diff options
Diffstat (limited to 'src/cpu')
27 files changed, 163 insertions, 329 deletions
diff --git a/src/cpu/base.cc b/src/cpu/base.cc index 6c8bf65fa..a54ed9349 100644 --- a/src/cpu/base.cc +++ b/src/cpu/base.cc @@ -90,7 +90,7 @@ CPUProgressEvent::process() const char * CPUProgressEvent::description() { - return "CPU Progress event"; + return "CPU Progress"; } #if FULL_SYSTEM @@ -343,9 +343,8 @@ BaseCPU::takeOverFrom(BaseCPU *oldCPU, Port *ic, Port *dc) for (int i = 0; i < threadContexts.size(); ++i) threadContexts[i]->profileClear(); - // The Sampler must take care of this! -// if (profileEvent) -// profileEvent->schedule(curTick); + if (profileEvent) + profileEvent->schedule(curTick); #endif // Connect new CPU to old CPU's memory only if new CPU isn't diff --git a/src/cpu/memtest/MemTest.py b/src/cpu/memtest/MemTest.py index 381519972..a328f4734 100644 --- a/src/cpu/memtest/MemTest.py +++ b/src/cpu/memtest/MemTest.py @@ -33,7 +33,7 @@ from m5 import build_env class MemTest(SimObject): type = 'MemTest' - max_loads = Param.Counter("number of loads to execute") + max_loads = Param.Counter(0, "number of loads to execute") atomic = Param.Bool(False, "Execute tester in atomic mode? (or timing)\n") memory_size = Param.Int(65536, "memory size") percent_dest_unaligned = Param.Percent(50, diff --git a/src/cpu/memtest/memtest.cc b/src/cpu/memtest/memtest.cc index 23e2297fe..83417c514 100644 --- a/src/cpu/memtest/memtest.cc +++ b/src/cpu/memtest/memtest.cc @@ -57,14 +57,22 @@ int TESTER_ALLOCATOR=0; bool MemTest::CpuPort::recvTiming(PacketPtr pkt) { - memtest->completeRequest(pkt); + if (pkt->isResponse()) { + memtest->completeRequest(pkt); + } else { + // must be snoop upcall + assert(pkt->isRequest()); + assert(pkt->getDest() == Packet::Broadcast); + } return true; } Tick MemTest::CpuPort::recvAtomic(PacketPtr pkt) { - panic("MemTest doesn't expect recvAtomic callback!"); + // must be snoop upcall + assert(pkt->isRequest()); + assert(pkt->getDest() == Packet::Broadcast); return curTick; } @@ -102,7 +110,6 @@ void MemTest::sendPkt(PacketPtr pkt) { if (atomic) { cachePort.sendAtomic(pkt); - pkt->makeAtomicResponse(); completeRequest(pkt); } else if (!cachePort.sendTiming(pkt)) { @@ -165,8 +172,6 @@ MemTest::MemTest(const string &name, tickEvent.schedule(0); id = TESTER_ALLOCATOR++; - if (TESTER_ALLOCATOR > 8) - panic("False sharing memtester only allows up to 8 testers"); accessRetry = false; } @@ -194,29 +199,25 @@ MemTest::init() // memory should be 0; no need to initialize them. } -static void -printData(ostream &os, uint8_t *data, int nbytes) -{ - os << hex << setfill('0'); - // assume little-endian: print bytes from highest address to lowest - for (uint8_t *dp = data + nbytes - 1; dp >= data; --dp) { - os << setw(2) << (unsigned)*dp; - } - os << dec; -} void MemTest::completeRequest(PacketPtr pkt) { + Request *req = pkt->req; + + DPRINTF(MemTest, "completing %s at address %x (blk %x)\n", + pkt->isWrite() ? "write" : "read", + req->getPaddr(), blockAddr(req->getPaddr())); + MemTestSenderState *state = dynamic_cast<MemTestSenderState *>(pkt->senderState); uint8_t *data = state->data; uint8_t *pkt_data = pkt->getPtr<uint8_t>(); - Request *req = pkt->req; //Remove the address from the list of outstanding - std::set<unsigned>::iterator removeAddr = outstandingAddrs.find(req->getPaddr()); + std::set<unsigned>::iterator removeAddr = + outstandingAddrs.find(req->getPaddr()); assert(removeAddr != outstandingAddrs.end()); outstandingAddrs.erase(removeAddr); @@ -224,15 +225,10 @@ MemTest::completeRequest(PacketPtr pkt) case MemCmd::ReadResp: if (memcmp(pkt_data, data, pkt->getSize()) != 0) { - cerr << name() << ": on read of 0x" << hex << req->getPaddr() - << " (0x" << hex << blockAddr(req->getPaddr()) << ")" - << "@ cycle " << dec << curTick - << ", cache returns 0x"; - printData(cerr, pkt_data, pkt->getSize()); - cerr << ", expected 0x"; - printData(cerr, data, pkt->getSize()); - cerr << endl; - fatal(""); + panic("%s: read of %x (blk %x) @ cycle %d " + "returns %x, expected %x\n", name(), + req->getPaddr(), blockAddr(req->getPaddr()), curTick, + *pkt_data, *data); } numReads++; @@ -244,40 +240,18 @@ MemTest::completeRequest(PacketPtr pkt) nextProgressMessage += progressInterval; } - if (numReads >= maxLoads) - exitSimLoop("Maximum number of loads