fetch_impl.hh revision 2863:2592e056dc5c
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]; 165 166 stalls[tid].decode = 0; 167 stalls[tid].rename = 0; 168 stalls[tid].iew = 0; 169 stalls[tid].commit = 0; 170 } 171 172 // Get the size of an instruction. 173 instSize = sizeof(MachInst); 174} 175 176template <class Impl> 177std::string 178DefaultFetch<Impl>::name() const 179{ 180 return cpu->name() + ".fetch"; 181} 182 183template <class Impl> 184void 185DefaultFetch<Impl>::regStats() 186{ 187 icacheStallCycles 188 .name(name() + ".icacheStallCycles") 189 .desc("Number of cycles fetch is stalled on an Icache miss") 190 .prereq(icacheStallCycles); 191 192 fetchedInsts 193 .name(name() + ".Insts") 194 .desc("Number of instructions fetch has processed") 195 .prereq(fetchedInsts); 196 197 fetchedBranches 198 .name(name() + ".Branches") 199 .desc("Number of branches that fetch encountered") 200 .prereq(fetchedBranches); 201 202 predictedBranches 203 .name(name() + ".predictedBranches") 204 .desc("Number of branches that fetch has predicted taken") 205 .prereq(predictedBranches); 206 207 fetchCycles 208 .name(name() + ".Cycles") 209 .desc("Number of cycles fetch has run and was not squashing or" 210 " blocked") 211 .prereq(fetchCycles); 212 213 fetchSquashCycles 214 .name(name() + ".SquashCycles") 215 .desc("Number of cycles fetch has spent squashing") 216 .prereq(fetchSquashCycles); 217 218 fetchIdleCycles 219 .name(name() + ".IdleCycles") 220 .desc("Number of cycles fetch was idle") 221 .prereq(fetchIdleCycles); 222 223 fetchBlockedCycles 224 .name(name() + ".BlockedCycles") 225 .desc("Number of cycles fetch has spent blocked") 226 .prereq(fetchBlockedCycles); 227 228 fetchedCacheLines 229 .name(name() + ".CacheLines") 230 .desc("Number of cache lines fetched") 231 .prereq(fetchedCacheLines); 232 233 fetchMiscStallCycles 234 .name(name() + ".MiscStallCycles") 235 .desc("Number of cycles fetch has spent waiting on interrupts, or " 236 "bad addresses, or out of MSHRs") 237 .prereq(fetchMiscStallCycles); 238 239 fetchIcacheSquashes 240 .name(name() + ".IcacheSquashes") 241 .desc("Number of outstanding Icache misses that were squashed") 242 .prereq(fetchIcacheSquashes); 243 244 fetchNisnDist 245 .init(/* base value */ 0, 246 /* last value */ fetchWidth, 247 /* bucket size */ 1) 248 .name(name() + ".rateDist") 249 .desc("Number of instructions fetched each cycle (Total)") 250 .flags(Stats::pdf); 251 252 idleRate 253 .name(name() + ".idleRate") 254 .desc("Percent of cycles fetch was idle") 255 .prereq(idleRate); 256 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 257 258 branchRate 259 .name(name() + ".branchRate") 260 .desc("Number of branch fetches per cycle") 261 .flags(Stats::total); 262 branchRate = fetchedBranches / cpu->numCycles; 263 264 fetchRate 265 .name(name() + ".rate") 266 .desc("Number of inst fetches per cycle") 267 .flags(Stats::total); 268 fetchRate = fetchedInsts / cpu->numCycles; 269 270 branchPred.regStats(); 271} 272 273template<class Impl> 274void 275DefaultFetch<Impl>::setCPU(O3CPU *cpu_ptr) 276{ 277 DPRINTF(Fetch, "Setting the CPU pointer.\n"); 278 cpu = cpu_ptr; 279 280 // Name is finally available, so create the port. 281 icachePort = new IcachePort(this); 282 283 Port *mem_dport = mem->getPort(""); 284 icachePort->setPeer(mem_dport); 285 mem_dport->setPeer(icachePort); 286 287#if USE_CHECKER 288 if (cpu->checker) { 289 cpu->checker->setIcachePort(icachePort); 290 } 291#endif 292 293 // Fetch needs to start fetching instructions at the very beginning, 294 // so it must start up in active state. 295 switchToActive(); 296} 297 298template<class Impl> 299void 300DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 301{ 302 DPRINTF(Fetch, "Setting the time buffer pointer.