cpu.cc revision 8737:770ccf3af571
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#include "cpu/checker/thread_context.hh" 72#endif 73 74#if THE_ISA == ALPHA_ISA 75#include "arch/alpha/osfpal.hh" 76#include "debug/Activity.hh" 77#endif 78 79struct BaseCPUParams; 80 81using namespace TheISA; 82using namespace std; 83 84BaseO3CPU::BaseO3CPU(BaseCPUParams *params) 85 : BaseCPU(params) 86{ 87} 88 89void 90BaseO3CPU::regStats() 91{ 92 BaseCPU::regStats(); 93} 94 95template<class Impl> 96bool 97FullO3CPU<Impl>::IcachePort::recvTiming(PacketPtr pkt) 98{ 99 DPRINTF(O3CPU, "Fetch unit received timing\n"); 100 if (pkt->isResponse()) { 101 // We shouldn't ever get a block in ownership state 102 assert(!(pkt->memInhibitAsserted() && !pkt->sharedAsserted())); 103 104 fetch->processCacheCompletion(pkt); 105 } 106 //else Snooped a coherence request, just return 107 return true; 108} 109 110template<class Impl> 111void 112FullO3CPU<Impl>::IcachePort::recvRetry() 113{ 114 fetch->recvRetry(); 115} 116 117template <class Impl> 118bool 119FullO3CPU<Impl>::DcachePort::recvTiming(PacketPtr pkt) 120{ 121 return lsq->recvTiming(pkt); 122} 123 124template <class Impl> 125void 126FullO3CPU<Impl>::DcachePort::recvRetry() 127{ 128 lsq->recvRetry(); 129} 130 131template <class Impl> 132FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 133 : Event(CPU_Tick_Pri), cpu(c) 134{ 135} 136 137template <class Impl> 138void 139FullO3CPU<Impl>::TickEvent::process() 140{ 141 cpu->tick(); 142} 143 144template <class Impl> 145const char * 146FullO3CPU<Impl>::TickEvent::description() const 147{ 148 return "FullO3CPU tick"; 149} 150 151template <class Impl> 152FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent() 153 : Event(CPU_Switch_Pri) 154{ 155} 156 157template <class Impl> 158void 159FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num, 160 FullO3CPU<Impl> *thread_cpu) 161{ 162 tid = thread_num; 163 cpu = thread_cpu; 164} 165 166template <class Impl> 167void 168FullO3CPU<Impl>::ActivateThreadEvent::process() 169{ 170 cpu->activateThread(tid); 171} 172 173template <class Impl> 174const char * 175FullO3CPU<Impl>::ActivateThreadEvent::description() const 176{ 177 return "FullO3CPU \"Activate Thread\""; 178} 179 180template <class Impl> 181FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent() 182 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL) 183{ 184} 185 186template <class Impl> 187void 188FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num, 189 FullO3CPU<Impl> *thread_cpu) 190{ 191 tid = thread_num; 192 cpu = thread_cpu; 193 remove = false; 194} 195 196template <class Impl> 197void 198FullO3CPU<Impl>::DeallocateContextEvent::process() 199{ 200 cpu->deactivateThread(tid); 201 if (remove) 202 cpu->removeThread(tid); 203} 204 205template <class Impl> 206const char * 207FullO3CPU<Impl>::DeallocateContextEvent::description() const 208{ 209 return "FullO3CPU \"Deallocate Context\""; 210} 211 212template <class Impl> 213FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params) 214 : BaseO3CPU(params), 215 itb(params->itb), 216 dtb(params->dtb), 217 tickEvent(this), 218#ifndef NDEBUG 219 instcount(0), 220#endif 221 removeInstsThisCycle(false), 222 fetch(this, params), 223 decode(this, params), 224 rename(this, params), 225 iew(this, params), 226 commit(this, params), 227 228 regFile(this, params->numPhysIntRegs, 229 params->numPhysFloatRegs), 230 231 freeList(params->numThreads, 232 TheISA::NumIntRegs, params->numPhysIntRegs, 233 TheISA::NumFloatRegs, params->numPhysFloatRegs), 234 235 rob(this, 236 params->numROBEntries, params->squashWidth, 237 params->smtROBPolicy, params->smtROBThreshold, 238 params->numThreads), 239 240 scoreboard(params->numThreads, 241 TheISA::NumIntRegs, params->numPhysIntRegs, 242 TheISA::NumFloatRegs, params->numPhysFloatRegs, 243 TheISA::NumMiscRegs * numThreads, 244 TheISA::ZeroReg), 245 246 icachePort(&fetch, this), 247 dcachePort(&iew.ldstQueue, this), 248 249 timeBuffer(params->backComSize, params->forwardComSize), 250 fetchQueue(params->backComSize, params->forwardComSize), 251 decodeQueue(params->backComSize, params->forwardComSize), 252 renameQueue(params->backComSize, params->forwardComSize), 253 iewQueue(params->backComSize, params->forwardComSize), 254 activityRec(name(), NumStages, 255 params->backComSize + params->forwardComSize, 256 params->activity), 257 258 globalSeqNum(1), 259 system(params->system), 260 drainCount(0), 261 deferRegistration(params->defer_registration) 262{ 263 if (!deferRegistration) { 264 _status = Running; 265 } else { 266 _status = Idle; 267 } 268 269#if USE_CHECKER 270 if (params->checker) { 271 BaseCPU *temp_checker = params->checker; 272 checker = dynamic_cast<Checker<Impl> *>(temp_checker); 273 checker->setIcachePort(&icachePort); 274#if FULL_SYSTEM 275 checker->setSystem(params->system); 276#endif 277 } else { 278 checker = NULL; 279 } 280#endif // USE_CHECKER 281 282#if !FULL_SYSTEM 283 thread.resize(numThreads); 284 tids.resize(numThreads); 285#endif 286 287 // The stages also need their CPU pointer setup. However this 288 // must be done at the upper level CPU because they have pointers 289 // to the upper level CPU, and not this FullO3CPU. 290 291 // Set up Pointers to the activeThreads list for each stage 292 fetch.setActiveThreads(&activeThreads); 293 decode.setActiveThreads(&activeThreads); 294 rename.setActiveThreads(&activeThreads); 295 iew.setActiveThreads(&activeThreads); 296 commit.setActiveThreads(&activeThreads); 297 298 // Give each of the stages the time buffer they will use. 299 fetch.setTimeBuffer(&timeBuffer); 300 decode.setTimeBuffer(&timeBuffer); 301 rename.setTimeBuffer(&timeBuffer); 302 iew.setTimeBuffer(&timeBuffer); 303 commit.setTimeBuffer(&timeBuffer); 304 305 // Also setup each of the stages' queues. 