cpu.cc revision 9384
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 isa(numThreads, NULL), 245 246 icachePort(&fetch, this), 247 dcachePort(&iew.ldstQueue, this), 248 249 timeBuffer(params->backComSize, params->forwardComSize), 250 fetchQueue(params->backComSize, params->forwardComSize), 251 decodeQueue(params->backComSize, params->forwardComSize), 252 renameQueue(params->backComSize, params->forwardComSize), 253 iewQueue(params->backComSize, params->forwardComSize), 254 activityRec(name(), NumStages, 255 params->backComSize + params->forwardComSize, 256 params->activity), 257 258 globalSeqNum(1), 259 system(params->system), 260 drainCount(0), 261 deferRegistration(params->defer_registration), 262 lastRunningCycle(curCycle()) 263{ 264 if (!deferRegistration) { 265 _status = Running; 266 } else { 267 _status = SwitchedOut; 268 } 269 270 if (params->checker) { 271 BaseCPU *temp_checker = params->checker; 272 checker = dynamic_cast<Checker<Impl> *>(temp_checker); 273 checker->setIcachePort(&icachePort); 274 checker->setSystem(params->system); 275 } else { 276 checker = NULL; 277 } 278 279 if (!FullSystem) { 280 thread.resize(numThreads); 281 tids.resize(numThreads); 282 } 283 284 // The stages also need their CPU pointer setup. However this 285 // must be done at the upper level CPU because they have pointers 286 // to the upper level CPU, and not this FullO3CPU. 287 288 // Set up Pointers to the activeThreads list for each stage 289 fetch.setActiveThreads(&activeThreads); 290 decode.setActiveThreads(&activeThreads); 291 rename.setActiveThreads(&activeThreads); 292 iew.setActiveThreads(&activeThreads); 293 commit.setActiveThreads(&activeThreads); 294 295 // Give each of the stages the time buffer they will use. 296 fetch.setTimeBuffer(&timeBuffer); 297 decode.setTimeBuffer(&timeBuffer); 298 rename.setTimeBuffer(&timeBuffer); 299 iew.setTimeBuffer(&timeBuffer); 300 commit.setTimeBuffer(&timeBuffer); 301 302 // Also setup each of the stages' queues. 303 fetch.setFetchQueue(&fetchQueue); 304 decode.setFetchQueue(&fetchQueue); 305 commit.setFetchQueue(&fetchQueue); 306 decode.setDecodeQueue(&decodeQueue); 307 rename.setDecodeQueue(&decodeQueue); 308 rename.setRenameQueue(&renameQueue); 309 iew.setRenameQueue(&renameQueue); 310 iew.setIEWQueue(&iewQueue); 311 commit.setIEWQueue(&iewQueue); 312 commit.setRenameQueue(&renameQueue); 313 314 commit.setIEWStage(&iew); 315 rename.setIEWStage(&iew); 316 rename.setCommitStage(&commit); 317 318 ThreadID active_threads; 319 if (FullSystem) { 320 active_threads = 1; 321 } else { 322 active_threads = params->workload.size(); 323 324 if (active_threads > Impl::MaxThreads) { 325 panic("Workload Size too large. Increase the 'MaxThreads' " 326 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) " 327 "or edit your workload size."); 328 } 329 } 330 331 //Make Sure That this a Valid Architeture 332 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs); 333 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs); 334 335 rename.setScoreboard(&scoreboard); 336 iew.setScoreboard(&scoreboard); 337 338 // Setup the rename map for whichever stages need it. 339 PhysRegIndex lreg_idx = 0; 340 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs 341 342 for (ThreadID tid = 0; tid < numThreads; tid++) { 343 bool bindRegs = (tid <= active_threads - 1); 344 345 isa[tid] = params->isa[tid]; 346 347 commitRenameMap[tid].init(TheISA::NumIntRegs, 348 params->numPhysIntRegs, 349 lreg_idx, //Index for Logical. Regs 350 351 TheISA::NumFloatRegs, 352 params->numPhysFloatRegs, 353 freg_idx, //Index for Float Regs 354 355 TheISA::NumMiscRegs, 356 357 TheISA::ZeroReg, 358 TheISA::ZeroReg, 359 360 tid, 361 false); 362 363 renameMap[tid].init(TheISA::NumIntRegs, 364 params->numPhysIntRegs, 365 lreg_idx, //Index for Logical. Regs 366 367 TheISA::NumFloatRegs, 368 params->numPhysFloatRegs, 369 freg_idx, //Index for Float Regs 370 371 TheISA::NumMiscRegs, 372 373 TheISA::ZeroReg, 374 TheISA::ZeroReg, 375 376 tid, 377 bindRegs); 378 379 activateThreadEvent[tid].init(tid, this); 380 deallocateContextEvent[tid].init(tid, this); 381 } 382 383 rename.setRenameMap(renameMap); 384 commit.setRenameMap(commitRenameMap); 385 386 // Give renameMap & rename stage access to the freeList; 387 for (ThreadID tid = 0; tid < numThreads; tid++) 388 renameMap[tid].setFreeList(&freeList); 389 rename.setFreeList(&freeList); 390 391 // Setup the ROB for whichever stages need it. 392 commit.setROB(&rob); 393 394 lastActivatedCycle = 0; 395#if 0 396 // Give renameMap & rename stage access to the freeList; 397 for (ThreadID tid = 0; tid < numThreads; tid++) 398 globalSeqNum[tid] = 1; 399#endif 400 401 contextSwitch = false; 402 DPRINTF(O3CPU, "Creating O3CPU object.\n"); 403 404 // Setup any thread state. 