cpu.cc

Go to the documentation of this file.

/*
 * Copyright (c) 2011-2012, 2014, 2016, 2017, 2019-2020, 2024 Arm Limited
 * Copyright (c) 2013 Advanced Micro Devices, Inc.
 * Copyright (c) 2022-2023 The University of Edinburgh
 * All rights reserved
 *
 * The license below extends only to copyright in the software and shall
 * not be construed as granting a license to any other intellectual
 * property including but not limited to intellectual property relating
 * to a hardware implementation of the functionality of the software
 * licensed hereunder.  You may use the software subject to the license
 * terms below provided that you ensure that this notice is replicated
 * unmodified and in its entirety in all distributions of the software,
 * modified or unmodified, in source code or in binary form.
 *
 * Copyright (c) 2004-2006 The Regents of The University of Michigan
 * Copyright (c) 2011 Regents of the University of California
 * 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.
 */
 
#include "cpu/o3/cpu.hh"
 
#include "cpu/activity.hh"
#include "cpu/checker/cpu.hh"
#include "cpu/checker/thread_context.hh"
#include "cpu/o3/dyn_inst.hh"
#include "cpu/o3/limits.hh"
#include "cpu/o3/thread_context.hh"
#include "cpu/simple_thread.hh"
#include "cpu/thread_context.hh"
#include "debug/Activity.hh"
#include "debug/Drain.hh"
#include "debug/O3CPU.hh"
#include "debug/Quiesce.hh"
#include "enums/MemoryMode.hh"
#include "sim/cur_tick.hh"
#include "sim/full_system.hh"
#include "sim/process.hh"
#include "sim/stat_control.hh"
#include "sim/system.hh"
 
namespace gem5
{
 
struct BaseCPUParams;
 
namespace o3
{
 
CPU::CPU(const BaseO3CPUParams &params)
    : BaseCPU(params),
      mmu(params.mmu),
      tickEvent([this] { tick(); }, "O3CPU tick", false, Event::CPU_Tick_Pri),
      threadExitEvent([this] { exitThreads(); }, "O3CPU exit threads", false,
                      Event::CPU_Exit_Pri),
#ifndef NDEBUG
      instcount(0),
#endif
      removeInstsThisCycle(false),
      bac(this, params),
      ftq(this, params),
      fetch(this, params),
      decode(this, params),
      rename(this, params),
      iew(this, params),
      commit(this, params),
 
      regFile(params.numPhysIntRegs, params.numPhysFloatRegs,
              params.numPhysVecRegs, params.numPhysVecPredRegs,
              params.numPhysMatRegs, params.numPhysCCRegs,
              params.isa[0]->regClasses()),
 
      freeList(name() + ".freelist", &regFile),
 
      rob(this, params),
 
      scoreboard(name() + ".scoreboard", regFile.totalNumPhysRegs()),
 
      isa(numThreads, NULL),
 
      timeBuffer(params.backComSize, params.forwardComSize),
      fetchQueue(params.backComSize, params.forwardComSize),
      decodeQueue(params.backComSize, params.forwardComSize),
      renameQueue(params.backComSize, params.forwardComSize),
      iewQueue(params.backComSize, params.forwardComSize),
      activityRec(name(), NumStages,
                  params.backComSize + params.forwardComSize, params.activity),
 
      globalSeqNum(1),
      globalFTSeqNum(1),
      system(params.system),
      lastRunningCycle(curCycle()),
      cpuStats(this)
{
    fatal_if(FullSystem && params.numThreads > 1,
            "SMT is not supported in O3 in full system mode currently.");
 
    fatal_if(!FullSystem && params.numThreads < params.workload.size(),
            "More workload items (%d) than threads (%d) on CPU %s.",
            params.workload.size(), params.numThreads, name());
 
    if (!params.switched_out) {
        _status = Running;
    } else {
        _status = SwitchedOut;
    }
 
    if (params.checker) {
        BaseCPU *temp_checker = params.checker;
        checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker);
        checker->setIcachePort(&fetch.getInstPort());
        checker->setSystem(params.system);
    } else {
        checker = NULL;
    }
 
    if (!FullSystem) {
        thread.resize(numThreads);
        tids.resize(numThreads);
    }
 
    // 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 level CPU, and not this CPU.
 
    // Set up Pointers to the activeThreads list for each stage
    bac.setActiveThreads(&activeThreads);
    fetch.setActiveThreads(&activeThreads);
    decode.setActiveThreads(&activeThreads);
    rename.setActiveThreads(&activeThreads);
    iew.setActiveThreads(&activeThreads);
    commit.setActiveThreads(&activeThreads);
 
    // Give each of the stages the time buffer they will use.
    bac.setTimeBuffer(&timeBuffer);
    fetch.setTimeBuffer(&timeBuffer);
    decode.setTimeBuffer(&timeBuffer);
    rename.setTimeBuffer(&timeBuffer);
    iew.setTimeBuffer(&timeBuffer);
    commit.setTimeBuffer(&timeBuffer);
 
    // Also setup each of the stages' queues.
    bac.setFetchTargetQueue(&ftq);
    fetch.setBACandFTQPtr(&bac, &ftq);
    fetch.setFetchQueue(&fetchQueue);
    decode.setFetchQueue(&fetchQueue);
    commit.setFetchQueue(&fetchQueue);
    decode.setDecodeQueue(&decodeQueue);
    rename.setDecodeQueue(&decodeQueue);
    rename.setRenameQueue(&renameQueue);
    iew.setRenameQueue(&renameQueue);
    iew.setIEWQueue(&iewQueue);
    commit.setIEWQueue(&iewQueue);
    commit.setRenameQueue(&renameQueue);
 
    commit.setIEWStage(&iew);
    rename.setIEWStage(&iew);
    rename.setCommitStage(&commit);
 
