diff options
Diffstat (limited to 'src/cpu/ozone')
-rw-r--r-- | src/cpu/ozone/cpu.cc | 9 | ||||
-rw-r--r-- | src/cpu/ozone/cpu.hh | 713 | ||||
-rw-r--r-- | src/cpu/ozone/cpu_impl.hh | 1029 |
3 files changed, 1372 insertions, 379 deletions
diff --git a/src/cpu/ozone/cpu.cc b/src/cpu/ozone/cpu.cc index cbeca9d3b..d2ea0164c 100644 --- a/src/cpu/ozone/cpu.cc +++ b/src/cpu/ozone/cpu.cc @@ -26,8 +26,9 @@ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ -#include "cpu/ooo_cpu/ooo_cpu_impl.hh" -#include "cpu/ooo_cpu/ooo_dyn_inst.hh" -#include "cpu/ooo_cpu/ooo_impl.hh" +#include "cpu/ozone/cpu_impl.hh" +#include "cpu/ozone/ozone_impl.hh" +#include "cpu/ozone/simple_impl.hh" -template class OoOCPU<OoOImpl>; +template class OzoneCPU<SimpleImpl>; +template class OzoneCPU<OzoneImpl>; diff --git a/src/cpu/ozone/cpu.hh b/src/cpu/ozone/cpu.hh index fa849bb09..5af2b02b2 100644 --- a/src/cpu/ozone/cpu.hh +++ b/src/cpu/ozone/cpu.hh @@ -26,15 +26,19 @@ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ -#ifndef __CPU_OOO_CPU_OOO_CPU_HH__ -#define __CPU_OOO_CPU_OOO_CPU_HH__ +#ifndef __CPU_OZONE_CPU_HH__ +#define __CPU_OZONE_CPU_HH__ + +#include <set> #include "base/statistics.hh" +#include "base/timebuf.hh" #include "config/full_system.hh" #include "cpu/base.hh" #include "cpu/exec_context.hh" -#include "encumbered/cpu/full/fu_pool.hh" -#include "cpu/ooo_cpu/ea_list.hh" +#include "cpu/inst_seq.hh" +#include "cpu/ozone/rename_table.hh" +#include "cpu/ozone/thread_state.hh" #include "cpu/pc_event.hh" #include "cpu/static_inst.hh" #include "mem/mem_interface.hh" @@ -42,14 +46,21 @@ // forward declarations #if FULL_SYSTEM -class Processor; +#include "arch/alpha/tlb.hh" + class AlphaITB; class AlphaDTB; class PhysicalMemory; +class MemoryController; +class Sampler; class RemoteGDB; class GDBListener; +namespace Kernel { + class Statistics; +}; + #else class Process; @@ -57,12 +68,16 @@ class Process; #endif // FULL_SYSTEM class Checkpoint; +class EndQuiesceEvent; class MemInterface; namespace Trace { class InstRecord; } +template <class> +class Checker; + /** * Declaration of Out-of-Order CPU class. Basically it is a SimpleCPU with * simple out-of-order capabilities added to it. It is still a 1 CPI machine @@ -72,23 +87,177 @@ namespace Trace { */ template <class Impl> -class OoOCPU : public BaseCPU +class OzoneCPU : public BaseCPU { private: + typedef typename Impl::FrontEnd FrontEnd; + typedef typename Impl::BackEnd BackEnd; typedef typename Impl::DynInst DynInst; typedef typename Impl::DynInstPtr DynInstPtr; + typedef TheISA::MiscReg MiscReg; + + public: + class OzoneXC : public ExecContext { + public: + OzoneCPU<Impl> *cpu; + + OzoneThreadState<Impl> *thread; + + BaseCPU *getCpuPtr(); + + void setCpuId(int id); + + int readCpuId() { return thread->cpuId; } + + FunctionalMemory *getMemPtr() { return thread->mem; } + +#if FULL_SYSTEM + System *getSystemPtr() { return cpu->system; } + + PhysicalMemory *getPhysMemPtr() { return cpu->physmem; } + + AlphaITB *getITBPtr() { return cpu->itb; } + + AlphaDTB * getDTBPtr() { return cpu->dtb; } + + Kernel::Statistics *getKernelStats() { return thread->kernelStats; } +#else + Process *getProcessPtr() { return thread->process; } +#endif + + Status status() const { return thread->_status; } + + void setStatus(Status new_status); + + /// Set the status to Active. Optional delay indicates number of + /// cycles to wait before beginning execution. + void activate(int delay = 1); + + /// Set the status to Suspended. + void suspend(); + + /// Set the status to Unallocated. + void deallocate(); + + /// Set the status to Halted. + void halt(); + +#if FULL_SYSTEM + void dumpFuncProfile(); +#endif + + void takeOverFrom(ExecContext *old_context); + + void regStats(const std::string &name); + + void serialize(std::ostream &os); + void unserialize(Checkpoint *cp, const std::string §ion); + +#if FULL_SYSTEM + EndQuiesceEvent *getQuiesceEvent(); + + Tick readLastActivate(); + Tick readLastSuspend(); + + void profileClear(); + void profileSample(); +#endif + + int getThreadNum(); + + // Also somewhat obnoxious. Really only used for the TLB fault. + TheISA::MachInst getInst(); + + void copyArchRegs(ExecContext *xc); + + void clearArchRegs(); + + uint64_t readIntReg(int reg_idx); + + float readFloatRegSingle(int reg_idx); + + double readFloatRegDouble(int reg_idx); + + uint64_t readFloatRegInt(int reg_idx); + + void setIntReg(int reg_idx, uint64_t val); + + void setFloatRegSingle(int reg_idx, float val); + + void setFloatRegDouble(int reg_idx, double val); + + void setFloatRegInt(int reg_idx, uint64_t val); + + uint64_t readPC() { return thread->PC; } + void setPC(Addr val); + + uint64_t readNextPC() { return thread->nextPC; } + void setNextPC(Addr val); + + public: + // ISA stuff: + MiscReg readMiscReg(int misc_reg); + + MiscReg readMiscRegWithEffect(int misc_reg, Fault &fault); + + Fault setMiscReg(int misc_reg, const MiscReg &val); + + Fault setMiscRegWithEffect(int misc_reg, const MiscReg &val); + + unsigned readStCondFailures() + { return thread->storeCondFailures; } + + void setStCondFailures(unsigned sc_failures) + { thread->storeCondFailures = sc_failures; } + +#if FULL_SYSTEM + bool inPalMode() { return cpu->inPalMode(); } +#endif + + bool misspeculating() { return false; } + +#if !FULL_SYSTEM + TheISA::IntReg getSyscallArg(int i) + { return thread->renameTable[TheISA::ArgumentReg0 + i]->readIntResult(); } + + // used to shift args for indirect syscall + void setSyscallArg(int i, TheISA::IntReg val) + { thread->renameTable[TheISA::ArgumentReg0 + i]->setIntResult(i); } + + void setSyscallReturn(SyscallReturn return_value) + { cpu->setSyscallReturn(return_value, thread->tid); } + + Counter readFuncExeInst() { return thread->funcExeInst; } + + void setFuncExeInst(Counter new_val) + { thread->funcExeInst = new_val; } +#endif + }; + + // execution