405 this->thread.resize(this->numThreads); 406 407 for (ThreadID tid = 0; tid < this->numThreads; ++tid) { 408 if (FullSystem) { 409 // SMT is not supported in FS mode yet. 410 assert(this->numThreads == 1); 411 this->thread[tid] = new Thread(this, 0, NULL); 412 } else { 413 if (tid < params->workload.size()) { 414 DPRINTF(O3CPU, "Workload[%i] process is %#x", 415 tid, this->thread[tid]); 416 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 417 (typename Impl::O3CPU *)(this), 418 tid, params->workload[tid]); 419 420 //usedTids[tid] = true; 421 //threadMap[tid] = tid; 422 } else { 423 //Allocate Empty thread so M5 can use later 424 //when scheduling threads to CPU 425 Process* dummy_proc = NULL; 426 427 this->thread[tid] = new typename FullO3CPU<Impl>::Thread( 428 (typename Impl::O3CPU *)(this), 429 tid, dummy_proc); 430 //usedTids[tid] = false; 431 } 432 } 433 434 ThreadContext *tc; 435 436 // Setup the TC that will serve as the interface to the threads/CPU. 437 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>; 438 439 tc = o3_tc; 440 441 // If we're using a checker, then the TC should be the 442 // CheckerThreadContext. 443 if (params->checker) { 444 tc = new CheckerThreadContext<O3ThreadContext<Impl> >( 445 o3_tc, this->checker); 446 } 447 448 o3_tc->cpu = (typename Impl::O3CPU *)(this); 449 assert(o3_tc->cpu); 450 o3_tc->thread = this->thread[tid]; 451 452 if (FullSystem) { 453 // Setup quiesce event. 454 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc); 455 } 456 // Give the thread the TC. 457 this->thread[tid]->tc = tc; 458 459 // Add the TC to the CPU's list of TC's. 460 this->threadContexts.push_back(tc); 461 } 462 463 // FullO3CPU always requires an interrupt controller. 464 if (!params->defer_registration && !interrupts) { 465 fatal("FullO3CPU %s has no interrupt controller.\n" 466 "Ensure createInterruptController() is called.\n", name()); 467 } 468 469 for (ThreadID tid = 0; tid < this->numThreads; tid++) 470 this->thread[tid]->setFuncExeInst(0); 471 472 lockAddr = 0; 473 lockFlag = false; 474} 475 476template <class Impl> 477FullO3CPU<Impl>::~FullO3CPU() 478{ 479} 480 481template <class Impl> 482void 483FullO3CPU<Impl>::regStats() 484{ 485 BaseO3CPU::regStats(); 486 487 // Register any of the O3CPU's stats here. 488 timesIdled 489 .name(name() + ".timesIdled") 490 .desc("Number of times that the entire CPU went into an idle state and" 491 " unscheduled itself") 492 .prereq(timesIdled); 493 494 idleCycles 495 .name(name() + ".idleCycles") 496 .desc("Total number of cycles that the CPU has spent unscheduled due " 497 "to idling") 498 .prereq(idleCycles); 499 500 quiesceCycles 501 .name(name() + ".quiesceCycles") 502 .desc("Total number of cycles that CPU has spent quiesced or waiting " 503 "for an interrupt") 504 .prereq(quiesceCycles); 505 506 // Number of Instructions simulated 507 // -------------------------------- 508 // Should probably be in Base CPU but need templated 509 // MaxThreads so put in here instead 510 committedInsts 511 .init(numThreads) 512 .name(name() + ".committedInsts") 513 .desc("Number of Instructions Simulated"); 514 515 committedOps 516 .init(numThreads) 517 .name(name() + ".committedOps") 518 .desc("Number of Ops (including micro ops) Simulated"); 519 520 totalCommittedInsts 521 .name(name() + ".committedInsts_total") 522 .desc("Number of Instructions Simulated"); 523 524 cpi 525 .name(name() + ".cpi") 526 .desc("CPI: Cycles Per Instruction") 527 .precision(6); 528 cpi = numCycles / committedInsts; 529 530 totalCpi 531 .name(name() + ".cpi_total") 532 .desc("CPI: Total CPI of All Threads") 533 .precision(6); 534 totalCpi = numCycles / totalCommittedInsts; 535 536 ipc 537 .name(name() + ".ipc") 538 .desc("IPC: Instructions Per Cycle") 539 .precision(6); 540 ipc = committedInsts / numCycles; 541 542 totalIpc 543 .name(name() + ".ipc_total") 544 .desc("IPC: Total IPC of All Threads") 545 .precision(6); 546 totalIpc = totalCommittedInsts / numCycles; 547 548 this->fetch.regStats(); 549 this->decode.regStats(); 550 this->rename.regStats(); 551 this->iew.regStats(); 552 this->commit.regStats(); 553 this->rob.regStats(); 554 555 intRegfileReads 556 .name(name() + ".int_regfile_reads") 557 .desc("number of integer regfile reads") 558 .prereq(intRegfileReads); 559 560 intRegfileWrites 561 .name(name() + ".int_regfile_writes") 562 .desc("number of integer regfile writes") 563 .prereq(intRegfileWrites); 564 565 fpRegfileReads 566 .name(name() + ".fp_regfile_reads") 567 .desc("number of floating regfile reads") 568 .prereq(fpRegfileReads); 569 570 fpRegfileWrites 571 .name(name() + ".fp_regfile_writes") 572 .desc("number of floating regfile writes") 573 .prereq(fpRegfileWrites); 574 575 miscRegfileReads 576 .name(name() + ".misc_regfile_reads") 577 .desc("number of misc regfile reads") 578 .prereq(miscRegfileReads); 579 580 miscRegfileWrites 581 .name(name() + ".misc_regfile_writes") 582 .desc("number of misc regfile writes") 583 .prereq(miscRegfileWrites); 584} 585 586template <class Impl> 587void 588FullO3CPU<Impl>::tick() 589{ 590 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n"); 591 592 ++numCycles; 593 594// activity = false; 595 596 //Tick each of the stages 597 fetch.tick(); 598 599 decode.tick(); 600 601 rename.tick(); 602 603 iew.tick(); 604 605 commit.tick(); 606 607 if (!FullSystem) 608 doContextSwitch(); 609 610 // Now advance the time buffers 611 timeBuffer.advance(); 612 613 fetchQueue.advance(); 614 decodeQueue.advance(); 615 renameQueue.advance(); 616 iewQueue.advance(); 617 618 activityRec.advance(); 619 620 if (removeInstsThisCycle) { 621 cleanUpRemovedInsts(); 622 } 623 624 if (!tickEvent.scheduled()) { 625 if (_status == SwitchedOut || 626 getDrainState() == Drainable::Drained) { 627 DPRINTF(O3CPU, "Switched out!