\n"); 303 timeBuffer = time_buffer; 304 305 // Create wires to get information from proper places in time buffer. 306 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 307 fromRename = timeBuffer->getWire(-renameToFetchDelay); 308 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 309 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 310} 311 312template<class Impl> 313void 314DefaultFetch<Impl>::setActiveThreads(list<unsigned> *at_ptr) 315{ 316 DPRINTF(Fetch, "Setting active threads list pointer.\n"); 317 activeThreads = at_ptr; 318} 319 320template<class Impl> 321void 322DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 323{ 324 DPRINTF(Fetch, "Setting the fetch queue pointer.\n"); 325 fetchQueue = fq_ptr; 326 327 // Create wire to write information to proper place in fetch queue. 328 toDecode = fetchQueue->getWire(0); 329} 330 331template<class Impl> 332void 333DefaultFetch<Impl>::initStage() 334{ 335 // Setup PC and nextPC with initial state. 336 for (int tid = 0; tid < numThreads; tid++) { 337 PC[tid] = cpu->readPC(tid); 338 nextPC[tid] = cpu->readNextPC(tid); 339#if THE_ISA != ALPHA_ISA 340 nextNPC[tid] = cpu->readNextNPC(tid); 341#endif 342 } 343} 344 345template<class Impl> 346void 347DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 348{ 349 unsigned tid = pkt->req->getThreadNum(); 350 351 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid); 352 353 // Only change the status if it's still waiting on the icache access 354 // to return. 355 if (fetchStatus[tid] != IcacheWaitResponse || 356 pkt->req != memReq[tid] || 357 isSwitchedOut() || 358 drainPending) { 359 ++fetchIcacheSquashes; 360 delete pkt->req; 361 delete pkt; 362 return; 363 } 364 365 // Wake up the CPU (if it went to sleep and was waiting on this completion 366 // event). 367 cpu->wakeCPU(); 368 369 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 370 tid); 371 372 switchToActive(); 373 374 // Only switch to IcacheAccessComplete if we're not stalled as well. 375 if (checkStall(tid)) { 376 fetchStatus[tid] = Blocked; 377 } else { 378 fetchStatus[tid] = IcacheAccessComplete; 379 } 380 381 // Reset the mem req to NULL. 382 delete pkt->req; 383 delete pkt; 384 memReq[tid] = NULL; 385} 386 387template <class Impl> 388bool 389DefaultFetch<Impl>::drain() 390{ 391 // Fetch is ready to drain at any time. 392 cpu->signalDrained(); 393 drainPending = true; 394 return true; 395} 396 397template <class Impl> 398void 399DefaultFetch<Impl>::resume() 400{ 401 drainPending = false; 402} 403 404template <class Impl> 405void 406DefaultFetch<Impl>::switchOut() 407{ 408 switchedOut = true; 409 // Branch predictor needs to have its state cleared. 410 branchPred.switchOut(); 411} 412 413template <class Impl> 414void 415DefaultFetch<Impl>::takeOverFrom() 416{ 417 // Reset all state 418 for (int i = 0; i < Impl::MaxThreads; ++i) { 419 stalls[i].decode = 0; 420 stalls[i].rename = 0; 421 stalls[i].iew = 0; 422 stalls[i].commit = 0; 423 PC[i] = cpu->readPC(i); 424 nextPC[i] = cpu->readNextPC(i); 425#if THE_ISA != ALPHA_ISA 426 nextNPC[i] = cpu->readNextNPC(i); 427#endif 428 fetchStatus[i] = Running; 429 } 430 numInst = 0; 431 wroteToTimeBuffer = false; 432 _status = Inactive; 433 switchedOut = false; 434 branchPred.takeOverFrom(); 435} 436 437template <class Impl> 438void 439DefaultFetch<Impl>::wakeFromQuiesce() 440{ 441 DPRINTF(Fetch, "Waking up from quiesce\n"); 442 // Hopefully this is safe 443 // @todo: Allow other threads to wake from quiesce. 444 fetchStatus[0] = Running; 445} 446 447template <class Impl> 448inline void 449DefaultFetch<Impl>::switchToActive() 450{ 451 if (_status == Inactive) { 452 DPRINTF(Activity, "Activating stage.