cpu.cc revision 8707:489489c67fd9
1/* 2 * Copyright (c) 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 * Copyright (c) 2011 Regents of the University of California 16 * All rights reserved. 17 * 18 * Redistribution and use in source and binary forms, with or without 19 * modification, are permitted provided that the following conditions are 20 * met: redistributions of source code must retain the above copyright 21 * notice, this list of conditions and the following disclaimer; 22 * redistributions in binary form must reproduce the above copyright 23 * notice, this list of conditions and the following disclaimer in the 24 * documentation and/or other materials provided with the distribution; 25 * neither the name of the copyright holders nor the names of its 26 * contributors may be used to endorse or promote products derived from 27 * this software without specific prior written permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 40 * 41 * Authors: Kevin Lim 42 * Korey Sewell 43 * Rick Strong 44 */ 45 46#include "config/full_system.hh" 47#include "config/the_isa.hh" 48#include "config/use_checker.hh" 49#include "cpu/o3/cpu.hh" 50#include "cpu/o3/isa_specific.hh" 51#include "cpu/o3/thread_context.hh" 52#include "cpu/activity.hh" 53#include "cpu/simple_thread.hh" 54#include "cpu/thread_context.hh" 55#include "debug/Activity.hh" 56#include "debug/O3CPU.hh" 57#include "debug/Quiesce.hh" 58#include "enums/MemoryMode.hh" 59#include "sim/core.hh" 60#include "sim/stat_control.hh" 61#include "sim/system.hh" 62 63#if FULL_SYSTEM 64#include "cpu/quiesce_event.hh" 65#else 66#include "sim/process.hh" 67#endif 68 69#if USE_CHECKER 70#include "cpu/checker/cpu.hh" 71#endif 72 73#if THE_ISA == ALPHA_ISA 74#include "arch/alpha/osfpal.hh" 75#include "debug/Activity.hh" 76#endif 77 78class BaseCPUParams; 79 80using namespace TheISA; 81using namespace std; 82 83BaseO3CPU::BaseO3CPU(BaseCPUParams *params) 84 : BaseCPU(params) 85{ 86} 87 88void 89BaseO3CPU::regStats() 90{ 91 BaseCPU::regStats(); 92} 93 94template<class Impl> 95bool 96FullO3CPU<Impl>::IcachePort::recvTiming(PacketPtr pkt) 97{ 98 DPRINTF(O3CPU, "Fetch unit received timing\n"); 99 if (pkt->isResponse()) { 100 // We shouldn't ever get a block in ownership state 101 assert(!(pkt->memInhibitAsserted() && !pkt->sharedAsserted())); 102 103 fetch->processCacheCompletion(pkt); 104 } 105 //else Snooped a coherence request, just return 106 return true; 107} 108 109template<class Impl> 110void 111FullO3CPU<Impl>::IcachePort::recvRetry() 112{ 113 fetch->recvRetry(); 114} 115 116template <class Impl> 117bool 118FullO3CPU<Impl>::DcachePort::recvTiming(PacketPtr pkt) 119{ 120 return lsq->recvTiming(pkt); 121} 122 123template <class Impl> 124void 125FullO3CPU<Impl>::DcachePort::recvRetry() 126{ 127 lsq->recvRetry(); 128} 129 130template <class Impl> 131FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 132 : Event(CPU_Tick_Pri), cpu(c) 133{ 134} 135 136template <class Impl> 137void 138FullO3CPU<Impl>::TickEvent::process() 139{ 140 cpu->tick(); 141} 142 143template <class Impl> 144const char * 145FullO3CPU<Impl>::TickEvent::description() const 146{ 147 return "FullO3CPU tick"; 148} 149 150template <class Impl> 151FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent() 152 : Event(CPU_Switch_Pri) 153{ 154} 155 156template <class Impl> 157void 158FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num, 159 FullO3CPU<Impl> *thread_cpu) 160{ 161 tid = thread_num; 162 cpu = thread_cpu; 163} 164 165template <class Impl> 166void 167FullO3CPU<Impl>::ActivateThreadEvent::process() 168{ 169 cpu->activateThread(tid); 170} 171 172template <class Impl> 173const char * 174FullO3CPU<Impl>::ActivateThreadEvent::description() const 175{ 176 return "FullO3CPU \"Activate Thread\""; 177} 178 179template <class Impl> 180FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent() 181 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL) 182{ 183} 184 185template <class Impl> 186void 187FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num, 188 FullO3CPU<Impl> *thread_cpu) 189{ 190 tid = thread_num; 191 cpu = thread_cpu; 192 remove = false; 193} 194 195template <class Impl> 196void 197FullO3CPU<Impl>::DeallocateContextEvent::process() 198{ 199 cpu->deactivateThread(tid); 200 if (remove) 201 cpu->removeThread(tid); 202} 203 204template <class Impl> 205const char * 206FullO3CPU<Impl>::DeallocateContextEvent::description() const 207{ 208 return "FullO3CPU \"Deallocate Context\""; 209} 210 211template <class Impl> 212FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params) 213 : BaseO3CPU(params), 214 itb(params->itb), 215 dtb(params->dtb), 216 tickEvent(this), 217#ifndef NDEBUG 218 instcount(0), 219#endif 220 removeInstsThisCycle(false), 221 fetch(this, params), 222 decode(this, params), 223 rename(this, params), 224 iew(this, params), 225 commit(this, params), 226 227 regFile(this, params->numPhysIntRegs, 228 params->numPhysFloatRegs), 229 230 freeList(params->numThreads, 231 TheISA::NumIntRegs, params->numPhysIntRegs, 232 TheISA::NumFloatRegs, params->numPhysFloatRegs), 233 234 rob(this, 235 params->numROBEntries, params->squashWidth, 236 params->smtROBPolicy, params->smtROBThreshold, 237 params->numThreads), 238 239 scoreboard(params->numThreads, 240 TheISA::NumIntRegs, params->numPhysIntRegs, 241 TheISA::NumFloatRegs, params->numPhysFloatRegs, 242 TheISA::NumMiscRegs * numThreads, 243 TheISA::ZeroReg), 244 245 icachePort(&fetch, this), 246 dcachePort(&iew.ldstQueue, this), 