/*
 * Copyright (c) 2002-2005 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.
 *
 * Authors: Steve Reinhardt
 */

#include "arch/locked_mem.hh"
#include "arch/mmaped_ipr.hh"
#include "arch/utility.hh"
#include "base/bigint.hh"
#include "cpu/exetrace.hh"
#include "cpu/simple/atomic.hh"
#include "mem/packet.hh"
#include "mem/packet_access.hh"
#include "params/AtomicSimpleCPU.hh"
#include "sim/system.hh"

using namespace std;
using namespace TheISA;

AtomicSimpleCPU::TickEvent::TickEvent(AtomicSimpleCPU *c)
    : Event(CPU_Tick_Pri), cpu(c)
{
}


void
AtomicSimpleCPU::TickEvent::process()
{
    cpu->tick();
}

const char *
AtomicSimpleCPU::TickEvent::description() const
{
    return "AtomicSimpleCPU tick";
}

Port *
AtomicSimpleCPU::getPort(const string &if_name, int idx)
{
    if (if_name == "dcache_port")
        return &dcachePort;
    else if (if_name == "icache_port")
        return &icachePort;
    else if (if_name == "physmem_port") {
        hasPhysMemPort = true;
        return &physmemPort;
    }
    else
        panic("No Such Port\n");
}

void
AtomicSimpleCPU::init()
{
    BaseCPU::init();
#if FULL_SYSTEM
    ThreadID size = threadContexts.size();
    for (ThreadID i = 0; i < size; ++i) {
        ThreadContext *tc = threadContexts[i];

        // initialize CPU, including PC
        TheISA::initCPU(tc, tc->contextId());
    }
#endif
    if (hasPhysMemPort) {
        bool snoop = false;
        AddrRangeList pmAddrList;
        physmemPort.getPeerAddressRanges(pmAddrList, snoop);
        physMemAddr = *pmAddrList.begin();
    }
    // Atomic doesn't do MT right now, so contextId == threadId
    ifetch_req.setThreadContext(_cpuId, 0); // Add thread ID if we add MT
    data_read_req.setThreadContext(_cpuId, 0); // Add thread ID here too
    data_write_req.setThreadContext(_cpuId, 0); // Add thread ID here too
}

bool
AtomicSimpleCPU::CpuPort::recvTiming(PacketPtr pkt)
{
    panic("AtomicSimpleCPU doesn't expect recvTiming callback!");
    return true;
}

Tick
AtomicSimpleCPU::CpuPort::recvAtomic(PacketPtr pkt)
{
    //Snooping a coherence request, just return
    return 0;
}

void
AtomicSimpleCPU::CpuPort::recvFunctional(PacketPtr pkt)
{
    //No internal storage to update, just return
    return;
}

void
AtomicSimpleCPU::CpuPort::recvStatusChange(Status status)
{
    if (status == RangeChange) {
        if (!snoopRangeSent) {
            snoopRangeSent = true;
            sendStatusChange(Port::RangeChange);
        }
        return;
    }

    panic("AtomicSimpleCPU doesn't expect recvStatusChange callback!");
}

void
AtomicSimpleCPU::CpuPort::recvRetry()
{
    panic("AtomicSimpleCPU doesn't expect recvRetry callback!");
}

void
AtomicSimpleCPU::DcachePort::setPeer(Port *port)
{
    Port::setPeer(port);

#if FULL_SYSTEM
    // Update the ThreadContext's memory ports (Functional/Virtual
    // Ports)
    cpu->tcBase()->connectMemPorts(cpu->tcBase());
#endif
}

AtomicSimpleCPU::AtomicSimpleCPU(AtomicSimpleCPUParams *p)
    : BaseSimpleCPU(p), tickEvent(this), width(p->width), locked(false),
      simulate_data_stalls(p->simulate_data_stalls),
      simulate_inst_stalls(p->simulate_inst_stalls),
      icachePort(name() + "-iport", this), dcachePort(name() + "-iport", this),
      physmemPort(name() + "-iport", this), hasPhysMemPort(false)
{
    _status = Idle;

    icachePort.snoopRangeSent = false;
    dcachePort.snoopRangeSent = false;

