fetch_impl.hh revision 8799:dac1e33e07b0
1/* 2 * Copyright (c) 2010-2011 ARM Limited 3 * All rights reserved. 4 * 5 * The license below extends only to copyright in the software and shall 6 * not be construed as granting a license to any other intellectual 7 * property including but not limited to intellectual property relating 8 * to a hardware implementation of the functionality of the software 9 * licensed hereunder. You may use the software subject to the license 10 * terms below provided that you ensure that this notice is replicated 11 * unmodified and in its entirety in all distributions of the software, 12 * modified or unmodified, in source code or in binary form. 13 * 14 * Copyright (c) 2004-2006 The Regents of The University of Michigan 15 * All rights reserved. 16 * 17 * Redistribution and use in source and binary forms, with or without 18 * modification, are permitted provided that the following conditions are 19 * met: redistributions of source code must retain the above copyright 20 * notice, this list of conditions and the following disclaimer; 21 * redistributions in binary form must reproduce the above copyright 22 * notice, this list of conditions and the following disclaimer in the 23 * documentation and/or other materials provided with the distribution; 24 * neither the name of the copyright holders nor the names of its 25 * contributors may be used to endorse or promote products derived from 26 * this software without specific prior written permission. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 39 * 40 * Authors: Kevin Lim 41 * Korey Sewell 42 */ 43 44#include <algorithm> 45#include <cstring> 46 47#include "arch/isa_traits.hh" 48#include "arch/tlb.hh" 49#include "arch/utility.hh" 50#include "arch/vtophys.hh" 51#include "base/types.hh" 52#include "config/the_isa.hh" 53#include "config/use_checker.hh" 54#include "cpu/base.hh" 55#include "cpu/checker/cpu.hh" 56#include "cpu/o3/fetch.hh" 57#include "cpu/exetrace.hh" 58#include "debug/Activity.hh" 59#include "debug/Fetch.hh" 60#include "mem/packet.hh" 61#include "mem/request.hh" 62#include "params/DerivO3CPU.hh" 63#include "sim/byteswap.hh" 64#include "sim/core.hh" 65#include "sim/eventq.hh" 66#include "sim/full_system.hh" 67#include "sim/system.hh" 68 69using namespace std; 70 71template<class Impl> 72DefaultFetch<Impl>::DefaultFetch(O3CPU *_cpu, DerivO3CPUParams *params) 73 : cpu(_cpu), 74 branchPred(params), 75 predecoder(NULL), 76 numInst(0), 77 decodeToFetchDelay(params->decodeToFetchDelay), 78 renameToFetchDelay(params->renameToFetchDelay), 79 iewToFetchDelay(params->iewToFetchDelay), 80 commitToFetchDelay(params->commitToFetchDelay), 81 fetchWidth(params->fetchWidth), 82 cacheBlocked(false), 83 retryPkt(NULL), 84 retryTid(InvalidThreadID), 85 numThreads(params->numThreads), 86 numFetchingThreads(params->smtNumFetchingThreads), 87 interruptPending(false), 88 drainPending(false), 89 switchedOut(false), 90 finishTranslationEvent(this) 91{ 92 if (numThreads > Impl::MaxThreads) 93 fatal("numThreads (%d) is larger than compiled limit (%d),\n" 94 "\tincrease MaxThreads in src/cpu/o3/impl.hh\n", 95 numThreads, static_cast<int>(Impl::MaxThreads)); 96 97 // Set fetch stage's status to inactive. 98 _status = Inactive; 99 100 std::string policy = params->smtFetchPolicy; 101 102 // Convert string to lowercase 103 std::transform(policy.begin(), policy.end(), policy.begin(), 104 (int(*)(int)) tolower); 105 106 // Figure out fetch policy 107 if (policy == "singlethread") { 108 fetchPolicy = SingleThread; 109 if (numThreads > 1) 110 panic("Invalid Fetch Policy for a SMT workload."); 111 } else if (policy == "roundrobin") { 112 fetchPolicy = RoundRobin; 113 DPRINTF(Fetch, "Fetch policy set to Round Robin\n"); 114 } else if (policy == "branch") { 115 fetchPolicy = Branch; 116 DPRINTF(Fetch, "Fetch policy set to Branch Count\n"); 117 } else if (policy == "iqcount") { 118 fetchPolicy = IQ; 119 DPRINTF(Fetch, "Fetch policy set to IQ count\n"); 120 } else if (policy == "lsqcount") { 121 fetchPolicy = LSQ; 122 DPRINTF(Fetch, "Fetch policy set to LSQ count\n"); 123 } else { 124 fatal("Invalid Fetch Policy. Options Are: {SingleThread," 125 " RoundRobin,LSQcount,IQcount}\n"); 126 } 127 128 // Get the size of an instruction. 129 instSize = sizeof(TheISA::MachInst); 130} 131 132template <class Impl> 133std::string 134DefaultFetch<Impl>::name() const 135{ 136 return cpu->name() + ".fetch"; 137} 138 139template <class Impl> 140void 141DefaultFetch<Impl>::regStats() 142{ 143 icacheStallCycles 144 .name(name() + ".icacheStallCycles") 145 .desc("Number of cycles fetch is stalled on an Icache miss") 146 .prereq(icacheStallCycles); 147 148 fetchedInsts 149 .name(name() + ".Insts") 150 .desc("Number of instructions fetch has processed") 151 .prereq(fetchedInsts); 152 153 fetchedBranches 154 .name(name() + ".Branches") 155 .desc("Number of branches that fetch encountered") 156 .prereq(fetchedBranches); 157 158 predictedBranches 159 .name(name() + ".predictedBranches") 160 .desc("Number of branches that fetch has predicted taken") 161 .prereq(predictedBranches); 162 163 fetchCycles 164 .name(name() + ".Cycles") 165 .desc("Number of cycles fetch has run and was not squashing or" 166 " blocked") 167 .prereq(fetchCycles); 168 169 fetchSquashCycles 170 .name(name() + ".SquashCycles") 171 .desc("Number of cycles fetch has spent squashing") 172 .prereq(fetchSquashCycles); 173 174 fetchTlbCycles 175 .name(name() + ".TlbCycles") 176 .desc("Number of cycles fetch has spent waiting for tlb") 177 .prereq(fetchTlbCycles); 178 179 fetchIdleCycles 180 .name(name() + ".IdleCycles") 181 .desc("Number of cycles fetch was idle") 182 .prereq(fetchIdleCycles); 