fetch_impl.hh revision 2906:3d65b80fdb11
1/* 2 * Copyright (c) 2004-2006 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: Kevin Lim 29 * Korey Sewell 30 */ 31 32#include "config/use_checker.hh" 33 34#include "arch/isa_traits.hh" 35#include "arch/utility.hh" 36#include "cpu/checker/cpu.hh" 37#include "cpu/exetrace.hh" 38#include "cpu/o3/fetch.hh" 39#include "mem/packet.hh" 40#include "mem/request.hh" 41#include "sim/byteswap.hh" 42#include "sim/host.hh" 43#include "sim/root.hh" 44 45#if FULL_SYSTEM 46#include "arch/tlb.hh" 47#include "arch/vtophys.hh" 48#include "base/remote_gdb.hh" 49#include "sim/system.hh" 50#endif // FULL_SYSTEM 51 52#include <algorithm> 53 54using namespace std; 55using namespace TheISA; 56 57template<class Impl> 58Tick 59DefaultFetch<Impl>::IcachePort::recvAtomic(PacketPtr pkt) 60{ 61 panic("DefaultFetch doesn't expect recvAtomic callback!"); 62 return curTick; 63} 64 65template<class Impl> 66void 67DefaultFetch<Impl>::IcachePort::recvFunctional(PacketPtr pkt) 68{ 69 panic("DefaultFetch doesn't expect recvFunctional callback!"); 70} 71 72template<class Impl> 73void 74DefaultFetch<Impl>::IcachePort::recvStatusChange(Status status) 75{ 76 if (status == RangeChange) 77 return; 78 79 panic("DefaultFetch doesn't expect recvStatusChange callback!"); 80} 81 82template<class Impl> 83bool 84DefaultFetch<Impl>::IcachePort::recvTiming(Packet *pkt) 85{ 86 fetch->processCacheCompletion(pkt); 87 return true; 88} 89 90template<class Impl> 91void 92DefaultFetch<Impl>::IcachePort::recvRetry() 93{ 94 fetch->recvRetry(); 95} 96 97template<class Impl> 98DefaultFetch<Impl>::DefaultFetch(Params *params) 99 : mem(params->mem), 100 branchPred(params), 101 decodeToFetchDelay(params->decodeToFetchDelay), 102 renameToFetchDelay(params->renameToFetchDelay), 103 iewToFetchDelay(params->iewToFetchDelay), 104 commitToFetchDelay(params->commitToFetchDelay), 105 fetchWidth(params->fetchWidth), 106 cacheBlocked(false), 107 retryPkt(NULL), 108 retryTid(-1), 109 numThreads(params->numberOfThreads), 110 numFetchingThreads(params->smtNumFetchingThreads), 111 interruptPending(false), 112 drainPending(false), 113 switchedOut(false) 114{ 115 if (numThreads > Impl::MaxThreads) 116 fatal("numThreads is not a valid value\n"); 117 118 // Set fetch stage's status to inactive. 119 _status = Inactive; 120 121 string policy = params->smtFetchPolicy; 122 123 // Convert string to lowercase 124 std::transform(policy.begin(), policy.end(), policy.begin(), 125 (int(*)(int)) tolower); 126 127 // Figure out fetch policy 128 if (policy == "singlethread") { 129 fetchPolicy = SingleThread; 130 if (numThreads > 1) 131 panic("Invalid Fetch Policy for a SMT workload."); 132 } else if (policy == "roundrobin") { 133 fetchPolicy = RoundRobin; 134 DPRINTF(Fetch, "Fetch policy set to Round Robin\n"); 135 } else if (policy == "branch") { 136 fetchPolicy = Branch; 137 DPRINTF(Fetch, "Fetch policy set to Branch Count\n"); 138 } else if (policy == "iqcount") { 139 fetchPolicy = IQ; 140 DPRINTF(Fetch, "Fetch policy set to IQ count\n"); 141 } else if (policy == "lsqcount") { 142 fetchPolicy = LSQ; 143 DPRINTF(Fetch, "Fetch policy set to LSQ count\n"); 144 } else { 145 fatal("Invalid Fetch Policy. Options Are: {SingleThread," 146 " RoundRobin,LSQcount,IQcount}\n"); 147 } 148 149 // Size of cache block. 150 cacheBlkSize = 64; 151 152 // Create mask to get rid of offset bits. 153 cacheBlkMask = (cacheBlkSize - 1); 154 155 for (int tid=0; tid < numThreads; tid++) { 156 157 fetchStatus[tid] = Running; 158 159 priorityList.push_back(tid); 160 161 memReq[tid] = NULL; 162 163 // Create space to store a cache line. 164 cacheData[tid] = new uint8_t[cacheBlkSize]; 165 cacheDataPC[tid] = 0; 166 cacheDataValid[tid] = false; 167 168 stalls[tid].decode = 0; 169 stalls[tid].rename = 0; 170 stalls[tid].iew = 0; 171 stalls[tid].commit = 0; 172 } 173 174 // Get the size of an instruction. 