fetch_impl.hh revision 2894:a83675362809
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#if USE_CHECKER 284 if (cpu->checker) { 285 cpu->checker->setIcachePort(icachePort); 286 } 287#endif 288 289 // Fetch needs to start fetching instructions at the very beginning, 290 // so it must start up in active state. 291 switchToActive(); 292} 293 294template<class Impl> 295void 296DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 297{ 298 DPRINTF(Fetch, "Setting the time buffer pointer.\n"); 299 timeBuffer = time_buffer; 300 301 // Create wires to get information from proper places in time buffer. 302 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 303 fromRename = timeBuffer->getWire(-renameToFetchDelay); 304 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 305 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 306} 307 308template<class Impl> 309void 310DefaultFetch<Impl>::setActiveThreads(list<unsigned> *at_ptr) 311{ 312 DPRINTF(Fetch, "Setting active threads list pointer.\n"); 313 activeThreads = at_ptr; 314} 315 316template<class Impl> 317void 318DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 319{ 320 DPRINTF(Fetch, "Setting the fetch queue pointer.\n"); 321 fetchQueue = fq_ptr; 322 323 // Create wire to write information to proper place in fetch queue. 324 toDecode = fetchQueue->getWire(0); 325} 326 327template<class Impl> 328void 329DefaultFetch<Impl>::initStage() 330{ 331 // Setup PC and nextPC with initial state. 332 for (int tid = 0; tid < numThreads; tid++) { 333 PC[tid] = cpu->readPC(tid); 334 nextPC[tid] = cpu->readNextPC(tid); 335#if THE_ISA != ALPHA_ISA 336 nextNPC[tid] = cpu->readNextNPC(tid); 337#endif 338 } 339} 340 341template<class Impl> 342void 343DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 344{ 345 unsigned tid = pkt->req->getThreadNum(); 346 347 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid); 348 349 // Only change the status if it's still waiting on the icache access 350 // to return. 351 if (fetchStatus[tid] != IcacheWaitResponse || 352 pkt->req != memReq[tid] || 353 isSwitchedOut()) { 354 ++fetchIcacheSquashes; 355 delete pkt->req; 356 delete pkt; 357 return; 358 } 359 360 memcpy(cacheData[tid], pkt->getPtr<uint8_t *>(), cacheBlkSize); 361 362 if (!drainPending) { 363 // Wake up the CPU (if it went to sleep and was waiting on 364 // this completion event). 365 cpu->wakeCPU(); 366 367 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 368 tid); 369 370 switchToActive(); 371 } 372 373 // Only switch to IcacheAccessComplete if we're not stalled as well. 374 if (checkStall(tid)) { 375 fetchStatus[tid] = Blocked; 376 } else { 377 fetchStatus[tid] = IcacheAccessComplete; 378 } 379 380 // Reset the mem req to NULL. 381 delete pkt->req; 382 delete pkt; 383 memReq[tid] = NULL; 384} 385 386template <class Impl> 387bool 388DefaultFetch<Impl>::drain() 389{ 390 // Fetch is ready to drain at any time. 391 cpu->signalDrained(); 392 drainPending = true; 393 return true; 394} 395 396template <class Impl> 397void 398DefaultFetch<Impl>::resume() 399{ 400 drainPending = false; 401} 402 403template <class Impl> 404void 405DefaultFetch<Impl>::switchOut() 406{ 407 switchedOut = true; 408 // Branch predictor needs to have its state cleared. 409 branchPred.switchOut(); 410} 411 412template <class Impl> 413void 414DefaultFetch<Impl>::takeOverFrom() 415{ 416 // Reset all state 417 for (int i = 0; i < Impl::MaxThreads; ++i) { 418 stalls[i].decode = 0; 419 stalls[i].rename = 0; 420 stalls[i].iew = 0; 421 stalls[i].commit = 0; 422 PC[i] = cpu->readPC(i); 423 nextPC[i] = cpu->readNextPC(i); 424#if THE_ISA != ALPHA_ISA 425 nextNPC[i] = cpu->readNextNPC(i); 426#endif 427 fetchStatus[i] = Running; 428 } 429 numInst = 0; 430 wroteToTimeBuffer = false; 431 _status = Inactive; 432 switchedOut = false; 433 branchPred.takeOverFrom(); 434} 435 436template <class Impl> 437void 438DefaultFetch<Impl>::wakeFromQuiesce() 439{ 440 DPRINTF(Fetch, "Waking up from quiesce\n"); 441 // Hopefully this is safe 442 // @todo: Allow other threads to wake from quiesce. 