cpu.cc revision 10487:5914229e6b16
1/* 2 * Copyright (c) 2011-2012, 2014 ARM Limited 3 * Copyright (c) 2013 Advanced Micro Devices, Inc. 4 * All rights reserved 5 * 6 * The license below extends only to copyright in the software and shall 7 * not be construed as granting a license to any other intellectual 8 * property including but not limited to intellectual property relating 9 * to a hardware implementation of the functionality of the software 10 * licensed hereunder. You may use the software subject to the license 11 * terms below provided that you ensure that this notice is replicated 12 * unmodified and in its entirety in all distributions of the software, 13 * modified or unmodified, in source code or in binary form. 14 * 15 * Copyright (c) 2004-2006 The Regents of The University of Michigan 16 * Copyright (c) 2011 Regents of the University of California 17 * All rights reserved. 18 * 19 * Redistribution and use in source and binary forms, with or without 20 * modification, are permitted provided that the following conditions are 21 * met: redistributions of source code must retain the above copyright 22 * notice, this list of conditions and the following disclaimer; 23 * redistributions in binary form must reproduce the above copyright 24 * notice, this list of conditions and the following disclaimer in the 25 * documentation and/or other materials provided with the distribution; 26 * neither the name of the copyright holders nor the names of its 27 * contributors may be used to endorse or promote products derived from 28 * this software without specific prior written permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 31 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 32 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 33 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 34 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 35 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 36 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 37 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 38 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 39 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 40 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 41 * 42 * Authors: Kevin Lim 43 * Korey Sewell 44 * Rick Strong 45 */ 46 47#include "arch/kernel_stats.hh" 48#include "config/the_isa.hh" 49#include "cpu/checker/cpu.hh" 50#include "cpu/checker/thread_context.hh" 51#include "cpu/o3/cpu.hh" 52#include "cpu/o3/isa_specific.hh" 53#include "cpu/o3/thread_context.hh" 54#include "cpu/activity.hh" 55#include "cpu/quiesce_event.hh" 56#include "cpu/simple_thread.hh" 57#include "cpu/thread_context.hh" 58#include "debug/Activity.hh" 59#include "debug/Drain.hh" 60#include "debug/O3CPU.hh" 61#include "debug/Quiesce.hh" 62#include "enums/MemoryMode.hh" 63#include "sim/core.hh" 64#include "sim/full_system.hh" 65#include "sim/process.hh" 66#include "sim/stat_control.hh" 67#include "sim/system.hh" 68 69#if THE_ISA == ALPHA_ISA 70#include "arch/alpha/osfpal.hh" 71#include "debug/Activity.hh" 72#endif 73 74struct BaseCPUParams; 75 76using namespace TheISA; 77using namespace std; 78 79BaseO3CPU::BaseO3CPU(BaseCPUParams *params) 80 : BaseCPU(params) 81{ 82} 83 84void 85BaseO3CPU::regStats() 86{ 87 BaseCPU::regStats(); 88} 89 90template<class Impl> 91bool 92FullO3CPU<Impl>::IcachePort::recvTimingResp(PacketPtr pkt) 93{ 94 DPRINTF(O3CPU, "Fetch unit received timing\n"); 95 // We shouldn't ever get a block in ownership state 96 assert(!(pkt->memInhibitAsserted() && !pkt->sharedAsserted())); 97 fetch->processCacheCompletion(pkt); 98 99 return true; 100} 101 102template<class Impl> 103void 104FullO3CPU<Impl>::IcachePort::recvRetry() 105{ 106 fetch->recvRetry(); 107} 108 109template <class Impl> 110bool 111FullO3CPU<Impl>::DcachePort::recvTimingResp(PacketPtr pkt) 112{ 113 return lsq->recvTimingResp(pkt); 114} 115 116template <class Impl> 117void 118FullO3CPU<Impl>::DcachePort::recvTimingSnoopReq(PacketPtr pkt) 119{ 120 lsq->recvTimingSnoopReq(pkt); 121} 122 123template <class Impl> 124void 125FullO3CPU<Impl>::DcachePort::recvRetry() 126{ 127 lsq->recvRetry(); 128} 129 130template <class Impl> 131FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 132 : Event(CPU_Tick_Pri), cpu(c) 133{ 134} 135 136template <class Impl> 137void 138FullO3CPU<Impl>::TickEvent::process() 139{ 140 cpu->tick(); 141} 142 143template <class Impl> 144const char * 145FullO3CPU<Impl>::TickEvent::description() const 146{ 147 return "FullO3CPU tick"; 148} 149 150template <class Impl> 151FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params) 152 : BaseO3CPU(params), 153 itb(params->itb), 154 dtb(params->dtb), 155 tickEvent(this), 156#ifndef NDEBUG 157 instcount(0), 158#endif 159 removeInstsThisCycle(false), 160 fetch(this, params), 161 decode(this, params), 162 rename(this, params), 163 iew(this, params), 164 commit(this, params), 165 166 regFile(params->numPhysIntRegs, 167 params->numPhysFloatRegs, 168 params->numPhysCCRegs), 169 170 freeList(name() + ".freelist", ®File), 171 172 rob(this, params), 173 174 scoreboard(name() + ".scoreboard", 175 regFile.totalNumPhysRegs(), TheISA::NumMiscRegs, 176 TheISA::ZeroReg, TheISA::ZeroReg), 177 178 isa(numThreads, NULL), 179 180 icachePort(&fetch, this), 181 dcachePort(&iew.ldstQueue, this), 182 183 timeBuffer(params->backComSize, params->forwardComSize), 184 fetchQueue(params->backComSize, params->forwardComSize), 185 decodeQueue(params->backComSize, params->forwardComSize), 186 renameQueue(params->backComSize, params->forwardComSize), 187 