183 184 fetchBlockedCycles 185 .name(name() + ".BlockedCycles") 186 .desc("Number of cycles fetch has spent blocked") 187 .prereq(fetchBlockedCycles); 188 189 fetchedCacheLines 190 .name(name() + ".CacheLines") 191 .desc("Number of cache lines fetched") 192 .prereq(fetchedCacheLines); 193 194 fetchMiscStallCycles 195 .name(name() + ".MiscStallCycles") 196 .desc("Number of cycles fetch has spent waiting on interrupts, or " 197 "bad addresses, or out of MSHRs") 198 .prereq(fetchMiscStallCycles); 199 200 fetchPendingDrainCycles 201 .name(name() + ".PendingDrainCycles") 202 .desc("Number of cycles fetch has spent waiting on pipes to drain") 203 .prereq(fetchPendingDrainCycles); 204 205 fetchNoActiveThreadStallCycles 206 .name(name() + ".NoActiveThreadStallCycles") 207 .desc("Number of stall cycles due to no active thread to fetch from") 208 .prereq(fetchNoActiveThreadStallCycles); 209 210 fetchPendingTrapStallCycles 211 .name(name() + ".PendingTrapStallCycles") 212 .desc("Number of stall cycles due to pending traps") 213 .prereq(fetchPendingTrapStallCycles); 214 215 fetchPendingQuiesceStallCycles 216 .name(name() + ".PendingQuiesceStallCycles") 217 .desc("Number of stall cycles due to pending quiesce instructions") 218 .prereq(fetchPendingQuiesceStallCycles); 219 220 fetchIcacheWaitRetryStallCycles 221 .name(name() + ".IcacheWaitRetryStallCycles") 222 .desc("Number of stall cycles due to full MSHR") 223 .prereq(fetchIcacheWaitRetryStallCycles); 224 225 fetchIcacheSquashes 226 .name(name() + ".IcacheSquashes") 227 .desc("Number of outstanding Icache misses that were squashed") 228 .prereq(fetchIcacheSquashes); 229 230 fetchTlbSquashes 231 .name(name() + ".ItlbSquashes") 232 .desc("Number of outstanding ITLB misses that were squashed") 233 .prereq(fetchTlbSquashes); 234 235 fetchNisnDist 236 .init(/* base value */ 0, 237 /* last value */ fetchWidth, 238 /* bucket size */ 1) 239 .name(name() + ".rateDist") 240 .desc("Number of instructions fetched each cycle (Total)") 241 .flags(Stats::pdf); 242 243 idleRate 244 .name(name() + ".idleRate") 245 .desc("Percent of cycles fetch was idle") 246 .prereq(idleRate); 247 idleRate = fetchIdleCycles * 100 / cpu->numCycles; 248 249 branchRate 250 .name(name() + ".branchRate") 251 .desc("Number of branch fetches per cycle") 252 .flags(Stats::total); 253 branchRate = fetchedBranches / cpu->numCycles; 254 255 fetchRate 256 .name(name() + ".rate") 257 .desc("Number of inst fetches per cycle") 258 .flags(Stats::total); 259 fetchRate = fetchedInsts / cpu->numCycles; 260 261 branchPred.regStats(); 262} 263 264template<class Impl> 265void 266DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer) 267{ 268 timeBuffer = time_buffer; 269 270 // Create wires to get information from proper places in time buffer. 271 fromDecode = timeBuffer->getWire(-decodeToFetchDelay); 272 fromRename = timeBuffer->getWire(-renameToFetchDelay); 273 fromIEW = timeBuffer->getWire(-iewToFetchDelay); 274 fromCommit = timeBuffer->getWire(-commitToFetchDelay); 275} 276 277template<class Impl> 278void 279DefaultFetch<Impl>::setActiveThreads(std::list<ThreadID> *at_ptr) 280{ 281 activeThreads = at_ptr; 282} 283 284template<class Impl> 285void 286DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr) 287{ 288 fetchQueue = fq_ptr; 289 290 // Create wire to write information to proper place in fetch queue. 291 toDecode = fetchQueue->getWire(0); 292} 293 294template<class Impl> 295void 296DefaultFetch<Impl>::initStage() 297{ 298 // Setup PC and nextPC with initial state. 299 for (ThreadID tid = 0; tid < numThreads; tid++) { 300 pc[tid] = cpu->pcState(tid); 301 fetchOffset[tid] = 0; 302 macroop[tid] = NULL; 303 delayedCommit[tid] = false; 304 } 305 306 for (ThreadID tid = 0; tid < numThreads; tid++) { 307 308 fetchStatus[tid] = Running; 309 310 priorityList.push_back(tid); 311 312 memReq[tid] = NULL; 313 314 stalls[tid].decode = false; 315 stalls[tid].rename = false; 316 stalls[tid].iew = false; 317 stalls[tid].commit = false; 318 } 319 320 // Schedule fetch to get the correct PC from the CPU 321 // scheduleFetchStartupEvent(1); 322 323 // Fetch needs to start fetching instructions at the very beginning, 324 // so it must start up in active state. 325 switchToActive(); 326} 327 328template<class Impl> 329void 330DefaultFetch<Impl>::setIcache() 331{ 332 assert(cpu->getIcachePort()->isConnected()); 333 334 // Size of cache block. 335 cacheBlkSize = cpu->getIcachePort()->peerBlockSize(); 336 337 // Create mask to get rid of offset bits. 338 cacheBlkMask = (cacheBlkSize - 1); 339 340 for (ThreadID tid = 0; tid < numThreads; tid++) { 341 // Create space to store a cache line. 342 cacheData[tid] = new uint8_t[cacheBlkSize]; 343 cacheDataPC[tid] = 0; 344 cacheDataValid[tid] = false; 345 } 346} 347 348template<class Impl> 349void 350DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt) 351{ 352 ThreadID tid = pkt->req->threadId(); 353 354 DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n", tid); 355 356 assert(!pkt->wasNacked()); 357 358 // Only change the status if it's still waiting on the icache access 359 // to return. 360 if (fetchStatus[tid] != IcacheWaitResponse || 361 pkt->req != memReq[tid] || 362 isSwitchedOut()) { 363 ++fetchIcacheSquashes; 364 delete pkt->req; 365 delete pkt; 366 return; 367 } 368 369 memcpy(cacheData[tid], pkt->getPtr<uint8_t>(), cacheBlkSize); 370 cacheDataValid[tid] = true; 371 372 if (!drainPending) { 373 // Wake up the CPU (if it went to sleep and was waiting on 374 // this completion event). 