\n"); 453 454 cpu->activateStage(O3CPU::FetchIdx); 455 456 _status = Active; 457 } 458} 459 460template <class Impl> 461inline void 462DefaultFetch<Impl>::switchToInactive() 463{ 464 if (_status == Active) { 465 DPRINTF(Activity, "Deactivating stage.\n"); 466 467 cpu->deactivateStage(O3CPU::FetchIdx); 468 469 _status = Inactive; 470 } 471} 472 473template <class Impl> 474bool 475DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC) 476{ 477 // Do branch prediction check here. 478 // A bit of a misnomer...next_PC is actually the current PC until 479 // this function updates it. 480 bool predict_taken; 481 482 if (!inst->isControl()) { 483 next_PC = next_PC + instSize; 484 inst->setPredTarg(next_PC); 485 return false; 486 } 487 488 predict_taken = branchPred.predict(inst, next_PC, inst->threadNumber); 489 490 ++fetchedBranches; 491 492 if (predict_taken) { 493 ++predictedBranches; 494 } 495 496 return predict_taken; 497} 498 499template <class Impl> 500bool 501DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid) 502{ 503 Fault fault = NoFault; 504 505#if FULL_SYSTEM 506 // Flag to say whether or not address is physical addr. 507 unsigned flags = cpu->inPalMode(fetch_PC) ? PHYSICAL : 0; 508#else 509 unsigned flags = 0; 510#endif // FULL_SYSTEM 511 512 if (cacheBlocked || (interruptPending && flags == 0) || drainPending) { 513 // Hold off fetch from getting new instructions when: 514 // Cache is blocked, or 515 // while an interrupt is pending and we're not in PAL mode, or 516 // fetch is switched out. 517 return false; 518 } 519 520 // Align the fetch PC so it's at the start of a cache block. 521 fetch_PC = icacheBlockAlignPC(fetch_PC); 522 523 // Setup the memReq to do a read of the first instruction's address. 524 // Set the appropriate read size and flags as well. 525 // Build request here. 526 RequestPtr mem_req = new Request(tid, fetch_PC, cacheBlkSize, flags, 527 fetch_PC, cpu->readCpuId(), tid); 528 529 memReq[tid] = mem_req; 530 531 // Translate the instruction request. 532 fault = cpu->translateInstReq(mem_req, cpu->thread[tid]); 533 534 // In the case of faults, the fetch stage may need to stall and wait 535 // for the ITB miss to be handled. 536 537 // If translation was successful, attempt to read the first 538 // instruction. 539 if (fault == NoFault) { 540#if 0 541 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || 542 memReq[tid]->flags & UNCACHEABLE) { 543 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " 544 "misspeculating path)!", 545 memReq[tid]->paddr); 546 ret_fault = TheISA::genMachineCheckFault(); 547 return false; 548 } 549#endif 550 551 // Build packet here. 552 PacketPtr data_pkt = new Packet(mem_req, 553 Packet::ReadReq, Packet::Broadcast); 554 data_pkt->dataStatic(cacheData[tid]); 555 556 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 557 558 fetchedCacheLines++; 559 560 // Now do the timing access to see whether or not the instruction 561 // exists within the cache. 562 if (!icachePort->sendTiming(data_pkt)) { 563 assert(retryPkt == NULL); 564 assert(retryTid == -1); 565 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 566 fetchStatus[tid] = IcacheWaitRetry; 567 retryPkt = data_pkt; 568 retryTid = tid; 569 cacheBlocked = true; 570 return false; 571 } 572 573 DPRINTF(Fetch, "[tid:%i]: Doing cache access.\n", tid); 574 575 lastIcacheStall[tid] = curTick; 576 577 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 578 "response.\n", tid); 579 580 fetchStatus[tid] = IcacheWaitResponse; 581 } else { 582 delete mem_req; 583 memReq[tid] = NULL; 584 } 585 586 ret_fault = fault; 587 return true; 588} 589 590template <class Impl> 591inline void 592DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid) 593{ 594 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x.