fetch_impl.hh revision 2696:30b38e36ff54
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 */ 30 31#include "arch/isa_traits.hh" 32#include "arch/utility.hh" 33#include "cpu/checker/cpu.hh" 34#include "cpu/exetrace.hh" 35#include "cpu/o3/fetch.hh" 36#include "mem/packet.hh" 37#include "mem/request.hh" 38#include "sim/byteswap.hh" 39#include "sim/host.hh" 40#include "sim/root.hh" 41 42#if FULL_SYSTEM 43#include "arch/tlb.hh" 44#include "arch/vtophys.hh" 45#include "base/remote_gdb.hh" 46#include "sim/system.hh" 47#endif // FULL_SYSTEM 48 49#include <algorithm> 50 51using namespace std; 52using namespace TheISA; 53 54template<class Impl> 55Tick 56DefaultFetch<Impl>::IcachePort::recvAtomic(PacketPtr pkt) 57{ 58 panic("DefaultFetch doesn't expect recvAtomic callback!"); 59 return curTick; 60} 61 62template<class Impl> 63void 64DefaultFetch<Impl>::IcachePort::recvFunctional(PacketPtr pkt) 65{ 66 panic("DefaultFetch doesn't expect recvFunctional callback!"); 67} 68 69template<class Impl> 70void 71DefaultFetch<Impl>::IcachePort::recvStatusChange(Status status) 72{ 73 if (status == RangeChange) 74 return; 75 76 panic("DefaultFetch doesn't expect recvStatusChange callback!"); 77} 78 79template<class Impl> 80bool 81DefaultFetch<Impl>::IcachePort::recvTiming(Packet *pkt) 82{ 83 fetch->processCacheCompletion(pkt); 84 return true; 85} 86 87template<class Impl> 88void 89DefaultFetch<Impl>::IcachePort::recvRetry() 90{ 91 fetch->recvRetry(); 92} 93 94template<class Impl> 95DefaultFetch<Impl>::DefaultFetch(Params *params) 96 : mem(params->mem), 97 branchPred(params), 98 decodeToFetchDelay(params->decodeToFetchDelay), 99 renameToFetchDelay(params->renameToFetchDelay), 100 iewToFetchDelay(params->iewToFetchDelay), 101 commitToFetchDelay(params->commitToFetchDelay), 102 fetchWidth(params->fetchWidth), 103 cacheBlocked(false), 104 retryPkt(NULL), 105 retryTid(-1), 106 numThreads(params->numberOfThreads), 107 numFetchingThreads(params->smtNumFetchingThreads), 108 interruptPending(false), 109 switchedOut(false) 110{ 111 if (numThreads > Impl::MaxThreads) 112 fatal("numThreads is not a valid value\n"); 113 114 DPRINTF(Fetch, "Fetch constructor called\n"); 115 116 // Set fetch stage's status to inactive. 117 _status = Inactive; 118 119 string policy = params->smtFetchPolicy; 120 121 // Convert string to lowercase 122 std::transform(policy.begin(), policy.end(), policy.begin(), 123 (int(*)(int)) tolower); 124 125 // Figure out fetch policy 126 if (policy == "singlethread") { 127 fetchPolicy = SingleThread; 128 } else if (policy == "roundrobin") { 129 fetchPolicy = RoundRobin; 130 DPRINTF(Fetch, "Fetch policy set to Round Robin\n"); 131 } else if (policy == "branch") { 132 fetchPolicy = Branch; 133 DPRINTF(Fetch, "Fetch policy set to Branch Count\n"); 134 } else if (policy == "iqcount") { 135 fetchPolicy = IQ; 136 DPRINTF(Fetch, "Fetch policy set to IQ count\n"); 137 } else if (policy == "lsqcount") { 138 fetchPolicy = LSQ; 139 DPRINTF(Fetch, "Fetch policy set to LSQ count\n"); 140 } else { 141 fatal("Invalid Fetch Policy. Options Are: {SingleThread," 142 " RoundRobin,LSQcount,IQcount}\n"); 143 } 144 145 // Size of cache block. 146 cacheBlkSize = 64; 147 148 // Create mask to get rid of offset bits. 149 cacheBlkMask = (cacheBlkSize - 1); 150 151 for (int tid=0; tid < numThreads; tid++) { 152 153 fetchStatus[tid] = Running; 154 155 priorityList.push_back(tid); 156 157 memReq[tid] = NULL; 158 159 // Create space to store a cache line. 160 cacheData[tid] = new uint8_t[cacheBlkSize]; 161 162 stalls[tid].decode = 0; 163 stalls[tid].rename = 0; 164 stalls[tid].iew = 0; 165 stalls[tid].commit = 0; 166 } 167 168 // Get the size of an instruction. 