375 cpu->wakeCPU(); 376 377 DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n", 378 tid); 379 380 switchToActive(); 381 } 382 383 // Only switch to IcacheAccessComplete if we're not stalled as well. 384 if (checkStall(tid)) { 385 fetchStatus[tid] = Blocked; 386 } else { 387 fetchStatus[tid] = IcacheAccessComplete; 388 } 389 390 // Reset the mem req to NULL. 391 delete pkt->req; 392 delete pkt; 393 memReq[tid] = NULL; 394} 395 396template <class Impl> 397bool 398DefaultFetch<Impl>::drain() 399{ 400 // Fetch is ready to drain at any time. 401 cpu->signalDrained(); 402 drainPending = true; 403 return true; 404} 405 406template <class Impl> 407void 408DefaultFetch<Impl>::resume() 409{ 410 drainPending = false; 411} 412 413template <class Impl> 414void 415DefaultFetch<Impl>::switchOut() 416{ 417 switchedOut = true; 418 // Branch predictor needs to have its state cleared. 419 branchPred.switchOut(); 420} 421 422template <class Impl> 423void 424DefaultFetch<Impl>::takeOverFrom() 425{ 426 // the instruction port is now connected so we can get the block 427 // size 428 setIcache(); 429 430 // Reset all state 431 for (ThreadID i = 0; i < Impl::MaxThreads; ++i) { 432 stalls[i].decode = 0; 433 stalls[i].rename = 0; 434 stalls[i].iew = 0; 435 stalls[i].commit = 0; 436 pc[i] = cpu->pcState(i); 437 fetchStatus[i] = Running; 438 } 439 numInst = 0; 440 wroteToTimeBuffer = false; 441 _status = Inactive; 442 switchedOut = false; 443 interruptPending = false; 444 branchPred.takeOverFrom(); 445} 446 447template <class Impl> 448void 449DefaultFetch<Impl>::wakeFromQuiesce() 450{ 451 DPRINTF(Fetch, "Waking up from quiesce\n"); 452 // Hopefully this is safe 453 // @todo: Allow other threads to wake from quiesce. 454 fetchStatus[0] = Running; 455} 456 457template <class Impl> 458inline void 459DefaultFetch<Impl>::switchToActive() 460{ 461 if (_status == Inactive) { 462 DPRINTF(Activity, "Activating stage.\n"); 463 464 cpu->activateStage(O3CPU::FetchIdx); 465 466 _status = Active; 467 } 468} 469 470template <class Impl> 471inline void 472DefaultFetch<Impl>::switchToInactive() 473{ 474 if (_status == Active) { 475 DPRINTF(Activity, "Deactivating stage.\n"); 476 477 cpu->deactivateStage(O3CPU::FetchIdx); 478 479 _status = Inactive; 480 } 481} 482 483template <class Impl> 484bool 485DefaultFetch<Impl>::lookupAndUpdateNextPC( 486 DynInstPtr &inst, TheISA::PCState &nextPC) 487{ 488 // Do branch prediction check here. 489 // A bit of a misnomer...next_PC is actually the current PC until 490 // this function updates it. 491 bool predict_taken; 492 493 if (!inst->isControl()) { 494 TheISA::advancePC(nextPC, inst->staticInst); 495 inst->setPredTarg(nextPC); 496 inst->setPredTaken(false); 497 return false; 498 } 499 500 ThreadID tid = inst->threadNumber; 501 predict_taken = branchPred.predict(inst, nextPC, tid); 502 503 if (predict_taken) { 504 DPRINTF(Fetch, "[tid:%i]: [sn:%i]: Branch predicted to be taken to %s.\n", 505 tid, inst->seqNum, nextPC); 506 } else { 507 DPRINTF(Fetch, "[tid:%i]: [sn:%i]:Branch predicted to be not taken.\n", 508 tid, inst->seqNum); 509 } 510 511 DPRINTF(Fetch, "[tid:%i]: [sn:%i] Branch predicted to go to %s.\n", 512 tid, inst->seqNum, nextPC); 513 inst->setPredTarg(nextPC); 514 inst->setPredTaken(predict_taken); 515 516 ++fetchedBranches; 517 518 if (predict_taken) { 519 ++predictedBranches; 520 } 521 522 return predict_taken; 523} 524 525template <class Impl> 526bool 527DefaultFetch<Impl>::fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc) 528{ 529 Fault fault = NoFault; 530 531 // @todo: not sure if these should block translation. 532 //AlphaDep 533 if (cacheBlocked) { 534 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, cache blocked\n", 535 tid); 536 return false; 537 } else if (isSwitchedOut()) { 538 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, switched out\n", 539 tid); 540 return false; 541 } else if (checkInterrupt(pc)) { 542 // Hold off fetch from getting new instructions when: 543 // Cache is blocked, or 544 // while an interrupt is pending and we're not in PAL mode, or 545 // fetch is switched out. 546 DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, interrupt pending\n", 547 tid); 548 return false; 549 } 550 551 // Align the fetch address so it's at the start of a cache block. 552 Addr block_PC = icacheBlockAlignPC(vaddr); 553 554 DPRINTF(Fetch, "[tid:%i] Fetching cache line %#x for addr %#x\n", 555 tid, block_PC, vaddr); 556 557 // Setup the memReq to do a read of the first instruction's address. 558 // Set the appropriate read size and flags as well. 559 // Build request here. 560 RequestPtr mem_req = 561 new Request(tid, block_PC, cacheBlkSize, Request::INST_FETCH, 562 pc, cpu->thread[tid]->contextId(), tid); 563 564 memReq[tid] = mem_req; 565 566 // Initiate translation of the icache block 567 fetchStatus[tid] = ItlbWait; 568 FetchTranslation *trans = new FetchTranslation(this); 569 cpu->itb->translateTiming(mem_req, cpu->thread[tid]->getTC(), 570 trans, BaseTLB::Execute); 571 return true; 572} 573 574template <class Impl> 575void 576DefaultFetch<Impl>::finishTranslation(Fault fault, RequestPtr mem_req) 577{ 578 ThreadID tid = mem_req->threadId(); 579 Addr block_PC = mem_req->getVaddr(); 580 581 // Wake up CPU if it was idle 582 cpu->wakeCPU(); 583 584 if (fetchStatus[tid] != ItlbWait || mem_req != memReq[tid] || 585 mem_req->getVaddr() != memReq[tid]->getVaddr() || isSwitchedOut()) { 586 DPRINTF(Fetch, "[tid:%i] Ignoring itlb completed after squash\n", 587 tid); 588 ++fetchTlbSquashes; 589 delete mem_req; 590 return; 591 } 592 593 594 // If translation was successful, attempt to read the icache block. 