cpu.cc revision 8733:64a7bf8fa56c
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 79class 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>::deallocateContext(ThreadID tid, bool remove, int delay) 770{ 771 // Schedule removal of thread data from CPU 772 if (delay){ 773 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 774 "on cycle %d\n", tid, curTick() + ticks(delay)); 775 scheduleDeallocateContextEvent(tid, remove, delay); 776 return false; 777 } else { 778 deactivateThread(tid); 779 if (remove) 780 removeThread(tid); 781 return true; 782 } 783} 784 785template <class Impl> 786void 787FullO3CPU<Impl>::suspendContext(ThreadID tid) 788{ 789 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 790 bool deallocated = deallocateContext(tid, false, 1); 791 // If this was the last thread then unschedule the tick event. 792 if ((activeThreads.size() == 1 && !deallocated) || 793 activeThreads.size() == 0) 794 unscheduleTickEvent(); 795 796 DPRINTF(Quiesce, "Suspending Context\n"); 797 lastRunningCycle = curTick(); 798 _status = Idle; 799} 800 801template <class Impl> 802void 803FullO3CPU<Impl>::haltContext(ThreadID tid) 804{ 805 //For now, this is the same as deallocate 806 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 807 deallocateContext(tid, true, 1); 808} 809 810template <class Impl> 811void 812FullO3CPU<Impl>::insertThread(ThreadID tid) 813{ 814 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 815 // Will change now that the PC and thread state is internal to the CPU 816 // and not in the ThreadContext. 817#if FULL_SYSTEM 818 ThreadContext *src_tc = system->threadContexts[tid]; 819#else 820 ThreadContext *src_tc = tcBase(tid); 821#endif 822 823 //Bind Int Regs to Rename Map 824 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 825 PhysRegIndex phys_reg = freeList.getIntReg(); 826 827 renameMap[tid].setEntry(ireg,phys_reg); 828 scoreboard.setReg(phys_reg); 829 } 830 831 //Bind Float Regs to Rename Map 832 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 833 PhysRegIndex phys_reg = freeList.getFloatReg(); 834 835 renameMap[tid].setEntry(freg,phys_reg); 836 scoreboard.setReg(phys_reg); 837 } 838 839 //Copy Thread Data Into RegFile 840 //this->copyFromTC(tid); 841 842 //Set PC/NPC/NNPC 843 pcState(src_tc->pcState(), tid); 844 845 src_tc->setStatus(ThreadContext::Active); 846 847 activateContext(tid,1); 848 849 //Reset ROB/IQ/LSQ Entries 850 commit.rob->resetEntries(); 851 iew.resetEntries(); 852} 853 854template <class Impl> 855void 856FullO3CPU<Impl>::removeThread(ThreadID tid) 857{ 858 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 859 860 // Copy Thread Data From RegFile 861 // If thread is suspended, it might be re-allocated 862 // this->copyToTC(tid); 863 864 865 // @todo: 2-27-2008: Fix how we free up rename mappings 866 // here to alleviate the case for double-freeing registers 867 // in SMT workloads. 868 869 // Unbind Int Regs from Rename Map 870 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 871 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 872 873 scoreboard.unsetReg(phys_reg); 874 freeList.addReg(phys_reg); 875 } 876 877 // Unbind Float Regs from Rename Map 878 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) { 879 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 880 881 scoreboard.unsetReg(phys_reg); 882 freeList.addReg(phys_reg); 883 } 884 885 // Squash Throughout Pipeline 886 DynInstPtr inst = commit.rob->readHeadInst(tid); 887 InstSeqNum squash_seq_num = inst->seqNum; 888 fetch.squash(0, squash_seq_num, inst, tid); 889 decode.squash(tid); 890 rename.squash(squash_seq_num, tid); 891 iew.squash(tid); 892 iew.ldstQueue.squash(squash_seq_num, tid); 893 commit.rob->squash(squash_seq_num, tid); 894 895 896 assert(iew.instQueue.getCount(tid) == 0); 897 assert(iew.ldstQueue.getCount(tid) == 0); 898 899 // Reset ROB/IQ/LSQ Entries 900 901 // Commented out for now. This should be possible to do by 902 // telling all the pipeline stages to drain first, and then 903 // checking until the drain completes. Once the pipeline is 904 // drained, call resetEntries(). - 10-09-06 ktlim 905/* 906 if (activeThreads.size() >= 1) { 907 commit.rob->resetEntries(); 908 iew.resetEntries(); 909 } 910*/ 911} 912 913 914template <class Impl> 915void 916FullO3CPU<Impl>::activateWhenReady(ThreadID tid) 917{ 918 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 919 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 920 tid); 921 922 bool ready = true; 923 924 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 925 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 926 "Phys. Int. Regs.