290#if USE_CHECKER 291 if (cpu->checker) { 292 cpu->checker->setIcachePort(icachePort); 293 } 294#endif 295 296 // Schedule fetch to get the correct PC from the CPU 297 // scheduleFetchStartupEvent(1); 298 299 // Fetch needs to start fetching instructions at the very beginning, 300 // so it must start up in active state. 301 switchToActive(); 302} 303 304template<class Impl> 305void 306DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 307{ 308 DPRINTF(Fetch, "Setting the time buffer pointer.\n"); 309 timeBuffer = time_buffer; 310 311 // Create wires to get information from proper places in time buffer. 312 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 313 fromRename = timeBuffer->getWire(-renameToFetchDelay); 314 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 315 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 316} 317 318template<class Impl> 319void 320DefaultFetch<Impl>::setActiveThreads(std::list<unsigned> *at_ptr) 321{ 322 DPRINTF(Fetch, "Setting active threads list pointer.\n"); 323 activeThreads = at_ptr; 324} 325 326template<class Impl> 327void 328DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 329{ 330 DPRINTF(Fetch, "Setting the fetch queue pointer.\n"); 331 fetchQueue = fq_ptr; 332 333 // Create wire to write information to proper place in fetch queue. 334 toDecode = fetchQueue->getWire(0); 335} 336 337template<class Impl> 338void 339DefaultFetch<Impl>::initStage() 340{ 341 // Setup PC and nextPC with initial state. 342 for (int tid = 0; tid < numThreads; tid++) { 343 PC[tid] = cpu->readPC(tid); 344 nextPC[tid] = cpu->readNextPC(tid); 345#if ISA_HAS_DELAY_SLOT 346 nextNPC[tid] = cpu->readNextNPC(tid); 347#endif 348 } 349} 350 351template<class Impl> 352void 353DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 354{ 355 unsigned tid = pkt->req->getThreadNum(); 356 357 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid); 358 359 // Only change the status if it's still waiting on the icache access 360 // to return. 361 if (fetchStatus[tid] != IcacheWaitResponse || 362 pkt->req != memReq[tid] || 363 isSwitchedOut()) { 364 ++fetchIcacheSquashes; 365 delete pkt->req; 366 delete pkt; 367 return; 368 } 369 370 memcpy(cacheData[tid], pkt->getPtr<uint8_t *>(), cacheBlkSize); 371 cacheDataValid[tid] = true; 372 373 if (!drainPending) { 374 // Wake up the CPU (if it went to sleep and was waiting on 375 // this completion event). 376 cpu->wakeCPU(); 377 378 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 379 tid); 380 381 switchToActive(); 382 } 383 384 // Only switch to IcacheAccessComplete if we're not stalled as well. 385 if (checkStall(tid)) { 386 fetchStatus[tid] = Blocked; 387 } else { 388 fetchStatus[tid] = IcacheAccessComplete; 389 } 390 391 // Reset the mem req to NULL. 392 delete pkt->req; 393 delete pkt; 394 memReq[tid] = NULL; 395} 396 397template <class Impl> 398bool 399DefaultFetch<Impl>::drain() 400{ 401 // Fetch is ready to drain at any time. 402 cpu->signalDrained(); 403 drainPending = true; 404 return true; 405} 406 407template <class Impl> 408void 409DefaultFetch<Impl>::resume() 410{ 411 drainPending = false; 412} 413 414template <class Impl> 415void 416DefaultFetch<Impl>::switchOut() 417{ 418 switchedOut = true; 419 // Branch predictor needs to have its state cleared. 420 branchPred.switchOut(); 421} 422 423template <class Impl> 424void 425DefaultFetch<Impl>::takeOverFrom() 426{ 427 // Reset all state 428 for (int i = 0; i < Impl::MaxThreads; ++i) { 429 stalls[i].decode = 0; 430 stalls[i].rename = 0; 431 stalls[i].iew = 0; 432 stalls[i].commit = 0; 433 PC[i] = cpu->readPC(i); 434 nextPC[i] = cpu->readNextPC(i); 435#if ISA_HAS_DELAY_SLOT 436 nextNPC[i] = cpu->readNextNPC(i); 437 delaySlotInfo[i].branchSeqNum = -1; 438 delaySlotInfo[i].numInsts = 0; 439 delaySlotInfo[i].targetAddr = 0; 440 delaySlotInfo[i].targetReady = false; 441#endif 442 fetchStatus[i] = Running; 443 } 444 numInst = 0; 445 wroteToTimeBuffer = false; 446 _status = Inactive; 447 switchedOut = false; 448 interruptPending = false; 449 branchPred.takeOverFrom(); 450} 451 452template <class Impl> 453void 454DefaultFetch<Impl>::wakeFromQuiesce() 455{ 456 DPRINTF(Fetch, "Waking up from quiesce\n"); 457 // Hopefully this is safe 458 // @todo: Allow other threads to wake from quiesce. 459 fetchStatus[0] = Running; 460} 461 462template <class Impl> 463inline void 464DefaultFetch<Impl>::switchToActive() 465{ 466 if (_status == Inactive) { 467 DPRINTF(Activity, "Activating stage.\n"); 468 469 cpu->activateStage(O3CPU::FetchIdx); 470 471 _status = Active; 472 } 473} 474 475template <class Impl> 476inline void 477DefaultFetch<Impl>::switchToInactive() 478{ 479 if (_status == Active) { 480 DPRINTF(Activity, "Deactivating stage.\n"); 481 482 cpu->deactivateStage(O3CPU::FetchIdx); 483 484 _status = Inactive; 485 } 486} 487 488template <class Impl> 489bool 490DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC, 491 Addr &next_NPC) 492{ 493 // Do branch prediction check here. 494 // A bit of a misnomer...next_PC is actually the current PC until 495 // this function updates it. 496 bool predict_taken; 497 498 if (!inst->isControl()) { 499#if ISA_HAS_DELAY_SLOT 500 Addr cur_PC = next_PC; 501 next_PC = cur_PC + instSize; //next_NPC; 502 next_NPC = cur_PC + (2 * instSize);//next_NPC + instSize; 503 inst->setPredTarg(next_NPC); 504#else 505 next_PC = next_PC + instSize; 506 inst->setPredTarg(next_PC); 507#endif 508 return false; 509 } 510 511 int tid = inst->threadNumber; 512#if ISA_HAS_DELAY_SLOT 513 Addr pred_PC = next_PC; 514 predict_taken = branchPred.predict(inst, pred_PC, tid); 515 516 if (predict_taken) { 517 DPRINTF(Fetch, "[tid:%i]: Branch predicted to be true.\n", tid); 518 } else { 519 DPRINTF(Fetch, "[tid:%i]: Branch predicted to be false.\n", tid); 520 } 521 522 if (predict_taken) { 523 next_PC = next_NPC; 524 next_NPC = pred_PC; 525 526 // Update delay slot info 527 ++delaySlotInfo[tid].numInsts; 528 delaySlotInfo[tid].targetAddr = pred_PC; 529 DPRINTF(Fetch, "[tid:%i]: %i delay slot inst(s) to process.\n", tid, 530 delaySlotInfo[tid].numInsts); 531 } else { // !predict_taken 532 if (inst->isCondDelaySlot()) { 533 next_PC = pred_PC; 534 // The delay slot is skipped here if there is on 535 // prediction 536 } else { 537 next_PC = next_NPC; 538 // No need to declare a delay slot here since 539 // there is no for the pred. target to jump 540 } 541 542 next_NPC = next_NPC + instSize; 543 } 544#else 545 predict_taken = branchPred.predict(inst, next_PC, tid); 546#endif 547 548 ++fetchedBranches; 549 550 if (predict_taken) { 551 ++predictedBranches; 552 } 553 554 return predict_taken; 555} 556 557template <class Impl> 558bool 559DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid) 560{ 561 Fault fault = NoFault; 562 563 //AlphaDep 564 if (cacheBlocked) { 565 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, cache blocked\n", 566 tid); 567 return false; 568 } else if (isSwitchedOut()) { 569 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, switched out\n", 570 tid); 571 return false; 572 } else if (interruptPending && !(fetch_PC & 0x3)) { 573 // Hold off fetch from getting new instructions when: 574 // Cache is blocked, or 575 // while an interrupt is pending and we're not in PAL mode, or 576 // fetch is switched out. 577 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, interrupt pending\n", 578 tid); 579 return false; 580 } 581 582 // Align the fetch PC so it's at the start of a cache block. 583 Addr block_PC = icacheBlockAlignPC(fetch_PC); 584 585 // If we've already got the block, no need to try to fetch it again. 586 if (cacheDataValid[tid] && block_PC == cacheDataPC[tid]) { 587 return true; 588 } 589 590 // Setup the memReq to do a read of the first instruction's address. 591 // Set the appropriate read size and flags as well. 592 // Build request here. 593 RequestPtr mem_req = new Request(tid, block_PC, cacheBlkSize, 0, 594 fetch_PC, cpu->readCpuId(), tid); 595 596 memReq[tid] = mem_req; 597 598 // Translate the instruction request. 