169 instSize = sizeof(MachInst); 170} 171 172template <class Impl> 173std::string 174DefaultFetch<Impl>::name() const 175{ 176 return cpu->name() + ".fetch"; 177} 178 179template <class Impl> 180void 181DefaultFetch<Impl>::regStats() 182{ 183 icacheStallCycles 184 .name(name() + ".icacheStallCycles") 185 .desc("Number of cycles fetch is stalled on an Icache miss") 186 .prereq(icacheStallCycles); 187 188 fetchedInsts 189 .name(name() + ".Insts") 190 .desc("Number of instructions fetch has processed") 191 .prereq(fetchedInsts); 192 193 fetchedBranches 194 .name(name() + ".Branches") 195 .desc("Number of branches that fetch encountered") 196 .prereq(fetchedBranches); 197 198 predictedBranches 199 .name(name() + ".predictedBranches") 200 .desc("Number of branches that fetch has predicted taken") 201 .prereq(predictedBranches); 202 203 fetchCycles 204 .name(name() + ".Cycles") 205 .desc("Number of cycles fetch has run and was not squashing or" 206 " blocked") 207 .prereq(fetchCycles); 208 209 fetchSquashCycles 210 .name(name() + ".SquashCycles") 211 .desc("Number of cycles fetch has spent squashing") 212 .prereq(fetchSquashCycles); 213 214 fetchIdleCycles 215 .name(name() + ".IdleCycles") 216 .desc("Number of cycles fetch was idle") 217 .prereq(fetchIdleCycles); 218 219 fetchBlockedCycles 220 .name(name() + ".BlockedCycles") 221 .desc("Number of cycles fetch has spent blocked") 222 .prereq(fetchBlockedCycles); 223 224 fetchedCacheLines 225 .name(name() + ".CacheLines") 226 .desc("Number of cache lines fetched") 227 .prereq(fetchedCacheLines); 228 229 fetchMiscStallCycles 230 .name(name() + ".MiscStallCycles") 231 .desc("Number of cycles fetch has spent waiting on interrupts, or " 232 "bad addresses, or out of MSHRs") 233 .prereq(fetchMiscStallCycles); 234 235 fetchIcacheSquashes 236 .name(name() + ".IcacheSquashes") 237 .desc("Number of outstanding Icache misses that were squashed") 238 .prereq(fetchIcacheSquashes); 239 240 fetchNisnDist 241 .init(/* base value */ 0, 242 /* last value */ fetchWidth, 243 /* bucket size */ 1) 244 .name(name() + ".rateDist") 245 .desc("Number of instructions fetched each cycle (Total)") 246 .flags(Stats::pdf); 247 248 idleRate 249 .name(name() + ".idleRate") 250 .desc("Percent of cycles fetch was idle") 251 .prereq(idleRate); 252 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 253 254 branchRate 255 .name(name() + ".branchRate") 256 .desc("Number of branch fetches per cycle") 257 .flags(Stats::total); 258 branchRate = fetchedBranches / cpu->numCycles; 259 260 fetchRate 261 .name(name() + ".rate") 262 .desc("Number of inst fetches per cycle") 263 .flags(Stats::total); 264 fetchRate = fetchedInsts / cpu->numCycles; 265 266 branchPred.regStats(); 267} 268 269template<class Impl> 270void 271DefaultFetch<Impl>::setCPU(FullCPU *cpu_ptr) 272{ 273 DPRINTF(Fetch, "Setting the CPU pointer.\n"); 274 cpu = cpu_ptr; 275 276 // Name is finally available, so create the port. 277 icachePort = new IcachePort(this); 278 279 Port *mem_dport = mem->getPort(""); 280 icachePort->setPeer(mem_dport); 281 mem_dport->setPeer(icachePort); 282 283 if (cpu->checker) { 284 cpu->checker->setIcachePort(icachePort); 285 } 286 287 // Fetch needs to start fetching instructions at the very beginning, 288 // so it must start up in active state. 289 switchToActive(); 290} 291 292template<class Impl> 293void 294DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 295{ 296 DPRINTF(Fetch, "Setting the time buffer pointer.\n"); 297 timeBuffer = time_buffer; 298 299 // Create wires to get information from proper places in time buffer. 300 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 301 fromRename = timeBuffer->getWire(-renameToFetchDelay); 302 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 303 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 304} 305 306template<class Impl> 307void 308DefaultFetch<Impl>::setActiveThreads(list<unsigned> *at_ptr) 309{ 310 DPRINTF(Fetch, "Setting active threads list pointer.\n"); 311 activeThreads = at_ptr; 312} 313 314template<class Impl> 315void 316DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 317{ 318 DPRINTF(Fetch, "Setting the fetch queue pointer.\n"); 319 fetchQueue = fq_ptr; 320 321 // Create wire to write information to proper place in fetch queue. 322 toDecode = fetchQueue->getWire(0); 323} 324 325#if 0 326template<class Impl> 327void 328DefaultFetch<Impl>::setPageTable(PageTable *pt_ptr) 329{ 330 DPRINTF(Fetch, "Setting the page table pointer.