cpu.cc revision 9342
1/* 2 * Copyright (c) 2011-2012 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 "cpu/checker/cpu.hh" 49#include "cpu/checker/thread_context.hh" 50#include "cpu/o3/cpu.hh" 51#include "cpu/o3/isa_specific.hh" 52#include "cpu/o3/thread_context.hh" 53#include "cpu/activity.hh" 54#include "cpu/quiesce_event.hh" 55#include "cpu/simple_thread.hh" 56#include "cpu/thread_context.hh" 57#include "debug/Activity.hh" 58#include "debug/Drain.hh" 59#include "debug/O3CPU.hh" 60#include "debug/Quiesce.hh" 61#include "enums/MemoryMode.hh" 62#include "sim/core.hh" 63#include "sim/full_system.hh" 64#include "sim/process.hh" 65#include "sim/stat_control.hh" 66#include "sim/system.hh" 67 68#if THE_ISA == ALPHA_ISA 69#include "arch/alpha/osfpal.hh" 70#include "debug/Activity.hh" 71#endif 72 73struct BaseCPUParams; 74 75using namespace TheISA; 76using namespace std; 77 78BaseO3CPU::BaseO3CPU(BaseCPUParams *params) 79 : BaseCPU(params) 80{ 81} 82 83void 84BaseO3CPU::regStats() 85{ 86 BaseCPU::regStats(); 87} 88 89template<class Impl> 90bool 91FullO3CPU<Impl>::IcachePort::recvTimingResp(PacketPtr pkt) 92{ 93 DPRINTF(O3CPU, "Fetch unit received timing\n"); 94 // We shouldn't ever get a block in ownership state 95 assert(!(pkt->memInhibitAsserted() && !pkt->sharedAsserted())); 96 fetch->processCacheCompletion(pkt); 97 98 return true; 99} 100 101template<class Impl> 102void 103FullO3CPU<Impl>::IcachePort::recvRetry() 104{ 105 fetch->recvRetry(); 106} 107 108template <class Impl> 109bool 110FullO3CPU<Impl>::DcachePort::recvTimingResp(PacketPtr pkt) 111{ 112 return lsq->recvTimingResp(pkt); 113} 114 115template <class Impl> 116void 117FullO3CPU<Impl>::DcachePort::recvTimingSnoopReq(PacketPtr pkt) 118{ 119 lsq->recvTimingSnoopReq(pkt); 120} 121 122template <class Impl> 123void 124FullO3CPU<Impl>::DcachePort::recvRetry() 125{ 126 lsq->recvRetry(); 127} 128 129template <class Impl> 130FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c) 131 : Event(CPU_Tick_Pri), cpu(c) 132{ 133} 134 135template <class Impl> 136void 137FullO3CPU<Impl>::TickEvent::process() 138{ 139 cpu->tick(); 140} 141 142template <class Impl> 143const char * 144FullO3CPU<Impl>::TickEvent::description() const 145{ 146 return "FullO3CPU tick"; 147} 148 149template <class Impl> 150FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent() 151 : Event(CPU_Switch_Pri) 152{ 153} 154 155template <class Impl> 156void 157FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num, 158 FullO3CPU<Impl> *thread_cpu) 159{ 160 tid = thread_num; 161 cpu = thread_cpu; 162} 163 164template <class Impl> 165void 166FullO3CPU<Impl>::ActivateThreadEvent::process() 167{ 168 cpu->activateThread(tid); 169} 170 171template <class Impl> 172const char * 173FullO3CPU<Impl>::ActivateThreadEvent::description() const 174{ 175 return "FullO3CPU \"Activate Thread\""; 176} 177 178template <class Impl> 179FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent() 180 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL) 181{ 182} 183 184template <class Impl> 185void 186FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num, 187 FullO3CPU<Impl> *thread_cpu) 188{ 189 tid = thread_num; 190 cpu = thread_cpu; 191 remove = false; 192} 193 194template <class Impl> 195void 196FullO3CPU<Impl>::DeallocateContextEvent::process() 197{ 198 cpu->deactivateThread(tid); 199 if (remove) 200 cpu->removeThread(tid); 201} 202 203template <class Impl> 204const char * 205FullO3CPU<Impl>::DeallocateContextEvent::description() const 206{ 207 return "FullO3CPU \"Deallocate Context\""; 208} 209 210template <class Impl> 211FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params) 212 : BaseO3CPU(params), 213 itb(params->itb), 214 dtb(params->dtb), 215 tickEvent(this), 216#ifndef NDEBUG 217 instcount(0), 218#endif 219 removeInstsThisCycle(false), 220 fetch(this, params), 221 decode(this, params), 222 rename(this, params), 223 iew(this, params), 224 commit(this, params), 225 226 regFile(this, params->numPhysIntRegs, 227 params->numPhysFloatRegs), 228 229 freeList(params->numThreads, 230 TheISA::NumIntRegs, params->numPhysIntRegs, 231 TheISA::NumFloatRegs, params->numPhysFloatRegs), 232 233 rob(this, 234 params->numROBEntries, params->squashWidth, 235 params->smtROBPolicy, params->smtROBThreshold, 236 params->numThreads), 237 238 scoreboard(params->numThreads, 239 TheISA::NumIntRegs, params->numPhysIntRegs, 240 TheISA::NumFloatRegs, params->numPhysFloatRegs, 241 TheISA::NumMiscRegs * numThreads, 242 TheISA::ZeroReg), 243 244 icachePort(&fetch, this), 245 dcachePort(&iew.ldstQueue, this), 246 247 timeBuffer(params->backComSize, params->forwardComSize), 248 fetchQueue(params->backComSize, params->forwardComSize), 249 decodeQueue(params->backComSize, params->forwardComSize), 250 renameQueue(params->backComSize, params->forwardComSize), 251 iewQueue(params->backComSize, params->forwardComSize), 252 activityRec(name(), NumStages, 253 params->backComSize + params->forwardComSize, 254 params->activity), 255 256 globalSeqNum(1), 257 system(params->system), 258 drainCount(0), 259 