iewQueue(params->backComSize, params->forwardComSize), 188 activityRec(name(), NumStages, 189 params->backComSize + params->forwardComSize, 190 params->activity), 191 192 globalSeqNum(1), 193 system(params->system), 194 drainManager(NULL), 195 lastRunningCycle(curCycle()) 196{ 197 if (!params->switched_out) { 198 _status = Running; 199 } else { 200 _status = SwitchedOut; 201 } 202 203 if (params->checker) { 204 BaseCPU *temp_checker = params->checker; 205 checker = dynamic_cast<Checker<Impl> *>(temp_checker); 206 checker->setIcachePort(&icachePort); 207 checker->setSystem(params->system); 208 } else { 209 checker = NULL; 210 } 211 212 if (!FullSystem) { 213 thread.resize(numThreads); 214 tids.resize(numThreads); 215 } 216 217 // The stages also need their CPU pointer setup. However this 218 // must be done at the upper level CPU because they have pointers 219 // to the upper level CPU, and not this FullO3CPU. 220 221 // Set up Pointers to the activeThreads list for each stage 222 fetch.setActiveThreads(&activeThreads); 223 decode.setActiveThreads(&activeThreads); 224 rename.setActiveThreads(&activeThreads); 225 iew.setActiveThreads(&activeThreads); 226 commit.setActiveThreads(&activeThreads); 227 228 // Give each of the stages the time buffer they will use. 229 fetch.setTimeBuffer(&timeBuffer); 230 decode.setTimeBuffer(&timeBuffer); 231 rename.setTimeBuffer(&timeBuffer); 232 iew.setTimeBuffer(&timeBuffer); 233 commit.setTimeBuffer(&timeBuffer); 234 235 // Also setup each of the stages' queues. 236 fetch.setFetchQueue(&fetchQueue); 237 decode.setFetchQueue(&fetchQueue); 238 commit.setFetchQueue(&fetchQueue); 239 decode.setDecodeQueue(&decodeQueue); 240 rename.setDecodeQueue(&decodeQueue); 241 rename.setRenameQueue(&renameQueue); 242 iew.setRenameQueue(&renameQueue); 243 iew.setIEWQueue(&iewQueue); 244 commit.setIEWQueue(&iewQueue); 245 commit.setRenameQueue(&renameQueue); 246 247 commit.setIEWStage(&iew); 248 rename.setIEWStage(&iew); 249 rename.setCommitStage(&commit); 250 251 ThreadID active_threads; 252 if (FullSystem) { 253 active_threads = 1; 254 } else { 255 active_threads = params->workload.size(); 256 257 if (active_threads > Impl::MaxThreads) { 258 panic("Workload Size too large. Increase the 'MaxThreads' " 259 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) " 260 "or edit your workload size."); 261 } 262 } 263 264 //Make Sure That this a Valid Architeture 265 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 266 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 267 assert(params->numPhysCCRegs >= numThreads * TheISA::NumCCRegs); 268 269 rename.setScoreboard(&scoreboard); 270 iew.setScoreboard(&scoreboard); 271 272 // Setup the rename map for whichever stages need it. 273 for (ThreadID tid = 0; tid < numThreads; tid++) { 274 isa[tid] = params->isa[tid]; 275 276 // Only Alpha has an FP zero register, so for other ISAs we 277 // use an invalid FP register index to avoid special treatment 278 // of any valid FP reg. 279 RegIndex invalidFPReg = TheISA::NumFloatRegs + 1; 280 RegIndex fpZeroReg = 281 (THE_ISA == ALPHA_ISA) ? TheISA::ZeroReg : invalidFPReg; 282 283 commitRenameMap[tid].init(®File, TheISA::ZeroReg, fpZeroReg, 284 &freeList); 285 286 renameMap[tid].init(®File, TheISA::ZeroReg, fpZeroReg, 287 &freeList); 288 } 289 290 // Initialize rename map to assign physical registers to the 291 // architectural registers for active threads only. 292 for (ThreadID tid = 0; tid < active_threads; tid++) { 293 for (RegIndex ridx = 0; ridx < TheISA::NumIntRegs; ++ridx) { 294 // Note that we can't use the rename() method because we don't 295 // want special treatment for the zero register at this point 296 PhysRegIndex phys_reg = freeList.getIntReg(); 297 renameMap[tid].setIntEntry(ridx, phys_reg); 298 commitRenameMap[tid].setIntEntry(ridx, phys_reg); 299 } 300 301 for (RegIndex ridx = 0; ridx < TheISA::NumFloatRegs; ++ridx) { 302 PhysRegIndex phys_reg = freeList.getFloatReg(); 303 renameMap[tid].setFloatEntry(ridx, phys_reg); 304 commitRenameMap[tid].setFloatEntry(ridx, phys_reg); 305 } 306 307 for (RegIndex ridx = 0; ridx < TheISA::NumCCRegs; ++ridx) { 308 PhysRegIndex phys_reg = freeList.getCCReg(); 309 renameMap[tid].setCCEntry(ridx, phys_reg); 310 commitRenameMap[tid].setCCEntry(ridx, phys_reg); 311 } 312 } 313 314 rename.setRenameMap(renameMap); 315 commit.setRenameMap(commitRenameMap); 316 rename.setFreeList(&freeList); 317 318 // Setup the ROB for whichever stages need it. 319 commit.setROB(&rob); 320 321 lastActivatedCycle = 0; 322#if 0 323 // Give renameMap & rename stage access to the freeList; 324 for (ThreadID tid = 0; tid < numThreads; tid++) 325 globalSeqNum[tid] = 1; 326#endif 327 328 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 329 330 // Setup any thread state. 331 this->thread.resize(this->numThreads); 332 333 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 334 if (FullSystem) { 335 // SMT is not supported in FS mode yet. 336 assert(this->numThreads == 1); 337 this->thread[tid] = new Thread(this, 0, NULL); 338 } else { 339 if (tid < params->workload.size()) { 340 DPRINTF(O3CPU, "Workload[%i] process is %#x", 341 tid, this->thread[tid]); 342 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 343 (typename Impl::O3CPU *)(this), 344 tid, params->workload[tid]); 345 346 //usedTids[tid] = true; 347 //threadMap[tid] = tid; 348 } else { 349 //Allocate Empty thread so M5 can use later 350 //when scheduling threads to CPU 351 Process* dummy_proc = NULL; 352 353 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 354 (typename Impl::O3CPU *)(this), 355 tid, dummy_proc); 356 //usedTids[tid] = false; 357 } 358 } 359 360 ThreadContext *tc; 361 362 // Setup the TC that will serve as the interface to the threads/CPU. 363 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 364 365 tc = o3_tc; 366 367 // If we're using a checker, then the TC should be the 368 // CheckerThreadContext. 