175 instSize = sizeof(MachInst); 176} 177 178template <class Impl> 179std::string 180DefaultFetch<Impl>::name() const 181{ 182 return cpu->name() + ".fetch"; 183} 184 185template <class Impl> 186void 187DefaultFetch<Impl>::regStats() 188{ 189 icacheStallCycles 190 .name(name() + ".icacheStallCycles") 191 .desc("Number of cycles fetch is stalled on an Icache miss") 192 .prereq(icacheStallCycles); 193 194 fetchedInsts 195 .name(name() + ".Insts") 196 .desc("Number of instructions fetch has processed") 197 .prereq(fetchedInsts); 198 199 fetchedBranches 200 .name(name() + ".Branches") 201 .desc("Number of branches that fetch encountered") 202 .prereq(fetchedBranches); 203 204 predictedBranches 205 .name(name() + ".predictedBranches") 206 .desc("Number of branches that fetch has predicted taken") 207 .prereq(predictedBranches); 208 209 fetchCycles 210 .name(name() + ".Cycles") 211 .desc("Number of cycles fetch has run and was not squashing or" 212 " blocked") 213 .prereq(fetchCycles); 214 215 fetchSquashCycles 216 .name(name() + ".SquashCycles") 217 .desc("Number of cycles fetch has spent squashing") 218 .prereq(fetchSquashCycles); 219 220 fetchIdleCycles 221 .name(name() + ".IdleCycles") 222 .desc("Number of cycles fetch was idle") 223 .prereq(fetchIdleCycles); 224 225 fetchBlockedCycles 226 .name(name() + ".BlockedCycles") 227 .desc("Number of cycles fetch has spent blocked") 228 .prereq(fetchBlockedCycles); 229 230 fetchedCacheLines 231 .name(name() + ".CacheLines") 232 .desc("Number of cache lines fetched") 233 .prereq(fetchedCacheLines); 234 235 fetchMiscStallCycles 236 .name(name() + ".MiscStallCycles") 237 .desc("Number of cycles fetch has spent waiting on interrupts, or " 238 "bad addresses, or out of MSHRs") 239 .prereq(fetchMiscStallCycles); 240 241 fetchIcacheSquashes 242 .name(name() + ".IcacheSquashes") 243 .desc("Number of outstanding Icache misses that were squashed") 244 .prereq(fetchIcacheSquashes); 245 246 fetchNisnDist 247 .init(/* base value */ 0, 248 /* last value */ fetchWidth, 249 /* bucket size */ 1) 250 .name(name() + ".rateDist") 251 .desc("Number of instructions fetched each cycle (Total)") 252 .flags(Stats::pdf); 253 254 idleRate 255 .name(name() + ".idleRate") 256 .desc("Percent of cycles fetch was idle") 257 .prereq(idleRate); 258 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 259 260 branchRate 261 .name(name() + ".branchRate") 262 .desc("Number of branch fetches per cycle") 263 .flags(Stats::total); 264 branchRate = fetchedBranches / cpu->numCycles; 265 266 fetchRate 267 .name(name() + ".rate") 268 .desc("Number of inst fetches per cycle") 269 .flags(Stats::total); 270 fetchRate = fetchedInsts / cpu->numCycles; 271 272 branchPred.regStats(); 273} 274 275template<class Impl> 276void 277DefaultFetch<Impl>::setCPU(O3CPU *cpu_ptr) 278{ 279 DPRINTF(Fetch, "Setting the CPU pointer.\n"); 280 cpu = cpu_ptr; 281 282 // Name is finally available, so create the port. 283 icachePort = new IcachePort(this); 284 285#if USE_CHECKER 286 if (cpu->checker) { 287 cpu->checker->setIcachePort(icachePort); 288 } 289#endif 290 291 // Fetch needs to start fetching instructions at the very beginning, 292 // so it must start up in active state. 293 switchToActive(); 294} 295 296template<class Impl> 297void 298DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 299{ 300 DPRINTF(Fetch, "Setting the time buffer pointer.\n"); 301 timeBuffer = time_buffer; 302 303 // Create wires to get information from proper places in time buffer. 304 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 305 fromRename = timeBuffer->getWire(-renameToFetchDelay); 306 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 307 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 308} 309 310template<class Impl> 311void 312DefaultFetch<Impl>::setActiveThreads(list<unsigned> *at_ptr) 313{ 314 DPRINTF(Fetch, "Setting active threads list pointer.\n"); 315 activeThreads = at_ptr; 316} 317 318template<class Impl> 319void 320DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 321{ 322 DPRINTF(Fetch, "Setting the fetch queue pointer.\n"); 323 fetchQueue = fq_ptr; 324 325 // Create wire to write information to proper place in fetch queue. 326 toDecode = fetchQueue->getWire(0); 327} 328 329template<class Impl> 330void 331DefaultFetch<Impl>::initStage() 332{ 333 // Setup PC and nextPC with initial state. 