599 fault = cpu->translateInstReq(mem_req, cpu->thread[tid]); 600 601 // In the case of faults, the fetch stage may need to stall and wait 602 // for the ITB miss to be handled. 603 604 // If translation was successful, attempt to read the first 605 // instruction. 606 if (fault == NoFault) { 607#if 0 608 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || 609 memReq[tid]->isUncacheable()) { 610 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " 611 "misspeculating path)!", 612 memReq[tid]->paddr); 613 ret_fault = TheISA::genMachineCheckFault(); 614 return false; 615 } 616#endif 617 618 // Build packet here. 619 PacketPtr data_pkt = new Packet(mem_req, 620 Packet::ReadReq, Packet::Broadcast); 621 data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]); 622 623 cacheDataPC[tid] = block_PC; 624 cacheDataValid[tid] = false; 625 626 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 627 628 fetchedCacheLines++; 629 630 // Now do the timing access to see whether or not the instruction 631 // exists within the cache. 632 if (!icachePort->sendTiming(data_pkt)) { 633 if (data_pkt->result == Packet::BadAddress) { 634 fault = TheISA::genMachineCheckFault(); 635 delete mem_req; 636 memReq[tid] = NULL; 637 } 638 assert(retryPkt == NULL); 639 assert(retryTid == -1); 640 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 641 fetchStatus[tid] = IcacheWaitRetry; 642 retryPkt = data_pkt; 643 retryTid = tid; 644 cacheBlocked = true; 645 return false; 646 } 647 648 DPRINTF(Fetch, "[tid:%i]: Doing cache access.\n", tid); 649 650 lastIcacheStall[tid] = curTick; 651 652 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 653 "response.\n", tid); 654 655 fetchStatus[tid] = IcacheWaitResponse; 656 } else { 657 delete mem_req; 658 memReq[tid] = NULL; 659 } 660 661 ret_fault = fault; 662 return true; 663} 664 665template <class Impl> 666inline void 667DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid) 668{ 669 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x.\n", 670 tid, new_PC); 671 672 PC[tid] = new_PC; 673 nextPC[tid] = new_PC + instSize; 674 nextNPC[tid] = new_PC + (2 * instSize); 675 676 // Clear the icache miss if it's outstanding. 677 if (fetchStatus[tid] == IcacheWaitResponse) { 678 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 679 tid); 680 memReq[tid] = NULL; 681 } 682 683 // Get rid of the retrying packet if it was from this thread. 684 if (retryTid == tid) { 685 assert(cacheBlocked); 686 cacheBlocked = false; 687 retryTid = -1; 688 delete retryPkt->req; 689 delete retryPkt; 690 retryPkt = NULL; 691 } 692 693 fetchStatus[tid] = Squashing; 694 695 ++fetchSquashCycles; 696} 697 698template<class Impl> 699void 700DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, 701 const InstSeqNum &seq_num, 702 unsigned tid) 703{ 704 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid); 705 706 doSquash(new_PC, tid); 707 708#if ISA_HAS_DELAY_SLOT 709 if (seq_num <= delaySlotInfo[tid].branchSeqNum) { 710 delaySlotInfo[tid].numInsts = 0; 711 delaySlotInfo[tid].targetAddr = 0; 712 delaySlotInfo[tid].targetReady = false; 713 } 714#endif 715 716 // Tell the CPU to remove any instructions that are in flight between 717 // fetch and decode. 718 cpu->removeInstsUntil(seq_num, tid); 719} 720 721template<class Impl> 722bool 723DefaultFetch<Impl>::checkStall(unsigned tid) const 724{ 725 bool ret_val = false; 726 727 if (cpu->contextSwitch) { 728 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 729 ret_val = true; 730 } else if (stalls[tid].decode) { 731 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 732 ret_val = true; 733 } else if (stalls[tid].rename) { 734 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 735 ret_val = true; 736 } else if (stalls[tid].iew) { 737 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 738 ret_val = true; 739 } else if (stalls[tid].commit) { 740 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 741 ret_val = true; 742 } 743 744 return ret_val; 745} 746 747template<class Impl> 748typename DefaultFetch<Impl>::FetchStatus 749DefaultFetch<Impl>::updateFetchStatus() 750{ 751 //Check Running 752 std::list<unsigned>::iterator threads = (*activeThreads).begin(); 753 754 while (threads != (*activeThreads).end()) { 755 756 unsigned tid = *threads++; 757 758 if (fetchStatus[tid] == Running || 759 fetchStatus[tid] == Squashing || 760 fetchStatus[tid] == IcacheAccessComplete) { 761 762 if (_status == Inactive) { 763 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 764 765 if (fetchStatus[tid] == IcacheAccessComplete) { 766 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 767 "completion\n",tid); 768 } 769 770 cpu->activateStage(O3CPU::FetchIdx); 771 } 772 773 return Active; 774 } 775 } 776 777 // Stage is switching from active to inactive, notify CPU of it. 778 if (_status == Active) { 779 DPRINTF(Activity, "Deactivating stage.\n"); 780 781 cpu->deactivateStage(O3CPU::FetchIdx); 782 } 783 784 return Inactive; 785} 786 787template <class Impl> 788void 789DefaultFetch<Impl>::squash(const Addr &new_PC, const InstSeqNum &seq_num, 790 bool squash_delay_slot, unsigned tid) 791{ 792 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid); 793 794 doSquash(new_PC, tid); 795 796#if ISA_HAS_DELAY_SLOT 797 if (seq_num <= delaySlotInfo[tid].branchSeqNum) { 798 delaySlotInfo[tid].numInsts = 0; 799 delaySlotInfo[tid].targetAddr = 0; 800 delaySlotInfo[tid].targetReady = false; 801 } 802 803 // Tell the CPU to remove any instructions that are not in the ROB. 804 cpu->removeInstsNotInROB(tid, squash_delay_slot, seq_num); 805#else 806 // Tell the CPU to remove any instructions that are not in the ROB. 807 cpu->removeInstsNotInROB(tid, true, 0); 808#endif 809} 810 811template <class Impl> 812void 813DefaultFetch<Impl>::tick() 814{ 815 std::list<unsigned>::iterator threads = (*activeThreads).begin(); 816 bool status_change = false; 817 818 wroteToTimeBuffer = false; 819 820 while (threads != (*activeThreads).end()) { 821 unsigned tid = *threads++; 822 823 // Check the signals for each thread to determine the proper status 824 // for each thread. 825 bool updated_status = checkSignalsAndUpdate(tid); 826 status_change = status_change || updated_status; 827 } 828 829 DPRINTF(Fetch, "Running stage.\n"); 830 831 // Reset the number of the instruction we're fetching. 832 numInst = 0; 833 834#if FULL_SYSTEM 835 if (fromCommit->commitInfo[0].interruptPending) { 836 interruptPending = true; 837 } 838 839 if (fromCommit->commitInfo[0].clearInterrupt) { 840 interruptPending = false; 841 } 842#endif 843 844 for (threadFetched = 0; threadFetched < numFetchingThreads; 845 threadFetched++) { 846 // Fetch each of the actively fetching threads. 847 fetch(status_change); 848 } 849 850 // Record number of instructions fetched this cycle for distribution. 851 fetchNisnDist.sample(numInst); 852 853 if (status_change) { 854 // Change the fetch stage status if there was a status change. 855 _status = updateFetchStatus(); 856 } 857 858 // If there was activity this cycle, inform the CPU of it. 859 if (wroteToTimeBuffer || cpu->contextSwitch) { 860 DPRINTF(Activity, "Activity this cycle.\n"); 861 862 cpu->activityThisCycle(); 863 } 864} 865 866template <class Impl> 867bool 868DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) 869{ 870 // Update the per thread stall statuses. 871 if (fromDecode->decodeBlock[tid]) { 872 stalls[tid].decode = true; 873 } 874 875 if (fromDecode->decodeUnblock[tid]) { 876 assert(stalls[tid].decode); 877 assert(!fromDecode->decodeBlock[tid]); 878 stalls[tid].decode = false; 879 } 880 881 if (fromRename->renameBlock[tid]) { 882 stalls[tid].rename = true; 883 } 884 885 if (fromRename->renameUnblock[tid]) { 886 assert(stalls[tid].rename); 887 assert(!fromRename->renameBlock[tid]); 888 stalls[tid].rename = false; 889 } 890 891 if (fromIEW->iewBlock[tid]) { 892 stalls[tid].iew = true; 893 } 894 895 if (fromIEW->iewUnblock[tid]) { 896 assert(stalls[tid].iew); 897 assert(!fromIEW->iewBlock[tid]); 898 stalls[tid].iew = false; 899 } 900 901 if (fromCommit->commitBlock[tid]) { 902 stalls[tid].commit = true; 903 } 904 905 if (fromCommit->commitUnblock[tid]) { 906 assert(stalls[tid].commit); 907 assert(!fromCommit->commitBlock[tid]); 908 stalls[tid].commit = false; 909 } 910 911 // Check squash signals from commit. 