369 if (params->checker) { 370 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 371 o3_tc, this->checker); 372 } 373 374 o3_tc->cpu = (typename Impl::O3CPU *)(this); 375 assert(o3_tc->cpu); 376 o3_tc->thread = this->thread[tid]; 377 378 if (FullSystem) { 379 // Setup quiesce event. 380 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 381 } 382 // Give the thread the TC. 383 this->thread[tid]->tc = tc; 384 385 // Add the TC to the CPU's list of TC's. 386 this->threadContexts.push_back(tc); 387 } 388 389 // FullO3CPU always requires an interrupt controller. 390 if (!params->switched_out && !interrupts) { 391 fatal("FullO3CPU %s has no interrupt controller.\n" 392 "Ensure createInterruptController() is called.\n", name()); 393 } 394 395 for (ThreadID tid = 0; tid < this->numThreads; tid++) 396 this->thread[tid]->setFuncExeInst(0); 397} 398 399template <class Impl> 400FullO3CPU<Impl>::~FullO3CPU() 401{ 402} 403 404template <class Impl> 405void 406FullO3CPU<Impl>::regProbePoints() 407{ 408 BaseCPU::regProbePoints(); 409 410 ppInstAccessComplete = new ProbePointArg<PacketPtr>(getProbeManager(), "InstAccessComplete"); 411 ppDataAccessComplete = new ProbePointArg<std::pair<DynInstPtr, PacketPtr> >(getProbeManager(), "DataAccessComplete"); 412 413 fetch.regProbePoints(); 414 iew.regProbePoints(); 415 commit.regProbePoints(); 416} 417 418template <class Impl> 419void 420FullO3CPU<Impl>::regStats() 421{ 422 BaseO3CPU::regStats(); 423 424 // Register any of the O3CPU's stats here. 425 timesIdled 426 .name(name() + ".timesIdled") 427 .desc("Number of times that the entire CPU went into an idle state and" 428 " unscheduled itself") 429 .prereq(timesIdled); 430 431 idleCycles 432 .name(name() + ".idleCycles") 433 .desc("Total number of cycles that the CPU has spent unscheduled due " 434 "to idling") 435 .prereq(idleCycles); 436 437 quiesceCycles 438 .name(name() + ".quiesceCycles") 439 .desc("Total number of cycles that CPU has spent quiesced or waiting " 440 "for an interrupt") 441 .prereq(quiesceCycles); 442 443 // Number of Instructions simulated 444 // -------------------------------- 445 // Should probably be in Base CPU but need templated 446 // MaxThreads so put in here instead 447 committedInsts 448 .init(numThreads) 449 .name(name() + ".committedInsts") 450 .desc("Number of Instructions Simulated") 451 .flags(Stats::total); 452 453 committedOps 454 .init(numThreads) 455 .name(name() + ".committedOps") 456 .desc("Number of Ops (including micro ops) Simulated") 457 .flags(Stats::total); 458 459 cpi 460 .name(name() + ".cpi") 461 .desc("CPI: Cycles Per Instruction") 462 .precision(6); 463 cpi = numCycles / committedInsts; 464 465 totalCpi 466 .name(name() + ".cpi_total") 467 .desc("CPI: Total CPI of All Threads") 468 .precision(6); 469 totalCpi = numCycles / sum(committedInsts); 470 471 ipc 472 .name(name() + ".ipc") 473 .desc("IPC: Instructions Per Cycle") 474 .precision(6); 475 ipc = committedInsts / numCycles; 476 477 totalIpc 478 .name(name() + ".ipc_total") 479 .desc("IPC: Total IPC of All Threads") 480 .precision(6); 481 totalIpc = sum(committedInsts) / numCycles; 482 483 this->fetch.regStats(); 484 this->decode.regStats(); 485 this->rename.regStats(); 486 this->iew.regStats(); 487 this->commit.regStats(); 488 this->rob.regStats(); 489 490 intRegfileReads 491 .name(name() + ".int_regfile_reads") 492 .desc("number of integer regfile reads") 493 .prereq(intRegfileReads); 494 495 intRegfileWrites 496 .name(name() + ".int_regfile_writes") 497 .desc("number of integer regfile writes") 498 .prereq(intRegfileWrites); 499 500 fpRegfileReads 501 .name(name() + ".fp_regfile_reads") 502 .desc("number of floating regfile reads") 503 .prereq(fpRegfileReads); 504 505 fpRegfileWrites 506 .name(name() + ".fp_regfile_writes") 507 .desc("number of floating regfile writes") 508 .prereq(fpRegfileWrites); 509 510 ccRegfileReads 511 .name(name() + ".cc_regfile_reads") 512 .desc("number of cc regfile reads") 513 .prereq(ccRegfileReads); 514 515 ccRegfileWrites 516 .name(name() + ".cc_regfile_writes") 517 .desc("number of cc regfile writes") 518 .prereq(ccRegfileWrites); 519 520 miscRegfileReads 521 .name(name() + ".misc_regfile_reads") 522 .desc("number of misc regfile reads") 523 .prereq(miscRegfileReads); 524 525 miscRegfileWrites 526 .name(name() + ".misc_regfile_writes") 527 .desc("number of misc regfile writes") 528 .prereq(miscRegfileWrites); 529} 530 531template <class Impl> 532void 533FullO3CPU<Impl>::tick() 534{ 535 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 536 assert(!switchedOut()); 537 assert(getDrainState() != Drainable::Drained); 538 539 ++numCycles; 540 ppCycles->notify(1); 541 542// activity = false; 543 544 //Tick each of the stages 545 fetch.tick(); 546 547 decode.tick(); 548 549 rename.tick(); 550 551 iew.tick(); 552 553 commit.tick(); 554 555 // Now advance the time buffers 556 timeBuffer.advance(); 557 558 fetchQueue.advance(); 559 decodeQueue.advance(); 560 renameQueue.advance(); 561 iewQueue.advance(); 562 563 activityRec.advance(); 564 565 if (removeInstsThisCycle) { 566 cleanUpRemovedInsts(); 567 } 568 569 if (!tickEvent.scheduled()) { 570 if (_status == SwitchedOut) { 571 DPRINTF(O3CPU, "Switched out!