334 for (int tid = 0; tid < numThreads; tid++) { 335 PC[tid] = cpu->readPC(tid); 336 nextPC[tid] = cpu->readNextPC(tid); 337#if THE_ISA != ALPHA_ISA 338 nextNPC[tid] = cpu->readNextNPC(tid); 339#endif 340 } 341} 342 343template<class Impl> 344void 345DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 346{ 347 unsigned tid = pkt->req->getThreadNum(); 348 349 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid); 350 351 // Only change the status if it's still waiting on the icache access 352 // to return. 353 if (fetchStatus[tid] != IcacheWaitResponse || 354 pkt->req != memReq[tid] || 355 isSwitchedOut()) { 356 ++fetchIcacheSquashes; 357 delete pkt->req; 358 delete pkt; 359 return; 360 } 361 362 memcpy(cacheData[tid], pkt->getPtr<uint8_t *>(), cacheBlkSize); 363 cacheDataValid[tid] = true; 364 365 if (!drainPending) { 366 // Wake up the CPU (if it went to sleep and was waiting on 367 // this completion event). 368 cpu->wakeCPU(); 369 370 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 371 tid); 372 373 switchToActive(); 374 } 375 376 // Only switch to IcacheAccessComplete if we're not stalled as well. 377 if (checkStall(tid)) { 378 fetchStatus[tid] = Blocked; 379 } else { 380 fetchStatus[tid] = IcacheAccessComplete; 381 } 382 383 // Reset the mem req to NULL. 384 delete pkt->req; 385 delete pkt; 386 memReq[tid] = NULL; 387} 388 389template <class Impl> 390bool 391DefaultFetch<Impl>::drain() 392{ 393 // Fetch is ready to drain at any time. 394 cpu->signalDrained(); 395 drainPending = true; 396 return true; 397} 398 399template <class Impl> 400void 401DefaultFetch<Impl>::resume() 402{ 403 drainPending = false; 404} 405 406template <class Impl> 407void 408DefaultFetch<Impl>::switchOut() 409{ 410 switchedOut = true; 411 // Branch predictor needs to have its state cleared. 412 branchPred.switchOut(); 413} 414 415template <class Impl> 416void 417DefaultFetch<Impl>::takeOverFrom() 418{ 419 // Reset all state 420 for (int i = 0; i < Impl::MaxThreads; ++i) { 421 stalls[i].decode = 0; 422 stalls[i].rename = 0; 423 stalls[i].iew = 0; 424 stalls[i].commit = 0; 425 PC[i] = cpu->readPC(i); 426 nextPC[i] = cpu->readNextPC(i); 427#if THE_ISA != ALPHA_ISA 428 nextNPC[i] = cpu->readNextNPC(i); 429#endif 430 fetchStatus[i] = Running; 431 } 432 numInst = 0; 433 wroteToTimeBuffer = false; 434 _status = Inactive; 435 switchedOut = false; 436 branchPred.takeOverFrom(); 437} 438 439template <class Impl> 440void 441DefaultFetch<Impl>::wakeFromQuiesce() 442{ 443 DPRINTF(Fetch, "Waking up from quiesce\n"); 444 // Hopefully this is safe 445 // @todo: Allow other threads to wake from quiesce. 446 fetchStatus[0] = Running; 447} 448 449template <class Impl> 450inline void 451DefaultFetch<Impl>::switchToActive() 452{ 453 if (_status == Inactive) { 454 DPRINTF(Activity, "Activating stage.\n"); 455 456 cpu->activateStage(O3CPU::FetchIdx); 457 458 _status = Active; 459 } 460} 461 462template <class Impl> 463inline void 464DefaultFetch<Impl>::switchToInactive() 465{ 466 if (_status == Active) { 467 DPRINTF(Activity, "Deactivating stage.\n"); 468 469 cpu->deactivateStage(O3CPU::FetchIdx); 470 471 _status = Inactive; 472 } 473} 474 475template <class Impl> 476bool 477DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC) 478{ 479 // Do branch prediction check here. 480 // A bit of a misnomer...next_PC is actually the current PC until 481 // this function updates it. 482 bool predict_taken; 483 484 if (!inst->isControl()) { 485 next_PC = next_PC + instSize; 486 inst->setPredTarg(next_PC); 487 return false; 488 } 489 490 predict_taken = branchPred.predict(inst, next_PC, inst->threadNumber); 491 492 ++fetchedBranches; 493 494 if (predict_taken) { 495 ++predictedBranches; 496 } 497 498 return predict_taken; 499} 500 501template <class Impl> 502bool 503DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid) 504{ 505 Fault fault = NoFault; 506 507#if FULL_SYSTEM 508 // Flag to say whether or not address is physical addr. 509 unsigned flags = cpu->inPalMode(fetch_PC) ? PHYSICAL : 0; 510#else 511 unsigned flags = 0; 512#endif // FULL_SYSTEM 513 514 if (cacheBlocked || (interruptPending && flags == 0)) { 515 // Hold off fetch from getting new instructions when: 516 // Cache is blocked, or 517 // while an interrupt is pending and we're not in PAL mode, or 518 // fetch is switched out. 519 return false; 520 } 521 522 // Align the fetch PC so it's at the start of a cache block. 