gem5/src/mem/coherent_xbar.cc

/*
 * Copyright (c) 2011-2020 ARM Limited
 * 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) 2006 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.
 */

/**
 * @file
 * Definition of a crossbar object.
 */

#include "mem/coherent_xbar.hh"

#include "base/logging.hh"
#include "base/trace.hh"
#include "debug/AddrRanges.hh"
#include "debug/CoherentXBar.hh"
#include "sim/system.hh"

CoherentXBar::CoherentXBar(const CoherentXBarParams &p)
    : BaseXBar(p), system(p.system), snoopFilter(p.snoop_filter),
      snoopResponseLatency(p.snoop_response_latency),
      maxOutstandingSnoopCheck(p.max_outstanding_snoops),
      maxRoutingTableSizeCheck(p.max_routing_table_size),
      pointOfCoherency(p.point_of_coherency),
      pointOfUnification(p.point_of_unification),

      ADD_STAT(snoops, UNIT_COUNT, "Total snoops"),
      ADD_STAT(snoopTraffic, UNIT_BYTE, "Total snoop traffic"),
      ADD_STAT(snoopFanout, UNIT_COUNT, "Request fanout histogram")
{
    // create the ports based on the size of the memory-side port and
    // CPU-side port vector ports, and the presence of the default port,
    // the ports are enumerated starting from zero
    for (int i = 0; i < p.port_mem_side_ports_connection_count; ++i) {
        std::string portName = csprintf("%s.mem_side_port[%d]", name(), i);
        RequestPort* bp = new CoherentXBarRequestPort(portName, *this, i);
        memSidePorts.push_back(bp);
        reqLayers.push_back(new ReqLayer(*bp, *this,
                                         csprintf("reqLayer%d", i)));
        snoopLayers.push_back(
                new SnoopRespLayer(*bp, *this, csprintf("snoopLayer%d", i)));
    }

    // see if we have a default CPU-side-port device connected and if so add
    // our corresponding memory-side port
    if (p.port_default_connection_count) {
        defaultPortID = memSidePorts.size();
        std::string portName = name() + ".default";
        RequestPort* bp = new CoherentXBarRequestPort(portName, *this,
                                                    defaultPortID);
        memSidePorts.push_back(bp);
        reqLayers.push_back(new ReqLayer(*bp, *this, csprintf("reqLayer%d",
                                         defaultPortID)));
        snoopLayers.push_back(new SnoopRespLayer(*bp, *this,
                                                 csprintf("snoopLayer%d",
                                                          defaultPortID)));
    }

    // create the CPU-side ports, once again starting at zero
    for (int i = 0; i < p.port_cpu_side_ports_connection_count; ++i) {
        std::string portName = csprintf("%s.cpu_side_port[%d]", name(), i);
        QueuedResponsePort* bp = new CoherentXBarResponsePort(portName,
                                                            *this, i);
        cpuSidePorts.push_back(bp);
        respLayers.push_back(new RespLayer(*bp, *this,
                                           csprintf("respLayer%d", i)));
        snoopRespPorts.push_back(new SnoopRespPort(*bp, *this));
    }
}

CoherentXBar::~CoherentXBar()
{
    for (auto l: reqLayers)
        delete l;
    for (auto l: respLayers)
        delete l;
    for (auto l: snoopLayers)
        delete l;
    for (auto p: snoopRespPorts)
        delete p;
}

void
CoherentXBar::init()
{
    BaseXBar::init();

    // iterate over our CPU-side ports and determine which of our
    // neighbouring memory-side ports are snooping and add them as snoopers
    for (const auto& p: cpuSidePorts) {
        // check if the connected memory-side port is snooping
        if (p->isSnooping()) {
            DPRINTF(AddrRanges, "Adding snooping requestor %s\n",
                    p->getPeer());
            snoopPorts.push_back(p);
        }
    }

    if (snoopPorts.empty())
        warn("CoherentXBar %s has no snooping ports attached!\n", name());

    // inform the snoop filter about the CPU-side ports so it can create
    // its own internal representation
    if (snoopFilter)
        snoopFilter->setCPUSidePorts(cpuSidePorts);
}

bool
CoherentXBar::recvTimingReq(PacketPtr pkt, PortID cpu_side_port_id)
{
    // determine the source port based on the id
    ResponsePort *src_port = cpuSidePorts[cpu_side_port_id];

    // remember if the packet is an express snoop
    bool is_express_snoop = pkt->isExpressSnoop();
    bool cache_responding = pkt->cacheResponding();
    // for normal requests, going downstream, the express snoop flag
    // and the cache responding flag should always be the same
    assert(is_express_snoop == cache_responding);

    // determine the destination based on the destination address range
    PortID mem_side_port_id = findPort(pkt->getAddrRange());

    // test if the crossbar should be considered occupied for the current
    // port, and exclude express snoops from the check
    if (!is_express_snoop &&
        !reqLayers[mem_side_port_id]->tryTiming(src_port)) {
        DPRINTF(CoherentXBar, "%s: src %s packet %s BUSY\n", __func__,
                src_port->name(), pkt->print());
        return false;
    }