247 248 timeBuffer(params->backComSize, params->forwardComSize), 249 fetchQueue(params->backComSize, params->forwardComSize), 250 decodeQueue(params->backComSize, params->forwardComSize), 251 renameQueue(params->backComSize, params->forwardComSize), 252 iewQueue(params->backComSize, params->forwardComSize), 253 activityRec(name(), NumStages, 254 params->backComSize + params->forwardComSize, 255 params->activity), 256 257 globalSeqNum(1), 258 system(params->system), 259 drainCount(0), 260 deferRegistration(params->defer_registration) 261{ 262 if (!deferRegistration) { 263 _status = Running; 264 } else { 265 _status = Idle; 266 } 267 268#if USE_CHECKER 269 if (params->checker) { 270 BaseCPU *temp_checker = params->checker; 271 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker); 272 checker->setIcachePort(&icachePort); 273#if FULL_SYSTEM 274 checker->setSystem(params->system); 275#endif 276 } else { 277 checker = NULL; 278 } 279#endif // USE_CHECKER 280 281#if !FULL_SYSTEM 282 thread.resize(numThreads); 283 tids.resize(numThreads); 284#endif 285 286 // The stages also need their CPU pointer setup. However this 287 // must be done at the upper level CPU because they have pointers 288 // to the upper level CPU, and not this FullO3CPU. 289 290 // Set up Pointers to the activeThreads list for each stage 291 fetch.setActiveThreads(&activeThreads); 292 decode.setActiveThreads(&activeThreads); 293 rename.setActiveThreads(&activeThreads); 294 iew.setActiveThreads(&activeThreads); 295 commit.setActiveThreads(&activeThreads); 296 297 // Give each of the stages the time buffer they will use. 298 fetch.setTimeBuffer(&timeBuffer); 299 decode.setTimeBuffer(&timeBuffer); 300 rename.setTimeBuffer(&timeBuffer); 301 iew.setTimeBuffer(&timeBuffer); 302 commit.setTimeBuffer(&timeBuffer); 303 304 // Also setup each of the stages' queues. 305 fetch.setFetchQueue(&fetchQueue); 306 decode.setFetchQueue(&fetchQueue); 307 commit.setFetchQueue(&fetchQueue); 308 decode.setDecodeQueue(&decodeQueue); 309 rename.setDecodeQueue(&decodeQueue); 310 rename.setRenameQueue(&renameQueue); 311 iew.setRenameQueue(&renameQueue); 312 iew.setIEWQueue(&iewQueue); 313 commit.setIEWQueue(&iewQueue); 314 commit.setRenameQueue(&renameQueue); 315 316 commit.setIEWStage(&iew); 317 rename.setIEWStage(&iew); 318 rename.setCommitStage(&commit); 319 320#if !FULL_SYSTEM 321 ThreadID active_threads = params->workload.size(); 322 323 if (active_threads > Impl::MaxThreads) { 324 panic("Workload Size too large. Increase the 'MaxThreads'" 325 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) or " 326 "edit your workload size."); 327 } 328#else 329 ThreadID active_threads = 1; 330#endif 331 332 //Make Sure That this a Valid Architeture 333 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 334 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 335 336 rename.setScoreboard(&scoreboard); 337 iew.setScoreboard(&scoreboard); 338 339 // Setup the rename map for whichever stages need it. 340 PhysRegIndex lreg_idx = 0; 341 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 342 343 for (ThreadID tid = 0; tid < numThreads; tid++) { 344 bool bindRegs = (tid <= active_threads - 1); 345 346 commitRenameMap[tid].init(TheISA::NumIntRegs, 347 params->numPhysIntRegs, 348 lreg_idx, //Index for Logical. Regs 349 350 TheISA::NumFloatRegs, 351 params->numPhysFloatRegs, 352 freg_idx, //Index for Float Regs 353 354 TheISA::NumMiscRegs, 355 356 TheISA::ZeroReg, 357 TheISA::ZeroReg, 358 359 tid, 360 false); 361 362 renameMap[tid].init(TheISA::NumIntRegs, 363 params->numPhysIntRegs, 364 lreg_idx, //Index for Logical. Regs 365 366 TheISA::NumFloatRegs, 367 params->numPhysFloatRegs, 368 freg_idx, //Index for Float Regs 369 370 TheISA::NumMiscRegs, 371 372 TheISA::ZeroReg, 373 TheISA::ZeroReg, 374 375 tid, 376 bindRegs); 377 378 activateThreadEvent[tid].init(tid, this); 379 deallocateContextEvent[tid].init(tid, this); 380 } 381 382 rename.setRenameMap(renameMap); 383 commit.setRenameMap(commitRenameMap); 384 385 // Give renameMap & rename stage access to the freeList; 386 for (ThreadID tid = 0; tid < numThreads; tid++) 387 renameMap[tid].setFreeList(&freeList); 388 rename.setFreeList(&freeList); 389 390 // Setup the ROB for whichever stages need it. 391 commit.setROB(&rob); 392 393 lastRunningCycle = curTick(); 394 395 lastActivatedCycle = -1; 396#if 0 397 // Give renameMap & rename stage access to the freeList; 398 for (ThreadID tid = 0; tid < numThreads; tid++) 399 globalSeqNum[tid] = 1; 400#endif 401 402 contextSwitch = false; 403 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 404 405 // Setup any thread state. 406 this->thread.resize(this->numThreads); 407 408 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 409#if FULL_SYSTEM 410 // SMT is not supported in FS mode yet. 411 assert(this->numThreads == 1); 412 this->thread[tid] = new Thread(this, 0); 413#else 414 if (tid < params->workload.size()) { 415 DPRINTF(O3CPU, "Workload[%i] process is %#x", 416 tid, this->thread[tid]); 417 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 418 (typename Impl::O3CPU *)(this), 419 tid, params->workload[tid]); 420 421 //usedTids[tid] = true; 422 //threadMap[tid] = tid; 423 } else { 424 //Allocate Empty thread so M5 can use later 425 //when scheduling threads to CPU 426 Process* dummy_proc = NULL; 427 428 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 429 (typename Impl::O3CPU *)(this), 430 tid, dummy_proc); 431 //usedTids[tid] = false; 432 } 433#endif // !FULL_SYSTEM 434 435 ThreadContext *tc; 436 437 // Setup the TC that will serve as the interface to the threads/CPU. 438 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 439 440 tc = o3_tc; 441 442 // If we're using a checker, then the TC should be the 443 // CheckerThreadContext. 