\n", 927 tid); 928 ready = false; 929 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 930 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 931 "Phys. Float. Regs.\n", 932 tid); 933 ready = false; 934 } else if (commit.rob->numFreeEntries() >= 935 commit.rob->entryAmount(activeThreads.size() + 1)) { 936 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 937 "ROB entries.\n", 938 tid); 939 ready = false; 940 } else if (iew.instQueue.numFreeEntries() >= 941 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 942 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 943 "IQ entries.\n", 944 tid); 945 ready = false; 946 } else if (iew.ldstQueue.numFreeEntries() >= 947 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 948 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 949 "LSQ entries.\n", 950 tid); 951 ready = false; 952 } 953 954 if (ready) { 955 insertThread(tid); 956 957 contextSwitch = false; 958 959 cpuWaitList.remove(tid); 960 } else { 961 suspendContext(tid); 962 963 //blocks fetch 964 contextSwitch = true; 965 966 //@todo: dont always add to waitlist 967 //do waitlist 968 cpuWaitList.push_back(tid); 969 } 970} 971 972#if FULL_SYSTEM 973template <class Impl> 974Fault 975FullO3CPU<Impl>::hwrei(ThreadID tid) 976{ 977#if THE_ISA == ALPHA_ISA 978 // Need to clear the lock flag upon returning from an interrupt. 979 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 980 981 this->thread[tid]->kernelStats->hwrei(); 982 983 // FIXME: XXX check for interrupts? XXX 984#endif 985 return NoFault; 986} 987 988template <class Impl> 989bool 990FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 991{ 992#if THE_ISA == ALPHA_ISA 993 if (this->thread[tid]->kernelStats) 994 this->thread[tid]->kernelStats->callpal(palFunc, 995 this->threadContexts[tid]); 996 997 switch (palFunc) { 998 case PAL::halt: 999 halt(); 1000 if (--System::numSystemsRunning == 0) 1001 exitSimLoop("all cpus halted"); 1002 break; 1003 1004 case PAL::bpt: 1005 case PAL::bugchk: 1006 if (this->system->breakpoint()) 1007 return false; 1008 break; 1009 } 1010#endif 1011 return true; 1012} 1013 1014template <class Impl> 1015Fault 1016FullO3CPU<Impl>::getInterrupts() 1017{ 1018 // Check if there are any outstanding interrupts 1019 return this->interrupts->getInterrupt(this->threadContexts[0]); 1020} 1021 1022template <class Impl> 1023void 1024FullO3CPU<Impl>::processInterrupts(Fault interrupt) 1025{ 1026 // Check for interrupts here. For now can copy the code that 1027 // exists within isa_fullsys_traits.hh. Also assume that thread 0 1028 // is the one that handles the interrupts. 1029 // @todo: Possibly consolidate the interrupt checking code. 1030 // @todo: Allow other threads to handle interrupts. 1031 1032 assert(interrupt != NoFault); 1033 this->interrupts->updateIntrInfo(this->threadContexts[0]); 1034 1035 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1036 this->trap(interrupt, 0, NULL); 1037} 1038 1039#endif 1040 1041template <class Impl> 1042void 1043FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst) 1044{ 1045 // Pass the thread's TC into the invoke method. 1046 fault->invoke(this->threadContexts[tid], inst); 1047} 1048 1049#if !FULL_SYSTEM 1050 1051template <class Impl> 1052void 1053FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid) 1054{ 1055 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1056 1057 DPRINTF(Activity,"Activity: syscall() called.\n"); 1058 1059 // Temporarily increase this by one to account for the syscall 1060 // instruction. 1061 ++(this->thread[tid]->funcExeInst); 1062 1063 // Execute the actual syscall. 