diff options
Diffstat (limited to 'src/cpu')
-rw-r--r-- | src/cpu/base.hh | 2 | ||||
-rw-r--r-- | src/cpu/base_dyn_inst_impl.hh | 2 | ||||
-rw-r--r-- | src/cpu/checker/cpu.cc | 8 | ||||
-rw-r--r-- | src/cpu/o3/fetch_impl.hh | 2 | ||||
-rw-r--r-- | src/cpu/o3/lsq_unit.hh | 4 | ||||
-rw-r--r-- | src/cpu/o3/lsq_unit_impl.hh | 6 | ||||
-rw-r--r-- | src/cpu/ozone/back_end.hh | 4 | ||||
-rw-r--r-- | src/cpu/ozone/back_end_impl.hh | 2 | ||||
-rw-r--r-- | src/cpu/ozone/cpu.hh | 12 | ||||
-rw-r--r-- | src/cpu/ozone/front_end_impl.hh | 2 | ||||
-rw-r--r-- | src/cpu/ozone/inorder_back_end.hh | 18 | ||||
-rw-r--r-- | src/cpu/ozone/lsq_unit.hh | 2 | ||||
-rw-r--r-- | src/cpu/ozone/lsq_unit_impl.hh | 6 | ||||
-rw-r--r-- | src/cpu/ozone/lw_lsq.hh | 4 | ||||
-rw-r--r-- | src/cpu/ozone/lw_lsq_impl.hh | 10 | ||||
-rw-r--r-- | src/cpu/simple/atomic.cc | 76 | ||||
-rw-r--r-- | src/cpu/simple/timing.cc | 104 | ||||
-rw-r--r-- | src/cpu/simple/timing.hh | 2 | ||||
-rw-r--r-- | src/cpu/simple_thread.hh | 6 |
19 files changed, 161 insertions, 111 deletions
diff --git a/src/cpu/base.hh b/src/cpu/base.hh index e02527371..75e0d86af 100644 --- a/src/cpu/base.hh +++ b/src/cpu/base.hh @@ -140,8 +140,8 @@ class BaseCPU : public MemObject bool functionTrace; Tick functionTraceStart; System *system; -#if FULL_SYSTEM int cpu_id; +#if FULL_SYSTEM Tick profile; #endif Tick progress_interval; diff --git a/src/cpu/base_dyn_inst_impl.hh b/src/cpu/base_dyn_inst_impl.hh index f2109e88d..d6cdff5c5 100644 --- a/src/cpu/base_dyn_inst_impl.hh +++ b/src/cpu/base_dyn_inst_impl.hh @@ -193,7 +193,7 @@ BaseDynInst<Impl>::prefetch(Addr addr, unsigned flags) // note this is a local, not BaseDynInst::fault Fault trans_fault = cpu->translateDataReadReq(req); - if (trans_fault == NoFault && !(req->flags & UNCACHEABLE)) { + if (trans_fault == NoFault && !(req->isUncacheable())) { // It's a valid address to cacheable space. Record key MemReq // parameters so we can generate another one just like it for // the timing access without calling translate() again (which diff --git a/src/cpu/checker/cpu.cc b/src/cpu/checker/cpu.cc index 1540a6b94..f6d56eef6 100644 --- a/src/cpu/checker/cpu.cc +++ b/src/cpu/checker/cpu.cc @@ -175,7 +175,7 @@ CheckerCPU::read(Addr addr, T &data, unsigned flags) pkt->dataStatic(&data); - if (!(memReq->getFlags() & UNCACHEABLE)) { + if (!(memReq->isUncacheable())) { // Access memory to see if we have the same data dcachePort->sendFunctional(pkt); } else { @@ -251,9 +251,9 @@ CheckerCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) // This is because the LSQ would have to be snooped in the CPU to // verify this data. if (unverifiedReq && - !(unverifiedReq->getFlags() & UNCACHEABLE) && - (!(unverifiedReq->getFlags() & LOCKED) || - ((unverifiedReq->getFlags() & LOCKED) && + !(unverifiedReq->isUncacheable()) && + (!(unverifiedReq->isLocked()) || + ((unverifiedReq->isLocked()) && unverifiedReq->getScResult() == 1))) { T inst_data; /* diff --git a/src/cpu/o3/fetch_impl.hh b/src/cpu/o3/fetch_impl.hh index 2d447bfe5..497179576 100644 --- a/src/cpu/o3/fetch_impl.hh +++ b/src/cpu/o3/fetch_impl.hh @@ -599,7 +599,7 @@ DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid if (fault == NoFault) { #if 0 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || - memReq[tid]->flags & UNCACHEABLE) { + memReq[tid]->isUncacheable()) { DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " "misspeculating path)!", memReq[tid]->paddr); diff --git a/src/cpu/o3/lsq_unit.hh