reached!"); + if (maxLoads != 0 && numReads >= maxLoads) + exitSimLoop("maximum number of loads reached"); break; case MemCmd::WriteResp: numWritesStat++; break; -/* - case Copy: - //Also remove dest from outstanding list - removeAddr = outstandingAddrs.find(req->dest); - assert(removeAddr != outstandingAddrs.end()); - outstandingAddrs.erase(removeAddr); - numCopiesStat++; - break; -*/ + default: panic("invalid command %s (%d)", pkt->cmdString(), pkt->cmd.toInt()); } - if (blockAddr(req->getPaddr()) == traceBlockAddr) { - cerr << name() << ": completed " - << (pkt->isWrite() ? "write" : "read") - << " access of " - << dec << pkt->getSize() << " bytes at address 0x" - << hex << req->getPaddr() - << " (0x" << hex << blockAddr(req->getPaddr()) << ")" - << ", value = 0x"; - printData(cerr, pkt_data, pkt->getSize()); - cerr << " @ cycle " << dec << curTick; - - cerr << endl; - } - noResponseCycles = 0; delete state; delete [] data; @@ -333,7 +307,7 @@ MemTest::tick() //mem tester //We can eliminate the lower bits of the offset, and then use the id //to offset within the blks - offset &= ~63; //Not the low order bits + offset = blockAddr(offset); offset += id; access_size = 0; @@ -359,71 +333,55 @@ MemTest::tick() if (cmd < percentReads) { // read - //For now we only allow one outstanding request per addreess per tester - //This means we assume CPU does write forwarding to reads that alias something - //in the cpu store buffer. + // For now we only allow one outstanding request per address + // per tester This means we assume CPU does write forwarding + // to reads that alias something in the cpu store buffer. if (outstandingAddrs.find(paddr) != outstandingAddrs.end()) { delete [] result; delete req; return; } - else outstandingAddrs.insert(paddr); + + outstandingAddrs.insert(paddr); // ***** NOTE FOR RON: I'm not sure how to access checkMem. - Kevin funcPort.readBlob(req->getPaddr(), result, req->getSize()); - if (blockAddr(paddr) == traceBlockAddr) { - cerr << name() - << ": initiating read " - << ((probe) ? "probe of " : "access of ") - << dec << req->getSize() << " bytes from addr 0x" - << hex << paddr - << " (0x" << hex << blockAddr(paddr) << ")" - << " at cycle " - << dec << curTick << endl; - } + DPRINTF(MemTest, + "initiating read at address %x (blk %x) expecting %x\n", + req->getPaddr(), blockAddr(req->getPaddr()), *result); PacketPtr pkt = new Packet(req, MemCmd::ReadReq, Packet::Broadcast); + pkt->setSrc(0); pkt->dataDynamicArray(new uint8_t[req->getSize()]); MemTestSenderState *state = new MemTestSenderState(result); pkt->senderState = state; if (probe) { cachePort.sendFunctional(pkt); - pkt->makeAtomicResponse(); completeRequest(pkt); } else { -// req->completionEvent = new MemCompleteEvent(req, result, this); sendPkt(pkt); } } else { // write - //For now we only allow one outstanding request per addreess per tester - //This means we assume CPU does write forwarding to reads that alias something - //in the cpu store buffer. + // For now we only allow one outstanding request per addreess + // per tester. This means we assume CPU does write forwarding + // to reads that alias something in the cpu store buffer. if (outstandingAddrs.find(paddr) != outstandingAddrs.end()) { delete [] result; delete req; return; } - else outstandingAddrs.insert(paddr); + outstandingAddrs.insert(paddr); + + DPRINTF(MemTest, "initiating write at address %x (blk %x) value %x\n", + req->getPaddr(), blockAddr(req->getPaddr()), data & 0xff); -/* - if (blockAddr(req->getPaddr()) == traceBlockAddr) { - cerr << name() << ": initiating write " - << ((probe)?"probe of ":"access of ") - << dec << req->getSize() << " bytes (value = 0x"; - printData(cerr, data_pkt->getPtr(), req->getSize()); - cerr << ") to addr 0x" - << hex << req->getPaddr() - << " (0x" << hex << blockAddr(req->getPaddr()) << ")" - << " at cycle " - << dec << curTick << endl; - } -*/ PacketPtr pkt = new Packet(req, MemCmd::WriteReq, Packet::Broadcast); + pkt->setSrc(0); uint8_t *pkt_data = new uint8_t[req->getSize()]; pkt->dataDynamicArray(pkt_data); memcpy(pkt_data, &data, req->getSize()); @@ -434,57 +392,11 @@ MemTest::tick() if (probe) { cachePort.sendFunctional(pkt); - pkt->makeAtomicResponse(); completeRequest(pkt); } else { -// req->completionEvent = new MemCompleteEvent(req, NULL, this); sendPkt(pkt); } } -/* else { - // copy - unsigned source_align = random() % 100; - unsigned dest_align = random() % 100; - unsigned offset2 = random() % size; - - Addr source = ((base) ? baseAddr1 : baseAddr2) + offset; - Addr dest = ((base) ? baseAddr2 : baseAddr1) + offset2; - if (outstandingAddrs.find(source) != outstandingAddrs.end()) return; - else outstandingAddrs.insert(source); - if (outstandingAddrs.find(dest) != outstandingAddrs.end()) return; - else outstandingAddrs.insert(dest); - - if (source_align >= percentSourceUnaligned) { - source = blockAddr(source); - } - if (dest_align >= percentDestUnaligned) { - dest = blockAddr(dest); - } - req->cmd = Copy; - req->flags &= ~UNCACHEABLE; - req->paddr = source; - req->dest = dest; - delete [] req->data; - req->data = new uint8_t[blockSize]; - req->size = blockSize; - if (source == traceBlockAddr || dest == traceBlockAddr) { - cerr << name() - << ": initiating copy of " - << dec << req->size << " bytes from addr 0x" - << hex << source - << " (0x" << hex << blockAddr(source) << ")" - << " to addr 0x" - << hex << dest - << " (0x" << hex << blockAddr(dest) << ")" - << " at cycle " - << dec << curTick << endl; - }* - cacheInterface->access(req); - uint8_t result[blockSize]; - checkMem->access(Read, source, &result, blockSize); - checkMem->access(Write, dest, &result, blockSize); - } -*/ } void diff --git a/src/cpu/memtest/memtest.hh b/src/cpu/memtest/memtest.hh index 123ee2a6c..f4713709a 100644 --- a/src/cpu/memtest/memtest.hh +++ b/src/cpu/memtest/memtest.hh @@ -35,8 +35,6 @@ #include <set> #include "base/statistics.hh" -//#include "mem/functional/functional.hh" -//#include "mem/mem_interface.hh" #include "sim/eventq.hh" #include "sim/sim_exit.hh" #include "sim/sim_object.hh" @@ -50,9 +48,6 @@ class MemTest : public MemObject public: MemTest(const std::string &name, -// MemInterface *_cache_interface, -// PhysicalMemory *main_mem, -// PhysicalMemory *check_mem, unsigned _memorySize, unsigned _percentReads, unsigned _percentFunctional, @@ -85,13 +80,13 @@ class MemTest : public MemObject TickEvent(MemTest *c) : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c) {} void process() {cpu->tick();} - virtual const char *description() { return "tick event"; } + virtual const char *description() { return "MemTest tick"; } }; TickEvent tickEvent; + class CpuPort : public Port { - MemTest *memtest; public: @@ -116,7 +111,7 @@ class MemTest : public MemObject virtual void getDeviceAddressRanges(AddrRangeList &resp, bool &snoop) - { resp.clear(); snoop = true; } + { resp.clear(); snoop = false; } }; CpuPort cachePort; @@ -136,12 +131,7 @@ class MemTest : public MemObject uint8_t *data; }; -// Request *dataReq; PacketPtr retryPkt; -// MemInterface *cacheInterface; -// PhysicalMemory *mainMem; -// PhysicalMemory *checkMem; -// SimpleThread *thread; bool accessRetry; diff --git a/src/cpu/o3/commit_impl.hh b/src/cpu/o3/commit_impl.hh index 9411c6c62..f263383ae 100644 --- a/src/cpu/o3/commit_impl.hh +++ b/src/cpu/o3/commit_impl.hh @@ -67,7 +67,7 @@ template <class Impl> const char * DefaultCommit<Impl>::TrapEvent::description() { - return "Trap event"; + return "Trap"; } template <class Impl> diff --git a/src/cpu/o3/cpu.cc b/src/cpu/o3/cpu.cc index 20ed5df63..cae6ae20c 100644 --- a/src/cpu/o3/cpu.cc +++ b/src/cpu/o3/cpu.cc @@ -83,7 +83,7 @@ template <class Impl> const char * FullO3CPU<Impl>::TickEvent::description() { - return "FullO3CPU tick event"; + return "FullO3CPU tick"; } template <class Impl> @@ -112,7 +112,7 @@ template <class Impl> const char * FullO3CPU<Impl>::ActivateThreadEvent::description() { - return "FullO3CPU \"Activate Thread\" event"; + return "FullO3CPU \"Activate Thread\""; } template <class Impl> @@ -144,7 +144,7 @@ template <class Impl> const char * FullO3CPU<Impl>::DeallocateContextEvent::description() { - return "FullO3CPU \"Deallocate Context\" event"; + return "FullO3CPU \"Deallocate Context\""; } template <class Impl> diff --git a/src/cpu/o3/fetch_impl.hh b/src/cpu/o3/fetch_impl.hh index d1c660258..ab411c7fe 100644 --- a/src/cpu/o3/fetch_impl.hh +++ b/src/cpu/o3/fetch_impl.hh @@ -628,12 +628,6 @@ DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid // Now do the timing access to see whether or not the instruction // exists within the cache. if (!icachePort->sendTiming(data_pkt)) { - if (data_pkt->result == Packet::BadAddress) { - fault = TheISA::genMachineCheckFault(); - delete mem_req; - memReq[tid] = NULL; - warn("Bad address!