src/mem/cache/cache_impl.hh

   1 /*
   2  * Copyright (c) 2002-2005 The Regents of The University of Michigan
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions are
   7  * met: redistributions of source code must retain the above copyright
   8  * notice, this list of conditions and the following disclaimer;
   9  * redistributions in binary form must reproduce the above copyright
  10  * notice, this list of conditions and the following disclaimer in the
  11  * documentation and/or other materials provided with the distribution;
  12  * neither the name of the copyright holders nor the names of its
  13  * contributors may be used to endorse or promote products derived from
  14  * this software without specific prior written permission.
  15  *
  16  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  17  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  18  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  19  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  20  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  22  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  26  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  27  *
  28  * Authors: Erik Hallnor
  29  *          Dave Greene
  30  *          Nathan Binkert
  31  */
  32
  33 /**
  34  * @file
  35  * Cache definitions.
  36  */
  37
  38 #include <assert.h>
  39 #include <math.h>
  40
  41 #include <cassert>
  42 #include <iostream>
  43 #include <string>
  44
  45 #include "sim/host.hh"
  46 #include "base/misc.hh"
  47 #include "cpu/smt.hh"
  48
  49 #include "mem/cache/cache.hh"
  50 #include "mem/cache/cache_blk.hh"
  51 #include "mem/cache/miss/mshr.hh"
  52 #include "mem/cache/prefetch/prefetcher.hh"
  53
  54 #include "sim/sim_exit.hh" // for SimExitEvent
  55
  56 template<class TagStore, class Buffering, class Coherence>
  57 bool
  58 Cache<TagStore,Buffering,Coherence>::
  59 doTimingAccess(Packet *pkt, CachePort *cachePort, bool isCpuSide)
  60 {
  61     if (isCpuSide)
  62     {
  63         if (pkt->isWrite() && (pkt->req->getFlags() & LOCKED)) {
  64             pkt->req->setScResult(1);
  65         }
  66         if (!(pkt->flags & SATISFIED)) {
  67             access(pkt);
  68         }
  69     }
  70     else
  71     {
  72         if (pkt->isResponse())
  73             handleResponse(pkt);
  74         else {
  75             //Check if we are in phase1
  76             if (!snoopPhase2) {
  77                 snoopPhase2 = true;
  78             }
  79             else {
  80                 //Check if we should do the snoop
  81                 if (pkt->flags && SNOOP_COMMIT)
  82                     snoop(pkt);
  83                 snoopPhase2 = false;
  84             }
  85         }
  86     }
  87     return true;
  88 }
  89
  90 template<class TagStore, class Buffering, class Coherence>
  91 Tick
  92 Cache<TagStore,Buffering,Coherence>::
  93 doAtomicAccess(Packet *pkt, bool isCpuSide)
  94 {
  95     if (isCpuSide)
  96     {
  97         //Temporary solution to LL/SC
  98         if (pkt->isWrite() && (pkt->req->getFlags() & LOCKED)) {
  99             pkt->req->setScResult(1);
 100         }
 101
 102         probe(pkt, true, NULL);
 103         //TEMP ALWAYS SUCCES FOR NOW
 104         pkt->result = Packet::Success;
 105     }
 106     else
 107     {
 108         if (pkt->isResponse())
 109             handleResponse(pkt);
 110         else
 111             snoopProbe(pkt);
 112     }
 113     //Fix this timing info
 114     return hitLatency;
 115 }
 116
 117 template<class TagStore, class Buffering, class Coherence>
 118 void
 119 Cache<TagStore,Buffering,Coherence>::
 120 doFunctionalAccess(Packet *pkt, bool isCpuSide)
 121 {
