gem5/src/mem/protocol/MOESI_CMP_token-dir.sm
Brad Beckmann fb2e0f56ef ruby: MOESI_CMP_token dma fixes
This patch fixes various protocol bugs regarding races between dma requests
and persistent requests.
2010-08-20 11:46:13 -07:00

1262 lines
43 KiB
Text

/*
* 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.
*/
/*
* $Id$
*/
machine(Directory, "Token protocol")
: DirectoryMemory * directory,
MemoryControl * memBuffer,
int l2_select_num_bits,
int directory_latency = 5,
bool distributed_persistent = true,
int fixed_timeout_latency = 100
{
MessageBuffer dmaResponseFromDir, network="To", virtual_network="5", ordered="true";
MessageBuffer responseFromDir, network="To", virtual_network="4", ordered="false";
MessageBuffer persistentFromDir, network="To", virtual_network="3", ordered="true";
MessageBuffer requestFromDir, network="To", virtual_network="1", ordered="false";
MessageBuffer responseToDir, network="From", virtual_network="4", ordered="false";
MessageBuffer persistentToDir, network="From", virtual_network="3", ordered="true";
MessageBuffer requestToDir, network="From", virtual_network="2", ordered="false";
MessageBuffer dmaRequestToDir, network="From", virtual_network="0", ordered="true";
// STATES
enumeration(State, desc="Directory states", default="Directory_State_O") {
// Base states
O, desc="Owner";
NO, desc="Not Owner";
L, desc="Locked";
// Memory wait states - can block all messages including persistent requests
O_W, desc="transitioning to Owner, waiting for memory write";
L_O_W, desc="transitioning to Locked, waiting for memory read, could eventually return to O";
L_NO_W, desc="transitioning to Locked, waiting for memory read, eventually return to NO";
DR_L_W, desc="transitioning to Locked underneath a DMA read, waiting for memory data";
DW_L_W, desc="transitioning to Locked underneath a DMA write, waiting for memory ack";
NO_W, desc="transitioning to Not Owner, waiting for memory read";
O_DW_W, desc="transitioning to Owner, waiting for memory before DMA ack";
O_DR_W, desc="transitioning to Owner, waiting for memory before DMA data";
// DMA request transient states - must respond to persistent requests
O_DW, desc="issued GETX for DMA write, waiting for all tokens";
NO_DW, desc="issued GETX for DMA write, waiting for all tokens";
NO_DR, desc="issued GETS for DMA read, waiting for data";
// DMA request in progress - competing with a CPU persistent request
DW_L, desc="issued GETX for DMA write, CPU persistent request must complete first";
DR_L, desc="issued GETS for DMA read, CPU persistent request must complete first";
}
// Events
enumeration(Event, desc="Directory events") {
GETX, desc="A GETX arrives";
GETS, desc="A GETS arrives";
Lockdown, desc="A lockdown request arrives";
Unlockdown, desc="An un-lockdown request arrives";
Own_Lock_or_Unlock, desc="own lock or unlock";
Own_Lock_or_Unlock_Tokens, desc="own lock or unlock with tokens";
Data_Owner, desc="Data arrive";
Data_All_Tokens, desc="Data and all tokens";
Ack_Owner, desc="Owner token arrived without data because it was clean";
Ack_Owner_All_Tokens, desc="All tokens including owner arrived without data because it was clean";
Tokens, desc="Tokens arrive";
Ack_All_Tokens, desc="All_Tokens arrive";
Request_Timeout, desc="A DMA request has timed out";
// Memory Controller
Memory_Data, desc="Fetched data from memory arrives";
Memory_Ack, desc="Writeback Ack from memory arrives";
// DMA requests
DMA_READ, desc="A DMA Read memory request";
DMA_WRITE, desc="A DMA Write memory request";
DMA_WRITE_All_Tokens, desc="A DMA Write memory request, directory has all tokens";
}
// TYPES
// DirectoryEntry
structure(Entry, desc="...", interface="AbstractEntry") {
State DirectoryState, desc="Directory state";
DataBlock DataBlk, desc="data for the block";
int Tokens, default="max_tokens()", desc="Number of tokens for the line we're holding";
// The following state is provided to allow for bandwidth
// efficient directory-like operation. However all of this state
// is 'soft state' that does not need to be correct (as long as
// you're eventually willing to resort to broadcast.)