deferRegistration(params->defer_registration), 260 lastRunningCycle(curCycle()) 261{ 262 if (!deferRegistration) { 263 _status = Running; 264 } else { 265 _status = SwitchedOut; 266 } 267 268 if (params->checker) { 269 BaseCPU *temp_checker = params->checker; 270 checker = dynamic_cast<Checker<Impl> *>(temp_checker); 271 checker->setIcachePort(&icachePort); 272 checker->setSystem(params->system); 273 } else { 274 checker = NULL; 275 } 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 lastActivatedCycle = 0; 391#if 0 392 // Give renameMap & rename stage access to the freeList; 393 for (ThreadID tid = 0; tid < numThreads; tid++) 394 globalSeqNum[tid] = 1; 395#endif 396 397 contextSwitch = false; 398 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 399 400 // Setup any thread state. 401 this->thread.resize(this->numThreads); 402 403 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 404 if (FullSystem) { 405 // SMT is not supported in FS mode yet. 406 assert(this->numThreads == 1); 407 this->thread[tid] = new Thread(this, 0, NULL); 408 } else { 409 if (tid < params->workload.size()) { 410 DPRINTF(O3CPU, "Workload[%i] process is %#x", 411 tid, this->thread[tid]); 412 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 413 (typename Impl::O3CPU *)(this), 414 tid, params->workload[tid]); 415 416 //usedTids[tid] = true; 417 //threadMap[tid] = tid; 418 } else { 419 //Allocate Empty thread so M5 can use later 420 //when scheduling threads to CPU 421 Process* dummy_proc = NULL; 422 423 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 424 (typename Impl::O3CPU *)(this), 425 tid, dummy_proc); 426 //usedTids[tid] = false; 427 } 428 } 429 430 ThreadContext *tc; 431 432 // Setup the TC that will serve as the interface to the threads/CPU. 433 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 434 435 tc = o3_tc; 436 437 // If we're using a checker, then the TC should be the 438 // CheckerThreadContext. 439 if (params->checker) { 440 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 441 o3_tc, this->checker); 442 } 443 444 o3_tc->cpu = (typename Impl::O3CPU *)(this); 445 assert(o3_tc->cpu); 446 o3_tc->thread = this->thread[tid]; 447 448 if (FullSystem) { 449 // Setup quiesce event. 450 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 451 } 452 // Give the thread the TC. 453 this->thread[tid]->tc = tc; 454 455 // Add the TC to the CPU's list of TC's. 456 this->threadContexts.push_back(tc); 457 } 458 459 // FullO3CPU always requires an interrupt controller. 460 if (!params->defer_registration && !interrupts) { 461 fatal("FullO3CPU %s has no interrupt controller.\n" 462 "Ensure createInterruptController() is called.\n", name()); 463 } 464 465 for (ThreadID tid = 0; tid < this->numThreads; tid++) 466 this->thread[tid]->setFuncExeInst(0); 467 468 lockAddr = 0; 469 lockFlag = false; 470} 471 472template <class Impl> 473FullO3CPU<Impl>::~FullO3CPU() 474{ 475} 476 477template <class Impl> 478void 479FullO3CPU<Impl>::regStats() 480{ 481 BaseO3CPU::regStats(); 482 483 // Register any of the O3CPU's stats here. 484 timesIdled 485 .name(name() + ".timesIdled") 486 .desc("Number of times that the entire CPU went into an idle state and" 487 " unscheduled itself") 488 .prereq(timesIdled); 489 490 idleCycles 491 .name(name() + ".idleCycles") 492 .desc("Total number of cycles that the CPU has spent unscheduled due " 493 "to idling") 494 .prereq(idleCycles); 495 496 quiesceCycles 497 .name(name() + ".quiesceCycles") 498 .desc("Total number of cycles that CPU has spent quiesced or waiting " 499 "for an interrupt") 500 .prereq(quiesceCycles); 501 502 // Number of Instructions simulated 503 // -------------------------------- 504 // Should probably be in Base CPU but need templated 505 // MaxThreads so put in here instead 506 committedInsts 507 .init(numThreads) 508 .name(name() + ".committedInsts") 509 .desc("Number of Instructions Simulated"); 510 511 committedOps 512 .init(numThreads) 513 .name(name() + ".committedOps") 514 .desc("Number of Ops (including micro ops) Simulated"); 515 516 totalCommittedInsts 517 .name(name() + ".committedInsts_total") 518 .desc("Number of Instructions Simulated"); 519 520 cpi 521 .name(name() + ".cpi") 522 .desc("CPI: Cycles Per Instruction") 523 .precision(6); 524 cpi = numCycles / committedInsts; 525 526 totalCpi 527 .name(name() + ".cpi_total") 528 .desc("CPI: Total CPI of All Threads") 529 .precision(6); 530 totalCpi = numCycles / totalCommittedInsts; 531 532 ipc 533 .name(name() + ".ipc") 534 .desc("IPC: Instructions Per Cycle") 535 .precision(6); 536 ipc = committedInsts / numCycles; 537 538 totalIpc 539 .name(name() + ".ipc_total") 540 .desc("IPC: Total IPC of All Threads") 541 .precision(6); 542 totalIpc = totalCommittedInsts / numCycles; 543 544 this->fetch.regStats(); 545 this->decode.regStats(); 546 this->rename.regStats(); 547 this->iew.regStats(); 548 this->commit.regStats(); 549 this->rob.regStats(); 550 551 intRegfileReads 552 .name(name() + ".int_regfile_reads") 553 .desc("number of integer regfile reads") 554 .prereq(intRegfileReads); 555 556 intRegfileWrites 557 .name(name() + ".int_regfile_writes") 558 .desc("number of integer regfile writes") 559 .prereq(intRegfileWrites); 560 561 fpRegfileReads 562 .name(name() + ".fp_regfile_reads") 563 .desc("number of floating regfile reads") 564 .prereq(fpRegfileReads); 565 566 fpRegfileWrites 567 .name(name() + ".fp_regfile_writes") 568 .desc("number of floating regfile writes") 569 .prereq(fpRegfileWrites); 570 571 miscRegfileReads 572 .name(name() + ".misc_regfile_reads") 573 .desc("number of misc regfile reads") 574 .prereq(miscRegfileReads); 575 576 miscRegfileWrites 577 .name(name() + ".misc_regfile_writes") 578 .desc("number of misc regfile writes") 579 .prereq(miscRegfileWrites); 580} 581 582template <class Impl> 583void 584FullO3CPU<Impl>::tick() 585{ 586 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 587 588 ++numCycles; 589 590// activity = false; 591 592 //Tick each of the stages 593 fetch.tick(); 594 595 decode.tick(); 596 597 rename.tick(); 598 599 iew.tick(); 600 601 commit.tick(); 602 603 if (!FullSystem) 604 doContextSwitch(); 605 606 // Now advance the time buffers 607 timeBuffer.advance(); 608 609 fetchQueue.advance(); 610 decodeQueue.advance(); 611 renameQueue.advance(); 612 iewQueue.advance(); 613 614 activityRec.advance(); 615 616 if (removeInstsThisCycle) { 617 cleanUpRemovedInsts(); 618 } 619 620 if (!tickEvent.scheduled()) { 621 if (_status == SwitchedOut || 622 getDrainState() == Drainable::Drained) { 623 DPRINTF(O3CPU, "Switched out!\n"); 624 // increment stat 625 lastRunningCycle = curCycle(); 626 } else if (!activityRec.active() || _status == Idle) { 627 DPRINTF(O3CPU, "Idle!\n"); 628 lastRunningCycle = curCycle(); 629 timesIdled++; 630 } else { 631 schedule(tickEvent, clockEdge(Cycles(1))); 632 DPRINTF(O3CPU, "Scheduling next tick!\n"); 633 } 634 } 635 636 if (!FullSystem) 637 updateThreadPriority(); 638} 639 640template <class Impl> 641void 642FullO3CPU<Impl>::init() 643{ 644 BaseCPU::init(); 645 646 for (ThreadID tid = 0; tid < numThreads; ++tid) { 647 // Set inSyscall so that the CPU doesn't squash when initially 648 // setting up registers. 649 thread[tid]->inSyscall = true; 650 // Initialise the ThreadContext's memory proxies 651 thread[tid]->initMemProxies(thread[tid]->getTC()); 652 } 653 654 // this CPU could still be unconnected if we are restoring from a 655 // checkpoint and this CPU is to be switched in, thus we can only 656 // do this here if the instruction port is actually connected, if 657 // not we have to do it as part of takeOverFrom 658 if (icachePort.isConnected()) 659 fetch.setIcache(); 660 661 if (FullSystem && !params()->defer_registration) { 662 for (ThreadID tid = 0; tid < numThreads; tid++) { 663 ThreadContext *src_tc = threadContexts[tid]; 664 TheISA::initCPU(src_tc, src_tc->contextId()); 665 } 666 } 667 668 // Clear inSyscall. 669 for (int tid = 0; tid < numThreads; ++tid) 670 thread[tid]->inSyscall = false; 671 672 // Initialize stages. 673 fetch.initStage(); 674 iew.initStage(); 675 rename.initStage(); 676 commit.initStage(); 677 678 commit.setThreads(thread); 679} 680 681template <class Impl> 682void 683FullO3CPU<Impl>::activateThread(ThreadID tid) 684{ 685 list<ThreadID>::iterator isActive = 686 std::find(activeThreads.begin(), activeThreads.end(), tid); 687 688 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 689 690 if (isActive == activeThreads.end()) { 691 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 692 tid); 693 694 activeThreads.push_back(tid); 695 } 696} 697 698template <class Impl> 699void 700FullO3CPU<Impl>::deactivateThread(ThreadID tid) 701{ 702 //Remove From Active List, if Active 703 list<ThreadID>::iterator thread_it = 704 std::find(activeThreads.begin(), activeThreads.end(), tid); 705 706 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 707 708 if (thread_it != activeThreads.end()) { 709 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 710 tid); 711 activeThreads.erase(thread_it); 712 } 713} 714 715template <class Impl> 716Counter 717FullO3CPU<Impl>::totalInsts() const 718{ 719 Counter total(0); 720 721 ThreadID size = thread.size(); 722 for (ThreadID i = 0; i < size; i++) 723 total += thread[i]->numInst; 724 725 return total; 726} 727 728template <class Impl> 729Counter 730FullO3CPU<Impl>::totalOps() const 731{ 732 Counter total(0); 733 734 ThreadID size = thread.size(); 735 for (ThreadID i = 0; i < size; i++) 736 total += thread[i]->numOp; 737 738 return total; 739} 740 741template <class Impl> 742void 743FullO3CPU<Impl>::activateContext(ThreadID tid, Cycles delay) 744{ 745 // Needs to set each stage to running as well. 746 if (delay){ 747 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 748 "on cycle %d\n", tid, clockEdge(delay)); 749 scheduleActivateThreadEvent(tid, delay); 750 } else { 751 activateThread(tid); 752 } 753 754 // If we are time 0 or if the last activation time is in the past, 755 // schedule the next tick and wake up the fetch unit 756 if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) { 757 scheduleTickEvent(delay); 758 759 // Be sure to signal that there's some activity so the CPU doesn't 760 // deschedule itself. 