    DPRINTF(CoherentXBar, "%s: src %s packet %s\n", __func__,
            src_port->name(), pkt->print());

    // store size and command as they might be modified when
    // forwarding the packet
    unsigned int pkt_size = pkt->hasData() ? pkt->getSize() : 0;
    unsigned int pkt_cmd = pkt->cmdToIndex();

    // store the old header delay so we can restore it if needed
    Tick old_header_delay = pkt->headerDelay;

    // a request sees the frontend and forward latency
    Tick xbar_delay = (frontendLatency + forwardLatency) * clockPeriod();

    // set the packet header and payload delay
    calcPacketTiming(pkt, xbar_delay);

    // determine how long to be crossbar layer is busy
    Tick packetFinishTime = clockEdge(headerLatency) + pkt->payloadDelay;

    // is this the destination point for this packet? (e.g. true if
    // this xbar is the PoC for a cache maintenance operation to the
    // PoC) otherwise the destination is any cache that can satisfy
    // the request
    const bool is_destination = isDestination(pkt);

    const bool snoop_caches = !system->bypassCaches() &&
        pkt->cmd != MemCmd::WriteClean;
    if (snoop_caches) {
        assert(pkt->snoopDelay == 0);

        if (pkt->isClean() && !is_destination) {
            // before snooping we need to make sure that the memory
            // below is not busy and the cache clean request can be
            // forwarded to it
            if (!memSidePorts[mem_side_port_id]->tryTiming(pkt)) {
                DPRINTF(CoherentXBar, "%s: src %s packet %s RETRY\n", __func__,
                        src_port->name(), pkt->print());

                // update the layer state and schedule an idle event
                reqLayers[mem_side_port_id]->failedTiming(src_port,
                                                        clockEdge(Cycles(1)));
                return false;
            }
        }


        // the packet is a memory-mapped request and should be
        // broadcasted to our snoopers but the source
        if (snoopFilter) {
            // check with the snoop filter where to forward this packet
            auto sf_res = snoopFilter->lookupRequest(pkt, *src_port);
            // the time required by a packet to be delivered through
            // the xbar has to be charged also with to lookup latency
            // of the snoop filter
            pkt->headerDelay += sf_res.second * clockPeriod();
            DPRINTF(CoherentXBar, "%s: src %s packet %s SF size: %i lat: %i\n",
                    __func__, src_port->name(), pkt->print(),
                    sf_res.first.size(), sf_res.second);

            if (pkt->isEviction()) {
                // for block-evicting packets, i.e. writebacks and
                // clean evictions, there is no need to snoop up, as
                // all we do is determine if the block is cached or
                // not, instead just set it here based on the snoop
                // filter result
                if (!sf_res.first.empty())
                    pkt->setBlockCached();
            } else {
                forwardTiming(pkt, cpu_side_port_id, sf_res.first);
            }
        } else {
            forwardTiming(pkt, cpu_side_port_id);
        }

        // add the snoop delay to our header delay, and then reset it
        pkt->headerDelay += pkt->snoopDelay;
        pkt->snoopDelay = 0;
    }

    // set up a sensible starting point
    bool success = true;

    // remember if the packet will generate a snoop response by
    // checking if a cache set the cacheResponding flag during the
    // snooping above
    const bool expect_snoop_resp = !cache_responding && pkt->cacheResponding();
    bool expect_response = pkt->needsResponse() && !pkt->cacheResponding();

    const bool sink_packet = sinkPacket(pkt);

    // in certain cases the crossbar is responsible for responding
    bool respond_directly = false;
    // store the original address as an address mapper could possibly
    // modify the address upon a sendTimingRequest
    const Addr addr(pkt->getAddr());
    if (sink_packet) {
        DPRINTF(CoherentXBar, "%s: Not forwarding %s\n", __func__,
                pkt->print());
    } else {
        // determine if we are forwarding the packet, or responding to
        // it
        if (forwardPacket(pkt)) {
            // if we are passing on, rather than sinking, a packet to
            // which an upstream cache has committed to responding,
            // the line was needs writable, and the responding only
            // had an Owned copy, so we need to immidiately let the
            // downstream caches know, bypass any flow control
            if (pkt->cacheResponding()) {
                pkt->setExpressSnoop();
            }

            // make sure that the write request (e.g., WriteClean)
            // will stop at the memory below if this crossbar is its
            // destination
            if (pkt->isWrite() && is_destination) {
                pkt->clearWriteThrough();
            }

            // since it is a normal request, attempt to send the packet
            success = memSidePorts[mem_side_port_id]->sendTimingReq(pkt);
        } else {
            // no need to forward, turn this packet around and respond
            // directly
            assert(pkt->needsResponse());

            respond_directly = true;
            assert(!expect_snoop_resp);
            expect_response = false;
        }
    }

    if (snoopFilter && snoop_caches) {
        // Let the snoop filter know about the success of the send operation
        snoopFilter->finishRequest(!success, addr, pkt->isSecure());
    }