Set Owner, desc="Probable Owner of the line. More accurately, the set of processors who need to see a GetS or GetO. We use a Set for convenience, but only one bit is set at a time.";
Set Sharers, desc="Probable sharers of the line. More accurately, the set of processors who need to see a GetX";
}
external_type(PersistentTable) {
void persistentRequestLock(Address, MachineID, AccessType);
void persistentRequestUnlock(Address, MachineID);
bool okToIssueStarving(Address, MachineID);
MachineID findSmallest(Address);
AccessType typeOfSmallest(Address);
void markEntries(Address);
bool isLocked(Address);
int countStarvingForAddress(Address);
int countReadStarvingForAddress(Address);
}
// TBE entries for DMA requests
structure(TBE, desc="TBE entries for outstanding DMA requests") {
Address PhysicalAddress, desc="physical address";
State TBEState, desc="Transient State";
DataBlock DmaDataBlk, desc="DMA Data to be written. Partial blocks need to merged with system memory";
DataBlock DataBlk, desc="The current view of system memory";
int Len, desc="...";
MachineID DmaRequestor, desc="DMA requestor";
bool WentPersistent, desc="Did the DMA request require a persistent request";
}
external_type(TBETable) {
TBE lookup(Address);
void allocate(Address);
void deallocate(Address);
bool isPresent(Address);
}
// ** OBJECTS **
PersistentTable persistentTable;
TimerTable reissueTimerTable;
TBETable TBEs, template_hack="<Directory_TBE>";
bool starving, default="false";
int l2_select_low_bit, default="RubySystem::getBlockSizeBits()";
Entry getDirectoryEntry(Address addr), return_by_ref="yes" {
return static_cast(Entry, directory[addr]);
}
State getState(Address addr) {
if (TBEs.isPresent(addr)) {
return TBEs[addr].TBEState;
} else {
return getDirectoryEntry(addr).DirectoryState;
}
}
void setState(Address addr, State state) {
if (TBEs.isPresent(addr)) {
TBEs[addr].TBEState := state;
}
getDirectoryEntry(addr).DirectoryState := state;
if (state == State:L || state == State:DW_L || state == State:DR_L) {
assert(getDirectoryEntry(addr).Tokens == 0);
}
// We have one or zero owners
assert((getDirectoryEntry(addr).Owner.count() == 0) || (getDirectoryEntry(addr).Owner.count() == 1));
// Make sure the token count is in range
assert(getDirectoryEntry(addr).Tokens >= 0);
assert(getDirectoryEntry(addr).Tokens <= max_tokens());
if (state == State:O || state == State:O_W || state == State:O_DW) {
assert(getDirectoryEntry(addr).Tokens >= 1); // Must have at least one token
// assert(getDirectoryEntry(addr).Tokens >= (max_tokens() / 2)); // Only mostly true; this might not always hold
}
}
bool okToIssueStarving(Address addr, MachineID machinID) {
return persistentTable.okToIssueStarving(addr, machineID);
}
void markPersistentEntries(Address addr) {
persistentTable.markEntries(addr);
}
// ** OUT_PORTS **
out_port(responseNetwork_out, ResponseMsg, responseFromDir);
out_port(persistentNetwork_out, PersistentMsg, persistentFromDir);
out_port(requestNetwork_out, RequestMsg, requestFromDir);
out_port(dmaResponseNetwork_out, DMAResponseMsg, dmaResponseFromDir);
//
// Memory buffer for memory controller to DIMM communication
//
out_port(memQueue_out, MemoryMsg, memBuffer);
// ** IN_PORTS **
// off-chip memory request/response is done
in_port(memQueue_in, MemoryMsg, memBuffer) {
if (memQueue_in.isReady()) {
peek(memQueue_in, MemoryMsg) {
if (in_msg.Type == MemoryRequestType:MEMORY_READ) {
trigger(Event:Memory_Data, in_msg.Address);
} else if (in_msg.Type == MemoryRequestType:MEMORY_WB) {
trigger(Event:Memory_Ack, in_msg.Address);
} else {
DEBUG_EXPR(in_msg.Type);
error("Invalid message");
}
}
}
}
// Reissue Timer
in_port(reissueTimerTable_in, Address, reissueTimerTable) {
if (reissueTimerTable_in.isReady()) {
trigger(Event:Request_Timeout, reissueTimerTable.readyAddress());
}
}
in_port(responseNetwork_in, ResponseMsg, responseToDir) {
if (responseNetwork_in.isReady()) {
peek(responseNetwork_in, ResponseMsg) {
assert(in_msg.Destination.isElement(machineID));
if (getDirectoryEntry(in_msg.Address).Tokens + in_msg.Tokens == max_tokens()) {
if ((in_msg.Type == CoherenceResponseType:DATA_OWNER) ||
(in_msg.Type == CoherenceResponseType:DATA_SHARED)) {
trigger(Event:Data_All_Tokens, in_msg.Address);
} else if (in_msg.Type == CoherenceResponseType:ACK_OWNER) {
trigger(Event:Ack_Owner_All_Tokens, in_msg.Address);
} else if (in_msg.Type == CoherenceResponseType:ACK) {
trigger(Event:Ack_All_Tokens, in_msg.Address);
} else {
DEBUG_EXPR(in_msg.Type);
error("Invalid message");
}
} else {
if (in_msg.Type == CoherenceResponseType:DATA_OWNER) {
trigger(Event:Data_Owner, in_msg.Address);
} else if ((in_msg.Type == CoherenceResponseType:ACK) ||
(in_msg.Type == CoherenceResponseType:DATA_SHARED)) {
trigger(Event:Tokens, in_msg.Address);
} else if (in_msg.Type == CoherenceResponseType:ACK_OWNER) {
trigger(Event:Ack_Owner, in_msg.Address);
} else {
DEBUG_EXPR(in_msg.Type);
error("Invalid message");
}
}
}
}
}
in_port(persistentNetwork_in, PersistentMsg, persistentToDir) {
if (persistentNetwork_in.isReady()) {
peek(persistentNetwork_in, PersistentMsg) {
assert(in_msg.Destination.isElement(machineID));
if (distributed_persistent) {
// Apply the lockdown or unlockdown message to the table
if (in_msg.Type == PersistentRequestType:GETX_PERSISTENT) {
persistentTable.persistentRequestLock(in_msg.Address, in_msg.Requestor, AccessType:Write);
} else if (in_msg.Type == PersistentRequestType:GETS_PERSISTENT) {
persistentTable.persistentRequestLock(in_msg.Address, in_msg.Requestor, AccessType:Read);