443 fetchStatus[0] = Running; 444} 445 446template <class Impl> 447inline void 448DefaultFetch<Impl>::switchToActive() 449{ 450 if (_status == Inactive) { 451 DPRINTF(Activity, "Activating stage.\n"); 452 453 cpu->activateStage(O3CPU::FetchIdx); 454 455 _status = Active; 456 } 457} 458 459template <class Impl> 460inline void 461DefaultFetch<Impl>::switchToInactive() 462{ 463 if (_status == Active) { 464 DPRINTF(Activity, "Deactivating stage.\n"); 465 466 cpu->deactivateStage(O3CPU::FetchIdx); 467 468 _status = Inactive; 469 } 470} 471 472template <class Impl> 473bool 474DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC) 475{ 476 // Do branch prediction check here. 477 // A bit of a misnomer...next_PC is actually the current PC until 478 // this function updates it. 479 bool predict_taken; 480 481 if (!inst->isControl()) { 482 next_PC = next_PC + instSize; 483 inst->setPredTarg(next_PC); 484 return false; 485 } 486 487 predict_taken = branchPred.predict(inst, next_PC, inst->threadNumber); 488 489 ++fetchedBranches; 490 491 if (predict_taken) { 492 ++predictedBranches; 493 } 494 495 return predict_taken; 496} 497 498template <class Impl> 499bool 500DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid) 501{ 502 Fault fault = NoFault; 503 504#if FULL_SYSTEM 505 // Flag to say whether or not address is physical addr. 506 unsigned flags = cpu->inPalMode(fetch_PC) ? PHYSICAL : 0; 507#else 508 unsigned flags = 0; 509#endif // FULL_SYSTEM 510 511 if (cacheBlocked || (interruptPending && flags == 0)) { 512 // Hold off fetch from getting new instructions when: 513 // Cache is blocked, or 514 // while an interrupt is pending and we're not in PAL mode, or 515 // fetch is switched out. 516 return false; 517 } 518 519 // Align the fetch PC so it's at the start of a cache block. 520 fetch_PC = icacheBlockAlignPC(fetch_PC); 521 522 // Setup the memReq to do a read of the first instruction's address. 523 // Set the appropriate read size and flags as well. 524 // Build request here. 525 RequestPtr mem_req = new Request(tid, fetch_PC, cacheBlkSize, flags, 526 fetch_PC, cpu->readCpuId(), tid); 527 528 memReq[tid] = mem_req; 529 530 // Translate the instruction request. 531 fault = cpu->translateInstReq(mem_req, cpu->thread[tid]); 532 533 // In the case of faults, the fetch stage may need to stall and wait 534 // for the ITB miss to be handled. 535 536 // If translation was successful, attempt to read the first 537 // instruction. 538 if (fault == NoFault) { 539#if 0 540 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || 541 memReq[tid]->flags & UNCACHEABLE) { 542 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " 543 "misspeculating path)!", 544 memReq[tid]->paddr); 545 ret_fault = TheISA::genMachineCheckFault(); 546 return false; 547 } 548#endif 549 550 // Build packet here. 551 PacketPtr data_pkt = new Packet(mem_req, 552 Packet::ReadReq, Packet::Broadcast); 553 data_pkt->dataDynamic(new uint8_t[cacheBlkSize]); 554 555 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 556 557 fetchedCacheLines++; 558 559 // Now do the timing access to see whether or not the instruction 560 // exists within the cache. 561 if (!icachePort->sendTiming(data_pkt)) { 562 assert(retryPkt == NULL); 563 assert(retryTid == -1); 564 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 565 fetchStatus[tid] = IcacheWaitRetry; 566 retryPkt = data_pkt; 567 retryTid = tid; 568 cacheBlocked = true; 569 return false; 570 } 571 572 DPRINTF(Fetch, "[tid:%i]: Doing cache access.