cpu.cc (8793:5f25086326ac) cpu.cc (8795:0909f8ed7aa0)
1/*
2 * Copyright (c) 2004-2006 The Regents of The University of Michigan
3 * Copyright (c) 2011 Regents of the University of California
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions are
8 * met: redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer;
10 * redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution;
13 * neither the name of the copyright holders nor the names of its
14 * contributors may be used to endorse or promote products derived from
15 * this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
18 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
19 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
20 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
21 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
22 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
23 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
29 * Authors: Kevin Lim
30 * Korey Sewell
31 * Rick Strong
32 */
33
34#include "arch/kernel_stats.hh"
35#include "config/the_isa.hh"
36#include "config/use_checker.hh"
37#include "cpu/o3/cpu.hh"
38#include "cpu/o3/isa_specific.hh"
39#include "cpu/o3/thread_context.hh"
40#include "cpu/activity.hh"
41#include "cpu/quiesce_event.hh"
42#include "cpu/simple_thread.hh"
43#include "cpu/thread_context.hh"
44#include "debug/Activity.hh"
45#include "debug/O3CPU.hh"
46#include "debug/Quiesce.hh"
47#include "enums/MemoryMode.hh"
48#include "sim/core.hh"
49#include "sim/full_system.hh"
50#include "sim/process.hh"
51#include "sim/stat_control.hh"
52#include "sim/system.hh"
53
54#if USE_CHECKER
55#include "cpu/checker/cpu.hh"
56#endif
57
58#if THE_ISA == ALPHA_ISA
59#include "arch/alpha/osfpal.hh"
60#include "debug/Activity.hh"
61#endif
62
63class BaseCPUParams;
64
65using namespace TheISA;
66using namespace std;
67
68BaseO3CPU::BaseO3CPU(BaseCPUParams *params)
69 : BaseCPU(params)
70{
71}
72
73void
74BaseO3CPU::regStats()
75{
76 BaseCPU::regStats();
77}
78
79template <class Impl>
80FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c)
81 : Event(CPU_Tick_Pri), cpu(c)
82{
83}
84
85template <class Impl>
86void
87FullO3CPU<Impl>::TickEvent::process()
88{
89 cpu->tick();
90}
91
92template <class Impl>
93const char *
94FullO3CPU<Impl>::TickEvent::description() const
95{
96 return "FullO3CPU tick";
97}
98
99template <class Impl>
100FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent()
101 : Event(CPU_Switch_Pri)
102{
103}
104
105template <class Impl>
106void
107FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num,
108 FullO3CPU<Impl> *thread_cpu)
109{
110 tid = thread_num;
111 cpu = thread_cpu;
112}
113
114template <class Impl>
115void
116FullO3CPU<Impl>::ActivateThreadEvent::process()
117{
118 cpu->activateThread(tid);
119}
120
121template <class Impl>
122const char *
123FullO3CPU<Impl>::ActivateThreadEvent::description() const
124{
125 return "FullO3CPU \"Activate Thread\"";
126}
127
128template <class Impl>
129FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent()
130 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL)
131{
132}
133
134template <class Impl>
135void
136FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num,
137 FullO3CPU<Impl> *thread_cpu)
138{
139 tid = thread_num;
140 cpu = thread_cpu;
141 remove = false;
142}
143
144template <class Impl>
145void
146FullO3CPU<Impl>::DeallocateContextEvent::process()
147{
148 cpu->deactivateThread(tid);
149 if (remove)
150 cpu->removeThread(tid);
151}
152
153template <class Impl>
154const char *
155FullO3CPU<Impl>::DeallocateContextEvent::description() const
156{
157 return "FullO3CPU \"Deallocate Context\"";
158}
159
160template <class Impl>
161FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params)
162 : BaseO3CPU(params),
163 itb(params->itb),
164 dtb(params->dtb),
165 tickEvent(this),
166#ifndef NDEBUG
167 instcount(0),
168#endif
169 removeInstsThisCycle(false),
170 fetch(this, params),
171 decode(this, params),
172 rename(this, params),
173 iew(this, params),
174 commit(this, params),
175
176 regFile(this, params->numPhysIntRegs,
177 params->numPhysFloatRegs),
178
179 freeList(params->numThreads,
180 TheISA::NumIntRegs, params->numPhysIntRegs,
181 TheISA::NumFloatRegs, params->numPhysFloatRegs),
182
183 rob(this,
184 params->numROBEntries, params->squashWidth,
185 params->smtROBPolicy, params->smtROBThreshold,
186 params->numThreads),
187
188 scoreboard(params->numThreads,
189 TheISA::NumIntRegs, params->numPhysIntRegs,
190 TheISA::NumFloatRegs, params->numPhysFloatRegs,
191 TheISA::NumMiscRegs * numThreads,
192 TheISA::ZeroReg),
193
194 timeBuffer(params->backComSize, params->forwardComSize),
195 fetchQueue(params->backComSize, params->forwardComSize),
196 decodeQueue(params->backComSize, params->forwardComSize),
197 renameQueue(params->backComSize, params->forwardComSize),
198 iewQueue(params->backComSize, params->forwardComSize),
199 activityRec(name(), NumStages,
200 params->backComSize + params->forwardComSize,
201 params->activity),
202
203 globalSeqNum(1),
204 system(params->system),
205 drainCount(0),
206 deferRegistration(params->defer_registration)
207{
208 if (!deferRegistration) {
209 _status = Running;
210 } else {
211 _status = Idle;
212 }
213
214#if USE_CHECKER
215 if (params->checker) {
216 BaseCPU *temp_checker = params->checker;
217 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker);
218 checker->setSystem(params->system);
219 } else {
220 checker = NULL;
221 }
222#endif // USE_CHECKER
223
224 if (!FullSystem) {
225 thread.resize(numThreads);
226 tids.resize(numThreads);
227 }
228
229 // The stages also need their CPU pointer setup. However this
230 // must be done at the upper level CPU because they have pointers
231 // to the upper level CPU, and not this FullO3CPU.
232
233 // Set up Pointers to the activeThreads list for each stage
234 fetch.setActiveThreads(&activeThreads);
235 decode.setActiveThreads(&activeThreads);
236 rename.setActiveThreads(&activeThreads);
237 iew.setActiveThreads(&activeThreads);
238 commit.setActiveThreads(&activeThreads);
239
240 // Give each of the stages the time buffer they will use.
241 fetch.setTimeBuffer(&timeBuffer);
242 decode.setTimeBuffer(&timeBuffer);
243 rename.setTimeBuffer(&timeBuffer);
244 iew.setTimeBuffer(&timeBuffer);
245 commit.setTimeBuffer(&timeBuffer);
246
247 // Also setup each of the stages' queues.
248 fetch.setFetchQueue(&fetchQueue);
249 decode.setFetchQueue(&fetchQueue);
250 commit.setFetchQueue(&fetchQueue);
251 decode.setDecodeQueue(&decodeQueue);
252 rename.setDecodeQueue(&decodeQueue);
253 rename.setRenameQueue(&renameQueue);
254 iew.setRenameQueue(&renameQueue);
255 iew.setIEWQueue(&iewQueue);
256 commit.setIEWQueue(&iewQueue);
257 commit.setRenameQueue(&renameQueue);
258
259 commit.setIEWStage(&iew);
260 rename.setIEWStage(&iew);
261 rename.setCommitStage(&commit);
262
263 ThreadID active_threads;
264 if (FullSystem) {
265 active_threads = 1;
266 } else {
267 active_threads = params->workload.size();
268
269 if (active_threads > Impl::MaxThreads) {
270 panic("Workload Size too large. Increase the 'MaxThreads' "
271 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) "
272 "or edit your workload size.");
273 }
274 }
275
276 //Make Sure That this a Valid Architeture
277 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs);
278 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs);
279
280 rename.setScoreboard(&scoreboard);
281 iew.setScoreboard(&scoreboard);
282
283 // Setup the rename map for whichever stages need it.
284 PhysRegIndex lreg_idx = 0;
285 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs
286
287 for (ThreadID tid = 0; tid < numThreads; tid++) {
288 bool bindRegs = (tid <= active_threads - 1);
289
290 commitRenameMap[tid].init(TheISA::NumIntRegs,
291 params->numPhysIntRegs,
292 lreg_idx, //Index for Logical. Regs
293
294 TheISA::NumFloatRegs,
295 params->numPhysFloatRegs,
296 freg_idx, //Index for Float Regs
297
298 TheISA::NumMiscRegs,
299
300 TheISA::ZeroReg,
301 TheISA::ZeroReg,
302
303 tid,
304 false);
305
306 renameMap[tid].init(TheISA::NumIntRegs,
307 params->numPhysIntRegs,
308 lreg_idx, //Index for Logical. Regs
309
310 TheISA::NumFloatRegs,
311 params->numPhysFloatRegs,
312 freg_idx, //Index for Float Regs
313
314 TheISA::NumMiscRegs,
315
316 TheISA::ZeroReg,
317 TheISA::ZeroReg,
318
319 tid,
320 bindRegs);
321
322 activateThreadEvent[tid].init(tid, this);
323 deallocateContextEvent[tid].init(tid, this);
324 }
325
326 rename.setRenameMap(renameMap);
327 commit.setRenameMap(commitRenameMap);
328
329 // Give renameMap & rename stage access to the freeList;
330 for (ThreadID tid = 0; tid < numThreads; tid++)
331 renameMap[tid].setFreeList(&freeList);
332 rename.setFreeList(&freeList);
333
334 // Setup the ROB for whichever stages need it.
335 commit.setROB(&rob);
336
337 lastRunningCycle = curTick();
338
339 lastActivatedCycle = -1;
340#if 0
341 // Give renameMap & rename stage access to the freeList;
342 for (ThreadID tid = 0; tid < numThreads; tid++)
343 globalSeqNum[tid] = 1;
344#endif
345
346 contextSwitch = false;
347 DPRINTF(O3CPU, "Creating O3CPU object.\n");
348
349 // Setup any thread state.
350 this->thread.resize(this->numThreads);
351
352 for (ThreadID tid = 0; tid < this->numThreads; ++tid) {
353 if (FullSystem) {
354 // SMT is not supported in FS mode yet.
355 assert(this->numThreads == 1);
356 this->thread[tid] = new Thread(this, 0, NULL);
357 } else {
358 if (tid < params->workload.size()) {
359 DPRINTF(O3CPU, "Workload[%i] process is %#x",
360 tid, this->thread[tid]);
361 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
362 (typename Impl::O3CPU *)(this),
363 tid, params->workload[tid]);
364
365 //usedTids[tid] = true;
366 //threadMap[tid] = tid;
367 } else {
368 //Allocate Empty thread so M5 can use later
369 //when scheduling threads to CPU
370 Process* dummy_proc = NULL;
371
372 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
373 (typename Impl::O3CPU *)(this),
374 tid, dummy_proc);
375 //usedTids[tid] = false;
376 }
377 }
378
379 ThreadContext *tc;
380
381 // Setup the TC that will serve as the interface to the threads/CPU.
382 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>;
383
384 tc = o3_tc;
385
386 // If we're using a checker, then the TC should be the
387 // CheckerThreadContext.
388#if USE_CHECKER
389 if (params->checker) {
390 tc = new CheckerThreadContext<O3ThreadContext<Impl> >(
391 o3_tc, this->checker);
392 }
393#endif
394
395 o3_tc->cpu = (typename Impl::O3CPU *)(this);
396 assert(o3_tc->cpu);
397 o3_tc->thread = this->thread[tid];
398
399 if (FullSystem) {
400 // Setup quiesce event.
401 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc);
402 }
403 // Give the thread the TC.
404 this->thread[tid]->tc = tc;
405
406 // Add the TC to the CPU's list of TC's.