} else if (in_msg.Type == PersistentRequestType:DEACTIVATE_PERSISTENT) {
persistentTable.persistentRequestUnlock(in_msg.Address, in_msg.Requestor);
} else {
error("Invalid message");
}
// React to the message based on the current state of the table
if (persistentTable.isLocked(in_msg.Address)) {
if (persistentTable.findSmallest(in_msg.Address) == machineID) {
if (getDirectoryEntry(in_msg.Address).Tokens > 0) {
trigger(Event:Own_Lock_or_Unlock_Tokens, in_msg.Address);
} else {
trigger(Event:Own_Lock_or_Unlock, in_msg.Address);
}
} else {
trigger(Event:Lockdown, in_msg.Address); // locked
}
} else {
trigger(Event:Unlockdown, in_msg.Address); // unlocked
}
}
else {
if (persistentTable.findSmallest(in_msg.Address) == machineID) {
if (getDirectoryEntry(in_msg.Address).Tokens > 0) {
trigger(Event:Own_Lock_or_Unlock_Tokens, in_msg.Address);
} else {
trigger(Event:Own_Lock_or_Unlock, in_msg.Address);
}
} else if (in_msg.Type == PersistentRequestType:GETX_PERSISTENT) {
trigger(Event:Lockdown, in_msg.Address); // locked
} else if (in_msg.Type == PersistentRequestType:GETS_PERSISTENT) {
trigger(Event:Lockdown, in_msg.Address); // locked
} else if (in_msg.Type == PersistentRequestType:DEACTIVATE_PERSISTENT) {
trigger(Event:Unlockdown, in_msg.Address); // unlocked
} else {
error("Invalid message");
}
}
}
}
}
in_port(requestNetwork_in, RequestMsg, requestToDir) {
if (requestNetwork_in.isReady()) {
peek(requestNetwork_in, RequestMsg) {
assert(in_msg.Destination.isElement(machineID));
if (in_msg.Type == CoherenceRequestType:GETS) {
trigger(Event:GETS, in_msg.Address);
} else if (in_msg.Type == CoherenceRequestType:GETX) {
trigger(Event:GETX, in_msg.Address);
} else {
error("Invalid message");
}
}
}
}
in_port(dmaRequestQueue_in, DMARequestMsg, dmaRequestToDir) {
if (dmaRequestQueue_in.isReady()) {
peek(dmaRequestQueue_in, DMARequestMsg) {
if (in_msg.Type == DMARequestType:READ) {
trigger(Event:DMA_READ, in_msg.LineAddress);
} else if (in_msg.Type == DMARequestType:WRITE) {
if (getDirectoryEntry(in_msg.LineAddress).Tokens == max_tokens()) {
trigger(Event:DMA_WRITE_All_Tokens, in_msg.LineAddress);
} else {
trigger(Event:DMA_WRITE, in_msg.LineAddress);
}
} else {
error("Invalid message");
}
}
}
}
// Actions
action(a_sendTokens, "a", desc="Send tokens to requestor") {
// Only send a message if we have tokens to send
if (getDirectoryEntry(address).Tokens > 0) {
peek(requestNetwork_in, RequestMsg) {
// enqueue(responseNetwork_out, ResponseMsg, latency="DIRECTORY_CACHE_LATENCY") {// FIXME?
enqueue(responseNetwork_out, ResponseMsg, latency=directory_latency) {// FIXME?
out_msg.Address := address;
out_msg.Type := CoherenceResponseType:ACK;
out_msg.Sender := machineID;
out_msg.Destination.add(in_msg.Requestor);
out_msg.Tokens := getDirectoryEntry(in_msg.Address).Tokens;
out_msg.MessageSize := MessageSizeType:Response_Control;
}
}
getDirectoryEntry(address).Tokens := 0;
}
}
action(px_tryIssuingPersistentGETXRequest, "px", desc="...") {
if (okToIssueStarving(address, machineID) && (starving == false)) {
enqueue(persistentNetwork_out, PersistentMsg, latency = "1") {
out_msg.Address := address;
out_msg.Type := PersistentRequestType:GETX_PERSISTENT;
out_msg.Requestor := machineID;
out_msg.Destination.broadcast(MachineType:L1Cache);
//
// Currently the configuration system limits the system to only one
// chip. Therefore, if we assume one shared L2 cache, then only one
// pertinent L2 cache exist.
//
//out_msg.Destination.addNetDest(getAllPertinentL2Banks(address));
out_msg.Destination.add(mapAddressToRange(address,
MachineType:L2Cache,
l2_select_low_bit,
l2_select_num_bits));
out_msg.Destination.add(map_Address_to_Directory(address));
out_msg.MessageSize := MessageSizeType:Persistent_Control;
out_msg.Prefetch := PrefetchBit:No;
out_msg.AccessMode := AccessModeType:SupervisorMode;
}
markPersistentEntries(address);
starving := true;
TBEs[address].WentPersistent := true;
// Do not schedule a wakeup, a persistent requests will always complete
} else {
// We'd like to issue a persistent request, but are not allowed
// to issue a P.R. right now. This, we do not increment the
// IssueCount.
// Set a wakeup timer
reissueTimerTable.set(address, 10);
}
}
action(bw_broadcastWrite, "bw", desc="Broadcast GETX if we need tokens") {
peek(dmaRequestQueue_in, DMARequestMsg) {
//
// Assser that we only send message if we don't already have all the tokens
//
assert(getDirectoryEntry(address).Tokens != max_tokens());
enqueue(requestNetwork_out, RequestMsg, latency = "1") {
out_msg.Address := address;
out_msg.Type := CoherenceRequestType:GETX;
out_msg.Requestor := machineID;
//
// Since only one chip, assuming all L1 caches are local
//
out_msg.Destination.broadcast(MachineType:L1Cache);
out_msg.Destination.add(mapAddressToRange(address,
MachineType:L2Cache,
l2_select_low_bit,
l2_select_num_bits));
out_msg.RetryNum := 0;
out_msg.MessageSize := MessageSizeType:Broadcast_Control;
out_msg.Prefetch := PrefetchBit:No;
out_msg.AccessMode := AccessModeType:SupervisorMode;
}
}
}
action(ps_tryIssuingPersistentGETSRequest, "ps", desc="...") {
if (okToIssueStarving(address, machineID) && (starving == false)) {
enqueue(persistentNetwork_out, PersistentMsg, latency = "1") {
out_msg.Address := address;
out_msg.Type := PersistentRequestType:GETS_PERSISTENT;
out_msg.Requestor := machineID;
out_msg.Destination.broadcast(MachineType:L1Cache);
//
// Currently the configuration system limits the system to only one
// chip. Therefore, if we assume one shared L2 cache, then only one
// pertinent L2 cache exist.