    ThreadID active_threads;
    if (FullSystem) {
        active_threads = 1;
    } else {
        active_threads = params.workload.size();
 
        if (active_threads > MaxThreads) {
            panic("Workload Size too large. Increase the 'MaxThreads' "
                  "constant in cpu/o3/limits.hh or edit your workload size.");
        }
    }
 
    // Make Sure That this a Valid Architeture
    assert(numThreads);
    const auto &regClasses = params.isa[0]->regClasses();
 
    panic_if(params.numPhysIntRegs <=
            numThreads * regClasses.at(IntRegClass)->numRegs() &&
            regClasses.at(IntRegClass)->numRegs() != 0,
            "Not enough physical registers, consider increasing "
            "numPhysIntRegs\n");
    panic_if(params.numPhysFloatRegs <=
            numThreads * regClasses.at(FloatRegClass)->numRegs() &&
            regClasses.at(FloatRegClass)->numRegs() != 0,
            "Not enough physical registers, consider increasing "
            "numPhysFloatRegs\n");
    panic_if(params.numPhysVecRegs <=
            numThreads * regClasses.at(VecRegClass)->numRegs() &&
            regClasses.at(VecRegClass)->numRegs() != 0,
            "Not enough physical registers, consider increasing "
            "numPhysVecRegs\n");
    panic_if(params.numPhysVecPredRegs <=
            numThreads * regClasses.at(VecPredRegClass)->numRegs() &&
            regClasses.at(VecPredRegClass)->numRegs() != 0,
            "Not enough physical registers, consider increasing "
            "numPhysVecPredRegs\n");
    panic_if(params.numPhysMatRegs <=
            numThreads * regClasses.at(MatRegClass)->numRegs() &&
            regClasses.at(MatRegClass)->numRegs() != 0,
            "Not enough physical registers, consider increasing "
            "numPhysMatRegs\n");
    panic_if(params.numPhysCCRegs <=
            numThreads * regClasses.at(CCRegClass)->numRegs() &&
            regClasses.at(CCRegClass)->numRegs() != 0,
            "Not enough physical registers, consider increasing "
            "numPhysCCRegs\n");
 
    // Just make this a warning and go ahead anyway, to keep from having to
    // add checks everywhere.
    warn_if(regClasses.at(CCRegClass)->numRegs() == 0 &&
            params.numPhysCCRegs != 0,
            "Non-zero number of physical CC regs specified, even though\n"
            "    ISA does not use them.");
 
    rename.setScoreboard(&scoreboard);
    iew.setScoreboard(&scoreboard);
 
    // Setup the rename map for whichever stages need it.
    for (ThreadID tid = 0; tid < numThreads; tid++) {
        isa[tid] = params.isa[tid];
        commitRenameMap[tid].init(regClasses, &regFile, &freeList);
        renameMap[tid].init(regClasses, &regFile, &freeList);
    }
 
    // Initialize rename map to assign physical registers to the
    // architectural registers for active threads only.
    for (ThreadID tid = 0; tid < active_threads; tid++) {
        for (auto type = (RegClassType)0; type <= CCRegClass;
                type = (RegClassType)(type + 1)) {
            for (auto &id: *regClasses.at(type)) {
                // Note that we can't use the rename() method because we don't
                // want special treatment for the zero register at this point
                PhysRegIdPtr phys_reg = freeList.getReg(type);
                renameMap[tid].setEntry(id, phys_reg);
                commitRenameMap[tid].setEntry(id, phys_reg);
            }
        }
    }
 
    rename.setRenameMap(renameMap);
    commit.setRenameMap(commitRenameMap);
    rename.setFreeList(&freeList);
 
    // Setup the ROB for whichever stages need it.
    commit.setROB(&rob);
 
    lastActivatedCycle = 0;
 
    DPRINTF(O3CPU, "Creating O3CPU object.\n");
 
    // Setup any thread state.
    thread.resize(numThreads);
 
    for (ThreadID tid = 0; tid < numThreads; ++tid) {
        if (FullSystem) {
            // SMT is not supported in FS mode yet.
            assert(numThreads == 1);
            thread[tid] = new ThreadState(this, 0, NULL);
        } else {
            if (tid < params.workload.size()) {
                DPRINTF(O3CPU, "Workload[%i] process is %#x", tid,
                        thread[tid]);
                thread[tid] = new ThreadState(this, tid, params.workload[tid]);
            } else {
                //Allocate Empty thread so M5 can use later
                //when scheduling threads to CPU
                Process* dummy_proc = NULL;
 
                thread[tid] = new ThreadState(this, tid, dummy_proc);
            }
        }
 
        gem5::ThreadContext *tc;
 
        // Setup the TC that will serve as the interface to the threads/CPU.
        auto *o3_tc = new ThreadContext;
 
        tc = o3_tc;
 
        // If we're using a checker, then the TC should be the
        // CheckerThreadContext.
        if (params.checker) {
            tc = new CheckerThreadContext<ThreadContext>(o3_tc, checker);
        }
 
        o3_tc->cpu = this;
        o3_tc->thread = thread[tid];
 