444#if USE_CHECKER 445 if (params->checker) { 446 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 447 o3_tc, this->checker); 448 } 449#endif 450 451 o3_tc->cpu = (typename Impl::O3CPU *)(this); 452 assert(o3_tc->cpu); 453 o3_tc->thread = this->thread[tid]; 454 455#if FULL_SYSTEM 456 // Setup quiesce event. 457 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 458#endif 459 // Give the thread the TC. 460 this->thread[tid]->tc = tc; 461 462 // Add the TC to the CPU's list of TC's. 463 this->threadContexts.push_back(tc); 464 } 465 466 for (ThreadID tid = 0; tid < this->numThreads; tid++) 467 this->thread[tid]->setFuncExeInst(0); 468 469 lockAddr = 0; 470 lockFlag = false; 471} 472 473template <class Impl> 474FullO3CPU<Impl>::~FullO3CPU() 475{ 476} 477 478template <class Impl> 479void 480FullO3CPU<Impl>::regStats() 481{ 482 BaseO3CPU::regStats(); 483 484 // Register any of the O3CPU's stats here. 485 timesIdled 486 .name(name() + ".timesIdled") 487 .desc("Number of times that the entire CPU went into an idle state and" 488 " unscheduled itself") 489 .prereq(timesIdled); 490 491 idleCycles 492 .name(name() + ".idleCycles") 493 .desc("Total number of cycles that the CPU has spent unscheduled due " 494 "to idling") 495 .prereq(idleCycles); 496 497 quiesceCycles 498 .name(name() + ".quiesceCycles") 499 .desc("Total number of cycles that CPU has spent quiesced or waiting " 500 "for an interrupt") 501 .prereq(quiesceCycles); 502 503 // Number of Instructions simulated 504 // -------------------------------- 505 // Should probably be in Base CPU but need templated 506 // MaxThreads so put in here instead 507 committedInsts 508 .init(numThreads) 509 .name(name() + ".committedInsts") 510 .desc("Number of Instructions Simulated"); 511 512 totalCommittedInsts 513 .name(name() + ".committedInsts_total") 514 .desc("Number of Instructions Simulated"); 515 516 cpi 517 .name(name() + ".cpi") 518 .desc("CPI: Cycles Per Instruction") 519 .precision(6); 520 cpi = numCycles / committedInsts; 521 522 totalCpi 523 .name(name() + ".cpi_total") 524 .desc("CPI: Total CPI of All Threads") 525 .precision(6); 526 totalCpi = numCycles / totalCommittedInsts; 527 528 ipc 529 .name(name() + ".ipc") 530 .desc("IPC: Instructions Per Cycle") 531 .precision(6); 532 ipc = committedInsts / numCycles; 533 534 totalIpc 535 .name(name() + ".ipc_total") 536 .desc("IPC: Total IPC of All Threads") 537 .precision(6); 538 totalIpc = totalCommittedInsts / numCycles; 539 540 this->fetch.regStats(); 541 this->decode.regStats(); 542 this->rename.regStats(); 543 this->iew.regStats(); 544 this->commit.regStats(); 545 this->rob.regStats(); 546 547 intRegfileReads 548 .name(name() + ".int_regfile_reads") 549 .desc("number of integer regfile reads") 550 .prereq(intRegfileReads); 551 552 intRegfileWrites 553 .name(name() + ".int_regfile_writes") 554 .desc("number of integer regfile writes") 555 .prereq(intRegfileWrites); 556 557 fpRegfileReads 558 .name(name() + ".fp_regfile_reads") 559 .desc("number of floating regfile reads") 560 .prereq(fpRegfileReads); 561 562 fpRegfileWrites 563 .name(name() + ".fp_regfile_writes") 564 .desc("number of floating regfile writes") 565 .prereq(fpRegfileWrites); 566 567 miscRegfileReads 568 .name(name() + ".misc_regfile_reads") 569 .desc("number of misc regfile reads") 570 .prereq(miscRegfileReads); 571 572 miscRegfileWrites 573 .name(name() + ".misc_regfile_writes") 574 .desc("number of misc regfile writes") 575 .prereq(miscRegfileWrites); 576} 577 578template <class Impl> 579Port * 580FullO3CPU<Impl>::getPort(const std::string &if_name, int idx) 581{ 582 if (if_name == "dcache_port") 583 return &dcachePort; 584 else if (if_name == "icache_port") 585 return &icachePort; 586 else 587 panic("No Such Port\n"); 588} 589 590template <class Impl> 591void 592FullO3CPU<Impl>::tick() 593{ 594 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 595 596 ++numCycles; 597 598// activity = false; 599 600 //Tick each of the stages 601 fetch.tick(); 602 603 decode.tick(); 604 605 rename.tick(); 606 607 iew.tick(); 608 609 commit.tick(); 610 611#if !FULL_SYSTEM 612 doContextSwitch(); 613#endif 614 615 // Now advance the time buffers 616 timeBuffer.advance(); 617 618 fetchQueue.advance(); 619 decodeQueue.advance(); 620 renameQueue.advance(); 621 iewQueue.advance(); 622 623 activityRec.advance(); 624 625 if (removeInstsThisCycle) { 626 cleanUpRemovedInsts(); 627 } 628 629 if (!tickEvent.scheduled()) { 630 if (_status == SwitchedOut || 631 getState() == SimObject::Drained) { 632 DPRINTF(O3CPU, "Switched out!\n"); 633 // increment stat 634 lastRunningCycle = curTick(); 635 } else if (!activityRec.active() || _status == Idle) { 636 DPRINTF(O3CPU, "Idle!\n"); 637 lastRunningCycle = curTick(); 638 timesIdled++; 639 } else { 640 schedule(tickEvent, nextCycle(curTick() + ticks(1))); 641 DPRINTF(O3CPU, "Scheduling next tick!\n"); 642 } 643 } 644 645#if !FULL_SYSTEM 646 updateThreadPriority(); 647#endif 648} 649 650template <class Impl> 651void 652FullO3CPU<Impl>::init() 653{ 654 BaseCPU::init(); 655 656 // Set inSyscall so that the CPU doesn't squash when initially 657 // setting up registers. 