//
//out_msg.Destination.addNetDest(getAllPertinentL2Banks(address));
out_msg.Destination.add(mapAddressToRange(address,
MachineType:L2Cache,
l2_select_low_bit,
l2_select_num_bits));
out_msg.Destination.add(map_Address_to_Directory(address));
out_msg.MessageSize := MessageSizeType:Persistent_Control;
out_msg.Prefetch := PrefetchBit:No;
out_msg.AccessMode := AccessModeType:SupervisorMode;
}
markPersistentEntries(address);
starving := true;
TBEs[address].WentPersistent := true;
// Do not schedule a wakeup, a persistent requests will always complete
} else {
// We'd like to issue a persistent request, but are not allowed
// to issue a P.R. right now. This, we do not increment the
// IssueCount.
// Set a wakeup timer
reissueTimerTable.set(address, 10);
}
}
action(br_broadcastRead, "br", desc="Broadcast GETS for data") {
peek(dmaRequestQueue_in, DMARequestMsg) {
enqueue(requestNetwork_out, RequestMsg, latency = "1") {
out_msg.Address := address;
out_msg.Type := CoherenceRequestType:GETS;
out_msg.Requestor := machineID;
//
// Since only one chip, assuming all L1 caches are local
//
out_msg.Destination.broadcast(MachineType:L1Cache);
out_msg.Destination.add(mapAddressToRange(address,
MachineType:L2Cache,
l2_select_low_bit,
l2_select_num_bits));
out_msg.RetryNum := 0;
out_msg.MessageSize := MessageSizeType:Broadcast_Control;
out_msg.Prefetch := PrefetchBit:No;
out_msg.AccessMode := AccessModeType:SupervisorMode;
}
}
}
action(aa_sendTokensToStarver, "\a", desc="Send tokens to starver") {
// Only send a message if we have tokens to send
if (getDirectoryEntry(address).Tokens > 0) {
// enqueue(responseNetwork_out, ResponseMsg, latency="DIRECTORY_CACHE_LATENCY") {// FIXME?
enqueue(responseNetwork_out, ResponseMsg, latency=directory_latency) {// FIXME?
out_msg.Address := address;
out_msg.Type := CoherenceResponseType:ACK;
out_msg.Sender := machineID;
out_msg.Destination.add(persistentTable.findSmallest(address));
out_msg.Tokens := getDirectoryEntry(address).Tokens;
out_msg.MessageSize := MessageSizeType:Response_Control;
}
getDirectoryEntry(address).Tokens := 0;
}
}
action(d_sendMemoryDataWithAllTokens, "d", desc="Send data and tokens to requestor") {
peek(memQueue_in, MemoryMsg) {
enqueue(responseNetwork_out, ResponseMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := CoherenceResponseType:DATA_OWNER;
out_msg.Sender := machineID;
out_msg.Destination.add(in_msg.OriginalRequestorMachId);
assert(getDirectoryEntry(address).Tokens > 0);
out_msg.Tokens := getDirectoryEntry(in_msg.Address).Tokens;
out_msg.DataBlk := getDirectoryEntry(in_msg.Address).DataBlk;
out_msg.Dirty := false;
out_msg.MessageSize := MessageSizeType:Response_Data;
}
}
getDirectoryEntry(address).Tokens := 0;
}
action(dd_sendMemDataToStarver, "\d", desc="Send data and tokens to starver") {
peek(memQueue_in, MemoryMsg) {
enqueue(responseNetwork_out, ResponseMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := CoherenceResponseType:DATA_OWNER;
out_msg.Sender := machineID;
out_msg.Destination.add(persistentTable.findSmallest(address));
assert(getDirectoryEntry(address).Tokens > 0);
out_msg.Tokens := getDirectoryEntry(address).Tokens;
out_msg.DataBlk := getDirectoryEntry(address).DataBlk;
out_msg.Dirty := false;
out_msg.MessageSize := MessageSizeType:Response_Data;
}
}
getDirectoryEntry(address).Tokens := 0;
}
action(de_sendTbeDataToStarver, "de", desc="Send data and tokens to starver") {
enqueue(responseNetwork_out, ResponseMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := CoherenceResponseType:DATA_OWNER;
out_msg.Sender := machineID;
out_msg.Destination.add(persistentTable.findSmallest(address));
assert(getDirectoryEntry(address).Tokens > 0);
out_msg.Tokens := getDirectoryEntry(address).Tokens;
out_msg.DataBlk := TBEs[address].DataBlk;
out_msg.Dirty := false;
out_msg.MessageSize := MessageSizeType:Response_Data;
}
getDirectoryEntry(address).Tokens := 0;
}
action(qf_queueMemoryFetchRequest, "qf", desc="Queue off-chip fetch request") {
peek(requestNetwork_in, RequestMsg) {
enqueue(memQueue_out, MemoryMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := MemoryRequestType:MEMORY_READ;
out_msg.Sender := machineID;
out_msg.OriginalRequestorMachId := in_msg.Requestor;
out_msg.MessageSize := in_msg.MessageSize;
out_msg.DataBlk := getDirectoryEntry(address).DataBlk;