306 fetch.setFetchQueue(&fetchQueue); 307 decode.setFetchQueue(&fetchQueue); 308 commit.setFetchQueue(&fetchQueue); 309 decode.setDecodeQueue(&decodeQueue); 310 rename.setDecodeQueue(&decodeQueue); 311 rename.setRenameQueue(&renameQueue); 312 iew.setRenameQueue(&renameQueue); 313 iew.setIEWQueue(&iewQueue); 314 commit.setIEWQueue(&iewQueue); 315 commit.setRenameQueue(&renameQueue); 316 317 commit.setIEWStage(&iew); 318 rename.setIEWStage(&iew); 319 rename.setCommitStage(&commit); 320 321#if !FULL_SYSTEM 322 ThreadID active_threads = params->workload.size(); 323 324 if (active_threads > Impl::MaxThreads) { 325 panic("Workload Size too large. Increase the 'MaxThreads'" 326 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) or " 327 "edit your workload size."); 328 } 329#else 330 ThreadID active_threads = 1; 331#endif 332 333 //Make Sure That this a Valid Architeture 334 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 335 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 336 337 rename.setScoreboard(&scoreboard); 338 iew.setScoreboard(&scoreboard); 339 340 // Setup the rename map for whichever stages need it. 341 PhysRegIndex lreg_idx = 0; 342 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 343 344 for (ThreadID tid = 0; tid < numThreads; tid++) { 345 bool bindRegs = (tid <= active_threads - 1); 346 347 commitRenameMap[tid].init(TheISA::NumIntRegs, 348 params->numPhysIntRegs, 349 lreg_idx, //Index for Logical. Regs 350 351 TheISA::NumFloatRegs, 352 params->numPhysFloatRegs, 353 freg_idx, //Index for Float Regs 354 355 TheISA::NumMiscRegs, 356 357 TheISA::ZeroReg, 358 TheISA::ZeroReg, 359 360 tid, 361 false); 362 363 renameMap[tid].init(TheISA::NumIntRegs, 364 params->numPhysIntRegs, 365 lreg_idx, //Index for Logical. Regs 366 367 TheISA::NumFloatRegs, 368 params->numPhysFloatRegs, 369 freg_idx, //Index for Float Regs 370 371 TheISA::NumMiscRegs, 372 373 TheISA::ZeroReg, 374 TheISA::ZeroReg, 375 376 tid, 377 bindRegs); 378 379 activateThreadEvent[tid].init(tid, this); 380 deallocateContextEvent[tid].init(tid, this); 381 } 382 383 rename.setRenameMap(renameMap); 384 commit.setRenameMap(commitRenameMap); 385 386 // Give renameMap & rename stage access to the freeList; 387 for (ThreadID tid = 0; tid < numThreads; tid++) 388 renameMap[tid].setFreeList(&freeList); 389 rename.setFreeList(&freeList); 390 391 // Setup the ROB for whichever stages need it. 392 commit.setROB(&rob); 393 394 lastRunningCycle = curTick(); 395 396 lastActivatedCycle = -1; 397#if 0 398 // Give renameMap & rename stage access to the freeList; 399 for (ThreadID tid = 0; tid < numThreads; tid++) 400 globalSeqNum[tid] = 1; 401#endif 402 403 contextSwitch = false; 404 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 405 406 // Setup any thread state. 407 this->thread.resize(this->numThreads); 408 409 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 410#if FULL_SYSTEM 411 // SMT is not supported in FS mode yet. 412 assert(this->numThreads == 1); 413 this->thread[tid] = new Thread(this, 0); 414#else 415 if (tid < params->workload.size()) { 416 DPRINTF(O3CPU, "Workload[%i] process is %#x", 417 tid, this->thread[tid]); 418 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 419 (typename Impl::O3CPU *)(this), 420 tid, params->workload[tid]); 421 422 //usedTids[tid] = true; 423 //threadMap[tid] = tid; 424 } else { 425 //Allocate Empty thread so M5 can use later 426 //when scheduling threads to CPU 427 Process* dummy_proc = NULL; 428 429 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 430 (typename Impl::O3CPU *)(this), 431 tid, dummy_proc); 432 //usedTids[tid] = false; 433 } 434#endif // !FULL_SYSTEM 435 436 ThreadContext *tc; 437 438 // Setup the TC that will serve as the interface to the threads/CPU. 439 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 440 441 tc = o3_tc; 442 443 // If we're using a checker, then the TC should be the 444 // CheckerThreadContext. 445#if USE_CHECKER 446 if (params->checker) { 447 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 448 o3_tc, this->checker); 449 } 450#endif 451 452 o3_tc->cpu = (typename Impl::O3CPU *)(this); 453 assert(o3_tc->cpu); 454 o3_tc->thread = this->thread[tid]; 455 456#if FULL_SYSTEM 457 // Setup quiesce event. 