base.cc

来自「M5,一个功能强大的多处理器系统模拟器.很多针对处理器架构,性能的研究都使用它作」· CC 代码 · 共 455 行

/*
 * Copyright (c) 2002, 2003, 2004, 2005
 * The Regents of The University of Michigan
 * All Rights Reserved
 *
 * This code is part of the M5 simulator.
 *
 * Permission is granted to use, copy, create derivative works and
 * redistribute this software and such derivative works for any
 * purpose, so long as the copyright notice above, this grant of
 * permission, and the disclaimer below appear in all copies made; and
 * so long as the name of The University of Michigan is not used in
 * any advertising or publicity pertaining to the use or distribution
 * of this software without specific, written prior authorization.
 *
 * THIS SOFTWARE IS PROVIDED AS IS, WITHOUT REPRESENTATION FROM THE
 * UNIVERSITY OF MICHIGAN AS TO ITS FITNESS FOR ANY PURPOSE, AND
 * WITHOUT WARRANTY BY THE UNIVERSITY OF MICHIGAN OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE. THE REGENTS OF THE UNIVERSITY OF MICHIGAN SHALL NOT BE
 * LIABLE FOR ANY DAMAGES, INCLUDING DIRECT, SPECIAL, INDIRECT,
 * INCIDENTAL, OR CONSEQUENTIAL DAMAGES, WITH RESPECT TO ANY CLAIM
 * ARISING OUT OF OR IN CONNECTION WITH THE USE OF THE SOFTWARE, EVEN
 * IF IT HAS BEEN OR IS HEREAFTER ADVISED OF THE POSSIBILITY OF SUCH
 * DAMAGES.
 *
 * Authors: Steven K. Reinhardt
 *          Nathan L. Binkert
 */

#include <iostream>
#include <string>
#include <sstream>

#include "base/cprintf.hh"
#include "base/loader/symtab.hh"
#include "base/misc.hh"
#include "base/output.hh"
#include "cpu/base.hh"
#include "cpu/cpuevent.hh"
#include "cpu/thread_context.hh"
#include "cpu/profile.hh"
#include "sim/sim_exit.hh"
#include "sim/process.hh"
#include "sim/sim_events.hh"
#include "sim/system.hh"

#include "base/trace.hh"

// Hack
#include "sim/stat_control.hh"

using namespace std;

vector<BaseCPU *> BaseCPU::cpuList;

// This variable reflects the max number of threads in any CPU.  Be
// careful to only use it once all the CPUs that you care about have
// been initialized
int maxThreadsPerCPU = 1;

CPUProgressEvent::CPUProgressEvent(EventQueue *q, Tick ival,
                                   BaseCPU *_cpu)
    : Event(q, Event::Progress_Event_Pri), interval(ival),
      lastNumInst(0), cpu(_cpu)
{
    if (interval)
        schedule(curTick + interval);
}

void
CPUProgressEvent::process()
{
    Counter temp = cpu->totalInstructions();
#ifndef NDEBUG
    double ipc = double(temp - lastNumInst) / (interval / cpu->ticks(1));

    DPRINTFN("%s progress event, instructions committed: %lli, IPC: %0.8d\n",
             cpu->name(), temp - lastNumInst, ipc);
    ipc = 0.0;
#else
    cprintf("%lli: %s progress event, instructions committed: %lli\n",
            curTick, cpu->name(), temp - lastNumInst);
#endif
    lastNumInst = temp;
    schedule(curTick + interval);
}

const char *
CPUProgressEvent::description() const
{
    return "CPU Progress";
}

#if FULL_SYSTEM
BaseCPU::BaseCPU(Params *p)
    : MemObject(makeParams(p->name)), clock(p->clock), instCnt(0),
      params(p), number_of_threads(p->numberOfThreads), system(p->system),
      phase(p->phase)
#else
BaseCPU::BaseCPU(Params *p)
    : MemObject(makeParams(p->name)), clock(p->clock), params(p),
      number_of_threads(p->numberOfThreads), system(p->system),
      phase(p->phase)
#endif
{
//    currentTick = curTick;

    // add self to global list of CPUs
    cpuList.push_back(this);

    if (number_of_threads > maxThreadsPerCPU)
        maxThreadsPerCPU = number_of_threads;

    // allocate per-thread instruction-based event queues
    comInstEventQueue = new EventQueue *[number_of_threads];
    for (int i = 0; i < number_of_threads; ++i)
        comInstEventQueue[i] = new EventQueue("instruction-based event queue");