595 if (fault == NoFault) { 596 // Check that we're not going off into random memory 597 // If we have, just wait around for commit to squash something and put 598 // us on the right track 599 if (!cpu->system->isMemory(mem_req->getPaddr())) { 600 warn("Address %#x is outside of physical memory, stopping fetch\n", 601 mem_req->getPaddr()); 602 fetchStatus[tid] = NoGoodAddr; 603 delete mem_req; 604 memReq[tid] = NULL; 605 return; 606 } 607 608 // Build packet here. 609 PacketPtr data_pkt = new Packet(mem_req, 610 MemCmd::ReadReq, Packet::Broadcast); 611 data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]); 612 613 cacheDataPC[tid] = block_PC; 614 cacheDataValid[tid] = false; 615 DPRINTF(Fetch, "Fetch: Doing instruction read.\n"); 616 617 fetchedCacheLines++; 618 619 // Access the cache. 620 if (!cpu->getIcachePort()->sendTiming(data_pkt)) { 621 assert(retryPkt == NULL); 622 assert(retryTid == InvalidThreadID); 623 DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid); 624 625 fetchStatus[tid] = IcacheWaitRetry; 626 retryPkt = data_pkt; 627 retryTid = tid; 628 cacheBlocked = true; 629 } else { 630 DPRINTF(Fetch, "[tid:%i]: Doing Icache access.\n", tid); 631 DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache " 632 "response.\n", tid); 633 634 lastIcacheStall[tid] = curTick(); 635 fetchStatus[tid] = IcacheWaitResponse; 636 } 637 } else { 638 if (!(numInst < fetchWidth)) { 639 assert(!finishTranslationEvent.scheduled()); 640 finishTranslationEvent.setFault(fault); 641 finishTranslationEvent.setReq(mem_req); 642 cpu->schedule(finishTranslationEvent, cpu->nextCycle(curTick() + cpu->ticks(1))); 643 return; 644 } 645 DPRINTF(Fetch, "[tid:%i] Got back req with addr %#x but expected %#x\n", 646 tid, mem_req->getVaddr(), memReq[tid]->getVaddr()); 647 // Translation faulted, icache request won't be sent. 648 delete mem_req; 649 memReq[tid] = NULL; 650 651 // Send the fault to commit. This thread will not do anything 652 // until commit handles the fault. The only other way it can 653 // wake up is if a squash comes along and changes the PC. 654 TheISA::PCState fetchPC = pc[tid]; 655 656 DPRINTF(Fetch, "[tid:%i]: Translation faulted, building noop.\n", tid); 657 // We will use a nop in ordier to carry the fault. 658 DynInstPtr instruction = buildInst(tid, 659 decoder.decode(TheISA::NoopMachInst, fetchPC.instAddr()), 660 NULL, fetchPC, fetchPC, false); 661 662 instruction->setPredTarg(fetchPC); 663 instruction->fault = fault; 664 wroteToTimeBuffer = true; 665 666 DPRINTF(Activity, "Activity this cycle.\n"); 667 cpu->activityThisCycle(); 668 669 fetchStatus[tid] = TrapPending; 670 671 DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n", tid); 672 DPRINTF(Fetch, "[tid:%i]: fault (%s) detected @ PC %s.\n", 673 tid, fault->name(), pc[tid]); 674 } 675 _status = updateFetchStatus(); 676} 677 678template <class Impl> 679inline void 680DefaultFetch<Impl>::doSquash(const TheISA::PCState &newPC, 681 const DynInstPtr squashInst, ThreadID tid) 682{ 683 DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %s.\n", 684 tid, newPC); 685 686 pc[tid] = newPC; 687 fetchOffset[tid] = 0; 688 if (squashInst && squashInst->pcState().instAddr() == newPC.instAddr()) 689 macroop[tid] = squashInst->macroop; 690 else 691 macroop[tid] = NULL; 692 predecoder.reset(); 693 694 // Clear the icache miss if it's outstanding. 695 if (fetchStatus[tid] == IcacheWaitResponse) { 696 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n", 697 tid); 698 memReq[tid] = NULL; 699 } else if (fetchStatus[tid] == ItlbWait) { 700 DPRINTF(Fetch, "[tid:%i]: Squashing outstanding ITLB miss.\n", 701 tid); 702 memReq[tid] = NULL; 703 } 704 705 // Get rid of the retrying packet if it was from this thread. 706 if (retryTid == tid) { 707 assert(cacheBlocked); 708 if (retryPkt) { 709 delete retryPkt->req; 710 delete retryPkt; 711 } 712 retryPkt = NULL; 713 retryTid = InvalidThreadID; 714 } 715 716 fetchStatus[tid] = Squashing; 717 718 ++fetchSquashCycles; 719} 720 721template<class Impl> 722void 723DefaultFetch<Impl>::squashFromDecode(const TheISA::PCState &newPC, 724 const DynInstPtr squashInst, 725 const InstSeqNum seq_num, ThreadID tid) 726{ 727 DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n", tid); 728 729 doSquash(newPC, squashInst, tid); 730 731 // Tell the CPU to remove any instructions that are in flight between 732 // fetch and decode. 733 cpu->removeInstsUntil(seq_num, tid); 734} 735 736template<class Impl> 737bool 738DefaultFetch<Impl>::checkStall(ThreadID tid) const 739{ 740 bool ret_val = false; 741 742 if (cpu->contextSwitch) { 743 DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid); 744 ret_val = true; 745 } else if (stalls[tid].decode) { 746 DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid); 747 ret_val = true; 748 } else if (stalls[tid].rename) { 749 DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid); 750 ret_val = true; 751 } else if (stalls[tid].iew) { 752 DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid); 753 ret_val = true; 754 } else if (stalls[tid].commit) { 755 DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid); 756 ret_val = true; 757 } 758 759 return ret_val; 760} 761 762template<class Impl> 763typename DefaultFetch<Impl>::FetchStatus 764DefaultFetch<Impl>::updateFetchStatus() 765{ 766 //Check Running 767 list<ThreadID>::iterator threads = activeThreads->begin(); 768 list<ThreadID>::iterator end = activeThreads->end(); 769 770 while (threads != end) { 771 ThreadID tid = *threads++; 772 773 if (fetchStatus[tid] == Running || 774 fetchStatus[tid] == Squashing || 775 fetchStatus[tid] == IcacheAccessComplete) { 776 777 if (_status == Inactive) { 778 DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid); 779 780 if (fetchStatus[tid] == IcacheAccessComplete) { 781 DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache" 782 "completion\n",tid); 783 } 784 785 cpu->activateStage(O3CPU::FetchIdx); 786 } 787 788 return Active; 789 } 790 } 791 792 // Stage is switching from active to inactive, notify CPU of it. 793 if (_status == Active) { 794 DPRINTF(Activity, "Deactivating stage.