        // Give the thread the TC.
        thread[tid]->tc = tc;
 
        // Add the TC to the CPU's list of TC's.
        threadContexts.push_back(tc);
    }
 
    // O3CPU always requires an interrupt controller.
    if (!params.switched_out && interrupts.empty()) {
        fatal("O3CPU %s has no interrupt controller.\n"
              "Ensure createInterruptController() is called.\n", name());
    }
}
 
void
CPU::regProbePoints()
{
    BaseCPU::regProbePoints();
 
    ppInstAccessComplete = new ProbePointArg<PacketPtr>(
            getProbeManager(), "InstAccessComplete");
    ppDataAccessComplete = new ProbePointArg<
        std::pair<DynInstPtr, PacketPtr>>(
                getProbeManager(), "DataAccessComplete");
 
    ftq.regProbePoints();
    bac.regProbePoints();
    fetch.regProbePoints();
    rename.regProbePoints();
    iew.regProbePoints();
    commit.regProbePoints();
}
 
CPU::CPUStats::CPUStats(CPU *cpu)
    : statistics::Group(cpu),
      ADD_STAT(timesIdled, statistics::units::Count::get(),
               "Number of times that the entire CPU went into an idle state "
               "and unscheduled itself"),
      ADD_STAT(idleCycles, statistics::units::Cycle::get(),
               "Total number of cycles that the CPU has spent unscheduled due "
               "to idling"),
      ADD_STAT(quiesceCycles, statistics::units::Cycle::get(),
               "Total number of cycles that CPU has spent quiesced or waiting "
               "for an interrupt")
{
    // Register any of the O3CPU's stats here.
    timesIdled
        .prereq(timesIdled);
 
    idleCycles
        .prereq(idleCycles);
 
    quiesceCycles
        .prereq(quiesceCycles);
}
 
void
CPU::tick()
{
    DPRINTF(O3CPU, "\n\nO3CPU: Ticking main, O3CPU.\n");
    assert(!switchedOut());
    assert(drainState() != DrainState::Drained);
 
    ++baseStats.numCycles;
    updateCycleCounters(BaseCPU::CPU_STATE_ON);
 
//    activity = false;
 
    //Tick each of the stages
    bac.tick();
 
    fetch.tick();
 
    decode.tick();
 
    rename.tick();
 
    iew.tick();
 
    commit.tick();
 
    // Now advance the time buffers
    timeBuffer.advance();
 
    fetchQueue.advance();
    decodeQueue.advance();
    renameQueue.advance();
    iewQueue.advance();
 
    activityRec.advance();
 
    if (removeInstsThisCycle) {
        cleanUpRemovedInsts();
    }
 
    if (!tickEvent.scheduled()) {
        if (_status == SwitchedOut) {
            DPRINTF(O3CPU, "Switched out!\n");
            // increment stat
            lastRunningCycle = curCycle();
        } else if (!activityRec.active() || _status == Idle) {
            DPRINTF(O3CPU, "Idle!\n");
            lastRunningCycle = curCycle();
            cpuStats.timesIdled++;
        } else {
            schedule(tickEvent, clockEdge(Cycles(1)));
            DPRINTF(O3CPU, "Scheduling next tick!\n");
        }
    }
 
    if (!FullSystem)
        updateThreadPriority();
 
    tryDrain();
}
 
void
CPU::init()
{
    BaseCPU::init();
 
    for (ThreadID tid = 0; tid < numThreads; ++tid) {
        // Set noSquashFromTC so that the CPU doesn't squash when initially
        // setting up registers.
        thread[tid]->noSquashFromTC = true;
    }
 
    // Clear noSquashFromTC.
    for (int tid = 0; tid < numThreads; ++tid)
        thread[tid]->noSquashFromTC = false;
 
    commit.setThreads(thread);
}
 
void
CPU::startup()
{
    BaseCPU::startup();
 
    bac.startupStage();
    fetch.startupStage();
    decode.startupStage();
    iew.startupStage();
    rename.startupStage();
    commit.startupStage();
}
 
void
CPU::activateThread(ThreadID tid)
{
    auto active_it = std::find(
            activeThreads.begin(), activeThreads.end(), tid);
 
    DPRINTF(O3CPU, "[tid:%i] Calling activate thread.\n", tid);
    assert(!switchedOut());
 
    if (active_it == activeThreads.end()) {
        DPRINTF(O3CPU, "[tid:%i] Adding to active threads list\n", tid);
 
        activeThreads.push_back(tid);
    }
}
 
void
CPU::deactivateThread(ThreadID tid)
{
    // hardware transactional memory
    // shouldn't deactivate thread in the middle of a transaction
    assert(!commit.executingHtmTransaction(tid));
 
    //Remove From Active List, if Active
    auto active_it = std::find(
            activeThreads.begin(), activeThreads.end(), tid);
 
    DPRINTF(O3CPU, "[tid:%i] Calling deactivate thread.\n", tid);
    assert(!switchedOut());
 
    if (active_it != activeThreads.end()) {
        DPRINTF(O3CPU,"[tid:%i] Removing from active threads list\n",
                tid);
        activeThreads.erase(active_it);
    }
 
    bac.deactivateThread(tid);
    fetch.deactivateThread(tid);
    commit.deactivateThread(tid);
}
 
Counter
CPU::totalInsts() const
{
    Counter total(0);
 