    //
    // set up instruction-count-based termination events, if any
    //
    if (p->max_insts_any_thread != 0)
        for (int i = 0; i < number_of_threads; ++i)
            schedExitSimLoop("a thread reached the max instruction count",
                             p->max_insts_any_thread, 0,
                             comInstEventQueue[i]);

    if (p->max_insts_all_threads != 0) {
        // allocate & initialize shared downcounter: each event will
        // decrement this when triggered; simulation will terminate
        // when counter reaches 0
        int *counter = new int;
        *counter = number_of_threads;
        for (int i = 0; i < number_of_threads; ++i)
            new CountedExitEvent(comInstEventQueue[i],
                "all threads reached the max instruction count",
                p->max_insts_all_threads, *counter);
    }

    // allocate per-thread load-based event queues
    comLoadEventQueue = new EventQueue *[number_of_threads];
    for (int i = 0; i < number_of_threads; ++i)
        comLoadEventQueue[i] = new EventQueue("load-based event queue");

    //
    // set up instruction-count-based termination events, if any
    //
    if (p->max_loads_any_thread != 0)
        for (int i = 0; i < number_of_threads; ++i)
            schedExitSimLoop("a thread reached the max load count",
                             p->max_loads_any_thread, 0,
                             comLoadEventQueue[i]);

    if (p->max_loads_all_threads != 0) {
        // allocate & initialize shared downcounter: each event will
        // decrement this when triggered; simulation will terminate
        // when counter reaches 0
        int *counter = new int;
        *counter = number_of_threads;
        for (int i = 0; i < number_of_threads; ++i)
            new CountedExitEvent(comLoadEventQueue[i],
                "all threads reached the max load count",
                p->max_loads_all_threads, *counter);
    }

    functionTracingEnabled = false;
    if (p->functionTrace) {
        functionTraceStream = simout.find(csprintf("ftrace.%s", name()));
        currentFunctionStart = currentFunctionEnd = 0;
        functionEntryTick = p->functionTraceStart;

        if (p->functionTraceStart == 0) {
            functionTracingEnabled = true;
        } else {
            new EventWrapper<BaseCPU, &BaseCPU::enableFunctionTrace>(this,
                                                                     p->functionTraceStart,
                                                                     true);
        }
    }
#if FULL_SYSTEM
    profileEvent = NULL;
    if (params->profile)
        profileEvent = new ProfileEvent(this, params->profile);
#endif
    tracer = params->tracer;
}

BaseCPU::Params::Params()
{
#if FULL_SYSTEM
    profile = false;
#endif
    checker = NULL;
    tracer = NULL;
}

void
BaseCPU::enableFunctionTrace()
{
    functionTracingEnabled = true;
}

BaseCPU::~BaseCPU()
{
}

void
BaseCPU::init()
{
    if (!params->deferRegistration)
        registerThreadContexts();
}

void
BaseCPU::startup()
{
#if FULL_SYSTEM
    if (!params->deferRegistration && profileEvent)
        profileEvent->schedule(curTick);
#endif

    if (params->progress_interval) {
        new CPUProgressEvent(&mainEventQueue,
                             ticks(params->progress_interval),
                             this);
    }
}


void
BaseCPU::regStats()
{
    using namespace Stats;

    numCycles
        .name(name() + ".numCycles")
        .desc("number of cpu cycles simulated")
        ;

    int size = threadContexts.size();
    if (size > 1) {
        for (int i = 0; i < size; ++i) {
            stringstream namestr;
            ccprintf(namestr, "%s.ctx%d", name(), i);
            threadContexts[i]->regStats(namestr.str());
        }
    } else if (size == 1)
        threadContexts[0]->regStats(name());

#if FULL_SYSTEM
#endif
}

Tick
BaseCPU::nextCycle()
{
    Tick next_tick = curTick - phase + clock - 1;
    next_tick -= (next_tick % clock);
    next_tick += phase;
    return next_tick;
}

Tick
BaseCPU::nextCycle(Tick begin_tick)
{
    Tick next_tick = begin_tick;
    if (next_tick % clock != 0)
        next_tick = next_tick - (next_tick % clock) + clock;
    next_tick += phase;

    assert(next_tick >= curTick);
    return next_tick;
}

void
BaseCPU::registerThreadContexts()
{
    for (int i = 0; i < threadContexts.size(); ++i) {
        ThreadContext *tc = threadContexts[i];