912 if (fromCommit->commitInfo[tid].squash) { 913 914 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 915 "from commit.\n",tid); 916 917#if ISA_HAS_DELAY_SLOT 918 InstSeqNum doneSeqNum = fromCommit->commitInfo[tid].bdelayDoneSeqNum; 919#else 920 InstSeqNum doneSeqNum = fromCommit->commitInfo[tid].doneSeqNum; 921#endif 922 // In any case, squash. 923 squash(fromCommit->commitInfo[tid].nextPC, 924 doneSeqNum, 925 fromCommit->commitInfo[tid].squashDelaySlot, 926 tid); 927 928 // Also check if there's a mispredict that happened. 929 if (fromCommit->commitInfo[tid].branchMispredict) { 930 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 931 fromCommit->commitInfo[tid].nextPC, 932 fromCommit->commitInfo[tid].branchTaken, 933 tid); 934 } else { 935 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 936 tid); 937 } 938 939 return true; 940 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 941 // Update the branch predictor if it wasn't a squashed instruction 942 // that was broadcasted. 943 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 944 } 945 946 // Check ROB squash signals from commit. 947 if (fromCommit->commitInfo[tid].robSquashing) { 948 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 949 950 // Continue to squash. 951 fetchStatus[tid] = Squashing; 952 953 return true; 954 } 955 956 // Check squash signals from decode. 957 if (fromDecode->decodeInfo[tid].squash) { 958 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 959 "from decode.\n",tid); 960 961 // Update the branch predictor. 962 if (fromDecode->decodeInfo[tid].branchMispredict) { 963 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 964 fromDecode->decodeInfo[tid].nextPC, 965 fromDecode->decodeInfo[tid].branchTaken, 966 tid); 967 } else { 968 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 969 tid); 970 } 971 972 if (fetchStatus[tid] != Squashing) { 973 974#if ISA_HAS_DELAY_SLOT 975 InstSeqNum doneSeqNum = fromDecode->decodeInfo[tid].bdelayDoneSeqNum; 976#else 977 InstSeqNum doneSeqNum = fromDecode->decodeInfo[tid].doneSeqNum; 978#endif 979 // Squash unless we're already squashing 980 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 981 doneSeqNum, 982 tid); 983 984 return true; 985 } 986 } 987 988 if (checkStall(tid) && 989 fetchStatus[tid] != IcacheWaitResponse && 990 fetchStatus[tid] != IcacheWaitRetry) { 991 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 992 993 fetchStatus[tid] = Blocked; 994 995 return true; 996 } 997 998 if (fetchStatus[tid] == Blocked || 999 fetchStatus[tid] == Squashing) { 1000 // Switch status to running if fetch isn't being told to block or 1001 // squash this cycle. 1002 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 1003 tid); 1004 1005 fetchStatus[tid] = Running; 1006 1007 return true; 1008 } 1009 1010 // If we've reached this point, we have not gotten any signals that 1011 // cause fetch to change its status. Fetch remains the same as before. 1012 return false; 1013} 1014 1015template<class Impl> 1016void 1017DefaultFetch<Impl>::fetch(bool &status_change) 1018{ 1019 ////////////////////////////////////////// 1020 // Start actual fetch 1021 ////////////////////////////////////////// 1022 int tid = getFetchingThread(fetchPolicy); 1023 1024 if (tid == -1 || drainPending) { 1025 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 1026 1027 // Breaks looping condition in tick() 1028 threadFetched = numFetchingThreads; 1029 return; 1030 } 1031 1032 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 1033 1034 // The current PC. 1035 Addr &fetch_PC = PC[tid]; 1036 1037 // Fault code for memory access. 1038 Fault fault = NoFault; 1039 1040 // If returning from the delay of a cache miss, then update the status 1041 // to running, otherwise do the cache access. Possibly move this up 1042 // to tick() function. 1043 if (fetchStatus[tid] == IcacheAccessComplete) { 1044 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", 1045 tid); 1046 1047 fetchStatus[tid] = Running; 1048 status_change = true; 1049 } else if (fetchStatus[tid] == Running) { 1050 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 1051 "instruction, starting at PC %08p.\n", 1052 tid, fetch_PC); 1053 1054 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid); 1055 if (!fetch_success) { 1056 if (cacheBlocked) { 1057 ++icacheStallCycles; 1058 } else { 1059 ++fetchMiscStallCycles; 1060 } 1061 return; 1062 } 1063 } else { 1064 if (fetchStatus[tid] == Idle) { 1065 ++fetchIdleCycles; 1066 DPRINTF(Fetch, "[tid:%i]: Fetch is idle!\n", tid); 1067 } else if (fetchStatus[tid] == Blocked) { 1068 ++fetchBlockedCycles; 1069 DPRINTF(Fetch, "[tid:%i]: Fetch is blocked!\n", tid); 1070 } else if (fetchStatus[tid] == Squashing) { 1071 ++fetchSquashCycles; 1072 DPRINTF(Fetch, "[tid:%i]: Fetch is squashing!\n", tid); 1073 } else if (fetchStatus[tid] == IcacheWaitResponse) { 1074 ++icacheStallCycles; 1075 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting cache response!\n", tid); 1076 } 1077 1078 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so 1079 // fetch should do nothing. 1080 return; 1081 } 1082 1083 ++fetchCycles; 1084 1085 // If we had a stall due to an icache miss, then return. 1086 if (fetchStatus[tid] == IcacheWaitResponse) { 1087 ++icacheStallCycles; 1088 status_change = true; 1089 return; 1090 } 1091 1092 Addr next_PC = fetch_PC; 1093 Addr next_NPC = next_PC + instSize; 1094 InstSeqNum inst_seq; 1095 MachInst inst; 1096 ExtMachInst ext_inst; 1097 // @todo: Fix this hack. 1098 unsigned offset = (fetch_PC & cacheBlkMask) & ~3; 1099 1100 if (fault == NoFault) { 1101 // If the read of the first instruction was successful, then grab the 1102 // instructions from the rest of the cache line and put them into the 1103 // queue heading to decode. 1104 1105 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 1106 "decode.\n",tid); 1107 1108 // Need to keep track of whether or not a predicted branch 1109 // ended this fetch block. 1110 bool predicted_branch = false; 1111 1112 // Need to keep track of whether or not a delay slot 1113 // instruction has been fetched 1114 1115 for (; 1116 offset < cacheBlkSize && 1117 numInst < fetchWidth && 1118 (!predicted_branch || delaySlotInfo[tid].numInsts > 0); 1119 ++numInst) { 1120 1121 // Get a sequence number. 1122 inst_seq = cpu->getAndIncrementInstSeq(); 1123 1124 // Make sure this is a valid index. 1125 assert(offset <= cacheBlkSize - instSize); 1126 1127 // Get the instruction from the array of the cache line. 1128 inst = TheISA::gtoh(*reinterpret_cast<TheISA::MachInst *> 1129 (&cacheData[tid][offset])); 1130 1131#if THE_ISA == ALPHA_ISA 1132 ext_inst = TheISA::makeExtMI(inst, fetch_PC); 1133#elif THE_ISA == SPARC_ISA 1134 ext_inst = TheISA::makeExtMI(inst, cpu->thread[tid]->getTC()); 1135#endif 1136 1137 // Create a new DynInst from the instruction fetched. 1138 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1139 next_PC, 1140 inst_seq, cpu); 1141 instruction->setTid(tid); 1142 1143 instruction->setASID(tid); 1144 1145 instruction->setThreadState(cpu->thread[tid]); 1146 1147 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created " 1148 "[sn:%lli]\n", 1149 tid, instruction->readPC(), inst_seq); 1150 1151 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", 1152 tid, instruction->staticInst->disassemble(fetch_PC)); 1153 1154 instruction->traceData = 1155 Trace::getInstRecord(curTick, cpu->tcBase(tid), 1156 instruction->staticInst, 1157 instruction->readPC()); 1158 1159 predicted_branch = lookupAndUpdateNextPC(instruction, next_PC, 1160 next_NPC); 1161 1162 // Add instruction to the CPU's list of instructions. 