\n"); 795 796 cpu->deactivateStage(O3CPU::FetchIdx); 797 } 798 799 return Inactive; 800} 801 802template <class Impl> 803void 804DefaultFetch<Impl>::squash(const TheISA::PCState &newPC, 805 const InstSeqNum seq_num, DynInstPtr squashInst, 806 ThreadID tid) 807{ 808 DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n", tid); 809 810 doSquash(newPC, squashInst, tid); 811 812 // Tell the CPU to remove any instructions that are not in the ROB. 813 cpu->removeInstsNotInROB(tid); 814} 815 816template <class Impl> 817void 818DefaultFetch<Impl>::tick() 819{ 820 list<ThreadID>::iterator threads = activeThreads->begin(); 821 list<ThreadID>::iterator end = activeThreads->end(); 822 bool status_change = false; 823 824 wroteToTimeBuffer = false; 825 826 for (ThreadID i = 0; i < Impl::MaxThreads; ++i) { 827 issuePipelinedIfetch[i] = false; 828 } 829 830 while (threads != end) { 831 ThreadID tid = *threads++; 832 833 // Check the signals for each thread to determine the proper status 834 // for each thread. 835 bool updated_status = checkSignalsAndUpdate(tid); 836 status_change = status_change || updated_status; 837 } 838 839 DPRINTF(Fetch, "Running stage.\n"); 840 841 if (FullSystem) { 842 if (fromCommit->commitInfo[0].interruptPending) { 843 interruptPending = true; 844 } 845 846 if (fromCommit->commitInfo[0].clearInterrupt) { 847 interruptPending = false; 848 } 849 } 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 // Issue the next I-cache request if possible. 873 for (ThreadID i = 0; i < Impl::MaxThreads; ++i) { 874 if (issuePipelinedIfetch[i]) { 875 pipelineIcacheAccesses(i); 876 } 877 } 878 879 // Reset the number of the instruction we've fetched. 880 numInst = 0; 881} 882 883template <class Impl> 884bool 885DefaultFetch<Impl>::checkSignalsAndUpdate(ThreadID tid) 886{ 887 // Update the per thread stall statuses. 888 if (fromDecode->decodeBlock[tid]) { 889 stalls[tid].decode = true; 890 } 891 892 if (fromDecode->decodeUnblock[tid]) { 893 assert(stalls[tid].decode); 894 assert(!fromDecode->decodeBlock[tid]); 895 stalls[tid].decode = false; 896 } 897 898 if (fromRename->renameBlock[tid]) { 899 stalls[tid].rename = true; 900 } 901 902 if (fromRename->renameUnblock[tid]) { 903 assert(stalls[tid].rename); 904 assert(!fromRename->renameBlock[tid]); 905 stalls[tid].rename = false; 906 } 907 908 if (fromIEW->iewBlock[tid]) { 909 stalls[tid].iew = true; 910 } 911 912 if (fromIEW->iewUnblock[tid]) { 913 assert(stalls[tid].iew); 914 assert(!fromIEW->iewBlock[tid]); 915 stalls[tid].iew = false; 916 } 917 918 if (fromCommit->commitBlock[tid]) { 919 stalls[tid].commit = true; 920 } 921 922 if (fromCommit->commitUnblock[tid]) { 923 assert(stalls[tid].commit); 924 assert(!fromCommit->commitBlock[tid]); 925 stalls[tid].commit = false; 926 } 927 928 // Check squash signals from commit. 929 if (fromCommit->commitInfo[tid].squash) { 930 931 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 932 "from commit.\n",tid); 933 // In any case, squash. 934 squash(fromCommit->commitInfo[tid].pc, 935 fromCommit->commitInfo[tid].doneSeqNum, 936 fromCommit->commitInfo[tid].squashInst, tid); 937 938 // If it was a branch mispredict on a control instruction, update the 939 // branch predictor with that instruction, otherwise just kill the 940 // invalid state we generated in after sequence number 941 if (fromCommit->commitInfo[tid].mispredictInst && 942 fromCommit->commitInfo[tid].mispredictInst->isControl()) { 943 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 944 fromCommit->commitInfo[tid].pc, 945 fromCommit->commitInfo[tid].branchTaken, 946 tid); 947 } else { 948 branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum, 949 tid); 950 } 951 952 return true; 953 } else if (fromCommit->commitInfo[tid].doneSeqNum) { 954 // Update the branch predictor if it wasn't a squashed instruction 955 // that was broadcasted. 956 branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid); 957 } 958 959 // Check ROB squash signals from commit. 960 if (fromCommit->commitInfo[tid].robSquashing) { 961 DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid); 962 963 // Continue to squash. 964 fetchStatus[tid] = Squashing; 965 966 return true; 967 } 968 969 // Check squash signals from decode. 970 if (fromDecode->decodeInfo[tid].squash) { 971 DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash " 972 "from decode.\n",tid); 973 974 // Update the branch predictor. 975 if (fromDecode->decodeInfo[tid].branchMispredict) { 976 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 977 fromDecode->decodeInfo[tid].nextPC, 978 fromDecode->decodeInfo[tid].branchTaken, 979 tid); 980 } else { 981 branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum, 982 tid); 983 } 984 985 if (fetchStatus[tid] != Squashing) { 986 987 DPRINTF(Fetch, "Squashing from decode with PC = %s\n", 988 fromDecode->decodeInfo[tid].nextPC); 989 // Squash unless we're already squashing 990 squashFromDecode(fromDecode->decodeInfo[tid].nextPC, 991 fromDecode->decodeInfo[tid].squashInst, 992 fromDecode->decodeInfo[tid].doneSeqNum, 993 tid); 994 995 return true; 996 } 997 } 998 999 if (checkStall(tid) && 1000 fetchStatus[tid] != IcacheWaitResponse && 1001 fetchStatus[tid] != IcacheWaitRetry) { 1002 DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid); 1003 1004 fetchStatus[tid] = Blocked; 1005 1006 return true; 1007 } 1008 1009 if (fetchStatus[tid] == Blocked || 1010 fetchStatus[tid] == Squashing) { 1011 // Switch status to running if fetch isn't being told to block or 1012 // squash this cycle. 