diff options
Diffstat (limited to 'cpu')
-rw-r--r-- | cpu/base.cc | 49 | ||||
-rw-r--r-- | cpu/base.hh | 5 | ||||
-rw-r--r-- | cpu/base_dyn_inst.cc | 32 | ||||
-rw-r--r-- | cpu/base_dyn_inst.hh | 8 | ||||
-rw-r--r-- | cpu/cpu_exec_context.cc (renamed from cpu/exec_context.cc) | 116 | ||||
-rw-r--r-- | cpu/cpu_exec_context.hh | 510 | ||||
-rw-r--r-- | cpu/exec_context.hh | 549 | ||||
-rw-r--r-- | cpu/exetrace.cc | 1 | ||||
-rw-r--r-- | cpu/intr_control.cc | 8 | ||||
-rw-r--r-- | cpu/intr_control.hh | 1 | ||||
-rw-r--r-- | cpu/memtest/memtest.cc | 6 | ||||
-rw-r--r-- | cpu/memtest/memtest.hh | 2 | ||||
-rw-r--r-- | cpu/o3/alpha_cpu.hh | 45 | ||||
-rw-r--r-- | cpu/o3/alpha_cpu_builder.cc | 1 | ||||
-rw-r--r-- | cpu/o3/alpha_cpu_impl.hh | 51 | ||||
-rw-r--r-- | cpu/o3/cpu.cc | 33 | ||||
-rw-r--r-- | cpu/o3/cpu.hh | 21 | ||||
-rw-r--r-- | cpu/pc_event.cc | 7 | ||||
-rw-r--r-- | cpu/simple/cpu.cc | 136 | ||||
-rw-r--r-- | cpu/simple/cpu.hh | 53 |
20 files changed, 1064 insertions, 570 deletions
diff --git a/cpu/base.cc b/cpu/base.cc index 5a7ecf152..e2a4c214a 100644 --- a/cpu/base.cc +++ b/cpu/base.cc @@ -39,10 +39,16 @@ #include "cpu/profile.hh" #include "cpu/sampler/sampler.hh" #include "sim/param.hh" +#include "sim/process.hh" #include "sim/sim_events.hh" +#include "sim/system.hh" #include "base/trace.hh" +#if FULL_SYSTEM +#include "kern/kernel_stats.hh" +#endif + using namespace std; vector<BaseCPU *> BaseCPU::cpuList; @@ -147,7 +153,10 @@ BaseCPU::BaseCPU(Params *p) profileEvent = NULL; if (params->profile) profileEvent = new ProfileEvent(this, params->profile); + + kernelStats = new Kernel::Statistics(system); #endif + } BaseCPU::Params::Params() @@ -165,6 +174,10 @@ BaseCPU::enableFunctionTrace() BaseCPU::~BaseCPU() { +#if FULL_SYSTEM + if (kernelStats) + delete kernelStats; +#endif } void @@ -203,6 +216,11 @@ BaseCPU::regStats() } } else if (size == 1) execContexts[0]->regStats(name()); + +#if FULL_SYSTEM + if (kernelStats) + kernelStats->regStats(name() + ".kern"); +#endif } @@ -216,9 +234,9 @@ BaseCPU::registerExecContexts() if (id != -1) id += i; - xc->cpu_id = system->registerExecContext(xc, id); + xc->setCpuId(system->registerExecContext(xc, id)); #else - xc->cpu_id = xc->process->registerExecContext(xc); + xc->setCpuId(xc->getProcessPtr()->registerExecContext(xc)); #endif } } @@ -240,12 +258,12 @@ BaseCPU::takeOverFrom(BaseCPU *oldCPU) ExecContext *oldXC = oldCPU->execContexts[i]; newXC->takeOverFrom(oldXC); - assert(newXC->cpu_id == oldXC->cpu_id); + assert(newXC->readCpuId() == oldXC->readCpuId()); #if FULL_SYSTEM - system->replaceExecContext(newXC, newXC->cpu_id); + system->replaceExecContext(newXC, newXC->readCpuId()); #else - assert(newXC->process == oldXC->process); - newXC->process->replaceExecContext(newXC, newXC->cpu_id); + assert(newXC->getProcessPtr() == oldXC->getProcessPtr()); + newXC->getProcessPtr()->replaceExecContext(newXC, newXC->readCpuId()); #endif } @@ -253,11 +271,11 @@ BaseCPU::takeOverFrom(BaseCPU *oldCPU) for (int i = 0; i < TheISA::NumInterruptLevels; ++i) interrupts[i] = oldCPU->interrupts[i]; intstatus = oldCPU->intstatus; - +/* for (int i = 0; i < execContexts.size(); ++i) if (execContexts[i]->profile) execContexts[i]->profile->clear(); - +*/ if (profileEvent) profileEvent->schedule(curTick); #endif @@ -272,11 +290,11 @@ BaseCPU::ProfileEvent::ProfileEvent(BaseCPU *_cpu, int _interval) void BaseCPU::ProfileEvent::process() { - for (int i = 0, size = cpu->execContexts.size(); i < size; ++i) { +/* for (int i = 0, size = cpu->execContexts.size(); i < size; ++i) { ExecContext *xc = cpu->execContexts[i]; xc->profile->sample(xc->profileNode, xc->profilePC); } - +*/ schedule(curTick + interval); } @@ -327,6 +345,12 @@ BaseCPU::serialize(std::ostream &os) { SERIALIZE_ARRAY(interrupts, TheISA::NumInterruptLevels); SERIALIZE_SCALAR(intstatus); + +#if FULL_SYSTEM + if (kernelStats) + kernelStats->serialize(os); +#endif + } void @@ -334,6 +358,11 @@ BaseCPU::unserialize(Checkpoint *cp, const std::string §ion) { UNSERIALIZE_ARRAY(interrupts, TheISA::NumInterruptLevels); UNSERIALIZE_SCALAR(intstatus); + +#if FULL_SYSTEM + if (kernelStats) + kernelStats->unserialize(cp, section); +#endif } #endif // FULL_SYSTEM diff --git a/cpu/base.hh b/cpu/base.hh index d5764d495..d9d5d2b88 100644 --- a/cpu/base.hh +++ b/cpu/base.hh @@ -40,6 +40,7 @@ #if FULL_SYSTEM class System; +namespace Kernel { class Statistics; } #endif class BranchPred; @@ -234,6 +235,10 @@ class BaseCPU : public SimObject public: // Number of CPU cycles simulated Stats::Scalar<> numCycles; + +#if FULL_SYSTEM + Kernel::Statistics *kernelStats; +#endif }; #endif // __CPU_BASE_HH__ diff --git a/cpu/base_dyn_inst.cc b/cpu/base_dyn_inst.cc index 86314bef1..a40b4a1dd 100644 --- a/cpu/base_dyn_inst.cc +++ b/cpu/base_dyn_inst.cc @@ -67,7 +67,7 @@ template <class Impl> BaseDynInst<Impl>::BaseDynInst(MachInst machInst, Addr inst_PC, Addr pred_PC, InstSeqNum seq_num, FullCPU *cpu) - : staticInst(machInst), traceData(NULL), cpu(cpu), xc(cpu->xcBase()) + : staticInst(machInst), traceData(NULL), cpu(cpu), cpuXC(cpu->cpuXCBase()) { seqNum = seq_num; @@ -138,14 +138,14 @@ BaseDynInst<Impl>::prefetch(Addr addr, unsigned flags) // state. // Generate a MemReq so we can translate the effective address. - MemReqPtr req = new MemReq(addr, xc, 1, flags); + MemReqPtr req = new MemReq(addr, cpuXC->getProxy(), 1, flags); req->asid = asid; // Prefetches never cause faults. fault = NoFault; // note this is a local, not BaseDynInst::fault - Fault trans_fault = xc->translateDataReadReq(req); + Fault trans_fault = cpuXC->translateDataReadReq(req); if (trans_fault == NoFault && !(req->flags & UNCACHEABLE)) { // It's a valid address to cacheable space. Record key MemReq @@ -183,10 +183,10 @@ BaseDynInst<Impl>::writeHint(Addr addr, int size, unsigned flags) // will casue a TLB miss trap if necessary... not sure whether // that's the best thing to do or not. We don't really need the // MemReq otherwise, since wh64 has no functional effect. - MemReqPtr req = new MemReq(addr, xc, size, flags); + MemReqPtr req = new MemReq(addr, cpuXC->getProxy(), size, flags); req->asid = asid; - fault = xc->translateDataWriteReq(req); + fault = cpuXC->translateDataWriteReq(req); if (fault == NoFault && !(req->flags & UNCACHEABLE)) { // Record key MemReq parameters so we can generate another one @@ -211,18 +211,18 @@ template <class Impl> Fault BaseDynInst<Impl>::copySrcTranslate(Addr src) { - MemReqPtr req = new MemReq(src, xc, 64); + MemReqPtr req = new MemReq(src, cpuXC->getProxy(), 64); req->asid = asid; // translate to physical address - Fault fault = xc->translateDataReadReq(req); + Fault fault = cpuXC->translateDataReadReq(req); if (fault == NoFault) { - xc->copySrcAddr = src; - xc->copySrcPhysAddr = req->paddr; + cpuXC->copySrcAddr = src; + cpuXC->copySrcPhysAddr = req->paddr; } else { - xc->copySrcAddr = 0; - xc->copySrcPhysAddr = 0; + cpuXC->copySrcAddr = 0; + cpuXC->copySrcPhysAddr = 0; } return fault; } @@ -235,18 +235,18 @@ Fault BaseDynInst<Impl>::copy(Addr dest) { uint8_t data[64]; - FunctionalMemory *mem = xc->mem; - assert(xc->copySrcPhysAddr || xc->misspeculating()); - MemReqPtr req = new MemReq(dest, xc, 64); + FunctionalMemory *mem = cpuXC->mem; + assert(cpuXC->copySrcPhysAddr || cpuXC->misspeculating()); + MemReqPtr req = new MemReq(dest, cpuXC->getProxy(), 64); req->asid = asid; // translate to physical address - Fault fault = xc->translateDataWriteReq(req); + Fault fault = cpuXC->translateDataWriteReq(req); if (fault == NoFault) { Addr dest_addr = req->paddr; // Need to read straight from memory since we have more than 8 bytes. - req->paddr = xc->copySrcPhysAddr; + req->paddr = cpuXC->copySrcPhysAddr; mem->read(req, data); req->paddr = dest_addr; mem->write(req, data); diff --git a/cpu/base_dyn_inst.hh b/cpu/base_dyn_inst.hh index e94c44151..3a7852f79 100644 --- a/cpu/base_dyn_inst.hh +++ b/cpu/base_dyn_inst.hh @@ -145,7 +145,7 @@ class BaseDynInst : public FastAlloc, public RefCounted FullCPU *cpu; /** Pointer to the exec context. Will not exist in the final version. */ - ExecContext *xc; + CPUExecContext *cpuXC; /** The kind of fault this instruction has generated. */ Fault fault; @@ -406,7 +406,7 @@ class BaseDynInst : public FastAlloc, public RefCounted /** Returns the exec context. * @todo: Remove this once the ExecContext is no longer used. */ - ExecContext *xcBase() { return xc; } + ExecContext *xcBase() { return cpuXC->getProxy(); } private: /** Instruction effective address. @@ -444,7 +444,7 @@ template<class T> inline Fault BaseDynInst<Impl>::read(Addr addr, T &data, unsigned flags) { - MemReqPtr req = new MemReq(addr, xc, sizeof(T), flags); + MemReqPtr req = new MemReq(addr, cpuXC->getProxy(), sizeof(T), flags); req->asid = asid; fault = cpu->translateDataReadReq(req); @@ -492,7 +492,7 @@ BaseDynInst<Impl>::write(T data, Addr addr, unsigned flags, uint64_t *res) traceData->setData(data); } - MemReqPtr req = new MemReq(addr, xc, sizeof(T), flags); + MemReqPtr req = new MemReq(addr, cpuXC->getProxy(), sizeof(T), flags); req->asid = asid; diff --git a/cpu/exec_context.cc b/cpu/cpu_exec_context.cc index 9bed3ba47..2fb3d5d53 100644 --- a/cpu/exec_context.cc +++ b/cpu/cpu_exec_context.cc @@ -29,6 +29,7 @@ #include <string> #include "cpu/base.hh" +#include "cpu/cpu_exec_context.hh" #include "cpu/exec_context.hh" #if FULL_SYSTEM @@ -49,23 +50,23 @@ using namespace std; // constructor #if FULL_SYSTEM -ExecContext::ExecContext(BaseCPU *_cpu, int _thread_num, System *_sys, +CPUExecContext::CPUExecContext(BaseCPU *_cpu, int _thread_num, System *_sys, AlphaITB *_itb, AlphaDTB *_dtb, FunctionalMemory *_mem) : _status(ExecContext::Unallocated), cpu(_cpu), thread_num(_thread_num), cpu_id(-1), mem(_mem), itb(_itb), dtb(_dtb), system(_sys), - memctrl(_sys->memctrl), physmem(_sys->physmem), - kernelBinning(system->kernelBinning), bin(kernelBinning->bin), - fnbin(kernelBinning->fnbin), profile(NULL), + memctrl(_sys->memctrl), physmem(_sys->physmem), profile(NULL), func_exe_inst(0), storeCondFailures(0) { - kernelStats = new Kernel::Statistics(this); + proxy = new ProxyExecContext<CPUExecContext>(this); + memset(®s, 0, sizeof(RegFile)); if (cpu->params->profile) { profile = new FunctionProfile(system->kernelSymtab); Callback *cb = - new MakeCallback<ExecContext, &ExecContext::dumpFuncProfile>(this); + new MakeCallback<CPUExecContext, + &CPUExecContext::dumpFuncProfile>(this); registerExitCallback(cb); } @@ -76,7 +77,7 @@ ExecContext::ExecContext(BaseCPU *_cpu, int _thread_num, System *_sys, profilePC = 3; } #else -ExecContext::ExecContext(BaseCPU *_cpu, int _thread_num, +CPUExecContext::CPUExecContext(BaseCPU *_cpu, int _thread_num, Process *_process, int _asid) : _status(ExecContext::Unallocated), cpu(_cpu), thread_num(_thread_num), cpu_id(-1), @@ -84,36 +85,46 @@ ExecContext::ExecContext(BaseCPU *_cpu, int _thread_num, func_exe_inst(0), storeCondFailures(0) { memset(®s, 0, sizeof(RegFile)); + proxy = new ProxyExecContext<CPUExecContext>(this); } -ExecContext::ExecContext(BaseCPU *_cpu, int _thread_num, +CPUExecContext::CPUExecContext(BaseCPU *_cpu, int _thread_num, FunctionalMemory *_mem, int _asid) : cpu(_cpu), thread_num(_thread_num), process(0), mem(_mem), asid(_asid), func_exe_inst(0), storeCondFailures(0) { memset(®s, 0, sizeof(RegFile)); + proxy = new ProxyExecContext<CPUExecContext>(this); } -#endif -ExecContext::~ExecContext() +CPUExecContext::CPUExecContext(RegFile *regFile) + : cpu(NULL), thread_num(-1), process(NULL), mem(NULL), asid(-1), + func_exe_inst(0), storeCondFailures(0) { -#if FULL_SYSTEM - delete kernelStats; + regs = *regFile; + proxy = new ProxyExecContext<CPUExecContext>(this); +} + #endif + +CPUExecContext::~CPUExecContext() +{ + delete proxy; } #if FULL_SYSTEM void -ExecContext::dumpFuncProfile() +CPUExecContext::dumpFuncProfile() { std::ostream *os = simout.create(csprintf("profile.%s.dat", cpu->name())); - profile->dump(this, *os); + profile->dump(proxy, *os); } #endif void -ExecContext::takeOverFrom(ExecContext *oldContext) +CPUExecContext::takeOverFrom(ExecContext *oldContext) { +/* // some things should already be set up assert(mem == oldContext->mem); #if FULL_SYSTEM @@ -130,98 +141,115 @@ ExecContext::takeOverFrom(ExecContext *oldContext) storeCondFailures = 0; - oldContext->_status = ExecContext::Unallocated; + oldContext->_status = CPUExecContext::Unallocated; +*/ } void -ExecContext::serialize(ostream &os) +CPUExecContext::serialize(ostream &os) { SERIALIZE_ENUM(_status); regs.serialize(os); // thread_num and cpu_id are deterministic from the config SERIALIZE_SCALAR(func_exe_inst); SERIALIZE_SCALAR(inst); - -#if FULL_SYSTEM - kernelStats->serialize(os); -#endif } void -ExecContext::unserialize(Checkpoint *cp, const std::string §ion) +CPUExecContext::unserialize(Checkpoint *cp, const std::string §ion) { UNSERIALIZE_ENUM(_status); regs.unserialize(cp, section); // thread_num and cpu_id are deterministic from the config UNSERIALIZE_SCALAR(func_exe_inst); UNSERIALIZE_SCALAR(inst); - -#if FULL_SYSTEM - kernelStats->unserialize(cp, section); -#endif } void -ExecContext::activate(int delay) +CPUExecContext::activate(int delay) { - if (status() == Active) + if (status() == ExecContext::Active) return; - _status = Active; + _status = ExecContext::Active; cpu->activateContext(thread_num, delay); } void -ExecContext::suspend() +CPUExecContext::suspend() { - if (status() == Suspended) + if (status() == ExecContext::Suspended) return; #if FULL_SYSTEM // Don't change the status from active if there are pending interrupts if (cpu->check_interrupts()) { - assert(status() == Active); + assert(status() == ExecContext::Active); return; } #endif - _status = Suspended; + _status = ExecContext::Suspended; cpu->suspendContext(thread_num); } void -ExecContext::deallocate() +CPUExecContext::deallocate() { - if (status() == Unallocated) + if (status() == ExecContext::Unallocated) return; - _status = Unallocated; + _status = ExecContext::Unallocated; cpu->deallocateContext(thread_num); } void -ExecContext::halt() +CPUExecContext::halt() { - if (status() == Halted) + if (status() == ExecContext::Halted) return; - _status = Halted; + _status = ExecContext::Halted; cpu->haltContext(thread_num); } void -ExecContext::regStats(const string &name) +CPUExecContext::regStats(const string &name) { -#if FULL_SYSTEM - kernelStats->regStats(name + ".kern"); -#endif } void -ExecContext::trap(Fault fault) +CPUExecContext::copyArchRegs(ExecContext *xc) +{ + // First loop through the integer registers. + for (int i = 0; i < AlphaISA::NumIntRegs; ++i) { + setIntReg(i, xc->readIntReg(i)); + } + + // Then loop through the floating point registers. + for (int i = 0; i < AlphaISA::NumFloatRegs; ++i) { + setFloatRegDouble(i, xc->readFloatRegDouble(i)); + setFloatRegInt(i, xc->readFloatRegInt(i)); + } + + // Copy misc. registers + setMiscReg(AlphaISA::Fpcr_DepTag, xc->readMiscReg(AlphaISA::Fpcr_DepTag)); + setMiscReg(AlphaISA::Uniq_DepTag, xc->readMiscReg(AlphaISA::Uniq_DepTag)); + setMiscReg(AlphaISA::Lock_Flag_DepTag, + xc->readMiscReg(AlphaISA::Lock_Flag_DepTag)); + setMiscReg(AlphaISA::Lock_Addr_DepTag, + xc->readMiscReg(AlphaISA::Lock_Addr_DepTag)); + + // Lastly copy PC/NPC + setPC(xc->readPC()); + setNextPC(xc->readNextPC()); +} + +void +CPUExecContext::trap(Fault fault) { //TheISA::trap(fault); //One possible way to do it... diff --git a/cpu/cpu_exec_context.hh b/cpu/cpu_exec_context.hh new file mode 100644 index 000000000..1ae24582d --- /dev/null +++ b/cpu/cpu_exec_context.hh @@ -0,0 +1,510 @@ +/* + * Copyright (c) 2001-2005 The Regents of The University of Michigan + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer; + * redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution; + * neither the name of the copyright holders nor the names of its + * contributors may be used to endorse or promote products derived from + * this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef __CPU_CPU_EXEC_CONTEXT_HH__ +#define __CPU_CPU_EXEC_CONTEXT_HH__ + +#include "arch/isa_traits.hh" +#include "config/full_system.hh" +#include "cpu/exec_context.hh" +#include "mem/functional/functional.hh" +#include "mem/mem_req.hh" +#include "sim/byteswap.hh" +#include "sim/host.hh" +#include "sim/serialize.hh" + +// forward declaration: see functional_memory.hh +class FunctionalMemory; +class PhysicalMemory; +class BaseCPU; + +#if FULL_SYSTEM + +#include "sim/system.hh" +#include "targetarch/alpha_memory.hh" + +class FunctionProfile; +class ProfileNode; +class MemoryController; + +#else // !FULL_SYSTEM + +#include "sim/process.hh" + +#endif // FULL_SYSTEM + +// +// The CPUExecContext object represents a functional context for +// instruction execution. It incorporates everything required for +// architecture-level functional simulation of a single thread. +// + +class CPUExecContext +{ + protected: + typedef TheISA::RegFile RegFile; + typedef TheISA::MachInst MachInst; + typedef TheISA::MiscRegFile MiscRegFile; + typedef TheISA::MiscReg MiscReg; + public: + typedef ExecContext::Status Status; + + private: + Status _status; + + public: + Status status() const { return _status; } + + void setStatus(Status newStatus) { _status = newStatus; } + + /// 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(); + + protected: + RegFile regs; // correct-path register context + + public: + // pointer to CPU associated with this context + BaseCPU *cpu; + + ProxyExecContext<CPUExecContext> *proxy; + + // Current instruction + MachInst inst; + + // Index of hardware thread context on the CPU that this represents. + int thread_num; + + // ID of this context w.r.t. the System or Process object to which + // it belongs. For full-system mode, this is the system CPU ID. + int cpu_id; + +#if FULL_SYSTEM + FunctionalMemory *mem; + 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; + + FunctionProfile *profile; + ProfileNode *profileNode; + Addr profilePC; + void dumpFuncProfile(); + +#else + Process *process; + + FunctionalMemory *mem; // functional storage for process address space + + // Address space ID. Note that this is used for TIMING cache + // simulation only; all functional memory accesses should use + // one of the FunctionalMemory pointers above. + short asid; + +#endif + + /** + * Temporary storage to pass the source address from copy_load to + * copy_store. + * @todo Remove this temporary when we have a better way to do it. + */ + Addr copySrcAddr; + /** + * Temp storage for the physical source address of a copy. + * @todo Remove this temporary when we have a better way to do it. + */ + Addr copySrcPhysAddr; + + + /* + * number of executed instructions, for matching with syscall trace + * points in EIO files. + */ + Counter func_exe_inst; + + // + // Count