1163 instruction->setInstListIt(cpu->addInst(instruction)); 1164 1165 // Write the instruction to the first slot in the queue 1166 // that heads to decode. 1167 toDecode->insts[numInst] = instruction; 1168 1169 toDecode->size++; 1170 1171 // Increment stat of fetched instructions. 1172 ++fetchedInsts; 1173 1174 // Move to the next instruction, unless we have a branch. 1175 fetch_PC = next_PC; 1176 1177 if (instruction->isQuiesce()) { 1178 DPRINTF(Fetch, "Quiesce instruction encountered, halting fetch!", 1179 curTick); 1180 fetchStatus[tid] = QuiescePending; 1181 ++numInst; 1182 status_change = true; 1183 break; 1184 } 1185 1186 offset += instSize; 1187 1188#if ISA_HAS_DELAY_SLOT 1189 if (predicted_branch) { 1190 delaySlotInfo[tid].branchSeqNum = inst_seq; 1191 1192 DPRINTF(Fetch, "[tid:%i]: Delay slot branch set to [sn:%i]\n", 1193 tid, inst_seq); 1194 continue; 1195 } else if (delaySlotInfo[tid].numInsts > 0) { 1196 --delaySlotInfo[tid].numInsts; 1197 1198 // It's OK to set PC to target of branch 1199 if (delaySlotInfo[tid].numInsts == 0) { 1200 delaySlotInfo[tid].targetReady = true; 1201 1202 // Break the looping condition 1203 predicted_branch = true; 1204 } 1205 1206 DPRINTF(Fetch, "[tid:%i]: %i delay slot inst(s) left to" 1207 " process.\n", tid, delaySlotInfo[tid].numInsts); 1208 } 1209#endif 1210 } 1211 1212 if (offset >= cacheBlkSize) { 1213 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache " 1214 "block.\n", tid); 1215 } else if (numInst >= fetchWidth) { 1216 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth " 1217 "for this cycle.\n", tid); 1218 } else if (predicted_branch && delaySlotInfo[tid].numInsts <= 0) { 1219 DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch " 1220 "instruction encountered.\n", tid); 1221 } 1222 } 1223 1224 if (numInst > 0) { 1225 wroteToTimeBuffer = true; 1226 } 1227 1228 // Now that fetching is completed, update the PC to signify what the next 1229 // cycle will be. 1230 if (fault == NoFault) { 1231#if ISA_HAS_DELAY_SLOT 1232 if (delaySlotInfo[tid].targetReady && 1233 delaySlotInfo[tid].numInsts == 0) { 1234 // Set PC to target 1235 PC[tid] = delaySlotInfo[tid].targetAddr; //next_PC 1236 nextPC[tid] = next_PC + instSize; //next_NPC 1237 nextNPC[tid] = next_PC + (2 * instSize); 1238 1239 delaySlotInfo[tid].targetReady = false; 1240 } else { 1241 PC[tid] = next_PC; 1242 nextPC[tid] = next_NPC; 1243 nextNPC[tid] = next_NPC + instSize; 1244 } 1245 1246 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n", tid, PC[tid]); 1247#else 1248 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n",tid, next_PC); 1249 PC[tid] = next_PC; 1250 nextPC[tid] = next_PC + instSize; 1251#endif 1252 } else { 1253 // We shouldn't be in an icache miss and also have a fault (an ITB 1254 // miss) 1255 if (fetchStatus[tid] == IcacheWaitResponse) { 1256 panic("Fetch should have exited prior to this!"); 1257 } 1258 1259 // Send the fault to commit. This thread will not do anything 1260 // until commit handles the fault. The only other way it can 1261 // wake up is if a squash comes along and changes the PC. 1262#if FULL_SYSTEM 1263 assert(numInst != fetchWidth); 1264 // Get a sequence number. 1265 inst_seq = cpu->getAndIncrementInstSeq(); 1266 // We will use a nop in order to carry the fault. 1267 ext_inst = TheISA::NoopMachInst; 1268 1269 // Create a new DynInst from the dummy nop. 1270 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1271 next_PC, 1272 inst_seq, cpu); 1273 instruction->setPredTarg(next_PC + instSize); 1274 instruction->setTid(tid); 1275 1276 instruction->setASID(tid); 1277 1278 instruction->setThreadState(cpu->thread[tid]); 1279 1280 instruction->traceData = NULL; 1281 1282 instruction->setInstListIt(cpu->addInst(instruction)); 1283 1284 instruction->fault = fault; 1285 1286 toDecode->insts[numInst] = instruction; 1287 toDecode->size++; 1288 1289 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid); 1290 1291 fetchStatus[tid] = TrapPending; 1292 status_change = true; 1293#else // !FULL_SYSTEM 1294 fetchStatus[tid] = TrapPending; 1295 status_change = true; 1296 1297#endif // FULL_SYSTEM 1298 DPRINTF(Fetch, "[tid:%i]: fault (%s) detected @ PC %08p", 1299 tid, fault->name(), PC[tid]); 1300 } 1301} 1302 1303template<class Impl> 1304void 1305DefaultFetch<Impl>::recvRetry() 1306{ 1307 if (retryPkt != NULL) { 1308 assert(cacheBlocked); 1309 assert(retryTid != -1); 1310 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1311 1312 if (icachePort->sendTiming(retryPkt)) { 1313 fetchStatus[retryTid] = IcacheWaitResponse; 1314 retryPkt = NULL; 1315 retryTid = -1; 1316 cacheBlocked = false; 1317 } 1318 } else { 1319 assert(retryTid == -1); 1320 // Access has been squashed since it was sent out. Just clear 1321 // the cache being blocked. 1322 cacheBlocked = false; 1323 } 1324} 1325 1326/////////////////////////////////////// 1327// // 1328// SMT FETCH POLICY MAINTAINED HERE // 1329// // 1330/////////////////////////////////////// 1331template<class Impl> 1332int 1333DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1334{ 1335 if (numThreads > 1) { 1336 switch (fetch_priority) { 1337 1338 case SingleThread: 1339 return 0; 1340 1341 case RoundRobin: 1342 return roundRobin(); 1343 1344 case IQ: 1345 return iqCount(); 1346 1347 case LSQ: 1348 return lsqCount(); 1349 1350 case Branch: 1351 return branchCount(); 1352 1353 default: 1354 return -1; 1355 } 1356 } else { 1357 int tid = *((*activeThreads).begin()); 1358 1359 if (fetchStatus[tid] == Running || 1360 fetchStatus[tid] == IcacheAccessComplete || 1361 fetchStatus[tid] == Idle) { 1362 return tid; 1363 } else { 1364 return -1; 1365 } 1366 } 1367 1368} 1369 1370 1371template<class Impl> 1372int 1373DefaultFetch<Impl>::roundRobin() 1374{ 1375 std::list<unsigned>::iterator pri_iter = priorityList.begin(); 1376 std::list<unsigned>::iterator end = priorityList.end(); 1377 1378 int high_pri; 1379 1380 while (pri_iter != end) { 1381 high_pri = *pri_iter; 1382 1383 assert(high_pri <= numThreads); 1384 1385 if (fetchStatus[high_pri] == Running || 1386 fetchStatus[high_pri] == IcacheAccessComplete || 1387 fetchStatus[high_pri] == Idle) { 1388 1389 priorityList.erase(pri_iter); 1390 priorityList.push_back(high_pri); 1391 1392 return high_pri; 1393 } 1394 1395 pri_iter++; 1396 } 1397 1398 return -1; 1399} 1400 1401template<class Impl> 1402int 1403DefaultFetch<Impl>::iqCount() 1404{ 1405 std::priority_queue<unsigned> PQ; 1406 1407 std::list<unsigned>::iterator threads = (*activeThreads).begin(); 1408 1409 while (threads != (*activeThreads).end()) { 1410 unsigned tid = *threads++; 1411 1412 PQ.push(fromIEW->iewInfo[tid].iqCount); 1413 } 1414 1415 while (!PQ.empty()) { 1416 1417 unsigned high_pri = PQ.top(); 1418 1419 if (fetchStatus[high_pri] == Running || 1420 fetchStatus[high_pri] == IcacheAccessComplete || 1421 fetchStatus[high_pri] == Idle) 1422 return high_pri; 1423 else 1424 PQ.pop(); 1425 1426 } 1427 1428 return -1; 1429} 1430 1431template<class Impl> 1432int 1433DefaultFetch<Impl>::lsqCount() 1434{ 1435 std::priority_queue<unsigned> PQ; 1436 1437 1438 std::list<unsigned>::iterator threads = (*activeThreads).begin(); 1439 1440 while (threads != (*activeThreads).end()) { 1441 unsigned tid = *threads++; 1442 1443 PQ.push(fromIEW->iewInfo[tid].ldstqCount); 1444 } 1445 1446 while (!PQ.empty()) { 1447 1448 unsigned high_pri = PQ.top(); 1449 1450 if (fetchStatus[high_pri] == Running || 1451 fetchStatus[high_pri] == IcacheAccessComplete || 1452 fetchStatus[high_pri] == Idle) 1453 return high_pri; 1454 else 1455 PQ.pop(); 1456 1457 } 1458 1459 return -1; 1460} 1461 1462template<class Impl> 1463int 1464DefaultFetch<Impl>::branchCount() 1465{ 1466 std::list<unsigned>::iterator threads = (*activeThreads).begin(); 1467 panic("Branch Count Fetch policy unimplemented\n"); 1468 return *threads; 1469}
