cpu.cc revision 8834
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 "arch/kernel_stats.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/quiesce_event.hh" 54#include "cpu/simple_thread.hh" 55#include "cpu/thread_context.hh" 56#include "debug/Activity.hh" 57#include "debug/O3CPU.hh" 58#include "debug/Quiesce.hh" 59#include "enums/MemoryMode.hh" 60#include "sim/core.hh" 61#include "sim/full_system.hh" 62#include "sim/process.hh" 63#include "sim/stat_control.hh" 64#include "sim/system.hh" 65 66#if USE_CHECKER 67#include "cpu/checker/cpu.hh" 68#include "cpu/checker/thread_context.hh" 69#endif 70 71#if THE_ISA == ALPHA_ISA 72#include "arch/alpha/osfpal.hh" 73#include "debug/Activity.hh" 74#endif 75 76struct BaseCPUParams; 77 78using namespace TheISA; 79using namespace std; 80 81BaseO3CPU::BaseO3CPU(BaseCPUParams *params) 82 : BaseCPU(params) 83{ 84} 85 86void 87BaseO3CPU::regStats() 88{ 89 BaseCPU::regStats(); 90} 91 92template<class Impl> 93bool 94FullO3CPU<Impl>::IcachePort::recvTiming(PacketPtr pkt) 95{ 96 DPRINTF(O3CPU, "Fetch unit received timing\n"); 97 if (pkt->isResponse()) { 98 // We shouldn't ever get a block in ownership state 99 assert(!(pkt->memInhibitAsserted() && !pkt->sharedAsserted())); 100 101 fetch->processCacheCompletion(pkt); 102 } 103 //else Snooped a coherence request, just return 104 return true; 105} 106 107template<class Impl> 108void 109FullO3CPU<Impl>::IcachePort::recvRetry() 110{ 111 fetch->recvRetry(); 112} 113 114template <class Impl> 115bool 116FullO3CPU<Impl>::DcachePort::recvTiming(PacketPtr pkt) 117{ 118 return lsq->recvTiming(pkt); 119} 120 121template <class Impl> 122void 123FullO3CPU<Impl>::DcachePort::recvRetry() 124{ 125 lsq->recvRetry(); 126} 127 128template <class Impl> 129FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 130 : Event(CPU_Tick_Pri), cpu(c) 131{ 132} 133 134template <class Impl> 135void 136FullO3CPU<Impl>::TickEvent::process() 137{ 138 cpu->tick(); 139} 140 141template <class Impl> 142const char * 143FullO3CPU<Impl>::TickEvent::description() const 144{ 145 return "FullO3CPU tick"; 146} 147 148template <class Impl> 149FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent() 150 : Event(CPU_Switch_Pri) 151{ 152} 153 154template <class Impl> 155void 156FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num, 157 FullO3CPU<Impl> *thread_cpu) 158{ 159 tid = thread_num; 160 cpu = thread_cpu; 161} 162 163template <class Impl> 164void 165FullO3CPU<Impl>::ActivateThreadEvent::process() 166{ 167 cpu->activateThread(tid); 168} 169 170template <class Impl> 171const char * 172FullO3CPU<Impl>::ActivateThreadEvent::description() const 173{ 174 return "FullO3CPU \"Activate Thread\""; 175} 176 177template <class Impl> 178FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent() 179 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL) 180{ 181} 182 183template <class Impl> 184void 185FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num, 186 FullO3CPU<Impl> *thread_cpu) 187{ 188 tid = thread_num; 189 cpu = thread_cpu; 190 remove = false; 191} 192 193template <class Impl> 194void 195FullO3CPU<Impl>::DeallocateContextEvent::process() 196{ 197 cpu->deactivateThread(tid); 198 if (remove) 199 cpu->removeThread(tid); 200} 201 202template <class Impl> 203const char * 204FullO3CPU<Impl>::DeallocateContextEvent::description() const 205{ 206 return "FullO3CPU \"Deallocate Context\""; 207} 208 209template <class Impl> 210FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params) 211 : BaseO3CPU(params), 212 itb(params->itb), 213 dtb(params->dtb), 214 tickEvent(this), 215#ifndef NDEBUG 216 instcount(0), 217#endif 218 removeInstsThisCycle(false), 219 fetch(this, params), 220 decode(this, params), 221 rename(this, params), 222 iew(this, params), 223 commit(this, params), 224 225 regFile(this, params->numPhysIntRegs, 226 params->numPhysFloatRegs), 227 228 freeList(params->numThreads, 229 TheISA::NumIntRegs, params->numPhysIntRegs, 230 TheISA::NumFloatRegs, params->numPhysFloatRegs), 231 232 rob(this, 233 params->numROBEntries, params->squashWidth, 234 params->smtROBPolicy, params->smtROBThreshold, 235 params->numThreads), 236 237 scoreboard(params->numThreads, 238 TheISA::NumIntRegs, params->numPhysIntRegs, 239 TheISA::NumFloatRegs, params->numPhysFloatRegs, 240 TheISA::NumMiscRegs * numThreads, 241 TheISA::ZeroReg), 242 243 icachePort(&fetch, this), 244 dcachePort(&iew.ldstQueue, this), 245 246 timeBuffer(params->backComSize, params->forwardComSize), 247 fetchQueue(params->backComSize, params->forwardComSize), 248 decodeQueue(params->backComSize, params->forwardComSize), 249 renameQueue(params->backComSize, params->forwardComSize), 250 iewQueue(params->backComSize, params->forwardComSize), 251 activityRec(name(), NumStages, 252 params->backComSize + params->forwardComSize, 253 params->activity), 254 255 globalSeqNum(1), 256 system(params->system), 257 drainCount(0), 258 deferRegistration(params->defer_registration) 259{ 260 if (!deferRegistration) { 261 _status = Running; 262 } else { 263 _status = Idle; 264 } 265 266#if