DEBUG_EXPR(out_msg);
}
}
}
action(qp_queueMemoryForPersistent, "qp", desc="Queue off-chip fetch request") {
enqueue(memQueue_out, MemoryMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := MemoryRequestType:MEMORY_READ;
out_msg.Sender := machineID;
out_msg.OriginalRequestorMachId := persistentTable.findSmallest(address);
out_msg.MessageSize := MessageSizeType:Request_Control;
out_msg.DataBlk := getDirectoryEntry(address).DataBlk;
DEBUG_EXPR(out_msg);
}
}
action(fd_memoryDma, "fd", desc="Queue off-chip fetch request") {
peek(dmaRequestQueue_in, DMARequestMsg) {
enqueue(memQueue_out, MemoryMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := MemoryRequestType:MEMORY_READ;
out_msg.Sender := machineID;
out_msg.OriginalRequestorMachId := in_msg.Requestor;
out_msg.MessageSize := in_msg.MessageSize;
out_msg.DataBlk := getDirectoryEntry(address).DataBlk;
DEBUG_EXPR(out_msg);
}
}
}
action(lq_queueMemoryWbRequest, "lq", desc="Write data to memory") {
enqueue(memQueue_out, MemoryMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := MemoryRequestType:MEMORY_WB;
DEBUG_EXPR(out_msg);
}
}
action(ld_queueMemoryDmaWriteFromTbe, "ld", desc="Write DMA data to memory") {
enqueue(memQueue_out, MemoryMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := MemoryRequestType:MEMORY_WB;
// first, initialize the data blk to the current version of system memory
out_msg.DataBlk := TBEs[address].DataBlk;
// then add the dma write data
out_msg.DataBlk.copyPartial(TBEs[address].DmaDataBlk, addressOffset(TBEs[address].PhysicalAddress), TBEs[address].Len);
DEBUG_EXPR(out_msg);
}
}
action(lr_queueMemoryDmaReadWriteback, "lr", desc="Write DMA data from read to memory") {
enqueue(memQueue_out, MemoryMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := MemoryRequestType:MEMORY_WB;
// first, initialize the data blk to the current version of system memory
out_msg.DataBlk := TBEs[address].DataBlk;
DEBUG_EXPR(out_msg);
}
}
action(vd_allocateDmaRequestInTBE, "vd", desc="Record Data in TBE") {
peek(dmaRequestQueue_in, DMARequestMsg) {
TBEs.allocate(address);
TBEs[address].DmaDataBlk := in_msg.DataBlk;
TBEs[address].PhysicalAddress := in_msg.PhysicalAddress;
TBEs[address].Len := in_msg.Len;
TBEs[address].DmaRequestor := in_msg.Requestor;
TBEs[address].WentPersistent := false;
}
}
action(s_deallocateTBE, "s", desc="Deallocate TBE") {
if (TBEs[address].WentPersistent) {
assert(starving == true);
enqueue(persistentNetwork_out, PersistentMsg, latency = "1") {
out_msg.Address := address;
out_msg.Type := PersistentRequestType:DEACTIVATE_PERSISTENT;
out_msg.Requestor := machineID;
out_msg.Destination.broadcast(MachineType:L1Cache);
//
// Currently the configuration system limits the system to only one
// chip. Therefore, if we assume one shared L2 cache, then only one
// pertinent L2 cache exist.
//
//out_msg.Destination.addNetDest(getAllPertinentL2Banks(address));
out_msg.Destination.add(mapAddressToRange(address,
MachineType:L2Cache,
l2_select_low_bit,
l2_select_num_bits));
out_msg.Destination.add(map_Address_to_Directory(address));
out_msg.MessageSize := MessageSizeType:Persistent_Control;
}
starving := false;
}
TBEs.deallocate(address);
}
action(rd_recordDataInTbe, "rd", desc="Record data in TBE") {
peek(responseNetwork_in, ResponseMsg) {
TBEs[address].DataBlk := in_msg.DataBlk;
}
}
action(cd_writeCleanDataToTbe, "cd", desc="Write clean memory data to TBE") {
TBEs[address].DataBlk := getDirectoryEntry(address).DataBlk;
}
action(dwt_writeDmaDataFromTBE, "dwt", desc="DMA Write data to memory from TBE") {
getDirectoryEntry(address).DataBlk := TBEs[address].DataBlk;
getDirectoryEntry(address).DataBlk.copyPartial(TBEs[address].DmaDataBlk, addressOffset(TBEs[address].PhysicalAddress), TBEs[address].Len);
}
action(f_incrementTokens, "f", desc="Increment the number of tokens we're tracking") {
peek(responseNetwork_in, ResponseMsg) {
assert(in_msg.Tokens >= 1);
getDirectoryEntry(address).Tokens := getDirectoryEntry(address).Tokens + in_msg.Tokens;
}
}
action(aat_assertAllTokens, "aat", desc="assert that we have all tokens") {
assert(getDirectoryEntry(address).Tokens == max_tokens());
}
action(j_popIncomingRequestQueue, "j", desc="Pop incoming request queue") {
requestNetwork_in.dequeue();
}
action(z_recycleRequest, "z", desc="Recycle the request queue") {
