cpu.cc (9524:d6ffa982a68b) cpu.cc (9648:f10eb34e3e38)
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 drainManager(NULL),
261 lastRunningCycle(curCycle())
262{
263 if (!params->switched_out) {
264 _status = Running;
265 } else {
266 _status = SwitchedOut;
267 }
268
269 if (params->checker) {
270 BaseCPU *temp_checker = params->checker;
271 checker = dynamic_cast<Checker<Impl> *>(temp_checker);
272 checker->setIcachePort(&icachePort);
273 checker->setSystem(params->system);
274 } else {
275 checker = NULL;
276 }
277
278 if (!FullSystem) {
279 thread.resize(numThreads);
280 tids.resize(numThreads);
281 }
282
283 // The stages also need their CPU pointer setup. However this
284 // must be done at the upper level CPU because they have pointers
285 // to the upper level CPU, and not this FullO3CPU.
286
287 // Set up Pointers to the activeThreads list for each stage
288 fetch.setActiveThreads(&activeThreads);
289 decode.setActiveThreads(&activeThreads);
290 rename.setActiveThreads(&activeThreads);
291 iew.setActiveThreads(&activeThreads);
292 commit.setActiveThreads(&activeThreads);
293
294 // Give each of the stages the time buffer they will use.
295 fetch.setTimeBuffer(&timeBuffer);
296 decode.setTimeBuffer(&timeBuffer);
297 rename.setTimeBuffer(&timeBuffer);
298 iew.setTimeBuffer(&timeBuffer);
299 commit.setTimeBuffer(&timeBuffer);
300
301 // Also setup each of the stages' queues.
302 fetch.setFetchQueue(&fetchQueue);
303 decode.setFetchQueue(&fetchQueue);
304 commit.setFetchQueue(&fetchQueue);
305 decode.setDecodeQueue(&decodeQueue);
306 rename.setDecodeQueue(&decodeQueue);
307 rename.setRenameQueue(&renameQueue);
308 iew.setRenameQueue(&renameQueue);
309 iew.setIEWQueue(&iewQueue);
310 commit.setIEWQueue(&iewQueue);
311 commit.setRenameQueue(&renameQueue);
312
313 commit.setIEWStage(&iew);
314 rename.setIEWStage(&iew);
315 rename.setCommitStage(&commit);
316
317 ThreadID active_threads;
318 if (FullSystem) {
319 active_threads = 1;
320 } else {
321 active_threads = params->workload.size();
322
323 if (active_threads > Impl::MaxThreads) {
324 panic("Workload Size too large. Increase the 'MaxThreads' "
325 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) "
326 "or edit your workload size.");
327 }
328 }
329
330 //Make Sure That this a Valid Architeture
331 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs);
332 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs);
333
334 rename.setScoreboard(&scoreboard);
335 iew.setScoreboard(&scoreboard);
336
337 // Setup the rename map for whichever stages need it.
338 PhysRegIndex lreg_idx = 0;
339 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs
340
341 for (ThreadID tid = 0; tid < numThreads; tid++) {
342 bool bindRegs = (tid <= active_threads - 1);
343
344 isa[tid] = params->isa[tid];
345
346 commitRenameMap[tid].init(TheISA::NumIntRegs,
347 params->numPhysIntRegs,
348 lreg_idx, //Index for Logical. Regs
349
350 TheISA::NumFloatRegs,
351 params->numPhysFloatRegs,
352 freg_idx, //Index for Float Regs
353
354 TheISA::NumMiscRegs,
355
356 TheISA::ZeroReg,
357 TheISA::ZeroReg,
358
359 tid,
360 false);
361
362 renameMap[tid].init(TheISA::NumIntRegs,
363 params->numPhysIntRegs,
364 lreg_idx, //Index for Logical. Regs
365
366 TheISA::NumFloatRegs,
367 params->numPhysFloatRegs,
368 freg_idx, //Index for Float Regs
369
370 TheISA::NumMiscRegs,
371
372 TheISA::ZeroReg,
373 TheISA::ZeroReg,
374
375 tid,
376 bindRegs);
377
378 activateThreadEvent[tid].init(tid, this);
379 deallocateContextEvent[tid].init(tid, this);
380 }
381
382 rename.setRenameMap(renameMap);
383 commit.setRenameMap(commitRenameMap);
384
385 // Give renameMap & rename stage access to the freeList;
386 for (ThreadID tid = 0; tid < numThreads; tid++)
387 renameMap[tid].setFreeList(&freeList);
388 rename.setFreeList(&freeList);
389
390 // Setup the ROB for whichever stages need it.
391 commit.setROB(&rob);
392
393 lastActivatedCycle = 0;
394#if 0
395 // Give renameMap & rename stage access to the freeList;
396 for (ThreadID tid = 0; tid < numThreads; tid++)
397 globalSeqNum[tid] = 1;
398#endif
399
400 contextSwitch = false;
401 DPRINTF(O3CPU, "Creating O3CPU object.\n");
402
403 // Setup any thread state.
404 this->thread.resize(this->numThreads);
405
406 for (ThreadID tid = 0; tid < this->numThreads; ++tid) {
407 if (FullSystem) {
408 // SMT is not supported in FS mode yet.
409 assert(this->numThreads == 1);
410 this->thread[tid] = new Thread(this, 0, NULL);
411 } else {
412 if (tid < params->workload.size()) {
413 DPRINTF(O3CPU, "Workload[%i] process is %#x",
414 tid, this->thread[tid]);
415 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
416 (typename Impl::O3CPU *)(this),
417 tid, params->workload[tid]);
418
419 //usedTids[tid] = true;
420 //threadMap[tid] = tid;
421 } else {
422 //Allocate Empty thread so M5 can use later
423 //when scheduling threads to CPU
424 Process* dummy_proc = NULL;
425
426 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
427 (typename Impl::O3CPU *)(this),
428 tid, dummy_proc);
429 //usedTids[tid] = false;
430 }
431 }
432
433 ThreadContext *tc;
434
435 // Setup the TC that will serve as the interface to the threads/CPU.
436 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>;
437
438 tc = o3_tc;
439
440 // If we're using a checker, then the TC should be the
441 // CheckerThreadContext.
442 if (params->checker) {
443 tc = new CheckerThreadContext<O3ThreadContext<Impl> >(
444 o3_tc, this->checker);
445 }
446
447 o3_tc->cpu = (typename Impl::O3CPU *)(this);
448 assert(o3_tc->cpu);
449 o3_tc->thread = this->thread[tid];
450
451 if (FullSystem) {
452 // Setup quiesce event.
453 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc);
454 }
455 // Give the thread the TC.
456 this->thread[tid]->tc = tc;
457
458 // Add the TC to the CPU's list of TC's.
459 this->threadContexts.push_back(tc);
460 }
461
462 // FullO3CPU always requires an interrupt controller.
463 if (!params->switched_out && !interrupts) {
464 fatal("FullO3CPU %s has no interrupt controller.\n"
465 "Ensure createInterruptController() is called.\n", name());
466 }
467
468 for (ThreadID tid = 0; tid < this->numThreads; tid++)
469 this->thread[tid]->setFuncExeInst(0);
470}
471
472template <class Impl>
473FullO3CPU<Impl>::~FullO3CPU()
474{
475}
476
477template <class Impl>
478void
479FullO3CPU<Impl>::regStats()
480{
481 BaseO3CPU::regStats();
482
483 // Register any of the O3CPU's stats here.
484 timesIdled
485 .name(name() + ".timesIdled")
486 .desc("Number of times that the entire CPU went into an idle state and"
487 " unscheduled itself")
488 .prereq(timesIdled);
489
490 idleCycles
491 .name(name() + ".idleCycles")
492 .desc("Total number of cycles that the CPU has spent unscheduled due "
493 "to idling")
494 .prereq(idleCycles);
495
496 quiesceCycles
497 .name(name() + ".quiesceCycles")
498 .desc("Total number of cycles that CPU has spent quiesced or waiting "
499 "for an interrupt")
500 .prereq(quiesceCycles);
501
502 // Number of Instructions simulated
503 // --------------------------------
504 // Should probably be in Base CPU but need templated
505 // MaxThreads so put in here instead
506 committedInsts
507 .init(numThreads)
508 .name(name() + ".committedInsts")
509 .desc("Number of Instructions Simulated");
510
511 committedOps
512 .init(numThreads)
513 .name(name() + ".committedOps")
514 .desc("Number of Ops (including micro ops) Simulated");
515
516 totalCommittedInsts
517 .name(name() + ".committedInsts_total")
518 .desc("Number of Instructions Simulated");
519
520 cpi
521 .name(name() + ".cpi")
522 .desc("CPI: Cycles Per Instruction")
523 .precision(6);
524 cpi = numCycles / committedInsts;
525
526 totalCpi
527 .name(name() + ".cpi_total")
528 .desc("CPI: Total CPI of All Threads")
529 .precision(6);
530 totalCpi = numCycles / totalCommittedInsts;
531
532 ipc
533 .name(name() + ".ipc")
534 .desc("IPC: Instructions Per Cycle")
535 .precision(6);
536 ipc = committedInsts / numCycles;
537
538 totalIpc
539 .name(name() + ".ipc_total")
540 .desc("IPC: Total IPC of All Threads")
541 .precision(6);
542 totalIpc = totalCommittedInsts / numCycles;
543
544 this->fetch.regStats();
545 this->decode.regStats();
546 this->rename.regStats();
547 this->iew.regStats();
548 this->commit.regStats();
549 this->rob.regStats();
550
551 intRegfileReads
552 .name(name() + ".int_regfile_reads")
553 .desc("number of integer regfile reads")
554 .prereq(intRegfileReads);
555
556 intRegfileWrites
557 .name(name() + ".int_regfile_writes")
558 .desc("number of integer regfile writes")
559 .prereq(intRegfileWrites);
560
561 fpRegfileReads
562 .name(name() + ".fp_regfile_reads")
563 .desc("number of floating regfile reads")
564 .prereq(fpRegfileReads);
565
566 fpRegfileWrites
567 .name(name() + ".fp_regfile_writes")
568 .desc("number of floating regfile writes")
569 .prereq(fpRegfileWrites);
570
571 miscRegfileReads
572 .name(name() + ".misc_regfile_reads")
573 .desc("number of misc regfile reads")
574 .prereq(miscRegfileReads);
575
576 miscRegfileWrites
577 .name(name() + ".misc_regfile_writes")
578 .desc("number of misc regfile writes")
579 .prereq(miscRegfileWrites);
580}
581
582template <class Impl>
583void
584FullO3CPU<Impl>::tick()
585{
586 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n");
587 assert(!switchedOut());
588 assert(getDrainState() != Drainable::Drained);
589
590 ++numCycles;
591
592// activity = false;
593
594 //Tick each of the stages
595 fetch.tick();
596
597 decode.tick();
598
599 rename.tick();
600
601 iew.tick();
602
603 commit.tick();
604
605 if (!FullSystem)
606 doContextSwitch();
607
608 // Now advance the time buffers
609 timeBuffer.advance();
610
611 fetchQueue.advance();
612 decodeQueue.advance();
613 renameQueue.advance();
614 iewQueue.advance();
615
616 activityRec.advance();
617
618 if (removeInstsThisCycle) {
619 cleanUpRemovedInsts();
620 }
621
622 if (!tickEvent.scheduled()) {
623 if (_status == SwitchedOut) {
624 DPRINTF(O3CPU, "Switched out!\n");
625 // increment stat
626 lastRunningCycle = curCycle();
627 } else if (!activityRec.active() || _status == Idle) {
628 DPRINTF(O3CPU, "Idle!\n");
629 lastRunningCycle = curCycle();
630 timesIdled++;
631 } else {
632 schedule(tickEvent, clockEdge(Cycles(1)));
633 DPRINTF(O3CPU, "Scheduling next tick!\n");
634 }
635 }
636
637 if (!FullSystem)
638 updateThreadPriority();
639
640 tryDrain();
641}
642
643template <class Impl>
644void
645FullO3CPU<Impl>::init()
646{
647 BaseCPU::init();
648
649 for (ThreadID tid = 0; tid < numThreads; ++tid) {
650 // Set noSquashFromTC so that the CPU doesn't squash when initially
651 // setting up registers.