USE_CHECKER 267 if (params->checker) { 268 BaseCPU *temp_checker = params->checker; 269 checker = dynamic_cast<Checker<Impl> *>(temp_checker); 270 checker->setIcachePort(&icachePort); 271 checker->setSystem(params->system); 272 } else { 273 checker = NULL; 274 } 275#endif // USE_CHECKER 276 277 if (!FullSystem) { 278 thread.resize(numThreads); 279 tids.resize(numThreads); 280 } 281 282 // The stages also need their CPU pointer setup. However this 283 // must be done at the upper level CPU because they have pointers 284 // to the upper level CPU, and not this FullO3CPU. 285 286 // Set up Pointers to the activeThreads list for each stage 287 fetch.setActiveThreads(&activeThreads); 288 decode.setActiveThreads(&activeThreads); 289 rename.setActiveThreads(&activeThreads); 290 iew.setActiveThreads(&activeThreads); 291 commit.setActiveThreads(&activeThreads); 292 293 // Give each of the stages the time buffer they will use. 294 fetch.setTimeBuffer(&timeBuffer); 295 decode.setTimeBuffer(&timeBuffer); 296 rename.setTimeBuffer(&timeBuffer); 297 iew.setTimeBuffer(&timeBuffer); 298 commit.setTimeBuffer(&timeBuffer); 299 300 // Also setup each of the stages' queues. 301 fetch.setFetchQueue(&fetchQueue); 302 decode.setFetchQueue(&fetchQueue); 303 commit.setFetchQueue(&fetchQueue); 304 decode.setDecodeQueue(&decodeQueue); 305 rename.setDecodeQueue(&decodeQueue); 306 rename.setRenameQueue(&renameQueue); 307 iew.setRenameQueue(&renameQueue); 308 iew.setIEWQueue(&iewQueue); 309 commit.setIEWQueue(&iewQueue); 310 commit.setRenameQueue(&renameQueue); 311 312 commit.setIEWStage(&iew); 313 rename.setIEWStage(&iew); 314 rename.setCommitStage(&commit); 315 316 ThreadID active_threads; 317 if (FullSystem) { 318 active_threads = 1; 319 } else { 320 active_threads = params->workload.size(); 321 322 if (active_threads > Impl::MaxThreads) { 323 panic("Workload Size too large. Increase the 'MaxThreads' " 324 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) " 325 "or edit your workload size."); 326 } 327 } 328 329 //Make Sure That this a Valid Architeture 330 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 331 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 332 333 rename.setScoreboard(&scoreboard); 334 iew.setScoreboard(&scoreboard); 335 336 // Setup the rename map for whichever stages need it. 337 PhysRegIndex lreg_idx = 0; 338 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 339 340 for (ThreadID tid = 0; tid < numThreads; tid++) { 341 bool bindRegs = (tid <= active_threads - 1); 342 343 commitRenameMap[tid].init(TheISA::NumIntRegs, 344 params->numPhysIntRegs, 345 lreg_idx, //Index for Logical. Regs 346 347 TheISA::NumFloatRegs, 348 params->numPhysFloatRegs, 349 freg_idx, //Index for Float Regs 350 351 TheISA::NumMiscRegs, 352 353 TheISA::ZeroReg, 354 TheISA::ZeroReg, 355 356 tid, 357 false); 358 359 renameMap[tid].init(TheISA::NumIntRegs, 360 params->numPhysIntRegs, 361 lreg_idx, //Index for Logical. Regs 362 363 TheISA::NumFloatRegs, 364 params->numPhysFloatRegs, 365 freg_idx, //Index for Float Regs 366 367 TheISA::NumMiscRegs, 368 369 TheISA::ZeroReg, 370 TheISA::ZeroReg, 371 372 tid, 373 bindRegs); 374 375 activateThreadEvent[tid].init(tid, this); 376 deallocateContextEvent[tid].init(tid, this); 377 } 378 379 rename.setRenameMap(renameMap); 380 commit.setRenameMap(commitRenameMap); 381 382 // Give renameMap & rename stage access to the freeList; 383 for (ThreadID tid = 0; tid < numThreads; tid++) 384 renameMap[tid].setFreeList(&freeList); 385 rename.setFreeList(&freeList); 386 387 // Setup the ROB for whichever stages need it. 388 commit.setROB(&rob); 389 390 lastRunningCycle = curTick(); 391 392 lastActivatedCycle = -1; 393#if 0 394 // Give renameMap & rename stage access to the freeList; 395 for (ThreadID tid = 0; tid < numThreads; tid++) 396 globalSeqNum[tid] = 1; 397#endif 398 399 contextSwitch = false; 400 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 401 402 // Setup any thread state. 403 this->thread.resize(this->numThreads); 404 405 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 406 if (FullSystem) { 407 // SMT is not supported in FS mode yet. 408 assert(this->numThreads == 1); 409 this->thread[tid] = new Thread(this, 0, NULL); 410 } else { 411 if (tid < params->workload.size()) { 412 DPRINTF(O3CPU, "Workload[%i] process is %#x", 413 tid, this->thread[tid]); 414 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 415 (typename Impl::O3CPU *)(this), 416 tid, params->workload[tid]); 417 418 //usedTids[tid] = true; 419 //threadMap[tid] = tid; 420 } else { 421 //Allocate Empty thread so M5 can use later 422 //when scheduling threads to CPU 423 Process* dummy_proc = NULL; 424 425 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 426 (typename Impl::O3CPU *)(this), 427 tid, dummy_proc); 428 //usedTids[tid] = false; 429 } 430 } 431 432 ThreadContext *tc; 433 434 // Setup the TC that will serve as the interface to the threads/CPU. 435 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 436 437 tc = o3_tc; 438 439 // If we're using a checker, then the TC should be the 440 // CheckerThreadContext. 