407 this->threadContexts.push_back(tc);
408 }
409
410 for (ThreadID tid = 0; tid < this->numThreads; tid++)
411 this->thread[tid]->setFuncExeInst(0);
412
413 lockAddr = 0;
414 lockFlag = false;
415}
416
417template <class Impl>
418FullO3CPU<Impl>::~FullO3CPU()
419{
420}
421
422template <class Impl>
423void
424FullO3CPU<Impl>::regStats()
425{
426 BaseO3CPU::regStats();
427
428 // Register any of the O3CPU's stats here.
429 timesIdled
430 .name(name() + ".timesIdled")
431 .desc("Number of times that the entire CPU went into an idle state and"
432 " unscheduled itself")
433 .prereq(timesIdled);
434
435 idleCycles
436 .name(name() + ".idleCycles")
437 .desc("Total number of cycles that the CPU has spent unscheduled due "
438 "to idling")
439 .prereq(idleCycles);
440
441 // Number of Instructions simulated
442 // --------------------------------
443 // Should probably be in Base CPU but need templated
444 // MaxThreads so put in here instead
445 committedInsts
446 .init(numThreads)
447 .name(name() + ".committedInsts")
448 .desc("Number of Instructions Simulated");
449
450 totalCommittedInsts
451 .name(name() + ".committedInsts_total")
452 .desc("Number of Instructions Simulated");
453
454 cpi
455 .name(name() + ".cpi")
456 .desc("CPI: Cycles Per Instruction")
457 .precision(6);
458 cpi = numCycles / committedInsts;
459
460 totalCpi
461 .name(name() + ".cpi_total")
462 .desc("CPI: Total CPI of All Threads")
463 .precision(6);
464 totalCpi = numCycles / totalCommittedInsts;
465
466 ipc
467 .name(name() + ".ipc")
468 .desc("IPC: Instructions Per Cycle")
469 .precision(6);
470 ipc = committedInsts / numCycles;
471
472 totalIpc
473 .name(name() + ".ipc_total")
474 .desc("IPC: Total IPC of All Threads")
475 .precision(6);
476 totalIpc = totalCommittedInsts / numCycles;
477
478 this->fetch.regStats();
479 this->decode.regStats();
480 this->rename.regStats();
481 this->iew.regStats();
482 this->commit.regStats();
483 this->rob.regStats();
484
485 intRegfileReads
486 .name(name() + ".int_regfile_reads")
487 .desc("number of integer regfile reads")
488 .prereq(intRegfileReads);
489
490 intRegfileWrites
491 .name(name() + ".int_regfile_writes")
492 .desc("number of integer regfile writes")
493 .prereq(intRegfileWrites);
494
495 fpRegfileReads
496 .name(name() + ".fp_regfile_reads")
497 .desc("number of floating regfile reads")
498 .prereq(fpRegfileReads);
499
500 fpRegfileWrites
501 .name(name() + ".fp_regfile_writes")
502 .desc("number of floating regfile writes")
503 .prereq(fpRegfileWrites);
504
505 miscRegfileReads
506 .name(name() + ".misc_regfile_reads")
507 .desc("number of misc regfile reads")
508 .prereq(miscRegfileReads);
509
510 miscRegfileWrites
511 .name(name() + ".misc_regfile_writes")
512 .desc("number of misc regfile writes")
513 .prereq(miscRegfileWrites);
514}
515
516template <class Impl>
517Port *
518FullO3CPU<Impl>::getPort(const std::string &if_name, int idx)
519{
520 if (if_name == "dcache_port")
521 return iew.getDcachePort();
522 else if (if_name == "icache_port")
523 return fetch.getIcachePort();
524 else
525 panic("No Such Port\n");
526}
527
528template <class Impl>
529void
530FullO3CPU<Impl>::tick()
531{
532 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n");
533
534 ++numCycles;
535
536// activity = false;
537
538 //Tick each of the stages
539 fetch.tick();
540
541 decode.tick();
542
543 rename.tick();
544
545 iew.tick();
546
547 commit.tick();
548
549 if (!FullSystem)
550 doContextSwitch();
551
552 // Now advance the time buffers
553 timeBuffer.advance();
554
555 fetchQueue.advance();
556 decodeQueue.advance();
557 renameQueue.advance();
558 iewQueue.advance();
559
560 activityRec.advance();
561
562 if (removeInstsThisCycle) {
563 cleanUpRemovedInsts();
564 }
565
566 if (!tickEvent.scheduled()) {
567 if (_status == SwitchedOut ||
568 getState() == SimObject::Drained) {
569 DPRINTF(O3CPU, "Switched out!\n");
570 // increment stat
571 lastRunningCycle = curTick();
572 } else if (!activityRec.active() || _status == Idle) {
573 DPRINTF(O3CPU, "Idle!\n");
574 lastRunningCycle = curTick();
575 timesIdled++;
576 } else {
577 schedule(tickEvent, nextCycle(curTick() + ticks(1)));
578 DPRINTF(O3CPU, "Scheduling next tick!\n");
579 }
580 }
581
582 if (!FullSystem)
583 updateThreadPriority();
584}
585
586template <class Impl>
587void
588FullO3CPU<Impl>::init()
589{
590 BaseCPU::init();
591
592 // Set inSyscall so that the CPU doesn't squash when initially
593 // setting up registers.
594 for (ThreadID tid = 0; tid < numThreads; ++tid)
595 thread[tid]->inSyscall = true;
596
597 if (FullSystem) {
598 for (ThreadID tid = 0; tid < numThreads; tid++) {
599 ThreadContext *src_tc = threadContexts[tid];
600 TheISA::initCPU(src_tc, src_tc->contextId());
601 }
602 }
603
604 // Clear inSyscall.
605 for (int tid = 0; tid < numThreads; ++tid)
606 thread[tid]->inSyscall = false;
607
608 // Initialize stages.
609 fetch.initStage();
610 iew.initStage();
611 rename.initStage();
612 commit.initStage();
613
614 commit.setThreads(thread);
615}
616
617template <class Impl>
618void
619FullO3CPU<Impl>::activateThread(ThreadID tid)
620{
621 list<ThreadID>::iterator isActive =
622 std::find(activeThreads.begin(), activeThreads.end(), tid);
623
624 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid);
625
626 if (isActive == activeThreads.end()) {
627 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n",
628 tid);
629
630 activeThreads.push_back(tid);
631 }
632}
633
634template <class Impl>
635void
636FullO3CPU<Impl>::deactivateThread(ThreadID tid)
637{
638 //Remove From Active List, if Active
639 list<ThreadID>::iterator thread_it =
640 std::find(activeThreads.begin(), activeThreads.end(), tid);
641
642 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid);
643
644 if (thread_it != activeThreads.end()) {
645 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n",
646 tid);
647 activeThreads.erase(thread_it);
648 }
649}
650
651template <class Impl>
652Counter
653FullO3CPU<Impl>::totalInstructions() const
654{
655 Counter total(0);
656
657 ThreadID size = thread.size();
658 for (ThreadID i = 0; i < size; i++)
659 total += thread[i]->numInst;
660
661 return total;
662}
663
664template <class Impl>
665void
666FullO3CPU<Impl>::activateContext(ThreadID tid, int delay)
667{
668 // Needs to set each stage to running as well.
669 if (delay){
670 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate "
671 "on cycle %d\n", tid, curTick() + ticks(delay));
672 scheduleActivateThreadEvent(tid, delay);
673 } else {
674 activateThread(tid);
675 }
676
677 if (lastActivatedCycle < curTick()) {
678 scheduleTickEvent(delay);
679
680 // Be sure to signal that there's some activity so the CPU doesn't
681 // deschedule itself.
682 activityRec.activity();
683 fetch.wakeFromQuiesce();
684
685 lastActivatedCycle = curTick();
686
687 _status = Running;
688 }
689}
690
691template <class Impl>
692bool
693FullO3CPU<Impl>::deallocateContext(ThreadID tid, bool remove, int delay)
694{
695 // Schedule removal of thread data from CPU
696 if (delay){
697 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate "
698 "on cycle %d\n", tid, curTick() + ticks(delay));
699 scheduleDeallocateContextEvent(tid, remove, delay);
700 return false;
701 } else {
702 deactivateThread(tid);
703 if (remove)
704 removeThread(tid);
705 return true;
706 }
707}
708
709template <class Impl>
710void
711FullO3CPU<Impl>::suspendContext(ThreadID tid)
712{
713 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid);
714 bool deallocated = deallocateContext(tid, false, 1);
715 // If this was the last thread then unschedule the tick event.
716 if ((activeThreads.size() == 1 && !deallocated) ||
717 activeThreads.size() == 0)
718 unscheduleTickEvent();
719 _status = Idle;
720}
721
722template <class Impl>
723void
724FullO3CPU<Impl>::haltContext(ThreadID tid)
725{
726 //For now, this is the same as deallocate
727 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid);
728 deallocateContext(tid, true, 1);
729}
730
731template <class Impl>
732void
733FullO3CPU<Impl>::insertThread(ThreadID tid)
734{
735 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU");
736 // Will change now that the PC and thread state is internal to the CPU
737 // and not in the ThreadContext.
738 ThreadContext *src_tc;
739 if (FullSystem)
740 src_tc = system->threadContexts[tid];
741 else
742 src_tc = tcBase(tid);
743
744 //Bind Int Regs to Rename Map
745 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
746 PhysRegIndex phys_reg = freeList.getIntReg();
747
748 renameMap[tid].setEntry(ireg,phys_reg);
749 scoreboard.setReg(phys_reg);
750 }
751
752 //Bind Float Regs to Rename Map
753 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) {
754 PhysRegIndex phys_reg = freeList.getFloatReg();
755
756 renameMap[tid].setEntry(freg,phys_reg);
757 scoreboard.setReg(phys_reg);
758 }
759
760 //Copy Thread Data Into RegFile
761 //this->copyFromTC(tid);
762
763 //Set PC/NPC/NNPC
764 pcState(src_tc->pcState(), tid);
765
766 src_tc->setStatus(ThreadContext::Active);
767
768 activateContext(tid,1);
769
770 //Reset ROB/IQ/LSQ Entries
771 commit.rob->resetEntries();
772 iew.resetEntries();
773}
774
775template <class Impl>
776void
777FullO3CPU<Impl>::removeThread(ThreadID tid)
778{
779 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid);
780
781 // Copy Thread Data From RegFile
782 // If thread is suspended, it might be re-allocated
783 // this->copyToTC(tid);
784
785
786 // @todo: 2-27-2008: Fix how we free up rename mappings
787 // here to alleviate the case for double-freeing registers
788 // in SMT workloads.