\n", tid); 573 574 lastIcacheStall[tid] = curTick; 575 576 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 577 "response.\n", tid); 578 579 fetchStatus[tid] = IcacheWaitResponse; 580 } else { 581 delete mem_req; 582 memReq[tid] = NULL; 583 } 584 585 ret_fault = fault; 586 return true; 587} 588 589template <class Impl> 590inline void 591DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid) 592{ 593 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x.\n", 594 tid, new_PC); 595 596 PC[tid] = new_PC; 597 nextPC[tid] = new_PC + instSize; 598 599 // Clear the icache miss if it's outstanding. 600 if (fetchStatus[tid] == IcacheWaitResponse) { 601 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 602 tid); 603 memReq[tid] = NULL; 604 } 605 606 // Get rid of the retrying packet if it was from this thread. 607 if (retryTid == tid) { 608 assert(cacheBlocked); 609 cacheBlocked = false; 610 retryTid = -1; 611 retryPkt = NULL; 612 delete retryPkt->req; 613 delete retryPkt; 614 } 615 616 fetchStatus[tid] = Squashing; 617 618 ++fetchSquashCycles; 619} 620 621template<class Impl> 622void 623DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, 624 const InstSeqNum &seq_num, 625 unsigned tid) 626{ 627 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid); 628 629 doSquash(new_PC, tid); 630 631 // Tell the CPU to remove any instructions that are in flight between 632 // fetch and decode. 633 cpu->removeInstsUntil(seq_num, tid); 634} 635 636template<class Impl> 637bool 638DefaultFetch<Impl>::checkStall(unsigned tid) const 639{ 640 bool ret_val = false; 641 642 if (cpu->contextSwitch) { 643 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 644 ret_val = true; 645 } else if (stalls[tid].decode) { 646 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 647 ret_val = true; 648 } else if (stalls[tid].rename) { 649 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 650 ret_val = true; 651 } else if (stalls[tid].iew) { 652 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 653 ret_val = true; 654 } else if (stalls[tid].commit) { 655 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 656 ret_val = true; 657 } 658 659 return ret_val; 660} 661 662template<class Impl> 663typename DefaultFetch<Impl>::FetchStatus 664DefaultFetch<Impl>::updateFetchStatus() 665{ 666 //Check Running 667 list<unsigned>::iterator threads = (*activeThreads).begin(); 668 669 while (threads != (*activeThreads).end()) { 670 671 unsigned tid = *threads++; 672 673 if (fetchStatus[tid] == Running || 674 fetchStatus[tid] == Squashing || 675 fetchStatus[tid] == IcacheAccessComplete) { 676 677 if (_status == Inactive) { 678 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 679 680 if (fetchStatus[tid] == IcacheAccessComplete) { 681 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 682 "completion\n",tid); 683 } 684 685 cpu->activateStage(O3CPU::FetchIdx); 686 } 687 688 return Active; 689 } 690 } 691 692 // Stage is switching from active to inactive, notify CPU of it. 693 if (_status == Active) { 694 DPRINTF(Activity, "Deactivating stage.\n"); 695 696 cpu->deactivateStage(O3CPU::FetchIdx); 697 } 698 699 return Inactive; 700} 701 702template <class Impl> 703void 704DefaultFetch<Impl>::squash(const Addr &new_PC, unsigned tid) 705{ 706 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid); 707 708 doSquash(new_PC, tid); 709 710 // Tell the CPU to remove any instructions that are not in the ROB. 711 cpu->removeInstsNotInROB(tid); 712} 713 714template <class Impl> 715void 716DefaultFetch<Impl>::tick() 717{ 718 list<unsigned>::iterator threads = (*activeThreads).begin(); 719 bool status_change = false; 720 721 wroteToTimeBuffer = false; 722 723 while (threads != (*activeThreads).end()) { 724 unsigned tid = *threads++; 725 726 // Check the signals for each thread to determine the proper status 727 // for each thread. 