\n", 595 tid, new_PC); 596 597 PC[tid] = new_PC; 598 nextPC[tid] = new_PC + instSize; 599 600 // Clear the icache miss if it's outstanding. 601 if (fetchStatus[tid] == IcacheWaitResponse) { 602 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 603 tid); 604 memReq[tid] = NULL; 605 } 606 607 // Get rid of the retrying packet if it was from this thread. 608 if (retryTid == tid) { 609 assert(cacheBlocked); 610 cacheBlocked = false; 611 retryTid = -1; 612 retryPkt = NULL; 613 delete retryPkt->req; 614 delete retryPkt; 615 } 616 617 fetchStatus[tid] = Squashing; 618 619 ++fetchSquashCycles; 620} 621 622template<class Impl> 623void 624DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, 625 const InstSeqNum &seq_num, 626 unsigned tid) 627{ 628 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid); 629 630 doSquash(new_PC, tid); 631 632 // Tell the CPU to remove any instructions that are in flight between 633 // fetch and decode. 634 cpu->removeInstsUntil(seq_num, tid); 635} 636 637template<class Impl> 638bool 639DefaultFetch<Impl>::checkStall(unsigned tid) const 640{ 641 bool ret_val = false; 642 643 if (cpu->contextSwitch) { 644 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 645 ret_val = true; 646 } else if (stalls[tid].decode) { 647 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 648 ret_val = true; 649 } else if (stalls[tid].rename) { 650 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 651 ret_val = true; 652 } else if (stalls[tid].iew) { 653 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 654 ret_val = true; 655 } else if (stalls[tid].commit) { 656 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 657 ret_val = true; 658 } 659 660 return ret_val; 661} 662 663template<class Impl> 664typename DefaultFetch<Impl>::FetchStatus 665DefaultFetch<Impl>::updateFetchStatus() 666{ 667 //Check Running 668 list<unsigned>::iterator threads = (*activeThreads).begin(); 669 670 while (threads != (*activeThreads).end()) { 671 672 unsigned tid = *threads++; 673 674 if (fetchStatus[tid] == Running || 675 fetchStatus[tid] == Squashing || 676 fetchStatus[tid] == IcacheAccessComplete) { 677 678 if (_status == Inactive) { 679 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 680 681 if (fetchStatus[tid] == IcacheAccessComplete) { 682 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 683 "completion\n",tid); 684 } 685 686 cpu->activateStage(O3CPU::FetchIdx); 687 } 688 689 return Active; 690 } 691 } 692 693 // Stage is switching from active to inactive, notify CPU of it. 694 if (_status == Active) { 695 DPRINTF(Activity, "Deactivating stage.\n"); 696 697 cpu->deactivateStage(O3CPU::FetchIdx); 698 } 699 700 return Inactive; 701} 702 703template <class Impl> 704void 705DefaultFetch<Impl>::squash(const Addr &new_PC, unsigned tid) 706{ 707 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid); 708 709 doSquash(new_PC, tid); 710 711 // Tell the CPU to remove any instructions that are not in the ROB. 712 cpu->removeInstsNotInROB(tid); 713} 714 715template <class Impl> 716void 717DefaultFetch<Impl>::tick() 718{ 719 list<unsigned>::iterator threads = (*activeThreads).begin(); 720 bool status_change = false; 721 722 wroteToTimeBuffer = false; 723 724 while (threads != (*activeThreads).end()) { 725 unsigned tid = *threads++; 726 727 // Check the signals for each thread to determine the proper status 728 // for each thread. 729 bool updated_status = checkSignalsAndUpdate(tid); 730 status_change = status_change || updated_status; 731 } 732 733 DPRINTF(Fetch, "Running stage.\n"); 734 735 // Reset the number of the instruction we're fetching. 