789
790 // Unbind Int Regs from Rename Map
791 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
792 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg);
793
794 scoreboard.unsetReg(phys_reg);
795 freeList.addReg(phys_reg);
796 }
797
798 // Unbind Float Regs from Rename Map
799 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) {
800 PhysRegIndex phys_reg = renameMap[tid].lookup(freg);
801
802 scoreboard.unsetReg(phys_reg);
803 freeList.addReg(phys_reg);
804 }
805
806 // Squash Throughout Pipeline
807 DynInstPtr inst = commit.rob->readHeadInst(tid);
808 InstSeqNum squash_seq_num = inst->seqNum;
809 fetch.squash(0, squash_seq_num, inst, tid);
810 decode.squash(tid);
811 rename.squash(squash_seq_num, tid);
812 iew.squash(tid);
813 iew.ldstQueue.squash(squash_seq_num, tid);
814 commit.rob->squash(squash_seq_num, tid);
815
816
817 assert(iew.instQueue.getCount(tid) == 0);
818 assert(iew.ldstQueue.getCount(tid) == 0);
819
820 // Reset ROB/IQ/LSQ Entries
821
822 // Commented out for now. This should be possible to do by
823 // telling all the pipeline stages to drain first, and then
824 // checking until the drain completes. Once the pipeline is
825 // drained, call resetEntries(). - 10-09-06 ktlim
826/*
827 if (activeThreads.size() >= 1) {
828 commit.rob->resetEntries();
829 iew.resetEntries();
830 }
831*/
832}
833
834
835template <class Impl>
836void
837FullO3CPU<Impl>::activateWhenReady(ThreadID tid)
838{
839 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming"
840 "(e.g. PhysRegs/ROB/IQ/LSQ) \n",
841 tid);
842
843 bool ready = true;
844
845 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) {
846 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
847 "Phys. Int. Regs.\n",
848 tid);
849 ready = false;
850 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) {
851 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
852 "Phys. Float. Regs.\n",
853 tid);
854 ready = false;
855 } else if (commit.rob->numFreeEntries() >=
856 commit.rob->entryAmount(activeThreads.size() + 1)) {
857 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
858 "ROB entries.\n",
859 tid);
860 ready = false;
861 } else if (iew.instQueue.numFreeEntries() >=
862 iew.instQueue.entryAmount(activeThreads.size() + 1)) {
863 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
864 "IQ entries.\n",
865 tid);
866 ready = false;
867 } else if (iew.ldstQueue.numFreeEntries() >=
868 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) {
869 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
870 "LSQ entries.\n",
871 tid);
872 ready = false;
873 }
874
875 if (ready) {
876 insertThread(tid);
877
878 contextSwitch = false;
879
880 cpuWaitList.remove(tid);
881 } else {
882 suspendContext(tid);
883
884 //blocks fetch
885 contextSwitch = true;
886
887 //@todo: dont always add to waitlist
888 //do waitlist
889 cpuWaitList.push_back(tid);
890 }
891}
892
893template <class Impl>
894Fault
895FullO3CPU<Impl>::hwrei(ThreadID tid)
896{
897#if THE_ISA == ALPHA_ISA
898 // Need to clear the lock flag upon returning from an interrupt.
899 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid);
900
901 this->thread[tid]->kernelStats->hwrei();
902
903 // FIXME: XXX check for interrupts? XXX
904#endif
905 return NoFault;
906}
907
908template <class Impl>
909bool
910FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid)
911{
912#if THE_ISA == ALPHA_ISA
913 if (this->thread[tid]->kernelStats)
914 this->thread[tid]->kernelStats->callpal(palFunc,
915 this->threadContexts[tid]);
916
917 switch (palFunc) {
918 case PAL::halt:
919 halt();
920 if (--System::numSystemsRunning == 0)
921 exitSimLoop("all cpus halted");
922 break;
923
924 case PAL::bpt:
925 case PAL::bugchk:
926 if (this->system->breakpoint())
927 return false;
928 break;
929 }
930#endif
931 return true;
932}
933
934template <class Impl>
935Fault
936FullO3CPU<Impl>::getInterrupts()
937{
938 // Check if there are any outstanding interrupts
939 return this->interrupts->getInterrupt(this->threadContexts[0]);
940}
941
942template <class Impl>
943void
944FullO3CPU<Impl>::processInterrupts(Fault interrupt)
945{
946 // Check for interrupts here. For now can copy the code that
947 // exists within isa_fullsys_traits.hh. Also assume that thread 0
948 // is the one that handles the interrupts.
949 // @todo: Possibly consolidate the interrupt checking code.
950 // @todo: Allow other threads to handle interrupts.
951
952 assert(interrupt != NoFault);
953 this->interrupts->updateIntrInfo(this->threadContexts[0]);
954
955 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name());
956 this->trap(interrupt, 0, NULL);
957}
958
959template <class Impl>
960void
961FullO3CPU<Impl>::updateMemPorts()
962{
963 // Update all ThreadContext's memory ports (Functional/Virtual
964 // Ports)
965 ThreadID size = thread.size();
966 for (ThreadID i = 0; i < size; ++i)
967 thread[i]->connectMemPorts(thread[i]->getTC());
968}
969
970template <class Impl>
971void
972FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst)
973{
974 // Pass the thread's TC into the invoke method.
975 fault->invoke(this->threadContexts[tid], inst);
976}
977
978template <class Impl>
979void
980FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid)
981{
982 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid);
983
984 DPRINTF(Activity,"Activity: syscall() called.\n");
985
986 // Temporarily increase this by one to account for the syscall
987 // instruction.
988 ++(this->thread[tid]->funcExeInst);
989
990 // Execute the actual syscall.
991 this->thread[tid]->syscall(callnum);
992
993 // Decrease funcExeInst by one as the normal commit will handle
994 // incrementing it.
995 --(this->thread[tid]->funcExeInst);
996}
997
998template <class Impl>
999void
1000FullO3CPU<Impl>::serialize(std::ostream &os)
1001{
1002 SimObject::State so_state = SimObject::getState();
1003 SERIALIZE_ENUM(so_state);
1004 BaseCPU::serialize(os);
1005 nameOut(os, csprintf("%s.tickEvent", name()));
1006 tickEvent.serialize(os);
1007
1008 // Use SimpleThread's ability to checkpoint to make it easier to
1009 // write out the registers. Also make this static so it doesn't
1010 // get instantiated multiple times (causes a panic in statistics).
1011 static SimpleThread temp;
1012
1013 ThreadID size = thread.size();
1014 for (ThreadID i = 0; i < size; i++) {
1015 nameOut(os, csprintf("%s.xc.%i", name(), i));
1016 temp.copyTC(thread[i]->getTC());
1017 temp.serialize(os);
1018 }
1019}
1020
1021template <class Impl>
1022void
1023FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string &section)
1024{
1025 SimObject::State so_state;
1026 UNSERIALIZE_ENUM(so_state);
1027 BaseCPU::unserialize(cp, section);
1028 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
1029
1030 // Use SimpleThread's ability to checkpoint to make it easier to
1031 // read in the registers. Also make this static so it doesn't
1032 // get instantiated multiple times (causes a panic in statistics).
1033 static SimpleThread temp;
1034
1035 ThreadID size = thread.size();
1036 for (ThreadID i = 0; i < size; i++) {
1037 temp.copyTC(thread[i]->getTC());
1038 temp.unserialize(cp, csprintf("%s.xc.%i", section, i));
1039 thread[i]->getTC()->copyArchRegs(temp.getTC());
1040 }
1041}
1042
1043template <class Impl>
1044unsigned int
1045FullO3CPU<Impl>::drain(Event *drain_event)
1046{
1047 DPRINTF(O3CPU, "Switching out\n");
1048
1049 // If the CPU isn't doing anything, then return immediately.
1050 if (_status == Idle || _status == SwitchedOut) {
1051 return 0;
1052 }
1053
1054 drainCount = 0;
1055 fetch.drain();
1056 decode.drain();
1057 rename.drain();
1058 iew.drain();
1059 commit.drain();
1060
1061 // Wake the CPU and record activity so everything can drain out if
1062 // the CPU was not able to immediately drain.
1063 if (getState() != SimObject::Drained) {
1064 // A bit of a hack...set the drainEvent after all the drain()
1065 // calls have been made, that way if all of the stages drain
1066 // immediately, the signalDrained() function knows not to call
1067 // process on the drain event.
1068 drainEvent = drain_event;
1069
1070 wakeCPU();
1071 activityRec.activity();
1072
1073 return 1;
1074 } else {
1075 return 0;
1076 }
1077}
1078
1079template <class Impl>
1080void
1081FullO3CPU<Impl>::resume()
1082{
1083 fetch.resume();
1084 decode.resume();
1085 rename.resume();
1086 iew.resume();
1087 commit.resume();
1088
1089 changeState(SimObject::Running);
1090
1091 if (_status == SwitchedOut || _status == Idle)
1092 return;
1093
1094 assert(system->getMemoryMode() == Enums::timing);
1095
1096 if (!tickEvent.scheduled())
1097 schedule(tickEvent, nextCycle());
1098 _status = Running;
1099}
1100
1101template <class Impl>
1102void
1103FullO3CPU<Impl>::signalDrained()
1104{
1105 if (++drainCount == NumStages) {
1106 if (tickEvent.scheduled())
1107 tickEvent.squash();
1108
1109 changeState(SimObject::Drained);
1110
1111 BaseCPU::switchOut();
1112
1113 if (drainEvent) {
1114 drainEvent->process();
1115 drainEvent = NULL;
1116 }
1117 }
1118 assert(drainCount <= 5);
1119}
1120
1121template <class Impl>
1122void
1123FullO3CPU<Impl>::switchOut()
1124{
1125 fetch.switchOut();
1126 rename.switchOut();
1127 iew.switchOut();
1128 commit.switchOut();
1129 instList.clear();
1130 while (!removeList.empty()) {
1131 removeList.pop();
1132 }
1133
1134 _status = SwitchedOut;
1135#if USE_CHECKER
1136 if (checker)
1137 checker->switchOut();
1138#endif
1139 if (tickEvent.scheduled())
1140 tickEvent.squash();
1141}
1142
1143template <class Impl>
1144void
1145FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU)
1146{
1147 // Flush out any old data from the time buffers.
1148 for (int i = 0; i < timeBuffer.getSize(); ++i) {
1149 timeBuffer.advance();
1150 fetchQueue.advance();
1151 decodeQueue.advance();
1152 renameQueue.advance();
1153 iewQueue.advance();
1154 }
1155
1156 activityRec.reset();
1157
1158 BaseCPU::takeOverFrom(oldCPU, fetch.getIcachePort(), iew.getDcachePort());
1159
1160 fetch.takeOverFrom();
1161 decode.takeOverFrom();
1162 rename.takeOverFrom();
1163 iew.takeOverFrom();
1164 commit.takeOverFrom();
1165
1166 assert(!tickEvent.scheduled() || tickEvent.squashed());
1167
1168 // @todo: Figure out how to properly select the tid to put onto
1169 // the active threads list.
1170 ThreadID tid = 0;
1171
1172 list<ThreadID>::iterator isActive =
1173 std::find(activeThreads.begin(), activeThreads.end(), tid);
1174
1175 if (isActive == activeThreads.end()) {
1176 //May Need to Re-code this if the delay variable is the delay
1177 //needed for thread to activate
1178 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n",
1179 tid);
1180
1181 activeThreads.push_back(tid);
1182 }
1183
1184 // Set all statuses to active, schedule the CPU's tick event.
