fetch_impl.hh revision 2674:6d4afef73a20
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/exetrace.hh" 34#include "cpu/o3/fetch.hh" 35#include "mem/packet.hh" 36#include "mem/request.hh" 37#include "sim/byteswap.hh" 38#include "sim/host.hh" 39#include "sim/root.hh" 40 41#if FULL_SYSTEM 42#include "arch/tlb.hh" 43#include "arch/vtophys.hh" 44#include "base/remote_gdb.hh" 45#include "mem/functional/memory_control.hh" 46#include "mem/functional/physical.hh" 47#include "sim/system.hh" 48#endif // FULL_SYSTEM 49 50#include <algorithm> 51 52using namespace std; 53using namespace TheISA; 54 55template<class Impl> 56Tick 57DefaultFetch<Impl>::IcachePort::recvAtomic(PacketPtr pkt) 58{ 59 panic("DefaultFetch doesn't expect recvAtomic callback!"); 60 return curTick; 61} 62 63template<class Impl> 64void 65DefaultFetch<Impl>::IcachePort::recvFunctional(PacketPtr pkt) 66{ 67 panic("DefaultFetch doesn't expect recvFunctional callback!"); 68} 69 70template<class Impl> 71void 72DefaultFetch<Impl>::IcachePort::recvStatusChange(Status status) 73{ 74 if (status == RangeChange) 75 return; 76 77 panic("DefaultFetch doesn't expect recvStatusChange callback!"); 78} 79 80template<class Impl> 81bool 82DefaultFetch<Impl>::IcachePort::recvTiming(Packet *pkt) 83{ 84 fetch->processCacheCompletion(pkt); 85 return true; 86} 87 88template<class Impl> 89void 90DefaultFetch<Impl>::IcachePort::recvRetry() 91{ 92 panic("DefaultFetch doesn't support retry yet."); 93 // we shouldn't get a retry unless we have a packet that we're 94 // waiting to transmit 95/* 96 assert(cpu->dcache_pkt != NULL); 97 assert(cpu->_status == DcacheRetry); 98 Packet *tmp = cpu->dcache_pkt; 99 if (sendTiming(tmp)) { 100 cpu->_status = DcacheWaitResponse; 101 cpu->dcache_pkt = NULL; 102 } 103*/ 104} 105 106template<class Impl> 107DefaultFetch<Impl>::DefaultFetch(Params *params) 108 : branchPred(params), 109 decodeToFetchDelay(params->decodeToFetchDelay), 110 renameToFetchDelay(params->renameToFetchDelay), 111 iewToFetchDelay(params->iewToFetchDelay), 112 commitToFetchDelay(params->commitToFetchDelay), 113 fetchWidth(params->fetchWidth), 114 numThreads(params->numberOfThreads), 115 numFetchingThreads(params->smtNumFetchingThreads), 116 interruptPending(false) 117{ 118 if (numThreads > Impl::MaxThreads) 119 fatal("numThreads is not a valid value\n"); 120 121 DPRINTF(Fetch, "Fetch constructor called\n"); 122 123 // Set fetch stage's status to inactive. 124 _status = Inactive; 125 126 string policy = params->smtFetchPolicy; 127 128 // Convert string to lowercase 129 std::transform(policy.begin(), policy.end(), policy.begin(), 130 (int(*)(int)) tolower); 131 132 // Figure out fetch policy 133 if (policy == "singlethread") { 134 fetchPolicy = SingleThread; 135 } else if (policy == "roundrobin") { 136 fetchPolicy = RoundRobin; 137 DPRINTF(Fetch, "Fetch policy set to Round Robin\n"); 138 } else if (policy == "branch") { 139 fetchPolicy = Branch; 140 DPRINTF(Fetch, "Fetch policy set to Branch Count\n"); 141 } else if (policy == "iqcount") { 142 fetchPolicy = IQ; 143 DPRINTF(Fetch, "Fetch policy set to IQ count\n"); 144 } else if (policy == "lsqcount") { 145 fetchPolicy = LSQ; 146 DPRINTF(Fetch, "Fetch policy set to LSQ count\n"); 147 } else { 148 fatal("Invalid Fetch Policy. Options Are: {SingleThread," 149 " RoundRobin,LSQcount,IQcount}\n"); 150 } 151 152 // Size of cache block. 153 cacheBlkSize = 64; 154 155 // Create mask to get rid of offset bits. 156 cacheBlkMask = (cacheBlkSize - 1); 157 158 for (int tid=0; tid < numThreads; tid++) { 159 160 fetchStatus[tid] = Running; 161 162 priorityList.push_back(tid); 163 164 memPkt[tid] = NULL; 165 166 // Create space to store a cache line. 167 cacheData[tid] = new uint8_t[cacheBlkSize]; 168 169 stalls[tid].decode = 0; 170 stalls[tid].rename = 0; 171 stalls[tid].iew = 0; 172 stalls[tid].commit = 0; 173 } 174 175 // Get the size of an instruction. 