 122     if (isCpuSide)
 123     {
 124         //TEMP USE CPU?THREAD 0 0
 125         pkt->req->setThreadContext(0,0);
 126
 127         //Temporary solution to LL/SC
 128         if (pkt->isWrite() && (pkt->req->getFlags() & LOCKED)) {
 129             assert("Can't handle LL/SC on functional path\n");
 130         }
 131
 132         probe(pkt, false, memSidePort);
 133         //TEMP ALWAYS SUCCESFUL FOR NOW
 134         pkt->result = Packet::Success;
 135     }
 136     else
 137     {
 138             probe(pkt, false, cpuSidePort);
 139     }
 140 }
 141
 142 template<class TagStore, class Buffering, class Coherence>
 143 void
 144 Cache<TagStore,Buffering,Coherence>::
 145 recvStatusChange(Port::Status status, bool isCpuSide)
 146 {
 147
 148 }
 149
 150
 151 template<class TagStore, class Buffering, class Coherence>
 152 Cache<TagStore,Buffering,Coherence>::
 153 Cache(const std::string &_name,
 154       Cache<TagStore,Buffering,Coherence>::Params &params)
 155     : BaseCache(_name, params.baseParams),
 156       prefetchAccess(params.prefetchAccess),
 157       tags(params.tags), missQueue(params.missQueue),
 158       coherence(params.coherence), prefetcher(params.prefetcher),
 159       doCopy(params.doCopy), blockOnCopy(params.blockOnCopy)
 160 {
 161 //FIX BUS POINTERS
 162 //    if (params.in == NULL) {
 163         topLevelCache = true;
 164 //    }
 165 //PLEASE FIX THIS, BUS SIZES NOT BEING USED
 166         tags->setCache(this, blkSize, 1/*params.out->width, params.out->clockRate*/);
 167     tags->setPrefetcher(prefetcher);
 168     missQueue->setCache(this);
 169     missQueue->setPrefetcher(prefetcher);
 170     coherence->setCache(this);
 171     prefetcher->setCache(this);
 172     prefetcher->setTags(tags);
 173     prefetcher->setBuffer(missQueue);
 174 #if 0
 175     invalidatePkt = new Packet;
 176     invalidatePkt->cmd = Packet::InvalidateReq;
 177 #endif
 178 }
 179
 180 template<class TagStore, class Buffering, class Coherence>
 181 void
 182 Cache<TagStore,Buffering,Coherence>::regStats()
 183 {
 184     BaseCache::regStats();
 185     tags->regStats(name());
 186     missQueue->regStats(name());
 187     coherence->regStats(name());
 188     prefetcher->regStats(name());
 189 }
 190
 191 template<class TagStore, class Buffering, class Coherence>
 192 bool
 193 Cache<TagStore,Buffering,Coherence>::access(PacketPtr &pkt)
 194 {
 195 //@todo Add back in MemDebug Calls
 196 //    MemDebug::cacheAccess(pkt);
 197     BlkType *blk = NULL;
 198     PacketList writebacks;
 199     int size = blkSize;
 200     int lat = hitLatency;
 201     if (prefetchAccess) {
 202         //We are determining prefetches on access stream, call prefetcher
 203         prefetcher->handleMiss(pkt, curTick);
 204     }
 205     if (!pkt->req->isUncacheable()) {
 206         if (pkt->isInvalidate() && !pkt->isRead()
 207             && !pkt->isWrite()) {
 208             //Upgrade or Invalidate
 209             //Look into what happens if two slave caches on bus
 210             DPRINTF(Cache, "%s %x ? blk_addr: %x\n", pkt->cmdString(),
 211                     pkt->getAddr() & (((ULL(1))<<48)-1),
 212                     pkt->getAddr() & ~((Addr)blkSize - 1));
 213
 214             //@todo Should this return latency have the hit latency in it?
 215 //          respond(pkt,curTick+lat);
 216             pkt->flags |= SATISFIED;
 217 //            return MA_HIT; //@todo, return values
 218             return true;
 219         }
 220         blk = tags->handleAccess(pkt, lat, writebacks);
 221     } else {
 222         size = pkt->getSize();
 223     }
 224     // If this is a block size write/hint (WH64) allocate the block here
 225     // if the coherence protocol allows it.