\n"); 331#if !FULL_SYSTEM 332 pTable = pt_ptr; 333#endif 334} 335#endif 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 } 346} 347 348template<class Impl> 349void 350DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 351{ 352 unsigned tid = pkt->req->getThreadNum(); 353 354 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid); 355 356 // Only change the status if it's still waiting on the icache access 357 // to return. 358 if (fetchStatus[tid] != IcacheWaitResponse || 359 pkt->req != memReq[tid] || 360 isSwitchedOut()) { 361 ++fetchIcacheSquashes; 362 delete pkt->req; 363 delete pkt; 364 memReq[tid] = NULL; 365 return; 366 } 367 368 // Wake up the CPU (if it went to sleep and was waiting on this completion 369 // event). 370 cpu->wakeCPU(); 371 372 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 373 tid); 374 375 switchToActive(); 376 377 // Only switch to IcacheAccessComplete if we're not stalled as well. 378 if (checkStall(tid)) { 379 fetchStatus[tid] = Blocked; 380 } else { 381 fetchStatus[tid] = IcacheAccessComplete; 382 } 383 384// memcpy(cacheData[tid], memReq[tid]->data, memReq[tid]->size); 385 386 // Reset the mem req to NULL. 387 delete pkt->req; 388 delete pkt; 389 memReq[tid] = NULL; 390} 391 392template <class Impl> 393void 394DefaultFetch<Impl>::switchOut() 395{ 396 // Fetch is ready to switch out at any time. 397 switchedOut = true; 398 cpu->signalSwitched(); 399} 400 401template <class Impl> 402void 403DefaultFetch<Impl>::doSwitchOut() 404{ 405 // Branch predictor needs to have its state cleared. 406 branchPred.switchOut(); 407} 408 409template <class Impl> 410void 411DefaultFetch<Impl>::takeOverFrom() 412{ 413 // Reset all state 414 for (int i = 0; i < Impl::MaxThreads; ++i) { 415 stalls[i].decode = 0; 416 stalls[i].rename = 0; 417 stalls[i].iew = 0; 418 stalls[i].commit = 0; 419 PC[i] = cpu->readPC(i); 420 nextPC[i] = cpu->readNextPC(i); 421 fetchStatus[i] = Running; 422 } 423 numInst = 0; 424 wroteToTimeBuffer = false; 425 _status = Inactive; 426 switchedOut = false; 427 branchPred.takeOverFrom(); 428} 429 430template <class Impl> 431void 432DefaultFetch<Impl>::wakeFromQuiesce() 433{ 434 DPRINTF(Fetch, "Waking up from quiesce\n"); 435 // Hopefully this is safe 436 // @todo: Allow other threads to wake from quiesce. 437 fetchStatus[0] = Running; 438} 439 440template <class Impl> 441inline void 442DefaultFetch<Impl>::switchToActive() 443{ 444 if (_status == Inactive) { 445 DPRINTF(Activity, "Activating stage.\n"); 446 447 cpu->activateStage(FullCPU::FetchIdx); 448 449 _status = Active; 450 } 451} 452 453template <class Impl> 454inline void 455DefaultFetch<Impl>::switchToInactive() 456{ 457 if (_status == Active) { 458 DPRINTF(Activity, "Deactivating stage.\n"); 459 460 cpu->deactivateStage(FullCPU::FetchIdx); 461 462 _status = Inactive; 463 } 464} 465 466template <class Impl> 467bool 468DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC) 469{ 470 // Do branch prediction check here. 471 // A bit of a misnomer...next_PC is actually the current PC until 472 // this function updates it. 473 bool predict_taken; 474 475 if (!inst->isControl()) { 476 next_PC = next_PC + instSize; 477 inst->setPredTarg(next_PC); 478 return false; 479 } 480 481 predict_taken = branchPred.predict(inst, next_PC, inst->threadNumber); 482 483 ++fetchedBranches; 484 485 if (predict_taken) { 486 ++predictedBranches; 487 } 488 489 return predict_taken; 490} 491 492template <class Impl> 493bool 494DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid) 495{ 496 Fault fault = NoFault; 497 498#if FULL_SYSTEM 499 // Flag to say whether or not address is physical addr. 500 unsigned flags = cpu->inPalMode(fetch_PC) ? PHYSICAL : 0; 501#else 502 unsigned flags = 0; 503#endif // FULL_SYSTEM 504 505 if (cacheBlocked || (interruptPending && flags == 0) || switchedOut) { 506 // Hold off fetch from getting new instructions when: 507 // Cache is blocked, or 508 // while an interrupt is pending and we're not in PAL mode, or 509 // fetch is switched out. 510 return false; 511 } 512 513 // Align the fetch PC so it's at the start of a cache block. 