\n"); 628 // increment stat 629 lastRunningCycle = curCycle(); 630 } else if (!activityRec.active() || _status == Idle) { 631 DPRINTF(O3CPU, "Idle!\n"); 632 lastRunningCycle = curCycle(); 633 timesIdled++; 634 } else { 635 schedule(tickEvent, clockEdge(Cycles(1))); 636 DPRINTF(O3CPU, "Scheduling next tick!\n"); 637 } 638 } 639 640 if (!FullSystem) 641 updateThreadPriority(); 642} 643 644template <class Impl> 645void 646FullO3CPU<Impl>::init() 647{ 648 BaseCPU::init(); 649 650 for (ThreadID tid = 0; tid < numThreads; ++tid) { 651 // Set noSquashFromTC so that the CPU doesn't squash when initially 652 // setting up registers. 653 thread[tid]->noSquashFromTC = true; 654 // Initialise the ThreadContext's memory proxies 655 thread[tid]->initMemProxies(thread[tid]->getTC()); 656 } 657 658 // this CPU could still be unconnected if we are restoring from a 659 // checkpoint and this CPU is to be switched in, thus we can only 660 // do this here if the instruction port is actually connected, if 661 // not we have to do it as part of takeOverFrom 662 if (icachePort.isConnected()) 663 fetch.setIcache(); 664 665 if (FullSystem && !params()->defer_registration) { 666 for (ThreadID tid = 0; tid < numThreads; tid++) { 667 ThreadContext *src_tc = threadContexts[tid]; 668 TheISA::initCPU(src_tc, src_tc->contextId()); 669 } 670 } 671 672 // Clear noSquashFromTC. 673 for (int tid = 0; tid < numThreads; ++tid) 674 thread[tid]->noSquashFromTC = false; 675 676 // Initialize stages. 677 fetch.initStage(); 678 iew.initStage(); 679 rename.initStage(); 680 commit.initStage(); 681 682 commit.setThreads(thread); 683} 684 685template <class Impl> 686void 687FullO3CPU<Impl>::activateThread(ThreadID tid) 688{ 689 list<ThreadID>::iterator isActive = 690 std::find(activeThreads.begin(), activeThreads.end(), tid); 691 692 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid); 693 694 if (isActive == activeThreads.end()) { 695 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n", 696 tid); 697 698 activeThreads.push_back(tid); 699 } 700} 701 702template <class Impl> 703void 704FullO3CPU<Impl>::deactivateThread(ThreadID tid) 705{ 706 //Remove From Active List, if Active 707 list<ThreadID>::iterator thread_it = 708 std::find(activeThreads.begin(), activeThreads.end(), tid); 709 710 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid); 711 712 if (thread_it != activeThreads.end()) { 713 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n", 714 tid); 715 activeThreads.erase(thread_it); 716 } 717} 718 719template <class Impl> 720Counter 721FullO3CPU<Impl>::totalInsts() const 722{ 723 Counter total(0); 724 725 ThreadID size = thread.size(); 726 for (ThreadID i = 0; i < size; i++) 727 total += thread[i]->numInst; 728 729 return total; 730} 731 732template <class Impl> 733Counter 734FullO3CPU<Impl>::totalOps() const 735{ 736 Counter total(0); 737 738 ThreadID size = thread.size(); 739 for (ThreadID i = 0; i < size; i++) 740 total += thread[i]->numOp; 741 742 return total; 743} 744 745template <class Impl> 746void 747FullO3CPU<Impl>::activateContext(ThreadID tid, Cycles delay) 748{ 749 // Needs to set each stage to running as well. 750 if (delay){ 751 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate " 752 "on cycle %d\n", tid, clockEdge(delay)); 753 scheduleActivateThreadEvent(tid, delay); 754 } else { 755 activateThread(tid); 756 } 757 758 // If we are time 0 or if the last activation time is in the past, 759 // schedule the next tick and wake up the fetch unit 760 if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) { 761 scheduleTickEvent(delay); 762 763 // Be sure to signal that there's some activity so the CPU doesn't 764 // deschedule itself. 765 activityRec.activity(); 766 fetch.wakeFromQuiesce(); 767 768 Cycles cycles(curCycle() - lastRunningCycle); 769 // @todo: This is an oddity that is only here to match the stats 770 if (cycles != 0) 771 --cycles; 772 quiesceCycles += cycles; 773 774 lastActivatedCycle = curTick(); 775 776 _status = Running; 777 } 778} 779 780template <class Impl> 781bool 782FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove, 783 Cycles delay) 784{ 785 // Schedule removal of thread data from CPU 786 if (delay){ 787 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate " 788 "on tick %d\n", tid, clockEdge(delay)); 789 scheduleDeallocateContextEvent(tid, remove, delay); 790 return false; 791 } else { 792 deactivateThread(tid); 793 if (remove) 794 removeThread(tid); 795 return true; 796 } 797} 798 799template <class Impl> 800void 801FullO3CPU<Impl>::suspendContext(ThreadID tid) 802{ 803 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid); 804 bool deallocated = scheduleDeallocateContext(tid, false, Cycles(1)); 805 // If this was the last thread then unschedule the tick event. 