\n"); - } assert(retryPkt == NULL); assert(retryTid == -1); DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); diff --git a/src/cpu/o3/inst_queue_impl.hh b/src/cpu/o3/inst_queue_impl.hh index bbdd4ddd5..47f019ebe 100644 --- a/src/cpu/o3/inst_queue_impl.hh +++ b/src/cpu/o3/inst_queue_impl.hh @@ -60,7 +60,7 @@ template <class Impl> const char * InstructionQueue<Impl>::FUCompletion::description() { - return "Functional unit completion event"; + return "Functional unit completion"; } template <class Impl> diff --git a/src/cpu/o3/lsq_impl.hh b/src/cpu/o3/lsq_impl.hh index b4a6a02da..10c0afd38 100644 --- a/src/cpu/o3/lsq_impl.hh +++ b/src/cpu/o3/lsq_impl.hh @@ -84,9 +84,10 @@ LSQ<Impl>::DcachePort::recvTiming(PacketPtr pkt) lsq->thread[pkt->req->getThreadNum()].completeDataAccess(pkt); } else { - //else it is a coherence request, maybe you need to do something - warn("Recieved a coherence request (Invalidate?), 03CPU doesn't" - "update LSQ for these\n"); + // must be a snoop + + // @TODO someday may need to process invalidations in LSQ here + // to provide stronger consistency model } return true; } diff --git a/src/cpu/o3/lsq_unit.hh b/src/cpu/o3/lsq_unit.hh index cc33e025d..be9224099 100644 --- a/src/cpu/o3/lsq_unit.hh +++ b/src/cpu/o3/lsq_unit.hh @@ -643,7 +643,10 @@ LSQUnit<Impl>::read(Request *req, T &data, int load_idx) // if we the cache is not blocked, do cache access if (!lsq->cacheBlocked()) { PacketPtr data_pkt = - new Packet(req, MemCmd::ReadReq, Packet::Broadcast); + new Packet(req, + (req->isLocked() ? + MemCmd::LoadLockedReq : MemCmd::ReadReq), + Packet::Broadcast); data_pkt->dataStatic(load_inst->memData); LSQSenderState *state = new LSQSenderState; @@ -653,8 +656,6 @@ LSQUnit<Impl>::read(Request *req, T &data, int load_idx) data_pkt->senderState = state; if (!dcachePort->sendTiming(data_pkt)) { - Packet::Result result = data_pkt->result; - // Delete state and data packet because a load retry // initiates a pipeline restart; it does not retry. delete state; @@ -663,10 +664,6 @@ LSQUnit<Impl>::read(Request *req, T &data, int load_idx) req = NULL; - if (result == Packet::BadAddress) { - return TheISA::genMachineCheckFault(); - } - // If the access didn't succeed, tell the LSQ by setting // the retry thread id. lsq->setRetryTid(lsqID); diff --git a/src/cpu/o3/lsq_unit_impl.hh b/src/cpu/o3/lsq_unit_impl.hh index bde4f8079..8b2e82d8e 100644 --- a/src/cpu/o3/lsq_unit_impl.hh +++ b/src/cpu/o3/lsq_unit_impl.hh @@ -69,7 +69,7 @@ template<class Impl> const char * LSQUnit<Impl>::WritebackEvent::description() { - return "Store writeback event"; + return "Store writeback"; } template<class Impl> @@ -647,7 +647,9 @@ LSQUnit<Impl>::writebackStores() memcpy(inst->memData, storeQueue[storeWBIdx].data, req->getSize()); - MemCmd command = req->isSwap() ? MemCmd::SwapReq : MemCmd::WriteReq; + MemCmd command = + req->isSwap() ? MemCmd::SwapReq : + (req->isLocked() ? MemCmd::StoreCondReq : MemCmd::WriteReq); PacketPtr data_pkt = new Packet(req, command, Packet::Broadcast); data_pkt->dataStatic(inst->memData); @@ -690,9 +692,6 @@ LSQUnit<Impl>::writebackStores() } if (!dcachePort->sendTiming(data_pkt)) { - if (data_pkt->result == Packet::BadAddress) { - panic("LSQ sent out a bad address for a completed store!"); - } // Need to handle becoming blocked on a store. DPRINTF(IEW, "D-Cache became blocked when writing [sn:%lli], will" "retry later\n", @@ -844,26 +843,6 @@ LSQUnit<Impl>::storePostSend(PacketPtr pkt) #endif } - if (pkt->result != Packet::Success) { - DPRINTF(LSQUnit,"D-Cache Write Miss on idx:%i!\n", - storeWBIdx); - - DPRINTF(Activity, "Active st accessing mem miss [sn:%lli]\n", - storeQueue[storeWBIdx].inst->seqNum); - - //mshrSeqNums.push_back(storeQueue[storeWBIdx].inst->seqNum); - - //DPRINTF(LSQUnit, "Added MSHR. count = %i\n",mshrSeqNums.size()); - - // @todo: Increment stat here. - } else { - DPRINTF(LSQUnit,"D-Cache: Write Hit on idx:%i !