523 fetch_PC = icacheBlockAlignPC(fetch_PC); 524 525 // If we've already got the block, no need to try to fetch it again. 526 if (cacheDataValid[tid] && fetch_PC == cacheDataPC[tid]) { 527 return true; 528 } 529 530 // Setup the memReq to do a read of the first instruction's address. 531 // Set the appropriate read size and flags as well. 532 // Build request here. 533 RequestPtr mem_req = new Request(tid, fetch_PC, cacheBlkSize, flags, 534 fetch_PC, cpu->readCpuId(), tid); 535 536 memReq[tid] = mem_req; 537 538 // Translate the instruction request. 539 fault = cpu->translateInstReq(mem_req, cpu->thread[tid]); 540 541 // In the case of faults, the fetch stage may need to stall and wait 542 // for the ITB miss to be handled. 543 544 // If translation was successful, attempt to read the first 545 // instruction. 546 if (fault == NoFault) { 547#if 0 548 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || 549 memReq[tid]->flags & UNCACHEABLE) { 550 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " 551 "misspeculating path)!", 552 memReq[tid]->paddr); 553 ret_fault = TheISA::genMachineCheckFault(); 554 return false; 555 } 556#endif 557 558 // Build packet here. 559 PacketPtr data_pkt = new Packet(mem_req, 560 Packet::ReadReq, Packet::Broadcast); 561 data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]); 562 563 cacheDataPC[tid] = fetch_PC; 564 cacheDataValid[tid] = false; 565 566 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 567 568 fetchedCacheLines++; 569 570 // Now do the timing access to see whether or not the instruction 571 // exists within the cache. 572 if (!icachePort->sendTiming(data_pkt)) { 573 assert(retryPkt == NULL); 574 assert(retryTid == -1); 575 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 576 fetchStatus[tid] = IcacheWaitRetry; 577 retryPkt = data_pkt; 578 retryTid = tid; 579 cacheBlocked = true; 580 return false; 581 } 582 583 DPRINTF(Fetch, "[tid:%i]: Doing cache access.\n", tid); 584 585 lastIcacheStall[tid] = curTick; 586 587 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 588 "response.\n", tid); 589 590 fetchStatus[tid] = IcacheWaitResponse; 591 } else { 592 delete mem_req; 593 memReq[tid] = NULL; 594 } 595 596 ret_fault = fault; 597 return true; 598} 599 600template <class Impl> 601inline void 602DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid) 603{ 604 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x.\n", 605 tid, new_PC); 606 607 PC[tid] = new_PC; 608 nextPC[tid] = new_PC + instSize; 609 610 // Clear the icache miss if it's outstanding. 611 if (fetchStatus[tid] == IcacheWaitResponse) { 612 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 613 tid); 614 memReq[tid] = NULL; 615 } 616 617 // Get rid of the retrying packet if it was from this thread. 618 if (retryTid == tid) { 619 assert(cacheBlocked); 620 cacheBlocked = false; 621 retryTid = -1; 622 retryPkt = NULL; 623 delete retryPkt->req; 624 delete retryPkt; 625 } 626 627 fetchStatus[tid] = Squashing; 628 629 ++fetchSquashCycles; 630} 631 632template<class Impl> 633void 634DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, 635 const InstSeqNum &seq_num, 636 unsigned tid) 637{ 638 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid); 639 640 doSquash(new_PC, tid); 641 642 // Tell the CPU to remove any instructions that are in flight between 643 // fetch and decode. 644 cpu->removeInstsUntil(seq_num, tid); 645} 646 647template<class Impl> 648bool 649DefaultFetch<Impl>::checkStall(unsigned tid) const 650{ 651 bool ret_val = false; 652 653 if (cpu->contextSwitch) { 654 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 655 ret_val = true; 656 } else if (stalls[tid].decode) { 657 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 658 ret_val = true; 659 } else if (stalls[tid].rename) { 660 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 661 ret_val = true; 662 } else if (stalls[tid].iew) { 663 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 664 ret_val = true; 665 } else if (stalls[tid].commit) { 666 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 667 ret_val = true; 668 } 669 670 return ret_val; 671} 672 673template<class Impl> 674typename DefaultFetch<Impl>::FetchStatus 675DefaultFetch<Impl>::updateFetchStatus() 676{ 677 //Check Running 678 list<unsigned>::iterator threads = (*activeThreads).begin(); 679 680 while (threads != (*activeThreads).end()) { 681 682 unsigned tid = *threads++; 683 684 if (fetchStatus[tid] == Running || 685 fetchStatus[tid] == Squashing || 686 fetchStatus[tid] == IcacheAccessComplete) { 687 688 if (_status == Inactive) { 689 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 690 691 if (fetchStatus[tid] == IcacheAccessComplete) { 692 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 693 "completion\n",tid); 694 } 695 696 cpu->activateStage(O3CPU::FetchIdx); 697 } 698 699 return Active; 700 } 701 } 702 703 // Stage is switching from active to inactive, notify CPU of it. 704 if (_status == Active) { 705 DPRINTF(Activity, "Deactivating stage.\n"); 706 707 cpu->deactivateStage(O3CPU::FetchIdx); 708 } 709 710 return Inactive; 711} 712 713template <class Impl> 714void 715DefaultFetch<Impl>::squash(const Addr &new_PC, unsigned tid) 716{ 717 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid); 718 719 doSquash(new_PC, tid); 720 721 // Tell the CPU to remove any instructions that are not in the ROB. 722 cpu->removeInstsNotInROB(tid); 723} 724 725template <class Impl> 726void 727DefaultFetch<Impl>::tick() 728{ 729 list<unsigned>::iterator threads = (*activeThreads).begin(); 730 bool status_change = false; 731 732 wroteToTimeBuffer = false; 733 734 while (threads != (*activeThreads).end()) { 735 unsigned tid = *threads++; 736 737 // Check the signals for each thread to determine the proper status 738 // for each thread. 739 bool updated_status = checkSignalsAndUpdate(tid); 740 status_change = status_change || updated_status; 741 } 742 743 DPRINTF(Fetch, "Running stage.\n"); 744 745 // Reset the number of the instruction we're fetching. 746 numInst = 0; 747 748#if FULL_SYSTEM 749 if (fromCommit->commitInfo[0].interruptPending) { 750 interruptPending = true; 751 } 752 753 if (fromCommit->commitInfo[0].clearInterrupt) { 754 interruptPending = false; 755 } 756#endif 757 758 for (threadFetched = 0; threadFetched < numFetchingThreads; 759 threadFetched++) { 760 // Fetch each of the actively fetching threads. 761 fetch(status_change); 762 } 763 764 // Record number of instructions fetched this cycle for distribution. 765 fetchNisnDist.sample(numInst); 766 767 if (status_change) { 768 // Change the fetch stage status if there was a status change. 769 _status = updateFetchStatus(); 770 } 771 772 // If there was activity this cycle, inform the CPU of it. 773 if (wroteToTimeBuffer || cpu->contextSwitch) { 774 DPRINTF(Activity, "Activity this cycle.\n"); 775 776 cpu->activityThisCycle(); 777 } 778} 779 780template <class Impl> 781bool 782DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) 783{ 784 // Update the per thread stall statuses. 785 if (fromDecode->decodeBlock[tid]) { 786 stalls[tid].decode = true; 787 } 788 789 if (fromDecode->decodeUnblock[tid]) { 790 assert(stalls[tid].decode); 791 assert(!fromDecode->decodeBlock[tid]); 792 stalls[tid].decode = false; 793 } 794 795 if (fromRename->renameBlock[tid]) { 796 stalls[tid].rename = true; 797 } 798 799 if (fromRename->renameUnblock[tid]) { 800 assert(stalls[tid].rename); 801 assert(!fromRename->renameBlock[tid]); 802 stalls[tid].rename = false; 803 } 804 805 if (fromIEW->iewBlock[tid]) { 806 stalls[tid].iew = true; 807 } 808 809 if (fromIEW->iewUnblock[tid]) { 810 assert(stalls[tid].iew); 811 assert(!fromIEW->iewBlock[tid]); 812 stalls[tid].iew = false; 813 } 814 815 if (fromCommit->commitBlock[tid]) { 816 stalls[tid].commit = true; 817 } 818 819 if (fromCommit->commitUnblock[tid]) { 820 assert(stalls[tid].commit); 821 assert(!fromCommit->commitBlock[tid]); 822 stalls[tid].commit = false; 823 } 824 825 // Check squash signals from commit. 826 if (fromCommit->commitInfo[tid].squash) { 827 828 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 829 "from commit.\n",tid); 830 831 // In any case, squash. 832 squash(fromCommit->commitInfo[tid].nextPC,tid); 833 834 // Also check if there's a mispredict that happened. 