728 bool updated_status = checkSignalsAndUpdate(tid); 729 status_change = status_change || updated_status; 730 } 731 732 DPRINTF(Fetch, "Running stage.\n"); 733 734 // Reset the number of the instruction we're fetching. 735 numInst = 0; 736 737#if FULL_SYSTEM 738 if (fromCommit->commitInfo[0].interruptPending) { 739 interruptPending = true; 740 } 741 742 if (fromCommit->commitInfo[0].clearInterrupt) { 743 interruptPending = false; 744 } 745#endif 746 747 for (threadFetched = 0; threadFetched < numFetchingThreads; 748 threadFetched++) { 749 // Fetch each of the actively fetching threads. 750 fetch(status_change); 751 } 752 753 // Record number of instructions fetched this cycle for distribution. 754 fetchNisnDist.sample(numInst); 755 756 if (status_change) { 757 // Change the fetch stage status if there was a status change. 758 _status = updateFetchStatus(); 759 } 760 761 // If there was activity this cycle, inform the CPU of it. 762 if (wroteToTimeBuffer || cpu->contextSwitch) { 763 DPRINTF(Activity, "Activity this cycle.\n"); 764 765 cpu->activityThisCycle(); 766 } 767} 768 769template <class Impl> 770bool 771DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) 772{ 773 // Update the per thread stall statuses. 774 if (fromDecode->decodeBlock[tid]) { 775 stalls[tid].decode = true; 776 } 777 778 if (fromDecode->decodeUnblock[tid]) { 779 assert(stalls[tid].decode); 780 assert(!fromDecode->decodeBlock[tid]); 781 stalls[tid].decode = false; 782 } 783 784 if (fromRename->renameBlock[tid]) { 785 stalls[tid].rename = true; 786 } 787 788 if (fromRename->renameUnblock[tid]) { 789 assert(stalls[tid].rename); 790 assert(!fromRename->renameBlock[tid]); 791 stalls[tid].rename = false; 792 } 793 794 if (fromIEW->iewBlock[tid]) { 795 stalls[tid].iew = true; 796 } 797 798 if (fromIEW->iewUnblock[tid]) { 799 assert(stalls[tid].iew); 800 assert(!fromIEW->iewBlock[tid]); 801 stalls[tid].iew = false; 802 } 803 804 if (fromCommit->commitBlock[tid]) { 805 stalls[tid].commit = true; 806 } 807 808 if (fromCommit->commitUnblock[tid]) { 809 assert(stalls[tid].commit); 810 assert(!fromCommit->commitBlock[tid]); 811 stalls[tid].commit = false; 812 } 813 814 // Check squash signals from commit. 815 if (fromCommit->commitInfo[tid].squash) { 816 817 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 818 "from commit.\n",tid); 819 820 // In any case, squash. 821 squash(fromCommit->commitInfo[tid].nextPC,tid); 822 823 // Also check if there's a mispredict that happened. 824 if (fromCommit->commitInfo[tid].branchMispredict) { 825 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 826 fromCommit->commitInfo[tid].nextPC, 827 fromCommit->commitInfo[tid].branchTaken, 828 tid); 829 } else { 830 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 831 tid); 832 } 833 834 return true; 835 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 836 // Update the branch predictor if it wasn't a squashed instruction 837 // that was broadcasted. 838 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 839 } 840 841 // Check ROB squash signals from commit. 842 if (fromCommit->commitInfo[tid].robSquashing) { 843 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 844 845 // Continue to squash. 846 fetchStatus[tid] = Squashing; 847 848 return true; 849 } 850 851 // Check squash signals from decode. 852 if (fromDecode->decodeInfo[tid].squash) { 853 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 854 "from decode.\n",tid); 855 856 // Update the branch predictor. 