652 thread[tid]->noSquashFromTC = true;
653 // Initialise the ThreadContext's memory proxies
654 thread[tid]->initMemProxies(thread[tid]->getTC());
655 }
656
657 if (FullSystem && !params()->switched_out) {
658 for (ThreadID tid = 0; tid < numThreads; tid++) {
659 ThreadContext *src_tc = threadContexts[tid];
660 TheISA::initCPU(src_tc, src_tc->contextId());
661 }
662 }
663
664 // Clear noSquashFromTC.
665 for (int tid = 0; tid < numThreads; ++tid)
666 thread[tid]->noSquashFromTC = false;
667
668 commit.setThreads(thread);
669}
670
671template <class Impl>
672void
673FullO3CPU<Impl>::startup()
674{
675 for (int tid = 0; tid < numThreads; ++tid)
676 isa[tid]->startup(threadContexts[tid]);
677
678 fetch.startupStage();
679 decode.startupStage();
680 iew.startupStage();
681 rename.startupStage();
682 commit.startupStage();
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 assert(!switchedOut());
694
695 if (isActive == activeThreads.end()) {
696 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n",
697 tid);
698
699 activeThreads.push_back(tid);
700 }
701}
702
703template <class Impl>
704void
705FullO3CPU<Impl>::deactivateThread(ThreadID tid)
706{
707 //Remove From Active List, if Active
708 list<ThreadID>::iterator thread_it =
709 std::find(activeThreads.begin(), activeThreads.end(), tid);
710
711 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid);
712 assert(!switchedOut());
713
714 if (thread_it != activeThreads.end()) {
715 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n",
716 tid);
717 activeThreads.erase(thread_it);
718 }
719}
720
721template <class Impl>
722Counter
723FullO3CPU<Impl>::totalInsts() const
724{
725 Counter total(0);
726
727 ThreadID size = thread.size();
728 for (ThreadID i = 0; i < size; i++)
729 total += thread[i]->numInst;
730
731 return total;
732}
733
734template <class Impl>
735Counter
736FullO3CPU<Impl>::totalOps() const
737{
738 Counter total(0);
739
740 ThreadID size = thread.size();
741 for (ThreadID i = 0; i < size; i++)
742 total += thread[i]->numOp;
743
744 return total;
745}
746
747template <class Impl>
748void
749FullO3CPU<Impl>::activateContext(ThreadID tid, Cycles delay)
750{
751 assert(!switchedOut());
752
753 // Needs to set each stage to running as well.
754 if (delay){
755 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate "
756 "on cycle %d\n", tid, clockEdge(delay));
757 scheduleActivateThreadEvent(tid, delay);
758 } else {
759 activateThread(tid);
760 }
761
762 // We don't want to wake the CPU if it is drained. In that case,
763 // we just want to flag the thread as active and schedule the tick
764 // event from drainResume() instead.
765 if (getDrainState() == Drainable::Drained)
766 return;
767
768 // If we are time 0 or if the last activation time is in the past,
769 // schedule the next tick and wake up the fetch unit
770 if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) {
771 scheduleTickEvent(delay);
772
773 // Be sure to signal that there's some activity so the CPU doesn't
774 // deschedule itself.
775 activityRec.activity();
776 fetch.wakeFromQuiesce();
777
778 Cycles cycles(curCycle() - lastRunningCycle);
779 // @todo: This is an oddity that is only here to match the stats
780 if (cycles != 0)
781 --cycles;
782 quiesceCycles += cycles;
783
784 lastActivatedCycle = curTick();
785
786 _status = Running;
787 }
788}
789
790template <class Impl>
791bool
792FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove,
793 Cycles delay)
794{
795 // Schedule removal of thread data from CPU
796 if (delay){
797 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate "
798 "on tick %d\n", tid, clockEdge(delay));
799 scheduleDeallocateContextEvent(tid, remove, delay);
800 return false;
801 } else {
802 deactivateThread(tid);
803 if (remove)
804 removeThread(tid);
805 return true;
806 }
807}
808
809template <class Impl>
810void
811FullO3CPU<Impl>::suspendContext(ThreadID tid)
812{
813 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid);
814 assert(!switchedOut());
815 bool deallocated = scheduleDeallocateContext(tid, false, Cycles(1));
816 // If this was the last thread then unschedule the tick event.
817 if ((activeThreads.size() == 1 && !deallocated) ||
818 activeThreads.size() == 0)
819 unscheduleTickEvent();
820
821 DPRINTF(Quiesce, "Suspending Context\n");
822 lastRunningCycle = curCycle();
823 _status = Idle;
824}
825
826template <class Impl>
827void
828FullO3CPU<Impl>::haltContext(ThreadID tid)
829{
830 //For now, this is the same as deallocate
831 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid);
832 assert(!switchedOut());
833 scheduleDeallocateContext(tid, true, Cycles(1));
834}
835
836template <class Impl>
837void
838FullO3CPU<Impl>::insertThread(ThreadID tid)
839{
840 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU");
841 // Will change now that the PC and thread state is internal to the CPU
842 // and not in the ThreadContext.
843 ThreadContext *src_tc;
844 if (FullSystem)
845 src_tc = system->threadContexts[tid];
846 else
847 src_tc = tcBase(tid);
848
849 //Bind Int Regs to Rename Map
850 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
851 PhysRegIndex phys_reg = freeList.getIntReg();
852
853 renameMap[tid].setEntry(ireg,phys_reg);
854 scoreboard.setReg(phys_reg);
855 }
856
857 //Bind Float Regs to Rename Map
858 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) {
859 PhysRegIndex phys_reg = freeList.getFloatReg();
860
861 renameMap[tid].setEntry(freg,phys_reg);
862 scoreboard.setReg(phys_reg);
863 }
864
865 //Copy Thread Data Into RegFile
866 //this->copyFromTC(tid);
867
868 //Set PC/NPC/NNPC
869 pcState(src_tc->pcState(), tid);
870
871 src_tc->setStatus(ThreadContext::Active);
872
873 activateContext(tid, Cycles(1));
874
875 //Reset ROB/IQ/LSQ Entries
876 commit.rob->resetEntries();
877 iew.resetEntries();
878}
879
880template <class Impl>
881void
882FullO3CPU<Impl>::removeThread(ThreadID tid)
883{
884 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid);
885
886 // Copy Thread Data From RegFile
887 // If thread is suspended, it might be re-allocated
888 // this->copyToTC(tid);
889
890
891 // @todo: 2-27-2008: Fix how we free up rename mappings
892 // here to alleviate the case for double-freeing registers
893 // in SMT workloads.
894
895 // Unbind Int Regs from Rename Map
896 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
897 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg);
898
899 scoreboard.unsetReg(phys_reg);
900 freeList.addReg(phys_reg);
901 }
902
903 // Unbind Float Regs from Rename Map
904 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) {
905 PhysRegIndex phys_reg = renameMap[tid].lookup(freg);
906
907 scoreboard.unsetReg(phys_reg);
908 freeList.addReg(phys_reg);
909 }
910
911 // Squash Throughout Pipeline
912 DynInstPtr inst = commit.rob->readHeadInst(tid);
913 InstSeqNum squash_seq_num = inst->seqNum;
914 fetch.squash(0, squash_seq_num, inst, tid);
915 decode.squash(tid);
916 rename.squash(squash_seq_num, tid);
917 iew.squash(tid);
918 iew.ldstQueue.squash(squash_seq_num, tid);
919 commit.rob->squash(squash_seq_num, tid);
920
921
922 assert(iew.instQueue.getCount(tid) == 0);
923 assert(iew.ldstQueue.getCount(tid) == 0);
924
925 // Reset ROB/IQ/LSQ Entries
926
927 // Commented out for now. This should be possible to do by
928 // telling all the pipeline stages to drain first, and then
929 // checking until the drain completes. Once the pipeline is
930 // drained, call resetEntries(). - 10-09-06 ktlim
931/*
932 if (activeThreads.size() >= 1) {
933 commit.rob->resetEntries();
934 iew.resetEntries();
935 }
936*/
937}
938
939
940template <class Impl>
941void
942FullO3CPU<Impl>::activateWhenReady(ThreadID tid)
943{
944 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming"
945 "(e.g. PhysRegs/ROB/IQ/LSQ) \n",
946 tid);
947
948 bool ready = true;
949
950 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) {
951 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
952 "Phys. Int. Regs.\n",
953 tid);
954 ready = false;
955 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) {
956 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
957 "Phys. Float. Regs.\n",
958 tid);
959 ready = false;
960 } else if (commit.rob->numFreeEntries() >=
961 commit.rob->entryAmount(activeThreads.size() + 1)) {
962 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
963 "ROB entries.\n",
964 tid);
965 ready = false;
966 } else if (iew.instQueue.numFreeEntries() >=
967 iew.instQueue.entryAmount(activeThreads.size() + 1)) {
968 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
969 "IQ entries.\n",
970 tid);
971 ready = false;
972 } else if (iew.ldstQueue.numFreeEntries() >=
973 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) {
974 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
975 "LSQ entries.\n",
976 tid);
977 ready = false;
978 }
979
980 if (ready) {
981 insertThread(tid);
982
983 contextSwitch = false;
984
985 cpuWaitList.remove(tid);
986 } else {
987 suspendContext(tid);
988
989 //blocks fetch
990 contextSwitch = true;
991
992 //@todo: dont always add to waitlist
993 //do waitlist
994 cpuWaitList.push_back(tid);
995 }
996}
997
998template <class Impl>
999Fault
1000FullO3CPU<Impl>::hwrei(ThreadID tid)
1001{
1002#if THE_ISA == ALPHA_ISA
1003 // Need to clear the lock flag upon returning from an interrupt.
1004 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid);
1005
1006 this->thread[tid]->kernelStats->hwrei();
1007
1008 // FIXME: XXX check for interrupts? XXX
1009#endif
1010 return NoFault;
1011}
1012
1013template <class Impl>
1014bool
1015FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid)
1016{
1017#if THE_ISA == ALPHA_ISA
1018 if (this->thread[tid]->kernelStats)
1019 this->thread[tid]->kernelStats->callpal(palFunc,
1020 this->threadContexts[tid]);
1021
1022 switch (palFunc) {
1023 case PAL::halt:
1024 halt();
1025 if (--System::numSystemsRunning == 0)
1026 exitSimLoop("all cpus halted");
1027 break;
1028
1029 case PAL::bpt:
1030 case PAL::bugchk:
1031 if (this->system->breakpoint())
1032 return false;
1033 break;
1034 }
1035#endif
1036 return true;
1037}
1038
1039template <class Impl>
1040Fault
1041FullO3CPU<Impl>::getInterrupts()
1042{
1043 // Check if there are any outstanding interrupts
1044 return this->interrupts->getInterrupt(this->threadContexts[0]);
1045}
1046
1047template <class Impl>
1048void
1049FullO3CPU<Impl>::processInterrupts(Fault interrupt)
1050{
1051 // Check for interrupts here. For now can copy the code that
1052 // exists within isa_fullsys_traits.hh. Also assume that thread 0
1053 // is the one that handles the interrupts.