    // check if we were successful in sending the packet onwards
    if (!success)  {
        // express snoops should never be forced to retry
        assert(!is_express_snoop);

        // restore the header delay
        pkt->headerDelay = old_header_delay;

        DPRINTF(CoherentXBar, "%s: src %s packet %s RETRY\n", __func__,
                src_port->name(), pkt->print());

        // update the layer state and schedule an idle event
        reqLayers[mem_side_port_id]->failedTiming(src_port,
                                                clockEdge(Cycles(1)));
    } else {
        // express snoops currently bypass the crossbar state entirely
        if (!is_express_snoop) {
            // if this particular request will generate a snoop
            // response
            if (expect_snoop_resp) {
                // we should never have an exsiting request outstanding
                assert(outstandingSnoop.find(pkt->req) ==
                       outstandingSnoop.end());
                outstandingSnoop.insert(pkt->req);

                // basic sanity check on the outstanding snoops
                panic_if(outstandingSnoop.size() > maxOutstandingSnoopCheck,
                         "%s: Outstanding snoop requests exceeded %d\n",
                         name(), maxOutstandingSnoopCheck);
            }

            // remember where to route the normal response to
            if (expect_response || expect_snoop_resp) {
                assert(routeTo.find(pkt->req) == routeTo.end());
                routeTo[pkt->req] = cpu_side_port_id;

                panic_if(routeTo.size() > maxRoutingTableSizeCheck,
                         "%s: Routing table exceeds %d packets\n",
                         name(), maxRoutingTableSizeCheck);
            }

            // update the layer state and schedule an idle event
            reqLayers[mem_side_port_id]->succeededTiming(packetFinishTime);
        }

        // stats updates only consider packets that were successfully sent
        pktCount[cpu_side_port_id][mem_side_port_id]++;
        pktSize[cpu_side_port_id][mem_side_port_id] += pkt_size;
        transDist[pkt_cmd]++;

        if (is_express_snoop) {
            snoops++;
            snoopTraffic += pkt_size;
        }
    }

    if (sink_packet)
        // queue the packet for deletion
        pendingDelete.reset(pkt);

    // normally we respond to the packet we just received if we need to
    PacketPtr rsp_pkt = pkt;
    PortID rsp_port_id = cpu_side_port_id;

    // If this is the destination of the cache clean operation the
    // crossbar is responsible for responding. This crossbar will
    // respond when the cache clean is complete. A cache clean
    // is complete either:
    // * direcly, if no cache above had a dirty copy of the block
    //   as indicated by the satisfied flag of the packet, or
    // * when the crossbar has seen both the cache clean request
    //   (CleanSharedReq, CleanInvalidReq) and the corresponding
    //   write (WriteClean) which updates the block in the memory
    //   below.
    if (success &&
        ((pkt->isClean() && pkt->satisfied()) ||
         pkt->cmd == MemCmd::WriteClean) &&
        is_destination) {
        PacketPtr deferred_rsp = pkt->isWrite() ? nullptr : pkt;
        auto cmo_lookup = outstandingCMO.find(pkt->id);
        if (cmo_lookup != outstandingCMO.end()) {
            // the cache clean request has already reached this xbar
            respond_directly = true;
            if (pkt->isWrite()) {
                rsp_pkt = cmo_lookup->second;
                assert(rsp_pkt);

                // determine the destination
                const auto route_lookup = routeTo.find(rsp_pkt->req);
                assert(route_lookup != routeTo.end());
                rsp_port_id = route_lookup->second;
                assert(rsp_port_id != InvalidPortID);
                assert(rsp_port_id < respLayers.size());
                // remove the request from the routing table
                routeTo.erase(route_lookup);
            }
            outstandingCMO.erase(cmo_lookup);
        } else {
            respond_directly = false;
            outstandingCMO.emplace(pkt->id, deferred_rsp);
            if (!pkt->isWrite()) {
                assert(routeTo.find(pkt->req) == routeTo.end());
                routeTo[pkt->req] = cpu_side_port_id;

                panic_if(routeTo.size() > maxRoutingTableSizeCheck,
                         "%s: Routing table exceeds %d packets\n",
                         name(), maxRoutingTableSizeCheck);
            }
        }
    }


    if (respond_directly) {
        assert(rsp_pkt->needsResponse());
        assert(success);

        rsp_pkt->makeResponse();

        if (snoopFilter && !system->bypassCaches()) {
            // let the snoop filter inspect the response and update its state
            snoopFilter->updateResponse(rsp_pkt, *cpuSidePorts[rsp_port_id]);
        }

        // we send the response after the current packet, even if the
        // response is not for this packet (e.g. cache clean operation
        // where both the request and the write packet have to cross
        // the destination xbar before the response is sent.)
        Tick response_time = clockEdge() + pkt->headerDelay;
        rsp_pkt->headerDelay = 0;

        cpuSidePorts[rsp_port_id]->schedTimingResp(rsp_pkt, response_time);
    }

    return success;
}

bool
CoherentXBar::recvTimingResp(PacketPtr pkt, PortID mem_side_port_id)
{
    // determine the source port based on the id
    RequestPort *src_port = memSidePorts[mem_side_port_id];