857 if (fromDecode->decodeInfo[tid].branchMispredict) { 858 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 859 fromDecode->decodeInfo[tid].nextPC, 860 fromDecode->decodeInfo[tid].branchTaken, 861 tid); 862 } else { 863 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 864 tid); 865 } 866 867 if (fetchStatus[tid] != Squashing) { 868 // Squash unless we're already squashing 869 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 870 fromDecode->decodeInfo[tid].doneSeqNum, 871 tid); 872 873 return true; 874 } 875 } 876 877 if (checkStall(tid) && fetchStatus[tid] != IcacheWaitResponse) { 878 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 879 880 fetchStatus[tid] = Blocked; 881 882 return true; 883 } 884 885 if (fetchStatus[tid] == Blocked || 886 fetchStatus[tid] == Squashing) { 887 // Switch status to running if fetch isn't being told to block or 888 // squash this cycle. 889 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 890 tid); 891 892 fetchStatus[tid] = Running; 893 894 return true; 895 } 896 897 // If we've reached this point, we have not gotten any signals that 898 // cause fetch to change its status. Fetch remains the same as before. 899 return false; 900} 901 902template<class Impl> 903void 904DefaultFetch<Impl>::fetch(bool &status_change) 905{ 906 ////////////////////////////////////////// 907 // Start actual fetch 908 ////////////////////////////////////////// 909 int tid = getFetchingThread(fetchPolicy); 910 911 if (tid == -1 || drainPending) { 912 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 913 914 // Breaks looping condition in tick() 915 threadFetched = numFetchingThreads; 916 return; 917 } 918 919 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 920 921 // The current PC. 922 Addr &fetch_PC = PC[tid]; 923 924 // Fault code for memory access. 925 Fault fault = NoFault; 926 927 // If returning from the delay of a cache miss, then update the status 928 // to running, otherwise do the cache access. Possibly move this up 929 // to tick() function. 930 if (fetchStatus[tid] == IcacheAccessComplete) { 931 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", 932 tid); 933 934 fetchStatus[tid] = Running; 935 status_change = true; 936 } else if (fetchStatus[tid] == Running) { 937 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 938 "instruction, starting at PC %08p.\n", 939 tid, fetch_PC); 940 941 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid); 942 if (!fetch_success) { 943 if (cacheBlocked) { 944 ++icacheStallCycles; 945 } else { 946 ++fetchMiscStallCycles; 947 } 948 return; 949 } 950 } else { 951 if (fetchStatus[tid] == Idle) { 952 ++fetchIdleCycles; 953 } else if (fetchStatus[tid] == Blocked) { 954 ++fetchBlockedCycles; 955 } else if (fetchStatus[tid] == Squashing) { 956 ++fetchSquashCycles; 957 } else if (fetchStatus[tid] == IcacheWaitResponse) { 958 ++icacheStallCycles; 959 } 960 961 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so 962 // fetch should do nothing. 963 return; 964 } 965 966 ++fetchCycles; 967 968 // If we had a stall due to an icache miss, then return. 969 if (fetchStatus[tid] == IcacheWaitResponse) { 970 ++icacheStallCycles; 971 status_change = true; 972 return; 973 } 974 975 Addr next_PC = fetch_PC; 976 InstSeqNum inst_seq; 977 MachInst inst; 978 ExtMachInst ext_inst; 979 // @todo: Fix this hack. 980 unsigned offset = (fetch_PC & cacheBlkMask) & ~3; 981 982 if (fault == NoFault) { 983 // If the read of the first instruction was successful, then grab the 984 // instructions from the rest of the cache line and put them into the 985 // queue heading to decode. 986 987 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 988 "decode.\n",tid); 989 990 // Need to keep track of whether or not a predicted branch 991 // ended this fetch block. 992 bool predicted_branch = false; 993 994 for (; 995 offset < cacheBlkSize && 996 numInst < fetchWidth && 997 !predicted_branch; 998 ++numInst) { 999 1000 // Get a sequence number. 