806 if ((activeThreads.size() == 1 && !deallocated) || 807 activeThreads.size() == 0) 808 unscheduleTickEvent(); 809 810 DPRINTF(Quiesce, "Suspending Context\n"); 811 lastRunningCycle = curCycle(); 812 _status = Idle; 813} 814 815template <class Impl> 816void 817FullO3CPU<Impl>::haltContext(ThreadID tid) 818{ 819 //For now, this is the same as deallocate 820 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid); 821 scheduleDeallocateContext(tid, true, Cycles(1)); 822} 823 824template <class Impl> 825void 826FullO3CPU<Impl>::insertThread(ThreadID tid) 827{ 828 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU"); 829 // Will change now that the PC and thread state is internal to the CPU 830 // and not in the ThreadContext. 831 ThreadContext *src_tc; 832 if (FullSystem) 833 src_tc = system->threadContexts[tid]; 834 else 835 src_tc = tcBase(tid); 836 837 //Bind Int Regs to Rename Map 838 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 839 PhysRegIndex phys_reg = freeList.getIntReg(); 840 841 renameMap[tid].setEntry(ireg,phys_reg); 842 scoreboard.setReg(phys_reg); 843 } 844 845 //Bind Float Regs to Rename Map 846 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) { 847 PhysRegIndex phys_reg = freeList.getFloatReg(); 848 849 renameMap[tid].setEntry(freg,phys_reg); 850 scoreboard.setReg(phys_reg); 851 } 852 853 //Copy Thread Data Into RegFile 854 //this->copyFromTC(tid); 855 856 //Set PC/NPC/NNPC 857 pcState(src_tc->pcState(), tid); 858 859 src_tc->setStatus(ThreadContext::Active); 860 861 activateContext(tid, Cycles(1)); 862 863 //Reset ROB/IQ/LSQ Entries 864 commit.rob->resetEntries(); 865 iew.resetEntries(); 866} 867 868template <class Impl> 869void 870FullO3CPU<Impl>::removeThread(ThreadID tid) 871{ 872 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid); 873 874 // Copy Thread Data From RegFile 875 // If thread is suspended, it might be re-allocated 876 // this->copyToTC(tid); 877 878 879 // @todo: 2-27-2008: Fix how we free up rename mappings 880 // here to alleviate the case for double-freeing registers 881 // in SMT workloads. 882 883 // Unbind Int Regs from Rename Map 884 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) { 885 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg); 886 887 scoreboard.unsetReg(phys_reg); 888 freeList.addReg(phys_reg); 889 } 890 891 // Unbind Float Regs from Rename Map 892 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) { 893 PhysRegIndex phys_reg = renameMap[tid].lookup(freg); 894 895 scoreboard.unsetReg(phys_reg); 896 freeList.addReg(phys_reg); 897 } 898 899 // Squash Throughout Pipeline 900 DynInstPtr inst = commit.rob->readHeadInst(tid); 901 InstSeqNum squash_seq_num = inst->seqNum; 902 fetch.squash(0, squash_seq_num, inst, tid); 903 decode.squash(tid); 904 rename.squash(squash_seq_num, tid); 905 iew.squash(tid); 906 iew.ldstQueue.squash(squash_seq_num, tid); 907 commit.rob->squash(squash_seq_num, tid); 908 909 910 assert(iew.instQueue.getCount(tid) == 0); 911 assert(iew.ldstQueue.getCount(tid) == 0); 912 913 // Reset ROB/IQ/LSQ Entries 914 915 // Commented out for now. This should be possible to do by 916 // telling all the pipeline stages to drain first, and then 917 // checking until the drain completes. Once the pipeline is 918 // drained, call resetEntries(). - 10-09-06 ktlim 919/* 920 if (activeThreads.size() >= 1) { 921 commit.rob->resetEntries(); 922 iew.resetEntries(); 923 } 924*/ 925} 926 927 928template <class Impl> 929void 930FullO3CPU<Impl>::activateWhenReady(ThreadID tid) 931{ 932 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming" 933 "(e.g. PhysRegs/ROB/IQ/LSQ) \n", 934 tid); 935 936 bool ready = true; 937 938 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) { 939 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 940 "Phys. Int. Regs.\n", 941 tid); 942 ready = false; 943 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) { 944 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 945 "Phys. Float. Regs.\n", 946 tid); 947 ready = false; 948 } else if (commit.rob->numFreeEntries() >= 949 commit.rob->entryAmount(activeThreads.size() + 1)) { 950 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 951 "ROB entries.\n", 952 tid); 953 ready = false; 954 } else if (iew.instQueue.numFreeEntries() >= 955 iew.instQueue.entryAmount(activeThreads.size() + 1)) { 956 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 957 "IQ entries.\n", 958 tid); 959 ready = false; 960 } else if (iew.ldstQueue.numFreeEntries() >= 961 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) { 962 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough " 963 "LSQ entries.