736 numInst = 0; 737 738#if FULL_SYSTEM 739 if (fromCommit->commitInfo[0].interruptPending) { 740 interruptPending = true; 741 } 742 743 if (fromCommit->commitInfo[0].clearInterrupt) { 744 interruptPending = false; 745 } 746#endif 747 748 for (threadFetched = 0; threadFetched < numFetchingThreads; 749 threadFetched++) { 750 // Fetch each of the actively fetching threads. 751 fetch(status_change); 752 } 753 754 // Record number of instructions fetched this cycle for distribution. 755 fetchNisnDist.sample(numInst); 756 757 if (status_change) { 758 // Change the fetch stage status if there was a status change. 759 _status = updateFetchStatus(); 760 } 761 762 // If there was activity this cycle, inform the CPU of it. 763 if (wroteToTimeBuffer || cpu->contextSwitch) { 764 DPRINTF(Activity, "Activity this cycle.\n"); 765 766 cpu->activityThisCycle(); 767 } 768} 769 770template <class Impl> 771bool 772DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) 773{ 774 // Update the per thread stall statuses. 775 if (fromDecode->decodeBlock[tid]) { 776 stalls[tid].decode = true; 777 } 778 779 if (fromDecode->decodeUnblock[tid]) { 780 assert(stalls[tid].decode); 781 assert(!fromDecode->decodeBlock[tid]); 782 stalls[tid].decode = false; 783 } 784 785 if (fromRename->renameBlock[tid]) { 786 stalls[tid].rename = true; 787 } 788 789 if (fromRename->renameUnblock[tid]) { 790 assert(stalls[tid].rename); 791 assert(!fromRename->renameBlock[tid]); 792 stalls[tid].rename = false; 793 } 794 795 if (fromIEW->iewBlock[tid]) { 796 stalls[tid].iew = true; 797 } 798 799 if (fromIEW->iewUnblock[tid]) { 800 assert(stalls[tid].iew); 801 assert(!fromIEW->iewBlock[tid]); 802 stalls[tid].iew = false; 803 } 804 805 if (fromCommit->commitBlock[tid]) { 806 stalls[tid].commit = true; 807 } 808 809 if (fromCommit->commitUnblock[tid]) { 810 assert(stalls[tid].commit); 811 assert(!fromCommit->commitBlock[tid]); 812 stalls[tid].commit = false; 813 } 814 815 // Check squash signals from commit. 816 if (fromCommit->commitInfo[tid].squash) { 817 818 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 819 "from commit.\n",tid); 820 821 // In any case, squash. 822 squash(fromCommit->commitInfo[tid].nextPC,tid); 823 824 // Also check if there's a mispredict that happened. 825 if (fromCommit->commitInfo[tid].branchMispredict) { 826 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 827 fromCommit->commitInfo[tid].nextPC, 828 fromCommit->commitInfo[tid].branchTaken, 829 tid); 830 } else { 831 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 832 tid); 833 } 834 835 return true; 836 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 837 // Update the branch predictor if it wasn't a squashed instruction 838 // that was broadcasted. 839 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 840 } 841 842 // Check ROB squash signals from commit. 843 if (fromCommit->commitInfo[tid].robSquashing) { 844 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 845 846 // Continue to squash. 847 fetchStatus[tid] = Squashing; 848 849 return true; 850 } 851 852 // Check squash signals from decode. 853 if (fromDecode->decodeInfo[tid].squash) { 854 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 855 "from decode.\n",tid); 856 857 // Update the branch predictor. 