\n"); 572 // increment stat 573 lastRunningCycle = curCycle(); 574 } else if (!activityRec.active() || _status == Idle) { 575 DPRINTF(O3CPU, "Idle!\n"); 576 lastRunningCycle = curCycle(); 577 timesIdled++; 578 } else { 579 schedule(tickEvent, clockEdge(Cycles(1))); 580 DPRINTF(O3CPU, "Scheduling next tick!\n"); 581 } 582 } 583 584 if (!FullSystem) 585 updateThreadPriority(); 586 587 tryDrain(); 588} 589 590template <class Impl> 591void 592FullO3CPU<Impl>::init() 593{ 594 BaseCPU::init(); 595 596 for (ThreadID tid = 0; tid < numThreads; ++tid) { 597 // Set noSquashFromTC so that the CPU doesn't squash when initially 598 // setting up registers. 599 thread[tid]->noSquashFromTC = true; 600 // Initialise the ThreadContext's memory proxies 601 thread[tid]->initMemProxies(thread[tid]->getTC()); 602 } 603 604 if (FullSystem && !params()->switched_out) { 605 for (ThreadID tid = 0; tid < numThreads; tid++) { 606 ThreadContext *src_tc = threadContexts[tid]; 607 TheISA::initCPU(src_tc, src_tc->contextId()); 608 } 609 } 610 611 // Clear noSquashFromTC. 612 for (int tid = 0; tid < numThreads; ++tid) 613 thread[tid]->noSquashFromTC = false; 614 615 commit.setThreads(thread); 616} 617 618template <class Impl> 619void 620FullO3CPU<Impl>::startup() 621{ 622 BaseCPU::startup(); 623 for (int tid = 0; tid < numThreads; ++tid) 624 isa[tid]->startup(threadContexts[tid]); 625 626 fetch.startupStage(); 627 decode.startupStage(); 628 iew.startupStage(); 629 rename.startupStage(); 630 commit.startupStage(); 631} 632 633template <class Impl> 634void 635FullO3CPU<Impl>::activateThread(ThreadID tid) 636{ 637 list<ThreadID>::iterator isActive = 638 std::find(activeThreads.begin(), activeThreads.end(), tid); 639 640 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 641 assert(!switchedOut()); 642 643 if (isActive == activeThreads.end()) { 644 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 645 tid); 646 647 activeThreads.push_back(tid); 648 } 649} 650 651template <class Impl> 652void 653FullO3CPU<Impl>::deactivateThread(ThreadID tid) 654{ 655 //Remove From Active List, if Active 656 list<ThreadID>::iterator thread_it = 657 std::find(activeThreads.begin(), activeThreads.end(), tid); 658 659 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 660 assert(!switchedOut()); 661 662 if (thread_it != activeThreads.end()) { 663 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 664 tid); 665 activeThreads.erase(thread_it); 666 } 667 668 fetch.deactivateThread(tid); 669 commit.deactivateThread(tid); 670} 671 672template <class Impl> 673Counter 674FullO3CPU<Impl>::totalInsts() const 675{ 676 Counter total(0); 677 678 ThreadID size = thread.size(); 679 for (ThreadID i = 0; i < size; i++) 680 total += thread[i]->numInst; 681 682 return total; 683} 684 685template <class Impl> 686Counter 687FullO3CPU<Impl>::totalOps() const 688{ 689 Counter total(0); 690 691 ThreadID size = thread.size(); 692 for (ThreadID i = 0; i < size; i++) 693 total += thread[i]->numOp; 694 695 return total; 696} 697 698template <class Impl> 699void 700FullO3CPU<Impl>::activateContext(ThreadID tid) 701{ 702 assert(!switchedOut()); 703 704 // Needs to set each stage to running as well. 705 activateThread(tid); 706 707 // We don't want to wake the CPU if it is drained. In that case, 708 // we just want to flag the thread as active and schedule the tick 709 // event from drainResume() instead. 710 if (getDrainState() == Drainable::Drained) 711 return; 712 713 // If we are time 0 or if the last activation time is in the past, 714 // schedule the next tick and wake up the fetch unit 715 if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) { 716 scheduleTickEvent(Cycles(0)); 717 718 // Be sure to signal that there's some activity so the CPU doesn't 719 // deschedule itself. 720 activityRec.activity(); 721 fetch.wakeFromQuiesce(); 722 723 Cycles cycles(curCycle() - lastRunningCycle); 724 // @todo: This is an oddity that is only here to match the stats 725 if (cycles != 0) 726 --cycles; 727 quiesceCycles += cycles; 728 729 lastActivatedCycle = curTick(); 730 731 _status = Running; 732 } 733} 734 735template <class Impl> 736void 737FullO3CPU<Impl>::suspendContext(ThreadID tid) 738{ 739 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 740 assert(!switchedOut()); 741 742 deactivateThread(tid); 743 744 // If this was the last thread then unschedule the tick event. 745 if (activeThreads.size() == 0) 746 unscheduleTickEvent(); 747 748 DPRINTF(Quiesce, "Suspending Context\n"); 749 lastRunningCycle = curCycle(); 750 _status = Idle; 751} 752 753template <class Impl> 754void 755FullO3CPU<Impl>::haltContext(ThreadID tid) 756{ 757 //For now, this is the same as deallocate 758 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 759 assert(!switchedOut()); 760 761 deactivateThread(tid); 762 removeThread(tid); 763} 764 765template <class Impl> 766void 767FullO3CPU<Impl>::insertThread(ThreadID tid) 768{ 769 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 770 // Will change now that the PC and thread state is internal to the CPU 771 // and not in the ThreadContext. 