\n", - storeWBIdx); - - DPRINTF(Activity, "Active st accessing mem hit [sn:%lli]\n", - storeQueue[storeWBIdx].inst->seqNum); - } - incrStIdx(storeWBIdx); } @@ -952,9 +931,6 @@ LSQUnit<Impl>::recvRetry() assert(retryPkt != NULL); if (dcachePort->sendTiming(retryPkt)) { - if (retryPkt->result == Packet::BadAddress) { - panic("LSQ sent out a bad address for a completed store!"); - } storePostSend(retryPkt); retryPkt = NULL; isStoreBlocked = false; diff --git a/src/cpu/ozone/back_end_impl.hh b/src/cpu/ozone/back_end_impl.hh index 4078699fe..27146ecf0 100644 --- a/src/cpu/ozone/back_end_impl.hh +++ b/src/cpu/ozone/back_end_impl.hh @@ -583,7 +583,7 @@ template<class Impl> const char * BackEnd<Impl>::LdWritebackEvent::description() { - return "Load writeback event"; + return "Load writeback"; } @@ -603,7 +603,7 @@ template <class Impl> const char * BackEnd<Impl>::DCacheCompletionEvent::description() { - return "Cache completion event"; + return "Cache completion"; } template <class Impl> diff --git a/src/cpu/ozone/cpu_impl.hh b/src/cpu/ozone/cpu_impl.hh index d1214223b..d73e5768a 100644 --- a/src/cpu/ozone/cpu_impl.hh +++ b/src/cpu/ozone/cpu_impl.hh @@ -84,7 +84,7 @@ template <class Impl> const char * OzoneCPU<Impl>::TickEvent::description() { - return "OzoneCPU tick event"; + return "OzoneCPU tick"; } template <class Impl> diff --git a/src/cpu/ozone/inorder_back_end_impl.hh b/src/cpu/ozone/inorder_back_end_impl.hh index 8d7ebb60e..c57fa0200 100644 --- a/src/cpu/ozone/inorder_back_end_impl.hh +++ b/src/cpu/ozone/inorder_back_end_impl.hh @@ -540,5 +540,5 @@ template <class Impl> const char * InorderBackEnd<Impl>::DCacheCompletionEvent::description() { - return "DCache completion event"; + return "DCache completion"; } diff --git a/src/cpu/ozone/inst_queue_impl.hh b/src/cpu/ozone/inst_queue_impl.hh index ea9d03c0d..461c7eb0f 100644 --- a/src/cpu/ozone/inst_queue_impl.hh +++ b/src/cpu/ozone/inst_queue_impl.hh @@ -64,7 +64,7 @@ template <class Impl> const char * InstQueue<Impl>::FUCompletion::description() { - return "Functional unit completion event"; + return "Functional unit completion"; } #endif template <class Impl> diff --git a/src/cpu/ozone/lsq_unit_impl.hh b/src/cpu/ozone/lsq_unit_impl.hh index c46eb90be..e08e54835 100644 --- a/src/cpu/ozone/lsq_unit_impl.hh +++ b/src/cpu/ozone/lsq_unit_impl.hh @@ -62,7 +62,7 @@ template <class Impl> const char * OzoneLSQ<Impl>::StoreCompletionEvent::description() { - return "LSQ store completion event"; + return "LSQ store completion"; } template <class Impl> diff --git a/src/cpu/ozone/lw_back_end_impl.hh b/src/cpu/ozone/lw_back_end_impl.hh index c0a9cad24..f84bda348 100644 --- a/src/cpu/ozone/lw_back_end_impl.hh +++ b/src/cpu/ozone/lw_back_end_impl.hh @@ -121,7 +121,7 @@ template <class Impl> const char * LWBackEnd<Impl>::TrapEvent::description() { - return "Trap event"; + return "Trap"; } template <class Impl> diff --git a/src/cpu/ozone/lw_lsq.hh b/src/cpu/ozone/lw_lsq.hh index 2048ad6bb..ba40e9ce1 100644 --- a/src/cpu/ozone/lw_lsq.hh +++ b/src/cpu/ozone/lw_lsq.hh @@ -632,7 +632,11 @@ OzoneLWLSQ<Impl>::read(RequestPtr req, T &data, int load_idx) DPRINTF(OzoneLSQ, "Doing timing access for inst PC %#x\n", inst->readPC()); - PacketPtr data_pkt = new Packet(req, Packet::ReadReq, Packet::Broadcast); + PacketPtr data_pkt = + new Packet(req, + (req->isLocked() ? + MemCmd::LoadLockedReq : Packet::ReadReq), + Packet::Broadcast); data_pkt->dataStatic(inst->memData); LSQSenderState *state = new LSQSenderState; @@ -661,16 +665,6 @@ OzoneLWLSQ<Impl>::read(RequestPtr req, T &data, int load_idx) cpu->lockFlag = true; } - if (data_pkt->result != Packet::Success) { - DPRINTF(OzoneLSQ, "OzoneLSQ: D-cache miss!\n"); - DPRINTF(Activity, "Activity: ld accessing mem miss [sn:%lli]\n", - inst->seqNum); - } else { - DPRINTF(OzoneLSQ, "OzoneLSQ: D-cache hit!\n"); - DPRINTF(Activity, "Activity: ld accessing mem hit [sn:%lli]\n", - inst->seqNum); - } - return NoFault; } diff --git a/src/cpu/ozone/lw_lsq_impl.hh b/src/cpu/ozone/lw_lsq_impl.hh index f26b06453..82191312a 100644 --- a/src/cpu/ozone/lw_lsq_impl.hh +++ b/src/cpu/ozone/lw_lsq_impl.hh @@ -57,7 +57,7 @@ template<class Impl> const char * OzoneLWLSQ<Impl>::WritebackEvent::description() { - return "Store writeback event"; + return "Store writeback"; } template <class Impl> @@ -587,7 +587,10 @@ OzoneLWLSQ<Impl>::writebackStores() memcpy(inst->memData, (uint8_t *)&(*sq_it).data, req->getSize()); - PacketPtr data_pkt = new Packet(req, Packet::WriteReq, Packet::Broadcast); + MemCmd command = + req->isSwap() ? MemCmd::SwapReq : + (req->isLocked() ? MemCmd::WriteReq : MemCmd::StoreCondReq); + PacketPtr data_pkt = new Packet(req, command, Packet::Broadcast); data_pkt->dataStatic(inst->memData); LSQSenderState *state = new LSQSenderState; @@ -853,24 +856,6 @@ OzoneLWLSQ<Impl>::storePostSend(PacketPtr pkt, DynInstPtr &inst) } #endif } - - if (pkt->result != Packet::Success) { - DPRINTF(OzoneLSQ,"D-Cache Write Miss!