\n", 964 tid); 965 ready = false; 966 } 967 968 if (ready) { 969 insertThread(tid); 970 971 contextSwitch = false; 972 973 cpuWaitList.remove(tid); 974 } else { 975 suspendContext(tid); 976 977 //blocks fetch 978 contextSwitch = true; 979 980 //@todo: dont always add to waitlist 981 //do waitlist 982 cpuWaitList.push_back(tid); 983 } 984} 985 986template <class Impl> 987Fault 988FullO3CPU<Impl>::hwrei(ThreadID tid) 989{ 990#if THE_ISA == ALPHA_ISA 991 // Need to clear the lock flag upon returning from an interrupt. 992 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid); 993 994 this->thread[tid]->kernelStats->hwrei(); 995 996 // FIXME: XXX check for interrupts? XXX 997#endif 998 return NoFault; 999} 1000 1001template <class Impl> 1002bool 1003FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid) 1004{ 1005#if THE_ISA == ALPHA_ISA 1006 if (this->thread[tid]->kernelStats) 1007 this->thread[tid]->kernelStats->callpal(palFunc, 1008 this->threadContexts[tid]); 1009 1010 switch (palFunc) { 1011 case PAL::halt: 1012 halt(); 1013 if (--System::numSystemsRunning == 0) 1014 exitSimLoop("all cpus halted"); 1015 break; 1016 1017 case PAL::bpt: 1018 case PAL::bugchk: 1019 if (this->system->breakpoint()) 1020 return false; 1021 break; 1022 } 1023#endif 1024 return true; 1025} 1026 1027template <class Impl> 1028Fault 1029FullO3CPU<Impl>::getInterrupts() 1030{ 1031 // Check if there are any outstanding interrupts 1032 return this->interrupts->getInterrupt(this->threadContexts[0]); 1033} 1034 1035template <class Impl> 1036void 1037FullO3CPU<Impl>::processInterrupts(Fault interrupt) 1038{ 1039 // Check for interrupts here. For now can copy the code that 1040 // exists within isa_fullsys_traits.hh. Also assume that thread 0 1041 // is the one that handles the interrupts. 1042 // @todo: Possibly consolidate the interrupt checking code. 1043 // @todo: Allow other threads to handle interrupts. 1044 1045 assert(interrupt != NoFault); 1046 this->interrupts->updateIntrInfo(this->threadContexts[0]); 1047 1048 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name()); 1049 this->trap(interrupt, 0, NULL); 1050} 1051 1052template <class Impl> 1053void 1054FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst) 1055{ 1056 // Pass the thread's TC into the invoke method. 1057 fault->invoke(this->threadContexts[tid], inst); 1058} 1059 1060template <class Impl> 1061void 1062FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid) 1063{ 1064 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid); 1065 1066 DPRINTF(Activity,"Activity: syscall() called.\n"); 1067 1068 // Temporarily increase this by one to account for the syscall 1069 // instruction. 1070 ++(this->thread[tid]->funcExeInst); 1071 1072 // Execute the actual syscall. 1073 this->thread[tid]->syscall(callnum); 1074 1075 // Decrease funcExeInst by one as the normal commit will handle 1076 // incrementing it. 1077 --(this->thread[tid]->funcExeInst); 1078} 1079 1080template <class Impl> 1081void 1082FullO3CPU<Impl>::serialize(std::ostream &os) 1083{ 1084 Drainable::State so_state(getDrainState()); 1085 SERIALIZE_ENUM(so_state); 1086 BaseCPU::serialize(os); 1087 nameOut(os, csprintf("%s.tickEvent", name())); 1088 tickEvent.serialize(os); 1089 1090 // Use SimpleThread's ability to checkpoint to make it easier to 1091 // write out the registers. Also make this static so it doesn't 1092 // get instantiated multiple times (causes a panic in statistics). 1093 static SimpleThread temp; 1094 1095 ThreadID size = thread.size(); 1096 for (ThreadID i = 0; i < size; i++) { 1097 nameOut(os, csprintf("%s.xc.%i", name(), i)); 1098 temp.copyTC(thread[i]->getTC()); 1099 temp.serialize(os); 1100 } 1101} 1102 1103template <class Impl> 1104void 1105FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string §ion) 1106{ 1107 Drainable::State so_state; 1108 UNSERIALIZE_ENUM(so_state); 1109 BaseCPU::unserialize(cp, section); 1110 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section)); 1111 1112 // Use SimpleThread's ability to checkpoint to make it easier to 1113 // read in the registers. Also make this static so it doesn't 1114 // get instantiated multiple times (causes a panic in statistics). 1115 static SimpleThread temp; 1116 1117 ThreadID size = thread.size(); 1118 for (ThreadID i = 0; i < size; i++) { 1119 temp.copyTC(thread[i]->getTC()); 1120 temp.unserialize(cp, csprintf("%s.xc.%i", section, i)); 1121 thread[i]->getTC()->copyArchRegs(temp.getTC()); 1122 } 1123} 1124 1125template <class Impl> 1126unsigned int 1127FullO3CPU<Impl>::drain(DrainManager *drain_manager) 1128{ 1129 DPRINTF(O3CPU, "Switching out\n"); 1130 1131 // If the CPU isn't doing anything, then return immediately. 1132 if (_status == SwitchedOut) 1133 return 0; 1134 1135 drainCount = 0; 1136 fetch.drain(); 1137 decode.drain(); 1138 rename.drain(); 1139 iew.drain(); 1140 commit.drain(); 1141 1142 // Wake the CPU and record activity so everything can drain out if 1143 // the CPU was not able to immediately drain. 