772 ThreadContext *src_tc; 773 if (FullSystem) 774 src_tc = system->threadContexts[tid]; 775 else 776 src_tc = tcBase(tid); 777 778 //Bind Int Regs to Rename Map 779 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 780 PhysRegIndex phys_reg = freeList.getIntReg(); 781 782 renameMap[tid].setEntry(ireg,phys_reg); 783 scoreboard.setReg(phys_reg); 784 } 785 786 //Bind Float Regs to Rename Map 787 int max_reg = TheISA::NumIntRegs + TheISA::NumFloatRegs; 788 for (int freg = TheISA::NumIntRegs; freg < max_reg; freg++) { 789 PhysRegIndex phys_reg = freeList.getFloatReg(); 790 791 renameMap[tid].setEntry(freg,phys_reg); 792 scoreboard.setReg(phys_reg); 793 } 794 795 //Bind condition-code Regs to Rename Map 796 max_reg = TheISA::NumIntRegs + TheISA::NumFloatRegs + TheISA::NumCCRegs; 797 for (int creg = TheISA::NumIntRegs + TheISA::NumFloatRegs; 798 creg < max_reg; creg++) { 799 PhysRegIndex phys_reg = freeList.getCCReg(); 800 801 renameMap[tid].setEntry(creg,phys_reg); 802 scoreboard.setReg(phys_reg); 803 } 804 805 //Copy Thread Data Into RegFile 806 //this->copyFromTC(tid); 807 808 //Set PC/NPC/NNPC 809 pcState(src_tc->pcState(), tid); 810 811 src_tc->setStatus(ThreadContext::Active); 812 813 activateContext(tid); 814 815 //Reset ROB/IQ/LSQ Entries 816 commit.rob->resetEntries(); 817 iew.resetEntries(); 818} 819 820template <class Impl> 821void 822FullO3CPU<Impl>::removeThread(ThreadID tid) 823{ 824 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 825 826 // Copy Thread Data From RegFile 827 // If thread is suspended, it might be re-allocated 828 // this->copyToTC(tid); 829 830 831 // @todo: 2-27-2008: Fix how we free up rename mappings 832 // here to alleviate the case for double-freeing registers 833 // in SMT workloads. 834 835 // Unbind Int Regs from Rename Map 836 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 837 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 838 scoreboard.unsetReg(phys_reg); 839 freeList.addReg(phys_reg); 840 } 841 842 // Unbind Float Regs from Rename Map 843 int max_reg = TheISA::FP_Reg_Base + TheISA::NumFloatRegs; 844 for (int freg = TheISA::FP_Reg_Base; freg < max_reg; freg++) { 845 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 846 scoreboard.unsetReg(phys_reg); 847 freeList.addReg(phys_reg); 848 } 849 850 // Unbind condition-code Regs from Rename Map 851 max_reg = TheISA::CC_Reg_Base + TheISA::NumCCRegs; 852 for (int creg = TheISA::CC_Reg_Base; creg < max_reg; creg++) { 853 PhysRegIndex phys_reg = renameMap[tid].lookup(creg); 854 scoreboard.unsetReg(phys_reg); 855 freeList.addReg(phys_reg); 856 } 857 858 // Squash Throughout Pipeline 859 DynInstPtr inst = commit.rob->readHeadInst(tid); 860 InstSeqNum squash_seq_num = inst->seqNum; 861 fetch.squash(0, squash_seq_num, inst, tid); 862 decode.squash(tid); 863 rename.squash(squash_seq_num, tid); 864 iew.squash(tid); 865 iew.ldstQueue.squash(squash_seq_num, tid); 866 commit.rob->squash(squash_seq_num, tid); 867 868 869 assert(iew.instQueue.getCount(tid) == 0); 870 assert(iew.ldstQueue.getCount(tid) == 0); 871 872 // Reset ROB/IQ/LSQ Entries 873 874 // Commented out for now. This should be possible to do by 875 // telling all the pipeline stages to drain first, and then 876 // checking until the drain completes. Once the pipeline is 877 // drained, call resetEntries(). - 10-09-06 ktlim 878/* 879 if (activeThreads.size() >= 1) { 880 commit.rob->resetEntries(); 881 iew.resetEntries(); 882 } 883*/ 884} 885 886template <class Impl> 887Fault 888FullO3CPU<Impl>::hwrei(ThreadID tid) 889{ 890#if THE_ISA == ALPHA_ISA 891 // Need to clear the lock flag upon returning from an interrupt. 892 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 893 894 this->thread[tid]->kernelStats->hwrei(); 895 896 // FIXME: XXX check for interrupts? XXX 897#endif 898 return NoFault; 899} 900 901template <class Impl> 902bool 903FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 904{ 905#if THE_ISA == ALPHA_ISA 906 if (this->thread[tid]->kernelStats) 907 this->thread[tid]->kernelStats->callpal(palFunc, 908 this->threadContexts[tid]); 909 910 switch (palFunc) { 911 case PAL::halt: 912 halt(); 913 if (--System::numSystemsRunning == 0) 914 exitSimLoop("all cpus halted"); 915 break; 916 917 case PAL::bpt: 918 case PAL::bugchk: 919 if (this->system->breakpoint()) 920 return false; 921 break; 922 } 923#endif 924 return true; 925} 926 927template <class Impl> 928Fault 929FullO3CPU<Impl>::getInterrupts() 930{ 931 // Check if there are any outstanding interrupts 932 return this->interrupts->getInterrupt(this->threadContexts[0]); 933} 934 935template <class Impl> 936void 937FullO3CPU<Impl>::processInterrupts(const Fault &interrupt) 938{ 939 // Check for interrupts here. For now can copy the code that 940 // exists within isa_fullsys_traits.hh. Also assume that thread 0 941 // is the one that handles the interrupts. 942 // @todo: Possibly consolidate the interrupt checking code. 943 // @todo: Allow other threads to handle interrupts. 944 945 assert(interrupt != NoFault); 946 this->interrupts->updateIntrInfo(this->threadContexts[0]); 947 948 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 949 this->trap(interrupt, 0, nullptr); 950} 951 952template <class Impl> 953void 954FullO3CPU<Impl>::trap(const Fault &fault, ThreadID tid, 955 const StaticInstPtr &inst) 956{ 957 // Pass the thread's TC into the invoke method. 958 fault->invoke(this->threadContexts[tid], inst); 959} 960 961template <class Impl> 962void 963FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid) 964{ 965 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 966 967 DPRINTF(Activity,"Activity: syscall() called.\n"); 968 969 // Temporarily increase this by one to account for the syscall 970 // instruction. 971 ++(this->thread[tid]->funcExeInst); 972 973 // Execute the actual syscall. 