mem: hmc: serial link model

This changeset adds a serial link model for the Hybrid Memory Cube (HMC).
SerialLink is a simple variation of the Bridge class, with the ability to
account for the latency of packet serialization. Also trySendTiming has been
modified to correctly model bandwidth.

Committed by: Nilay Vaish <nilay@cs.wisc.edu>

1 files changed, 437 insertions, 0 deletions

diff --git a/src/mem/serial_link.cc b/src/mem/serial_link.cc
new file mode 100644
index 000000000..577339bb8
--- /dev/null
+++ b/src/mem/serial_link.cc
@@ -0,0 +1,437 @@
+/*
+ * Copyright (c) 2011-2013 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
+ * Copyright (c) 2015 The University of Bologna
+ * 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: Ali Saidi
+ *          Steve Reinhardt
+ *          Andreas Hansson
+ *          Erfan Azarkhish
+ */
+
+/**
+ * @file
+ * Implementation of the SerialLink Class, modeling Hybrid-Memory-Cube's
+ * serial interface.
+ */
+
+#include "mem/serial_link.hh"
+
+#include "base/trace.hh"
+#include "debug/SerialLink.hh"
+#include "params/SerialLink.hh"
+
+
+SerialLink::SerialLinkSlavePort::SerialLinkSlavePort(const std::string& _name,
+                                         SerialLink& _serial_link,
+                                         SerialLinkMasterPort& _masterPort,
+                                         Cycles _delay, int _resp_limit,
+                                         const std::vector<AddrRange>&
+                                         _ranges)
+    : SlavePort(_name, &_serial_link), serial_link(_serial_link),
+      masterPort(_masterPort), delay(_delay),
+      ranges(_ranges.begin(), _ranges.end()),
+      outstandingResponses(0), retryReq(false),
+      respQueueLimit(_resp_limit), sendEvent(*this)
+{
+}
+
+SerialLink::SerialLinkMasterPort::SerialLinkMasterPort(const std::string&
+                                           _name, SerialLink& _serial_link,
+                                           SerialLinkSlavePort& _slavePort,
+                                           Cycles _delay, int _req_limit)
+    : MasterPort(_name, &_serial_link), serial_link(_serial_link),
+      slavePort(_slavePort), delay(_delay), reqQueueLimit(_req_limit),
+      sendEvent(*this)
+{
+}
+
+SerialLink::SerialLink(SerialLinkParams *p)
+    : MemObject(p),
+      slavePort(p->name + ".slave", *this, masterPort,
+                ticksToCycles(p->delay), p->resp_size, p->ranges),
+      masterPort(p->name + ".master", *this, slavePort,
+                 ticksToCycles(p->delay), p->req_size),
+      num_lanes(p->num_lanes)
+{
+}
+
+BaseMasterPort&
+SerialLink::getMasterPort(const std::string &if_name, PortID idx)
+{
+    if (if_name == "master")
+        return masterPort;
+    else
+        // pass it along to our super class
+        return MemObject::getMasterPort(if_name, idx);
+}
+
+BaseSlavePort&
+SerialLink::getSlavePort(const std::string &if_name, PortID idx)
+{
+    if (if_name == "slave")
+        return slavePort;
+    else
+        // pass it along to our super class
+        return MemObject::getSlavePort(if_name, idx);
+}
+
+void
+SerialLink::init()
+{
+    // make sure both sides are connected and have the same block size
+    if (!slavePort.isConnected() || !masterPort.isConnected())
+        fatal("Both ports of a serial_link must be connected.\n");
+
+    // notify the master side  of our address ranges
+    slavePort.sendRangeChange();
+}
+
+bool
+SerialLink::SerialLinkSlavePort::respQueueFull() const
+{
+    return outstandingResponses == respQueueLimit;
+}
+
+bool
+SerialLink::SerialLinkMasterPort::reqQueueFull() const
+{
+    return transmitList.size() == reqQueueLimit;
+}
+
+bool
+SerialLink::SerialLinkMasterPort::recvTimingResp(PacketPtr pkt)
+{
+    // all checks are done when the request is accepted on the slave
+    // side, so we are guaranteed to have space for the response
+    DPRINTF(SerialLink, "recvTimingResp: %s addr 0x%x\n",
+            pkt->cmdString(), pkt->getAddr());
+
+    DPRINTF(SerialLink, "Request queue size: %d\n", transmitList.size());
+
+    // @todo: We need to pay for this and not just zero it out
+    pkt->headerDelay = pkt->payloadDelay = 0;
+
+    // This is similar to what happens for the request packets:
+    // The serializer will start serialization as soon as it receives the
+    // first flit, but the deserializer (at the host side in this case), will
+    // have to wait to receive the whole packet. So we only account for the
+    // deserialization latency.