requestNetwork_in.recycle();
}
action(k_popIncomingResponseQueue, "k", desc="Pop incoming response queue") {
responseNetwork_in.dequeue();
}
action(kz_recycleResponse, "kz", desc="Recycle incoming response queue") {
responseNetwork_in.recycle();
}
action(l_popIncomingPersistentQueue, "l", desc="Pop incoming persistent queue") {
persistentNetwork_in.dequeue();
}
action(p_popDmaRequestQueue, "pd", desc="pop dma request queue") {
dmaRequestQueue_in.dequeue();
}
action(y_recycleDmaRequestQueue, "y", desc="recycle dma request queue") {
dmaRequestQueue_in.recycle();
}
action(l_popMemQueue, "q", desc="Pop off-chip request queue") {
memQueue_in.dequeue();
}
action(m_writeDataToMemory, "m", desc="Write dirty writeback to memory") {
peek(responseNetwork_in, ResponseMsg) {
getDirectoryEntry(in_msg.Address).DataBlk := in_msg.DataBlk;
DEBUG_EXPR(in_msg.Address);
DEBUG_EXPR(in_msg.DataBlk);
}
}
action(n_checkData, "n", desc="Check incoming clean data message") {
peek(responseNetwork_in, ResponseMsg) {
assert(getDirectoryEntry(in_msg.Address).DataBlk == in_msg.DataBlk);
}
}
action(r_bounceResponse, "r", desc="Bounce response to starving processor") {
peek(responseNetwork_in, ResponseMsg) {
enqueue(responseNetwork_out, ResponseMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := in_msg.Type;
out_msg.Sender := machineID;
out_msg.Destination.add(persistentTable.findSmallest(address));
out_msg.Tokens := in_msg.Tokens;
out_msg.MessageSize := in_msg.MessageSize;
out_msg.DataBlk := in_msg.DataBlk;
out_msg.Dirty := in_msg.Dirty;
}
}
}
action(rs_resetScheduleTimeout, "rs", desc="Reschedule Schedule Timeout") {
//
// currently only support a fixed timeout latency
//
if (reissueTimerTable.isSet(address)) {
reissueTimerTable.unset(address);
reissueTimerTable.set(address, fixed_timeout_latency);
}
}
action(st_scheduleTimeout, "st", desc="Schedule Timeout") {
//
// currently only support a fixed timeout latency
//
reissueTimerTable.set(address, fixed_timeout_latency);
}
action(ut_unsetReissueTimer, "ut", desc="Unset reissue timer.") {
if (reissueTimerTable.isSet(address)) {
reissueTimerTable.unset(address);
}
}
action(bd_bounceDatalessOwnerToken, "bd", desc="Bounce clean owner token to starving processor") {
peek(responseNetwork_in, ResponseMsg) {
assert(in_msg.Type == CoherenceResponseType:ACK_OWNER);
assert(in_msg.Dirty == false);
assert(in_msg.MessageSize == MessageSizeType:Writeback_Control);
// NOTE: The following check would not be valid in a real
// implementation. We include the data in the "dataless"
// message so we can assert the clean data matches the datablock
// in memory
assert(getDirectoryEntry(in_msg.Address).DataBlk == in_msg.DataBlk);
// Bounce the message, but "re-associate" the data and the owner
// token. In essence we're converting an ACK_OWNER message to a
// DATA_OWNER message, keeping the number of tokens the same.
enqueue(responseNetwork_out, ResponseMsg, latency="1") {
out_msg.Address := address;
out_msg.Type := CoherenceResponseType:DATA_OWNER;
out_msg.Sender := machineID;
out_msg.Destination.add(persistentTable.findSmallest(address));
out_msg.Tokens := in_msg.Tokens;
out_msg.DataBlk := getDirectoryEntry(in_msg.Address).DataBlk;
out_msg.Dirty := in_msg.Dirty;
out_msg.MessageSize := MessageSizeType:Response_Data;
}
}
}
action(da_sendDmaAck, "da", desc="Send Ack to DMA controller") {
enqueue(dmaResponseNetwork_out, DMAResponseMsg, latency="1") {
out_msg.PhysicalAddress := address;
out_msg.LineAddress := address;
out_msg.Type := DMAResponseType:ACK;
out_msg.Destination.add(TBEs[address].DmaRequestor);
out_msg.MessageSize := MessageSizeType:Writeback_Control;
}
}
action(dm_sendMemoryDataToDma, "dm", desc="Send Data to DMA controller from memory") {
peek(memQueue_in, MemoryMsg) {
enqueue(dmaResponseNetwork_out, DMAResponseMsg, latency="1") {
out_msg.PhysicalAddress := address;
out_msg.LineAddress := address;
out_msg.Type := DMAResponseType:DATA;
//
// we send the entire data block and rely on the dma controller to
// split it up if need be
//
out_msg.DataBlk := in_msg.DataBlk;
out_msg.Destination.add(TBEs[address].DmaRequestor);
out_msg.MessageSize := MessageSizeType:Response_Data;
}
}
}
action(dd_sendDmaData, "dd", desc="Send Data to DMA controller") {
peek(responseNetwork_in, ResponseMsg) {
enqueue(dmaResponseNetwork_out, DMAResponseMsg, latency="1") {