1185 // @todo: Fix up statuses so this is handled properly
1186 ThreadID size = threadContexts.size();
1187 for (ThreadID i = 0; i < size; ++i) {
1188 ThreadContext *tc = threadContexts[i];
1189 if (tc->status() == ThreadContext::Active && _status != Running) {
1190 _status = Running;
1191 reschedule(tickEvent, nextCycle(), true);
1192 }
1193 }
1194 if (!tickEvent.scheduled())
1195 schedule(tickEvent, nextCycle());
1196}
1197
1198template <class Impl>
1199TheISA::MiscReg
1200FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid)
1201{
1202 return this->isa[tid].readMiscRegNoEffect(misc_reg);
1203}
1204
1205template <class Impl>
1206TheISA::MiscReg
1207FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid)
1208{
1209 miscRegfileReads++;
1210 return this->isa[tid].readMiscReg(misc_reg, tcBase(tid));
1211}
1212
1213template <class Impl>
1214void
1215FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg,
1216 const TheISA::MiscReg &val, ThreadID tid)
1217{
1218 this->isa[tid].setMiscRegNoEffect(misc_reg, val);
1219}
1220
1221template <class Impl>
1222void
1223FullO3CPU<Impl>::setMiscReg(int misc_reg,
1224 const TheISA::MiscReg &val, ThreadID tid)
1225{
1226 miscRegfileWrites++;
1227 this->isa[tid].setMiscReg(misc_reg, val, tcBase(tid));
1228}
1229
1230template <class Impl>
1231uint64_t
1232FullO3CPU<Impl>::readIntReg(int reg_idx)
1233{
1234 intRegfileReads++;
1235 return regFile.readIntReg(reg_idx);
1236}
1237
1238template <class Impl>
1239FloatReg
1240FullO3CPU<Impl>::readFloatReg(int reg_idx)
1241{
1242 fpRegfileReads++;
1243 return regFile.readFloatReg(reg_idx);
1244}
1245
1246template <class Impl>
1247FloatRegBits
1248FullO3CPU<Impl>::readFloatRegBits(int reg_idx)
1249{
1250 fpRegfileReads++;
1251 return regFile.readFloatRegBits(reg_idx);
1252}
1253
1254template <class Impl>
1255void
1256FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val)
1257{
1258 intRegfileWrites++;
1259 regFile.setIntReg(reg_idx, val);
1260}
1261
1262template <class Impl>
1263void
1264FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val)
1265{
1266 fpRegfileWrites++;
1267 regFile.setFloatReg(reg_idx, val);
1268}
1269
1270template <class Impl>
1271void
1272FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val)
1273{
1274 fpRegfileWrites++;
1275 regFile.setFloatRegBits(reg_idx, val);
1276}
1277
1278template <class Impl>
1279uint64_t
1280FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid)
1281{
1282 intRegfileReads++;
1283 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1284
1285 return regFile.readIntReg(phys_reg);
1286}
1287
1288template <class Impl>
1289float
1290FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid)
1291{
1292 fpRegfileReads++;
1293 int idx = reg_idx + TheISA::NumIntRegs;
1294 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1295
1296 return regFile.readFloatReg(phys_reg);
1297}
1298
1299template <class Impl>
1300uint64_t
1301FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid)
1302{
1303 fpRegfileReads++;
1304 int idx = reg_idx + TheISA::NumIntRegs;
1305 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1306
1307 return regFile.readFloatRegBits(phys_reg);
1308}
1309
1310template <class Impl>
1311void
1312FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid)
1313{
1314 intRegfileWrites++;
1315 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1316
1317 regFile.setIntReg(phys_reg, val);
1318}
1319
1320template <class Impl>
1321void
1322FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid)
1323{
1324 fpRegfileWrites++;
1325 int idx = reg_idx + TheISA::NumIntRegs;
1326 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1327
1328 regFile.setFloatReg(phys_reg, val);
1329}
1330
1331template <class Impl>
1332void
1333FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid)
1334{
1335 fpRegfileWrites++;
1336 int idx = reg_idx + TheISA::NumIntRegs;
1337 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1338
1339 regFile.setFloatRegBits(phys_reg, val);
1340}
1341
1342template <class Impl>
1343TheISA::PCState
1344FullO3CPU<Impl>::pcState(ThreadID tid)
1345{
1346 return commit.pcState(tid);
1347}
1348
1349template <class Impl>
1350void
1351FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid)
1352{
1353 commit.pcState(val, tid);
1354}
1355
1356template <class Impl>
1357Addr
1358FullO3CPU<Impl>::instAddr(ThreadID tid)
1359{
1360 return commit.instAddr(tid);
1361}
1362
1363template <class Impl>
1364Addr
1365FullO3CPU<Impl>::nextInstAddr(ThreadID tid)
1366{
1367 return commit.nextInstAddr(tid);
1368}
1369
1370template <class Impl>
1371MicroPC
1372FullO3CPU<Impl>::microPC(ThreadID tid)
1373{
1374 return commit.microPC(tid);
1375}
1376
1377template <class Impl>
1378void
1379FullO3CPU<Impl>::squashFromTC(ThreadID tid)
1380{
1381 this->thread[tid]->inSyscall = true;
1382 this->commit.generateTCEvent(tid);
1383}
1384
1385template <class Impl>
1386typename FullO3CPU<Impl>::ListIt
1387FullO3CPU<Impl>::addInst(DynInstPtr &inst)
1388{
1389 instList.push_back(inst);
1390
1391 return --(instList.end());
1392}
1393
1394template <class Impl>
1395void
1396FullO3CPU<Impl>::instDone(ThreadID tid)
1397{
1398 // Keep an instruction count.
1399 thread[tid]->numInst++;
1400 thread[tid]->numInsts++;
1401 committedInsts[tid]++;
1402 totalCommittedInsts++;
1403 system->totalNumInsts++;
1404 // Check for instruction-count-based events.
1405 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst);
1406 system->instEventQueue.serviceEvents(system->totalNumInsts);
1407}
1408
1409template <class Impl>
1410void
1411FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst)
1412{
1413 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s "
1414 "[sn:%lli]\n",
1415 inst->threadNumber, inst->pcState(), inst->seqNum);
1416
1417 removeInstsThisCycle = true;
1418
1419 // Remove the front instruction.
1420 removeList.push(inst->getInstListIt());
1421}
1422
1423template <class Impl>
1424void
1425FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid)
1426{
1427 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction"
1428 " list.\n", tid);
1429
1430 ListIt end_it;
1431
1432 bool rob_empty = false;
1433
1434 if (instList.empty()) {
1435 return;
1436 } else if (rob.isEmpty(/*tid*/)) {
1437 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n");
1438 end_it = instList.begin();
1439 rob_empty = true;
1440 } else {
1441 end_it = (rob.readTailInst(tid))->getInstListIt();
1442 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n");
1443 }
1444
1445 removeInstsThisCycle = true;
1446
1447 ListIt inst_it = instList.end();
1448
1449 inst_it--;
1450
1451 // Walk through the instruction list, removing any instructions
1452 // that were inserted after the given instruction iterator, end_it.
1453 while (inst_it != end_it) {
1454 assert(!instList.empty());
1455
1456 squashInstIt(inst_it, tid);
1457
1458 inst_it--;
1459 }
1460
1461 // If the ROB was empty, then we actually need to remove the first
1462 // instruction as well.
1463 if (rob_empty) {
1464 squashInstIt(inst_it, tid);
1465 }
1466}
1467
1468template <class Impl>
1469void
1470FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid)
1471{
1472 assert(!instList.empty());
1473
1474 removeInstsThisCycle = true;
1475
1476 ListIt inst_iter = instList.end();
1477
1478 inst_iter--;
1479
1480 DPRINTF(O3CPU, "Deleting instructions from instruction "
1481 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
1482 tid, seq_num, (*inst_iter)->seqNum);
1483
1484 while ((*inst_iter)->seqNum > seq_num) {
1485
1486 bool break_loop = (inst_iter == instList.begin());
1487
1488 squashInstIt(inst_iter, tid);
1489
1490 inst_iter--;
1491
1492 if (break_loop)
1493 break;
1494 }
1495}
1496
1497template <class Impl>
1498inline void
1499FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid)
1500{
1501 if ((*instIt)->threadNumber == tid) {
1502 DPRINTF(O3CPU, "Squashing instruction, "
1503 "[tid:%i] [sn:%lli] PC %s\n",
1504 (*instIt)->threadNumber,
1505 (*instIt)->seqNum,
1506 (*instIt)->pcState());
1507
1508 // Mark it as squashed.
1509 (*instIt)->setSquashed();
1510
1511 // @todo: Formulate a consistent method for deleting
1512 // instructions from the instruction list
1513 // Remove the instruction from the list.
1514 removeList.push(instIt);
1515 }
1516}
1517
1518template <class Impl>
1519void
1520FullO3CPU<Impl>::cleanUpRemovedInsts()
1521{
1522 while (!removeList.empty()) {
1523 DPRINTF(O3CPU, "Removing instruction, "
1524 "[tid:%i] [sn:%lli] PC %s\n",
1525 (*removeList.front())->threadNumber,
1526 (*removeList.front())->seqNum,
1527 (*removeList.front())->pcState());
1528
1529 instList.erase(removeList.front());
1530
1531 removeList.pop();
1532 }
1533
1534 removeInstsThisCycle = false;
1535}
1536/*
1537template <class Impl>
1538void
1539FullO3CPU<Impl>::removeAllInsts()
1540{
1541 instList.clear();
1542}
1543*/
1544template <class Impl>
1545void
1546FullO3CPU<Impl>::dumpInsts()
1547{
1548 int num = 0;
1549
1550 ListIt inst_list_it = instList.begin();
1551
1552 cprintf("Dumping Instruction List\n");
1553
1554 while (inst_list_it != instList.end()) {
1555 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
1556 "Squashed:%i\n\n",
1557 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber,
1558 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(),
1559 (*inst_list_it)->isSquashed());
1560 inst_list_it++;
1561 ++num;
1562 }
1563}
1564/*
1565template <class Impl>
1566void
1567FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst)
1568{
1569 iew.wakeDependents(inst);
1570}
1571*/
1572template <class Impl>
1573void
1574FullO3CPU<Impl>::wakeCPU()
1575{
1576 if (activityRec.active() || tickEvent.scheduled()) {
1577 DPRINTF(Activity, "CPU already running.\n");
1578 return;
1579 }
1580
1581 DPRINTF(Activity, "Waking up CPU\n");
1582
1583 idleCycles += tickToCycles((curTick() - 1) - lastRunningCycle);
1584 numCycles += tickToCycles((curTick() - 1) - lastRunningCycle);
1585
1586 schedule(tickEvent, nextCycle());
1587}
1588
1589template <class Impl>
1590void
1591FullO3CPU<Impl>::wakeup()
1592{
1593 if (this->thread[0]->status() != ThreadContext::Suspended)
1594 return;
1595
1596 this->wakeCPU();
1597
1598 DPRINTF(Quiesce, "Suspended Processor woken\n");
1599 this->threadContexts[0]->activate();
1600}
1601
1602template <class Impl>
1603ThreadID
1604FullO3CPU<Impl>::getFreeTid()
1605{
1606 for (ThreadID tid = 0; tid < numThreads; tid++) {
1607 if (!tids[tid]) {
1608 tids[tid] = true;
1609 return tid;
1610 }
1611 }
1612
1613 return InvalidThreadID;
1614}
1615
1616template <class Impl>
1617void
1618FullO3CPU<Impl>::doContextSwitch()
1619{
1620 if (contextSwitch) {
1621
1622 //ADD CODE TO DEACTIVE THREAD HERE (???)
