPEND_TRANSITION_COMMENT(in_msg.Type); + APPEND_TRANSITION_COMMENT(" Mode: "); + APPEND_TRANSITION_COMMENT(in_msg.AccessMode); + APPEND_TRANSITION_COMMENT(" PF: "); + APPEND_TRANSITION_COMMENT(in_msg.Prefetch); + } + } + + //********************************END*************************** + + action(d_sendDataToRequestor, "d", desc="Send data from cache to reqeustor") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_RESPONSE_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceResponseType:L2_DATA; + out_msg.Sender := machineID; + out_msg.Destination.add(in_msg.Requestor); + out_msg.DataBlk := getL2CacheEntry(address).DataBlk; + out_msg.Dirty := getL2CacheEntry(address).Dirty; + out_msg.MessageSize := MessageSizeType:Response_Data; + + out_msg.AckCount := 0 - getL2CacheEntry(address).Sharers.count(); + if (getL2CacheEntry(address).Sharers.isElement(in_msg.Requestor)) { + out_msg.AckCount := out_msg.AckCount + 1; + } + APPEND_TRANSITION_COMMENT(" AckCount: "); + APPEND_TRANSITION_COMMENT(out_msg.AckCount); + } + } + } + + // use DATA instead of L2_DATA because L1 doesn't need to wait for acks from L1 filters in this case + action(ds_sendSharedDataToRequestor, "ds", desc="Send data from cache to reqeustor") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_RESPONSE_LATENCY") { + out_msg.Address := address; + out_msg.Type := CoherenceResponseType:L2_DATA; + out_msg.Sender := machineID; + out_msg.Destination.add(in_msg.Requestor); + out_msg.DataBlk := getL2CacheEntry(address).DataBlk; + out_msg.Dirty := getL2CacheEntry(address).Dirty; + out_msg.MessageSize := MessageSizeType:Response_Data; + // no ACKS needed because no possible conflicts + out_msg.AckCount := 0; + } + } + } + + action(f_sendInvToSharers, "fsi", desc="invalidate sharers for L2 replacement") { + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.Type := CoherenceRequestType:REPLACE; + out_msg.Requestor := machineID; + out_msg.Destination := L2cacheMemory[address].Sharers; + out_msg.MessageSize := MessageSizeType:Request_Control; + } + } + + action(fwm_sendFwdInvToSharersMinusRequestor, "fwm", desc="invalidate sharers for request, requestor is sharer") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(L1RequestIntraChipL2Network_out, RequestMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceRequestType:INV; + out_msg.Requestor := in_msg.Requestor; + out_msg.Destination := L2cacheMemory[address].Sharers; + out_msg.Destination.remove(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Request_Control; + //also pass along timestamp + out_msg.Timestamp := in_msg.Timestamp; + APPEND_TRANSITION_COMMENT(" Sharers: "); + APPEND_TRANSITION_COMMENT(L2cacheMemory[address].Sharers); + } + } + } + + // OTHER ACTIONS + action(i_allocateTBE, "i", desc="Allocate TBE for internal/external request(isPrefetch=0, number of invalidates=0)") { + check_allocate(L2_TBEs); + L2_TBEs.allocate(address); + L2_TBEs[address].L1_GetS_IDs.clear(); + L2_TBEs[address].DataBlk := getL2CacheEntry(address).DataBlk; + L2_TBEs[address].Dirty := getL2CacheEntry(address).Dirty; + L2_TBEs[address].pendingAcks := getL2CacheEntry(address).Sharers.count(); + } + + action(i_setTBEPhysicalAddress, "ia", desc="Sets the physical address field of the TBE"){ + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + L2_TBEs[address].PhysicalAddress := in_msg.PhysicalAddress; + } + } + + action(s_deallocateTBE, "s", desc="Deallocate external TBE") { + L2_TBEs.deallocate(address); + } + + action(jj_popL1RequestQueue, "\j", desc="Pop incoming L1 request queue") { + profileMsgDelay(0, L1RequestIntraChipL2Network_in.dequeue_getDelayCycles()); + } + + action(k_popUnblockQueue, "k", desc="Pop incoming unblock queue") { + profileMsgDelay(0, L1unblockNetwork_in.dequeue_getDelayCycles()); + } + + + action(o_popIncomingResponseQueue, "o", desc="Pop Incoming Response