1144 if (getDrainState() != Drainable::Drained) { 1145 // A bit of a hack...set the drainManager after all the drain() 1146 // calls have been made, that way if all of the stages drain 1147 // immediately, the signalDrained() function knows not to call 1148 // process on the drain event. 1149 drainManager = drain_manager; 1150 1151 wakeCPU(); 1152 activityRec.activity(); 1153 1154 DPRINTF(Drain, "CPU not drained\n"); 1155 1156 return 1; 1157 } else { 1158 return 0; 1159 } 1160} 1161 1162template <class Impl> 1163void 1164FullO3CPU<Impl>::drainResume() 1165{ 1166 fetch.resume(); 1167 decode.resume(); 1168 rename.resume(); 1169 iew.resume(); 1170 commit.resume(); 1171 1172 setDrainState(Drainable::Running); 1173 1174 if (_status == SwitchedOut) 1175 return; 1176 1177 assert(system->getMemoryMode() == Enums::timing); 1178 1179 if (!tickEvent.scheduled()) 1180 schedule(tickEvent, nextCycle()); 1181 _status = Running; 1182} 1183 1184template <class Impl> 1185void 1186FullO3CPU<Impl>::signalDrained() 1187{ 1188 if (++drainCount == NumStages) { 1189 if (tickEvent.scheduled()) 1190 tickEvent.squash(); 1191 1192 setDrainState(Drainable::Drained); 1193 1194 BaseCPU::switchOut(); 1195 1196 if (drainManager) { 1197 DPRINTF(Drain, "CPU done draining, processing drain event\n"); 1198 drainManager->signalDrainDone(); 1199 drainManager = NULL; 1200 } 1201 } 1202 assert(drainCount <= 5); 1203} 1204 1205template <class Impl> 1206void 1207FullO3CPU<Impl>::switchOut() 1208{ 1209 fetch.switchOut(); 1210 rename.switchOut(); 1211 iew.switchOut(); 1212 commit.switchOut(); 1213 instList.clear(); 1214 while (!removeList.empty()) { 1215 removeList.pop(); 1216 } 1217 1218 _status = SwitchedOut; 1219 1220 if (checker) 1221 checker->switchOut(); 1222 1223 if (tickEvent.scheduled()) 1224 tickEvent.squash(); 1225} 1226 1227template <class Impl> 1228void 1229FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU) 1230{ 1231 // Flush out any old data from the time buffers. 1232 for (int i = 0; i < timeBuffer.getSize(); ++i) { 1233 timeBuffer.advance(); 1234 fetchQueue.advance(); 1235 decodeQueue.advance(); 1236 renameQueue.advance(); 1237 iewQueue.advance(); 1238 } 1239 1240 activityRec.reset(); 1241 1242 BaseCPU::takeOverFrom(oldCPU); 1243 1244 fetch.takeOverFrom(); 1245 decode.takeOverFrom(); 1246 rename.takeOverFrom(); 1247 iew.takeOverFrom(); 1248 commit.takeOverFrom(); 1249 1250 assert(!tickEvent.scheduled() || tickEvent.squashed()); 1251 1252 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU); 1253 if (oldO3CPU) 1254 globalSeqNum = oldO3CPU->globalSeqNum; 1255 1256 // @todo: Figure out how to properly select the tid to put onto 1257 // the active threads list. 1258 ThreadID tid = 0; 1259 1260 list<ThreadID>::iterator isActive = 1261 std::find(activeThreads.begin(), activeThreads.end(), tid); 1262 1263 if (isActive == activeThreads.end()) { 1264 //May Need to Re-code this if the delay variable is the delay 1265 //needed for thread to activate 1266 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n", 1267 tid); 1268 1269 activeThreads.push_back(tid); 1270 } 1271 1272 // Set all statuses to active, schedule the CPU's tick event. 1273 // @todo: Fix up statuses so this is handled properly 1274 ThreadID size = threadContexts.size(); 1275 for (ThreadID i = 0; i < size; ++i) { 1276 ThreadContext *tc = threadContexts[i]; 1277 if (tc->status() == ThreadContext::Active && _status != Running) { 1278 _status = Running; 1279 reschedule(tickEvent, nextCycle(), true); 1280 } 1281 } 1282 if (!tickEvent.scheduled()) 1283 schedule(tickEvent, nextCycle()); 1284 1285 lastRunningCycle = curCycle(); 1286} 1287 1288template <class Impl> 1289TheISA::MiscReg 1290FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid) 1291{ 1292 return this->isa[tid]->readMiscRegNoEffect(misc_reg); 1293} 1294 1295template <class Impl> 1296TheISA::MiscReg 1297FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid) 1298{ 1299 miscRegfileReads++; 1300 return this->isa[tid]->readMiscReg(misc_reg, tcBase(tid)); 1301} 1302 1303template <class Impl> 1304void 1305FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg, 1306 const TheISA::MiscReg &val, ThreadID tid) 1307{ 1308 this->isa[tid]->setMiscRegNoEffect(misc_reg, val); 1309} 1310 1311template <class Impl> 1312void 1313FullO3CPU<Impl>::setMiscReg(int misc_reg, 1314 const TheISA::MiscReg &val, ThreadID tid) 1315{ 1316 miscRegfileWrites++; 1317 this->isa[tid]->setMiscReg(misc_reg, val, tcBase(tid)); 1318} 1319 1320template <class Impl> 1321uint64_t 1322FullO3CPU<Impl>::readIntReg(int reg_idx) 1323{ 1324 intRegfileReads++; 1325 return regFile.readIntReg(reg_idx); 1326} 1327 1328template <class Impl> 1329FloatReg 1330FullO3CPU<Impl>::readFloatReg(int reg_idx) 1331{ 1332 fpRegfileReads++; 1333 return regFile.readFloatReg(reg_idx); 1334} 1335 1336template <class Impl> 1337FloatRegBits 1338FullO3CPU<Impl>::readFloatRegBits(int