835 if (fromCommit->commitInfo[tid].branchMispredict) { 836 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 837 fromCommit->commitInfo[tid].nextPC, 838 fromCommit->commitInfo[tid].branchTaken, 839 tid); 840 } else { 841 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 842 tid); 843 } 844 845 return true; 846 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 847 // Update the branch predictor if it wasn't a squashed instruction 848 // that was broadcasted. 849 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 850 } 851 852 // Check ROB squash signals from commit. 853 if (fromCommit->commitInfo[tid].robSquashing) { 854 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 855 856 // Continue to squash. 857 fetchStatus[tid] = Squashing; 858 859 return true; 860 } 861 862 // Check squash signals from decode. 863 if (fromDecode->decodeInfo[tid].squash) { 864 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 865 "from decode.\n",tid); 866 867 // Update the branch predictor. 868 if (fromDecode->decodeInfo[tid].branchMispredict) { 869 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 870 fromDecode->decodeInfo[tid].nextPC, 871 fromDecode->decodeInfo[tid].branchTaken, 872 tid); 873 } else { 874 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 875 tid); 876 } 877 878 if (fetchStatus[tid] != Squashing) { 879 // Squash unless we're already squashing 880 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 881 fromDecode->decodeInfo[tid].doneSeqNum, 882 tid); 883 884 return true; 885 } 886 } 887 888 if (checkStall(tid) && fetchStatus[tid] != IcacheWaitResponse) { 889 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 890 891 fetchStatus[tid] = Blocked; 892 893 return true; 894 } 895 896 if (fetchStatus[tid] == Blocked || 897 fetchStatus[tid] == Squashing) { 898 // Switch status to running if fetch isn't being told to block or 899 // squash this cycle. 900 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 901 tid); 902 903 fetchStatus[tid] = Running; 904 905 return true; 906 } 907 908 // If we've reached this point, we have not gotten any signals that 909 // cause fetch to change its status. Fetch remains the same as before. 910 return false; 911} 912 913template<class Impl> 914void 915DefaultFetch<Impl>::fetch(bool &status_change) 916{ 917 ////////////////////////////////////////// 918 // Start actual fetch 919 ////////////////////////////////////////// 920 int tid = getFetchingThread(fetchPolicy); 921 922 if (tid == -1 || drainPending) { 923 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 924 925 // Breaks looping condition in tick() 926 threadFetched = numFetchingThreads; 927 return; 928 } 929 930 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 931 932 // The current PC. 933 Addr &fetch_PC = PC[tid]; 934 935 // Fault code for memory access. 936 Fault fault = NoFault; 937 938 // If returning from the delay of a cache miss, then update the status 939 // to running, otherwise do the cache access. Possibly move this up 940 // to tick() function. 941 if (fetchStatus[tid] == IcacheAccessComplete) { 942 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", 943 tid); 944 945 fetchStatus[tid] = Running; 946 status_change = true; 947 } else if (fetchStatus[tid] == Running) { 948 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 949 "instruction, starting at PC %08p.\n", 950 tid, fetch_PC); 951 952 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid); 953 if (!fetch_success) { 954 if (cacheBlocked) { 955 ++icacheStallCycles; 956 } else { 957 ++fetchMiscStallCycles; 958 } 959 return; 960 } 961 } else { 962 if (fetchStatus[tid] == Idle) { 963 ++fetchIdleCycles; 964 } else if (fetchStatus[tid] == Blocked) { 965 ++fetchBlockedCycles; 966 } else if (fetchStatus[tid] == Squashing) { 967 ++fetchSquashCycles; 968 } else if (fetchStatus[tid] == IcacheWaitResponse) { 969 ++icacheStallCycles; 970 } 971 972 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so 973 // fetch should do nothing. 974 return; 975 } 976 977 ++fetchCycles; 978 979 // If we had a stall due to an icache miss, then return. 