1013 DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n", 1014 tid); 1015 1016 fetchStatus[tid] = Running; 1017 1018 return true; 1019 } 1020 1021 // If we've reached this point, we have not gotten any signals that 1022 // cause fetch to change its status. Fetch remains the same as before. 1023 return false; 1024} 1025 1026template<class Impl> 1027typename Impl::DynInstPtr 1028DefaultFetch<Impl>::buildInst(ThreadID tid, StaticInstPtr staticInst, 1029 StaticInstPtr curMacroop, TheISA::PCState thisPC, 1030 TheISA::PCState nextPC, bool trace) 1031{ 1032 // Get a sequence number. 1033 InstSeqNum seq = cpu->getAndIncrementInstSeq(); 1034 1035 // Create a new DynInst from the instruction fetched. 1036 DynInstPtr instruction = 1037 new DynInst(staticInst, curMacroop, thisPC, nextPC, seq, cpu); 1038 instruction->setTid(tid); 1039 1040 instruction->setASID(tid); 1041 1042 instruction->setThreadState(cpu->thread[tid]); 1043 1044 DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x (%d) created " 1045 "[sn:%lli].\n", tid, thisPC.instAddr(), 1046 thisPC.microPC(), seq); 1047 1048 DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", tid, 1049 instruction->staticInst-> 1050 disassemble(thisPC.instAddr())); 1051 1052#if TRACING_ON 1053 if (trace) { 1054 instruction->traceData = 1055 cpu->getTracer()->getInstRecord(curTick(), cpu->tcBase(tid), 1056 instruction->staticInst, thisPC, curMacroop); 1057 } 1058#else 1059 instruction->traceData = NULL; 1060#endif 1061 1062 // Add instruction to the CPU's list of instructions. 1063 instruction->setInstListIt(cpu->addInst(instruction)); 1064 1065 // Write the instruction to the first slot in the queue 1066 // that heads to decode. 1067 assert(numInst < fetchWidth); 1068 toDecode->insts[toDecode->size++] = instruction; 1069 1070 // Keep track of if we can take an interrupt at this boundary 1071 delayedCommit[tid] = instruction->isDelayedCommit(); 1072 1073 return instruction; 1074} 1075 1076template<class Impl> 1077void 1078DefaultFetch<Impl>::fetch(bool &status_change) 1079{ 1080 ////////////////////////////////////////// 1081 // Start actual fetch 1082 ////////////////////////////////////////// 1083 ThreadID tid = getFetchingThread(fetchPolicy); 1084 1085 if (tid == InvalidThreadID || drainPending) { 1086 // Breaks looping condition in tick() 1087 threadFetched = numFetchingThreads; 1088 1089 if (numThreads == 1) { // @todo Per-thread stats 1090 profileStall(0); 1091 } 1092 1093 return; 1094 } 1095 1096 DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid); 1097 1098 // The current PC. 1099 TheISA::PCState thisPC = pc[tid]; 1100 1101 Addr pcOffset = fetchOffset[tid]; 1102 Addr fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1103 1104 bool inRom = isRomMicroPC(thisPC.microPC()); 1105 1106 // If returning from the delay of a cache miss, then update the status 1107 // to running, otherwise do the cache access. Possibly move this up 1108 // to tick() function. 1109 if (fetchStatus[tid] == IcacheAccessComplete) { 1110 DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", tid); 1111 1112 fetchStatus[tid] = Running; 1113 status_change = true; 1114 } else if (fetchStatus[tid] == Running) { 1115 // Align the fetch PC so its at the start of a cache block. 1116 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1117 1118 // If buffer is no longer valid or fetchAddr has moved to point 1119 // to the next cache block, AND we have no remaining ucode 1120 // from a macro-op, then start fetch from icache. 1121 if (!(cacheDataValid[tid] && block_PC == cacheDataPC[tid]) 1122 && !inRom && !macroop[tid]) { 1123 DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read " 1124 "instruction, starting at PC %s.\n", tid, thisPC); 1125 1126 fetchCacheLine(fetchAddr, tid, thisPC.instAddr()); 1127 1128 if (fetchStatus[tid] == IcacheWaitResponse) 1129 ++icacheStallCycles; 1130 else if (fetchStatus[tid] == ItlbWait) 1131 ++fetchTlbCycles; 1132 else 1133 ++fetchMiscStallCycles; 1134 return; 1135 } else if ((checkInterrupt(thisPC.instAddr()) && !delayedCommit[tid]) 1136 || isSwitchedOut()) { 1137 // Stall CPU if an interrupt is posted and we're not issuing 1138 // an delayed commit micro-op currently (delayed commit instructions 1139 // are not interruptable by interrupts, only faults) 1140 ++fetchMiscStallCycles; 1141 return; 1142 } 1143 } else { 1144 if (fetchStatus[tid] == Idle) { 1145 ++fetchIdleCycles; 1146 DPRINTF(Fetch, "[tid:%i]: Fetch is idle!\n", tid); 1147 } 1148 1149 // Status is Idle, so fetch should do nothing. 1150 return; 1151 } 1152 1153 ++fetchCycles; 1154 1155 TheISA::PCState nextPC = thisPC; 1156 1157 StaticInstPtr staticInst = NULL; 1158 StaticInstPtr curMacroop = macroop[tid]; 1159 1160 // If the read of the first instruction was successful, then grab the 1161 // instructions from the rest of the cache line and put them into the 1162 // queue heading to decode. 1163 1164 DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to " 1165 "decode.\n", tid); 1166 1167 // Need to keep track of whether or not a predicted branch 1168 // ended this fetch block. 1169 bool predictedBranch = false; 1170 1171 TheISA::MachInst *cacheInsts = 1172 reinterpret_cast<TheISA::MachInst *>(cacheData[tid]); 1173 1174 const unsigned numInsts = cacheBlkSize / instSize; 1175 unsigned blkOffset = (fetchAddr - cacheDataPC[tid]) / instSize; 1176 1177 // Loop through instruction memory from the cache. 1178 // Keep issuing while fetchWidth is available and branch is not 1179 // predicted taken 1180 while (numInst < fetchWidth && !predictedBranch) { 1181 1182 // We need to process more memory if we aren't going to get a 1183 // StaticInst from the rom, the current macroop, or what's already 1184 // in the predecoder. 