queue") { + profileMsgDelay(3, responseIntraChipL2Network_in.dequeue_getDelayCycles()); + } + + + action(m_writeDataToCache, "m", desc="Write data from response queue to cache") { + peek(responseIntraChipL2Network_in, ResponseMsg) { + getL2CacheEntry(address).DataBlk := in_msg.DataBlk; + getL2CacheEntry(address).Dirty := in_msg.Dirty; + // reset the L2 miss bit + getL2CacheEntry(address).L2Miss := false; + } + } + + // Sets the L2Miss bit in the L2 entry - indicates data was sourced from memory + action(m_markL2MissBit, "mi", desc="Set the entry's L2 Miss bit") { + getL2CacheEntry(address).L2Miss := true; + } + + action(m_copyNackersIntoSharers, "mn", desc="Copy the NACKers list into our sharers list") { + peek(L1unblockNetwork_in, ResponseMsg) { + assert(in_msg.Nackers.count() > 0); + getL2CacheEntry(address).Sharers.clear(); + // only need to copy into sharers list if we are in special state of "multicast" filter checks + if(getL2CacheEntry(address).L2Miss == true){ + getL2CacheEntry(address).Sharers := in_msg.Nackers; + APPEND_TRANSITION_COMMENT(" Unblocker: "); + APPEND_TRANSITION_COMMENT(in_msg.Sender); + APPEND_TRANSITION_COMMENT(" Nackers: "); + APPEND_TRANSITION_COMMENT(getL2CacheEntry(address).Sharers); + } + } + } + + action(mr_writeDataToCacheFromRequest, "mr", desc="Write data from response queue to cache") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + getL2CacheEntry(address).DataBlk := in_msg.DataBlk; + getL2CacheEntry(address).Dirty := in_msg.Dirty; + // reset the L2 miss bit + getL2CacheEntry(address).L2Miss := false; + } + } + + action(q_updateAck, "q", desc="update pending ack count") { + peek(responseIntraChipL2Network_in, ResponseMsg) { + L2_TBEs[address].pendingAcks := L2_TBEs[address].pendingAcks - in_msg.AckCount; + APPEND_TRANSITION_COMMENT(in_msg.AckCount); + APPEND_TRANSITION_COMMENT(" p: "); + APPEND_TRANSITION_COMMENT(L2_TBEs[address].pendingAcks); + } + } + + // For transactional memory. If received NACK instead of ACK + action(q_updateNack, "qn", desc="update pending ack count") { + peek(responseIntraChipL2Network_in, ResponseMsg) { + // set flag indicating we have seen NACK + L2_TBEs[address].nack := true; + L2_TBEs[address].pendingAcks := L2_TBEs[address].pendingAcks - in_msg.AckCount; + APPEND_TRANSITION_COMMENT(in_msg.AckCount); + APPEND_TRANSITION_COMMENT(" p: "); + APPEND_TRANSITION_COMMENT(L2_TBEs[address].pendingAcks); + } + } + + action(qq_writeDataToTBE, "\qq", desc="Write data from response queue to TBE") { + peek(responseIntraChipL2Network_in, ResponseMsg) { + L2_TBEs[address].DataBlk := in_msg.DataBlk; + L2_TBEs[address].Dirty := in_msg.Dirty; + } + } + + + action(z_stall, "z", desc="Stall") { + } + + + action(ss_recordGetSL1ID, "\s", desc="Record L1 GetS for load response") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + L2_TBEs[address].L1_GetS_IDs.add(in_msg.Requestor); + } + } + + action(xx_recordGetXL1ID, "\x", desc="Record L1 GetX for store response") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + L2_TBEs[address].L1_GetX_ID := in_msg.Requestor; + } + } + + action(set_setMRU, "\set", desc="set the MRU entry") { + L2cacheMemory.setMRU(address); + } + + action(qq_allocateL2CacheBlock, "\q", desc="Set L2 cache tag equal to tag of block B.") { + if (L2cacheMemory.isTagPresent(address) == false) { + L2cacheMemory.allocate(address); + } + } + + action(rr_deallocateL2CacheBlock, "\r", desc="Deallocate L2 cache block. Sets the cache to not present, allowing a replacement in parallel with a fetch.") { + L2cacheMemory.deallocate(address); + } + + action(t_sendWBAck, "t", desc="Send writeback ACK") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.Type := CoherenceResponseType:WB_ACK; + out_msg.Sender := machineID; + out_msg.Destination.add(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Response_Control; + } + } + } + + action(ts_sendInvAckToUpgrader, "ts", desc="Send