458 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 459#endif 460 // Give the thread the TC. 461 this->thread[tid]->tc = tc; 462 463 // Add the TC to the CPU's list of TC's. 464 this->threadContexts.push_back(tc); 465 } 466 467 for (ThreadID tid = 0; tid < this->numThreads; tid++) 468 this->thread[tid]->setFuncExeInst(0); 469 470 lockAddr = 0; 471 lockFlag = false; 472} 473 474template <class Impl> 475FullO3CPU<Impl>::~FullO3CPU() 476{ 477} 478 479template <class Impl> 480void 481FullO3CPU<Impl>::regStats() 482{ 483 BaseO3CPU::regStats(); 484 485 // Register any of the O3CPU's stats here. 486 timesIdled 487 .name(name() + ".timesIdled") 488 .desc("Number of times that the entire CPU went into an idle state and" 489 " unscheduled itself") 490 .prereq(timesIdled); 491 492 idleCycles 493 .name(name() + ".idleCycles") 494 .desc("Total number of cycles that the CPU has spent unscheduled due " 495 "to idling") 496 .prereq(idleCycles); 497 498 quiesceCycles 499 .name(name() + ".quiesceCycles") 500 .desc("Total number of cycles that CPU has spent quiesced or waiting " 501 "for an interrupt") 502 .prereq(quiesceCycles); 503 504 // Number of Instructions simulated 505 // -------------------------------- 506 // Should probably be in Base CPU but need templated 507 // MaxThreads so put in here instead 508 committedInsts 509 .init(numThreads) 510 .name(name() + ".committedInsts") 511 .desc("Number of Instructions Simulated"); 512 513 totalCommittedInsts 514 .name(name() + ".committedInsts_total") 515 .desc("Number of Instructions Simulated"); 516 517 cpi 518 .name(name() + ".cpi") 519 .desc("CPI: Cycles Per Instruction") 520 .precision(6); 521 cpi = numCycles / committedInsts; 522 523 totalCpi 524 .name(name() + ".cpi_total") 525 .desc("CPI: Total CPI of All Threads") 526 .precision(6); 527 totalCpi = numCycles / totalCommittedInsts; 528 529 ipc 530 .name(name() + ".ipc") 531 .desc("IPC: Instructions Per Cycle") 532 .precision(6); 533 ipc = committedInsts / numCycles; 534 535 totalIpc 536 .name(name() + ".ipc_total") 537 .desc("IPC: Total IPC of All Threads") 538 .precision(6); 539 totalIpc = totalCommittedInsts / numCycles; 540 541 this->fetch.regStats(); 542 this->decode.regStats(); 543 this->rename.regStats(); 544 this->iew.regStats(); 545 this->commit.regStats(); 546 this->rob.regStats(); 547 548 intRegfileReads 549 .name(name() + ".int_regfile_reads") 550 .desc("number of integer regfile reads") 551 .prereq(intRegfileReads); 552 553 intRegfileWrites 554 .name(name() + ".int_regfile_writes") 555 .desc("number of integer regfile writes") 556 .prereq(intRegfileWrites); 557 558 fpRegfileReads 559 .name(name() + ".fp_regfile_reads") 560 .desc("number of floating regfile reads") 561 .prereq(fpRegfileReads); 562 563 fpRegfileWrites 564 .name(name() + ".fp_regfile_writes") 565 .desc("number of floating regfile writes") 566 .prereq(fpRegfileWrites); 567 568 miscRegfileReads 569 .name(name() + ".misc_regfile_reads") 570 .desc("number of misc regfile reads") 571 .prereq(miscRegfileReads); 572 573 miscRegfileWrites 574 .name(name() + ".misc_regfile_writes") 575 .desc("number of misc regfile writes") 576 .prereq(miscRegfileWrites); 577} 578 579template <class Impl> 580Port * 581FullO3CPU<Impl>::getPort(const std::string &if_name, int idx) 582{ 583 if (if_name == "dcache_port") 584 return &dcachePort; 585 else if (if_name == "icache_port") 586 return &icachePort; 587 else 588 panic("No Such Port\n"); 589} 590 591template <class Impl> 592void 593FullO3CPU<Impl>::tick() 594{ 595 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 596 597 ++numCycles; 598 599// activity = false; 600 601 //Tick each of the stages 602 fetch.tick(); 603 604 decode.tick(); 605 606 rename.tick(); 607 608 iew.tick(); 609 610 commit.tick(); 611 612#if !FULL_SYSTEM 613 doContextSwitch(); 614#endif 615 616 // Now advance the time buffers 617 timeBuffer.advance(); 618 619 fetchQueue.advance(); 620 decodeQueue.advance(); 621 renameQueue.advance(); 622 iewQueue.advance(); 623 624 activityRec.advance(); 625 626 if (removeInstsThisCycle) { 627 cleanUpRemovedInsts(); 628 } 629 630 if (!tickEvent.scheduled()) { 631 if (_status == SwitchedOut || 632 getState() == SimObject::Drained) { 633 DPRINTF(O3CPU, "Switched out!\n"); 634 // increment stat 635 lastRunningCycle = curTick(); 636 } else if (!activityRec.active() || _status == Idle) { 637 DPRINTF(O3CPU, "Idle!\n"); 638 lastRunningCycle = curTick(); 639 timesIdled++; 640 } else { 641 schedule(tickEvent, nextCycle(curTick() + ticks(1))); 642 DPRINTF(O3CPU, "Scheduling next tick!\n"); 643 } 644 } 645 646#if !FULL_SYSTEM 647 updateThreadPriority(); 648#endif 649} 650 651template <class Impl> 652void 653FullO3CPU<Impl>::init() 654{ 655 BaseCPU::init(); 656 657 // Set inSyscall so that the CPU doesn't squash when initially 658 // setting up registers. 659 for (ThreadID tid = 0; tid < numThreads; ++tid) 660 thread[tid]->inSyscall = true; 661 662 // this CPU could still be unconnected if we are restoring from a 663 // checkpoint and this CPU is to be switched in, thus we can only 664 // do this here if the instruction port is actually connected, if 665 // not we have to do it as part of takeOverFrom 666 if (icachePort.isConnected()) 667 fetch.setIcache(); 668 669#if FULL_SYSTEM 670 for (ThreadID tid = 0; tid < numThreads; tid++) { 671 ThreadContext *src_tc = threadContexts[tid]; 672 TheISA::initCPU(src_tc, src_tc->contextId()); 673 // Initialise the ThreadContext's memory proxies 674 thread[tid]->initMemProxies(thread[tid]->getTC()); 675 } 676#endif 677 678 // Clear inSyscall. 679 for (int tid = 0; tid < numThreads; ++tid) 680 thread[tid]->inSyscall = false; 681 682 // Initialize stages. 