1054 // @todo: Possibly consolidate the interrupt checking code.
1055 // @todo: Allow other threads to handle interrupts.
1056
1057 assert(interrupt != NoFault);
1058 this->interrupts->updateIntrInfo(this->threadContexts[0]);
1059
1060 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name());
1061 this->trap(interrupt, 0, NULL);
1062}
1063
1064template <class Impl>
1065void
1066FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst)
1067{
1068 // Pass the thread's TC into the invoke method.
1069 fault->invoke(this->threadContexts[tid], inst);
1070}
1071
1072template <class Impl>
1073void
1074FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid)
1075{
1076 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid);
1077
1078 DPRINTF(Activity,"Activity: syscall() called.\n");
1079
1080 // Temporarily increase this by one to account for the syscall
1081 // instruction.
1082 ++(this->thread[tid]->funcExeInst);
1083
1084 // Execute the actual syscall.
1085 this->thread[tid]->syscall(callnum);
1086
1087 // Decrease funcExeInst by one as the normal commit will handle
1088 // incrementing it.
1089 --(this->thread[tid]->funcExeInst);
1090}
1091
1092template <class Impl>
1093void
1094FullO3CPU<Impl>::serializeThread(std::ostream &os, ThreadID tid)
1095{
1096 thread[tid]->serialize(os);
1097}
1098
1099template <class Impl>
1100void
1101FullO3CPU<Impl>::unserializeThread(Checkpoint *cp, const std::string &section,
1102 ThreadID tid)
1103{
1104 thread[tid]->unserialize(cp, section);
1105}
1106
1107template <class Impl>
1108unsigned int
1109FullO3CPU<Impl>::drain(DrainManager *drain_manager)
1110{
1111 // If the CPU isn't doing anything, then return immediately.
1112 if (switchedOut()) {
1113 setDrainState(Drainable::Drained);
1114 return 0;
1115 }
1116
1117 DPRINTF(Drain, "Draining...\n");
1118 setDrainState(Drainable::Draining);
1119
1120 // We only need to signal a drain to the commit stage as this
1121 // initiates squashing controls the draining. Once the commit
1122 // stage commits an instruction where it is safe to stop, it'll
1123 // squash the rest of the instructions in the pipeline and force
1124 // the fetch stage to stall. The pipeline will be drained once all
1125 // in-flight instructions have retired.
1126 commit.drain();
1127
1128 // Wake the CPU and record activity so everything can drain out if
1129 // the CPU was not able to immediately drain.
1130 if (!isDrained()) {
1131 drainManager = drain_manager;
1132
1133 wakeCPU();
1134 activityRec.activity();
1135
1136 DPRINTF(Drain, "CPU not drained\n");
1137
1138 return 1;
1139 } else {
1140 setDrainState(Drainable::Drained);
1141 DPRINTF(Drain, "CPU is already drained\n");
1142 if (tickEvent.scheduled())
1143 deschedule(tickEvent);
1144
1145 // Flush out any old data from the time buffers. In
1146 // particular, there might be some data in flight from the
1147 // fetch stage that isn't visible in any of the CPU buffers we
1148 // test in isDrained().
1149 for (int i = 0; i < timeBuffer.getSize(); ++i) {
1150 timeBuffer.advance();
1151 fetchQueue.advance();
1152 decodeQueue.advance();
1153 renameQueue.advance();
1154 iewQueue.advance();
1155 }
1156
1157 drainSanityCheck();
1158 return 0;
1159 }
1160}
1161
1162template <class Impl>
1163bool
1164FullO3CPU<Impl>::tryDrain()
1165{
1166 if (!drainManager || !isDrained())
1167 return false;
1168
1169 if (tickEvent.scheduled())
1170 deschedule(tickEvent);
1171
1172 DPRINTF(Drain, "CPU done draining, processing drain event\n");
1173 drainManager->signalDrainDone();
1174 drainManager = NULL;
1175
1176 return true;
1177}
1178
1179template <class Impl>
1180void
1181FullO3CPU<Impl>::drainSanityCheck() const
1182{
1183 assert(isDrained());
1184 fetch.drainSanityCheck();
1185 decode.drainSanityCheck();
1186 rename.drainSanityCheck();
1187 iew.drainSanityCheck();
1188 commit.drainSanityCheck();
1189}
1190
1191template <class Impl>
1192bool
1193FullO3CPU<Impl>::isDrained() const
1194{
1195 bool drained(true);
1196
1197 for (ThreadID i = 0; i < thread.size(); ++i) {
1198 if (activateThreadEvent[i].scheduled()) {
1199 DPRINTF(Drain, "CPU not drained, tread %i has a "
1200 "pending activate event\n", i);
1201 drained = false;
1202 }
1203 if (deallocateContextEvent[i].scheduled()) {
1204 DPRINTF(Drain, "CPU not drained, tread %i has a "
1205 "pending deallocate context event\n", i);
1206 drained = false;
1207 }
1208 }
1209
1210 if (!instList.empty() || !removeList.empty()) {
1211 DPRINTF(Drain, "Main CPU structures not drained.\n");
1212 drained = false;
1213 }
1214
1215 if (!fetch.isDrained()) {
1216 DPRINTF(Drain, "Fetch not drained.\n");
1217 drained = false;
1218 }
1219
1220 if (!decode.isDrained()) {
1221 DPRINTF(Drain, "Decode not drained.\n");
1222 drained = false;
1223 }
1224
1225 if (!rename.isDrained()) {
1226 DPRINTF(Drain, "Rename not drained.\n");
1227 drained = false;
1228 }
1229
1230 if (!iew.isDrained()) {
1231 DPRINTF(Drain, "IEW not drained.\n");
1232 drained = false;
1233 }
1234
1235 if (!commit.isDrained()) {
1236 DPRINTF(Drain, "Commit not drained.\n");
1237 drained = false;
1238 }
1239
1240 return drained;
1241}
1242
1243template <class Impl>
1244void
1245FullO3CPU<Impl>::commitDrained(ThreadID tid)
1246{
1247 fetch.drainStall(tid);
1248}
1249
1250template <class Impl>
1251void
1252FullO3CPU<Impl>::drainResume()
1253{
1254 setDrainState(Drainable::Running);
1255 if (switchedOut())
1256 return;
1257
1258 DPRINTF(Drain, "Resuming...\n");
1259 verifyMemoryMode();
1260
1261 fetch.drainResume();
1262 commit.drainResume();
1263
1264 _status = Idle;
1265 for (ThreadID i = 0; i < thread.size(); i++) {
1266 if (thread[i]->status() == ThreadContext::Active) {
1267 DPRINTF(Drain, "Activating thread: %i\n", i);
1268 activateThread(i);
1269 _status = Running;
1270 }
1271 }
1272
1273 assert(!tickEvent.scheduled());
1274 if (_status == Running)
1275 schedule(tickEvent, nextCycle());
1276}
1277
1278template <class Impl>
1279void
1280FullO3CPU<Impl>::switchOut()
1281{
1282 DPRINTF(O3CPU, "Switching out\n");
1283 BaseCPU::switchOut();
1284
1285 activityRec.reset();
1286
1287 _status = SwitchedOut;
1288
1289 if (checker)
1290 checker->switchOut();
1291}
1292
1293template <class Impl>
1294void
1295FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU)
1296{
1297 BaseCPU::takeOverFrom(oldCPU);
1298
1299 fetch.takeOverFrom();
1300 decode.takeOverFrom();
1301 rename.takeOverFrom();
1302 iew.takeOverFrom();
1303 commit.takeOverFrom();
1304
1305 assert(!tickEvent.scheduled());
1306
1307 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU);
1308 if (oldO3CPU)
1309 globalSeqNum = oldO3CPU->globalSeqNum;
1310
1311 lastRunningCycle = curCycle();
1312 _status = Idle;
1313}
1314
1315template <class Impl>
1316void
1317FullO3CPU<Impl>::verifyMemoryMode() const
1318{
1319 if (!system->isTimingMode()) {
1320 fatal("The O3 CPU requires the memory system to be in "
1321 "'timing' mode.\n");
1322 }
1323}
1324
1325template <class Impl>
1326TheISA::MiscReg
1327FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid)
1328{
1329 return this->isa[tid]->readMiscRegNoEffect(misc_reg);
1330}
1331
1332template <class Impl>
1333TheISA::MiscReg
1334FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid)
1335{
1336 miscRegfileReads++;
1337 return this->isa[tid]->readMiscReg(misc_reg, tcBase(tid));
1338}
1339
1340template <class Impl>
1341void
1342FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg,
1343 const TheISA::MiscReg &val, ThreadID tid)
1344{
1345 this->isa[tid]->setMiscRegNoEffect(misc_reg, val);
1346}
1347
1348template <class Impl>
1349void
1350FullO3CPU<Impl>::setMiscReg(int misc_reg,
1351 const TheISA::MiscReg &val, ThreadID tid)
1352{
1353 miscRegfileWrites++;
1354 this->isa[tid]->setMiscReg(misc_reg, val, tcBase(tid));
1355}
1356
1357template <class Impl>
1358uint64_t
1359FullO3CPU<Impl>::readIntReg(int reg_idx)
1360{
1361 intRegfileReads++;
1362 return regFile.readIntReg(reg_idx);
1363}
1364
1365template <class Impl>
1366FloatReg
1367FullO3CPU<Impl>::readFloatReg(int reg_idx)
1368{
1369 fpRegfileReads++;
1370 return regFile.readFloatReg(reg_idx);
1371}
1372
1373template <class Impl>
1374FloatRegBits
1375FullO3CPU<Impl>::readFloatRegBits(int reg_idx)
1376{
1377 fpRegfileReads++;
1378 return regFile.readFloatRegBits(reg_idx);
1379}
1380
1381template <class Impl>
1382void
1383FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val)
1384{
1385 intRegfileWrites++;
1386 regFile.setIntReg(reg_idx, val);
1387}
1388
1389template <class Impl>
1390void
1391FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val)
1392{
1393 fpRegfileWrites++;
1394 regFile.setFloatReg(reg_idx, val);
1395}
1396
1397template <class Impl>
1398void
1399FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val)
1400{
1401 fpRegfileWrites++;
1402 regFile.setFloatRegBits(reg_idx, val);
1403}
1404
1405template <class Impl>
1406uint64_t
1407FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid)
1408{
1409 intRegfileReads++;
1410 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1411
1412 return regFile.readIntReg(phys_reg);
1413}
1414
1415template <class Impl>
1416float
1417FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid)
1418{
1419 fpRegfileReads++;
1420 int idx = reg_idx + TheISA::NumIntRegs;
1421 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1422
1423 return regFile.readFloatReg(phys_reg);
1424}
1425
1426template <class Impl>
1427uint64_t
1428FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid)
1429{
1430 fpRegfileReads++;
1431 int idx = reg_idx + TheISA::NumIntRegs;
1432 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1433
1434 return regFile.readFloatRegBits(phys_reg);
1435}
1436
1437template <class Impl>
1438void
1439FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid)
1440{
1441 intRegfileWrites++;
1442 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1443
1444 regFile.setIntReg(phys_reg, val);
1445}
1446
1447template <class Impl>
1448void
1449FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid)
1450{
1451 fpRegfileWrites++;
1452 int idx = reg_idx + TheISA::NumIntRegs;
1453 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1454
1455 regFile.setFloatReg(phys_reg, val);
1456}
1457
1458template <class Impl>
1459void
1460FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid)
1461{
1462 fpRegfileWrites++;
1463 int idx = reg_idx + TheISA::NumIntRegs;
1464 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1465
1466 regFile.setFloatRegBits(phys_reg, val);
1467}
1468
1469template <class Impl>
1470TheISA::PCState
1471FullO3CPU<Impl>::pcState(ThreadID tid)
1472{
1473 return commit.pcState(tid);
1474}
1475
1476template <class Impl>
1477void
1478FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid)
1479{
1480 commit.pcState(val, tid);
1481}
1482
1483template <class Impl>
1484Addr
1485FullO3CPU<Impl>::instAddr(ThreadID tid)
1486{
1487 return commit.instAddr(tid);
1488}
1489
1490template <class Impl>
1491Addr
1492FullO3CPU<Impl>::nextInstAddr(ThreadID tid)
1493{
1494 return commit.nextInstAddr(tid);
1495}
1496
1497template <class Impl>
1498MicroPC
1499FullO3CPU<Impl>::microPC(ThreadID tid)
1500{
1501 return commit.microPC(tid);
1502}
1503
1504template <class Impl>
1505void
1506FullO3CPU<Impl>::squashFromTC(ThreadID tid)
1507{
1508 this->thread[tid]->noSquashFromTC = true;
1509 this->commit.generateTCEvent(tid);
1510}
1511
1512template <class Impl>
1513typename FullO3CPU<Impl>::ListIt
1514FullO3CPU<Impl>::addInst(DynInstPtr &inst)
1515{
1516 instList.push_back(inst);
1517
1518 return --(instList.end());
1519}
1520
1521template <class Impl>
1522void
1523FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst)
1524{
1525 // Keep an instruction count.