1185 bool needMem = !inRom && !curMacroop && !predecoder.extMachInstReady(); 1186 1187 if (needMem) { 1188 if (blkOffset >= numInsts) { 1189 // We need to process more memory, but we've run out of the 1190 // current block. 1191 break; 1192 } 1193 1194 if (ISA_HAS_DELAY_SLOT && pcOffset == 0) { 1195 // Walk past any annulled delay slot instructions. 1196 Addr pcAddr = thisPC.instAddr() & BaseCPU::PCMask; 1197 while (fetchAddr != pcAddr && blkOffset < numInsts) { 1198 blkOffset++; 1199 fetchAddr += instSize; 1200 } 1201 if (blkOffset >= numInsts) 1202 break; 1203 } 1204 MachInst inst = TheISA::gtoh(cacheInsts[blkOffset]); 1205 1206 predecoder.setTC(cpu->thread[tid]->getTC()); 1207 predecoder.moreBytes(thisPC, fetchAddr, inst); 1208 1209 if (predecoder.needMoreBytes()) { 1210 blkOffset++; 1211 fetchAddr += instSize; 1212 pcOffset += instSize; 1213 } 1214 } 1215 1216 // Extract as many instructions and/or microops as we can from 1217 // the memory we've processed so far. 1218 do { 1219 if (!(curMacroop || inRom)) { 1220 if (predecoder.extMachInstReady()) { 1221 ExtMachInst extMachInst = 1222 predecoder.getExtMachInst(thisPC); 1223 staticInst = 1224 decoder.decode(extMachInst, thisPC.instAddr()); 1225 1226 // Increment stat of fetched instructions. 1227 ++fetchedInsts; 1228 1229 if (staticInst->isMacroop()) { 1230 curMacroop = staticInst; 1231 } else { 1232 pcOffset = 0; 1233 } 1234 } else { 1235 // We need more bytes for this instruction so blkOffset and 1236 // pcOffset will be updated 1237 break; 1238 } 1239 } 1240 // Whether we're moving to a new macroop because we're at the 1241 // end of the current one, or the branch predictor incorrectly 1242 // thinks we are... 1243 bool newMacro = false; 1244 if (curMacroop || inRom) { 1245 if (inRom) { 1246 staticInst = cpu->microcodeRom.fetchMicroop( 1247 thisPC.microPC(), curMacroop); 1248 } else { 1249 staticInst = curMacroop->fetchMicroop(thisPC.microPC()); 1250 } 1251 newMacro |= staticInst->isLastMicroop(); 1252 } 1253 1254 DynInstPtr instruction = 1255 buildInst(tid, staticInst, curMacroop, 1256 thisPC, nextPC, true); 1257 1258 numInst++; 1259 1260#if TRACING_ON 1261 instruction->fetchTick = curTick(); 1262#endif 1263 1264 nextPC = thisPC; 1265 1266 // If we're branching after this instruction, quite fetching 1267 // from the same block then. 1268 predictedBranch |= thisPC.branching(); 1269 predictedBranch |= 1270 lookupAndUpdateNextPC(instruction, nextPC); 1271 if (predictedBranch) { 1272 DPRINTF(Fetch, "Branch detected with PC = %s\n", thisPC); 1273 } 1274 1275 newMacro |= thisPC.instAddr() != nextPC.instAddr(); 1276 1277 // Move to the next instruction, unless we have a branch. 1278 thisPC = nextPC; 1279 1280 if (newMacro) { 1281 fetchAddr = thisPC.instAddr() & BaseCPU::PCMask; 1282 blkOffset = (fetchAddr - cacheDataPC[tid]) / instSize; 1283 pcOffset = 0; 1284 curMacroop = NULL; 1285 } 1286 1287 if (instruction->isQuiesce()) { 1288 DPRINTF(Fetch, 1289 "Quiesce instruction encountered, halting fetch!"); 1290 fetchStatus[tid] = QuiescePending; 1291 status_change = true; 1292 break; 1293 } 1294 } while ((curMacroop || predecoder.extMachInstReady()) && 1295 numInst < fetchWidth); 1296 } 1297 1298 if (predictedBranch) { 1299 DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch " 1300 "instruction encountered.\n", tid); 1301 } else if (numInst >= fetchWidth) { 1302 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth " 1303 "for this cycle.\n", tid); 1304 } else if (blkOffset >= cacheBlkSize) { 1305 DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache " 1306 "block.\n", tid); 1307 } 1308 1309 macroop[tid] = curMacroop; 1310 fetchOffset[tid] = pcOffset; 1311 1312 if (numInst > 0) { 1313 wroteToTimeBuffer = true; 1314 } 1315 1316 pc[tid] = thisPC; 1317 1318 // pipeline a fetch if we're crossing a cache boundary and not in 1319 // a state that would preclude fetching 1320 fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1321 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1322 issuePipelinedIfetch[tid] = block_PC != cacheDataPC[tid] && 1323 fetchStatus[tid] != IcacheWaitResponse && 1324 fetchStatus[tid] != ItlbWait && 1325 fetchStatus[tid] != IcacheWaitRetry && 1326 fetchStatus[tid] != QuiescePending && 1327 !curMacroop; 1328} 1329 1330template<class Impl> 1331void 1332DefaultFetch<Impl>::recvRetry() 1333{ 1334 if (retryPkt != NULL) { 1335 assert(cacheBlocked); 1336 assert(retryTid != InvalidThreadID); 1337 assert(fetchStatus[retryTid] == IcacheWaitRetry); 1338 1339 if (cpu->getIcachePort()->sendTiming(retryPkt)) { 1340 fetchStatus[retryTid] = IcacheWaitResponse; 1341 retryPkt = NULL; 1342 retryTid = InvalidThreadID; 1343 cacheBlocked = false; 1344 } 1345 } else { 1346 assert(retryTid == InvalidThreadID); 1347 // Access has been squashed since it was sent out. Just clear 1348 // the cache being blocked. 