 226     /** @todo make the fast write alloc (wh64) work with coherence. */
 227     /** @todo Do we want to do fast writes for writebacks as well? */
 228     if (!blk && pkt->getSize() >= blkSize && coherence->allowFastWrites() &&
 229         (pkt->cmd == Packet::WriteReq || pkt->cmd == Packet::WriteInvalidateReq) ) {
 230         // not outstanding misses, can do this
 231         MSHR* outstanding_miss = missQueue->findMSHR(pkt->getAddr());
 232         if (pkt->cmd == Packet::WriteInvalidateReq || !outstanding_miss) {
 233             if (outstanding_miss) {
 234                 warn("WriteInv doing a fastallocate"
 235                      "with an outstanding miss to the same address\n");
 236             }
 237             blk = tags->handleFill(NULL, pkt, BlkValid | BlkWritable,
 238                                    writebacks);
 239             ++fastWrites;
 240         }
 241     }
 242     while (!writebacks.empty()) {
 243         missQueue->doWriteback(writebacks.front());
 244         writebacks.pop_front();
 245     }
 246     DPRINTF(Cache, "%s %x %s blk_addr: %x pc %x\n", pkt->cmdString(),
 247             pkt->getAddr() & (((ULL(1))<<48)-1), (blk) ? "hit" : "miss",
 248             pkt->getAddr() & ~((Addr)blkSize - 1), pkt->req->getPC());
 249     if (blk) {
 250         // Hit
 251         hits[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
 252         // clear dirty bit if write through
 253         if (pkt->needsResponse())
 254             respond(pkt, curTick+lat);
 255 //      return MA_HIT;
 256         return true;
 257     }
 258
 259     // Miss
 260     if (!pkt->req->isUncacheable()) {
 261         misses[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
 262         /** @todo Move miss count code into BaseCache */
 263         if (missCount) {
 264             --missCount;
 265             if (missCount == 0)
 266                 exitSimLoop("A cache reached the maximum miss count");
 267         }
 268     }
 269     missQueue->handleMiss(pkt, size, curTick + hitLatency);
 270 //    return MA_CACHE_MISS;
 271     return true;
 272 }
 273
 274
 275 template<class TagStore, class Buffering, class Coherence>
 276 Packet *
 277 Cache<TagStore,Buffering,Coherence>::getPacket()
 278 {
 279     Packet * pkt = missQueue->getPacket();
 280     if (pkt) {
 281         if (!pkt->req->isUncacheable()) {
 282             if (pkt->cmd == Packet::HardPFReq) misses[Packet::HardPFReq][0/*pkt->req->getThreadNum()*/]++;
 283             BlkType *blk = tags->findBlock(pkt);
 284             Packet::Command cmd = coherence->getBusCmd(pkt->cmd,
 285                                               (blk)? blk->status : 0);
 286             missQueue->setBusCmd(pkt, cmd);
 287         }
 288     }
 289
 290     assert(!doMasterRequest() || missQueue->havePending());
 291     assert(!pkt || pkt->time <= curTick);
 292     return pkt;
 293 }
 294
 295 template<class TagStore, class Buffering, class Coherence>
 296 void
 297 Cache<TagStore,Buffering,Coherence>::sendResult(PacketPtr &pkt, bool success)
 298 {
 299     if (success) {
 300         missQueue->markInService(pkt);
 301           //Temp Hack for UPGRADES
 302           if (pkt->cmd == Packet::UpgradeReq) {
 303               handleResponse(pkt);
 304           }
 305     } else if (pkt && !pkt->req->isUncacheable()) {
 306         missQueue->restoreOrigCmd(pkt);
 307     }
 308 }
 309
 310 template<class TagStore, class Buffering, class Coherence>
 311 void
 312 Cache<TagStore,Buffering,Coherence>::handleResponse(Packet * &pkt)
 313 {
 314     BlkType *blk = NULL;
 315     if (pkt->senderState) {
 316 //      MemDebug::cacheResponse(pkt);
 317         DPRINTF(Cache, "Handling reponse to %x, blk addr: %x\n",pkt->getAddr(),
 318                 pkt->getAddr() & (((ULL(1))<<48)-1));
 319
 320         if (pkt->isCacheFill() && !pkt->isNoAllocate()) {
 321             blk = tags->findBlock(pkt);
 322             CacheBlk::State old_state = (blk) ? blk->status : 0;