1064 this->thread[tid]->syscall(callnum); 1065 1066 // Decrease funcExeInst by one as the normal commit will handle 1067 // incrementing it. 1068 --(this->thread[tid]->funcExeInst); 1069} 1070 1071#endif 1072 1073template <class Impl> 1074void 1075FullO3CPU<Impl>::serialize(std::ostream &os) 1076{ 1077 SimObject::State so_state = SimObject::getState(); 1078 SERIALIZE_ENUM(so_state); 1079 BaseCPU::serialize(os); 1080 nameOut(os, csprintf("%s.tickEvent", name())); 1081 tickEvent.serialize(os); 1082 1083 // Use SimpleThread's ability to checkpoint to make it easier to 1084 // write out the registers. Also make this static so it doesn't 1085 // get instantiated multiple times (causes a panic in statistics). 1086 static SimpleThread temp; 1087 1088 ThreadID size = thread.size(); 1089 for (ThreadID i = 0; i < size; i++) { 1090 nameOut(os, csprintf("%s.xc.%i", name(), i)); 1091 temp.copyTC(thread[i]->getTC()); 1092 temp.serialize(os); 1093 } 1094} 1095 1096template <class Impl> 1097void 1098FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 1099{ 1100 SimObject::State so_state; 1101 UNSERIALIZE_ENUM(so_state); 1102 BaseCPU::unserialize(cp, section); 1103 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 1104 1105 // Use SimpleThread's ability to checkpoint to make it easier to 1106 // read in the registers. Also make this static so it doesn't 1107 // get instantiated multiple times (causes a panic in statistics). 1108 static SimpleThread temp; 1109 1110 ThreadID size = thread.size(); 1111 for (ThreadID i = 0; i < size; i++) { 1112 temp.copyTC(thread[i]->getTC()); 1113 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 1114 thread[i]->getTC()->copyArchRegs(temp.getTC()); 1115 } 1116} 1117 1118template <class Impl> 1119unsigned int 1120FullO3CPU<Impl>::drain(Event *drain_event) 1121{ 1122 DPRINTF(O3CPU, "Switching out\n"); 1123 1124 // If the CPU isn't doing anything, then return immediately. 1125 if (_status == Idle || _status == SwitchedOut) { 1126 return 0; 1127 } 1128 1129 drainCount = 0; 1130 fetch.drain(); 1131 decode.drain(); 1132 rename.drain(); 1133 iew.drain(); 1134 commit.drain(); 1135 1136 // Wake the CPU and record activity so everything can drain out if 1137 // the CPU was not able to immediately drain. 1138 if (getState() != SimObject::Drained) { 1139 // A bit of a hack...set the drainEvent after all the drain() 1140 // calls have been made, that way if all of the stages drain 1141 // immediately, the signalDrained() function knows not to call 1142 // process on the drain event. 1143 drainEvent = drain_event; 1144 1145 wakeCPU(); 1146 activityRec.activity(); 1147 1148 return 1; 1149 } else { 1150 return 0; 1151 } 1152} 1153 1154template <class Impl> 1155void 1156FullO3CPU<Impl>::resume() 1157{ 1158 fetch.resume(); 1159 decode.resume(); 1160 rename.resume(); 1161 iew.resume(); 1162 commit.resume(); 1163 1164 changeState(SimObject::Running); 1165 1166 if (_status == SwitchedOut || _status == Idle) 1167 return; 1168 1169 assert(system->getMemoryMode() == Enums::timing); 1170 1171 if (!tickEvent.scheduled()) 1172 schedule(tickEvent, nextCycle()); 1173 _status = Running; 1174} 1175 1176template <class Impl> 1177void 1178FullO3CPU<Impl>::signalDrained() 1179{ 1180 if (++drainCount == NumStages) { 1181 if (tickEvent.scheduled()) 1182 tickEvent.squash(); 1183 1184 changeState(SimObject::Drained); 1185 1186 BaseCPU::switchOut(); 1187 1188 if (drainEvent) { 1189 drainEvent->process(); 1190 drainEvent = NULL; 1191 } 1192 } 1193 assert(drainCount <= 5); 1194} 1195 1196template <class Impl> 1197void 1198FullO3CPU<Impl>::switchOut() 1199{ 1200 fetch.switchOut(); 1201 rename.switchOut(); 1202 iew.switchOut(); 1203 commit.switchOut(); 1204 instList.clear(); 1205 while (!removeList.empty()) { 1206 removeList.pop(); 1207 } 1208 1209 _status = SwitchedOut; 1210#if USE_CHECKER 1211 if (checker) 1212 checker->switchOut(); 1213#endif 1214 if (tickEvent.scheduled()) 1215 tickEvent.squash(); 1216} 1217 1218template <class Impl> 1219void 1220FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1221{ 1222 // Flush out any old data from the time buffers. 