1349 cacheBlocked = false; 1350 } 1351} 1352 1353/////////////////////////////////////// 1354// // 1355// SMT FETCH POLICY MAINTAINED HERE // 1356// // 1357/////////////////////////////////////// 1358template<class Impl> 1359ThreadID 1360DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority) 1361{ 1362 if (numThreads > 1) { 1363 switch (fetch_priority) { 1364 1365 case SingleThread: 1366 return 0; 1367 1368 case RoundRobin: 1369 return roundRobin(); 1370 1371 case IQ: 1372 return iqCount(); 1373 1374 case LSQ: 1375 return lsqCount(); 1376 1377 case Branch: 1378 return branchCount(); 1379 1380 default: 1381 return InvalidThreadID; 1382 } 1383 } else { 1384 list<ThreadID>::iterator thread = activeThreads->begin(); 1385 if (thread == activeThreads->end()) { 1386 return InvalidThreadID; 1387 } 1388 1389 ThreadID tid = *thread; 1390 1391 if (fetchStatus[tid] == Running || 1392 fetchStatus[tid] == IcacheAccessComplete || 1393 fetchStatus[tid] == Idle) { 1394 return tid; 1395 } else { 1396 return InvalidThreadID; 1397 } 1398 } 1399} 1400 1401 1402template<class Impl> 1403ThreadID 1404DefaultFetch<Impl>::roundRobin() 1405{ 1406 list<ThreadID>::iterator pri_iter = priorityList.begin(); 1407 list<ThreadID>::iterator end = priorityList.end(); 1408 1409 ThreadID high_pri; 1410 1411 while (pri_iter != end) { 1412 high_pri = *pri_iter; 1413 1414 assert(high_pri <= numThreads); 1415 1416 if (fetchStatus[high_pri] == Running || 1417 fetchStatus[high_pri] == IcacheAccessComplete || 1418 fetchStatus[high_pri] == Idle) { 1419 1420 priorityList.erase(pri_iter); 1421 priorityList.push_back(high_pri); 1422 1423 return high_pri; 1424 } 1425 1426 pri_iter++; 1427 } 1428 1429 return InvalidThreadID; 1430} 1431 1432template<class Impl> 1433ThreadID 1434DefaultFetch<Impl>::iqCount() 1435{ 1436 std::priority_queue<unsigned> PQ; 1437 std::map<unsigned, ThreadID> threadMap; 1438 1439 list<ThreadID>::iterator threads = activeThreads->begin(); 1440 list<ThreadID>::iterator end = activeThreads->end(); 1441 1442 while (threads != end) { 1443 ThreadID tid = *threads++; 1444 unsigned iqCount = fromIEW->iewInfo[tid].iqCount; 1445 1446 PQ.push(iqCount); 1447 threadMap[iqCount] = tid; 1448 } 1449 1450 while (!PQ.empty()) { 1451 ThreadID high_pri = threadMap[PQ.top()]; 1452 1453 if (fetchStatus[high_pri] == Running || 1454 fetchStatus[high_pri] == IcacheAccessComplete || 1455 fetchStatus[high_pri] == Idle) 1456 return high_pri; 1457 else 1458 PQ.pop(); 1459 1460 } 1461 1462 return InvalidThreadID; 1463} 1464 1465template<class Impl> 1466ThreadID 1467DefaultFetch<Impl>::lsqCount() 1468{ 1469 std::priority_queue<unsigned> PQ; 1470 std::map<unsigned, ThreadID> threadMap; 1471 1472 list<ThreadID>::iterator threads = activeThreads->begin(); 1473 list<ThreadID>::iterator end = activeThreads->end(); 1474 1475 while (threads != end) { 1476 ThreadID tid = *threads++; 1477 unsigned ldstqCount = fromIEW->iewInfo[tid].ldstqCount; 1478 1479 PQ.push(ldstqCount); 1480 threadMap[ldstqCount] = tid; 1481 } 1482 1483 while (!PQ.empty()) { 1484 ThreadID high_pri = threadMap[PQ.top()]; 1485 1486 if (fetchStatus[high_pri] == Running || 1487 fetchStatus[high_pri] == IcacheAccessComplete || 1488 fetchStatus[high_pri] == Idle) 1489 return high_pri; 1490 else 1491 PQ.pop(); 1492 } 1493 1494 return InvalidThreadID; 1495} 1496 1497template<class Impl> 1498ThreadID 1499DefaultFetch<Impl>::branchCount() 1500{ 1501#if 0 1502 list<ThreadID>::iterator thread = activeThreads->begin(); 1503 assert(thread != activeThreads->end()); 1504 ThreadID tid = *thread; 1505#endif 1506 1507 panic("Branch Count Fetch policy unimplemented\n"); 1508 return InvalidThreadID; 1509} 1510 1511template<class Impl> 1512void 1513DefaultFetch<Impl>::pipelineIcacheAccesses(ThreadID tid) 1514{ 1515 if (!issuePipelinedIfetch[tid]) { 1516 return; 1517 } 1518 1519 // The next PC to access. 1520 TheISA::PCState thisPC = pc[tid]; 1521 1522 if (isRomMicroPC(thisPC.microPC())) { 1523 return; 1524 } 1525 1526 Addr pcOffset = fetchOffset[tid]; 1527 Addr fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask; 1528 1529 // Align the fetch PC so its at the start of a cache block. 1530 Addr block_PC = icacheBlockAlignPC(fetchAddr); 1531 1532 // Unless buffer already got the block, fetch it from icache. 1533 if (!(cacheDataValid[tid] && block_PC == cacheDataPC[tid])) { 1534 DPRINTF(Fetch, "[tid:%i]: Issuing a pipelined I-cache access, " 1535 "starting at PC %s.\n", tid, thisPC); 1536 1537 fetchCacheLine(fetchAddr, tid, thisPC.instAddr()); 1538 } 1539} 1540 1541template<class Impl> 1542void 1543DefaultFetch<Impl>::profileStall(ThreadID tid) { 1544 DPRINTF(Fetch,"There are no more threads available to fetch from.\n"); 1545 1546 // @todo Per-thread stats 1547 1548 if (drainPending) { 1549 ++fetchPendingDrainCycles; 1550 DPRINTF(Fetch, "Fetch is waiting for a drain!\n"); 1551 } else if (activeThreads->empty()) { 1552 ++fetchNoActiveThreadStallCycles; 1553 DPRINTF(Fetch, "Fetch has no active thread!\n"); 1554 } else if (fetchStatus[tid] == Blocked) { 1555 ++fetchBlockedCycles; 1556 DPRINTF(Fetch, "[tid:%i]: Fetch is blocked!\n", tid); 1557 } else if (fetchStatus[tid] == Squashing) { 1558 ++fetchSquashCycles; 1559 DPRINTF(Fetch, "[tid:%i]: Fetch is squashing!\n", tid); 1560 } else if (fetchStatus[tid] == IcacheWaitResponse) { 1561 ++icacheStallCycles; 1562 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting cache response!\n", 1563 tid); 1564 } else if (fetchStatus[tid] == ItlbWait) { 1565 ++fetchTlbCycles; 1566 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting ITLB walk to " 1567 "finish!\n", tid); 1568 } else if (fetchStatus[tid] == TrapPending) { 1569 ++fetchPendingTrapStallCycles; 1570 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for a pending trap!\n", 1571 tid); 1572 } else if (fetchStatus[tid] == QuiescePending) { 1573 ++fetchPendingQuiesceStallCycles; 1574 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for a pending quiesce " 1575 "instruction!\n", tid); 1576 } else if (fetchStatus[tid] == IcacheWaitRetry) { 1577 ++fetchIcacheWaitRetryStallCycles; 1578 DPRINTF(Fetch, "[tid:%i]: Fetch is waiting for an I-cache retry!\n", 1579 tid); 1580 } else if (fetchStatus[tid] == NoGoodAddr) { 1581 DPRINTF(Fetch, "[tid:%i]: Fetch predicted non-executable address\n", 1582 tid); 1583 } else { 1584 DPRINTF(Fetch, "[tid:%i]: Unexpected fetch stall reason (Status: %i).\n", 1585 tid, fetchStatus[tid]); 1586 } 1587} 1588