    ThreadID size = thread.size();
    for (ThreadID i = 0; i < size; i++)
        total += thread[i]->numInst;
 
    return total;
}
 
Counter
CPU::totalOps() const
{
    Counter total(0);
 
    ThreadID size = thread.size();
    for (ThreadID i = 0; i < size; i++)
        total += thread[i]->numOp;
 
    return total;
}
 
void
CPU::activateContext(ThreadID tid)
{
    assert(!switchedOut());
 
    // Needs to set each stage to running as well.
    activateThread(tid);
 
    // We don't want to wake the CPU if it is drained. In that case,
    // we just want to flag the thread as active and schedule the tick
    // event from drainResume() instead.
    if (drainState() == DrainState::Drained)
        return;
 
    // If we are time 0 or if the last activation time is in the past,
    // schedule the next tick and wake up the fetch unit
    if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) {
        scheduleTickEvent(Cycles(0));
 
        // Be sure to signal that there's some activity so the CPU doesn't
        // deschedule itself.
        activityRec.activity();
        fetch.wakeFromQuiesce();
 
        Cycles cycles(curCycle() - lastRunningCycle);
        // @todo: This is an oddity that is only here to match the stats
        if (cycles != 0)
            --cycles;
        cpuStats.quiesceCycles += cycles;
 
        lastActivatedCycle = curTick();
 
        _status = Running;
 
        BaseCPU::activateContext(tid);
    }
}
 
void
CPU::suspendContext(ThreadID tid)
{
    DPRINTF(O3CPU,"[tid:%i] Suspending Thread Context.\n", tid);
    assert(!switchedOut());
 
    deactivateThread(tid);
 
    // If this was the last thread then unschedule the tick event.
    if (activeThreads.size() == 0) {
        unscheduleTickEvent();
        lastRunningCycle = curCycle();
        _status = Idle;
    }
 
    DPRINTF(Quiesce, "Suspending Context\n");
 
    BaseCPU::suspendContext(tid);
}
 
void
CPU::haltContext(ThreadID tid)
{
    //For now, this is the same as deallocate
    DPRINTF(O3CPU,"[tid:%i] Halt Context called. Deallocating\n", tid);
    assert(!switchedOut());
 
    deactivateThread(tid);
    removeThread(tid);
 
    // If this was the last thread then unschedule the tick event.
    if (activeThreads.size() == 0) {
        if (tickEvent.scheduled())
        {
            unscheduleTickEvent();
        }
        lastRunningCycle = curCycle();
        _status = Idle;
    }
    updateCycleCounters(BaseCPU::CPU_STATE_SLEEP);
}
 
void
CPU::insertThread(ThreadID tid)
{
    DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU");
    // Will change now that the PC and thread state is internal to the CPU
    // and not in the ThreadContext.
    gem5::ThreadContext *src_tc;
    if (FullSystem)
        src_tc = system->threads[tid];
    else
        src_tc = tcBase(tid);
 
    //Bind Int Regs to Rename Map
    const auto &regClasses = isa[tid]->regClasses();
 
    for (auto type = (RegClassType)0; type <= CCRegClass;
            type = (RegClassType)(type + 1)) {
        for (auto &id: *regClasses.at(type)) {
            PhysRegIdPtr phys_reg = freeList.getReg(type);
            renameMap[tid].setEntry(id, phys_reg);
            scoreboard.setReg(phys_reg);
        }
    }
 
    //Copy Thread Data Into RegFile
    //copyFromTC(tid);
 
    //Set PC/NPC/NNPC
    pcState(src_tc->pcState(), tid);
 
    src_tc->setStatus(gem5::ThreadContext::Active);
 
    activateContext(tid);
 
    //Reset ROB/IQ/LSQ Entries
    commit.rob->resetEntries();
}
 
void
CPU::removeThread(ThreadID tid)
{
    DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid);
 
    // Copy Thread Data From RegFile
    // If thread is suspended, it might be re-allocated
    // copyToTC(tid);
 
 
    // @todo: 2-27-2008: Fix how we free up rename mappings
    // here to alleviate the case for double-freeing registers
    // in SMT workloads.
 
    // clear all thread-specific states in each stage of the pipeline
    // since this thread is going to be completely removed from the CPU
    commit.clearStates(tid);
    bac.clearStates(tid);
    fetch.clearStates(tid);
    decode.clearStates(tid);
    rename.clearStates(tid);
    iew.clearStates(tid);
 
    // at this step, all instructions in the pipeline should be already
    // either committed successfully or squashed. All thread-specific
    // queues in the pipeline must be empty.
    assert(iew.instQueue.getCount(tid) == 0);
    assert(iew.ldstQueue.getCount(tid) == 0);
    assert(commit.rob->isEmpty(tid));
    assert(ftq.isEmpty(tid));
 
    // Reset ROB/IQ/LSQ Entries
 
    // Commented out for now.  This should be possible to do by
    // telling all the pipeline stages to drain first, and then
    // checking until the drain completes.  Once the pipeline is
    // drained, call resetEntries(). - 10-09-06 ktlim
/*
    if (activeThreads.size() >= 1) {
        commit.rob->resetEntries();
        iew.resetEntries();
    }
*/
}
 
Fault
CPU::getInterrupts()
{
    // Check if there are any outstanding interrupts
    return interrupts[0]->getInterrupt();
}
 
void
CPU::processInterrupts(const Fault &interrupt)
{
    // 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.
    // @todo: Possibly consolidate the interrupt checking code.
    // @todo: Allow other threads to handle interrupts.
 