974 this->thread[tid]->syscall(callnum); 975 976 // Decrease funcExeInst by one as the normal commit will handle 977 // incrementing it. 978 --(this->thread[tid]->funcExeInst); 979} 980 981template <class Impl> 982void 983FullO3CPU<Impl>::serializeThread(std::ostream &os, ThreadID tid) 984{ 985 thread[tid]->serialize(os); 986} 987 988template <class Impl> 989void 990FullO3CPU<Impl>::unserializeThread(Checkpoint *cp, const std::string §ion, 991 ThreadID tid) 992{ 993 thread[tid]->unserialize(cp, section); 994} 995 996template <class Impl> 997unsigned int 998FullO3CPU<Impl>::drain(DrainManager *drain_manager) 999{ 1000 // If the CPU isn't doing anything, then return immediately. 1001 if (switchedOut()) { 1002 setDrainState(Drainable::Drained); 1003 return 0; 1004 } 1005 1006 DPRINTF(Drain, "Draining...\n"); 1007 setDrainState(Drainable::Draining); 1008 1009 // We only need to signal a drain to the commit stage as this 1010 // initiates squashing controls the draining. Once the commit 1011 // stage commits an instruction where it is safe to stop, it'll 1012 // squash the rest of the instructions in the pipeline and force 1013 // the fetch stage to stall. The pipeline will be drained once all 1014 // in-flight instructions have retired. 1015 commit.drain(); 1016 1017 // Wake the CPU and record activity so everything can drain out if 1018 // the CPU was not able to immediately drain. 1019 if (!isDrained()) { 1020 drainManager = drain_manager; 1021 1022 wakeCPU(); 1023 activityRec.activity(); 1024 1025 DPRINTF(Drain, "CPU not drained\n"); 1026 1027 return 1; 1028 } else { 1029 setDrainState(Drainable::Drained); 1030 DPRINTF(Drain, "CPU is already drained\n"); 1031 if (tickEvent.scheduled()) 1032 deschedule(tickEvent); 1033 1034 // Flush out any old data from the time buffers. In 1035 // particular, there might be some data in flight from the 1036 // fetch stage that isn't visible in any of the CPU buffers we 1037 // test in isDrained(). 1038 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1039 timeBuffer.advance(); 1040 fetchQueue.advance(); 1041 decodeQueue.advance(); 1042 renameQueue.advance(); 1043 iewQueue.advance(); 1044 } 1045 1046 drainSanityCheck(); 1047 return 0; 1048 } 1049} 1050 1051template <class Impl> 1052bool 1053FullO3CPU<Impl>::tryDrain() 1054{ 1055 if (!drainManager || !isDrained()) 1056 return false; 1057 1058 if (tickEvent.scheduled()) 1059 deschedule(tickEvent); 1060 1061 DPRINTF(Drain, "CPU done draining, processing drain event\n"); 1062 drainManager->signalDrainDone(); 1063 drainManager = NULL; 1064 1065 return true; 1066} 1067 1068template <class Impl> 1069void 1070FullO3CPU<Impl>::drainSanityCheck() const 1071{ 1072 assert(isDrained()); 1073 fetch.drainSanityCheck(); 1074 decode.drainSanityCheck(); 1075 rename.drainSanityCheck(); 1076 iew.drainSanityCheck(); 1077 commit.drainSanityCheck(); 1078} 1079 1080template <class Impl> 1081bool 1082FullO3CPU<Impl>::isDrained() const 1083{ 1084 bool drained(true); 1085 1086 if (!instList.empty() || !removeList.empty()) { 1087 DPRINTF(Drain, "Main CPU structures not drained.\n"); 1088 drained = false; 1089 } 1090 1091 if (!fetch.isDrained()) { 1092 DPRINTF(Drain, "Fetch not drained.\n"); 1093 drained = false; 1094 } 1095 1096 if (!decode.isDrained()) { 1097 DPRINTF(Drain, "Decode not drained.\n"); 1098 drained = false; 1099 } 1100 1101 if (!rename.isDrained()) { 1102 DPRINTF(Drain, "Rename not drained.\n"); 1103 drained = false; 1104 } 1105 1106 if (!iew.isDrained()) { 1107 DPRINTF(Drain, "IEW not drained.\n"); 1108 drained = false; 1109 } 1110 1111 if (!commit.isDrained()) { 1112 DPRINTF(Drain, "Commit not drained.\n"); 1113 drained = false; 1114 } 1115 1116 return drained; 1117} 1118 1119template <class Impl> 1120void 1121FullO3CPU<Impl>::commitDrained(ThreadID tid) 1122{ 1123 fetch.drainStall(tid); 1124} 1125 1126template <class Impl> 1127void 1128FullO3CPU<Impl>::drainResume() 1129{ 1130 setDrainState(Drainable::Running); 1131 if (switchedOut()) 1132 return; 1133 1134 DPRINTF(Drain, "Resuming...\n"); 1135 verifyMemoryMode(); 1136 1137 fetch.drainResume(); 1138 commit.drainResume(); 1139 1140 _status = Idle; 1141 for (ThreadID i = 0; i < thread.size(); i++) { 1142 if (thread[i]->status() == ThreadContext::Active) { 1143 DPRINTF(Drain, "Activating thread: %i\n", i); 1144 activateThread(i); 1145 _status = Running; 1146 } 1147 } 1148 1149 assert(!tickEvent.scheduled()); 1150 if (_status == Running) 1151 schedule(tickEvent, nextCycle()); 1152} 1153 1154template <class Impl> 1155void 1156FullO3CPU<Impl>::switchOut() 1157{ 1158 DPRINTF(O3CPU, "Switching out\n"); 1159 BaseCPU::switchOut(); 1160 1161 activityRec.reset(); 1162 1163 _status = SwitchedOut; 1164 1165 if (checker) 1166 checker->switchOut(); 1167} 1168 1169template <class Impl> 1170void 1171FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1172{ 1173 BaseCPU::takeOverFrom(oldCPU); 1174 1175 fetch.takeOverFrom(); 1176 decode.takeOverFrom(); 1177 rename.takeOverFrom(); 1178 iew.takeOverFrom(); 1179 commit.takeOverFrom(); 1180 1181 assert(!tickEvent.scheduled()); 1182 1183 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU); 1184 if (oldO3CPU) 1185 globalSeqNum = oldO3CPU->globalSeqNum; 1186 1187 lastRunningCycle = curCycle(); 1188 _status = Idle; 1189} 1190 1191template <class Impl> 1192void 1193FullO3CPU<Impl>::verifyMemoryMode() const 1194{ 1195 if (!system->isTimingMode()) { 1196 fatal("The O3 CPU requires the memory system to be in " 1197 "'timing' mode.