}


AtomicSimpleCPU::~AtomicSimpleCPU()
{
}

void
AtomicSimpleCPU::serialize(ostream &os)
{
    SimObject::State so_state = SimObject::getState();
    SERIALIZE_ENUM(so_state);
    SERIALIZE_SCALAR(locked);
    BaseSimpleCPU::serialize(os);
    nameOut(os, csprintf("%s.tickEvent", name()));
    tickEvent.serialize(os);
}

void
AtomicSimpleCPU::unserialize(Checkpoint *cp, const string &section)
{
    SimObject::State so_state;
    UNSERIALIZE_ENUM(so_state);
    UNSERIALIZE_SCALAR(locked);
    BaseSimpleCPU::unserialize(cp, section);
    tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
}

void
AtomicSimpleCPU::resume()
{
    if (_status == Idle || _status == SwitchedOut)
        return;

    DPRINTF(SimpleCPU, "Resume\n");
    assert(system->getMemoryMode() == Enums::atomic);

    changeState(SimObject::Running);
    if (thread->status() == ThreadContext::Active) {
        if (!tickEvent.scheduled())
            schedule(tickEvent, nextCycle());
    }
}

void
AtomicSimpleCPU::switchOut()
{
    assert(_status == Running || _status == Idle);
    _status = SwitchedOut;

    tickEvent.squash();
}


void
AtomicSimpleCPU::takeOverFrom(BaseCPU *oldCPU)
{
    BaseCPU::takeOverFrom(oldCPU, &icachePort, &dcachePort);

    assert(!tickEvent.scheduled());

    // if any of this CPU's ThreadContexts are active, mark the CPU as
    // running and schedule its tick event.
    ThreadID size = threadContexts.size();
    for (ThreadID i = 0; i < size; ++i) {
        ThreadContext *tc = threadContexts[i];
        if (tc->status() == ThreadContext::Active && _status != Running) {
            _status = Running;
            schedule(tickEvent, nextCycle());
            break;
        }
    }
    if (_status != Running) {
        _status = Idle;
    }
    assert(threadContexts.size() == 1);
    ifetch_req.setThreadContext(_cpuId, 0); // Add thread ID if we add MT
    data_read_req.setThreadContext(_cpuId, 0); // Add thread ID here too
    data_write_req.setThreadContext(_cpuId, 0); // Add thread ID here too
}


void
AtomicSimpleCPU::activateContext(int thread_num, int delay)
{
    DPRINTF(SimpleCPU, "ActivateContext %d (%d cycles)\n", thread_num, delay);

    assert(thread_num == 0);
    assert(thread);

    assert(_status == Idle);
    assert(!tickEvent.scheduled());

    notIdleFraction++;
    numCycles += tickToCycles(thread->lastActivate - thread->lastSuspend);

    //Make sure ticks are still on multiples of cycles
    schedule(tickEvent, nextCycle(curTick + ticks(delay)));
    _status = Running;
}


void
AtomicSimpleCPU::suspendContext(int thread_num)
{
    DPRINTF(SimpleCPU, "SuspendContext %d\n", thread_num);

    assert(thread_num == 0);
    assert(thread);

    if (_status == Idle)
        return;

    assert(_status == Running);

    // tick event may not be scheduled if this gets called from inside
    // an instruction's execution, e.g. "quiesce"
    if (tickEvent.scheduled())
        deschedule(tickEvent);

    notIdleFraction--;
    _status = Idle;
}


template <class T>
Fault
AtomicSimpleCPU::read(Addr addr, T &data, unsigned flags)
{
    // use the CPU's statically allocated read request and packet objects
    Request *req = &data_read_req;

    if (traceData) {
        traceData->setAddr(addr);
    }

    //The block size of our peer.
    unsigned blockSize = dcachePort.peerBlockSize();
    //The size of the data we're trying to read.
    int dataSize = sizeof(T);

    uint8_t * dataPtr = (uint8_t *)&data;