    assert(interrupt != NoFault);
    interrupts[0]->updateIntrInfo();
 
    DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name());
    trap(interrupt, 0, nullptr);
}
 
void
CPU::trap(const Fault &fault, ThreadID tid, const StaticInstPtr &inst)
{
    // Pass the thread's TC into the invoke method.
    fault->invoke(threadContexts[tid], inst);
}
 
void
CPU::serializeThread(CheckpointOut &cp, ThreadID tid) const
{
    thread[tid]->serialize(cp);
}
 
void
CPU::unserializeThread(CheckpointIn &cp, ThreadID tid)
{
    thread[tid]->unserialize(cp);
}
 
DrainState
CPU::drain()
{
    // Deschedule any power gating event (if any)
    deschedulePowerGatingEvent();
 
    // If the CPU isn't doing anything, then return immediately.
    if (switchedOut())
        return DrainState::Drained;
 
    DPRINTF(Drain, "Draining...\n");
 
    // We only need to signal a drain to the commit stage as this
    // initiates squashing controls the draining. Once the commit
    // stage commits an instruction where it is safe to stop, it'll
    // squash the rest of the instructions in the pipeline and force
    // the fetch stage to stall. The pipeline will be drained once all
    // in-flight instructions have retired.
    commit.drain();
 
    // Wake the CPU and record activity so everything can drain out if
    // the CPU was not able to immediately drain.
    if (!isCpuDrained())  {
        // If a thread is suspended, wake it up so it can be drained
        for (auto t : threadContexts) {
            if (t->status() == gem5::ThreadContext::Suspended){
                DPRINTF(Drain, "Currently suspended so activate %i \n",
                        t->threadId());
                t->activate();
                // As the thread is now active, change the power state as well
                activateContext(t->threadId());
            }
        }
 
        wakeCPU();
        activityRec.activity();
 
        DPRINTF(Drain, "CPU not drained\n");
 
        return DrainState::Draining;
    } else {
        DPRINTF(Drain, "CPU is already drained\n");
        if (tickEvent.scheduled())
            deschedule(tickEvent);
 
        // Flush out any old data from the time buffers.  In
        // particular, there might be some data in flight from the
        // fetch stage that isn't visible in any of the CPU buffers we
        // test in isCpuDrained().
        for (int i = 0; i < timeBuffer.getSize(); ++i) {
            timeBuffer.advance();
            fetchQueue.advance();
            decodeQueue.advance();
            renameQueue.advance();
            iewQueue.advance();
        }
 
        drainSanityCheck();
        return DrainState::Drained;
    }
}
 
bool
CPU::tryDrain()
{
    if (drainState() != DrainState::Draining || !isCpuDrained())
        return false;
 
    if (tickEvent.scheduled())
        deschedule(tickEvent);
 
    DPRINTF(Drain, "CPU done draining, processing drain event\n");
    signalDrainDone();
 
    return true;
}
 
void
CPU::drainSanityCheck() const
{
    assert(isCpuDrained());
    bac.drainSanityCheck();
    fetch.drainSanityCheck();
    decode.drainSanityCheck();
    rename.drainSanityCheck();
    iew.drainSanityCheck();
    commit.drainSanityCheck();
}
 
bool
CPU::isCpuDrained() const
{
    bool drained(true);
 
    if (!instList.empty() || !removeList.empty()) {
        DPRINTF(Drain, "Main CPU structures not drained.\n");
        drained = false;
    }
 
    if (!bac.isDrained()) {
        DPRINTF(Drain, "BAC not drained.\n");
        drained = false;
    }
 
    if (!fetch.isDrained()) {
        DPRINTF(Drain, "Fetch not drained.\n");
        drained = false;
    }
 
    if (!decode.isDrained()) {
        DPRINTF(Drain, "Decode not drained.\n");
        drained = false;
    }
 
    if (!rename.isDrained()) {
        DPRINTF(Drain, "Rename not drained.\n");
        drained = false;
    }
 
    if (!iew.isDrained()) {
        DPRINTF(Drain, "IEW not drained.\n");
        drained = false;
    }
 
    if (!commit.isDrained()) {
        DPRINTF(Drain, "Commit not drained.\n");
        drained = false;
    }
 
    return drained;
}
 
void
CPU::commitDrained(ThreadID tid)
{
    bac.drainStall(tid);
    fetch.drainStall(tid);
}
 
void
CPU::drainResume()
{
    if (switchedOut())
        return;
 
    DPRINTF(Drain, "Resuming...\n");
    verifyMemoryMode();
 
    bac.drainResume();
    fetch.drainResume();
    commit.drainResume();
 
    _status = Idle;
    for (ThreadID i = 0; i < thread.size(); i++) {
        if (thread[i]->status() == gem5::ThreadContext::Active) {
            DPRINTF(Drain, "Activating thread: %i\n", i);
            activateThread(i);
            _status = Running;
        }
    }
 
    assert(!tickEvent.scheduled());
    if (_status == Running)
        schedule(tickEvent, nextCycle());
 
    // Reschedule any power gating event (if any)
    schedulePowerGatingEvent();
}
 
void
CPU::switchOut()
{
    DPRINTF(O3CPU, "Switching out\n");
    BaseCPU::switchOut();
 
    activityRec.reset();
 