    // determine the destination
    const auto route_lookup = routeTo.find(pkt->req);
    assert(route_lookup != routeTo.end());
    const PortID cpu_side_port_id = route_lookup->second;
    assert(cpu_side_port_id != InvalidPortID);
    assert(cpu_side_port_id < respLayers.size());

    // test if the crossbar should be considered occupied for the
    // current port
    if (!respLayers[cpu_side_port_id]->tryTiming(src_port)) {
        DPRINTF(CoherentXBar, "%s: src %s packet %s BUSY\n", __func__,
                src_port->name(), pkt->print());
        return false;
    }

    DPRINTF(CoherentXBar, "%s: src %s packet %s\n", __func__,
            src_port->name(), pkt->print());

    // store size and command as they might be modified when
    // forwarding the packet
    unsigned int pkt_size = pkt->hasData() ? pkt->getSize() : 0;
    unsigned int pkt_cmd = pkt->cmdToIndex();

    // a response sees the response latency
    Tick xbar_delay = responseLatency * clockPeriod();

    // set the packet header and payload delay
    calcPacketTiming(pkt, xbar_delay);

    // determine how long to be crossbar layer is busy
    Tick packetFinishTime = clockEdge(headerLatency) + pkt->payloadDelay;

    if (snoopFilter && !system->bypassCaches()) {
        // let the snoop filter inspect the response and update its state
        snoopFilter->updateResponse(pkt, *cpuSidePorts[cpu_side_port_id]);
    }

    // send the packet through the destination CPU-side port and pay for
    // any outstanding header delay
    Tick latency = pkt->headerDelay;
    pkt->headerDelay = 0;
    cpuSidePorts[cpu_side_port_id]->schedTimingResp(pkt, curTick()
                                        + latency);

    // remove the request from the routing table
    routeTo.erase(route_lookup);

    respLayers[cpu_side_port_id]->succeededTiming(packetFinishTime);

    // stats updates
    pktCount[cpu_side_port_id][mem_side_port_id]++;
    pktSize[cpu_side_port_id][mem_side_port_id] += pkt_size;
    transDist[pkt_cmd]++;

    return true;
}

void
CoherentXBar::recvTimingSnoopReq(PacketPtr pkt, PortID mem_side_port_id)
{
    DPRINTF(CoherentXBar, "%s: src %s packet %s\n", __func__,
            memSidePorts[mem_side_port_id]->name(), pkt->print());

    // update stats here as we know the forwarding will succeed
    unsigned int pkt_size = pkt->hasData() ? pkt->getSize() : 0;
    transDist[pkt->cmdToIndex()]++;
    snoops++;
    snoopTraffic += pkt_size;

    // we should only see express snoops from caches
    assert(pkt->isExpressSnoop());

    // set the packet header and payload delay, for now use forward latency
    // @todo Assess the choice of latency further
    calcPacketTiming(pkt, forwardLatency * clockPeriod());

    // remember if a cache has already committed to responding so we
    // can see if it changes during the snooping
    const bool cache_responding = pkt->cacheResponding();

    assert(pkt->snoopDelay == 0);

    if (snoopFilter) {
        // let the Snoop Filter work its magic and guide probing
        auto sf_res = snoopFilter->lookupSnoop(pkt);
        // the time required by a packet to be delivered through
        // the xbar has to be charged also with to lookup latency
        // of the snoop filter
        pkt->headerDelay += sf_res.second * clockPeriod();
        DPRINTF(CoherentXBar, "%s: src %s packet %s SF size: %i lat: %i\n",
                __func__, memSidePorts[mem_side_port_id]->name(),
                pkt->print(), sf_res.first.size(), sf_res.second);

        // forward to all snoopers
        forwardTiming(pkt, InvalidPortID, sf_res.first);
    } else {
        forwardTiming(pkt, InvalidPortID);
    }

    // add the snoop delay to our header delay, and then reset it
    pkt->headerDelay += pkt->snoopDelay;
    pkt->snoopDelay = 0;

    // if we can expect a response, remember how to route it
    if (!cache_responding && pkt->cacheResponding()) {
        assert(routeTo.find(pkt->req) == routeTo.end());
        routeTo[pkt->req] = mem_side_port_id;
    }

    // a snoop request came from a connected CPU-side-port device (one of
    // our memory-side ports), and if it is not coming from the CPU-side-port
    // device responsible for the address range something is
    // wrong, hence there is nothing further to do as the packet
    // would be going back to where it came from
    assert(findPort(pkt->getAddrRange()) == mem_side_port_id);
}

bool
CoherentXBar::recvTimingSnoopResp(PacketPtr pkt, PortID cpu_side_port_id)
{
    // determine the source port based on the id
    ResponsePort* src_port = cpuSidePorts[cpu_side_port_id];