b/src/cpu/o3/lsq_unit.hh index 90d1a3d53..58945f04e 100644 --- a/src/cpu/o3/lsq_unit.hh +++ b/src/cpu/o3/lsq_unit.hh @@ -492,7 +492,7 @@ LSQUnit<Impl>::read(Request *req, T &data, int load_idx) // A bit of a hackish way to get uncached accesses to work only if they're // at the head of the LSQ and are ready to commit (at the head of the ROB // too). - if (req->getFlags() & UNCACHEABLE && + if (req->isUncacheable() && (load_idx != loadHead || !load_inst->isAtCommit())) { iewStage->rescheduleMemInst(load_inst); ++lsqRescheduledLoads; @@ -509,7 +509,7 @@ LSQUnit<Impl>::read(Request *req, T &data, int load_idx) load_idx, store_idx, storeHead, req->getPaddr()); #if FULL_SYSTEM - if (req->getFlags() & LOCKED) { + if (req->isLocked()) { cpu->lockAddr = req->getPaddr(); cpu->lockFlag = true; } diff --git a/src/cpu/o3/lsq_unit_impl.hh b/src/cpu/o3/lsq_unit_impl.hh index 98bea74fb..63ffcece1 100644 --- a/src/cpu/o3/lsq_unit_impl.hh +++ b/src/cpu/o3/lsq_unit_impl.hh @@ -416,7 +416,7 @@ LSQUnit<Impl>::executeLoad(DynInstPtr &inst) // realizes there is activity. // Mark it as executed unless it is an uncached load that // needs to hit the head of commit. - if (!(inst->req->getFlags() & UNCACHEABLE) || inst->isAtCommit()) { + if (!(inst->req->isUncacheable()) || inst->isAtCommit()) { inst->setExecuted(); } iewStage->instToCommit(inst); @@ -613,8 +613,8 @@ LSQUnit<Impl>::writebackStores() storeQueue[storeWBIdx].inst->seqNum); // @todo: Remove this SC hack once the memory system handles it. - if (req->getFlags() & LOCKED) { - if (req->getFlags() & UNCACHEABLE) { + if (req->isLocked()) { + if (req->isUncacheable()) { req->setScResult(2); } else { if (cpu->lockFlag) { diff --git a/src/cpu/ozone/back_end.hh b/src/cpu/ozone/back_end.hh index 9bab6a964..8debd277d 100644 --- a/src/cpu/ozone/back_end.hh +++ b/src/cpu/ozone/back_end.hh @@ -493,7 +493,7 @@ BackEnd<Impl>::read(RequestPtr req, T &data, int load_idx) } */ /* - if (!dcacheInterface && (memReq->flags & UNCACHEABLE)) + if (!dcacheInterface && (memReq->isUncacheable())) recordEvent("Uncached Read"); */ return LSQ.read(req, data, load_idx); @@ -534,7 +534,7 @@ BackEnd<Impl>::write(RequestPtr req, T &data, int store_idx) *res = memReq->result; */ /* - if (!dcacheInterface && (memReq->flags & UNCACHEABLE)) + if (!dcacheInterface && (memReq->isUncacheable())) recordEvent("Uncached Write"); */ return LSQ.write(req, data, store_idx); diff --git a/src/cpu/ozone/back_end_impl.hh b/src/cpu/ozone/back_end_impl.hh index ac3218c02..4078699fe 100644 --- a/src/cpu/ozone/back_end_impl.hh +++ b/src/cpu/ozone/back_end_impl.hh @@ -1256,7 +1256,7 @@ BackEnd<Impl>::executeInsts() // ++iewExecStoreInsts; - if (!(inst->req->flags & LOCKED)) { + if (!(inst->req->isLocked())) { inst->setExecuted(); instToCommit(inst); diff --git a/src/cpu/ozone/cpu.hh b/src/cpu/ozone/cpu.hh index 8c5be9424..70ec1d101 100644 --- a/src/cpu/ozone/cpu.hh +++ b/src/cpu/ozone/cpu.hh @@ -455,12 +455,12 @@ class OzoneCPU : public BaseCPU { #if 0 #if FULL_SYSTEM && defined(TARGET_ALPHA) - if (req->flags & LOCKED) { + if (req->isLocked()) { req->xc->setMiscReg(TheISA::Lock_Addr_DepTag, req->paddr); req->xc->setMiscReg(TheISA::Lock_Flag_DepTag, true); } #endif - if (req->flags & LOCKED) { + if (req->isLocked()) { lockAddrList.insert(req->paddr); lockFlag = true; } @@ -489,10 +489,10 @@ class OzoneCPU : public BaseCPU ExecContext *xc; // If this is a store conditional, act appropriately - if (req->flags & LOCKED) { + if (req->isLocked()) { xc = req->xc; - if (req->flags & UNCACHEABLE) { + if (req->isUncacheable()) { // Don't update result register (see stq_c in isa_desc) req->result = 2; xc->setStCondFailures(0);//Needed? [RGD] @@ -532,8 +532,8 @@ class OzoneCPU : public BaseCPU #endif - if (req->flags & LOCKED) { - if (req->flags & UNCACHEABLE) { + if (req->isLocked()) { + if (req->isUncacheable()) { req->result = 2; } else { if (this->lockFlag) { diff --git a/src/cpu/ozone/front_end_impl.hh b/src/cpu/ozone/front_end_impl.hh index d34716de6..5956c5cba 100644 --- a/src/cpu/ozone/front_end_impl.hh +++ b/src/cpu/ozone/front_end_impl.hh @@ -493,7 +493,7 @@ FrontEnd<Impl>::fetchCacheLine() if (fault == NoFault) { #if 0 if (cpu->system->memctrl->badaddr(memReq->paddr) || - memReq->flags & UNCACHEABLE) { + memReq->isUncacheable()) { DPRINTF(FE, "Fetch: Bad address %#x (hopefully on a " "misspeculating path!", memReq->paddr); diff --git a/src/cpu/ozone/inorder_back_end.hh b/src/cpu/ozone/inorder_back_end.hh index ffdba2f6c..76eef6fad 100644 --- a/src/cpu/ozone/inorder_back_end.hh +++ b/src/cpu/ozone/inorder_back_end.hh @@ -231,7 +231,7 @@ InorderBackEnd<Impl>::read(Addr addr, T &data, unsigned flags) } } /* - if (!dcacheInterface && (memReq->flags & UNCACHEABLE)) + if (!dcacheInterface && (memReq->isUncacheable())) recordEvent("Uncached Read"); */ return fault; @@ -243,7 +243,7 @@ Fault InorderBackEnd<Impl>::read(MemReqPtr &req, T &data) { #if FULL_SYSTEM && defined(TARGET_ALPHA) - if (req->flags & LOCKED) { + if (req->isLocked()) { req->xc->setMiscReg(TheISA::Lock_Addr_DepTag, req->paddr); req->xc->setMiscReg(TheISA::Lock_Flag_DepTag, true); } @@ -291,7 +291,7 @@ InorderBackEnd<Impl>::write(T data, Addr addr, unsigned flags, uint64_t *res) if (res && (fault == NoFault)) *res = memReq->result; /* - if (!dcacheInterface && (memReq->flags & UNCACHEABLE)) + if (!dcacheInterface && (memReq->isUncacheable())) recordEvent("Uncached Write"); */ return fault; @@ -306,10 +306,10 @@ InorderBackEnd<Impl>::write(MemReqPtr &req, T &data) ExecContext *xc; // If this is a store conditional, act appropriately - if (req->flags & LOCKED) { + if (req->isLocked()) { xc = req->xc; - if (req->flags & UNCACHEABLE) { + if (req->isUncacheable()) { // Don't update result register (see stq_c in isa_desc) req->result = 2; xc->setStCondFailures(0);//Needed? [RGD] @@ -391,7 +391,7 @@ InorderBackEnd<Impl>::read(MemReqPtr &req, T &data, int load_idx) } /* - if (!dcacheInterface && (req->flags & UNCACHEABLE)) + if (!dcacheInterface && (req->isUncacheable())) recordEvent("Uncached Read"); */ return NoFault; @@ -455,8 +455,8 @@ InorderBackEnd<Impl>::write(MemReqPtr &req, T &data, int store_idx) } } /* - if (req->flags & LOCKED) { - if (req->flags & UNCACHEABLE) { + if (req->isLocked()) { + if (req->isUncacheable()) { // Don't update result register (see stq_c in isa_desc) req->result = 2; } else { @@ -469,7 +469,7 @@ InorderBackEnd<Impl>::write(MemReqPtr &req, T &data, int store_idx) *res = req->result; */ /* - if (!dcacheInterface && (req->flags & UNCACHEABLE)) + if (!dcacheInterface && (req->isUncacheable())) recordEvent("Uncached Write"); */ return NoFault; diff --git a/src/cpu/ozone/lsq_unit.hh b/src/cpu/ozone/lsq_unit.hh index 38c1c09a2..056c79521 100644 --- a/src/cpu/ozone/lsq_unit.hh +++ b/src/cpu/ozone/lsq_unit.hh @@ -426,7 +426,7 @@ OzoneLSQ<Impl>::read(MemReqPtr &req, T &data, int load_idx) // at the head of the LSQ and are ready to commit (at the head of the ROB // too). // @todo: Fix