    _status = SwitchedOut;
 
    if (checker)
        checker->switchOut();
}
 
void
CPU::takeOverFrom(BaseCPU *oldCPU)
{
    BaseCPU::takeOverFrom(oldCPU);
 
    bac.takeOverFrom();
    fetch.takeOverFrom();
    decode.takeOverFrom();
    rename.takeOverFrom();
    iew.takeOverFrom();
    commit.takeOverFrom();
 
    assert(!tickEvent.scheduled());
 
    auto *oldO3CPU = dynamic_cast<CPU *>(oldCPU);
    if (oldO3CPU)
        globalSeqNum = oldO3CPU->globalSeqNum;
 
    lastRunningCycle = curCycle();
    _status = Idle;
}
 
void
CPU::verifyMemoryMode() const
{
    if (!system->isTimingMode()) {
        fatal("The O3 CPU requires the memory system to be in "
              "'timing' mode.\n");
    }
}
 
RegVal
CPU::readMiscRegNoEffect(int misc_reg, ThreadID tid) const
{
    return isa[tid]->readMiscRegNoEffect(misc_reg);
}
 
RegVal
CPU::readMiscReg(int misc_reg, ThreadID tid)
{
    executeStats[tid]->numMiscRegReads++;
    return isa[tid]->readMiscReg(misc_reg);
}
 
void
CPU::setMiscRegNoEffect(int misc_reg, RegVal val, ThreadID tid)
{
    isa[tid]->setMiscRegNoEffect(misc_reg, val);
}
 
void
CPU::setMiscReg(int misc_reg, RegVal val, ThreadID tid)
{
    executeStats[tid]->numMiscRegWrites++;
    isa[tid]->setMiscReg(misc_reg, val);
}
 
RegVal
CPU::getReg(PhysRegIdPtr phys_reg, ThreadID tid)
{
    switch (phys_reg->classValue()) {
      case IntRegClass:
        executeStats[tid]->numIntRegReads++;
        break;
      case FloatRegClass:
        executeStats[tid]->numFpRegReads++;
        break;
      case CCRegClass:
        executeStats[tid]->numCCRegReads++;
        break;
      case VecRegClass:
      case VecElemClass:
        executeStats[tid]->numVecRegReads++;
        break;
      case VecPredRegClass:
        executeStats[tid]->numVecPredRegReads++;
        break;
      default:
        break;
    }
    return regFile.getReg(phys_reg);
}
 
void
CPU::getReg(PhysRegIdPtr phys_reg, void *val, ThreadID tid)
{
    switch (phys_reg->classValue()) {
      case IntRegClass:
        executeStats[tid]->numIntRegReads++;
        break;
      case FloatRegClass:
        executeStats[tid]->numFpRegReads++;
        break;
      case CCRegClass:
        executeStats[tid]->numCCRegReads++;
        break;
      case VecRegClass:
      case VecElemClass:
        executeStats[tid]->numVecRegReads++;
        break;
      case VecPredRegClass:
        executeStats[tid]->numVecPredRegReads++;
        break;
      default:
        break;
    }
    regFile.getReg(phys_reg, val);
}
 
void *
CPU::getWritableReg(PhysRegIdPtr phys_reg, ThreadID tid)
{
    switch (phys_reg->classValue()) {
      case VecRegClass:
        executeStats[tid]->numVecRegWrites++;
        break;
      case VecPredRegClass:
        executeStats[tid]->numVecPredRegWrites++;
        break;
      default:
        break;
    }
    return regFile.getWritableReg(phys_reg);
}
 
void
CPU::setReg(PhysRegIdPtr phys_reg, RegVal val, ThreadID tid)
{
    switch (phys_reg->classValue()) {
      case IntRegClass:
        executeStats[tid]->numIntRegWrites++;
        break;
      case FloatRegClass:
        executeStats[tid]->numFpRegWrites++;
        break;
      case CCRegClass:
        executeStats[tid]->numCCRegWrites++;
        break;
      case VecRegClass:
      case VecElemClass:
        executeStats[tid]->numVecRegWrites++;
        break;
      case VecPredRegClass:
        executeStats[tid]->numVecPredRegWrites++;
        break;
      default:
        break;
    }
    regFile.setReg(phys_reg, val);
}
 
void
CPU::setReg(PhysRegIdPtr phys_reg, const void *val, ThreadID tid)
{
    switch (phys_reg->classValue()) {
      case IntRegClass:
        executeStats[tid]->numIntRegWrites++;
        break;
      case FloatRegClass:
        executeStats[tid]->numFpRegWrites++;
        break;
      case CCRegClass:
        executeStats[tid]->numCCRegWrites++;
        break;
      case VecRegClass:
      case VecElemClass:
        executeStats[tid]->numVecRegWrites++;
        break;
      case VecPredRegClass:
        executeStats[tid]->numVecPredRegWrites++;
        break;
      default:
        break;
    }
    regFile.setReg(phys_reg, val);
}
 