 323             PacketList writebacks;
 324             blk = tags->handleFill(blk, (MSHR*)pkt->senderState,
 325                                    coherence->getNewState(pkt,old_state),
 326                                    writebacks, pkt);
 327             while (!writebacks.empty()) {
 328                     missQueue->doWriteback(writebacks.front());
 329                     writebacks.pop_front();
 330             }
 331         }
 332         missQueue->handleResponse(pkt, curTick + hitLatency);
 333     }
 334 }
 335
 336 template<class TagStore, class Buffering, class Coherence>
 337 void
 338 Cache<TagStore,Buffering,Coherence>::pseudoFill(Addr addr)
 339 {
 340     // Need to temporarily move this blk into MSHRs
 341     MSHR *mshr = missQueue->allocateTargetList(addr);
 342     int lat;
 343     PacketList dummy;
 344     // Read the data into the mshr
 345     BlkType *blk = tags->handleAccess(mshr->pkt, lat, dummy, false);
 346     assert(dummy.empty());
 347     assert(mshr->pkt->flags & SATISFIED);
 348     // can overload order since it isn't used on non pending blocks
 349     mshr->order = blk->status;
 350     // temporarily remove the block from the cache.
 351     tags->invalidateBlk(addr);
 352 }
 353
 354 template<class TagStore, class Buffering, class Coherence>
 355 void
 356 Cache<TagStore,Buffering,Coherence>::pseudoFill(MSHR *mshr)
 357 {
 358     // Need to temporarily move this blk into MSHRs
 359     assert(mshr->pkt->cmd == Packet::ReadReq);
 360     int lat;
 361     PacketList dummy;
 362     // Read the data into the mshr
 363     BlkType *blk = tags->handleAccess(mshr->pkt, lat, dummy, false);
 364     assert(dummy.empty());
 365     assert(mshr->pkt->flags & SATISFIED);
 366     // can overload order since it isn't used on non pending blocks
 367     mshr->order = blk->status;
 368     // temporarily remove the block from the cache.
 369     tags->invalidateBlk(mshr->pkt->getAddr());
 370 }
 371
 372
 373 template<class TagStore, class Buffering, class Coherence>
 374 Packet *
 375 Cache<TagStore,Buffering,Coherence>::getCoherencePacket()
 376 {
 377     return coherence->getPacket();
 378 }
 379
 380
 381 template<class TagStore, class Buffering, class Coherence>
 382 void
 383 Cache<TagStore,Buffering,Coherence>::snoop(Packet * &pkt)
 384 {
 385     Addr blk_addr = pkt->getAddr() & ~(Addr(blkSize-1));
 386     BlkType *blk = tags->findBlock(pkt);
 387     MSHR *mshr = missQueue->findMSHR(blk_addr);
 388     if (isTopLevel() && coherence->hasProtocol()) { //@todo Move this into handle bus req
 389         //If we find an mshr, and it is in service, we need to NACK or invalidate
 390         if (mshr) {
 391             if (mshr->inService) {
 392                 if ((mshr->pkt->isInvalidate() || !mshr->pkt->isCacheFill())
 393                     && (pkt->cmd != Packet::InvalidateReq && pkt->cmd != Packet::WriteInvalidateReq)) {
 394                     //If the outstanding request was an invalidate (upgrade,readex,..)
 395                     //Then we need to ACK the request until we get the data
 396                     //Also NACK if the outstanding request is not a cachefill (writeback)
 397                     pkt->flags |= SATISFIED;
 398                     pkt->flags |= NACKED_LINE;
 399                     assert("Don't detect these on the other side yet\n");
 400                     respondToSnoop(pkt, curTick + hitLatency);
 401                     return;
 402                 }
 403                 else {
 404                     //The supplier will be someone else, because we are waiting for
 405                     //the data.  This should cause this cache to be forced to go to
 406                     //the shared state, not the exclusive even though the shared line
 407                     //won't be asserted.  But for now we will just invlidate ourselves
 408                     //and allow the other cache to go into the exclusive state.