514 fetch_PC = icacheBlockAlignPC(fetch_PC); 515 516 // Setup the memReq to do a read of the first instruction's address. 517 // Set the appropriate read size and flags as well. 518 // Build request here. 519 RequestPtr mem_req = new Request(tid, fetch_PC, cacheBlkSize, flags, 520 fetch_PC, cpu->readCpuId(), tid); 521 522 memReq[tid] = mem_req; 523 524 // Translate the instruction request. 525 fault = cpu->translateInstReq(mem_req, cpu->thread[tid]); 526 527 // In the case of faults, the fetch stage may need to stall and wait 528 // for the ITB miss to be handled. 529 530 // If translation was successful, attempt to read the first 531 // instruction. 532 if (fault == NoFault) { 533#if 0 534 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || 535 memReq[tid]->flags & UNCACHEABLE) { 536 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " 537 "misspeculating path)!", 538 memReq[tid]->paddr); 539 ret_fault = TheISA::genMachineCheckFault(); 540 return false; 541 } 542#endif 543 544 // Build packet here. 545 PacketPtr data_pkt = new Packet(mem_req, 546 Packet::ReadReq, Packet::Broadcast); 547 data_pkt->dataStatic(cacheData[tid]); 548 549 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 550 551 fetchedCacheLines++; 552 553 // Now do the timing access to see whether or not the instruction 554 // exists within the cache. 555 if (!icachePort->sendTiming(data_pkt)) { 556 assert(retryPkt == NULL); 557 assert(retryTid == -1); 558 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 559 fetchStatus[tid] = IcacheWaitRetry; 560 retryPkt = data_pkt; 561 retryTid = tid; 562 cacheBlocked = true; 563 return false; 564 } 565 566 DPRINTF(Fetch, "Doing cache access.\n"); 567 568 lastIcacheStall[tid] = curTick; 569 570 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 571 "response.\n", tid); 572 573 fetchStatus[tid] = IcacheWaitResponse; 574 } else { 575 delete mem_req; 576 memReq[tid] = NULL; 577 } 578 579 ret_fault = fault; 580 return true; 581} 582 583template <class Impl> 584inline void 585DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid) 586{ 587 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x.\n", 588 tid, new_PC); 589 590 PC[tid] = new_PC; 591 nextPC[tid] = new_PC + instSize; 592 593 // Clear the icache miss if it's outstanding. 594 if (fetchStatus[tid] == IcacheWaitResponse) { 595 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 596 tid); 597 // Should I delete this here or when it comes back from the cache? 598// delete memReq[tid]; 599 memReq[tid] = NULL; 600 } 601 602 // Get rid of the retrying packet if it was from this thread. 603 if (retryTid == tid) { 604 assert(cacheBlocked); 605 cacheBlocked = false; 606 retryTid = -1; 607 retryPkt = NULL; 608 delete retryPkt->req; 609 delete retryPkt; 610 } 611 612 fetchStatus[tid] = Squashing; 613 614 ++fetchSquashCycles; 615} 616 617template<class Impl> 618void 619DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, 620 const InstSeqNum &seq_num, 621 unsigned tid) 622{ 623 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid); 624 625 doSquash(new_PC, tid); 626 627 // Tell the CPU to remove any instructions that are in flight between 628 // fetch and decode. 629 cpu->removeInstsUntil(seq_num, tid); 630} 631 632template<class Impl> 633bool 634DefaultFetch<Impl>::checkStall(unsigned tid) const 635{ 636 bool ret_val = false; 637 638 if (cpu->contextSwitch) { 639 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 640 ret_val = true; 641 } else if (stalls[tid].decode) { 642 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 643 ret_val = true; 644 } else if (stalls[tid].rename) { 645 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 646 ret_val = true; 647 } else if (stalls[tid].iew) { 648 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 649 ret_val = true; 650 } else if (stalls[tid].commit) { 651 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 652 ret_val = true; 653 } 654 655 return