| 297#if USE_CHECKER 298 if (cpu->checker) { 299 cpu->checker->setIcachePort(icachePort); 300 } 301#endif 302 303 // Schedule fetch to get the correct PC from the CPU 304 // scheduleFetchStartupEvent(1); 305 306 // Fetch needs to start fetching instructions at the very beginning, 307 // so it must start up in active state. 308 switchToActive(); 309} 310 311template<class Impl> 312void 313DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 314{ 315 DPRINTF(Fetch, "Setting the time buffer pointer.\n"); 316 timeBuffer = time_buffer; 317 318 // Create wires to get information from proper places in time buffer. 319 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 320 fromRename = timeBuffer->getWire(-renameToFetchDelay); 321 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 322 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 323} 324 325template<class Impl> 326void 327DefaultFetch<Impl>::setActiveThreads(std::list<unsigned> *at_ptr) 328{ 329 DPRINTF(Fetch, "Setting active threads list pointer.\n"); 330 activeThreads = at_ptr; 331} 332 333template<class Impl> 334void 335DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 336{ 337 DPRINTF(Fetch, "Setting the fetch queue pointer.\n"); 338 fetchQueue = fq_ptr; 339 340 // Create wire to write information to proper place in fetch queue. 341 toDecode = fetchQueue->getWire(0); 342} 343 344template<class Impl> 345void 346DefaultFetch<Impl>::initStage() 347{ 348 // Setup PC and nextPC with initial state. 349 for (int tid = 0; tid < numThreads; tid++) { 350 PC[tid] = cpu->readPC(tid); 351 nextPC[tid] = cpu->readNextPC(tid); 352#if ISA_HAS_DELAY_SLOT 353 nextNPC[tid] = cpu->readNextNPC(tid); 354#endif 355 } 356} 357 358template<class Impl> 359void 360DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 361{ 362 unsigned tid = pkt->req->getThreadNum(); 363 364 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid); 365 366 // Only change the status if it's still waiting on the icache access 367 // to return. 368 if (fetchStatus[tid] != IcacheWaitResponse || 369 pkt->req != memReq[tid] || 370 isSwitchedOut()) { 371 ++fetchIcacheSquashes; 372 delete pkt->req; 373 delete pkt; 374 return; 375 } 376 377 memcpy(cacheData[tid], pkt->getPtr<uint8_t *>(), cacheBlkSize); 378 cacheDataValid[tid] = true; 379 380 if (!drainPending) { 381 // Wake up the CPU (if it went to sleep and was waiting on 382 // this completion event). 383 cpu->wakeCPU(); 384 385 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 386 tid); 387 388 switchToActive(); 389 } 390 391 // Only switch to IcacheAccessComplete if we're not stalled as well. 392 if (checkStall(tid)) { 393 fetchStatus[tid] = Blocked; 394 } else { 395 fetchStatus[tid] = IcacheAccessComplete; 396 } 397 398 // Reset the mem req to NULL. 399 delete pkt->req; 400 delete pkt; 401 memReq[tid] = NULL; 402} 403 404template <class Impl> 405bool 406DefaultFetch<Impl>::drain() 407{ 408 // Fetch is ready to drain at any time. 409 cpu->signalDrained(); 410 drainPending = true; 411 return true; 412} 413 414template <class Impl> 415void 416DefaultFetch<Impl>::resume() 417{ 418 drainPending = false; 419} 420 421template <class Impl> 422void 423DefaultFetch<Impl>::switchOut() 424{ 425 switchedOut = true; 426 // Branch predictor needs to have its state cleared. 427 branchPred.switchOut(); 428} 429 430template <class Impl> 431void 432DefaultFetch<Impl>::takeOverFrom() 433{ 434 // Reset all state 435 for (int i = 0; i < Impl::MaxThreads; ++i) { 436 stalls[i].decode = 0; 437 stalls[i].rename = 0; 438 stalls[i].iew = 0; 439 stalls[i].commit = 0; 440 PC[i] = cpu->readPC(i); 441 nextPC[i] = cpu->readNextPC(i); 442#if ISA_HAS_DELAY_SLOT 443 nextNPC[i] = cpu->readNextNPC(i); 444 delaySlotInfo[i].branchSeqNum = -1; 445 delaySlotInfo[i].numInsts = 0; 446 delaySlotInfo[i].targetAddr = 0; 447 delaySlotInfo[i].targetReady = false; 448#endif 449 fetchStatus[i] = Running; 450 } 451 numInst = 0; 452 wroteToTimeBuffer = false; 453 _status = Inactive; 454 switchedOut = false; 455 interruptPending = false; 456 branchPred.takeOverFrom(); 457} 458 459template <class Impl> 460void 461DefaultFetch<Impl>::wakeFromQuiesce() 462{ 463 DPRINTF(Fetch, "Waking up from quiesce\n"); 464 // Hopefully this is safe 465 // @todo: Allow other threads to wake from quiesce. 466 fetchStatus[0] = Running; 467} 468 469template <class Impl> 470inline void 471DefaultFetch<Impl>::switchToActive() 472{ 473 if (_status == Inactive) { 474 DPRINTF(Activity, "Activating stage.\n"); 475 476 cpu->activateStage(O3CPU::FetchIdx); 477 478 _status = Active; 479 } 480} 481 482template <class Impl> 483inline void 484DefaultFetch<Impl>::switchToInactive() 485{ 486 if (_status == Active) { 487 DPRINTF(Activity, "Deactivating stage.\n"); 488 489 cpu->deactivateStage(O3CPU::FetchIdx); 490 491 _status = Inactive; 492 } 493} 494 495template <class Impl> 496bool 497DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC, 498 Addr &next_NPC) 499{ 500 // Do branch prediction check here. 501 // A bit of a misnomer...next_PC is actually the current PC until 502 // this function updates it. 503 bool predict_taken; 504 505 if (!inst->isControl()) { 506#if ISA_HAS_DELAY_SLOT 507 Addr cur_PC = next_PC; 508 next_PC = cur_PC + instSize; //next_NPC; 509 next_NPC = cur_PC + (2 * instSize);//next_NPC + instSize; 510 inst->setPredTarg(next_NPC); 511#else 512 next_PC = next_PC + instSize; 513 inst->setPredTarg(next_PC); 514#endif 515 return false; 516 } 517 518 int tid = inst->threadNumber; 519#if ISA_HAS_DELAY_SLOT 520 Addr pred_PC = next_PC; 521 predict_taken = branchPred.predict(inst, pred_PC, tid); 522 523 if (predict_taken) { 524 DPRINTF(Fetch, "[tid:%i]: Branch predicted to be true.\n", tid); 525 } else { 526 DPRINTF(Fetch, "[tid:%i]: Branch predicted to be false.\n", tid); 527 } 528 529 if (predict_taken) { 530 next_PC = next_NPC; 531 next_NPC = pred_PC; 532 533 // Update delay slot info 534 ++delaySlotInfo[tid].numInsts; 535 delaySlotInfo[tid].targetAddr = pred_PC; 536 DPRINTF(Fetch, "[tid:%i]: %i delay slot inst(s) to process.\n", tid, 537 delaySlotInfo[tid].numInsts); 538 } else { // !predict_taken 539 if (inst->isCondDelaySlot()) { 540 next_PC = pred_PC; 541 // The delay slot is skipped here if there is on 542 // prediction 543 } else { 544 next_PC = next_NPC; 545 // No need to declare a delay slot here since 546 // there is no for the pred. target to jump 547 } 548 549 next_NPC = next_NPC + instSize; 550 } 551#else 552 predict_taken = branchPred.predict(inst, next_PC, tid); 553#endif 554 555 ++fetchedBranches; 556 557 if (predict_taken) { 558 ++predictedBranches; 559 } 560 561 return predict_taken; 562} 563 564template <class Impl> 565bool 566DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid) 567{ 568 Fault fault = NoFault; 569 570 //AlphaDep 571 if (cacheBlocked) { 572 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, cache blocked\n", 573 tid); 574 return false; 575 } else if (isSwitchedOut()) { 576 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, switched out\n", 577 tid); 578 return false; 579 } else if (interruptPending && !(fetch_PC & 0x3)) { 580 // Hold off fetch from getting new instructions when: 581 // Cache is blocked, or 582 // while an interrupt is pending and we're not in PAL mode, or 583 // fetch is switched out. 584 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, interrupt pending\n", 585 tid); 586 return false; 587 } 588 589 // Align the fetch PC so it's at the start of a cache block. 590 Addr block_PC = icacheBlockAlignPC(fetch_PC); 591 592 // If we've already got the block, no need to try to fetch it again. 593 if (cacheDataValid[tid] && block_PC == cacheDataPC[tid]) { 594 return true; 595 } 596 597 // Setup the memReq to do a read of the first instruction's address. 598 // Set the appropriate read size and flags as well. 599 // Build request here. 600 RequestPtr mem_req = new Request(tid, block_PC, cacheBlkSize, 0, 601 fetch_PC, cpu->readCpuId(), tid); 602 603 memReq[tid] = mem_req; 604 605 // Translate the instruction request. 