176 instSize = sizeof(MachInst); 177} 178 179template <class Impl> 180std::string 181DefaultFetch<Impl>::name() const 182{ 183 return cpu->name() + ".fetch"; 184} 185 186template <class Impl> 187void 188DefaultFetch<Impl>::regStats() 189{ 190 icacheStallCycles 191 .name(name() + ".icacheStallCycles") 192 .desc("Number of cycles fetch is stalled on an Icache miss") 193 .prereq(icacheStallCycles); 194 195 fetchedInsts 196 .name(name() + ".Insts") 197 .desc("Number of instructions fetch has processed") 198 .prereq(fetchedInsts); 199 200 fetchedBranches 201 .name(name() + ".Branches") 202 .desc("Number of branches that fetch encountered") 203 .prereq(fetchedBranches); 204 205 predictedBranches 206 .name(name() + ".predictedBranches") 207 .desc("Number of branches that fetch has predicted taken") 208 .prereq(predictedBranches); 209 210 fetchCycles 211 .name(name() + ".Cycles") 212 .desc("Number of cycles fetch has run and was not squashing or" 213 " blocked") 214 .prereq(fetchCycles); 215 216 fetchSquashCycles 217 .name(name() + ".SquashCycles") 218 .desc("Number of cycles fetch has spent squashing") 219 .prereq(fetchSquashCycles); 220 221 fetchIdleCycles 222 .name(name() + ".IdleCycles") 223 .desc("Number of cycles fetch was idle") 224 .prereq(fetchIdleCycles); 225 226 fetchBlockedCycles 227 .name(name() + ".BlockedCycles") 228 .desc("Number of cycles fetch has spent blocked") 229 .prereq(fetchBlockedCycles); 230 231 fetchedCacheLines 232 .name(name() + ".CacheLines") 233 .desc("Number of cache lines fetched") 234 .prereq(fetchedCacheLines); 235 236 fetchMiscStallCycles 237 .name(name() + ".MiscStallCycles") 238 .desc("Number of cycles fetch has spent waiting on interrupts, or " 239 "bad addresses, or out of MSHRs") 240 .prereq(fetchMiscStallCycles); 241 242 fetchIcacheSquashes 243 .name(name() + ".IcacheSquashes") 244 .desc("Number of outstanding Icache misses that were squashed") 245 .prereq(fetchIcacheSquashes); 246 247 fetchNisnDist 248 .init(/* base value */ 0, 249 /* last value */ fetchWidth, 250 /* bucket size */ 1) 251 .name(name() + ".rateDist") 252 .desc("Number of instructions fetched each cycle (Total)") 253 .flags(Stats::pdf); 254 255 idleRate 256 .name(name() + ".idleRate") 257 .desc("Percent of cycles fetch was idle") 258 .prereq(idleRate); 259 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 260 261 branchRate 262 .name(name() + ".branchRate") 263 .desc("Number of branch fetches per cycle") 264 .flags(Stats::total); 265 branchRate = fetchedBranches / cpu->numCycles; 266 267 fetchRate 268 .name(name() + ".rate") 269 .desc("Number of inst fetches per cycle") 270 .flags(Stats::total); 271 fetchRate = fetchedInsts / cpu->numCycles; 272 273 branchPred.regStats(); 274} 275 276template<class Impl> 277void 278DefaultFetch<Impl>::setCPU(FullCPU *cpu_ptr) 279{ 280 DPRINTF(Fetch, "Setting the CPU pointer.\n"); 281 cpu = cpu_ptr; 282 283 // Name is finally available, so create the port. 284 icachePort = new IcachePort(this); 285 286 // Fetch needs to start fetching instructions at the very beginning, 287 // so it must start up in active state. 288 switchToActive(); 289} 290 291template<class Impl> 292void 293DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 294{ 295 DPRINTF(Fetch, "Setting the time buffer pointer.\n"); 296 timeBuffer = time_buffer; 297 298 // Create wires to get information from proper places in time buffer. 299 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 300 fromRename = timeBuffer->getWire(-renameToFetchDelay); 301 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 302 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 303} 304 305template<class Impl> 306void 307DefaultFetch<Impl>::setActiveThreads(list<unsigned> *at_ptr) 308{ 309 DPRINTF(Fetch, "Setting active threads list pointer.\n"); 310 activeThreads = at_ptr; 311} 312 313template<class Impl> 314void 315DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 316{ 317 DPRINTF(Fetch, "Setting the fetch queue pointer.