1001 inst_seq = cpu->getAndIncrementInstSeq(); 1002 1003 // Make sure this is a valid index. 1004 assert(offset <= cacheBlkSize - instSize); 1005 1006 // Get the instruction from the array of the cache line. 1007 inst = gtoh(*reinterpret_cast<MachInst *> 1008 (&cacheData[tid][offset])); 1009 1010 ext_inst = TheISA::makeExtMI(inst, fetch_PC); 1011 1012 // Create a new DynInst from the instruction fetched. 1013 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1014 next_PC, 1015 inst_seq, cpu); 1016 instruction->setTid(tid); 1017 1018 instruction->setASID(tid); 1019 1020 instruction->setThreadState(cpu->thread[tid]); 1021 1022 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created " 1023 "[sn:%lli]\n", 1024 tid, instruction->readPC(), inst_seq); 1025 1026 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", 1027 tid, instruction->staticInst->disassemble(fetch_PC)); 1028 1029 instruction->traceData = 1030 Trace::getInstRecord(curTick, cpu->tcBase(tid), cpu, 1031 instruction->staticInst, 1032 instruction->readPC(),tid); 1033 1034 predicted_branch = lookupAndUpdateNextPC(instruction, next_PC); 1035 1036 // Add instruction to the CPU's list of instructions. 1037 instruction->setInstListIt(cpu->addInst(instruction)); 1038 1039 // Write the instruction to the first slot in the queue 1040 // that heads to decode. 1041 toDecode->insts[numInst] = instruction; 1042 1043 toDecode->size++; 1044 1045 // Increment stat of fetched instructions. 1046 ++fetchedInsts; 1047 1048 // Move to the next instruction, unless we have a branch. 1049 fetch_PC = next_PC; 1050 1051 if (instruction->isQuiesce()) { 1052 warn("cycle %lli: Quiesce instruction encountered, halting fetch!", 1053 curTick); 1054 fetchStatus[tid] = QuiescePending; 1055 ++numInst; 1056 status_change = true; 1057 break; 1058 } 1059 1060 offset+= instSize; 1061 } 1062 } 1063 1064 if (numInst > 0) { 1065 wroteToTimeBuffer = true; 1066 } 1067 1068 // Now that fetching is completed, update the PC to signify what the next 1069 // cycle will be. 1070 if (fault == NoFault) { 1071 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n",tid, next_PC); 1072 1073#if THE_ISA == ALPHA_ISA 1074 PC[tid] = next_PC; 1075 nextPC[tid] = next_PC + instSize; 1076#else 1077 PC[tid] = next_PC; 1078 nextPC[tid] = next_PC + instSize; 1079 nextPC[tid] = next_PC + instSize; 1080 1081 thread->setNextPC(thread->readNextNPC()); 1082 thread->setNextNPC(thread->readNextNPC() + sizeof(MachInst)); 1083#endif 1084 } else { 1085 // We shouldn't be in an icache miss and also have a fault (an ITB 1086 // miss) 1087 if (fetchStatus[tid] == IcacheWaitResponse) { 1088 panic("Fetch should have exited prior to this!"); 1089 } 1090 1091 // Send the fault to commit. This thread will not do anything 1092 // until commit handles the fault. The only other way it can 1093 // wake up is if a squash comes along and changes the PC. 1094#if FULL_SYSTEM 1095 assert(numInst != fetchWidth); 1096 // Get a sequence number. 1097 inst_seq = cpu->getAndIncrementInstSeq(); 1098 // We will use a nop in order to carry the fault. 1099 ext_inst = TheISA::NoopMachInst; 1100 1101 // Create a new DynInst from the dummy nop. 1102 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1103 next_PC, 1104 inst_seq, cpu); 1105 instruction->setPredTarg(next_PC + instSize); 1106 instruction->setTid(tid); 1107 1108 instruction->setASID(tid); 1109 1110 instruction->setThreadState(cpu->thread[tid]); 1111 1112 instruction->traceData = NULL; 1113 1114 instruction->setInstListIt(cpu->addInst(instruction)); 1115 1116 instruction->fault = fault; 1117 1118 toDecode->insts[numInst] = instruction; 1119 toDecode->size++; 1120 1121 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid); 1122 1123 fetchStatus[tid] = TrapPending; 1124 status_change = true; 1125 1126 