606 fault = cpu->translateInstReq(mem_req, cpu->thread[tid]); 607 608 // In the case of faults, the fetch stage may need to stall and wait 609 // for the ITB miss to be handled. 610 611 // If translation was successful, attempt to read the first 612 // instruction. 613 if (fault == NoFault) { 614#if 0 615 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || 616 memReq[tid]->isUncacheable()) { 617 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " 618 "misspeculating path)!", 619 memReq[tid]->paddr); 620 ret_fault = TheISA::genMachineCheckFault(); 621 return false; 622 } 623#endif 624 625 // Build packet here. 626 PacketPtr data_pkt = new Packet(mem_req, 627 Packet::ReadReq, Packet::Broadcast); 628 data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]); 629 630 cacheDataPC[tid] = block_PC; 631 cacheDataValid[tid] = false; 632 633 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 634 635 fetchedCacheLines++; 636 637 // Now do the timing access to see whether or not the instruction 638 // exists within the cache. 639 if (!icachePort->sendTiming(data_pkt)) { 640 if (data_pkt->result == Packet::BadAddress) { 641 fault = TheISA::genMachineCheckFault(); 642 delete mem_req; 643 memReq[tid] = NULL; 644 } 645 assert(retryPkt == NULL); 646 assert(retryTid == -1); 647 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 648 fetchStatus[tid] = IcacheWaitRetry; 649 retryPkt = data_pkt; 650 retryTid = tid; 651 cacheBlocked = true; 652 return false; 653 } 654 655 DPRINTF(Fetch, "[tid:%i]: Doing cache access.\n", tid); 656 657 lastIcacheStall[tid] = curTick; 658 659 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 660 "response.\n", tid); 661 662 fetchStatus[tid] = IcacheWaitResponse; 663 } else { 664 delete mem_req; 665 memReq[tid] = NULL; 666 } 667 668 ret_fault = fault; 669 return true; 670} 671 672template <class Impl> 673inline void 674DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid) 675{ 676 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x.\n", 677 tid, new_PC); 678 679 PC[tid] = new_PC; 680 nextPC[tid] = new_PC + instSize; 681 nextNPC[tid] = new_PC + (2 * instSize); 682 683 // Clear the icache miss if it's outstanding. 684 if (fetchStatus[tid] == IcacheWaitResponse) { 685 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 686 tid); 687 memReq[tid] = NULL; 688 } 689 690 // Get rid of the retrying packet if it was from this thread. 691 if (retryTid == tid) { 692 assert(cacheBlocked); 693 cacheBlocked = false; 694 retryTid = -1; 695 delete retryPkt->req; 696 delete retryPkt; 697 retryPkt = NULL; 698 } 699 700 fetchStatus[tid] = Squashing; 701 702 ++fetchSquashCycles; 703} 704 705template<class Impl> 706void 707DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, 708 const InstSeqNum &seq_num, 709 unsigned tid) 710{ 711 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid); 712 713 doSquash(new_PC, tid); 714 715#if ISA_HAS_DELAY_SLOT 716 if (seq_num <= delaySlotInfo[tid].branchSeqNum) { 717 delaySlotInfo[tid].numInsts = 0; 718 delaySlotInfo[tid].targetAddr = 0; 719 delaySlotInfo[tid].targetReady = false; 720 } 721#endif 722 723 // Tell the CPU to remove any instructions that are in flight between 724 // fetch and decode. 725 cpu->removeInstsUntil(seq_num, tid); 726} 727 728template<class Impl> 729bool 730DefaultFetch<Impl>::checkStall(unsigned tid) const 731{ 732 bool ret_val = false; 733 734 if (cpu->contextSwitch) { 735 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 736 ret_val = true; 737 } else if (stalls[tid].decode) { 738 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 739 ret_val = true; 740 } else if (stalls[tid].rename) { 741 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 742 ret_val = true; 743 } else if (stalls[tid].iew) { 744 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 745 ret_val = true; 746 } else if (stalls[tid].commit) { 747 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 748 ret_val = true; 749 } 750 751 return ret_val; 752} 753 754template<class Impl> 755typename DefaultFetch<Impl>::FetchStatus 756DefaultFetch<Impl>::updateFetchStatus() 757{ 758 //Check Running 759 std::list<unsigned>::iterator threads = (*activeThreads).begin(); 760 761 while (threads != (*activeThreads).end()) { 762 763 unsigned tid = *threads++; 764 765 if (fetchStatus[tid] == Running || 766 fetchStatus[tid] == Squashing || 767 fetchStatus[tid] == IcacheAccessComplete) { 768 769 if (_status == Inactive) { 770 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 771 772 if (fetchStatus[tid] == IcacheAccessComplete) { 773 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 774 "completion\n",tid); 775 } 776 777 cpu->activateStage(O3CPU::FetchIdx); 778 } 779 780 return Active; 781 } 782 } 783 784 // Stage is switching from active to inactive, notify CPU of it. 785 if (_status == Active) { 786 DPRINTF(Activity, "Deactivating stage.\n"); 787 788 cpu->deactivateStage(O3CPU::FetchIdx); 789 } 790 791 return Inactive; 792} 793 794template <class Impl> 795void 796DefaultFetch<Impl>::squash(const Addr &new_PC, const InstSeqNum &seq_num, 797 bool squash_delay_slot, unsigned tid) 798{ 799 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid); 800 801 doSquash(new_PC, tid); 802 803#if ISA_HAS_DELAY_SLOT 804 if (seq_num <= delaySlotInfo[tid].branchSeqNum) { 805 delaySlotInfo[tid].numInsts = 0; 806 delaySlotInfo[tid].targetAddr = 0; 807 delaySlotInfo[tid].targetReady = false; 808 } 809 810 // Tell the CPU to remove any instructions that are not in the ROB. 811 cpu->removeInstsNotInROB(tid, squash_delay_slot, seq_num); 812#else 813 // Tell the CPU to remove any instructions that are not in the ROB. 814 cpu->removeInstsNotInROB(tid, true, 0); 815#endif 816} 817 818template <class Impl> 819void 820DefaultFetch<Impl>::tick() 821{ 822 std::list<unsigned>::iterator threads = (*activeThreads).begin(); 823 bool status_change = false; 824 825 wroteToTimeBuffer = false; 826 827 while (threads != (*activeThreads).end()) { 828 unsigned tid = *threads++; 829 830 // Check the signals for each thread to determine the proper status 831 // for each thread. 832 bool updated_status = checkSignalsAndUpdate(tid); 833 status_change = status_change || updated_status; 834 } 835 836 DPRINTF(Fetch, "Running stage.\n"); 837 838 // Reset the number of the instruction we're fetching. 839 numInst = 0; 840 841#if FULL_SYSTEM 842 if (fromCommit->commitInfo[0].interruptPending) { 843 interruptPending = true; 844 } 845 846 if (fromCommit->commitInfo[0].clearInterrupt) { 847 interruptPending = false; 848 } 849#endif 850 851 for (threadFetched = 0; threadFetched < numFetchingThreads; 852 threadFetched++) { 853 // Fetch each of the actively fetching threads. 854 fetch(status_change); 855 } 856 857 // Record number of instructions fetched this cycle for distribution. 858 fetchNisnDist.sample(numInst); 859 860 if (status_change) { 861 // Change the fetch stage status if there was a status change. 862 _status = updateFetchStatus(); 863 } 864 865 // If there was activity this cycle, inform the CPU of it. 866 if (wroteToTimeBuffer || cpu->contextSwitch) { 867 DPRINTF(Activity, "Activity this cycle.\n"); 868 869 cpu->activityThisCycle(); 870 } 871} 872 873template <class Impl> 874bool 875DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) 876{ 877 // Update the per thread stall statuses. 878 if (fromDecode->decodeBlock[tid]) { 879 stalls[tid].decode = true; 880 } 881 882 if (fromDecode->decodeUnblock[tid]) { 883 assert(stalls[tid].decode); 884 assert(!fromDecode->decodeBlock[tid]); 885 stalls[tid].decode = false; 886 } 887 888 if (fromRename->renameBlock[tid]) { 889 stalls[tid].rename = true; 890 } 891 892 if (fromRename->renameUnblock[tid]) { 893 assert(stalls[tid].rename); 894 assert(!fromRename->renameBlock[tid]); 895 stalls[tid].rename = false; 896 } 897 898 if (fromIEW->iewBlock[tid]) { 899 stalls[tid].iew = true; 900 } 901 902 if (fromIEW->iewUnblock[tid]) { 903 assert(stalls[tid].iew); 904 assert(!fromIEW->iewBlock[tid]); 905 stalls[tid].iew = false; 906 } 907 908 if (fromCommit->commitBlock[tid]) { 909 stalls[tid].commit = true; 910 } 911 912 if (fromCommit->commitUnblock[tid]) { 913 assert(stalls[tid].commit); 914 assert(!fromCommit->commitBlock[tid]); 915 stalls[tid].commit = false; 916 } 917 918 // Check squash signals from commit. 