1223 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1224 timeBuffer.advance(); 1225 fetchQueue.advance(); 1226 decodeQueue.advance(); 1227 renameQueue.advance(); 1228 iewQueue.advance(); 1229 } 1230 1231 activityRec.reset(); 1232 1233 BaseCPU::takeOverFrom(oldCPU, &icachePort, &dcachePort); 1234 1235 fetch.takeOverFrom(); 1236 decode.takeOverFrom(); 1237 rename.takeOverFrom(); 1238 iew.takeOverFrom(); 1239 commit.takeOverFrom(); 1240 1241 assert(!tickEvent.scheduled() || tickEvent.squashed()); 1242 1243 // @todo: Figure out how to properly select the tid to put onto 1244 // the active threads list. 1245 ThreadID tid = 0; 1246 1247 list<ThreadID>::iterator isActive = 1248 std::find(activeThreads.begin(), activeThreads.end(), tid); 1249 1250 if (isActive == activeThreads.end()) { 1251 //May Need to Re-code this if the delay variable is the delay 1252 //needed for thread to activate 1253 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 1254 tid); 1255 1256 activeThreads.push_back(tid); 1257 } 1258 1259 // Set all statuses to active, schedule the CPU's tick event. 1260 // @todo: Fix up statuses so this is handled properly 1261 ThreadID size = threadContexts.size(); 1262 for (ThreadID i = 0; i < size; ++i) { 1263 ThreadContext *tc = threadContexts[i]; 1264 if (tc->status() == ThreadContext::Active && _status != Running) { 1265 _status = Running; 1266 reschedule(tickEvent, nextCycle(), true); 1267 } 1268 } 1269 if (!tickEvent.scheduled()) 1270 schedule(tickEvent, nextCycle()); 1271 1272 lastRunningCycle = curTick(); 1273} 1274 1275template <class Impl> 1276TheISA::MiscReg 1277FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) 1278{ 1279 return this->isa[tid].readMiscRegNoEffect(misc_reg); 1280} 1281 1282template <class Impl> 1283TheISA::MiscReg 1284FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1285{ 1286 miscRegfileReads++; 1287 return this->isa[tid].readMiscReg(misc_reg, tcBase(tid)); 1288} 1289 1290template <class Impl> 1291void 1292FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1293 const TheISA::MiscReg &val, ThreadID tid) 1294{ 1295 this->isa[tid].setMiscRegNoEffect(misc_reg, val); 1296} 1297 1298template <class Impl> 1299void 1300FullO3CPU<Impl>::setMiscReg(int misc_reg, 1301 const TheISA::MiscReg &val, ThreadID tid) 1302{ 1303 miscRegfileWrites++; 1304 this->isa[tid].setMiscReg(misc_reg, val, tcBase(tid)); 1305} 1306 1307template <class Impl> 1308uint64_t 1309FullO3CPU<Impl>::readIntReg(int reg_idx) 1310{ 1311 intRegfileReads++; 1312 return regFile.readIntReg(reg_idx); 1313} 1314 1315template <class Impl> 1316FloatReg 1317FullO3CPU<Impl>::readFloatReg(int reg_idx) 1318{ 1319 fpRegfileReads++; 1320 return regFile.readFloatReg(reg_idx); 1321} 1322 1323template <class Impl> 1324FloatRegBits 1325FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1326{ 1327 fpRegfileReads++; 1328 return regFile.readFloatRegBits(reg_idx); 1329} 1330 1331template <class Impl> 1332void 1333FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1334{ 1335 intRegfileWrites++; 1336 regFile.setIntReg(reg_idx, val); 1337} 1338 1339template <class Impl> 1340void 1341FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1342{ 1343 fpRegfileWrites++; 1344 regFile.setFloatReg(reg_idx, val); 1345} 1346 1347template <class Impl> 1348void 1349FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1350{ 1351 fpRegfileWrites++; 1352 regFile.setFloatRegBits(reg_idx, val); 1353} 1354 1355template <class Impl> 1356uint64_t 1357FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1358{ 1359 intRegfileReads++; 1360 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1361 1362 return regFile.readIntReg(phys_reg); 1363} 1364 1365template <class Impl> 1366float 1367FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1368{ 1369 fpRegfileReads++; 1370 int idx = reg_idx + TheISA::NumIntRegs; 1371 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1372 1373 return regFile.readFloatReg(phys_reg); 1374} 1375 1376template <class Impl> 1377uint64_t 1378FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1379{ 1380 fpRegfileReads++; 1381 int idx = reg_idx + TheISA::NumIntRegs; 1382 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1383 