441#if USE_CHECKER 442 if (params->checker) { 443 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 444 o3_tc, this->checker); 445 } 446#endif 447 448 o3_tc->cpu = (typename Impl::O3CPU *)(this); 449 assert(o3_tc->cpu); 450 o3_tc->thread = this->thread[tid]; 451 452 if (FullSystem) { 453 // Setup quiesce event. 454 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 455 } 456 // Give the thread the TC. 457 this->thread[tid]->tc = tc; 458 459 // Add the TC to the CPU's list of TC's. 460 this->threadContexts.push_back(tc); 461 } 462 463 for (ThreadID tid = 0; tid < this->numThreads; tid++) 464 this->thread[tid]->setFuncExeInst(0); 465 466 lockAddr = 0; 467 lockFlag = false; 468} 469 470template <class Impl> 471FullO3CPU<Impl>::~FullO3CPU() 472{ 473} 474 475template <class Impl> 476void 477FullO3CPU<Impl>::regStats() 478{ 479 BaseO3CPU::regStats(); 480 481 // Register any of the O3CPU's stats here. 482 timesIdled 483 .name(name() + ".timesIdled") 484 .desc("Number of times that the entire CPU went into an idle state and" 485 " unscheduled itself") 486 .prereq(timesIdled); 487 488 idleCycles 489 .name(name() + ".idleCycles") 490 .desc("Total number of cycles that the CPU has spent unscheduled due " 491 "to idling") 492 .prereq(idleCycles); 493 494 quiesceCycles 495 .name(name() + ".quiesceCycles") 496 .desc("Total number of cycles that CPU has spent quiesced or waiting " 497 "for an interrupt") 498 .prereq(quiesceCycles); 499 500 // Number of Instructions simulated 501 // -------------------------------- 502 // Should probably be in Base CPU but need templated 503 // MaxThreads so put in here instead 504 committedInsts 505 .init(numThreads) 506 .name(name() + ".committedInsts") 507 .desc("Number of Instructions Simulated"); 508 509 committedOps 510 .init(numThreads) 511 .name(name() + ".committedOps") 512 .desc("Number of Ops (including micro ops) Simulated"); 513 514 totalCommittedInsts 515 .name(name() + ".committedInsts_total") 516 .desc("Number of Instructions Simulated"); 517 518 cpi 519 .name(name() + ".cpi") 520 .desc("CPI: Cycles Per Instruction") 521 .precision(6); 522 cpi = numCycles / committedInsts; 523 524 totalCpi 525 .name(name() + ".cpi_total") 526 .desc("CPI: Total CPI of All Threads") 527 .precision(6); 528 totalCpi = numCycles / totalCommittedInsts; 529 530 ipc 531 .name(name() + ".ipc") 532 .desc("IPC: Instructions Per Cycle") 533 .precision(6); 534 ipc = committedInsts / numCycles; 535 536 totalIpc 537 .name(name() + ".ipc_total") 538 .desc("IPC: Total IPC of All Threads") 539 .precision(6); 540 totalIpc = totalCommittedInsts / numCycles; 541 542 this->fetch.regStats(); 543 this->decode.regStats(); 544 this->rename.regStats(); 545 this->iew.regStats(); 546 this->commit.regStats(); 547 this->rob.regStats(); 548 549 intRegfileReads 550 .name(name() + ".int_regfile_reads") 551 .desc("number of integer regfile reads") 552 .prereq(intRegfileReads); 553 554 intRegfileWrites 555 .name(name() + ".int_regfile_writes") 556 .desc("number of integer regfile writes") 557 .prereq(intRegfileWrites); 558 559 fpRegfileReads 560 .name(name() + ".fp_regfile_reads") 561 .desc("number of floating regfile reads") 562 .prereq(fpRegfileReads); 563 564 fpRegfileWrites 565 .name(name() + ".fp_regfile_writes") 566 .desc("number of floating regfile writes") 567 .prereq(fpRegfileWrites); 568 569 miscRegfileReads 570 .name(name() + ".misc_regfile_reads") 571 .desc("number of misc regfile reads") 572 .prereq(miscRegfileReads); 573 574 miscRegfileWrites 575 .name(name() + ".misc_regfile_writes") 576 .desc("number of misc regfile writes") 577 .prereq(miscRegfileWrites); 578} 579 580template <class Impl> 581Port * 582FullO3CPU<Impl>::getPort(const std::string &if_name, int idx) 583{ 584 if (if_name == "dcache_port") 585 return &dcachePort; 586 else if (if_name == "icache_port") 587 return &icachePort; 588 else 589 panic("No Such Port\n"); 590} 591 592template <class Impl> 593void 594FullO3CPU<Impl>::tick() 595{ 596 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 597 598 ++numCycles; 599 600// activity = false; 601 602 //Tick each of the stages 603 fetch.tick(); 604 605 decode.tick(); 606 607 rename.tick(); 608 609 iew.tick(); 610 611 commit.tick(); 612 613 if (!FullSystem) 614 doContextSwitch(); 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 (!FullSystem) 647 updateThreadPriority(); 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 (FullSystem) { 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 } 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>::totalInsts() 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> 738Counter 739FullO3CPU<Impl>::totalOps() const 740{ 741 Counter total(0); 742 743 ThreadID size = thread.size(); 744 for (ThreadID i = 0; i < size; i++) 745 total += thread[i]->numOp; 746 747 return total; 748} 749 750template <class Impl> 751void 752FullO3CPU<Impl>::activateContext(ThreadID tid, int delay) 753{ 754 // Needs to set each stage to running as well. 755 if (delay){ 756 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 757 "on cycle %d\n", tid, curTick() + ticks(delay)); 758 scheduleActivateThreadEvent(tid, delay); 759 } else { 760 activateThread(tid); 761 } 762 763 if (lastActivatedCycle < curTick()) { 764 scheduleTickEvent(delay); 765 766 // Be sure to signal that there's some activity so the CPU doesn't 767 // deschedule itself. 