\n"); 318 fetchQueue = fq_ptr; 319 320 // Create wire to write information to proper place in fetch queue. 321 toDecode = fetchQueue->getWire(0); 322} 323 324#if 0 325template<class Impl> 326void 327DefaultFetch<Impl>::setPageTable(PageTable *pt_ptr) 328{ 329 DPRINTF(Fetch, "Setting the page table pointer.\n"); 330#if !FULL_SYSTEM 331 pTable = pt_ptr; 332#endif 333} 334#endif 335 336template<class Impl> 337void 338DefaultFetch<Impl>::initStage() 339{ 340 // Setup PC and nextPC with initial state. 341 for (int tid = 0; tid < numThreads; tid++) { 342 PC[tid] = cpu->readPC(tid); 343 nextPC[tid] = cpu->readNextPC(tid); 344 } 345} 346 347template<class Impl> 348void 349DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 350{ 351 unsigned tid = pkt->req->getThreadNum(); 352 353 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid); 354 355 // Only change the status if it's still waiting on the icache access 356 // to return. 357 if (fetchStatus[tid] != IcacheWaitResponse || 358 pkt != memPkt[tid] || 359 isSwitchedOut()) { 360 ++fetchIcacheSquashes; 361 delete pkt; 362 return; 363 } 364 365 // Wake up the CPU (if it went to sleep and was waiting on this completion 366 // event). 367 cpu->wakeCPU(); 368 369 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 370 tid); 371 372 switchToActive(); 373 374 // Only switch to IcacheAccessComplete if we're not stalled as well. 375 if (checkStall(tid)) { 376 fetchStatus[tid] = Blocked; 377 } else { 378 fetchStatus[tid] = IcacheAccessComplete; 379 } 380 381// memcpy(cacheData[tid], memReq[tid]->data, memReq[tid]->size); 382 383 // Reset the mem req to NULL. 384 delete pkt->req; 385 delete pkt; 386 memPkt[tid] = NULL; 387} 388 389template <class Impl> 390void 391DefaultFetch<Impl>::switchOut() 392{ 393 // Fetch is ready to switch out at any time. 394 switchedOut = true; 395 cpu->signalSwitched(); 396} 397 398template <class Impl> 399void 400DefaultFetch<Impl>::doSwitchOut() 401{ 402 // Branch predictor needs to have its state cleared. 403 branchPred.switchOut(); 404} 405 406template <class Impl> 407void 408DefaultFetch<Impl>::takeOverFrom() 409{ 410 // Reset all state 411 for (int i = 0; i < Impl::MaxThreads; ++i) { 412 stalls[i].decode = 0; 413 stalls[i].rename = 0; 414 stalls[i].iew = 0; 415 stalls[i].commit = 0; 416 PC[i] = cpu->readPC(i); 417 nextPC[i] = cpu->readNextPC(i); 418 fetchStatus[i] = Running; 419 } 420 numInst = 0; 421 wroteToTimeBuffer = false; 422 _status = Inactive; 423 switchedOut = false; 424 branchPred.takeOverFrom(); 425} 426 427template <class Impl> 428void 429DefaultFetch<Impl>::wakeFromQuiesce() 430{ 431 DPRINTF(Fetch, "Waking up from quiesce\n"); 432 // Hopefully this is safe 433 // @todo: Allow other threads to wake from quiesce. 434 fetchStatus[0] = Running; 435} 436 437template <class Impl> 438inline void 439DefaultFetch<Impl>::switchToActive() 440{ 441 if (_status == Inactive) { 442 DPRINTF(Activity, "Activating stage.\n"); 443 444 cpu->activateStage(FullCPU::FetchIdx); 445 446 _status = Active; 447 } 448} 449 450template <class Impl> 451inline void 452DefaultFetch<Impl>::switchToInactive() 453{ 454 if (_status == Active) { 455 DPRINTF(Activity, "Deactivating stage.\n"); 456 457 cpu->deactivateStage(FullCPU::FetchIdx); 458 459 _status = Inactive; 460 } 461} 462 463template <class Impl> 464bool 465DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC) 466{ 467 // Do branch prediction check here. 468 // A bit of a misnomer...next_PC is actually the current PC until 469 // this function updates it. 470 bool predict_taken; 471 472 if (!inst->isControl()) { 473 next_PC = next_PC + instSize; 474 inst->setPredTarg(next_PC); 475 return false; 476 } 477 478 predict_taken = branchPred.predict(inst, next_PC, inst->threadNumber); 479 480 ++fetchedBranches; 481 482 if (predict_taken) { 483 ++predictedBranches; 484 } 485 486 return predict_taken; 487} 488 489template <class Impl> 490bool 491DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid) 492{ 493 Fault fault = NoFault; 494 495#if FULL_SYSTEM 496 // Flag to say whether or not address is physical addr. 497 unsigned flags = cpu->inPalMode(fetch_PC) ? PHYSICAL : 0; 498#else 499 unsigned flags = 0; 500#endif // FULL_SYSTEM 501 502 if (interruptPending && flags == 0 || switchedOut) { 503 // Hold off fetch from getting new instructions while an interrupt 504 // is pending. 