1623
1624 ThreadID size = cpuWaitList.size();
1625 for (ThreadID tid = 0; tid < size; tid++) {
1626 activateWhenReady(tid);
1627 }
1628
1629 if (cpuWaitList.size() == 0)
1630 contextSwitch = true;
1631 }
1632}
1633
1634template <class Impl>
1635void
1636FullO3CPU<Impl>::updateThreadPriority()
1637{
1638 if (activeThreads.size() > 1) {
1639 //DEFAULT TO ROUND ROBIN SCHEME
1640 //e.g. Move highest priority to end of thread list
1641 list<ThreadID>::iterator list_begin = activeThreads.begin();
1/*
2 * Copyright (c) 2004-2006 The Regents of The University of Michigan
3 * Copyright (c) 2011 Regents of the University of California
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions are
8 * met: redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer;
10 * redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution;
13 * neither the name of the copyright holders nor the names of its
14 * contributors may be used to endorse or promote products derived from
15 * this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
18 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
19 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
20 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
21 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
22 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
23 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
29 * Authors: Kevin Lim
30 * Korey Sewell
31 * Rick Strong
32 */
33
34#include "arch/kernel_stats.hh"
35#include "config/the_isa.hh"
36#include "config/use_checker.hh"
37#include "cpu/o3/cpu.hh"
38#include "cpu/o3/isa_specific.hh"
39#include "cpu/o3/thread_context.hh"
40#include "cpu/activity.hh"
41#include "cpu/quiesce_event.hh"
42#include "cpu/simple_thread.hh"
43#include "cpu/thread_context.hh"
44#include "debug/Activity.hh"
45#include "debug/O3CPU.hh"
46#include "debug/Quiesce.hh"
47#include "enums/MemoryMode.hh"
48#include "sim/core.hh"
49#include "sim/full_system.hh"
50#include "sim/process.hh"
51#include "sim/stat_control.hh"
52#include "sim/system.hh"
53
54#if USE_CHECKER
55#include "cpu/checker/cpu.hh"
56#endif
57
58#if THE_ISA == ALPHA_ISA
59#include "arch/alpha/osfpal.hh"
60#include "debug/Activity.hh"
61#endif
62
63class BaseCPUParams;
64
65using namespace TheISA;
66using namespace std;
67
68BaseO3CPU::BaseO3CPU(BaseCPUParams *params)
69 : BaseCPU(params)
70{
71}
72
73void
74BaseO3CPU::regStats()
75{
76 BaseCPU::regStats();
77}
78
79template <class Impl>
80FullO3CPU<Impl>::TickEvent::TickEvent(FullO3CPU<Impl> *c)
81 : Event(CPU_Tick_Pri), cpu(c)
82{
83}
84
85template <class Impl>
86void
87FullO3CPU<Impl>::TickEvent::process()
88{
89 cpu->tick();
90}
91
92template <class Impl>
93const char *
94FullO3CPU<Impl>::TickEvent::description() const
95{
96 return "FullO3CPU tick";
97}
98
99template <class Impl>
100FullO3CPU<Impl>::ActivateThreadEvent::ActivateThreadEvent()
101 : Event(CPU_Switch_Pri)
102{
103}
104
105template <class Impl>
106void
107FullO3CPU<Impl>::ActivateThreadEvent::init(int thread_num,
108 FullO3CPU<Impl> *thread_cpu)
109{
110 tid = thread_num;
111 cpu = thread_cpu;
112}
113
114template <class Impl>
115void
116FullO3CPU<Impl>::ActivateThreadEvent::process()
117{
118 cpu->activateThread(tid);
119}
120
121template <class Impl>
122const char *
123FullO3CPU<Impl>::ActivateThreadEvent::description() const
124{
125 return "FullO3CPU \"Activate Thread\"";
126}
127
128template <class Impl>
129FullO3CPU<Impl>::DeallocateContextEvent::DeallocateContextEvent()
130 : Event(CPU_Tick_Pri), tid(0), remove(false), cpu(NULL)
131{
132}
133
134template <class Impl>
135void
136FullO3CPU<Impl>::DeallocateContextEvent::init(int thread_num,
137 FullO3CPU<Impl> *thread_cpu)
138{
139 tid = thread_num;
140 cpu = thread_cpu;
141 remove = false;
142}
143
144template <class Impl>
145void
146FullO3CPU<Impl>::DeallocateContextEvent::process()
147{
148 cpu->deactivateThread(tid);
149 if (remove)
150 cpu->removeThread(tid);
151}
152
153template <class Impl>
154const char *
155FullO3CPU<Impl>::DeallocateContextEvent::description() const
156{
157 return "FullO3CPU \"Deallocate Context\"";
158}
159
160template <class Impl>
161FullO3CPU<Impl>::FullO3CPU(DerivO3CPUParams *params)
162 : BaseO3CPU(params),
163 itb(params->itb),
164 dtb(params->dtb),
165 tickEvent(this),
166#ifndef NDEBUG
167 instcount(0),
168#endif
169 removeInstsThisCycle(false),
170 fetch(this, params),
171 decode(this, params),
172 rename(this, params),
173 iew(this, params),
174 commit(this, params),
175
176 regFile(this, params->numPhysIntRegs,
177 params->numPhysFloatRegs),
178
179 freeList(params->numThreads,
180 TheISA::NumIntRegs, params->numPhysIntRegs,
181 TheISA::NumFloatRegs, params->numPhysFloatRegs),
182
183 rob(this,
184 params->numROBEntries, params->squashWidth,
185 params->smtROBPolicy, params->smtROBThreshold,
186 params->numThreads),
187
188 scoreboard(params->numThreads,
189 TheISA::NumIntRegs, params->numPhysIntRegs,
190 TheISA::NumFloatRegs, params->numPhysFloatRegs,
191 TheISA::NumMiscRegs * numThreads,
192 TheISA::ZeroReg),
193
194 timeBuffer(params->backComSize, params->forwardComSize),
195 fetchQueue(params->backComSize, params->forwardComSize),
196 decodeQueue(params->backComSize, params->forwardComSize),
197 renameQueue(params->backComSize, params->forwardComSize),
198 iewQueue(params->backComSize, params->forwardComSize),
199 activityRec(name(), NumStages,
200 params->backComSize + params->forwardComSize,
201 params->activity),
202
203 globalSeqNum(1),
204 system(params->system),
205 drainCount(0),
206 deferRegistration(params->defer_registration)
207{
208 if (!deferRegistration) {
209 _status = Running;
210 } else {
211 _status = Idle;
212 }
213
214#if USE_CHECKER
215 if (params->checker) {
216 BaseCPU *temp_checker = params->checker;
217 checker = dynamic_cast<Checker<DynInstPtr> *>(temp_checker);
218 checker->setSystem(params->system);
219 } else {
220 checker = NULL;
221 }
222#endif // USE_CHECKER
223
224 if (!FullSystem) {
225 thread.resize(numThreads);
226 tids.resize(numThreads);
227 }
228
229 // The stages also need their CPU pointer setup. However this
230 // must be done at the upper level CPU because they have pointers
231 // to the upper level CPU, and not this FullO3CPU.
232
233 // Set up Pointers to the activeThreads list for each stage
234 fetch.setActiveThreads(&activeThreads);
235 decode.setActiveThreads(&activeThreads);
236 rename.setActiveThreads(&activeThreads);
237 iew.setActiveThreads(&activeThreads);
238 commit.setActiveThreads(&activeThreads);
239
240 // Give each of the stages the time buffer they will use.
241 fetch.setTimeBuffer(&timeBuffer);
242 decode.setTimeBuffer(&timeBuffer);
243 rename.setTimeBuffer(&timeBuffer);
244 iew.setTimeBuffer(&timeBuffer);
245 commit.setTimeBuffer(&timeBuffer);
246
247 // Also setup each of the stages' queues.
248 fetch.setFetchQueue(&fetchQueue);
249 decode.setFetchQueue(&fetchQueue);
250 commit.setFetchQueue(&fetchQueue);
251 decode.setDecodeQueue(&decodeQueue);
252 rename.setDecodeQueue(&decodeQueue);
253 rename.setRenameQueue(&renameQueue);
254 iew.setRenameQueue(&renameQueue);
255 iew.setIEWQueue(&iewQueue);
256 commit.setIEWQueue(&iewQueue);
257 commit.setRenameQueue(&renameQueue);
258
259 commit.setIEWStage(&iew);
260 rename.setIEWStage(&iew);
261 rename.setCommitStage(&commit);
262
263 ThreadID active_threads;
264 if (FullSystem) {
265 active_threads = 1;
266 } else {
267 active_threads = params->workload.size();
268
269 if (active_threads > Impl::MaxThreads) {
270 panic("Workload Size too large. Increase the 'MaxThreads' "
271 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) "
272 "or edit your workload size.");
273 }
274 }
275
276 //Make Sure That this a Valid Architeture
277 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs);
278 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs);
279
280 rename.setScoreboard(&scoreboard);
281 iew.setScoreboard(&scoreboard);
282
283 // Setup the rename map for whichever stages need it.
284 PhysRegIndex lreg_idx = 0;
285 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs
286
287 for (ThreadID tid = 0; tid < numThreads; tid++) {
288 bool bindRegs = (tid <= active_threads - 1);
289
290 commitRenameMap[tid].init(TheISA::NumIntRegs,
291 params->numPhysIntRegs,
292 lreg_idx, //Index for Logical. Regs
293
294 TheISA::NumFloatRegs,
295 params->numPhysFloatRegs,
296 freg_idx, //Index for Float Regs
297
298 TheISA::NumMiscRegs,
299
300 TheISA::ZeroReg,
301 TheISA::ZeroReg,
302
303 tid,
304 false);
305
306 renameMap[tid].init(TheISA::NumIntRegs,
307 params->numPhysIntRegs,
308 lreg_idx, //Index for Logical. Regs
309
310 TheISA::NumFloatRegs,
311 params->numPhysFloatRegs,
312 freg_idx, //Index for Float Regs
313
314 TheISA::NumMiscRegs,
315
316 TheISA::ZeroReg,
317 TheISA::ZeroReg,
318
319 tid,
320 bindRegs);
321
322 activateThreadEvent[tid].init(tid, this);
323 deallocateContextEvent[tid].init(tid, this);
324 }
325
326 rename.setRenameMap(renameMap);
327 commit.setRenameMap(commitRenameMap);
328
329 // Give renameMap & rename stage access to the freeList;
330 for (ThreadID tid = 0; tid < numThreads; tid++)
331 renameMap[tid].setFreeList(&freeList);
332 rename.setFreeList(&freeList);
333
334 // Setup the ROB for whichever stages need it.
335 commit.setROB(&rob);
336
337 lastRunningCycle = curTick();
338
339 lastActivatedCycle = -1;
340#if 0
341 // Give renameMap & rename stage access to the freeList;
342 for (ThreadID tid = 0; tid < numThreads; tid++)
343 globalSeqNum[tid] = 1;
344#endif
345
346 contextSwitch = false;
347 DPRINTF(O3CPU, "Creating O3CPU object.\n");
348
349 // Setup any thread state.
350 this->thread.resize(this->numThreads);
351
352 for (ThreadID tid = 0; tid < this->numThreads; ++tid) {
353 if (FullSystem) {
354 // SMT is not supported in FS mode yet.
355 assert(this->numThreads == 1);
356 this->thread[tid] = new Thread(this, 0, NULL);
357 } else {
358 if (tid < params->workload.size()) {
359 DPRINTF(O3CPU, "Workload[%i] process is %#x",
360 tid, this->thread[tid]);
361 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
362 (typename Impl::O3CPU *)(this),
363 tid, params->workload[tid]);
364
365 //usedTids[tid] = true;
366 //threadMap[tid] = tid;
367 } else {
368 //Allocate Empty thread so M5 can use later
369 //when scheduling threads to CPU
370 Process* dummy_proc = NULL;
371
372 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
373 (typename Impl::O3CPU *)(this),
374 tid, dummy_proc);
375 //usedTids[tid] = false;
376 }
377 }
378
379 ThreadContext *tc;
380
381 // Setup the TC that will serve as the interface to the threads/CPU.
382 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>;
383
384 tc = o3_tc;
385
386 // If we're using a checker, then the TC should be the
387 // CheckerThreadContext.
388#if USE_CHECKER
389 if (params->checker) {
390 tc = new CheckerThreadContext<O3ThreadContext<Impl> >(
391 o3_tc, this->checker);
392 }
393#endif
394
395 o3_tc->cpu = (typename Impl::O3CPU *)(this);
396 assert(o3_tc->cpu);
397 o3_tc->thread = this->thread[tid];
398
399 if (FullSystem) {
400 // Setup quiesce event.
401 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc);
402 }
403 // Give the thread the TC.
404 this->thread[tid]->tc = tc;
405
406 // Add the TC to the CPU's list of TC's.