658 for (ThreadID tid = 0; tid < numThreads; ++tid) 659 thread[tid]->inSyscall = true; 660 661 // this CPU could still be unconnected if we are restoring from a 662 // checkpoint and this CPU is to be switched in, thus we can only 663 // do this here if the instruction port is actually connected, if 664 // not we have to do it as part of takeOverFrom 665 if (icachePort.isConnected()) 666 fetch.setIcache(); 667 668#if FULL_SYSTEM 669 for (ThreadID tid = 0; tid < numThreads; tid++) { 670 ThreadContext *src_tc = threadContexts[tid]; 671 TheISA::initCPU(src_tc, src_tc->contextId()); 672 // Initialise the ThreadContext's memory proxies 673 thread[tid]->initMemProxies(thread[tid]->getTC()); 674 } 675#endif 676 677 // Clear inSyscall. 678 for (int tid = 0; tid < numThreads; ++tid) 679 thread[tid]->inSyscall = false; 680 681 // Initialize stages. 682 fetch.initStage(); 683 iew.initStage(); 684 rename.initStage(); 685 commit.initStage(); 686 687 commit.setThreads(thread); 688} 689 690template <class Impl> 691void 692FullO3CPU<Impl>::activateThread(ThreadID tid) 693{ 694 list<ThreadID>::iterator isActive = 695 std::find(activeThreads.begin(), activeThreads.end(), tid); 696 697 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 698 699 if (isActive == activeThreads.end()) { 700 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 701 tid); 702 703 activeThreads.push_back(tid); 704 } 705} 706 707template <class Impl> 708void 709FullO3CPU<Impl>::deactivateThread(ThreadID tid) 710{ 711 //Remove From Active List, if Active 712 list<ThreadID>::iterator thread_it = 713 std::find(activeThreads.begin(), activeThreads.end(), tid); 714 715 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 716 717 if (thread_it != activeThreads.end()) { 718 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 719 tid); 720 activeThreads.erase(thread_it); 721 } 722} 723 724template <class Impl> 725Counter 726FullO3CPU<Impl>::totalInstructions() const 727{ 728 Counter total(0); 729 730 ThreadID size = thread.size(); 731 for (ThreadID i = 0; i < size; i++) 732 total += thread[i]->numInst; 733 734 return total; 735} 736 737template <class Impl> 738void 739FullO3CPU<Impl>::activateContext(ThreadID tid, int delay) 740{ 741 // Needs to set each stage to running as well. 742 if (delay){ 743 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 744 "on cycle %d\n", tid, curTick() + ticks(delay)); 745 scheduleActivateThreadEvent(tid, delay); 746 } else { 747 activateThread(tid); 748 } 749 750 if (lastActivatedCycle < curTick()) { 751 scheduleTickEvent(delay); 752 753 // Be sure to signal that there's some activity so the CPU doesn't 754 // deschedule itself. 755 activityRec.activity(); 756 fetch.wakeFromQuiesce(); 757 758 quiesceCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 759 760 lastActivatedCycle = curTick(); 761 762 _status = Running; 763 } 764} 765 766template <class Impl> 767bool 768FullO3CPU<Impl>::deallocateContext(ThreadID tid, bool remove, int delay) 769{ 770 // Schedule removal of thread data from CPU 771 if (delay){ 772 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 773 "on cycle %d\n", tid, curTick() + ticks(delay)); 774 scheduleDeallocateContextEvent(tid, remove, delay); 775 return false; 776 } else { 777 deactivateThread(tid); 778 if (remove) 779 removeThread(tid); 780 return true; 781 } 782} 783 784template <class Impl> 785void 786FullO3CPU<Impl>::suspendContext(ThreadID tid) 787{ 788 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 789 bool deallocated = deallocateContext(tid, false, 1); 790 // If this was the last thread then unschedule the tick event. 791 if ((activeThreads.size() == 1 && !deallocated) || 792 activeThreads.size() == 0) 793 unscheduleTickEvent(); 794 795 DPRINTF(Quiesce, "Suspending Context\n"); 796 lastRunningCycle = curTick(); 797 _status = Idle; 798} 799 800template <class Impl> 801void 802FullO3CPU<Impl>::haltContext(ThreadID tid) 803{ 804 //For now, this is the same as deallocate 805 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 806 deallocateContext(tid, true, 1); 807} 808 809template <class Impl> 810void 811FullO3CPU<Impl>::insertThread(ThreadID tid) 812{ 813 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 814 // Will change now that the PC and thread state is internal to the CPU 815 // and not in the ThreadContext. 816#if FULL_SYSTEM 817 ThreadContext *src_tc = system->threadContexts[tid]; 818#else 819 ThreadContext *src_tc = tcBase(tid); 820#endif 821 822 //Bind Int Regs to Rename Map 823 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 824 PhysRegIndex phys_reg = freeList.getIntReg(); 825 826 renameMap[tid].setEntry(ireg,phys_reg); 827 scoreboard.setReg(phys_reg); 828 } 829 830 //Bind Float Regs to Rename Map 831 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 832 PhysRegIndex phys_reg = freeList.getFloatReg(); 833 834 renameMap[tid].setEntry(freg,phys_reg); 835 scoreboard.setReg(phys_reg); 836 } 837 838 //Copy Thread Data Into RegFile 839 //this->copyFromTC(tid); 840 841 //Set PC/NPC/NNPC 842 pcState(src_tc->pcState(), tid); 843 844 src_tc->setStatus(ThreadContext::Active); 845 846 activateContext(tid,1); 847 848 //Reset ROB/IQ/LSQ Entries 849 commit.rob->resetEntries(); 850 iew.resetEntries(); 851} 852 853template <class Impl> 854void 855FullO3CPU<Impl>::removeThread(ThreadID tid) 856{ 857 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 858 859 // Copy Thread Data From RegFile 860 // If thread is suspended, it might be re-allocated 861 // this->copyToTC(tid); 862 863 864 // @todo: 2-27-2008: Fix how we free up rename mappings 865 // here to alleviate the case for double-freeing registers 866 // in SMT workloads. 