505 return false; 506 } 507 508 // Align the fetch PC so it's at the start of a cache block. 509 fetch_PC = icacheBlockAlignPC(fetch_PC); 510 511 // Setup the memReq to do a read of the first instruction's address. 512 // Set the appropriate read size and flags as well. 513 // Build request here. 514 RequestPtr mem_req = new Request(tid, fetch_PC, cacheBlkSize, flags, 515 fetch_PC, cpu->readCpuId(), tid); 516 517 memPkt[tid] = NULL; 518 519 // Translate the instruction request. 520//#if FULL_SYSTEM 521 fault = cpu->translateInstReq(mem_req); 522//#else 523// fault = pTable->translate(memReq[tid]); 524//#endif 525 526 // In the case of faults, the fetch stage may need to stall and wait 527 // for the ITB miss to be handled. 528 529 // If translation was successful, attempt to read the first 530 // instruction. 531 if (fault == NoFault) { 532#if FULL_SYSTEM 533 if (cpu->system->memctrl->badaddr(memReq[tid]->paddr) || 534 memReq[tid]->flags & UNCACHEABLE) { 535 DPRINTF(Fetch, "Fetch: Bad address %#x (hopefully on a " 536 "misspeculating path)!", 537 memReq[tid]->paddr); 538 ret_fault = TheISA::genMachineCheckFault(); 539 return false; 540 } 541#endif 542 543 // Build packet here. 544 PacketPtr data_pkt = new Packet(mem_req, 545 Packet::ReadReq, Packet::Broadcast); 546 data_pkt->dataStatic(cacheData[tid]); 547 548 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 549 550 fetchedCacheLines++; 551 552 // Now do the timing access to see whether or not the instruction 553 // exists within the cache. 554 if (!icachePort->sendTiming(data_pkt)) { 555 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 556 ret_fault = NoFault; 557 return false; 558 } 559 560 DPRINTF(Fetch, "Doing cache access.\n"); 561 562 lastIcacheStall[tid] = curTick; 563 564 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 565 "response.\n", tid); 566 567 fetchStatus[tid] = IcacheWaitResponse; 568 } 569 570 ret_fault = fault; 571 return true; 572} 573 574template <class Impl> 575inline void 576DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid) 577{ 578 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x.\n", 579 tid, new_PC); 580 581 PC[tid] = new_PC; 582 nextPC[tid] = new_PC + instSize; 583 584 // Clear the icache miss if it's outstanding. 585 if (fetchStatus[tid] == IcacheWaitResponse) { 586 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 587 tid); 588 delete memPkt[tid]; 589 memPkt[tid] = NULL; 590 } 591 592 fetchStatus[tid] = Squashing; 593 594 ++fetchSquashCycles; 595} 596 597template<class Impl> 598void 599DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, 600 const InstSeqNum &seq_num, 601 unsigned tid) 602{ 603 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid); 604 605 doSquash(new_PC, tid); 606 607 // Tell the CPU to remove any instructions that are in flight between 608 // fetch and decode. 609 cpu->removeInstsUntil(seq_num, tid); 610} 611 612template<class Impl> 613bool 614DefaultFetch<Impl>::checkStall(unsigned tid) const 615{ 616 bool ret_val = false; 617 618 if (cpu->contextSwitch) { 619 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 620 ret_val = true; 621 } else if (stalls[tid].decode) { 622 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 623 ret_val = true; 624 } else if (stalls[tid].rename) { 625 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 626 ret_val = true; 627 } else if (stalls[tid].iew) { 628 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 629 ret_val = true; 630 } else if (stalls[tid].commit) { 