1526 if (!inst->isMicroop() || inst->isLastMicroop()) {
1527 thread[tid]->numInst++;
1528 thread[tid]->numInsts++;
1529 committedInsts[tid]++;
1530 totalCommittedInsts++;
1531 }
1532 thread[tid]->numOp++;
1533 thread[tid]->numOps++;
1534 committedOps[tid]++;
1535
1536 system->totalNumInsts++;
1537 // Check for instruction-count-based events.
1538 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst);
1539 system->instEventQueue.serviceEvents(system->totalNumInsts);
1540}
1541
1542template <class Impl>
1543void
1544FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst)
1545{
1546 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s "
1547 "[sn:%lli]\n",
1548 inst->threadNumber, inst->pcState(), inst->seqNum);
1549
1550 removeInstsThisCycle = true;
1551
1552 // Remove the front instruction.
1553 removeList.push(inst->getInstListIt());
1554}
1555
1556template <class Impl>
1557void
1558FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid)
1559{
1560 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction"
1561 " list.\n", tid);
1562
1563 ListIt end_it;
1564
1565 bool rob_empty = false;
1566
1567 if (instList.empty()) {
1568 return;
1569 } else if (rob.isEmpty(/*tid*/)) {
1570 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n");
1571 end_it = instList.begin();
1572 rob_empty = true;
1573 } else {
1574 end_it = (rob.readTailInst(tid))->getInstListIt();
1575 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n");
1576 }
1577
1578 removeInstsThisCycle = true;
1579
1580 ListIt inst_it = instList.end();
1581
1582 inst_it--;
1583
1584 // Walk through the instruction list, removing any instructions
1585 // that were inserted after the given instruction iterator, end_it.
1586 while (inst_it != end_it) {
1587 assert(!instList.empty());
1588
1589 squashInstIt(inst_it, tid);
1590
1591 inst_it--;
1592 }
1593
1594 // If the ROB was empty, then we actually need to remove the first
1595 // instruction as well.
1596 if (rob_empty) {
1597 squashInstIt(inst_it, tid);
1598 }
1599}
1600
1601template <class Impl>
1602void
1603FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid)
1604{
1605 assert(!instList.empty());
1606
1607 removeInstsThisCycle = true;
1608
1609 ListIt inst_iter = instList.end();
1610
1611 inst_iter--;
1612
1613 DPRINTF(O3CPU, "Deleting instructions from instruction "
1614 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
1615 tid, seq_num, (*inst_iter)->seqNum);
1616
1617 while ((*inst_iter)->seqNum > seq_num) {
1618
1619 bool break_loop = (inst_iter == instList.begin());
1620
1621 squashInstIt(inst_iter, tid);
1622
1623 inst_iter--;
1624
1625 if (break_loop)
1626 break;
1627 }
1628}
1629
1630template <class Impl>
1631inline void
1632FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid)
1633{
1634 if ((*instIt)->threadNumber == tid) {
1635 DPRINTF(O3CPU, "Squashing instruction, "
1636 "[tid:%i] [sn:%lli] PC %s\n",
1637 (*instIt)->threadNumber,
1638 (*instIt)->seqNum,
1639 (*instIt)->pcState());
1640
1641 // Mark it as squashed.
1642 (*instIt)->setSquashed();
1643
1644 // @todo: Formulate a consistent method for deleting
1645 // instructions from the instruction list
1646 // Remove the instruction from the list.
1647 removeList.push(instIt);
1648 }
1649}
1650
1651template <class Impl>
1652void
1653FullO3CPU<Impl>::cleanUpRemovedInsts()
1654{
1655 while (!removeList.empty()) {
1656 DPRINTF(O3CPU, "Removing instruction, "
1657 "[tid:%i] [sn:%lli] PC %s\n",
1658 (*removeList.front())->threadNumber,
1659 (*removeList.front())->seqNum,
1660 (*removeList.front())->pcState());
1661
1662 instList.erase(removeList.front());
1663
1664 removeList.pop();
1665 }
1666
1667 removeInstsThisCycle = false;
1668}
1669/*
1670template <class Impl>
1671void
1672FullO3CPU<Impl>::removeAllInsts()
1673{
1674 instList.clear();
1675}
1676*/
1677template <class Impl>
1678void
1679FullO3CPU<Impl>::dumpInsts()
1680{
1681 int num = 0;
1682
1683 ListIt inst_list_it = instList.begin();
1684
1685 cprintf("Dumping Instruction List\n");
1686
1687 while (inst_list_it != instList.end()) {
1688 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
1689 "Squashed:%i\n\n",
1690 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber,
1691 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(),
1692 (*inst_list_it)->isSquashed());
1693 inst_list_it++;
1694 ++num;
1695 }
1696}
1697/*
1698template <class Impl>
1699void
1700FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst)
1701{
1702 iew.wakeDependents(inst);
1703}
1704*/
1705template <class Impl>
1706void
1707FullO3CPU<Impl>::wakeCPU()
1708{
1709 if (activityRec.active() || tickEvent.scheduled()) {
1710 DPRINTF(Activity, "CPU already running.\n");
1711 return;
1712 }
1713
1714 DPRINTF(Activity, "Waking up CPU\n");
1715
1716 Cycles cycles(curCycle() - lastRunningCycle);
1717 // @todo: This is an oddity that is only here to match the stats
1718 if (cycles != 0)
1719 --cycles;
1720 idleCycles += cycles;
1721 numCycles += cycles;
1722
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 drainManager(NULL),
261 lastRunningCycle(curCycle())
262{
263 if (!params->switched_out) {
264 _status = Running;
265 } else {
266 _status = SwitchedOut;
267 }
268
269 if (params->checker) {
270 BaseCPU *temp_checker = params->checker;
271 checker = dynamic_cast<Checker<Impl> *>(temp_checker);
272 checker->setIcachePort(&icachePort);
273 checker->setSystem(params->system);
274 } else {
275 checker = NULL;
276 }
277
278 if (!FullSystem) {
279 thread.resize(numThreads);
280 tids.resize(numThreads);
281 }
282
283 // The stages also need their CPU pointer setup. However this
284 // must be done at the upper level CPU because they have pointers
285 // to the upper level CPU, and not this FullO3CPU.
286
287 // Set up Pointers to the activeThreads list for each stage
288 fetch.setActiveThreads(&activeThreads);
289 decode.setActiveThreads(&activeThreads);
290 rename.setActiveThreads(&activeThreads);
291 iew.setActiveThreads(&activeThreads);
292 commit.setActiveThreads(&activeThreads);
293
294 // Give each of the stages the time buffer they will use.
295 fetch.setTimeBuffer(&timeBuffer);
296 decode.setTimeBuffer(&timeBuffer);
297 rename.setTimeBuffer(&timeBuffer);
298 iew.setTimeBuffer(&timeBuffer);
299 commit.setTimeBuffer(&timeBuffer);
300
301 // Also setup each of the stages' queues.
302 fetch.setFetchQueue(&fetchQueue);
303 decode.setFetchQueue(&fetchQueue);
304 commit.setFetchQueue(&fetchQueue);
305 decode.setDecodeQueue(&decodeQueue);
306 rename.setDecodeQueue(&decodeQueue);
307 rename.setRenameQueue(&renameQueue);
308 iew.setRenameQueue(&renameQueue);
309 iew.setIEWQueue(&iewQueue);
310 commit.setIEWQueue(&iewQueue);
311 commit.setRenameQueue(&renameQueue);
312
313 commit.setIEWStage(&iew);
314 rename.setIEWStage(&iew);
315 rename.setCommitStage(&commit);
316
317 ThreadID active_threads;
318 if (FullSystem) {
319 active_threads = 1;
320 } else {
321 active_threads = params->workload.size();
322
323 if (active_threads > Impl::MaxThreads) {
324 panic("Workload Size too large. Increase the 'MaxThreads' "
325 "constant in your O3CPU impl. file (e.g. o3/alpha/impl.hh) "
326 "or edit your workload size.");
327 }
328 }
329
330 //Make Sure That this a Valid Architeture
331 assert(params->numPhysIntRegs >= numThreads * TheISA::NumIntRegs);
332 assert(params->numPhysFloatRegs >= numThreads * TheISA::NumFloatRegs);
333
334 rename.setScoreboard(&scoreboard);
335 iew.setScoreboard(&scoreboard);
336
337 // Setup the rename map for whichever stages need it.