 409                     //@todo Make it so a read to a pending read doesn't invalidate.
 410                     //@todo Make it so that a read to a pending read can't be exclusive now.
 411
 412                     //Set the address so find match works
 413                     assert("Don't have invalidates yet\n");
 414                     invalidatePkt->addrOverride(pkt->getAddr());
 415
 416                     //Append the invalidate on
 417                     missQueue->addTarget(mshr,invalidatePkt);
 418                     DPRINTF(Cache, "Appending Invalidate to blk_addr: %x\n", pkt->getAddr() & (((ULL(1))<<48)-1));
 419                     return;
 420                 }
 421             }
 422         }
 423         //We also need to check the writeback buffers and handle those
 424         std::vector<MSHR *> writebacks;
 425         if (missQueue->findWrites(blk_addr, writebacks)) {
 426             DPRINTF(Cache, "Snoop hit in writeback to blk_addr: %x\n", pkt->getAddr() & (((ULL(1))<<48)-1));
 427
 428             //Look through writebacks for any non-uncachable writes, use that
 429             for (int i=0; i<writebacks.size(); i++) {
 430                 mshr = writebacks[i];
 431
 432                 if (!mshr->pkt->req->isUncacheable()) {
 433                     if (pkt->isRead()) {
 434                         //Only Upgrades don't get here
 435                         //Supply the data
 436                         pkt->flags |= SATISFIED;
 437
 438                         //If we are in an exclusive protocol, make it ask again
 439                         //to get write permissions (upgrade), signal shared
 440                         pkt->flags |= SHARED_LINE;
 441
 442                         assert(pkt->isRead());
 443                         Addr offset = pkt->getAddr() & ~(blkSize - 1);
 444                         assert(offset < blkSize);
 445                         assert(pkt->getSize() <= blkSize);
 446                         assert(offset + pkt->getSize() <=blkSize);
 447                         memcpy(pkt->getPtr<uint8_t>(), mshr->pkt->getPtr<uint8_t>() + offset, pkt->getSize());
 448
 449                         respondToSnoop(pkt, curTick + hitLatency);
 450                     }
 451
 452                     if (pkt->isInvalidate()) {
 453                         //This must be an upgrade or other cache will take ownership
 454                         missQueue->markInService(mshr->pkt);
 455                     }
 456                     return;
 457                 }
 458             }
 459         }
 460     }
 461     CacheBlk::State new_state;
 462     bool satisfy = coherence->handleBusRequest(pkt,blk,mshr, new_state);
 463     if (satisfy) {
 464         tags->handleSnoop(blk, new_state, pkt);
 465         respondToSnoop(pkt, curTick + hitLatency);
 466         return;
 467     }
 468     tags->handleSnoop(blk, new_state);
 469 }
 470
 471 template<class TagStore, class Buffering, class Coherence>
 472 void
 473 Cache<TagStore,Buffering,Coherence>::snoopResponse(Packet * &pkt)
 474 {
 475     //Need to handle the response, if NACKED
 476     if (pkt->flags & NACKED_LINE) {
 477         //Need to mark it as not in service, and retry for bus
 478         assert(0); //Yeah, we saw a NACK come through
 479
 480         //For now this should never get called, we return false when we see a NACK
 481         //instead, by doing this we allow the bus_blocked mechanism to handle the retry
 482         //For now it retrys in just 2 cycles, need to figure out how to change that
 483         //Eventually we will want to also have success come in as a parameter
 484         //Need to make sure that we handle the functionality that happens on successufl
 485         //return of the sendAddr function
 486     }
 487 }
 488
 489 template<class TagStore, class Buffering, class Coherence>
 490 void
 491 Cache<TagStore,Buffering,Coherence>::invalidateBlk(Addr addr)
 492 {
 493     tags->invalidateBlk(addr);
 494 }
 495
 496
 497 /**
 498  * @todo Fix to not assume write allocate
 499  */
 500 template<class TagStore, class Buffering, class Coherence>
 501 Tick
 502 Cache<TagStore,Buffering,Coherence>::probe(Packet * &pkt, bool update, CachePort* otherSidePort)
 503 {
 504 //    MemDebug::cacheProbe(pkt);
 505     if (!pkt->req->isUncacheable()) {
 506         if (pkt->isInvalidate() && !pkt->isRead()
 507             && !pkt->isWrite()) {
 508             //Upgrade or Invalidate, satisfy it, don't forward
 509             DPRINTF(Cache, "%s %x ? blk_addr: %x\n", pkt->cmdString(),
 510                     pkt->getAddr() & (((ULL(1))<<48)-1),
 511                     pkt->getAddr() & ~((Addr)blkSize - 1));
 512             pkt->flags |= SATISFIED;
 513             return 0;
 514         }
 515     }
 516
 517     PacketList writebacks;
 518     int lat;