ret_val; 656} 657 658template<class Impl> 659typename DefaultFetch<Impl>::FetchStatus 660DefaultFetch<Impl>::updateFetchStatus() 661{ 662 //Check Running 663 list<unsigned>::iterator threads = (*activeThreads).begin(); 664 665 while (threads != (*activeThreads).end()) { 666 667 unsigned tid = *threads++; 668 669 if (fetchStatus[tid] == Running || 670 fetchStatus[tid] == Squashing || 671 fetchStatus[tid] == IcacheAccessComplete) { 672 673 if (_status == Inactive) { 674 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 675 676 if (fetchStatus[tid] == IcacheAccessComplete) { 677 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 678 "completion\n",tid); 679 } 680 681 cpu->activateStage(FullCPU::FetchIdx); 682 } 683 684 return Active; 685 } 686 } 687 688 // Stage is switching from active to inactive, notify CPU of it. 689 if (_status == Active) { 690 DPRINTF(Activity, "Deactivating stage.\n"); 691 692 cpu->deactivateStage(FullCPU::FetchIdx); 693 } 694 695 return Inactive; 696} 697 698template <class Impl> 699void 700DefaultFetch<Impl>::squash(const Addr &new_PC, unsigned tid) 701{ 702 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid); 703 704 doSquash(new_PC, tid); 705 706 // Tell the CPU to remove any instructions that are not in the ROB. 707 cpu->removeInstsNotInROB(tid); 708} 709 710template <class Impl> 711void 712DefaultFetch<Impl>::tick() 713{ 714 list<unsigned>::iterator threads = (*activeThreads).begin(); 715 bool status_change = false; 716 717 wroteToTimeBuffer = false; 718 719 while (threads != (*activeThreads).end()) { 720 unsigned tid = *threads++; 721 722 // Check the signals for each thread to determine the proper status 723 // for each thread. 724 bool updated_status = checkSignalsAndUpdate(tid); 725 status_change = status_change || updated_status; 726 } 727 728 DPRINTF(Fetch, "Running stage.\n"); 729 730 // Reset the number of the instruction we're fetching. 731 numInst = 0; 732 733 if (fromCommit->commitInfo[0].interruptPending) { 734 interruptPending = true; 735 } 736 if (fromCommit->commitInfo[0].clearInterrupt) { 737 interruptPending = false; 738 } 739 740 for (threadFetched = 0; threadFetched < numFetchingThreads; 741 threadFetched++) { 742 // Fetch each of the actively fetching threads. 743 fetch(status_change); 744 } 745 746 // Record number of instructions fetched this cycle for distribution. 747 fetchNisnDist.sample(numInst); 748 749 if (status_change) { 750 // Change the fetch stage status if there was a status change. 751 _status = updateFetchStatus(); 752 } 753 754 // If there was activity this cycle, inform the CPU of it. 755 if (wroteToTimeBuffer || cpu->contextSwitch) { 756 DPRINTF(Activity, "Activity this cycle.\n"); 757 758 cpu->activityThisCycle(); 759 } 760} 761 762template <class Impl> 763bool 764DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) 765{ 766 // Update the per thread stall statuses. 767 if (fromDecode->decodeBlock[tid]) { 768 stalls[tid].decode = true; 769 } 770 771 if (fromDecode->decodeUnblock[tid]) { 772 assert(stalls[tid].decode); 773 assert(!fromDecode->decodeBlock[tid]); 774 stalls[tid].decode = false; 775 } 776 777 if (fromRename->renameBlock[tid]) { 778 stalls[tid].rename = true; 779 } 780 781 if (fromRename->renameUnblock[tid]) { 782 assert(stalls[tid].rename); 783 assert(!fromRename->renameBlock[tid]); 784 stalls[tid].rename = false; 785 } 786 787 if (fromIEW->iewBlock[tid]) { 788 stalls[tid].iew = true; 789 } 790 791 if (fromIEW->iewUnblock[tid]) { 792 assert(stalls[tid].iew); 793 assert(!fromIEW->iewBlock[tid]); 794 stalls[tid].iew = false; 795 } 796 797 if (fromCommit->commitBlock[tid]) { 798 stalls[tid].commit = true; 799 } 800 801 if (fromCommit->commitUnblock[tid]) { 802 assert(stalls[tid].commit); 803 assert(!fromCommit->commitBlock[tid]); 804 stalls[tid].commit = false; 805 } 806 807 // Check squash signals from commit. 808 if (fromCommit->commitInfo[tid].squash) { 809 810 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 811 "from commit.\n",tid); 812 813 // In any case, squash. 