338 PhysRegIndex lreg_idx = 0;
339 PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs
340
341 for (ThreadID tid = 0; tid < numThreads; tid++) {
342 bool bindRegs = (tid <= active_threads - 1);
343
344 isa[tid] = params->isa[tid];
345
346 commitRenameMap[tid].init(TheISA::NumIntRegs,
347 params->numPhysIntRegs,
348 lreg_idx, //Index for Logical. Regs
349
350 TheISA::NumFloatRegs,
351 params->numPhysFloatRegs,
352 freg_idx, //Index for Float Regs
353
354 TheISA::NumMiscRegs,
355
356 TheISA::ZeroReg,
357 TheISA::ZeroReg,
358
359 tid,
360 false);
361
362 renameMap[tid].init(TheISA::NumIntRegs,
363 params->numPhysIntRegs,
364 lreg_idx, //Index for Logical. Regs
365
366 TheISA::NumFloatRegs,
367 params->numPhysFloatRegs,
368 freg_idx, //Index for Float Regs
369
370 TheISA::NumMiscRegs,
371
372 TheISA::ZeroReg,
373 TheISA::ZeroReg,
374
375 tid,
376 bindRegs);
377
378 activateThreadEvent[tid].init(tid, this);
379 deallocateContextEvent[tid].init(tid, this);
380 }
381
382 rename.setRenameMap(renameMap);
383 commit.setRenameMap(commitRenameMap);
384
385 // Give renameMap & rename stage access to the freeList;
386 for (ThreadID tid = 0; tid < numThreads; tid++)
387 renameMap[tid].setFreeList(&freeList);
388 rename.setFreeList(&freeList);
389
390 // Setup the ROB for whichever stages need it.
391 commit.setROB(&rob);
392
393 lastActivatedCycle = 0;
394#if 0
395 // Give renameMap & rename stage access to the freeList;
396 for (ThreadID tid = 0; tid < numThreads; tid++)
397 globalSeqNum[tid] = 1;
398#endif
399
400 contextSwitch = false;
401 DPRINTF(O3CPU, "Creating O3CPU object.\n");
402
403 // Setup any thread state.
404 this->thread.resize(this->numThreads);
405
406 for (ThreadID tid = 0; tid < this->numThreads; ++tid) {
407 if (FullSystem) {
408 // SMT is not supported in FS mode yet.
409 assert(this->numThreads == 1);
410 this->thread[tid] = new Thread(this, 0, NULL);
411 } else {
412 if (tid < params->workload.size()) {
413 DPRINTF(O3CPU, "Workload[%i] process is %#x",
414 tid, this->thread[tid]);
415 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
416 (typename Impl::O3CPU *)(this),
417 tid, params->workload[tid]);
418
419 //usedTids[tid] = true;
420 //threadMap[tid] = tid;
421 } else {
422 //Allocate Empty thread so M5 can use later
423 //when scheduling threads to CPU
424 Process* dummy_proc = NULL;
425
426 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
427 (typename Impl::O3CPU *)(this),
428 tid, dummy_proc);
429 //usedTids[tid] = false;
430 }
431 }
432
433 ThreadContext *tc;
434
435 // Setup the TC that will serve as the interface to the threads/CPU.
436 O3ThreadContext<Impl> *o3_tc = new O3ThreadContext<Impl>;
437
438 tc = o3_tc;
439
440 // If we're using a checker, then the TC should be the
441 // CheckerThreadContext.
442 if (params->checker) {
443 tc = new CheckerThreadContext<O3ThreadContext<Impl> >(
444 o3_tc, this->checker);
445 }
446
447 o3_tc->cpu = (typename Impl::O3CPU *)(this);
448 assert(o3_tc->cpu);
449 o3_tc->thread = this->thread[tid];
450
451 if (FullSystem) {
452 // Setup quiesce event.
453 this->thread[tid]->quiesceEvent = new EndQuiesceEvent(tc);
454 }
455 // Give the thread the TC.
456 this->thread[tid]->tc = tc;
457
458 // Add the TC to the CPU's list of TC's.
459 this->threadContexts.push_back(tc);
460 }
461
462 // FullO3CPU always requires an interrupt controller.
463 if (!params->switched_out && !interrupts) {
464 fatal("FullO3CPU %s has no interrupt controller.\n"
465 "Ensure createInterruptController() is called.\n", name());
466 }
467
468 for (ThreadID tid = 0; tid < this->numThreads; tid++)
469 this->thread[tid]->setFuncExeInst(0);
470}
471
472template <class Impl>
473FullO3CPU<Impl>::~FullO3CPU()
474{
475}
476
477template <class Impl>
478void
479FullO3CPU<Impl>::regStats()
480{
481 BaseO3CPU::regStats();
482
483 // Register any of the O3CPU's stats here.
484 timesIdled
485 .name(name() + ".timesIdled")
486 .desc("Number of times that the entire CPU went into an idle state and"
487 " unscheduled itself")
488 .prereq(timesIdled);
489
490 idleCycles
491 .name(name() + ".idleCycles")
492 .desc("Total number of cycles that the CPU has spent unscheduled due "
493 "to idling")
494 .prereq(idleCycles);
495
496 quiesceCycles
497 .name(name() + ".quiesceCycles")
498 .desc("Total number of cycles that CPU has spent quiesced or waiting "
499 "for an interrupt")
500 .prereq(quiesceCycles);
501
502 // Number of Instructions simulated
503 // --------------------------------
504 // Should probably be in Base CPU but need templated
505 // MaxThreads so put in here instead
506 committedInsts
507 .init(numThreads)
508 .name(name() + ".committedInsts")
509 .desc("Number of Instructions Simulated");
510
511 committedOps
512 .init(numThreads)
513 .name(name() + ".committedOps")
514 .desc("Number of Ops (including micro ops) Simulated");
515
516 totalCommittedInsts
517 .name(name() + ".committedInsts_total")
518 .desc("Number of Instructions Simulated");
519
520 cpi
521 .name(name() + ".cpi")
522 .desc("CPI: Cycles Per Instruction")
523 .precision(6);
524 cpi = numCycles / committedInsts;
525
526 totalCpi
527 .name(name() + ".cpi_total")
528 .desc("CPI: Total CPI of All Threads")
529 .precision(6);
530 totalCpi = numCycles / totalCommittedInsts;
531
532 ipc
533 .name(name() + ".ipc")
534 .desc("IPC: Instructions Per Cycle")
535 .precision(6);
536 ipc = committedInsts / numCycles;
537
538 totalIpc
539 .name(name() + ".ipc_total")
540 .desc("IPC: Total IPC of All Threads")
541 .precision(6);
542 totalIpc = totalCommittedInsts / numCycles;
543
544 this->fetch.regStats();
545 this->decode.regStats();
546 this->rename.regStats();
547 this->iew.regStats();
548 this->commit.regStats();
549 this->rob.regStats();
550
551 intRegfileReads
552 .name(name() + ".int_regfile_reads")
553 .desc("number of integer regfile reads")
554 .prereq(intRegfileReads);
555
556 intRegfileWrites
557 .name(name() + ".int_regfile_writes")
558 .desc("number of integer regfile writes")
559 .prereq(intRegfileWrites);
560
561 fpRegfileReads
562 .name(name() + ".fp_regfile_reads")
563 .desc("number of floating regfile reads")
564 .prereq(fpRegfileReads);
565
566 fpRegfileWrites
567 .name(name() + ".fp_regfile_writes")
568 .desc("number of floating regfile writes")
569 .prereq(fpRegfileWrites);
570
571 miscRegfileReads
572 .name(name() + ".misc_regfile_reads")
573 .desc("number of misc regfile reads")
574 .prereq(miscRegfileReads);
575
576 miscRegfileWrites
577 .name(name() + ".misc_regfile_writes")
578 .desc("number of misc regfile writes")
579 .prereq(miscRegfileWrites);
580}
581
582template <class Impl>
583void
584FullO3CPU<Impl>::tick()
585{
586 DPRINTF(O3CPU, "\n\nFullO3CPU: Ticking main, FullO3CPU.\n");
587 assert(!switchedOut());
588 assert(getDrainState() != Drainable::Drained);
589
590 ++numCycles;
591
592// activity = false;
593
594 //Tick each of the stages
595 fetch.tick();
596
597 decode.tick();
598
599 rename.tick();
600
601 iew.tick();
602
603 commit.tick();
604
605 if (!FullSystem)
606 doContextSwitch();
607
608 // Now advance the time buffers
609 timeBuffer.advance();
610
611 fetchQueue.advance();
612 decodeQueue.advance();
613 renameQueue.advance();
614 iewQueue.advance();
615
616 activityRec.advance();
617
618 if (removeInstsThisCycle) {
619 cleanUpRemovedInsts();
620 }
621
622 if (!tickEvent.scheduled()) {
623 if (_status == SwitchedOut) {
624 DPRINTF(O3CPU, "Switched out!\n");
625 // increment stat
626 lastRunningCycle = curCycle();
627 } else if (!activityRec.active() || _status == Idle) {
628 DPRINTF(O3CPU, "Idle!\n");
629 lastRunningCycle = curCycle();
630 timesIdled++;
631 } else {
632 schedule(tickEvent, clockEdge(Cycles(1)));
633 DPRINTF(O3CPU, "Scheduling next tick!\n");
634 }
635 }
636
637 if (!FullSystem)
638 updateThreadPriority();
639
640 tryDrain();
641}
642
643template <class Impl>
644void
645FullO3CPU<Impl>::init()
646{
647 BaseCPU::init();
648
649 for (ThreadID tid = 0; tid < numThreads; ++tid) {
650 // Set noSquashFromTC so that the CPU doesn't squash when initially
651 // setting up registers.
652 thread[tid]->noSquashFromTC = true;
653 // Initialise the ThreadContext's memory proxies
654 thread[tid]->initMemProxies(thread[tid]->getTC());
655 }
656
657 if (FullSystem && !params()->switched_out) {
658 for (ThreadID tid = 0; tid < numThreads; tid++) {
659 ThreadContext *src_tc = threadContexts[tid];
660 TheISA::initCPU(src_tc, src_tc->contextId());
661 }
662 }
663
664 // Clear noSquashFromTC.
665 for (int tid = 0; tid < numThreads; ++tid)
666 thread[tid]->noSquashFromTC = false;
667
668 commit.setThreads(thread);
669}
670
671template <class Impl>
672void
673FullO3CPU<Impl>::startup()
674{
675 for (int tid = 0; tid < numThreads; ++tid)
676 isa[tid]->startup(threadContexts[tid]);
677
678 fetch.startupStage();
679 decode.startupStage();
680 iew.startupStage();
681 rename.startupStage();
682 commit.startupStage();
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 assert(!switchedOut());
694
695 if (isActive == activeThreads.end()) {
696 DPRINTF(O3CPU, "[tid:%i]: Adding to active threads list\n",
697 tid);
698
699 activeThreads.push_back(tid);
700 }
701}
702
703template <class Impl>
704void
705FullO3CPU<Impl>::deactivateThread(ThreadID tid)
706{
707 //Remove From Active List, if Active
708 list<ThreadID>::iterator thread_it =
709 std::find(activeThreads.begin(), activeThreads.end(), tid);
710
711 DPRINTF(O3CPU, "[tid:%i]: Calling deactivate thread.\n", tid);
712 assert(!switchedOut());
713
714 if (thread_it != activeThreads.end()) {
715 DPRINTF(O3CPU,"[tid:%i]: Removing from active threads list\n",
716 tid);
717 activeThreads.erase(thread_it);
718 }
719}
720
721template <class Impl>
722Counter
723FullO3CPU<Impl>::totalInsts() const
724{
725 Counter total(0);
726
727 ThreadID size = thread.size();
728 for (ThreadID i = 0; i < size; i++)
729 total += thread[i]->numInst;
730
731 return total;
732}
733
734template <class Impl>
735Counter
736FullO3CPU<Impl>::totalOps() const
737{
738 Counter total(0);
739
740 ThreadID size = thread.size();
741 for (ThreadID i = 0; i < size; i++)
742 total += thread[i]->numOp;
743
744 return total;
745}
746
747template <class Impl>
748void
749FullO3CPU<Impl>::activateContext(ThreadID tid, Cycles delay)
750{
751 assert(!switchedOut());
752
753 // Needs to set each stage to running as well.