    // get the destination
    const auto route_lookup = routeTo.find(pkt->req);
    assert(route_lookup != routeTo.end());
    const PortID dest_port_id = route_lookup->second;
    assert(dest_port_id != InvalidPortID);

    // determine if the response is from a snoop request we
    // created as the result of a normal request (in which case it
    // should be in the outstandingSnoop), or if we merely forwarded
    // someone else's snoop request
    const bool forwardAsSnoop = outstandingSnoop.find(pkt->req) ==
        outstandingSnoop.end();

    // test if the crossbar should be considered occupied for the
    // current port, note that the check is bypassed if the response
    // is being passed on as a normal response since this is occupying
    // the response layer rather than the snoop response layer
    if (forwardAsSnoop) {
        assert(dest_port_id < snoopLayers.size());
        if (!snoopLayers[dest_port_id]->tryTiming(src_port)) {
            DPRINTF(CoherentXBar, "%s: src %s packet %s BUSY\n", __func__,
                    src_port->name(), pkt->print());
            return false;
        }
    } else {
        // get the memory-side port that mirrors this CPU-side port internally
        RequestPort* snoop_port = snoopRespPorts[cpu_side_port_id];
        assert(dest_port_id < respLayers.size());
        if (!respLayers[dest_port_id]->tryTiming(snoop_port)) {
            DPRINTF(CoherentXBar, "%s: src %s packet %s BUSY\n", __func__,
                    snoop_port->name(), pkt->print());
            return false;
        }
    }

    DPRINTF(CoherentXBar, "%s: src %s packet %s\n", __func__,
            src_port->name(), pkt->print());

    // store size and command as they might be modified when
    // forwarding the packet
    unsigned int pkt_size = pkt->hasData() ? pkt->getSize() : 0;
    unsigned int pkt_cmd = pkt->cmdToIndex();

    // responses are never express snoops
    assert(!pkt->isExpressSnoop());

    // a snoop response sees the snoop response latency, and if it is
    // forwarded as a normal response, the response latency
    Tick xbar_delay =
        (forwardAsSnoop ? snoopResponseLatency : responseLatency) *
        clockPeriod();

    // set the packet header and payload delay
    calcPacketTiming(pkt, xbar_delay);

    // determine how long to be crossbar layer is busy
    Tick packetFinishTime = clockEdge(headerLatency) + pkt->payloadDelay;

    // forward it either as a snoop response or a normal response
    if (forwardAsSnoop) {
        // this is a snoop response to a snoop request we forwarded,
        // e.g. coming from the L1 and going to the L2, and it should
        // be forwarded as a snoop response

        if (snoopFilter) {
            // update the probe filter so that it can properly track the line
            snoopFilter->updateSnoopForward(pkt,
                            *cpuSidePorts[cpu_side_port_id],
                            *memSidePorts[dest_port_id]);
        }

        GEM5_VAR_USED bool success =
            memSidePorts[dest_port_id]->sendTimingSnoopResp(pkt);
        pktCount[cpu_side_port_id][dest_port_id]++;
        pktSize[cpu_side_port_id][dest_port_id] += pkt_size;
        assert(success);

        snoopLayers[dest_port_id]->succeededTiming(packetFinishTime);
    } else {
        // we got a snoop response on one of our CPU-side ports,
        // i.e. from a coherent requestor connected to the crossbar, and
        // since we created the snoop request as part of recvTiming,
        // this should now be a normal response again
        outstandingSnoop.erase(pkt->req);

        // this is a snoop response from a coherent requestor, hence it
        // should never go back to where the snoop response came from,
        // but instead to where the original request came from
        assert(cpu_side_port_id != dest_port_id);

        if (snoopFilter) {
            // update the probe filter so that it can properly track
            // the line
            snoopFilter->updateSnoopResponse(pkt,
                        *cpuSidePorts[cpu_side_port_id],
                        *cpuSidePorts[dest_port_id]);
        }

        DPRINTF(CoherentXBar, "%s: src %s packet %s FWD RESP\n", __func__,
                src_port->name(), pkt->print());

        // as a normal response, it should go back to a requestor through
        // one of our CPU-side ports, we also pay for any outstanding
        // header latency
        Tick latency = pkt->headerDelay;
        pkt->headerDelay = 0;
        cpuSidePorts[dest_port_id]->schedTimingResp(pkt,
                                    curTick() + latency);

        respLayers[dest_port_id]->succeededTiming(packetFinishTime);
    }

    // remove the request from the routing table
    routeTo.erase(route_lookup);

    // stats updates
    transDist[pkt_cmd]++;
    snoops++;
    snoopTraffic += pkt_size;

    return true;
}


void
CoherentXBar::forwardTiming(PacketPtr pkt, PortID exclude_cpu_side_port_id,
                           const std::vector<QueuedResponsePort*>& dests)
{
    DPRINTF(CoherentXBar, "%s for %s\n", __func__, pkt->print());