 519     BlkType *blk = tags->handleAccess(pkt, lat, writebacks, update);
 520
 521     if (!blk) {
 522         // Need to check for outstanding misses and writes
 523         Addr blk_addr = pkt->getAddr() & ~(blkSize - 1);
 524
 525         // There can only be one matching outstanding miss.
 526         MSHR* mshr = missQueue->findMSHR(blk_addr);
 527
 528         // There can be many matching outstanding writes.
 529         std::vector<MSHR*> writes;
 530         missQueue->findWrites(blk_addr, writes);
 531
 532         if (!update) {
 533                 otherSidePort->sendFunctional(pkt);
 534
 535             // Check for data in MSHR and writebuffer.
 536             if (mshr) {
 537                 warn("Found outstanding miss on an non-update probe");
 538                 MSHR::TargetList *targets = mshr->getTargetList();
 539                 MSHR::TargetList::iterator i = targets->begin();
 540                 MSHR::TargetList::iterator end = targets->end();
 541                 for (; i != end; ++i) {
 542                     Packet * target = *i;
 543                     // If the target contains data, and it overlaps the
 544                     // probed request, need to update data
 545                     if (target->isWrite() && target->intersect(pkt)) {
 546                         uint8_t* pkt_data;
 547                         uint8_t* write_data;
 548                         int data_size;
 549                         if (target->getAddr() < pkt->getAddr()) {
 550                             int offset = pkt->getAddr() - target->getAddr();
 551                             pkt_data = pkt->getPtr<uint8_t>();
 552                             write_data = target->getPtr<uint8_t>() + offset;
 553                             data_size = target->getSize() - offset;
 554                             assert(data_size > 0);
 555                             if (data_size > pkt->getSize())
 556                                 data_size = pkt->getSize();
 557                         } else {
 558                             int offset = target->getAddr() - pkt->getAddr();
 559                             pkt_data = pkt->getPtr<uint8_t>() + offset;
 560                             write_data = target->getPtr<uint8_t>();
 561                             data_size = pkt->getSize() - offset;
 562                             assert(data_size > pkt->getSize());
 563                             if (data_size > target->getSize())
 564                                 data_size = target->getSize();
 565                         }
 566
 567                         if (pkt->isWrite()) {
 568                             memcpy(pkt_data, write_data, data_size);
 569                         } else {
 570                             memcpy(write_data, pkt_data, data_size);
 571                         }
 572                     }
 573                 }
 574             }
 575             for (int i = 0; i < writes.size(); ++i) {
 576                 Packet * write = writes[i]->pkt;
 577                 if (write->intersect(pkt)) {
 578                     warn("Found outstanding write on an non-update probe");
 579                     uint8_t* pkt_data;
 580                     uint8_t* write_data;
 581                     int data_size;
 582                     if (write->getAddr() < pkt->getAddr()) {
 583                         int offset = pkt->getAddr() - write->getAddr();
 584                         pkt_data = pkt->getPtr<uint8_t>();
 585                         write_data = write->getPtr<uint8_t>() + offset;
 586                         data_size = write->getSize() - offset;
 587                         assert(data_size > 0);
 588                         if (data_size > pkt->getSize())
 589                             data_size = pkt->getSize();
 590                     } else {
 591                         int offset = write->getAddr() - pkt->getAddr();
 592                         pkt_data = pkt->getPtr<uint8_t>() + offset;
 593                         write_data = write->getPtr<uint8_t>();
 594                         data_size = pkt->getSize() - offset;
 595                         assert(data_size > pkt->getSize());
 596                         if (data_size > write->getSize())
 597                             data_size = write->getSize();
 598                     }
 599
 600                     if (pkt->isWrite()) {
 601                         memcpy(pkt_data, write_data, data_size);
 602                     } else {
 603                         memcpy(write_data, pkt_data, data_size);
 604                     }
 605
 606                 }
 607             }
 608             return 0;
 609         } else {
 610             // update the cache state and statistics
 611             if (mshr || !writes.empty()){
 612                 // Can't handle it, return pktuest unsatisfied.