631 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 632 ret_val = true; 633 } 634 635 return ret_val; 636} 637 638template<class Impl> 639typename DefaultFetch<Impl>::FetchStatus 640DefaultFetch<Impl>::updateFetchStatus() 641{ 642 //Check Running 643 list<unsigned>::iterator threads = (*activeThreads).begin(); 644 645 while (threads != (*activeThreads).end()) { 646 647 unsigned tid = *threads++; 648 649 if (fetchStatus[tid] == Running || 650 fetchStatus[tid] == Squashing || 651 fetchStatus[tid] == IcacheAccessComplete) { 652 653 if (_status == Inactive) { 654 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 655 656 if (fetchStatus[tid] == IcacheAccessComplete) { 657 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 658 "completion\n",tid); 659 } 660 661 cpu->activateStage(FullCPU::FetchIdx); 662 } 663 664 return Active; 665 } 666 } 667 668 // Stage is switching from active to inactive, notify CPU of it. 669 if (_status == Active) { 670 DPRINTF(Activity, "Deactivating stage.\n"); 671 672 cpu->deactivateStage(FullCPU::FetchIdx); 673 } 674 675 return Inactive; 676} 677 678template <class Impl> 679void 680DefaultFetch<Impl>::squash(const Addr &new_PC, unsigned tid) 681{ 682 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid); 683 684 doSquash(new_PC, tid); 685 686 // Tell the CPU to remove any instructions that are not in the ROB. 687 cpu->removeInstsNotInROB(tid); 688} 689 690template <class Impl> 691void 692DefaultFetch<Impl>::tick() 693{ 694 list<unsigned>::iterator threads = (*activeThreads).begin(); 695 bool status_change = false; 696 697 wroteToTimeBuffer = false; 698 699 while (threads != (*activeThreads).end()) { 700 unsigned tid = *threads++; 701 702 // Check the signals for each thread to determine the proper status 703 // for each thread. 704 bool updated_status = checkSignalsAndUpdate(tid); 705 status_change = status_change || updated_status; 706 } 707 708 DPRINTF(Fetch, "Running stage.\n"); 709 710 // Reset the number of the instruction we're fetching. 711 numInst = 0; 712 713 if (fromCommit->commitInfo[0].interruptPending) { 714 interruptPending = true; 715 } 716 if (fromCommit->commitInfo[0].clearInterrupt) { 717 interruptPending = false; 718 } 719 720 for (threadFetched = 0; threadFetched < numFetchingThreads; 721 threadFetched++) { 722 // Fetch each of the actively fetching threads. 723 fetch(status_change); 724 } 725 726 // Record number of instructions fetched this cycle for distribution. 727 fetchNisnDist.sample(numInst); 728 729 if (status_change) { 730 // Change the fetch stage status if there was a status change. 731 _status = updateFetchStatus(); 732 } 733 734 // If there was activity this cycle, inform the CPU of it. 735 if (wroteToTimeBuffer || cpu->contextSwitch) { 736 DPRINTF(Activity, "Activity this cycle.\n"); 737 738 cpu->activityThisCycle(); 739 } 740} 741 742template <class Impl> 743bool 744DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) 745{ 746 // Update the per thread stall statuses. 747 if (fromDecode->decodeBlock[tid]) { 748 stalls[tid].decode = true; 749 } 750 751 if (fromDecode->decodeUnblock[tid]) { 752 assert(stalls[tid].decode); 753 assert(!fromDecode->decodeBlock[tid]); 754 stalls[tid].decode = false; 755 } 756 757 if (fromRename->renameBlock[tid]) { 758 stalls[tid].rename = true; 759 } 760 761 if (fromRename->renameUnblock[tid]) { 762 assert(stalls[tid].rename); 763 assert(!fromRename->renameBlock[tid]); 764 stalls[tid].rename = false; 765 } 766 767 if (fromIEW->iewBlock[tid]) { 768 stalls[tid].iew = true; 769 } 770 771 if (fromIEW->iewUnblock[tid]) { 772 assert(stalls[tid].iew); 773 assert(!fromIEW->iewBlock[tid]); 774 stalls[tid].iew = false; 775 } 776 777 if (fromCommit->commitBlock[tid]) { 778 stalls[tid].commit = true; 779 } 780 781 if (fromCommit->commitUnblock[tid]) { 782 assert(stalls[tid].commit); 783 assert(!fromCommit->commitBlock[tid]); 784 stalls[tid].commit = false; 785 } 786 787 // Check squash signals from commit. 