warn("cycle %lli: fault (%d) detected @ PC %08p", curTick, fault, PC[tid]); 1127#else // !FULL_SYSTEM 1128 warn("cycle %lli: fault (%d) detected @ PC %08p", curTick, fault, PC[tid]); 1129#endif // FULL_SYSTEM 1130 } 1131} 1132 1133template<class Impl> 1134void 1135DefaultFetch<Impl>::recvRetry() 1136{ 1137 assert(cacheBlocked); 1138 if (retryPkt != NULL) { 1139 assert(retryTid != -1); 1140 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1141 1142 if (icachePort->sendTiming(retryPkt)) { 1143 fetchStatus[retryTid] = IcacheWaitResponse; 1144 retryPkt = NULL; 1145 retryTid = -1; 1146 cacheBlocked = false; 1147 } 1148 } else { 1149 assert(retryTid == -1); 1150 // Access has been squashed since it was sent out. Just clear 1151 // the cache being blocked. 1152 cacheBlocked = false; 1153 } 1154} 1155 1156/////////////////////////////////////// 1157// // 1158// SMT FETCH POLICY MAINTAINED HERE // 1159// // 1160/////////////////////////////////////// 1161template<class Impl> 1162int 1163DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1164{ 1165 if (numThreads > 1) { 1166 switch (fetch_priority) { 1167 1168 case SingleThread: 1169 return 0; 1170 1171 case RoundRobin: 1172 return roundRobin(); 1173 1174 case IQ: 1175 return iqCount(); 1176 1177 case LSQ: 1178 return lsqCount(); 1179 1180 case Branch: 1181 return branchCount(); 1182 1183 default: 1184 return -1; 1185 } 1186 } else { 1187 int tid = *((*activeThreads).begin()); 1188 1189 if (fetchStatus[tid] == Running || 1190 fetchStatus[tid] == IcacheAccessComplete || 1191 fetchStatus[tid] == Idle) { 1192 return tid; 1193 } else { 1194 return -1; 1195 } 1196 } 1197 1198} 1199 1200 1201template<class Impl> 1202int 1203DefaultFetch<Impl>::roundRobin() 1204{ 1205 list<unsigned>::iterator pri_iter = priorityList.begin(); 1206 list<unsigned>::iterator end = priorityList.end(); 1207 1208 int high_pri; 1209 1210 while (pri_iter != end) { 1211 high_pri = *pri_iter; 1212 1213 assert(high_pri <= numThreads); 1214 1215 if (fetchStatus[high_pri] == Running || 1216 fetchStatus[high_pri] == IcacheAccessComplete || 1217 fetchStatus[high_pri] == Idle) { 1218 1219 priorityList.erase(pri_iter); 1220 priorityList.push_back(high_pri); 1221 1222 return high_pri; 1223 } 1224 1225 pri_iter++; 1226 } 1227 1228 return -1; 1229} 1230 1231template<class Impl> 1232int 1233DefaultFetch<Impl>::iqCount() 1234{ 1235 priority_queue<unsigned> PQ; 1236 1237 list<unsigned>::iterator threads = (*activeThreads).begin(); 1238 1239 while (threads != (*activeThreads).end()) { 1240 unsigned tid = *threads++; 1241 1242 PQ.push(fromIEW->iewInfo[tid].iqCount); 1243 } 1244 1245 while (!PQ.empty()) { 1246 1247 unsigned high_pri = PQ.top(); 1248 1249 if (fetchStatus[high_pri] == Running || 1250 fetchStatus[high_pri] == IcacheAccessComplete || 1251 fetchStatus[high_pri] == Idle) 1252 return high_pri; 1253 else 1254 PQ.pop(); 1255 1256 } 1257 1258 return -1; 1259} 1260 1261template<class Impl> 1262int 1263DefaultFetch<Impl>::lsqCount() 1264{ 1265 priority_queue<unsigned> PQ; 1266 1267 1268 list<unsigned>::iterator threads = (*activeThreads).begin(); 1269 1270 while (threads != (*activeThreads).end()) { 1271 unsigned tid = *threads++; 1272 1273 PQ.push(fromIEW->iewInfo[tid].ldstqCount); 1274 } 1275 1276 while (!PQ.empty()) { 1277 1278 unsigned high_pri = PQ.top(); 1279 1280 if (fetchStatus[high_pri] == Running || 1281 fetchStatus[high_pri] == IcacheAccessComplete || 1282 fetchStatus[high_pri] == Idle) 1283 return high_pri; 1284 else 1285 PQ.pop(); 1286 1287 } 1288 1289 return -1; 1290} 1291 1292template<class Impl> 1293int 1294DefaultFetch<Impl>::branchCount() 1295{ 1296 list<unsigned>::iterator threads = (*activeThreads).begin(); 1297 panic("Branch Count Fetch policy unimplemented\n"); 1298 return *threads; 1299} 1300