814 squash(fromCommit->commitInfo[tid].nextPC,tid); 815 816 // Also check if there's a mispredict that happened. 817 if (fromCommit->commitInfo[tid].branchMispredict) { 818 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 819 fromCommit->commitInfo[tid].nextPC, 820 fromCommit->commitInfo[tid].branchTaken, 821 tid); 822 } else { 823 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 824 tid); 825 } 826 827 return true; 828 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 829 // Update the branch predictor if it wasn't a squashed instruction 830 // that was broadcasted. 831 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 832 } 833 834 // Check ROB squash signals from commit. 835 if (fromCommit->commitInfo[tid].robSquashing) { 836 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing Thread %u.\n", tid); 837 838 // Continue to squash. 839 fetchStatus[tid] = Squashing; 840 841 return true; 842 } 843 844 // Check squash signals from decode. 845 if (fromDecode->decodeInfo[tid].squash) { 846 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 847 "from decode.\n",tid); 848 849 // Update the branch predictor. 850 if (fromDecode->decodeInfo[tid].branchMispredict) { 851 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 852 fromDecode->decodeInfo[tid].nextPC, 853 fromDecode->decodeInfo[tid].branchTaken, 854 tid); 855 } else { 856 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 857 tid); 858 } 859 860 if (fetchStatus[tid] != Squashing) { 861 // Squash unless we're already squashing 862 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 863 fromDecode->decodeInfo[tid].doneSeqNum, 864 tid); 865 866 return true; 867 } 868 } 869 870 if (checkStall(tid) && fetchStatus[tid] != IcacheWaitResponse) { 871 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 872 873 fetchStatus[tid] = Blocked; 874 875 return true; 876 } 877 878 if (fetchStatus[tid] == Blocked || 879 fetchStatus[tid] == Squashing) { 880 // Switch status to running if fetch isn't being told to block or 881 // squash this cycle. 882 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 883 tid); 884 885 fetchStatus[tid] = Running; 886 887 return true; 888 } 889 890 // If we've reached this point, we have not gotten any signals that 891 // cause fetch to change its status. Fetch remains the same as before. 892 return false; 893} 894 895template<class Impl> 896void 897DefaultFetch<Impl>::fetch(bool &status_change) 898{ 899 ////////////////////////////////////////// 900 // Start actual fetch 901 ////////////////////////////////////////// 902 int tid = getFetchingThread(fetchPolicy); 903 904 if (tid == -1) { 905 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 906 907 // Breaks looping condition in tick() 908 threadFetched = numFetchingThreads; 909 return; 910 } 911 912 // The current PC. 913 Addr &fetch_PC = PC[tid]; 914 915 // Fault code for memory access. 916 Fault fault = NoFault; 917 918 // If returning from the delay of a cache miss, then update the status 919 // to running, otherwise do the cache access. Possibly move this up 920 // to tick() function. 921 if (fetchStatus[tid] == IcacheAccessComplete) { 922 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", 923 tid); 924 925 fetchStatus[tid] = Running; 926 status_change = true; 927 } else if (fetchStatus[tid] == Running) { 928 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 929 "instruction, starting at PC %08p.\n", 930 tid, fetch_PC); 931 932 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid); 933 if (!fetch_success) { 934 ++fetchMiscStallCycles; 935 return; 936 } 937 } else { 938 if (fetchStatus[tid] == Idle) { 939 ++fetchIdleCycles; 940 } else if (fetchStatus[tid] == Blocked) { 941 ++fetchBlockedCycles; 942 } else if (fetchStatus[tid] == Squashing) { 943 ++fetchSquashCycles; 944 } else if (fetchStatus[tid] == IcacheWaitResponse) { 945 ++icacheStallCycles; 946 } 947 948 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so 949 // fetch should do nothing. 950 return; 951 } 952 953 ++fetchCycles; 954 955 // If we had a stall due to an icache miss, then return. 