788 if (fromCommit->commitInfo[tid].squash) { 789 790 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 791 "from commit.\n",tid); 792 793 // In any case, squash. 794 squash(fromCommit->commitInfo[tid].nextPC,tid); 795 796 // Also check if there's a mispredict that happened. 797 if (fromCommit->commitInfo[tid].branchMispredict) { 798 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 799 fromCommit->commitInfo[tid].nextPC, 800 fromCommit->commitInfo[tid].branchTaken, 801 tid); 802 } else { 803 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 804 tid); 805 } 806 807 return true; 808 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 809 // Update the branch predictor if it wasn't a squashed instruction 810 // that was broadcasted. 811 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 812 } 813 814 // Check ROB squash signals from commit. 815 if (fromCommit->commitInfo[tid].robSquashing) { 816 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing Thread %u.\n", tid); 817 818 // Continue to squash. 819 fetchStatus[tid] = Squashing; 820 821 return true; 822 } 823 824 // Check squash signals from decode. 825 if (fromDecode->decodeInfo[tid].squash) { 826 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 827 "from decode.\n",tid); 828 829 // Update the branch predictor. 830 if (fromDecode->decodeInfo[tid].branchMispredict) { 831 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 832 fromDecode->decodeInfo[tid].nextPC, 833 fromDecode->decodeInfo[tid].branchTaken, 834 tid); 835 } else { 836 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 837 tid); 838 } 839 840 if (fetchStatus[tid] != Squashing) { 841 // Squash unless we're already squashing 842 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 843 fromDecode->decodeInfo[tid].doneSeqNum, 844 tid); 845 846 return true; 847 } 848 } 849 850 if (checkStall(tid) && fetchStatus[tid] != IcacheWaitResponse) { 851 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 852 853 fetchStatus[tid] = Blocked; 854 855 return true; 856 } 857 858 if (fetchStatus[tid] == Blocked || 859 fetchStatus[tid] == Squashing) { 860 // Switch status to running if fetch isn't being told to block or 861 // squash this cycle. 862 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 863 tid); 864 865 fetchStatus[tid] = Running; 866 867 return true; 868 } 869 870 // If we've reached this point, we have not gotten any signals that 871 // cause fetch to change its status. Fetch remains the same as before. 872 return false; 873} 874 875template<class Impl> 876void 877DefaultFetch<Impl>::fetch(bool &status_change) 878{ 879 ////////////////////////////////////////// 880 // Start actual fetch 881 ////////////////////////////////////////// 882 int tid = getFetchingThread(fetchPolicy); 883 884 if (tid == -1) { 885 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 886 887 // Breaks looping condition in tick() 888 threadFetched = numFetchingThreads; 889 return; 890 } 891 892 // The current PC. 893 Addr &fetch_PC = PC[tid]; 894 895 // Fault code for memory access. 896 Fault fault = NoFault; 897 898 // If returning from the delay of a cache miss, then update the status 899 // to running, otherwise do the cache access. Possibly move this up 900 // to tick() function. 901 if (fetchStatus[tid] == IcacheAccessComplete) { 902 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", 903 tid); 904 905 fetchStatus[tid] = Running; 906 status_change = true; 907 } else if (fetchStatus[tid] == Running) { 908 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 909 "instruction, starting at PC %08p.\n", 910 tid, fetch_PC); 911 912 bool fetch_success = fetchCacheLine(fetch_PC, fault, tid); 913 if (!fetch_success) { 914 ++fetchMiscStallCycles; 915 return; 916 } 917 } else { 918 if (fetchStatus[tid] == Idle) { 919 ++fetchIdleCycles; 920 } else if (fetchStatus[tid] == Blocked) { 921 ++fetchBlockedCycles; 922 } else if (fetchStatus[tid] == Squashing) { 923 ++fetchSquashCycles; 924 } else if (fetchStatus[tid] == IcacheWaitResponse) { 925 ++icacheStallCycles; 926 } 927 928 // Status is Idle, Squashing, Blocked, or IcacheWaitResponse, so 929 // fetch should do nothing. 