    //The address of the second part of this access if it needs to be split
    //across a cache line boundary.
    Addr secondAddr = roundDown(addr + dataSize - 1, blockSize);

    if(secondAddr > addr)
        dataSize = secondAddr - addr;

    dcache_latency = 0;

    while(1) {
        req->setVirt(0, addr, dataSize, flags, thread->readPC());

        // translate to physical address
        Fault fault = thread->dtb->translateAtomic(req, tc, BaseTLB::Read);

        // Now do the access.
        if (fault == NoFault) {
            Packet pkt = Packet(req,
                    req->isLLSC() ? MemCmd::LoadLockedReq : MemCmd::ReadReq,
                    Packet::Broadcast);
            pkt.dataStatic(dataPtr);

            if (req->isMmapedIpr())
                dcache_latency += TheISA::handleIprRead(thread->getTC(), &pkt);
            else {
                if (hasPhysMemPort && pkt.getAddr() == physMemAddr)
                    dcache_latency += physmemPort.sendAtomic(&pkt);
                else
                    dcache_latency += dcachePort.sendAtomic(&pkt);
            }
            dcache_access = true;

            assert(!pkt.isError());

            if (req->isLLSC()) {
                TheISA::handleLockedRead(thread, req);
            }
        }

        // This will need a new way to tell if it has a dcache attached.
        if (req->isUncacheable())
            recordEvent("Uncached Read");

        //If there's a fault, return it
        if (fault != NoFault)
            return fault;
        //If we don't need to access a second cache line, stop now.
        if (secondAddr <= addr)
        {
            data = gtoh(data);
            if (traceData) {
                traceData->setData(data);
            }
            if (req->isLocked() && fault == NoFault) {
                assert(!locked);
                locked = true;
            }
            return fault;
        }

        /*
         * Set up for accessing the second cache line.
         */

        //Move the pointer we're reading into to the correct location.
        dataPtr += dataSize;
        //Adjust the size to get the remaining bytes.
        dataSize = addr + sizeof(T) - secondAddr;
        //And access the right address.
        addr = secondAddr;
    }
}

#ifndef DOXYGEN_SHOULD_SKIP_THIS

template
Fault
AtomicSimpleCPU::read(Addr addr, Twin32_t &data, unsigned flags);

template
Fault
AtomicSimpleCPU::read(Addr addr, Twin64_t &data, unsigned flags);

template
Fault
AtomicSimpleCPU::read(Addr addr, uint64_t &data, unsigned flags);

template
Fault
AtomicSimpleCPU::read(Addr addr, uint32_t &data, unsigned flags);

template
Fault
AtomicSimpleCPU::read(Addr addr, uint16_t &data, unsigned flags);

template
Fault
AtomicSimpleCPU::read(Addr addr, uint8_t &data, unsigned flags);

#endif //DOXYGEN_SHOULD_SKIP_THIS

template<>
Fault
AtomicSimpleCPU::read(Addr addr, double &data, unsigned flags)
{
    return read(addr, *(uint64_t*)&data, flags);
}

template<>
Fault
AtomicSimpleCPU::read(Addr addr, float &data, unsigned flags)
{
    return read(addr, *(uint32_t*)&data, flags);
}


template<>
Fault
AtomicSimpleCPU::read(Addr addr, int32_t &data, unsigned flags)
{
    return read(addr, (uint32_t&)data, flags);
}


template <class T>
Fault
AtomicSimpleCPU::write(T data, Addr addr, unsigned flags, uint64_t *res)
{
    // use the CPU's statically allocated write request and packet objects
    Request *req = &data_write_req;

    if (traceData) {
        traceData->setAddr(addr);
    }

    //The block size of our peer.
    unsigned blockSize = dcachePort.peerBlockSize();
    //The size of the data we're trying to read.
    int dataSize = sizeof(T);

    uint8_t * dataPtr = (uint8_t *)&data;