context proxy + OzoneXC ozoneXC; + ExecContext *xcProxy; + ExecContext *checkerXC; + + typedef OzoneThreadState<Impl> ImplState; + + private: + OzoneThreadState<Impl> thread; + public: // main simulation loop (one cycle) void tick(); + std::set<InstSeqNum> snList; + std::set<Addr> lockAddrList; private: struct TickEvent : public Event { - OoOCPU *cpu; + OzoneCPU *cpu; int width; - TickEvent(OoOCPU *c, int w); + TickEvent(OzoneCPU *c, int w); void process(); const char *description(); }; @@ -99,9 +268,9 @@ class OoOCPU : public BaseCPU void scheduleTickEvent(int delay) { if (tickEvent.squashed()) - tickEvent.reschedule(curTick + delay); + tickEvent.reschedule(curTick + cycles(delay)); else if (!tickEvent.scheduled()) - tickEvent.schedule(curTick + delay); + tickEvent.schedule(curTick + cycles(delay)); } /// Unschedule tick event, regardless of its current state. @@ -118,20 +287,17 @@ class OoOCPU : public BaseCPU void trace_data(T data); public: - // enum Status { Running, Idle, - IcacheMiss, - IcacheMissComplete, - DcacheMissStall, SwitchedOut }; - private: Status _status; public: + bool checkInterrupts; + void post_interrupt(int int_num, int index); void zero_fill_64(Addr addr) { @@ -142,41 +308,45 @@ class OoOCPU : public BaseCPU } }; - struct Params : public BaseCPU::Params - { - MemInterface *icache_interface; - MemInterface *dcache_interface; - int width; -#if FULL_SYSTEM - AlphaITB *itb; - AlphaDTB *dtb; - FunctionalMemory *mem; -#else - Process *process; -#endif - int issueWidth; - }; + typedef typename Impl::Params Params; - OoOCPU(Params *params); + OzoneCPU(Params *params); - virtual ~OoOCPU(); + virtual ~OzoneCPU(); void init(); - private: - void copyFromXC(); - public: - // execution context - ExecContext *xc; + BaseCPU *getCpuPtr() { return this; } + + void setCpuId(int id) { cpuId = id; } + + int readCpuId() { return cpuId; } + + int cpuId; - void switchOut(); + void switchOut(Sampler *sampler); + void signalSwitched(); void takeOverFrom(BaseCPU *oldCPU); + Sampler *sampler; + + int switchCount; + #if FULL_SYSTEM Addr dbg_vtophys(Addr addr); bool interval_stats; + + AlphaITB *itb; + AlphaDTB *dtb; + System *system; + + // the following two fields are redundant, since we can always + // look them up through the system pointer, but we'll leave them + // here for now for convenience + MemoryController *memctrl; + PhysicalMemory *physmem; #endif // L1 instruction cache @@ -185,54 +355,15 @@ class OoOCPU : public BaseCPU // L1 data cache MemInterface *dcacheInterface; - FuncUnitPool *fuPool; + /** Pointer to memory. */ + FunctionalMemory *mem; - // Refcounted pointer to the one memory request. - MemReqPtr cacheMemReq; - - class ICacheCompletionEvent : public Event - { - private: - OoOCPU *cpu; - - public: - ICacheCompletionEvent(OoOCPU *_cpu); - - virtual void process(); - virtual const char *description(); - }; - - // Will need to create a cache completion event upon any memory miss. - ICacheCompletionEvent iCacheCompletionEvent; - - class DCacheCompletionEvent; - - typedef typename - std::list<DCacheCompletionEvent>::iterator DCacheCompEventIt; - - class DCacheCompletionEvent : public Event - { - private: - OoOCPU *cpu; - DynInstPtr inst; - DCacheCompEventIt dcceIt; - - public: - DCacheCompletionEvent(OoOCPU *_cpu, DynInstPtr &_inst, - DCacheCompEventIt &_dcceIt); - - virtual void process(); - virtual const char *description(); - }; - - friend class DCacheCompletionEvent; - - protected: - std::list<DCacheCompletionEvent> dCacheCompList; - DCacheCompEventIt dcceIt; + FrontEnd *frontEnd; + BackEnd *backEnd; private: Status status() const { return _status; } + void setStatus(Status new_status) { _status = new_status; } virtual void activateContext(int thread_num, int delay); virtual void suspendContext(int thread_num); @@ -244,18 +375,16 @@ class OoOCPU : public BaseCPU virtual void resetStats(); // number of simulated instructions + public: Counter numInst; Counter startNumInst; - Stats::Scalar<> numInsts; virtual Counter totalInstructions() const { return numInst - startNumInst; } - // number of simulated memory references - Stats::Scalar<> numMemRefs; - + private: // number of simulated loads Counter numLoad; Counter startNumLoad; @@ -263,27 +392,15 @@ class OoOCPU : public BaseCPU // number of idle cycles Stats::Average<> notIdleFraction; Stats::Formula idleFraction; - - // number of cycles stalled for I-cache misses - Stats::Scalar<> icacheStallCycles; - Counter lastIcacheStall; - - // number of cycles stalled for D-cache misses - Stats::Scalar<> dcacheStallCycles; - Counter lastDcacheStall; - - void processICacheCompletion(); - public: virtual void serialize(std::ostream &os); virtual void unserialize(Checkpoint *cp, const std::string §ion); + #if FULL_SYSTEM bool validInstAddr(Addr addr) { return true; } bool validDataAddr(Addr addr) { return true; } - int getInstAsid() { return xc->regs.instAsid(); } - int getDataAsid() { return xc->regs.dataAsid(); } Fault translateInstReq(MemReqPtr &req) { @@ -302,13 +419,13 @@ class OoOCPU : public BaseCPU #else bool validInstAddr(Addr addr) - { return xc->validInstAddr(addr); } + { return true; } bool validDataAddr(Addr addr) - { return xc->validDataAddr(addr); } + { return true; } - int getInstAsid() { return xc->asid; } - int getDataAsid() { return xc->asid; } + int getInstAsid() { return thread.asid; } + int getDataAsid() { return thread.asid; } Fault dummyTranslation(MemReqPtr &req) { @@ -321,318 +438,192 @@ class OoOCPU : public BaseCPU req->paddr = req->paddr | (Addr)req->asid << sizeof(Addr) * 8 - 16; return NoFault; } + + /** Translates instruction requestion in syscall emulation mode. */ Fault translateInstReq(MemReqPtr &req) { return dummyTranslation(req); } + + /** Translates data read request in syscall emulation mode. */ Fault translateDataReadReq(MemReqPtr &req) { return