    // snoops should only happen if the system isn't bypassing caches
    assert(!system->bypassCaches());

    unsigned fanout = 0;

    for (const auto& p: dests) {
        // we could have gotten this request from a snooping requestor
        // (corresponding to our own CPU-side port that is also in
        // snoopPorts) and should not send it back to where it came
        // from
        if (exclude_cpu_side_port_id == InvalidPortID ||
            p->getId() != exclude_cpu_side_port_id) {
            // cache is not allowed to refuse snoop
            p->sendTimingSnoopReq(pkt);
            fanout++;
        }
    }

    // Stats for fanout of this forward operation
    snoopFanout.sample(fanout);
}

void
CoherentXBar::recvReqRetry(PortID mem_side_port_id)
{
    // responses and snoop responses never block on forwarding them,
    // so the retry will always be coming from a port to which we
    // tried to forward a request
    reqLayers[mem_side_port_id]->recvRetry();
}

Tick
CoherentXBar::recvAtomicBackdoor(PacketPtr pkt, PortID cpu_side_port_id,
                                 MemBackdoorPtr *backdoor)
{
    DPRINTF(CoherentXBar, "%s: src %s packet %s\n", __func__,
            cpuSidePorts[cpu_side_port_id]->name(), pkt->print());

    unsigned int pkt_size = pkt->hasData() ? pkt->getSize() : 0;
    unsigned int pkt_cmd = pkt->cmdToIndex();

    MemCmd snoop_response_cmd = MemCmd::InvalidCmd;
    Tick snoop_response_latency = 0;

    // is this the destination point for this packet? (e.g. true if
    // this xbar is the PoC for a cache maintenance operation to the
    // PoC) otherwise the destination is any cache that can satisfy
    // the request
    const bool is_destination = isDestination(pkt);

    const bool snoop_caches = !system->bypassCaches() &&
        pkt->cmd != MemCmd::WriteClean;
    if (snoop_caches) {
        // forward to all snoopers but the source
        std::pair<MemCmd, Tick> snoop_result;
        if (snoopFilter) {
            // check with the snoop filter where to forward this packet
            auto sf_res =
                snoopFilter->lookupRequest(pkt,
                *cpuSidePorts [cpu_side_port_id]);
            snoop_response_latency += sf_res.second * clockPeriod();
            DPRINTF(CoherentXBar, "%s: src %s packet %s SF size: %i lat: %i\n",
                    __func__, cpuSidePorts[cpu_side_port_id]->name(),
                    pkt->print(), sf_res.first.size(), sf_res.second);

            // let the snoop filter know about the success of the send
            // operation, and do it even before sending it onwards to
            // avoid situations where atomic upward snoops sneak in
            // between and change the filter state
            snoopFilter->finishRequest(false, pkt->getAddr(), pkt->isSecure());

            if (pkt->isEviction()) {
                // for block-evicting packets, i.e. writebacks and
                // clean evictions, there is no need to snoop up, as
                // all we do is determine if the block is cached or
                // not, instead just set it here based on the snoop
                // filter result
                if (!sf_res.first.empty())
                    pkt->setBlockCached();
            } else {
                snoop_result = forwardAtomic(pkt, cpu_side_port_id,
                                            InvalidPortID, sf_res.first);
            }
        } else {
            snoop_result = forwardAtomic(pkt, cpu_side_port_id);
        }
        snoop_response_cmd = snoop_result.first;
        snoop_response_latency += snoop_result.second;
    }

    // set up a sensible default value
    Tick response_latency = 0;

    const bool sink_packet = sinkPacket(pkt);

    // even if we had a snoop response, we must continue and also
    // perform the actual request at the destination
    PortID mem_side_port_id = findPort(pkt->getAddrRange());

    if (sink_packet) {
        DPRINTF(CoherentXBar, "%s: Not forwarding %s\n", __func__,
                pkt->print());
    } else {
        if (forwardPacket(pkt)) {
            // make sure that the write request (e.g., WriteClean)
            // will stop at the memory below if this crossbar is its
            // destination
            if (pkt->isWrite() && is_destination) {
                pkt->clearWriteThrough();
            }

            // forward the request to the appropriate destination
            auto mem_side_port = memSidePorts[mem_side_port_id];
            response_latency = backdoor ?
                mem_side_port->sendAtomicBackdoor(pkt, *backdoor) :
                mem_side_port->sendAtomic(pkt);
        } else {
            // if it does not need a response we sink the packet above
            assert(pkt->needsResponse());

            pkt->makeResponse();
        }
    }

    // stats updates for the request
    pktCount[cpu_side_port_id][mem_side_port_id]++;
    pktSize[cpu_side_port_id][mem_side_port_id] += pkt_size;
    transDist[pkt_cmd]++;


    // if lower levels have replied, tell the snoop filter
    if (!system->bypassCaches() && snoopFilter && pkt->isResponse()) {
        snoopFilter->updateResponse(pkt, *cpuSidePorts[cpu_side_port_id]);
    }

    // if we got a response from a snooper, restore it here
    if (snoop_response_cmd != MemCmd::InvalidCmd) {
        // no one else should have responded
        assert(!pkt->isResponse());
        pkt->cmd = snoop_response_cmd;
        response_latency = snoop_response_latency;
    }