407 this->threadContexts.push_back(tc);
408 }
409
410 for (ThreadID tid = 0; tid < this->numThreads; tid++)
411 this->thread[tid]->setFuncExeInst(0);
412
413 lockAddr = 0;
414 lockFlag = false;
415}
416
417template <class Impl>
418FullO3CPU<Impl>::~FullO3CPU()
419{
420}
421
422template <class Impl>
423void
424FullO3CPU<Impl>::regStats()
425{
426 BaseO3CPU::regStats();
427
428 // Register any of the O3CPU's stats here.
429 timesIdled
430 .name(name() + ".timesIdled")
431 .desc("Number of times that the entire CPU went into an idle state and"
432 " unscheduled itself")
433 .prereq(timesIdled);
434
435 idleCycles
436 .name(name() + ".idleCycles")
437 .desc("Total number of cycles that the CPU has spent unscheduled due "
438 "to idling")
439 .prereq(idleCycles);
440
441 // Number of Instructions simulated
442 // --------------------------------
443 // Should probably be in Base CPU but need templated
444 // MaxThreads so put in here instead
445 committedInsts
446 .init(numThreads)
447 .name(name() + ".committedInsts")
448 .desc("Number of Instructions Simulated");
449
450 totalCommittedInsts
451 .name(name() + ".committedInsts_total")
452 .desc("Number of Instructions Simulated");
453
454 cpi
455 .name(name() + ".cpi")
456 .desc("CPI: Cycles Per Instruction")
457 .precision(6);
458 cpi = numCycles / committedInsts;
459
460 totalCpi
461 .name(name() + ".cpi_total")
462 .desc("CPI: Total CPI of All Threads")
463 .precision(6);
464 totalCpi = numCycles / totalCommittedInsts;
465
466 ipc
467 .name(name() + ".ipc")
468 .desc("IPC: Instructions Per Cycle")
469 .precision(6);
470 ipc = committedInsts / numCycles;
471
472 totalIpc
473 .name(name() + ".ipc_total")
474 .desc("IPC: Total IPC of All Threads")
475 .precision(6);
476 totalIpc = totalCommittedInsts / numCycles;
477
478 this->fetch.regStats();
479 this->decode.regStats();
480 this->rename.regStats();
481 this->iew.regStats();
482 this->commit.regStats();
483 this->rob.regStats();
484
485 intRegfileReads
486 .name(name() + ".int_regfile_reads")
487 .desc("number of integer regfile reads")
488 .prereq(intRegfileReads);
489
490 intRegfileWrites
491 .name(name() + ".int_regfile_writes")
492 .desc("number of integer regfile writes")
493 .prereq(intRegfileWrites);
494
495 fpRegfileReads
496 .name(name() + ".fp_regfile_reads")
497 .desc("number of floating regfile reads")
498 .prereq(fpRegfileReads);
499
500 fpRegfileWrites
501 .name(name() + ".fp_regfile_writes")
502 .desc("number of floating regfile writes")
503 .prereq(fpRegfileWrites);
504
505 miscRegfileReads
506 .name(name() + ".misc_regfile_reads")
507 .desc("number of misc regfile reads")
508 .prereq(miscRegfileReads);
509
510 miscRegfileWrites
511 .name(name() + ".misc_regfile_writes")
512 .desc("number of misc regfile writes")
513 .prereq(miscRegfileWrites);
514}
515
516template <class Impl>
517Port *
518FullO3CPU<Impl>::getPort(const std::string &if_name, int idx)
519{
520 if (if_name == "dcache_port")
521 return iew.getDcachePort();
522 else if (if_name == "icache_port")
523 return fetch.getIcachePort();
524 else
525 panic("No Such Port\n");
526}
527
528template <class Impl>
529void
530FullO3CPU<Impl>::tick()
531{
532 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n");
533
534 ++numCycles;
535
536// activity = false;
537
538 //Tick each of the stages
539 fetch.tick();
540
541 decode.tick();
542
543 rename.tick();
544
545 iew.tick();
546
547 commit.tick();
548
549 if (!FullSystem)
550 doContextSwitch();
551
552 // Now advance the time buffers
553 timeBuffer.advance();
554
555 fetchQueue.advance();
556 decodeQueue.advance();
557 renameQueue.advance();
558 iewQueue.advance();
559
560 activityRec.advance();
561
562 if (removeInstsThisCycle) {
563 cleanUpRemovedInsts();
564 }
565
566 if (!tickEvent.scheduled()) {
567 if (_status == SwitchedOut ||
568 getState() == SimObject::Drained) {
569 DPRINTF(O3CPU, "Switched out!\n");
570 // increment stat
571 lastRunningCycle = curTick();
572 } else if (!activityRec.active() || _status == Idle) {
573 DPRINTF(O3CPU, "Idle!\n");
574 lastRunningCycle = curTick();
575 timesIdled++;
576 } else {
577 schedule(tickEvent, nextCycle(curTick() + ticks(1)));
578 DPRINTF(O3CPU, "Scheduling next tick!\n");
579 }
580 }
581
582 if (!FullSystem)
583 updateThreadPriority();
584}
585
586template <class Impl>
587void
588FullO3CPU<Impl>::init()
589{
590 BaseCPU::init();
591
592 // Set inSyscall so that the CPU doesn't squash when initially
593 // setting up registers.
594 for (ThreadID tid = 0; tid < numThreads; ++tid)
595 thread[tid]->inSyscall = true;
596
597 if (FullSystem) {
598 for (ThreadID tid = 0; tid < numThreads; tid++) {
599 ThreadContext *src_tc = threadContexts[tid];
600 TheISA::initCPU(src_tc, src_tc->contextId());
601 }
602 }
603
604 // Clear inSyscall.
605 for (int tid = 0; tid < numThreads; ++tid)
606 thread[tid]->inSyscall = false;
607
608 // Initialize stages.
609 fetch.initStage();
610 iew.initStage();
611 rename.initStage();
612 commit.initStage();
613
614 commit.setThreads(thread);
615}
616
617template <class Impl>
618void
619FullO3CPU<Impl>::activateThread(ThreadID tid)
620{
621 list<ThreadID>::iterator isActive =
622 std::find(activeThreads.begin(), activeThreads.end(), tid);
623
624 DPRINTF(O3CPU, "[tid:%i]: Calling activate thread.\n", tid);
625
626 if (isActive == activeThreads.end()) {
627 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n",
628 tid);
629
630 activeThreads.push_back(tid);
631 }
632}
633
634template <class Impl>
635void
636FullO3CPU<Impl>::deactivateThread(ThreadID tid)
637{
638 //Remove From Active List, if Active
639 list<ThreadID>::iterator thread_it =
640 std::find(activeThreads.begin(), activeThreads.end(), tid);
641
642 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid);
643
644 if (thread_it != activeThreads.end()) {
645 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n",
646 tid);
647 activeThreads.erase(thread_it);
648 }
649}
650
651template <class Impl>
652Counter
653FullO3CPU<Impl>::totalInstructions() const
654{
655 Counter total(0);
656
657 ThreadID size = thread.size();
658 for (ThreadID i = 0; i < size; i++)
659 total += thread[i]->numInst;
660
661 return total;
662}
663
664template <class Impl>
665void
666FullO3CPU<Impl>::activateContext(ThreadID tid, int delay)
667{
668 // Needs to set each stage to running as well.
669 if (delay){
670 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate "
671 "on cycle %d\n", tid, curTick() + ticks(delay));
672 scheduleActivateThreadEvent(tid, delay);
673 } else {
674 activateThread(tid);
675 }
676
677 if (lastActivatedCycle < curTick()) {
678 scheduleTickEvent(delay);
679
680 // Be sure to signal that there's some activity so the CPU doesn't
681 // deschedule itself.
682 activityRec.activity();
683 fetch.wakeFromQuiesce();
684
685 lastActivatedCycle = curTick();
686
687 _status = Running;
688 }
689}
690
691template <class Impl>
692bool
693FullO3CPU<Impl>::deallocateContext(ThreadID tid, bool remove, int delay)
694{
695 // Schedule removal of thread data from CPU
696 if (delay){
697 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate "
698 "on cycle %d\n", tid, curTick() + ticks(delay));
699 scheduleDeallocateContextEvent(tid, remove, delay);
700 return false;
701 } else {
702 deactivateThread(tid);
703 if (remove)
704 removeThread(tid);
705 return true;
706 }
707}
708
709template <class Impl>
710void
711FullO3CPU<Impl>::suspendContext(ThreadID tid)
712{
713 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid);
714 bool deallocated = deallocateContext(tid, false, 1);
715 // If this was the last thread then unschedule the tick event.
716 if ((activeThreads.size() == 1 && !deallocated) ||
717 activeThreads.size() == 0)
718 unscheduleTickEvent();
719 _status = Idle;
720}
721
722template <class Impl>
723void
724FullO3CPU<Impl>::haltContext(ThreadID tid)
725{
726 //For now, this is the same as deallocate
727 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid);
728 deallocateContext(tid, true, 1);
729}
730
731template <class Impl>
732void
733FullO3CPU<Impl>::insertThread(ThreadID tid)
734{
735 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU");
736 // Will change now that the PC and thread state is internal to the CPU
737 // and not in the ThreadContext.
738 ThreadContext *src_tc;
739 if (FullSystem)
740 src_tc = system->threadContexts[tid];
741 else
742 src_tc = tcBase(tid);
743
744 //Bind Int Regs to Rename Map
745 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
746 PhysRegIndex phys_reg = freeList.getIntReg();
747
748 renameMap[tid].setEntry(ireg,phys_reg);
749 scoreboard.setReg(phys_reg);
750 }
751
752 //Bind Float Regs to Rename Map
753 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) {
754 PhysRegIndex phys_reg = freeList.getFloatReg();
755
756 renameMap[tid].setEntry(freg,phys_reg);
757 scoreboard.setReg(phys_reg);
758 }
759
760 //Copy Thread Data Into RegFile
761 //this->copyFromTC(tid);
762
763 //Set PC/NPC/NNPC
764 pcState(src_tc->pcState(), tid);
765
766 src_tc->setStatus(ThreadContext::Active);
767
768 activateContext(tid,1);
769
770 //Reset ROB/IQ/LSQ Entries
771 commit.rob->resetEntries();
772 iew.resetEntries();
773}
774
775template <class Impl>
776void
777FullO3CPU<Impl>::removeThread(ThreadID tid)
778{
779 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid);
780
781 // Copy Thread Data From RegFile
782 // If thread is suspended, it might be re-allocated
783 // this->copyToTC(tid);
784
785
786 // @todo: 2-27-2008: Fix how we free up rename mappings
787 // here to alleviate the case for double-freeing registers
788 // in SMT workloads.