dummyTranslation(req); } + + /** Translates data write request in syscall emulation mode. */ Fault translateDataWriteReq(MemReqPtr &req) { return dummyTranslation(req); } - #endif + /** Old CPU read from memory function. No longer used. */ template <class T> - Fault read(Addr addr, T &data, unsigned flags, DynInstPtr inst); - - template <class T> - Fault write(T data, Addr addr, unsigned flags, - uint64_t *res, DynInstPtr inst); - - void prefetch(Addr addr, unsigned flags) + Fault read(MemReqPtr &req, T &data) { - // need to do this... - } - - void writeHint(Addr addr, int size, unsigned flags) - { - // need to do this... - } - - Fault copySrcTranslate(Addr src); - - Fault copy(Addr dest); - - private: - bool executeInst(DynInstPtr &inst); - - void renameInst(DynInstPtr &inst); - - void addInst(DynInstPtr &inst); - - void commitHeadInst(); - - bool getOneInst(); - - Fault fetchCacheLine(); - - InstSeqNum getAndIncrementInstSeq(); - - bool ambigMemAddr; - - private: - InstSeqNum globalSeqNum; - - DynInstPtr renameTable[TheISA::TotalNumRegs]; - DynInstPtr commitTable[TheISA::TotalNumRegs]; - - // Might need a table of the shadow registers as well. -#if FULL_SYSTEM - DynInstPtr palShadowTable[TheISA::NumIntRegs]; +#if 0 +#if FULL_SYSTEM && defined(TARGET_ALPHA) + if (req->flags & LOCKED) { + req->xc->setMiscReg(TheISA::Lock_Addr_DepTag, req->paddr); + req->xc->setMiscReg(TheISA::Lock_Flag_DepTag, true); + } +#endif #endif + Fault error; + if (req->flags & LOCKED) { + lockAddrList.insert(req->paddr); + lockFlag = true; + } - public: - // The register accessor methods provide the index of the - // instruction's operand (e.g., 0 or 1), not the architectural - // register index, to simplify the implementation of register - // renaming. We find the architectural register index by indexing - // into the instruction's own operand index table. Note that a - // raw pointer to the StaticInst is provided instead of a - // ref-counted StaticInstPtr to redice overhead. This is fine as - // long as these methods don't copy the pointer into any long-term - // storage (which is pretty hard to imagine they would have reason - // to do). - - // In the OoO case these shouldn't read from the XC but rather from the - // rename table of DynInsts. Also these likely shouldn't be called very - // often, other than when adding things into the xc during say a syscall. - - uint64_t readIntReg(StaticInst *si, int idx) - { - return xc->readIntReg(si->srcRegIdx(idx)); + error = this->mem->read(req, data); + data = gtoh(data); + return error; } - FloatReg readFloatReg(StaticInst *si, int idx, width) - { - int reg_idx = si->srcRegIdx(idx) - TheISA::FP_Base_DepTag; - return xc->readFloatReg(reg_idx, width); - } - FloatReg readFloatReg(StaticInst *si, int idx) + /** CPU read function, forwards read to LSQ. */ + template <class T> + Fault read(MemReqPtr &req, T &data, int load_idx) { - int reg_idx = si->srcRegIdx(idx) - TheISA::FP_Base_DepTag; - return xc->readFloatReg(reg_idx); + return backEnd->read(req, data, load_idx); } - FloatRegBits readFloatRegBits(StaticInst *si, int idx, int width) + /** Old CPU write to memory function. No longer used. */ + template <class T> + Fault write(MemReqPtr &req, T &data) { - int reg_idx = si->srcRegIdx(idx) - TheISA::FP_Base_DepTag; - return xc->readFloatRegBits(reg_idx, width); - } +#if 0 +#if FULL_SYSTEM && defined(TARGET_ALPHA) + ExecContext *xc; + + // If this is a store conditional, act appropriately + if (req->flags & LOCKED) { + xc = req->xc; + + if (req->flags & UNCACHEABLE) { + // Don't update result register (see stq_c in isa_desc) + req->result = 2; + xc->setStCondFailures(0);//Needed? [RGD] + } else { + bool lock_flag = xc->readMiscReg(TheISA::Lock_Flag_DepTag); + Addr lock_addr = xc->readMiscReg(TheISA::Lock_Addr_DepTag); + req->result = lock_flag; + if (!lock_flag || + ((lock_addr & ~0xf) != (req->paddr & ~0xf))) { + xc->setMiscReg(TheISA::Lock_Flag_DepTag, false); + xc->setStCondFailures(xc->readStCondFailures() + 1); + if (((xc->readStCondFailures()) % 100000) == 0) { + std::cerr << "Warning: " + << xc->readStCondFailures() + << " consecutive store conditional failures " + << "on cpu " << req->xc->readCpuId() + << std::endl; + } + return NoFault; + } + else xc->setStCondFailures(0); + } + } - FloatRegBits readFloatRegBits(StaticInst *si, int idx) - { - int reg_idx = si->srcRegIdx(idx) - TheISA::FP_Base_DepTag; - return xc->readFloatRegBits(reg_idx); - } + // Need to clear any locked flags on other proccessors for + // this address. Only do this for succsful Store Conditionals + // and all other stores (WH64?). Unsuccessful Store + // Conditionals would have returned above, and wouldn't fall + // through. + for (int i = 0; i < this->system->execContexts.size(); i++){ + xc = this->system->execContexts[i]; + if ((xc->readMiscReg(TheISA::Lock_Addr_DepTag) & ~0xf) == + (req->paddr & ~0xf)) { + xc->setMiscReg(TheISA::Lock_Flag_DepTag, false); + } + } - void setIntReg(StaticInst *si, int idx, uint64_t val) - { - xc->setIntReg(si->destRegIdx(idx), val); - } +#endif +#endif - void setFloatReg(StaticInst *si, int idx, FloatReg val, int width) - { - int reg_idx = si->destRegIdx(idx) - TheISA::FP_Base_DepTag; - xc->setFloatReg(reg_idx, val, width); - } + if (req->flags & LOCKED) { + if (req->flags & UNCACHEABLE) { + req->result = 2; + } else { + if (this->lockFlag) { + if (lockAddrList.find(req->paddr) != + lockAddrList.end()) { + req->result = 1; + } else { + req->result = 0; + return NoFault; + } + } else { + req->result = 0; + return NoFault; + } + } + } - void setFloatReg(StaticInst *si, int idx, FloatReg val) - { - int reg_idx = si->destRegIdx(idx) - TheISA::FP_Base_DepTag; - xc->setFloatReg(reg_idx, val); + return this->mem->write(req, (T)htog(data)); } - void setFloatRegBits(StaticInst *si, int idx, FloatRegBits val, int width) + /** CPU write