1384 return regFile.readFloatRegBits(phys_reg); 1385} 1386 1387template <class Impl> 1388void 1389FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1390{ 1391 intRegfileWrites++; 1392 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1393 1394 regFile.setIntReg(phys_reg, val); 1395} 1396 1397template <class Impl> 1398void 1399FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1400{ 1401 fpRegfileWrites++; 1402 int idx = reg_idx + TheISA::NumIntRegs; 1403 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1404 1405 regFile.setFloatReg(phys_reg, val); 1406} 1407 1408template <class Impl> 1409void 1410FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1411{ 1412 fpRegfileWrites++; 1413 int idx = reg_idx + TheISA::NumIntRegs; 1414 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1415 1416 regFile.setFloatRegBits(phys_reg, val); 1417} 1418 1419template <class Impl> 1420TheISA::PCState 1421FullO3CPU<Impl>::pcState(ThreadID tid) 1422{ 1423 return commit.pcState(tid); 1424} 1425 1426template <class Impl> 1427void 1428FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1429{ 1430 commit.pcState(val, tid); 1431} 1432 1433template <class Impl> 1434Addr 1435FullO3CPU<Impl>::instAddr(ThreadID tid) 1436{ 1437 return commit.instAddr(tid); 1438} 1439 1440template <class Impl> 1441Addr 1442FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1443{ 1444 return commit.nextInstAddr(tid); 1445} 1446 1447template <class Impl> 1448MicroPC 1449FullO3CPU<Impl>::microPC(ThreadID tid) 1450{ 1451 return commit.microPC(tid); 1452} 1453 1454template <class Impl> 1455void 1456FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1457{ 1458 this->thread[tid]->inSyscall = true; 1459 this->commit.generateTCEvent(tid); 1460} 1461 1462template <class Impl> 1463typename FullO3CPU<Impl>::ListIt 1464FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1465{ 1466 instList.push_back(inst); 1467 1468 return --(instList.end()); 1469} 1470 1471template <class Impl> 1472void 1473FullO3CPU<Impl>::instDone(ThreadID tid) 1474{ 1475 // Keep an instruction count. 1476 thread[tid]->numInst++; 1477 thread[tid]->numInsts++; 1478 committedInsts[tid]++; 1479 totalCommittedInsts++; 1480 system->totalNumInsts++; 1481 // Check for instruction-count-based events. 1482 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1483 system->instEventQueue.serviceEvents(system->totalNumInsts); 1484} 1485 1486template <class Impl> 1487void 1488FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1489{ 1490 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1491 "[sn:%lli]\n", 1492 inst->threadNumber, inst->pcState(), inst->seqNum); 1493 1494 removeInstsThisCycle = true; 1495 1496 // Remove the front instruction. 1497 removeList.push(inst->getInstListIt()); 1498} 1499 1500template <class Impl> 1501void 1502FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1503{ 1504 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1505 " list.\n", tid); 1506 1507 ListIt end_it; 1508 1509 bool rob_empty = false; 1510 1511 if (instList.empty()) { 1512 return; 1513 } else if (rob.isEmpty(/*tid*/)) { 1514 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1515 end_it = instList.begin(); 1516 rob_empty = true; 1517 } else { 1518 end_it = (rob.readTailInst(tid))->getInstListIt(); 1519 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1520 } 1521 1522 removeInstsThisCycle = true; 1523 1524 ListIt inst_it = instList.end(); 1525 1526 inst_it--; 1527 1528 // Walk through the instruction list, removing any instructions 1529 // that were inserted after the given instruction iterator, end_it. 1530 while (inst_it != end_it) { 1531 assert(!instList.empty()); 1532 1533 squashInstIt(inst_it, tid); 1534 1535 inst_it--; 1536 } 1537 1538 // If the ROB was empty, then we actually need to remove the first 1539 // instruction as well. 1540 if (rob_empty) { 1541 squashInstIt(inst_it, tid); 1542 } 1543} 1544 1545template <class Impl> 1546void 1547FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1548{ 1549 assert(!instList.empty()); 1550 1551 removeInstsThisCycle = true; 1552 1553 ListIt inst_iter = instList.end(); 