867 868 // Unbind Int Regs from Rename Map 869 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 870 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 871 872 scoreboard.unsetReg(phys_reg); 873 freeList.addReg(phys_reg); 874 } 875 876 // Unbind Float Regs from Rename Map 877 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) { 878 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 879 880 scoreboard.unsetReg(phys_reg); 881 freeList.addReg(phys_reg); 882 } 883 884 // Squash Throughout Pipeline 885 DynInstPtr inst = commit.rob->readHeadInst(tid); 886 InstSeqNum squash_seq_num = inst->seqNum; 887 fetch.squash(0, squash_seq_num, inst, tid); 888 decode.squash(tid); 889 rename.squash(squash_seq_num, tid); 890 iew.squash(tid); 891 iew.ldstQueue.squash(squash_seq_num, tid); 892 commit.rob->squash(squash_seq_num, tid); 893 894 895 assert(iew.instQueue.getCount(tid) == 0); 896 assert(iew.ldstQueue.getCount(tid) == 0); 897 898 // Reset ROB/IQ/LSQ Entries 899 900 // Commented out for now. This should be possible to do by 901 // telling all the pipeline stages to drain first, and then 902 // checking until the drain completes. Once the pipeline is 903 // drained, call resetEntries(). - 10-09-06 ktlim 904/* 905 if (activeThreads.size() >= 1) { 906 commit.rob->resetEntries(); 907 iew.resetEntries(); 908 } 909*/ 910} 911 912 913template <class Impl> 914void 915FullO3CPU<Impl>::activateWhenReady(ThreadID tid) 916{ 917 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 918 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 919 tid); 920 921 bool ready = true; 922 923 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 924 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 925 "Phys. Int. Regs.\n", 926 tid); 927 ready = false; 928 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 929 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 930 "Phys. Float. Regs.\n", 931 tid); 932 ready = false; 933 } else if (commit.rob->numFreeEntries() >= 934 commit.rob->entryAmount(activeThreads.size() + 1)) { 935 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 936 "ROB entries.\n", 937 tid); 938 ready = false; 939 } else if (iew.instQueue.numFreeEntries() >= 940 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 941 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 942 "IQ entries.\n", 943 tid); 944 ready = false; 945 } else if (iew.ldstQueue.numFreeEntries() >= 946 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 947 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 948 "LSQ entries.\n", 949 tid); 950 ready = false; 951 } 952 953 if (ready) { 954 insertThread(tid); 955 956 contextSwitch = false; 957 958 cpuWaitList.remove(tid); 959 } else { 960 suspendContext(tid); 961 962 //blocks fetch 963 contextSwitch = true; 964 965 //@todo: dont always add to waitlist 966 //do waitlist 967 cpuWaitList.push_back(tid); 968 } 969} 970 971#if FULL_SYSTEM 972template <class Impl> 973Fault 974FullO3CPU<Impl>::hwrei(ThreadID tid) 975{ 976#if THE_ISA == ALPHA_ISA 977 // Need to clear the lock flag upon returning from an interrupt. 978 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 979 980 this->thread[tid]->kernelStats->hwrei(); 981 982 // FIXME: XXX check for interrupts? XXX 983#endif 984 return NoFault; 985} 986 987template <class Impl> 988bool 989FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 990{ 991#if THE_ISA == ALPHA_ISA 992 if (this->thread[tid]->kernelStats) 993 this->thread[tid]->kernelStats->callpal(palFunc, 994 this->threadContexts[tid]); 995 996 switch (palFunc) { 997 case PAL::halt: 998 halt(); 999 if (--System::numSystemsRunning == 0) 1000 exitSimLoop("all cpus halted"); 1001 break; 1002 1003 case PAL::bpt: 1004 case PAL::bugchk: 1005 if (this->system->breakpoint()) 1006 return false; 1007 break; 1008 } 1009#endif 1010 return true; 1011} 1012 1013template <class Impl> 1014Fault 1015FullO3CPU<Impl>::getInterrupts() 1016{ 1017 // Check if there are any outstanding interrupts 1018 return this->interrupts->getInterrupt(this->threadContexts[0]); 1019} 1020 1021template <class Impl> 1022void 1023FullO3CPU<Impl>::processInterrupts(Fault interrupt) 1024{ 1025 // Check for interrupts here. For now can copy the code that 1026 // exists within isa_fullsys_traits.hh. Also assume that thread 0 1027 // is the one that handles the interrupts. 1028 // @todo: Possibly consolidate the interrupt checking code. 1029 // @todo: Allow other threads to handle interrupts. 1030 1031 assert(interrupt != NoFault); 1032 this->interrupts->updateIntrInfo(this->threadContexts[0]); 1033 1034 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1035 this->trap(interrupt, 0, NULL); 1036} 1037 1038#endif 1039 1040template <class Impl> 1041void 1042FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst) 1043{ 1044 // Pass the thread's TC into the invoke method. 1045 fault->invoke(this->threadContexts[tid], inst); 1046} 1047 1048#if !FULL_SYSTEM 1049 1050template <class Impl> 1051void 1052FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid) 1053{ 1054 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1055 1056 DPRINTF(Activity,"Activity: syscall() called.\n"); 1057 1058 // Temporarily increase this by one to account for the syscall 1059 // instruction. 1060 ++(this->thread[tid]->funcExeInst); 1061 1062 // Execute the actual syscall. 