930 return; 931 } 932 933 ++fetchCycles; 934 935 // If we had a stall due to an icache miss, then return. 936 if (fetchStatus[tid] == IcacheWaitResponse) { 937 ++icacheStallCycles; 938 status_change = true; 939 return; 940 } 941 942 Addr next_PC = fetch_PC; 943 InstSeqNum inst_seq; 944 MachInst inst; 945 ExtMachInst ext_inst; 946 // @todo: Fix this hack. 947 unsigned offset = (fetch_PC & cacheBlkMask) & ~3; 948 949 if (fault == NoFault) { 950 // If the read of the first instruction was successful, then grab the 951 // instructions from the rest of the cache line and put them into the 952 // queue heading to decode. 953 954 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 955 "decode.\n",tid); 956 957 // Need to keep track of whether or not a predicted branch 958 // ended this fetch block. 959 bool predicted_branch = false; 960 961 for (; 962 offset < cacheBlkSize && 963 numInst < fetchWidth && 964 !predicted_branch; 965 ++numInst) { 966 967 // Get a sequence number. 968 inst_seq = cpu->getAndIncrementInstSeq(); 969 970 // Make sure this is a valid index. 971 assert(offset <= cacheBlkSize - instSize); 972 973 // Get the instruction from the array of the cache line. 974 inst = gtoh(*reinterpret_cast<MachInst *> 975 (&cacheData[tid][offset])); 976 977 ext_inst = TheISA::makeExtMI(inst, fetch_PC); 978 979 // Create a new DynInst from the instruction fetched. 980 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 981 next_PC, 982 inst_seq, cpu); 983 instruction->setThread(tid); 984 985 instruction->setASID(tid); 986 987 instruction->setState(cpu->thread[tid]); 988 989 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x created " 990 "[sn:%lli]\n", 991 tid, instruction->readPC(), inst_seq); 992 993 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", 994 tid, instruction->staticInst->disassemble(fetch_PC)); 995 996 instruction->traceData = 997 Trace::getInstRecord(curTick, cpu->xcBase(tid), cpu, 998 instruction->staticInst, 999 instruction->readPC(),tid); 1000 1001 predicted_branch = lookupAndUpdateNextPC(instruction, next_PC); 1002 1003 // Add instruction to the CPU's list of instructions. 1004 instruction->setInstListIt(cpu->addInst(instruction)); 1005 1006 // Write the instruction to the first slot in the queue 1007 // that heads to decode. 1008 toDecode->insts[numInst] = instruction; 1009 1010 toDecode->size++; 1011 1012 // Increment stat of fetched instructions. 1013 ++fetchedInsts; 1014 1015 // Move to the next instruction, unless we have a branch. 1016 fetch_PC = next_PC; 1017 1018 if (instruction->isQuiesce()) { 1019 warn("%lli: Quiesce instruction encountered, halting fetch!", 1020 curTick); 1021 fetchStatus[tid] = QuiescePending; 1022 ++numInst; 1023 status_change = true; 1024 break; 1025 } 1026 1027 offset+= instSize; 1028 } 1029 } 1030 1031 if (numInst > 0) { 1032 wroteToTimeBuffer = true; 1033 } 1034 1035 // Now that fetching is completed, update the PC to signify what the next 1036 // cycle will be. 1037 if (fault == NoFault) { 1038 DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n",tid, next_PC); 1039 1040 PC[tid] = next_PC; 1041 nextPC[tid] = next_PC + instSize; 1042 } else { 1043 // We shouldn't be in an icache miss and also have a fault (an ITB 1044 // miss) 1045 if (fetchStatus[tid] == IcacheWaitResponse) { 1046 panic("Fetch should have exited prior to this!"); 1047 } 1048 1049 // Send the fault to commit. This thread will not do anything 1050 // until commit handles the fault. The only other way it can 1051 // wake up is if a squash comes along and changes the PC. 1052#if FULL_SYSTEM 1053 assert(numInst != fetchWidth); 1054 // Get a sequence number. 