754 if (delay){
755 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to activate "
756 "on cycle %d\n", tid, clockEdge(delay));
757 scheduleActivateThreadEvent(tid, delay);
758 } else {
759 activateThread(tid);
760 }
761
762 // We don't want to wake the CPU if it is drained. In that case,
763 // we just want to flag the thread as active and schedule the tick
764 // event from drainResume() instead.
765 if (getDrainState() == Drainable::Drained)
766 return;
767
768 // If we are time 0 or if the last activation time is in the past,
769 // schedule the next tick and wake up the fetch unit
770 if (lastActivatedCycle == 0 || lastActivatedCycle < curTick()) {
771 scheduleTickEvent(delay);
772
773 // Be sure to signal that there's some activity so the CPU doesn't
774 // deschedule itself.
775 activityRec.activity();
776 fetch.wakeFromQuiesce();
777
778 Cycles cycles(curCycle() - lastRunningCycle);
779 // @todo: This is an oddity that is only here to match the stats
780 if (cycles != 0)
781 --cycles;
782 quiesceCycles += cycles;
783
784 lastActivatedCycle = curTick();
785
786 _status = Running;
787 }
788}
789
790template <class Impl>
791bool
792FullO3CPU<Impl>::scheduleDeallocateContext(ThreadID tid, bool remove,
793 Cycles delay)
794{
795 // Schedule removal of thread data from CPU
796 if (delay){
797 DPRINTF(O3CPU, "[tid:%i]: Scheduling thread context to deallocate "
798 "on tick %d\n", tid, clockEdge(delay));
799 scheduleDeallocateContextEvent(tid, remove, delay);
800 return false;
801 } else {
802 deactivateThread(tid);
803 if (remove)
804 removeThread(tid);
805 return true;
806 }
807}
808
809template <class Impl>
810void
811FullO3CPU<Impl>::suspendContext(ThreadID tid)
812{
813 DPRINTF(O3CPU,"[tid: %i]: Suspending Thread Context.\n", tid);
814 assert(!switchedOut());
815 bool deallocated = scheduleDeallocateContext(tid, false, Cycles(1));
816 // If this was the last thread then unschedule the tick event.
817 if ((activeThreads.size() == 1 && !deallocated) ||
818 activeThreads.size() == 0)
819 unscheduleTickEvent();
820
821 DPRINTF(Quiesce, "Suspending Context\n");
822 lastRunningCycle = curCycle();
823 _status = Idle;
824}
825
826template <class Impl>
827void
828FullO3CPU<Impl>::haltContext(ThreadID tid)
829{
830 //For now, this is the same as deallocate
831 DPRINTF(O3CPU,"[tid:%i]: Halt Context called. Deallocating", tid);
832 assert(!switchedOut());
833 scheduleDeallocateContext(tid, true, Cycles(1));
834}
835
836template <class Impl>
837void
838FullO3CPU<Impl>::insertThread(ThreadID tid)
839{
840 DPRINTF(O3CPU,"[tid:%i] Initializing thread into CPU");
841 // Will change now that the PC and thread state is internal to the CPU
842 // and not in the ThreadContext.
843 ThreadContext *src_tc;
844 if (FullSystem)
845 src_tc = system->threadContexts[tid];
846 else
847 src_tc = tcBase(tid);
848
849 //Bind Int Regs to Rename Map
850 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
851 PhysRegIndex phys_reg = freeList.getIntReg();
852
853 renameMap[tid].setEntry(ireg,phys_reg);
854 scoreboard.setReg(phys_reg);
855 }
856
857 //Bind Float Regs to Rename Map
858 for (int freg = 0; freg < TheISA::NumFloatRegs; freg++) {
859 PhysRegIndex phys_reg = freeList.getFloatReg();
860
861 renameMap[tid].setEntry(freg,phys_reg);
862 scoreboard.setReg(phys_reg);
863 }
864
865 //Copy Thread Data Into RegFile
866 //this->copyFromTC(tid);
867
868 //Set PC/NPC/NNPC
869 pcState(src_tc->pcState(), tid);
870
871 src_tc->setStatus(ThreadContext::Active);
872
873 activateContext(tid, Cycles(1));
874
875 //Reset ROB/IQ/LSQ Entries
876 commit.rob->resetEntries();
877 iew.resetEntries();
878}
879
880template <class Impl>
881void
882FullO3CPU<Impl>::removeThread(ThreadID tid)
883{
884 DPRINTF(O3CPU,"[tid:%i] Removing thread context from CPU.\n", tid);
885
886 // Copy Thread Data From RegFile
887 // If thread is suspended, it might be re-allocated
888 // this->copyToTC(tid);
889
890
891 // @todo: 2-27-2008: Fix how we free up rename mappings
892 // here to alleviate the case for double-freeing registers
893 // in SMT workloads.
894
895 // Unbind Int Regs from Rename Map
896 for (int ireg = 0; ireg < TheISA::NumIntRegs; ireg++) {
897 PhysRegIndex phys_reg = renameMap[tid].lookup(ireg);
898
899 scoreboard.unsetReg(phys_reg);
900 freeList.addReg(phys_reg);
901 }
902
903 // Unbind Float Regs from Rename Map
904 for (int freg = TheISA::NumIntRegs; freg < TheISA::NumFloatRegs; freg++) {
905 PhysRegIndex phys_reg = renameMap[tid].lookup(freg);
906
907 scoreboard.unsetReg(phys_reg);
908 freeList.addReg(phys_reg);
909 }
910
911 // Squash Throughout Pipeline
912 DynInstPtr inst = commit.rob->readHeadInst(tid);
913 InstSeqNum squash_seq_num = inst->seqNum;
914 fetch.squash(0, squash_seq_num, inst, tid);
915 decode.squash(tid);
916 rename.squash(squash_seq_num, tid);
917 iew.squash(tid);
918 iew.ldstQueue.squash(squash_seq_num, tid);
919 commit.rob->squash(squash_seq_num, tid);
920
921
922 assert(iew.instQueue.getCount(tid) == 0);
923 assert(iew.ldstQueue.getCount(tid) == 0);
924
925 // Reset ROB/IQ/LSQ Entries
926
927 // Commented out for now. This should be possible to do by
928 // telling all the pipeline stages to drain first, and then
929 // checking until the drain completes. Once the pipeline is
930 // drained, call resetEntries(). - 10-09-06 ktlim
931/*
932 if (activeThreads.size() >= 1) {
933 commit.rob->resetEntries();
934 iew.resetEntries();
935 }
936*/
937}
938
939
940template <class Impl>
941void
942FullO3CPU<Impl>::activateWhenReady(ThreadID tid)
943{
944 DPRINTF(O3CPU,"[tid:%i]: Checking if resources are available for incoming"
945 "(e.g. PhysRegs/ROB/IQ/LSQ) \n",
946 tid);
947
948 bool ready = true;
949
950 if (freeList.numFreeIntRegs() >= TheISA::NumIntRegs) {
951 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
952 "Phys. Int. Regs.\n",
953 tid);
954 ready = false;
955 } else if (freeList.numFreeFloatRegs() >= TheISA::NumFloatRegs) {
956 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
957 "Phys. Float. Regs.\n",
958 tid);
959 ready = false;
960 } else if (commit.rob->numFreeEntries() >=
961 commit.rob->entryAmount(activeThreads.size() + 1)) {
962 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
963 "ROB entries.\n",
964 tid);
965 ready = false;
966 } else if (iew.instQueue.numFreeEntries() >=
967 iew.instQueue.entryAmount(activeThreads.size() + 1)) {
968 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
969 "IQ entries.\n",
970 tid);
971 ready = false;
972 } else if (iew.ldstQueue.numFreeEntries() >=
973 iew.ldstQueue.entryAmount(activeThreads.size() + 1)) {
974 DPRINTF(O3CPU,"[tid:%i] Suspending thread due to not enough "
975 "LSQ entries.\n",
976 tid);
977 ready = false;
978 }
979
980 if (ready) {
981 insertThread(tid);
982
983 contextSwitch = false;
984
985 cpuWaitList.remove(tid);
986 } else {
987 suspendContext(tid);
988
989 //blocks fetch
990 contextSwitch = true;
991
992 //@todo: dont always add to waitlist
993 //do waitlist
994 cpuWaitList.push_back(tid);
995 }
996}
997
998template <class Impl>
999Fault
1000FullO3CPU<Impl>::hwrei(ThreadID tid)
1001{
1002#if THE_ISA == ALPHA_ISA
1003 // Need to clear the lock flag upon returning from an interrupt.
1004 this->setMiscRegNoEffect(AlphaISA::MISCREG_LOCKFLAG, false, tid);
1005
1006 this->thread[tid]->kernelStats->hwrei();
1007
1008 // FIXME: XXX check for interrupts? XXX
1009#endif
1010 return NoFault;
1011}
1012
1013template <class Impl>
1014bool
1015FullO3CPU<Impl>::simPalCheck(int palFunc, ThreadID tid)
1016{
1017#if THE_ISA == ALPHA_ISA
1018 if (this->thread[tid]->kernelStats)
1019 this->thread[tid]->kernelStats->callpal(palFunc,
1020 this->threadContexts[tid]);
1021
1022 switch (palFunc) {
1023 case PAL::halt:
1024 halt();
1025 if (--System::numSystemsRunning == 0)
1026 exitSimLoop("all cpus halted");
1027 break;
1028
1029 case PAL::bpt:
1030 case PAL::bugchk:
1031 if (this->system->breakpoint())
1032 return false;
1033 break;
1034 }
1035#endif
1036 return true;
1037}
1038
1039template <class Impl>
1040Fault
1041FullO3CPU<Impl>::getInterrupts()
1042{
1043 // Check if there are any outstanding interrupts
1044 return this->interrupts->getInterrupt(this->threadContexts[0]);
1045}
1046
1047template <class Impl>
1048void
1049FullO3CPU<Impl>::processInterrupts(Fault interrupt)
1050{
1051 // Check for interrupts here. For now can copy the code that
1052 // exists within isa_fullsys_traits.hh. Also assume that thread 0
1053 // is the one that handles the interrupts.
1054 // @todo: Possibly consolidate the interrupt checking code.
1055 // @todo: Allow other threads to handle interrupts.
1056
1057 assert(interrupt != NoFault);
1058 this->interrupts->updateIntrInfo(this->threadContexts[0]);
1059
1060 DPRINTF(O3CPU, "Interrupt %s being handled\n", interrupt->name());
1061 this->trap(interrupt, 0, NULL);
1062}
1063
1064template <class Impl>
1065void
1066FullO3CPU<Impl>::trap(Fault fault, ThreadID tid, StaticInstPtr inst)
1067{
1068 // Pass the thread's TC into the invoke method.
1069 fault->invoke(this->threadContexts[tid], inst);
1070}
1071
1072template <class Impl>
1073void
1074FullO3CPU<Impl>::syscall(int64_t callnum, ThreadID tid)
1075{
1076 DPRINTF(O3CPU, "[tid:%i] Executing syscall().\n\n", tid);
1077
1078 DPRINTF(Activity,"Activity: syscall() called.\n");
1079
1080 // Temporarily increase this by one to account for the syscall
1081 // instruction.
1082 ++(this->thread[tid]->funcExeInst);
1083
1084 // Execute the actual syscall.