1063 this->thread[tid]->syscall(callnum); 1064 1065 // Decrease funcExeInst by one as the normal commit will handle 1066 // incrementing it. 1067 --(this->thread[tid]->funcExeInst); 1068} 1069 1070#endif 1071 1072template <class Impl> 1073void 1074FullO3CPU<Impl>::serialize(std::ostream &os) 1075{ 1076 SimObject::State so_state = SimObject::getState(); 1077 SERIALIZE_ENUM(so_state); 1078 BaseCPU::serialize(os); 1079 nameOut(os, csprintf("%s.tickEvent", name())); 1080 tickEvent.serialize(os); 1081 1082 // Use SimpleThread's ability to checkpoint to make it easier to 1083 // write out the registers. Also make this static so it doesn't 1084 // get instantiated multiple times (causes a panic in statistics). 1085 static SimpleThread temp; 1086 1087 ThreadID size = thread.size(); 1088 for (ThreadID i = 0; i < size; i++) { 1089 nameOut(os, csprintf("%s.xc.%i", name(), i)); 1090 temp.copyTC(thread[i]->getTC()); 1091 temp.serialize(os); 1092 } 1093} 1094 1095template <class Impl> 1096void 1097FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 1098{ 1099 SimObject::State so_state; 1100 UNSERIALIZE_ENUM(so_state); 1101 BaseCPU::unserialize(cp, section); 1102 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 1103 1104 // Use SimpleThread's ability to checkpoint to make it easier to 1105 // read in the registers. Also make this static so it doesn't 1106 // get instantiated multiple times (causes a panic in statistics). 1107 static SimpleThread temp; 1108 1109 ThreadID size = thread.size(); 1110 for (ThreadID i = 0; i < size; i++) { 1111 temp.copyTC(thread[i]->getTC()); 1112 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 1113 thread[i]->getTC()->copyArchRegs(temp.getTC()); 1114 } 1115} 1116 1117template <class Impl> 1118unsigned int 1119FullO3CPU<Impl>::drain(Event *drain_event) 1120{ 1121 DPRINTF(O3CPU, "Switching out\n"); 1122 1123 // If the CPU isn't doing anything, then return immediately. 1124 if (_status == Idle || _status == SwitchedOut) { 1125 return 0; 1126 } 1127 1128 drainCount = 0; 1129 fetch.drain(); 1130 decode.drain(); 1131 rename.drain(); 1132 iew.drain(); 1133 commit.drain(); 1134 1135 // Wake the CPU and record activity so everything can drain out if 1136 // the CPU was not able to immediately drain. 1137 if (getState() != SimObject::Drained) { 1138 // A bit of a hack...set the drainEvent after all the drain() 1139 // calls have been made, that way if all of the stages drain 1140 // immediately, the signalDrained() function knows not to call 1141 // process on the drain event. 1142 drainEvent = drain_event; 1143 1144 wakeCPU(); 1145 activityRec.activity(); 1146 1147 return 1; 1148 } else { 1149 return 0; 1150 } 1151} 1152 1153template <class Impl> 1154void 1155FullO3CPU<Impl>::resume() 1156{ 1157 fetch.resume(); 1158 decode.resume(); 1159 rename.resume(); 1160 iew.resume(); 1161 commit.resume(); 1162 1163 changeState(SimObject::Running); 1164 1165 if (_status == SwitchedOut || _status == Idle) 1166 return; 1167 1168 assert(system->getMemoryMode() == Enums::timing); 1169 1170 if (!tickEvent.scheduled()) 1171 schedule(tickEvent, nextCycle()); 1172 _status = Running; 1173} 1174 1175template <class Impl> 1176void 1177FullO3CPU<Impl>::signalDrained() 1178{ 1179 if (++drainCount == NumStages) { 1180 if (tickEvent.scheduled()) 1181 tickEvent.squash(); 1182 1183 changeState(SimObject::Drained); 1184 1185 BaseCPU::switchOut(); 1186 1187 if (drainEvent) { 1188 drainEvent->process(); 1189 drainEvent = NULL; 1190 } 1191 } 1192 assert(drainCount <= 5); 1193} 1194 1195template <class Impl> 1196void 1197FullO3CPU<Impl>::switchOut() 1198{ 1199 fetch.switchOut(); 1200 rename.switchOut(); 1201 iew.switchOut(); 1202 commit.switchOut(); 1203 instList.clear(); 1204 while (!removeList.empty()) { 1205 removeList.pop(); 1206 } 1207 1208 _status = SwitchedOut; 1209#if USE_CHECKER 1210 if (checker) 1211 checker->switchOut(); 1212#endif 1213 if (tickEvent.scheduled()) 1214 tickEvent.squash(); 1215} 1216 1217template <class Impl> 1218void 1219FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1220{ 1221 // Flush out any old data from the time buffers. 1222 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1223 timeBuffer.advance(); 1224 fetchQueue.advance(); 1225 decodeQueue.advance(); 1226 renameQueue.advance(); 1227 iewQueue.advance(); 1228 } 1229 1230 activityRec.reset(); 1231 1232 BaseCPU::takeOverFrom(oldCPU, &icachePort, &dcachePort); 1233 1234 fetch.takeOverFrom(); 1235 decode.takeOverFrom(); 1236 rename.takeOverFrom(); 1237 iew.takeOverFrom(); 1238 commit.takeOverFrom(); 1239 1240 assert(!tickEvent.scheduled() || tickEvent.squashed()); 1241 1242 // @todo: Figure out how to properly select the tid to put onto 1243 // the active threads list. 1244 ThreadID tid = 0; 1245 1246 list<ThreadID>::iterator isActive = 1247 std::find(activeThreads.begin(), activeThreads.end(), tid); 1248 1249 if (isActive == activeThreads.end()) { 1250 //May Need to Re-code this if the delay variable is the delay 1251 //needed for thread to activate 1252 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 1253 tid); 1254 1255 activeThreads.push_back(tid); 1256 } 1257 1258 // Set all statuses to active, schedule the CPU's tick event. 