    //The address of the second part of this access if it needs to be split
    //across a cache line boundary.
    Addr secondAddr = roundDown(addr + dataSize - 1, blockSize);

    if(secondAddr > addr)
        dataSize = secondAddr - addr;

    dcache_latency = 0;

    while(1) {
        req->setVirt(0, addr, dataSize, flags, thread->readPC());

        // translate to physical address
        Fault fault = thread->dtb->translateAtomic(req, tc, BaseTLB::Write);

        // Now do the access.
        if (fault == NoFault) {
            MemCmd cmd = MemCmd::WriteReq; // default
            bool do_access = true;  // flag to suppress cache access

            if (req->isLLSC()) {
                cmd = MemCmd::StoreCondReq;
                do_access = TheISA::handleLockedWrite(thread, req);
            } else if (req->isSwap()) {
                cmd = MemCmd::SwapReq;
                if (req->isCondSwap()) {
                    assert(res);
                    req->setExtraData(*res);
                }
            }

            if (do_access) {
                Packet pkt = Packet(req, cmd, Packet::Broadcast);
                pkt.dataStatic(dataPtr);

                if (req->isMmapedIpr()) {
                    dcache_latency +=
                        TheISA::handleIprWrite(thread->getTC(), &pkt);
                } else {
                    //XXX This needs to be outside of the loop in order to
                    //work properly for cache line boundary crossing
                    //accesses in transendian simulations.
                    data = htog(data);
                    if (hasPhysMemPort && pkt.getAddr() == physMemAddr)
                        dcache_latency += physmemPort.sendAtomic(&pkt);
                    else
                        dcache_latency += dcachePort.sendAtomic(&pkt);
                }
                dcache_access = true;
                assert(!pkt.isError());

                if (req->isSwap()) {
                    assert(res);
                    *res = pkt.get<T>();
                }
            }

            if (res && !req->isSwap()) {
                *res = req->getExtraData();
            }
        }

        // This will need a new way to tell if it's hooked up to a cache or not.
        if (req->isUncacheable())
            recordEvent("Uncached Write");

        //If there's a fault or we don't need to access a second cache line,
        //stop now.
        if (fault != NoFault || secondAddr <= addr)
        {
            // If the write needs to have a fault on the access, consider
            // calling changeStatus() and changing it to "bad addr write"
            // or something.
            if (traceData) {
                traceData->setData(gtoh(data));
            }
            if (req->isLocked() && fault == NoFault) {
                assert(locked);
                locked = false;
            }
            return fault;
        }

        /*
         * Set up for accessing the second cache line.
         */

        //Move the pointer we're reading into to the correct location.
        dataPtr += dataSize;
        //Adjust the size to get the remaining bytes.
        dataSize = addr + sizeof(T) - secondAddr;
        //And access the right address.
        addr = secondAddr;
    }
}


#ifndef DOXYGEN_SHOULD_SKIP_THIS

template
Fault
AtomicSimpleCPU::write(Twin32_t data, Addr addr,
                       unsigned flags, uint64_t *res);

template
Fault
AtomicSimpleCPU::write(Twin64_t data, Addr addr,
                       unsigned flags, uint64_t *res);

template
Fault
AtomicSimpleCPU::write(uint64_t data, Addr addr,
                       unsigned flags, uint64_t *res);

template
Fault
AtomicSimpleCPU::write(uint32_t data, Addr addr,
                       unsigned flags, uint64_t *res);

template
Fault
AtomicSimpleCPU::write(uint16_t data, Addr addr,
                       unsigned flags, uint64_t *res);

template
Fault
AtomicSimpleCPU::write(uint8_t data, Addr addr,
                       unsigned flags, uint64_t *res);