1085 this->thread[tid]->syscall(callnum);
1086
1087 // Decrease funcExeInst by one as the normal commit will handle
1088 // incrementing it.
1089 --(this->thread[tid]->funcExeInst);
1090}
1091
1092template <class Impl>
1093void
1094FullO3CPU<Impl>::serializeThread(std::ostream &os, ThreadID tid)
1095{
1096 thread[tid]->serialize(os);
1097}
1098
1099template <class Impl>
1100void
1101FullO3CPU<Impl>::unserializeThread(Checkpoint *cp, const std::string &section,
1102 ThreadID tid)
1103{
1104 thread[tid]->unserialize(cp, section);
1105}
1106
1107template <class Impl>
1108unsigned int
1109FullO3CPU<Impl>::drain(DrainManager *drain_manager)
1110{
1111 // If the CPU isn't doing anything, then return immediately.
1112 if (switchedOut()) {
1113 setDrainState(Drainable::Drained);
1114 return 0;
1115 }
1116
1117 DPRINTF(Drain, "Draining...\n");
1118 setDrainState(Drainable::Draining);
1119
1120 // We only need to signal a drain to the commit stage as this
1121 // initiates squashing controls the draining. Once the commit
1122 // stage commits an instruction where it is safe to stop, it'll
1123 // squash the rest of the instructions in the pipeline and force
1124 // the fetch stage to stall. The pipeline will be drained once all
1125 // in-flight instructions have retired.
1126 commit.drain();
1127
1128 // Wake the CPU and record activity so everything can drain out if
1129 // the CPU was not able to immediately drain.
1130 if (!isDrained()) {
1131 drainManager = drain_manager;
1132
1133 wakeCPU();
1134 activityRec.activity();
1135
1136 DPRINTF(Drain, "CPU not drained\n");
1137
1138 return 1;
1139 } else {
1140 setDrainState(Drainable::Drained);
1141 DPRINTF(Drain, "CPU is already drained\n");
1142 if (tickEvent.scheduled())
1143 deschedule(tickEvent);
1144
1145 // Flush out any old data from the time buffers. In
1146 // particular, there might be some data in flight from the
1147 // fetch stage that isn't visible in any of the CPU buffers we
1148 // test in isDrained().
1149 for (int i = 0; i < timeBuffer.getSize(); ++i) {
1150 timeBuffer.advance();
1151 fetchQueue.advance();
1152 decodeQueue.advance();
1153 renameQueue.advance();
1154 iewQueue.advance();
1155 }
1156
1157 drainSanityCheck();
1158 return 0;
1159 }
1160}
1161
1162template <class Impl>
1163bool
1164FullO3CPU<Impl>::tryDrain()
1165{
1166 if (!drainManager || !isDrained())
1167 return false;
1168
1169 if (tickEvent.scheduled())
1170 deschedule(tickEvent);
1171
1172 DPRINTF(Drain, "CPU done draining, processing drain event\n");
1173 drainManager->signalDrainDone();
1174 drainManager = NULL;
1175
1176 return true;
1177}
1178
1179template <class Impl>
1180void
1181FullO3CPU<Impl>::drainSanityCheck() const
1182{
1183 assert(isDrained());
1184 fetch.drainSanityCheck();
1185 decode.drainSanityCheck();
1186 rename.drainSanityCheck();
1187 iew.drainSanityCheck();
1188 commit.drainSanityCheck();
1189}
1190
1191template <class Impl>
1192bool
1193FullO3CPU<Impl>::isDrained() const
1194{
1195 bool drained(true);
1196
1197 for (ThreadID i = 0; i < thread.size(); ++i) {
1198 if (activateThreadEvent[i].scheduled()) {
1199 DPRINTF(Drain, "CPU not drained, tread %i has a "
1200 "pending activate event\n", i);
1201 drained = false;
1202 }
1203 if (deallocateContextEvent[i].scheduled()) {
1204 DPRINTF(Drain, "CPU not drained, tread %i has a "
1205 "pending deallocate context event\n", i);
1206 drained = false;
1207 }
1208 }
1209
1210 if (!instList.empty() || !removeList.empty()) {
1211 DPRINTF(Drain, "Main CPU structures not drained.\n");
1212 drained = false;
1213 }
1214
1215 if (!fetch.isDrained()) {
1216 DPRINTF(Drain, "Fetch not drained.\n");
1217 drained = false;
1218 }
1219
1220 if (!decode.isDrained()) {
1221 DPRINTF(Drain, "Decode not drained.\n");
1222 drained = false;
1223 }
1224
1225 if (!rename.isDrained()) {
1226 DPRINTF(Drain, "Rename not drained.\n");
1227 drained = false;
1228 }
1229
1230 if (!iew.isDrained()) {
1231 DPRINTF(Drain, "IEW not drained.\n");
1232 drained = false;
1233 }
1234
1235 if (!commit.isDrained()) {
1236 DPRINTF(Drain, "Commit not drained.\n");
1237 drained = false;
1238 }
1239
1240 return drained;
1241}
1242
1243template <class Impl>
1244void
1245FullO3CPU<Impl>::commitDrained(ThreadID tid)
1246{
1247 fetch.drainStall(tid);
1248}
1249
1250template <class Impl>
1251void
1252FullO3CPU<Impl>::drainResume()
1253{
1254 setDrainState(Drainable::Running);
1255 if (switchedOut())
1256 return;
1257
1258 DPRINTF(Drain, "Resuming...\n");
1259 verifyMemoryMode();
1260
1261 fetch.drainResume();
1262 commit.drainResume();
1263
1264 _status = Idle;
1265 for (ThreadID i = 0; i < thread.size(); i++) {
1266 if (thread[i]->status() == ThreadContext::Active) {
1267 DPRINTF(Drain, "Activating thread: %i\n", i);
1268 activateThread(i);
1269 _status = Running;
1270 }
1271 }
1272
1273 assert(!tickEvent.scheduled());
1274 if (_status == Running)
1275 schedule(tickEvent, nextCycle());
1276}
1277
1278template <class Impl>
1279void
1280FullO3CPU<Impl>::switchOut()
1281{
1282 DPRINTF(O3CPU, "Switching out\n");
1283 BaseCPU::switchOut();
1284
1285 activityRec.reset();
1286
1287 _status = SwitchedOut;
1288
1289 if (checker)
1290 checker->switchOut();
1291}
1292
1293template <class Impl>
1294void
1295FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU)
1296{
1297 BaseCPU::takeOverFrom(oldCPU);
1298
1299 fetch.takeOverFrom();
1300 decode.takeOverFrom();
1301 rename.takeOverFrom();
1302 iew.takeOverFrom();
1303 commit.takeOverFrom();
1304
1305 assert(!tickEvent.scheduled());
1306
1307 FullO3CPU<Impl> *oldO3CPU = dynamic_cast<FullO3CPU<Impl>*>(oldCPU);
1308 if (oldO3CPU)
1309 globalSeqNum = oldO3CPU->globalSeqNum;
1310
1311 lastRunningCycle = curCycle();
1312 _status = Idle;
1313}
1314
1315template <class Impl>
1316void
1317FullO3CPU<Impl>::verifyMemoryMode() const
1318{
1319 if (!system->isTimingMode()) {
1320 fatal("The O3 CPU requires the memory system to be in "
1321 "'timing' mode.\n");
1322 }
1323}
1324
1325template <class Impl>
1326TheISA::MiscReg
1327FullO3CPU<Impl>::readMiscRegNoEffect(int misc_reg, ThreadID tid)
1328{
1329 return this->isa[tid]->readMiscRegNoEffect(misc_reg);
1330}
1331
1332template <class Impl>
1333TheISA::MiscReg
1334FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid)
1335{
1336 miscRegfileReads++;
1337 return this->isa[tid]->readMiscReg(misc_reg, tcBase(tid));
1338}
1339
1340template <class Impl>
1341void
1342FullO3CPU<Impl>::setMiscRegNoEffect(int misc_reg,
1343 const TheISA::MiscReg &val, ThreadID tid)
1344{
1345 this->isa[tid]->setMiscRegNoEffect(misc_reg, val);
1346}
1347
1348template <class Impl>
1349void
1350FullO3CPU<Impl>::setMiscReg(int misc_reg,
1351 const TheISA::MiscReg &val, ThreadID tid)
1352{
1353 miscRegfileWrites++;
1354 this->isa[tid]->setMiscReg(misc_reg, val, tcBase(tid));
1355}
1356
1357template <class Impl>
1358uint64_t
1359FullO3CPU<Impl>::readIntReg(int reg_idx)
1360{
1361 intRegfileReads++;
1362 return regFile.readIntReg(reg_idx);
1363}
1364
1365template <class Impl>
1366FloatReg
1367FullO3CPU<Impl>::readFloatReg(int reg_idx)
1368{
1369 fpRegfileReads++;
1370 return regFile.readFloatReg(reg_idx);
1371}
1372
1373template <class Impl>
1374FloatRegBits
1375FullO3CPU<Impl>::readFloatRegBits(int reg_idx)
1376{
1377 fpRegfileReads++;
1378 return regFile.readFloatRegBits(reg_idx);
1379}
1380
1381template <class Impl>
1382void
1383FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val)
1384{
1385 intRegfileWrites++;
1386 regFile.setIntReg(reg_idx, val);
1387}
1388
1389template <class Impl>
1390void
1391FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val)
1392{
1393 fpRegfileWrites++;
1394 regFile.setFloatReg(reg_idx, val);
1395}
1396
1397template <class Impl>
1398void
1399FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val)
1400{
1401 fpRegfileWrites++;
1402 regFile.setFloatRegBits(reg_idx, val);
1403}
1404
1405template <class Impl>
1406uint64_t
1407FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid)
1408{
1409 intRegfileReads++;
1410 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1411
1412 return regFile.readIntReg(phys_reg);
1413}
1414
1415template <class Impl>
1416float
1417FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid)
1418{
1419 fpRegfileReads++;
1420 int idx = reg_idx + TheISA::NumIntRegs;
1421 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1422
1423 return regFile.readFloatReg(phys_reg);
1424}
1425
1426template <class Impl>
1427uint64_t
1428FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid)
1429{
1430 fpRegfileReads++;
1431 int idx = reg_idx + TheISA::NumIntRegs;
1432 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1433
1434 return regFile.readFloatRegBits(phys_reg);
1435}
1436
1437template <class Impl>
1438void
1439FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid)
1440{
1441 intRegfileWrites++;
1442 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
1443
1444 regFile.setIntReg(phys_reg, val);
1445}
1446
1447template <class Impl>
1448void
1449FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid)
1450{
1451 fpRegfileWrites++;
1452 int idx = reg_idx + TheISA::NumIntRegs;
1453 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1454
1455 regFile.setFloatReg(phys_reg, val);
1456}
1457
1458template <class Impl>
1459void
1460FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid)
1461{
1462 fpRegfileWrites++;
1463 int idx = reg_idx + TheISA::NumIntRegs;
1464 PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
1465
1466 regFile.setFloatRegBits(phys_reg, val);
1467}
1468
1469template <class Impl>
1470TheISA::PCState
1471FullO3CPU<Impl>::pcState(ThreadID tid)
1472{
1473 return commit.pcState(tid);
1474}
1475
1476template <class Impl>
1477void
1478FullO3CPU<Impl>::pcState(const TheISA::PCState &val, ThreadID tid)
1479{
1480 commit.pcState(val, tid);
1481}
1482
1483template <class Impl>
1484Addr
1485FullO3CPU<Impl>::instAddr(ThreadID tid)
1486{
1487 return commit.instAddr(tid);
1488}
1489
1490template <class Impl>
1491Addr
1492FullO3CPU<Impl>::nextInstAddr(ThreadID tid)
1493{
1494 return commit.nextInstAddr(tid);
1495}
1496
1497template <class Impl>
1498MicroPC
1499FullO3CPU<Impl>::microPC(ThreadID tid)
1500{
1501 return commit.microPC(tid);
1502}
1503
1504template <class Impl>
1505void
1506FullO3CPU<Impl>::squashFromTC(ThreadID tid)
1507{
1508 this->thread[tid]->noSquashFromTC = true;
1509 this->commit.generateTCEvent(tid);
1510}
1511
1512template <class Impl>
1513typename FullO3CPU<Impl>::ListIt
1514FullO3CPU<Impl>::addInst(DynInstPtr &inst)
1515{
1516 instList.push_back(inst);
1517
1518 return --(instList.end());
1519}
1520
1521template <class Impl>
1522void
1523FullO3CPU<Impl>::instDone(ThreadID tid, DynInstPtr &inst)
1524{
1525 // Keep an instruction count.