    // If this is the destination of the cache clean operation the
    // crossbar is responsible for responding. This crossbar will
    // respond when the cache clean is complete. An atomic cache clean
    // is complete when the crossbars receives the cache clean
    // request (CleanSharedReq, CleanInvalidReq), as either:
    // * no cache above had a dirty copy of the block as indicated by
    //   the satisfied flag of the packet, or
    // * the crossbar has already seen the corresponding write
    //   (WriteClean) which updates the block in the memory below.
    if (pkt->isClean() && isDestination(pkt) && pkt->satisfied()) {
        auto it = outstandingCMO.find(pkt->id);
        assert(it != outstandingCMO.end());
        // we are responding right away
        outstandingCMO.erase(it);
    } else if (pkt->cmd == MemCmd::WriteClean && isDestination(pkt)) {
        // if this is the destination of the operation, the xbar
        // sends the responce to the cache clean operation only
        // after having encountered the cache clean request
        GEM5_VAR_USED auto ret = outstandingCMO.emplace(pkt->id, nullptr);
        // in atomic mode we know that the WriteClean packet should
        // precede the clean request
        assert(ret.second);
    }

    // add the response data
    if (pkt->isResponse()) {
        pkt_size = pkt->hasData() ? pkt->getSize() : 0;
        pkt_cmd = pkt->cmdToIndex();

        // stats updates
        pktCount[cpu_side_port_id][mem_side_port_id]++;
        pktSize[cpu_side_port_id][mem_side_port_id] += pkt_size;
        transDist[pkt_cmd]++;
    }

    // @todo: Not setting header time
    pkt->payloadDelay = response_latency;
    return response_latency;
}

Tick
CoherentXBar::recvAtomicSnoop(PacketPtr pkt, PortID mem_side_port_id)
{
    DPRINTF(CoherentXBar, "%s: src %s packet %s\n", __func__,
            memSidePorts[mem_side_port_id]->name(), pkt->print());

    // add the request snoop data
    unsigned int pkt_size = pkt->hasData() ? pkt->getSize() : 0;
    snoops++;
    snoopTraffic += pkt_size;

    // forward to all snoopers
    std::pair<MemCmd, Tick> snoop_result;
    Tick snoop_response_latency = 0;
    if (snoopFilter) {
        auto sf_res = snoopFilter->lookupSnoop(pkt);
        snoop_response_latency += sf_res.second * clockPeriod();
        DPRINTF(CoherentXBar, "%s: src %s packet %s SF size: %i lat: %i\n",
                __func__, memSidePorts[mem_side_port_id]->name(),
                pkt->print(), sf_res.first.size(), sf_res.second);
        snoop_result = forwardAtomic(pkt, InvalidPortID, mem_side_port_id,
                                     sf_res.first);
    } else {
        snoop_result = forwardAtomic(pkt, InvalidPortID);
    }
    MemCmd snoop_response_cmd = snoop_result.first;
    snoop_response_latency += snoop_result.second;

    if (snoop_response_cmd != MemCmd::InvalidCmd)
        pkt->cmd = snoop_response_cmd;

    // add the response snoop data
    if (pkt->isResponse()) {
        snoops++;
    }

    // @todo: Not setting header time
    pkt->payloadDelay = snoop_response_latency;
    return snoop_response_latency;
}

std::pair<MemCmd, Tick>
CoherentXBar::forwardAtomic(PacketPtr pkt, PortID exclude_cpu_side_port_id,
                           PortID source_mem_side_port_id,
                           const std::vector<QueuedResponsePort*>& dests)
{
    // the packet may be changed on snoops, record the original
    // command to enable us to restore it between snoops so that
    // additional snoops can take place properly
    MemCmd orig_cmd = pkt->cmd;
    MemCmd snoop_response_cmd = MemCmd::InvalidCmd;
    Tick snoop_response_latency = 0;

    // snoops should only happen if the system isn't bypassing caches
    assert(!system->bypassCaches());

    unsigned fanout = 0;

    for (const auto& p: dests) {
        // we could have gotten this request from a snooping memory-side port
        // (corresponding to our own CPU-side port that is also in
        // snoopPorts) and should not send it back to where it came
        // from
        if (exclude_cpu_side_port_id != InvalidPortID &&
            p->getId() == exclude_cpu_side_port_id)
            continue;

        Tick latency = p->sendAtomicSnoop(pkt);
        fanout++;

        // in contrast to a functional access, we have to keep on
        // going as all snoopers must be updated even if we get a
        // response
        if (!pkt->isResponse())
            continue;