761 activityRec.activity(); 762 fetch.wakeFromQuiesce(); 763 764 Cycles cycles(curCycle() - lastRunningCycle); 765 // @todo: This is an oddity that is only here to match the stats 766 if (cycles != 0) 767 --cycles; 768 quiesceCycles += cycles; 769 770 lastActivatedCycle = curTick(); 771 772 _status = Running; 773 } 774} 775 776template <class Impl> 777bool 778FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove, 779 Cycles delay) 780{ 781 // Schedule removal of thread data from CPU 782 if (delay){ 783 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 784 "on tick %d\n", tid, clockEdge(delay)); 785 scheduleDeallocateContextEvent(tid, remove, delay); 786 return false; 787 } else { 788 deactivateThread(tid); 789 if (remove) 790 removeThread(tid); 791 return true; 792 } 793} 794 795template <class Impl> 796void 797FullO3CPU<Impl>::suspendContext(ThreadID tid) 798{ 799 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 800 bool deallocated = scheduleDeallocateContext(tid, false, Cycles(1)); 801 // If this was the last thread then unschedule the tick event. 802 if ((activeThreads.size() == 1 && !deallocated) || 803 activeThreads.size() == 0) 804 unscheduleTickEvent(); 805 806 DPRINTF(Quiesce, "Suspending Context\n"); 807 lastRunningCycle = curCycle(); 808 _status = Idle; 809} 810 811template <class Impl> 812void 813FullO3CPU<Impl>::haltContext(ThreadID tid) 814{ 815 //For now, this is the same as deallocate 816 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 817 scheduleDeallocateContext(tid, true, Cycles(1)); 818} 819 820template <class Impl> 821void 822FullO3CPU<Impl>::insertThread(ThreadID tid) 823{ 824 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 825 // Will change now that the PC and thread state is internal to the CPU 826 // and not in the ThreadContext. 827 ThreadContext *src_tc; 828 if (FullSystem) 829 src_tc = system->threadContexts[tid]; 830 else 831 src_tc = tcBase(tid); 832 833 //Bind Int Regs to Rename Map 834 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 835 PhysRegIndex phys_reg = freeList.getIntReg(); 836 837 renameMap[tid].setEntry(ireg,phys_reg); 838 scoreboard.setReg(phys_reg); 839 } 840 841 //Bind Float Regs to Rename Map 842 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 843 PhysRegIndex phys_reg = freeList.getFloatReg(); 844 845 renameMap[tid].setEntry(freg,phys_reg); 846 scoreboard.setReg(phys_reg); 847 } 848 849 //Copy Thread Data Into RegFile 850 //this->copyFromTC(tid); 851 852 //Set PC/NPC/NNPC 853 pcState(src_tc->pcState(), tid); 854 855 src_tc->setStatus(ThreadContext::Active); 856 857 activateContext(tid, Cycles(1)); 858 859 //Reset ROB/IQ/LSQ Entries 860 commit.rob->resetEntries(); 861 iew.resetEntries(); 862} 863 864template <class Impl> 865void 866FullO3CPU<Impl>::removeThread(ThreadID tid) 867{ 868 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 869 870 // Copy Thread Data From RegFile 871 // If thread is suspended, it might be re-allocated 872 // this->copyToTC(tid); 873 874 875 // @todo: 2-27-2008: Fix how we free up rename mappings 876 // here to alleviate the case for double-freeing registers 877 // in SMT workloads. 878 879 // Unbind Int Regs from Rename Map 880 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 881 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 882 883 scoreboard.unsetReg(phys_reg); 884 freeList.addReg(phys_reg); 885 } 886 887 // Unbind Float Regs from Rename Map 888 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) { 889 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 890 891 scoreboard.unsetReg(phys_reg); 892 freeList.addReg(phys_reg); 893 } 894 895 // Squash Throughout Pipeline 896 DynInstPtr inst = commit.rob->readHeadInst(tid); 897 InstSeqNum squash_seq_num = inst->seqNum; 898 fetch.squash(0, squash_seq_num, inst, tid); 899 decode.squash(tid); 900 rename.squash(squash_seq_num, tid); 901 iew.squash(tid); 902 iew.ldstQueue.squash(squash_seq_num, tid); 903 commit.rob->squash(squash_seq_num, tid); 904 905 906 assert(iew.instQueue.getCount(tid) == 0); 907 assert(iew.ldstQueue.getCount(tid) == 0); 908 909 // Reset ROB/IQ/LSQ Entries 910 911 // Commented out for now. This should be possible to do by 912 // telling all the pipeline stages to drain first, and then 913 // checking until the drain completes. Once the pipeline is 914 // drained, call resetEntries(). - 10-09-06 ktlim 915/* 916 if (activeThreads.size() >= 1) { 917 commit.rob->resetEntries(); 918 iew.resetEntries(); 919 } 920*/ 921} 922 923 924template <class Impl> 925void 926FullO3CPU<Impl>::activateWhenReady(ThreadID tid) 927{ 928 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 929 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 930 tid); 931 932 bool ready = true; 933 934 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 935 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 936 "Phys. Int. Regs.\n", 937 tid); 938 ready = false; 939 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 940 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 941 "Phys. Float. Regs.