1526 if (!inst->isMicroop() || inst->isLastMicroop()) {
1527 thread[tid]->numInst++;
1528 thread[tid]->numInsts++;
1529 committedInsts[tid]++;
1530 totalCommittedInsts++;
1531 }
1532 thread[tid]->numOp++;
1533 thread[tid]->numOps++;
1534 committedOps[tid]++;
1535
1536 system->totalNumInsts++;
1537 // Check for instruction-count-based events.
1538 comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst);
1539 system->instEventQueue.serviceEvents(system->totalNumInsts);
1540}
1541
1542template <class Impl>
1543void
1544FullO3CPU<Impl>::removeFrontInst(DynInstPtr &inst)
1545{
1546 DPRINTF(O3CPU, "Removing committed instruction [tid:%i] PC %s "
1547 "[sn:%lli]\n",
1548 inst->threadNumber, inst->pcState(), inst->seqNum);
1549
1550 removeInstsThisCycle = true;
1551
1552 // Remove the front instruction.
1553 removeList.push(inst->getInstListIt());
1554}
1555
1556template <class Impl>
1557void
1558FullO3CPU<Impl>::removeInstsNotInROB(ThreadID tid)
1559{
1560 DPRINTF(O3CPU, "Thread %i: Deleting instructions from instruction"
1561 " list.\n", tid);
1562
1563 ListIt end_it;
1564
1565 bool rob_empty = false;
1566
1567 if (instList.empty()) {
1568 return;
1569 } else if (rob.isEmpty(/*tid*/)) {
1570 DPRINTF(O3CPU, "ROB is empty, squashing all insts.\n");
1571 end_it = instList.begin();
1572 rob_empty = true;
1573 } else {
1574 end_it = (rob.readTailInst(tid))->getInstListIt();
1575 DPRINTF(O3CPU, "ROB is not empty, squashing insts not in ROB.\n");
1576 }
1577
1578 removeInstsThisCycle = true;
1579
1580 ListIt inst_it = instList.end();
1581
1582 inst_it--;
1583
1584 // Walk through the instruction list, removing any instructions
1585 // that were inserted after the given instruction iterator, end_it.
1586 while (inst_it != end_it) {
1587 assert(!instList.empty());
1588
1589 squashInstIt(inst_it, tid);
1590
1591 inst_it--;
1592 }
1593
1594 // If the ROB was empty, then we actually need to remove the first
1595 // instruction as well.
1596 if (rob_empty) {
1597 squashInstIt(inst_it, tid);
1598 }
1599}
1600
1601template <class Impl>
1602void
1603FullO3CPU<Impl>::removeInstsUntil(const InstSeqNum &seq_num, ThreadID tid)
1604{
1605 assert(!instList.empty());
1606
1607 removeInstsThisCycle = true;
1608
1609 ListIt inst_iter = instList.end();
1610
1611 inst_iter--;
1612
1613 DPRINTF(O3CPU, "Deleting instructions from instruction "
1614 "list that are from [tid:%i] and above [sn:%lli] (end=%lli).\n",
1615 tid, seq_num, (*inst_iter)->seqNum);
1616
1617 while ((*inst_iter)->seqNum > seq_num) {
1618
1619 bool break_loop = (inst_iter == instList.begin());
1620
1621 squashInstIt(inst_iter, tid);
1622
1623 inst_iter--;
1624
1625 if (break_loop)
1626 break;
1627 }
1628}
1629
1630template <class Impl>
1631inline void
1632FullO3CPU<Impl>::squashInstIt(const ListIt &instIt, ThreadID tid)
1633{
1634 if ((*instIt)->threadNumber == tid) {
1635 DPRINTF(O3CPU, "Squashing instruction, "
1636 "[tid:%i] [sn:%lli] PC %s\n",
1637 (*instIt)->threadNumber,
1638 (*instIt)->seqNum,
1639 (*instIt)->pcState());
1640
1641 // Mark it as squashed.
1642 (*instIt)->setSquashed();
1643
1644 // @todo: Formulate a consistent method for deleting
1645 // instructions from the instruction list
1646 // Remove the instruction from the list.
1647 removeList.push(instIt);
1648 }
1649}
1650
1651template <class Impl>
1652void
1653FullO3CPU<Impl>::cleanUpRemovedInsts()
1654{
1655 while (!removeList.empty()) {
1656 DPRINTF(O3CPU, "Removing instruction, "
1657 "[tid:%i] [sn:%lli] PC %s\n",
1658 (*removeList.front())->threadNumber,
1659 (*removeList.front())->seqNum,
1660 (*removeList.front())->pcState());
1661
1662 instList.erase(removeList.front());
1663
1664 removeList.pop();
1665 }
1666
1667 removeInstsThisCycle = false;
1668}
1669/*
1670template <class Impl>
1671void
1672FullO3CPU<Impl>::removeAllInsts()
1673{
1674 instList.clear();
1675}
1676*/
1677template <class Impl>
1678void
1679FullO3CPU<Impl>::dumpInsts()
1680{
1681 int num = 0;
1682
1683 ListIt inst_list_it = instList.begin();
1684
1685 cprintf("Dumping Instruction List\n");
1686
1687 while (inst_list_it != instList.end()) {
1688 cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
1689 "Squashed:%i\n\n",
1690 num, (*inst_list_it)->instAddr(), (*inst_list_it)->threadNumber,
1691 (*inst_list_it)->seqNum, (*inst_list_it)->isIssued(),
1692 (*inst_list_it)->isSquashed());
1693 inst_list_it++;
1694 ++num;
1695 }
1696}
1697/*
1698template <class Impl>
1699void
1700FullO3CPU<Impl>::wakeDependents(DynInstPtr &inst)
1701{
1702 iew.wakeDependents(inst);
1703}
1704*/
1705template <class Impl>
1706void
1707FullO3CPU<Impl>::wakeCPU()
1708{
1709 if (activityRec.active() || tickEvent.scheduled()) {
1710 DPRINTF(Activity, "CPU already running.\n");
1711 return;
1712 }
1713
1714 DPRINTF(Activity, "Waking up CPU\n");
1715
1716 Cycles cycles(curCycle() - lastRunningCycle);
1717 // @todo: This is an oddity that is only here to match the stats
1718 if (cycles != 0)
1719 --cycles;
1720 idleCycles += cycles;
1721 numCycles += cycles;
1722
1723 schedule(tickEvent, nextCycle());
1723 schedule(tickEvent, clockEdge());
1724}
1725
1726template <class Impl>
1727void
1728FullO3CPU<Impl>::wakeup()
1729{
1730 if (this->thread[0]->status() != ThreadContext::Suspended)
1731 return;
1732
1733 this->wakeCPU();
1734
1735 DPRINTF(Quiesce, "Suspended Processor woken\n");
1736 this->threadContexts[0]->activate();
1737}
1738
1739template <class Impl>
1740ThreadID
1741FullO3CPU<Impl>::getFreeTid()
1742{
1743 for (ThreadID tid = 0; tid < numThreads; tid++) {
1744 if (!tids[tid]) {
1745 tids[tid] = true;
1746 return tid;
1747 }
1748 }
1749
1750 return InvalidThreadID;
1751}
1752
1753template <class Impl>
1754void
1755FullO3CPU<Impl>::doContextSwitch()
1756{
1757 if (contextSwitch) {
1758
1759 //ADD CODE TO DEACTIVE THREAD HERE (???)
1760
1761 ThreadID size = cpuWaitList.size();
1762 for (ThreadID tid = 0; tid < size; tid++) {
1763 activateWhenReady(tid);
1764 }
1765
1766 if (cpuWaitList.size() == 0)
1767 contextSwitch = true;
1768 }
1769}
1770
1771template <class Impl>
1772void
1773FullO3CPU<Impl>::updateThreadPriority()
1774{
1775 if (activeThreads.size() > 1) {
1776 //DEFAULT TO ROUND ROBIN SCHEME
1777 //e.g. Move highest priority to end of thread list
1778 list<ThreadID>::iterator list_begin = activeThreads.begin();
1779
1780 unsigned high_thread = *list_begin;
1781
1782 activeThreads.erase(list_begin);
1783
1784 activeThreads.push_back(high_thread);
1785 }
1786}
1787
1788// Forward declaration of FullO3CPU.
1789template class FullO3CPU<O3CPUImpl>;
1724}
1725
1726template <class Impl>
1727void
1728FullO3CPU<Impl>::wakeup()
1729{
1730 if (this->thread[0]->status() != ThreadContext::Suspended)
1731 return;
1732
1733 this->wakeCPU();
1734
1735 DPRINTF(Quiesce, "Suspended Processor woken\n");
1736 this->threadContexts[0]->activate();
1737}
1738
1739template <class Impl>
1740ThreadID
1741FullO3CPU<Impl>::getFreeTid()
1742{
1743 for (ThreadID tid = 0; tid < numThreads; tid++) {
1744 if (!tids[tid]) {
1745 tids[tid] = true;
1746 return tid;
1747 }
1748 }
1749
1750 return InvalidThreadID;
1751}
1752
1753template <class Impl>
1754void
1755FullO3CPU<Impl>::doContextSwitch()
1756{
1757 if (contextSwitch) {
1758
1759 //ADD CODE TO DEACTIVE THREAD HERE (???)
1760
1761 ThreadID size = cpuWaitList.size();
1762 for (ThreadID tid = 0; tid < size; tid++) {
1763 activateWhenReady(tid);
1764 }
1765
1766 if (cpuWaitList.size() == 0)
1767 contextSwitch = true;
1768 }
1769}
1770
1771template <class Impl>
1772void
1773FullO3CPU<Impl>::updateThreadPriority()
1774{
1775 if (activeThreads.size() > 1) {
1776 //DEFAULT TO ROUND ROBIN SCHEME
1777 //e.g. Move highest priority to end of thread list
1778 list<ThreadID>::iterator list_begin = activeThreads.begin();
1779
1780 unsigned high_thread = *list_begin;
1781
1782 activeThreads.erase(list_begin);
1783
1784 activeThreads.push_back(high_thread);
1785 }
1786}
1787
1788// Forward declaration of FullO3CPU.
1789template class FullO3CPU<O3CPUImpl>;