        // response from snoop agent
        assert(pkt->cmd != orig_cmd);
        assert(pkt->cacheResponding());
        // should only happen once
        assert(snoop_response_cmd == MemCmd::InvalidCmd);
        // save response state
        snoop_response_cmd = pkt->cmd;
        snoop_response_latency = latency;

        if (snoopFilter) {
            // Handle responses by the snoopers and differentiate between
            // responses to requests from above and snoops from below
            if (source_mem_side_port_id != InvalidPortID) {
                // Getting a response for a snoop from below
                assert(exclude_cpu_side_port_id == InvalidPortID);
                snoopFilter->updateSnoopForward(pkt, *p,
                             *memSidePorts[source_mem_side_port_id]);
            } else {
                // Getting a response for a request from above
                assert(source_mem_side_port_id == InvalidPortID);
                snoopFilter->updateSnoopResponse(pkt, *p,
                             *cpuSidePorts[exclude_cpu_side_port_id]);
            }
        }
        // restore original packet state for remaining snoopers
        pkt->cmd = orig_cmd;
    }

    // Stats for fanout
    snoopFanout.sample(fanout);

    // the packet is restored as part of the loop and any potential
    // snoop response is part of the returned pair
    return std::make_pair(snoop_response_cmd, snoop_response_latency);
}

void
CoherentXBar::recvFunctional(PacketPtr pkt, PortID cpu_side_port_id)
{
    if (!pkt->isPrint()) {
        // don't do DPRINTFs on PrintReq as it clutters up the output
        DPRINTF(CoherentXBar, "%s: src %s packet %s\n", __func__,
                cpuSidePorts[cpu_side_port_id]->name(), pkt->print());
    }

    if (!system->bypassCaches()) {
        // forward to all snoopers but the source
        forwardFunctional(pkt, cpu_side_port_id);
    }

    // there is no need to continue if the snooping has found what we
    // were looking for and the packet is already a response
    if (!pkt->isResponse()) {
        // since our CPU-side ports are queued ports we need to check
        // them as well
        for (const auto& p : cpuSidePorts) {
            // if we find a response that has the data, then the
            // downstream caches/memories may be out of date, so simply stop
            // here
            if (p->trySatisfyFunctional(pkt)) {
                if (pkt->needsResponse())
                    pkt->makeResponse();
                return;
            }
        }

        PortID dest_id = findPort(pkt->getAddrRange());

        memSidePorts[dest_id]->sendFunctional(pkt);
    }
}

void
CoherentXBar::recvFunctionalSnoop(PacketPtr pkt, PortID mem_side_port_id)
{
    if (!pkt->isPrint()) {
        // don't do DPRINTFs on PrintReq as it clutters up the output
        DPRINTF(CoherentXBar, "%s: src %s packet %s\n", __func__,
                memSidePorts[mem_side_port_id]->name(), pkt->print());
    }

    for (const auto& p : cpuSidePorts) {
        if (p->trySatisfyFunctional(pkt)) {
            if (pkt->needsResponse())
                pkt->makeResponse();
            return;
        }
    }

    // forward to all snoopers
    forwardFunctional(pkt, InvalidPortID);
}

void
CoherentXBar::forwardFunctional(PacketPtr pkt, PortID exclude_cpu_side_port_id)
{
    // snoops should only happen if the system isn't bypassing caches
    assert(!system->bypassCaches());

    for (const auto& p: snoopPorts) {
        // we could have gotten this request from a snooping requestor
        // (corresponding to our own CPU-side port that is also in
        // snoopPorts) and should not send it back to where it came
        // from
        if (exclude_cpu_side_port_id == InvalidPortID ||
            p->getId() != exclude_cpu_side_port_id)
            p->sendFunctionalSnoop(pkt);

        // if we get a response we are done
        if (pkt->isResponse()) {
            break;
        }
    }
}

bool
CoherentXBar::sinkPacket(const PacketPtr pkt) const
{
    // we can sink the packet if:
    // 1) the crossbar is the point of coherency, and a cache is
    //    responding after being snooped
    // 2) the crossbar is the point of coherency, and the packet is a
    //    coherency packet (not a read or a write) that does not
    //    require a response
    // 3) this is a clean evict or clean writeback, but the packet is
    //    found in a cache above this crossbar
    // 4) a cache is responding after being snooped, and the packet
    //    either does not need the block to be writable, or the cache
    //    that has promised to respond (setting the cache responding
    //    flag) is providing writable and thus had a Modified block,
    //    and no further action is needed
    return (pointOfCoherency && pkt->cacheResponding()) ||
        (pointOfCoherency && !(pkt->isRead() || pkt->isWrite()) &&
         !pkt->needsResponse()) ||
        (pkt->isCleanEviction() && pkt->isBlockCached()) ||
        (pkt->cacheResponding() &&
         (!pkt->needsWritable() || pkt->responderHadWritable()));
}

bool
CoherentXBar::forwardPacket(const PacketPtr pkt)
{
    // we are forwarding the packet if:
    // 1) this is a cache clean request to the PoU/PoC and this
    //    crossbar is above the PoU/PoC
    // 2) this is a read or a write
    // 3) this crossbar is above the point of coherency
    if (pkt->isClean()) {
        return !isDestination(pkt);
    }
    return pkt->isRead() || pkt->isWrite() || !pointOfCoherency;
}


void
CoherentXBar::regStats()
{
    BaseXBar::regStats();

    snoopFanout.init(0, snoopPorts.size(), 1);
}