919 if (fromCommit->commitInfo[tid].squash) { 920 921 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 922 "from commit.\n",tid); 923 924#if ISA_HAS_DELAY_SLOT 925 InstSeqNum doneSeqNum = fromCommit->commitInfo[tid].bdelayDoneSeqNum; 926#else 927 InstSeqNum doneSeqNum = fromCommit->commitInfo[tid].doneSeqNum; 928#endif 929 // In any case, squash. 930 squash(fromCommit->commitInfo[tid].nextPC, 931 doneSeqNum, 932 fromCommit->commitInfo[tid].squashDelaySlot, 933 tid); 934 935 // Also check if there's a mispredict that happened. 936 if (fromCommit->commitInfo[tid].branchMispredict) { 937 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 938 fromCommit->commitInfo[tid].nextPC, 939 fromCommit->commitInfo[tid].branchTaken, 940 tid); 941 } else { 942 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 943 tid); 944 } 945 946 return true; 947 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 948 // Update the branch predictor if it wasn't a squashed instruction 949 // that was broadcasted. 950 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 951 } 952 953 // Check ROB squash signals from commit. 954 if (fromCommit->commitInfo[tid].robSquashing) { 955 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 956 957 // Continue to squash. 958 fetchStatus[tid] = Squashing; 959 960 return true; 961 } 962 963 // Check squash signals from decode. 964 if (fromDecode->decodeInfo[tid].squash) { 965 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 966 "from decode.\n",tid); 967 968 // Update the branch predictor. 969 if (fromDecode->decodeInfo[tid].branchMispredict) { 970 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 971 fromDecode->decodeInfo[tid].nextPC, 972 fromDecode->decodeInfo[tid].branchTaken, 973 tid); 974 } else { 975 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 976 tid); 977 } 978 979 if (fetchStatus[tid] != Squashing) { 980 981#if ISA_HAS_DELAY_SLOT 982 InstSeqNum doneSeqNum = fromDecode->decodeInfo[tid].bdelayDoneSeqNum; 983#else 984 InstSeqNum doneSeqNum = fromDecode->decodeInfo[tid].doneSeqNum; 985#endif 986 // Squash unless we're already squashing 987 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 988 doneSeqNum, 989 tid); 990 991 return true; 992 } 993 } 994 995 if (checkStall(tid) && 996 fetchStatus[tid] != IcacheWaitResponse && 997 fetchStatus[tid] != IcacheWaitRetry) { 998 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 999 1000 fetchStatus[tid] = Blocked; 1001 1002 return true; 1003 } 1004 1005 if (fetchStatus[tid] == Blocked || 1006 fetchStatus[tid] == Squashing) { 1007 // Switch status to running if fetch isn't being told to block or 1008 // squash this cycle. 1009 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 1010 tid); 1011 1012 fetchStatus[tid] = Running; 1013 1014 return true; 1015 } 1016 1017 // If we've reached this point, we have not gotten any signals that 1018 // cause fetch to change its status. Fetch remains the same as before. 1019 return false; 1020} 1021 1022template<class Impl> 1023void 1024DefaultFetch<Impl>::fetch(bool &status_change) 1025{ 1026 ////////////////////////////////////////// 1027 // Start actual fetch 1028 ////////////////////////////////////////// 1029 int tid = getFetchingThread(fetchPolicy); 1030 1031 if (tid == -1 || drainPending) { 1032 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 1033 1034 // Breaks looping condition in tick() 1035 threadFetched = numFetchingThreads; 1036 return; 1037 } 1038 1039 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 1040 1041 // The current PC. 1042 Addr &fetch_PC = PC[tid]; 1043 1044 // Fault code for memory access. 1045 Fault fault = NoFault; 1046 1047 // If returning from the delay of a cache miss, then update the status 1048 // to running, otherwise do the cache access. Possibly move this up 1049 // to tick() function. 1050 if (fetchStatus[tid] == IcacheAccessComplete) { 1051 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", 1052 tid); 1053 1054 fetchStatus[tid] = Running; 1055 status_change = true; 1056 } else if (fetchStatus[tid] == Running) { 1057 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 1058 "instruction, starting at PC %08p.\n", 1059 tid, fetch_PC); 1060 1061 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid); 1062 if (!fetch_success) { 1063 if (cacheBlocked) { 1064 ++icacheStallCycles; 1065 } else { 1066 ++fetchMiscStallCycles; 1067 } 1068 return; 1069 } 1070 } else { 1071 if (fetchStatus[tid] == Idle) { 1072 ++fetchIdleCycles; 1073 DPRINTF(Fetch, "[tid:%i]: Fetch is idle!\n", tid); 1074 } else if (fetchStatus[tid] == Blocked) { 1075 ++fetchBlockedCycles; 1076 DPRINTF(Fetch, "[tid:%i]: Fetch is blocked!\n", tid); 1077 } else if (fetchStatus[tid] == Squashing) { 1078 ++fetchSquashCycles; 1079 DPRINTF(Fetch, "[tid:%i]: Fetch is squashing!\n", tid); 1080 } else if (fetchStatus[tid] == IcacheWaitResponse) { 1081 ++icacheStallCycles; 1082 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting cache response!\n", tid); 1083 } 1084 1085 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so 1086 // fetch should do nothing. 1087 return; 1088 } 1089 1090 ++fetchCycles; 1091 1092 // If we had a stall due to an icache miss, then return. 1093 if (fetchStatus[tid] == IcacheWaitResponse) { 1094 ++icacheStallCycles; 1095 status_change = true; 1096 return; 1097 } 1098 1099 Addr next_PC = fetch_PC; 1100 Addr next_NPC = next_PC + instSize; 1101 InstSeqNum inst_seq; 1102 MachInst inst; 1103 ExtMachInst ext_inst; 1104 // @todo: Fix this hack. 1105 unsigned offset = (fetch_PC & cacheBlkMask) & ~3; 1106 1107 if (fault == NoFault) { 1108 // If the read of the first instruction was successful, then grab the 1109 // instructions from the rest of the cache line and put them into the 1110 // queue heading to decode. 1111 1112 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 1113 "decode.\n",tid); 1114 1115 // Need to keep track of whether or not a predicted branch 1116 // ended this fetch block. 1117 bool predicted_branch = false; 1118 1119 // Need to keep track of whether or not a delay slot 1120 // instruction has been fetched 1121 1122 for (; 1123 offset < cacheBlkSize && 1124 numInst < fetchWidth && 1125 (!predicted_branch || delaySlotInfo[tid].numInsts > 0); 1126 ++numInst) { 1127 1128 // Get a sequence number. 1129 inst_seq = cpu->getAndIncrementInstSeq(); 1130 1131 // Make sure this is a valid index. 1132 assert(offset <= cacheBlkSize - instSize); 1133 1134 // Get the instruction from the array of the cache line. 1135 inst = TheISA::gtoh(*reinterpret_cast<TheISA::MachInst *> 1136 (&cacheData[tid][offset])); 1137 1138#if THE_ISA == ALPHA_ISA 1139 ext_inst = TheISA::makeExtMI(inst, fetch_PC); 1140#elif THE_ISA == SPARC_ISA 1141 ext_inst = TheISA::makeExtMI(inst, cpu->thread[tid]->getTC()); 1142#endif 1143 1144 // Create a new DynInst from the instruction fetched. 1145 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1146 next_PC, 1147 inst_seq, cpu); 1148 instruction->setTid(tid); 1149 1150 instruction->setASID(tid); 1151 1152 instruction->setThreadState(cpu->thread[tid]); 1153 1154 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created " 1155 "[sn:%lli]\n", 1156 tid, instruction->readPC(), inst_seq); 1157 1158 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", 1159 tid, instruction->staticInst->disassemble(fetch_PC)); 1160 1161 instruction->traceData = 1162 Trace::getInstRecord(curTick, cpu->tcBase(tid), 1163 instruction->staticInst, 1164 instruction->readPC()); 1165 1166 predicted_branch = lookupAndUpdateNextPC(instruction, next_PC, 1167 next_NPC); 1168 1169 // Add instruction to the CPU's list of instructions. 