\n", 942 tid); 943 ready = false; 944 } else if (commit.rob->numFreeEntries() >= 945 commit.rob->entryAmount(activeThreads.size() + 1)) { 946 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 947 "ROB entries.\n", 948 tid); 949 ready = false; 950 } else if (iew.instQueue.numFreeEntries() >= 951 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 952 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 953 "IQ entries.\n", 954 tid); 955 ready = false; 956 } else if (iew.ldstQueue.numFreeEntries() >= 957 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 958 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 959 "LSQ entries.\n", 960 tid); 961 ready = false; 962 } 963 964 if (ready) { 965 insertThread(tid); 966 967 contextSwitch = false; 968 969 cpuWaitList.remove(tid); 970 } else { 971 suspendContext(tid); 972 973 //blocks fetch 974 contextSwitch = true; 975 976 //@todo: dont always add to waitlist 977 //do waitlist 978 cpuWaitList.push_back(tid); 979 } 980} 981 982template <class Impl> 983Fault 984FullO3CPU<Impl>::hwrei(ThreadID tid) 985{ 986#if THE_ISA == ALPHA_ISA 987 // Need to clear the lock flag upon returning from an interrupt. 988 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 989 990 this->thread[tid]->kernelStats->hwrei(); 991 992 // FIXME: XXX check for interrupts? XXX 993#endif 994 return NoFault; 995} 996 997template <class Impl> 998bool 999FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 1000{ 1001#if THE_ISA == ALPHA_ISA 1002 if (this->thread[tid]->kernelStats) 1003 this->thread[tid]->kernelStats->callpal(palFunc, 1004 this->threadContexts[tid]); 1005 1006 switch (palFunc) { 1007 case PAL::halt: 1008 halt(); 1009 if (--System::numSystemsRunning == 0) 1010 exitSimLoop("all cpus halted"); 1011 break; 1012 1013 case PAL::bpt: 1014 case PAL::bugchk: 1015 if (this->system->breakpoint()) 1016 return false; 1017 break; 1018 } 1019#endif 1020 return true; 1021} 1022 1023template <class Impl> 1024Fault 1025FullO3CPU<Impl>::getInterrupts() 1026{ 1027 // Check if there are any outstanding interrupts 1028 return this->interrupts->getInterrupt(this->threadContexts[0]); 1029} 1030 1031template <class Impl> 1032void 1033FullO3CPU<Impl>::processInterrupts(Fault interrupt) 1034{ 1035 // Check for interrupts here. For now can copy the code that 1036 // exists within isa_fullsys_traits.hh. Also assume that thread 0 1037 // is the one that handles the interrupts. 1038 // @todo: Possibly consolidate the interrupt checking code. 1039 // @todo: Allow other threads to handle interrupts. 1040 1041 assert(interrupt != NoFault); 1042 this->interrupts->updateIntrInfo(this->threadContexts[0]); 1043 1044 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1045 this->trap(interrupt, 0, NULL); 1046} 1047 1048template <class Impl> 1049void 1050FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst) 1051{ 1052 // Pass the thread's TC into the invoke method. 1053 fault->invoke(this->threadContexts[tid], inst); 1054} 1055 1056template <class Impl> 1057void 1058FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid) 1059{ 1060 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1061 1062 DPRINTF(Activity,"Activity: syscall() called.\n"); 1063 1064 // Temporarily increase this by one to account for the syscall 1065 // instruction. 1066 ++(this->thread[tid]->funcExeInst); 1067 1068 // Execute the actual syscall. 1069 this->thread[tid]->syscall(callnum); 1070 1071 // Decrease funcExeInst by one as the normal commit will handle 1072 // incrementing it. 1073 --(this->thread[tid]->funcExeInst); 1074} 1075 1076template <class Impl> 1077void 1078FullO3CPU<Impl>::serialize(std::ostream &os) 1079{ 1080 Drainable::State so_state(getDrainState()); 1081 SERIALIZE_ENUM(so_state); 1082 BaseCPU::serialize(os); 1083 nameOut(os, csprintf("%s.tickEvent", name())); 1084 tickEvent.serialize(os); 1085 1086 // Use SimpleThread's ability to checkpoint to make it easier to 1087 // write out the registers. Also make this static so it doesn't 1088 // get instantiated multiple times (causes a panic in statistics). 1089 static SimpleThread temp; 1090 1091 ThreadID size = thread.size(); 1092 for (ThreadID i = 0; i < size; i++) { 1093 nameOut(os, csprintf("%s.xc.%i", name(), i)); 1094 temp.copyTC(thread[i]->getTC()); 1095 temp.serialize(os); 1096 } 1097} 1098 1099template <class Impl> 1100void 1101FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 1102{ 1103 Drainable::State so_state; 1104 UNSERIALIZE_ENUM(so_state); 1105 BaseCPU::unserialize(cp, section); 1106 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 1107 1108 // Use SimpleThread's ability to checkpoint to make it easier to 1109 // read in the registers. Also make this static so it doesn't 1110 // get instantiated multiple times (causes a panic in statistics). 1111 static SimpleThread temp; 1112 1113 ThreadID size = thread.size(); 1114 for (ThreadID i = 0; i < size; i++) { 1115 temp.copyTC(thread[i]->getTC()); 1116 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 1117 thread[i]->getTC()->copyArchRegs(temp.getTC()); 1118 } 1119} 1120 1121template <class Impl> 1122unsigned int 1123FullO3CPU<Impl>::drain(DrainManager *drain_manager) 1124{ 1125 DPRINTF(O3CPU, "Switching out\n"); 1126 1127 // If the CPU isn't doing anything, then return immediately. 