956 if (fetchStatus[tid] == IcacheWaitResponse) { 957 ++icacheStallCycles; 958 status_change = true; 959 return; 960 } 961 962 Addr next_PC = fetch_PC; 963 InstSeqNum inst_seq; 964 MachInst inst; 965 ExtMachInst ext_inst; 966 // @todo: Fix this hack. 967 unsigned offset = (fetch_PC & cacheBlkMask) & ~3; 968 969 if (fault == NoFault) { 970 // If the read of the first instruction was successful, then grab the 971 // instructions from the rest of the cache line and put them into the 972 // queue heading to decode. 973 974 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 975 "decode.\n",tid); 976 977 // Need to keep track of whether or not a predicted branch 978 // ended this fetch block. 979 bool predicted_branch = false; 980 981 for (; 982 offset < cacheBlkSize && 983 numInst < fetchWidth && 984 !predicted_branch; 985 ++numInst) { 986 987 // Get a sequence number. 988 inst_seq = cpu->getAndIncrementInstSeq(); 989 990 // Make sure this is a valid index. 991 assert(offset <= cacheBlkSize - instSize); 992 993 // Get the instruction from the array of the cache line. 994 inst = gtoh(*reinterpret_cast<MachInst *> 995 (&cacheData[tid][offset])); 996 997 ext_inst = TheISA::makeExtMI(inst, fetch_PC); 998 999 // Create a new DynInst from the instruction fetched. 1000 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1001 next_PC, 1002 inst_seq, cpu); 1003 instruction->setThread(tid); 1004 1005 instruction->setASID(tid); 1006 1007 instruction->setState(cpu->thread[tid]); 1008 1009 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created " 1010 "[sn:%lli]\n", 1011 tid, instruction->readPC(), inst_seq); 1012 1013 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", 1014 tid, instruction->staticInst->disassemble(fetch_PC)); 1015 1016 instruction->traceData = 1017 Trace::getInstRecord(curTick, cpu->tcBase(tid), cpu, 1018 instruction->staticInst, 1019 instruction->readPC(),tid); 1020 1021 predicted_branch = lookupAndUpdateNextPC(instruction, next_PC); 1022 1023 // Add instruction to the CPU's list of instructions. 1024 instruction->setInstListIt(cpu->addInst(instruction)); 1025 1026 // Write the instruction to the first slot in the queue 1027 // that heads to decode. 1028 toDecode->insts[numInst] = instruction; 1029 1030 toDecode->size++; 1031 1032 // Increment stat of fetched instructions. 1033 ++fetchedInsts; 1034 1035 // Move to the next instruction, unless we have a branch. 1036 fetch_PC = next_PC; 1037 1038 if (instruction->isQuiesce()) { 1039 warn("%lli: Quiesce instruction encountered, halting fetch!", 1040 curTick); 1041 fetchStatus[tid] = QuiescePending; 1042 ++numInst; 1043 status_change = true; 1044 break; 1045 } 1046 1047 offset+= instSize; 1048 } 1049 } 1050 1051 if (numInst > 0) { 1052 wroteToTimeBuffer = true; 1053 } 1054 1055 // Now that fetching is completed, update the PC to signify what the next 1056 // cycle will be. 1057 if (fault == NoFault) { 1058 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n",tid, next_PC); 1059 1060 PC[tid] = next_PC; 1061 nextPC[tid] = next_PC + instSize; 1062 } else { 1063 // We shouldn't be in an icache miss and also have a fault (an ITB 1064 // miss) 1065 if (fetchStatus[tid] == IcacheWaitResponse) { 1066 panic("Fetch should have exited prior to this!"); 1067 } 1068 1069 // Send the fault to commit. This thread will not do anything 1070 // until commit handles the fault. The only other way it can 1071 // wake up is if a squash comes along and changes the PC. 1072#if FULL_SYSTEM 1073 assert(numInst != fetchWidth); 1074 // Get a sequence number. 1075 inst_seq = cpu->getAndIncrementInstSeq(); 1076 // We will use a nop in order to carry the fault. 