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