1128 if (_status == SwitchedOut) 1129 return 0; 1130 1131 drainCount = 0; 1132 fetch.drain(); 1133 decode.drain(); 1134 rename.drain(); 1135 iew.drain(); 1136 commit.drain(); 1137 1138 // Wake the CPU and record activity so everything can drain out if 1139 // the CPU was not able to immediately drain. 1140 if (getDrainState() != Drainable::Drained) { 1141 // A bit of a hack...set the drainManager after all the drain() 1142 // calls have been made, that way if all of the stages drain 1143 // immediately, the signalDrained() function knows not to call 1144 // process on the drain event. 1145 drainManager = drain_manager; 1146 1147 wakeCPU(); 1148 activityRec.activity(); 1149 1150 DPRINTF(Drain, "CPU not drained\n"); 1151 1152 return 1; 1153 } else { 1154 return 0; 1155 } 1156} 1157 1158template <class Impl> 1159void 1160FullO3CPU<Impl>::drainResume() 1161{ 1162 fetch.resume(); 1163 decode.resume(); 1164 rename.resume(); 1165 iew.resume(); 1166 commit.resume(); 1167 1168 setDrainState(Drainable::Running); 1169 1170 if (_status == SwitchedOut) 1171 return; 1172 1173 assert(system->getMemoryMode() == Enums::timing); 1174 1175 if (!tickEvent.scheduled()) 1176 schedule(tickEvent, nextCycle()); 1177 _status = Running; 1178} 1179 1180template <class Impl> 1181void 1182FullO3CPU<Impl>::signalDrained() 1183{ 1184 if (++drainCount == NumStages) { 1185 if (tickEvent.scheduled()) 1186 tickEvent.squash(); 1187 1188 setDrainState(Drainable::Drained); 1189 1190 BaseCPU::switchOut(); 1191 1192 if (drainManager) { 1193 DPRINTF(Drain, "CPU done draining, processing drain event\n"); 1194 drainManager->signalDrainDone(); 1195 drainManager = NULL; 1196 } 1197 } 1198 assert(drainCount <= 5); 1199} 1200 1201template <class Impl> 1202void 1203FullO3CPU<Impl>::switchOut() 1204{ 1205 fetch.switchOut(); 1206 rename.switchOut(); 1207 iew.switchOut(); 1208 commit.switchOut(); 1209 instList.clear(); 1210 while (!removeList.empty()) { 1211 removeList.pop(); 1212 } 1213 1214 _status = SwitchedOut; 1215 1216 if (checker) 1217 checker->switchOut(); 1218 1219 if (tickEvent.scheduled()) 1220 tickEvent.squash(); 1221} 1222 1223template <class Impl> 1224void 1225FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1226{ 1227 // Flush out any old data from the time buffers. 1228 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1229 timeBuffer.advance(); 1230 fetchQueue.advance(); 1231 decodeQueue.advance(); 1232 renameQueue.advance(); 1233 iewQueue.advance(); 1234 } 1235 1236 activityRec.reset(); 1237 1238 BaseCPU::takeOverFrom(oldCPU); 1239 1240 fetch.takeOverFrom(); 1241 decode.takeOverFrom(); 1242 rename.takeOverFrom(); 1243 iew.takeOverFrom(); 1244 commit.takeOverFrom(); 1245 1246 assert(!tickEvent.scheduled() || tickEvent.squashed()); 1247 1248 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU); 1249 if (oldO3CPU) 1250 globalSeqNum = oldO3CPU->globalSeqNum; 1251 1252 // @todo: Figure out how to properly select the tid to put onto 1253 // the active threads list. 1254 ThreadID tid = 0; 1255 1256 list<ThreadID>::iterator isActive = 1257 std::find(activeThreads.begin(), activeThreads.end(), tid); 1258 1259 if (isActive == activeThreads.end()) { 1260 //May Need to Re-code this if the delay variable is the delay 1261 //needed for thread to activate 1262 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 1263 tid); 1264 1265 activeThreads.push_back(tid); 1266 } 1267 1268 // Set all statuses to active, schedule the CPU's tick event. 1269 // @todo: Fix up statuses so this is handled properly 1270 ThreadID size = threadContexts.size(); 1271 for (ThreadID i = 0; i < size; ++i) { 1272 ThreadContext *tc = threadContexts[i]; 1273 if (tc->status() == ThreadContext::Active && _status != Running) { 1274 _status = Running; 1275 reschedule(tickEvent, nextCycle(), true); 1276 } 1277 } 1278 if (!tickEvent.scheduled()) 1279 schedule(tickEvent, nextCycle()); 1280 1281 lastRunningCycle = curCycle(); 1282} 1283 1284template <class Impl> 1285TheISA::MiscReg 1286FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) 1287{ 1288 return this->isa[tid].readMiscRegNoEffect(misc_reg); 1289} 1290 1291template <class Impl> 1292TheISA::MiscReg 1293FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1294{ 1295 miscRegfileReads++; 1296 return this->isa[tid].readMiscReg(misc_reg, tcBase(tid)); 1297} 1298 1299template <class Impl> 1300void 1301FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1302 const TheISA::MiscReg &val, ThreadID tid) 1303{ 1304 this->isa[tid].setMiscRegNoEffect(misc_reg, val); 1305} 1306 1307template <class Impl> 1308void 1309FullO3CPU<Impl>::setMiscReg(int misc_reg, 1310 const TheISA::MiscReg &val, ThreadID tid) 1311{ 1312 miscRegfileWrites++; 1313 this->isa[tid].setMiscReg(misc_reg, val, tcBase(tid)); 1314} 1315 1316template <class Impl> 1317uint64_t 1318FullO3CPU<Impl>::readIntReg(int reg_idx) 1319{ 1320 intRegfileReads++; 