980 if (fetchStatus[tid] == IcacheWaitResponse) { 981 ++icacheStallCycles; 982 status_change = true; 983 return; 984 } 985 986 Addr next_PC = fetch_PC; 987 InstSeqNum inst_seq; 988 MachInst inst; 989 ExtMachInst ext_inst; 990 // @todo: Fix this hack. 991 unsigned offset = (fetch_PC & cacheBlkMask) & ~3; 992 993 if (fault == NoFault) { 994 // If the read of the first instruction was successful, then grab the 995 // instructions from the rest of the cache line and put them into the 996 // queue heading to decode. 997 998 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 999 "decode.\n",tid); 1000 1001 // Need to keep track of whether or not a predicted branch 1002 // ended this fetch block. 1003 bool predicted_branch = false; 1004 1005 for (; 1006 offset < cacheBlkSize && 1007 numInst < fetchWidth && 1008 !predicted_branch; 1009 ++numInst) { 1010 1011 // Get a sequence number. 1012 inst_seq = cpu->getAndIncrementInstSeq(); 1013 1014 // Make sure this is a valid index. 1015 assert(offset <= cacheBlkSize - instSize); 1016 1017 // Get the instruction from the array of the cache line. 1018 inst = gtoh(*reinterpret_cast<MachInst *> 1019 (&cacheData[tid][offset])); 1020 1021 ext_inst = TheISA::makeExtMI(inst, fetch_PC); 1022 1023 // Create a new DynInst from the instruction fetched. 1024 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1025 next_PC, 1026 inst_seq, cpu); 1027 instruction->setTid(tid); 1028 1029 instruction->setASID(tid); 1030 1031 instruction->setThreadState(cpu->thread[tid]); 1032 1033 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created " 1034 "[sn:%lli]\n", 1035 tid, instruction->readPC(), inst_seq); 1036 1037 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", 1038 tid, instruction->staticInst->disassemble(fetch_PC)); 1039 1040 instruction->traceData = 1041 Trace::getInstRecord(curTick, cpu->tcBase(tid), cpu, 1042 instruction->staticInst, 1043 instruction->readPC(),tid); 1044 1045 predicted_branch = lookupAndUpdateNextPC(instruction, next_PC); 1046 1047 // Add instruction to the CPU's list of instructions. 1048 instruction->setInstListIt(cpu->addInst(instruction)); 1049 1050 // Write the instruction to the first slot in the queue 1051 // that heads to decode. 1052 toDecode->insts[numInst] = instruction; 1053 1054 toDecode->size++; 1055 1056 // Increment stat of fetched instructions. 1057 ++fetchedInsts; 1058 1059 // Move to the next instruction, unless we have a branch. 1060 fetch_PC = next_PC; 1061 1062 if (instruction->isQuiesce()) { 1063 warn("cycle %lli: Quiesce instruction encountered, halting fetch!", 1064 curTick); 1065 fetchStatus[tid] = QuiescePending; 1066 ++numInst; 1067 status_change = true; 1068 break; 1069 } 1070 1071 offset+= instSize; 1072 } 1073 } 1074 1075 if (numInst > 0) { 1076 wroteToTimeBuffer = true; 1077 } 1078 1079 // Now that fetching is completed, update the PC to signify what the next 1080 // cycle will be. 1081 if (fault == NoFault) { 1082 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n",tid, next_PC); 1083 1084#if THE_ISA == ALPHA_ISA 1085 PC[tid] = next_PC; 1086 nextPC[tid] = next_PC + instSize; 1087#else 1088 PC[tid] = next_PC; 1089 nextPC[tid] = next_PC + instSize; 1090 nextPC[tid] = next_PC + instSize; 1091 1092 thread->setNextPC(thread->readNextNPC()); 1093 thread->setNextNPC(thread->readNextNPC() + sizeof(MachInst)); 1094#endif 1095 } else { 1096 // We shouldn't be in an icache miss and also have a fault (an ITB 1097 // miss) 1098 if (fetchStatus[tid] == IcacheWaitResponse) { 1099 panic("Fetch should have exited prior to this!"); 1100 } 1101 1102 // Send the fault to commit. This thread will not do anything 1103 // until commit handles the fault. The only other way it can 1104 // wake up is if a squash comes along and changes the PC. 1105#if FULL_SYSTEM 1106 assert(numInst != fetchWidth); 1107 // Get a sequence number. 1108 inst_seq = cpu->getAndIncrementInstSeq(); 1109 // We will use a nop in order to carry the fault. 