uncached accesses. - if (req->flags & UNCACHEABLE && + if (req->isUncacheable() && (load_idx != loadHead || !loadQueue[load_idx]->readyToCommit())) { return TheISA::genMachineCheckFault(); diff --git a/src/cpu/ozone/lsq_unit_impl.hh b/src/cpu/ozone/lsq_unit_impl.hh index ee0804036..c46eb90be 100644 --- a/src/cpu/ozone/lsq_unit_impl.hh +++ b/src/cpu/ozone/lsq_unit_impl.hh @@ -577,7 +577,7 @@ OzoneLSQ<Impl>::writebackStores() MemAccessResult result = dcacheInterface->access(req); //@todo temp fix for LL/SC (works fine for 1 CPU) - if (req->flags & LOCKED) { + if (req->isLocked()) { req->result=1; panic("LL/SC! oh no no support!!!"); } @@ -596,7 +596,7 @@ OzoneLSQ<Impl>::writebackStores() Event *wb = NULL; /* typename IEW::LdWritebackEvent *wb = NULL; - if (req->flags & LOCKED) { + if (req->isLocked()) { // Stx_C does not generate a system port transaction. req->result=0; wb = new typename IEW::LdWritebackEvent(storeQueue[storeWBIdx].inst, @@ -630,7 +630,7 @@ OzoneLSQ<Impl>::writebackStores() // DPRINTF(Activity, "Active st accessing mem hit [sn:%lli]\n", // storeQueue[storeWBIdx].inst->seqNum); - if (req->flags & LOCKED) { + if (req->isLocked()) { // Stx_C does not generate a system port transaction. req->result=1; typename BackEnd::LdWritebackEvent *wb = diff --git a/src/cpu/ozone/lw_lsq.hh b/src/cpu/ozone/lw_lsq.hh index 6640a9f34..347f4569b 100644 --- a/src/cpu/ozone/lw_lsq.hh +++ b/src/cpu/ozone/lw_lsq.hh @@ -507,7 +507,7 @@ OzoneLWLSQ<Impl>::read(RequestPtr req, T &data, int load_idx) // at the head of the LSQ and are ready to commit (at the head of the ROB // too). // @todo: Fix uncached accesses. - if (req->getFlags() & UNCACHEABLE && + if (req->isUncacheable() && (inst != loadQueue.back() || !inst->isAtCommit())) { DPRINTF(OzoneLSQ, "[sn:%lli] Uncached load and not head of " "commit/LSQ!\n", @@ -659,7 +659,7 @@ OzoneLWLSQ<Impl>::read(RequestPtr req, T &data, int load_idx) return NoFault; } - if (req->getFlags() & LOCKED) { + if (req->isLocked()) { cpu->lockFlag = true; } diff --git a/src/cpu/ozone/lw_lsq_impl.hh b/src/cpu/ozone/lw_lsq_impl.hh index 4c96ad149..9d17b027f 100644 --- a/src/cpu/ozone/lw_lsq_impl.hh +++ b/src/cpu/ozone/lw_lsq_impl.hh @@ -394,7 +394,7 @@ OzoneLWLSQ<Impl>::executeLoad(DynInstPtr &inst) // Actually probably want the oldest faulting load if (load_fault != NoFault) { DPRINTF(OzoneLSQ, "Load [sn:%lli] has a fault\n", inst->seqNum); - if (!(inst->req->getFlags() & UNCACHEABLE && !inst->isAtCommit())) { + if (!(inst->req->isUncacheable() && !inst->isAtCommit())) { inst->setExecuted(); } // Maybe just set it as can commit here, although that might cause @@ -605,8 +605,8 @@ OzoneLWLSQ<Impl>::writebackStores() inst->seqNum); // @todo: Remove this SC hack once the memory system handles it. - if (req->getFlags() & LOCKED) { - if (req->getFlags() & UNCACHEABLE) { + if (req->isLocked()) { + if (req->isUncacheable()) { req->setScResult(2); } else { if (cpu->lockFlag) { @@ -663,7 +663,7 @@ OzoneLWLSQ<Impl>::writebackStores() if (result != MA_HIT && dcacheInterface->doEvents()) { store_event->miss = true; typename BackEnd::LdWritebackEvent *wb = NULL; - if (req->flags & LOCKED) { + if (req->isLocked()) { wb = new typename BackEnd::LdWritebackEvent(inst, be); store_event->wbEvent = wb; @@ -690,7 +690,7 @@ OzoneLWLSQ<Impl>::writebackStores() // DPRINTF(Activity, "Active st accessing mem hit [sn:%lli]\n", // inst->seqNum); - if (req->flags & LOCKED) { + if (req->isLocked()) { // Stx_C does not generate a system