function, forwards write to LSQ. */ + template <class T> + Fault write(MemReqPtr &req, T &data, int store_idx) { - int reg_idx = si->destRegIdx(idx) - TheISA::FP_Base_DepTag; - xc->setFloatRegBits(reg_idx, val, width); + return backEnd->write(req, data, store_idx); } - void setFloatRegBits(StaticInst *si, int idx, FloatRegBits val) + void prefetch(Addr addr, unsigned flags) { - int reg_idx = si->destRegIdx(idx) - TheISA::FP_Base_DepTag; - xc->setFloatRegBits(reg_idx, val); + // need to do this... } - uint64_t readPC() { return PC; } - void setNextPC(Addr val) { nextPC = val; } - - private: - Addr PC; - Addr nextPC; - - unsigned issueWidth; - - bool fetchRedirExcp; - bool fetchRedirBranch; - - /** Mask to get a cache block's address. */ - Addr cacheBlkMask; - - unsigned cacheBlkSize; - - Addr cacheBlkPC; - - /** The cache line being fetched. */ - uint8_t *cacheData; - - protected: - bool cacheBlkValid; - - private: - - // Align an address (typically a PC) to the start of an I-cache block. - // We fold in the PISA 64- to 32-bit conversion here as well. - Addr icacheBlockAlignPC(Addr addr) + void writeHint(Addr addr, int size, unsigned flags) { - addr = TheISA::realPCToFetchPC(addr); - return (addr & ~(cacheBlkMask)); + // need to do this... } - unsigned instSize; + Fault copySrcTranslate(Addr src); - // ROB tracking stuff. - DynInstPtr robHeadPtr; - DynInstPtr robTailPtr; - unsigned robSize; - unsigned robInsts; + Fault copy(Addr dest); - // List of outstanding EA instructions. - protected: - EAList eaList; + InstSeqNum globalSeqNum; public: - void branchToTarget(Addr val) - { - if (!fetchRedirExcp) { - fetchRedirBranch = true; - PC = val; - } - } - - // ISA stuff: - uint64_t readUniq() { return xc->readUniq(); } - void setUniq(uint64_t val) { xc->setUniq(val); } + void squashFromXC(); - uint64_t readFpcr() { return xc->readFpcr(); } - void setFpcr(uint64_t val) { xc->setFpcr(val); } + // @todo: This can be a useful debug function. Implement it. + void dumpInsts() { frontEnd->dumpInsts(); } #if FULL_SYSTEM - uint64_t readIpr(int idx, Fault &fault) { return xc->readIpr(idx, fault); } - Fault setIpr(int idx, uint64_t val) { return xc->setIpr(idx, val); } - Fault hwrei() { return xc->hwrei(); } - int readIntrFlag() { return xc->readIntrFlag(); } - void setIntrFlag(int val) { xc->setIntrFlag(val); } - bool inPalMode() { return xc->inPalMode(); } - void trap(Fault fault) { fault->invoke(xc); } - bool simPalCheck(int palFunc) { return xc->simPalCheck(palFunc); } + Fault hwrei(); + int readIntrFlag() { return thread.regs.intrflag; } + void setIntrFlag(int val) { thread.regs.intrflag = val; } + bool inPalMode() { return AlphaISA::PcPAL(thread.PC); } + bool inPalMode(Addr pc) { return AlphaISA::PcPAL(pc); } + bool simPalCheck(int palFunc); + void processInterrupts(); #else - void syscall() { xc->syscall(); } -#endif - - ExecContext *xcBase() { return xc; } -}; - - -// precise architected memory state accessor macros -template <class Impl> -template <class T> -Fault -OoOCPU<Impl>::read(Addr addr, T &data, unsigned flags, DynInstPtr inst) -{ - MemReqPtr readReq = new MemReq(); - readReq->xc = xc; - readReq->asid = 0; - readReq->data = new uint8_t[64]; - - readReq->reset(addr, sizeof(T), flags); - - // translate to physical address - This might be an ISA impl call - Fault fault = translateDataReadReq(readReq); - - // do functional access - if (fault == NoFault) - fault = xc->mem->read(readReq, data); -#if 0 - if (traceData) { - traceData->setAddr(addr); - if (fault == NoFault) - traceData->setData(data); - } + void syscall(); + void setSyscallReturn(SyscallReturn return_value, int tid); #endif - // if we have a cache, do cache access too - if (fault == NoFault && dcacheInterface) { - readReq->cmd = Read; - readReq->completionEvent = NULL; - readReq->time = curTick; - /*MemAccessResult result = */dcacheInterface->access(readReq); + ExecContext *xcBase() { return xcProxy; } - if (dcacheInterface->doEvents()) { - readReq->completionEvent = new DCacheCompletionEvent(this, inst, - dcceIt); - } - } - - if (!dcacheInterface && (readReq->flags & UNCACHEABLE)) - recordEvent("Uncached Read"); - - return fault; -} - -template <class Impl> -template <class T> -Fault -OoOCPU<Impl>::write(T data, Addr addr, unsigned flags, - uint64_t *res, DynInstPtr inst) -{ - MemReqPtr writeReq = new MemReq(); - writeReq->xc = xc; - writeReq->asid = 0; - writeReq->data = new uint8_t[64]; + bool decoupledFrontEnd; + struct CommStruct { + InstSeqNum doneSeqNum; + InstSeqNum nonSpecSeqNum; + bool uncached; + unsigned lqIdx; -#if 0 - if (traceData) { - traceData->setAddr(addr); - traceData->setData(data); - } -#endif - - writeReq->reset(addr, sizeof(T), flags); - - // translate to physical address - Fault fault = translateDataWriteReq(writeReq); - - // do functional access - if (fault == NoFault) - fault = xc->write(writeReq, data); - - if (fault == NoFault && dcacheInterface) { - writeReq->cmd = Write; - memcpy(writeReq->data,(uint8_t *)&data,writeReq->size); - writeReq->completionEvent = NULL; - writeReq->time = curTick; - /*MemAccessResult result = */dcacheInterface->access(writeReq); - - if (dcacheInterface->doEvents()) { - writeReq->completionEvent = new DCacheCompletionEvent(this, inst, - dcceIt); - } - } - - if (res && (fault == NoFault)) - *res = writeReq->result; + bool stall; + }; + TimeBuffer<CommStruct> comm; - if (!dcacheInterface && (writeReq->flags & UNCACHEABLE)) - recordEvent("Uncached Write"); + bool lockFlag; - return fault; -} + Stats::Scalar<> quiesceCycles; + Checker<DynInstPtr> *checker; +}; -#endif // __CPU_OOO_CPU_OOO_CPU_HH__ +#endif // __CPU_OZONE_CPU_HH__ diff --git a/src/cpu/ozone/cpu_impl.hh b/src/cpu/ozone/cpu_impl.hh index e7ed3cfe0..5675da3a8 100644 --- a/src/cpu/ozone/cpu_impl.hh +++ b/src/cpu/ozone/cpu_impl.hh @@ -1,5 +1,5 @@ /* - * Copyright (c) 2005 The Regents