683 fetch.initStage(); 684 iew.initStage(); 685 rename.initStage(); 686 commit.initStage(); 687 688 commit.setThreads(thread); 689} 690 691template <class Impl> 692void 693FullO3CPU<Impl>::activateThread(ThreadID tid) 694{ 695 list<ThreadID>::iterator isActive = 696 std::find(activeThreads.begin(), activeThreads.end(), tid); 697 698 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 699 700 if (isActive == activeThreads.end()) { 701 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 702 tid); 703 704 activeThreads.push_back(tid); 705 } 706} 707 708template <class Impl> 709void 710FullO3CPU<Impl>::deactivateThread(ThreadID tid) 711{ 712 //Remove From Active List, if Active 713 list<ThreadID>::iterator thread_it = 714 std::find(activeThreads.begin(), activeThreads.end(), tid); 715 716 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 717 718 if (thread_it != activeThreads.end()) { 719 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 720 tid); 721 activeThreads.erase(thread_it); 722 } 723} 724 725template <class Impl> 726Counter 727FullO3CPU<Impl>::totalInstructions() const 728{ 729 Counter total(0); 730 731 ThreadID size = thread.size(); 732 for (ThreadID i = 0; i < size; i++) 733 total += thread[i]->numInst; 734 735 return total; 736} 737 738template <class Impl> 739void 740FullO3CPU<Impl>::activateContext(ThreadID tid, int delay) 741{ 742 // Needs to set each stage to running as well. 743 if (delay){ 744 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 745 "on cycle %d\n", tid, curTick() + ticks(delay)); 746 scheduleActivateThreadEvent(tid, delay); 747 } else { 748 activateThread(tid); 749 } 750 751 if (lastActivatedCycle < curTick()) { 752 scheduleTickEvent(delay); 753 754 // Be sure to signal that there's some activity so the CPU doesn't 755 // deschedule itself. 756 activityRec.activity(); 757 fetch.wakeFromQuiesce(); 758 759 quiesceCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 760 761 lastActivatedCycle = curTick(); 762 763 _status = Running; 764 } 765} 766 767template <class Impl> 768bool 769FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove, 770 int delay) 771{ 772 // Schedule removal of thread data from CPU 773 if (delay){ 774 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 775 "on cycle %d\n", tid, curTick() + ticks(delay)); 776 scheduleDeallocateContextEvent(tid, remove, delay); 777 return false; 778 } else { 779 deactivateThread(tid); 780 if (remove) 781 removeThread(tid); 782 return true; 783 } 784} 785 786template <class Impl> 787void 788FullO3CPU<Impl>::suspendContext(ThreadID tid) 789{ 790 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 791 bool deallocated = scheduleDeallocateContext(tid, false, 1); 792 // If this was the last thread then unschedule the tick event. 793 if ((activeThreads.size() == 1 && !deallocated) || 794 activeThreads.size() == 0) 795 unscheduleTickEvent(); 796 797 DPRINTF(Quiesce, "Suspending Context\n"); 798 lastRunningCycle = curTick(); 799 _status = Idle; 800} 801 802template <class Impl> 803void 804FullO3CPU<Impl>::haltContext(ThreadID tid) 805{ 806 //For now, this is the same as deallocate 807 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 808 scheduleDeallocateContext(tid, true, 1); 809} 810 811template <class Impl> 812void 813FullO3CPU<Impl>::insertThread(ThreadID tid) 814{ 815 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 816 // Will change now that the PC and thread state is internal to the CPU 817 // and not in the ThreadContext. 818#if FULL_SYSTEM 819 ThreadContext *src_tc = system->threadContexts[tid]; 820#else 821 ThreadContext *src_tc = tcBase(tid); 822#endif 823 824 //Bind Int Regs to Rename Map 825 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 826 PhysRegIndex phys_reg = freeList.getIntReg(); 827 828 renameMap[tid].setEntry(ireg,phys_reg); 829 scoreboard.setReg(phys_reg); 830 } 831 832 //Bind Float Regs to Rename Map 833 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 834 PhysRegIndex phys_reg = freeList.getFloatReg(); 835 836 renameMap[tid].setEntry(freg,phys_reg); 837 scoreboard.setReg(phys_reg); 838 } 839 840 //Copy Thread Data Into RegFile 841 //this->copyFromTC(tid); 842 843 //Set PC/NPC/NNPC 844 pcState(src_tc->pcState(), tid); 845 846 src_tc->setStatus(ThreadContext::Active); 847 848 activateContext(tid,1); 849 850 //Reset ROB/IQ/LSQ Entries 851 commit.rob->resetEntries(); 852 iew.resetEntries(); 853} 854 855template <class Impl> 856void 857FullO3CPU<Impl>::removeThread(ThreadID tid) 858{ 859 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 860 861 // Copy Thread Data From RegFile 862 // If thread is suspended, it might be re-allocated 863 // this->copyToTC(tid); 864 865 866 // @todo: 2-27-2008: Fix how we free up rename mappings 867 // here to alleviate the case for double-freeing registers 868 // in SMT workloads. 