#endif //DOXYGEN_SHOULD_SKIP_THIS

template<>
Fault
AtomicSimpleCPU::write(double data, Addr addr, unsigned flags, uint64_t *res)
{
    return write(*(uint64_t*)&data, addr, flags, res);
}

template<>
Fault
AtomicSimpleCPU::write(float data, Addr addr, unsigned flags, uint64_t *res)
{
    return write(*(uint32_t*)&data, addr, flags, res);
}


template<>
Fault
AtomicSimpleCPU::write(int32_t data, Addr addr, unsigned flags, uint64_t *res)
{
    return write((uint32_t)data, addr, flags, res);
}


void
AtomicSimpleCPU::tick()
{
    DPRINTF(SimpleCPU, "Tick\n");

    Tick latency = 0;

    for (int i = 0; i < width || locked; ++i) {
        numCycles++;

        if (!curStaticInst || !curStaticInst->isDelayedCommit())
            checkForInterrupts();

        checkPcEventQueue();

        Fault fault = NoFault;

        bool fromRom = isRomMicroPC(thread->readMicroPC());
        if (!fromRom && !curMacroStaticInst) {
            setupFetchRequest(&ifetch_req);
            fault = thread->itb->translateAtomic(&ifetch_req, tc,
                                                 BaseTLB::Execute);
        }

        if (fault == NoFault) {
            Tick icache_latency = 0;
            bool icache_access = false;
            dcache_access = false; // assume no dcache access

            if (!fromRom && !curMacroStaticInst) {
                // This is commented out because the predecoder would act like
                // a tiny cache otherwise. It wouldn't be flushed when needed
                // like the I cache. It should be flushed, and when that works
                // this code should be uncommented.
                //Fetch more instruction memory if necessary
                //if(predecoder.needMoreBytes())
                //{
                    icache_access = true;
                    Packet ifetch_pkt = Packet(&ifetch_req, MemCmd::ReadReq,
                                               Packet::Broadcast);
                    ifetch_pkt.dataStatic(&inst);

                    if (hasPhysMemPort && ifetch_pkt.getAddr() == physMemAddr)
                        icache_latency = physmemPort.sendAtomic(&ifetch_pkt);
                    else
                        icache_latency = icachePort.sendAtomic(&ifetch_pkt);

                    assert(!ifetch_pkt.isError());

                    // ifetch_req is initialized to read the instruction directly
                    // into the CPU object's inst field.
                //}
            }

            preExecute();

            if (curStaticInst) {
                fault = curStaticInst->execute(this, traceData);

                // keep an instruction count
                if (fault == NoFault)
                    countInst();
                else if (traceData) {
                    // If there was a fault, we should trace this instruction.
                    delete traceData;
                    traceData = NULL;
                }

                postExecute();
            }

            // @todo remove me after debugging with legion done
            if (curStaticInst && (!curStaticInst->isMicroop() ||
                        curStaticInst->isFirstMicroop()))
                instCnt++;

            Tick stall_ticks = 0;
            if (simulate_inst_stalls && icache_access)
                stall_ticks += icache_latency;

            if (simulate_data_stalls && dcache_access)
                stall_ticks += dcache_latency;

            if (stall_ticks) {
                Tick stall_cycles = stall_ticks / ticks(1);
                Tick aligned_stall_ticks = ticks(stall_cycles);

                if (aligned_stall_ticks < stall_ticks)
                    aligned_stall_ticks += 1;

                latency += aligned_stall_ticks;
            }

        }
        if(fault != NoFault || !stayAtPC)
            advancePC(fault);
    }

    // instruction takes at least one cycle
    if (latency < ticks(1))
        latency = ticks(1);

    if (_status != Idle)
        schedule(tickEvent, curTick + latency);
}


void
AtomicSimpleCPU::printAddr(Addr a)
{
    dcachePort.printAddr(a);
}


////////////////////////////////////////////////////////////////////////
//
//  AtomicSimpleCPU Simulation Object
//
AtomicSimpleCPU *
AtomicSimpleCPUParams::create()
{
    numThreads = 1;
#if !FULL_SYSTEM
    if (workload.size() != 1)
        panic("only one workload allowed");
#endif
    return new AtomicSimpleCPU(this);
}