ACK to upgrader") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.PhysicalAddress := in_msg.PhysicalAddress; + out_msg.Type := CoherenceResponseType:ACK; + out_msg.Sender := machineID; + out_msg.Destination.add(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Response_Control; + // upgrader doesn't get ack from itself, hence the + 1 + out_msg.AckCount := 0 - getL2CacheEntry(address).Sharers.count() + 1; + APPEND_TRANSITION_COMMENT(" "); + APPEND_TRANSITION_COMMENT(in_msg.PhysicalAddress); + } + } + } + + // same as above, but send NACK instead of ACK + action(ts_sendInvNackToUpgrader, "tsn", desc="Send NACK to upgrader") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + enqueue(responseIntraChipL2Network_out, ResponseMsg, latency="L2_TAG_LATENCY") { + out_msg.Address := address; + out_msg.Type := CoherenceResponseType:NACK; + out_msg.Sender := machineID; + out_msg.Destination.add(in_msg.Requestor); + out_msg.MessageSize := MessageSizeType:Response_Control; + // upgrader doesn't get ack from itself, hence the + 1 + out_msg.AckCount := 0 - getL2CacheEntry(address).Sharers.count() + 1; + } + } + } + + action(uu_profileMiss, "\u", desc="Profile the demand miss") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + profile_L2Cache_miss(convertToGenericType(in_msg.Type), in_msg.AccessMode, MessageSizeTypeToInt(in_msg.MessageSize), in_msg.Prefetch, L1CacheMachIDToProcessorNum(in_msg.Requestor)); + } + } + + action(ww_profileMissNoDir, "\w", desc="Profile this transition at the L2 because Dir won't see the request") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + // profile_request(in_msg.L1CacheStateStr, getStateStr(address), "NA", getCoherenceRequestTypeStr(in_msg.Type)); + } + } + + + + action(nn_addSharer, "\n", desc="Add L1 sharer to list") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + addSharer(address, in_msg.Requestor); + APPEND_TRANSITION_COMMENT( getL2CacheEntry(address).Sharers ); + } + } + + action(nnu_addSharerFromUnblock, "\nu", desc="Add L1 sharer to list") { + peek(L1unblockNetwork_in, ResponseMsg) { + addSharer(address, in_msg.Sender); + if (in_msg.RemoveLastOwnerFromDir == true) { + // We do this to solve some races with PUTX + APPEND_TRANSITION_COMMENT("Last owner removed, it was "); + APPEND_TRANSITION_COMMENT(in_msg.LastOwnerID); + L2cacheMemory[address].Sharers.remove(in_msg.LastOwnerID); + assert(in_msg.LastOwnerID == L2cacheMemory[address].Exclusive); + } + } + } + + + action(kk_removeRequestSharer, "\k", desc="Remove L1 Request sharer from list") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + L2cacheMemory[address].Sharers.remove(in_msg.Requestor); + } + } + + action(ll_clearSharers, "\l", desc="Remove all L1 sharers from list") { + L2cacheMemory[address].Sharers.clear(); + } + + action(mmu_markExclusiveFromUnblock, "\mu", desc="set the exclusive owner") { + peek(L1unblockNetwork_in, ResponseMsg) { + if (in_msg.RemoveLastOwnerFromDir == true) { + // We do this to solve some races with PUTX + APPEND_TRANSITION_COMMENT(" Last owner removed, it was "); + APPEND_TRANSITION_COMMENT(in_msg.LastOwnerID); + assert(in_msg.LastOwnerID == L2cacheMemory[address].Exclusive); + } + L2cacheMemory[address].Sharers.clear(); + L2cacheMemory[address].Exclusive := in_msg.Sender; + addSharer(address, in_msg.Sender); + } + } + + action(zz_recycleL1RequestQueue, "zz", desc="recycle L1 request queue") { + peek(L1RequestIntraChipL2Network_in, RequestMsg) { + if (in_msg.Type == CoherenceRequestType:PUTX || in_msg.Type == CoherenceRequestType:PUTS) { + if (L2cacheMemory.isTagPresent(in_msg.Address)) { + getL2CacheEntry(in_msg.Address).L1PutsPending := getL2CacheEntry(in_msg.Address).L1PutsPending + 1; + DEBUG_EXPR("RECYCLE PutSPending "); + DEBUG_EXPR(getL2CacheEntry(in_msg.Address).L1PutsPending); + DEBUG_EXPR(in_msg.Type); + DEBUG_EXPR(in_msg.Requestor); + } + } + } + L1RequestIntraChipL2Network_in.recycle(); + } + + //***************************************************** + // TRANSITIONS + //***************************************************** + + /* Recycle while waiting for PUT */ + transition({PB_MT, PB_MT_IB, PB_SS}, {L1_GETS, L1_GET_INSTR, L1_GETX, L1_UPGRADE, L1_GETS_ESCAPE, L1_GETX_ESCAPE, L1_GET_INSTR_ESCAPE, L2_Replacement, L2_Replacement_clean, L2_Replacement_XACT, L2_Replacement_clean_XACT}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + transition({IM, IS, ISS, SS_MB, M_MB, ISS_MB, IS_SSB, MT_MB, MT_IIB, MT_IB, MT_SB, M_SSB, SS_SSB}, + {L2_Replacement, L2_Replacement_clean, L2_Replacement_XACT, L2_Replacement_clean_XACT}) { + zz_recycleL1RequestQueue; + } + + transition({SS_MB, M_MB, ISS_MB, IS_SSB, MT_MB, MT_IIB, MT_IB, MT_SB, M_SSB, SS_SSB}, + {L1_GETS, L1_GET_INSTR, L1_GETX, L1_UPGRADE, L1_GETS_ESCAPE, L1_GETX_ESCAPE, L1_GET_INSTR_ESCAPE}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + transition({NP, SS, M, M_I, MT_I, MCT_I, I_I, S_I, ISS, IS, IM, /*SS_MB,*/ SS_SSB, /* MT_MB, M_MB, ISS_MB,*/ IS_SSB, M_SSB, /*MT_IIB, */MT_IB/*, MT_SB*/}, {L1_PUTX,L1_PUTS}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + // occurs when L2 replacement raced with L1 replacement, and L2 finished its replacement first + transition({NP, M_I, MCT_I, I_I, S_I, IS, ISS, IM, SS, M, MT, IS_SSB, MT_IB, M_SSB, SS_SSB}, {L1_PUTX_old, L1_PUTS_old}){ + f_profileRequestor; + jj_popL1RequestQueue; + } + + // PUT from current (last) exclusive owner, that was replacing the line when it received Fwd req + transition(MT_I, {L1_PUTX_old, L1_PUTS_old}) { + f_profileRequestor; + jj_popL1RequestQueue; + } + + transition({SS, M, MT}, {L1_PUT_PENDING}) { // L1_PUT_ msg pending for the block, don't accept new requests until PUT is processed */ + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + //=============================================== + // BASE STATE - I + + // Transitions from I (Idle) + + // When L2 doesn't have block, need to send broadcasst to all L1s to check appropriate filter(s) + transition(NP, L1_GETS, ISS) { + p_profileRequest; + f_profileRequestor; + qq_allocateL2CacheBlock; + ll_clearSharers; + // will mark as exclusive when we get unblocked with success + //nn_addSharer; + i_allocateTBE; + i_setTBEPhysicalAddress; + ss_recordGetSL1ID; + a_issueFetchToMemory; + // for correctness we need to query both read + write filters + a_checkL1ReadWriteFiltersExceptRequestor; + uu_profileMiss; + jj_popL1RequestQueue; + } + + // no need to check filters, send accumulated ACK to requestor + transition(NP, L1_GETS_ESCAPE, ISS) { + p_profileRequest; + f_profileRequestor; + qq_allocateL2CacheBlock; + ll_clearSharers; + // will mark as exclusive when we get unblocked with success + //nn_addSharer; + i_allocateTBE; + i_setTBEPhysicalAddress; + ss_recordGetSL1ID; + a_issueFetchToMemory; + // send accumulated ACK + f_sendAccumulatedAckToRequestor; + uu_profileMiss; + jj_popL1RequestQueue; + } + + transition(NP, L1_GET_INSTR, IS) { + p_profileRequest; + f_profileRequestor; + qq_allocateL2CacheBlock; + ll_clearSharers; + nn_addSharer; + i_allocateTBE; + i_setTBEPhysicalAddress; + ss_recordGetSL1ID; + a_issueFetchToMemory; + // for correctness query the read + write filters + a_checkL1ReadWriteFiltersExceptRequestor; + uu_profileMiss; + jj_popL1RequestQueue; + } + + // no need to query filters, send accumluated ACK to requestor + transition(NP, L1_GET_INSTR_ESCAPE, IS) { + p_profileRequest; + f_profileRequestor; + qq_allocateL2CacheBlock; + ll_clearSharers; + nn_addSharer; + i_allocateTBE; + i_setTBEPhysicalAddress; + ss_recordGetSL1ID; + a_issueFetchToMemory; + // send accumulated ACK + f_sendAccumulatedAckToRequestor; + uu_profileMiss; + jj_popL1RequestQueue; + } + + transition(NP, L1_GETX, IM) { + p_profileRequest; + f_profileRequestor; + qq_allocateL2CacheBlock; + ll_clearSharers; + // nn_addSharer; + i_allocateTBE; + i_setTBEPhysicalAddress; + xx_recordGetXL1ID; + a_issueFetchToMemory; + // also query the L1 write and read filters + a_checkL1ReadWriteFiltersExceptRequestor; + uu_profileMiss; + jj_popL1RequestQueue; + } + + // don't check filters + transition(NP, L1_GETX_ESCAPE, IM) { + p_profileRequest; + f_profileRequestor; + qq_allocateL2CacheBlock; + ll_clearSharers; + // nn_addSharer; + i_allocateTBE; + i_setTBEPhysicalAddress; + xx_recordGetXL1ID; + a_issueFetchToMemory; + // send accumulated ACK to requestor + f_sendAccumulatedAckToRequestor; + uu_profileMiss; + jj_popL1RequestQueue; + } + + + // transitions from IS/IM + + // force L1s to respond success or failure + transition(ISS, Mem_Data, ISS_MB){ + m_writeDataToCache; + m_markL2MissBit; + // send exclusive data but force L1 to wait for filter responses + f_sendExclusiveDataToGetSRequestor; + s_deallocateTBE; + o_popIncomingResponseQueue; + } + + transition(IS, Mem_Data, IS_SSB){ + m_writeDataToCache; + m_markL2MissBit; + // send data but force L1 to wait for filter responses + f_sendDataToGetSRequestor; + s_deallocateTBE; + o_popIncomingResponseQueue; + } + + transition(IM, Mem_Data, ISS_MB){ + m_writeDataToCache; + m_markL2MissBit; + // send data but force L1 to wait for filter responses + f_sendDataToGetXRequestor; + s_deallocateTBE; + o_popIncomingResponseQueue; + } + + // disallow grouping of requestors. There is a correctness problem if we check the wrong + // filters as indicated by the original requestor. + transition({IS, ISS}, {L1_GETX, L1_GETS, L1_GET_INSTR, L1_GETX_ESCAPE, L1_GETS_ESCAPE, L1_GET_INSTR_ESCAPE}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + transition(IM, {L1_GETX, L1_GETS, L1_GET_INSTR, L1_GETX_ESCAPE, L1_GETS_ESCAPE, L1_GET_INSTR_ESCAPE}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + // transitions from SS + transition(SS, {L1_GETS, L1_GET_INSTR, L1_GETS_ESCAPE, L1_GET_INSTR_ESCAPE}, SS_SSB) { + p_profileRequest; + f_profileRequestor; + ds_sendSharedDataToRequestor; + nn_addSharer; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + // For isolation the L1 filters might return NACKs to the requestor + transition(SS, L1_GETX, SS_MB) { + p_profileRequest; + f_profileRequestor; + d_sendDataToRequestor; + fwm_sendFwdInvToSharersMinusRequestor; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + // send special INV to sharers - they have to invalidate + transition(SS, L1_GETX_ESCAPE, SS_MB) { + p_profileRequest; + f_profileRequestor; + d_sendDataToRequestor; + fwm_sendFwdInvEscapeToSharersMinusRequestor; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + // For isolation the L1 filters might return NACKs to the requestor + transition(SS, L1_UPGRADE, SS_MB) { + f_profileRequestor; + fwm_sendFwdInvToSharersMinusRequestor; + ts_sendInvAckToUpgrader; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + transition(SS, L2_Replacement_clean, I_I) { + i_allocateTBE; + f_sendInvToSharers; + rr_deallocateL2CacheBlock; + } + + transition(SS, L2_Replacement_clean_XACT, I_I) { + q_profileOverflow; + i_allocateTBE; + f_sendInvToSharers; + rr_deallocateL2CacheBlock; + } + + transition(SS, L2_Replacement, S_I) { + i_allocateTBE; + f_sendInvToSharers; + rr_deallocateL2CacheBlock; + } + + transition(SS, L2_Replacement_XACT, S_I) { + q_profileOverflow; + i_allocateTBE; + f_sendInvToSharers; + rr_deallocateL2CacheBlock; + } + + // Transitions from M + + // send data, but force L1 to wait for filter responses + transition(M, L1_GETS, M_MB) { + p_profileRequest; + f_profileRequestor; + f_sendExclusiveDataToRequestor; + // selective filter checks, but need to check both read+write in case nackers put NP block into M state + a_checkNackerL1ReadWriteFiltersExceptRequestor; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + // don't care