768 activityRec.activity(); 769 fetch.wakeFromQuiesce(); 770 771 quiesceCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 772 773 lastActivatedCycle = curTick(); 774 775 _status = Running; 776 } 777} 778 779template <class Impl> 780bool 781FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove, 782 int delay) 783{ 784 // Schedule removal of thread data from CPU 785 if (delay){ 786 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 787 "on cycle %d\n", tid, curTick() + ticks(delay)); 788 scheduleDeallocateContextEvent(tid, remove, delay); 789 return false; 790 } else { 791 deactivateThread(tid); 792 if (remove) 793 removeThread(tid); 794 return true; 795 } 796} 797 798template <class Impl> 799void 800FullO3CPU<Impl>::suspendContext(ThreadID tid) 801{ 802 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 803 bool deallocated = scheduleDeallocateContext(tid, false, 1); 804 // If this was the last thread then unschedule the tick event. 805 if ((activeThreads.size() == 1 && !deallocated) || 806 activeThreads.size() == 0) 807 unscheduleTickEvent(); 808 809 DPRINTF(Quiesce, "Suspending Context\n"); 810 lastRunningCycle = curTick(); 811 _status = Idle; 812} 813 814template <class Impl> 815void 816FullO3CPU<Impl>::haltContext(ThreadID tid) 817{ 818 //For now, this is the same as deallocate 819 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 820 scheduleDeallocateContext(tid, true, 1); 821} 822 823template <class Impl> 824void 825FullO3CPU<Impl>::insertThread(ThreadID tid) 826{ 827 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 828 // Will change now that the PC and thread state is internal to the CPU 829 // and not in the ThreadContext. 830 ThreadContext *src_tc; 831 if (FullSystem) 832 src_tc = system->threadContexts[tid]; 833 else 834 src_tc = tcBase(tid); 835 836 //Bind Int Regs to Rename Map 837 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 838 PhysRegIndex phys_reg = freeList.getIntReg(); 839 840 renameMap[tid].setEntry(ireg,phys_reg); 841 scoreboard.setReg(phys_reg); 842 } 843 844 //Bind Float Regs to Rename Map 845 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 846 PhysRegIndex phys_reg = freeList.getFloatReg(); 847 848 renameMap[tid].setEntry(freg,phys_reg); 849 scoreboard.setReg(phys_reg); 850 } 851 852 //Copy Thread Data Into RegFile 853 //this->copyFromTC(tid); 854 855 //Set PC/NPC/NNPC 856 pcState(src_tc->pcState(), tid); 857 858 src_tc->setStatus(ThreadContext::Active); 859 860 activateContext(tid,1); 861 862 //Reset ROB/IQ/LSQ Entries 863 commit.rob->resetEntries(); 864 iew.resetEntries(); 865} 866 867template <class Impl> 868void 869FullO3CPU<Impl>::removeThread(ThreadID tid) 870{ 871 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 872 873 // Copy Thread Data From RegFile 874 // If thread is suspended, it might be re-allocated 875 // this->copyToTC(tid); 876 877 878 // @todo: 2-27-2008: Fix how we free up rename mappings 879 // here to alleviate the case for double-freeing registers 880 // in SMT workloads. 881 882 // Unbind Int Regs from Rename Map 883 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 884 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 885 886 scoreboard.unsetReg(phys_reg); 887 freeList.addReg(phys_reg); 888 } 889 890 // Unbind Float Regs from Rename Map 891 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) { 892 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 893 894 scoreboard.unsetReg(phys_reg); 895 freeList.addReg(phys_reg); 896 } 897 898 // Squash Throughout Pipeline 899 DynInstPtr inst = commit.rob->readHeadInst(tid); 900 InstSeqNum squash_seq_num = inst->seqNum; 901 fetch.squash(0, squash_seq_num, inst, tid); 902 decode.squash(tid); 903 rename.squash(squash_seq_num, tid); 904 iew.squash(tid); 905 iew.ldstQueue.squash(squash_seq_num, tid); 906 commit.rob->squash(squash_seq_num, tid); 907 908 909 assert(iew.instQueue.getCount(tid) == 0); 910 assert(iew.ldstQueue.getCount(tid) == 0); 911 912 // Reset ROB/IQ/LSQ Entries 913 914 // Commented out for now. This should be possible to do by 915 // telling all the pipeline stages to drain first, and then 916 // checking until the drain completes. Once the pipeline is 917 // drained, call resetEntries(). - 10-09-06 ktlim 918/* 919 if (activeThreads.size() >= 1) { 920 commit.rob->resetEntries(); 921 iew.resetEntries(); 922 } 923*/ 924} 925 926 927template <class Impl> 928void 929FullO3CPU<Impl>::activateWhenReady(ThreadID tid) 930{ 931 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 932 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 933 tid); 934 935 bool ready = true; 936 937 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 938 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 939 "Phys. Int. Regs.