port // transaction in the 21264, but that might be // hard to accomplish in this model. diff --git a/src/cpu/simple/atomic.cc b/src/cpu/simple/atomic.cc index 88698bfee..42b0e9783 100644 --- a/src/cpu/simple/atomic.cc +++ b/src/cpu/simple/atomic.cc @@ -28,6 +28,7 @@ * Authors: Steve Reinhardt */ +#include "arch/locked_mem.hh" #include "arch/utility.hh" #include "cpu/exetrace.hh" #include "cpu/simple/atomic.hh" @@ -133,20 +134,19 @@ AtomicSimpleCPU::AtomicSimpleCPU(Params *p) { _status = Idle; - // @todo fix me and get the real cpu id & thread number!!! ifetch_req = new Request(); - ifetch_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE + ifetch_req->setThreadContext(p->cpu_id, 0); // Add thread ID if we add MT ifetch_pkt = new Packet(ifetch_req, Packet::ReadReq, Packet::Broadcast); ifetch_pkt->dataStatic(&inst); data_read_req = new Request(); - data_read_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE + data_read_req->setThreadContext(p->cpu_id, 0); // Add thread ID here too data_read_pkt = new Packet(data_read_req, Packet::ReadReq, Packet::Broadcast); data_read_pkt->dataStatic(&dataReg); data_write_req = new Request(); - data_write_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE + data_write_req->setThreadContext(p->cpu_id, 0); // Add thread ID here too data_write_pkt = new Packet(data_write_req, Packet::WriteReq, Packet::Broadcast); } @@ -253,29 +253,36 @@ template <class T> Fault AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags) { - data_read_req->setVirt(0, addr, sizeof(T), flags, thread->readPC()); + // use the CPU's statically allocated read request and packet objects + Request *req = data_read_req; + Packet *pkt = data_read_pkt; + + req->setVirt(0, addr, sizeof(T), flags, thread->readPC()); if (traceData) { traceData->setAddr(addr); } // translate to physical address - Fault fault = thread->translateDataReadReq(data_read_req); + Fault fault = thread->translateDataReadReq(req); // Now do the access. if (fault == NoFault) { - data_read_pkt->reinitFromRequest(); + pkt->reinitFromRequest(); - dcache_latency = dcachePort.sendAtomic(data_read_pkt); + dcache_latency = dcachePort.sendAtomic(pkt); dcache_access = true; - assert(data_read_pkt->result == Packet::Success); - data = data_read_pkt->get<T>(); + assert(pkt->result == Packet::Success); + data = pkt->get<T>(); + if (req->isLocked()) { + TheISA::handleLockedRead(thread, req); + } } // This will need a new way to tell if it has a dcache attached. - if (data_read_req->getFlags() & UNCACHEABLE) + if (req->isUncacheable()) recordEvent("Uncached Read"); return fault; @@ -328,33 +335,52 @@ template <class T> Fault AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) { - data_write_req->setVirt(0, addr, sizeof(T), flags, thread->readPC()); + // use the CPU's statically allocated write request and packet objects + Request *req = data_write_req; + Packet *pkt = data_write_pkt; + + req->setVirt(0, addr, sizeof(T), flags, thread->readPC()); if (traceData) { traceData->setAddr(addr); } // translate to physical address - Fault fault = thread->translateDataWriteReq(data_write_req); + Fault fault = thread->translateDataWriteReq(req); // Now do the access. if (fault == NoFault) { - data = htog(data); - data_write_pkt->reinitFromRequest(); - data_write_pkt->dataStatic(&data); + bool do_access = true; // flag to suppress cache access - dcache_latency = dcachePort.sendAtomic(data_write_pkt); - dcache_access = true; + if (req->isLocked()) { + do_access = TheISA::handleLockedWrite(thread, req); + } + + if (do_access) { + data = htog(data); + pkt->reinitFromRequest(); + pkt->dataStatic(&data); - assert(data_write_pkt->result == Packet::Success); + dcache_latency = dcachePort.sendAtomic(pkt); + dcache_access = true; + + assert(pkt->result == Packet::Success); + } - if (res && data_write_req->getFlags() & LOCKED) { - *res = data_write_req->getScResult(); + if (req->isLocked()) { + uint64_t scResult = req->getScResult(); + if (scResult != 0) { + // clear failure counter + thread->setStCondFailures(0); + } + if (res) { + *res = req->getScResult(); + } } } // This will need a new way to tell if it's hooked up to a cache or not. - if (data_write_req->getFlags() & UNCACHEABLE) + if (req->isUncacheable()) recordEvent("Uncached Write"); // If the write needs to have a fault on the access, consider calling @@ -467,11 +493,11 @@ BEGIN_DECLARE_SIM_OBJECT_PARAMS(AtomicSimpleCPU) Param<Tick> progress_interval; SimObjectParam<MemObject *> mem; SimObjectParam<System *> system; + Param<int> cpu_id; #if FULL_SYSTEM SimObjectParam<AlphaITB *> itb; SimObjectParam<AlphaDTB *> dtb; - Param<int> cpu_id; Param<Tick> profile; #else SimObjectParam<Process *> workload; @@ -500,11 +526,11 @@ BEGIN_INIT_SIM_OBJECT_PARAMS(AtomicSimpleCPU) INIT_PARAM(progress_interval, "Progress interval"), INIT_PARAM(mem, "memory"), INIT_PARAM(system, "system object"), + INIT_PARAM(cpu_id, "processor ID"), #if FULL_SYSTEM INIT_PARAM(itb, "Instruction TLB"), INIT_PARAM(dtb, "Data TLB"), - INIT_PARAM(cpu_id, "processor ID"), INIT_PARAM(profile, ""), #else INIT_PARAM(workload, "processes to run"), @@ -538,11 +564,11 @@ CREATE_SIM_OBJECT(AtomicSimpleCPU) params->simulate_stalls = simulate_stalls; params->mem = mem; params->system = system; + params->cpu_id = cpu_id; #if FULL_SYSTEM params->itb = itb; params->dtb = dtb; - params->cpu_id = cpu_id; params->profile = profile; #else params->process = workload; diff --git a/src/cpu/simple/timing.cc b/src/cpu/simple/timing.cc index 03ee27e04..a394468b9 100644 --- a/src/cpu/simple/timing.cc +++ b/src/cpu/simple/timing.cc @@ -28,6 +28,7 @@ * Authors: Steve Reinhardt */ +#include "arch/locked_mem.hh" #include "arch/utility.hh" #include "cpu/exetrace.hh" #include "cpu/simple/timing.hh" @@ -94,7 +95,8 @@ TimingSimpleCPU::CpuPort::TickEvent::schedule(Packet *_pkt, Tick t) } TimingSimpleCPU::TimingSimpleCPU(Params *p) - : BaseSimpleCPU(p), icachePort(this, p->clock), dcachePort(this, p->clock) + : BaseSimpleCPU(p), icachePort(this, p->clock), dcachePort(this, p->clock), + cpu_id(p->cpu_id) { _status = Idle; ifetch_pkt = dcache_pkt = NULL; @@ -227,35 +229,35 @@ template <class T> Fault TimingSimpleCPU::read(Addr addr, T &data, unsigned flags) { - // need to fill in CPU & thread IDs here - Request *data_read_req = new Request(); - data_read_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE - data_read_req->setVirt(0, addr, sizeof(T), flags, thread->readPC()); + Request *req = + new Request(/* asid */ 0, addr, sizeof(T), flags, thread->readPC(), + cpu_id, /* thread ID */ 0); if (traceData) { - traceData->setAddr(data_read_req->getVaddr()); + traceData->setAddr(req->getVaddr()); } // translate to physical address - Fault fault = thread->translateDataReadReq(data_read_req); + Fault fault = thread->translateDataReadReq(req); // Now do the access. if (fault == NoFault) { - Packet *data_read_pkt = - new Packet(data_read_req, Packet::ReadReq, Packet::Broadcast); - data_read_pkt->dataDynamic<T>(new