\n"); - - DPRINTF(Activity, "Active st accessing mem miss [sn:%lli]\n", - inst->seqNum); - - //mshrSeqNums.push_back(storeQueue[storeWBIdx].inst->seqNum); - - //DPRINTF(OzoneLWLSQ, "Added MSHR. count = %i\n",mshrSeqNums.size()); - - // @todo: Increment stat here. - } else { - DPRINTF(OzoneLSQ,"D-Cache: Write Hit!\n"); - - DPRINTF(Activity, "Active st accessing mem hit [sn:%lli]\n", - inst->seqNum); - } } template <class Impl> diff --git a/src/cpu/quiesce_event.cc b/src/cpu/quiesce_event.cc index fa79e6d1e..3495a0e52 100644 --- a/src/cpu/quiesce_event.cc +++ b/src/cpu/quiesce_event.cc @@ -47,5 +47,5 @@ EndQuiesceEvent::process() const char* EndQuiesceEvent::description() { - return "End Quiesce Event."; + return "End Quiesce"; } diff --git a/src/cpu/simple/atomic.cc b/src/cpu/simple/atomic.cc index 888ef4960..704b65f36 100644 --- a/src/cpu/simple/atomic.cc +++ b/src/cpu/simple/atomic.cc @@ -57,7 +57,7 @@ AtomicSimpleCPU::TickEvent::process() const char * AtomicSimpleCPU::TickEvent::description() { - return "AtomicSimpleCPU tick event"; + return "AtomicSimpleCPU tick"; } Port * @@ -148,23 +148,9 @@ AtomicSimpleCPU::AtomicSimpleCPU(Params *p) icachePort.snoopRangeSent = false; dcachePort.snoopRangeSent = false; - ifetch_req = new Request(); - ifetch_req->setThreadContext(p->cpu_id, 0); // Add thread ID if we add MT - ifetch_pkt = new Packet(ifetch_req, MemCmd::ReadReq, Packet::Broadcast); - ifetch_pkt->dataStatic(&inst); - - data_read_req = new Request(); - data_read_req->setThreadContext(p->cpu_id, 0); // Add thread ID here too - data_read_pkt = new Packet(data_read_req, MemCmd::ReadReq, - Packet::Broadcast); - data_read_pkt->dataStatic(&dataReg); - - data_write_req = new Request(); - data_write_req->setThreadContext(p->cpu_id, 0); // Add thread ID here too - data_write_pkt = new Packet(data_write_req, MemCmd::WriteReq, - Packet::Broadcast); - data_swap_pkt = new Packet(data_write_req, MemCmd::SwapReq, - Packet::Broadcast); + ifetch_req.setThreadContext(p->cpu_id, 0); // Add thread ID if we add MT + data_read_req.setThreadContext(p->cpu_id, 0); // Add thread ID here too + data_write_req.setThreadContext(p->cpu_id, 0); // Add thread ID here too } @@ -197,6 +183,7 @@ AtomicSimpleCPU::unserialize(Checkpoint *cp, const string §ion) void AtomicSimpleCPU::resume() { + DPRINTF(SimpleCPU, "Resume\n"); if (_status != SwitchedOut && _status != Idle) { assert(system->getMemoryMode() == Enums::atomic); @@ -245,6 +232,8 @@ AtomicSimpleCPU::takeOverFrom(BaseCPU *oldCPU) void AtomicSimpleCPU::activateContext(int thread_num, int delay) { + DPRINTF(SimpleCPU, "ActivateContext %d (%d cycles)\n", thread_num, delay); + assert(thread_num == 0); assert(thread); @@ -262,6 +251,8 @@ AtomicSimpleCPU::activateContext(int thread_num, int delay) void AtomicSimpleCPU::suspendContext(int thread_num) { + DPRINTF(SimpleCPU, "SuspendContext %d\n", thread_num); + assert(thread_num == 0); assert(thread); @@ -282,9 +273,7 @@ Fault AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags) { // use the CPU's statically allocated read request and packet objects - Request *req = data_read_req; - PacketPtr pkt = data_read_pkt; - + Request *req = &data_read_req; req->setVirt(0, addr, sizeof(T), flags, thread->readPC()); if (traceData) { @@ -296,19 +285,20 @@ AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags) // Now do the access. if (fault == NoFault) { - pkt->reinitFromRequest(); + Packet pkt = + Packet(req, + req->isLocked() ? MemCmd::LoadLockedReq : MemCmd::ReadReq, + Packet::Broadcast); + pkt.dataStatic(&data); if (req->isMmapedIpr()) - dcache_latency = TheISA::handleIprRead(thread->getTC(),pkt); + dcache_latency = TheISA::handleIprRead(thread->getTC(), &pkt); else - dcache_latency = dcachePort.sendAtomic(pkt); + dcache_latency = dcachePort.sendAtomic(&pkt); dcache_access = true; -#if !defined(NDEBUG) - if (pkt->result != Packet::Success) - panic("Unable to find responder