858 if (fromDecode->decodeInfo[tid].branchMispredict) { 859 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 860 fromDecode->decodeInfo[tid].nextPC, 861 fromDecode->decodeInfo[tid].branchTaken, 862 tid); 863 } else { 864 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 865 tid); 866 } 867 868 if (fetchStatus[tid] != Squashing) { 869 // Squash unless we're already squashing 870 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 871 fromDecode->decodeInfo[tid].doneSeqNum, 872 tid); 873 874 return true; 875 } 876 } 877 878 if (checkStall(tid) && fetchStatus[tid] != IcacheWaitResponse) { 879 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 880 881 fetchStatus[tid] = Blocked; 882 883 return true; 884 } 885 886 if (fetchStatus[tid] == Blocked || 887 fetchStatus[tid] == Squashing) { 888 // Switch status to running if fetch isn't being told to block or 889 // squash this cycle. 890 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 891 tid); 892 893 fetchStatus[tid] = Running; 894 895 return true; 896 } 897 898 // If we've reached this point, we have not gotten any signals that 899 // cause fetch to change its status. Fetch remains the same as before. 900 return false; 901} 902 903template<class Impl> 904void 905DefaultFetch<Impl>::fetch(bool &status_change) 906{ 907 ////////////////////////////////////////// 908 // Start actual fetch 909 ////////////////////////////////////////// 910 int tid = getFetchingThread(fetchPolicy); 911 912 if (tid == -1) { 913 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 914 915 // Breaks looping condition in tick() 916 threadFetched = numFetchingThreads; 917 return; 918 } 919 920 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 921 922 // The current PC. 923 Addr &fetch_PC = PC[tid]; 924 925 // Fault code for memory access. 926 Fault fault = NoFault; 927 928 // If returning from the delay of a cache miss, then update the status 929 // to running, otherwise do the cache access. Possibly move this up 930 // to tick() function. 931 if (fetchStatus[tid] == IcacheAccessComplete) { 932 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", 933 tid); 934 935 fetchStatus[tid] = Running; 936 status_change = true; 937 } else if (fetchStatus[tid] == Running) { 938 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 939 "instruction, starting at PC %08p.\n", 940 tid, fetch_PC); 941 942 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid); 943 if (!fetch_success) { 944 if (cacheBlocked) { 945 ++icacheStallCycles; 946 } else { 947 ++fetchMiscStallCycles; 948 } 949 return; 950 } 951 } else { 952 if (fetchStatus[tid] == Idle) { 953 ++fetchIdleCycles; 954 } else if (fetchStatus[tid] == Blocked) { 955 ++fetchBlockedCycles; 956 } else if (fetchStatus[tid] == Squashing) { 957 ++fetchSquashCycles; 958 } else if (fetchStatus[tid] == IcacheWaitResponse) { 959 ++icacheStallCycles; 960 } 961 962 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so 963 // fetch should do nothing. 964 return; 965 } 966 967 ++fetchCycles; 968 969 // If we had a stall due to an icache miss, then return. 970 if (fetchStatus[tid] == IcacheWaitResponse) { 971 ++icacheStallCycles; 972 status_change = true; 973 return; 974 } 975 976 Addr next_PC = fetch_PC; 977 InstSeqNum inst_seq; 978 MachInst inst; 979 ExtMachInst ext_inst; 980 // @todo: Fix this hack. 981 unsigned offset = (fetch_PC & cacheBlkMask) & ~3; 982 983 if (fault == NoFault) { 984 // If the read of the first instruction was successful, then grab the 985 // instructions from the rest of the cache line and put them into the 986 // queue heading to decode. 987 988 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 989 "decode.\n",tid); 990 991 // Need to keep track of whether or not a predicted branch 992 // ended this fetch block. 993 bool predicted_branch = false; 994 995 for (; 996 offset < cacheBlkSize && 997 numInst < fetchWidth && 998 !predicted_branch; 999 ++numInst) { 1000 1001 // Get a sequence number. 1002 inst_seq = cpu->getAndIncrementInstSeq(); 1003 1004 // Make sure this is a valid index. 