1055 inst_seq = cpu->getAndIncrementInstSeq(); 1056 // We will use a nop in order to carry the fault. 1057 ext_inst = TheISA::NoopMachInst; 1058 1059 // Create a new DynInst from the dummy nop. 1060 DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, 1061 next_PC, 1062 inst_seq, cpu); 1063 instruction->setPredTarg(next_PC + instSize); 1064 instruction->setThread(tid); 1065 1066 instruction->setASID(tid); 1067 1068 instruction->setState(cpu->thread[tid]); 1069 1070 instruction->traceData = NULL; 1071 1072 instruction->setInstListIt(cpu->addInst(instruction)); 1073 1074 instruction->fault = fault; 1075 1076 toDecode->insts[numInst] = instruction; 1077 toDecode->size++; 1078 1079 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n",tid); 1080 1081 fetchStatus[tid] = TrapPending; 1082 status_change = true; 1083 1084 warn("%lli fault (%d) detected @ PC %08p", curTick, fault, PC[tid]); 1085#else // !FULL_SYSTEM 1086 fatal("fault (%d) detected @ PC %08p", fault, PC[tid]); 1087#endif // FULL_SYSTEM 1088 } 1089} 1090 1091 1092/////////////////////////////////////// 1093// // 1094// SMT FETCH POLICY MAINTAINED HERE // 1095// // 1096/////////////////////////////////////// 1097template<class Impl> 1098int 1099DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1100{ 1101 if (numThreads > 1) { 1102 switch (fetch_priority) { 1103 1104 case SingleThread: 1105 return 0; 1106 1107 case RoundRobin: 1108 return roundRobin(); 1109 1110 case IQ: 1111 return iqCount(); 1112 1113 case LSQ: 1114 return lsqCount(); 1115 1116 case Branch: 1117 return branchCount(); 1118 1119 default: 1120 return -1; 1121 } 1122 } else { 1123 int tid = *((*activeThreads).begin()); 1124 1125 if (fetchStatus[tid] == Running || 1126 fetchStatus[tid] == IcacheAccessComplete || 1127 fetchStatus[tid] == Idle) { 1128 return tid; 1129 } else { 1130 return -1; 1131 } 1132 } 1133 1134} 1135 1136 1137template<class Impl> 1138int 1139DefaultFetch<Impl>::roundRobin() 1140{ 1141 list<unsigned>::iterator pri_iter = priorityList.begin(); 1142 list<unsigned>::iterator end = priorityList.end(); 1143 1144 int high_pri; 1145 1146 while (pri_iter != end) { 1147 high_pri = *pri_iter; 1148 1149 assert(high_pri <= numThreads); 1150 1151 if (fetchStatus[high_pri] == Running || 1152 fetchStatus[high_pri] == IcacheAccessComplete || 1153 fetchStatus[high_pri] == Idle) { 1154 1155 priorityList.erase(pri_iter); 1156 priorityList.push_back(high_pri); 1157 1158 return high_pri; 1159 } 1160 1161 pri_iter++; 1162 } 1163 1164 return -1; 1165} 1166 1167template<class Impl> 1168int 1169DefaultFetch<Impl>::iqCount() 1170{ 1171 priority_queue<unsigned> PQ; 1172 1173 list<unsigned>::iterator threads = (*activeThreads).begin(); 1174 1175 while (threads != (*activeThreads).end()) { 1176 unsigned tid = *threads++; 1177 1178 PQ.push(fromIEW->iewInfo[tid].iqCount); 1179 } 1180 1181 while (!PQ.empty()) { 1182 1183 unsigned high_pri = PQ.top(); 1184 1185 if (fetchStatus[high_pri] == Running || 1186 fetchStatus[high_pri] == IcacheAccessComplete || 1187 fetchStatus[high_pri] == Idle) 1188 return high_pri; 1189 else 1190 PQ.pop(); 1191 1192 } 1193 1194 return -1; 1195} 1196 1197template<class Impl> 1198int 1199DefaultFetch<Impl>::lsqCount() 1200{ 1201 priority_queue<unsigned> PQ; 1202 1203 1204 list<unsigned>::iterator threads = (*activeThreads).begin(); 1205 1206 while (threads != (*activeThreads).end()) { 1207 unsigned tid = *threads++; 1208 1209 PQ.push(fromIEW->iewInfo[tid].ldstqCount); 1210 } 1211 1212 while (!PQ.empty()) { 1213 1214 unsigned high_pri = PQ.top(); 1215 1216 if (fetchStatus[high_pri] == Running || 1217 fetchStatus[high_pri] == IcacheAccessComplete || 1218 fetchStatus[high_pri] == Idle) 1219 return high_pri; 1220 else 1221 PQ.pop(); 1222 1223 } 1224 1225 return -1; 1226} 1227 1228template<class Impl> 1229int 1230DefaultFetch<Impl>::branchCount() 1231{ 1232 list<unsigned>::iterator threads = (*activeThreads).begin(); 1233 1234 return *threads; 1235} 1236