T); + Packet *pkt = + new Packet(req, Packet::ReadReq, Packet::Broadcast); + pkt->dataDynamic<T>(new T); - if (!dcachePort.sendTiming(data_read_pkt)) { + if (!dcachePort.sendTiming(pkt)) { _status = DcacheRetry; - dcache_pkt = data_read_pkt; + dcache_pkt = pkt; } else { _status = DcacheWaitResponse; + // memory system takes ownership of packet dcache_pkt = NULL; } } // This will need a new way to tell if it has a dcache attached. - if (data_read_req->getFlags() & UNCACHEABLE) + if (req->isUncacheable()) recordEvent("Uncached Read"); return fault; @@ -308,31 +310,39 @@ template <class T> Fault TimingSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) { - // need to fill in CPU & thread IDs here - Request *data_write_req = new Request(); - data_write_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE - data_write_req->setVirt(0, addr, sizeof(T), flags, thread->readPC()); + Request *req = + new Request(/* asid */ 0, addr, sizeof(T), flags, thread->readPC(), + cpu_id, /* thread ID */ 0); // translate to physical address - Fault fault = thread->translateDataWriteReq(data_write_req); + Fault fault = thread->translateDataWriteReq(req); + // Now do the access. if (fault == NoFault) { - Packet *data_write_pkt = - new Packet(data_write_req, Packet::WriteReq, Packet::Broadcast); - data_write_pkt->allocate(); - data_write_pkt->set(data); + assert(dcache_pkt == NULL); + dcache_pkt = new Packet(req, Packet::WriteReq, Packet::Broadcast); + dcache_pkt->allocate(); + dcache_pkt->set(data); - if (!dcachePort.sendTiming(data_write_pkt)) { - _status = DcacheRetry; - dcache_pkt = data_write_pkt; - } else { - _status = DcacheWaitResponse; - dcache_pkt = NULL; + bool do_access = true; // flag to suppress cache access + + if (req->isLocked()) { + do_access = TheISA::handleLockedWrite(thread, req); + } + + if (do_access) { + if (!dcachePort.sendTiming(dcache_pkt)) { + _status = DcacheRetry; + } else { + _status = DcacheWaitResponse; + // memory system takes ownership of packet + dcache_pkt = NULL; + } } } // This will need a new way to tell if it's hooked up to a cache or not. - if (data_write_req->getFlags() & UNCACHEABLE) + if (req->isUncacheable()) recordEvent("Uncached Write"); // If the write needs to have a fault on the access, consider calling @@ -392,9 +402,8 @@ TimingSimpleCPU::fetch() { checkForInterrupts(); - // need to fill in CPU & thread IDs here Request *ifetch_req = new Request(); - ifetch_req->setThreadContext(0,0); //Need CPU/Thread IDS HERE + ifetch_req->setThreadContext(cpu_id, /* thread ID */ 0); Fault fault = setupFetchRequest(ifetch_req); ifetch_pkt = new Packet(ifetch_req, Packet::ReadReq, Packet::Broadcast); @@ -453,12 +462,20 @@ TimingSimpleCPU::completeIfetch(Packet *pkt) if (curStaticInst->isMemRef() && !curStaticInst->isDataPrefetch()) { // load or store: just send to dcache Fault fault = curStaticInst->initiateAcc(this, traceData); - if (fault == NoFault) { - // successfully initiated access: instruction will - // complete in dcache response callback - assert(_status == DcacheWaitResponse); + if (_status != Running) { + // instruction will complete in dcache response callback + assert(_status == DcacheWaitResponse || _status == DcacheRetry); + assert(fault == NoFault); } else { - // fault: complete now to invoke fault handler + if (fault == NoFault) { + // early fail on store conditional: complete now + assert(dcache_pkt != NULL); + fault = curStaticInst->completeAcc(dcache_pkt, this, + traceData); + delete