reg_idx) 1339{ 1340 fpRegfileReads++; 1341 return regFile.readFloatRegBits(reg_idx); 1342} 1343 1344template <class Impl> 1345void 1346FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val) 1347{ 1348 intRegfileWrites++; 1349 regFile.setIntReg(reg_idx, val); 1350} 1351 1352template <class Impl> 1353void 1354FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val) 1355{ 1356 fpRegfileWrites++; 1357 regFile.setFloatReg(reg_idx, val); 1358} 1359 1360template <class Impl> 1361void 1362FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val) 1363{ 1364 fpRegfileWrites++; 1365 regFile.setFloatRegBits(reg_idx, val); 1366} 1367 1368template <class Impl> 1369uint64_t 1370FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid) 1371{ 1372 intRegfileReads++; 1373 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1374 1375 return regFile.readIntReg(phys_reg); 1376} 1377 1378template <class Impl> 1379float 1380FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid) 1381{ 1382 fpRegfileReads++; 1383 int idx = reg_idx + TheISA::NumIntRegs; 1384 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1385 1386 return regFile.readFloatReg(phys_reg); 1387} 1388 1389template <class Impl> 1390uint64_t 1391FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid) 1392{ 1393 fpRegfileReads++; 1394 int idx = reg_idx + TheISA::NumIntRegs; 1395 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1396 1397 return regFile.readFloatRegBits(phys_reg); 1398} 1399 1400template <class Impl> 1401void 1402FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid) 1403{ 1404 intRegfileWrites++; 1405 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx); 1406 1407 regFile.setIntReg(phys_reg, val); 1408} 1409 1410template <class Impl> 1411void 1412FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid) 1413{ 1414 fpRegfileWrites++; 1415 int idx = reg_idx + TheISA::NumIntRegs; 1416 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1417 1418 regFile.setFloatReg(phys_reg, val); 1419} 1420 1421template <class Impl> 1422void 1423FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid) 1424{ 1425 fpRegfileWrites++; 1426 int idx = reg_idx + TheISA::NumIntRegs; 1427 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx); 1428 1429 regFile.setFloatRegBits(phys_reg, val); 1430} 1431 1432template <class Impl> 1433TheISA::PCState 1434FullO3CPU<Impl>::pcState(ThreadID tid) 1435{ 1436 return commit.pcState(tid); 1437} 1438 1439template <class Impl> 1440void 1441FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid) 1442{ 1443 commit.pcState(val, tid); 1444} 1445 1446template <class Impl> 1447Addr 1448FullO3CPU<Impl>::instAddr(ThreadID tid) 1449{ 1450 return commit.instAddr(tid); 1451} 1452 1453template <class Impl> 1454Addr 1455FullO3CPU<Impl>::nextInstAddr(ThreadID tid) 1456{ 1457 return commit.nextInstAddr(tid); 1458} 1459 1460template <class Impl> 1461MicroPC 1462FullO3CPU<Impl>::microPC(ThreadID tid) 1463{ 1464 return commit.microPC(tid); 1465} 1466 1467template <class Impl> 1468void 1469FullO3CPU<Impl>::squashFromTC(ThreadID tid) 1470{ 1471 this->thread[tid]->noSquashFromTC = true; 1472 this->commit.generateTCEvent(tid); 1473} 1474 1475template <class Impl> 1476typename FullO3CPU<Impl>::ListIt 1477FullO3CPU<Impl>::addInst(DynInstPtr &inst) 1478{ 1479 instList.push_back(inst); 1480 1481 return --(instList.end()); 1482} 1483 1484template <class Impl> 1485void 1486FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst) 1487{ 1488 // Keep an instruction count. 1489 if (!inst->isMicroop() || inst->isLastMicroop()) { 1490 thread[tid]->numInst++; 1491 thread[tid]->numInsts++; 1492 committedInsts[tid]++; 1493 totalCommittedInsts++; 1494 } 1495 thread[tid]->numOp++; 1496 thread[tid]->numOps++; 1497 committedOps[tid]++; 1498 1499 system->totalNumInsts++; 1500 // Check for instruction-count-based events. 1501 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst); 1502 system->instEventQueue.serviceEvents(system->totalNumInsts); 1503} 1504 1505template <class Impl> 1506void 1507FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst) 1508{ 1509 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s " 1510 "[sn:%lli]\n", 1511 inst->threadNumber, inst->pcState(), inst->seqNum); 1512 1513 removeInstsThisCycle = true; 1514 1515 // Remove the front instruction. 1516 removeList.push(inst->getInstListIt()); 1517} 1518 1519template <class Impl> 1520void 1521FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid) 1522{ 1523 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction" 1524 " list.