1321 return regFile.readIntReg(reg_idx); 1322} 1323 1324template <class Impl> 1325FloatReg 1326FullO3CPU<Impl>::readFloatReg(int reg_idx) 1327{ 1328 fpRegfileReads++; 1329 return regFile.readFloatReg(reg_idx); 1330} 1331 1332template <class Impl> 1333FloatRegBits 1334FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1335{ 1336 fpRegfileReads++; 1337 return regFile.readFloatRegBits(reg_idx); 1338} 1339 1340template <class Impl> 1341void 1342FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1343{ 1344 intRegfileWrites++; 1345 regFile.setIntReg(reg_idx, val); 1346} 1347 1348template <class Impl> 1349void 1350FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1351{ 1352 fpRegfileWrites++; 1353 regFile.setFloatReg(reg_idx, val); 1354} 1355 1356template <class Impl> 1357void 1358FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1359{ 1360 fpRegfileWrites++; 1361 regFile.setFloatRegBits(reg_idx, val); 1362} 1363 1364template <class Impl> 1365uint64_t 1366FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1367{ 1368 intRegfileReads++; 1369 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1370 1371 return regFile.readIntReg(phys_reg); 1372} 1373 1374template <class Impl> 1375float 1376FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1377{ 1378 fpRegfileReads++; 1379 int idx = reg_idx + TheISA::NumIntRegs; 1380 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1381 1382 return regFile.readFloatReg(phys_reg); 1383} 1384 1385template <class Impl> 1386uint64_t 1387FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1388{ 1389 fpRegfileReads++; 1390 int idx = reg_idx + TheISA::NumIntRegs; 1391 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1392 1393 return regFile.readFloatRegBits(phys_reg); 1394} 1395 1396template <class Impl> 1397void 1398FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1399{ 1400 intRegfileWrites++; 1401 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1402 1403 regFile.setIntReg(phys_reg, val); 1404} 1405 1406template <class Impl> 1407void 1408FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1409{ 1410 fpRegfileWrites++; 1411 int idx = reg_idx + TheISA::NumIntRegs; 1412 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1413 1414 regFile.setFloatReg(phys_reg, val); 1415} 1416 1417template <class Impl> 1418void 1419FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1420{ 1421 fpRegfileWrites++; 1422 int idx = reg_idx + TheISA::NumIntRegs; 1423 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1424 1425 regFile.setFloatRegBits(phys_reg, val); 1426} 1427 1428template <class Impl> 1429TheISA::PCState 1430FullO3CPU<Impl>::pcState(ThreadID tid) 1431{ 1432 return commit.pcState(tid); 1433} 1434 1435template <class Impl> 1436void 1437FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1438{ 1439 commit.pcState(val, tid); 1440} 1441 1442template <class Impl> 1443Addr 1444FullO3CPU<Impl>::instAddr(ThreadID tid) 1445{ 1446 return commit.instAddr(tid); 1447} 1448 1449template <class Impl> 1450Addr 1451FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1452{ 1453 return commit.nextInstAddr(tid); 1454} 1455 1456template <class Impl> 1457MicroPC 1458FullO3CPU<Impl>::microPC(ThreadID tid) 1459{ 1460 return commit.microPC(tid); 1461} 1462 1463template <class Impl> 1464void 1465FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1466{ 1467 this->thread[tid]->inSyscall = true; 1468 this->commit.generateTCEvent(tid); 1469} 1470 1471template <class Impl> 1472typename FullO3CPU<Impl>::ListIt 1473FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1474{ 1475 instList.push_back(inst); 1476 1477 return --(instList.end()); 1478} 1479 1480template <class Impl> 1481void 1482FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst) 1483{ 1484 // Keep an instruction count. 1485 if (!inst->isMicroop() || inst->isLastMicroop()) { 1486 thread[tid]->numInst++; 1487 thread[tid]->numInsts++; 1488 committedInsts[tid]++; 1489 totalCommittedInsts++; 1490 } 1491 thread[tid]->numOp++; 1492 thread[tid]->numOps++; 1493 committedOps[tid]++; 1494 1495 system->totalNumInsts++; 1496 // Check for instruction-count-based events. 1497 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1498 system->instEventQueue.serviceEvents(system->totalNumInsts); 1499} 1500 1501template <class Impl> 1502void 1503FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1504{ 1505 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1506 "[sn:%lli]\n", 1507 inst->threadNumber, inst->pcState(), inst->seqNum); 1508 1509 removeInstsThisCycle = true; 1510 1511 // Remove the front instruction. 1512 removeList.push(inst->getInstListIt()); 1513} 1514 1515template <class Impl> 1516void 1517FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1518{ 1519 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1520 " list.