869 870 // Unbind Int Regs from Rename Map 871 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 872 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 873 874 scoreboard.unsetReg(phys_reg); 875 freeList.addReg(phys_reg); 876 } 877 878 // Unbind Float Regs from Rename Map 879 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) { 880 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 881 882 scoreboard.unsetReg(phys_reg); 883 freeList.addReg(phys_reg); 884 } 885 886 // Squash Throughout Pipeline 887 DynInstPtr inst = commit.rob->readHeadInst(tid); 888 InstSeqNum squash_seq_num = inst->seqNum; 889 fetch.squash(0, squash_seq_num, inst, tid); 890 decode.squash(tid); 891 rename.squash(squash_seq_num, tid); 892 iew.squash(tid); 893 iew.ldstQueue.squash(squash_seq_num, tid); 894 commit.rob->squash(squash_seq_num, tid); 895 896 897 assert(iew.instQueue.getCount(tid) == 0); 898 assert(iew.ldstQueue.getCount(tid) == 0); 899 900 // Reset ROB/IQ/LSQ Entries 901 902 // Commented out for now. This should be possible to do by 903 // telling all the pipeline stages to drain first, and then 904 // checking until the drain completes. Once the pipeline is 905 // drained, call resetEntries(). - 10-09-06 ktlim 906/* 907 if (activeThreads.size() >= 1) { 908 commit.rob->resetEntries(); 909 iew.resetEntries(); 910 } 911*/ 912} 913 914 915template <class Impl> 916void 917FullO3CPU<Impl>::activateWhenReady(ThreadID tid) 918{ 919 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 920 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 921 tid); 922 923 bool ready = true; 924 925 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 926 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 927 "Phys. Int. Regs.\n", 928 tid); 929 ready = false; 930 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 931 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 932 "Phys. Float. Regs.\n", 933 tid); 934 ready = false; 935 } else if (commit.rob->numFreeEntries() >= 936 commit.rob->entryAmount(activeThreads.size() + 1)) { 937 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 938 "ROB entries.\n", 939 tid); 940 ready = false; 941 } else if (iew.instQueue.numFreeEntries() >= 942 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 943 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 944 "IQ entries.\n", 945 tid); 946 ready = false; 947 } else if (iew.ldstQueue.numFreeEntries() >= 948 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 949 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 950 "LSQ entries.\n", 951 tid); 952 ready = false; 953 } 954 955 if (ready) { 956 insertThread(tid); 957 958 contextSwitch = false; 959 960 cpuWaitList.remove(tid); 961 } else { 962 suspendContext(tid); 963 964 //blocks fetch 965 contextSwitch = true; 966 967 //@todo: dont always add to waitlist 968 //do waitlist 969 cpuWaitList.push_back(tid); 970 } 971} 972 973#if FULL_SYSTEM 974template <class Impl> 975Fault 976FullO3CPU<Impl>::hwrei(ThreadID tid) 977{ 978#if THE_ISA == ALPHA_ISA 979 // Need to clear the lock flag upon returning from an interrupt. 980 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 981 982 this->thread[tid]->kernelStats->hwrei(); 983 984 // FIXME: XXX check for interrupts? XXX 985#endif 986 return NoFault; 987} 988 989template <class Impl> 990bool 991FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 992{ 993#if THE_ISA == ALPHA_ISA 994 if (this->thread[tid]->kernelStats) 995 this->thread[tid]->kernelStats->callpal(palFunc, 996 this->threadContexts[tid]); 997 998 switch (palFunc) { 999 case PAL::halt: 1000 halt(); 1001 if (--System::numSystemsRunning == 0) 1002 exitSimLoop("all cpus halted"); 1003 break; 1004 1005 case PAL::bpt: 1006 case PAL::bugchk: 1007 if (this->system->breakpoint()) 1008 return false; 1009 break; 1010 } 1011#endif 1012 return true; 1013} 1014 1015template <class Impl> 1016Fault 1017FullO3CPU<Impl>::getInterrupts() 1018{ 1019 // Check if there are any outstanding interrupts 1020 return this->interrupts->getInterrupt(this->threadContexts[0]); 1021} 1022 1023template <class Impl> 1024void 1025FullO3CPU<Impl>::processInterrupts(Fault interrupt) 1026{ 1027 // Check for interrupts here. For now can copy the code that 1028 // exists within isa_fullsys_traits.hh. Also assume that thread 0 1029 // is the one that handles the interrupts. 1030 // @todo: Possibly consolidate the interrupt checking code. 1031 // @todo: Allow other threads to handle interrupts. 1032 1033 assert(interrupt != NoFault); 1034 this->interrupts->updateIntrInfo(this->threadContexts[0]); 1035 1036 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1037 this->trap(interrupt, 0, NULL); 1038} 1039 1040#endif 1041 1042template <class Impl> 1043void 1044FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst) 1045{ 1046 // Pass the thread's TC into the invoke method. 1047 fault->invoke(this->threadContexts[tid], inst); 1048} 1049 1050#if !FULL_SYSTEM 1051 1052template <class Impl> 1053void 1054FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid) 1055{ 1056 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1057 1058 DPRINTF(Activity,"Activity: syscall() called.\n"); 1059 1060 // Temporarily increase this by one to account for the syscall 1061 // instruction. 1062 ++(this->thread[tid]->funcExeInst); 1063 1064 // Execute the actual syscall. 1065 this->thread[tid]->syscall(callnum); 1066 1067 // Decrease funcExeInst by one as the normal commit will handle 1068 // incrementing it. 1069 --(this->thread[tid]->funcExeInst); 1070} 1071 1072#endif 1073 1074template <class Impl> 1075void 1076FullO3CPU<Impl>::serialize(std::ostream &os) 1077{ 1078 SimObject::State so_state = SimObject::getState(); 1079 SERIALIZE_ENUM(so_state); 1080 BaseCPU::serialize(os); 1081 nameOut(os, csprintf("%s.tickEvent", name())); 1082 tickEvent.serialize(os); 1083 1084 // Use SimpleThread's ability to checkpoint to make it easier to 1085 // write out the registers. Also make this static so it doesn't 1086 // get instantiated multiple times (causes a panic in statistics). 