of The University of Michigan + * Copyright (c) 2006 The Regents of The University of Michigan * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -26,23 +26,1024 @@ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ -#ifndef __CPU_OOO_CPU_OOO_IMPL_HH__ -#define __CPU_OOO_CPU_OOO_IMPL_HH__ +//#include <cstdio> +//#include <cstdlib> -#include "arch/isa_traits.hh" +#include "arch/isa_traits.hh" // For MachInst +#include "base/trace.hh" +#include "config/full_system.hh" +#include "cpu/base.hh" +#include "cpu/checker/exec_context.hh" +#include "cpu/exec_context.hh" +#include "cpu/exetrace.hh" +#include "cpu/ozone/cpu.hh" +#include "cpu/quiesce_event.hh" +#include "cpu/static_inst.hh" +//#include "mem/base_mem.hh" +#include "mem/mem_interface.hh" +#include "sim/sim_object.hh" +#include "sim/stats.hh" + +#if FULL_SYSTEM +#include "arch/faults.hh" +#include "arch/alpha/osfpal.hh" +#include "arch/alpha/tlb.hh" +#include "arch/vtophys.hh" +#include "base/callback.hh" +//#include "base/remote_gdb.hh" +#include "cpu/profile.hh" +#include "kern/kernel_stats.hh" +#include "mem/functional/memory_control.hh" +#include "mem/functional/physical.hh" +#include "sim/faults.hh" +#include "sim/sim_events.hh" +#include "sim/sim_exit.hh" +#include "sim/system.hh" +#else // !FULL_SYSTEM +#include "mem/functional/functional.hh" +#include "sim/process.hh" +#endif // FULL_SYSTEM + +using namespace TheISA; + +template <class Impl> +template<typename T> +void +OzoneCPU<Impl>::trace_data(T data) { + if (traceData) { + traceData->setData(data); + } +} + +template <class Impl> +OzoneCPU<Impl>::TickEvent::TickEvent(OzoneCPU *c, int w) + : Event(&mainEventQueue, CPU_Tick_Pri), cpu(c), width(w) +{ +} + +template <class Impl> +void +OzoneCPU<Impl>::TickEvent::process() +{ + cpu->tick(); +} + +template <class Impl> +const char * +OzoneCPU<Impl>::TickEvent::description() +{ + return "OzoneCPU tick event"; +} + +template <class Impl> +OzoneCPU<Impl>::OzoneCPU(Params *p) +#if FULL_SYSTEM + : BaseCPU(p), thread(this, 0, p->mem), tickEvent(this, p->width), + mem(p->mem), +#else + : BaseCPU(p), thread(this, 0, p->workload[0], 0), tickEvent(this, p->width), + mem(p->workload[0]->getMemory()), +#endif + comm(5, 5) +{ + frontEnd = new FrontEnd(p); + backEnd = new BackEnd(p); + + _status = Idle; + + if (p->checker) { + BaseCPU *temp_checker = p->checker; + checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker); + checker->setMemory(mem); +#if FULL_SYSTEM + checker->setSystem(p->system); +#endif + checkerXC = new CheckerExecContext<OzoneXC>(&ozoneXC, checker); + thread.xcProxy = checkerXC; + xcProxy = checkerXC; + } else { + checker = NULL; + thread.xcProxy = &ozoneXC; + xcProxy = &ozoneXC; + } + + ozoneXC.cpu = this; + ozoneXC.thread = &thread; + + thread.inSyscall = false; + + thread.setStatus(ExecContext::Suspended); +#if FULL_SYSTEM + /***** All thread state stuff *****/ + thread.cpu = this; + thread.tid = 0; + thread.mem = p->mem; + + thread.quiesceEvent = new EndQuiesceEvent(xcProxy); + + system = p->system; + itb = p->itb; + dtb = p->dtb; + memctrl = p->system->memctrl; + physmem = p->system->physmem; + + if (p->profile) { + thread.profile = new FunctionProfile(p->system->kernelSymtab); + // @todo: This might be better as an ExecContext instead of OzoneXC + Callback *cb = + new MakeCallback<OzoneXC, + &OzoneXC::dumpFuncProfile>(&ozoneXC); + registerExitCallback(cb); + } + + // let's fill with a dummy node for now so we don't get a segfault + // on the first cycle when there's no node available. + static ProfileNode dummyNode; + thread.profileNode = &dummyNode; + thread.profilePC = 3; +#else + thread.cpu = this; + thread.tid = 0; + thread.process = p->workload[0]; + thread.asid = 0; +#endif // !FULL_SYSTEM + + numInst = 0; + startNumInst = 0; + + execContexts.push_back(xcProxy); + + frontEnd->setCPU(this); + backEnd->setCPU(this); + + frontEnd->setXC(xcProxy); + backEnd->setXC(xcProxy); + + frontEnd->setThreadState(&thread); + backEnd->setThreadState(&thread); + + frontEnd->setCommBuffer(&comm); + backEnd->setCommBuffer(&comm); + + frontEnd->setBackEnd(backEnd); + backEnd->setFrontEnd(frontEnd); + + decoupledFrontEnd = p->decoupledFrontEnd; + + globalSeqNum = 1; + + checkInterrupts = false; + + for (int i = 0; i < TheISA::TotalNumRegs; ++i) { + thread.renameTable[i] = new DynInst(this); + thread.renameTable[i]->setResultReady(); + } + + frontEnd->renameTable.copyFrom(thread.renameTable); + backEnd->renameTable.copyFrom(thread.renameTable); + +#if !FULL_SYSTEM +// pTable = p->pTable; +#endif + + lockFlag = 0; + + DPRINTF(OzoneCPU, "OzoneCPU: Created Ozone cpu object.\n"); +} + +template <class Impl> +OzoneCPU<Impl>::~OzoneCPU() +{ +} + +template <class Impl> +void +OzoneCPU<Impl>::switchOut(Sampler *_sampler) +{ + sampler = _sampler; + switchCount = 0; + // Front end needs state from back end, so switch out the back end first. + backEnd->switchOut(); + frontEnd->switchOut(); +} + +template <class Impl> +void +OzoneCPU<Impl>::signalSwitched() +{ + if (++switchCount == 2) { + backEnd->doSwitchOut(); + frontEnd->doSwitchOut(); + if (checker) + checker->switchOut(sampler); + _status = SwitchedOut; + if (tickEvent.scheduled()) + tickEvent.squash(); + sampler->signalSwitched(); + } + assert(switchCount <= 2); +} + +template <class Impl> +void +OzoneCPU<Impl>::takeOverFrom(BaseCPU *oldCPU) +{ + BaseCPU::takeOverFrom(oldCPU); + + backEnd->takeOverFrom(); + frontEnd->takeOverFrom(); + assert(!tickEvent.scheduled()); + + // @todo: Fix hardcoded number + // Clear out any old information in time buffer. + for (int i = 0; i < 6; ++i) { + comm.advance(); + } + + // if any of this CPU's ExecContexts are active, mark the CPU as + // running and schedule its tick event. + for (int i = 0; i < execContexts.size(); ++i) { + ExecContext *xc = execContexts[i]; + if (xc->status() == ExecContext::Active && + _status != Running) { + _status = Running; + tickEvent.schedule(curTick); + } + } + // Nothing running, change status to reflect that we're no longer + // switched out. + if (_status == SwitchedOut) { + _status = Idle; + } +} + +template <class Impl> +void +OzoneCPU<Impl>::activateContext(int thread_num, int delay) +{ + // Eventually change this in SMT. + assert(thread_num == 0); + + assert(_status == Idle); + notIdleFraction++; + scheduleTickEvent(delay); + _status = Running; + thread._status = ExecContext::Active; + frontEnd->wakeFromQuiesce(); +} + +template <class Impl> +void +OzoneCPU<Impl>::suspendContext(int thread_num) +{ + // Eventually change this in SMT. + assert(thread_num == 0); + // @todo: Figure out how to initially set the status properly so + // this is running. +// assert(_status == Running); + notIdleFraction--; + unscheduleTickEvent(); + _status = Idle; +} + +template <class Impl> +void +OzoneCPU<Impl>::deallocateContext(int thread_num) +{ + // for now, these are equivalent + suspendContext(thread_num); +} template <class Impl> -class OoOCPU; +void +OzoneCPU<Impl>::haltContext(int thread_num) +{ + // for now, these are equivalent + suspendContext(thread_num); +} + +template <class Impl> +void +OzoneCPU<Impl>::regStats() +{ + using namespace Stats; + + BaseCPU::regStats(); + + thread.numInsts + .name(name() + ".num_insts") + .desc("Number of instructions executed") + ; + + thread.numMemRefs + .name(name() + ".num_refs") + .desc("Number of memory references") + ; + + notIdleFraction + .name(name() + ".not_idle_fraction") + .desc("Percentage of non-idle cycles") + ; + + idleFraction + .name(name() + ".idle_fraction") + .desc("Percentage of idle cycles") + ; + + quiesceCycles + .name(name() + ".quiesce_cycles") + .desc("Number of cycles spent in quiesce") + ; + + idleFraction = constant(1.0) - notIdleFraction; + + frontEnd->regStats(); + backEnd->regStats(); +} + +template <class Impl> +void +OzoneCPU<Impl>::resetStats() +{ + startNumInst = numInst; + notIdleFraction = (_status != Idle); +} + +template <class Impl> +void +OzoneCPU<Impl>::init() +{ + BaseCPU::init(); + + // Mark this as in syscall so it won't need to squash + thread.inSyscall = true; +#if FULL_SYSTEM + for (int i = 0; i < execContexts.size(); ++i) { + ExecContext *xc = execContexts[i]; + + // initialize CPU, including PC + TheISA::initCPU(xc, xc->readCpuId()); + } +#endif + frontEnd->renameTable.copyFrom(thread.renameTable); + backEnd->renameTable.copyFrom(thread.renameTable); + + thread.inSyscall = false; +} + +template <class Impl> +void +OzoneCPU<Impl>::serialize(std::ostream &os) +{ + BaseCPU::serialize(os); + SERIALIZE_ENUM(_status); + nameOut(os, csprintf("%s.xc", name())); + ozoneXC.serialize(os); + nameOut(os, csprintf("%s.tickEvent", name())); + tickEvent.serialize(os); +} + +template <class Impl> +void +OzoneCPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) +{ + BaseCPU::unserialize(cp, section); + UNSERIALIZE_ENUM(_status); + ozoneXC.unserialize(cp, csprintf("%s.xc", section)); + tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); +} + +template <class Impl> +Fault +OzoneCPU<Impl>::copySrcTranslate(Addr src) +{ + panic("Copy not implemented!\n"); + return NoFault; +#if 0 + static bool no_warn = true; + int blk_size = (dcacheInterface) ? dcacheInterface->getBlockSize() : 64; + // Only support block sizes of 64 atm. + assert(blk_size == 64); + int offset = src & (blk_size - 1); + + // Make sure block doesn't span page + if (no_warn && + (src & TheISA::PageMask) != ((src + blk_size) & TheISA::PageMask) && + (src >> 40) != 0xfffffc) { + warn("Copied block source spans pages %x.", src); + no_warn = false; + } + + memReq->reset(src & ~(blk_size - 1), blk_size); + + // translate to physical address + Fault fault = xc->translateDataReadReq(memReq); + + assert(fault != Alignment_Fault); + + if (fault == NoFault) { + xc->copySrcAddr = src; + xc->copySrcPhysAddr = memReq->paddr + offset; + } else { + xc->copySrcAddr = 0; + xc->copySrcPhysAddr = 0; + } + return fault; +#endif +} + +template <class Impl> +Fault +OzoneCPU<Impl>::copy(Addr dest) +{ + panic("Copy not implemented!\n"); + return NoFault; +#if 0 + static bool no_warn = true; + int blk_size = (dcacheInterface) ? dcacheInterface->getBlockSize() : 64; + // Only support block sizes of 64 atm. + assert(blk_size == 64); + uint8_t data[blk_size]; + //assert(xc->copySrcAddr); + int offset = dest & (blk_size - 1); + + // Make sure block doesn't span page + if (no_warn && + (dest & TheISA::PageMask) != ((dest + blk_size) & TheISA::PageMask) && + (dest >> 40) != 0xfffffc) { + no_warn = false; + warn("Copied block destination spans pages %x. ", dest); + } + + memReq->reset(dest & ~(blk_size -1), blk_size); + // translate to physical address + Fault fault = xc->translateDataWriteReq(memReq); + + assert(fault != Alignment_Fault); + + if (fault == NoFault) { + Addr dest_addr = memReq->paddr + offset; + // Need to read straight from memory since we have more than 8 bytes. + memReq->paddr = xc->copySrcPhysAddr; + xc->mem->read(memReq, data); + memReq->paddr = dest_addr; + xc->mem->write(memReq, data); + if (dcacheInterface) { + memReq->cmd = Copy; + memReq->completionEvent = NULL; + memReq->paddr = xc->copySrcPhysAddr; + memReq->dest = dest_addr; + memReq->size = 64; + memReq->time = curTick; + dcacheInterface->access(memReq); + } + } + return fault; +#endif +} + +#if FULL_SYSTEM +template <class Impl> +Addr +OzoneCPU<Impl>::dbg_vtophys(Addr addr) +{ + return vtophys(xcProxy, addr); +} +#endif // FULL_SYSTEM + +#if FULL_SYSTEM +template <class Impl> +void +OzoneCPU<Impl>::post_interrupt(int int_num, int index) +{ + BaseCPU::post_interrupt(int_num, index); + + if (_status == Idle) { + DPRINTF(IPI,"Suspended Processor awoke\n"); +// thread.activate(); + // Hack for now. Otherwise might have to go through the xcProxy, or + // I need to figure out what's the right thing to call. + activateContext(thread.tid, 1); + } +} +#endif // FULL_SYSTEM + +/* start simulation, program loaded, processor precise state initialized */ +template <class Impl> +void +OzoneCPU<Impl>::tick() +{ + DPRINTF(OzoneCPU, "\n\nOzoneCPU: Ticking cpu.