+    Cycles cycles = delay;
+    cycles += Cycles(divCeil(pkt->getSize() * 8, serial_link.num_lanes));
+    Tick t = serial_link.clockEdge(cycles);
+
+    //@todo: If the processor sends two uncached requests towards HMC and the
+    // second one is smaller than the first one. It may happen that the second
+    // one crosses this link faster than the first one (because the packet
+    // waits in the link based on its size). This can reorder the received
+    // response.
+    slavePort.schedTimingResp(pkt, t);
+
+    return true;
+}
+
+bool
+SerialLink::SerialLinkSlavePort::recvTimingReq(PacketPtr pkt)
+{
+    DPRINTF(SerialLink, "recvTimingReq: %s addr 0x%x\n",
+            pkt->cmdString(), pkt->getAddr());
+
+    // we should not see a timing request if we are already in a retry
+    assert(!retryReq);
+
+    DPRINTF(SerialLink, "Response queue size: %d outresp: %d\n",
+            transmitList.size(), outstandingResponses);
+
+    // if the request queue is full then there is no hope
+    if (masterPort.reqQueueFull()) {
+        DPRINTF(SerialLink, "Request queue full\n");
+        retryReq = true;
+    } else if ( !retryReq ) {
+        // look at the response queue if we expect to see a response
+        bool expects_response = pkt->needsResponse() &&
+            !pkt->memInhibitAsserted();
+        if (expects_response) {
+            if (respQueueFull()) {
+                DPRINTF(SerialLink, "Response queue full\n");
+                retryReq = true;
+            } else {
+                // ok to send the request with space for the response
+                DPRINTF(SerialLink, "Reserving space for response\n");
+                assert(outstandingResponses != respQueueLimit);
+                ++outstandingResponses;
+
+                // no need to set retryReq to false as this is already the
+                // case
+            }
+        }
+
+        if (!retryReq) {
+            // @todo: We need to pay for this and not just zero it out
+            pkt->headerDelay = pkt->payloadDelay = 0;
+
+            // We assume that the serializer component at the transmitter side
+            // does not need to receive the whole packet to start the
+            // serialization (this assumption is consistent with the HMC
+            // standard). But the deserializer waits for the complete packet
+            // to check its integrity first. So everytime a packet crosses a
+            // serial link, we should account for its deserialization latency
+            // only.
+            Cycles cycles = delay;
+            cycles += Cycles(divCeil(pkt->getSize() * 8,
+                serial_link.num_lanes));
+            Tick t = serial_link.clockEdge(cycles);
+
+            //@todo: If the processor sends two uncached requests towards HMC
+            // and the second one is smaller than the first one. It may happen
+            // that the second one crosses this link faster than the first one
+            // (because the packet waits in the link based on its size).
+            // This can reorder the received response.
+            masterPort.schedTimingReq(pkt, t);
+        }
+    }
+
+    // remember that we are now stalling a packet and that we have to
+    // tell the sending master to retry once space becomes available,
+    // we make no distinction whether the stalling is due to the
+    // request queue or response queue being full
+    return !retryReq;
+}
+
+void
+SerialLink::SerialLinkSlavePort::retryStalledReq()
+{
+    if (retryReq) {
+        DPRINTF(SerialLink, "Request waiting for retry, now retrying\n");
+        retryReq = false;
+        sendRetryReq();
+    }
+}
+
+void
+SerialLink::SerialLinkMasterPort::schedTimingReq(PacketPtr pkt, Tick when)
+{
+    // If we're about to put this packet at the head of the queue, we
+    // need to schedule an event to do the transmit.  Otherwise there
+    // should already be an event scheduled for sending the head
+    // packet.
+    if (transmitList.empty()) {
+        serial_link.schedule(sendEvent, when);
+    }
+
+    assert(transmitList.size() != reqQueueLimit);
+
+    transmitList.emplace_back(DeferredPacket(pkt, when));
+}
+
+
+void
+SerialLink::SerialLinkSlavePort::schedTimingResp(PacketPtr pkt, Tick when)
+{
+    // If we're about to put this packet at the head of the queue, we
+    // need to schedule an event to do the transmit.  Otherwise there
+    // should already be an event scheduled for sending the head
+    // packet.
+    if (transmitList.empty()) {
+        serial_link.schedule(sendEvent, when);
+    }
+
+    transmitList.emplace_back(DeferredPacket(pkt, when));
+}
+
+void
+SerialLink::SerialLinkMasterPort::trySendTiming()
+{
+    assert(!transmitList.empty());
+
+    DeferredPacket req = transmitList.front();
+
+    assert(req.tick <= curTick());
+
+    PacketPtr pkt = req.pkt;
+
+    DPRINTF(SerialLink, "trySend request addr 0x%x, queue size %d\n",
+            pkt->getAddr(), transmitList.size());
+
+    if (sendTimingReq(pkt)) {
+        // send successful
+        transmitList.pop_front();
+
+        DPRINTF(SerialLink, "trySend request successful\n");
+
+        // If there are more packets to send, schedule event to try again.