1087 static SimpleThread temp; 1088 1089 ThreadID size = thread.size(); 1090 for (ThreadID i = 0; i < size; i++) { 1091 nameOut(os, csprintf("%s.xc.%i", name(), i)); 1092 temp.copyTC(thread[i]->getTC()); 1093 temp.serialize(os); 1094 } 1095} 1096 1097template <class Impl> 1098void 1099FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 1100{ 1101 SimObject::State so_state; 1102 UNSERIALIZE_ENUM(so_state); 1103 BaseCPU::unserialize(cp, section); 1104 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 1105 1106 // Use SimpleThread's ability to checkpoint to make it easier to 1107 // read in the registers. Also make this static so it doesn't 1108 // get instantiated multiple times (causes a panic in statistics). 1109 static SimpleThread temp; 1110 1111 ThreadID size = thread.size(); 1112 for (ThreadID i = 0; i < size; i++) { 1113 temp.copyTC(thread[i]->getTC()); 1114 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 1115 thread[i]->getTC()->copyArchRegs(temp.getTC()); 1116 } 1117} 1118 1119template <class Impl> 1120unsigned int 1121FullO3CPU<Impl>::drain(Event *drain_event) 1122{ 1123 DPRINTF(O3CPU, "Switching out\n"); 1124 1125 // If the CPU isn't doing anything, then return immediately. 1126 if (_status == Idle || _status == SwitchedOut) { 1127 return 0; 1128 } 1129 1130 drainCount = 0; 1131 fetch.drain(); 1132 decode.drain(); 1133 rename.drain(); 1134 iew.drain(); 1135 commit.drain(); 1136 1137 // Wake the CPU and record activity so everything can drain out if 1138 // the CPU was not able to immediately drain. 1139 if (getState() != SimObject::Drained) { 1140 // A bit of a hack...set the drainEvent after all the drain() 1141 // calls have been made, that way if all of the stages drain 1142 // immediately, the signalDrained() function knows not to call 1143 // process on the drain event. 1144 drainEvent = drain_event; 1145 1146 wakeCPU(); 1147 activityRec.activity(); 1148 1149 return 1; 1150 } else { 1151 return 0; 1152 } 1153} 1154 1155template <class Impl> 1156void 1157FullO3CPU<Impl>::resume() 1158{ 1159 fetch.resume(); 1160 decode.resume(); 1161 rename.resume(); 1162 iew.resume(); 1163 commit.resume(); 1164 1165 changeState(SimObject::Running); 1166 1167 if (_status == SwitchedOut || _status == Idle) 1168 return; 1169 1170 assert(system->getMemoryMode() == Enums::timing); 1171 1172 if (!tickEvent.scheduled()) 1173 schedule(tickEvent, nextCycle()); 1174 _status = Running; 1175} 1176 1177template <class Impl> 1178void 1179FullO3CPU<Impl>::signalDrained() 1180{ 1181 if (++drainCount == NumStages) { 1182 if (tickEvent.scheduled()) 1183 tickEvent.squash(); 1184 1185 changeState(SimObject::Drained); 1186 1187 BaseCPU::switchOut(); 1188 1189 if (drainEvent) { 1190 drainEvent->process(); 1191 drainEvent = NULL; 1192 } 1193 } 1194 assert(drainCount <= 5); 1195} 1196 1197template <class Impl> 1198void 1199FullO3CPU<Impl>::switchOut() 1200{ 1201 fetch.switchOut(); 1202 rename.switchOut(); 1203 iew.switchOut(); 1204 commit.switchOut(); 1205 instList.clear(); 1206 while (!removeList.empty()) { 1207 removeList.pop(); 1208 } 1209 1210 _status = SwitchedOut; 1211#if USE_CHECKER 1212 if (checker) 1213 checker->switchOut(); 1214#endif 1215 if (tickEvent.scheduled()) 1216 tickEvent.squash(); 1217} 1218 1219template <class Impl> 1220void 1221FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1222{ 1223 // Flush out any old data from the time buffers. 1224 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1225 timeBuffer.advance(); 1226 fetchQueue.advance(); 1227 decodeQueue.advance(); 1228 renameQueue.advance(); 1229 iewQueue.advance(); 1230 } 1231 1232 activityRec.reset(); 1233 1234 BaseCPU::takeOverFrom(oldCPU); 1235 1236 fetch.takeOverFrom(); 1237 decode.takeOverFrom(); 1238 rename.takeOverFrom(); 1239 iew.takeOverFrom(); 1240 commit.takeOverFrom(); 1241 1242 assert(!tickEvent.scheduled() || tickEvent.squashed()); 1243 1244 // @todo: Figure out how to properly select the tid to put onto 1245 // the active threads list. 1246 ThreadID tid = 0; 1247 1248 list<ThreadID>::iterator isActive = 1249 std::find(activeThreads.begin(), activeThreads.end(), tid); 1250 1251 if (isActive == activeThreads.end()) { 1252 //May Need to Re-code this if the delay variable is the delay 1253 //needed for thread to activate 1254 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 1255 tid); 1256 1257 activeThreads.push_back(tid); 1258 } 1259 1260 // Set all statuses to active, schedule the CPU's tick event. 