1005 assert(offset <= cacheBlkSize - instSize); 1006 1007 // Get the instruction from the array of the cache line. 1008 inst = gtoh(*reinterpret_cast<MachInst *> 1009 (&cacheData[tid][offset])); 1010 1011 ext_inst = TheISA::makeExtMI(inst, fetch_PC); 1012 1013 // Create a new DynInst from the instruction fetched. 1014 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1015 next_PC, 1016 inst_seq, cpu); 1017 instruction->setTid(tid); 1018 1019 instruction->setASID(tid); 1020 1021 instruction->setThreadState(cpu->thread[tid]); 1022 1023 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created " 1024 "[sn:%lli]\n", 1025 tid, instruction->readPC(), inst_seq); 1026 1027 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", 1028 tid, instruction->staticInst->disassemble(fetch_PC)); 1029 1030 instruction->traceData = 1031 Trace::getInstRecord(curTick, cpu->tcBase(tid), cpu, 1032 instruction->staticInst, 1033 instruction->readPC(),tid); 1034 1035 predicted_branch = lookupAndUpdateNextPC(instruction, next_PC); 1036 1037 // Add instruction to the CPU's list of instructions. 1038 instruction->setInstListIt(cpu->addInst(instruction)); 1039 1040 // Write the instruction to the first slot in the queue 1041 // that heads to decode. 1042 toDecode->insts[numInst] = instruction; 1043 1044 toDecode->size++; 1045 1046 // Increment stat of fetched instructions. 1047 ++fetchedInsts; 1048 1049 // Move to the next instruction, unless we have a branch. 1050 fetch_PC = next_PC; 1051 1052 if (instruction->isQuiesce()) { 1053 warn("cycle %lli: Quiesce instruction encountered, halting fetch!", 1054 curTick); 1055 fetchStatus[tid] = QuiescePending; 1056 ++numInst; 1057 status_change = true; 1058 break; 1059 } 1060 1061 offset+= instSize; 1062 } 1063 } 1064 1065 if (numInst > 0) { 1066 wroteToTimeBuffer = true; 1067 } 1068 1069 // Now that fetching is completed, update the PC to signify what the next 1070 // cycle will be. 1071 if (fault == NoFault) { 1072 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n",tid, next_PC); 1073 1074#if THE_ISA == ALPHA_ISA 1075 PC[tid] = next_PC; 1076 nextPC[tid] = next_PC + instSize; 1077#else 1078 PC[tid] = next_PC; 1079 nextPC[tid] = next_PC + instSize; 1080 nextPC[tid] = next_PC + instSize; 1081 1082 thread->setNextPC(thread->readNextNPC()); 1083 thread->setNextNPC(thread->readNextNPC() + sizeof(MachInst)); 1084#endif 1085 } else { 1086 // We shouldn't be in an icache miss and also have a fault (an ITB 1087 // miss) 1088 if (fetchStatus[tid] == IcacheWaitResponse) { 1089 panic("Fetch should have exited prior to this!"); 1090 } 1091 1092 // Send the fault to commit. This thread will not do anything 1093 // until commit handles the fault. The only other way it can 1094 // wake up is if a squash comes along and changes the PC. 1095#if FULL_SYSTEM 1096 assert(numInst != fetchWidth); 1097 // Get a sequence number. 1098 inst_seq = cpu->getAndIncrementInstSeq(); 1099 // We will use a nop in order to carry the fault. 1100 ext_inst = TheISA::NoopMachInst; 1101 1102 // Create a new DynInst from the dummy nop. 1103 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1104 next_PC, 1105 inst_seq, cpu); 1106 instruction->setPredTarg(next_PC + instSize); 1107 instruction->setTid(tid); 1108 1109 instruction->setASID(tid); 1110 1111 instruction->setThreadState(cpu->thread[tid]); 1112 1113 instruction->traceData = NULL; 1114 1115 instruction->setInstListIt(cpu->addInst(instruction)); 1116 1117 instruction->fault = fault; 1118 1119 toDecode->insts[numInst] = instruction; 1120 toDecode->size++; 1121 1122 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid); 1123 1124 fetchStatus[tid] = TrapPending; 1125 status_change = true; 1126 1127 warn("cycle %lli: fault (%d) detected @ PC %08p", curTick, fault, PC[tid]); 