1259 // @todo: Fix up statuses so this is handled properly 1260 ThreadID size = threadContexts.size(); 1261 for (ThreadID i = 0; i < size; ++i) { 1262 ThreadContext *tc = threadContexts[i]; 1263 if (tc->status() == ThreadContext::Active && _status != Running) { 1264 _status = Running; 1265 reschedule(tickEvent, nextCycle(), true); 1266 } 1267 } 1268 if (!tickEvent.scheduled()) 1269 schedule(tickEvent, nextCycle()); 1270 1271 lastRunningCycle = curTick(); 1272} 1273 1274template <class Impl> 1275TheISA::MiscReg 1276FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) 1277{ 1278 return this->isa[tid].readMiscRegNoEffect(misc_reg); 1279} 1280 1281template <class Impl> 1282TheISA::MiscReg 1283FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1284{ 1285 miscRegfileReads++; 1286 return this->isa[tid].readMiscReg(misc_reg, tcBase(tid)); 1287} 1288 1289template <class Impl> 1290void 1291FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1292 const TheISA::MiscReg &val, ThreadID tid) 1293{ 1294 this->isa[tid].setMiscRegNoEffect(misc_reg, val); 1295} 1296 1297template <class Impl> 1298void 1299FullO3CPU<Impl>::setMiscReg(int misc_reg, 1300 const TheISA::MiscReg &val, ThreadID tid) 1301{ 1302 miscRegfileWrites++; 1303 this->isa[tid].setMiscReg(misc_reg, val, tcBase(tid)); 1304} 1305 1306template <class Impl> 1307uint64_t 1308FullO3CPU<Impl>::readIntReg(int reg_idx) 1309{ 1310 intRegfileReads++; 1311 return regFile.readIntReg(reg_idx); 1312} 1313 1314template <class Impl> 1315FloatReg 1316FullO3CPU<Impl>::readFloatReg(int reg_idx) 1317{ 1318 fpRegfileReads++; 1319 return regFile.readFloatReg(reg_idx); 1320} 1321 1322template <class Impl> 1323FloatRegBits 1324FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1325{ 1326 fpRegfileReads++; 1327 return regFile.readFloatRegBits(reg_idx); 1328} 1329 1330template <class Impl> 1331void 1332FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1333{ 1334 intRegfileWrites++; 1335 regFile.setIntReg(reg_idx, val); 1336} 1337 1338template <class Impl> 1339void 1340FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1341{ 1342 fpRegfileWrites++; 1343 regFile.setFloatReg(reg_idx, val); 1344} 1345 1346template <class Impl> 1347void 1348FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1349{ 1350 fpRegfileWrites++; 1351 regFile.setFloatRegBits(reg_idx, val); 1352} 1353 1354template <class Impl> 1355uint64_t 1356FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1357{ 1358 intRegfileReads++; 1359 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1360 1361 return regFile.readIntReg(phys_reg); 1362} 1363 1364template <class Impl> 1365float 1366FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1367{ 1368 fpRegfileReads++; 1369 int idx = reg_idx + TheISA::NumIntRegs; 1370 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1371 1372 return regFile.readFloatReg(phys_reg); 1373} 1374 1375template <class Impl> 1376uint64_t 1377FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1378{ 1379 fpRegfileReads++; 1380 int idx = reg_idx + TheISA::NumIntRegs; 1381 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1382 1383 return regFile.readFloatRegBits(phys_reg); 1384} 1385 1386template <class Impl> 1387void 1388FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1389{ 1390 intRegfileWrites++; 1391 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1392 1393 regFile.setIntReg(phys_reg, val); 1394} 1395 1396template <class Impl> 1397void 1398FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1399{ 1400 fpRegfileWrites++; 1401 int idx = reg_idx + TheISA::NumIntRegs; 1402 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1403 1404 regFile.setFloatReg(phys_reg, val); 1405} 1406 1407template <class Impl> 1408void 1409FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1410{ 1411 fpRegfileWrites++; 1412 int idx = reg_idx + TheISA::NumIntRegs; 1413 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1414 1415 regFile.setFloatRegBits(phys_reg, val); 1416} 1417 1418template <class Impl> 1419TheISA::PCState 1420FullO3CPU<Impl>::pcState(ThreadID tid) 1421{ 1422 return commit.pcState(tid); 1423} 1424 1425template <class Impl> 1426void 1427FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1428{ 1429 commit.pcState(val, tid); 1430} 1431 1432template <class Impl> 1433Addr 1434FullO3CPU<Impl>::instAddr(ThreadID tid) 1435{ 1436 return commit.instAddr(tid); 1437} 1438 1439template <class Impl> 1440Addr 1441FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1442{ 1443 return commit.nextInstAddr(tid); 1444} 1445 1446template <class Impl> 1447MicroPC 1448FullO3CPU<Impl>::microPC(ThreadID tid) 1449{ 1450 return commit.microPC(tid); 1451} 1452 1453template <class Impl> 1454void 1455FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1456{ 1457 this->thread[tid]->inSyscall = true; 1458 this->commit.generateTCEvent(tid); 1459} 1460 1461template <class Impl> 1462typename FullO3CPU<Impl>::ListIt 1463FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1464{ 1465 instList.push_back(inst); 1466 1467 return --(instList.end()); 1468} 1469 1470template <class Impl> 1471void 1472FullO3CPU<Impl>::instDone(ThreadID tid) 1473{ 1474 // Keep an instruction count. 1475 thread[tid]->numInst++; 1476 thread[tid]->numInsts++; 1477 committedInsts[tid]++; 1478 totalCommittedInsts++; 1479 system->totalNumInsts++; 1480 // Check for instruction-count-based events. 1481 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1482 system->instEventQueue.serviceEvents(system->totalNumInsts); 1483} 1484 1485template <class Impl> 1486void 1487FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1488{ 1489 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1490 "[sn:%lli]\n", 1491 inst->threadNumber, inst->pcState(), inst->seqNum); 1492 1493 removeInstsThisCycle = true; 1494 1495 // Remove the front instruction. 