about filters + transition(M, L1_GETS_ESCAPE, M_MB) { + p_profileRequest; + f_profileRequestor; + f_sendExclusiveDataToRequestor; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + transition(M, L1_GET_INSTR, M_SSB) { + p_profileRequest; + f_profileRequestor; + f_sendDataToRequestor; + // NEW - selective filter checks, but need to check both read+write in case nackers put NP block into M state + a_checkNackerL1ReadWriteFiltersExceptRequestor; + // This should always be _after_ f_sendDataToRequestor and a_checkNackerL1WriteFiltersExceptRequestor, since they + // explicitly look at the sharers list! + nn_addSharer; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + // don't care about filters + transition(M, L1_GET_INSTR_ESCAPE, M_SSB) { + p_profileRequest; + f_profileRequestor; + f_sendDataToRequestor; + nn_addSharer; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + transition(M, L1_GETX, M_MB) { + p_profileRequest; + f_profileRequestor; + f_sendDataToRequestor; + // selective filter checks + a_checkNackerL1ReadWriteFiltersExceptRequestor; + // issue filter checks + //a_checkL1ReadWriteFiltersExceptRequestor; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + // don't care about filters + transition(M, L1_GETX_ESCAPE, M_MB) { + p_profileRequest; + f_profileRequestor; + f_sendDataToRequestor; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + transition(M, L2_Replacement, M_I) { + i_allocateTBE; + c_exclusiveReplacement; + rr_deallocateL2CacheBlock; + } + + transition(M, L2_Replacement_clean, M_I) { + rr_deallocateL2CacheBlock; + } + + transition(M, L2_Replacement_XACT, M_I) { + q_profileOverflow; + i_allocateTBE; + c_exclusiveReplacement; + rr_deallocateL2CacheBlock; + } + + transition(M, L2_Replacement_clean_XACT, M_I) { + q_profileOverflow; + rr_deallocateL2CacheBlock; + } + + + // transitions from MT + transition(MT, {L1_GETX, L1_GETX_ESCAPE}, MT_MB) { + p_profileRequest; + f_profileRequestor; + b_forwardRequestToExclusive; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + + transition(MT, {L1_GETS, L1_GET_INSTR, L1_GETS_ESCAPE, L1_GET_INSTR_ESCAPE}, MT_IIB) { + p_profileRequest; + f_profileRequestor; + b_forwardRequestToExclusive; + uu_profileMiss; + set_setMRU; + jj_popL1RequestQueue; + } + + transition(MT, L2_Replacement, MT_I) { + i_allocateTBE; + f_sendInvToSharers; + rr_deallocateL2CacheBlock; + } + + transition(MT, L2_Replacement_clean, MCT_I) { + i_allocateTBE; + f_sendInvToSharers; + rr_deallocateL2CacheBlock; + } + + transition(MT, L2_Replacement_XACT, MT_I) { + q_profileOverflow; + i_allocateTBE; + f_sendInvToSharers; + rr_deallocateL2CacheBlock; + } + + transition(MT, L2_Replacement_clean_XACT, MCT_I) { + q_profileOverflow; + i_allocateTBE; + f_sendInvToSharers; + rr_deallocateL2CacheBlock; + } + + transition(MT, L1_PUTX, M) { + f_profileRequestor; + // this doesn't affect exlusive ptr + ll_clearSharers; + mr_writeDataToCacheFromRequest; + t_sendWBAck; + jj_popL1RequestQueue; + } + + // This is for the case of transactional read line in E state being replaced from L1. We need to maintain isolation on this + // in the event of a future transactional store from another proc, so we maintain this transactional sharer on the list + transition(MT, L1_PUTS, SS) { + f_profileRequestor; + ll_clearSharers; + // maintain transactional read isolation + nn_addSharer; + mr_writeDataToCacheFromRequest; + t_sendWBAck; + jj_popL1RequestQueue; + } + + // transitions from blocking states + transition(SS_MB, Unblock_Cancel, SS) { + k_popUnblockQueue; + } + + transition(M_SSB, Unblock_Cancel, M) { + ll_clearSharers; + // copy NACKers list from unblock message to our sharers list + m_copyNackersIntoSharers; + k_popUnblockQueue; + } + + transition(MT_MB, Unblock_Cancel, MT) { + k_popUnblockQueue; + } + + transition(MT_IB, Unblock_Cancel, MT) { + k_popUnblockQueue; + } + + transition(MT_IIB, Unblock_Cancel, MT){ + k_popUnblockQueue; + } + + // L2 just got the data from memory, but we have Nackers. We can let nacked block reside in M, but GETS request needs to check read+write + // signatures to avoid atomicity violations. + transition({ISS_MB, IS_SSB}, Unblock_Cancel, M){ + //rr_deallocateL2CacheBlock; + // copy NACKers list from unblock message to our sharers list + m_copyNackersIntoSharers; + k_popUnblockQueue; + } + + transition(M_MB, Unblock_Cancel, M) { + // copy NACKers list from unblock message to our sharers list + m_copyNackersIntoSharers; + k_popUnblockQueue; + } + + transition(SS_MB, Exclusive_Unblock, MT) { + // update actual directory + mmu_markExclusiveFromUnblock; + // mark block as trans if needed + f_markBlockTransIfTrans; + k_popUnblockQueue; + } + + // PUT from next exclusive surpassed its own ExclusiveUnblock + // Perceived as PUTX_old because the directory is outdated + transition(SS_MB, {L1_PUTX_old, L1_PUTS_old}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + // PUT from current (old) exclusive, can't do anything with it in this state + // Don't know whether exclusive was replacing or not, so wait to see what Unblock says + transition(SS_MB, {L1_PUTX, L1_PUTS}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + // Next exclusive informs that last owner was replacing the line when it received Fwd req + // Thus, expect a PUTX_old from previous owner + transition(SS_MB, Exclusive_Unblock_WaitPUTold, PB_MT) { + // update actual directory + mmu_markExclusiveFromUnblock; + // mark block as trans if needed + f_markBlockTransIfTrans; + k_popUnblockQueue; + } + + transition(PB_MT, {L1_PUTX_old, L1_PUTS_old}, MT) { // OK, PUT_old received, go to MT + f_profileRequestor; + jj_popL1RequestQueue; + } + + // PUT from current (next) exclusive, so recycle + // Expecting PUT_old, won't take in new PUT until previous PUT arrives + transition(PB_MT, {L1_PUTX, L1_PUTS}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + // L2 blocks on GETS requests in SS state + transition(SS_SSB, Unblock, SS) { + // mark block as trans if needed + f_markBlockTransIfTrans; + k_popUnblockQueue; + } + + transition({M_SSB, IS_SSB}, Unblock, SS) { + // we already added the sharer when we received original request + // mark block as trans if needed + f_markBlockTransIfTrans; + k_popUnblockQueue; + } + + transition({M_MB, MT_MB, ISS_MB}, Exclusive_Unblock, MT) { + // update actual directory + mmu_markExclusiveFromUnblock; + // mark block as trans if needed + f_markBlockTransIfTrans; + k_popUnblockQueue; + } + + transition({M_MB, MT_MB, ISS_MB}, Exclusive_Unblock_WaitPUTold, PB_MT) { + // update actual directory + mmu_markExclusiveFromUnblock; + // mark block as trans if needed + f_markBlockTransIfTrans; + k_popUnblockQueue; + } + + // PUT from (not yet) next exclusive surpassed its own ExclusiveUnblock + // thus became PUTX_old (since directory is not up-to-date) + transition({M_MB, MT_MB, ISS_MB}, {L1_PUTX_old, L1_PUTS_old}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + // PUT from current (previous) owner: recycle until unblock arrives + // We don't know whether replacing cache is waiting for WB_Ack or it was replacing when fwd arrived + transition({M_MB, MT_MB, ISS_MB}, {L1_PUTX, L1_PUTS}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + // L1 requestor received data from exclusive L1, but writeback data from exclusive L1 hasn't arrived yet + transition(MT_IIB, Unblock, MT_IB) { + nnu_addSharerFromUnblock; + // mark block as trans if needed + f_markBlockTransIfTrans; + k_popUnblockQueue; + } + + // PUT from current (previous) owner: recycle + // We don't know whether replacing cache is waiting for WB_Ack or it was replacing when fwd arrived + transition(MT_IIB, {L1_PUTX, L1_PUTS}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + transition(MT_IB, {WB_Data, WB_Data_clean}, SS) { + m_writeDataToCache; + o_popIncomingResponseQueue; + } + + // PUT from (not yet) next exclusive, but unblock hasn't arrived yet, so it became PUT_old: recycle + transition(MT_IIB, {L1_PUTX_old, L1_PUTS_old}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + transition(MT_IIB, Unblock_WaitPUTold, PB_MT_IB) { // Now arrives Unblock, wait for PUT and WB_Data + nnu_addSharerFromUnblock; + // mark block as trans if needed + f_markBlockTransIfTrans; + k_popUnblockQueue; + } + + // L1 requestor has not received data from exclusive L1, but we received writeback data from exclusive L1 + transition(MT_IIB, {WB_Data, WB_Data_clean}, MT_SB) { + m_writeDataToCache; + o_popIncomingResponseQueue; + } + + // PUT_old from previous owner, that was replacing when it received Fwd req + transition(PB_MT_IB, {L1_PUTX_old, L1_PUTS_old}, MT_IB) { // Go to MT_IB, and wait for WB_Data + f_profileRequestor; + jj_popL1RequestQueue; + } + + transition(PB_MT_IB, {L1_PUTX, L1_PUTS}) { // Waiting for PUT_old, don't take new PUT in + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + // WB_data from previous owner, we already received unblock, just wait for PUT_old to go to SS + transition(PB_MT_IB, {WB_Data, WB_Data_clean}, PB_SS) { // Received Unblock, now arrives WB_Data, wait for PUT + m_writeDataToCache; + o_popIncomingResponseQueue; + } + + transition(PB_SS, {L1_PUTX_old, L1_PUTS_old}, SS) { // Received Unblock and WB_Data, now arrives PUT, go to SS + f_profileRequestor; + jj_popL1RequestQueue; + } + + // PUT from new exclusive owner, while waiting for PUT from previous exclusive owner: recycle + transition(PB_SS, {L1_PUTX, L1_PUTS}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + transition(MT_SB, Unblock, SS) { + nnu_addSharerFromUnblock; + // mark block as trans if needed + f_markBlockTransIfTrans; + k_popUnblockQueue; + } + + transition(MT_SB, Unblock_WaitPUTold, PB_SS) { // Received WB_Data, now arriving Unblock, wait for PUT + nnu_addSharerFromUnblock; + // mark block as trans if needed + f_markBlockTransIfTrans; + k_popUnblockQueue; + } + + // PUT from (not yet) new exclusive owner, before we receive Unblock from it (became PUT_old because directory is not up-to-date) + transition(MT_SB, {L1_PUTX_old, L1_PUTS_old}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + // PUT from current (last) exclusive owner, that was replacing the line when it received Fwd req + transition(MT_SB, {L1_PUTX, L1_PUTS}) { + kk_removeRequestSharer; // When Unblock arrives, it'll trigger Unblock, not Unblock_WaitPUTold + f_profileRequestor; + jj_popL1RequestQueue; + } + + // writeback states + transition({I_I, S_I, MT_I, MCT_I, M_I}, {L1_GETX, L1_UPGRADE, L1_GETS, L1_GET_INSTR, L1_GETX_ESCAPE, L1_GETS_ESCAPE, L1_GET_INSTR_ESCAPE}) { + f_profileRequestor; + zz_recycleL1RequestQueue; + } + + transition(I_I, Ack) { + q_updateAck; + o_popIncomingResponseQueue; + } + + transition(I_I, Ack_all, NP) { + s_deallocateTBE; + o_popIncomingResponseQueue; + } + + transition({MT_I, MCT_I}, WB_Data, M_I) { + qq_writeDataToTBE; + ct_exclusiveReplacementFromTBE; + o_popIncomingResponseQueue; + } + + transition(MCT_I, WB_Data_clean, NP) { + s_deallocateTBE; + o_popIncomingResponseQueue; + } + + // L1 never changed Dirty data + transition(MT_I, Ack_all, M_I) { + ct_exclusiveReplacementFromTBE; + o_popIncomingResponseQueue; + } + + // clean data that L1 exclusive never wrote + transition(MCT_I, Ack_all, NP) { + s_deallocateTBE; + o_popIncomingResponseQueue; + } + + transition(MT_I, WB_Data_clean, NP) { + s_deallocateTBE; + o_popIncomingResponseQueue; + } + + transition(S_I, Ack) { + q_updateAck; + o_popIncomingResponseQueue; + } + + transition(S_I, Ack_all, M_I) { + ct_exclusiveReplacementFromTBE; + o_popIncomingResponseQueue; + } + + transition(M_I, Mem_Ack, NP) { + s_deallocateTBE; + o_popIncomingResponseQueue; + } +} + |