RegVal
CPU::getArchReg(const RegId &reg, ThreadID tid)
{
    const RegId flat = reg.flatten(*isa[tid]);
    PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup(flat);
    return regFile.getReg(phys_reg);
}
 
void
CPU::getArchReg(const RegId &reg, void *val, ThreadID tid)
{
    const RegId flat = reg.flatten(*isa[tid]);
    PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup(flat);
    regFile.getReg(phys_reg, val);
}
 
void *
CPU::getWritableArchReg(const RegId &reg, ThreadID tid)
{
    const RegId flat = reg.flatten(*isa[tid]);
    PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup(flat);
    return regFile.getWritableReg(phys_reg);
}
 
void
CPU::setArchReg(const RegId &reg, RegVal val, ThreadID tid)
{
    const RegId flat = reg.flatten(*isa[tid]);
    PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup(flat);
    regFile.setReg(phys_reg, val);
}
 
void
CPU::setArchReg(const RegId &reg, const void *val, ThreadID tid)
{
    const RegId flat = reg.flatten(*isa[tid]);
    PhysRegIdPtr phys_reg = commitRenameMap[tid].lookup(flat);
    regFile.setReg(phys_reg, val);
}
 
const PCStateBase &
CPU::pcState(ThreadID tid)
{
    return commit.pcState(tid);
}
 
void
CPU::pcState(const PCStateBase &val, ThreadID tid)
{
    commit.pcState(val, tid);
}
 
void
CPU::squashFromTC(ThreadID tid)
{
    thread[tid]->noSquashFromTC = true;
    commit.generateTCEvent(tid);
}
 
CPU::ListIt
CPU::addInst(const DynInstPtr &inst)
{
    instList.push_back(inst);
 
    return --(instList.end());
}
 
void
CPU::instDone(ThreadID tid, const DynInstPtr &inst)
{
    // Keep an instruction count.
    if (!inst->isMicroop() || inst->isLastMicroop()) {
        thread[tid]->numInst++;
        thread[tid]->threadStats.numInsts++;
        commitStats[tid]->numInstsNotNOP++;
 
        // Check for instruction-count-based events.
        thread[tid]->comInstEventQueue.serviceEvents(thread[tid]->numInst);
    }
 
    if (inst->isMemRef()) {
        thread[tid]->threadStats.numMemRefs++;
    }
 
    thread[tid]->numOp++;
    thread[tid]->threadStats.numOps++;
    commitStats[tid]->numOpsNotNOP++;
 
    probeInstCommit(inst->staticInst, inst->pcState().instAddr());
}
 
void
CPU::removeFrontInst(const DynInstPtr &inst)
{
    DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s "
            "[sn:%lli]\n",
            inst->threadNumber, inst->pcState(), inst->seqNum);
 
    removeInstsThisCycle = true;
 
    // Remove the front instruction.
    removeList.push(inst->getInstListIt());
}
 
void
CPU::removeInstsNotInROB(ThreadID tid)
{
    DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction"
            " list.\n", tid);
 
    ListIt end_it;
 
    bool rob_empty = false;
 
    if (instList.empty()) {
        return;
    } else if (rob.isEmpty(tid)) {
        DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n");
        end_it = instList.begin();
        rob_empty = true;
    } else {
        end_it = (rob.readTailInst(tid))->getInstListIt();
        DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n");
    }
 
    removeInstsThisCycle = true;
 
    ListIt inst_it = instList.end();
 
    inst_it--;
 
    // Walk through the instruction list, removing any instructions
    // that were inserted after the given instruction iterator, end_it.
    while (inst_it != end_it) {
        assert(!instList.empty());
 
        squashInstIt(inst_it, tid);
 
        inst_it--;
    }
 
    // If the ROB was empty, then we actually need to remove the first
    // instruction as well.
    if (rob_empty) {
        squashInstIt(inst_it, tid);
    }
}
 
void
CPU::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid)
{
    assert(!instList.empty());
 
    removeInstsThisCycle = true;
 
    ListIt inst_iter = instList.end();
 
    inst_iter--;
 
    DPRINTF(O3CPU, "Deleting instructions from instruction "
            "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
            tid, seq_num, (*inst_iter)->seqNum);
 
    while ((*inst_iter)->seqNum > seq_num) {
 
        bool break_loop = (inst_iter == instList.begin());
 
        squashInstIt(inst_iter, tid);
 
        inst_iter--;
 
        if (break_loop)
            break;
    }
}
 
void
CPU::squashInstIt(const ListIt &instIt, ThreadID tid)
{
    if ((*instIt)->threadNumber == tid) {
        DPRINTF(O3CPU, "Squashing instruction, "
                "[tid:%i] [sn:%lli] PC %s\n",
                (*instIt)->threadNumber,
                (*instIt)->seqNum,
                (*instIt)->pcState());
 
        // Mark it as squashed.
        (*instIt)->setSquashed();
 
        // @todo: Formulate a consistent method for deleting
        // instructions from the instruction list
        // Remove the instruction from the list.
        removeList.push(instIt);
    }
}
 
void
CPU::cleanUpRemovedInsts()
{
    while (!removeList.empty()) {
        DPRINTF(O3CPU, "Removing instruction, "
                "[tid:%i] [sn:%lli] PC %s\n",
                (*removeList.front())->threadNumber,
                (*removeList.front())->seqNum,
                (*removeList.front())->pcState());
 
        instList.erase(removeList.front());
 
        removeList.pop();
    }
 
    removeInstsThisCycle = false;
}
/*
void
CPU::removeAllInsts()
{
    instList.clear();
}
*/
void
CPU::dumpInsts()
{
    int num = 0;
 