\n"); + + _status = Running; + thread.renameTable[ZeroReg]->setIntResult(0); + thread.renameTable[ZeroReg+TheISA::FP_Base_DepTag]-> + setDoubleResult(0.0); + + comm.advance(); + frontEnd->tick(); + backEnd->tick(); + + // check for instruction-count-based events + comInstEventQueue[0]->serviceEvents(numInst); + + if (!tickEvent.scheduled() && _status == Running) + tickEvent.schedule(curTick + cycles(1)); +} + +template <class Impl> +void +OzoneCPU<Impl>::squashFromXC() +{ + thread.inSyscall = true; + backEnd->generateXCEvent(); +} + +#if !FULL_SYSTEM +template <class Impl> +void +OzoneCPU<Impl>::syscall() +{ + // Not sure this copy is needed, depending on how the XC proxy is made. + thread.renameTable.copyFrom(backEnd->renameTable); + + thread.inSyscall = true; + + thread.funcExeInst++; + + DPRINTF(OzoneCPU, "FuncExeInst: %i\n", thread.funcExeInst); + + thread.process->syscall(xcProxy); + + thread.funcExeInst--; + + thread.inSyscall = false; + + frontEnd->renameTable.copyFrom(thread.renameTable); + backEnd->renameTable.copyFrom(thread.renameTable); +} + +template <class Impl> +void +OzoneCPU<Impl>::setSyscallReturn(SyscallReturn return_value, int tid) +{ + // check for error condition. Alpha syscall convention is to + // indicate success/failure in reg a3 (r19) and put the + // return value itself in the standard return value reg (v0). + if (return_value.successful()) { + // no error + thread.renameTable[SyscallSuccessReg]->setIntResult(0); + thread.renameTable[ReturnValueReg]->setIntResult( + return_value.value()); + } else { + // got an error, return details + thread.renameTable[SyscallSuccessReg]->setIntResult((IntReg) -1); + thread.renameTable[ReturnValueReg]->setIntResult( + -return_value.value()); + } +} +#else +template <class Impl> +Fault +OzoneCPU<Impl>::hwrei() +{ + // Need to move this to ISA code + // May also need to make this per thread + + lockFlag = false; + lockAddrList.clear(); + thread.kernelStats->hwrei(); + + checkInterrupts = true; + + // FIXME: XXX check for interrupts? XXX + return NoFault; +} + +template <class Impl> +void +OzoneCPU<Impl>::processInterrupts() +{ + // Check for interrupts here. For now can copy the code that + // exists within isa_fullsys_traits.hh. Also assume that thread 0 + // is the one that handles the interrupts. + + // Check if there are any outstanding interrupts + //Handle the interrupts + int ipl = 0; + int summary = 0; + + checkInterrupts = false; + + if (thread.readMiscReg(IPR_ASTRR)) + panic("asynchronous traps not implemented\n"); + + if (thread.readMiscReg(IPR_SIRR)) { + for (int i = INTLEVEL_SOFTWARE_MIN; + i < INTLEVEL_SOFTWARE_MAX; i++) { + if (thread.readMiscReg(IPR_SIRR) & (ULL(1) << i)) { + // See table 4-19 of the 21164 hardware reference + ipl = (i - INTLEVEL_SOFTWARE_MIN) + 1; + summary |= (ULL(1) << i); + } + } + } + + uint64_t interrupts = intr_status(); + + if (interrupts) { + for (int i = INTLEVEL_EXTERNAL_MIN; + i < INTLEVEL_EXTERNAL_MAX; i++) { + if (interrupts & (ULL(1) << i)) { + // See table 4-19 of the 21164 hardware reference + ipl = i; + summary |= (ULL(1) << i); + } + } + } + + if (ipl && ipl > thread.readMiscReg(IPR_IPLR)) { + thread.setMiscReg(IPR_ISR, summary); + thread.setMiscReg(IPR_INTID, ipl); + // @todo: Make this more transparent + if (checker) { + checker->cpuXCBase()->setMiscReg(IPR_ISR, summary); + checker->cpuXCBase()->setMiscReg(IPR_INTID, ipl); + } + Fault fault = new InterruptFault; + fault->invoke(thread.getXCProxy()); + DPRINTF(Flow, "Interrupt! IPLR=%d ipl=%d summary=%x\n", + thread.readMiscReg(IPR_IPLR), ipl, summary); + } +} + +template <class Impl> +bool +OzoneCPU<Impl>::simPalCheck(int palFunc) +{ + // Need to move this to ISA code + // May also need to make this per thread + thread.kernelStats->callpal(palFunc, xcProxy); + + switch (palFunc) { + case PAL::halt: + haltContext(thread.tid); + if (--System::numSystemsRunning == 0) + new SimExitEvent("all cpus halted"); + break; + + case PAL::bpt: + case PAL::bugchk: + if (system->breakpoint()) + return false; + break; + } + + return true; +} +#endif + +template <class Impl> +BaseCPU * +OzoneCPU<Impl>::OzoneXC::getCpuPtr() +{ + return cpu; +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::setCpuId(int id) +{ + cpu->cpuId = id; + thread->cpuId = id; +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::setStatus(Status new_status) +{ + thread->_status = new_status; +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::activate(int delay) +{ + cpu->activateContext(thread->tid, delay); +} + +/// Set the status to Suspended. +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::suspend() +{ + cpu->suspendContext(thread->tid); +} + +/// Set the status to Unallocated. +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::deallocate() +{ + cpu->deallocateContext(thread->tid); +} + +/// Set the status to Halted. +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::halt() +{ + cpu->haltContext(thread->tid); +} + +#if FULL_SYSTEM +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::dumpFuncProfile() +{ } +#endif + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::takeOverFrom(ExecContext *old_context) +{ + // some things should already be set up + assert(getMemPtr() == old_context->getMemPtr()); +#if FULL_SYSTEM + assert(getSystemPtr() == old_context->getSystemPtr()); +#else + assert(getProcessPtr() == old_context->getProcessPtr()); +#endif + + // copy over functional state + setStatus(old_context->status()); + copyArchRegs(old_context); + setCpuId(old_context->readCpuId()); + +#if !FULL_SYSTEM + setFuncExeInst(old_context->readFuncExeInst()); +#else + EndQuiesceEvent *other_quiesce = old_context->getQuiesceEvent(); + if (other_quiesce) { + // Point the quiesce event's XC at this XC so that it wakes up + // the proper CPU. + other_quiesce->xc = this; + } + if (thread->quiesceEvent) { + thread->quiesceEvent->xc = this; + } + + thread->kernelStats = old_context->getKernelStats(); +// storeCondFailures = 0; + cpu->lockFlag = false; +#endif + + old_context->setStatus(ExecContext::Unallocated); +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::regStats(const std::string &name) +{ +#if FULL_SYSTEM + thread->kernelStats = new Kernel::Statistics(cpu->system); + thread->kernelStats->regStats(name + ".kern"); +#endif +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::serialize(std::ostream &os) +{ } + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::unserialize(Checkpoint *cp, const std::string §ion) +{ } + +#if FULL_SYSTEM +template <class Impl> +EndQuiesceEvent * +OzoneCPU<Impl>::OzoneXC::getQuiesceEvent() +{ + return thread->quiesceEvent; +} + +template <class Impl> +Tick +OzoneCPU<Impl>::OzoneXC::readLastActivate() +{ + return thread->lastActivate; +} + +template <class Impl> +Tick +OzoneCPU<Impl>::OzoneXC::readLastSuspend() +{ + return thread->lastSuspend; +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::profileClear() +{ + if (thread->profile) + thread->profile->clear(); +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::profileSample() +{ + if (thread->profile) + thread->profile->sample(thread->profileNode, thread->profilePC); +} +#endif + +template <class Impl> +int +OzoneCPU<Impl>::OzoneXC::getThreadNum() +{ + return thread->tid; +} + +// Also somewhat obnoxious. Really only used for the TLB fault. +template <class Impl> +TheISA::MachInst +OzoneCPU<Impl>::OzoneXC::getInst() +{ + return thread->inst; +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::copyArchRegs(ExecContext *xc) +{ + thread->PC = xc->readPC(); + thread->nextPC = xc->readNextPC(); + + cpu->frontEnd->setPC(thread->PC); + cpu->frontEnd->setNextPC(thread->nextPC); + + for (int i = 0; i < TheISA::TotalNumRegs; ++i) { + if (i < TheISA::FP_Base_DepTag) { + thread->renameTable[i]->setIntResult(xc->readIntReg(i)); + } else if (i < (TheISA::FP_Base_DepTag + TheISA::NumFloatRegs)) { + int fp_idx = i - TheISA::FP_Base_DepTag; + thread->renameTable[i]->setDoubleResult( + xc->readFloatRegDouble(fp_idx)); + } + } + +#if !FULL_SYSTEM + thread->funcExeInst = xc->readFuncExeInst(); +#endif + + // Need to copy the XC values into the current rename table, + // copy the misc regs. + thread->regs.miscRegs.copyMiscRegs(xc); +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::clearArchRegs() +{ + panic("Unimplemented!"); +} + +template <class Impl> +uint64_t +OzoneCPU<Impl>::OzoneXC::readIntReg(int reg_idx) +{ + return thread->renameTable[reg_idx]->readIntResult(); +} + +template <class Impl> +float +OzoneCPU<Impl>::OzoneXC::readFloatRegSingle(int reg_idx) +{ + int idx = reg_idx + TheISA::FP_Base_DepTag; + return thread->renameTable[idx]->readFloatResult(); +} + +template <class Impl> +double +OzoneCPU<Impl>::OzoneXC::readFloatRegDouble(int reg_idx) +{ + int idx = reg_idx + TheISA::FP_Base_DepTag; + return thread->renameTable[idx]->readDoubleResult(); +} + +template <class Impl> +uint64_t +OzoneCPU<Impl>::OzoneXC::readFloatRegInt(int reg_idx) +{ + int idx = reg_idx + TheISA::FP_Base_DepTag; + return thread->renameTable[idx]->readIntResult(); +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::setIntReg(int reg_idx, uint64_t val) +{ + thread->renameTable[reg_idx]->setIntResult(val); + + if (!thread->inSyscall) { + cpu->squashFromXC(); + } +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::setFloatRegSingle(int reg_idx, float val) +{ + panic("Unimplemented!"); +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::setFloatRegDouble(int reg_idx, double val) +{ + int idx = reg_idx + TheISA::FP_Base_DepTag; + + thread->renameTable[idx]->setDoubleResult(val); + + if (!thread->inSyscall) { + cpu->squashFromXC(); + } +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::setFloatRegInt(int reg_idx, uint64_t val) +{ + panic("Unimplemented!"); +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::setPC(Addr val) +{ + thread->PC = val; + cpu->frontEnd->setPC(val); + + if (!thread->inSyscall) { + cpu->squashFromXC(); + } +} + +template <class Impl> +void +OzoneCPU<Impl>::OzoneXC::setNextPC(Addr val) +{ + thread->nextPC = val; + cpu->frontEnd->setNextPC(val); + + if (!thread->inSyscall) { + cpu->squashFromXC(); + } +} + +template <class Impl> +TheISA::MiscReg +OzoneCPU<Impl>::OzoneXC::readMiscReg(int misc_reg) +{ + return thread->regs.miscRegs.readReg(misc_reg); +} + +template <class Impl> +TheISA::MiscReg +OzoneCPU<Impl>::OzoneXC::readMiscRegWithEffect(int misc_reg, Fault &fault) +{ + return thread->regs.miscRegs.readRegWithEffect(misc_reg, + fault, this); +} + +template <class Impl> +Fault +OzoneCPU<Impl>::OzoneXC::setMiscReg(int misc_reg, const MiscReg &val) +{ + // Needs to setup a squash event unless we're in syscall mode + Fault ret_fault = thread->regs.miscRegs.setReg(misc_reg, val); + + if (!thread->inSyscall) { + cpu->squashFromXC(); + } + + return ret_fault; +} template <class Impl> -class OoODynInst; +Fault +OzoneCPU<Impl>::OzoneXC::setMiscRegWithEffect(int misc_reg, const MiscReg &val) +{ + // Needs to setup a squash event unless we're in syscall mode + Fault ret_fault = thread->regs.miscRegs.setRegWithEffect(misc_reg, val, + this); -struct OoOImpl { - typedef AlphaISA ISA; - typedef OoOCPU<OoOImpl> OoOCPU; - typedef OoOCPU FullCPU; - typedef OoODynInst<OoOImpl> DynInst; - typedef RefCountingPtr<DynInst> DynInstPtr; -}; + if (!thread->inSyscall) { + cpu->squashFromXC(); + } -#endif // __CPU_OOO_CPU_OOO_IMPL_HH__ + return ret_fault; +} |