#if FULL_SYSTEM
        int id = params->cpu_id;
        if (id != -1)
            id += i;

        tc->setCpuId(system->registerThreadContext(tc, id));
#else
        tc->setCpuId(tc->getProcessPtr()->registerThreadContext(tc));
#endif
    }
}


int
BaseCPU::findContext(ThreadContext *tc)
{
    for (int i = 0; i < threadContexts.size(); ++i) {
        if (tc == threadContexts[i])
            return i;
    }
    return 0;
}

void
BaseCPU::switchOut()
{
//    panic("This CPU doesn't support sampling!");
#if FULL_SYSTEM
    if (profileEvent && profileEvent->scheduled())
        profileEvent->deschedule();
#endif
}

void
BaseCPU::takeOverFrom(BaseCPU *oldCPU, Port *ic, Port *dc)
{
    assert(threadContexts.size() == oldCPU->threadContexts.size());

    for (int i = 0; i < threadContexts.size(); ++i) {
        ThreadContext *newTC = threadContexts[i];
        ThreadContext *oldTC = oldCPU->threadContexts[i];

        newTC->takeOverFrom(oldTC);

        CpuEvent::replaceThreadContext(oldTC, newTC);

        assert(newTC->readCpuId() == oldTC->readCpuId());
#if FULL_SYSTEM
        system->replaceThreadContext(newTC, newTC->readCpuId());
#else
        assert(newTC->getProcessPtr() == oldTC->getProcessPtr());
        newTC->getProcessPtr()->replaceThreadContext(newTC, newTC->readCpuId());
#endif

        if (DTRACE(Context))
            ThreadContext::compare(oldTC, newTC);
    }

#if FULL_SYSTEM
    interrupts = oldCPU->interrupts;

    for (int i = 0; i < threadContexts.size(); ++i)
        threadContexts[i]->profileClear();

    if (profileEvent)
        profileEvent->schedule(curTick);
#endif

    // Connect new CPU to old CPU's memory only if new CPU isn't
    // connected to anything.  Also connect old CPU's memory to new
    // CPU.
    Port *peer;
    if (ic->getPeer() == NULL || ic->getPeer()->isDefaultPort()) {
        peer = oldCPU->getPort("icache_port")->getPeer();
        ic->setPeer(peer);
    } else {
        peer = ic->getPeer();
    }
    peer->setPeer(ic);

    if (dc->getPeer() == NULL || dc->getPeer()->isDefaultPort()) {
        peer = oldCPU->getPort("dcache_port")->getPeer();
        dc->setPeer(peer);
    } else {
        peer = dc->getPeer();
    }
    peer->setPeer(dc);
}


#if FULL_SYSTEM
BaseCPU::ProfileEvent::ProfileEvent(BaseCPU *_cpu, int _interval)
    : Event(&mainEventQueue), cpu(_cpu), interval(_interval)
{ }

void
BaseCPU::ProfileEvent::process()
{
    for (int i = 0, size = cpu->threadContexts.size(); i < size; ++i) {
        ThreadContext *tc = cpu->threadContexts[i];
        tc->profileSample();
    }

    schedule(curTick + interval);
}

void
BaseCPU::post_interrupt(int int_num, int index)
{
    interrupts.post(int_num, index);
}

void
BaseCPU::clear_interrupt(int int_num, int index)
{
    interrupts.clear(int_num, index);
}

void
BaseCPU::clear_interrupts()
{
    interrupts.clear_all();
}

uint64_t
BaseCPU::get_interrupts(int int_num)
{
    return interrupts.get_vec(int_num);
}

void
BaseCPU::serialize(std::ostream &os)
{
    SERIALIZE_SCALAR(instCnt);
    interrupts.serialize(os);
}

void
BaseCPU::unserialize(Checkpoint *cp, const std::string &section)
{
    UNSERIALIZE_SCALAR(instCnt);
    interrupts.unserialize(cp, section);
}

#endif // FULL_SYSTEM

void
BaseCPU::traceFunctionsInternal(Addr pc)
{
    if (!debugSymbolTable)
        return;

    // if pc enters different function, print new function symbol and
    // update saved range.  Otherwise do nothing.
    if (pc < currentFunctionStart || pc >= currentFunctionEnd) {
        string sym_str;
        bool found = debugSymbolTable->findNearestSymbol(pc, sym_str,
                                                         currentFunctionStart,
                                                         currentFunctionEnd);

        if (!found) {
            // no symbol found: use addr as label
            sym_str = csprintf("0x%x", pc);
            currentFunctionStart = pc;
            currentFunctionEnd = pc + 1;
        }

        ccprintf(*functionTraceStream, " (%d)\n%d: %s",
                 curTick - functionEntryTick, curTick, sym_str);
        functionEntryTick = curTick;
    }
}