\n", 940 tid); 941 ready = false; 942 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 943 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 944 "Phys. Float. Regs.\n", 945 tid); 946 ready = false; 947 } else if (commit.rob->numFreeEntries() >= 948 commit.rob->entryAmount(activeThreads.size() + 1)) { 949 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 950 "ROB entries.\n", 951 tid); 952 ready = false; 953 } else if (iew.instQueue.numFreeEntries() >= 954 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 955 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 956 "IQ entries.\n", 957 tid); 958 ready = false; 959 } else if (iew.ldstQueue.numFreeEntries() >= 960 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 961 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 962 "LSQ entries.\n", 963 tid); 964 ready = false; 965 } 966 967 if (ready) { 968 insertThread(tid); 969 970 contextSwitch = false; 971 972 cpuWaitList.remove(tid); 973 } else { 974 suspendContext(tid); 975 976 //blocks fetch 977 contextSwitch = true; 978 979 //@todo: dont always add to waitlist 980 //do waitlist 981 cpuWaitList.push_back(tid); 982 } 983} 984 985template <class Impl> 986Fault 987FullO3CPU<Impl>::hwrei(ThreadID tid) 988{ 989#if THE_ISA == ALPHA_ISA 990 // Need to clear the lock flag upon returning from an interrupt. 991 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 992 993 this->thread[tid]->kernelStats->hwrei(); 994 995 // FIXME: XXX check for interrupts? XXX 996#endif 997 return NoFault; 998} 999 1000template <class Impl> 1001bool 1002FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 1003{ 1004#if THE_ISA == ALPHA_ISA 1005 if (this->thread[tid]->kernelStats) 1006 this->thread[tid]->kernelStats->callpal(palFunc, 1007 this->threadContexts[tid]); 1008 1009 switch (palFunc) { 1010 case PAL::halt: 1011 halt(); 1012 if (--System::numSystemsRunning == 0) 1013 exitSimLoop("all cpus halted"); 1014 break; 1015 1016 case PAL::bpt: 1017 case PAL::bugchk: 1018 if (this->system->breakpoint()) 1019 return false; 1020 break; 1021 } 1022#endif 1023 return true; 1024} 1025 1026template <class Impl> 1027Fault 1028FullO3CPU<Impl>::getInterrupts() 1029{ 1030 // Check if there are any outstanding interrupts 1031 return this->interrupts->getInterrupt(this->threadContexts[0]); 1032} 1033 1034template <class Impl> 1035void 1036FullO3CPU<Impl>::processInterrupts(Fault interrupt) 1037{ 1038 // Check for interrupts here. For now can copy the code that 1039 // exists within isa_fullsys_traits.hh. Also assume that thread 0 1040 // is the one that handles the interrupts. 1041 // @todo: Possibly consolidate the interrupt checking code. 1042 // @todo: Allow other threads to handle interrupts. 1043 1044 assert(interrupt != NoFault); 1045 this->interrupts->updateIntrInfo(this->threadContexts[0]); 1046 1047 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1048 this->trap(interrupt, 0, NULL); 1049} 1050 1051template <class Impl> 1052void 1053FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst) 1054{ 1055 // Pass the thread's TC into the invoke method. 1056 fault->invoke(this->threadContexts[tid], inst); 1057} 1058 1059template <class Impl> 1060void 1061FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid) 1062{ 1063 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1064 1065 DPRINTF(Activity,"Activity: syscall() called.\n"); 1066 1067 // Temporarily increase this by one to account for the syscall 1068 // instruction. 1069 ++(this->thread[tid]->funcExeInst); 1070 1071 // Execute the actual syscall. 1072 this->thread[tid]->syscall(callnum); 1073 1074 // Decrease funcExeInst by one as the normal commit will handle 1075 // incrementing it. 1076 --(this->thread[tid]->funcExeInst); 1077} 1078 1079template <class Impl> 1080void 1081FullO3CPU<Impl>::serialize(std::ostream &os) 1082{ 1083 SimObject::State so_state = SimObject::getState(); 1084 SERIALIZE_ENUM(so_state); 1085 BaseCPU::serialize(os); 1086 nameOut(os, csprintf("%s.tickEvent", name())); 1087 tickEvent.serialize(os); 1088 1089 // Use SimpleThread's ability to checkpoint to make it easier to 1090 // write out the registers. Also make this static so it doesn't 1091 // get instantiated multiple times (causes a panic in statistics). 1092 static SimpleThread temp; 1093 1094 ThreadID size = thread.size(); 1095 for (ThreadID i = 0; i < size; i++) { 1096 nameOut(os, csprintf("%s.xc.%i", name(), i)); 1097 temp.copyTC(thread[i]->getTC()); 1098 temp.serialize(os); 1099 } 1100} 1101 1102template <class Impl> 1103void 1104FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 1105{ 1106 SimObject::State so_state; 1107 UNSERIALIZE_ENUM(so_state); 1108 BaseCPU::unserialize(cp, section); 1109 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 1110 1111 // Use SimpleThread's ability to checkpoint to make it easier to 1112 // read in the registers. Also make this static so it doesn't 1113 // get instantiated multiple times (causes a panic in statistics). 