1261 // @todo: Fix up statuses so this is handled properly 1262 ThreadID size = threadContexts.size(); 1263 for (ThreadID i = 0; i < size; ++i) { 1264 ThreadContext *tc = threadContexts[i]; 1265 if (tc->status() == ThreadContext::Active && _status != Running) { 1266 _status = Running; 1267 reschedule(tickEvent, nextCycle(), true); 1268 } 1269 } 1270 if (!tickEvent.scheduled()) 1271 schedule(tickEvent, nextCycle()); 1272 1273 lastRunningCycle = curTick(); 1274} 1275 1276template <class Impl> 1277TheISA::MiscReg 1278FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) 1279{ 1280 return this->isa[tid].readMiscRegNoEffect(misc_reg); 1281} 1282 1283template <class Impl> 1284TheISA::MiscReg 1285FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1286{ 1287 miscRegfileReads++; 1288 return this->isa[tid].readMiscReg(misc_reg, tcBase(tid)); 1289} 1290 1291template <class Impl> 1292void 1293FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1294 const TheISA::MiscReg &val, ThreadID tid) 1295{ 1296 this->isa[tid].setMiscRegNoEffect(misc_reg, val); 1297} 1298 1299template <class Impl> 1300void 1301FullO3CPU<Impl>::setMiscReg(int misc_reg, 1302 const TheISA::MiscReg &val, ThreadID tid) 1303{ 1304 miscRegfileWrites++; 1305 this->isa[tid].setMiscReg(misc_reg, val, tcBase(tid)); 1306} 1307 1308template <class Impl> 1309uint64_t 1310FullO3CPU<Impl>::readIntReg(int reg_idx) 1311{ 1312 intRegfileReads++; 1313 return regFile.readIntReg(reg_idx); 1314} 1315 1316template <class Impl> 1317FloatReg 1318FullO3CPU<Impl>::readFloatReg(int reg_idx) 1319{ 1320 fpRegfileReads++; 1321 return regFile.readFloatReg(reg_idx); 1322} 1323 1324template <class Impl> 1325FloatRegBits 1326FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1327{ 1328 fpRegfileReads++; 1329 return regFile.readFloatRegBits(reg_idx); 1330} 1331 1332template <class Impl> 1333void 1334FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1335{ 1336 intRegfileWrites++; 1337 regFile.setIntReg(reg_idx, val); 1338} 1339 1340template <class Impl> 1341void 1342FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1343{ 1344 fpRegfileWrites++; 1345 regFile.setFloatReg(reg_idx, val); 1346} 1347 1348template <class Impl> 1349void 1350FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1351{ 1352 fpRegfileWrites++; 1353 regFile.setFloatRegBits(reg_idx, val); 1354} 1355 1356template <class Impl> 1357uint64_t 1358FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1359{ 1360 intRegfileReads++; 1361 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1362 1363 return regFile.readIntReg(phys_reg); 1364} 1365 1366template <class Impl> 1367float 1368FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1369{ 1370 fpRegfileReads++; 1371 int idx = reg_idx + TheISA::NumIntRegs; 1372 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1373 1374 return regFile.readFloatReg(phys_reg); 1375} 1376 1377template <class Impl> 1378uint64_t 1379FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1380{ 1381 fpRegfileReads++; 1382 int idx = reg_idx + TheISA::NumIntRegs; 1383 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1384 1385 return regFile.readFloatRegBits(phys_reg); 1386} 1387 1388template <class Impl> 1389void 1390FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1391{ 1392 intRegfileWrites++; 1393 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1394 1395 regFile.setIntReg(phys_reg, val); 1396} 1397 1398template <class Impl> 1399void 1400FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1401{ 1402 fpRegfileWrites++; 1403 int idx = reg_idx + TheISA::NumIntRegs; 1404 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1405 1406 regFile.setFloatReg(phys_reg, val); 1407} 1408 1409template <class Impl> 1410void 1411FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1412{ 1413 fpRegfileWrites++; 1414 int idx = reg_idx + TheISA::NumIntRegs; 1415 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1416 1417 regFile.setFloatRegBits(phys_reg, val); 1418} 1419 1420template <class Impl> 1421TheISA::PCState 1422FullO3CPU<Impl>::pcState(ThreadID tid) 1423{ 1424 return commit.pcState(tid); 1425} 1426 1427template <class Impl> 1428void 1429FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1430{ 1431 commit.pcState(val, tid); 1432} 1433 1434template <class Impl> 1435Addr 1436FullO3CPU<Impl>::instAddr(ThreadID tid) 1437{ 1438 return commit.instAddr(tid); 1439} 1440 1441template <class Impl> 1442Addr 1443FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1444{ 1445 return commit.nextInstAddr(tid); 1446} 1447 1448template <class Impl> 1449MicroPC 1450FullO3CPU<Impl>::microPC(ThreadID tid) 1451{ 1452 return commit.microPC(tid); 1453} 1454 1455template <class Impl> 1456void 1457FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1458{ 1459 this->thread[tid]->inSyscall = true; 1460 this->commit.generateTCEvent(tid); 1461} 1462 1463template <class Impl> 1464typename FullO3CPU<Impl>::ListIt 1465FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1466{ 1467 instList.push_back(inst); 1468 1469 return --(instList.end()); 1470} 1471 1472template <class Impl> 1473void 1474FullO3CPU<Impl>::instDone(ThreadID tid) 1475{ 1476 // Keep an instruction count. 1477 thread[tid]->numInst++; 1478 thread[tid]->numInsts++; 1479 committedInsts[tid]++; 1480 totalCommittedInsts++; 1481 system->totalNumInsts++; 1482 // Check for instruction-count-based events. 1483 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1484 system->instEventQueue.serviceEvents(system->totalNumInsts); 1485} 1486 1487template <class Impl> 1488void 1489FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1490{ 1491 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1492 "[sn:%lli]\n", 1493 inst->threadNumber, inst->pcState(), inst->seqNum); 1494 1495 removeInstsThisCycle = true; 1496 1497 // Remove the front instruction. 1498 removeList.push(inst->getInstListIt()); 1499} 1500 1501template <class Impl> 1502void 1503FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1504{ 1505 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1506 " list.