+        if (!transmitList.empty()) {
+            DeferredPacket next_req = transmitList.front();
+            DPRINTF(SerialLink, "Scheduling next send\n");
+
+            // Make sure bandwidth limitation is met
+            Cycles cycles = Cycles(divCeil(pkt->getSize() * 8,
+                serial_link.num_lanes));
+            Tick t = serial_link.clockEdge(cycles);
+            serial_link.schedule(sendEvent, std::max(next_req.tick, t));
+        }
+
+        // if we have stalled a request due to a full request queue,
+        // then send a retry at this point, also note that if the
+        // request we stalled was waiting for the response queue
+        // rather than the request queue we might stall it again
+        slavePort.retryStalledReq();
+    }
+
+    // if the send failed, then we try again once we receive a retry,
+    // and therefore there is no need to take any action
+}
+
+void
+SerialLink::SerialLinkSlavePort::trySendTiming()
+{
+    assert(!transmitList.empty());
+
+    DeferredPacket resp = transmitList.front();
+
+    assert(resp.tick <= curTick());
+
+    PacketPtr pkt = resp.pkt;
+
+    DPRINTF(SerialLink, "trySend response addr 0x%x, outstanding %d\n",
+            pkt->getAddr(), outstandingResponses);
+
+    if (sendTimingResp(pkt)) {
+        // send successful
+        transmitList.pop_front();
+        DPRINTF(SerialLink, "trySend response successful\n");
+
+        assert(outstandingResponses != 0);
+        --outstandingResponses;
+
+        // If there are more packets to send, schedule event to try again.
+        if (!transmitList.empty()) {
+            DeferredPacket next_resp = transmitList.front();
+            DPRINTF(SerialLink, "Scheduling next send\n");
+
+            // Make sure bandwidth limitation is met
+            Cycles cycles = Cycles(divCeil(pkt->getSize() * 8,
+                serial_link.num_lanes));
+            Tick t = serial_link.clockEdge(cycles);
+            serial_link.schedule(sendEvent, std::max(next_resp.tick, t));
+        }
+
+        // if there is space in the request queue and we were stalling
+        // a request, it will definitely be possible to accept it now
+        // since there is guaranteed space in the response queue
+        if (!masterPort.reqQueueFull() && retryReq) {
+            DPRINTF(SerialLink, "Request waiting for retry, now retrying\n");
+            retryReq = false;
+            sendRetryReq();
+        }
+    }
+
+    // if the send failed, then we try again once we receive a retry,
+    // and therefore there is no need to take any action
+}
+
+void
+SerialLink::SerialLinkMasterPort::recvReqRetry()
+{
+    trySendTiming();
+}
+
+void
+SerialLink::SerialLinkSlavePort::recvRespRetry()
+{
+    trySendTiming();
+}
+
+Tick
+SerialLink::SerialLinkSlavePort::recvAtomic(PacketPtr pkt)
+{
+    return delay * serial_link.clockPeriod() + masterPort.sendAtomic(pkt);
+}
+
+void
+SerialLink::SerialLinkSlavePort::recvFunctional(PacketPtr pkt)
+{
+    pkt->pushLabel(name());
+
+    // check the response queue
+    for (auto i = transmitList.begin();  i != transmitList.end(); ++i) {
+        if (pkt->checkFunctional((*i).pkt)) {
+            pkt->makeResponse();
+            return;
+        }
+    }
+
+    // also check the master port's request queue
+    if (masterPort.checkFunctional(pkt)) {
+        return;
+    }
+
+    pkt->popLabel();
+
+    // fall through if pkt still not satisfied
+    masterPort.sendFunctional(pkt);
+}
+
+bool
+SerialLink::SerialLinkMasterPort::checkFunctional(PacketPtr pkt)
+{
+    bool found = false;
+    auto i = transmitList.begin();
+
+    while(i != transmitList.end() && !found) {
+        if (pkt->checkFunctional((*i).pkt)) {
+            pkt->makeResponse();
+            found = true;
+        }
+        ++i;
+    }
+
+    return found;
+}
+
+AddrRangeList
+SerialLink::SerialLinkSlavePort::getAddrRanges() const
+{
+    return ranges;
+}
+
+SerialLink *
+SerialLinkParams::create()
+{
+    return new SerialLink(this);
+}