1496 removeList.push(inst->getInstListIt()); 1497} 1498 1499template <class Impl> 1500void 1501FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1502{ 1503 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1504 " list.\n", tid); 1505 1506 ListIt end_it; 1507 1508 bool rob_empty = false; 1509 1510 if (instList.empty()) { 1511 return; 1512 } else if (rob.isEmpty(/*tid*/)) { 1513 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1514 end_it = instList.begin(); 1515 rob_empty = true; 1516 } else { 1517 end_it = (rob.readTailInst(tid))->getInstListIt(); 1518 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1519 } 1520 1521 removeInstsThisCycle = true; 1522 1523 ListIt inst_it = instList.end(); 1524 1525 inst_it--; 1526 1527 // Walk through the instruction list, removing any instructions 1528 // that were inserted after the given instruction iterator, end_it. 1529 while (inst_it != end_it) { 1530 assert(!instList.empty()); 1531 1532 squashInstIt(inst_it, tid); 1533 1534 inst_it--; 1535 } 1536 1537 // If the ROB was empty, then we actually need to remove the first 1538 // instruction as well. 1539 if (rob_empty) { 1540 squashInstIt(inst_it, tid); 1541 } 1542} 1543 1544template <class Impl> 1545void 1546FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1547{ 1548 assert(!instList.empty()); 1549 1550 removeInstsThisCycle = true; 1551 1552 ListIt inst_iter = instList.end(); 1553 1554 inst_iter--; 1555 1556 DPRINTF(O3CPU, "Deleting instructions from instruction " 1557 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1558 tid, seq_num, (*inst_iter)->seqNum); 1559 1560 while ((*inst_iter)->seqNum > seq_num) { 1561 1562 bool break_loop = (inst_iter == instList.begin()); 1563 1564 squashInstIt(inst_iter, tid); 1565 1566 inst_iter--; 1567 1568 if (break_loop) 1569 break; 1570 } 1571} 1572 1573template <class Impl> 1574inline void 1575FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1576{ 1577 if ((*instIt)->threadNumber == tid) { 1578 DPRINTF(O3CPU, "Squashing instruction, " 1579 "[tid:%i] [sn:%lli] PC %s\n", 1580 (*instIt)->threadNumber, 1581 (*instIt)->seqNum, 1582 (*instIt)->pcState()); 1583 1584 // Mark it as squashed. 1585 (*instIt)->setSquashed(); 1586 1587 // @todo: Formulate a consistent method for deleting 1588 // instructions from the instruction list 1589 // Remove the instruction from the list. 1590 removeList.push(instIt); 1591 } 1592} 1593 1594template <class Impl> 1595void 1596FullO3CPU<Impl>::cleanUpRemovedInsts() 1597{ 1598 while (!removeList.empty()) { 1599 DPRINTF(O3CPU, "Removing instruction, " 1600 "[tid:%i] [sn:%lli] PC %s\n", 1601 (*removeList.front())->threadNumber, 1602 (*removeList.front())->seqNum, 1603 (*removeList.front())->pcState()); 1604 1605 instList.erase(removeList.front()); 1606 1607 removeList.pop(); 1608 } 1609 1610 removeInstsThisCycle = false; 1611} 1612/* 1613template <class Impl> 1614void 1615FullO3CPU<Impl>::removeAllInsts() 1616{ 1617 instList.clear(); 1618} 1619*/ 1620template <class Impl> 1621void 1622FullO3CPU<Impl>::dumpInsts() 1623{ 1624 int num = 0; 1625 1626 ListIt inst_list_it = instList.begin(); 1627 1628 cprintf("Dumping Instruction List\n"); 1629 1630 while (inst_list_it != instList.end()) { 1631 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1632 "Squashed:%i\n\n", 1633 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1634 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1635 (*inst_list_it)->isSquashed()); 1636 inst_list_it++; 1637 ++num; 1638 } 1639} 1640/* 1641template <class Impl> 1642void 1643FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1644{ 1645 iew.wakeDependents(inst); 1646} 1647*/ 1648template <class Impl> 1649void 1650FullO3CPU<Impl>::wakeCPU() 1651{ 1652 if (activityRec.active() || tickEvent.scheduled()) { 1653 DPRINTF(Activity, "CPU already running.\n"); 1654 return; 1655 } 1656 1657 DPRINTF(Activity, "Waking up CPU\n"); 1658 1659 idleCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 1660 numCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 1661 1662 schedule(tickEvent, nextCycle()); 1663} 1664 1665#if FULL_SYSTEM 1666template <class Impl> 1667void 1668FullO3CPU<Impl>::wakeup() 1669{ 1670 if (this->thread[0]->status() != ThreadContext::Suspended) 1671 return; 1672 1673 this->wakeCPU(); 1674 1675 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1676 this->threadContexts[0]->activate(); 1677} 1678#endif 1679 1680template <class Impl> 1681ThreadID 1682FullO3CPU<Impl>::getFreeTid() 1683{ 1684 for (ThreadID tid = 0; tid < numThreads; tid++) { 1685 if (!tids[tid]) { 1686 tids[tid] = true; 1687 return tid; 1688 } 1689 } 1690 1691 return InvalidThreadID; 1692} 1693 1694template <class Impl> 1695void 1696FullO3CPU<Impl>::doContextSwitch() 1697{ 1698 if (contextSwitch) { 1699 1700 //ADD CODE TO DEACTIVE THREAD HERE (???) 1701 1702 ThreadID size = cpuWaitList.size(); 1703 for (ThreadID tid = 0; tid < size; tid++) { 1704 activateWhenReady(tid); 1705 } 1706 1707 if (cpuWaitList.size() == 0) 1708 contextSwitch = true; 1709 } 1710} 1711 1712template <class Impl> 1713void 1714FullO3CPU<Impl>::updateThreadPriority() 1715{ 1716 if (activeThreads.size() > 1) { 1717 //DEFAULT TO ROUND ROBIN SCHEME 1718 //e.g. Move highest priority to end of thread list 1719 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1720 1721 unsigned high_thread = *list_begin; 1722 1723 activeThreads.erase(list_begin); 1724 1725 activeThreads.push_back(high_thread); 1726 } 1727} 1728 1729// Forward declaration of FullO3CPU. 1730template class FullO3CPU<O3CPUImpl>; 1731