1114 static SimpleThread temp; 1115 1116 ThreadID size = thread.size(); 1117 for (ThreadID i = 0; i < size; i++) { 1118 temp.copyTC(thread[i]->getTC()); 1119 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 1120 thread[i]->getTC()->copyArchRegs(temp.getTC()); 1121 } 1122} 1123 1124template <class Impl> 1125unsigned int 1126FullO3CPU<Impl>::drain(Event *drain_event) 1127{ 1128 DPRINTF(O3CPU, "Switching out\n"); 1129 1130 // If the CPU isn't doing anything, then return immediately. 1131 if (_status == Idle || _status == SwitchedOut) { 1132 return 0; 1133 } 1134 1135 drainCount = 0; 1136 fetch.drain(); 1137 decode.drain(); 1138 rename.drain(); 1139 iew.drain(); 1140 commit.drain(); 1141 1142 // Wake the CPU and record activity so everything can drain out if 1143 // the CPU was not able to immediately drain. 1144 if (getState() != SimObject::Drained) { 1145 // A bit of a hack...set the drainEvent after all the drain() 1146 // calls have been made, that way if all of the stages drain 1147 // immediately, the signalDrained() function knows not to call 1148 // process on the drain event. 1149 drainEvent = drain_event; 1150 1151 wakeCPU(); 1152 activityRec.activity(); 1153 1154 return 1; 1155 } else { 1156 return 0; 1157 } 1158} 1159 1160template <class Impl> 1161void 1162FullO3CPU<Impl>::resume() 1163{ 1164 fetch.resume(); 1165 decode.resume(); 1166 rename.resume(); 1167 iew.resume(); 1168 commit.resume(); 1169 1170 changeState(SimObject::Running); 1171 1172 if (_status == SwitchedOut || _status == Idle) 1173 return; 1174 1175 assert(system->getMemoryMode() == Enums::timing); 1176 1177 if (!tickEvent.scheduled()) 1178 schedule(tickEvent, nextCycle()); 1179 _status = Running; 1180} 1181 1182template <class Impl> 1183void 1184FullO3CPU<Impl>::signalDrained() 1185{ 1186 if (++drainCount == NumStages) { 1187 if (tickEvent.scheduled()) 1188 tickEvent.squash(); 1189 1190 changeState(SimObject::Drained); 1191 1192 BaseCPU::switchOut(); 1193 1194 if (drainEvent) { 1195 drainEvent->process(); 1196 drainEvent = NULL; 1197 } 1198 } 1199 assert(drainCount <= 5); 1200} 1201 1202template <class Impl> 1203void 1204FullO3CPU<Impl>::switchOut() 1205{ 1206 fetch.switchOut(); 1207 rename.switchOut(); 1208 iew.switchOut(); 1209 commit.switchOut(); 1210 instList.clear(); 1211 while (!removeList.empty()) { 1212 removeList.pop(); 1213 } 1214 1215 _status = SwitchedOut; 1216#if USE_CHECKER 1217 if (checker) 1218 checker->switchOut(); 1219#endif 1220 if (tickEvent.scheduled()) 1221 tickEvent.squash(); 1222} 1223 1224template <class Impl> 1225void 1226FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1227{ 1228 // Flush out any old data from the time buffers. 1229 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1230 timeBuffer.advance(); 1231 fetchQueue.advance(); 1232 decodeQueue.advance(); 1233 renameQueue.advance(); 1234 iewQueue.advance(); 1235 } 1236 1237 activityRec.reset(); 1238 1239 BaseCPU::takeOverFrom(oldCPU); 1240 1241 fetch.takeOverFrom(); 1242 decode.takeOverFrom(); 1243 rename.takeOverFrom(); 1244 iew.takeOverFrom(); 1245 commit.takeOverFrom(); 1246 1247 assert(!tickEvent.scheduled() || tickEvent.squashed()); 1248 1249 // @todo: Figure out how to properly select the tid to put onto 1250 // the active threads list. 1251 ThreadID tid = 0; 1252 1253 list<ThreadID>::iterator isActive = 1254 std::find(activeThreads.begin(), activeThreads.end(), tid); 1255 1256 if (isActive == activeThreads.end()) { 1257 //May Need to Re-code this if the delay variable is the delay 1258 //needed for thread to activate 1259 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 1260 tid); 1261 1262 activeThreads.push_back(tid); 1263 } 1264 1265 // Set all statuses to active, schedule the CPU's tick event. 1266 // @todo: Fix up statuses so this is handled properly 1267 ThreadID size = threadContexts.size(); 1268 for (ThreadID i = 0; i < size; ++i) { 1269 ThreadContext *tc = threadContexts[i]; 1270 if (tc->status() == ThreadContext::Active && _status != Running) { 1271 _status = Running; 1272 reschedule(tickEvent, nextCycle(), true); 1273 } 1274 } 1275 if (!tickEvent.scheduled()) 1276 schedule(tickEvent, nextCycle()); 1277 1278 lastRunningCycle = curTick(); 1279} 1280 1281template <class Impl> 1282TheISA::MiscReg 1283FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) 1284{ 1285 return this->isa[tid].readMiscRegNoEffect(misc_reg); 1286} 1287 1288template <class Impl> 1289TheISA::MiscReg 1290FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1291{ 1292 miscRegfileReads++; 1293 return this->isa[tid].readMiscReg(misc_reg, tcBase(tid)); 1294} 1295 1296template <class Impl> 1297void 1298FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1299 const TheISA::MiscReg &val, ThreadID tid) 1300{ 1301 this->isa[tid].setMiscRegNoEffect(misc_reg, val); 1302} 1303 1304template <class Impl> 1305void 1306FullO3CPU<Impl>::setMiscReg(int misc_reg, 1307 const TheISA::MiscReg &val, ThreadID tid) 1308{ 1309 miscRegfileWrites++; 1310 this->isa[tid].setMiscReg(misc_reg, val, tcBase(tid)); 1311} 1312 1313template <class Impl> 1314uint64_t 1315FullO3CPU<Impl>::readIntReg(int reg_idx) 