dcache_pkt->req; + delete dcache_pkt; + dcache_pkt = NULL; + } postExecute(); advanceInst(fault); } @@ -479,8 +496,7 @@ TimingSimpleCPU::IcachePort::ITickEvent::process() bool TimingSimpleCPU::IcachePort::recvTiming(Packet *pkt) { - // These next few lines could be replaced with something faster - // who knows what though + // delay processing of returned data until next CPU clock edge Tick time = pkt->req->getTime(); while (time < curTick) time += lat; @@ -527,6 +543,10 @@ TimingSimpleCPU::completeDataAccess(Packet *pkt) Fault fault = curStaticInst->completeAcc(pkt, this, traceData); + if (pkt->isRead() && pkt->req->isLocked()) { + TheISA::handleLockedRead(thread, pkt->req); + } + delete pkt->req; delete pkt; @@ -546,6 +566,7 @@ TimingSimpleCPU::completeDrain() bool TimingSimpleCPU::DcachePort::recvTiming(Packet *pkt) { + // delay processing of returned data until next CPU clock edge Tick time = pkt->req->getTime(); while (time < curTick) time += lat; @@ -574,6 +595,7 @@ TimingSimpleCPU::DcachePort::recvRetry() Packet *tmp = cpu->dcache_pkt; if (sendTiming(tmp)) { cpu->_status = DcacheWaitResponse; + // memory system takes ownership of packet cpu->dcache_pkt = NULL; } } @@ -592,11 +614,11 @@ BEGIN_DECLARE_SIM_OBJECT_PARAMS(TimingSimpleCPU) Param<Tick> progress_interval; SimObjectParam<MemObject *> mem; SimObjectParam<System *> system; + Param<int> cpu_id; #if FULL_SYSTEM SimObjectParam<AlphaITB *> itb; SimObjectParam<AlphaDTB *> dtb; - Param<int> cpu_id; Param<Tick> profile; #else SimObjectParam<Process *> workload; @@ -625,11 +647,11 @@ BEGIN_INIT_SIM_OBJECT_PARAMS(TimingSimpleCPU) INIT_PARAM(progress_interval, "Progress interval"), INIT_PARAM(mem, "memory"), INIT_PARAM(system, "system object"), + INIT_PARAM(cpu_id, "processor ID"), #if FULL_SYSTEM INIT_PARAM(itb, "Instruction TLB"), INIT_PARAM(dtb, "Data TLB"), - INIT_PARAM(cpu_id, "processor ID"), INIT_PARAM(profile, ""), #else INIT_PARAM(workload, "processes to run"), @@ -661,11 +683,11 @@ CREATE_SIM_OBJECT(TimingSimpleCPU) params->functionTraceStart = function_trace_start; params->mem = mem; params->system = system; + params->cpu_id = cpu_id; #if FULL_SYSTEM params->itb = itb; params->dtb = dtb; - params->cpu_id = cpu_id; params->profile = profile; #else params->process = workload; diff --git a/src/cpu/simple/timing.hh b/src/cpu/simple/timing.hh index d03fa4bc0..b65eebd99 100644 --- a/src/cpu/simple/timing.hh +++ b/src/cpu/simple/timing.hh @@ -166,6 +166,8 @@ class TimingSimpleCPU : public BaseSimpleCPU Packet *ifetch_pkt; Packet *dcache_pkt; + int cpu_id; + public: virtual Port *getPort(const std::string &if_name, int idx = -1); diff --git a/src/cpu/simple_thread.hh b/src/cpu/simple_thread.hh index 242cfd0e1..6fa6500bd 100644 --- a/src/cpu/simple_thread.hh +++ b/src/cpu/simple_thread.hh @@ -237,7 +237,7 @@ class SimpleThread : public ThreadState Fault read(RequestPtr &req, T &data) { #if FULL_SYSTEM && THE_ISA == ALPHA_ISA - if (req->flags & LOCKED) { + if (req->isLocked()) { req->xc->setMiscReg(TheISA::Lock_Addr_DepTag, req->paddr); req->xc->setMiscReg(TheISA::Lock_Flag_DepTag, true); } @@ -256,10 +256,10 @@ class SimpleThread : public ThreadState ExecContext *xc; // If this is a store conditional, act appropriately - if (req->flags & LOCKED) { + if (req->isLocked()) { xc = req->xc; - if (req->flags & UNCACHEABLE) { + if (req->isUncacheable()) { // Don't update result register (see stq_c in isa_desc) req->result = 2; xc->setStCondFailures(0);//Needed? [RGD] |