1077 ext_inst = TheISA::NoopMachInst; 1078 1079 // Create a new DynInst from the dummy nop. 1080 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1081 next_PC, 1082 inst_seq, cpu); 1083 instruction->setPredTarg(next_PC + instSize); 1084 instruction->setThread(tid); 1085 1086 instruction->setASID(tid); 1087 1088 instruction->setState(cpu->thread[tid]); 1089 1090 instruction->traceData = NULL; 1091 1092 instruction->setInstListIt(cpu->addInst(instruction)); 1093 1094 instruction->fault = fault; 1095 1096 toDecode->insts[numInst] = instruction; 1097 toDecode->size++; 1098 1099 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid); 1100 1101 fetchStatus[tid] = TrapPending; 1102 status_change = true; 1103 1104 warn("%lli fault (%d) detected @ PC %08p", curTick, fault, PC[tid]); 1105#else // !FULL_SYSTEM 1106 warn("%lli fault (%d) detected @ PC %08p", curTick, fault, PC[tid]); 1107#endif // FULL_SYSTEM 1108 } 1109} 1110 1111template<class Impl> 1112void 1113DefaultFetch<Impl>::recvRetry() 1114{ 1115 assert(cacheBlocked); 1116 if (retryPkt != NULL) { 1117 assert(retryTid != -1); 1118 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1119 1120 if (icachePort->sendTiming(retryPkt)) { 1121 fetchStatus[retryTid] = IcacheWaitResponse; 1122 retryPkt = NULL; 1123 retryTid = -1; 1124 cacheBlocked = false; 1125 } 1126 } else { 1127 assert(retryTid == -1); 1128 // Access has been squashed since it was sent out. Just clear 1129 // the cache being blocked. 1130 cacheBlocked = false; 1131 } 1132} 1133 1134/////////////////////////////////////// 1135// // 1136// SMT FETCH POLICY MAINTAINED HERE // 1137// // 1138/////////////////////////////////////// 1139template<class Impl> 1140int 1141DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1142{ 1143 if (numThreads > 1) { 1144 switch (fetch_priority) { 1145 1146 case SingleThread: 1147 return 0; 1148 1149 case RoundRobin: 1150 return roundRobin(); 1151 1152 case IQ: 1153 return iqCount(); 1154 1155 case LSQ: 1156 return lsqCount(); 1157 1158 case Branch: 1159 return branchCount(); 1160 1161 default: 1162 return -1; 1163 } 1164 } else { 1165 int tid = *((*activeThreads).begin()); 1166 1167 if (fetchStatus[tid] == Running || 1168 fetchStatus[tid] == IcacheAccessComplete || 1169 fetchStatus[tid] == Idle) { 1170 return tid; 1171 } else { 1172 return -1; 1173 } 1174 } 1175 1176} 1177 1178 1179template<class Impl> 1180int 1181DefaultFetch<Impl>::roundRobin() 1182{ 1183 list<unsigned>::iterator pri_iter = priorityList.begin(); 1184 list<unsigned>::iterator end = priorityList.end(); 1185 1186 int high_pri; 1187 1188 while (pri_iter != end) { 1189 high_pri = *pri_iter; 1190 1191 assert(high_pri <= numThreads); 1192 1193 if (fetchStatus[high_pri] == Running || 1194 fetchStatus[high_pri] == IcacheAccessComplete || 1195 fetchStatus[high_pri] == Idle) { 1196 1197 priorityList.erase(pri_iter); 1198 priorityList.push_back(high_pri); 1199 1200 return high_pri; 1201 } 1202 1203 pri_iter++; 1204 } 1205 1206 return -1; 1207} 1208 1209template<class Impl> 1210int 1211DefaultFetch<Impl>::iqCount() 1212{ 1213 priority_queue<unsigned> PQ; 1214 1215 list<unsigned>::iterator threads = (*activeThreads).begin(); 1216 1217 while (threads != (*activeThreads).end()) { 1218 unsigned tid = *threads++; 1219 1220 PQ.push(fromIEW->iewInfo[tid].iqCount); 1221 } 1222 1223 while (!PQ.empty()) { 1224 1225 unsigned high_pri = PQ.top(); 1226 1227 if (fetchStatus[high_pri] == Running || 1228 fetchStatus[high_pri] == IcacheAccessComplete || 1229 fetchStatus[high_pri] == Idle) 1230 return high_pri; 1231 else 1232 PQ.pop(); 1233 1234 } 1235 1236 return -1; 1237} 1238 1239template<class Impl> 1240int 1241DefaultFetch<Impl>::lsqCount() 1242{ 1243 priority_queue<unsigned> PQ; 1244 1245 1246 list<unsigned>::iterator threads = (*activeThreads).begin(); 1247 1248 while (threads != (*activeThreads).end()) { 1249 unsigned tid = *threads++; 1250 1251 PQ.push(fromIEW->iewInfo[tid].ldstqCount); 1252 } 1253 1254 while (!PQ.empty()) { 1255 1256 unsigned high_pri = PQ.top(); 1257 1258 if (fetchStatus[high_pri] == Running || 1259 fetchStatus[high_pri] == IcacheAccessComplete || 1260 fetchStatus[high_pri] == Idle) 1261 return high_pri; 1262 else 1263 PQ.pop(); 1264 1265 } 1266 1267 return -1; 1268} 1269 1270template<class Impl> 1271int 1272DefaultFetch<Impl>::branchCount() 1273{ 1274 list<unsigned>::iterator threads = (*activeThreads).begin(); 1275 1276 return *threads; 1277} 1278