\n", tid); 1521 1522 ListIt end_it; 1523 1524 bool rob_empty = false; 1525 1526 if (instList.empty()) { 1527 return; 1528 } else if (rob.isEmpty(/*tid*/)) { 1529 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1530 end_it = instList.begin(); 1531 rob_empty = true; 1532 } else { 1533 end_it = (rob.readTailInst(tid))->getInstListIt(); 1534 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1535 } 1536 1537 removeInstsThisCycle = true; 1538 1539 ListIt inst_it = instList.end(); 1540 1541 inst_it--; 1542 1543 // Walk through the instruction list, removing any instructions 1544 // that were inserted after the given instruction iterator, end_it. 1545 while (inst_it != end_it) { 1546 assert(!instList.empty()); 1547 1548 squashInstIt(inst_it, tid); 1549 1550 inst_it--; 1551 } 1552 1553 // If the ROB was empty, then we actually need to remove the first 1554 // instruction as well. 1555 if (rob_empty) { 1556 squashInstIt(inst_it, tid); 1557 } 1558} 1559 1560template <class Impl> 1561void 1562FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1563{ 1564 assert(!instList.empty()); 1565 1566 removeInstsThisCycle = true; 1567 1568 ListIt inst_iter = instList.end(); 1569 1570 inst_iter--; 1571 1572 DPRINTF(O3CPU, "Deleting instructions from instruction " 1573 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1574 tid, seq_num, (*inst_iter)->seqNum); 1575 1576 while ((*inst_iter)->seqNum > seq_num) { 1577 1578 bool break_loop = (inst_iter == instList.begin()); 1579 1580 squashInstIt(inst_iter, tid); 1581 1582 inst_iter--; 1583 1584 if (break_loop) 1585 break; 1586 } 1587} 1588 1589template <class Impl> 1590inline void 1591FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1592{ 1593 if ((*instIt)->threadNumber == tid) { 1594 DPRINTF(O3CPU, "Squashing instruction, " 1595 "[tid:%i] [sn:%lli] PC %s\n", 1596 (*instIt)->threadNumber, 1597 (*instIt)->seqNum, 1598 (*instIt)->pcState()); 1599 1600 // Mark it as squashed. 1601 (*instIt)->setSquashed(); 1602 1603 // @todo: Formulate a consistent method for deleting 1604 // instructions from the instruction list 1605 // Remove the instruction from the list. 1606 removeList.push(instIt); 1607 } 1608} 1609 1610template <class Impl> 1611void 1612FullO3CPU<Impl>::cleanUpRemovedInsts() 1613{ 1614 while (!removeList.empty()) { 1615 DPRINTF(O3CPU, "Removing instruction, " 1616 "[tid:%i] [sn:%lli] PC %s\n", 1617 (*removeList.front())->threadNumber, 1618 (*removeList.front())->seqNum, 1619 (*removeList.front())->pcState()); 1620 1621 instList.erase(removeList.front()); 1622 1623 removeList.pop(); 1624 } 1625 1626 removeInstsThisCycle = false; 1627} 1628/* 1629template <class Impl> 1630void 1631FullO3CPU<Impl>::removeAllInsts() 1632{ 1633 instList.clear(); 1634} 1635*/ 1636template <class Impl> 1637void 1638FullO3CPU<Impl>::dumpInsts() 1639{ 1640 int num = 0; 1641 1642 ListIt inst_list_it = instList.begin(); 1643 1644 cprintf("Dumping Instruction List\n"); 1645 1646 while (inst_list_it != instList.end()) { 1647 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1648 "Squashed:%i\n\n", 1649 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1650 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1651 (*inst_list_it)->isSquashed()); 1652 inst_list_it++; 1653 ++num; 1654 } 1655} 1656/* 1657template <class Impl> 1658void 1659FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1660{ 1661 iew.wakeDependents(inst); 1662} 1663*/ 1664template <class Impl> 1665void 1666FullO3CPU<Impl>::wakeCPU() 1667{ 1668 if (activityRec.active() || tickEvent.scheduled()) { 1669 DPRINTF(Activity, "CPU already running.\n"); 1670 return; 1671 } 1672 1673 DPRINTF(Activity, "Waking up CPU\n"); 1674 1675 Cycles cycles(curCycle() - lastRunningCycle); 1676 // @todo: This is an oddity that is only here to match the stats 1677 if (cycles != 0) 1678 --cycles; 1679 idleCycles += cycles; 1680 numCycles += cycles; 1681 1682 schedule(tickEvent, nextCycle()); 1683} 1684 1685template <class Impl> 1686void 1687FullO3CPU<Impl>::wakeup() 1688{ 1689 if (this->thread[0]->status() != ThreadContext::Suspended) 1690 return; 1691 1692 this->wakeCPU(); 1693 1694 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1695 this->threadContexts[0]->activate(); 1696} 1697 1698template <class Impl> 1699ThreadID 1700FullO3CPU<Impl>::getFreeTid() 1701{ 1702 for (ThreadID tid = 0; tid < numThreads; tid++) { 1703 if (!tids[tid]) { 1704 tids[tid] = true; 1705 return tid; 1706 } 1707 } 1708 1709 return InvalidThreadID; 1710} 1711 1712template <class Impl> 1713void 1714FullO3CPU<Impl>::doContextSwitch() 1715{ 1716 if (contextSwitch) { 1717 1718 //ADD CODE TO DEACTIVE THREAD HERE (???) 1719 1720 ThreadID size = cpuWaitList.size(); 1721 for (ThreadID tid = 0; tid < size; tid++) { 1722 activateWhenReady(tid); 1723 } 1724 1725 if (cpuWaitList.size() == 0) 1726 contextSwitch = true; 1727 } 1728} 1729 1730template <class Impl> 1731void 1732FullO3CPU<Impl>::updateThreadPriority() 1733{ 1734 if (activeThreads.size() > 1) { 1735 //DEFAULT TO ROUND ROBIN SCHEME 1736 //e.g. Move highest priority to end of thread list 1737 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1738 1739 unsigned high_thread = *list_begin; 1740 1741 activeThreads.erase(list_begin); 1742 1743 activeThreads.push_back(high_thread); 1744 } 1745} 1746 1747// Forward declaration of FullO3CPU. 1748template class FullO3CPU<O3CPUImpl>; 1749