1128#else // !FULL_SYSTEM 1129 warn("cycle %lli: fault (%d) detected @ PC %08p", curTick, fault, PC[tid]); 1130#endif // FULL_SYSTEM 1131 } 1132} 1133 1134template<class Impl> 1135void 1136DefaultFetch<Impl>::recvRetry() 1137{ 1138 assert(cacheBlocked); 1139 if (retryPkt != NULL) { 1140 assert(retryTid != -1); 1141 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1142 1143 if (icachePort->sendTiming(retryPkt)) { 1144 fetchStatus[retryTid] = IcacheWaitResponse; 1145 retryPkt = NULL; 1146 retryTid = -1; 1147 cacheBlocked = false; 1148 } 1149 } else { 1150 assert(retryTid == -1); 1151 // Access has been squashed since it was sent out. Just clear 1152 // the cache being blocked. 1153 cacheBlocked = false; 1154 } 1155} 1156 1157/////////////////////////////////////// 1158// // 1159// SMT FETCH POLICY MAINTAINED HERE // 1160// // 1161/////////////////////////////////////// 1162template<class Impl> 1163int 1164DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1165{ 1166 if (numThreads > 1) { 1167 switch (fetch_priority) { 1168 1169 case SingleThread: 1170 return 0; 1171 1172 case RoundRobin: 1173 return roundRobin(); 1174 1175 case IQ: 1176 return iqCount(); 1177 1178 case LSQ: 1179 return lsqCount(); 1180 1181 case Branch: 1182 return branchCount(); 1183 1184 default: 1185 return -1; 1186 } 1187 } else { 1188 int tid = *((*activeThreads).begin()); 1189 1190 if (fetchStatus[tid] == Running || 1191 fetchStatus[tid] == IcacheAccessComplete || 1192 fetchStatus[tid] == Idle) { 1193 return tid; 1194 } else { 1195 return -1; 1196 } 1197 } 1198 1199} 1200 1201 1202template<class Impl> 1203int 1204DefaultFetch<Impl>::roundRobin() 1205{ 1206 list<unsigned>::iterator pri_iter = priorityList.begin(); 1207 list<unsigned>::iterator end = priorityList.end(); 1208 1209 int high_pri; 1210 1211 while (pri_iter != end) { 1212 high_pri = *pri_iter; 1213 1214 assert(high_pri <= numThreads); 1215 1216 if (fetchStatus[high_pri] == Running || 1217 fetchStatus[high_pri] == IcacheAccessComplete || 1218 fetchStatus[high_pri] == Idle) { 1219 1220 priorityList.erase(pri_iter); 1221 priorityList.push_back(high_pri); 1222 1223 return high_pri; 1224 } 1225 1226 pri_iter++; 1227 } 1228 1229 return -1; 1230} 1231 1232template<class Impl> 1233int 1234DefaultFetch<Impl>::iqCount() 1235{ 1236 priority_queue<unsigned> PQ; 1237 1238 list<unsigned>::iterator threads = (*activeThreads).begin(); 1239 1240 while (threads != (*activeThreads).end()) { 1241 unsigned tid = *threads++; 1242 1243 PQ.push(fromIEW->iewInfo[tid].iqCount); 1244 } 1245 1246 while (!PQ.empty()) { 1247 1248 unsigned high_pri = PQ.top(); 1249 1250 if (fetchStatus[high_pri] == Running || 1251 fetchStatus[high_pri] == IcacheAccessComplete || 1252 fetchStatus[high_pri] == Idle) 1253 return high_pri; 1254 else 1255 PQ.pop(); 1256 1257 } 1258 1259 return -1; 1260} 1261 1262template<class Impl> 1263int 1264DefaultFetch<Impl>::lsqCount() 1265{ 1266 priority_queue<unsigned> PQ; 1267 1268 1269 list<unsigned>::iterator threads = (*activeThreads).begin(); 1270 1271 while (threads != (*activeThreads).end()) { 1272 unsigned tid = *threads++; 1273 1274 PQ.push(fromIEW->iewInfo[tid].ldstqCount); 1275 } 1276 1277 while (!PQ.empty()) { 1278 1279 unsigned high_pri = PQ.top(); 1280 1281 if (fetchStatus[high_pri] == Running || 1282 fetchStatus[high_pri] == IcacheAccessComplete || 1283 fetchStatus[high_pri] == Idle) 1284 return high_pri; 1285 else 1286 PQ.pop(); 1287 1288 } 1289 1290 return -1; 1291} 1292 1293template<class Impl> 1294int 1295DefaultFetch<Impl>::branchCount() 1296{ 1297 list<unsigned>::iterator threads = (*activeThreads).begin(); 1298 panic("Branch Count Fetch policy unimplemented\n"); 1299 return *threads; 1300} 1301