1170 instruction->setInstListIt(cpu->addInst(instruction)); 1171 1172 // Write the instruction to the first slot in the queue 1173 // that heads to decode. 1174 toDecode->insts[numInst] = instruction; 1175 1176 toDecode->size++; 1177 1178 // Increment stat of fetched instructions. 1179 ++fetchedInsts; 1180 1181 // Move to the next instruction, unless we have a branch. 1182 fetch_PC = next_PC; 1183 1184 if (instruction->isQuiesce()) { 1185 DPRINTF(Fetch, "Quiesce instruction encountered, halting fetch!", 1186 curTick); 1187 fetchStatus[tid] = QuiescePending; 1188 ++numInst; 1189 status_change = true; 1190 break; 1191 } 1192 1193 offset += instSize; 1194 1195#if ISA_HAS_DELAY_SLOT 1196 if (predicted_branch) { 1197 delaySlotInfo[tid].branchSeqNum = inst_seq; 1198 1199 DPRINTF(Fetch, "[tid:%i]: Delay slot branch set to [sn:%i]\n", 1200 tid, inst_seq); 1201 continue; 1202 } else if (delaySlotInfo[tid].numInsts > 0) { 1203 --delaySlotInfo[tid].numInsts; 1204 1205 // It's OK to set PC to target of branch 1206 if (delaySlotInfo[tid].numInsts == 0) { 1207 delaySlotInfo[tid].targetReady = true; 1208 1209 // Break the looping condition 1210 predicted_branch = true; 1211 } 1212 1213 DPRINTF(Fetch, "[tid:%i]: %i delay slot inst(s) left to" 1214 " process.\n", tid, delaySlotInfo[tid].numInsts); 1215 } 1216#endif 1217 } 1218 1219 if (offset >= cacheBlkSize) { 1220 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache " 1221 "block.\n", tid); 1222 } else if (numInst >= fetchWidth) { 1223 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth " 1224 "for this cycle.\n", tid); 1225 } else if (predicted_branch && delaySlotInfo[tid].numInsts <= 0) { 1226 DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch " 1227 "instruction encountered.\n", tid); 1228 } 1229 } 1230 1231 if (numInst > 0) { 1232 wroteToTimeBuffer = true; 1233 } 1234 1235 // Now that fetching is completed, update the PC to signify what the next 1236 // cycle will be. 1237 if (fault == NoFault) { 1238#if ISA_HAS_DELAY_SLOT 1239 if (delaySlotInfo[tid].targetReady && 1240 delaySlotInfo[tid].numInsts == 0) { 1241 // Set PC to target 1242 PC[tid] = delaySlotInfo[tid].targetAddr; //next_PC 1243 nextPC[tid] = next_PC + instSize; //next_NPC 1244 nextNPC[tid] = next_PC + (2 * instSize); 1245 1246 delaySlotInfo[tid].targetReady = false; 1247 } else { 1248 PC[tid] = next_PC; 1249 nextPC[tid] = next_NPC; 1250 nextNPC[tid] = next_NPC + instSize; 1251 } 1252 1253 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n", tid, PC[tid]); 1254#else 1255 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n",tid, next_PC); 1256 PC[tid] = next_PC; 1257 nextPC[tid] = next_PC + instSize; 1258#endif 1259 } else { 1260 // We shouldn't be in an icache miss and also have a fault (an ITB 1261 // miss) 1262 if (fetchStatus[tid] == IcacheWaitResponse) { 1263 panic("Fetch should have exited prior to this!"); 1264 } 1265 1266 // Send the fault to commit. This thread will not do anything 1267 // until commit handles the fault. The only other way it can 1268 // wake up is if a squash comes along and changes the PC. 1269#if FULL_SYSTEM 1270 assert(numInst != fetchWidth); 1271 // Get a sequence number. 1272 inst_seq = cpu->getAndIncrementInstSeq(); 1273 // We will use a nop in order to carry the fault. 1274 ext_inst = TheISA::NoopMachInst; 1275 1276 // Create a new DynInst from the dummy nop. 1277 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1278 next_PC, 1279 inst_seq, cpu); 1280 instruction->setPredTarg(next_PC + instSize); 1281 instruction->setTid(tid); 1282 1283 instruction->setASID(tid); 1284 1285 instruction->setThreadState(cpu->thread[tid]); 1286 1287 instruction->traceData = NULL; 1288 1289 instruction->setInstListIt(cpu->addInst(instruction)); 1290 1291 instruction->fault = fault; 1292 1293 toDecode->insts[numInst] = instruction; 1294 toDecode->size++; 1295 1296 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid); 1297 1298 fetchStatus[tid] = TrapPending; 1299 status_change = true; 1300#else // !FULL_SYSTEM 1301 fetchStatus[tid] = TrapPending; 1302 status_change = true; 1303 1304#endif // FULL_SYSTEM 1305 DPRINTF(Fetch, "[tid:%i]: fault (%s) detected @ PC %08p", 1306 tid, fault->name(), PC[tid]); 1307 } 1308} 1309 1310template<class Impl> 1311void 1312DefaultFetch<Impl>::recvRetry() 1313{ 1314 if (retryPkt != NULL) { 1315 assert(cacheBlocked); 1316 assert(retryTid != -1); 1317 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1318 1319 if (icachePort->sendTiming(retryPkt)) { 1320 fetchStatus[retryTid] = IcacheWaitResponse; 1321 retryPkt = NULL; 1322 retryTid = -1; 1323 cacheBlocked = false; 1324 } 1325 } else { 1326 assert(retryTid == -1); 1327 // Access has been squashed since it was sent out. Just clear 1328 // the cache being blocked. 1329 cacheBlocked = false; 1330 } 1331} 1332 1333/////////////////////////////////////// 1334// // 1335// SMT FETCH POLICY MAINTAINED HERE // 1336// // 1337/////////////////////////////////////// 1338template<class Impl> 1339int 1340DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1341{ 1342 if (numThreads > 1) { 1343 switch (fetch_priority) { 1344 1345 case SingleThread: 1346 return 0; 1347 1348 case RoundRobin: 1349 return roundRobin(); 1350 1351 case IQ: 1352 return iqCount(); 1353 1354 case LSQ: 1355 return lsqCount(); 1356 1357 case Branch: 1358 return branchCount(); 1359 1360 default: 1361 return -1; 1362 } 1363 } else { 1364 int tid = *((*activeThreads).begin()); 1365 1366 if (fetchStatus[tid] == Running || 1367 fetchStatus[tid] == IcacheAccessComplete || 1368 fetchStatus[tid] == Idle) { 1369 return tid; 1370 } else { 1371 return -1; 1372 } 1373 } 1374 1375} 1376 1377 1378template<class Impl> 1379int 1380DefaultFetch<Impl>::roundRobin() 1381{ 1382 std::list<unsigned>::iterator pri_iter = priorityList.begin(); 1383 std::list<unsigned>::iterator end = priorityList.end(); 1384 1385 int high_pri; 1386 1387 while (pri_iter != end) { 1388 high_pri = *pri_iter; 1389 1390 assert(high_pri <= numThreads); 1391 1392 if (fetchStatus[high_pri] == Running || 1393 fetchStatus[high_pri] == IcacheAccessComplete || 1394 fetchStatus[high_pri] == Idle) { 1395 1396 priorityList.erase(pri_iter); 1397 priorityList.push_back(high_pri); 1398 1399 return high_pri; 1400 } 1401 1402 pri_iter++; 1403 } 1404 1405 return -1; 1406} 1407 1408template<class Impl> 1409int 1410DefaultFetch<Impl>::iqCount() 1411{ 1412 std::priority_queue<unsigned> PQ; 1413 1414 std::list<unsigned>::iterator threads = (*activeThreads).begin(); 1415 1416 while (threads != (*activeThreads).end()) { 1417 unsigned tid = *threads++; 1418 1419 PQ.push(fromIEW->iewInfo[tid].iqCount); 1420 } 1421 1422 while (!PQ.empty()) { 1423 1424 unsigned high_pri = PQ.top(); 1425 1426 if (fetchStatus[high_pri] == Running || 1427 fetchStatus[high_pri] == IcacheAccessComplete || 1428 fetchStatus[high_pri] == Idle) 1429 return high_pri; 1430 else 1431 PQ.pop(); 1432 1433 } 1434 1435 return -1; 1436} 1437 1438template<class Impl> 1439int 1440DefaultFetch<Impl>::lsqCount() 1441{ 1442 std::priority_queue<unsigned> PQ; 1443 1444 1445 std::list<unsigned>::iterator threads = (*activeThreads).begin(); 1446 1447 while (threads != (*activeThreads).end()) { 1448 unsigned tid = *threads++; 1449 1450 PQ.push(fromIEW->iewInfo[tid].ldstqCount); 1451 } 1452 1453 while (!PQ.empty()) { 1454 1455 unsigned high_pri = PQ.top(); 1456 1457 if (fetchStatus[high_pri] == Running || 1458 fetchStatus[high_pri] == IcacheAccessComplete || 1459 fetchStatus[high_pri] == Idle) 1460 return high_pri; 1461 else 1462 PQ.pop(); 1463 1464 } 1465 1466 return -1; 1467} 1468 1469template<class Impl> 1470int 1471DefaultFetch<Impl>::branchCount() 1472{ 1473 std::list<unsigned>::iterator threads = (*activeThreads).begin(); 1474 panic("Branch Count Fetch policy unimplemented\n"); 1475 return *threads; 1476}
|