out_msg.PhysicalAddress := address;
out_msg.LineAddress := address;
out_msg.Type := DMAResponseType:DATA;
//
// we send the entire data block and rely on the dma controller to
// split it up if need be
//
out_msg.DataBlk := in_msg.DataBlk;
out_msg.Destination.add(TBEs[address].DmaRequestor);
out_msg.MessageSize := MessageSizeType:Response_Data;
}
}
}
// TRANSITIONS
//
// Trans. from base state O
// the directory has valid data
//
transition(O, GETX, NO_W) {
qf_queueMemoryFetchRequest;
j_popIncomingRequestQueue;
}
transition(O, DMA_WRITE, O_DW) {
vd_allocateDmaRequestInTBE;
cd_writeCleanDataToTbe;
bw_broadcastWrite;
st_scheduleTimeout;
p_popDmaRequestQueue;
}
transition(O, DMA_WRITE_All_Tokens, O_DW_W) {
vd_allocateDmaRequestInTBE;
cd_writeCleanDataToTbe;
dwt_writeDmaDataFromTBE;
ld_queueMemoryDmaWriteFromTbe;
p_popDmaRequestQueue;
}
transition(O, GETS, NO_W) {
qf_queueMemoryFetchRequest;
j_popIncomingRequestQueue;
}
transition(O, DMA_READ, O_DR_W) {
vd_allocateDmaRequestInTBE;
fd_memoryDma;
st_scheduleTimeout;
p_popDmaRequestQueue;
}
transition(O, Lockdown, L_O_W) {
qp_queueMemoryForPersistent;
l_popIncomingPersistentQueue;
}
transition(O, {Tokens, Ack_All_Tokens}) {
f_incrementTokens;
k_popIncomingResponseQueue;
}
transition(O, {Data_Owner, Data_All_Tokens}) {
n_checkData;
f_incrementTokens;
k_popIncomingResponseQueue;
}
transition({O, NO}, Unlockdown) {
l_popIncomingPersistentQueue;
}
//
// transitioning to Owner, waiting for memory before DMA ack
// All other events should recycle/stall
//
transition(O_DR_W, Memory_Data, O) {
dm_sendMemoryDataToDma;
ut_unsetReissueTimer;
s_deallocateTBE;
l_popMemQueue;
}
//
// issued GETX for DMA write, waiting for all tokens
//
transition(O_DW, Request_Timeout) {
ut_unsetReissueTimer;
px_tryIssuingPersistentGETXRequest;
}
transition(O_DW, Tokens) {
f_incrementTokens;
k_popIncomingResponseQueue;
}
transition(O_DW, Data_Owner) {
f_incrementTokens;
rd_recordDataInTbe;
k_popIncomingResponseQueue;
}
transition(O_DW, Ack_Owner) {
f_incrementTokens;
cd_writeCleanDataToTbe;
k_popIncomingResponseQueue;
}
transition(O_DW, Lockdown, DW_L) {
de_sendTbeDataToStarver;
l_popIncomingPersistentQueue;
}
transition({NO_DW, O_DW}, Data_All_Tokens, O_DW_W) {
f_incrementTokens;
rd_recordDataInTbe;
dwt_writeDmaDataFromTBE;
ld_queueMemoryDmaWriteFromTbe;
ut_unsetReissueTimer;
k_popIncomingResponseQueue;
}
transition(O_DW, Ack_All_Tokens, O_DW_W) {
f_incrementTokens;
dwt_writeDmaDataFromTBE;
ld_queueMemoryDmaWriteFromTbe;
ut_unsetReissueTimer;
k_popIncomingResponseQueue;
}
transition(O_DW, Ack_Owner_All_Tokens, O_DW_W) {
f_incrementTokens;
cd_writeCleanDataToTbe;
dwt_writeDmaDataFromTBE;
ld_queueMemoryDmaWriteFromTbe;
ut_unsetReissueTimer;
k_popIncomingResponseQueue;
}
transition(O_DW_W, Memory_Ack, O) {
da_sendDmaAck;
s_deallocateTBE;
l_popMemQueue;
}
//
// Trans. from NO
// The direcotry does not have valid data, but may have some tokens
//
transition(NO, GETX) {
a_sendTokens;
j_popIncomingRequestQueue;
}
transition(NO, DMA_WRITE, NO_DW) {
vd_allocateDmaRequestInTBE;
bw_broadcastWrite;
st_scheduleTimeout;
p_popDmaRequestQueue;
}
transition(NO, GETS) {
j_popIncomingRequestQueue;
}
transition(NO, DMA_READ, NO_DR) {
vd_allocateDmaRequestInTBE;
br_broadcastRead;
st_scheduleTimeout;
p_popDmaRequestQueue;
}
transition(NO, Lockdown, L) {
aa_sendTokensToStarver;
l_popIncomingPersistentQueue;
}
transition(NO, {Data_Owner, Data_All_Tokens}, O_W) {
m_writeDataToMemory;
f_incrementTokens;
lq_queueMemoryWbRequest;
k_popIncomingResponseQueue;
}
transition(NO, {Ack_Owner, Ack_Owner_All_Tokens}, O) {
n_checkData;
f_incrementTokens;
k_popIncomingResponseQueue;
}
transition(NO, Tokens) {
f_incrementTokens;
k_popIncomingResponseQueue;
}
transition(NO_W, Memory_Data, NO) {
d_sendMemoryDataWithAllTokens;
l_popMemQueue;
}
// Trans. from NO_DW
transition(NO_DW, Request_Timeout) {
ut_unsetReissueTimer;
px_tryIssuingPersistentGETXRequest;
}
transition(NO_DW, Lockdown, DW_L) {
aa_sendTokensToStarver;
l_popIncomingPersistentQueue;
}
// Note: NO_DW, Data_All_Tokens transition is combined with O_DW
// Note: NO_DW should not receive the action Ack_All_Tokens because the
// directory does not have valid data
transition(NO_DW, Data_Owner, O_DW) {
f_incrementTokens;
rd_recordDataInTbe;
k_popIncomingResponseQueue;
}