for address pa = %#X va = %#X\n", - pkt->req->getPaddr(), pkt->req->getVaddr()); -#endif - data = pkt->get<T>(); + assert(!pkt.isError()); + + data = gtoh(data); if (req->isLocked()) { TheISA::handleLockedRead(thread, req); @@ -378,16 +368,9 @@ Fault AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) { // use the CPU's statically allocated write request and packet objects - Request *req = data_write_req; - PacketPtr pkt; - + Request *req = &data_write_req; req->setVirt(0, addr, sizeof(T), flags, thread->readPC()); - if (req->isSwap()) - pkt = data_swap_pkt; - else - pkt = data_write_pkt; - if (traceData) { traceData->setAddr(addr); } @@ -397,40 +380,40 @@ AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) // Now do the access. if (fault == NoFault) { + MemCmd cmd = MemCmd::WriteReq; // default bool do_access = true; // flag to suppress cache access if (req->isLocked()) { + cmd = MemCmd::StoreCondReq; do_access = TheISA::handleLockedWrite(thread, req); + } else if (req->isSwap()) { + cmd = MemCmd::SwapReq; + if (req->isCondSwap()) { + assert(res); + req->setExtraData(*res); + } } - if (req->isCondSwap()) { - assert(res); - req->setExtraData(*res); - } - if (do_access) { - pkt->reinitFromRequest(); - pkt->dataStatic(&data); + Packet pkt = Packet(req, cmd, Packet::Broadcast); + pkt.dataStatic(&data); if (req->isMmapedIpr()) { - dcache_latency = TheISA::handleIprWrite(thread->getTC(), pkt); + dcache_latency = TheISA::handleIprWrite(thread->getTC(), &pkt); } else { data = htog(data); - dcache_latency = dcachePort.sendAtomic(pkt); + dcache_latency = dcachePort.sendAtomic(&pkt); } dcache_access = true; + assert(!pkt.isError()); -#if !defined(NDEBUG) - if (pkt->result != Packet::Success) - panic("Unable to find responder for address pa = %#X va = %#X\n", - pkt->req->getPaddr(), pkt->req->getVaddr()); -#endif + if (req->isSwap()) { + assert(res); + *res = pkt.get<T>(); + } } - if (req->isSwap()) { - assert(res); - *res = pkt->get<T>(); - } else if (res) { + if (res && !req->isSwap()) { *res = req->getExtraData(); } } @@ -505,6 +488,8 @@ AtomicSimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res) void AtomicSimpleCPU::tick() { + DPRINTF(SimpleCPU, "Tick\n"); + Tick latency = cycles(1); // instruction takes one cycle by default for (int i = 0; i < width; ++i) { @@ -513,7 +498,7 @@ AtomicSimpleCPU::tick() if (!curStaticInst || !curStaticInst->isDelayedCommit()) checkForInterrupts(); - Fault fault = setupFetchRequest(ifetch_req); + Fault fault = setupFetchRequest(&ifetch_req); if (fault == NoFault) { Tick icache_latency = 0; @@ -524,9 +509,11 @@ AtomicSimpleCPU::tick() //if(predecoder.needMoreBytes()) //{ icache_access = true; - ifetch_pkt->reinitFromRequest(); + Packet ifetch_pkt = Packet(&ifetch_req, MemCmd::ReadReq, + Packet::Broadcast); + ifetch_pkt.dataStatic(&inst); - icache_latency = icachePort.sendAtomic(ifetch_pkt); + icache_latency = icachePort.sendAtomic(&ifetch_pkt); // ifetch_req is initialized to read the instruction directly // into the CPU object's inst field. //} diff --git a/src/cpu/simple/atomic.hh b/src/cpu/simple/atomic.hh index b127e3791..28e883b24 100644 --- a/src/cpu/simple/atomic.hh +++ b/src/cpu/simple/atomic.hh @@ -121,13 +121,9 @@ class AtomicSimpleCPU : public BaseSimpleCPU }; DcachePort dcachePort; - Request *ifetch_req; - PacketPtr ifetch_pkt; - Request *data_read_req; - PacketPtr data_read_pkt; - Request *data_write_req; - PacketPtr data_write_pkt; - PacketPtr data_swap_pkt; + Request ifetch_req; + Request data_read_req; + Request data_write_req; bool dcache_access; Tick dcache_latency; diff --git a/src/cpu/simple/base.hh b/src/cpu/simple/base.hh index 243167db0..0550aa036 100644 --- a/src/cpu/simple/base.hh +++ b/src/cpu/simple/base.hh @@ -131,9 +131,6 @@ class BaseSimpleCPU : public BaseCPU // The predecoder TheISA::Predecoder predecoder; - // Static data storage - TheISA::LargestRead dataReg; - StaticInstPtr curStaticInst; StaticInstPtr curMacroStaticInst; diff --git a/src/cpu/simple/timing.cc b/src/cpu/simple/timing.cc index 855aaab59..a70ca7c75 100644 --- a/src/cpu/simple/timing.cc +++ b/src/cpu/simple/timing.cc @@ -260,7 +260,10 @@ TimingSimpleCPU::read(Addr addr, T &data, unsigned flags) // Now do the access. if (fault == NoFault) { PacketPtr pkt = - new Packet(req, MemCmd::ReadReq, Packet::Broadcast); + new