 613                 return 0;
 614             }
 615             if (!pkt->req->isUncacheable()) {
 616                 // Fetch the cache block to fill
 617                 BlkType *blk = tags->findBlock(pkt);
 618                 Packet::Command temp_cmd = coherence->getBusCmd(pkt->cmd,
 619                                                    (blk)? blk->status : 0);
 620
 621                 Packet * busPkt = new Packet(pkt->req,temp_cmd, -1, blkSize);
 622
 623                 busPkt->allocate();
 624
 625                 busPkt->time = curTick;
 626
 627                 lat = memSidePort->sendAtomic(busPkt);
 628
 629                 //Be sure to flip the response to a request for coherence
 630                 busPkt->makeAtomicResponse();
 631
 632 /*              if (!(busPkt->flags & SATISFIED)) {
 633                     // blocked at a higher level, just return
 634                     return 0;
 635                 }
 636
 637 */              misses[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
 638
 639                 CacheBlk::State old_state = (blk) ? blk->status : 0;
 640                 tags->handleFill(blk, busPkt,
 641                                  coherence->getNewState(busPkt, old_state),
 642                                  writebacks, pkt);
 643                 // Handle writebacks if needed
 644                 while (!writebacks.empty()){
 645                     memSidePort->sendAtomic(writebacks.front());
 646                     writebacks.pop_front();
 647                 }
 648                 return lat + hitLatency;
 649             } else {
 650                 return memSidePort->sendAtomic(pkt);
 651             }
 652         }
 653     } else {
 654         // There was a cache hit.
 655         // Handle writebacks if needed
 656         while (!writebacks.empty()){
 657             memSidePort->sendAtomic(writebacks.front());
 658             writebacks.pop_front();
 659         }
 660
 661         if (update) {
 662             hits[pkt->cmdToIndex()][0/*pkt->req->getThreadNum()*/]++;
 663         } else if (pkt->isWrite()) {
 664             // Still need to change data in all locations.
 665             return otherSidePort->sendAtomic(pkt);
 666         }
 667         return curTick + lat;
 668     }
 669     fatal("Probe not handled.\n");
 670     return 0;
 671 }
 672
 673 template<class TagStore, class Buffering, class Coherence>
 674 Tick
 675 Cache<TagStore,Buffering,Coherence>::snoopProbe(PacketPtr &pkt)
 676 {
 677         Addr blk_addr = pkt->getAddr() & ~(Addr(blkSize-1));
 678         BlkType *blk = tags->findBlock(pkt);
 679         MSHR *mshr = missQueue->findMSHR(blk_addr);
 680         CacheBlk::State new_state = 0;
 681         bool satisfy = coherence->handleBusRequest(pkt,blk,mshr, new_state);
 682         if (satisfy) {
 683             tags->handleSnoop(blk, new_state, pkt);
 684             return hitLatency;
 685         }
 686         tags->handleSnoop(blk, new_state);
 687         return 0;
 688 }
 689