\n", tid); 1507 1508 ListIt end_it; 1509 1510 bool rob_empty = false; 1511 1512 if (instList.empty()) { 1513 return; 1514 } else if (rob.isEmpty(/*tid*/)) { 1515 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1516 end_it = instList.begin(); 1517 rob_empty = true; 1518 } else { 1519 end_it = (rob.readTailInst(tid))->getInstListIt(); 1520 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1521 } 1522 1523 removeInstsThisCycle = true; 1524 1525 ListIt inst_it = instList.end(); 1526 1527 inst_it--; 1528 1529 // Walk through the instruction list, removing any instructions 1530 // that were inserted after the given instruction iterator, end_it. 1531 while (inst_it != end_it) { 1532 assert(!instList.empty()); 1533 1534 squashInstIt(inst_it, tid); 1535 1536 inst_it--; 1537 } 1538 1539 // If the ROB was empty, then we actually need to remove the first 1540 // instruction as well. 1541 if (rob_empty) { 1542 squashInstIt(inst_it, tid); 1543 } 1544} 1545 1546template <class Impl> 1547void 1548FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1549{ 1550 assert(!instList.empty()); 1551 1552 removeInstsThisCycle = true; 1553 1554 ListIt inst_iter = instList.end(); 1555 1556 inst_iter--; 1557 1558 DPRINTF(O3CPU, "Deleting instructions from instruction " 1559 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1560 tid, seq_num, (*inst_iter)->seqNum); 1561 1562 while ((*inst_iter)->seqNum > seq_num) { 1563 1564 bool break_loop = (inst_iter == instList.begin()); 1565 1566 squashInstIt(inst_iter, tid); 1567 1568 inst_iter--; 1569 1570 if (break_loop) 1571 break; 1572 } 1573} 1574 1575template <class Impl> 1576inline void 1577FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1578{ 1579 if ((*instIt)->threadNumber == tid) { 1580 DPRINTF(O3CPU, "Squashing instruction, " 1581 "[tid:%i] [sn:%lli] PC %s\n", 1582 (*instIt)->threadNumber, 1583 (*instIt)->seqNum, 1584 (*instIt)->pcState()); 1585 1586 // Mark it as squashed. 1587 (*instIt)->setSquashed(); 1588 1589 // @todo: Formulate a consistent method for deleting 1590 // instructions from the instruction list 1591 // Remove the instruction from the list. 1592 removeList.push(instIt); 1593 } 1594} 1595 1596template <class Impl> 1597void 1598FullO3CPU<Impl>::cleanUpRemovedInsts() 1599{ 1600 while (!removeList.empty()) { 1601 DPRINTF(O3CPU, "Removing instruction, " 1602 "[tid:%i] [sn:%lli] PC %s\n", 1603 (*removeList.front())->threadNumber, 1604 (*removeList.front())->seqNum, 1605 (*removeList.front())->pcState()); 1606 1607 instList.erase(removeList.front()); 1608 1609 removeList.pop(); 1610 } 1611 1612 removeInstsThisCycle = false; 1613} 1614/* 1615template <class Impl> 1616void 1617FullO3CPU<Impl>::removeAllInsts() 1618{ 1619 instList.clear(); 1620} 1621*/ 1622template <class Impl> 1623void 1624FullO3CPU<Impl>::dumpInsts() 1625{ 1626 int num = 0; 1627 1628 ListIt inst_list_it = instList.begin(); 1629 1630 cprintf("Dumping Instruction List\n"); 1631 1632 while (inst_list_it != instList.end()) { 1633 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1634 "Squashed:%i\n\n", 1635 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1636 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1637 (*inst_list_it)->isSquashed()); 1638 inst_list_it++; 1639 ++num; 1640 } 1641} 1642/* 1643template <class Impl> 1644void 1645FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1646{ 1647 iew.wakeDependents(inst); 1648} 1649*/ 1650template <class Impl> 1651void 1652FullO3CPU<Impl>::wakeCPU() 1653{ 1654 if (activityRec.active() || tickEvent.scheduled()) { 1655 DPRINTF(Activity, "CPU already running.\n"); 1656 return; 1657 } 1658 1659 DPRINTF(Activity, "Waking up CPU\n"); 1660 1661 idleCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 1662 numCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 1663 1664 schedule(tickEvent, nextCycle()); 1665} 1666 1667#if FULL_SYSTEM 1668template <class Impl> 1669void 1670FullO3CPU<Impl>::wakeup() 1671{ 1672 if (this->thread[0]->status() != ThreadContext::Suspended) 1673 return; 1674 1675 this->wakeCPU(); 1676 1677 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1678 this->threadContexts[0]->activate(); 1679} 1680#endif 1681 1682template <class Impl> 1683ThreadID 1684FullO3CPU<Impl>::getFreeTid() 1685{ 1686 for (ThreadID tid = 0; tid < numThreads; tid++) { 1687 if (!tids[tid]) { 1688 tids[tid] = true; 1689 return tid; 1690 } 1691 } 1692 1693 return InvalidThreadID; 1694} 1695 1696template <class Impl> 1697void 1698FullO3CPU<Impl>::doContextSwitch() 1699{ 1700 if (contextSwitch) { 1701 1702 //ADD CODE TO DEACTIVE THREAD HERE (???) 1703 1704 ThreadID size = cpuWaitList.size(); 1705 for (ThreadID tid = 0; tid < size; tid++) { 1706 activateWhenReady(tid); 1707 } 1708 1709 if (cpuWaitList.size() == 0) 1710 contextSwitch = true; 1711 } 1712} 1713 1714template <class Impl> 1715void 1716FullO3CPU<Impl>::updateThreadPriority() 1717{ 1718 if (activeThreads.size() > 1) { 1719 //DEFAULT TO ROUND ROBIN SCHEME 1720 //e.g. Move highest priority to end of thread list 1721 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1722 1723 unsigned high_thread = *list_begin; 1724 1725 activeThreads.erase(list_begin); 1726 1727 activeThreads.push_back(high_thread); 1728 } 1729} 1730 1731// Forward declaration of FullO3CPU. 1732template class FullO3CPU<O3CPUImpl>; 1733