Packet(req, + (req->isLocked() ? + MemCmd::LoadLockedReq : MemCmd::ReadReq), + Packet::Broadcast); pkt->dataDynamic<T>(new T); if (!dcachePort.sendTiming(pkt)) { @@ -350,23 +353,26 @@ TimingSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) // Now do the access. if (fault == NoFault) { - assert(dcache_pkt == NULL); - if (req->isSwap()) - dcache_pkt = new Packet(req, MemCmd::SwapReq, Packet::Broadcast); - else - dcache_pkt = new Packet(req, MemCmd::WriteReq, Packet::Broadcast); - dcache_pkt->allocate(); - dcache_pkt->set(data); - + MemCmd cmd = MemCmd::WriteReq; // default bool do_access = true; // flag to suppress cache access if (req->isLocked()) { + cmd = MemCmd::StoreCondReq; do_access = TheISA::handleLockedWrite(thread, req); + } else if (req->isSwap()) { + cmd = MemCmd::SwapReq; + if (req->isCondSwap()) { + assert(res); + req->setExtraData(*res); + } } - if (req->isCondSwap()) { - assert(res); - req->setExtraData(*res); - } + + // Note: need to allocate dcache_pkt even if do_access is + // false, as it's used unconditionally to call completeAcc(). + assert(dcache_pkt == NULL); + dcache_pkt = new Packet(req, cmd, Packet::Broadcast); + dcache_pkt->allocate(); + dcache_pkt->set(data); if (do_access) { if (!dcachePort.sendTiming(dcache_pkt)) { @@ -501,7 +507,7 @@ TimingSimpleCPU::completeIfetch(PacketPtr pkt) { // received a response from the icache: execute the received // instruction - assert(pkt->result == Packet::Success); + assert(!pkt->isError()); assert(_status == IcacheWaitResponse); _status = Running; @@ -569,7 +575,7 @@ TimingSimpleCPU::IcachePort::recvTiming(PacketPtr pkt) return true; } - else if (pkt->result == Packet::Nacked) { + else if (pkt->wasNacked()) { assert(cpu->_status == IcacheWaitResponse); pkt->reinitNacked(); if (!sendTiming(pkt)) { @@ -600,7 +606,7 @@ TimingSimpleCPU::completeDataAccess(PacketPtr pkt) { // received a response from the dcache: complete the load or store // instruction - assert(pkt->result == Packet::Success); + assert(!pkt->isError()); assert(_status == DcacheWaitResponse); _status = Running; @@ -609,7 +615,7 @@ TimingSimpleCPU::completeDataAccess(PacketPtr pkt) Fault fault = curStaticInst->completeAcc(pkt, this, traceData); - if (pkt->isRead() && pkt->req->isLocked()) { + if (pkt->isRead() && pkt->isLocked()) { TheISA::handleLockedRead(thread, pkt->req); } @@ -663,7 +669,7 @@ TimingSimpleCPU::DcachePort::recvTiming(PacketPtr pkt) return true; } - else if (pkt->result == Packet::Nacked) { + else if (pkt->wasNacked()) { assert(cpu->_status == DcacheWaitResponse); pkt->reinitNacked(); if (!sendTiming(pkt)) { diff --git a/src/cpu/simple/timing.hh b/src/cpu/simple/timing.hh index 39958bfb6..ba194b3fa 100644 --- a/src/cpu/simple/timing.hh +++ b/src/cpu/simple/timing.hh @@ -101,7 +101,7 @@ class TimingSimpleCPU : public BaseSimpleCPU TickEvent(TimingSimpleCPU *_cpu) :Event(&mainEventQueue), cpu(_cpu) {} - const char *description() { return "Timing CPU clock event"; } + const char *description() { return "Timing CPU tick"; } void schedule(PacketPtr _pkt, Tick t); }; @@ -127,7 +127,7 @@ class TimingSimpleCPU : public BaseSimpleCPU ITickEvent(TimingSimpleCPU *_cpu) : TickEvent(_cpu) {} void process(); - const char *description() { return "Timing CPU clock event"; } + const char *description() { return "Timing CPU icache tick"; } }; ITickEvent tickEvent; @@ -155,7 +155,7 @@ class TimingSimpleCPU : public BaseSimpleCPU DTickEvent(TimingSimpleCPU *_cpu) : TickEvent(_cpu) {} void process(); - const char *description() { return "Timing CPU clock event"; } + const char *description() { return "Timing CPU dcache tick"; } }; DTickEvent tickEvent; diff --git a/src/cpu/trace/opt_cpu.cc b/src/cpu/trace/opt_cpu.cc index 467984252..33da3d870 100644 --- a/src/cpu/trace/opt_cpu.cc +++ b/src/cpu/trace/opt_cpu.cc @@ -206,7 +206,7 @@ OptCPU::TickEvent::process() const char * OptCPU::TickEvent::description() { - return "OptCPU tick event"; + return "OptCPU tick"; } diff --git a/src/cpu/trace/trace_cpu.cc b/src/cpu/trace/trace_cpu.cc index a3164221b..e5739b2ce 100644 --- a/src/cpu/trace/trace_cpu.cc +++ b/src/cpu/trace/trace_cpu.cc @@ -148,7 +148,7 @@ TraceCPU::TickEvent::process() const char * TraceCPU::TickEvent::description() { - return "TraceCPU tick event"; + return "TraceCPU tick"; } TraceCPU * |