\n"); 1198 } 1199} 1200 1201template <class Impl> 1202TheISA::MiscReg 1203FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) 1204{ 1205 return this->isa[tid]->readMiscRegNoEffect(misc_reg); 1206} 1207 1208template <class Impl> 1209TheISA::MiscReg 1210FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1211{ 1212 miscRegfileReads++; 1213 return this->isa[tid]->readMiscReg(misc_reg, tcBase(tid)); 1214} 1215 1216template <class Impl> 1217void 1218FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1219 const TheISA::MiscReg &val, ThreadID tid) 1220{ 1221 this->isa[tid]->setMiscRegNoEffect(misc_reg, val); 1222} 1223 1224template <class Impl> 1225void 1226FullO3CPU<Impl>::setMiscReg(int misc_reg, 1227 const TheISA::MiscReg &val, ThreadID tid) 1228{ 1229 miscRegfileWrites++; 1230 this->isa[tid]->setMiscReg(misc_reg, val, tcBase(tid)); 1231} 1232 1233template <class Impl> 1234uint64_t 1235FullO3CPU<Impl>::readIntReg(int reg_idx) 1236{ 1237 intRegfileReads++; 1238 return regFile.readIntReg(reg_idx); 1239} 1240 1241template <class Impl> 1242FloatReg 1243FullO3CPU<Impl>::readFloatReg(int reg_idx) 1244{ 1245 fpRegfileReads++; 1246 return regFile.readFloatReg(reg_idx); 1247} 1248 1249template <class Impl> 1250FloatRegBits 1251FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1252{ 1253 fpRegfileReads++; 1254 return regFile.readFloatRegBits(reg_idx); 1255} 1256 1257template <class Impl> 1258CCReg 1259FullO3CPU<Impl>::readCCReg(int reg_idx) 1260{ 1261 ccRegfileReads++; 1262 return regFile.readCCReg(reg_idx); 1263} 1264 1265template <class Impl> 1266void 1267FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1268{ 1269 intRegfileWrites++; 1270 regFile.setIntReg(reg_idx, val); 1271} 1272 1273template <class Impl> 1274void 1275FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1276{ 1277 fpRegfileWrites++; 1278 regFile.setFloatReg(reg_idx, val); 1279} 1280 1281template <class Impl> 1282void 1283FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1284{ 1285 fpRegfileWrites++; 1286 regFile.setFloatRegBits(reg_idx, val); 1287} 1288 1289template <class Impl> 1290void 1291FullO3CPU<Impl>::setCCReg(int reg_idx, CCReg val) 1292{ 1293 ccRegfileWrites++; 1294 regFile.setCCReg(reg_idx, val); 1295} 1296 1297template <class Impl> 1298uint64_t 1299FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1300{ 1301 intRegfileReads++; 1302 PhysRegIndex phys_reg = commitRenameMap[tid].lookupInt(reg_idx); 1303 1304 return regFile.readIntReg(phys_reg); 1305} 1306 1307template <class Impl> 1308float 1309FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1310{ 1311 fpRegfileReads++; 1312 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1313 1314 return regFile.readFloatReg(phys_reg); 1315} 1316 1317template <class Impl> 1318uint64_t 1319FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1320{ 1321 fpRegfileReads++; 1322 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1323 1324 return regFile.readFloatRegBits(phys_reg); 1325} 1326 1327template <class Impl> 1328CCReg 1329FullO3CPU<Impl>::readArchCCReg(int reg_idx, ThreadID tid) 1330{ 1331 ccRegfileReads++; 1332 PhysRegIndex phys_reg = commitRenameMap[tid].lookupCC(reg_idx); 1333 1334 return regFile.readCCReg(phys_reg); 1335} 1336 1337template <class Impl> 1338void 1339FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1340{ 1341 intRegfileWrites++; 1342 PhysRegIndex phys_reg = commitRenameMap[tid].lookupInt(reg_idx); 1343 1344 regFile.setIntReg(phys_reg, val); 1345} 1346 1347template <class Impl> 1348void 1349FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1350{ 1351 fpRegfileWrites++; 1352 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1353 1354 regFile.setFloatReg(phys_reg, val); 1355} 1356 1357template <class Impl> 1358void 1359FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1360{ 1361 fpRegfileWrites++; 1362 PhysRegIndex phys_reg = commitRenameMap[tid].lookupFloat(reg_idx); 1363 1364 regFile.setFloatRegBits(phys_reg, val); 1365} 1366 1367template <class Impl> 1368void 1369FullO3CPU<Impl>::setArchCCReg(int reg_idx, CCReg val, ThreadID tid) 1370{ 1371 ccRegfileWrites++; 1372 PhysRegIndex phys_reg = commitRenameMap[tid].lookupCC(reg_idx); 1373 1374 regFile.setCCReg(phys_reg, val); 1375} 1376 1377template <class Impl> 1378TheISA::PCState 1379FullO3CPU<Impl>::pcState(ThreadID tid) 1380{ 1381 return commit.pcState(tid); 1382} 1383 1384template <class Impl> 1385void 1386FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1387{ 1388 commit.pcState(val, tid); 1389} 1390 1391template <class Impl> 1392Addr 1393FullO3CPU<Impl>::instAddr(ThreadID tid) 1394{ 1395 return commit.instAddr(tid); 1396} 1397 1398template <class Impl> 1399Addr 1400FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1401{ 1402 return commit.nextInstAddr(tid); 1403} 1404 1405template <class Impl> 1406MicroPC 1407FullO3CPU<Impl>::microPC(ThreadID tid) 1408{ 1409 return commit.microPC(tid); 1410} 1411 1412template <class Impl> 1413void 1414FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1415{ 1416 this->thread[tid]->noSquashFromTC = true; 1417 this->commit.generateTCEvent(tid); 1418} 1419 1420template <class Impl> 1421typename FullO3CPU<Impl>::ListIt 1422FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1423{ 1424 instList.push_back(inst); 1425 1426 return --(instList.end()); 1427} 1428 1429template <class Impl> 1430void 1431FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst) 1432{ 1433 // Keep an instruction count. 1434 if (!inst->isMicroop() || inst->isLastMicroop()) { 1435 thread[tid]->numInst++; 1436 thread[tid]->numInsts++; 1437 committedInsts[tid]++; 1438 } 1439 thread[tid]->numOp++; 1440 thread[tid]->numOps++; 1441 committedOps[tid]++; 1442 1443 system->totalNumInsts++; 1444 // Check for instruction-count-based events. 