1110 ext_inst = TheISA::NoopMachInst; 1111 1112 // Create a new DynInst from the dummy nop. 1113 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1114 next_PC, 1115 inst_seq, cpu); 1116 instruction->setPredTarg(next_PC + instSize); 1117 instruction->setTid(tid); 1118 1119 instruction->setASID(tid); 1120 1121 instruction->setThreadState(cpu->thread[tid]); 1122 1123 instruction->traceData = NULL; 1124 1125 instruction->setInstListIt(cpu->addInst(instruction)); 1126 1127 instruction->fault = fault; 1128 1129 toDecode->insts[numInst] = instruction; 1130 toDecode->size++; 1131 1132 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid); 1133 1134 fetchStatus[tid] = TrapPending; 1135 status_change = true; 1136 1137 warn("cycle %lli: fault (%d) detected @ PC %08p", curTick, fault, PC[tid]); 1138#else // !FULL_SYSTEM 1139 warn("cycle %lli: fault (%d) detected @ PC %08p", curTick, fault, PC[tid]); 1140#endif // FULL_SYSTEM 1141 } 1142} 1143 1144template<class Impl> 1145void 1146DefaultFetch<Impl>::recvRetry() 1147{ 1148 assert(cacheBlocked); 1149 if (retryPkt != NULL) { 1150 assert(retryTid != -1); 1151 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1152 1153 if (icachePort->sendTiming(retryPkt)) { 1154 fetchStatus[retryTid] = IcacheWaitResponse; 1155 retryPkt = NULL; 1156 retryTid = -1; 1157 cacheBlocked = false; 1158 } 1159 } else { 1160 assert(retryTid == -1); 1161 // Access has been squashed since it was sent out. Just clear 1162 // the cache being blocked. 1163 cacheBlocked = false; 1164 } 1165} 1166 1167/////////////////////////////////////// 1168// // 1169// SMT FETCH POLICY MAINTAINED HERE // 1170// // 1171/////////////////////////////////////// 1172template<class Impl> 1173int 1174DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1175{ 1176 if (numThreads > 1) { 1177 switch (fetch_priority) { 1178 1179 case SingleThread: 1180 return 0; 1181 1182 case RoundRobin: 1183 return roundRobin(); 1184 1185 case IQ: 1186 return iqCount(); 1187 1188 case LSQ: 1189 return lsqCount(); 1190 1191 case Branch: 1192 return branchCount(); 1193 1194 default: 1195 return -1; 1196 } 1197 } else { 1198 int tid = *((*activeThreads).begin()); 1199 1200 if (fetchStatus[tid] == Running || 1201 fetchStatus[tid] == IcacheAccessComplete || 1202 fetchStatus[tid] == Idle) { 1203 return tid; 1204 } else { 1205 return -1; 1206 } 1207 } 1208 1209} 1210 1211 1212template<class Impl> 1213int 1214DefaultFetch<Impl>::roundRobin() 1215{ 1216 list<unsigned>::iterator pri_iter = priorityList.begin(); 1217 list<unsigned>::iterator end = priorityList.end(); 1218 1219 int high_pri; 1220 1221 while (pri_iter != end) { 1222 high_pri = *pri_iter; 1223 1224 assert(high_pri <= numThreads); 1225 1226 if (fetchStatus[high_pri] == Running || 1227 fetchStatus[high_pri] == IcacheAccessComplete || 1228 fetchStatus[high_pri] == Idle) { 1229 1230 priorityList.erase(pri_iter); 1231 priorityList.push_back(high_pri); 1232 1233 return high_pri; 1234 } 1235 1236 pri_iter++; 1237 } 1238 1239 return -1; 1240} 1241 1242template<class Impl> 1243int 1244DefaultFetch<Impl>::iqCount() 1245{ 1246 priority_queue<unsigned> PQ; 1247 1248 list<unsigned>::iterator threads = (*activeThreads).begin(); 1249 1250 while (threads != (*activeThreads).end()) { 1251 unsigned tid = *threads++; 1252 1253 PQ.push(fromIEW->iewInfo[tid].iqCount); 1254 } 1255 1256 while (!PQ.empty()) { 1257 1258 unsigned high_pri = PQ.top(); 1259 1260 if (fetchStatus[high_pri] == Running || 1261 fetchStatus[high_pri] == IcacheAccessComplete || 1262 fetchStatus[high_pri] == Idle) 1263 return high_pri; 1264 else 1265 PQ.pop(); 1266 1267 } 1268 1269 return -1; 1270} 1271 1272template<class Impl> 1273int 1274DefaultFetch<Impl>::lsqCount() 1275{ 1276 priority_queue<unsigned> PQ; 1277 1278 1279 list<unsigned>::iterator threads = (*activeThreads).begin(); 1280 1281 while (threads != (*activeThreads).end()) { 1282 unsigned tid = *threads++; 1283 1284 PQ.push(fromIEW->iewInfo[tid].ldstqCount); 1285 } 1286 1287 while (!PQ.empty()) { 1288 1289 unsigned high_pri = PQ.top(); 1290 1291 if (fetchStatus[high_pri] == Running || 1292 fetchStatus[high_pri] == IcacheAccessComplete || 1293 fetchStatus[high_pri] == Idle) 1294 return high_pri; 1295 else 1296 PQ.pop(); 1297 1298 } 1299 1300 return -1; 1301} 1302 1303template<class Impl> 1304int 1305DefaultFetch<Impl>::branchCount() 1306{ 1307 list<unsigned>::iterator threads = (*activeThreads).begin(); 1308 panic("Branch Count Fetch policy unimplemented\n"); 1309 return *threads; 1310} 1311