\n", tid); 1525 1526 ListIt end_it; 1527 1528 bool rob_empty = false; 1529 1530 if (instList.empty()) { 1531 return; 1532 } else if (rob.isEmpty(/*tid*/)) { 1533 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n"); 1534 end_it = instList.begin(); 1535 rob_empty = true; 1536 } else { 1537 end_it = (rob.readTailInst(tid))->getInstListIt(); 1538 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n"); 1539 } 1540 1541 removeInstsThisCycle = true; 1542 1543 ListIt inst_it = instList.end(); 1544 1545 inst_it--; 1546 1547 // Walk through the instruction list, removing any instructions 1548 // that were inserted after the given instruction iterator, end_it. 1549 while (inst_it != end_it) { 1550 assert(!instList.empty()); 1551 1552 squashInstIt(inst_it, tid); 1553 1554 inst_it--; 1555 } 1556 1557 // If the ROB was empty, then we actually need to remove the first 1558 // instruction as well. 1559 if (rob_empty) { 1560 squashInstIt(inst_it, tid); 1561 } 1562} 1563 1564template <class Impl> 1565void 1566FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid) 1567{ 1568 assert(!instList.empty()); 1569 1570 removeInstsThisCycle = true; 1571 1572 ListIt inst_iter = instList.end(); 1573 1574 inst_iter--; 1575 1576 DPRINTF(O3CPU, "Deleting instructions from instruction " 1577 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n", 1578 tid, seq_num, (*inst_iter)->seqNum); 1579 1580 while ((*inst_iter)->seqNum > seq_num) { 1581 1582 bool break_loop = (inst_iter == instList.begin()); 1583 1584 squashInstIt(inst_iter, tid); 1585 1586 inst_iter--; 1587 1588 if (break_loop) 1589 break; 1590 } 1591} 1592 1593template <class Impl> 1594inline void 1595FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid) 1596{ 1597 if ((*instIt)->threadNumber == tid) { 1598 DPRINTF(O3CPU, "Squashing instruction, " 1599 "[tid:%i] [sn:%lli] PC %s\n", 1600 (*instIt)->threadNumber, 1601 (*instIt)->seqNum, 1602 (*instIt)->pcState()); 1603 1604 // Mark it as squashed. 1605 (*instIt)->setSquashed(); 1606 1607 // @todo: Formulate a consistent method for deleting 1608 // instructions from the instruction list 1609 // Remove the instruction from the list. 1610 removeList.push(instIt); 1611 } 1612} 1613 1614template <class Impl> 1615void 1616FullO3CPU<Impl>::cleanUpRemovedInsts() 1617{ 1618 while (!removeList.empty()) { 1619 DPRINTF(O3CPU, "Removing instruction, " 1620 "[tid:%i] [sn:%lli] PC %s\n", 1621 (*removeList.front())->threadNumber, 1622 (*removeList.front())->seqNum, 1623 (*removeList.front())->pcState()); 1624 1625 instList.erase(removeList.front()); 1626 1627 removeList.pop(); 1628 } 1629 1630 removeInstsThisCycle = false; 1631} 1632/* 1633template <class Impl> 1634void 1635FullO3CPU<Impl>::removeAllInsts() 1636{ 1637 instList.clear(); 1638} 1639*/ 1640template <class Impl> 1641void 1642FullO3CPU<Impl>::dumpInsts() 1643{ 1644 int num = 0; 1645 1646 ListIt inst_list_it = instList.begin(); 1647 1648 cprintf("Dumping Instruction List\n"); 1649 1650 while (inst_list_it != instList.end()) { 1651 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n" 1652 "Squashed:%i\n\n", 1653 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber, 1654 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(), 1655 (*inst_list_it)->isSquashed()); 1656 inst_list_it++; 1657 ++num; 1658 } 1659} 1660/* 1661template <class Impl> 1662void 1663FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst) 1664{ 1665 iew.wakeDependents(inst); 1666} 1667*/ 1668template <class Impl> 1669void 1670FullO3CPU<Impl>::wakeCPU() 1671{ 1672 if (activityRec.active() || tickEvent.scheduled()) { 1673 DPRINTF(Activity, "CPU already running.\n"); 1674 return; 1675 } 1676 1677 DPRINTF(Activity, "Waking up CPU\n"); 1678 1679 Cycles cycles(curCycle() - lastRunningCycle); 1680 // @todo: This is an oddity that is only here to match the stats 1681 if (cycles != 0) 1682 --cycles; 1683 idleCycles += cycles; 1684 numCycles += cycles; 1685 1686 schedule(tickEvent, nextCycle()); 1687} 1688 1689template <class Impl> 1690void 1691FullO3CPU<Impl>::wakeup() 1692{ 1693 if (this->thread[0]->status() != ThreadContext::Suspended) 1694 return; 1695 1696 this->wakeCPU(); 1697 1698 DPRINTF(Quiesce, "Suspended Processor woken\n"); 1699 this->threadContexts[0]->activate(); 1700} 1701 1702template <class Impl> 1703ThreadID 1704FullO3CPU<Impl>::getFreeTid() 1705{ 1706 for (ThreadID tid = 0; tid < numThreads; tid++) { 1707 if (!tids[tid]) { 1708 tids[tid] = true; 1709 return tid; 1710 } 1711 } 1712 1713 return InvalidThreadID; 1714} 1715 1716template <class Impl> 1717void 1718FullO3CPU<Impl>::doContextSwitch() 1719{ 1720 if (contextSwitch) { 1721 1722 //ADD CODE TO DEACTIVE THREAD HERE (???) 1723 1724 ThreadID size = cpuWaitList.size(); 1725 for (ThreadID tid = 0; tid < size; tid++) { 1726 activateWhenReady(tid); 1727 } 1728 1729 if (cpuWaitList.size() == 0) 1730 contextSwitch = true; 1731 } 1732} 1733 1734template <class Impl> 1735void 1736FullO3CPU<Impl>::updateThreadPriority() 1737{ 1738 if (activeThreads.size() > 1) { 1739 //DEFAULT TO ROUND ROBIN SCHEME 1740 //e.g. Move highest priority to end of thread list 1741 list<ThreadID>::iterator list_begin = activeThreads.begin(); 1742 1743 unsigned high_thread = *list_begin; 1744 1745 activeThreads.erase(list_begin); 1746 1747 activeThreads.push_back(high_thread); 1748 } 1749} 1750 1751// Forward declaration of FullO3CPU. 1752template class FullO3CPU<O3CPUImpl>; 1753