1554 1555 inst_iter--; 1556 1557 DPRINTF(O3CPU, "Deleting instructions from instruction " 1558 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1559 tid, seq_num, (*inst_iter)->seqNum); 1560 1561 while ((*inst_iter)->seqNum > seq_num) { 1562 1563 bool break_loop = (inst_iter == instList.begin()); 1564 1565 squashInstIt(inst_iter, tid); 1566 1567 inst_iter--; 1568 1569 if (break_loop) 1570 break; 1571 } 1572} 1573 1574template <class Impl> 1575inline void 1576FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1577{ 1578 if ((*instIt)->threadNumber == tid) { 1579 DPRINTF(O3CPU, "Squashing instruction, " 1580 "[tid:%i] [sn:%lli] PC %s\n", 1581 (*instIt)->threadNumber, 1582 (*instIt)->seqNum, 1583 (*instIt)->pcState()); 1584 1585 // Mark it as squashed. 1586 (*instIt)->setSquashed(); 1587 1588 // @todo: Formulate a consistent method for deleting 1589 // instructions from the instruction list 1590 // Remove the instruction from the list. 1591 removeList.push(instIt); 1592 } 1593} 1594 1595template <class Impl> 1596void 1597FullO3CPU<Impl>::cleanUpRemovedInsts() 1598{ 1599 while (!removeList.empty()) { 1600 DPRINTF(O3CPU, "Removing instruction, " 1601 "[tid:%i] [sn:%lli] PC %s\n", 1602 (*removeList.front())->threadNumber, 1603 (*removeList.front())->seqNum, 1604 (*removeList.front())->pcState()); 1605 1606 instList.erase(removeList.front()); 1607 1608 removeList.pop(); 1609 } 1610 1611 removeInstsThisCycle = false; 1612} 1613/* 1614template <class Impl> 1615void 1616FullO3CPU<Impl>::removeAllInsts() 1617{ 1618 instList.clear(); 1619} 1620*/ 1621template <class Impl> 1622void 1623FullO3CPU<Impl>::dumpInsts() 1624{ 1625 int num = 0; 1626 1627 ListIt inst_list_it = instList.begin(); 1628 1629 cprintf("Dumping Instruction List\n"); 1630 1631 while (inst_list_it != instList.end()) { 1632 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1633 "Squashed:%i\n\n", 1634 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1635 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1636 (*inst_list_it)->isSquashed()); 1637 inst_list_it++; 1638 ++num; 1639 } 1640} 1641/* 1642template <class Impl> 1643void 1644FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1645{ 1646 iew.wakeDependents(inst); 1647} 1648*/ 1649template <class Impl> 1650void 1651FullO3CPU<Impl>::wakeCPU() 1652{ 1653 if (activityRec.active() || tickEvent.scheduled()) { 1654 DPRINTF(Activity, "CPU already running.\n"); 1655 return; 1656 } 1657 1658 DPRINTF(Activity, "Waking up CPU\n"); 1659 1660 idleCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 1661 numCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 1662 1663 schedule(tickEvent, nextCycle()); 1664} 1665 1666#if FULL_SYSTEM 1667template <class Impl> 1668void 1669FullO3CPU<Impl>::wakeup() 1670{ 1671 if (this->thread[0]->status() != ThreadContext::Suspended) 1672 return; 1673 1674 this->wakeCPU(); 1675 1676 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1677 this->threadContexts[0]->activate(); 1678} 1679#endif 1680 1681template <class Impl> 1682ThreadID 1683FullO3CPU<Impl>::getFreeTid() 1684{ 1685 for (ThreadID tid = 0; tid < numThreads; tid++) { 1686 if (!tids[tid]) { 1687 tids[tid] = true; 1688 return tid; 1689 } 1690 } 1691 1692 return InvalidThreadID; 1693} 1694 1695template <class Impl> 1696void 1697FullO3CPU<Impl>::doContextSwitch() 1698{ 1699 if (contextSwitch) { 1700 1701 //ADD CODE TO DEACTIVE THREAD HERE (???) 1702 1703 ThreadID size = cpuWaitList.size(); 1704 for (ThreadID tid = 0; tid < size; tid++) { 1705 activateWhenReady(tid); 1706 } 1707 1708 if (cpuWaitList.size() == 0) 1709 contextSwitch = true; 1710 } 1711} 1712 1713template <class Impl> 1714void 1715FullO3CPU<Impl>::updateThreadPriority() 1716{ 1717 if (activeThreads.size() > 1) { 1718 //DEFAULT TO ROUND ROBIN SCHEME 1719 //e.g. Move highest priority to end of thread list 1720 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1721 1722 unsigned high_thread = *list_begin; 1723 1724 activeThreads.erase(list_begin); 1725 1726 activeThreads.push_back(high_thread); 1727 } 1728} 1729 1730// Forward declaration of FullO3CPU. 1731template class FullO3CPU<O3CPUImpl>; 1732