1316{ 1317 intRegfileReads++; 1318 return regFile.readIntReg(reg_idx); 1319} 1320 1321template <class Impl> 1322FloatReg 1323FullO3CPU<Impl>::readFloatReg(int reg_idx) 1324{ 1325 fpRegfileReads++; 1326 return regFile.readFloatReg(reg_idx); 1327} 1328 1329template <class Impl> 1330FloatRegBits 1331FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1332{ 1333 fpRegfileReads++; 1334 return regFile.readFloatRegBits(reg_idx); 1335} 1336 1337template <class Impl> 1338void 1339FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1340{ 1341 intRegfileWrites++; 1342 regFile.setIntReg(reg_idx, val); 1343} 1344 1345template <class Impl> 1346void 1347FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1348{ 1349 fpRegfileWrites++; 1350 regFile.setFloatReg(reg_idx, val); 1351} 1352 1353template <class Impl> 1354void 1355FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1356{ 1357 fpRegfileWrites++; 1358 regFile.setFloatRegBits(reg_idx, val); 1359} 1360 1361template <class Impl> 1362uint64_t 1363FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1364{ 1365 intRegfileReads++; 1366 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1367 1368 return regFile.readIntReg(phys_reg); 1369} 1370 1371template <class Impl> 1372float 1373FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1374{ 1375 fpRegfileReads++; 1376 int idx = reg_idx + TheISA::NumIntRegs; 1377 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1378 1379 return regFile.readFloatReg(phys_reg); 1380} 1381 1382template <class Impl> 1383uint64_t 1384FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1385{ 1386 fpRegfileReads++; 1387 int idx = reg_idx + TheISA::NumIntRegs; 1388 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1389 1390 return regFile.readFloatRegBits(phys_reg); 1391} 1392 1393template <class Impl> 1394void 1395FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1396{ 1397 intRegfileWrites++; 1398 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1399 1400 regFile.setIntReg(phys_reg, val); 1401} 1402 1403template <class Impl> 1404void 1405FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1406{ 1407 fpRegfileWrites++; 1408 int idx = reg_idx + TheISA::NumIntRegs; 1409 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1410 1411 regFile.setFloatReg(phys_reg, val); 1412} 1413 1414template <class Impl> 1415void 1416FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1417{ 1418 fpRegfileWrites++; 1419 int idx = reg_idx + TheISA::NumIntRegs; 1420 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1421 1422 regFile.setFloatRegBits(phys_reg, val); 1423} 1424 1425template <class Impl> 1426TheISA::PCState 1427FullO3CPU<Impl>::pcState(ThreadID tid) 1428{ 1429 return commit.pcState(tid); 1430} 1431 1432template <class Impl> 1433void 1434FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1435{ 1436 commit.pcState(val, tid); 1437} 1438 1439template <class Impl> 1440Addr 1441FullO3CPU<Impl>::instAddr(ThreadID tid) 1442{ 1443 return commit.instAddr(tid); 1444} 1445 1446template <class Impl> 1447Addr 1448FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1449{ 1450 return commit.nextInstAddr(tid); 1451} 1452 1453template <class Impl> 1454MicroPC 1455FullO3CPU<Impl>::microPC(ThreadID tid) 1456{ 1457 return commit.microPC(tid); 1458} 1459 1460template <class Impl> 1461void 1462FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1463{ 1464 this->thread[tid]->inSyscall = true; 1465 this->commit.generateTCEvent(tid); 1466} 1467 1468template <class Impl> 1469typename FullO3CPU<Impl>::ListIt 1470FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1471{ 1472 instList.push_back(inst); 1473 1474 return --(instList.end()); 1475} 1476 1477template <class Impl> 1478void 1479FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst) 1480{ 1481 // Keep an instruction count. 1482 if (!inst->isMicroop() || inst->isLastMicroop()) { 1483 thread[tid]->numInst++; 1484 thread[tid]->numInsts++; 1485 committedInsts[tid]++; 1486 totalCommittedInsts++; 1487 } 1488 thread[tid]->numOp++; 1489 thread[tid]->numOps++; 1490 committedOps[tid]++; 1491 1492 system->totalNumInsts++; 1493 // Check for instruction-count-based events. 1494 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1495 system->instEventQueue.serviceEvents(system->totalNumInsts); 1496} 1497 1498template <class Impl> 1499void 1500FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1501{ 1502 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1503 "[sn:%lli]\n", 1504 inst->threadNumber, inst->pcState(), inst->seqNum); 1505 1506 removeInstsThisCycle = true; 1507 1508 // Remove the front instruction. 1509 removeList.push(inst->getInstListIt()); 1510} 1511 1512template <class Impl> 1513void 1514FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1515{ 1516 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1517 " list.