1445 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1446 system->instEventQueue.serviceEvents(system->totalNumInsts); 1447 1448 probeInstCommit(inst->staticInst); 1449} 1450 1451template <class Impl> 1452void 1453FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1454{ 1455 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1456 "[sn:%lli]\n", 1457 inst->threadNumber, inst->pcState(), inst->seqNum); 1458 1459 removeInstsThisCycle = true; 1460 1461 // Remove the front instruction. 1462 removeList.push(inst->getInstListIt()); 1463} 1464 1465template <class Impl> 1466void 1467FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1468{ 1469 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1470 " list.\n", tid); 1471 1472 ListIt end_it; 1473 1474 bool rob_empty = false; 1475 1476 if (instList.empty()) { 1477 return; 1478 } else if (rob.isEmpty(tid)) { 1479 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1480 end_it = instList.begin(); 1481 rob_empty = true; 1482 } else { 1483 end_it = (rob.readTailInst(tid))->getInstListIt(); 1484 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1485 } 1486 1487 removeInstsThisCycle = true; 1488 1489 ListIt inst_it = instList.end(); 1490 1491 inst_it--; 1492 1493 // Walk through the instruction list, removing any instructions 1494 // that were inserted after the given instruction iterator, end_it. 1495 while (inst_it != end_it) { 1496 assert(!instList.empty()); 1497 1498 squashInstIt(inst_it, tid); 1499 1500 inst_it--; 1501 } 1502 1503 // If the ROB was empty, then we actually need to remove the first 1504 // instruction as well. 1505 if (rob_empty) { 1506 squashInstIt(inst_it, tid); 1507 } 1508} 1509 1510template <class Impl> 1511void 1512FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1513{ 1514 assert(!instList.empty()); 1515 1516 removeInstsThisCycle = true; 1517 1518 ListIt inst_iter = instList.end(); 1519 1520 inst_iter--; 1521 1522 DPRINTF(O3CPU, "Deleting instructions from instruction " 1523 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1524 tid, seq_num, (*inst_iter)->seqNum); 1525 1526 while ((*inst_iter)->seqNum > seq_num) { 1527 1528 bool break_loop = (inst_iter == instList.begin()); 1529 1530 squashInstIt(inst_iter, tid); 1531 1532 inst_iter--; 1533 1534 if (break_loop) 1535 break; 1536 } 1537} 1538 1539template <class Impl> 1540inline void 1541FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1542{ 1543 if ((*instIt)->threadNumber == tid) { 1544 DPRINTF(O3CPU, "Squashing instruction, " 1545 "[tid:%i] [sn:%lli] PC %s\n", 1546 (*instIt)->threadNumber, 1547 (*instIt)->seqNum, 1548 (*instIt)->pcState()); 1549 1550 // Mark it as squashed. 1551 (*instIt)->setSquashed(); 1552 1553 // @todo: Formulate a consistent method for deleting 1554 // instructions from the instruction list 1555 // Remove the instruction from the list. 1556 removeList.push(instIt); 1557 } 1558} 1559 1560template <class Impl> 1561void 1562FullO3CPU<Impl>::cleanUpRemovedInsts() 1563{ 1564 while (!removeList.empty()) { 1565 DPRINTF(O3CPU, "Removing instruction, " 1566 "[tid:%i] [sn:%lli] PC %s\n", 1567 (*removeList.front())->threadNumber, 1568 (*removeList.front())->seqNum, 1569 (*removeList.front())->pcState()); 1570 1571 instList.erase(removeList.front()); 1572 1573 removeList.pop(); 1574 } 1575 1576 removeInstsThisCycle = false; 1577} 1578/* 1579template <class Impl> 1580void 1581FullO3CPU<Impl>::removeAllInsts() 1582{ 1583 instList.clear(); 1584} 1585*/ 1586template <class Impl> 1587void 1588FullO3CPU<Impl>::dumpInsts() 1589{ 1590 int num = 0; 1591 1592 ListIt inst_list_it = instList.begin(); 1593 1594 cprintf("Dumping Instruction List\n"); 1595 1596 while (inst_list_it != instList.end()) { 1597 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1598 "Squashed:%i\n\n", 1599 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1600 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1601 (*inst_list_it)->isSquashed()); 1602 inst_list_it++; 1603 ++num; 1604 } 1605} 1606/* 1607template <class Impl> 1608void 1609FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1610{ 1611 iew.wakeDependents(inst); 1612} 1613*/ 1614template <class Impl> 1615void 1616FullO3CPU<Impl>::wakeCPU() 1617{ 1618 if (activityRec.active() || tickEvent.scheduled()) { 1619 DPRINTF(Activity, "CPU already running.\n"); 1620 return; 1621 } 1622 1623 DPRINTF(Activity, "Waking up CPU\n"); 1624 1625 Cycles cycles(curCycle() - lastRunningCycle); 1626 // @todo: This is an oddity that is only here to match the stats 1627 if (cycles > 1) { 1628 --cycles; 1629 idleCycles += cycles; 1630 numCycles += cycles; 1631 ppCycles->notify(cycles); 1632 } 1633 1634 schedule(tickEvent, clockEdge()); 1635} 1636 1637template <class Impl> 1638void 1639FullO3CPU<Impl>::wakeup() 1640{ 1641 if (this->thread[0]->status() != ThreadContext::Suspended) 1642 return; 1643 1644 this->wakeCPU(); 1645 1646 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1647 this->threadContexts[0]->activate(); 1648} 1649 1650template <class Impl> 1651ThreadID 1652FullO3CPU<Impl>::getFreeTid() 1653{ 1654 for (ThreadID tid = 0; tid < numThreads; tid++) { 1655 if (!tids[tid]) { 1656 tids[tid] = true; 1657 return tid; 1658 } 1659 } 1660 1661 return InvalidThreadID; 1662} 1663 1664template <class Impl> 1665void 1666FullO3CPU<Impl>::updateThreadPriority() 1667{ 1668 if (activeThreads.size() > 1) { 1669 //DEFAULT TO ROUND ROBIN SCHEME 1670 //e.g. Move highest priority to end of thread list 1671 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1672 1673 unsigned high_thread = *list_begin; 1674 1675 activeThreads.erase(list_begin); 1676 1677 activeThreads.push_back(high_thread); 1678 } 1679} 1680 1681// Forward declaration of FullO3CPU. 1682template class FullO3CPU<O3CPUImpl>; 1683