transition({NO_DW, NO_DR}, Tokens) {
f_incrementTokens;
k_popIncomingResponseQueue;
}
// Trans. from NO_DR
transition(NO_DR, Request_Timeout) {
ut_unsetReissueTimer;
ps_tryIssuingPersistentGETSRequest;
}
transition(NO_DR, Lockdown, DR_L) {
aa_sendTokensToStarver;
l_popIncomingPersistentQueue;
}
transition(NO_DR, {Data_Owner, Data_All_Tokens}, O_W) {
m_writeDataToMemory;
f_incrementTokens;
dd_sendDmaData;
lr_queueMemoryDmaReadWriteback;
ut_unsetReissueTimer;
s_deallocateTBE;
k_popIncomingResponseQueue;
}
// Trans. from L
transition({L, DW_L, DR_L}, {GETX, GETS}) {
j_popIncomingRequestQueue;
}
transition({L, DW_L, DR_L, L_O_W, L_NO_W, DR_L_W, DW_L_W}, Lockdown) {
l_popIncomingPersistentQueue;
}
//
// Received data for lockdown blocks
// For blocks with outstanding dma requests to them
// ...we could change this to write the data to memory and send it cleanly
// ...we could also proactively complete our DMA requests
// However, to keep my mind from spinning out-of-control, we won't for now :)
//
transition({DW_L, DR_L, L}, {Data_Owner, Data_All_Tokens}) {
r_bounceResponse;
k_popIncomingResponseQueue;
}
transition({DW_L, DR_L, L}, Tokens) {
r_bounceResponse;
k_popIncomingResponseQueue;
}
transition({DW_L, DR_L, L}, {Ack_Owner_All_Tokens, Ack_Owner}) {
bd_bounceDatalessOwnerToken;
k_popIncomingResponseQueue;
}
transition(L, {Unlockdown, Own_Lock_or_Unlock}, NO) {
l_popIncomingPersistentQueue;
}
transition(L, Own_Lock_or_Unlock_Tokens, O) {
l_popIncomingPersistentQueue;
}
transition({L_NO_W, L_O_W}, Memory_Data, L) {
dd_sendMemDataToStarver;
l_popMemQueue;
}
transition(L_O_W, Memory_Ack) {
qp_queueMemoryForPersistent;
l_popMemQueue;
}
transition(L_O_W, {Unlockdown, Own_Lock_or_Unlock, Own_Lock_or_Unlock_Tokens}, O_W) {
l_popIncomingPersistentQueue;
}
transition(L_NO_W, {Unlockdown, Own_Lock_or_Unlock, Own_Lock_or_Unlock_Tokens}, NO_W) {
l_popIncomingPersistentQueue;
}
transition(DR_L_W, Memory_Data, DR_L) {
dd_sendMemDataToStarver;
l_popMemQueue;
}
transition(DW_L_W, Memory_Ack, L) {
aat_assertAllTokens;
da_sendDmaAck;
s_deallocateTBE;
dd_sendMemDataToStarver;
l_popMemQueue;
}
transition(DW_L, {Unlockdown, Own_Lock_or_Unlock, Own_Lock_or_Unlock_Tokens}, NO_DW) {
l_popIncomingPersistentQueue;
}
transition(DR_L_W, {Unlockdown, Own_Lock_or_Unlock, Own_Lock_or_Unlock_Tokens}, O_DR_W) {
l_popIncomingPersistentQueue;
}
transition(DW_L_W, {Unlockdown, Own_Lock_or_Unlock, Own_Lock_or_Unlock_Tokens}, O_DW_W) {
l_popIncomingPersistentQueue;
}
transition({DW_L, DR_L_W, DW_L_W}, Request_Timeout) {
ut_unsetReissueTimer;
px_tryIssuingPersistentGETXRequest;
}
transition(DR_L, {Unlockdown, Own_Lock_or_Unlock, Own_Lock_or_Unlock_Tokens}, NO_DR) {
l_popIncomingPersistentQueue;
}
transition(DR_L, Request_Timeout) {
ut_unsetReissueTimer;
ps_tryIssuingPersistentGETSRequest;
}
//
// The O_W + Memory_Data > O transistion is confusing, but it can happen if a
// presistent request is issued and resolve before memory returns with data
//
transition(O_W, {Memory_Ack, Memory_Data}, O) {
l_popMemQueue;
}
transition({O, NO}, {Own_Lock_or_Unlock, Own_Lock_or_Unlock_Tokens}) {
l_popIncomingPersistentQueue;
}
// Blocked states
transition({NO_W, O_W, L_O_W, L_NO_W, DR_L_W, DW_L_W, O_DW_W, O_DR_W, O_DW, NO_DW, NO_DR}, {GETX, GETS}) {
z_recycleRequest;
}
transition({NO_W, O_W, L_O_W, L_NO_W, DR_L_W, DW_L_W, O_DW_W, O_DR_W, O_DW, NO_DW, NO_DR, L, DW_L, DR_L}, {DMA_READ, DMA_WRITE, DMA_WRITE_All_Tokens}) {
y_recycleDmaRequestQueue;
}
transition({NO_W, O_W, L_O_W, L_NO_W, DR_L_W, DW_L_W, O_DW_W, O_DR_W}, {Data_Owner, Ack_Owner, Tokens, Data_All_Tokens, Ack_All_Tokens}) {
kz_recycleResponse;
}
//
// If we receive a request timeout while waiting for memory, it is likely that
// the request will be satisfied and issuing a presistent request will do us
// no good. Just wait.
//
transition({O_DW_W, O_DR_W}, Request_Timeout) {
rs_resetScheduleTimeout;
}
transition(NO_W, Lockdown, L_NO_W) {
l_popIncomingPersistentQueue;
}
transition(O_W, Lockdown, L_O_W) {
l_popIncomingPersistentQueue;
}
transition(O_DR_W, Lockdown, DR_L_W) {
l_popIncomingPersistentQueue;
}
transition(O_DW_W, Lockdown, DW_L_W) {
l_popIncomingPersistentQueue;
}
transition({NO_W, O_W, O_DR_W, O_DW_W, O_DW, NO_DR, NO_DW}, {Unlockdown, Own_Lock_or_Unlock, Own_Lock_or_Unlock_Tokens}) {
l_popIncomingPersistentQueue;
}
}