789
790 // Unbind Int Regs from Rename Map
791 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
792 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg);
793
794 scoreboard.unsetReg(phys_reg);
795 freeList.addReg(phys_reg);
796 }
797
798 // Unbind Float Regs from Rename Map
799 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) {
800 PhysRegIndex phys_reg = renameMap[tid].lookup(freg);
801
802 scoreboard.unsetReg(phys_reg);
803 freeList.addReg(phys_reg);
804 }
805
806 // Squash Throughout Pipeline
807 DynInstPtr inst = commit.rob->readHeadInst(tid);
808 InstSeqNum squash_seq_num = inst->seqNum;
809 fetch.squash(0, squash_seq_num, inst, tid);
810 decode.squash(tid);
811 rename.squash(squash_seq_num, tid);
812 iew.squash(tid);
813 iew.ldstQueue.squash(squash_seq_num, tid);
814 commit.rob->squash(squash_seq_num, tid);
815
816
817 assert(iew.instQueue.getCount(tid) == 0);
818 assert(iew.ldstQueue.getCount(tid) == 0);
819
820 // Reset ROB/IQ/LSQ Entries
821
822 // Commented out for now. This should be possible to do by
823 // telling all the pipeline stages to drain first, and then
824 // checking until the drain completes. Once the pipeline is
825 // drained, call resetEntries(). - 10-09-06 ktlim
826/*
827 if (activeThreads.size() >= 1) {
828 commit.rob->resetEntries();
829 iew.resetEntries();
830 }
831*/
832}
833
834
835template <class Impl>
836void
837FullO3CPU<Impl>::activateWhenReady(ThreadID tid)
838{
839 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming"
840 "(e.g. PhysRegs/ROB/IQ/LSQ) \n",
841 tid);
842
843 bool ready = true;
844
845 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) {
846 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
847 "Phys. Int. Regs.\n",
848 tid);
849 ready = false;
850 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) {
851 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
852 "Phys. Float. Regs.\n",
853 tid);
854 ready = false;
855 } else if (commit.rob->numFreeEntries() >=
856 commit.rob->entryAmount(activeThreads.size() + 1)) {
857 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
858 "ROB entries.\n",
859 tid);
860 ready = false;
861 } else if (iew.instQueue.numFreeEntries() >=
862 iew.instQueue.entryAmount(activeThreads.size() + 1)) {
863 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
864 "IQ entries.\n",
865 tid);
866 ready = false;
867 } else if (iew.ldstQueue.numFreeEntries() >=
868 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) {
869 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
870 "LSQ entries.\n",
871 tid);
872 ready = false;
873 }
874
875 if (ready) {
876 insertThread(tid);
877
878 contextSwitch = false;
879
880 cpuWaitList.remove(tid);
881 } else {
882 suspendContext(tid);
883
884 //blocks fetch
885 contextSwitch = true;
886
887 //@todo: dont always add to waitlist
888 //do waitlist
889 cpuWaitList.push_back(tid);
890 }
891}
892
893template <class Impl>
894Fault
895FullO3CPU<Impl>::hwrei(ThreadID tid)
896{
897#if THE_ISA == ALPHA_ISA
898 // Need to clear the lock flag upon returning from an interrupt.
899 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid);
900
901 this->thread[tid]->kernelStats->hwrei();
902
903 // FIXME: XXX check for interrupts? XXX
904#endif
905 return NoFault;
906}
907
908template <class Impl>
909bool
910FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid)
911{
912#if THE_ISA == ALPHA_ISA
913 if (this->thread[tid]->kernelStats)
914 this->thread[tid]->kernelStats->callpal(palFunc,
915 this->threadContexts[tid]);
916
917 switch (palFunc) {
918 case PAL::halt:
919 halt();
920 if (--System::numSystemsRunning == 0)
921 exitSimLoop("all cpus halted");
922 break;
923
924 case PAL::bpt:
925 case PAL::bugchk:
926 if (this->system->breakpoint())
927 return false;
928 break;
929 }
930#endif
931 return true;
932}
933
934template <class Impl>
935Fault
936FullO3CPU<Impl>::getInterrupts()
937{
938 // Check if there are any outstanding interrupts
939 return this->interrupts->getInterrupt(this->threadContexts[0]);
940}
941
942template <class Impl>
943void
944FullO3CPU<Impl>::processInterrupts(Fault interrupt)
945{
946 // Check for interrupts here. For now can copy the code that
947 // exists within isa_fullsys_traits.hh. Also assume that thread 0
948 // is the one that handles the interrupts.
949 // @todo: Possibly consolidate the interrupt checking code.
950 // @todo: Allow other threads to handle interrupts.
951
952 assert(interrupt != NoFault);
953 this->interrupts->updateIntrInfo(this->threadContexts[0]);
954
955 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name());
956 this->trap(interrupt, 0, NULL);
957}
958
959template <class Impl>
960void
961FullO3CPU<Impl>::updateMemPorts()
962{
963 // Update all ThreadContext's memory ports (Functional/Virtual
964 // Ports)
965 ThreadID size = thread.size();
966 for (ThreadID i = 0; i < size; ++i)
967 thread[i]->connectMemPorts(thread[i]->getTC());
968}
969
970template <class Impl>
971void
972FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst)
973{
974 // Pass the thread's TC into the invoke method.
975 fault->invoke(this->threadContexts[tid], inst);
976}
977
978template <class Impl>
979void
980FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid)
981{
982 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid);
983
984 DPRINTF(Activity,"Activity: syscall() called.\n");
985
986 // Temporarily increase this by one to account for the syscall
987 // instruction.
988 ++(this->thread[tid]->funcExeInst);
989
990 // Execute the actual syscall.
991 this->thread[tid]->syscall(callnum);
992
993 // Decrease funcExeInst by one as the normal commit will handle
994 // incrementing it.
995 --(this->thread[tid]->funcExeInst);
996}
997
998template <class Impl>
999void
1000FullO3CPU<Impl>::serialize(std::ostream &os)
1001{
1002 SimObject::State so_state = SimObject::getState();
1003 SERIALIZE_ENUM(so_state);
1004 BaseCPU::serialize(os);
1005 nameOut(os, csprintf("%s.tickEvent", name()));
1006 tickEvent.serialize(os);
1007
1008 // Use SimpleThread's ability to checkpoint to make it easier to
1009 // write out the registers. Also make this static so it doesn't
1010 // get instantiated multiple times (causes a panic in statistics).
1011 static SimpleThread temp;
1012
1013 ThreadID size = thread.size();
1014 for (ThreadID i = 0; i < size; i++) {
1015 nameOut(os, csprintf("%s.xc.%i", name(), i));
1016 temp.copyTC(thread[i]->getTC());
1017 temp.serialize(os);
1018 }
1019}
1020
1021template <class Impl>
1022void
1023FullO3CPU<Impl>::unserialize(Checkpoint *cp, const std::string &section)
1024{
1025 SimObject::State so_state;
1026 UNSERIALIZE_ENUM(so_state);
1027 BaseCPU::unserialize(cp, section);
1028 tickEvent.unserialize(cp, csprintf("%s.tickEvent", section));
1029
1030 // Use SimpleThread's ability to checkpoint to make it easier to
1031 // read in the registers. Also make this static so it doesn't
1032 // get instantiated multiple times (causes a panic in statistics).
1033 static SimpleThread temp;
1034
1035 ThreadID size = thread.size();
1036 for (ThreadID i = 0; i < size; i++) {
1037 temp.copyTC(thread[i]->getTC());
1038 temp.unserialize(cp, csprintf("%s.xc.%i", section, i));
1039 thread[i]->getTC()->copyArchRegs(temp.getTC());
1040 }
1041}
1042
1043template <class Impl>
1044unsigned int
1045FullO3CPU<Impl>::drain(Event *drain_event)
1046{
1047 DPRINTF(O3CPU, "Switching out\n");
1048
1049 // If the CPU isn't doing anything, then return immediately.
1050 if (_status == Idle || _status == SwitchedOut) {
1051 return 0;
1052 }
1053
1054 drainCount = 0;
1055 fetch.drain();
1056 decode.drain();
1057 rename.drain();
1058 iew.drain();
1059 commit.drain();
1060
1061 // Wake the CPU and record activity so everything can drain out if
1062 // the CPU was not able to immediately drain.
1063 if (getState() != SimObject::Drained) {
1064 // A bit of a hack...set the drainEvent after all the drain()
1065 // calls have been made, that way if all of the stages drain
1066 // immediately, the signalDrained() function knows not to call
1067 // process on the drain event.
1068 drainEvent = drain_event;
1069
1070 wakeCPU();
1071 activityRec.activity();
1072
1073 return 1;
1074 } else {
1075 return 0;
1076 }
1077}
1078
1079template <class Impl>
1080void
1081FullO3CPU<Impl>::resume()
1082{
1083 fetch.resume();
1084 decode.resume();
1085 rename.resume();
1086 iew.resume();
1087 commit.resume();
1088
1089 changeState(SimObject::Running);
1090
1091 if (_status == SwitchedOut || _status == Idle)
1092 return;
1093
1094 assert(system->getMemoryMode() == Enums::timing);
1095
1096 if (!tickEvent.scheduled())
1097 schedule(tickEvent, nextCycle());
1098 _status = Running;
1099}
1100
1101template <class Impl>
1102void
1103FullO3CPU<Impl>::signalDrained()
1104{
1105 if (++drainCount == NumStages) {
1106 if (tickEvent.scheduled())
1107 tickEvent.squash();
1108
1109 changeState(SimObject::Drained);
1110
1111 BaseCPU::switchOut();
1112
1113 if (drainEvent) {
1114 drainEvent->process();
1115 drainEvent = NULL;
1116 }
1117 }
1118 assert(drainCount <= 5);
1119}
1120
1121template <class Impl>
1122void
1123FullO3CPU<Impl>::switchOut()
1124{
1125 fetch.switchOut();
1126 rename.switchOut();
1127 iew.switchOut();
1128 commit.switchOut();
1129 instList.clear();
1130 while (!removeList.empty()) {
1131 removeList.pop();
1132 }
1133
1134 _status = SwitchedOut;
1135#if USE_CHECKER
1136 if (checker)
1137 checker->switchOut();
1138#endif
1139 if (tickEvent.scheduled())
1140 tickEvent.squash();
1141}
1142
1143template <class Impl>
1144void
1145FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU)
1146{
1147 // Flush out any old data from the time buffers.
1148 for (int i = 0; i < timeBuffer.getSize(); ++i) {
1149 timeBuffer.advance();
1150 fetchQueue.advance();
1151 decodeQueue.advance();
1152 renameQueue.advance();
1153 iewQueue.advance();
1154 }
1155
1156 activityRec.reset();
1157
1158 BaseCPU::takeOverFrom(oldCPU, fetch.getIcachePort(), iew.getDcachePort());
1159
1160 fetch.takeOverFrom();
1161 decode.takeOverFrom();
1162 rename.takeOverFrom();
1163 iew.takeOverFrom();
1164 commit.takeOverFrom();
1165
1166 assert(!tickEvent.scheduled() || tickEvent.squashed());
1167
1168 // @todo: Figure out how to properly select the tid to put onto
1169 // the active threads list.
1170 ThreadID tid = 0;
1171
1172 list<ThreadID>::iterator isActive =
1173 std::find(activeThreads.begin(), activeThreads.end(), tid);
1174
1175 if (isActive == activeThreads.end()) {
1176 //May Need to Re-code this if the delay variable is the delay
1177 //needed for thread to activate
1178 DPRINTF(O3CPU, "Adding Thread %i to active threads list\n",
1179 tid);
1180
1181 activeThreads.push_back(tid);
1182 }
1183
1184 // Set all statuses to active, schedule the CPU's tick event.