failed store conditionals so we can warn of apparent + // application deadlock situations. + unsigned storeCondFailures; + + // constructor: initialize context from given process structure +#if FULL_SYSTEM + CPUExecContext(BaseCPU *_cpu, int _thread_num, System *_system, + AlphaITB *_itb, AlphaDTB *_dtb, FunctionalMemory *_dem); +#else + CPUExecContext(BaseCPU *_cpu, int _thread_num, Process *_process, int _asid); + CPUExecContext(BaseCPU *_cpu, int _thread_num, FunctionalMemory *_mem, + int _asid); + // Constructor to use XC to pass reg file around. Not used for anything + // else. + CPUExecContext(RegFile *regFile); +#endif + virtual ~CPUExecContext(); + + virtual void takeOverFrom(ExecContext *oldContext); + + void regStats(const std::string &name); + + void serialize(std::ostream &os); + void unserialize(Checkpoint *cp, const std::string §ion); + + BaseCPU *getCpuPtr() { return cpu; } + + ExecContext *getProxy() { return proxy; } + + int getThreadNum() { return thread_num; } + +#if FULL_SYSTEM + System *getSystemPtr() { return system; } + + PhysicalMemory *getPhysMemPtr() { return physmem; } + + AlphaITB *getITBPtr() { return itb; } + + AlphaDTB *getDTBPtr() { return dtb; } + + bool validInstAddr(Addr addr) { return true; } + bool validDataAddr(Addr addr) { return true; } + int getInstAsid() { return regs.instAsid(); } + int getDataAsid() { return regs.dataAsid(); } + + Fault translateInstReq(MemReqPtr &req) + { + return itb->translate(req); + } + + Fault translateDataReadReq(MemReqPtr &req) + { + return dtb->translate(req, false); + } + + Fault translateDataWriteReq(MemReqPtr &req) + { + return dtb->translate(req, true); + } + +#else + Process *getProcessPtr() { return process; } + + bool validInstAddr(Addr addr) + { return process->validInstAddr(addr); } + + bool validDataAddr(Addr addr) + { return process->validDataAddr(addr); } + + int getInstAsid() { return asid; } + int getDataAsid() { return asid; } + + Fault dummyTranslation(MemReqPtr &req) + { +#if 0 + assert((req->vaddr >> 48 & 0xffff) == 0); +#endif + + // put the asid in the upper 16 bits of the paddr + req->paddr = req->vaddr & ~((Addr)0xffff << sizeof(Addr) * 8 - 16); + req->paddr = req->paddr | (Addr)req->asid << sizeof(Addr) * 8 - 16; + return NoFault; + } + Fault translateInstReq(MemReqPtr &req) + { + return dummyTranslation(req); + } + Fault translateDataReadReq(MemReqPtr &req) + { + return dummyTranslation(req); + } + Fault translateDataWriteReq(MemReqPtr &req) + { + return dummyTranslation(req); + } + +#endif + + template <class T> + Fault read(MemReqPtr &req, T &data) + { +#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 + + Fault error; + error = mem->read(req, data); + data = LittleEndianGuest::gtoh(data); + return error; + } + + template <class T> + Fault write(MemReqPtr &req, T &data) + { +#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); + } + } + + // 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 < system->execContexts.size(); i++){ + xc = system->execContexts[i]; + if ((xc->readMiscReg(TheISA::Lock_Addr_DepTag) & ~0xf) == + (req->paddr & ~0xf)) { + xc->setMiscReg(TheISA::Lock_Flag_DepTag, false); + } + } + +#endif + return mem->write(req, (T)LittleEndianGuest::htog(data)); + } + + virtual bool misspeculating(); + + + MachInst getInst() { return inst; } + + void setInst(MachInst new_inst) + { + inst = new_inst; + } + + Fault instRead(MemReqPtr &req) + { + return mem->read(req, inst); + } + + void setCpuId(int id) { cpu_id = id; } + + int readCpuId() { return cpu_id; } + + FunctionalMemory *getMemPtr() { return mem; } + + void copyArchRegs(ExecContext *xc); + + // + // New accessors for new decoder. + // + uint64_t readIntReg(int reg_idx) + { + return regs.intRegFile[reg_idx]; + } + + float readFloatRegSingle(int reg_idx) + { + return (float)regs.floatRegFile.d[reg_idx]; + } + + double readFloatRegDouble(int reg_idx) + { + return regs.floatRegFile.d[reg_idx]; + } + + uint64_t readFloatRegInt(int reg_idx) + { + return regs.floatRegFile.q[reg_idx]; + } + + void setIntReg(int reg_idx, uint64_t val) + { + regs.intRegFile[reg_idx] = val; + } + + void setFloatRegSingle(int reg_idx, float val) + { + regs.floatRegFile.d[reg_idx] = (double)val; + } + + void setFloatRegDouble(int reg_idx, double val) + { + regs.floatRegFile.d[reg_idx] = val; + } + + void setFloatRegInt(int reg_idx, uint64_t val) + { + regs.floatRegFile.q[reg_idx] = val; + } + + uint64_t readPC() + { + return regs.pc; + } + + void setPC(uint64_t val) + { + regs.pc = val; + } + + uint64_t readNextPC() + { + return regs.npc; + } + + void setNextPC(uint64_t val) + { + regs.npc = val; + } + + MiscReg readMiscReg(int misc_reg) + { + return regs.miscRegs.readReg(misc_reg); + } + + MiscReg readMiscRegWithEffect(int misc_reg, Fault &fault) + { + return regs.miscRegs.readRegWithEffect(misc_reg, fault, proxy); + } + + Fault setMiscReg(int misc_reg, const MiscReg &val) + { + return regs.miscRegs.setReg(misc_reg, val); + } + + Fault setMiscRegWithEffect(int misc_reg, const MiscReg &val) + { + return regs.miscRegs.setRegWithEffect(misc_reg, val, proxy); + } + + unsigned readStCondFailures() { return storeCondFailures; } + + void setStCondFailures(unsigned sc_failures) + { storeCondFailures = sc_failures; } + + void clearArchRegs() { memset(®s, 0, sizeof(regs)); } + +#if FULL_SYSTEM + int readIntrFlag() { return regs.intrflag; } + void setIntrFlag(int val) { regs.intrflag = val; } + Fault hwrei(); + bool inPalMode() { return AlphaISA::PcPAL(regs.pc); } + void ev5_trap(Fault fault); + bool simPalCheck(int palFunc); +#endif + + /** Meant to be more generic trap function to be + * called when an instruction faults. + * @param fault The fault generated by executing the instruction. + * @todo How to do this properly so it's dependent upon ISA only? + */ + + void trap(Fault fault); + +#if !FULL_SYSTEM + TheISA::IntReg getSyscallArg(int i) + { + return regs.intRegFile[TheISA::ArgumentReg0 + i]; + } + + // used to shift args for indirect syscall + void setSyscallArg(int i, TheISA::IntReg val) + { + regs.intRegFile[TheISA::ArgumentReg0 + i] = val; + } + + void setSyscallReturn(SyscallReturn return_value) + { + // 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). + const int RegA3 = 19; // only place this is used + if (return_value.successful()) { + // no error + regs.intRegFile[RegA3] = 0; + regs.intRegFile[TheISA::ReturnValueReg] = return_value.value(); + } else { + // got an error, return details + regs.intRegFile[RegA3] = (TheISA::IntReg) -1; + regs.intRegFile[TheISA::ReturnValueReg] = -return_value.value(); + } + } + + void syscall() + { + process->syscall(proxy); + } + + Counter readFuncExeInst() { return func_exe_inst; } + + void setFuncExeInst(Counter new_val) { func_exe_inst = new_val; } +#endif +}; + + +// for non-speculative execution context, spec_mode is always false +inline bool +CPUExecContext::misspeculating() +{ + return false; +} + +#endif // __CPU_CPU_EXEC_CONTEXT_HH__ diff --git a/cpu/exec_context.hh b/cpu/exec_context.hh index 88b12c301..9c96b5c42 100644 --- a/cpu/exec_context.hh +++ b/cpu/exec_context.hh @@ -1,5 +1,5 @@ /* - * Copyright (c) 2001-2005 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 @@ -30,46 +30,29 @@ #define __CPU_EXEC_CONTEXT_HH__ #include "config/full_system.hh" -#include "mem/functional/functional.hh" #include "mem/mem_req.hh" +#include "sim/faults.hh" #include "sim/host.hh" #include "sim/serialize.hh" -#include "arch/isa_traits.hh" -//#include "arch/isa_registers.hh" #include "sim/byteswap.hh" // forward declaration: see functional_memory.hh +// @todo: Figure out a more architecture independent way to obtain the ITB and +// DTB pointers. +class AlphaDTB; +class AlphaITB; +class BaseCPU; class FunctionalMemory; class PhysicalMemory; -class BaseCPU; - -#if FULL_SYSTEM - -#include "sim/system.hh" -#include "targetarch/alpha_memory.hh" - -class FunctionProfile; -class ProfileNode; -class MemoryController; -namespace Kernel { class Binning; class Statistics; } - -#else // !FULL_SYSTEM - -#include "sim/process.hh" - -#endif // FULL_SYSTEM - -// -// The ExecContext object represents a functional context for -// instruction execution. It incorporates everything required for -// architecture-level functional simulation of a single thread. -// +class Process; +class System; class ExecContext { protected: typedef TheISA::RegFile RegFile; typedef TheISA::MachInst MachInst; + typedef TheISA::IntReg IntReg; typedef TheISA::MiscRegFile MiscRegFile; typedef TheISA::MiscReg MiscReg; public: @@ -86,7 +69,7 @@ class ExecContext Active, /// Temporarily inactive. Entered while waiting for - /// initialization,synchronization, etc. + /// synchronization, etc. Suspended, /// Permanently shut down. Entered when target executes @@ -95,392 +78,326 @@ class ExecContext Halted }; - private: - Status _status; + virtual ~ExecContext() { }; - public: - Status status() const { return _status; } + virtual BaseCPU *getCpuPtr() = 0; + + virtual void setCpuId(int id) = 0; - void setStatus(Status newStatus) { _status = newStatus; } + virtual int readCpuId() = 0; + + virtual FunctionalMemory *getMemPtr() = 0; + +#if FULL_SYSTEM + virtual System *getSystemPtr() = 0; + + virtual PhysicalMemory *getPhysMemPtr() = 0; + + virtual AlphaITB *getITBPtr() = 0; + + virtual AlphaDTB * getDTBPtr() = 0; +#else + virtual Process *getProcessPtr() = 0; +#endif + + virtual Status status() const = 0; /// Set the status to Active. Optional delay indicates number of /// cycles to wait before beginning execution. - void activate(int delay = 1); + virtual void activate(int delay = 1) = 0; /// Set the status to Suspended. - void suspend(); + virtual void suspend() = 0; /// Set the status to Unallocated. - void deallocate(); + virtual void deallocate() = 0; /// Set the status to Halted. - void halt(); + virtual void halt() = 0; - public: - RegFile regs; // correct-path register context +#if FULL_SYSTEM + virtual void dumpFuncProfile() = 0; +#endif - // pointer to CPU associated with this context - BaseCPU *cpu; + virtual void takeOverFrom(ExecContext *oldContext) = 0; - // Current instruction - MachInst inst; + virtual void regStats(const std::string &name) = 0; - // Index of hardware thread context on the CPU that this represents. - int thread_num; + virtual void serialize(std::ostream &os) = 0; + virtual void unserialize(Checkpoint *cp, const std::string §ion) = 0; - // ID of this context w.r.t. the System or Process object to which - // it belongs. For full-system mode, this is the system CPU ID. - int cpu_id; + virtual int getThreadNum() = 0; -#if FULL_SYSTEM - FunctionalMemory *mem; - 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; - - Kernel::Binning *kernelBinning; - Kernel::Statistics *kernelStats; - bool bin; - bool fnbin; - - FunctionProfile *profile; - ProfileNode *profileNode; - Addr profilePC; - void dumpFuncProfile(); + virtual bool validInstAddr(Addr addr) = 0; + virtual bool validDataAddr(Addr addr) = 0; + virtual int getInstAsid() = 0; + virtual int getDataAsid() = 0; -#else - Process *process; + virtual Fault translateInstReq(MemReqPtr &req) = 0; + + virtual Fault translateDataReadReq(MemReqPtr &req) = 0; + + virtual Fault translateDataWriteReq(MemReqPtr &req) = 0; + + virtual TheISA::MachInst getInst() = 0; + + virtual void copyArchRegs(ExecContext *xc) = 0; + + virtual void clearArchRegs() = 0; + + // + // New accessors for new decoder. + // + virtual uint64_t readIntReg(int reg_idx) = 0; + + virtual float readFloatRegSingle(int reg_idx) = 0; - FunctionalMemory *mem; // functional storage for process address space + virtual double readFloatRegDouble(int reg_idx) = 0; - // Address space ID. Note that this is used for TIMING cache - // simulation only; all functional memory accesses should use - // one of the FunctionalMemory pointers above. - short asid; + virtual uint64_t readFloatRegInt(int reg_idx) = 0; + virtual void setIntReg(int reg_idx, uint64_t val) = 0; + + virtual void setFloatRegSingle(int reg_idx, float val) = 0; + + virtual void setFloatRegDouble(int reg_idx, double val) = 0; + + virtual void setFloatRegInt(int reg_idx, uint64_t val) = 0; + + virtual uint64_t readPC() = 0; + + virtual void setPC(uint64_t val) = 0; + + virtual uint64_t readNextPC() = 0; + + virtual void setNextPC(uint64_t val) = 0; + + virtual MiscReg readMiscReg(int misc_reg) = 0; + + virtual MiscReg readMiscRegWithEffect(int misc_reg, Fault &fault) = 0; + + virtual Fault setMiscReg(int misc_reg, const MiscReg &val) = 0; + + virtual Fault setMiscRegWithEffect(int misc_reg, const MiscReg &val) = 0; + + virtual unsigned readStCondFailures() = 0; + + virtual void setStCondFailures(unsigned sc_failures) = 0; + +#if FULL_SYSTEM + virtual int readIntrFlag() = 0; + virtual void setIntrFlag(int val) = 0; + virtual Fault hwrei() = 0; + virtual bool inPalMode() = 0; + virtual void ev5_trap(Fault fault) = 0; + virtual bool simPalCheck(int palFunc) = 0; #endif - /** - * Temporary storage to pass the source address from copy_load to - * copy_store. - * @todo Remove this temporary when we have a better way to do it. - */ - Addr copySrcAddr; - /** - * Temp storage for the physical source address of a copy. - * @todo Remove this temporary when we have a better way to do it. + virtual bool misspeculating() = 0; + + /** Meant to be more generic trap function to be + * called when an instruction faults. + * @param fault The fault generated by executing the instruction. + * @todo How to do this properly so it's dependent upon ISA only? */ - Addr copySrcPhysAddr; + virtual void trap(Fault fault) = 0; - /* - * number of executed instructions, for matching with syscall trace - * points in EIO files. - */ - Counter func_exe_inst; +#if !FULL_SYSTEM + virtual IntReg getSyscallArg(int i) = 0; - // - // Count failed store conditionals so we can warn of apparent - // application deadlock situations. - unsigned storeCondFailures; + // used to shift args for indirect syscall + virtual void setSyscallArg(int i, IntReg val) = 0; - // constructor: initialize context from given process structure -#if FULL_SYSTEM - ExecContext(BaseCPU *_cpu, int _thread_num, System *_system, - AlphaITB *_itb, AlphaDTB *_dtb, FunctionalMemory *_dem); -#else - ExecContext(BaseCPU *_cpu, int _thread_num, Process *_process, int _asid); - ExecContext(BaseCPU *_cpu, int _thread_num, FunctionalMemory *_mem, - int _asid); + virtual void setSyscallReturn(SyscallReturn return_value) = 0; + + virtual void syscall() = 0; + + virtual Counter readFuncExeInst() = 0; + + virtual void setFuncExeInst(Counter new_val) = 0; #endif - virtual ~ExecContext(); +}; - virtual void takeOverFrom(ExecContext *oldContext); +template <class XC> +class ProxyExecContext : public ExecContext +{ + public: + ProxyExecContext(XC *actual_xc) + { actualXC = actual_xc; } - void regStats(const std::string &name); + private: + XC *actualXC; - void serialize(std::ostream &os); - void unserialize(Checkpoint *cp, const std::string §ion); + public: -#if FULL_SYSTEM - bool validInstAddr(Addr addr) { return true; } - bool validDataAddr(Addr addr) { return true; } - int getInstAsid() { return regs.instAsid(); } - int getDataAsid() { return regs.dataAsid(); } + BaseCPU *getCpuPtr() { return actualXC->getCpuPtr(); } - Fault translateInstReq(MemReqPtr &req) - { - return itb->translate(req); - } + void setCpuId(int id) { actualXC->setCpuId(id); } - Fault translateDataReadReq(MemReqPtr &req) - { - return dtb->translate(req, false); - } + int readCpuId() { return actualXC->readCpuId(); } - Fault translateDataWriteReq(MemReqPtr &req) - { - return dtb->translate(req, true); - } + FunctionalMemory *getMemPtr() { return actualXC->getMemPtr(); } -#else - bool validInstAddr(Addr addr) - { return process->validInstAddr(addr); } +#if FULL_SYSTEM + System *getSystemPtr() { return actualXC->getSystemPtr(); } - bool validDataAddr(Addr addr) - { return process->validDataAddr(addr); } + PhysicalMemory *getPhysMemPtr() { return actualXC->getPhysMemPtr(); } - int getInstAsid() { return asid; } - int getDataAsid() { return asid; } + AlphaITB *getITBPtr() { return actualXC->getITBPtr(); } - Fault dummyTranslation(MemReqPtr &req) - { -#if 0 - assert((req->vaddr >> 48 & 0xffff) == 0); + AlphaDTB *getDTBPtr() { return actualXC->getDTBPtr(); } +#else + Process *getProcessPtr() { return actualXC->getProcessPtr(); } #endif - // put the asid in the upper 16 bits of the paddr - req->paddr = req->vaddr & ~((Addr)0xffff << sizeof(Addr) * 8 - 16); - req->paddr = req->paddr | (Addr)req->asid << sizeof(Addr) * 8 - 16; - return NoFault; - } - Fault translateInstReq(MemReqPtr &req) - { - return dummyTranslation(req); - } - Fault translateDataReadReq(MemReqPtr &req) - { - return dummyTranslation(req); - } - Fault translateDataWriteReq(MemReqPtr &req) - { - return dummyTranslation(req); - } + Status status() const { return actualXC->status(); } -#endif + /// Set the status to Active. Optional delay indicates number of + /// cycles to wait before beginning execution. + void activate(int delay = 1) { actualXC->activate(delay); } - template <class T> - Fault read(MemReqPtr &req, T &data) - { -#if FULL_SYSTEM && defined(TARGET_ALPHA) - if (req->flags & LOCKED) { - MiscRegFile *cregs = &req->xc->regs.miscRegs; - cregs->setReg(TheISA::Lock_Addr_DepTag, req->paddr); - cregs->setReg(TheISA::Lock_Flag_DepTag, true); - } -#endif + /// Set the status to Suspended. + void suspend() { actualXC->suspend(); } - Fault error; - error = mem->read(req, data); - data = LittleEndianGuest::gtoh(data); - return error; - } + /// Set the status to Unallocated. + void deallocate() { actualXC->deallocate(); } - template <class T> - Fault write(MemReqPtr &req, T &data) - { -#if FULL_SYSTEM && defined(TARGET_ALPHA) - - MiscRegFile *cregs; - - // If this is a store conditional, act appropriately - if (req->flags & LOCKED) { - cregs = &req->xc->regs.miscRegs; - - if (req->flags & UNCACHEABLE) { - // Don't update result register (see stq_c in isa_desc) - req->result = 2; - req->xc->storeCondFailures = 0;//Needed? [RGD] - } else { - bool lock_flag = cregs->readReg(TheISA::Lock_Flag_DepTag); - Addr lock_addr = cregs->readReg(TheISA::Lock_Addr_DepTag); - req->result = lock_flag; - if (!lock_flag || - ((lock_addr & ~0xf) != (req->paddr & ~0xf))) { - cregs->setReg(TheISA::Lock_Flag_DepTag, false); - if (((++req->xc->storeCondFailures) % 100000) == 0) { - std::cerr << "Warning: " - << req->xc->storeCondFailures - << " consecutive store conditional failures " - << "on cpu " << req->xc->cpu_id - << std::endl; - } - return NoFault; - } - else req->xc->storeCondFailures = 0; - } - } - - // 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 < system->execContexts.size(); i++){ - cregs = &system->execContexts[i]->regs.miscRegs; - if ((cregs->readReg(TheISA::Lock_Addr_DepTag) & ~0xf) == - (req->paddr & ~0xf)) { - cregs->setReg(TheISA::Lock_Flag_DepTag, false); - } - } + /// Set the status to Halted. + void halt() { actualXC->halt(); } +#if FULL_SYSTEM + void dumpFuncProfile() { actualXC->dumpFuncProfile(); } #endif - return mem->write(req, (T)LittleEndianGuest::htog(data)); - } - virtual bool misspeculating(); + void takeOverFrom(ExecContext *oldContext) + { actualXC->takeOverFrom(oldContext); } + void regStats(const std::string &name) { actualXC->regStats(name); } - MachInst getInst() { return inst; } + void serialize(std::ostream &os) { actualXC->serialize(os); } + void unserialize(Checkpoint *cp, const std::string §ion) + { actualXC->unserialize(cp, section); } - void setInst(MachInst new_inst) - { - inst = new_inst; - } + int getThreadNum() { return actualXC->getThreadNum(); } - Fault instRead(MemReqPtr &req) - { - return mem->read(req, inst); - } + bool validInstAddr(Addr addr) { return actualXC->validInstAddr(addr); } + bool validDataAddr(Addr addr) { return actualXC->validDataAddr(addr); } + int getInstAsid() { return actualXC->getInstAsid(); } + int getDataAsid() { return actualXC->getDataAsid(); } + + Fault translateInstReq(MemReqPtr &req) + { return actualXC->translateInstReq(req); } + + Fault translateDataReadReq(MemReqPtr &req) + { return actualXC->translateDataReadReq(req); } + + Fault translateDataWriteReq(MemReqPtr &req) + { return actualXC->translateDataWriteReq(req); } + + // @todo: Do I need this? + MachInst getInst() { return actualXC->getInst(); } + + // @todo: Do I need this? + void copyArchRegs(ExecContext *xc) { actualXC->copyArchRegs(xc); } + + void clearArchRegs() { actualXC->clearArchRegs(); } // // New accessors for new decoder. // uint64_t readIntReg(int reg_idx) - { - return regs.intRegFile[reg_idx]; - } + { return actualXC->readIntReg(reg_idx); } float readFloatRegSingle(int reg_idx) - { - return (float)regs.floatRegFile.d[reg_idx]; - } + { return actualXC->readFloatRegSingle(reg_idx); } double readFloatRegDouble(int reg_idx) - { - return regs.floatRegFile.d[reg_idx]; - } + { return actualXC->readFloatRegDouble(reg_idx); } uint64_t readFloatRegInt(int reg_idx) - { - return regs.floatRegFile.q[reg_idx]; - } + { return actualXC->readFloatRegInt(reg_idx); } void setIntReg(int reg_idx, uint64_t val) - { - regs.intRegFile[reg_idx] = val; - } + { actualXC->setIntReg(reg_idx, val); } void setFloatRegSingle(int reg_idx, float val) - { - regs.floatRegFile.d[reg_idx] = (double)val; - } + { actualXC->setFloatRegSingle(reg_idx, val); } void setFloatRegDouble(int reg_idx, double val) - { - regs.floatRegFile.d[reg_idx] = val; - } + { actualXC->setFloatRegDouble(reg_idx, val); } void setFloatRegInt(int reg_idx, uint64_t val) - { - regs.floatRegFile.q[reg_idx] = val; - } + { actualXC->setFloatRegInt(reg_idx, val); } - uint64_t readPC() - { - return regs.pc; - } + uint64_t readPC() { return actualXC->readPC(); } - void setNextPC(uint64_t val) - { - regs.npc = val; - } + void setPC(uint64_t val) { actualXC->setPC(val); } + + uint64_t readNextPC() { return actualXC->readNextPC(); } + + void setNextPC(uint64_t val) { actualXC->setNextPC(val); } MiscReg readMiscReg(int misc_reg) - { - return regs.miscRegs.readReg(misc_reg); - } + { return actualXC->readMiscReg(misc_reg); } MiscReg readMiscRegWithEffect(int misc_reg, Fault &fault) - { - return regs.miscRegs.readRegWithEffect(misc_reg, fault, this); - } + { return actualXC->readMiscRegWithEffect(misc_reg, fault); } Fault setMiscReg(int misc_reg, const MiscReg &val) - { - return regs.miscRegs.setReg(misc_reg, val); - } + { return actualXC->setMiscReg(misc_reg, val); } Fault setMiscRegWithEffect(int misc_reg, const MiscReg &val) - { - return regs.miscRegs.setRegWithEffect(misc_reg, val, this); - } + { return actualXC->setMiscRegWithEffect(misc_reg, val); } + + unsigned readStCondFailures() + { return actualXC->readStCondFailures(); } + + void setStCondFailures(unsigned sc_failures) + { actualXC->setStCondFailures(sc_failures); } #if FULL_SYSTEM - int readIntrFlag() { return regs.intrflag; } - void setIntrFlag(int val) { regs.intrflag = val; } - Fault hwrei(); - bool inPalMode() { return AlphaISA::PcPAL(regs.pc); } - void ev5_trap(Fault fault); - bool simPalCheck(int palFunc); + int readIntrFlag() { return actualXC->readIntrFlag(); } + + void setIntrFlag(int val) { actualXC->setIntrFlag(val); } + + Fault hwrei() { return actualXC->hwrei(); } + + bool inPalMode() { return actualXC->inPalMode(); } + + void ev5_trap(Fault fault) { actualXC->ev5_trap(fault); } + + bool simPalCheck(int palFunc) { return actualXC->simPalCheck(palFunc); } #endif + // @todo: Fix this! + bool misspeculating() { return false; } + /** Meant to be more generic trap function to be * called when an instruction faults. * @param fault The fault generated by executing the instruction. * @todo How to do this properly so it's dependent upon ISA only? */ - void trap(Fault fault); + void trap(Fault fault) { actualXC->trap(fault); } #if !FULL_SYSTEM - TheISA::IntReg getSyscallArg(int i) - { - return regs.intRegFile[TheISA::ArgumentReg0 + i]; - } + IntReg getSyscallArg(int i) { return actualXC->getSyscallArg(i); } // used to shift args for indirect syscall - void setSyscallArg(int i, TheISA::IntReg val) - { - regs.intRegFile[TheISA::ArgumentReg0 + i] = val; - } + void setSyscallArg(int i, IntReg val) + { actualXC->setSyscallArg(i, val); } void setSyscallReturn(SyscallReturn return_value) - { - // 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). - const int RegA3 = 19; // only place this is used - if (return_value.successful()) { - // no error - regs.intRegFile[RegA3] = 0; - regs.intRegFile[TheISA::ReturnValueReg] = return_value.value(); - } else { - // got an error, return details - regs.intRegFile[RegA3] = (TheISA::IntReg) -1; - regs.intRegFile[TheISA::ReturnValueReg] = -return_value.value(); - } - } - - void syscall() - { - process->syscall(this); - } -#endif -}; + { actualXC->setSyscallReturn(return_value); } + void syscall() { actualXC->syscall(); } -// for non-speculative execution context, spec_mode is always false -inline bool -ExecContext::misspeculating() -{ - return false; -} + Counter readFuncExeInst() { return actualXC->readFuncExeInst(); } -#endif // __CPU_EXEC_CONTEXT_HH__ + void setFuncExeInst(Counter new_val) + { return actualXC->setFuncExeInst(new_val); } +#endif +}; + +#endif diff --git a/cpu/exetrace.cc b/cpu/exetrace.cc index 8393a1b85..84b5eacf7 100644 --- a/cpu/exetrace.cc +++ b/cpu/exetrace.cc @@ -34,7 +34,6 @@ #include "encumbered/cpu/full/spec_state.hh" #include "encumbered/cpu/full/issue.hh" #include "cpu/exetrace.hh" -#include "cpu/exec_context.hh" #include "base/loader/symtab.hh" #include "cpu/base.hh" #include "cpu/static_inst.hh" diff --git a/cpu/intr_control.cc b/cpu/intr_control.cc index 5f17c7212..d1866a0c4 100644 --- a/cpu/intr_control.cc +++ b/cpu/intr_control.cc @@ -48,7 +48,7 @@ void IntrControl::post(int int_num, int index) { std::vector<ExecContext *> &xcvec = cpu->system->execContexts; - BaseCPU *temp = xcvec[0]->cpu; + BaseCPU *temp = xcvec[0]->getCpuPtr(); temp->post_interrupt(int_num, index); } @@ -56,7 +56,7 @@ void IntrControl::post(int cpu_id, int int_num, int index) { std::vector<ExecContext *> &xcvec = cpu->system->execContexts; - BaseCPU *temp = xcvec[cpu_id]->cpu; + BaseCPU *temp = xcvec[cpu_id]->getCpuPtr(); temp->post_interrupt(int_num, index); } @@ -64,7 +64,7 @@ void IntrControl::clear(int int_num, int index) { std::vector<ExecContext *> &xcvec = cpu->system->execContexts; - BaseCPU *temp = xcvec[0]->cpu; + BaseCPU *temp = xcvec[0]->getCpuPtr(); temp->clear_interrupt(int_num, index); } @@ -72,7 +72,7 @@ void IntrControl::clear(int cpu_id, int int_num, int index) { std::vector<ExecContext *> &xcvec = cpu->system->execContexts; - BaseCPU *temp = xcvec[cpu_id]->cpu; + BaseCPU *temp = xcvec[cpu_id]->getCpuPtr(); temp->clear_interrupt(int_num, index); } diff --git a/cpu/intr_control.hh b/cpu/intr_control.hh index 2a57a8dfc..5ec4e14cb 100644 --- a/cpu/intr_control.hh +++ b/cpu/intr_control.hh @@ -34,7 +34,6 @@ #include "cpu/base.hh" #include "sim/sim_object.hh" #include "sim/system.hh" -#include "cpu/exec_context.hh" class IntrControl : public SimObject diff --git a/cpu/memtest/memtest.cc b/cpu/memtest/memtest.cc index 5a4024587..94b66b70b 100644 --- a/cpu/memtest/memtest.cc +++ b/cpu/memtest/memtest.cc @@ -36,7 +36,7 @@ #include "base/misc.hh" #include "base/statistics.hh" -#include "cpu/exec_context.hh" +#include "cpu/cpu_exec_context.hh" #include "cpu/memtest/memtest.hh" #include "mem/cache/base_cache.hh" #include "sim/builder.hh" @@ -79,7 +79,7 @@ MemTest::MemTest(const string &name, vector<string> cmd; cmd.push_back("/bin/ls"); vector<string> null_vec; - xc = new ExecContext(NULL, 0, mainMem, 0); + cpuXC = new CPUExecContext(NULL, 0, mainMem, 0); blockSize = cacheInterface->getBlockSize(); blockAddrMask = blockSize - 1; @@ -269,7 +269,7 @@ MemTest::tick() req->data = new uint8_t[req->size]; req->paddr &= ~(req->size - 1); req->time = curTick; - req->xc = xc; + req->xc = cpuXC->getProxy(); if (cmd < percentReads) { // read diff --git a/cpu/memtest/memtest.hh b/cpu/memtest/memtest.hh index 7abcf017a..cdb40a26a 100644 --- a/cpu/memtest/memtest.hh +++ b/cpu/memtest/memtest.hh @@ -83,7 +83,7 @@ class MemTest : public SimObject MemInterface *cacheInterface; FunctionalMemory *mainMem; FunctionalMemory *checkMem; - ExecContext *xc; + CPUExecContext *cpuXC; unsigned size; // size of testing memory region diff --git a/cpu/o3/alpha_cpu.hh b/cpu/o3/alpha_cpu.hh index 47ea532a6..75a4d72c2 100644 --- a/cpu/o3/alpha_cpu.hh +++ b/cpu/o3/alpha_cpu.hh @@ -152,13 +152,13 @@ class AlphaFullCPU : public FullO3CPU<Impl> // set the register. IntReg getSyscallArg(int i) { - return this->xc->regs.intRegFile[AlphaISA::ArgumentReg0 + i]; + return this->cpuXC->readIntReg(AlphaISA::ArgumentReg0 + i); } // used to shift args for indirect syscall void setSyscallArg(int i, IntReg val) { - this->xc->regs.intRegFile[AlphaISA::ArgumentReg0 + i] = val; + this->cpuXC->setIntReg(AlphaISA::ArgumentReg0 + i, val); } void setSyscallReturn(int64_t return_value) @@ -169,12 +169,12 @@ class AlphaFullCPU : public FullO3CPU<Impl> const int RegA3 = 19; // only place this is used if (return_value >= 0) { // no error - this->xc->regs.intRegFile[RegA3] = 0; - this->xc->regs.intRegFile[AlphaISA::ReturnValueReg] = return_value; + this->cpuXC->setIntReg(RegA3, 0); + this->cpuXC->setIntReg(AlphaISA::ReturnValueReg, return_value); } else { // got an error, return details - this->xc->regs.intRegFile[RegA3] = (IntReg) -1; - this->xc->regs.intRegFile[AlphaISA::ReturnValueReg] = -return_value; + this->cpuXC->setIntReg(RegA3, (IntReg) -1); + this->cpuXC->setIntReg(AlphaISA::ReturnValueReg, -return_value); } } @@ -208,9 +208,8 @@ class AlphaFullCPU : public FullO3CPU<Impl> { #if FULL_SYSTEM && defined(TARGET_ALPHA) if (req->flags & LOCKED) { - MiscRegFile *cregs = &req->xc->regs.miscRegs; - cregs->setReg(TheISA::Lock_Addr_DepTag, req->paddr); - cregs->setReg(TheISA::Lock_Flag_DepTag, true); + req->xc->setMiscReg(TheISA::Lock_Addr_DepTag, req->paddr); + req->xc->setMiscReg(TheISA::Lock_Flag_DepTag, true); } #endif @@ -230,34 +229,34 @@ class AlphaFullCPU : public FullO3CPU<Impl> Fault write(MemReqPtr &req, T &data) { #if FULL_SYSTEM && defined(TARGET_ALPHA) - - MiscRegFile *cregs; + ExecContext *xc; // If this is a store conditional, act appropriately if (req->flags & LOCKED) { - cregs = &req->xc->regs.miscRegs; + xc = req->xc; if (req->flags & UNCACHEABLE) { // Don't update result register (see stq_c in isa_desc) req->result = 2; - req->xc->storeCondFailures = 0;//Needed? [RGD] + xc->setStCondFailures(0);//Needed? [RGD] } else { - bool lock_flag = cregs->readReg(TheISA::Lock_Flag_DepTag); - Addr lock_addr = cregs->readReg(TheISA::Lock_Addr_DepTag); + 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))) { - cregs->setReg(TheISA::Lock_Flag_DepTag, false); - if (((++req->xc->storeCondFailures) % 100000) == 0) { + xc->setMiscReg(TheISA::Lock_Flag_DepTag, false); + xc->setStCondFailures(xc->readStCondFailures() + 1); + if (((xc->readStCondFailures()) % 100000) == 0) { std::cerr << "Warning: " - << req->xc->storeCondFailures + << xc->readStCondFailures() << " consecutive store conditional failures " - << "on cpu " << req->xc->cpu_id + << "on cpu " << req->xc->readCpuId() << std::endl; } return NoFault; } - else req->xc->storeCondFailures = 0; + else xc->setStCondFailures(0); } } @@ -267,10 +266,10 @@ class AlphaFullCPU : public FullO3CPU<Impl> // Conditionals would have returned above, and wouldn't fall // through. for (int i = 0; i < this->system->execContexts.size(); i++){ - cregs = &this->system->execContexts[i]->regs.miscRegs; - if ((cregs->readReg(TheISA::Lock_Addr_DepTag) & ~0xf) == + xc = this->system->execContexts[i]; + if ((xc->readMiscReg(TheISA::Lock_Addr_DepTag) & ~0xf) == (req->paddr & ~0xf)) { - cregs->setReg(TheISA::Lock_Flag_DepTag, false); + xc->setMiscReg(TheISA::Lock_Flag_DepTag, false); } } diff --git a/cpu/o3/alpha_cpu_builder.cc b/cpu/o3/alpha_cpu_builder.cc index 3547fb1b5..7e687ad2b 100644 --- a/cpu/o3/alpha_cpu_builder.cc +++ b/cpu/o3/alpha_cpu_builder.cc @@ -30,7 +30,6 @@ #include "base/loader/symtab.hh" #include "base/misc.hh" #include "cpu/base.hh" -#include "cpu/exec_context.hh" #include "cpu/exetrace.hh" #include "cpu/o3/alpha_cpu.hh" #include "cpu/o3/alpha_impl.hh" diff --git a/cpu/o3/alpha_cpu_impl.hh b/cpu/o3/alpha_cpu_impl.hh index bd4e34914..271b542ab 100644 --- a/cpu/o3/alpha_cpu_impl.hh +++ b/cpu/o3/alpha_cpu_impl.hh @@ -165,7 +165,7 @@ AlphaFullCPU<Impl>::copyToXC() for (int i = 0; i < AlphaISA::NumIntRegs; ++i) { renamed_reg = this->renameMap.lookup(i); - this->xc->regs.intRegFile[i] = this->regFile.readIntReg(renamed_reg); + this->cpuXC->setIntReg(i, this->regFile.readIntReg(renamed_reg)); DPRINTF(FullCPU, "FullCPU: Copying register %i, has data %lli.\n", renamed_reg, this->regFile.intRegFile[renamed_reg]); } @@ -174,21 +174,23 @@ AlphaFullCPU<Impl>::copyToXC() for (int i = 0; i < AlphaISA::NumFloatRegs; ++i) { renamed_reg = this->renameMap.lookup(i + AlphaISA::FP_Base_DepTag); - this->xc->regs.floatRegFile.d[i] = - this->regFile.readFloatRegDouble(renamed_reg); - this->xc->regs.floatRegFile.q[i] = - this->regFile.readFloatRegInt(renamed_reg); + this->cpuXC->setFloatRegDouble(i, + this->regFile.readFloatRegDouble(renamed_reg)); + this->cpuXC->setFloatRegInt(i, + this->regFile.readFloatRegInt(renamed_reg)); } /* - this->xc->regs.miscRegs.fpcr = this->regFile.miscRegs.fpcr; - this->xc->regs.miscRegs.uniq = this->regFile.miscRegs.uniq; - this->xc->regs.miscRegs.lock_flag = this->regFile.miscRegs.lock_flag; - this->xc->regs.miscRegs.lock_addr = this->regFile.miscRegs.lock_addr; + this->cpuXC->regs.miscRegs.fpcr = this->regFile.miscRegs.fpcr; + this->cpuXC->regs.miscRegs.uniq = this->regFile.miscRegs.uniq; + this->cpuXC->regs.miscRegs.lock_flag = this->regFile.miscRegs.lock_flag; + this->cpuXC->regs.miscRegs.lock_addr = this->regFile.miscRegs.lock_addr; */ - this->xc->regs.pc = this->rob.readHeadPC(); - this->xc->regs.npc = this->xc->regs.pc+4; + this->cpuXC->setPC(this->rob.readHeadPC()); + this->cpuXC->setNextPC(this->cpuXC->readPC()+4); - this->xc->func_exe_inst = this->funcExeInst; +#if !FULL_SYSTEM + this->cpuXC->setFuncExeInst(this->funcExeInst); +#endif } // This function will probably mess things up unless the ROB is empty and @@ -207,9 +209,9 @@ AlphaFullCPU<Impl>::copyFromXC() DPRINTF(FullCPU, "FullCPU: Copying over register %i, had data %lli, " "now has data %lli.\n", renamed_reg, this->regFile.intRegFile[renamed_reg], - this->xc->regs.intRegFile[i]); + this->cpuXC->readIntReg(i)); - this->regFile.setIntReg(renamed_reg, this->xc->regs.intRegFile[i]); + this->regFile.setIntReg(renamed_reg, this->cpuXC->readIntReg(i)); } // Then loop through the floating point registers. @@ -217,22 +219,23 @@ AlphaFullCPU<Impl>::copyFromXC() { renamed_reg = this->renameMap.lookup(i + AlphaISA::FP_Base_DepTag); this->regFile.setFloatRegDouble(renamed_reg, - this->xc->regs.floatRegFile.d[i]); + this->cpuXC->readFloatRegDouble(i)); this->regFile.setFloatRegInt(renamed_reg, - this->xc->regs.floatRegFile.q[i]); + this->cpuXC->readFloatRegInt(i)); } /* // Then loop through the misc registers. - this->regFile.miscRegs.fpcr = this->xc->regs.miscRegs.fpcr; - this->regFile.miscRegs.uniq = this->xc->regs.miscRegs.uniq; - this->regFile.miscRegs.lock_flag = this->xc->regs.miscRegs.lock_flag; - this->regFile.miscRegs.lock_addr = this->xc->regs.miscRegs.lock_addr; + this->regFile.miscRegs.fpcr = this->cpuXC->regs.miscRegs.fpcr; + this->regFile.miscRegs.uniq = this->cpuXC->regs.miscRegs.uniq; + this->regFile.miscRegs.lock_flag = this->cpuXC->regs.miscRegs.lock_flag; + this->regFile.miscRegs.lock_addr = this->cpuXC->regs.miscRegs.lock_addr; */ // Then finally set the PC and the next PC. -// regFile.pc = xc->regs.pc; -// regFile.npc = xc->regs.npc; - - this->funcExeInst = this->xc->func_exe_inst; +// regFile.pc = cpuXC->regs.pc; +// regFile.npc = cpuXC->regs.npc; +#if !FULL_SYSTEM + this->funcExeInst = this->cpuXC->readFuncExeInst(); +#endif } #if FULL_SYSTEM diff --git a/cpu/o3/cpu.cc b/cpu/o3/cpu.cc index a8c620028..62d68bb33 100644 --- a/cpu/o3/cpu.cc +++ b/cpu/o3/cpu.cc @@ -35,10 +35,11 @@ #endif #include "sim/root.hh" +#include "cpu/cpu_exec_context.hh" +#include "cpu/exec_context.hh" #include "cpu/o3/alpha_dyn_inst.hh" #include "cpu/o3/alpha_impl.hh" #include "cpu/o3/cpu.hh" -#include "cpu/exec_context.hh" using namespace std; @@ -103,7 +104,7 @@ FullO3CPU<Impl>::FullO3CPU(Params ¶ms) renameQueue(5, 5), iewQueue(5, 5), - xc(NULL), + cpuXC(NULL), globalSeqNum(1), @@ -134,8 +135,8 @@ FullO3CPU<Impl>::FullO3CPU(Params ¶ms) for (int i = 0; i < this->number_of_threads; ++i) { #if FULL_SYSTEM assert(i == 0); - system->execContexts[i] = - new ExecContext(this, i, system, itb, dtb, mem); + thread[i] = new CPUExecContext(this, 0, system, itb, dtb, mem); + system->execContexts[i] = thread[i]->getProxy(); execContexts.push_back(system->execContexts[i]); #else @@ -143,21 +144,17 @@ FullO3CPU<Impl>::FullO3CPU(Params ¶ms) DPRINTF(FullCPU, "FullCPU: Workload[%i]'s starting PC is %#x, " "process is %#x", i, params.workload[i]->prog_entry, thread[i]); - thread[i] = new ExecContext(this, i, params.workload[i], i); + thread[i] = new CPUExecContext(this, i, params.workload[i], i); } assert(params.workload[i]->getMemory() != NULL); assert(mem != NULL); - execContexts.push_back(thread[i]); + execContexts.push_back(thread[i]->getProxy()); #endif // !FULL_SYSTEM } // Note that this is a hack so that my code which still uses xc-> will // still work. I should remove this eventually -#if FULL_SYSTEM - xc = system->execContexts[0]; -#else - xc = thread[0]; -#endif + cpuXC = thread[0]; // The stages also need their CPU pointer setup. However this must be // done at the upper level CPU because they have pointers to the upper @@ -248,21 +245,21 @@ FullO3CPU<Impl>::init() // that it can start properly. #if FULL_SYSTEM ExecContext *src_xc = system->execContexts[0]; - TheISA::initCPU(&src_xc->regs, src_xc->cpu_id); + TheISA::initCPU(src_xc, src_xc->readCpuId()); #else - ExecContext *src_xc = thread[0]; + ExecContext *src_xc = thread[0]->getProxy(); #endif // First loop through the integer registers. for (int i = 0; i < TheISA::NumIntRegs; ++i) { - regFile.intRegFile[i] = src_xc->regs.intRegFile[i]; + regFile.intRegFile[i] = src_xc->readIntReg(i); } // Then loop through the floating point registers. for (int i = 0; i < TheISA::NumFloatRegs; ++i) { - regFile.floatRegFile[i].d = src_xc->regs.floatRegFile.d[i]; - regFile.floatRegFile[i].q = src_xc->regs.floatRegFile.q[i]; + regFile.floatRegFile[i].d = src_xc->readFloatRegDouble(i); + regFile.floatRegFile[i].q = src_xc->readFloatRegInt(i); } /* // Then loop through the misc registers. @@ -272,8 +269,8 @@ FullO3CPU<Impl>::init() regFile.miscRegs.lock_addr = src_xc->regs.miscRegs.lock_addr; */ // Then finally set the PC and the next PC. - regFile.pc = src_xc->regs.pc; - regFile.npc = src_xc->regs.npc; + regFile.pc = src_xc->readPC(); + regFile.npc = src_xc->readNextPC(); } } diff --git a/cpu/o3/cpu.hh b/cpu/o3/cpu.hh index 09d9c3d66..45e21db7f 100644 --- a/cpu/o3/cpu.hh +++ b/cpu/o3/cpu.hh @@ -44,9 +44,9 @@ #include "base/timebuf.hh" #include "config/full_system.hh" #include "cpu/base.hh" +#include "cpu/cpu_exec_context.hh" #include "cpu/o3/comm.hh" #include "cpu/o3/cpu_policy.hh" -#include "cpu/exec_context.hh" #include "sim/process.hh" #if FULL_SYSTEM @@ -54,6 +54,7 @@ using namespace EV5; #endif +class ExecContext; class FunctionalMemory; class Process; @@ -164,8 +165,8 @@ class FullO3CPU : public BaseFullCPU bool validDataAddr(Addr addr) { return thread[0]->validDataAddr(addr); } - int getInstAsid() { return thread[0]->asid; } - int getDataAsid() { return thread[0]->asid; } + int getInstAsid() { return thread[0]->getInstAsid(); } + int getDataAsid() { return thread[0]->getDataAsid(); } #endif @@ -320,16 +321,17 @@ class FullO3CPU : public BaseFullCPU public: /** The temporary exec context to support older accessors. */ - ExecContext *xc; + CPUExecContext *cpuXC; /** Temporary function to get pointer to exec context. */ ExecContext *xcBase() { -#if FULL_SYSTEM - return system->execContexts[0]; -#else + return thread[0]->getProxy(); + } + + CPUExecContext *cpuXCBase() + { return thread[0]; -#endif } InstSeqNum globalSeqNum; @@ -344,9 +346,8 @@ class FullO3CPU : public BaseFullCPU AlphaDTB *dtb; // SWContext *swCtx; -#else - std::vector<ExecContext *> thread; #endif + std::vector<CPUExecContext *> thread; FunctionalMemory *mem; diff --git a/cpu/pc_event.cc b/cpu/pc_event.cc index 83fbc3e2d..050bf1a88 100644 --- a/cpu/pc_event.cc +++ b/cpu/pc_event.cc @@ -38,6 +38,7 @@ #include "cpu/pc_event.hh" #include "sim/debug.hh" #include "sim/root.hh" +#include "sim/system.hh" using namespace std; @@ -79,7 +80,7 @@ PCEventQueue::schedule(PCEvent *event) bool PCEventQueue::doService(ExecContext *xc) { - Addr pc = xc->regs.pc & ~0x3; + Addr pc = xc->readPC() & ~0x3; int serviced = 0; range_t range = equal_range(pc); for (iterator i = range.first; i != range.second; ++i) { @@ -87,7 +88,7 @@ PCEventQueue::doService(ExecContext *xc) // another event. This for example, prevents two invocations // of the SkipFuncEvent. Maybe we should have separate PC // event queues for each processor? - if (pc != (xc->regs.pc & ~0x3)) + if (pc != (xc->readPC() & ~0x3)) continue; DPRINTF(PCEvent, "PC based event serviced at %#x: %s\n", @@ -126,7 +127,7 @@ BreakPCEvent::BreakPCEvent(PCEventQueue *q, const std::string &desc, Addr addr, void BreakPCEvent::process(ExecContext *xc) { - StringWrap name(xc->cpu->name() + ".break_event"); + StringWrap name(xc->getCpuPtr()->name() + ".break_event"); DPRINTFN("break event %s triggered\n", descr()); debug_break(); if (remove) diff --git a/cpu/simple/cpu.cc b/cpu/simple/cpu.cc index dd2d53c17..7c60b242f 100644 --- a/cpu/simple/cpu.cc +++ b/cpu/simple/cpu.cc @@ -44,6 +44,7 @@ #include "base/stats/events.hh" #include "base/trace.hh" #include "cpu/base.hh" +#include "cpu/cpu_exec_context.hh" #include "cpu/exec_context.hh" #include "cpu/exetrace.hh" #include "cpu/profile.hh" @@ -94,7 +95,7 @@ SimpleCPU::init() ExecContext *xc = execContexts[i]; // initialize CPU, including PC - TheISA::initCPU(&xc->regs, xc->cpu_id); + TheISA::initCPU(xc, xc->readCpuId()); } #endif } @@ -132,22 +133,24 @@ SimpleCPU::CacheCompletionEvent::description() } SimpleCPU::SimpleCPU(Params *p) - : BaseCPU(p), tickEvent(this, p->width), xc(NULL), + : BaseCPU(p), tickEvent(this, p->width), cpuXC(NULL), cacheCompletionEvent(this) { _status = Idle; #if FULL_SYSTEM - xc = new ExecContext(this, 0, p->system, p->itb, p->dtb, p->mem); + cpuXC = new CPUExecContext(this, 0, p->system, p->itb, p->dtb, p->mem); #else - xc = new ExecContext(this, /* thread_num */ 0, p->process, /* asid */ 0); + cpuXC = new CPUExecContext(this, /* thread_num */ 0, p->process, + /* asid */ 0); #endif // !FULL_SYSTEM + xcProxy = cpuXC->getProxy(); icacheInterface = p->icache_interface; dcacheInterface = p->dcache_interface; memReq = new MemReq(); - memReq->xc = xc; + memReq->xc = xcProxy; memReq->asid = 0; memReq->data = new uint8_t[64]; @@ -158,7 +161,7 @@ SimpleCPU::SimpleCPU(Params *p) lastIcacheStall = 0; lastDcacheStall = 0; - execContexts.push_back(xc); + execContexts.push_back(xcProxy); } SimpleCPU::~SimpleCPU() @@ -207,7 +210,7 @@ void SimpleCPU::activateContext(int thread_num, int delay) { assert(thread_num == 0); - assert(xc); + assert(cpuXC); assert(_status == Idle); notIdleFraction++; @@ -220,7 +223,7 @@ void SimpleCPU::suspendContext(int thread_num) { assert(thread_num == 0); - assert(xc); + assert(cpuXC); assert(_status == Running); notIdleFraction--; @@ -301,7 +304,7 @@ SimpleCPU::serialize(ostream &os) SERIALIZE_ENUM(_status); SERIALIZE_SCALAR(inst); nameOut(os, csprintf("%s.xc", name())); - xc->serialize(os); + cpuXC->serialize(os); nameOut(os, csprintf("%s.tickEvent", name())); tickEvent.serialize(os); nameOut(os, csprintf("%s.cacheCompletionEvent", name())); @@ -314,7 +317,7 @@ SimpleCPU::unserialize(Checkpoint *cp, const string §ion) BaseCPU::unserialize(cp, section); UNSERIALIZE_ENUM(_status); UNSERIALIZE_SCALAR(inst); - xc->unserialize(cp, csprintf("%s.xc", section)); + cpuXC->unserialize(cp, csprintf("%s.xc", section)); tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); cacheCompletionEvent .unserialize(cp, csprintf("%s.cacheCompletionEvent", section)); @@ -345,16 +348,16 @@ SimpleCPU::copySrcTranslate(Addr src) memReq->reset(src & ~(blk_size - 1), blk_size); // translate to physical address - Fault fault = xc->translateDataReadReq(memReq); + Fault fault = cpuXC->translateDataReadReq(memReq); assert(fault != AlignmentFault); if (fault == NoFault) { - xc->copySrcAddr = src; - xc->copySrcPhysAddr = memReq->paddr + offset; + cpuXC->copySrcAddr = src; + cpuXC->copySrcPhysAddr = memReq->paddr + offset; } else { - xc->copySrcAddr = 0; - xc->copySrcPhysAddr = 0; + cpuXC->copySrcAddr = 0; + cpuXC->copySrcPhysAddr = 0; } return fault; } @@ -367,7 +370,7 @@ SimpleCPU::copy(Addr dest) // Only support block sizes of 64 atm. assert(blk_size == 64); uint8_t data[blk_size]; - //assert(xc->copySrcAddr); + //assert(cpuXC->copySrcAddr); int offset = dest & (blk_size - 1); // Make sure block doesn't span page @@ -380,21 +383,21 @@ SimpleCPU::copy(Addr dest) memReq->reset(dest & ~(blk_size -1), blk_size); // translate to physical address - Fault fault = xc->translateDataWriteReq(memReq); + Fault fault = cpuXC->translateDataWriteReq(memReq); assert(fault != AlignmentFault); 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 = cpuXC->copySrcPhysAddr; + cpuXC->mem->read(memReq, data); memReq->paddr = dest_addr; - xc->mem->write(memReq, data); + cpuXC->mem->write(memReq, data); if (dcacheInterface) { memReq->cmd = Copy; memReq->completionEvent = NULL; - memReq->paddr = xc->copySrcPhysAddr; + memReq->paddr = cpuXC->copySrcPhysAddr; memReq->dest = dest_addr; memReq->size = 64; memReq->time = curTick; @@ -411,7 +414,7 @@ Fault SimpleCPU::read(Addr addr, T &data, unsigned flags) { if (status() == DcacheMissStall || status() == DcacheMissSwitch) { - Fault fault = xc->read(memReq,data); + Fault fault = cpuXC->read(memReq,data); if (traceData) { traceData->setAddr(addr); @@ -422,7 +425,7 @@ SimpleCPU::read(Addr addr, T &data, unsigned flags) memReq->reset(addr, sizeof(T), flags); // translate to physical address - Fault fault = xc->translateDataReadReq(memReq); + Fault fault = cpuXC->translateDataReadReq(memReq); // if we have a cache, do cache access too if (fault == NoFault && dcacheInterface) { @@ -442,12 +445,12 @@ SimpleCPU::read(Addr addr, T &data, unsigned flags) _status = DcacheMissStall; } else { // do functional access - fault = xc->read(memReq, data); + fault = cpuXC->read(memReq, data); } } else if(fault == NoFault) { // do functional access - fault = xc->read(memReq, data); + fault = cpuXC->read(memReq, data); } @@ -507,11 +510,11 @@ SimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res) memReq->reset(addr, sizeof(T), flags); // translate to physical address - Fault fault = xc->translateDataWriteReq(memReq); + Fault fault = cpuXC->translateDataWriteReq(memReq); // do functional access if (fault == NoFault) - fault = xc->write(memReq, data); + fault = cpuXC->write(memReq, data); if (fault == NoFault && dcacheInterface) { memReq->cmd = Write; @@ -588,7 +591,7 @@ SimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res) Addr SimpleCPU::dbg_vtophys(Addr addr) { - return vtophys(xc, addr); + return vtophys(xcProxy, addr); } #endif // FULL_SYSTEM @@ -636,9 +639,9 @@ SimpleCPU::post_interrupt(int int_num, int index) { BaseCPU::post_interrupt(int_num, index); - if (xc->status() == ExecContext::Suspended) { + if (cpuXC->status() == ExecContext::Suspended) { DPRINTF(IPI,"Suspended Processor awoke\n"); - xc->activate(); + cpuXC->activate(); } } #endif // FULL_SYSTEM @@ -654,16 +657,16 @@ SimpleCPU::tick() Fault fault = NoFault; #if FULL_SYSTEM - if (checkInterrupts && check_interrupts() && !xc->inPalMode() && + if (checkInterrupts && check_interrupts() && !cpuXC->inPalMode() && status() != IcacheMissComplete) { int ipl = 0; int summary = 0; checkInterrupts = false; - if (xc->readMiscReg(IPR_SIRR)) { + if (cpuXC->readMiscReg(IPR_SIRR)) { for (int i = INTLEVEL_SOFTWARE_MIN; i < INTLEVEL_SOFTWARE_MAX; i++) { - if (xc->readMiscReg(IPR_SIRR) & (ULL(1) << i)) { + if (cpuXC->readMiscReg(IPR_SIRR) & (ULL(1) << i)) { // See table 4-19 of 21164 hardware reference ipl = (i - INTLEVEL_SOFTWARE_MIN) + 1; summary |= (ULL(1) << i); @@ -671,7 +674,7 @@ SimpleCPU::tick() } } - uint64_t interrupts = xc->cpu->intr_status(); + uint64_t interrupts = cpuXC->cpu->intr_status(); for (int i = INTLEVEL_EXTERNAL_MIN; i < INTLEVEL_EXTERNAL_MAX; i++) { if (interrupts & (ULL(1) << i)) { @@ -681,24 +684,24 @@ SimpleCPU::tick() } } - if (xc->readMiscReg(IPR_ASTRR)) + if (cpuXC->readMiscReg(IPR_ASTRR)) panic("asynchronous traps not implemented\n"); - if (ipl && ipl > xc->readMiscReg(IPR_IPLR)) { - xc->setMiscReg(IPR_ISR, summary); - xc->setMiscReg(IPR_INTID, ipl); - xc->ev5_trap(InterruptFault); + if (ipl && ipl > cpuXC->readMiscReg(IPR_IPLR)) { + cpuXC->setMiscReg(IPR_ISR, summary); + cpuXC->setMiscReg(IPR_INTID, ipl); + cpuXC->ev5_trap(InterruptFault); DPRINTF(Flow, "Interrupt! IPLR=%d ipl=%d summary=%x\n", - xc->readMiscReg(IPR_IPLR), ipl, summary); + cpuXC->readMiscReg(IPR_IPLR), ipl, summary); } } #endif // maintain $r0 semantics - xc->regs.intRegFile[ZeroReg] = 0; + cpuXC->setIntReg(ZeroReg, 0); #ifdef TARGET_ALPHA - xc->regs.floatRegFile.d[ZeroReg] = 0.0; + cpuXC->setFloatRegDouble(ZeroReg, 0.0); #endif // TARGET_ALPHA if (status() == IcacheMissComplete) { @@ -720,13 +723,13 @@ SimpleCPU::tick() #endif memReq->cmd = Read; - memReq->reset(xc->regs.pc & ~3, sizeof(uint32_t), - IFETCH_FLAGS(xc->regs.pc)); + memReq->reset(cpuXC->readPC() & ~3, sizeof(uint32_t), + IFETCH_FLAGS(cpuXC->readPC())); - fault = xc->translateInstReq(memReq); + fault = cpuXC->translateInstReq(memReq); if (fault == NoFault) - fault = xc->mem->read(memReq, inst); + fault = cpuXC->mem->read(memReq, inst); if (icacheInterface && fault == NoFault) { memReq->completionEvent = NULL; @@ -763,29 +766,30 @@ SimpleCPU::tick() inst = gtoh(inst); curStaticInst = StaticInst::decode(inst); - traceData = Trace::getInstRecord(curTick, xc, this, curStaticInst, - xc->regs.pc); + traceData = Trace::getInstRecord(curTick, xcProxy, this, curStaticInst, + cpuXC->readPC()); #if FULL_SYSTEM - xc->setInst(inst); + cpuXC->setInst(inst); #endif // FULL_SYSTEM - xc->func_exe_inst++; + cpuXC->func_exe_inst++; fault = curStaticInst->execute(this, traceData); #if FULL_SYSTEM - if (xc->fnbin) { - assert(xc->kernelStats); - system->kernelBinning->execute(xc, inst); + if (system->kernelBinning->fnbin) { + assert(kernelStats); + system->kernelBinning->execute(xcProxy, inst); } - if (xc->profile) { - bool usermode = (xc->readMiscReg(AlphaISA::IPR_DTB_CM) & 0x18) != 0; - xc->profilePC = usermode ? 1 : xc->regs.pc; - ProfileNode *node = xc->profile->consume(xc, inst); + if (cpuXC->profile) { + bool usermode = + (cpuXC->readMiscReg(AlphaISA::IPR_DTB_CM) & 0x18) != 0; + cpuXC->profilePC = usermode ? 1 : cpuXC->readPC(); + ProfileNode *node = cpuXC->profile->consume(xcProxy, inst); if (node) - xc->profileNode = node; + cpuXC->profileNode = node; } #endif @@ -805,29 +809,29 @@ SimpleCPU::tick() traceData->finalize(); } - traceFunctions(xc->regs.pc); + traceFunctions(cpuXC->readPC()); } // if (fault == NoFault) if (fault != NoFault) { #if FULL_SYSTEM - xc->ev5_trap(fault); + cpuXC->ev5_trap(fault); #else // !FULL_SYSTEM - fatal("fault (%d) detected @ PC 0x%08p", fault, xc->regs.pc); + fatal("fault (%d) detected @ PC 0x%08p", fault, cpuXC->readPC()); #endif // FULL_SYSTEM } else { // go to the next instruction - xc->regs.pc = xc->regs.npc; - xc->regs.npc += sizeof(MachInst); + cpuXC->setPC(cpuXC->readNextPC()); + cpuXC->setNextPC(cpuXC->readNextPC() + sizeof(MachInst)); } #if FULL_SYSTEM Addr oldpc; do { - oldpc = xc->regs.pc; - system->pcEventQueue.service(xc); - } while (oldpc != xc->regs.pc); + oldpc = cpuXC->readPC(); + system->pcEventQueue.service(xcProxy); + } while (oldpc != cpuXC->readPC()); #endif assert(status() == Running || diff --git a/cpu/simple/cpu.hh b/cpu/simple/cpu.hh index 3bc905be1..1441a8fcd 100644 --- a/cpu/simple/cpu.hh +++ b/cpu/simple/cpu.hh @@ -32,7 +32,7 @@ #include "base/statistics.hh" #include "config/full_system.hh" #include "cpu/base.hh" -#include "cpu/exec_context.hh" +#include "cpu/cpu_exec_context.hh" #include "cpu/pc_event.hh" #include "cpu/sampler/sampler.hh" #include "cpu/static_inst.hh" @@ -54,6 +54,7 @@ class Process; #endif // FULL_SYSTEM +class ExecContext; class MemInterface; class Checkpoint; @@ -148,7 +149,9 @@ class SimpleCPU : public BaseCPU public: // execution context - ExecContext *xc; + CPUExecContext *cpuXC; + + ExecContext *xcProxy; void switchOut(Sampler *s); void takeOverFrom(BaseCPU *oldCPU); @@ -275,86 +278,86 @@ class SimpleCPU : public BaseCPU uint64_t readIntReg(const StaticInst *si, int idx) { - return xc->readIntReg(si->srcRegIdx(idx)); + return cpuXC->readIntReg(si->srcRegIdx(idx)); } float readFloatRegSingle(const StaticInst *si, int idx) { int reg_idx = si->srcRegIdx(idx) - TheISA::FP_Base_DepTag; - return xc->readFloatRegSingle(reg_idx); + return cpuXC->readFloatRegSingle(reg_idx); } double readFloatRegDouble(const StaticInst *si, int idx) { int reg_idx = si->srcRegIdx(idx) - TheISA::FP_Base_DepTag; - return xc->readFloatRegDouble(reg_idx); + return cpuXC->readFloatRegDouble(reg_idx); } uint64_t readFloatRegInt(const StaticInst *si, int idx) { int reg_idx = si->srcRegIdx(idx) - TheISA::FP_Base_DepTag; - return xc->readFloatRegInt(reg_idx); + return cpuXC->readFloatRegInt(reg_idx); } void setIntReg(const StaticInst *si, int idx, uint64_t val) { - xc->setIntReg(si->destRegIdx(idx), val); + cpuXC->setIntReg(si->destRegIdx(idx), val); } void setFloatRegSingle(const StaticInst *si, int idx, float val) { int reg_idx = si->destRegIdx(idx) - TheISA::FP_Base_DepTag; - xc->setFloatRegSingle(reg_idx, val); + cpuXC->setFloatRegSingle(reg_idx, val); } void setFloatRegDouble(const StaticInst *si, int idx, double val) { int reg_idx = si->destRegIdx(idx) - TheISA::FP_Base_DepTag; - xc->setFloatRegDouble(reg_idx, val); + cpuXC->setFloatRegDouble(reg_idx, val); } void setFloatRegInt(const StaticInst *si, int idx, uint64_t val) { int reg_idx = si->destRegIdx(idx) - TheISA::FP_Base_DepTag; - xc->setFloatRegInt(reg_idx, val); + cpuXC->setFloatRegInt(reg_idx, val); } - uint64_t readPC() { return xc->readPC(); } - void setNextPC(uint64_t val) { xc->setNextPC(val); } + uint64_t readPC() { return cpuXC->readPC(); } + void setNextPC(uint64_t val) { cpuXC->setNextPC(val); } MiscReg readMiscReg(int misc_reg) { - return xc->readMiscReg(misc_reg); + return cpuXC->readMiscReg(misc_reg); } MiscReg readMiscRegWithEffect(int misc_reg, Fault &fault) { - return xc->readMiscRegWithEffect(misc_reg, fault); + return cpuXC->readMiscRegWithEffect(misc_reg, fault); } Fault setMiscReg(int misc_reg, const MiscReg &val) { - return xc->setMiscReg(misc_reg, val); + return cpuXC->setMiscReg(misc_reg, val); } Fault setMiscRegWithEffect(int misc_reg, const MiscReg &val) { - return xc->setMiscRegWithEffect(misc_reg, val); + return cpuXC->setMiscRegWithEffect(misc_reg, val); } #if FULL_SYSTEM - Fault hwrei() { return xc->hwrei(); } - int readIntrFlag() { return xc->readIntrFlag(); } - void setIntrFlag(int val) { xc->setIntrFlag(val); } - bool inPalMode() { return xc->inPalMode(); } - void ev5_trap(Fault fault) { xc->ev5_trap(fault); } - bool simPalCheck(int palFunc) { return xc->simPalCheck(palFunc); } + Fault hwrei() { return cpuXC->hwrei(); } + int readIntrFlag() { return cpuXC->readIntrFlag(); } + void setIntrFlag(int val) { cpuXC->setIntrFlag(val); } + bool inPalMode() { return cpuXC->inPalMode(); } + void ev5_trap(Fault fault) { cpuXC->ev5_trap(fault); } + bool simPalCheck(int palFunc) { return cpuXC->simPalCheck(palFunc); } #else - void syscall() { xc->syscall(); } + void syscall() { cpuXC->syscall(); } #endif - bool misspeculating() { return xc->misspeculating(); } - ExecContext *xcBase() { return xc; } + bool misspeculating() { return cpuXC->misspeculating(); } + ExecContext *xcBase() { return xcProxy; } }; #endif // __CPU_SIMPLE_CPU_SIMPLE_CPU_HH__ |