    ListIt inst_list_it = instList.begin();
 
    cprintf("Dumping Instruction List\n");
 
    while (inst_list_it != instList.end()) {
        cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
                "Squashed:%i\n\n",
                num, (*inst_list_it)->pcState().instAddr(),
                (*inst_list_it)->threadNumber,
                (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(),
                (*inst_list_it)->isSquashed());
        inst_list_it++;
        ++num;
    }
}
/*
void
CPU::wakeDependents(const DynInstPtr &inst)
{
    iew.wakeDependents(inst);
}
*/
void
CPU::wakeCPU()
{
    if (activityRec.active() || tickEvent.scheduled()) {
        DPRINTF(Activity, "CPU already running.\n");
        return;
    }
 
    DPRINTF(Activity, "Waking up CPU\n");
 
    Cycles cycles(curCycle() - lastRunningCycle);
    // @todo: This is an oddity that is only here to match the stats
    if (cycles > 1) {
        --cycles;
        cpuStats.idleCycles += cycles;
        baseStats.numCycles += cycles;
    }
 
    schedule(tickEvent, clockEdge());
}
 
void
CPU::wakeup(ThreadID tid)
{
    if (thread[tid]->status() != gem5::ThreadContext::Suspended)
        return;
 
    wakeCPU();
 
    DPRINTF(Quiesce, "Suspended Processor woken\n");
    threadContexts[tid]->activate();
}
 
ThreadID
CPU::getFreeTid()
{
    for (ThreadID tid = 0; tid < numThreads; tid++) {
        if (!tids[tid]) {
            tids[tid] = true;
            return tid;
        }
    }
 
    return InvalidThreadID;
}
 
bool
CPU::inUserMode(ThreadID tid)
{
    return isa[tid]->inUserMode();
}
 
void
CPU::updateThreadPriority()
{
    if (activeThreads.size() > 1) {
        //DEFAULT TO ROUND ROBIN SCHEME
        //e.g. Move highest priority to end of thread list
        auto list_begin = activeThreads.begin();
 
        unsigned high_thread = *list_begin;
 
        activeThreads.erase(list_begin);
 
        activeThreads.push_back(high_thread);
    }
}
 
void
CPU::addThreadToExitingList(ThreadID tid)
{
    DPRINTF(O3CPU, "Thread %d is inserted to exitingThreads list\n", tid);
 
    // the thread trying to exit can't be already halted
    assert(tcBase(tid)->status() != gem5::ThreadContext::Halted);
 
    // make sure the thread has not been added to the list yet
    assert(exitingThreads.count(tid) == 0);
 
    // add the thread to exitingThreads list to mark that this thread is
    // trying to exit. The boolean value in the pair denotes if a thread is
    // ready to exit. The thread is not ready to exit until the corresponding
    // exit trap event is processed in the future. Until then, it'll be still
    // an active thread that is trying to exit.
    exitingThreads.emplace(std::make_pair(tid, false));
}
 
bool
CPU::isThreadExiting(ThreadID tid) const
{
    return exitingThreads.count(tid) == 1;
}
 
void
CPU::scheduleThreadExitEvent(ThreadID tid)
{
    assert(exitingThreads.count(tid) == 1);
 
    // exit trap event has been processed. Now, the thread is ready to exit
    // and be removed from the CPU.
    exitingThreads[tid] = true;
 
    // we schedule a threadExitEvent in the next cycle to properly clean
    // up the thread's states in the pipeline. threadExitEvent has lower
    // priority than tickEvent, so the cleanup will happen at the very end
    // of the next cycle after all pipeline stages complete their operations.
    // We want all stages to complete squashing instructions before doing
    // the cleanup.
    if (!threadExitEvent.scheduled()) {
        schedule(threadExitEvent, nextCycle());
    }
}
 
void
CPU::exitThreads()
{
    // there must be at least one thread trying to exit
    assert(exitingThreads.size() > 0);
 
    // terminate all threads that are ready to exit
    auto it = exitingThreads.begin();
    while (it != exitingThreads.end()) {
        ThreadID thread_id = it->first;
        bool readyToExit = it->second;
 
        if (readyToExit) {
            DPRINTF(O3CPU, "Exiting thread %d\n", thread_id);
            haltContext(thread_id);
            tcBase(thread_id)->setStatus(gem5::ThreadContext::Halted);
            it = exitingThreads.erase(it);
        } else {
            it++;
        }
    }
}
 
void
CPU::htmSendAbortSignal(ThreadID tid, uint64_t htm_uid,
        HtmFailureFaultCause cause)
{
    const Addr addr = 0x0ul;
    const int size = 8;
    const Request::Flags flags =
      Request::PHYSICAL|Request::STRICT_ORDER|Request::HTM_ABORT;
 
    // O3-specific actions
    iew.ldstQueue.resetHtmStartsStops(tid);
    commit.resetHtmStartsStops(tid);
 
    // notify l1 d-cache (ruby) that core has aborted transaction
    RequestPtr req =
        std::make_shared<Request>(addr, size, flags, _dataRequestorId);
 
    req->taskId(taskId());
    req->setContext(thread[tid]->contextId());
    req->setHtmAbortCause(cause);
 
    assert(req->isHTMAbort());
 
    PacketPtr abort_pkt = Packet::createRead(req);
    uint8_t *memData = new uint8_t[8];
    assert(memData);
    abort_pkt->dataStatic(memData);
    abort_pkt->setHtmTransactional(htm_uid);
 
    // TODO include correct error handling here
    if (!iew.ldstQueue.getDataPort().sendTimingReq(abort_pkt)) {
        panic("HTM abort signal was not sent to the memory subsystem.");
    }
}
 
} // namespace o3
} // namespace gem5