1185 // @todo: Fix up statuses so this is handled properly
1186 ThreadID size = threadContexts.size();
1187 for (ThreadID i = 0; i < size; ++i) {
1188 ThreadContext *tc = threadContexts[i];
1189 if (tc->status() == ThreadContext::Active && _status != Running) {
1190 _status = Running;
1191 reschedule(tickEvent, nextCycle(), true);
1192 }
1193 }
1194 if (!tickEvent.scheduled())
1195 schedule(tickEvent, nextCycle());
1196}
1197
1198template <class Impl>
1199TheISA::MiscReg
1200FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid)
1201{
1202 return this->isa[tid].readMiscRegNoEffect(misc_reg);
1203}
1204
1205template <class Impl>
1206TheISA::MiscReg
1207FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid)
1208{
1209 miscRegfileReads++;
1210 return this->isa[tid].readMiscReg(misc_reg, tcBase(tid));
1211}
1212
1213template <class Impl>
1214void
1215FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg,
1216 const TheISA::MiscReg &val, ThreadID tid)
1217{
1218 this->isa[tid].setMiscRegNoEffect(misc_reg, val);
1219}
1220
1221template <class Impl>
1222void
1223FullO3CPU<Impl>::setMiscReg(int misc_reg,
1224 const TheISA::MiscReg &val, ThreadID tid)
1225{
1226 miscRegfileWrites++;
1227 this->isa[tid].setMiscReg(misc_reg, val, tcBase(tid));
1228}
1229
1230template <class Impl>
1231uint64_t
1232FullO3CPU<Impl>::readIntReg(int reg_idx)
1233{
1234 intRegfileReads++;
1235 return regFile.readIntReg(reg_idx);
1236}
1237
1238template <class Impl>
1239FloatReg
1240FullO3CPU<Impl>::readFloatReg(int reg_idx)
1241{
1242 fpRegfileReads++;
1243 return regFile.readFloatReg(reg_idx);
1244}
1245
1246template <class Impl>
1247FloatRegBits
1248FullO3CPU<Impl>::readFloatRegBits(int reg_idx)
1249{
1250 fpRegfileReads++;
1251 return regFile.readFloatRegBits(reg_idx);
1252}
1253
1254template <class Impl>
1255void
1256FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val)
1257{
1258 intRegfileWrites++;
1259 regFile.setIntReg(reg_idx, val);
1260}
1261
1262template <class Impl>
1263void
1264FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val)
1265{
1266 fpRegfileWrites++;
1267 regFile.setFloatReg(reg_idx, val);
1268}
1269
1270template <class Impl>
1271void
1272FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val)
1273{
1274 fpRegfileWrites++;
1275 regFile.setFloatRegBits(reg_idx, val);
1276}
1277
1278template <class Impl>
1279uint64_t
1280FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid)
1281{
1282 intRegfileReads++;
1283 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1284
1285 return regFile.readIntReg(phys_reg);
1286}
1287
1288template <class Impl>
1289float
1290FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid)
1291{
1292 fpRegfileReads++;
1293 int idx = reg_idx + TheISA::NumIntRegs;
1294 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1295
1296 return regFile.readFloatReg(phys_reg);
1297}
1298
1299template <class Impl>
1300uint64_t
1301FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid)
1302{
1303 fpRegfileReads++;
1304 int idx = reg_idx + TheISA::NumIntRegs;
1305 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1306
1307 return regFile.readFloatRegBits(phys_reg);
1308}
1309
1310template <class Impl>
1311void
1312FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid)
1313{
1314 intRegfileWrites++;
1315 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1316
1317 regFile.setIntReg(phys_reg, val);
1318}
1319
1320template <class Impl>
1321void
1322FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid)
1323{
1324 fpRegfileWrites++;
1325 int idx = reg_idx + TheISA::NumIntRegs;
1326 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1327
1328 regFile.setFloatReg(phys_reg, val);
1329}
1330
1331template <class Impl>
1332void
1333FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid)
1334{
1335 fpRegfileWrites++;
1336 int idx = reg_idx + TheISA::NumIntRegs;
1337 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1338
1339 regFile.setFloatRegBits(phys_reg, val);
1340}
1341
1342template <class Impl>
1343TheISA::PCState
1344FullO3CPU<Impl>::pcState(ThreadID tid)
1345{
1346 return commit.pcState(tid);
1347}
1348
1349template <class Impl>
1350void
1351FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid)
1352{
1353 commit.pcState(val, tid);
1354}
1355
1356template <class Impl>
1357Addr
1358FullO3CPU<Impl>::instAddr(ThreadID tid)
1359{
1360 return commit.instAddr(tid);
1361}
1362
1363template <class Impl>
1364Addr
1365FullO3CPU<Impl>::nextInstAddr(ThreadID tid)
1366{
1367 return commit.nextInstAddr(tid);
1368}
1369
1370template <class Impl>
1371MicroPC
1372FullO3CPU<Impl>::microPC(ThreadID tid)
1373{
1374 return commit.microPC(tid);
1375}
1376
1377template <class Impl>
1378void
1379FullO3CPU<Impl>::squashFromTC(ThreadID tid)
1380{
1381 this->thread[tid]->inSyscall = true;
1382 this->commit.generateTCEvent(tid);
1383}
1384
1385template <class Impl>
1386typename FullO3CPU<Impl>::ListIt
1387FullO3CPU<Impl>::addInst(DynInstPtr &inst)
1388{
1389 instList.push_back(inst);
1390
1391 return --(instList.end());
1392}
1393
1394template <class Impl>
1395void
1396FullO3CPU<Impl>::instDone(ThreadID tid)
1397{
1398 // Keep an instruction count.
1399 thread[tid]->numInst++;
1400 thread[tid]->numInsts++;
1401 committedInsts[tid]++;
1402 totalCommittedInsts++;
1403 system->totalNumInsts++;
1404 // Check for instruction-count-based events.
1405 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst);
1406 system->instEventQueue.serviceEvents(system->totalNumInsts);
1407}
1408
1409template <class Impl>
1410void
1411FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst)
1412{
1413 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s "
1414 "[sn:%lli]\n",
1415 inst->threadNumber, inst->pcState(), inst->seqNum);
1416
1417 removeInstsThisCycle = true;
1418
1419 // Remove the front instruction.
1420 removeList.push(inst->getInstListIt());
1421}
1422
1423template <class Impl>
1424void
1425FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid)
1426{
1427 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction"
1428 " list.\n", tid);
1429
1430 ListIt end_it;
1431
1432 bool rob_empty = false;
1433
1434 if (instList.empty()) {
1435 return;
1436 } else if (rob.isEmpty(/*tid*/)) {
1437 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n");
1438 end_it = instList.begin();
1439 rob_empty = true;
1440 } else {
1441 end_it = (rob.readTailInst(tid))->getInstListIt();
1442 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n");
1443 }
1444
1445 removeInstsThisCycle = true;
1446
1447 ListIt inst_it = instList.end();
1448
1449 inst_it--;
1450
1451 // Walk through the instruction list, removing any instructions
1452 // that were inserted after the given instruction iterator, end_it.
1453 while (inst_it != end_it) {
1454 assert(!instList.empty());
1455
1456 squashInstIt(inst_it, tid);
1457
1458 inst_it--;
1459 }
1460
1461 // If the ROB was empty, then we actually need to remove the first
1462 // instruction as well.
1463 if (rob_empty) {
1464 squashInstIt(inst_it, tid);
1465 }
1466}
1467
1468template <class Impl>
1469void
1470FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid)
1471{
1472 assert(!instList.empty());
1473
1474 removeInstsThisCycle = true;
1475
1476 ListIt inst_iter = instList.end();
1477
1478 inst_iter--;
1479
1480 DPRINTF(O3CPU, "Deleting instructions from instruction "
1481 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
1482 tid, seq_num, (*inst_iter)->seqNum);
1483
1484 while ((*inst_iter)->seqNum > seq_num) {
1485
1486 bool break_loop = (inst_iter == instList.begin());
1487
1488 squashInstIt(inst_iter, tid);
1489
1490 inst_iter--;
1491
1492 if (break_loop)
1493 break;
1494 }
1495}
1496
1497template <class Impl>
1498inline void
1499FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid)
1500{
1501 if ((*instIt)->threadNumber == tid) {
1502 DPRINTF(O3CPU, "Squashing instruction, "
1503 "[tid:%i] [sn:%lli] PC %s\n",
1504 (*instIt)->threadNumber,
1505 (*instIt)->seqNum,
1506 (*instIt)->pcState());
1507
1508 // Mark it as squashed.
1509 (*instIt)->setSquashed();
1510
1511 // @todo: Formulate a consistent method for deleting
1512 // instructions from the instruction list
1513 // Remove the instruction from the list.
1514 removeList.push(instIt);
1515 }
1516}
1517
1518template <class Impl>
1519void
1520FullO3CPU<Impl>::cleanUpRemovedInsts()
1521{
1522 while (!removeList.empty()) {
1523 DPRINTF(O3CPU, "Removing instruction, "
1524 "[tid:%i] [sn:%lli] PC %s\n",
1525 (*removeList.front())->threadNumber,
1526 (*removeList.front())->seqNum,
1527 (*removeList.front())->pcState());
1528
1529 instList.erase(removeList.front());
1530
1531 removeList.pop();
1532 }
1533
1534 removeInstsThisCycle = false;
1535}
1536/*
1537template <class Impl>
1538void
1539FullO3CPU<Impl>::removeAllInsts()
1540{
1541 instList.clear();
1542}
1543*/
1544template <class Impl>
1545void
1546FullO3CPU<Impl>::dumpInsts()
1547{
1548 int num = 0;
1549
1550 ListIt inst_list_it = instList.begin();
1551
1552 cprintf("Dumping Instruction List\n");
1553
1554 while (inst_list_it != instList.end()) {
1555 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
1556 "Squashed:%i\n\n",
1557 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber,
1558 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(),
1559 (*inst_list_it)->isSquashed());
1560 inst_list_it++;
1561 ++num;
1562 }
1563}
1564/*
1565template <class Impl>
1566void
1567FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst)
1568{
1569 iew.wakeDependents(inst);
1570}
1571*/
1572template <class Impl>
1573void
1574FullO3CPU<Impl>::wakeCPU()
1575{
1576 if (activityRec.active() || tickEvent.scheduled()) {
1577 DPRINTF(Activity, "CPU already running.\n");
1578 return;
1579 }
1580
1581 DPRINTF(Activity, "Waking up CPU\n");
1582
1583 idleCycles += tickToCycles((curTick() - 1) - lastRunningCycle);
1584 numCycles += tickToCycles((curTick() - 1) - lastRunningCycle);
1585
1586 schedule(tickEvent, nextCycle());
1587}
1588
1589template <class Impl>
1590void
1591FullO3CPU<Impl>::wakeup()
1592{
1593 if (this->thread[0]->status() != ThreadContext::Suspended)
1594 return;
1595
1596 this->wakeCPU();
1597
1598 DPRINTF(Quiesce, "Suspended Processor woken\n");
1599 this->threadContexts[0]->activate();
1600}
1601
1602template <class Impl>
1603ThreadID
1604FullO3CPU<Impl>::getFreeTid()
1605{
1606 for (ThreadID tid = 0; tid < numThreads; tid++) {
1607 if (!tids[tid]) {
1608 tids[tid] = true;
1609 return tid;
1610 }
1611 }
1612
1613 return InvalidThreadID;
1614}
1615
1616template <class Impl>
1617void
1618FullO3CPU<Impl>::doContextSwitch()
1619{
1620 if (contextSwitch) {
1621
1622 //ADD CODE TO DEACTIVE THREAD HERE (???)
1623
1624 ThreadID size = cpuWaitList.size();
1625 for (ThreadID tid = 0; tid < size; tid++) {
1626 activateWhenReady(tid);
1627 }
1628
1629 if (cpuWaitList.size() == 0)
1630 contextSwitch = true;
1631 }
1632}
1633
1634template <class Impl>
1635void
1636FullO3CPU<Impl>::updateThreadPriority()
1637{
1638 if (activeThreads.size() > 1) {
1639 //DEFAULT TO ROUND ROBIN SCHEME
1640 //e.g. Move highest priority to end of thread list
1641 list<ThreadID>::iterator list_begin = activeThreads.begin();
1642 list<ThreadID>::iterator list_end = activeThreads.end();
1643
1644 unsigned high_thread = *list_begin;
1645
1646 activeThreads.erase(list_begin);
1647
1648 activeThreads.push_back(high_thread);
1649 }
1650}
1651
1652// Forward declaration of FullO3CPU.
1653template class FullO3CPU<O3CPUImpl>;
1642
1643 unsigned high_thread = *list_begin;
1644
1645 activeThreads.erase(list_begin);
1646
1647 activeThreads.push_back(high_thread);
1648 }
1649}
1650
1651// Forward declaration of FullO3CPU.
1652template class FullO3CPU<O3CPUImpl>;