\n", tid); 1518 1519 ListIt end_it; 1520 1521 bool rob_empty = false; 1522 1523 if (instList.empty()) { 1524 return; 1525 } else if (rob.isEmpty(/*tid*/)) { 1526 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1527 end_it = instList.begin(); 1528 rob_empty = true; 1529 } else { 1530 end_it = (rob.readTailInst(tid))->getInstListIt(); 1531 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1532 } 1533 1534 removeInstsThisCycle = true; 1535 1536 ListIt inst_it = instList.end(); 1537 1538 inst_it--; 1539 1540 // Walk through the instruction list, removing any instructions 1541 // that were inserted after the given instruction iterator, end_it. 1542 while (inst_it != end_it) { 1543 assert(!instList.empty()); 1544 1545 squashInstIt(inst_it, tid); 1546 1547 inst_it--; 1548 } 1549 1550 // If the ROB was empty, then we actually need to remove the first 1551 // instruction as well. 1552 if (rob_empty) { 1553 squashInstIt(inst_it, tid); 1554 } 1555} 1556 1557template <class Impl> 1558void 1559FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1560{ 1561 assert(!instList.empty()); 1562 1563 removeInstsThisCycle = true; 1564 1565 ListIt inst_iter = instList.end(); 1566 1567 inst_iter--; 1568 1569 DPRINTF(O3CPU, "Deleting instructions from instruction " 1570 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1571 tid, seq_num, (*inst_iter)->seqNum); 1572 1573 while ((*inst_iter)->seqNum > seq_num) { 1574 1575 bool break_loop = (inst_iter == instList.begin()); 1576 1577 squashInstIt(inst_iter, tid); 1578 1579 inst_iter--; 1580 1581 if (break_loop) 1582 break; 1583 } 1584} 1585 1586template <class Impl> 1587inline void 1588FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1589{ 1590 if ((*instIt)->threadNumber == tid) { 1591 DPRINTF(O3CPU, "Squashing instruction, " 1592 "[tid:%i] [sn:%lli] PC %s\n", 1593 (*instIt)->threadNumber, 1594 (*instIt)->seqNum, 1595 (*instIt)->pcState()); 1596 1597 // Mark it as squashed. 1598 (*instIt)->setSquashed(); 1599 1600 // @todo: Formulate a consistent method for deleting 1601 // instructions from the instruction list 1602 // Remove the instruction from the list. 1603 removeList.push(instIt); 1604 } 1605} 1606 1607template <class Impl> 1608void 1609FullO3CPU<Impl>::cleanUpRemovedInsts() 1610{ 1611 while (!removeList.empty()) { 1612 DPRINTF(O3CPU, "Removing instruction, " 1613 "[tid:%i] [sn:%lli] PC %s\n", 1614 (*removeList.front())->threadNumber, 1615 (*removeList.front())->seqNum, 1616 (*removeList.front())->pcState()); 1617 1618 instList.erase(removeList.front()); 1619 1620 removeList.pop(); 1621 } 1622 1623 removeInstsThisCycle = false; 1624} 1625/* 1626template <class Impl> 1627void 1628FullO3CPU<Impl>::removeAllInsts() 1629{ 1630 instList.clear(); 1631} 1632*/ 1633template <class Impl> 1634void 1635FullO3CPU<Impl>::dumpInsts() 1636{ 1637 int num = 0; 1638 1639 ListIt inst_list_it = instList.begin(); 1640 1641 cprintf("Dumping Instruction List\n"); 1642 1643 while (inst_list_it != instList.end()) { 1644 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1645 "Squashed:%i\n\n", 1646 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1647 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1648 (*inst_list_it)->isSquashed()); 1649 inst_list_it++; 1650 ++num; 1651 } 1652} 1653/* 1654template <class Impl> 1655void 1656FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1657{ 1658 iew.wakeDependents(inst); 1659} 1660*/ 1661template <class Impl> 1662void 1663FullO3CPU<Impl>::wakeCPU() 1664{ 1665 if (activityRec.active() || tickEvent.scheduled()) { 1666 DPRINTF(Activity, "CPU already running.\n"); 1667 return; 1668 } 1669 1670 DPRINTF(Activity, "Waking up CPU\n"); 1671 1672 idleCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 1673 numCycles += tickToCycles((curTick() - 1) - lastRunningCycle); 1674 1675 schedule(tickEvent, nextCycle()); 1676} 1677 1678template <class Impl> 1679void 1680FullO3CPU<Impl>::wakeup() 1681{ 1682 if (this->thread[0]->status() != ThreadContext::Suspended) 1683 return; 1684 1685 this->wakeCPU(); 1686 1687 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1688 this->threadContexts[0]->activate(); 1689} 1690 1691template <class Impl> 1692ThreadID 1693FullO3CPU<Impl>::getFreeTid() 1694{ 1695 for (ThreadID tid = 0; tid < numThreads; tid++) { 1696 if (!tids[tid]) { 1697 tids[tid] = true; 1698 return tid; 1699 } 1700 } 1701 1702 return InvalidThreadID; 1703} 1704 1705template <class Impl> 1706void 1707FullO3CPU<Impl>::doContextSwitch() 1708{ 1709 if (contextSwitch) { 1710 1711 //ADD CODE TO DEACTIVE THREAD HERE (???) 1712 1713 ThreadID size = cpuWaitList.size(); 1714 for (ThreadID tid = 0; tid < size; tid++) { 1715 activateWhenReady(tid); 1716 } 1717 1718 if (cpuWaitList.size() == 0) 1719 contextSwitch = true; 1720 } 1721} 1722 1723template <class Impl> 1724void 1725FullO3CPU<Impl>::updateThreadPriority() 1726{ 1727 if (activeThreads.size() > 1) { 1728 //DEFAULT TO ROUND ROBIN SCHEME 1729 //e.g. Move highest priority to end of thread list 1730 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1731 1732 unsigned high_thread = *list_begin; 1733 1734 activeThreads.erase(list_begin); 1735 1736 activeThreads.push_back(high_thread); 1737 } 1738} 1739 1740// Forward declaration of FullO3CPU. 1741template class FullO3CPU<O3CPUImpl>; 1742